kernel-aufs5.patch

   1 SPDX-License-Identifier: GPL-2.0
   2 aufs5.15.5 kbuild patch
   3
   4 diff --git a/fs/Kconfig b/fs/Kconfig
   5 index a6313a969bc5f..aca4b89d41a14 100644
   6 --- a/fs/Kconfig
   7 +++ b/fs/Kconfig
   8 @@ -312,6 +312,7 @@ source "fs/sysv/Kconfig"
   9  source "fs/ufs/Kconfig"
  10  source "fs/erofs/Kconfig"
  11  source "fs/vboxsf/Kconfig"
  12 +source "fs/aufs/Kconfig"
  13
  14  endif # MISC_FILESYSTEMS
  15
  16 diff --git a/fs/Makefile b/fs/Makefile
  17 index 84c5e4cdfee5a..b4fcdad8412ed 100644
  18 --- a/fs/Makefile
  19 +++ b/fs/Makefile
  20 @@ -138,3 +138,4 @@ obj-$(CONFIG_EFIVAR_FS)             += efivarfs/
  21  obj-$(CONFIG_EROFS_FS)         += erofs/
  22  obj-$(CONFIG_VBOXSF_FS)                += vboxsf/
  23  obj-$(CONFIG_ZONEFS_FS)                += zonefs/
  24 +obj-$(CONFIG_AUFS_FS)           += aufs/
  25 SPDX-License-Identifier: GPL-2.0
  26 aufs5.15.5 base patch
  27
  28 diff --git a/MAINTAINERS b/MAINTAINERS
  29 index 3b79fd441dde8..3fb0a57f61b05 100644
  30 --- a/MAINTAINERS
  31 +++ b/MAINTAINERS
  32 @@ -3122,6 +3122,19 @@ F:       include/uapi/linux/audit.h
  33  F:     kernel/audit*
  34  F:     lib/*audit.c
  35
  36 +AUFS (advanced multi layered unification filesystem) FILESYSTEM
  37 +M:     "J. R. Okajima" <hooanon05g@gmail.com>
  38 +L:     aufs-users@lists.sourceforge.net (members only)
  39 +L:     linux-unionfs@vger.kernel.org
  40 +S:     Supported
  41 +W:     http://aufs.sourceforge.net
  42 +T:     git://github.com/sfjro/aufs4-linux.git
  43 +F:     Documentation/ABI/testing/debugfs-aufs
  44 +F:     Documentation/ABI/testing/sysfs-aufs
  45 +F:     Documentation/filesystems/aufs/
  46 +F:     fs/aufs/
  47 +F:     include/uapi/linux/aufs_type.h
  48 +
  49  AUXILIARY DISPLAY DRIVERS
  50  M:     Miguel Ojeda <ojeda@kernel.org>
  51  S:     Maintained
  52 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
  53 index dfc72a1f6500d..c9d1ef703673c 100644
  54 --- a/drivers/block/loop.c
  55 +++ b/drivers/block/loop.c
  56 @@ -798,6 +798,24 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
  57         return error;
  58  }
  59
  60 +/*
  61 + * for AUFS
  62 + * no get/put for file.
  63 + */
  64 +struct file *loop_backing_file(struct super_block *sb)
  65 +{
  66 +       struct file *ret;
  67 +       struct loop_device *l;
  68 +
  69 +       ret = NULL;
  70 +       if (MAJOR(sb->s_dev) == LOOP_MAJOR) {
  71 +               l = sb->s_bdev->bd_disk->private_data;
  72 +               ret = l->lo_backing_file;
  73 +       }
  74 +       return ret;
  75 +}
  76 +EXPORT_SYMBOL_GPL(loop_backing_file);
  77 +
  78  /* loop sysfs attributes */
  79
  80  static ssize_t loop_attr_show(struct device *dev, char *page,
  81 diff --git a/fs/dcache.c b/fs/dcache.c
  82 index cf871a81f4fdc..bc5095b734f58 100644
  83 --- a/fs/dcache.c
  84 +++ b/fs/dcache.c
  85 @@ -1320,7 +1320,7 @@ enum d_walk_ret {
  86   *
  87   * The @enter() callbacks are called with d_lock held.
  88   */
  89 -static void d_walk(struct dentry *parent, void *data,
  90 +void d_walk(struct dentry *parent, void *data,
  91                    enum d_walk_ret (*enter)(void *, struct dentry *))
  92  {
  93         struct dentry *this_parent;
  94 diff --git a/fs/fcntl.c b/fs/fcntl.c
  95 index 9c6c6a3e2de51..02382fa9bd341 100644
  96 --- a/fs/fcntl.c
  97 +++ b/fs/fcntl.c
  98 @@ -33,7 +33,7 @@
  99
 100  #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
 101
 102 -static int setfl(int fd, struct file * filp, unsigned long arg)
 103 +int setfl(int fd, struct file *filp, unsigned long arg)
 104  {
 105         struct inode * inode = file_inode(filp);
 106         int error = 0;
 107 @@ -64,6 +64,8 @@ static int setfl(int fd, struct file * filp, unsigned long arg)
 108
 109         if (filp->f_op->check_flags)
 110                 error = filp->f_op->check_flags(arg);
 111 +       if (!error && filp->f_op->setfl)
 112 +               error = filp->f_op->setfl(filp, arg);
 113         if (error)
 114                 return error;
 115
 116 diff --git a/fs/namespace.c b/fs/namespace.c
 117 index 659a8f39c61af..1283670737826 100644
 118 --- a/fs/namespace.c
 119 +++ b/fs/namespace.c
 120 @@ -808,6 +808,12 @@ static inline int check_mnt(struct mount *mnt)
 121         return mnt->mnt_ns == current->nsproxy->mnt_ns;
 122  }
 123
 124 +/* for aufs, CONFIG_AUFS_BR_FUSE */
 125 +int is_current_mnt_ns(struct vfsmount *mnt)
 126 +{
 127 +       return check_mnt(real_mount(mnt));
 128 +}
 129 +
 130  /*
 131   * vfsmount lock must be held for write
 132   */
 133 diff --git a/fs/splice.c b/fs/splice.c
 134 index 5dbce4dcc1a7d..3e6ba363b7775 100644
 135 --- a/fs/splice.c
 136 +++ b/fs/splice.c
 137 @@ -759,8 +759,8 @@ static int warn_unsupported(struct file *file, const char *op)
 138  /*
 139   * Attempt to initiate a splice from pipe to file.
 140   */
 141 -static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 142 -                          loff_t *ppos, size_t len, unsigned int flags)
 143 +long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 144 +                   loff_t *ppos, size_t len, unsigned int flags)
 145  {
 146         if (unlikely(!out->f_op->splice_write))
 147                 return warn_unsupported(out, "write");
 148 @@ -770,9 +770,9 @@ static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 149  /*
 150   * Attempt to initiate a splice from a file to a pipe.
 151   */
 152 -static long do_splice_to(struct file *in, loff_t *ppos,
 153 -                        struct pipe_inode_info *pipe, size_t len,
 154 -                        unsigned int flags)
 155 +long do_splice_to(struct file *in, loff_t *ppos,
 156 +                 struct pipe_inode_info *pipe, size_t len,
 157 +                 unsigned int flags)
 158  {
 159         unsigned int p_space;
 160         int ret;
 161 diff --git a/fs/sync.c b/fs/sync.c
 162 index 1373a610dc784..b7b5a0a0df6ff 100644
 163 --- a/fs/sync.c
 164 +++ b/fs/sync.c
 165 @@ -28,7 +28,7 @@
 166   * wait == 1 case since in that case write_inode() functions do
 167   * sync_dirty_buffer() and thus effectively write one block at a time.
 168   */
 169 -static int __sync_filesystem(struct super_block *sb, int wait)
 170 +int __sync_filesystem(struct super_block *sb, int wait)
 171  {
 172         if (wait)
 173                 sync_inodes_sb(sb);
 174 diff --git a/include/linux/fs.h b/include/linux/fs.h
 175 index 56eba723477e3..e60d8ad854004 100644
 176 --- a/include/linux/fs.h
 177 +++ b/include/linux/fs.h
 178 @@ -1381,6 +1381,7 @@ extern void fasync_free(struct fasync_struct *);
 179  /* can be called from interrupts */
 180  extern void kill_fasync(struct fasync_struct **, int, int);
 181
 182 +extern int setfl(int fd, struct file *filp, unsigned long arg);
 183  extern void __f_setown(struct file *filp, struct pid *, enum pid_type, int force);
 184  extern int f_setown(struct file *filp, unsigned long arg, int force);
 185  extern void f_delown(struct file *filp);
 186 @@ -2092,6 +2093,7 @@ struct file_operations {
 187         ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
 188         unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
 189         int (*check_flags)(int);
 190 +       int (*setfl)(struct file *, unsigned long);
 191         int (*flock) (struct file *, int, struct file_lock *);
 192         ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
 193         ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
 194 @@ -2615,6 +2617,7 @@ extern int current_umask(void);
 195  extern void ihold(struct inode * inode);
 196  extern void iput(struct inode *);
 197  extern int generic_update_time(struct inode *, struct timespec64 *, int);
 198 +extern int update_time(struct inode *, struct timespec64 *, int);
 199
 200  /* /sys/fs */
 201  extern struct kobject *fs_kobj;
 202 @@ -2778,6 +2781,7 @@ static inline bool sb_is_blkdev_sb(struct super_block *sb)
 203  }
 204
 205  void emergency_thaw_all(void);
 206 +extern int __sync_filesystem(struct super_block *, int);
 207  extern int sync_filesystem(struct super_block *);
 208  extern const struct file_operations def_blk_fops;
 209  extern const struct file_operations def_chr_fops;
 210 diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h
 211 index 9fe165beb0f9e..e47f7e15eeaf7 100644
 212 --- a/include/linux/lockdep.h
 213 +++ b/include/linux/lockdep.h
 214 @@ -248,6 +248,8 @@ static inline int lockdep_match_key(struct lockdep_map *lock,
 215         return lock->key == key;
 216  }
 217
 218 +struct lock_class *lockdep_hlock_class(struct held_lock *hlock);
 219 +
 220  /*
 221   * Acquire a lock.
 222   *
 223 diff --git a/include/linux/mnt_namespace.h b/include/linux/mnt_namespace.h
 224 index 8f882f5881e87..6b9808f098435 100644
 225 --- a/include/linux/mnt_namespace.h
 226 +++ b/include/linux/mnt_namespace.h
 227 @@ -7,12 +7,15 @@ struct mnt_namespace;
 228  struct fs_struct;
 229  struct user_namespace;
 230  struct ns_common;
 231 +struct vfsmount;
 232
 233  extern struct mnt_namespace *copy_mnt_ns(unsigned long, struct mnt_namespace *,
 234                 struct user_namespace *, struct fs_struct *);
 235  extern void put_mnt_ns(struct mnt_namespace *ns);
 236  extern struct ns_common *from_mnt_ns(struct mnt_namespace *);
 237
 238 +extern int is_current_mnt_ns(struct vfsmount *mnt);
 239 +
 240  extern const struct file_operations proc_mounts_operations;
 241  extern const struct file_operations proc_mountinfo_operations;
 242  extern const struct file_operations proc_mountstats_operations;
 243 diff --git a/include/linux/splice.h b/include/linux/splice.h
 244 index a55179fd60fc3..8e21c53cf8831 100644
 245 --- a/include/linux/splice.h
 246 +++ b/include/linux/splice.h
 247 @@ -93,4 +93,10 @@ extern void splice_shrink_spd(struct splice_pipe_desc *);
 248
 249  extern const struct pipe_buf_operations page_cache_pipe_buf_ops;
 250  extern const struct pipe_buf_operations default_pipe_buf_ops;
 251 +
 252 +extern long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 253 +                          loff_t *ppos, size_t len, unsigned int flags);
 254 +extern long do_splice_to(struct file *in, loff_t *ppos,
 255 +                        struct pipe_inode_info *pipe, size_t len,
 256 +                        unsigned int flags);
 257  #endif
 258 diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
 259 index d624231eab2bb..eff2ca26e6c58 100644
 260 --- a/kernel/locking/lockdep.c
 261 +++ b/kernel/locking/lockdep.c
 262 @@ -189,7 +189,7 @@ static
 263  struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
 264  static DECLARE_BITMAP(lock_classes_in_use, MAX_LOCKDEP_KEYS);
 265
 266 -static inline struct lock_class *hlock_class(struct held_lock *hlock)
 267 +inline struct lock_class *lockdep_hlock_class(struct held_lock *hlock)
 268  {
 269         unsigned int class_idx = hlock->class_idx;
 270
 271 @@ -210,6 +210,7 @@ static inline struct lock_class *hlock_class(struct held_lock *hlock)
 272          */
 273         return lock_classes + class_idx;
 274  }
 275 +#define hlock_class(hlock) lockdep_hlock_class(hlock)
 276
 277  #ifdef CONFIG_LOCK_STAT
 278  static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
 279 SPDX-License-Identifier: GPL-2.0
 280 aufs5.15.5 mmap patch
 281
 282 diff --git a/fs/proc/base.c b/fs/proc/base.c
 283 index 1f394095eb880..93f2479ef319e 100644
 284 --- a/fs/proc/base.c
 285 +++ b/fs/proc/base.c
 286 @@ -2189,7 +2189,7 @@ static int map_files_get_link(struct dentry *dentry, struct path *path)
 287         rc = -ENOENT;
 288         vma = find_exact_vma(mm, vm_start, vm_end);
 289         if (vma && vma->vm_file) {
 290 -               *path = vma->vm_file->f_path;
 291 +               *path = vma_pr_or_file(vma)->f_path;
 292                 path_get(path);
 293                 rc = 0;
 294         }
 295 diff --git a/fs/proc/nommu.c b/fs/proc/nommu.c
 296 index 13452b32e2bd5..38acccfef9d49 100644
 297 --- a/fs/proc/nommu.c
 298 +++ b/fs/proc/nommu.c
 299 @@ -40,7 +40,10 @@ static int nommu_region_show(struct seq_file *m, struct vm_region *region)
 300         file = region->vm_file;
 301
 302         if (file) {
 303 -               struct inode *inode = file_inode(region->vm_file);
 304 +               struct inode *inode;
 305 +
 306 +               file = vmr_pr_or_file(region);
 307 +               inode = file_inode(file);
 308                 dev = inode->i_sb->s_dev;
 309                 ino = inode->i_ino;
 310         }
 311 diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
 312 index cf25be3e03212..70af1c25ffa22 100644
 313 --- a/fs/proc/task_mmu.c
 314 +++ b/fs/proc/task_mmu.c
 315 @@ -280,7 +280,10 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma)
 316         const char *name = NULL;
 317
 318         if (file) {
 319 -               struct inode *inode = file_inode(vma->vm_file);
 320 +               struct inode *inode;
 321 +
 322 +               file = vma_pr_or_file(vma);
 323 +               inode = file_inode(file);
 324                 dev = inode->i_sb->s_dev;
 325                 ino = inode->i_ino;
 326                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 327 @@ -1865,7 +1868,7 @@ static int show_numa_map(struct seq_file *m, void *v)
 328         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
 329         struct vm_area_struct *vma = v;
 330         struct numa_maps *md = &numa_priv->md;
 331 -       struct file *file = vma->vm_file;
 332 +       struct file *file = vma_pr_or_file(vma);
 333         struct mm_struct *mm = vma->vm_mm;
 334         struct mempolicy *pol;
 335         char buffer[64];
 336 diff --git a/fs/proc/task_nommu.c b/fs/proc/task_nommu.c
 337 index a6d21fc0033c6..02c2de31196e0 100644
 338 --- a/fs/proc/task_nommu.c
 339 +++ b/fs/proc/task_nommu.c
 340 @@ -155,7 +155,10 @@ static int nommu_vma_show(struct seq_file *m, struct vm_area_struct *vma)
 341         file = vma->vm_file;
 342
 343         if (file) {
 344 -               struct inode *inode = file_inode(vma->vm_file);
 345 +               struct inode *inode;
 346 +
 347 +               file = vma_pr_or_file(vma);
 348 +               inode = file_inode(file);
 349                 dev = inode->i_sb->s_dev;
 350                 ino = inode->i_ino;
 351                 pgoff = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
 352 diff --git a/include/linux/mm.h b/include/linux/mm.h
 353 index 73a52aba448f9..5dd42acf0707b 100644
 354 --- a/include/linux/mm.h
 355 +++ b/include/linux/mm.h
 356 @@ -1806,6 +1806,28 @@ static inline void unmap_shared_mapping_range(struct address_space *mapping,
 357         unmap_mapping_range(mapping, holebegin, holelen, 0);
 358  }
 359
 360 +extern void vma_do_file_update_time(struct vm_area_struct *, const char[], int);
 361 +extern struct file *vma_do_pr_or_file(struct vm_area_struct *, const char[],
 362 +                                     int);
 363 +extern void vma_do_get_file(struct vm_area_struct *, const char[], int);
 364 +extern void vma_do_fput(struct vm_area_struct *, const char[], int);
 365 +
 366 +#define vma_file_update_time(vma)      vma_do_file_update_time(vma, __func__, \
 367 +                                                               __LINE__)
 368 +#define vma_pr_or_file(vma)            vma_do_pr_or_file(vma, __func__, \
 369 +                                                         __LINE__)
 370 +#define vma_get_file(vma)              vma_do_get_file(vma, __func__, __LINE__)
 371 +#define vma_fput(vma)                  vma_do_fput(vma, __func__, __LINE__)
 372 +
 373 +#ifndef CONFIG_MMU
 374 +extern struct file *vmr_do_pr_or_file(struct vm_region *, const char[], int);
 375 +extern void vmr_do_fput(struct vm_region *, const char[], int);
 376 +
 377 +#define vmr_pr_or_file(region)         vmr_do_pr_or_file(region, __func__, \
 378 +                                                         __LINE__)
 379 +#define vmr_fput(region)               vmr_do_fput(region, __func__, __LINE__)
 380 +#endif /* !CONFIG_MMU */
 381 +
 382  extern int access_process_vm(struct task_struct *tsk, unsigned long addr,
 383                 void *buf, int len, unsigned int gup_flags);
 384  extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
 385 diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
 386 index 7f8ee09c711f4..3a9a798a4ae1f 100644
 387 --- a/include/linux/mm_types.h
 388 +++ b/include/linux/mm_types.h
 389 @@ -294,6 +294,7 @@ struct vm_region {
 390         unsigned long   vm_top;         /* region allocated to here */
 391         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 392         struct file     *vm_file;       /* the backing file or NULL */
 393 +       struct file     *vm_prfile;     /* the virtual backing file or NULL */
 394
 395         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 396         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 397 @@ -373,6 +374,7 @@ struct vm_area_struct {
 398         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 399                                            units */
 400         struct file * vm_file;          /* File we map to (can be NULL). */
 401 +       struct file *vm_prfile;         /* shadow of vm_file */
 402         void * vm_private_data;         /* was vm_pte (shared mem) */
 403
 404  #ifdef CONFIG_SWAP
 405 diff --git a/kernel/fork.c b/kernel/fork.c
 406 index 10885c649ca42..f3a9cd12011b3 100644
 407 --- a/kernel/fork.c
 408 +++ b/kernel/fork.c
 409 @@ -573,7 +573,7 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
 410                 if (file) {
 411                         struct address_space *mapping = file->f_mapping;
 412
 413 -                       get_file(file);
 414 +                       vma_get_file(tmp);
 415                         i_mmap_lock_write(mapping);
 416                         if (tmp->vm_flags & VM_SHARED)
 417                                 mapping_allow_writable(mapping);
 418 diff --git a/mm/Makefile b/mm/Makefile
 419 index fc60a40ce954b..c715b0138237b 100644
 420 --- a/mm/Makefile
 421 +++ b/mm/Makefile
 422 @@ -52,7 +52,7 @@ obj-y                 := filemap.o mempool.o oom_kill.o fadvise.o \
 423                            mm_init.o percpu.o slab_common.o \
 424                            compaction.o vmacache.o \
 425                            interval_tree.o list_lru.o workingset.o \
 426 -                          debug.o gup.o mmap_lock.o $(mmu-y)
 427 +                          prfile.o debug.o gup.o mmap_lock.o $(mmu-y)
 428
 429  # Give 'page_alloc' its own module-parameter namespace
 430  page-alloc-y := page_alloc.o
 431 diff --git a/mm/filemap.c b/mm/filemap.c
 432 index 82a17c35eb96b..348e22067abdc 100644
 433 --- a/mm/filemap.c
 434 +++ b/mm/filemap.c
 435 @@ -3349,7 +3349,7 @@ vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
 436         vm_fault_t ret = VM_FAULT_LOCKED;
 437
 438         sb_start_pagefault(mapping->host->i_sb);
 439 -       file_update_time(vmf->vma->vm_file);
 440 +       vma_file_update_time(vmf->vma);
 441         lock_page(page);
 442         if (page->mapping != mapping) {
 443                 unlock_page(page);
 444 diff --git a/mm/mmap.c b/mm/mmap.c
 445 index 88dcc5c252255..6c276614ca962 100644
 446 --- a/mm/mmap.c
 447 +++ b/mm/mmap.c
 448 @@ -183,7 +183,7 @@ static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 449         if (vma->vm_ops && vma->vm_ops->close)
 450                 vma->vm_ops->close(vma);
 451         if (vma->vm_file)
 452 -               fput(vma->vm_file);
 453 +               vma_fput(vma);
 454         mpol_put(vma_policy(vma));
 455         vm_area_free(vma);
 456         return next;
 457 @@ -952,7 +952,7 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 458         if (remove_next) {
 459                 if (file) {
 460                         uprobe_munmap(next, next->vm_start, next->vm_end);
 461 -                       fput(file);
 462 +                       vma_fput(vma);
 463                 }
 464                 if (next->anon_vma)
 465                         anon_vma_merge(vma, next);
 466 @@ -1873,7 +1873,7 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
 467         return addr;
 468
 469  unmap_and_free_vma:
 470 -       fput(vma->vm_file);
 471 +       vma_fput(vma);
 472         vma->vm_file = NULL;
 473
 474         /* Undo any partial mapping done by a device driver. */
 475 @@ -2731,7 +2731,7 @@ int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
 476                 goto out_free_mpol;
 477
 478         if (new->vm_file)
 479 -               get_file(new->vm_file);
 480 +               vma_get_file(new);
 481
 482         if (new->vm_ops && new->vm_ops->open)
 483                 new->vm_ops->open(new);
 484 @@ -2750,7 +2750,7 @@ int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
 485         if (new->vm_ops && new->vm_ops->close)
 486                 new->vm_ops->close(new);
 487         if (new->vm_file)
 488 -               fput(new->vm_file);
 489 +               vma_fput(new);
 490         unlink_anon_vmas(new);
 491   out_free_mpol:
 492         mpol_put(vma_policy(new));
 493 @@ -2945,7 +2945,7 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 494         struct vm_area_struct *vma;
 495         unsigned long populate = 0;
 496         unsigned long ret = -EINVAL;
 497 -       struct file *file;
 498 +       struct file *file, *prfile;
 499
 500         pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.rst.\n",
 501                      current->comm, current->pid);
 502 @@ -3001,10 +3001,27 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 503         if (vma->vm_flags & VM_LOCKED)
 504                 flags |= MAP_LOCKED;
 505
 506 -       file = get_file(vma->vm_file);
 507 +       vma_get_file(vma);
 508 +       file = vma->vm_file;
 509 +       prfile = vma->vm_prfile;
 510         ret = do_mmap(vma->vm_file, start, size,
 511                         prot, flags, pgoff, &populate, NULL);
 512 +       if (!IS_ERR_VALUE(ret) && file && prfile) {
 513 +               struct vm_area_struct *new_vma;
 514 +
 515 +               new_vma = find_vma(mm, ret);
 516 +               if (!new_vma->vm_prfile)
 517 +                       new_vma->vm_prfile = prfile;
 518 +               if (new_vma != vma)
 519 +                       get_file(prfile);
 520 +       }
 521 +       /*
 522 +        * two fput()s instead of vma_fput(vma),
 523 +        * coz vma may not be available anymore.
 524 +        */
 525         fput(file);
 526 +       if (prfile)
 527 +               fput(prfile);
 528  out:
 529         mmap_write_unlock(mm);
 530         if (populate)
 531 @@ -3285,7 +3302,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
 532                 if (anon_vma_clone(new_vma, vma))
 533                         goto out_free_mempol;
 534                 if (new_vma->vm_file)
 535 -                       get_file(new_vma->vm_file);
 536 +                       vma_get_file(new_vma);
 537                 if (new_vma->vm_ops && new_vma->vm_ops->open)
 538                         new_vma->vm_ops->open(new_vma);
 539                 vma_link(mm, new_vma, prev, rb_link, rb_parent);
 540 diff --git a/mm/nommu.c b/mm/nommu.c
 541 index 02d2427b8f9e8..a7419302ab4ed 100644
 542 --- a/mm/nommu.c
 543 +++ b/mm/nommu.c
 544 @@ -523,7 +523,7 @@ static void __put_nommu_region(struct vm_region *region)
 545                 up_write(&nommu_region_sem);
 546
 547                 if (region->vm_file)
 548 -                       fput(region->vm_file);
 549 +                       vmr_fput(region);
 550
 551                 /* IO memory and memory shared directly out of the pagecache
 552                  * from ramfs/tmpfs mustn't be released here */
 553 @@ -655,7 +655,7 @@ static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 554         if (vma->vm_ops && vma->vm_ops->close)
 555                 vma->vm_ops->close(vma);
 556         if (vma->vm_file)
 557 -               fput(vma->vm_file);
 558 +               vma_fput(vma);
 559         put_nommu_region(vma->vm_region);
 560         vm_area_free(vma);
 561  }
 562 @@ -1175,7 +1175,7 @@ unsigned long do_mmap(struct file *file,
 563                                         goto error_just_free;
 564                                 }
 565                         }
 566 -                       fput(region->vm_file);
 567 +                       vmr_fput(region);
 568                         kmem_cache_free(vm_region_jar, region);
 569                         region = pregion;
 570                         result = start;
 571 @@ -1252,10 +1252,10 @@ unsigned long do_mmap(struct file *file,
 572         up_write(&nommu_region_sem);
 573  error:
 574         if (region->vm_file)
 575 -               fput(region->vm_file);
 576 +               vmr_fput(region);
 577         kmem_cache_free(vm_region_jar, region);
 578         if (vma->vm_file)
 579 -               fput(vma->vm_file);
 580 +               vma_fput(vma);
 581         vm_area_free(vma);
 582         return ret;
 583
 584 diff --git a/mm/prfile.c b/mm/prfile.c
 585 new file mode 100644
 586 index 0000000000000..511543ab1b418
 587 --- /dev/null
 588 +++ b/mm/prfile.c
 589 @@ -0,0 +1,86 @@
 590 +// SPDX-License-Identifier: GPL-2.0
 591 +/*
 592 + * Mainly for aufs which mmap(2) different file and wants to print different
 593 + * path in /proc/PID/maps.
 594 + * Call these functions via macros defined in linux/mm.h.
 595 + *
 596 + * See Documentation/filesystems/aufs/design/06mmap.txt
 597 + *
 598 + * Copyright (c) 2014-2021 Junjro R. Okajima
 599 + * Copyright (c) 2014 Ian Campbell
 600 + */
 601 +
 602 +#include <linux/mm.h>
 603 +#include <linux/file.h>
 604 +#include <linux/fs.h>
 605 +
 606 +/* #define PRFILE_TRACE */
 607 +static inline void prfile_trace(struct file *f, struct file *pr,
 608 +                             const char func[], int line, const char func2[])
 609 +{
 610 +#ifdef PRFILE_TRACE
 611 +       if (pr)
 612 +               pr_info("%s:%d: %s, %pD2\n", func, line, func2, f);
 613 +#endif
 614 +}
 615 +
 616 +void vma_do_file_update_time(struct vm_area_struct *vma, const char func[],
 617 +                            int line)
 618 +{
 619 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
 620 +
 621 +       prfile_trace(f, pr, func, line, __func__);
 622 +       file_update_time(f);
 623 +       if (f && pr)
 624 +               file_update_time(pr);
 625 +}
 626 +
 627 +struct file *vma_do_pr_or_file(struct vm_area_struct *vma, const char func[],
 628 +                              int line)
 629 +{
 630 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
 631 +
 632 +       prfile_trace(f, pr, func, line, __func__);
 633 +       return (f && pr) ? pr : f;
 634 +}
 635 +
 636 +void vma_do_get_file(struct vm_area_struct *vma, const char func[], int line)
 637 +{
 638 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
 639 +
 640 +       prfile_trace(f, pr, func, line, __func__);
 641 +       get_file(f);
 642 +       if (f && pr)
 643 +               get_file(pr);
 644 +}
 645 +
 646 +void vma_do_fput(struct vm_area_struct *vma, const char func[], int line)
 647 +{
 648 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
 649 +
 650 +       prfile_trace(f, pr, func, line, __func__);
 651 +       fput(f);
 652 +       if (f && pr)
 653 +               fput(pr);
 654 +}
 655 +
 656 +#ifndef CONFIG_MMU
 657 +struct file *vmr_do_pr_or_file(struct vm_region *region, const char func[],
 658 +                              int line)
 659 +{
 660 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
 661 +
 662 +       prfile_trace(f, pr, func, line, __func__);
 663 +       return (f && pr) ? pr : f;
 664 +}
 665 +
 666 +void vmr_do_fput(struct vm_region *region, const char func[], int line)
 667 +{
 668 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
 669 +
 670 +       prfile_trace(f, pr, func, line, __func__);
 671 +       fput(f);
 672 +       if (f && pr)
 673 +               fput(pr);
 674 +}
 675 +#endif /* !CONFIG_MMU */
 676 SPDX-License-Identifier: GPL-2.0
 677 aufs5.15.5 standalone patch
 678
 679 diff --git a/fs/dcache.c b/fs/dcache.c
 680 index bc5095b734f58..9508bd57a3bc0 100644
 681 --- a/fs/dcache.c
 682 +++ b/fs/dcache.c
 683 @@ -1425,6 +1425,7 @@ void d_walk(struct dentry *parent, void *data,
 684         seq = 1;
 685         goto again;
 686  }
 687 +EXPORT_SYMBOL_GPL(d_walk);
 688
 689  struct check_mount {
 690         struct vfsmount *mnt;
 691 @@ -2970,6 +2971,7 @@ void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
 692
 693         write_sequnlock(&rename_lock);
 694  }
 695 +EXPORT_SYMBOL_GPL(d_exchange);
 696
 697  /**
 698   * d_ancestor - search for an ancestor
 699 diff --git a/fs/exec.c b/fs/exec.c
 700 index ac7b51b51f38a..52a8be4ebc1e1 100644
 701 --- a/fs/exec.c
 702 +++ b/fs/exec.c
 703 @@ -111,6 +111,7 @@ bool path_noexec(const struct path *path)
 704         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 705                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 706  }
 707 +EXPORT_SYMBOL_GPL(path_noexec);
 708
 709  #ifdef CONFIG_USELIB
 710  /*
 711 diff --git a/fs/fcntl.c b/fs/fcntl.c
 712 index 02382fa9bd341..3418c60b90146 100644
 713 --- a/fs/fcntl.c
 714 +++ b/fs/fcntl.c
 715 @@ -86,6 +86,7 @@ int setfl(int fd, struct file *filp, unsigned long arg)
 716   out:
 717         return error;
 718  }
 719 +EXPORT_SYMBOL_GPL(setfl);
 720
 721  static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 722                       int force)
 723 diff --git a/fs/file_table.c b/fs/file_table.c
 724 index 45437f8e1003e..786af52904fcf 100644
 725 --- a/fs/file_table.c
 726 +++ b/fs/file_table.c
 727 @@ -161,6 +161,7 @@ struct file *alloc_empty_file(int flags, const struct cred *cred)
 728         }
 729         return ERR_PTR(-ENFILE);
 730  }
 731 +EXPORT_SYMBOL_GPL(alloc_empty_file);
 732
 733  /*
 734   * Variant of alloc_empty_file() that doesn't check and modify nr_files.
 735 @@ -375,6 +376,7 @@ void __fput_sync(struct file *file)
 736  }
 737
 738  EXPORT_SYMBOL(fput);
 739 +EXPORT_SYMBOL_GPL(__fput_sync);
 740
 741  void __init files_init(void)
 742  {
 743 diff --git a/fs/namespace.c b/fs/namespace.c
 744 index 1283670737826..db99365620118 100644
 745 --- a/fs/namespace.c
 746 +++ b/fs/namespace.c
 747 @@ -439,6 +439,7 @@ void __mnt_drop_write(struct vfsmount *mnt)
 748         mnt_dec_writers(real_mount(mnt));
 749         preempt_enable();
 750  }
 751 +EXPORT_SYMBOL_GPL(__mnt_drop_write);
 752
 753  /**
 754   * mnt_drop_write - give up write access to a mount
 755 @@ -813,6 +814,7 @@ int is_current_mnt_ns(struct vfsmount *mnt)
 756  {
 757         return check_mnt(real_mount(mnt));
 758  }
 759 +EXPORT_SYMBOL_GPL(is_current_mnt_ns);
 760
 761  /*
 762   * vfsmount lock must be held for write
 763 @@ -2011,6 +2013,7 @@ int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
 764         }
 765         return 0;
 766  }
 767 +EXPORT_SYMBOL_GPL(iterate_mounts);
 768
 769  static void lock_mnt_tree(struct mount *mnt)
 770  {
 771 diff --git a/fs/notify/group.c b/fs/notify/group.c
 772 index fb89c351295d6..460ad19c2570a 100644
 773 --- a/fs/notify/group.c
 774 +++ b/fs/notify/group.c
 775 @@ -100,6 +100,7 @@ void fsnotify_get_group(struct fsnotify_group *group)
 776  {
 777         refcount_inc(&group->refcnt);
 778  }
 779 +EXPORT_SYMBOL_GPL(fsnotify_get_group);
 780
 781  /*
 782   * Drop a reference to a group.  Free it if it's through.
 783 diff --git a/fs/open.c b/fs/open.c
 784 index e0df1536eb69f..81b2d7c83addb 100644
 785 --- a/fs/open.c
 786 +++ b/fs/open.c
 787 @@ -65,6 +65,7 @@ int do_truncate(struct user_namespace *mnt_userns, struct dentry *dentry,
 788         inode_unlock(dentry->d_inode);
 789         return ret;
 790  }
 791 +EXPORT_SYMBOL_GPL(do_truncate);
 792
 793  long vfs_truncate(const struct path *path, loff_t length)
 794  {
 795 diff --git a/fs/read_write.c b/fs/read_write.c
 796 index af057c57bdc64..76017f8331fbf 100644
 797 --- a/fs/read_write.c
 798 +++ b/fs/read_write.c
 799 @@ -492,6 +492,7 @@ ssize_t vfs_read(struct file *file, char __user *buf, size_t count, loff_t *pos)
 800         inc_syscr(current);
 801         return ret;
 802  }
 803 +EXPORT_SYMBOL_GPL(vfs_read);
 804
 805  static ssize_t new_sync_write(struct file *filp, const char __user *buf, size_t len, loff_t *ppos)
 806  {
 807 @@ -602,6 +603,7 @@ ssize_t vfs_write(struct file *file, const char __user *buf, size_t count, loff_
 808         file_end_write(file);
 809         return ret;
 810  }
 811 +EXPORT_SYMBOL_GPL(vfs_write);
 812
 813  /* file_ppos returns &file->f_pos or NULL if file is stream */
 814  static inline loff_t *file_ppos(struct file *file)
 815 diff --git a/fs/splice.c b/fs/splice.c
 816 index 3e6ba363b7775..7c1be373eb7cd 100644
 817 --- a/fs/splice.c
 818 +++ b/fs/splice.c
 819 @@ -766,6 +766,7 @@ long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 820                 return warn_unsupported(out, "write");
 821         return out->f_op->splice_write(pipe, out, ppos, len, flags);
 822  }
 823 +EXPORT_SYMBOL_GPL(do_splice_from);
 824
 825  /*
 826   * Attempt to initiate a splice from a file to a pipe.
 827 @@ -795,6 +796,7 @@ long do_splice_to(struct file *in, loff_t *ppos,
 828                 return warn_unsupported(in, "read");
 829         return in->f_op->splice_read(in, ppos, pipe, len, flags);
 830  }
 831 +EXPORT_SYMBOL_GPL(do_splice_to);
 832
 833  /**
 834   * splice_direct_to_actor - splices data directly between two non-pipes
 835 diff --git a/fs/sync.c b/fs/sync.c
 836 index b7b5a0a0df6ff..fa5c7fba7f1ba 100644
 837 --- a/fs/sync.c
 838 +++ b/fs/sync.c
 839 @@ -39,6 +39,7 @@ int __sync_filesystem(struct super_block *sb, int wait)
 840                 sb->s_op->sync_fs(sb, wait);
 841         return __sync_blockdev(sb->s_bdev, wait);
 842  }
 843 +EXPORT_SYMBOL_GPL(__sync_filesystem);
 844
 845  /*
 846   * Write out and wait upon all dirty data associated with this
 847 diff --git a/fs/xattr.c b/fs/xattr.c
 848 index 5c8c5175b385c..ff7e9ff774b73 100644
 849 --- a/fs/xattr.c
 850 +++ b/fs/xattr.c
 851 @@ -384,6 +384,7 @@ vfs_getxattr_alloc(struct user_namespace *mnt_userns, struct dentry *dentry,
 852         *xattr_value = value;
 853         return error;
 854  }
 855 +EXPORT_SYMBOL_GPL(vfs_getxattr_alloc);
 856
 857  ssize_t
 858  __vfs_getxattr(struct dentry *dentry, struct inode *inode, const char *name,
 859 diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
 860 index eff2ca26e6c58..0627584f7872a 100644
 861 --- a/kernel/locking/lockdep.c
 862 +++ b/kernel/locking/lockdep.c
 863 @@ -210,6 +210,7 @@ inline struct lock_class *lockdep_hlock_class(struct held_lock *hlock)
 864          */
 865         return lock_classes + class_idx;
 866  }
 867 +EXPORT_SYMBOL_GPL(lockdep_hlock_class);
 868  #define hlock_class(hlock) lockdep_hlock_class(hlock)
 869
 870  #ifdef CONFIG_LOCK_STAT
 871 diff --git a/kernel/task_work.c b/kernel/task_work.c
 872 index 1698fbe6f0e13..081b05acadf82 100644
 873 --- a/kernel/task_work.c
 874 +++ b/kernel/task_work.c
 875 @@ -167,3 +167,4 @@ void task_work_run(void)
 876                 } while (work);
 877         }
 878  }
 879 +EXPORT_SYMBOL_GPL(task_work_run);
 880 diff --git a/security/security.c b/security/security.c
 881 index 67264cb08fb31..0d78065d71fe1 100644
 882 --- a/security/security.c
 883 +++ b/security/security.c
 884 @@ -1147,6 +1147,7 @@ int security_path_rmdir(const struct path *dir, struct dentry *dentry)
 885                 return 0;
 886         return call_int_hook(path_rmdir, 0, dir, dentry);
 887  }
 888 +EXPORT_SYMBOL_GPL(security_path_rmdir);
 889
 890  int security_path_unlink(const struct path *dir, struct dentry *dentry)
 891  {
 892 @@ -1163,6 +1164,7 @@ int security_path_symlink(const struct path *dir, struct dentry *dentry,
 893                 return 0;
 894         return call_int_hook(path_symlink, 0, dir, dentry, old_name);
 895  }
 896 +EXPORT_SYMBOL_GPL(security_path_symlink);
 897
 898  int security_path_link(struct dentry *old_dentry, const struct path *new_dir,
 899                        struct dentry *new_dentry)
 900 @@ -1171,6 +1173,7 @@ int security_path_link(struct dentry *old_dentry, const struct path *new_dir,
 901                 return 0;
 902         return call_int_hook(path_link, 0, old_dentry, new_dir, new_dentry);
 903  }
 904 +EXPORT_SYMBOL_GPL(security_path_link);
 905
 906  int security_path_rename(const struct path *old_dir, struct dentry *old_dentry,
 907                          const struct path *new_dir, struct dentry *new_dentry,
 908 @@ -1198,6 +1201,7 @@ int security_path_truncate(const struct path *path)
 909                 return 0;
 910         return call_int_hook(path_truncate, 0, path);
 911  }
 912 +EXPORT_SYMBOL_GPL(security_path_truncate);
 913
 914  int security_path_chmod(const struct path *path, umode_t mode)
 915  {
 916 @@ -1205,6 +1209,7 @@ int security_path_chmod(const struct path *path, umode_t mode)
 917                 return 0;
 918         return call_int_hook(path_chmod, 0, path, mode);
 919  }
 920 +EXPORT_SYMBOL_GPL(security_path_chmod);
 921
 922  int security_path_chown(const struct path *path, kuid_t uid, kgid_t gid)
 923  {
 924 @@ -1212,6 +1217,7 @@ int security_path_chown(const struct path *path, kuid_t uid, kgid_t gid)
 925                 return 0;
 926         return call_int_hook(path_chown, 0, path, uid, gid);
 927  }
 928 +EXPORT_SYMBOL_GPL(security_path_chown);
 929
 930  int security_path_chroot(const struct path *path)
 931  {
 932 @@ -1312,6 +1318,7 @@ int security_inode_permission(struct inode *inode, int mask)
 933                 return 0;
 934         return call_int_hook(inode_permission, 0, inode, mask);
 935  }
 936 +EXPORT_SYMBOL_GPL(security_inode_permission);
 937
 938  int security_inode_setattr(struct dentry *dentry, struct iattr *attr)
 939  {
 940 @@ -1509,6 +1516,7 @@ int security_file_permission(struct file *file, int mask)
 941
 942         return fsnotify_perm(file, mask);
 943  }
 944 +EXPORT_SYMBOL_GPL(security_file_permission);
 945
 946  int security_file_alloc(struct file *file)
 947  {
 948 diff -urN /usr/share/empty/Documentation/ABI/testing/debugfs-aufs linux/Documentation/ABI/testing/debugfs-aufs
 949 --- /usr/share/empty/Documentation/ABI/testing/debugfs-aufs     1970-01-01 01:00:00.000000000 +0100
 950 +++ linux/Documentation/ABI/testing/debugfs-aufs        2021-12-03 15:38:59.926647310 +0100
 951 @@ -0,0 +1,55 @@
 952 +What:          /debug/aufs/si_<id>/
 953 +Date:          March 2009
 954 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
 955 +Description:
 956 +               Under /debug/aufs, a directory named si_<id> is created
 957 +               per aufs mount, where <id> is a unique id generated
 958 +               internally.
 959 +
 960 +What:          /debug/aufs/si_<id>/plink
 961 +Date:          Apr 2013
 962 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
 963 +Description:
 964 +               It has three lines and shows the information about the
 965 +               pseudo-link. The first line is a single number
 966 +               representing a number of buckets. The second line is a
 967 +               number of pseudo-links per buckets (separated by a
 968 +               blank). The last line is a single number representing a
 969 +               total number of psedo-links.
 970 +               When the aufs mount option 'noplink' is specified, it
 971 +               will show "1\n0\n0\n".
 972 +
 973 +What:          /debug/aufs/si_<id>/xib
 974 +Date:          March 2009
 975 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
 976 +Description:
 977 +               It shows the consumed blocks by xib (External Inode Number
 978 +               Bitmap), its block size and file size.
 979 +               When the aufs mount option 'noxino' is specified, it
 980 +               will be empty. About XINO files, see the aufs manual.
 981 +
 982 +What:          /debug/aufs/si_<id>/xi<branch-index>
 983 +Date:          March 2009
 984 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
 985 +Description:
 986 +               It shows the consumed blocks by xino (External Inode Number
 987 +               Translation Table), its link count, block size and file
 988 +               size.
 989 +               Due to the file size limit, there may exist multiple
 990 +               xino files per branch.  In this case, "-N" is added to
 991 +               the filename and it corresponds to the index of the
 992 +               internal xino array.  "-0" is omitted.
 993 +               When the aufs mount option 'noxino' is specified, Those
 994 +               entries won't exist.  About XINO files, see the aufs
 995 +               manual.
 996 +
 997 +What:          /debug/aufs/si_<id>/xigen
 998 +Date:          March 2009
 999 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1000 +Description:
1001 +               It shows the consumed blocks by xigen (External Inode
1002 +               Generation Table), its block size and file size.
1003 +               If CONFIG_AUFS_EXPORT is disabled, this entry will not
1004 +               be created.
1005 +               When the aufs mount option 'noxino' is specified, it
1006 +               will be empty. About XINO files, see the aufs manual.
1007 diff -urN /usr/share/empty/Documentation/ABI/testing/sysfs-aufs linux/Documentation/ABI/testing/sysfs-aufs
1008 --- /usr/share/empty/Documentation/ABI/testing/sysfs-aufs       1970-01-01 01:00:00.000000000 +0100
1009 +++ linux/Documentation/ABI/testing/sysfs-aufs  2021-12-03 15:38:59.926647310 +0100
1010 @@ -0,0 +1,31 @@
1011 +What:          /sys/fs/aufs/si_<id>/
1012 +Date:          March 2009
1013 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1014 +Description:
1015 +               Under /sys/fs/aufs, a directory named si_<id> is created
1016 +               per aufs mount, where <id> is a unique id generated
1017 +               internally.
1018 +
1019 +What:          /sys/fs/aufs/si_<id>/br<idx>
1020 +Date:          March 2009
1021 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1022 +Description:
1023 +               It shows the abolute path of a member directory (which
1024 +               is called branch) in aufs, and its permission.
1025 +
1026 +What:          /sys/fs/aufs/si_<id>/brid<idx>
1027 +Date:          July 2013
1028 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1029 +Description:
1030 +               It shows the id of a member directory (which is called
1031 +               branch) in aufs.
1032 +
1033 +What:          /sys/fs/aufs/si_<id>/xi_path
1034 +Date:          March 2009
1035 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1036 +Description:
1037 +               It shows the abolute path of XINO (External Inode Number
1038 +               Bitmap, Translation Table and Generation Table) file
1039 +               even if it is the default path.
1040 +               When the aufs mount option 'noxino' is specified, it
1041 +               will be empty. About XINO files, see the aufs manual.
1042 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt linux/Documentation/filesystems/aufs/design/01intro.txt
1043 --- /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt  1970-01-01 01:00:00.000000000 +0100
1044 +++ linux/Documentation/filesystems/aufs/design/01intro.txt     2021-12-03 15:38:59.929980643 +0100
1045 @@ -0,0 +1,171 @@
1046 +
1047 +# Copyright (C) 2005-2021 Junjiro R. Okajima
1048 +#
1049 +# This program is free software; you can redistribute it and/or modify
1050 +# it under the terms of the GNU General Public License as published by
1051 +# the Free Software Foundation; either version 2 of the License, or
1052 +# (at your option) any later version.
1053 +#
1054 +# This program is distributed in the hope that it will be useful,
1055 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1056 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1057 +# GNU General Public License for more details.
1058 +#
1059 +# You should have received a copy of the GNU General Public License
1060 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1061 +
1062 +Introduction
1063 +----------------------------------------
1064 +
1065 +aufs [ei ju: ef es] | /ey-yoo-ef-es/ | [a u f s]
1066 +1. abbrev. for "advanced multi-layered unification filesystem".
1067 +2. abbrev. for "another unionfs".
1068 +3. abbrev. for "auf das" in German which means "on the" in English.
1069 +   Ex. "Butter aufs Brot"(G) means "butter onto bread"(E).
1070 +   But "Filesystem aufs Filesystem" is hard to understand.
1071 +4. abbrev. for "African Urban Fashion Show".
1072 +
1073 +AUFS is a filesystem with features:
1074 +- multi layered stackable unification filesystem, the member directory
1075 +  is called as a branch.
1076 +- branch permission and attribute, 'readonly', 'real-readonly',
1077 +  'readwrite', 'whiteout-able', 'link-able whiteout', etc. and their
1078 +  combination.
1079 +- internal "file copy-on-write".
1080 +- logical deletion, whiteout.
1081 +- dynamic branch manipulation, adding, deleting and changing permission.
1082 +- allow bypassing aufs, user's direct branch access.
1083 +- external inode number translation table and bitmap which maintains the
1084 +  persistent aufs inode number.
1085 +- seekable directory, including NFS readdir.
1086 +- file mapping, mmap and sharing pages.
1087 +- pseudo-link, hardlink over branches.
1088 +- loopback mounted filesystem as a branch.
1089 +- several policies to select one among multiple writable branches.
1090 +- revert a single systemcall when an error occurs in aufs.
1091 +- and more...
1092 +
1093 +
1094 +Multi Layered Stackable Unification Filesystem
1095 +----------------------------------------------------------------------
1096 +Most people already knows what it is.
1097 +It is a filesystem which unifies several directories and provides a
1098 +merged single directory. When users access a file, the access will be
1099 +passed/re-directed/converted (sorry, I am not sure which English word is
1100 +correct) to the real file on the member filesystem. The member
1101 +filesystem is called 'lower filesystem' or 'branch' and has a mode
1102 +'readonly' and 'readwrite.' And the deletion for a file on the lower
1103 +readonly branch is handled by creating 'whiteout' on the upper writable
1104 +branch.
1105 +
1106 +On LKML, there have been discussions about UnionMount (Jan Blunck,
1107 +Bharata B Rao and Valerie Aurora) and Unionfs (Erez Zadok). They took
1108 +different approaches to implement the merged-view.
1109 +The former tries putting it into VFS, and the latter implements as a
1110 +separate filesystem.
1111 +(If I misunderstand about these implementations, please let me know and
1112 +I shall correct it. Because it is a long time ago when I read their
1113 +source files last time).
1114 +
1115 +UnionMount's approach will be able to small, but may be hard to share
1116 +branches between several UnionMount since the whiteout in it is
1117 +implemented in the inode on branch filesystem and always
1118 +shared. According to Bharata's post, readdir does not seems to be
1119 +finished yet.
1120 +There are several missing features known in this implementations such as
1121 +- for users, the inode number may change silently. eg. copy-up.
1122 +- link(2) may break by copy-up.
1123 +- read(2) may get an obsoleted filedata (fstat(2) too).
1124 +- fcntl(F_SETLK) may be broken by copy-up.
1125 +- unnecessary copy-up may happen, for example mmap(MAP_PRIVATE) after
1126 +  open(O_RDWR).
1127 +
1128 +In linux-3.18, "overlay" filesystem (formerly known as "overlayfs") was
1129 +merged into mainline. This is another implementation of UnionMount as a
1130 +separated filesystem. All the limitations and known problems which
1131 +UnionMount are equally inherited to "overlay" filesystem.
1132 +
1133 +Unionfs has a longer history. When I started implementing a stackable
1134 +filesystem (Aug 2005), it already existed. It has virtual super_block,
1135 +inode, dentry and file objects and they have an array pointing lower
1136 +same kind objects. After contributing many patches for Unionfs, I
1137 +re-started my project AUFS (Jun 2006).
1138 +
1139 +In AUFS, the structure of filesystem resembles to Unionfs, but I
1140 +implemented my own ideas, approaches and enhancements and it became
1141 +totally different one.
1142 +
1143 +Comparing DM snapshot and fs based implementation
1144 +- the number of bytes to be copied between devices is much smaller.
1145 +- the type of filesystem must be one and only.
1146 +- the fs must be writable, no readonly fs, even for the lower original
1147 +  device. so the compression fs will not be usable. but if we use
1148 +  loopback mount, we may address this issue.
1149 +  for instance,
1150 +       mount /cdrom/squashfs.img /sq
1151 +       losetup /sq/ext2.img
1152 +       losetup /somewhere/cow
1153 +       dmsetup "snapshot /dev/loop0 /dev/loop1 ..."
1154 +- it will be difficult (or needs more operations) to extract the
1155 +  difference between the original device and COW.
1156 +- DM snapshot-merge may help a lot when users try merging. in the
1157 +  fs-layer union, users will use rsync(1).
1158 +
1159 +You may want to read my old paper "Filesystems in LiveCD"
1160 +(http://aufs.sourceforge.net/aufs2/report/sq/sq.pdf).
1161 +
1162 +
1163 +Several characters/aspects/persona of aufs
1164 +----------------------------------------------------------------------
1165 +
1166 +Aufs has several characters, aspects or persona.
1167 +1. a filesystem, callee of VFS helper
1168 +2. sub-VFS, caller of VFS helper for branches
1169 +3. a virtual filesystem which maintains persistent inode number
1170 +4. reader/writer of files on branches such like an application
1171 +
1172 +1. Callee of VFS Helper
1173 +As an ordinary linux filesystem, aufs is a callee of VFS. For instance,
1174 +unlink(2) from an application reaches sys_unlink() kernel function and
1175 +then vfs_unlink() is called. vfs_unlink() is one of VFS helper and it
1176 +calls filesystem specific unlink operation. Actually aufs implements the
1177 +unlink operation but it behaves like a redirector.
1178 +
1179 +2. Caller of VFS Helper for Branches
1180 +aufs_unlink() passes the unlink request to the branch filesystem as if
1181 +it were called from VFS. So the called unlink operation of the branch
1182 +filesystem acts as usual. As a caller of VFS helper, aufs should handle
1183 +every necessary pre/post operation for the branch filesystem.
1184 +- acquire the lock for the parent dir on a branch
1185 +- lookup in a branch
1186 +- revalidate dentry on a branch
1187 +- mnt_want_write() for a branch
1188 +- vfs_unlink() for a branch
1189 +- mnt_drop_write() for a branch
1190 +- release the lock on a branch
1191 +
1192 +3. Persistent Inode Number
1193 +One of the most important issue for a filesystem is to maintain inode
1194 +numbers. This is particularly important to support exporting a
1195 +filesystem via NFS. Aufs is a virtual filesystem which doesn't have a
1196 +backend block device for its own. But some storage is necessary to
1197 +keep and maintain the inode numbers. It may be a large space and may not
1198 +suit to keep in memory. Aufs rents some space from its first writable
1199 +branch filesystem (by default) and creates file(s) on it. These files
1200 +are created by aufs internally and removed soon (currently) keeping
1201 +opened.
1202 +Note: Because these files are removed, they are totally gone after
1203 +      unmounting aufs. It means the inode numbers are not persistent
1204 +      across unmount or reboot. I have a plan to make them really
1205 +      persistent which will be important for aufs on NFS server.
1206 +
1207 +4. Read/Write Files Internally (copy-on-write)
1208 +Because a branch can be readonly, when you write a file on it, aufs will
1209 +"copy-up" it to the upper writable branch internally. And then write the
1210 +originally requested thing to the file. Generally kernel doesn't
1211 +open/read/write file actively. In aufs, even a single write may cause a
1212 +internal "file copy". This behaviour is very similar to cp(1) command.
1213 +
1214 +Some people may think it is better to pass such work to user space
1215 +helper, instead of doing in kernel space. Actually I am still thinking
1216 +about it. But currently I have implemented it in kernel space.
1217 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt linux/Documentation/filesystems/aufs/design/02struct.txt
1218 --- /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt 1970-01-01 01:00:00.000000000 +0100
1219 +++ linux/Documentation/filesystems/aufs/design/02struct.txt    2021-12-03 15:38:59.929980643 +0100
1220 @@ -0,0 +1,258 @@
1221 +
1222 +# Copyright (C) 2005-2021 Junjiro R. Okajima
1223 +#
1224 +# This program is free software; you can redistribute it and/or modify
1225 +# it under the terms of the GNU General Public License as published by
1226 +# the Free Software Foundation; either version 2 of the License, or
1227 +# (at your option) any later version.
1228 +#
1229 +# This program is distributed in the hope that it will be useful,
1230 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1231 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1232 +# GNU General Public License for more details.
1233 +#
1234 +# You should have received a copy of the GNU General Public License
1235 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1236 +
1237 +Basic Aufs Internal Structure
1238 +
1239 +Superblock/Inode/Dentry/File Objects
1240 +----------------------------------------------------------------------
1241 +As like an ordinary filesystem, aufs has its own
1242 +superblock/inode/dentry/file objects. All these objects have a
1243 +dynamically allocated array and store the same kind of pointers to the
1244 +lower filesystem, branch.
1245 +For example, when you build a union with one readwrite branch and one
1246 +readonly, mounted /au, /rw and /ro respectively.
1247 +- /au = /rw + /ro
1248 +- /ro/fileA exists but /rw/fileA
1249 +
1250 +Aufs lookup operation finds /ro/fileA and gets dentry for that. These
1251 +pointers are stored in a aufs dentry. The array in aufs dentry will be,
1252 +- [0] = NULL (because /rw/fileA doesn't exist)
1253 +- [1] = /ro/fileA
1254 +
1255 +This style of an array is essentially same to the aufs
1256 +superblock/inode/dentry/file objects.
1257 +
1258 +Because aufs supports manipulating branches, ie. add/delete/change
1259 +branches dynamically, these objects has its own generation. When
1260 +branches are changed, the generation in aufs superblock is
1261 +incremented. And a generation in other object are compared when it is
1262 +accessed. When a generation in other objects are obsoleted, aufs
1263 +refreshes the internal array.
1264 +
1265 +
1266 +Superblock
1267 +----------------------------------------------------------------------
1268 +Additionally aufs superblock has some data for policies to select one
1269 +among multiple writable branches, XIB files, pseudo-links and kobject.
1270 +See below in detail.
1271 +About the policies which supports copy-down a directory, see
1272 +wbr_policy.txt too.
1273 +
1274 +
1275 +Branch and XINO(External Inode Number Translation Table)
1276 +----------------------------------------------------------------------
1277 +Every branch has its own xino (external inode number translation table)
1278 +file. The xino file is created and unlinked by aufs internally. When two
1279 +members of a union exist on the same filesystem, they share the single
1280 +xino file.
1281 +The struct of a xino file is simple, just a sequence of aufs inode
1282 +numbers which is indexed by the lower inode number.
1283 +In the above sample, assume the inode number of /ro/fileA is i111 and
1284 +aufs assigns the inode number i999 for fileA. Then aufs writes 999 as
1285 +4(8) bytes at 111 * 4(8) bytes offset in the xino file.
1286 +
1287 +When the inode numbers are not contiguous, the xino file will be sparse
1288 +which has a hole in it and doesn't consume as much disk space as it
1289 +might appear. If your branch filesystem consumes disk space for such
1290 +holes, then you should specify 'xino=' option at mounting aufs.
1291 +
1292 +Aufs has a mount option to free the disk blocks for such holes in XINO
1293 +files on tmpfs or ramdisk. But it is not so effective actually. If you
1294 +meet a problem of disk shortage due to XINO files, then you should try
1295 +"tmpfs-ino.patch" (and "vfs-ino.patch" too) in aufs4-standalone.git.
1296 +The patch localizes the assignment inumbers per tmpfs-mount and avoid
1297 +the holes in XINO files.
1298 +
1299 +Also a writable branch has three kinds of "whiteout bases". All these
1300 +are existed when the branch is joined to aufs, and their names are
1301 +whiteout-ed doubly, so that users will never see their names in aufs
1302 +hierarchy.
1303 +1. a regular file which will be hardlinked to all whiteouts.
1304 +2. a directory to store a pseudo-link.
1305 +3. a directory to store an "orphan"-ed file temporary.
1306 +
1307 +1. Whiteout Base
1308 +   When you remove a file on a readonly branch, aufs handles it as a
1309 +   logical deletion and creates a whiteout on the upper writable branch
1310 +   as a hardlink of this file in order not to consume inode on the
1311 +   writable branch.
1312 +2. Pseudo-link Dir
1313 +   See below, Pseudo-link.
1314 +3. Step-Parent Dir
1315 +   When "fileC" exists on the lower readonly branch only and it is
1316 +   opened and removed with its parent dir, and then user writes
1317 +   something into it, then aufs copies-up fileC to this
1318 +   directory. Because there is no other dir to store fileC. After
1319 +   creating a file under this dir, the file is unlinked.
1320 +
1321 +Because aufs supports manipulating branches, ie. add/delete/change
1322 +dynamically, a branch has its own id. When the branch order changes,
1323 +aufs finds the new index by searching the branch id.
1324 +
1325 +
1326 +Pseudo-link
1327 +----------------------------------------------------------------------
1328 +Assume "fileA" exists on the lower readonly branch only and it is
1329 +hardlinked to "fileB" on the branch. When you write something to fileA,
1330 +aufs copies-up it to the upper writable branch. Additionally aufs
1331 +creates a hardlink under the Pseudo-link Directory of the writable
1332 +branch. The inode of a pseudo-link is kept in aufs super_block as a
1333 +simple list. If fileB is read after unlinking fileA, aufs returns
1334 +filedata from the pseudo-link instead of the lower readonly
1335 +branch. Because the pseudo-link is based upon the inode, to keep the
1336 +inode number by xino (see above) is essentially necessary.
1337 +
1338 +All the hardlinks under the Pseudo-link Directory of the writable branch
1339 +should be restored in a proper location later. Aufs provides a utility
1340 +to do this. The userspace helpers executed at remounting and unmounting
1341 +aufs by default.
1342 +During this utility is running, it puts aufs into the pseudo-link
1343 +maintenance mode. In this mode, only the process which began the
1344 +maintenance mode (and its child processes) is allowed to operate in
1345 +aufs. Some other processes which are not related to the pseudo-link will
1346 +be allowed to run too, but the rest have to return an error or wait
1347 +until the maintenance mode ends. If a process already acquires an inode
1348 +mutex (in VFS), it has to return an error.
1349 +
1350 +
1351 +XIB(external inode number bitmap)
1352 +----------------------------------------------------------------------
1353 +Addition to the xino file per a branch, aufs has an external inode number
1354 +bitmap in a superblock object. It is also an internal file such like a
1355 +xino file.
1356 +It is a simple bitmap to mark whether the aufs inode number is in-use or
1357 +not.
1358 +To reduce the file I/O, aufs prepares a single memory page to cache xib.
1359 +
1360 +As well as XINO files, aufs has a feature to truncate/refresh XIB to
1361 +reduce the number of consumed disk blocks for these files.
1362 +
1363 +
1364 +Virtual or Vertical Dir, and Readdir in Userspace
1365 +----------------------------------------------------------------------
1366 +In order to support multiple layers (branches), aufs readdir operation
1367 +constructs a virtual dir block on memory. For readdir, aufs calls
1368 +vfs_readdir() internally for each dir on branches, merges their entries
1369 +with eliminating the whiteout-ed ones, and sets it to file (dir)
1370 +object. So the file object has its entry list until it is closed. The
1371 +entry list will be updated when the file position is zero and becomes
1372 +obsoleted. This decision is made in aufs automatically.
1373 +
1374 +The dynamically allocated memory block for the name of entries has a
1375 +unit of 512 bytes (by default) and stores the names contiguously (no
1376 +padding). Another block for each entry is handled by kmem_cache too.
1377 +During building dir blocks, aufs creates hash list and judging whether
1378 +the entry is whiteouted by its upper branch or already listed.
1379 +The merged result is cached in the corresponding inode object and
1380 +maintained by a customizable life-time option.
1381 +
1382 +Some people may call it can be a security hole or invite DoS attack
1383 +since the opened and once readdir-ed dir (file object) holds its entry
1384 +list and becomes a pressure for system memory. But I'd say it is similar
1385 +to files under /proc or /sys. The virtual files in them also holds a
1386 +memory page (generally) while they are opened. When an idea to reduce
1387 +memory for them is introduced, it will be applied to aufs too.
1388 +For those who really hate this situation, I've developed readdir(3)
1389 +library which operates this merging in userspace. You just need to set
1390 +LD_PRELOAD environment variable, and aufs will not consume no memory in
1391 +kernel space for readdir(3).
1392 +
1393 +
1394 +Workqueue
1395 +----------------------------------------------------------------------
1396 +Aufs sometimes requires privilege access to a branch. For instance,
1397 +in copy-up/down operation. When a user process is going to make changes
1398 +to a file which exists in the lower readonly branch only, and the mode
1399 +of one of ancestor directories may not be writable by a user
1400 +process. Here aufs copy-up the file with its ancestors and they may
1401 +require privilege to set its owner/group/mode/etc.
1402 +This is a typical case of a application character of aufs (see
1403 +Introduction).
1404 +
1405 +Aufs uses workqueue synchronously for this case. It creates its own
1406 +workqueue. The workqueue is a kernel thread and has privilege. Aufs
1407 +passes the request to call mkdir or write (for example), and wait for
1408 +its completion. This approach solves a problem of a signal handler
1409 +simply.
1410 +If aufs didn't adopt the workqueue and changed the privilege of the
1411 +process, then the process may receive the unexpected SIGXFSZ or other
1412 +signals.
1413 +
1414 +Also aufs uses the system global workqueue ("events" kernel thread) too
1415 +for asynchronous tasks, such like handling inotify/fsnotify, re-creating a
1416 +whiteout base and etc. This is unrelated to a privilege.
1417 +Most of aufs operation tries acquiring a rw_semaphore for aufs
1418 +superblock at the beginning, at the same time waits for the completion
1419 +of all queued asynchronous tasks.
1420 +
1421 +
1422 +Whiteout
1423 +----------------------------------------------------------------------
1424 +The whiteout in aufs is very similar to Unionfs's. That is represented
1425 +by its filename. UnionMount takes an approach of a file mode, but I am
1426 +afraid several utilities (find(1) or something) will have to support it.
1427 +
1428 +Basically the whiteout represents "logical deletion" which stops aufs to
1429 +lookup further, but also it represents "dir is opaque" which also stop
1430 +further lookup.
1431 +
1432 +In aufs, rmdir(2) and rename(2) for dir uses whiteout alternatively.
1433 +In order to make several functions in a single systemcall to be
1434 +revertible, aufs adopts an approach to rename a directory to a temporary
1435 +unique whiteouted name.
1436 +For example, in rename(2) dir where the target dir already existed, aufs
1437 +renames the target dir to a temporary unique whiteouted name before the
1438 +actual rename on a branch, and then handles other actions (make it opaque,
1439 +update the attributes, etc). If an error happens in these actions, aufs
1440 +simply renames the whiteouted name back and returns an error. If all are
1441 +succeeded, aufs registers a function to remove the whiteouted unique
1442 +temporary name completely and asynchronously to the system global
1443 +workqueue.
1444 +
1445 +
1446 +Copy-up
1447 +----------------------------------------------------------------------
1448 +It is a well-known feature or concept.
1449 +When user modifies a file on a readonly branch, aufs operate "copy-up"
1450 +internally and makes change to the new file on the upper writable branch.
1451 +When the trigger systemcall does not update the timestamps of the parent
1452 +dir, aufs reverts it after copy-up.
1453 +
1454 +
1455 +Move-down (aufs3.9 and later)
1456 +----------------------------------------------------------------------
1457 +"Copy-up" is one of the essential feature in aufs. It copies a file from
1458 +the lower readonly branch to the upper writable branch when a user
1459 +changes something about the file.
1460 +"Move-down" is an opposite action of copy-up. Basically this action is
1461 +ran manually instead of automatically and internally.
1462 +For desgin and implementation, aufs has to consider these issues.
1463 +- whiteout for the file may exist on the lower branch.
1464 +- ancestor directories may not exist on the lower branch.
1465 +- diropq for the ancestor directories may exist on the upper branch.
1466 +- free space on the lower branch will reduce.
1467 +- another access to the file may happen during moving-down, including
1468 +  UDBA (see "Revalidate Dentry and UDBA").
1469 +- the file should not be hard-linked nor pseudo-linked. they should be
1470 +  handled by auplink utility later.
1471 +
1472 +Sometimes users want to move-down a file from the upper writable branch
1473 +to the lower readonly or writable branch. For instance,
1474 +- the free space of the upper writable branch is going to run out.
1475 +- create a new intermediate branch between the upper and lower branch.
1476 +- etc.
1477 +
1478 +For this purpose, use "aumvdown" command in aufs-util.git.
1479 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03atomic_open.txt linux/Documentation/filesystems/aufs/design/03atomic_open.txt
1480 --- /usr/share/empty/Documentation/filesystems/aufs/design/03atomic_open.txt    1970-01-01 01:00:00.000000000 +0100
1481 +++ linux/Documentation/filesystems/aufs/design/03atomic_open.txt       2021-12-03 15:38:59.929980643 +0100
1482 @@ -0,0 +1,85 @@
1483 +
1484 +# Copyright (C) 2015-2021 Junjiro R. Okajima
1485 +#
1486 +# This program is free software; you can redistribute it and/or modify
1487 +# it under the terms of the GNU General Public License as published by
1488 +# the Free Software Foundation; either version 2 of the License, or
1489 +# (at your option) any later version.
1490 +#
1491 +# This program is distributed in the hope that it will be useful,
1492 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1493 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1494 +# GNU General Public License for more details.
1495 +#
1496 +# You should have received a copy of the GNU General Public License
1497 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1498 +
1499 +Support for a branch who has its ->atomic_open()
1500 +----------------------------------------------------------------------
1501 +The filesystems who implement its ->atomic_open() are not majority. For
1502 +example NFSv4 does, and aufs should call NFSv4 ->atomic_open,
1503 +particularly for open(O_CREAT|O_EXCL, 0400) case. Other than
1504 +->atomic_open(), NFSv4 returns an error for this open(2). While I am not
1505 +sure whether all filesystems who have ->atomic_open() behave like this,
1506 +but NFSv4 surely returns the error.
1507 +
1508 +In order to support ->atomic_open() for aufs, there are a few
1509 +approaches.
1510 +
1511 +A. Introduce aufs_atomic_open()
1512 +   - calls one of VFS:do_last(), lookup_open() or atomic_open() for
1513 +     branch fs.
1514 +B. Introduce aufs_atomic_open() calling create, open and chmod. this is
1515 +   an aufs user Pip Cet's approach
1516 +   - calls aufs_create(), VFS finish_open() and notify_change().
1517 +   - pass fake-mode to finish_open(), and then correct the mode by
1518 +     notify_change().
1519 +C. Extend aufs_open() to call branch fs's ->atomic_open()
1520 +   - no aufs_atomic_open().
1521 +   - aufs_lookup() registers the TID to an aufs internal object.
1522 +   - aufs_create() does nothing when the matching TID is registered, but
1523 +     registers the mode.
1524 +   - aufs_open() calls branch fs's ->atomic_open() when the matching
1525 +     TID is registered.
1526 +D. Extend aufs_open() to re-try branch fs's ->open() with superuser's
1527 +   credential
1528 +   - no aufs_atomic_open().
1529 +   - aufs_create() registers the TID to an internal object. this info
1530 +     represents "this process created this file just now."
1531 +   - when aufs gets EACCES from branch fs's ->open(), then confirm the
1532 +     registered TID and re-try open() with superuser's credential.
1533 +
1534 +Pros and cons for each approach.
1535 +
1536 +A.
1537 +   - straightforward but highly depends upon VFS internal.
1538 +   - the atomic behavaiour is kept.
1539 +   - some of parameters such as nameidata are hard to reproduce for
1540 +     branch fs.
1541 +   - large overhead.
1542 +B.
1543 +   - easy to implement.
1544 +   - the atomic behavaiour is lost.
1545 +C.
1546 +   - the atomic behavaiour is kept.
1547 +   - dirty and tricky.
1548 +   - VFS checks whether the file is created correctly after calling
1549 +     ->create(), which means this approach doesn't work.
1550 +D.
1551 +   - easy to implement.
1552 +   - the atomic behavaiour is lost.
1553 +   - to open a file with superuser's credential and give it to a user
1554 +     process is a bad idea, since the file object keeps the credential
1555 +     in it. It may affect LSM or something. This approach doesn't work
1556 +     either.
1557 +
1558 +The approach A is ideal, but it hard to implement. So here is a
1559 +variation of A, which is to be implemented.
1560 +
1561 +A-1. Introduce aufs_atomic_open()
1562 +     - calls branch fs ->atomic_open() if exists. otherwise calls
1563 +       vfs_create() and finish_open().
1564 +     - the demerit is that the several checks after branch fs
1565 +       ->atomic_open() are lost. in the ordinary case, the checks are
1566 +       done by VFS:do_last(), lookup_open() and atomic_open(). some can
1567 +       be implemented in aufs, but not all I am afraid.
1568 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt linux/Documentation/filesystems/aufs/design/03lookup.txt
1569 --- /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt 1970-01-01 01:00:00.000000000 +0100
1570 +++ linux/Documentation/filesystems/aufs/design/03lookup.txt    2021-12-03 15:38:59.929980643 +0100
1571 @@ -0,0 +1,113 @@
1572 +
1573 +# Copyright (C) 2005-2021 Junjiro R. Okajima
1574 +#
1575 +# This program is free software; you can redistribute it and/or modify
1576 +# it under the terms of the GNU General Public License as published by
1577 +# the Free Software Foundation; either version 2 of the License, or
1578 +# (at your option) any later version.
1579 +#
1580 +# This program is distributed in the hope that it will be useful,
1581 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1582 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1583 +# GNU General Public License for more details.
1584 +#
1585 +# You should have received a copy of the GNU General Public License
1586 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1587 +
1588 +Lookup in a Branch
1589 +----------------------------------------------------------------------
1590 +Since aufs has a character of sub-VFS (see Introduction), it operates
1591 +lookup for branches as VFS does. It may be a heavy work. But almost all
1592 +lookup operation in aufs is the simplest case, ie. lookup only an entry
1593 +directly connected to its parent. Digging down the directory hierarchy
1594 +is unnecessary. VFS has a function lookup_one_len() for that use, and
1595 +aufs calls it.
1596 +
1597 +When a branch is a remote filesystem, aufs basically relies upon its
1598 +->d_revalidate(), also aufs forces the hardest revalidate tests for
1599 +them.
1600 +For d_revalidate, aufs implements three levels of revalidate tests. See
1601 +"Revalidate Dentry and UDBA" in detail.
1602 +
1603 +
1604 +Test Only the Highest One for the Directory Permission (dirperm1 option)
1605 +----------------------------------------------------------------------
1606 +Let's try case study.
1607 +- aufs has two branches, upper readwrite and lower readonly.
1608 +  /au = /rw + /ro
1609 +- "dirA" exists under /ro, but /rw. and its mode is 0700.
1610 +- user invoked "chmod a+rx /au/dirA"
1611 +- the internal copy-up is activated and "/rw/dirA" is created and its
1612 +  permission bits are set to world readable.
1613 +- then "/au/dirA" becomes world readable?
1614 +
1615 +In this case, /ro/dirA is still 0700 since it exists in readonly branch,
1616 +or it may be a natively readonly filesystem. If aufs respects the lower
1617 +branch, it should not respond readdir request from other users. But user
1618 +allowed it by chmod. Should really aufs rejects showing the entries
1619 +under /ro/dirA?
1620 +
1621 +To be honest, I don't have a good solution for this case. So aufs
1622 +implements 'dirperm1' and 'nodirperm1' mount options, and leave it to
1623 +users.
1624 +When dirperm1 is specified, aufs checks only the highest one for the
1625 +directory permission, and shows the entries. Otherwise, as usual, checks
1626 +every dir existing on all branches and rejects the request.
1627 +
1628 +As a side effect, dirperm1 option improves the performance of aufs
1629 +because the number of permission check is reduced when the number of
1630 +branch is many.
1631 +
1632 +
1633 +Revalidate Dentry and UDBA (User's Direct Branch Access)
1634 +----------------------------------------------------------------------
1635 +Generally VFS helpers re-validate a dentry as a part of lookup.
1636 +0. digging down the directory hierarchy.
1637 +1. lock the parent dir by its i_mutex.
1638 +2. lookup the final (child) entry.
1639 +3. revalidate it.
1640 +4. call the actual operation (create, unlink, etc.)
1641 +5. unlock the parent dir
1642 +
1643 +If the filesystem implements its ->d_revalidate() (step 3), then it is
1644 +called. Actually aufs implements it and checks the dentry on a branch is
1645 +still valid.
1646 +But it is not enough. Because aufs has to release the lock for the
1647 +parent dir on a branch at the end of ->lookup() (step 2) and
1648 +->d_revalidate() (step 3) while the i_mutex of the aufs dir is still
1649 +held by VFS.
1650 +If the file on a branch is changed directly, eg. bypassing aufs, after
1651 +aufs released the lock, then the subsequent operation may cause
1652 +something unpleasant result.
1653 +
1654 +This situation is a result of VFS architecture, ->lookup() and
1655 +->d_revalidate() is separated. But I never say it is wrong. It is a good
1656 +design from VFS's point of view. It is just not suitable for sub-VFS
1657 +character in aufs.
1658 +
1659 +Aufs supports such case by three level of revalidation which is
1660 +selectable by user.
1661 +1. Simple Revalidate
1662 +   Addition to the native flow in VFS's, confirm the child-parent
1663 +   relationship on the branch just after locking the parent dir on the
1664 +   branch in the "actual operation" (step 4). When this validation
1665 +   fails, aufs returns EBUSY. ->d_revalidate() (step 3) in aufs still
1666 +   checks the validation of the dentry on branches.
1667 +2. Monitor Changes Internally by Inotify/Fsnotify
1668 +   Addition to above, in the "actual operation" (step 4) aufs re-lookup
1669 +   the dentry on the branch, and returns EBUSY if it finds different
1670 +   dentry.
1671 +   Additionally, aufs sets the inotify/fsnotify watch for every dir on branches
1672 +   during it is in cache. When the event is notified, aufs registers a
1673 +   function to kernel 'events' thread by schedule_work(). And the
1674 +   function sets some special status to the cached aufs dentry and inode
1675 +   private data. If they are not cached, then aufs has nothing to
1676 +   do. When the same file is accessed through aufs (step 0-3) later,
1677 +   aufs will detect the status and refresh all necessary data.
1678 +   In this mode, aufs has to ignore the event which is fired by aufs
1679 +   itself.
1680 +3. No Extra Validation
1681 +   This is the simplest test and doesn't add any additional revalidation
1682 +   test, and skip the revalidation in step 4. It is useful and improves
1683 +   aufs performance when system surely hide the aufs branches from user,
1684 +   by over-mounting something (or another method).
1685 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt linux/Documentation/filesystems/aufs/design/04branch.txt
1686 --- /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt 1970-01-01 01:00:00.000000000 +0100
1687 +++ linux/Documentation/filesystems/aufs/design/04branch.txt    2021-12-03 15:38:59.929980643 +0100
1688 @@ -0,0 +1,74 @@
1689 +
1690 +# Copyright (C) 2005-2021 Junjiro R. Okajima
1691 +#
1692 +# This program is free software; you can redistribute it and/or modify
1693 +# it under the terms of the GNU General Public License as published by
1694 +# the Free Software Foundation; either version 2 of the License, or
1695 +# (at your option) any later version.
1696 +#
1697 +# This program is distributed in the hope that it will be useful,
1698 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1699 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1700 +# GNU General Public License for more details.
1701 +#
1702 +# You should have received a copy of the GNU General Public License
1703 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1704 +
1705 +Branch Manipulation
1706 +
1707 +Since aufs supports dynamic branch manipulation, ie. add/remove a branch
1708 +and changing its permission/attribute, there are a lot of works to do.
1709 +
1710 +
1711 +Add a Branch
1712 +----------------------------------------------------------------------
1713 +o Confirm the adding dir exists outside of aufs, including loopback
1714 +  mount, and its various attributes.
1715 +o Initialize the xino file and whiteout bases if necessary.
1716 +  See struct.txt.
1717 +
1718 +o Check the owner/group/mode of the directory
1719 +  When the owner/group/mode of the adding directory differs from the
1720 +  existing branch, aufs issues a warning because it may impose a
1721 +  security risk.
1722 +  For example, when a upper writable branch has a world writable empty
1723 +  top directory, a malicious user can create any files on the writable
1724 +  branch directly, like copy-up and modify manually. If something like
1725 +  /etc/{passwd,shadow} exists on the lower readonly branch but the upper
1726 +  writable branch, and the writable branch is world-writable, then a
1727 +  malicious guy may create /etc/passwd on the writable branch directly
1728 +  and the infected file will be valid in aufs.
1729 +  I am afraid it can be a security issue, but aufs can do nothing except
1730 +  producing a warning.
1731 +
1732 +
1733 +Delete a Branch
1734 +----------------------------------------------------------------------
1735 +o Confirm the deleting branch is not busy
1736 +  To be general, there is one merit to adopt "remount" interface to
1737 +  manipulate branches. It is to discard caches. At deleting a branch,
1738 +  aufs checks the still cached (and connected) dentries and inodes. If
1739 +  there are any, then they are all in-use. An inode without its
1740 +  corresponding dentry can be alive alone (for example, inotify/fsnotify case).
1741 +
1742 +  For the cached one, aufs checks whether the same named entry exists on
1743 +  other branches.
1744 +  If the cached one is a directory, because aufs provides a merged view
1745 +  to users, as long as one dir is left on any branch aufs can show the
1746 +  dir to users. In this case, the branch can be removed from aufs.
1747 +  Otherwise aufs rejects deleting the branch.
1748 +
1749 +  If any file on the deleting branch is opened by aufs, then aufs
1750 +  rejects deleting.
1751 +
1752 +
1753 +Modify the Permission of a Branch
1754 +----------------------------------------------------------------------
1755 +o Re-initialize or remove the xino file and whiteout bases if necessary.
1756 +  See struct.txt.
1757 +
1758 +o rw --> ro: Confirm the modifying branch is not busy
1759 +  Aufs rejects the request if any of these conditions are true.
1760 +  - a file on the branch is mmap-ed.
1761 +  - a regular file on the branch is opened for write and there is no
1762 +    same named entry on the upper branch.
1763 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt linux/Documentation/filesystems/aufs/design/05wbr_policy.txt
1764 --- /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt     1970-01-01 01:00:00.000000000 +0100
1765 +++ linux/Documentation/filesystems/aufs/design/05wbr_policy.txt        2021-12-03 15:38:59.929980643 +0100
1766 @@ -0,0 +1,64 @@
1767 +
1768 +# Copyright (C) 2005-2021 Junjiro R. Okajima
1769 +#
1770 +# This program is free software; you can redistribute it and/or modify
1771 +# it under the terms of the GNU General Public License as published by
1772 +# the Free Software Foundation; either version 2 of the License, or
1773 +# (at your option) any later version.
1774 +#
1775 +# This program is distributed in the hope that it will be useful,
1776 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1777 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1778 +# GNU General Public License for more details.
1779 +#
1780 +# You should have received a copy of the GNU General Public License
1781 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1782 +
1783 +Policies to Select One among Multiple Writable Branches
1784 +----------------------------------------------------------------------
1785 +When the number of writable branch is more than one, aufs has to decide
1786 +the target branch for file creation or copy-up. By default, the highest
1787 +writable branch which has the parent (or ancestor) dir of the target
1788 +file is chosen (top-down-parent policy).
1789 +By user's request, aufs implements some other policies to select the
1790 +writable branch, for file creation several policies, round-robin,
1791 +most-free-space, and other policies. For copy-up, top-down-parent,
1792 +bottom-up-parent, bottom-up and others.
1793 +
1794 +As expected, the round-robin policy selects the branch in circular. When
1795 +you have two writable branches and creates 10 new files, 5 files will be
1796 +created for each branch. mkdir(2) systemcall is an exception. When you
1797 +create 10 new directories, all will be created on the same branch.
1798 +And the most-free-space policy selects the one which has most free
1799 +space among the writable branches. The amount of free space will be
1800 +checked by aufs internally, and users can specify its time interval.
1801 +
1802 +The policies for copy-up is more simple,
1803 +top-down-parent is equivalent to the same named on in create policy,
1804 +bottom-up-parent selects the writable branch where the parent dir
1805 +exists and the nearest upper one from the copyup-source,
1806 +bottom-up selects the nearest upper writable branch from the
1807 +copyup-source, regardless the existence of the parent dir.
1808 +
1809 +There are some rules or exceptions to apply these policies.
1810 +- If there is a readonly branch above the policy-selected branch and
1811 +  the parent dir is marked as opaque (a variation of whiteout), or the
1812 +  target (creating) file is whiteout-ed on the upper readonly branch,
1813 +  then the result of the policy is ignored and the target file will be
1814 +  created on the nearest upper writable branch than the readonly branch.
1815 +- If there is a writable branch above the policy-selected branch and
1816 +  the parent dir is marked as opaque or the target file is whiteouted
1817 +  on the branch, then the result of the policy is ignored and the target
1818 +  file will be created on the highest one among the upper writable
1819 +  branches who has diropq or whiteout. In case of whiteout, aufs removes
1820 +  it as usual.
1821 +- link(2) and rename(2) systemcalls are exceptions in every policy.
1822 +  They try selecting the branch where the source exists as possible
1823 +  since copyup a large file will take long time. If it can't be,
1824 +  ie. the branch where the source exists is readonly, then they will
1825 +  follow the copyup policy.
1826 +- There is an exception for rename(2) when the target exists.
1827 +  If the rename target exists, aufs compares the index of the branches
1828 +  where the source and the target exists and selects the higher
1829 +  one. If the selected branch is readonly, then aufs follows the
1830 +  copyup policy.
1831 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06dirren.dot linux/Documentation/filesystems/aufs/design/06dirren.dot
1832 --- /usr/share/empty/Documentation/filesystems/aufs/design/06dirren.dot 1970-01-01 01:00:00.000000000 +0100
1833 +++ linux/Documentation/filesystems/aufs/design/06dirren.dot    2021-12-03 15:38:59.929980643 +0100
1834 @@ -0,0 +1,31 @@
1835 +
1836 +// to view this graph, run dot(1) command in GRAPHVIZ.
1837 +
1838 +digraph G {
1839 +node [shape=box];
1840 +whinfo [label="detailed info file\n(lower_brid_root-hinum, h_inum, namelen, old name)"];
1841 +
1842 +node [shape=oval];
1843 +
1844 +aufs_rename -> whinfo [label="store/remove"];
1845 +
1846 +node [shape=oval];
1847 +inode_list [label="h_inum list in branch\ncache"];
1848 +
1849 +node [shape=box];
1850 +whinode [label="h_inum list file"];
1851 +
1852 +node [shape=oval];
1853 +brmgmt [label="br_add/del/mod/umount"];
1854 +
1855 +brmgmt -> inode_list [label="create/remove"];
1856 +brmgmt -> whinode [label="load/store"];
1857 +
1858 +inode_list -> whinode [style=dashed,dir=both];
1859 +
1860 +aufs_rename -> inode_list [label="add/del"];
1861 +
1862 +aufs_lookup -> inode_list [label="search"];
1863 +
1864 +aufs_lookup -> whinfo [label="load/remove"];
1865 +}
1866 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06dirren.txt linux/Documentation/filesystems/aufs/design/06dirren.txt
1867 --- /usr/share/empty/Documentation/filesystems/aufs/design/06dirren.txt 1970-01-01 01:00:00.000000000 +0100
1868 +++ linux/Documentation/filesystems/aufs/design/06dirren.txt    2021-12-03 15:38:59.929980643 +0100
1869 @@ -0,0 +1,102 @@
1870 +
1871 +# Copyright (C) 2017-2021 Junjiro R. Okajima
1872 +#
1873 +# This program is free software; you can redistribute it and/or modify
1874 +# it under the terms of the GNU General Public License as published by
1875 +# the Free Software Foundation; either version 2 of the License, or
1876 +# (at your option) any later version.
1877 +#
1878 +# This program is distributed in the hope that it will be useful,
1879 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1880 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1881 +# GNU General Public License for more details.
1882 +#
1883 +# You should have received a copy of the GNU General Public License
1884 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1885 +
1886 +Special handling for renaming a directory (DIRREN)
1887 +----------------------------------------------------------------------
1888 +First, let's assume we have a simple usecase.
1889 +
1890 +- /u = /rw + /ro
1891 +- /rw/dirA exists
1892 +- /ro/dirA and /ro/dirA/file exist too
1893 +- there is no dirB on both branches
1894 +- a user issues rename("dirA", "dirB")
1895 +
1896 +Now, what should aufs behave against this rename(2)?
1897 +There are a few possible cases.
1898 +
1899 +A. returns EROFS.
1900 +   since dirA exists on a readonly branch which cannot be renamed.
1901 +B. returns EXDEV.
1902 +   it is possible to copy-up dirA (only the dir itself), but the child
1903 +   entries ("file" in this case) should not be. it must be a bad
1904 +   approach to copy-up recursively.
1905 +C. returns a success.
1906 +   even the branch /ro is readonly, aufs tries renaming it. Obviously it
1907 +   is a violation of aufs' policy.
1908 +D. construct an extra information which indicates that /ro/dirA should
1909 +   be handled as the name of dirB.
1910 +   overlayfs has a similar feature called REDIRECT.
1911 +
1912 +Until now, aufs implements the case B only which returns EXDEV, and
1913 +expects the userspace application behaves like mv(1) which tries
1914 +issueing rename(2) recursively.
1915 +
1916 +A new aufs feature called DIRREN is introduced which implements the case
1917 +D. There are several "extra information" added.
1918 +
1919 +1. detailed info per renamed directory
1920 +   path: /rw/dirB/$AUFS_WH_DR_INFO_PFX.<lower branch-id>
1921 +2. the inode-number list of directories on a branch
1922 +   path: /rw/dirB/$AUFS_WH_DR_BRHINO
1923 +
1924 +The filename of "detailed info per directory" represents the lower
1925 +branch, and its format is
1926 +- a type of the branch id
1927 +  one of these.
1928 +  + uuid (not implemented yet)
1929 +  + fsid
1930 +  + dev
1931 +- the inode-number of the branch root dir
1932 +
1933 +And it contains these info in a single regular file.
1934 +- magic number
1935 +- branch's inode-number of the logically renamed dir
1936 +- the name of the before-renamed dir
1937 +
1938 +The "detailed info per directory" file is created in aufs rename(2), and
1939 +loaded in any lookup.
1940 +The info is considered in lookup for the matching case only. Here
1941 +"matching" means that the root of branch (in the info filename) is same
1942 +to the current looking-up branch. After looking-up the before-renamed
1943 +name, the inode-number is compared. And the matched dentry is used.
1944 +
1945 +The "inode-number list of directories" is a regular file which contains
1946 +simply the inode-numbers on the branch. The file is created or updated
1947 +in removing the branch, and loaded in adding the branch. Its lifetime is
1948 +equal to the branch.
1949 +The list is refered in lookup, and when the current target inode is
1950 +found in the list, the aufs tries loading the "detailed info per
1951 +directory" and get the changed and valid name of the dir.
1952 +
1953 +Theoretically these "extra informaiton" may be able to be put into XATTR
1954 +in the dir inode. But aufs doesn't choose this way because
1955 +1. XATTR may not be supported by the branch (or its configuration)
1956 +2. XATTR may have its size limit.
1957 +3. XATTR may be less easy to convert than a regular file, when the
1958 +   format of the info is changed in the future.
1959 +At the same time, I agree that the regular file approach is much slower
1960 +than XATTR approach. So, in the future, aufs may take the XATTR or other
1961 +better approach.
1962 +
1963 +This DIRREN feature is enabled by aufs configuration, and is activated
1964 +by a new mount option.
1965 +
1966 +For the more complicated case, there is a work with UDBA option, which
1967 +is to dected the direct access to the branches (by-passing aufs) and to
1968 +maintain the cashes in aufs. Since a single cached aufs dentry may
1969 +contains two names, before- and after-rename, the name comparision in
1970 +UDBA handler may not work correctly. In this case, the behaviour will be
1971 +equivalen to udba=reval case.
1972 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt linux/Documentation/filesystems/aufs/design/06fhsm.txt
1973 --- /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt   1970-01-01 01:00:00.000000000 +0100
1974 +++ linux/Documentation/filesystems/aufs/design/06fhsm.txt      2021-12-03 15:38:59.929980643 +0100
1975 @@ -0,0 +1,120 @@
1976 +
1977 +# Copyright (C) 2011-2021 Junjiro R. Okajima
1978 +#
1979 +# This program is free software; you can redistribute it and/or modify
1980 +# it under the terms of the GNU General Public License as published by
1981 +# the Free Software Foundation; either version 2 of the License, or
1982 +# (at your option) any later version.
1983 +#
1984 +# This program is distributed in the hope that it will be useful,
1985 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1986 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1987 +# GNU General Public License for more details.
1988 +#
1989 +# You should have received a copy of the GNU General Public License
1990 +# along with this program; if not, write to the Free Software
1991 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
1992 +
1993 +
1994 +File-based Hierarchical Storage Management (FHSM)
1995 +----------------------------------------------------------------------
1996 +Hierarchical Storage Management (or HSM) is a well-known feature in the
1997 +storage world. Aufs provides this feature as file-based with multiple
1998 +writable branches, based upon the principle of "Colder, the Lower".
1999 +Here the word "colder" means that the less used files, and "lower" means
2000 +that the position in the order of the stacked branches vertically.
2001 +These multiple writable branches are prioritized, ie. the topmost one
2002 +should be the fastest drive and be used heavily.
2003 +
2004 +o Characters in aufs FHSM story
2005 +- aufs itself and a new branch attribute.
2006 +- a new ioctl interface to move-down and to establish a connection with
2007 +  the daemon ("move-down" is a converse of "copy-up").
2008 +- userspace tool and daemon.
2009 +
2010 +The userspace daemon establishes a connection with aufs and waits for
2011 +the notification. The notified information is very similar to struct
2012 +statfs containing the number of consumed blocks and inodes.
2013 +When the consumed blocks/inodes of a branch exceeds the user-specified
2014 +upper watermark, the daemon activates its move-down process until the
2015 +consumed blocks/inodes reaches the user-specified lower watermark.
2016 +
2017 +The actual move-down is done by aufs based upon the request from
2018 +user-space since we need to maintain the inode number and the internal
2019 +pointer arrays in aufs.
2020 +
2021 +Currently aufs FHSM handles the regular files only. Additionally they
2022 +must not be hard-linked nor pseudo-linked.
2023 +
2024 +
2025 +o Cowork of aufs and the user-space daemon
2026 +  During the userspace daemon established the connection, aufs sends a
2027 +  small notification to it whenever aufs writes something into the
2028 +  writable branch. But it may cost high since aufs issues statfs(2)
2029 +  internally. So user can specify a new option to cache the
2030 +  info. Actually the notification is controlled by these factors.
2031 +  + the specified cache time.
2032 +  + classified as "force" by aufs internally.
2033 +  Until the specified time expires, aufs doesn't send the info
2034 +  except the forced cases. When aufs decide forcing, the info is always
2035 +  notified to userspace.
2036 +  For example, the number of free inodes is generally large enough and
2037 +  the shortage of it happens rarely. So aufs doesn't force the
2038 +  notification when creating a new file, directory and others. This is
2039 +  the typical case which aufs doesn't force.
2040 +  When aufs writes the actual filedata and the files consumes any of new
2041 +  blocks, the aufs forces notifying.
2042 +
2043 +
2044 +o Interfaces in aufs
2045 +- New branch attribute.
2046 +  + fhsm
2047 +    Specifies that the branch is managed by FHSM feature. In other word,
2048 +    participant in the FHSM.
2049 +    When nofhsm is set to the branch, it will not be the source/target
2050 +    branch of the move-down operation. This attribute is set
2051 +    independently from coo and moo attributes, and if you want full
2052 +    FHSM, you should specify them as well.
2053 +- New mount option.
2054 +  + fhsm_sec
2055 +    Specifies a second to suppress many less important info to be
2056 +    notified.
2057 +- New ioctl.
2058 +  + AUFS_CTL_FHSM_FD
2059 +    create a new file descriptor which userspace can read the notification
2060 +    (a subset of struct statfs) from aufs.
2061 +- Module parameter 'brs'
2062 +  It has to be set to 1. Otherwise the new mount option 'fhsm' will not
2063 +  be set.
2064 +- mount helpers /sbin/mount.aufs and /sbin/umount.aufs
2065 +  When there are two or more branches with fhsm attributes,
2066 +  /sbin/mount.aufs invokes the user-space daemon and /sbin/umount.aufs
2067 +  terminates it. As a result of remounting and branch-manipulation, the
2068 +  number of branches with fhsm attribute can be one. In this case,
2069 +  /sbin/mount.aufs will terminate the user-space daemon.
2070 +
2071 +
2072 +Finally the operation is done as these steps in kernel-space.
2073 +- make sure that,
2074 +  + no one else is using the file.
2075 +  + the file is not hard-linked.
2076 +  + the file is not pseudo-linked.
2077 +  + the file is a regular file.
2078 +  + the parent dir is not opaqued.
2079 +- find the target writable branch.
2080 +- make sure the file is not whiteout-ed by the upper (than the target)
2081 +  branch.
2082 +- make the parent dir on the target branch.
2083 +- mutex lock the inode on the branch.
2084 +- unlink the whiteout on the target branch (if exists).
2085 +- lookup and create the whiteout-ed temporary name on the target branch.
2086 +- copy the file as the whiteout-ed temporary name on the target branch.
2087 +- rename the whiteout-ed temporary name to the original name.
2088 +- unlink the file on the source branch.
2089 +- maintain the internal pointer array and the external inode number
2090 +  table (XINO).
2091 +- maintain the timestamps and other attributes of the parent dir and the
2092 +  file.
2093 +
2094 +And of course, in every step, an error may happen. So the operation
2095 +should restore the original file state after an error happens.
2096 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt linux/Documentation/filesystems/aufs/design/06mmap.txt
2097 --- /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt   1970-01-01 01:00:00.000000000 +0100
2098 +++ linux/Documentation/filesystems/aufs/design/06mmap.txt      2021-12-03 15:38:59.929980643 +0100
2099 @@ -0,0 +1,72 @@
2100 +
2101 +# Copyright (C) 2005-2021 Junjiro R. Okajima
2102 +#
2103 +# This program is free software; you can redistribute it and/or modify
2104 +# it under the terms of the GNU General Public License as published by
2105 +# the Free Software Foundation; either version 2 of the License, or
2106 +# (at your option) any later version.
2107 +#
2108 +# This program is distributed in the hope that it will be useful,
2109 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2110 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2111 +# GNU General Public License for more details.
2112 +#
2113 +# You should have received a copy of the GNU General Public License
2114 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2115 +
2116 +mmap(2) -- File Memory Mapping
2117 +----------------------------------------------------------------------
2118 +In aufs, the file-mapped pages are handled by a branch fs directly, no
2119 +interaction with aufs. It means aufs_mmap() calls the branch fs's
2120 +->mmap().
2121 +This approach is simple and good, but there is one problem.
2122 +Under /proc, several entries show the mmapped files by its path (with
2123 +device and inode number), and the printed path will be the path on the
2124 +branch fs's instead of virtual aufs's.
2125 +This is not a problem in most cases, but some utilities lsof(1) (and its
2126 +user) may expect the path on aufs.
2127 +
2128 +To address this issue, aufs adds a new member called vm_prfile in struct
2129 +vm_area_struct (and struct vm_region). The original vm_file points to
2130 +the file on the branch fs in order to handle everything correctly as
2131 +usual. The new vm_prfile points to a virtual file in aufs, and the
2132 +show-functions in procfs refers to vm_prfile if it is set.
2133 +Also we need to maintain several other places where touching vm_file
2134 +such like
2135 +- fork()/clone() copies vma and the reference count of vm_file is
2136 +  incremented.
2137 +- merging vma maintains the ref count too.
2138 +
2139 +This is not a good approach. It just fakes the printed path. But it
2140 +leaves all behaviour around f_mapping unchanged. This is surely an
2141 +advantage.
2142 +Actually aufs had adopted another complicated approach which calls
2143 +generic_file_mmap() and handles struct vm_operations_struct. In this
2144 +approach, aufs met a hard problem and I could not solve it without
2145 +switching the approach.
2146 +
2147 +There may be one more another approach which is
2148 +- bind-mount the branch-root onto the aufs-root internally
2149 +- grab the new vfsmount (ie. struct mount)
2150 +- lazy-umount the branch-root internally
2151 +- in open(2) the aufs-file, open the branch-file with the hidden
2152 +  vfsmount (instead of the original branch's vfsmount)
2153 +- ideally this "bind-mount and lazy-umount" should be done atomically,
2154 +  but it may be possible from userspace by the mount helper.
2155 +
2156 +Adding the internal hidden vfsmount and using it in opening a file, the
2157 +file path under /proc will be printed correctly. This approach looks
2158 +smarter, but is not possible I am afraid.
2159 +- aufs-root may be bind-mount later. when it happens, another hidden
2160 +  vfsmount will be required.
2161 +- it is hard to get the chance to bind-mount and lazy-umount
2162 +  + in kernel-space, FS can have vfsmount in open(2) via
2163 +    file->f_path, and aufs can know its vfsmount. But several locks are
2164 +    already acquired, and if aufs tries to bind-mount and lazy-umount
2165 +    here, then it may cause a deadlock.
2166 +  + in user-space, bind-mount doesn't invoke the mount helper.
2167 +- since /proc shows dev and ino, aufs has to give vma these info. it
2168 +  means a new member vm_prinode will be necessary. this is essentially
2169 +  equivalent to vm_prfile described above.
2170 +
2171 +I have to give up this "looks-smater" approach.
2172 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt linux/Documentation/filesystems/aufs/design/06xattr.txt
2173 --- /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt  1970-01-01 01:00:00.000000000 +0100
2174 +++ linux/Documentation/filesystems/aufs/design/06xattr.txt     2021-12-03 15:38:59.929980643 +0100
2175 @@ -0,0 +1,96 @@
2176 +
2177 +# Copyright (C) 2014-2021 Junjiro R. Okajima
2178 +#
2179 +# This program is free software; you can redistribute it and/or modify
2180 +# it under the terms of the GNU General Public License as published by
2181 +# the Free Software Foundation; either version 2 of the License, or
2182 +# (at your option) any later version.
2183 +#
2184 +# This program is distributed in the hope that it will be useful,
2185 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2186 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2187 +# GNU General Public License for more details.
2188 +#
2189 +# You should have received a copy of the GNU General Public License
2190 +# along with this program; if not, write to the Free Software
2191 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
2192 +
2193 +
2194 +Listing XATTR/EA and getting the value
2195 +----------------------------------------------------------------------
2196 +For the inode standard attributes (owner, group, timestamps, etc.), aufs
2197 +shows the values from the topmost existing file. This behaviour is good
2198 +for the non-dir entries since the bahaviour exactly matches the shown
2199 +information. But for the directories, aufs considers all the same named
2200 +entries on the lower branches. Which means, if one of the lower entry
2201 +rejects readdir call, then aufs returns an error even if the topmost
2202 +entry allows it. This behaviour is necessary to respect the branch fs's
2203 +security, but can make users confused since the user-visible standard
2204 +attributes don't match the behaviour.
2205 +To address this issue, aufs has a mount option called dirperm1 which
2206 +checks the permission for the topmost entry only, and ignores the lower
2207 +entry's permission.
2208 +
2209 +A similar issue can happen around XATTR.
2210 +getxattr(2) and listxattr(2) families behave as if dirperm1 option is
2211 +always set. Otherwise these very unpleasant situation would happen.
2212 +- listxattr(2) may return the duplicated entries.
2213 +- users may not be able to remove or reset the XATTR forever,
2214 +
2215 +
2216 +XATTR/EA support in the internal (copy,move)-(up,down)
2217 +----------------------------------------------------------------------
2218 +Generally the extended attributes of inode are categorized as these.
2219 +- "security" for LSM and capability.
2220 +- "system" for posix ACL, 'acl' mount option is required for the branch
2221 +  fs generally.
2222 +- "trusted" for userspace, CAP_SYS_ADMIN is required.
2223 +- "user" for userspace, 'user_xattr' mount option is required for the
2224 +  branch fs generally.
2225 +
2226 +Moreover there are some other categories. Aufs handles these rather
2227 +unpopular categories as the ordinary ones, ie. there is no special
2228 +condition nor exception.
2229 +
2230 +In copy-up, the support for XATTR on the dst branch may differ from the
2231 +src branch. In this case, the copy-up operation will get an error and
2232 +the original user operation which triggered the copy-up will fail. It
2233 +can happen that even all copy-up will fail.
2234 +When both of src and dst branches support XATTR and if an error occurs
2235 +during copying XATTR, then the copy-up should fail obviously. That is a
2236 +good reason and aufs should return an error to userspace. But when only
2237 +the src branch support that XATTR, aufs should not return an error.
2238 +For example, the src branch supports ACL but the dst branch doesn't
2239 +because the dst branch may natively un-support it or temporary
2240 +un-support it due to "noacl" mount option. Of course, the dst branch fs
2241 +may NOT return an error even if the XATTR is not supported. It is
2242 +totally up to the branch fs.
2243 +
2244 +Anyway when the aufs internal copy-up gets an error from the dst branch
2245 +fs, then aufs tries removing the just copied entry and returns the error
2246 +to the userspace. The worst case of this situation will be all copy-up
2247 +will fail.
2248 +
2249 +For the copy-up operation, there two basic approaches.
2250 +- copy the specified XATTR only (by category above), and return the
2251 +  error unconditionally if it happens.
2252 +- copy all XATTR, and ignore the error on the specified category only.
2253 +
2254 +In order to support XATTR and to implement the correct behaviour, aufs
2255 +chooses the latter approach and introduces some new branch attributes,
2256 +"icexsec", "icexsys", "icextr", "icexusr", and "icexoth".
2257 +They correspond to the XATTR namespaces (see above). Additionally, to be
2258 +convenient, "icex" is also provided which means all "icex*" attributes
2259 +are set (here the word "icex" stands for "ignore copy-error on XATTR").
2260 +
2261 +The meaning of these attributes is to ignore the error from setting
2262 +XATTR on that branch.
2263 +Note that aufs tries copying all XATTR unconditionally, and ignores the
2264 +error from the dst branch according to the specified attributes.
2265 +
2266 +Some XATTR may have its default value. The default value may come from
2267 +the parent dir or the environment. If the default value is set at the
2268 +file creating-time, it will be overwritten by copy-up.
2269 +Some contradiction may happen I am afraid.
2270 +Do we need another attribute to stop copying XATTR? I am unsure. For
2271 +now, aufs implements the branch attributes to ignore the error.
2272 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt linux/Documentation/filesystems/aufs/design/07export.txt
2273 --- /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt 1970-01-01 01:00:00.000000000 +0100
2274 +++ linux/Documentation/filesystems/aufs/design/07export.txt    2021-12-03 15:38:59.929980643 +0100
2275 @@ -0,0 +1,58 @@
2276 +
2277 +# Copyright (C) 2005-2021 Junjiro R. Okajima
2278 +#
2279 +# This program is free software; you can redistribute it and/or modify
2280 +# it under the terms of the GNU General Public License as published by
2281 +# the Free Software Foundation; either version 2 of the License, or
2282 +# (at your option) any later version.
2283 +#
2284 +# This program is distributed in the hope that it will be useful,
2285 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2286 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2287 +# GNU General Public License for more details.
2288 +#
2289 +# You should have received a copy of the GNU General Public License
2290 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2291 +
2292 +Export Aufs via NFS
2293 +----------------------------------------------------------------------
2294 +Here is an approach.
2295 +- like xino/xib, add a new file 'xigen' which stores aufs inode
2296 +  generation.
2297 +- iget_locked(): initialize aufs inode generation for a new inode, and
2298 +  store it in xigen file.
2299 +- destroy_inode(): increment aufs inode generation and store it in xigen
2300 +  file. it is necessary even if it is not unlinked, because any data of
2301 +  inode may be changed by UDBA.
2302 +- encode_fh(): for a root dir, simply return FILEID_ROOT. otherwise
2303 +  build file handle by
2304 +  + branch id (4 bytes)
2305 +  + superblock generation (4 bytes)
2306 +  + inode number (4 or 8 bytes)
2307 +  + parent dir inode number (4 or 8 bytes)
2308 +  + inode generation (4 bytes))
2309 +  + return value of exportfs_encode_fh() for the parent on a branch (4
2310 +    bytes)
2311 +  + file handle for a branch (by exportfs_encode_fh())
2312 +- fh_to_dentry():
2313 +  + find the index of a branch from its id in handle, and check it is
2314 +    still exist in aufs.
2315 +  + 1st level: get the inode number from handle and search it in cache.
2316 +  + 2nd level: if not found in cache, get the parent inode number from
2317 +    the handle and search it in cache. and then open the found parent
2318 +    dir, find the matching inode number by vfs_readdir() and get its
2319 +    name, and call lookup_one_len() for the target dentry.
2320 +  + 3rd level: if the parent dir is not cached, call
2321 +    exportfs_decode_fh() for a branch and get the parent on a branch,
2322 +    build a pathname of it, convert it a pathname in aufs, call
2323 +    path_lookup(). now aufs gets a parent dir dentry, then handle it as
2324 +    the 2nd level.
2325 +  + to open the dir, aufs needs struct vfsmount. aufs keeps vfsmount
2326 +    for every branch, but not itself. to get this, (currently) aufs
2327 +    searches in current->nsproxy->mnt_ns list. it may not be a good
2328 +    idea, but I didn't get other approach.
2329 +  + test the generation of the gotten inode.
2330 +- every inode operation: they may get EBUSY due to UDBA. in this case,
2331 +  convert it into ESTALE for NFSD.
2332 +- readdir(): call lockdep_on/off() because filldir in NFSD calls
2333 +  lookup_one_len(), vfs_getattr(), encode_fh() and others.
2334 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt linux/Documentation/filesystems/aufs/design/08shwh.txt
2335 --- /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt   1970-01-01 01:00:00.000000000 +0100
2336 +++ linux/Documentation/filesystems/aufs/design/08shwh.txt      2021-12-03 15:38:59.929980643 +0100
2337 @@ -0,0 +1,52 @@
2338 +
2339 +# Copyright (C) 2005-2021 Junjiro R. Okajima
2340 +#
2341 +# This program is free software; you can redistribute it and/or modify
2342 +# it under the terms of the GNU General Public License as published by
2343 +# the Free Software Foundation; either version 2 of the License, or
2344 +# (at your option) any later version.
2345 +#
2346 +# This program is distributed in the hope that it will be useful,
2347 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2348 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2349 +# GNU General Public License for more details.
2350 +#
2351 +# You should have received a copy of the GNU General Public License
2352 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2353 +
2354 +Show Whiteout Mode (shwh)
2355 +----------------------------------------------------------------------
2356 +Generally aufs hides the name of whiteouts. But in some cases, to show
2357 +them is very useful for users. For instance, creating a new middle layer
2358 +(branch) by merging existing layers.
2359 +
2360 +(borrowing aufs1 HOW-TO from a user, Michael Towers)
2361 +When you have three branches,
2362 +- Bottom: 'system', squashfs (underlying base system), read-only
2363 +- Middle: 'mods', squashfs, read-only
2364 +- Top: 'overlay', ram (tmpfs), read-write
2365 +
2366 +The top layer is loaded at boot time and saved at shutdown, to preserve
2367 +the changes made to the system during the session.
2368 +When larger changes have been made, or smaller changes have accumulated,
2369 +the size of the saved top layer data grows. At this point, it would be
2370 +nice to be able to merge the two overlay branches ('mods' and 'overlay')
2371 +and rewrite the 'mods' squashfs, clearing the top layer and thus
2372 +restoring save and load speed.
2373 +
2374 +This merging is simplified by the use of another aufs mount, of just the
2375 +two overlay branches using the 'shwh' option.
2376 +# mount -t aufs -o ro,shwh,br:/livesys/overlay=ro+wh:/livesys/mods=rr+wh \
2377 +       aufs /livesys/merge_union
2378 +
2379 +A merged view of these two branches is then available at
2380 +/livesys/merge_union, and the new feature is that the whiteouts are
2381 +visible!
2382 +Note that in 'shwh' mode the aufs mount must be 'ro', which will disable
2383 +writing to all branches. Also the default mode for all branches is 'ro'.
2384 +It is now possible to save the combined contents of the two overlay
2385 +branches to a new squashfs, e.g.:
2386 +# mksquashfs /livesys/merge_union /path/to/newmods.squash
2387 +
2388 +This new squashfs archive can be stored on the boot device and the
2389 +initramfs will use it to replace the old one at the next boot.
2390 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt linux/Documentation/filesystems/aufs/design/10dynop.txt
2391 --- /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt  1970-01-01 01:00:00.000000000 +0100
2392 +++ linux/Documentation/filesystems/aufs/design/10dynop.txt     2021-12-03 15:38:59.929980643 +0100
2393 @@ -0,0 +1,47 @@
2394 +
2395 +# Copyright (C) 2010-2021 Junjiro R. Okajima
2396 +#
2397 +# This program is free software; you can redistribute it and/or modify
2398 +# it under the terms of the GNU General Public License as published by
2399 +# the Free Software Foundation; either version 2 of the License, or
2400 +# (at your option) any later version.
2401 +#
2402 +# This program is distributed in the hope that it will be useful,
2403 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2404 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2405 +# GNU General Public License for more details.
2406 +#
2407 +# You should have received a copy of the GNU General Public License
2408 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2409 +
2410 +Dynamically customizable FS operations
2411 +----------------------------------------------------------------------
2412 +Generally FS operations (struct inode_operations, struct
2413 +address_space_operations, struct file_operations, etc.) are defined as
2414 +"static const", but it never means that FS have only one set of
2415 +operation. Some FS have multiple sets of them. For instance, ext2 has
2416 +three sets, one for XIP, for NOBH, and for normal.
2417 +Since aufs overrides and redirects these operations, sometimes aufs has
2418 +to change its behaviour according to the branch FS type. More importantly
2419 +VFS acts differently if a function (member in the struct) is set or
2420 +not. It means aufs should have several sets of operations and select one
2421 +among them according to the branch FS definition.
2422 +
2423 +In order to solve this problem and not to affect the behaviour of VFS,
2424 +aufs defines these operations dynamically. For instance, aufs defines
2425 +dummy direct_IO function for struct address_space_operations, but it may
2426 +not be set to the address_space_operations actually. When the branch FS
2427 +doesn't have it, aufs doesn't set it to its address_space_operations
2428 +while the function definition itself is still alive. So the behaviour
2429 +itself will not change, and it will return an error when direct_IO is
2430 +not set.
2431 +
2432 +The lifetime of these dynamically generated operation object is
2433 +maintained by aufs branch object. When the branch is removed from aufs,
2434 +the reference counter of the object is decremented. When it reaches
2435 +zero, the dynamically generated operation object will be freed.
2436 +
2437 +This approach is designed to support AIO (io_submit), Direct I/O and
2438 +XIP (DAX) mainly.
2439 +Currently this approach is applied to address_space_operations for
2440 +regular files only.
2441 diff -urN /usr/share/empty/Documentation/filesystems/aufs/README linux/Documentation/filesystems/aufs/README
2442 --- /usr/share/empty/Documentation/filesystems/aufs/README      1970-01-01 01:00:00.000000000 +0100
2443 +++ linux/Documentation/filesystems/aufs/README 2021-12-03 15:40:58.233313963 +0100
2444 @@ -0,0 +1,396 @@
2445 +
2446 +Aufs5 -- advanced multi layered unification filesystem version 5.x
2447 +http://aufs.sf.net
2448 +Junjiro R. Okajima
2449 +
2450 +
2451 +0. Introduction
2452 +----------------------------------------
2453 +In the early days, aufs was entirely re-designed and re-implemented
2454 +Unionfs Version 1.x series. Adding many original ideas, approaches,
2455 +improvements and implementations, it became totally different from
2456 +Unionfs while keeping the basic features.
2457 +Later, Unionfs Version 2.x series began taking some of the same
2458 +approaches to aufs1's.
2459 +Unionfs was being developed by Professor Erez Zadok at Stony Brook
2460 +University and his team.
2461 +
2462 +Aufs5 supports linux-v5.0 and later, If you want older kernel version
2463 +support,
2464 +- for linux-v4.x series, try aufs4-linux.git or aufs4-standalone.git
2465 +- for linux-v3.x series, try aufs3-linux.git or aufs3-standalone.git
2466 +- for linux-v2.6.16 and later, try aufs2-2.6.git, aufs2-standalone.git
2467 +  or aufs1 from CVS on SourceForge.
2468 +
2469 +Note: it becomes clear that "Aufs was rejected. Let's give it up."
2470 +      According to Christoph Hellwig, linux rejects all union-type
2471 +      filesystems but UnionMount.
2472 +<http://marc.info/?l=linux-kernel&m=123938533724484&w=2>
2473 +
2474 +PS. Al Viro seems have a plan to merge aufs as well as overlayfs and
2475 +    UnionMount, and he pointed out an issue around a directory mutex
2476 +    lock and aufs addressed it. But it is still unsure whether aufs will
2477 +    be merged (or any other union solution).
2478 +<http://marc.info/?l=linux-kernel&m=136312705029295&w=1>
2479 +
2480 +
2481 +1. Features
2482 +----------------------------------------
2483 +- unite several directories into a single virtual filesystem. The member
2484 +  directory is called as a branch.
2485 +- you can specify the permission flags to the branch, which are 'readonly',
2486 +  'readwrite' and 'whiteout-able.'
2487 +- by upper writable branch, internal copyup and whiteout, files/dirs on
2488 +  readonly branch are modifiable logically.
2489 +- dynamic branch manipulation, add, del.
2490 +- etc...
2491 +
2492 +Also there are many enhancements in aufs, such as:
2493 +- test only the highest one for the directory permission (dirperm1)
2494 +- copyup on open (coo=)
2495 +- 'move' policy for copy-up between two writable branches, after
2496 +  checking free space.
2497 +- xattr, acl
2498 +- readdir(3) in userspace.
2499 +- keep inode number by external inode number table
2500 +- keep the timestamps of file/dir in internal copyup operation
2501 +- seekable directory, supporting NFS readdir.
2502 +- whiteout is hardlinked in order to reduce the consumption of inodes
2503 +  on branch
2504 +- do not copyup, nor create a whiteout when it is unnecessary
2505 +- revert a single systemcall when an error occurs in aufs
2506 +- remount interface instead of ioctl
2507 +- maintain /etc/mtab by an external command, /sbin/mount.aufs.
2508 +- loopback mounted filesystem as a branch
2509 +- kernel thread for removing the dir who has a plenty of whiteouts
2510 +- support copyup sparse file (a file which has a 'hole' in it)
2511 +- default permission flags for branches
2512 +- selectable permission flags for ro branch, whether whiteout can
2513 +  exist or not
2514 +- export via NFS.
2515 +- support <sysfs>/fs/aufs and <debugfs>/aufs.
2516 +- support multiple writable branches, some policies to select one
2517 +  among multiple writable branches.
2518 +- a new semantics for link(2) and rename(2) to support multiple
2519 +  writable branches.
2520 +- no glibc changes are required.
2521 +- pseudo hardlink (hardlink over branches)
2522 +- allow a direct access manually to a file on branch, e.g. bypassing aufs.
2523 +  including NFS or remote filesystem branch.
2524 +- userspace wrapper for pathconf(3)/fpathconf(3) with _PC_LINK_MAX.
2525 +- and more...
2526 +
2527 +Currently these features are dropped temporary from aufs5.
2528 +See design/08plan.txt in detail.
2529 +- nested mount, i.e. aufs as readonly no-whiteout branch of another aufs
2530 +  (robr)
2531 +- statistics of aufs thread (/sys/fs/aufs/stat)
2532 +
2533 +Features or just an idea in the future (see also design/*.txt),
2534 +- reorder the branch index without del/re-add.
2535 +- permanent xino files for NFSD
2536 +- an option for refreshing the opened files after add/del branches
2537 +- light version, without branch manipulation. (unnecessary?)
2538 +- copyup in userspace
2539 +- inotify in userspace
2540 +- readv/writev
2541 +
2542 +
2543 +2. Download
2544 +----------------------------------------
2545 +There are three GIT trees for aufs5, aufs5-linux.git,
2546 +aufs5-standalone.git, and aufs-util.git. Note that there is no "5" in
2547 +"aufs-util.git."
2548 +While the aufs-util is always necessary, you need either of aufs5-linux
2549 +or aufs5-standalone.
2550 +
2551 +The aufs5-linux tree includes the whole linux mainline GIT tree,
2552 +git://git.kernel.org/.../torvalds/linux.git.
2553 +And you cannot select CONFIG_AUFS_FS=m for this version, eg. you cannot
2554 +build aufs5 as an external kernel module.
2555 +Several extra patches are not included in this tree. Only
2556 +aufs5-standalone tree contains them. They are described in the later
2557 +section "Configuration and Compilation."
2558 +
2559 +On the other hand, the aufs5-standalone tree has only aufs source files
2560 +and necessary patches, and you can select CONFIG_AUFS_FS=m.
2561 +But you need to apply all aufs patches manually.
2562 +
2563 +You will find GIT branches whose name is in form of "aufs5.x" where "x"
2564 +represents the linux kernel version, "linux-5.x". For instance,
2565 +"aufs5.0" is for linux-5.0. For latest "linux-5.x-rcN", use
2566 +"aufs5.x-rcN" branch.
2567 +
2568 +o aufs5-linux tree
2569 +$ git clone --reference /your/linux/git/tree \
2570 +       git://github.com/sfjro/aufs5-linux.git aufs5-linux.git
2571 +- if you don't have linux GIT tree, then remove "--reference ..."
2572 +$ cd aufs5-linux.git
2573 +$ git checkout origin/aufs5.0
2574 +
2575 +Or You may want to directly git-pull aufs into your linux GIT tree, and
2576 +leave the patch-work to GIT.
2577 +$ cd /your/linux/git/tree
2578 +$ git remote add aufs5 git://github.com/sfjro/aufs5-linux.git
2579 +$ git fetch aufs5
2580 +$ git checkout -b my5.0 v5.0
2581 +$ (add your local change...)
2582 +$ git pull aufs5 aufs5.0
2583 +- now you have v5.0 + your_changes + aufs5.0 in you my5.0 branch.
2584 +- you may need to solve some conflicts between your_changes and
2585 +  aufs5.0. in this case, git-rerere is recommended so that you can
2586 +  solve the similar conflicts automatically when you upgrade to 5.1 or
2587 +  later in the future.
2588 +
2589 +o aufs5-standalone tree
2590 +$ git clone git://github.com/sfjro/aufs5-standalone.git aufs5-standalone.git
2591 +$ cd aufs5-standalone.git
2592 +$ git checkout origin/aufs5.0
2593 +
2594 +o aufs-util tree
2595 +$ git clone git://git.code.sf.net/p/aufs/aufs-util aufs-util.git
2596 +- note that the public aufs-util.git is on SourceForge instead of
2597 +  GitHUB.
2598 +$ cd aufs-util.git
2599 +$ git checkout origin/aufs5.0
2600 +
2601 +Note: The 5.x-rcN branch is to be used with `rc' kernel versions ONLY.
2602 +The minor version number, 'x' in '5.x', of aufs may not always
2603 +follow the minor version number of the kernel.
2604 +Because changes in the kernel that cause the use of a new
2605 +minor version number do not always require changes to aufs-util.
2606 +
2607 +Since aufs-util has its own minor version number, you may not be
2608 +able to find a GIT branch in aufs-util for your kernel's
2609 +exact minor version number.
2610 +In this case, you should git-checkout the branch for the
2611 +nearest lower number.
2612 +
2613 +For (an unreleased) example:
2614 +If you are using "linux-5.10" and the "aufs5.10" branch
2615 +does not exist in aufs-util repository, then "aufs5.9", "aufs5.8"
2616 +or something numerically smaller is the branch for your kernel.
2617 +
2618 +Also you can view all branches by
2619 +       $ git branch -a
2620 +
2621 +
2622 +3. Configuration and Compilation
2623 +----------------------------------------
2624 +Make sure you have git-checkout'ed the correct branch.
2625 +
2626 +For aufs5-linux tree,
2627 +- enable CONFIG_AUFS_FS.
2628 +- set other aufs configurations if necessary.
2629 +- for aufs5.13 and later
2630 +  Because aufs is not only an ordinary filesystem (callee of VFS), but
2631 +  also a caller of VFS functions for branch filesystems, subclassing of
2632 +  the internal locks for LOCKDEP is necessary. LOCKDEP is a debugging
2633 +  feature of linux kernel. If you enable CONFIG_LOCKDEP, then you will
2634 +  need to customize some LOCKDEP numbers. Here are what I use on my
2635 +  test environment.
2636 +       CONFIG_LOCKDEP_BITS=21
2637 +       CONFIG_LOCKDEP_CHAINS_BITS=21
2638 +       CONFIG_LOCKDEP_STACK_TRACE_BITS=24
2639 +
2640 +For aufs5-standalone tree,
2641 +There are several ways to build.
2642 +
2643 +1.
2644 +- apply ./aufs5-kbuild.patch to your kernel source files.
2645 +- apply ./aufs5-base.patch too.
2646 +- apply ./aufs5-mmap.patch too.
2647 +- apply ./aufs5-standalone.patch too, if you have a plan to set
2648 +  CONFIG_AUFS_FS=m. otherwise you don't need ./aufs5-standalone.patch.
2649 +- copy ./{Documentation,fs,include/uapi/linux/aufs_type.h} files to your
2650 +  kernel source tree. Never copy $PWD/include/uapi/linux/Kbuild.
2651 +- enable CONFIG_AUFS_FS, you can select either
2652 +  =m or =y.
2653 +- and build your kernel as usual.
2654 +- install the built kernel.
2655 +- install the header files too by "make headers_install" to the
2656 +  directory where you specify. By default, it is $PWD/usr.
2657 +  "make help" shows a brief note for headers_install.
2658 +- and reboot your system.
2659 +
2660 +2.
2661 +- module only (CONFIG_AUFS_FS=m).
2662 +- apply ./aufs5-base.patch to your kernel source files.
2663 +- apply ./aufs5-mmap.patch too.
2664 +- apply ./aufs5-standalone.patch too.
2665 +- build your kernel, don't forget "make headers_install", and reboot.
2666 +- edit ./config.mk and set other aufs configurations if necessary.
2667 +  Note: You should read $PWD/fs/aufs/Kconfig carefully which describes
2668 +  every aufs configurations.
2669 +- build the module by simple "make".
2670 +- you can specify ${KDIR} make variable which points to your kernel
2671 +  source tree.
2672 +- install the files
2673 +  + run "make install" to install the aufs module, or copy the built
2674 +    $PWD/aufs.ko to /lib/modules/... and run depmod -a (or reboot simply).
2675 +  + run "make install_headers" (instead of headers_install) to install
2676 +    the modified aufs header file (you can specify DESTDIR which is
2677 +    available in aufs standalone version's Makefile only), or copy
2678 +    $PWD/usr/include/linux/aufs_type.h to /usr/include/linux or wherever
2679 +    you like manually. By default, the target directory is $PWD/usr.
2680 +- no need to apply aufs5-kbuild.patch, nor copying source files to your
2681 +  kernel source tree.
2682 +
2683 +Note: The header file aufs_type.h is necessary to build aufs-util
2684 +      as well as "make headers_install" in the kernel source tree.
2685 +      headers_install is subject to be forgotten, but it is essentially
2686 +      necessary, not only for building aufs-util.
2687 +      You may not meet problems without headers_install in some older
2688 +      version though.
2689 +
2690 +And then,
2691 +- read README in aufs-util, build and install it
2692 +- note that your distribution may contain an obsoleted version of
2693 +  aufs_type.h in /usr/include/linux or something. When you build aufs
2694 +  utilities, make sure that your compiler refers the correct aufs header
2695 +  file which is built by "make headers_install."
2696 +- if you want to use readdir(3) in userspace or pathconf(3) wrapper,
2697 +  then run "make install_ulib" too. And refer to the aufs manual in
2698 +  detail.
2699 +
2700 +There several other patches in aufs5-standalone.git. They are all
2701 +optional. When you meet some problems, they will help you.
2702 +- aufs5-loopback.patch
2703 +  Supports a nested loopback mount in a branch-fs. This patch is
2704 +  unnecessary until aufs produces a message like "you may want to try
2705 +  another patch for loopback file".
2706 +- vfs-ino.patch
2707 +  Modifies a system global kernel internal function get_next_ino() in
2708 +  order to stop assigning 0 for an inode-number. Not directly related to
2709 +  aufs, but recommended generally.
2710 +- tmpfs-idr.patch
2711 +  Keeps the tmpfs inode number as the lowest value. Effective to reduce
2712 +  the size of aufs XINO files for tmpfs branch. Also it prevents the
2713 +  duplication of inode number, which is important for backup tools and
2714 +  other utilities. When you find aufs XINO files for tmpfs branch
2715 +  growing too much, try this patch.
2716 +
2717 +
2718 +4. Usage
2719 +----------------------------------------
2720 +At first, make sure aufs-util are installed, and please read the aufs
2721 +manual, aufs.5 in aufs-util.git tree.
2722 +$ man -l aufs.5
2723 +
2724 +And then,
2725 +$ mkdir /tmp/rw /tmp/aufs
2726 +# mount -t aufs -o br=/tmp/rw:${HOME} none /tmp/aufs
2727 +
2728 +Here is another example. The result is equivalent.
2729 +# mount -t aufs -o br=/tmp/rw=rw:${HOME}=ro none /tmp/aufs
2730 +  Or
2731 +# mount -t aufs -o br:/tmp/rw none /tmp/aufs
2732 +# mount -o remount,append:${HOME} /tmp/aufs
2733 +
2734 +Then, you can see whole tree of your home dir through /tmp/aufs. If
2735 +you modify a file under /tmp/aufs, the one on your home directory is
2736 +not affected, instead the same named file will be newly created under
2737 +/tmp/rw. And all of your modification to a file will be applied to
2738 +the one under /tmp/rw. This is called the file based Copy on Write
2739 +(COW) method.
2740 +Aufs mount options are described in aufs.5.
2741 +If you run chroot or something and make your aufs as a root directory,
2742 +then you need to customize the shutdown script. See the aufs manual in
2743 +detail.
2744 +
2745 +Additionally, there are some sample usages of aufs which are a
2746 +diskless system with network booting, and LiveCD over NFS.
2747 +See sample dir in CVS tree on SourceForge.
2748 +
2749 +
2750 +5. Contact
2751 +----------------------------------------
2752 +When you have any problems or strange behaviour in aufs, please let me
2753 +know with:
2754 +- /proc/mounts (instead of the output of mount(8))
2755 +- /sys/module/aufs/*
2756 +- /sys/fs/aufs/* (if you have them)
2757 +- /debug/aufs/* (if you have them)
2758 +- linux kernel version
2759 +  if your kernel is not plain, for example modified by distributor,
2760 +  the url where i can download its source is necessary too.
2761 +- aufs version which was printed at loading the module or booting the
2762 +  system, instead of the date you downloaded.
2763 +- configuration (define/undefine CONFIG_AUFS_xxx)
2764 +- kernel configuration or /proc/config.gz (if you have it)
2765 +- LSM (linux security module, if you are using)
2766 +- behaviour which you think to be incorrect
2767 +- actual operation, reproducible one is better
2768 +- mailto: aufs-users at lists.sourceforge.net
2769 +
2770 +Usually, I don't watch the Public Areas(Bugs, Support Requests, Patches,
2771 +and Feature Requests) on SourceForge. Please join and write to
2772 +aufs-users ML.
2773 +
2774 +
2775 +6. Acknowledgements
2776 +----------------------------------------
2777 +Thanks to everyone who have tried and are using aufs, whoever
2778 +have reported a bug or any feedback.
2779 +
2780 +Especially donators:
2781 +Tomas Matejicek(slax.org) made a donation (much more than once).
2782 +       Since Apr 2010, Tomas M (the author of Slax and Linux Live
2783 +       scripts) is making "doubling" donations.
2784 +       Unfortunately I cannot list all of the donators, but I really
2785 +       appreciate.
2786 +       It ends Aug 2010, but the ordinary donation URL is still available.
2787 +       <http://sourceforge.net/donate/index.php?group_id=167503>
2788 +Dai Itasaka made a donation (2007/8).
2789 +Chuck Smith made a donation (2008/4, 10 and 12).
2790 +Henk Schoneveld made a donation (2008/9).
2791 +Chih-Wei Huang, ASUS, CTC donated Eee PC 4G (2008/10).
2792 +Francois Dupoux made a donation (2008/11).
2793 +Bruno Cesar Ribas and Luis Carlos Erpen de Bona, C3SL serves public
2794 +       aufs2 GIT tree (2009/2).
2795 +William Grant made a donation (2009/3).
2796 +Patrick Lane made a donation (2009/4).
2797 +The Mail Archive (mail-archive.com) made donations (2009/5).
2798 +Nippy Networks (Ed Wildgoose) made a donation (2009/7).
2799 +New Dream Network, LLC (www.dreamhost.com) made a donation (2009/11).
2800 +Pavel Pronskiy made a donation (2011/2).
2801 +Iridium and Inmarsat satellite phone retailer (www.mailasail.com), Nippy
2802 +       Networks (Ed Wildgoose) made a donation for hardware (2011/3).
2803 +Max Lekomcev (DOM-TV project) made a donation (2011/7, 12, 2012/3, 6 and
2804 +11).
2805 +Sam Liddicott made a donation (2011/9).
2806 +Era Scarecrow made a donation (2013/4).
2807 +Bor Ratajc made a donation (2013/4).
2808 +Alessandro Gorreta made a donation (2013/4).
2809 +POIRETTE Marc made a donation (2013/4).
2810 +Alessandro Gorreta made a donation (2013/4).
2811 +lauri kasvandik made a donation (2013/5).
2812 +"pemasu from Finland" made a donation (2013/7).
2813 +The Parted Magic Project made a donation (2013/9 and 11).
2814 +Pavel Barta made a donation (2013/10).
2815 +Nikolay Pertsev made a donation (2014/5).
2816 +James B made a donation (2014/7 and 2015/7).
2817 +Stefano Di Biase made a donation (2014/8).
2818 +Daniel Epellei made a donation (2015/1).
2819 +OmegaPhil made a donation (2016/1, 2018/4).
2820 +Tomasz Szewczyk made a donation (2016/4).
2821 +James Burry made a donation (2016/12).
2822 +Carsten Rose made a donation (2018/9).
2823 +Porteus Kiosk made a donation (2018/10).
2824 +
2825 +Thank you very much.
2826 +Donations are always, including future donations, very important and
2827 +helpful for me to keep on developing aufs.
2828 +
2829 +
2830 +7.
2831 +----------------------------------------
2832 +If you are an experienced user, no explanation is needed. Aufs is
2833 +just a linux filesystem.
2834 +
2835 +
2836 +Enjoy!
2837 +
2838 +# Local variables: ;
2839 +# mode: text;
2840 +# End: ;
2841 diff -urN /usr/share/empty/fs/aufs/aufs.h linux/fs/aufs/aufs.h
2842 --- /usr/share/empty/fs/aufs/aufs.h     1970-01-01 01:00:00.000000000 +0100
2843 +++ linux/fs/aufs/aufs.h        2021-12-03 15:38:59.929980643 +0100
2844 @@ -0,0 +1,62 @@
2845 +/* SPDX-License-Identifier: GPL-2.0 */
2846 +/*
2847 + * Copyright (C) 2005-2021 Junjiro R. Okajima
2848 + *
2849 + * This program, aufs is free software; you can redistribute it and/or modify
2850 + * it under the terms of the GNU General Public License as published by
2851 + * the Free Software Foundation; either version 2 of the License, or
2852 + * (at your option) any later version.
2853 + *
2854 + * This program is distributed in the hope that it will be useful,
2855 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2856 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2857 + * GNU General Public License for more details.
2858 + *
2859 + * You should have received a copy of the GNU General Public License
2860 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2861 + */
2862 +
2863 +/*
2864 + * all header files
2865 + */
2866 +
2867 +#ifndef __AUFS_H__
2868 +#define __AUFS_H__
2869 +
2870 +#ifdef __KERNEL__
2871 +
2872 +#define AuStub(type, name, body, ...) \
2873 +       static inline type name(__VA_ARGS__) { body; }
2874 +
2875 +#define AuStubVoid(name, ...) \
2876 +       AuStub(void, name, , __VA_ARGS__)
2877 +#define AuStubInt0(name, ...) \
2878 +       AuStub(int, name, return 0, __VA_ARGS__)
2879 +
2880 +#include "debug.h"
2881 +
2882 +#include "branch.h"
2883 +#include "cpup.h"
2884 +#include "dcsub.h"
2885 +#include "dbgaufs.h"
2886 +#include "dentry.h"
2887 +#include "dir.h"
2888 +#include "dirren.h"
2889 +#include "dynop.h"
2890 +#include "file.h"
2891 +#include "fstype.h"
2892 +#include "hbl.h"
2893 +#include "inode.h"
2894 +#include "lcnt.h"
2895 +#include "loop.h"
2896 +#include "module.h"
2897 +#include "opts.h"
2898 +#include "rwsem.h"
2899 +#include "super.h"
2900 +#include "sysaufs.h"
2901 +#include "vfsub.h"
2902 +#include "whout.h"
2903 +#include "wkq.h"
2904 +
2905 +#endif /* __KERNEL__ */
2906 +#endif /* __AUFS_H__ */
2907 diff -urN /usr/share/empty/fs/aufs/branch.c linux/fs/aufs/branch.c
2908 --- /usr/share/empty/fs/aufs/branch.c   1970-01-01 01:00:00.000000000 +0100
2909 +++ linux/fs/aufs/branch.c      2021-12-03 15:38:59.933313976 +0100
2910 @@ -0,0 +1,1427 @@
2911 +// SPDX-License-Identifier: GPL-2.0
2912 +/*
2913 + * Copyright (C) 2005-2021 Junjiro R. Okajima
2914 + *
2915 + * This program, aufs is free software; you can redistribute it and/or modify
2916 + * it under the terms of the GNU General Public License as published by
2917 + * the Free Software Foundation; either version 2 of the License, or
2918 + * (at your option) any later version.
2919 + *
2920 + * This program is distributed in the hope that it will be useful,
2921 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2922 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2923 + * GNU General Public License for more details.
2924 + *
2925 + * You should have received a copy of the GNU General Public License
2926 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2927 + */
2928 +
2929 +/*
2930 + * branch management
2931 + */
2932 +
2933 +#include <linux/compat.h>
2934 +#include <linux/statfs.h>
2935 +#include "aufs.h"
2936 +
2937 +/*
2938 + * free a single branch
2939 + */
2940 +static void au_br_do_free(struct au_branch *br)
2941 +{
2942 +       int i;
2943 +       struct au_wbr *wbr;
2944 +       struct au_dykey **key;
2945 +
2946 +       au_hnotify_fin_br(br);
2947 +       /* always, regardless the mount option */
2948 +       au_dr_hino_free(&br->br_dirren);
2949 +       au_xino_put(br);
2950 +
2951 +       AuLCntZero(au_lcnt_read(&br->br_nfiles, /*do_rev*/0));
2952 +       au_lcnt_fin(&br->br_nfiles, /*do_sync*/0);
2953 +       AuLCntZero(au_lcnt_read(&br->br_count, /*do_rev*/0));
2954 +       au_lcnt_fin(&br->br_count, /*do_sync*/0);
2955 +
2956 +       wbr = br->br_wbr;
2957 +       if (wbr) {
2958 +               for (i = 0; i < AuBrWh_Last; i++)
2959 +                       dput(wbr->wbr_wh[i]);
2960 +               AuDebugOn(atomic_read(&wbr->wbr_wh_running));
2961 +               AuRwDestroy(&wbr->wbr_wh_rwsem);
2962 +       }
2963 +
2964 +       if (br->br_fhsm) {
2965 +               au_br_fhsm_fin(br->br_fhsm);
2966 +               au_kfree_try_rcu(br->br_fhsm);
2967 +       }
2968 +
2969 +       key = br->br_dykey;
2970 +       for (i = 0; i < AuBrDynOp; i++, key++)
2971 +               if (*key)
2972 +                       au_dy_put(*key);
2973 +               else
2974 +                       break;
2975 +
2976 +       /* recursive lock, s_umount of branch's */
2977 +       /* synchronize_rcu(); */ /* why? */
2978 +       lockdep_off();
2979 +       path_put(&br->br_path);
2980 +       lockdep_on();
2981 +       au_kfree_rcu(wbr);
2982 +       au_lcnt_wait_for_fin(&br->br_nfiles);
2983 +       au_lcnt_wait_for_fin(&br->br_count);
2984 +       /* I don't know why, but percpu_refcount requires this */
2985 +       /* synchronize_rcu(); */
2986 +       au_kfree_rcu(br);
2987 +}
2988 +
2989 +/*
2990 + * frees all branches
2991 + */
2992 +void au_br_free(struct au_sbinfo *sbinfo)
2993 +{
2994 +       aufs_bindex_t bmax;
2995 +       struct au_branch **br;
2996 +
2997 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
2998 +
2999 +       bmax = sbinfo->si_bbot + 1;
3000 +       br = sbinfo->si_branch;
3001 +       while (bmax--)
3002 +               au_br_do_free(*br++);
3003 +}
3004 +
3005 +/*
3006 + * find the index of a branch which is specified by @br_id.
3007 + */
3008 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id)
3009 +{
3010 +       aufs_bindex_t bindex, bbot;
3011 +
3012 +       bbot = au_sbbot(sb);
3013 +       for (bindex = 0; bindex <= bbot; bindex++)
3014 +               if (au_sbr_id(sb, bindex) == br_id)
3015 +                       return bindex;
3016 +       return -1;
3017 +}
3018 +
3019 +/* ---------------------------------------------------------------------- */
3020 +
3021 +/*
3022 + * add a branch
3023 + */
3024 +
3025 +static int test_overlap(struct super_block *sb, struct dentry *h_adding,
3026 +                       struct dentry *h_root)
3027 +{
3028 +       if (unlikely(h_adding == h_root
3029 +                    || au_test_loopback_overlap(sb, h_adding)))
3030 +               return 1;
3031 +       if (h_adding->d_sb != h_root->d_sb)
3032 +               return 0;
3033 +       return au_test_subdir(h_adding, h_root)
3034 +               || au_test_subdir(h_root, h_adding);
3035 +}
3036 +
3037 +/*
3038 + * returns a newly allocated branch. @new_nbranch is a number of branches
3039 + * after adding a branch.
3040 + */
3041 +static struct au_branch *au_br_alloc(struct super_block *sb, int new_nbranch,
3042 +                                    int perm)
3043 +{
3044 +       struct au_branch *add_branch;
3045 +       struct dentry *root;
3046 +       struct inode *inode;
3047 +       int err;
3048 +
3049 +       err = -ENOMEM;
3050 +       add_branch = kzalloc(sizeof(*add_branch), GFP_NOFS);
3051 +       if (unlikely(!add_branch))
3052 +               goto out;
3053 +       add_branch->br_xino = au_xino_alloc(/*nfile*/1);
3054 +       if (unlikely(!add_branch->br_xino))
3055 +               goto out_br;
3056 +       err = au_hnotify_init_br(add_branch, perm);
3057 +       if (unlikely(err))
3058 +               goto out_xino;
3059 +
3060 +       if (au_br_writable(perm)) {
3061 +               /* may be freed separately at changing the branch permission */
3062 +               add_branch->br_wbr = kzalloc(sizeof(*add_branch->br_wbr),
3063 +                                            GFP_NOFS);
3064 +               if (unlikely(!add_branch->br_wbr))
3065 +                       goto out_hnotify;
3066 +       }
3067 +
3068 +       if (au_br_fhsm(perm)) {
3069 +               err = au_fhsm_br_alloc(add_branch);
3070 +               if (unlikely(err))
3071 +                       goto out_wbr;
3072 +       }
3073 +
3074 +       root = sb->s_root;
3075 +       err = au_sbr_realloc(au_sbi(sb), new_nbranch, /*may_shrink*/0);
3076 +       if (!err)
3077 +               err = au_di_realloc(au_di(root), new_nbranch, /*may_shrink*/0);
3078 +       if (!err) {
3079 +               inode = d_inode(root);
3080 +               err = au_hinode_realloc(au_ii(inode), new_nbranch,
3081 +                                       /*may_shrink*/0);
3082 +       }
3083 +       if (!err)
3084 +               return add_branch; /* success */
3085 +
3086 +out_wbr:
3087 +       au_kfree_rcu(add_branch->br_wbr);
3088 +out_hnotify:
3089 +       au_hnotify_fin_br(add_branch);
3090 +out_xino:
3091 +       au_xino_put(add_branch);
3092 +out_br:
3093 +       au_kfree_rcu(add_branch);
3094 +out:
3095 +       return ERR_PTR(err);
3096 +}
3097 +
3098 +/*
3099 + * test if the branch permission is legal or not.
3100 + */
3101 +static int test_br(struct inode *inode, int brperm, char *path)
3102 +{
3103 +       int err;
3104 +
3105 +       err = (au_br_writable(brperm) && IS_RDONLY(inode));
3106 +       if (!err)
3107 +               goto out;
3108 +
3109 +       err = -EINVAL;
3110 +       pr_err("write permission for readonly mount or inode, %s\n", path);
3111 +
3112 +out:
3113 +       return err;
3114 +}
3115 +
3116 +/*
3117 + * returns:
3118 + * 0: success, the caller will add it
3119 + * plus: success, it is already unified, the caller should ignore it
3120 + * minus: error
3121 + */
3122 +static int test_add(struct super_block *sb, struct au_opt_add *add, int remount)
3123 +{
3124 +       int err;
3125 +       aufs_bindex_t bbot, bindex;
3126 +       struct dentry *root, *h_dentry;
3127 +       struct inode *inode, *h_inode;
3128 +
3129 +       root = sb->s_root;
3130 +       bbot = au_sbbot(sb);
3131 +       if (unlikely(bbot >= 0
3132 +                    && au_find_dbindex(root, add->path.dentry) >= 0)) {
3133 +               err = 1;
3134 +               if (!remount) {
3135 +                       err = -EINVAL;
3136 +                       pr_err("%s duplicated\n", add->pathname);
3137 +               }
3138 +               goto out;
3139 +       }
3140 +
3141 +       err = -ENOSPC; /* -E2BIG; */
3142 +       if (unlikely(AUFS_BRANCH_MAX <= add->bindex
3143 +                    || AUFS_BRANCH_MAX - 1 <= bbot)) {
3144 +               pr_err("number of branches exceeded %s\n", add->pathname);
3145 +               goto out;
3146 +       }
3147 +
3148 +       err = -EDOM;
3149 +       if (unlikely(add->bindex < 0 || bbot + 1 < add->bindex)) {
3150 +               pr_err("bad index %d\n", add->bindex);
3151 +               goto out;
3152 +       }
3153 +
3154 +       inode = d_inode(add->path.dentry);
3155 +       err = -ENOENT;
3156 +       if (unlikely(!inode->i_nlink)) {
3157 +               pr_err("no existence %s\n", add->pathname);
3158 +               goto out;
3159 +       }
3160 +
3161 +       err = -EINVAL;
3162 +       if (unlikely(inode->i_sb == sb)) {
3163 +               pr_err("%s must be outside\n", add->pathname);
3164 +               goto out;
3165 +       }
3166 +
3167 +       if (unlikely(au_test_fs_unsuppoted(inode->i_sb))) {
3168 +               pr_err("unsupported filesystem, %s (%s)\n",
3169 +                      add->pathname, au_sbtype(inode->i_sb));
3170 +               goto out;
3171 +       }
3172 +
3173 +       if (unlikely(inode->i_sb->s_stack_depth)) {
3174 +               pr_err("already stacked, %s (%s)\n",
3175 +                      add->pathname, au_sbtype(inode->i_sb));
3176 +               goto out;
3177 +       }
3178 +
3179 +       err = test_br(d_inode(add->path.dentry), add->perm, add->pathname);
3180 +       if (unlikely(err))
3181 +               goto out;
3182 +
3183 +       if (bbot < 0)
3184 +               return 0; /* success */
3185 +
3186 +       err = -EINVAL;
3187 +       for (bindex = 0; bindex <= bbot; bindex++)
3188 +               if (unlikely(test_overlap(sb, add->path.dentry,
3189 +                                         au_h_dptr(root, bindex)))) {
3190 +                       pr_err("%s is overlapped\n", add->pathname);
3191 +                       goto out;
3192 +               }
3193 +
3194 +       err = 0;
3195 +       if (au_opt_test(au_mntflags(sb), WARN_PERM)) {
3196 +               h_dentry = au_h_dptr(root, 0);
3197 +               h_inode = d_inode(h_dentry);
3198 +               if ((h_inode->i_mode & S_IALLUGO) != (inode->i_mode & S_IALLUGO)
3199 +                   || !uid_eq(h_inode->i_uid, inode->i_uid)
3200 +                   || !gid_eq(h_inode->i_gid, inode->i_gid))
3201 +                       pr_warn("uid/gid/perm %s %u/%u/0%o, %u/%u/0%o\n",
3202 +                               add->pathname,
3203 +                               i_uid_read(inode), i_gid_read(inode),
3204 +                               (inode->i_mode & S_IALLUGO),
3205 +                               i_uid_read(h_inode), i_gid_read(h_inode),
3206 +                               (h_inode->i_mode & S_IALLUGO));
3207 +       }
3208 +
3209 +out:
3210 +       return err;
3211 +}
3212 +
3213 +/*
3214 + * initialize or clean the whiteouts for an adding branch
3215 + */
3216 +static int au_br_init_wh(struct super_block *sb, struct au_branch *br,
3217 +                        int new_perm)
3218 +{
3219 +       int err, old_perm;
3220 +       aufs_bindex_t bindex;
3221 +       struct inode *h_inode;
3222 +       struct au_wbr *wbr;
3223 +       struct au_hinode *hdir;
3224 +       struct dentry *h_dentry;
3225 +
3226 +       err = vfsub_mnt_want_write(au_br_mnt(br));
3227 +       if (unlikely(err))
3228 +               goto out;
3229 +
3230 +       wbr = br->br_wbr;
3231 +       old_perm = br->br_perm;
3232 +       br->br_perm = new_perm;
3233 +       hdir = NULL;
3234 +       h_inode = NULL;
3235 +       bindex = au_br_index(sb, br->br_id);
3236 +       if (0 <= bindex) {
3237 +               hdir = au_hi(d_inode(sb->s_root), bindex);
3238 +               au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
3239 +       } else {
3240 +               h_dentry = au_br_dentry(br);
3241 +               h_inode = d_inode(h_dentry);
3242 +               inode_lock_nested(h_inode, AuLsc_I_PARENT);
3243 +       }
3244 +       if (!wbr)
3245 +               err = au_wh_init(br, sb);
3246 +       else {
3247 +               wbr_wh_write_lock(wbr);
3248 +               err = au_wh_init(br, sb);
3249 +               wbr_wh_write_unlock(wbr);
3250 +       }
3251 +       if (hdir)
3252 +               au_hn_inode_unlock(hdir);
3253 +       else
3254 +               inode_unlock(h_inode);
3255 +       vfsub_mnt_drop_write(au_br_mnt(br));
3256 +       br->br_perm = old_perm;
3257 +
3258 +       if (!err && wbr && !au_br_writable(new_perm)) {
3259 +               au_kfree_rcu(wbr);
3260 +               br->br_wbr = NULL;
3261 +       }
3262 +
3263 +out:
3264 +       return err;
3265 +}
3266 +
3267 +static int au_wbr_init(struct au_branch *br, struct super_block *sb,
3268 +                      int perm)
3269 +{
3270 +       int err;
3271 +       struct kstatfs kst;
3272 +       struct au_wbr *wbr;
3273 +
3274 +       wbr = br->br_wbr;
3275 +       au_rw_init(&wbr->wbr_wh_rwsem);
3276 +       atomic_set(&wbr->wbr_wh_running, 0);
3277 +
3278 +       /*
3279 +        * a limit for rmdir/rename a dir
3280 +        * cf. AUFS_MAX_NAMELEN in include/uapi/linux/aufs_type.h
3281 +        */
3282 +       err = vfs_statfs(&br->br_path, &kst);
3283 +       if (unlikely(err))
3284 +               goto out;
3285 +       err = -EINVAL;
3286 +       if (kst.f_namelen >= NAME_MAX)
3287 +               err = au_br_init_wh(sb, br, perm);
3288 +       else
3289 +               pr_err("%pd(%s), unsupported namelen %ld\n",
3290 +                      au_br_dentry(br),
3291 +                      au_sbtype(au_br_dentry(br)->d_sb), kst.f_namelen);
3292 +
3293 +out:
3294 +       return err;
3295 +}
3296 +
3297 +/* initialize a new branch */
3298 +static int au_br_init(struct au_branch *br, struct super_block *sb,
3299 +                     struct au_opt_add *add)
3300 +{
3301 +       int err;
3302 +       struct au_branch *brbase;
3303 +       struct file *xf;
3304 +       struct inode *h_inode;
3305 +
3306 +       err = 0;
3307 +       br->br_perm = add->perm;
3308 +       br->br_path = add->path; /* set first, path_get() later */
3309 +       spin_lock_init(&br->br_dykey_lock);
3310 +       au_lcnt_init(&br->br_nfiles, /*release*/NULL);
3311 +       au_lcnt_init(&br->br_count, /*release*/NULL);
3312 +       br->br_id = au_new_br_id(sb);
3313 +       AuDebugOn(br->br_id < 0);
3314 +
3315 +       /* always, regardless the given option */
3316 +       err = au_dr_br_init(sb, br, &add->path);
3317 +       if (unlikely(err))
3318 +               goto out_err;
3319 +
3320 +       if (au_br_writable(add->perm)) {
3321 +               err = au_wbr_init(br, sb, add->perm);
3322 +               if (unlikely(err))
3323 +                       goto out_err;
3324 +       }
3325 +
3326 +       if (au_opt_test(au_mntflags(sb), XINO)) {
3327 +               brbase = au_sbr(sb, 0);
3328 +               xf = au_xino_file(brbase->br_xino, /*idx*/-1);
3329 +               AuDebugOn(!xf);
3330 +               h_inode = d_inode(add->path.dentry);
3331 +               err = au_xino_init_br(sb, br, h_inode->i_ino, &xf->f_path);
3332 +               if (unlikely(err)) {
3333 +                       AuDebugOn(au_xino_file(br->br_xino, /*idx*/-1));
3334 +                       goto out_err;
3335 +               }
3336 +       }
3337 +
3338 +       sysaufs_br_init(br);
3339 +       path_get(&br->br_path);
3340 +       goto out; /* success */
3341 +
3342 +out_err:
3343 +       memset(&br->br_path, 0, sizeof(br->br_path));
3344 +out:
3345 +       return err;
3346 +}
3347 +
3348 +static void au_br_do_add_brp(struct au_sbinfo *sbinfo, aufs_bindex_t bindex,
3349 +                            struct au_branch *br, aufs_bindex_t bbot,
3350 +                            aufs_bindex_t amount)
3351 +{
3352 +       struct au_branch **brp;
3353 +
3354 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3355 +
3356 +       brp = sbinfo->si_branch + bindex;
3357 +       memmove(brp + 1, brp, sizeof(*brp) * amount);
3358 +       *brp = br;
3359 +       sbinfo->si_bbot++;
3360 +       if (unlikely(bbot < 0))
3361 +               sbinfo->si_bbot = 0;
3362 +}
3363 +
3364 +static void au_br_do_add_hdp(struct au_dinfo *dinfo, aufs_bindex_t bindex,
3365 +                            aufs_bindex_t bbot, aufs_bindex_t amount)
3366 +{
3367 +       struct au_hdentry *hdp;
3368 +
3369 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3370 +
3371 +       hdp = au_hdentry(dinfo, bindex);
3372 +       memmove(hdp + 1, hdp, sizeof(*hdp) * amount);
3373 +       au_h_dentry_init(hdp);
3374 +       dinfo->di_bbot++;
3375 +       if (unlikely(bbot < 0))
3376 +               dinfo->di_btop = 0;
3377 +}
3378 +
3379 +static void au_br_do_add_hip(struct au_iinfo *iinfo, aufs_bindex_t bindex,
3380 +                            aufs_bindex_t bbot, aufs_bindex_t amount)
3381 +{
3382 +       struct au_hinode *hip;
3383 +
3384 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3385 +
3386 +       hip = au_hinode(iinfo, bindex);
3387 +       memmove(hip + 1, hip, sizeof(*hip) * amount);
3388 +       au_hinode_init(hip);
3389 +       iinfo->ii_bbot++;
3390 +       if (unlikely(bbot < 0))
3391 +               iinfo->ii_btop = 0;
3392 +}
3393 +
3394 +static void au_br_do_add(struct super_block *sb, struct au_branch *br,
3395 +                        aufs_bindex_t bindex)
3396 +{
3397 +       struct dentry *root, *h_dentry;
3398 +       struct inode *root_inode, *h_inode;
3399 +       aufs_bindex_t bbot, amount;
3400 +
3401 +       root = sb->s_root;
3402 +       root_inode = d_inode(root);
3403 +       bbot = au_sbbot(sb);
3404 +       amount = bbot + 1 - bindex;
3405 +       h_dentry = au_br_dentry(br);
3406 +       au_sbilist_lock();
3407 +       au_br_do_add_brp(au_sbi(sb), bindex, br, bbot, amount);
3408 +       au_br_do_add_hdp(au_di(root), bindex, bbot, amount);
3409 +       au_br_do_add_hip(au_ii(root_inode), bindex, bbot, amount);
3410 +       au_set_h_dptr(root, bindex, dget(h_dentry));
3411 +       h_inode = d_inode(h_dentry);
3412 +       au_set_h_iptr(root_inode, bindex, au_igrab(h_inode), /*flags*/0);
3413 +       au_sbilist_unlock();
3414 +}
3415 +
3416 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount)
3417 +{
3418 +       int err;
3419 +       aufs_bindex_t bbot, add_bindex;
3420 +       struct dentry *root, *h_dentry;
3421 +       struct inode *root_inode;
3422 +       struct au_branch *add_branch;
3423 +
3424 +       root = sb->s_root;
3425 +       root_inode = d_inode(root);
3426 +       IMustLock(root_inode);
3427 +       IiMustWriteLock(root_inode);
3428 +       err = test_add(sb, add, remount);
3429 +       if (unlikely(err < 0))
3430 +               goto out;
3431 +       if (err) {
3432 +               err = 0;
3433 +               goto out; /* success */
3434 +       }
3435 +
3436 +       bbot = au_sbbot(sb);
3437 +       add_branch = au_br_alloc(sb, bbot + 2, add->perm);
3438 +       err = PTR_ERR(add_branch);
3439 +       if (IS_ERR(add_branch))
3440 +               goto out;
3441 +
3442 +       err = au_br_init(add_branch, sb, add);
3443 +       if (unlikely(err)) {
3444 +               au_br_do_free(add_branch);
3445 +               goto out;
3446 +       }
3447 +
3448 +       add_bindex = add->bindex;
3449 +       sysaufs_brs_del(sb, add_bindex);        /* remove successors */
3450 +       au_br_do_add(sb, add_branch, add_bindex);
3451 +       sysaufs_brs_add(sb, add_bindex);        /* append successors */
3452 +       dbgaufs_brs_add(sb, add_bindex, /*topdown*/0);  /* rename successors */
3453 +
3454 +       h_dentry = add->path.dentry;
3455 +       if (!add_bindex) {
3456 +               au_cpup_attr_all(root_inode, /*force*/1);
3457 +               sb->s_maxbytes = h_dentry->d_sb->s_maxbytes;
3458 +       } else
3459 +               au_add_nlink(root_inode, d_inode(h_dentry));
3460 +
3461 +out:
3462 +       return err;
3463 +}
3464 +
3465 +/* ---------------------------------------------------------------------- */
3466 +
3467 +static unsigned long long au_farray_cb(struct super_block *sb, void *a,
3468 +                                      unsigned long long max __maybe_unused,
3469 +                                      void *arg)
3470 +{
3471 +       unsigned long long n;
3472 +       struct file **p, *f;
3473 +       struct hlist_bl_head *files;
3474 +       struct hlist_bl_node *pos;
3475 +       struct au_finfo *finfo;
3476 +
3477 +       n = 0;
3478 +       p = a;
3479 +       files = &au_sbi(sb)->si_files;
3480 +       hlist_bl_lock(files);
3481 +       hlist_bl_for_each_entry(finfo, pos, files, fi_hlist) {
3482 +               f = finfo->fi_file;
3483 +               if (file_count(f)
3484 +                   && !special_file(file_inode(f)->i_mode)) {
3485 +                       get_file(f);
3486 +                       *p++ = f;
3487 +                       n++;
3488 +                       AuDebugOn(n > max);
3489 +               }
3490 +       }
3491 +       hlist_bl_unlock(files);
3492 +
3493 +       return n;
3494 +}
3495 +
3496 +static struct file **au_farray_alloc(struct super_block *sb,
3497 +                                    unsigned long long *max)
3498 +{
3499 +       struct au_sbinfo *sbi;
3500 +
3501 +       sbi = au_sbi(sb);
3502 +       *max = au_lcnt_read(&sbi->si_nfiles, /*do_rev*/1);
3503 +       return au_array_alloc(max, au_farray_cb, sb, /*arg*/NULL);
3504 +}
3505 +
3506 +static void au_farray_free(struct file **a, unsigned long long max)
3507 +{
3508 +       unsigned long long ull;
3509 +
3510 +       for (ull = 0; ull < max; ull++)
3511 +               if (a[ull])
3512 +                       fput(a[ull]);
3513 +       kvfree(a);
3514 +}
3515 +
3516 +/* ---------------------------------------------------------------------- */
3517 +
3518 +/*
3519 + * delete a branch
3520 + */
3521 +
3522 +/* to show the line number, do not make it inlined function */
3523 +#define AuVerbose(do_info, fmt, ...) do { \
3524 +       if (do_info) \
3525 +               pr_info(fmt, ##__VA_ARGS__); \
3526 +} while (0)
3527 +
3528 +static int au_test_ibusy(struct inode *inode, aufs_bindex_t btop,
3529 +                        aufs_bindex_t bbot)
3530 +{
3531 +       return (inode && !S_ISDIR(inode->i_mode)) || btop == bbot;
3532 +}
3533 +
3534 +static int au_test_dbusy(struct dentry *dentry, aufs_bindex_t btop,
3535 +                        aufs_bindex_t bbot)
3536 +{
3537 +       return au_test_ibusy(d_inode(dentry), btop, bbot);
3538 +}
3539 +
3540 +/*
3541 + * test if the branch is deletable or not.
3542 + */
3543 +static int test_dentry_busy(struct dentry *root, aufs_bindex_t bindex,
3544 +                           unsigned int sigen, const unsigned int verbose)
3545 +{
3546 +       int err, i, j, ndentry;
3547 +       aufs_bindex_t btop, bbot;
3548 +       struct au_dcsub_pages dpages;
3549 +       struct au_dpage *dpage;
3550 +       struct dentry *d;
3551 +
3552 +       err = au_dpages_init(&dpages, GFP_NOFS);
3553 +       if (unlikely(err))
3554 +               goto out;
3555 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
3556 +       if (unlikely(err))
3557 +               goto out_dpages;
3558 +
3559 +       for (i = 0; !err && i < dpages.ndpage; i++) {
3560 +               dpage = dpages.dpages + i;
3561 +               ndentry = dpage->ndentry;
3562 +               for (j = 0; !err && j < ndentry; j++) {
3563 +                       d = dpage->dentries[j];
3564 +                       AuDebugOn(au_dcount(d) <= 0);
3565 +                       if (!au_digen_test(d, sigen)) {
3566 +                               di_read_lock_child(d, AuLock_IR);
3567 +                               if (unlikely(au_dbrange_test(d))) {
3568 +                                       di_read_unlock(d, AuLock_IR);
3569 +                                       continue;
3570 +                               }
3571 +                       } else {
3572 +                               di_write_lock_child(d);
3573 +                               if (unlikely(au_dbrange_test(d))) {
3574 +                                       di_write_unlock(d);
3575 +                                       continue;
3576 +                               }
3577 +                               err = au_reval_dpath(d, sigen);
3578 +                               if (!err)
3579 +                                       di_downgrade_lock(d, AuLock_IR);
3580 +                               else {
3581 +                                       di_write_unlock(d);
3582 +                                       break;
3583 +                               }
3584 +                       }
3585 +
3586 +                       /* AuDbgDentry(d); */
3587 +                       btop = au_dbtop(d);
3588 +                       bbot = au_dbbot(d);
3589 +                       if (btop <= bindex
3590 +                           && bindex <= bbot
3591 +                           && au_h_dptr(d, bindex)
3592 +                           && au_test_dbusy(d, btop, bbot)) {
3593 +                               err = -EBUSY;
3594 +                               AuVerbose(verbose, "busy %pd\n", d);
3595 +                               AuDbgDentry(d);
3596 +                       }
3597 +                       di_read_unlock(d, AuLock_IR);
3598 +               }
3599 +       }
3600 +
3601 +out_dpages:
3602 +       au_dpages_free(&dpages);
3603 +out:
3604 +       return err;
3605 +}
3606 +
3607 +static int test_inode_busy(struct super_block *sb, aufs_bindex_t bindex,
3608 +                          unsigned int sigen, const unsigned int verbose)
3609 +{
3610 +       int err;
3611 +       unsigned long long max, ull;
3612 +       struct inode *i, **array;
3613 +       aufs_bindex_t btop, bbot;
3614 +
3615 +       array = au_iarray_alloc(sb, &max);
3616 +       err = PTR_ERR(array);
3617 +       if (IS_ERR(array))
3618 +               goto out;
3619 +
3620 +       err = 0;
3621 +       AuDbg("b%d\n", bindex);
3622 +       for (ull = 0; !err && ull < max; ull++) {
3623 +               i = array[ull];
3624 +               if (unlikely(!i))
3625 +                       break;
3626 +               if (i->i_ino == AUFS_ROOT_INO)
3627 +                       continue;
3628 +
3629 +               /* AuDbgInode(i); */
3630 +               if (au_iigen(i, NULL) == sigen)
3631 +                       ii_read_lock_child(i);
3632 +               else {
3633 +                       ii_write_lock_child(i);
3634 +                       err = au_refresh_hinode_self(i);
3635 +                       au_iigen_dec(i);
3636 +                       if (!err)
3637 +                               ii_downgrade_lock(i);
3638 +                       else {
3639 +                               ii_write_unlock(i);
3640 +                               break;
3641 +                       }
3642 +               }
3643 +
3644 +               btop = au_ibtop(i);
3645 +               bbot = au_ibbot(i);
3646 +               if (btop <= bindex
3647 +                   && bindex <= bbot
3648 +                   && au_h_iptr(i, bindex)
3649 +                   && au_test_ibusy(i, btop, bbot)) {
3650 +                       err = -EBUSY;
3651 +                       AuVerbose(verbose, "busy i%lu\n", i->i_ino);
3652 +                       AuDbgInode(i);
3653 +               }
3654 +               ii_read_unlock(i);
3655 +       }
3656 +       au_iarray_free(array, max);
3657 +
3658 +out:
3659 +       return err;
3660 +}
3661 +
3662 +static int test_children_busy(struct dentry *root, aufs_bindex_t bindex,
3663 +                             const unsigned int verbose)
3664 +{
3665 +       int err;
3666 +       unsigned int sigen;
3667 +
3668 +       sigen = au_sigen(root->d_sb);
3669 +       DiMustNoWaiters(root);
3670 +       IiMustNoWaiters(d_inode(root));
3671 +       di_write_unlock(root);
3672 +       err = test_dentry_busy(root, bindex, sigen, verbose);
3673 +       if (!err)
3674 +               err = test_inode_busy(root->d_sb, bindex, sigen, verbose);
3675 +       di_write_lock_child(root); /* aufs_write_lock() calls ..._child() */
3676 +
3677 +       return err;
3678 +}
3679 +
3680 +static int test_dir_busy(struct file *file, aufs_bindex_t br_id,
3681 +                        struct file **to_free, int *idx)
3682 +{
3683 +       int err;
3684 +       unsigned char matched, root;
3685 +       aufs_bindex_t bindex, bbot;
3686 +       struct au_fidir *fidir;
3687 +       struct au_hfile *hfile;
3688 +
3689 +       err = 0;
3690 +       root = IS_ROOT(file->f_path.dentry);
3691 +       if (root) {
3692 +               get_file(file);
3693 +               to_free[*idx] = file;
3694 +               (*idx)++;
3695 +               goto out;
3696 +       }
3697 +
3698 +       matched = 0;
3699 +       fidir = au_fi(file)->fi_hdir;
3700 +       AuDebugOn(!fidir);
3701 +       bbot = au_fbbot_dir(file);
3702 +       for (bindex = au_fbtop(file); bindex <= bbot; bindex++) {
3703 +               hfile = fidir->fd_hfile + bindex;
3704 +               if (!hfile->hf_file)
3705 +                       continue;
3706 +
3707 +               if (hfile->hf_br->br_id == br_id) {
3708 +                       matched = 1;
3709 +                       break;
3710 +               }
3711 +       }
3712 +       if (matched)
3713 +               err = -EBUSY;
3714 +
3715 +out:
3716 +       return err;
3717 +}
3718 +
3719 +static int test_file_busy(struct super_block *sb, aufs_bindex_t br_id,
3720 +                         struct file **to_free, int opened)
3721 +{
3722 +       int err, idx;
3723 +       unsigned long long ull, max;
3724 +       aufs_bindex_t btop;
3725 +       struct file *file, **array;
3726 +       struct dentry *root;
3727 +       struct au_hfile *hfile;
3728 +
3729 +       array = au_farray_alloc(sb, &max);
3730 +       err = PTR_ERR(array);
3731 +       if (IS_ERR(array))
3732 +               goto out;
3733 +
3734 +       err = 0;
3735 +       idx = 0;
3736 +       root = sb->s_root;
3737 +       di_write_unlock(root);
3738 +       for (ull = 0; ull < max; ull++) {
3739 +               file = array[ull];
3740 +               if (unlikely(!file))
3741 +                       break;
3742 +
3743 +               /* AuDbg("%pD\n", file); */
3744 +               fi_read_lock(file);
3745 +               btop = au_fbtop(file);
3746 +               if (!d_is_dir(file->f_path.dentry)) {
3747 +                       hfile = &au_fi(file)->fi_htop;
3748 +                       if (hfile->hf_br->br_id == br_id)
3749 +                               err = -EBUSY;
3750 +               } else
3751 +                       err = test_dir_busy(file, br_id, to_free, &idx);
3752 +               fi_read_unlock(file);
3753 +               if (unlikely(err))
3754 +                       break;
3755 +       }
3756 +       di_write_lock_child(root);
3757 +       au_farray_free(array, max);
3758 +       AuDebugOn(idx > opened);
3759 +
3760 +out:
3761 +       return err;
3762 +}
3763 +
3764 +static void br_del_file(struct file **to_free, unsigned long long opened,
3765 +                       aufs_bindex_t br_id)
3766 +{
3767 +       unsigned long long ull;
3768 +       aufs_bindex_t bindex, btop, bbot, bfound;
3769 +       struct file *file;
3770 +       struct au_fidir *fidir;
3771 +       struct au_hfile *hfile;
3772 +
3773 +       for (ull = 0; ull < opened; ull++) {
3774 +               file = to_free[ull];
3775 +               if (unlikely(!file))
3776 +                       break;
3777 +
3778 +               /* AuDbg("%pD\n", file); */
3779 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
3780 +               bfound = -1;
3781 +               fidir = au_fi(file)->fi_hdir;
3782 +               AuDebugOn(!fidir);
3783 +               fi_write_lock(file);
3784 +               btop = au_fbtop(file);
3785 +               bbot = au_fbbot_dir(file);
3786 +               for (bindex = btop; bindex <= bbot; bindex++) {
3787 +                       hfile = fidir->fd_hfile + bindex;
3788 +                       if (!hfile->hf_file)
3789 +                               continue;
3790 +
3791 +                       if (hfile->hf_br->br_id == br_id) {
3792 +                               bfound = bindex;
3793 +                               break;
3794 +                       }
3795 +               }
3796 +               AuDebugOn(bfound < 0);
3797 +               au_set_h_fptr(file, bfound, NULL);
3798 +               if (bfound == btop) {
3799 +                       for (btop++; btop <= bbot; btop++)
3800 +                               if (au_hf_dir(file, btop)) {
3801 +                                       au_set_fbtop(file, btop);
3802 +                                       break;
3803 +                               }
3804 +               }
3805 +               fi_write_unlock(file);
3806 +       }
3807 +}
3808 +
3809 +static void au_br_do_del_brp(struct au_sbinfo *sbinfo,
3810 +                            const aufs_bindex_t bindex,
3811 +                            const aufs_bindex_t bbot)
3812 +{
3813 +       struct au_branch **brp, **p;
3814 +
3815 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3816 +
3817 +       brp = sbinfo->si_branch + bindex;
3818 +       if (bindex < bbot)
3819 +               memmove(brp, brp + 1, sizeof(*brp) * (bbot - bindex));
3820 +       sbinfo->si_branch[0 + bbot] = NULL;
3821 +       sbinfo->si_bbot--;
3822 +
3823 +       p = au_krealloc(sbinfo->si_branch, sizeof(*p) * bbot, AuGFP_SBILIST,
3824 +                       /*may_shrink*/1);
3825 +       if (p)
3826 +               sbinfo->si_branch = p;
3827 +       /* harmless error */
3828 +}
3829 +
3830 +static void au_br_do_del_hdp(struct au_dinfo *dinfo, const aufs_bindex_t bindex,
3831 +                            const aufs_bindex_t bbot)
3832 +{
3833 +       struct au_hdentry *hdp, *p;
3834 +
3835 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3836 +
3837 +       hdp = au_hdentry(dinfo, bindex);
3838 +       if (bindex < bbot)
3839 +               memmove(hdp, hdp + 1, sizeof(*hdp) * (bbot - bindex));
3840 +       /* au_h_dentry_init(au_hdentry(dinfo, bbot); */
3841 +       dinfo->di_bbot--;
3842 +
3843 +       p = au_krealloc(dinfo->di_hdentry, sizeof(*p) * bbot, AuGFP_SBILIST,
3844 +                       /*may_shrink*/1);
3845 +       if (p)
3846 +               dinfo->di_hdentry = p;
3847 +       /* harmless error */
3848 +}
3849 +
3850 +static void au_br_do_del_hip(struct au_iinfo *iinfo, const aufs_bindex_t bindex,
3851 +                            const aufs_bindex_t bbot)
3852 +{
3853 +       struct au_hinode *hip, *p;
3854 +
3855 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3856 +
3857 +       hip = au_hinode(iinfo, bindex);
3858 +       if (bindex < bbot)
3859 +               memmove(hip, hip + 1, sizeof(*hip) * (bbot - bindex));
3860 +       /* au_hinode_init(au_hinode(iinfo, bbot)); */
3861 +       iinfo->ii_bbot--;
3862 +
3863 +       p = au_krealloc(iinfo->ii_hinode, sizeof(*p) * bbot, AuGFP_SBILIST,
3864 +                       /*may_shrink*/1);
3865 +       if (p)
3866 +               iinfo->ii_hinode = p;
3867 +       /* harmless error */
3868 +}
3869 +
3870 +static void au_br_do_del(struct super_block *sb, aufs_bindex_t bindex,
3871 +                        struct au_branch *br)
3872 +{
3873 +       aufs_bindex_t bbot;
3874 +       struct au_sbinfo *sbinfo;
3875 +       struct dentry *root, *h_root;
3876 +       struct inode *inode, *h_inode;
3877 +       struct au_hinode *hinode;
3878 +
3879 +       SiMustWriteLock(sb);
3880 +
3881 +       root = sb->s_root;
3882 +       inode = d_inode(root);
3883 +       sbinfo = au_sbi(sb);
3884 +       bbot = sbinfo->si_bbot;
3885 +
3886 +       h_root = au_h_dptr(root, bindex);
3887 +       hinode = au_hi(inode, bindex);
3888 +       h_inode = au_igrab(hinode->hi_inode);
3889 +       au_hiput(hinode);
3890 +
3891 +       au_sbilist_lock();
3892 +       au_br_do_del_brp(sbinfo, bindex, bbot);
3893 +       au_br_do_del_hdp(au_di(root), bindex, bbot);
3894 +       au_br_do_del_hip(au_ii(inode), bindex, bbot);
3895 +       au_sbilist_unlock();
3896 +
3897 +       /* ignore an error */
3898 +       au_dr_br_fin(sb, br); /* always, regardless the mount option */
3899 +
3900 +       dput(h_root);
3901 +       iput(h_inode);
3902 +       au_br_do_free(br);
3903 +}
3904 +
3905 +static unsigned long long empty_cb(struct super_block *sb, void *array,
3906 +                                  unsigned long long max, void *arg)
3907 +{
3908 +       return max;
3909 +}
3910 +
3911 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount)
3912 +{
3913 +       int err, rerr, i;
3914 +       unsigned long long opened;
3915 +       unsigned int mnt_flags;
3916 +       aufs_bindex_t bindex, bbot, br_id;
3917 +       unsigned char do_wh, verbose;
3918 +       struct au_branch *br;
3919 +       struct au_wbr *wbr;
3920 +       struct dentry *root;
3921 +       struct file **to_free;
3922 +
3923 +       err = 0;
3924 +       opened = 0;
3925 +       to_free = NULL;
3926 +       root = sb->s_root;
3927 +       bindex = au_find_dbindex(root, del->h_path.dentry);
3928 +       if (bindex < 0) {
3929 +               if (remount)
3930 +                       goto out; /* success */
3931 +               err = -ENOENT;
3932 +               pr_err("%s no such branch\n", del->pathname);
3933 +               goto out;
3934 +       }
3935 +       AuDbg("bindex b%d\n", bindex);
3936 +
3937 +       err = -EBUSY;
3938 +       mnt_flags = au_mntflags(sb);
3939 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3940 +       bbot = au_sbbot(sb);
3941 +       if (unlikely(!bbot)) {
3942 +               AuVerbose(verbose, "no more branches left\n");
3943 +               goto out;
3944 +       }
3945 +
3946 +       br = au_sbr(sb, bindex);
3947 +       AuDebugOn(!path_equal(&br->br_path, &del->h_path));
3948 +       if (unlikely(au_lcnt_read(&br->br_count, /*do_rev*/1))) {
3949 +               AuVerbose(verbose, "br %pd2 is busy now\n", del->h_path.dentry);
3950 +               goto out;
3951 +       }
3952 +
3953 +       br_id = br->br_id;
3954 +       opened = au_lcnt_read(&br->br_nfiles, /*do_rev*/1);
3955 +       if (unlikely(opened)) {
3956 +               to_free = au_array_alloc(&opened, empty_cb, sb, NULL);
3957 +               err = PTR_ERR(to_free);
3958 +               if (IS_ERR(to_free))
3959 +                       goto out;
3960 +
3961 +               err = test_file_busy(sb, br_id, to_free, opened);
3962 +               if (unlikely(err)) {
3963 +                       AuVerbose(verbose, "%llu file(s) opened\n", opened);
3964 +                       goto out;
3965 +               }
3966 +       }
3967 +
3968 +       wbr = br->br_wbr;
3969 +       do_wh = wbr && (wbr->wbr_whbase || wbr->wbr_plink || wbr->wbr_orph);
3970 +       if (do_wh) {
3971 +               /* instead of WbrWhMustWriteLock(wbr) */
3972 +               SiMustWriteLock(sb);
3973 +               for (i = 0; i < AuBrWh_Last; i++) {
3974 +                       dput(wbr->wbr_wh[i]);
3975 +                       wbr->wbr_wh[i] = NULL;
3976 +               }
3977 +       }
3978 +
3979 +       err = test_children_busy(root, bindex, verbose);
3980 +       if (unlikely(err)) {
3981 +               if (do_wh)
3982 +                       goto out_wh;
3983 +               goto out;
3984 +       }
3985 +
3986 +       err = 0;
3987 +       if (to_free) {
3988 +               /*
3989 +                * now we confirmed the branch is deletable.
3990 +                * let's free the remaining opened dirs on the branch.
3991 +                */
3992 +               di_write_unlock(root);
3993 +               br_del_file(to_free, opened, br_id);
3994 +               di_write_lock_child(root);
3995 +       }
3996 +
3997 +       sysaufs_brs_del(sb, bindex);    /* remove successors */
3998 +       dbgaufs_xino_del(br);           /* remove one */
3999 +       au_br_do_del(sb, bindex, br);
4000 +       sysaufs_brs_add(sb, bindex);    /* append successors */
4001 +       dbgaufs_brs_add(sb, bindex, /*topdown*/1);      /* rename successors */
4002 +
4003 +       if (!bindex) {
4004 +               au_cpup_attr_all(d_inode(root), /*force*/1);
4005 +               sb->s_maxbytes = au_sbr_sb(sb, 0)->s_maxbytes;
4006 +       } else
4007 +               au_sub_nlink(d_inode(root), d_inode(del->h_path.dentry));
4008 +       if (au_opt_test(mnt_flags, PLINK))
4009 +               au_plink_half_refresh(sb, br_id);
4010 +
4011 +       goto out; /* success */
4012 +
4013 +out_wh:
4014 +       /* revert */
4015 +       rerr = au_br_init_wh(sb, br, br->br_perm);
4016 +       if (rerr)
4017 +               pr_warn("failed re-creating base whiteout, %s. (%d)\n",
4018 +                       del->pathname, rerr);
4019 +out:
4020 +       if (to_free)
4021 +               au_farray_free(to_free, opened);
4022 +       return err;
4023 +}
4024 +
4025 +/* ---------------------------------------------------------------------- */
4026 +
4027 +static int au_ibusy(struct super_block *sb, struct aufs_ibusy __user *arg)
4028 +{
4029 +       int err;
4030 +       aufs_bindex_t btop, bbot;
4031 +       struct aufs_ibusy ibusy;
4032 +       struct inode *inode, *h_inode;
4033 +
4034 +       err = -EPERM;
4035 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
4036 +               goto out;
4037 +
4038 +       err = copy_from_user(&ibusy, arg, sizeof(ibusy));
4039 +       if (!err)
4040 +               /* VERIFY_WRITE */
4041 +               err = !access_ok(&arg->h_ino, sizeof(arg->h_ino));
4042 +       if (unlikely(err)) {
4043 +               err = -EFAULT;
4044 +               AuTraceErr(err);
4045 +               goto out;
4046 +       }
4047 +
4048 +       err = -EINVAL;
4049 +       si_read_lock(sb, AuLock_FLUSH);
4050 +       if (unlikely(ibusy.bindex < 0 || ibusy.bindex > au_sbbot(sb)))
4051 +               goto out_unlock;
4052 +
4053 +       err = 0;
4054 +       ibusy.h_ino = 0; /* invalid */
4055 +       inode = ilookup(sb, ibusy.ino);
4056 +       if (!inode
4057 +           || inode->i_ino == AUFS_ROOT_INO
4058 +           || au_is_bad_inode(inode))
4059 +               goto out_unlock;
4060 +
4061 +       ii_read_lock_child(inode);
4062 +       btop = au_ibtop(inode);
4063 +       bbot = au_ibbot(inode);
4064 +       if (btop <= ibusy.bindex && ibusy.bindex <= bbot) {
4065 +               h_inode = au_h_iptr(inode, ibusy.bindex);
4066 +               if (h_inode && au_test_ibusy(inode, btop, bbot))
4067 +                       ibusy.h_ino = h_inode->i_ino;
4068 +       }
4069 +       ii_read_unlock(inode);
4070 +       iput(inode);
4071 +
4072 +out_unlock:
4073 +       si_read_unlock(sb);
4074 +       if (!err) {
4075 +               err = __put_user(ibusy.h_ino, &arg->h_ino);
4076 +               if (unlikely(err)) {
4077 +                       err = -EFAULT;
4078 +                       AuTraceErr(err);
4079 +               }
4080 +       }
4081 +out:
4082 +       return err;
4083 +}
4084 +
4085 +long au_ibusy_ioctl(struct file *file, unsigned long arg)
4086 +{
4087 +       return au_ibusy(file->f_path.dentry->d_sb, (void __user *)arg);
4088 +}
4089 +
4090 +#ifdef CONFIG_COMPAT
4091 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg)
4092 +{
4093 +       return au_ibusy(file->f_path.dentry->d_sb, compat_ptr(arg));
4094 +}
4095 +#endif
4096 +
4097 +/* ---------------------------------------------------------------------- */
4098 +
4099 +/*
4100 + * change a branch permission
4101 + */
4102 +
4103 +static void au_warn_ima(void)
4104 +{
4105 +#ifdef CONFIG_IMA
4106 +       /* since it doesn't support mark_files_ro() */
4107 +       AuWarn1("RW -> RO makes IMA to produce wrong message\n");
4108 +#endif
4109 +}
4110 +
4111 +static int do_need_sigen_inc(int a, int b)
4112 +{
4113 +       return au_br_whable(a) && !au_br_whable(b);
4114 +}
4115 +
4116 +static int need_sigen_inc(int old, int new)
4117 +{
4118 +       return do_need_sigen_inc(old, new)
4119 +               || do_need_sigen_inc(new, old);
4120 +}
4121 +
4122 +static int au_br_mod_files_ro(struct super_block *sb, aufs_bindex_t bindex)
4123 +{
4124 +       int err, do_warn;
4125 +       unsigned int mnt_flags;
4126 +       unsigned long long ull, max;
4127 +       aufs_bindex_t br_id;
4128 +       unsigned char verbose, writer;
4129 +       struct file *file, *hf, **array;
4130 +       struct au_hfile *hfile;
4131 +       struct inode *h_inode;
4132 +
4133 +       mnt_flags = au_mntflags(sb);
4134 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
4135 +
4136 +       array = au_farray_alloc(sb, &max);
4137 +       err = PTR_ERR(array);
4138 +       if (IS_ERR(array))
4139 +               goto out;
4140 +
4141 +       do_warn = 0;
4142 +       br_id = au_sbr_id(sb, bindex);
4143 +       for (ull = 0; ull < max; ull++) {
4144 +               file = array[ull];
4145 +               if (unlikely(!file))
4146 +                       break;
4147 +
4148 +               /* AuDbg("%pD\n", file); */
4149 +               fi_read_lock(file);
4150 +               if (unlikely(au_test_mmapped(file))) {
4151 +                       err = -EBUSY;
4152 +                       AuVerbose(verbose, "mmapped %pD\n", file);
4153 +                       AuDbgFile(file);
4154 +                       FiMustNoWaiters(file);
4155 +                       fi_read_unlock(file);
4156 +                       goto out_array;
4157 +               }
4158 +
4159 +               hfile = &au_fi(file)->fi_htop;
4160 +               hf = hfile->hf_file;
4161 +               if (!d_is_reg(file->f_path.dentry)
4162 +                   || !(file->f_mode & FMODE_WRITE)
4163 +                   || hfile->hf_br->br_id != br_id
4164 +                   || !(hf->f_mode & FMODE_WRITE))
4165 +                       array[ull] = NULL;
4166 +               else {
4167 +                       do_warn = 1;
4168 +                       get_file(file);
4169 +               }
4170 +
4171 +               FiMustNoWaiters(file);
4172 +               fi_read_unlock(file);
4173 +               fput(file);
4174 +       }
4175 +
4176 +       err = 0;
4177 +       if (do_warn)
4178 +               au_warn_ima();
4179 +
4180 +       for (ull = 0; ull < max; ull++) {
4181 +               file = array[ull];
4182 +               if (!file)
4183 +                       continue;
4184 +
4185 +               /* todo: already flushed? */
4186 +               /*
4187 +                * fs/super.c:mark_files_ro() is gone, but aufs keeps its
4188 +                * approach which resets f_mode and calls mnt_drop_write() and
4189 +                * file_release_write() for each file, because the branch
4190 +                * attribute in aufs world is totally different from the native
4191 +                * fs rw/ro mode.
4192 +               */
4193 +               /* fi_read_lock(file); */
4194 +               hfile = &au_fi(file)->fi_htop;
4195 +               hf = hfile->hf_file;
4196 +               /* fi_read_unlock(file); */
4197 +               spin_lock(&hf->f_lock);
4198 +               writer = !!(hf->f_mode & FMODE_WRITER);
4199 +               hf->f_mode &= ~(FMODE_WRITE | FMODE_WRITER);
4200 +               spin_unlock(&hf->f_lock);
4201 +               if (writer) {
4202 +                       h_inode = file_inode(hf);
4203 +                       if (hf->f_mode & FMODE_READ)
4204 +                               i_readcount_inc(h_inode);
4205 +                       put_write_access(h_inode);
4206 +                       __mnt_drop_write(hf->f_path.mnt);
4207 +               }
4208 +       }
4209 +
4210 +out_array:
4211 +       au_farray_free(array, max);
4212 +out:
4213 +       AuTraceErr(err);
4214 +       return err;
4215 +}
4216 +
4217 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4218 +             int *do_refresh)
4219 +{
4220 +       int err, rerr;
4221 +       aufs_bindex_t bindex;
4222 +       struct dentry *root;
4223 +       struct au_branch *br;
4224 +       struct au_br_fhsm *bf;
4225 +
4226 +       root = sb->s_root;
4227 +       bindex = au_find_dbindex(root, mod->h_root);
4228 +       if (bindex < 0) {
4229 +               if (remount)
4230 +                       return 0; /* success */
4231 +               err = -ENOENT;
4232 +               pr_err("%s no such branch\n", mod->path);
4233 +               goto out;
4234 +       }
4235 +       AuDbg("bindex b%d\n", bindex);
4236 +
4237 +       err = test_br(d_inode(mod->h_root), mod->perm, mod->path);
4238 +       if (unlikely(err))
4239 +               goto out;
4240 +
4241 +       br = au_sbr(sb, bindex);
4242 +       AuDebugOn(mod->h_root != au_br_dentry(br));
4243 +       if (br->br_perm == mod->perm)
4244 +               return 0; /* success */
4245 +
4246 +       /* pre-allocate for non-fhsm --> fhsm */
4247 +       bf = NULL;
4248 +       if (!au_br_fhsm(br->br_perm) && au_br_fhsm(mod->perm)) {
4249 +               err = au_fhsm_br_alloc(br);
4250 +               if (unlikely(err))
4251 +                       goto out;
4252 +               bf = br->br_fhsm;
4253 +               br->br_fhsm = NULL;
4254 +       }
4255 +
4256 +       if (au_br_writable(br->br_perm)) {
4257 +               /* remove whiteout base */
4258 +               err = au_br_init_wh(sb, br, mod->perm);
4259 +               if (unlikely(err))
4260 +                       goto out_bf;
4261 +
4262 +               if (!au_br_writable(mod->perm)) {
4263 +                       /* rw --> ro, file might be mmapped */
4264 +                       DiMustNoWaiters(root);
4265 +                       IiMustNoWaiters(d_inode(root));
4266 +                       di_write_unlock(root);
4267 +                       err = au_br_mod_files_ro(sb, bindex);
4268 +                       /* aufs_write_lock() calls ..._child() */
4269 +                       di_write_lock_child(root);
4270 +
4271 +                       if (unlikely(err)) {
4272 +                               rerr = -ENOMEM;
4273 +                               br->br_wbr = kzalloc(sizeof(*br->br_wbr),
4274 +                                                    GFP_NOFS);
4275 +                               if (br->br_wbr)
4276 +                                       rerr = au_wbr_init(br, sb, br->br_perm);
4277 +                               if (unlikely(rerr)) {
4278 +                                       AuIOErr("nested error %d (%d)\n",
4279 +                                               rerr, err);
4280 +                                       br->br_perm = mod->perm;
4281 +                               }
4282 +                       }
4283 +               }
4284 +       } else if (au_br_writable(mod->perm)) {
4285 +               /* ro --> rw */
4286 +               err = -ENOMEM;
4287 +               br->br_wbr = kzalloc(sizeof(*br->br_wbr), GFP_NOFS);
4288 +               if (br->br_wbr) {
4289 +                       err = au_wbr_init(br, sb, mod->perm);
4290 +                       if (unlikely(err)) {
4291 +                               au_kfree_rcu(br->br_wbr);
4292 +                               br->br_wbr = NULL;
4293 +                       }
4294 +               }
4295 +       }
4296 +       if (unlikely(err))
4297 +               goto out_bf;
4298 +
4299 +       if (au_br_fhsm(br->br_perm)) {
4300 +               if (!au_br_fhsm(mod->perm)) {
4301 +                       /* fhsm --> non-fhsm */
4302 +                       au_br_fhsm_fin(br->br_fhsm);
4303 +                       au_kfree_rcu(br->br_fhsm);
4304 +                       br->br_fhsm = NULL;
4305 +               }
4306 +       } else if (au_br_fhsm(mod->perm))
4307 +               /* non-fhsm --> fhsm */
4308 +               br->br_fhsm = bf;
4309 +
4310 +       *do_refresh |= need_sigen_inc(br->br_perm, mod->perm);
4311 +       br->br_perm = mod->perm;
4312 +       goto out; /* success */
4313 +
4314 +out_bf:
4315 +       au_kfree_try_rcu(bf);
4316 +out:
4317 +       AuTraceErr(err);
4318 +       return err;
4319 +}
4320 +
4321 +/* ---------------------------------------------------------------------- */
4322 +
4323 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs)
4324 +{
4325 +       int err;
4326 +       struct kstatfs kstfs;
4327 +
4328 +       err = vfs_statfs(&br->br_path, &kstfs);
4329 +       if (!err) {
4330 +               stfs->f_blocks = kstfs.f_blocks;
4331 +               stfs->f_bavail = kstfs.f_bavail;
4332 +               stfs->f_files = kstfs.f_files;
4333 +               stfs->f_ffree = kstfs.f_ffree;
4334 +       }
4335 +
4336 +       return err;
4337 +}
4338 diff -urN /usr/share/empty/fs/aufs/branch.h linux/fs/aufs/branch.h
4339 --- /usr/share/empty/fs/aufs/branch.h   1970-01-01 01:00:00.000000000 +0100
4340 +++ linux/fs/aufs/branch.h      2021-12-03 15:40:58.233313963 +0100
4341 @@ -0,0 +1,375 @@
4342 +/* SPDX-License-Identifier: GPL-2.0 */
4343 +/*
4344 + * Copyright (C) 2005-2021 Junjiro R. Okajima
4345 + *
4346 + * This program, aufs is free software; you can redistribute it and/or modify
4347 + * it under the terms of the GNU General Public License as published by
4348 + * the Free Software Foundation; either version 2 of the License, or
4349 + * (at your option) any later version.
4350 + *
4351 + * This program is distributed in the hope that it will be useful,
4352 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4353 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4354 + * GNU General Public License for more details.
4355 + *
4356 + * You should have received a copy of the GNU General Public License
4357 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4358 + */
4359 +
4360 +/*
4361 + * branch filesystems and xino for them
4362 + */
4363 +
4364 +#ifndef __AUFS_BRANCH_H__
4365 +#define __AUFS_BRANCH_H__
4366 +
4367 +#ifdef __KERNEL__
4368 +
4369 +#include <linux/mount.h>
4370 +#include "dirren.h"
4371 +#include "dynop.h"
4372 +#include "lcnt.h"
4373 +#include "rwsem.h"
4374 +#include "super.h"
4375 +
4376 +/* ---------------------------------------------------------------------- */
4377 +
4378 +/* a xino file */
4379 +struct au_xino {
4380 +       struct file             **xi_file;
4381 +       unsigned int            xi_nfile;
4382 +
4383 +       struct {
4384 +               spinlock_t              spin;
4385 +               ino_t                   *array;
4386 +               int                     total;
4387 +               /* reserved for future use */
4388 +               /* unsigned long        *bitmap; */
4389 +               wait_queue_head_t       wqh;
4390 +       } xi_nondir;
4391 +
4392 +       struct mutex            xi_mtx; /* protects xi_file array */
4393 +       struct hlist_bl_head    xi_writing;
4394 +
4395 +       atomic_t                xi_truncating;
4396 +
4397 +       struct kref             xi_kref;
4398 +};
4399 +
4400 +/* File-based Hierarchical Storage Management */
4401 +struct au_br_fhsm {
4402 +#ifdef CONFIG_AUFS_FHSM
4403 +       struct mutex            bf_lock;
4404 +       unsigned long           bf_jiffy;
4405 +       struct aufs_stfs        bf_stfs;
4406 +       int                     bf_readable;
4407 +#endif
4408 +};
4409 +
4410 +/* members for writable branch only */
4411 +enum {AuBrWh_BASE, AuBrWh_PLINK, AuBrWh_ORPH, AuBrWh_Last};
4412 +struct au_wbr {
4413 +       struct au_rwsem         wbr_wh_rwsem;
4414 +       struct dentry           *wbr_wh[AuBrWh_Last];
4415 +       atomic_t                wbr_wh_running;
4416 +#define wbr_whbase             wbr_wh[AuBrWh_BASE]     /* whiteout base */
4417 +#define wbr_plink              wbr_wh[AuBrWh_PLINK]    /* pseudo-link dir */
4418 +#define wbr_orph               wbr_wh[AuBrWh_ORPH]     /* dir for orphans */
4419 +
4420 +       /* mfs mode */
4421 +       unsigned long long      wbr_bytes;
4422 +};
4423 +
4424 +/* ext2 has 3 types of operations at least, ext3 has 4 */
4425 +#define AuBrDynOp (AuDyLast * 4)
4426 +
4427 +#ifdef CONFIG_AUFS_HFSNOTIFY
4428 +/* support for asynchronous destruction */
4429 +struct au_br_hfsnotify {
4430 +       struct fsnotify_group   *hfsn_group;
4431 +};
4432 +#endif
4433 +
4434 +/* sysfs entries */
4435 +struct au_brsysfs {
4436 +       char                    name[16];
4437 +       struct attribute        attr;
4438 +};
4439 +
4440 +enum {
4441 +       AuBrSysfs_BR,
4442 +       AuBrSysfs_BRID,
4443 +       AuBrSysfs_Last
4444 +};
4445 +
4446 +/* protected by superblock rwsem */
4447 +struct au_branch {
4448 +       struct au_xino          *br_xino;
4449 +
4450 +       aufs_bindex_t           br_id;
4451 +
4452 +       int                     br_perm;
4453 +       struct path             br_path;
4454 +       spinlock_t              br_dykey_lock;
4455 +       struct au_dykey         *br_dykey[AuBrDynOp];
4456 +       au_lcnt_t               br_nfiles;      /* opened files */
4457 +       au_lcnt_t               br_count;       /* in-use for other */
4458 +
4459 +       struct au_wbr           *br_wbr;
4460 +       struct au_br_fhsm       *br_fhsm;
4461 +
4462 +#ifdef CONFIG_AUFS_HFSNOTIFY
4463 +       struct au_br_hfsnotify  *br_hfsn;
4464 +#endif
4465 +
4466 +#ifdef CONFIG_SYSFS
4467 +       /* entries under sysfs per mount-point */
4468 +       struct au_brsysfs       br_sysfs[AuBrSysfs_Last];
4469 +#endif
4470 +
4471 +#ifdef CONFIG_DEBUG_FS
4472 +       struct dentry            *br_dbgaufs; /* xino */
4473 +#endif
4474 +
4475 +       struct au_dr_br         br_dirren;
4476 +};
4477 +
4478 +/* ---------------------------------------------------------------------- */
4479 +
4480 +static inline struct vfsmount *au_br_mnt(struct au_branch *br)
4481 +{
4482 +       return br->br_path.mnt;
4483 +}
4484 +
4485 +static inline struct dentry *au_br_dentry(struct au_branch *br)
4486 +{
4487 +       return br->br_path.dentry;
4488 +}
4489 +
4490 +static inline struct user_namespace *au_br_userns(struct au_branch *br)
4491 +{
4492 +       return mnt_user_ns(br->br_path.mnt);
4493 +}
4494 +
4495 +static inline struct super_block *au_br_sb(struct au_branch *br)
4496 +{
4497 +       return au_br_mnt(br)->mnt_sb;
4498 +}
4499 +
4500 +static inline int au_br_rdonly(struct au_branch *br)
4501 +{
4502 +       return (sb_rdonly(au_br_sb(br))
4503 +               || !au_br_writable(br->br_perm))
4504 +               ? -EROFS : 0;
4505 +}
4506 +
4507 +static inline int au_br_hnotifyable(int brperm __maybe_unused)
4508 +{
4509 +#ifdef CONFIG_AUFS_HNOTIFY
4510 +       return !(brperm & AuBrPerm_RR);
4511 +#else
4512 +       return 0;
4513 +#endif
4514 +}
4515 +
4516 +static inline int au_br_test_oflag(int oflag, struct au_branch *br)
4517 +{
4518 +       int err, exec_flag;
4519 +
4520 +       err = 0;
4521 +       exec_flag = oflag & __FMODE_EXEC;
4522 +       if (unlikely(exec_flag && path_noexec(&br->br_path)))
4523 +               err = -EACCES;
4524 +
4525 +       return err;
4526 +}
4527 +
4528 +static inline void au_xino_get(struct au_branch *br)
4529 +{
4530 +       struct au_xino *xi;
4531 +
4532 +       xi = br->br_xino;
4533 +       if (xi)
4534 +               kref_get(&xi->xi_kref);
4535 +}
4536 +
4537 +static inline int au_xino_count(struct au_branch *br)
4538 +{
4539 +       int v;
4540 +       struct au_xino *xi;
4541 +
4542 +       v = 0;
4543 +       xi = br->br_xino;
4544 +       if (xi)
4545 +               v = kref_read(&xi->xi_kref);
4546 +
4547 +       return v;
4548 +}
4549 +
4550 +/* ---------------------------------------------------------------------- */
4551 +
4552 +/* branch.c */
4553 +struct au_sbinfo;
4554 +void au_br_free(struct au_sbinfo *sinfo);
4555 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id);
4556 +struct au_opt_add;
4557 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount);
4558 +struct au_opt_del;
4559 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount);
4560 +long au_ibusy_ioctl(struct file *file, unsigned long arg);
4561 +#ifdef CONFIG_COMPAT
4562 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg);
4563 +#endif
4564 +struct au_opt_mod;
4565 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4566 +             int *do_refresh);
4567 +struct aufs_stfs;
4568 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs);
4569 +
4570 +/* xino.c */
4571 +static const loff_t au_loff_max = LLONG_MAX;
4572 +
4573 +aufs_bindex_t au_xi_root(struct super_block *sb, struct dentry *dentry);
4574 +struct file *au_xino_create(struct super_block *sb, char *fpath, int silent,
4575 +                           int wbrtop);
4576 +struct file *au_xino_create2(struct super_block *sb, struct path *base,
4577 +                            struct file *copy_src);
4578 +struct au_xi_new {
4579 +       struct au_xino *xi;     /* switch between xino and xigen */
4580 +       int idx;
4581 +       struct path *base;
4582 +       struct file *copy_src;
4583 +};
4584 +struct file *au_xi_new(struct super_block *sb, struct au_xi_new *xinew);
4585 +
4586 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4587 +                ino_t *ino);
4588 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4589 +                 ino_t ino);
4590 +ssize_t xino_fread(struct file *file, void *buf, size_t size, loff_t *pos);
4591 +ssize_t xino_fwrite(struct file *file, void *buf, size_t size, loff_t *pos);
4592 +
4593 +int au_xib_trunc(struct super_block *sb);
4594 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex, int idx_begin);
4595 +
4596 +struct au_xino *au_xino_alloc(unsigned int nfile);
4597 +int au_xino_put(struct au_branch *br);
4598 +struct file *au_xino_file1(struct au_xino *xi);
4599 +
4600 +struct au_opt_xino;
4601 +void au_xino_clr(struct super_block *sb);
4602 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xiopt, int remount);
4603 +struct file *au_xino_def(struct super_block *sb);
4604 +int au_xino_init_br(struct super_block *sb, struct au_branch *br, ino_t hino,
4605 +                   struct path *base);
4606 +
4607 +ino_t au_xino_new_ino(struct super_block *sb);
4608 +void au_xino_delete_inode(struct inode *inode, const int unlinked);
4609 +
4610 +void au_xinondir_leave(struct super_block *sb, aufs_bindex_t bindex,
4611 +                      ino_t h_ino, int idx);
4612 +int au_xinondir_enter(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4613 +                     int *idx);
4614 +
4615 +int au_xino_path(struct seq_file *seq, struct file *file);
4616 +
4617 +/* ---------------------------------------------------------------------- */
4618 +
4619 +/* @idx is signed to accept -1 meaning the first file */
4620 +static inline struct file *au_xino_file(struct au_xino *xi, int idx)
4621 +{
4622 +       struct file *file;
4623 +
4624 +       file = NULL;
4625 +       if (!xi)
4626 +               goto out;
4627 +
4628 +       if (idx >= 0) {
4629 +               if (idx < xi->xi_nfile)
4630 +                       file = xi->xi_file[idx];
4631 +       } else
4632 +               file = au_xino_file1(xi);
4633 +
4634 +out:
4635 +       return file;
4636 +}
4637 +
4638 +/* ---------------------------------------------------------------------- */
4639 +
4640 +/* Superblock to branch */
4641 +static inline
4642 +aufs_bindex_t au_sbr_id(struct super_block *sb, aufs_bindex_t bindex)
4643 +{
4644 +       return au_sbr(sb, bindex)->br_id;
4645 +}
4646 +
4647 +static inline
4648 +struct vfsmount *au_sbr_mnt(struct super_block *sb, aufs_bindex_t bindex)
4649 +{
4650 +       return au_br_mnt(au_sbr(sb, bindex));
4651 +}
4652 +
4653 +static inline
4654 +struct user_namespace *au_sbr_userns(struct super_block *sb, aufs_bindex_t bindex)
4655 +{
4656 +       return au_br_userns(au_sbr(sb, bindex));
4657 +}
4658 +
4659 +static inline
4660 +struct super_block *au_sbr_sb(struct super_block *sb, aufs_bindex_t bindex)
4661 +{
4662 +       return au_br_sb(au_sbr(sb, bindex));
4663 +}
4664 +
4665 +static inline int au_sbr_perm(struct super_block *sb, aufs_bindex_t bindex)
4666 +{
4667 +       return au_sbr(sb, bindex)->br_perm;
4668 +}
4669 +
4670 +static inline int au_sbr_whable(struct super_block *sb, aufs_bindex_t bindex)
4671 +{
4672 +       return au_br_whable(au_sbr_perm(sb, bindex));
4673 +}
4674 +
4675 +/* ---------------------------------------------------------------------- */
4676 +
4677 +#define wbr_wh_read_lock(wbr)  au_rw_read_lock(&(wbr)->wbr_wh_rwsem)
4678 +#define wbr_wh_write_lock(wbr) au_rw_write_lock(&(wbr)->wbr_wh_rwsem)
4679 +#define wbr_wh_read_trylock(wbr)       au_rw_read_trylock(&(wbr)->wbr_wh_rwsem)
4680 +#define wbr_wh_write_trylock(wbr) au_rw_write_trylock(&(wbr)->wbr_wh_rwsem)
4681 +/*
4682 +#define wbr_wh_read_trylock_nested(wbr) \
4683 +       au_rw_read_trylock_nested(&(wbr)->wbr_wh_rwsem)
4684 +#define wbr_wh_write_trylock_nested(wbr) \
4685 +       au_rw_write_trylock_nested(&(wbr)->wbr_wh_rwsem)
4686 +*/
4687 +
4688 +#define wbr_wh_read_unlock(wbr)        au_rw_read_unlock(&(wbr)->wbr_wh_rwsem)
4689 +#define wbr_wh_write_unlock(wbr)       au_rw_write_unlock(&(wbr)->wbr_wh_rwsem)
4690 +#define wbr_wh_downgrade_lock(wbr)     au_rw_dgrade_lock(&(wbr)->wbr_wh_rwsem)
4691 +
4692 +#define WbrWhMustNoWaiters(wbr)        AuRwMustNoWaiters(&(wbr)->wbr_wh_rwsem)
4693 +#define WbrWhMustAnyLock(wbr)  AuRwMustAnyLock(&(wbr)->wbr_wh_rwsem)
4694 +#define WbrWhMustWriteLock(wbr)        AuRwMustWriteLock(&(wbr)->wbr_wh_rwsem)
4695 +
4696 +/* ---------------------------------------------------------------------- */
4697 +
4698 +#ifdef CONFIG_AUFS_FHSM
4699 +static inline void au_br_fhsm_init(struct au_br_fhsm *brfhsm)
4700 +{
4701 +       mutex_init(&brfhsm->bf_lock);
4702 +       brfhsm->bf_jiffy = 0;
4703 +       brfhsm->bf_readable = 0;
4704 +}
4705 +
4706 +static inline void au_br_fhsm_fin(struct au_br_fhsm *brfhsm)
4707 +{
4708 +       mutex_destroy(&brfhsm->bf_lock);
4709 +}
4710 +#else
4711 +AuStubVoid(au_br_fhsm_init, struct au_br_fhsm *brfhsm)
4712 +AuStubVoid(au_br_fhsm_fin, struct au_br_fhsm *brfhsm)
4713 +#endif
4714 +
4715 +#endif /* __KERNEL__ */
4716 +#endif /* __AUFS_BRANCH_H__ */
4717 diff -urN /usr/share/empty/fs/aufs/conf.mk linux/fs/aufs/conf.mk
4718 --- /usr/share/empty/fs/aufs/conf.mk    1970-01-01 01:00:00.000000000 +0100
4719 +++ linux/fs/aufs/conf.mk       2021-12-03 15:38:59.933313976 +0100
4720 @@ -0,0 +1,40 @@
4721 +# SPDX-License-Identifier: GPL-2.0
4722 +
4723 +AuConfStr = CONFIG_AUFS_FS=${CONFIG_AUFS_FS}
4724 +
4725 +define AuConf
4726 +ifdef ${1}
4727 +AuConfStr += ${1}=${${1}}
4728 +endif
4729 +endef
4730 +
4731 +AuConfAll = BRANCH_MAX_127 BRANCH_MAX_511 BRANCH_MAX_1023 BRANCH_MAX_32767 \
4732 +       SBILIST \
4733 +       HNOTIFY HFSNOTIFY \
4734 +       EXPORT INO_T_64 \
4735 +       XATTR \
4736 +       FHSM \
4737 +       RDU \
4738 +       DIRREN \
4739 +       SHWH \
4740 +       BR_RAMFS \
4741 +       BR_FUSE POLL \
4742 +       BR_HFSPLUS \
4743 +       BDEV_LOOP \
4744 +       DEBUG MAGIC_SYSRQ
4745 +$(foreach i, ${AuConfAll}, \
4746 +       $(eval $(call AuConf,CONFIG_AUFS_${i})))
4747 +
4748 +AuConfName = ${obj}/conf.str
4749 +${AuConfName}.tmp: FORCE
4750 +       @echo ${AuConfStr} | tr ' ' '\n' | sed -e 's/^/"/' -e 's/$$/\\n"/' > $@
4751 +${AuConfName}: ${AuConfName}.tmp
4752 +       @diff -q $< $@ > /dev/null 2>&1 || { \
4753 +       echo '  GEN    ' $@; \
4754 +       cp -p $< $@; \
4755 +       }
4756 +FORCE:
4757 +clean-files += ${AuConfName} ${AuConfName}.tmp
4758 +${obj}/sysfs.o: ${AuConfName}
4759 +
4760 +-include ${srctree}/${src}/conf_priv.mk
4761 diff -urN /usr/share/empty/fs/aufs/cpup.c linux/fs/aufs/cpup.c
4762 --- /usr/share/empty/fs/aufs/cpup.c     1970-01-01 01:00:00.000000000 +0100
4763 +++ linux/fs/aufs/cpup.c        2021-12-03 15:40:58.233313963 +0100
4764 @@ -0,0 +1,1459 @@
4765 +// SPDX-License-Identifier: GPL-2.0
4766 +/*
4767 + * Copyright (C) 2005-2021 Junjiro R. Okajima
4768 + *
4769 + * This program, aufs is free software; you can redistribute it and/or modify
4770 + * it under the terms of the GNU General Public License as published by
4771 + * the Free Software Foundation; either version 2 of the License, or
4772 + * (at your option) any later version.
4773 + *
4774 + * This program is distributed in the hope that it will be useful,
4775 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4776 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4777 + * GNU General Public License for more details.
4778 + *
4779 + * You should have received a copy of the GNU General Public License
4780 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4781 + */
4782 +
4783 +/*
4784 + * copy-up functions, see wbr_policy.c for copy-down
4785 + */
4786 +
4787 +#include <linux/fs_stack.h>
4788 +#include <linux/mm.h>
4789 +#include <linux/task_work.h>
4790 +#include "aufs.h"
4791 +
4792 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags)
4793 +{
4794 +       const unsigned int mask = S_DEAD | S_SWAPFILE | S_PRIVATE
4795 +               | S_NOATIME | S_NOCMTIME | S_AUTOMOUNT;
4796 +
4797 +       BUILD_BUG_ON(sizeof(iflags) != sizeof(dst->i_flags));
4798 +
4799 +       dst->i_flags |= iflags & ~mask;
4800 +       if (au_test_fs_notime(dst->i_sb))
4801 +               dst->i_flags |= S_NOATIME | S_NOCMTIME;
4802 +}
4803 +
4804 +void au_cpup_attr_timesizes(struct inode *inode)
4805 +{
4806 +       struct inode *h_inode;
4807 +
4808 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4809 +       fsstack_copy_attr_times(inode, h_inode);
4810 +       fsstack_copy_inode_size(inode, h_inode);
4811 +}
4812 +
4813 +void au_cpup_attr_nlink(struct inode *inode, int force)
4814 +{
4815 +       struct inode *h_inode;
4816 +       struct super_block *sb;
4817 +       aufs_bindex_t bindex, bbot;
4818 +
4819 +       sb = inode->i_sb;
4820 +       bindex = au_ibtop(inode);
4821 +       h_inode = au_h_iptr(inode, bindex);
4822 +       if (!force
4823 +           && !S_ISDIR(h_inode->i_mode)
4824 +           && au_opt_test(au_mntflags(sb), PLINK)
4825 +           && au_plink_test(inode))
4826 +               return;
4827 +
4828 +       /*
4829 +        * 0 can happen in revalidating.
4830 +        * h_inode->i_mutex may not be held here, but it is harmless since once
4831 +        * i_nlink reaches 0, it will never become positive except O_TMPFILE
4832 +        * case.
4833 +        * todo: O_TMPFILE+linkat(AT_SYMLINK_FOLLOW) bypassing aufs may cause
4834 +        *       the incorrect link count.
4835 +        */
4836 +       set_nlink(inode, h_inode->i_nlink);
4837 +
4838 +       /*
4839 +        * fewer nlink makes find(1) noisy, but larger nlink doesn't.
4840 +        * it may includes whplink directory.
4841 +        */
4842 +       if (S_ISDIR(h_inode->i_mode)) {
4843 +               bbot = au_ibbot(inode);
4844 +               for (bindex++; bindex <= bbot; bindex++) {
4845 +                       h_inode = au_h_iptr(inode, bindex);
4846 +                       if (h_inode)
4847 +                               au_add_nlink(inode, h_inode);
4848 +               }
4849 +       }
4850 +}
4851 +
4852 +void au_cpup_attr_changeable(struct inode *inode)
4853 +{
4854 +       struct inode *h_inode;
4855 +
4856 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4857 +       inode->i_mode = h_inode->i_mode;
4858 +       inode->i_uid = h_inode->i_uid;
4859 +       inode->i_gid = h_inode->i_gid;
4860 +       au_cpup_attr_timesizes(inode);
4861 +       au_cpup_attr_flags(inode, h_inode->i_flags);
4862 +}
4863 +
4864 +void au_cpup_igen(struct inode *inode, struct inode *h_inode)
4865 +{
4866 +       struct au_iinfo *iinfo = au_ii(inode);
4867 +
4868 +       IiMustWriteLock(inode);
4869 +
4870 +       iinfo->ii_higen = h_inode->i_generation;
4871 +       iinfo->ii_hsb1 = h_inode->i_sb;
4872 +}
4873 +
4874 +void au_cpup_attr_all(struct inode *inode, int force)
4875 +{
4876 +       struct inode *h_inode;
4877 +
4878 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4879 +       au_cpup_attr_changeable(inode);
4880 +       if (inode->i_nlink > 0)
4881 +               au_cpup_attr_nlink(inode, force);
4882 +       inode->i_rdev = h_inode->i_rdev;
4883 +       inode->i_blkbits = h_inode->i_blkbits;
4884 +       au_cpup_igen(inode, h_inode);
4885 +}
4886 +
4887 +/* ---------------------------------------------------------------------- */
4888 +
4889 +/* Note: dt_dentry and dt_h_dentry are not dget/dput-ed */
4890 +
4891 +/* keep the timestamps of the parent dir when cpup */
4892 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
4893 +                   struct path *h_path)
4894 +{
4895 +       struct inode *h_inode;
4896 +
4897 +       dt->dt_dentry = dentry;
4898 +       dt->dt_h_path = *h_path;
4899 +       h_inode = d_inode(h_path->dentry);
4900 +       dt->dt_atime = h_inode->i_atime;
4901 +       dt->dt_mtime = h_inode->i_mtime;
4902 +       /* smp_mb(); */
4903 +}
4904 +
4905 +void au_dtime_revert(struct au_dtime *dt)
4906 +{
4907 +       struct iattr attr;
4908 +       int err;
4909 +
4910 +       attr.ia_atime = dt->dt_atime;
4911 +       attr.ia_mtime = dt->dt_mtime;
4912 +       attr.ia_valid = ATTR_FORCE | ATTR_MTIME | ATTR_MTIME_SET
4913 +               | ATTR_ATIME | ATTR_ATIME_SET;
4914 +
4915 +       /* no delegation since this is a directory */
4916 +       err = vfsub_notify_change(&dt->dt_h_path, &attr, /*delegated*/NULL);
4917 +       if (unlikely(err))
4918 +               pr_warn("restoring timestamps failed(%d). ignored\n", err);
4919 +}
4920 +
4921 +/* ---------------------------------------------------------------------- */
4922 +
4923 +/* internal use only */
4924 +struct au_cpup_reg_attr {
4925 +       int             valid;
4926 +       struct kstat    st;
4927 +       unsigned int    iflags; /* inode->i_flags */
4928 +};
4929 +
4930 +static noinline_for_stack
4931 +int cpup_iattr(struct dentry *dst, aufs_bindex_t bindex, struct path *h_src,
4932 +              struct au_cpup_reg_attr *h_src_attr)
4933 +{
4934 +       int err, sbits, icex;
4935 +       unsigned int mnt_flags;
4936 +       unsigned char verbose;
4937 +       struct iattr ia;
4938 +       struct path h_path;
4939 +       struct inode *h_isrc, *h_idst;
4940 +       struct kstat *h_st;
4941 +       struct au_branch *br;
4942 +
4943 +       br = au_sbr(dst->d_sb, bindex);
4944 +       h_path.mnt = au_br_mnt(br);
4945 +       h_path.dentry = au_h_dptr(dst, bindex);
4946 +       h_idst = d_inode(h_path.dentry);
4947 +       h_isrc = d_inode(h_src->dentry);
4948 +       ia.ia_valid = ATTR_FORCE | ATTR_UID | ATTR_GID
4949 +               | ATTR_ATIME | ATTR_MTIME
4950 +               | ATTR_ATIME_SET | ATTR_MTIME_SET;
4951 +       if (h_src_attr && h_src_attr->valid) {
4952 +               h_st = &h_src_attr->st;
4953 +               ia.ia_uid = h_st->uid;
4954 +               ia.ia_gid = h_st->gid;
4955 +               ia.ia_atime = h_st->atime;
4956 +               ia.ia_mtime = h_st->mtime;
4957 +               if (h_idst->i_mode != h_st->mode
4958 +                   && !S_ISLNK(h_idst->i_mode)) {
4959 +                       ia.ia_valid |= ATTR_MODE;
4960 +                       ia.ia_mode = h_st->mode;
4961 +               }
4962 +               sbits = !!(h_st->mode & (S_ISUID | S_ISGID));
4963 +               au_cpup_attr_flags(h_idst, h_src_attr->iflags);
4964 +       } else {
4965 +               ia.ia_uid = h_isrc->i_uid;
4966 +               ia.ia_gid = h_isrc->i_gid;
4967 +               ia.ia_atime = h_isrc->i_atime;
4968 +               ia.ia_mtime = h_isrc->i_mtime;
4969 +               if (h_idst->i_mode != h_isrc->i_mode
4970 +                   && !S_ISLNK(h_idst->i_mode)) {
4971 +                       ia.ia_valid |= ATTR_MODE;
4972 +                       ia.ia_mode = h_isrc->i_mode;
4973 +               }
4974 +               sbits = !!(h_isrc->i_mode & (S_ISUID | S_ISGID));
4975 +               au_cpup_attr_flags(h_idst, h_isrc->i_flags);
4976 +       }
4977 +       /* no delegation since it is just created */
4978 +       err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4979 +
4980 +       /* is this nfs only? */
4981 +       if (!err && sbits && au_test_nfs(h_path.dentry->d_sb)) {
4982 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
4983 +               ia.ia_mode = h_isrc->i_mode;
4984 +               err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4985 +       }
4986 +
4987 +       icex = br->br_perm & AuBrAttr_ICEX;
4988 +       if (!err) {
4989 +               mnt_flags = au_mntflags(dst->d_sb);
4990 +               verbose = !!au_opt_test(mnt_flags, VERBOSE);
4991 +               err = au_cpup_xattr(&h_path, h_src, icex, verbose);
4992 +       }
4993 +
4994 +       return err;
4995 +}
4996 +
4997 +/* ---------------------------------------------------------------------- */
4998 +
4999 +static int au_do_copy_file(struct file *dst, struct file *src, loff_t len,
5000 +                          char *buf, unsigned long blksize)
5001 +{
5002 +       int err;
5003 +       size_t sz, rbytes, wbytes;
5004 +       unsigned char all_zero;
5005 +       char *p, *zp;
5006 +       struct inode *h_inode;
5007 +       /* reduce stack usage */
5008 +       struct iattr *ia;
5009 +
5010 +       zp = page_address(ZERO_PAGE(0));
5011 +       if (unlikely(!zp))
5012 +               return -ENOMEM; /* possible? */
5013 +
5014 +       err = 0;
5015 +       all_zero = 0;
5016 +       while (len) {
5017 +               AuDbg("len %lld\n", len);
5018 +               sz = blksize;
5019 +               if (len < blksize)
5020 +                       sz = len;
5021 +
5022 +               rbytes = 0;
5023 +               /* todo: signal_pending? */
5024 +               while (!rbytes || err == -EAGAIN || err == -EINTR) {
5025 +                       rbytes = vfsub_read_k(src, buf, sz, &src->f_pos);
5026 +                       err = rbytes;
5027 +               }
5028 +               if (unlikely(err < 0))
5029 +                       break;
5030 +
5031 +               all_zero = 0;
5032 +               if (len >= rbytes && rbytes == blksize)
5033 +                       all_zero = !memcmp(buf, zp, rbytes);
5034 +               if (!all_zero) {
5035 +                       wbytes = rbytes;
5036 +                       p = buf;
5037 +                       while (wbytes) {
5038 +                               size_t b;
5039 +
5040 +                               b = vfsub_write_k(dst, p, wbytes, &dst->f_pos);
5041 +                               err = b;
5042 +                               /* todo: signal_pending? */
5043 +                               if (unlikely(err == -EAGAIN || err == -EINTR))
5044 +                                       continue;
5045 +                               if (unlikely(err < 0))
5046 +                                       break;
5047 +                               wbytes -= b;
5048 +                               p += b;
5049 +                       }
5050 +                       if (unlikely(err < 0))
5051 +                               break;
5052 +               } else {
5053 +                       loff_t res;
5054 +
5055 +                       AuLabel(hole);
5056 +                       res = vfsub_llseek(dst, rbytes, SEEK_CUR);
5057 +                       err = res;
5058 +                       if (unlikely(res < 0))
5059 +                               break;
5060 +               }
5061 +               len -= rbytes;
5062 +               err = 0;
5063 +       }
5064 +
5065 +       /* the last block may be a hole */
5066 +       if (!err && all_zero) {
5067 +               AuLabel(last hole);
5068 +
5069 +               err = 1;
5070 +               if (au_test_nfs(dst->f_path.dentry->d_sb)) {
5071 +                       /* nfs requires this step to make last hole */
5072 +                       /* is this only nfs? */
5073 +                       do {
5074 +                               /* todo: signal_pending? */
5075 +                               err = vfsub_write_k(dst, "\0", 1, &dst->f_pos);
5076 +                       } while (err == -EAGAIN || err == -EINTR);
5077 +                       if (err == 1)
5078 +                               dst->f_pos--;
5079 +               }
5080 +
5081 +               if (err == 1) {
5082 +                       ia = (void *)buf;
5083 +                       ia->ia_size = dst->f_pos;
5084 +                       ia->ia_valid = ATTR_SIZE | ATTR_FILE;
5085 +                       ia->ia_file = dst;
5086 +                       h_inode = file_inode(dst);
5087 +                       inode_lock_nested(h_inode, AuLsc_I_CHILD2);
5088 +                       /* no delegation since it is just created */
5089 +                       err = vfsub_notify_change(&dst->f_path, ia,
5090 +                                                 /*delegated*/NULL);
5091 +                       inode_unlock(h_inode);
5092 +               }
5093 +       }
5094 +
5095 +       return err;
5096 +}
5097 +
5098 +int au_copy_file(struct file *dst, struct file *src, loff_t len)
5099 +{
5100 +       int err;
5101 +       unsigned long blksize;
5102 +       unsigned char do_kfree;
5103 +       char *buf;
5104 +       struct super_block *h_sb;
5105 +
5106 +       err = -ENOMEM;
5107 +       h_sb = file_inode(dst)->i_sb;
5108 +       blksize = h_sb->s_blocksize;
5109 +       if (!blksize || PAGE_SIZE < blksize)
5110 +               blksize = PAGE_SIZE;
5111 +       AuDbg("blksize %lu\n", blksize);
5112 +       do_kfree = (blksize != PAGE_SIZE && blksize >= sizeof(struct iattr *));
5113 +       if (do_kfree)
5114 +               buf = kmalloc(blksize, GFP_NOFS);
5115 +       else
5116 +               buf = (void *)__get_free_page(GFP_NOFS);
5117 +       if (unlikely(!buf))
5118 +               goto out;
5119 +
5120 +       if (len > (1 << 22))
5121 +               AuDbg("copying a large file %lld\n", (long long)len);
5122 +
5123 +       src->f_pos = 0;
5124 +       dst->f_pos = 0;
5125 +       err = au_do_copy_file(dst, src, len, buf, blksize);
5126 +       if (do_kfree) {
5127 +               AuDebugOn(!au_kfree_do_sz_test(blksize));
5128 +               au_kfree_do_rcu(buf);
5129 +       } else
5130 +               free_page((unsigned long)buf);
5131 +
5132 +out:
5133 +       return err;
5134 +}
5135 +
5136 +static int au_do_copy(struct file *dst, struct file *src, loff_t len)
5137 +{
5138 +       int err;
5139 +       struct super_block *h_src_sb;
5140 +       struct inode *h_src_inode;
5141 +
5142 +       h_src_inode = file_inode(src);
5143 +       h_src_sb = h_src_inode->i_sb;
5144 +
5145 +       /* XFS acquires inode_lock */
5146 +       if (!au_test_xfs(h_src_sb))
5147 +               err = au_copy_file(dst, src, len);
5148 +       else {
5149 +               inode_unlock_shared(h_src_inode);
5150 +               err = au_copy_file(dst, src, len);
5151 +               inode_lock_shared_nested(h_src_inode, AuLsc_I_CHILD);
5152 +       }
5153 +
5154 +       return err;
5155 +}
5156 +
5157 +static int au_clone_or_copy(struct file *dst, struct file *src, loff_t len)
5158 +{
5159 +       int err;
5160 +       loff_t lo;
5161 +       struct super_block *h_src_sb;
5162 +       struct inode *h_src_inode;
5163 +
5164 +       h_src_inode = file_inode(src);
5165 +       h_src_sb = h_src_inode->i_sb;
5166 +       if (h_src_sb != file_inode(dst)->i_sb
5167 +           || !dst->f_op->remap_file_range) {
5168 +               err = au_do_copy(dst, src, len);
5169 +               goto out;
5170 +       }
5171 +
5172 +       if (!au_test_nfs(h_src_sb)) {
5173 +               inode_unlock_shared(h_src_inode);
5174 +               lo = vfsub_clone_file_range(src, dst, len);
5175 +               inode_lock_shared_nested(h_src_inode, AuLsc_I_CHILD);
5176 +       } else
5177 +               lo = vfsub_clone_file_range(src, dst, len);
5178 +       if (lo == len) {
5179 +               err = 0;
5180 +               goto out; /* success */
5181 +       } else if (lo >= 0)
5182 +               /* todo: possible? */
5183 +               /* paritially succeeded */
5184 +               AuDbg("lo %lld, len %lld. Retrying.\n", lo, len);
5185 +       else if (lo != -EOPNOTSUPP) {
5186 +               /* older XFS has a condition in cloning */
5187 +               err = lo;
5188 +               goto out;
5189 +       }
5190 +
5191 +       /* the backend fs on NFS may not support cloning */
5192 +       err = au_do_copy(dst, src, len);
5193 +
5194 +out:
5195 +       AuTraceErr(err);
5196 +       return err;
5197 +}
5198 +
5199 +/*
5200 + * to support a sparse file which is opened with O_APPEND,
5201 + * we need to close the file.
5202 + */
5203 +static int au_cp_regular(struct au_cp_generic *cpg)
5204 +{
5205 +       int err, i;
5206 +       enum { SRC, DST };
5207 +       struct {
5208 +               aufs_bindex_t bindex;
5209 +               unsigned int flags;
5210 +               struct dentry *dentry;
5211 +               int force_wr;
5212 +               struct file *file;
5213 +       } *f, file[] = {
5214 +               {
5215 +                       .bindex = cpg->bsrc,
5216 +                       .flags = O_RDONLY | O_NOATIME | O_LARGEFILE,
5217 +               },
5218 +               {
5219 +                       .bindex = cpg->bdst,
5220 +                       .flags = O_WRONLY | O_NOATIME | O_LARGEFILE,
5221 +                       .force_wr = !!au_ftest_cpup(cpg->flags, RWDST),
5222 +               }
5223 +       };
5224 +       struct au_branch *br;
5225 +       struct super_block *sb, *h_src_sb;
5226 +       struct inode *h_src_inode;
5227 +       struct task_struct *tsk = current;
5228 +
5229 +       /* bsrc branch can be ro/rw. */
5230 +       sb = cpg->dentry->d_sb;
5231 +       f = file;
5232 +       for (i = 0; i < 2; i++, f++) {
5233 +               f->dentry = au_h_dptr(cpg->dentry, f->bindex);
5234 +               f->file = au_h_open(cpg->dentry, f->bindex, f->flags,
5235 +                                   /*file*/NULL, f->force_wr);
5236 +               if (IS_ERR(f->file)) {
5237 +                       err = PTR_ERR(f->file);
5238 +                       if (i == SRC)
5239 +                               goto out;
5240 +                       else
5241 +                               goto out_src;
5242 +               }
5243 +       }
5244 +
5245 +       /* try stopping to update while we copyup */
5246 +       h_src_inode = d_inode(file[SRC].dentry);
5247 +       h_src_sb = h_src_inode->i_sb;
5248 +       if (!au_test_nfs(h_src_sb))
5249 +               IMustLock(h_src_inode);
5250 +       err = au_clone_or_copy(file[DST].file, file[SRC].file, cpg->len);
5251 +
5252 +       /* i wonder if we had O_NO_DELAY_FPUT flag */
5253 +       if (tsk->flags & PF_KTHREAD)
5254 +               __fput_sync(file[DST].file);
5255 +       else {
5256 +               /* it happened actually */
5257 +               fput(file[DST].file);
5258 +               /*
5259 +                * too bad.
5260 +                * we have to call both since we don't know which place the file
5261 +                * was added to.
5262 +                */
5263 +               task_work_run();
5264 +               flush_delayed_fput();
5265 +       }
5266 +       br = au_sbr(sb, file[DST].bindex);
5267 +       au_lcnt_dec(&br->br_nfiles);
5268 +
5269 +out_src:
5270 +       fput(file[SRC].file);
5271 +       br = au_sbr(sb, file[SRC].bindex);
5272 +       au_lcnt_dec(&br->br_nfiles);
5273 +out:
5274 +       return err;
5275 +}
5276 +
5277 +static int au_do_cpup_regular(struct au_cp_generic *cpg,
5278 +                             struct au_cpup_reg_attr *h_src_attr)
5279 +{
5280 +       int err, rerr;
5281 +       loff_t l;
5282 +       struct path h_path;
5283 +       struct inode *h_src_inode, *h_dst_inode;
5284 +
5285 +       err = 0;
5286 +       h_src_inode = au_h_iptr(d_inode(cpg->dentry), cpg->bsrc);
5287 +       l = i_size_read(h_src_inode);
5288 +       if (cpg->len == -1 || l < cpg->len)
5289 +               cpg->len = l;
5290 +       if (cpg->len) {
5291 +               /* try stopping to update while we are referencing */
5292 +               inode_lock_shared_nested(h_src_inode, AuLsc_I_CHILD);
5293 +               au_pin_hdir_unlock(cpg->pin);
5294 +
5295 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5296 +               h_path.mnt = au_sbr_mnt(cpg->dentry->d_sb, cpg->bsrc);
5297 +               h_src_attr->iflags = h_src_inode->i_flags;
5298 +               if (!au_test_nfs(h_src_inode->i_sb))
5299 +                       err = vfsub_getattr(&h_path, &h_src_attr->st);
5300 +               else {
5301 +                       inode_unlock_shared(h_src_inode);
5302 +                       err = vfsub_getattr(&h_path, &h_src_attr->st);
5303 +                       inode_lock_shared_nested(h_src_inode, AuLsc_I_CHILD);
5304 +               }
5305 +               if (unlikely(err)) {
5306 +                       inode_unlock_shared(h_src_inode);
5307 +                       goto out;
5308 +               }
5309 +               h_src_attr->valid = 1;
5310 +               if (!au_test_nfs(h_src_inode->i_sb)) {
5311 +                       err = au_cp_regular(cpg);
5312 +                       inode_unlock_shared(h_src_inode);
5313 +               } else {
5314 +                       inode_unlock_shared(h_src_inode);
5315 +                       err = au_cp_regular(cpg);
5316 +               }
5317 +               rerr = au_pin_hdir_relock(cpg->pin);
5318 +               if (!err && rerr)
5319 +                       err = rerr;
5320 +       }
5321 +       if (!err && (h_src_inode->i_state & I_LINKABLE)) {
5322 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bdst);
5323 +               h_dst_inode = d_inode(h_path.dentry);
5324 +               spin_lock(&h_dst_inode->i_lock);
5325 +               h_dst_inode->i_state |= I_LINKABLE;
5326 +               spin_unlock(&h_dst_inode->i_lock);
5327 +       }
5328 +
5329 +out:
5330 +       return err;
5331 +}
5332 +
5333 +static int au_do_cpup_symlink(struct path *h_path, struct dentry *h_src,
5334 +                             struct inode *h_dir)
5335 +{
5336 +       int err;
5337 +       DEFINE_DELAYED_CALL(done);
5338 +       const char *sym;
5339 +
5340 +       sym = vfs_get_link(h_src, &done);
5341 +       err = PTR_ERR(sym);
5342 +       if (IS_ERR(sym))
5343 +               goto out;
5344 +
5345 +       err = vfsub_symlink(h_dir, h_path, sym);
5346 +
5347 +out:
5348 +       do_delayed_call(&done);
5349 +       return err;
5350 +}
5351 +
5352 +/*
5353 + * regardless 'acl' option, reset all ACL.
5354 + * All ACL will be copied up later from the original entry on the lower branch.
5355 + */
5356 +static int au_reset_acl(struct inode *h_dir, struct path *h_path, umode_t mode)
5357 +{
5358 +       int err;
5359 +       struct dentry *h_dentry;
5360 +       struct inode *h_inode;
5361 +       struct user_namespace *h_userns;
5362 +
5363 +       h_userns = mnt_user_ns(h_path->mnt);
5364 +       h_dentry = h_path->dentry;
5365 +       h_inode = d_inode(h_dentry);
5366 +       /* forget_all_cached_acls(h_inode)); */
5367 +       err = vfsub_removexattr(h_userns, h_dentry, XATTR_NAME_POSIX_ACL_ACCESS);
5368 +       AuTraceErr(err);
5369 +       if (err == -EOPNOTSUPP)
5370 +               err = 0;
5371 +       if (!err)
5372 +               err = vfsub_acl_chmod(h_userns, h_inode, mode);
5373 +
5374 +       AuTraceErr(err);
5375 +       return err;
5376 +}
5377 +
5378 +static int au_do_cpup_dir(struct au_cp_generic *cpg, struct dentry *dst_parent,
5379 +                         struct inode *h_dir, struct path *h_path)
5380 +{
5381 +       int err;
5382 +       struct inode *dir, *inode;
5383 +       struct user_namespace *h_userns;
5384 +
5385 +       h_userns = mnt_user_ns(h_path->mnt);
5386 +       err = vfsub_removexattr(h_userns, h_path->dentry,
5387 +                               XATTR_NAME_POSIX_ACL_DEFAULT);
5388 +       AuTraceErr(err);
5389 +       if (err == -EOPNOTSUPP)
5390 +               err = 0;
5391 +       if (unlikely(err))
5392 +               goto out;
5393 +
5394 +       /*
5395 +        * strange behaviour from the users view,
5396 +        * particularly setattr case
5397 +        */
5398 +       dir = d_inode(dst_parent);
5399 +       if (au_ibtop(dir) == cpg->bdst)
5400 +               au_cpup_attr_nlink(dir, /*force*/1);
5401 +       inode = d_inode(cpg->dentry);
5402 +       au_cpup_attr_nlink(inode, /*force*/1);
5403 +
5404 +out:
5405 +       return err;
5406 +}
5407 +
5408 +static noinline_for_stack
5409 +int cpup_entry(struct au_cp_generic *cpg, struct dentry *dst_parent,
5410 +              struct au_cpup_reg_attr *h_src_attr)
5411 +{
5412 +       int err;
5413 +       umode_t mode;
5414 +       unsigned int mnt_flags;
5415 +       unsigned char isdir, isreg, force;
5416 +       const unsigned char do_dt = !!au_ftest_cpup(cpg->flags, DTIME);
5417 +       struct au_dtime dt;
5418 +       struct path h_path;
5419 +       struct dentry *h_src, *h_dst, *h_parent;
5420 +       struct inode *h_inode, *h_dir;
5421 +       struct super_block *sb;
5422 +
5423 +       /* bsrc branch can be ro/rw. */
5424 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5425 +       h_inode = d_inode(h_src);
5426 +       AuDebugOn(h_inode != au_h_iptr(d_inode(cpg->dentry), cpg->bsrc));
5427 +
5428 +       /* try stopping to be referenced while we are creating */
5429 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5430 +       if (au_ftest_cpup(cpg->flags, RENAME))
5431 +               AuDebugOn(strncmp(h_dst->d_name.name, AUFS_WH_PFX,
5432 +                                 AUFS_WH_PFX_LEN));
5433 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5434 +       h_dir = d_inode(h_parent);
5435 +       IMustLock(h_dir);
5436 +       AuDebugOn(h_parent != h_dst->d_parent);
5437 +
5438 +       sb = cpg->dentry->d_sb;
5439 +       h_path.mnt = au_sbr_mnt(sb, cpg->bdst);
5440 +       if (do_dt) {
5441 +               h_path.dentry = h_parent;
5442 +               au_dtime_store(&dt, dst_parent, &h_path);
5443 +       }
5444 +       h_path.dentry = h_dst;
5445 +
5446 +       isreg = 0;
5447 +       isdir = 0;
5448 +       mode = h_inode->i_mode;
5449 +       switch (mode & S_IFMT) {
5450 +       case S_IFREG:
5451 +               isreg = 1;
5452 +               err = vfsub_create(h_dir, &h_path, 0600, /*want_excl*/true);
5453 +               if (!err)
5454 +                       err = au_do_cpup_regular(cpg, h_src_attr);
5455 +               break;
5456 +       case S_IFDIR:
5457 +               isdir = 1;
5458 +               err = vfsub_mkdir(h_dir, &h_path, mode);
5459 +               if (!err)
5460 +                       err = au_do_cpup_dir(cpg, dst_parent, h_dir, &h_path);
5461 +               break;
5462 +       case S_IFLNK:
5463 +               err = au_do_cpup_symlink(&h_path, h_src, h_dir);
5464 +               break;
5465 +       case S_IFCHR:
5466 +       case S_IFBLK:
5467 +               AuDebugOn(!capable(CAP_MKNOD));
5468 +               fallthrough;
5469 +       case S_IFIFO:
5470 +       case S_IFSOCK:
5471 +               err = vfsub_mknod(h_dir, &h_path, mode, h_inode->i_rdev);
5472 +               break;
5473 +       default:
5474 +               AuIOErr("Unknown inode type 0%o\n", mode);
5475 +               err = -EIO;
5476 +       }
5477 +       if (!err)
5478 +               err = au_reset_acl(h_dir, &h_path, mode);
5479 +
5480 +       mnt_flags = au_mntflags(sb);
5481 +       if (!au_opt_test(mnt_flags, UDBA_NONE)
5482 +           && !isdir
5483 +           && au_opt_test(mnt_flags, XINO)
5484 +           && (h_inode->i_nlink == 1
5485 +               || (h_inode->i_state & I_LINKABLE))
5486 +           /* todo: unnecessary? */
5487 +           /* && d_inode(cpg->dentry)->i_nlink == 1 */
5488 +           && cpg->bdst < cpg->bsrc
5489 +           && !au_ftest_cpup(cpg->flags, KEEPLINO))
5490 +               au_xino_write(sb, cpg->bsrc, h_inode->i_ino, /*ino*/0);
5491 +               /* ignore this error */
5492 +
5493 +       if (!err) {
5494 +               force = 0;
5495 +               if (isreg) {
5496 +                       force = !!cpg->len;
5497 +                       if (cpg->len == -1)
5498 +                               force = !!i_size_read(h_inode);
5499 +               }
5500 +               au_fhsm_wrote(sb, cpg->bdst, force);
5501 +       }
5502 +
5503 +       if (do_dt)
5504 +               au_dtime_revert(&dt);
5505 +       return err;
5506 +}
5507 +
5508 +static int au_do_ren_after_cpup(struct au_cp_generic *cpg, struct path *h_path)
5509 +{
5510 +       int err;
5511 +       struct dentry *dentry, *h_dentry, *h_parent, *parent;
5512 +       struct path h_ppath;
5513 +       struct inode *h_dir;
5514 +       aufs_bindex_t bdst;
5515 +
5516 +       dentry = cpg->dentry;
5517 +       bdst = cpg->bdst;
5518 +       h_ppath.mnt = au_sbr_mnt(dentry->d_sb, bdst);
5519 +       h_dentry = au_h_dptr(dentry, bdst);
5520 +       if (!au_ftest_cpup(cpg->flags, OVERWRITE)) {
5521 +               dget(h_dentry);
5522 +               au_set_h_dptr(dentry, bdst, NULL);
5523 +               err = au_lkup_neg(dentry, bdst, /*wh*/0);
5524 +               if (!err)
5525 +                       h_path->dentry = dget(au_h_dptr(dentry, bdst));
5526 +               au_set_h_dptr(dentry, bdst, h_dentry);
5527 +       } else {
5528 +               err = 0;
5529 +               parent = dget_parent(dentry);
5530 +               h_ppath.dentry = au_h_dptr(parent, bdst);
5531 +               dput(parent);
5532 +               h_path->dentry = vfsub_lkup_one(&dentry->d_name, &h_ppath);
5533 +               if (IS_ERR(h_path->dentry))
5534 +                       err = PTR_ERR(h_path->dentry);
5535 +       }
5536 +       if (unlikely(err))
5537 +               goto out;
5538 +
5539 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
5540 +       h_dir = d_inode(h_parent);
5541 +       IMustLock(h_dir);
5542 +       AuDbg("%pd %pd\n", h_dentry, h_path->dentry);
5543 +       /* no delegation since it is just created */
5544 +       err = vfsub_rename(h_dir, h_dentry, h_dir, h_path, /*delegated*/NULL,
5545 +                          /*flags*/0);
5546 +       dput(h_path->dentry);
5547 +
5548 +out:
5549 +       return err;
5550 +}
5551 +
5552 +/*
5553 + * copyup the @dentry from @bsrc to @bdst.
5554 + * the caller must set the both of lower dentries.
5555 + * @len is for truncating when it is -1 copyup the entire file.
5556 + * in link/rename cases, @dst_parent may be different from the real one.
5557 + * basic->bsrc can be larger than basic->bdst.
5558 + * aufs doesn't touch the credential so
5559 + * security_inode_copy_up{,_xattr}() are unnecessary.
5560 + */
5561 +static int au_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5562 +{
5563 +       int err, rerr;
5564 +       aufs_bindex_t old_ibtop;
5565 +       unsigned char isdir, plink;
5566 +       struct dentry *h_src, *h_dst, *h_parent;
5567 +       struct inode *dst_inode, *h_dir, *inode, *delegated, *src_inode;
5568 +       struct super_block *sb;
5569 +       struct au_branch *br;
5570 +       struct path h_src_path;
5571 +       /* to reduce stack size */
5572 +       struct {
5573 +               struct au_dtime dt;
5574 +               struct path h_path;
5575 +               struct au_cpup_reg_attr h_src_attr;
5576 +       } *a;
5577 +
5578 +       err = -ENOMEM;
5579 +       a = kmalloc(sizeof(*a), GFP_NOFS);
5580 +       if (unlikely(!a))
5581 +               goto out;
5582 +       a->h_src_attr.valid = 0;
5583 +
5584 +       sb = cpg->dentry->d_sb;
5585 +       br = au_sbr(sb, cpg->bdst);
5586 +       a->h_path.mnt = au_br_mnt(br);
5587 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5588 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5589 +       h_dir = d_inode(h_parent);
5590 +       IMustLock(h_dir);
5591 +
5592 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5593 +       inode = d_inode(cpg->dentry);
5594 +
5595 +       if (!dst_parent)
5596 +               dst_parent = dget_parent(cpg->dentry);
5597 +       else
5598 +               dget(dst_parent);
5599 +
5600 +       plink = !!au_opt_test(au_mntflags(sb), PLINK);
5601 +       dst_inode = au_h_iptr(inode, cpg->bdst);
5602 +       if (dst_inode) {
5603 +               if (unlikely(!plink)) {
5604 +                       err = -EIO;
5605 +                       AuIOErr("hi%lu(i%lu) exists on b%d "
5606 +                               "but plink is disabled\n",
5607 +                               dst_inode->i_ino, inode->i_ino, cpg->bdst);
5608 +                       goto out_parent;
5609 +               }
5610 +
5611 +               if (dst_inode->i_nlink) {
5612 +                       const int do_dt = au_ftest_cpup(cpg->flags, DTIME);
5613 +
5614 +                       h_src = au_plink_lkup(inode, cpg->bdst);
5615 +                       err = PTR_ERR(h_src);
5616 +                       if (IS_ERR(h_src))
5617 +                               goto out_parent;
5618 +                       if (unlikely(d_is_negative(h_src))) {
5619 +                               err = -EIO;
5620 +                               AuIOErr("i%lu exists on b%d "
5621 +                                       "but not pseudo-linked\n",
5622 +                                       inode->i_ino, cpg->bdst);
5623 +                               dput(h_src);
5624 +                               goto out_parent;
5625 +                       }
5626 +
5627 +                       if (do_dt) {
5628 +                               a->h_path.dentry = h_parent;
5629 +                               au_dtime_store(&a->dt, dst_parent, &a->h_path);
5630 +                       }
5631 +
5632 +                       a->h_path.dentry = h_dst;
5633 +                       delegated = NULL;
5634 +                       err = vfsub_link(h_src, h_dir, &a->h_path, &delegated);
5635 +                       if (!err && au_ftest_cpup(cpg->flags, RENAME))
5636 +                               err = au_do_ren_after_cpup(cpg, &a->h_path);
5637 +                       if (do_dt)
5638 +                               au_dtime_revert(&a->dt);
5639 +                       if (unlikely(err == -EWOULDBLOCK)) {
5640 +                               pr_warn("cannot retry for NFSv4 delegation"
5641 +                                       " for an internal link\n");
5642 +                               iput(delegated);
5643 +                       }
5644 +                       dput(h_src);
5645 +                       goto out_parent;
5646 +               } else
5647 +                       /* todo: cpup_wh_file? */
5648 +                       /* udba work */
5649 +                       au_update_ibrange(inode, /*do_put_zero*/1);
5650 +       }
5651 +
5652 +       isdir = S_ISDIR(inode->i_mode);
5653 +       old_ibtop = au_ibtop(inode);
5654 +       err = cpup_entry(cpg, dst_parent, &a->h_src_attr);
5655 +       if (unlikely(err))
5656 +               goto out_rev;
5657 +       dst_inode = d_inode(h_dst);
5658 +       inode_lock_nested(dst_inode, AuLsc_I_CHILD2);
5659 +       /* todo: necessary? */
5660 +       /* au_pin_hdir_unlock(cpg->pin); */
5661 +
5662 +       h_src_path.dentry = h_src;
5663 +       h_src_path.mnt = au_sbr_mnt(sb, cpg->bsrc);
5664 +       err = cpup_iattr(cpg->dentry, cpg->bdst, &h_src_path, &a->h_src_attr);
5665 +       if (unlikely(err)) {
5666 +               /* todo: necessary? */
5667 +               /* au_pin_hdir_relock(cpg->pin); */ /* ignore an error */
5668 +               inode_unlock(dst_inode);
5669 +               goto out_rev;
5670 +       }
5671 +
5672 +       if (cpg->bdst < old_ibtop) {
5673 +               if (S_ISREG(inode->i_mode)) {
5674 +                       err = au_dy_iaop(inode, cpg->bdst, dst_inode);
5675 +                       if (unlikely(err)) {
5676 +                               /* ignore an error */
5677 +                               /* au_pin_hdir_relock(cpg->pin); */
5678 +                               inode_unlock(dst_inode);
5679 +                               goto out_rev;
5680 +                       }
5681 +               }
5682 +               au_set_ibtop(inode, cpg->bdst);
5683 +       } else
5684 +               au_set_ibbot(inode, cpg->bdst);
5685 +       au_set_h_iptr(inode, cpg->bdst, au_igrab(dst_inode),
5686 +                     au_hi_flags(inode, isdir));
5687 +
5688 +       /* todo: necessary? */
5689 +       /* err = au_pin_hdir_relock(cpg->pin); */
5690 +       inode_unlock(dst_inode);
5691 +       if (unlikely(err))
5692 +               goto out_rev;
5693 +
5694 +       src_inode = d_inode(h_src);
5695 +       if (!isdir
5696 +           && (src_inode->i_nlink > 1
5697 +               || src_inode->i_state & I_LINKABLE)
5698 +           && plink)
5699 +               au_plink_append(inode, cpg->bdst, h_dst);
5700 +
5701 +       if (au_ftest_cpup(cpg->flags, RENAME)) {
5702 +               a->h_path.dentry = h_dst;
5703 +               err = au_do_ren_after_cpup(cpg, &a->h_path);
5704 +       }
5705 +       if (!err)
5706 +               goto out_parent; /* success */
5707 +
5708 +       /* revert */
5709 +out_rev:
5710 +       a->h_path.dentry = h_parent;
5711 +       au_dtime_store(&a->dt, dst_parent, &a->h_path);
5712 +       a->h_path.dentry = h_dst;
5713 +       rerr = 0;
5714 +       if (d_is_positive(h_dst)) {
5715 +               if (!isdir) {
5716 +                       /* no delegation since it is just created */
5717 +                       rerr = vfsub_unlink(h_dir, &a->h_path,
5718 +                                           /*delegated*/NULL, /*force*/0);
5719 +               } else
5720 +                       rerr = vfsub_rmdir(h_dir, &a->h_path);
5721 +       }
5722 +       au_dtime_revert(&a->dt);
5723 +       if (rerr) {
5724 +               AuIOErr("failed removing broken entry(%d, %d)\n", err, rerr);
5725 +               err = -EIO;
5726 +       }
5727 +out_parent:
5728 +       dput(dst_parent);
5729 +       au_kfree_rcu(a);
5730 +out:
5731 +       return err;
5732 +}
5733 +
5734 +#if 0 /* reserved */
5735 +struct au_cpup_single_args {
5736 +       int *errp;
5737 +       struct au_cp_generic *cpg;
5738 +       struct dentry *dst_parent;
5739 +};
5740 +
5741 +static void au_call_cpup_single(void *args)
5742 +{
5743 +       struct au_cpup_single_args *a = args;
5744 +
5745 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5746 +       *a->errp = au_cpup_single(a->cpg, a->dst_parent);
5747 +       au_pin_hdir_release(a->cpg->pin);
5748 +}
5749 +#endif
5750 +
5751 +/*
5752 + * prevent SIGXFSZ in copy-up.
5753 + * testing CAP_MKNOD is for generic fs,
5754 + * but CAP_FSETID is for xfs only, currently.
5755 + */
5756 +static int au_cpup_sio_test(struct au_pin *pin, umode_t mode)
5757 +{
5758 +       int do_sio;
5759 +       struct super_block *sb;
5760 +       struct inode *h_dir;
5761 +
5762 +       do_sio = 0;
5763 +       sb = au_pinned_parent(pin)->d_sb;
5764 +       if (!au_wkq_test()
5765 +           && (!au_sbi(sb)->si_plink_maint_pid
5766 +               || au_plink_maint(sb, AuLock_NOPLM))) {
5767 +               switch (mode & S_IFMT) {
5768 +               case S_IFREG:
5769 +                       /* no condition about RLIMIT_FSIZE and the file size */
5770 +                       do_sio = 1;
5771 +                       break;
5772 +               case S_IFCHR:
5773 +               case S_IFBLK:
5774 +                       do_sio = !capable(CAP_MKNOD);
5775 +                       break;
5776 +               }
5777 +               if (!do_sio)
5778 +                       do_sio = ((mode & (S_ISUID | S_ISGID))
5779 +                                 && !capable(CAP_FSETID));
5780 +               /* this workaround may be removed in the future */
5781 +               if (!do_sio) {
5782 +                       h_dir = au_pinned_h_dir(pin);
5783 +                       do_sio = h_dir->i_mode & S_ISVTX;
5784 +               }
5785 +       }
5786 +
5787 +       return do_sio;
5788 +}
5789 +
5790 +#if 0 /* reserved */
5791 +int au_sio_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5792 +{
5793 +       int err, wkq_err;
5794 +       struct dentry *h_dentry;
5795 +
5796 +       h_dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5797 +       if (!au_cpup_sio_test(pin, d_inode(h_dentry)->i_mode))
5798 +               err = au_cpup_single(cpg, dst_parent);
5799 +       else {
5800 +               struct au_cpup_single_args args = {
5801 +                       .errp           = &err,
5802 +                       .cpg            = cpg,
5803 +                       .dst_parent     = dst_parent
5804 +               };
5805 +               wkq_err = au_wkq_wait(au_call_cpup_single, &args);
5806 +               if (unlikely(wkq_err))
5807 +                       err = wkq_err;
5808 +       }
5809 +
5810 +       return err;
5811 +}
5812 +#endif
5813 +
5814 +/*
5815 + * copyup the @dentry from the first active lower branch to @bdst,
5816 + * using au_cpup_single().
5817 + */
5818 +static int au_cpup_simple(struct au_cp_generic *cpg)
5819 +{
5820 +       int err;
5821 +       unsigned int flags_orig;
5822 +       struct dentry *dentry;
5823 +
5824 +       AuDebugOn(cpg->bsrc < 0);
5825 +
5826 +       dentry = cpg->dentry;
5827 +       DiMustWriteLock(dentry);
5828 +
5829 +       err = au_lkup_neg(dentry, cpg->bdst, /*wh*/1);
5830 +       if (!err) {
5831 +               flags_orig = cpg->flags;
5832 +               au_fset_cpup(cpg->flags, RENAME);
5833 +               err = au_cpup_single(cpg, NULL);
5834 +               cpg->flags = flags_orig;
5835 +               if (!err)
5836 +                       return 0; /* success */
5837 +
5838 +               /* revert */
5839 +               au_set_h_dptr(dentry, cpg->bdst, NULL);
5840 +               au_set_dbtop(dentry, cpg->bsrc);
5841 +       }
5842 +
5843 +       return err;
5844 +}
5845 +
5846 +struct au_cpup_simple_args {
5847 +       int *errp;
5848 +       struct au_cp_generic *cpg;
5849 +};
5850 +
5851 +static void au_call_cpup_simple(void *args)
5852 +{
5853 +       struct au_cpup_simple_args *a = args;
5854 +
5855 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5856 +       *a->errp = au_cpup_simple(a->cpg);
5857 +       au_pin_hdir_release(a->cpg->pin);
5858 +}
5859 +
5860 +static int au_do_sio_cpup_simple(struct au_cp_generic *cpg)
5861 +{
5862 +       int err, wkq_err;
5863 +       struct dentry *dentry, *parent;
5864 +       struct file *h_file;
5865 +       struct inode *h_dir;
5866 +       struct user_namespace *h_userns;
5867 +
5868 +       dentry = cpg->dentry;
5869 +       h_file = NULL;
5870 +       if (au_ftest_cpup(cpg->flags, HOPEN)) {
5871 +               AuDebugOn(cpg->bsrc < 0);
5872 +               h_file = au_h_open_pre(dentry, cpg->bsrc, /*force_wr*/0);
5873 +               err = PTR_ERR(h_file);
5874 +               if (IS_ERR(h_file))
5875 +                       goto out;
5876 +       }
5877 +
5878 +       parent = dget_parent(dentry);
5879 +       h_dir = au_h_iptr(d_inode(parent), cpg->bdst);
5880 +       h_userns = au_sbr_userns(dentry->d_sb, cpg->bdst);
5881 +       if (!au_test_h_perm_sio(h_userns, h_dir, MAY_EXEC | MAY_WRITE)
5882 +           && !au_cpup_sio_test(cpg->pin, d_inode(dentry)->i_mode))
5883 +               err = au_cpup_simple(cpg);
5884 +       else {
5885 +               struct au_cpup_simple_args args = {
5886 +                       .errp           = &err,
5887 +                       .cpg            = cpg
5888 +               };
5889 +               wkq_err = au_wkq_wait(au_call_cpup_simple, &args);
5890 +               if (unlikely(wkq_err))
5891 +                       err = wkq_err;
5892 +       }
5893 +
5894 +       dput(parent);
5895 +       if (h_file)
5896 +               au_h_open_post(dentry, cpg->bsrc, h_file);
5897 +
5898 +out:
5899 +       return err;
5900 +}
5901 +
5902 +int au_sio_cpup_simple(struct au_cp_generic *cpg)
5903 +{
5904 +       aufs_bindex_t bsrc, bbot;
5905 +       struct dentry *dentry, *h_dentry;
5906 +
5907 +       if (cpg->bsrc < 0) {
5908 +               dentry = cpg->dentry;
5909 +               bbot = au_dbbot(dentry);
5910 +               for (bsrc = cpg->bdst + 1; bsrc <= bbot; bsrc++) {
5911 +                       h_dentry = au_h_dptr(dentry, bsrc);
5912 +                       if (h_dentry) {
5913 +                               AuDebugOn(d_is_negative(h_dentry));
5914 +                               break;
5915 +                       }
5916 +               }
5917 +               AuDebugOn(bsrc > bbot);
5918 +               cpg->bsrc = bsrc;
5919 +       }
5920 +       AuDebugOn(cpg->bsrc <= cpg->bdst);
5921 +       return au_do_sio_cpup_simple(cpg);
5922 +}
5923 +
5924 +int au_sio_cpdown_simple(struct au_cp_generic *cpg)
5925 +{
5926 +       AuDebugOn(cpg->bdst <= cpg->bsrc);
5927 +       return au_do_sio_cpup_simple(cpg);
5928 +}
5929 +
5930 +/* ---------------------------------------------------------------------- */
5931 +
5932 +/*
5933 + * copyup the deleted file for writing.
5934 + */
5935 +static int au_do_cpup_wh(struct au_cp_generic *cpg, struct dentry *wh_dentry,
5936 +                        struct file *file)
5937 +{
5938 +       int err;
5939 +       unsigned int flags_orig;
5940 +       aufs_bindex_t bsrc_orig;
5941 +       struct au_dinfo *dinfo;
5942 +       struct {
5943 +               struct au_hdentry *hd;
5944 +               struct dentry *h_dentry;
5945 +       } hdst, hsrc;
5946 +
5947 +       dinfo = au_di(cpg->dentry);
5948 +       AuRwMustWriteLock(&dinfo->di_rwsem);
5949 +
5950 +       bsrc_orig = cpg->bsrc;
5951 +       cpg->bsrc = dinfo->di_btop;
5952 +       hdst.hd = au_hdentry(dinfo, cpg->bdst);
5953 +       hdst.h_dentry = hdst.hd->hd_dentry;
5954 +       hdst.hd->hd_dentry = wh_dentry;
5955 +       dinfo->di_btop = cpg->bdst;
5956 +
5957 +       hsrc.h_dentry = NULL;
5958 +       if (file) {
5959 +               hsrc.hd = au_hdentry(dinfo, cpg->bsrc);
5960 +               hsrc.h_dentry = hsrc.hd->hd_dentry;
5961 +               hsrc.hd->hd_dentry = au_hf_top(file)->f_path.dentry;
5962 +       }
5963 +       flags_orig = cpg->flags;
5964 +       cpg->flags = !AuCpup_DTIME;
5965 +       err = au_cpup_single(cpg, /*h_parent*/NULL);
5966 +       cpg->flags = flags_orig;
5967 +       if (file) {
5968 +               if (!err)
5969 +                       err = au_reopen_nondir(file);
5970 +               hsrc.hd->hd_dentry = hsrc.h_dentry;
5971 +       }
5972 +       hdst.hd->hd_dentry = hdst.h_dentry;
5973 +       dinfo->di_btop = cpg->bsrc;
5974 +       cpg->bsrc = bsrc_orig;
5975 +
5976 +       return err;
5977 +}
5978 +
5979 +static int au_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5980 +{
5981 +       int err;
5982 +       aufs_bindex_t bdst;
5983 +       struct au_dtime dt;
5984 +       struct dentry *dentry, *parent, *h_parent, *wh_dentry;
5985 +       struct au_branch *br;
5986 +       struct path h_path;
5987 +
5988 +       dentry = cpg->dentry;
5989 +       bdst = cpg->bdst;
5990 +       br = au_sbr(dentry->d_sb, bdst);
5991 +       parent = dget_parent(dentry);
5992 +       h_parent = au_h_dptr(parent, bdst);
5993 +       wh_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
5994 +       err = PTR_ERR(wh_dentry);
5995 +       if (IS_ERR(wh_dentry))
5996 +               goto out;
5997 +
5998 +       h_path.dentry = h_parent;
5999 +       h_path.mnt = au_br_mnt(br);
6000 +       au_dtime_store(&dt, parent, &h_path);
6001 +       err = au_do_cpup_wh(cpg, wh_dentry, file);
6002 +       if (unlikely(err))
6003 +               goto out_wh;
6004 +
6005 +       dget(wh_dentry);
6006 +       h_path.dentry = wh_dentry;
6007 +       if (!d_is_dir(wh_dentry)) {
6008 +               /* no delegation since it is just created */
6009 +               err = vfsub_unlink(d_inode(h_parent), &h_path,
6010 +                                  /*delegated*/NULL, /*force*/0);
6011 +       } else
6012 +               err = vfsub_rmdir(d_inode(h_parent), &h_path);
6013 +       if (unlikely(err)) {
6014 +               AuIOErr("failed remove copied-up tmp file %pd(%d)\n",
6015 +                       wh_dentry, err);
6016 +               err = -EIO;
6017 +       }
6018 +       au_dtime_revert(&dt);
6019 +       au_set_hi_wh(d_inode(dentry), bdst, wh_dentry);
6020 +
6021 +out_wh:
6022 +       dput(wh_dentry);
6023 +out:
6024 +       dput(parent);
6025 +       return err;
6026 +}
6027 +
6028 +struct au_cpup_wh_args {
6029 +       int *errp;
6030 +       struct au_cp_generic *cpg;
6031 +       struct file *file;
6032 +};
6033 +
6034 +static void au_call_cpup_wh(void *args)
6035 +{
6036 +       struct au_cpup_wh_args *a = args;
6037 +
6038 +       au_pin_hdir_acquire_nest(a->cpg->pin);
6039 +       *a->errp = au_cpup_wh(a->cpg, a->file);
6040 +       au_pin_hdir_release(a->cpg->pin);
6041 +}
6042 +
6043 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file)
6044 +{
6045 +       int err, wkq_err;
6046 +       aufs_bindex_t bdst;
6047 +       struct dentry *dentry, *parent, *h_orph, *h_parent;
6048 +       struct inode *dir, *h_dir, *h_tmpdir;
6049 +       struct au_wbr *wbr;
6050 +       struct au_pin wh_pin, *pin_orig;
6051 +       struct user_namespace *h_userns;
6052 +
6053 +       dentry = cpg->dentry;
6054 +       bdst = cpg->bdst;
6055 +       parent = dget_parent(dentry);
6056 +       dir = d_inode(parent);
6057 +       h_orph = NULL;
6058 +       h_parent = NULL;
6059 +       h_dir = au_igrab(au_h_iptr(dir, bdst));
6060 +       h_tmpdir = h_dir;
6061 +       pin_orig = NULL;
6062 +       if (!h_dir->i_nlink) {
6063 +               wbr = au_sbr(dentry->d_sb, bdst)->br_wbr;
6064 +               h_orph = wbr->wbr_orph;
6065 +
6066 +               h_parent = dget(au_h_dptr(parent, bdst));
6067 +               au_set_h_dptr(parent, bdst, dget(h_orph));
6068 +               h_tmpdir = d_inode(h_orph);
6069 +               au_set_h_iptr(dir, bdst, au_igrab(h_tmpdir), /*flags*/0);
6070 +
6071 +               inode_lock_nested(h_tmpdir, AuLsc_I_PARENT3);
6072 +               /* todo: au_h_open_pre()? */
6073 +
6074 +               pin_orig = cpg->pin;
6075 +               au_pin_init(&wh_pin, dentry, bdst, AuLsc_DI_PARENT,
6076 +                           AuLsc_I_PARENT3, cpg->pin->udba, AuPin_DI_LOCKED);
6077 +               cpg->pin = &wh_pin;
6078 +       }
6079 +
6080 +       h_userns = au_sbr_userns(dentry->d_sb, bdst);
6081 +       if (!au_test_h_perm_sio(h_userns, h_tmpdir, MAY_EXEC | MAY_WRITE)
6082 +           && !au_cpup_sio_test(cpg->pin, d_inode(dentry)->i_mode))
6083 +               err = au_cpup_wh(cpg, file);
6084 +       else {
6085 +               struct au_cpup_wh_args args = {
6086 +                       .errp   = &err,
6087 +                       .cpg    = cpg,
6088 +                       .file   = file
6089 +               };
6090 +               wkq_err = au_wkq_wait(au_call_cpup_wh, &args);
6091 +               if (unlikely(wkq_err))
6092 +                       err = wkq_err;
6093 +       }
6094 +
6095 +       if (h_orph) {
6096 +               inode_unlock(h_tmpdir);
6097 +               /* todo: au_h_open_post()? */
6098 +               au_set_h_iptr(dir, bdst, au_igrab(h_dir), /*flags*/0);
6099 +               au_set_h_dptr(parent, bdst, h_parent);
6100 +               AuDebugOn(!pin_orig);
6101 +               cpg->pin = pin_orig;
6102 +       }
6103 +       iput(h_dir);
6104 +       dput(parent);
6105 +
6106 +       return err;
6107 +}
6108 +
6109 +/* ---------------------------------------------------------------------- */
6110 +
6111 +/*
6112 + * generic routine for both of copy-up and copy-down.
6113 + */
6114 +/* cf. revalidate function in file.c */
6115 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
6116 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
6117 +                        struct au_pin *pin,
6118 +                        struct dentry *h_parent, void *arg),
6119 +              void *arg)
6120 +{
6121 +       int err;
6122 +       struct au_pin pin;
6123 +       struct dentry *d, *parent, *h_parent, *real_parent, *h_dentry;
6124 +
6125 +       err = 0;
6126 +       parent = dget_parent(dentry);
6127 +       if (IS_ROOT(parent))
6128 +               goto out;
6129 +
6130 +       au_pin_init(&pin, dentry, bdst, AuLsc_DI_PARENT2, AuLsc_I_PARENT2,
6131 +                   au_opt_udba(dentry->d_sb), AuPin_MNT_WRITE);
6132 +
6133 +       /* do not use au_dpage */
6134 +       real_parent = parent;
6135 +       while (1) {
6136 +               dput(parent);
6137 +               parent = dget_parent(dentry);
6138 +               h_parent = au_h_dptr(parent, bdst);
6139 +               if (h_parent)
6140 +                       goto out; /* success */
6141 +
6142 +               /* find top dir which is necessary to cpup */
6143 +               do {
6144 +                       d = parent;
6145 +                       dput(parent);
6146 +                       parent = dget_parent(d);
6147 +                       di_read_lock_parent3(parent, !AuLock_IR);
6148 +                       h_parent = au_h_dptr(parent, bdst);
6149 +                       di_read_unlock(parent, !AuLock_IR);
6150 +               } while (!h_parent);
6151 +
6152 +               if (d != real_parent)
6153 +                       di_write_lock_child3(d);
6154 +
6155 +               /* somebody else might create while we were sleeping */
6156 +               h_dentry = au_h_dptr(d, bdst);
6157 +               if (!h_dentry || d_is_negative(h_dentry)) {
6158 +                       if (h_dentry)
6159 +                               au_update_dbtop(d);
6160 +
6161 +                       au_pin_set_dentry(&pin, d);
6162 +                       err = au_do_pin(&pin);
6163 +                       if (!err) {
6164 +                               err = cp(d, bdst, &pin, h_parent, arg);
6165 +                               au_unpin(&pin);
6166 +                       }
6167 +               }
6168 +
6169 +               if (d != real_parent)
6170 +                       di_write_unlock(d);
6171 +               if (unlikely(err))
6172 +                       break;
6173 +       }
6174 +
6175 +out:
6176 +       dput(parent);
6177 +       return err;
6178 +}
6179 +
6180 +static int au_cpup_dir(struct dentry *dentry, aufs_bindex_t bdst,
6181 +                      struct au_pin *pin,
6182 +                      struct dentry *h_parent __maybe_unused,
6183 +                      void *arg __maybe_unused)
6184 +{
6185 +       struct au_cp_generic cpg = {
6186 +               .dentry = dentry,
6187 +               .bdst   = bdst,
6188 +               .bsrc   = -1,
6189 +               .len    = 0,
6190 +               .pin    = pin,
6191 +               .flags  = AuCpup_DTIME
6192 +       };
6193 +       return au_sio_cpup_simple(&cpg);
6194 +}
6195 +
6196 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
6197 +{
6198 +       return au_cp_dirs(dentry, bdst, au_cpup_dir, NULL);
6199 +}
6200 +
6201 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
6202 +{
6203 +       int err;
6204 +       struct dentry *parent;
6205 +       struct inode *dir;
6206 +
6207 +       parent = dget_parent(dentry);
6208 +       dir = d_inode(parent);
6209 +       err = 0;
6210 +       if (au_h_iptr(dir, bdst))
6211 +               goto out;
6212 +
6213 +       di_read_unlock(parent, AuLock_IR);
6214 +       di_write_lock_parent(parent);
6215 +       /* someone else might change our inode while we were sleeping */
6216 +       if (!au_h_iptr(dir, bdst))
6217 +               err = au_cpup_dirs(dentry, bdst);
6218 +       di_downgrade_lock(parent, AuLock_IR);
6219 +
6220 +out:
6221 +       dput(parent);
6222 +       return err;
6223 +}
6224 diff -urN /usr/share/empty/fs/aufs/cpup.h linux/fs/aufs/cpup.h
6225 --- /usr/share/empty/fs/aufs/cpup.h     1970-01-01 01:00:00.000000000 +0100
6226 +++ linux/fs/aufs/cpup.h        2021-12-03 15:38:59.933313976 +0100
6227 @@ -0,0 +1,100 @@
6228 +/* SPDX-License-Identifier: GPL-2.0 */
6229 +/*
6230 + * Copyright (C) 2005-2021 Junjiro R. Okajima
6231 + *
6232 + * This program, aufs is free software; you can redistribute it and/or modify
6233 + * it under the terms of the GNU General Public License as published by
6234 + * the Free Software Foundation; either version 2 of the License, or
6235 + * (at your option) any later version.
6236 + *
6237 + * This program is distributed in the hope that it will be useful,
6238 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6239 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6240 + * GNU General Public License for more details.
6241 + *
6242 + * You should have received a copy of the GNU General Public License
6243 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6244 + */
6245 +
6246 +/*
6247 + * copy-up/down functions
6248 + */
6249 +
6250 +#ifndef __AUFS_CPUP_H__
6251 +#define __AUFS_CPUP_H__
6252 +
6253 +#ifdef __KERNEL__
6254 +
6255 +#include <linux/path.h>
6256 +
6257 +struct inode;
6258 +struct file;
6259 +struct au_pin;
6260 +
6261 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags);
6262 +void au_cpup_attr_timesizes(struct inode *inode);
6263 +void au_cpup_attr_nlink(struct inode *inode, int force);
6264 +void au_cpup_attr_changeable(struct inode *inode);
6265 +void au_cpup_igen(struct inode *inode, struct inode *h_inode);
6266 +void au_cpup_attr_all(struct inode *inode, int force);
6267 +
6268 +/* ---------------------------------------------------------------------- */
6269 +
6270 +struct au_cp_generic {
6271 +       struct dentry   *dentry;
6272 +       aufs_bindex_t   bdst, bsrc;
6273 +       loff_t          len;
6274 +       struct au_pin   *pin;
6275 +       unsigned int    flags;
6276 +};
6277 +
6278 +/* cpup flags */
6279 +#define AuCpup_DTIME           1               /* do dtime_store/revert */
6280 +#define AuCpup_KEEPLINO                (1 << 1)        /* do not clear the lower xino,
6281 +                                                  for link(2) */
6282 +#define AuCpup_RENAME          (1 << 2)        /* rename after cpup */
6283 +#define AuCpup_HOPEN           (1 << 3)        /* call h_open_pre/post() in
6284 +                                                  cpup */
6285 +#define AuCpup_OVERWRITE       (1 << 4)        /* allow overwriting the
6286 +                                                  existing entry */
6287 +#define AuCpup_RWDST           (1 << 5)        /* force write target even if
6288 +                                                  the branch is marked as RO */
6289 +
6290 +#ifndef CONFIG_AUFS_BR_HFSPLUS
6291 +#undef AuCpup_HOPEN
6292 +#define AuCpup_HOPEN           0
6293 +#endif
6294 +
6295 +#define au_ftest_cpup(flags, name)     ((flags) & AuCpup_##name)
6296 +#define au_fset_cpup(flags, name) \
6297 +       do { (flags) |= AuCpup_##name; } while (0)
6298 +#define au_fclr_cpup(flags, name) \
6299 +       do { (flags) &= ~AuCpup_##name; } while (0)
6300 +
6301 +int au_copy_file(struct file *dst, struct file *src, loff_t len);
6302 +int au_sio_cpup_simple(struct au_cp_generic *cpg);
6303 +int au_sio_cpdown_simple(struct au_cp_generic *cpg);
6304 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file);
6305 +
6306 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
6307 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
6308 +                        struct au_pin *pin,
6309 +                        struct dentry *h_parent, void *arg),
6310 +              void *arg);
6311 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
6312 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
6313 +
6314 +/* ---------------------------------------------------------------------- */
6315 +
6316 +/* keep timestamps when copyup */
6317 +struct au_dtime {
6318 +       struct dentry *dt_dentry;
6319 +       struct path dt_h_path;
6320 +       struct timespec64 dt_atime, dt_mtime;
6321 +};
6322 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
6323 +                   struct path *h_path);
6324 +void au_dtime_revert(struct au_dtime *dt);
6325 +
6326 +#endif /* __KERNEL__ */
6327 +#endif /* __AUFS_CPUP_H__ */
6328 diff -urN /usr/share/empty/fs/aufs/dbgaufs.c linux/fs/aufs/dbgaufs.c
6329 --- /usr/share/empty/fs/aufs/dbgaufs.c  1970-01-01 01:00:00.000000000 +0100
6330 +++ linux/fs/aufs/dbgaufs.c     2021-12-03 15:38:59.933313976 +0100
6331 @@ -0,0 +1,526 @@
6332 +// SPDX-License-Identifier: GPL-2.0
6333 +/*
6334 + * Copyright (C) 2005-2021 Junjiro R. Okajima
6335 + *
6336 + * This program, aufs is free software; you can redistribute it and/or modify
6337 + * it under the terms of the GNU General Public License as published by
6338 + * the Free Software Foundation; either version 2 of the License, or
6339 + * (at your option) any later version.
6340 + *
6341 + * This program is distributed in the hope that it will be useful,
6342 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6343 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6344 + * GNU General Public License for more details.
6345 + *
6346 + * You should have received a copy of the GNU General Public License
6347 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6348 + */
6349 +
6350 +/*
6351 + * debugfs interface
6352 + */
6353 +
6354 +#include <linux/debugfs.h>
6355 +#include "aufs.h"
6356 +
6357 +#ifndef CONFIG_SYSFS
6358 +#error DEBUG_FS depends upon SYSFS
6359 +#endif
6360 +
6361 +static struct dentry *dbgaufs;
6362 +static const mode_t dbgaufs_mode = 0444;
6363 +
6364 +/* 20 is max digits length of ulong 64 */
6365 +struct dbgaufs_arg {
6366 +       int n;
6367 +       char a[20 * 4];
6368 +};
6369 +
6370 +/*
6371 + * common function for all XINO files
6372 + */
6373 +static int dbgaufs_xi_release(struct inode *inode __maybe_unused,
6374 +                             struct file *file)
6375 +{
6376 +       void *p;
6377 +
6378 +       p = file->private_data;
6379 +       if (p) {
6380 +               /* this is struct dbgaufs_arg */
6381 +               AuDebugOn(!au_kfree_sz_test(p));
6382 +               au_kfree_do_rcu(p);
6383 +       }
6384 +       return 0;
6385 +}
6386 +
6387 +static int dbgaufs_xi_open(struct file *xf, struct file *file, int do_fcnt,
6388 +                          int cnt)
6389 +{
6390 +       int err;
6391 +       struct kstat st;
6392 +       struct dbgaufs_arg *p;
6393 +
6394 +       err = -ENOMEM;
6395 +       p = kmalloc(sizeof(*p), GFP_NOFS);
6396 +       if (unlikely(!p))
6397 +               goto out;
6398 +
6399 +       err = 0;
6400 +       p->n = 0;
6401 +       file->private_data = p;
6402 +       if (!xf)
6403 +               goto out;
6404 +
6405 +       err = vfsub_getattr(&xf->f_path, &st);
6406 +       if (!err) {
6407 +               if (do_fcnt)
6408 +                       p->n = snprintf
6409 +                               (p->a, sizeof(p->a), "%d, %llux%u %lld\n",
6410 +                                cnt, st.blocks, st.blksize,
6411 +                                (long long)st.size);
6412 +               else
6413 +                       p->n = snprintf(p->a, sizeof(p->a), "%llux%u %lld\n",
6414 +                                       st.blocks, st.blksize,
6415 +                                       (long long)st.size);
6416 +               AuDebugOn(p->n >= sizeof(p->a));
6417 +       } else {
6418 +               p->n = snprintf(p->a, sizeof(p->a), "err %d\n", err);
6419 +               err = 0;
6420 +       }
6421 +
6422 +out:
6423 +       return err;
6424 +}
6425 +
6426 +static ssize_t dbgaufs_xi_read(struct file *file, char __user *buf,
6427 +                              size_t count, loff_t *ppos)
6428 +{
6429 +       struct dbgaufs_arg *p;
6430 +
6431 +       p = file->private_data;
6432 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
6433 +}
6434 +
6435 +/* ---------------------------------------------------------------------- */
6436 +
6437 +struct dbgaufs_plink_arg {
6438 +       int n;
6439 +       char a[];
6440 +};
6441 +
6442 +static int dbgaufs_plink_release(struct inode *inode __maybe_unused,
6443 +                                struct file *file)
6444 +{
6445 +       free_page((unsigned long)file->private_data);
6446 +       return 0;
6447 +}
6448 +
6449 +static int dbgaufs_plink_open(struct inode *inode, struct file *file)
6450 +{
6451 +       int err, i, limit;
6452 +       unsigned long n, sum;
6453 +       struct dbgaufs_plink_arg *p;
6454 +       struct au_sbinfo *sbinfo;
6455 +       struct super_block *sb;
6456 +       struct hlist_bl_head *hbl;
6457 +
6458 +       err = -ENOMEM;
6459 +       p = (void *)get_zeroed_page(GFP_NOFS);
6460 +       if (unlikely(!p))
6461 +               goto out;
6462 +
6463 +       err = -EFBIG;
6464 +       sbinfo = inode->i_private;
6465 +       sb = sbinfo->si_sb;
6466 +       si_noflush_read_lock(sb);
6467 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
6468 +               limit = PAGE_SIZE - sizeof(p->n);
6469 +
6470 +               /* the number of buckets */
6471 +               n = snprintf(p->a + p->n, limit, "%d\n", AuPlink_NHASH);
6472 +               p->n += n;
6473 +               limit -= n;
6474 +
6475 +               sum = 0;
6476 +               for (i = 0, hbl = sbinfo->si_plink; i < AuPlink_NHASH;
6477 +                    i++, hbl++) {
6478 +                       n = au_hbl_count(hbl);
6479 +                       sum += n;
6480 +
6481 +                       n = snprintf(p->a + p->n, limit, "%lu ", n);
6482 +                       p->n += n;
6483 +                       limit -= n;
6484 +                       if (unlikely(limit <= 0))
6485 +                               goto out_free;
6486 +               }
6487 +               p->a[p->n - 1] = '\n';
6488 +
6489 +               /* the sum of plinks */
6490 +               n = snprintf(p->a + p->n, limit, "%lu\n", sum);
6491 +               p->n += n;
6492 +               limit -= n;
6493 +               if (unlikely(limit <= 0))
6494 +                       goto out_free;
6495 +       } else {
6496 +#define str "1\n0\n0\n"
6497 +               p->n = sizeof(str) - 1;
6498 +               strcpy(p->a, str);
6499 +#undef str
6500 +       }
6501 +       si_read_unlock(sb);
6502 +
6503 +       err = 0;
6504 +       file->private_data = p;
6505 +       goto out; /* success */
6506 +
6507 +out_free:
6508 +       free_page((unsigned long)p);
6509 +out:
6510 +       return err;
6511 +}
6512 +
6513 +static ssize_t dbgaufs_plink_read(struct file *file, char __user *buf,
6514 +                                 size_t count, loff_t *ppos)
6515 +{
6516 +       struct dbgaufs_plink_arg *p;
6517 +
6518 +       p = file->private_data;
6519 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
6520 +}
6521 +
6522 +static const struct file_operations dbgaufs_plink_fop = {
6523 +       .owner          = THIS_MODULE,
6524 +       .open           = dbgaufs_plink_open,
6525 +       .release        = dbgaufs_plink_release,
6526 +       .read           = dbgaufs_plink_read
6527 +};
6528 +
6529 +/* ---------------------------------------------------------------------- */
6530 +
6531 +static int dbgaufs_xib_open(struct inode *inode, struct file *file)
6532 +{
6533 +       int err;
6534 +       struct au_sbinfo *sbinfo;
6535 +       struct super_block *sb;
6536 +
6537 +       sbinfo = inode->i_private;
6538 +       sb = sbinfo->si_sb;
6539 +       si_noflush_read_lock(sb);
6540 +       err = dbgaufs_xi_open(sbinfo->si_xib, file, /*do_fcnt*/0, /*cnt*/0);
6541 +       si_read_unlock(sb);
6542 +       return err;
6543 +}
6544 +
6545 +static const struct file_operations dbgaufs_xib_fop = {
6546 +       .owner          = THIS_MODULE,
6547 +       .open           = dbgaufs_xib_open,
6548 +       .release        = dbgaufs_xi_release,
6549 +       .read           = dbgaufs_xi_read
6550 +};
6551 +
6552 +/* ---------------------------------------------------------------------- */
6553 +
6554 +#define DbgaufsXi_PREFIX "xi"
6555 +
6556 +static int dbgaufs_xino_open(struct inode *inode, struct file *file)
6557 +{
6558 +       int err, idx;
6559 +       long l;
6560 +       aufs_bindex_t bindex;
6561 +       char *p, a[sizeof(DbgaufsXi_PREFIX) + 8];
6562 +       struct au_sbinfo *sbinfo;
6563 +       struct super_block *sb;
6564 +       struct au_xino *xi;
6565 +       struct file *xf;
6566 +       struct qstr *name;
6567 +       struct au_branch *br;
6568 +
6569 +       err = -ENOENT;
6570 +       name = &file->f_path.dentry->d_name;
6571 +       if (unlikely(name->len < sizeof(DbgaufsXi_PREFIX)
6572 +                    || memcmp(name->name, DbgaufsXi_PREFIX,
6573 +                              sizeof(DbgaufsXi_PREFIX) - 1)))
6574 +               goto out;
6575 +
6576 +       AuDebugOn(name->len >= sizeof(a));
6577 +       memcpy(a, name->name, name->len);
6578 +       a[name->len] = '\0';
6579 +       p = strchr(a, '-');
6580 +       if (p)
6581 +               *p = '\0';
6582 +       err = kstrtol(a + sizeof(DbgaufsXi_PREFIX) - 1, 10, &l);
6583 +       if (unlikely(err))
6584 +               goto out;
6585 +       bindex = l;
6586 +       idx = 0;
6587 +       if (p) {
6588 +               err = kstrtol(p + 1, 10, &l);
6589 +               if (unlikely(err))
6590 +                       goto out;
6591 +               idx = l;
6592 +       }
6593 +
6594 +       err = -ENOENT;
6595 +       sbinfo = inode->i_private;
6596 +       sb = sbinfo->si_sb;
6597 +       si_noflush_read_lock(sb);
6598 +       if (unlikely(bindex < 0 || bindex > au_sbbot(sb)))
6599 +               goto out_si;
6600 +       br = au_sbr(sb, bindex);
6601 +       xi = br->br_xino;
6602 +       if (unlikely(idx >= xi->xi_nfile))
6603 +               goto out_si;
6604 +       xf = au_xino_file(xi, idx);
6605 +       if (xf)
6606 +               err = dbgaufs_xi_open(xf, file, /*do_fcnt*/1,
6607 +                                     au_xino_count(br));
6608 +
6609 +out_si:
6610 +       si_read_unlock(sb);
6611 +out:
6612 +       AuTraceErr(err);
6613 +       return err;
6614 +}
6615 +
6616 +static const struct file_operations dbgaufs_xino_fop = {
6617 +       .owner          = THIS_MODULE,
6618 +       .open           = dbgaufs_xino_open,
6619 +       .release        = dbgaufs_xi_release,
6620 +       .read           = dbgaufs_xi_read
6621 +};
6622 +
6623 +void dbgaufs_xino_del(struct au_branch *br)
6624 +{
6625 +       struct dentry *dbgaufs;
6626 +
6627 +       dbgaufs = br->br_dbgaufs;
6628 +       if (!dbgaufs)
6629 +               return;
6630 +
6631 +       br->br_dbgaufs = NULL;
6632 +       /* debugfs acquires the parent i_mutex */
6633 +       lockdep_off();
6634 +       debugfs_remove(dbgaufs);
6635 +       lockdep_on();
6636 +}
6637 +
6638 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
6639 +{
6640 +       aufs_bindex_t bbot;
6641 +       struct au_branch *br;
6642 +
6643 +       if (!au_sbi(sb)->si_dbgaufs)
6644 +               return;
6645 +
6646 +       bbot = au_sbbot(sb);
6647 +       for (; bindex <= bbot; bindex++) {
6648 +               br = au_sbr(sb, bindex);
6649 +               dbgaufs_xino_del(br);
6650 +       }
6651 +}
6652 +
6653 +static void dbgaufs_br_do_add(struct super_block *sb, aufs_bindex_t bindex,
6654 +                             unsigned int idx, struct dentry *parent,
6655 +                             struct au_sbinfo *sbinfo)
6656 +{
6657 +       struct au_branch *br;
6658 +       struct dentry *d;
6659 +       /* "xi" bindex(5) "-" idx(2) NULL */
6660 +       char name[sizeof(DbgaufsXi_PREFIX) + 8];
6661 +
6662 +       if (!idx)
6663 +               snprintf(name, sizeof(name), DbgaufsXi_PREFIX "%d", bindex);
6664 +       else
6665 +               snprintf(name, sizeof(name), DbgaufsXi_PREFIX "%d-%u",
6666 +                        bindex, idx);
6667 +       br = au_sbr(sb, bindex);
6668 +       if (br->br_dbgaufs) {
6669 +               struct qstr qstr = QSTR_INIT(name, strlen(name));
6670 +
6671 +               if (!au_qstreq(&br->br_dbgaufs->d_name, &qstr)) {
6672 +                       /* debugfs acquires the parent i_mutex */
6673 +                       lockdep_off();
6674 +                       d = debugfs_rename(parent, br->br_dbgaufs, parent,
6675 +                                          name);
6676 +                       lockdep_on();
6677 +                       if (unlikely(!d))
6678 +                               pr_warn("failed renaming %pd/%s, ignored.\n",
6679 +                                       parent, name);
6680 +               }
6681 +       } else {
6682 +               lockdep_off();
6683 +               br->br_dbgaufs = debugfs_create_file(name, dbgaufs_mode, parent,
6684 +                                                    sbinfo, &dbgaufs_xino_fop);
6685 +               lockdep_on();
6686 +               if (unlikely(!br->br_dbgaufs))
6687 +                       pr_warn("failed creating %pd/%s, ignored.\n",
6688 +                               parent, name);
6689 +       }
6690 +}
6691 +
6692 +static void dbgaufs_br_add(struct super_block *sb, aufs_bindex_t bindex,
6693 +                          struct dentry *parent, struct au_sbinfo *sbinfo)
6694 +{
6695 +       struct au_branch *br;
6696 +       struct au_xino *xi;
6697 +       unsigned int u;
6698 +
6699 +       br = au_sbr(sb, bindex);
6700 +       xi = br->br_xino;
6701 +       for (u = 0; u < xi->xi_nfile; u++)
6702 +               dbgaufs_br_do_add(sb, bindex, u, parent, sbinfo);
6703 +}
6704 +
6705 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex, int topdown)
6706 +{
6707 +       struct au_sbinfo *sbinfo;
6708 +       struct dentry *parent;
6709 +       aufs_bindex_t bbot;
6710 +
6711 +       if (!au_opt_test(au_mntflags(sb), XINO))
6712 +               return;
6713 +
6714 +       sbinfo = au_sbi(sb);
6715 +       parent = sbinfo->si_dbgaufs;
6716 +       if (!parent)
6717 +               return;
6718 +
6719 +       bbot = au_sbbot(sb);
6720 +       if (topdown)
6721 +               for (; bindex <= bbot; bindex++)
6722 +                       dbgaufs_br_add(sb, bindex, parent, sbinfo);
6723 +       else
6724 +               for (; bbot >= bindex; bbot--)
6725 +                       dbgaufs_br_add(sb, bbot, parent, sbinfo);
6726 +}
6727 +
6728 +/* ---------------------------------------------------------------------- */
6729 +
6730 +#ifdef CONFIG_AUFS_EXPORT
6731 +static int dbgaufs_xigen_open(struct inode *inode, struct file *file)
6732 +{
6733 +       int err;
6734 +       struct au_sbinfo *sbinfo;
6735 +       struct super_block *sb;
6736 +
6737 +       sbinfo = inode->i_private;
6738 +       sb = sbinfo->si_sb;
6739 +       si_noflush_read_lock(sb);
6740 +       err = dbgaufs_xi_open(sbinfo->si_xigen, file, /*do_fcnt*/0, /*cnt*/0);
6741 +       si_read_unlock(sb);
6742 +       return err;
6743 +}
6744 +
6745 +static const struct file_operations dbgaufs_xigen_fop = {
6746 +       .owner          = THIS_MODULE,
6747 +       .open           = dbgaufs_xigen_open,
6748 +       .release        = dbgaufs_xi_release,
6749 +       .read           = dbgaufs_xi_read
6750 +};
6751 +
6752 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6753 +{
6754 +       int err;
6755 +
6756 +       /*
6757 +        * This function is a dynamic '__init' function actually,
6758 +        * so the tiny check for si_rwsem is unnecessary.
6759 +        */
6760 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6761 +
6762 +       err = -EIO;
6763 +       sbinfo->si_dbgaufs_xigen = debugfs_create_file
6764 +               ("xigen", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6765 +                &dbgaufs_xigen_fop);
6766 +       if (sbinfo->si_dbgaufs_xigen)
6767 +               err = 0;
6768 +
6769 +       return err;
6770 +}
6771 +#else
6772 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6773 +{
6774 +       return 0;
6775 +}
6776 +#endif /* CONFIG_AUFS_EXPORT */
6777 +
6778 +/* ---------------------------------------------------------------------- */
6779 +
6780 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo)
6781 +{
6782 +       /*
6783 +        * This function is a dynamic '__fin' function actually,
6784 +        * so the tiny check for si_rwsem is unnecessary.
6785 +        */
6786 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6787 +
6788 +       debugfs_remove_recursive(sbinfo->si_dbgaufs);
6789 +       sbinfo->si_dbgaufs = NULL;
6790 +}
6791 +
6792 +int dbgaufs_si_init(struct au_sbinfo *sbinfo)
6793 +{
6794 +       int err;
6795 +       char name[SysaufsSiNameLen];
6796 +
6797 +       /*
6798 +        * This function is a dynamic '__init' function actually,
6799 +        * so the tiny check for si_rwsem is unnecessary.
6800 +        */
6801 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6802 +
6803 +       err = -ENOENT;
6804 +       if (!dbgaufs) {
6805 +               AuErr1("/debug/aufs is uninitialized\n");
6806 +               goto out;
6807 +       }
6808 +
6809 +       err = -EIO;
6810 +       sysaufs_name(sbinfo, name);
6811 +       sbinfo->si_dbgaufs = debugfs_create_dir(name, dbgaufs);
6812 +       if (unlikely(!sbinfo->si_dbgaufs))
6813 +               goto out;
6814 +
6815 +       /* regardless plink/noplink option */
6816 +       sbinfo->si_dbgaufs_plink = debugfs_create_file
6817 +               ("plink", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6818 +                &dbgaufs_plink_fop);
6819 +       if (unlikely(!sbinfo->si_dbgaufs_plink))
6820 +               goto out_dir;
6821 +
6822 +       /* regardless xino/noxino option */
6823 +       sbinfo->si_dbgaufs_xib = debugfs_create_file
6824 +               ("xib", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6825 +                &dbgaufs_xib_fop);
6826 +       if (unlikely(!sbinfo->si_dbgaufs_xib))
6827 +               goto out_dir;
6828 +
6829 +       err = dbgaufs_xigen_init(sbinfo);
6830 +       if (!err)
6831 +               goto out; /* success */
6832 +
6833 +out_dir:
6834 +       dbgaufs_si_fin(sbinfo);
6835 +out:
6836 +       if (unlikely(err))
6837 +               pr_err("debugfs/aufs failed\n");
6838 +       return err;
6839 +}
6840 +
6841 +/* ---------------------------------------------------------------------- */
6842 +
6843 +void dbgaufs_fin(void)
6844 +{
6845 +       debugfs_remove(dbgaufs);
6846 +}
6847 +
6848 +int __init dbgaufs_init(void)
6849 +{
6850 +       int err;
6851 +
6852 +       err = -EIO;
6853 +       dbgaufs = debugfs_create_dir(AUFS_NAME, NULL);
6854 +       if (dbgaufs)
6855 +               err = 0;
6856 +       return err;
6857 +}
6858 diff -urN /usr/share/empty/fs/aufs/dbgaufs.h linux/fs/aufs/dbgaufs.h
6859 --- /usr/share/empty/fs/aufs/dbgaufs.h  1970-01-01 01:00:00.000000000 +0100
6860 +++ linux/fs/aufs/dbgaufs.h     2021-12-03 15:38:59.933313976 +0100
6861 @@ -0,0 +1,53 @@
6862 +/* SPDX-License-Identifier: GPL-2.0 */
6863 +/*
6864 + * Copyright (C) 2005-2021 Junjiro R. Okajima
6865 + *
6866 + * This program, aufs is free software; you can redistribute it and/or modify
6867 + * it under the terms of the GNU General Public License as published by
6868 + * the Free Software Foundation; either version 2 of the License, or
6869 + * (at your option) any later version.
6870 + *
6871 + * This program is distributed in the hope that it will be useful,
6872 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6873 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6874 + * GNU General Public License for more details.
6875 + *
6876 + * You should have received a copy of the GNU General Public License
6877 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6878 + */
6879 +
6880 +/*
6881 + * debugfs interface
6882 + */
6883 +
6884 +#ifndef __DBGAUFS_H__
6885 +#define __DBGAUFS_H__
6886 +
6887 +#ifdef __KERNEL__
6888 +
6889 +struct super_block;
6890 +struct au_sbinfo;
6891 +struct au_branch;
6892 +
6893 +#ifdef CONFIG_DEBUG_FS
6894 +/* dbgaufs.c */
6895 +void dbgaufs_xino_del(struct au_branch *br);
6896 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
6897 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex, int topdown);
6898 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo);
6899 +int dbgaufs_si_init(struct au_sbinfo *sbinfo);
6900 +void dbgaufs_fin(void);
6901 +int __init dbgaufs_init(void);
6902 +#else
6903 +AuStubVoid(dbgaufs_xino_del, struct au_branch *br)
6904 +AuStubVoid(dbgaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
6905 +AuStubVoid(dbgaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex,
6906 +          int topdown)
6907 +AuStubVoid(dbgaufs_si_fin, struct au_sbinfo *sbinfo)
6908 +AuStubInt0(dbgaufs_si_init, struct au_sbinfo *sbinfo)
6909 +AuStubVoid(dbgaufs_fin, void)
6910 +AuStubInt0(__init dbgaufs_init, void)
6911 +#endif /* CONFIG_DEBUG_FS */
6912 +
6913 +#endif /* __KERNEL__ */
6914 +#endif /* __DBGAUFS_H__ */
6915 diff -urN /usr/share/empty/fs/aufs/dcsub.c linux/fs/aufs/dcsub.c
6916 --- /usr/share/empty/fs/aufs/dcsub.c    1970-01-01 01:00:00.000000000 +0100
6917 +++ linux/fs/aufs/dcsub.c       2021-12-03 15:38:59.933313976 +0100
6918 @@ -0,0 +1,225 @@
6919 +// SPDX-License-Identifier: GPL-2.0
6920 +/*
6921 + * Copyright (C) 2005-2021 Junjiro R. Okajima
6922 + *
6923 + * This program, aufs is free software; you can redistribute it and/or modify
6924 + * it under the terms of the GNU General Public License as published by
6925 + * the Free Software Foundation; either version 2 of the License, or
6926 + * (at your option) any later version.
6927 + *
6928 + * This program is distributed in the hope that it will be useful,
6929 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6930 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6931 + * GNU General Public License for more details.
6932 + *
6933 + * You should have received a copy of the GNU General Public License
6934 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6935 + */
6936 +
6937 +/*
6938 + * sub-routines for dentry cache
6939 + */
6940 +
6941 +#include "aufs.h"
6942 +
6943 +static void au_dpage_free(struct au_dpage *dpage)
6944 +{
6945 +       int i;
6946 +       struct dentry **p;
6947 +
6948 +       p = dpage->dentries;
6949 +       for (i = 0; i < dpage->ndentry; i++)
6950 +               dput(*p++);
6951 +       free_page((unsigned long)dpage->dentries);
6952 +}
6953 +
6954 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp)
6955 +{
6956 +       int err;
6957 +       void *p;
6958 +
6959 +       err = -ENOMEM;
6960 +       dpages->dpages = kmalloc(sizeof(*dpages->dpages), gfp);
6961 +       if (unlikely(!dpages->dpages))
6962 +               goto out;
6963 +
6964 +       p = (void *)__get_free_page(gfp);
6965 +       if (unlikely(!p))
6966 +               goto out_dpages;
6967 +
6968 +       dpages->dpages[0].ndentry = 0;
6969 +       dpages->dpages[0].dentries = p;
6970 +       dpages->ndpage = 1;
6971 +       return 0; /* success */
6972 +
6973 +out_dpages:
6974 +       au_kfree_try_rcu(dpages->dpages);
6975 +out:
6976 +       return err;
6977 +}
6978 +
6979 +void au_dpages_free(struct au_dcsub_pages *dpages)
6980 +{
6981 +       int i;
6982 +       struct au_dpage *p;
6983 +
6984 +       p = dpages->dpages;
6985 +       for (i = 0; i < dpages->ndpage; i++)
6986 +               au_dpage_free(p++);
6987 +       au_kfree_try_rcu(dpages->dpages);
6988 +}
6989 +
6990 +static int au_dpages_append(struct au_dcsub_pages *dpages,
6991 +                           struct dentry *dentry, gfp_t gfp)
6992 +{
6993 +       int err, sz;
6994 +       struct au_dpage *dpage;
6995 +       void *p;
6996 +
6997 +       dpage = dpages->dpages + dpages->ndpage - 1;
6998 +       sz = PAGE_SIZE / sizeof(dentry);
6999 +       if (unlikely(dpage->ndentry >= sz)) {
7000 +               AuLabel(new dpage);
7001 +               err = -ENOMEM;
7002 +               sz = dpages->ndpage * sizeof(*dpages->dpages);
7003 +               p = au_kzrealloc(dpages->dpages, sz,
7004 +                                sz + sizeof(*dpages->dpages), gfp,
7005 +                                /*may_shrink*/0);
7006 +               if (unlikely(!p))
7007 +                       goto out;
7008 +
7009 +               dpages->dpages = p;
7010 +               dpage = dpages->dpages + dpages->ndpage;
7011 +               p = (void *)__get_free_page(gfp);
7012 +               if (unlikely(!p))
7013 +                       goto out;
7014 +
7015 +               dpage->ndentry = 0;
7016 +               dpage->dentries = p;
7017 +               dpages->ndpage++;
7018 +       }
7019 +
7020 +       AuDebugOn(au_dcount(dentry) <= 0);
7021 +       dpage->dentries[dpage->ndentry++] = dget_dlock(dentry);
7022 +       return 0; /* success */
7023 +
7024 +out:
7025 +       return err;
7026 +}
7027 +
7028 +/* todo: BAD approach */
7029 +/* copied from linux/fs/dcache.c */
7030 +enum d_walk_ret {
7031 +       D_WALK_CONTINUE,
7032 +       D_WALK_QUIT,
7033 +       D_WALK_NORETRY,
7034 +       D_WALK_SKIP,
7035 +};
7036 +
7037 +extern void d_walk(struct dentry *parent, void *data,
7038 +                  enum d_walk_ret (*enter)(void *, struct dentry *));
7039 +
7040 +struct ac_dpages_arg {
7041 +       int err;
7042 +       struct au_dcsub_pages *dpages;
7043 +       struct super_block *sb;
7044 +       au_dpages_test test;
7045 +       void *arg;
7046 +};
7047 +
7048 +static enum d_walk_ret au_call_dpages_append(void *_arg, struct dentry *dentry)
7049 +{
7050 +       enum d_walk_ret ret;
7051 +       struct ac_dpages_arg *arg = _arg;
7052 +
7053 +       ret = D_WALK_CONTINUE;
7054 +       if (dentry->d_sb == arg->sb
7055 +           && !IS_ROOT(dentry)
7056 +           && au_dcount(dentry) > 0
7057 +           && au_di(dentry)
7058 +           && (!arg->test || arg->test(dentry, arg->arg))) {
7059 +               arg->err = au_dpages_append(arg->dpages, dentry, GFP_ATOMIC);
7060 +               if (unlikely(arg->err))
7061 +                       ret = D_WALK_QUIT;
7062 +       }
7063 +
7064 +       return ret;
7065 +}
7066 +
7067 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
7068 +                  au_dpages_test test, void *arg)
7069 +{
7070 +       struct ac_dpages_arg args = {
7071 +               .err    = 0,
7072 +               .dpages = dpages,
7073 +               .sb     = root->d_sb,
7074 +               .test   = test,
7075 +               .arg    = arg
7076 +       };
7077 +
7078 +       d_walk(root, &args, au_call_dpages_append);
7079 +
7080 +       return args.err;
7081 +}
7082 +
7083 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
7084 +                      int do_include, au_dpages_test test, void *arg)
7085 +{
7086 +       int err;
7087 +
7088 +       err = 0;
7089 +       write_seqlock(&rename_lock);
7090 +       spin_lock(&dentry->d_lock);
7091 +       if (do_include
7092 +           && au_dcount(dentry) > 0
7093 +           && (!test || test(dentry, arg)))
7094 +               err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
7095 +       spin_unlock(&dentry->d_lock);
7096 +       if (unlikely(err))
7097 +               goto out;
7098 +
7099 +       /*
7100 +        * RCU for vfsmount is unnecessary since this is a traverse in a single
7101 +        * mount
7102 +        */
7103 +       while (!IS_ROOT(dentry)) {
7104 +               dentry = dentry->d_parent; /* rename_lock is locked */
7105 +               spin_lock(&dentry->d_lock);
7106 +               if (au_dcount(dentry) > 0
7107 +                   && (!test || test(dentry, arg)))
7108 +                       err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
7109 +               spin_unlock(&dentry->d_lock);
7110 +               if (unlikely(err))
7111 +                       break;
7112 +       }
7113 +
7114 +out:
7115 +       write_sequnlock(&rename_lock);
7116 +       return err;
7117 +}
7118 +
7119 +static inline int au_dcsub_dpages_aufs(struct dentry *dentry, void *arg)
7120 +{
7121 +       return au_di(dentry) && dentry->d_sb == arg;
7122 +}
7123 +
7124 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
7125 +                           struct dentry *dentry, int do_include)
7126 +{
7127 +       return au_dcsub_pages_rev(dpages, dentry, do_include,
7128 +                                 au_dcsub_dpages_aufs, dentry->d_sb);
7129 +}
7130 +
7131 +int au_test_subdir(struct dentry *d1, struct dentry *d2)
7132 +{
7133 +       struct path path[2] = {
7134 +               {
7135 +                       .dentry = d1
7136 +               },
7137 +               {
7138 +                       .dentry = d2
7139 +               }
7140 +       };
7141 +
7142 +       return path_is_under(path + 0, path + 1);
7143 +}
7144 diff -urN /usr/share/empty/fs/aufs/dcsub.h linux/fs/aufs/dcsub.h
7145 --- /usr/share/empty/fs/aufs/dcsub.h    1970-01-01 01:00:00.000000000 +0100
7146 +++ linux/fs/aufs/dcsub.h       2021-12-03 15:38:59.933313976 +0100
7147 @@ -0,0 +1,137 @@
7148 +/* SPDX-License-Identifier: GPL-2.0 */
7149 +/*
7150 + * Copyright (C) 2005-2021 Junjiro R. Okajima
7151 + *
7152 + * This program, aufs is free software; you can redistribute it and/or modify
7153 + * it under the terms of the GNU General Public License as published by
7154 + * the Free Software Foundation; either version 2 of the License, or
7155 + * (at your option) any later version.
7156 + *
7157 + * This program is distributed in the hope that it will be useful,
7158 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7159 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7160 + * GNU General Public License for more details.
7161 + *
7162 + * You should have received a copy of the GNU General Public License
7163 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7164 + */
7165 +
7166 +/*
7167 + * sub-routines for dentry cache
7168 + */
7169 +
7170 +#ifndef __AUFS_DCSUB_H__
7171 +#define __AUFS_DCSUB_H__
7172 +
7173 +#ifdef __KERNEL__
7174 +
7175 +#include <linux/dcache.h>
7176 +#include <linux/fs.h>
7177 +
7178 +struct au_dpage {
7179 +       int ndentry;
7180 +       struct dentry **dentries;
7181 +};
7182 +
7183 +struct au_dcsub_pages {
7184 +       int ndpage;
7185 +       struct au_dpage *dpages;
7186 +};
7187 +
7188 +/* ---------------------------------------------------------------------- */
7189 +
7190 +/* dcsub.c */
7191 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp);
7192 +void au_dpages_free(struct au_dcsub_pages *dpages);
7193 +typedef int (*au_dpages_test)(struct dentry *dentry, void *arg);
7194 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
7195 +                  au_dpages_test test, void *arg);
7196 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
7197 +                      int do_include, au_dpages_test test, void *arg);
7198 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
7199 +                           struct dentry *dentry, int do_include);
7200 +int au_test_subdir(struct dentry *d1, struct dentry *d2);
7201 +
7202 +/* ---------------------------------------------------------------------- */
7203 +
7204 +/*
7205 + * todo: in linux-3.13, several similar (but faster) helpers are added to
7206 + * include/linux/dcache.h. Try them (in the future).
7207 + */
7208 +
7209 +static inline int au_d_hashed_positive(struct dentry *d)
7210 +{
7211 +       int err;
7212 +       struct inode *inode = d_inode(d);
7213 +
7214 +       err = 0;
7215 +       if (unlikely(d_unhashed(d)
7216 +                    || d_is_negative(d)
7217 +                    || !inode->i_nlink))
7218 +               err = -ENOENT;
7219 +       return err;
7220 +}
7221 +
7222 +static inline int au_d_linkable(struct dentry *d)
7223 +{
7224 +       int err;
7225 +       struct inode *inode = d_inode(d);
7226 +
7227 +       err = au_d_hashed_positive(d);
7228 +       if (err
7229 +           && d_is_positive(d)
7230 +           && (inode->i_state & I_LINKABLE))
7231 +               err = 0;
7232 +       return err;
7233 +}
7234 +
7235 +static inline int au_d_alive(struct dentry *d)
7236 +{
7237 +       int err;
7238 +       struct inode *inode;
7239 +
7240 +       err = 0;
7241 +       if (!IS_ROOT(d))
7242 +               err = au_d_hashed_positive(d);
7243 +       else {
7244 +               inode = d_inode(d);
7245 +               if (unlikely(d_unlinked(d)
7246 +                            || d_is_negative(d)
7247 +                            || !inode->i_nlink))
7248 +                       err = -ENOENT;
7249 +       }
7250 +       return err;
7251 +}
7252 +
7253 +static inline int au_alive_dir(struct dentry *d)
7254 +{
7255 +       int err;
7256 +
7257 +       err = au_d_alive(d);
7258 +       if (unlikely(err || IS_DEADDIR(d_inode(d))))
7259 +               err = -ENOENT;
7260 +       return err;
7261 +}
7262 +
7263 +static inline int au_qstreq(struct qstr *a, struct qstr *b)
7264 +{
7265 +       return a->len == b->len
7266 +               && !memcmp(a->name, b->name, a->len);
7267 +}
7268 +
7269 +/*
7270 + * by the commit
7271 + * 360f547 2015-01-25 dcache: let the dentry count go down to zero without
7272 + *                     taking d_lock
7273 + * the type of d_lockref.count became int, but the inlined function d_count()
7274 + * still returns unsigned int.
7275 + * I don't know why. Maybe it is for every d_count() users?
7276 + * Anyway au_dcount() lives on.
7277 + */
7278 +static inline int au_dcount(struct dentry *d)
7279 +{
7280 +       return (int)d_count(d);
7281 +}
7282 +
7283 +#endif /* __KERNEL__ */
7284 +#endif /* __AUFS_DCSUB_H__ */
7285 diff -urN /usr/share/empty/fs/aufs/debug.c linux/fs/aufs/debug.c
7286 --- /usr/share/empty/fs/aufs/debug.c    1970-01-01 01:00:00.000000000 +0100
7287 +++ linux/fs/aufs/debug.c       2021-12-03 15:38:59.933313976 +0100
7288 @@ -0,0 +1,444 @@
7289 +// SPDX-License-Identifier: GPL-2.0
7290 +/*
7291 + * Copyright (C) 2005-2021 Junjiro R. Okajima
7292 + *
7293 + * This program, aufs is free software; you can redistribute it and/or modify
7294 + * it under the terms of the GNU General Public License as published by
7295 + * the Free Software Foundation; either version 2 of the License, or
7296 + * (at your option) any later version.
7297 + *
7298 + * This program is distributed in the hope that it will be useful,
7299 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7300 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7301 + * GNU General Public License for more details.
7302 + *
7303 + * You should have received a copy of the GNU General Public License
7304 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7305 + */
7306 +
7307 +/*
7308 + * debug print functions
7309 + */
7310 +
7311 +#include <linux/iversion.h>
7312 +#include "aufs.h"
7313 +
7314 +/* Returns 0, or -errno.  arg is in kp->arg. */
7315 +static int param_atomic_t_set(const char *val, const struct kernel_param *kp)
7316 +{
7317 +       int err, n;
7318 +
7319 +       err = kstrtoint(val, 0, &n);
7320 +       if (!err) {
7321 +               if (n > 0)
7322 +                       au_debug_on();
7323 +               else
7324 +                       au_debug_off();
7325 +       }
7326 +       return err;
7327 +}
7328 +
7329 +/* Returns length written or -errno.  Buffer is 4k (ie. be short!) */
7330 +static int param_atomic_t_get(char *buffer, const struct kernel_param *kp)
7331 +{
7332 +       atomic_t *a;
7333 +
7334 +       a = kp->arg;
7335 +       return sprintf(buffer, "%d", atomic_read(a));
7336 +}
7337 +
7338 +static struct kernel_param_ops param_ops_atomic_t = {
7339 +       .set = param_atomic_t_set,
7340 +       .get = param_atomic_t_get
7341 +       /* void (*free)(void *arg) */
7342 +};
7343 +
7344 +atomic_t aufs_debug = ATOMIC_INIT(0);
7345 +MODULE_PARM_DESC(debug, "debug print");
7346 +module_param_named(debug, aufs_debug, atomic_t, 0664);
7347 +
7348 +DEFINE_MUTEX(au_dbg_mtx);      /* just to serialize the dbg msgs */
7349 +char *au_plevel = KERN_DEBUG;
7350 +#define dpri(fmt, ...) do {                                    \
7351 +       if ((au_plevel                                          \
7352 +            && strcmp(au_plevel, KERN_DEBUG))                  \
7353 +           || au_debug_test())                                 \
7354 +               printk("%s" fmt, au_plevel, ##__VA_ARGS__);     \
7355 +} while (0)
7356 +
7357 +/* ---------------------------------------------------------------------- */
7358 +
7359 +void au_dpri_whlist(struct au_nhash *whlist)
7360 +{
7361 +       unsigned long ul, n;
7362 +       struct hlist_head *head;
7363 +       struct au_vdir_wh *pos;
7364 +
7365 +       n = whlist->nh_num;
7366 +       head = whlist->nh_head;
7367 +       for (ul = 0; ul < n; ul++) {
7368 +               hlist_for_each_entry(pos, head, wh_hash)
7369 +                       dpri("b%d, %.*s, %d\n",
7370 +                            pos->wh_bindex,
7371 +                            pos->wh_str.len, pos->wh_str.name,
7372 +                            pos->wh_str.len);
7373 +               head++;
7374 +       }
7375 +}
7376 +
7377 +void au_dpri_vdir(struct au_vdir *vdir)
7378 +{
7379 +       unsigned long ul;
7380 +       union au_vdir_deblk_p p;
7381 +       unsigned char *o;
7382 +
7383 +       if (!vdir || IS_ERR(vdir)) {
7384 +               dpri("err %ld\n", PTR_ERR(vdir));
7385 +               return;
7386 +       }
7387 +
7388 +       dpri("deblk %u, nblk %lu, deblk %p, last{%lu, %p}, ver %llu\n",
7389 +            vdir->vd_deblk_sz, vdir->vd_nblk, vdir->vd_deblk,
7390 +            vdir->vd_last.ul, vdir->vd_last.p.deblk, vdir->vd_version);
7391 +       for (ul = 0; ul < vdir->vd_nblk; ul++) {
7392 +               p.deblk = vdir->vd_deblk[ul];
7393 +               o = p.deblk;
7394 +               dpri("[%lu]: %p\n", ul, o);
7395 +       }
7396 +}
7397 +
7398 +static int do_pri_inode(aufs_bindex_t bindex, struct inode *inode, int hn,
7399 +                       struct dentry *wh)
7400 +{
7401 +       char *n = NULL;
7402 +       int l = 0;
7403 +
7404 +       if (!inode || IS_ERR(inode)) {
7405 +               dpri("i%d: err %ld\n", bindex, PTR_ERR(inode));
7406 +               return -1;
7407 +       }
7408 +
7409 +       /* the type of i_blocks depends upon CONFIG_LBDAF */
7410 +       BUILD_BUG_ON(sizeof(inode->i_blocks) != sizeof(unsigned long)
7411 +                    && sizeof(inode->i_blocks) != sizeof(u64));
7412 +       if (wh) {
7413 +               n = (void *)wh->d_name.name;
7414 +               l = wh->d_name.len;
7415 +       }
7416 +
7417 +       dpri("i%d: %p, i%lu, %s, cnt %d, nl %u, 0%o, sz %llu, blk %llu,"
7418 +            " hn %d, ct %lld, np %lu, st 0x%lx, f 0x%x, v %llu, g %x%s%.*s\n",
7419 +            bindex, inode,
7420 +            inode->i_ino, inode->i_sb ? au_sbtype(inode->i_sb) : "??",
7421 +            atomic_read(&inode->i_count), inode->i_nlink, inode->i_mode,
7422 +            i_size_read(inode), (unsigned long long)inode->i_blocks,
7423 +            hn, (long long)timespec64_to_ns(&inode->i_ctime) & 0x0ffff,
7424 +            inode->i_mapping ? inode->i_mapping->nrpages : 0,
7425 +            inode->i_state, inode->i_flags, inode_peek_iversion(inode),
7426 +            inode->i_generation,
7427 +            l ? ", wh " : "", l, n);
7428 +       return 0;
7429 +}
7430 +
7431 +void au_dpri_inode(struct inode *inode)
7432 +{
7433 +       struct au_iinfo *iinfo;
7434 +       struct au_hinode *hi;
7435 +       aufs_bindex_t bindex;
7436 +       int err, hn;
7437 +
7438 +       err = do_pri_inode(-1, inode, -1, NULL);
7439 +       if (err || !au_test_aufs(inode->i_sb) || au_is_bad_inode(inode))
7440 +               return;
7441 +
7442 +       iinfo = au_ii(inode);
7443 +       dpri("i-1: btop %d, bbot %d, gen %d\n",
7444 +            iinfo->ii_btop, iinfo->ii_bbot, au_iigen(inode, NULL));
7445 +       if (iinfo->ii_btop < 0)
7446 +               return;
7447 +       hn = 0;
7448 +       for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot; bindex++) {
7449 +               hi = au_hinode(iinfo, bindex);
7450 +               hn = !!au_hn(hi);
7451 +               do_pri_inode(bindex, hi->hi_inode, hn, hi->hi_whdentry);
7452 +       }
7453 +}
7454 +
7455 +void au_dpri_dalias(struct inode *inode)
7456 +{
7457 +       struct dentry *d;
7458 +
7459 +       spin_lock(&inode->i_lock);
7460 +       hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias)
7461 +               au_dpri_dentry(d);
7462 +       spin_unlock(&inode->i_lock);
7463 +}
7464 +
7465 +static int do_pri_dentry(aufs_bindex_t bindex, struct dentry *dentry)
7466 +{
7467 +       struct dentry *wh = NULL;
7468 +       int hn;
7469 +       struct inode *inode;
7470 +       struct au_iinfo *iinfo;
7471 +       struct au_hinode *hi;
7472 +
7473 +       if (!dentry || IS_ERR(dentry)) {
7474 +               dpri("d%d: err %ld\n", bindex, PTR_ERR(dentry));
7475 +               return -1;
7476 +       }
7477 +       /* do not call dget_parent() here */
7478 +       /* note: access d_xxx without d_lock */
7479 +       dpri("d%d: %p, %pd2?, %s, cnt %d, flags 0x%x, %shashed\n",
7480 +            bindex, dentry, dentry,
7481 +            dentry->d_sb ? au_sbtype(dentry->d_sb) : "??",
7482 +            au_dcount(dentry), dentry->d_flags,
7483 +            d_unhashed(dentry) ? "un" : "");
7484 +       hn = -1;
7485 +       inode = NULL;
7486 +       if (d_is_positive(dentry))
7487 +               inode = d_inode(dentry);
7488 +       if (inode
7489 +           && au_test_aufs(dentry->d_sb)
7490 +           && bindex >= 0
7491 +           && !au_is_bad_inode(inode)) {
7492 +               iinfo = au_ii(inode);
7493 +               hi = au_hinode(iinfo, bindex);
7494 +               hn = !!au_hn(hi);
7495 +               wh = hi->hi_whdentry;
7496 +       }
7497 +       do_pri_inode(bindex, inode, hn, wh);
7498 +       return 0;
7499 +}
7500 +
7501 +void au_dpri_dentry(struct dentry *dentry)
7502 +{
7503 +       struct au_dinfo *dinfo;
7504 +       aufs_bindex_t bindex;
7505 +       int err;
7506 +
7507 +       err = do_pri_dentry(-1, dentry);
7508 +       if (err || !au_test_aufs(dentry->d_sb))
7509 +               return;
7510 +
7511 +       dinfo = au_di(dentry);
7512 +       if (!dinfo)
7513 +               return;
7514 +       dpri("d-1: btop %d, bbot %d, bwh %d, bdiropq %d, gen %d, tmp %d\n",
7515 +            dinfo->di_btop, dinfo->di_bbot,
7516 +            dinfo->di_bwh, dinfo->di_bdiropq, au_digen(dentry),
7517 +            dinfo->di_tmpfile);
7518 +       if (dinfo->di_btop < 0)
7519 +               return;
7520 +       for (bindex = dinfo->di_btop; bindex <= dinfo->di_bbot; bindex++)
7521 +               do_pri_dentry(bindex, au_hdentry(dinfo, bindex)->hd_dentry);
7522 +}
7523 +
7524 +static int do_pri_file(aufs_bindex_t bindex, struct file *file)
7525 +{
7526 +       char a[32];
7527 +
7528 +       if (!file || IS_ERR(file)) {
7529 +               dpri("f%d: err %ld\n", bindex, PTR_ERR(file));
7530 +               return -1;
7531 +       }
7532 +       a[0] = 0;
7533 +       if (bindex < 0
7534 +           && !IS_ERR_OR_NULL(file->f_path.dentry)
7535 +           && au_test_aufs(file->f_path.dentry->d_sb)
7536 +           && au_fi(file))
7537 +               snprintf(a, sizeof(a), ", gen %d, mmapped %d",
7538 +                        au_figen(file), atomic_read(&au_fi(file)->fi_mmapped));
7539 +       dpri("f%d: mode 0x%x, flags 0%o, cnt %ld, v %llu, pos %llu%s\n",
7540 +            bindex, file->f_mode, file->f_flags, (long)file_count(file),
7541 +            file->f_version, file->f_pos, a);
7542 +       if (!IS_ERR_OR_NULL(file->f_path.dentry))
7543 +               do_pri_dentry(bindex, file->f_path.dentry);
7544 +       return 0;
7545 +}
7546 +
7547 +void au_dpri_file(struct file *file)
7548 +{
7549 +       struct au_finfo *finfo;
7550 +       struct au_fidir *fidir;
7551 +       struct au_hfile *hfile;
7552 +       aufs_bindex_t bindex;
7553 +       int err;
7554 +
7555 +       err = do_pri_file(-1, file);
7556 +       if (err
7557 +           || IS_ERR_OR_NULL(file->f_path.dentry)
7558 +           || !au_test_aufs(file->f_path.dentry->d_sb))
7559 +               return;
7560 +
7561 +       finfo = au_fi(file);
7562 +       if (!finfo)
7563 +               return;
7564 +       if (finfo->fi_btop < 0)
7565 +               return;
7566 +       fidir = finfo->fi_hdir;
7567 +       if (!fidir)
7568 +               do_pri_file(finfo->fi_btop, finfo->fi_htop.hf_file);
7569 +       else
7570 +               for (bindex = finfo->fi_btop;
7571 +                    bindex >= 0 && bindex <= fidir->fd_bbot;
7572 +                    bindex++) {
7573 +                       hfile = fidir->fd_hfile + bindex;
7574 +                       do_pri_file(bindex, hfile ? hfile->hf_file : NULL);
7575 +               }
7576 +}
7577 +
7578 +static int do_pri_br(aufs_bindex_t bindex, struct au_branch *br)
7579 +{
7580 +       struct vfsmount *mnt;
7581 +       struct super_block *sb;
7582 +
7583 +       if (!br || IS_ERR(br))
7584 +               goto out;
7585 +       mnt = au_br_mnt(br);
7586 +       if (!mnt || IS_ERR(mnt))
7587 +               goto out;
7588 +       sb = mnt->mnt_sb;
7589 +       if (!sb || IS_ERR(sb))
7590 +               goto out;
7591 +
7592 +       dpri("s%d: {perm 0x%x, id %d, wbr %p}, "
7593 +            "%s, dev 0x%02x%02x, flags 0x%lx, cnt %d, active %d, "
7594 +            "xino %d\n",
7595 +            bindex, br->br_perm, br->br_id, br->br_wbr,
7596 +            au_sbtype(sb), MAJOR(sb->s_dev), MINOR(sb->s_dev),
7597 +            sb->s_flags, sb->s_count,
7598 +            atomic_read(&sb->s_active),
7599 +            !!au_xino_file(br->br_xino, /*idx*/-1));
7600 +       return 0;
7601 +
7602 +out:
7603 +       dpri("s%d: err %ld\n", bindex, PTR_ERR(br));
7604 +       return -1;
7605 +}
7606 +
7607 +void au_dpri_sb(struct super_block *sb)
7608 +{
7609 +       struct au_sbinfo *sbinfo;
7610 +       aufs_bindex_t bindex;
7611 +       int err;
7612 +       /* to reduce stack size */
7613 +       struct {
7614 +               struct vfsmount mnt;
7615 +               struct au_branch fake;
7616 +       } *a;
7617 +
7618 +       /* this function can be called from magic sysrq */
7619 +       a = kzalloc(sizeof(*a), GFP_ATOMIC);
7620 +       if (unlikely(!a)) {
7621 +               dpri("no memory\n");
7622 +               return;
7623 +       }
7624 +
7625 +       a->mnt.mnt_sb = sb;
7626 +       a->fake.br_path.mnt = &a->mnt;
7627 +       err = do_pri_br(-1, &a->fake);
7628 +       au_kfree_rcu(a);
7629 +       dpri("dev 0x%x\n", sb->s_dev);
7630 +       if (err || !au_test_aufs(sb))
7631 +               return;
7632 +
7633 +       sbinfo = au_sbi(sb);
7634 +       if (!sbinfo)
7635 +               return;
7636 +       dpri("nw %d, gen %u, kobj %d\n",
7637 +            atomic_read(&sbinfo->si_nowait.nw_len), sbinfo->si_generation,
7638 +            kref_read(&sbinfo->si_kobj.kref));
7639 +       for (bindex = 0; bindex <= sbinfo->si_bbot; bindex++)
7640 +               do_pri_br(bindex, sbinfo->si_branch[0 + bindex]);
7641 +}
7642 +
7643 +/* ---------------------------------------------------------------------- */
7644 +
7645 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line)
7646 +{
7647 +       struct inode *h_inode, *inode = d_inode(dentry);
7648 +       struct dentry *h_dentry;
7649 +       aufs_bindex_t bindex, bbot, bi;
7650 +
7651 +       if (!inode /* || au_di(dentry)->di_lsc == AuLsc_DI_TMP */)
7652 +               return;
7653 +
7654 +       bbot = au_dbbot(dentry);
7655 +       bi = au_ibbot(inode);
7656 +       if (bi < bbot)
7657 +               bbot = bi;
7658 +       bindex = au_dbtop(dentry);
7659 +       bi = au_ibtop(inode);
7660 +       if (bi > bindex)
7661 +               bindex = bi;
7662 +
7663 +       for (; bindex <= bbot; bindex++) {
7664 +               h_dentry = au_h_dptr(dentry, bindex);
7665 +               if (!h_dentry)
7666 +                       continue;
7667 +               h_inode = au_h_iptr(inode, bindex);
7668 +               if (unlikely(h_inode != d_inode(h_dentry))) {
7669 +                       au_debug_on();
7670 +                       AuDbg("b%d, %s:%d\n", bindex, func, line);
7671 +                       AuDbgDentry(dentry);
7672 +                       AuDbgInode(inode);
7673 +                       au_debug_off();
7674 +                       if (au_test_fuse(h_inode->i_sb))
7675 +                               WARN_ON_ONCE(1);
7676 +                       else
7677 +                               BUG();
7678 +               }
7679 +       }
7680 +}
7681 +
7682 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen)
7683 +{
7684 +       int err, i, j;
7685 +       struct au_dcsub_pages dpages;
7686 +       struct au_dpage *dpage;
7687 +       struct dentry **dentries;
7688 +
7689 +       err = au_dpages_init(&dpages, GFP_NOFS);
7690 +       AuDebugOn(err);
7691 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/1);
7692 +       AuDebugOn(err);
7693 +       for (i = dpages.ndpage - 1; !err && i >= 0; i--) {
7694 +               dpage = dpages.dpages + i;
7695 +               dentries = dpage->dentries;
7696 +               for (j = dpage->ndentry - 1; !err && j >= 0; j--)
7697 +                       AuDebugOn(au_digen_test(dentries[j], sigen));
7698 +       }
7699 +       au_dpages_free(&dpages);
7700 +}
7701 +
7702 +void au_dbg_verify_kthread(void)
7703 +{
7704 +       if (au_wkq_test()) {
7705 +               au_dbg_blocked();
7706 +               /*
7707 +                * It may be recursive, but udba=notify between two aufs mounts,
7708 +                * where a single ro branch is shared, is not a problem.
7709 +                */
7710 +               /* WARN_ON(1); */
7711 +       }
7712 +}
7713 +
7714 +/* ---------------------------------------------------------------------- */
7715 +
7716 +int __init au_debug_init(void)
7717 +{
7718 +       aufs_bindex_t bindex;
7719 +       struct au_vdir_destr destr;
7720 +
7721 +       bindex = -1;
7722 +       AuDebugOn(bindex >= 0);
7723 +
7724 +       destr.len = -1;
7725 +       AuDebugOn(destr.len < NAME_MAX);
7726 +
7727 +#ifdef CONFIG_4KSTACKS
7728 +       pr_warn("CONFIG_4KSTACKS is defined.\n");
7729 +#endif
7730 +
7731 +       return 0;
7732 +}
7733 diff -urN /usr/share/empty/fs/aufs/debug.h linux/fs/aufs/debug.h
7734 --- /usr/share/empty/fs/aufs/debug.h    1970-01-01 01:00:00.000000000 +0100
7735 +++ linux/fs/aufs/debug.h       2021-12-03 15:38:59.933313976 +0100
7736 @@ -0,0 +1,226 @@
7737 +/* SPDX-License-Identifier: GPL-2.0 */
7738 +/*
7739 + * Copyright (C) 2005-2021 Junjiro R. Okajima
7740 + *
7741 + * This program, aufs is free software; you can redistribute it and/or modify
7742 + * it under the terms of the GNU General Public License as published by
7743 + * the Free Software Foundation; either version 2 of the License, or
7744 + * (at your option) any later version.
7745 + *
7746 + * This program is distributed in the hope that it will be useful,
7747 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7748 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7749 + * GNU General Public License for more details.
7750 + *
7751 + * You should have received a copy of the GNU General Public License
7752 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7753 + */
7754 +
7755 +/*
7756 + * debug print functions
7757 + */
7758 +
7759 +#ifndef __AUFS_DEBUG_H__
7760 +#define __AUFS_DEBUG_H__
7761 +
7762 +#ifdef __KERNEL__
7763 +
7764 +#include <linux/atomic.h>
7765 +#include <linux/module.h>
7766 +#include <linux/kallsyms.h>
7767 +#include <linux/sysrq.h>
7768 +
7769 +#ifdef CONFIG_AUFS_DEBUG
7770 +#define AuDebugOn(a)           BUG_ON(a)
7771 +
7772 +/* module parameter */
7773 +extern atomic_t aufs_debug;
7774 +static inline void au_debug_on(void)
7775 +{
7776 +       atomic_inc(&aufs_debug);
7777 +}
7778 +static inline void au_debug_off(void)
7779 +{
7780 +       atomic_dec_if_positive(&aufs_debug);
7781 +}
7782 +
7783 +static inline int au_debug_test(void)
7784 +{
7785 +       return atomic_read(&aufs_debug) > 0;
7786 +}
7787 +#else
7788 +#define AuDebugOn(a)           do {} while (0)
7789 +AuStubVoid(au_debug_on, void)
7790 +AuStubVoid(au_debug_off, void)
7791 +AuStubInt0(au_debug_test, void)
7792 +#endif /* CONFIG_AUFS_DEBUG */
7793 +
7794 +#define param_check_atomic_t(name, p) __param_check(name, p, atomic_t)
7795 +
7796 +/* ---------------------------------------------------------------------- */
7797 +
7798 +/* debug print */
7799 +
7800 +#define AuDbg(fmt, ...) do { \
7801 +       if (au_debug_test()) \
7802 +               pr_debug("DEBUG: " fmt, ##__VA_ARGS__); \
7803 +} while (0)
7804 +#define AuLabel(l)             AuDbg(#l "\n")
7805 +#define AuIOErr(fmt, ...)      pr_err("I/O Error, " fmt, ##__VA_ARGS__)
7806 +#define AuWarn1(fmt, ...) do { \
7807 +       static unsigned char _c; \
7808 +       if (!_c++) \
7809 +               pr_warn(fmt, ##__VA_ARGS__); \
7810 +} while (0)
7811 +
7812 +#define AuErr1(fmt, ...) do { \
7813 +       static unsigned char _c; \
7814 +       if (!_c++) \
7815 +               pr_err(fmt, ##__VA_ARGS__); \
7816 +} while (0)
7817 +
7818 +#define AuIOErr1(fmt, ...) do { \
7819 +       static unsigned char _c; \
7820 +       if (!_c++) \
7821 +               AuIOErr(fmt, ##__VA_ARGS__); \
7822 +} while (0)
7823 +
7824 +#define AuUnsupportMsg "This operation is not supported." \
7825 +                       " Please report this application to aufs-users ML."
7826 +#define AuUnsupport(fmt, ...) do { \
7827 +       pr_err(AuUnsupportMsg "\n" fmt, ##__VA_ARGS__); \
7828 +       dump_stack(); \
7829 +} while (0)
7830 +
7831 +#define AuTraceErr(e) do { \
7832 +       if (unlikely((e) < 0)) \
7833 +               AuDbg("err %d\n", (int)(e)); \
7834 +} while (0)
7835 +
7836 +#define AuTraceErrPtr(p) do { \
7837 +       if (IS_ERR(p)) \
7838 +               AuDbg("err %ld\n", PTR_ERR(p)); \
7839 +} while (0)
7840 +
7841 +/* dirty macros for debug print, use with "%.*s" and caution */
7842 +#define AuLNPair(qstr)         (qstr)->len, (qstr)->name
7843 +
7844 +/* ---------------------------------------------------------------------- */
7845 +
7846 +struct dentry;
7847 +#ifdef CONFIG_AUFS_DEBUG
7848 +extern struct mutex au_dbg_mtx;
7849 +extern char *au_plevel;
7850 +struct au_nhash;
7851 +void au_dpri_whlist(struct au_nhash *whlist);
7852 +struct au_vdir;
7853 +void au_dpri_vdir(struct au_vdir *vdir);
7854 +struct inode;
7855 +void au_dpri_inode(struct inode *inode);
7856 +void au_dpri_dalias(struct inode *inode);
7857 +void au_dpri_dentry(struct dentry *dentry);
7858 +struct file;
7859 +void au_dpri_file(struct file *filp);
7860 +struct super_block;
7861 +void au_dpri_sb(struct super_block *sb);
7862 +
7863 +#define au_dbg_verify_dinode(d) __au_dbg_verify_dinode(d, __func__, __LINE__)
7864 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
7865 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen);
7866 +void au_dbg_verify_kthread(void);
7867 +
7868 +int __init au_debug_init(void);
7869 +
7870 +#define AuDbgWhlist(w) do { \
7871 +       mutex_lock(&au_dbg_mtx); \
7872 +       AuDbg(#w "\n"); \
7873 +       au_dpri_whlist(w); \
7874 +       mutex_unlock(&au_dbg_mtx); \
7875 +} while (0)
7876 +
7877 +#define AuDbgVdir(v) do { \
7878 +       mutex_lock(&au_dbg_mtx); \
7879 +       AuDbg(#v "\n"); \
7880 +       au_dpri_vdir(v); \
7881 +       mutex_unlock(&au_dbg_mtx); \
7882 +} while (0)
7883 +
7884 +#define AuDbgInode(i) do { \
7885 +       mutex_lock(&au_dbg_mtx); \
7886 +       AuDbg(#i "\n"); \
7887 +       au_dpri_inode(i); \
7888 +       mutex_unlock(&au_dbg_mtx); \
7889 +} while (0)
7890 +
7891 +#define AuDbgDAlias(i) do { \
7892 +       mutex_lock(&au_dbg_mtx); \
7893 +       AuDbg(#i "\n"); \
7894 +       au_dpri_dalias(i); \
7895 +       mutex_unlock(&au_dbg_mtx); \
7896 +} while (0)
7897 +
7898 +#define AuDbgDentry(d) do { \
7899 +       mutex_lock(&au_dbg_mtx); \
7900 +       AuDbg(#d "\n"); \
7901 +       au_dpri_dentry(d); \
7902 +       mutex_unlock(&au_dbg_mtx); \
7903 +} while (0)
7904 +
7905 +#define AuDbgFile(f) do { \
7906 +       mutex_lock(&au_dbg_mtx); \
7907 +       AuDbg(#f "\n"); \
7908 +       au_dpri_file(f); \
7909 +       mutex_unlock(&au_dbg_mtx); \
7910 +} while (0)
7911 +
7912 +#define AuDbgSb(sb) do { \
7913 +       mutex_lock(&au_dbg_mtx); \
7914 +       AuDbg(#sb "\n"); \
7915 +       au_dpri_sb(sb); \
7916 +       mutex_unlock(&au_dbg_mtx); \
7917 +} while (0)
7918 +
7919 +#define AuDbgSym(addr) do {                            \
7920 +       char sym[KSYM_SYMBOL_LEN];                      \
7921 +       sprint_symbol(sym, (unsigned long)addr);        \
7922 +       AuDbg("%s\n", sym);                             \
7923 +} while (0)
7924 +#else
7925 +AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
7926 +AuStubVoid(au_dbg_verify_gen, struct dentry *parent, unsigned int sigen)
7927 +AuStubVoid(au_dbg_verify_kthread, void)
7928 +AuStubInt0(__init au_debug_init, void)
7929 +
7930 +#define AuDbgWhlist(w)         do {} while (0)
7931 +#define AuDbgVdir(v)           do {} while (0)
7932 +#define AuDbgInode(i)          do {} while (0)
7933 +#define AuDbgDAlias(i)         do {} while (0)
7934 +#define AuDbgDentry(d)         do {} while (0)
7935 +#define AuDbgFile(f)           do {} while (0)
7936 +#define AuDbgSb(sb)            do {} while (0)
7937 +#define AuDbgSym(addr)         do {} while (0)
7938 +#endif /* CONFIG_AUFS_DEBUG */
7939 +
7940 +/* ---------------------------------------------------------------------- */
7941 +
7942 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
7943 +int __init au_sysrq_init(void);
7944 +void au_sysrq_fin(void);
7945 +
7946 +#ifdef CONFIG_HW_CONSOLE
7947 +#define au_dbg_blocked() do { \
7948 +       WARN_ON(1); \
7949 +       handle_sysrq('w'); \
7950 +} while (0)
7951 +#else
7952 +AuStubVoid(au_dbg_blocked, void)
7953 +#endif
7954 +
7955 +#else
7956 +AuStubInt0(__init au_sysrq_init, void)
7957 +AuStubVoid(au_sysrq_fin, void)
7958 +AuStubVoid(au_dbg_blocked, void)
7959 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
7960 +
7961 +#endif /* __KERNEL__ */
7962 +#endif /* __AUFS_DEBUG_H__ */
7963 diff -urN /usr/share/empty/fs/aufs/dentry.c linux/fs/aufs/dentry.c
7964 --- /usr/share/empty/fs/aufs/dentry.c   1970-01-01 01:00:00.000000000 +0100
7965 +++ linux/fs/aufs/dentry.c      2021-12-03 15:40:58.233313963 +0100
7966 @@ -0,0 +1,1169 @@
7967 +// SPDX-License-Identifier: GPL-2.0
7968 +/*
7969 + * Copyright (C) 2005-2021 Junjiro R. Okajima
7970 + *
7971 + * This program, aufs is free software; you can redistribute it and/or modify
7972 + * it under the terms of the GNU General Public License as published by
7973 + * the Free Software Foundation; either version 2 of the License, or
7974 + * (at your option) any later version.
7975 + *
7976 + * This program is distributed in the hope that it will be useful,
7977 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7978 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7979 + * GNU General Public License for more details.
7980 + *
7981 + * You should have received a copy of the GNU General Public License
7982 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7983 + */
7984 +
7985 +/*
7986 + * lookup and dentry operations
7987 + */
7988 +
7989 +#include <linux/iversion.h>
7990 +#include <linux/namei.h>
7991 +#include "aufs.h"
7992 +
7993 +/*
7994 + * returns positive/negative dentry, NULL or an error.
7995 + * NULL means whiteout-ed or not-found.
7996 + */
7997 +static struct dentry*
7998 +au_do_lookup(struct dentry *h_parent, struct dentry *dentry,
7999 +            aufs_bindex_t bindex, struct au_do_lookup_args *args)
8000 +{
8001 +       struct dentry *h_dentry;
8002 +       struct inode *h_inode;
8003 +       struct au_branch *br;
8004 +       struct user_namespace *h_userns;
8005 +       struct path h_path;
8006 +       int wh_found, opq;
8007 +       unsigned char wh_able;
8008 +       const unsigned char allow_neg = !!au_ftest_lkup(args->flags, ALLOW_NEG);
8009 +       const unsigned char ignore_perm = !!au_ftest_lkup(args->flags,
8010 +                                                         IGNORE_PERM);
8011 +
8012 +       wh_found = 0;
8013 +       br = au_sbr(dentry->d_sb, bindex);
8014 +       h_path.dentry = h_parent;
8015 +       h_path.mnt = au_br_mnt(br);
8016 +       h_userns = au_br_userns(br);
8017 +       wh_able = !!au_br_whable(br->br_perm);
8018 +       if (wh_able)
8019 +               wh_found = au_wh_test(h_userns, &h_path, &args->whname,
8020 +                                     ignore_perm);
8021 +       h_dentry = ERR_PTR(wh_found);
8022 +       if (!wh_found)
8023 +               goto real_lookup;
8024 +       if (unlikely(wh_found < 0))
8025 +               goto out;
8026 +
8027 +       /* We found a whiteout */
8028 +       /* au_set_dbbot(dentry, bindex); */
8029 +       au_set_dbwh(dentry, bindex);
8030 +       if (!allow_neg)
8031 +               return NULL; /* success */
8032 +
8033 +real_lookup:
8034 +       if (!ignore_perm)
8035 +               h_dentry = vfsub_lkup_one(args->name, &h_path);
8036 +       else
8037 +               h_dentry = au_sio_lkup_one(h_userns, args->name, &h_path);
8038 +       if (IS_ERR(h_dentry)) {
8039 +               if (PTR_ERR(h_dentry) == -ENAMETOOLONG
8040 +                   && !allow_neg)
8041 +                       h_dentry = NULL;
8042 +               goto out;
8043 +       }
8044 +
8045 +       h_inode = d_inode(h_dentry);
8046 +       if (d_is_negative(h_dentry)) {
8047 +               if (!allow_neg)
8048 +                       goto out_neg;
8049 +       } else if (wh_found
8050 +                  || (args->type && args->type != (h_inode->i_mode & S_IFMT)))
8051 +               goto out_neg;
8052 +       else if (au_ftest_lkup(args->flags, DIRREN)
8053 +                /* && h_inode */
8054 +                && !au_dr_lkup_h_ino(args, bindex, h_inode->i_ino)) {
8055 +               AuDbg("b%d %pd ignored hi%llu\n", bindex, h_dentry,
8056 +                     (unsigned long long)h_inode->i_ino);
8057 +               goto out_neg;
8058 +       }
8059 +
8060 +       if (au_dbbot(dentry) <= bindex)
8061 +               au_set_dbbot(dentry, bindex);
8062 +       if (au_dbtop(dentry) < 0 || bindex < au_dbtop(dentry))
8063 +               au_set_dbtop(dentry, bindex);
8064 +       au_set_h_dptr(dentry, bindex, h_dentry);
8065 +
8066 +       if (!d_is_dir(h_dentry)
8067 +           || !wh_able
8068 +           || (d_really_is_positive(dentry) && !d_is_dir(dentry)))
8069 +               goto out; /* success */
8070 +
8071 +       h_path.dentry = h_dentry;
8072 +       inode_lock_shared_nested(h_inode, AuLsc_I_CHILD);
8073 +       opq = au_diropq_test(h_userns, &h_path);
8074 +       inode_unlock_shared(h_inode);
8075 +       if (opq > 0)
8076 +               au_set_dbdiropq(dentry, bindex);
8077 +       else if (unlikely(opq < 0)) {
8078 +               au_set_h_dptr(dentry, bindex, NULL);
8079 +               h_dentry = ERR_PTR(opq);
8080 +       }
8081 +       goto out;
8082 +
8083 +out_neg:
8084 +       dput(h_dentry);
8085 +       h_dentry = NULL;
8086 +out:
8087 +       return h_dentry;
8088 +}
8089 +
8090 +static int au_test_shwh(struct super_block *sb, const struct qstr *name)
8091 +{
8092 +       if (unlikely(!au_opt_test(au_mntflags(sb), SHWH)
8093 +                    && !strncmp(name->name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)))
8094 +               return -EPERM;
8095 +       return 0;
8096 +}
8097 +
8098 +/*
8099 + * returns the number of lower positive dentries,
8100 + * otherwise an error.
8101 + * can be called at unlinking with @type is zero.
8102 + */
8103 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t btop,
8104 +                  unsigned int flags)
8105 +{
8106 +       int npositive, err;
8107 +       aufs_bindex_t bindex, btail, bdiropq;
8108 +       unsigned char isdir, dirperm1, dirren;
8109 +       struct au_do_lookup_args args = {
8110 +               .flags          = flags,
8111 +               .name           = &dentry->d_name
8112 +       };
8113 +       struct dentry *parent;
8114 +       struct super_block *sb;
8115 +
8116 +       sb = dentry->d_sb;
8117 +       err = au_test_shwh(sb, args.name);
8118 +       if (unlikely(err))
8119 +               goto out;
8120 +
8121 +       err = au_wh_name_alloc(&args.whname, args.name);
8122 +       if (unlikely(err))
8123 +               goto out;
8124 +
8125 +       isdir = !!d_is_dir(dentry);
8126 +       dirperm1 = !!au_opt_test(au_mntflags(sb), DIRPERM1);
8127 +       dirren = !!au_opt_test(au_mntflags(sb), DIRREN);
8128 +       if (dirren)
8129 +               au_fset_lkup(args.flags, DIRREN);
8130 +
8131 +       npositive = 0;
8132 +       parent = dget_parent(dentry);
8133 +       btail = au_dbtaildir(parent);
8134 +       for (bindex = btop; bindex <= btail; bindex++) {
8135 +               struct dentry *h_parent, *h_dentry;
8136 +               struct inode *h_inode, *h_dir;
8137 +               struct au_branch *br;
8138 +
8139 +               h_dentry = au_h_dptr(dentry, bindex);
8140 +               if (h_dentry) {
8141 +                       if (d_is_positive(h_dentry))
8142 +                               npositive++;
8143 +                       break;
8144 +               }
8145 +               h_parent = au_h_dptr(parent, bindex);
8146 +               if (!h_parent || !d_is_dir(h_parent))
8147 +                       continue;
8148 +
8149 +               if (dirren) {
8150 +                       /* if the inum matches, then use the prepared name */
8151 +                       err = au_dr_lkup_name(&args, bindex);
8152 +                       if (unlikely(err))
8153 +                               goto out_parent;
8154 +               }
8155 +
8156 +               h_dir = d_inode(h_parent);
8157 +               inode_lock_shared_nested(h_dir, AuLsc_I_PARENT);
8158 +               h_dentry = au_do_lookup(h_parent, dentry, bindex, &args);
8159 +               inode_unlock_shared(h_dir);
8160 +               err = PTR_ERR(h_dentry);
8161 +               if (IS_ERR(h_dentry))
8162 +                       goto out_parent;
8163 +               if (h_dentry)
8164 +                       au_fclr_lkup(args.flags, ALLOW_NEG);
8165 +               if (dirperm1)
8166 +                       au_fset_lkup(args.flags, IGNORE_PERM);
8167 +
8168 +               if (au_dbwh(dentry) == bindex)
8169 +                       break;
8170 +               if (!h_dentry)
8171 +                       continue;
8172 +               if (d_is_negative(h_dentry))
8173 +                       continue;
8174 +               h_inode = d_inode(h_dentry);
8175 +               npositive++;
8176 +               if (!args.type)
8177 +                       args.type = h_inode->i_mode & S_IFMT;
8178 +               if (args.type != S_IFDIR)
8179 +                       break;
8180 +               else if (isdir) {
8181 +                       /* the type of lower may be different */
8182 +                       bdiropq = au_dbdiropq(dentry);
8183 +                       if (bdiropq >= 0 && bdiropq <= bindex)
8184 +                               break;
8185 +               }
8186 +               br = au_sbr(sb, bindex);
8187 +               if (dirren
8188 +                   && au_dr_hino_test_add(&br->br_dirren, h_inode->i_ino,
8189 +                                          /*add_ent*/NULL)) {
8190 +                       /* prepare next name to lookup */
8191 +                       err = au_dr_lkup(&args, dentry, bindex);
8192 +                       if (unlikely(err))
8193 +                               goto out_parent;
8194 +               }
8195 +       }
8196 +
8197 +       if (npositive) {
8198 +               AuLabel(positive);
8199 +               au_update_dbtop(dentry);
8200 +       }
8201 +       err = npositive;
8202 +       if (unlikely(!au_opt_test(au_mntflags(sb), UDBA_NONE)
8203 +                    && au_dbtop(dentry) < 0)) {
8204 +               err = -EIO;
8205 +               AuIOErr("both of real entry and whiteout found, %pd, err %d\n",
8206 +                       dentry, err);
8207 +       }
8208 +
8209 +out_parent:
8210 +       dput(parent);
8211 +       au_kfree_try_rcu(args.whname.name);
8212 +       if (dirren)
8213 +               au_dr_lkup_fin(&args);
8214 +out:
8215 +       return err;
8216 +}
8217 +
8218 +struct dentry *au_sio_lkup_one(struct user_namespace *userns, struct qstr *name,
8219 +                              struct path *ppath)
8220 +{
8221 +       struct dentry *dentry;
8222 +       int wkq_err;
8223 +
8224 +       if (!au_test_h_perm_sio(userns, d_inode(ppath->dentry), MAY_EXEC))
8225 +               dentry = vfsub_lkup_one(name, ppath);
8226 +       else {
8227 +               struct vfsub_lkup_one_args args = {
8228 +                       .errp   = &dentry,
8229 +                       .name   = name,
8230 +                       .ppath  = ppath
8231 +               };
8232 +
8233 +               wkq_err = au_wkq_wait(vfsub_call_lkup_one, &args);
8234 +               if (unlikely(wkq_err))
8235 +                       dentry = ERR_PTR(wkq_err);
8236 +       }
8237 +
8238 +       return dentry;
8239 +}
8240 +
8241 +/*
8242 + * lookup @dentry on @bindex which should be negative.
8243 + */
8244 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh)
8245 +{
8246 +       int err;
8247 +       struct dentry *parent, *h_dentry;
8248 +       struct au_branch *br;
8249 +       struct user_namespace *h_userns;
8250 +       struct path h_ppath;
8251 +
8252 +       parent = dget_parent(dentry);
8253 +       br = au_sbr(dentry->d_sb, bindex);
8254 +       h_ppath.dentry = au_h_dptr(parent, bindex);
8255 +       h_ppath.mnt = au_br_mnt(br);
8256 +       h_userns = au_br_userns(br);
8257 +       if (wh)
8258 +               h_dentry = au_whtmp_lkup(h_ppath.dentry, br, &dentry->d_name);
8259 +       else
8260 +               h_dentry = au_sio_lkup_one(h_userns, &dentry->d_name, &h_ppath);
8261 +       err = PTR_ERR(h_dentry);
8262 +       if (IS_ERR(h_dentry))
8263 +               goto out;
8264 +       if (unlikely(d_is_positive(h_dentry))) {
8265 +               err = -EIO;
8266 +               AuIOErr("%pd should be negative on b%d.\n", h_dentry, bindex);
8267 +               dput(h_dentry);
8268 +               goto out;
8269 +       }
8270 +
8271 +       err = 0;
8272 +       if (bindex < au_dbtop(dentry))
8273 +               au_set_dbtop(dentry, bindex);
8274 +       if (au_dbbot(dentry) < bindex)
8275 +               au_set_dbbot(dentry, bindex);
8276 +       au_set_h_dptr(dentry, bindex, h_dentry);
8277 +
8278 +out:
8279 +       dput(parent);
8280 +       return err;
8281 +}
8282 +
8283 +/* ---------------------------------------------------------------------- */
8284 +
8285 +/* subset of struct inode */
8286 +struct au_iattr {
8287 +       unsigned long           i_ino;
8288 +       /* unsigned int         i_nlink; */
8289 +       kuid_t                  i_uid;
8290 +       kgid_t                  i_gid;
8291 +       u64                     i_version;
8292 +/*
8293 +       loff_t                  i_size;
8294 +       blkcnt_t                i_blocks;
8295 +*/
8296 +       umode_t                 i_mode;
8297 +};
8298 +
8299 +static void au_iattr_save(struct au_iattr *ia, struct inode *h_inode)
8300 +{
8301 +       ia->i_ino = h_inode->i_ino;
8302 +       /* ia->i_nlink = h_inode->i_nlink; */
8303 +       ia->i_uid = h_inode->i_uid;
8304 +       ia->i_gid = h_inode->i_gid;
8305 +       ia->i_version = inode_query_iversion(h_inode);
8306 +/*
8307 +       ia->i_size = h_inode->i_size;
8308 +       ia->i_blocks = h_inode->i_blocks;
8309 +*/
8310 +       ia->i_mode = (h_inode->i_mode & S_IFMT);
8311 +}
8312 +
8313 +static int au_iattr_test(struct au_iattr *ia, struct inode *h_inode)
8314 +{
8315 +       return ia->i_ino != h_inode->i_ino
8316 +               /* || ia->i_nlink != h_inode->i_nlink */
8317 +               || !uid_eq(ia->i_uid, h_inode->i_uid)
8318 +               || !gid_eq(ia->i_gid, h_inode->i_gid)
8319 +               || !inode_eq_iversion(h_inode, ia->i_version)
8320 +/*
8321 +               || ia->i_size != h_inode->i_size
8322 +               || ia->i_blocks != h_inode->i_blocks
8323 +*/
8324 +               || ia->i_mode != (h_inode->i_mode & S_IFMT);
8325 +}
8326 +
8327 +static int au_h_verify_dentry(struct dentry *h_dentry, struct dentry *h_parent,
8328 +                             struct au_branch *br)
8329 +{
8330 +       int err;
8331 +       struct au_iattr ia;
8332 +       struct inode *h_inode;
8333 +       struct dentry *h_d;
8334 +       struct super_block *h_sb;
8335 +       struct path h_ppath;
8336 +
8337 +       err = 0;
8338 +       memset(&ia, -1, sizeof(ia));
8339 +       h_sb = h_dentry->d_sb;
8340 +       h_inode = NULL;
8341 +       if (d_is_positive(h_dentry)) {
8342 +               h_inode = d_inode(h_dentry);
8343 +               au_iattr_save(&ia, h_inode);
8344 +       } else if (au_test_nfs(h_sb) || au_test_fuse(h_sb))
8345 +               /* nfs d_revalidate may return 0 for negative dentry */
8346 +               /* fuse d_revalidate always return 0 for negative dentry */
8347 +               goto out;
8348 +
8349 +       /* main purpose is namei.c:cached_lookup() and d_revalidate */
8350 +       h_ppath.dentry = h_parent;
8351 +       h_ppath.mnt = au_br_mnt(br);
8352 +       h_d = vfsub_lkup_one(&h_dentry->d_name, &h_ppath);
8353 +       err = PTR_ERR(h_d);
8354 +       if (IS_ERR(h_d))
8355 +               goto out;
8356 +
8357 +       err = 0;
8358 +       if (unlikely(h_d != h_dentry
8359 +                    || d_inode(h_d) != h_inode
8360 +                    || (h_inode && au_iattr_test(&ia, h_inode))))
8361 +               err = au_busy_or_stale();
8362 +       dput(h_d);
8363 +
8364 +out:
8365 +       AuTraceErr(err);
8366 +       return err;
8367 +}
8368 +
8369 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
8370 +               struct dentry *h_parent, struct au_branch *br)
8371 +{
8372 +       int err;
8373 +
8374 +       err = 0;
8375 +       if (udba == AuOpt_UDBA_REVAL
8376 +           && !au_test_fs_remote(h_dentry->d_sb)) {
8377 +               IMustLock(h_dir);
8378 +               err = (d_inode(h_dentry->d_parent) != h_dir);
8379 +       } else if (udba != AuOpt_UDBA_NONE)
8380 +               err = au_h_verify_dentry(h_dentry, h_parent, br);
8381 +
8382 +       return err;
8383 +}
8384 +
8385 +/* ---------------------------------------------------------------------- */
8386 +
8387 +static int au_do_refresh_hdentry(struct dentry *dentry, struct dentry *parent)
8388 +{
8389 +       int err;
8390 +       aufs_bindex_t new_bindex, bindex, bbot, bwh, bdiropq;
8391 +       struct au_hdentry tmp, *p, *q;
8392 +       struct au_dinfo *dinfo;
8393 +       struct super_block *sb;
8394 +
8395 +       DiMustWriteLock(dentry);
8396 +
8397 +       sb = dentry->d_sb;
8398 +       dinfo = au_di(dentry);
8399 +       bbot = dinfo->di_bbot;
8400 +       bwh = dinfo->di_bwh;
8401 +       bdiropq = dinfo->di_bdiropq;
8402 +       bindex = dinfo->di_btop;
8403 +       p = au_hdentry(dinfo, bindex);
8404 +       for (; bindex <= bbot; bindex++, p++) {
8405 +               if (!p->hd_dentry)
8406 +                       continue;
8407 +
8408 +               new_bindex = au_br_index(sb, p->hd_id);
8409 +               if (new_bindex == bindex)
8410 +                       continue;
8411 +
8412 +               if (dinfo->di_bwh == bindex)
8413 +                       bwh = new_bindex;
8414 +               if (dinfo->di_bdiropq == bindex)
8415 +                       bdiropq = new_bindex;
8416 +               if (new_bindex < 0) {
8417 +                       au_hdput(p);
8418 +                       p->hd_dentry = NULL;
8419 +                       continue;
8420 +               }
8421 +
8422 +               /* swap two lower dentries, and loop again */
8423 +               q = au_hdentry(dinfo, new_bindex);
8424 +               tmp = *q;
8425 +               *q = *p;
8426 +               *p = tmp;
8427 +               if (tmp.hd_dentry) {
8428 +                       bindex--;
8429 +                       p--;
8430 +               }
8431 +       }
8432 +
8433 +       dinfo->di_bwh = -1;
8434 +       if (bwh >= 0 && bwh <= au_sbbot(sb) && au_sbr_whable(sb, bwh))
8435 +               dinfo->di_bwh = bwh;
8436 +
8437 +       dinfo->di_bdiropq = -1;
8438 +       if (bdiropq >= 0
8439 +           && bdiropq <= au_sbbot(sb)
8440 +           && au_sbr_whable(sb, bdiropq))
8441 +               dinfo->di_bdiropq = bdiropq;
8442 +
8443 +       err = -EIO;
8444 +       dinfo->di_btop = -1;
8445 +       dinfo->di_bbot = -1;
8446 +       bbot = au_dbbot(parent);
8447 +       bindex = 0;
8448 +       p = au_hdentry(dinfo, bindex);
8449 +       for (; bindex <= bbot; bindex++, p++)
8450 +               if (p->hd_dentry) {
8451 +                       dinfo->di_btop = bindex;
8452 +                       break;
8453 +               }
8454 +
8455 +       if (dinfo->di_btop >= 0) {
8456 +               bindex = bbot;
8457 +               p = au_hdentry(dinfo, bindex);
8458 +               for (; bindex >= 0; bindex--, p--)
8459 +                       if (p->hd_dentry) {
8460 +                               dinfo->di_bbot = bindex;
8461 +                               err = 0;
8462 +                               break;
8463 +                       }
8464 +       }
8465 +
8466 +       return err;
8467 +}
8468 +
8469 +static void au_do_hide(struct dentry *dentry)
8470 +{
8471 +       struct inode *inode;
8472 +
8473 +       if (d_really_is_positive(dentry)) {
8474 +               inode = d_inode(dentry);
8475 +               if (!d_is_dir(dentry)) {
8476 +                       if (inode->i_nlink && !d_unhashed(dentry))
8477 +                               drop_nlink(inode);
8478 +               } else {
8479 +                       clear_nlink(inode);
8480 +                       /* stop next lookup */
8481 +                       inode->i_flags |= S_DEAD;
8482 +               }
8483 +               smp_mb(); /* necessary? */
8484 +       }
8485 +       d_drop(dentry);
8486 +}
8487 +
8488 +static int au_hide_children(struct dentry *parent)
8489 +{
8490 +       int err, i, j, ndentry;
8491 +       struct au_dcsub_pages dpages;
8492 +       struct au_dpage *dpage;
8493 +       struct dentry *dentry;
8494 +
8495 +       err = au_dpages_init(&dpages, GFP_NOFS);
8496 +       if (unlikely(err))
8497 +               goto out;
8498 +       err = au_dcsub_pages(&dpages, parent, NULL, NULL);
8499 +       if (unlikely(err))
8500 +               goto out_dpages;
8501 +
8502 +       /* in reverse order */
8503 +       for (i = dpages.ndpage - 1; i >= 0; i--) {
8504 +               dpage = dpages.dpages + i;
8505 +               ndentry = dpage->ndentry;
8506 +               for (j = ndentry - 1; j >= 0; j--) {
8507 +                       dentry = dpage->dentries[j];
8508 +                       if (dentry != parent)
8509 +                               au_do_hide(dentry);
8510 +               }
8511 +       }
8512 +
8513 +out_dpages:
8514 +       au_dpages_free(&dpages);
8515 +out:
8516 +       return err;
8517 +}
8518 +
8519 +static void au_hide(struct dentry *dentry)
8520 +{
8521 +       int err;
8522 +
8523 +       AuDbgDentry(dentry);
8524 +       if (d_is_dir(dentry)) {
8525 +               /* shrink_dcache_parent(dentry); */
8526 +               err = au_hide_children(dentry);
8527 +               if (unlikely(err))
8528 +                       AuIOErr("%pd, failed hiding children, ignored %d\n",
8529 +                               dentry, err);
8530 +       }
8531 +       au_do_hide(dentry);
8532 +}
8533 +
8534 +/*
8535 + * By adding a dirty branch, a cached dentry may be affected in various ways.
8536 + *
8537 + * a dirty branch is added
8538 + * - on the top of layers
8539 + * - in the middle of layers
8540 + * - to the bottom of layers
8541 + *
8542 + * on the added branch there exists
8543 + * - a whiteout
8544 + * - a diropq
8545 + * - a same named entry
8546 + *   + exist
8547 + *     * negative --> positive
8548 + *     * positive --> positive
8549 + *      - type is unchanged
8550 + *      - type is changed
8551 + *   + doesn't exist
8552 + *     * negative --> negative
8553 + *     * positive --> negative (rejected by au_br_del() for non-dir case)
8554 + * - none
8555 + */
8556 +static int au_refresh_by_dinfo(struct dentry *dentry, struct au_dinfo *dinfo,
8557 +                              struct au_dinfo *tmp)
8558 +{
8559 +       int err;
8560 +       aufs_bindex_t bindex, bbot;
8561 +       struct {
8562 +               struct dentry *dentry;
8563 +               struct inode *inode;
8564 +               mode_t mode;
8565 +       } orig_h, tmp_h = {
8566 +               .dentry = NULL
8567 +       };
8568 +       struct au_hdentry *hd;
8569 +       struct inode *inode, *h_inode;
8570 +       struct dentry *h_dentry;
8571 +
8572 +       err = 0;
8573 +       AuDebugOn(dinfo->di_btop < 0);
8574 +       orig_h.mode = 0;
8575 +       orig_h.dentry = au_hdentry(dinfo, dinfo->di_btop)->hd_dentry;
8576 +       orig_h.inode = NULL;
8577 +       if (d_is_positive(orig_h.dentry)) {
8578 +               orig_h.inode = d_inode(orig_h.dentry);
8579 +               orig_h.mode = orig_h.inode->i_mode & S_IFMT;
8580 +       }
8581 +       if (tmp->di_btop >= 0) {
8582 +               tmp_h.dentry = au_hdentry(tmp, tmp->di_btop)->hd_dentry;
8583 +               if (d_is_positive(tmp_h.dentry)) {
8584 +                       tmp_h.inode = d_inode(tmp_h.dentry);
8585 +                       tmp_h.mode = tmp_h.inode->i_mode & S_IFMT;
8586 +               }
8587 +       }
8588 +
8589 +       inode = NULL;
8590 +       if (d_really_is_positive(dentry))
8591 +               inode = d_inode(dentry);
8592 +       if (!orig_h.inode) {
8593 +               AuDbg("negative originally\n");
8594 +               if (inode) {
8595 +                       au_hide(dentry);
8596 +                       goto out;
8597 +               }
8598 +               AuDebugOn(inode);
8599 +               AuDebugOn(dinfo->di_btop != dinfo->di_bbot);
8600 +               AuDebugOn(dinfo->di_bdiropq != -1);
8601 +
8602 +               if (!tmp_h.inode) {
8603 +                       AuDbg("negative --> negative\n");
8604 +                       /* should have only one negative lower */
8605 +                       if (tmp->di_btop >= 0
8606 +                           && tmp->di_btop < dinfo->di_btop) {
8607 +                               AuDebugOn(tmp->di_btop != tmp->di_bbot);
8608 +                               AuDebugOn(dinfo->di_btop != dinfo->di_bbot);
8609 +                               au_set_h_dptr(dentry, dinfo->di_btop, NULL);
8610 +                               au_di_cp(dinfo, tmp);
8611 +                               hd = au_hdentry(tmp, tmp->di_btop);
8612 +                               au_set_h_dptr(dentry, tmp->di_btop,
8613 +                                             dget(hd->hd_dentry));
8614 +                       }
8615 +                       au_dbg_verify_dinode(dentry);
8616 +               } else {
8617 +                       AuDbg("negative --> positive\n");
8618 +                       /*
8619 +                        * similar to the behaviour of creating with bypassing
8620 +                        * aufs.
8621 +                        * unhash it in order to force an error in the
8622 +                        * succeeding create operation.
8623 +                        * we should not set S_DEAD here.
8624 +                        */
8625 +                       d_drop(dentry);
8626 +                       /* au_di_swap(tmp, dinfo); */
8627 +                       au_dbg_verify_dinode(dentry);
8628 +               }
8629 +       } else {
8630 +               AuDbg("positive originally\n");
8631 +               /* inode may be NULL */
8632 +               AuDebugOn(inode && (inode->i_mode & S_IFMT) != orig_h.mode);
8633 +               if (!tmp_h.inode) {
8634 +                       AuDbg("positive --> negative\n");
8635 +                       /* or bypassing aufs */
8636 +                       au_hide(dentry);
8637 +                       if (tmp->di_bwh >= 0 && tmp->di_bwh <= dinfo->di_btop)
8638 +                               dinfo->di_bwh = tmp->di_bwh;
8639 +                       if (inode)
8640 +                               err = au_refresh_hinode_self(inode);
8641 +                       au_dbg_verify_dinode(dentry);
8642 +               } else if (orig_h.mode == tmp_h.mode) {
8643 +                       AuDbg("positive --> positive, same type\n");
8644 +                       if (!S_ISDIR(orig_h.mode)
8645 +                           && dinfo->di_btop > tmp->di_btop) {
8646 +                               /*
8647 +                                * similar to the behaviour of removing and
8648 +                                * creating.
8649 +                                */
8650 +                               au_hide(dentry);
8651 +                               if (inode)
8652 +                                       err = au_refresh_hinode_self(inode);
8653 +                               au_dbg_verify_dinode(dentry);
8654 +                       } else {
8655 +                               /* fill empty slots */
8656 +                               if (dinfo->di_btop > tmp->di_btop)
8657 +                                       dinfo->di_btop = tmp->di_btop;
8658 +                               if (dinfo->di_bbot < tmp->di_bbot)
8659 +                                       dinfo->di_bbot = tmp->di_bbot;
8660 +                               dinfo->di_bwh = tmp->di_bwh;
8661 +                               dinfo->di_bdiropq = tmp->di_bdiropq;
8662 +                               bbot = dinfo->di_bbot;
8663 +                               bindex = tmp->di_btop;
8664 +                               hd = au_hdentry(tmp, bindex);
8665 +                               for (; bindex <= bbot; bindex++, hd++) {
8666 +                                       if (au_h_dptr(dentry, bindex))
8667 +                                               continue;
8668 +                                       h_dentry = hd->hd_dentry;
8669 +                                       if (!h_dentry)
8670 +                                               continue;
8671 +                                       AuDebugOn(d_is_negative(h_dentry));
8672 +                                       h_inode = d_inode(h_dentry);
8673 +                                       AuDebugOn(orig_h.mode
8674 +                                                 != (h_inode->i_mode
8675 +                                                     & S_IFMT));
8676 +                                       au_set_h_dptr(dentry, bindex,
8677 +                                                     dget(h_dentry));
8678 +                               }
8679 +                               if (inode)
8680 +                                       err = au_refresh_hinode(inode, dentry);
8681 +                               au_dbg_verify_dinode(dentry);
8682 +                       }
8683 +               } else {
8684 +                       AuDbg("positive --> positive, different type\n");
8685 +                       /* similar to the behaviour of removing and creating */
8686 +                       au_hide(dentry);
8687 +                       if (inode)
8688 +                               err = au_refresh_hinode_self(inode);
8689 +                       au_dbg_verify_dinode(dentry);
8690 +               }
8691 +       }
8692 +
8693 +out:
8694 +       return err;
8695 +}
8696 +
8697 +void au_refresh_dop(struct dentry *dentry, int force_reval)
8698 +{
8699 +       const struct dentry_operations *dop
8700 +               = force_reval ? &aufs_dop : dentry->d_sb->s_d_op;
8701 +       static const unsigned int mask
8702 +               = DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE;
8703 +
8704 +       BUILD_BUG_ON(sizeof(mask) != sizeof(dentry->d_flags));
8705 +
8706 +       if (dentry->d_op == dop)
8707 +               return;
8708 +
8709 +       AuDbg("%pd\n", dentry);
8710 +       spin_lock(&dentry->d_lock);
8711 +       if (dop == &aufs_dop)
8712 +               dentry->d_flags |= mask;
8713 +       else
8714 +               dentry->d_flags &= ~mask;
8715 +       dentry->d_op = dop;
8716 +       spin_unlock(&dentry->d_lock);
8717 +}
8718 +
8719 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent)
8720 +{
8721 +       int err, ebrange, nbr;
8722 +       unsigned int sigen;
8723 +       struct au_dinfo *dinfo, *tmp;
8724 +       struct super_block *sb;
8725 +       struct inode *inode;
8726 +
8727 +       DiMustWriteLock(dentry);
8728 +       AuDebugOn(IS_ROOT(dentry));
8729 +       AuDebugOn(d_really_is_negative(parent));
8730 +
8731 +       sb = dentry->d_sb;
8732 +       sigen = au_sigen(sb);
8733 +       err = au_digen_test(parent, sigen);
8734 +       if (unlikely(err))
8735 +               goto out;
8736 +
8737 +       nbr = au_sbbot(sb) + 1;
8738 +       dinfo = au_di(dentry);
8739 +       err = au_di_realloc(dinfo, nbr, /*may_shrink*/0);
8740 +       if (unlikely(err))
8741 +               goto out;
8742 +       ebrange = au_dbrange_test(dentry);
8743 +       if (!ebrange)
8744 +               ebrange = au_do_refresh_hdentry(dentry, parent);
8745 +
8746 +       if (d_unhashed(dentry) || ebrange /* || dinfo->di_tmpfile */) {
8747 +               AuDebugOn(au_dbtop(dentry) < 0 && au_dbbot(dentry) >= 0);
8748 +               if (d_really_is_positive(dentry)) {
8749 +                       inode = d_inode(dentry);
8750 +                       err = au_refresh_hinode_self(inode);
8751 +               }
8752 +               au_dbg_verify_dinode(dentry);
8753 +               if (!err)
8754 +                       goto out_dgen; /* success */
8755 +               goto out;
8756 +       }
8757 +
8758 +       /* temporary dinfo */
8759 +       AuDbgDentry(dentry);
8760 +       err = -ENOMEM;
8761 +       tmp = au_di_alloc(sb, AuLsc_DI_TMP);
8762 +       if (unlikely(!tmp))
8763 +               goto out;
8764 +       au_di_swap(tmp, dinfo);
8765 +       /* returns the number of positive dentries */
8766 +       /*
8767 +        * if current working dir is removed, it returns an error.
8768 +        * but the dentry is legal.
8769 +        */
8770 +       err = au_lkup_dentry(dentry, /*btop*/0, AuLkup_ALLOW_NEG);
8771 +       AuDbgDentry(dentry);
8772 +       au_di_swap(tmp, dinfo);
8773 +       if (err == -ENOENT)
8774 +               err = 0;
8775 +       if (err >= 0) {
8776 +               /* compare/refresh by dinfo */
8777 +               AuDbgDentry(dentry);
8778 +               err = au_refresh_by_dinfo(dentry, dinfo, tmp);
8779 +               au_dbg_verify_dinode(dentry);
8780 +               AuTraceErr(err);
8781 +       }
8782 +       au_di_realloc(dinfo, nbr, /*may_shrink*/1); /* harmless if err */
8783 +       au_rw_write_unlock(&tmp->di_rwsem);
8784 +       au_di_free(tmp);
8785 +       if (unlikely(err))
8786 +               goto out;
8787 +
8788 +out_dgen:
8789 +       au_update_digen(dentry);
8790 +out:
8791 +       if (unlikely(err && !(dentry->d_flags & DCACHE_NFSFS_RENAMED))) {
8792 +               AuIOErr("failed refreshing %pd, %d\n", dentry, err);
8793 +               AuDbgDentry(dentry);
8794 +       }
8795 +       AuTraceErr(err);
8796 +       return err;
8797 +}
8798 +
8799 +static int au_do_h_d_reval(struct dentry *h_dentry, unsigned int flags,
8800 +                          struct dentry *dentry, aufs_bindex_t bindex)
8801 +{
8802 +       int err, valid;
8803 +
8804 +       err = 0;
8805 +       if (!(h_dentry->d_flags & DCACHE_OP_REVALIDATE))
8806 +               goto out;
8807 +
8808 +       AuDbg("b%d\n", bindex);
8809 +       /*
8810 +        * gave up supporting LOOKUP_CREATE/OPEN for lower fs,
8811 +        * due to whiteout and branch permission.
8812 +        */
8813 +       flags &= ~(/*LOOKUP_PARENT |*/ LOOKUP_OPEN | LOOKUP_CREATE
8814 +                  | LOOKUP_FOLLOW | LOOKUP_EXCL);
8815 +       /* it may return tri-state */
8816 +       valid = h_dentry->d_op->d_revalidate(h_dentry, flags);
8817 +
8818 +       if (unlikely(valid < 0))
8819 +               err = valid;
8820 +       else if (!valid)
8821 +               err = -EINVAL;
8822 +
8823 +out:
8824 +       AuTraceErr(err);
8825 +       return err;
8826 +}
8827 +
8828 +/* todo: remove this */
8829 +static int h_d_revalidate(struct dentry *dentry, struct inode *inode,
8830 +                         unsigned int flags, int do_udba, int dirren)
8831 +{
8832 +       int err;
8833 +       umode_t mode, h_mode;
8834 +       aufs_bindex_t bindex, btail, btop, ibs, ibe;
8835 +       unsigned char plus, unhashed, is_root, h_plus, h_nfs, tmpfile;
8836 +       struct inode *h_inode, *h_cached_inode;
8837 +       struct dentry *h_dentry;
8838 +       struct qstr *name, *h_name;
8839 +
8840 +       err = 0;
8841 +       plus = 0;
8842 +       mode = 0;
8843 +       ibs = -1;
8844 +       ibe = -1;
8845 +       unhashed = !!d_unhashed(dentry);
8846 +       is_root = !!IS_ROOT(dentry);
8847 +       name = &dentry->d_name;
8848 +       tmpfile = au_di(dentry)->di_tmpfile;
8849 +
8850 +       /*
8851 +        * Theoretically, REVAL test should be unnecessary in case of
8852 +        * {FS,I}NOTIFY.
8853 +        * But {fs,i}notify doesn't fire some necessary events,
8854 +        *      IN_ATTRIB for atime/nlink/pageio
8855 +        * Let's do REVAL test too.
8856 +        */
8857 +       if (do_udba && inode) {
8858 +               mode = (inode->i_mode & S_IFMT);
8859 +               plus = (inode->i_nlink > 0);
8860 +               ibs = au_ibtop(inode);
8861 +               ibe = au_ibbot(inode);
8862 +       }
8863 +
8864 +       btop = au_dbtop(dentry);
8865 +       btail = btop;
8866 +       if (inode && S_ISDIR(inode->i_mode))
8867 +               btail = au_dbtaildir(dentry);
8868 +       for (bindex = btop; bindex <= btail; bindex++) {
8869 +               h_dentry = au_h_dptr(dentry, bindex);
8870 +               if (!h_dentry)
8871 +                       continue;
8872 +
8873 +               AuDbg("b%d, %pd\n", bindex, h_dentry);
8874 +               h_nfs = !!au_test_nfs(h_dentry->d_sb);
8875 +               spin_lock(&h_dentry->d_lock);
8876 +               h_name = &h_dentry->d_name;
8877 +               if (unlikely(do_udba
8878 +                            && !is_root
8879 +                            && ((!h_nfs
8880 +                                 && (unhashed != !!d_unhashed(h_dentry)
8881 +                                     || (!tmpfile && !dirren
8882 +                                         && !au_qstreq(name, h_name))
8883 +                                         ))
8884 +                                || (h_nfs
8885 +                                    && !(flags & LOOKUP_OPEN)
8886 +                                    && (h_dentry->d_flags
8887 +                                        & DCACHE_NFSFS_RENAMED)))
8888 +                           )) {
8889 +                       int h_unhashed;
8890 +
8891 +                       h_unhashed = d_unhashed(h_dentry);
8892 +                       spin_unlock(&h_dentry->d_lock);
8893 +                       AuDbg("unhash 0x%x 0x%x, %pd %pd\n",
8894 +                             unhashed, h_unhashed, dentry, h_dentry);
8895 +                       goto err;
8896 +               }
8897 +               spin_unlock(&h_dentry->d_lock);
8898 +
8899 +               err = au_do_h_d_reval(h_dentry, flags, dentry, bindex);
8900 +               if (unlikely(err))
8901 +                       /* do not goto err, to keep the errno */
8902 +                       break;
8903 +
8904 +               /* todo: plink too? */
8905 +               if (!do_udba)
8906 +                       continue;
8907 +
8908 +               /* UDBA tests */
8909 +               if (unlikely(!!inode != d_is_positive(h_dentry)))
8910 +                       goto err;
8911 +
8912 +               h_inode = NULL;
8913 +               if (d_is_positive(h_dentry))
8914 +                       h_inode = d_inode(h_dentry);
8915 +               h_plus = plus;
8916 +               h_mode = mode;
8917 +               h_cached_inode = h_inode;
8918 +               if (h_inode) {
8919 +                       h_mode = (h_inode->i_mode & S_IFMT);
8920 +                       h_plus = (h_inode->i_nlink > 0);
8921 +               }
8922 +               if (inode && ibs <= bindex && bindex <= ibe)
8923 +                       h_cached_inode = au_h_iptr(inode, bindex);
8924 +
8925 +               if (!h_nfs) {
8926 +                       if (unlikely(plus != h_plus && !tmpfile))
8927 +                               goto err;
8928 +               } else {
8929 +                       if (unlikely(!(h_dentry->d_flags & DCACHE_NFSFS_RENAMED)
8930 +                                    && !is_root
8931 +                                    && !IS_ROOT(h_dentry)
8932 +                                    && unhashed != d_unhashed(h_dentry)))
8933 +                               goto err;
8934 +               }
8935 +               if (unlikely(mode != h_mode
8936 +                            || h_cached_inode != h_inode))
8937 +                       goto err;
8938 +               continue;
8939 +
8940 +err:
8941 +               err = -EINVAL;
8942 +               break;
8943 +       }
8944 +
8945 +       AuTraceErr(err);
8946 +       return err;
8947 +}
8948 +
8949 +/* todo: consolidate with do_refresh() and au_reval_for_attr() */
8950 +static int simple_reval_dpath(struct dentry *dentry, unsigned int sigen)
8951 +{
8952 +       int err;
8953 +       struct dentry *parent;
8954 +
8955 +       if (!au_digen_test(dentry, sigen))
8956 +               return 0;
8957 +
8958 +       parent = dget_parent(dentry);
8959 +       di_read_lock_parent(parent, AuLock_IR);
8960 +       AuDebugOn(au_digen_test(parent, sigen));
8961 +       au_dbg_verify_gen(parent, sigen);
8962 +       err = au_refresh_dentry(dentry, parent);
8963 +       di_read_unlock(parent, AuLock_IR);
8964 +       dput(parent);
8965 +       AuTraceErr(err);
8966 +       return err;
8967 +}
8968 +
8969 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen)
8970 +{
8971 +       int err;
8972 +       struct dentry *d, *parent;
8973 +
8974 +       if (!au_ftest_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR))
8975 +               return simple_reval_dpath(dentry, sigen);
8976 +
8977 +       /* slow loop, keep it simple and stupid */
8978 +       /* cf: au_cpup_dirs() */
8979 +       err = 0;
8980 +       parent = NULL;
8981 +       while (au_digen_test(dentry, sigen)) {
8982 +               d = dentry;
8983 +               while (1) {
8984 +                       dput(parent);
8985 +                       parent = dget_parent(d);
8986 +                       if (!au_digen_test(parent, sigen))
8987 +                               break;
8988 +                       d = parent;
8989 +               }
8990 +
8991 +               if (d != dentry)
8992 +                       di_write_lock_child2(d);
8993 +
8994 +               /* someone might update our dentry while we were sleeping */
8995 +               if (au_digen_test(d, sigen)) {
8996 +                       /*
8997 +                        * todo: consolidate with simple_reval_dpath(),
8998 +                        * do_refresh() and au_reval_for_attr().
8999 +                        */
9000 +                       di_read_lock_parent(parent, AuLock_IR);
9001 +                       err = au_refresh_dentry(d, parent);
9002 +                       di_read_unlock(parent, AuLock_IR);
9003 +               }
9004 +
9005 +               if (d != dentry)
9006 +                       di_write_unlock(d);
9007 +               dput(parent);
9008 +               if (unlikely(err))
9009 +                       break;
9010 +       }
9011 +
9012 +       return err;
9013 +}
9014 +
9015 +/*
9016 + * if valid returns 1, otherwise 0.
9017 + */
9018 +static int aufs_d_revalidate(struct dentry *dentry, unsigned int flags)
9019 +{
9020 +       int valid, err;
9021 +       unsigned int sigen;
9022 +       unsigned char do_udba, dirren;
9023 +       struct super_block *sb;
9024 +       struct inode *inode;
9025 +
9026 +       /* todo: support rcu-walk? */
9027 +       if (flags & LOOKUP_RCU)
9028 +               return -ECHILD;
9029 +
9030 +       valid = 0;
9031 +       if (unlikely(!au_di(dentry)))
9032 +               goto out;
9033 +
9034 +       valid = 1;
9035 +       sb = dentry->d_sb;
9036 +       /*
9037 +        * todo: very ugly
9038 +        * i_mutex of parent dir may be held,
9039 +        * but we should not return 'invalid' due to busy.
9040 +        */
9041 +       err = aufs_read_lock(dentry, AuLock_FLUSH | AuLock_DW | AuLock_NOPLM);
9042 +       if (unlikely(err)) {
9043 +               valid = err;
9044 +               AuTraceErr(err);
9045 +               goto out;
9046 +       }
9047 +       inode = NULL;
9048 +       if (d_really_is_positive(dentry))
9049 +               inode = d_inode(dentry);
9050 +       if (unlikely(inode && au_is_bad_inode(inode))) {
9051 +               err = -EINVAL;
9052 +               AuTraceErr(err);
9053 +               goto out_dgrade;
9054 +       }
9055 +       if (unlikely(au_dbrange_test(dentry))) {
9056 +               err = -EINVAL;
9057 +               AuTraceErr(err);
9058 +               goto out_dgrade;
9059 +       }
9060 +
9061 +       sigen = au_sigen(sb);
9062 +       if (au_digen_test(dentry, sigen)) {
9063 +               AuDebugOn(IS_ROOT(dentry));
9064 +               err = au_reval_dpath(dentry, sigen);
9065 +               if (unlikely(err)) {
9066 +                       AuTraceErr(err);
9067 +                       goto out_dgrade;
9068 +               }
9069 +       }
9070 +       di_downgrade_lock(dentry, AuLock_IR);
9071 +
9072 +       err = -EINVAL;
9073 +       if (!(flags & (LOOKUP_OPEN | LOOKUP_EMPTY))
9074 +           && inode
9075 +           && !(inode->i_state && I_LINKABLE)
9076 +           && (IS_DEADDIR(inode) || !inode->i_nlink)) {
9077 +               AuTraceErr(err);
9078 +               goto out_inval;
9079 +       }
9080 +
9081 +       do_udba = !au_opt_test(au_mntflags(sb), UDBA_NONE);
9082 +       if (do_udba && inode) {
9083 +               aufs_bindex_t btop = au_ibtop(inode);
9084 +               struct inode *h_inode;
9085 +
9086 +               if (btop >= 0) {
9087 +                       h_inode = au_h_iptr(inode, btop);
9088 +                       if (h_inode && au_test_higen(inode, h_inode)) {
9089 +                               AuTraceErr(err);
9090 +                               goto out_inval;
9091 +                       }
9092 +               }
9093 +       }
9094 +
9095 +       dirren = !!au_opt_test(au_mntflags(sb), DIRREN);
9096 +       err = h_d_revalidate(dentry, inode, flags, do_udba, dirren);
9097 +       if (unlikely(!err && do_udba && au_dbtop(dentry) < 0)) {
9098 +               err = -EIO;
9099 +               AuDbg("both of real entry and whiteout found, %p, err %d\n",
9100 +                     dentry, err);
9101 +       }
9102 +       goto out_inval;
9103 +
9104 +out_dgrade:
9105 +       di_downgrade_lock(dentry, AuLock_IR);
9106 +out_inval:
9107 +       aufs_read_unlock(dentry, AuLock_IR);
9108 +       AuTraceErr(err);
9109 +       valid = !err;
9110 +out:
9111 +       if (!valid) {
9112 +               AuDbg("%pd invalid, %d\n", dentry, valid);
9113 +               d_drop(dentry);
9114 +       }
9115 +       return valid;
9116 +}
9117 +
9118 +static void aufs_d_release(struct dentry *dentry)
9119 +{
9120 +       if (au_di(dentry)) {
9121 +               au_di_fin(dentry);
9122 +               au_hn_di_reinit(dentry);
9123 +       }
9124 +}
9125 +
9126 +const struct dentry_operations aufs_dop = {
9127 +       .d_revalidate           = aufs_d_revalidate,
9128 +       .d_weak_revalidate      = aufs_d_revalidate,
9129 +       .d_release              = aufs_d_release
9130 +};
9131 +
9132 +/* aufs_dop without d_revalidate */
9133 +const struct dentry_operations aufs_dop_noreval = {
9134 +       .d_release              = aufs_d_release
9135 +};
9136 diff -urN /usr/share/empty/fs/aufs/dentry.h linux/fs/aufs/dentry.h
9137 --- /usr/share/empty/fs/aufs/dentry.h   1970-01-01 01:00:00.000000000 +0100
9138 +++ linux/fs/aufs/dentry.h      2021-12-03 15:40:58.233313963 +0100
9139 @@ -0,0 +1,269 @@
9140 +/* SPDX-License-Identifier: GPL-2.0 */
9141 +/*
9142 + * Copyright (C) 2005-2021 Junjiro R. Okajima
9143 + *
9144 + * This program, aufs is free software; you can redistribute it and/or modify
9145 + * it under the terms of the GNU General Public License as published by
9146 + * the Free Software Foundation; either version 2 of the License, or
9147 + * (at your option) any later version.
9148 + *
9149 + * This program is distributed in the hope that it will be useful,
9150 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9151 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9152 + * GNU General Public License for more details.
9153 + *
9154 + * You should have received a copy of the GNU General Public License
9155 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9156 + */
9157 +
9158 +/*
9159 + * lookup and dentry operations
9160 + */
9161 +
9162 +#ifndef __AUFS_DENTRY_H__
9163 +#define __AUFS_DENTRY_H__
9164 +
9165 +#ifdef __KERNEL__
9166 +
9167 +#include <linux/dcache.h>
9168 +#include "dirren.h"
9169 +#include "rwsem.h"
9170 +
9171 +struct au_hdentry {
9172 +       struct dentry           *hd_dentry;
9173 +       aufs_bindex_t           hd_id;
9174 +};
9175 +
9176 +struct au_dinfo {
9177 +       atomic_t                di_generation;
9178 +
9179 +       struct au_rwsem         di_rwsem;
9180 +       aufs_bindex_t           di_btop, di_bbot, di_bwh, di_bdiropq;
9181 +       unsigned char           di_tmpfile; /* to allow the different name */
9182 +       struct au_hdentry       *di_hdentry;
9183 +       struct rcu_head         rcu;
9184 +} ____cacheline_aligned_in_smp;
9185 +
9186 +/* ---------------------------------------------------------------------- */
9187 +
9188 +/* flags for au_lkup_dentry() */
9189 +#define AuLkup_ALLOW_NEG       1
9190 +#define AuLkup_IGNORE_PERM     (1 << 1)
9191 +#define AuLkup_DIRREN          (1 << 2)
9192 +#define au_ftest_lkup(flags, name)     ((flags) & AuLkup_##name)
9193 +#define au_fset_lkup(flags, name) \
9194 +       do { (flags) |= AuLkup_##name; } while (0)
9195 +#define au_fclr_lkup(flags, name) \
9196 +       do { (flags) &= ~AuLkup_##name; } while (0)
9197 +
9198 +#ifndef CONFIG_AUFS_DIRREN
9199 +#undef AuLkup_DIRREN
9200 +#define AuLkup_DIRREN 0
9201 +#endif
9202 +
9203 +struct au_do_lookup_args {
9204 +       unsigned int            flags;
9205 +       mode_t                  type;
9206 +       struct qstr             whname, *name;
9207 +       struct au_dr_lookup     dirren;
9208 +};
9209 +
9210 +/* ---------------------------------------------------------------------- */
9211 +
9212 +/* dentry.c */
9213 +extern const struct dentry_operations aufs_dop, aufs_dop_noreval;
9214 +struct au_branch;
9215 +struct dentry *au_sio_lkup_one(struct user_namespace *userns, struct qstr *name,
9216 +                              struct path *ppath);
9217 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
9218 +               struct dentry *h_parent, struct au_branch *br);
9219 +
9220 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t btop,
9221 +                  unsigned int flags);
9222 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh);
9223 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent);
9224 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen);
9225 +void au_refresh_dop(struct dentry *dentry, int force_reval);
9226 +
9227 +/* dinfo.c */
9228 +void au_di_init_once(void *_di);
9229 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc);
9230 +void au_di_free(struct au_dinfo *dinfo);
9231 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b);
9232 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src);
9233 +int au_di_init(struct dentry *dentry);
9234 +void au_di_fin(struct dentry *dentry);
9235 +int au_di_realloc(struct au_dinfo *dinfo, int nbr, int may_shrink);
9236 +
9237 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc);
9238 +void di_read_unlock(struct dentry *d, int flags);
9239 +void di_downgrade_lock(struct dentry *d, int flags);
9240 +void di_write_lock(struct dentry *d, unsigned int lsc);
9241 +void di_write_unlock(struct dentry *d);
9242 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir);
9243 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir);
9244 +void di_write_unlock2(struct dentry *d1, struct dentry *d2);
9245 +
9246 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex);
9247 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex);
9248 +aufs_bindex_t au_dbtail(struct dentry *dentry);
9249 +aufs_bindex_t au_dbtaildir(struct dentry *dentry);
9250 +
9251 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
9252 +                  struct dentry *h_dentry);
9253 +int au_digen_test(struct dentry *dentry, unsigned int sigen);
9254 +int au_dbrange_test(struct dentry *dentry);
9255 +void au_update_digen(struct dentry *dentry);
9256 +void au_update_dbrange(struct dentry *dentry, int do_put_zero);
9257 +void au_update_dbtop(struct dentry *dentry);
9258 +void au_update_dbbot(struct dentry *dentry);
9259 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry);
9260 +
9261 +/* ---------------------------------------------------------------------- */
9262 +
9263 +static inline struct au_dinfo *au_di(struct dentry *dentry)
9264 +{
9265 +       return dentry->d_fsdata;
9266 +}
9267 +
9268 +/* ---------------------------------------------------------------------- */
9269 +
9270 +/* lock subclass for dinfo */
9271 +enum {
9272 +       AuLsc_DI_CHILD,         /* child first */
9273 +       AuLsc_DI_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
9274 +       AuLsc_DI_CHILD3,        /* copyup dirs */
9275 +       AuLsc_DI_PARENT,
9276 +       AuLsc_DI_PARENT2,
9277 +       AuLsc_DI_PARENT3,
9278 +       AuLsc_DI_TMP            /* temp for replacing dinfo */
9279 +};
9280 +
9281 +/*
9282 + * di_read_lock_child, di_write_lock_child,
9283 + * di_read_lock_child2, di_write_lock_child2,
9284 + * di_read_lock_child3, di_write_lock_child3,
9285 + * di_read_lock_parent, di_write_lock_parent,
9286 + * di_read_lock_parent2, di_write_lock_parent2,
9287 + * di_read_lock_parent3, di_write_lock_parent3,
9288 + */
9289 +#define AuReadLockFunc(name, lsc) \
9290 +static inline void di_read_lock_##name(struct dentry *d, int flags) \
9291 +{ di_read_lock(d, flags, AuLsc_DI_##lsc); }
9292 +
9293 +#define AuWriteLockFunc(name, lsc) \
9294 +static inline void di_write_lock_##name(struct dentry *d) \
9295 +{ di_write_lock(d, AuLsc_DI_##lsc); }
9296 +
9297 +#define AuRWLockFuncs(name, lsc) \
9298 +       AuReadLockFunc(name, lsc) \
9299 +       AuWriteLockFunc(name, lsc)
9300 +
9301 +AuRWLockFuncs(child, CHILD);
9302 +AuRWLockFuncs(child2, CHILD2);
9303 +AuRWLockFuncs(child3, CHILD3);
9304 +AuRWLockFuncs(parent, PARENT);
9305 +AuRWLockFuncs(parent2, PARENT2);
9306 +AuRWLockFuncs(parent3, PARENT3);
9307 +
9308 +#undef AuReadLockFunc
9309 +#undef AuWriteLockFunc
9310 +#undef AuRWLockFuncs
9311 +
9312 +#define DiMustNoWaiters(d)     AuRwMustNoWaiters(&au_di(d)->di_rwsem)
9313 +#define DiMustAnyLock(d)       AuRwMustAnyLock(&au_di(d)->di_rwsem)
9314 +#define DiMustWriteLock(d)     AuRwMustWriteLock(&au_di(d)->di_rwsem)
9315 +
9316 +/* ---------------------------------------------------------------------- */
9317 +
9318 +/* todo: memory barrier? */
9319 +static inline unsigned int au_digen(struct dentry *d)
9320 +{
9321 +       return atomic_read(&au_di(d)->di_generation);
9322 +}
9323 +
9324 +static inline void au_h_dentry_init(struct au_hdentry *hdentry)
9325 +{
9326 +       hdentry->hd_dentry = NULL;
9327 +}
9328 +
9329 +static inline struct au_hdentry *au_hdentry(struct au_dinfo *di,
9330 +                                           aufs_bindex_t bindex)
9331 +{
9332 +       return di->di_hdentry + bindex;
9333 +}
9334 +
9335 +static inline void au_hdput(struct au_hdentry *hd)
9336 +{
9337 +       if (hd)
9338 +               dput(hd->hd_dentry);
9339 +}
9340 +
9341 +static inline aufs_bindex_t au_dbtop(struct dentry *dentry)
9342 +{
9343 +       DiMustAnyLock(dentry);
9344 +       return au_di(dentry)->di_btop;
9345 +}
9346 +
9347 +static inline aufs_bindex_t au_dbbot(struct dentry *dentry)
9348 +{
9349 +       DiMustAnyLock(dentry);
9350 +       return au_di(dentry)->di_bbot;
9351 +}
9352 +
9353 +static inline aufs_bindex_t au_dbwh(struct dentry *dentry)
9354 +{
9355 +       DiMustAnyLock(dentry);
9356 +       return au_di(dentry)->di_bwh;
9357 +}
9358 +
9359 +static inline aufs_bindex_t au_dbdiropq(struct dentry *dentry)
9360 +{
9361 +       DiMustAnyLock(dentry);
9362 +       return au_di(dentry)->di_bdiropq;
9363 +}
9364 +
9365 +/* todo: hard/soft set? */
9366 +static inline void au_set_dbtop(struct dentry *dentry, aufs_bindex_t bindex)
9367 +{
9368 +       DiMustWriteLock(dentry);
9369 +       au_di(dentry)->di_btop = bindex;
9370 +}
9371 +
9372 +static inline void au_set_dbbot(struct dentry *dentry, aufs_bindex_t bindex)
9373 +{
9374 +       DiMustWriteLock(dentry);
9375 +       au_di(dentry)->di_bbot = bindex;
9376 +}
9377 +
9378 +static inline void au_set_dbwh(struct dentry *dentry, aufs_bindex_t bindex)
9379 +{
9380 +       DiMustWriteLock(dentry);
9381 +       /* dbwh can be outside of btop - bbot range */
9382 +       au_di(dentry)->di_bwh = bindex;
9383 +}
9384 +
9385 +static inline void au_set_dbdiropq(struct dentry *dentry, aufs_bindex_t bindex)
9386 +{
9387 +       DiMustWriteLock(dentry);
9388 +       au_di(dentry)->di_bdiropq = bindex;
9389 +}
9390 +
9391 +/* ---------------------------------------------------------------------- */
9392 +
9393 +#ifdef CONFIG_AUFS_HNOTIFY
9394 +static inline void au_digen_dec(struct dentry *d)
9395 +{
9396 +       atomic_dec(&au_di(d)->di_generation);
9397 +}
9398 +
9399 +static inline void au_hn_di_reinit(struct dentry *dentry)
9400 +{
9401 +       dentry->d_fsdata = NULL;
9402 +}
9403 +#else
9404 +AuStubVoid(au_hn_di_reinit, struct dentry *dentry __maybe_unused)
9405 +#endif /* CONFIG_AUFS_HNOTIFY */
9406 +
9407 +#endif /* __KERNEL__ */
9408 +#endif /* __AUFS_DENTRY_H__ */
9409 diff -urN /usr/share/empty/fs/aufs/dinfo.c linux/fs/aufs/dinfo.c
9410 --- /usr/share/empty/fs/aufs/dinfo.c    1970-01-01 01:00:00.000000000 +0100
9411 +++ linux/fs/aufs/dinfo.c       2021-12-03 15:38:59.933313976 +0100
9412 @@ -0,0 +1,554 @@
9413 +// SPDX-License-Identifier: GPL-2.0
9414 +/*
9415 + * Copyright (C) 2005-2021 Junjiro R. Okajima
9416 + *
9417 + * This program, aufs is free software; you can redistribute it and/or modify
9418 + * it under the terms of the GNU General Public License as published by
9419 + * the Free Software Foundation; either version 2 of the License, or
9420 + * (at your option) any later version.
9421 + *
9422 + * This program is distributed in the hope that it will be useful,
9423 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9424 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9425 + * GNU General Public License for more details.
9426 + *
9427 + * You should have received a copy of the GNU General Public License
9428 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9429 + */
9430 +
9431 +/*
9432 + * dentry private data
9433 + */
9434 +
9435 +#include "aufs.h"
9436 +
9437 +void au_di_init_once(void *_dinfo)
9438 +{
9439 +       struct au_dinfo *dinfo = _dinfo;
9440 +
9441 +       au_rw_init(&dinfo->di_rwsem);
9442 +}
9443 +
9444 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc)
9445 +{
9446 +       struct au_dinfo *dinfo;
9447 +       int nbr, i;
9448 +
9449 +       dinfo = au_cache_alloc_dinfo();
9450 +       if (unlikely(!dinfo))
9451 +               goto out;
9452 +
9453 +       nbr = au_sbbot(sb) + 1;
9454 +       if (nbr <= 0)
9455 +               nbr = 1;
9456 +       dinfo->di_hdentry = kcalloc(nbr, sizeof(*dinfo->di_hdentry), GFP_NOFS);
9457 +       if (dinfo->di_hdentry) {
9458 +               au_rw_write_lock_nested(&dinfo->di_rwsem, lsc);
9459 +               dinfo->di_btop = -1;
9460 +               dinfo->di_bbot = -1;
9461 +               dinfo->di_bwh = -1;
9462 +               dinfo->di_bdiropq = -1;
9463 +               dinfo->di_tmpfile = 0;
9464 +               for (i = 0; i < nbr; i++)
9465 +                       dinfo->di_hdentry[i].hd_id = -1;
9466 +               goto out;
9467 +       }
9468 +
9469 +       au_cache_free_dinfo(dinfo);
9470 +       dinfo = NULL;
9471 +
9472 +out:
9473 +       return dinfo;
9474 +}
9475 +
9476 +void au_di_free(struct au_dinfo *dinfo)
9477 +{
9478 +       struct au_hdentry *p;
9479 +       aufs_bindex_t bbot, bindex;
9480 +
9481 +       /* dentry may not be revalidated */
9482 +       bindex = dinfo->di_btop;
9483 +       if (bindex >= 0) {
9484 +               bbot = dinfo->di_bbot;
9485 +               p = au_hdentry(dinfo, bindex);
9486 +               while (bindex++ <= bbot)
9487 +                       au_hdput(p++);
9488 +       }
9489 +       au_kfree_try_rcu(dinfo->di_hdentry);
9490 +       au_cache_free_dinfo(dinfo);
9491 +}
9492 +
9493 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b)
9494 +{
9495 +       struct au_hdentry *p;
9496 +       aufs_bindex_t bi;
9497 +
9498 +       AuRwMustWriteLock(&a->di_rwsem);
9499 +       AuRwMustWriteLock(&b->di_rwsem);
9500 +
9501 +#define DiSwap(v, name)                                \
9502 +       do {                                    \
9503 +               v = a->di_##name;               \
9504 +               a->di_##name = b->di_##name;    \
9505 +               b->di_##name = v;               \
9506 +       } while (0)
9507 +
9508 +       DiSwap(p, hdentry);
9509 +       DiSwap(bi, btop);
9510 +       DiSwap(bi, bbot);
9511 +       DiSwap(bi, bwh);
9512 +       DiSwap(bi, bdiropq);
9513 +       /* smp_mb(); */
9514 +
9515 +#undef DiSwap
9516 +}
9517 +
9518 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src)
9519 +{
9520 +       AuRwMustWriteLock(&dst->di_rwsem);
9521 +       AuRwMustWriteLock(&src->di_rwsem);
9522 +
9523 +       dst->di_btop = src->di_btop;
9524 +       dst->di_bbot = src->di_bbot;
9525 +       dst->di_bwh = src->di_bwh;
9526 +       dst->di_bdiropq = src->di_bdiropq;
9527 +       /* smp_mb(); */
9528 +}
9529 +
9530 +int au_di_init(struct dentry *dentry)
9531 +{
9532 +       int err;
9533 +       struct super_block *sb;
9534 +       struct au_dinfo *dinfo;
9535 +
9536 +       err = 0;
9537 +       sb = dentry->d_sb;
9538 +       dinfo = au_di_alloc(sb, AuLsc_DI_CHILD);
9539 +       if (dinfo) {
9540 +               atomic_set(&dinfo->di_generation, au_sigen(sb));
9541 +               /* smp_mb(); */ /* atomic_set */
9542 +               dentry->d_fsdata = dinfo;
9543 +       } else
9544 +               err = -ENOMEM;
9545 +
9546 +       return err;
9547 +}
9548 +
9549 +void au_di_fin(struct dentry *dentry)
9550 +{
9551 +       struct au_dinfo *dinfo;
9552 +
9553 +       dinfo = au_di(dentry);
9554 +       AuRwDestroy(&dinfo->di_rwsem);
9555 +       au_di_free(dinfo);
9556 +}
9557 +
9558 +int au_di_realloc(struct au_dinfo *dinfo, int nbr, int may_shrink)
9559 +{
9560 +       int err, sz;
9561 +       struct au_hdentry *hdp;
9562 +
9563 +       AuRwMustWriteLock(&dinfo->di_rwsem);
9564 +
9565 +       err = -ENOMEM;
9566 +       sz = sizeof(*hdp) * (dinfo->di_bbot + 1);
9567 +       if (!sz)
9568 +               sz = sizeof(*hdp);
9569 +       hdp = au_kzrealloc(dinfo->di_hdentry, sz, sizeof(*hdp) * nbr, GFP_NOFS,
9570 +                          may_shrink);
9571 +       if (hdp) {
9572 +               dinfo->di_hdentry = hdp;
9573 +               err = 0;
9574 +       }
9575 +
9576 +       return err;
9577 +}
9578 +
9579 +/* ---------------------------------------------------------------------- */
9580 +
9581 +static void do_ii_write_lock(struct inode *inode, unsigned int lsc)
9582 +{
9583 +       switch (lsc) {
9584 +       case AuLsc_DI_CHILD:
9585 +               ii_write_lock_child(inode);
9586 +               break;
9587 +       case AuLsc_DI_CHILD2:
9588 +               ii_write_lock_child2(inode);
9589 +               break;
9590 +       case AuLsc_DI_CHILD3:
9591 +               ii_write_lock_child3(inode);
9592 +               break;
9593 +       case AuLsc_DI_PARENT:
9594 +               ii_write_lock_parent(inode);
9595 +               break;
9596 +       case AuLsc_DI_PARENT2:
9597 +               ii_write_lock_parent2(inode);
9598 +               break;
9599 +       case AuLsc_DI_PARENT3:
9600 +               ii_write_lock_parent3(inode);
9601 +               break;
9602 +       default:
9603 +               BUG();
9604 +       }
9605 +}
9606 +
9607 +static void do_ii_read_lock(struct inode *inode, unsigned int lsc)
9608 +{
9609 +       switch (lsc) {
9610 +       case AuLsc_DI_CHILD:
9611 +               ii_read_lock_child(inode);
9612 +               break;
9613 +       case AuLsc_DI_CHILD2:
9614 +               ii_read_lock_child2(inode);
9615 +               break;
9616 +       case AuLsc_DI_CHILD3:
9617 +               ii_read_lock_child3(inode);
9618 +               break;
9619 +       case AuLsc_DI_PARENT:
9620 +               ii_read_lock_parent(inode);
9621 +               break;
9622 +       case AuLsc_DI_PARENT2:
9623 +               ii_read_lock_parent2(inode);
9624 +               break;
9625 +       case AuLsc_DI_PARENT3:
9626 +               ii_read_lock_parent3(inode);
9627 +               break;
9628 +       default:
9629 +               BUG();
9630 +       }
9631 +}
9632 +
9633 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc)
9634 +{
9635 +       struct inode *inode;
9636 +
9637 +       au_rw_read_lock_nested(&au_di(d)->di_rwsem, lsc);
9638 +       if (d_really_is_positive(d)) {
9639 +               inode = d_inode(d);
9640 +               if (au_ftest_lock(flags, IW))
9641 +                       do_ii_write_lock(inode, lsc);
9642 +               else if (au_ftest_lock(flags, IR))
9643 +                       do_ii_read_lock(inode, lsc);
9644 +       }
9645 +}
9646 +
9647 +void di_read_unlock(struct dentry *d, int flags)
9648 +{
9649 +       struct inode *inode;
9650 +
9651 +       if (d_really_is_positive(d)) {
9652 +               inode = d_inode(d);
9653 +               if (au_ftest_lock(flags, IW)) {
9654 +                       au_dbg_verify_dinode(d);
9655 +                       ii_write_unlock(inode);
9656 +               } else if (au_ftest_lock(flags, IR)) {
9657 +                       au_dbg_verify_dinode(d);
9658 +                       ii_read_unlock(inode);
9659 +               }
9660 +       }
9661 +       au_rw_read_unlock(&au_di(d)->di_rwsem);
9662 +}
9663 +
9664 +void di_downgrade_lock(struct dentry *d, int flags)
9665 +{
9666 +       if (d_really_is_positive(d) && au_ftest_lock(flags, IR))
9667 +               ii_downgrade_lock(d_inode(d));
9668 +       au_rw_dgrade_lock(&au_di(d)->di_rwsem);
9669 +}
9670 +
9671 +void di_write_lock(struct dentry *d, unsigned int lsc)
9672 +{
9673 +       au_rw_write_lock_nested(&au_di(d)->di_rwsem, lsc);
9674 +       if (d_really_is_positive(d))
9675 +               do_ii_write_lock(d_inode(d), lsc);
9676 +}
9677 +
9678 +void di_write_unlock(struct dentry *d)
9679 +{
9680 +       au_dbg_verify_dinode(d);
9681 +       if (d_really_is_positive(d))
9682 +               ii_write_unlock(d_inode(d));
9683 +       au_rw_write_unlock(&au_di(d)->di_rwsem);
9684 +}
9685 +
9686 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir)
9687 +{
9688 +       AuDebugOn(d1 == d2
9689 +                 || d_inode(d1) == d_inode(d2)
9690 +                 || d1->d_sb != d2->d_sb);
9691 +
9692 +       if ((isdir && au_test_subdir(d1, d2))
9693 +           || d1 < d2) {
9694 +               di_write_lock_child(d1);
9695 +               di_write_lock_child2(d2);
9696 +       } else {
9697 +               di_write_lock_child(d2);
9698 +               di_write_lock_child2(d1);
9699 +       }
9700 +}
9701 +
9702 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir)
9703 +{
9704 +       AuDebugOn(d1 == d2
9705 +                 || d_inode(d1) == d_inode(d2)
9706 +                 || d1->d_sb != d2->d_sb);
9707 +
9708 +       if ((isdir && au_test_subdir(d1, d2))
9709 +           || d1 < d2) {
9710 +               di_write_lock_parent(d1);
9711 +               di_write_lock_parent2(d2);
9712 +       } else {
9713 +               di_write_lock_parent(d2);
9714 +               di_write_lock_parent2(d1);
9715 +       }
9716 +}
9717 +
9718 +void di_write_unlock2(struct dentry *d1, struct dentry *d2)
9719 +{
9720 +       di_write_unlock(d1);
9721 +       if (d_inode(d1) == d_inode(d2))
9722 +               au_rw_write_unlock(&au_di(d2)->di_rwsem);
9723 +       else
9724 +               di_write_unlock(d2);
9725 +}
9726 +
9727 +/* ---------------------------------------------------------------------- */
9728 +
9729 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex)
9730 +{
9731 +       struct dentry *d;
9732 +
9733 +       DiMustAnyLock(dentry);
9734 +
9735 +       if (au_dbtop(dentry) < 0 || bindex < au_dbtop(dentry))
9736 +               return NULL;
9737 +       AuDebugOn(bindex < 0);
9738 +       d = au_hdentry(au_di(dentry), bindex)->hd_dentry;
9739 +       AuDebugOn(d && au_dcount(d) <= 0);
9740 +       return d;
9741 +}
9742 +
9743 +/*
9744 + * extended version of au_h_dptr().
9745 + * returns a hashed and positive (or linkable) h_dentry in bindex, NULL, or
9746 + * error.
9747 + */
9748 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex)
9749 +{
9750 +       struct dentry *h_dentry;
9751 +       struct inode *inode, *h_inode;
9752 +
9753 +       AuDebugOn(d_really_is_negative(dentry));
9754 +
9755 +       h_dentry = NULL;
9756 +       if (au_dbtop(dentry) <= bindex
9757 +           && bindex <= au_dbbot(dentry))
9758 +               h_dentry = au_h_dptr(dentry, bindex);
9759 +       if (h_dentry && !au_d_linkable(h_dentry)) {
9760 +               dget(h_dentry);
9761 +               goto out; /* success */
9762 +       }
9763 +
9764 +       inode = d_inode(dentry);
9765 +       AuDebugOn(bindex < au_ibtop(inode));
9766 +       AuDebugOn(au_ibbot(inode) < bindex);
9767 +       h_inode = au_h_iptr(inode, bindex);
9768 +       h_dentry = d_find_alias(h_inode);
9769 +       if (h_dentry) {
9770 +               if (!IS_ERR(h_dentry)) {
9771 +                       if (!au_d_linkable(h_dentry))
9772 +                               goto out; /* success */
9773 +                       dput(h_dentry);
9774 +               } else
9775 +                       goto out;
9776 +       }
9777 +
9778 +       if (au_opt_test(au_mntflags(dentry->d_sb), PLINK)) {
9779 +               h_dentry = au_plink_lkup(inode, bindex);
9780 +               AuDebugOn(!h_dentry);
9781 +               if (!IS_ERR(h_dentry)) {
9782 +                       if (!au_d_hashed_positive(h_dentry))
9783 +                               goto out; /* success */
9784 +                       dput(h_dentry);
9785 +                       h_dentry = NULL;
9786 +               }
9787 +       }
9788 +
9789 +out:
9790 +       AuDbgDentry(h_dentry);
9791 +       return h_dentry;
9792 +}
9793 +
9794 +aufs_bindex_t au_dbtail(struct dentry *dentry)
9795 +{
9796 +       aufs_bindex_t bbot, bwh;
9797 +
9798 +       bbot = au_dbbot(dentry);
9799 +       if (0 <= bbot) {
9800 +               bwh = au_dbwh(dentry);
9801 +               if (!bwh)
9802 +                       return bwh;
9803 +               if (0 < bwh && bwh < bbot)
9804 +                       return bwh - 1;
9805 +       }
9806 +       return bbot;
9807 +}
9808 +
9809 +aufs_bindex_t au_dbtaildir(struct dentry *dentry)
9810 +{
9811 +       aufs_bindex_t bbot, bopq;
9812 +
9813 +       bbot = au_dbtail(dentry);
9814 +       if (0 <= bbot) {
9815 +               bopq = au_dbdiropq(dentry);
9816 +               if (0 <= bopq && bopq < bbot)
9817 +                       bbot = bopq;
9818 +       }
9819 +       return bbot;
9820 +}
9821 +
9822 +/* ---------------------------------------------------------------------- */
9823 +
9824 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
9825 +                  struct dentry *h_dentry)
9826 +{
9827 +       struct au_dinfo *dinfo;
9828 +       struct au_hdentry *hd;
9829 +       struct au_branch *br;
9830 +
9831 +       DiMustWriteLock(dentry);
9832 +
9833 +       dinfo = au_di(dentry);
9834 +       hd = au_hdentry(dinfo, bindex);
9835 +       au_hdput(hd);
9836 +       hd->hd_dentry = h_dentry;
9837 +       if (h_dentry) {
9838 +               br = au_sbr(dentry->d_sb, bindex);
9839 +               hd->hd_id = br->br_id;
9840 +       }
9841 +}
9842 +
9843 +int au_dbrange_test(struct dentry *dentry)
9844 +{
9845 +       int err;
9846 +       aufs_bindex_t btop, bbot;
9847 +
9848 +       err = 0;
9849 +       btop = au_dbtop(dentry);
9850 +       bbot = au_dbbot(dentry);
9851 +       if (btop >= 0)
9852 +               AuDebugOn(bbot < 0 && btop > bbot);
9853 +       else {
9854 +               err = -EIO;
9855 +               AuDebugOn(bbot >= 0);
9856 +       }
9857 +
9858 +       return err;
9859 +}
9860 +
9861 +int au_digen_test(struct dentry *dentry, unsigned int sigen)
9862 +{
9863 +       int err;
9864 +
9865 +       err = 0;
9866 +       if (unlikely(au_digen(dentry) != sigen
9867 +                    || au_iigen_test(d_inode(dentry), sigen)))
9868 +               err = -EIO;
9869 +
9870 +       return err;
9871 +}
9872 +
9873 +void au_update_digen(struct dentry *dentry)
9874 +{
9875 +       atomic_set(&au_di(dentry)->di_generation, au_sigen(dentry->d_sb));
9876 +       /* smp_mb(); */ /* atomic_set */
9877 +}
9878 +
9879 +void au_update_dbrange(struct dentry *dentry, int do_put_zero)
9880 +{
9881 +       struct au_dinfo *dinfo;
9882 +       struct dentry *h_d;
9883 +       struct au_hdentry *hdp;
9884 +       aufs_bindex_t bindex, bbot;
9885 +
9886 +       DiMustWriteLock(dentry);
9887 +
9888 +       dinfo = au_di(dentry);
9889 +       if (!dinfo || dinfo->di_btop < 0)
9890 +               return;
9891 +
9892 +       if (do_put_zero) {
9893 +               bbot = dinfo->di_bbot;
9894 +               bindex = dinfo->di_btop;
9895 +               hdp = au_hdentry(dinfo, bindex);
9896 +               for (; bindex <= bbot; bindex++, hdp++) {
9897 +                       h_d = hdp->hd_dentry;
9898 +                       if (h_d && d_is_negative(h_d))
9899 +                               au_set_h_dptr(dentry, bindex, NULL);
9900 +               }
9901 +       }
9902 +
9903 +       dinfo->di_btop = 0;
9904 +       hdp = au_hdentry(dinfo, dinfo->di_btop);
9905 +       for (; dinfo->di_btop <= dinfo->di_bbot; dinfo->di_btop++, hdp++)
9906 +               if (hdp->hd_dentry)
9907 +                       break;
9908 +       if (dinfo->di_btop > dinfo->di_bbot) {
9909 +               dinfo->di_btop = -1;
9910 +               dinfo->di_bbot = -1;
9911 +               return;
9912 +       }
9913 +
9914 +       hdp = au_hdentry(dinfo, dinfo->di_bbot);
9915 +       for (; dinfo->di_bbot >= 0; dinfo->di_bbot--, hdp--)
9916 +               if (hdp->hd_dentry)
9917 +                       break;
9918 +       AuDebugOn(dinfo->di_btop > dinfo->di_bbot || dinfo->di_bbot < 0);
9919 +}
9920 +
9921 +void au_update_dbtop(struct dentry *dentry)
9922 +{
9923 +       aufs_bindex_t bindex, bbot;
9924 +       struct dentry *h_dentry;
9925 +
9926 +       bbot = au_dbbot(dentry);
9927 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++) {
9928 +               h_dentry = au_h_dptr(dentry, bindex);
9929 +               if (!h_dentry)
9930 +                       continue;
9931 +               if (d_is_positive(h_dentry)) {
9932 +                       au_set_dbtop(dentry, bindex);
9933 +                       return;
9934 +               }
9935 +               au_set_h_dptr(dentry, bindex, NULL);
9936 +       }
9937 +}
9938 +
9939 +void au_update_dbbot(struct dentry *dentry)
9940 +{
9941 +       aufs_bindex_t bindex, btop;
9942 +       struct dentry *h_dentry;
9943 +
9944 +       btop = au_dbtop(dentry);
9945 +       for (bindex = au_dbbot(dentry); bindex >= btop; bindex--) {
9946 +               h_dentry = au_h_dptr(dentry, bindex);
9947 +               if (!h_dentry)
9948 +                       continue;
9949 +               if (d_is_positive(h_dentry)) {
9950 +                       au_set_dbbot(dentry, bindex);
9951 +                       return;
9952 +               }
9953 +               au_set_h_dptr(dentry, bindex, NULL);
9954 +       }
9955 +}
9956 +
9957 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry)
9958 +{
9959 +       aufs_bindex_t bindex, bbot;
9960 +
9961 +       bbot = au_dbbot(dentry);
9962 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++)
9963 +               if (au_h_dptr(dentry, bindex) == h_dentry)
9964 +                       return bindex;
9965 +       return -1;
9966 +}
9967 diff -urN /usr/share/empty/fs/aufs/dir.c linux/fs/aufs/dir.c
9968 --- /usr/share/empty/fs/aufs/dir.c      1970-01-01 01:00:00.000000000 +0100
9969 +++ linux/fs/aufs/dir.c 2021-12-03 15:40:58.233313963 +0100
9970 @@ -0,0 +1,765 @@
9971 +// SPDX-License-Identifier: GPL-2.0
9972 +/*
9973 + * Copyright (C) 2005-2021 Junjiro R. Okajima
9974 + *
9975 + * This program, aufs is free software; you can redistribute it and/or modify
9976 + * it under the terms of the GNU General Public License as published by
9977 + * the Free Software Foundation; either version 2 of the License, or
9978 + * (at your option) any later version.
9979 + *
9980 + * This program is distributed in the hope that it will be useful,
9981 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9982 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9983 + * GNU General Public License for more details.
9984 + *
9985 + * You should have received a copy of the GNU General Public License
9986 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9987 + */
9988 +
9989 +/*
9990 + * directory operations
9991 + */
9992 +
9993 +#include <linux/fs_stack.h>
9994 +#include <linux/iversion.h>
9995 +#include "aufs.h"
9996 +
9997 +void au_add_nlink(struct inode *dir, struct inode *h_dir)
9998 +{
9999 +       unsigned int nlink;
10000 +
10001 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
10002 +
10003 +       nlink = dir->i_nlink;
10004 +       nlink += h_dir->i_nlink - 2;
10005 +       if (h_dir->i_nlink < 2)
10006 +               nlink += 2;
10007 +       smp_mb(); /* for i_nlink */
10008 +       /* 0 can happen in revaliding */
10009 +       set_nlink(dir, nlink);
10010 +}
10011 +
10012 +void au_sub_nlink(struct inode *dir, struct inode *h_dir)
10013 +{
10014 +       unsigned int nlink;
10015 +
10016 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
10017 +
10018 +       nlink = dir->i_nlink;
10019 +       nlink -= h_dir->i_nlink - 2;
10020 +       if (h_dir->i_nlink < 2)
10021 +               nlink -= 2;
10022 +       smp_mb(); /* for i_nlink */
10023 +       /* nlink == 0 means the branch-fs is broken */
10024 +       set_nlink(dir, nlink);
10025 +}
10026 +
10027 +loff_t au_dir_size(struct file *file, struct dentry *dentry)
10028 +{
10029 +       loff_t sz;
10030 +       aufs_bindex_t bindex, bbot;
10031 +       struct file *h_file;
10032 +       struct dentry *h_dentry;
10033 +
10034 +       sz = 0;
10035 +       if (file) {
10036 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
10037 +
10038 +               bbot = au_fbbot_dir(file);
10039 +               for (bindex = au_fbtop(file);
10040 +                    bindex <= bbot && sz < KMALLOC_MAX_SIZE;
10041 +                    bindex++) {
10042 +                       h_file = au_hf_dir(file, bindex);
10043 +                       if (h_file && file_inode(h_file))
10044 +                               sz += vfsub_f_size_read(h_file);
10045 +               }
10046 +       } else {
10047 +               AuDebugOn(!dentry);
10048 +               AuDebugOn(!d_is_dir(dentry));
10049 +
10050 +               bbot = au_dbtaildir(dentry);
10051 +               for (bindex = au_dbtop(dentry);
10052 +                    bindex <= bbot && sz < KMALLOC_MAX_SIZE;
10053 +                    bindex++) {
10054 +                       h_dentry = au_h_dptr(dentry, bindex);
10055 +                       if (h_dentry && d_is_positive(h_dentry))
10056 +                               sz += i_size_read(d_inode(h_dentry));
10057 +               }
10058 +       }
10059 +       if (sz < KMALLOC_MAX_SIZE)
10060 +               sz = roundup_pow_of_two(sz);
10061 +       if (sz > KMALLOC_MAX_SIZE)
10062 +               sz = KMALLOC_MAX_SIZE;
10063 +       else if (sz < NAME_MAX) {
10064 +               BUILD_BUG_ON(AUFS_RDBLK_DEF < NAME_MAX);
10065 +               sz = AUFS_RDBLK_DEF;
10066 +       }
10067 +       return sz;
10068 +}
10069 +
10070 +struct au_dir_ts_arg {
10071 +       struct dentry *dentry;
10072 +       aufs_bindex_t brid;
10073 +};
10074 +
10075 +static void au_do_dir_ts(void *arg)
10076 +{
10077 +       struct au_dir_ts_arg *a = arg;
10078 +       struct au_dtime dt;
10079 +       struct path h_path;
10080 +       struct inode *dir, *h_dir;
10081 +       struct super_block *sb;
10082 +       struct au_branch *br;
10083 +       struct au_hinode *hdir;
10084 +       int err;
10085 +       aufs_bindex_t btop, bindex;
10086 +
10087 +       sb = a->dentry->d_sb;
10088 +       if (d_really_is_negative(a->dentry))
10089 +               goto out;
10090 +       /* no dir->i_mutex lock */
10091 +       aufs_read_lock(a->dentry, AuLock_DW); /* noflush */
10092 +
10093 +       dir = d_inode(a->dentry);
10094 +       btop = au_ibtop(dir);
10095 +       bindex = au_br_index(sb, a->brid);
10096 +       if (bindex < btop)
10097 +               goto out_unlock;
10098 +
10099 +       br = au_sbr(sb, bindex);
10100 +       h_path.dentry = au_h_dptr(a->dentry, bindex);
10101 +       if (!h_path.dentry)
10102 +               goto out_unlock;
10103 +       h_path.mnt = au_br_mnt(br);
10104 +       au_dtime_store(&dt, a->dentry, &h_path);
10105 +
10106 +       br = au_sbr(sb, btop);
10107 +       if (!au_br_writable(br->br_perm))
10108 +               goto out_unlock;
10109 +       h_path.dentry = au_h_dptr(a->dentry, btop);
10110 +       h_path.mnt = au_br_mnt(br);
10111 +       err = vfsub_mnt_want_write(h_path.mnt);
10112 +       if (err)
10113 +               goto out_unlock;
10114 +       hdir = au_hi(dir, btop);
10115 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
10116 +       h_dir = au_h_iptr(dir, btop);
10117 +       if (h_dir->i_nlink
10118 +           && timespec64_compare(&h_dir->i_mtime, &dt.dt_mtime) < 0) {
10119 +               dt.dt_h_path = h_path;
10120 +               au_dtime_revert(&dt);
10121 +       }
10122 +       au_hn_inode_unlock(hdir);
10123 +       vfsub_mnt_drop_write(h_path.mnt);
10124 +       au_cpup_attr_timesizes(dir);
10125 +
10126 +out_unlock:
10127 +       aufs_read_unlock(a->dentry, AuLock_DW);
10128 +out:
10129 +       dput(a->dentry);
10130 +       au_nwt_done(&au_sbi(sb)->si_nowait);
10131 +       au_kfree_try_rcu(arg);
10132 +}
10133 +
10134 +void au_dir_ts(struct inode *dir, aufs_bindex_t bindex)
10135 +{
10136 +       int perm, wkq_err;
10137 +       aufs_bindex_t btop;
10138 +       struct au_dir_ts_arg *arg;
10139 +       struct dentry *dentry;
10140 +       struct super_block *sb;
10141 +
10142 +       IMustLock(dir);
10143 +
10144 +       dentry = d_find_any_alias(dir);
10145 +       AuDebugOn(!dentry);
10146 +       sb = dentry->d_sb;
10147 +       btop = au_ibtop(dir);
10148 +       if (btop == bindex) {
10149 +               au_cpup_attr_timesizes(dir);
10150 +               goto out;
10151 +       }
10152 +
10153 +       perm = au_sbr_perm(sb, btop);
10154 +       if (!au_br_writable(perm))
10155 +               goto out;
10156 +
10157 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
10158 +       if (!arg)
10159 +               goto out;
10160 +
10161 +       arg->dentry = dget(dentry); /* will be dput-ted by au_do_dir_ts() */
10162 +       arg->brid = au_sbr_id(sb, bindex);
10163 +       wkq_err = au_wkq_nowait(au_do_dir_ts, arg, sb, /*flags*/0);
10164 +       if (unlikely(wkq_err)) {
10165 +               pr_err("wkq %d\n", wkq_err);
10166 +               dput(dentry);
10167 +               au_kfree_try_rcu(arg);
10168 +       }
10169 +
10170 +out:
10171 +       dput(dentry);
10172 +}
10173 +
10174 +/* ---------------------------------------------------------------------- */
10175 +
10176 +static int reopen_dir(struct file *file)
10177 +{
10178 +       int err;
10179 +       unsigned int flags;
10180 +       aufs_bindex_t bindex, btail, btop;
10181 +       struct dentry *dentry, *h_dentry;
10182 +       struct file *h_file;
10183 +
10184 +       /* open all lower dirs */
10185 +       dentry = file->f_path.dentry;
10186 +       btop = au_dbtop(dentry);
10187 +       for (bindex = au_fbtop(file); bindex < btop; bindex++)
10188 +               au_set_h_fptr(file, bindex, NULL);
10189 +       au_set_fbtop(file, btop);
10190 +
10191 +       btail = au_dbtaildir(dentry);
10192 +       for (bindex = au_fbbot_dir(file); btail < bindex; bindex--)
10193 +               au_set_h_fptr(file, bindex, NULL);
10194 +       au_set_fbbot_dir(file, btail);
10195 +
10196 +       flags = vfsub_file_flags(file);
10197 +       for (bindex = btop; bindex <= btail; bindex++) {
10198 +               h_dentry = au_h_dptr(dentry, bindex);
10199 +               if (!h_dentry)
10200 +                       continue;
10201 +               h_file = au_hf_dir(file, bindex);
10202 +               if (h_file)
10203 +                       continue;
10204 +
10205 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
10206 +               err = PTR_ERR(h_file);
10207 +               if (IS_ERR(h_file))
10208 +                       goto out; /* close all? */
10209 +               au_set_h_fptr(file, bindex, h_file);
10210 +       }
10211 +       au_update_figen(file);
10212 +       /* todo: necessary? */
10213 +       /* file->f_ra = h_file->f_ra; */
10214 +       err = 0;
10215 +
10216 +out:
10217 +       return err;
10218 +}
10219 +
10220 +static int do_open_dir(struct file *file, int flags, struct file *h_file)
10221 +{
10222 +       int err;
10223 +       aufs_bindex_t bindex, btail;
10224 +       struct dentry *dentry, *h_dentry;
10225 +       struct vfsmount *mnt;
10226 +
10227 +       FiMustWriteLock(file);
10228 +       AuDebugOn(h_file);
10229 +
10230 +       err = 0;
10231 +       mnt = file->f_path.mnt;
10232 +       dentry = file->f_path.dentry;
10233 +       file->f_version = inode_query_iversion(d_inode(dentry));
10234 +       bindex = au_dbtop(dentry);
10235 +       au_set_fbtop(file, bindex);
10236 +       btail = au_dbtaildir(dentry);
10237 +       au_set_fbbot_dir(file, btail);
10238 +       for (; !err && bindex <= btail; bindex++) {
10239 +               h_dentry = au_h_dptr(dentry, bindex);
10240 +               if (!h_dentry)
10241 +                       continue;
10242 +
10243 +               err = vfsub_test_mntns(mnt, h_dentry->d_sb);
10244 +               if (unlikely(err))
10245 +                       break;
10246 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
10247 +               if (IS_ERR(h_file)) {
10248 +                       err = PTR_ERR(h_file);
10249 +                       break;
10250 +               }
10251 +               au_set_h_fptr(file, bindex, h_file);
10252 +       }
10253 +       au_update_figen(file);
10254 +       /* todo: necessary? */
10255 +       /* file->f_ra = h_file->f_ra; */
10256 +       if (!err)
10257 +               return 0; /* success */
10258 +
10259 +       /* close all */
10260 +       for (bindex = au_fbtop(file); bindex <= btail; bindex++)
10261 +               au_set_h_fptr(file, bindex, NULL);
10262 +       au_set_fbtop(file, -1);
10263 +       au_set_fbbot_dir(file, -1);
10264 +
10265 +       return err;
10266 +}
10267 +
10268 +static int aufs_open_dir(struct inode *inode __maybe_unused,
10269 +                        struct file *file)
10270 +{
10271 +       int err;
10272 +       struct super_block *sb;
10273 +       struct au_fidir *fidir;
10274 +
10275 +       err = -ENOMEM;
10276 +       sb = file->f_path.dentry->d_sb;
10277 +       si_read_lock(sb, AuLock_FLUSH);
10278 +       fidir = au_fidir_alloc(sb);
10279 +       if (fidir) {
10280 +               struct au_do_open_args args = {
10281 +                       .open   = do_open_dir,
10282 +                       .fidir  = fidir
10283 +               };
10284 +               err = au_do_open(file, &args);
10285 +               if (unlikely(err))
10286 +                       au_kfree_rcu(fidir);
10287 +       }
10288 +       si_read_unlock(sb);
10289 +       return err;
10290 +}
10291 +
10292 +static int aufs_release_dir(struct inode *inode __maybe_unused,
10293 +                           struct file *file)
10294 +{
10295 +       struct au_vdir *vdir_cache;
10296 +       struct au_finfo *finfo;
10297 +       struct au_fidir *fidir;
10298 +       struct au_hfile *hf;
10299 +       aufs_bindex_t bindex, bbot;
10300 +
10301 +       finfo = au_fi(file);
10302 +       fidir = finfo->fi_hdir;
10303 +       if (fidir) {
10304 +               au_hbl_del(&finfo->fi_hlist,
10305 +                          &au_sbi(file->f_path.dentry->d_sb)->si_files);
10306 +               vdir_cache = fidir->fd_vdir_cache; /* lock-free */
10307 +               if (vdir_cache)
10308 +                       au_vdir_free(vdir_cache);
10309 +
10310 +               bindex = finfo->fi_btop;
10311 +               if (bindex >= 0) {
10312 +                       hf = fidir->fd_hfile + bindex;
10313 +                       /*
10314 +                        * calls fput() instead of filp_close(),
10315 +                        * since no dnotify or lock for the lower file.
10316 +                        */
10317 +                       bbot = fidir->fd_bbot;
10318 +                       for (; bindex <= bbot; bindex++, hf++)
10319 +                               if (hf->hf_file)
10320 +                                       au_hfput(hf, /*execed*/0);
10321 +               }
10322 +               au_kfree_rcu(fidir);
10323 +               finfo->fi_hdir = NULL;
10324 +       }
10325 +       au_finfo_fin(file);
10326 +       return 0;
10327 +}
10328 +
10329 +/* ---------------------------------------------------------------------- */
10330 +
10331 +static int au_do_flush_dir(struct file *file, fl_owner_t id)
10332 +{
10333 +       int err;
10334 +       aufs_bindex_t bindex, bbot;
10335 +       struct file *h_file;
10336 +
10337 +       err = 0;
10338 +       bbot = au_fbbot_dir(file);
10339 +       for (bindex = au_fbtop(file); !err && bindex <= bbot; bindex++) {
10340 +               h_file = au_hf_dir(file, bindex);
10341 +               if (h_file)
10342 +                       err = vfsub_flush(h_file, id);
10343 +       }
10344 +       return err;
10345 +}
10346 +
10347 +static int aufs_flush_dir(struct file *file, fl_owner_t id)
10348 +{
10349 +       return au_do_flush(file, id, au_do_flush_dir);
10350 +}
10351 +
10352 +/* ---------------------------------------------------------------------- */
10353 +
10354 +static int au_do_fsync_dir_no_file(struct dentry *dentry, int datasync)
10355 +{
10356 +       int err;
10357 +       aufs_bindex_t bbot, bindex;
10358 +       struct inode *inode;
10359 +       struct super_block *sb;
10360 +
10361 +       err = 0;
10362 +       sb = dentry->d_sb;
10363 +       inode = d_inode(dentry);
10364 +       IMustLock(inode);
10365 +       bbot = au_dbbot(dentry);
10366 +       for (bindex = au_dbtop(dentry); !err && bindex <= bbot; bindex++) {
10367 +               struct path h_path;
10368 +
10369 +               if (au_test_ro(sb, bindex, inode))
10370 +                       continue;
10371 +               h_path.dentry = au_h_dptr(dentry, bindex);
10372 +               if (!h_path.dentry)
10373 +                       continue;
10374 +
10375 +               h_path.mnt = au_sbr_mnt(sb, bindex);
10376 +               err = vfsub_fsync(NULL, &h_path, datasync);
10377 +       }
10378 +
10379 +       return err;
10380 +}
10381 +
10382 +static int au_do_fsync_dir(struct file *file, int datasync)
10383 +{
10384 +       int err;
10385 +       aufs_bindex_t bbot, bindex;
10386 +       struct file *h_file;
10387 +       struct super_block *sb;
10388 +       struct inode *inode;
10389 +
10390 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1, /*fi_lsc*/0);
10391 +       if (unlikely(err))
10392 +               goto out;
10393 +
10394 +       inode = file_inode(file);
10395 +       sb = inode->i_sb;
10396 +       bbot = au_fbbot_dir(file);
10397 +       for (bindex = au_fbtop(file); !err && bindex <= bbot; bindex++) {
10398 +               h_file = au_hf_dir(file, bindex);
10399 +               if (!h_file || au_test_ro(sb, bindex, inode))
10400 +                       continue;
10401 +
10402 +               err = vfsub_fsync(h_file, &h_file->f_path, datasync);
10403 +       }
10404 +
10405 +out:
10406 +       return err;
10407 +}
10408 +
10409 +/*
10410 + * @file may be NULL
10411 + */
10412 +static int aufs_fsync_dir(struct file *file, loff_t start, loff_t end,
10413 +                         int datasync)
10414 +{
10415 +       int err;
10416 +       struct dentry *dentry;
10417 +       struct inode *inode;
10418 +       struct super_block *sb;
10419 +
10420 +       err = 0;
10421 +       dentry = file->f_path.dentry;
10422 +       inode = d_inode(dentry);
10423 +       inode_lock(inode);
10424 +       sb = dentry->d_sb;
10425 +       si_noflush_read_lock(sb);
10426 +       if (file)
10427 +               err = au_do_fsync_dir(file, datasync);
10428 +       else {
10429 +               di_write_lock_child(dentry);
10430 +               err = au_do_fsync_dir_no_file(dentry, datasync);
10431 +       }
10432 +       au_cpup_attr_timesizes(inode);
10433 +       di_write_unlock(dentry);
10434 +       if (file)
10435 +               fi_write_unlock(file);
10436 +
10437 +       si_read_unlock(sb);
10438 +       inode_unlock(inode);
10439 +       return err;
10440 +}
10441 +
10442 +/* ---------------------------------------------------------------------- */
10443 +
10444 +static int aufs_iterate_shared(struct file *file, struct dir_context *ctx)
10445 +{
10446 +       int err;
10447 +       struct dentry *dentry;
10448 +       struct inode *inode, *h_inode;
10449 +       struct super_block *sb;
10450 +
10451 +       AuDbg("%pD, ctx{%ps, %llu}\n", file, ctx->actor, ctx->pos);
10452 +
10453 +       dentry = file->f_path.dentry;
10454 +       inode = d_inode(dentry);
10455 +       IMustLock(inode);
10456 +
10457 +       sb = dentry->d_sb;
10458 +       si_read_lock(sb, AuLock_FLUSH);
10459 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1, /*fi_lsc*/0);
10460 +       if (unlikely(err))
10461 +               goto out;
10462 +       err = au_alive_dir(dentry);
10463 +       if (!err)
10464 +               err = au_vdir_init(file);
10465 +       di_downgrade_lock(dentry, AuLock_IR);
10466 +       if (unlikely(err))
10467 +               goto out_unlock;
10468 +
10469 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
10470 +       if (!au_test_nfsd()) {
10471 +               err = au_vdir_fill_de(file, ctx);
10472 +               fsstack_copy_attr_atime(inode, h_inode);
10473 +       } else {
10474 +               /*
10475 +                * nfsd filldir may call lookup_one_len(), vfs_getattr(),
10476 +                * encode_fh() and others.
10477 +                */
10478 +               atomic_inc(&h_inode->i_count);
10479 +               di_read_unlock(dentry, AuLock_IR);
10480 +               si_read_unlock(sb);
10481 +               err = au_vdir_fill_de(file, ctx);
10482 +               fsstack_copy_attr_atime(inode, h_inode);
10483 +               fi_write_unlock(file);
10484 +               iput(h_inode);
10485 +
10486 +               AuTraceErr(err);
10487 +               return err;
10488 +       }
10489 +
10490 +out_unlock:
10491 +       di_read_unlock(dentry, AuLock_IR);
10492 +       fi_write_unlock(file);
10493 +out:
10494 +       si_read_unlock(sb);
10495 +       return err;
10496 +}
10497 +
10498 +/* ---------------------------------------------------------------------- */
10499 +
10500 +#define AuTestEmpty_WHONLY     1
10501 +#define AuTestEmpty_CALLED     (1 << 1)
10502 +#define AuTestEmpty_SHWH       (1 << 2)
10503 +#define au_ftest_testempty(flags, name)        ((flags) & AuTestEmpty_##name)
10504 +#define au_fset_testempty(flags, name) \
10505 +       do { (flags) |= AuTestEmpty_##name; } while (0)
10506 +#define au_fclr_testempty(flags, name) \
10507 +       do { (flags) &= ~AuTestEmpty_##name; } while (0)
10508 +
10509 +#ifndef CONFIG_AUFS_SHWH
10510 +#undef AuTestEmpty_SHWH
10511 +#define AuTestEmpty_SHWH       0
10512 +#endif
10513 +
10514 +struct test_empty_arg {
10515 +       struct dir_context ctx;
10516 +       struct au_nhash *whlist;
10517 +       unsigned int flags;
10518 +       int err;
10519 +       aufs_bindex_t bindex;
10520 +};
10521 +
10522 +static int test_empty_cb(struct dir_context *ctx, const char *__name,
10523 +                        int namelen, loff_t offset __maybe_unused, u64 ino,
10524 +                        unsigned int d_type)
10525 +{
10526 +       struct test_empty_arg *arg = container_of(ctx, struct test_empty_arg,
10527 +                                                 ctx);
10528 +       char *name = (void *)__name;
10529 +
10530 +       arg->err = 0;
10531 +       au_fset_testempty(arg->flags, CALLED);
10532 +       /* smp_mb(); */
10533 +       if (name[0] == '.'
10534 +           && (namelen == 1 || (name[1] == '.' && namelen == 2)))
10535 +               goto out; /* success */
10536 +
10537 +       if (namelen <= AUFS_WH_PFX_LEN
10538 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
10539 +               if (au_ftest_testempty(arg->flags, WHONLY)
10540 +                   && !au_nhash_test_known_wh(arg->whlist, name, namelen))
10541 +                       arg->err = -ENOTEMPTY;
10542 +               goto out;
10543 +       }
10544 +
10545 +       name += AUFS_WH_PFX_LEN;
10546 +       namelen -= AUFS_WH_PFX_LEN;
10547 +       if (!au_nhash_test_known_wh(arg->whlist, name, namelen))
10548 +               arg->err = au_nhash_append_wh
10549 +                       (arg->whlist, name, namelen, ino, d_type, arg->bindex,
10550 +                        au_ftest_testempty(arg->flags, SHWH));
10551 +
10552 +out:
10553 +       /* smp_mb(); */
10554 +       AuTraceErr(arg->err);
10555 +       return arg->err;
10556 +}
10557 +
10558 +static int do_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
10559 +{
10560 +       int err;
10561 +       struct file *h_file;
10562 +       struct au_branch *br;
10563 +
10564 +       h_file = au_h_open(dentry, arg->bindex,
10565 +                          O_RDONLY | O_NONBLOCK | O_DIRECTORY | O_LARGEFILE,
10566 +                          /*file*/NULL, /*force_wr*/0);
10567 +       err = PTR_ERR(h_file);
10568 +       if (IS_ERR(h_file))
10569 +               goto out;
10570 +
10571 +       err = 0;
10572 +       if (!au_opt_test(au_mntflags(dentry->d_sb), UDBA_NONE)
10573 +           && !file_inode(h_file)->i_nlink)
10574 +               goto out_put;
10575 +
10576 +       do {
10577 +               arg->err = 0;
10578 +               au_fclr_testempty(arg->flags, CALLED);
10579 +               /* smp_mb(); */
10580 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
10581 +               if (err >= 0)
10582 +                       err = arg->err;
10583 +       } while (!err && au_ftest_testempty(arg->flags, CALLED));
10584 +
10585 +out_put:
10586 +       fput(h_file);
10587 +       br = au_sbr(dentry->d_sb, arg->bindex);
10588 +       au_lcnt_dec(&br->br_nfiles);
10589 +out:
10590 +       return err;
10591 +}
10592 +
10593 +struct do_test_empty_args {
10594 +       int *errp;
10595 +       struct dentry *dentry;
10596 +       struct test_empty_arg *arg;
10597 +};
10598 +
10599 +static void call_do_test_empty(void *args)
10600 +{
10601 +       struct do_test_empty_args *a = args;
10602 +       *a->errp = do_test_empty(a->dentry, a->arg);
10603 +}
10604 +
10605 +static int sio_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
10606 +{
10607 +       int err, wkq_err;
10608 +       struct dentry *h_dentry;
10609 +       struct inode *h_inode;
10610 +       struct user_namespace *h_userns;
10611 +
10612 +       h_userns = au_sbr_userns(dentry->d_sb, arg->bindex);
10613 +       h_dentry = au_h_dptr(dentry, arg->bindex);
10614 +       h_inode = d_inode(h_dentry);
10615 +       /* todo: i_mode changes anytime? */
10616 +       inode_lock_shared_nested(h_inode, AuLsc_I_CHILD);
10617 +       err = au_test_h_perm_sio(h_userns, h_inode, MAY_EXEC | MAY_READ);
10618 +       inode_unlock_shared(h_inode);
10619 +       if (!err)
10620 +               err = do_test_empty(dentry, arg);
10621 +       else {
10622 +               struct do_test_empty_args args = {
10623 +                       .errp   = &err,
10624 +                       .dentry = dentry,
10625 +                       .arg    = arg
10626 +               };
10627 +               unsigned int flags = arg->flags;
10628 +
10629 +               wkq_err = au_wkq_wait(call_do_test_empty, &args);
10630 +               if (unlikely(wkq_err))
10631 +                       err = wkq_err;
10632 +               arg->flags = flags;
10633 +       }
10634 +
10635 +       return err;
10636 +}
10637 +
10638 +int au_test_empty_lower(struct dentry *dentry)
10639 +{
10640 +       int err;
10641 +       unsigned int rdhash;
10642 +       aufs_bindex_t bindex, btop, btail;
10643 +       struct au_nhash whlist;
10644 +       struct test_empty_arg arg = {
10645 +               .ctx = {
10646 +                       .actor = test_empty_cb
10647 +               }
10648 +       };
10649 +       int (*test_empty)(struct dentry *dentry, struct test_empty_arg *arg);
10650 +
10651 +       SiMustAnyLock(dentry->d_sb);
10652 +
10653 +       rdhash = au_sbi(dentry->d_sb)->si_rdhash;
10654 +       if (!rdhash)
10655 +               rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, dentry));
10656 +       err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
10657 +       if (unlikely(err))
10658 +               goto out;
10659 +
10660 +       arg.flags = 0;
10661 +       arg.whlist = &whlist;
10662 +       btop = au_dbtop(dentry);
10663 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
10664 +               au_fset_testempty(arg.flags, SHWH);
10665 +       test_empty = do_test_empty;
10666 +       if (au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1))
10667 +               test_empty = sio_test_empty;
10668 +       arg.bindex = btop;
10669 +       err = test_empty(dentry, &arg);
10670 +       if (unlikely(err))
10671 +               goto out_whlist;
10672 +
10673 +       au_fset_testempty(arg.flags, WHONLY);
10674 +       btail = au_dbtaildir(dentry);
10675 +       for (bindex = btop + 1; !err && bindex <= btail; bindex++) {
10676 +               struct dentry *h_dentry;
10677 +
10678 +               h_dentry = au_h_dptr(dentry, bindex);
10679 +               if (h_dentry && d_is_positive(h_dentry)) {
10680 +                       arg.bindex = bindex;
10681 +                       err = test_empty(dentry, &arg);
10682 +               }
10683 +       }
10684 +
10685 +out_whlist:
10686 +       au_nhash_wh_free(&whlist);
10687 +out:
10688 +       return err;
10689 +}
10690 +
10691 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist)
10692 +{
10693 +       int err;
10694 +       struct test_empty_arg arg = {
10695 +               .ctx = {
10696 +                       .actor = test_empty_cb
10697 +               }
10698 +       };
10699 +       aufs_bindex_t bindex, btail;
10700 +
10701 +       err = 0;
10702 +       arg.whlist = whlist;
10703 +       arg.flags = AuTestEmpty_WHONLY;
10704 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
10705 +               au_fset_testempty(arg.flags, SHWH);
10706 +       btail = au_dbtaildir(dentry);
10707 +       for (bindex = au_dbtop(dentry); !err && bindex <= btail; bindex++) {
10708 +               struct dentry *h_dentry;
10709 +
10710 +               h_dentry = au_h_dptr(dentry, bindex);
10711 +               if (h_dentry && d_is_positive(h_dentry)) {
10712 +                       arg.bindex = bindex;
10713 +                       err = sio_test_empty(dentry, &arg);
10714 +               }
10715 +       }
10716 +
10717 +       return err;
10718 +}
10719 +
10720 +/* ---------------------------------------------------------------------- */
10721 +
10722 +const struct file_operations aufs_dir_fop = {
10723 +       .owner          = THIS_MODULE,
10724 +       .llseek         = default_llseek,
10725 +       .read           = generic_read_dir,
10726 +       .iterate_shared = aufs_iterate_shared,
10727 +       .unlocked_ioctl = aufs_ioctl_dir,
10728 +#ifdef CONFIG_COMPAT
10729 +       .compat_ioctl   = aufs_compat_ioctl_dir,
10730 +#endif
10731 +       .open           = aufs_open_dir,
10732 +       .release        = aufs_release_dir,
10733 +       .flush          = aufs_flush_dir,
10734 +       .fsync          = aufs_fsync_dir
10735 +};
10736 diff -urN /usr/share/empty/fs/aufs/dir.h linux/fs/aufs/dir.h
10737 --- /usr/share/empty/fs/aufs/dir.h      1970-01-01 01:00:00.000000000 +0100
10738 +++ linux/fs/aufs/dir.h 2021-12-03 15:38:59.933313976 +0100
10739 @@ -0,0 +1,134 @@
10740 +/* SPDX-License-Identifier: GPL-2.0 */
10741 +/*
10742 + * Copyright (C) 2005-2021 Junjiro R. Okajima
10743 + *
10744 + * This program, aufs is free software; you can redistribute it and/or modify
10745 + * it under the terms of the GNU General Public License as published by
10746 + * the Free Software Foundation; either version 2 of the License, or
10747 + * (at your option) any later version.
10748 + *
10749 + * This program is distributed in the hope that it will be useful,
10750 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10751 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10752 + * GNU General Public License for more details.
10753 + *
10754 + * You should have received a copy of the GNU General Public License
10755 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10756 + */
10757 +
10758 +/*
10759 + * directory operations
10760 + */
10761 +
10762 +#ifndef __AUFS_DIR_H__
10763 +#define __AUFS_DIR_H__
10764 +
10765 +#ifdef __KERNEL__
10766 +
10767 +#include <linux/fs.h>
10768 +
10769 +/* ---------------------------------------------------------------------- */
10770 +
10771 +/* need to be faster and smaller */
10772 +
10773 +struct au_nhash {
10774 +       unsigned int            nh_num;
10775 +       struct hlist_head       *nh_head;
10776 +};
10777 +
10778 +struct au_vdir_destr {
10779 +       unsigned char   len;
10780 +       unsigned char   name[];
10781 +} __packed;
10782 +
10783 +struct au_vdir_dehstr {
10784 +       struct hlist_node       hash;
10785 +       struct au_vdir_destr    *str;
10786 +       struct rcu_head         rcu;
10787 +} ____cacheline_aligned_in_smp;
10788 +
10789 +struct au_vdir_de {
10790 +       ino_t                   de_ino;
10791 +       unsigned char           de_type;
10792 +       /* caution: packed */
10793 +       struct au_vdir_destr    de_str;
10794 +} __packed;
10795 +
10796 +struct au_vdir_wh {
10797 +       struct hlist_node       wh_hash;
10798 +#ifdef CONFIG_AUFS_SHWH
10799 +       ino_t                   wh_ino;
10800 +       aufs_bindex_t           wh_bindex;
10801 +       unsigned char           wh_type;
10802 +#else
10803 +       aufs_bindex_t           wh_bindex;
10804 +#endif
10805 +       /* caution: packed */
10806 +       struct au_vdir_destr    wh_str;
10807 +} __packed;
10808 +
10809 +union au_vdir_deblk_p {
10810 +       unsigned char           *deblk;
10811 +       struct au_vdir_de       *de;
10812 +};
10813 +
10814 +struct au_vdir {
10815 +       unsigned char   **vd_deblk;
10816 +       unsigned long   vd_nblk;
10817 +       struct {
10818 +               unsigned long           ul;
10819 +               union au_vdir_deblk_p   p;
10820 +       } vd_last;
10821 +
10822 +       u64             vd_version;
10823 +       unsigned int    vd_deblk_sz;
10824 +       unsigned long   vd_jiffy;
10825 +       struct rcu_head rcu;
10826 +} ____cacheline_aligned_in_smp;
10827 +
10828 +/* ---------------------------------------------------------------------- */
10829 +
10830 +/* dir.c */
10831 +extern const struct file_operations aufs_dir_fop;
10832 +void au_add_nlink(struct inode *dir, struct inode *h_dir);
10833 +void au_sub_nlink(struct inode *dir, struct inode *h_dir);
10834 +loff_t au_dir_size(struct file *file, struct dentry *dentry);
10835 +void au_dir_ts(struct inode *dir, aufs_bindex_t bsrc);
10836 +int au_test_empty_lower(struct dentry *dentry);
10837 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist);
10838 +
10839 +/* vdir.c */
10840 +unsigned int au_rdhash_est(loff_t sz);
10841 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp);
10842 +void au_nhash_wh_free(struct au_nhash *whlist);
10843 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
10844 +                           int limit);
10845 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen);
10846 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
10847 +                      unsigned int d_type, aufs_bindex_t bindex,
10848 +                      unsigned char shwh);
10849 +void au_vdir_free(struct au_vdir *vdir);
10850 +int au_vdir_init(struct file *file);
10851 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx);
10852 +
10853 +/* ioctl.c */
10854 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg);
10855 +
10856 +#ifdef CONFIG_AUFS_RDU
10857 +/* rdu.c */
10858 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
10859 +#ifdef CONFIG_COMPAT
10860 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd,
10861 +                        unsigned long arg);
10862 +#endif
10863 +#else
10864 +AuStub(long, au_rdu_ioctl, return -EINVAL, struct file *file,
10865 +       unsigned int cmd, unsigned long arg)
10866 +#ifdef CONFIG_COMPAT
10867 +AuStub(long, au_rdu_compat_ioctl, return -EINVAL, struct file *file,
10868 +       unsigned int cmd, unsigned long arg)
10869 +#endif
10870 +#endif
10871 +
10872 +#endif /* __KERNEL__ */
10873 +#endif /* __AUFS_DIR_H__ */
10874 diff -urN /usr/share/empty/fs/aufs/dirren.c linux/fs/aufs/dirren.c
10875 --- /usr/share/empty/fs/aufs/dirren.c   1970-01-01 01:00:00.000000000 +0100
10876 +++ linux/fs/aufs/dirren.c      2021-12-03 15:38:59.933313976 +0100
10877 @@ -0,0 +1,1315 @@
10878 +// SPDX-License-Identifier: GPL-2.0
10879 +/*
10880 + * Copyright (C) 2017-2021 Junjiro R. Okajima
10881 + *
10882 + * This program, aufs is free software; you can redistribute it and/or modify
10883 + * it under the terms of the GNU General Public License as published by
10884 + * the Free Software Foundation; either version 2 of the License, or
10885 + * (at your option) any later version.
10886 + *
10887 + * This program is distributed in the hope that it will be useful,
10888 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10889 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10890 + * GNU General Public License for more details.
10891 + *
10892 + * You should have received a copy of the GNU General Public License
10893 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10894 + */
10895 +
10896 +/*
10897 + * special handling in renaming a directory
10898 + * in order to support looking-up the before-renamed name on the lower readonly
10899 + * branches
10900 + */
10901 +
10902 +#include <linux/byteorder/generic.h>
10903 +#include "aufs.h"
10904 +
10905 +static void au_dr_hino_del(struct au_dr_br *dr, struct au_dr_hino *ent)
10906 +{
10907 +       int idx;
10908 +
10909 +       idx = au_dr_ihash(ent->dr_h_ino);
10910 +       au_hbl_del(&ent->dr_hnode, dr->dr_h_ino + idx);
10911 +}
10912 +
10913 +static int au_dr_hino_test_empty(struct au_dr_br *dr)
10914 +{
10915 +       int ret, i;
10916 +       struct hlist_bl_head *hbl;
10917 +
10918 +       ret = 1;
10919 +       for (i = 0; ret && i < AuDirren_NHASH; i++) {
10920 +               hbl = dr->dr_h_ino + i;
10921 +               hlist_bl_lock(hbl);
10922 +               ret &= hlist_bl_empty(hbl);
10923 +               hlist_bl_unlock(hbl);
10924 +       }
10925 +
10926 +       return ret;
10927 +}
10928 +
10929 +static struct au_dr_hino *au_dr_hino_find(struct au_dr_br *dr, ino_t ino)
10930 +{
10931 +       struct au_dr_hino *found, *ent;
10932 +       struct hlist_bl_head *hbl;
10933 +       struct hlist_bl_node *pos;
10934 +       int idx;
10935 +
10936 +       found = NULL;
10937 +       idx = au_dr_ihash(ino);
10938 +       hbl = dr->dr_h_ino + idx;
10939 +       hlist_bl_lock(hbl);
10940 +       hlist_bl_for_each_entry(ent, pos, hbl, dr_hnode)
10941 +               if (ent->dr_h_ino == ino) {
10942 +                       found = ent;
10943 +                       break;
10944 +               }
10945 +       hlist_bl_unlock(hbl);
10946 +
10947 +       return found;
10948 +}
10949 +
10950 +int au_dr_hino_test_add(struct au_dr_br *dr, ino_t ino,
10951 +                       struct au_dr_hino *add_ent)
10952 +{
10953 +       int found, idx;
10954 +       struct hlist_bl_head *hbl;
10955 +       struct hlist_bl_node *pos;
10956 +       struct au_dr_hino *ent;
10957 +
10958 +       found = 0;
10959 +       idx = au_dr_ihash(ino);
10960 +       hbl = dr->dr_h_ino + idx;
10961 +#if 0 /* debug print */
10962 +       {
10963 +               struct hlist_bl_node *tmp;
10964 +
10965 +               hlist_bl_for_each_entry_safe(ent, pos, tmp, hbl, dr_hnode)
10966 +                       AuDbg("hi%llu\n", (unsigned long long)ent->dr_h_ino);
10967 +       }
10968 +#endif
10969 +       hlist_bl_lock(hbl);
10970 +       hlist_bl_for_each_entry(ent, pos, hbl, dr_hnode)
10971 +               if (ent->dr_h_ino == ino) {
10972 +                       found = 1;
10973 +                       break;
10974 +               }
10975 +       if (!found && add_ent)
10976 +               hlist_bl_add_head(&add_ent->dr_hnode, hbl);
10977 +       hlist_bl_unlock(hbl);
10978 +
10979 +       if (!found && add_ent)
10980 +               AuDbg("i%llu added\n", (unsigned long long)add_ent->dr_h_ino);
10981 +
10982 +       return found;
10983 +}
10984 +
10985 +void au_dr_hino_free(struct au_dr_br *dr)
10986 +{
10987 +       int i;
10988 +       struct hlist_bl_head *hbl;
10989 +       struct hlist_bl_node *pos, *tmp;
10990 +       struct au_dr_hino *ent;
10991 +
10992 +       /* SiMustWriteLock(sb); */
10993 +
10994 +       for (i = 0; i < AuDirren_NHASH; i++) {
10995 +               hbl = dr->dr_h_ino + i;
10996 +               /* no spinlock since sbinfo must be write-locked */
10997 +               hlist_bl_for_each_entry_safe(ent, pos, tmp, hbl, dr_hnode)
10998 +                       au_kfree_rcu(ent);
10999 +               INIT_HLIST_BL_HEAD(hbl);
11000 +       }
11001 +}
11002 +
11003 +/* returns the number of inodes or an error */
11004 +static int au_dr_hino_store(struct super_block *sb, struct au_branch *br,
11005 +                           struct file *hinofile)
11006 +{
11007 +       int err, i;
11008 +       ssize_t ssz;
11009 +       loff_t pos, oldsize;
11010 +       __be64 u64;
11011 +       struct inode *hinoinode;
11012 +       struct hlist_bl_head *hbl;
11013 +       struct hlist_bl_node *n1, *n2;
11014 +       struct au_dr_hino *ent;
11015 +
11016 +       SiMustWriteLock(sb);
11017 +       AuDebugOn(!au_br_writable(br->br_perm));
11018 +
11019 +       hinoinode = file_inode(hinofile);
11020 +       oldsize = i_size_read(hinoinode);
11021 +
11022 +       err = 0;
11023 +       pos = 0;
11024 +       hbl = br->br_dirren.dr_h_ino;
11025 +       for (i = 0; !err && i < AuDirren_NHASH; i++, hbl++) {
11026 +               /* no bit-lock since sbinfo must be write-locked */
11027 +               hlist_bl_for_each_entry_safe(ent, n1, n2, hbl, dr_hnode) {
11028 +                       AuDbg("hi%llu, %pD2\n",
11029 +                             (unsigned long long)ent->dr_h_ino, hinofile);
11030 +                       u64 = cpu_to_be64(ent->dr_h_ino);
11031 +                       ssz = vfsub_write_k(hinofile, &u64, sizeof(u64), &pos);
11032 +                       if (ssz == sizeof(u64))
11033 +                               continue;
11034 +
11035 +                       /* write error */
11036 +                       pr_err("ssz %zd, %pD2\n", ssz, hinofile);
11037 +                       err = -ENOSPC;
11038 +                       if (ssz < 0)
11039 +                               err = ssz;
11040 +                       break;
11041 +               }
11042 +       }
11043 +       /* regardless the error */
11044 +       if (pos < oldsize) {
11045 +               err = vfsub_trunc(&hinofile->f_path, pos, /*attr*/0, hinofile);
11046 +               AuTraceErr(err);
11047 +       }
11048 +
11049 +       AuTraceErr(err);
11050 +       return err;
11051 +}
11052 +
11053 +static int au_dr_hino_load(struct au_dr_br *dr, struct file *hinofile)
11054 +{
11055 +       int err, hidx;
11056 +       ssize_t ssz;
11057 +       size_t sz, n;
11058 +       loff_t pos;
11059 +       uint64_t u64;
11060 +       struct au_dr_hino *ent;
11061 +       struct inode *hinoinode;
11062 +       struct hlist_bl_head *hbl;
11063 +
11064 +       err = 0;
11065 +       pos = 0;
11066 +       hbl = dr->dr_h_ino;
11067 +       hinoinode = file_inode(hinofile);
11068 +       sz = i_size_read(hinoinode);
11069 +       AuDebugOn(sz % sizeof(u64));
11070 +       n = sz / sizeof(u64);
11071 +       while (n--) {
11072 +               ssz = vfsub_read_k(hinofile, &u64, sizeof(u64), &pos);
11073 +               if (unlikely(ssz != sizeof(u64))) {
11074 +                       pr_err("ssz %zd, %pD2\n", ssz, hinofile);
11075 +                       err = -EINVAL;
11076 +                       if (ssz < 0)
11077 +                               err = ssz;
11078 +                       goto out_free;
11079 +               }
11080 +
11081 +               ent = kmalloc(sizeof(*ent), GFP_NOFS);
11082 +               if (!ent) {
11083 +                       err = -ENOMEM;
11084 +                       AuTraceErr(err);
11085 +                       goto out_free;
11086 +               }
11087 +               ent->dr_h_ino = be64_to_cpu((__force __be64)u64);
11088 +               AuDbg("hi%llu, %pD2\n",
11089 +                     (unsigned long long)ent->dr_h_ino, hinofile);
11090 +               hidx = au_dr_ihash(ent->dr_h_ino);
11091 +               au_hbl_add(&ent->dr_hnode, hbl + hidx);
11092 +       }
11093 +       goto out; /* success */
11094 +
11095 +out_free:
11096 +       au_dr_hino_free(dr);
11097 +out:
11098 +       AuTraceErr(err);
11099 +       return err;
11100 +}
11101 +
11102 +/*
11103 + * @bindex/@br is a switch to distinguish whether suspending hnotify or not.
11104 + * @path is a switch to distinguish load and store.
11105 + */
11106 +static int au_dr_hino(struct super_block *sb, aufs_bindex_t bindex,
11107 +                     struct au_branch *br, const struct path *path)
11108 +{
11109 +       int err, flags;
11110 +       unsigned char load, suspend;
11111 +       struct file *hinofile;
11112 +       struct au_hinode *hdir;
11113 +       struct inode *dir, *delegated;
11114 +       struct path hinopath;
11115 +       struct qstr hinoname = QSTR_INIT(AUFS_WH_DR_BRHINO,
11116 +                                        sizeof(AUFS_WH_DR_BRHINO) - 1);
11117 +
11118 +       AuDebugOn(bindex < 0 && !br);
11119 +       AuDebugOn(bindex >= 0 && br);
11120 +
11121 +       err = -EINVAL;
11122 +       suspend = !br;
11123 +       if (suspend)
11124 +               br = au_sbr(sb, bindex);
11125 +       load = !!path;
11126 +       if (!load) {
11127 +               path = &br->br_path;
11128 +               AuDebugOn(!au_br_writable(br->br_perm));
11129 +               if (unlikely(!au_br_writable(br->br_perm)))
11130 +                       goto out;
11131 +       }
11132 +
11133 +       hdir = NULL;
11134 +       if (suspend) {
11135 +               dir = d_inode(sb->s_root);
11136 +               hdir = au_hinode(au_ii(dir), bindex);
11137 +               dir = hdir->hi_inode;
11138 +               au_hn_inode_lock_nested(hdir, AuLsc_I_CHILD);
11139 +       } else {
11140 +               dir = d_inode(path->dentry);
11141 +               inode_lock_nested(dir, AuLsc_I_CHILD);
11142 +       }
11143 +       hinopath.mnt = path->mnt;
11144 +       hinopath.dentry = vfsub_lkup_one(&hinoname, (struct path *)path);
11145 +       err = PTR_ERR(hinopath.dentry);
11146 +       if (IS_ERR(hinopath.dentry))
11147 +               goto out_unlock;
11148 +
11149 +       err = 0;
11150 +       flags = O_RDONLY;
11151 +       if (load) {
11152 +               if (d_is_negative(hinopath.dentry))
11153 +                       goto out_dput; /* success */
11154 +       } else {
11155 +               if (au_dr_hino_test_empty(&br->br_dirren)) {
11156 +                       if (d_is_positive(hinopath.dentry)) {
11157 +                               delegated = NULL;
11158 +                               err = vfsub_unlink(dir, &hinopath, &delegated,
11159 +                                                  /*force*/0);
11160 +                               AuTraceErr(err);
11161 +                               if (unlikely(err))
11162 +                                       pr_err("ignored err %d, %pd2\n",
11163 +                                              err, hinopath.dentry);
11164 +                               if (unlikely(err == -EWOULDBLOCK))
11165 +                                       iput(delegated);
11166 +                               err = 0;
11167 +                       }
11168 +                       goto out_dput;
11169 +               } else if (!d_is_positive(hinopath.dentry)) {
11170 +                       err = vfsub_create(dir, &hinopath, 0600,
11171 +                                          /*want_excl*/false);
11172 +                       AuTraceErr(err);
11173 +                       if (unlikely(err))
11174 +                               goto out_dput;
11175 +               }
11176 +               flags = O_WRONLY;
11177 +       }
11178 +       hinofile = vfsub_dentry_open(&hinopath, flags);
11179 +       if (suspend)
11180 +               au_hn_inode_unlock(hdir);
11181 +       else
11182 +               inode_unlock(dir);
11183 +       dput(hinopath.dentry);
11184 +       AuTraceErrPtr(hinofile);
11185 +       if (IS_ERR(hinofile)) {
11186 +               err = PTR_ERR(hinofile);
11187 +               goto out;
11188 +       }
11189 +
11190 +       if (load)
11191 +               err = au_dr_hino_load(&br->br_dirren, hinofile);
11192 +       else
11193 +               err = au_dr_hino_store(sb, br, hinofile);
11194 +       fput(hinofile);
11195 +       goto out;
11196 +
11197 +out_dput:
11198 +       dput(hinopath.dentry);
11199 +out_unlock:
11200 +       if (suspend)
11201 +               au_hn_inode_unlock(hdir);
11202 +       else
11203 +               inode_unlock(dir);
11204 +out:
11205 +       AuTraceErr(err);
11206 +       return err;
11207 +}
11208 +
11209 +/* ---------------------------------------------------------------------- */
11210 +
11211 +static int au_dr_brid_init(struct au_dr_brid *brid, const struct path *path)
11212 +{
11213 +       int err;
11214 +       struct kstatfs kstfs;
11215 +       dev_t dev;
11216 +       struct dentry *dentry;
11217 +       struct super_block *sb;
11218 +
11219 +       err = vfs_statfs((void *)path, &kstfs);
11220 +       AuTraceErr(err);
11221 +       if (unlikely(err))
11222 +               goto out;
11223 +
11224 +       /* todo: support for UUID */
11225 +
11226 +       if (kstfs.f_fsid.val[0] || kstfs.f_fsid.val[1]) {
11227 +               brid->type = AuBrid_FSID;
11228 +               brid->fsid = kstfs.f_fsid;
11229 +       } else {
11230 +               dentry = path->dentry;
11231 +               sb = dentry->d_sb;
11232 +               dev = sb->s_dev;
11233 +               if (dev) {
11234 +                       brid->type = AuBrid_DEV;
11235 +                       brid->dev = dev;
11236 +               }
11237 +       }
11238 +
11239 +out:
11240 +       return err;
11241 +}
11242 +
11243 +int au_dr_br_init(struct super_block *sb, struct au_branch *br,
11244 +                 const struct path *path)
11245 +{
11246 +       int err, i;
11247 +       struct au_dr_br *dr;
11248 +       struct hlist_bl_head *hbl;
11249 +
11250 +       dr = &br->br_dirren;
11251 +       hbl = dr->dr_h_ino;
11252 +       for (i = 0; i < AuDirren_NHASH; i++, hbl++)
11253 +               INIT_HLIST_BL_HEAD(hbl);
11254 +
11255 +       err = au_dr_brid_init(&dr->dr_brid, path);
11256 +       if (unlikely(err))
11257 +               goto out;
11258 +
11259 +       if (au_opt_test(au_mntflags(sb), DIRREN))
11260 +               err = au_dr_hino(sb, /*bindex*/-1, br, path);
11261 +
11262 +out:
11263 +       AuTraceErr(err);
11264 +       return err;
11265 +}
11266 +
11267 +int au_dr_br_fin(struct super_block *sb, struct au_branch *br)
11268 +{
11269 +       int err;
11270 +
11271 +       err = 0;
11272 +       if (au_br_writable(br->br_perm))
11273 +               err = au_dr_hino(sb, /*bindex*/-1, br, /*path*/NULL);
11274 +       if (!err)
11275 +               au_dr_hino_free(&br->br_dirren);
11276 +
11277 +       return err;
11278 +}
11279 +
11280 +/* ---------------------------------------------------------------------- */
11281 +
11282 +static int au_brid_str(struct au_dr_brid *brid, struct inode *h_inode,
11283 +                      char *buf, size_t sz)
11284 +{
11285 +       int err;
11286 +       unsigned int major, minor;
11287 +       char *p;
11288 +
11289 +       p = buf;
11290 +       err = snprintf(p, sz, "%d_", brid->type);
11291 +       AuDebugOn(err > sz);
11292 +       p += err;
11293 +       sz -= err;
11294 +       switch (brid->type) {
11295 +       case AuBrid_Unset:
11296 +               return -EINVAL;
11297 +       case AuBrid_UUID:
11298 +               err = snprintf(p, sz, "%pU", brid->uuid.b);
11299 +               break;
11300 +       case AuBrid_FSID:
11301 +               err = snprintf(p, sz, "%08x-%08x",
11302 +                              brid->fsid.val[0], brid->fsid.val[1]);
11303 +               break;
11304 +       case AuBrid_DEV:
11305 +               major = MAJOR(brid->dev);
11306 +               minor = MINOR(brid->dev);
11307 +               if (major <= 0xff && minor <= 0xff)
11308 +                       err = snprintf(p, sz, "%02x%02x", major, minor);
11309 +               else
11310 +                       err = snprintf(p, sz, "%03x:%05x", major, minor);
11311 +               break;
11312 +       }
11313 +       AuDebugOn(err > sz);
11314 +       p += err;
11315 +       sz -= err;
11316 +       err = snprintf(p, sz, "_%llu", (unsigned long long)h_inode->i_ino);
11317 +       AuDebugOn(err > sz);
11318 +       p += err;
11319 +       sz -= err;
11320 +
11321 +       return p - buf;
11322 +}
11323 +
11324 +static int au_drinfo_name(struct au_branch *br, char *name, int len)
11325 +{
11326 +       int rlen;
11327 +       struct dentry *br_dentry;
11328 +       struct inode *br_inode;
11329 +
11330 +       br_dentry = au_br_dentry(br);
11331 +       br_inode = d_inode(br_dentry);
11332 +       rlen = au_brid_str(&br->br_dirren.dr_brid, br_inode, name, len);
11333 +       AuDebugOn(rlen >= AUFS_DIRREN_ENV_VAL_SZ);
11334 +       AuDebugOn(rlen > len);
11335 +
11336 +       return rlen;
11337 +}
11338 +
11339 +/* ---------------------------------------------------------------------- */
11340 +
11341 +/*
11342 + * from the given @h_dentry, construct drinfo at @*fdata.
11343 + * when the size of @*fdata is not enough, reallocate and return new @fdata and
11344 + * @allocated.
11345 + */
11346 +static int au_drinfo_construct(struct au_drinfo_fdata **fdata,
11347 +                              struct dentry *h_dentry,
11348 +                              unsigned char *allocated)
11349 +{
11350 +       int err, v;
11351 +       struct au_drinfo_fdata *f, *p;
11352 +       struct au_drinfo *drinfo;
11353 +       struct inode *h_inode;
11354 +       struct qstr *qname;
11355 +
11356 +       err = 0;
11357 +       f = *fdata;
11358 +       h_inode = d_inode(h_dentry);
11359 +       qname = &h_dentry->d_name;
11360 +       drinfo = &f->drinfo;
11361 +       drinfo->ino = (__force uint64_t)cpu_to_be64(h_inode->i_ino);
11362 +       drinfo->oldnamelen = qname->len;
11363 +       if (*allocated < sizeof(*f) + qname->len) {
11364 +               v = roundup_pow_of_two(*allocated + qname->len);
11365 +               p = au_krealloc(f, v, GFP_NOFS, /*may_shrink*/0);
11366 +               if (unlikely(!p)) {
11367 +                       err = -ENOMEM;
11368 +                       AuTraceErr(err);
11369 +                       goto out;
11370 +               }
11371 +               f = p;
11372 +               *fdata = f;
11373 +               *allocated = v;
11374 +               drinfo = &f->drinfo;
11375 +       }
11376 +       memcpy(drinfo->oldname, qname->name, qname->len);
11377 +       AuDbg("i%llu, %.*s\n",
11378 +             be64_to_cpu((__force __be64)drinfo->ino), drinfo->oldnamelen,
11379 +             drinfo->oldname);
11380 +
11381 +out:
11382 +       AuTraceErr(err);
11383 +       return err;
11384 +}
11385 +
11386 +/* callers have to free the return value */
11387 +static struct au_drinfo *au_drinfo_read_k(struct file *file, ino_t h_ino)
11388 +{
11389 +       struct au_drinfo *ret, *drinfo;
11390 +       struct au_drinfo_fdata fdata;
11391 +       int len;
11392 +       loff_t pos;
11393 +       ssize_t ssz;
11394 +
11395 +       ret = ERR_PTR(-EIO);
11396 +       pos = 0;
11397 +       ssz = vfsub_read_k(file, &fdata, sizeof(fdata), &pos);
11398 +       if (unlikely(ssz != sizeof(fdata))) {
11399 +               AuIOErr("ssz %zd, %u, %pD2\n",
11400 +                       ssz, (unsigned int)sizeof(fdata), file);
11401 +               goto out;
11402 +       }
11403 +
11404 +       fdata.magic = ntohl((__force __be32)fdata.magic);
11405 +       switch (fdata.magic) {
11406 +       case AUFS_DRINFO_MAGIC_V1:
11407 +               break;
11408 +       default:
11409 +               AuIOErr("magic-num 0x%x, 0x%x, %pD2\n",
11410 +                       fdata.magic, AUFS_DRINFO_MAGIC_V1, file);
11411 +               goto out;
11412 +       }
11413 +
11414 +       drinfo = &fdata.drinfo;
11415 +       len = drinfo->oldnamelen;
11416 +       if (!len) {
11417 +               AuIOErr("broken drinfo %pD2\n", file);
11418 +               goto out;
11419 +       }
11420 +
11421 +       ret = NULL;
11422 +       drinfo->ino = be64_to_cpu((__force __be64)drinfo->ino);
11423 +       if (unlikely(h_ino && drinfo->ino != h_ino)) {
11424 +               AuDbg("ignored i%llu, i%llu, %pD2\n",
11425 +                     (unsigned long long)drinfo->ino,
11426 +                     (unsigned long long)h_ino, file);
11427 +               goto out; /* success */
11428 +       }
11429 +
11430 +       ret = kmalloc(sizeof(*ret) + len, GFP_NOFS);
11431 +       if (unlikely(!ret)) {
11432 +               ret = ERR_PTR(-ENOMEM);
11433 +               AuTraceErrPtr(ret);
11434 +               goto out;
11435 +       }
11436 +
11437 +       *ret = *drinfo;
11438 +       ssz = vfsub_read_k(file, (void *)ret->oldname, len, &pos);
11439 +       if (unlikely(ssz != len)) {
11440 +               au_kfree_rcu(ret);
11441 +               ret = ERR_PTR(-EIO);
11442 +               AuIOErr("ssz %zd, %u, %pD2\n", ssz, len, file);
11443 +               goto out;
11444 +       }
11445 +
11446 +       AuDbg("oldname %.*s\n", ret->oldnamelen, ret->oldname);
11447 +
11448 +out:
11449 +       return ret;
11450 +}
11451 +
11452 +/* ---------------------------------------------------------------------- */
11453 +
11454 +/* in order to be revertible */
11455 +struct au_drinfo_rev_elm {
11456 +       int                     created;
11457 +       struct dentry           *info_dentry;
11458 +       struct au_drinfo        *info_last;
11459 +};
11460 +
11461 +struct au_drinfo_rev {
11462 +       unsigned char                   already;
11463 +       aufs_bindex_t                   nelm;
11464 +       struct au_drinfo_rev_elm        elm[];
11465 +};
11466 +
11467 +/* todo: isn't it too large? */
11468 +struct au_drinfo_store {
11469 +       struct path h_ppath;
11470 +       struct dentry *h_dentry;
11471 +       struct au_drinfo_fdata *fdata;
11472 +       char *infoname;                 /* inside of whname, just after PFX */
11473 +       char whname[sizeof(AUFS_WH_DR_INFO_PFX) + AUFS_DIRREN_ENV_VAL_SZ];
11474 +       aufs_bindex_t btgt, btail;
11475 +       unsigned char no_sio,
11476 +               allocated,              /* current size of *fdata */
11477 +               infonamelen,            /* room size for p */
11478 +               whnamelen,              /* length of the generated name */
11479 +               renameback;             /* renamed back */
11480 +};
11481 +
11482 +/* on rename(2) error, the caller should revert it using @elm */
11483 +static int au_drinfo_do_store(struct au_drinfo_store *w,
11484 +                             struct au_drinfo_rev_elm *elm)
11485 +{
11486 +       int err, len;
11487 +       ssize_t ssz;
11488 +       loff_t pos;
11489 +       struct path infopath = {
11490 +               .mnt = w->h_ppath.mnt
11491 +       };
11492 +       struct inode *h_dir, *h_inode, *delegated;
11493 +       struct file *infofile;
11494 +       struct qstr *qname;
11495 +
11496 +       AuDebugOn(elm
11497 +                 && memcmp(elm, page_address(ZERO_PAGE(0)), sizeof(*elm)));
11498 +
11499 +       infopath.dentry = vfsub_lookup_one_len(w->whname, &w->h_ppath,
11500 +                                              w->whnamelen);
11501 +       AuTraceErrPtr(infopath.dentry);
11502 +       if (IS_ERR(infopath.dentry)) {
11503 +               err = PTR_ERR(infopath.dentry);
11504 +               goto out;
11505 +       }
11506 +
11507 +       err = 0;
11508 +       h_dir = d_inode(w->h_ppath.dentry);
11509 +       if (elm && d_is_negative(infopath.dentry)) {
11510 +               err = vfsub_create(h_dir, &infopath, 0600, /*want_excl*/true);
11511 +               AuTraceErr(err);
11512 +               if (unlikely(err))
11513 +                       goto out_dput;
11514 +               elm->created = 1;
11515 +               elm->info_dentry = dget(infopath.dentry);
11516 +       }
11517 +
11518 +       infofile = vfsub_dentry_open(&infopath, O_RDWR);
11519 +       AuTraceErrPtr(infofile);
11520 +       if (IS_ERR(infofile)) {
11521 +               err = PTR_ERR(infofile);
11522 +               goto out_dput;
11523 +       }
11524 +
11525 +       h_inode = d_inode(infopath.dentry);
11526 +       if (elm && i_size_read(h_inode)) {
11527 +               h_inode = d_inode(w->h_dentry);
11528 +               elm->info_last = au_drinfo_read_k(infofile, h_inode->i_ino);
11529 +               AuTraceErrPtr(elm->info_last);
11530 +               if (IS_ERR(elm->info_last)) {
11531 +                       err = PTR_ERR(elm->info_last);
11532 +                       elm->info_last = NULL;
11533 +                       AuDebugOn(elm->info_dentry);
11534 +                       goto out_fput;
11535 +               }
11536 +       }
11537 +
11538 +       if (elm && w->renameback) {
11539 +               delegated = NULL;
11540 +               err = vfsub_unlink(h_dir, &infopath, &delegated, /*force*/0);
11541 +               AuTraceErr(err);
11542 +               if (unlikely(err == -EWOULDBLOCK))
11543 +                       iput(delegated);
11544 +               goto out_fput;
11545 +       }
11546 +
11547 +       pos = 0;
11548 +       qname = &w->h_dentry->d_name;
11549 +       len = sizeof(*w->fdata) + qname->len;
11550 +       if (!elm)
11551 +               len = sizeof(*w->fdata) + w->fdata->drinfo.oldnamelen;
11552 +       ssz = vfsub_write_k(infofile, w->fdata, len, &pos);
11553 +       if (ssz == len) {
11554 +               AuDbg("hi%llu, %.*s\n", w->fdata->drinfo.ino,
11555 +                     w->fdata->drinfo.oldnamelen, w->fdata->drinfo.oldname);
11556 +               goto out_fput; /* success */
11557 +       } else {
11558 +               err = -EIO;
11559 +               if (ssz < 0)
11560 +                       err = ssz;
11561 +               /* the caller should revert it using @elm */
11562 +       }
11563 +
11564 +out_fput:
11565 +       fput(infofile);
11566 +out_dput:
11567 +       dput(infopath.dentry);
11568 +out:
11569 +       AuTraceErr(err);
11570 +       return err;
11571 +}
11572 +
11573 +struct au_call_drinfo_do_store_args {
11574 +       int *errp;
11575 +       struct au_drinfo_store *w;
11576 +       struct au_drinfo_rev_elm *elm;
11577 +};
11578 +
11579 +static void au_call_drinfo_do_store(void *args)
11580 +{
11581 +       struct au_call_drinfo_do_store_args *a = args;
11582 +
11583 +       *a->errp = au_drinfo_do_store(a->w, a->elm);
11584 +}
11585 +
11586 +static int au_drinfo_store_sio(struct au_drinfo_store *w,
11587 +                              struct au_drinfo_rev_elm *elm)
11588 +{
11589 +       int err, wkq_err;
11590 +
11591 +       if (w->no_sio)
11592 +               err = au_drinfo_do_store(w, elm);
11593 +       else {
11594 +               struct au_call_drinfo_do_store_args a = {
11595 +                       .errp   = &err,
11596 +                       .w      = w,
11597 +                       .elm    = elm
11598 +               };
11599 +               wkq_err = au_wkq_wait(au_call_drinfo_do_store, &a);
11600 +               if (unlikely(wkq_err))
11601 +                       err = wkq_err;
11602 +       }
11603 +       AuTraceErr(err);
11604 +
11605 +       return err;
11606 +}
11607 +
11608 +static int au_drinfo_store_work_init(struct au_drinfo_store *w,
11609 +                                    aufs_bindex_t btgt)
11610 +{
11611 +       int err;
11612 +
11613 +       memset(w, 0, sizeof(*w));
11614 +       w->allocated = roundup_pow_of_two(sizeof(*w->fdata) + 40);
11615 +       strcpy(w->whname, AUFS_WH_DR_INFO_PFX);
11616 +       w->infoname = w->whname + sizeof(AUFS_WH_DR_INFO_PFX) - 1;
11617 +       w->infonamelen = sizeof(w->whname) - sizeof(AUFS_WH_DR_INFO_PFX);
11618 +       w->btgt = btgt;
11619 +       w->no_sio = !!uid_eq(current_fsuid(), GLOBAL_ROOT_UID);
11620 +
11621 +       err = -ENOMEM;
11622 +       w->fdata = kcalloc(1, w->allocated, GFP_NOFS);
11623 +       if (unlikely(!w->fdata)) {
11624 +               AuTraceErr(err);
11625 +               goto out;
11626 +       }
11627 +       w->fdata->magic = (__force uint32_t)htonl(AUFS_DRINFO_MAGIC_V1);
11628 +       err = 0;
11629 +
11630 +out:
11631 +       return err;
11632 +}
11633 +
11634 +static void au_drinfo_store_work_fin(struct au_drinfo_store *w)
11635 +{
11636 +       au_kfree_rcu(w->fdata);
11637 +}
11638 +
11639 +static void au_drinfo_store_rev(struct au_drinfo_rev *rev,
11640 +                               struct au_drinfo_store *w)
11641 +{
11642 +       struct au_drinfo_rev_elm *elm;
11643 +       struct inode *h_dir, *delegated;
11644 +       int err, nelm;
11645 +       struct path infopath = {
11646 +               .mnt = w->h_ppath.mnt
11647 +       };
11648 +
11649 +       h_dir = d_inode(w->h_ppath.dentry);
11650 +       IMustLock(h_dir);
11651 +
11652 +       err = 0;
11653 +       elm = rev->elm;
11654 +       for (nelm = rev->nelm; nelm > 0; nelm--, elm++) {
11655 +               AuDebugOn(elm->created && elm->info_last);
11656 +               if (elm->created) {
11657 +                       AuDbg("here\n");
11658 +                       delegated = NULL;
11659 +                       infopath.dentry = elm->info_dentry;
11660 +                       err = vfsub_unlink(h_dir, &infopath, &delegated,
11661 +                                          !w->no_sio);
11662 +                       AuTraceErr(err);
11663 +                       if (unlikely(err == -EWOULDBLOCK))
11664 +                               iput(delegated);
11665 +                       dput(elm->info_dentry);
11666 +               } else if (elm->info_last) {
11667 +                       AuDbg("here\n");
11668 +                       w->fdata->drinfo = *elm->info_last;
11669 +                       memcpy(w->fdata->drinfo.oldname,
11670 +                              elm->info_last->oldname,
11671 +                              elm->info_last->oldnamelen);
11672 +                       err = au_drinfo_store_sio(w, /*elm*/NULL);
11673 +                       au_kfree_rcu(elm->info_last);
11674 +               }
11675 +               if (unlikely(err))
11676 +                       AuIOErr("%d, %s\n", err, w->whname);
11677 +               /* go on even if err */
11678 +       }
11679 +}
11680 +
11681 +/* caller has to call au_dr_rename_fin() later */
11682 +static int au_drinfo_store(struct dentry *dentry, aufs_bindex_t btgt,
11683 +                          struct qstr *dst_name, void *_rev)
11684 +{
11685 +       int err, sz, nelm;
11686 +       aufs_bindex_t bindex, btail;
11687 +       struct au_drinfo_store work;
11688 +       struct au_drinfo_rev *rev, **p;
11689 +       struct au_drinfo_rev_elm *elm;
11690 +       struct super_block *sb;
11691 +       struct au_branch *br;
11692 +       struct au_hinode *hdir;
11693 +
11694 +       err = au_drinfo_store_work_init(&work, btgt);
11695 +       AuTraceErr(err);
11696 +       if (unlikely(err))
11697 +               goto out;
11698 +
11699 +       err = -ENOMEM;
11700 +       btail = au_dbtaildir(dentry);
11701 +       nelm = btail - btgt;
11702 +       sz = sizeof(*rev) + sizeof(*elm) * nelm;
11703 +       rev = kcalloc(1, sz, GFP_NOFS);
11704 +       if (unlikely(!rev)) {
11705 +               AuTraceErr(err);
11706 +               goto out_args;
11707 +       }
11708 +       rev->nelm = nelm;
11709 +       elm = rev->elm;
11710 +       p = _rev;
11711 +       *p = rev;
11712 +
11713 +       err = 0;
11714 +       sb = dentry->d_sb;
11715 +       work.h_ppath.dentry = au_h_dptr(dentry, btgt);
11716 +       work.h_ppath.mnt = au_sbr_mnt(sb, btgt);
11717 +       hdir = au_hi(d_inode(dentry), btgt);
11718 +       au_hn_inode_lock_nested(hdir, AuLsc_I_CHILD);
11719 +       for (bindex = btgt + 1; bindex <= btail; bindex++, elm++) {
11720 +               work.h_dentry = au_h_dptr(dentry, bindex);
11721 +               if (!work.h_dentry)
11722 +                       continue;
11723 +
11724 +               err = au_drinfo_construct(&work.fdata, work.h_dentry,
11725 +                                         &work.allocated);
11726 +               AuTraceErr(err);
11727 +               if (unlikely(err))
11728 +                       break;
11729 +
11730 +               work.renameback = au_qstreq(&work.h_dentry->d_name, dst_name);
11731 +               br = au_sbr(sb, bindex);
11732 +               work.whnamelen = sizeof(AUFS_WH_DR_INFO_PFX) - 1;
11733 +               work.whnamelen += au_drinfo_name(br, work.infoname,
11734 +                                                work.infonamelen);
11735 +               AuDbg("whname %.*s, i%llu, %.*s\n",
11736 +                     work.whnamelen, work.whname,
11737 +                     be64_to_cpu((__force __be64)work.fdata->drinfo.ino),
11738 +                     work.fdata->drinfo.oldnamelen,
11739 +                     work.fdata->drinfo.oldname);
11740 +
11741 +               err = au_drinfo_store_sio(&work, elm);
11742 +               AuTraceErr(err);
11743 +               if (unlikely(err))
11744 +                       break;
11745 +       }
11746 +       if (unlikely(err)) {
11747 +               /* revert all drinfo */
11748 +               au_drinfo_store_rev(rev, &work);
11749 +               au_kfree_try_rcu(rev);
11750 +               *p = NULL;
11751 +       }
11752 +       au_hn_inode_unlock(hdir);
11753 +
11754 +out_args:
11755 +       au_drinfo_store_work_fin(&work);
11756 +out:
11757 +       return err;
11758 +}
11759 +
11760 +/* ---------------------------------------------------------------------- */
11761 +
11762 +int au_dr_rename(struct dentry *src, aufs_bindex_t bindex,
11763 +                struct qstr *dst_name, void *_rev)
11764 +{
11765 +       int err, already;
11766 +       ino_t ino;
11767 +       struct super_block *sb;
11768 +       struct au_branch *br;
11769 +       struct au_dr_br *dr;
11770 +       struct dentry *h_dentry;
11771 +       struct inode *h_inode;
11772 +       struct au_dr_hino *ent;
11773 +       struct au_drinfo_rev *rev, **p;
11774 +
11775 +       AuDbg("bindex %d\n", bindex);
11776 +
11777 +       err = -ENOMEM;
11778 +       ent = kmalloc(sizeof(*ent), GFP_NOFS);
11779 +       if (unlikely(!ent))
11780 +               goto out;
11781 +
11782 +       sb = src->d_sb;
11783 +       br = au_sbr(sb, bindex);
11784 +       dr = &br->br_dirren;
11785 +       h_dentry = au_h_dptr(src, bindex);
11786 +       h_inode = d_inode(h_dentry);
11787 +       ino = h_inode->i_ino;
11788 +       ent->dr_h_ino = ino;
11789 +       already = au_dr_hino_test_add(dr, ino, ent);
11790 +       AuDbg("b%d, hi%llu, already %d\n",
11791 +             bindex, (unsigned long long)ino, already);
11792 +
11793 +       err = au_drinfo_store(src, bindex, dst_name, _rev);
11794 +       AuTraceErr(err);
11795 +       if (!err) {
11796 +               p = _rev;
11797 +               rev = *p;
11798 +               rev->already = already;
11799 +               goto out; /* success */
11800 +       }
11801 +
11802 +       /* revert */
11803 +       if (!already)
11804 +               au_dr_hino_del(dr, ent);
11805 +       au_kfree_rcu(ent);
11806 +
11807 +out:
11808 +       AuTraceErr(err);
11809 +       return err;
11810 +}
11811 +
11812 +void au_dr_rename_fin(struct dentry *src, aufs_bindex_t btgt, void *_rev)
11813 +{
11814 +       struct au_drinfo_rev *rev;
11815 +       struct au_drinfo_rev_elm *elm;
11816 +       int nelm;
11817 +
11818 +       rev = _rev;
11819 +       elm = rev->elm;
11820 +       for (nelm = rev->nelm; nelm > 0; nelm--, elm++) {
11821 +               dput(elm->info_dentry);
11822 +               au_kfree_rcu(elm->info_last);
11823 +       }
11824 +       au_kfree_try_rcu(rev);
11825 +}
11826 +
11827 +void au_dr_rename_rev(struct dentry *src, aufs_bindex_t btgt, void *_rev)
11828 +{
11829 +       int err;
11830 +       struct au_drinfo_store work;
11831 +       struct au_drinfo_rev *rev = _rev;
11832 +       struct super_block *sb;
11833 +       struct au_branch *br;
11834 +       struct inode *h_inode;
11835 +       struct au_dr_br *dr;
11836 +       struct au_dr_hino *ent;
11837 +
11838 +       err = au_drinfo_store_work_init(&work, btgt);
11839 +       if (unlikely(err))
11840 +               goto out;
11841 +
11842 +       sb = src->d_sb;
11843 +       br = au_sbr(sb, btgt);
11844 +       work.h_ppath.dentry = au_h_dptr(src, btgt);
11845 +       work.h_ppath.mnt = au_br_mnt(br);
11846 +       au_drinfo_store_rev(rev, &work);
11847 +       au_drinfo_store_work_fin(&work);
11848 +       if (rev->already)
11849 +               goto out;
11850 +
11851 +       dr = &br->br_dirren;
11852 +       h_inode = d_inode(work.h_ppath.dentry);
11853 +       ent = au_dr_hino_find(dr, h_inode->i_ino);
11854 +       BUG_ON(!ent);
11855 +       au_dr_hino_del(dr, ent);
11856 +       au_kfree_rcu(ent);
11857 +
11858 +out:
11859 +       au_kfree_try_rcu(rev);
11860 +       if (unlikely(err))
11861 +               pr_err("failed to remove dirren info\n");
11862 +}
11863 +
11864 +/* ---------------------------------------------------------------------- */
11865 +
11866 +static struct au_drinfo *au_drinfo_do_load(struct path *h_ppath,
11867 +                                          char *whname, int whnamelen,
11868 +                                          struct dentry **info_dentry)
11869 +{
11870 +       struct au_drinfo *drinfo;
11871 +       struct file *f;
11872 +       struct inode *h_dir;
11873 +       struct path infopath;
11874 +       int unlocked;
11875 +
11876 +       AuDbg("%pd/%.*s\n", h_ppath->dentry, whnamelen, whname);
11877 +
11878 +       *info_dentry = NULL;
11879 +       drinfo = NULL;
11880 +       unlocked = 0;
11881 +       h_dir = d_inode(h_ppath->dentry);
11882 +       inode_lock_shared_nested(h_dir, AuLsc_I_PARENT);
11883 +       infopath.dentry = vfsub_lookup_one_len(whname, h_ppath, whnamelen);
11884 +       if (IS_ERR(infopath.dentry)) {
11885 +               drinfo = (void *)infopath.dentry;
11886 +               goto out;
11887 +       }
11888 +
11889 +       if (d_is_negative(infopath.dentry))
11890 +               goto out_dput; /* success */
11891 +
11892 +       infopath.mnt = h_ppath->mnt;
11893 +       f = vfsub_dentry_open(&infopath, O_RDONLY);
11894 +       inode_unlock_shared(h_dir);
11895 +       unlocked = 1;
11896 +       if (IS_ERR(f)) {
11897 +               drinfo = (void *)f;
11898 +               goto out_dput;
11899 +       }
11900 +
11901 +       drinfo = au_drinfo_read_k(f, /*h_ino*/0);
11902 +       if (IS_ERR_OR_NULL(drinfo))
11903 +               goto out_fput;
11904 +
11905 +       AuDbg("oldname %.*s\n", drinfo->oldnamelen, drinfo->oldname);
11906 +       *info_dentry = dget(infopath.dentry); /* keep it alive */
11907 +
11908 +out_fput:
11909 +       fput(f);
11910 +out_dput:
11911 +       dput(infopath.dentry);
11912 +out:
11913 +       if (!unlocked)
11914 +               inode_unlock_shared(h_dir);
11915 +       AuTraceErrPtr(drinfo);
11916 +       return drinfo;
11917 +}
11918 +
11919 +struct au_drinfo_do_load_args {
11920 +       struct au_drinfo **drinfop;
11921 +       struct path *h_ppath;
11922 +       char *whname;
11923 +       int whnamelen;
11924 +       struct dentry **info_dentry;
11925 +};
11926 +
11927 +static void au_call_drinfo_do_load(void *args)
11928 +{
11929 +       struct au_drinfo_do_load_args *a = args;
11930 +
11931 +       *a->drinfop = au_drinfo_do_load(a->h_ppath, a->whname, a->whnamelen,
11932 +                                       a->info_dentry);
11933 +}
11934 +
11935 +struct au_drinfo_load {
11936 +       struct path h_ppath;
11937 +       struct qstr *qname;
11938 +       unsigned char no_sio;
11939 +
11940 +       aufs_bindex_t ninfo;
11941 +       struct au_drinfo **drinfo;
11942 +};
11943 +
11944 +static int au_drinfo_load(struct au_drinfo_load *w, aufs_bindex_t bindex,
11945 +                         struct au_branch *br)
11946 +{
11947 +       int err, wkq_err, whnamelen, e;
11948 +       char whname[sizeof(AUFS_WH_DR_INFO_PFX) + AUFS_DIRREN_ENV_VAL_SZ]
11949 +               = AUFS_WH_DR_INFO_PFX;
11950 +       struct au_drinfo *drinfo;
11951 +       struct qstr oldname;
11952 +       struct inode *h_dir, *delegated;
11953 +       struct dentry *info_dentry;
11954 +       struct path infopath;
11955 +
11956 +       whnamelen = sizeof(AUFS_WH_DR_INFO_PFX) - 1;
11957 +       whnamelen += au_drinfo_name(br, whname + whnamelen,
11958 +                                   sizeof(whname) - whnamelen);
11959 +       if (w->no_sio)
11960 +               drinfo = au_drinfo_do_load(&w->h_ppath, whname, whnamelen,
11961 +                                          &info_dentry);
11962 +       else {
11963 +               struct au_drinfo_do_load_args args = {
11964 +                       .drinfop        = &drinfo,
11965 +                       .h_ppath        = &w->h_ppath,
11966 +                       .whname         = whname,
11967 +                       .whnamelen      = whnamelen,
11968 +                       .info_dentry    = &info_dentry
11969 +               };
11970 +               wkq_err = au_wkq_wait(au_call_drinfo_do_load, &args);
11971 +               if (unlikely(wkq_err))
11972 +                       drinfo = ERR_PTR(wkq_err);
11973 +       }
11974 +       err = PTR_ERR(drinfo);
11975 +       if (IS_ERR_OR_NULL(drinfo))
11976 +               goto out;
11977 +
11978 +       err = 0;
11979 +       oldname.len = drinfo->oldnamelen;
11980 +       oldname.name = drinfo->oldname;
11981 +       if (au_qstreq(w->qname, &oldname)) {
11982 +               /* the name is renamed back */
11983 +               au_kfree_rcu(drinfo);
11984 +               drinfo = NULL;
11985 +
11986 +               infopath.dentry = info_dentry;
11987 +               infopath.mnt = w->h_ppath.mnt;
11988 +               h_dir = d_inode(w->h_ppath.dentry);
11989 +               delegated = NULL;
11990 +               inode_lock_nested(h_dir, AuLsc_I_PARENT);
11991 +               e = vfsub_unlink(h_dir, &infopath, &delegated, !w->no_sio);
11992 +               inode_unlock(h_dir);
11993 +               if (unlikely(e))
11994 +                       AuIOErr("ignored %d, %pd2\n", e, &infopath.dentry);
11995 +               if (unlikely(e == -EWOULDBLOCK))
11996 +                       iput(delegated);
11997 +       }
11998 +       au_kfree_rcu(w->drinfo[bindex]);
11999 +       w->drinfo[bindex] = drinfo;
12000 +       dput(info_dentry);
12001 +
12002 +out:
12003 +       AuTraceErr(err);
12004 +       return err;
12005 +}
12006 +
12007 +/* ---------------------------------------------------------------------- */
12008 +
12009 +static void au_dr_lkup_free(struct au_drinfo **drinfo, int n)
12010 +{
12011 +       struct au_drinfo **p = drinfo;
12012 +
12013 +       while (n-- > 0)
12014 +               au_kfree_rcu(*drinfo++);
12015 +       au_kfree_try_rcu(p);
12016 +}
12017 +
12018 +int au_dr_lkup(struct au_do_lookup_args *lkup, struct dentry *dentry,
12019 +              aufs_bindex_t btgt)
12020 +{
12021 +       int err, ninfo;
12022 +       struct au_drinfo_load w;
12023 +       aufs_bindex_t bindex, bbot;
12024 +       struct au_branch *br;
12025 +       struct inode *h_dir;
12026 +       struct au_dr_hino *ent;
12027 +       struct super_block *sb;
12028 +
12029 +       AuDbg("%.*s, name %.*s, whname %.*s, b%d\n",
12030 +             AuLNPair(&dentry->d_name), AuLNPair(&lkup->dirren.dr_name),
12031 +             AuLNPair(&lkup->whname), btgt);
12032 +
12033 +       sb = dentry->d_sb;
12034 +       bbot = au_sbbot(sb);
12035 +       w.ninfo = bbot + 1;
12036 +       if (!lkup->dirren.drinfo) {
12037 +               lkup->dirren.drinfo = kcalloc(w.ninfo,
12038 +                                             sizeof(*lkup->dirren.drinfo),
12039 +                                             GFP_NOFS);
12040 +               if (unlikely(!lkup->dirren.drinfo)) {
12041 +                       err = -ENOMEM;
12042 +                       goto out;
12043 +               }
12044 +               lkup->dirren.ninfo = w.ninfo;
12045 +       }
12046 +       w.drinfo = lkup->dirren.drinfo;
12047 +       w.no_sio = !!uid_eq(current_fsuid(), GLOBAL_ROOT_UID);
12048 +       w.h_ppath.dentry = au_h_dptr(dentry, btgt);
12049 +       AuDebugOn(!w.h_ppath.dentry);
12050 +       w.h_ppath.mnt = au_sbr_mnt(sb, btgt);
12051 +       w.qname = &dentry->d_name;
12052 +
12053 +       ninfo = 0;
12054 +       for (bindex = btgt + 1; bindex <= bbot; bindex++) {
12055 +               br = au_sbr(sb, bindex);
12056 +               err = au_drinfo_load(&w, bindex, br);
12057 +               if (unlikely(err))
12058 +                       goto out_free;
12059 +               if (w.drinfo[bindex])
12060 +                       ninfo++;
12061 +       }
12062 +       if (!ninfo) {
12063 +               br = au_sbr(sb, btgt);
12064 +               h_dir = d_inode(w.h_ppath.dentry);
12065 +               ent = au_dr_hino_find(&br->br_dirren, h_dir->i_ino);
12066 +               AuDebugOn(!ent);
12067 +               au_dr_hino_del(&br->br_dirren, ent);
12068 +               au_kfree_rcu(ent);
12069 +       }
12070 +       goto out; /* success */
12071 +
12072 +out_free:
12073 +       au_dr_lkup_free(lkup->dirren.drinfo, lkup->dirren.ninfo);
12074 +       lkup->dirren.ninfo = 0;
12075 +       lkup->dirren.drinfo = NULL;
12076 +out:
12077 +       AuTraceErr(err);
12078 +       return err;
12079 +}
12080 +
12081 +void au_dr_lkup_fin(struct au_do_lookup_args *lkup)
12082 +{
12083 +       au_dr_lkup_free(lkup->dirren.drinfo, lkup->dirren.ninfo);
12084 +}
12085 +
12086 +int au_dr_lkup_name(struct au_do_lookup_args *lkup, aufs_bindex_t btgt)
12087 +{
12088 +       int err;
12089 +       struct au_drinfo *drinfo;
12090 +
12091 +       err = 0;
12092 +       if (!lkup->dirren.drinfo)
12093 +               goto out;
12094 +       AuDebugOn(lkup->dirren.ninfo <= btgt);
12095 +       drinfo = lkup->dirren.drinfo[btgt];
12096 +       if (!drinfo)
12097 +               goto out;
12098 +
12099 +       au_kfree_try_rcu(lkup->whname.name);
12100 +       lkup->whname.name = NULL;
12101 +       lkup->dirren.dr_name.len = drinfo->oldnamelen;
12102 +       lkup->dirren.dr_name.name = drinfo->oldname;
12103 +       lkup->name = &lkup->dirren.dr_name;
12104 +       err = au_wh_name_alloc(&lkup->whname, lkup->name);
12105 +       if (!err)
12106 +               AuDbg("name %.*s, whname %.*s, b%d\n",
12107 +                     AuLNPair(lkup->name), AuLNPair(&lkup->whname),
12108 +                     btgt);
12109 +
12110 +out:
12111 +       AuTraceErr(err);
12112 +       return err;
12113 +}
12114 +
12115 +int au_dr_lkup_h_ino(struct au_do_lookup_args *lkup, aufs_bindex_t bindex,
12116 +                    ino_t h_ino)
12117 +{
12118 +       int match;
12119 +       struct au_drinfo *drinfo;
12120 +
12121 +       match = 1;
12122 +       if (!lkup->dirren.drinfo)
12123 +               goto out;
12124 +       AuDebugOn(lkup->dirren.ninfo <= bindex);
12125 +       drinfo = lkup->dirren.drinfo[bindex];
12126 +       if (!drinfo)
12127 +               goto out;
12128 +
12129 +       match = (drinfo->ino == h_ino);
12130 +       AuDbg("match %d\n", match);
12131 +
12132 +out:
12133 +       return match;
12134 +}
12135 +
12136 +/* ---------------------------------------------------------------------- */
12137 +
12138 +int au_dr_opt_set(struct super_block *sb)
12139 +{
12140 +       int err;
12141 +       aufs_bindex_t bindex, bbot;
12142 +       struct au_branch *br;
12143 +
12144 +       err = 0;
12145 +       bbot = au_sbbot(sb);
12146 +       for (bindex = 0; !err && bindex <= bbot; bindex++) {
12147 +               br = au_sbr(sb, bindex);
12148 +               err = au_dr_hino(sb, bindex, /*br*/NULL, &br->br_path);
12149 +       }
12150 +
12151 +       return err;
12152 +}
12153 +
12154 +int au_dr_opt_flush(struct super_block *sb)
12155 +{
12156 +       int err;
12157 +       aufs_bindex_t bindex, bbot;
12158 +       struct au_branch *br;
12159 +
12160 +       err = 0;
12161 +       bbot = au_sbbot(sb);
12162 +       for (bindex = 0; !err && bindex <= bbot; bindex++) {
12163 +               br = au_sbr(sb, bindex);
12164 +               if (au_br_writable(br->br_perm))
12165 +                       err = au_dr_hino(sb, bindex, /*br*/NULL, /*path*/NULL);
12166 +       }
12167 +
12168 +       return err;
12169 +}
12170 +
12171 +int au_dr_opt_clr(struct super_block *sb, int no_flush)
12172 +{
12173 +       int err;
12174 +       aufs_bindex_t bindex, bbot;
12175 +       struct au_branch *br;
12176 +
12177 +       err = 0;
12178 +       if (!no_flush) {
12179 +               err = au_dr_opt_flush(sb);
12180 +               if (unlikely(err))
12181 +                       goto out;
12182 +       }
12183 +
12184 +       bbot = au_sbbot(sb);
12185 +       for (bindex = 0; bindex <= bbot; bindex++) {
12186 +               br = au_sbr(sb, bindex);
12187 +               au_dr_hino_free(&br->br_dirren);
12188 +       }
12189 +
12190 +out:
12191 +       return err;
12192 +}
12193 diff -urN /usr/share/empty/fs/aufs/dirren.h linux/fs/aufs/dirren.h
12194 --- /usr/share/empty/fs/aufs/dirren.h   1970-01-01 01:00:00.000000000 +0100
12195 +++ linux/fs/aufs/dirren.h      2021-12-03 15:38:59.933313976 +0100
12196 @@ -0,0 +1,140 @@
12197 +/* SPDX-License-Identifier: GPL-2.0 */
12198 +/*
12199 + * Copyright (C) 2017-2021 Junjiro R. Okajima
12200 + *
12201 + * This program, aufs is free software; you can redistribute it and/or modify
12202 + * it under the terms of the GNU General Public License as published by
12203 + * the Free Software Foundation; either version 2 of the License, or
12204 + * (at your option) any later version.
12205 + *
12206 + * This program is distributed in the hope that it will be useful,
12207 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12208 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12209 + * GNU General Public License for more details.
12210 + *
12211 + * You should have received a copy of the GNU General Public License
12212 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12213 + */
12214 +
12215 +/*
12216 + * renamed dir info
12217 + */
12218 +
12219 +#ifndef __AUFS_DIRREN_H__
12220 +#define __AUFS_DIRREN_H__
12221 +
12222 +#ifdef __KERNEL__
12223 +
12224 +#include <linux/dcache.h>
12225 +#include <linux/statfs.h>
12226 +#include <linux/uuid.h>
12227 +#include "hbl.h"
12228 +
12229 +#define AuDirren_NHASH 100
12230 +
12231 +#ifdef CONFIG_AUFS_DIRREN
12232 +enum au_brid_type {
12233 +       AuBrid_Unset,
12234 +       AuBrid_UUID,
12235 +       AuBrid_FSID,
12236 +       AuBrid_DEV
12237 +};
12238 +
12239 +struct au_dr_brid {
12240 +       enum au_brid_type       type;
12241 +       union {
12242 +               uuid_t  uuid;   /* unimplemented yet */
12243 +               fsid_t  fsid;
12244 +               dev_t   dev;
12245 +       };
12246 +};
12247 +
12248 +/* 20 is the max digits length of ulong 64 */
12249 +/* brid-type "_" uuid "_" inum */
12250 +#define AUFS_DIRREN_FNAME_SZ   (1 + 1 + UUID_STRING_LEN + 20)
12251 +#define AUFS_DIRREN_ENV_VAL_SZ (AUFS_DIRREN_FNAME_SZ + 1 + 20)
12252 +
12253 +struct au_dr_hino {
12254 +       struct hlist_bl_node    dr_hnode;
12255 +       ino_t                   dr_h_ino;
12256 +};
12257 +
12258 +struct au_dr_br {
12259 +       struct hlist_bl_head    dr_h_ino[AuDirren_NHASH];
12260 +       struct au_dr_brid       dr_brid;
12261 +};
12262 +
12263 +struct au_dr_lookup {
12264 +       /* dr_name is pointed by struct au_do_lookup_args.name */
12265 +       struct qstr             dr_name; /* subset of dr_info */
12266 +       aufs_bindex_t           ninfo;
12267 +       struct au_drinfo        **drinfo;
12268 +};
12269 +#else
12270 +struct au_dr_hino;
12271 +/* empty */
12272 +struct au_dr_br { };
12273 +struct au_dr_lookup { };
12274 +#endif
12275 +
12276 +/* ---------------------------------------------------------------------- */
12277 +
12278 +struct au_branch;
12279 +struct au_do_lookup_args;
12280 +struct au_hinode;
12281 +#ifdef CONFIG_AUFS_DIRREN
12282 +int au_dr_hino_test_add(struct au_dr_br *dr, ino_t h_ino,
12283 +                       struct au_dr_hino *add_ent);
12284 +void au_dr_hino_free(struct au_dr_br *dr);
12285 +int au_dr_br_init(struct super_block *sb, struct au_branch *br,
12286 +                 const struct path *path);
12287 +int au_dr_br_fin(struct super_block *sb, struct au_branch *br);
12288 +int au_dr_rename(struct dentry *src, aufs_bindex_t bindex,
12289 +                struct qstr *dst_name, void *_rev);
12290 +void au_dr_rename_fin(struct dentry *src, aufs_bindex_t btgt, void *rev);
12291 +void au_dr_rename_rev(struct dentry *src, aufs_bindex_t bindex, void *rev);
12292 +int au_dr_lkup(struct au_do_lookup_args *lkup, struct dentry *dentry,
12293 +              aufs_bindex_t bindex);
12294 +int au_dr_lkup_name(struct au_do_lookup_args *lkup, aufs_bindex_t btgt);
12295 +int au_dr_lkup_h_ino(struct au_do_lookup_args *lkup, aufs_bindex_t bindex,
12296 +                    ino_t h_ino);
12297 +void au_dr_lkup_fin(struct au_do_lookup_args *lkup);
12298 +int au_dr_opt_set(struct super_block *sb);
12299 +int au_dr_opt_flush(struct super_block *sb);
12300 +int au_dr_opt_clr(struct super_block *sb, int no_flush);
12301 +#else
12302 +AuStubInt0(au_dr_hino_test_add, struct au_dr_br *dr, ino_t h_ino,
12303 +          struct au_dr_hino *add_ent);
12304 +AuStubVoid(au_dr_hino_free, struct au_dr_br *dr);
12305 +AuStubInt0(au_dr_br_init, struct super_block *sb, struct au_branch *br,
12306 +          const struct path *path);
12307 +AuStubInt0(au_dr_br_fin, struct super_block *sb, struct au_branch *br);
12308 +AuStubInt0(au_dr_rename, struct dentry *src, aufs_bindex_t bindex,
12309 +          struct qstr *dst_name, void *_rev);
12310 +AuStubVoid(au_dr_rename_fin, struct dentry *src, aufs_bindex_t btgt, void *rev);
12311 +AuStubVoid(au_dr_rename_rev, struct dentry *src, aufs_bindex_t bindex,
12312 +          void *rev);
12313 +AuStubInt0(au_dr_lkup, struct au_do_lookup_args *lkup, struct dentry *dentry,
12314 +          aufs_bindex_t bindex);
12315 +AuStubInt0(au_dr_lkup_name, struct au_do_lookup_args *lkup, aufs_bindex_t btgt);
12316 +AuStubInt0(au_dr_lkup_h_ino, struct au_do_lookup_args *lkup,
12317 +          aufs_bindex_t bindex, ino_t h_ino);
12318 +AuStubVoid(au_dr_lkup_fin, struct au_do_lookup_args *lkup);
12319 +AuStubInt0(au_dr_opt_set, struct super_block *sb);
12320 +AuStubInt0(au_dr_opt_flush, struct super_block *sb);
12321 +AuStubInt0(au_dr_opt_clr, struct super_block *sb, int no_flush);
12322 +#endif
12323 +
12324 +/* ---------------------------------------------------------------------- */
12325 +
12326 +#ifdef CONFIG_AUFS_DIRREN
12327 +static inline int au_dr_ihash(ino_t h_ino)
12328 +{
12329 +       return h_ino % AuDirren_NHASH;
12330 +}
12331 +#else
12332 +AuStubInt0(au_dr_ihash, ino_t h_ino);
12333 +#endif
12334 +
12335 +#endif /* __KERNEL__ */
12336 +#endif /* __AUFS_DIRREN_H__ */
12337 diff -urN /usr/share/empty/fs/aufs/dynop.c linux/fs/aufs/dynop.c
12338 --- /usr/share/empty/fs/aufs/dynop.c    1970-01-01 01:00:00.000000000 +0100
12339 +++ linux/fs/aufs/dynop.c       2021-12-03 15:38:59.933313976 +0100
12340 @@ -0,0 +1,368 @@
12341 +// SPDX-License-Identifier: GPL-2.0
12342 +/*
12343 + * Copyright (C) 2010-2021 Junjiro R. Okajima
12344 + *
12345 + * This program, aufs is free software; you can redistribute it and/or modify
12346 + * it under the terms of the GNU General Public License as published by
12347 + * the Free Software Foundation; either version 2 of the License, or
12348 + * (at your option) any later version.
12349 + *
12350 + * This program is distributed in the hope that it will be useful,
12351 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12352 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12353 + * GNU General Public License for more details.
12354 + *
12355 + * You should have received a copy of the GNU General Public License
12356 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12357 + */
12358 +
12359 +/*
12360 + * dynamically customizable operations for regular files
12361 + */
12362 +
12363 +#include "aufs.h"
12364 +
12365 +#define DyPrSym(key)   AuDbgSym(key->dk_op.dy_hop)
12366 +
12367 +/*
12368 + * How large will these lists be?
12369 + * Usually just a few elements, 20-30 at most for each, I guess.
12370 + */
12371 +static struct hlist_bl_head dynop[AuDyLast];
12372 +
12373 +static struct au_dykey *dy_gfind_get(struct hlist_bl_head *hbl,
12374 +                                    const void *h_op)
12375 +{
12376 +       struct au_dykey *key, *tmp;
12377 +       struct hlist_bl_node *pos;
12378 +
12379 +       key = NULL;
12380 +       hlist_bl_lock(hbl);
12381 +       hlist_bl_for_each_entry(tmp, pos, hbl, dk_hnode)
12382 +               if (tmp->dk_op.dy_hop == h_op) {
12383 +                       if (kref_get_unless_zero(&tmp->dk_kref))
12384 +                               key = tmp;
12385 +                       break;
12386 +               }
12387 +       hlist_bl_unlock(hbl);
12388 +
12389 +       return key;
12390 +}
12391 +
12392 +static struct au_dykey *dy_bradd(struct au_branch *br, struct au_dykey *key)
12393 +{
12394 +       struct au_dykey **k, *found;
12395 +       const void *h_op = key->dk_op.dy_hop;
12396 +       int i;
12397 +
12398 +       found = NULL;
12399 +       k = br->br_dykey;
12400 +       for (i = 0; i < AuBrDynOp; i++)
12401 +               if (k[i]) {
12402 +                       if (k[i]->dk_op.dy_hop == h_op) {
12403 +                               found = k[i];
12404 +                               break;
12405 +                       }
12406 +               } else
12407 +                       break;
12408 +       if (!found) {
12409 +               spin_lock(&br->br_dykey_lock);
12410 +               for (; i < AuBrDynOp; i++)
12411 +                       if (k[i]) {
12412 +                               if (k[i]->dk_op.dy_hop == h_op) {
12413 +                                       found = k[i];
12414 +                                       break;
12415 +                               }
12416 +                       } else {
12417 +                               k[i] = key;
12418 +                               break;
12419 +                       }
12420 +               spin_unlock(&br->br_dykey_lock);
12421 +               BUG_ON(i == AuBrDynOp); /* expand the array */
12422 +       }
12423 +
12424 +       return found;
12425 +}
12426 +
12427 +/* kref_get() if @key is already added */
12428 +static struct au_dykey *dy_gadd(struct hlist_bl_head *hbl, struct au_dykey *key)
12429 +{
12430 +       struct au_dykey *tmp, *found;
12431 +       struct hlist_bl_node *pos;
12432 +       const void *h_op = key->dk_op.dy_hop;
12433 +
12434 +       found = NULL;
12435 +       hlist_bl_lock(hbl);
12436 +       hlist_bl_for_each_entry(tmp, pos, hbl, dk_hnode)
12437 +               if (tmp->dk_op.dy_hop == h_op) {
12438 +                       if (kref_get_unless_zero(&tmp->dk_kref))
12439 +                               found = tmp;
12440 +                       break;
12441 +               }
12442 +       if (!found)
12443 +               hlist_bl_add_head(&key->dk_hnode, hbl);
12444 +       hlist_bl_unlock(hbl);
12445 +
12446 +       if (!found)
12447 +               DyPrSym(key);
12448 +       return found;
12449 +}
12450 +
12451 +static void dy_free_rcu(struct rcu_head *rcu)
12452 +{
12453 +       struct au_dykey *key;
12454 +
12455 +       key = container_of(rcu, struct au_dykey, dk_rcu);
12456 +       DyPrSym(key);
12457 +       kfree(key);
12458 +}
12459 +
12460 +static void dy_free(struct kref *kref)
12461 +{
12462 +       struct au_dykey *key;
12463 +       struct hlist_bl_head *hbl;
12464 +
12465 +       key = container_of(kref, struct au_dykey, dk_kref);
12466 +       hbl = dynop + key->dk_op.dy_type;
12467 +       au_hbl_del(&key->dk_hnode, hbl);
12468 +       call_rcu(&key->dk_rcu, dy_free_rcu);
12469 +}
12470 +
12471 +void au_dy_put(struct au_dykey *key)
12472 +{
12473 +       kref_put(&key->dk_kref, dy_free);
12474 +}
12475 +
12476 +/* ---------------------------------------------------------------------- */
12477 +
12478 +#define DyDbgSize(cnt, op)     AuDebugOn(cnt != sizeof(op)/sizeof(void *))
12479 +
12480 +#ifdef CONFIG_AUFS_DEBUG
12481 +#define DyDbgDeclare(cnt)      unsigned int cnt = 0
12482 +#define DyDbgInc(cnt)          do { cnt++; } while (0)
12483 +#else
12484 +#define DyDbgDeclare(cnt)      do {} while (0)
12485 +#define DyDbgInc(cnt)          do {} while (0)
12486 +#endif
12487 +
12488 +#define DySet(func, dst, src, h_op, h_sb) do {                         \
12489 +       DyDbgInc(cnt);                                                  \
12490 +       if (h_op->func) {                                               \
12491 +               if (src.func)                                           \
12492 +                       dst.func = src.func;                            \
12493 +               else                                                    \
12494 +                       AuDbg("%s %s\n", au_sbtype(h_sb), #func);       \
12495 +       }                                                               \
12496 +} while (0)
12497 +
12498 +#define DySetForce(func, dst, src) do {                \
12499 +       AuDebugOn(!src.func);                   \
12500 +       DyDbgInc(cnt);                          \
12501 +       dst.func = src.func;                    \
12502 +} while (0)
12503 +
12504 +#define DySetAop(func) \
12505 +       DySet(func, dyaop->da_op, aufs_aop, h_aop, h_sb)
12506 +#define DySetAopForce(func) \
12507 +       DySetForce(func, dyaop->da_op, aufs_aop)
12508 +
12509 +static void dy_aop(struct au_dykey *key, const void *h_op,
12510 +                  struct super_block *h_sb __maybe_unused)
12511 +{
12512 +       struct au_dyaop *dyaop = (void *)key;
12513 +       const struct address_space_operations *h_aop = h_op;
12514 +       DyDbgDeclare(cnt);
12515 +
12516 +       AuDbg("%s\n", au_sbtype(h_sb));
12517 +
12518 +       DySetAop(writepage);
12519 +       DySetAopForce(readpage);        /* force */
12520 +       DySetAop(writepages);
12521 +       DySetAop(set_page_dirty);
12522 +       DySetAop(readpages);
12523 +       DySetAop(readahead);
12524 +       DySetAop(write_begin);
12525 +       DySetAop(write_end);
12526 +       DySetAop(bmap);
12527 +       DySetAop(invalidatepage);
12528 +       DySetAop(releasepage);
12529 +       DySetAop(freepage);
12530 +       /* this one will be changed according to an aufs mount option */
12531 +       DySetAop(direct_IO);
12532 +       DySetAop(migratepage);
12533 +       DySetAop(isolate_page);
12534 +       DySetAop(putback_page);
12535 +       DySetAop(launder_page);
12536 +       DySetAop(is_partially_uptodate);
12537 +       DySetAop(is_dirty_writeback);
12538 +       DySetAop(error_remove_page);
12539 +       DySetAop(swap_activate);
12540 +       DySetAop(swap_deactivate);
12541 +
12542 +       DyDbgSize(cnt, *h_aop);
12543 +}
12544 +
12545 +/* ---------------------------------------------------------------------- */
12546 +
12547 +static void dy_bug(struct kref *kref)
12548 +{
12549 +       BUG();
12550 +}
12551 +
12552 +static struct au_dykey *dy_get(struct au_dynop *op, struct au_branch *br)
12553 +{
12554 +       struct au_dykey *key, *old;
12555 +       struct hlist_bl_head *hbl;
12556 +       struct op {
12557 +               unsigned int sz;
12558 +               void (*set)(struct au_dykey *key, const void *h_op,
12559 +                           struct super_block *h_sb __maybe_unused);
12560 +       };
12561 +       static const struct op a[] = {
12562 +               [AuDy_AOP] = {
12563 +                       .sz     = sizeof(struct au_dyaop),
12564 +                       .set    = dy_aop
12565 +               }
12566 +       };
12567 +       const struct op *p;
12568 +
12569 +       hbl = dynop + op->dy_type;
12570 +       key = dy_gfind_get(hbl, op->dy_hop);
12571 +       if (key)
12572 +               goto out_add; /* success */
12573 +
12574 +       p = a + op->dy_type;
12575 +       key = kzalloc(p->sz, GFP_NOFS);
12576 +       if (unlikely(!key)) {
12577 +               key = ERR_PTR(-ENOMEM);
12578 +               goto out;
12579 +       }
12580 +
12581 +       key->dk_op.dy_hop = op->dy_hop;
12582 +       kref_init(&key->dk_kref);
12583 +       p->set(key, op->dy_hop, au_br_sb(br));
12584 +       old = dy_gadd(hbl, key);
12585 +       if (old) {
12586 +               au_kfree_rcu(key);
12587 +               key = old;
12588 +       }
12589 +
12590 +out_add:
12591 +       old = dy_bradd(br, key);
12592 +       if (old)
12593 +               /* its ref-count should never be zero here */
12594 +               kref_put(&key->dk_kref, dy_bug);
12595 +out:
12596 +       return key;
12597 +}
12598 +
12599 +/* ---------------------------------------------------------------------- */
12600 +/*
12601 + * Aufs prohibits O_DIRECT by default even if the branch supports it.
12602 + * This behaviour is necessary to return an error from open(O_DIRECT) instead
12603 + * of the succeeding I/O. The dio mount option enables O_DIRECT and makes
12604 + * open(O_DIRECT) always succeed, but the succeeding I/O may return an error.
12605 + * See the aufs manual in detail.
12606 + */
12607 +static void dy_adx(struct au_dyaop *dyaop, int do_dx)
12608 +{
12609 +       if (!do_dx)
12610 +               dyaop->da_op.direct_IO = NULL;
12611 +       else
12612 +               dyaop->da_op.direct_IO = aufs_aop.direct_IO;
12613 +}
12614 +
12615 +static struct au_dyaop *dy_aget(struct au_branch *br,
12616 +                               const struct address_space_operations *h_aop,
12617 +                               int do_dx)
12618 +{
12619 +       struct au_dyaop *dyaop;
12620 +       struct au_dynop op;
12621 +
12622 +       op.dy_type = AuDy_AOP;
12623 +       op.dy_haop = h_aop;
12624 +       dyaop = (void *)dy_get(&op, br);
12625 +       if (IS_ERR(dyaop))
12626 +               goto out;
12627 +       dy_adx(dyaop, do_dx);
12628 +
12629 +out:
12630 +       return dyaop;
12631 +}
12632 +
12633 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
12634 +               struct inode *h_inode)
12635 +{
12636 +       int err, do_dx;
12637 +       struct super_block *sb;
12638 +       struct au_branch *br;
12639 +       struct au_dyaop *dyaop;
12640 +
12641 +       AuDebugOn(!S_ISREG(h_inode->i_mode));
12642 +       IiMustWriteLock(inode);
12643 +
12644 +       sb = inode->i_sb;
12645 +       br = au_sbr(sb, bindex);
12646 +       do_dx = !!au_opt_test(au_mntflags(sb), DIO);
12647 +       dyaop = dy_aget(br, h_inode->i_mapping->a_ops, do_dx);
12648 +       err = PTR_ERR(dyaop);
12649 +       if (IS_ERR(dyaop))
12650 +               /* unnecessary to call dy_fput() */
12651 +               goto out;
12652 +
12653 +       err = 0;
12654 +       inode->i_mapping->a_ops = &dyaop->da_op;
12655 +
12656 +out:
12657 +       return err;
12658 +}
12659 +
12660 +/*
12661 + * Is it safe to replace a_ops during the inode/file is in operation?
12662 + * Yes, I hope so.
12663 + */
12664 +int au_dy_irefresh(struct inode *inode)
12665 +{
12666 +       int err;
12667 +       aufs_bindex_t btop;
12668 +       struct inode *h_inode;
12669 +
12670 +       err = 0;
12671 +       if (S_ISREG(inode->i_mode)) {
12672 +               btop = au_ibtop(inode);
12673 +               h_inode = au_h_iptr(inode, btop);
12674 +               err = au_dy_iaop(inode, btop, h_inode);
12675 +       }
12676 +       return err;
12677 +}
12678 +
12679 +void au_dy_arefresh(int do_dx)
12680 +{
12681 +       struct hlist_bl_head *hbl;
12682 +       struct hlist_bl_node *pos;
12683 +       struct au_dykey *key;
12684 +
12685 +       hbl = dynop + AuDy_AOP;
12686 +       hlist_bl_lock(hbl);
12687 +       hlist_bl_for_each_entry(key, pos, hbl, dk_hnode)
12688 +               dy_adx((void *)key, do_dx);
12689 +       hlist_bl_unlock(hbl);
12690 +}
12691 +
12692 +/* ---------------------------------------------------------------------- */
12693 +
12694 +void __init au_dy_init(void)
12695 +{
12696 +       int i;
12697 +
12698 +       for (i = 0; i < AuDyLast; i++)
12699 +               INIT_HLIST_BL_HEAD(dynop + i);
12700 +}
12701 +
12702 +void au_dy_fin(void)
12703 +{
12704 +       int i;
12705 +
12706 +       for (i = 0; i < AuDyLast; i++)
12707 +               WARN_ON(!hlist_bl_empty(dynop + i));
12708 +}
12709 diff -urN /usr/share/empty/fs/aufs/dynop.h linux/fs/aufs/dynop.h
12710 --- /usr/share/empty/fs/aufs/dynop.h    1970-01-01 01:00:00.000000000 +0100
12711 +++ linux/fs/aufs/dynop.h       2021-12-03 15:38:59.933313976 +0100
12712 @@ -0,0 +1,77 @@
12713 +/* SPDX-License-Identifier: GPL-2.0 */
12714 +/*
12715 + * Copyright (C) 2010-2021 Junjiro R. Okajima
12716 + *
12717 + * This program, aufs is free software; you can redistribute it and/or modify
12718 + * it under the terms of the GNU General Public License as published by
12719 + * the Free Software Foundation; either version 2 of the License, or
12720 + * (at your option) any later version.
12721 + *
12722 + * This program is distributed in the hope that it will be useful,
12723 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12724 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12725 + * GNU General Public License for more details.
12726 + *
12727 + * You should have received a copy of the GNU General Public License
12728 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12729 + */
12730 +
12731 +/*
12732 + * dynamically customizable operations (for regular files only)
12733 + */
12734 +
12735 +#ifndef __AUFS_DYNOP_H__
12736 +#define __AUFS_DYNOP_H__
12737 +
12738 +#ifdef __KERNEL__
12739 +
12740 +#include <linux/fs.h>
12741 +#include <linux/kref.h>
12742 +
12743 +enum {AuDy_AOP, AuDyLast};
12744 +
12745 +struct au_dynop {
12746 +       int                                             dy_type;
12747 +       union {
12748 +               const void                              *dy_hop;
12749 +               const struct address_space_operations   *dy_haop;
12750 +       };
12751 +};
12752 +
12753 +struct au_dykey {
12754 +       union {
12755 +               struct hlist_bl_node    dk_hnode;
12756 +               struct rcu_head         dk_rcu;
12757 +       };
12758 +       struct au_dynop         dk_op;
12759 +
12760 +       /*
12761 +        * during I am in the branch local array, kref is gotten. when the
12762 +        * branch is removed, kref is put.
12763 +        */
12764 +       struct kref             dk_kref;
12765 +};
12766 +
12767 +/* stop unioning since their sizes are very different from each other */
12768 +struct au_dyaop {
12769 +       struct au_dykey                 da_key;
12770 +       struct address_space_operations da_op; /* not const */
12771 +};
12772 +/* make sure that 'struct au_dykey *' can be any type */
12773 +static_assert(!offsetof(struct au_dyaop, da_key));
12774 +
12775 +/* ---------------------------------------------------------------------- */
12776 +
12777 +/* dynop.c */
12778 +struct au_branch;
12779 +void au_dy_put(struct au_dykey *key);
12780 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
12781 +               struct inode *h_inode);
12782 +int au_dy_irefresh(struct inode *inode);
12783 +void au_dy_arefresh(int do_dio);
12784 +
12785 +void __init au_dy_init(void);
12786 +void au_dy_fin(void);
12787 +
12788 +#endif /* __KERNEL__ */
12789 +#endif /* __AUFS_DYNOP_H__ */
12790 diff -urN /usr/share/empty/fs/aufs/export.c linux/fs/aufs/export.c
12791 --- /usr/share/empty/fs/aufs/export.c   1970-01-01 01:00:00.000000000 +0100
12792 +++ linux/fs/aufs/export.c      2021-12-03 15:40:58.233313963 +0100
12793 @@ -0,0 +1,831 @@
12794 +// SPDX-License-Identifier: GPL-2.0
12795 +/*
12796 + * Copyright (C) 2005-2021 Junjiro R. Okajima
12797 + *
12798 + * This program, aufs is free software; you can redistribute it and/or modify
12799 + * it under the terms of the GNU General Public License as published by
12800 + * the Free Software Foundation; either version 2 of the License, or
12801 + * (at your option) any later version.
12802 + *
12803 + * This program is distributed in the hope that it will be useful,
12804 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12805 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12806 + * GNU General Public License for more details.
12807 + *
12808 + * You should have received a copy of the GNU General Public License
12809 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12810 + */
12811 +
12812 +/*
12813 + * export via nfs
12814 + */
12815 +
12816 +#include <linux/exportfs.h>
12817 +#include <linux/fs_struct.h>
12818 +#include <linux/namei.h>
12819 +#include <linux/nsproxy.h>
12820 +#include <linux/random.h>
12821 +#include <linux/writeback.h>
12822 +#include "aufs.h"
12823 +
12824 +union conv {
12825 +#ifdef CONFIG_AUFS_INO_T_64
12826 +       __u32 a[2];
12827 +#else
12828 +       __u32 a[1];
12829 +#endif
12830 +       ino_t ino;
12831 +};
12832 +
12833 +static ino_t decode_ino(__u32 *a)
12834 +{
12835 +       union conv u;
12836 +
12837 +       BUILD_BUG_ON(sizeof(u.ino) != sizeof(u.a));
12838 +       u.a[0] = a[0];
12839 +#ifdef CONFIG_AUFS_INO_T_64
12840 +       u.a[1] = a[1];
12841 +#endif
12842 +       return u.ino;
12843 +}
12844 +
12845 +static void encode_ino(__u32 *a, ino_t ino)
12846 +{
12847 +       union conv u;
12848 +
12849 +       u.ino = ino;
12850 +       a[0] = u.a[0];
12851 +#ifdef CONFIG_AUFS_INO_T_64
12852 +       a[1] = u.a[1];
12853 +#endif
12854 +}
12855 +
12856 +/* NFS file handle */
12857 +enum {
12858 +       Fh_br_id,
12859 +       Fh_sigen,
12860 +#ifdef CONFIG_AUFS_INO_T_64
12861 +       /* support 64bit inode number */
12862 +       Fh_ino1,
12863 +       Fh_ino2,
12864 +       Fh_dir_ino1,
12865 +       Fh_dir_ino2,
12866 +#else
12867 +       Fh_ino1,
12868 +       Fh_dir_ino1,
12869 +#endif
12870 +       Fh_igen,
12871 +       Fh_h_type,
12872 +       Fh_tail,
12873 +
12874 +       Fh_ino = Fh_ino1,
12875 +       Fh_dir_ino = Fh_dir_ino1
12876 +};
12877 +
12878 +static int au_test_anon(struct dentry *dentry)
12879 +{
12880 +       /* note: read d_flags without d_lock */
12881 +       return !!(dentry->d_flags & DCACHE_DISCONNECTED);
12882 +}
12883 +
12884 +int au_test_nfsd(void)
12885 +{
12886 +       int ret;
12887 +       struct task_struct *tsk = current;
12888 +       char comm[sizeof(tsk->comm)];
12889 +
12890 +       ret = 0;
12891 +       if (tsk->flags & PF_KTHREAD) {
12892 +               get_task_comm(comm, tsk);
12893 +               ret = !strcmp(comm, "nfsd");
12894 +       }
12895 +
12896 +       return ret;
12897 +}
12898 +
12899 +/* ---------------------------------------------------------------------- */
12900 +/* inode generation external table */
12901 +
12902 +void au_xigen_inc(struct inode *inode)
12903 +{
12904 +       loff_t pos;
12905 +       ssize_t sz;
12906 +       __u32 igen;
12907 +       struct super_block *sb;
12908 +       struct au_sbinfo *sbinfo;
12909 +
12910 +       sb = inode->i_sb;
12911 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
12912 +
12913 +       sbinfo = au_sbi(sb);
12914 +       pos = inode->i_ino;
12915 +       pos *= sizeof(igen);
12916 +       igen = inode->i_generation + 1;
12917 +       sz = xino_fwrite(sbinfo->si_xigen, &igen, sizeof(igen), &pos);
12918 +       if (sz == sizeof(igen))
12919 +               return; /* success */
12920 +
12921 +       if (unlikely(sz >= 0))
12922 +               AuIOErr("xigen error (%zd)\n", sz);
12923 +}
12924 +
12925 +int au_xigen_new(struct inode *inode)
12926 +{
12927 +       int err;
12928 +       loff_t pos;
12929 +       ssize_t sz;
12930 +       struct super_block *sb;
12931 +       struct au_sbinfo *sbinfo;
12932 +       struct file *file;
12933 +
12934 +       err = 0;
12935 +       /* todo: dirty, at mount time */
12936 +       if (inode->i_ino == AUFS_ROOT_INO)
12937 +               goto out;
12938 +       sb = inode->i_sb;
12939 +       SiMustAnyLock(sb);
12940 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
12941 +               goto out;
12942 +
12943 +       err = -EFBIG;
12944 +       pos = inode->i_ino;
12945 +       if (unlikely(au_loff_max / sizeof(inode->i_generation) - 1 < pos)) {
12946 +               AuIOErr1("too large i%lld\n", pos);
12947 +               goto out;
12948 +       }
12949 +       pos *= sizeof(inode->i_generation);
12950 +
12951 +       err = 0;
12952 +       sbinfo = au_sbi(sb);
12953 +       file = sbinfo->si_xigen;
12954 +       BUG_ON(!file);
12955 +
12956 +       if (vfsub_f_size_read(file)
12957 +           < pos + sizeof(inode->i_generation)) {
12958 +               inode->i_generation = atomic_inc_return(&sbinfo->si_xigen_next);
12959 +               sz = xino_fwrite(file, &inode->i_generation,
12960 +                                sizeof(inode->i_generation), &pos);
12961 +       } else
12962 +               sz = xino_fread(file, &inode->i_generation,
12963 +                               sizeof(inode->i_generation), &pos);
12964 +       if (sz == sizeof(inode->i_generation))
12965 +               goto out; /* success */
12966 +
12967 +       err = sz;
12968 +       if (unlikely(sz >= 0)) {
12969 +               err = -EIO;
12970 +               AuIOErr("xigen error (%zd)\n", sz);
12971 +       }
12972 +
12973 +out:
12974 +       return err;
12975 +}
12976 +
12977 +int au_xigen_set(struct super_block *sb, struct path *path)
12978 +{
12979 +       int err;
12980 +       struct au_sbinfo *sbinfo;
12981 +       struct file *file;
12982 +
12983 +       SiMustWriteLock(sb);
12984 +
12985 +       sbinfo = au_sbi(sb);
12986 +       file = au_xino_create2(sb, path, sbinfo->si_xigen);
12987 +       err = PTR_ERR(file);
12988 +       if (IS_ERR(file))
12989 +               goto out;
12990 +       err = 0;
12991 +       if (sbinfo->si_xigen)
12992 +               fput(sbinfo->si_xigen);
12993 +       sbinfo->si_xigen = file;
12994 +
12995 +out:
12996 +       AuTraceErr(err);
12997 +       return err;
12998 +}
12999 +
13000 +void au_xigen_clr(struct super_block *sb)
13001 +{
13002 +       struct au_sbinfo *sbinfo;
13003 +
13004 +       SiMustWriteLock(sb);
13005 +
13006 +       sbinfo = au_sbi(sb);
13007 +       if (sbinfo->si_xigen) {
13008 +               fput(sbinfo->si_xigen);
13009 +               sbinfo->si_xigen = NULL;
13010 +       }
13011 +}
13012 +
13013 +/* ---------------------------------------------------------------------- */
13014 +
13015 +static struct dentry *decode_by_ino(struct super_block *sb, ino_t ino,
13016 +                                   ino_t dir_ino)
13017 +{
13018 +       struct dentry *dentry, *d;
13019 +       struct inode *inode;
13020 +       unsigned int sigen;
13021 +
13022 +       dentry = NULL;
13023 +       inode = ilookup(sb, ino);
13024 +       if (!inode)
13025 +               goto out;
13026 +
13027 +       dentry = ERR_PTR(-ESTALE);
13028 +       sigen = au_sigen(sb);
13029 +       if (unlikely(au_is_bad_inode(inode)
13030 +                    || IS_DEADDIR(inode)
13031 +                    || sigen != au_iigen(inode, NULL)))
13032 +               goto out_iput;
13033 +
13034 +       dentry = NULL;
13035 +       if (!dir_ino || S_ISDIR(inode->i_mode))
13036 +               dentry = d_find_alias(inode);
13037 +       else {
13038 +               spin_lock(&inode->i_lock);
13039 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
13040 +                       spin_lock(&d->d_lock);
13041 +                       if (!au_test_anon(d)
13042 +                           && d_inode(d->d_parent)->i_ino == dir_ino) {
13043 +                               dentry = dget_dlock(d);
13044 +                               spin_unlock(&d->d_lock);
13045 +                               break;
13046 +                       }
13047 +                       spin_unlock(&d->d_lock);
13048 +               }
13049 +               spin_unlock(&inode->i_lock);
13050 +       }
13051 +       if (unlikely(dentry && au_digen_test(dentry, sigen))) {
13052 +               /* need to refresh */
13053 +               dput(dentry);
13054 +               dentry = NULL;
13055 +       }
13056 +
13057 +out_iput:
13058 +       iput(inode);
13059 +out:
13060 +       AuTraceErrPtr(dentry);
13061 +       return dentry;
13062 +}
13063 +
13064 +/* ---------------------------------------------------------------------- */
13065 +
13066 +/* todo: dirty? */
13067 +/* if exportfs_decode_fh() passed vfsmount*, we could be happy */
13068 +
13069 +struct au_compare_mnt_args {
13070 +       /* input */
13071 +       struct super_block *sb;
13072 +
13073 +       /* output */
13074 +       struct vfsmount *mnt;
13075 +};
13076 +
13077 +static int au_compare_mnt(struct vfsmount *mnt, void *arg)
13078 +{
13079 +       struct au_compare_mnt_args *a = arg;
13080 +
13081 +       if (mnt->mnt_sb != a->sb)
13082 +               return 0;
13083 +       a->mnt = mntget(mnt);
13084 +       return 1;
13085 +}
13086 +
13087 +static struct vfsmount *au_mnt_get(struct super_block *sb)
13088 +{
13089 +       int err;
13090 +       struct path root;
13091 +       struct au_compare_mnt_args args = {
13092 +               .sb = sb
13093 +       };
13094 +
13095 +       get_fs_root(current->fs, &root);
13096 +       rcu_read_lock();
13097 +       err = iterate_mounts(au_compare_mnt, &args, root.mnt);
13098 +       rcu_read_unlock();
13099 +       path_put(&root);
13100 +       AuDebugOn(!err);
13101 +       AuDebugOn(!args.mnt);
13102 +       return args.mnt;
13103 +}
13104 +
13105 +struct au_nfsd_si_lock {
13106 +       unsigned int sigen;
13107 +       aufs_bindex_t bindex, br_id;
13108 +       unsigned char force_lock;
13109 +};
13110 +
13111 +static int si_nfsd_read_lock(struct super_block *sb,
13112 +                            struct au_nfsd_si_lock *nsi_lock)
13113 +{
13114 +       int err;
13115 +       aufs_bindex_t bindex;
13116 +
13117 +       si_read_lock(sb, AuLock_FLUSH);
13118 +
13119 +       /* branch id may be wrapped around */
13120 +       err = 0;
13121 +       bindex = au_br_index(sb, nsi_lock->br_id);
13122 +       if (bindex >= 0 && nsi_lock->sigen + AUFS_BRANCH_MAX > au_sigen(sb))
13123 +               goto out; /* success */
13124 +
13125 +       err = -ESTALE;
13126 +       bindex = -1;
13127 +       if (!nsi_lock->force_lock)
13128 +               si_read_unlock(sb);
13129 +
13130 +out:
13131 +       nsi_lock->bindex = bindex;
13132 +       return err;
13133 +}
13134 +
13135 +struct find_name_by_ino {
13136 +       struct dir_context ctx;
13137 +       int called, found;
13138 +       ino_t ino;
13139 +       char *name;
13140 +       int namelen;
13141 +};
13142 +
13143 +static int
13144 +find_name_by_ino(struct dir_context *ctx, const char *name, int namelen,
13145 +                loff_t offset, u64 ino, unsigned int d_type)
13146 +{
13147 +       struct find_name_by_ino *a = container_of(ctx, struct find_name_by_ino,
13148 +                                                 ctx);
13149 +
13150 +       a->called++;
13151 +       if (a->ino != ino)
13152 +               return 0;
13153 +
13154 +       memcpy(a->name, name, namelen);
13155 +       a->namelen = namelen;
13156 +       a->found = 1;
13157 +       return 1;
13158 +}
13159 +
13160 +static struct dentry *au_lkup_by_ino(struct path *path, ino_t ino,
13161 +                                    struct au_nfsd_si_lock *nsi_lock)
13162 +{
13163 +       struct dentry *dentry, *parent;
13164 +       struct file *file;
13165 +       struct inode *dir;
13166 +       struct find_name_by_ino arg = {
13167 +               .ctx = {
13168 +                       .actor = find_name_by_ino
13169 +               }
13170 +       };
13171 +       int err;
13172 +
13173 +       parent = path->dentry;
13174 +       if (nsi_lock)
13175 +               si_read_unlock(parent->d_sb);
13176 +       file = vfsub_dentry_open(path, au_dir_roflags);
13177 +       dentry = (void *)file;
13178 +       if (IS_ERR(file))
13179 +               goto out;
13180 +
13181 +       dentry = ERR_PTR(-ENOMEM);
13182 +       arg.name = (void *)__get_free_page(GFP_NOFS);
13183 +       if (unlikely(!arg.name))
13184 +               goto out_file;
13185 +       arg.ino = ino;
13186 +       arg.found = 0;
13187 +       do {
13188 +               arg.called = 0;
13189 +               /* smp_mb(); */
13190 +               err = vfsub_iterate_dir(file, &arg.ctx);
13191 +       } while (!err && !arg.found && arg.called);
13192 +       dentry = ERR_PTR(err);
13193 +       if (unlikely(err))
13194 +               goto out_name;
13195 +       /* instead of ENOENT */
13196 +       dentry = ERR_PTR(-ESTALE);
13197 +       if (!arg.found)
13198 +               goto out_name;
13199 +
13200 +       /* do not call vfsub_lkup_one() */
13201 +       dir = d_inode(parent);
13202 +       dentry = vfsub_lookup_one_len_unlocked(arg.name, path, arg.namelen);
13203 +       AuTraceErrPtr(dentry);
13204 +       if (IS_ERR(dentry))
13205 +               goto out_name;
13206 +       AuDebugOn(au_test_anon(dentry));
13207 +       if (unlikely(d_really_is_negative(dentry))) {
13208 +               dput(dentry);
13209 +               dentry = ERR_PTR(-ENOENT);
13210 +       }
13211 +
13212 +out_name:
13213 +       free_page((unsigned long)arg.name);
13214 +out_file:
13215 +       fput(file);
13216 +out:
13217 +       if (unlikely(nsi_lock
13218 +                    && si_nfsd_read_lock(parent->d_sb, nsi_lock) < 0))
13219 +               if (!IS_ERR(dentry)) {
13220 +                       dput(dentry);
13221 +                       dentry = ERR_PTR(-ESTALE);
13222 +               }
13223 +       AuTraceErrPtr(dentry);
13224 +       return dentry;
13225 +}
13226 +
13227 +static struct dentry *decode_by_dir_ino(struct super_block *sb, ino_t ino,
13228 +                                       ino_t dir_ino,
13229 +                                       struct au_nfsd_si_lock *nsi_lock)
13230 +{
13231 +       struct dentry *dentry;
13232 +       struct path path;
13233 +
13234 +       if (dir_ino != AUFS_ROOT_INO) {
13235 +               path.dentry = decode_by_ino(sb, dir_ino, 0);
13236 +               dentry = path.dentry;
13237 +               if (!path.dentry || IS_ERR(path.dentry))
13238 +                       goto out;
13239 +               AuDebugOn(au_test_anon(path.dentry));
13240 +       } else
13241 +               path.dentry = dget(sb->s_root);
13242 +
13243 +       path.mnt = au_mnt_get(sb);
13244 +       dentry = au_lkup_by_ino(&path, ino, nsi_lock);
13245 +       path_put(&path);
13246 +
13247 +out:
13248 +       AuTraceErrPtr(dentry);
13249 +       return dentry;
13250 +}
13251 +
13252 +/* ---------------------------------------------------------------------- */
13253 +
13254 +static int h_acceptable(void *expv, struct dentry *dentry)
13255 +{
13256 +       return 1;
13257 +}
13258 +
13259 +static char *au_build_path(struct dentry *h_parent, struct path *h_rootpath,
13260 +                          char *buf, int len, struct super_block *sb)
13261 +{
13262 +       char *p;
13263 +       int n;
13264 +       struct path path;
13265 +
13266 +       p = d_path(h_rootpath, buf, len);
13267 +       if (IS_ERR(p))
13268 +               goto out;
13269 +       n = strlen(p);
13270 +
13271 +       path.mnt = h_rootpath->mnt;
13272 +       path.dentry = h_parent;
13273 +       p = d_path(&path, buf, len);
13274 +       if (IS_ERR(p))
13275 +               goto out;
13276 +       if (n != 1)
13277 +               p += n;
13278 +
13279 +       path.mnt = au_mnt_get(sb);
13280 +       path.dentry = sb->s_root;
13281 +       p = d_path(&path, buf, len - strlen(p));
13282 +       mntput(path.mnt);
13283 +       if (IS_ERR(p))
13284 +               goto out;
13285 +       if (n != 1)
13286 +               p[strlen(p)] = '/';
13287 +
13288 +out:
13289 +       AuTraceErrPtr(p);
13290 +       return p;
13291 +}
13292 +
13293 +static
13294 +struct dentry *decode_by_path(struct super_block *sb, ino_t ino, __u32 *fh,
13295 +                             int fh_len, struct au_nfsd_si_lock *nsi_lock)
13296 +{
13297 +       struct dentry *dentry, *h_parent, *root;
13298 +       struct super_block *h_sb;
13299 +       char *pathname, *p;
13300 +       struct vfsmount *h_mnt;
13301 +       struct au_branch *br;
13302 +       int err;
13303 +       struct path path;
13304 +
13305 +       br = au_sbr(sb, nsi_lock->bindex);
13306 +       h_mnt = au_br_mnt(br);
13307 +       h_sb = h_mnt->mnt_sb;
13308 +       /* todo: call lower fh_to_dentry()? fh_to_parent()? */
13309 +       lockdep_off();
13310 +       h_parent = exportfs_decode_fh(h_mnt, (void *)(fh + Fh_tail),
13311 +                                     fh_len - Fh_tail, fh[Fh_h_type],
13312 +                                     h_acceptable, /*context*/NULL);
13313 +       lockdep_on();
13314 +       dentry = h_parent;
13315 +       if (unlikely(!h_parent || IS_ERR(h_parent))) {
13316 +               AuWarn1("%s decode_fh failed, %ld\n",
13317 +                       au_sbtype(h_sb), PTR_ERR(h_parent));
13318 +               goto out;
13319 +       }
13320 +       dentry = NULL;
13321 +       if (unlikely(au_test_anon(h_parent))) {
13322 +               AuWarn1("%s decode_fh returned a disconnected dentry\n",
13323 +                       au_sbtype(h_sb));
13324 +               goto out_h_parent;
13325 +       }
13326 +
13327 +       dentry = ERR_PTR(-ENOMEM);
13328 +       pathname = (void *)__get_free_page(GFP_NOFS);
13329 +       if (unlikely(!pathname))
13330 +               goto out_h_parent;
13331 +
13332 +       root = sb->s_root;
13333 +       path.mnt = h_mnt;
13334 +       di_read_lock_parent(root, !AuLock_IR);
13335 +       path.dentry = au_h_dptr(root, nsi_lock->bindex);
13336 +       di_read_unlock(root, !AuLock_IR);
13337 +       p = au_build_path(h_parent, &path, pathname, PAGE_SIZE, sb);
13338 +       dentry = (void *)p;
13339 +       if (IS_ERR(p))
13340 +               goto out_pathname;
13341 +
13342 +       si_read_unlock(sb);
13343 +       err = vfsub_kern_path(p, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &path);
13344 +       dentry = ERR_PTR(err);
13345 +       if (unlikely(err))
13346 +               goto out_relock;
13347 +
13348 +       dentry = ERR_PTR(-ENOENT);
13349 +       AuDebugOn(au_test_anon(path.dentry));
13350 +       if (unlikely(d_really_is_negative(path.dentry)))
13351 +               goto out_path;
13352 +
13353 +       if (ino != d_inode(path.dentry)->i_ino)
13354 +               dentry = au_lkup_by_ino(&path, ino, /*nsi_lock*/NULL);
13355 +       else
13356 +               dentry = dget(path.dentry);
13357 +
13358 +out_path:
13359 +       path_put(&path);
13360 +out_relock:
13361 +       if (unlikely(si_nfsd_read_lock(sb, nsi_lock) < 0))
13362 +               if (!IS_ERR(dentry)) {
13363 +                       dput(dentry);
13364 +                       dentry = ERR_PTR(-ESTALE);
13365 +               }
13366 +out_pathname:
13367 +       free_page((unsigned long)pathname);
13368 +out_h_parent:
13369 +       dput(h_parent);
13370 +out:
13371 +       AuTraceErrPtr(dentry);
13372 +       return dentry;
13373 +}
13374 +
13375 +/* ---------------------------------------------------------------------- */
13376 +
13377 +static struct dentry *
13378 +aufs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
13379 +                 int fh_type)
13380 +{
13381 +       struct dentry *dentry;
13382 +       __u32 *fh = fid->raw;
13383 +       struct au_branch *br;
13384 +       ino_t ino, dir_ino;
13385 +       struct au_nfsd_si_lock nsi_lock = {
13386 +               .force_lock     = 0
13387 +       };
13388 +
13389 +       dentry = ERR_PTR(-ESTALE);
13390 +       /* it should never happen, but the file handle is unreliable */
13391 +       if (unlikely(fh_len < Fh_tail))
13392 +               goto out;
13393 +       nsi_lock.sigen = fh[Fh_sigen];
13394 +       nsi_lock.br_id = fh[Fh_br_id];
13395 +
13396 +       /* branch id may be wrapped around */
13397 +       br = NULL;
13398 +       if (unlikely(si_nfsd_read_lock(sb, &nsi_lock)))
13399 +               goto out;
13400 +       nsi_lock.force_lock = 1;
13401 +
13402 +       /* is this inode still cached? */
13403 +       ino = decode_ino(fh + Fh_ino);
13404 +       /* it should never happen */
13405 +       if (unlikely(ino == AUFS_ROOT_INO))
13406 +               goto out_unlock;
13407 +
13408 +       dir_ino = decode_ino(fh + Fh_dir_ino);
13409 +       dentry = decode_by_ino(sb, ino, dir_ino);
13410 +       if (IS_ERR(dentry))
13411 +               goto out_unlock;
13412 +       if (dentry)
13413 +               goto accept;
13414 +
13415 +       /* is the parent dir cached? */
13416 +       br = au_sbr(sb, nsi_lock.bindex);
13417 +       au_lcnt_inc(&br->br_nfiles);
13418 +       dentry = decode_by_dir_ino(sb, ino, dir_ino, &nsi_lock);
13419 +       if (IS_ERR(dentry))
13420 +               goto out_unlock;
13421 +       if (dentry)
13422 +               goto accept;
13423 +
13424 +       /* lookup path */
13425 +       dentry = decode_by_path(sb, ino, fh, fh_len, &nsi_lock);
13426 +       if (IS_ERR(dentry))
13427 +               goto out_unlock;
13428 +       if (unlikely(!dentry))
13429 +               /* todo?: make it ESTALE */
13430 +               goto out_unlock;
13431 +
13432 +accept:
13433 +       if (!au_digen_test(dentry, au_sigen(sb))
13434 +           && d_inode(dentry)->i_generation == fh[Fh_igen])
13435 +               goto out_unlock; /* success */
13436 +
13437 +       dput(dentry);
13438 +       dentry = ERR_PTR(-ESTALE);
13439 +out_unlock:
13440 +       if (br)
13441 +               au_lcnt_dec(&br->br_nfiles);
13442 +       si_read_unlock(sb);
13443 +out:
13444 +       AuTraceErrPtr(dentry);
13445 +       return dentry;
13446 +}
13447 +
13448 +#if 0 /* reserved for future use */
13449 +/* support subtreecheck option */
13450 +static struct dentry *aufs_fh_to_parent(struct super_block *sb, struct fid *fid,
13451 +                                       int fh_len, int fh_type)
13452 +{
13453 +       struct dentry *parent;
13454 +       __u32 *fh = fid->raw;
13455 +       ino_t dir_ino;
13456 +
13457 +       dir_ino = decode_ino(fh + Fh_dir_ino);
13458 +       parent = decode_by_ino(sb, dir_ino, 0);
13459 +       if (IS_ERR(parent))
13460 +               goto out;
13461 +       if (!parent)
13462 +               parent = decode_by_path(sb, au_br_index(sb, fh[Fh_br_id]),
13463 +                                       dir_ino, fh, fh_len);
13464 +
13465 +out:
13466 +       AuTraceErrPtr(parent);
13467 +       return parent;
13468 +}
13469 +#endif
13470 +
13471 +/* ---------------------------------------------------------------------- */
13472 +
13473 +static int aufs_encode_fh(struct inode *inode, __u32 *fh, int *max_len,
13474 +                         struct inode *dir)
13475 +{
13476 +       int err;
13477 +       aufs_bindex_t bindex;
13478 +       struct super_block *sb, *h_sb;
13479 +       struct dentry *dentry, *parent, *h_parent;
13480 +       struct inode *h_dir;
13481 +       struct au_branch *br;
13482 +
13483 +       err = -ENOSPC;
13484 +       if (unlikely(*max_len <= Fh_tail)) {
13485 +               AuWarn1("NFSv2 client (max_len %d)?\n", *max_len);
13486 +               goto out;
13487 +       }
13488 +
13489 +       err = FILEID_ROOT;
13490 +       if (inode->i_ino == AUFS_ROOT_INO) {
13491 +               AuDebugOn(inode->i_ino != AUFS_ROOT_INO);
13492 +               goto out;
13493 +       }
13494 +
13495 +       h_parent = NULL;
13496 +       sb = inode->i_sb;
13497 +       err = si_read_lock(sb, AuLock_FLUSH);
13498 +       if (unlikely(err))
13499 +               goto out;
13500 +
13501 +#ifdef CONFIG_AUFS_DEBUG
13502 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
13503 +               AuWarn1("NFS-exporting requires xino\n");
13504 +#endif
13505 +       err = -EIO;
13506 +       parent = NULL;
13507 +       ii_read_lock_child(inode);
13508 +       bindex = au_ibtop(inode);
13509 +       if (!dir) {
13510 +               dentry = d_find_any_alias(inode);
13511 +               if (unlikely(!dentry))
13512 +                       goto out_unlock;
13513 +               AuDebugOn(au_test_anon(dentry));
13514 +               parent = dget_parent(dentry);
13515 +               dput(dentry);
13516 +               if (unlikely(!parent))
13517 +                       goto out_unlock;
13518 +               if (d_really_is_positive(parent))
13519 +                       dir = d_inode(parent);
13520 +       }
13521 +
13522 +       ii_read_lock_parent(dir);
13523 +       h_dir = au_h_iptr(dir, bindex);
13524 +       ii_read_unlock(dir);
13525 +       if (unlikely(!h_dir))
13526 +               goto out_parent;
13527 +       h_parent = d_find_any_alias(h_dir);
13528 +       if (unlikely(!h_parent))
13529 +               goto out_hparent;
13530 +
13531 +       err = -EPERM;
13532 +       br = au_sbr(sb, bindex);
13533 +       h_sb = au_br_sb(br);
13534 +       if (unlikely(!h_sb->s_export_op)) {
13535 +               AuErr1("%s branch is not exportable\n", au_sbtype(h_sb));
13536 +               goto out_hparent;
13537 +       }
13538 +
13539 +       fh[Fh_br_id] = br->br_id;
13540 +       fh[Fh_sigen] = au_sigen(sb);
13541 +       encode_ino(fh + Fh_ino, inode->i_ino);
13542 +       encode_ino(fh + Fh_dir_ino, dir->i_ino);
13543 +       fh[Fh_igen] = inode->i_generation;
13544 +
13545 +       *max_len -= Fh_tail;
13546 +       fh[Fh_h_type] = exportfs_encode_fh(h_parent, (void *)(fh + Fh_tail),
13547 +                                          max_len,
13548 +                                          /*connectable or subtreecheck*/0);
13549 +       err = fh[Fh_h_type];
13550 +       *max_len += Fh_tail;
13551 +       /* todo: macros? */
13552 +       if (err != FILEID_INVALID)
13553 +               err = 99;
13554 +       else
13555 +               AuWarn1("%s encode_fh failed\n", au_sbtype(h_sb));
13556 +
13557 +out_hparent:
13558 +       dput(h_parent);
13559 +out_parent:
13560 +       dput(parent);
13561 +out_unlock:
13562 +       ii_read_unlock(inode);
13563 +       si_read_unlock(sb);
13564 +out:
13565 +       if (unlikely(err < 0))
13566 +               err = FILEID_INVALID;
13567 +       return err;
13568 +}
13569 +
13570 +/* ---------------------------------------------------------------------- */
13571 +
13572 +static int aufs_commit_metadata(struct inode *inode)
13573 +{
13574 +       int err;
13575 +       aufs_bindex_t bindex;
13576 +       struct super_block *sb;
13577 +       struct inode *h_inode;
13578 +       int (*f)(struct inode *inode);
13579 +
13580 +       sb = inode->i_sb;
13581 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13582 +       ii_write_lock_child(inode);
13583 +       bindex = au_ibtop(inode);
13584 +       AuDebugOn(bindex < 0);
13585 +       h_inode = au_h_iptr(inode, bindex);
13586 +
13587 +       f = h_inode->i_sb->s_export_op->commit_metadata;
13588 +       if (f)
13589 +               err = f(h_inode);
13590 +       else
13591 +               err = sync_inode_metadata(h_inode, /*wait*/1);
13592 +
13593 +       au_cpup_attr_timesizes(inode);
13594 +       ii_write_unlock(inode);
13595 +       si_read_unlock(sb);
13596 +       return err;
13597 +}
13598 +
13599 +/* ---------------------------------------------------------------------- */
13600 +
13601 +static struct export_operations aufs_export_op = {
13602 +       .fh_to_dentry           = aufs_fh_to_dentry,
13603 +       /* .fh_to_parent        = aufs_fh_to_parent, */
13604 +       .encode_fh              = aufs_encode_fh,
13605 +       .commit_metadata        = aufs_commit_metadata
13606 +};
13607 +
13608 +void au_export_init(struct super_block *sb)
13609 +{
13610 +       struct au_sbinfo *sbinfo;
13611 +       __u32 u;
13612 +
13613 +       BUILD_BUG_ON_MSG(IS_BUILTIN(CONFIG_AUFS_FS)
13614 +                        && IS_MODULE(CONFIG_EXPORTFS),
13615 +                        AUFS_NAME ": unsupported configuration "
13616 +                        "CONFIG_EXPORTFS=m and CONFIG_AUFS_FS=y");
13617 +
13618 +       sb->s_export_op = &aufs_export_op;
13619 +       sbinfo = au_sbi(sb);
13620 +       sbinfo->si_xigen = NULL;
13621 +       get_random_bytes(&u, sizeof(u));
13622 +       BUILD_BUG_ON(sizeof(u) != sizeof(int));
13623 +       atomic_set(&sbinfo->si_xigen_next, u);
13624 +}
13625 diff -urN /usr/share/empty/fs/aufs/fhsm.c linux/fs/aufs/fhsm.c
13626 --- /usr/share/empty/fs/aufs/fhsm.c     1970-01-01 01:00:00.000000000 +0100
13627 +++ linux/fs/aufs/fhsm.c        2021-12-03 15:38:59.933313976 +0100
13628 @@ -0,0 +1,427 @@
13629 +// SPDX-License-Identifier: GPL-2.0
13630 +/*
13631 + * Copyright (C) 2011-2021 Junjiro R. Okajima
13632 + *
13633 + * This program, aufs is free software; you can redistribute it and/or modify
13634 + * it under the terms of the GNU General Public License as published by
13635 + * the Free Software Foundation; either version 2 of the License, or
13636 + * (at your option) any later version.
13637 + *
13638 + * This program is distributed in the hope that it will be useful,
13639 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13640 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13641 + * GNU General Public License for more details.
13642 + *
13643 + * You should have received a copy of the GNU General Public License
13644 + * along with this program; if not, write to the Free Software
13645 + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
13646 + */
13647 +
13648 +/*
13649 + * File-based Hierarchy Storage Management
13650 + */
13651 +
13652 +#include <linux/anon_inodes.h>
13653 +#include <linux/poll.h>
13654 +#include <linux/seq_file.h>
13655 +#include <linux/statfs.h>
13656 +#include "aufs.h"
13657 +
13658 +static aufs_bindex_t au_fhsm_bottom(struct super_block *sb)
13659 +{
13660 +       struct au_sbinfo *sbinfo;
13661 +       struct au_fhsm *fhsm;
13662 +
13663 +       SiMustAnyLock(sb);
13664 +
13665 +       sbinfo = au_sbi(sb);
13666 +       fhsm = &sbinfo->si_fhsm;
13667 +       AuDebugOn(!fhsm);
13668 +       return fhsm->fhsm_bottom;
13669 +}
13670 +
13671 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex)
13672 +{
13673 +       struct au_sbinfo *sbinfo;
13674 +       struct au_fhsm *fhsm;
13675 +
13676 +       SiMustWriteLock(sb);
13677 +
13678 +       sbinfo = au_sbi(sb);
13679 +       fhsm = &sbinfo->si_fhsm;
13680 +       AuDebugOn(!fhsm);
13681 +       fhsm->fhsm_bottom = bindex;
13682 +}
13683 +
13684 +/* ---------------------------------------------------------------------- */
13685 +
13686 +static int au_fhsm_test_jiffy(struct au_sbinfo *sbinfo, struct au_branch *br)
13687 +{
13688 +       struct au_br_fhsm *bf;
13689 +
13690 +       bf = br->br_fhsm;
13691 +       MtxMustLock(&bf->bf_lock);
13692 +
13693 +       return !bf->bf_readable
13694 +               || time_after(jiffies,
13695 +                             bf->bf_jiffy + sbinfo->si_fhsm.fhsm_expire);
13696 +}
13697 +
13698 +/* ---------------------------------------------------------------------- */
13699 +
13700 +static void au_fhsm_notify(struct super_block *sb, int val)
13701 +{
13702 +       struct au_sbinfo *sbinfo;
13703 +       struct au_fhsm *fhsm;
13704 +
13705 +       SiMustAnyLock(sb);
13706 +
13707 +       sbinfo = au_sbi(sb);
13708 +       fhsm = &sbinfo->si_fhsm;
13709 +       if (au_fhsm_pid(fhsm)
13710 +           && atomic_read(&fhsm->fhsm_readable) != -1) {
13711 +               atomic_set(&fhsm->fhsm_readable, val);
13712 +               if (val)
13713 +                       wake_up(&fhsm->fhsm_wqh);
13714 +       }
13715 +}
13716 +
13717 +static int au_fhsm_stfs(struct super_block *sb, aufs_bindex_t bindex,
13718 +                       struct aufs_stfs *rstfs, int do_lock, int do_notify)
13719 +{
13720 +       int err;
13721 +       struct au_branch *br;
13722 +       struct au_br_fhsm *bf;
13723 +
13724 +       br = au_sbr(sb, bindex);
13725 +       AuDebugOn(au_br_rdonly(br));
13726 +       bf = br->br_fhsm;
13727 +       AuDebugOn(!bf);
13728 +
13729 +       if (do_lock)
13730 +               mutex_lock(&bf->bf_lock);
13731 +       else
13732 +               MtxMustLock(&bf->bf_lock);
13733 +
13734 +       /* sb->s_root for NFS is unreliable */
13735 +       err = au_br_stfs(br, &bf->bf_stfs);
13736 +       if (unlikely(err)) {
13737 +               AuErr1("FHSM failed (%d), b%d, ignored.\n", bindex, err);
13738 +               goto out;
13739 +       }
13740 +
13741 +       bf->bf_jiffy = jiffies;
13742 +       bf->bf_readable = 1;
13743 +       if (do_notify)
13744 +               au_fhsm_notify(sb, /*val*/1);
13745 +       if (rstfs)
13746 +               *rstfs = bf->bf_stfs;
13747 +
13748 +out:
13749 +       if (do_lock)
13750 +               mutex_unlock(&bf->bf_lock);
13751 +       au_fhsm_notify(sb, /*val*/1);
13752 +
13753 +       return err;
13754 +}
13755 +
13756 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force)
13757 +{
13758 +       int err;
13759 +       struct au_sbinfo *sbinfo;
13760 +       struct au_fhsm *fhsm;
13761 +       struct au_branch *br;
13762 +       struct au_br_fhsm *bf;
13763 +
13764 +       AuDbg("b%d, force %d\n", bindex, force);
13765 +       SiMustAnyLock(sb);
13766 +
13767 +       sbinfo = au_sbi(sb);
13768 +       fhsm = &sbinfo->si_fhsm;
13769 +       if (!au_ftest_si(sbinfo, FHSM)
13770 +           || fhsm->fhsm_bottom == bindex)
13771 +               return;
13772 +
13773 +       br = au_sbr(sb, bindex);
13774 +       bf = br->br_fhsm;
13775 +       AuDebugOn(!bf);
13776 +       mutex_lock(&bf->bf_lock);
13777 +       if (force
13778 +           || au_fhsm_pid(fhsm)
13779 +           || au_fhsm_test_jiffy(sbinfo, br))
13780 +               err = au_fhsm_stfs(sb, bindex, /*rstfs*/NULL, /*do_lock*/0,
13781 +                                 /*do_notify*/1);
13782 +       mutex_unlock(&bf->bf_lock);
13783 +}
13784 +
13785 +void au_fhsm_wrote_all(struct super_block *sb, int force)
13786 +{
13787 +       aufs_bindex_t bindex, bbot;
13788 +       struct au_branch *br;
13789 +
13790 +       /* exclude the bottom */
13791 +       bbot = au_fhsm_bottom(sb);
13792 +       for (bindex = 0; bindex < bbot; bindex++) {
13793 +               br = au_sbr(sb, bindex);
13794 +               if (au_br_fhsm(br->br_perm))
13795 +                       au_fhsm_wrote(sb, bindex, force);
13796 +       }
13797 +}
13798 +
13799 +/* ---------------------------------------------------------------------- */
13800 +
13801 +static __poll_t au_fhsm_poll(struct file *file, struct poll_table_struct *wait)
13802 +{
13803 +       __poll_t mask;
13804 +       struct au_sbinfo *sbinfo;
13805 +       struct au_fhsm *fhsm;
13806 +
13807 +       mask = 0;
13808 +       sbinfo = file->private_data;
13809 +       fhsm = &sbinfo->si_fhsm;
13810 +       poll_wait(file, &fhsm->fhsm_wqh, wait);
13811 +       if (atomic_read(&fhsm->fhsm_readable))
13812 +               mask = EPOLLIN /* | EPOLLRDNORM */;
13813 +
13814 +       if (!mask)
13815 +               AuDbg("mask 0x%x\n", mask);
13816 +       return mask;
13817 +}
13818 +
13819 +static int au_fhsm_do_read_one(struct aufs_stbr __user *stbr,
13820 +                             struct aufs_stfs *stfs, __s16 brid)
13821 +{
13822 +       int err;
13823 +
13824 +       err = copy_to_user(&stbr->stfs, stfs, sizeof(*stfs));
13825 +       if (!err)
13826 +               err = __put_user(brid, &stbr->brid);
13827 +       if (unlikely(err))
13828 +               err = -EFAULT;
13829 +
13830 +       return err;
13831 +}
13832 +
13833 +static ssize_t au_fhsm_do_read(struct super_block *sb,
13834 +                              struct aufs_stbr __user *stbr, size_t count)
13835 +{
13836 +       ssize_t err;
13837 +       int nstbr;
13838 +       aufs_bindex_t bindex, bbot;
13839 +       struct au_branch *br;
13840 +       struct au_br_fhsm *bf;
13841 +
13842 +       /* except the bottom branch */
13843 +       err = 0;
13844 +       nstbr = 0;
13845 +       bbot = au_fhsm_bottom(sb);
13846 +       for (bindex = 0; !err && bindex < bbot; bindex++) {
13847 +               br = au_sbr(sb, bindex);
13848 +               if (!au_br_fhsm(br->br_perm))
13849 +                       continue;
13850 +
13851 +               bf = br->br_fhsm;
13852 +               mutex_lock(&bf->bf_lock);
13853 +               if (bf->bf_readable) {
13854 +                       err = -EFAULT;
13855 +                       if (count >= sizeof(*stbr))
13856 +                               err = au_fhsm_do_read_one(stbr++, &bf->bf_stfs,
13857 +                                                         br->br_id);
13858 +                       if (!err) {
13859 +                               bf->bf_readable = 0;
13860 +                               count -= sizeof(*stbr);
13861 +                               nstbr++;
13862 +                       }
13863 +               }
13864 +               mutex_unlock(&bf->bf_lock);
13865 +       }
13866 +       if (!err)
13867 +               err = sizeof(*stbr) * nstbr;
13868 +
13869 +       return err;
13870 +}
13871 +
13872 +static ssize_t au_fhsm_read(struct file *file, char __user *buf, size_t count,
13873 +                          loff_t *pos)
13874 +{
13875 +       ssize_t err;
13876 +       int readable;
13877 +       aufs_bindex_t nfhsm, bindex, bbot;
13878 +       struct au_sbinfo *sbinfo;
13879 +       struct au_fhsm *fhsm;
13880 +       struct au_branch *br;
13881 +       struct super_block *sb;
13882 +
13883 +       err = 0;
13884 +       sbinfo = file->private_data;
13885 +       fhsm = &sbinfo->si_fhsm;
13886 +need_data:
13887 +       spin_lock_irq(&fhsm->fhsm_wqh.lock);
13888 +       if (!atomic_read(&fhsm->fhsm_readable)) {
13889 +               if (vfsub_file_flags(file) & O_NONBLOCK)
13890 +                       err = -EAGAIN;
13891 +               else
13892 +                       err = wait_event_interruptible_locked_irq
13893 +                               (fhsm->fhsm_wqh,
13894 +                                atomic_read(&fhsm->fhsm_readable));
13895 +       }
13896 +       spin_unlock_irq(&fhsm->fhsm_wqh.lock);
13897 +       if (unlikely(err))
13898 +               goto out;
13899 +
13900 +       /* sb may already be dead */
13901 +       au_rw_read_lock(&sbinfo->si_rwsem);
13902 +       readable = atomic_read(&fhsm->fhsm_readable);
13903 +       if (readable > 0) {
13904 +               sb = sbinfo->si_sb;
13905 +               AuDebugOn(!sb);
13906 +               /* exclude the bottom branch */
13907 +               nfhsm = 0;
13908 +               bbot = au_fhsm_bottom(sb);
13909 +               for (bindex = 0; bindex < bbot; bindex++) {
13910 +                       br = au_sbr(sb, bindex);
13911 +                       if (au_br_fhsm(br->br_perm))
13912 +                               nfhsm++;
13913 +               }
13914 +               err = -EMSGSIZE;
13915 +               if (nfhsm * sizeof(struct aufs_stbr) <= count) {
13916 +                       atomic_set(&fhsm->fhsm_readable, 0);
13917 +                       err = au_fhsm_do_read(sbinfo->si_sb, (void __user *)buf,
13918 +                                            count);
13919 +               }
13920 +       }
13921 +       au_rw_read_unlock(&sbinfo->si_rwsem);
13922 +       if (!readable)
13923 +               goto need_data;
13924 +
13925 +out:
13926 +       return err;
13927 +}
13928 +
13929 +static int au_fhsm_release(struct inode *inode, struct file *file)
13930 +{
13931 +       struct au_sbinfo *sbinfo;
13932 +       struct au_fhsm *fhsm;
13933 +
13934 +       /* sb may already be dead */
13935 +       sbinfo = file->private_data;
13936 +       fhsm = &sbinfo->si_fhsm;
13937 +       spin_lock(&fhsm->fhsm_spin);
13938 +       fhsm->fhsm_pid = 0;
13939 +       spin_unlock(&fhsm->fhsm_spin);
13940 +       kobject_put(&sbinfo->si_kobj);
13941 +
13942 +       return 0;
13943 +}
13944 +
13945 +static const struct file_operations au_fhsm_fops = {
13946 +       .owner          = THIS_MODULE,
13947 +       .llseek         = noop_llseek,
13948 +       .read           = au_fhsm_read,
13949 +       .poll           = au_fhsm_poll,
13950 +       .release        = au_fhsm_release
13951 +};
13952 +
13953 +int au_fhsm_fd(struct super_block *sb, int oflags)
13954 +{
13955 +       int err, fd;
13956 +       struct au_sbinfo *sbinfo;
13957 +       struct au_fhsm *fhsm;
13958 +
13959 +       err = -EPERM;
13960 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
13961 +               goto out;
13962 +
13963 +       err = -EINVAL;
13964 +       if (unlikely(oflags & ~(O_CLOEXEC | O_NONBLOCK)))
13965 +               goto out;
13966 +
13967 +       err = 0;
13968 +       sbinfo = au_sbi(sb);
13969 +       fhsm = &sbinfo->si_fhsm;
13970 +       spin_lock(&fhsm->fhsm_spin);
13971 +       if (!fhsm->fhsm_pid)
13972 +               fhsm->fhsm_pid = current->pid;
13973 +       else
13974 +               err = -EBUSY;
13975 +       spin_unlock(&fhsm->fhsm_spin);
13976 +       if (unlikely(err))
13977 +               goto out;
13978 +
13979 +       oflags |= O_RDONLY;
13980 +       /* oflags |= FMODE_NONOTIFY; */
13981 +       fd = anon_inode_getfd("[aufs_fhsm]", &au_fhsm_fops, sbinfo, oflags);
13982 +       err = fd;
13983 +       if (unlikely(fd < 0))
13984 +               goto out_pid;
13985 +
13986 +       /* succeed regardless 'fhsm' status */
13987 +       kobject_get(&sbinfo->si_kobj);
13988 +       si_noflush_read_lock(sb);
13989 +       if (au_ftest_si(sbinfo, FHSM))
13990 +               au_fhsm_wrote_all(sb, /*force*/0);
13991 +       si_read_unlock(sb);
13992 +       goto out; /* success */
13993 +
13994 +out_pid:
13995 +       spin_lock(&fhsm->fhsm_spin);
13996 +       fhsm->fhsm_pid = 0;
13997 +       spin_unlock(&fhsm->fhsm_spin);
13998 +out:
13999 +       AuTraceErr(err);
14000 +       return err;
14001 +}
14002 +
14003 +/* ---------------------------------------------------------------------- */
14004 +
14005 +int au_fhsm_br_alloc(struct au_branch *br)
14006 +{
14007 +       int err;
14008 +
14009 +       err = 0;
14010 +       br->br_fhsm = kmalloc(sizeof(*br->br_fhsm), GFP_NOFS);
14011 +       if (br->br_fhsm)
14012 +               au_br_fhsm_init(br->br_fhsm);
14013 +       else
14014 +               err = -ENOMEM;
14015 +
14016 +       return err;
14017 +}
14018 +
14019 +/* ---------------------------------------------------------------------- */
14020 +
14021 +void au_fhsm_fin(struct super_block *sb)
14022 +{
14023 +       au_fhsm_notify(sb, /*val*/-1);
14024 +}
14025 +
14026 +void au_fhsm_init(struct au_sbinfo *sbinfo)
14027 +{
14028 +       struct au_fhsm *fhsm;
14029 +
14030 +       fhsm = &sbinfo->si_fhsm;
14031 +       spin_lock_init(&fhsm->fhsm_spin);
14032 +       init_waitqueue_head(&fhsm->fhsm_wqh);
14033 +       atomic_set(&fhsm->fhsm_readable, 0);
14034 +       fhsm->fhsm_expire
14035 +               = msecs_to_jiffies(AUFS_FHSM_CACHE_DEF_SEC * MSEC_PER_SEC);
14036 +       fhsm->fhsm_bottom = -1;
14037 +}
14038 +
14039 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec)
14040 +{
14041 +       sbinfo->si_fhsm.fhsm_expire
14042 +               = msecs_to_jiffies(sec * MSEC_PER_SEC);
14043 +}
14044 +
14045 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo)
14046 +{
14047 +       unsigned int u;
14048 +
14049 +       if (!au_ftest_si(sbinfo, FHSM))
14050 +               return;
14051 +
14052 +       u = jiffies_to_msecs(sbinfo->si_fhsm.fhsm_expire) / MSEC_PER_SEC;
14053 +       if (u != AUFS_FHSM_CACHE_DEF_SEC)
14054 +               seq_printf(seq, ",fhsm_sec=%u", u);
14055 +}
14056 diff -urN /usr/share/empty/fs/aufs/file.c linux/fs/aufs/file.c
14057 --- /usr/share/empty/fs/aufs/file.c     1970-01-01 01:00:00.000000000 +0100
14058 +++ linux/fs/aufs/file.c        2021-12-03 15:38:59.936647310 +0100
14059 @@ -0,0 +1,863 @@
14060 +// SPDX-License-Identifier: GPL-2.0
14061 +/*
14062 + * Copyright (C) 2005-2021 Junjiro R. Okajima
14063 + *
14064 + * This program, aufs is free software; you can redistribute it and/or modify
14065 + * it under the terms of the GNU General Public License as published by
14066 + * the Free Software Foundation; either version 2 of the License, or
14067 + * (at your option) any later version.
14068 + *
14069 + * This program is distributed in the hope that it will be useful,
14070 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14071 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14072 + * GNU General Public License for more details.
14073 + *
14074 + * You should have received a copy of the GNU General Public License
14075 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14076 + */
14077 +
14078 +/*
14079 + * handling file/dir, and address_space operation
14080 + */
14081 +
14082 +#ifdef CONFIG_AUFS_DEBUG
14083 +#include <linux/migrate.h>
14084 +#endif
14085 +#include <linux/pagemap.h>
14086 +#include "aufs.h"
14087 +
14088 +/* drop flags for writing */
14089 +unsigned int au_file_roflags(unsigned int flags)
14090 +{
14091 +       flags &= ~(O_WRONLY | O_RDWR | O_APPEND | O_CREAT | O_TRUNC);
14092 +       flags |= O_RDONLY | O_NOATIME;
14093 +       return flags;
14094 +}
14095 +
14096 +/* common functions to regular file and dir */
14097 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
14098 +                      struct file *file, int force_wr)
14099 +{
14100 +       struct file *h_file;
14101 +       struct dentry *h_dentry;
14102 +       struct inode *h_inode;
14103 +       struct super_block *sb;
14104 +       struct au_branch *br;
14105 +       struct path h_path;
14106 +       int err;
14107 +
14108 +       /* a race condition can happen between open and unlink/rmdir */
14109 +       h_file = ERR_PTR(-ENOENT);
14110 +       h_dentry = au_h_dptr(dentry, bindex);
14111 +       if (au_test_nfsd() && (!h_dentry || d_is_negative(h_dentry)))
14112 +               goto out;
14113 +       h_inode = d_inode(h_dentry);
14114 +       spin_lock(&h_dentry->d_lock);
14115 +       err = (!d_unhashed(dentry) && d_unlinked(h_dentry))
14116 +               /* || !d_inode(dentry)->i_nlink */
14117 +               ;
14118 +       spin_unlock(&h_dentry->d_lock);
14119 +       if (unlikely(err))
14120 +               goto out;
14121 +
14122 +       sb = dentry->d_sb;
14123 +       br = au_sbr(sb, bindex);
14124 +       err = au_br_test_oflag(flags, br);
14125 +       h_file = ERR_PTR(err);
14126 +       if (unlikely(err))
14127 +               goto out;
14128 +
14129 +       /* drop flags for writing */
14130 +       if (au_test_ro(sb, bindex, d_inode(dentry))) {
14131 +               if (force_wr && !(flags & O_WRONLY))
14132 +                       force_wr = 0;
14133 +               flags = au_file_roflags(flags);
14134 +               if (force_wr) {
14135 +                       h_file = ERR_PTR(-EROFS);
14136 +                       flags = au_file_roflags(flags);
14137 +                       if (unlikely(vfsub_native_ro(h_inode)
14138 +                                    || IS_APPEND(h_inode)))
14139 +                               goto out;
14140 +                       flags &= ~O_ACCMODE;
14141 +                       flags |= O_WRONLY;
14142 +               }
14143 +       }
14144 +       flags &= ~O_CREAT;
14145 +       au_lcnt_inc(&br->br_nfiles);
14146 +       h_path.dentry = h_dentry;
14147 +       h_path.mnt = au_br_mnt(br);
14148 +       h_file = vfsub_dentry_open(&h_path, flags);
14149 +       if (IS_ERR(h_file))
14150 +               goto out_br;
14151 +
14152 +       if (flags & __FMODE_EXEC) {
14153 +               err = deny_write_access(h_file);
14154 +               if (unlikely(err)) {
14155 +                       fput(h_file);
14156 +                       h_file = ERR_PTR(err);
14157 +                       goto out_br;
14158 +               }
14159 +       }
14160 +       fsnotify_open(h_file);
14161 +       goto out; /* success */
14162 +
14163 +out_br:
14164 +       au_lcnt_dec(&br->br_nfiles);
14165 +out:
14166 +       return h_file;
14167 +}
14168 +
14169 +static int au_cmoo(struct dentry *dentry)
14170 +{
14171 +       int err, cmoo, matched;
14172 +       unsigned int udba;
14173 +       struct path h_path;
14174 +       struct au_pin pin;
14175 +       struct au_cp_generic cpg = {
14176 +               .dentry = dentry,
14177 +               .bdst   = -1,
14178 +               .bsrc   = -1,
14179 +               .len    = -1,
14180 +               .pin    = &pin,
14181 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
14182 +       };
14183 +       struct inode *delegated;
14184 +       struct super_block *sb;
14185 +       struct au_sbinfo *sbinfo;
14186 +       struct au_fhsm *fhsm;
14187 +       pid_t pid;
14188 +       struct au_branch *br;
14189 +       struct dentry *parent;
14190 +       struct au_hinode *hdir;
14191 +
14192 +       DiMustWriteLock(dentry);
14193 +       IiMustWriteLock(d_inode(dentry));
14194 +
14195 +       err = 0;
14196 +       if (IS_ROOT(dentry))
14197 +               goto out;
14198 +       cpg.bsrc = au_dbtop(dentry);
14199 +       if (!cpg.bsrc)
14200 +               goto out;
14201 +
14202 +       sb = dentry->d_sb;
14203 +       sbinfo = au_sbi(sb);
14204 +       fhsm = &sbinfo->si_fhsm;
14205 +       pid = au_fhsm_pid(fhsm);
14206 +       rcu_read_lock();
14207 +       matched = (pid
14208 +                  && (current->pid == pid
14209 +                      || rcu_dereference(current->real_parent)->pid == pid));
14210 +       rcu_read_unlock();
14211 +       if (matched)
14212 +               goto out;
14213 +
14214 +       br = au_sbr(sb, cpg.bsrc);
14215 +       cmoo = au_br_cmoo(br->br_perm);
14216 +       if (!cmoo)
14217 +               goto out;
14218 +       if (!d_is_reg(dentry))
14219 +               cmoo &= AuBrAttr_COO_ALL;
14220 +       if (!cmoo)
14221 +               goto out;
14222 +
14223 +       parent = dget_parent(dentry);
14224 +       di_write_lock_parent(parent);
14225 +       err = au_wbr_do_copyup_bu(dentry, cpg.bsrc - 1);
14226 +       cpg.bdst = err;
14227 +       if (unlikely(err < 0)) {
14228 +               err = 0;        /* there is no upper writable branch */
14229 +               goto out_dgrade;
14230 +       }
14231 +       AuDbg("bsrc %d, bdst %d\n", cpg.bsrc, cpg.bdst);
14232 +
14233 +       /* do not respect the coo attrib for the target branch */
14234 +       err = au_cpup_dirs(dentry, cpg.bdst);
14235 +       if (unlikely(err))
14236 +               goto out_dgrade;
14237 +
14238 +       di_downgrade_lock(parent, AuLock_IR);
14239 +       udba = au_opt_udba(sb);
14240 +       err = au_pin(&pin, dentry, cpg.bdst, udba,
14241 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
14242 +       if (unlikely(err))
14243 +               goto out_parent;
14244 +
14245 +       err = au_sio_cpup_simple(&cpg);
14246 +       au_unpin(&pin);
14247 +       if (unlikely(err))
14248 +               goto out_parent;
14249 +       if (!(cmoo & AuBrWAttr_MOO))
14250 +               goto out_parent; /* success */
14251 +
14252 +       err = au_pin(&pin, dentry, cpg.bsrc, udba,
14253 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
14254 +       if (unlikely(err))
14255 +               goto out_parent;
14256 +
14257 +       h_path.mnt = au_br_mnt(br);
14258 +       h_path.dentry = au_h_dptr(dentry, cpg.bsrc);
14259 +       hdir = au_hi(d_inode(parent), cpg.bsrc);
14260 +       delegated = NULL;
14261 +       err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated, /*force*/1);
14262 +       au_unpin(&pin);
14263 +       /* todo: keep h_dentry or not? */
14264 +       if (unlikely(err == -EWOULDBLOCK)) {
14265 +               pr_warn("cannot retry for NFSv4 delegation"
14266 +                       " for an internal unlink\n");
14267 +               iput(delegated);
14268 +       }
14269 +       if (unlikely(err)) {
14270 +               pr_err("unlink %pd after coo failed (%d), ignored\n",
14271 +                      dentry, err);
14272 +               err = 0;
14273 +       }
14274 +       goto out_parent; /* success */
14275 +
14276 +out_dgrade:
14277 +       di_downgrade_lock(parent, AuLock_IR);
14278 +out_parent:
14279 +       di_read_unlock(parent, AuLock_IR);
14280 +       dput(parent);
14281 +out:
14282 +       AuTraceErr(err);
14283 +       return err;
14284 +}
14285 +
14286 +int au_do_open(struct file *file, struct au_do_open_args *args)
14287 +{
14288 +       int err, aopen = args->aopen;
14289 +       struct dentry *dentry;
14290 +       struct au_finfo *finfo;
14291 +
14292 +       if (!aopen)
14293 +               err = au_finfo_init(file, args->fidir);
14294 +       else {
14295 +               lockdep_off();
14296 +               err = au_finfo_init(file, args->fidir);
14297 +               lockdep_on();
14298 +       }
14299 +       if (unlikely(err))
14300 +               goto out;
14301 +
14302 +       dentry = file->f_path.dentry;
14303 +       AuDebugOn(IS_ERR_OR_NULL(dentry));
14304 +       di_write_lock_child(dentry);
14305 +       err = au_cmoo(dentry);
14306 +       di_downgrade_lock(dentry, AuLock_IR);
14307 +       if (!err) {
14308 +               if (!aopen)
14309 +                       err = args->open(file, vfsub_file_flags(file), NULL);
14310 +               else {
14311 +                       lockdep_off();
14312 +                       err = args->open(file, vfsub_file_flags(file),
14313 +                                        args->h_file);
14314 +                       lockdep_on();
14315 +               }
14316 +       }
14317 +       di_read_unlock(dentry, AuLock_IR);
14318 +
14319 +       finfo = au_fi(file);
14320 +       if (!err) {
14321 +               finfo->fi_file = file;
14322 +               au_hbl_add(&finfo->fi_hlist,
14323 +                          &au_sbi(file->f_path.dentry->d_sb)->si_files);
14324 +       }
14325 +       if (!aopen)
14326 +               fi_write_unlock(file);
14327 +       else {
14328 +               lockdep_off();
14329 +               fi_write_unlock(file);
14330 +               lockdep_on();
14331 +       }
14332 +       if (unlikely(err)) {
14333 +               finfo->fi_hdir = NULL;
14334 +               au_finfo_fin(file);
14335 +       }
14336 +
14337 +out:
14338 +       AuTraceErr(err);
14339 +       return err;
14340 +}
14341 +
14342 +int au_reopen_nondir(struct file *file)
14343 +{
14344 +       int err;
14345 +       aufs_bindex_t btop;
14346 +       struct dentry *dentry;
14347 +       struct au_branch *br;
14348 +       struct file *h_file, *h_file_tmp;
14349 +
14350 +       dentry = file->f_path.dentry;
14351 +       btop = au_dbtop(dentry);
14352 +       br = au_sbr(dentry->d_sb, btop);
14353 +       h_file_tmp = NULL;
14354 +       if (au_fbtop(file) == btop) {
14355 +               h_file = au_hf_top(file);
14356 +               if (file->f_mode == h_file->f_mode)
14357 +                       return 0; /* success */
14358 +               h_file_tmp = h_file;
14359 +               get_file(h_file_tmp);
14360 +               au_lcnt_inc(&br->br_nfiles);
14361 +               au_set_h_fptr(file, btop, NULL);
14362 +       }
14363 +       AuDebugOn(au_fi(file)->fi_hdir);
14364 +       /*
14365 +        * it can happen
14366 +        * file exists on both of rw and ro
14367 +        * open --> dbtop and fbtop are both 0
14368 +        * prepend a branch as rw, "rw" become ro
14369 +        * remove rw/file
14370 +        * delete the top branch, "rw" becomes rw again
14371 +        *      --> dbtop is 1, fbtop is still 0
14372 +        * write --> fbtop is 0 but dbtop is 1
14373 +        */
14374 +       /* AuDebugOn(au_fbtop(file) < btop); */
14375 +
14376 +       h_file = au_h_open(dentry, btop, vfsub_file_flags(file) & ~O_TRUNC,
14377 +                          file, /*force_wr*/0);
14378 +       err = PTR_ERR(h_file);
14379 +       if (IS_ERR(h_file)) {
14380 +               if (h_file_tmp) {
14381 +                       /* revert */
14382 +                       au_set_h_fptr(file, btop, h_file_tmp);
14383 +                       h_file_tmp = NULL;
14384 +               }
14385 +               goto out; /* todo: close all? */
14386 +       }
14387 +
14388 +       err = 0;
14389 +       au_set_fbtop(file, btop);
14390 +       au_set_h_fptr(file, btop, h_file);
14391 +       au_update_figen(file);
14392 +       /* todo: necessary? */
14393 +       /* file->f_ra = h_file->f_ra; */
14394 +
14395 +out:
14396 +       if (h_file_tmp) {
14397 +               fput(h_file_tmp);
14398 +               au_lcnt_dec(&br->br_nfiles);
14399 +       }
14400 +       return err;
14401 +}
14402 +
14403 +/* ---------------------------------------------------------------------- */
14404 +
14405 +static int au_reopen_wh(struct file *file, aufs_bindex_t btgt,
14406 +                       struct dentry *hi_wh)
14407 +{
14408 +       int err;
14409 +       aufs_bindex_t btop;
14410 +       struct au_dinfo *dinfo;
14411 +       struct dentry *h_dentry;
14412 +       struct au_hdentry *hdp;
14413 +
14414 +       dinfo = au_di(file->f_path.dentry);
14415 +       AuRwMustWriteLock(&dinfo->di_rwsem);
14416 +
14417 +       btop = dinfo->di_btop;
14418 +       dinfo->di_btop = btgt;
14419 +       hdp = au_hdentry(dinfo, btgt);
14420 +       h_dentry = hdp->hd_dentry;
14421 +       hdp->hd_dentry = hi_wh;
14422 +       err = au_reopen_nondir(file);
14423 +       hdp->hd_dentry = h_dentry;
14424 +       dinfo->di_btop = btop;
14425 +
14426 +       return err;
14427 +}
14428 +
14429 +static int au_ready_to_write_wh(struct file *file, loff_t len,
14430 +                               aufs_bindex_t bcpup, struct au_pin *pin)
14431 +{
14432 +       int err;
14433 +       struct inode *inode, *h_inode;
14434 +       struct dentry *h_dentry, *hi_wh;
14435 +       struct au_cp_generic cpg = {
14436 +               .dentry = file->f_path.dentry,
14437 +               .bdst   = bcpup,
14438 +               .bsrc   = -1,
14439 +               .len    = len,
14440 +               .pin    = pin
14441 +       };
14442 +
14443 +       au_update_dbtop(cpg.dentry);
14444 +       inode = d_inode(cpg.dentry);
14445 +       h_inode = NULL;
14446 +       if (au_dbtop(cpg.dentry) <= bcpup
14447 +           && au_dbbot(cpg.dentry) >= bcpup) {
14448 +               h_dentry = au_h_dptr(cpg.dentry, bcpup);
14449 +               if (h_dentry && d_is_positive(h_dentry))
14450 +                       h_inode = d_inode(h_dentry);
14451 +       }
14452 +       hi_wh = au_hi_wh(inode, bcpup);
14453 +       if (!hi_wh && !h_inode)
14454 +               err = au_sio_cpup_wh(&cpg, file);
14455 +       else
14456 +               /* already copied-up after unlink */
14457 +               err = au_reopen_wh(file, bcpup, hi_wh);
14458 +
14459 +       if (!err
14460 +           && (inode->i_nlink > 1
14461 +               || (inode->i_state & I_LINKABLE))
14462 +           && au_opt_test(au_mntflags(cpg.dentry->d_sb), PLINK))
14463 +               au_plink_append(inode, bcpup, au_h_dptr(cpg.dentry, bcpup));
14464 +
14465 +       return err;
14466 +}
14467 +
14468 +/*
14469 + * prepare the @file for writing.
14470 + */
14471 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin)
14472 +{
14473 +       int err;
14474 +       aufs_bindex_t dbtop;
14475 +       struct dentry *parent;
14476 +       struct inode *inode;
14477 +       struct super_block *sb;
14478 +       struct file *h_file;
14479 +       struct au_cp_generic cpg = {
14480 +               .dentry = file->f_path.dentry,
14481 +               .bdst   = -1,
14482 +               .bsrc   = -1,
14483 +               .len    = len,
14484 +               .pin    = pin,
14485 +               .flags  = AuCpup_DTIME
14486 +       };
14487 +
14488 +       sb = cpg.dentry->d_sb;
14489 +       inode = d_inode(cpg.dentry);
14490 +       cpg.bsrc = au_fbtop(file);
14491 +       err = au_test_ro(sb, cpg.bsrc, inode);
14492 +       if (!err && (au_hf_top(file)->f_mode & FMODE_WRITE)) {
14493 +               err = au_pin(pin, cpg.dentry, cpg.bsrc, AuOpt_UDBA_NONE,
14494 +                            /*flags*/0);
14495 +               goto out;
14496 +       }
14497 +
14498 +       /* need to cpup or reopen */
14499 +       parent = dget_parent(cpg.dentry);
14500 +       di_write_lock_parent(parent);
14501 +       err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
14502 +       cpg.bdst = err;
14503 +       if (unlikely(err < 0))
14504 +               goto out_dgrade;
14505 +       err = 0;
14506 +
14507 +       if (!d_unhashed(cpg.dentry) && !au_h_dptr(parent, cpg.bdst)) {
14508 +               err = au_cpup_dirs(cpg.dentry, cpg.bdst);
14509 +               if (unlikely(err))
14510 +                       goto out_dgrade;
14511 +       }
14512 +
14513 +       err = au_pin(pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
14514 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
14515 +       if (unlikely(err))
14516 +               goto out_dgrade;
14517 +
14518 +       dbtop = au_dbtop(cpg.dentry);
14519 +       if (dbtop <= cpg.bdst)
14520 +               cpg.bsrc = cpg.bdst;
14521 +
14522 +       if (dbtop <= cpg.bdst           /* just reopen */
14523 +           || !d_unhashed(cpg.dentry)  /* copyup and reopen */
14524 +               ) {
14525 +               h_file = au_h_open_pre(cpg.dentry, cpg.bsrc, /*force_wr*/0);
14526 +               if (IS_ERR(h_file))
14527 +                       err = PTR_ERR(h_file);
14528 +               else {
14529 +                       di_downgrade_lock(parent, AuLock_IR);
14530 +                       if (dbtop > cpg.bdst)
14531 +                               err = au_sio_cpup_simple(&cpg);
14532 +                       if (!err)
14533 +                               err = au_reopen_nondir(file);
14534 +                       au_h_open_post(cpg.dentry, cpg.bsrc, h_file);
14535 +               }
14536 +       } else {                        /* copyup as wh and reopen */
14537 +               /*
14538 +                * since writable hfsplus branch is not supported,
14539 +                * h_open_pre/post() are unnecessary.
14540 +                */
14541 +               err = au_ready_to_write_wh(file, len, cpg.bdst, pin);
14542 +               di_downgrade_lock(parent, AuLock_IR);
14543 +       }
14544 +
14545 +       if (!err) {
14546 +               au_pin_set_parent_lflag(pin, /*lflag*/0);
14547 +               goto out_dput; /* success */
14548 +       }
14549 +       au_unpin(pin);
14550 +       goto out_unlock;
14551 +
14552 +out_dgrade:
14553 +       di_downgrade_lock(parent, AuLock_IR);
14554 +out_unlock:
14555 +       di_read_unlock(parent, AuLock_IR);
14556 +out_dput:
14557 +       dput(parent);
14558 +out:
14559 +       return err;
14560 +}
14561 +
14562 +/* ---------------------------------------------------------------------- */
14563 +
14564 +int au_do_flush(struct file *file, fl_owner_t id,
14565 +               int (*flush)(struct file *file, fl_owner_t id))
14566 +{
14567 +       int err;
14568 +       struct super_block *sb;
14569 +       struct inode *inode;
14570 +
14571 +       inode = file_inode(file);
14572 +       sb = inode->i_sb;
14573 +       si_noflush_read_lock(sb);
14574 +       fi_read_lock(file);
14575 +       ii_read_lock_child(inode);
14576 +
14577 +       err = flush(file, id);
14578 +       au_cpup_attr_timesizes(inode);
14579 +
14580 +       ii_read_unlock(inode);
14581 +       fi_read_unlock(file);
14582 +       si_read_unlock(sb);
14583 +       return err;
14584 +}
14585 +
14586 +/* ---------------------------------------------------------------------- */
14587 +
14588 +static int au_file_refresh_by_inode(struct file *file, int *need_reopen)
14589 +{
14590 +       int err;
14591 +       struct au_pin pin;
14592 +       struct au_finfo *finfo;
14593 +       struct dentry *parent, *hi_wh;
14594 +       struct inode *inode;
14595 +       struct super_block *sb;
14596 +       struct au_cp_generic cpg = {
14597 +               .dentry = file->f_path.dentry,
14598 +               .bdst   = -1,
14599 +               .bsrc   = -1,
14600 +               .len    = -1,
14601 +               .pin    = &pin,
14602 +               .flags  = AuCpup_DTIME
14603 +       };
14604 +
14605 +       FiMustWriteLock(file);
14606 +
14607 +       err = 0;
14608 +       finfo = au_fi(file);
14609 +       sb = cpg.dentry->d_sb;
14610 +       inode = d_inode(cpg.dentry);
14611 +       cpg.bdst = au_ibtop(inode);
14612 +       if (cpg.bdst == finfo->fi_btop || IS_ROOT(cpg.dentry))
14613 +               goto out;
14614 +
14615 +       parent = dget_parent(cpg.dentry);
14616 +       if (au_test_ro(sb, cpg.bdst, inode)) {
14617 +               di_read_lock_parent(parent, !AuLock_IR);
14618 +               err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
14619 +               cpg.bdst = err;
14620 +               di_read_unlock(parent, !AuLock_IR);
14621 +               if (unlikely(err < 0))
14622 +                       goto out_parent;
14623 +               err = 0;
14624 +       }
14625 +
14626 +       di_read_lock_parent(parent, AuLock_IR);
14627 +       hi_wh = au_hi_wh(inode, cpg.bdst);
14628 +       if (!S_ISDIR(inode->i_mode)
14629 +           && au_opt_test(au_mntflags(sb), PLINK)
14630 +           && au_plink_test(inode)
14631 +           && !d_unhashed(cpg.dentry)
14632 +           && cpg.bdst < au_dbtop(cpg.dentry)) {
14633 +               err = au_test_and_cpup_dirs(cpg.dentry, cpg.bdst);
14634 +               if (unlikely(err))
14635 +                       goto out_unlock;
14636 +
14637 +               /* always superio. */
14638 +               err = au_pin(&pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
14639 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
14640 +               if (!err) {
14641 +                       err = au_sio_cpup_simple(&cpg);
14642 +                       au_unpin(&pin);
14643 +               }
14644 +       } else if (hi_wh) {
14645 +               /* already copied-up after unlink */
14646 +               err = au_reopen_wh(file, cpg.bdst, hi_wh);
14647 +               *need_reopen = 0;
14648 +       }
14649 +
14650 +out_unlock:
14651 +       di_read_unlock(parent, AuLock_IR);
14652 +out_parent:
14653 +       dput(parent);
14654 +out:
14655 +       return err;
14656 +}
14657 +
14658 +static void au_do_refresh_dir(struct file *file)
14659 +{
14660 +       aufs_bindex_t bindex, bbot, new_bindex, brid;
14661 +       struct au_hfile *p, tmp, *q;
14662 +       struct au_finfo *finfo;
14663 +       struct super_block *sb;
14664 +       struct au_fidir *fidir;
14665 +
14666 +       FiMustWriteLock(file);
14667 +
14668 +       sb = file->f_path.dentry->d_sb;
14669 +       finfo = au_fi(file);
14670 +       fidir = finfo->fi_hdir;
14671 +       AuDebugOn(!fidir);
14672 +       p = fidir->fd_hfile + finfo->fi_btop;
14673 +       brid = p->hf_br->br_id;
14674 +       bbot = fidir->fd_bbot;
14675 +       for (bindex = finfo->fi_btop; bindex <= bbot; bindex++, p++) {
14676 +               if (!p->hf_file)
14677 +                       continue;
14678 +
14679 +               new_bindex = au_br_index(sb, p->hf_br->br_id);
14680 +               if (new_bindex == bindex)
14681 +                       continue;
14682 +               if (new_bindex < 0) {
14683 +                       au_set_h_fptr(file, bindex, NULL);
14684 +                       continue;
14685 +               }
14686 +
14687 +               /* swap two lower inode, and loop again */
14688 +               q = fidir->fd_hfile + new_bindex;
14689 +               tmp = *q;
14690 +               *q = *p;
14691 +               *p = tmp;
14692 +               if (tmp.hf_file) {
14693 +                       bindex--;
14694 +                       p--;
14695 +               }
14696 +       }
14697 +
14698 +       p = fidir->fd_hfile;
14699 +       if (!au_test_mmapped(file) && !d_unlinked(file->f_path.dentry)) {
14700 +               bbot = au_sbbot(sb);
14701 +               for (finfo->fi_btop = 0; finfo->fi_btop <= bbot;
14702 +                    finfo->fi_btop++, p++)
14703 +                       if (p->hf_file) {
14704 +                               if (file_inode(p->hf_file))
14705 +                                       break;
14706 +                               au_hfput(p, /*execed*/0);
14707 +                       }
14708 +       } else {
14709 +               bbot = au_br_index(sb, brid);
14710 +               for (finfo->fi_btop = 0; finfo->fi_btop < bbot;
14711 +                    finfo->fi_btop++, p++)
14712 +                       if (p->hf_file)
14713 +                               au_hfput(p, /*execed*/0);
14714 +               bbot = au_sbbot(sb);
14715 +       }
14716 +
14717 +       p = fidir->fd_hfile + bbot;
14718 +       for (fidir->fd_bbot = bbot; fidir->fd_bbot >= finfo->fi_btop;
14719 +            fidir->fd_bbot--, p--)
14720 +               if (p->hf_file) {
14721 +                       if (file_inode(p->hf_file))
14722 +                               break;
14723 +                       au_hfput(p, /*execed*/0);
14724 +               }
14725 +       AuDebugOn(fidir->fd_bbot < finfo->fi_btop);
14726 +}
14727 +
14728 +/*
14729 + * after branch manipulating, refresh the file.
14730 + */
14731 +static int refresh_file(struct file *file, int (*reopen)(struct file *file))
14732 +{
14733 +       int err, need_reopen, nbr;
14734 +       aufs_bindex_t bbot, bindex;
14735 +       struct dentry *dentry;
14736 +       struct super_block *sb;
14737 +       struct au_finfo *finfo;
14738 +       struct au_hfile *hfile;
14739 +
14740 +       dentry = file->f_path.dentry;
14741 +       sb = dentry->d_sb;
14742 +       nbr = au_sbbot(sb) + 1;
14743 +       finfo = au_fi(file);
14744 +       if (!finfo->fi_hdir) {
14745 +               hfile = &finfo->fi_htop;
14746 +               AuDebugOn(!hfile->hf_file);
14747 +               bindex = au_br_index(sb, hfile->hf_br->br_id);
14748 +               AuDebugOn(bindex < 0);
14749 +               if (bindex != finfo->fi_btop)
14750 +                       au_set_fbtop(file, bindex);
14751 +       } else {
14752 +               err = au_fidir_realloc(finfo, nbr, /*may_shrink*/0);
14753 +               if (unlikely(err))
14754 +                       goto out;
14755 +               au_do_refresh_dir(file);
14756 +       }
14757 +
14758 +       err = 0;
14759 +       need_reopen = 1;
14760 +       if (!au_test_mmapped(file))
14761 +               err = au_file_refresh_by_inode(file, &need_reopen);
14762 +       if (finfo->fi_hdir)
14763 +               /* harmless if err */
14764 +               au_fidir_realloc(finfo, nbr, /*may_shrink*/1);
14765 +       if (!err && need_reopen && !d_unlinked(dentry))
14766 +               err = reopen(file);
14767 +       if (!err) {
14768 +               au_update_figen(file);
14769 +               goto out; /* success */
14770 +       }
14771 +
14772 +       /* error, close all lower files */
14773 +       if (finfo->fi_hdir) {
14774 +               bbot = au_fbbot_dir(file);
14775 +               for (bindex = au_fbtop(file); bindex <= bbot; bindex++)
14776 +                       au_set_h_fptr(file, bindex, NULL);
14777 +       }
14778 +
14779 +out:
14780 +       return err;
14781 +}
14782 +
14783 +/* common function to regular file and dir */
14784 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
14785 +                         int wlock, unsigned int fi_lsc)
14786 +{
14787 +       int err;
14788 +       unsigned int sigen, figen;
14789 +       aufs_bindex_t btop;
14790 +       unsigned char pseudo_link;
14791 +       struct dentry *dentry;
14792 +       struct inode *inode;
14793 +
14794 +       err = 0;
14795 +       dentry = file->f_path.dentry;
14796 +       inode = d_inode(dentry);
14797 +       sigen = au_sigen(dentry->d_sb);
14798 +       fi_write_lock_nested(file, fi_lsc);
14799 +       figen = au_figen(file);
14800 +       if (!fi_lsc)
14801 +               di_write_lock_child(dentry);
14802 +       else
14803 +               di_write_lock_child2(dentry);
14804 +       btop = au_dbtop(dentry);
14805 +       pseudo_link = (btop != au_ibtop(inode));
14806 +       if (sigen == figen && !pseudo_link && au_fbtop(file) == btop) {
14807 +               if (!wlock) {
14808 +                       di_downgrade_lock(dentry, AuLock_IR);
14809 +                       fi_downgrade_lock(file);
14810 +               }
14811 +               goto out; /* success */
14812 +       }
14813 +
14814 +       AuDbg("sigen %d, figen %d\n", sigen, figen);
14815 +       if (au_digen_test(dentry, sigen)) {
14816 +               err = au_reval_dpath(dentry, sigen);
14817 +               AuDebugOn(!err && au_digen_test(dentry, sigen));
14818 +       }
14819 +
14820 +       if (!err)
14821 +               err = refresh_file(file, reopen);
14822 +       if (!err) {
14823 +               if (!wlock) {
14824 +                       di_downgrade_lock(dentry, AuLock_IR);
14825 +                       fi_downgrade_lock(file);
14826 +               }
14827 +       } else {
14828 +               di_write_unlock(dentry);
14829 +               fi_write_unlock(file);
14830 +       }
14831 +
14832 +out:
14833 +       return err;
14834 +}
14835 +
14836 +/* ---------------------------------------------------------------------- */
14837 +
14838 +/* cf. aufs_nopage() */
14839 +/* for madvise(2) */
14840 +static int aufs_readpage(struct file *file __maybe_unused, struct page *page)
14841 +{
14842 +       unlock_page(page);
14843 +       return 0;
14844 +}
14845 +
14846 +/* it will never be called, but necessary to support O_DIRECT */
14847 +static ssize_t aufs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
14848 +{ BUG(); return 0; }
14849 +
14850 +/* they will never be called. */
14851 +#ifdef CONFIG_AUFS_DEBUG
14852 +static int aufs_write_begin(struct file *file, struct address_space *mapping,
14853 +                           loff_t pos, unsigned len, unsigned flags,
14854 +                           struct page **pagep, void **fsdata)
14855 +{ AuUnsupport(); return 0; }
14856 +static int aufs_write_end(struct file *file, struct address_space *mapping,
14857 +                         loff_t pos, unsigned len, unsigned copied,
14858 +                         struct page *page, void *fsdata)
14859 +{ AuUnsupport(); return 0; }
14860 +static int aufs_writepage(struct page *page, struct writeback_control *wbc)
14861 +{ AuUnsupport(); return 0; }
14862 +
14863 +static int aufs_set_page_dirty(struct page *page)
14864 +{ AuUnsupport(); return 0; }
14865 +static void aufs_invalidatepage(struct page *page, unsigned int offset,
14866 +                               unsigned int length)
14867 +{ AuUnsupport(); }
14868 +static int aufs_releasepage(struct page *page, gfp_t gfp)
14869 +{ AuUnsupport(); return 0; }
14870 +#if 0 /* called by memory compaction regardless file */
14871 +static int aufs_migratepage(struct address_space *mapping, struct page *newpage,
14872 +                           struct page *page, enum migrate_mode mode)
14873 +{ AuUnsupport(); return 0; }
14874 +#endif
14875 +static bool aufs_isolate_page(struct page *page, isolate_mode_t mode)
14876 +{ AuUnsupport(); return true; }
14877 +static void aufs_putback_page(struct page *page)
14878 +{ AuUnsupport(); }
14879 +static int aufs_launder_page(struct page *page)
14880 +{ AuUnsupport(); return 0; }
14881 +static int aufs_is_partially_uptodate(struct page *page,
14882 +                                     unsigned long from,
14883 +                                     unsigned long count)
14884 +{ AuUnsupport(); return 0; }
14885 +static void aufs_is_dirty_writeback(struct page *page, bool *dirty,
14886 +                                   bool *writeback)
14887 +{ AuUnsupport(); }
14888 +static int aufs_error_remove_page(struct address_space *mapping,
14889 +                                 struct page *page)
14890 +{ AuUnsupport(); return 0; }
14891 +static int aufs_swap_activate(struct swap_info_struct *sis, struct file *file,
14892 +                             sector_t *span)
14893 +{ AuUnsupport(); return 0; }
14894 +static void aufs_swap_deactivate(struct file *file)
14895 +{ AuUnsupport(); }
14896 +#endif /* CONFIG_AUFS_DEBUG */
14897 +
14898 +const struct address_space_operations aufs_aop = {
14899 +       .readpage               = aufs_readpage,
14900 +       .direct_IO              = aufs_direct_IO,
14901 +#ifdef CONFIG_AUFS_DEBUG
14902 +       .writepage              = aufs_writepage,
14903 +       /* no writepages, because of writepage */
14904 +       .set_page_dirty         = aufs_set_page_dirty,
14905 +       /* no readpages, because of readpage */
14906 +       .write_begin            = aufs_write_begin,
14907 +       .write_end              = aufs_write_end,
14908 +       /* no bmap, no block device */
14909 +       .invalidatepage         = aufs_invalidatepage,
14910 +       .releasepage            = aufs_releasepage,
14911 +       /* is fallback_migrate_page ok? */
14912 +       /* .migratepage         = aufs_migratepage, */
14913 +       .isolate_page           = aufs_isolate_page,
14914 +       .putback_page           = aufs_putback_page,
14915 +       .launder_page           = aufs_launder_page,
14916 +       .is_partially_uptodate  = aufs_is_partially_uptodate,
14917 +       .is_dirty_writeback     = aufs_is_dirty_writeback,
14918 +       .error_remove_page      = aufs_error_remove_page,
14919 +       .swap_activate          = aufs_swap_activate,
14920 +       .swap_deactivate        = aufs_swap_deactivate
14921 +#endif /* CONFIG_AUFS_DEBUG */
14922 +};
14923 diff -urN /usr/share/empty/fs/aufs/file.h linux/fs/aufs/file.h
14924 --- /usr/share/empty/fs/aufs/file.h     1970-01-01 01:00:00.000000000 +0100
14925 +++ linux/fs/aufs/file.h        2021-12-03 15:38:59.936647310 +0100
14926 @@ -0,0 +1,342 @@
14927 +/* SPDX-License-Identifier: GPL-2.0 */
14928 +/*
14929 + * Copyright (C) 2005-2021 Junjiro R. Okajima
14930 + *
14931 + * This program, aufs is free software; you can redistribute it and/or modify
14932 + * it under the terms of the GNU General Public License as published by
14933 + * the Free Software Foundation; either version 2 of the License, or
14934 + * (at your option) any later version.
14935 + *
14936 + * This program is distributed in the hope that it will be useful,
14937 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14938 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14939 + * GNU General Public License for more details.
14940 + *
14941 + * You should have received a copy of the GNU General Public License
14942 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14943 + */
14944 +
14945 +/*
14946 + * file operations
14947 + */
14948 +
14949 +#ifndef __AUFS_FILE_H__
14950 +#define __AUFS_FILE_H__
14951 +
14952 +#ifdef __KERNEL__
14953 +
14954 +#include <linux/file.h>
14955 +#include <linux/fs.h>
14956 +#include <linux/mm_types.h>
14957 +#include <linux/poll.h>
14958 +#include "rwsem.h"
14959 +
14960 +struct au_branch;
14961 +struct au_hfile {
14962 +       struct file             *hf_file;
14963 +       struct au_branch        *hf_br;
14964 +};
14965 +
14966 +struct au_vdir;
14967 +struct au_fidir {
14968 +       aufs_bindex_t           fd_bbot;
14969 +       aufs_bindex_t           fd_nent;
14970 +       struct au_vdir          *fd_vdir_cache;
14971 +       struct au_hfile         fd_hfile[];
14972 +};
14973 +
14974 +static inline int au_fidir_sz(int nent)
14975 +{
14976 +       AuDebugOn(nent < 0);
14977 +       return sizeof(struct au_fidir) + sizeof(struct au_hfile) * nent;
14978 +}
14979 +
14980 +struct au_finfo {
14981 +       atomic_t                fi_generation;
14982 +
14983 +       struct au_rwsem         fi_rwsem;
14984 +       aufs_bindex_t           fi_btop;
14985 +
14986 +       /* do not union them */
14987 +       struct {                                /* for non-dir */
14988 +               struct au_hfile                 fi_htop;
14989 +               atomic_t                        fi_mmapped;
14990 +       };
14991 +       struct au_fidir         *fi_hdir;       /* for dir only */
14992 +
14993 +       struct hlist_bl_node    fi_hlist;
14994 +       struct file             *fi_file;       /* very ugly */
14995 +       struct rcu_head         rcu;
14996 +} ____cacheline_aligned_in_smp;
14997 +
14998 +/* ---------------------------------------------------------------------- */
14999 +
15000 +/* file.c */
15001 +extern const struct address_space_operations aufs_aop;
15002 +unsigned int au_file_roflags(unsigned int flags);
15003 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
15004 +                      struct file *file, int force_wr);
15005 +struct au_do_open_args {
15006 +       int             aopen;
15007 +       int             (*open)(struct file *file, int flags,
15008 +                               struct file *h_file);
15009 +       struct au_fidir *fidir;
15010 +       struct file     *h_file;
15011 +};
15012 +int au_do_open(struct file *file, struct au_do_open_args *args);
15013 +int au_reopen_nondir(struct file *file);
15014 +struct au_pin;
15015 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin);
15016 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
15017 +                         int wlock, unsigned int fi_lsc);
15018 +int au_do_flush(struct file *file, fl_owner_t id,
15019 +               int (*flush)(struct file *file, fl_owner_t id));
15020 +
15021 +/* poll.c */
15022 +#ifdef CONFIG_AUFS_POLL
15023 +__poll_t aufs_poll(struct file *file, struct poll_table_struct *pt);
15024 +#endif
15025 +
15026 +#ifdef CONFIG_AUFS_BR_HFSPLUS
15027 +/* hfsplus.c */
15028 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
15029 +                          int force_wr);
15030 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
15031 +                   struct file *h_file);
15032 +#else
15033 +AuStub(struct file *, au_h_open_pre, return NULL, struct dentry *dentry,
15034 +       aufs_bindex_t bindex, int force_wr)
15035 +AuStubVoid(au_h_open_post, struct dentry *dentry, aufs_bindex_t bindex,
15036 +          struct file *h_file);
15037 +#endif
15038 +
15039 +/* f_op.c */
15040 +extern const struct file_operations aufs_file_fop;
15041 +int au_do_open_nondir(struct file *file, int flags, struct file *h_file);
15042 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file);
15043 +struct file *au_read_pre(struct file *file, int keep_fi, unsigned int lsc);
15044 +
15045 +/* finfo.c */
15046 +void au_hfput(struct au_hfile *hf, int execed);
15047 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex,
15048 +                  struct file *h_file);
15049 +
15050 +void au_update_figen(struct file *file);
15051 +struct au_fidir *au_fidir_alloc(struct super_block *sb);
15052 +int au_fidir_realloc(struct au_finfo *finfo, int nbr, int may_shrink);
15053 +
15054 +void au_fi_init_once(void *_fi);
15055 +void au_finfo_fin(struct file *file);
15056 +int au_finfo_init(struct file *file, struct au_fidir *fidir);
15057 +
15058 +/* ioctl.c */
15059 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg);
15060 +#ifdef CONFIG_COMPAT
15061 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
15062 +                          unsigned long arg);
15063 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
15064 +                             unsigned long arg);
15065 +#endif
15066 +
15067 +/* ---------------------------------------------------------------------- */
15068 +
15069 +static inline struct au_finfo *au_fi(struct file *file)
15070 +{
15071 +       return file->private_data;
15072 +}
15073 +
15074 +/* ---------------------------------------------------------------------- */
15075 +
15076 +#define fi_read_lock(f)        au_rw_read_lock(&au_fi(f)->fi_rwsem)
15077 +#define fi_write_lock(f)       au_rw_write_lock(&au_fi(f)->fi_rwsem)
15078 +#define fi_read_trylock(f)     au_rw_read_trylock(&au_fi(f)->fi_rwsem)
15079 +#define fi_write_trylock(f)    au_rw_write_trylock(&au_fi(f)->fi_rwsem)
15080 +/*
15081 +#define fi_read_trylock_nested(f) \
15082 +       au_rw_read_trylock_nested(&au_fi(f)->fi_rwsem)
15083 +#define fi_write_trylock_nested(f) \
15084 +       au_rw_write_trylock_nested(&au_fi(f)->fi_rwsem)
15085 +*/
15086 +
15087 +#define fi_read_unlock(f)      au_rw_read_unlock(&au_fi(f)->fi_rwsem)
15088 +#define fi_write_unlock(f)     au_rw_write_unlock(&au_fi(f)->fi_rwsem)
15089 +#define fi_downgrade_lock(f)   au_rw_dgrade_lock(&au_fi(f)->fi_rwsem)
15090 +
15091 +/* lock subclass for finfo */
15092 +enum {
15093 +       AuLsc_FI_1,
15094 +       AuLsc_FI_2
15095 +};
15096 +
15097 +static inline void fi_read_lock_nested(struct file *f, unsigned int lsc)
15098 +{
15099 +       au_rw_read_lock_nested(&au_fi(f)->fi_rwsem, lsc);
15100 +}
15101 +
15102 +static inline void fi_write_lock_nested(struct file *f, unsigned int lsc)
15103 +{
15104 +       au_rw_write_lock_nested(&au_fi(f)->fi_rwsem, lsc);
15105 +}
15106 +
15107 +/*
15108 + * fi_read_lock_1, fi_write_lock_1,
15109 + * fi_read_lock_2, fi_write_lock_2
15110 + */
15111 +#define AuReadLockFunc(name) \
15112 +static inline void fi_read_lock_##name(struct file *f) \
15113 +{ fi_read_lock_nested(f, AuLsc_FI_##name); }
15114 +
15115 +#define AuWriteLockFunc(name) \
15116 +static inline void fi_write_lock_##name(struct file *f) \
15117 +{ fi_write_lock_nested(f, AuLsc_FI_##name); }
15118 +
15119 +#define AuRWLockFuncs(name) \
15120 +       AuReadLockFunc(name) \
15121 +       AuWriteLockFunc(name)
15122 +
15123 +AuRWLockFuncs(1);
15124 +AuRWLockFuncs(2);
15125 +
15126 +#undef AuReadLockFunc
15127 +#undef AuWriteLockFunc
15128 +#undef AuRWLockFuncs
15129 +
15130 +#define FiMustNoWaiters(f)     AuRwMustNoWaiters(&au_fi(f)->fi_rwsem)
15131 +#define FiMustAnyLock(f)       AuRwMustAnyLock(&au_fi(f)->fi_rwsem)
15132 +#define FiMustWriteLock(f)     AuRwMustWriteLock(&au_fi(f)->fi_rwsem)
15133 +
15134 +/* ---------------------------------------------------------------------- */
15135 +
15136 +/* todo: hard/soft set? */
15137 +static inline aufs_bindex_t au_fbtop(struct file *file)
15138 +{
15139 +       FiMustAnyLock(file);
15140 +       return au_fi(file)->fi_btop;
15141 +}
15142 +
15143 +static inline aufs_bindex_t au_fbbot_dir(struct file *file)
15144 +{
15145 +       FiMustAnyLock(file);
15146 +       AuDebugOn(!au_fi(file)->fi_hdir);
15147 +       return au_fi(file)->fi_hdir->fd_bbot;
15148 +}
15149 +
15150 +static inline struct au_vdir *au_fvdir_cache(struct file *file)
15151 +{
15152 +       FiMustAnyLock(file);
15153 +       AuDebugOn(!au_fi(file)->fi_hdir);
15154 +       return au_fi(file)->fi_hdir->fd_vdir_cache;
15155 +}
15156 +
15157 +static inline void au_set_fbtop(struct file *file, aufs_bindex_t bindex)
15158 +{
15159 +       FiMustWriteLock(file);
15160 +       au_fi(file)->fi_btop = bindex;
15161 +}
15162 +
15163 +static inline void au_set_fbbot_dir(struct file *file, aufs_bindex_t bindex)
15164 +{
15165 +       FiMustWriteLock(file);
15166 +       AuDebugOn(!au_fi(file)->fi_hdir);
15167 +       au_fi(file)->fi_hdir->fd_bbot = bindex;
15168 +}
15169 +
15170 +static inline void au_set_fvdir_cache(struct file *file,
15171 +                                     struct au_vdir *vdir_cache)
15172 +{
15173 +       FiMustWriteLock(file);
15174 +       AuDebugOn(!au_fi(file)->fi_hdir);
15175 +       au_fi(file)->fi_hdir->fd_vdir_cache = vdir_cache;
15176 +}
15177 +
15178 +static inline struct file *au_hf_top(struct file *file)
15179 +{
15180 +       FiMustAnyLock(file);
15181 +       AuDebugOn(au_fi(file)->fi_hdir);
15182 +       return au_fi(file)->fi_htop.hf_file;
15183 +}
15184 +
15185 +static inline struct file *au_hf_dir(struct file *file, aufs_bindex_t bindex)
15186 +{
15187 +       FiMustAnyLock(file);
15188 +       AuDebugOn(!au_fi(file)->fi_hdir);
15189 +       return au_fi(file)->fi_hdir->fd_hfile[0 + bindex].hf_file;
15190 +}
15191 +
15192 +/* todo: memory barrier? */
15193 +static inline unsigned int au_figen(struct file *f)
15194 +{
15195 +       return atomic_read(&au_fi(f)->fi_generation);
15196 +}
15197 +
15198 +static inline void au_set_mmapped(struct file *f)
15199 +{
15200 +       if (atomic_inc_return(&au_fi(f)->fi_mmapped))
15201 +               return;
15202 +       pr_warn("fi_mmapped wrapped around\n");
15203 +       while (!atomic_inc_return(&au_fi(f)->fi_mmapped))
15204 +               ;
15205 +}
15206 +
15207 +static inline void au_unset_mmapped(struct file *f)
15208 +{
15209 +       atomic_dec(&au_fi(f)->fi_mmapped);
15210 +}
15211 +
15212 +static inline int au_test_mmapped(struct file *f)
15213 +{
15214 +       return atomic_read(&au_fi(f)->fi_mmapped);
15215 +}
15216 +
15217 +/* customize vma->vm_file */
15218 +
15219 +static inline void au_do_vm_file_reset(struct vm_area_struct *vma,
15220 +                                      struct file *file)
15221 +{
15222 +       struct file *f;
15223 +
15224 +       f = vma->vm_file;
15225 +       get_file(file);
15226 +       vma->vm_file = file;
15227 +       fput(f);
15228 +}
15229 +
15230 +#ifdef CONFIG_MMU
15231 +#define AuDbgVmRegion(file, vma) do {} while (0)
15232 +
15233 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
15234 +                                   struct file *file)
15235 +{
15236 +       au_do_vm_file_reset(vma, file);
15237 +}
15238 +#else
15239 +#define AuDbgVmRegion(file, vma) \
15240 +       AuDebugOn((vma)->vm_region && (vma)->vm_region->vm_file != (file))
15241 +
15242 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
15243 +                                   struct file *file)
15244 +{
15245 +       struct file *f;
15246 +
15247 +       au_do_vm_file_reset(vma, file);
15248 +       f = vma->vm_region->vm_file;
15249 +       get_file(file);
15250 +       vma->vm_region->vm_file = file;
15251 +       fput(f);
15252 +}
15253 +#endif /* CONFIG_MMU */
15254 +
15255 +/* handle vma->vm_prfile */
15256 +static inline void au_vm_prfile_set(struct vm_area_struct *vma,
15257 +                                   struct file *file)
15258 +{
15259 +       get_file(file);
15260 +       vma->vm_prfile = file;
15261 +#ifndef CONFIG_MMU
15262 +       get_file(file);
15263 +       vma->vm_region->vm_prfile = file;
15264 +#endif
15265 +}
15266 +
15267 +#endif /* __KERNEL__ */
15268 +#endif /* __AUFS_FILE_H__ */
15269 diff -urN /usr/share/empty/fs/aufs/finfo.c linux/fs/aufs/finfo.c
15270 --- /usr/share/empty/fs/aufs/finfo.c    1970-01-01 01:00:00.000000000 +0100
15271 +++ linux/fs/aufs/finfo.c       2021-12-03 15:38:59.936647310 +0100
15272 @@ -0,0 +1,149 @@
15273 +// SPDX-License-Identifier: GPL-2.0
15274 +/*
15275 + * Copyright (C) 2005-2021 Junjiro R. Okajima
15276 + *
15277 + * This program, aufs is free software; you can redistribute it and/or modify
15278 + * it under the terms of the GNU General Public License as published by
15279 + * the Free Software Foundation; either version 2 of the License, or
15280 + * (at your option) any later version.
15281 + *
15282 + * This program is distributed in the hope that it will be useful,
15283 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15284 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15285 + * GNU General Public License for more details.
15286 + *
15287 + * You should have received a copy of the GNU General Public License
15288 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15289 + */
15290 +
15291 +/*
15292 + * file private data
15293 + */
15294 +
15295 +#include "aufs.h"
15296 +
15297 +void au_hfput(struct au_hfile *hf, int execed)
15298 +{
15299 +       if (execed)
15300 +               allow_write_access(hf->hf_file);
15301 +       fput(hf->hf_file);
15302 +       hf->hf_file = NULL;
15303 +       au_lcnt_dec(&hf->hf_br->br_nfiles);
15304 +       hf->hf_br = NULL;
15305 +}
15306 +
15307 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex, struct file *val)
15308 +{
15309 +       struct au_finfo *finfo = au_fi(file);
15310 +       struct au_hfile *hf;
15311 +       struct au_fidir *fidir;
15312 +
15313 +       fidir = finfo->fi_hdir;
15314 +       if (!fidir) {
15315 +               AuDebugOn(finfo->fi_btop != bindex);
15316 +               hf = &finfo->fi_htop;
15317 +       } else
15318 +               hf = fidir->fd_hfile + bindex;
15319 +
15320 +       if (hf && hf->hf_file)
15321 +               au_hfput(hf, vfsub_file_execed(file));
15322 +       if (val) {
15323 +               FiMustWriteLock(file);
15324 +               AuDebugOn(IS_ERR_OR_NULL(file->f_path.dentry));
15325 +               hf->hf_file = val;
15326 +               hf->hf_br = au_sbr(file->f_path.dentry->d_sb, bindex);
15327 +       }
15328 +}
15329 +
15330 +void au_update_figen(struct file *file)
15331 +{
15332 +       atomic_set(&au_fi(file)->fi_generation, au_digen(file->f_path.dentry));
15333 +       /* smp_mb(); */ /* atomic_set */
15334 +}
15335 +
15336 +/* ---------------------------------------------------------------------- */
15337 +
15338 +struct au_fidir *au_fidir_alloc(struct super_block *sb)
15339 +{
15340 +       struct au_fidir *fidir;
15341 +       int nbr;
15342 +
15343 +       nbr = au_sbbot(sb) + 1;
15344 +       if (nbr < 2)
15345 +               nbr = 2; /* initial allocate for 2 branches */
15346 +       fidir = kzalloc(au_fidir_sz(nbr), GFP_NOFS);
15347 +       if (fidir) {
15348 +               fidir->fd_bbot = -1;
15349 +               fidir->fd_nent = nbr;
15350 +       }
15351 +
15352 +       return fidir;
15353 +}
15354 +
15355 +int au_fidir_realloc(struct au_finfo *finfo, int nbr, int may_shrink)
15356 +{
15357 +       int err;
15358 +       struct au_fidir *fidir, *p;
15359 +
15360 +       AuRwMustWriteLock(&finfo->fi_rwsem);
15361 +       fidir = finfo->fi_hdir;
15362 +       AuDebugOn(!fidir);
15363 +
15364 +       err = -ENOMEM;
15365 +       p = au_kzrealloc(fidir, au_fidir_sz(fidir->fd_nent), au_fidir_sz(nbr),
15366 +                        GFP_NOFS, may_shrink);
15367 +       if (p) {
15368 +               p->fd_nent = nbr;
15369 +               finfo->fi_hdir = p;
15370 +               err = 0;
15371 +       }
15372 +
15373 +       return err;
15374 +}
15375 +
15376 +/* ---------------------------------------------------------------------- */
15377 +
15378 +void au_finfo_fin(struct file *file)
15379 +{
15380 +       struct au_finfo *finfo;
15381 +
15382 +       au_lcnt_dec(&au_sbi(file->f_path.dentry->d_sb)->si_nfiles);
15383 +
15384 +       finfo = au_fi(file);
15385 +       AuDebugOn(finfo->fi_hdir);
15386 +       AuRwDestroy(&finfo->fi_rwsem);
15387 +       au_cache_free_finfo(finfo);
15388 +}
15389 +
15390 +void au_fi_init_once(void *_finfo)
15391 +{
15392 +       struct au_finfo *finfo = _finfo;
15393 +
15394 +       au_rw_init(&finfo->fi_rwsem);
15395 +}
15396 +
15397 +int au_finfo_init(struct file *file, struct au_fidir *fidir)
15398 +{
15399 +       int err;
15400 +       struct au_finfo *finfo;
15401 +       struct dentry *dentry;
15402 +
15403 +       err = -ENOMEM;
15404 +       dentry = file->f_path.dentry;
15405 +       finfo = au_cache_alloc_finfo();
15406 +       if (unlikely(!finfo))
15407 +               goto out;
15408 +
15409 +       err = 0;
15410 +       au_lcnt_inc(&au_sbi(dentry->d_sb)->si_nfiles);
15411 +       au_rw_write_lock(&finfo->fi_rwsem);
15412 +       finfo->fi_btop = -1;
15413 +       finfo->fi_hdir = fidir;
15414 +       atomic_set(&finfo->fi_generation, au_digen(dentry));
15415 +       /* smp_mb(); */ /* atomic_set */
15416 +
15417 +       file->private_data = finfo;
15418 +
15419 +out:
15420 +       return err;
15421 +}
15422 diff -urN /usr/share/empty/fs/aufs/f_op.c linux/fs/aufs/f_op.c
15423 --- /usr/share/empty/fs/aufs/f_op.c     1970-01-01 01:00:00.000000000 +0100
15424 +++ linux/fs/aufs/f_op.c        2021-12-03 15:38:59.933313976 +0100
15425 @@ -0,0 +1,771 @@
15426 +// SPDX-License-Identifier: GPL-2.0
15427 +/*
15428 + * Copyright (C) 2005-2021 Junjiro R. Okajima
15429 + *
15430 + * This program, aufs is free software; you can redistribute it and/or modify
15431 + * it under the terms of the GNU General Public License as published by
15432 + * the Free Software Foundation; either version 2 of the License, or
15433 + * (at your option) any later version.
15434 + *
15435 + * This program is distributed in the hope that it will be useful,
15436 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15437 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15438 + * GNU General Public License for more details.
15439 + *
15440 + * You should have received a copy of the GNU General Public License
15441 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15442 + */
15443 +
15444 +/*
15445 + * file and vm operations
15446 + */
15447 +
15448 +#include <linux/aio.h>
15449 +#include <linux/fs_stack.h>
15450 +#include <linux/mman.h>
15451 +#include <linux/security.h>
15452 +#include "aufs.h"
15453 +
15454 +int au_do_open_nondir(struct file *file, int flags, struct file *h_file)
15455 +{
15456 +       int err;
15457 +       aufs_bindex_t bindex;
15458 +       struct dentry *dentry, *h_dentry;
15459 +       struct au_finfo *finfo;
15460 +       struct inode *h_inode;
15461 +
15462 +       FiMustWriteLock(file);
15463 +
15464 +       err = 0;
15465 +       dentry = file->f_path.dentry;
15466 +       AuDebugOn(IS_ERR_OR_NULL(dentry));
15467 +       finfo = au_fi(file);
15468 +       memset(&finfo->fi_htop, 0, sizeof(finfo->fi_htop));
15469 +       atomic_set(&finfo->fi_mmapped, 0);
15470 +       bindex = au_dbtop(dentry);
15471 +       if (!h_file) {
15472 +               h_dentry = au_h_dptr(dentry, bindex);
15473 +               err = vfsub_test_mntns(file->f_path.mnt, h_dentry->d_sb);
15474 +               if (unlikely(err))
15475 +                       goto out;
15476 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
15477 +               if (IS_ERR(h_file)) {
15478 +                       err = PTR_ERR(h_file);
15479 +                       goto out;
15480 +               }
15481 +       } else {
15482 +               h_dentry = h_file->f_path.dentry;
15483 +               err = vfsub_test_mntns(file->f_path.mnt, h_dentry->d_sb);
15484 +               if (unlikely(err))
15485 +                       goto out;
15486 +               /* br ref is already inc-ed */
15487 +       }
15488 +
15489 +       if ((flags & __O_TMPFILE)
15490 +           && !(flags & O_EXCL)) {
15491 +               h_inode = file_inode(h_file);
15492 +               spin_lock(&h_inode->i_lock);
15493 +               h_inode->i_state |= I_LINKABLE;
15494 +               spin_unlock(&h_inode->i_lock);
15495 +       }
15496 +       au_set_fbtop(file, bindex);
15497 +       au_set_h_fptr(file, bindex, h_file);
15498 +       au_update_figen(file);
15499 +       /* todo: necessary? */
15500 +       /* file->f_ra = h_file->f_ra; */
15501 +
15502 +out:
15503 +       return err;
15504 +}
15505 +
15506 +static int aufs_open_nondir(struct inode *inode __maybe_unused,
15507 +                           struct file *file)
15508 +{
15509 +       int err;
15510 +       struct super_block *sb;
15511 +       struct au_do_open_args args = {
15512 +               .open   = au_do_open_nondir
15513 +       };
15514 +
15515 +       AuDbg("%pD, f_flags 0x%x, f_mode 0x%x\n",
15516 +             file, vfsub_file_flags(file), file->f_mode);
15517 +
15518 +       sb = file->f_path.dentry->d_sb;
15519 +       si_read_lock(sb, AuLock_FLUSH);
15520 +       err = au_do_open(file, &args);
15521 +       si_read_unlock(sb);
15522 +       return err;
15523 +}
15524 +
15525 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file)
15526 +{
15527 +       struct au_finfo *finfo;
15528 +       aufs_bindex_t bindex;
15529 +
15530 +       finfo = au_fi(file);
15531 +       au_hbl_del(&finfo->fi_hlist,
15532 +                  &au_sbi(file->f_path.dentry->d_sb)->si_files);
15533 +       bindex = finfo->fi_btop;
15534 +       if (bindex >= 0)
15535 +               au_set_h_fptr(file, bindex, NULL);
15536 +
15537 +       au_finfo_fin(file);
15538 +       return 0;
15539 +}
15540 +
15541 +/* ---------------------------------------------------------------------- */
15542 +
15543 +static int au_do_flush_nondir(struct file *file, fl_owner_t id)
15544 +{
15545 +       int err;
15546 +       struct file *h_file;
15547 +
15548 +       err = 0;
15549 +       h_file = au_hf_top(file);
15550 +       if (h_file)
15551 +               err = vfsub_flush(h_file, id);
15552 +       return err;
15553 +}
15554 +
15555 +static int aufs_flush_nondir(struct file *file, fl_owner_t id)
15556 +{
15557 +       return au_do_flush(file, id, au_do_flush_nondir);
15558 +}
15559 +
15560 +/* ---------------------------------------------------------------------- */
15561 +/*
15562 + * read and write functions acquire [fdi]_rwsem once, but release before
15563 + * mmap_sem. This is because to stop a race condition between mmap(2).
15564 + * Releasing these aufs-rwsem should be safe, no branch-management (by keeping
15565 + * si_rwsem), no harmful copy-up should happen. Actually copy-up may happen in
15566 + * read functions after [fdi]_rwsem are released, but it should be harmless.
15567 + */
15568 +
15569 +/* Callers should call au_read_post() or fput() in the end */
15570 +struct file *au_read_pre(struct file *file, int keep_fi, unsigned int lsc)
15571 +{
15572 +       struct file *h_file;
15573 +       int err;
15574 +
15575 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0, lsc);
15576 +       if (!err) {
15577 +               di_read_unlock(file->f_path.dentry, AuLock_IR);
15578 +               h_file = au_hf_top(file);
15579 +               get_file(h_file);
15580 +               if (!keep_fi)
15581 +                       fi_read_unlock(file);
15582 +       } else
15583 +               h_file = ERR_PTR(err);
15584 +
15585 +       return h_file;
15586 +}
15587 +
15588 +static void au_read_post(struct inode *inode, struct file *h_file)
15589 +{
15590 +       /* update without lock, I don't think it a problem */
15591 +       fsstack_copy_attr_atime(inode, file_inode(h_file));
15592 +       fput(h_file);
15593 +}
15594 +
15595 +struct au_write_pre {
15596 +       /* input */
15597 +       unsigned int lsc;
15598 +
15599 +       /* output */
15600 +       blkcnt_t blks;
15601 +       aufs_bindex_t btop;
15602 +};
15603 +
15604 +/*
15605 + * return with iinfo is write-locked
15606 + * callers should call au_write_post() or iinfo_write_unlock() + fput() in the
15607 + * end
15608 + */
15609 +static struct file *au_write_pre(struct file *file, int do_ready,
15610 +                                struct au_write_pre *wpre)
15611 +{
15612 +       struct file *h_file;
15613 +       struct dentry *dentry;
15614 +       int err;
15615 +       unsigned int lsc;
15616 +       struct au_pin pin;
15617 +
15618 +       lsc = 0;
15619 +       if (wpre)
15620 +               lsc = wpre->lsc;
15621 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1, lsc);
15622 +       h_file = ERR_PTR(err);
15623 +       if (unlikely(err))
15624 +               goto out;
15625 +
15626 +       dentry = file->f_path.dentry;
15627 +       if (do_ready) {
15628 +               err = au_ready_to_write(file, -1, &pin);
15629 +               if (unlikely(err)) {
15630 +                       h_file = ERR_PTR(err);
15631 +                       di_write_unlock(dentry);
15632 +                       goto out_fi;
15633 +               }
15634 +       }
15635 +
15636 +       di_downgrade_lock(dentry, /*flags*/0);
15637 +       if (wpre)
15638 +               wpre->btop = au_fbtop(file);
15639 +       h_file = au_hf_top(file);
15640 +       get_file(h_file);
15641 +       if (wpre)
15642 +               wpre->blks = file_inode(h_file)->i_blocks;
15643 +       if (do_ready)
15644 +               au_unpin(&pin);
15645 +       di_read_unlock(dentry, /*flags*/0);
15646 +
15647 +out_fi:
15648 +       fi_write_unlock(file);
15649 +out:
15650 +       return h_file;
15651 +}
15652 +
15653 +static void au_write_post(struct inode *inode, struct file *h_file,
15654 +                         struct au_write_pre *wpre, ssize_t written)
15655 +{
15656 +       struct inode *h_inode;
15657 +
15658 +       au_cpup_attr_timesizes(inode);
15659 +       AuDebugOn(au_ibtop(inode) != wpre->btop);
15660 +       h_inode = file_inode(h_file);
15661 +       inode->i_mode = h_inode->i_mode;
15662 +       ii_write_unlock(inode);
15663 +       /* AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks); */
15664 +       if (written > 0)
15665 +               au_fhsm_wrote(inode->i_sb, wpre->btop,
15666 +                             /*force*/h_inode->i_blocks > wpre->blks);
15667 +       fput(h_file);
15668 +}
15669 +
15670 +/*
15671 + * todo: very ugly
15672 + * it locks both of i_mutex and si_rwsem for read in safe.
15673 + * if the plink maintenance mode continues forever (that is the problem),
15674 + * may loop forever.
15675 + */
15676 +static void au_mtx_and_read_lock(struct inode *inode)
15677 +{
15678 +       int err;
15679 +       struct super_block *sb = inode->i_sb;
15680 +
15681 +       while (1) {
15682 +               inode_lock(inode);
15683 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
15684 +               if (!err)
15685 +                       break;
15686 +               inode_unlock(inode);
15687 +               si_read_lock(sb, AuLock_NOPLMW);
15688 +               si_read_unlock(sb);
15689 +       }
15690 +}
15691 +
15692 +static ssize_t au_do_iter(struct file *h_file, int rw, struct kiocb *kio,
15693 +                         struct iov_iter *iov_iter)
15694 +{
15695 +       ssize_t err;
15696 +       struct file *file;
15697 +       ssize_t (*iter)(struct kiocb *, struct iov_iter *);
15698 +
15699 +       err = security_file_permission(h_file, rw);
15700 +       if (unlikely(err))
15701 +               goto out;
15702 +
15703 +       err = -ENOSYS;  /* the branch doesn't have its ->(read|write)_iter() */
15704 +       iter = NULL;
15705 +       if (rw == MAY_READ)
15706 +               iter = h_file->f_op->read_iter;
15707 +       else if (rw == MAY_WRITE)
15708 +               iter = h_file->f_op->write_iter;
15709 +
15710 +       file = kio->ki_filp;
15711 +       kio->ki_filp = h_file;
15712 +       if (iter) {
15713 +               lockdep_off();
15714 +               err = iter(kio, iov_iter);
15715 +               lockdep_on();
15716 +       } else
15717 +               /* currently there is no such fs */
15718 +               WARN_ON_ONCE(1);
15719 +       kio->ki_filp = file;
15720 +
15721 +out:
15722 +       return err;
15723 +}
15724 +
15725 +static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
15726 +{
15727 +       ssize_t err;
15728 +       struct file *file, *h_file;
15729 +       struct inode *inode;
15730 +       struct super_block *sb;
15731 +
15732 +       file = kio->ki_filp;
15733 +       inode = file_inode(file);
15734 +       sb = inode->i_sb;
15735 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
15736 +
15737 +       h_file = au_read_pre(file, /*keep_fi*/1, /*lsc*/0);
15738 +       err = PTR_ERR(h_file);
15739 +       if (IS_ERR(h_file))
15740 +               goto out;
15741 +
15742 +       if (au_test_loopback_kthread()) {
15743 +               au_warn_loopback(h_file->f_path.dentry->d_sb);
15744 +               if (file->f_mapping != h_file->f_mapping) {
15745 +                       file->f_mapping = h_file->f_mapping;
15746 +                       smp_mb(); /* unnecessary? */
15747 +               }
15748 +       }
15749 +       fi_read_unlock(file);
15750 +
15751 +       err = au_do_iter(h_file, MAY_READ, kio, iov_iter);
15752 +       /* todo: necessary? */
15753 +       /* file->f_ra = h_file->f_ra; */
15754 +       au_read_post(inode, h_file);
15755 +
15756 +out:
15757 +       si_read_unlock(sb);
15758 +       return err;
15759 +}
15760 +
15761 +static ssize_t aufs_write_iter(struct kiocb *kio, struct iov_iter *iov_iter)
15762 +{
15763 +       ssize_t err;
15764 +       struct au_write_pre wpre;
15765 +       struct inode *inode;
15766 +       struct file *file, *h_file;
15767 +
15768 +       file = kio->ki_filp;
15769 +       inode = file_inode(file);
15770 +       au_mtx_and_read_lock(inode);
15771 +
15772 +       wpre.lsc = 0;
15773 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
15774 +       err = PTR_ERR(h_file);
15775 +       if (IS_ERR(h_file))
15776 +               goto out;
15777 +
15778 +       err = au_do_iter(h_file, MAY_WRITE, kio, iov_iter);
15779 +       au_write_post(inode, h_file, &wpre, err);
15780 +
15781 +out:
15782 +       si_read_unlock(inode->i_sb);
15783 +       inode_unlock(inode);
15784 +       return err;
15785 +}
15786 +
15787 +/*
15788 + * We may be able to remove aufs_splice_{read,write}() since almost all FSes
15789 + * don't have their own .splice_{read,write} implimentations, and they use
15790 + * generic_file_splice_read() and iter_file_splice_write() who can act like the
15791 + * simple converters to f_op->iter_read() and ->iter_write().
15792 + * But we keep our own implementations because some non-mainlined FSes may have
15793 + * their own .splice_{read,write} implimentations and aufs doesn't want to take
15794 + * away an opportunity to co-work with aufs from them.
15795 + */
15796 +static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
15797 +                               struct pipe_inode_info *pipe, size_t len,
15798 +                               unsigned int flags)
15799 +{
15800 +       ssize_t err;
15801 +       struct file *h_file;
15802 +       struct inode *inode;
15803 +       struct super_block *sb;
15804 +
15805 +       inode = file_inode(file);
15806 +       sb = inode->i_sb;
15807 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
15808 +
15809 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
15810 +       err = PTR_ERR(h_file);
15811 +       if (IS_ERR(h_file))
15812 +               goto out;
15813 +
15814 +       err = vfsub_splice_to(h_file, ppos, pipe, len, flags);
15815 +       /* todo: necessary? */
15816 +       /* file->f_ra = h_file->f_ra; */
15817 +       au_read_post(inode, h_file);
15818 +
15819 +out:
15820 +       si_read_unlock(sb);
15821 +       return err;
15822 +}
15823 +
15824 +static ssize_t
15825 +aufs_splice_write(struct pipe_inode_info *pipe, struct file *file, loff_t *ppos,
15826 +                 size_t len, unsigned int flags)
15827 +{
15828 +       ssize_t err;
15829 +       struct au_write_pre wpre;
15830 +       struct inode *inode;
15831 +       struct file *h_file;
15832 +
15833 +       inode = file_inode(file);
15834 +       au_mtx_and_read_lock(inode);
15835 +
15836 +       wpre.lsc = 0;
15837 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
15838 +       err = PTR_ERR(h_file);
15839 +       if (IS_ERR(h_file))
15840 +               goto out;
15841 +
15842 +       err = vfsub_splice_from(pipe, h_file, ppos, len, flags);
15843 +       au_write_post(inode, h_file, &wpre, err);
15844 +
15845 +out:
15846 +       si_read_unlock(inode->i_sb);
15847 +       inode_unlock(inode);
15848 +       return err;
15849 +}
15850 +
15851 +static long aufs_fallocate(struct file *file, int mode, loff_t offset,
15852 +                          loff_t len)
15853 +{
15854 +       long err;
15855 +       struct au_write_pre wpre;
15856 +       struct inode *inode;
15857 +       struct file *h_file;
15858 +
15859 +       inode = file_inode(file);
15860 +       au_mtx_and_read_lock(inode);
15861 +
15862 +       wpre.lsc = 0;
15863 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
15864 +       err = PTR_ERR(h_file);
15865 +       if (IS_ERR(h_file))
15866 +               goto out;
15867 +
15868 +       lockdep_off();
15869 +       err = vfs_fallocate(h_file, mode, offset, len);
15870 +       lockdep_on();
15871 +       au_write_post(inode, h_file, &wpre, /*written*/1);
15872 +
15873 +out:
15874 +       si_read_unlock(inode->i_sb);
15875 +       inode_unlock(inode);
15876 +       return err;
15877 +}
15878 +
15879 +static ssize_t aufs_copy_file_range(struct file *src, loff_t src_pos,
15880 +                                   struct file *dst, loff_t dst_pos,
15881 +                                   size_t len, unsigned int flags)
15882 +{
15883 +       ssize_t err;
15884 +       struct au_write_pre wpre;
15885 +       enum { SRC, DST };
15886 +       struct {
15887 +               struct inode *inode;
15888 +               struct file *h_file;
15889 +               struct super_block *h_sb;
15890 +       } a[2];
15891 +#define a_src  a[SRC]
15892 +#define a_dst  a[DST]
15893 +
15894 +       err = -EINVAL;
15895 +       a_src.inode = file_inode(src);
15896 +       if (unlikely(!S_ISREG(a_src.inode->i_mode)))
15897 +               goto out;
15898 +       a_dst.inode = file_inode(dst);
15899 +       if (unlikely(!S_ISREG(a_dst.inode->i_mode)))
15900 +               goto out;
15901 +
15902 +       au_mtx_and_read_lock(a_dst.inode);
15903 +       /*
15904 +        * in order to match the order in di_write_lock2_{child,parent}(),
15905 +        * use f_path.dentry for this comparison.
15906 +        */
15907 +       if (src->f_path.dentry < dst->f_path.dentry) {
15908 +               a_src.h_file = au_read_pre(src, /*keep_fi*/1, AuLsc_FI_1);
15909 +               err = PTR_ERR(a_src.h_file);
15910 +               if (IS_ERR(a_src.h_file))
15911 +                       goto out_si;
15912 +
15913 +               wpre.lsc = AuLsc_FI_2;
15914 +               a_dst.h_file = au_write_pre(dst, /*do_ready*/1, &wpre);
15915 +               err = PTR_ERR(a_dst.h_file);
15916 +               if (IS_ERR(a_dst.h_file)) {
15917 +                       au_read_post(a_src.inode, a_src.h_file);
15918 +                       goto out_si;
15919 +               }
15920 +       } else {
15921 +               wpre.lsc = AuLsc_FI_1;
15922 +               a_dst.h_file = au_write_pre(dst, /*do_ready*/1, &wpre);
15923 +               err = PTR_ERR(a_dst.h_file);
15924 +               if (IS_ERR(a_dst.h_file))
15925 +                       goto out_si;
15926 +
15927 +               a_src.h_file = au_read_pre(src, /*keep_fi*/1, AuLsc_FI_2);
15928 +               err = PTR_ERR(a_src.h_file);
15929 +               if (IS_ERR(a_src.h_file)) {
15930 +                       au_write_post(a_dst.inode, a_dst.h_file, &wpre,
15931 +                                     /*written*/0);
15932 +                       goto out_si;
15933 +               }
15934 +       }
15935 +
15936 +       err = -EXDEV;
15937 +       a_src.h_sb = file_inode(a_src.h_file)->i_sb;
15938 +       a_dst.h_sb = file_inode(a_dst.h_file)->i_sb;
15939 +       if (unlikely(a_src.h_sb != a_dst.h_sb)) {
15940 +               AuDbgFile(src);
15941 +               AuDbgFile(dst);
15942 +               goto out_file;
15943 +       }
15944 +
15945 +       err = vfsub_copy_file_range(a_src.h_file, src_pos, a_dst.h_file,
15946 +                                   dst_pos, len, flags);
15947 +
15948 +out_file:
15949 +       au_write_post(a_dst.inode, a_dst.h_file, &wpre, err);
15950 +       fi_read_unlock(src);
15951 +       au_read_post(a_src.inode, a_src.h_file);
15952 +out_si:
15953 +       si_read_unlock(a_dst.inode->i_sb);
15954 +       inode_unlock(a_dst.inode);
15955 +out:
15956 +       return err;
15957 +#undef a_src
15958 +#undef a_dst
15959 +}
15960 +
15961 +/* ---------------------------------------------------------------------- */
15962 +
15963 +/*
15964 + * The locking order around current->mmap_sem.
15965 + * - in most and regular cases
15966 + *   file I/O syscall -- aufs_read() or something
15967 + *     -- si_rwsem for read -- mmap_sem
15968 + *     (Note that [fdi]i_rwsem are released before mmap_sem).
15969 + * - in mmap case
15970 + *   mmap(2) -- mmap_sem -- aufs_mmap() -- si_rwsem for read -- [fdi]i_rwsem
15971 + * This AB-BA order is definitely bad, but is not a problem since "si_rwsem for
15972 + * read" allows multiple processes to acquire it and [fdi]i_rwsem are not held
15973 + * in file I/O. Aufs needs to stop lockdep in aufs_mmap() though.
15974 + * It means that when aufs acquires si_rwsem for write, the process should never
15975 + * acquire mmap_sem.
15976 + *
15977 + * Actually aufs_iterate() holds [fdi]i_rwsem before mmap_sem, but this is not a
15978 + * problem either since any directory is not able to be mmap-ed.
15979 + * The similar scenario is applied to aufs_readlink() too.
15980 + */
15981 +
15982 +#if 0 /* stop calling security_file_mmap() */
15983 +/* cf. linux/include/linux/mman.h: calc_vm_prot_bits() */
15984 +#define AuConv_VM_PROT(f, b)   _calc_vm_trans(f, VM_##b, PROT_##b)
15985 +
15986 +static unsigned long au_arch_prot_conv(unsigned long flags)
15987 +{
15988 +       /* currently ppc64 only */
15989 +#ifdef CONFIG_PPC64
15990 +       /* cf. linux/arch/powerpc/include/asm/mman.h */
15991 +       AuDebugOn(arch_calc_vm_prot_bits(-1) != VM_SAO);
15992 +       return AuConv_VM_PROT(flags, SAO);
15993 +#else
15994 +       AuDebugOn(arch_calc_vm_prot_bits(-1));
15995 +       return 0;
15996 +#endif
15997 +}
15998 +
15999 +static unsigned long au_prot_conv(unsigned long flags)
16000 +{
16001 +       return AuConv_VM_PROT(flags, READ)
16002 +               | AuConv_VM_PROT(flags, WRITE)
16003 +               | AuConv_VM_PROT(flags, EXEC)
16004 +               | au_arch_prot_conv(flags);
16005 +}
16006 +
16007 +/* cf. linux/include/linux/mman.h: calc_vm_flag_bits() */
16008 +#define AuConv_VM_MAP(f, b)    _calc_vm_trans(f, VM_##b, MAP_##b)
16009 +
16010 +static unsigned long au_flag_conv(unsigned long flags)
16011 +{
16012 +       return AuConv_VM_MAP(flags, GROWSDOWN)
16013 +               | AuConv_VM_MAP(flags, DENYWRITE)
16014 +               | AuConv_VM_MAP(flags, LOCKED);
16015 +}
16016 +#endif
16017 +
16018 +static int aufs_mmap(struct file *file, struct vm_area_struct *vma)
16019 +{
16020 +       int err;
16021 +       const unsigned char wlock
16022 +               = (file->f_mode & FMODE_WRITE) && (vma->vm_flags & VM_SHARED);
16023 +       struct super_block *sb;
16024 +       struct file *h_file;
16025 +       struct inode *inode;
16026 +
16027 +       AuDbgVmRegion(file, vma);
16028 +
16029 +       inode = file_inode(file);
16030 +       sb = inode->i_sb;
16031 +       lockdep_off();
16032 +       si_read_lock(sb, AuLock_NOPLMW);
16033 +
16034 +       h_file = au_write_pre(file, wlock, /*wpre*/NULL);
16035 +       lockdep_on();
16036 +       err = PTR_ERR(h_file);
16037 +       if (IS_ERR(h_file))
16038 +               goto out;
16039 +
16040 +       err = 0;
16041 +       au_set_mmapped(file);
16042 +       au_vm_file_reset(vma, h_file);
16043 +       /*
16044 +        * we cannot call security_mmap_file() here since it may acquire
16045 +        * mmap_sem or i_mutex.
16046 +        *
16047 +        * err = security_mmap_file(h_file, au_prot_conv(vma->vm_flags),
16048 +        *                       au_flag_conv(vma->vm_flags));
16049 +        */
16050 +       if (!err)
16051 +               err = call_mmap(h_file, vma);
16052 +       if (!err) {
16053 +               au_vm_prfile_set(vma, file);
16054 +               fsstack_copy_attr_atime(inode, file_inode(h_file));
16055 +               goto out_fput; /* success */
16056 +       }
16057 +       au_unset_mmapped(file);
16058 +       au_vm_file_reset(vma, file);
16059 +
16060 +out_fput:
16061 +       lockdep_off();
16062 +       ii_write_unlock(inode);
16063 +       lockdep_on();
16064 +       fput(h_file);
16065 +out:
16066 +       lockdep_off();
16067 +       si_read_unlock(sb);
16068 +       lockdep_on();
16069 +       AuTraceErr(err);
16070 +       return err;
16071 +}
16072 +
16073 +/* ---------------------------------------------------------------------- */
16074 +
16075 +static int aufs_fsync_nondir(struct file *file, loff_t start, loff_t end,
16076 +                            int datasync)
16077 +{
16078 +       int err;
16079 +       struct au_write_pre wpre;
16080 +       struct inode *inode;
16081 +       struct file *h_file;
16082 +
16083 +       err = 0; /* -EBADF; */ /* posix? */
16084 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
16085 +               goto out;
16086 +
16087 +       inode = file_inode(file);
16088 +       au_mtx_and_read_lock(inode);
16089 +
16090 +       wpre.lsc = 0;
16091 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
16092 +       err = PTR_ERR(h_file);
16093 +       if (IS_ERR(h_file))
16094 +               goto out_unlock;
16095 +
16096 +       err = vfsub_fsync(h_file, &h_file->f_path, datasync);
16097 +       au_write_post(inode, h_file, &wpre, /*written*/0);
16098 +
16099 +out_unlock:
16100 +       si_read_unlock(inode->i_sb);
16101 +       inode_unlock(inode);
16102 +out:
16103 +       return err;
16104 +}
16105 +
16106 +static int aufs_fasync(int fd, struct file *file, int flag)
16107 +{
16108 +       int err;
16109 +       struct file *h_file;
16110 +       struct super_block *sb;
16111 +
16112 +       sb = file->f_path.dentry->d_sb;
16113 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
16114 +
16115 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
16116 +       err = PTR_ERR(h_file);
16117 +       if (IS_ERR(h_file))
16118 +               goto out;
16119 +
16120 +       if (h_file->f_op->fasync)
16121 +               err = h_file->f_op->fasync(fd, h_file, flag);
16122 +       fput(h_file); /* instead of au_read_post() */
16123 +
16124 +out:
16125 +       si_read_unlock(sb);
16126 +       return err;
16127 +}
16128 +
16129 +static int aufs_setfl(struct file *file, unsigned long arg)
16130 +{
16131 +       int err;
16132 +       struct file *h_file;
16133 +       struct super_block *sb;
16134 +
16135 +       sb = file->f_path.dentry->d_sb;
16136 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
16137 +
16138 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
16139 +       err = PTR_ERR(h_file);
16140 +       if (IS_ERR(h_file))
16141 +               goto out;
16142 +
16143 +       /* stop calling h_file->fasync */
16144 +       arg |= vfsub_file_flags(file) & FASYNC;
16145 +       err = setfl(/*unused fd*/-1, h_file, arg);
16146 +       fput(h_file); /* instead of au_read_post() */
16147 +
16148 +out:
16149 +       si_read_unlock(sb);
16150 +       return err;
16151 +}
16152 +
16153 +/* ---------------------------------------------------------------------- */
16154 +
16155 +/* no one supports this operation, currently */
16156 +#if 0 /* reserved for future use */
16157 +static ssize_t aufs_sendpage(struct file *file, struct page *page, int offset,
16158 +                            size_t len, loff_t *pos, int more)
16159 +{
16160 +}
16161 +#endif
16162 +
16163 +/* ---------------------------------------------------------------------- */
16164 +
16165 +const struct file_operations aufs_file_fop = {
16166 +       .owner          = THIS_MODULE,
16167 +
16168 +       .llseek         = default_llseek,
16169 +
16170 +       .read_iter      = aufs_read_iter,
16171 +       .write_iter     = aufs_write_iter,
16172 +
16173 +#ifdef CONFIG_AUFS_POLL
16174 +       .poll           = aufs_poll,
16175 +#endif
16176 +       .unlocked_ioctl = aufs_ioctl_nondir,
16177 +#ifdef CONFIG_COMPAT
16178 +       .compat_ioctl   = aufs_compat_ioctl_nondir,
16179 +#endif
16180 +       .mmap           = aufs_mmap,
16181 +       .open           = aufs_open_nondir,
16182 +       .flush          = aufs_flush_nondir,
16183 +       .release        = aufs_release_nondir,
16184 +       .fsync          = aufs_fsync_nondir,
16185 +       .fasync         = aufs_fasync,
16186 +       /* .sendpage    = aufs_sendpage, */
16187 +       .setfl          = aufs_setfl,
16188 +       .splice_write   = aufs_splice_write,
16189 +       .splice_read    = aufs_splice_read,
16190 +#if 0 /* reserved for future use */
16191 +       .aio_splice_write = aufs_aio_splice_write,
16192 +       .aio_splice_read  = aufs_aio_splice_read,
16193 +#endif
16194 +       .fallocate      = aufs_fallocate,
16195 +       .copy_file_range = aufs_copy_file_range
16196 +};
16197 diff -urN /usr/share/empty/fs/aufs/fstype.h linux/fs/aufs/fstype.h
16198 --- /usr/share/empty/fs/aufs/fstype.h   1970-01-01 01:00:00.000000000 +0100
16199 +++ linux/fs/aufs/fstype.h      2021-12-03 15:38:59.936647310 +0100
16200 @@ -0,0 +1,401 @@
16201 +/* SPDX-License-Identifier: GPL-2.0 */
16202 +/*
16203 + * Copyright (C) 2005-2021 Junjiro R. Okajima
16204 + *
16205 + * This program, aufs is free software; you can redistribute it and/or modify
16206 + * it under the terms of the GNU General Public License as published by
16207 + * the Free Software Foundation; either version 2 of the License, or
16208 + * (at your option) any later version.
16209 + *
16210 + * This program is distributed in the hope that it will be useful,
16211 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16212 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16213 + * GNU General Public License for more details.
16214 + *
16215 + * You should have received a copy of the GNU General Public License
16216 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16217 + */
16218 +
16219 +/*
16220 + * judging filesystem type
16221 + */
16222 +
16223 +#ifndef __AUFS_FSTYPE_H__
16224 +#define __AUFS_FSTYPE_H__
16225 +
16226 +#ifdef __KERNEL__
16227 +
16228 +#include <linux/fs.h>
16229 +#include <linux/magic.h>
16230 +#include <linux/nfs_fs.h>
16231 +#include <linux/romfs_fs.h>
16232 +
16233 +static inline int au_test_aufs(struct super_block *sb)
16234 +{
16235 +       return sb->s_magic == AUFS_SUPER_MAGIC;
16236 +}
16237 +
16238 +static inline const char *au_sbtype(struct super_block *sb)
16239 +{
16240 +       return sb->s_type->name;
16241 +}
16242 +
16243 +static inline int au_test_iso9660(struct super_block *sb __maybe_unused)
16244 +{
16245 +#if IS_ENABLED(CONFIG_ISO9660_FS)
16246 +       return sb->s_magic == ISOFS_SUPER_MAGIC;
16247 +#else
16248 +       return 0;
16249 +#endif
16250 +}
16251 +
16252 +static inline int au_test_romfs(struct super_block *sb __maybe_unused)
16253 +{
16254 +#if IS_ENABLED(CONFIG_ROMFS_FS)
16255 +       return sb->s_magic == ROMFS_MAGIC;
16256 +#else
16257 +       return 0;
16258 +#endif
16259 +}
16260 +
16261 +static inline int au_test_cramfs(struct super_block *sb __maybe_unused)
16262 +{
16263 +#if IS_ENABLED(CONFIG_CRAMFS)
16264 +       return sb->s_magic == CRAMFS_MAGIC;
16265 +#endif
16266 +       return 0;
16267 +}
16268 +
16269 +static inline int au_test_nfs(struct super_block *sb __maybe_unused)
16270 +{
16271 +#if IS_ENABLED(CONFIG_NFS_FS)
16272 +       return sb->s_magic == NFS_SUPER_MAGIC;
16273 +#else
16274 +       return 0;
16275 +#endif
16276 +}
16277 +
16278 +static inline int au_test_fuse(struct super_block *sb __maybe_unused)
16279 +{
16280 +#if IS_ENABLED(CONFIG_FUSE_FS)
16281 +       return sb->s_magic == FUSE_SUPER_MAGIC;
16282 +#else
16283 +       return 0;
16284 +#endif
16285 +}
16286 +
16287 +static inline int au_test_xfs(struct super_block *sb __maybe_unused)
16288 +{
16289 +#if IS_ENABLED(CONFIG_XFS_FS)
16290 +       return sb->s_magic == XFS_SB_MAGIC;
16291 +#else
16292 +       return 0;
16293 +#endif
16294 +}
16295 +
16296 +static inline int au_test_tmpfs(struct super_block *sb __maybe_unused)
16297 +{
16298 +#ifdef CONFIG_TMPFS
16299 +       return sb->s_magic == TMPFS_MAGIC;
16300 +#else
16301 +       return 0;
16302 +#endif
16303 +}
16304 +
16305 +static inline int au_test_ecryptfs(struct super_block *sb __maybe_unused)
16306 +{
16307 +#if IS_ENABLED(CONFIG_ECRYPT_FS)
16308 +       return !strcmp(au_sbtype(sb), "ecryptfs");
16309 +#else
16310 +       return 0;
16311 +#endif
16312 +}
16313 +
16314 +static inline int au_test_ramfs(struct super_block *sb)
16315 +{
16316 +       return sb->s_magic == RAMFS_MAGIC;
16317 +}
16318 +
16319 +static inline int au_test_ubifs(struct super_block *sb __maybe_unused)
16320 +{
16321 +#if IS_ENABLED(CONFIG_UBIFS_FS)
16322 +       return sb->s_magic == UBIFS_SUPER_MAGIC;
16323 +#else
16324 +       return 0;
16325 +#endif
16326 +}
16327 +
16328 +static inline int au_test_procfs(struct super_block *sb __maybe_unused)
16329 +{
16330 +#ifdef CONFIG_PROC_FS
16331 +       return sb->s_magic == PROC_SUPER_MAGIC;
16332 +#else
16333 +       return 0;
16334 +#endif
16335 +}
16336 +
16337 +static inline int au_test_sysfs(struct super_block *sb __maybe_unused)
16338 +{
16339 +#ifdef CONFIG_SYSFS
16340 +       return sb->s_magic == SYSFS_MAGIC;
16341 +#else
16342 +       return 0;
16343 +#endif
16344 +}
16345 +
16346 +static inline int au_test_configfs(struct super_block *sb __maybe_unused)
16347 +{
16348 +#if IS_ENABLED(CONFIG_CONFIGFS_FS)
16349 +       return sb->s_magic == CONFIGFS_MAGIC;
16350 +#else
16351 +       return 0;
16352 +#endif
16353 +}
16354 +
16355 +static inline int au_test_minix(struct super_block *sb __maybe_unused)
16356 +{
16357 +#if IS_ENABLED(CONFIG_MINIX_FS)
16358 +       return sb->s_magic == MINIX3_SUPER_MAGIC
16359 +               || sb->s_magic == MINIX2_SUPER_MAGIC
16360 +               || sb->s_magic == MINIX2_SUPER_MAGIC2
16361 +               || sb->s_magic == MINIX_SUPER_MAGIC
16362 +               || sb->s_magic == MINIX_SUPER_MAGIC2;
16363 +#else
16364 +       return 0;
16365 +#endif
16366 +}
16367 +
16368 +static inline int au_test_fat(struct super_block *sb __maybe_unused)
16369 +{
16370 +#if IS_ENABLED(CONFIG_FAT_FS)
16371 +       return sb->s_magic == MSDOS_SUPER_MAGIC;
16372 +#else
16373 +       return 0;
16374 +#endif
16375 +}
16376 +
16377 +static inline int au_test_msdos(struct super_block *sb)
16378 +{
16379 +       return au_test_fat(sb);
16380 +}
16381 +
16382 +static inline int au_test_vfat(struct super_block *sb)
16383 +{
16384 +       return au_test_fat(sb);
16385 +}
16386 +
16387 +static inline int au_test_securityfs(struct super_block *sb __maybe_unused)
16388 +{
16389 +#ifdef CONFIG_SECURITYFS
16390 +       return sb->s_magic == SECURITYFS_MAGIC;
16391 +#else
16392 +       return 0;
16393 +#endif
16394 +}
16395 +
16396 +static inline int au_test_squashfs(struct super_block *sb __maybe_unused)
16397 +{
16398 +#if IS_ENABLED(CONFIG_SQUASHFS)
16399 +       return sb->s_magic == SQUASHFS_MAGIC;
16400 +#else
16401 +       return 0;
16402 +#endif
16403 +}
16404 +
16405 +static inline int au_test_btrfs(struct super_block *sb __maybe_unused)
16406 +{
16407 +#if IS_ENABLED(CONFIG_BTRFS_FS)
16408 +       return sb->s_magic == BTRFS_SUPER_MAGIC;
16409 +#else
16410 +       return 0;
16411 +#endif
16412 +}
16413 +
16414 +static inline int au_test_xenfs(struct super_block *sb __maybe_unused)
16415 +{
16416 +#if IS_ENABLED(CONFIG_XENFS)
16417 +       return sb->s_magic == XENFS_SUPER_MAGIC;
16418 +#else
16419 +       return 0;
16420 +#endif
16421 +}
16422 +
16423 +static inline int au_test_debugfs(struct super_block *sb __maybe_unused)
16424 +{
16425 +#ifdef CONFIG_DEBUG_FS
16426 +       return sb->s_magic == DEBUGFS_MAGIC;
16427 +#else
16428 +       return 0;
16429 +#endif
16430 +}
16431 +
16432 +static inline int au_test_nilfs(struct super_block *sb __maybe_unused)
16433 +{
16434 +#if IS_ENABLED(CONFIG_NILFS)
16435 +       return sb->s_magic == NILFS_SUPER_MAGIC;
16436 +#else
16437 +       return 0;
16438 +#endif
16439 +}
16440 +
16441 +static inline int au_test_hfsplus(struct super_block *sb __maybe_unused)
16442 +{
16443 +#if IS_ENABLED(CONFIG_HFSPLUS_FS)
16444 +       return sb->s_magic == HFSPLUS_SUPER_MAGIC;
16445 +#else
16446 +       return 0;
16447 +#endif
16448 +}
16449 +
16450 +/* ---------------------------------------------------------------------- */
16451 +/*
16452 + * they can't be an aufs branch.
16453 + */
16454 +static inline int au_test_fs_unsuppoted(struct super_block *sb)
16455 +{
16456 +       return
16457 +#ifndef CONFIG_AUFS_BR_RAMFS
16458 +               au_test_ramfs(sb) ||
16459 +#endif
16460 +               au_test_procfs(sb)
16461 +               || au_test_sysfs(sb)
16462 +               || au_test_configfs(sb)
16463 +               || au_test_debugfs(sb)
16464 +               || au_test_securityfs(sb)
16465 +               || au_test_xenfs(sb)
16466 +               || au_test_ecryptfs(sb)
16467 +               /* || !strcmp(au_sbtype(sb), "unionfs") */
16468 +               || au_test_aufs(sb); /* will be supported in next version */
16469 +}
16470 +
16471 +static inline int au_test_fs_remote(struct super_block *sb)
16472 +{
16473 +       return !au_test_tmpfs(sb)
16474 +#ifdef CONFIG_AUFS_BR_RAMFS
16475 +               && !au_test_ramfs(sb)
16476 +#endif
16477 +               && !(sb->s_type->fs_flags & FS_REQUIRES_DEV);
16478 +}
16479 +
16480 +/* ---------------------------------------------------------------------- */
16481 +
16482 +/*
16483 + * Note: these functions (below) are created after reading ->getattr() in all
16484 + * filesystems under linux/fs. it means we have to do so in every update...
16485 + */
16486 +
16487 +/*
16488 + * some filesystems require getattr to refresh the inode attributes before
16489 + * referencing.
16490 + * in most cases, we can rely on the inode attribute in NFS (or every remote fs)
16491 + * and leave the work for d_revalidate()
16492 + */
16493 +static inline int au_test_fs_refresh_iattr(struct super_block *sb)
16494 +{
16495 +       return au_test_nfs(sb)
16496 +               || au_test_fuse(sb)
16497 +               /* || au_test_btrfs(sb) */      /* untested */
16498 +               ;
16499 +}
16500 +
16501 +/*
16502 + * filesystems which don't maintain i_size or i_blocks.
16503 + */
16504 +static inline int au_test_fs_bad_iattr_size(struct super_block *sb)
16505 +{
16506 +       return au_test_xfs(sb)
16507 +               || au_test_btrfs(sb)
16508 +               || au_test_ubifs(sb)
16509 +               || au_test_hfsplus(sb)  /* maintained, but incorrect */
16510 +               /* || au_test_minix(sb) */      /* untested */
16511 +               ;
16512 +}
16513 +
16514 +/*
16515 + * filesystems which don't store the correct value in some of their inode
16516 + * attributes.
16517 + */
16518 +static inline int au_test_fs_bad_iattr(struct super_block *sb)
16519 +{
16520 +       return au_test_fs_bad_iattr_size(sb)
16521 +               || au_test_fat(sb)
16522 +               || au_test_msdos(sb)
16523 +               || au_test_vfat(sb);
16524 +}
16525 +
16526 +/* they don't check i_nlink in link(2) */
16527 +static inline int au_test_fs_no_limit_nlink(struct super_block *sb)
16528 +{
16529 +       return au_test_tmpfs(sb)
16530 +#ifdef CONFIG_AUFS_BR_RAMFS
16531 +               || au_test_ramfs(sb)
16532 +#endif
16533 +               || au_test_ubifs(sb)
16534 +               || au_test_hfsplus(sb);
16535 +}
16536 +
16537 +/*
16538 + * filesystems which sets S_NOATIME and S_NOCMTIME.
16539 + */
16540 +static inline int au_test_fs_notime(struct super_block *sb)
16541 +{
16542 +       return au_test_nfs(sb)
16543 +               || au_test_fuse(sb)
16544 +               || au_test_ubifs(sb)
16545 +               ;
16546 +}
16547 +
16548 +/* temporary support for i#1 in cramfs */
16549 +static inline int au_test_fs_unique_ino(struct inode *inode)
16550 +{
16551 +       if (au_test_cramfs(inode->i_sb))
16552 +               return inode->i_ino != 1;
16553 +       return 1;
16554 +}
16555 +
16556 +/* ---------------------------------------------------------------------- */
16557 +
16558 +/*
16559 + * the filesystem where the xino files placed must support i/o after unlink and
16560 + * maintain i_size and i_blocks.
16561 + */
16562 +static inline int au_test_fs_bad_xino(struct super_block *sb)
16563 +{
16564 +       return au_test_fs_remote(sb)
16565 +               || au_test_fs_bad_iattr_size(sb)
16566 +               /* don't want unnecessary work for xino */
16567 +               || au_test_aufs(sb)
16568 +               || au_test_ecryptfs(sb)
16569 +               || au_test_nilfs(sb);
16570 +}
16571 +
16572 +static inline int au_test_fs_trunc_xino(struct super_block *sb)
16573 +{
16574 +       return au_test_tmpfs(sb)
16575 +               || au_test_ramfs(sb);
16576 +}
16577 +
16578 +/*
16579 + * test if the @sb is real-readonly.
16580 + */
16581 +static inline int au_test_fs_rr(struct super_block *sb)
16582 +{
16583 +       return au_test_squashfs(sb)
16584 +               || au_test_iso9660(sb)
16585 +               || au_test_cramfs(sb)
16586 +               || au_test_romfs(sb);
16587 +}
16588 +
16589 +/*
16590 + * test if the @inode is nfs with 'noacl' option
16591 + * NFS always sets SB_POSIXACL regardless its mount option 'noacl.'
16592 + */
16593 +static inline int au_test_nfs_noacl(struct inode *inode)
16594 +{
16595 +       return au_test_nfs(inode->i_sb)
16596 +               /* && IS_POSIXACL(inode) */
16597 +               && !nfs_server_capable(inode, NFS_CAP_ACLS);
16598 +}
16599 +
16600 +#endif /* __KERNEL__ */
16601 +#endif /* __AUFS_FSTYPE_H__ */
16602 diff -urN /usr/share/empty/fs/aufs/hbl.h linux/fs/aufs/hbl.h
16603 --- /usr/share/empty/fs/aufs/hbl.h      1970-01-01 01:00:00.000000000 +0100
16604 +++ linux/fs/aufs/hbl.h 2021-12-03 15:38:59.936647310 +0100
16605 @@ -0,0 +1,65 @@
16606 +/* SPDX-License-Identifier: GPL-2.0 */
16607 +/*
16608 + * Copyright (C) 2017-2021 Junjiro R. Okajima
16609 + *
16610 + * This program, aufs is free software; you can redistribute it and/or modify
16611 + * it under the terms of the GNU General Public License as published by
16612 + * the Free Software Foundation; either version 2 of the License, or
16613 + * (at your option) any later version.
16614 + *
16615 + * This program is distributed in the hope that it will be useful,
16616 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16617 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16618 + * GNU General Public License for more details.
16619 + *
16620 + * You should have received a copy of the GNU General Public License
16621 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16622 + */
16623 +
16624 +/*
16625 + * helpers for hlist_bl.h
16626 + */
16627 +
16628 +#ifndef __AUFS_HBL_H__
16629 +#define __AUFS_HBL_H__
16630 +
16631 +#ifdef __KERNEL__
16632 +
16633 +#include <linux/list_bl.h>
16634 +
16635 +static inline void au_hbl_add(struct hlist_bl_node *node,
16636 +                             struct hlist_bl_head *hbl)
16637 +{
16638 +       hlist_bl_lock(hbl);
16639 +       hlist_bl_add_head(node, hbl);
16640 +       hlist_bl_unlock(hbl);
16641 +}
16642 +
16643 +static inline void au_hbl_del(struct hlist_bl_node *node,
16644 +                             struct hlist_bl_head *hbl)
16645 +{
16646 +       hlist_bl_lock(hbl);
16647 +       hlist_bl_del(node);
16648 +       hlist_bl_unlock(hbl);
16649 +}
16650 +
16651 +#define au_hbl_for_each(pos, head)                                     \
16652 +       for (pos = hlist_bl_first(head);                                \
16653 +            pos;                                                       \
16654 +            pos = pos->next)
16655 +
16656 +static inline unsigned long au_hbl_count(struct hlist_bl_head *hbl)
16657 +{
16658 +       unsigned long cnt;
16659 +       struct hlist_bl_node *pos;
16660 +
16661 +       cnt = 0;
16662 +       hlist_bl_lock(hbl);
16663 +       au_hbl_for_each(pos, hbl)
16664 +               cnt++;
16665 +       hlist_bl_unlock(hbl);
16666 +       return cnt;
16667 +}
16668 +
16669 +#endif /* __KERNEL__ */
16670 +#endif /* __AUFS_HBL_H__ */
16671 diff -urN /usr/share/empty/fs/aufs/hfsnotify.c linux/fs/aufs/hfsnotify.c
16672 --- /usr/share/empty/fs/aufs/hfsnotify.c        1970-01-01 01:00:00.000000000 +0100
16673 +++ linux/fs/aufs/hfsnotify.c   2021-12-03 15:38:59.936647310 +0100
16674 @@ -0,0 +1,288 @@
16675 +// SPDX-License-Identifier: GPL-2.0
16676 +/*
16677 + * Copyright (C) 2005-2021 Junjiro R. Okajima
16678 + *
16679 + * This program, aufs is free software; you can redistribute it and/or modify
16680 + * it under the terms of the GNU General Public License as published by
16681 + * the Free Software Foundation; either version 2 of the License, or
16682 + * (at your option) any later version.
16683 + *
16684 + * This program is distributed in the hope that it will be useful,
16685 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16686 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16687 + * GNU General Public License for more details.
16688 + *
16689 + * You should have received a copy of the GNU General Public License
16690 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16691 + */
16692 +
16693 +/*
16694 + * fsnotify for the lower directories
16695 + */
16696 +
16697 +#include "aufs.h"
16698 +
16699 +/* FS_IN_IGNORED is unnecessary */
16700 +static const __u32 AuHfsnMask = (FS_MOVED_TO | FS_MOVED_FROM | FS_DELETE
16701 +                                | FS_CREATE | FS_EVENT_ON_CHILD);
16702 +static DECLARE_WAIT_QUEUE_HEAD(au_hfsn_wq);
16703 +static __cacheline_aligned_in_smp atomic64_t au_hfsn_ifree = ATOMIC64_INIT(0);
16704 +
16705 +static void au_hfsn_free_mark(struct fsnotify_mark *mark)
16706 +{
16707 +       struct au_hnotify *hn = container_of(mark, struct au_hnotify,
16708 +                                            hn_mark);
16709 +       /* AuDbg("here\n"); */
16710 +       au_cache_free_hnotify(hn);
16711 +       smp_mb__before_atomic(); /* for atomic64_dec */
16712 +       if (atomic64_dec_and_test(&au_hfsn_ifree))
16713 +               wake_up(&au_hfsn_wq);
16714 +}
16715 +
16716 +static int au_hfsn_alloc(struct au_hinode *hinode)
16717 +{
16718 +       int err;
16719 +       struct au_hnotify *hn;
16720 +       struct super_block *sb;
16721 +       struct au_branch *br;
16722 +       struct fsnotify_mark *mark;
16723 +       aufs_bindex_t bindex;
16724 +
16725 +       hn = hinode->hi_notify;
16726 +       sb = hn->hn_aufs_inode->i_sb;
16727 +       bindex = au_br_index(sb, hinode->hi_id);
16728 +       br = au_sbr(sb, bindex);
16729 +       AuDebugOn(!br->br_hfsn);
16730 +
16731 +       mark = &hn->hn_mark;
16732 +       fsnotify_init_mark(mark, br->br_hfsn->hfsn_group);
16733 +       mark->mask = AuHfsnMask;
16734 +       /*
16735 +        * by udba rename or rmdir, aufs assign a new inode to the known
16736 +        * h_inode, so specify 1 to allow dups.
16737 +        */
16738 +       lockdep_off();
16739 +       err = fsnotify_add_inode_mark(mark, hinode->hi_inode, /*allow_dups*/1);
16740 +       lockdep_on();
16741 +
16742 +       return err;
16743 +}
16744 +
16745 +static int au_hfsn_free(struct au_hinode *hinode, struct au_hnotify *hn)
16746 +{
16747 +       struct fsnotify_mark *mark;
16748 +       unsigned long long ull;
16749 +       struct fsnotify_group *group;
16750 +
16751 +       ull = atomic64_inc_return(&au_hfsn_ifree);
16752 +       BUG_ON(!ull);
16753 +
16754 +       mark = &hn->hn_mark;
16755 +       spin_lock(&mark->lock);
16756 +       group = mark->group;
16757 +       fsnotify_get_group(group);
16758 +       spin_unlock(&mark->lock);
16759 +       lockdep_off();
16760 +       fsnotify_destroy_mark(mark, group);
16761 +       fsnotify_put_mark(mark);
16762 +       fsnotify_put_group(group);
16763 +       lockdep_on();
16764 +
16765 +       /* free hn by myself */
16766 +       return 0;
16767 +}
16768 +
16769 +/* ---------------------------------------------------------------------- */
16770 +
16771 +static void au_hfsn_ctl(struct au_hinode *hinode, int do_set)
16772 +{
16773 +       struct fsnotify_mark *mark;
16774 +
16775 +       mark = &hinode->hi_notify->hn_mark;
16776 +       spin_lock(&mark->lock);
16777 +       if (do_set) {
16778 +               AuDebugOn(mark->mask & AuHfsnMask);
16779 +               mark->mask |= AuHfsnMask;
16780 +       } else {
16781 +               AuDebugOn(!(mark->mask & AuHfsnMask));
16782 +               mark->mask &= ~AuHfsnMask;
16783 +       }
16784 +       spin_unlock(&mark->lock);
16785 +       /* fsnotify_recalc_inode_mask(hinode->hi_inode); */
16786 +}
16787 +
16788 +/* ---------------------------------------------------------------------- */
16789 +
16790 +/* #define AuDbgHnotify */
16791 +#ifdef AuDbgHnotify
16792 +static char *au_hfsn_name(u32 mask)
16793 +{
16794 +#ifdef CONFIG_AUFS_DEBUG
16795 +#define test_ret(flag)                         \
16796 +       do {                                    \
16797 +               if (mask & flag)                \
16798 +                       return #flag;           \
16799 +       } while (0)
16800 +       test_ret(FS_ACCESS);
16801 +       test_ret(FS_MODIFY);
16802 +       test_ret(FS_ATTRIB);
16803 +       test_ret(FS_CLOSE_WRITE);
16804 +       test_ret(FS_CLOSE_NOWRITE);
16805 +       test_ret(FS_OPEN);
16806 +       test_ret(FS_MOVED_FROM);
16807 +       test_ret(FS_MOVED_TO);
16808 +       test_ret(FS_CREATE);
16809 +       test_ret(FS_DELETE);
16810 +       test_ret(FS_DELETE_SELF);
16811 +       test_ret(FS_MOVE_SELF);
16812 +       test_ret(FS_UNMOUNT);
16813 +       test_ret(FS_Q_OVERFLOW);
16814 +       test_ret(FS_IN_IGNORED);
16815 +       test_ret(FS_ISDIR);
16816 +       test_ret(FS_IN_ONESHOT);
16817 +       test_ret(FS_EVENT_ON_CHILD);
16818 +       return "";
16819 +#undef test_ret
16820 +#else
16821 +       return "??";
16822 +#endif
16823 +}
16824 +#endif
16825 +
16826 +/* ---------------------------------------------------------------------- */
16827 +
16828 +static void au_hfsn_free_group(struct fsnotify_group *group)
16829 +{
16830 +       struct au_br_hfsnotify *hfsn = group->private;
16831 +
16832 +       /* AuDbg("here\n"); */
16833 +       au_kfree_try_rcu(hfsn);
16834 +}
16835 +
16836 +static int au_hfsn_handle_event(struct fsnotify_group *group,
16837 +                               u32 mask, const void *data, int data_type,
16838 +                               struct inode *dir,
16839 +                               const struct qstr *file_name, u32 cookie,
16840 +                               struct fsnotify_iter_info *iter_info)
16841 +{
16842 +       int err;
16843 +       struct au_hnotify *hnotify;
16844 +       struct inode *h_dir, *h_inode;
16845 +       struct fsnotify_mark *inode_mark;
16846 +
16847 +       AuDebugOn(data_type != FSNOTIFY_EVENT_INODE);
16848 +
16849 +       err = 0;
16850 +       /* if FS_UNMOUNT happens, there must be another bug */
16851 +       AuDebugOn(mask & FS_UNMOUNT);
16852 +       if (mask & (FS_IN_IGNORED | FS_UNMOUNT))
16853 +               goto out;
16854 +
16855 +       h_dir = dir;
16856 +       h_inode = NULL;
16857 +#ifdef AuDbgHnotify
16858 +       au_debug_on();
16859 +       if (1 || h_child_qstr.len != sizeof(AUFS_XINO_FNAME) - 1
16860 +           || strncmp(h_child_qstr.name, AUFS_XINO_FNAME, h_child_qstr.len)) {
16861 +               AuDbg("i%lu, mask 0x%x %s, hcname %.*s, hi%lu\n",
16862 +                     h_dir->i_ino, mask, au_hfsn_name(mask),
16863 +                     AuLNPair(&h_child_qstr), h_inode ? h_inode->i_ino : 0);
16864 +               /* WARN_ON(1); */
16865 +       }
16866 +       au_debug_off();
16867 +#endif
16868 +
16869 +       inode_mark = fsnotify_iter_inode_mark(iter_info);
16870 +       AuDebugOn(!inode_mark);
16871 +       hnotify = container_of(inode_mark, struct au_hnotify, hn_mark);
16872 +       err = au_hnotify(h_dir, hnotify, mask, file_name, h_inode);
16873 +
16874 +out:
16875 +       return err;
16876 +}
16877 +
16878 +static struct fsnotify_ops au_hfsn_ops = {
16879 +       .handle_event           = au_hfsn_handle_event,
16880 +       .free_group_priv        = au_hfsn_free_group,
16881 +       .free_mark              = au_hfsn_free_mark
16882 +};
16883 +
16884 +/* ---------------------------------------------------------------------- */
16885 +
16886 +static void au_hfsn_fin_br(struct au_branch *br)
16887 +{
16888 +       struct au_br_hfsnotify *hfsn;
16889 +
16890 +       hfsn = br->br_hfsn;
16891 +       if (hfsn) {
16892 +               lockdep_off();
16893 +               fsnotify_put_group(hfsn->hfsn_group);
16894 +               lockdep_on();
16895 +       }
16896 +}
16897 +
16898 +static int au_hfsn_init_br(struct au_branch *br, int perm)
16899 +{
16900 +       int err;
16901 +       struct fsnotify_group *group;
16902 +       struct au_br_hfsnotify *hfsn;
16903 +
16904 +       err = 0;
16905 +       br->br_hfsn = NULL;
16906 +       if (!au_br_hnotifyable(perm))
16907 +               goto out;
16908 +
16909 +       err = -ENOMEM;
16910 +       hfsn = kmalloc(sizeof(*hfsn), GFP_NOFS);
16911 +       if (unlikely(!hfsn))
16912 +               goto out;
16913 +
16914 +       err = 0;
16915 +       group = fsnotify_alloc_group(&au_hfsn_ops);
16916 +       if (IS_ERR(group)) {
16917 +               err = PTR_ERR(group);
16918 +               pr_err("fsnotify_alloc_group() failed, %d\n", err);
16919 +               goto out_hfsn;
16920 +       }
16921 +
16922 +       group->private = hfsn;
16923 +       hfsn->hfsn_group = group;
16924 +       br->br_hfsn = hfsn;
16925 +       goto out; /* success */
16926 +
16927 +out_hfsn:
16928 +       au_kfree_try_rcu(hfsn);
16929 +out:
16930 +       return err;
16931 +}
16932 +
16933 +static int au_hfsn_reset_br(unsigned int udba, struct au_branch *br, int perm)
16934 +{
16935 +       int err;
16936 +
16937 +       err = 0;
16938 +       if (!br->br_hfsn)
16939 +               err = au_hfsn_init_br(br, perm);
16940 +
16941 +       return err;
16942 +}
16943 +
16944 +/* ---------------------------------------------------------------------- */
16945 +
16946 +static void au_hfsn_fin(void)
16947 +{
16948 +       AuDbg("au_hfsn_ifree %lld\n", (long long)atomic64_read(&au_hfsn_ifree));
16949 +       wait_event(au_hfsn_wq, !atomic64_read(&au_hfsn_ifree));
16950 +}
16951 +
16952 +const struct au_hnotify_op au_hnotify_op = {
16953 +       .ctl            = au_hfsn_ctl,
16954 +       .alloc          = au_hfsn_alloc,
16955 +       .free           = au_hfsn_free,
16956 +
16957 +       .fin            = au_hfsn_fin,
16958 +
16959 +       .reset_br       = au_hfsn_reset_br,
16960 +       .fin_br         = au_hfsn_fin_br,
16961 +       .init_br        = au_hfsn_init_br
16962 +};
16963 diff -urN /usr/share/empty/fs/aufs/hfsplus.c linux/fs/aufs/hfsplus.c
16964 --- /usr/share/empty/fs/aufs/hfsplus.c  1970-01-01 01:00:00.000000000 +0100
16965 +++ linux/fs/aufs/hfsplus.c     2021-12-03 15:38:59.936647310 +0100
16966 @@ -0,0 +1,60 @@
16967 +// SPDX-License-Identifier: GPL-2.0
16968 +/*
16969 + * Copyright (C) 2010-2021 Junjiro R. Okajima
16970 + *
16971 + * This program, aufs is free software; you can redistribute it and/or modify
16972 + * it under the terms of the GNU General Public License as published by
16973 + * the Free Software Foundation; either version 2 of the License, or
16974 + * (at your option) any later version.
16975 + *
16976 + * This program is distributed in the hope that it will be useful,
16977 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16978 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16979 + * GNU General Public License for more details.
16980 + *
16981 + * You should have received a copy of the GNU General Public License
16982 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16983 + */
16984 +
16985 +/*
16986 + * special support for filesystems which acquires an inode mutex
16987 + * at final closing a file, eg, hfsplus.
16988 + *
16989 + * This trick is very simple and stupid, just to open the file before really
16990 + * necessary open to tell hfsplus that this is not the final closing.
16991 + * The caller should call au_h_open_pre() after acquiring the inode mutex,
16992 + * and au_h_open_post() after releasing it.
16993 + */
16994 +
16995 +#include "aufs.h"
16996 +
16997 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
16998 +                          int force_wr)
16999 +{
17000 +       struct file *h_file;
17001 +       struct dentry *h_dentry;
17002 +
17003 +       h_dentry = au_h_dptr(dentry, bindex);
17004 +       AuDebugOn(!h_dentry);
17005 +       AuDebugOn(d_is_negative(h_dentry));
17006 +
17007 +       h_file = NULL;
17008 +       if (au_test_hfsplus(h_dentry->d_sb)
17009 +           && d_is_reg(h_dentry))
17010 +               h_file = au_h_open(dentry, bindex,
17011 +                                  O_RDONLY | O_NOATIME | O_LARGEFILE,
17012 +                                  /*file*/NULL, force_wr);
17013 +       return h_file;
17014 +}
17015 +
17016 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
17017 +                   struct file *h_file)
17018 +{
17019 +       struct au_branch *br;
17020 +
17021 +       if (h_file) {
17022 +               fput(h_file);
17023 +               br = au_sbr(dentry->d_sb, bindex);
17024 +               au_lcnt_dec(&br->br_nfiles);
17025 +       }
17026 +}
17027 diff -urN /usr/share/empty/fs/aufs/hnotify.c linux/fs/aufs/hnotify.c
17028 --- /usr/share/empty/fs/aufs/hnotify.c  1970-01-01 01:00:00.000000000 +0100
17029 +++ linux/fs/aufs/hnotify.c     2021-12-03 15:38:59.936647310 +0100
17030 @@ -0,0 +1,715 @@
17031 +// SPDX-License-Identifier: GPL-2.0
17032 +/*
17033 + * Copyright (C) 2005-2021 Junjiro R. Okajima
17034 + *
17035 + * This program, aufs is free software; you can redistribute it and/or modify
17036 + * it under the terms of the GNU General Public License as published by
17037 + * the Free Software Foundation; either version 2 of the License, or
17038 + * (at your option) any later version.
17039 + *
17040 + * This program is distributed in the hope that it will be useful,
17041 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17042 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17043 + * GNU General Public License for more details.
17044 + *
17045 + * You should have received a copy of the GNU General Public License
17046 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17047 + */
17048 +
17049 +/*
17050 + * abstraction to notify the direct changes on lower directories
17051 + */
17052 +
17053 +/* #include <linux/iversion.h> */
17054 +#include "aufs.h"
17055 +
17056 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode)
17057 +{
17058 +       int err;
17059 +       struct au_hnotify *hn;
17060 +
17061 +       err = -ENOMEM;
17062 +       hn = au_cache_alloc_hnotify();
17063 +       if (hn) {
17064 +               hn->hn_aufs_inode = inode;
17065 +               hinode->hi_notify = hn;
17066 +               err = au_hnotify_op.alloc(hinode);
17067 +               AuTraceErr(err);
17068 +               if (unlikely(err)) {
17069 +                       hinode->hi_notify = NULL;
17070 +                       au_cache_free_hnotify(hn);
17071 +                       /*
17072 +                        * The upper dir was removed by udba, but the same named
17073 +                        * dir left. In this case, aufs assigns a new inode
17074 +                        * number and set the monitor again.
17075 +                        * For the lower dir, the old monitor is still left.
17076 +                        */
17077 +                       if (err == -EEXIST)
17078 +                               err = 0;
17079 +               }
17080 +       }
17081 +
17082 +       AuTraceErr(err);
17083 +       return err;
17084 +}
17085 +
17086 +void au_hn_free(struct au_hinode *hinode)
17087 +{
17088 +       struct au_hnotify *hn;
17089 +
17090 +       hn = hinode->hi_notify;
17091 +       if (hn) {
17092 +               hinode->hi_notify = NULL;
17093 +               if (au_hnotify_op.free(hinode, hn))
17094 +                       au_cache_free_hnotify(hn);
17095 +       }
17096 +}
17097 +
17098 +/* ---------------------------------------------------------------------- */
17099 +
17100 +void au_hn_ctl(struct au_hinode *hinode, int do_set)
17101 +{
17102 +       if (hinode->hi_notify)
17103 +               au_hnotify_op.ctl(hinode, do_set);
17104 +}
17105 +
17106 +void au_hn_reset(struct inode *inode, unsigned int flags)
17107 +{
17108 +       aufs_bindex_t bindex, bbot;
17109 +       struct inode *hi;
17110 +       struct dentry *iwhdentry;
17111 +
17112 +       bbot = au_ibbot(inode);
17113 +       for (bindex = au_ibtop(inode); bindex <= bbot; bindex++) {
17114 +               hi = au_h_iptr(inode, bindex);
17115 +               if (!hi)
17116 +                       continue;
17117 +
17118 +               /* inode_lock_nested(hi, AuLsc_I_CHILD); */
17119 +               iwhdentry = au_hi_wh(inode, bindex);
17120 +               if (iwhdentry)
17121 +                       dget(iwhdentry);
17122 +               au_igrab(hi);
17123 +               au_set_h_iptr(inode, bindex, NULL, 0);
17124 +               au_set_h_iptr(inode, bindex, au_igrab(hi),
17125 +                             flags & ~AuHi_XINO);
17126 +               iput(hi);
17127 +               dput(iwhdentry);
17128 +               /* inode_unlock(hi); */
17129 +       }
17130 +}
17131 +
17132 +/* ---------------------------------------------------------------------- */
17133 +
17134 +static int hn_xino(struct inode *inode, struct inode *h_inode)
17135 +{
17136 +       int err;
17137 +       aufs_bindex_t bindex, bbot, bfound, btop;
17138 +       struct inode *h_i;
17139 +
17140 +       err = 0;
17141 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
17142 +               pr_warn("branch root dir was changed\n");
17143 +               goto out;
17144 +       }
17145 +
17146 +       bfound = -1;
17147 +       bbot = au_ibbot(inode);
17148 +       btop = au_ibtop(inode);
17149 +#if 0 /* reserved for future use */
17150 +       if (bindex == bbot) {
17151 +               /* keep this ino in rename case */
17152 +               goto out;
17153 +       }
17154 +#endif
17155 +       for (bindex = btop; bindex <= bbot; bindex++)
17156 +               if (au_h_iptr(inode, bindex) == h_inode) {
17157 +                       bfound = bindex;
17158 +                       break;
17159 +               }
17160 +       if (bfound < 0)
17161 +               goto out;
17162 +
17163 +       for (bindex = btop; bindex <= bbot; bindex++) {
17164 +               h_i = au_h_iptr(inode, bindex);
17165 +               if (!h_i)
17166 +                       continue;
17167 +
17168 +               err = au_xino_write(inode->i_sb, bindex, h_i->i_ino, /*ino*/0);
17169 +               /* ignore this error */
17170 +               /* bad action? */
17171 +       }
17172 +
17173 +       /* children inode number will be broken */
17174 +
17175 +out:
17176 +       AuTraceErr(err);
17177 +       return err;
17178 +}
17179 +
17180 +static int hn_gen_tree(struct dentry *dentry)
17181 +{
17182 +       int err, i, j, ndentry;
17183 +       struct au_dcsub_pages dpages;
17184 +       struct au_dpage *dpage;
17185 +       struct dentry **dentries;
17186 +
17187 +       err = au_dpages_init(&dpages, GFP_NOFS);
17188 +       if (unlikely(err))
17189 +               goto out;
17190 +       err = au_dcsub_pages(&dpages, dentry, NULL, NULL);
17191 +       if (unlikely(err))
17192 +               goto out_dpages;
17193 +
17194 +       for (i = 0; i < dpages.ndpage; i++) {
17195 +               dpage = dpages.dpages + i;
17196 +               dentries = dpage->dentries;
17197 +               ndentry = dpage->ndentry;
17198 +               for (j = 0; j < ndentry; j++) {
17199 +                       struct dentry *d;
17200 +
17201 +                       d = dentries[j];
17202 +                       if (IS_ROOT(d))
17203 +                               continue;
17204 +
17205 +                       au_digen_dec(d);
17206 +                       if (d_really_is_positive(d))
17207 +                               /* todo: reset children xino?
17208 +                                  cached children only? */
17209 +                               au_iigen_dec(d_inode(d));
17210 +               }
17211 +       }
17212 +
17213 +out_dpages:
17214 +       au_dpages_free(&dpages);
17215 +out:
17216 +       return err;
17217 +}
17218 +
17219 +/*
17220 + * return 0 if processed.
17221 + */
17222 +static int hn_gen_by_inode(char *name, unsigned int nlen, struct inode *inode,
17223 +                          const unsigned int isdir)
17224 +{
17225 +       int err;
17226 +       struct dentry *d;
17227 +       struct qstr *dname;
17228 +
17229 +       err = 1;
17230 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
17231 +               pr_warn("branch root dir was changed\n");
17232 +               err = 0;
17233 +               goto out;
17234 +       }
17235 +
17236 +       if (!isdir) {
17237 +               AuDebugOn(!name);
17238 +               au_iigen_dec(inode);
17239 +               spin_lock(&inode->i_lock);
17240 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
17241 +                       spin_lock(&d->d_lock);
17242 +                       dname = &d->d_name;
17243 +                       if (dname->len != nlen
17244 +                           && memcmp(dname->name, name, nlen)) {
17245 +                               spin_unlock(&d->d_lock);
17246 +                               continue;
17247 +                       }
17248 +                       err = 0;
17249 +                       au_digen_dec(d);
17250 +                       spin_unlock(&d->d_lock);
17251 +                       break;
17252 +               }
17253 +               spin_unlock(&inode->i_lock);
17254 +       } else {
17255 +               au_fset_si(au_sbi(inode->i_sb), FAILED_REFRESH_DIR);
17256 +               d = d_find_any_alias(inode);
17257 +               if (!d) {
17258 +                       au_iigen_dec(inode);
17259 +                       goto out;
17260 +               }
17261 +
17262 +               spin_lock(&d->d_lock);
17263 +               dname = &d->d_name;
17264 +               if (dname->len == nlen && !memcmp(dname->name, name, nlen)) {
17265 +                       spin_unlock(&d->d_lock);
17266 +                       err = hn_gen_tree(d);
17267 +                       spin_lock(&d->d_lock);
17268 +               }
17269 +               spin_unlock(&d->d_lock);
17270 +               dput(d);
17271 +       }
17272 +
17273 +out:
17274 +       AuTraceErr(err);
17275 +       return err;
17276 +}
17277 +
17278 +static int hn_gen_by_name(struct dentry *dentry, const unsigned int isdir)
17279 +{
17280 +       int err;
17281 +
17282 +       if (IS_ROOT(dentry)) {
17283 +               pr_warn("branch root dir was changed\n");
17284 +               return 0;
17285 +       }
17286 +
17287 +       err = 0;
17288 +       if (!isdir) {
17289 +               au_digen_dec(dentry);
17290 +               if (d_really_is_positive(dentry))
17291 +                       au_iigen_dec(d_inode(dentry));
17292 +       } else {
17293 +               au_fset_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR);
17294 +               if (d_really_is_positive(dentry))
17295 +                       err = hn_gen_tree(dentry);
17296 +       }
17297 +
17298 +       AuTraceErr(err);
17299 +       return err;
17300 +}
17301 +
17302 +/* ---------------------------------------------------------------------- */
17303 +
17304 +/* hnotify job flags */
17305 +#define AuHnJob_XINO0          1
17306 +#define AuHnJob_GEN            (1 << 1)
17307 +#define AuHnJob_DIRENT         (1 << 2)
17308 +#define AuHnJob_ISDIR          (1 << 3)
17309 +#define AuHnJob_TRYXINO0       (1 << 4)
17310 +#define AuHnJob_MNTPNT         (1 << 5)
17311 +#define au_ftest_hnjob(flags, name)    ((flags) & AuHnJob_##name)
17312 +#define au_fset_hnjob(flags, name) \
17313 +       do { (flags) |= AuHnJob_##name; } while (0)
17314 +#define au_fclr_hnjob(flags, name) \
17315 +       do { (flags) &= ~AuHnJob_##name; } while (0)
17316 +
17317 +enum {
17318 +       AuHn_CHILD,
17319 +       AuHn_PARENT,
17320 +       AuHnLast
17321 +};
17322 +
17323 +struct au_hnotify_args {
17324 +       struct inode *h_dir, *dir, *h_child_inode;
17325 +       u32 mask;
17326 +       unsigned int flags[AuHnLast];
17327 +       unsigned int h_child_nlen;
17328 +       char h_child_name[];
17329 +};
17330 +
17331 +struct hn_job_args {
17332 +       unsigned int flags;
17333 +       struct inode *inode, *h_inode, *dir, *h_dir;
17334 +       struct dentry *dentry;
17335 +       char *h_name;
17336 +       int h_nlen;
17337 +};
17338 +
17339 +static int hn_job(struct hn_job_args *a)
17340 +{
17341 +       const unsigned int isdir = au_ftest_hnjob(a->flags, ISDIR);
17342 +       int e;
17343 +
17344 +       /* reset xino */
17345 +       if (au_ftest_hnjob(a->flags, XINO0) && a->inode)
17346 +               hn_xino(a->inode, a->h_inode); /* ignore this error */
17347 +
17348 +       if (au_ftest_hnjob(a->flags, TRYXINO0)
17349 +           && a->inode
17350 +           && a->h_inode) {
17351 +               inode_lock_shared_nested(a->h_inode, AuLsc_I_CHILD);
17352 +               if (!a->h_inode->i_nlink
17353 +                   && !(a->h_inode->i_state & I_LINKABLE))
17354 +                       hn_xino(a->inode, a->h_inode); /* ignore this error */
17355 +               inode_unlock_shared(a->h_inode);
17356 +       }
17357 +
17358 +       /* make the generation obsolete */
17359 +       if (au_ftest_hnjob(a->flags, GEN)) {
17360 +               e = -1;
17361 +               if (a->inode)
17362 +                       e = hn_gen_by_inode(a->h_name, a->h_nlen, a->inode,
17363 +                                             isdir);
17364 +               if (e && a->dentry)
17365 +                       hn_gen_by_name(a->dentry, isdir);
17366 +               /* ignore this error */
17367 +       }
17368 +
17369 +       /* make dir entries obsolete */
17370 +       if (au_ftest_hnjob(a->flags, DIRENT) && a->inode) {
17371 +               struct au_vdir *vdir;
17372 +
17373 +               vdir = au_ivdir(a->inode);
17374 +               if (vdir)
17375 +                       vdir->vd_jiffy = 0;
17376 +               /* IMustLock(a->inode); */
17377 +               /* inode_inc_iversion(a->inode); */
17378 +       }
17379 +
17380 +       /* can do nothing but warn */
17381 +       if (au_ftest_hnjob(a->flags, MNTPNT)
17382 +           && a->dentry
17383 +           && d_mountpoint(a->dentry))
17384 +               pr_warn("mount-point %pd is removed or renamed\n", a->dentry);
17385 +
17386 +       return 0;
17387 +}
17388 +
17389 +/* ---------------------------------------------------------------------- */
17390 +
17391 +static struct dentry *lookup_wlock_by_name(char *name, unsigned int nlen,
17392 +                                          struct inode *dir)
17393 +{
17394 +       struct dentry *dentry, *d, *parent;
17395 +       struct qstr *dname;
17396 +
17397 +       parent = d_find_any_alias(dir);
17398 +       if (!parent)
17399 +               return NULL;
17400 +
17401 +       dentry = NULL;
17402 +       spin_lock(&parent->d_lock);
17403 +       list_for_each_entry(d, &parent->d_subdirs, d_child) {
17404 +               /* AuDbg("%pd\n", d); */
17405 +               spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
17406 +               dname = &d->d_name;
17407 +               if (dname->len != nlen || memcmp(dname->name, name, nlen))
17408 +                       goto cont_unlock;
17409 +               if (au_di(d))
17410 +                       au_digen_dec(d);
17411 +               else
17412 +                       goto cont_unlock;
17413 +               if (au_dcount(d) > 0) {
17414 +                       dentry = dget_dlock(d);
17415 +                       spin_unlock(&d->d_lock);
17416 +                       break;
17417 +               }
17418 +
17419 +cont_unlock:
17420 +               spin_unlock(&d->d_lock);
17421 +       }
17422 +       spin_unlock(&parent->d_lock);
17423 +       dput(parent);
17424 +
17425 +       if (dentry)
17426 +               di_write_lock_child(dentry);
17427 +
17428 +       return dentry;
17429 +}
17430 +
17431 +static struct inode *lookup_wlock_by_ino(struct super_block *sb,
17432 +                                        aufs_bindex_t bindex, ino_t h_ino)
17433 +{
17434 +       struct inode *inode;
17435 +       ino_t ino;
17436 +       int err;
17437 +
17438 +       inode = NULL;
17439 +       err = au_xino_read(sb, bindex, h_ino, &ino);
17440 +       if (!err && ino)
17441 +               inode = ilookup(sb, ino);
17442 +       if (!inode)
17443 +               goto out;
17444 +
17445 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
17446 +               pr_warn("wrong root branch\n");
17447 +               iput(inode);
17448 +               inode = NULL;
17449 +               goto out;
17450 +       }
17451 +
17452 +       ii_write_lock_child(inode);
17453 +
17454 +out:
17455 +       return inode;
17456 +}
17457 +
17458 +static void au_hn_bh(void *_args)
17459 +{
17460 +       struct au_hnotify_args *a = _args;
17461 +       struct super_block *sb;
17462 +       aufs_bindex_t bindex, bbot, bfound;
17463 +       unsigned char xino, try_iput;
17464 +       int err;
17465 +       struct inode *inode;
17466 +       ino_t h_ino;
17467 +       struct hn_job_args args;
17468 +       struct dentry *dentry;
17469 +       struct au_sbinfo *sbinfo;
17470 +
17471 +       AuDebugOn(!_args);
17472 +       AuDebugOn(!a->h_dir);
17473 +       AuDebugOn(!a->dir);
17474 +       AuDebugOn(!a->mask);
17475 +       AuDbg("mask 0x%x, i%lu, hi%lu, hci%lu\n",
17476 +             a->mask, a->dir->i_ino, a->h_dir->i_ino,
17477 +             a->h_child_inode ? a->h_child_inode->i_ino : 0);
17478 +
17479 +       inode = NULL;
17480 +       dentry = NULL;
17481 +       /*
17482 +        * do not lock a->dir->i_mutex here
17483 +        * because of d_revalidate() may cause a deadlock.
17484 +        */
17485 +       sb = a->dir->i_sb;
17486 +       AuDebugOn(!sb);
17487 +       sbinfo = au_sbi(sb);
17488 +       AuDebugOn(!sbinfo);
17489 +       si_write_lock(sb, AuLock_NOPLMW);
17490 +
17491 +       if (au_opt_test(sbinfo->si_mntflags, DIRREN))
17492 +               switch (a->mask & FS_EVENTS_POSS_ON_CHILD) {
17493 +               case FS_MOVED_FROM:
17494 +               case FS_MOVED_TO:
17495 +                       AuWarn1("DIRREN with UDBA may not work correctly "
17496 +                               "for the direct rename(2)\n");
17497 +               }
17498 +
17499 +       ii_read_lock_parent(a->dir);
17500 +       bfound = -1;
17501 +       bbot = au_ibbot(a->dir);
17502 +       for (bindex = au_ibtop(a->dir); bindex <= bbot; bindex++)
17503 +               if (au_h_iptr(a->dir, bindex) == a->h_dir) {
17504 +                       bfound = bindex;
17505 +                       break;
17506 +               }
17507 +       ii_read_unlock(a->dir);
17508 +       if (unlikely(bfound < 0))
17509 +               goto out;
17510 +
17511 +       xino = !!au_opt_test(au_mntflags(sb), XINO);
17512 +       h_ino = 0;
17513 +       if (a->h_child_inode)
17514 +               h_ino = a->h_child_inode->i_ino;
17515 +
17516 +       if (a->h_child_nlen
17517 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], GEN)
17518 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], MNTPNT)))
17519 +               dentry = lookup_wlock_by_name(a->h_child_name, a->h_child_nlen,
17520 +                                             a->dir);
17521 +       try_iput = 0;
17522 +       if (dentry && d_really_is_positive(dentry))
17523 +               inode = d_inode(dentry);
17524 +       if (xino && !inode && h_ino
17525 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], XINO0)
17526 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], TRYXINO0)
17527 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], GEN))) {
17528 +               inode = lookup_wlock_by_ino(sb, bfound, h_ino);
17529 +               try_iput = 1;
17530 +       }
17531 +
17532 +       args.flags = a->flags[AuHn_CHILD];
17533 +       args.dentry = dentry;
17534 +       args.inode = inode;
17535 +       args.h_inode = a->h_child_inode;
17536 +       args.dir = a->dir;
17537 +       args.h_dir = a->h_dir;
17538 +       args.h_name = a->h_child_name;
17539 +       args.h_nlen = a->h_child_nlen;
17540 +       err = hn_job(&args);
17541 +       if (dentry) {
17542 +               if (au_di(dentry))
17543 +                       di_write_unlock(dentry);
17544 +               dput(dentry);
17545 +       }
17546 +       if (inode && try_iput) {
17547 +               ii_write_unlock(inode);
17548 +               iput(inode);
17549 +       }
17550 +
17551 +       ii_write_lock_parent(a->dir);
17552 +       args.flags = a->flags[AuHn_PARENT];
17553 +       args.dentry = NULL;
17554 +       args.inode = a->dir;
17555 +       args.h_inode = a->h_dir;
17556 +       args.dir = NULL;
17557 +       args.h_dir = NULL;
17558 +       args.h_name = NULL;
17559 +       args.h_nlen = 0;
17560 +       err = hn_job(&args);
17561 +       ii_write_unlock(a->dir);
17562 +
17563 +out:
17564 +       iput(a->h_child_inode);
17565 +       iput(a->h_dir);
17566 +       iput(a->dir);
17567 +       si_write_unlock(sb);
17568 +       au_nwt_done(&sbinfo->si_nowait);
17569 +       au_kfree_rcu(a);
17570 +}
17571 +
17572 +/* ---------------------------------------------------------------------- */
17573 +
17574 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
17575 +              const struct qstr *h_child_qstr, struct inode *h_child_inode)
17576 +{
17577 +       int err, len;
17578 +       unsigned int flags[AuHnLast], f;
17579 +       unsigned char isdir, isroot, wh;
17580 +       struct inode *dir;
17581 +       struct au_hnotify_args *args;
17582 +       char *p, *h_child_name;
17583 +
17584 +       err = 0;
17585 +       AuDebugOn(!hnotify || !hnotify->hn_aufs_inode);
17586 +       dir = igrab(hnotify->hn_aufs_inode);
17587 +       if (!dir)
17588 +               goto out;
17589 +
17590 +       isroot = (dir->i_ino == AUFS_ROOT_INO);
17591 +       wh = 0;
17592 +       h_child_name = (void *)h_child_qstr->name;
17593 +       len = h_child_qstr->len;
17594 +       if (h_child_name) {
17595 +               if (len > AUFS_WH_PFX_LEN
17596 +                   && !memcmp(h_child_name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
17597 +                       h_child_name += AUFS_WH_PFX_LEN;
17598 +                       len -= AUFS_WH_PFX_LEN;
17599 +                       wh = 1;
17600 +               }
17601 +       }
17602 +
17603 +       isdir = 0;
17604 +       if (h_child_inode)
17605 +               isdir = !!S_ISDIR(h_child_inode->i_mode);
17606 +       flags[AuHn_PARENT] = AuHnJob_ISDIR;
17607 +       flags[AuHn_CHILD] = 0;
17608 +       if (isdir)
17609 +               flags[AuHn_CHILD] = AuHnJob_ISDIR;
17610 +       au_fset_hnjob(flags[AuHn_PARENT], DIRENT);
17611 +       au_fset_hnjob(flags[AuHn_CHILD], GEN);
17612 +       switch (mask & ALL_FSNOTIFY_DIRENT_EVENTS) {
17613 +       case FS_MOVED_FROM:
17614 +       case FS_MOVED_TO:
17615 +               au_fset_hnjob(flags[AuHn_CHILD], XINO0);
17616 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
17617 +               fallthrough;
17618 +       case FS_CREATE:
17619 +               AuDebugOn(!h_child_name);
17620 +               break;
17621 +
17622 +       case FS_DELETE:
17623 +               /*
17624 +                * aufs never be able to get this child inode.
17625 +                * revalidation should be in d_revalidate()
17626 +                * by checking i_nlink, i_generation or d_unhashed().
17627 +                */
17628 +               AuDebugOn(!h_child_name);
17629 +               au_fset_hnjob(flags[AuHn_CHILD], TRYXINO0);
17630 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
17631 +               break;
17632 +
17633 +       default:
17634 +               AuDebugOn(1);
17635 +       }
17636 +
17637 +       if (wh)
17638 +               h_child_inode = NULL;
17639 +
17640 +       err = -ENOMEM;
17641 +       /* iput() and kfree() will be called in au_hnotify() */
17642 +       args = kmalloc(sizeof(*args) + len + 1, GFP_NOFS);
17643 +       if (unlikely(!args)) {
17644 +               AuErr1("no memory\n");
17645 +               iput(dir);
17646 +               goto out;
17647 +       }
17648 +       args->flags[AuHn_PARENT] = flags[AuHn_PARENT];
17649 +       args->flags[AuHn_CHILD] = flags[AuHn_CHILD];
17650 +       args->mask = mask;
17651 +       args->dir = dir;
17652 +       args->h_dir = igrab(h_dir);
17653 +       if (h_child_inode)
17654 +               h_child_inode = igrab(h_child_inode); /* can be NULL */
17655 +       args->h_child_inode = h_child_inode;
17656 +       args->h_child_nlen = len;
17657 +       if (len) {
17658 +               p = (void *)args;
17659 +               p += sizeof(*args);
17660 +               memcpy(p, h_child_name, len);
17661 +               p[len] = 0;
17662 +       }
17663 +
17664 +       /* NFS fires the event for silly-renamed one from kworker */
17665 +       f = 0;
17666 +       if (!dir->i_nlink
17667 +           || (au_test_nfs(h_dir->i_sb) && (mask & FS_DELETE)))
17668 +               f = AuWkq_NEST;
17669 +       err = au_wkq_nowait(au_hn_bh, args, dir->i_sb, f);
17670 +       if (unlikely(err)) {
17671 +               pr_err("wkq %d\n", err);
17672 +               iput(args->h_child_inode);
17673 +               iput(args->h_dir);
17674 +               iput(args->dir);
17675 +               au_kfree_rcu(args);
17676 +       }
17677 +
17678 +out:
17679 +       return err;
17680 +}
17681 +
17682 +/* ---------------------------------------------------------------------- */
17683 +
17684 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm)
17685 +{
17686 +       int err;
17687 +
17688 +       AuDebugOn(!(udba & AuOptMask_UDBA));
17689 +
17690 +       err = 0;
17691 +       if (au_hnotify_op.reset_br)
17692 +               err = au_hnotify_op.reset_br(udba, br, perm);
17693 +
17694 +       return err;
17695 +}
17696 +
17697 +int au_hnotify_init_br(struct au_branch *br, int perm)
17698 +{
17699 +       int err;
17700 +
17701 +       err = 0;
17702 +       if (au_hnotify_op.init_br)
17703 +               err = au_hnotify_op.init_br(br, perm);
17704 +
17705 +       return err;
17706 +}
17707 +
17708 +void au_hnotify_fin_br(struct au_branch *br)
17709 +{
17710 +       if (au_hnotify_op.fin_br)
17711 +               au_hnotify_op.fin_br(br);
17712 +}
17713 +
17714 +static void au_hn_destroy_cache(void)
17715 +{
17716 +       kmem_cache_destroy(au_cache[AuCache_HNOTIFY]);
17717 +       au_cache[AuCache_HNOTIFY] = NULL;
17718 +}
17719 +
17720 +int __init au_hnotify_init(void)
17721 +{
17722 +       int err;
17723 +
17724 +       err = -ENOMEM;
17725 +       au_cache[AuCache_HNOTIFY] = AuCache(au_hnotify);
17726 +       if (au_cache[AuCache_HNOTIFY]) {
17727 +               err = 0;
17728 +               if (au_hnotify_op.init)
17729 +                       err = au_hnotify_op.init();
17730 +               if (unlikely(err))
17731 +                       au_hn_destroy_cache();
17732 +       }
17733 +       AuTraceErr(err);
17734 +       return err;
17735 +}
17736 +
17737 +void au_hnotify_fin(void)
17738 +{
17739 +       if (au_hnotify_op.fin)
17740 +               au_hnotify_op.fin();
17741 +
17742 +       /* cf. au_cache_fin() */
17743 +       if (au_cache[AuCache_HNOTIFY])
17744 +               au_hn_destroy_cache();
17745 +}
17746 diff -urN /usr/share/empty/fs/aufs/iinfo.c linux/fs/aufs/iinfo.c
17747 --- /usr/share/empty/fs/aufs/iinfo.c    1970-01-01 01:00:00.000000000 +0100
17748 +++ linux/fs/aufs/iinfo.c       2021-12-03 15:38:59.936647310 +0100
17749 @@ -0,0 +1,286 @@
17750 +// SPDX-License-Identifier: GPL-2.0
17751 +/*
17752 + * Copyright (C) 2005-2021 Junjiro R. Okajima
17753 + *
17754 + * This program, aufs is free software; you can redistribute it and/or modify
17755 + * it under the terms of the GNU General Public License as published by
17756 + * the Free Software Foundation; either version 2 of the License, or
17757 + * (at your option) any later version.
17758 + *
17759 + * This program is distributed in the hope that it will be useful,
17760 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17761 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17762 + * GNU General Public License for more details.
17763 + *
17764 + * You should have received a copy of the GNU General Public License
17765 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17766 + */
17767 +
17768 +/*
17769 + * inode private data
17770 + */
17771 +
17772 +#include "aufs.h"
17773 +
17774 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex)
17775 +{
17776 +       struct inode *h_inode;
17777 +       struct au_hinode *hinode;
17778 +
17779 +       IiMustAnyLock(inode);
17780 +
17781 +       hinode = au_hinode(au_ii(inode), bindex);
17782 +       h_inode = hinode->hi_inode;
17783 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
17784 +       return h_inode;
17785 +}
17786 +
17787 +/* todo: hard/soft set? */
17788 +void au_hiput(struct au_hinode *hinode)
17789 +{
17790 +       au_hn_free(hinode);
17791 +       dput(hinode->hi_whdentry);
17792 +       iput(hinode->hi_inode);
17793 +}
17794 +
17795 +unsigned int au_hi_flags(struct inode *inode, int isdir)
17796 +{
17797 +       unsigned int flags;
17798 +       const unsigned int mnt_flags = au_mntflags(inode->i_sb);
17799 +
17800 +       flags = 0;
17801 +       if (au_opt_test(mnt_flags, XINO))
17802 +               au_fset_hi(flags, XINO);
17803 +       if (isdir && au_opt_test(mnt_flags, UDBA_HNOTIFY))
17804 +               au_fset_hi(flags, HNOTIFY);
17805 +       return flags;
17806 +}
17807 +
17808 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
17809 +                  struct inode *h_inode, unsigned int flags)
17810 +{
17811 +       struct au_hinode *hinode;
17812 +       struct inode *hi;
17813 +       struct au_iinfo *iinfo = au_ii(inode);
17814 +
17815 +       IiMustWriteLock(inode);
17816 +
17817 +       hinode = au_hinode(iinfo, bindex);
17818 +       hi = hinode->hi_inode;
17819 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
17820 +
17821 +       if (hi)
17822 +               au_hiput(hinode);
17823 +       hinode->hi_inode = h_inode;
17824 +       if (h_inode) {
17825 +               int err;
17826 +               struct super_block *sb = inode->i_sb;
17827 +               struct au_branch *br;
17828 +
17829 +               AuDebugOn(inode->i_mode
17830 +                         && (h_inode->i_mode & S_IFMT)
17831 +                         != (inode->i_mode & S_IFMT));
17832 +               if (bindex == iinfo->ii_btop)
17833 +                       au_cpup_igen(inode, h_inode);
17834 +               br = au_sbr(sb, bindex);
17835 +               hinode->hi_id = br->br_id;
17836 +               if (au_ftest_hi(flags, XINO)) {
17837 +                       err = au_xino_write(sb, bindex, h_inode->i_ino,
17838 +                                           inode->i_ino);
17839 +                       if (unlikely(err))
17840 +                               AuIOErr1("failed au_xino_write() %d\n", err);
17841 +               }
17842 +
17843 +               if (au_ftest_hi(flags, HNOTIFY)
17844 +                   && au_br_hnotifyable(br->br_perm)) {
17845 +                       err = au_hn_alloc(hinode, inode);
17846 +                       if (unlikely(err))
17847 +                               AuIOErr1("au_hn_alloc() %d\n", err);
17848 +               }
17849 +       }
17850 +}
17851 +
17852 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
17853 +                 struct dentry *h_wh)
17854 +{
17855 +       struct au_hinode *hinode;
17856 +
17857 +       IiMustWriteLock(inode);
17858 +
17859 +       hinode = au_hinode(au_ii(inode), bindex);
17860 +       AuDebugOn(hinode->hi_whdentry);
17861 +       hinode->hi_whdentry = h_wh;
17862 +}
17863 +
17864 +void au_update_iigen(struct inode *inode, int half)
17865 +{
17866 +       struct au_iinfo *iinfo;
17867 +       struct au_iigen *iigen;
17868 +       unsigned int sigen;
17869 +
17870 +       sigen = au_sigen(inode->i_sb);
17871 +       iinfo = au_ii(inode);
17872 +       iigen = &iinfo->ii_generation;
17873 +       spin_lock(&iigen->ig_spin);
17874 +       iigen->ig_generation = sigen;
17875 +       if (half)
17876 +               au_ig_fset(iigen->ig_flags, HALF_REFRESHED);
17877 +       else
17878 +               au_ig_fclr(iigen->ig_flags, HALF_REFRESHED);
17879 +       spin_unlock(&iigen->ig_spin);
17880 +}
17881 +
17882 +/* it may be called at remount time, too */
17883 +void au_update_ibrange(struct inode *inode, int do_put_zero)
17884 +{
17885 +       struct au_iinfo *iinfo;
17886 +       aufs_bindex_t bindex, bbot;
17887 +
17888 +       AuDebugOn(au_is_bad_inode(inode));
17889 +       IiMustWriteLock(inode);
17890 +
17891 +       iinfo = au_ii(inode);
17892 +       if (do_put_zero && iinfo->ii_btop >= 0) {
17893 +               for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot;
17894 +                    bindex++) {
17895 +                       struct inode *h_i;
17896 +
17897 +                       h_i = au_hinode(iinfo, bindex)->hi_inode;
17898 +                       if (h_i
17899 +                           && !h_i->i_nlink
17900 +                           && !(h_i->i_state & I_LINKABLE))
17901 +                               au_set_h_iptr(inode, bindex, NULL, 0);
17902 +               }
17903 +       }
17904 +
17905 +       iinfo->ii_btop = -1;
17906 +       iinfo->ii_bbot = -1;
17907 +       bbot = au_sbbot(inode->i_sb);
17908 +       for (bindex = 0; bindex <= bbot; bindex++)
17909 +               if (au_hinode(iinfo, bindex)->hi_inode) {
17910 +                       iinfo->ii_btop = bindex;
17911 +                       break;
17912 +               }
17913 +       if (iinfo->ii_btop >= 0)
17914 +               for (bindex = bbot; bindex >= iinfo->ii_btop; bindex--)
17915 +                       if (au_hinode(iinfo, bindex)->hi_inode) {
17916 +                               iinfo->ii_bbot = bindex;
17917 +                               break;
17918 +                       }
17919 +       AuDebugOn(iinfo->ii_btop > iinfo->ii_bbot);
17920 +}
17921 +
17922 +/* ---------------------------------------------------------------------- */
17923 +
17924 +void au_icntnr_init_once(void *_c)
17925 +{
17926 +       struct au_icntnr *c = _c;
17927 +       struct au_iinfo *iinfo = &c->iinfo;
17928 +
17929 +       spin_lock_init(&iinfo->ii_generation.ig_spin);
17930 +       au_rw_init(&iinfo->ii_rwsem);
17931 +       inode_init_once(&c->vfs_inode);
17932 +}
17933 +
17934 +void au_hinode_init(struct au_hinode *hinode)
17935 +{
17936 +       hinode->hi_inode = NULL;
17937 +       hinode->hi_id = -1;
17938 +       au_hn_init(hinode);
17939 +       hinode->hi_whdentry = NULL;
17940 +}
17941 +
17942 +int au_iinfo_init(struct inode *inode)
17943 +{
17944 +       struct au_iinfo *iinfo;
17945 +       struct super_block *sb;
17946 +       struct au_hinode *hi;
17947 +       int nbr, i;
17948 +
17949 +       sb = inode->i_sb;
17950 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
17951 +       nbr = au_sbbot(sb) + 1;
17952 +       if (unlikely(nbr <= 0))
17953 +               nbr = 1;
17954 +       hi = kmalloc_array(nbr, sizeof(*iinfo->ii_hinode), GFP_NOFS);
17955 +       if (hi) {
17956 +               au_lcnt_inc(&au_sbi(sb)->si_ninodes);
17957 +
17958 +               iinfo->ii_hinode = hi;
17959 +               for (i = 0; i < nbr; i++, hi++)
17960 +                       au_hinode_init(hi);
17961 +
17962 +               iinfo->ii_generation.ig_generation = au_sigen(sb);
17963 +               iinfo->ii_btop = -1;
17964 +               iinfo->ii_bbot = -1;
17965 +               iinfo->ii_vdir = NULL;
17966 +               return 0;
17967 +       }
17968 +       return -ENOMEM;
17969 +}
17970 +
17971 +int au_hinode_realloc(struct au_iinfo *iinfo, int nbr, int may_shrink)
17972 +{
17973 +       int err, i;
17974 +       struct au_hinode *hip;
17975 +
17976 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
17977 +
17978 +       err = -ENOMEM;
17979 +       hip = au_krealloc(iinfo->ii_hinode, sizeof(*hip) * nbr, GFP_NOFS,
17980 +                         may_shrink);
17981 +       if (hip) {
17982 +               iinfo->ii_hinode = hip;
17983 +               i = iinfo->ii_bbot + 1;
17984 +               hip += i;
17985 +               for (; i < nbr; i++, hip++)
17986 +                       au_hinode_init(hip);
17987 +               err = 0;
17988 +       }
17989 +
17990 +       return err;
17991 +}
17992 +
17993 +void au_iinfo_fin(struct inode *inode)
17994 +{
17995 +       struct au_iinfo *iinfo;
17996 +       struct au_hinode *hi;
17997 +       struct super_block *sb;
17998 +       aufs_bindex_t bindex, bbot;
17999 +       const unsigned char unlinked = !inode->i_nlink;
18000 +
18001 +       AuDebugOn(au_is_bad_inode(inode));
18002 +
18003 +       sb = inode->i_sb;
18004 +       au_lcnt_dec(&au_sbi(sb)->si_ninodes);
18005 +       if (si_pid_test(sb))
18006 +               au_xino_delete_inode(inode, unlinked);
18007 +       else {
18008 +               /*
18009 +                * it is safe to hide the dependency between sbinfo and
18010 +                * sb->s_umount.
18011 +                */
18012 +               lockdep_off();
18013 +               si_noflush_read_lock(sb);
18014 +               au_xino_delete_inode(inode, unlinked);
18015 +               si_read_unlock(sb);
18016 +               lockdep_on();
18017 +       }
18018 +
18019 +       iinfo = au_ii(inode);
18020 +       if (iinfo->ii_vdir)
18021 +               au_vdir_free(iinfo->ii_vdir);
18022 +
18023 +       bindex = iinfo->ii_btop;
18024 +       if (bindex >= 0) {
18025 +               hi = au_hinode(iinfo, bindex);
18026 +               bbot = iinfo->ii_bbot;
18027 +               while (bindex++ <= bbot) {
18028 +                       if (hi->hi_inode)
18029 +                               au_hiput(hi);
18030 +                       hi++;
18031 +               }
18032 +       }
18033 +       au_kfree_rcu(iinfo->ii_hinode);
18034 +       AuRwDestroy(&iinfo->ii_rwsem);
18035 +}
18036 diff -urN /usr/share/empty/fs/aufs/inode.c linux/fs/aufs/inode.c
18037 --- /usr/share/empty/fs/aufs/inode.c    1970-01-01 01:00:00.000000000 +0100
18038 +++ linux/fs/aufs/inode.c       2021-12-03 15:40:58.236647297 +0100
18039 @@ -0,0 +1,531 @@
18040 +// SPDX-License-Identifier: GPL-2.0
18041 +/*
18042 + * Copyright (C) 2005-2021 Junjiro R. Okajima
18043 + *
18044 + * This program, aufs is free software; you can redistribute it and/or modify
18045 + * it under the terms of the GNU General Public License as published by
18046 + * the Free Software Foundation; either version 2 of the License, or
18047 + * (at your option) any later version.
18048 + *
18049 + * This program is distributed in the hope that it will be useful,
18050 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18051 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18052 + * GNU General Public License for more details.
18053 + *
18054 + * You should have received a copy of the GNU General Public License
18055 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18056 + */
18057 +
18058 +/*
18059 + * inode functions
18060 + */
18061 +
18062 +#include <linux/iversion.h>
18063 +#include "aufs.h"
18064 +
18065 +struct inode *au_igrab(struct inode *inode)
18066 +{
18067 +       if (inode) {
18068 +               AuDebugOn(!atomic_read(&inode->i_count));
18069 +               ihold(inode);
18070 +       }
18071 +       return inode;
18072 +}
18073 +
18074 +static void au_refresh_hinode_attr(struct inode *inode, int do_version)
18075 +{
18076 +       au_cpup_attr_all(inode, /*force*/0);
18077 +       au_update_iigen(inode, /*half*/1);
18078 +       if (do_version)
18079 +               inode_inc_iversion(inode);
18080 +}
18081 +
18082 +static int au_ii_refresh(struct inode *inode, int *update)
18083 +{
18084 +       int err, e, nbr;
18085 +       umode_t type;
18086 +       aufs_bindex_t bindex, new_bindex;
18087 +       struct super_block *sb;
18088 +       struct au_iinfo *iinfo;
18089 +       struct au_hinode *p, *q, tmp;
18090 +
18091 +       AuDebugOn(au_is_bad_inode(inode));
18092 +       IiMustWriteLock(inode);
18093 +
18094 +       *update = 0;
18095 +       sb = inode->i_sb;
18096 +       nbr = au_sbbot(sb) + 1;
18097 +       type = inode->i_mode & S_IFMT;
18098 +       iinfo = au_ii(inode);
18099 +       err = au_hinode_realloc(iinfo, nbr, /*may_shrink*/0);
18100 +       if (unlikely(err))
18101 +               goto out;
18102 +
18103 +       AuDebugOn(iinfo->ii_btop < 0);
18104 +       p = au_hinode(iinfo, iinfo->ii_btop);
18105 +       for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot;
18106 +            bindex++, p++) {
18107 +               if (!p->hi_inode)
18108 +                       continue;
18109 +
18110 +               AuDebugOn(type != (p->hi_inode->i_mode & S_IFMT));
18111 +               new_bindex = au_br_index(sb, p->hi_id);
18112 +               if (new_bindex == bindex)
18113 +                       continue;
18114 +
18115 +               if (new_bindex < 0) {
18116 +                       *update = 1;
18117 +                       au_hiput(p);
18118 +                       p->hi_inode = NULL;
18119 +                       continue;
18120 +               }
18121 +
18122 +               if (new_bindex < iinfo->ii_btop)
18123 +                       iinfo->ii_btop = new_bindex;
18124 +               if (iinfo->ii_bbot < new_bindex)
18125 +                       iinfo->ii_bbot = new_bindex;
18126 +               /* swap two lower inode, and loop again */
18127 +               q = au_hinode(iinfo, new_bindex);
18128 +               tmp = *q;
18129 +               *q = *p;
18130 +               *p = tmp;
18131 +               if (tmp.hi_inode) {
18132 +                       bindex--;
18133 +                       p--;
18134 +               }
18135 +       }
18136 +       au_update_ibrange(inode, /*do_put_zero*/0);
18137 +       au_hinode_realloc(iinfo, nbr, /*may_shrink*/1); /* harmless if err */
18138 +       e = au_dy_irefresh(inode);
18139 +       if (unlikely(e && !err))
18140 +               err = e;
18141 +
18142 +out:
18143 +       AuTraceErr(err);
18144 +       return err;
18145 +}
18146 +
18147 +void au_refresh_iop(struct inode *inode, int force_getattr)
18148 +{
18149 +       int type;
18150 +       struct au_sbinfo *sbi = au_sbi(inode->i_sb);
18151 +       const struct inode_operations *iop
18152 +               = force_getattr ? aufs_iop : sbi->si_iop_array;
18153 +
18154 +       if (inode->i_op == iop)
18155 +               return;
18156 +
18157 +       switch (inode->i_mode & S_IFMT) {
18158 +       case S_IFDIR:
18159 +               type = AuIop_DIR;
18160 +               break;
18161 +       case S_IFLNK:
18162 +               type = AuIop_SYMLINK;
18163 +               break;
18164 +       default:
18165 +               type = AuIop_OTHER;
18166 +               break;
18167 +       }
18168 +
18169 +       inode->i_op = iop + type;
18170 +       /* unnecessary smp_wmb() */
18171 +}
18172 +
18173 +int au_refresh_hinode_self(struct inode *inode)
18174 +{
18175 +       int err, update;
18176 +
18177 +       err = au_ii_refresh(inode, &update);
18178 +       if (!err)
18179 +               au_refresh_hinode_attr(inode, update && S_ISDIR(inode->i_mode));
18180 +
18181 +       AuTraceErr(err);
18182 +       return err;
18183 +}
18184 +
18185 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry)
18186 +{
18187 +       int err, e, update;
18188 +       unsigned int flags;
18189 +       umode_t mode;
18190 +       aufs_bindex_t bindex, bbot;
18191 +       unsigned char isdir;
18192 +       struct au_hinode *p;
18193 +       struct au_iinfo *iinfo;
18194 +
18195 +       err = au_ii_refresh(inode, &update);
18196 +       if (unlikely(err))
18197 +               goto out;
18198 +
18199 +       update = 0;
18200 +       iinfo = au_ii(inode);
18201 +       p = au_hinode(iinfo, iinfo->ii_btop);
18202 +       mode = (inode->i_mode & S_IFMT);
18203 +       isdir = S_ISDIR(mode);
18204 +       flags = au_hi_flags(inode, isdir);
18205 +       bbot = au_dbbot(dentry);
18206 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++) {
18207 +               struct inode *h_i, *h_inode;
18208 +               struct dentry *h_d;
18209 +
18210 +               h_d = au_h_dptr(dentry, bindex);
18211 +               if (!h_d || d_is_negative(h_d))
18212 +                       continue;
18213 +
18214 +               h_inode = d_inode(h_d);
18215 +               AuDebugOn(mode != (h_inode->i_mode & S_IFMT));
18216 +               if (iinfo->ii_btop <= bindex && bindex <= iinfo->ii_bbot) {
18217 +                       h_i = au_h_iptr(inode, bindex);
18218 +                       if (h_i) {
18219 +                               if (h_i == h_inode)
18220 +                                       continue;
18221 +                               err = -EIO;
18222 +                               break;
18223 +                       }
18224 +               }
18225 +               if (bindex < iinfo->ii_btop)
18226 +                       iinfo->ii_btop = bindex;
18227 +               if (iinfo->ii_bbot < bindex)
18228 +                       iinfo->ii_bbot = bindex;
18229 +               au_set_h_iptr(inode, bindex, au_igrab(h_inode), flags);
18230 +               update = 1;
18231 +       }
18232 +       au_update_ibrange(inode, /*do_put_zero*/0);
18233 +       e = au_dy_irefresh(inode);
18234 +       if (unlikely(e && !err))
18235 +               err = e;
18236 +       if (!err)
18237 +               au_refresh_hinode_attr(inode, update && isdir);
18238 +
18239 +out:
18240 +       AuTraceErr(err);
18241 +       return err;
18242 +}
18243 +
18244 +static int set_inode(struct inode *inode, struct dentry *dentry)
18245 +{
18246 +       int err;
18247 +       unsigned int flags;
18248 +       umode_t mode;
18249 +       aufs_bindex_t bindex, btop, btail;
18250 +       unsigned char isdir;
18251 +       struct dentry *h_dentry;
18252 +       struct inode *h_inode;
18253 +       struct au_iinfo *iinfo;
18254 +       const struct inode_operations *iop;
18255 +
18256 +       IiMustWriteLock(inode);
18257 +
18258 +       err = 0;
18259 +       isdir = 0;
18260 +       iop = au_sbi(inode->i_sb)->si_iop_array;
18261 +       btop = au_dbtop(dentry);
18262 +       h_dentry = au_h_dptr(dentry, btop);
18263 +       h_inode = d_inode(h_dentry);
18264 +       mode = h_inode->i_mode;
18265 +       switch (mode & S_IFMT) {
18266 +       case S_IFREG:
18267 +               btail = au_dbtail(dentry);
18268 +               inode->i_op = iop + AuIop_OTHER;
18269 +               inode->i_fop = &aufs_file_fop;
18270 +               err = au_dy_iaop(inode, btop, h_inode);
18271 +               if (unlikely(err))
18272 +                       goto out;
18273 +               break;
18274 +       case S_IFDIR:
18275 +               isdir = 1;
18276 +               btail = au_dbtaildir(dentry);
18277 +               inode->i_op = iop + AuIop_DIR;
18278 +               inode->i_fop = &aufs_dir_fop;
18279 +               break;
18280 +       case S_IFLNK:
18281 +               btail = au_dbtail(dentry);
18282 +               inode->i_op = iop + AuIop_SYMLINK;
18283 +               break;
18284 +       case S_IFBLK:
18285 +       case S_IFCHR:
18286 +       case S_IFIFO:
18287 +       case S_IFSOCK:
18288 +               btail = au_dbtail(dentry);
18289 +               inode->i_op = iop + AuIop_OTHER;
18290 +               init_special_inode(inode, mode, h_inode->i_rdev);
18291 +               break;
18292 +       default:
18293 +               AuIOErr("Unknown file type 0%o\n", mode);
18294 +               err = -EIO;
18295 +               goto out;
18296 +       }
18297 +
18298 +       /* do not set hnotify for whiteouted dirs (SHWH mode) */
18299 +       flags = au_hi_flags(inode, isdir);
18300 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH)
18301 +           && au_ftest_hi(flags, HNOTIFY)
18302 +           && dentry->d_name.len > AUFS_WH_PFX_LEN
18303 +           && !memcmp(dentry->d_name.name, AUFS_WH_PFX, AUFS_WH_PFX_LEN))
18304 +               au_fclr_hi(flags, HNOTIFY);
18305 +       iinfo = au_ii(inode);
18306 +       iinfo->ii_btop = btop;
18307 +       iinfo->ii_bbot = btail;
18308 +       for (bindex = btop; bindex <= btail; bindex++) {
18309 +               h_dentry = au_h_dptr(dentry, bindex);
18310 +               if (h_dentry)
18311 +                       au_set_h_iptr(inode, bindex,
18312 +                                     au_igrab(d_inode(h_dentry)), flags);
18313 +       }
18314 +       au_cpup_attr_all(inode, /*force*/1);
18315 +       /*
18316 +        * to force calling aufs_get_acl() every time,
18317 +        * do not call cache_no_acl() for aufs inode.
18318 +        */
18319 +
18320 +out:
18321 +       return err;
18322 +}
18323 +
18324 +/*
18325 + * successful returns with iinfo write_locked
18326 + * minus: errno
18327 + * zero: success, matched
18328 + * plus: no error, but unmatched
18329 + */
18330 +static int reval_inode(struct inode *inode, struct dentry *dentry)
18331 +{
18332 +       int err;
18333 +       unsigned int gen, igflags;
18334 +       aufs_bindex_t bindex, bbot;
18335 +       struct inode *h_inode, *h_dinode;
18336 +       struct dentry *h_dentry;
18337 +
18338 +       /*
18339 +        * before this function, if aufs got any iinfo lock, it must be only
18340 +        * one, the parent dir.
18341 +        * it can happen by UDBA and the obsoleted inode number.
18342 +        */
18343 +       err = -EIO;
18344 +       if (unlikely(inode->i_ino == parent_ino(dentry)))
18345 +               goto out;
18346 +
18347 +       err = 1;
18348 +       ii_write_lock_new_child(inode);
18349 +       h_dentry = au_h_dptr(dentry, au_dbtop(dentry));
18350 +       h_dinode = d_inode(h_dentry);
18351 +       bbot = au_ibbot(inode);
18352 +       for (bindex = au_ibtop(inode); bindex <= bbot; bindex++) {
18353 +               h_inode = au_h_iptr(inode, bindex);
18354 +               if (!h_inode || h_inode != h_dinode)
18355 +                       continue;
18356 +
18357 +               err = 0;
18358 +               gen = au_iigen(inode, &igflags);
18359 +               if (gen == au_digen(dentry)
18360 +                   && !au_ig_ftest(igflags, HALF_REFRESHED))
18361 +                       break;
18362 +
18363 +               /* fully refresh inode using dentry */
18364 +               err = au_refresh_hinode(inode, dentry);
18365 +               if (!err)
18366 +                       au_update_iigen(inode, /*half*/0);
18367 +               break;
18368 +       }
18369 +
18370 +       if (unlikely(err))
18371 +               ii_write_unlock(inode);
18372 +out:
18373 +       return err;
18374 +}
18375 +
18376 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
18377 +          unsigned int d_type, ino_t *ino)
18378 +{
18379 +       int err, idx;
18380 +       const int isnondir = d_type != DT_DIR;
18381 +
18382 +       /* prevent hardlinked inode number from race condition */
18383 +       if (isnondir) {
18384 +               err = au_xinondir_enter(sb, bindex, h_ino, &idx);
18385 +               if (unlikely(err))
18386 +                       goto out;
18387 +       }
18388 +
18389 +       err = au_xino_read(sb, bindex, h_ino, ino);
18390 +       if (unlikely(err))
18391 +               goto out_xinondir;
18392 +
18393 +       if (!*ino) {
18394 +               err = -EIO;
18395 +               *ino = au_xino_new_ino(sb);
18396 +               if (unlikely(!*ino))
18397 +                       goto out_xinondir;
18398 +               err = au_xino_write(sb, bindex, h_ino, *ino);
18399 +               if (unlikely(err))
18400 +                       goto out_xinondir;
18401 +       }
18402 +
18403 +out_xinondir:
18404 +       if (isnondir && idx >= 0)
18405 +               au_xinondir_leave(sb, bindex, h_ino, idx);
18406 +out:
18407 +       return err;
18408 +}
18409 +
18410 +/* successful returns with iinfo write_locked */
18411 +/* todo: return with unlocked? */
18412 +struct inode *au_new_inode(struct dentry *dentry, int must_new)
18413 +{
18414 +       struct inode *inode, *h_inode;
18415 +       struct dentry *h_dentry;
18416 +       struct super_block *sb;
18417 +       ino_t h_ino, ino;
18418 +       int err, idx, hlinked;
18419 +       aufs_bindex_t btop;
18420 +
18421 +       sb = dentry->d_sb;
18422 +       btop = au_dbtop(dentry);
18423 +       h_dentry = au_h_dptr(dentry, btop);
18424 +       h_inode = d_inode(h_dentry);
18425 +       h_ino = h_inode->i_ino;
18426 +       hlinked = !d_is_dir(h_dentry) && h_inode->i_nlink > 1;
18427 +
18428 +new_ino:
18429 +       /*
18430 +        * stop 'race'-ing between hardlinks under different
18431 +        * parents.
18432 +        */
18433 +       if (hlinked) {
18434 +               err = au_xinondir_enter(sb, btop, h_ino, &idx);
18435 +               inode = ERR_PTR(err);
18436 +               if (unlikely(err))
18437 +                       goto out;
18438 +       }
18439 +
18440 +       err = au_xino_read(sb, btop, h_ino, &ino);
18441 +       inode = ERR_PTR(err);
18442 +       if (unlikely(err))
18443 +               goto out_xinondir;
18444 +
18445 +       if (!ino) {
18446 +               ino = au_xino_new_ino(sb);
18447 +               if (unlikely(!ino)) {
18448 +                       inode = ERR_PTR(-EIO);
18449 +                       goto out_xinondir;
18450 +               }
18451 +       }
18452 +
18453 +       AuDbg("i%lu\n", (unsigned long)ino);
18454 +       inode = au_iget_locked(sb, ino);
18455 +       err = PTR_ERR(inode);
18456 +       if (IS_ERR(inode))
18457 +               goto out_xinondir;
18458 +
18459 +       AuDbg("%lx, new %d\n", inode->i_state, !!(inode->i_state & I_NEW));
18460 +       if (inode->i_state & I_NEW) {
18461 +               ii_write_lock_new_child(inode);
18462 +               err = set_inode(inode, dentry);
18463 +               if (!err) {
18464 +                       unlock_new_inode(inode);
18465 +                       goto out_xinondir; /* success */
18466 +               }
18467 +
18468 +               /*
18469 +                * iget_failed() calls iput(), but we need to call
18470 +                * ii_write_unlock() after iget_failed(). so dirty hack for
18471 +                * i_count.
18472 +                */
18473 +               atomic_inc(&inode->i_count);
18474 +               iget_failed(inode);
18475 +               ii_write_unlock(inode);
18476 +               au_xino_write(sb, btop, h_ino, /*ino*/0);
18477 +               /* ignore this error */
18478 +               goto out_iput;
18479 +       } else if (!must_new && !IS_DEADDIR(inode) && inode->i_nlink) {
18480 +               /*
18481 +                * horrible race condition between lookup, readdir and copyup
18482 +                * (or something).
18483 +                */
18484 +               if (hlinked && idx >= 0)
18485 +                       au_xinondir_leave(sb, btop, h_ino, idx);
18486 +               err = reval_inode(inode, dentry);
18487 +               if (unlikely(err < 0)) {
18488 +                       hlinked = 0;
18489 +                       goto out_iput;
18490 +               }
18491 +               if (!err)
18492 +                       goto out; /* success */
18493 +               else if (hlinked && idx >= 0) {
18494 +                       err = au_xinondir_enter(sb, btop, h_ino, &idx);
18495 +                       if (unlikely(err)) {
18496 +                               iput(inode);
18497 +                               inode = ERR_PTR(err);
18498 +                               goto out;
18499 +                       }
18500 +               }
18501 +       }
18502 +
18503 +       if (unlikely(au_test_fs_unique_ino(h_inode)))
18504 +               AuWarn1("Warning: Un-notified UDBA or repeatedly renamed dir,"
18505 +                       " b%d, %s, %pd, hi%lu, i%lu.\n",
18506 +                       btop, au_sbtype(h_dentry->d_sb), dentry,
18507 +                       (unsigned long)h_ino, (unsigned long)ino);
18508 +       ino = 0;
18509 +       err = au_xino_write(sb, btop, h_ino, /*ino*/0);
18510 +       if (!err) {
18511 +               iput(inode);
18512 +               if (hlinked && idx >= 0)
18513 +                       au_xinondir_leave(sb, btop, h_ino, idx);
18514 +               goto new_ino;
18515 +       }
18516 +
18517 +out_iput:
18518 +       iput(inode);
18519 +       inode = ERR_PTR(err);
18520 +out_xinondir:
18521 +       if (hlinked && idx >= 0)
18522 +               au_xinondir_leave(sb, btop, h_ino, idx);
18523 +out:
18524 +       return inode;
18525 +}
18526 +
18527 +/* ---------------------------------------------------------------------- */
18528 +
18529 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
18530 +              struct inode *inode)
18531 +{
18532 +       int err;
18533 +       struct inode *hi;
18534 +
18535 +       err = au_br_rdonly(au_sbr(sb, bindex));
18536 +
18537 +       /* pseudo-link after flushed may happen out of bounds */
18538 +       if (!err
18539 +           && inode
18540 +           && au_ibtop(inode) <= bindex
18541 +           && bindex <= au_ibbot(inode)) {
18542 +               /*
18543 +                * permission check is unnecessary since vfsub routine
18544 +                * will be called later
18545 +                */
18546 +               hi = au_h_iptr(inode, bindex);
18547 +               if (hi)
18548 +                       err = IS_IMMUTABLE(hi) ? -EROFS : 0;
18549 +       }
18550 +
18551 +       return err;
18552 +}
18553 +
18554 +int au_test_h_perm(struct user_namespace *h_userns, struct inode *h_inode,
18555 +                  int mask)
18556 +{
18557 +       if (uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
18558 +               return 0;
18559 +       return inode_permission(h_userns, h_inode, mask);
18560 +}
18561 +
18562 +int au_test_h_perm_sio(struct user_namespace *h_userns, struct inode *h_inode,
18563 +                      int mask)
18564 +{
18565 +       if (au_test_nfs(h_inode->i_sb)
18566 +           && (mask & MAY_WRITE)
18567 +           && S_ISDIR(h_inode->i_mode))
18568 +               mask |= MAY_READ; /* force permission check */
18569 +       return au_test_h_perm(h_userns, h_inode, mask);
18570 +}
18571 diff -urN /usr/share/empty/fs/aufs/inode.h linux/fs/aufs/inode.h
18572 --- /usr/share/empty/fs/aufs/inode.h    1970-01-01 01:00:00.000000000 +0100
18573 +++ linux/fs/aufs/inode.h       2021-12-03 15:40:58.236647297 +0100
18574 @@ -0,0 +1,705 @@
18575 +/* SPDX-License-Identifier: GPL-2.0 */
18576 +/*
18577 + * Copyright (C) 2005-2021 Junjiro R. Okajima
18578 + *
18579 + * This program, aufs is free software; you can redistribute it and/or modify
18580 + * it under the terms of the GNU General Public License as published by
18581 + * the Free Software Foundation; either version 2 of the License, or
18582 + * (at your option) any later version.
18583 + *
18584 + * This program is distributed in the hope that it will be useful,
18585 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18586 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18587 + * GNU General Public License for more details.
18588 + *
18589 + * You should have received a copy of the GNU General Public License
18590 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18591 + */
18592 +
18593 +/*
18594 + * inode operations
18595 + */
18596 +
18597 +#ifndef __AUFS_INODE_H__
18598 +#define __AUFS_INODE_H__
18599 +
18600 +#ifdef __KERNEL__
18601 +
18602 +#include <linux/fsnotify.h>
18603 +#include "rwsem.h"
18604 +
18605 +struct vfsmount;
18606 +
18607 +struct au_hnotify {
18608 +#ifdef CONFIG_AUFS_HNOTIFY
18609 +#ifdef CONFIG_AUFS_HFSNOTIFY
18610 +       /* never use fsnotify_add_vfsmount_mark() */
18611 +       struct fsnotify_mark            hn_mark;
18612 +#endif
18613 +       struct inode            *hn_aufs_inode; /* no get/put */
18614 +       struct rcu_head         rcu;
18615 +#endif
18616 +} ____cacheline_aligned_in_smp;
18617 +
18618 +struct au_hinode {
18619 +       struct inode            *hi_inode;
18620 +       aufs_bindex_t           hi_id;
18621 +#ifdef CONFIG_AUFS_HNOTIFY
18622 +       struct au_hnotify       *hi_notify;
18623 +#endif
18624 +
18625 +       /* reference to the copied-up whiteout with get/put */
18626 +       struct dentry           *hi_whdentry;
18627 +};
18628 +
18629 +/* ig_flags */
18630 +#define AuIG_HALF_REFRESHED            1
18631 +#define au_ig_ftest(flags, name)       ((flags) & AuIG_##name)
18632 +#define au_ig_fset(flags, name) \
18633 +       do { (flags) |= AuIG_##name; } while (0)
18634 +#define au_ig_fclr(flags, name) \
18635 +       do { (flags) &= ~AuIG_##name; } while (0)
18636 +
18637 +struct au_iigen {
18638 +       spinlock_t      ig_spin;
18639 +       __u32           ig_generation, ig_flags;
18640 +};
18641 +
18642 +struct au_vdir;
18643 +struct au_iinfo {
18644 +       struct au_iigen         ii_generation;
18645 +       struct super_block      *ii_hsb1;       /* no get/put */
18646 +
18647 +       struct au_rwsem         ii_rwsem;
18648 +       aufs_bindex_t           ii_btop, ii_bbot;
18649 +       __u32                   ii_higen;
18650 +       struct au_hinode        *ii_hinode;
18651 +       struct au_vdir          *ii_vdir;
18652 +};
18653 +
18654 +struct au_icntnr {
18655 +       struct au_iinfo         iinfo;
18656 +       struct inode            vfs_inode;
18657 +       struct hlist_bl_node    plink;
18658 +       struct rcu_head         rcu;
18659 +} ____cacheline_aligned_in_smp;
18660 +
18661 +/* au_pin flags */
18662 +#define AuPin_DI_LOCKED                1
18663 +#define AuPin_MNT_WRITE                (1 << 1)
18664 +#define au_ftest_pin(flags, name)      ((flags) & AuPin_##name)
18665 +#define au_fset_pin(flags, name) \
18666 +       do { (flags) |= AuPin_##name; } while (0)
18667 +#define au_fclr_pin(flags, name) \
18668 +       do { (flags) &= ~AuPin_##name; } while (0)
18669 +
18670 +struct au_pin {
18671 +       /* input */
18672 +       struct dentry *dentry;
18673 +       unsigned int udba;
18674 +       unsigned char lsc_di, lsc_hi, flags;
18675 +       aufs_bindex_t bindex;
18676 +
18677 +       /* output */
18678 +       struct dentry *parent;
18679 +       struct au_hinode *hdir;
18680 +       struct vfsmount *h_mnt;
18681 +
18682 +       /* temporary unlock/relock for copyup */
18683 +       struct dentry *h_dentry, *h_parent;
18684 +       struct au_branch *br;
18685 +       struct task_struct *task;
18686 +};
18687 +
18688 +void au_pin_hdir_unlock(struct au_pin *p);
18689 +int au_pin_hdir_lock(struct au_pin *p);
18690 +int au_pin_hdir_relock(struct au_pin *p);
18691 +void au_pin_hdir_acquire_nest(struct au_pin *p);
18692 +void au_pin_hdir_release(struct au_pin *p);
18693 +
18694 +/* ---------------------------------------------------------------------- */
18695 +
18696 +static inline struct au_iinfo *au_ii(struct inode *inode)
18697 +{
18698 +       BUG_ON(is_bad_inode(inode));
18699 +       return &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
18700 +}
18701 +
18702 +/* ---------------------------------------------------------------------- */
18703 +
18704 +/* inode.c */
18705 +struct inode *au_igrab(struct inode *inode);
18706 +void au_refresh_iop(struct inode *inode, int force_getattr);
18707 +int au_refresh_hinode_self(struct inode *inode);
18708 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry);
18709 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
18710 +          unsigned int d_type, ino_t *ino);
18711 +struct inode *au_new_inode(struct dentry *dentry, int must_new);
18712 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
18713 +              struct inode *inode);
18714 +int au_test_h_perm(struct user_namespace *h_userns, struct inode *h_inode,
18715 +                  int mask);
18716 +int au_test_h_perm_sio(struct user_namespace *h_userns, struct inode *h_inode,
18717 +                      int mask);
18718 +
18719 +static inline int au_wh_ino(struct super_block *sb, aufs_bindex_t bindex,
18720 +                           ino_t h_ino, unsigned int d_type, ino_t *ino)
18721 +{
18722 +#ifdef CONFIG_AUFS_SHWH
18723 +       return au_ino(sb, bindex, h_ino, d_type, ino);
18724 +#else
18725 +       return 0;
18726 +#endif
18727 +}
18728 +
18729 +/* i_op.c */
18730 +enum {
18731 +       AuIop_SYMLINK,
18732 +       AuIop_DIR,
18733 +       AuIop_OTHER,
18734 +       AuIop_Last
18735 +};
18736 +extern struct inode_operations aufs_iop[AuIop_Last], /* not const */
18737 +       aufs_iop_nogetattr[AuIop_Last];
18738 +
18739 +/* au_wr_dir flags */
18740 +#define AuWrDir_ADD_ENTRY      1
18741 +#define AuWrDir_ISDIR          (1 << 1)
18742 +#define AuWrDir_TMPFILE                (1 << 2)
18743 +#define au_ftest_wrdir(flags, name)    ((flags) & AuWrDir_##name)
18744 +#define au_fset_wrdir(flags, name) \
18745 +       do { (flags) |= AuWrDir_##name; } while (0)
18746 +#define au_fclr_wrdir(flags, name) \
18747 +       do { (flags) &= ~AuWrDir_##name; } while (0)
18748 +
18749 +struct au_wr_dir_args {
18750 +       aufs_bindex_t force_btgt;
18751 +       unsigned char flags;
18752 +};
18753 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
18754 +             struct au_wr_dir_args *args);
18755 +
18756 +struct dentry *au_pinned_h_parent(struct au_pin *pin);
18757 +void au_pin_init(struct au_pin *pin, struct dentry *dentry,
18758 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
18759 +                unsigned int udba, unsigned char flags);
18760 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
18761 +          unsigned int udba, unsigned char flags) __must_check;
18762 +int au_do_pin(struct au_pin *pin) __must_check;
18763 +void au_unpin(struct au_pin *pin);
18764 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen);
18765 +
18766 +#define AuIcpup_DID_CPUP       1
18767 +#define au_ftest_icpup(flags, name)    ((flags) & AuIcpup_##name)
18768 +#define au_fset_icpup(flags, name) \
18769 +       do { (flags) |= AuIcpup_##name; } while (0)
18770 +#define au_fclr_icpup(flags, name) \
18771 +       do { (flags) &= ~AuIcpup_##name; } while (0)
18772 +
18773 +struct au_icpup_args {
18774 +       unsigned char flags;
18775 +       unsigned char pin_flags;
18776 +       aufs_bindex_t btgt;
18777 +       unsigned int udba;
18778 +       struct au_pin pin;
18779 +       struct path h_path;
18780 +       struct inode *h_inode;
18781 +};
18782 +
18783 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
18784 +                    struct au_icpup_args *a);
18785 +
18786 +int au_h_path_getattr(struct dentry *dentry, struct inode *inode, int force,
18787 +                     struct path *h_path, int locked);
18788 +
18789 +/* i_op_add.c */
18790 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
18791 +              struct dentry *h_parent, int isdir);
18792 +int aufs_mknod(struct user_namespace *userns, struct inode *dir,
18793 +              struct dentry *dentry, umode_t mode, dev_t dev);
18794 +int aufs_symlink(struct user_namespace *userns, struct inode *dir,
18795 +                struct dentry *dentry, const char *symname);
18796 +int aufs_create(struct user_namespace *userns, struct inode *dir,
18797 +               struct dentry *dentry, umode_t mode, bool want_excl);
18798 +struct vfsub_aopen_args;
18799 +int au_aopen_or_create(struct inode *dir, struct dentry *dentry,
18800 +                      struct vfsub_aopen_args *args);
18801 +int aufs_tmpfile(struct user_namespace *userns, struct inode *dir,
18802 +                struct dentry *dentry, umode_t mode);
18803 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
18804 +             struct dentry *dentry);
18805 +int aufs_mkdir(struct user_namespace *userns, struct inode *dir,
18806 +              struct dentry *dentry, umode_t mode);
18807 +
18808 +/* i_op_del.c */
18809 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup);
18810 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
18811 +              struct dentry *h_parent, int isdir);
18812 +int aufs_unlink(struct inode *dir, struct dentry *dentry);
18813 +int aufs_rmdir(struct inode *dir, struct dentry *dentry);
18814 +
18815 +/* i_op_ren.c */
18816 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt);
18817 +int aufs_rename(struct user_namespace *userns,
18818 +               struct inode *_src_dir, struct dentry *_src_dentry,
18819 +               struct inode *_dst_dir, struct dentry *_dst_dentry,
18820 +               unsigned int _flags);
18821 +
18822 +/* iinfo.c */
18823 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex);
18824 +void au_hiput(struct au_hinode *hinode);
18825 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
18826 +                 struct dentry *h_wh);
18827 +unsigned int au_hi_flags(struct inode *inode, int isdir);
18828 +
18829 +/* hinode flags */
18830 +#define AuHi_XINO      1
18831 +#define AuHi_HNOTIFY   (1 << 1)
18832 +#define au_ftest_hi(flags, name)       ((flags) & AuHi_##name)
18833 +#define au_fset_hi(flags, name) \
18834 +       do { (flags) |= AuHi_##name; } while (0)
18835 +#define au_fclr_hi(flags, name) \
18836 +       do { (flags) &= ~AuHi_##name; } while (0)
18837 +
18838 +#ifndef CONFIG_AUFS_HNOTIFY
18839 +#undef AuHi_HNOTIFY
18840 +#define AuHi_HNOTIFY   0
18841 +#endif
18842 +
18843 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
18844 +                  struct inode *h_inode, unsigned int flags);
18845 +
18846 +void au_update_iigen(struct inode *inode, int half);
18847 +void au_update_ibrange(struct inode *inode, int do_put_zero);
18848 +
18849 +void au_icntnr_init_once(void *_c);
18850 +void au_hinode_init(struct au_hinode *hinode);
18851 +int au_iinfo_init(struct inode *inode);
18852 +void au_iinfo_fin(struct inode *inode);
18853 +int au_hinode_realloc(struct au_iinfo *iinfo, int nbr, int may_shrink);
18854 +
18855 +#ifdef CONFIG_PROC_FS
18856 +/* plink.c */
18857 +int au_plink_maint(struct super_block *sb, int flags);
18858 +struct au_sbinfo;
18859 +void au_plink_maint_leave(struct au_sbinfo *sbinfo);
18860 +int au_plink_maint_enter(struct super_block *sb);
18861 +#ifdef CONFIG_AUFS_DEBUG
18862 +void au_plink_list(struct super_block *sb);
18863 +#else
18864 +AuStubVoid(au_plink_list, struct super_block *sb)
18865 +#endif
18866 +int au_plink_test(struct inode *inode);
18867 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex);
18868 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
18869 +                    struct dentry *h_dentry);
18870 +void au_plink_put(struct super_block *sb, int verbose);
18871 +void au_plink_clean(struct super_block *sb, int verbose);
18872 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id);
18873 +#else
18874 +AuStubInt0(au_plink_maint, struct super_block *sb, int flags);
18875 +AuStubVoid(au_plink_maint_leave, struct au_sbinfo *sbinfo);
18876 +AuStubInt0(au_plink_maint_enter, struct super_block *sb);
18877 +AuStubVoid(au_plink_list, struct super_block *sb);
18878 +AuStubInt0(au_plink_test, struct inode *inode);
18879 +AuStub(struct dentry *, au_plink_lkup, return NULL,
18880 +       struct inode *inode, aufs_bindex_t bindex);
18881 +AuStubVoid(au_plink_append, struct inode *inode, aufs_bindex_t bindex,
18882 +          struct dentry *h_dentry);
18883 +AuStubVoid(au_plink_put, struct super_block *sb, int verbose);
18884 +AuStubVoid(au_plink_clean, struct super_block *sb, int verbose);
18885 +AuStubVoid(au_plink_half_refresh, struct super_block *sb, aufs_bindex_t br_id);
18886 +#endif /* CONFIG_PROC_FS */
18887 +
18888 +#ifdef CONFIG_AUFS_XATTR
18889 +/* xattr.c */
18890 +int au_cpup_xattr(struct path *h_dst, struct path *h_src, int ignore_flags,
18891 +                 unsigned int verbose);
18892 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size);
18893 +void au_xattr_init(struct super_block *sb);
18894 +#else
18895 +AuStubInt0(au_cpup_xattr, struct path *h_dst, struct path *h_src,
18896 +          int ignore_flags, unsigned int verbose);
18897 +AuStubVoid(au_xattr_init, struct super_block *sb);
18898 +#endif
18899 +
18900 +#ifdef CONFIG_FS_POSIX_ACL
18901 +struct posix_acl *aufs_get_acl(struct inode *inode, int type, bool rcu);
18902 +int aufs_set_acl(struct user_namespace *userns, struct inode *inode,
18903 +                struct posix_acl *acl, int type);
18904 +#endif
18905 +
18906 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
18907 +enum {
18908 +       AU_XATTR_SET,
18909 +       AU_ACL_SET
18910 +};
18911 +
18912 +struct au_sxattr {
18913 +       int type;
18914 +       union {
18915 +               struct {
18916 +                       const char      *name;
18917 +                       const void      *value;
18918 +                       size_t          size;
18919 +                       int             flags;
18920 +               } set;
18921 +               struct {
18922 +                       struct posix_acl *acl;
18923 +                       int             type;
18924 +               } acl_set;
18925 +       } u;
18926 +};
18927 +ssize_t au_sxattr(struct dentry *dentry, struct inode *inode,
18928 +                 struct au_sxattr *arg);
18929 +#endif
18930 +
18931 +/* ---------------------------------------------------------------------- */
18932 +
18933 +/* lock subclass for iinfo */
18934 +enum {
18935 +       AuLsc_II_CHILD,         /* child first */
18936 +       AuLsc_II_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
18937 +       AuLsc_II_CHILD3,        /* copyup dirs */
18938 +       AuLsc_II_PARENT,        /* see AuLsc_I_PARENT in vfsub.h */
18939 +       AuLsc_II_PARENT2,
18940 +       AuLsc_II_PARENT3,       /* copyup dirs */
18941 +       AuLsc_II_NEW_CHILD
18942 +};
18943 +
18944 +/*
18945 + * ii_read_lock_child, ii_write_lock_child,
18946 + * ii_read_lock_child2, ii_write_lock_child2,
18947 + * ii_read_lock_child3, ii_write_lock_child3,
18948 + * ii_read_lock_parent, ii_write_lock_parent,
18949 + * ii_read_lock_parent2, ii_write_lock_parent2,
18950 + * ii_read_lock_parent3, ii_write_lock_parent3,
18951 + * ii_read_lock_new_child, ii_write_lock_new_child,
18952 + */
18953 +#define AuReadLockFunc(name, lsc) \
18954 +static inline void ii_read_lock_##name(struct inode *i) \
18955 +{ \
18956 +       au_rw_read_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
18957 +}
18958 +
18959 +#define AuWriteLockFunc(name, lsc) \
18960 +static inline void ii_write_lock_##name(struct inode *i) \
18961 +{ \
18962 +       au_rw_write_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
18963 +}
18964 +
18965 +#define AuRWLockFuncs(name, lsc) \
18966 +       AuReadLockFunc(name, lsc) \
18967 +       AuWriteLockFunc(name, lsc)
18968 +
18969 +AuRWLockFuncs(child, CHILD);
18970 +AuRWLockFuncs(child2, CHILD2);
18971 +AuRWLockFuncs(child3, CHILD3);
18972 +AuRWLockFuncs(parent, PARENT);
18973 +AuRWLockFuncs(parent2, PARENT2);
18974 +AuRWLockFuncs(parent3, PARENT3);
18975 +AuRWLockFuncs(new_child, NEW_CHILD);
18976 +
18977 +#undef AuReadLockFunc
18978 +#undef AuWriteLockFunc
18979 +#undef AuRWLockFuncs
18980 +
18981 +#define ii_read_unlock(i)      au_rw_read_unlock(&au_ii(i)->ii_rwsem)
18982 +#define ii_write_unlock(i)     au_rw_write_unlock(&au_ii(i)->ii_rwsem)
18983 +#define ii_downgrade_lock(i)   au_rw_dgrade_lock(&au_ii(i)->ii_rwsem)
18984 +
18985 +#define IiMustNoWaiters(i)     AuRwMustNoWaiters(&au_ii(i)->ii_rwsem)
18986 +#define IiMustAnyLock(i)       AuRwMustAnyLock(&au_ii(i)->ii_rwsem)
18987 +#define IiMustWriteLock(i)     AuRwMustWriteLock(&au_ii(i)->ii_rwsem)
18988 +
18989 +/* ---------------------------------------------------------------------- */
18990 +
18991 +static inline void au_icntnr_init(struct au_icntnr *c)
18992 +{
18993 +#ifdef CONFIG_AUFS_DEBUG
18994 +       c->vfs_inode.i_mode = 0;
18995 +#endif
18996 +}
18997 +
18998 +static inline unsigned int au_iigen(struct inode *inode, unsigned int *igflags)
18999 +{
19000 +       unsigned int gen;
19001 +       struct au_iinfo *iinfo;
19002 +       struct au_iigen *iigen;
19003 +
19004 +       iinfo = au_ii(inode);
19005 +       iigen = &iinfo->ii_generation;
19006 +       spin_lock(&iigen->ig_spin);
19007 +       if (igflags)
19008 +               *igflags = iigen->ig_flags;
19009 +       gen = iigen->ig_generation;
19010 +       spin_unlock(&iigen->ig_spin);
19011 +
19012 +       return gen;
19013 +}
19014 +
19015 +/* tiny test for inode number */
19016 +/* tmpfs generation is too rough */
19017 +static inline int au_test_higen(struct inode *inode, struct inode *h_inode)
19018 +{
19019 +       struct au_iinfo *iinfo;
19020 +
19021 +       iinfo = au_ii(inode);
19022 +       AuRwMustAnyLock(&iinfo->ii_rwsem);
19023 +       return !(iinfo->ii_hsb1 == h_inode->i_sb
19024 +                && iinfo->ii_higen == h_inode->i_generation);
19025 +}
19026 +
19027 +static inline void au_iigen_dec(struct inode *inode)
19028 +{
19029 +       struct au_iinfo *iinfo;
19030 +       struct au_iigen *iigen;
19031 +
19032 +       iinfo = au_ii(inode);
19033 +       iigen = &iinfo->ii_generation;
19034 +       spin_lock(&iigen->ig_spin);
19035 +       iigen->ig_generation--;
19036 +       spin_unlock(&iigen->ig_spin);
19037 +}
19038 +
19039 +static inline int au_iigen_test(struct inode *inode, unsigned int sigen)
19040 +{
19041 +       int err;
19042 +
19043 +       err = 0;
19044 +       if (unlikely(inode && au_iigen(inode, NULL) != sigen))
19045 +               err = -EIO;
19046 +
19047 +       return err;
19048 +}
19049 +
19050 +/* ---------------------------------------------------------------------- */
19051 +
19052 +static inline struct au_hinode *au_hinode(struct au_iinfo *iinfo,
19053 +                                         aufs_bindex_t bindex)
19054 +{
19055 +       return iinfo->ii_hinode + bindex;
19056 +}
19057 +
19058 +static inline int au_is_bad_inode(struct inode *inode)
19059 +{
19060 +       return !!(is_bad_inode(inode) || !au_hinode(au_ii(inode), 0));
19061 +}
19062 +
19063 +static inline aufs_bindex_t au_ii_br_id(struct inode *inode,
19064 +                                       aufs_bindex_t bindex)
19065 +{
19066 +       IiMustAnyLock(inode);
19067 +       return au_hinode(au_ii(inode), bindex)->hi_id;
19068 +}
19069 +
19070 +static inline aufs_bindex_t au_ibtop(struct inode *inode)
19071 +{
19072 +       IiMustAnyLock(inode);
19073 +       return au_ii(inode)->ii_btop;
19074 +}
19075 +
19076 +static inline aufs_bindex_t au_ibbot(struct inode *inode)
19077 +{
19078 +       IiMustAnyLock(inode);
19079 +       return au_ii(inode)->ii_bbot;
19080 +}
19081 +
19082 +static inline struct au_vdir *au_ivdir(struct inode *inode)
19083 +{
19084 +       IiMustAnyLock(inode);
19085 +       return au_ii(inode)->ii_vdir;
19086 +}
19087 +
19088 +static inline struct dentry *au_hi_wh(struct inode *inode, aufs_bindex_t bindex)
19089 +{
19090 +       IiMustAnyLock(inode);
19091 +       return au_hinode(au_ii(inode), bindex)->hi_whdentry;
19092 +}
19093 +
19094 +static inline void au_set_ibtop(struct inode *inode, aufs_bindex_t bindex)
19095 +{
19096 +       IiMustWriteLock(inode);
19097 +       au_ii(inode)->ii_btop = bindex;
19098 +}
19099 +
19100 +static inline void au_set_ibbot(struct inode *inode, aufs_bindex_t bindex)
19101 +{
19102 +       IiMustWriteLock(inode);
19103 +       au_ii(inode)->ii_bbot = bindex;
19104 +}
19105 +
19106 +static inline void au_set_ivdir(struct inode *inode, struct au_vdir *vdir)
19107 +{
19108 +       IiMustWriteLock(inode);
19109 +       au_ii(inode)->ii_vdir = vdir;
19110 +}
19111 +
19112 +static inline struct au_hinode *au_hi(struct inode *inode, aufs_bindex_t bindex)
19113 +{
19114 +       IiMustAnyLock(inode);
19115 +       return au_hinode(au_ii(inode), bindex);
19116 +}
19117 +
19118 +/* ---------------------------------------------------------------------- */
19119 +
19120 +static inline struct dentry *au_pinned_parent(struct au_pin *pin)
19121 +{
19122 +       if (pin)
19123 +               return pin->parent;
19124 +       return NULL;
19125 +}
19126 +
19127 +static inline struct inode *au_pinned_h_dir(struct au_pin *pin)
19128 +{
19129 +       if (pin && pin->hdir)
19130 +               return pin->hdir->hi_inode;
19131 +       return NULL;
19132 +}
19133 +
19134 +static inline struct au_hinode *au_pinned_hdir(struct au_pin *pin)
19135 +{
19136 +       if (pin)
19137 +               return pin->hdir;
19138 +       return NULL;
19139 +}
19140 +
19141 +static inline void au_pin_set_dentry(struct au_pin *pin, struct dentry *dentry)
19142 +{
19143 +       if (pin)
19144 +               pin->dentry = dentry;
19145 +}
19146 +
19147 +static inline void au_pin_set_parent_lflag(struct au_pin *pin,
19148 +                                          unsigned char lflag)
19149 +{
19150 +       if (pin) {
19151 +               if (lflag)
19152 +                       au_fset_pin(pin->flags, DI_LOCKED);
19153 +               else
19154 +                       au_fclr_pin(pin->flags, DI_LOCKED);
19155 +       }
19156 +}
19157 +
19158 +#if 0 /* reserved */
19159 +static inline void au_pin_set_parent(struct au_pin *pin, struct dentry *parent)
19160 +{
19161 +       if (pin) {
19162 +               dput(pin->parent);
19163 +               pin->parent = dget(parent);
19164 +       }
19165 +}
19166 +#endif
19167 +
19168 +/* ---------------------------------------------------------------------- */
19169 +
19170 +struct au_branch;
19171 +#ifdef CONFIG_AUFS_HNOTIFY
19172 +struct au_hnotify_op {
19173 +       void (*ctl)(struct au_hinode *hinode, int do_set);
19174 +       int (*alloc)(struct au_hinode *hinode);
19175 +
19176 +       /*
19177 +        * if it returns true, the caller should free hinode->hi_notify,
19178 +        * otherwise ->free() frees it.
19179 +        */
19180 +       int (*free)(struct au_hinode *hinode,
19181 +                   struct au_hnotify *hn) __must_check;
19182 +
19183 +       void (*fin)(void);
19184 +       int (*init)(void);
19185 +
19186 +       int (*reset_br)(unsigned int udba, struct au_branch *br, int perm);
19187 +       void (*fin_br)(struct au_branch *br);
19188 +       int (*init_br)(struct au_branch *br, int perm);
19189 +};
19190 +
19191 +/* hnotify.c */
19192 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode);
19193 +void au_hn_free(struct au_hinode *hinode);
19194 +void au_hn_ctl(struct au_hinode *hinode, int do_set);
19195 +void au_hn_reset(struct inode *inode, unsigned int flags);
19196 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
19197 +              const struct qstr *h_child_qstr, struct inode *h_child_inode);
19198 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm);
19199 +int au_hnotify_init_br(struct au_branch *br, int perm);
19200 +void au_hnotify_fin_br(struct au_branch *br);
19201 +int __init au_hnotify_init(void);
19202 +void au_hnotify_fin(void);
19203 +
19204 +/* hfsnotify.c */
19205 +extern const struct au_hnotify_op au_hnotify_op;
19206 +
19207 +static inline
19208 +void au_hn_init(struct au_hinode *hinode)
19209 +{
19210 +       hinode->hi_notify = NULL;
19211 +}
19212 +
19213 +static inline struct au_hnotify *au_hn(struct au_hinode *hinode)
19214 +{
19215 +       return hinode->hi_notify;
19216 +}
19217 +
19218 +#else
19219 +AuStub(int, au_hn_alloc, return -EOPNOTSUPP,
19220 +       struct au_hinode *hinode __maybe_unused,
19221 +       struct inode *inode __maybe_unused)
19222 +AuStub(struct au_hnotify *, au_hn, return NULL, struct au_hinode *hinode)
19223 +AuStubVoid(au_hn_free, struct au_hinode *hinode __maybe_unused)
19224 +AuStubVoid(au_hn_ctl, struct au_hinode *hinode __maybe_unused,
19225 +          int do_set __maybe_unused)
19226 +AuStubVoid(au_hn_reset, struct inode *inode __maybe_unused,
19227 +          unsigned int flags __maybe_unused)
19228 +AuStubInt0(au_hnotify_reset_br, unsigned int udba __maybe_unused,
19229 +          struct au_branch *br __maybe_unused,
19230 +          int perm __maybe_unused)
19231 +AuStubInt0(au_hnotify_init_br, struct au_branch *br __maybe_unused,
19232 +          int perm __maybe_unused)
19233 +AuStubVoid(au_hnotify_fin_br, struct au_branch *br __maybe_unused)
19234 +AuStubInt0(__init au_hnotify_init, void)
19235 +AuStubVoid(au_hnotify_fin, void)
19236 +AuStubVoid(au_hn_init, struct au_hinode *hinode __maybe_unused)
19237 +#endif /* CONFIG_AUFS_HNOTIFY */
19238 +
19239 +static inline void au_hn_suspend(struct au_hinode *hdir)
19240 +{
19241 +       au_hn_ctl(hdir, /*do_set*/0);
19242 +}
19243 +
19244 +static inline void au_hn_resume(struct au_hinode *hdir)
19245 +{
19246 +       au_hn_ctl(hdir, /*do_set*/1);
19247 +}
19248 +
19249 +static inline void au_hn_inode_lock(struct au_hinode *hdir)
19250 +{
19251 +       inode_lock(hdir->hi_inode);
19252 +       au_hn_suspend(hdir);
19253 +}
19254 +
19255 +static inline void au_hn_inode_lock_nested(struct au_hinode *hdir,
19256 +                                         unsigned int sc __maybe_unused)
19257 +{
19258 +       inode_lock_nested(hdir->hi_inode, sc);
19259 +       au_hn_suspend(hdir);
19260 +}
19261 +
19262 +#if 0 /* unused */
19263 +#include "vfsub.h"
19264 +static inline void au_hn_inode_lock_shared_nested(struct au_hinode *hdir,
19265 +                                                 unsigned int sc)
19266 +{
19267 +       inode_lock_shared_nested(hdir->hi_inode, sc);
19268 +       au_hn_suspend(hdir);
19269 +}
19270 +#endif
19271 +
19272 +static inline void au_hn_inode_unlock(struct au_hinode *hdir)
19273 +{
19274 +       au_hn_resume(hdir);
19275 +       inode_unlock(hdir->hi_inode);
19276 +}
19277 +
19278 +#endif /* __KERNEL__ */
19279 +#endif /* __AUFS_INODE_H__ */
19280 diff -urN /usr/share/empty/fs/aufs/ioctl.c linux/fs/aufs/ioctl.c
19281 --- /usr/share/empty/fs/aufs/ioctl.c    1970-01-01 01:00:00.000000000 +0100
19282 +++ linux/fs/aufs/ioctl.c       2021-12-03 15:38:59.936647310 +0100
19283 @@ -0,0 +1,220 @@
19284 +// SPDX-License-Identifier: GPL-2.0
19285 +/*
19286 + * Copyright (C) 2005-2021 Junjiro R. Okajima
19287 + *
19288 + * This program, aufs is free software; you can redistribute it and/or modify
19289 + * it under the terms of the GNU General Public License as published by
19290 + * the Free Software Foundation; either version 2 of the License, or
19291 + * (at your option) any later version.
19292 + *
19293 + * This program is distributed in the hope that it will be useful,
19294 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19295 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19296 + * GNU General Public License for more details.
19297 + *
19298 + * You should have received a copy of the GNU General Public License
19299 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19300 + */
19301 +
19302 +/*
19303 + * ioctl
19304 + * plink-management and readdir in userspace.
19305 + * assist the pathconf(3) wrapper library.
19306 + * move-down
19307 + * File-based Hierarchical Storage Management.
19308 + */
19309 +
19310 +#include <linux/compat.h>
19311 +#include <linux/file.h>
19312 +#include "aufs.h"
19313 +
19314 +static int au_wbr_fd(struct path *path, struct aufs_wbr_fd __user *arg)
19315 +{
19316 +       int err, fd;
19317 +       aufs_bindex_t wbi, bindex, bbot;
19318 +       struct file *h_file;
19319 +       struct super_block *sb;
19320 +       struct dentry *root;
19321 +       struct au_branch *br;
19322 +       struct aufs_wbr_fd wbrfd = {
19323 +               .oflags = au_dir_roflags,
19324 +               .brid   = -1
19325 +       };
19326 +       const int valid = O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_DIRECTORY
19327 +               | O_NOATIME | O_CLOEXEC;
19328 +
19329 +       AuDebugOn(wbrfd.oflags & ~valid);
19330 +
19331 +       if (arg) {
19332 +               err = copy_from_user(&wbrfd, arg, sizeof(wbrfd));
19333 +               if (unlikely(err)) {
19334 +                       err = -EFAULT;
19335 +                       goto out;
19336 +               }
19337 +
19338 +               err = -EINVAL;
19339 +               AuDbg("wbrfd{0%o, %d}\n", wbrfd.oflags, wbrfd.brid);
19340 +               wbrfd.oflags |= au_dir_roflags;
19341 +               AuDbg("0%o\n", wbrfd.oflags);
19342 +               if (unlikely(wbrfd.oflags & ~valid))
19343 +                       goto out;
19344 +       }
19345 +
19346 +       fd = get_unused_fd_flags(0);
19347 +       err = fd;
19348 +       if (unlikely(fd < 0))
19349 +               goto out;
19350 +
19351 +       h_file = ERR_PTR(-EINVAL);
19352 +       wbi = 0;
19353 +       br = NULL;
19354 +       sb = path->dentry->d_sb;
19355 +       root = sb->s_root;
19356 +       aufs_read_lock(root, AuLock_IR);
19357 +       bbot = au_sbbot(sb);
19358 +       if (wbrfd.brid >= 0) {
19359 +               wbi = au_br_index(sb, wbrfd.brid);
19360 +               if (unlikely(wbi < 0 || wbi > bbot))
19361 +                       goto out_unlock;
19362 +       }
19363 +
19364 +       h_file = ERR_PTR(-ENOENT);
19365 +       br = au_sbr(sb, wbi);
19366 +       if (!au_br_writable(br->br_perm)) {
19367 +               if (arg)
19368 +                       goto out_unlock;
19369 +
19370 +               bindex = wbi + 1;
19371 +               wbi = -1;
19372 +               for (; bindex <= bbot; bindex++) {
19373 +                       br = au_sbr(sb, bindex);
19374 +                       if (au_br_writable(br->br_perm)) {
19375 +                               wbi = bindex;
19376 +                               br = au_sbr(sb, wbi);
19377 +                               break;
19378 +                       }
19379 +               }
19380 +       }
19381 +       AuDbg("wbi %d\n", wbi);
19382 +       if (wbi >= 0)
19383 +               h_file = au_h_open(root, wbi, wbrfd.oflags, NULL,
19384 +                                  /*force_wr*/0);
19385 +
19386 +out_unlock:
19387 +       aufs_read_unlock(root, AuLock_IR);
19388 +       err = PTR_ERR(h_file);
19389 +       if (IS_ERR(h_file))
19390 +               goto out_fd;
19391 +
19392 +       au_lcnt_dec(&br->br_nfiles); /* cf. au_h_open() */
19393 +       fd_install(fd, h_file);
19394 +       err = fd;
19395 +       goto out; /* success */
19396 +
19397 +out_fd:
19398 +       put_unused_fd(fd);
19399 +out:
19400 +       AuTraceErr(err);
19401 +       return err;
19402 +}
19403 +
19404 +/* ---------------------------------------------------------------------- */
19405 +
19406 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg)
19407 +{
19408 +       long err;
19409 +       struct dentry *dentry;
19410 +
19411 +       switch (cmd) {
19412 +       case AUFS_CTL_RDU:
19413 +       case AUFS_CTL_RDU_INO:
19414 +               err = au_rdu_ioctl(file, cmd, arg);
19415 +               break;
19416 +
19417 +       case AUFS_CTL_WBR_FD:
19418 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
19419 +               break;
19420 +
19421 +       case AUFS_CTL_IBUSY:
19422 +               err = au_ibusy_ioctl(file, arg);
19423 +               break;
19424 +
19425 +       case AUFS_CTL_BRINFO:
19426 +               err = au_brinfo_ioctl(file, arg);
19427 +               break;
19428 +
19429 +       case AUFS_CTL_FHSM_FD:
19430 +               dentry = file->f_path.dentry;
19431 +               if (IS_ROOT(dentry))
19432 +                       err = au_fhsm_fd(dentry->d_sb, arg);
19433 +               else
19434 +                       err = -ENOTTY;
19435 +               break;
19436 +
19437 +       default:
19438 +               /* do not call the lower */
19439 +               AuDbg("0x%x\n", cmd);
19440 +               err = -ENOTTY;
19441 +       }
19442 +
19443 +       AuTraceErr(err);
19444 +       return err;
19445 +}
19446 +
19447 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg)
19448 +{
19449 +       long err;
19450 +
19451 +       switch (cmd) {
19452 +       case AUFS_CTL_MVDOWN:
19453 +               err = au_mvdown(file->f_path.dentry, (void __user *)arg);
19454 +               break;
19455 +
19456 +       case AUFS_CTL_WBR_FD:
19457 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
19458 +               break;
19459 +
19460 +       default:
19461 +               /* do not call the lower */
19462 +               AuDbg("0x%x\n", cmd);
19463 +               err = -ENOTTY;
19464 +       }
19465 +
19466 +       AuTraceErr(err);
19467 +       return err;
19468 +}
19469 +
19470 +#ifdef CONFIG_COMPAT
19471 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
19472 +                          unsigned long arg)
19473 +{
19474 +       long err;
19475 +
19476 +       switch (cmd) {
19477 +       case AUFS_CTL_RDU:
19478 +       case AUFS_CTL_RDU_INO:
19479 +               err = au_rdu_compat_ioctl(file, cmd, arg);
19480 +               break;
19481 +
19482 +       case AUFS_CTL_IBUSY:
19483 +               err = au_ibusy_compat_ioctl(file, arg);
19484 +               break;
19485 +
19486 +       case AUFS_CTL_BRINFO:
19487 +               err = au_brinfo_compat_ioctl(file, arg);
19488 +               break;
19489 +
19490 +       default:
19491 +               err = aufs_ioctl_dir(file, cmd, arg);
19492 +       }
19493 +
19494 +       AuTraceErr(err);
19495 +       return err;
19496 +}
19497 +
19498 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
19499 +                             unsigned long arg)
19500 +{
19501 +       return aufs_ioctl_nondir(file, cmd, (unsigned long)compat_ptr(arg));
19502 +}
19503 +#endif
19504 diff -urN /usr/share/empty/fs/aufs/i_op_add.c linux/fs/aufs/i_op_add.c
19505 --- /usr/share/empty/fs/aufs/i_op_add.c 1970-01-01 01:00:00.000000000 +0100
19506 +++ linux/fs/aufs/i_op_add.c    2021-12-03 15:40:58.236647297 +0100
19507 @@ -0,0 +1,941 @@
19508 +// SPDX-License-Identifier: GPL-2.0
19509 +/*
19510 + * Copyright (C) 2005-2021 Junjiro R. Okajima
19511 + *
19512 + * This program, aufs is free software; you can redistribute it and/or modify
19513 + * it under the terms of the GNU General Public License as published by
19514 + * the Free Software Foundation; either version 2 of the License, or
19515 + * (at your option) any later version.
19516 + *
19517 + * This program is distributed in the hope that it will be useful,
19518 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19519 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19520 + * GNU General Public License for more details.
19521 + *
19522 + * You should have received a copy of the GNU General Public License
19523 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19524 + */
19525 +
19526 +/*
19527 + * inode operations (add entry)
19528 + */
19529 +
19530 +#include <linux/iversion.h>
19531 +#include "aufs.h"
19532 +
19533 +/*
19534 + * final procedure of adding a new entry, except link(2).
19535 + * remove whiteout, instantiate, copyup the parent dir's times and size
19536 + * and update version.
19537 + * if it failed, re-create the removed whiteout.
19538 + */
19539 +static int epilog(struct inode *dir, aufs_bindex_t bindex,
19540 +                 struct dentry *wh_dentry, struct dentry *dentry)
19541 +{
19542 +       int err, rerr;
19543 +       aufs_bindex_t bwh;
19544 +       struct path h_path;
19545 +       struct super_block *sb;
19546 +       struct inode *inode, *h_dir;
19547 +       struct dentry *wh;
19548 +
19549 +       bwh = -1;
19550 +       sb = dir->i_sb;
19551 +       if (wh_dentry) {
19552 +               h_dir = d_inode(wh_dentry->d_parent); /* dir inode is locked */
19553 +               IMustLock(h_dir);
19554 +               AuDebugOn(au_h_iptr(dir, bindex) != h_dir);
19555 +               bwh = au_dbwh(dentry);
19556 +               h_path.dentry = wh_dentry;
19557 +               h_path.mnt = au_sbr_mnt(sb, bindex);
19558 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path,
19559 +                                         dentry);
19560 +               if (unlikely(err))
19561 +                       goto out;
19562 +       }
19563 +
19564 +       inode = au_new_inode(dentry, /*must_new*/1);
19565 +       if (!IS_ERR(inode)) {
19566 +               d_instantiate(dentry, inode);
19567 +               dir = d_inode(dentry->d_parent); /* dir inode is locked */
19568 +               IMustLock(dir);
19569 +               au_dir_ts(dir, bindex);
19570 +               inode_inc_iversion(dir);
19571 +               au_fhsm_wrote(sb, bindex, /*force*/0);
19572 +               return 0; /* success */
19573 +       }
19574 +
19575 +       err = PTR_ERR(inode);
19576 +       if (!wh_dentry)
19577 +               goto out;
19578 +
19579 +       /* revert */
19580 +       /* dir inode is locked */
19581 +       wh = au_wh_create(dentry, bwh, wh_dentry->d_parent);
19582 +       rerr = PTR_ERR(wh);
19583 +       if (IS_ERR(wh)) {
19584 +               AuIOErr("%pd reverting whiteout failed(%d, %d)\n",
19585 +                       dentry, err, rerr);
19586 +               err = -EIO;
19587 +       } else
19588 +               dput(wh);
19589 +
19590 +out:
19591 +       return err;
19592 +}
19593 +
19594 +static int au_d_may_add(struct dentry *dentry)
19595 +{
19596 +       int err;
19597 +
19598 +       err = 0;
19599 +       if (unlikely(d_unhashed(dentry)))
19600 +               err = -ENOENT;
19601 +       if (unlikely(d_really_is_positive(dentry)))
19602 +               err = -EEXIST;
19603 +       return err;
19604 +}
19605 +
19606 +/*
19607 + * simple tests for the adding inode operations.
19608 + * following the checks in vfs, plus the parent-child relationship.
19609 + */
19610 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
19611 +              struct dentry *h_parent, int isdir)
19612 +{
19613 +       int err;
19614 +       umode_t h_mode;
19615 +       struct dentry *h_dentry;
19616 +       struct inode *h_inode;
19617 +
19618 +       err = -ENAMETOOLONG;
19619 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
19620 +               goto out;
19621 +
19622 +       h_dentry = au_h_dptr(dentry, bindex);
19623 +       if (d_really_is_negative(dentry)) {
19624 +               err = -EEXIST;
19625 +               if (unlikely(d_is_positive(h_dentry)))
19626 +                       goto out;
19627 +       } else {
19628 +               /* rename(2) case */
19629 +               err = -EIO;
19630 +               if (unlikely(d_is_negative(h_dentry)))
19631 +                       goto out;
19632 +               h_inode = d_inode(h_dentry);
19633 +               if (unlikely(!h_inode->i_nlink))
19634 +                       goto out;
19635 +
19636 +               h_mode = h_inode->i_mode;
19637 +               if (!isdir) {
19638 +                       err = -EISDIR;
19639 +                       if (unlikely(S_ISDIR(h_mode)))
19640 +                               goto out;
19641 +               } else if (unlikely(!S_ISDIR(h_mode))) {
19642 +                       err = -ENOTDIR;
19643 +                       goto out;
19644 +               }
19645 +       }
19646 +
19647 +       err = 0;
19648 +       /* expected parent dir is locked */
19649 +       if (unlikely(h_parent != h_dentry->d_parent))
19650 +               err = -EIO;
19651 +
19652 +out:
19653 +       AuTraceErr(err);
19654 +       return err;
19655 +}
19656 +
19657 +/*
19658 + * initial procedure of adding a new entry.
19659 + * prepare writable branch and the parent dir, lock it,
19660 + * and lookup whiteout for the new entry.
19661 + */
19662 +static struct dentry*
19663 +lock_hdir_lkup_wh(struct dentry *dentry, struct au_dtime *dt,
19664 +                 struct dentry *src_dentry, struct au_pin *pin,
19665 +                 struct au_wr_dir_args *wr_dir_args)
19666 +{
19667 +       struct dentry *wh_dentry, *h_parent;
19668 +       struct super_block *sb;
19669 +       struct au_branch *br;
19670 +       int err;
19671 +       unsigned int udba;
19672 +       aufs_bindex_t bcpup;
19673 +
19674 +       AuDbg("%pd\n", dentry);
19675 +
19676 +       err = au_wr_dir(dentry, src_dentry, wr_dir_args);
19677 +       bcpup = err;
19678 +       wh_dentry = ERR_PTR(err);
19679 +       if (unlikely(err < 0))
19680 +               goto out;
19681 +
19682 +       sb = dentry->d_sb;
19683 +       udba = au_opt_udba(sb);
19684 +       err = au_pin(pin, dentry, bcpup, udba,
19685 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
19686 +       wh_dentry = ERR_PTR(err);
19687 +       if (unlikely(err))
19688 +               goto out;
19689 +
19690 +       h_parent = au_pinned_h_parent(pin);
19691 +       if (udba != AuOpt_UDBA_NONE
19692 +           && au_dbtop(dentry) == bcpup)
19693 +               err = au_may_add(dentry, bcpup, h_parent,
19694 +                                au_ftest_wrdir(wr_dir_args->flags, ISDIR));
19695 +       else if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
19696 +               err = -ENAMETOOLONG;
19697 +       wh_dentry = ERR_PTR(err);
19698 +       if (unlikely(err))
19699 +               goto out_unpin;
19700 +
19701 +       br = au_sbr(sb, bcpup);
19702 +       if (dt) {
19703 +               struct path tmp = {
19704 +                       .dentry = h_parent,
19705 +                       .mnt    = au_br_mnt(br)
19706 +               };
19707 +               au_dtime_store(dt, au_pinned_parent(pin), &tmp);
19708 +       }
19709 +
19710 +       wh_dentry = NULL;
19711 +       if (bcpup != au_dbwh(dentry))
19712 +               goto out; /* success */
19713 +
19714 +       /*
19715 +        * ENAMETOOLONG here means that if we allowed create such name, then it
19716 +        * would not be able to removed in the future. So we don't allow such
19717 +        * name here and we don't handle ENAMETOOLONG differently here.
19718 +        */
19719 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
19720 +
19721 +out_unpin:
19722 +       if (IS_ERR(wh_dentry))
19723 +               au_unpin(pin);
19724 +out:
19725 +       return wh_dentry;
19726 +}
19727 +
19728 +/* ---------------------------------------------------------------------- */
19729 +
19730 +enum { Mknod, Symlink, Creat };
19731 +struct simple_arg {
19732 +       int type;
19733 +       union {
19734 +               struct {
19735 +                       umode_t                 mode;
19736 +                       bool                    want_excl;
19737 +                       bool                    try_aopen;
19738 +                       struct vfsub_aopen_args *aopen;
19739 +               } c;
19740 +               struct {
19741 +                       const char *symname;
19742 +               } s;
19743 +               struct {
19744 +                       umode_t mode;
19745 +                       dev_t dev;
19746 +               } m;
19747 +       } u;
19748 +};
19749 +
19750 +static int add_simple(struct inode *dir, struct dentry *dentry,
19751 +                     struct simple_arg *arg)
19752 +{
19753 +       int err, rerr;
19754 +       aufs_bindex_t btop;
19755 +       unsigned char created;
19756 +       const unsigned char try_aopen
19757 +               = (arg->type == Creat && arg->u.c.try_aopen);
19758 +       struct vfsub_aopen_args *aopen = arg->u.c.aopen;
19759 +       struct dentry *wh_dentry, *parent;
19760 +       struct inode *h_dir;
19761 +       struct super_block *sb;
19762 +       struct au_branch *br;
19763 +       /* to reduce stack size */
19764 +       struct {
19765 +               struct au_dtime dt;
19766 +               struct au_pin pin;
19767 +               struct path h_path;
19768 +               struct au_wr_dir_args wr_dir_args;
19769 +       } *a;
19770 +
19771 +       AuDbg("%pd\n", dentry);
19772 +       IMustLock(dir);
19773 +
19774 +       err = -ENOMEM;
19775 +       a = kmalloc(sizeof(*a), GFP_NOFS);
19776 +       if (unlikely(!a))
19777 +               goto out;
19778 +       a->wr_dir_args.force_btgt = -1;
19779 +       a->wr_dir_args.flags = AuWrDir_ADD_ENTRY;
19780 +
19781 +       parent = dentry->d_parent; /* dir inode is locked */
19782 +       if (!try_aopen) {
19783 +               err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
19784 +               if (unlikely(err))
19785 +                       goto out_free;
19786 +       }
19787 +       err = au_d_may_add(dentry);
19788 +       if (unlikely(err))
19789 +               goto out_unlock;
19790 +       if (!try_aopen)
19791 +               di_write_lock_parent(parent);
19792 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
19793 +                                     &a->pin, &a->wr_dir_args);
19794 +       err = PTR_ERR(wh_dentry);
19795 +       if (IS_ERR(wh_dentry))
19796 +               goto out_parent;
19797 +
19798 +       btop = au_dbtop(dentry);
19799 +       sb = dentry->d_sb;
19800 +       br = au_sbr(sb, btop);
19801 +       a->h_path.dentry = au_h_dptr(dentry, btop);
19802 +       a->h_path.mnt = au_br_mnt(br);
19803 +       h_dir = au_pinned_h_dir(&a->pin);
19804 +       switch (arg->type) {
19805 +       case Creat:
19806 +               if (!try_aopen || !h_dir->i_op->atomic_open) {
19807 +                       err = vfsub_create(h_dir, &a->h_path, arg->u.c.mode,
19808 +                                          arg->u.c.want_excl);
19809 +                       created = !err;
19810 +                       if (!err && try_aopen)
19811 +                               aopen->file->f_mode |= FMODE_CREATED;
19812 +               } else {
19813 +                       aopen->br = br;
19814 +                       err = vfsub_atomic_open(h_dir, a->h_path.dentry, aopen);
19815 +                       AuDbg("err %d\n", err);
19816 +                       AuDbgFile(aopen->file);
19817 +                       created = err >= 0
19818 +                               && !!(aopen->file->f_mode & FMODE_CREATED);
19819 +               }
19820 +               break;
19821 +       case Symlink:
19822 +               err = vfsub_symlink(h_dir, &a->h_path, arg->u.s.symname);
19823 +               created = !err;
19824 +               break;
19825 +       case Mknod:
19826 +               err = vfsub_mknod(h_dir, &a->h_path, arg->u.m.mode,
19827 +                                 arg->u.m.dev);
19828 +               created = !err;
19829 +               break;
19830 +       default:
19831 +               BUG();
19832 +       }
19833 +       if (unlikely(err < 0))
19834 +               goto out_unpin;
19835 +
19836 +       err = epilog(dir, btop, wh_dentry, dentry);
19837 +       if (!err)
19838 +               goto out_unpin; /* success */
19839 +
19840 +       /* revert */
19841 +       if (created /* && d_is_positive(a->h_path.dentry) */) {
19842 +               /* no delegation since it is just created */
19843 +               rerr = vfsub_unlink(h_dir, &a->h_path, /*delegated*/NULL,
19844 +                                   /*force*/0);
19845 +               if (rerr) {
19846 +                       AuIOErr("%pd revert failure(%d, %d)\n",
19847 +                               dentry, err, rerr);
19848 +                       err = -EIO;
19849 +               }
19850 +               au_dtime_revert(&a->dt);
19851 +       }
19852 +       if (try_aopen && h_dir->i_op->atomic_open
19853 +           && (aopen->file->f_mode & FMODE_OPENED))
19854 +               /* aopen->file is still opened */
19855 +               au_lcnt_dec(&aopen->br->br_nfiles);
19856 +
19857 +out_unpin:
19858 +       au_unpin(&a->pin);
19859 +       dput(wh_dentry);
19860 +out_parent:
19861 +       if (!try_aopen)
19862 +               di_write_unlock(parent);
19863 +out_unlock:
19864 +       if (unlikely(err)) {
19865 +               au_update_dbtop(dentry);
19866 +               d_drop(dentry);
19867 +       }
19868 +       if (!try_aopen)
19869 +               aufs_read_unlock(dentry, AuLock_DW);
19870 +out_free:
19871 +       au_kfree_rcu(a);
19872 +out:
19873 +       return err;
19874 +}
19875 +
19876 +int aufs_mknod(struct user_namespace *userns, struct inode *dir,
19877 +              struct dentry *dentry, umode_t mode, dev_t dev)
19878 +{
19879 +       struct simple_arg arg = {
19880 +               .type = Mknod,
19881 +               .u.m = {
19882 +                       .mode   = mode,
19883 +                       .dev    = dev
19884 +               }
19885 +       };
19886 +       return add_simple(dir, dentry, &arg);
19887 +}
19888 +
19889 +int aufs_symlink(struct user_namespace *userns, struct inode *dir,
19890 +                struct dentry *dentry, const char *symname)
19891 +{
19892 +       struct simple_arg arg = {
19893 +               .type = Symlink,
19894 +               .u.s.symname = symname
19895 +       };
19896 +       return add_simple(dir, dentry, &arg);
19897 +}
19898 +
19899 +int aufs_create(struct user_namespace *userns, struct inode *dir,
19900 +               struct dentry *dentry, umode_t mode, bool want_excl)
19901 +{
19902 +       struct simple_arg arg = {
19903 +               .type = Creat,
19904 +               .u.c = {
19905 +                       .mode           = mode,
19906 +                       .want_excl      = want_excl
19907 +               }
19908 +       };
19909 +       return add_simple(dir, dentry, &arg);
19910 +}
19911 +
19912 +int au_aopen_or_create(struct inode *dir, struct dentry *dentry,
19913 +                      struct vfsub_aopen_args *aopen_args)
19914 +{
19915 +       struct simple_arg arg = {
19916 +               .type = Creat,
19917 +               .u.c = {
19918 +                       .mode           = aopen_args->create_mode,
19919 +                       .want_excl      = aopen_args->open_flag & O_EXCL,
19920 +                       .try_aopen      = true,
19921 +                       .aopen          = aopen_args
19922 +               }
19923 +       };
19924 +       return add_simple(dir, dentry, &arg);
19925 +}
19926 +
19927 +int aufs_tmpfile(struct user_namespace *userns, struct inode *dir,
19928 +                struct dentry *dentry, umode_t mode)
19929 +{
19930 +       int err;
19931 +       aufs_bindex_t bindex;
19932 +       struct super_block *sb;
19933 +       struct dentry *parent, *h_parent, *h_dentry;
19934 +       struct inode *h_dir, *inode;
19935 +       struct vfsmount *h_mnt;
19936 +       struct user_namespace *h_userns;
19937 +       struct au_wr_dir_args wr_dir_args = {
19938 +               .force_btgt     = -1,
19939 +               .flags          = AuWrDir_TMPFILE
19940 +       };
19941 +
19942 +       /* copy-up may happen */
19943 +       inode_lock(dir);
19944 +
19945 +       sb = dir->i_sb;
19946 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19947 +       if (unlikely(err))
19948 +               goto out;
19949 +
19950 +       err = au_di_init(dentry);
19951 +       if (unlikely(err))
19952 +               goto out_si;
19953 +
19954 +       err = -EBUSY;
19955 +       parent = d_find_any_alias(dir);
19956 +       AuDebugOn(!parent);
19957 +       di_write_lock_parent(parent);
19958 +       if (unlikely(d_inode(parent) != dir))
19959 +               goto out_parent;
19960 +
19961 +       err = au_digen_test(parent, au_sigen(sb));
19962 +       if (unlikely(err))
19963 +               goto out_parent;
19964 +
19965 +       bindex = au_dbtop(parent);
19966 +       au_set_dbtop(dentry, bindex);
19967 +       au_set_dbbot(dentry, bindex);
19968 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
19969 +       bindex = err;
19970 +       if (unlikely(err < 0))
19971 +               goto out_parent;
19972 +
19973 +       err = -EOPNOTSUPP;
19974 +       h_dir = au_h_iptr(dir, bindex);
19975 +       if (unlikely(!h_dir->i_op->tmpfile))
19976 +               goto out_parent;
19977 +
19978 +       h_mnt = au_sbr_mnt(sb, bindex);
19979 +       err = vfsub_mnt_want_write(h_mnt);
19980 +       if (unlikely(err))
19981 +               goto out_parent;
19982 +
19983 +       h_userns = mnt_user_ns(h_mnt);
19984 +       h_parent = au_h_dptr(parent, bindex);
19985 +       h_dentry = vfs_tmpfile(h_userns, h_parent, mode, /*open_flag*/0);
19986 +       if (IS_ERR(h_dentry)) {
19987 +               err = PTR_ERR(h_dentry);
19988 +               goto out_mnt;
19989 +       }
19990 +
19991 +       au_set_dbtop(dentry, bindex);
19992 +       au_set_dbbot(dentry, bindex);
19993 +       au_set_h_dptr(dentry, bindex, dget(h_dentry));
19994 +       inode = au_new_inode(dentry, /*must_new*/1);
19995 +       if (IS_ERR(inode)) {
19996 +               err = PTR_ERR(inode);
19997 +               au_set_h_dptr(dentry, bindex, NULL);
19998 +               au_set_dbtop(dentry, -1);
19999 +               au_set_dbbot(dentry, -1);
20000 +       } else {
20001 +               if (!inode->i_nlink)
20002 +                       set_nlink(inode, 1);
20003 +               d_tmpfile(dentry, inode);
20004 +               au_di(dentry)->di_tmpfile = 1;
20005 +
20006 +               /* update without i_mutex */
20007 +               if (au_ibtop(dir) == au_dbtop(dentry))
20008 +                       au_cpup_attr_timesizes(dir);
20009 +       }
20010 +       dput(h_dentry);
20011 +
20012 +out_mnt:
20013 +       vfsub_mnt_drop_write(h_mnt);
20014 +out_parent:
20015 +       di_write_unlock(parent);
20016 +       dput(parent);
20017 +       di_write_unlock(dentry);
20018 +       if (unlikely(err)) {
20019 +               au_di_fin(dentry);
20020 +               dentry->d_fsdata = NULL;
20021 +       }
20022 +out_si:
20023 +       si_read_unlock(sb);
20024 +out:
20025 +       inode_unlock(dir);
20026 +       return err;
20027 +}
20028 +
20029 +/* ---------------------------------------------------------------------- */
20030 +
20031 +struct au_link_args {
20032 +       aufs_bindex_t bdst, bsrc;
20033 +       struct au_pin pin;
20034 +       struct path h_path;
20035 +       struct dentry *src_parent, *parent;
20036 +};
20037 +
20038 +static int au_cpup_before_link(struct dentry *src_dentry,
20039 +                              struct au_link_args *a)
20040 +{
20041 +       int err;
20042 +       struct dentry *h_src_dentry;
20043 +       struct au_cp_generic cpg = {
20044 +               .dentry = src_dentry,
20045 +               .bdst   = a->bdst,
20046 +               .bsrc   = a->bsrc,
20047 +               .len    = -1,
20048 +               .pin    = &a->pin,
20049 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN /* | AuCpup_KEEPLINO */
20050 +       };
20051 +
20052 +       di_read_lock_parent(a->src_parent, AuLock_IR);
20053 +       err = au_test_and_cpup_dirs(src_dentry, a->bdst);
20054 +       if (unlikely(err))
20055 +               goto out;
20056 +
20057 +       h_src_dentry = au_h_dptr(src_dentry, a->bsrc);
20058 +       err = au_pin(&a->pin, src_dentry, a->bdst,
20059 +                    au_opt_udba(src_dentry->d_sb),
20060 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
20061 +       if (unlikely(err))
20062 +               goto out;
20063 +
20064 +       err = au_sio_cpup_simple(&cpg);
20065 +       au_unpin(&a->pin);
20066 +
20067 +out:
20068 +       di_read_unlock(a->src_parent, AuLock_IR);
20069 +       return err;
20070 +}
20071 +
20072 +static int au_cpup_or_link(struct dentry *src_dentry, struct dentry *dentry,
20073 +                          struct au_link_args *a)
20074 +{
20075 +       int err;
20076 +       unsigned char plink;
20077 +       aufs_bindex_t bbot;
20078 +       struct dentry *h_src_dentry;
20079 +       struct inode *h_inode, *inode, *delegated;
20080 +       struct super_block *sb;
20081 +       struct file *h_file;
20082 +
20083 +       plink = 0;
20084 +       h_inode = NULL;
20085 +       sb = src_dentry->d_sb;
20086 +       inode = d_inode(src_dentry);
20087 +       if (au_ibtop(inode) <= a->bdst)
20088 +               h_inode = au_h_iptr(inode, a->bdst);
20089 +       if (!h_inode || !h_inode->i_nlink) {
20090 +               /* copyup src_dentry as the name of dentry. */
20091 +               bbot = au_dbbot(dentry);
20092 +               if (bbot < a->bsrc)
20093 +                       au_set_dbbot(dentry, a->bsrc);
20094 +               au_set_h_dptr(dentry, a->bsrc,
20095 +                             dget(au_h_dptr(src_dentry, a->bsrc)));
20096 +               dget(a->h_path.dentry);
20097 +               au_set_h_dptr(dentry, a->bdst, NULL);
20098 +               AuDbg("temporary d_inode...\n");
20099 +               spin_lock(&dentry->d_lock);
20100 +               dentry->d_inode = d_inode(src_dentry); /* tmp */
20101 +               spin_unlock(&dentry->d_lock);
20102 +               h_file = au_h_open_pre(dentry, a->bsrc, /*force_wr*/0);
20103 +               if (IS_ERR(h_file))
20104 +                       err = PTR_ERR(h_file);
20105 +               else {
20106 +                       struct au_cp_generic cpg = {
20107 +                               .dentry = dentry,
20108 +                               .bdst   = a->bdst,
20109 +                               .bsrc   = -1,
20110 +                               .len    = -1,
20111 +                               .pin    = &a->pin,
20112 +                               .flags  = AuCpup_KEEPLINO
20113 +                       };
20114 +                       err = au_sio_cpup_simple(&cpg);
20115 +                       au_h_open_post(dentry, a->bsrc, h_file);
20116 +                       if (!err) {
20117 +                               dput(a->h_path.dentry);
20118 +                               a->h_path.dentry = au_h_dptr(dentry, a->bdst);
20119 +                       } else
20120 +                               au_set_h_dptr(dentry, a->bdst,
20121 +                                             a->h_path.dentry);
20122 +               }
20123 +               spin_lock(&dentry->d_lock);
20124 +               dentry->d_inode = NULL; /* restore */
20125 +               spin_unlock(&dentry->d_lock);
20126 +               AuDbg("temporary d_inode...done\n");
20127 +               au_set_h_dptr(dentry, a->bsrc, NULL);
20128 +               au_set_dbbot(dentry, bbot);
20129 +       } else {
20130 +               /* the inode of src_dentry already exists on a.bdst branch */
20131 +               h_src_dentry = d_find_alias(h_inode);
20132 +               if (!h_src_dentry && au_plink_test(inode)) {
20133 +                       plink = 1;
20134 +                       h_src_dentry = au_plink_lkup(inode, a->bdst);
20135 +                       err = PTR_ERR(h_src_dentry);
20136 +                       if (IS_ERR(h_src_dentry))
20137 +                               goto out;
20138 +
20139 +                       if (unlikely(d_is_negative(h_src_dentry))) {
20140 +                               dput(h_src_dentry);
20141 +                               h_src_dentry = NULL;
20142 +                       }
20143 +
20144 +               }
20145 +               if (h_src_dentry) {
20146 +                       delegated = NULL;
20147 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
20148 +                                        &a->h_path, &delegated);
20149 +                       if (unlikely(err == -EWOULDBLOCK)) {
20150 +                               pr_warn("cannot retry for NFSv4 delegation"
20151 +                                       " for an internal link\n");
20152 +                               iput(delegated);
20153 +                       }
20154 +                       dput(h_src_dentry);
20155 +               } else {
20156 +                       AuIOErr("no dentry found for hi%lu on b%d\n",
20157 +                               h_inode->i_ino, a->bdst);
20158 +                       err = -EIO;
20159 +               }
20160 +       }
20161 +
20162 +       if (!err && !plink)
20163 +               au_plink_append(inode, a->bdst, a->h_path.dentry);
20164 +
20165 +out:
20166 +       AuTraceErr(err);
20167 +       return err;
20168 +}
20169 +
20170 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
20171 +             struct dentry *dentry)
20172 +{
20173 +       int err, rerr;
20174 +       struct au_dtime dt;
20175 +       struct au_link_args *a;
20176 +       struct dentry *wh_dentry, *h_src_dentry;
20177 +       struct inode *inode, *delegated;
20178 +       struct super_block *sb;
20179 +       struct au_wr_dir_args wr_dir_args = {
20180 +               /* .force_btgt  = -1, */
20181 +               .flags          = AuWrDir_ADD_ENTRY
20182 +       };
20183 +
20184 +       IMustLock(dir);
20185 +       inode = d_inode(src_dentry);
20186 +       IMustLock(inode);
20187 +
20188 +       err = -ENOMEM;
20189 +       a = kzalloc(sizeof(*a), GFP_NOFS);
20190 +       if (unlikely(!a))
20191 +               goto out;
20192 +
20193 +       a->parent = dentry->d_parent; /* dir inode is locked */
20194 +       err = aufs_read_and_write_lock2(dentry, src_dentry,
20195 +                                       AuLock_NOPLM | AuLock_GEN);
20196 +       if (unlikely(err))
20197 +               goto out_kfree;
20198 +       err = au_d_linkable(src_dentry);
20199 +       if (unlikely(err))
20200 +               goto out_unlock;
20201 +       err = au_d_may_add(dentry);
20202 +       if (unlikely(err))
20203 +               goto out_unlock;
20204 +
20205 +       a->src_parent = dget_parent(src_dentry);
20206 +       wr_dir_args.force_btgt = au_ibtop(inode);
20207 +
20208 +       di_write_lock_parent(a->parent);
20209 +       wr_dir_args.force_btgt = au_wbr(dentry, wr_dir_args.force_btgt);
20210 +       wh_dentry = lock_hdir_lkup_wh(dentry, &dt, src_dentry, &a->pin,
20211 +                                     &wr_dir_args);
20212 +       err = PTR_ERR(wh_dentry);
20213 +       if (IS_ERR(wh_dentry))
20214 +               goto out_parent;
20215 +
20216 +       err = 0;
20217 +       sb = dentry->d_sb;
20218 +       a->bdst = au_dbtop(dentry);
20219 +       a->h_path.dentry = au_h_dptr(dentry, a->bdst);
20220 +       a->h_path.mnt = au_sbr_mnt(sb, a->bdst);
20221 +       a->bsrc = au_ibtop(inode);
20222 +       h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
20223 +       if (!h_src_dentry && au_di(src_dentry)->di_tmpfile)
20224 +               h_src_dentry = dget(au_hi_wh(inode, a->bsrc));
20225 +       if (!h_src_dentry) {
20226 +               a->bsrc = au_dbtop(src_dentry);
20227 +               h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
20228 +               AuDebugOn(!h_src_dentry);
20229 +       } else if (IS_ERR(h_src_dentry)) {
20230 +               err = PTR_ERR(h_src_dentry);
20231 +               goto out_parent;
20232 +       }
20233 +
20234 +       /*
20235 +        * aufs doesn't touch the credential so
20236 +        * security_dentry_create_files_as() is unnecessary.
20237 +        */
20238 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
20239 +               if (a->bdst < a->bsrc
20240 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */)
20241 +                       err = au_cpup_or_link(src_dentry, dentry, a);
20242 +               else {
20243 +                       delegated = NULL;
20244 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
20245 +                                        &a->h_path, &delegated);
20246 +                       if (unlikely(err == -EWOULDBLOCK)) {
20247 +                               pr_warn("cannot retry for NFSv4 delegation"
20248 +                                       " for an internal link\n");
20249 +                               iput(delegated);
20250 +                       }
20251 +               }
20252 +               dput(h_src_dentry);
20253 +       } else {
20254 +               /*
20255 +                * copyup src_dentry to the branch we process,
20256 +                * and then link(2) to it.
20257 +                */
20258 +               dput(h_src_dentry);
20259 +               if (a->bdst < a->bsrc
20260 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */) {
20261 +                       au_unpin(&a->pin);
20262 +                       di_write_unlock(a->parent);
20263 +                       err = au_cpup_before_link(src_dentry, a);
20264 +                       di_write_lock_parent(a->parent);
20265 +                       if (!err)
20266 +                               err = au_pin(&a->pin, dentry, a->bdst,
20267 +                                            au_opt_udba(sb),
20268 +                                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
20269 +                       if (unlikely(err))
20270 +                               goto out_wh;
20271 +               }
20272 +               if (!err) {
20273 +                       h_src_dentry = au_h_dptr(src_dentry, a->bdst);
20274 +                       err = -ENOENT;
20275 +                       if (h_src_dentry && d_is_positive(h_src_dentry)) {
20276 +                               delegated = NULL;
20277 +                               err = vfsub_link(h_src_dentry,
20278 +                                                au_pinned_h_dir(&a->pin),
20279 +                                                &a->h_path, &delegated);
20280 +                               if (unlikely(err == -EWOULDBLOCK)) {
20281 +                                       pr_warn("cannot retry"
20282 +                                               " for NFSv4 delegation"
20283 +                                               " for an internal link\n");
20284 +                                       iput(delegated);
20285 +                               }
20286 +                       }
20287 +               }
20288 +       }
20289 +       if (unlikely(err))
20290 +               goto out_unpin;
20291 +
20292 +       if (wh_dentry) {
20293 +               a->h_path.dentry = wh_dentry;
20294 +               err = au_wh_unlink_dentry(au_pinned_h_dir(&a->pin), &a->h_path,
20295 +                                         dentry);
20296 +               if (unlikely(err))
20297 +                       goto out_revert;
20298 +       }
20299 +
20300 +       au_dir_ts(dir, a->bdst);
20301 +       inode_inc_iversion(dir);
20302 +       inc_nlink(inode);
20303 +       inode->i_ctime = dir->i_ctime;
20304 +       d_instantiate(dentry, au_igrab(inode));
20305 +       if (d_unhashed(a->h_path.dentry))
20306 +               /* some filesystem calls d_drop() */
20307 +               d_drop(dentry);
20308 +       /* some filesystems consume an inode even hardlink */
20309 +       au_fhsm_wrote(sb, a->bdst, /*force*/0);
20310 +       goto out_unpin; /* success */
20311 +
20312 +out_revert:
20313 +       /* no delegation since it is just created */
20314 +       rerr = vfsub_unlink(au_pinned_h_dir(&a->pin), &a->h_path,
20315 +                           /*delegated*/NULL, /*force*/0);
20316 +       if (unlikely(rerr)) {
20317 +               AuIOErr("%pd reverting failed(%d, %d)\n", dentry, err, rerr);
20318 +               err = -EIO;
20319 +       }
20320 +       au_dtime_revert(&dt);
20321 +out_unpin:
20322 +       au_unpin(&a->pin);
20323 +out_wh:
20324 +       dput(wh_dentry);
20325 +out_parent:
20326 +       di_write_unlock(a->parent);
20327 +       dput(a->src_parent);
20328 +out_unlock:
20329 +       if (unlikely(err)) {
20330 +               au_update_dbtop(dentry);
20331 +               d_drop(dentry);
20332 +       }
20333 +       aufs_read_and_write_unlock2(dentry, src_dentry);
20334 +out_kfree:
20335 +       au_kfree_rcu(a);
20336 +out:
20337 +       AuTraceErr(err);
20338 +       return err;
20339 +}
20340 +
20341 +int aufs_mkdir(struct user_namespace *userns, struct inode *dir,
20342 +              struct dentry *dentry, umode_t mode)
20343 +{
20344 +       int err, rerr;
20345 +       aufs_bindex_t bindex;
20346 +       unsigned char diropq;
20347 +       struct path h_path;
20348 +       struct dentry *wh_dentry, *parent, *opq_dentry;
20349 +       struct inode *h_inode;
20350 +       struct super_block *sb;
20351 +       struct {
20352 +               struct au_pin pin;
20353 +               struct au_dtime dt;
20354 +       } *a; /* reduce the stack usage */
20355 +       struct au_wr_dir_args wr_dir_args = {
20356 +               .force_btgt     = -1,
20357 +               .flags          = AuWrDir_ADD_ENTRY | AuWrDir_ISDIR
20358 +       };
20359 +
20360 +       IMustLock(dir);
20361 +
20362 +       err = -ENOMEM;
20363 +       a = kmalloc(sizeof(*a), GFP_NOFS);
20364 +       if (unlikely(!a))
20365 +               goto out;
20366 +
20367 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
20368 +       if (unlikely(err))
20369 +               goto out_free;
20370 +       err = au_d_may_add(dentry);
20371 +       if (unlikely(err))
20372 +               goto out_unlock;
20373 +
20374 +       parent = dentry->d_parent; /* dir inode is locked */
20375 +       di_write_lock_parent(parent);
20376 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
20377 +                                     &a->pin, &wr_dir_args);
20378 +       err = PTR_ERR(wh_dentry);
20379 +       if (IS_ERR(wh_dentry))
20380 +               goto out_parent;
20381 +
20382 +       sb = dentry->d_sb;
20383 +       bindex = au_dbtop(dentry);
20384 +       h_path.dentry = au_h_dptr(dentry, bindex);
20385 +       h_path.mnt = au_sbr_mnt(sb, bindex);
20386 +       err = vfsub_mkdir(au_pinned_h_dir(&a->pin), &h_path, mode);
20387 +       if (unlikely(err))
20388 +               goto out_unpin;
20389 +
20390 +       /* make the dir opaque */
20391 +       diropq = 0;
20392 +       h_inode = d_inode(h_path.dentry);
20393 +       if (wh_dentry
20394 +           || au_opt_test(au_mntflags(sb), ALWAYS_DIROPQ)) {
20395 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
20396 +               opq_dentry = au_diropq_create(dentry, bindex);
20397 +               inode_unlock(h_inode);
20398 +               err = PTR_ERR(opq_dentry);
20399 +               if (IS_ERR(opq_dentry))
20400 +                       goto out_dir;
20401 +               dput(opq_dentry);
20402 +               diropq = 1;
20403 +       }
20404 +
20405 +       err = epilog(dir, bindex, wh_dentry, dentry);
20406 +       if (!err) {
20407 +               inc_nlink(dir);
20408 +               goto out_unpin; /* success */
20409 +       }
20410 +
20411 +       /* revert */
20412 +       if (diropq) {
20413 +               AuLabel(revert opq);
20414 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
20415 +               rerr = au_diropq_remove(dentry, bindex);
20416 +               inode_unlock(h_inode);
20417 +               if (rerr) {
20418 +                       AuIOErr("%pd reverting diropq failed(%d, %d)\n",
20419 +                               dentry, err, rerr);
20420 +                       err = -EIO;
20421 +               }
20422 +       }
20423 +
20424 +out_dir:
20425 +       AuLabel(revert dir);
20426 +       rerr = vfsub_rmdir(au_pinned_h_dir(&a->pin), &h_path);
20427 +       if (rerr) {
20428 +               AuIOErr("%pd reverting dir failed(%d, %d)\n",
20429 +                       dentry, err, rerr);
20430 +               err = -EIO;
20431 +       }
20432 +       au_dtime_revert(&a->dt);
20433 +out_unpin:
20434 +       au_unpin(&a->pin);
20435 +       dput(wh_dentry);
20436 +out_parent:
20437 +       di_write_unlock(parent);
20438 +out_unlock:
20439 +       if (unlikely(err)) {
20440 +               au_update_dbtop(dentry);
20441 +               d_drop(dentry);
20442 +       }
20443 +       aufs_read_unlock(dentry, AuLock_DW);
20444 +out_free:
20445 +       au_kfree_rcu(a);
20446 +out:
20447 +       return err;
20448 +}
20449 diff -urN /usr/share/empty/fs/aufs/i_op.c linux/fs/aufs/i_op.c
20450 --- /usr/share/empty/fs/aufs/i_op.c     1970-01-01 01:00:00.000000000 +0100
20451 +++ linux/fs/aufs/i_op.c        2021-12-03 15:40:58.236647297 +0100
20452 @@ -0,0 +1,1513 @@
20453 +// SPDX-License-Identifier: GPL-2.0
20454 +/*
20455 + * Copyright (C) 2005-2021 Junjiro R. Okajima
20456 + *
20457 + * This program, aufs is free software; you can redistribute it and/or modify
20458 + * it under the terms of the GNU General Public License as published by
20459 + * the Free Software Foundation; either version 2 of the License, or
20460 + * (at your option) any later version.
20461 + *
20462 + * This program is distributed in the hope that it will be useful,
20463 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
20464 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20465 + * GNU General Public License for more details.
20466 + *
20467 + * You should have received a copy of the GNU General Public License
20468 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20469 + */
20470 +
20471 +/*
20472 + * inode operations (except add/del/rename)
20473 + */
20474 +
20475 +#include <linux/device_cgroup.h>
20476 +#include <linux/fs_stack.h>
20477 +#include <linux/iversion.h>
20478 +#include <linux/namei.h>
20479 +#include <linux/security.h>
20480 +#include "aufs.h"
20481 +
20482 +static int h_permission(struct inode *h_inode, int mask,
20483 +                       struct path *h_path, int brperm)
20484 +{
20485 +       int err;
20486 +       const unsigned char write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
20487 +       struct user_namespace *h_userns;
20488 +
20489 +       err = -EPERM;
20490 +       if (write_mask && IS_IMMUTABLE(h_inode))
20491 +               goto out;
20492 +
20493 +       err = -EACCES;
20494 +       if (((mask & MAY_EXEC)
20495 +            && S_ISREG(h_inode->i_mode)
20496 +            && (path_noexec(h_path)
20497 +                || !(h_inode->i_mode & 0111))))
20498 +               goto out;
20499 +
20500 +       /*
20501 +        * - skip the lower fs test in the case of write to ro branch.
20502 +        * - nfs dir permission write check is optimized, but a policy for
20503 +        *   link/rename requires a real check.
20504 +        * - nfs always sets SB_POSIXACL regardless its mount option 'noacl.'
20505 +        *   in this case, generic_permission() returns -EOPNOTSUPP.
20506 +        */
20507 +       h_userns = mnt_user_ns(h_path->mnt);
20508 +       if ((write_mask && !au_br_writable(brperm))
20509 +           || (au_test_nfs(h_inode->i_sb) && S_ISDIR(h_inode->i_mode)
20510 +               && write_mask && !(mask & MAY_READ))
20511 +           || !h_inode->i_op->permission) {
20512 +               /* AuLabel(generic_permission); */
20513 +               /* AuDbg("get_acl %ps\n", h_inode->i_op->get_acl); */
20514 +               err = generic_permission(h_userns, h_inode, mask);
20515 +               if (err == -EOPNOTSUPP && au_test_nfs_noacl(h_inode))
20516 +                       err = h_inode->i_op->permission(h_userns, h_inode,
20517 +                                                       mask);
20518 +               AuTraceErr(err);
20519 +       } else {
20520 +               /* AuLabel(h_inode->permission); */
20521 +               err = h_inode->i_op->permission(h_userns, h_inode, mask);
20522 +               AuTraceErr(err);
20523 +       }
20524 +
20525 +       if (!err)
20526 +               err = devcgroup_inode_permission(h_inode, mask);
20527 +       if (!err)
20528 +               err = security_inode_permission(h_inode, mask);
20529 +
20530 +out:
20531 +       return err;
20532 +}
20533 +
20534 +static int aufs_permission(struct user_namespace *userns, struct inode *inode,
20535 +                          int mask)
20536 +{
20537 +       int err;
20538 +       aufs_bindex_t bindex, bbot;
20539 +       const unsigned char isdir = !!S_ISDIR(inode->i_mode),
20540 +               write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
20541 +       struct inode *h_inode;
20542 +       struct super_block *sb;
20543 +       struct au_branch *br;
20544 +
20545 +       /* todo: support rcu-walk? */
20546 +       if (mask & MAY_NOT_BLOCK)
20547 +               return -ECHILD;
20548 +
20549 +       sb = inode->i_sb;
20550 +       si_read_lock(sb, AuLock_FLUSH);
20551 +       ii_read_lock_child(inode);
20552 +#if 0 /* reserved for future use */
20553 +       /*
20554 +        * This test may be rather 'too much' since the test is essentially done
20555 +        * in the aufs_lookup().  Theoretically it is possible that the inode
20556 +        * generation doesn't match to the superblock's here.  But it isn't a
20557 +        * big deal I suppose.
20558 +        */
20559 +       err = au_iigen_test(inode, au_sigen(sb));
20560 +       if (unlikely(err))
20561 +               goto out;
20562 +#endif
20563 +
20564 +       if (!isdir
20565 +           || write_mask
20566 +           || au_opt_test(au_mntflags(sb), DIRPERM1)) {
20567 +               err = au_busy_or_stale();
20568 +               h_inode = au_h_iptr(inode, au_ibtop(inode));
20569 +               if (unlikely(!h_inode
20570 +                            || (h_inode->i_mode & S_IFMT)
20571 +                            != (inode->i_mode & S_IFMT)))
20572 +                       goto out;
20573 +
20574 +               err = 0;
20575 +               bindex = au_ibtop(inode);
20576 +               br = au_sbr(sb, bindex);
20577 +               err = h_permission(h_inode, mask, &br->br_path, br->br_perm);
20578 +               if (write_mask
20579 +                   && !err
20580 +                   && !special_file(h_inode->i_mode)) {
20581 +                       /* test whether the upper writable branch exists */
20582 +                       err = -EROFS;
20583 +                       for (; bindex >= 0; bindex--)
20584 +                               if (!au_br_rdonly(au_sbr(sb, bindex))) {
20585 +                                       err = 0;
20586 +                                       break;
20587 +                               }
20588 +               }
20589 +               goto out;
20590 +       }
20591 +
20592 +       /* non-write to dir */
20593 +       err = 0;
20594 +       bbot = au_ibbot(inode);
20595 +       for (bindex = au_ibtop(inode); !err && bindex <= bbot; bindex++) {
20596 +               h_inode = au_h_iptr(inode, bindex);
20597 +               if (h_inode) {
20598 +                       err = au_busy_or_stale();
20599 +                       if (unlikely(!S_ISDIR(h_inode->i_mode)))
20600 +                               break;
20601 +
20602 +                       br = au_sbr(sb, bindex);
20603 +                       err = h_permission(h_inode, mask, &br->br_path,
20604 +                                          br->br_perm);
20605 +               }
20606 +       }
20607 +
20608 +out:
20609 +       ii_read_unlock(inode);
20610 +       si_read_unlock(sb);
20611 +       return err;
20612 +}
20613 +
20614 +/* ---------------------------------------------------------------------- */
20615 +
20616 +static struct dentry *aufs_lookup(struct inode *dir, struct dentry *dentry,
20617 +                                 unsigned int flags)
20618 +{
20619 +       struct dentry *ret, *parent;
20620 +       struct inode *inode;
20621 +       struct super_block *sb;
20622 +       int err, npositive;
20623 +
20624 +       IMustLock(dir);
20625 +
20626 +       /* todo: support rcu-walk? */
20627 +       ret = ERR_PTR(-ECHILD);
20628 +       if (flags & LOOKUP_RCU)
20629 +               goto out;
20630 +
20631 +       ret = ERR_PTR(-ENAMETOOLONG);
20632 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
20633 +               goto out;
20634 +
20635 +       sb = dir->i_sb;
20636 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
20637 +       ret = ERR_PTR(err);
20638 +       if (unlikely(err))
20639 +               goto out;
20640 +
20641 +       err = au_di_init(dentry);
20642 +       ret = ERR_PTR(err);
20643 +       if (unlikely(err))
20644 +               goto out_si;
20645 +
20646 +       inode = NULL;
20647 +       npositive = 0; /* suppress a warning */
20648 +       parent = dentry->d_parent; /* dir inode is locked */
20649 +       di_read_lock_parent(parent, AuLock_IR);
20650 +       err = au_alive_dir(parent);
20651 +       if (!err)
20652 +               err = au_digen_test(parent, au_sigen(sb));
20653 +       if (!err) {
20654 +               /* regardless LOOKUP_CREATE, always ALLOW_NEG */
20655 +               npositive = au_lkup_dentry(dentry, au_dbtop(parent),
20656 +                                          AuLkup_ALLOW_NEG);
20657 +               err = npositive;
20658 +       }
20659 +       di_read_unlock(parent, AuLock_IR);
20660 +       ret = ERR_PTR(err);
20661 +       if (unlikely(err < 0))
20662 +               goto out_unlock;
20663 +
20664 +       if (npositive) {
20665 +               inode = au_new_inode(dentry, /*must_new*/0);
20666 +               if (IS_ERR(inode)) {
20667 +                       ret = (void *)inode;
20668 +                       inode = NULL;
20669 +                       goto out_unlock;
20670 +               }
20671 +       }
20672 +
20673 +       if (inode)
20674 +               atomic_inc(&inode->i_count);
20675 +       ret = d_splice_alias(inode, dentry);
20676 +#if 0 /* reserved for future use */
20677 +       if (unlikely(d_need_lookup(dentry))) {
20678 +               spin_lock(&dentry->d_lock);
20679 +               dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
20680 +               spin_unlock(&dentry->d_lock);
20681 +       } else
20682 +#endif
20683 +       if (inode) {
20684 +               if (!IS_ERR(ret)) {
20685 +                       iput(inode);
20686 +                       if (ret && ret != dentry)
20687 +                               ii_write_unlock(inode);
20688 +               } else {
20689 +                       ii_write_unlock(inode);
20690 +                       iput(inode);
20691 +                       inode = NULL;
20692 +               }
20693 +       }
20694 +
20695 +out_unlock:
20696 +       di_write_unlock(dentry);
20697 +out_si:
20698 +       si_read_unlock(sb);
20699 +out:
20700 +       return ret;
20701 +}
20702 +
20703 +/* ---------------------------------------------------------------------- */
20704 +
20705 +/*
20706 + * very dirty and complicated aufs ->atomic_open().
20707 + * aufs_atomic_open()
20708 + * + au_aopen_or_create()
20709 + *   + add_simple()
20710 + *     + vfsub_atomic_open()
20711 + *       + branch fs ->atomic_open()
20712 + *        may call the actual 'open' for h_file
20713 + *       + inc br_nfiles only if opened
20714 + * + au_aopen_no_open() or au_aopen_do_open()
20715 + *
20716 + * au_aopen_do_open()
20717 + * + finish_open()
20718 + *   + au_do_aopen()
20719 + *     + au_do_open() the body of all 'open'
20720 + *       + au_do_open_nondir()
20721 + *        set the passed h_file
20722 + *
20723 + * au_aopen_no_open()
20724 + * + finish_no_open()
20725 + */
20726 +
20727 +struct aopen_node {
20728 +       struct hlist_bl_node hblist;
20729 +       struct file *file, *h_file;
20730 +};
20731 +
20732 +static int au_do_aopen(struct inode *inode, struct file *file)
20733 +{
20734 +       struct hlist_bl_head *aopen;
20735 +       struct hlist_bl_node *pos;
20736 +       struct aopen_node *node;
20737 +       struct au_do_open_args args = {
20738 +               .aopen  = 1,
20739 +               .open   = au_do_open_nondir
20740 +       };
20741 +
20742 +       aopen = &au_sbi(inode->i_sb)->si_aopen;
20743 +       hlist_bl_lock(aopen);
20744 +       hlist_bl_for_each_entry(node, pos, aopen, hblist)
20745 +               if (node->file == file) {
20746 +                       args.h_file = node->h_file;
20747 +                       break;
20748 +               }
20749 +       hlist_bl_unlock(aopen);
20750 +       /* AuDebugOn(!args.h_file); */
20751 +
20752 +       return au_do_open(file, &args);
20753 +}
20754 +
20755 +static int au_aopen_do_open(struct file *file, struct dentry *dentry,
20756 +                           struct aopen_node *aopen_node)
20757 +{
20758 +       int err;
20759 +       struct hlist_bl_head *aopen;
20760 +
20761 +       AuLabel(here);
20762 +       aopen = &au_sbi(dentry->d_sb)->si_aopen;
20763 +       au_hbl_add(&aopen_node->hblist, aopen);
20764 +       err = finish_open(file, dentry, au_do_aopen);
20765 +       au_hbl_del(&aopen_node->hblist, aopen);
20766 +       /* AuDbgFile(file); */
20767 +       AuDbg("%pd%s%s\n", dentry,
20768 +             (file->f_mode & FMODE_CREATED) ? " created" : "",
20769 +             (file->f_mode & FMODE_OPENED) ? " opened" : "");
20770 +
20771 +       AuTraceErr(err);
20772 +       return err;
20773 +}
20774 +
20775 +static int au_aopen_no_open(struct file *file, struct dentry *dentry)
20776 +{
20777 +       int err;
20778 +
20779 +       AuLabel(here);
20780 +       dget(dentry);
20781 +       err = finish_no_open(file, dentry);
20782 +
20783 +       AuTraceErr(err);
20784 +       return err;
20785 +}
20786 +
20787 +static int aufs_atomic_open(struct inode *dir, struct dentry *dentry,
20788 +                           struct file *file, unsigned int open_flag,
20789 +                           umode_t create_mode)
20790 +{
20791 +       int err, did_open;
20792 +       unsigned int lkup_flags;
20793 +       aufs_bindex_t bindex;
20794 +       struct super_block *sb;
20795 +       struct dentry *parent, *d;
20796 +       struct vfsub_aopen_args args = {
20797 +               .open_flag      = open_flag,
20798 +               .create_mode    = create_mode
20799 +       };
20800 +       struct aopen_node aopen_node = {
20801 +               .file   = file
20802 +       };
20803 +
20804 +       IMustLock(dir);
20805 +       AuDbg("open_flag 0%o\n", open_flag);
20806 +       AuDbgDentry(dentry);
20807 +
20808 +       err = 0;
20809 +       if (!au_di(dentry)) {
20810 +               lkup_flags = LOOKUP_OPEN;
20811 +               if (open_flag & O_CREAT)
20812 +                       lkup_flags |= LOOKUP_CREATE;
20813 +               d = aufs_lookup(dir, dentry, lkup_flags);
20814 +               if (IS_ERR(d)) {
20815 +                       err = PTR_ERR(d);
20816 +                       AuTraceErr(err);
20817 +                       goto out;
20818 +               } else if (d) {
20819 +                       /*
20820 +                        * obsoleted dentry found.
20821 +                        * another error will be returned later.
20822 +                        */
20823 +                       d_drop(d);
20824 +                       AuDbgDentry(d);
20825 +                       dput(d);
20826 +               }
20827 +               AuDbgDentry(dentry);
20828 +       }
20829 +
20830 +       if (d_is_positive(dentry)
20831 +           || d_unhashed(dentry)
20832 +           || d_unlinked(dentry)
20833 +           || !(open_flag & O_CREAT)) {
20834 +               err = au_aopen_no_open(file, dentry);
20835 +               goto out; /* success */
20836 +       }
20837 +
20838 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
20839 +       if (unlikely(err))
20840 +               goto out;
20841 +
20842 +       sb = dentry->d_sb;
20843 +       parent = dentry->d_parent;      /* dir is locked */
20844 +       di_write_lock_parent(parent);
20845 +       err = au_lkup_dentry(dentry, /*btop*/0, AuLkup_ALLOW_NEG);
20846 +       if (unlikely(err < 0))
20847 +               goto out_parent;
20848 +
20849 +       AuDbgDentry(dentry);
20850 +       if (d_is_positive(dentry)) {
20851 +               err = au_aopen_no_open(file, dentry);
20852 +               goto out_parent; /* success */
20853 +       }
20854 +
20855 +       args.file = alloc_empty_file(file->f_flags, current_cred());
20856 +       err = PTR_ERR(args.file);
20857 +       if (IS_ERR(args.file))
20858 +               goto out_parent;
20859 +
20860 +       bindex = au_dbtop(dentry);
20861 +       err = au_aopen_or_create(dir, dentry, &args);
20862 +       AuTraceErr(err);
20863 +       AuDbgFile(args.file);
20864 +       file->f_mode = args.file->f_mode & ~FMODE_OPENED;
20865 +       did_open = !!(args.file->f_mode & FMODE_OPENED);
20866 +       if (!did_open) {
20867 +               fput(args.file);
20868 +               args.file = NULL;
20869 +       }
20870 +       di_write_unlock(parent);
20871 +       di_write_unlock(dentry);
20872 +       if (unlikely(err < 0)) {
20873 +               if (args.file)
20874 +                       fput(args.file);
20875 +               goto out_sb;
20876 +       }
20877 +
20878 +       if (!did_open)
20879 +               err = au_aopen_no_open(file, dentry);
20880 +       else {
20881 +               aopen_node.h_file = args.file;
20882 +               err = au_aopen_do_open(file, dentry, &aopen_node);
20883 +       }
20884 +       if (unlikely(err < 0)) {
20885 +               if (args.file)
20886 +                       fput(args.file);
20887 +               if (did_open)
20888 +                       au_lcnt_dec(&args.br->br_nfiles);
20889 +       }
20890 +       goto out_sb; /* success */
20891 +
20892 +out_parent:
20893 +       di_write_unlock(parent);
20894 +       di_write_unlock(dentry);
20895 +out_sb:
20896 +       si_read_unlock(sb);
20897 +out:
20898 +       AuTraceErr(err);
20899 +       AuDbgFile(file);
20900 +       return err;
20901 +}
20902 +
20903 +
20904 +/* ---------------------------------------------------------------------- */
20905 +
20906 +static int au_wr_dir_cpup(struct dentry *dentry, struct dentry *parent,
20907 +                         const unsigned char add_entry, aufs_bindex_t bcpup,
20908 +                         aufs_bindex_t btop)
20909 +{
20910 +       int err;
20911 +       struct dentry *h_parent;
20912 +       struct inode *h_dir;
20913 +
20914 +       if (add_entry)
20915 +               IMustLock(d_inode(parent));
20916 +       else
20917 +               di_write_lock_parent(parent);
20918 +
20919 +       err = 0;
20920 +       if (!au_h_dptr(parent, bcpup)) {
20921 +               if (btop > bcpup)
20922 +                       err = au_cpup_dirs(dentry, bcpup);
20923 +               else if (btop < bcpup)
20924 +                       err = au_cpdown_dirs(dentry, bcpup);
20925 +               else
20926 +                       BUG();
20927 +       }
20928 +       if (!err && add_entry && !au_ftest_wrdir(add_entry, TMPFILE)) {
20929 +               h_parent = au_h_dptr(parent, bcpup);
20930 +               h_dir = d_inode(h_parent);
20931 +               inode_lock_shared_nested(h_dir, AuLsc_I_PARENT);
20932 +               err = au_lkup_neg(dentry, bcpup, /*wh*/0);
20933 +               /* todo: no unlock here */
20934 +               inode_unlock_shared(h_dir);
20935 +
20936 +               AuDbg("bcpup %d\n", bcpup);
20937 +               if (!err) {
20938 +                       if (d_really_is_negative(dentry))
20939 +                               au_set_h_dptr(dentry, btop, NULL);
20940 +                       au_update_dbrange(dentry, /*do_put_zero*/0);
20941 +               }
20942 +       }
20943 +
20944 +       if (!add_entry)
20945 +               di_write_unlock(parent);
20946 +       if (!err)
20947 +               err = bcpup; /* success */
20948 +
20949 +       AuTraceErr(err);
20950 +       return err;
20951 +}
20952 +
20953 +/*
20954 + * decide the branch and the parent dir where we will create a new entry.
20955 + * returns new bindex or an error.
20956 + * copyup the parent dir if needed.
20957 + */
20958 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
20959 +             struct au_wr_dir_args *args)
20960 +{
20961 +       int err;
20962 +       unsigned int flags;
20963 +       aufs_bindex_t bcpup, btop, src_btop;
20964 +       const unsigned char add_entry
20965 +               = au_ftest_wrdir(args->flags, ADD_ENTRY)
20966 +               | au_ftest_wrdir(args->flags, TMPFILE);
20967 +       struct super_block *sb;
20968 +       struct dentry *parent;
20969 +       struct au_sbinfo *sbinfo;
20970 +
20971 +       sb = dentry->d_sb;
20972 +       sbinfo = au_sbi(sb);
20973 +       parent = dget_parent(dentry);
20974 +       btop = au_dbtop(dentry);
20975 +       bcpup = btop;
20976 +       if (args->force_btgt < 0) {
20977 +               if (src_dentry) {
20978 +                       src_btop = au_dbtop(src_dentry);
20979 +                       if (src_btop < btop)
20980 +                               bcpup = src_btop;
20981 +               } else if (add_entry) {
20982 +                       flags = 0;
20983 +                       if (au_ftest_wrdir(args->flags, ISDIR))
20984 +                               au_fset_wbr(flags, DIR);
20985 +                       err = AuWbrCreate(sbinfo, dentry, flags);
20986 +                       bcpup = err;
20987 +               }
20988 +
20989 +               if (bcpup < 0 || au_test_ro(sb, bcpup, d_inode(dentry))) {
20990 +                       if (add_entry)
20991 +                               err = AuWbrCopyup(sbinfo, dentry);
20992 +                       else {
20993 +                               if (!IS_ROOT(dentry)) {
20994 +                                       di_read_lock_parent(parent, !AuLock_IR);
20995 +                                       err = AuWbrCopyup(sbinfo, dentry);
20996 +                                       di_read_unlock(parent, !AuLock_IR);
20997 +                               } else
20998 +                                       err = AuWbrCopyup(sbinfo, dentry);
20999 +                       }
21000 +                       bcpup = err;
21001 +                       if (unlikely(err < 0))
21002 +                               goto out;
21003 +               }
21004 +       } else {
21005 +               bcpup = args->force_btgt;
21006 +               AuDebugOn(au_test_ro(sb, bcpup, d_inode(dentry)));
21007 +       }
21008 +
21009 +       AuDbg("btop %d, bcpup %d\n", btop, bcpup);
21010 +       err = bcpup;
21011 +       if (bcpup == btop)
21012 +               goto out; /* success */
21013 +
21014 +       /* copyup the new parent into the branch we process */
21015 +       err = au_wr_dir_cpup(dentry, parent, add_entry, bcpup, btop);
21016 +       if (err >= 0) {
21017 +               if (d_really_is_negative(dentry)) {
21018 +                       au_set_h_dptr(dentry, btop, NULL);
21019 +                       au_set_dbtop(dentry, bcpup);
21020 +                       au_set_dbbot(dentry, bcpup);
21021 +               }
21022 +               AuDebugOn(add_entry
21023 +                         && !au_ftest_wrdir(args->flags, TMPFILE)
21024 +                         && !au_h_dptr(dentry, bcpup));
21025 +       }
21026 +
21027 +out:
21028 +       dput(parent);
21029 +       return err;
21030 +}
21031 +
21032 +/* ---------------------------------------------------------------------- */
21033 +
21034 +void au_pin_hdir_unlock(struct au_pin *p)
21035 +{
21036 +       if (p->hdir)
21037 +               au_hn_inode_unlock(p->hdir);
21038 +}
21039 +
21040 +int au_pin_hdir_lock(struct au_pin *p)
21041 +{
21042 +       int err;
21043 +
21044 +       err = 0;
21045 +       if (!p->hdir)
21046 +               goto out;
21047 +
21048 +       /* even if an error happens later, keep this lock */
21049 +       au_hn_inode_lock_nested(p->hdir, p->lsc_hi);
21050 +
21051 +       err = -EBUSY;
21052 +       if (unlikely(p->hdir->hi_inode != d_inode(p->h_parent)))
21053 +               goto out;
21054 +
21055 +       err = 0;
21056 +       if (p->h_dentry)
21057 +               err = au_h_verify(p->h_dentry, p->udba, p->hdir->hi_inode,
21058 +                                 p->h_parent, p->br);
21059 +
21060 +out:
21061 +       return err;
21062 +}
21063 +
21064 +int au_pin_hdir_relock(struct au_pin *p)
21065 +{
21066 +       int err, i;
21067 +       struct inode *h_i;
21068 +       struct dentry *h_d[] = {
21069 +               p->h_dentry,
21070 +               p->h_parent
21071 +       };
21072 +
21073 +       err = au_pin_hdir_lock(p);
21074 +       if (unlikely(err))
21075 +               goto out;
21076 +
21077 +       for (i = 0; !err && i < sizeof(h_d)/sizeof(*h_d); i++) {
21078 +               if (!h_d[i])
21079 +                       continue;
21080 +               if (d_is_positive(h_d[i])) {
21081 +                       h_i = d_inode(h_d[i]);
21082 +                       err = !h_i->i_nlink;
21083 +               }
21084 +       }
21085 +
21086 +out:
21087 +       return err;
21088 +}
21089 +
21090 +static void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task)
21091 +{
21092 +       atomic_long_set(&p->hdir->hi_inode->i_rwsem.owner, (long)task);
21093 +}
21094 +
21095 +void au_pin_hdir_acquire_nest(struct au_pin *p)
21096 +{
21097 +       if (p->hdir) {
21098 +               rwsem_acquire_nest(&p->hdir->hi_inode->i_rwsem.dep_map,
21099 +                                  p->lsc_hi, 0, NULL, _RET_IP_);
21100 +               au_pin_hdir_set_owner(p, current);
21101 +       }
21102 +}
21103 +
21104 +void au_pin_hdir_release(struct au_pin *p)
21105 +{
21106 +       if (p->hdir) {
21107 +               au_pin_hdir_set_owner(p, p->task);
21108 +               rwsem_release(&p->hdir->hi_inode->i_rwsem.dep_map, _RET_IP_);
21109 +       }
21110 +}
21111 +
21112 +struct dentry *au_pinned_h_parent(struct au_pin *pin)
21113 +{
21114 +       if (pin && pin->parent)
21115 +               return au_h_dptr(pin->parent, pin->bindex);
21116 +       return NULL;
21117 +}
21118 +
21119 +void au_unpin(struct au_pin *p)
21120 +{
21121 +       if (p->hdir)
21122 +               au_pin_hdir_unlock(p);
21123 +       if (p->h_mnt && au_ftest_pin(p->flags, MNT_WRITE))
21124 +               vfsub_mnt_drop_write(p->h_mnt);
21125 +       if (!p->hdir)
21126 +               return;
21127 +
21128 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
21129 +               di_read_unlock(p->parent, AuLock_IR);
21130 +       iput(p->hdir->hi_inode);
21131 +       dput(p->parent);
21132 +       p->parent = NULL;
21133 +       p->hdir = NULL;
21134 +       p->h_mnt = NULL;
21135 +       /* do not clear p->task */
21136 +}
21137 +
21138 +int au_do_pin(struct au_pin *p)
21139 +{
21140 +       int err;
21141 +       struct super_block *sb;
21142 +       struct inode *h_dir;
21143 +
21144 +       err = 0;
21145 +       sb = p->dentry->d_sb;
21146 +       p->br = au_sbr(sb, p->bindex);
21147 +       if (IS_ROOT(p->dentry)) {
21148 +               if (au_ftest_pin(p->flags, MNT_WRITE)) {
21149 +                       p->h_mnt = au_br_mnt(p->br);
21150 +                       err = vfsub_mnt_want_write(p->h_mnt);
21151 +                       if (unlikely(err)) {
21152 +                               au_fclr_pin(p->flags, MNT_WRITE);
21153 +                               goto out_err;
21154 +                       }
21155 +               }
21156 +               goto out;
21157 +       }
21158 +
21159 +       p->h_dentry = NULL;
21160 +       if (p->bindex <= au_dbbot(p->dentry))
21161 +               p->h_dentry = au_h_dptr(p->dentry, p->bindex);
21162 +
21163 +       p->parent = dget_parent(p->dentry);
21164 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
21165 +               di_read_lock(p->parent, AuLock_IR, p->lsc_di);
21166 +
21167 +       h_dir = NULL;
21168 +       p->h_parent = au_h_dptr(p->parent, p->bindex);
21169 +       p->hdir = au_hi(d_inode(p->parent), p->bindex);
21170 +       if (p->hdir)
21171 +               h_dir = p->hdir->hi_inode;
21172 +
21173 +       /*
21174 +        * udba case, or
21175 +        * if DI_LOCKED is not set, then p->parent may be different
21176 +        * and h_parent can be NULL.
21177 +        */
21178 +       if (unlikely(!p->hdir || !h_dir || !p->h_parent)) {
21179 +               err = -EBUSY;
21180 +               if (!au_ftest_pin(p->flags, DI_LOCKED))
21181 +                       di_read_unlock(p->parent, AuLock_IR);
21182 +               dput(p->parent);
21183 +               p->parent = NULL;
21184 +               goto out_err;
21185 +       }
21186 +
21187 +       if (au_ftest_pin(p->flags, MNT_WRITE)) {
21188 +               p->h_mnt = au_br_mnt(p->br);
21189 +               err = vfsub_mnt_want_write(p->h_mnt);
21190 +               if (unlikely(err)) {
21191 +                       au_fclr_pin(p->flags, MNT_WRITE);
21192 +                       if (!au_ftest_pin(p->flags, DI_LOCKED))
21193 +                               di_read_unlock(p->parent, AuLock_IR);
21194 +                       dput(p->parent);
21195 +                       p->parent = NULL;
21196 +                       goto out_err;
21197 +               }
21198 +       }
21199 +
21200 +       au_igrab(h_dir);
21201 +       err = au_pin_hdir_lock(p);
21202 +       if (!err)
21203 +               goto out; /* success */
21204 +
21205 +       au_unpin(p);
21206 +
21207 +out_err:
21208 +       pr_err("err %d\n", err);
21209 +       err = au_busy_or_stale();
21210 +out:
21211 +       return err;
21212 +}
21213 +
21214 +void au_pin_init(struct au_pin *p, struct dentry *dentry,
21215 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
21216 +                unsigned int udba, unsigned char flags)
21217 +{
21218 +       p->dentry = dentry;
21219 +       p->udba = udba;
21220 +       p->lsc_di = lsc_di;
21221 +       p->lsc_hi = lsc_hi;
21222 +       p->flags = flags;
21223 +       p->bindex = bindex;
21224 +
21225 +       p->parent = NULL;
21226 +       p->hdir = NULL;
21227 +       p->h_mnt = NULL;
21228 +
21229 +       p->h_dentry = NULL;
21230 +       p->h_parent = NULL;
21231 +       p->br = NULL;
21232 +       p->task = current;
21233 +}
21234 +
21235 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
21236 +          unsigned int udba, unsigned char flags)
21237 +{
21238 +       au_pin_init(pin, dentry, bindex, AuLsc_DI_PARENT, AuLsc_I_PARENT2,
21239 +                   udba, flags);
21240 +       return au_do_pin(pin);
21241 +}
21242 +
21243 +/* ---------------------------------------------------------------------- */
21244 +
21245 +/*
21246 + * ->setattr() and ->getattr() are called in various cases.
21247 + * chmod, stat: dentry is revalidated.
21248 + * fchmod, fstat: file and dentry are not revalidated, additionally they may be
21249 + *               unhashed.
21250 + * for ->setattr(), ia->ia_file is passed from ftruncate only.
21251 + */
21252 +/* todo: consolidate with do_refresh() and simple_reval_dpath() */
21253 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen)
21254 +{
21255 +       int err;
21256 +       struct dentry *parent;
21257 +
21258 +       err = 0;
21259 +       if (au_digen_test(dentry, sigen)) {
21260 +               parent = dget_parent(dentry);
21261 +               di_read_lock_parent(parent, AuLock_IR);
21262 +               err = au_refresh_dentry(dentry, parent);
21263 +               di_read_unlock(parent, AuLock_IR);
21264 +               dput(parent);
21265 +       }
21266 +
21267 +       AuTraceErr(err);
21268 +       return err;
21269 +}
21270 +
21271 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
21272 +                    struct au_icpup_args *a)
21273 +{
21274 +       int err;
21275 +       loff_t sz;
21276 +       aufs_bindex_t btop, ibtop;
21277 +       struct dentry *hi_wh, *parent;
21278 +       struct inode *inode;
21279 +       struct au_wr_dir_args wr_dir_args = {
21280 +               .force_btgt     = -1,
21281 +               .flags          = 0
21282 +       };
21283 +
21284 +       if (d_is_dir(dentry))
21285 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
21286 +       /* plink or hi_wh() case */
21287 +       btop = au_dbtop(dentry);
21288 +       inode = d_inode(dentry);
21289 +       ibtop = au_ibtop(inode);
21290 +       if (btop != ibtop && !au_test_ro(inode->i_sb, ibtop, inode))
21291 +               wr_dir_args.force_btgt = ibtop;
21292 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
21293 +       if (unlikely(err < 0))
21294 +               goto out;
21295 +       a->btgt = err;
21296 +       if (err != btop)
21297 +               au_fset_icpup(a->flags, DID_CPUP);
21298 +
21299 +       err = 0;
21300 +       a->pin_flags = AuPin_MNT_WRITE;
21301 +       parent = NULL;
21302 +       if (!IS_ROOT(dentry)) {
21303 +               au_fset_pin(a->pin_flags, DI_LOCKED);
21304 +               parent = dget_parent(dentry);
21305 +               di_write_lock_parent(parent);
21306 +       }
21307 +
21308 +       err = au_pin(&a->pin, dentry, a->btgt, a->udba, a->pin_flags);
21309 +       if (unlikely(err))
21310 +               goto out_parent;
21311 +
21312 +       sz = -1;
21313 +       a->h_path.dentry = au_h_dptr(dentry, btop);
21314 +       a->h_inode = d_inode(a->h_path.dentry);
21315 +       if (ia && (ia->ia_valid & ATTR_SIZE)) {
21316 +               inode_lock_shared_nested(a->h_inode, AuLsc_I_CHILD);
21317 +               if (ia->ia_size < i_size_read(a->h_inode))
21318 +                       sz = ia->ia_size;
21319 +               inode_unlock_shared(a->h_inode);
21320 +       }
21321 +
21322 +       hi_wh = NULL;
21323 +       if (au_ftest_icpup(a->flags, DID_CPUP) && d_unlinked(dentry)) {
21324 +               hi_wh = au_hi_wh(inode, a->btgt);
21325 +               if (!hi_wh) {
21326 +                       struct au_cp_generic cpg = {
21327 +                               .dentry = dentry,
21328 +                               .bdst   = a->btgt,
21329 +                               .bsrc   = -1,
21330 +                               .len    = sz,
21331 +                               .pin    = &a->pin
21332 +                       };
21333 +                       err = au_sio_cpup_wh(&cpg, /*file*/NULL);
21334 +                       if (unlikely(err))
21335 +                               goto out_unlock;
21336 +                       hi_wh = au_hi_wh(inode, a->btgt);
21337 +                       /* todo: revalidate hi_wh? */
21338 +               }
21339 +       }
21340 +
21341 +       if (parent) {
21342 +               au_pin_set_parent_lflag(&a->pin, /*lflag*/0);
21343 +               di_downgrade_lock(parent, AuLock_IR);
21344 +               dput(parent);
21345 +               parent = NULL;
21346 +       }
21347 +       if (!au_ftest_icpup(a->flags, DID_CPUP))
21348 +               goto out; /* success */
21349 +
21350 +       if (!d_unhashed(dentry)) {
21351 +               struct au_cp_generic cpg = {
21352 +                       .dentry = dentry,
21353 +                       .bdst   = a->btgt,
21354 +                       .bsrc   = btop,
21355 +                       .len    = sz,
21356 +                       .pin    = &a->pin,
21357 +                       .flags  = AuCpup_DTIME | AuCpup_HOPEN
21358 +               };
21359 +               err = au_sio_cpup_simple(&cpg);
21360 +               if (!err)
21361 +                       a->h_path.dentry = au_h_dptr(dentry, a->btgt);
21362 +       } else if (!hi_wh)
21363 +               a->h_path.dentry = au_h_dptr(dentry, a->btgt);
21364 +       else
21365 +               a->h_path.dentry = hi_wh; /* do not dget here */
21366 +
21367 +out_unlock:
21368 +       a->h_inode = d_inode(a->h_path.dentry);
21369 +       if (!err)
21370 +               goto out; /* success */
21371 +       au_unpin(&a->pin);
21372 +out_parent:
21373 +       if (parent) {
21374 +               di_write_unlock(parent);
21375 +               dput(parent);
21376 +       }
21377 +out:
21378 +       if (!err)
21379 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
21380 +       return err;
21381 +}
21382 +
21383 +static int aufs_setattr(struct user_namespace *userns, struct dentry *dentry,
21384 +                       struct iattr *ia)
21385 +{
21386 +       int err;
21387 +       struct inode *inode, *delegated;
21388 +       struct super_block *sb;
21389 +       struct file *file;
21390 +       struct au_icpup_args *a;
21391 +       struct user_namespace *h_userns;
21392 +
21393 +       inode = d_inode(dentry);
21394 +       IMustLock(inode);
21395 +
21396 +       err = setattr_prepare(userns, dentry, ia);
21397 +       if (unlikely(err))
21398 +               goto out;
21399 +
21400 +       err = -ENOMEM;
21401 +       a = kzalloc(sizeof(*a), GFP_NOFS);
21402 +       if (unlikely(!a))
21403 +               goto out;
21404 +
21405 +       if (ia->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
21406 +               ia->ia_valid &= ~ATTR_MODE;
21407 +
21408 +       file = NULL;
21409 +       sb = dentry->d_sb;
21410 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
21411 +       if (unlikely(err))
21412 +               goto out_kfree;
21413 +
21414 +       if (ia->ia_valid & ATTR_FILE) {
21415 +               /* currently ftruncate(2) only */
21416 +               AuDebugOn(!d_is_reg(dentry));
21417 +               file = ia->ia_file;
21418 +               err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1,
21419 +                                           /*fi_lsc*/0);
21420 +               if (unlikely(err))
21421 +                       goto out_si;
21422 +               ia->ia_file = au_hf_top(file);
21423 +               a->udba = AuOpt_UDBA_NONE;
21424 +       } else {
21425 +               /* fchmod() doesn't pass ia_file */
21426 +               a->udba = au_opt_udba(sb);
21427 +               di_write_lock_child(dentry);
21428 +               /* no d_unlinked(), to set UDBA_NONE for root */
21429 +               if (d_unhashed(dentry))
21430 +                       a->udba = AuOpt_UDBA_NONE;
21431 +               if (a->udba != AuOpt_UDBA_NONE) {
21432 +                       AuDebugOn(IS_ROOT(dentry));
21433 +                       err = au_reval_for_attr(dentry, au_sigen(sb));
21434 +                       if (unlikely(err))
21435 +                               goto out_dentry;
21436 +               }
21437 +       }
21438 +
21439 +       err = au_pin_and_icpup(dentry, ia, a);
21440 +       if (unlikely(err < 0))
21441 +               goto out_dentry;
21442 +       if (au_ftest_icpup(a->flags, DID_CPUP)) {
21443 +               ia->ia_file = NULL;
21444 +               ia->ia_valid &= ~ATTR_FILE;
21445 +       }
21446 +
21447 +       a->h_path.mnt = au_sbr_mnt(sb, a->btgt);
21448 +       if ((ia->ia_valid & (ATTR_MODE | ATTR_CTIME))
21449 +           == (ATTR_MODE | ATTR_CTIME)) {
21450 +               err = security_path_chmod(&a->h_path, ia->ia_mode);
21451 +               if (unlikely(err))
21452 +                       goto out_unlock;
21453 +       } else if ((ia->ia_valid & (ATTR_UID | ATTR_GID))
21454 +                  && (ia->ia_valid & ATTR_CTIME)) {
21455 +               err = security_path_chown(&a->h_path, ia->ia_uid, ia->ia_gid);
21456 +               if (unlikely(err))
21457 +                       goto out_unlock;
21458 +       }
21459 +
21460 +       if (ia->ia_valid & ATTR_SIZE) {
21461 +               struct file *f;
21462 +
21463 +               if (ia->ia_size < i_size_read(inode))
21464 +                       /* unmap only */
21465 +                       truncate_setsize(inode, ia->ia_size);
21466 +
21467 +               f = NULL;
21468 +               if (ia->ia_valid & ATTR_FILE)
21469 +                       f = ia->ia_file;
21470 +               inode_unlock(a->h_inode);
21471 +               err = vfsub_trunc(&a->h_path, ia->ia_size, ia->ia_valid, f);
21472 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
21473 +       } else {
21474 +               delegated = NULL;
21475 +               while (1) {
21476 +                       err = vfsub_notify_change(&a->h_path, ia, &delegated);
21477 +                       if (delegated) {
21478 +                               err = break_deleg_wait(&delegated);
21479 +                               if (!err)
21480 +                                       continue;
21481 +                       }
21482 +                       break;
21483 +               }
21484 +       }
21485 +       /*
21486 +        * regardless aufs 'acl' option setting.
21487 +        * why don't all acl-aware fs call this func from their ->setattr()?
21488 +        */
21489 +       if (!err && (ia->ia_valid & ATTR_MODE)) {
21490 +               h_userns = mnt_user_ns(a->h_path.mnt);
21491 +               err = vfsub_acl_chmod(h_userns, a->h_inode, ia->ia_mode);
21492 +       }
21493 +       if (!err)
21494 +               au_cpup_attr_changeable(inode);
21495 +
21496 +out_unlock:
21497 +       inode_unlock(a->h_inode);
21498 +       au_unpin(&a->pin);
21499 +       if (unlikely(err))
21500 +               au_update_dbtop(dentry);
21501 +out_dentry:
21502 +       di_write_unlock(dentry);
21503 +       if (file) {
21504 +               fi_write_unlock(file);
21505 +               ia->ia_file = file;
21506 +               ia->ia_valid |= ATTR_FILE;
21507 +       }
21508 +out_si:
21509 +       si_read_unlock(sb);
21510 +out_kfree:
21511 +       au_kfree_rcu(a);
21512 +out:
21513 +       AuTraceErr(err);
21514 +       return err;
21515 +}
21516 +
21517 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
21518 +static int au_h_path_to_set_attr(struct dentry *dentry,
21519 +                                struct au_icpup_args *a, struct path *h_path)
21520 +{
21521 +       int err;
21522 +       struct super_block *sb;
21523 +
21524 +       sb = dentry->d_sb;
21525 +       a->udba = au_opt_udba(sb);
21526 +       /* no d_unlinked(), to set UDBA_NONE for root */
21527 +       if (d_unhashed(dentry))
21528 +               a->udba = AuOpt_UDBA_NONE;
21529 +       if (a->udba != AuOpt_UDBA_NONE) {
21530 +               AuDebugOn(IS_ROOT(dentry));
21531 +               err = au_reval_for_attr(dentry, au_sigen(sb));
21532 +               if (unlikely(err))
21533 +                       goto out;
21534 +       }
21535 +       err = au_pin_and_icpup(dentry, /*ia*/NULL, a);
21536 +       if (unlikely(err < 0))
21537 +               goto out;
21538 +
21539 +       h_path->dentry = a->h_path.dentry;
21540 +       h_path->mnt = au_sbr_mnt(sb, a->btgt);
21541 +
21542 +out:
21543 +       return err;
21544 +}
21545 +
21546 +ssize_t au_sxattr(struct dentry *dentry, struct inode *inode,
21547 +                 struct au_sxattr *arg)
21548 +{
21549 +       int err;
21550 +       struct path h_path;
21551 +       struct super_block *sb;
21552 +       struct au_icpup_args *a;
21553 +       struct inode *h_inode;
21554 +       struct user_namespace *h_userns;
21555 +
21556 +       IMustLock(inode);
21557 +
21558 +       err = -ENOMEM;
21559 +       a = kzalloc(sizeof(*a), GFP_NOFS);
21560 +       if (unlikely(!a))
21561 +               goto out;
21562 +
21563 +       sb = dentry->d_sb;
21564 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
21565 +       if (unlikely(err))
21566 +               goto out_kfree;
21567 +
21568 +       h_path.dentry = NULL;   /* silence gcc */
21569 +       di_write_lock_child(dentry);
21570 +       err = au_h_path_to_set_attr(dentry, a, &h_path);
21571 +       if (unlikely(err))
21572 +               goto out_di;
21573 +       h_userns = mnt_user_ns(h_path.mnt);
21574 +
21575 +       inode_unlock(a->h_inode);
21576 +       switch (arg->type) {
21577 +       case AU_XATTR_SET:
21578 +               AuDebugOn(d_is_negative(h_path.dentry));
21579 +               err = vfsub_setxattr(h_userns, h_path.dentry,
21580 +                                    arg->u.set.name, arg->u.set.value,
21581 +                                    arg->u.set.size, arg->u.set.flags);
21582 +               break;
21583 +       case AU_ACL_SET:
21584 +               err = -EOPNOTSUPP;
21585 +               h_inode = d_inode(h_path.dentry);
21586 +               if (h_inode->i_op->set_acl) {
21587 +                       /* this will call posix_acl_update_mode */
21588 +                       err = h_inode->i_op->set_acl(h_userns, h_inode,
21589 +                                                    arg->u.acl_set.acl,
21590 +                                                    arg->u.acl_set.type);
21591 +               }
21592 +               break;
21593 +       }
21594 +       if (!err)
21595 +               au_cpup_attr_timesizes(inode);
21596 +
21597 +       au_unpin(&a->pin);
21598 +       if (unlikely(err))
21599 +               au_update_dbtop(dentry);
21600 +
21601 +out_di:
21602 +       di_write_unlock(dentry);
21603 +       si_read_unlock(sb);
21604 +out_kfree:
21605 +       au_kfree_rcu(a);
21606 +out:
21607 +       AuTraceErr(err);
21608 +       return err;
21609 +}
21610 +#endif
21611 +
21612 +static void au_refresh_iattr(struct inode *inode, struct kstat *st,
21613 +                            unsigned int nlink)
21614 +{
21615 +       unsigned int n;
21616 +
21617 +       inode->i_mode = st->mode;
21618 +       /* don't i_[ug]id_write() here */
21619 +       inode->i_uid = st->uid;
21620 +       inode->i_gid = st->gid;
21621 +       inode->i_atime = st->atime;
21622 +       inode->i_mtime = st->mtime;
21623 +       inode->i_ctime = st->ctime;
21624 +
21625 +       au_cpup_attr_nlink(inode, /*force*/0);
21626 +       if (S_ISDIR(inode->i_mode)) {
21627 +               n = inode->i_nlink;
21628 +               n -= nlink;
21629 +               n += st->nlink;
21630 +               smp_mb(); /* for i_nlink */
21631 +               /* 0 can happen */
21632 +               set_nlink(inode, n);
21633 +       }
21634 +
21635 +       spin_lock(&inode->i_lock);
21636 +       inode->i_blocks = st->blocks;
21637 +       i_size_write(inode, st->size);
21638 +       spin_unlock(&inode->i_lock);
21639 +}
21640 +
21641 +/*
21642 + * common routine for aufs_getattr() and au_getxattr().
21643 + * returns zero or negative (an error).
21644 + * @dentry will be read-locked in success.
21645 + */
21646 +int au_h_path_getattr(struct dentry *dentry, struct inode *inode, int force,
21647 +                     struct path *h_path, int locked)
21648 +{
21649 +       int err;
21650 +       unsigned int mnt_flags, sigen;
21651 +       unsigned char udba_none;
21652 +       aufs_bindex_t bindex;
21653 +       struct super_block *sb, *h_sb;
21654 +
21655 +       h_path->mnt = NULL;
21656 +       h_path->dentry = NULL;
21657 +
21658 +       err = 0;
21659 +       sb = dentry->d_sb;
21660 +       mnt_flags = au_mntflags(sb);
21661 +       udba_none = !!au_opt_test(mnt_flags, UDBA_NONE);
21662 +
21663 +       if (unlikely(locked))
21664 +               goto body; /* skip locking dinfo */
21665 +
21666 +       /* support fstat(2) */
21667 +       if (!d_unlinked(dentry) && !udba_none) {
21668 +               sigen = au_sigen(sb);
21669 +               err = au_digen_test(dentry, sigen);
21670 +               if (!err) {
21671 +                       di_read_lock_child(dentry, AuLock_IR);
21672 +                       err = au_dbrange_test(dentry);
21673 +                       if (unlikely(err)) {
21674 +                               di_read_unlock(dentry, AuLock_IR);
21675 +                               goto out;
21676 +                       }
21677 +               } else {
21678 +                       AuDebugOn(IS_ROOT(dentry));
21679 +                       di_write_lock_child(dentry);
21680 +                       err = au_dbrange_test(dentry);
21681 +                       if (!err)
21682 +                               err = au_reval_for_attr(dentry, sigen);
21683 +                       if (!err)
21684 +                               di_downgrade_lock(dentry, AuLock_IR);
21685 +                       else {
21686 +                               di_write_unlock(dentry);
21687 +                               goto out;
21688 +                       }
21689 +               }
21690 +       } else
21691 +               di_read_lock_child(dentry, AuLock_IR);
21692 +
21693 +body:
21694 +       if (!inode) {
21695 +               inode = d_inode(dentry);
21696 +               if (unlikely(!inode))
21697 +                       goto out;
21698 +       }
21699 +       bindex = au_ibtop(inode);
21700 +       h_path->mnt = au_sbr_mnt(sb, bindex);
21701 +       h_sb = h_path->mnt->mnt_sb;
21702 +       if (!force
21703 +           && !au_test_fs_bad_iattr(h_sb)
21704 +           && udba_none)
21705 +               goto out; /* success */
21706 +
21707 +       if (au_dbtop(dentry) == bindex)
21708 +               h_path->dentry = au_h_dptr(dentry, bindex);
21709 +       else if (au_opt_test(mnt_flags, PLINK) && au_plink_test(inode)) {
21710 +               h_path->dentry = au_plink_lkup(inode, bindex);
21711 +               if (IS_ERR(h_path->dentry))
21712 +                       /* pretending success */
21713 +                       h_path->dentry = NULL;
21714 +               else
21715 +                       dput(h_path->dentry);
21716 +       }
21717 +
21718 +out:
21719 +       return err;
21720 +}
21721 +
21722 +static int aufs_getattr(struct user_namespace *userns, const struct path *path,
21723 +                       struct kstat *st, u32 request, unsigned int query)
21724 +{
21725 +       int err;
21726 +       unsigned char positive;
21727 +       struct path h_path;
21728 +       struct dentry *dentry;
21729 +       struct inode *inode;
21730 +       struct super_block *sb;
21731 +
21732 +       dentry = path->dentry;
21733 +       inode = d_inode(dentry);
21734 +       sb = dentry->d_sb;
21735 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
21736 +       if (unlikely(err))
21737 +               goto out;
21738 +       err = au_h_path_getattr(dentry, /*inode*/NULL, /*force*/0, &h_path,
21739 +                               /*locked*/0);
21740 +       if (unlikely(err))
21741 +               goto out_si;
21742 +       if (unlikely(!h_path.dentry))
21743 +               /* illegally overlapped or something */
21744 +               goto out_fill; /* pretending success */
21745 +
21746 +       positive = d_is_positive(h_path.dentry);
21747 +       if (positive)
21748 +               /* no vfsub version */
21749 +               err = vfs_getattr(&h_path, st, request, query);
21750 +       if (!err) {
21751 +               if (positive)
21752 +                       au_refresh_iattr(inode, st,
21753 +                                        d_inode(h_path.dentry)->i_nlink);
21754 +               goto out_fill; /* success */
21755 +       }
21756 +       AuTraceErr(err);
21757 +       goto out_di;
21758 +
21759 +out_fill:
21760 +       generic_fillattr(userns, inode, st);
21761 +out_di:
21762 +       di_read_unlock(dentry, AuLock_IR);
21763 +out_si:
21764 +       si_read_unlock(sb);
21765 +out:
21766 +       AuTraceErr(err);
21767 +       return err;
21768 +}
21769 +
21770 +/* ---------------------------------------------------------------------- */
21771 +
21772 +static const char *aufs_get_link(struct dentry *dentry, struct inode *inode,
21773 +                                struct delayed_call *done)
21774 +{
21775 +       const char *ret;
21776 +       struct dentry *h_dentry;
21777 +       struct inode *h_inode;
21778 +       int err;
21779 +       aufs_bindex_t bindex;
21780 +
21781 +       ret = NULL; /* suppress a warning */
21782 +       err = -ECHILD;
21783 +       if (!dentry)
21784 +               goto out;
21785 +
21786 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
21787 +       if (unlikely(err))
21788 +               goto out;
21789 +
21790 +       err = au_d_hashed_positive(dentry);
21791 +       if (unlikely(err))
21792 +               goto out_unlock;
21793 +
21794 +       err = -EINVAL;
21795 +       inode = d_inode(dentry);
21796 +       bindex = au_ibtop(inode);
21797 +       h_inode = au_h_iptr(inode, bindex);
21798 +       if (unlikely(!h_inode->i_op->get_link))
21799 +               goto out_unlock;
21800 +
21801 +       err = -EBUSY;
21802 +       h_dentry = NULL;
21803 +       if (au_dbtop(dentry) <= bindex) {
21804 +               h_dentry = au_h_dptr(dentry, bindex);
21805 +               if (h_dentry)
21806 +                       dget(h_dentry);
21807 +       }
21808 +       if (!h_dentry) {
21809 +               h_dentry = d_find_any_alias(h_inode);
21810 +               if (IS_ERR(h_dentry)) {
21811 +                       err = PTR_ERR(h_dentry);
21812 +                       goto out_unlock;
21813 +               }
21814 +       }
21815 +       if (unlikely(!h_dentry))
21816 +               goto out_unlock;
21817 +
21818 +       err = 0;
21819 +       AuDbg("%ps\n", h_inode->i_op->get_link);
21820 +       AuDbgDentry(h_dentry);
21821 +       ret = vfs_get_link(h_dentry, done);
21822 +       dput(h_dentry);
21823 +       if (IS_ERR(ret))
21824 +               err = PTR_ERR(ret);
21825 +
21826 +out_unlock:
21827 +       aufs_read_unlock(dentry, AuLock_IR);
21828 +out:
21829 +       if (unlikely(err))
21830 +               ret = ERR_PTR(err);
21831 +       AuTraceErrPtr(ret);
21832 +       return ret;
21833 +}
21834 +
21835 +/* ---------------------------------------------------------------------- */
21836 +
21837 +static int au_is_special(struct inode *inode)
21838 +{
21839 +       return (inode->i_mode & (S_IFBLK | S_IFCHR | S_IFIFO | S_IFSOCK));
21840 +}
21841 +
21842 +static int aufs_update_time(struct inode *inode, struct timespec64 *ts,
21843 +                           int flags)
21844 +{
21845 +       int err;
21846 +       aufs_bindex_t bindex;
21847 +       struct super_block *sb;
21848 +       struct inode *h_inode;
21849 +       struct vfsmount *h_mnt;
21850 +
21851 +       sb = inode->i_sb;
21852 +       WARN_ONCE((flags & S_ATIME) && !IS_NOATIME(inode),
21853 +                 "unexpected s_flags 0x%lx", sb->s_flags);
21854 +
21855 +       /* mmap_sem might be acquired already, cf. aufs_mmap() */
21856 +       lockdep_off();
21857 +       si_read_lock(sb, AuLock_FLUSH);
21858 +       ii_write_lock_child(inode);
21859 +
21860 +       err = 0;
21861 +       bindex = au_ibtop(inode);
21862 +       h_inode = au_h_iptr(inode, bindex);
21863 +       if (!au_test_ro(sb, bindex, inode)) {
21864 +               h_mnt = au_sbr_mnt(sb, bindex);
21865 +               err = vfsub_mnt_want_write(h_mnt);
21866 +               if (!err) {
21867 +                       err = vfsub_update_time(h_inode, ts, flags);
21868 +                       vfsub_mnt_drop_write(h_mnt);
21869 +               }
21870 +       } else if (au_is_special(h_inode)) {
21871 +               /*
21872 +                * Never copy-up here.
21873 +                * These special files may already be opened and used for
21874 +                * communicating. If we copied it up, then the communication
21875 +                * would be corrupted.
21876 +                */
21877 +               AuWarn1("timestamps for i%lu are ignored "
21878 +                       "since it is on readonly branch (hi%lu).\n",
21879 +                       inode->i_ino, h_inode->i_ino);
21880 +       } else if (flags & ~S_ATIME) {
21881 +               err = -EIO;
21882 +               AuIOErr1("unexpected flags 0x%x\n", flags);
21883 +               AuDebugOn(1);
21884 +       }
21885 +
21886 +       if (!err)
21887 +               au_cpup_attr_timesizes(inode);
21888 +       ii_write_unlock(inode);
21889 +       si_read_unlock(sb);
21890 +       lockdep_on();
21891 +
21892 +       if (!err && (flags & S_VERSION))
21893 +               inode_inc_iversion(inode);
21894 +
21895 +       return err;
21896 +}
21897 +
21898 +/* ---------------------------------------------------------------------- */
21899 +
21900 +/* no getattr version will be set by module.c:aufs_init() */
21901 +struct inode_operations aufs_iop_nogetattr[AuIop_Last],
21902 +       aufs_iop[] = {
21903 +       [AuIop_SYMLINK] = {
21904 +               .permission     = aufs_permission,
21905 +#ifdef CONFIG_FS_POSIX_ACL
21906 +               .get_acl        = aufs_get_acl,
21907 +               .set_acl        = aufs_set_acl, /* unsupport for symlink? */
21908 +#endif
21909 +
21910 +               .setattr        = aufs_setattr,
21911 +               .getattr        = aufs_getattr,
21912 +
21913 +#ifdef CONFIG_AUFS_XATTR
21914 +               .listxattr      = aufs_listxattr,
21915 +#endif
21916 +
21917 +               .get_link       = aufs_get_link,
21918 +
21919 +               /* .update_time = aufs_update_time */
21920 +       },
21921 +       [AuIop_DIR] = {
21922 +               .create         = aufs_create,
21923 +               .lookup         = aufs_lookup,
21924 +               .link           = aufs_link,
21925 +               .unlink         = aufs_unlink,
21926 +               .symlink        = aufs_symlink,
21927 +               .mkdir          = aufs_mkdir,
21928 +               .rmdir          = aufs_rmdir,
21929 +               .mknod          = aufs_mknod,
21930 +               .rename         = aufs_rename,
21931 +
21932 +               .permission     = aufs_permission,
21933 +#ifdef CONFIG_FS_POSIX_ACL
21934 +               .get_acl        = aufs_get_acl,
21935 +               .set_acl        = aufs_set_acl,
21936 +#endif
21937 +
21938 +               .setattr        = aufs_setattr,
21939 +               .getattr        = aufs_getattr,
21940 +
21941 +#ifdef CONFIG_AUFS_XATTR
21942 +               .listxattr      = aufs_listxattr,
21943 +#endif
21944 +
21945 +               .update_time    = aufs_update_time,
21946 +               .atomic_open    = aufs_atomic_open,
21947 +               .tmpfile        = aufs_tmpfile
21948 +       },
21949 +       [AuIop_OTHER] = {
21950 +               .permission     = aufs_permission,
21951 +#ifdef CONFIG_FS_POSIX_ACL
21952 +               .get_acl        = aufs_get_acl,
21953 +               .set_acl        = aufs_set_acl,
21954 +#endif
21955 +
21956 +               .setattr        = aufs_setattr,
21957 +               .getattr        = aufs_getattr,
21958 +
21959 +#ifdef CONFIG_AUFS_XATTR
21960 +               .listxattr      = aufs_listxattr,
21961 +#endif
21962 +
21963 +               .update_time    = aufs_update_time
21964 +       }
21965 +};
21966 diff -urN /usr/share/empty/fs/aufs/i_op_del.c linux/fs/aufs/i_op_del.c
21967 --- /usr/share/empty/fs/aufs/i_op_del.c 1970-01-01 01:00:00.000000000 +0100
21968 +++ linux/fs/aufs/i_op_del.c    2021-12-03 15:40:58.236647297 +0100
21969 @@ -0,0 +1,522 @@
21970 +// SPDX-License-Identifier: GPL-2.0
21971 +/*
21972 + * Copyright (C) 2005-2021 Junjiro R. Okajima
21973 + *
21974 + * This program, aufs is free software; you can redistribute it and/or modify
21975 + * it under the terms of the GNU General Public License as published by
21976 + * the Free Software Foundation; either version 2 of the License, or
21977 + * (at your option) any later version.
21978 + *
21979 + * This program is distributed in the hope that it will be useful,
21980 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21981 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21982 + * GNU General Public License for more details.
21983 + *
21984 + * You should have received a copy of the GNU General Public License
21985 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21986 + */
21987 +
21988 +/*
21989 + * inode operations (del entry)
21990 + */
21991 +
21992 +#include <linux/iversion.h>
21993 +#include "aufs.h"
21994 +
21995 +/*
21996 + * decide if a new whiteout for @dentry is necessary or not.
21997 + * when it is necessary, prepare the parent dir for the upper branch whose
21998 + * branch index is @bcpup for creation. the actual creation of the whiteout will
21999 + * be done by caller.
22000 + * return value:
22001 + * 0: wh is unnecessary
22002 + * plus: wh is necessary
22003 + * minus: error
22004 + */
22005 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup)
22006 +{
22007 +       int need_wh, err;
22008 +       aufs_bindex_t btop;
22009 +       struct super_block *sb;
22010 +
22011 +       sb = dentry->d_sb;
22012 +       btop = au_dbtop(dentry);
22013 +       if (*bcpup < 0) {
22014 +               *bcpup = btop;
22015 +               if (au_test_ro(sb, btop, d_inode(dentry))) {
22016 +                       err = AuWbrCopyup(au_sbi(sb), dentry);
22017 +                       *bcpup = err;
22018 +                       if (unlikely(err < 0))
22019 +                               goto out;
22020 +               }
22021 +       } else
22022 +               AuDebugOn(btop < *bcpup
22023 +                         || au_test_ro(sb, *bcpup, d_inode(dentry)));
22024 +       AuDbg("bcpup %d, btop %d\n", *bcpup, btop);
22025 +
22026 +       if (*bcpup != btop) {
22027 +               err = au_cpup_dirs(dentry, *bcpup);
22028 +               if (unlikely(err))
22029 +                       goto out;
22030 +               need_wh = 1;
22031 +       } else {
22032 +               struct au_dinfo *dinfo, *tmp;
22033 +
22034 +               need_wh = -ENOMEM;
22035 +               dinfo = au_di(dentry);
22036 +               tmp = au_di_alloc(sb, AuLsc_DI_TMP);
22037 +               if (tmp) {
22038 +                       au_di_cp(tmp, dinfo);
22039 +                       au_di_swap(tmp, dinfo);
22040 +                       /* returns the number of positive dentries */
22041 +                       need_wh = au_lkup_dentry(dentry, btop + 1,
22042 +                                                /* AuLkup_IGNORE_PERM */ 0);
22043 +                       au_di_swap(tmp, dinfo);
22044 +                       au_rw_write_unlock(&tmp->di_rwsem);
22045 +                       au_di_free(tmp);
22046 +               }
22047 +       }
22048 +       AuDbg("need_wh %d\n", need_wh);
22049 +       err = need_wh;
22050 +
22051 +out:
22052 +       return err;
22053 +}
22054 +
22055 +/*
22056 + * simple tests for the del-entry operations.
22057 + * following the checks in vfs, plus the parent-child relationship.
22058 + */
22059 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
22060 +              struct dentry *h_parent, int isdir)
22061 +{
22062 +       int err;
22063 +       umode_t h_mode;
22064 +       struct dentry *h_dentry, *h_latest;
22065 +       struct inode *h_inode;
22066 +       struct path h_ppath;
22067 +       struct super_block *sb;
22068 +       struct au_branch *br;
22069 +       struct user_namespace *h_userns;
22070 +
22071 +       h_dentry = au_h_dptr(dentry, bindex);
22072 +       if (d_really_is_positive(dentry)) {
22073 +               err = -ENOENT;
22074 +               if (unlikely(d_is_negative(h_dentry)))
22075 +                       goto out;
22076 +               h_inode = d_inode(h_dentry);
22077 +               if (unlikely(!h_inode->i_nlink))
22078 +                       goto out;
22079 +
22080 +               h_mode = h_inode->i_mode;
22081 +               if (!isdir) {
22082 +                       err = -EISDIR;
22083 +                       if (unlikely(S_ISDIR(h_mode)))
22084 +                               goto out;
22085 +               } else if (unlikely(!S_ISDIR(h_mode))) {
22086 +                       err = -ENOTDIR;
22087 +                       goto out;
22088 +               }
22089 +       } else {
22090 +               /* rename(2) case */
22091 +               err = -EIO;
22092 +               if (unlikely(d_is_positive(h_dentry)))
22093 +                       goto out;
22094 +       }
22095 +
22096 +       err = -ENOENT;
22097 +       /* expected parent dir is locked */
22098 +       if (unlikely(h_parent != h_dentry->d_parent))
22099 +               goto out;
22100 +       err = 0;
22101 +
22102 +       /*
22103 +        * rmdir a dir may break the consistency on some filesystem.
22104 +        * let's try heavy test.
22105 +        */
22106 +       err = -EACCES;
22107 +       sb = dentry->d_sb;
22108 +       br = au_sbr(sb, bindex);
22109 +       h_userns = au_br_userns(br);
22110 +       if (unlikely(!au_opt_test(au_mntflags(sb), DIRPERM1)
22111 +                    && au_test_h_perm(h_userns, d_inode(h_parent),
22112 +                                      MAY_EXEC | MAY_WRITE)))
22113 +               goto out;
22114 +
22115 +       h_ppath.dentry = h_parent;
22116 +       h_ppath.mnt = au_br_mnt(br);
22117 +       h_latest = au_sio_lkup_one(h_userns, &dentry->d_name, &h_ppath);
22118 +       err = -EIO;
22119 +       if (IS_ERR(h_latest))
22120 +               goto out;
22121 +       if (h_latest == h_dentry)
22122 +               err = 0;
22123 +       dput(h_latest);
22124 +
22125 +out:
22126 +       return err;
22127 +}
22128 +
22129 +/*
22130 + * decide the branch where we operate for @dentry. the branch index will be set
22131 + * @rbcpup. after deciding it, 'pin' it and store the timestamps of the parent
22132 + * dir for reverting.
22133 + * when a new whiteout is necessary, create it.
22134 + */
22135 +static struct dentry*
22136 +lock_hdir_create_wh(struct dentry *dentry, int isdir, aufs_bindex_t *rbcpup,
22137 +                   struct au_dtime *dt, struct au_pin *pin)
22138 +{
22139 +       struct dentry *wh_dentry;
22140 +       struct super_block *sb;
22141 +       struct path h_path;
22142 +       int err, need_wh;
22143 +       unsigned int udba;
22144 +       aufs_bindex_t bcpup;
22145 +
22146 +       need_wh = au_wr_dir_need_wh(dentry, isdir, rbcpup);
22147 +       wh_dentry = ERR_PTR(need_wh);
22148 +       if (unlikely(need_wh < 0))
22149 +               goto out;
22150 +
22151 +       sb = dentry->d_sb;
22152 +       udba = au_opt_udba(sb);
22153 +       bcpup = *rbcpup;
22154 +       err = au_pin(pin, dentry, bcpup, udba,
22155 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
22156 +       wh_dentry = ERR_PTR(err);
22157 +       if (unlikely(err))
22158 +               goto out;
22159 +
22160 +       h_path.dentry = au_pinned_h_parent(pin);
22161 +       if (udba != AuOpt_UDBA_NONE
22162 +           && au_dbtop(dentry) == bcpup) {
22163 +               err = au_may_del(dentry, bcpup, h_path.dentry, isdir);
22164 +               wh_dentry = ERR_PTR(err);
22165 +               if (unlikely(err))
22166 +                       goto out_unpin;
22167 +       }
22168 +
22169 +       h_path.mnt = au_sbr_mnt(sb, bcpup);
22170 +       au_dtime_store(dt, au_pinned_parent(pin), &h_path);
22171 +       wh_dentry = NULL;
22172 +       if (!need_wh)
22173 +               goto out; /* success, no need to create whiteout */
22174 +
22175 +       wh_dentry = au_wh_create(dentry, bcpup, h_path.dentry);
22176 +       if (IS_ERR(wh_dentry))
22177 +               goto out_unpin;
22178 +
22179 +       /* returns with the parent is locked and wh_dentry is dget-ed */
22180 +       goto out; /* success */
22181 +
22182 +out_unpin:
22183 +       au_unpin(pin);
22184 +out:
22185 +       return wh_dentry;
22186 +}
22187 +
22188 +/*
22189 + * when removing a dir, rename it to a unique temporary whiteout-ed name first
22190 + * in order to be revertible and save time for removing many child whiteouts
22191 + * under the dir.
22192 + * returns 1 when there are too many child whiteout and caller should remove
22193 + * them asynchronously. returns 0 when the number of children is enough small to
22194 + * remove now or the branch fs is a remote fs.
22195 + * otherwise return an error.
22196 + */
22197 +static int renwh_and_rmdir(struct dentry *dentry, aufs_bindex_t bindex,
22198 +                          struct au_nhash *whlist, struct inode *dir)
22199 +{
22200 +       int rmdir_later, err, dirwh;
22201 +       struct dentry *h_dentry;
22202 +       struct super_block *sb;
22203 +       struct inode *inode;
22204 +
22205 +       sb = dentry->d_sb;
22206 +       SiMustAnyLock(sb);
22207 +       h_dentry = au_h_dptr(dentry, bindex);
22208 +       err = au_whtmp_ren(h_dentry, au_sbr(sb, bindex));
22209 +       if (unlikely(err))
22210 +               goto out;
22211 +
22212 +       /* stop monitoring */
22213 +       inode = d_inode(dentry);
22214 +       au_hn_free(au_hi(inode, bindex));
22215 +
22216 +       if (!au_test_fs_remote(h_dentry->d_sb)) {
22217 +               dirwh = au_sbi(sb)->si_dirwh;
22218 +               rmdir_later = (dirwh <= 1);
22219 +               if (!rmdir_later)
22220 +                       rmdir_later = au_nhash_test_longer_wh(whlist, bindex,
22221 +                                                             dirwh);
22222 +               if (rmdir_later)
22223 +                       return rmdir_later;
22224 +       }
22225 +
22226 +       err = au_whtmp_rmdir(dir, bindex, h_dentry, whlist);
22227 +       if (unlikely(err)) {
22228 +               AuIOErr("rmdir %pd, b%d failed, %d. ignored\n",
22229 +                       h_dentry, bindex, err);
22230 +               err = 0;
22231 +       }
22232 +
22233 +out:
22234 +       AuTraceErr(err);
22235 +       return err;
22236 +}
22237 +
22238 +/*
22239 + * final procedure for deleting a entry.
22240 + * maintain dentry and iattr.
22241 + */
22242 +static void epilog(struct inode *dir, struct dentry *dentry,
22243 +                  aufs_bindex_t bindex)
22244 +{
22245 +       struct inode *inode;
22246 +
22247 +       inode = d_inode(dentry);
22248 +       d_drop(dentry);
22249 +       inode->i_ctime = dir->i_ctime;
22250 +
22251 +       au_dir_ts(dir, bindex);
22252 +       inode_inc_iversion(dir);
22253 +}
22254 +
22255 +/*
22256 + * when an error happened, remove the created whiteout and revert everything.
22257 + */
22258 +static int do_revert(int err, struct inode *dir, aufs_bindex_t bindex,
22259 +                    aufs_bindex_t bwh, struct dentry *wh_dentry,
22260 +                    struct dentry *dentry, struct au_dtime *dt)
22261 +{
22262 +       int rerr;
22263 +       struct path h_path = {
22264 +               .dentry = wh_dentry,
22265 +               .mnt    = au_sbr_mnt(dir->i_sb, bindex)
22266 +       };
22267 +
22268 +       rerr = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path, dentry);
22269 +       if (!rerr) {
22270 +               au_set_dbwh(dentry, bwh);
22271 +               au_dtime_revert(dt);
22272 +               return 0;
22273 +       }
22274 +
22275 +       AuIOErr("%pd reverting whiteout failed(%d, %d)\n", dentry, err, rerr);
22276 +       return -EIO;
22277 +}
22278 +
22279 +/* ---------------------------------------------------------------------- */
22280 +
22281 +int aufs_unlink(struct inode *dir, struct dentry *dentry)
22282 +{
22283 +       int err;
22284 +       aufs_bindex_t bwh, bindex, btop;
22285 +       struct inode *inode, *h_dir, *delegated;
22286 +       struct dentry *parent, *wh_dentry;
22287 +       /* to reduce stack size */
22288 +       struct {
22289 +               struct au_dtime dt;
22290 +               struct au_pin pin;
22291 +               struct path h_path;
22292 +       } *a;
22293 +
22294 +       IMustLock(dir);
22295 +
22296 +       err = -ENOMEM;
22297 +       a = kmalloc(sizeof(*a), GFP_NOFS);
22298 +       if (unlikely(!a))
22299 +               goto out;
22300 +
22301 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
22302 +       if (unlikely(err))
22303 +               goto out_free;
22304 +       err = au_d_hashed_positive(dentry);
22305 +       if (unlikely(err))
22306 +               goto out_unlock;
22307 +       inode = d_inode(dentry);
22308 +       IMustLock(inode);
22309 +       err = -EISDIR;
22310 +       if (unlikely(d_is_dir(dentry)))
22311 +               goto out_unlock; /* possible? */
22312 +
22313 +       btop = au_dbtop(dentry);
22314 +       bwh = au_dbwh(dentry);
22315 +       bindex = -1;
22316 +       parent = dentry->d_parent; /* dir inode is locked */
22317 +       di_write_lock_parent(parent);
22318 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/0, &bindex, &a->dt,
22319 +                                       &a->pin);
22320 +       err = PTR_ERR(wh_dentry);
22321 +       if (IS_ERR(wh_dentry))
22322 +               goto out_parent;
22323 +
22324 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, btop);
22325 +       a->h_path.dentry = au_h_dptr(dentry, btop);
22326 +       dget(a->h_path.dentry);
22327 +       if (bindex == btop) {
22328 +               h_dir = au_pinned_h_dir(&a->pin);
22329 +               delegated = NULL;
22330 +               err = vfsub_unlink(h_dir, &a->h_path, &delegated, /*force*/0);
22331 +               if (unlikely(err == -EWOULDBLOCK)) {
22332 +                       pr_warn("cannot retry for NFSv4 delegation"
22333 +                               " for an internal unlink\n");
22334 +                       iput(delegated);
22335 +               }
22336 +       } else {
22337 +               /* dir inode is locked */
22338 +               h_dir = d_inode(wh_dentry->d_parent);
22339 +               IMustLock(h_dir);
22340 +               err = 0;
22341 +       }
22342 +
22343 +       if (!err) {
22344 +               vfsub_drop_nlink(inode);
22345 +               epilog(dir, dentry, bindex);
22346 +
22347 +               /* update target timestamps */
22348 +               if (bindex == btop) {
22349 +                       vfsub_update_h_iattr(&a->h_path, /*did*/NULL);
22350 +                       /*ignore*/
22351 +                       inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
22352 +               } else
22353 +                       /* todo: this timestamp may be reverted later */
22354 +                       inode->i_ctime = h_dir->i_ctime;
22355 +               goto out_unpin; /* success */
22356 +       }
22357 +
22358 +       /* revert */
22359 +       if (wh_dentry) {
22360 +               int rerr;
22361 +
22362 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
22363 +                                &a->dt);
22364 +               if (rerr)
22365 +                       err = rerr;
22366 +       }
22367 +
22368 +out_unpin:
22369 +       au_unpin(&a->pin);
22370 +       dput(wh_dentry);
22371 +       dput(a->h_path.dentry);
22372 +out_parent:
22373 +       di_write_unlock(parent);
22374 +out_unlock:
22375 +       aufs_read_unlock(dentry, AuLock_DW);
22376 +out_free:
22377 +       au_kfree_rcu(a);
22378 +out:
22379 +       return err;
22380 +}
22381 +
22382 +int aufs_rmdir(struct inode *dir, struct dentry *dentry)
22383 +{
22384 +       int err, rmdir_later;
22385 +       aufs_bindex_t bwh, bindex, btop;
22386 +       struct inode *inode;
22387 +       struct dentry *parent, *wh_dentry, *h_dentry;
22388 +       struct au_whtmp_rmdir *args;
22389 +       /* to reduce stack size */
22390 +       struct {
22391 +               struct au_dtime dt;
22392 +               struct au_pin pin;
22393 +       } *a;
22394 +
22395 +       IMustLock(dir);
22396 +
22397 +       err = -ENOMEM;
22398 +       a = kmalloc(sizeof(*a), GFP_NOFS);
22399 +       if (unlikely(!a))
22400 +               goto out;
22401 +
22402 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
22403 +       if (unlikely(err))
22404 +               goto out_free;
22405 +       err = au_alive_dir(dentry);
22406 +       if (unlikely(err))
22407 +               goto out_unlock;
22408 +       inode = d_inode(dentry);
22409 +       IMustLock(inode);
22410 +       err = -ENOTDIR;
22411 +       if (unlikely(!d_is_dir(dentry)))
22412 +               goto out_unlock; /* possible? */
22413 +
22414 +       err = -ENOMEM;
22415 +       args = au_whtmp_rmdir_alloc(dir->i_sb, GFP_NOFS);
22416 +       if (unlikely(!args))
22417 +               goto out_unlock;
22418 +
22419 +       parent = dentry->d_parent; /* dir inode is locked */
22420 +       di_write_lock_parent(parent);
22421 +       err = au_test_empty(dentry, &args->whlist);
22422 +       if (unlikely(err))
22423 +               goto out_parent;
22424 +
22425 +       btop = au_dbtop(dentry);
22426 +       bwh = au_dbwh(dentry);
22427 +       bindex = -1;
22428 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/1, &bindex, &a->dt,
22429 +                                       &a->pin);
22430 +       err = PTR_ERR(wh_dentry);
22431 +       if (IS_ERR(wh_dentry))
22432 +               goto out_parent;
22433 +
22434 +       h_dentry = au_h_dptr(dentry, btop);
22435 +       dget(h_dentry);
22436 +       rmdir_later = 0;
22437 +       if (bindex == btop) {
22438 +               err = renwh_and_rmdir(dentry, btop, &args->whlist, dir);
22439 +               if (err > 0) {
22440 +                       rmdir_later = err;
22441 +                       err = 0;
22442 +               }
22443 +       } else {
22444 +               /* stop monitoring */
22445 +               au_hn_free(au_hi(inode, btop));
22446 +
22447 +               /* dir inode is locked */
22448 +               IMustLock(d_inode(wh_dentry->d_parent));
22449 +               err = 0;
22450 +       }
22451 +
22452 +       if (!err) {
22453 +               vfsub_dead_dir(inode);
22454 +               au_set_dbdiropq(dentry, -1);
22455 +               epilog(dir, dentry, bindex);
22456 +
22457 +               if (rmdir_later) {
22458 +                       au_whtmp_kick_rmdir(dir, btop, h_dentry, args);
22459 +                       args = NULL;
22460 +               }
22461 +
22462 +               goto out_unpin; /* success */
22463 +       }
22464 +
22465 +       /* revert */
22466 +       AuLabel(revert);
22467 +       if (wh_dentry) {
22468 +               int rerr;
22469 +
22470 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
22471 +                                &a->dt);
22472 +               if (rerr)
22473 +                       err = rerr;
22474 +       }
22475 +
22476 +out_unpin:
22477 +       au_unpin(&a->pin);
22478 +       dput(wh_dentry);
22479 +       dput(h_dentry);
22480 +out_parent:
22481 +       di_write_unlock(parent);
22482 +       if (args)
22483 +               au_whtmp_rmdir_free(args);
22484 +out_unlock:
22485 +       aufs_read_unlock(dentry, AuLock_DW);
22486 +out_free:
22487 +       au_kfree_rcu(a);
22488 +out:
22489 +       AuTraceErr(err);
22490 +       return err;
22491 +}
22492 diff -urN /usr/share/empty/fs/aufs/i_op_ren.c linux/fs/aufs/i_op_ren.c
22493 --- /usr/share/empty/fs/aufs/i_op_ren.c 1970-01-01 01:00:00.000000000 +0100
22494 +++ linux/fs/aufs/i_op_ren.c    2021-12-03 15:40:58.236647297 +0100
22495 @@ -0,0 +1,1257 @@
22496 +// SPDX-License-Identifier: GPL-2.0
22497 +/*
22498 + * Copyright (C) 2005-2021 Junjiro R. Okajima
22499 + *
22500 + * This program, aufs is free software; you can redistribute it and/or modify
22501 + * it under the terms of the GNU General Public License as published by
22502 + * the Free Software Foundation; either version 2 of the License, or
22503 + * (at your option) any later version.
22504 + *
22505 + * This program is distributed in the hope that it will be useful,
22506 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22507 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22508 + * GNU General Public License for more details.
22509 + *
22510 + * You should have received a copy of the GNU General Public License
22511 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22512 + */
22513 +
22514 +/*
22515 + * inode operation (rename entry)
22516 + * todo: this is crazy monster
22517 + */
22518 +
22519 +#include <linux/iversion.h>
22520 +#include "aufs.h"
22521 +
22522 +enum { AuSRC, AuDST, AuSrcDst };
22523 +enum { AuPARENT, AuCHILD, AuParentChild };
22524 +
22525 +#define AuRen_ISDIR_SRC                1
22526 +#define AuRen_ISDIR_DST                (1 << 1)
22527 +#define AuRen_ISSAMEDIR                (1 << 2)
22528 +#define AuRen_WHSRC            (1 << 3)
22529 +#define AuRen_WHDST            (1 << 4)
22530 +#define AuRen_MNT_WRITE                (1 << 5)
22531 +#define AuRen_DT_DSTDIR                (1 << 6)
22532 +#define AuRen_DIROPQ_SRC       (1 << 7)
22533 +#define AuRen_DIROPQ_DST       (1 << 8)
22534 +#define AuRen_DIRREN           (1 << 9)
22535 +#define AuRen_DROPPED_SRC      (1 << 10)
22536 +#define AuRen_DROPPED_DST      (1 << 11)
22537 +#define au_ftest_ren(flags, name)      ((flags) & AuRen_##name)
22538 +#define au_fset_ren(flags, name) \
22539 +       do { (flags) |= AuRen_##name; } while (0)
22540 +#define au_fclr_ren(flags, name) \
22541 +       do { (flags) &= ~AuRen_##name; } while (0)
22542 +
22543 +#ifndef CONFIG_AUFS_DIRREN
22544 +#undef AuRen_DIRREN
22545 +#define AuRen_DIRREN           0
22546 +#endif
22547 +
22548 +struct au_ren_args {
22549 +       struct {
22550 +               struct dentry *dentry, *h_dentry, *parent, *h_parent,
22551 +                       *wh_dentry;
22552 +               struct inode *dir, *inode;
22553 +               struct au_hinode *hdir, *hinode;
22554 +               struct au_dtime dt[AuParentChild];
22555 +               aufs_bindex_t btop, bdiropq;
22556 +       } sd[AuSrcDst];
22557 +
22558 +#define src_dentry     sd[AuSRC].dentry
22559 +#define src_dir                sd[AuSRC].dir
22560 +#define src_inode      sd[AuSRC].inode
22561 +#define src_h_dentry   sd[AuSRC].h_dentry
22562 +#define src_parent     sd[AuSRC].parent
22563 +#define src_h_parent   sd[AuSRC].h_parent
22564 +#define src_wh_dentry  sd[AuSRC].wh_dentry
22565 +#define src_hdir       sd[AuSRC].hdir
22566 +#define src_hinode     sd[AuSRC].hinode
22567 +#define src_h_dir      sd[AuSRC].hdir->hi_inode
22568 +#define src_dt         sd[AuSRC].dt
22569 +#define src_btop       sd[AuSRC].btop
22570 +#define src_bdiropq    sd[AuSRC].bdiropq
22571 +
22572 +#define dst_dentry     sd[AuDST].dentry
22573 +#define dst_dir                sd[AuDST].dir
22574 +#define dst_inode      sd[AuDST].inode
22575 +#define dst_h_dentry   sd[AuDST].h_dentry
22576 +#define dst_parent     sd[AuDST].parent
22577 +#define dst_h_parent   sd[AuDST].h_parent
22578 +#define dst_wh_dentry  sd[AuDST].wh_dentry
22579 +#define dst_hdir       sd[AuDST].hdir
22580 +#define dst_hinode     sd[AuDST].hinode
22581 +#define dst_h_dir      sd[AuDST].hdir->hi_inode
22582 +#define dst_dt         sd[AuDST].dt
22583 +#define dst_btop       sd[AuDST].btop
22584 +#define dst_bdiropq    sd[AuDST].bdiropq
22585 +
22586 +       struct dentry *h_trap;
22587 +       struct au_branch *br;
22588 +       struct path h_path;
22589 +       struct au_nhash whlist;
22590 +       aufs_bindex_t btgt, src_bwh;
22591 +
22592 +       struct {
22593 +               unsigned short auren_flags;
22594 +               unsigned char flags;    /* syscall parameter */
22595 +               unsigned char exchange;
22596 +       } __packed;
22597 +
22598 +       struct au_whtmp_rmdir *thargs;
22599 +       struct dentry *h_dst;
22600 +       struct au_hinode *h_root;
22601 +};
22602 +
22603 +/* ---------------------------------------------------------------------- */
22604 +
22605 +/*
22606 + * functions for reverting.
22607 + * when an error happened in a single rename systemcall, we should revert
22608 + * everything as if nothing happened.
22609 + * we don't need to revert the copied-up/down the parent dir since they are
22610 + * harmless.
22611 + */
22612 +
22613 +#define RevertFailure(fmt, ...) do { \
22614 +       AuIOErr("revert failure: " fmt " (%d, %d)\n", \
22615 +               ##__VA_ARGS__, err, rerr); \
22616 +       err = -EIO; \
22617 +} while (0)
22618 +
22619 +static void au_ren_do_rev_diropq(int err, struct au_ren_args *a, int idx)
22620 +{
22621 +       int rerr;
22622 +       struct dentry *d;
22623 +#define src_or_dst(member) a->sd[idx].member
22624 +
22625 +       d = src_or_dst(dentry); /* {src,dst}_dentry */
22626 +       au_hn_inode_lock_nested(src_or_dst(hinode), AuLsc_I_CHILD);
22627 +       rerr = au_diropq_remove(d, a->btgt);
22628 +       au_hn_inode_unlock(src_or_dst(hinode));
22629 +       au_set_dbdiropq(d, src_or_dst(bdiropq));
22630 +       if (rerr)
22631 +               RevertFailure("remove diropq %pd", d);
22632 +
22633 +#undef src_or_dst_
22634 +}
22635 +
22636 +static void au_ren_rev_diropq(int err, struct au_ren_args *a)
22637 +{
22638 +       if (au_ftest_ren(a->auren_flags, DIROPQ_SRC))
22639 +               au_ren_do_rev_diropq(err, a, AuSRC);
22640 +       if (au_ftest_ren(a->auren_flags, DIROPQ_DST))
22641 +               au_ren_do_rev_diropq(err, a, AuDST);
22642 +}
22643 +
22644 +static void au_ren_rev_rename(int err, struct au_ren_args *a)
22645 +{
22646 +       int rerr;
22647 +       struct inode *delegated;
22648 +       struct path h_ppath = {
22649 +               .dentry = a->src_h_parent,
22650 +               .mnt    = a->h_path.mnt
22651 +       };
22652 +
22653 +       a->h_path.dentry = vfsub_lkup_one(&a->src_dentry->d_name, &h_ppath);
22654 +       rerr = PTR_ERR(a->h_path.dentry);
22655 +       if (IS_ERR(a->h_path.dentry)) {
22656 +               RevertFailure("lkup one %pd", a->src_dentry);
22657 +               return;
22658 +       }
22659 +
22660 +       delegated = NULL;
22661 +       rerr = vfsub_rename(a->dst_h_dir,
22662 +                           au_h_dptr(a->src_dentry, a->btgt),
22663 +                           a->src_h_dir, &a->h_path, &delegated, a->flags);
22664 +       if (unlikely(rerr == -EWOULDBLOCK)) {
22665 +               pr_warn("cannot retry for NFSv4 delegation"
22666 +                       " for an internal rename\n");
22667 +               iput(delegated);
22668 +       }
22669 +       d_drop(a->h_path.dentry);
22670 +       dput(a->h_path.dentry);
22671 +       /* au_set_h_dptr(a->src_dentry, a->btgt, NULL); */
22672 +       if (rerr)
22673 +               RevertFailure("rename %pd", a->src_dentry);
22674 +}
22675 +
22676 +static void au_ren_rev_whtmp(int err, struct au_ren_args *a)
22677 +{
22678 +       int rerr;
22679 +       struct inode *delegated;
22680 +       struct path h_ppath = {
22681 +               .dentry = a->dst_h_parent,
22682 +               .mnt    = a->h_path.mnt
22683 +       };
22684 +
22685 +       a->h_path.dentry = vfsub_lkup_one(&a->dst_dentry->d_name, &h_ppath);
22686 +       rerr = PTR_ERR(a->h_path.dentry);
22687 +       if (IS_ERR(a->h_path.dentry)) {
22688 +               RevertFailure("lkup one %pd", a->dst_dentry);
22689 +               return;
22690 +       }
22691 +       if (d_is_positive(a->h_path.dentry)) {
22692 +               d_drop(a->h_path.dentry);
22693 +               dput(a->h_path.dentry);
22694 +               return;
22695 +       }
22696 +
22697 +       delegated = NULL;
22698 +       rerr = vfsub_rename(a->dst_h_dir, a->h_dst, a->dst_h_dir, &a->h_path,
22699 +                           &delegated, a->flags);
22700 +       if (unlikely(rerr == -EWOULDBLOCK)) {
22701 +               pr_warn("cannot retry for NFSv4 delegation"
22702 +                       " for an internal rename\n");
22703 +               iput(delegated);
22704 +       }
22705 +       d_drop(a->h_path.dentry);
22706 +       dput(a->h_path.dentry);
22707 +       if (!rerr)
22708 +               au_set_h_dptr(a->dst_dentry, a->btgt, dget(a->h_dst));
22709 +       else
22710 +               RevertFailure("rename %pd", a->h_dst);
22711 +}
22712 +
22713 +static void au_ren_rev_whsrc(int err, struct au_ren_args *a)
22714 +{
22715 +       int rerr;
22716 +
22717 +       a->h_path.dentry = a->src_wh_dentry;
22718 +       rerr = au_wh_unlink_dentry(a->src_h_dir, &a->h_path, a->src_dentry);
22719 +       au_set_dbwh(a->src_dentry, a->src_bwh);
22720 +       if (rerr)
22721 +               RevertFailure("unlink %pd", a->src_wh_dentry);
22722 +}
22723 +#undef RevertFailure
22724 +
22725 +/* ---------------------------------------------------------------------- */
22726 +
22727 +/*
22728 + * when we have to copyup the renaming entry, do it with the rename-target name
22729 + * in order to minimize the cost (the later actual rename is unnecessary).
22730 + * otherwise rename it on the target branch.
22731 + */
22732 +static int au_ren_or_cpup(struct au_ren_args *a)
22733 +{
22734 +       int err;
22735 +       struct dentry *d;
22736 +       struct inode *delegated;
22737 +
22738 +       d = a->src_dentry;
22739 +       if (au_dbtop(d) == a->btgt) {
22740 +               a->h_path.dentry = a->dst_h_dentry;
22741 +               AuDebugOn(au_dbtop(d) != a->btgt);
22742 +               delegated = NULL;
22743 +               err = vfsub_rename(a->src_h_dir, au_h_dptr(d, a->btgt),
22744 +                                  a->dst_h_dir, &a->h_path, &delegated,
22745 +                                  a->flags);
22746 +               if (unlikely(err == -EWOULDBLOCK)) {
22747 +                       pr_warn("cannot retry for NFSv4 delegation"
22748 +                               " for an internal rename\n");
22749 +                       iput(delegated);
22750 +               }
22751 +       } else
22752 +               BUG();
22753 +
22754 +       if (!err && a->h_dst)
22755 +               /* it will be set to dinfo later */
22756 +               dget(a->h_dst);
22757 +
22758 +       return err;
22759 +}
22760 +
22761 +/* cf. aufs_rmdir() */
22762 +static int au_ren_del_whtmp(struct au_ren_args *a)
22763 +{
22764 +       int err;
22765 +       struct inode *dir;
22766 +
22767 +       dir = a->dst_dir;
22768 +       SiMustAnyLock(dir->i_sb);
22769 +       if (!au_nhash_test_longer_wh(&a->whlist, a->btgt,
22770 +                                    au_sbi(dir->i_sb)->si_dirwh)
22771 +           || au_test_fs_remote(a->h_dst->d_sb)) {
22772 +               err = au_whtmp_rmdir(dir, a->btgt, a->h_dst, &a->whlist);
22773 +               if (unlikely(err))
22774 +                       pr_warn("failed removing whtmp dir %pd (%d), "
22775 +                               "ignored.\n", a->h_dst, err);
22776 +       } else {
22777 +               au_nhash_wh_free(&a->thargs->whlist);
22778 +               a->thargs->whlist = a->whlist;
22779 +               a->whlist.nh_num = 0;
22780 +               au_whtmp_kick_rmdir(dir, a->btgt, a->h_dst, a->thargs);
22781 +               dput(a->h_dst);
22782 +               a->thargs = NULL;
22783 +       }
22784 +
22785 +       return 0;
22786 +}
22787 +
22788 +/* make it 'opaque' dir. */
22789 +static int au_ren_do_diropq(struct au_ren_args *a, int idx)
22790 +{
22791 +       int err;
22792 +       struct dentry *d, *diropq;
22793 +#define src_or_dst(member) a->sd[idx].member
22794 +
22795 +       err = 0;
22796 +       d = src_or_dst(dentry); /* {src,dst}_dentry */
22797 +       src_or_dst(bdiropq) = au_dbdiropq(d);
22798 +       src_or_dst(hinode) = au_hi(src_or_dst(inode), a->btgt);
22799 +       au_hn_inode_lock_nested(src_or_dst(hinode), AuLsc_I_CHILD);
22800 +       diropq = au_diropq_create(d, a->btgt);
22801 +       au_hn_inode_unlock(src_or_dst(hinode));
22802 +       if (IS_ERR(diropq))
22803 +               err = PTR_ERR(diropq);
22804 +       else
22805 +               dput(diropq);
22806 +
22807 +#undef src_or_dst_
22808 +       return err;
22809 +}
22810 +
22811 +static int au_ren_diropq(struct au_ren_args *a)
22812 +{
22813 +       int err;
22814 +       unsigned char always;
22815 +       struct dentry *d;
22816 +
22817 +       err = 0;
22818 +       d = a->dst_dentry; /* already renamed on the branch */
22819 +       always = !!au_opt_test(au_mntflags(d->d_sb), ALWAYS_DIROPQ);
22820 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)
22821 +           && !au_ftest_ren(a->auren_flags, DIRREN)
22822 +           && a->btgt != au_dbdiropq(a->src_dentry)
22823 +           && (a->dst_wh_dentry
22824 +               || a->btgt <= au_dbdiropq(d)
22825 +               /* hide the lower to keep xino */
22826 +               /* the lowers may not be a dir, but we hide them anyway */
22827 +               || a->btgt < au_dbbot(d)
22828 +               || always)) {
22829 +               AuDbg("here\n");
22830 +               err = au_ren_do_diropq(a, AuSRC);
22831 +               if (unlikely(err))
22832 +                       goto out;
22833 +               au_fset_ren(a->auren_flags, DIROPQ_SRC);
22834 +       }
22835 +       if (!a->exchange)
22836 +               goto out; /* success */
22837 +
22838 +       d = a->src_dentry; /* already renamed on the branch */
22839 +       if (au_ftest_ren(a->auren_flags, ISDIR_DST)
22840 +           && a->btgt != au_dbdiropq(a->dst_dentry)
22841 +           && (a->btgt < au_dbdiropq(d)
22842 +               || a->btgt < au_dbbot(d)
22843 +               || always)) {
22844 +               AuDbgDentry(a->src_dentry);
22845 +               AuDbgDentry(a->dst_dentry);
22846 +               err = au_ren_do_diropq(a, AuDST);
22847 +               if (unlikely(err))
22848 +                       goto out_rev_src;
22849 +               au_fset_ren(a->auren_flags, DIROPQ_DST);
22850 +       }
22851 +       goto out; /* success */
22852 +
22853 +out_rev_src:
22854 +       AuDbg("err %d, reverting src\n", err);
22855 +       au_ren_rev_diropq(err, a);
22856 +out:
22857 +       return err;
22858 +}
22859 +
22860 +static int do_rename(struct au_ren_args *a)
22861 +{
22862 +       int err;
22863 +       struct dentry *d, *h_d;
22864 +
22865 +       if (!a->exchange) {
22866 +               /* prepare workqueue args for asynchronous rmdir */
22867 +               h_d = a->dst_h_dentry;
22868 +               if (au_ftest_ren(a->auren_flags, ISDIR_DST)
22869 +                   /* && !au_ftest_ren(a->auren_flags, DIRREN) */
22870 +                   && d_is_positive(h_d)) {
22871 +                       err = -ENOMEM;
22872 +                       a->thargs = au_whtmp_rmdir_alloc(a->src_dentry->d_sb,
22873 +                                                        GFP_NOFS);
22874 +                       if (unlikely(!a->thargs))
22875 +                               goto out;
22876 +                       a->h_dst = dget(h_d);
22877 +               }
22878 +
22879 +               /* create whiteout for src_dentry */
22880 +               if (au_ftest_ren(a->auren_flags, WHSRC)) {
22881 +                       a->src_bwh = au_dbwh(a->src_dentry);
22882 +                       AuDebugOn(a->src_bwh >= 0);
22883 +                       a->src_wh_dentry = au_wh_create(a->src_dentry, a->btgt,
22884 +                                                       a->src_h_parent);
22885 +                       err = PTR_ERR(a->src_wh_dentry);
22886 +                       if (IS_ERR(a->src_wh_dentry))
22887 +                               goto out_thargs;
22888 +               }
22889 +
22890 +               /* lookup whiteout for dentry */
22891 +               if (au_ftest_ren(a->auren_flags, WHDST)) {
22892 +                       h_d = au_wh_lkup(a->dst_h_parent,
22893 +                                        &a->dst_dentry->d_name, a->br);
22894 +                       err = PTR_ERR(h_d);
22895 +                       if (IS_ERR(h_d))
22896 +                               goto out_whsrc;
22897 +                       if (d_is_negative(h_d))
22898 +                               dput(h_d);
22899 +                       else
22900 +                               a->dst_wh_dentry = h_d;
22901 +               }
22902 +
22903 +               /* rename dentry to tmpwh */
22904 +               if (a->thargs) {
22905 +                       err = au_whtmp_ren(a->dst_h_dentry, a->br);
22906 +                       if (unlikely(err))
22907 +                               goto out_whdst;
22908 +
22909 +                       d = a->dst_dentry;
22910 +                       au_set_h_dptr(d, a->btgt, NULL);
22911 +                       err = au_lkup_neg(d, a->btgt, /*wh*/0);
22912 +                       if (unlikely(err))
22913 +                               goto out_whtmp;
22914 +                       a->dst_h_dentry = au_h_dptr(d, a->btgt);
22915 +               }
22916 +       }
22917 +
22918 +       BUG_ON(d_is_positive(a->dst_h_dentry) && a->src_btop != a->btgt);
22919 +#if 0 /* debugging */
22920 +       BUG_ON(!au_ftest_ren(a->auren_flags, DIRREN)
22921 +              && d_is_positive(a->dst_h_dentry)
22922 +              && a->src_btop != a->btgt);
22923 +#endif
22924 +
22925 +       /* rename by vfs_rename or cpup */
22926 +       err = au_ren_or_cpup(a);
22927 +       if (unlikely(err))
22928 +               /* leave the copied-up one */
22929 +               goto out_whtmp;
22930 +
22931 +       /* make dir opaque */
22932 +       err = au_ren_diropq(a);
22933 +       if (unlikely(err))
22934 +               goto out_rename;
22935 +
22936 +       /* update target timestamps */
22937 +       if (a->exchange) {
22938 +               AuDebugOn(au_dbtop(a->dst_dentry) != a->btgt);
22939 +               a->h_path.dentry = au_h_dptr(a->dst_dentry, a->btgt);
22940 +               vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
22941 +               a->dst_inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
22942 +       }
22943 +       AuDebugOn(au_dbtop(a->src_dentry) != a->btgt);
22944 +       a->h_path.dentry = au_h_dptr(a->src_dentry, a->btgt);
22945 +       vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
22946 +       a->src_inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
22947 +
22948 +       if (!a->exchange) {
22949 +               /* remove whiteout for dentry */
22950 +               if (a->dst_wh_dentry) {
22951 +                       a->h_path.dentry = a->dst_wh_dentry;
22952 +                       err = au_wh_unlink_dentry(a->dst_h_dir, &a->h_path,
22953 +                                                 a->dst_dentry);
22954 +                       if (unlikely(err))
22955 +                               goto out_diropq;
22956 +               }
22957 +
22958 +               /* remove whtmp */
22959 +               if (a->thargs)
22960 +                       au_ren_del_whtmp(a); /* ignore this error */
22961 +
22962 +               au_fhsm_wrote(a->src_dentry->d_sb, a->btgt, /*force*/0);
22963 +       }
22964 +       err = 0;
22965 +       goto out_success;
22966 +
22967 +out_diropq:
22968 +       au_ren_rev_diropq(err, a);
22969 +out_rename:
22970 +       au_ren_rev_rename(err, a);
22971 +       dput(a->h_dst);
22972 +out_whtmp:
22973 +       if (a->thargs)
22974 +               au_ren_rev_whtmp(err, a);
22975 +out_whdst:
22976 +       dput(a->dst_wh_dentry);
22977 +       a->dst_wh_dentry = NULL;
22978 +out_whsrc:
22979 +       if (a->src_wh_dentry)
22980 +               au_ren_rev_whsrc(err, a);
22981 +out_success:
22982 +       dput(a->src_wh_dentry);
22983 +       dput(a->dst_wh_dentry);
22984 +out_thargs:
22985 +       if (a->thargs) {
22986 +               dput(a->h_dst);
22987 +               au_whtmp_rmdir_free(a->thargs);
22988 +               a->thargs = NULL;
22989 +       }
22990 +out:
22991 +       return err;
22992 +}
22993 +
22994 +/* ---------------------------------------------------------------------- */
22995 +
22996 +/*
22997 + * test if @dentry dir can be rename destination or not.
22998 + * success means, it is a logically empty dir.
22999 + */
23000 +static int may_rename_dstdir(struct dentry *dentry, struct au_nhash *whlist)
23001 +{
23002 +       return au_test_empty(dentry, whlist);
23003 +}
23004 +
23005 +/*
23006 + * test if @a->src_dentry dir can be rename source or not.
23007 + * if it can, return 0.
23008 + * success means,
23009 + * - it is a logically empty dir.
23010 + * - or, it exists on writable branch and has no children including whiteouts
23011 + *   on the lower branch unless DIRREN is on.
23012 + */
23013 +static int may_rename_srcdir(struct au_ren_args *a)
23014 +{
23015 +       int err;
23016 +       unsigned int rdhash;
23017 +       aufs_bindex_t btop, btgt;
23018 +       struct dentry *dentry;
23019 +       struct super_block *sb;
23020 +       struct au_sbinfo *sbinfo;
23021 +
23022 +       dentry = a->src_dentry;
23023 +       sb = dentry->d_sb;
23024 +       sbinfo = au_sbi(sb);
23025 +       if (au_opt_test(sbinfo->si_mntflags, DIRREN))
23026 +               au_fset_ren(a->auren_flags, DIRREN);
23027 +
23028 +       btgt = a->btgt;
23029 +       btop = au_dbtop(dentry);
23030 +       if (btop != btgt) {
23031 +               struct au_nhash whlist;
23032 +
23033 +               SiMustAnyLock(sb);
23034 +               rdhash = sbinfo->si_rdhash;
23035 +               if (!rdhash)
23036 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL,
23037 +                                                          dentry));
23038 +               err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
23039 +               if (unlikely(err))
23040 +                       goto out;
23041 +               err = au_test_empty(dentry, &whlist);
23042 +               au_nhash_wh_free(&whlist);
23043 +               goto out;
23044 +       }
23045 +
23046 +       if (btop == au_dbtaildir(dentry))
23047 +               return 0; /* success */
23048 +
23049 +       err = au_test_empty_lower(dentry);
23050 +
23051 +out:
23052 +       if (err == -ENOTEMPTY) {
23053 +               if (au_ftest_ren(a->auren_flags, DIRREN)) {
23054 +                       err = 0;
23055 +               } else {
23056 +                       AuWarn1("renaming dir who has child(ren) on multiple "
23057 +                               "branches, is not supported\n");
23058 +                       err = -EXDEV;
23059 +               }
23060 +       }
23061 +       return err;
23062 +}
23063 +
23064 +/* side effect: sets whlist and h_dentry */
23065 +static int au_ren_may_dir(struct au_ren_args *a)
23066 +{
23067 +       int err;
23068 +       unsigned int rdhash;
23069 +       struct dentry *d;
23070 +
23071 +       d = a->dst_dentry;
23072 +       SiMustAnyLock(d->d_sb);
23073 +
23074 +       err = 0;
23075 +       if (au_ftest_ren(a->auren_flags, ISDIR_DST) && a->dst_inode) {
23076 +               rdhash = au_sbi(d->d_sb)->si_rdhash;
23077 +               if (!rdhash)
23078 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, d));
23079 +               err = au_nhash_alloc(&a->whlist, rdhash, GFP_NOFS);
23080 +               if (unlikely(err))
23081 +                       goto out;
23082 +
23083 +               if (!a->exchange) {
23084 +                       au_set_dbtop(d, a->dst_btop);
23085 +                       err = may_rename_dstdir(d, &a->whlist);
23086 +                       au_set_dbtop(d, a->btgt);
23087 +               } else
23088 +                       err = may_rename_srcdir(a);
23089 +       }
23090 +       a->dst_h_dentry = au_h_dptr(d, au_dbtop(d));
23091 +       if (unlikely(err))
23092 +               goto out;
23093 +
23094 +       d = a->src_dentry;
23095 +       a->src_h_dentry = au_h_dptr(d, au_dbtop(d));
23096 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)) {
23097 +               err = may_rename_srcdir(a);
23098 +               if (unlikely(err)) {
23099 +                       au_nhash_wh_free(&a->whlist);
23100 +                       a->whlist.nh_num = 0;
23101 +               }
23102 +       }
23103 +out:
23104 +       return err;
23105 +}
23106 +
23107 +/* ---------------------------------------------------------------------- */
23108 +
23109 +/*
23110 + * simple tests for rename.
23111 + * following the checks in vfs, plus the parent-child relationship.
23112 + */
23113 +static int au_may_ren(struct au_ren_args *a)
23114 +{
23115 +       int err, isdir;
23116 +       struct inode *h_inode;
23117 +
23118 +       if (a->src_btop == a->btgt) {
23119 +               err = au_may_del(a->src_dentry, a->btgt, a->src_h_parent,
23120 +                                au_ftest_ren(a->auren_flags, ISDIR_SRC));
23121 +               if (unlikely(err))
23122 +                       goto out;
23123 +               err = -EINVAL;
23124 +               if (unlikely(a->src_h_dentry == a->h_trap))
23125 +                       goto out;
23126 +       }
23127 +
23128 +       err = 0;
23129 +       if (a->dst_btop != a->btgt)
23130 +               goto out;
23131 +
23132 +       err = -ENOTEMPTY;
23133 +       if (unlikely(a->dst_h_dentry == a->h_trap))
23134 +               goto out;
23135 +
23136 +       err = -EIO;
23137 +       isdir = !!au_ftest_ren(a->auren_flags, ISDIR_DST);
23138 +       if (d_really_is_negative(a->dst_dentry)) {
23139 +               if (d_is_negative(a->dst_h_dentry))
23140 +                       err = au_may_add(a->dst_dentry, a->btgt,
23141 +                                        a->dst_h_parent, isdir);
23142 +       } else {
23143 +               if (unlikely(d_is_negative(a->dst_h_dentry)))
23144 +                       goto out;
23145 +               h_inode = d_inode(a->dst_h_dentry);
23146 +               if (h_inode->i_nlink)
23147 +                       err = au_may_del(a->dst_dentry, a->btgt,
23148 +                                        a->dst_h_parent, isdir);
23149 +       }
23150 +
23151 +out:
23152 +       if (unlikely(err == -ENOENT || err == -EEXIST))
23153 +               err = -EIO;
23154 +       AuTraceErr(err);
23155 +       return err;
23156 +}
23157 +
23158 +/* ---------------------------------------------------------------------- */
23159 +
23160 +/*
23161 + * locking order
23162 + * (VFS)
23163 + * - src_dir and dir by lock_rename()
23164 + * - inode if exists
23165 + * (aufs)
23166 + * - lock all
23167 + *   + src_dentry and dentry by aufs_read_and_write_lock2() which calls,
23168 + *     + si_read_lock
23169 + *     + di_write_lock2_child()
23170 + *       + di_write_lock_child()
23171 + *        + ii_write_lock_child()
23172 + *       + di_write_lock_child2()
23173 + *        + ii_write_lock_child2()
23174 + *     + src_parent and parent
23175 + *       + di_write_lock_parent()
23176 + *        + ii_write_lock_parent()
23177 + *       + di_write_lock_parent2()
23178 + *        + ii_write_lock_parent2()
23179 + *   + lower src_dir and dir by vfsub_lock_rename()
23180 + *   + verify the every relationships between child and parent. if any
23181 + *     of them failed, unlock all and return -EBUSY.
23182 + */
23183 +static void au_ren_unlock(struct au_ren_args *a)
23184 +{
23185 +       vfsub_unlock_rename(a->src_h_parent, a->src_hdir,
23186 +                           a->dst_h_parent, a->dst_hdir);
23187 +       if (au_ftest_ren(a->auren_flags, DIRREN)
23188 +           && a->h_root)
23189 +               au_hn_inode_unlock(a->h_root);
23190 +       if (au_ftest_ren(a->auren_flags, MNT_WRITE))
23191 +               vfsub_mnt_drop_write(au_br_mnt(a->br));
23192 +}
23193 +
23194 +static int au_ren_lock(struct au_ren_args *a)
23195 +{
23196 +       int err;
23197 +       unsigned int udba;
23198 +
23199 +       err = 0;
23200 +       a->src_h_parent = au_h_dptr(a->src_parent, a->btgt);
23201 +       a->src_hdir = au_hi(a->src_dir, a->btgt);
23202 +       a->dst_h_parent = au_h_dptr(a->dst_parent, a->btgt);
23203 +       a->dst_hdir = au_hi(a->dst_dir, a->btgt);
23204 +
23205 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
23206 +       if (unlikely(err))
23207 +               goto out;
23208 +       au_fset_ren(a->auren_flags, MNT_WRITE);
23209 +       if (au_ftest_ren(a->auren_flags, DIRREN)) {
23210 +               struct dentry *root;
23211 +               struct inode *dir;
23212 +
23213 +               /*
23214 +                * sbinfo is already locked, so this ii_read_lock is
23215 +                * unnecessary. but our debugging feature checks it.
23216 +                */
23217 +               root = a->src_inode->i_sb->s_root;
23218 +               if (root != a->src_parent && root != a->dst_parent) {
23219 +                       dir = d_inode(root);
23220 +                       ii_read_lock_parent3(dir);
23221 +                       a->h_root = au_hi(dir, a->btgt);
23222 +                       ii_read_unlock(dir);
23223 +                       au_hn_inode_lock_nested(a->h_root, AuLsc_I_PARENT3);
23224 +               }
23225 +       }
23226 +       a->h_trap = vfsub_lock_rename(a->src_h_parent, a->src_hdir,
23227 +                                     a->dst_h_parent, a->dst_hdir);
23228 +       udba = au_opt_udba(a->src_dentry->d_sb);
23229 +       if (unlikely(a->src_hdir->hi_inode != d_inode(a->src_h_parent)
23230 +                    || a->dst_hdir->hi_inode != d_inode(a->dst_h_parent)))
23231 +               err = au_busy_or_stale();
23232 +       if (!err && au_dbtop(a->src_dentry) == a->btgt)
23233 +               err = au_h_verify(a->src_h_dentry, udba,
23234 +                                 d_inode(a->src_h_parent), a->src_h_parent,
23235 +                                 a->br);
23236 +       if (!err && au_dbtop(a->dst_dentry) == a->btgt)
23237 +               err = au_h_verify(a->dst_h_dentry, udba,
23238 +                                 d_inode(a->dst_h_parent), a->dst_h_parent,
23239 +                                 a->br);
23240 +       if (!err)
23241 +               goto out; /* success */
23242 +
23243 +       err = au_busy_or_stale();
23244 +       au_ren_unlock(a);
23245 +
23246 +out:
23247 +       return err;
23248 +}
23249 +
23250 +/* ---------------------------------------------------------------------- */
23251 +
23252 +static void au_ren_refresh_dir(struct au_ren_args *a)
23253 +{
23254 +       struct inode *dir;
23255 +
23256 +       dir = a->dst_dir;
23257 +       inode_inc_iversion(dir);
23258 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)) {
23259 +               /* is this updating defined in POSIX? */
23260 +               au_cpup_attr_timesizes(a->src_inode);
23261 +               au_cpup_attr_nlink(dir, /*force*/1);
23262 +       }
23263 +       au_dir_ts(dir, a->btgt);
23264 +
23265 +       if (a->exchange) {
23266 +               dir = a->src_dir;
23267 +               inode_inc_iversion(dir);
23268 +               if (au_ftest_ren(a->auren_flags, ISDIR_DST)) {
23269 +                       /* is this updating defined in POSIX? */
23270 +                       au_cpup_attr_timesizes(a->dst_inode);
23271 +                       au_cpup_attr_nlink(dir, /*force*/1);
23272 +               }
23273 +               au_dir_ts(dir, a->btgt);
23274 +       }
23275 +
23276 +       if (au_ftest_ren(a->auren_flags, ISSAMEDIR))
23277 +               return;
23278 +
23279 +       dir = a->src_dir;
23280 +       inode_inc_iversion(dir);
23281 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC))
23282 +               au_cpup_attr_nlink(dir, /*force*/1);
23283 +       au_dir_ts(dir, a->btgt);
23284 +}
23285 +
23286 +static void au_ren_refresh(struct au_ren_args *a)
23287 +{
23288 +       aufs_bindex_t bbot, bindex;
23289 +       struct dentry *d, *h_d;
23290 +       struct inode *i, *h_i;
23291 +       struct super_block *sb;
23292 +
23293 +       d = a->dst_dentry;
23294 +       d_drop(d);
23295 +       if (a->h_dst)
23296 +               /* already dget-ed by au_ren_or_cpup() */
23297 +               au_set_h_dptr(d, a->btgt, a->h_dst);
23298 +
23299 +       i = a->dst_inode;
23300 +       if (i) {
23301 +               if (!a->exchange) {
23302 +                       if (!au_ftest_ren(a->auren_flags, ISDIR_DST))
23303 +                               vfsub_drop_nlink(i);
23304 +                       else {
23305 +                               vfsub_dead_dir(i);
23306 +                               au_cpup_attr_timesizes(i);
23307 +                       }
23308 +                       au_update_dbrange(d, /*do_put_zero*/1);
23309 +               } else
23310 +                       au_cpup_attr_nlink(i, /*force*/1);
23311 +       } else {
23312 +               bbot = a->btgt;
23313 +               for (bindex = au_dbtop(d); bindex < bbot; bindex++)
23314 +                       au_set_h_dptr(d, bindex, NULL);
23315 +               bbot = au_dbbot(d);
23316 +               for (bindex = a->btgt + 1; bindex <= bbot; bindex++)
23317 +                       au_set_h_dptr(d, bindex, NULL);
23318 +               au_update_dbrange(d, /*do_put_zero*/0);
23319 +       }
23320 +
23321 +       if (a->exchange
23322 +           || au_ftest_ren(a->auren_flags, DIRREN)) {
23323 +               d_drop(a->src_dentry);
23324 +               if (au_ftest_ren(a->auren_flags, DIRREN))
23325 +                       au_set_dbwh(a->src_dentry, -1);
23326 +               return;
23327 +       }
23328 +
23329 +       d = a->src_dentry;
23330 +       au_set_dbwh(d, -1);
23331 +       bbot = au_dbbot(d);
23332 +       for (bindex = a->btgt + 1; bindex <= bbot; bindex++) {
23333 +               h_d = au_h_dptr(d, bindex);
23334 +               if (h_d)
23335 +                       au_set_h_dptr(d, bindex, NULL);
23336 +       }
23337 +       au_set_dbbot(d, a->btgt);
23338 +
23339 +       sb = d->d_sb;
23340 +       i = a->src_inode;
23341 +       if (au_opt_test(au_mntflags(sb), PLINK) && au_plink_test(i))
23342 +               return; /* success */
23343 +
23344 +       bbot = au_ibbot(i);
23345 +       for (bindex = a->btgt + 1; bindex <= bbot; bindex++) {
23346 +               h_i = au_h_iptr(i, bindex);
23347 +               if (h_i) {
23348 +                       au_xino_write(sb, bindex, h_i->i_ino, /*ino*/0);
23349 +                       /* ignore this error */
23350 +                       au_set_h_iptr(i, bindex, NULL, 0);
23351 +               }
23352 +       }
23353 +       au_set_ibbot(i, a->btgt);
23354 +}
23355 +
23356 +/* ---------------------------------------------------------------------- */
23357 +
23358 +/* mainly for link(2) and rename(2) */
23359 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt)
23360 +{
23361 +       aufs_bindex_t bdiropq, bwh;
23362 +       struct dentry *parent;
23363 +       struct au_branch *br;
23364 +
23365 +       parent = dentry->d_parent;
23366 +       IMustLock(d_inode(parent)); /* dir is locked */
23367 +
23368 +       bdiropq = au_dbdiropq(parent);
23369 +       bwh = au_dbwh(dentry);
23370 +       br = au_sbr(dentry->d_sb, btgt);
23371 +       if (au_br_rdonly(br)
23372 +           || (0 <= bdiropq && bdiropq < btgt)
23373 +           || (0 <= bwh && bwh < btgt))
23374 +               btgt = -1;
23375 +
23376 +       AuDbg("btgt %d\n", btgt);
23377 +       return btgt;
23378 +}
23379 +
23380 +/* sets src_btop, dst_btop and btgt */
23381 +static int au_ren_wbr(struct au_ren_args *a)
23382 +{
23383 +       int err;
23384 +       struct au_wr_dir_args wr_dir_args = {
23385 +               /* .force_btgt  = -1, */
23386 +               .flags          = AuWrDir_ADD_ENTRY
23387 +       };
23388 +
23389 +       a->src_btop = au_dbtop(a->src_dentry);
23390 +       a->dst_btop = au_dbtop(a->dst_dentry);
23391 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)
23392 +           || au_ftest_ren(a->auren_flags, ISDIR_DST))
23393 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
23394 +       wr_dir_args.force_btgt = a->src_btop;
23395 +       if (a->dst_inode && a->dst_btop < a->src_btop)
23396 +               wr_dir_args.force_btgt = a->dst_btop;
23397 +       wr_dir_args.force_btgt = au_wbr(a->dst_dentry, wr_dir_args.force_btgt);
23398 +       err = au_wr_dir(a->dst_dentry, a->src_dentry, &wr_dir_args);
23399 +       a->btgt = err;
23400 +       if (a->exchange)
23401 +               au_update_dbtop(a->dst_dentry);
23402 +
23403 +       return err;
23404 +}
23405 +
23406 +static void au_ren_dt(struct au_ren_args *a)
23407 +{
23408 +       a->h_path.dentry = a->src_h_parent;
23409 +       au_dtime_store(a->src_dt + AuPARENT, a->src_parent, &a->h_path);
23410 +       if (!au_ftest_ren(a->auren_flags, ISSAMEDIR)) {
23411 +               a->h_path.dentry = a->dst_h_parent;
23412 +               au_dtime_store(a->dst_dt + AuPARENT, a->dst_parent, &a->h_path);
23413 +       }
23414 +
23415 +       au_fclr_ren(a->auren_flags, DT_DSTDIR);
23416 +       if (!au_ftest_ren(a->auren_flags, ISDIR_SRC)
23417 +           && !a->exchange)
23418 +               return;
23419 +
23420 +       a->h_path.dentry = a->src_h_dentry;
23421 +       au_dtime_store(a->src_dt + AuCHILD, a->src_dentry, &a->h_path);
23422 +       if (d_is_positive(a->dst_h_dentry)) {
23423 +               au_fset_ren(a->auren_flags, DT_DSTDIR);
23424 +               a->h_path.dentry = a->dst_h_dentry;
23425 +               au_dtime_store(a->dst_dt + AuCHILD, a->dst_dentry, &a->h_path);
23426 +       }
23427 +}
23428 +
23429 +static void au_ren_rev_dt(int err, struct au_ren_args *a)
23430 +{
23431 +       struct dentry *h_d;
23432 +       struct inode *h_inode;
23433 +
23434 +       au_dtime_revert(a->src_dt + AuPARENT);
23435 +       if (!au_ftest_ren(a->auren_flags, ISSAMEDIR))
23436 +               au_dtime_revert(a->dst_dt + AuPARENT);
23437 +
23438 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC) && err != -EIO) {
23439 +               h_d = a->src_dt[AuCHILD].dt_h_path.dentry;
23440 +               h_inode = d_inode(h_d);
23441 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
23442 +               au_dtime_revert(a->src_dt + AuCHILD);
23443 +               inode_unlock(h_inode);
23444 +
23445 +               if (au_ftest_ren(a->auren_flags, DT_DSTDIR)) {
23446 +                       h_d = a->dst_dt[AuCHILD].dt_h_path.dentry;
23447 +                       h_inode = d_inode(h_d);
23448 +                       inode_lock_nested(h_inode, AuLsc_I_CHILD);
23449 +                       au_dtime_revert(a->dst_dt + AuCHILD);
23450 +                       inode_unlock(h_inode);
23451 +               }
23452 +       }
23453 +}
23454 +
23455 +/* ---------------------------------------------------------------------- */
23456 +
23457 +int aufs_rename(struct user_namespace *userns,
23458 +               struct inode *_src_dir, struct dentry *_src_dentry,
23459 +               struct inode *_dst_dir, struct dentry *_dst_dentry,
23460 +               unsigned int _flags)
23461 +{
23462 +       int err, lock_flags;
23463 +       void *rev;
23464 +       /* reduce stack space */
23465 +       struct au_ren_args *a;
23466 +       struct au_pin pin;
23467 +
23468 +       AuDbg("%pd, %pd, 0x%x\n", _src_dentry, _dst_dentry, _flags);
23469 +       IMustLock(_src_dir);
23470 +       IMustLock(_dst_dir);
23471 +
23472 +       err = -EINVAL;
23473 +       if (unlikely(_flags & RENAME_WHITEOUT))
23474 +               goto out;
23475 +
23476 +       err = -ENOMEM;
23477 +       BUILD_BUG_ON(sizeof(*a) > PAGE_SIZE);
23478 +       a = kzalloc(sizeof(*a), GFP_NOFS);
23479 +       if (unlikely(!a))
23480 +               goto out;
23481 +
23482 +       a->flags = _flags;
23483 +       BUILD_BUG_ON(sizeof(a->exchange) == sizeof(u8)
23484 +                    && RENAME_EXCHANGE > U8_MAX);
23485 +       a->exchange = _flags & RENAME_EXCHANGE;
23486 +       a->src_dir = _src_dir;
23487 +       a->src_dentry = _src_dentry;
23488 +       a->src_inode = NULL;
23489 +       if (d_really_is_positive(a->src_dentry))
23490 +               a->src_inode = d_inode(a->src_dentry);
23491 +       a->src_parent = a->src_dentry->d_parent; /* dir inode is locked */
23492 +       a->dst_dir = _dst_dir;
23493 +       a->dst_dentry = _dst_dentry;
23494 +       a->dst_inode = NULL;
23495 +       if (d_really_is_positive(a->dst_dentry))
23496 +               a->dst_inode = d_inode(a->dst_dentry);
23497 +       a->dst_parent = a->dst_dentry->d_parent; /* dir inode is locked */
23498 +       if (a->dst_inode) {
23499 +               /*
23500 +                * if EXCHANGE && src is non-dir && dst is dir,
23501 +                * dst is not locked.
23502 +                */
23503 +               /* IMustLock(a->dst_inode); */
23504 +               au_igrab(a->dst_inode);
23505 +       }
23506 +
23507 +       err = -ENOTDIR;
23508 +       lock_flags = AuLock_FLUSH | AuLock_NOPLM | AuLock_GEN;
23509 +       if (d_is_dir(a->src_dentry)) {
23510 +               au_fset_ren(a->auren_flags, ISDIR_SRC);
23511 +               if (unlikely(!a->exchange
23512 +                            && d_really_is_positive(a->dst_dentry)
23513 +                            && !d_is_dir(a->dst_dentry)))
23514 +                       goto out_free;
23515 +               lock_flags |= AuLock_DIRS;
23516 +       }
23517 +       if (a->dst_inode && d_is_dir(a->dst_dentry)) {
23518 +               au_fset_ren(a->auren_flags, ISDIR_DST);
23519 +               if (unlikely(!a->exchange
23520 +                            && d_really_is_positive(a->src_dentry)
23521 +                            && !d_is_dir(a->src_dentry)))
23522 +                       goto out_free;
23523 +               lock_flags |= AuLock_DIRS;
23524 +       }
23525 +       err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry,
23526 +                                       lock_flags);
23527 +       if (unlikely(err))
23528 +               goto out_free;
23529 +
23530 +       err = au_d_hashed_positive(a->src_dentry);
23531 +       if (unlikely(err))
23532 +               goto out_unlock;
23533 +       err = -ENOENT;
23534 +       if (a->dst_inode) {
23535 +               /*
23536 +                * If it is a dir, VFS unhash it before this
23537 +                * function. It means we cannot rely upon d_unhashed().
23538 +                */
23539 +               if (unlikely(!a->dst_inode->i_nlink))
23540 +                       goto out_unlock;
23541 +               if (!au_ftest_ren(a->auren_flags, ISDIR_DST)) {
23542 +                       err = au_d_hashed_positive(a->dst_dentry);
23543 +                       if (unlikely(err && !a->exchange))
23544 +                               goto out_unlock;
23545 +               } else if (unlikely(IS_DEADDIR(a->dst_inode)))
23546 +                       goto out_unlock;
23547 +       } else if (unlikely(d_unhashed(a->dst_dentry)))
23548 +               goto out_unlock;
23549 +
23550 +       /*
23551 +        * is it possible?
23552 +        * yes, it happened (in linux-3.3-rcN) but I don't know why.
23553 +        * there may exist a problem somewhere else.
23554 +        */
23555 +       err = -EINVAL;
23556 +       if (unlikely(d_inode(a->dst_parent) == d_inode(a->src_dentry)))
23557 +               goto out_unlock;
23558 +
23559 +       au_fset_ren(a->auren_flags, ISSAMEDIR); /* temporary */
23560 +       di_write_lock_parent(a->dst_parent);
23561 +
23562 +       /* which branch we process */
23563 +       err = au_ren_wbr(a);
23564 +       if (unlikely(err < 0))
23565 +               goto out_parent;
23566 +       a->br = au_sbr(a->dst_dentry->d_sb, a->btgt);
23567 +       a->h_path.mnt = au_br_mnt(a->br);
23568 +
23569 +       /* are they available to be renamed */
23570 +       err = au_ren_may_dir(a);
23571 +       if (unlikely(err))
23572 +               goto out_children;
23573 +
23574 +       /* prepare the writable parent dir on the same branch */
23575 +       if (a->dst_btop == a->btgt) {
23576 +               au_fset_ren(a->auren_flags, WHDST);
23577 +       } else {
23578 +               err = au_cpup_dirs(a->dst_dentry, a->btgt);
23579 +               if (unlikely(err))
23580 +                       goto out_children;
23581 +       }
23582 +
23583 +       err = 0;
23584 +       if (!a->exchange) {
23585 +               if (a->src_dir != a->dst_dir) {
23586 +                       /*
23587 +                        * this temporary unlock is safe,
23588 +                        * because both dir->i_mutex are locked.
23589 +                        */
23590 +                       di_write_unlock(a->dst_parent);
23591 +                       di_write_lock_parent(a->src_parent);
23592 +                       err = au_wr_dir_need_wh(a->src_dentry,
23593 +                                               au_ftest_ren(a->auren_flags,
23594 +                                                            ISDIR_SRC),
23595 +                                               &a->btgt);
23596 +                       di_write_unlock(a->src_parent);
23597 +                       di_write_lock2_parent(a->src_parent, a->dst_parent,
23598 +                                             /*isdir*/1);
23599 +                       au_fclr_ren(a->auren_flags, ISSAMEDIR);
23600 +               } else
23601 +                       err = au_wr_dir_need_wh(a->src_dentry,
23602 +                                               au_ftest_ren(a->auren_flags,
23603 +                                                            ISDIR_SRC),
23604 +                                               &a->btgt);
23605 +       }
23606 +       if (unlikely(err < 0))
23607 +               goto out_children;
23608 +       if (err)
23609 +               au_fset_ren(a->auren_flags, WHSRC);
23610 +
23611 +       /* cpup src */
23612 +       if (a->src_btop != a->btgt) {
23613 +               err = au_pin(&pin, a->src_dentry, a->btgt,
23614 +                            au_opt_udba(a->src_dentry->d_sb),
23615 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
23616 +               if (!err) {
23617 +                       struct au_cp_generic cpg = {
23618 +                               .dentry = a->src_dentry,
23619 +                               .bdst   = a->btgt,
23620 +                               .bsrc   = a->src_btop,
23621 +                               .len    = -1,
23622 +                               .pin    = &pin,
23623 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
23624 +                       };
23625 +                       AuDebugOn(au_dbtop(a->src_dentry) != a->src_btop);
23626 +                       err = au_sio_cpup_simple(&cpg);
23627 +                       au_unpin(&pin);
23628 +               }
23629 +               if (unlikely(err))
23630 +                       goto out_children;
23631 +               a->src_btop = a->btgt;
23632 +               a->src_h_dentry = au_h_dptr(a->src_dentry, a->btgt);
23633 +               if (!a->exchange)
23634 +                       au_fset_ren(a->auren_flags, WHSRC);
23635 +       }
23636 +
23637 +       /* cpup dst */
23638 +       if (a->exchange && a->dst_inode
23639 +           && a->dst_btop != a->btgt) {
23640 +               err = au_pin(&pin, a->dst_dentry, a->btgt,
23641 +                            au_opt_udba(a->dst_dentry->d_sb),
23642 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
23643 +               if (!err) {
23644 +                       struct au_cp_generic cpg = {
23645 +                               .dentry = a->dst_dentry,
23646 +                               .bdst   = a->btgt,
23647 +                               .bsrc   = a->dst_btop,
23648 +                               .len    = -1,
23649 +                               .pin    = &pin,
23650 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
23651 +                       };
23652 +                       err = au_sio_cpup_simple(&cpg);
23653 +                       au_unpin(&pin);
23654 +               }
23655 +               if (unlikely(err))
23656 +                       goto out_children;
23657 +               a->dst_btop = a->btgt;
23658 +               a->dst_h_dentry = au_h_dptr(a->dst_dentry, a->btgt);
23659 +       }
23660 +
23661 +       /* lock them all */
23662 +       err = au_ren_lock(a);
23663 +       if (unlikely(err))
23664 +               /* leave the copied-up one */
23665 +               goto out_children;
23666 +
23667 +       if (!a->exchange) {
23668 +               if (!au_opt_test(au_mntflags(a->dst_dir->i_sb), UDBA_NONE))
23669 +                       err = au_may_ren(a);
23670 +               else if (unlikely(a->dst_dentry->d_name.len > AUFS_MAX_NAMELEN))
23671 +                       err = -ENAMETOOLONG;
23672 +               if (unlikely(err))
23673 +                       goto out_hdir;
23674 +       }
23675 +
23676 +       /* store timestamps to be revertible */
23677 +       au_ren_dt(a);
23678 +
23679 +       /* store dirren info */
23680 +       if (au_ftest_ren(a->auren_flags, DIRREN)) {
23681 +               err = au_dr_rename(a->src_dentry, a->btgt,
23682 +                                  &a->dst_dentry->d_name, &rev);
23683 +               AuTraceErr(err);
23684 +               if (unlikely(err))
23685 +                       goto out_dt;
23686 +       }
23687 +
23688 +       /* here we go */
23689 +       err = do_rename(a);
23690 +       if (unlikely(err))
23691 +               goto out_dirren;
23692 +
23693 +       if (au_ftest_ren(a->auren_flags, DIRREN))
23694 +               au_dr_rename_fin(a->src_dentry, a->btgt, rev);
23695 +
23696 +       /* update dir attributes */
23697 +       au_ren_refresh_dir(a);
23698 +
23699 +       /* dput/iput all lower dentries */
23700 +       au_ren_refresh(a);
23701 +
23702 +       goto out_hdir; /* success */
23703 +
23704 +out_dirren:
23705 +       if (au_ftest_ren(a->auren_flags, DIRREN))
23706 +               au_dr_rename_rev(a->src_dentry, a->btgt, rev);
23707 +out_dt:
23708 +       au_ren_rev_dt(err, a);
23709 +out_hdir:
23710 +       au_ren_unlock(a);
23711 +out_children:
23712 +       au_nhash_wh_free(&a->whlist);
23713 +       if (err && a->dst_inode && a->dst_btop != a->btgt) {
23714 +               AuDbg("btop %d, btgt %d\n", a->dst_btop, a->btgt);
23715 +               au_set_h_dptr(a->dst_dentry, a->btgt, NULL);
23716 +               au_set_dbtop(a->dst_dentry, a->dst_btop);
23717 +       }
23718 +out_parent:
23719 +       if (!err) {
23720 +               if (d_unhashed(a->src_dentry))
23721 +                       au_fset_ren(a->auren_flags, DROPPED_SRC);
23722 +               if (d_unhashed(a->dst_dentry))
23723 +                       au_fset_ren(a->auren_flags, DROPPED_DST);
23724 +               if (!a->exchange)
23725 +                       d_move(a->src_dentry, a->dst_dentry);
23726 +               else {
23727 +                       d_exchange(a->src_dentry, a->dst_dentry);
23728 +                       if (au_ftest_ren(a->auren_flags, DROPPED_DST))
23729 +                               d_drop(a->dst_dentry);
23730 +               }
23731 +               if (au_ftest_ren(a->auren_flags, DROPPED_SRC))
23732 +                       d_drop(a->src_dentry);
23733 +       } else {
23734 +               au_update_dbtop(a->dst_dentry);
23735 +               if (!a->dst_inode)
23736 +                       d_drop(a->dst_dentry);
23737 +       }
23738 +       if (au_ftest_ren(a->auren_flags, ISSAMEDIR))
23739 +               di_write_unlock(a->dst_parent);
23740 +       else
23741 +               di_write_unlock2(a->src_parent, a->dst_parent);
23742 +out_unlock:
23743 +       aufs_read_and_write_unlock2(a->dst_dentry, a->src_dentry);
23744 +out_free:
23745 +       iput(a->dst_inode);
23746 +       if (a->thargs)
23747 +               au_whtmp_rmdir_free(a->thargs);
23748 +       au_kfree_rcu(a);
23749 +out:
23750 +       AuTraceErr(err);
23751 +       return err;
23752 +}
23753 diff -urN /usr/share/empty/fs/aufs/Kconfig linux/fs/aufs/Kconfig
23754 --- /usr/share/empty/fs/aufs/Kconfig    1970-01-01 01:00:00.000000000 +0100
23755 +++ linux/fs/aufs/Kconfig       2021-12-03 15:38:59.929980643 +0100
23756 @@ -0,0 +1,199 @@
23757 +# SPDX-License-Identifier: GPL-2.0
23758 +config AUFS_FS
23759 +       tristate "Aufs (Advanced multi layered unification filesystem) support"
23760 +       help
23761 +       Aufs is a stackable unification filesystem such as Unionfs,
23762 +       which unifies several directories and provides a merged single
23763 +       directory.
23764 +       In the early days, aufs was entirely re-designed and
23765 +       re-implemented Unionfs Version 1.x series. Introducing many
23766 +       original ideas, approaches and improvements, it becomes totally
23767 +       different from Unionfs while keeping the basic features.
23768 +
23769 +if AUFS_FS
23770 +choice
23771 +       prompt "Maximum number of branches"
23772 +       default AUFS_BRANCH_MAX_127
23773 +       help
23774 +       Specifies the maximum number of branches (or member directories)
23775 +       in a single aufs. The larger value consumes more system
23776 +       resources and has a minor impact to performance.
23777 +config AUFS_BRANCH_MAX_127
23778 +       bool "127"
23779 +       help
23780 +       Specifies the maximum number of branches (or member directories)
23781 +       in a single aufs. The larger value consumes more system
23782 +       resources and has a minor impact to performance.
23783 +config AUFS_BRANCH_MAX_511
23784 +       bool "511"
23785 +       help
23786 +       Specifies the maximum number of branches (or member directories)
23787 +       in a single aufs. The larger value consumes more system
23788 +       resources and has a minor impact to performance.
23789 +config AUFS_BRANCH_MAX_1023
23790 +       bool "1023"
23791 +       help
23792 +       Specifies the maximum number of branches (or member directories)
23793 +       in a single aufs. The larger value consumes more system
23794 +       resources and has a minor impact to performance.
23795 +config AUFS_BRANCH_MAX_32767
23796 +       bool "32767"
23797 +       help
23798 +       Specifies the maximum number of branches (or member directories)
23799 +       in a single aufs. The larger value consumes more system
23800 +       resources and has a minor impact to performance.
23801 +endchoice
23802 +
23803 +config AUFS_SBILIST
23804 +       bool
23805 +       depends on AUFS_MAGIC_SYSRQ || PROC_FS
23806 +       default y
23807 +       help
23808 +       Automatic configuration for internal use.
23809 +       When aufs supports Magic SysRq or /proc, enabled automatically.
23810 +
23811 +config AUFS_HNOTIFY
23812 +       bool "Detect direct branch access (bypassing aufs)"
23813 +       help
23814 +       If you want to modify files on branches directly, eg. bypassing aufs,
23815 +       and want aufs to detect the changes of them fully, then enable this
23816 +       option and use 'udba=notify' mount option.
23817 +       Currently there is only one available configuration, "fsnotify".
23818 +       It will have a negative impact to the performance.
23819 +       See detail in aufs.5.
23820 +
23821 +choice
23822 +       prompt "method" if AUFS_HNOTIFY
23823 +       default AUFS_HFSNOTIFY
23824 +config AUFS_HFSNOTIFY
23825 +       bool "fsnotify"
23826 +       select FSNOTIFY
23827 +endchoice
23828 +
23829 +config AUFS_EXPORT
23830 +       bool "NFS-exportable aufs"
23831 +       depends on EXPORTFS
23832 +       help
23833 +       If you want to export your mounted aufs via NFS, then enable this
23834 +       option. There are several requirements for this configuration.
23835 +       See detail in aufs.5.
23836 +
23837 +config AUFS_INO_T_64
23838 +       bool
23839 +       depends on AUFS_EXPORT
23840 +       depends on 64BIT && !(ALPHA || S390)
23841 +       default y
23842 +       help
23843 +       Automatic configuration for internal use.
23844 +       /* typedef unsigned long/int __kernel_ino_t */
23845 +       /* alpha and s390x are int */
23846 +
23847 +config AUFS_XATTR
23848 +       bool "support for XATTR/EA (including Security Labels)"
23849 +       help
23850 +       If your branch fs supports XATTR/EA and you want to make them
23851 +       available in aufs too, then enable this opsion and specify the
23852 +       branch attributes for EA.
23853 +       See detail in aufs.5.
23854 +
23855 +config AUFS_FHSM
23856 +       bool "File-based Hierarchical Storage Management"
23857 +       help
23858 +       Hierarchical Storage Management (or HSM) is a well-known feature
23859 +       in the storage world. Aufs provides this feature as file-based.
23860 +       with multiple branches.
23861 +       These multiple branches are prioritized, ie. the topmost one
23862 +       should be the fastest drive and be used heavily.
23863 +
23864 +config AUFS_RDU
23865 +       bool "Readdir in userspace"
23866 +       help
23867 +       Aufs has two methods to provide a merged view for a directory,
23868 +       by a user-space library and by kernel-space natively. The latter
23869 +       is always enabled but sometimes large and slow.
23870 +       If you enable this option, install the library in aufs2-util
23871 +       package, and set some environment variables for your readdir(3),
23872 +       then the work will be handled in user-space which generally
23873 +       shows better performance in most cases.
23874 +       See detail in aufs.5.
23875 +
23876 +config AUFS_DIRREN
23877 +       bool "Workaround for rename(2)-ing a directory"
23878 +       help
23879 +       By default, aufs returns EXDEV error in renameing a dir who has
23880 +       his child on the lower branch, since it is a bad idea to issue
23881 +       rename(2) internally for every lower branch. But user may not
23882 +       accept this behaviour. So here is a workaround to allow such
23883 +       rename(2) and store some extra infromation on the writable
23884 +       branch. Obviously this costs high (and I don't like it).
23885 +       To use this feature, you need to enable this configuration AND
23886 +       to specify the mount option `dirren.'
23887 +       See details in aufs.5 and the design documents.
23888 +
23889 +config AUFS_SHWH
23890 +       bool "Show whiteouts"
23891 +       help
23892 +       If you want to make the whiteouts in aufs visible, then enable
23893 +       this option and specify 'shwh' mount option. Although it may
23894 +       sounds like philosophy or something, but in technically it
23895 +       simply shows the name of whiteout with keeping its behaviour.
23896 +
23897 +config AUFS_BR_RAMFS
23898 +       bool "Ramfs (initramfs/rootfs) as an aufs branch"
23899 +       help
23900 +       If you want to use ramfs as an aufs branch fs, then enable this
23901 +       option. Generally tmpfs is recommended.
23902 +       Aufs prohibited them to be a branch fs by default, because
23903 +       initramfs becomes unusable after switch_root or something
23904 +       generally. If you sets initramfs as an aufs branch and boot your
23905 +       system by switch_root, you will meet a problem easily since the
23906 +       files in initramfs may be inaccessible.
23907 +       Unless you are going to use ramfs as an aufs branch fs without
23908 +       switch_root or something, leave it N.
23909 +
23910 +config AUFS_BR_FUSE
23911 +       bool "Fuse fs as an aufs branch"
23912 +       depends on FUSE_FS
23913 +       select AUFS_POLL
23914 +       help
23915 +       If you want to use fuse-based userspace filesystem as an aufs
23916 +       branch fs, then enable this option.
23917 +       It implements the internal poll(2) operation which is
23918 +       implemented by fuse only (curretnly).
23919 +
23920 +config AUFS_POLL
23921 +       bool
23922 +       help
23923 +       Automatic configuration for internal use.
23924 +
23925 +config AUFS_BR_HFSPLUS
23926 +       bool "Hfsplus as an aufs branch"
23927 +       depends on HFSPLUS_FS
23928 +       default y
23929 +       help
23930 +       If you want to use hfsplus fs as an aufs branch fs, then enable
23931 +       this option. This option introduces a small overhead at
23932 +       copying-up a file on hfsplus.
23933 +
23934 +config AUFS_BDEV_LOOP
23935 +       bool
23936 +       depends on BLK_DEV_LOOP
23937 +       default y
23938 +       help
23939 +       Automatic configuration for internal use.
23940 +       Convert =[ym] into =y.
23941 +
23942 +config AUFS_DEBUG
23943 +       bool "Debug aufs"
23944 +       help
23945 +       Enable this to compile aufs internal debug code.
23946 +       It will have a negative impact to the performance.
23947 +
23948 +config AUFS_MAGIC_SYSRQ
23949 +       bool
23950 +       depends on AUFS_DEBUG && MAGIC_SYSRQ
23951 +       default y
23952 +       help
23953 +       Automatic configuration for internal use.
23954 +       When aufs supports Magic SysRq, enabled automatically.
23955 +endif
23956 diff -urN /usr/share/empty/fs/aufs/lcnt.h linux/fs/aufs/lcnt.h
23957 --- /usr/share/empty/fs/aufs/lcnt.h     1970-01-01 01:00:00.000000000 +0100
23958 +++ linux/fs/aufs/lcnt.h        2021-12-03 15:38:59.936647310 +0100
23959 @@ -0,0 +1,186 @@
23960 +/* SPDX-License-Identifier: GPL-2.0 */
23961 +/*
23962 + * Copyright (C) 2018-2021 Junjiro R. Okajima
23963 + *
23964 + * This program, aufs is free software; you can redistribute it and/or modify
23965 + * it under the terms of the GNU General Public License as published by
23966 + * the Free Software Foundation; either version 2 of the License, or
23967 + * (at your option) any later version.
23968 + *
23969 + * This program is distributed in the hope that it will be useful,
23970 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
23971 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
23972 + * GNU General Public License for more details.
23973 + *
23974 + * You should have received a copy of the GNU General Public License
23975 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
23976 + */
23977 +
23978 +/*
23979 + * simple long counter wrapper
23980 + */
23981 +
23982 +#ifndef __AUFS_LCNT_H__
23983 +#define __AUFS_LCNT_H__
23984 +
23985 +#ifdef __KERNEL__
23986 +
23987 +#include "debug.h"
23988 +
23989 +#define AuLCntATOMIC   1
23990 +#define AuLCntPCPUCNT  2
23991 +/*
23992 + * why does percpu_refcount require extra synchronize_rcu()s in
23993 + * au_br_do_free()
23994 + */
23995 +#define AuLCntPCPUREF  3
23996 +
23997 +/* #define AuLCntChosen        AuLCntATOMIC */
23998 +#define AuLCntChosen   AuLCntPCPUCNT
23999 +/* #define AuLCntChosen        AuLCntPCPUREF */
24000 +
24001 +#if AuLCntChosen == AuLCntATOMIC
24002 +#include <linux/atomic.h>
24003 +
24004 +typedef atomic_long_t au_lcnt_t;
24005 +
24006 +static inline int au_lcnt_init(au_lcnt_t *cnt, void *release __maybe_unused)
24007 +{
24008 +       atomic_long_set(cnt, 0);
24009 +       return 0;
24010 +}
24011 +
24012 +static inline void au_lcnt_wait_for_fin(au_lcnt_t *cnt __maybe_unused)
24013 +{
24014 +       /* empty */
24015 +}
24016 +
24017 +static inline void au_lcnt_fin(au_lcnt_t *cnt __maybe_unused,
24018 +                              int do_sync __maybe_unused)
24019 +{
24020 +       /* empty */
24021 +}
24022 +
24023 +static inline void au_lcnt_inc(au_lcnt_t *cnt)
24024 +{
24025 +       atomic_long_inc(cnt);
24026 +}
24027 +
24028 +static inline void au_lcnt_dec(au_lcnt_t *cnt)
24029 +{
24030 +       atomic_long_dec(cnt);
24031 +}
24032 +
24033 +static inline long au_lcnt_read(au_lcnt_t *cnt, int do_rev __maybe_unused)
24034 +{
24035 +       return atomic_long_read(cnt);
24036 +}
24037 +#endif
24038 +
24039 +#if AuLCntChosen == AuLCntPCPUCNT
24040 +#include <linux/percpu_counter.h>
24041 +
24042 +typedef struct percpu_counter au_lcnt_t;
24043 +
24044 +static inline int au_lcnt_init(au_lcnt_t *cnt, void *release __maybe_unused)
24045 +{
24046 +       return percpu_counter_init(cnt, 0, GFP_NOFS);
24047 +}
24048 +
24049 +static inline void au_lcnt_wait_for_fin(au_lcnt_t *cnt __maybe_unused)
24050 +{
24051 +       /* empty */
24052 +}
24053 +
24054 +static inline void au_lcnt_fin(au_lcnt_t *cnt, int do_sync __maybe_unused)
24055 +{
24056 +       percpu_counter_destroy(cnt);
24057 +}
24058 +
24059 +static inline void au_lcnt_inc(au_lcnt_t *cnt)
24060 +{
24061 +       percpu_counter_inc(cnt);
24062 +}
24063 +
24064 +static inline void au_lcnt_dec(au_lcnt_t *cnt)
24065 +{
24066 +       percpu_counter_dec(cnt);
24067 +}
24068 +
24069 +static inline long au_lcnt_read(au_lcnt_t *cnt, int do_rev __maybe_unused)
24070 +{
24071 +       s64 n;
24072 +
24073 +       n = percpu_counter_sum(cnt);
24074 +       BUG_ON(n < 0);
24075 +       if (LONG_MAX != LLONG_MAX
24076 +           && n > LONG_MAX)
24077 +               AuWarn1("%s\n", "wrap-around");
24078 +
24079 +       return n;
24080 +}
24081 +#endif
24082 +
24083 +#if AuLCntChosen == AuLCntPCPUREF
24084 +#include <linux/percpu-refcount.h>
24085 +
24086 +typedef struct percpu_ref au_lcnt_t;
24087 +
24088 +static inline int au_lcnt_init(au_lcnt_t *cnt, percpu_ref_func_t *release)
24089 +{
24090 +       if (!release)
24091 +               release = percpu_ref_exit;
24092 +       return percpu_ref_init(cnt, release, /*percpu mode*/0, GFP_NOFS);
24093 +}
24094 +
24095 +static inline void au_lcnt_wait_for_fin(au_lcnt_t *cnt __maybe_unused)
24096 +{
24097 +       synchronize_rcu();
24098 +}
24099 +
24100 +static inline void au_lcnt_fin(au_lcnt_t *cnt, int do_sync)
24101 +{
24102 +       percpu_ref_kill(cnt);
24103 +       if (do_sync)
24104 +               au_lcnt_wait_for_fin(cnt);
24105 +}
24106 +
24107 +static inline void au_lcnt_inc(au_lcnt_t *cnt)
24108 +{
24109 +       percpu_ref_get(cnt);
24110 +}
24111 +
24112 +static inline void au_lcnt_dec(au_lcnt_t *cnt)
24113 +{
24114 +       percpu_ref_put(cnt);
24115 +}
24116 +
24117 +/*
24118 + * avoid calling this func as possible.
24119 + */
24120 +static inline long au_lcnt_read(au_lcnt_t *cnt, int do_rev)
24121 +{
24122 +       long l;
24123 +
24124 +       percpu_ref_switch_to_atomic_sync(cnt);
24125 +       l = atomic_long_read(&cnt->count);
24126 +       if (do_rev)
24127 +               percpu_ref_switch_to_percpu(cnt);
24128 +
24129 +       /* percpu_ref is initialized by 1 instead of 0 */
24130 +       return l - 1;
24131 +}
24132 +#endif
24133 +
24134 +#ifdef CONFIG_AUFS_DEBUG
24135 +#define AuLCntZero(val) do {                   \
24136 +       long l = val;                           \
24137 +       if (l)                                  \
24138 +               AuDbg("%s = %ld\n", #val, l);   \
24139 +} while (0)
24140 +#else
24141 +#define AuLCntZero(val)                do {} while (0)
24142 +#endif
24143 +
24144 +#endif /* __KERNEL__ */
24145 +#endif /* __AUFS_LCNT_H__ */
24146 diff -urN /usr/share/empty/fs/aufs/loop.c linux/fs/aufs/loop.c
24147 --- /usr/share/empty/fs/aufs/loop.c     1970-01-01 01:00:00.000000000 +0100
24148 +++ linux/fs/aufs/loop.c        2021-12-03 15:38:59.936647310 +0100
24149 @@ -0,0 +1,148 @@
24150 +// SPDX-License-Identifier: GPL-2.0
24151 +/*
24152 + * Copyright (C) 2005-2021 Junjiro R. Okajima
24153 + *
24154 + * This program, aufs is free software; you can redistribute it and/or modify
24155 + * it under the terms of the GNU General Public License as published by
24156 + * the Free Software Foundation; either version 2 of the License, or
24157 + * (at your option) any later version.
24158 + *
24159 + * This program is distributed in the hope that it will be useful,
24160 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24161 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24162 + * GNU General Public License for more details.
24163 + *
24164 + * You should have received a copy of the GNU General Public License
24165 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24166 + */
24167 +
24168 +/*
24169 + * support for loopback block device as a branch
24170 + */
24171 +
24172 +#include "aufs.h"
24173 +
24174 +/* added into drivers/block/loop.c */
24175 +static struct file *(*backing_file_func)(struct super_block *sb);
24176 +
24177 +/*
24178 + * test if two lower dentries have overlapping branches.
24179 + */
24180 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding)
24181 +{
24182 +       struct super_block *h_sb;
24183 +       struct file *backing_file;
24184 +
24185 +       if (unlikely(!backing_file_func)) {
24186 +               /* don't load "loop" module here */
24187 +               backing_file_func = symbol_get(loop_backing_file);
24188 +               if (unlikely(!backing_file_func))
24189 +                       /* "loop" module is not loaded */
24190 +                       return 0;
24191 +       }
24192 +
24193 +       h_sb = h_adding->d_sb;
24194 +       backing_file = backing_file_func(h_sb);
24195 +       if (!backing_file)
24196 +               return 0;
24197 +
24198 +       h_adding = backing_file->f_path.dentry;
24199 +       /*
24200 +        * h_adding can be local NFS.
24201 +        * in this case aufs cannot detect the loop.
24202 +        */
24203 +       if (unlikely(h_adding->d_sb == sb))
24204 +               return 1;
24205 +       return !!au_test_subdir(h_adding, sb->s_root);
24206 +}
24207 +
24208 +/* true if a kernel thread named 'loop[0-9].*' accesses a file */
24209 +int au_test_loopback_kthread(void)
24210 +{
24211 +       int ret;
24212 +       struct task_struct *tsk = current;
24213 +       char c, comm[sizeof(tsk->comm)];
24214 +
24215 +       ret = 0;
24216 +       if (tsk->flags & PF_KTHREAD) {
24217 +               get_task_comm(comm, tsk);
24218 +               c = comm[4];
24219 +               ret = ('0' <= c && c <= '9'
24220 +                      && !strncmp(comm, "loop", 4));
24221 +       }
24222 +
24223 +       return ret;
24224 +}
24225 +
24226 +/* ---------------------------------------------------------------------- */
24227 +
24228 +#define au_warn_loopback_step  16
24229 +static int au_warn_loopback_nelem = au_warn_loopback_step;
24230 +static unsigned long *au_warn_loopback_array;
24231 +
24232 +void au_warn_loopback(struct super_block *h_sb)
24233 +{
24234 +       int i, new_nelem;
24235 +       unsigned long *a, magic;
24236 +       static DEFINE_SPINLOCK(spin);
24237 +
24238 +       magic = h_sb->s_magic;
24239 +       spin_lock(&spin);
24240 +       a = au_warn_loopback_array;
24241 +       for (i = 0; i < au_warn_loopback_nelem && *a; i++)
24242 +               if (a[i] == magic) {
24243 +                       spin_unlock(&spin);
24244 +                       return;
24245 +               }
24246 +
24247 +       /* h_sb is new to us, print it */
24248 +       if (i < au_warn_loopback_nelem) {
24249 +               a[i] = magic;
24250 +               goto pr;
24251 +       }
24252 +
24253 +       /* expand the array */
24254 +       new_nelem = au_warn_loopback_nelem + au_warn_loopback_step;
24255 +       a = au_kzrealloc(au_warn_loopback_array,
24256 +                        au_warn_loopback_nelem * sizeof(unsigned long),
24257 +                        new_nelem * sizeof(unsigned long), GFP_ATOMIC,
24258 +                        /*may_shrink*/0);
24259 +       if (a) {
24260 +               au_warn_loopback_nelem = new_nelem;
24261 +               au_warn_loopback_array = a;
24262 +               a[i] = magic;
24263 +               goto pr;
24264 +       }
24265 +
24266 +       spin_unlock(&spin);
24267 +       AuWarn1("realloc failed, ignored\n");
24268 +       return;
24269 +
24270 +pr:
24271 +       spin_unlock(&spin);
24272 +       pr_warn("you may want to try another patch for loopback file "
24273 +               "on %s(0x%lx) branch\n", au_sbtype(h_sb), magic);
24274 +}
24275 +
24276 +int au_loopback_init(void)
24277 +{
24278 +       int err;
24279 +       struct super_block *sb __maybe_unused;
24280 +
24281 +       BUILD_BUG_ON(sizeof(sb->s_magic) != sizeof(*au_warn_loopback_array));
24282 +
24283 +       err = 0;
24284 +       au_warn_loopback_array = kcalloc(au_warn_loopback_step,
24285 +                                        sizeof(unsigned long), GFP_NOFS);
24286 +       if (unlikely(!au_warn_loopback_array))
24287 +               err = -ENOMEM;
24288 +
24289 +       return err;
24290 +}
24291 +
24292 +void au_loopback_fin(void)
24293 +{
24294 +       if (backing_file_func)
24295 +               symbol_put(loop_backing_file);
24296 +       au_kfree_try_rcu(au_warn_loopback_array);
24297 +}
24298 diff -urN /usr/share/empty/fs/aufs/loop.h linux/fs/aufs/loop.h
24299 --- /usr/share/empty/fs/aufs/loop.h     1970-01-01 01:00:00.000000000 +0100
24300 +++ linux/fs/aufs/loop.h        2021-12-03 15:38:59.936647310 +0100
24301 @@ -0,0 +1,55 @@
24302 +/* SPDX-License-Identifier: GPL-2.0 */
24303 +/*
24304 + * Copyright (C) 2005-2021 Junjiro R. Okajima
24305 + *
24306 + * This program, aufs is free software; you can redistribute it and/or modify
24307 + * it under the terms of the GNU General Public License as published by
24308 + * the Free Software Foundation; either version 2 of the License, or
24309 + * (at your option) any later version.
24310 + *
24311 + * This program is distributed in the hope that it will be useful,
24312 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24313 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24314 + * GNU General Public License for more details.
24315 + *
24316 + * You should have received a copy of the GNU General Public License
24317 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24318 + */
24319 +
24320 +/*
24321 + * support for loopback mount as a branch
24322 + */
24323 +
24324 +#ifndef __AUFS_LOOP_H__
24325 +#define __AUFS_LOOP_H__
24326 +
24327 +#ifdef __KERNEL__
24328 +
24329 +struct dentry;
24330 +struct super_block;
24331 +
24332 +#ifdef CONFIG_AUFS_BDEV_LOOP
24333 +/* drivers/block/loop.c */
24334 +struct file *loop_backing_file(struct super_block *sb);
24335 +
24336 +/* loop.c */
24337 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding);
24338 +int au_test_loopback_kthread(void);
24339 +void au_warn_loopback(struct super_block *h_sb);
24340 +
24341 +int au_loopback_init(void);
24342 +void au_loopback_fin(void);
24343 +#else
24344 +AuStub(struct file *, loop_backing_file, return NULL, struct super_block *sb)
24345 +
24346 +AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
24347 +          struct dentry *h_adding)
24348 +AuStubInt0(au_test_loopback_kthread, void)
24349 +AuStubVoid(au_warn_loopback, struct super_block *h_sb)
24350 +
24351 +AuStubInt0(au_loopback_init, void)
24352 +AuStubVoid(au_loopback_fin, void)
24353 +#endif /* BLK_DEV_LOOP */
24354 +
24355 +#endif /* __KERNEL__ */
24356 +#endif /* __AUFS_LOOP_H__ */
24357 diff -urN /usr/share/empty/fs/aufs/magic.mk linux/fs/aufs/magic.mk
24358 --- /usr/share/empty/fs/aufs/magic.mk   1970-01-01 01:00:00.000000000 +0100
24359 +++ linux/fs/aufs/magic.mk      2021-12-03 15:38:59.936647310 +0100
24360 @@ -0,0 +1,31 @@
24361 +# SPDX-License-Identifier: GPL-2.0
24362 +
24363 +# defined in ${srctree}/fs/fuse/inode.c
24364 +# tristate
24365 +ifdef CONFIG_FUSE_FS
24366 +ccflags-y += -DFUSE_SUPER_MAGIC=0x65735546
24367 +endif
24368 +
24369 +# defined in ${srctree}/fs/xfs/xfs_sb.h
24370 +# tristate
24371 +ifdef CONFIG_XFS_FS
24372 +ccflags-y += -DXFS_SB_MAGIC=0x58465342
24373 +endif
24374 +
24375 +# defined in ${srctree}/fs/configfs/mount.c
24376 +# tristate
24377 +ifdef CONFIG_CONFIGFS_FS
24378 +ccflags-y += -DCONFIGFS_MAGIC=0x62656570
24379 +endif
24380 +
24381 +# defined in ${srctree}/fs/ubifs/ubifs.h
24382 +# tristate
24383 +ifdef CONFIG_UBIFS_FS
24384 +ccflags-y += -DUBIFS_SUPER_MAGIC=0x24051905
24385 +endif
24386 +
24387 +# defined in ${srctree}/fs/hfsplus/hfsplus_raw.h
24388 +# tristate
24389 +ifdef CONFIG_HFSPLUS_FS
24390 +ccflags-y += -DHFSPLUS_SUPER_MAGIC=0x482b
24391 +endif
24392 diff -urN /usr/share/empty/fs/aufs/Makefile linux/fs/aufs/Makefile
24393 --- /usr/share/empty/fs/aufs/Makefile   1970-01-01 01:00:00.000000000 +0100
24394 +++ linux/fs/aufs/Makefile      2021-12-03 15:38:59.929980643 +0100
24395 @@ -0,0 +1,46 @@
24396 +# SPDX-License-Identifier: GPL-2.0
24397 +
24398 +include ${src}/magic.mk
24399 +ifeq (${CONFIG_AUFS_FS},m)
24400 +include ${src}/conf.mk
24401 +endif
24402 +-include ${src}/priv_def.mk
24403 +
24404 +# cf. include/linux/kernel.h
24405 +# enable pr_debug
24406 +ccflags-y += -DDEBUG
24407 +# sparse requires the full pathname
24408 +ifdef M
24409 +ccflags-y += -include ${M}/../../include/uapi/linux/aufs_type.h
24410 +else
24411 +ccflags-y += -include ${srctree}/include/uapi/linux/aufs_type.h
24412 +endif
24413 +
24414 +obj-$(CONFIG_AUFS_FS) += aufs.o
24415 +aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
24416 +       wkq.o vfsub.o dcsub.o \
24417 +       cpup.o whout.o wbr_policy.o \
24418 +       dinfo.o dentry.o \
24419 +       dynop.o \
24420 +       finfo.o file.o f_op.o \
24421 +       dir.o vdir.o \
24422 +       iinfo.o inode.o i_op.o i_op_add.o i_op_del.o i_op_ren.o \
24423 +       mvdown.o ioctl.o
24424 +
24425 +# all are boolean
24426 +aufs-$(CONFIG_PROC_FS) += procfs.o plink.o
24427 +aufs-$(CONFIG_SYSFS) += sysfs.o
24428 +aufs-$(CONFIG_DEBUG_FS) += dbgaufs.o
24429 +aufs-$(CONFIG_AUFS_BDEV_LOOP) += loop.o
24430 +aufs-$(CONFIG_AUFS_HNOTIFY) += hnotify.o
24431 +aufs-$(CONFIG_AUFS_HFSNOTIFY) += hfsnotify.o
24432 +aufs-$(CONFIG_AUFS_EXPORT) += export.o
24433 +aufs-$(CONFIG_AUFS_XATTR) += xattr.o
24434 +aufs-$(CONFIG_FS_POSIX_ACL) += posix_acl.o
24435 +aufs-$(CONFIG_AUFS_DIRREN) += dirren.o
24436 +aufs-$(CONFIG_AUFS_FHSM) += fhsm.o
24437 +aufs-$(CONFIG_AUFS_POLL) += poll.o
24438 +aufs-$(CONFIG_AUFS_RDU) += rdu.o
24439 +aufs-$(CONFIG_AUFS_BR_HFSPLUS) += hfsplus.o
24440 +aufs-$(CONFIG_AUFS_DEBUG) += debug.o
24441 +aufs-$(CONFIG_AUFS_MAGIC_SYSRQ) += sysrq.o
24442 diff -urN /usr/share/empty/fs/aufs/module.c linux/fs/aufs/module.c
24443 --- /usr/share/empty/fs/aufs/module.c   1970-01-01 01:00:00.000000000 +0100
24444 +++ linux/fs/aufs/module.c      2021-12-03 15:38:59.936647310 +0100
24445 @@ -0,0 +1,273 @@
24446 +// SPDX-License-Identifier: GPL-2.0
24447 +/*
24448 + * Copyright (C) 2005-2021 Junjiro R. Okajima
24449 + *
24450 + * This program, aufs is free software; you can redistribute it and/or modify
24451 + * it under the terms of the GNU General Public License as published by
24452 + * the Free Software Foundation; either version 2 of the License, or
24453 + * (at your option) any later version.
24454 + *
24455 + * This program is distributed in the hope that it will be useful,
24456 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24457 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24458 + * GNU General Public License for more details.
24459 + *
24460 + * You should have received a copy of the GNU General Public License
24461 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24462 + */
24463 +
24464 +/*
24465 + * module global variables and operations
24466 + */
24467 +
24468 +#include <linux/module.h>
24469 +#include <linux/seq_file.h>
24470 +#include "aufs.h"
24471 +
24472 +/* shrinkable realloc */
24473 +void *au_krealloc(void *p, unsigned int new_sz, gfp_t gfp, int may_shrink)
24474 +{
24475 +       size_t sz;
24476 +       int diff;
24477 +
24478 +       sz = 0;
24479 +       diff = -1;
24480 +       if (p) {
24481 +#if 0 /* unused */
24482 +               if (!new_sz) {
24483 +                       au_kfree_rcu(p);
24484 +                       p = NULL;
24485 +                       goto out;
24486 +               }
24487 +#else
24488 +               AuDebugOn(!new_sz);
24489 +#endif
24490 +               sz = ksize(p);
24491 +               diff = au_kmidx_sub(sz, new_sz);
24492 +       }
24493 +       if (sz && !diff)
24494 +               goto out;
24495 +
24496 +       if (sz < new_sz)
24497 +               /* expand or SLOB */
24498 +               p = krealloc(p, new_sz, gfp);
24499 +       else if (new_sz < sz && may_shrink) {
24500 +               /* shrink */
24501 +               void *q;
24502 +
24503 +               q = kmalloc(new_sz, gfp);
24504 +               if (q) {
24505 +                       if (p) {
24506 +                               memcpy(q, p, new_sz);
24507 +                               au_kfree_try_rcu(p);
24508 +                       }
24509 +                       p = q;
24510 +               } else
24511 +                       p = NULL;
24512 +       }
24513 +
24514 +out:
24515 +       return p;
24516 +}
24517 +
24518 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp,
24519 +                  int may_shrink)
24520 +{
24521 +       p = au_krealloc(p, new_sz, gfp, may_shrink);
24522 +       if (p && new_sz > nused)
24523 +               memset(p + nused, 0, new_sz - nused);
24524 +       return p;
24525 +}
24526 +
24527 +/* ---------------------------------------------------------------------- */
24528 +/*
24529 + * aufs caches
24530 + */
24531 +struct kmem_cache *au_cache[AuCache_Last];
24532 +
24533 +static void au_cache_fin(void)
24534 +{
24535 +       int i;
24536 +
24537 +       /*
24538 +        * Make sure all delayed rcu free inodes are flushed before we
24539 +        * destroy cache.
24540 +        */
24541 +       rcu_barrier();
24542 +
24543 +       /* excluding AuCache_HNOTIFY */
24544 +       BUILD_BUG_ON(AuCache_HNOTIFY + 1 != AuCache_Last);
24545 +       for (i = 0; i < AuCache_HNOTIFY; i++) {
24546 +               kmem_cache_destroy(au_cache[i]);
24547 +               au_cache[i] = NULL;
24548 +       }
24549 +}
24550 +
24551 +static int __init au_cache_init(void)
24552 +{
24553 +       au_cache[AuCache_DINFO] = AuCacheCtor(au_dinfo, au_di_init_once);
24554 +       if (au_cache[AuCache_DINFO])
24555 +               /* SLAB_DESTROY_BY_RCU */
24556 +               au_cache[AuCache_ICNTNR] = AuCacheCtor(au_icntnr,
24557 +                                                      au_icntnr_init_once);
24558 +       if (au_cache[AuCache_ICNTNR])
24559 +               au_cache[AuCache_FINFO] = AuCacheCtor(au_finfo,
24560 +                                                     au_fi_init_once);
24561 +       if (au_cache[AuCache_FINFO])
24562 +               au_cache[AuCache_VDIR] = AuCache(au_vdir);
24563 +       if (au_cache[AuCache_VDIR])
24564 +               au_cache[AuCache_DEHSTR] = AuCache(au_vdir_dehstr);
24565 +       if (au_cache[AuCache_DEHSTR])
24566 +               return 0;
24567 +
24568 +       au_cache_fin();
24569 +       return -ENOMEM;
24570 +}
24571 +
24572 +/* ---------------------------------------------------------------------- */
24573 +
24574 +int au_dir_roflags;
24575 +
24576 +#ifdef CONFIG_AUFS_SBILIST
24577 +/*
24578 + * iterate_supers_type() doesn't protect us from
24579 + * remounting (branch management)
24580 + */
24581 +struct hlist_bl_head au_sbilist;
24582 +#endif
24583 +
24584 +/*
24585 + * functions for module interface.
24586 + */
24587 +MODULE_LICENSE("GPL");
24588 +/* MODULE_LICENSE("GPL v2"); */
24589 +MODULE_AUTHOR("Junjiro R. Okajima <aufs-users@lists.sourceforge.net>");
24590 +MODULE_DESCRIPTION(AUFS_NAME
24591 +       " -- Advanced multi layered unification filesystem");
24592 +MODULE_VERSION(AUFS_VERSION);
24593 +MODULE_ALIAS_FS(AUFS_NAME);
24594 +
24595 +/* this module parameter has no meaning when SYSFS is disabled */
24596 +int sysaufs_brs = 1;
24597 +MODULE_PARM_DESC(brs, "use <sysfs>/fs/aufs/si_*/brN");
24598 +module_param_named(brs, sysaufs_brs, int, 0444);
24599 +
24600 +/* this module parameter has no meaning when USER_NS is disabled */
24601 +bool au_userns;
24602 +MODULE_PARM_DESC(allow_userns, "allow unprivileged to mount under userns");
24603 +module_param_named(allow_userns, au_userns, bool, 0444);
24604 +
24605 +/* ---------------------------------------------------------------------- */
24606 +
24607 +static char au_esc_chars[0x20 + 3]; /* 0x01-0x20, backslash, del, and NULL */
24608 +
24609 +int au_seq_path(struct seq_file *seq, struct path *path)
24610 +{
24611 +       int err;
24612 +
24613 +       err = seq_path(seq, path, au_esc_chars);
24614 +       if (err >= 0)
24615 +               err = 0;
24616 +       else
24617 +               err = -ENOMEM;
24618 +
24619 +       return err;
24620 +}
24621 +
24622 +/* ---------------------------------------------------------------------- */
24623 +
24624 +static int __init aufs_init(void)
24625 +{
24626 +       int err, i;
24627 +       char *p;
24628 +
24629 +       p = au_esc_chars;
24630 +       for (i = 1; i <= ' '; i++)
24631 +               *p++ = i;
24632 +       *p++ = '\\';
24633 +       *p++ = '\x7f';
24634 +       *p = 0;
24635 +
24636 +       au_dir_roflags = au_file_roflags(O_DIRECTORY | O_LARGEFILE);
24637 +
24638 +       memcpy(aufs_iop_nogetattr, aufs_iop, sizeof(aufs_iop));
24639 +       for (i = 0; i < AuIop_Last; i++)
24640 +               aufs_iop_nogetattr[i].getattr = NULL;
24641 +
24642 +       memset(au_cache, 0, sizeof(au_cache));  /* including hnotify */
24643 +
24644 +       au_sbilist_init();
24645 +       sysaufs_brs_init();
24646 +       au_debug_init();
24647 +       au_dy_init();
24648 +       err = sysaufs_init();
24649 +       if (unlikely(err))
24650 +               goto out;
24651 +       err = dbgaufs_init();
24652 +       if (unlikely(err))
24653 +               goto out_sysaufs;
24654 +       err = au_procfs_init();
24655 +       if (unlikely(err))
24656 +               goto out_dbgaufs;
24657 +       err = au_wkq_init();
24658 +       if (unlikely(err))
24659 +               goto out_procfs;
24660 +       err = au_loopback_init();
24661 +       if (unlikely(err))
24662 +               goto out_wkq;
24663 +       err = au_hnotify_init();
24664 +       if (unlikely(err))
24665 +               goto out_loopback;
24666 +       err = au_sysrq_init();
24667 +       if (unlikely(err))
24668 +               goto out_hin;
24669 +       err = au_cache_init();
24670 +       if (unlikely(err))
24671 +               goto out_sysrq;
24672 +
24673 +       aufs_fs_type.fs_flags |= au_userns ? FS_USERNS_MOUNT : 0;
24674 +       err = register_filesystem(&aufs_fs_type);
24675 +       if (unlikely(err))
24676 +               goto out_cache;
24677 +
24678 +       /* since we define pr_fmt, call printk directly */
24679 +       printk(KERN_INFO AUFS_NAME " " AUFS_VERSION "\n");
24680 +       goto out; /* success */
24681 +
24682 +out_cache:
24683 +       au_cache_fin();
24684 +out_sysrq:
24685 +       au_sysrq_fin();
24686 +out_hin:
24687 +       au_hnotify_fin();
24688 +out_loopback:
24689 +       au_loopback_fin();
24690 +out_wkq:
24691 +       au_wkq_fin();
24692 +out_procfs:
24693 +       au_procfs_fin();
24694 +out_dbgaufs:
24695 +       dbgaufs_fin();
24696 +out_sysaufs:
24697 +       sysaufs_fin();
24698 +       au_dy_fin();
24699 +out:
24700 +       return err;
24701 +}
24702 +
24703 +static void __exit aufs_exit(void)
24704 +{
24705 +       unregister_filesystem(&aufs_fs_type);
24706 +       au_cache_fin();
24707 +       au_sysrq_fin();
24708 +       au_hnotify_fin();
24709 +       au_loopback_fin();
24710 +       au_wkq_fin();
24711 +       au_procfs_fin();
24712 +       dbgaufs_fin();
24713 +       sysaufs_fin();
24714 +       au_dy_fin();
24715 +}
24716 +
24717 +module_init(aufs_init);
24718 +module_exit(aufs_exit);
24719 diff -urN /usr/share/empty/fs/aufs/module.h linux/fs/aufs/module.h
24720 --- /usr/share/empty/fs/aufs/module.h   1970-01-01 01:00:00.000000000 +0100
24721 +++ linux/fs/aufs/module.h      2021-12-03 15:40:58.236647297 +0100
24722 @@ -0,0 +1,166 @@
24723 +/* SPDX-License-Identifier: GPL-2.0 */
24724 +/*
24725 + * Copyright (C) 2005-2021 Junjiro R. Okajima
24726 + *
24727 + * This program, aufs is free software; you can redistribute it and/or modify
24728 + * it under the terms of the GNU General Public License as published by
24729 + * the Free Software Foundation; either version 2 of the License, or
24730 + * (at your option) any later version.
24731 + *
24732 + * This program is distributed in the hope that it will be useful,
24733 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24734 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24735 + * GNU General Public License for more details.
24736 + *
24737 + * You should have received a copy of the GNU General Public License
24738 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24739 + */
24740 +
24741 +/*
24742 + * module initialization and module-global
24743 + */
24744 +
24745 +#ifndef __AUFS_MODULE_H__
24746 +#define __AUFS_MODULE_H__
24747 +
24748 +#ifdef __KERNEL__
24749 +
24750 +#include <linux/slab.h>
24751 +#include "debug.h"
24752 +#include "dentry.h"
24753 +#include "dir.h"
24754 +#include "file.h"
24755 +#include "inode.h"
24756 +
24757 +struct path;
24758 +struct seq_file;
24759 +
24760 +/* module parameters */
24761 +extern int sysaufs_brs;
24762 +extern bool au_userns;
24763 +
24764 +/* ---------------------------------------------------------------------- */
24765 +
24766 +extern int au_dir_roflags;
24767 +
24768 +void *au_krealloc(void *p, unsigned int new_sz, gfp_t gfp, int may_shrink);
24769 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp,
24770 +                  int may_shrink);
24771 +
24772 +/*
24773 + * Comparing the size of the object with sizeof(struct rcu_head)
24774 + * case 1: object is always larger
24775 + *     --> au_kfree_rcu() or au_kfree_do_rcu()
24776 + * case 2: object is always smaller
24777 + *     --> au_kfree_small()
24778 + * case 3: object can be any size
24779 + *     --> au_kfree_try_rcu()
24780 + */
24781 +
24782 +static inline void au_kfree_do_rcu(const void *p)
24783 +{
24784 +       struct {
24785 +               struct rcu_head rcu;
24786 +       } *a = (void *)p;
24787 +
24788 +       kfree_rcu(a, rcu);
24789 +}
24790 +
24791 +#define au_kfree_rcu(_p) do {                                          \
24792 +               typeof(_p) p = (_p);                                    \
24793 +               BUILD_BUG_ON(sizeof(*p) < sizeof(struct rcu_head));     \
24794 +               if (p)                                                  \
24795 +                       au_kfree_do_rcu(p);                             \
24796 +       } while (0)
24797 +
24798 +#define au_kfree_do_sz_test(sz)        (sz >= sizeof(struct rcu_head))
24799 +#define au_kfree_sz_test(p)    (p && au_kfree_do_sz_test(ksize(p)))
24800 +
24801 +static inline void au_kfree_try_rcu(const void *p)
24802 +{
24803 +       if (!p)
24804 +               return;
24805 +       if (au_kfree_sz_test(p))
24806 +               au_kfree_do_rcu(p);
24807 +       else
24808 +               kfree(p);
24809 +}
24810 +
24811 +static inline void au_kfree_small(const void *p)
24812 +{
24813 +       if (!p)
24814 +               return;
24815 +       AuDebugOn(au_kfree_sz_test(p));
24816 +       kfree(p);
24817 +}
24818 +
24819 +static inline int au_kmidx_sub(size_t sz, size_t new_sz)
24820 +{
24821 +#ifndef CONFIG_SLOB
24822 +       return __kmalloc_index(sz, false) - __kmalloc_index(new_sz, false);
24823 +#else
24824 +       return -1; /* SLOB is untested */
24825 +#endif
24826 +}
24827 +
24828 +int au_seq_path(struct seq_file *seq, struct path *path);
24829 +
24830 +#ifdef CONFIG_PROC_FS
24831 +/* procfs.c */
24832 +int __init au_procfs_init(void);
24833 +void au_procfs_fin(void);
24834 +#else
24835 +AuStubInt0(au_procfs_init, void);
24836 +AuStubVoid(au_procfs_fin, void);
24837 +#endif
24838 +
24839 +/* ---------------------------------------------------------------------- */
24840 +
24841 +/* kmem cache */
24842 +enum {
24843 +       AuCache_DINFO,
24844 +       AuCache_ICNTNR,
24845 +       AuCache_FINFO,
24846 +       AuCache_VDIR,
24847 +       AuCache_DEHSTR,
24848 +       AuCache_HNOTIFY, /* must be last */
24849 +       AuCache_Last
24850 +};
24851 +
24852 +extern struct kmem_cache *au_cache[AuCache_Last];
24853 +
24854 +#define AuCacheFlags           (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD)
24855 +#define AuCache(type)          KMEM_CACHE(type, AuCacheFlags)
24856 +#define AuCacheCtor(type, ctor)        \
24857 +       kmem_cache_create(#type, sizeof(struct type), \
24858 +                         __alignof__(struct type), AuCacheFlags, ctor)
24859 +
24860 +#define AuCacheFuncs(name, index)                                      \
24861 +       static inline struct au_##name *au_cache_alloc_##name(void)     \
24862 +       { return kmem_cache_alloc(au_cache[AuCache_##index], GFP_NOFS); } \
24863 +       static inline void au_cache_free_##name##_norcu(struct au_##name *p) \
24864 +       { kmem_cache_free(au_cache[AuCache_##index], p); }              \
24865 +                                                                       \
24866 +       static inline void au_cache_free_##name##_rcu_cb(struct rcu_head *rcu) \
24867 +       { void *p = rcu;                                                \
24868 +               p -= offsetof(struct au_##name, rcu);                   \
24869 +               kmem_cache_free(au_cache[AuCache_##index], p); }        \
24870 +       static inline void au_cache_free_##name##_rcu(struct au_##name *p) \
24871 +       { BUILD_BUG_ON(sizeof(struct au_##name) < sizeof(struct rcu_head)); \
24872 +               call_rcu(&p->rcu, au_cache_free_##name##_rcu_cb); }     \
24873 +                                                                       \
24874 +       static inline void au_cache_free_##name(struct au_##name *p)    \
24875 +       { /* au_cache_free_##name##_norcu(p); */                        \
24876 +               au_cache_free_##name##_rcu(p); }
24877 +
24878 +AuCacheFuncs(dinfo, DINFO);
24879 +AuCacheFuncs(icntnr, ICNTNR);
24880 +AuCacheFuncs(finfo, FINFO);
24881 +AuCacheFuncs(vdir, VDIR);
24882 +AuCacheFuncs(vdir_dehstr, DEHSTR);
24883 +#ifdef CONFIG_AUFS_HNOTIFY
24884 +AuCacheFuncs(hnotify, HNOTIFY);
24885 +#endif
24886 +
24887 +#endif /* __KERNEL__ */
24888 +#endif /* __AUFS_MODULE_H__ */
24889 diff -urN /usr/share/empty/fs/aufs/mvdown.c linux/fs/aufs/mvdown.c
24890 --- /usr/share/empty/fs/aufs/mvdown.c   1970-01-01 01:00:00.000000000 +0100
24891 +++ linux/fs/aufs/mvdown.c      2021-12-03 15:38:59.939980643 +0100
24892 @@ -0,0 +1,706 @@
24893 +// SPDX-License-Identifier: GPL-2.0
24894 +/*
24895 + * Copyright (C) 2011-2021 Junjiro R. Okajima
24896 + *
24897 + * This program, aufs is free software; you can redistribute it and/or modify
24898 + * it under the terms of the GNU General Public License as published by
24899 + * the Free Software Foundation; either version 2 of the License, or
24900 + * (at your option) any later version.
24901 + *
24902 + * This program is distributed in the hope that it will be useful,
24903 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24904 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24905 + * GNU General Public License for more details.
24906 + *
24907 + * You should have received a copy of the GNU General Public License
24908 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24909 + */
24910 +
24911 +/*
24912 + * move-down, opposite of copy-up
24913 + */
24914 +
24915 +#include "aufs.h"
24916 +
24917 +struct au_mvd_args {
24918 +       struct {
24919 +               struct super_block *h_sb;
24920 +               struct dentry *h_parent;
24921 +               struct au_hinode *hdir;
24922 +               struct inode *h_dir, *h_inode;
24923 +               struct au_pin pin;
24924 +       } info[AUFS_MVDOWN_NARRAY];
24925 +
24926 +       struct aufs_mvdown mvdown;
24927 +       struct dentry *dentry, *parent;
24928 +       struct inode *inode, *dir;
24929 +       struct super_block *sb;
24930 +       aufs_bindex_t bopq, bwh, bfound;
24931 +       unsigned char rename_lock;
24932 +};
24933 +
24934 +#define mvd_errno              mvdown.au_errno
24935 +#define mvd_bsrc               mvdown.stbr[AUFS_MVDOWN_UPPER].bindex
24936 +#define mvd_src_brid           mvdown.stbr[AUFS_MVDOWN_UPPER].brid
24937 +#define mvd_bdst               mvdown.stbr[AUFS_MVDOWN_LOWER].bindex
24938 +#define mvd_dst_brid           mvdown.stbr[AUFS_MVDOWN_LOWER].brid
24939 +
24940 +#define mvd_h_src_sb           info[AUFS_MVDOWN_UPPER].h_sb
24941 +#define mvd_h_src_parent       info[AUFS_MVDOWN_UPPER].h_parent
24942 +#define mvd_hdir_src           info[AUFS_MVDOWN_UPPER].hdir
24943 +#define mvd_h_src_dir          info[AUFS_MVDOWN_UPPER].h_dir
24944 +#define mvd_h_src_inode                info[AUFS_MVDOWN_UPPER].h_inode
24945 +#define mvd_pin_src            info[AUFS_MVDOWN_UPPER].pin
24946 +
24947 +#define mvd_h_dst_sb           info[AUFS_MVDOWN_LOWER].h_sb
24948 +#define mvd_h_dst_parent       info[AUFS_MVDOWN_LOWER].h_parent
24949 +#define mvd_hdir_dst           info[AUFS_MVDOWN_LOWER].hdir
24950 +#define mvd_h_dst_dir          info[AUFS_MVDOWN_LOWER].h_dir
24951 +#define mvd_h_dst_inode                info[AUFS_MVDOWN_LOWER].h_inode
24952 +#define mvd_pin_dst            info[AUFS_MVDOWN_LOWER].pin
24953 +
24954 +#define AU_MVD_PR(flag, ...) do {                      \
24955 +               if (flag)                               \
24956 +                       pr_err(__VA_ARGS__);            \
24957 +       } while (0)
24958 +
24959 +static int find_lower_writable(struct au_mvd_args *a)
24960 +{
24961 +       struct super_block *sb;
24962 +       aufs_bindex_t bindex, bbot;
24963 +       struct au_branch *br;
24964 +
24965 +       sb = a->sb;
24966 +       bindex = a->mvd_bsrc;
24967 +       bbot = au_sbbot(sb);
24968 +       if (a->mvdown.flags & AUFS_MVDOWN_FHSM_LOWER)
24969 +               for (bindex++; bindex <= bbot; bindex++) {
24970 +                       br = au_sbr(sb, bindex);
24971 +                       if (au_br_fhsm(br->br_perm)
24972 +                           && !sb_rdonly(au_br_sb(br)))
24973 +                               return bindex;
24974 +               }
24975 +       else if (!(a->mvdown.flags & AUFS_MVDOWN_ROLOWER))
24976 +               for (bindex++; bindex <= bbot; bindex++) {
24977 +                       br = au_sbr(sb, bindex);
24978 +                       if (!au_br_rdonly(br))
24979 +                               return bindex;
24980 +               }
24981 +       else
24982 +               for (bindex++; bindex <= bbot; bindex++) {
24983 +                       br = au_sbr(sb, bindex);
24984 +                       if (!sb_rdonly(au_br_sb(br))) {
24985 +                               if (au_br_rdonly(br))
24986 +                                       a->mvdown.flags
24987 +                                               |= AUFS_MVDOWN_ROLOWER_R;
24988 +                               return bindex;
24989 +                       }
24990 +               }
24991 +
24992 +       return -1;
24993 +}
24994 +
24995 +/* make the parent dir on bdst */
24996 +static int au_do_mkdir(const unsigned char dmsg, struct au_mvd_args *a)
24997 +{
24998 +       int err;
24999 +
25000 +       err = 0;
25001 +       a->mvd_hdir_src = au_hi(a->dir, a->mvd_bsrc);
25002 +       a->mvd_hdir_dst = au_hi(a->dir, a->mvd_bdst);
25003 +       a->mvd_h_src_parent = au_h_dptr(a->parent, a->mvd_bsrc);
25004 +       a->mvd_h_dst_parent = NULL;
25005 +       if (au_dbbot(a->parent) >= a->mvd_bdst)
25006 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
25007 +       if (!a->mvd_h_dst_parent) {
25008 +               err = au_cpdown_dirs(a->dentry, a->mvd_bdst);
25009 +               if (unlikely(err)) {
25010 +                       AU_MVD_PR(dmsg, "cpdown_dirs failed\n");
25011 +                       goto out;
25012 +               }
25013 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
25014 +       }
25015 +
25016 +out:
25017 +       AuTraceErr(err);
25018 +       return err;
25019 +}
25020 +
25021 +/* lock them all */
25022 +static int au_do_lock(const unsigned char dmsg, struct au_mvd_args *a)
25023 +{
25024 +       int err;
25025 +       struct dentry *h_trap;
25026 +
25027 +       a->mvd_h_src_sb = au_sbr_sb(a->sb, a->mvd_bsrc);
25028 +       a->mvd_h_dst_sb = au_sbr_sb(a->sb, a->mvd_bdst);
25029 +       err = au_pin(&a->mvd_pin_dst, a->dentry, a->mvd_bdst,
25030 +                    au_opt_udba(a->sb),
25031 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
25032 +       AuTraceErr(err);
25033 +       if (unlikely(err)) {
25034 +               AU_MVD_PR(dmsg, "pin_dst failed\n");
25035 +               goto out;
25036 +       }
25037 +
25038 +       if (a->mvd_h_src_sb != a->mvd_h_dst_sb) {
25039 +               a->rename_lock = 0;
25040 +               au_pin_init(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
25041 +                           AuLsc_DI_PARENT, AuLsc_I_PARENT3,
25042 +                           au_opt_udba(a->sb),
25043 +                           AuPin_MNT_WRITE | AuPin_DI_LOCKED);
25044 +               err = au_do_pin(&a->mvd_pin_src);
25045 +               AuTraceErr(err);
25046 +               a->mvd_h_src_dir = d_inode(a->mvd_h_src_parent);
25047 +               if (unlikely(err)) {
25048 +                       AU_MVD_PR(dmsg, "pin_src failed\n");
25049 +                       goto out_dst;
25050 +               }
25051 +               goto out; /* success */
25052 +       }
25053 +
25054 +       a->rename_lock = 1;
25055 +       au_pin_hdir_unlock(&a->mvd_pin_dst);
25056 +       err = au_pin(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
25057 +                    au_opt_udba(a->sb),
25058 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
25059 +       AuTraceErr(err);
25060 +       a->mvd_h_src_dir = d_inode(a->mvd_h_src_parent);
25061 +       if (unlikely(err)) {
25062 +               AU_MVD_PR(dmsg, "pin_src failed\n");
25063 +               au_pin_hdir_lock(&a->mvd_pin_dst);
25064 +               goto out_dst;
25065 +       }
25066 +       au_pin_hdir_unlock(&a->mvd_pin_src);
25067 +       h_trap = vfsub_lock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
25068 +                                  a->mvd_h_dst_parent, a->mvd_hdir_dst);
25069 +       if (h_trap) {
25070 +               err = (h_trap != a->mvd_h_src_parent);
25071 +               if (err)
25072 +                       err = (h_trap != a->mvd_h_dst_parent);
25073 +       }
25074 +       BUG_ON(err); /* it should never happen */
25075 +       if (unlikely(a->mvd_h_src_dir != au_pinned_h_dir(&a->mvd_pin_src))) {
25076 +               err = -EBUSY;
25077 +               AuTraceErr(err);
25078 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
25079 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
25080 +               au_pin_hdir_lock(&a->mvd_pin_src);
25081 +               au_unpin(&a->mvd_pin_src);
25082 +               au_pin_hdir_lock(&a->mvd_pin_dst);
25083 +               goto out_dst;
25084 +       }
25085 +       goto out; /* success */
25086 +
25087 +out_dst:
25088 +       au_unpin(&a->mvd_pin_dst);
25089 +out:
25090 +       AuTraceErr(err);
25091 +       return err;
25092 +}
25093 +
25094 +static void au_do_unlock(const unsigned char dmsg, struct au_mvd_args *a)
25095 +{
25096 +       if (!a->rename_lock)
25097 +               au_unpin(&a->mvd_pin_src);
25098 +       else {
25099 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
25100 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
25101 +               au_pin_hdir_lock(&a->mvd_pin_src);
25102 +               au_unpin(&a->mvd_pin_src);
25103 +               au_pin_hdir_lock(&a->mvd_pin_dst);
25104 +       }
25105 +       au_unpin(&a->mvd_pin_dst);
25106 +}
25107 +
25108 +/* copy-down the file */
25109 +static int au_do_cpdown(const unsigned char dmsg, struct au_mvd_args *a)
25110 +{
25111 +       int err;
25112 +       struct au_cp_generic cpg = {
25113 +               .dentry = a->dentry,
25114 +               .bdst   = a->mvd_bdst,
25115 +               .bsrc   = a->mvd_bsrc,
25116 +               .len    = -1,
25117 +               .pin    = &a->mvd_pin_dst,
25118 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
25119 +       };
25120 +
25121 +       AuDbg("b%d, b%d\n", cpg.bsrc, cpg.bdst);
25122 +       if (a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
25123 +               au_fset_cpup(cpg.flags, OVERWRITE);
25124 +       if (a->mvdown.flags & AUFS_MVDOWN_ROLOWER)
25125 +               au_fset_cpup(cpg.flags, RWDST);
25126 +       err = au_sio_cpdown_simple(&cpg);
25127 +       if (unlikely(err))
25128 +               AU_MVD_PR(dmsg, "cpdown failed\n");
25129 +
25130 +       AuTraceErr(err);
25131 +       return err;
25132 +}
25133 +
25134 +/*
25135 + * unlink the whiteout on bdst if exist which may be created by UDBA while we
25136 + * were sleeping
25137 + */
25138 +static int au_do_unlink_wh(const unsigned char dmsg, struct au_mvd_args *a)
25139 +{
25140 +       int err;
25141 +       struct path h_path;
25142 +       struct au_branch *br;
25143 +       struct inode *delegated;
25144 +
25145 +       br = au_sbr(a->sb, a->mvd_bdst);
25146 +       h_path.dentry = au_wh_lkup(a->mvd_h_dst_parent, &a->dentry->d_name, br);
25147 +       err = PTR_ERR(h_path.dentry);
25148 +       if (IS_ERR(h_path.dentry)) {
25149 +               AU_MVD_PR(dmsg, "wh_lkup failed\n");
25150 +               goto out;
25151 +       }
25152 +
25153 +       err = 0;
25154 +       if (d_is_positive(h_path.dentry)) {
25155 +               h_path.mnt = au_br_mnt(br);
25156 +               delegated = NULL;
25157 +               err = vfsub_unlink(d_inode(a->mvd_h_dst_parent), &h_path,
25158 +                                  &delegated, /*force*/0);
25159 +               if (unlikely(err == -EWOULDBLOCK)) {
25160 +                       pr_warn("cannot retry for NFSv4 delegation"
25161 +                               " for an internal unlink\n");
25162 +                       iput(delegated);
25163 +               }
25164 +               if (unlikely(err))
25165 +                       AU_MVD_PR(dmsg, "wh_unlink failed\n");
25166 +       }
25167 +       dput(h_path.dentry);
25168 +
25169 +out:
25170 +       AuTraceErr(err);
25171 +       return err;
25172 +}
25173 +
25174 +/*
25175 + * unlink the topmost h_dentry
25176 + */
25177 +static int au_do_unlink(const unsigned char dmsg, struct au_mvd_args *a)
25178 +{
25179 +       int err;
25180 +       struct path h_path;
25181 +       struct inode *delegated;
25182 +
25183 +       h_path.mnt = au_sbr_mnt(a->sb, a->mvd_bsrc);
25184 +       h_path.dentry = au_h_dptr(a->dentry, a->mvd_bsrc);
25185 +       delegated = NULL;
25186 +       err = vfsub_unlink(a->mvd_h_src_dir, &h_path, &delegated, /*force*/0);
25187 +       if (unlikely(err == -EWOULDBLOCK)) {
25188 +               pr_warn("cannot retry for NFSv4 delegation"
25189 +                       " for an internal unlink\n");
25190 +               iput(delegated);
25191 +       }
25192 +       if (unlikely(err))
25193 +               AU_MVD_PR(dmsg, "unlink failed\n");
25194 +
25195 +       AuTraceErr(err);
25196 +       return err;
25197 +}
25198 +
25199 +/* Since mvdown succeeded, we ignore an error of this function */
25200 +static void au_do_stfs(const unsigned char dmsg, struct au_mvd_args *a)
25201 +{
25202 +       int err;
25203 +       struct au_branch *br;
25204 +
25205 +       a->mvdown.flags |= AUFS_MVDOWN_STFS_FAILED;
25206 +       br = au_sbr(a->sb, a->mvd_bsrc);
25207 +       err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_UPPER].stfs);
25208 +       if (!err) {
25209 +               br = au_sbr(a->sb, a->mvd_bdst);
25210 +               a->mvdown.stbr[AUFS_MVDOWN_LOWER].brid = br->br_id;
25211 +               err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_LOWER].stfs);
25212 +       }
25213 +       if (!err)
25214 +               a->mvdown.flags &= ~AUFS_MVDOWN_STFS_FAILED;
25215 +       else
25216 +               AU_MVD_PR(dmsg, "statfs failed (%d), ignored\n", err);
25217 +}
25218 +
25219 +/*
25220 + * copy-down the file and unlink the bsrc file.
25221 + * - unlink the bdst whout if exist
25222 + * - copy-down the file (with whtmp name and rename)
25223 + * - unlink the bsrc file
25224 + */
25225 +static int au_do_mvdown(const unsigned char dmsg, struct au_mvd_args *a)
25226 +{
25227 +       int err;
25228 +
25229 +       err = au_do_mkdir(dmsg, a);
25230 +       if (!err)
25231 +               err = au_do_lock(dmsg, a);
25232 +       if (unlikely(err))
25233 +               goto out;
25234 +
25235 +       /*
25236 +        * do not revert the activities we made on bdst since they should be
25237 +        * harmless in aufs.
25238 +        */
25239 +
25240 +       err = au_do_cpdown(dmsg, a);
25241 +       if (!err)
25242 +               err = au_do_unlink_wh(dmsg, a);
25243 +       if (!err && !(a->mvdown.flags & AUFS_MVDOWN_KUPPER))
25244 +               err = au_do_unlink(dmsg, a);
25245 +       if (unlikely(err))
25246 +               goto out_unlock;
25247 +
25248 +       AuDbg("%pd2, 0x%x, %d --> %d\n",
25249 +             a->dentry, a->mvdown.flags, a->mvd_bsrc, a->mvd_bdst);
25250 +       if (find_lower_writable(a) < 0)
25251 +               a->mvdown.flags |= AUFS_MVDOWN_BOTTOM;
25252 +
25253 +       if (a->mvdown.flags & AUFS_MVDOWN_STFS)
25254 +               au_do_stfs(dmsg, a);
25255 +
25256 +       /* maintain internal array */
25257 +       if (!(a->mvdown.flags & AUFS_MVDOWN_KUPPER)) {
25258 +               au_set_h_dptr(a->dentry, a->mvd_bsrc, NULL);
25259 +               au_set_dbtop(a->dentry, a->mvd_bdst);
25260 +               au_set_h_iptr(a->inode, a->mvd_bsrc, NULL, /*flags*/0);
25261 +               au_set_ibtop(a->inode, a->mvd_bdst);
25262 +       } else {
25263 +               /* hide the lower */
25264 +               au_set_h_dptr(a->dentry, a->mvd_bdst, NULL);
25265 +               au_set_dbbot(a->dentry, a->mvd_bsrc);
25266 +               au_set_h_iptr(a->inode, a->mvd_bdst, NULL, /*flags*/0);
25267 +               au_set_ibbot(a->inode, a->mvd_bsrc);
25268 +       }
25269 +       if (au_dbbot(a->dentry) < a->mvd_bdst)
25270 +               au_set_dbbot(a->dentry, a->mvd_bdst);
25271 +       if (au_ibbot(a->inode) < a->mvd_bdst)
25272 +               au_set_ibbot(a->inode, a->mvd_bdst);
25273 +
25274 +out_unlock:
25275 +       au_do_unlock(dmsg, a);
25276 +out:
25277 +       AuTraceErr(err);
25278 +       return err;
25279 +}
25280 +
25281 +/* ---------------------------------------------------------------------- */
25282 +
25283 +/* make sure the file is idle */
25284 +static int au_mvd_args_busy(const unsigned char dmsg, struct au_mvd_args *a)
25285 +{
25286 +       int err, plinked;
25287 +
25288 +       err = 0;
25289 +       plinked = !!au_opt_test(au_mntflags(a->sb), PLINK);
25290 +       if (au_dbtop(a->dentry) == a->mvd_bsrc
25291 +           && au_dcount(a->dentry) == 1
25292 +           && atomic_read(&a->inode->i_count) == 1
25293 +           /* && a->mvd_h_src_inode->i_nlink == 1 */
25294 +           && (!plinked || !au_plink_test(a->inode))
25295 +           && a->inode->i_nlink == 1)
25296 +               goto out;
25297 +
25298 +       err = -EBUSY;
25299 +       AU_MVD_PR(dmsg,
25300 +                 "b%d, d{b%d, c%d?}, i{c%d?, l%u}, hi{l%u}, p{%d, %d}\n",
25301 +                 a->mvd_bsrc, au_dbtop(a->dentry), au_dcount(a->dentry),
25302 +                 atomic_read(&a->inode->i_count), a->inode->i_nlink,
25303 +                 a->mvd_h_src_inode->i_nlink,
25304 +                 plinked, plinked ? au_plink_test(a->inode) : 0);
25305 +
25306 +out:
25307 +       AuTraceErr(err);
25308 +       return err;
25309 +}
25310 +
25311 +/* make sure the parent dir is fine */
25312 +static int au_mvd_args_parent(const unsigned char dmsg,
25313 +                             struct au_mvd_args *a)
25314 +{
25315 +       int err;
25316 +       aufs_bindex_t bindex;
25317 +
25318 +       err = 0;
25319 +       if (unlikely(au_alive_dir(a->parent))) {
25320 +               err = -ENOENT;
25321 +               AU_MVD_PR(dmsg, "parent dir is dead\n");
25322 +               goto out;
25323 +       }
25324 +
25325 +       a->bopq = au_dbdiropq(a->parent);
25326 +       bindex = au_wbr_nonopq(a->dentry, a->mvd_bdst);
25327 +       AuDbg("b%d\n", bindex);
25328 +       if (unlikely((bindex >= 0 && bindex < a->mvd_bdst)
25329 +                    || (a->bopq != -1 && a->bopq < a->mvd_bdst))) {
25330 +               err = -EINVAL;
25331 +               a->mvd_errno = EAU_MVDOWN_OPAQUE;
25332 +               AU_MVD_PR(dmsg, "ancestor is opaque b%d, b%d\n",
25333 +                         a->bopq, a->mvd_bdst);
25334 +       }
25335 +
25336 +out:
25337 +       AuTraceErr(err);
25338 +       return err;
25339 +}
25340 +
25341 +static int au_mvd_args_intermediate(const unsigned char dmsg,
25342 +                                   struct au_mvd_args *a)
25343 +{
25344 +       int err;
25345 +       struct au_dinfo *dinfo, *tmp;
25346 +
25347 +       /* lookup the next lower positive entry */
25348 +       err = -ENOMEM;
25349 +       tmp = au_di_alloc(a->sb, AuLsc_DI_TMP);
25350 +       if (unlikely(!tmp))
25351 +               goto out;
25352 +
25353 +       a->bfound = -1;
25354 +       a->bwh = -1;
25355 +       dinfo = au_di(a->dentry);
25356 +       au_di_cp(tmp, dinfo);
25357 +       au_di_swap(tmp, dinfo);
25358 +
25359 +       /* returns the number of positive dentries */
25360 +       err = au_lkup_dentry(a->dentry, a->mvd_bsrc + 1,
25361 +                            /* AuLkup_IGNORE_PERM */ 0);
25362 +       if (!err)
25363 +               a->bwh = au_dbwh(a->dentry);
25364 +       else if (err > 0)
25365 +               a->bfound = au_dbtop(a->dentry);
25366 +
25367 +       au_di_swap(tmp, dinfo);
25368 +       au_rw_write_unlock(&tmp->di_rwsem);
25369 +       au_di_free(tmp);
25370 +       if (unlikely(err < 0))
25371 +               AU_MVD_PR(dmsg, "failed look-up lower\n");
25372 +
25373 +       /*
25374 +        * here, we have these cases.
25375 +        * bfound == -1
25376 +        *      no positive dentry under bsrc. there are more sub-cases.
25377 +        *      bwh < 0
25378 +        *              there no whiteout, we can safely move-down.
25379 +        *      bwh <= bsrc
25380 +        *              impossible
25381 +        *      bsrc < bwh && bwh < bdst
25382 +        *              there is a whiteout on RO branch. cannot proceed.
25383 +        *      bwh == bdst
25384 +        *              there is a whiteout on the RW target branch. it should
25385 +        *              be removed.
25386 +        *      bdst < bwh
25387 +        *              there is a whiteout somewhere unrelated branch.
25388 +        * -1 < bfound && bfound <= bsrc
25389 +        *      impossible.
25390 +        * bfound < bdst
25391 +        *      found, but it is on RO branch between bsrc and bdst. cannot
25392 +        *      proceed.
25393 +        * bfound == bdst
25394 +        *      found, replace it if AUFS_MVDOWN_FORCE is set. otherwise return
25395 +        *      error.
25396 +        * bdst < bfound
25397 +        *      found, after we create the file on bdst, it will be hidden.
25398 +        */
25399 +
25400 +       AuDebugOn(a->bfound == -1
25401 +                 && a->bwh != -1
25402 +                 && a->bwh <= a->mvd_bsrc);
25403 +       AuDebugOn(-1 < a->bfound
25404 +                 && a->bfound <= a->mvd_bsrc);
25405 +
25406 +       err = -EINVAL;
25407 +       if (a->bfound == -1
25408 +           && a->mvd_bsrc < a->bwh
25409 +           && a->bwh != -1
25410 +           && a->bwh < a->mvd_bdst) {
25411 +               a->mvd_errno = EAU_MVDOWN_WHITEOUT;
25412 +               AU_MVD_PR(dmsg, "bsrc %d, bdst %d, bfound %d, bwh %d\n",
25413 +                         a->mvd_bsrc, a->mvd_bdst, a->bfound, a->bwh);
25414 +               goto out;
25415 +       } else if (a->bfound != -1 && a->bfound < a->mvd_bdst) {
25416 +               a->mvd_errno = EAU_MVDOWN_UPPER;
25417 +               AU_MVD_PR(dmsg, "bdst %d, bfound %d\n",
25418 +                         a->mvd_bdst, a->bfound);
25419 +               goto out;
25420 +       }
25421 +
25422 +       err = 0; /* success */
25423 +
25424 +out:
25425 +       AuTraceErr(err);
25426 +       return err;
25427 +}
25428 +
25429 +static int au_mvd_args_exist(const unsigned char dmsg, struct au_mvd_args *a)
25430 +{
25431 +       int err;
25432 +
25433 +       err = 0;
25434 +       if (!(a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
25435 +           && a->bfound == a->mvd_bdst)
25436 +               err = -EEXIST;
25437 +       AuTraceErr(err);
25438 +       return err;
25439 +}
25440 +
25441 +static int au_mvd_args(const unsigned char dmsg, struct au_mvd_args *a)
25442 +{
25443 +       int err;
25444 +       struct au_branch *br;
25445 +
25446 +       err = -EISDIR;
25447 +       if (unlikely(S_ISDIR(a->inode->i_mode)))
25448 +               goto out;
25449 +
25450 +       err = -EINVAL;
25451 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_UPPER))
25452 +               a->mvd_bsrc = au_ibtop(a->inode);
25453 +       else {
25454 +               a->mvd_bsrc = au_br_index(a->sb, a->mvd_src_brid);
25455 +               if (unlikely(a->mvd_bsrc < 0
25456 +                            || (a->mvd_bsrc < au_dbtop(a->dentry)
25457 +                                || au_dbbot(a->dentry) < a->mvd_bsrc
25458 +                                || !au_h_dptr(a->dentry, a->mvd_bsrc))
25459 +                            || (a->mvd_bsrc < au_ibtop(a->inode)
25460 +                                || au_ibbot(a->inode) < a->mvd_bsrc
25461 +                                || !au_h_iptr(a->inode, a->mvd_bsrc)))) {
25462 +                       a->mvd_errno = EAU_MVDOWN_NOUPPER;
25463 +                       AU_MVD_PR(dmsg, "no upper\n");
25464 +                       goto out;
25465 +               }
25466 +       }
25467 +       if (unlikely(a->mvd_bsrc == au_sbbot(a->sb))) {
25468 +               a->mvd_errno = EAU_MVDOWN_BOTTOM;
25469 +               AU_MVD_PR(dmsg, "on the bottom\n");
25470 +               goto out;
25471 +       }
25472 +       a->mvd_h_src_inode = au_h_iptr(a->inode, a->mvd_bsrc);
25473 +       br = au_sbr(a->sb, a->mvd_bsrc);
25474 +       err = au_br_rdonly(br);
25475 +       if (!(a->mvdown.flags & AUFS_MVDOWN_ROUPPER)) {
25476 +               if (unlikely(err))
25477 +                       goto out;
25478 +       } else if (!(vfsub_native_ro(a->mvd_h_src_inode)
25479 +                    || IS_APPEND(a->mvd_h_src_inode))) {
25480 +               if (err)
25481 +                       a->mvdown.flags |= AUFS_MVDOWN_ROUPPER_R;
25482 +               /* go on */
25483 +       } else
25484 +               goto out;
25485 +
25486 +       err = -EINVAL;
25487 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_LOWER)) {
25488 +               a->mvd_bdst = find_lower_writable(a);
25489 +               if (unlikely(a->mvd_bdst < 0)) {
25490 +                       a->mvd_errno = EAU_MVDOWN_BOTTOM;
25491 +                       AU_MVD_PR(dmsg, "no writable lower branch\n");
25492 +                       goto out;
25493 +               }
25494 +       } else {
25495 +               a->mvd_bdst = au_br_index(a->sb, a->mvd_dst_brid);
25496 +               if (unlikely(a->mvd_bdst < 0
25497 +                            || au_sbbot(a->sb) < a->mvd_bdst)) {
25498 +                       a->mvd_errno = EAU_MVDOWN_NOLOWERBR;
25499 +                       AU_MVD_PR(dmsg, "no lower brid\n");
25500 +                       goto out;
25501 +               }
25502 +       }
25503 +
25504 +       err = au_mvd_args_busy(dmsg, a);
25505 +       if (!err)
25506 +               err = au_mvd_args_parent(dmsg, a);
25507 +       if (!err)
25508 +               err = au_mvd_args_intermediate(dmsg, a);
25509 +       if (!err)
25510 +               err = au_mvd_args_exist(dmsg, a);
25511 +       if (!err)
25512 +               AuDbg("b%d, b%d\n", a->mvd_bsrc, a->mvd_bdst);
25513 +
25514 +out:
25515 +       AuTraceErr(err);
25516 +       return err;
25517 +}
25518 +
25519 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *uarg)
25520 +{
25521 +       int err, e;
25522 +       unsigned char dmsg;
25523 +       struct au_mvd_args *args;
25524 +       struct inode *inode;
25525 +
25526 +       inode = d_inode(dentry);
25527 +       err = -EPERM;
25528 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
25529 +               goto out;
25530 +
25531 +       err = -ENOMEM;
25532 +       args = kmalloc(sizeof(*args), GFP_NOFS);
25533 +       if (unlikely(!args))
25534 +               goto out;
25535 +
25536 +       err = copy_from_user(&args->mvdown, uarg, sizeof(args->mvdown));
25537 +       if (!err)
25538 +               /* VERIFY_WRITE */
25539 +               err = !access_ok(uarg, sizeof(*uarg));
25540 +       if (unlikely(err)) {
25541 +               err = -EFAULT;
25542 +               AuTraceErr(err);
25543 +               goto out_free;
25544 +       }
25545 +       AuDbg("flags 0x%x\n", args->mvdown.flags);
25546 +       args->mvdown.flags &= ~(AUFS_MVDOWN_ROLOWER_R | AUFS_MVDOWN_ROUPPER_R);
25547 +       args->mvdown.au_errno = 0;
25548 +       args->dentry = dentry;
25549 +       args->inode = inode;
25550 +       args->sb = dentry->d_sb;
25551 +
25552 +       err = -ENOENT;
25553 +       dmsg = !!(args->mvdown.flags & AUFS_MVDOWN_DMSG);
25554 +       args->parent = dget_parent(dentry);
25555 +       args->dir = d_inode(args->parent);
25556 +       inode_lock_nested(args->dir, I_MUTEX_PARENT);
25557 +       dput(args->parent);
25558 +       if (unlikely(args->parent != dentry->d_parent)) {
25559 +               AU_MVD_PR(dmsg, "parent dir is moved\n");
25560 +               goto out_dir;
25561 +       }
25562 +
25563 +       inode_lock_nested(inode, I_MUTEX_CHILD);
25564 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_NOPLMW);
25565 +       if (unlikely(err))
25566 +               goto out_inode;
25567 +
25568 +       di_write_lock_parent(args->parent);
25569 +       err = au_mvd_args(dmsg, args);
25570 +       if (unlikely(err))
25571 +               goto out_parent;
25572 +
25573 +       err = au_do_mvdown(dmsg, args);
25574 +       if (unlikely(err))
25575 +               goto out_parent;
25576 +
25577 +       au_cpup_attr_timesizes(args->dir);
25578 +       au_cpup_attr_timesizes(inode);
25579 +       if (!(args->mvdown.flags & AUFS_MVDOWN_KUPPER))
25580 +               au_cpup_igen(inode, au_h_iptr(inode, args->mvd_bdst));
25581 +       /* au_digen_dec(dentry); */
25582 +
25583 +out_parent:
25584 +       di_write_unlock(args->parent);
25585 +       aufs_read_unlock(dentry, AuLock_DW);
25586 +out_inode:
25587 +       inode_unlock(inode);
25588 +out_dir:
25589 +       inode_unlock(args->dir);
25590 +out_free:
25591 +       e = copy_to_user(uarg, &args->mvdown, sizeof(args->mvdown));
25592 +       if (unlikely(e))
25593 +               err = -EFAULT;
25594 +       au_kfree_rcu(args);
25595 +out:
25596 +       AuTraceErr(err);
25597 +       return err;
25598 +}
25599 diff -urN /usr/share/empty/fs/aufs/opts.c linux/fs/aufs/opts.c
25600 --- /usr/share/empty/fs/aufs/opts.c     1970-01-01 01:00:00.000000000 +0100
25601 +++ linux/fs/aufs/opts.c        2021-12-03 15:38:59.939980643 +0100
25602 @@ -0,0 +1,1880 @@
25603 +// SPDX-License-Identifier: GPL-2.0
25604 +/*
25605 + * Copyright (C) 2005-2021 Junjiro R. Okajima
25606 + *
25607 + * This program, aufs is free software; you can redistribute it and/or modify
25608 + * it under the terms of the GNU General Public License as published by
25609 + * the Free Software Foundation; either version 2 of the License, or
25610 + * (at your option) any later version.
25611 + *
25612 + * This program is distributed in the hope that it will be useful,
25613 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25614 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25615 + * GNU General Public License for more details.
25616 + *
25617 + * You should have received a copy of the GNU General Public License
25618 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25619 + */
25620 +
25621 +/*
25622 + * mount options/flags
25623 + */
25624 +
25625 +#include <linux/namei.h>
25626 +#include <linux/types.h> /* a distribution requires */
25627 +#include <linux/parser.h>
25628 +#include "aufs.h"
25629 +
25630 +/* ---------------------------------------------------------------------- */
25631 +
25632 +enum {
25633 +       Opt_br,
25634 +       Opt_add, Opt_del, Opt_mod, Opt_append, Opt_prepend,
25635 +       Opt_idel, Opt_imod,
25636 +       Opt_dirwh, Opt_rdcache, Opt_rdblk, Opt_rdhash,
25637 +       Opt_rdblk_def, Opt_rdhash_def,
25638 +       Opt_xino, Opt_noxino,
25639 +       Opt_trunc_xino, Opt_trunc_xino_v, Opt_notrunc_xino,
25640 +       Opt_trunc_xino_path, Opt_itrunc_xino,
25641 +       Opt_trunc_xib, Opt_notrunc_xib,
25642 +       Opt_shwh, Opt_noshwh,
25643 +       Opt_plink, Opt_noplink, Opt_list_plink,
25644 +       Opt_udba,
25645 +       Opt_dio, Opt_nodio,
25646 +       Opt_diropq_a, Opt_diropq_w,
25647 +       Opt_warn_perm, Opt_nowarn_perm,
25648 +       Opt_wbr_copyup, Opt_wbr_create,
25649 +       Opt_fhsm_sec,
25650 +       Opt_verbose, Opt_noverbose,
25651 +       Opt_sum, Opt_nosum, Opt_wsum,
25652 +       Opt_dirperm1, Opt_nodirperm1,
25653 +       Opt_dirren, Opt_nodirren,
25654 +       Opt_acl, Opt_noacl,
25655 +       Opt_tail, Opt_ignore, Opt_ignore_silent, Opt_err
25656 +};
25657 +
25658 +static match_table_t options = {
25659 +       {Opt_br, "br=%s"},
25660 +       {Opt_br, "br:%s"},
25661 +
25662 +       {Opt_add, "add=%d:%s"},
25663 +       {Opt_add, "add:%d:%s"},
25664 +       {Opt_add, "ins=%d:%s"},
25665 +       {Opt_add, "ins:%d:%s"},
25666 +       {Opt_append, "append=%s"},
25667 +       {Opt_append, "append:%s"},
25668 +       {Opt_prepend, "prepend=%s"},
25669 +       {Opt_prepend, "prepend:%s"},
25670 +
25671 +       {Opt_del, "del=%s"},
25672 +       {Opt_del, "del:%s"},
25673 +       /* {Opt_idel, "idel:%d"}, */
25674 +       {Opt_mod, "mod=%s"},
25675 +       {Opt_mod, "mod:%s"},
25676 +       /* {Opt_imod, "imod:%d:%s"}, */
25677 +
25678 +       {Opt_dirwh, "dirwh=%d"},
25679 +
25680 +       {Opt_xino, "xino=%s"},
25681 +       {Opt_noxino, "noxino"},
25682 +       {Opt_trunc_xino, "trunc_xino"},
25683 +       {Opt_trunc_xino_v, "trunc_xino_v=%d:%d"},
25684 +       {Opt_notrunc_xino, "notrunc_xino"},
25685 +       {Opt_trunc_xino_path, "trunc_xino=%s"},
25686 +       {Opt_itrunc_xino, "itrunc_xino=%d"},
25687 +       /* {Opt_zxino, "zxino=%s"}, */
25688 +       {Opt_trunc_xib, "trunc_xib"},
25689 +       {Opt_notrunc_xib, "notrunc_xib"},
25690 +
25691 +#ifdef CONFIG_PROC_FS
25692 +       {Opt_plink, "plink"},
25693 +#else
25694 +       {Opt_ignore_silent, "plink"},
25695 +#endif
25696 +
25697 +       {Opt_noplink, "noplink"},
25698 +
25699 +#ifdef CONFIG_AUFS_DEBUG
25700 +       {Opt_list_plink, "list_plink"},
25701 +#endif
25702 +
25703 +       {Opt_udba, "udba=%s"},
25704 +
25705 +       {Opt_dio, "dio"},
25706 +       {Opt_nodio, "nodio"},
25707 +
25708 +#ifdef CONFIG_AUFS_DIRREN
25709 +       {Opt_dirren, "dirren"},
25710 +       {Opt_nodirren, "nodirren"},
25711 +#else
25712 +       {Opt_ignore, "dirren"},
25713 +       {Opt_ignore_silent, "nodirren"},
25714 +#endif
25715 +
25716 +#ifdef CONFIG_AUFS_FHSM
25717 +       {Opt_fhsm_sec, "fhsm_sec=%d"},
25718 +#else
25719 +       {Opt_ignore, "fhsm_sec=%d"},
25720 +#endif
25721 +
25722 +       {Opt_diropq_a, "diropq=always"},
25723 +       {Opt_diropq_a, "diropq=a"},
25724 +       {Opt_diropq_w, "diropq=whiteouted"},
25725 +       {Opt_diropq_w, "diropq=w"},
25726 +
25727 +       {Opt_warn_perm, "warn_perm"},
25728 +       {Opt_nowarn_perm, "nowarn_perm"},
25729 +
25730 +       /* keep them temporary */
25731 +       {Opt_ignore_silent, "nodlgt"},
25732 +       {Opt_ignore, "clean_plink"},
25733 +
25734 +#ifdef CONFIG_AUFS_SHWH
25735 +       {Opt_shwh, "shwh"},
25736 +#endif
25737 +       {Opt_noshwh, "noshwh"},
25738 +
25739 +       {Opt_dirperm1, "dirperm1"},
25740 +       {Opt_nodirperm1, "nodirperm1"},
25741 +
25742 +       {Opt_verbose, "verbose"},
25743 +       {Opt_verbose, "v"},
25744 +       {Opt_noverbose, "noverbose"},
25745 +       {Opt_noverbose, "quiet"},
25746 +       {Opt_noverbose, "q"},
25747 +       {Opt_noverbose, "silent"},
25748 +
25749 +       {Opt_sum, "sum"},
25750 +       {Opt_nosum, "nosum"},
25751 +       {Opt_wsum, "wsum"},
25752 +
25753 +       {Opt_rdcache, "rdcache=%d"},
25754 +       {Opt_rdblk, "rdblk=%d"},
25755 +       {Opt_rdblk_def, "rdblk=def"},
25756 +       {Opt_rdhash, "rdhash=%d"},
25757 +       {Opt_rdhash_def, "rdhash=def"},
25758 +
25759 +       {Opt_wbr_create, "create=%s"},
25760 +       {Opt_wbr_create, "create_policy=%s"},
25761 +       {Opt_wbr_copyup, "cpup=%s"},
25762 +       {Opt_wbr_copyup, "copyup=%s"},
25763 +       {Opt_wbr_copyup, "copyup_policy=%s"},
25764 +
25765 +       /* generic VFS flag */
25766 +#ifdef CONFIG_FS_POSIX_ACL
25767 +       {Opt_acl, "acl"},
25768 +       {Opt_noacl, "noacl"},
25769 +#else
25770 +       {Opt_ignore, "acl"},
25771 +       {Opt_ignore_silent, "noacl"},
25772 +#endif
25773 +
25774 +       /* internal use for the scripts */
25775 +       {Opt_ignore_silent, "si=%s"},
25776 +
25777 +       {Opt_br, "dirs=%s"},
25778 +       {Opt_ignore, "debug=%d"},
25779 +       {Opt_ignore, "delete=whiteout"},
25780 +       {Opt_ignore, "delete=all"},
25781 +       {Opt_ignore, "imap=%s"},
25782 +
25783 +       /* temporary workaround, due to old mount(8)? */
25784 +       {Opt_ignore_silent, "relatime"},
25785 +
25786 +       {Opt_err, NULL}
25787 +};
25788 +
25789 +/* ---------------------------------------------------------------------- */
25790 +
25791 +static const char *au_parser_pattern(int val, match_table_t tbl)
25792 +{
25793 +       struct match_token *p;
25794 +
25795 +       p = tbl;
25796 +       while (p->pattern) {
25797 +               if (p->token == val)
25798 +                       return p->pattern;
25799 +               p++;
25800 +       }
25801 +       BUG();
25802 +       return "??";
25803 +}
25804 +
25805 +static const char *au_optstr(int *val, match_table_t tbl)
25806 +{
25807 +       struct match_token *p;
25808 +       int v;
25809 +
25810 +       v = *val;
25811 +       if (!v)
25812 +               goto out;
25813 +       p = tbl;
25814 +       while (p->pattern) {
25815 +               if (p->token
25816 +                   && (v & p->token) == p->token) {
25817 +                       *val &= ~p->token;
25818 +                       return p->pattern;
25819 +               }
25820 +               p++;
25821 +       }
25822 +
25823 +out:
25824 +       return NULL;
25825 +}
25826 +
25827 +/* ---------------------------------------------------------------------- */
25828 +
25829 +static match_table_t brperm = {
25830 +       {AuBrPerm_RO, AUFS_BRPERM_RO},
25831 +       {AuBrPerm_RR, AUFS_BRPERM_RR},
25832 +       {AuBrPerm_RW, AUFS_BRPERM_RW},
25833 +       {0, NULL}
25834 +};
25835 +
25836 +static match_table_t brattr = {
25837 +       /* general */
25838 +       {AuBrAttr_COO_REG, AUFS_BRATTR_COO_REG},
25839 +       {AuBrAttr_COO_ALL, AUFS_BRATTR_COO_ALL},
25840 +       /* 'unpin' attrib is meaningless since linux-3.18-rc1 */
25841 +       {AuBrAttr_UNPIN, AUFS_BRATTR_UNPIN},
25842 +#ifdef CONFIG_AUFS_FHSM
25843 +       {AuBrAttr_FHSM, AUFS_BRATTR_FHSM},
25844 +#endif
25845 +#ifdef CONFIG_AUFS_XATTR
25846 +       {AuBrAttr_ICEX, AUFS_BRATTR_ICEX},
25847 +       {AuBrAttr_ICEX_SEC, AUFS_BRATTR_ICEX_SEC},
25848 +       {AuBrAttr_ICEX_SYS, AUFS_BRATTR_ICEX_SYS},
25849 +       {AuBrAttr_ICEX_TR, AUFS_BRATTR_ICEX_TR},
25850 +       {AuBrAttr_ICEX_USR, AUFS_BRATTR_ICEX_USR},
25851 +       {AuBrAttr_ICEX_OTH, AUFS_BRATTR_ICEX_OTH},
25852 +#endif
25853 +
25854 +       /* ro/rr branch */
25855 +       {AuBrRAttr_WH, AUFS_BRRATTR_WH},
25856 +
25857 +       /* rw branch */
25858 +       {AuBrWAttr_MOO, AUFS_BRWATTR_MOO},
25859 +       {AuBrWAttr_NoLinkWH, AUFS_BRWATTR_NLWH},
25860 +
25861 +       {0, NULL}
25862 +};
25863 +
25864 +static int br_attr_val(char *str, match_table_t table, substring_t args[])
25865 +{
25866 +       int attr, v;
25867 +       char *p;
25868 +
25869 +       attr = 0;
25870 +       do {
25871 +               p = strchr(str, '+');
25872 +               if (p)
25873 +                       *p = 0;
25874 +               v = match_token(str, table, args);
25875 +               if (v) {
25876 +                       if (v & AuBrAttr_CMOO_Mask)
25877 +                               attr &= ~AuBrAttr_CMOO_Mask;
25878 +                       attr |= v;
25879 +               } else {
25880 +                       if (p)
25881 +                               *p = '+';
25882 +                       pr_warn("ignored branch attribute %s\n", str);
25883 +                       break;
25884 +               }
25885 +               if (p)
25886 +                       str = p + 1;
25887 +       } while (p);
25888 +
25889 +       return attr;
25890 +}
25891 +
25892 +static int au_do_optstr_br_attr(au_br_perm_str_t *str, int perm)
25893 +{
25894 +       int sz;
25895 +       const char *p;
25896 +       char *q;
25897 +
25898 +       q = str->a;
25899 +       *q = 0;
25900 +       p = au_optstr(&perm, brattr);
25901 +       if (p) {
25902 +               sz = strlen(p);
25903 +               memcpy(q, p, sz + 1);
25904 +               q += sz;
25905 +       } else
25906 +               goto out;
25907 +
25908 +       do {
25909 +               p = au_optstr(&perm, brattr);
25910 +               if (p) {
25911 +                       *q++ = '+';
25912 +                       sz = strlen(p);
25913 +                       memcpy(q, p, sz + 1);
25914 +                       q += sz;
25915 +               }
25916 +       } while (p);
25917 +
25918 +out:
25919 +       return q - str->a;
25920 +}
25921 +
25922 +static int noinline_for_stack br_perm_val(char *perm)
25923 +{
25924 +       int val, bad, sz;
25925 +       char *p;
25926 +       substring_t args[MAX_OPT_ARGS];
25927 +       au_br_perm_str_t attr;
25928 +
25929 +       p = strchr(perm, '+');
25930 +       if (p)
25931 +               *p = 0;
25932 +       val = match_token(perm, brperm, args);
25933 +       if (!val) {
25934 +               if (p)
25935 +                       *p = '+';
25936 +               pr_warn("ignored branch permission %s\n", perm);
25937 +               val = AuBrPerm_RO;
25938 +               goto out;
25939 +       }
25940 +       if (!p)
25941 +               goto out;
25942 +
25943 +       val |= br_attr_val(p + 1, brattr, args);
25944 +
25945 +       bad = 0;
25946 +       switch (val & AuBrPerm_Mask) {
25947 +       case AuBrPerm_RO:
25948 +       case AuBrPerm_RR:
25949 +               bad = val & AuBrWAttr_Mask;
25950 +               val &= ~AuBrWAttr_Mask;
25951 +               break;
25952 +       case AuBrPerm_RW:
25953 +               bad = val & AuBrRAttr_Mask;
25954 +               val &= ~AuBrRAttr_Mask;
25955 +               break;
25956 +       }
25957 +
25958 +       /*
25959 +        * 'unpin' attrib becomes meaningless since linux-3.18-rc1, but aufs
25960 +        * does not treat it as an error, just warning.
25961 +        * this is a tiny guard for the user operation.
25962 +        */
25963 +       if (val & AuBrAttr_UNPIN) {
25964 +               bad |= AuBrAttr_UNPIN;
25965 +               val &= ~AuBrAttr_UNPIN;
25966 +       }
25967 +
25968 +       if (unlikely(bad)) {
25969 +               sz = au_do_optstr_br_attr(&attr, bad);
25970 +               AuDebugOn(!sz);
25971 +               pr_warn("ignored branch attribute %s\n", attr.a);
25972 +       }
25973 +
25974 +out:
25975 +       return val;
25976 +}
25977 +
25978 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm)
25979 +{
25980 +       au_br_perm_str_t attr;
25981 +       const char *p;
25982 +       char *q;
25983 +       int sz;
25984 +
25985 +       q = str->a;
25986 +       p = au_optstr(&perm, brperm);
25987 +       AuDebugOn(!p || !*p);
25988 +       sz = strlen(p);
25989 +       memcpy(q, p, sz + 1);
25990 +       q += sz;
25991 +
25992 +       sz = au_do_optstr_br_attr(&attr, perm);
25993 +       if (sz) {
25994 +               *q++ = '+';
25995 +               memcpy(q, attr.a, sz + 1);
25996 +       }
25997 +
25998 +       AuDebugOn(strlen(str->a) >= sizeof(str->a));
25999 +}
26000 +
26001 +/* ---------------------------------------------------------------------- */
26002 +
26003 +static match_table_t udbalevel = {
26004 +       {AuOpt_UDBA_REVAL, "reval"},
26005 +       {AuOpt_UDBA_NONE, "none"},
26006 +#ifdef CONFIG_AUFS_HNOTIFY
26007 +       {AuOpt_UDBA_HNOTIFY, "notify"}, /* abstraction */
26008 +#ifdef CONFIG_AUFS_HFSNOTIFY
26009 +       {AuOpt_UDBA_HNOTIFY, "fsnotify"},
26010 +#endif
26011 +#endif
26012 +       {-1, NULL}
26013 +};
26014 +
26015 +static int noinline_for_stack udba_val(char *str)
26016 +{
26017 +       substring_t args[MAX_OPT_ARGS];
26018 +
26019 +       return match_token(str, udbalevel, args);
26020 +}
26021 +
26022 +const char *au_optstr_udba(int udba)
26023 +{
26024 +       return au_parser_pattern(udba, udbalevel);
26025 +}
26026 +
26027 +/* ---------------------------------------------------------------------- */
26028 +
26029 +static match_table_t au_wbr_create_policy = {
26030 +       {AuWbrCreate_TDP, "tdp"},
26031 +       {AuWbrCreate_TDP, "top-down-parent"},
26032 +       {AuWbrCreate_RR, "rr"},
26033 +       {AuWbrCreate_RR, "round-robin"},
26034 +       {AuWbrCreate_MFS, "mfs"},
26035 +       {AuWbrCreate_MFS, "most-free-space"},
26036 +       {AuWbrCreate_MFSV, "mfs:%d"},
26037 +       {AuWbrCreate_MFSV, "most-free-space:%d"},
26038 +
26039 +       /* top-down regardless the parent, and then mfs */
26040 +       {AuWbrCreate_TDMFS, "tdmfs:%d"},
26041 +       {AuWbrCreate_TDMFSV, "tdmfs:%d:%d"},
26042 +
26043 +       {AuWbrCreate_MFSRR, "mfsrr:%d"},
26044 +       {AuWbrCreate_MFSRRV, "mfsrr:%d:%d"},
26045 +       {AuWbrCreate_PMFS, "pmfs"},
26046 +       {AuWbrCreate_PMFSV, "pmfs:%d"},
26047 +       {AuWbrCreate_PMFSRR, "pmfsrr:%d"},
26048 +       {AuWbrCreate_PMFSRRV, "pmfsrr:%d:%d"},
26049 +
26050 +       {-1, NULL}
26051 +};
26052 +
26053 +static int au_wbr_mfs_wmark(substring_t *arg, char *str,
26054 +                           struct au_opt_wbr_create *create)
26055 +{
26056 +       int err;
26057 +       unsigned long long ull;
26058 +
26059 +       err = 0;
26060 +       if (!match_u64(arg, &ull))
26061 +               create->mfsrr_watermark = ull;
26062 +       else {
26063 +               pr_err("bad integer in %s\n", str);
26064 +               err = -EINVAL;
26065 +       }
26066 +
26067 +       return err;
26068 +}
26069 +
26070 +static int au_wbr_mfs_sec(substring_t *arg, char *str,
26071 +                         struct au_opt_wbr_create *create)
26072 +{
26073 +       int n, err;
26074 +
26075 +       err = 0;
26076 +       if (!match_int(arg, &n) && 0 <= n && n <= AUFS_MFS_MAX_SEC)
26077 +               create->mfs_second = n;
26078 +       else {
26079 +               pr_err("bad integer in %s\n", str);
26080 +               err = -EINVAL;
26081 +       }
26082 +
26083 +       return err;
26084 +}
26085 +
26086 +static int noinline_for_stack
26087 +au_wbr_create_val(char *str, struct au_opt_wbr_create *create)
26088 +{
26089 +       int err, e;
26090 +       substring_t args[MAX_OPT_ARGS];
26091 +
26092 +       err = match_token(str, au_wbr_create_policy, args);
26093 +       create->wbr_create = err;
26094 +       switch (err) {
26095 +       case AuWbrCreate_MFSRRV:
26096 +       case AuWbrCreate_TDMFSV:
26097 +       case AuWbrCreate_PMFSRRV:
26098 +               e = au_wbr_mfs_wmark(&args[0], str, create);
26099 +               if (!e)
26100 +                       e = au_wbr_mfs_sec(&args[1], str, create);
26101 +               if (unlikely(e))
26102 +                       err = e;
26103 +               break;
26104 +       case AuWbrCreate_MFSRR:
26105 +       case AuWbrCreate_TDMFS:
26106 +       case AuWbrCreate_PMFSRR:
26107 +               e = au_wbr_mfs_wmark(&args[0], str, create);
26108 +               if (unlikely(e)) {
26109 +                       err = e;
26110 +                       break;
26111 +               }
26112 +               fallthrough;
26113 +       case AuWbrCreate_MFS:
26114 +       case AuWbrCreate_PMFS:
26115 +               create->mfs_second = AUFS_MFS_DEF_SEC;
26116 +               break;
26117 +       case AuWbrCreate_MFSV:
26118 +       case AuWbrCreate_PMFSV:
26119 +               e = au_wbr_mfs_sec(&args[0], str, create);
26120 +               if (unlikely(e))
26121 +                       err = e;
26122 +               break;
26123 +       }
26124 +
26125 +       return err;
26126 +}
26127 +
26128 +const char *au_optstr_wbr_create(int wbr_create)
26129 +{
26130 +       return au_parser_pattern(wbr_create, au_wbr_create_policy);
26131 +}
26132 +
26133 +static match_table_t au_wbr_copyup_policy = {
26134 +       {AuWbrCopyup_TDP, "tdp"},
26135 +       {AuWbrCopyup_TDP, "top-down-parent"},
26136 +       {AuWbrCopyup_BUP, "bup"},
26137 +       {AuWbrCopyup_BUP, "bottom-up-parent"},
26138 +       {AuWbrCopyup_BU, "bu"},
26139 +       {AuWbrCopyup_BU, "bottom-up"},
26140 +       {-1, NULL}
26141 +};
26142 +
26143 +static int noinline_for_stack au_wbr_copyup_val(char *str)
26144 +{
26145 +       substring_t args[MAX_OPT_ARGS];
26146 +
26147 +       return match_token(str, au_wbr_copyup_policy, args);
26148 +}
26149 +
26150 +const char *au_optstr_wbr_copyup(int wbr_copyup)
26151 +{
26152 +       return au_parser_pattern(wbr_copyup, au_wbr_copyup_policy);
26153 +}
26154 +
26155 +/* ---------------------------------------------------------------------- */
26156 +
26157 +static const int lkup_dirflags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
26158 +
26159 +static void dump_opts(struct au_opts *opts)
26160 +{
26161 +#ifdef CONFIG_AUFS_DEBUG
26162 +       /* reduce stack space */
26163 +       union {
26164 +               struct au_opt_add *add;
26165 +               struct au_opt_del *del;
26166 +               struct au_opt_mod *mod;
26167 +               struct au_opt_xino *xino;
26168 +               struct au_opt_xino_itrunc *xino_itrunc;
26169 +               struct au_opt_wbr_create *create;
26170 +       } u;
26171 +       struct au_opt *opt;
26172 +
26173 +       opt = opts->opt;
26174 +       while (opt->type != Opt_tail) {
26175 +               switch (opt->type) {
26176 +               case Opt_add:
26177 +                       u.add = &opt->add;
26178 +                       AuDbg("add {b%d, %s, 0x%x, %p}\n",
26179 +                                 u.add->bindex, u.add->pathname, u.add->perm,
26180 +                                 u.add->path.dentry);
26181 +                       break;
26182 +               case Opt_del:
26183 +               case Opt_idel:
26184 +                       u.del = &opt->del;
26185 +                       AuDbg("del {%s, %p}\n",
26186 +                             u.del->pathname, u.del->h_path.dentry);
26187 +                       break;
26188 +               case Opt_mod:
26189 +               case Opt_imod:
26190 +                       u.mod = &opt->mod;
26191 +                       AuDbg("mod {%s, 0x%x, %p}\n",
26192 +                                 u.mod->path, u.mod->perm, u.mod->h_root);
26193 +                       break;
26194 +               case Opt_append:
26195 +                       u.add = &opt->add;
26196 +                       AuDbg("append {b%d, %s, 0x%x, %p}\n",
26197 +                                 u.add->bindex, u.add->pathname, u.add->perm,
26198 +                                 u.add->path.dentry);
26199 +                       break;
26200 +               case Opt_prepend:
26201 +                       u.add = &opt->add;
26202 +                       AuDbg("prepend {b%d, %s, 0x%x, %p}\n",
26203 +                                 u.add->bindex, u.add->pathname, u.add->perm,
26204 +                                 u.add->path.dentry);
26205 +                       break;
26206 +               case Opt_dirwh:
26207 +                       AuDbg("dirwh %d\n", opt->dirwh);
26208 +                       break;
26209 +               case Opt_rdcache:
26210 +                       AuDbg("rdcache %d\n", opt->rdcache);
26211 +                       break;
26212 +               case Opt_rdblk:
26213 +                       AuDbg("rdblk %u\n", opt->rdblk);
26214 +                       break;
26215 +               case Opt_rdblk_def:
26216 +                       AuDbg("rdblk_def\n");
26217 +                       break;
26218 +               case Opt_rdhash:
26219 +                       AuDbg("rdhash %u\n", opt->rdhash);
26220 +                       break;
26221 +               case Opt_rdhash_def:
26222 +                       AuDbg("rdhash_def\n");
26223 +                       break;
26224 +               case Opt_xino:
26225 +                       u.xino = &opt->xino;
26226 +                       AuDbg("xino {%s %pD}\n", u.xino->path, u.xino->file);
26227 +                       break;
26228 +               case Opt_trunc_xino:
26229 +                       AuLabel(trunc_xino);
26230 +                       break;
26231 +               case Opt_notrunc_xino:
26232 +                       AuLabel(notrunc_xino);
26233 +                       break;
26234 +               case Opt_trunc_xino_path:
26235 +               case Opt_itrunc_xino:
26236 +                       u.xino_itrunc = &opt->xino_itrunc;
26237 +                       AuDbg("trunc_xino %d\n", u.xino_itrunc->bindex);
26238 +                       break;
26239 +               case Opt_noxino:
26240 +                       AuLabel(noxino);
26241 +                       break;
26242 +               case Opt_trunc_xib:
26243 +                       AuLabel(trunc_xib);
26244 +                       break;
26245 +               case Opt_notrunc_xib:
26246 +                       AuLabel(notrunc_xib);
26247 +                       break;
26248 +               case Opt_shwh:
26249 +                       AuLabel(shwh);
26250 +                       break;
26251 +               case Opt_noshwh:
26252 +                       AuLabel(noshwh);
26253 +                       break;
26254 +               case Opt_dirperm1:
26255 +                       AuLabel(dirperm1);
26256 +                       break;
26257 +               case Opt_nodirperm1:
26258 +                       AuLabel(nodirperm1);
26259 +                       break;
26260 +               case Opt_plink:
26261 +                       AuLabel(plink);
26262 +                       break;
26263 +               case Opt_noplink:
26264 +                       AuLabel(noplink);
26265 +                       break;
26266 +               case Opt_list_plink:
26267 +                       AuLabel(list_plink);
26268 +                       break;
26269 +               case Opt_udba:
26270 +                       AuDbg("udba %d, %s\n",
26271 +                                 opt->udba, au_optstr_udba(opt->udba));
26272 +                       break;
26273 +               case Opt_dio:
26274 +                       AuLabel(dio);
26275 +                       break;
26276 +               case Opt_nodio:
26277 +                       AuLabel(nodio);
26278 +                       break;
26279 +               case Opt_diropq_a:
26280 +                       AuLabel(diropq_a);
26281 +                       break;
26282 +               case Opt_diropq_w:
26283 +                       AuLabel(diropq_w);
26284 +                       break;
26285 +               case Opt_warn_perm:
26286 +                       AuLabel(warn_perm);
26287 +                       break;
26288 +               case Opt_nowarn_perm:
26289 +                       AuLabel(nowarn_perm);
26290 +                       break;
26291 +               case Opt_verbose:
26292 +                       AuLabel(verbose);
26293 +                       break;
26294 +               case Opt_noverbose:
26295 +                       AuLabel(noverbose);
26296 +                       break;
26297 +               case Opt_sum:
26298 +                       AuLabel(sum);
26299 +                       break;
26300 +               case Opt_nosum:
26301 +                       AuLabel(nosum);
26302 +                       break;
26303 +               case Opt_wsum:
26304 +                       AuLabel(wsum);
26305 +                       break;
26306 +               case Opt_wbr_create:
26307 +                       u.create = &opt->wbr_create;
26308 +                       AuDbg("create %d, %s\n", u.create->wbr_create,
26309 +                                 au_optstr_wbr_create(u.create->wbr_create));
26310 +                       switch (u.create->wbr_create) {
26311 +                       case AuWbrCreate_MFSV:
26312 +                       case AuWbrCreate_PMFSV:
26313 +                               AuDbg("%d sec\n", u.create->mfs_second);
26314 +                               break;
26315 +                       case AuWbrCreate_MFSRR:
26316 +                       case AuWbrCreate_TDMFS:
26317 +                               AuDbg("%llu watermark\n",
26318 +                                         u.create->mfsrr_watermark);
26319 +                               break;
26320 +                       case AuWbrCreate_MFSRRV:
26321 +                       case AuWbrCreate_TDMFSV:
26322 +                       case AuWbrCreate_PMFSRRV:
26323 +                               AuDbg("%llu watermark, %d sec\n",
26324 +                                         u.create->mfsrr_watermark,
26325 +                                         u.create->mfs_second);
26326 +                               break;
26327 +                       }
26328 +                       break;
26329 +               case Opt_wbr_copyup:
26330 +                       AuDbg("copyup %d, %s\n", opt->wbr_copyup,
26331 +                                 au_optstr_wbr_copyup(opt->wbr_copyup));
26332 +                       break;
26333 +               case Opt_fhsm_sec:
26334 +                       AuDbg("fhsm_sec %u\n", opt->fhsm_second);
26335 +                       break;
26336 +               case Opt_dirren:
26337 +                       AuLabel(dirren);
26338 +                       break;
26339 +               case Opt_nodirren:
26340 +                       AuLabel(nodirren);
26341 +                       break;
26342 +               case Opt_acl:
26343 +                       AuLabel(acl);
26344 +                       break;
26345 +               case Opt_noacl:
26346 +                       AuLabel(noacl);
26347 +                       break;
26348 +               default:
26349 +                       BUG();
26350 +               }
26351 +               opt++;
26352 +       }
26353 +#endif
26354 +}
26355 +
26356 +void au_opts_free(struct au_opts *opts)
26357 +{
26358 +       struct au_opt *opt;
26359 +
26360 +       opt = opts->opt;
26361 +       while (opt->type != Opt_tail) {
26362 +               switch (opt->type) {
26363 +               case Opt_add:
26364 +               case Opt_append:
26365 +               case Opt_prepend:
26366 +                       path_put(&opt->add.path);
26367 +                       break;
26368 +               case Opt_del:
26369 +               case Opt_idel:
26370 +                       path_put(&opt->del.h_path);
26371 +                       break;
26372 +               case Opt_mod:
26373 +               case Opt_imod:
26374 +                       dput(opt->mod.h_root);
26375 +                       break;
26376 +               case Opt_xino:
26377 +                       fput(opt->xino.file);
26378 +                       break;
26379 +               }
26380 +               opt++;
26381 +       }
26382 +}
26383 +
26384 +static int opt_add(struct au_opt *opt, char *opt_str, unsigned long sb_flags,
26385 +                  aufs_bindex_t bindex)
26386 +{
26387 +       int err;
26388 +       struct au_opt_add *add = &opt->add;
26389 +       char *p;
26390 +
26391 +       add->bindex = bindex;
26392 +       add->perm = AuBrPerm_RO;
26393 +       add->pathname = opt_str;
26394 +       p = strchr(opt_str, '=');
26395 +       if (p) {
26396 +               *p++ = 0;
26397 +               if (*p)
26398 +                       add->perm = br_perm_val(p);
26399 +       }
26400 +
26401 +       err = vfsub_kern_path(add->pathname, lkup_dirflags, &add->path);
26402 +       if (!err) {
26403 +               if (!p) {
26404 +                       add->perm = AuBrPerm_RO;
26405 +                       if (au_test_fs_rr(add->path.dentry->d_sb))
26406 +                               add->perm = AuBrPerm_RR;
26407 +                       else if (!bindex && !(sb_flags & SB_RDONLY))
26408 +                               add->perm = AuBrPerm_RW;
26409 +               }
26410 +               opt->type = Opt_add;
26411 +               goto out;
26412 +       }
26413 +       pr_err("lookup failed %s (%d)\n", add->pathname, err);
26414 +       err = -EINVAL;
26415 +
26416 +out:
26417 +       return err;
26418 +}
26419 +
26420 +static int au_opts_parse_del(struct au_opt_del *del, substring_t args[])
26421 +{
26422 +       int err;
26423 +
26424 +       del->pathname = args[0].from;
26425 +       AuDbg("del path %s\n", del->pathname);
26426 +
26427 +       err = vfsub_kern_path(del->pathname, lkup_dirflags, &del->h_path);
26428 +       if (unlikely(err))
26429 +               pr_err("lookup failed %s (%d)\n", del->pathname, err);
26430 +
26431 +       return err;
26432 +}
26433 +
26434 +#if 0 /* reserved for future use */
26435 +static int au_opts_parse_idel(struct super_block *sb, aufs_bindex_t bindex,
26436 +                             struct au_opt_del *del, substring_t args[])
26437 +{
26438 +       int err;
26439 +       struct dentry *root;
26440 +
26441 +       err = -EINVAL;
26442 +       root = sb->s_root;
26443 +       aufs_read_lock(root, AuLock_FLUSH);
26444 +       if (bindex < 0 || au_sbbot(sb) < bindex) {
26445 +               pr_err("out of bounds, %d\n", bindex);
26446 +               goto out;
26447 +       }
26448 +
26449 +       err = 0;
26450 +       del->h_path.dentry = dget(au_h_dptr(root, bindex));
26451 +       del->h_path.mnt = mntget(au_sbr_mnt(sb, bindex));
26452 +
26453 +out:
26454 +       aufs_read_unlock(root, !AuLock_IR);
26455 +       return err;
26456 +}
26457 +#endif
26458 +
26459 +static int noinline_for_stack
26460 +au_opts_parse_mod(struct au_opt_mod *mod, substring_t args[])
26461 +{
26462 +       int err;
26463 +       struct path path;
26464 +       char *p;
26465 +
26466 +       err = -EINVAL;
26467 +       mod->path = args[0].from;
26468 +       p = strchr(mod->path, '=');
26469 +       if (unlikely(!p)) {
26470 +               pr_err("no permission %s\n", args[0].from);
26471 +               goto out;
26472 +       }
26473 +
26474 +       *p++ = 0;
26475 +       err = vfsub_kern_path(mod->path, lkup_dirflags, &path);
26476 +       if (unlikely(err)) {
26477 +               pr_err("lookup failed %s (%d)\n", mod->path, err);
26478 +               goto out;
26479 +       }
26480 +
26481 +       mod->perm = br_perm_val(p);
26482 +       AuDbg("mod path %s, perm 0x%x, %s\n", mod->path, mod->perm, p);
26483 +       mod->h_root = dget(path.dentry);
26484 +       path_put(&path);
26485 +
26486 +out:
26487 +       return err;
26488 +}
26489 +
26490 +#if 0 /* reserved for future use */
26491 +static int au_opts_parse_imod(struct super_block *sb, aufs_bindex_t bindex,
26492 +                             struct au_opt_mod *mod, substring_t args[])
26493 +{
26494 +       int err;
26495 +       struct dentry *root;
26496 +
26497 +       err = -EINVAL;
26498 +       root = sb->s_root;
26499 +       aufs_read_lock(root, AuLock_FLUSH);
26500 +       if (bindex < 0 || au_sbbot(sb) < bindex) {
26501 +               pr_err("out of bounds, %d\n", bindex);
26502 +               goto out;
26503 +       }
26504 +
26505 +       err = 0;
26506 +       mod->perm = br_perm_val(args[1].from);
26507 +       AuDbg("mod path %s, perm 0x%x, %s\n",
26508 +             mod->path, mod->perm, args[1].from);
26509 +       mod->h_root = dget(au_h_dptr(root, bindex));
26510 +
26511 +out:
26512 +       aufs_read_unlock(root, !AuLock_IR);
26513 +       return err;
26514 +}
26515 +#endif
26516 +
26517 +static int au_opts_parse_xino(struct super_block *sb, struct au_opt_xino *xino,
26518 +                             substring_t args[])
26519 +{
26520 +       int err;
26521 +       struct file *file;
26522 +
26523 +       file = au_xino_create(sb, args[0].from, /*silent*/0, /*wbrtop*/0);
26524 +       err = PTR_ERR(file);
26525 +       if (IS_ERR(file))
26526 +               goto out;
26527 +
26528 +       err = -EINVAL;
26529 +       if (unlikely(file->f_path.dentry->d_sb == sb)) {
26530 +               fput(file);
26531 +               pr_err("%s must be outside\n", args[0].from);
26532 +               goto out;
26533 +       }
26534 +
26535 +       err = 0;
26536 +       xino->file = file;
26537 +       xino->path = args[0].from;
26538 +
26539 +out:
26540 +       return err;
26541 +}
26542 +
26543 +static int noinline_for_stack
26544 +au_opts_parse_xino_itrunc_path(struct super_block *sb,
26545 +                              struct au_opt_xino_itrunc *xino_itrunc,
26546 +                              substring_t args[])
26547 +{
26548 +       int err;
26549 +       aufs_bindex_t bbot, bindex;
26550 +       struct path path;
26551 +       struct dentry *root;
26552 +
26553 +       err = vfsub_kern_path(args[0].from, lkup_dirflags, &path);
26554 +       if (unlikely(err)) {
26555 +               pr_err("lookup failed %s (%d)\n", args[0].from, err);
26556 +               goto out;
26557 +       }
26558 +
26559 +       xino_itrunc->bindex = -1;
26560 +       root = sb->s_root;
26561 +       aufs_read_lock(root, AuLock_FLUSH);
26562 +       bbot = au_sbbot(sb);
26563 +       for (bindex = 0; bindex <= bbot; bindex++) {
26564 +               if (au_h_dptr(root, bindex) == path.dentry) {
26565 +                       xino_itrunc->bindex = bindex;
26566 +                       break;
26567 +               }
26568 +       }
26569 +       aufs_read_unlock(root, !AuLock_IR);
26570 +       path_put(&path);
26571 +
26572 +       if (unlikely(xino_itrunc->bindex < 0)) {
26573 +               pr_err("no such branch %s\n", args[0].from);
26574 +               err = -EINVAL;
26575 +       }
26576 +
26577 +out:
26578 +       return err;
26579 +}
26580 +
26581 +/* called without aufs lock */
26582 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
26583 +{
26584 +       int err, n, token;
26585 +       aufs_bindex_t bindex;
26586 +       unsigned char skipped;
26587 +       struct dentry *root;
26588 +       struct au_opt *opt, *opt_tail;
26589 +       char *opt_str;
26590 +       /* reduce the stack space */
26591 +       union {
26592 +               struct au_opt_xino_itrunc *xino_itrunc;
26593 +               struct au_opt_wbr_create *create;
26594 +       } u;
26595 +       struct {
26596 +               substring_t args[MAX_OPT_ARGS];
26597 +       } *a;
26598 +
26599 +       err = -ENOMEM;
26600 +       a = kmalloc(sizeof(*a), GFP_NOFS);
26601 +       if (unlikely(!a))
26602 +               goto out;
26603 +
26604 +       root = sb->s_root;
26605 +       err = 0;
26606 +       bindex = 0;
26607 +       opt = opts->opt;
26608 +       opt_tail = opt + opts->max_opt - 1;
26609 +       opt->type = Opt_tail;
26610 +       while (!err && (opt_str = strsep(&str, ",")) && *opt_str) {
26611 +               err = -EINVAL;
26612 +               skipped = 0;
26613 +               token = match_token(opt_str, options, a->args);
26614 +               switch (token) {
26615 +               case Opt_br:
26616 +                       err = 0;
26617 +                       while (!err && (opt_str = strsep(&a->args[0].from, ":"))
26618 +                              && *opt_str) {
26619 +                               err = opt_add(opt, opt_str, opts->sb_flags,
26620 +                                             bindex++);
26621 +                               if (unlikely(!err && ++opt > opt_tail)) {
26622 +                                       err = -E2BIG;
26623 +                                       break;
26624 +                               }
26625 +                               opt->type = Opt_tail;
26626 +                               skipped = 1;
26627 +                       }
26628 +                       break;
26629 +               case Opt_add:
26630 +                       if (unlikely(match_int(&a->args[0], &n))) {
26631 +                               pr_err("bad integer in %s\n", opt_str);
26632 +                               break;
26633 +                       }
26634 +                       bindex = n;
26635 +                       err = opt_add(opt, a->args[1].from, opts->sb_flags,
26636 +                                     bindex);
26637 +                       if (!err)
26638 +                               opt->type = token;
26639 +                       break;
26640 +               case Opt_append:
26641 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
26642 +                                     /*dummy bindex*/1);
26643 +                       if (!err)
26644 +                               opt->type = token;
26645 +                       break;
26646 +               case Opt_prepend:
26647 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
26648 +                                     /*bindex*/0);
26649 +                       if (!err)
26650 +                               opt->type = token;
26651 +                       break;
26652 +               case Opt_del:
26653 +                       err = au_opts_parse_del(&opt->del, a->args);
26654 +                       if (!err)
26655 +                               opt->type = token;
26656 +                       break;
26657 +#if 0 /* reserved for future use */
26658 +               case Opt_idel:
26659 +                       del->pathname = "(indexed)";
26660 +                       if (unlikely(match_int(&args[0], &n))) {
26661 +                               pr_err("bad integer in %s\n", opt_str);
26662 +                               break;
26663 +                       }
26664 +                       err = au_opts_parse_idel(sb, n, &opt->del, a->args);
26665 +                       if (!err)
26666 +                               opt->type = token;
26667 +                       break;
26668 +#endif
26669 +               case Opt_mod:
26670 +                       err = au_opts_parse_mod(&opt->mod, a->args);
26671 +                       if (!err)
26672 +                               opt->type = token;
26673 +                       break;
26674 +#ifdef IMOD /* reserved for future use */
26675 +               case Opt_imod:
26676 +                       u.mod->path = "(indexed)";
26677 +                       if (unlikely(match_int(&a->args[0], &n))) {
26678 +                               pr_err("bad integer in %s\n", opt_str);
26679 +                               break;
26680 +                       }
26681 +                       err = au_opts_parse_imod(sb, n, &opt->mod, a->args);
26682 +                       if (!err)
26683 +                               opt->type = token;
26684 +                       break;
26685 +#endif
26686 +               case Opt_xino:
26687 +                       err = au_opts_parse_xino(sb, &opt->xino, a->args);
26688 +                       if (!err)
26689 +                               opt->type = token;
26690 +                       break;
26691 +
26692 +               case Opt_trunc_xino_path:
26693 +                       err = au_opts_parse_xino_itrunc_path
26694 +                               (sb, &opt->xino_itrunc, a->args);
26695 +                       if (!err)
26696 +                               opt->type = token;
26697 +                       break;
26698 +
26699 +               case Opt_itrunc_xino:
26700 +                       u.xino_itrunc = &opt->xino_itrunc;
26701 +                       if (unlikely(match_int(&a->args[0], &n))) {
26702 +                               pr_err("bad integer in %s\n", opt_str);
26703 +                               break;
26704 +                       }
26705 +                       u.xino_itrunc->bindex = n;
26706 +                       aufs_read_lock(root, AuLock_FLUSH);
26707 +                       if (n < 0 || au_sbbot(sb) < n) {
26708 +                               pr_err("out of bounds, %d\n", n);
26709 +                               aufs_read_unlock(root, !AuLock_IR);
26710 +                               break;
26711 +                       }
26712 +                       aufs_read_unlock(root, !AuLock_IR);
26713 +                       err = 0;
26714 +                       opt->type = token;
26715 +                       break;
26716 +
26717 +               case Opt_dirwh:
26718 +                       if (unlikely(match_int(&a->args[0], &opt->dirwh)))
26719 +                               break;
26720 +                       err = 0;
26721 +                       opt->type = token;
26722 +                       break;
26723 +
26724 +               case Opt_rdcache:
26725 +                       if (unlikely(match_int(&a->args[0], &n))) {
26726 +                               pr_err("bad integer in %s\n", opt_str);
26727 +                               break;
26728 +                       }
26729 +                       if (unlikely(n > AUFS_RDCACHE_MAX)) {
26730 +                               pr_err("rdcache must be smaller than %d\n",
26731 +                                      AUFS_RDCACHE_MAX);
26732 +                               break;
26733 +                       }
26734 +                       opt->rdcache = n;
26735 +                       err = 0;
26736 +                       opt->type = token;
26737 +                       break;
26738 +               case Opt_rdblk:
26739 +                       if (unlikely(match_int(&a->args[0], &n)
26740 +                                    || n < 0
26741 +                                    || n > KMALLOC_MAX_SIZE)) {
26742 +                               pr_err("bad integer in %s\n", opt_str);
26743 +                               break;
26744 +                       }
26745 +                       if (unlikely(n && n < NAME_MAX)) {
26746 +                               pr_err("rdblk must be larger than %d\n",
26747 +                                      NAME_MAX);
26748 +                               break;
26749 +                       }
26750 +                       opt->rdblk = n;
26751 +                       err = 0;
26752 +                       opt->type = token;
26753 +                       break;
26754 +               case Opt_rdhash:
26755 +                       if (unlikely(match_int(&a->args[0], &n)
26756 +                                    || n < 0
26757 +                                    || n * sizeof(struct hlist_head)
26758 +                                    > KMALLOC_MAX_SIZE)) {
26759 +                               pr_err("bad integer in %s\n", opt_str);
26760 +                               break;
26761 +                       }
26762 +                       opt->rdhash = n;
26763 +                       err = 0;
26764 +                       opt->type = token;
26765 +                       break;
26766 +
26767 +               case Opt_trunc_xino:
26768 +               case Opt_notrunc_xino:
26769 +               case Opt_noxino:
26770 +               case Opt_trunc_xib:
26771 +               case Opt_notrunc_xib:
26772 +               case Opt_shwh:
26773 +               case Opt_noshwh:
26774 +               case Opt_dirperm1:
26775 +               case Opt_nodirperm1:
26776 +               case Opt_plink:
26777 +               case Opt_noplink:
26778 +               case Opt_list_plink:
26779 +               case Opt_dio:
26780 +               case Opt_nodio:
26781 +               case Opt_diropq_a:
26782 +               case Opt_diropq_w:
26783 +               case Opt_warn_perm:
26784 +               case Opt_nowarn_perm:
26785 +               case Opt_verbose:
26786 +               case Opt_noverbose:
26787 +               case Opt_sum:
26788 +               case Opt_nosum:
26789 +               case Opt_wsum:
26790 +               case Opt_rdblk_def:
26791 +               case Opt_rdhash_def:
26792 +               case Opt_dirren:
26793 +               case Opt_nodirren:
26794 +               case Opt_acl:
26795 +               case Opt_noacl:
26796 +                       err = 0;
26797 +                       opt->type = token;
26798 +                       break;
26799 +
26800 +               case Opt_udba:
26801 +                       opt->udba = udba_val(a->args[0].from);
26802 +                       if (opt->udba >= 0) {
26803 +                               err = 0;
26804 +                               opt->type = token;
26805 +                       } else
26806 +                               pr_err("wrong value, %s\n", opt_str);
26807 +                       break;
26808 +
26809 +               case Opt_wbr_create:
26810 +                       u.create = &opt->wbr_create;
26811 +                       u.create->wbr_create
26812 +                               = au_wbr_create_val(a->args[0].from, u.create);
26813 +                       if (u.create->wbr_create >= 0) {
26814 +                               err = 0;
26815 +                               opt->type = token;
26816 +                       } else
26817 +                               pr_err("wrong value, %s\n", opt_str);
26818 +                       break;
26819 +               case Opt_wbr_copyup:
26820 +                       opt->wbr_copyup = au_wbr_copyup_val(a->args[0].from);
26821 +                       if (opt->wbr_copyup >= 0) {
26822 +                               err = 0;
26823 +                               opt->type = token;
26824 +                       } else
26825 +                               pr_err("wrong value, %s\n", opt_str);
26826 +                       break;
26827 +
26828 +               case Opt_fhsm_sec:
26829 +                       if (unlikely(match_int(&a->args[0], &n)
26830 +                                    || n < 0)) {
26831 +                               pr_err("bad integer in %s\n", opt_str);
26832 +                               break;
26833 +                       }
26834 +                       if (sysaufs_brs) {
26835 +                               opt->fhsm_second = n;
26836 +                               opt->type = token;
26837 +                       } else
26838 +                               pr_warn("ignored %s\n", opt_str);
26839 +                       err = 0;
26840 +                       break;
26841 +
26842 +               case Opt_ignore:
26843 +                       pr_warn("ignored %s\n", opt_str);
26844 +                       fallthrough;
26845 +               case Opt_ignore_silent:
26846 +                       skipped = 1;
26847 +                       err = 0;
26848 +                       break;
26849 +               case Opt_err:
26850 +                       pr_err("unknown option %s\n", opt_str);
26851 +                       break;
26852 +               }
26853 +
26854 +               if (!err && !skipped) {
26855 +                       if (unlikely(++opt > opt_tail)) {
26856 +                               err = -E2BIG;
26857 +                               opt--;
26858 +                               opt->type = Opt_tail;
26859 +                               break;
26860 +                       }
26861 +                       opt->type = Opt_tail;
26862 +               }
26863 +       }
26864 +
26865 +       au_kfree_rcu(a);
26866 +       dump_opts(opts);
26867 +       if (unlikely(err))
26868 +               au_opts_free(opts);
26869 +
26870 +out:
26871 +       return err;
26872 +}
26873 +
26874 +static int au_opt_wbr_create(struct super_block *sb,
26875 +                            struct au_opt_wbr_create *create)
26876 +{
26877 +       int err;
26878 +       struct au_sbinfo *sbinfo;
26879 +
26880 +       SiMustWriteLock(sb);
26881 +
26882 +       err = 1; /* handled */
26883 +       sbinfo = au_sbi(sb);
26884 +       if (sbinfo->si_wbr_create_ops->fin) {
26885 +               err = sbinfo->si_wbr_create_ops->fin(sb);
26886 +               if (!err)
26887 +                       err = 1;
26888 +       }
26889 +
26890 +       sbinfo->si_wbr_create = create->wbr_create;
26891 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + create->wbr_create;
26892 +       switch (create->wbr_create) {
26893 +       case AuWbrCreate_MFSRRV:
26894 +       case AuWbrCreate_MFSRR:
26895 +       case AuWbrCreate_TDMFS:
26896 +       case AuWbrCreate_TDMFSV:
26897 +       case AuWbrCreate_PMFSRR:
26898 +       case AuWbrCreate_PMFSRRV:
26899 +               sbinfo->si_wbr_mfs.mfsrr_watermark = create->mfsrr_watermark;
26900 +               fallthrough;
26901 +       case AuWbrCreate_MFS:
26902 +       case AuWbrCreate_MFSV:
26903 +       case AuWbrCreate_PMFS:
26904 +       case AuWbrCreate_PMFSV:
26905 +               sbinfo->si_wbr_mfs.mfs_expire
26906 +                       = msecs_to_jiffies(create->mfs_second * MSEC_PER_SEC);
26907 +               break;
26908 +       }
26909 +
26910 +       if (sbinfo->si_wbr_create_ops->init)
26911 +               sbinfo->si_wbr_create_ops->init(sb); /* ignore */
26912 +
26913 +       return err;
26914 +}
26915 +
26916 +/*
26917 + * returns,
26918 + * plus: processed without an error
26919 + * zero: unprocessed
26920 + */
26921 +static int au_opt_simple(struct super_block *sb, struct au_opt *opt,
26922 +                        struct au_opts *opts)
26923 +{
26924 +       int err;
26925 +       struct au_sbinfo *sbinfo;
26926 +
26927 +       SiMustWriteLock(sb);
26928 +
26929 +       err = 1; /* handled */
26930 +       sbinfo = au_sbi(sb);
26931 +       switch (opt->type) {
26932 +       case Opt_udba:
26933 +               sbinfo->si_mntflags &= ~AuOptMask_UDBA;
26934 +               sbinfo->si_mntflags |= opt->udba;
26935 +               opts->given_udba |= opt->udba;
26936 +               break;
26937 +
26938 +       case Opt_plink:
26939 +               au_opt_set(sbinfo->si_mntflags, PLINK);
26940 +               break;
26941 +       case Opt_noplink:
26942 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
26943 +                       au_plink_put(sb, /*verbose*/1);
26944 +               au_opt_clr(sbinfo->si_mntflags, PLINK);
26945 +               break;
26946 +       case Opt_list_plink:
26947 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
26948 +                       au_plink_list(sb);
26949 +               break;
26950 +
26951 +       case Opt_dio:
26952 +               au_opt_set(sbinfo->si_mntflags, DIO);
26953 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
26954 +               break;
26955 +       case Opt_nodio:
26956 +               au_opt_clr(sbinfo->si_mntflags, DIO);
26957 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
26958 +               break;
26959 +
26960 +       case Opt_fhsm_sec:
26961 +               au_fhsm_set(sbinfo, opt->fhsm_second);
26962 +               break;
26963 +
26964 +       case Opt_diropq_a:
26965 +               au_opt_set(sbinfo->si_mntflags, ALWAYS_DIROPQ);
26966 +               break;
26967 +       case Opt_diropq_w:
26968 +               au_opt_clr(sbinfo->si_mntflags, ALWAYS_DIROPQ);
26969 +               break;
26970 +
26971 +       case Opt_warn_perm:
26972 +               au_opt_set(sbinfo->si_mntflags, WARN_PERM);
26973 +               break;
26974 +       case Opt_nowarn_perm:
26975 +               au_opt_clr(sbinfo->si_mntflags, WARN_PERM);
26976 +               break;
26977 +
26978 +       case Opt_verbose:
26979 +               au_opt_set(sbinfo->si_mntflags, VERBOSE);
26980 +               break;
26981 +       case Opt_noverbose:
26982 +               au_opt_clr(sbinfo->si_mntflags, VERBOSE);
26983 +               break;
26984 +
26985 +       case Opt_sum:
26986 +               au_opt_set(sbinfo->si_mntflags, SUM);
26987 +               break;
26988 +       case Opt_wsum:
26989 +               au_opt_clr(sbinfo->si_mntflags, SUM);
26990 +               au_opt_set(sbinfo->si_mntflags, SUM_W);
26991 +               break;
26992 +       case Opt_nosum:
26993 +               au_opt_clr(sbinfo->si_mntflags, SUM);
26994 +               au_opt_clr(sbinfo->si_mntflags, SUM_W);
26995 +               break;
26996 +
26997 +       case Opt_wbr_create:
26998 +               err = au_opt_wbr_create(sb, &opt->wbr_create);
26999 +               break;
27000 +       case Opt_wbr_copyup:
27001 +               sbinfo->si_wbr_copyup = opt->wbr_copyup;
27002 +               sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + opt->wbr_copyup;
27003 +               break;
27004 +
27005 +       case Opt_dirwh:
27006 +               sbinfo->si_dirwh = opt->dirwh;
27007 +               break;
27008 +
27009 +       case Opt_rdcache:
27010 +               sbinfo->si_rdcache
27011 +                       = msecs_to_jiffies(opt->rdcache * MSEC_PER_SEC);
27012 +               break;
27013 +       case Opt_rdblk:
27014 +               sbinfo->si_rdblk = opt->rdblk;
27015 +               break;
27016 +       case Opt_rdblk_def:
27017 +               sbinfo->si_rdblk = AUFS_RDBLK_DEF;
27018 +               break;
27019 +       case Opt_rdhash:
27020 +               sbinfo->si_rdhash = opt->rdhash;
27021 +               break;
27022 +       case Opt_rdhash_def:
27023 +               sbinfo->si_rdhash = AUFS_RDHASH_DEF;
27024 +               break;
27025 +
27026 +       case Opt_shwh:
27027 +               au_opt_set(sbinfo->si_mntflags, SHWH);
27028 +               break;
27029 +       case Opt_noshwh:
27030 +               au_opt_clr(sbinfo->si_mntflags, SHWH);
27031 +               break;
27032 +
27033 +       case Opt_dirperm1:
27034 +               au_opt_set(sbinfo->si_mntflags, DIRPERM1);
27035 +               break;
27036 +       case Opt_nodirperm1:
27037 +               au_opt_clr(sbinfo->si_mntflags, DIRPERM1);
27038 +               break;
27039 +
27040 +       case Opt_trunc_xino:
27041 +               au_opt_set(sbinfo->si_mntflags, TRUNC_XINO);
27042 +               break;
27043 +       case Opt_notrunc_xino:
27044 +               au_opt_clr(sbinfo->si_mntflags, TRUNC_XINO);
27045 +               break;
27046 +
27047 +       case Opt_trunc_xino_path:
27048 +       case Opt_itrunc_xino:
27049 +               err = au_xino_trunc(sb, opt->xino_itrunc.bindex,
27050 +                                   /*idx_begin*/0);
27051 +               if (!err)
27052 +                       err = 1;
27053 +               break;
27054 +
27055 +       case Opt_trunc_xib:
27056 +               au_fset_opts(opts->flags, TRUNC_XIB);
27057 +               break;
27058 +       case Opt_notrunc_xib:
27059 +               au_fclr_opts(opts->flags, TRUNC_XIB);
27060 +               break;
27061 +
27062 +       case Opt_dirren:
27063 +               err = 1;
27064 +               if (!au_opt_test(sbinfo->si_mntflags, DIRREN)) {
27065 +                       err = au_dr_opt_set(sb);
27066 +                       if (!err)
27067 +                               err = 1;
27068 +               }
27069 +               if (err == 1)
27070 +                       au_opt_set(sbinfo->si_mntflags, DIRREN);
27071 +               break;
27072 +       case Opt_nodirren:
27073 +               err = 1;
27074 +               if (au_opt_test(sbinfo->si_mntflags, DIRREN)) {
27075 +                       err = au_dr_opt_clr(sb, au_ftest_opts(opts->flags,
27076 +                                                             DR_FLUSHED));
27077 +                       if (!err)
27078 +                               err = 1;
27079 +               }
27080 +               if (err == 1)
27081 +                       au_opt_clr(sbinfo->si_mntflags, DIRREN);
27082 +               break;
27083 +
27084 +       case Opt_acl:
27085 +               sb->s_flags |= SB_POSIXACL;
27086 +               break;
27087 +       case Opt_noacl:
27088 +               sb->s_flags &= ~SB_POSIXACL;
27089 +               break;
27090 +
27091 +       default:
27092 +               err = 0;
27093 +               break;
27094 +       }
27095 +
27096 +       return err;
27097 +}
27098 +
27099 +/*
27100 + * returns tri-state.
27101 + * plus: processed without an error
27102 + * zero: unprocessed
27103 + * minus: error
27104 + */
27105 +static int au_opt_br(struct super_block *sb, struct au_opt *opt,
27106 +                    struct au_opts *opts)
27107 +{
27108 +       int err, do_refresh;
27109 +
27110 +       err = 0;
27111 +       switch (opt->type) {
27112 +       case Opt_append:
27113 +               opt->add.bindex = au_sbbot(sb) + 1;
27114 +               if (opt->add.bindex < 0)
27115 +                       opt->add.bindex = 0;
27116 +               goto add;
27117 +               /* Always goto add, not fallthrough */
27118 +       case Opt_prepend:
27119 +               opt->add.bindex = 0;
27120 +               fallthrough;
27121 +       add: /* indented label */
27122 +       case Opt_add:
27123 +               err = au_br_add(sb, &opt->add,
27124 +                               au_ftest_opts(opts->flags, REMOUNT));
27125 +               if (!err) {
27126 +                       err = 1;
27127 +                       au_fset_opts(opts->flags, REFRESH);
27128 +               }
27129 +               break;
27130 +
27131 +       case Opt_del:
27132 +       case Opt_idel:
27133 +               err = au_br_del(sb, &opt->del,
27134 +                               au_ftest_opts(opts->flags, REMOUNT));
27135 +               if (!err) {
27136 +                       err = 1;
27137 +                       au_fset_opts(opts->flags, TRUNC_XIB);
27138 +                       au_fset_opts(opts->flags, REFRESH);
27139 +               }
27140 +               break;
27141 +
27142 +       case Opt_mod:
27143 +       case Opt_imod:
27144 +               err = au_br_mod(sb, &opt->mod,
27145 +                               au_ftest_opts(opts->flags, REMOUNT),
27146 +                               &do_refresh);
27147 +               if (!err) {
27148 +                       err = 1;
27149 +                       if (do_refresh)
27150 +                               au_fset_opts(opts->flags, REFRESH);
27151 +               }
27152 +               break;
27153 +       }
27154 +       return err;
27155 +}
27156 +
27157 +static int au_opt_xino(struct super_block *sb, struct au_opt *opt,
27158 +                      struct au_opt_xino **opt_xino,
27159 +                      struct au_opts *opts)
27160 +{
27161 +       int err;
27162 +
27163 +       err = 0;
27164 +       switch (opt->type) {
27165 +       case Opt_xino:
27166 +               err = au_xino_set(sb, &opt->xino,
27167 +                                 !!au_ftest_opts(opts->flags, REMOUNT));
27168 +               if (unlikely(err))
27169 +                       break;
27170 +
27171 +               *opt_xino = &opt->xino;
27172 +               break;
27173 +
27174 +       case Opt_noxino:
27175 +               au_xino_clr(sb);
27176 +               *opt_xino = (void *)-1;
27177 +               break;
27178 +       }
27179 +
27180 +       return err;
27181 +}
27182 +
27183 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
27184 +                  unsigned int pending)
27185 +{
27186 +       int err, fhsm;
27187 +       aufs_bindex_t bindex, bbot;
27188 +       unsigned char do_plink, skip, do_free, can_no_dreval;
27189 +       struct au_branch *br;
27190 +       struct au_wbr *wbr;
27191 +       struct dentry *root, *dentry;
27192 +       struct inode *dir, *h_dir;
27193 +       struct au_sbinfo *sbinfo;
27194 +       struct au_hinode *hdir;
27195 +
27196 +       SiMustAnyLock(sb);
27197 +
27198 +       sbinfo = au_sbi(sb);
27199 +       AuDebugOn(!(sbinfo->si_mntflags & AuOptMask_UDBA));
27200 +
27201 +       if (!(sb_flags & SB_RDONLY)) {
27202 +               if (unlikely(!au_br_writable(au_sbr_perm(sb, 0))))
27203 +                       pr_warn("first branch should be rw\n");
27204 +               if (unlikely(au_opt_test(sbinfo->si_mntflags, SHWH)))
27205 +                       pr_warn_once("shwh should be used with ro\n");
27206 +       }
27207 +
27208 +       if (au_opt_test((sbinfo->si_mntflags | pending), UDBA_HNOTIFY)
27209 +           && !au_opt_test(sbinfo->si_mntflags, XINO))
27210 +               pr_warn_once("udba=*notify requires xino\n");
27211 +
27212 +       if (au_opt_test(sbinfo->si_mntflags, DIRPERM1))
27213 +               pr_warn_once("dirperm1 breaks the protection"
27214 +                            " by the permission bits on the lower branch\n");
27215 +
27216 +       err = 0;
27217 +       fhsm = 0;
27218 +       root = sb->s_root;
27219 +       dir = d_inode(root);
27220 +       do_plink = !!au_opt_test(sbinfo->si_mntflags, PLINK);
27221 +       can_no_dreval = !!au_opt_test((sbinfo->si_mntflags | pending),
27222 +                                     UDBA_NONE);
27223 +       bbot = au_sbbot(sb);
27224 +       for (bindex = 0; !err && bindex <= bbot; bindex++) {
27225 +               skip = 0;
27226 +               h_dir = au_h_iptr(dir, bindex);
27227 +               br = au_sbr(sb, bindex);
27228 +
27229 +               if ((br->br_perm & AuBrAttr_ICEX)
27230 +                   && !h_dir->i_op->listxattr)
27231 +                       br->br_perm &= ~AuBrAttr_ICEX;
27232 +#if 0 /* untested */
27233 +               if ((br->br_perm & AuBrAttr_ICEX_SEC)
27234 +                   && (au_br_sb(br)->s_flags & SB_NOSEC))
27235 +                       br->br_perm &= ~AuBrAttr_ICEX_SEC;
27236 +#endif
27237 +
27238 +               do_free = 0;
27239 +               wbr = br->br_wbr;
27240 +               if (wbr)
27241 +                       wbr_wh_read_lock(wbr);
27242 +
27243 +               if (!au_br_writable(br->br_perm)) {
27244 +                       do_free = !!wbr;
27245 +                       skip = (!wbr
27246 +                               || (!wbr->wbr_whbase
27247 +                                   && !wbr->wbr_plink
27248 +                                   && !wbr->wbr_orph));
27249 +               } else if (!au_br_wh_linkable(br->br_perm)) {
27250 +                       /* skip = (!br->br_whbase && !br->br_orph); */
27251 +                       skip = (!wbr || !wbr->wbr_whbase);
27252 +                       if (skip && wbr) {
27253 +                               if (do_plink)
27254 +                                       skip = !!wbr->wbr_plink;
27255 +                               else
27256 +                                       skip = !wbr->wbr_plink;
27257 +                       }
27258 +               } else {
27259 +                       /* skip = (br->br_whbase && br->br_ohph); */
27260 +                       skip = (wbr && wbr->wbr_whbase);
27261 +                       if (skip) {
27262 +                               if (do_plink)
27263 +                                       skip = !!wbr->wbr_plink;
27264 +                               else
27265 +                                       skip = !wbr->wbr_plink;
27266 +                       }
27267 +               }
27268 +               if (wbr)
27269 +                       wbr_wh_read_unlock(wbr);
27270 +
27271 +               if (can_no_dreval) {
27272 +                       dentry = br->br_path.dentry;
27273 +                       spin_lock(&dentry->d_lock);
27274 +                       if (dentry->d_flags &
27275 +                           (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE))
27276 +                               can_no_dreval = 0;
27277 +                       spin_unlock(&dentry->d_lock);
27278 +               }
27279 +
27280 +               if (au_br_fhsm(br->br_perm)) {
27281 +                       fhsm++;
27282 +                       AuDebugOn(!br->br_fhsm);
27283 +               }
27284 +
27285 +               if (skip)
27286 +                       continue;
27287 +
27288 +               hdir = au_hi(dir, bindex);
27289 +               au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
27290 +               if (wbr)
27291 +                       wbr_wh_write_lock(wbr);
27292 +               err = au_wh_init(br, sb);
27293 +               if (wbr)
27294 +                       wbr_wh_write_unlock(wbr);
27295 +               au_hn_inode_unlock(hdir);
27296 +
27297 +               if (!err && do_free) {
27298 +                       au_kfree_rcu(wbr);
27299 +                       br->br_wbr = NULL;
27300 +               }
27301 +       }
27302 +
27303 +       if (can_no_dreval)
27304 +               au_fset_si(sbinfo, NO_DREVAL);
27305 +       else
27306 +               au_fclr_si(sbinfo, NO_DREVAL);
27307 +
27308 +       if (fhsm >= 2) {
27309 +               au_fset_si(sbinfo, FHSM);
27310 +               for (bindex = bbot; bindex >= 0; bindex--) {
27311 +                       br = au_sbr(sb, bindex);
27312 +                       if (au_br_fhsm(br->br_perm)) {
27313 +                               au_fhsm_set_bottom(sb, bindex);
27314 +                               break;
27315 +                       }
27316 +               }
27317 +       } else {
27318 +               au_fclr_si(sbinfo, FHSM);
27319 +               au_fhsm_set_bottom(sb, -1);
27320 +       }
27321 +
27322 +       return err;
27323 +}
27324 +
27325 +int au_opts_mount(struct super_block *sb, struct au_opts *opts)
27326 +{
27327 +       int err;
27328 +       unsigned int tmp;
27329 +       aufs_bindex_t bindex, bbot;
27330 +       struct au_opt *opt;
27331 +       struct au_opt_xino *opt_xino, xino;
27332 +       struct au_sbinfo *sbinfo;
27333 +       struct au_branch *br;
27334 +       struct inode *dir;
27335 +
27336 +       SiMustWriteLock(sb);
27337 +
27338 +       err = 0;
27339 +       opt_xino = NULL;
27340 +       opt = opts->opt;
27341 +       while (err >= 0 && opt->type != Opt_tail)
27342 +               err = au_opt_simple(sb, opt++, opts);
27343 +       if (err > 0)
27344 +               err = 0;
27345 +       else if (unlikely(err < 0))
27346 +               goto out;
27347 +
27348 +       /* disable xino and udba temporary */
27349 +       sbinfo = au_sbi(sb);
27350 +       tmp = sbinfo->si_mntflags;
27351 +       au_opt_clr(sbinfo->si_mntflags, XINO);
27352 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_REVAL);
27353 +
27354 +       opt = opts->opt;
27355 +       while (err >= 0 && opt->type != Opt_tail)
27356 +               err = au_opt_br(sb, opt++, opts);
27357 +       if (err > 0)
27358 +               err = 0;
27359 +       else if (unlikely(err < 0))
27360 +               goto out;
27361 +
27362 +       bbot = au_sbbot(sb);
27363 +       if (unlikely(bbot < 0)) {
27364 +               err = -EINVAL;
27365 +               pr_err("no branches\n");
27366 +               goto out;
27367 +       }
27368 +
27369 +       if (au_opt_test(tmp, XINO))
27370 +               au_opt_set(sbinfo->si_mntflags, XINO);
27371 +       opt = opts->opt;
27372 +       while (!err && opt->type != Opt_tail)
27373 +               err = au_opt_xino(sb, opt++, &opt_xino, opts);
27374 +       if (unlikely(err))
27375 +               goto out;
27376 +
27377 +       err = au_opts_verify(sb, sb->s_flags, tmp);
27378 +       if (unlikely(err))
27379 +               goto out;
27380 +
27381 +       /* restore xino */
27382 +       if (au_opt_test(tmp, XINO) && !opt_xino) {
27383 +               xino.file = au_xino_def(sb);
27384 +               err = PTR_ERR(xino.file);
27385 +               if (IS_ERR(xino.file))
27386 +                       goto out;
27387 +
27388 +               err = au_xino_set(sb, &xino, /*remount*/0);
27389 +               fput(xino.file);
27390 +               if (unlikely(err))
27391 +                       goto out;
27392 +       }
27393 +
27394 +       /* restore udba */
27395 +       tmp &= AuOptMask_UDBA;
27396 +       sbinfo->si_mntflags &= ~AuOptMask_UDBA;
27397 +       sbinfo->si_mntflags |= tmp;
27398 +       bbot = au_sbbot(sb);
27399 +       for (bindex = 0; bindex <= bbot; bindex++) {
27400 +               br = au_sbr(sb, bindex);
27401 +               err = au_hnotify_reset_br(tmp, br, br->br_perm);
27402 +               if (unlikely(err))
27403 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
27404 +                               bindex, err);
27405 +               /* go on even if err */
27406 +       }
27407 +       if (au_opt_test(tmp, UDBA_HNOTIFY)) {
27408 +               dir = d_inode(sb->s_root);
27409 +               au_hn_reset(dir, au_hi_flags(dir, /*isdir*/1) & ~AuHi_XINO);
27410 +       }
27411 +
27412 +out:
27413 +       return err;
27414 +}
27415 +
27416 +int au_opts_remount(struct super_block *sb, struct au_opts *opts)
27417 +{
27418 +       int err, rerr;
27419 +       unsigned char no_dreval;
27420 +       struct inode *dir;
27421 +       struct au_opt_xino *opt_xino;
27422 +       struct au_opt *opt;
27423 +       struct au_sbinfo *sbinfo;
27424 +
27425 +       SiMustWriteLock(sb);
27426 +
27427 +       err = au_dr_opt_flush(sb);
27428 +       if (unlikely(err))
27429 +               goto out;
27430 +       au_fset_opts(opts->flags, DR_FLUSHED);
27431 +
27432 +       dir = d_inode(sb->s_root);
27433 +       sbinfo = au_sbi(sb);
27434 +       opt_xino = NULL;
27435 +       opt = opts->opt;
27436 +       while (err >= 0 && opt->type != Opt_tail) {
27437 +               err = au_opt_simple(sb, opt, opts);
27438 +               if (!err)
27439 +                       err = au_opt_br(sb, opt, opts);
27440 +               if (!err)
27441 +                       err = au_opt_xino(sb, opt, &opt_xino, opts);
27442 +               opt++;
27443 +       }
27444 +       if (err > 0)
27445 +               err = 0;
27446 +       AuTraceErr(err);
27447 +       /* go on even err */
27448 +
27449 +       no_dreval = !!au_ftest_si(sbinfo, NO_DREVAL);
27450 +       rerr = au_opts_verify(sb, opts->sb_flags, /*pending*/0);
27451 +       if (unlikely(rerr && !err))
27452 +               err = rerr;
27453 +
27454 +       if (no_dreval != !!au_ftest_si(sbinfo, NO_DREVAL))
27455 +               au_fset_opts(opts->flags, REFRESH_IDOP);
27456 +
27457 +       if (au_ftest_opts(opts->flags, TRUNC_XIB)) {
27458 +               rerr = au_xib_trunc(sb);
27459 +               if (unlikely(rerr && !err))
27460 +                       err = rerr;
27461 +       }
27462 +
27463 +       /* will be handled by the caller */
27464 +       if (!au_ftest_opts(opts->flags, REFRESH)
27465 +           && (opts->given_udba
27466 +               || au_opt_test(sbinfo->si_mntflags, XINO)
27467 +               || au_ftest_opts(opts->flags, REFRESH_IDOP)
27468 +                   ))
27469 +               au_fset_opts(opts->flags, REFRESH);
27470 +
27471 +       AuDbg("status 0x%x\n", opts->flags);
27472 +
27473 +out:
27474 +       return err;
27475 +}
27476 +
27477 +/* ---------------------------------------------------------------------- */
27478 +
27479 +unsigned int au_opt_udba(struct super_block *sb)
27480 +{
27481 +       return au_mntflags(sb) & AuOptMask_UDBA;
27482 +}
27483 diff -urN /usr/share/empty/fs/aufs/opts.h linux/fs/aufs/opts.h
27484 --- /usr/share/empty/fs/aufs/opts.h     1970-01-01 01:00:00.000000000 +0100
27485 +++ linux/fs/aufs/opts.h        2021-12-03 15:38:59.939980643 +0100
27486 @@ -0,0 +1,225 @@
27487 +/* SPDX-License-Identifier: GPL-2.0 */
27488 +/*
27489 + * Copyright (C) 2005-2021 Junjiro R. Okajima
27490 + *
27491 + * This program, aufs is free software; you can redistribute it and/or modify
27492 + * it under the terms of the GNU General Public License as published by
27493 + * the Free Software Foundation; either version 2 of the License, or
27494 + * (at your option) any later version.
27495 + *
27496 + * This program is distributed in the hope that it will be useful,
27497 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27498 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27499 + * GNU General Public License for more details.
27500 + *
27501 + * You should have received a copy of the GNU General Public License
27502 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27503 + */
27504 +
27505 +/*
27506 + * mount options/flags
27507 + */
27508 +
27509 +#ifndef __AUFS_OPTS_H__
27510 +#define __AUFS_OPTS_H__
27511 +
27512 +#ifdef __KERNEL__
27513 +
27514 +#include <linux/path.h>
27515 +
27516 +struct file;
27517 +
27518 +/* ---------------------------------------------------------------------- */
27519 +
27520 +/* mount flags */
27521 +#define AuOpt_XINO             1               /* external inode number bitmap
27522 +                                                  and translation table */
27523 +#define AuOpt_TRUNC_XINO       (1 << 1)        /* truncate xino files */
27524 +#define AuOpt_UDBA_NONE                (1 << 2)        /* users direct branch access */
27525 +#define AuOpt_UDBA_REVAL       (1 << 3)
27526 +#define AuOpt_UDBA_HNOTIFY     (1 << 4)
27527 +#define AuOpt_SHWH             (1 << 5)        /* show whiteout */
27528 +#define AuOpt_PLINK            (1 << 6)        /* pseudo-link */
27529 +#define AuOpt_DIRPERM1         (1 << 7)        /* ignore the lower dir's perm
27530 +                                                  bits */
27531 +#define AuOpt_ALWAYS_DIROPQ    (1 << 9)        /* policy to creating diropq */
27532 +#define AuOpt_SUM              (1 << 10)       /* summation for statfs(2) */
27533 +#define AuOpt_SUM_W            (1 << 11)       /* unimplemented */
27534 +#define AuOpt_WARN_PERM                (1 << 12)       /* warn when add-branch */
27535 +#define AuOpt_VERBOSE          (1 << 13)       /* print the cause of error */
27536 +#define AuOpt_DIO              (1 << 14)       /* direct io */
27537 +#define AuOpt_DIRREN           (1 << 15)       /* directory rename */
27538 +
27539 +#ifndef CONFIG_AUFS_HNOTIFY
27540 +#undef AuOpt_UDBA_HNOTIFY
27541 +#define AuOpt_UDBA_HNOTIFY     0
27542 +#endif
27543 +#ifndef CONFIG_AUFS_DIRREN
27544 +#undef AuOpt_DIRREN
27545 +#define AuOpt_DIRREN           0
27546 +#endif
27547 +#ifndef CONFIG_AUFS_SHWH
27548 +#undef AuOpt_SHWH
27549 +#define AuOpt_SHWH             0
27550 +#endif
27551 +
27552 +#define AuOpt_Def      (AuOpt_XINO \
27553 +                        | AuOpt_UDBA_REVAL \
27554 +                        | AuOpt_PLINK \
27555 +                        /* | AuOpt_DIRPERM1 */ \
27556 +                        | AuOpt_WARN_PERM)
27557 +#define AuOptMask_UDBA (AuOpt_UDBA_NONE \
27558 +                        | AuOpt_UDBA_REVAL \
27559 +                        | AuOpt_UDBA_HNOTIFY)
27560 +
27561 +#define au_opt_test(flags, name)       (flags & AuOpt_##name)
27562 +#define au_opt_set(flags, name) do { \
27563 +       BUILD_BUG_ON(AuOpt_##name & AuOptMask_UDBA); \
27564 +       ((flags) |= AuOpt_##name); \
27565 +} while (0)
27566 +#define au_opt_set_udba(flags, name) do { \
27567 +       (flags) &= ~AuOptMask_UDBA; \
27568 +       ((flags) |= AuOpt_##name); \
27569 +} while (0)
27570 +#define au_opt_clr(flags, name) do { \
27571 +       ((flags) &= ~AuOpt_##name); \
27572 +} while (0)
27573 +
27574 +static inline unsigned int au_opts_plink(unsigned int mntflags)
27575 +{
27576 +#ifdef CONFIG_PROC_FS
27577 +       return mntflags;
27578 +#else
27579 +       return mntflags & ~AuOpt_PLINK;
27580 +#endif
27581 +}
27582 +
27583 +/* ---------------------------------------------------------------------- */
27584 +
27585 +/* policies to select one among multiple writable branches */
27586 +enum {
27587 +       AuWbrCreate_TDP,        /* top down parent */
27588 +       AuWbrCreate_RR,         /* round robin */
27589 +       AuWbrCreate_MFS,        /* most free space */
27590 +       AuWbrCreate_MFSV,       /* mfs with seconds */
27591 +       AuWbrCreate_MFSRR,      /* mfs then rr */
27592 +       AuWbrCreate_MFSRRV,     /* mfs then rr with seconds */
27593 +       AuWbrCreate_TDMFS,      /* top down regardless parent and mfs */
27594 +       AuWbrCreate_TDMFSV,     /* top down regardless parent and mfs */
27595 +       AuWbrCreate_PMFS,       /* parent and mfs */
27596 +       AuWbrCreate_PMFSV,      /* parent and mfs with seconds */
27597 +       AuWbrCreate_PMFSRR,     /* parent, mfs and round-robin */
27598 +       AuWbrCreate_PMFSRRV,    /* plus seconds */
27599 +
27600 +       AuWbrCreate_Def = AuWbrCreate_TDP
27601 +};
27602 +
27603 +enum {
27604 +       AuWbrCopyup_TDP,        /* top down parent */
27605 +       AuWbrCopyup_BUP,        /* bottom up parent */
27606 +       AuWbrCopyup_BU,         /* bottom up */
27607 +
27608 +       AuWbrCopyup_Def = AuWbrCopyup_TDP
27609 +};
27610 +
27611 +/* ---------------------------------------------------------------------- */
27612 +
27613 +struct au_opt_add {
27614 +       aufs_bindex_t   bindex;
27615 +       char            *pathname;
27616 +       int             perm;
27617 +       struct path     path;
27618 +};
27619 +
27620 +struct au_opt_del {
27621 +       char            *pathname;
27622 +       struct path     h_path;
27623 +};
27624 +
27625 +struct au_opt_mod {
27626 +       char            *path;
27627 +       int             perm;
27628 +       struct dentry   *h_root;
27629 +};
27630 +
27631 +struct au_opt_xino {
27632 +       char            *path;
27633 +       struct file     *file;
27634 +};
27635 +
27636 +struct au_opt_xino_itrunc {
27637 +       aufs_bindex_t   bindex;
27638 +};
27639 +
27640 +struct au_opt_wbr_create {
27641 +       int                     wbr_create;
27642 +       int                     mfs_second;
27643 +       unsigned long long      mfsrr_watermark;
27644 +};
27645 +
27646 +struct au_opt {
27647 +       int type;
27648 +       union {
27649 +               struct au_opt_xino      xino;
27650 +               struct au_opt_xino_itrunc xino_itrunc;
27651 +               struct au_opt_add       add;
27652 +               struct au_opt_del       del;
27653 +               struct au_opt_mod       mod;
27654 +               int                     dirwh;
27655 +               int                     rdcache;
27656 +               unsigned int            rdblk;
27657 +               unsigned int            rdhash;
27658 +               int                     udba;
27659 +               struct au_opt_wbr_create wbr_create;
27660 +               int                     wbr_copyup;
27661 +               unsigned int            fhsm_second;
27662 +       };
27663 +};
27664 +
27665 +/* opts flags */
27666 +#define AuOpts_REMOUNT         1
27667 +#define AuOpts_REFRESH         (1 << 1)
27668 +#define AuOpts_TRUNC_XIB       (1 << 2)
27669 +#define AuOpts_REFRESH_DYAOP   (1 << 3)
27670 +#define AuOpts_REFRESH_IDOP    (1 << 4)
27671 +#define AuOpts_DR_FLUSHED      (1 << 5)
27672 +#define au_ftest_opts(flags, name)     ((flags) & AuOpts_##name)
27673 +#define au_fset_opts(flags, name) \
27674 +       do { (flags) |= AuOpts_##name; } while (0)
27675 +#define au_fclr_opts(flags, name) \
27676 +       do { (flags) &= ~AuOpts_##name; } while (0)
27677 +
27678 +#ifndef CONFIG_AUFS_DIRREN
27679 +#undef AuOpts_DR_FLUSHED
27680 +#define AuOpts_DR_FLUSHED      0
27681 +#endif
27682 +
27683 +struct au_opts {
27684 +       struct au_opt   *opt;
27685 +       int             max_opt;
27686 +
27687 +       unsigned int    given_udba;
27688 +       unsigned int    flags;
27689 +       unsigned long   sb_flags;
27690 +};
27691 +
27692 +/* ---------------------------------------------------------------------- */
27693 +
27694 +/* opts.c */
27695 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm);
27696 +const char *au_optstr_udba(int udba);
27697 +const char *au_optstr_wbr_copyup(int wbr_copyup);
27698 +const char *au_optstr_wbr_create(int wbr_create);
27699 +
27700 +void au_opts_free(struct au_opts *opts);
27701 +struct super_block;
27702 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts);
27703 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
27704 +                  unsigned int pending);
27705 +int au_opts_mount(struct super_block *sb, struct au_opts *opts);
27706 +int au_opts_remount(struct super_block *sb, struct au_opts *opts);
27707 +
27708 +unsigned int au_opt_udba(struct super_block *sb);
27709 +
27710 +#endif /* __KERNEL__ */
27711 +#endif /* __AUFS_OPTS_H__ */
27712 diff -urN /usr/share/empty/fs/aufs/plink.c linux/fs/aufs/plink.c
27713 --- /usr/share/empty/fs/aufs/plink.c    1970-01-01 01:00:00.000000000 +0100
27714 +++ linux/fs/aufs/plink.c       2021-12-03 15:38:59.939980643 +0100
27715 @@ -0,0 +1,516 @@
27716 +// SPDX-License-Identifier: GPL-2.0
27717 +/*
27718 + * Copyright (C) 2005-2021 Junjiro R. Okajima
27719 + *
27720 + * This program, aufs is free software; you can redistribute it and/or modify
27721 + * it under the terms of the GNU General Public License as published by
27722 + * the Free Software Foundation; either version 2 of the License, or
27723 + * (at your option) any later version.
27724 + *
27725 + * This program is distributed in the hope that it will be useful,
27726 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27727 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27728 + * GNU General Public License for more details.
27729 + *
27730 + * You should have received a copy of the GNU General Public License
27731 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27732 + */
27733 +
27734 +/*
27735 + * pseudo-link
27736 + */
27737 +
27738 +#include "aufs.h"
27739 +
27740 +/*
27741 + * the pseudo-link maintenance mode.
27742 + * during a user process maintains the pseudo-links,
27743 + * prohibit adding a new plink and branch manipulation.
27744 + *
27745 + * Flags
27746 + * NOPLM:
27747 + *     For entry functions which will handle plink, and i_mutex is already held
27748 + *     in VFS.
27749 + *     They cannot wait and should return an error at once.
27750 + *     Callers has to check the error.
27751 + * NOPLMW:
27752 + *     For entry functions which will handle plink, but i_mutex is not held
27753 + *     in VFS.
27754 + *     They can wait the plink maintenance mode to finish.
27755 + *
27756 + * They behave like F_SETLK and F_SETLKW.
27757 + * If the caller never handle plink, then both flags are unnecessary.
27758 + */
27759 +
27760 +int au_plink_maint(struct super_block *sb, int flags)
27761 +{
27762 +       int err;
27763 +       pid_t pid, ppid;
27764 +       struct task_struct *parent, *prev;
27765 +       struct au_sbinfo *sbi;
27766 +
27767 +       SiMustAnyLock(sb);
27768 +
27769 +       err = 0;
27770 +       if (!au_opt_test(au_mntflags(sb), PLINK))
27771 +               goto out;
27772 +
27773 +       sbi = au_sbi(sb);
27774 +       pid = sbi->si_plink_maint_pid;
27775 +       if (!pid || pid == current->pid)
27776 +               goto out;
27777 +
27778 +       /* todo: it highly depends upon /sbin/mount.aufs */
27779 +       prev = NULL;
27780 +       parent = current;
27781 +       ppid = 0;
27782 +       rcu_read_lock();
27783 +       while (1) {
27784 +               parent = rcu_dereference(parent->real_parent);
27785 +               if (parent == prev)
27786 +                       break;
27787 +               ppid = task_pid_vnr(parent);
27788 +               if (pid == ppid) {
27789 +                       rcu_read_unlock();
27790 +                       goto out;
27791 +               }
27792 +               prev = parent;
27793 +       }
27794 +       rcu_read_unlock();
27795 +
27796 +       if (au_ftest_lock(flags, NOPLMW)) {
27797 +               /* if there is no i_mutex lock in VFS, we don't need to wait */
27798 +               /* AuDebugOn(!lockdep_depth(current)); */
27799 +               while (sbi->si_plink_maint_pid) {
27800 +                       si_read_unlock(sb);
27801 +                       /* gave up wake_up_bit() */
27802 +                       wait_event(sbi->si_plink_wq, !sbi->si_plink_maint_pid);
27803 +
27804 +                       if (au_ftest_lock(flags, FLUSH))
27805 +                               au_nwt_flush(&sbi->si_nowait);
27806 +                       si_noflush_read_lock(sb);
27807 +               }
27808 +       } else if (au_ftest_lock(flags, NOPLM)) {
27809 +               AuDbg("ppid %d, pid %d\n", ppid, pid);
27810 +               err = -EAGAIN;
27811 +       }
27812 +
27813 +out:
27814 +       return err;
27815 +}
27816 +
27817 +void au_plink_maint_leave(struct au_sbinfo *sbinfo)
27818 +{
27819 +       spin_lock(&sbinfo->si_plink_maint_lock);
27820 +       sbinfo->si_plink_maint_pid = 0;
27821 +       spin_unlock(&sbinfo->si_plink_maint_lock);
27822 +       wake_up_all(&sbinfo->si_plink_wq);
27823 +}
27824 +
27825 +int au_plink_maint_enter(struct super_block *sb)
27826 +{
27827 +       int err;
27828 +       struct au_sbinfo *sbinfo;
27829 +
27830 +       err = 0;
27831 +       sbinfo = au_sbi(sb);
27832 +       /* make sure i am the only one in this fs */
27833 +       si_write_lock(sb, AuLock_FLUSH);
27834 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
27835 +               spin_lock(&sbinfo->si_plink_maint_lock);
27836 +               if (!sbinfo->si_plink_maint_pid)
27837 +                       sbinfo->si_plink_maint_pid = current->pid;
27838 +               else
27839 +                       err = -EBUSY;
27840 +               spin_unlock(&sbinfo->si_plink_maint_lock);
27841 +       }
27842 +       si_write_unlock(sb);
27843 +
27844 +       return err;
27845 +}
27846 +
27847 +/* ---------------------------------------------------------------------- */
27848 +
27849 +#ifdef CONFIG_AUFS_DEBUG
27850 +void au_plink_list(struct super_block *sb)
27851 +{
27852 +       int i;
27853 +       struct au_sbinfo *sbinfo;
27854 +       struct hlist_bl_head *hbl;
27855 +       struct hlist_bl_node *pos;
27856 +       struct au_icntnr *icntnr;
27857 +
27858 +       SiMustAnyLock(sb);
27859 +
27860 +       sbinfo = au_sbi(sb);
27861 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
27862 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
27863 +
27864 +       for (i = 0; i < AuPlink_NHASH; i++) {
27865 +               hbl = sbinfo->si_plink + i;
27866 +               hlist_bl_lock(hbl);
27867 +               hlist_bl_for_each_entry(icntnr, pos, hbl, plink)
27868 +                       AuDbg("%lu\n", icntnr->vfs_inode.i_ino);
27869 +               hlist_bl_unlock(hbl);
27870 +       }
27871 +}
27872 +#endif
27873 +
27874 +/* is the inode pseudo-linked? */
27875 +int au_plink_test(struct inode *inode)
27876 +{
27877 +       int found, i;
27878 +       struct au_sbinfo *sbinfo;
27879 +       struct hlist_bl_head *hbl;
27880 +       struct hlist_bl_node *pos;
27881 +       struct au_icntnr *icntnr;
27882 +
27883 +       sbinfo = au_sbi(inode->i_sb);
27884 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
27885 +       AuDebugOn(!au_opt_test(au_mntflags(inode->i_sb), PLINK));
27886 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
27887 +
27888 +       found = 0;
27889 +       i = au_plink_hash(inode->i_ino);
27890 +       hbl =  sbinfo->si_plink + i;
27891 +       hlist_bl_lock(hbl);
27892 +       hlist_bl_for_each_entry(icntnr, pos, hbl, plink)
27893 +               if (&icntnr->vfs_inode == inode) {
27894 +                       found = 1;
27895 +                       break;
27896 +               }
27897 +       hlist_bl_unlock(hbl);
27898 +       return found;
27899 +}
27900 +
27901 +/* ---------------------------------------------------------------------- */
27902 +
27903 +/*
27904 + * generate a name for plink.
27905 + * the file will be stored under AUFS_WH_PLINKDIR.
27906 + */
27907 +/* 20 is max digits length of ulong 64 */
27908 +#define PLINK_NAME_LEN ((20 + 1) * 2)
27909 +
27910 +static int plink_name(char *name, int len, struct inode *inode,
27911 +                     aufs_bindex_t bindex)
27912 +{
27913 +       int rlen;
27914 +       struct inode *h_inode;
27915 +
27916 +       h_inode = au_h_iptr(inode, bindex);
27917 +       rlen = snprintf(name, len, "%lu.%lu", inode->i_ino, h_inode->i_ino);
27918 +       return rlen;
27919 +}
27920 +
27921 +struct au_do_plink_lkup_args {
27922 +       struct dentry **errp;
27923 +       struct qstr *tgtname;
27924 +       struct path *h_ppath;
27925 +};
27926 +
27927 +static struct dentry *au_do_plink_lkup(struct qstr *tgtname,
27928 +                                      struct path *h_ppath)
27929 +{
27930 +       struct dentry *h_dentry;
27931 +       struct inode *h_inode;
27932 +
27933 +       h_inode = d_inode(h_ppath->dentry);
27934 +       inode_lock_shared_nested(h_inode, AuLsc_I_CHILD2);
27935 +       h_dentry = vfsub_lkup_one(tgtname, h_ppath);
27936 +       inode_unlock_shared(h_inode);
27937 +
27938 +       return h_dentry;
27939 +}
27940 +
27941 +static void au_call_do_plink_lkup(void *args)
27942 +{
27943 +       struct au_do_plink_lkup_args *a = args;
27944 +       *a->errp = au_do_plink_lkup(a->tgtname, a->h_ppath);
27945 +}
27946 +
27947 +/* lookup the plink-ed @inode under the branch at @bindex */
27948 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex)
27949 +{
27950 +       struct dentry *h_dentry;
27951 +       struct au_branch *br;
27952 +       struct path h_ppath;
27953 +       int wkq_err;
27954 +       char a[PLINK_NAME_LEN];
27955 +       struct qstr tgtname = QSTR_INIT(a, 0);
27956 +
27957 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
27958 +
27959 +       br = au_sbr(inode->i_sb, bindex);
27960 +       h_ppath.dentry = br->br_wbr->wbr_plink;
27961 +       h_ppath.mnt = au_br_mnt(br);
27962 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
27963 +
27964 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
27965 +               struct au_do_plink_lkup_args args = {
27966 +                       .errp           = &h_dentry,
27967 +                       .tgtname        = &tgtname,
27968 +                       .h_ppath        = &h_ppath
27969 +               };
27970 +
27971 +               wkq_err = au_wkq_wait(au_call_do_plink_lkup, &args);
27972 +               if (unlikely(wkq_err))
27973 +                       h_dentry = ERR_PTR(wkq_err);
27974 +       } else
27975 +               h_dentry = au_do_plink_lkup(&tgtname, &h_ppath);
27976 +
27977 +       return h_dentry;
27978 +}
27979 +
27980 +/* create a pseudo-link */
27981 +static int do_whplink(struct qstr *tgt, struct path *h_ppath,
27982 +                     struct dentry *h_dentry)
27983 +{
27984 +       int err;
27985 +       struct path h_path;
27986 +       struct inode *h_dir, *delegated;
27987 +
27988 +       h_dir = d_inode(h_ppath->dentry);
27989 +       inode_lock_nested(h_dir, AuLsc_I_CHILD2);
27990 +       h_path.mnt = h_ppath->mnt;
27991 +again:
27992 +       h_path.dentry = vfsub_lkup_one(tgt, h_ppath);
27993 +       err = PTR_ERR(h_path.dentry);
27994 +       if (IS_ERR(h_path.dentry))
27995 +               goto out;
27996 +
27997 +       err = 0;
27998 +       /* wh.plink dir is not monitored */
27999 +       /* todo: is it really safe? */
28000 +       if (d_is_positive(h_path.dentry)
28001 +           && d_inode(h_path.dentry) != d_inode(h_dentry)) {
28002 +               delegated = NULL;
28003 +               err = vfsub_unlink(h_dir, &h_path, &delegated, /*force*/0);
28004 +               if (unlikely(err == -EWOULDBLOCK)) {
28005 +                       pr_warn("cannot retry for NFSv4 delegation"
28006 +                               " for an internal unlink\n");
28007 +                       iput(delegated);
28008 +               }
28009 +               dput(h_path.dentry);
28010 +               h_path.dentry = NULL;
28011 +               if (!err)
28012 +                       goto again;
28013 +       }
28014 +       if (!err && d_is_negative(h_path.dentry)) {
28015 +               delegated = NULL;
28016 +               err = vfsub_link(h_dentry, h_dir, &h_path, &delegated);
28017 +               if (unlikely(err == -EWOULDBLOCK)) {
28018 +                       pr_warn("cannot retry for NFSv4 delegation"
28019 +                               " for an internal link\n");
28020 +                       iput(delegated);
28021 +               }
28022 +       }
28023 +       dput(h_path.dentry);
28024 +
28025 +out:
28026 +       inode_unlock(h_dir);
28027 +       return err;
28028 +}
28029 +
28030 +struct do_whplink_args {
28031 +       int *errp;
28032 +       struct qstr *tgt;
28033 +       struct path *h_ppath;
28034 +       struct dentry *h_dentry;
28035 +};
28036 +
28037 +static void call_do_whplink(void *args)
28038 +{
28039 +       struct do_whplink_args *a = args;
28040 +       *a->errp = do_whplink(a->tgt, a->h_ppath, a->h_dentry);
28041 +}
28042 +
28043 +static int whplink(struct dentry *h_dentry, struct inode *inode,
28044 +                  aufs_bindex_t bindex)
28045 +{
28046 +       int err, wkq_err;
28047 +       struct au_branch *br;
28048 +       struct au_wbr *wbr;
28049 +       struct path h_ppath;
28050 +       char a[PLINK_NAME_LEN];
28051 +       struct qstr tgtname = QSTR_INIT(a, 0);
28052 +
28053 +       br = au_sbr(inode->i_sb, bindex);
28054 +       wbr = br->br_wbr;
28055 +       h_ppath.dentry = wbr->wbr_plink;
28056 +       h_ppath.mnt = au_br_mnt(br);
28057 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
28058 +
28059 +       /* always superio. */
28060 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
28061 +               struct do_whplink_args args = {
28062 +                       .errp           = &err,
28063 +                       .tgt            = &tgtname,
28064 +                       .h_ppath        = &h_ppath,
28065 +                       .h_dentry       = h_dentry
28066 +               };
28067 +               wkq_err = au_wkq_wait(call_do_whplink, &args);
28068 +               if (unlikely(wkq_err))
28069 +                       err = wkq_err;
28070 +       } else
28071 +               err = do_whplink(&tgtname, &h_ppath, h_dentry);
28072 +
28073 +       return err;
28074 +}
28075 +
28076 +/*
28077 + * create a new pseudo-link for @h_dentry on @bindex.
28078 + * the linked inode is held in aufs @inode.
28079 + */
28080 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
28081 +                    struct dentry *h_dentry)
28082 +{
28083 +       struct super_block *sb;
28084 +       struct au_sbinfo *sbinfo;
28085 +       struct hlist_bl_head *hbl;
28086 +       struct hlist_bl_node *pos;
28087 +       struct au_icntnr *icntnr;
28088 +       int found, err, cnt, i;
28089 +
28090 +       sb = inode->i_sb;
28091 +       sbinfo = au_sbi(sb);
28092 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
28093 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
28094 +
28095 +       found = au_plink_test(inode);
28096 +       if (found)
28097 +               return;
28098 +
28099 +       i = au_plink_hash(inode->i_ino);
28100 +       hbl = sbinfo->si_plink + i;
28101 +       au_igrab(inode);
28102 +
28103 +       hlist_bl_lock(hbl);
28104 +       hlist_bl_for_each_entry(icntnr, pos, hbl, plink) {
28105 +               if (&icntnr->vfs_inode == inode) {
28106 +                       found = 1;
28107 +                       break;
28108 +               }
28109 +       }
28110 +       if (!found) {
28111 +               icntnr = container_of(inode, struct au_icntnr, vfs_inode);
28112 +               hlist_bl_add_head(&icntnr->plink, hbl);
28113 +       }
28114 +       hlist_bl_unlock(hbl);
28115 +       if (!found) {
28116 +               cnt = au_hbl_count(hbl);
28117 +#define msg "unexpectedly unbalanced or too many pseudo-links"
28118 +               if (cnt > AUFS_PLINK_WARN)
28119 +                       AuWarn1(msg ", %d\n", cnt);
28120 +#undef msg
28121 +               err = whplink(h_dentry, inode, bindex);
28122 +               if (unlikely(err)) {
28123 +                       pr_warn("err %d, damaged pseudo link.\n", err);
28124 +                       au_hbl_del(&icntnr->plink, hbl);
28125 +                       iput(&icntnr->vfs_inode);
28126 +               }
28127 +       } else
28128 +               iput(&icntnr->vfs_inode);
28129 +}
28130 +
28131 +/* free all plinks */
28132 +void au_plink_put(struct super_block *sb, int verbose)
28133 +{
28134 +       int i, warned;
28135 +       struct au_sbinfo *sbinfo;
28136 +       struct hlist_bl_head *hbl;
28137 +       struct hlist_bl_node *pos, *tmp;
28138 +       struct au_icntnr *icntnr;
28139 +
28140 +       SiMustWriteLock(sb);
28141 +
28142 +       sbinfo = au_sbi(sb);
28143 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
28144 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
28145 +
28146 +       /* no spin_lock since sbinfo is write-locked */
28147 +       warned = 0;
28148 +       for (i = 0; i < AuPlink_NHASH; i++) {
28149 +               hbl = sbinfo->si_plink + i;
28150 +               if (!warned && verbose && !hlist_bl_empty(hbl)) {
28151 +                       pr_warn("pseudo-link is not flushed");
28152 +                       warned = 1;
28153 +               }
28154 +               hlist_bl_for_each_entry_safe(icntnr, pos, tmp, hbl, plink)
28155 +                       iput(&icntnr->vfs_inode);
28156 +               INIT_HLIST_BL_HEAD(hbl);
28157 +       }
28158 +}
28159 +
28160 +void au_plink_clean(struct super_block *sb, int verbose)
28161 +{
28162 +       struct dentry *root;
28163 +
28164 +       root = sb->s_root;
28165 +       aufs_write_lock(root);
28166 +       if (au_opt_test(au_mntflags(sb), PLINK))
28167 +               au_plink_put(sb, verbose);
28168 +       aufs_write_unlock(root);
28169 +}
28170 +
28171 +static int au_plink_do_half_refresh(struct inode *inode, aufs_bindex_t br_id)
28172 +{
28173 +       int do_put;
28174 +       aufs_bindex_t btop, bbot, bindex;
28175 +
28176 +       do_put = 0;
28177 +       btop = au_ibtop(inode);
28178 +       bbot = au_ibbot(inode);
28179 +       if (btop >= 0) {
28180 +               for (bindex = btop; bindex <= bbot; bindex++) {
28181 +                       if (!au_h_iptr(inode, bindex)
28182 +                           || au_ii_br_id(inode, bindex) != br_id)
28183 +                               continue;
28184 +                       au_set_h_iptr(inode, bindex, NULL, 0);
28185 +                       do_put = 1;
28186 +                       break;
28187 +               }
28188 +               if (do_put)
28189 +                       for (bindex = btop; bindex <= bbot; bindex++)
28190 +                               if (au_h_iptr(inode, bindex)) {
28191 +                                       do_put = 0;
28192 +                                       break;
28193 +                               }
28194 +       } else
28195 +               do_put = 1;
28196 +
28197 +       return do_put;
28198 +}
28199 +
28200 +/* free the plinks on a branch specified by @br_id */
28201 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id)
28202 +{
28203 +       struct au_sbinfo *sbinfo;
28204 +       struct hlist_bl_head *hbl;
28205 +       struct hlist_bl_node *pos, *tmp;
28206 +       struct au_icntnr *icntnr;
28207 +       struct inode *inode;
28208 +       int i, do_put;
28209 +
28210 +       SiMustWriteLock(sb);
28211 +
28212 +       sbinfo = au_sbi(sb);
28213 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
28214 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
28215 +
28216 +       /* no bit_lock since sbinfo is write-locked */
28217 +       for (i = 0; i < AuPlink_NHASH; i++) {
28218 +               hbl = sbinfo->si_plink + i;
28219 +               hlist_bl_for_each_entry_safe(icntnr, pos, tmp, hbl, plink) {
28220 +                       inode = au_igrab(&icntnr->vfs_inode);
28221 +                       ii_write_lock_child(inode);
28222 +                       do_put = au_plink_do_half_refresh(inode, br_id);
28223 +                       if (do_put) {
28224 +                               hlist_bl_del(&icntnr->plink);
28225 +                               iput(inode);
28226 +                       }
28227 +                       ii_write_unlock(inode);
28228 +                       iput(inode);
28229 +               }
28230 +       }
28231 +}
28232 diff -urN /usr/share/empty/fs/aufs/poll.c linux/fs/aufs/poll.c
28233 --- /usr/share/empty/fs/aufs/poll.c     1970-01-01 01:00:00.000000000 +0100
28234 +++ linux/fs/aufs/poll.c        2021-12-03 15:38:59.939980643 +0100
28235 @@ -0,0 +1,51 @@
28236 +// SPDX-License-Identifier: GPL-2.0
28237 +/*
28238 + * Copyright (C) 2005-2021 Junjiro R. Okajima
28239 + *
28240 + * This program, aufs is free software; you can redistribute it and/or modify
28241 + * it under the terms of the GNU General Public License as published by
28242 + * the Free Software Foundation; either version 2 of the License, or
28243 + * (at your option) any later version.
28244 + *
28245 + * This program is distributed in the hope that it will be useful,
28246 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28247 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28248 + * GNU General Public License for more details.
28249 + *
28250 + * You should have received a copy of the GNU General Public License
28251 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28252 + */
28253 +
28254 +/*
28255 + * poll operation
28256 + * There is only one filesystem which implements ->poll operation, currently.
28257 + */
28258 +
28259 +#include "aufs.h"
28260 +
28261 +__poll_t aufs_poll(struct file *file, struct poll_table_struct *pt)
28262 +{
28263 +       __poll_t mask;
28264 +       struct file *h_file;
28265 +       struct super_block *sb;
28266 +
28267 +       /* We should pretend an error happened. */
28268 +       mask = EPOLLERR /* | EPOLLIN | EPOLLOUT */;
28269 +       sb = file->f_path.dentry->d_sb;
28270 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
28271 +
28272 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
28273 +       if (IS_ERR(h_file)) {
28274 +               AuDbg("h_file %ld\n", PTR_ERR(h_file));
28275 +               goto out;
28276 +       }
28277 +
28278 +       mask = vfs_poll(h_file, pt);
28279 +       fput(h_file); /* instead of au_read_post() */
28280 +
28281 +out:
28282 +       si_read_unlock(sb);
28283 +       if (mask & EPOLLERR)
28284 +               AuDbg("mask 0x%x\n", mask);
28285 +       return mask;
28286 +}
28287 diff -urN /usr/share/empty/fs/aufs/posix_acl.c linux/fs/aufs/posix_acl.c
28288 --- /usr/share/empty/fs/aufs/posix_acl.c        1970-01-01 01:00:00.000000000 +0100
28289 +++ linux/fs/aufs/posix_acl.c   2021-12-03 15:40:58.236647297 +0100
28290 @@ -0,0 +1,111 @@
28291 +// SPDX-License-Identifier: GPL-2.0
28292 +/*
28293 + * Copyright (C) 2014-2021 Junjiro R. Okajima
28294 + *
28295 + * This program, aufs is free software; you can redistribute it and/or modify
28296 + * it under the terms of the GNU General Public License as published by
28297 + * the Free Software Foundation; either version 2 of the License, or
28298 + * (at your option) any later version.
28299 + *
28300 + * This program is distributed in the hope that it will be useful,
28301 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28302 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28303 + * GNU General Public License for more details.
28304 + *
28305 + * You should have received a copy of the GNU General Public License
28306 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28307 + */
28308 +
28309 +/*
28310 + * posix acl operations
28311 + */
28312 +
28313 +#include <linux/fs.h>
28314 +#include "aufs.h"
28315 +
28316 +struct posix_acl *aufs_get_acl(struct inode *inode, int type, bool rcu)
28317 +{
28318 +       struct posix_acl *acl;
28319 +       int err;
28320 +       aufs_bindex_t bindex;
28321 +       struct inode *h_inode;
28322 +       struct super_block *sb;
28323 +
28324 +       acl = ERR_PTR(-ECHILD);
28325 +       if (rcu)
28326 +               goto out;
28327 +
28328 +       acl = NULL;
28329 +       sb = inode->i_sb;
28330 +       si_read_lock(sb, AuLock_FLUSH);
28331 +       ii_read_lock_child(inode);
28332 +       if (!(sb->s_flags & SB_POSIXACL))
28333 +               goto unlock;
28334 +
28335 +       bindex = au_ibtop(inode);
28336 +       h_inode = au_h_iptr(inode, bindex);
28337 +       if (unlikely(!h_inode
28338 +                    || ((h_inode->i_mode & S_IFMT)
28339 +                        != (inode->i_mode & S_IFMT)))) {
28340 +               err = au_busy_or_stale();
28341 +               acl = ERR_PTR(err);
28342 +               goto unlock;
28343 +       }
28344 +
28345 +       /* always topmost only */
28346 +       acl = get_acl(h_inode, type);
28347 +       if (IS_ERR(acl))
28348 +               forget_cached_acl(inode, type);
28349 +       else
28350 +               set_cached_acl(inode, type, acl);
28351 +
28352 +unlock:
28353 +       ii_read_unlock(inode);
28354 +       si_read_unlock(sb);
28355 +
28356 +out:
28357 +       AuTraceErrPtr(acl);
28358 +       return acl;
28359 +}
28360 +
28361 +int aufs_set_acl(struct user_namespace *userns, struct inode *inode,
28362 +                struct posix_acl *acl, int type)
28363 +{
28364 +       int err;
28365 +       ssize_t ssz;
28366 +       struct dentry *dentry;
28367 +       struct au_sxattr arg = {
28368 +               .type = AU_ACL_SET,
28369 +               .u.acl_set = {
28370 +                       .acl    = acl,
28371 +                       .type   = type
28372 +               },
28373 +       };
28374 +
28375 +       IMustLock(inode);
28376 +
28377 +       if (inode->i_ino == AUFS_ROOT_INO)
28378 +               dentry = dget(inode->i_sb->s_root);
28379 +       else {
28380 +               dentry = d_find_alias(inode);
28381 +               if (!dentry)
28382 +                       dentry = d_find_any_alias(inode);
28383 +               if (!dentry) {
28384 +                       pr_warn("cannot handle this inode, "
28385 +                               "please report to aufs-users ML\n");
28386 +                       err = -ENOENT;
28387 +                       goto out;
28388 +               }
28389 +       }
28390 +
28391 +       ssz = au_sxattr(dentry, inode, &arg);
28392 +       /* forget even it if succeeds since the branch might set differently */
28393 +       forget_cached_acl(inode, type);
28394 +       dput(dentry);
28395 +       err = ssz;
28396 +       if (ssz >= 0)
28397 +               err = 0;
28398 +
28399 +out:
28400 +       return err;
28401 +}
28402 diff -urN /usr/share/empty/fs/aufs/procfs.c linux/fs/aufs/procfs.c
28403 --- /usr/share/empty/fs/aufs/procfs.c   1970-01-01 01:00:00.000000000 +0100
28404 +++ linux/fs/aufs/procfs.c      2021-12-03 15:38:59.939980643 +0100
28405 @@ -0,0 +1,170 @@
28406 +// SPDX-License-Identifier: GPL-2.0
28407 +/*
28408 + * Copyright (C) 2010-2021 Junjiro R. Okajima
28409 + *
28410 + * This program, aufs is free software; you can redistribute it and/or modify
28411 + * it under the terms of the GNU General Public License as published by
28412 + * the Free Software Foundation; either version 2 of the License, or
28413 + * (at your option) any later version.
28414 + *
28415 + * This program is distributed in the hope that it will be useful,
28416 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28417 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28418 + * GNU General Public License for more details.
28419 + *
28420 + * You should have received a copy of the GNU General Public License
28421 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28422 + */
28423 +
28424 +/*
28425 + * procfs interfaces
28426 + */
28427 +
28428 +#include <linux/proc_fs.h>
28429 +#include "aufs.h"
28430 +
28431 +static int au_procfs_plm_release(struct inode *inode, struct file *file)
28432 +{
28433 +       struct au_sbinfo *sbinfo;
28434 +
28435 +       sbinfo = file->private_data;
28436 +       if (sbinfo) {
28437 +               au_plink_maint_leave(sbinfo);
28438 +               kobject_put(&sbinfo->si_kobj);
28439 +       }
28440 +
28441 +       return 0;
28442 +}
28443 +
28444 +static void au_procfs_plm_write_clean(struct file *file)
28445 +{
28446 +       struct au_sbinfo *sbinfo;
28447 +
28448 +       sbinfo = file->private_data;
28449 +       if (sbinfo)
28450 +               au_plink_clean(sbinfo->si_sb, /*verbose*/0);
28451 +}
28452 +
28453 +static int au_procfs_plm_write_si(struct file *file, unsigned long id)
28454 +{
28455 +       int err;
28456 +       struct super_block *sb;
28457 +       struct au_sbinfo *sbinfo;
28458 +       struct hlist_bl_node *pos;
28459 +
28460 +       err = -EBUSY;
28461 +       if (unlikely(file->private_data))
28462 +               goto out;
28463 +
28464 +       sb = NULL;
28465 +       /* don't use au_sbilist_lock() here */
28466 +       hlist_bl_lock(&au_sbilist);
28467 +       hlist_bl_for_each_entry(sbinfo, pos, &au_sbilist, si_list)
28468 +               if (id == sysaufs_si_id(sbinfo)) {
28469 +                       if (kobject_get_unless_zero(&sbinfo->si_kobj))
28470 +                               sb = sbinfo->si_sb;
28471 +                       break;
28472 +               }
28473 +       hlist_bl_unlock(&au_sbilist);
28474 +
28475 +       err = -EINVAL;
28476 +       if (unlikely(!sb))
28477 +               goto out;
28478 +
28479 +       err = au_plink_maint_enter(sb);
28480 +       if (!err)
28481 +               /* keep kobject_get() */
28482 +               file->private_data = sbinfo;
28483 +       else
28484 +               kobject_put(&sbinfo->si_kobj);
28485 +out:
28486 +       return err;
28487 +}
28488 +
28489 +/*
28490 + * Accept a valid "si=xxxx" only.
28491 + * Once it is accepted successfully, accept "clean" too.
28492 + */
28493 +static ssize_t au_procfs_plm_write(struct file *file, const char __user *ubuf,
28494 +                                  size_t count, loff_t *ppos)
28495 +{
28496 +       ssize_t err;
28497 +       unsigned long id;
28498 +       /* last newline is allowed */
28499 +       char buf[3 + sizeof(unsigned long) * 2 + 1];
28500 +
28501 +       err = -EACCES;
28502 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
28503 +               goto out;
28504 +
28505 +       err = -EINVAL;
28506 +       if (unlikely(count > sizeof(buf)))
28507 +               goto out;
28508 +
28509 +       err = copy_from_user(buf, ubuf, count);
28510 +       if (unlikely(err)) {
28511 +               err = -EFAULT;
28512 +               goto out;
28513 +       }
28514 +       buf[count] = 0;
28515 +
28516 +       err = -EINVAL;
28517 +       if (!strcmp("clean", buf)) {
28518 +               au_procfs_plm_write_clean(file);
28519 +               goto out_success;
28520 +       } else if (unlikely(strncmp("si=", buf, 3)))
28521 +               goto out;
28522 +
28523 +       err = kstrtoul(buf + 3, 16, &id);
28524 +       if (unlikely(err))
28525 +               goto out;
28526 +
28527 +       err = au_procfs_plm_write_si(file, id);
28528 +       if (unlikely(err))
28529 +               goto out;
28530 +
28531 +out_success:
28532 +       err = count; /* success */
28533 +out:
28534 +       return err;
28535 +}
28536 +
28537 +static const struct proc_ops au_procfs_plm_op = {
28538 +       .proc_write     = au_procfs_plm_write,
28539 +       .proc_release   = au_procfs_plm_release
28540 +};
28541 +
28542 +/* ---------------------------------------------------------------------- */
28543 +
28544 +static struct proc_dir_entry *au_procfs_dir;
28545 +
28546 +void au_procfs_fin(void)
28547 +{
28548 +       remove_proc_entry(AUFS_PLINK_MAINT_NAME, au_procfs_dir);
28549 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
28550 +}
28551 +
28552 +int __init au_procfs_init(void)
28553 +{
28554 +       int err;
28555 +       struct proc_dir_entry *entry;
28556 +
28557 +       err = -ENOMEM;
28558 +       au_procfs_dir = proc_mkdir(AUFS_PLINK_MAINT_DIR, NULL);
28559 +       if (unlikely(!au_procfs_dir))
28560 +               goto out;
28561 +
28562 +       entry = proc_create(AUFS_PLINK_MAINT_NAME, S_IFREG | 0200,
28563 +                           au_procfs_dir, &au_procfs_plm_op);
28564 +       if (unlikely(!entry))
28565 +               goto out_dir;
28566 +
28567 +       err = 0;
28568 +       goto out; /* success */
28569 +
28570 +
28571 +out_dir:
28572 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
28573 +out:
28574 +       return err;
28575 +}
28576 diff -urN /usr/share/empty/fs/aufs/rdu.c linux/fs/aufs/rdu.c
28577 --- /usr/share/empty/fs/aufs/rdu.c      1970-01-01 01:00:00.000000000 +0100
28578 +++ linux/fs/aufs/rdu.c 2021-12-03 15:38:59.939980643 +0100
28579 @@ -0,0 +1,384 @@
28580 +// SPDX-License-Identifier: GPL-2.0
28581 +/*
28582 + * Copyright (C) 2005-2021 Junjiro R. Okajima
28583 + *
28584 + * This program, aufs is free software; you can redistribute it and/or modify
28585 + * it under the terms of the GNU General Public License as published by
28586 + * the Free Software Foundation; either version 2 of the License, or
28587 + * (at your option) any later version.
28588 + *
28589 + * This program is distributed in the hope that it will be useful,
28590 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28591 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28592 + * GNU General Public License for more details.
28593 + *
28594 + * You should have received a copy of the GNU General Public License
28595 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28596 + */
28597 +
28598 +/*
28599 + * readdir in userspace.
28600 + */
28601 +
28602 +#include <linux/compat.h>
28603 +#include <linux/fs_stack.h>
28604 +#include <linux/security.h>
28605 +#include "aufs.h"
28606 +
28607 +/* bits for struct aufs_rdu.flags */
28608 +#define        AuRdu_CALLED    1
28609 +#define        AuRdu_CONT      (1 << 1)
28610 +#define        AuRdu_FULL      (1 << 2)
28611 +#define au_ftest_rdu(flags, name)      ((flags) & AuRdu_##name)
28612 +#define au_fset_rdu(flags, name) \
28613 +       do { (flags) |= AuRdu_##name; } while (0)
28614 +#define au_fclr_rdu(flags, name) \
28615 +       do { (flags) &= ~AuRdu_##name; } while (0)
28616 +
28617 +struct au_rdu_arg {
28618 +       struct dir_context              ctx;
28619 +       struct aufs_rdu                 *rdu;
28620 +       union au_rdu_ent_ul             ent;
28621 +       unsigned long                   end;
28622 +
28623 +       struct super_block              *sb;
28624 +       int                             err;
28625 +};
28626 +
28627 +static int au_rdu_fill(struct dir_context *ctx, const char *name, int nlen,
28628 +                      loff_t offset, u64 h_ino, unsigned int d_type)
28629 +{
28630 +       int err, len;
28631 +       struct au_rdu_arg *arg = container_of(ctx, struct au_rdu_arg, ctx);
28632 +       struct aufs_rdu *rdu = arg->rdu;
28633 +       struct au_rdu_ent ent;
28634 +
28635 +       err = 0;
28636 +       arg->err = 0;
28637 +       au_fset_rdu(rdu->cookie.flags, CALLED);
28638 +       len = au_rdu_len(nlen);
28639 +       if (arg->ent.ul + len  < arg->end) {
28640 +               ent.ino = h_ino;
28641 +               ent.bindex = rdu->cookie.bindex;
28642 +               ent.type = d_type;
28643 +               ent.nlen = nlen;
28644 +               if (unlikely(nlen > AUFS_MAX_NAMELEN))
28645 +                       ent.type = DT_UNKNOWN;
28646 +
28647 +               /* unnecessary to support mmap_sem since this is a dir */
28648 +               err = -EFAULT;
28649 +               if (copy_to_user(arg->ent.e, &ent, sizeof(ent)))
28650 +                       goto out;
28651 +               if (copy_to_user(arg->ent.e->name, name, nlen))
28652 +                       goto out;
28653 +               /* the terminating NULL */
28654 +               if (__put_user(0, arg->ent.e->name + nlen))
28655 +                       goto out;
28656 +               err = 0;
28657 +               /* AuDbg("%p, %.*s\n", arg->ent.p, nlen, name); */
28658 +               arg->ent.ul += len;
28659 +               rdu->rent++;
28660 +       } else {
28661 +               err = -EFAULT;
28662 +               au_fset_rdu(rdu->cookie.flags, FULL);
28663 +               rdu->full = 1;
28664 +               rdu->tail = arg->ent;
28665 +       }
28666 +
28667 +out:
28668 +       /* AuTraceErr(err); */
28669 +       return err;
28670 +}
28671 +
28672 +static int au_rdu_do(struct file *h_file, struct au_rdu_arg *arg)
28673 +{
28674 +       int err;
28675 +       loff_t offset;
28676 +       struct au_rdu_cookie *cookie = &arg->rdu->cookie;
28677 +
28678 +       /* we don't have to care (FMODE_32BITHASH | FMODE_64BITHASH) for ext4 */
28679 +       offset = vfsub_llseek(h_file, cookie->h_pos, SEEK_SET);
28680 +       err = offset;
28681 +       if (unlikely(offset != cookie->h_pos))
28682 +               goto out;
28683 +
28684 +       err = 0;
28685 +       do {
28686 +               arg->err = 0;
28687 +               au_fclr_rdu(cookie->flags, CALLED);
28688 +               /* smp_mb(); */
28689 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
28690 +               if (err >= 0)
28691 +                       err = arg->err;
28692 +       } while (!err
28693 +                && au_ftest_rdu(cookie->flags, CALLED)
28694 +                && !au_ftest_rdu(cookie->flags, FULL));
28695 +       cookie->h_pos = h_file->f_pos;
28696 +
28697 +out:
28698 +       AuTraceErr(err);
28699 +       return err;
28700 +}
28701 +
28702 +static int au_rdu(struct file *file, struct aufs_rdu *rdu)
28703 +{
28704 +       int err;
28705 +       aufs_bindex_t bbot;
28706 +       struct au_rdu_arg arg = {
28707 +               .ctx = {
28708 +                       .actor = au_rdu_fill
28709 +               }
28710 +       };
28711 +       struct dentry *dentry;
28712 +       struct inode *inode;
28713 +       struct file *h_file;
28714 +       struct au_rdu_cookie *cookie = &rdu->cookie;
28715 +
28716 +       /* VERIFY_WRITE */
28717 +       err = !access_ok(rdu->ent.e, rdu->sz);
28718 +       if (unlikely(err)) {
28719 +               err = -EFAULT;
28720 +               AuTraceErr(err);
28721 +               goto out;
28722 +       }
28723 +       rdu->rent = 0;
28724 +       rdu->tail = rdu->ent;
28725 +       rdu->full = 0;
28726 +       arg.rdu = rdu;
28727 +       arg.ent = rdu->ent;
28728 +       arg.end = arg.ent.ul;
28729 +       arg.end += rdu->sz;
28730 +
28731 +       err = -ENOTDIR;
28732 +       if (unlikely(!file->f_op->iterate && !file->f_op->iterate_shared))
28733 +               goto out;
28734 +
28735 +       err = security_file_permission(file, MAY_READ);
28736 +       AuTraceErr(err);
28737 +       if (unlikely(err))
28738 +               goto out;
28739 +
28740 +       dentry = file->f_path.dentry;
28741 +       inode = d_inode(dentry);
28742 +       inode_lock_shared(inode);
28743 +
28744 +       arg.sb = inode->i_sb;
28745 +       err = si_read_lock(arg.sb, AuLock_FLUSH | AuLock_NOPLM);
28746 +       if (unlikely(err))
28747 +               goto out_mtx;
28748 +       err = au_alive_dir(dentry);
28749 +       if (unlikely(err))
28750 +               goto out_si;
28751 +       /* todo: reval? */
28752 +       fi_read_lock(file);
28753 +
28754 +       err = -EAGAIN;
28755 +       if (unlikely(au_ftest_rdu(cookie->flags, CONT)
28756 +                    && cookie->generation != au_figen(file)))
28757 +               goto out_unlock;
28758 +
28759 +       err = 0;
28760 +       if (!rdu->blk) {
28761 +               rdu->blk = au_sbi(arg.sb)->si_rdblk;
28762 +               if (!rdu->blk)
28763 +                       rdu->blk = au_dir_size(file, /*dentry*/NULL);
28764 +       }
28765 +       bbot = au_fbtop(file);
28766 +       if (cookie->bindex < bbot)
28767 +               cookie->bindex = bbot;
28768 +       bbot = au_fbbot_dir(file);
28769 +       /* AuDbg("b%d, b%d\n", cookie->bindex, bbot); */
28770 +       for (; !err && cookie->bindex <= bbot;
28771 +            cookie->bindex++, cookie->h_pos = 0) {
28772 +               h_file = au_hf_dir(file, cookie->bindex);
28773 +               if (!h_file)
28774 +                       continue;
28775 +
28776 +               au_fclr_rdu(cookie->flags, FULL);
28777 +               err = au_rdu_do(h_file, &arg);
28778 +               AuTraceErr(err);
28779 +               if (unlikely(au_ftest_rdu(cookie->flags, FULL) || err))
28780 +                       break;
28781 +       }
28782 +       AuDbg("rent %llu\n", rdu->rent);
28783 +
28784 +       if (!err && !au_ftest_rdu(cookie->flags, CONT)) {
28785 +               rdu->shwh = !!au_opt_test(au_sbi(arg.sb)->si_mntflags, SHWH);
28786 +               au_fset_rdu(cookie->flags, CONT);
28787 +               cookie->generation = au_figen(file);
28788 +       }
28789 +
28790 +       ii_read_lock_child(inode);
28791 +       fsstack_copy_attr_atime(inode, au_h_iptr(inode, au_ibtop(inode)));
28792 +       ii_read_unlock(inode);
28793 +
28794 +out_unlock:
28795 +       fi_read_unlock(file);
28796 +out_si:
28797 +       si_read_unlock(arg.sb);
28798 +out_mtx:
28799 +       inode_unlock_shared(inode);
28800 +out:
28801 +       AuTraceErr(err);
28802 +       return err;
28803 +}
28804 +
28805 +static int au_rdu_ino(struct file *file, struct aufs_rdu *rdu)
28806 +{
28807 +       int err;
28808 +       ino_t ino;
28809 +       unsigned long long nent;
28810 +       union au_rdu_ent_ul *u;
28811 +       struct au_rdu_ent ent;
28812 +       struct super_block *sb;
28813 +
28814 +       err = 0;
28815 +       nent = rdu->nent;
28816 +       u = &rdu->ent;
28817 +       sb = file->f_path.dentry->d_sb;
28818 +       si_read_lock(sb, AuLock_FLUSH);
28819 +       while (nent-- > 0) {
28820 +               /* unnecessary to support mmap_sem since this is a dir */
28821 +               err = copy_from_user(&ent, u->e, sizeof(ent));
28822 +               if (!err)
28823 +                       /* VERIFY_WRITE */
28824 +                       err = !access_ok(&u->e->ino, sizeof(ino));
28825 +               if (unlikely(err)) {
28826 +                       err = -EFAULT;
28827 +                       AuTraceErr(err);
28828 +                       break;
28829 +               }
28830 +
28831 +               /* AuDbg("b%d, i%llu\n", ent.bindex, ent.ino); */
28832 +               if (!ent.wh)
28833 +                       err = au_ino(sb, ent.bindex, ent.ino, ent.type, &ino);
28834 +               else
28835 +                       err = au_wh_ino(sb, ent.bindex, ent.ino, ent.type,
28836 +                                       &ino);
28837 +               if (unlikely(err)) {
28838 +                       AuTraceErr(err);
28839 +                       break;
28840 +               }
28841 +
28842 +               err = __put_user(ino, &u->e->ino);
28843 +               if (unlikely(err)) {
28844 +                       err = -EFAULT;
28845 +                       AuTraceErr(err);
28846 +                       break;
28847 +               }
28848 +               u->ul += au_rdu_len(ent.nlen);
28849 +       }
28850 +       si_read_unlock(sb);
28851 +
28852 +       return err;
28853 +}
28854 +
28855 +/* ---------------------------------------------------------------------- */
28856 +
28857 +static int au_rdu_verify(struct aufs_rdu *rdu)
28858 +{
28859 +       AuDbg("rdu{%llu, %p, %u | %u | %llu, %u, %u | "
28860 +             "%llu, b%d, 0x%x, g%u}\n",
28861 +             rdu->sz, rdu->ent.e, rdu->verify[AufsCtlRduV_SZ],
28862 +             rdu->blk,
28863 +             rdu->rent, rdu->shwh, rdu->full,
28864 +             rdu->cookie.h_pos, rdu->cookie.bindex, rdu->cookie.flags,
28865 +             rdu->cookie.generation);
28866 +
28867 +       if (rdu->verify[AufsCtlRduV_SZ] == sizeof(*rdu))
28868 +               return 0;
28869 +
28870 +       AuDbg("%u:%u\n",
28871 +             rdu->verify[AufsCtlRduV_SZ], (unsigned int)sizeof(*rdu));
28872 +       return -EINVAL;
28873 +}
28874 +
28875 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
28876 +{
28877 +       long err, e;
28878 +       struct aufs_rdu rdu;
28879 +       void __user *p = (void __user *)arg;
28880 +
28881 +       err = copy_from_user(&rdu, p, sizeof(rdu));
28882 +       if (unlikely(err)) {
28883 +               err = -EFAULT;
28884 +               AuTraceErr(err);
28885 +               goto out;
28886 +       }
28887 +       err = au_rdu_verify(&rdu);
28888 +       if (unlikely(err))
28889 +               goto out;
28890 +
28891 +       switch (cmd) {
28892 +       case AUFS_CTL_RDU:
28893 +               err = au_rdu(file, &rdu);
28894 +               if (unlikely(err))
28895 +                       break;
28896 +
28897 +               e = copy_to_user(p, &rdu, sizeof(rdu));
28898 +               if (unlikely(e)) {
28899 +                       err = -EFAULT;
28900 +                       AuTraceErr(err);
28901 +               }
28902 +               break;
28903 +       case AUFS_CTL_RDU_INO:
28904 +               err = au_rdu_ino(file, &rdu);
28905 +               break;
28906 +
28907 +       default:
28908 +               /* err = -ENOTTY; */
28909 +               err = -EINVAL;
28910 +       }
28911 +
28912 +out:
28913 +       AuTraceErr(err);
28914 +       return err;
28915 +}
28916 +
28917 +#ifdef CONFIG_COMPAT
28918 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
28919 +{
28920 +       long err, e;
28921 +       struct aufs_rdu rdu;
28922 +       void __user *p = compat_ptr(arg);
28923 +
28924 +       /* todo: get_user()? */
28925 +       err = copy_from_user(&rdu, p, sizeof(rdu));
28926 +       if (unlikely(err)) {
28927 +               err = -EFAULT;
28928 +               AuTraceErr(err);
28929 +               goto out;
28930 +       }
28931 +       rdu.ent.e = compat_ptr(rdu.ent.ul);
28932 +       err = au_rdu_verify(&rdu);
28933 +       if (unlikely(err))
28934 +               goto out;
28935 +
28936 +       switch (cmd) {
28937 +       case AUFS_CTL_RDU:
28938 +               err = au_rdu(file, &rdu);
28939 +               if (unlikely(err))
28940 +                       break;
28941 +
28942 +               rdu.ent.ul = ptr_to_compat(rdu.ent.e);
28943 +               rdu.tail.ul = ptr_to_compat(rdu.tail.e);
28944 +               e = copy_to_user(p, &rdu, sizeof(rdu));
28945 +               if (unlikely(e)) {
28946 +                       err = -EFAULT;
28947 +                       AuTraceErr(err);
28948 +               }
28949 +               break;
28950 +       case AUFS_CTL_RDU_INO:
28951 +               err = au_rdu_ino(file, &rdu);
28952 +               break;
28953 +
28954 +       default:
28955 +               /* err = -ENOTTY; */
28956 +               err = -EINVAL;
28957 +       }
28958 +
28959 +out:
28960 +       AuTraceErr(err);
28961 +       return err;
28962 +}
28963 +#endif
28964 diff -urN /usr/share/empty/fs/aufs/rwsem.h linux/fs/aufs/rwsem.h
28965 --- /usr/share/empty/fs/aufs/rwsem.h    1970-01-01 01:00:00.000000000 +0100
28966 +++ linux/fs/aufs/rwsem.h       2021-12-03 15:40:58.236647297 +0100
28967 @@ -0,0 +1,85 @@
28968 +/* SPDX-License-Identifier: GPL-2.0 */
28969 +/*
28970 + * Copyright (C) 2005-2021 Junjiro R. Okajima
28971 + *
28972 + * This program, aufs is free software; you can redistribute it and/or modify
28973 + * it under the terms of the GNU General Public License as published by
28974 + * the Free Software Foundation; either version 2 of the License, or
28975 + * (at your option) any later version.
28976 + *
28977 + * This program is distributed in the hope that it will be useful,
28978 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28979 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28980 + * GNU General Public License for more details.
28981 + *
28982 + * You should have received a copy of the GNU General Public License
28983 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28984 + */
28985 +
28986 +/*
28987 + * simple read-write semaphore wrappers
28988 + */
28989 +
28990 +#ifndef __AUFS_RWSEM_H__
28991 +#define __AUFS_RWSEM_H__
28992 +
28993 +#ifdef __KERNEL__
28994 +
28995 +#include "debug.h"
28996 +
28997 +/* in the future, the name 'au_rwsem' will be totally gone */
28998 +#define au_rwsem       rw_semaphore
28999 +
29000 +/* to debug easier, do not make them inlined functions */
29001 +#define AuRwMustNoWaiters(rw)  AuDebugOn(rwsem_is_contended(rw))
29002 +
29003 +#ifdef CONFIG_LOCKDEP
29004 +/* rwsem_is_locked() is unusable */
29005 +#define AuRwMustReadLock(rw)   AuDebugOn(IS_ENABLED(CONFIG_LOCKDEP)    \
29006 +                                         && !lockdep_recursing(current) \
29007 +                                         && debug_locks                \
29008 +                                         && !lockdep_is_held_type(rw, 1))
29009 +#define AuRwMustWriteLock(rw)  AuDebugOn(IS_ENABLED(CONFIG_LOCKDEP)    \
29010 +                                         && !lockdep_recursing(current) \
29011 +                                         && debug_locks                \
29012 +                                         && !lockdep_is_held_type(rw, 0))
29013 +#define AuRwMustAnyLock(rw)    AuDebugOn(IS_ENABLED(CONFIG_LOCKDEP)    \
29014 +                                         && !lockdep_recursing(current) \
29015 +                                         && debug_locks                \
29016 +                                         && !lockdep_is_held(rw))
29017 +#define AuRwDestroy(rw)                AuDebugOn(IS_ENABLED(CONFIG_LOCKDEP)    \
29018 +                                         && !lockdep_recursing(current) \
29019 +                                         && debug_locks                \
29020 +                                         && lockdep_is_held(rw))
29021 +#else
29022 +#define AuRwMustReadLock(rw)   do {} while (0)
29023 +#define AuRwMustWriteLock(rw)  do {} while (0)
29024 +#define AuRwMustAnyLock(rw)    do {} while (0)
29025 +#define AuRwDestroy(rw)                do {} while (0)
29026 +#endif
29027 +
29028 +#define au_rw_init(rw) init_rwsem(rw)
29029 +
29030 +#define au_rw_init_wlock(rw) do {              \
29031 +               au_rw_init(rw);                 \
29032 +               down_write(rw);                 \
29033 +       } while (0)
29034 +
29035 +#define au_rw_init_wlock_nested(rw, lsc) do {  \
29036 +               au_rw_init(rw);                 \
29037 +               down_write_nested(rw, lsc);     \
29038 +       } while (0)
29039 +
29040 +#define au_rw_read_lock(rw)            down_read(rw)
29041 +#define au_rw_read_lock_nested(rw, lsc)        down_read_nested(rw, lsc)
29042 +#define au_rw_read_unlock(rw)          up_read(rw)
29043 +#define au_rw_dgrade_lock(rw)          downgrade_write(rw)
29044 +#define au_rw_write_lock(rw)           down_write(rw)
29045 +#define au_rw_write_lock_nested(rw, lsc) down_write_nested(rw, lsc)
29046 +#define au_rw_write_unlock(rw)         up_write(rw)
29047 +/* why is not _nested version defined? */
29048 +#define au_rw_read_trylock(rw)         down_read_trylock(rw)
29049 +#define au_rw_write_trylock(rw)                down_write_trylock(rw)
29050 +
29051 +#endif /* __KERNEL__ */
29052 +#endif /* __AUFS_RWSEM_H__ */
29053 diff -urN /usr/share/empty/fs/aufs/sbinfo.c linux/fs/aufs/sbinfo.c
29054 --- /usr/share/empty/fs/aufs/sbinfo.c   1970-01-01 01:00:00.000000000 +0100
29055 +++ linux/fs/aufs/sbinfo.c      2021-12-03 15:38:59.939980643 +0100
29056 @@ -0,0 +1,314 @@
29057 +// SPDX-License-Identifier: GPL-2.0
29058 +/*
29059 + * Copyright (C) 2005-2021 Junjiro R. Okajima
29060 + *
29061 + * This program, aufs is free software; you can redistribute it and/or modify
29062 + * it under the terms of the GNU General Public License as published by
29063 + * the Free Software Foundation; either version 2 of the License, or
29064 + * (at your option) any later version.
29065 + *
29066 + * This program is distributed in the hope that it will be useful,
29067 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29068 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29069 + * GNU General Public License for more details.
29070 + *
29071 + * You should have received a copy of the GNU General Public License
29072 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29073 + */
29074 +
29075 +/*
29076 + * superblock private data
29077 + */
29078 +
29079 +#include <linux/iversion.h>
29080 +#include "aufs.h"
29081 +
29082 +/*
29083 + * they are necessary regardless sysfs is disabled.
29084 + */
29085 +void au_si_free(struct kobject *kobj)
29086 +{
29087 +       int i;
29088 +       struct au_sbinfo *sbinfo;
29089 +       char *locked __maybe_unused; /* debug only */
29090 +
29091 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
29092 +       for (i = 0; i < AuPlink_NHASH; i++)
29093 +               AuDebugOn(!hlist_bl_empty(sbinfo->si_plink + i));
29094 +       AuDebugOn(atomic_read(&sbinfo->si_nowait.nw_len));
29095 +
29096 +       AuLCntZero(au_lcnt_read(&sbinfo->si_ninodes, /*do_rev*/0));
29097 +       au_lcnt_fin(&sbinfo->si_ninodes, /*do_sync*/0);
29098 +       AuLCntZero(au_lcnt_read(&sbinfo->si_nfiles, /*do_rev*/0));
29099 +       au_lcnt_fin(&sbinfo->si_nfiles, /*do_sync*/0);
29100 +
29101 +       dbgaufs_si_fin(sbinfo);
29102 +       au_rw_write_lock(&sbinfo->si_rwsem);
29103 +       au_br_free(sbinfo);
29104 +       au_rw_write_unlock(&sbinfo->si_rwsem);
29105 +
29106 +       au_kfree_try_rcu(sbinfo->si_branch);
29107 +       mutex_destroy(&sbinfo->si_xib_mtx);
29108 +       AuRwDestroy(&sbinfo->si_rwsem);
29109 +
29110 +       au_lcnt_wait_for_fin(&sbinfo->si_ninodes);
29111 +       /* si_nfiles is waited too */
29112 +       au_kfree_rcu(sbinfo);
29113 +}
29114 +
29115 +int au_si_alloc(struct super_block *sb)
29116 +{
29117 +       int err, i;
29118 +       struct au_sbinfo *sbinfo;
29119 +
29120 +       err = -ENOMEM;
29121 +       sbinfo = kzalloc(sizeof(*sbinfo), GFP_NOFS);
29122 +       if (unlikely(!sbinfo))
29123 +               goto out;
29124 +
29125 +       /* will be reallocated separately */
29126 +       sbinfo->si_branch = kzalloc(sizeof(*sbinfo->si_branch), GFP_NOFS);
29127 +       if (unlikely(!sbinfo->si_branch))
29128 +               goto out_sbinfo;
29129 +
29130 +       err = sysaufs_si_init(sbinfo);
29131 +       if (!err) {
29132 +               dbgaufs_si_null(sbinfo);
29133 +               err = dbgaufs_si_init(sbinfo);
29134 +               if (unlikely(err))
29135 +                       kobject_put(&sbinfo->si_kobj);
29136 +       }
29137 +       if (unlikely(err))
29138 +               goto out_br;
29139 +
29140 +       au_nwt_init(&sbinfo->si_nowait);
29141 +       au_rw_init_wlock(&sbinfo->si_rwsem);
29142 +
29143 +       au_lcnt_init(&sbinfo->si_ninodes, /*release*/NULL);
29144 +       au_lcnt_init(&sbinfo->si_nfiles, /*release*/NULL);
29145 +
29146 +       sbinfo->si_bbot = -1;
29147 +       sbinfo->si_last_br_id = AUFS_BRANCH_MAX / 2;
29148 +
29149 +       sbinfo->si_wbr_copyup = AuWbrCopyup_Def;
29150 +       sbinfo->si_wbr_create = AuWbrCreate_Def;
29151 +       sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + sbinfo->si_wbr_copyup;
29152 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + sbinfo->si_wbr_create;
29153 +
29154 +       au_fhsm_init(sbinfo);
29155 +
29156 +       sbinfo->si_mntflags = au_opts_plink(AuOpt_Def);
29157 +
29158 +       sbinfo->si_xino_jiffy = jiffies;
29159 +       sbinfo->si_xino_expire
29160 +               = msecs_to_jiffies(AUFS_XINO_DEF_SEC * MSEC_PER_SEC);
29161 +       mutex_init(&sbinfo->si_xib_mtx);
29162 +       /* leave si_xib_last_pindex and si_xib_next_bit */
29163 +
29164 +       INIT_HLIST_BL_HEAD(&sbinfo->si_aopen);
29165 +
29166 +       sbinfo->si_rdcache = msecs_to_jiffies(AUFS_RDCACHE_DEF * MSEC_PER_SEC);
29167 +       sbinfo->si_rdblk = AUFS_RDBLK_DEF;
29168 +       sbinfo->si_rdhash = AUFS_RDHASH_DEF;
29169 +       sbinfo->si_dirwh = AUFS_DIRWH_DEF;
29170 +
29171 +       for (i = 0; i < AuPlink_NHASH; i++)
29172 +               INIT_HLIST_BL_HEAD(sbinfo->si_plink + i);
29173 +       init_waitqueue_head(&sbinfo->si_plink_wq);
29174 +       spin_lock_init(&sbinfo->si_plink_maint_lock);
29175 +
29176 +       INIT_HLIST_BL_HEAD(&sbinfo->si_files);
29177 +
29178 +       /* with getattr by default */
29179 +       sbinfo->si_iop_array = aufs_iop;
29180 +
29181 +       /* leave other members for sysaufs and si_mnt. */
29182 +       sbinfo->si_sb = sb;
29183 +       sb->s_fs_info = sbinfo;
29184 +       si_pid_set(sb);
29185 +       return 0; /* success */
29186 +
29187 +out_br:
29188 +       au_kfree_try_rcu(sbinfo->si_branch);
29189 +out_sbinfo:
29190 +       au_kfree_rcu(sbinfo);
29191 +out:
29192 +       return err;
29193 +}
29194 +
29195 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr, int may_shrink)
29196 +{
29197 +       int err, sz;
29198 +       struct au_branch **brp;
29199 +
29200 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
29201 +
29202 +       err = -ENOMEM;
29203 +       sz = sizeof(*brp) * (sbinfo->si_bbot + 1);
29204 +       if (unlikely(!sz))
29205 +               sz = sizeof(*brp);
29206 +       brp = au_kzrealloc(sbinfo->si_branch, sz, sizeof(*brp) * nbr, GFP_NOFS,
29207 +                          may_shrink);
29208 +       if (brp) {
29209 +               sbinfo->si_branch = brp;
29210 +               err = 0;
29211 +       }
29212 +
29213 +       return err;
29214 +}
29215 +
29216 +/* ---------------------------------------------------------------------- */
29217 +
29218 +unsigned int au_sigen_inc(struct super_block *sb)
29219 +{
29220 +       unsigned int gen;
29221 +       struct inode *inode;
29222 +
29223 +       SiMustWriteLock(sb);
29224 +
29225 +       gen = ++au_sbi(sb)->si_generation;
29226 +       au_update_digen(sb->s_root);
29227 +       inode = d_inode(sb->s_root);
29228 +       au_update_iigen(inode, /*half*/0);
29229 +       inode_inc_iversion(inode);
29230 +       return gen;
29231 +}
29232 +
29233 +aufs_bindex_t au_new_br_id(struct super_block *sb)
29234 +{
29235 +       aufs_bindex_t br_id;
29236 +       int i;
29237 +       struct au_sbinfo *sbinfo;
29238 +
29239 +       SiMustWriteLock(sb);
29240 +
29241 +       sbinfo = au_sbi(sb);
29242 +       for (i = 0; i <= AUFS_BRANCH_MAX; i++) {
29243 +               br_id = ++sbinfo->si_last_br_id;
29244 +               AuDebugOn(br_id < 0);
29245 +               if (br_id && au_br_index(sb, br_id) < 0)
29246 +                       return br_id;
29247 +       }
29248 +
29249 +       return -1;
29250 +}
29251 +
29252 +/* ---------------------------------------------------------------------- */
29253 +
29254 +/* it is ok that new 'nwt' tasks are appended while we are sleeping */
29255 +int si_read_lock(struct super_block *sb, int flags)
29256 +{
29257 +       int err;
29258 +
29259 +       err = 0;
29260 +       if (au_ftest_lock(flags, FLUSH))
29261 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
29262 +
29263 +       si_noflush_read_lock(sb);
29264 +       err = au_plink_maint(sb, flags);
29265 +       if (unlikely(err))
29266 +               si_read_unlock(sb);
29267 +
29268 +       return err;
29269 +}
29270 +
29271 +int si_write_lock(struct super_block *sb, int flags)
29272 +{
29273 +       int err;
29274 +
29275 +       if (au_ftest_lock(flags, FLUSH))
29276 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
29277 +
29278 +       si_noflush_write_lock(sb);
29279 +       err = au_plink_maint(sb, flags);
29280 +       if (unlikely(err))
29281 +               si_write_unlock(sb);
29282 +
29283 +       return err;
29284 +}
29285 +
29286 +/* dentry and super_block lock. call at entry point */
29287 +int aufs_read_lock(struct dentry *dentry, int flags)
29288 +{
29289 +       int err;
29290 +       struct super_block *sb;
29291 +
29292 +       sb = dentry->d_sb;
29293 +       err = si_read_lock(sb, flags);
29294 +       if (unlikely(err))
29295 +               goto out;
29296 +
29297 +       if (au_ftest_lock(flags, DW))
29298 +               di_write_lock_child(dentry);
29299 +       else
29300 +               di_read_lock_child(dentry, flags);
29301 +
29302 +       if (au_ftest_lock(flags, GEN)) {
29303 +               err = au_digen_test(dentry, au_sigen(sb));
29304 +               if (!au_opt_test(au_mntflags(sb), UDBA_NONE))
29305 +                       AuDebugOn(!err && au_dbrange_test(dentry));
29306 +               else if (!err)
29307 +                       err = au_dbrange_test(dentry);
29308 +               if (unlikely(err))
29309 +                       aufs_read_unlock(dentry, flags);
29310 +       }
29311 +
29312 +out:
29313 +       return err;
29314 +}
29315 +
29316 +void aufs_read_unlock(struct dentry *dentry, int flags)
29317 +{
29318 +       if (au_ftest_lock(flags, DW))
29319 +               di_write_unlock(dentry);
29320 +       else
29321 +               di_read_unlock(dentry, flags);
29322 +       si_read_unlock(dentry->d_sb);
29323 +}
29324 +
29325 +void aufs_write_lock(struct dentry *dentry)
29326 +{
29327 +       si_write_lock(dentry->d_sb, AuLock_FLUSH | AuLock_NOPLMW);
29328 +       di_write_lock_child(dentry);
29329 +}
29330 +
29331 +void aufs_write_unlock(struct dentry *dentry)
29332 +{
29333 +       di_write_unlock(dentry);
29334 +       si_write_unlock(dentry->d_sb);
29335 +}
29336 +
29337 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags)
29338 +{
29339 +       int err;
29340 +       unsigned int sigen;
29341 +       struct super_block *sb;
29342 +
29343 +       sb = d1->d_sb;
29344 +       err = si_read_lock(sb, flags);
29345 +       if (unlikely(err))
29346 +               goto out;
29347 +
29348 +       di_write_lock2_child(d1, d2, au_ftest_lock(flags, DIRS));
29349 +
29350 +       if (au_ftest_lock(flags, GEN)) {
29351 +               sigen = au_sigen(sb);
29352 +               err = au_digen_test(d1, sigen);
29353 +               AuDebugOn(!err && au_dbrange_test(d1));
29354 +               if (!err) {
29355 +                       err = au_digen_test(d2, sigen);
29356 +                       AuDebugOn(!err && au_dbrange_test(d2));
29357 +               }
29358 +               if (unlikely(err))
29359 +                       aufs_read_and_write_unlock2(d1, d2);
29360 +       }
29361 +
29362 +out:
29363 +       return err;
29364 +}
29365 +
29366 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2)
29367 +{
29368 +       di_write_unlock2(d1, d2);
29369 +       si_read_unlock(d1->d_sb);
29370 +}
29371 diff -urN /usr/share/empty/fs/aufs/super.c linux/fs/aufs/super.c
29372 --- /usr/share/empty/fs/aufs/super.c    1970-01-01 01:00:00.000000000 +0100
29373 +++ linux/fs/aufs/super.c       2021-12-03 15:40:58.236647297 +0100
29374 @@ -0,0 +1,1050 @@
29375 +// SPDX-License-Identifier: GPL-2.0
29376 +/*
29377 + * Copyright (C) 2005-2021 Junjiro R. Okajima
29378 + *
29379 + * This program, aufs is free software; you can redistribute it and/or modify
29380 + * it under the terms of the GNU General Public License as published by
29381 + * the Free Software Foundation; either version 2 of the License, or
29382 + * (at your option) any later version.
29383 + *
29384 + * This program is distributed in the hope that it will be useful,
29385 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29386 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29387 + * GNU General Public License for more details.
29388 + *
29389 + * You should have received a copy of the GNU General Public License
29390 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29391 + */
29392 +
29393 +/*
29394 + * mount and super_block operations
29395 + */
29396 +
29397 +#include <linux/iversion.h>
29398 +#include <linux/mm.h>
29399 +#include <linux/seq_file.h>
29400 +#include <linux/statfs.h>
29401 +#include <linux/vmalloc.h>
29402 +#include "aufs.h"
29403 +
29404 +/*
29405 + * super_operations
29406 + */
29407 +static struct inode *aufs_alloc_inode(struct super_block *sb __maybe_unused)
29408 +{
29409 +       struct au_icntnr *c;
29410 +
29411 +       c = au_cache_alloc_icntnr();
29412 +       if (c) {
29413 +               au_icntnr_init(c);
29414 +               inode_set_iversion(&c->vfs_inode, 1); /* sigen(sb); */
29415 +               c->iinfo.ii_hinode = NULL;
29416 +               return &c->vfs_inode;
29417 +       }
29418 +       return NULL;
29419 +}
29420 +
29421 +static void aufs_destroy_inode(struct inode *inode)
29422 +{
29423 +       if (!au_is_bad_inode(inode))
29424 +               au_iinfo_fin(inode);
29425 +}
29426 +
29427 +static void aufs_free_inode(struct inode *inode)
29428 +{
29429 +       au_cache_free_icntnr(container_of(inode, struct au_icntnr, vfs_inode));
29430 +}
29431 +
29432 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino)
29433 +{
29434 +       struct inode *inode;
29435 +       int err;
29436 +
29437 +       inode = iget_locked(sb, ino);
29438 +       if (unlikely(!inode)) {
29439 +               inode = ERR_PTR(-ENOMEM);
29440 +               goto out;
29441 +       }
29442 +       if (!(inode->i_state & I_NEW))
29443 +               goto out;
29444 +
29445 +       err = au_xigen_new(inode);
29446 +       if (!err)
29447 +               err = au_iinfo_init(inode);
29448 +       if (!err)
29449 +               inode_inc_iversion(inode);
29450 +       else {
29451 +               iget_failed(inode);
29452 +               inode = ERR_PTR(err);
29453 +       }
29454 +
29455 +out:
29456 +       /* never return NULL */
29457 +       AuDebugOn(!inode);
29458 +       AuTraceErrPtr(inode);
29459 +       return inode;
29460 +}
29461 +
29462 +/* lock free root dinfo */
29463 +static int au_show_brs(struct seq_file *seq, struct super_block *sb)
29464 +{
29465 +       int err;
29466 +       aufs_bindex_t bindex, bbot;
29467 +       struct path path;
29468 +       struct au_hdentry *hdp;
29469 +       struct au_branch *br;
29470 +       au_br_perm_str_t perm;
29471 +
29472 +       err = 0;
29473 +       bbot = au_sbbot(sb);
29474 +       bindex = 0;
29475 +       hdp = au_hdentry(au_di(sb->s_root), bindex);
29476 +       for (; !err && bindex <= bbot; bindex++, hdp++) {
29477 +               br = au_sbr(sb, bindex);
29478 +               path.mnt = au_br_mnt(br);
29479 +               path.dentry = hdp->hd_dentry;
29480 +               err = au_seq_path(seq, &path);
29481 +               if (!err) {
29482 +                       au_optstr_br_perm(&perm, br->br_perm);
29483 +                       seq_printf(seq, "=%s", perm.a);
29484 +                       if (bindex != bbot)
29485 +                               seq_putc(seq, ':');
29486 +               }
29487 +       }
29488 +       if (unlikely(err || seq_has_overflowed(seq)))
29489 +               err = -E2BIG;
29490 +
29491 +       return err;
29492 +}
29493 +
29494 +static void au_gen_fmt(char *fmt, int len __maybe_unused, const char *pat,
29495 +                      const char *append)
29496 +{
29497 +       char *p;
29498 +
29499 +       p = fmt;
29500 +       while (*pat != ':')
29501 +               *p++ = *pat++;
29502 +       *p++ = *pat++;
29503 +       strcpy(p, append);
29504 +       AuDebugOn(strlen(fmt) >= len);
29505 +}
29506 +
29507 +static void au_show_wbr_create(struct seq_file *m, int v,
29508 +                              struct au_sbinfo *sbinfo)
29509 +{
29510 +       const char *pat;
29511 +       char fmt[32];
29512 +       struct au_wbr_mfs *mfs;
29513 +
29514 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
29515 +
29516 +       seq_puts(m, ",create=");
29517 +       pat = au_optstr_wbr_create(v);
29518 +       mfs = &sbinfo->si_wbr_mfs;
29519 +       switch (v) {
29520 +       case AuWbrCreate_TDP:
29521 +       case AuWbrCreate_RR:
29522 +       case AuWbrCreate_MFS:
29523 +       case AuWbrCreate_PMFS:
29524 +               seq_puts(m, pat);
29525 +               break;
29526 +       case AuWbrCreate_MFSRR:
29527 +       case AuWbrCreate_TDMFS:
29528 +       case AuWbrCreate_PMFSRR:
29529 +               au_gen_fmt(fmt, sizeof(fmt), pat, "%llu");
29530 +               seq_printf(m, fmt, mfs->mfsrr_watermark);
29531 +               break;
29532 +       case AuWbrCreate_MFSV:
29533 +       case AuWbrCreate_PMFSV:
29534 +               au_gen_fmt(fmt, sizeof(fmt), pat, "%lu");
29535 +               seq_printf(m, fmt,
29536 +                          jiffies_to_msecs(mfs->mfs_expire)
29537 +                          / MSEC_PER_SEC);
29538 +               break;
29539 +       case AuWbrCreate_MFSRRV:
29540 +       case AuWbrCreate_TDMFSV:
29541 +       case AuWbrCreate_PMFSRRV:
29542 +               au_gen_fmt(fmt, sizeof(fmt), pat, "%llu:%lu");
29543 +               seq_printf(m, fmt, mfs->mfsrr_watermark,
29544 +                          jiffies_to_msecs(mfs->mfs_expire) / MSEC_PER_SEC);
29545 +               break;
29546 +       default:
29547 +               BUG();
29548 +       }
29549 +}
29550 +
29551 +static int au_show_xino(struct seq_file *seq, struct super_block *sb)
29552 +{
29553 +#ifdef CONFIG_SYSFS
29554 +       return 0;
29555 +#else
29556 +       int err;
29557 +       const int len = sizeof(AUFS_XINO_FNAME) - 1;
29558 +       aufs_bindex_t bindex, brid;
29559 +       struct qstr *name;
29560 +       struct file *f;
29561 +       struct dentry *d, *h_root;
29562 +       struct au_branch *br;
29563 +
29564 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
29565 +
29566 +       err = 0;
29567 +       f = au_sbi(sb)->si_xib;
29568 +       if (!f)
29569 +               goto out;
29570 +
29571 +       /* stop printing the default xino path on the first writable branch */
29572 +       h_root = NULL;
29573 +       bindex = au_xi_root(sb, f->f_path.dentry);
29574 +       if (bindex >= 0) {
29575 +               br = au_sbr_sb(sb, bindex);
29576 +               h_root = au_br_dentry(br);
29577 +       }
29578 +
29579 +       d = f->f_path.dentry;
29580 +       name = &d->d_name;
29581 +       /* safe ->d_parent because the file is unlinked */
29582 +       if (d->d_parent == h_root
29583 +           && name->len == len
29584 +           && !memcmp(name->name, AUFS_XINO_FNAME, len))
29585 +               goto out;
29586 +
29587 +       seq_puts(seq, ",xino=");
29588 +       err = au_xino_path(seq, f);
29589 +
29590 +out:
29591 +       return err;
29592 +#endif
29593 +}
29594 +
29595 +/* seq_file will re-call me in case of too long string */
29596 +static int aufs_show_options(struct seq_file *m, struct dentry *dentry)
29597 +{
29598 +       int err;
29599 +       unsigned int mnt_flags, v;
29600 +       struct super_block *sb;
29601 +       struct au_sbinfo *sbinfo;
29602 +
29603 +#define AuBool(name, str) do { \
29604 +       v = au_opt_test(mnt_flags, name); \
29605 +       if (v != au_opt_test(AuOpt_Def, name)) \
29606 +               seq_printf(m, ",%s" #str, v ? "" : "no"); \
29607 +} while (0)
29608 +
29609 +#define AuStr(name, str) do { \
29610 +       v = mnt_flags & AuOptMask_##name; \
29611 +       if (v != (AuOpt_Def & AuOptMask_##name)) \
29612 +               seq_printf(m, "," #str "=%s", au_optstr_##str(v)); \
29613 +} while (0)
29614 +
29615 +#define AuUInt(name, str, val) do { \
29616 +       if (val != AUFS_##name##_DEF) \
29617 +               seq_printf(m, "," #str "=%u", val); \
29618 +} while (0)
29619 +
29620 +       sb = dentry->d_sb;
29621 +       if (sb->s_flags & SB_POSIXACL)
29622 +               seq_puts(m, ",acl");
29623 +#if 0 /* reserved for future use */
29624 +       if (sb->s_flags & SB_I_VERSION)
29625 +               seq_puts(m, ",i_version");
29626 +#endif
29627 +
29628 +       /* lock free root dinfo */
29629 +       si_noflush_read_lock(sb);
29630 +       sbinfo = au_sbi(sb);
29631 +       seq_printf(m, ",si=%lx", sysaufs_si_id(sbinfo));
29632 +
29633 +       mnt_flags = au_mntflags(sb);
29634 +       if (au_opt_test(mnt_flags, XINO)) {
29635 +               err = au_show_xino(m, sb);
29636 +               if (unlikely(err))
29637 +                       goto out;
29638 +       } else
29639 +               seq_puts(m, ",noxino");
29640 +
29641 +       AuBool(TRUNC_XINO, trunc_xino);
29642 +       AuStr(UDBA, udba);
29643 +       AuBool(SHWH, shwh);
29644 +       AuBool(PLINK, plink);
29645 +       AuBool(DIO, dio);
29646 +       AuBool(DIRPERM1, dirperm1);
29647 +
29648 +       v = sbinfo->si_wbr_create;
29649 +       if (v != AuWbrCreate_Def)
29650 +               au_show_wbr_create(m, v, sbinfo);
29651 +
29652 +       v = sbinfo->si_wbr_copyup;
29653 +       if (v != AuWbrCopyup_Def)
29654 +               seq_printf(m, ",cpup=%s", au_optstr_wbr_copyup(v));
29655 +
29656 +       v = au_opt_test(mnt_flags, ALWAYS_DIROPQ);
29657 +       if (v != au_opt_test(AuOpt_Def, ALWAYS_DIROPQ))
29658 +               seq_printf(m, ",diropq=%c", v ? 'a' : 'w');
29659 +
29660 +       AuUInt(DIRWH, dirwh, sbinfo->si_dirwh);
29661 +
29662 +       v = jiffies_to_msecs(sbinfo->si_rdcache) / MSEC_PER_SEC;
29663 +       AuUInt(RDCACHE, rdcache, v);
29664 +
29665 +       AuUInt(RDBLK, rdblk, sbinfo->si_rdblk);
29666 +       AuUInt(RDHASH, rdhash, sbinfo->si_rdhash);
29667 +
29668 +       au_fhsm_show(m, sbinfo);
29669 +
29670 +       AuBool(DIRREN, dirren);
29671 +       AuBool(SUM, sum);
29672 +       /* AuBool(SUM_W, wsum); */
29673 +       AuBool(WARN_PERM, warn_perm);
29674 +       AuBool(VERBOSE, verbose);
29675 +
29676 +out:
29677 +       /* be sure to print "br:" last */
29678 +       if (!sysaufs_brs) {
29679 +               seq_puts(m, ",br:");
29680 +               au_show_brs(m, sb);
29681 +       }
29682 +       si_read_unlock(sb);
29683 +       return 0;
29684 +
29685 +#undef AuBool
29686 +#undef AuStr
29687 +#undef AuUInt
29688 +}
29689 +
29690 +/* ---------------------------------------------------------------------- */
29691 +
29692 +/* sum mode which returns the summation for statfs(2) */
29693 +
29694 +static u64 au_add_till_max(u64 a, u64 b)
29695 +{
29696 +       u64 old;
29697 +
29698 +       old = a;
29699 +       a += b;
29700 +       if (old <= a)
29701 +               return a;
29702 +       return ULLONG_MAX;
29703 +}
29704 +
29705 +static u64 au_mul_till_max(u64 a, long mul)
29706 +{
29707 +       u64 old;
29708 +
29709 +       old = a;
29710 +       a *= mul;
29711 +       if (old <= a)
29712 +               return a;
29713 +       return ULLONG_MAX;
29714 +}
29715 +
29716 +static int au_statfs_sum(struct super_block *sb, struct kstatfs *buf)
29717 +{
29718 +       int err;
29719 +       long bsize, factor;
29720 +       u64 blocks, bfree, bavail, files, ffree;
29721 +       aufs_bindex_t bbot, bindex, i;
29722 +       unsigned char shared;
29723 +       struct path h_path;
29724 +       struct super_block *h_sb;
29725 +
29726 +       err = 0;
29727 +       bsize = LONG_MAX;
29728 +       files = 0;
29729 +       ffree = 0;
29730 +       blocks = 0;
29731 +       bfree = 0;
29732 +       bavail = 0;
29733 +       bbot = au_sbbot(sb);
29734 +       for (bindex = 0; bindex <= bbot; bindex++) {
29735 +               h_path.mnt = au_sbr_mnt(sb, bindex);
29736 +               h_sb = h_path.mnt->mnt_sb;
29737 +               shared = 0;
29738 +               for (i = 0; !shared && i < bindex; i++)
29739 +                       shared = (au_sbr_sb(sb, i) == h_sb);
29740 +               if (shared)
29741 +                       continue;
29742 +
29743 +               /* sb->s_root for NFS is unreliable */
29744 +               h_path.dentry = h_path.mnt->mnt_root;
29745 +               err = vfs_statfs(&h_path, buf);
29746 +               if (unlikely(err))
29747 +                       goto out;
29748 +
29749 +               if (bsize > buf->f_bsize) {
29750 +                       /*
29751 +                        * we will reduce bsize, so we have to expand blocks
29752 +                        * etc. to match them again
29753 +                        */
29754 +                       factor = (bsize / buf->f_bsize);
29755 +                       blocks = au_mul_till_max(blocks, factor);
29756 +                       bfree = au_mul_till_max(bfree, factor);
29757 +                       bavail = au_mul_till_max(bavail, factor);
29758 +                       bsize = buf->f_bsize;
29759 +               }
29760 +
29761 +               factor = (buf->f_bsize / bsize);
29762 +               blocks = au_add_till_max(blocks,
29763 +                               au_mul_till_max(buf->f_blocks, factor));
29764 +               bfree = au_add_till_max(bfree,
29765 +                               au_mul_till_max(buf->f_bfree, factor));
29766 +               bavail = au_add_till_max(bavail,
29767 +                               au_mul_till_max(buf->f_bavail, factor));
29768 +               files = au_add_till_max(files, buf->f_files);
29769 +               ffree = au_add_till_max(ffree, buf->f_ffree);
29770 +       }
29771 +
29772 +       buf->f_bsize = bsize;
29773 +       buf->f_blocks = blocks;
29774 +       buf->f_bfree = bfree;
29775 +       buf->f_bavail = bavail;
29776 +       buf->f_files = files;
29777 +       buf->f_ffree = ffree;
29778 +       buf->f_frsize = 0;
29779 +
29780 +out:
29781 +       return err;
29782 +}
29783 +
29784 +static int aufs_statfs(struct dentry *dentry, struct kstatfs *buf)
29785 +{
29786 +       int err;
29787 +       struct path h_path;
29788 +       struct super_block *sb;
29789 +
29790 +       /* lock free root dinfo */
29791 +       sb = dentry->d_sb;
29792 +       si_noflush_read_lock(sb);
29793 +       if (!au_opt_test(au_mntflags(sb), SUM)) {
29794 +               /* sb->s_root for NFS is unreliable */
29795 +               h_path.mnt = au_sbr_mnt(sb, 0);
29796 +               h_path.dentry = h_path.mnt->mnt_root;
29797 +               err = vfs_statfs(&h_path, buf);
29798 +       } else
29799 +               err = au_statfs_sum(sb, buf);
29800 +       si_read_unlock(sb);
29801 +
29802 +       if (!err) {
29803 +               buf->f_type = AUFS_SUPER_MAGIC;
29804 +               buf->f_namelen = AUFS_MAX_NAMELEN;
29805 +               memset(&buf->f_fsid, 0, sizeof(buf->f_fsid));
29806 +       }
29807 +       /* buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1; */
29808 +
29809 +       return err;
29810 +}
29811 +
29812 +/* ---------------------------------------------------------------------- */
29813 +
29814 +static int aufs_sync_fs(struct super_block *sb, int wait)
29815 +{
29816 +       int err, e;
29817 +       aufs_bindex_t bbot, bindex;
29818 +       struct au_branch *br;
29819 +       struct super_block *h_sb;
29820 +
29821 +       err = 0;
29822 +       si_noflush_read_lock(sb);
29823 +       bbot = au_sbbot(sb);
29824 +       for (bindex = 0; bindex <= bbot; bindex++) {
29825 +               br = au_sbr(sb, bindex);
29826 +               if (!au_br_writable(br->br_perm))
29827 +                       continue;
29828 +
29829 +               h_sb = au_sbr_sb(sb, bindex);
29830 +               e = vfsub_sync_filesystem(h_sb, wait);
29831 +               if (unlikely(e && !err))
29832 +                       err = e;
29833 +               /* go on even if an error happens */
29834 +       }
29835 +       si_read_unlock(sb);
29836 +
29837 +       return err;
29838 +}
29839 +
29840 +/* ---------------------------------------------------------------------- */
29841 +
29842 +/* final actions when unmounting a file system */
29843 +static void aufs_put_super(struct super_block *sb)
29844 +{
29845 +       struct au_sbinfo *sbinfo;
29846 +
29847 +       sbinfo = au_sbi(sb);
29848 +       if (sbinfo)
29849 +               kobject_put(&sbinfo->si_kobj);
29850 +}
29851 +
29852 +/* ---------------------------------------------------------------------- */
29853 +
29854 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb,
29855 +                    struct super_block *sb, void *arg)
29856 +{
29857 +       void *array;
29858 +       unsigned long long n, sz;
29859 +
29860 +       array = NULL;
29861 +       n = 0;
29862 +       if (!*hint)
29863 +               goto out;
29864 +
29865 +       if (*hint > ULLONG_MAX / sizeof(array)) {
29866 +               array = ERR_PTR(-EMFILE);
29867 +               pr_err("hint %llu\n", *hint);
29868 +               goto out;
29869 +       }
29870 +
29871 +       sz = sizeof(array) * *hint;
29872 +       array = kzalloc(sz, GFP_NOFS);
29873 +       if (unlikely(!array))
29874 +               array = vzalloc(sz);
29875 +       if (unlikely(!array)) {
29876 +               array = ERR_PTR(-ENOMEM);
29877 +               goto out;
29878 +       }
29879 +
29880 +       n = cb(sb, array, *hint, arg);
29881 +       AuDebugOn(n > *hint);
29882 +
29883 +out:
29884 +       *hint = n;
29885 +       return array;
29886 +}
29887 +
29888 +static unsigned long long au_iarray_cb(struct super_block *sb, void *a,
29889 +                                      unsigned long long max __maybe_unused,
29890 +                                      void *arg)
29891 +{
29892 +       unsigned long long n;
29893 +       struct inode **p, *inode;
29894 +       struct list_head *head;
29895 +
29896 +       n = 0;
29897 +       p = a;
29898 +       head = arg;
29899 +       spin_lock(&sb->s_inode_list_lock);
29900 +       list_for_each_entry(inode, head, i_sb_list) {
29901 +               if (!au_is_bad_inode(inode)
29902 +                   && au_ii(inode)->ii_btop >= 0) {
29903 +                       spin_lock(&inode->i_lock);
29904 +                       if (atomic_read(&inode->i_count)) {
29905 +                               au_igrab(inode);
29906 +                               *p++ = inode;
29907 +                               n++;
29908 +                               AuDebugOn(n > max);
29909 +                       }
29910 +                       spin_unlock(&inode->i_lock);
29911 +               }
29912 +       }
29913 +       spin_unlock(&sb->s_inode_list_lock);
29914 +
29915 +       return n;
29916 +}
29917 +
29918 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max)
29919 +{
29920 +       struct au_sbinfo *sbi;
29921 +
29922 +       sbi = au_sbi(sb);
29923 +       *max = au_lcnt_read(&sbi->si_ninodes, /*do_rev*/1);
29924 +       return au_array_alloc(max, au_iarray_cb, sb, &sb->s_inodes);
29925 +}
29926 +
29927 +void au_iarray_free(struct inode **a, unsigned long long max)
29928 +{
29929 +       unsigned long long ull;
29930 +
29931 +       for (ull = 0; ull < max; ull++)
29932 +               iput(a[ull]);
29933 +       kvfree(a);
29934 +}
29935 +
29936 +/* ---------------------------------------------------------------------- */
29937 +
29938 +/*
29939 + * refresh dentry and inode at remount time.
29940 + */
29941 +/* todo: consolidate with simple_reval_dpath() and au_reval_for_attr() */
29942 +static int au_do_refresh(struct dentry *dentry, unsigned int dir_flags,
29943 +                     struct dentry *parent)
29944 +{
29945 +       int err;
29946 +
29947 +       di_write_lock_child(dentry);
29948 +       di_read_lock_parent(parent, AuLock_IR);
29949 +       err = au_refresh_dentry(dentry, parent);
29950 +       if (!err && dir_flags)
29951 +               au_hn_reset(d_inode(dentry), dir_flags);
29952 +       di_read_unlock(parent, AuLock_IR);
29953 +       di_write_unlock(dentry);
29954 +
29955 +       return err;
29956 +}
29957 +
29958 +static int au_do_refresh_d(struct dentry *dentry, unsigned int sigen,
29959 +                          struct au_sbinfo *sbinfo,
29960 +                          const unsigned int dir_flags, unsigned int do_idop)
29961 +{
29962 +       int err;
29963 +       struct dentry *parent;
29964 +
29965 +       err = 0;
29966 +       parent = dget_parent(dentry);
29967 +       if (!au_digen_test(parent, sigen) && au_digen_test(dentry, sigen)) {
29968 +               if (d_really_is_positive(dentry)) {
29969 +                       if (!d_is_dir(dentry))
29970 +                               err = au_do_refresh(dentry, /*dir_flags*/0,
29971 +                                                parent);
29972 +                       else {
29973 +                               err = au_do_refresh(dentry, dir_flags, parent);
29974 +                               if (unlikely(err))
29975 +                                       au_fset_si(sbinfo, FAILED_REFRESH_DIR);
29976 +                       }
29977 +               } else
29978 +                       err = au_do_refresh(dentry, /*dir_flags*/0, parent);
29979 +               AuDbgDentry(dentry);
29980 +       }
29981 +       dput(parent);
29982 +
29983 +       if (!err) {
29984 +               if (do_idop)
29985 +                       au_refresh_dop(dentry, /*force_reval*/0);
29986 +       } else
29987 +               au_refresh_dop(dentry, /*force_reval*/1);
29988 +
29989 +       AuTraceErr(err);
29990 +       return err;
29991 +}
29992 +
29993 +static int au_refresh_d(struct super_block *sb, unsigned int do_idop)
29994 +{
29995 +       int err, i, j, ndentry, e;
29996 +       unsigned int sigen;
29997 +       struct au_dcsub_pages dpages;
29998 +       struct au_dpage *dpage;
29999 +       struct dentry **dentries, *d;
30000 +       struct au_sbinfo *sbinfo;
30001 +       struct dentry *root = sb->s_root;
30002 +       const unsigned int dir_flags = au_hi_flags(d_inode(root), /*isdir*/1);
30003 +
30004 +       if (do_idop)
30005 +               au_refresh_dop(root, /*force_reval*/0);
30006 +
30007 +       err = au_dpages_init(&dpages, GFP_NOFS);
30008 +       if (unlikely(err))
30009 +               goto out;
30010 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
30011 +       if (unlikely(err))
30012 +               goto out_dpages;
30013 +
30014 +       sigen = au_sigen(sb);
30015 +       sbinfo = au_sbi(sb);
30016 +       for (i = 0; i < dpages.ndpage; i++) {
30017 +               dpage = dpages.dpages + i;
30018 +               dentries = dpage->dentries;
30019 +               ndentry = dpage->ndentry;
30020 +               for (j = 0; j < ndentry; j++) {
30021 +                       d = dentries[j];
30022 +                       e = au_do_refresh_d(d, sigen, sbinfo, dir_flags,
30023 +                                           do_idop);
30024 +                       if (unlikely(e && !err))
30025 +                               err = e;
30026 +                       /* go on even err */
30027 +               }
30028 +       }
30029 +
30030 +out_dpages:
30031 +       au_dpages_free(&dpages);
30032 +out:
30033 +       return err;
30034 +}
30035 +
30036 +static int au_refresh_i(struct super_block *sb, unsigned int do_idop)
30037 +{
30038 +       int err, e;
30039 +       unsigned int sigen;
30040 +       unsigned long long max, ull;
30041 +       struct inode *inode, **array;
30042 +
30043 +       array = au_iarray_alloc(sb, &max);
30044 +       err = PTR_ERR(array);
30045 +       if (IS_ERR(array))
30046 +               goto out;
30047 +
30048 +       err = 0;
30049 +       sigen = au_sigen(sb);
30050 +       for (ull = 0; ull < max; ull++) {
30051 +               inode = array[ull];
30052 +               if (unlikely(!inode))
30053 +                       break;
30054 +
30055 +               e = 0;
30056 +               ii_write_lock_child(inode);
30057 +               if (au_iigen(inode, NULL) != sigen) {
30058 +                       e = au_refresh_hinode_self(inode);
30059 +                       if (unlikely(e)) {
30060 +                               au_refresh_iop(inode, /*force_getattr*/1);
30061 +                               pr_err("error %d, i%lu\n", e, inode->i_ino);
30062 +                               if (!err)
30063 +                                       err = e;
30064 +                               /* go on even if err */
30065 +                       }
30066 +               }
30067 +               if (!e && do_idop)
30068 +                       au_refresh_iop(inode, /*force_getattr*/0);
30069 +               ii_write_unlock(inode);
30070 +       }
30071 +
30072 +       au_iarray_free(array, max);
30073 +
30074 +out:
30075 +       return err;
30076 +}
30077 +
30078 +static void au_remount_refresh(struct super_block *sb, unsigned int do_idop)
30079 +{
30080 +       int err, e;
30081 +       unsigned int udba;
30082 +       aufs_bindex_t bindex, bbot;
30083 +       struct dentry *root;
30084 +       struct inode *inode;
30085 +       struct au_branch *br;
30086 +       struct au_sbinfo *sbi;
30087 +
30088 +       au_sigen_inc(sb);
30089 +       sbi = au_sbi(sb);
30090 +       au_fclr_si(sbi, FAILED_REFRESH_DIR);
30091 +
30092 +       root = sb->s_root;
30093 +       DiMustNoWaiters(root);
30094 +       inode = d_inode(root);
30095 +       IiMustNoWaiters(inode);
30096 +
30097 +       udba = au_opt_udba(sb);
30098 +       bbot = au_sbbot(sb);
30099 +       for (bindex = 0; bindex <= bbot; bindex++) {
30100 +               br = au_sbr(sb, bindex);
30101 +               err = au_hnotify_reset_br(udba, br, br->br_perm);
30102 +               if (unlikely(err))
30103 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
30104 +                               bindex, err);
30105 +               /* go on even if err */
30106 +       }
30107 +       au_hn_reset(inode, au_hi_flags(inode, /*isdir*/1));
30108 +
30109 +       if (do_idop) {
30110 +               if (au_ftest_si(sbi, NO_DREVAL)) {
30111 +                       AuDebugOn(sb->s_d_op == &aufs_dop_noreval);
30112 +                       sb->s_d_op = &aufs_dop_noreval;
30113 +                       AuDebugOn(sbi->si_iop_array == aufs_iop_nogetattr);
30114 +                       sbi->si_iop_array = aufs_iop_nogetattr;
30115 +               } else {
30116 +                       AuDebugOn(sb->s_d_op == &aufs_dop);
30117 +                       sb->s_d_op = &aufs_dop;
30118 +                       AuDebugOn(sbi->si_iop_array == aufs_iop);
30119 +                       sbi->si_iop_array = aufs_iop;
30120 +               }
30121 +               pr_info("reset to %ps and %ps\n",
30122 +                       sb->s_d_op, sbi->si_iop_array);
30123 +       }
30124 +
30125 +       di_write_unlock(root);
30126 +       err = au_refresh_d(sb, do_idop);
30127 +       e = au_refresh_i(sb, do_idop);
30128 +       if (unlikely(e && !err))
30129 +               err = e;
30130 +       /* aufs_write_lock() calls ..._child() */
30131 +       di_write_lock_child(root);
30132 +
30133 +       au_cpup_attr_all(inode, /*force*/1);
30134 +
30135 +       if (unlikely(err))
30136 +               AuIOErr("refresh failed, ignored, %d\n", err);
30137 +}
30138 +
30139 +/* stop extra interpretation of errno in mount(8), and strange error messages */
30140 +static int cvt_err(int err)
30141 +{
30142 +       AuTraceErr(err);
30143 +
30144 +       switch (err) {
30145 +       case -ENOENT:
30146 +       case -ENOTDIR:
30147 +       case -EEXIST:
30148 +       case -EIO:
30149 +               err = -EINVAL;
30150 +       }
30151 +       return err;
30152 +}
30153 +
30154 +static int aufs_remount_fs(struct super_block *sb, int *flags, char *data)
30155 +{
30156 +       int err, do_dx;
30157 +       unsigned int mntflags;
30158 +       struct au_opts opts = {
30159 +               .opt = NULL
30160 +       };
30161 +       struct dentry *root;
30162 +       struct inode *inode;
30163 +       struct au_sbinfo *sbinfo;
30164 +
30165 +       err = 0;
30166 +       root = sb->s_root;
30167 +       if (!data || !*data) {
30168 +               err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
30169 +               if (!err) {
30170 +                       di_write_lock_child(root);
30171 +                       err = au_opts_verify(sb, *flags, /*pending*/0);
30172 +                       aufs_write_unlock(root);
30173 +               }
30174 +               goto out;
30175 +       }
30176 +
30177 +       err = -ENOMEM;
30178 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
30179 +       if (unlikely(!opts.opt))
30180 +               goto out;
30181 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
30182 +       opts.flags = AuOpts_REMOUNT;
30183 +       opts.sb_flags = *flags;
30184 +
30185 +       /* parse it before aufs lock */
30186 +       err = au_opts_parse(sb, data, &opts);
30187 +       if (unlikely(err))
30188 +               goto out_opts;
30189 +
30190 +       sbinfo = au_sbi(sb);
30191 +       inode = d_inode(root);
30192 +       inode_lock(inode);
30193 +       err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
30194 +       if (unlikely(err))
30195 +               goto out_mtx;
30196 +       di_write_lock_child(root);
30197 +
30198 +       /* au_opts_remount() may return an error */
30199 +       err = au_opts_remount(sb, &opts);
30200 +       au_opts_free(&opts);
30201 +
30202 +       if (au_ftest_opts(opts.flags, REFRESH))
30203 +               au_remount_refresh(sb, au_ftest_opts(opts.flags, REFRESH_IDOP));
30204 +
30205 +       if (au_ftest_opts(opts.flags, REFRESH_DYAOP)) {
30206 +               mntflags = au_mntflags(sb);
30207 +               do_dx = !!au_opt_test(mntflags, DIO);
30208 +               au_dy_arefresh(do_dx);
30209 +       }
30210 +
30211 +       au_fhsm_wrote_all(sb, /*force*/1); /* ?? */
30212 +       aufs_write_unlock(root);
30213 +
30214 +out_mtx:
30215 +       inode_unlock(inode);
30216 +out_opts:
30217 +       free_page((unsigned long)opts.opt);
30218 +out:
30219 +       err = cvt_err(err);
30220 +       AuTraceErr(err);
30221 +       return err;
30222 +}
30223 +
30224 +static const struct super_operations aufs_sop = {
30225 +       .alloc_inode    = aufs_alloc_inode,
30226 +       .destroy_inode  = aufs_destroy_inode,
30227 +       .free_inode     = aufs_free_inode,
30228 +       /* always deleting, no clearing */
30229 +       .drop_inode     = generic_delete_inode,
30230 +       .show_options   = aufs_show_options,
30231 +       .statfs         = aufs_statfs,
30232 +       .put_super      = aufs_put_super,
30233 +       .sync_fs        = aufs_sync_fs,
30234 +       .remount_fs     = aufs_remount_fs
30235 +};
30236 +
30237 +/* ---------------------------------------------------------------------- */
30238 +
30239 +static int alloc_root(struct super_block *sb)
30240 +{
30241 +       int err;
30242 +       struct inode *inode;
30243 +       struct dentry *root;
30244 +
30245 +       err = -ENOMEM;
30246 +       inode = au_iget_locked(sb, AUFS_ROOT_INO);
30247 +       err = PTR_ERR(inode);
30248 +       if (IS_ERR(inode))
30249 +               goto out;
30250 +
30251 +       inode->i_op = aufs_iop + AuIop_DIR; /* with getattr by default */
30252 +       inode->i_fop = &aufs_dir_fop;
30253 +       inode->i_mode = S_IFDIR;
30254 +       set_nlink(inode, 2);
30255 +       unlock_new_inode(inode);
30256 +
30257 +       root = d_make_root(inode);
30258 +       if (unlikely(!root))
30259 +               goto out;
30260 +       err = PTR_ERR(root);
30261 +       if (IS_ERR(root))
30262 +               goto out;
30263 +
30264 +       err = au_di_init(root);
30265 +       if (!err) {
30266 +               sb->s_root = root;
30267 +               return 0; /* success */
30268 +       }
30269 +       dput(root);
30270 +
30271 +out:
30272 +       return err;
30273 +}
30274 +
30275 +static int aufs_fill_super(struct super_block *sb, void *raw_data,
30276 +                          int silent __maybe_unused)
30277 +{
30278 +       int err;
30279 +       struct au_opts opts = {
30280 +               .opt = NULL
30281 +       };
30282 +       struct au_sbinfo *sbinfo;
30283 +       struct dentry *root;
30284 +       struct inode *inode;
30285 +       char *arg = raw_data;
30286 +
30287 +       if (unlikely(!arg || !*arg)) {
30288 +               err = -EINVAL;
30289 +               pr_err("no arg\n");
30290 +               goto out;
30291 +       }
30292 +
30293 +       err = -ENOMEM;
30294 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
30295 +       if (unlikely(!opts.opt))
30296 +               goto out;
30297 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
30298 +       opts.sb_flags = sb->s_flags;
30299 +
30300 +       err = au_si_alloc(sb);
30301 +       if (unlikely(err))
30302 +               goto out_opts;
30303 +       sbinfo = au_sbi(sb);
30304 +
30305 +       /* all timestamps always follow the ones on the branch */
30306 +       sb->s_flags |= SB_NOATIME | SB_NODIRATIME;
30307 +       sb->s_flags |= SB_I_VERSION; /* do we really need this? */
30308 +       sb->s_op = &aufs_sop;
30309 +       sb->s_d_op = &aufs_dop;
30310 +       sb->s_magic = AUFS_SUPER_MAGIC;
30311 +       sb->s_maxbytes = 0;
30312 +       sb->s_stack_depth = 1;
30313 +       au_export_init(sb);
30314 +       au_xattr_init(sb);
30315 +
30316 +       err = alloc_root(sb);
30317 +       if (unlikely(err)) {
30318 +               si_write_unlock(sb);
30319 +               goto out_info;
30320 +       }
30321 +       root = sb->s_root;
30322 +       inode = d_inode(root);
30323 +
30324 +       /*
30325 +        * actually we can parse options regardless aufs lock here.
30326 +        * but at remount time, parsing must be done before aufs lock.
30327 +        * so we follow the same rule.
30328 +        */
30329 +       ii_write_lock_parent(inode);
30330 +       aufs_write_unlock(root);
30331 +       err = au_opts_parse(sb, arg, &opts);
30332 +       if (unlikely(err))
30333 +               goto out_root;
30334 +
30335 +       /* lock vfs_inode first, then aufs. */
30336 +       inode_lock(inode);
30337 +       aufs_write_lock(root);
30338 +       err = au_opts_mount(sb, &opts);
30339 +       au_opts_free(&opts);
30340 +       if (!err && au_ftest_si(sbinfo, NO_DREVAL)) {
30341 +               sb->s_d_op = &aufs_dop_noreval;
30342 +               pr_info("%ps\n", sb->s_d_op);
30343 +               au_refresh_dop(root, /*force_reval*/0);
30344 +               sbinfo->si_iop_array = aufs_iop_nogetattr;
30345 +               au_refresh_iop(inode, /*force_getattr*/0);
30346 +       }
30347 +       aufs_write_unlock(root);
30348 +       inode_unlock(inode);
30349 +       if (!err)
30350 +               goto out_opts; /* success */
30351 +
30352 +out_root:
30353 +       dput(root);
30354 +       sb->s_root = NULL;
30355 +out_info:
30356 +       kobject_put(&sbinfo->si_kobj);
30357 +       sb->s_fs_info = NULL;
30358 +out_opts:
30359 +       free_page((unsigned long)opts.opt);
30360 +out:
30361 +       AuTraceErr(err);
30362 +       err = cvt_err(err);
30363 +       AuTraceErr(err);
30364 +       return err;
30365 +}
30366 +
30367 +/* ---------------------------------------------------------------------- */
30368 +
30369 +static struct dentry *aufs_mount(struct file_system_type *fs_type, int flags,
30370 +                                const char *dev_name __maybe_unused,
30371 +                                void *raw_data)
30372 +{
30373 +       struct dentry *root;
30374 +
30375 +       /* all timestamps always follow the ones on the branch */
30376 +       /* mnt->mnt_flags |= MNT_NOATIME | MNT_NODIRATIME; */
30377 +       root = mount_nodev(fs_type, flags, raw_data, aufs_fill_super);
30378 +       if (IS_ERR(root))
30379 +               goto out;
30380 +
30381 +       au_sbilist_add(root->d_sb);
30382 +
30383 +out:
30384 +       return root;
30385 +}
30386 +
30387 +static void aufs_kill_sb(struct super_block *sb)
30388 +{
30389 +       struct au_sbinfo *sbinfo;
30390 +
30391 +       sbinfo = au_sbi(sb);
30392 +       if (sbinfo) {
30393 +               au_sbilist_del(sb);
30394 +               aufs_write_lock(sb->s_root);
30395 +               au_fhsm_fin(sb);
30396 +               if (sbinfo->si_wbr_create_ops->fin)
30397 +                       sbinfo->si_wbr_create_ops->fin(sb);
30398 +               if (au_opt_test(sbinfo->si_mntflags, UDBA_HNOTIFY)) {
30399 +                       au_opt_set_udba(sbinfo->si_mntflags, UDBA_NONE);
30400 +                       au_remount_refresh(sb, /*do_idop*/0);
30401 +               }
30402 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
30403 +                       au_plink_put(sb, /*verbose*/1);
30404 +               au_xino_clr(sb);
30405 +               au_dr_opt_flush(sb);
30406 +               sbinfo->si_sb = NULL;
30407 +               aufs_write_unlock(sb->s_root);
30408 +               au_nwt_flush(&sbinfo->si_nowait);
30409 +       }
30410 +       kill_anon_super(sb);
30411 +}
30412 +
30413 +struct file_system_type aufs_fs_type = {
30414 +       .name           = AUFS_FSTYPE,
30415 +       /* a race between rename and others */
30416 +       .fs_flags       = FS_RENAME_DOES_D_MOVE
30417 +                               /* untested */
30418 +                               /*| FS_ALLOW_IDMAP*/
30419 +                               ,
30420 +       .mount          = aufs_mount,
30421 +       .kill_sb        = aufs_kill_sb,
30422 +       /* no need to __module_get() and module_put(). */
30423 +       .owner          = THIS_MODULE,
30424 +};
30425 diff -urN /usr/share/empty/fs/aufs/super.h linux/fs/aufs/super.h
30426 --- /usr/share/empty/fs/aufs/super.h    1970-01-01 01:00:00.000000000 +0100
30427 +++ linux/fs/aufs/super.h       2021-12-03 15:38:59.939980643 +0100
30428 @@ -0,0 +1,587 @@
30429 +/* SPDX-License-Identifier: GPL-2.0 */
30430 +/*
30431 + * Copyright (C) 2005-2021 Junjiro R. Okajima
30432 + *
30433 + * This program, aufs is free software; you can redistribute it and/or modify
30434 + * it under the terms of the GNU General Public License as published by
30435 + * the Free Software Foundation; either version 2 of the License, or
30436 + * (at your option) any later version.
30437 + *
30438 + * This program is distributed in the hope that it will be useful,
30439 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30440 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30441 + * GNU General Public License for more details.
30442 + *
30443 + * You should have received a copy of the GNU General Public License
30444 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30445 + */
30446 +
30447 +/*
30448 + * super_block operations
30449 + */
30450 +
30451 +#ifndef __AUFS_SUPER_H__
30452 +#define __AUFS_SUPER_H__
30453 +
30454 +#ifdef __KERNEL__
30455 +
30456 +#include <linux/fs.h>
30457 +#include <linux/kobject.h>
30458 +#include "hbl.h"
30459 +#include "lcnt.h"
30460 +#include "rwsem.h"
30461 +#include "wkq.h"
30462 +
30463 +/* policies to select one among multiple writable branches */
30464 +struct au_wbr_copyup_operations {
30465 +       int (*copyup)(struct dentry *dentry);
30466 +};
30467 +
30468 +#define AuWbr_DIR      1               /* target is a dir */
30469 +#define AuWbr_PARENT   (1 << 1)        /* always require a parent */
30470 +
30471 +#define au_ftest_wbr(flags, name)      ((flags) & AuWbr_##name)
30472 +#define au_fset_wbr(flags, name)       { (flags) |= AuWbr_##name; }
30473 +#define au_fclr_wbr(flags, name)       { (flags) &= ~AuWbr_##name; }
30474 +
30475 +struct au_wbr_create_operations {
30476 +       int (*create)(struct dentry *dentry, unsigned int flags);
30477 +       int (*init)(struct super_block *sb);
30478 +       int (*fin)(struct super_block *sb);
30479 +};
30480 +
30481 +struct au_wbr_mfs {
30482 +       struct mutex    mfs_lock; /* protect this structure */
30483 +       unsigned long   mfs_jiffy;
30484 +       unsigned long   mfs_expire;
30485 +       aufs_bindex_t   mfs_bindex;
30486 +
30487 +       unsigned long long      mfsrr_bytes;
30488 +       unsigned long long      mfsrr_watermark;
30489 +};
30490 +
30491 +#define AuPlink_NHASH 100
30492 +static inline int au_plink_hash(ino_t ino)
30493 +{
30494 +       return ino % AuPlink_NHASH;
30495 +}
30496 +
30497 +/* File-based Hierarchical Storage Management */
30498 +struct au_fhsm {
30499 +#ifdef CONFIG_AUFS_FHSM
30500 +       /* allow only one process who can receive the notification */
30501 +       spinlock_t              fhsm_spin;
30502 +       pid_t                   fhsm_pid;
30503 +       wait_queue_head_t       fhsm_wqh;
30504 +       atomic_t                fhsm_readable;
30505 +
30506 +       /* these are protected by si_rwsem */
30507 +       unsigned long           fhsm_expire;
30508 +       aufs_bindex_t           fhsm_bottom;
30509 +#endif
30510 +};
30511 +
30512 +struct au_branch;
30513 +struct au_sbinfo {
30514 +       /* nowait tasks in the system-wide workqueue */
30515 +       struct au_nowait_tasks  si_nowait;
30516 +
30517 +       /*
30518 +        * tried sb->s_umount, but failed due to the dependency between i_mutex.
30519 +        * rwsem for au_sbinfo is necessary.
30520 +        */
30521 +       struct au_rwsem         si_rwsem;
30522 +
30523 +       /*
30524 +        * dirty approach to protect sb->sb_inodes and ->s_files (gone) from
30525 +        * remount.
30526 +        */
30527 +       au_lcnt_t               si_ninodes, si_nfiles;
30528 +
30529 +       /* branch management */
30530 +       unsigned int            si_generation;
30531 +
30532 +       /* see AuSi_ flags */
30533 +       unsigned char           au_si_status;
30534 +
30535 +       aufs_bindex_t           si_bbot;
30536 +
30537 +       /* dirty trick to keep br_id plus */
30538 +       unsigned int            si_last_br_id :
30539 +                               sizeof(aufs_bindex_t) * BITS_PER_BYTE - 1;
30540 +       struct au_branch        **si_branch;
30541 +
30542 +       /* policy to select a writable branch */
30543 +       unsigned char           si_wbr_copyup;
30544 +       unsigned char           si_wbr_create;
30545 +       struct au_wbr_copyup_operations *si_wbr_copyup_ops;
30546 +       struct au_wbr_create_operations *si_wbr_create_ops;
30547 +
30548 +       /* round robin */
30549 +       atomic_t                si_wbr_rr_next;
30550 +
30551 +       /* most free space */
30552 +       struct au_wbr_mfs       si_wbr_mfs;
30553 +
30554 +       /* File-based Hierarchical Storage Management */
30555 +       struct au_fhsm          si_fhsm;
30556 +
30557 +       /* mount flags */
30558 +       /* include/asm-ia64/siginfo.h defines a macro named si_flags */
30559 +       unsigned int            si_mntflags;
30560 +
30561 +       /* external inode number (bitmap and translation table) */
30562 +       loff_t                  si_ximaxent;    /* max entries in a xino */
30563 +
30564 +       struct file             *si_xib;
30565 +       struct mutex            si_xib_mtx; /* protect xib members */
30566 +       unsigned long           *si_xib_buf;
30567 +       unsigned long           si_xib_last_pindex;
30568 +       int                     si_xib_next_bit;
30569 +
30570 +       unsigned long           si_xino_jiffy;
30571 +       unsigned long           si_xino_expire;
30572 +       /* reserved for future use */
30573 +       /* unsigned long long   si_xib_limit; */        /* Max xib file size */
30574 +
30575 +#ifdef CONFIG_AUFS_EXPORT
30576 +       /* i_generation */
30577 +       /* todo: make xigen file an array to support many inode numbers */
30578 +       struct file             *si_xigen;
30579 +       atomic_t                si_xigen_next;
30580 +#endif
30581 +
30582 +       /* dirty trick to support atomic_open */
30583 +       struct hlist_bl_head    si_aopen;
30584 +
30585 +       /* vdir parameters */
30586 +       unsigned long           si_rdcache;     /* max cache time in jiffies */
30587 +       unsigned int            si_rdblk;       /* deblk size */
30588 +       unsigned int            si_rdhash;      /* hash size */
30589 +
30590 +       /*
30591 +        * If the number of whiteouts are larger than si_dirwh, leave all of
30592 +        * them after au_whtmp_ren to reduce the cost of rmdir(2).
30593 +        * future fsck.aufs or kernel thread will remove them later.
30594 +        * Otherwise, remove all whiteouts and the dir in rmdir(2).
30595 +        */
30596 +       unsigned int            si_dirwh;
30597 +
30598 +       /* pseudo_link list */
30599 +       struct hlist_bl_head    si_plink[AuPlink_NHASH];
30600 +       wait_queue_head_t       si_plink_wq;
30601 +       spinlock_t              si_plink_maint_lock;
30602 +       pid_t                   si_plink_maint_pid;
30603 +
30604 +       /* file list */
30605 +       struct hlist_bl_head    si_files;
30606 +
30607 +       /* with/without getattr, brother of sb->s_d_op */
30608 +       const struct inode_operations *si_iop_array;
30609 +
30610 +       /*
30611 +        * sysfs and lifetime management.
30612 +        * this is not a small structure and it may be a waste of memory in case
30613 +        * of sysfs is disabled, particularly when many aufs-es are mounted.
30614 +        * but using sysfs is majority.
30615 +        */
30616 +       struct kobject          si_kobj;
30617 +#ifdef CONFIG_DEBUG_FS
30618 +       struct dentry            *si_dbgaufs;
30619 +       struct dentry            *si_dbgaufs_plink;
30620 +       struct dentry            *si_dbgaufs_xib;
30621 +#ifdef CONFIG_AUFS_EXPORT
30622 +       struct dentry            *si_dbgaufs_xigen;
30623 +#endif
30624 +#endif
30625 +
30626 +#ifdef CONFIG_AUFS_SBILIST
30627 +       struct hlist_bl_node    si_list;
30628 +#endif
30629 +
30630 +       /* dirty, necessary for unmounting, sysfs and sysrq */
30631 +       struct super_block      *si_sb;
30632 +};
30633 +
30634 +/* sbinfo status flags */
30635 +/*
30636 + * set true when refresh_dirs() failed at remount time.
30637 + * then try refreshing dirs at access time again.
30638 + * if it is false, refreshing dirs at access time is unnecessary
30639 + */
30640 +#define AuSi_FAILED_REFRESH_DIR        1
30641 +#define AuSi_FHSM              (1 << 1)        /* fhsm is active now */
30642 +#define AuSi_NO_DREVAL         (1 << 2)        /* disable all d_revalidate */
30643 +
30644 +#ifndef CONFIG_AUFS_FHSM
30645 +#undef AuSi_FHSM
30646 +#define AuSi_FHSM              0
30647 +#endif
30648 +
30649 +static inline unsigned char au_do_ftest_si(struct au_sbinfo *sbi,
30650 +                                          unsigned int flag)
30651 +{
30652 +       AuRwMustAnyLock(&sbi->si_rwsem);
30653 +       return sbi->au_si_status & flag;
30654 +}
30655 +#define au_ftest_si(sbinfo, name)      au_do_ftest_si(sbinfo, AuSi_##name)
30656 +#define au_fset_si(sbinfo, name) do { \
30657 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
30658 +       (sbinfo)->au_si_status |= AuSi_##name; \
30659 +} while (0)
30660 +#define au_fclr_si(sbinfo, name) do { \
30661 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
30662 +       (sbinfo)->au_si_status &= ~AuSi_##name; \
30663 +} while (0)
30664 +
30665 +/* ---------------------------------------------------------------------- */
30666 +
30667 +/* policy to select one among writable branches */
30668 +#define AuWbrCopyup(sbinfo, ...) \
30669 +       ((sbinfo)->si_wbr_copyup_ops->copyup(__VA_ARGS__))
30670 +#define AuWbrCreate(sbinfo, ...) \
30671 +       ((sbinfo)->si_wbr_create_ops->create(__VA_ARGS__))
30672 +
30673 +/* flags for si_read_lock()/aufs_read_lock()/di_read_lock() */
30674 +#define AuLock_DW              1               /* write-lock dentry */
30675 +#define AuLock_IR              (1 << 1)        /* read-lock inode */
30676 +#define AuLock_IW              (1 << 2)        /* write-lock inode */
30677 +#define AuLock_FLUSH           (1 << 3)        /* wait for 'nowait' tasks */
30678 +#define AuLock_DIRS            (1 << 4)        /* target is a pair of dirs */
30679 +                                               /* except RENAME_EXCHANGE */
30680 +#define AuLock_NOPLM           (1 << 5)        /* return err in plm mode */
30681 +#define AuLock_NOPLMW          (1 << 6)        /* wait for plm mode ends */
30682 +#define AuLock_GEN             (1 << 7)        /* test digen/iigen */
30683 +#define au_ftest_lock(flags, name)     ((flags) & AuLock_##name)
30684 +#define au_fset_lock(flags, name) \
30685 +       do { (flags) |= AuLock_##name; } while (0)
30686 +#define au_fclr_lock(flags, name) \
30687 +       do { (flags) &= ~AuLock_##name; } while (0)
30688 +
30689 +/* ---------------------------------------------------------------------- */
30690 +
30691 +/* super.c */
30692 +extern struct file_system_type aufs_fs_type;
30693 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino);
30694 +typedef unsigned long long (*au_arraycb_t)(struct super_block *sb, void *array,
30695 +                                          unsigned long long max, void *arg);
30696 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb,
30697 +                    struct super_block *sb, void *arg);
30698 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max);
30699 +void au_iarray_free(struct inode **a, unsigned long long max);
30700 +
30701 +/* sbinfo.c */
30702 +void au_si_free(struct kobject *kobj);
30703 +int au_si_alloc(struct super_block *sb);
30704 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr, int may_shrink);
30705 +
30706 +unsigned int au_sigen_inc(struct super_block *sb);
30707 +aufs_bindex_t au_new_br_id(struct super_block *sb);
30708 +
30709 +int si_read_lock(struct super_block *sb, int flags);
30710 +int si_write_lock(struct super_block *sb, int flags);
30711 +int aufs_read_lock(struct dentry *dentry, int flags);
30712 +void aufs_read_unlock(struct dentry *dentry, int flags);
30713 +void aufs_write_lock(struct dentry *dentry);
30714 +void aufs_write_unlock(struct dentry *dentry);
30715 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags);
30716 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2);
30717 +
30718 +/* wbr_policy.c */
30719 +extern struct au_wbr_copyup_operations au_wbr_copyup_ops[];
30720 +extern struct au_wbr_create_operations au_wbr_create_ops[];
30721 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst);
30722 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex);
30723 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t btop);
30724 +
30725 +/* mvdown.c */
30726 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *arg);
30727 +
30728 +#ifdef CONFIG_AUFS_FHSM
30729 +/* fhsm.c */
30730 +
30731 +static inline pid_t au_fhsm_pid(struct au_fhsm *fhsm)
30732 +{
30733 +       pid_t pid;
30734 +
30735 +       spin_lock(&fhsm->fhsm_spin);
30736 +       pid = fhsm->fhsm_pid;
30737 +       spin_unlock(&fhsm->fhsm_spin);
30738 +
30739 +       return pid;
30740 +}
30741 +
30742 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force);
30743 +void au_fhsm_wrote_all(struct super_block *sb, int force);
30744 +int au_fhsm_fd(struct super_block *sb, int oflags);
30745 +int au_fhsm_br_alloc(struct au_branch *br);
30746 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex);
30747 +void au_fhsm_fin(struct super_block *sb);
30748 +void au_fhsm_init(struct au_sbinfo *sbinfo);
30749 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec);
30750 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo);
30751 +#else
30752 +AuStubVoid(au_fhsm_wrote, struct super_block *sb, aufs_bindex_t bindex,
30753 +          int force)
30754 +AuStubVoid(au_fhsm_wrote_all, struct super_block *sb, int force)
30755 +AuStub(int, au_fhsm_fd, return -EOPNOTSUPP, struct super_block *sb, int oflags)
30756 +AuStub(pid_t, au_fhsm_pid, return 0, struct au_fhsm *fhsm)
30757 +AuStubInt0(au_fhsm_br_alloc, struct au_branch *br)
30758 +AuStubVoid(au_fhsm_set_bottom, struct super_block *sb, aufs_bindex_t bindex)
30759 +AuStubVoid(au_fhsm_fin, struct super_block *sb)
30760 +AuStubVoid(au_fhsm_init, struct au_sbinfo *sbinfo)
30761 +AuStubVoid(au_fhsm_set, struct au_sbinfo *sbinfo, unsigned int sec)
30762 +AuStubVoid(au_fhsm_show, struct seq_file *seq, struct au_sbinfo *sbinfo)
30763 +#endif
30764 +
30765 +/* ---------------------------------------------------------------------- */
30766 +
30767 +static inline struct au_sbinfo *au_sbi(struct super_block *sb)
30768 +{
30769 +       return sb->s_fs_info;
30770 +}
30771 +
30772 +/* ---------------------------------------------------------------------- */
30773 +
30774 +#ifdef CONFIG_AUFS_EXPORT
30775 +int au_test_nfsd(void);
30776 +void au_export_init(struct super_block *sb);
30777 +void au_xigen_inc(struct inode *inode);
30778 +int au_xigen_new(struct inode *inode);
30779 +int au_xigen_set(struct super_block *sb, struct path *path);
30780 +void au_xigen_clr(struct super_block *sb);
30781 +
30782 +static inline int au_busy_or_stale(void)
30783 +{
30784 +       if (!au_test_nfsd())
30785 +               return -EBUSY;
30786 +       return -ESTALE;
30787 +}
30788 +#else
30789 +AuStubInt0(au_test_nfsd, void)
30790 +AuStubVoid(au_export_init, struct super_block *sb)
30791 +AuStubVoid(au_xigen_inc, struct inode *inode)
30792 +AuStubInt0(au_xigen_new, struct inode *inode)
30793 +AuStubInt0(au_xigen_set, struct super_block *sb, struct path *path)
30794 +AuStubVoid(au_xigen_clr, struct super_block *sb)
30795 +AuStub(int, au_busy_or_stale, return -EBUSY, void)
30796 +#endif /* CONFIG_AUFS_EXPORT */
30797 +
30798 +/* ---------------------------------------------------------------------- */
30799 +
30800 +#ifdef CONFIG_AUFS_SBILIST
30801 +/* module.c */
30802 +extern struct hlist_bl_head au_sbilist;
30803 +
30804 +static inline void au_sbilist_init(void)
30805 +{
30806 +       INIT_HLIST_BL_HEAD(&au_sbilist);
30807 +}
30808 +
30809 +static inline void au_sbilist_add(struct super_block *sb)
30810 +{
30811 +       au_hbl_add(&au_sbi(sb)->si_list, &au_sbilist);
30812 +}
30813 +
30814 +static inline void au_sbilist_del(struct super_block *sb)
30815 +{
30816 +       au_hbl_del(&au_sbi(sb)->si_list, &au_sbilist);
30817 +}
30818 +
30819 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
30820 +static inline void au_sbilist_lock(void)
30821 +{
30822 +       hlist_bl_lock(&au_sbilist);
30823 +}
30824 +
30825 +static inline void au_sbilist_unlock(void)
30826 +{
30827 +       hlist_bl_unlock(&au_sbilist);
30828 +}
30829 +#define AuGFP_SBILIST  GFP_ATOMIC
30830 +#else
30831 +AuStubVoid(au_sbilist_lock, void)
30832 +AuStubVoid(au_sbilist_unlock, void)
30833 +#define AuGFP_SBILIST  GFP_NOFS
30834 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
30835 +#else
30836 +AuStubVoid(au_sbilist_init, void)
30837 +AuStubVoid(au_sbilist_add, struct super_block *sb)
30838 +AuStubVoid(au_sbilist_del, struct super_block *sb)
30839 +AuStubVoid(au_sbilist_lock, void)
30840 +AuStubVoid(au_sbilist_unlock, void)
30841 +#define AuGFP_SBILIST  GFP_NOFS
30842 +#endif
30843 +
30844 +/* ---------------------------------------------------------------------- */
30845 +
30846 +static inline void dbgaufs_si_null(struct au_sbinfo *sbinfo)
30847 +{
30848 +       /*
30849 +        * This function is a dynamic '__init' function actually,
30850 +        * so the tiny check for si_rwsem is unnecessary.
30851 +        */
30852 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
30853 +#ifdef CONFIG_DEBUG_FS
30854 +       sbinfo->si_dbgaufs = NULL;
30855 +       sbinfo->si_dbgaufs_plink = NULL;
30856 +       sbinfo->si_dbgaufs_xib = NULL;
30857 +#ifdef CONFIG_AUFS_EXPORT
30858 +       sbinfo->si_dbgaufs_xigen = NULL;
30859 +#endif
30860 +#endif
30861 +}
30862 +
30863 +/* ---------------------------------------------------------------------- */
30864 +
30865 +/* current->atomic_flags */
30866 +/* this value should never corrupt the ones defined in linux/sched.h */
30867 +#define PFA_AUFS       0x10
30868 +
30869 +TASK_PFA_TEST(AUFS, test_aufs) /* task_test_aufs */
30870 +TASK_PFA_SET(AUFS, aufs)       /* task_set_aufs */
30871 +TASK_PFA_CLEAR(AUFS, aufs)     /* task_clear_aufs */
30872 +
30873 +static inline int si_pid_test(struct super_block *sb)
30874 +{
30875 +       return !!task_test_aufs(current);
30876 +}
30877 +
30878 +static inline void si_pid_clr(struct super_block *sb)
30879 +{
30880 +       AuDebugOn(!task_test_aufs(current));
30881 +       task_clear_aufs(current);
30882 +}
30883 +
30884 +static inline void si_pid_set(struct super_block *sb)
30885 +{
30886 +       AuDebugOn(task_test_aufs(current));
30887 +       task_set_aufs(current);
30888 +}
30889 +
30890 +/* ---------------------------------------------------------------------- */
30891 +
30892 +/* lock superblock. mainly for entry point functions */
30893 +#define __si_read_lock(sb)     au_rw_read_lock(&au_sbi(sb)->si_rwsem)
30894 +#define __si_write_lock(sb)    au_rw_write_lock(&au_sbi(sb)->si_rwsem)
30895 +#define __si_read_trylock(sb)  au_rw_read_trylock(&au_sbi(sb)->si_rwsem)
30896 +#define __si_write_trylock(sb) au_rw_write_trylock(&au_sbi(sb)->si_rwsem)
30897 +/*
30898 +#define __si_read_trylock_nested(sb) \
30899 +       au_rw_read_trylock_nested(&au_sbi(sb)->si_rwsem)
30900 +#define __si_write_trylock_nested(sb) \
30901 +       au_rw_write_trylock_nested(&au_sbi(sb)->si_rwsem)
30902 +*/
30903 +
30904 +#define __si_read_unlock(sb)   au_rw_read_unlock(&au_sbi(sb)->si_rwsem)
30905 +#define __si_write_unlock(sb)  au_rw_write_unlock(&au_sbi(sb)->si_rwsem)
30906 +#define __si_downgrade_lock(sb)        au_rw_dgrade_lock(&au_sbi(sb)->si_rwsem)
30907 +
30908 +#define SiMustNoWaiters(sb)    AuRwMustNoWaiters(&au_sbi(sb)->si_rwsem)
30909 +#define SiMustAnyLock(sb)      AuRwMustAnyLock(&au_sbi(sb)->si_rwsem)
30910 +#define SiMustWriteLock(sb)    AuRwMustWriteLock(&au_sbi(sb)->si_rwsem)
30911 +
30912 +static inline void si_noflush_read_lock(struct super_block *sb)
30913 +{
30914 +       __si_read_lock(sb);
30915 +       si_pid_set(sb);
30916 +}
30917 +
30918 +static inline int si_noflush_read_trylock(struct super_block *sb)
30919 +{
30920 +       int locked;
30921 +
30922 +       locked = __si_read_trylock(sb);
30923 +       if (locked)
30924 +               si_pid_set(sb);
30925 +       return locked;
30926 +}
30927 +
30928 +static inline void si_noflush_write_lock(struct super_block *sb)
30929 +{
30930 +       __si_write_lock(sb);
30931 +       si_pid_set(sb);
30932 +}
30933 +
30934 +static inline int si_noflush_write_trylock(struct super_block *sb)
30935 +{
30936 +       int locked;
30937 +
30938 +       locked = __si_write_trylock(sb);
30939 +       if (locked)
30940 +               si_pid_set(sb);
30941 +       return locked;
30942 +}
30943 +
30944 +#if 0 /* reserved */
30945 +static inline int si_read_trylock(struct super_block *sb, int flags)
30946 +{
30947 +       if (au_ftest_lock(flags, FLUSH))
30948 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
30949 +       return si_noflush_read_trylock(sb);
30950 +}
30951 +#endif
30952 +
30953 +static inline void si_read_unlock(struct super_block *sb)
30954 +{
30955 +       si_pid_clr(sb);
30956 +       __si_read_unlock(sb);
30957 +}
30958 +
30959 +#if 0 /* reserved */
30960 +static inline int si_write_trylock(struct super_block *sb, int flags)
30961 +{
30962 +       if (au_ftest_lock(flags, FLUSH))
30963 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
30964 +       return si_noflush_write_trylock(sb);
30965 +}
30966 +#endif
30967 +
30968 +static inline void si_write_unlock(struct super_block *sb)
30969 +{
30970 +       si_pid_clr(sb);
30971 +       __si_write_unlock(sb);
30972 +}
30973 +
30974 +#if 0 /* reserved */
30975 +static inline void si_downgrade_lock(struct super_block *sb)
30976 +{
30977 +       __si_downgrade_lock(sb);
30978 +}
30979 +#endif
30980 +
30981 +/* ---------------------------------------------------------------------- */
30982 +
30983 +static inline aufs_bindex_t au_sbbot(struct super_block *sb)
30984 +{
30985 +       SiMustAnyLock(sb);
30986 +       return au_sbi(sb)->si_bbot;
30987 +}
30988 +
30989 +static inline unsigned int au_mntflags(struct super_block *sb)
30990 +{
30991 +       SiMustAnyLock(sb);
30992 +       return au_sbi(sb)->si_mntflags;
30993 +}
30994 +
30995 +static inline unsigned int au_sigen(struct super_block *sb)
30996 +{
30997 +       SiMustAnyLock(sb);
30998 +       return au_sbi(sb)->si_generation;
30999 +}
31000 +
31001 +static inline struct au_branch *au_sbr(struct super_block *sb,
31002 +                                      aufs_bindex_t bindex)
31003 +{
31004 +       SiMustAnyLock(sb);
31005 +       return au_sbi(sb)->si_branch[0 + bindex];
31006 +}
31007 +
31008 +static inline loff_t au_xi_maxent(struct super_block *sb)
31009 +{
31010 +       SiMustAnyLock(sb);
31011 +       return au_sbi(sb)->si_ximaxent;
31012 +}
31013 +
31014 +#endif /* __KERNEL__ */
31015 +#endif /* __AUFS_SUPER_H__ */
31016 diff -urN /usr/share/empty/fs/aufs/sysaufs.c linux/fs/aufs/sysaufs.c
31017 --- /usr/share/empty/fs/aufs/sysaufs.c  1970-01-01 01:00:00.000000000 +0100
31018 +++ linux/fs/aufs/sysaufs.c     2021-12-03 15:38:59.939980643 +0100
31019 @@ -0,0 +1,93 @@
31020 +// SPDX-License-Identifier: GPL-2.0
31021 +/*
31022 + * Copyright (C) 2005-2021 Junjiro R. Okajima
31023 + *
31024 + * This program, aufs is free software; you can redistribute it and/or modify
31025 + * it under the terms of the GNU General Public License as published by
31026 + * the Free Software Foundation; either version 2 of the License, or
31027 + * (at your option) any later version.
31028 + *
31029 + * This program is distributed in the hope that it will be useful,
31030 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31031 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31032 + * GNU General Public License for more details.
31033 + *
31034 + * You should have received a copy of the GNU General Public License
31035 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31036 + */
31037 +
31038 +/*
31039 + * sysfs interface and lifetime management
31040 + * they are necessary regardless sysfs is disabled.
31041 + */
31042 +
31043 +#include <linux/random.h>
31044 +#include "aufs.h"
31045 +
31046 +unsigned long sysaufs_si_mask;
31047 +struct kset *sysaufs_kset;
31048 +
31049 +#define AuSiAttr(_name) { \
31050 +       .attr   = { .name = __stringify(_name), .mode = 0444 }, \
31051 +       .show   = sysaufs_si_##_name,                           \
31052 +}
31053 +
31054 +static struct sysaufs_si_attr sysaufs_si_attr_xi_path = AuSiAttr(xi_path);
31055 +struct attribute *sysaufs_si_attrs[] = {
31056 +       &sysaufs_si_attr_xi_path.attr,
31057 +       NULL,
31058 +};
31059 +
31060 +static const struct sysfs_ops au_sbi_ops = {
31061 +       .show   = sysaufs_si_show
31062 +};
31063 +
31064 +static struct kobj_type au_sbi_ktype = {
31065 +       .release        = au_si_free,
31066 +       .sysfs_ops      = &au_sbi_ops,
31067 +       .default_attrs  = sysaufs_si_attrs
31068 +};
31069 +
31070 +/* ---------------------------------------------------------------------- */
31071 +
31072 +int sysaufs_si_init(struct au_sbinfo *sbinfo)
31073 +{
31074 +       int err;
31075 +
31076 +       sbinfo->si_kobj.kset = sysaufs_kset;
31077 +       /* cf. sysaufs_name() */
31078 +       err = kobject_init_and_add
31079 +               (&sbinfo->si_kobj, &au_sbi_ktype, /*&sysaufs_kset->kobj*/NULL,
31080 +                SysaufsSiNamePrefix "%lx", sysaufs_si_id(sbinfo));
31081 +
31082 +       return err;
31083 +}
31084 +
31085 +void sysaufs_fin(void)
31086 +{
31087 +       sysfs_remove_group(&sysaufs_kset->kobj, sysaufs_attr_group);
31088 +       kset_unregister(sysaufs_kset);
31089 +}
31090 +
31091 +int __init sysaufs_init(void)
31092 +{
31093 +       int err;
31094 +
31095 +       do {
31096 +               get_random_bytes(&sysaufs_si_mask, sizeof(sysaufs_si_mask));
31097 +       } while (!sysaufs_si_mask);
31098 +
31099 +       err = -EINVAL;
31100 +       sysaufs_kset = kset_create_and_add(AUFS_NAME, NULL, fs_kobj);
31101 +       if (unlikely(!sysaufs_kset))
31102 +               goto out;
31103 +       err = PTR_ERR(sysaufs_kset);
31104 +       if (IS_ERR(sysaufs_kset))
31105 +               goto out;
31106 +       err = sysfs_create_group(&sysaufs_kset->kobj, sysaufs_attr_group);
31107 +       if (unlikely(err))
31108 +               kset_unregister(sysaufs_kset);
31109 +
31110 +out:
31111 +       return err;
31112 +}
31113 diff -urN /usr/share/empty/fs/aufs/sysaufs.h linux/fs/aufs/sysaufs.h
31114 --- /usr/share/empty/fs/aufs/sysaufs.h  1970-01-01 01:00:00.000000000 +0100
31115 +++ linux/fs/aufs/sysaufs.h     2021-12-03 15:38:59.939980643 +0100
31116 @@ -0,0 +1,102 @@
31117 +/* SPDX-License-Identifier: GPL-2.0 */
31118 +/*
31119 + * Copyright (C) 2005-2021 Junjiro R. Okajima
31120 + *
31121 + * This program, aufs is free software; you can redistribute it and/or modify
31122 + * it under the terms of the GNU General Public License as published by
31123 + * the Free Software Foundation; either version 2 of the License, or
31124 + * (at your option) any later version.
31125 + *
31126 + * This program is distributed in the hope that it will be useful,
31127 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31128 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31129 + * GNU General Public License for more details.
31130 + *
31131 + * You should have received a copy of the GNU General Public License
31132 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31133 + */
31134 +
31135 +/*
31136 + * sysfs interface and mount lifetime management
31137 + */
31138 +
31139 +#ifndef __SYSAUFS_H__
31140 +#define __SYSAUFS_H__
31141 +
31142 +#ifdef __KERNEL__
31143 +
31144 +#include <linux/sysfs.h>
31145 +#include "module.h"
31146 +
31147 +struct super_block;
31148 +struct au_sbinfo;
31149 +
31150 +struct sysaufs_si_attr {
31151 +       struct attribute attr;
31152 +       int (*show)(struct seq_file *seq, struct super_block *sb);
31153 +};
31154 +
31155 +/* ---------------------------------------------------------------------- */
31156 +
31157 +/* sysaufs.c */
31158 +extern unsigned long sysaufs_si_mask;
31159 +extern struct kset *sysaufs_kset;
31160 +extern struct attribute *sysaufs_si_attrs[];
31161 +int sysaufs_si_init(struct au_sbinfo *sbinfo);
31162 +int __init sysaufs_init(void);
31163 +void sysaufs_fin(void);
31164 +
31165 +/* ---------------------------------------------------------------------- */
31166 +
31167 +/* some people doesn't like to show a pointer in kernel */
31168 +static inline unsigned long sysaufs_si_id(struct au_sbinfo *sbinfo)
31169 +{
31170 +       return sysaufs_si_mask ^ (unsigned long)sbinfo;
31171 +}
31172 +
31173 +#define SysaufsSiNamePrefix    "si_"
31174 +#define SysaufsSiNameLen       (sizeof(SysaufsSiNamePrefix) + 16)
31175 +static inline void sysaufs_name(struct au_sbinfo *sbinfo, char *name)
31176 +{
31177 +       snprintf(name, SysaufsSiNameLen, SysaufsSiNamePrefix "%lx",
31178 +                sysaufs_si_id(sbinfo));
31179 +}
31180 +
31181 +struct au_branch;
31182 +#ifdef CONFIG_SYSFS
31183 +/* sysfs.c */
31184 +extern struct attribute_group *sysaufs_attr_group;
31185 +
31186 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb);
31187 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
31188 +                        char *buf);
31189 +long au_brinfo_ioctl(struct file *file, unsigned long arg);
31190 +#ifdef CONFIG_COMPAT
31191 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg);
31192 +#endif
31193 +
31194 +void sysaufs_br_init(struct au_branch *br);
31195 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
31196 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
31197 +
31198 +#define sysaufs_brs_init()     do {} while (0)
31199 +
31200 +#else
31201 +#define sysaufs_attr_group     NULL
31202 +
31203 +AuStubInt0(sysaufs_si_xi_path, struct seq_file *seq, struct super_block *sb)
31204 +AuStub(ssize_t, sysaufs_si_show, return 0, struct kobject *kobj,
31205 +       struct attribute *attr, char *buf)
31206 +AuStubVoid(sysaufs_br_init, struct au_branch *br)
31207 +AuStubVoid(sysaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
31208 +AuStubVoid(sysaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
31209 +
31210 +static inline void sysaufs_brs_init(void)
31211 +{
31212 +       sysaufs_brs = 0;
31213 +}
31214 +
31215 +#endif /* CONFIG_SYSFS */
31216 +
31217 +#endif /* __KERNEL__ */
31218 +#endif /* __SYSAUFS_H__ */
31219 diff -urN /usr/share/empty/fs/aufs/sysfs.c linux/fs/aufs/sysfs.c
31220 --- /usr/share/empty/fs/aufs/sysfs.c    1970-01-01 01:00:00.000000000 +0100
31221 +++ linux/fs/aufs/sysfs.c       2021-12-03 15:38:59.939980643 +0100
31222 @@ -0,0 +1,374 @@
31223 +// SPDX-License-Identifier: GPL-2.0
31224 +/*
31225 + * Copyright (C) 2005-2021 Junjiro R. Okajima
31226 + *
31227 + * This program, aufs is free software; you can redistribute it and/or modify
31228 + * it under the terms of the GNU General Public License as published by
31229 + * the Free Software Foundation; either version 2 of the License, or
31230 + * (at your option) any later version.
31231 + *
31232 + * This program is distributed in the hope that it will be useful,
31233 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31234 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31235 + * GNU General Public License for more details.
31236 + *
31237 + * You should have received a copy of the GNU General Public License
31238 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31239 + */
31240 +
31241 +/*
31242 + * sysfs interface
31243 + */
31244 +
31245 +#include <linux/compat.h>
31246 +#include <linux/seq_file.h>
31247 +#include "aufs.h"
31248 +
31249 +#ifdef CONFIG_AUFS_FS_MODULE
31250 +/* this entry violates the "one line per file" policy of sysfs */
31251 +static ssize_t config_show(struct kobject *kobj, struct kobj_attribute *attr,
31252 +                          char *buf)
31253 +{
31254 +       ssize_t err;
31255 +       static char *conf =
31256 +/* this file is generated at compiling */
31257 +#include "conf.str"
31258 +               ;
31259 +
31260 +       err = snprintf(buf, PAGE_SIZE, conf);
31261 +       if (unlikely(err >= PAGE_SIZE))
31262 +               err = -EFBIG;
31263 +       return err;
31264 +}
31265 +
31266 +static struct kobj_attribute au_config_attr = __ATTR_RO(config);
31267 +#endif
31268 +
31269 +static struct attribute *au_attr[] = {
31270 +#ifdef CONFIG_AUFS_FS_MODULE
31271 +       &au_config_attr.attr,
31272 +#endif
31273 +       NULL,   /* need to NULL terminate the list of attributes */
31274 +};
31275 +
31276 +static struct attribute_group sysaufs_attr_group_body = {
31277 +       .attrs = au_attr
31278 +};
31279 +
31280 +struct attribute_group *sysaufs_attr_group = &sysaufs_attr_group_body;
31281 +
31282 +/* ---------------------------------------------------------------------- */
31283 +
31284 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb)
31285 +{
31286 +       int err;
31287 +
31288 +       SiMustAnyLock(sb);
31289 +
31290 +       err = 0;
31291 +       if (au_opt_test(au_mntflags(sb), XINO)) {
31292 +               err = au_xino_path(seq, au_sbi(sb)->si_xib);
31293 +               seq_putc(seq, '\n');
31294 +       }
31295 +       return err;
31296 +}
31297 +
31298 +/*
31299 + * the lifetime of branch is independent from the entry under sysfs.
31300 + * sysfs handles the lifetime of the entry, and never call ->show() after it is
31301 + * unlinked.
31302 + */
31303 +static int sysaufs_si_br(struct seq_file *seq, struct super_block *sb,
31304 +                        aufs_bindex_t bindex, int idx)
31305 +{
31306 +       int err;
31307 +       struct path path;
31308 +       struct dentry *root;
31309 +       struct au_branch *br;
31310 +       au_br_perm_str_t perm;
31311 +
31312 +       AuDbg("b%d\n", bindex);
31313 +
31314 +       err = 0;
31315 +       root = sb->s_root;
31316 +       di_read_lock_parent(root, !AuLock_IR);
31317 +       br = au_sbr(sb, bindex);
31318 +
31319 +       switch (idx) {
31320 +       case AuBrSysfs_BR:
31321 +               path.mnt = au_br_mnt(br);
31322 +               path.dentry = au_h_dptr(root, bindex);
31323 +               err = au_seq_path(seq, &path);
31324 +               if (!err) {
31325 +                       au_optstr_br_perm(&perm, br->br_perm);
31326 +                       seq_printf(seq, "=%s\n", perm.a);
31327 +               }
31328 +               break;
31329 +       case AuBrSysfs_BRID:
31330 +               seq_printf(seq, "%d\n", br->br_id);
31331 +               break;
31332 +       }
31333 +       di_read_unlock(root, !AuLock_IR);
31334 +       if (unlikely(err || seq_has_overflowed(seq)))
31335 +               err = -E2BIG;
31336 +
31337 +       return err;
31338 +}
31339 +
31340 +/* ---------------------------------------------------------------------- */
31341 +
31342 +static struct seq_file *au_seq(char *p, ssize_t len)
31343 +{
31344 +       struct seq_file *seq;
31345 +
31346 +       seq = kzalloc(sizeof(*seq), GFP_NOFS);
31347 +       if (seq) {
31348 +               /* mutex_init(&seq.lock); */
31349 +               seq->buf = p;
31350 +               seq->size = len;
31351 +               return seq; /* success */
31352 +       }
31353 +
31354 +       seq = ERR_PTR(-ENOMEM);
31355 +       return seq;
31356 +}
31357 +
31358 +#define SysaufsBr_PREFIX       "br"
31359 +#define SysaufsBrid_PREFIX     "brid"
31360 +
31361 +/* todo: file size may exceed PAGE_SIZE */
31362 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
31363 +                       char *buf)
31364 +{
31365 +       ssize_t err;
31366 +       int idx;
31367 +       long l;
31368 +       aufs_bindex_t bbot;
31369 +       struct au_sbinfo *sbinfo;
31370 +       struct super_block *sb;
31371 +       struct seq_file *seq;
31372 +       char *name;
31373 +       struct attribute **cattr;
31374 +
31375 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
31376 +       sb = sbinfo->si_sb;
31377 +
31378 +       /*
31379 +        * prevent a race condition between sysfs and aufs.
31380 +        * for instance, sysfs_file_read() calls sysfs_get_active_two() which
31381 +        * prohibits maintaining the sysfs entries.
31382 +        * hew we acquire read lock after sysfs_get_active_two().
31383 +        * on the other hand, the remount process may maintain the sysfs/aufs
31384 +        * entries after acquiring write lock.
31385 +        * it can cause a deadlock.
31386 +        * simply we gave up processing read here.
31387 +        */
31388 +       err = -EBUSY;
31389 +       if (unlikely(!si_noflush_read_trylock(sb)))
31390 +               goto out;
31391 +
31392 +       seq = au_seq(buf, PAGE_SIZE);
31393 +       err = PTR_ERR(seq);
31394 +       if (IS_ERR(seq))
31395 +               goto out_unlock;
31396 +
31397 +       name = (void *)attr->name;
31398 +       cattr = sysaufs_si_attrs;
31399 +       while (*cattr) {
31400 +               if (!strcmp(name, (*cattr)->name)) {
31401 +                       err = container_of(*cattr, struct sysaufs_si_attr, attr)
31402 +                               ->show(seq, sb);
31403 +                       goto out_seq;
31404 +               }
31405 +               cattr++;
31406 +       }
31407 +
31408 +       if (!strncmp(name, SysaufsBrid_PREFIX,
31409 +                    sizeof(SysaufsBrid_PREFIX) - 1)) {
31410 +               idx = AuBrSysfs_BRID;
31411 +               name += sizeof(SysaufsBrid_PREFIX) - 1;
31412 +       } else if (!strncmp(name, SysaufsBr_PREFIX,
31413 +                           sizeof(SysaufsBr_PREFIX) - 1)) {
31414 +               idx = AuBrSysfs_BR;
31415 +               name += sizeof(SysaufsBr_PREFIX) - 1;
31416 +       } else
31417 +                 BUG();
31418 +
31419 +       err = kstrtol(name, 10, &l);
31420 +       if (!err) {
31421 +               bbot = au_sbbot(sb);
31422 +               if (l <= bbot)
31423 +                       err = sysaufs_si_br(seq, sb, (aufs_bindex_t)l, idx);
31424 +               else
31425 +                       err = -ENOENT;
31426 +       }
31427 +
31428 +out_seq:
31429 +       if (!err) {
31430 +               err = seq->count;
31431 +               /* sysfs limit */
31432 +               if (unlikely(err == PAGE_SIZE))
31433 +                       err = -EFBIG;
31434 +       }
31435 +       au_kfree_rcu(seq);
31436 +out_unlock:
31437 +       si_read_unlock(sb);
31438 +out:
31439 +       return err;
31440 +}
31441 +
31442 +/* ---------------------------------------------------------------------- */
31443 +
31444 +static int au_brinfo(struct super_block *sb, union aufs_brinfo __user *arg)
31445 +{
31446 +       int err;
31447 +       int16_t brid;
31448 +       aufs_bindex_t bindex, bbot;
31449 +       size_t sz;
31450 +       char *buf;
31451 +       struct seq_file *seq;
31452 +       struct au_branch *br;
31453 +
31454 +       si_read_lock(sb, AuLock_FLUSH);
31455 +       bbot = au_sbbot(sb);
31456 +       err = bbot + 1;
31457 +       if (!arg)
31458 +               goto out;
31459 +
31460 +       err = -ENOMEM;
31461 +       buf = (void *)__get_free_page(GFP_NOFS);
31462 +       if (unlikely(!buf))
31463 +               goto out;
31464 +
31465 +       seq = au_seq(buf, PAGE_SIZE);
31466 +       err = PTR_ERR(seq);
31467 +       if (IS_ERR(seq))
31468 +               goto out_buf;
31469 +
31470 +       sz = sizeof(*arg) - offsetof(union aufs_brinfo, path);
31471 +       for (bindex = 0; bindex <= bbot; bindex++, arg++) {
31472 +               /* VERIFY_WRITE */
31473 +               err = !access_ok(arg, sizeof(*arg));
31474 +               if (unlikely(err))
31475 +                       break;
31476 +
31477 +               br = au_sbr(sb, bindex);
31478 +               brid = br->br_id;
31479 +               BUILD_BUG_ON(sizeof(brid) != sizeof(arg->id));
31480 +               err = __put_user(brid, &arg->id);
31481 +               if (unlikely(err))
31482 +                       break;
31483 +
31484 +               BUILD_BUG_ON(sizeof(br->br_perm) != sizeof(arg->perm));
31485 +               err = __put_user(br->br_perm, &arg->perm);
31486 +               if (unlikely(err))
31487 +                       break;
31488 +
31489 +               err = au_seq_path(seq, &br->br_path);
31490 +               if (unlikely(err))
31491 +                       break;
31492 +               seq_putc(seq, '\0');
31493 +               if (!seq_has_overflowed(seq)) {
31494 +                       err = copy_to_user(arg->path, seq->buf, seq->count);
31495 +                       seq->count = 0;
31496 +                       if (unlikely(err))
31497 +                               break;
31498 +               } else {
31499 +                       err = -E2BIG;
31500 +                       goto out_seq;
31501 +               }
31502 +       }
31503 +       if (unlikely(err))
31504 +               err = -EFAULT;
31505 +
31506 +out_seq:
31507 +       au_kfree_rcu(seq);
31508 +out_buf:
31509 +       free_page((unsigned long)buf);
31510 +out:
31511 +       si_read_unlock(sb);
31512 +       return err;
31513 +}
31514 +
31515 +long au_brinfo_ioctl(struct file *file, unsigned long arg)
31516 +{
31517 +       return au_brinfo(file->f_path.dentry->d_sb, (void __user *)arg);
31518 +}
31519 +
31520 +#ifdef CONFIG_COMPAT
31521 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg)
31522 +{
31523 +       return au_brinfo(file->f_path.dentry->d_sb, compat_ptr(arg));
31524 +}
31525 +#endif
31526 +
31527 +/* ---------------------------------------------------------------------- */
31528 +
31529 +void sysaufs_br_init(struct au_branch *br)
31530 +{
31531 +       int i;
31532 +       struct au_brsysfs *br_sysfs;
31533 +       struct attribute *attr;
31534 +
31535 +       br_sysfs = br->br_sysfs;
31536 +       for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
31537 +               attr = &br_sysfs->attr;
31538 +               sysfs_attr_init(attr);
31539 +               attr->name = br_sysfs->name;
31540 +               attr->mode = 0444;
31541 +               br_sysfs++;
31542 +       }
31543 +}
31544 +
31545 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
31546 +{
31547 +       struct au_branch *br;
31548 +       struct kobject *kobj;
31549 +       struct au_brsysfs *br_sysfs;
31550 +       int i;
31551 +       aufs_bindex_t bbot;
31552 +
31553 +       if (!sysaufs_brs)
31554 +               return;
31555 +
31556 +       kobj = &au_sbi(sb)->si_kobj;
31557 +       bbot = au_sbbot(sb);
31558 +       for (; bindex <= bbot; bindex++) {
31559 +               br = au_sbr(sb, bindex);
31560 +               br_sysfs = br->br_sysfs;
31561 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
31562 +                       sysfs_remove_file(kobj, &br_sysfs->attr);
31563 +                       br_sysfs++;
31564 +               }
31565 +       }
31566 +}
31567 +
31568 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
31569 +{
31570 +       int err, i;
31571 +       aufs_bindex_t bbot;
31572 +       struct kobject *kobj;
31573 +       struct au_branch *br;
31574 +       struct au_brsysfs *br_sysfs;
31575 +
31576 +       if (!sysaufs_brs)
31577 +               return;
31578 +
31579 +       kobj = &au_sbi(sb)->si_kobj;
31580 +       bbot = au_sbbot(sb);
31581 +       for (; bindex <= bbot; bindex++) {
31582 +               br = au_sbr(sb, bindex);
31583 +               br_sysfs = br->br_sysfs;
31584 +               snprintf(br_sysfs[AuBrSysfs_BR].name, sizeof(br_sysfs->name),
31585 +                        SysaufsBr_PREFIX "%d", bindex);
31586 +               snprintf(br_sysfs[AuBrSysfs_BRID].name, sizeof(br_sysfs->name),
31587 +                        SysaufsBrid_PREFIX "%d", bindex);
31588 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
31589 +                       err = sysfs_create_file(kobj, &br_sysfs->attr);
31590 +                       if (unlikely(err))
31591 +                               pr_warn("failed %s under sysfs(%d)\n",
31592 +                                       br_sysfs->name, err);
31593 +                       br_sysfs++;
31594 +               }
31595 +       }
31596 +}
31597 diff -urN /usr/share/empty/fs/aufs/sysrq.c linux/fs/aufs/sysrq.c
31598 --- /usr/share/empty/fs/aufs/sysrq.c    1970-01-01 01:00:00.000000000 +0100
31599 +++ linux/fs/aufs/sysrq.c       2021-12-03 15:38:59.939980643 +0100
31600 @@ -0,0 +1,149 @@
31601 +// SPDX-License-Identifier: GPL-2.0
31602 +/*
31603 + * Copyright (C) 2005-2021 Junjiro R. Okajima
31604 + *
31605 + * This program, aufs is free software; you can redistribute it and/or modify
31606 + * it under the terms of the GNU General Public License as published by
31607 + * the Free Software Foundation; either version 2 of the License, or
31608 + * (at your option) any later version.
31609 + *
31610 + * This program is distributed in the hope that it will be useful,
31611 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31612 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31613 + * GNU General Public License for more details.
31614 + *
31615 + * You should have received a copy of the GNU General Public License
31616 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31617 + */
31618 +
31619 +/*
31620 + * magic sysrq handler
31621 + */
31622 +
31623 +/* #include <linux/sysrq.h> */
31624 +#include <linux/writeback.h>
31625 +#include "aufs.h"
31626 +
31627 +/* ---------------------------------------------------------------------- */
31628 +
31629 +static void sysrq_sb(struct super_block *sb)
31630 +{
31631 +       char *plevel;
31632 +       struct au_sbinfo *sbinfo;
31633 +       struct file *file;
31634 +       struct hlist_bl_head *files;
31635 +       struct hlist_bl_node *pos;
31636 +       struct au_finfo *finfo;
31637 +       struct inode *i;
31638 +
31639 +       plevel = au_plevel;
31640 +       au_plevel = KERN_WARNING;
31641 +
31642 +       /* since we define pr_fmt, call printk directly */
31643 +#define pr(str) printk(KERN_WARNING AUFS_NAME ": " str)
31644 +
31645 +       sbinfo = au_sbi(sb);
31646 +       printk(KERN_WARNING "si=%lx\n", sysaufs_si_id(sbinfo));
31647 +       pr("superblock\n");
31648 +       au_dpri_sb(sb);
31649 +
31650 +#if 0 /* reserved */
31651 +       do {
31652 +               int err, i, j, ndentry;
31653 +               struct au_dcsub_pages dpages;
31654 +               struct au_dpage *dpage;
31655 +
31656 +               err = au_dpages_init(&dpages, GFP_ATOMIC);
31657 +               if (unlikely(err))
31658 +                       break;
31659 +               err = au_dcsub_pages(&dpages, sb->s_root, NULL, NULL);
31660 +               if (!err)
31661 +                       for (i = 0; i < dpages.ndpage; i++) {
31662 +                               dpage = dpages.dpages + i;
31663 +                               ndentry = dpage->ndentry;
31664 +                               for (j = 0; j < ndentry; j++)
31665 +                                       au_dpri_dentry(dpage->dentries[j]);
31666 +                       }
31667 +               au_dpages_free(&dpages);
31668 +       } while (0);
31669 +#endif
31670 +
31671 +       pr("isolated inode\n");
31672 +       spin_lock(&sb->s_inode_list_lock);
31673 +       list_for_each_entry(i, &sb->s_inodes, i_sb_list) {
31674 +               spin_lock(&i->i_lock);
31675 +               if (hlist_empty(&i->i_dentry))
31676 +                       au_dpri_inode(i);
31677 +               spin_unlock(&i->i_lock);
31678 +       }
31679 +       spin_unlock(&sb->s_inode_list_lock);
31680 +
31681 +       pr("files\n");
31682 +       files = &au_sbi(sb)->si_files;
31683 +       hlist_bl_lock(files);
31684 +       hlist_bl_for_each_entry(finfo, pos, files, fi_hlist) {
31685 +               umode_t mode;
31686 +
31687 +               file = finfo->fi_file;
31688 +               mode = file_inode(file)->i_mode;
31689 +               if (!special_file(mode))
31690 +                       au_dpri_file(file);
31691 +       }
31692 +       hlist_bl_unlock(files);
31693 +       pr("done\n");
31694 +
31695 +#undef pr
31696 +       au_plevel = plevel;
31697 +}
31698 +
31699 +/* ---------------------------------------------------------------------- */
31700 +
31701 +/* module parameter */
31702 +static char *aufs_sysrq_key = "a";
31703 +module_param_named(sysrq, aufs_sysrq_key, charp, 0444);
31704 +MODULE_PARM_DESC(sysrq, "MagicSysRq key for " AUFS_NAME);
31705 +
31706 +static void au_sysrq(int key __maybe_unused)
31707 +{
31708 +       struct au_sbinfo *sbinfo;
31709 +       struct hlist_bl_node *pos;
31710 +
31711 +       lockdep_off();
31712 +       au_sbilist_lock();
31713 +       hlist_bl_for_each_entry(sbinfo, pos, &au_sbilist, si_list)
31714 +               sysrq_sb(sbinfo->si_sb);
31715 +       au_sbilist_unlock();
31716 +       lockdep_on();
31717 +}
31718 +
31719 +static struct sysrq_key_op au_sysrq_op = {
31720 +       .handler        = au_sysrq,
31721 +       .help_msg       = "Aufs",
31722 +       .action_msg     = "Aufs",
31723 +       .enable_mask    = SYSRQ_ENABLE_DUMP
31724 +};
31725 +
31726 +/* ---------------------------------------------------------------------- */
31727 +
31728 +int __init au_sysrq_init(void)
31729 +{
31730 +       int err;
31731 +       char key;
31732 +
31733 +       err = -1;
31734 +       key = *aufs_sysrq_key;
31735 +       if ('a' <= key && key <= 'z')
31736 +               err = register_sysrq_key(key, &au_sysrq_op);
31737 +       if (unlikely(err))
31738 +               pr_err("err %d, sysrq=%c\n", err, key);
31739 +       return err;
31740 +}
31741 +
31742 +void au_sysrq_fin(void)
31743 +{
31744 +       int err;
31745 +
31746 +       err = unregister_sysrq_key(*aufs_sysrq_key, &au_sysrq_op);
31747 +       if (unlikely(err))
31748 +               pr_err("err %d (ignored)\n", err);
31749 +}
31750 diff -urN /usr/share/empty/fs/aufs/vdir.c linux/fs/aufs/vdir.c
31751 --- /usr/share/empty/fs/aufs/vdir.c     1970-01-01 01:00:00.000000000 +0100
31752 +++ linux/fs/aufs/vdir.c        2021-12-03 15:38:59.939980643 +0100
31753 @@ -0,0 +1,896 @@
31754 +// SPDX-License-Identifier: GPL-2.0
31755 +/*
31756 + * Copyright (C) 2005-2021 Junjiro R. Okajima
31757 + *
31758 + * This program, aufs is free software; you can redistribute it and/or modify
31759 + * it under the terms of the GNU General Public License as published by
31760 + * the Free Software Foundation; either version 2 of the License, or
31761 + * (at your option) any later version.
31762 + *
31763 + * This program is distributed in the hope that it will be useful,
31764 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31765 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31766 + * GNU General Public License for more details.
31767 + *
31768 + * You should have received a copy of the GNU General Public License
31769 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31770 + */
31771 +
31772 +/*
31773 + * virtual or vertical directory
31774 + */
31775 +
31776 +#include <linux/iversion.h>
31777 +#include "aufs.h"
31778 +
31779 +static unsigned int calc_size(int nlen)
31780 +{
31781 +       return ALIGN(sizeof(struct au_vdir_de) + nlen, sizeof(ino_t));
31782 +}
31783 +
31784 +static int set_deblk_end(union au_vdir_deblk_p *p,
31785 +                        union au_vdir_deblk_p *deblk_end)
31786 +{
31787 +       if (calc_size(0) <= deblk_end->deblk - p->deblk) {
31788 +               p->de->de_str.len = 0;
31789 +               /* smp_mb(); */
31790 +               return 0;
31791 +       }
31792 +       return -1; /* error */
31793 +}
31794 +
31795 +/* returns true or false */
31796 +static int is_deblk_end(union au_vdir_deblk_p *p,
31797 +                       union au_vdir_deblk_p *deblk_end)
31798 +{
31799 +       if (calc_size(0) <= deblk_end->deblk - p->deblk)
31800 +               return !p->de->de_str.len;
31801 +       return 1;
31802 +}
31803 +
31804 +static unsigned char *last_deblk(struct au_vdir *vdir)
31805 +{
31806 +       return vdir->vd_deblk[vdir->vd_nblk - 1];
31807 +}
31808 +
31809 +/* ---------------------------------------------------------------------- */
31810 +
31811 +/* estimate the appropriate size for name hash table */
31812 +unsigned int au_rdhash_est(loff_t sz)
31813 +{
31814 +       unsigned int n;
31815 +
31816 +       n = UINT_MAX;
31817 +       sz >>= 10;
31818 +       if (sz < n)
31819 +               n = sz;
31820 +       if (sz < AUFS_RDHASH_DEF)
31821 +               n = AUFS_RDHASH_DEF;
31822 +       /* pr_info("n %u\n", n); */
31823 +       return n;
31824 +}
31825 +
31826 +/*
31827 + * the allocated memory has to be freed by
31828 + * au_nhash_wh_free() or au_nhash_de_free().
31829 + */
31830 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp)
31831 +{
31832 +       struct hlist_head *head;
31833 +       unsigned int u;
31834 +       size_t sz;
31835 +
31836 +       sz = sizeof(*nhash->nh_head) * num_hash;
31837 +       head = kmalloc(sz, gfp);
31838 +       if (head) {
31839 +               nhash->nh_num = num_hash;
31840 +               nhash->nh_head = head;
31841 +               for (u = 0; u < num_hash; u++)
31842 +                       INIT_HLIST_HEAD(head++);
31843 +               return 0; /* success */
31844 +       }
31845 +
31846 +       return -ENOMEM;
31847 +}
31848 +
31849 +static void nhash_count(struct hlist_head *head)
31850 +{
31851 +#if 0 /* debugging */
31852 +       unsigned long n;
31853 +       struct hlist_node *pos;
31854 +
31855 +       n = 0;
31856 +       hlist_for_each(pos, head)
31857 +               n++;
31858 +       pr_info("%lu\n", n);
31859 +#endif
31860 +}
31861 +
31862 +static void au_nhash_wh_do_free(struct hlist_head *head)
31863 +{
31864 +       struct au_vdir_wh *pos;
31865 +       struct hlist_node *node;
31866 +
31867 +       hlist_for_each_entry_safe(pos, node, head, wh_hash)
31868 +               au_kfree_rcu(pos);
31869 +}
31870 +
31871 +static void au_nhash_de_do_free(struct hlist_head *head)
31872 +{
31873 +       struct au_vdir_dehstr *pos;
31874 +       struct hlist_node *node;
31875 +
31876 +       hlist_for_each_entry_safe(pos, node, head, hash)
31877 +               au_cache_free_vdir_dehstr(pos);
31878 +}
31879 +
31880 +static void au_nhash_do_free(struct au_nhash *nhash,
31881 +                            void (*free)(struct hlist_head *head))
31882 +{
31883 +       unsigned int n;
31884 +       struct hlist_head *head;
31885 +
31886 +       n = nhash->nh_num;
31887 +       if (!n)
31888 +               return;
31889 +
31890 +       head = nhash->nh_head;
31891 +       while (n-- > 0) {
31892 +               nhash_count(head);
31893 +               free(head++);
31894 +       }
31895 +       au_kfree_try_rcu(nhash->nh_head);
31896 +}
31897 +
31898 +void au_nhash_wh_free(struct au_nhash *whlist)
31899 +{
31900 +       au_nhash_do_free(whlist, au_nhash_wh_do_free);
31901 +}
31902 +
31903 +static void au_nhash_de_free(struct au_nhash *delist)
31904 +{
31905 +       au_nhash_do_free(delist, au_nhash_de_do_free);
31906 +}
31907 +
31908 +/* ---------------------------------------------------------------------- */
31909 +
31910 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
31911 +                           int limit)
31912 +{
31913 +       int num;
31914 +       unsigned int u, n;
31915 +       struct hlist_head *head;
31916 +       struct au_vdir_wh *pos;
31917 +
31918 +       num = 0;
31919 +       n = whlist->nh_num;
31920 +       head = whlist->nh_head;
31921 +       for (u = 0; u < n; u++, head++)
31922 +               hlist_for_each_entry(pos, head, wh_hash)
31923 +                       if (pos->wh_bindex == btgt && ++num > limit)
31924 +                               return 1;
31925 +       return 0;
31926 +}
31927 +
31928 +static struct hlist_head *au_name_hash(struct au_nhash *nhash,
31929 +                                      unsigned char *name,
31930 +                                      unsigned int len)
31931 +{
31932 +       unsigned int v;
31933 +       /* const unsigned int magic_bit = 12; */
31934 +
31935 +       AuDebugOn(!nhash->nh_num || !nhash->nh_head);
31936 +
31937 +       v = 0;
31938 +       if (len > 8)
31939 +               len = 8;
31940 +       while (len--)
31941 +               v += *name++;
31942 +       /* v = hash_long(v, magic_bit); */
31943 +       v %= nhash->nh_num;
31944 +       return nhash->nh_head + v;
31945 +}
31946 +
31947 +static int au_nhash_test_name(struct au_vdir_destr *str, const char *name,
31948 +                             int nlen)
31949 +{
31950 +       return str->len == nlen && !memcmp(str->name, name, nlen);
31951 +}
31952 +
31953 +/* returns found or not */
31954 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen)
31955 +{
31956 +       struct hlist_head *head;
31957 +       struct au_vdir_wh *pos;
31958 +       struct au_vdir_destr *str;
31959 +
31960 +       head = au_name_hash(whlist, name, nlen);
31961 +       hlist_for_each_entry(pos, head, wh_hash) {
31962 +               str = &pos->wh_str;
31963 +               AuDbg("%.*s\n", str->len, str->name);
31964 +               if (au_nhash_test_name(str, name, nlen))
31965 +                       return 1;
31966 +       }
31967 +       return 0;
31968 +}
31969 +
31970 +/* returns found(true) or not */
31971 +static int test_known(struct au_nhash *delist, char *name, int nlen)
31972 +{
31973 +       struct hlist_head *head;
31974 +       struct au_vdir_dehstr *pos;
31975 +       struct au_vdir_destr *str;
31976 +
31977 +       head = au_name_hash(delist, name, nlen);
31978 +       hlist_for_each_entry(pos, head, hash) {
31979 +               str = pos->str;
31980 +               AuDbg("%.*s\n", str->len, str->name);
31981 +               if (au_nhash_test_name(str, name, nlen))
31982 +                       return 1;
31983 +       }
31984 +       return 0;
31985 +}
31986 +
31987 +static void au_shwh_init_wh(struct au_vdir_wh *wh, ino_t ino,
31988 +                           unsigned char d_type)
31989 +{
31990 +#ifdef CONFIG_AUFS_SHWH
31991 +       wh->wh_ino = ino;
31992 +       wh->wh_type = d_type;
31993 +#endif
31994 +}
31995 +
31996 +/* ---------------------------------------------------------------------- */
31997 +
31998 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
31999 +                      unsigned int d_type, aufs_bindex_t bindex,
32000 +                      unsigned char shwh)
32001 +{
32002 +       int err;
32003 +       struct au_vdir_destr *str;
32004 +       struct au_vdir_wh *wh;
32005 +
32006 +       AuDbg("%.*s\n", nlen, name);
32007 +       AuDebugOn(!whlist->nh_num || !whlist->nh_head);
32008 +
32009 +       err = -ENOMEM;
32010 +       wh = kmalloc(sizeof(*wh) + nlen, GFP_NOFS);
32011 +       if (unlikely(!wh))
32012 +               goto out;
32013 +
32014 +       err = 0;
32015 +       wh->wh_bindex = bindex;
32016 +       if (shwh)
32017 +               au_shwh_init_wh(wh, ino, d_type);
32018 +       str = &wh->wh_str;
32019 +       str->len = nlen;
32020 +       memcpy(str->name, name, nlen);
32021 +       hlist_add_head(&wh->wh_hash, au_name_hash(whlist, name, nlen));
32022 +       /* smp_mb(); */
32023 +
32024 +out:
32025 +       return err;
32026 +}
32027 +
32028 +static int append_deblk(struct au_vdir *vdir)
32029 +{
32030 +       int err;
32031 +       unsigned long ul;
32032 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
32033 +       union au_vdir_deblk_p p, deblk_end;
32034 +       unsigned char **o;
32035 +
32036 +       err = -ENOMEM;
32037 +       o = au_krealloc(vdir->vd_deblk, sizeof(*o) * (vdir->vd_nblk + 1),
32038 +                       GFP_NOFS, /*may_shrink*/0);
32039 +       if (unlikely(!o))
32040 +               goto out;
32041 +
32042 +       vdir->vd_deblk = o;
32043 +       p.deblk = kmalloc(deblk_sz, GFP_NOFS);
32044 +       if (p.deblk) {
32045 +               ul = vdir->vd_nblk++;
32046 +               vdir->vd_deblk[ul] = p.deblk;
32047 +               vdir->vd_last.ul = ul;
32048 +               vdir->vd_last.p.deblk = p.deblk;
32049 +               deblk_end.deblk = p.deblk + deblk_sz;
32050 +               err = set_deblk_end(&p, &deblk_end);
32051 +       }
32052 +
32053 +out:
32054 +       return err;
32055 +}
32056 +
32057 +static int append_de(struct au_vdir *vdir, char *name, int nlen, ino_t ino,
32058 +                    unsigned int d_type, struct au_nhash *delist)
32059 +{
32060 +       int err;
32061 +       unsigned int sz;
32062 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
32063 +       union au_vdir_deblk_p p, *room, deblk_end;
32064 +       struct au_vdir_dehstr *dehstr;
32065 +
32066 +       p.deblk = last_deblk(vdir);
32067 +       deblk_end.deblk = p.deblk + deblk_sz;
32068 +       room = &vdir->vd_last.p;
32069 +       AuDebugOn(room->deblk < p.deblk || deblk_end.deblk <= room->deblk
32070 +                 || !is_deblk_end(room, &deblk_end));
32071 +
32072 +       sz = calc_size(nlen);
32073 +       if (unlikely(sz > deblk_end.deblk - room->deblk)) {
32074 +               err = append_deblk(vdir);
32075 +               if (unlikely(err))
32076 +                       goto out;
32077 +
32078 +               p.deblk = last_deblk(vdir);
32079 +               deblk_end.deblk = p.deblk + deblk_sz;
32080 +               /* smp_mb(); */
32081 +               AuDebugOn(room->deblk != p.deblk);
32082 +       }
32083 +
32084 +       err = -ENOMEM;
32085 +       dehstr = au_cache_alloc_vdir_dehstr();
32086 +       if (unlikely(!dehstr))
32087 +               goto out;
32088 +
32089 +       dehstr->str = &room->de->de_str;
32090 +       hlist_add_head(&dehstr->hash, au_name_hash(delist, name, nlen));
32091 +       room->de->de_ino = ino;
32092 +       room->de->de_type = d_type;
32093 +       room->de->de_str.len = nlen;
32094 +       memcpy(room->de->de_str.name, name, nlen);
32095 +
32096 +       err = 0;
32097 +       room->deblk += sz;
32098 +       if (unlikely(set_deblk_end(room, &deblk_end)))
32099 +               err = append_deblk(vdir);
32100 +       /* smp_mb(); */
32101 +
32102 +out:
32103 +       return err;
32104 +}
32105 +
32106 +/* ---------------------------------------------------------------------- */
32107 +
32108 +void au_vdir_free(struct au_vdir *vdir)
32109 +{
32110 +       unsigned char **deblk;
32111 +
32112 +       deblk = vdir->vd_deblk;
32113 +       while (vdir->vd_nblk--)
32114 +               au_kfree_try_rcu(*deblk++);
32115 +       au_kfree_try_rcu(vdir->vd_deblk);
32116 +       au_cache_free_vdir(vdir);
32117 +}
32118 +
32119 +static struct au_vdir *alloc_vdir(struct file *file)
32120 +{
32121 +       struct au_vdir *vdir;
32122 +       struct super_block *sb;
32123 +       int err;
32124 +
32125 +       sb = file->f_path.dentry->d_sb;
32126 +       SiMustAnyLock(sb);
32127 +
32128 +       err = -ENOMEM;
32129 +       vdir = au_cache_alloc_vdir();
32130 +       if (unlikely(!vdir))
32131 +               goto out;
32132 +
32133 +       vdir->vd_deblk = kzalloc(sizeof(*vdir->vd_deblk), GFP_NOFS);
32134 +       if (unlikely(!vdir->vd_deblk))
32135 +               goto out_free;
32136 +
32137 +       vdir->vd_deblk_sz = au_sbi(sb)->si_rdblk;
32138 +       if (!vdir->vd_deblk_sz) {
32139 +               /* estimate the appropriate size for deblk */
32140 +               vdir->vd_deblk_sz = au_dir_size(file, /*dentry*/NULL);
32141 +               /* pr_info("vd_deblk_sz %u\n", vdir->vd_deblk_sz); */
32142 +       }
32143 +       vdir->vd_nblk = 0;
32144 +       vdir->vd_version = 0;
32145 +       vdir->vd_jiffy = 0;
32146 +       err = append_deblk(vdir);
32147 +       if (!err)
32148 +               return vdir; /* success */
32149 +
32150 +       au_kfree_try_rcu(vdir->vd_deblk);
32151 +
32152 +out_free:
32153 +       au_cache_free_vdir(vdir);
32154 +out:
32155 +       vdir = ERR_PTR(err);
32156 +       return vdir;
32157 +}
32158 +
32159 +static int reinit_vdir(struct au_vdir *vdir)
32160 +{
32161 +       int err;
32162 +       union au_vdir_deblk_p p, deblk_end;
32163 +
32164 +       while (vdir->vd_nblk > 1) {
32165 +               au_kfree_try_rcu(vdir->vd_deblk[vdir->vd_nblk - 1]);
32166 +               /* vdir->vd_deblk[vdir->vd_nblk - 1] = NULL; */
32167 +               vdir->vd_nblk--;
32168 +       }
32169 +       p.deblk = vdir->vd_deblk[0];
32170 +       deblk_end.deblk = p.deblk + vdir->vd_deblk_sz;
32171 +       err = set_deblk_end(&p, &deblk_end);
32172 +       /* keep vd_dblk_sz */
32173 +       vdir->vd_last.ul = 0;
32174 +       vdir->vd_last.p.deblk = vdir->vd_deblk[0];
32175 +       vdir->vd_version = 0;
32176 +       vdir->vd_jiffy = 0;
32177 +       /* smp_mb(); */
32178 +       return err;
32179 +}
32180 +
32181 +/* ---------------------------------------------------------------------- */
32182 +
32183 +#define AuFillVdir_CALLED      1
32184 +#define AuFillVdir_WHABLE      (1 << 1)
32185 +#define AuFillVdir_SHWH                (1 << 2)
32186 +#define au_ftest_fillvdir(flags, name) ((flags) & AuFillVdir_##name)
32187 +#define au_fset_fillvdir(flags, name) \
32188 +       do { (flags) |= AuFillVdir_##name; } while (0)
32189 +#define au_fclr_fillvdir(flags, name) \
32190 +       do { (flags) &= ~AuFillVdir_##name; } while (0)
32191 +
32192 +#ifndef CONFIG_AUFS_SHWH
32193 +#undef AuFillVdir_SHWH
32194 +#define AuFillVdir_SHWH                0
32195 +#endif
32196 +
32197 +struct fillvdir_arg {
32198 +       struct dir_context      ctx;
32199 +       struct file             *file;
32200 +       struct au_vdir          *vdir;
32201 +       struct au_nhash         delist;
32202 +       struct au_nhash         whlist;
32203 +       aufs_bindex_t           bindex;
32204 +       unsigned int            flags;
32205 +       int                     err;
32206 +};
32207 +
32208 +static int fillvdir(struct dir_context *ctx, const char *__name, int nlen,
32209 +                   loff_t offset __maybe_unused, u64 h_ino,
32210 +                   unsigned int d_type)
32211 +{
32212 +       struct fillvdir_arg *arg = container_of(ctx, struct fillvdir_arg, ctx);
32213 +       char *name = (void *)__name;
32214 +       struct super_block *sb;
32215 +       ino_t ino;
32216 +       const unsigned char shwh = !!au_ftest_fillvdir(arg->flags, SHWH);
32217 +
32218 +       arg->err = 0;
32219 +       sb = arg->file->f_path.dentry->d_sb;
32220 +       au_fset_fillvdir(arg->flags, CALLED);
32221 +       /* smp_mb(); */
32222 +       if (nlen <= AUFS_WH_PFX_LEN
32223 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
32224 +               if (test_known(&arg->delist, name, nlen)
32225 +                   || au_nhash_test_known_wh(&arg->whlist, name, nlen))
32226 +                       goto out; /* already exists or whiteouted */
32227 +
32228 +               arg->err = au_ino(sb, arg->bindex, h_ino, d_type, &ino);
32229 +               if (!arg->err) {
32230 +                       if (unlikely(nlen > AUFS_MAX_NAMELEN))
32231 +                               d_type = DT_UNKNOWN;
32232 +                       arg->err = append_de(arg->vdir, name, nlen, ino,
32233 +                                            d_type, &arg->delist);
32234 +               }
32235 +       } else if (au_ftest_fillvdir(arg->flags, WHABLE)) {
32236 +               name += AUFS_WH_PFX_LEN;
32237 +               nlen -= AUFS_WH_PFX_LEN;
32238 +               if (au_nhash_test_known_wh(&arg->whlist, name, nlen))
32239 +                       goto out; /* already whiteouted */
32240 +
32241 +               ino = 0; /* just to suppress a warning */
32242 +               if (shwh)
32243 +                       arg->err = au_wh_ino(sb, arg->bindex, h_ino, d_type,
32244 +                                            &ino);
32245 +               if (!arg->err) {
32246 +                       if (nlen <= AUFS_MAX_NAMELEN + AUFS_WH_PFX_LEN)
32247 +                               d_type = DT_UNKNOWN;
32248 +                       arg->err = au_nhash_append_wh
32249 +                               (&arg->whlist, name, nlen, ino, d_type,
32250 +                                arg->bindex, shwh);
32251 +               }
32252 +       }
32253 +
32254 +out:
32255 +       if (!arg->err)
32256 +               arg->vdir->vd_jiffy = jiffies;
32257 +       /* smp_mb(); */
32258 +       AuTraceErr(arg->err);
32259 +       return arg->err;
32260 +}
32261 +
32262 +static int au_handle_shwh(struct super_block *sb, struct au_vdir *vdir,
32263 +                         struct au_nhash *whlist, struct au_nhash *delist)
32264 +{
32265 +#ifdef CONFIG_AUFS_SHWH
32266 +       int err;
32267 +       unsigned int nh, u;
32268 +       struct hlist_head *head;
32269 +       struct au_vdir_wh *pos;
32270 +       struct hlist_node *n;
32271 +       char *p, *o;
32272 +       struct au_vdir_destr *destr;
32273 +
32274 +       AuDebugOn(!au_opt_test(au_mntflags(sb), SHWH));
32275 +
32276 +       err = -ENOMEM;
32277 +       o = p = (void *)__get_free_page(GFP_NOFS);
32278 +       if (unlikely(!p))
32279 +               goto out;
32280 +
32281 +       err = 0;
32282 +       nh = whlist->nh_num;
32283 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
32284 +       p += AUFS_WH_PFX_LEN;
32285 +       for (u = 0; u < nh; u++) {
32286 +               head = whlist->nh_head + u;
32287 +               hlist_for_each_entry_safe(pos, n, head, wh_hash) {
32288 +                       destr = &pos->wh_str;
32289 +                       memcpy(p, destr->name, destr->len);
32290 +                       err = append_de(vdir, o, destr->len + AUFS_WH_PFX_LEN,
32291 +                                       pos->wh_ino, pos->wh_type, delist);
32292 +                       if (unlikely(err))
32293 +                               break;
32294 +               }
32295 +       }
32296 +
32297 +       free_page((unsigned long)o);
32298 +
32299 +out:
32300 +       AuTraceErr(err);
32301 +       return err;
32302 +#else
32303 +       return 0;
32304 +#endif
32305 +}
32306 +
32307 +static int au_do_read_vdir(struct fillvdir_arg *arg)
32308 +{
32309 +       int err;
32310 +       unsigned int rdhash;
32311 +       loff_t offset;
32312 +       aufs_bindex_t bbot, bindex, btop;
32313 +       unsigned char shwh;
32314 +       struct file *hf, *file;
32315 +       struct super_block *sb;
32316 +
32317 +       file = arg->file;
32318 +       sb = file->f_path.dentry->d_sb;
32319 +       SiMustAnyLock(sb);
32320 +
32321 +       rdhash = au_sbi(sb)->si_rdhash;
32322 +       if (!rdhash)
32323 +               rdhash = au_rdhash_est(au_dir_size(file, /*dentry*/NULL));
32324 +       err = au_nhash_alloc(&arg->delist, rdhash, GFP_NOFS);
32325 +       if (unlikely(err))
32326 +               goto out;
32327 +       err = au_nhash_alloc(&arg->whlist, rdhash, GFP_NOFS);
32328 +       if (unlikely(err))
32329 +               goto out_delist;
32330 +
32331 +       err = 0;
32332 +       arg->flags = 0;
32333 +       shwh = 0;
32334 +       if (au_opt_test(au_mntflags(sb), SHWH)) {
32335 +               shwh = 1;
32336 +               au_fset_fillvdir(arg->flags, SHWH);
32337 +       }
32338 +       btop = au_fbtop(file);
32339 +       bbot = au_fbbot_dir(file);
32340 +       for (bindex = btop; !err && bindex <= bbot; bindex++) {
32341 +               hf = au_hf_dir(file, bindex);
32342 +               if (!hf)
32343 +                       continue;
32344 +
32345 +               offset = vfsub_llseek(hf, 0, SEEK_SET);
32346 +               err = offset;
32347 +               if (unlikely(offset))
32348 +                       break;
32349 +
32350 +               arg->bindex = bindex;
32351 +               au_fclr_fillvdir(arg->flags, WHABLE);
32352 +               if (shwh
32353 +                   || (bindex != bbot
32354 +                       && au_br_whable(au_sbr_perm(sb, bindex))))
32355 +                       au_fset_fillvdir(arg->flags, WHABLE);
32356 +               do {
32357 +                       arg->err = 0;
32358 +                       au_fclr_fillvdir(arg->flags, CALLED);
32359 +                       /* smp_mb(); */
32360 +                       err = vfsub_iterate_dir(hf, &arg->ctx);
32361 +                       if (err >= 0)
32362 +                               err = arg->err;
32363 +               } while (!err && au_ftest_fillvdir(arg->flags, CALLED));
32364 +
32365 +               /*
32366 +                * dir_relax() may be good for concurrency, but aufs should not
32367 +                * use it since it will cause a lockdep problem.
32368 +                */
32369 +       }
32370 +
32371 +       if (!err && shwh)
32372 +               err = au_handle_shwh(sb, arg->vdir, &arg->whlist, &arg->delist);
32373 +
32374 +       au_nhash_wh_free(&arg->whlist);
32375 +
32376 +out_delist:
32377 +       au_nhash_de_free(&arg->delist);
32378 +out:
32379 +       return err;
32380 +}
32381 +
32382 +static int read_vdir(struct file *file, int may_read)
32383 +{
32384 +       int err;
32385 +       unsigned long expire;
32386 +       unsigned char do_read;
32387 +       struct fillvdir_arg arg = {
32388 +               .ctx = {
32389 +                       .actor = fillvdir
32390 +               }
32391 +       };
32392 +       struct inode *inode;
32393 +       struct au_vdir *vdir, *allocated;
32394 +
32395 +       err = 0;
32396 +       inode = file_inode(file);
32397 +       IMustLock(inode);
32398 +       IiMustWriteLock(inode);
32399 +       SiMustAnyLock(inode->i_sb);
32400 +
32401 +       allocated = NULL;
32402 +       do_read = 0;
32403 +       expire = au_sbi(inode->i_sb)->si_rdcache;
32404 +       vdir = au_ivdir(inode);
32405 +       if (!vdir) {
32406 +               do_read = 1;
32407 +               vdir = alloc_vdir(file);
32408 +               err = PTR_ERR(vdir);
32409 +               if (IS_ERR(vdir))
32410 +                       goto out;
32411 +               err = 0;
32412 +               allocated = vdir;
32413 +       } else if (may_read
32414 +                  && (!inode_eq_iversion(inode, vdir->vd_version)
32415 +                      || time_after(jiffies, vdir->vd_jiffy + expire))) {
32416 +               do_read = 1;
32417 +               err = reinit_vdir(vdir);
32418 +               if (unlikely(err))
32419 +                       goto out;
32420 +       }
32421 +
32422 +       if (!do_read)
32423 +               return 0; /* success */
32424 +
32425 +       arg.file = file;
32426 +       arg.vdir = vdir;
32427 +       err = au_do_read_vdir(&arg);
32428 +       if (!err) {
32429 +               /* file->f_pos = 0; */ /* todo: ctx->pos? */
32430 +               vdir->vd_version = inode_query_iversion(inode);
32431 +               vdir->vd_last.ul = 0;
32432 +               vdir->vd_last.p.deblk = vdir->vd_deblk[0];
32433 +               if (allocated)
32434 +                       au_set_ivdir(inode, allocated);
32435 +       } else if (allocated)
32436 +               au_vdir_free(allocated);
32437 +
32438 +out:
32439 +       return err;
32440 +}
32441 +
32442 +static int copy_vdir(struct au_vdir *tgt, struct au_vdir *src)
32443 +{
32444 +       int err, rerr;
32445 +       unsigned long ul, n;
32446 +       const unsigned int deblk_sz = src->vd_deblk_sz;
32447 +
32448 +       AuDebugOn(tgt->vd_nblk != 1);
32449 +
32450 +       err = -ENOMEM;
32451 +       if (tgt->vd_nblk < src->vd_nblk) {
32452 +               unsigned char **p;
32453 +
32454 +               p = au_krealloc(tgt->vd_deblk, sizeof(*p) * src->vd_nblk,
32455 +                               GFP_NOFS, /*may_shrink*/0);
32456 +               if (unlikely(!p))
32457 +                       goto out;
32458 +               tgt->vd_deblk = p;
32459 +       }
32460 +
32461 +       if (tgt->vd_deblk_sz != deblk_sz) {
32462 +               unsigned char *p;
32463 +
32464 +               tgt->vd_deblk_sz = deblk_sz;
32465 +               p = au_krealloc(tgt->vd_deblk[0], deblk_sz, GFP_NOFS,
32466 +                               /*may_shrink*/1);
32467 +               if (unlikely(!p))
32468 +                       goto out;
32469 +               tgt->vd_deblk[0] = p;
32470 +       }
32471 +       memcpy(tgt->vd_deblk[0], src->vd_deblk[0], deblk_sz);
32472 +       tgt->vd_version = src->vd_version;
32473 +       tgt->vd_jiffy = src->vd_jiffy;
32474 +
32475 +       n = src->vd_nblk;
32476 +       for (ul = 1; ul < n; ul++) {
32477 +               tgt->vd_deblk[ul] = kmemdup(src->vd_deblk[ul], deblk_sz,
32478 +                                           GFP_NOFS);
32479 +               if (unlikely(!tgt->vd_deblk[ul]))
32480 +                       goto out;
32481 +               tgt->vd_nblk++;
32482 +       }
32483 +       tgt->vd_nblk = n;
32484 +       tgt->vd_last.ul = tgt->vd_last.ul;
32485 +       tgt->vd_last.p.deblk = tgt->vd_deblk[tgt->vd_last.ul];
32486 +       tgt->vd_last.p.deblk += src->vd_last.p.deblk
32487 +               - src->vd_deblk[src->vd_last.ul];
32488 +       /* smp_mb(); */
32489 +       return 0; /* success */
32490 +
32491 +out:
32492 +       rerr = reinit_vdir(tgt);
32493 +       BUG_ON(rerr);
32494 +       return err;
32495 +}
32496 +
32497 +int au_vdir_init(struct file *file)
32498 +{
32499 +       int err;
32500 +       struct inode *inode;
32501 +       struct au_vdir *vdir_cache, *allocated;
32502 +
32503 +       /* test file->f_pos here instead of ctx->pos */
32504 +       err = read_vdir(file, !file->f_pos);
32505 +       if (unlikely(err))
32506 +               goto out;
32507 +
32508 +       allocated = NULL;
32509 +       vdir_cache = au_fvdir_cache(file);
32510 +       if (!vdir_cache) {
32511 +               vdir_cache = alloc_vdir(file);
32512 +               err = PTR_ERR(vdir_cache);
32513 +               if (IS_ERR(vdir_cache))
32514 +                       goto out;
32515 +               allocated = vdir_cache;
32516 +       } else if (!file->f_pos && vdir_cache->vd_version != file->f_version) {
32517 +               /* test file->f_pos here instead of ctx->pos */
32518 +               err = reinit_vdir(vdir_cache);
32519 +               if (unlikely(err))
32520 +                       goto out;
32521 +       } else
32522 +               return 0; /* success */
32523 +
32524 +       inode = file_inode(file);
32525 +       err = copy_vdir(vdir_cache, au_ivdir(inode));
32526 +       if (!err) {
32527 +               file->f_version = inode_query_iversion(inode);
32528 +               if (allocated)
32529 +                       au_set_fvdir_cache(file, allocated);
32530 +       } else if (allocated)
32531 +               au_vdir_free(allocated);
32532 +
32533 +out:
32534 +       return err;
32535 +}
32536 +
32537 +static loff_t calc_offset(struct au_vdir *vdir)
32538 +{
32539 +       loff_t offset;
32540 +       union au_vdir_deblk_p p;
32541 +
32542 +       p.deblk = vdir->vd_deblk[vdir->vd_last.ul];
32543 +       offset = vdir->vd_last.p.deblk - p.deblk;
32544 +       offset += vdir->vd_deblk_sz * vdir->vd_last.ul;
32545 +       return offset;
32546 +}
32547 +
32548 +/* returns true or false */
32549 +static int seek_vdir(struct file *file, struct dir_context *ctx)
32550 +{
32551 +       int valid;
32552 +       unsigned int deblk_sz;
32553 +       unsigned long ul, n;
32554 +       loff_t offset;
32555 +       union au_vdir_deblk_p p, deblk_end;
32556 +       struct au_vdir *vdir_cache;
32557 +
32558 +       valid = 1;
32559 +       vdir_cache = au_fvdir_cache(file);
32560 +       offset = calc_offset(vdir_cache);
32561 +       AuDbg("offset %lld\n", offset);
32562 +       if (ctx->pos == offset)
32563 +               goto out;
32564 +
32565 +       vdir_cache->vd_last.ul = 0;
32566 +       vdir_cache->vd_last.p.deblk = vdir_cache->vd_deblk[0];
32567 +       if (!ctx->pos)
32568 +               goto out;
32569 +
32570 +       valid = 0;
32571 +       deblk_sz = vdir_cache->vd_deblk_sz;
32572 +       ul = div64_u64(ctx->pos, deblk_sz);
32573 +       AuDbg("ul %lu\n", ul);
32574 +       if (ul >= vdir_cache->vd_nblk)
32575 +               goto out;
32576 +
32577 +       n = vdir_cache->vd_nblk;
32578 +       for (; ul < n; ul++) {
32579 +               p.deblk = vdir_cache->vd_deblk[ul];
32580 +               deblk_end.deblk = p.deblk + deblk_sz;
32581 +               offset = ul;
32582 +               offset *= deblk_sz;
32583 +               while (!is_deblk_end(&p, &deblk_end) && offset < ctx->pos) {
32584 +                       unsigned int l;
32585 +
32586 +                       l = calc_size(p.de->de_str.len);
32587 +                       offset += l;
32588 +                       p.deblk += l;
32589 +               }
32590 +               if (!is_deblk_end(&p, &deblk_end)) {
32591 +                       valid = 1;
32592 +                       vdir_cache->vd_last.ul = ul;
32593 +                       vdir_cache->vd_last.p = p;
32594 +                       break;
32595 +               }
32596 +       }
32597 +
32598 +out:
32599 +       /* smp_mb(); */
32600 +       if (!valid)
32601 +               AuDbg("valid %d\n", !valid);
32602 +       return valid;
32603 +}
32604 +
32605 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx)
32606 +{
32607 +       unsigned int l, deblk_sz;
32608 +       union au_vdir_deblk_p deblk_end;
32609 +       struct au_vdir *vdir_cache;
32610 +       struct au_vdir_de *de;
32611 +
32612 +       if (!seek_vdir(file, ctx))
32613 +               return 0;
32614 +
32615 +       vdir_cache = au_fvdir_cache(file);
32616 +       deblk_sz = vdir_cache->vd_deblk_sz;
32617 +       while (1) {
32618 +               deblk_end.deblk = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
32619 +               deblk_end.deblk += deblk_sz;
32620 +               while (!is_deblk_end(&vdir_cache->vd_last.p, &deblk_end)) {
32621 +                       de = vdir_cache->vd_last.p.de;
32622 +                       AuDbg("%.*s, off%lld, i%lu, dt%d\n",
32623 +                             de->de_str.len, de->de_str.name, ctx->pos,
32624 +                             (unsigned long)de->de_ino, de->de_type);
32625 +                       if (unlikely(!dir_emit(ctx, de->de_str.name,
32626 +                                              de->de_str.len, de->de_ino,
32627 +                                              de->de_type))) {
32628 +                               /* todo: ignore the error caused by udba? */
32629 +                               /* return err; */
32630 +                               return 0;
32631 +                       }
32632 +
32633 +                       l = calc_size(de->de_str.len);
32634 +                       vdir_cache->vd_last.p.deblk += l;
32635 +                       ctx->pos += l;
32636 +               }
32637 +               if (vdir_cache->vd_last.ul < vdir_cache->vd_nblk - 1) {
32638 +                       vdir_cache->vd_last.ul++;
32639 +                       vdir_cache->vd_last.p.deblk
32640 +                               = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
32641 +                       ctx->pos = deblk_sz * vdir_cache->vd_last.ul;
32642 +                       continue;
32643 +               }
32644 +               break;
32645 +       }
32646 +
32647 +       /* smp_mb(); */
32648 +       return 0;
32649 +}
32650 diff -urN /usr/share/empty/fs/aufs/vfsub.c linux/fs/aufs/vfsub.c
32651 --- /usr/share/empty/fs/aufs/vfsub.c    1970-01-01 01:00:00.000000000 +0100
32652 +++ linux/fs/aufs/vfsub.c       2021-12-03 15:40:58.236647297 +0100
32653 @@ -0,0 +1,919 @@
32654 +// SPDX-License-Identifier: GPL-2.0
32655 +/*
32656 + * Copyright (C) 2005-2021 Junjiro R. Okajima
32657 + *
32658 + * This program, aufs is free software; you can redistribute it and/or modify
32659 + * it under the terms of the GNU General Public License as published by
32660 + * the Free Software Foundation; either version 2 of the License, or
32661 + * (at your option) any later version.
32662 + *
32663 + * This program is distributed in the hope that it will be useful,
32664 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32665 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32666 + * GNU General Public License for more details.
32667 + *
32668 + * You should have received a copy of the GNU General Public License
32669 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32670 + */
32671 +
32672 +/*
32673 + * sub-routines for VFS
32674 + */
32675 +
32676 +#include <linux/mnt_namespace.h>
32677 +#include <linux/namei.h>
32678 +#include <linux/nsproxy.h>
32679 +#include <linux/security.h>
32680 +#include <linux/splice.h>
32681 +#include "aufs.h"
32682 +
32683 +#ifdef CONFIG_AUFS_BR_FUSE
32684 +int vfsub_test_mntns(struct vfsmount *mnt, struct super_block *h_sb)
32685 +{
32686 +       if (!au_test_fuse(h_sb) || !au_userns)
32687 +               return 0;
32688 +
32689 +       return is_current_mnt_ns(mnt) ? 0 : -EACCES;
32690 +}
32691 +#endif
32692 +
32693 +int vfsub_sync_filesystem(struct super_block *h_sb, int wait)
32694 +{
32695 +       int err;
32696 +
32697 +       lockdep_off();
32698 +       down_read(&h_sb->s_umount);
32699 +       err = __sync_filesystem(h_sb, wait);
32700 +       up_read(&h_sb->s_umount);
32701 +       lockdep_on();
32702 +
32703 +       return err;
32704 +}
32705 +
32706 +/* ---------------------------------------------------------------------- */
32707 +
32708 +int vfsub_update_h_iattr(struct path *h_path, int *did)
32709 +{
32710 +       int err;
32711 +       struct kstat st;
32712 +       struct super_block *h_sb;
32713 +
32714 +       /*
32715 +        * Always needs h_path->mnt for LSM or FUSE branch.
32716 +        */
32717 +       AuDebugOn(!h_path->mnt);
32718 +
32719 +       /* for remote fs, leave work for its getattr or d_revalidate */
32720 +       /* for bad i_attr fs, handle them in aufs_getattr() */
32721 +       /* still some fs may acquire i_mutex. we need to skip them */
32722 +       err = 0;
32723 +       if (!did)
32724 +               did = &err;
32725 +       h_sb = h_path->dentry->d_sb;
32726 +       *did = (!au_test_fs_remote(h_sb) && au_test_fs_refresh_iattr(h_sb));
32727 +       if (*did)
32728 +               err = vfsub_getattr(h_path, &st);
32729 +
32730 +       return err;
32731 +}
32732 +
32733 +/* ---------------------------------------------------------------------- */
32734 +
32735 +struct file *vfsub_dentry_open(struct path *path, int flags)
32736 +{
32737 +       return dentry_open(path, flags /* | __FMODE_NONOTIFY */,
32738 +                          current_cred());
32739 +}
32740 +
32741 +struct file *vfsub_filp_open(const char *path, int oflags, int mode)
32742 +{
32743 +       struct file *file;
32744 +
32745 +       lockdep_off();
32746 +       file = filp_open(path,
32747 +                        oflags /* | __FMODE_NONOTIFY */,
32748 +                        mode);
32749 +       lockdep_on();
32750 +       if (IS_ERR(file))
32751 +               goto out;
32752 +       vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
32753 +
32754 +out:
32755 +       return file;
32756 +}
32757 +
32758 +/*
32759 + * Ideally this function should call VFS:do_last() in order to keep all its
32760 + * checkings. But it is very hard for aufs to regenerate several VFS internal
32761 + * structure such as nameidata. This is a second (or third) best approach.
32762 + * cf. linux/fs/namei.c:do_last(), lookup_open() and atomic_open().
32763 + */
32764 +int vfsub_atomic_open(struct inode *dir, struct dentry *dentry,
32765 +                     struct vfsub_aopen_args *args)
32766 +{
32767 +       int err;
32768 +       struct au_branch *br = args->br;
32769 +       struct file *file = args->file;
32770 +       /* copied from linux/fs/namei.c:atomic_open() */
32771 +       struct dentry *const DENTRY_NOT_SET = (void *)-1UL;
32772 +
32773 +       IMustLock(dir);
32774 +       AuDebugOn(!dir->i_op->atomic_open);
32775 +
32776 +       err = au_br_test_oflag(args->open_flag, br);
32777 +       if (unlikely(err))
32778 +               goto out;
32779 +
32780 +       au_lcnt_inc(&br->br_nfiles);
32781 +       file->f_path.dentry = DENTRY_NOT_SET;
32782 +       file->f_path.mnt = au_br_mnt(br);
32783 +       AuDbg("%ps\n", dir->i_op->atomic_open);
32784 +       err = dir->i_op->atomic_open(dir, dentry, file, args->open_flag,
32785 +                                    args->create_mode);
32786 +       if (unlikely(err < 0)) {
32787 +               au_lcnt_dec(&br->br_nfiles);
32788 +               goto out;
32789 +       }
32790 +
32791 +       /* temporary workaround for nfsv4 branch */
32792 +       if (au_test_nfs(dir->i_sb))
32793 +               nfs_mark_for_revalidate(dir);
32794 +
32795 +       if (file->f_mode & FMODE_CREATED)
32796 +               fsnotify_create(dir, dentry);
32797 +       if (!(file->f_mode & FMODE_OPENED)) {
32798 +               au_lcnt_dec(&br->br_nfiles);
32799 +               goto out;
32800 +       }
32801 +
32802 +       /* todo: call VFS:may_open() here */
32803 +       /* todo: ima_file_check() too? */
32804 +       if (!err && (args->open_flag & __FMODE_EXEC))
32805 +               err = deny_write_access(file);
32806 +       if (!err)
32807 +               fsnotify_open(file);
32808 +       else
32809 +               au_lcnt_dec(&br->br_nfiles);
32810 +       /* note that the file is created and still opened */
32811 +
32812 +out:
32813 +       return err;
32814 +}
32815 +
32816 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path)
32817 +{
32818 +       int err;
32819 +
32820 +       err = kern_path(name, flags, path);
32821 +       if (!err && d_is_positive(path->dentry))
32822 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
32823 +       return err;
32824 +}
32825 +
32826 +struct dentry *vfsub_lookup_one_len_unlocked(const char *name,
32827 +                                            struct path *ppath, int len)
32828 +{
32829 +       struct path path;
32830 +
32831 +       path.dentry = lookup_one_len_unlocked(name, ppath->dentry, len);
32832 +       if (IS_ERR(path.dentry))
32833 +               goto out;
32834 +       if (d_is_positive(path.dentry)) {
32835 +               path.mnt = ppath->mnt;
32836 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
32837 +       }
32838 +
32839 +out:
32840 +       AuTraceErrPtr(path.dentry);
32841 +       return path.dentry;
32842 +}
32843 +
32844 +struct dentry *vfsub_lookup_one_len(const char *name, struct path *ppath,
32845 +                                   int len)
32846 +{
32847 +       struct path path;
32848 +
32849 +       /* VFS checks it too, but by WARN_ON_ONCE() */
32850 +       IMustLock(d_inode(ppath->dentry));
32851 +
32852 +       path.dentry = lookup_one_len(name, ppath->dentry, len);
32853 +       if (IS_ERR(path.dentry))
32854 +               goto out;
32855 +       if (d_is_positive(path.dentry)) {
32856 +               path.mnt = ppath->mnt;
32857 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
32858 +       }
32859 +
32860 +out:
32861 +       AuTraceErrPtr(path.dentry);
32862 +       return path.dentry;
32863 +}
32864 +
32865 +void vfsub_call_lkup_one(void *args)
32866 +{
32867 +       struct vfsub_lkup_one_args *a = args;
32868 +       *a->errp = vfsub_lkup_one(a->name, a->ppath);
32869 +}
32870 +
32871 +/* ---------------------------------------------------------------------- */
32872 +
32873 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
32874 +                                struct dentry *d2, struct au_hinode *hdir2)
32875 +{
32876 +       struct dentry *d;
32877 +
32878 +       lockdep_off();
32879 +       d = lock_rename(d1, d2);
32880 +       lockdep_on();
32881 +       au_hn_suspend(hdir1);
32882 +       if (hdir1 != hdir2)
32883 +               au_hn_suspend(hdir2);
32884 +
32885 +       return d;
32886 +}
32887 +
32888 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
32889 +                        struct dentry *d2, struct au_hinode *hdir2)
32890 +{
32891 +       au_hn_resume(hdir1);
32892 +       if (hdir1 != hdir2)
32893 +               au_hn_resume(hdir2);
32894 +       lockdep_off();
32895 +       unlock_rename(d1, d2);
32896 +       lockdep_on();
32897 +}
32898 +
32899 +/* ---------------------------------------------------------------------- */
32900 +
32901 +int vfsub_create(struct inode *dir, struct path *path, int mode, bool want_excl)
32902 +{
32903 +       int err;
32904 +       struct dentry *d;
32905 +       struct user_namespace *userns;
32906 +
32907 +       IMustLock(dir);
32908 +
32909 +       d = path->dentry;
32910 +       path->dentry = d->d_parent;
32911 +       err = security_path_mknod(path, d, mode, 0);
32912 +       path->dentry = d;
32913 +       if (unlikely(err))
32914 +               goto out;
32915 +       userns = mnt_user_ns(path->mnt);
32916 +
32917 +       lockdep_off();
32918 +       err = vfs_create(userns, dir, path->dentry, mode, want_excl);
32919 +       lockdep_on();
32920 +       if (!err) {
32921 +               struct path tmp = *path;
32922 +               int did;
32923 +
32924 +               vfsub_update_h_iattr(&tmp, &did);
32925 +               if (did) {
32926 +                       tmp.dentry = path->dentry->d_parent;
32927 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
32928 +               }
32929 +               /*ignore*/
32930 +       }
32931 +
32932 +out:
32933 +       return err;
32934 +}
32935 +
32936 +int vfsub_symlink(struct inode *dir, struct path *path, const char *symname)
32937 +{
32938 +       int err;
32939 +       struct dentry *d;
32940 +       struct user_namespace *userns;
32941 +
32942 +       IMustLock(dir);
32943 +
32944 +       d = path->dentry;
32945 +       path->dentry = d->d_parent;
32946 +       err = security_path_symlink(path, d, symname);
32947 +       path->dentry = d;
32948 +       if (unlikely(err))
32949 +               goto out;
32950 +       userns = mnt_user_ns(path->mnt);
32951 +
32952 +       lockdep_off();
32953 +       err = vfs_symlink(userns, dir, path->dentry, symname);
32954 +       lockdep_on();
32955 +       if (!err) {
32956 +               struct path tmp = *path;
32957 +               int did;
32958 +
32959 +               vfsub_update_h_iattr(&tmp, &did);
32960 +               if (did) {
32961 +                       tmp.dentry = path->dentry->d_parent;
32962 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
32963 +               }
32964 +               /*ignore*/
32965 +       }
32966 +
32967 +out:
32968 +       return err;
32969 +}
32970 +
32971 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev)
32972 +{
32973 +       int err;
32974 +       struct dentry *d;
32975 +       struct user_namespace *userns;
32976 +
32977 +       IMustLock(dir);
32978 +
32979 +       d = path->dentry;
32980 +       path->dentry = d->d_parent;
32981 +       err = security_path_mknod(path, d, mode, new_encode_dev(dev));
32982 +       path->dentry = d;
32983 +       if (unlikely(err))
32984 +               goto out;
32985 +       userns = mnt_user_ns(path->mnt);
32986 +
32987 +       lockdep_off();
32988 +       err = vfs_mknod(userns, dir, path->dentry, mode, dev);
32989 +       lockdep_on();
32990 +       if (!err) {
32991 +               struct path tmp = *path;
32992 +               int did;
32993 +
32994 +               vfsub_update_h_iattr(&tmp, &did);
32995 +               if (did) {
32996 +                       tmp.dentry = path->dentry->d_parent;
32997 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
32998 +               }
32999 +               /*ignore*/
33000 +       }
33001 +
33002 +out:
33003 +       return err;
33004 +}
33005 +
33006 +static int au_test_nlink(struct inode *inode)
33007 +{
33008 +       const unsigned int link_max = UINT_MAX >> 1; /* rough margin */
33009 +
33010 +       if (!au_test_fs_no_limit_nlink(inode->i_sb)
33011 +           || inode->i_nlink < link_max)
33012 +               return 0;
33013 +       return -EMLINK;
33014 +}
33015 +
33016 +int vfsub_link(struct dentry *src_dentry, struct inode *dir, struct path *path,
33017 +              struct inode **delegated_inode)
33018 +{
33019 +       int err;
33020 +       struct dentry *d;
33021 +       struct user_namespace *userns;
33022 +
33023 +       IMustLock(dir);
33024 +
33025 +       err = au_test_nlink(d_inode(src_dentry));
33026 +       if (unlikely(err))
33027 +               return err;
33028 +
33029 +       /* we don't call may_linkat() */
33030 +       d = path->dentry;
33031 +       path->dentry = d->d_parent;
33032 +       err = security_path_link(src_dentry, path, d);
33033 +       path->dentry = d;
33034 +       if (unlikely(err))
33035 +               goto out;
33036 +       userns = mnt_user_ns(path->mnt);
33037 +
33038 +       lockdep_off();
33039 +       err = vfs_link(src_dentry, userns, dir, path->dentry, delegated_inode);
33040 +       lockdep_on();
33041 +       if (!err) {
33042 +               struct path tmp = *path;
33043 +               int did;
33044 +
33045 +               /* fuse has different memory inode for the same inumber */
33046 +               vfsub_update_h_iattr(&tmp, &did);
33047 +               if (did) {
33048 +                       tmp.dentry = path->dentry->d_parent;
33049 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
33050 +                       tmp.dentry = src_dentry;
33051 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
33052 +               }
33053 +               /*ignore*/
33054 +       }
33055 +
33056 +out:
33057 +       return err;
33058 +}
33059 +
33060 +int vfsub_rename(struct inode *src_dir, struct dentry *src_dentry,
33061 +                struct inode *dir, struct path *path,
33062 +                struct inode **delegated_inode, unsigned int flags)
33063 +{
33064 +       int err;
33065 +       struct renamedata rd;
33066 +       struct path tmp = {
33067 +               .mnt    = path->mnt
33068 +       };
33069 +       struct dentry *d;
33070 +
33071 +       IMustLock(dir);
33072 +       IMustLock(src_dir);
33073 +
33074 +       d = path->dentry;
33075 +       path->dentry = d->d_parent;
33076 +       tmp.dentry = src_dentry->d_parent;
33077 +       err = security_path_rename(&tmp, src_dentry, path, d, /*flags*/0);
33078 +       path->dentry = d;
33079 +       if (unlikely(err))
33080 +               goto out;
33081 +
33082 +       rd.old_mnt_userns = mnt_user_ns(path->mnt);
33083 +       rd.old_dir = src_dir;
33084 +       rd.old_dentry = src_dentry;
33085 +       rd.new_mnt_userns = rd.old_mnt_userns;
33086 +       rd.new_dir = dir;
33087 +       rd.new_dentry = path->dentry;
33088 +       rd.delegated_inode = delegated_inode;
33089 +       rd.flags = flags;
33090 +       lockdep_off();
33091 +       err = vfs_rename(&rd);
33092 +       lockdep_on();
33093 +       if (!err) {
33094 +               int did;
33095 +
33096 +               tmp.dentry = d->d_parent;
33097 +               vfsub_update_h_iattr(&tmp, &did);
33098 +               if (did) {
33099 +                       tmp.dentry = src_dentry;
33100 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
33101 +                       tmp.dentry = src_dentry->d_parent;
33102 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
33103 +               }
33104 +               /*ignore*/
33105 +       }
33106 +
33107 +out:
33108 +       return err;
33109 +}
33110 +
33111 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode)
33112 +{
33113 +       int err;
33114 +       struct dentry *d;
33115 +       struct user_namespace *userns;
33116 +
33117 +       IMustLock(dir);
33118 +
33119 +       d = path->dentry;
33120 +       path->dentry = d->d_parent;
33121 +       err = security_path_mkdir(path, d, mode);
33122 +       path->dentry = d;
33123 +       if (unlikely(err))
33124 +               goto out;
33125 +       userns = mnt_user_ns(path->mnt);
33126 +
33127 +       lockdep_off();
33128 +       err = vfs_mkdir(userns, dir, path->dentry, mode);
33129 +       lockdep_on();
33130 +       if (!err) {
33131 +               struct path tmp = *path;
33132 +               int did;
33133 +
33134 +               vfsub_update_h_iattr(&tmp, &did);
33135 +               if (did) {
33136 +                       tmp.dentry = path->dentry->d_parent;
33137 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
33138 +               }
33139 +               /*ignore*/
33140 +       }
33141 +
33142 +out:
33143 +       return err;
33144 +}
33145 +
33146 +int vfsub_rmdir(struct inode *dir, struct path *path)
33147 +{
33148 +       int err;
33149 +       struct dentry *d;
33150 +       struct user_namespace *userns;
33151 +
33152 +       IMustLock(dir);
33153 +
33154 +       d = path->dentry;
33155 +       path->dentry = d->d_parent;
33156 +       err = security_path_rmdir(path, d);
33157 +       path->dentry = d;
33158 +       if (unlikely(err))
33159 +               goto out;
33160 +       userns = mnt_user_ns(path->mnt);
33161 +
33162 +       lockdep_off();
33163 +       err = vfs_rmdir(userns, dir, path->dentry);
33164 +       lockdep_on();
33165 +       if (!err) {
33166 +               struct path tmp = {
33167 +                       .dentry = path->dentry->d_parent,
33168 +                       .mnt    = path->mnt
33169 +               };
33170 +
33171 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
33172 +       }
33173 +
33174 +out:
33175 +       return err;
33176 +}
33177 +
33178 +/* ---------------------------------------------------------------------- */
33179 +
33180 +/* todo: support mmap_sem? */
33181 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
33182 +                    loff_t *ppos)
33183 +{
33184 +       ssize_t err;
33185 +
33186 +       lockdep_off();
33187 +       err = vfs_read(file, ubuf, count, ppos);
33188 +       lockdep_on();
33189 +       if (err >= 0)
33190 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
33191 +       return err;
33192 +}
33193 +
33194 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
33195 +                    loff_t *ppos)
33196 +{
33197 +       ssize_t err;
33198 +
33199 +       lockdep_off();
33200 +       err = kernel_read(file, kbuf, count, ppos);
33201 +       lockdep_on();
33202 +       AuTraceErr(err);
33203 +       if (err >= 0)
33204 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
33205 +       return err;
33206 +}
33207 +
33208 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
33209 +                     loff_t *ppos)
33210 +{
33211 +       ssize_t err;
33212 +
33213 +       lockdep_off();
33214 +       err = vfs_write(file, ubuf, count, ppos);
33215 +       lockdep_on();
33216 +       if (err >= 0)
33217 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
33218 +       return err;
33219 +}
33220 +
33221 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count, loff_t *ppos)
33222 +{
33223 +       ssize_t err;
33224 +
33225 +       lockdep_off();
33226 +       err = kernel_write(file, kbuf, count, ppos);
33227 +       lockdep_on();
33228 +       if (err >= 0)
33229 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
33230 +       return err;
33231 +}
33232 +
33233 +int vfsub_flush(struct file *file, fl_owner_t id)
33234 +{
33235 +       int err;
33236 +
33237 +       err = 0;
33238 +       if (file->f_op->flush) {
33239 +               if (!au_test_nfs(file->f_path.dentry->d_sb))
33240 +                       err = file->f_op->flush(file, id);
33241 +               else {
33242 +                       lockdep_off();
33243 +                       err = file->f_op->flush(file, id);
33244 +                       lockdep_on();
33245 +               }
33246 +               if (!err)
33247 +                       vfsub_update_h_iattr(&file->f_path, /*did*/NULL);
33248 +               /*ignore*/
33249 +       }
33250 +       return err;
33251 +}
33252 +
33253 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx)
33254 +{
33255 +       int err;
33256 +
33257 +       AuDbg("%pD, ctx{%ps, %llu}\n", file, ctx->actor, ctx->pos);
33258 +
33259 +       lockdep_off();
33260 +       err = iterate_dir(file, ctx);
33261 +       lockdep_on();
33262 +       if (err >= 0)
33263 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
33264 +
33265 +       return err;
33266 +}
33267 +
33268 +long vfsub_splice_to(struct file *in, loff_t *ppos,
33269 +                    struct pipe_inode_info *pipe, size_t len,
33270 +                    unsigned int flags)
33271 +{
33272 +       long err;
33273 +
33274 +       lockdep_off();
33275 +       err = do_splice_to(in, ppos, pipe, len, flags);
33276 +       lockdep_on();
33277 +       file_accessed(in);
33278 +       if (err >= 0)
33279 +               vfsub_update_h_iattr(&in->f_path, /*did*/NULL); /*ignore*/
33280 +       return err;
33281 +}
33282 +
33283 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
33284 +                      loff_t *ppos, size_t len, unsigned int flags)
33285 +{
33286 +       long err;
33287 +
33288 +       lockdep_off();
33289 +       err = do_splice_from(pipe, out, ppos, len, flags);
33290 +       lockdep_on();
33291 +       if (err >= 0)
33292 +               vfsub_update_h_iattr(&out->f_path, /*did*/NULL); /*ignore*/
33293 +       return err;
33294 +}
33295 +
33296 +int vfsub_fsync(struct file *file, struct path *path, int datasync)
33297 +{
33298 +       int err;
33299 +
33300 +       /* file can be NULL */
33301 +       lockdep_off();
33302 +       err = vfs_fsync(file, datasync);
33303 +       lockdep_on();
33304 +       if (!err) {
33305 +               if (!path) {
33306 +                       AuDebugOn(!file);
33307 +                       path = &file->f_path;
33308 +               }
33309 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
33310 +       }
33311 +       return err;
33312 +}
33313 +
33314 +/* cf. open.c:do_sys_truncate() and do_sys_ftruncate() */
33315 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
33316 +               struct file *h_file)
33317 +{
33318 +       int err;
33319 +       struct inode *h_inode;
33320 +       struct super_block *h_sb;
33321 +       struct user_namespace *h_userns;
33322 +
33323 +       if (!h_file) {
33324 +               err = vfsub_truncate(h_path, length);
33325 +               goto out;
33326 +       }
33327 +
33328 +       h_inode = d_inode(h_path->dentry);
33329 +       h_sb = h_inode->i_sb;
33330 +       lockdep_off();
33331 +       sb_start_write(h_sb);
33332 +       lockdep_on();
33333 +       err = security_path_truncate(h_path);
33334 +       if (!err) {
33335 +               h_userns = mnt_user_ns(h_path->mnt);
33336 +               lockdep_off();
33337 +               err = do_truncate(h_userns, h_path->dentry, length, attr,
33338 +                                 h_file);
33339 +               lockdep_on();
33340 +       }
33341 +       lockdep_off();
33342 +       sb_end_write(h_sb);
33343 +       lockdep_on();
33344 +
33345 +out:
33346 +       return err;
33347 +}
33348 +
33349 +/* ---------------------------------------------------------------------- */
33350 +
33351 +struct au_vfsub_mkdir_args {
33352 +       int *errp;
33353 +       struct inode *dir;
33354 +       struct path *path;
33355 +       int mode;
33356 +};
33357 +
33358 +static void au_call_vfsub_mkdir(void *args)
33359 +{
33360 +       struct au_vfsub_mkdir_args *a = args;
33361 +       *a->errp = vfsub_mkdir(a->dir, a->path, a->mode);
33362 +}
33363 +
33364 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode)
33365 +{
33366 +       int err, do_sio, wkq_err;
33367 +       struct user_namespace *userns;
33368 +
33369 +       userns = mnt_user_ns(path->mnt);
33370 +       do_sio = au_test_h_perm_sio(userns, dir, MAY_EXEC | MAY_WRITE);
33371 +       if (!do_sio) {
33372 +               lockdep_off();
33373 +               err = vfsub_mkdir(dir, path, mode);
33374 +               lockdep_on();
33375 +       } else {
33376 +               struct au_vfsub_mkdir_args args = {
33377 +                       .errp   = &err,
33378 +                       .dir    = dir,
33379 +                       .path   = path,
33380 +                       .mode   = mode
33381 +               };
33382 +               wkq_err = au_wkq_wait(au_call_vfsub_mkdir, &args);
33383 +               if (unlikely(wkq_err))
33384 +                       err = wkq_err;
33385 +       }
33386 +
33387 +       return err;
33388 +}
33389 +
33390 +struct au_vfsub_rmdir_args {
33391 +       int *errp;
33392 +       struct inode *dir;
33393 +       struct path *path;
33394 +};
33395 +
33396 +static void au_call_vfsub_rmdir(void *args)
33397 +{
33398 +       struct au_vfsub_rmdir_args *a = args;
33399 +       *a->errp = vfsub_rmdir(a->dir, a->path);
33400 +}
33401 +
33402 +int vfsub_sio_rmdir(struct inode *dir, struct path *path)
33403 +{
33404 +       int err, do_sio, wkq_err;
33405 +       struct user_namespace *userns;
33406 +
33407 +       userns = mnt_user_ns(path->mnt);
33408 +       do_sio = au_test_h_perm_sio(userns, dir, MAY_EXEC | MAY_WRITE);
33409 +       if (!do_sio) {
33410 +               lockdep_off();
33411 +               err = vfsub_rmdir(dir, path);
33412 +               lockdep_on();
33413 +       } else {
33414 +               struct au_vfsub_rmdir_args args = {
33415 +                       .errp   = &err,
33416 +                       .dir    = dir,
33417 +                       .path   = path
33418 +               };
33419 +               wkq_err = au_wkq_wait(au_call_vfsub_rmdir, &args);
33420 +               if (unlikely(wkq_err))
33421 +                       err = wkq_err;
33422 +       }
33423 +
33424 +       return err;
33425 +}
33426 +
33427 +/* ---------------------------------------------------------------------- */
33428 +
33429 +struct notify_change_args {
33430 +       int *errp;
33431 +       struct path *path;
33432 +       struct iattr *ia;
33433 +       struct inode **delegated_inode;
33434 +};
33435 +
33436 +static void call_notify_change(void *args)
33437 +{
33438 +       struct notify_change_args *a = args;
33439 +       struct inode *h_inode;
33440 +       struct user_namespace *userns;
33441 +
33442 +       h_inode = d_inode(a->path->dentry);
33443 +       IMustLock(h_inode);
33444 +
33445 +       *a->errp = -EPERM;
33446 +       if (!IS_IMMUTABLE(h_inode) && !IS_APPEND(h_inode)) {
33447 +               userns = mnt_user_ns(a->path->mnt);
33448 +               lockdep_off();
33449 +               *a->errp = notify_change(userns, a->path->dentry, a->ia,
33450 +                                        a->delegated_inode);
33451 +               lockdep_on();
33452 +               if (!*a->errp)
33453 +                       vfsub_update_h_iattr(a->path, /*did*/NULL); /*ignore*/
33454 +       }
33455 +       AuTraceErr(*a->errp);
33456 +}
33457 +
33458 +int vfsub_notify_change(struct path *path, struct iattr *ia,
33459 +                       struct inode **delegated_inode)
33460 +{
33461 +       int err;
33462 +       struct notify_change_args args = {
33463 +               .errp                   = &err,
33464 +               .path                   = path,
33465 +               .ia                     = ia,
33466 +               .delegated_inode        = delegated_inode
33467 +       };
33468 +
33469 +       call_notify_change(&args);
33470 +
33471 +       return err;
33472 +}
33473 +
33474 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
33475 +                           struct inode **delegated_inode)
33476 +{
33477 +       int err, wkq_err;
33478 +       struct notify_change_args args = {
33479 +               .errp                   = &err,
33480 +               .path                   = path,
33481 +               .ia                     = ia,
33482 +               .delegated_inode        = delegated_inode
33483 +       };
33484 +
33485 +       wkq_err = au_wkq_wait(call_notify_change, &args);
33486 +       if (unlikely(wkq_err))
33487 +               err = wkq_err;
33488 +
33489 +       return err;
33490 +}
33491 +
33492 +/* ---------------------------------------------------------------------- */
33493 +
33494 +struct unlink_args {
33495 +       int *errp;
33496 +       struct inode *dir;
33497 +       struct path *path;
33498 +       struct inode **delegated_inode;
33499 +};
33500 +
33501 +static void call_unlink(void *args)
33502 +{
33503 +       struct unlink_args *a = args;
33504 +       struct dentry *d = a->path->dentry;
33505 +       struct inode *h_inode;
33506 +       struct user_namespace *userns;
33507 +       const int stop_sillyrename = (au_test_nfs(d->d_sb)
33508 +                                     && au_dcount(d) == 1);
33509 +
33510 +       IMustLock(a->dir);
33511 +
33512 +       a->path->dentry = d->d_parent;
33513 +       *a->errp = security_path_unlink(a->path, d);
33514 +       a->path->dentry = d;
33515 +       if (unlikely(*a->errp))
33516 +               return;
33517 +
33518 +       if (!stop_sillyrename)
33519 +               dget(d);
33520 +       h_inode = NULL;
33521 +       if (d_is_positive(d)) {
33522 +               h_inode = d_inode(d);
33523 +               ihold(h_inode);
33524 +       }
33525 +
33526 +       userns = mnt_user_ns(a->path->mnt);
33527 +       lockdep_off();
33528 +       *a->errp = vfs_unlink(userns, a->dir, d, a->delegated_inode);
33529 +       lockdep_on();
33530 +       if (!*a->errp) {
33531 +               struct path tmp = {
33532 +                       .dentry = d->d_parent,
33533 +                       .mnt    = a->path->mnt
33534 +               };
33535 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
33536 +       }
33537 +
33538 +       if (!stop_sillyrename)
33539 +               dput(d);
33540 +       if (h_inode)
33541 +               iput(h_inode);
33542 +
33543 +       AuTraceErr(*a->errp);
33544 +}
33545 +
33546 +/*
33547 + * @dir: must be locked.
33548 + * @dentry: target dentry.
33549 + */
33550 +int vfsub_unlink(struct inode *dir, struct path *path,
33551 +                struct inode **delegated_inode, int force)
33552 +{
33553 +       int err;
33554 +       struct unlink_args args = {
33555 +               .errp                   = &err,
33556 +               .dir                    = dir,
33557 +               .path                   = path,
33558 +               .delegated_inode        = delegated_inode
33559 +       };
33560 +
33561 +       if (!force)
33562 +               call_unlink(&args);
33563 +       else {
33564 +               int wkq_err;
33565 +
33566 +               wkq_err = au_wkq_wait(call_unlink, &args);
33567 +               if (unlikely(wkq_err))
33568 +                       err = wkq_err;
33569 +       }
33570 +
33571 +       return err;
33572 +}
33573 diff -urN /usr/share/empty/fs/aufs/vfsub.h linux/fs/aufs/vfsub.h
33574 --- /usr/share/empty/fs/aufs/vfsub.h    1970-01-01 01:00:00.000000000 +0100
33575 +++ linux/fs/aufs/vfsub.h       2021-12-03 15:40:58.236647297 +0100
33576 @@ -0,0 +1,358 @@
33577 +/* SPDX-License-Identifier: GPL-2.0 */
33578 +/*
33579 + * Copyright (C) 2005-2021 Junjiro R. Okajima
33580 + *
33581 + * This program, aufs is free software; you can redistribute it and/or modify
33582 + * it under the terms of the GNU General Public License as published by
33583 + * the Free Software Foundation; either version 2 of the License, or
33584 + * (at your option) any later version.
33585 + *
33586 + * This program is distributed in the hope that it will be useful,
33587 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33588 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33589 + * GNU General Public License for more details.
33590 + *
33591 + * You should have received a copy of the GNU General Public License
33592 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33593 + */
33594 +
33595 +/*
33596 + * sub-routines for VFS
33597 + */
33598 +
33599 +#ifndef __AUFS_VFSUB_H__
33600 +#define __AUFS_VFSUB_H__
33601 +
33602 +#ifdef __KERNEL__
33603 +
33604 +#include <linux/fs.h>
33605 +#include <linux/mount.h>
33606 +#include <linux/posix_acl.h>
33607 +#include <linux/xattr.h>
33608 +#include "debug.h"
33609 +
33610 +/* copied from linux/fs/internal.h */
33611 +/* todo: BAD approach!! */
33612 +extern void __mnt_drop_write(struct vfsmount *);
33613 +extern struct file *alloc_empty_file(int, const struct cred *);
33614 +
33615 +/* ---------------------------------------------------------------------- */
33616 +
33617 +/* lock subclass for lower inode */
33618 +/* default MAX_LOCKDEP_SUBCLASSES(8) is not enough */
33619 +/* reduce? gave up. */
33620 +enum {
33621 +       AuLsc_I_Begin = I_MUTEX_PARENT2, /* 5 */
33622 +       AuLsc_I_PARENT,         /* lower inode, parent first */
33623 +       AuLsc_I_PARENT2,        /* copyup dirs */
33624 +       AuLsc_I_PARENT3,        /* copyup wh */
33625 +       AuLsc_I_CHILD,
33626 +       AuLsc_I_CHILD2,
33627 +       AuLsc_I_End
33628 +};
33629 +
33630 +/* to debug easier, do not make them inlined functions */
33631 +#define MtxMustLock(mtx)       AuDebugOn(!mutex_is_locked(mtx))
33632 +#define IMustLock(i)           AuDebugOn(!inode_is_locked(i))
33633 +
33634 +/* ---------------------------------------------------------------------- */
33635 +
33636 +static inline void vfsub_drop_nlink(struct inode *inode)
33637 +{
33638 +       AuDebugOn(!inode->i_nlink);
33639 +       drop_nlink(inode);
33640 +}
33641 +
33642 +static inline void vfsub_dead_dir(struct inode *inode)
33643 +{
33644 +       AuDebugOn(!S_ISDIR(inode->i_mode));
33645 +       inode->i_flags |= S_DEAD;
33646 +       clear_nlink(inode);
33647 +}
33648 +
33649 +static inline int vfsub_native_ro(struct inode *inode)
33650 +{
33651 +       return sb_rdonly(inode->i_sb)
33652 +               || IS_RDONLY(inode)
33653 +               /* || IS_APPEND(inode) */
33654 +               || IS_IMMUTABLE(inode);
33655 +}
33656 +
33657 +#ifdef CONFIG_AUFS_BR_FUSE
33658 +int vfsub_test_mntns(struct vfsmount *mnt, struct super_block *h_sb);
33659 +#else
33660 +AuStubInt0(vfsub_test_mntns, struct vfsmount *mnt, struct super_block *h_sb);
33661 +#endif
33662 +
33663 +int vfsub_sync_filesystem(struct super_block *h_sb, int wait);
33664 +
33665 +/* ---------------------------------------------------------------------- */
33666 +
33667 +int vfsub_update_h_iattr(struct path *h_path, int *did);
33668 +struct file *vfsub_dentry_open(struct path *path, int flags);
33669 +struct file *vfsub_filp_open(const char *path, int oflags, int mode);
33670 +struct au_branch;
33671 +struct vfsub_aopen_args {
33672 +       struct file             *file;
33673 +       unsigned int            open_flag;
33674 +       umode_t                 create_mode;
33675 +       struct au_branch        *br;
33676 +};
33677 +int vfsub_atomic_open(struct inode *dir, struct dentry *dentry,
33678 +                     struct vfsub_aopen_args *args);
33679 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path);
33680 +
33681 +struct dentry *vfsub_lookup_one_len_unlocked(const char *name,
33682 +                                            struct path *ppath, int len);
33683 +struct dentry *vfsub_lookup_one_len(const char *name, struct path *ppath,
33684 +                                   int len);
33685 +
33686 +struct vfsub_lkup_one_args {
33687 +       struct dentry **errp;
33688 +       struct qstr *name;
33689 +       struct path *ppath;
33690 +};
33691 +
33692 +static inline struct dentry *vfsub_lkup_one(struct qstr *name,
33693 +                                           struct path *ppath)
33694 +{
33695 +       return vfsub_lookup_one_len(name->name, ppath, name->len);
33696 +}
33697 +
33698 +void vfsub_call_lkup_one(void *args);
33699 +
33700 +/* ---------------------------------------------------------------------- */
33701 +
33702 +static inline int vfsub_mnt_want_write(struct vfsmount *mnt)
33703 +{
33704 +       int err;
33705 +
33706 +       lockdep_off();
33707 +       err = mnt_want_write(mnt);
33708 +       lockdep_on();
33709 +       return err;
33710 +}
33711 +
33712 +static inline void vfsub_mnt_drop_write(struct vfsmount *mnt)
33713 +{
33714 +       lockdep_off();
33715 +       mnt_drop_write(mnt);
33716 +       lockdep_on();
33717 +}
33718 +
33719 +#if 0 /* reserved */
33720 +static inline void vfsub_mnt_drop_write_file(struct file *file)
33721 +{
33722 +       lockdep_off();
33723 +       mnt_drop_write_file(file);
33724 +       lockdep_on();
33725 +}
33726 +#endif
33727 +
33728 +/* ---------------------------------------------------------------------- */
33729 +
33730 +struct au_hinode;
33731 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
33732 +                                struct dentry *d2, struct au_hinode *hdir2);
33733 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
33734 +                        struct dentry *d2, struct au_hinode *hdir2);
33735 +
33736 +int vfsub_create(struct inode *dir, struct path *path, int mode,
33737 +                bool want_excl);
33738 +int vfsub_symlink(struct inode *dir, struct path *path,
33739 +                 const char *symname);
33740 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev);
33741 +int vfsub_link(struct dentry *src_dentry, struct inode *dir,
33742 +              struct path *path, struct inode **delegated_inode);
33743 +int vfsub_rename(struct inode *src_hdir, struct dentry *src_dentry,
33744 +                struct inode *hdir, struct path *path,
33745 +                struct inode **delegated_inode, unsigned int flags);
33746 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode);
33747 +int vfsub_rmdir(struct inode *dir, struct path *path);
33748 +
33749 +/* ---------------------------------------------------------------------- */
33750 +
33751 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
33752 +                    loff_t *ppos);
33753 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
33754 +                       loff_t *ppos);
33755 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
33756 +                     loff_t *ppos);
33757 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count,
33758 +                     loff_t *ppos);
33759 +int vfsub_flush(struct file *file, fl_owner_t id);
33760 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx);
33761 +
33762 +static inline loff_t vfsub_f_size_read(struct file *file)
33763 +{
33764 +       return i_size_read(file_inode(file));
33765 +}
33766 +
33767 +static inline unsigned int vfsub_file_flags(struct file *file)
33768 +{
33769 +       unsigned int flags;
33770 +
33771 +       spin_lock(&file->f_lock);
33772 +       flags = file->f_flags;
33773 +       spin_unlock(&file->f_lock);
33774 +
33775 +       return flags;
33776 +}
33777 +
33778 +static inline int vfsub_file_execed(struct file *file)
33779 +{
33780 +       /* todo: direct access f_flags */
33781 +       return !!(vfsub_file_flags(file) & __FMODE_EXEC);
33782 +}
33783 +
33784 +#if 0 /* reserved */
33785 +static inline void vfsub_file_accessed(struct file *h_file)
33786 +{
33787 +       file_accessed(h_file);
33788 +       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL); /*ignore*/
33789 +}
33790 +#endif
33791 +
33792 +#if 0 /* reserved */
33793 +static inline void vfsub_touch_atime(struct vfsmount *h_mnt,
33794 +                                    struct dentry *h_dentry)
33795 +{
33796 +       struct path h_path = {
33797 +               .dentry = h_dentry,
33798 +               .mnt    = h_mnt
33799 +       };
33800 +       touch_atime(&h_path);
33801 +       vfsub_update_h_iattr(&h_path, /*did*/NULL); /*ignore*/
33802 +}
33803 +#endif
33804 +
33805 +static inline int vfsub_update_time(struct inode *h_inode,
33806 +                                   struct timespec64 *ts, int flags)
33807 +{
33808 +       return inode_update_time(h_inode, ts, flags);
33809 +       /* no vfsub_update_h_iattr() since we don't have struct path */
33810 +}
33811 +
33812 +#ifdef CONFIG_FS_POSIX_ACL
33813 +static inline int vfsub_acl_chmod(struct user_namespace *h_userns,
33814 +                                 struct inode *h_inode, umode_t h_mode)
33815 +{
33816 +       int err;
33817 +
33818 +       err = posix_acl_chmod(h_userns, h_inode, h_mode);
33819 +       if (err == -EOPNOTSUPP)
33820 +               err = 0;
33821 +       return err;
33822 +}
33823 +#else
33824 +AuStubInt0(vfsub_acl_chmod, struct user_namespace *h_userns,
33825 +          struct inode *h_inode, umode_t h_mode);
33826 +#endif
33827 +
33828 +long vfsub_splice_to(struct file *in, loff_t *ppos,
33829 +                    struct pipe_inode_info *pipe, size_t len,
33830 +                    unsigned int flags);
33831 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
33832 +                      loff_t *ppos, size_t len, unsigned int flags);
33833 +
33834 +static inline long vfsub_truncate(struct path *path, loff_t length)
33835 +{
33836 +       long err;
33837 +
33838 +       lockdep_off();
33839 +       err = vfs_truncate(path, length);
33840 +       lockdep_on();
33841 +       return err;
33842 +}
33843 +
33844 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
33845 +               struct file *h_file);
33846 +int vfsub_fsync(struct file *file, struct path *path, int datasync);
33847 +
33848 +/*
33849 + * re-use branch fs's ioctl(FICLONE) while aufs itself doesn't support such
33850 + * ioctl.
33851 + */
33852 +static inline loff_t vfsub_clone_file_range(struct file *src, struct file *dst,
33853 +                                           loff_t len)
33854 +{
33855 +       loff_t err;
33856 +
33857 +       lockdep_off();
33858 +       err = vfs_clone_file_range(src, 0, dst, 0, len, /*remap_flags*/0);
33859 +       lockdep_on();
33860 +
33861 +       return err;
33862 +}
33863 +
33864 +/* copy_file_range(2) is a systemcall */
33865 +static inline ssize_t vfsub_copy_file_range(struct file *src, loff_t src_pos,
33866 +                                           struct file *dst, loff_t dst_pos,
33867 +                                           size_t len, unsigned int flags)
33868 +{
33869 +       ssize_t ssz;
33870 +
33871 +       lockdep_off();
33872 +       ssz = vfs_copy_file_range(src, src_pos, dst, dst_pos, len, flags);
33873 +       lockdep_on();
33874 +
33875 +       return ssz;
33876 +}
33877 +
33878 +/* ---------------------------------------------------------------------- */
33879 +
33880 +static inline loff_t vfsub_llseek(struct file *file, loff_t offset, int origin)
33881 +{
33882 +       loff_t err;
33883 +
33884 +       lockdep_off();
33885 +       err = vfs_llseek(file, offset, origin);
33886 +       lockdep_on();
33887 +       return err;
33888 +}
33889 +
33890 +/* ---------------------------------------------------------------------- */
33891 +
33892 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode);
33893 +int vfsub_sio_rmdir(struct inode *dir, struct path *path);
33894 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
33895 +                           struct inode **delegated_inode);
33896 +int vfsub_notify_change(struct path *path, struct iattr *ia,
33897 +                       struct inode **delegated_inode);
33898 +int vfsub_unlink(struct inode *dir, struct path *path,
33899 +                struct inode **delegated_inode, int force);
33900 +
33901 +static inline int vfsub_getattr(const struct path *path, struct kstat *st)
33902 +{
33903 +       return vfs_getattr(path, st, STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
33904 +}
33905 +
33906 +/* ---------------------------------------------------------------------- */
33907 +
33908 +static inline int vfsub_setxattr(struct user_namespace *userns,
33909 +                                struct dentry *dentry, const char *name,
33910 +                                const void *value, size_t size, int flags)
33911 +{
33912 +       int err;
33913 +
33914 +       lockdep_off();
33915 +       err = vfs_setxattr(userns, dentry, name, value, size, flags);
33916 +       lockdep_on();
33917 +
33918 +       return err;
33919 +}
33920 +
33921 +static inline int vfsub_removexattr(struct user_namespace *userns,
33922 +                                   struct dentry *dentry, const char *name)
33923 +{
33924 +       int err;
33925 +
33926 +       lockdep_off();
33927 +       err = vfs_removexattr(userns, dentry, name);
33928 +       lockdep_on();
33929 +
33930 +       return err;
33931 +}
33932 +
33933 +#endif /* __KERNEL__ */
33934 +#endif /* __AUFS_VFSUB_H__ */
33935 diff -urN /usr/share/empty/fs/aufs/wbr_policy.c linux/fs/aufs/wbr_policy.c
33936 --- /usr/share/empty/fs/aufs/wbr_policy.c       1970-01-01 01:00:00.000000000 +0100
33937 +++ linux/fs/aufs/wbr_policy.c  2021-12-03 15:38:59.939980643 +0100
33938 @@ -0,0 +1,830 @@
33939 +// SPDX-License-Identifier: GPL-2.0
33940 +/*
33941 + * Copyright (C) 2005-2021 Junjiro R. Okajima
33942 + *
33943 + * This program, aufs is free software; you can redistribute it and/or modify
33944 + * it under the terms of the GNU General Public License as published by
33945 + * the Free Software Foundation; either version 2 of the License, or
33946 + * (at your option) any later version.
33947 + *
33948 + * This program is distributed in the hope that it will be useful,
33949 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33950 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33951 + * GNU General Public License for more details.
33952 + *
33953 + * You should have received a copy of the GNU General Public License
33954 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33955 + */
33956 +
33957 +/*
33958 + * policies for selecting one among multiple writable branches
33959 + */
33960 +
33961 +#include <linux/statfs.h>
33962 +#include "aufs.h"
33963 +
33964 +/* subset of cpup_attr() */
33965 +static noinline_for_stack
33966 +int au_cpdown_attr(struct path *h_path, struct dentry *h_src)
33967 +{
33968 +       int err, sbits;
33969 +       struct iattr ia;
33970 +       struct inode *h_isrc;
33971 +
33972 +       h_isrc = d_inode(h_src);
33973 +       ia.ia_valid = ATTR_FORCE | ATTR_MODE | ATTR_UID | ATTR_GID;
33974 +       ia.ia_mode = h_isrc->i_mode;
33975 +       ia.ia_uid = h_isrc->i_uid;
33976 +       ia.ia_gid = h_isrc->i_gid;
33977 +       sbits = !!(ia.ia_mode & (S_ISUID | S_ISGID));
33978 +       au_cpup_attr_flags(d_inode(h_path->dentry), h_isrc->i_flags);
33979 +       /* no delegation since it is just created */
33980 +       err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
33981 +
33982 +       /* is this nfs only? */
33983 +       if (!err && sbits && au_test_nfs(h_path->dentry->d_sb)) {
33984 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
33985 +               ia.ia_mode = h_isrc->i_mode;
33986 +               err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
33987 +       }
33988 +
33989 +       return err;
33990 +}
33991 +
33992 +#define AuCpdown_PARENT_OPQ    1
33993 +#define AuCpdown_WHED          (1 << 1)
33994 +#define AuCpdown_MADE_DIR      (1 << 2)
33995 +#define AuCpdown_DIROPQ                (1 << 3)
33996 +#define au_ftest_cpdown(flags, name)   ((flags) & AuCpdown_##name)
33997 +#define au_fset_cpdown(flags, name) \
33998 +       do { (flags) |= AuCpdown_##name; } while (0)
33999 +#define au_fclr_cpdown(flags, name) \
34000 +       do { (flags) &= ~AuCpdown_##name; } while (0)
34001 +
34002 +static int au_cpdown_dir_opq(struct dentry *dentry, aufs_bindex_t bdst,
34003 +                            unsigned int *flags)
34004 +{
34005 +       int err;
34006 +       struct dentry *opq_dentry;
34007 +
34008 +       opq_dentry = au_diropq_create(dentry, bdst);
34009 +       err = PTR_ERR(opq_dentry);
34010 +       if (IS_ERR(opq_dentry))
34011 +               goto out;
34012 +       dput(opq_dentry);
34013 +       au_fset_cpdown(*flags, DIROPQ);
34014 +
34015 +out:
34016 +       return err;
34017 +}
34018 +
34019 +static int au_cpdown_dir_wh(struct dentry *dentry, struct dentry *h_parent,
34020 +                           struct inode *dir, aufs_bindex_t bdst)
34021 +{
34022 +       int err;
34023 +       struct path h_path;
34024 +       struct au_branch *br;
34025 +
34026 +       br = au_sbr(dentry->d_sb, bdst);
34027 +       h_path.dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
34028 +       err = PTR_ERR(h_path.dentry);
34029 +       if (IS_ERR(h_path.dentry))
34030 +               goto out;
34031 +
34032 +       err = 0;
34033 +       if (d_is_positive(h_path.dentry)) {
34034 +               h_path.mnt = au_br_mnt(br);
34035 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bdst), &h_path,
34036 +                                         dentry);
34037 +       }
34038 +       dput(h_path.dentry);
34039 +
34040 +out:
34041 +       return err;
34042 +}
34043 +
34044 +static int au_cpdown_dir(struct dentry *dentry, aufs_bindex_t bdst,
34045 +                        struct au_pin *pin,
34046 +                        struct dentry *h_parent, void *arg)
34047 +{
34048 +       int err, rerr;
34049 +       aufs_bindex_t bopq, btop;
34050 +       struct path h_path;
34051 +       struct dentry *parent;
34052 +       struct inode *h_dir, *h_inode, *inode, *dir;
34053 +       unsigned int *flags = arg;
34054 +
34055 +       btop = au_dbtop(dentry);
34056 +       /* dentry is di-locked */
34057 +       parent = dget_parent(dentry);
34058 +       dir = d_inode(parent);
34059 +       h_dir = d_inode(h_parent);
34060 +       AuDebugOn(h_dir != au_h_iptr(dir, bdst));
34061 +       IMustLock(h_dir);
34062 +
34063 +       err = au_lkup_neg(dentry, bdst, /*wh*/0);
34064 +       if (unlikely(err < 0))
34065 +               goto out;
34066 +       h_path.dentry = au_h_dptr(dentry, bdst);
34067 +       h_path.mnt = au_sbr_mnt(dentry->d_sb, bdst);
34068 +       err = vfsub_sio_mkdir(au_h_iptr(dir, bdst), &h_path, 0755);
34069 +       if (unlikely(err))
34070 +               goto out_put;
34071 +       au_fset_cpdown(*flags, MADE_DIR);
34072 +
34073 +       bopq = au_dbdiropq(dentry);
34074 +       au_fclr_cpdown(*flags, WHED);
34075 +       au_fclr_cpdown(*flags, DIROPQ);
34076 +       if (au_dbwh(dentry) == bdst)
34077 +               au_fset_cpdown(*flags, WHED);
34078 +       if (!au_ftest_cpdown(*flags, PARENT_OPQ) && bopq <= bdst)
34079 +               au_fset_cpdown(*flags, PARENT_OPQ);
34080 +       h_inode = d_inode(h_path.dentry);
34081 +       inode_lock_nested(h_inode, AuLsc_I_CHILD);
34082 +       if (au_ftest_cpdown(*flags, WHED)) {
34083 +               err = au_cpdown_dir_opq(dentry, bdst, flags);
34084 +               if (unlikely(err)) {
34085 +                       inode_unlock(h_inode);
34086 +                       goto out_dir;
34087 +               }
34088 +       }
34089 +
34090 +       err = au_cpdown_attr(&h_path, au_h_dptr(dentry, btop));
34091 +       inode_unlock(h_inode);
34092 +       if (unlikely(err))
34093 +               goto out_opq;
34094 +
34095 +       if (au_ftest_cpdown(*flags, WHED)) {
34096 +               err = au_cpdown_dir_wh(dentry, h_parent, dir, bdst);
34097 +               if (unlikely(err))
34098 +                       goto out_opq;
34099 +       }
34100 +
34101 +       inode = d_inode(dentry);
34102 +       if (au_ibbot(inode) < bdst)
34103 +               au_set_ibbot(inode, bdst);
34104 +       au_set_h_iptr(inode, bdst, au_igrab(h_inode),
34105 +                     au_hi_flags(inode, /*isdir*/1));
34106 +       au_fhsm_wrote(dentry->d_sb, bdst, /*force*/0);
34107 +       goto out; /* success */
34108 +
34109 +       /* revert */
34110 +out_opq:
34111 +       if (au_ftest_cpdown(*flags, DIROPQ)) {
34112 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
34113 +               rerr = au_diropq_remove(dentry, bdst);
34114 +               inode_unlock(h_inode);
34115 +               if (unlikely(rerr)) {
34116 +                       AuIOErr("failed removing diropq for %pd b%d (%d)\n",
34117 +                               dentry, bdst, rerr);
34118 +                       err = -EIO;
34119 +                       goto out;
34120 +               }
34121 +       }
34122 +out_dir:
34123 +       if (au_ftest_cpdown(*flags, MADE_DIR)) {
34124 +               rerr = vfsub_sio_rmdir(au_h_iptr(dir, bdst), &h_path);
34125 +               if (unlikely(rerr)) {
34126 +                       AuIOErr("failed removing %pd b%d (%d)\n",
34127 +                               dentry, bdst, rerr);
34128 +                       err = -EIO;
34129 +               }
34130 +       }
34131 +out_put:
34132 +       au_set_h_dptr(dentry, bdst, NULL);
34133 +       if (au_dbbot(dentry) == bdst)
34134 +               au_update_dbbot(dentry);
34135 +out:
34136 +       dput(parent);
34137 +       return err;
34138 +}
34139 +
34140 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst)
34141 +{
34142 +       int err;
34143 +       unsigned int flags;
34144 +
34145 +       flags = 0;
34146 +       err = au_cp_dirs(dentry, bdst, au_cpdown_dir, &flags);
34147 +
34148 +       return err;
34149 +}
34150 +
34151 +/* ---------------------------------------------------------------------- */
34152 +
34153 +/* policies for create */
34154 +
34155 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex)
34156 +{
34157 +       int err, i, j, ndentry;
34158 +       aufs_bindex_t bopq;
34159 +       struct au_dcsub_pages dpages;
34160 +       struct au_dpage *dpage;
34161 +       struct dentry **dentries, *parent, *d;
34162 +
34163 +       err = au_dpages_init(&dpages, GFP_NOFS);
34164 +       if (unlikely(err))
34165 +               goto out;
34166 +       parent = dget_parent(dentry);
34167 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/0);
34168 +       if (unlikely(err))
34169 +               goto out_free;
34170 +
34171 +       err = bindex;
34172 +       for (i = 0; i < dpages.ndpage; i++) {
34173 +               dpage = dpages.dpages + i;
34174 +               dentries = dpage->dentries;
34175 +               ndentry = dpage->ndentry;
34176 +               for (j = 0; j < ndentry; j++) {
34177 +                       d = dentries[j];
34178 +                       di_read_lock_parent2(d, !AuLock_IR);
34179 +                       bopq = au_dbdiropq(d);
34180 +                       di_read_unlock(d, !AuLock_IR);
34181 +                       if (bopq >= 0 && bopq < err)
34182 +                               err = bopq;
34183 +               }
34184 +       }
34185 +
34186 +out_free:
34187 +       dput(parent);
34188 +       au_dpages_free(&dpages);
34189 +out:
34190 +       return err;
34191 +}
34192 +
34193 +static int au_wbr_bu(struct super_block *sb, aufs_bindex_t bindex)
34194 +{
34195 +       for (; bindex >= 0; bindex--)
34196 +               if (!au_br_rdonly(au_sbr(sb, bindex)))
34197 +                       return bindex;
34198 +       return -EROFS;
34199 +}
34200 +
34201 +/* top down parent */
34202 +static int au_wbr_create_tdp(struct dentry *dentry,
34203 +                            unsigned int flags __maybe_unused)
34204 +{
34205 +       int err;
34206 +       aufs_bindex_t btop, bindex;
34207 +       struct super_block *sb;
34208 +       struct dentry *parent, *h_parent;
34209 +
34210 +       sb = dentry->d_sb;
34211 +       btop = au_dbtop(dentry);
34212 +       err = btop;
34213 +       if (!au_br_rdonly(au_sbr(sb, btop)))
34214 +               goto out;
34215 +
34216 +       err = -EROFS;
34217 +       parent = dget_parent(dentry);
34218 +       for (bindex = au_dbtop(parent); bindex < btop; bindex++) {
34219 +               h_parent = au_h_dptr(parent, bindex);
34220 +               if (!h_parent || d_is_negative(h_parent))
34221 +                       continue;
34222 +
34223 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
34224 +                       err = bindex;
34225 +                       break;
34226 +               }
34227 +       }
34228 +       dput(parent);
34229 +
34230 +       /* bottom up here */
34231 +       if (unlikely(err < 0)) {
34232 +               err = au_wbr_bu(sb, btop - 1);
34233 +               if (err >= 0)
34234 +                       err = au_wbr_nonopq(dentry, err);
34235 +       }
34236 +
34237 +out:
34238 +       AuDbg("b%d\n", err);
34239 +       return err;
34240 +}
34241 +
34242 +/* ---------------------------------------------------------------------- */
34243 +
34244 +/* an exception for the policy other than tdp */
34245 +static int au_wbr_create_exp(struct dentry *dentry)
34246 +{
34247 +       int err;
34248 +       aufs_bindex_t bwh, bdiropq;
34249 +       struct dentry *parent;
34250 +
34251 +       err = -1;
34252 +       bwh = au_dbwh(dentry);
34253 +       parent = dget_parent(dentry);
34254 +       bdiropq = au_dbdiropq(parent);
34255 +       if (bwh >= 0) {
34256 +               if (bdiropq >= 0)
34257 +                       err = min(bdiropq, bwh);
34258 +               else
34259 +                       err = bwh;
34260 +               AuDbg("%d\n", err);
34261 +       } else if (bdiropq >= 0) {
34262 +               err = bdiropq;
34263 +               AuDbg("%d\n", err);
34264 +       }
34265 +       dput(parent);
34266 +
34267 +       if (err >= 0)
34268 +               err = au_wbr_nonopq(dentry, err);
34269 +
34270 +       if (err >= 0 && au_br_rdonly(au_sbr(dentry->d_sb, err)))
34271 +               err = -1;
34272 +
34273 +       AuDbg("%d\n", err);
34274 +       return err;
34275 +}
34276 +
34277 +/* ---------------------------------------------------------------------- */
34278 +
34279 +/* round robin */
34280 +static int au_wbr_create_init_rr(struct super_block *sb)
34281 +{
34282 +       int err;
34283 +
34284 +       err = au_wbr_bu(sb, au_sbbot(sb));
34285 +       atomic_set(&au_sbi(sb)->si_wbr_rr_next, -err); /* less important */
34286 +       /* smp_mb(); */
34287 +
34288 +       AuDbg("b%d\n", err);
34289 +       return err;
34290 +}
34291 +
34292 +static int au_wbr_create_rr(struct dentry *dentry, unsigned int flags)
34293 +{
34294 +       int err, nbr;
34295 +       unsigned int u;
34296 +       aufs_bindex_t bindex, bbot;
34297 +       struct super_block *sb;
34298 +       atomic_t *next;
34299 +
34300 +       err = au_wbr_create_exp(dentry);
34301 +       if (err >= 0)
34302 +               goto out;
34303 +
34304 +       sb = dentry->d_sb;
34305 +       next = &au_sbi(sb)->si_wbr_rr_next;
34306 +       bbot = au_sbbot(sb);
34307 +       nbr = bbot + 1;
34308 +       for (bindex = 0; bindex <= bbot; bindex++) {
34309 +               if (!au_ftest_wbr(flags, DIR)) {
34310 +                       err = atomic_dec_return(next) + 1;
34311 +                       /* modulo for 0 is meaningless */
34312 +                       if (unlikely(!err))
34313 +                               err = atomic_dec_return(next) + 1;
34314 +               } else
34315 +                       err = atomic_read(next);
34316 +               AuDbg("%d\n", err);
34317 +               u = err;
34318 +               err = u % nbr;
34319 +               AuDbg("%d\n", err);
34320 +               if (!au_br_rdonly(au_sbr(sb, err)))
34321 +                       break;
34322 +               err = -EROFS;
34323 +       }
34324 +
34325 +       if (err >= 0)
34326 +               err = au_wbr_nonopq(dentry, err);
34327 +
34328 +out:
34329 +       AuDbg("%d\n", err);
34330 +       return err;
34331 +}
34332 +
34333 +/* ---------------------------------------------------------------------- */
34334 +
34335 +/* most free space */
34336 +static void au_mfs(struct dentry *dentry, struct dentry *parent)
34337 +{
34338 +       struct super_block *sb;
34339 +       struct au_branch *br;
34340 +       struct au_wbr_mfs *mfs;
34341 +       struct dentry *h_parent;
34342 +       aufs_bindex_t bindex, bbot;
34343 +       int err;
34344 +       unsigned long long b, bavail;
34345 +       struct path h_path;
34346 +       /* reduce the stack usage */
34347 +       struct kstatfs *st;
34348 +
34349 +       st = kmalloc(sizeof(*st), GFP_NOFS);
34350 +       if (unlikely(!st)) {
34351 +               AuWarn1("failed updating mfs(%d), ignored\n", -ENOMEM);
34352 +               return;
34353 +       }
34354 +
34355 +       bavail = 0;
34356 +       sb = dentry->d_sb;
34357 +       mfs = &au_sbi(sb)->si_wbr_mfs;
34358 +       MtxMustLock(&mfs->mfs_lock);
34359 +       mfs->mfs_bindex = -EROFS;
34360 +       mfs->mfsrr_bytes = 0;
34361 +       if (!parent) {
34362 +               bindex = 0;
34363 +               bbot = au_sbbot(sb);
34364 +       } else {
34365 +               bindex = au_dbtop(parent);
34366 +               bbot = au_dbtaildir(parent);
34367 +       }
34368 +
34369 +       for (; bindex <= bbot; bindex++) {
34370 +               if (parent) {
34371 +                       h_parent = au_h_dptr(parent, bindex);
34372 +                       if (!h_parent || d_is_negative(h_parent))
34373 +                               continue;
34374 +               }
34375 +               br = au_sbr(sb, bindex);
34376 +               if (au_br_rdonly(br))
34377 +                       continue;
34378 +
34379 +               /* sb->s_root for NFS is unreliable */
34380 +               h_path.mnt = au_br_mnt(br);
34381 +               h_path.dentry = h_path.mnt->mnt_root;
34382 +               err = vfs_statfs(&h_path, st);
34383 +               if (unlikely(err)) {
34384 +                       AuWarn1("failed statfs, b%d, %d\n", bindex, err);
34385 +                       continue;
34386 +               }
34387 +
34388 +               /* when the available size is equal, select the lower one */
34389 +               BUILD_BUG_ON(sizeof(b) < sizeof(st->f_bavail)
34390 +                            || sizeof(b) < sizeof(st->f_bsize));
34391 +               b = st->f_bavail * st->f_bsize;
34392 +               br->br_wbr->wbr_bytes = b;
34393 +               if (b >= bavail) {
34394 +                       bavail = b;
34395 +                       mfs->mfs_bindex = bindex;
34396 +                       mfs->mfs_jiffy = jiffies;
34397 +               }
34398 +       }
34399 +
34400 +       mfs->mfsrr_bytes = bavail;
34401 +       AuDbg("b%d\n", mfs->mfs_bindex);
34402 +       au_kfree_rcu(st);
34403 +}
34404 +
34405 +static int au_wbr_create_mfs(struct dentry *dentry, unsigned int flags)
34406 +{
34407 +       int err;
34408 +       struct dentry *parent;
34409 +       struct super_block *sb;
34410 +       struct au_wbr_mfs *mfs;
34411 +
34412 +       err = au_wbr_create_exp(dentry);
34413 +       if (err >= 0)
34414 +               goto out;
34415 +
34416 +       sb = dentry->d_sb;
34417 +       parent = NULL;
34418 +       if (au_ftest_wbr(flags, PARENT))
34419 +               parent = dget_parent(dentry);
34420 +       mfs = &au_sbi(sb)->si_wbr_mfs;
34421 +       mutex_lock(&mfs->mfs_lock);
34422 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
34423 +           || mfs->mfs_bindex < 0
34424 +           || au_br_rdonly(au_sbr(sb, mfs->mfs_bindex)))
34425 +               au_mfs(dentry, parent);
34426 +       mutex_unlock(&mfs->mfs_lock);
34427 +       err = mfs->mfs_bindex;
34428 +       dput(parent);
34429 +
34430 +       if (err >= 0)
34431 +               err = au_wbr_nonopq(dentry, err);
34432 +
34433 +out:
34434 +       AuDbg("b%d\n", err);
34435 +       return err;
34436 +}
34437 +
34438 +static int au_wbr_create_init_mfs(struct super_block *sb)
34439 +{
34440 +       struct au_wbr_mfs *mfs;
34441 +
34442 +       mfs = &au_sbi(sb)->si_wbr_mfs;
34443 +       mutex_init(&mfs->mfs_lock);
34444 +       mfs->mfs_jiffy = 0;
34445 +       mfs->mfs_bindex = -EROFS;
34446 +
34447 +       return 0;
34448 +}
34449 +
34450 +static int au_wbr_create_fin_mfs(struct super_block *sb __maybe_unused)
34451 +{
34452 +       mutex_destroy(&au_sbi(sb)->si_wbr_mfs.mfs_lock);
34453 +       return 0;
34454 +}
34455 +
34456 +/* ---------------------------------------------------------------------- */
34457 +
34458 +/* top down regardless parent, and then mfs */
34459 +static int au_wbr_create_tdmfs(struct dentry *dentry,
34460 +                              unsigned int flags __maybe_unused)
34461 +{
34462 +       int err;
34463 +       aufs_bindex_t bwh, btail, bindex, bfound, bmfs;
34464 +       unsigned long long watermark;
34465 +       struct super_block *sb;
34466 +       struct au_wbr_mfs *mfs;
34467 +       struct au_branch *br;
34468 +       struct dentry *parent;
34469 +
34470 +       sb = dentry->d_sb;
34471 +       mfs = &au_sbi(sb)->si_wbr_mfs;
34472 +       mutex_lock(&mfs->mfs_lock);
34473 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
34474 +           || mfs->mfs_bindex < 0)
34475 +               au_mfs(dentry, /*parent*/NULL);
34476 +       watermark = mfs->mfsrr_watermark;
34477 +       bmfs = mfs->mfs_bindex;
34478 +       mutex_unlock(&mfs->mfs_lock);
34479 +
34480 +       /* another style of au_wbr_create_exp() */
34481 +       bwh = au_dbwh(dentry);
34482 +       parent = dget_parent(dentry);
34483 +       btail = au_dbtaildir(parent);
34484 +       if (bwh >= 0 && bwh < btail)
34485 +               btail = bwh;
34486 +
34487 +       err = au_wbr_nonopq(dentry, btail);
34488 +       if (unlikely(err < 0))
34489 +               goto out;
34490 +       btail = err;
34491 +       bfound = -1;
34492 +       for (bindex = 0; bindex <= btail; bindex++) {
34493 +               br = au_sbr(sb, bindex);
34494 +               if (au_br_rdonly(br))
34495 +                       continue;
34496 +               if (br->br_wbr->wbr_bytes > watermark) {
34497 +                       bfound = bindex;
34498 +                       break;
34499 +               }
34500 +       }
34501 +       err = bfound;
34502 +       if (err < 0)
34503 +               err = bmfs;
34504 +
34505 +out:
34506 +       dput(parent);
34507 +       AuDbg("b%d\n", err);
34508 +       return err;
34509 +}
34510 +
34511 +/* ---------------------------------------------------------------------- */
34512 +
34513 +/* most free space and then round robin */
34514 +static int au_wbr_create_mfsrr(struct dentry *dentry, unsigned int flags)
34515 +{
34516 +       int err;
34517 +       struct au_wbr_mfs *mfs;
34518 +
34519 +       err = au_wbr_create_mfs(dentry, flags);
34520 +       if (err >= 0) {
34521 +               mfs = &au_sbi(dentry->d_sb)->si_wbr_mfs;
34522 +               mutex_lock(&mfs->mfs_lock);
34523 +               if (mfs->mfsrr_bytes < mfs->mfsrr_watermark)
34524 +                       err = au_wbr_create_rr(dentry, flags);
34525 +               mutex_unlock(&mfs->mfs_lock);
34526 +       }
34527 +
34528 +       AuDbg("b%d\n", err);
34529 +       return err;
34530 +}
34531 +
34532 +static int au_wbr_create_init_mfsrr(struct super_block *sb)
34533 +{
34534 +       int err;
34535 +
34536 +       au_wbr_create_init_mfs(sb); /* ignore */
34537 +       err = au_wbr_create_init_rr(sb);
34538 +
34539 +       return err;
34540 +}
34541 +
34542 +/* ---------------------------------------------------------------------- */
34543 +
34544 +/* top down parent and most free space */
34545 +static int au_wbr_create_pmfs(struct dentry *dentry, unsigned int flags)
34546 +{
34547 +       int err, e2;
34548 +       unsigned long long b;
34549 +       aufs_bindex_t bindex, btop, bbot;
34550 +       struct super_block *sb;
34551 +       struct dentry *parent, *h_parent;
34552 +       struct au_branch *br;
34553 +
34554 +       err = au_wbr_create_tdp(dentry, flags);
34555 +       if (unlikely(err < 0))
34556 +               goto out;
34557 +       parent = dget_parent(dentry);
34558 +       btop = au_dbtop(parent);
34559 +       bbot = au_dbtaildir(parent);
34560 +       if (btop == bbot)
34561 +               goto out_parent; /* success */
34562 +
34563 +       e2 = au_wbr_create_mfs(dentry, flags);
34564 +       if (e2 < 0)
34565 +               goto out_parent; /* success */
34566 +
34567 +       /* when the available size is equal, select upper one */
34568 +       sb = dentry->d_sb;
34569 +       br = au_sbr(sb, err);
34570 +       b = br->br_wbr->wbr_bytes;
34571 +       AuDbg("b%d, %llu\n", err, b);
34572 +
34573 +       for (bindex = btop; bindex <= bbot; bindex++) {
34574 +               h_parent = au_h_dptr(parent, bindex);
34575 +               if (!h_parent || d_is_negative(h_parent))
34576 +                       continue;
34577 +
34578 +               br = au_sbr(sb, bindex);
34579 +               if (!au_br_rdonly(br) && br->br_wbr->wbr_bytes > b) {
34580 +                       b = br->br_wbr->wbr_bytes;
34581 +                       err = bindex;
34582 +                       AuDbg("b%d, %llu\n", err, b);
34583 +               }
34584 +       }
34585 +
34586 +       if (err >= 0)
34587 +               err = au_wbr_nonopq(dentry, err);
34588 +
34589 +out_parent:
34590 +       dput(parent);
34591 +out:
34592 +       AuDbg("b%d\n", err);
34593 +       return err;
34594 +}
34595 +
34596 +/* ---------------------------------------------------------------------- */
34597 +
34598 +/*
34599 + * - top down parent
34600 + * - most free space with parent
34601 + * - most free space round-robin regardless parent
34602 + */
34603 +static int au_wbr_create_pmfsrr(struct dentry *dentry, unsigned int flags)
34604 +{
34605 +       int err;
34606 +       unsigned long long watermark;
34607 +       struct super_block *sb;
34608 +       struct au_branch *br;
34609 +       struct au_wbr_mfs *mfs;
34610 +
34611 +       err = au_wbr_create_pmfs(dentry, flags | AuWbr_PARENT);
34612 +       if (unlikely(err < 0))
34613 +               goto out;
34614 +
34615 +       sb = dentry->d_sb;
34616 +       br = au_sbr(sb, err);
34617 +       mfs = &au_sbi(sb)->si_wbr_mfs;
34618 +       mutex_lock(&mfs->mfs_lock);
34619 +       watermark = mfs->mfsrr_watermark;
34620 +       mutex_unlock(&mfs->mfs_lock);
34621 +       if (br->br_wbr->wbr_bytes < watermark)
34622 +               /* regardless the parent dir */
34623 +               err = au_wbr_create_mfsrr(dentry, flags);
34624 +
34625 +out:
34626 +       AuDbg("b%d\n", err);
34627 +       return err;
34628 +}
34629 +
34630 +/* ---------------------------------------------------------------------- */
34631 +
34632 +/* policies for copyup */
34633 +
34634 +/* top down parent */
34635 +static int au_wbr_copyup_tdp(struct dentry *dentry)
34636 +{
34637 +       return au_wbr_create_tdp(dentry, /*flags, anything is ok*/0);
34638 +}
34639 +
34640 +/* bottom up parent */
34641 +static int au_wbr_copyup_bup(struct dentry *dentry)
34642 +{
34643 +       int err;
34644 +       aufs_bindex_t bindex, btop;
34645 +       struct dentry *parent, *h_parent;
34646 +       struct super_block *sb;
34647 +
34648 +       err = -EROFS;
34649 +       sb = dentry->d_sb;
34650 +       parent = dget_parent(dentry);
34651 +       btop = au_dbtop(parent);
34652 +       for (bindex = au_dbtop(dentry); bindex >= btop; bindex--) {
34653 +               h_parent = au_h_dptr(parent, bindex);
34654 +               if (!h_parent || d_is_negative(h_parent))
34655 +                       continue;
34656 +
34657 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
34658 +                       err = bindex;
34659 +                       break;
34660 +               }
34661 +       }
34662 +       dput(parent);
34663 +
34664 +       /* bottom up here */
34665 +       if (unlikely(err < 0))
34666 +               err = au_wbr_bu(sb, btop - 1);
34667 +
34668 +       AuDbg("b%d\n", err);
34669 +       return err;
34670 +}
34671 +
34672 +/* bottom up */
34673 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t btop)
34674 +{
34675 +       int err;
34676 +
34677 +       err = au_wbr_bu(dentry->d_sb, btop);
34678 +       AuDbg("b%d\n", err);
34679 +       if (err > btop)
34680 +               err = au_wbr_nonopq(dentry, err);
34681 +
34682 +       AuDbg("b%d\n", err);
34683 +       return err;
34684 +}
34685 +
34686 +static int au_wbr_copyup_bu(struct dentry *dentry)
34687 +{
34688 +       int err;
34689 +       aufs_bindex_t btop;
34690 +
34691 +       btop = au_dbtop(dentry);
34692 +       err = au_wbr_do_copyup_bu(dentry, btop);
34693 +       return err;
34694 +}
34695 +
34696 +/* ---------------------------------------------------------------------- */
34697 +
34698 +struct au_wbr_copyup_operations au_wbr_copyup_ops[] = {
34699 +       [AuWbrCopyup_TDP] = {
34700 +               .copyup = au_wbr_copyup_tdp
34701 +       },
34702 +       [AuWbrCopyup_BUP] = {
34703 +               .copyup = au_wbr_copyup_bup
34704 +       },
34705 +       [AuWbrCopyup_BU] = {
34706 +               .copyup = au_wbr_copyup_bu
34707 +       }
34708 +};
34709 +
34710 +struct au_wbr_create_operations au_wbr_create_ops[] = {
34711 +       [AuWbrCreate_TDP] = {
34712 +               .create = au_wbr_create_tdp
34713 +       },
34714 +       [AuWbrCreate_RR] = {
34715 +               .create = au_wbr_create_rr,
34716 +               .init   = au_wbr_create_init_rr
34717 +       },
34718 +       [AuWbrCreate_MFS] = {
34719 +               .create = au_wbr_create_mfs,
34720 +               .init   = au_wbr_create_init_mfs,
34721 +               .fin    = au_wbr_create_fin_mfs
34722 +       },
34723 +       [AuWbrCreate_MFSV] = {
34724 +               .create = au_wbr_create_mfs,
34725 +               .init   = au_wbr_create_init_mfs,
34726 +               .fin    = au_wbr_create_fin_mfs
34727 +       },
34728 +       [AuWbrCreate_MFSRR] = {
34729 +               .create = au_wbr_create_mfsrr,
34730 +               .init   = au_wbr_create_init_mfsrr,
34731 +               .fin    = au_wbr_create_fin_mfs
34732 +       },
34733 +       [AuWbrCreate_MFSRRV] = {
34734 +               .create = au_wbr_create_mfsrr,
34735 +               .init   = au_wbr_create_init_mfsrr,
34736 +               .fin    = au_wbr_create_fin_mfs
34737 +       },
34738 +       [AuWbrCreate_TDMFS] = {
34739 +               .create = au_wbr_create_tdmfs,
34740 +               .init   = au_wbr_create_init_mfs,
34741 +               .fin    = au_wbr_create_fin_mfs
34742 +       },
34743 +       [AuWbrCreate_TDMFSV] = {
34744 +               .create = au_wbr_create_tdmfs,
34745 +               .init   = au_wbr_create_init_mfs,
34746 +               .fin    = au_wbr_create_fin_mfs
34747 +       },
34748 +       [AuWbrCreate_PMFS] = {
34749 +               .create = au_wbr_create_pmfs,
34750 +               .init   = au_wbr_create_init_mfs,
34751 +               .fin    = au_wbr_create_fin_mfs
34752 +       },
34753 +       [AuWbrCreate_PMFSV] = {
34754 +               .create = au_wbr_create_pmfs,
34755 +               .init   = au_wbr_create_init_mfs,
34756 +               .fin    = au_wbr_create_fin_mfs
34757 +       },
34758 +       [AuWbrCreate_PMFSRR] = {
34759 +               .create = au_wbr_create_pmfsrr,
34760 +               .init   = au_wbr_create_init_mfsrr,
34761 +               .fin    = au_wbr_create_fin_mfs
34762 +       },
34763 +       [AuWbrCreate_PMFSRRV] = {
34764 +               .create = au_wbr_create_pmfsrr,
34765 +               .init   = au_wbr_create_init_mfsrr,
34766 +               .fin    = au_wbr_create_fin_mfs
34767 +       }
34768 +};
34769 diff -urN /usr/share/empty/fs/aufs/whout.c linux/fs/aufs/whout.c
34770 --- /usr/share/empty/fs/aufs/whout.c    1970-01-01 01:00:00.000000000 +0100
34771 +++ linux/fs/aufs/whout.c       2021-12-03 15:40:58.236647297 +0100
34772 @@ -0,0 +1,1072 @@
34773 +// SPDX-License-Identifier: GPL-2.0
34774 +/*
34775 + * Copyright (C) 2005-2021 Junjiro R. Okajima
34776 + *
34777 + * This program, aufs is free software; you can redistribute it and/or modify
34778 + * it under the terms of the GNU General Public License as published by
34779 + * the Free Software Foundation; either version 2 of the License, or
34780 + * (at your option) any later version.
34781 + *
34782 + * This program is distributed in the hope that it will be useful,
34783 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
34784 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
34785 + * GNU General Public License for more details.
34786 + *
34787 + * You should have received a copy of the GNU General Public License
34788 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
34789 + */
34790 +
34791 +/*
34792 + * whiteout for logical deletion and opaque directory
34793 + */
34794 +
34795 +#include "aufs.h"
34796 +
34797 +#define WH_MASK                        0444
34798 +
34799 +/*
34800 + * If a directory contains this file, then it is opaque.  We start with the
34801 + * .wh. flag so that it is blocked by lookup.
34802 + */
34803 +static struct qstr diropq_name = QSTR_INIT(AUFS_WH_DIROPQ,
34804 +                                          sizeof(AUFS_WH_DIROPQ) - 1);
34805 +
34806 +/*
34807 + * generate whiteout name, which is NOT terminated by NULL.
34808 + * @name: original d_name.name
34809 + * @len: original d_name.len
34810 + * @wh: whiteout qstr
34811 + * returns zero when succeeds, otherwise error.
34812 + * succeeded value as wh->name should be freed by kfree().
34813 + */
34814 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name)
34815 +{
34816 +       char *p;
34817 +
34818 +       if (unlikely(name->len > PATH_MAX - AUFS_WH_PFX_LEN))
34819 +               return -ENAMETOOLONG;
34820 +
34821 +       wh->len = name->len + AUFS_WH_PFX_LEN;
34822 +       p = kmalloc(wh->len, GFP_NOFS);
34823 +       wh->name = p;
34824 +       if (p) {
34825 +               memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
34826 +               memcpy(p + AUFS_WH_PFX_LEN, name->name, name->len);
34827 +               /* smp_mb(); */
34828 +               return 0;
34829 +       }
34830 +       return -ENOMEM;
34831 +}
34832 +
34833 +/* ---------------------------------------------------------------------- */
34834 +
34835 +/*
34836 + * test if the @wh_name exists under @h_ppath.
34837 + * @try_sio specifies the necessary of super-io.
34838 + */
34839 +int au_wh_test(struct user_namespace *h_userns, struct path *h_ppath,
34840 +              struct qstr *wh_name, int try_sio)
34841 +{
34842 +       int err;
34843 +       struct dentry *wh_dentry;
34844 +
34845 +       if (!try_sio)
34846 +               wh_dentry = vfsub_lkup_one(wh_name, h_ppath);
34847 +       else
34848 +               wh_dentry = au_sio_lkup_one(h_userns, wh_name, h_ppath);
34849 +       err = PTR_ERR(wh_dentry);
34850 +       if (IS_ERR(wh_dentry)) {
34851 +               if (err == -ENAMETOOLONG)
34852 +                       err = 0;
34853 +               goto out;
34854 +       }
34855 +
34856 +       err = 0;
34857 +       if (d_is_negative(wh_dentry))
34858 +               goto out_wh; /* success */
34859 +
34860 +       err = 1;
34861 +       if (d_is_reg(wh_dentry))
34862 +               goto out_wh; /* success */
34863 +
34864 +       err = -EIO;
34865 +       AuIOErr("%pd Invalid whiteout entry type 0%o.\n",
34866 +               wh_dentry, d_inode(wh_dentry)->i_mode);
34867 +
34868 +out_wh:
34869 +       dput(wh_dentry);
34870 +out:
34871 +       return err;
34872 +}
34873 +
34874 +/*
34875 + * test if the @h_path->dentry sets opaque or not.
34876 + */
34877 +int au_diropq_test(struct user_namespace *h_userns, struct path *h_path)
34878 +{
34879 +       int err;
34880 +       struct inode *h_dir;
34881 +
34882 +       h_dir = d_inode(h_path->dentry);
34883 +       err = au_wh_test(h_userns, h_path, &diropq_name,
34884 +                        au_test_h_perm_sio(h_userns, h_dir, MAY_EXEC));
34885 +       return err;
34886 +}
34887 +
34888 +/*
34889 + * returns a negative dentry whose name is unique and temporary.
34890 + */
34891 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
34892 +                            struct qstr *prefix)
34893 +{
34894 +       struct dentry *dentry;
34895 +       int i;
34896 +       char defname[NAME_MAX - AUFS_MAX_NAMELEN + DNAME_INLINE_LEN + 1],
34897 +               *name, *p;
34898 +       /* strict atomic_t is unnecessary here */
34899 +       static unsigned short cnt;
34900 +       struct qstr qs;
34901 +       struct path h_ppath;
34902 +       struct user_namespace *h_userns;
34903 +
34904 +       BUILD_BUG_ON(sizeof(cnt) * 2 > AUFS_WH_TMP_LEN);
34905 +
34906 +       name = defname;
34907 +       qs.len = sizeof(defname) - DNAME_INLINE_LEN + prefix->len - 1;
34908 +       if (unlikely(prefix->len > DNAME_INLINE_LEN)) {
34909 +               dentry = ERR_PTR(-ENAMETOOLONG);
34910 +               if (unlikely(qs.len > NAME_MAX))
34911 +                       goto out;
34912 +               dentry = ERR_PTR(-ENOMEM);
34913 +               name = kmalloc(qs.len + 1, GFP_NOFS);
34914 +               if (unlikely(!name))
34915 +                       goto out;
34916 +       }
34917 +
34918 +       /* doubly whiteout-ed */
34919 +       memcpy(name, AUFS_WH_PFX AUFS_WH_PFX, AUFS_WH_PFX_LEN * 2);
34920 +       p = name + AUFS_WH_PFX_LEN * 2;
34921 +       memcpy(p, prefix->name, prefix->len);
34922 +       p += prefix->len;
34923 +       *p++ = '.';
34924 +       AuDebugOn(name + qs.len + 1 - p <= AUFS_WH_TMP_LEN);
34925 +
34926 +       h_ppath.dentry = h_parent;
34927 +       h_ppath.mnt = au_br_mnt(br);
34928 +       h_userns = au_br_userns(br);
34929 +       qs.name = name;
34930 +       for (i = 0; i < 3; i++) {
34931 +               sprintf(p, "%.*x", AUFS_WH_TMP_LEN, cnt++);
34932 +               dentry = au_sio_lkup_one(h_userns, &qs, &h_ppath);
34933 +               if (IS_ERR(dentry) || d_is_negative(dentry))
34934 +                       goto out_name;
34935 +               dput(dentry);
34936 +       }
34937 +       /* pr_warn("could not get random name\n"); */
34938 +       dentry = ERR_PTR(-EEXIST);
34939 +       AuDbg("%.*s\n", AuLNPair(&qs));
34940 +       BUG();
34941 +
34942 +out_name:
34943 +       if (name != defname)
34944 +               au_kfree_try_rcu(name);
34945 +out:
34946 +       AuTraceErrPtr(dentry);
34947 +       return dentry;
34948 +}
34949 +
34950 +/*
34951 + * rename the @h_dentry on @br to the whiteouted temporary name.
34952 + */
34953 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br)
34954 +{
34955 +       int err;
34956 +       struct path h_path = {
34957 +               .mnt = au_br_mnt(br)
34958 +       };
34959 +       struct inode *h_dir, *delegated;
34960 +       struct dentry *h_parent;
34961 +
34962 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
34963 +       h_dir = d_inode(h_parent);
34964 +       IMustLock(h_dir);
34965 +
34966 +       h_path.dentry = au_whtmp_lkup(h_parent, br, &h_dentry->d_name);
34967 +       err = PTR_ERR(h_path.dentry);
34968 +       if (IS_ERR(h_path.dentry))
34969 +               goto out;
34970 +
34971 +       /* under the same dir, no need to lock_rename() */
34972 +       delegated = NULL;
34973 +       err = vfsub_rename(h_dir, h_dentry, h_dir, &h_path, &delegated,
34974 +                          /*flags*/0);
34975 +       AuTraceErr(err);
34976 +       if (unlikely(err == -EWOULDBLOCK)) {
34977 +               pr_warn("cannot retry for NFSv4 delegation"
34978 +                       " for an internal rename\n");
34979 +               iput(delegated);
34980 +       }
34981 +       dput(h_path.dentry);
34982 +
34983 +out:
34984 +       AuTraceErr(err);
34985 +       return err;
34986 +}
34987 +
34988 +/* ---------------------------------------------------------------------- */
34989 +/*
34990 + * functions for removing a whiteout
34991 + */
34992 +
34993 +static int do_unlink_wh(struct inode *h_dir, struct path *h_path)
34994 +{
34995 +       int err, force;
34996 +       struct inode *delegated;
34997 +
34998 +       /*
34999 +        * forces superio when the dir has a sticky bit.
35000 +        * this may be a violation of unix fs semantics.
35001 +        */
35002 +       force = (h_dir->i_mode & S_ISVTX)
35003 +               && !uid_eq(current_fsuid(), d_inode(h_path->dentry)->i_uid);
35004 +       delegated = NULL;
35005 +       err = vfsub_unlink(h_dir, h_path, &delegated, force);
35006 +       if (unlikely(err == -EWOULDBLOCK)) {
35007 +               pr_warn("cannot retry for NFSv4 delegation"
35008 +                       " for an internal unlink\n");
35009 +               iput(delegated);
35010 +       }
35011 +       return err;
35012 +}
35013 +
35014 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
35015 +                       struct dentry *dentry)
35016 +{
35017 +       int err;
35018 +
35019 +       err = do_unlink_wh(h_dir, h_path);
35020 +       if (!err && dentry)
35021 +               au_set_dbwh(dentry, -1);
35022 +
35023 +       return err;
35024 +}
35025 +
35026 +static int unlink_wh_name(struct path *h_ppath, struct qstr *wh)
35027 +{
35028 +       int err;
35029 +       struct path h_path;
35030 +
35031 +       err = 0;
35032 +       h_path.dentry = vfsub_lkup_one(wh, h_ppath);
35033 +       if (IS_ERR(h_path.dentry))
35034 +               err = PTR_ERR(h_path.dentry);
35035 +       else {
35036 +               if (d_is_reg(h_path.dentry)) {
35037 +                       h_path.mnt = h_ppath->mnt;
35038 +                       err = do_unlink_wh(d_inode(h_ppath->dentry), &h_path);
35039 +               }
35040 +               dput(h_path.dentry);
35041 +       }
35042 +
35043 +       return err;
35044 +}
35045 +
35046 +/* ---------------------------------------------------------------------- */
35047 +/*
35048 + * initialize/clean whiteout for a branch
35049 + */
35050 +
35051 +static void au_wh_clean(struct inode *h_dir, struct path *whpath,
35052 +                       const int isdir)
35053 +{
35054 +       int err;
35055 +       struct inode *delegated;
35056 +
35057 +       if (d_is_negative(whpath->dentry))
35058 +               return;
35059 +
35060 +       if (isdir)
35061 +               err = vfsub_rmdir(h_dir, whpath);
35062 +       else {
35063 +               delegated = NULL;
35064 +               err = vfsub_unlink(h_dir, whpath, &delegated, /*force*/0);
35065 +               if (unlikely(err == -EWOULDBLOCK)) {
35066 +                       pr_warn("cannot retry for NFSv4 delegation"
35067 +                               " for an internal unlink\n");
35068 +                       iput(delegated);
35069 +               }
35070 +       }
35071 +       if (unlikely(err))
35072 +               pr_warn("failed removing %pd (%d), ignored.\n",
35073 +                       whpath->dentry, err);
35074 +}
35075 +
35076 +static int test_linkable(struct dentry *h_root)
35077 +{
35078 +       struct inode *h_dir = d_inode(h_root);
35079 +
35080 +       if (h_dir->i_op->link)
35081 +               return 0;
35082 +
35083 +       pr_err("%pd (%s) doesn't support link(2), use noplink and rw+nolwh\n",
35084 +              h_root, au_sbtype(h_root->d_sb));
35085 +       return -ENOSYS; /* the branch doesn't have its ->link() */
35086 +}
35087 +
35088 +/* todo: should this mkdir be done in /sbin/mount.aufs helper? */
35089 +static int au_whdir(struct inode *h_dir, struct path *path)
35090 +{
35091 +       int err;
35092 +
35093 +       err = -EEXIST;
35094 +       if (d_is_negative(path->dentry)) {
35095 +               int mode = 0700;
35096 +
35097 +               if (au_test_nfs(path->dentry->d_sb))
35098 +                       mode |= 0111;
35099 +               err = vfsub_mkdir(h_dir, path, mode);
35100 +       } else if (d_is_dir(path->dentry))
35101 +               err = 0;
35102 +       else
35103 +               pr_err("unknown %pd exists\n", path->dentry);
35104 +
35105 +       return err;
35106 +}
35107 +
35108 +struct au_wh_base {
35109 +       const struct qstr *name;
35110 +       struct dentry *dentry;
35111 +};
35112 +
35113 +static void au_wh_init_ro(struct inode *h_dir, struct au_wh_base base[],
35114 +                         struct path *h_path)
35115 +{
35116 +       h_path->dentry = base[AuBrWh_BASE].dentry;
35117 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
35118 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
35119 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
35120 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
35121 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
35122 +}
35123 +
35124 +/*
35125 + * returns tri-state,
35126 + * minus: error, caller should print the message
35127 + * zero: success
35128 + * plus: error, caller should NOT print the message
35129 + */
35130 +static int au_wh_init_rw_nolink(struct dentry *h_root, struct au_wbr *wbr,
35131 +                               int do_plink, struct au_wh_base base[],
35132 +                               struct path *h_path)
35133 +{
35134 +       int err;
35135 +       struct inode *h_dir;
35136 +
35137 +       h_dir = d_inode(h_root);
35138 +       h_path->dentry = base[AuBrWh_BASE].dentry;
35139 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
35140 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
35141 +       if (do_plink) {
35142 +               err = test_linkable(h_root);
35143 +               if (unlikely(err)) {
35144 +                       err = 1;
35145 +                       goto out;
35146 +               }
35147 +
35148 +               err = au_whdir(h_dir, h_path);
35149 +               if (unlikely(err))
35150 +                       goto out;
35151 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
35152 +       } else
35153 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
35154 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
35155 +       err = au_whdir(h_dir, h_path);
35156 +       if (unlikely(err))
35157 +               goto out;
35158 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
35159 +
35160 +out:
35161 +       return err;
35162 +}
35163 +
35164 +/*
35165 + * for the moment, aufs supports the branch filesystem which does not support
35166 + * link(2). testing on FAT which does not support i_op->setattr() fully either,
35167 + * copyup failed. finally, such filesystem will not be used as the writable
35168 + * branch.
35169 + *
35170 + * returns tri-state, see above.
35171 + */
35172 +static int au_wh_init_rw(struct dentry *h_root, struct au_wbr *wbr,
35173 +                        int do_plink, struct au_wh_base base[],
35174 +                        struct path *h_path)
35175 +{
35176 +       int err;
35177 +       struct inode *h_dir;
35178 +
35179 +       WbrWhMustWriteLock(wbr);
35180 +
35181 +       err = test_linkable(h_root);
35182 +       if (unlikely(err)) {
35183 +               err = 1;
35184 +               goto out;
35185 +       }
35186 +
35187 +       /*
35188 +        * todo: should this create be done in /sbin/mount.aufs helper?
35189 +        */
35190 +       err = -EEXIST;
35191 +       h_dir = d_inode(h_root);
35192 +       if (d_is_negative(base[AuBrWh_BASE].dentry)) {
35193 +               h_path->dentry = base[AuBrWh_BASE].dentry;
35194 +               err = vfsub_create(h_dir, h_path, WH_MASK, /*want_excl*/true);
35195 +       } else if (d_is_reg(base[AuBrWh_BASE].dentry))
35196 +               err = 0;
35197 +       else
35198 +               pr_err("unknown %pd2 exists\n", base[AuBrWh_BASE].dentry);
35199 +       if (unlikely(err))
35200 +               goto out;
35201 +
35202 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
35203 +       if (do_plink) {
35204 +               err = au_whdir(h_dir, h_path);
35205 +               if (unlikely(err))
35206 +                       goto out;
35207 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
35208 +       } else
35209 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
35210 +       wbr->wbr_whbase = dget(base[AuBrWh_BASE].dentry);
35211 +
35212 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
35213 +       err = au_whdir(h_dir, h_path);
35214 +       if (unlikely(err))
35215 +               goto out;
35216 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
35217 +
35218 +out:
35219 +       return err;
35220 +}
35221 +
35222 +/*
35223 + * initialize the whiteout base file/dir for @br.
35224 + */
35225 +int au_wh_init(struct au_branch *br, struct super_block *sb)
35226 +{
35227 +       int err, i;
35228 +       const unsigned char do_plink
35229 +               = !!au_opt_test(au_mntflags(sb), PLINK);
35230 +       struct inode *h_dir;
35231 +       struct path path = br->br_path;
35232 +       struct dentry *h_root = path.dentry;
35233 +       struct au_wbr *wbr = br->br_wbr;
35234 +       static const struct qstr base_name[] = {
35235 +               [AuBrWh_BASE] = QSTR_INIT(AUFS_BASE_NAME,
35236 +                                         sizeof(AUFS_BASE_NAME) - 1),
35237 +               [AuBrWh_PLINK] = QSTR_INIT(AUFS_PLINKDIR_NAME,
35238 +                                          sizeof(AUFS_PLINKDIR_NAME) - 1),
35239 +               [AuBrWh_ORPH] = QSTR_INIT(AUFS_ORPHDIR_NAME,
35240 +                                         sizeof(AUFS_ORPHDIR_NAME) - 1)
35241 +       };
35242 +       struct au_wh_base base[] = {
35243 +               [AuBrWh_BASE] = {
35244 +                       .name   = base_name + AuBrWh_BASE,
35245 +                       .dentry = NULL
35246 +               },
35247 +               [AuBrWh_PLINK] = {
35248 +                       .name   = base_name + AuBrWh_PLINK,
35249 +                       .dentry = NULL
35250 +               },
35251 +               [AuBrWh_ORPH] = {
35252 +                       .name   = base_name + AuBrWh_ORPH,
35253 +                       .dentry = NULL
35254 +               }
35255 +       };
35256 +
35257 +       if (wbr)
35258 +               WbrWhMustWriteLock(wbr);
35259 +
35260 +       for (i = 0; i < AuBrWh_Last; i++) {
35261 +               /* doubly whiteouted */
35262 +               struct dentry *d;
35263 +
35264 +               d = au_wh_lkup(h_root, (void *)base[i].name, br);
35265 +               err = PTR_ERR(d);
35266 +               if (IS_ERR(d))
35267 +                       goto out;
35268 +
35269 +               base[i].dentry = d;
35270 +               AuDebugOn(wbr
35271 +                         && wbr->wbr_wh[i]
35272 +                         && wbr->wbr_wh[i] != base[i].dentry);
35273 +       }
35274 +
35275 +       if (wbr)
35276 +               for (i = 0; i < AuBrWh_Last; i++) {
35277 +                       dput(wbr->wbr_wh[i]);
35278 +                       wbr->wbr_wh[i] = NULL;
35279 +               }
35280 +
35281 +       err = 0;
35282 +       if (!au_br_writable(br->br_perm)) {
35283 +               h_dir = d_inode(h_root);
35284 +               au_wh_init_ro(h_dir, base, &path);
35285 +       } else if (!au_br_wh_linkable(br->br_perm)) {
35286 +               err = au_wh_init_rw_nolink(h_root, wbr, do_plink, base, &path);
35287 +               if (err > 0)
35288 +                       goto out;
35289 +               else if (err)
35290 +                       goto out_err;
35291 +       } else {
35292 +               err = au_wh_init_rw(h_root, wbr, do_plink, base, &path);
35293 +               if (err > 0)
35294 +                       goto out;
35295 +               else if (err)
35296 +                       goto out_err;
35297 +       }
35298 +       goto out; /* success */
35299 +
35300 +out_err:
35301 +       pr_err("an error(%d) on the writable branch %pd(%s)\n",
35302 +              err, h_root, au_sbtype(h_root->d_sb));
35303 +out:
35304 +       for (i = 0; i < AuBrWh_Last; i++)
35305 +               dput(base[i].dentry);
35306 +       return err;
35307 +}
35308 +
35309 +/* ---------------------------------------------------------------------- */
35310 +/*
35311 + * whiteouts are all hard-linked usually.
35312 + * when its link count reaches a ceiling, we create a new whiteout base
35313 + * asynchronously.
35314 + */
35315 +
35316 +struct reinit_br_wh {
35317 +       struct super_block *sb;
35318 +       struct au_branch *br;
35319 +};
35320 +
35321 +static void reinit_br_wh(void *arg)
35322 +{
35323 +       int err;
35324 +       aufs_bindex_t bindex;
35325 +       struct path h_path;
35326 +       struct reinit_br_wh *a = arg;
35327 +       struct au_wbr *wbr;
35328 +       struct inode *dir, *delegated;
35329 +       struct dentry *h_root;
35330 +       struct au_hinode *hdir;
35331 +
35332 +       err = 0;
35333 +       wbr = a->br->br_wbr;
35334 +       /* big aufs lock */
35335 +       si_noflush_write_lock(a->sb);
35336 +       if (!au_br_writable(a->br->br_perm))
35337 +               goto out;
35338 +       bindex = au_br_index(a->sb, a->br->br_id);
35339 +       if (unlikely(bindex < 0))
35340 +               goto out;
35341 +
35342 +       di_read_lock_parent(a->sb->s_root, AuLock_IR);
35343 +       dir = d_inode(a->sb->s_root);
35344 +       hdir = au_hi(dir, bindex);
35345 +       h_root = au_h_dptr(a->sb->s_root, bindex);
35346 +       AuDebugOn(h_root != au_br_dentry(a->br));
35347 +
35348 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
35349 +       wbr_wh_write_lock(wbr);
35350 +       err = au_h_verify(wbr->wbr_whbase, au_opt_udba(a->sb), hdir->hi_inode,
35351 +                         h_root, a->br);
35352 +       if (!err) {
35353 +               h_path.dentry = wbr->wbr_whbase;
35354 +               h_path.mnt = au_br_mnt(a->br);
35355 +               delegated = NULL;
35356 +               err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated,
35357 +                                  /*force*/0);
35358 +               if (unlikely(err == -EWOULDBLOCK)) {
35359 +                       pr_warn("cannot retry for NFSv4 delegation"
35360 +                               " for an internal unlink\n");
35361 +                       iput(delegated);
35362 +               }
35363 +       } else {
35364 +               pr_warn("%pd is moved, ignored\n", wbr->wbr_whbase);
35365 +               err = 0;
35366 +       }
35367 +       dput(wbr->wbr_whbase);
35368 +       wbr->wbr_whbase = NULL;
35369 +       if (!err)
35370 +               err = au_wh_init(a->br, a->sb);
35371 +       wbr_wh_write_unlock(wbr);
35372 +       au_hn_inode_unlock(hdir);
35373 +       di_read_unlock(a->sb->s_root, AuLock_IR);
35374 +       if (!err)
35375 +               au_fhsm_wrote(a->sb, bindex, /*force*/0);
35376 +
35377 +out:
35378 +       if (wbr)
35379 +               atomic_dec(&wbr->wbr_wh_running);
35380 +       au_lcnt_dec(&a->br->br_count);
35381 +       si_write_unlock(a->sb);
35382 +       au_nwt_done(&au_sbi(a->sb)->si_nowait);
35383 +       au_kfree_rcu(a);
35384 +       if (unlikely(err))
35385 +               AuIOErr("err %d\n", err);
35386 +}
35387 +
35388 +static void kick_reinit_br_wh(struct super_block *sb, struct au_branch *br)
35389 +{
35390 +       int do_dec, wkq_err;
35391 +       struct reinit_br_wh *arg;
35392 +
35393 +       do_dec = 1;
35394 +       if (atomic_inc_return(&br->br_wbr->wbr_wh_running) != 1)
35395 +               goto out;
35396 +
35397 +       /* ignore ENOMEM */
35398 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
35399 +       if (arg) {
35400 +               /*
35401 +                * dec(wh_running), kfree(arg) and dec(br_count)
35402 +                * in reinit function
35403 +                */
35404 +               arg->sb = sb;
35405 +               arg->br = br;
35406 +               au_lcnt_inc(&br->br_count);
35407 +               wkq_err = au_wkq_nowait(reinit_br_wh, arg, sb, /*flags*/0);
35408 +               if (unlikely(wkq_err)) {
35409 +                       atomic_dec(&br->br_wbr->wbr_wh_running);
35410 +                       au_lcnt_dec(&br->br_count);
35411 +                       au_kfree_rcu(arg);
35412 +               }
35413 +               do_dec = 0;
35414 +       }
35415 +
35416 +out:
35417 +       if (do_dec)
35418 +               atomic_dec(&br->br_wbr->wbr_wh_running);
35419 +}
35420 +
35421 +/* ---------------------------------------------------------------------- */
35422 +
35423 +/*
35424 + * create the whiteout @wh.
35425 + */
35426 +static int link_or_create_wh(struct super_block *sb, aufs_bindex_t bindex,
35427 +                            struct dentry *wh)
35428 +{
35429 +       int err;
35430 +       struct path h_path = {
35431 +               .dentry = wh
35432 +       };
35433 +       struct au_branch *br;
35434 +       struct au_wbr *wbr;
35435 +       struct dentry *h_parent;
35436 +       struct inode *h_dir, *delegated;
35437 +
35438 +       h_parent = wh->d_parent; /* dir inode is locked */
35439 +       h_dir = d_inode(h_parent);
35440 +       IMustLock(h_dir);
35441 +
35442 +       br = au_sbr(sb, bindex);
35443 +       h_path.mnt = au_br_mnt(br);
35444 +       wbr = br->br_wbr;
35445 +       wbr_wh_read_lock(wbr);
35446 +       if (wbr->wbr_whbase) {
35447 +               delegated = NULL;
35448 +               err = vfsub_link(wbr->wbr_whbase, h_dir, &h_path, &delegated);
35449 +               if (unlikely(err == -EWOULDBLOCK)) {
35450 +                       pr_warn("cannot retry for NFSv4 delegation"
35451 +                               " for an internal link\n");
35452 +                       iput(delegated);
35453 +               }
35454 +               if (!err || err != -EMLINK)
35455 +                       goto out;
35456 +
35457 +               /* link count full. re-initialize br_whbase. */
35458 +               kick_reinit_br_wh(sb, br);
35459 +       }
35460 +
35461 +       /* return this error in this context */
35462 +       err = vfsub_create(h_dir, &h_path, WH_MASK, /*want_excl*/true);
35463 +       if (!err)
35464 +               au_fhsm_wrote(sb, bindex, /*force*/0);
35465 +
35466 +out:
35467 +       wbr_wh_read_unlock(wbr);
35468 +       return err;
35469 +}
35470 +
35471 +/* ---------------------------------------------------------------------- */
35472 +
35473 +/*
35474 + * create or remove the diropq.
35475 + */
35476 +static struct dentry *do_diropq(struct dentry *dentry, aufs_bindex_t bindex,
35477 +                               unsigned int flags)
35478 +{
35479 +       struct dentry *opq_dentry;
35480 +       struct super_block *sb;
35481 +       struct au_branch *br;
35482 +       struct path h_path;
35483 +       int err;
35484 +
35485 +       sb = dentry->d_sb;
35486 +       br = au_sbr(sb, bindex);
35487 +       h_path.dentry = au_h_dptr(dentry, bindex);
35488 +       h_path.mnt = au_br_mnt(br);
35489 +       opq_dentry = vfsub_lkup_one(&diropq_name, &h_path);
35490 +       if (IS_ERR(opq_dentry))
35491 +               goto out;
35492 +
35493 +       if (au_ftest_diropq(flags, CREATE)) {
35494 +               err = link_or_create_wh(sb, bindex, opq_dentry);
35495 +               if (!err) {
35496 +                       au_set_dbdiropq(dentry, bindex);
35497 +                       goto out; /* success */
35498 +               }
35499 +       } else {
35500 +               h_path.dentry = opq_dentry;
35501 +               err = do_unlink_wh(au_h_iptr(d_inode(dentry), bindex), &h_path);
35502 +               if (!err)
35503 +                       au_set_dbdiropq(dentry, -1);
35504 +       }
35505 +       dput(opq_dentry);
35506 +       opq_dentry = ERR_PTR(err);
35507 +
35508 +out:
35509 +       return opq_dentry;
35510 +}
35511 +
35512 +struct do_diropq_args {
35513 +       struct dentry **errp;
35514 +       struct dentry *dentry;
35515 +       aufs_bindex_t bindex;
35516 +       unsigned int flags;
35517 +};
35518 +
35519 +static void call_do_diropq(void *args)
35520 +{
35521 +       struct do_diropq_args *a = args;
35522 +       *a->errp = do_diropq(a->dentry, a->bindex, a->flags);
35523 +}
35524 +
35525 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
35526 +                            unsigned int flags)
35527 +{
35528 +       struct dentry *diropq, *h_dentry;
35529 +       struct user_namespace *h_userns;
35530 +
35531 +       h_userns = au_sbr_userns(dentry->d_sb, bindex);
35532 +       h_dentry = au_h_dptr(dentry, bindex);
35533 +       if (!au_test_h_perm_sio(h_userns, d_inode(h_dentry),
35534 +                               MAY_EXEC | MAY_WRITE))
35535 +               diropq = do_diropq(dentry, bindex, flags);
35536 +       else {
35537 +               int wkq_err;
35538 +               struct do_diropq_args args = {
35539 +                       .errp           = &diropq,
35540 +                       .dentry         = dentry,
35541 +                       .bindex         = bindex,
35542 +                       .flags          = flags
35543 +               };
35544 +
35545 +               wkq_err = au_wkq_wait(call_do_diropq, &args);
35546 +               if (unlikely(wkq_err))
35547 +                       diropq = ERR_PTR(wkq_err);
35548 +       }
35549 +
35550 +       return diropq;
35551 +}
35552 +
35553 +/* ---------------------------------------------------------------------- */
35554 +
35555 +/*
35556 + * lookup whiteout dentry.
35557 + * @h_parent: lower parent dentry which must exist and be locked
35558 + * @base_name: name of dentry which will be whiteouted
35559 + * returns dentry for whiteout.
35560 + */
35561 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
35562 +                         struct au_branch *br)
35563 +{
35564 +       int err;
35565 +       struct qstr wh_name;
35566 +       struct dentry *wh_dentry;
35567 +       struct path h_path;
35568 +
35569 +       err = au_wh_name_alloc(&wh_name, base_name);
35570 +       wh_dentry = ERR_PTR(err);
35571 +       if (!err) {
35572 +               h_path.dentry = h_parent;
35573 +               h_path.mnt = au_br_mnt(br);
35574 +               wh_dentry = vfsub_lkup_one(&wh_name, &h_path);
35575 +               au_kfree_try_rcu(wh_name.name);
35576 +       }
35577 +       return wh_dentry;
35578 +}
35579 +
35580 +/*
35581 + * link/create a whiteout for @dentry on @bindex.
35582 + */
35583 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
35584 +                           struct dentry *h_parent)
35585 +{
35586 +       struct dentry *wh_dentry;
35587 +       struct super_block *sb;
35588 +       int err;
35589 +
35590 +       sb = dentry->d_sb;
35591 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, au_sbr(sb, bindex));
35592 +       if (!IS_ERR(wh_dentry) && d_is_negative(wh_dentry)) {
35593 +               err = link_or_create_wh(sb, bindex, wh_dentry);
35594 +               if (!err) {
35595 +                       au_set_dbwh(dentry, bindex);
35596 +                       au_fhsm_wrote(sb, bindex, /*force*/0);
35597 +               } else {
35598 +                       dput(wh_dentry);
35599 +                       wh_dentry = ERR_PTR(err);
35600 +               }
35601 +       }
35602 +
35603 +       return wh_dentry;
35604 +}
35605 +
35606 +/* ---------------------------------------------------------------------- */
35607 +
35608 +/* Delete all whiteouts in this directory on branch bindex. */
35609 +static int del_wh_children(struct path *h_path, struct au_nhash *whlist,
35610 +                          aufs_bindex_t bindex)
35611 +{
35612 +       int err;
35613 +       unsigned long ul, n;
35614 +       struct qstr wh_name;
35615 +       char *p;
35616 +       struct hlist_head *head;
35617 +       struct au_vdir_wh *pos;
35618 +       struct au_vdir_destr *str;
35619 +
35620 +       err = -ENOMEM;
35621 +       p = (void *)__get_free_page(GFP_NOFS);
35622 +       wh_name.name = p;
35623 +       if (unlikely(!wh_name.name))
35624 +               goto out;
35625 +
35626 +       err = 0;
35627 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
35628 +       p += AUFS_WH_PFX_LEN;
35629 +       n = whlist->nh_num;
35630 +       head = whlist->nh_head;
35631 +       for (ul = 0; !err && ul < n; ul++, head++) {
35632 +               hlist_for_each_entry(pos, head, wh_hash) {
35633 +                       if (pos->wh_bindex != bindex)
35634 +                               continue;
35635 +
35636 +                       str = &pos->wh_str;
35637 +                       if (str->len + AUFS_WH_PFX_LEN <= PATH_MAX) {
35638 +                               memcpy(p, str->name, str->len);
35639 +                               wh_name.len = AUFS_WH_PFX_LEN + str->len;
35640 +                               err = unlink_wh_name(h_path, &wh_name);
35641 +                               if (!err)
35642 +                                       continue;
35643 +                               break;
35644 +                       }
35645 +                       AuIOErr("whiteout name too long %.*s\n",
35646 +                               str->len, str->name);
35647 +                       err = -EIO;
35648 +                       break;
35649 +               }
35650 +       }
35651 +       free_page((unsigned long)wh_name.name);
35652 +
35653 +out:
35654 +       return err;
35655 +}
35656 +
35657 +struct del_wh_children_args {
35658 +       int *errp;
35659 +       struct path *h_path;
35660 +       struct au_nhash *whlist;
35661 +       aufs_bindex_t bindex;
35662 +};
35663 +
35664 +static void call_del_wh_children(void *args)
35665 +{
35666 +       struct del_wh_children_args *a = args;
35667 +       *a->errp = del_wh_children(a->h_path, a->whlist, a->bindex);
35668 +}
35669 +
35670 +/* ---------------------------------------------------------------------- */
35671 +
35672 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp)
35673 +{
35674 +       struct au_whtmp_rmdir *whtmp;
35675 +       int err;
35676 +       unsigned int rdhash;
35677 +
35678 +       SiMustAnyLock(sb);
35679 +
35680 +       whtmp = kzalloc(sizeof(*whtmp), gfp);
35681 +       if (unlikely(!whtmp)) {
35682 +               whtmp = ERR_PTR(-ENOMEM);
35683 +               goto out;
35684 +       }
35685 +
35686 +       /* no estimation for dir size */
35687 +       rdhash = au_sbi(sb)->si_rdhash;
35688 +       if (!rdhash)
35689 +               rdhash = AUFS_RDHASH_DEF;
35690 +       err = au_nhash_alloc(&whtmp->whlist, rdhash, gfp);
35691 +       if (unlikely(err)) {
35692 +               au_kfree_rcu(whtmp);
35693 +               whtmp = ERR_PTR(err);
35694 +       }
35695 +
35696 +out:
35697 +       return whtmp;
35698 +}
35699 +
35700 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp)
35701 +{
35702 +       if (whtmp->br)
35703 +               au_lcnt_dec(&whtmp->br->br_count);
35704 +       dput(whtmp->wh_dentry);
35705 +       iput(whtmp->dir);
35706 +       au_nhash_wh_free(&whtmp->whlist);
35707 +       au_kfree_rcu(whtmp);
35708 +}
35709 +
35710 +/*
35711 + * rmdir the whiteouted temporary named dir @h_dentry.
35712 + * @whlist: whiteouted children.
35713 + */
35714 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
35715 +                  struct dentry *wh_dentry, struct au_nhash *whlist)
35716 +{
35717 +       int err;
35718 +       unsigned int h_nlink;
35719 +       struct path wh_path;
35720 +       struct inode *wh_inode, *h_dir;
35721 +       struct au_branch *br;
35722 +       struct user_namespace *h_userns;
35723 +
35724 +       h_dir = d_inode(wh_dentry->d_parent); /* dir inode is locked */
35725 +       IMustLock(h_dir);
35726 +
35727 +       br = au_sbr(dir->i_sb, bindex);
35728 +       wh_path.dentry = wh_dentry;
35729 +       wh_path.mnt = au_br_mnt(br);
35730 +       h_userns = au_br_userns(br);
35731 +       wh_inode = d_inode(wh_dentry);
35732 +       inode_lock_nested(wh_inode, AuLsc_I_CHILD);
35733 +
35734 +       /*
35735 +        * someone else might change some whiteouts while we were sleeping.
35736 +        * it means this whlist may have an obsoleted entry.
35737 +        */
35738 +       if (!au_test_h_perm_sio(h_userns, wh_inode, MAY_EXEC | MAY_WRITE))
35739 +               err = del_wh_children(&wh_path, whlist, bindex);
35740 +       else {
35741 +               int wkq_err;
35742 +               struct del_wh_children_args args = {
35743 +                       .errp           = &err,
35744 +                       .h_path         = &wh_path,
35745 +                       .whlist         = whlist,
35746 +                       .bindex         = bindex
35747 +               };
35748 +
35749 +               wkq_err = au_wkq_wait(call_del_wh_children, &args);
35750 +               if (unlikely(wkq_err))
35751 +                       err = wkq_err;
35752 +       }
35753 +       inode_unlock(wh_inode);
35754 +
35755 +       if (!err) {
35756 +               h_nlink = h_dir->i_nlink;
35757 +               err = vfsub_rmdir(h_dir, &wh_path);
35758 +               /* some fs doesn't change the parent nlink in some cases */
35759 +               h_nlink -= h_dir->i_nlink;
35760 +       }
35761 +
35762 +       if (!err) {
35763 +               if (au_ibtop(dir) == bindex) {
35764 +                       /* todo: dir->i_mutex is necessary */
35765 +                       au_cpup_attr_timesizes(dir);
35766 +                       if (h_nlink)
35767 +                               vfsub_drop_nlink(dir);
35768 +               }
35769 +               return 0; /* success */
35770 +       }
35771 +
35772 +       pr_warn("failed removing %pd(%d), ignored\n", wh_dentry, err);
35773 +       return err;
35774 +}
35775 +
35776 +static void call_rmdir_whtmp(void *args)
35777 +{
35778 +       int err;
35779 +       aufs_bindex_t bindex;
35780 +       struct au_whtmp_rmdir *a = args;
35781 +       struct super_block *sb;
35782 +       struct dentry *h_parent;
35783 +       struct inode *h_dir;
35784 +       struct au_hinode *hdir;
35785 +
35786 +       /* rmdir by nfsd may cause deadlock with this i_mutex */
35787 +       /* inode_lock(a->dir); */
35788 +       err = -EROFS;
35789 +       sb = a->dir->i_sb;
35790 +       si_read_lock(sb, !AuLock_FLUSH);
35791 +       if (!au_br_writable(a->br->br_perm))
35792 +               goto out;
35793 +       bindex = au_br_index(sb, a->br->br_id);
35794 +       if (unlikely(bindex < 0))
35795 +               goto out;
35796 +
35797 +       err = -EIO;
35798 +       ii_write_lock_parent(a->dir);
35799 +       h_parent = dget_parent(a->wh_dentry);
35800 +       h_dir = d_inode(h_parent);
35801 +       hdir = au_hi(a->dir, bindex);
35802 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
35803 +       if (unlikely(err))
35804 +               goto out_mnt;
35805 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
35806 +       err = au_h_verify(a->wh_dentry, au_opt_udba(sb), h_dir, h_parent,
35807 +                         a->br);
35808 +       if (!err)
35809 +               err = au_whtmp_rmdir(a->dir, bindex, a->wh_dentry, &a->whlist);
35810 +       au_hn_inode_unlock(hdir);
35811 +       vfsub_mnt_drop_write(au_br_mnt(a->br));
35812 +
35813 +out_mnt:
35814 +       dput(h_parent);
35815 +       ii_write_unlock(a->dir);
35816 +out:
35817 +       /* inode_unlock(a->dir); */
35818 +       au_whtmp_rmdir_free(a);
35819 +       si_read_unlock(sb);
35820 +       au_nwt_done(&au_sbi(sb)->si_nowait);
35821 +       if (unlikely(err))
35822 +               AuIOErr("err %d\n", err);
35823 +}
35824 +
35825 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
35826 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args)
35827 +{
35828 +       int wkq_err;
35829 +       struct super_block *sb;
35830 +
35831 +       IMustLock(dir);
35832 +
35833 +       /* all post-process will be done in do_rmdir_whtmp(). */
35834 +       sb = dir->i_sb;
35835 +       args->dir = au_igrab(dir);
35836 +       args->br = au_sbr(sb, bindex);
35837 +       au_lcnt_inc(&args->br->br_count);
35838 +       args->wh_dentry = dget(wh_dentry);
35839 +       wkq_err = au_wkq_nowait(call_rmdir_whtmp, args, sb, /*flags*/0);
35840 +       if (unlikely(wkq_err)) {
35841 +               pr_warn("rmdir error %pd (%d), ignored\n", wh_dentry, wkq_err);
35842 +               au_whtmp_rmdir_free(args);
35843 +       }
35844 +}
35845 diff -urN /usr/share/empty/fs/aufs/whout.h linux/fs/aufs/whout.h
35846 --- /usr/share/empty/fs/aufs/whout.h    1970-01-01 01:00:00.000000000 +0100
35847 +++ linux/fs/aufs/whout.h       2021-12-03 15:40:58.236647297 +0100
35848 @@ -0,0 +1,87 @@
35849 +/* SPDX-License-Identifier: GPL-2.0 */
35850 +/*
35851 + * Copyright (C) 2005-2021 Junjiro R. Okajima
35852 + *
35853 + * This program, aufs is free software; you can redistribute it and/or modify
35854 + * it under the terms of the GNU General Public License as published by
35855 + * the Free Software Foundation; either version 2 of the License, or
35856 + * (at your option) any later version.
35857 + *
35858 + * This program is distributed in the hope that it will be useful,
35859 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
35860 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
35861 + * GNU General Public License for more details.
35862 + *
35863 + * You should have received a copy of the GNU General Public License
35864 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
35865 + */
35866 +
35867 +/*
35868 + * whiteout for logical deletion and opaque directory
35869 + */
35870 +
35871 +#ifndef __AUFS_WHOUT_H__
35872 +#define __AUFS_WHOUT_H__
35873 +
35874 +#ifdef __KERNEL__
35875 +
35876 +#include "dir.h"
35877 +
35878 +/* whout.c */
35879 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name);
35880 +int au_wh_test(struct user_namespace *h_userns, struct path *h_ppath,
35881 +              struct qstr *wh_name, int try_sio);
35882 +int au_diropq_test(struct user_namespace *h_userns, struct path *h_path);
35883 +struct au_branch;
35884 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
35885 +                            struct qstr *prefix);
35886 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br);
35887 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
35888 +                       struct dentry *dentry);
35889 +int au_wh_init(struct au_branch *br, struct super_block *sb);
35890 +
35891 +/* diropq flags */
35892 +#define AuDiropq_CREATE        1
35893 +#define au_ftest_diropq(flags, name)   ((flags) & AuDiropq_##name)
35894 +#define au_fset_diropq(flags, name) \
35895 +       do { (flags) |= AuDiropq_##name; } while (0)
35896 +#define au_fclr_diropq(flags, name) \
35897 +       do { (flags) &= ~AuDiropq_##name; } while (0)
35898 +
35899 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
35900 +                            unsigned int flags);
35901 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
35902 +                         struct au_branch *br);
35903 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
35904 +                           struct dentry *h_parent);
35905 +
35906 +/* real rmdir for the whiteout-ed dir */
35907 +struct au_whtmp_rmdir {
35908 +       struct inode *dir;
35909 +       struct au_branch *br;
35910 +       struct dentry *wh_dentry;
35911 +       struct au_nhash whlist;
35912 +};
35913 +
35914 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp);
35915 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp);
35916 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
35917 +                  struct dentry *wh_dentry, struct au_nhash *whlist);
35918 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
35919 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args);
35920 +
35921 +/* ---------------------------------------------------------------------- */
35922 +
35923 +static inline struct dentry *au_diropq_create(struct dentry *dentry,
35924 +                                             aufs_bindex_t bindex)
35925 +{
35926 +       return au_diropq_sio(dentry, bindex, AuDiropq_CREATE);
35927 +}
35928 +
35929 +static inline int au_diropq_remove(struct dentry *dentry, aufs_bindex_t bindex)
35930 +{
35931 +       return PTR_ERR(au_diropq_sio(dentry, bindex, !AuDiropq_CREATE));
35932 +}
35933 +
35934 +#endif /* __KERNEL__ */
35935 +#endif /* __AUFS_WHOUT_H__ */
35936 diff -urN /usr/share/empty/fs/aufs/wkq.c linux/fs/aufs/wkq.c
35937 --- /usr/share/empty/fs/aufs/wkq.c      1970-01-01 01:00:00.000000000 +0100
35938 +++ linux/fs/aufs/wkq.c 2021-12-03 15:38:59.943313977 +0100
35939 @@ -0,0 +1,372 @@
35940 +// SPDX-License-Identifier: GPL-2.0
35941 +/*
35942 + * Copyright (C) 2005-2021 Junjiro R. Okajima
35943 + *
35944 + * This program, aufs is free software; you can redistribute it and/or modify
35945 + * it under the terms of the GNU General Public License as published by
35946 + * the Free Software Foundation; either version 2 of the License, or
35947 + * (at your option) any later version.
35948 + *
35949 + * This program is distributed in the hope that it will be useful,
35950 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
35951 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
35952 + * GNU General Public License for more details.
35953 + *
35954 + * You should have received a copy of the GNU General Public License
35955 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
35956 + */
35957 +
35958 +/*
35959 + * workqueue for asynchronous/super-io operations
35960 + * todo: try new credential scheme
35961 + */
35962 +
35963 +#include <linux/module.h>
35964 +#include "aufs.h"
35965 +
35966 +/* internal workqueue named AUFS_WKQ_NAME */
35967 +
35968 +static struct workqueue_struct *au_wkq;
35969 +
35970 +struct au_wkinfo {
35971 +       struct work_struct wk;
35972 +       struct kobject *kobj;
35973 +
35974 +       unsigned int flags; /* see wkq.h */
35975 +
35976 +       au_wkq_func_t func;
35977 +       void *args;
35978 +
35979 +#ifdef CONFIG_LOCKDEP
35980 +       int dont_check;
35981 +       struct held_lock **hlock;
35982 +#endif
35983 +
35984 +       struct completion *comp;
35985 +};
35986 +
35987 +/* ---------------------------------------------------------------------- */
35988 +/*
35989 + * Aufs passes some operations to the workqueue such as the internal copyup.
35990 + * This scheme looks rather unnatural for LOCKDEP debugging feature, since the
35991 + * job run by workqueue depends upon the locks acquired in the other task.
35992 + * Delegating a small operation to the workqueue, aufs passes its lockdep
35993 + * information too. And the job in the workqueue restores the info in order to
35994 + * pretend as if it acquired those locks. This is just to make LOCKDEP work
35995 + * correctly and expectedly.
35996 + */
35997 +
35998 +#ifndef CONFIG_LOCKDEP
35999 +AuStubInt0(au_wkq_lockdep_alloc, struct au_wkinfo *wkinfo);
36000 +AuStubVoid(au_wkq_lockdep_free, struct au_wkinfo *wkinfo);
36001 +AuStubVoid(au_wkq_lockdep_pre, struct au_wkinfo *wkinfo);
36002 +AuStubVoid(au_wkq_lockdep_post, struct au_wkinfo *wkinfo);
36003 +AuStubVoid(au_wkq_lockdep_init, struct au_wkinfo *wkinfo);
36004 +#else
36005 +static void au_wkq_lockdep_init(struct au_wkinfo *wkinfo)
36006 +{
36007 +       wkinfo->hlock = NULL;
36008 +       wkinfo->dont_check = 0;
36009 +}
36010 +
36011 +/*
36012 + * 1: matched
36013 + * 0: unmatched
36014 + */
36015 +static int au_wkq_lockdep_test(struct lock_class_key *key, const char *name)
36016 +{
36017 +       static DEFINE_SPINLOCK(spin);
36018 +       static struct {
36019 +               char *name;
36020 +               struct lock_class_key *key;
36021 +       } a[] = {
36022 +               { .name = "&sbinfo->si_rwsem" },
36023 +               { .name = "&finfo->fi_rwsem" },
36024 +               { .name = "&dinfo->di_rwsem" },
36025 +               { .name = "&iinfo->ii_rwsem" }
36026 +       };
36027 +       static int set;
36028 +       int i;
36029 +
36030 +       /* lockless read from 'set.' see below */
36031 +       if (set == ARRAY_SIZE(a)) {
36032 +               for (i = 0; i < ARRAY_SIZE(a); i++)
36033 +                       if (a[i].key == key)
36034 +                               goto match;
36035 +               goto unmatch;
36036 +       }
36037 +
36038 +       spin_lock(&spin);
36039 +       if (set)
36040 +               for (i = 0; i < ARRAY_SIZE(a); i++)
36041 +                       if (a[i].key == key) {
36042 +                               spin_unlock(&spin);
36043 +                               goto match;
36044 +                       }
36045 +       for (i = 0; i < ARRAY_SIZE(a); i++) {
36046 +               if (a[i].key) {
36047 +                       if (unlikely(a[i].key == key)) { /* rare but possible */
36048 +                               spin_unlock(&spin);
36049 +                               goto match;
36050 +                       } else
36051 +                               continue;
36052 +               }
36053 +               if (strstr(a[i].name, name)) {
36054 +                       /*
36055 +                        * the order of these three lines is important for the
36056 +                        * lockless read above.
36057 +                        */
36058 +                       a[i].key = key;
36059 +                       spin_unlock(&spin);
36060 +                       set++;
36061 +                       /* AuDbg("%d, %s\n", set, name); */
36062 +                       goto match;
36063 +               }
36064 +       }
36065 +       spin_unlock(&spin);
36066 +       goto unmatch;
36067 +
36068 +match:
36069 +       return 1;
36070 +unmatch:
36071 +       return 0;
36072 +}
36073 +
36074 +static int au_wkq_lockdep_alloc(struct au_wkinfo *wkinfo)
36075 +{
36076 +       int err, n;
36077 +       struct task_struct *curr;
36078 +       struct held_lock **hl, *held_locks, *p;
36079 +
36080 +       err = 0;
36081 +       curr = current;
36082 +       wkinfo->dont_check = lockdep_recursing(curr);
36083 +       if (wkinfo->dont_check)
36084 +               goto out;
36085 +       n = curr->lockdep_depth;
36086 +       if (!n)
36087 +               goto out;
36088 +
36089 +       err = -ENOMEM;
36090 +       wkinfo->hlock = kmalloc_array(n + 1, sizeof(*wkinfo->hlock), GFP_NOFS);
36091 +       if (unlikely(!wkinfo->hlock))
36092 +               goto out;
36093 +
36094 +       err = 0;
36095 +#if 0 /* left for debugging */
36096 +       if (0 && au_debug_test())
36097 +               lockdep_print_held_locks(curr);
36098 +#endif
36099 +       held_locks = curr->held_locks;
36100 +       hl = wkinfo->hlock;
36101 +       while (n--) {
36102 +               p = held_locks++;
36103 +               if (au_wkq_lockdep_test(p->instance->key, p->instance->name))
36104 +                       *hl++ = p;
36105 +       }
36106 +       *hl = NULL;
36107 +
36108 +out:
36109 +       return err;
36110 +}
36111 +
36112 +static void au_wkq_lockdep_free(struct au_wkinfo *wkinfo)
36113 +{
36114 +       au_kfree_try_rcu(wkinfo->hlock);
36115 +}
36116 +
36117 +static void au_wkq_lockdep_pre(struct au_wkinfo *wkinfo)
36118 +{
36119 +       struct held_lock *p, **hl = wkinfo->hlock;
36120 +       int subclass;
36121 +
36122 +       if (wkinfo->dont_check)
36123 +               lockdep_off();
36124 +       if (!hl)
36125 +               return;
36126 +       while ((p = *hl++)) { /* assignment */
36127 +               subclass = lockdep_hlock_class(p)->subclass;
36128 +               /* AuDbg("%s, %d\n", p->instance->name, subclass); */
36129 +               if (p->read)
36130 +                       rwsem_acquire_read(p->instance, subclass, 0,
36131 +                                          /*p->acquire_ip*/_RET_IP_);
36132 +               else
36133 +                       rwsem_acquire(p->instance, subclass, 0,
36134 +                                     /*p->acquire_ip*/_RET_IP_);
36135 +       }
36136 +}
36137 +
36138 +static void au_wkq_lockdep_post(struct au_wkinfo *wkinfo)
36139 +{
36140 +       struct held_lock *p, **hl = wkinfo->hlock;
36141 +
36142 +       if (wkinfo->dont_check)
36143 +               lockdep_on();
36144 +       if (!hl)
36145 +               return;
36146 +       while ((p = *hl++)) /* assignment */
36147 +               rwsem_release(p->instance, /*p->acquire_ip*/_RET_IP_);
36148 +}
36149 +#endif
36150 +
36151 +static void wkq_func(struct work_struct *wk)
36152 +{
36153 +       struct au_wkinfo *wkinfo = container_of(wk, struct au_wkinfo, wk);
36154 +
36155 +       AuDebugOn(!uid_eq(current_fsuid(), GLOBAL_ROOT_UID));
36156 +       AuDebugOn(rlimit(RLIMIT_FSIZE) != RLIM_INFINITY);
36157 +
36158 +       au_wkq_lockdep_pre(wkinfo);
36159 +       wkinfo->func(wkinfo->args);
36160 +       au_wkq_lockdep_post(wkinfo);
36161 +       if (au_ftest_wkq(wkinfo->flags, WAIT))
36162 +               complete(wkinfo->comp);
36163 +       else {
36164 +               kobject_put(wkinfo->kobj);
36165 +               module_put(THIS_MODULE); /* todo: ?? */
36166 +               au_kfree_rcu(wkinfo);
36167 +       }
36168 +}
36169 +
36170 +/*
36171 + * Since struct completion is large, try allocating it dynamically.
36172 + */
36173 +#define AuWkqCompDeclare(name) struct completion *comp = NULL
36174 +
36175 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
36176 +{
36177 +       *comp = kmalloc(sizeof(**comp), GFP_NOFS);
36178 +       if (*comp) {
36179 +               init_completion(*comp);
36180 +               wkinfo->comp = *comp;
36181 +               return 0;
36182 +       }
36183 +       return -ENOMEM;
36184 +}
36185 +
36186 +static void au_wkq_comp_free(struct completion *comp)
36187 +{
36188 +       au_kfree_rcu(comp);
36189 +}
36190 +
36191 +static void au_wkq_run(struct au_wkinfo *wkinfo)
36192 +{
36193 +       if (au_ftest_wkq(wkinfo->flags, NEST)) {
36194 +               if (au_wkq_test()) {
36195 +                       AuWarn1("wkq from wkq, unless silly-rename on NFS,"
36196 +                               " due to a dead dir by UDBA,"
36197 +                               " or async xino write?\n");
36198 +                       AuDebugOn(au_ftest_wkq(wkinfo->flags, WAIT));
36199 +               }
36200 +       } else
36201 +               au_dbg_verify_kthread();
36202 +
36203 +       if (au_ftest_wkq(wkinfo->flags, WAIT)) {
36204 +               INIT_WORK_ONSTACK(&wkinfo->wk, wkq_func);
36205 +               queue_work(au_wkq, &wkinfo->wk);
36206 +       } else {
36207 +               INIT_WORK(&wkinfo->wk, wkq_func);
36208 +               schedule_work(&wkinfo->wk);
36209 +       }
36210 +}
36211 +
36212 +/*
36213 + * Be careful. It is easy to make deadlock happen.
36214 + * processA: lock, wkq and wait
36215 + * processB: wkq and wait, lock in wkq
36216 + * --> deadlock
36217 + */
36218 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args)
36219 +{
36220 +       int err;
36221 +       AuWkqCompDeclare(comp);
36222 +       struct au_wkinfo wkinfo = {
36223 +               .flags  = flags,
36224 +               .func   = func,
36225 +               .args   = args
36226 +       };
36227 +
36228 +       err = au_wkq_comp_alloc(&wkinfo, &comp);
36229 +       if (unlikely(err))
36230 +               goto out;
36231 +       err = au_wkq_lockdep_alloc(&wkinfo);
36232 +       if (unlikely(err))
36233 +               goto out_comp;
36234 +       if (!err) {
36235 +               au_wkq_run(&wkinfo);
36236 +               /* no timeout, no interrupt */
36237 +               wait_for_completion(wkinfo.comp);
36238 +       }
36239 +       au_wkq_lockdep_free(&wkinfo);
36240 +
36241 +out_comp:
36242 +       au_wkq_comp_free(comp);
36243 +out:
36244 +       destroy_work_on_stack(&wkinfo.wk);
36245 +       return err;
36246 +}
36247 +
36248 +/*
36249 + * Note: dget/dput() in func for aufs dentries are not supported. It will be a
36250 + * problem in a concurrent umounting.
36251 + */
36252 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
36253 +                 unsigned int flags)
36254 +{
36255 +       int err;
36256 +       struct au_wkinfo *wkinfo;
36257 +
36258 +       atomic_inc(&au_sbi(sb)->si_nowait.nw_len);
36259 +
36260 +       /*
36261 +        * wkq_func() must free this wkinfo.
36262 +        * it highly depends upon the implementation of workqueue.
36263 +        */
36264 +       err = 0;
36265 +       wkinfo = kmalloc(sizeof(*wkinfo), GFP_NOFS);
36266 +       if (wkinfo) {
36267 +               wkinfo->kobj = &au_sbi(sb)->si_kobj;
36268 +               wkinfo->flags = flags & ~AuWkq_WAIT;
36269 +               wkinfo->func = func;
36270 +               wkinfo->args = args;
36271 +               wkinfo->comp = NULL;
36272 +               au_wkq_lockdep_init(wkinfo);
36273 +               kobject_get(wkinfo->kobj);
36274 +               __module_get(THIS_MODULE); /* todo: ?? */
36275 +
36276 +               au_wkq_run(wkinfo);
36277 +       } else {
36278 +               err = -ENOMEM;
36279 +               au_nwt_done(&au_sbi(sb)->si_nowait);
36280 +       }
36281 +
36282 +       return err;
36283 +}
36284 +
36285 +/* ---------------------------------------------------------------------- */
36286 +
36287 +void au_nwt_init(struct au_nowait_tasks *nwt)
36288 +{
36289 +       atomic_set(&nwt->nw_len, 0);
36290 +       /* smp_mb(); */ /* atomic_set */
36291 +       init_waitqueue_head(&nwt->nw_wq);
36292 +}
36293 +
36294 +void au_wkq_fin(void)
36295 +{
36296 +       destroy_workqueue(au_wkq);
36297 +}
36298 +
36299 +int __init au_wkq_init(void)
36300 +{
36301 +       int err;
36302 +
36303 +       err = 0;
36304 +       au_wkq = alloc_workqueue(AUFS_WKQ_NAME, 0, WQ_DFL_ACTIVE);
36305 +       if (IS_ERR(au_wkq))
36306 +               err = PTR_ERR(au_wkq);
36307 +       else if (!au_wkq)
36308 +               err = -ENOMEM;
36309 +
36310 +       return err;
36311 +}
36312 diff -urN /usr/share/empty/fs/aufs/wkq.h linux/fs/aufs/wkq.h
36313 --- /usr/share/empty/fs/aufs/wkq.h      1970-01-01 01:00:00.000000000 +0100
36314 +++ linux/fs/aufs/wkq.h 2021-12-03 15:38:59.943313977 +0100
36315 @@ -0,0 +1,89 @@
36316 +/* SPDX-License-Identifier: GPL-2.0 */
36317 +/*
36318 + * Copyright (C) 2005-2021 Junjiro R. Okajima
36319 + *
36320 + * This program, aufs is free software; you can redistribute it and/or modify
36321 + * it under the terms of the GNU General Public License as published by
36322 + * the Free Software Foundation; either version 2 of the License, or
36323 + * (at your option) any later version.
36324 + *
36325 + * This program is distributed in the hope that it will be useful,
36326 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
36327 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
36328 + * GNU General Public License for more details.
36329 + *
36330 + * You should have received a copy of the GNU General Public License
36331 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
36332 + */
36333 +
36334 +/*
36335 + * workqueue for asynchronous/super-io operations
36336 + * todo: try new credentials management scheme
36337 + */
36338 +
36339 +#ifndef __AUFS_WKQ_H__
36340 +#define __AUFS_WKQ_H__
36341 +
36342 +#ifdef __KERNEL__
36343 +
36344 +#include <linux/wait.h>
36345 +
36346 +struct super_block;
36347 +
36348 +/* ---------------------------------------------------------------------- */
36349 +
36350 +/*
36351 + * in the next operation, wait for the 'nowait' tasks in system-wide workqueue
36352 + */
36353 +struct au_nowait_tasks {
36354 +       atomic_t                nw_len;
36355 +       wait_queue_head_t       nw_wq;
36356 +};
36357 +
36358 +/* ---------------------------------------------------------------------- */
36359 +
36360 +typedef void (*au_wkq_func_t)(void *args);
36361 +
36362 +/* wkq flags */
36363 +#define AuWkq_WAIT     1
36364 +#define AuWkq_NEST     (1 << 1)
36365 +#define au_ftest_wkq(flags, name)      ((flags) & AuWkq_##name)
36366 +#define au_fset_wkq(flags, name) \
36367 +       do { (flags) |= AuWkq_##name; } while (0)
36368 +#define au_fclr_wkq(flags, name) \
36369 +       do { (flags) &= ~AuWkq_##name; } while (0)
36370 +
36371 +/* wkq.c */
36372 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args);
36373 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
36374 +                 unsigned int flags);
36375 +void au_nwt_init(struct au_nowait_tasks *nwt);
36376 +int __init au_wkq_init(void);
36377 +void au_wkq_fin(void);
36378 +
36379 +/* ---------------------------------------------------------------------- */
36380 +
36381 +static inline int au_wkq_test(void)
36382 +{
36383 +       return current->flags & PF_WQ_WORKER;
36384 +}
36385 +
36386 +static inline int au_wkq_wait(au_wkq_func_t func, void *args)
36387 +{
36388 +       return au_wkq_do_wait(AuWkq_WAIT, func, args);
36389 +}
36390 +
36391 +static inline void au_nwt_done(struct au_nowait_tasks *nwt)
36392 +{
36393 +       if (atomic_dec_and_test(&nwt->nw_len))
36394 +               wake_up_all(&nwt->nw_wq);
36395 +}
36396 +
36397 +static inline int au_nwt_flush(struct au_nowait_tasks *nwt)
36398 +{
36399 +       wait_event(nwt->nw_wq, !atomic_read(&nwt->nw_len));
36400 +       return 0;
36401 +}
36402 +
36403 +#endif /* __KERNEL__ */
36404 +#endif /* __AUFS_WKQ_H__ */
36405 diff -urN /usr/share/empty/fs/aufs/xattr.c linux/fs/aufs/xattr.c
36406 --- /usr/share/empty/fs/aufs/xattr.c    1970-01-01 01:00:00.000000000 +0100
36407 +++ linux/fs/aufs/xattr.c       2021-12-03 15:40:58.236647297 +0100
36408 @@ -0,0 +1,368 @@
36409 +// SPDX-License-Identifier: GPL-2.0
36410 +/*
36411 + * Copyright (C) 2014-2021 Junjiro R. Okajima
36412 + *
36413 + * This program, aufs is free software; you can redistribute it and/or modify
36414 + * it under the terms of the GNU General Public License as published by
36415 + * the Free Software Foundation; either version 2 of the License, or
36416 + * (at your option) any later version.
36417 + *
36418 + * This program is distributed in the hope that it will be useful,
36419 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
36420 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
36421 + * GNU General Public License for more details.
36422 + *
36423 + * You should have received a copy of the GNU General Public License
36424 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
36425 + */
36426 +
36427 +/*
36428 + * handling xattr functions
36429 + */
36430 +
36431 +#include <linux/fs.h>
36432 +#include <linux/posix_acl_xattr.h>
36433 +#include <linux/xattr.h>
36434 +#include "aufs.h"
36435 +
36436 +static int au_xattr_ignore(int err, char *name, unsigned int ignore_flags)
36437 +{
36438 +       if (!ignore_flags)
36439 +               goto out;
36440 +       switch (err) {
36441 +       case -ENOMEM:
36442 +       case -EDQUOT:
36443 +               goto out;
36444 +       }
36445 +
36446 +       if ((ignore_flags & AuBrAttr_ICEX) == AuBrAttr_ICEX) {
36447 +               err = 0;
36448 +               goto out;
36449 +       }
36450 +
36451 +#define cmp(brattr, prefix) do {                                       \
36452 +               if (!strncmp(name, XATTR_##prefix##_PREFIX,             \
36453 +                            XATTR_##prefix##_PREFIX_LEN)) {            \
36454 +                       if (ignore_flags & AuBrAttr_ICEX_##brattr)      \
36455 +                               err = 0;                                \
36456 +                       goto out;                                       \
36457 +               }                                                       \
36458 +       } while (0)
36459 +
36460 +       cmp(SEC, SECURITY);
36461 +       cmp(SYS, SYSTEM);
36462 +       cmp(TR, TRUSTED);
36463 +       cmp(USR, USER);
36464 +#undef cmp
36465 +
36466 +       if (ignore_flags & AuBrAttr_ICEX_OTH)
36467 +               err = 0;
36468 +
36469 +out:
36470 +       return err;
36471 +}
36472 +
36473 +static const int au_xattr_out_of_list = AuBrAttr_ICEX_OTH << 1;
36474 +
36475 +static int au_do_cpup_xattr(struct path *h_dst, struct path *h_src,
36476 +                           char *name, char **buf, unsigned int ignore_flags,
36477 +                           unsigned int verbose)
36478 +{
36479 +       int err;
36480 +       ssize_t ssz;
36481 +       struct inode *h_idst;
36482 +       struct dentry *h_dst_dentry, *h_src_dentry;
36483 +       struct user_namespace *h_dst_userns, *h_src_userns;
36484 +
36485 +       h_src_userns = mnt_user_ns(h_src->mnt);
36486 +       h_src_dentry = h_src->dentry;
36487 +       ssz = vfs_getxattr_alloc(h_src_userns, h_src_dentry, name, buf, 0,
36488 +                                GFP_NOFS);
36489 +       err = ssz;
36490 +       if (unlikely(err <= 0)) {
36491 +               if (err == -ENODATA
36492 +                   || (err == -EOPNOTSUPP
36493 +                       && ((ignore_flags & au_xattr_out_of_list)
36494 +                           || (au_test_nfs_noacl(d_inode(h_src_dentry))
36495 +                               && (!strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS)
36496 +                                   || !strcmp(name,
36497 +                                              XATTR_NAME_POSIX_ACL_DEFAULT))))
36498 +                           ))
36499 +                       err = 0;
36500 +               if (err && (verbose || au_debug_test()))
36501 +                       pr_err("%s, err %d\n", name, err);
36502 +               goto out;
36503 +       }
36504 +
36505 +       /* unlock it temporary */
36506 +       h_dst_userns = mnt_user_ns(h_dst->mnt);
36507 +       h_dst_dentry = h_dst->dentry;
36508 +       h_idst = d_inode(h_dst_dentry);
36509 +       inode_unlock(h_idst);
36510 +       err = vfsub_setxattr(h_dst_userns, h_dst_dentry, name, *buf, ssz,
36511 +                            /*flags*/0);
36512 +       inode_lock_nested(h_idst, AuLsc_I_CHILD2);
36513 +       if (unlikely(err)) {
36514 +               if (verbose || au_debug_test())
36515 +                       pr_err("%s, err %d\n", name, err);
36516 +               err = au_xattr_ignore(err, name, ignore_flags);
36517 +       }
36518 +
36519 +out:
36520 +       return err;
36521 +}
36522 +
36523 +int au_cpup_xattr(struct path *h_dst, struct path *h_src, int ignore_flags,
36524 +                 unsigned int verbose)
36525 +{
36526 +       int err, unlocked, acl_access, acl_default;
36527 +       ssize_t ssz;
36528 +       struct dentry *h_dst_dentry, *h_src_dentry;
36529 +       struct inode *h_isrc, *h_idst;
36530 +       char *value, *p, *o, *e;
36531 +
36532 +       /* try stopping to update the source inode while we are referencing */
36533 +       /* there should not be the parent-child relationship between them */
36534 +       h_dst_dentry = h_dst->dentry;
36535 +       h_idst = d_inode(h_dst_dentry);
36536 +       h_src_dentry = h_src->dentry;
36537 +       h_isrc = d_inode(h_src_dentry);
36538 +       inode_unlock(h_idst);
36539 +       inode_lock_shared_nested(h_isrc, AuLsc_I_CHILD);
36540 +       inode_lock_nested(h_idst, AuLsc_I_CHILD2);
36541 +       unlocked = 0;
36542 +
36543 +       /* some filesystems don't list POSIX ACL, for example tmpfs */
36544 +       ssz = vfs_listxattr(h_src_dentry, NULL, 0);
36545 +       err = ssz;
36546 +       if (unlikely(err < 0)) {
36547 +               AuTraceErr(err);
36548 +               if (err == -ENODATA
36549 +                   || err == -EOPNOTSUPP)
36550 +                       err = 0;        /* ignore */
36551 +               goto out;
36552 +       }
36553 +
36554 +       err = 0;
36555 +       p = NULL;
36556 +       o = NULL;
36557 +       if (ssz) {
36558 +               err = -ENOMEM;
36559 +               p = kmalloc(ssz, GFP_NOFS);
36560 +               o = p;
36561 +               if (unlikely(!p))
36562 +                       goto out;
36563 +               err = vfs_listxattr(h_src_dentry, p, ssz);
36564 +       }
36565 +       inode_unlock_shared(h_isrc);
36566 +       unlocked = 1;
36567 +       AuDbg("err %d, ssz %zd\n", err, ssz);
36568 +       if (unlikely(err < 0))
36569 +               goto out_free;
36570 +
36571 +       err = 0;
36572 +       e = p + ssz;
36573 +       value = NULL;
36574 +       acl_access = 0;
36575 +       acl_default = 0;
36576 +       while (!err && p < e) {
36577 +               acl_access |= !strncmp(p, XATTR_NAME_POSIX_ACL_ACCESS,
36578 +                                      sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1);
36579 +               acl_default |= !strncmp(p, XATTR_NAME_POSIX_ACL_DEFAULT,
36580 +                                       sizeof(XATTR_NAME_POSIX_ACL_DEFAULT)
36581 +                                       - 1);
36582 +               err = au_do_cpup_xattr(h_dst, h_src, p, &value, ignore_flags,
36583 +                                      verbose);
36584 +               p += strlen(p) + 1;
36585 +       }
36586 +       AuTraceErr(err);
36587 +       ignore_flags |= au_xattr_out_of_list;
36588 +       if (!err && !acl_access) {
36589 +               err = au_do_cpup_xattr(h_dst, h_src,
36590 +                                      XATTR_NAME_POSIX_ACL_ACCESS, &value,
36591 +                                      ignore_flags, verbose);
36592 +               AuTraceErr(err);
36593 +       }
36594 +       if (!err && !acl_default) {
36595 +               err = au_do_cpup_xattr(h_dst, h_src,
36596 +                                      XATTR_NAME_POSIX_ACL_DEFAULT, &value,
36597 +                                      ignore_flags, verbose);
36598 +               AuTraceErr(err);
36599 +       }
36600 +
36601 +       au_kfree_try_rcu(value);
36602 +
36603 +out_free:
36604 +       au_kfree_try_rcu(o);
36605 +out:
36606 +       if (!unlocked)
36607 +               inode_unlock_shared(h_isrc);
36608 +       AuTraceErr(err);
36609 +       return err;
36610 +}
36611 +
36612 +/* ---------------------------------------------------------------------- */
36613 +
36614 +static int au_smack_reentering(struct super_block *sb)
36615 +{
36616 +#if IS_ENABLED(CONFIG_SECURITY_SMACK) || IS_ENABLED(CONFIG_SECURITY_SELINUX)
36617 +       /*
36618 +        * as a part of lookup, smack_d_instantiate() is called, and it calls
36619 +        * i_op->getxattr(). ouch.
36620 +        */
36621 +       return si_pid_test(sb);
36622 +#else
36623 +       return 0;
36624 +#endif
36625 +}
36626 +
36627 +enum {
36628 +       AU_XATTR_LIST,
36629 +       AU_XATTR_GET
36630 +};
36631 +
36632 +struct au_lgxattr {
36633 +       int type;
36634 +       union {
36635 +               struct {
36636 +                       char    *list;
36637 +                       size_t  size;
36638 +               } list;
36639 +               struct {
36640 +                       const char      *name;
36641 +                       void            *value;
36642 +                       size_t          size;
36643 +               } get;
36644 +       } u;
36645 +};
36646 +
36647 +static ssize_t au_lgxattr(struct dentry *dentry, struct inode *inode,
36648 +                         struct au_lgxattr *arg)
36649 +{
36650 +       ssize_t err;
36651 +       int reenter;
36652 +       struct path h_path;
36653 +       struct super_block *sb;
36654 +
36655 +       sb = dentry->d_sb;
36656 +       reenter = au_smack_reentering(sb);
36657 +       if (!reenter) {
36658 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
36659 +               if (unlikely(err))
36660 +                       goto out;
36661 +       }
36662 +       err = au_h_path_getattr(dentry, inode, /*force*/1, &h_path, reenter);
36663 +       if (unlikely(err))
36664 +               goto out_si;
36665 +       if (unlikely(!h_path.dentry))
36666 +               /* illegally overlapped or something */
36667 +               goto out_di; /* pretending success */
36668 +
36669 +       /* always topmost entry only */
36670 +       switch (arg->type) {
36671 +       case AU_XATTR_LIST:
36672 +               err = vfs_listxattr(h_path.dentry,
36673 +                                   arg->u.list.list, arg->u.list.size);
36674 +               break;
36675 +       case AU_XATTR_GET:
36676 +               AuDebugOn(d_is_negative(h_path.dentry));
36677 +               err = vfs_getxattr(mnt_user_ns(h_path.mnt), h_path.dentry,
36678 +                                  arg->u.get.name, arg->u.get.value,
36679 +                                  arg->u.get.size);
36680 +               break;
36681 +       }
36682 +
36683 +out_di:
36684 +       if (!reenter)
36685 +               di_read_unlock(dentry, AuLock_IR);
36686 +out_si:
36687 +       if (!reenter)
36688 +               si_read_unlock(sb);
36689 +out:
36690 +       AuTraceErr(err);
36691 +       return err;
36692 +}
36693 +
36694 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size)
36695 +{
36696 +       struct au_lgxattr arg = {
36697 +               .type = AU_XATTR_LIST,
36698 +               .u.list = {
36699 +                       .list   = list,
36700 +                       .size   = size
36701 +               },
36702 +       };
36703 +
36704 +       return au_lgxattr(dentry, /*inode*/NULL, &arg);
36705 +}
36706 +
36707 +static ssize_t au_getxattr(struct dentry *dentry, struct inode *inode,
36708 +                          const char *name, void *value, size_t size)
36709 +{
36710 +       struct au_lgxattr arg = {
36711 +               .type = AU_XATTR_GET,
36712 +               .u.get = {
36713 +                       .name   = name,
36714 +                       .value  = value,
36715 +                       .size   = size
36716 +               },
36717 +       };
36718 +
36719 +       return au_lgxattr(dentry, inode, &arg);
36720 +}
36721 +
36722 +static int au_setxattr(struct dentry *dentry, struct inode *inode,
36723 +                      const char *name, const void *value, size_t size,
36724 +                      int flags)
36725 +{
36726 +       struct au_sxattr arg = {
36727 +               .type = AU_XATTR_SET,
36728 +               .u.set = {
36729 +                       .name   = name,
36730 +                       .value  = value,
36731 +                       .size   = size,
36732 +                       .flags  = flags
36733 +               },
36734 +       };
36735 +
36736 +       return au_sxattr(dentry, inode, &arg);
36737 +}
36738 +
36739 +/* ---------------------------------------------------------------------- */
36740 +
36741 +static int au_xattr_get(const struct xattr_handler *handler,
36742 +                       struct dentry *dentry, struct inode *inode,
36743 +                       const char *name, void *buffer, size_t size)
36744 +{
36745 +       return au_getxattr(dentry, inode, name, buffer, size);
36746 +}
36747 +
36748 +static int au_xattr_set(const struct xattr_handler *handler,
36749 +                       struct user_namespace *userns,
36750 +                       struct dentry *dentry, struct inode *inode,
36751 +                       const char *name, const void *value, size_t size,
36752 +                       int flags)
36753 +{
36754 +       return au_setxattr(dentry, inode, name, value, size, flags);
36755 +}
36756 +
36757 +static const struct xattr_handler au_xattr_handler = {
36758 +       .name   = "",
36759 +       .prefix = "",
36760 +       .get    = au_xattr_get,
36761 +       .set    = au_xattr_set
36762 +};
36763 +
36764 +static const struct xattr_handler *au_xattr_handlers[] = {
36765 +#ifdef CONFIG_FS_POSIX_ACL
36766 +       &posix_acl_access_xattr_handler,
36767 +       &posix_acl_default_xattr_handler,
36768 +#endif
36769 +       &au_xattr_handler, /* must be last */
36770 +       NULL
36771 +};
36772 +
36773 +void au_xattr_init(struct super_block *sb)
36774 +{
36775 +       sb->s_xattr = au_xattr_handlers;
36776 +}
36777 diff -urN /usr/share/empty/fs/aufs/xino.c linux/fs/aufs/xino.c
36778 --- /usr/share/empty/fs/aufs/xino.c     1970-01-01 01:00:00.000000000 +0100
36779 +++ linux/fs/aufs/xino.c        2021-12-03 15:40:58.236647297 +0100
36780 @@ -0,0 +1,1926 @@
36781 +// SPDX-License-Identifier: GPL-2.0
36782 +/*
36783 + * Copyright (C) 2005-2021 Junjiro R. Okajima
36784 + *
36785 + * This program, aufs is free software; you can redistribute it and/or modify
36786 + * it under the terms of the GNU General Public License as published by
36787 + * the Free Software Foundation; either version 2 of the License, or
36788 + * (at your option) any later version.
36789 + *
36790 + * This program is distributed in the hope that it will be useful,
36791 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
36792 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
36793 + * GNU General Public License for more details.
36794 + *
36795 + * You should have received a copy of the GNU General Public License
36796 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
36797 + */
36798 +
36799 +/*
36800 + * external inode number translation table and bitmap
36801 + *
36802 + * things to consider
36803 + * - the lifetime
36804 + *   + au_xino object
36805 + *   + XINO files (xino, xib, xigen)
36806 + *   + dynamic debugfs entries (xiN)
36807 + *   + static debugfs entries (xib, xigen)
36808 + *   + static sysfs entry (xi_path)
36809 + * - several entry points to handle them.
36810 + *   + mount(2) without xino option (default)
36811 + *   + mount(2) with xino option
36812 + *   + mount(2) with noxino option
36813 + *   + umount(2)
36814 + *   + remount with add/del branches
36815 + *   + remount with xino/noxino options
36816 + */
36817 +
36818 +#include <linux/seq_file.h>
36819 +#include <linux/statfs.h>
36820 +#include "aufs.h"
36821 +
36822 +static aufs_bindex_t sbr_find_shared(struct super_block *sb, aufs_bindex_t btop,
36823 +                                    aufs_bindex_t bbot,
36824 +                                    struct super_block *h_sb)
36825 +{
36826 +       /* todo: try binary-search if the branches are many */
36827 +       for (; btop <= bbot; btop++)
36828 +               if (h_sb == au_sbr_sb(sb, btop))
36829 +                       return btop;
36830 +       return -1;
36831 +}
36832 +
36833 +/*
36834 + * find another branch who is on the same filesystem of the specified
36835 + * branch{@btgt}. search until @bbot.
36836 + */
36837 +static aufs_bindex_t is_sb_shared(struct super_block *sb, aufs_bindex_t btgt,
36838 +                                 aufs_bindex_t bbot)
36839 +{
36840 +       aufs_bindex_t bindex;
36841 +       struct super_block *tgt_sb;
36842 +
36843 +       tgt_sb = au_sbr_sb(sb, btgt);
36844 +       bindex = sbr_find_shared(sb, /*btop*/0, btgt - 1, tgt_sb);
36845 +       if (bindex < 0)
36846 +               bindex = sbr_find_shared(sb, btgt + 1, bbot, tgt_sb);
36847 +
36848 +       return bindex;
36849 +}
36850 +
36851 +/* ---------------------------------------------------------------------- */
36852 +
36853 +/*
36854 + * stop unnecessary notify events at creating xino files
36855 + */
36856 +
36857 +aufs_bindex_t au_xi_root(struct super_block *sb, struct dentry *dentry)
36858 +{
36859 +       aufs_bindex_t bfound, bindex, bbot;
36860 +       struct dentry *parent;
36861 +       struct au_branch *br;
36862 +
36863 +       bfound = -1;
36864 +       parent = dentry->d_parent; /* safe d_parent access */
36865 +       bbot = au_sbbot(sb);
36866 +       for (bindex = 0; bindex <= bbot; bindex++) {
36867 +               br = au_sbr(sb, bindex);
36868 +               if (au_br_dentry(br) == parent) {
36869 +                       bfound = bindex;
36870 +                       break;
36871 +               }
36872 +       }
36873 +
36874 +       AuDbg("bfound b%d\n", bfound);
36875 +       return bfound;
36876 +}
36877 +
36878 +struct au_xino_lock_dir {
36879 +       struct au_hinode *hdir;
36880 +       struct dentry *parent;
36881 +       struct inode *dir;
36882 +};
36883 +
36884 +static struct dentry *au_dget_parent_lock(struct dentry *dentry,
36885 +                                         unsigned int lsc)
36886 +{
36887 +       struct dentry *parent;
36888 +       struct inode *dir;
36889 +
36890 +       parent = dget_parent(dentry);
36891 +       dir = d_inode(parent);
36892 +       inode_lock_nested(dir, lsc);
36893 +#if 0 /* it should not happen */
36894 +       spin_lock(&dentry->d_lock);
36895 +       if (unlikely(dentry->d_parent != parent)) {
36896 +               spin_unlock(&dentry->d_lock);
36897 +               inode_unlock(dir);
36898 +               dput(parent);
36899 +               parent = NULL;
36900 +               goto out;
36901 +       }
36902 +       spin_unlock(&dentry->d_lock);
36903 +
36904 +out:
36905 +#endif
36906 +       return parent;
36907 +}
36908 +
36909 +static void au_xino_lock_dir(struct super_block *sb, struct path *xipath,
36910 +                            struct au_xino_lock_dir *ldir)
36911 +{
36912 +       aufs_bindex_t bindex;
36913 +
36914 +       ldir->hdir = NULL;
36915 +       bindex = au_xi_root(sb, xipath->dentry);
36916 +       if (bindex >= 0) {
36917 +               /* rw branch root */
36918 +               ldir->hdir = au_hi(d_inode(sb->s_root), bindex);
36919 +               au_hn_inode_lock_nested(ldir->hdir, AuLsc_I_PARENT);
36920 +       } else {
36921 +               /* other */
36922 +               ldir->parent = au_dget_parent_lock(xipath->dentry,
36923 +                                                  AuLsc_I_PARENT);
36924 +               ldir->dir = d_inode(ldir->parent);
36925 +       }
36926 +}
36927 +
36928 +static void au_xino_unlock_dir(struct au_xino_lock_dir *ldir)
36929 +{
36930 +       if (ldir->hdir)
36931 +               au_hn_inode_unlock(ldir->hdir);
36932 +       else {
36933 +               inode_unlock(ldir->dir);
36934 +               dput(ldir->parent);
36935 +       }
36936 +}
36937 +
36938 +/* ---------------------------------------------------------------------- */
36939 +
36940 +/*
36941 + * create and set a new xino file
36942 + */
36943 +struct file *au_xino_create(struct super_block *sb, char *fpath, int silent,
36944 +                           int wbrtop)
36945 +{
36946 +       struct file *file;
36947 +       struct dentry *h_parent, *d;
36948 +       struct inode *h_dir, *inode;
36949 +       int err;
36950 +       static DEFINE_MUTEX(mtx);
36951 +
36952 +       /*
36953 +        * at mount-time, and the xino file is the default path,
36954 +        * hnotify is disabled so we have no notify events to ignore.
36955 +        * when a user specified the xino, we cannot get au_hdir to be ignored.
36956 +        */
36957 +       if (!wbrtop)
36958 +               mutex_lock(&mtx);
36959 +       file = vfsub_filp_open(fpath, O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
36960 +                              /* | __FMODE_NONOTIFY */,
36961 +                              0666);
36962 +       if (IS_ERR(file)) {
36963 +               if (!wbrtop)
36964 +                       mutex_unlock(&mtx);
36965 +               if (!silent)
36966 +                       pr_err("open %s(%ld)\n", fpath, PTR_ERR(file));
36967 +               return file;
36968 +       }
36969 +
36970 +       /* keep file count */
36971 +       err = 0;
36972 +       d = file->f_path.dentry;
36973 +       h_parent = au_dget_parent_lock(d, AuLsc_I_PARENT);
36974 +       if (!wbrtop)
36975 +               mutex_unlock(&mtx);
36976 +       /* mnt_want_write() is unnecessary here */
36977 +       h_dir = d_inode(h_parent);
36978 +       inode = file_inode(file);
36979 +       /* no delegation since it is just created */
36980 +       if (inode->i_nlink)
36981 +               err = vfsub_unlink(h_dir, &file->f_path, /*delegated*/NULL,
36982 +                                  /*force*/0);
36983 +       inode_unlock(h_dir);
36984 +       dput(h_parent);
36985 +       if (unlikely(err)) {
36986 +               if (!silent)
36987 +                       pr_err("unlink %s(%d)\n", fpath, err);
36988 +               goto out;
36989 +       }
36990 +
36991 +       err = -EINVAL;
36992 +       if (unlikely(sb == d->d_sb)) {
36993 +               if (!silent)
36994 +                       pr_err("%s must be outside\n", fpath);
36995 +               goto out;
36996 +       }
36997 +       if (unlikely(au_test_fs_bad_xino(d->d_sb))) {
36998 +               if (!silent)
36999 +                       pr_err("xino doesn't support %s(%s)\n",
37000 +                              fpath, au_sbtype(d->d_sb));
37001 +               goto out;
37002 +       }
37003 +       return file; /* success */
37004 +
37005 +out:
37006 +       fput(file);
37007 +       file = ERR_PTR(err);
37008 +       return file;
37009 +}
37010 +
37011 +/*
37012 + * create a new xinofile at the same place/path as @base.
37013 + */
37014 +struct file *au_xino_create2(struct super_block *sb, struct path *base,
37015 +                            struct file *copy_src)
37016 +{
37017 +       struct file *file;
37018 +       struct dentry *dentry;
37019 +       struct inode *dir, *delegated;
37020 +       struct qstr *name;
37021 +       struct path ppath, path;
37022 +       int err, do_unlock;
37023 +       struct au_xino_lock_dir ldir;
37024 +
37025 +       do_unlock = 1;
37026 +       au_xino_lock_dir(sb, base, &ldir);
37027 +       dentry = base->dentry;
37028 +       ppath.dentry = dentry->d_parent; /* dir inode is locked */
37029 +       ppath.mnt = base->mnt;
37030 +       dir = d_inode(ppath.dentry);
37031 +       IMustLock(dir);
37032 +
37033 +       name = &dentry->d_name;
37034 +       path.dentry = vfsub_lookup_one_len(name->name, &ppath, name->len);
37035 +       if (IS_ERR(path.dentry)) {
37036 +               file = (void *)path.dentry;
37037 +               pr_err("%pd lookup err %ld\n", dentry, PTR_ERR(path.dentry));
37038 +               goto out;
37039 +       }
37040 +
37041 +       /* no need to mnt_want_write() since we call dentry_open() later */
37042 +       err = vfs_create(mnt_user_ns(base->mnt), dir, path.dentry, 0666, NULL);
37043 +       if (unlikely(err)) {
37044 +               file = ERR_PTR(err);
37045 +               pr_err("%pd create err %d\n", dentry, err);
37046 +               goto out_dput;
37047 +       }
37048 +
37049 +       path.mnt = base->mnt;
37050 +       file = vfsub_dentry_open(&path,
37051 +                                O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
37052 +                                /* | __FMODE_NONOTIFY */);
37053 +       if (IS_ERR(file)) {
37054 +               pr_err("%pd open err %ld\n", dentry, PTR_ERR(file));
37055 +               goto out_dput;
37056 +       }
37057 +
37058 +       delegated = NULL;
37059 +       err = vfsub_unlink(dir, &file->f_path, &delegated, /*force*/0);
37060 +       au_xino_unlock_dir(&ldir);
37061 +       do_unlock = 0;
37062 +       if (unlikely(err == -EWOULDBLOCK)) {
37063 +               pr_warn("cannot retry for NFSv4 delegation"
37064 +                       " for an internal unlink\n");
37065 +               iput(delegated);
37066 +       }
37067 +       if (unlikely(err)) {
37068 +               pr_err("%pd unlink err %d\n", dentry, err);
37069 +               goto out_fput;
37070 +       }
37071 +
37072 +       if (copy_src) {
37073 +               /* no one can touch copy_src xino */
37074 +               err = au_copy_file(file, copy_src, vfsub_f_size_read(copy_src));
37075 +               if (unlikely(err)) {
37076 +                       pr_err("%pd copy err %d\n", dentry, err);
37077 +                       goto out_fput;
37078 +               }
37079 +       }
37080 +       goto out_dput; /* success */
37081 +
37082 +out_fput:
37083 +       fput(file);
37084 +       file = ERR_PTR(err);
37085 +out_dput:
37086 +       dput(path.dentry);
37087 +out:
37088 +       if (do_unlock)
37089 +               au_xino_unlock_dir(&ldir);
37090 +       return file;
37091 +}
37092 +
37093 +struct file *au_xino_file1(struct au_xino *xi)
37094 +{
37095 +       struct file *file;
37096 +       unsigned int u, nfile;
37097 +
37098 +       file = NULL;
37099 +       nfile = xi->xi_nfile;
37100 +       for (u = 0; u < nfile; u++) {
37101 +               file = xi->xi_file[u];
37102 +               if (file)
37103 +                       break;
37104 +       }
37105 +
37106 +       return file;
37107 +}
37108 +
37109 +static int au_xino_file_set(struct au_xino *xi, int idx, struct file *file)
37110 +{
37111 +       int err;
37112 +       struct file *f;
37113 +       void *p;
37114 +
37115 +       if (file)
37116 +               get_file(file);
37117 +
37118 +       err = 0;
37119 +       f = NULL;
37120 +       if (idx < xi->xi_nfile) {
37121 +               f = xi->xi_file[idx];
37122 +               if (f)
37123 +                       fput(f);
37124 +       } else {
37125 +               p = au_kzrealloc(xi->xi_file,
37126 +                                sizeof(*xi->xi_file) * xi->xi_nfile,
37127 +                                sizeof(*xi->xi_file) * (idx + 1),
37128 +                                GFP_NOFS, /*may_shrink*/0);
37129 +               if (p) {
37130 +                       MtxMustLock(&xi->xi_mtx);
37131 +                       xi->xi_file = p;
37132 +                       xi->xi_nfile = idx + 1;
37133 +               } else {
37134 +                       err = -ENOMEM;
37135 +                       if (file)
37136 +                               fput(file);
37137 +                       goto out;
37138 +               }
37139 +       }
37140 +       xi->xi_file[idx] = file;
37141 +
37142 +out:
37143 +       return err;
37144 +}
37145 +
37146 +/*
37147 + * if @xinew->xi is not set, then create new xigen file.
37148 + */
37149 +struct file *au_xi_new(struct super_block *sb, struct au_xi_new *xinew)
37150 +{
37151 +       struct file *file;
37152 +       int err;
37153 +
37154 +       SiMustAnyLock(sb);
37155 +
37156 +       file = au_xino_create2(sb, xinew->base, xinew->copy_src);
37157 +       if (IS_ERR(file)) {
37158 +               err = PTR_ERR(file);
37159 +               pr_err("%s[%d], err %d\n",
37160 +                      xinew->xi ? "xino" : "xigen",
37161 +                      xinew->idx, err);
37162 +               goto out;
37163 +       }
37164 +
37165 +       if (xinew->xi)
37166 +               err = au_xino_file_set(xinew->xi, xinew->idx, file);
37167 +       else {
37168 +               BUG();
37169 +               /* todo: make xigen file an array */
37170 +               /* err = au_xigen_file_set(sb, xinew->idx, file); */
37171 +       }
37172 +       fput(file);
37173 +       if (unlikely(err))
37174 +               file = ERR_PTR(err);
37175 +
37176 +out:
37177 +       return file;
37178 +}
37179 +
37180 +/* ---------------------------------------------------------------------- */
37181 +
37182 +/*
37183 + * truncate xino files
37184 + */
37185 +static int au_xino_do_trunc(struct super_block *sb, aufs_bindex_t bindex,
37186 +                           int idx, struct kstatfs *st)
37187 +{
37188 +       int err;
37189 +       blkcnt_t blocks;
37190 +       struct file *file, *new_xino;
37191 +       struct au_xi_new xinew = {
37192 +               .idx = idx
37193 +       };
37194 +
37195 +       err = 0;
37196 +       xinew.xi = au_sbr(sb, bindex)->br_xino;
37197 +       file = au_xino_file(xinew.xi, idx);
37198 +       if (!file)
37199 +               goto out;
37200 +
37201 +       xinew.base = &file->f_path;
37202 +       err = vfs_statfs(xinew.base, st);
37203 +       if (unlikely(err)) {
37204 +               AuErr1("statfs err %d, ignored\n", err);
37205 +               err = 0;
37206 +               goto out;
37207 +       }
37208 +
37209 +       blocks = file_inode(file)->i_blocks;
37210 +       pr_info("begin truncating xino(b%d-%d), ib%llu, %llu/%llu free blks\n",
37211 +               bindex, idx, (u64)blocks, st->f_bfree, st->f_blocks);
37212 +
37213 +       xinew.copy_src = file;
37214 +       new_xino = au_xi_new(sb, &xinew);
37215 +       if (IS_ERR(new_xino)) {
37216 +               err = PTR_ERR(new_xino);
37217 +               pr_err("xino(b%d-%d), err %d, ignored\n", bindex, idx, err);
37218 +               goto out;
37219 +       }
37220 +
37221 +       err = vfs_statfs(&new_xino->f_path, st);
37222 +       if (!err)
37223 +               pr_info("end truncating xino(b%d-%d), ib%llu, %llu/%llu free blks\n",
37224 +                       bindex, idx, (u64)file_inode(new_xino)->i_blocks,
37225 +                       st->f_bfree, st->f_blocks);
37226 +       else {
37227 +               AuErr1("statfs err %d, ignored\n", err);
37228 +               err = 0;
37229 +       }
37230 +
37231 +out:
37232 +       return err;
37233 +}
37234 +
37235 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex, int idx_begin)
37236 +{
37237 +       int err, i;
37238 +       unsigned long jiffy;
37239 +       aufs_bindex_t bbot;
37240 +       struct kstatfs *st;
37241 +       struct au_branch *br;
37242 +       struct au_xino *xi;
37243 +
37244 +       err = -ENOMEM;
37245 +       st = kmalloc(sizeof(*st), GFP_NOFS);
37246 +       if (unlikely(!st))
37247 +               goto out;
37248 +
37249 +       err = -EINVAL;
37250 +       bbot = au_sbbot(sb);
37251 +       if (unlikely(bindex < 0 || bbot < bindex))
37252 +               goto out_st;
37253 +
37254 +       err = 0;
37255 +       jiffy = jiffies;
37256 +       br = au_sbr(sb, bindex);
37257 +       xi = br->br_xino;
37258 +       for (i = idx_begin; !err && i < xi->xi_nfile; i++)
37259 +               err = au_xino_do_trunc(sb, bindex, i, st);
37260 +       if (!err)
37261 +               au_sbi(sb)->si_xino_jiffy = jiffy;
37262 +
37263 +out_st:
37264 +       au_kfree_rcu(st);
37265 +out:
37266 +       return err;
37267 +}
37268 +
37269 +struct xino_do_trunc_args {
37270 +       struct super_block *sb;
37271 +       struct au_branch *br;
37272 +       int idx;
37273 +};
37274 +
37275 +static void xino_do_trunc(void *_args)
37276 +{
37277 +       struct xino_do_trunc_args *args = _args;
37278 +       struct super_block *sb;
37279 +       struct au_branch *br;
37280 +       struct inode *dir;
37281 +       int err, idx;
37282 +       aufs_bindex_t bindex;
37283 +
37284 +       err = 0;
37285 +       sb = args->sb;
37286 +       dir = d_inode(sb->s_root);
37287 +       br = args->br;
37288 +       idx = args->idx;
37289 +
37290 +       si_noflush_write_lock(sb);
37291 +       ii_read_lock_parent(dir);
37292 +       bindex = au_br_index(sb, br->br_id);
37293 +       err = au_xino_trunc(sb, bindex, idx);
37294 +       ii_read_unlock(dir);
37295 +       if (unlikely(err))
37296 +               pr_warn("err b%d, (%d)\n", bindex, err);
37297 +       atomic_dec(&br->br_xino->xi_truncating);
37298 +       au_lcnt_dec(&br->br_count);
37299 +       si_write_unlock(sb);
37300 +       au_nwt_done(&au_sbi(sb)->si_nowait);
37301 +       au_kfree_rcu(args);
37302 +}
37303 +
37304 +/*
37305 + * returns the index in the xi_file array whose corresponding file is necessary
37306 + * to truncate, or -1 which means no need to truncate.
37307 + */
37308 +static int xino_trunc_test(struct super_block *sb, struct au_branch *br)
37309 +{
37310 +       int err;
37311 +       unsigned int u;
37312 +       struct kstatfs st;
37313 +       struct au_sbinfo *sbinfo;
37314 +       struct au_xino *xi;
37315 +       struct file *file;
37316 +
37317 +       /* todo: si_xino_expire and the ratio should be customizable */
37318 +       sbinfo = au_sbi(sb);
37319 +       if (time_before(jiffies,
37320 +                       sbinfo->si_xino_jiffy + sbinfo->si_xino_expire))
37321 +               return -1;
37322 +
37323 +       /* truncation border */
37324 +       xi = br->br_xino;
37325 +       for (u = 0; u < xi->xi_nfile; u++) {
37326 +               file = au_xino_file(xi, u);
37327 +               if (!file)
37328 +                       continue;
37329 +
37330 +               err = vfs_statfs(&file->f_path, &st);
37331 +               if (unlikely(err)) {
37332 +                       AuErr1("statfs err %d, ignored\n", err);
37333 +                       return -1;
37334 +               }
37335 +               if (div64_u64(st.f_bfree * 100, st.f_blocks)
37336 +                   >= AUFS_XINO_DEF_TRUNC)
37337 +                       return u;
37338 +       }
37339 +
37340 +       return -1;
37341 +}
37342 +
37343 +static void xino_try_trunc(struct super_block *sb, struct au_branch *br)
37344 +{
37345 +       int idx;
37346 +       struct xino_do_trunc_args *args;
37347 +       int wkq_err;
37348 +
37349 +       idx = xino_trunc_test(sb, br);
37350 +       if (idx < 0)
37351 +               return;
37352 +
37353 +       if (atomic_inc_return(&br->br_xino->xi_truncating) > 1)
37354 +               goto out;
37355 +
37356 +       /* lock and kfree() will be called in trunc_xino() */
37357 +       args = kmalloc(sizeof(*args), GFP_NOFS);
37358 +       if (unlikely(!args)) {
37359 +               AuErr1("no memory\n");
37360 +               goto out;
37361 +       }
37362 +
37363 +       au_lcnt_inc(&br->br_count);
37364 +       args->sb = sb;
37365 +       args->br = br;
37366 +       args->idx = idx;
37367 +       wkq_err = au_wkq_nowait(xino_do_trunc, args, sb, /*flags*/0);
37368 +       if (!wkq_err)
37369 +               return; /* success */
37370 +
37371 +       pr_err("wkq %d\n", wkq_err);
37372 +       au_lcnt_dec(&br->br_count);
37373 +       au_kfree_rcu(args);
37374 +
37375 +out:
37376 +       atomic_dec(&br->br_xino->xi_truncating);
37377 +}
37378 +
37379 +/* ---------------------------------------------------------------------- */
37380 +
37381 +struct au_xi_calc {
37382 +       int idx;
37383 +       loff_t pos;
37384 +};
37385 +
37386 +static void au_xi_calc(struct super_block *sb, ino_t h_ino,
37387 +                      struct au_xi_calc *calc)
37388 +{
37389 +       loff_t maxent;
37390 +
37391 +       maxent = au_xi_maxent(sb);
37392 +       calc->idx = div64_u64_rem(h_ino, maxent, &calc->pos);
37393 +       calc->pos *= sizeof(ino_t);
37394 +}
37395 +
37396 +static int au_xino_do_new_async(struct super_block *sb, struct au_branch *br,
37397 +                               struct au_xi_calc *calc)
37398 +{
37399 +       int err;
37400 +       struct file *file;
37401 +       struct au_xino *xi = br->br_xino;
37402 +       struct au_xi_new xinew = {
37403 +               .xi = xi
37404 +       };
37405 +
37406 +       SiMustAnyLock(sb);
37407 +
37408 +       err = 0;
37409 +       if (!xi)
37410 +               goto out;
37411 +
37412 +       mutex_lock(&xi->xi_mtx);
37413 +       file = au_xino_file(xi, calc->idx);
37414 +       if (file)
37415 +               goto out_mtx;
37416 +
37417 +       file = au_xino_file(xi, /*idx*/-1);
37418 +       AuDebugOn(!file);
37419 +       xinew.idx = calc->idx;
37420 +       xinew.base = &file->f_path;
37421 +       /* xinew.copy_src = NULL; */
37422 +       file = au_xi_new(sb, &xinew);
37423 +       if (IS_ERR(file))
37424 +               err = PTR_ERR(file);
37425 +
37426 +out_mtx:
37427 +       mutex_unlock(&xi->xi_mtx);
37428 +out:
37429 +       return err;
37430 +}
37431 +
37432 +struct au_xino_do_new_async_args {
37433 +       struct super_block *sb;
37434 +       struct au_branch *br;
37435 +       struct au_xi_calc calc;
37436 +       ino_t ino;
37437 +};
37438 +
37439 +struct au_xi_writing {
37440 +       struct hlist_bl_node node;
37441 +       ino_t h_ino, ino;
37442 +};
37443 +
37444 +static int au_xino_do_write(struct file *file, struct au_xi_calc *calc,
37445 +                           ino_t ino);
37446 +
37447 +static void au_xino_call_do_new_async(void *args)
37448 +{
37449 +       struct au_xino_do_new_async_args *a = args;
37450 +       struct au_branch *br;
37451 +       struct super_block *sb;
37452 +       struct au_sbinfo *sbi;
37453 +       struct inode *root;
37454 +       struct file *file;
37455 +       struct au_xi_writing *del, *p;
37456 +       struct hlist_bl_head *hbl;
37457 +       struct hlist_bl_node *pos;
37458 +       int err;
37459 +
37460 +       br = a->br;
37461 +       sb = a->sb;
37462 +       sbi = au_sbi(sb);
37463 +       si_noflush_read_lock(sb);
37464 +       root = d_inode(sb->s_root);
37465 +       ii_read_lock_child(root);
37466 +       err = au_xino_do_new_async(sb, br, &a->calc);
37467 +       if (unlikely(err)) {
37468 +               AuIOErr("err %d\n", err);
37469 +               goto out;
37470 +       }
37471 +
37472 +       file = au_xino_file(br->br_xino, a->calc.idx);
37473 +       AuDebugOn(!file);
37474 +       err = au_xino_do_write(file, &a->calc, a->ino);
37475 +       if (unlikely(err)) {
37476 +               AuIOErr("err %d\n", err);
37477 +               goto out;
37478 +       }
37479 +
37480 +       del = NULL;
37481 +       hbl = &br->br_xino->xi_writing;
37482 +       hlist_bl_lock(hbl);
37483 +       au_hbl_for_each(pos, hbl) {
37484 +               p = container_of(pos, struct au_xi_writing, node);
37485 +               if (p->ino == a->ino) {
37486 +                       del = p;
37487 +                       hlist_bl_del(&p->node);
37488 +                       break;
37489 +               }
37490 +       }
37491 +       hlist_bl_unlock(hbl);
37492 +       au_kfree_rcu(del);
37493 +
37494 +out:
37495 +       au_lcnt_dec(&br->br_count);
37496 +       ii_read_unlock(root);
37497 +       si_read_unlock(sb);
37498 +       au_nwt_done(&sbi->si_nowait);
37499 +       au_kfree_rcu(a);
37500 +}
37501 +
37502 +/*
37503 + * create a new xino file asynchronously
37504 + */
37505 +static int au_xino_new_async(struct super_block *sb, struct au_branch *br,
37506 +                            struct au_xi_calc *calc, ino_t ino)
37507 +{
37508 +       int err;
37509 +       struct au_xino_do_new_async_args *arg;
37510 +
37511 +       err = -ENOMEM;
37512 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
37513 +       if (unlikely(!arg))
37514 +               goto out;
37515 +
37516 +       arg->sb = sb;
37517 +       arg->br = br;
37518 +       arg->calc = *calc;
37519 +       arg->ino = ino;
37520 +       au_lcnt_inc(&br->br_count);
37521 +       err = au_wkq_nowait(au_xino_call_do_new_async, arg, sb, AuWkq_NEST);
37522 +       if (unlikely(err)) {
37523 +               pr_err("wkq %d\n", err);
37524 +               au_lcnt_dec(&br->br_count);
37525 +               au_kfree_rcu(arg);
37526 +       }
37527 +
37528 +out:
37529 +       return err;
37530 +}
37531 +
37532 +/*
37533 + * read @ino from xinofile for the specified branch{@sb, @bindex}
37534 + * at the position of @h_ino.
37535 + */
37536 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
37537 +                ino_t *ino)
37538 +{
37539 +       int err;
37540 +       ssize_t sz;
37541 +       struct au_xi_calc calc;
37542 +       struct au_sbinfo *sbinfo;
37543 +       struct file *file;
37544 +       struct au_xino *xi;
37545 +       struct hlist_bl_head *hbl;
37546 +       struct hlist_bl_node *pos;
37547 +       struct au_xi_writing *p;
37548 +
37549 +       *ino = 0;
37550 +       if (!au_opt_test(au_mntflags(sb), XINO))
37551 +               return 0; /* no xino */
37552 +
37553 +       err = 0;
37554 +       au_xi_calc(sb, h_ino, &calc);
37555 +       xi = au_sbr(sb, bindex)->br_xino;
37556 +       file = au_xino_file(xi, calc.idx);
37557 +       if (!file) {
37558 +               hbl = &xi->xi_writing;
37559 +               hlist_bl_lock(hbl);
37560 +               au_hbl_for_each(pos, hbl) {
37561 +                       p = container_of(pos, struct au_xi_writing, node);
37562 +                       if (p->h_ino == h_ino) {
37563 +                               AuDbg("hi%llu, i%llu, found\n",
37564 +                                     (u64)p->h_ino, (u64)p->ino);
37565 +                               *ino = p->ino;
37566 +                               break;
37567 +                       }
37568 +               }
37569 +               hlist_bl_unlock(hbl);
37570 +               return 0;
37571 +       } else if (vfsub_f_size_read(file) < calc.pos + sizeof(*ino))
37572 +               return 0; /* no xino */
37573 +
37574 +       sbinfo = au_sbi(sb);
37575 +       sz = xino_fread(file, ino, sizeof(*ino), &calc.pos);
37576 +       if (sz == sizeof(*ino))
37577 +               return 0; /* success */
37578 +
37579 +       err = sz;
37580 +       if (unlikely(sz >= 0)) {
37581 +               err = -EIO;
37582 +               AuIOErr("xino read error (%zd)\n", sz);
37583 +       }
37584 +       return err;
37585 +}
37586 +
37587 +static int au_xino_do_write(struct file *file, struct au_xi_calc *calc,
37588 +                           ino_t ino)
37589 +{
37590 +       ssize_t sz;
37591 +
37592 +       sz = xino_fwrite(file, &ino, sizeof(ino), &calc->pos);
37593 +       if (sz == sizeof(ino))
37594 +               return 0; /* success */
37595 +
37596 +       AuIOErr("write failed (%zd)\n", sz);
37597 +       return -EIO;
37598 +}
37599 +
37600 +/*
37601 + * write @ino to the xinofile for the specified branch{@sb, @bindex}
37602 + * at the position of @h_ino.
37603 + * even if @ino is zero, it is written to the xinofile and means no entry.
37604 + * if the size of the xino file on a specific filesystem exceeds the watermark,
37605 + * try truncating it.
37606 + */
37607 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
37608 +                 ino_t ino)
37609 +{
37610 +       int err;
37611 +       unsigned int mnt_flags;
37612 +       struct au_xi_calc calc;
37613 +       struct file *file;
37614 +       struct au_branch *br;
37615 +       struct au_xino *xi;
37616 +       struct au_xi_writing *p;
37617 +
37618 +       SiMustAnyLock(sb);
37619 +
37620 +       mnt_flags = au_mntflags(sb);
37621 +       if (!au_opt_test(mnt_flags, XINO))
37622 +               return 0;
37623 +
37624 +       au_xi_calc(sb, h_ino, &calc);
37625 +       br = au_sbr(sb, bindex);
37626 +       xi = br->br_xino;
37627 +       file = au_xino_file(xi, calc.idx);
37628 +       if (!file) {
37629 +               /* store the inum pair into the list */
37630 +               p = kmalloc(sizeof(*p), GFP_NOFS | __GFP_NOFAIL);
37631 +               p->h_ino = h_ino;
37632 +               p->ino = ino;
37633 +               au_hbl_add(&p->node, &xi->xi_writing);
37634 +
37635 +               /* create and write a new xino file asynchronously */
37636 +               err = au_xino_new_async(sb, br, &calc, ino);
37637 +               if (!err)
37638 +                       return 0; /* success */
37639 +               goto out;
37640 +       }
37641 +
37642 +       err = au_xino_do_write(file, &calc, ino);
37643 +       if (!err) {
37644 +               br = au_sbr(sb, bindex);
37645 +               if (au_opt_test(mnt_flags, TRUNC_XINO)
37646 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
37647 +                       xino_try_trunc(sb, br);
37648 +               return 0; /* success */
37649 +       }
37650 +
37651 +out:
37652 +       AuIOErr("write failed (%d)\n", err);
37653 +       return -EIO;
37654 +}
37655 +
37656 +static ssize_t xino_fread_wkq(struct file *file, void *buf, size_t size,
37657 +                             loff_t *pos);
37658 +
37659 +/* todo: unnecessary to support mmap_sem since kernel-space? */
37660 +ssize_t xino_fread(struct file *file, void *kbuf, size_t size, loff_t *pos)
37661 +{
37662 +       ssize_t err;
37663 +       int i;
37664 +       const int prevent_endless = 10;
37665 +
37666 +       i = 0;
37667 +       do {
37668 +               err = vfsub_read_k(file, kbuf, size, pos);
37669 +               if (err == -EINTR
37670 +                   && !au_wkq_test()
37671 +                   && fatal_signal_pending(current)) {
37672 +                       err = xino_fread_wkq(file, kbuf, size, pos);
37673 +                       BUG_ON(err == -EINTR);
37674 +               }
37675 +       } while (i++ < prevent_endless
37676 +                && (err == -EAGAIN || err == -EINTR));
37677 +
37678 +#if 0 /* reserved for future use */
37679 +       if (err > 0)
37680 +               fsnotify_access(file->f_path.dentry);
37681 +#endif
37682 +
37683 +       return err;
37684 +}
37685 +
37686 +struct xino_fread_args {
37687 +       ssize_t *errp;
37688 +       struct file *file;
37689 +       void *buf;
37690 +       size_t size;
37691 +       loff_t *pos;
37692 +};
37693 +
37694 +static void call_xino_fread(void *args)
37695 +{
37696 +       struct xino_fread_args *a = args;
37697 +       *a->errp = xino_fread(a->file, a->buf, a->size, a->pos);
37698 +}
37699 +
37700 +static ssize_t xino_fread_wkq(struct file *file, void *buf, size_t size,
37701 +                             loff_t *pos)
37702 +{
37703 +       ssize_t err;
37704 +       int wkq_err;
37705 +       struct xino_fread_args args = {
37706 +               .errp   = &err,
37707 +               .file   = file,
37708 +               .buf    = buf,
37709 +               .size   = size,
37710 +               .pos    = pos
37711 +       };
37712 +
37713 +       wkq_err = au_wkq_wait(call_xino_fread, &args);
37714 +       if (unlikely(wkq_err))
37715 +               err = wkq_err;
37716 +
37717 +       return err;
37718 +}
37719 +
37720 +static ssize_t xino_fwrite_wkq(struct file *file, void *buf, size_t size,
37721 +                              loff_t *pos);
37722 +
37723 +static ssize_t do_xino_fwrite(struct file *file, void *kbuf, size_t size,
37724 +                             loff_t *pos)
37725 +{
37726 +       ssize_t err;
37727 +       int i;
37728 +       const int prevent_endless = 10;
37729 +
37730 +       i = 0;
37731 +       do {
37732 +               err = vfsub_write_k(file, kbuf, size, pos);
37733 +               if (err == -EINTR
37734 +                   && !au_wkq_test()
37735 +                   && fatal_signal_pending(current)) {
37736 +                       err = xino_fwrite_wkq(file, kbuf, size, pos);
37737 +                       BUG_ON(err == -EINTR);
37738 +               }
37739 +       } while (i++ < prevent_endless
37740 +                && (err == -EAGAIN || err == -EINTR));
37741 +
37742 +#if 0 /* reserved for future use */
37743 +       if (err > 0)
37744 +               fsnotify_modify(file->f_path.dentry);
37745 +#endif
37746 +
37747 +       return err;
37748 +}
37749 +
37750 +struct do_xino_fwrite_args {
37751 +       ssize_t *errp;
37752 +       struct file *file;
37753 +       void *buf;
37754 +       size_t size;
37755 +       loff_t *pos;
37756 +};
37757 +
37758 +static void call_do_xino_fwrite(void *args)
37759 +{
37760 +       struct do_xino_fwrite_args *a = args;
37761 +       *a->errp = do_xino_fwrite(a->file, a->buf, a->size, a->pos);
37762 +}
37763 +
37764 +static ssize_t xino_fwrite_wkq(struct file *file, void *buf, size_t size,
37765 +                              loff_t *pos)
37766 +{
37767 +       ssize_t err;
37768 +       int wkq_err;
37769 +       struct do_xino_fwrite_args args = {
37770 +               .errp   = &err,
37771 +               .file   = file,
37772 +               .buf    = buf,
37773 +               .size   = size,
37774 +               .pos    = pos
37775 +       };
37776 +
37777 +       /*
37778 +        * it breaks RLIMIT_FSIZE and normal user's limit,
37779 +        * users should care about quota and real 'filesystem full.'
37780 +        */
37781 +       wkq_err = au_wkq_wait(call_do_xino_fwrite, &args);
37782 +       if (unlikely(wkq_err))
37783 +               err = wkq_err;
37784 +
37785 +       return err;
37786 +}
37787 +
37788 +ssize_t xino_fwrite(struct file *file, void *buf, size_t size, loff_t *pos)
37789 +{
37790 +       ssize_t err;
37791 +
37792 +       if (rlimit(RLIMIT_FSIZE) == RLIM_INFINITY) {
37793 +               lockdep_off();
37794 +               err = do_xino_fwrite(file, buf, size, pos);
37795 +               lockdep_on();
37796 +       } else {
37797 +               lockdep_off();
37798 +               err = xino_fwrite_wkq(file, buf, size, pos);
37799 +               lockdep_on();
37800 +       }
37801 +
37802 +       return err;
37803 +}
37804 +
37805 +/* ---------------------------------------------------------------------- */
37806 +
37807 +/*
37808 + * inode number bitmap
37809 + */
37810 +static const int page_bits = (int)PAGE_SIZE * BITS_PER_BYTE;
37811 +static ino_t xib_calc_ino(unsigned long pindex, int bit)
37812 +{
37813 +       ino_t ino;
37814 +
37815 +       AuDebugOn(bit < 0 || page_bits <= bit);
37816 +       ino = AUFS_FIRST_INO + pindex * page_bits + bit;
37817 +       return ino;
37818 +}
37819 +
37820 +static void xib_calc_bit(ino_t ino, unsigned long *pindex, int *bit)
37821 +{
37822 +       AuDebugOn(ino < AUFS_FIRST_INO);
37823 +       ino -= AUFS_FIRST_INO;
37824 +       *pindex = ino / page_bits;
37825 +       *bit = ino % page_bits;
37826 +}
37827 +
37828 +static int xib_pindex(struct super_block *sb, unsigned long pindex)
37829 +{
37830 +       int err;
37831 +       loff_t pos;
37832 +       ssize_t sz;
37833 +       struct au_sbinfo *sbinfo;
37834 +       struct file *xib;
37835 +       unsigned long *p;
37836 +
37837 +       sbinfo = au_sbi(sb);
37838 +       MtxMustLock(&sbinfo->si_xib_mtx);
37839 +       AuDebugOn(pindex > ULONG_MAX / PAGE_SIZE
37840 +                 || !au_opt_test(sbinfo->si_mntflags, XINO));
37841 +
37842 +       if (pindex == sbinfo->si_xib_last_pindex)
37843 +               return 0;
37844 +
37845 +       xib = sbinfo->si_xib;
37846 +       p = sbinfo->si_xib_buf;
37847 +       pos = sbinfo->si_xib_last_pindex;
37848 +       pos *= PAGE_SIZE;
37849 +       sz = xino_fwrite(xib, p, PAGE_SIZE, &pos);
37850 +       if (unlikely(sz != PAGE_SIZE))
37851 +               goto out;
37852 +
37853 +       pos = pindex;
37854 +       pos *= PAGE_SIZE;
37855 +       if (vfsub_f_size_read(xib) >= pos + PAGE_SIZE)
37856 +               sz = xino_fread(xib, p, PAGE_SIZE, &pos);
37857 +       else {
37858 +               memset(p, 0, PAGE_SIZE);
37859 +               sz = xino_fwrite(xib, p, PAGE_SIZE, &pos);
37860 +       }
37861 +       if (sz == PAGE_SIZE) {
37862 +               sbinfo->si_xib_last_pindex = pindex;
37863 +               return 0; /* success */
37864 +       }
37865 +
37866 +out:
37867 +       AuIOErr1("write failed (%zd)\n", sz);
37868 +       err = sz;
37869 +       if (sz >= 0)
37870 +               err = -EIO;
37871 +       return err;
37872 +}
37873 +
37874 +static void au_xib_clear_bit(struct inode *inode)
37875 +{
37876 +       int err, bit;
37877 +       unsigned long pindex;
37878 +       struct super_block *sb;
37879 +       struct au_sbinfo *sbinfo;
37880 +
37881 +       AuDebugOn(inode->i_nlink);
37882 +
37883 +       sb = inode->i_sb;
37884 +       xib_calc_bit(inode->i_ino, &pindex, &bit);
37885 +       AuDebugOn(page_bits <= bit);
37886 +       sbinfo = au_sbi(sb);
37887 +       mutex_lock(&sbinfo->si_xib_mtx);
37888 +       err = xib_pindex(sb, pindex);
37889 +       if (!err) {
37890 +               clear_bit(bit, sbinfo->si_xib_buf);
37891 +               sbinfo->si_xib_next_bit = bit;
37892 +       }
37893 +       mutex_unlock(&sbinfo->si_xib_mtx);
37894 +}
37895 +
37896 +/* ---------------------------------------------------------------------- */
37897 +
37898 +/*
37899 + * truncate a xino bitmap file
37900 + */
37901 +
37902 +/* todo: slow */
37903 +static int do_xib_restore(struct super_block *sb, struct file *file, void *page)
37904 +{
37905 +       int err, bit;
37906 +       ssize_t sz;
37907 +       unsigned long pindex;
37908 +       loff_t pos, pend;
37909 +       struct au_sbinfo *sbinfo;
37910 +       ino_t *ino;
37911 +       unsigned long *p;
37912 +
37913 +       err = 0;
37914 +       sbinfo = au_sbi(sb);
37915 +       MtxMustLock(&sbinfo->si_xib_mtx);
37916 +       p = sbinfo->si_xib_buf;
37917 +       pend = vfsub_f_size_read(file);
37918 +       pos = 0;
37919 +       while (pos < pend) {
37920 +               sz = xino_fread(file, page, PAGE_SIZE, &pos);
37921 +               err = sz;
37922 +               if (unlikely(sz <= 0))
37923 +                       goto out;
37924 +
37925 +               err = 0;
37926 +               for (ino = page; sz > 0; ino++, sz -= sizeof(ino)) {
37927 +                       if (unlikely(*ino < AUFS_FIRST_INO))
37928 +                               continue;
37929 +
37930 +                       xib_calc_bit(*ino, &pindex, &bit);
37931 +                       AuDebugOn(page_bits <= bit);
37932 +                       err = xib_pindex(sb, pindex);
37933 +                       if (!err)
37934 +                               set_bit(bit, p);
37935 +                       else
37936 +                               goto out;
37937 +               }
37938 +       }
37939 +
37940 +out:
37941 +       return err;
37942 +}
37943 +
37944 +static int xib_restore(struct super_block *sb)
37945 +{
37946 +       int err, i;
37947 +       unsigned int nfile;
37948 +       aufs_bindex_t bindex, bbot;
37949 +       void *page;
37950 +       struct au_branch *br;
37951 +       struct au_xino *xi;
37952 +       struct file *file;
37953 +
37954 +       err = -ENOMEM;
37955 +       page = (void *)__get_free_page(GFP_NOFS);
37956 +       if (unlikely(!page))
37957 +               goto out;
37958 +
37959 +       err = 0;
37960 +       bbot = au_sbbot(sb);
37961 +       for (bindex = 0; !err && bindex <= bbot; bindex++)
37962 +               if (!bindex || is_sb_shared(sb, bindex, bindex - 1) < 0) {
37963 +                       br = au_sbr(sb, bindex);
37964 +                       xi = br->br_xino;
37965 +                       nfile = xi->xi_nfile;
37966 +                       for (i = 0; i < nfile; i++) {
37967 +                               file = au_xino_file(xi, i);
37968 +                               if (file)
37969 +                                       err = do_xib_restore(sb, file, page);
37970 +                       }
37971 +               } else
37972 +                       AuDbg("skip shared b%d\n", bindex);
37973 +       free_page((unsigned long)page);
37974 +
37975 +out:
37976 +       return err;
37977 +}
37978 +
37979 +int au_xib_trunc(struct super_block *sb)
37980 +{
37981 +       int err;
37982 +       ssize_t sz;
37983 +       loff_t pos;
37984 +       struct au_sbinfo *sbinfo;
37985 +       unsigned long *p;
37986 +       struct file *file;
37987 +
37988 +       SiMustWriteLock(sb);
37989 +
37990 +       err = 0;
37991 +       sbinfo = au_sbi(sb);
37992 +       if (!au_opt_test(sbinfo->si_mntflags, XINO))
37993 +               goto out;
37994 +
37995 +       file = sbinfo->si_xib;
37996 +       if (vfsub_f_size_read(file) <= PAGE_SIZE)
37997 +               goto out;
37998 +
37999 +       file = au_xino_create2(sb, &sbinfo->si_xib->f_path, NULL);
38000 +       err = PTR_ERR(file);
38001 +       if (IS_ERR(file))
38002 +               goto out;
38003 +       fput(sbinfo->si_xib);
38004 +       sbinfo->si_xib = file;
38005 +
38006 +       p = sbinfo->si_xib_buf;
38007 +       memset(p, 0, PAGE_SIZE);
38008 +       pos = 0;
38009 +       sz = xino_fwrite(sbinfo->si_xib, p, PAGE_SIZE, &pos);
38010 +       if (unlikely(sz != PAGE_SIZE)) {
38011 +               err = sz;
38012 +               AuIOErr("err %d\n", err);
38013 +               if (sz >= 0)
38014 +                       err = -EIO;
38015 +               goto out;
38016 +       }
38017 +
38018 +       mutex_lock(&sbinfo->si_xib_mtx);
38019 +       /* mnt_want_write() is unnecessary here */
38020 +       err = xib_restore(sb);
38021 +       mutex_unlock(&sbinfo->si_xib_mtx);
38022 +
38023 +out:
38024 +       return err;
38025 +}
38026 +
38027 +/* ---------------------------------------------------------------------- */
38028 +
38029 +struct au_xino *au_xino_alloc(unsigned int nfile)
38030 +{
38031 +       struct au_xino *xi;
38032 +
38033 +       xi = kzalloc(sizeof(*xi), GFP_NOFS);
38034 +       if (unlikely(!xi))
38035 +               goto out;
38036 +       xi->xi_nfile = nfile;
38037 +       xi->xi_file = kcalloc(nfile, sizeof(*xi->xi_file), GFP_NOFS);
38038 +       if (unlikely(!xi->xi_file))
38039 +               goto out_free;
38040 +
38041 +       xi->xi_nondir.total = 8; /* initial size */
38042 +       xi->xi_nondir.array = kcalloc(xi->xi_nondir.total, sizeof(ino_t),
38043 +                                     GFP_NOFS);
38044 +       if (unlikely(!xi->xi_nondir.array))
38045 +               goto out_file;
38046 +
38047 +       spin_lock_init(&xi->xi_nondir.spin);
38048 +       init_waitqueue_head(&xi->xi_nondir.wqh);
38049 +       mutex_init(&xi->xi_mtx);
38050 +       INIT_HLIST_BL_HEAD(&xi->xi_writing);
38051 +       atomic_set(&xi->xi_truncating, 0);
38052 +       kref_init(&xi->xi_kref);
38053 +       goto out; /* success */
38054 +
38055 +out_file:
38056 +       au_kfree_try_rcu(xi->xi_file);
38057 +out_free:
38058 +       au_kfree_rcu(xi);
38059 +       xi = NULL;
38060 +out:
38061 +       return xi;
38062 +}
38063 +
38064 +static int au_xino_init(struct au_branch *br, int idx, struct file *file)
38065 +{
38066 +       int err;
38067 +       struct au_xino *xi;
38068 +
38069 +       err = 0;
38070 +       xi = au_xino_alloc(idx + 1);
38071 +       if (unlikely(!xi)) {
38072 +               err = -ENOMEM;
38073 +               goto out;
38074 +       }
38075 +
38076 +       if (file)
38077 +               get_file(file);
38078 +       xi->xi_file[idx] = file;
38079 +       AuDebugOn(br->br_xino);
38080 +       br->br_xino = xi;
38081 +
38082 +out:
38083 +       return err;
38084 +}
38085 +
38086 +static void au_xino_release(struct kref *kref)
38087 +{
38088 +       struct au_xino *xi;
38089 +       int i;
38090 +       unsigned long ul;
38091 +       struct hlist_bl_head *hbl;
38092 +       struct hlist_bl_node *pos, *n;
38093 +       struct au_xi_writing *p;
38094 +
38095 +       xi = container_of(kref, struct au_xino, xi_kref);
38096 +       for (i = 0; i < xi->xi_nfile; i++)
38097 +               if (xi->xi_file[i])
38098 +                       fput(xi->xi_file[i]);
38099 +       for (i = xi->xi_nondir.total - 1; i >= 0; i--)
38100 +               AuDebugOn(xi->xi_nondir.array[i]);
38101 +       mutex_destroy(&xi->xi_mtx);
38102 +       hbl = &xi->xi_writing;
38103 +       ul = au_hbl_count(hbl);
38104 +       if (unlikely(ul)) {
38105 +               pr_warn("xi_writing %lu\n", ul);
38106 +               hlist_bl_lock(hbl);
38107 +               hlist_bl_for_each_entry_safe(p, pos, n, hbl, node) {
38108 +                       hlist_bl_del(&p->node);
38109 +                       /* kmemleak reported au_kfree_rcu() doesn't free it */
38110 +                       kfree(p);
38111 +               }
38112 +               hlist_bl_unlock(hbl);
38113 +       }
38114 +       au_kfree_try_rcu(xi->xi_file);
38115 +       au_kfree_try_rcu(xi->xi_nondir.array);
38116 +       au_kfree_rcu(xi);
38117 +}
38118 +
38119 +int au_xino_put(struct au_branch *br)
38120 +{
38121 +       int ret;
38122 +       struct au_xino *xi;
38123 +
38124 +       ret = 0;
38125 +       xi = br->br_xino;
38126 +       if (xi) {
38127 +               br->br_xino = NULL;
38128 +               ret = kref_put(&xi->xi_kref, au_xino_release);
38129 +       }
38130 +
38131 +       return ret;
38132 +}
38133 +
38134 +/* ---------------------------------------------------------------------- */
38135 +
38136 +/*
38137 + * xino mount option handlers
38138 + */
38139 +
38140 +/* xino bitmap */
38141 +static void xino_clear_xib(struct super_block *sb)
38142 +{
38143 +       struct au_sbinfo *sbinfo;
38144 +
38145 +       SiMustWriteLock(sb);
38146 +
38147 +       sbinfo = au_sbi(sb);
38148 +       if (sbinfo->si_xib)
38149 +               fput(sbinfo->si_xib);
38150 +       sbinfo->si_xib = NULL;
38151 +       if (sbinfo->si_xib_buf)
38152 +               free_page((unsigned long)sbinfo->si_xib_buf);
38153 +       sbinfo->si_xib_buf = NULL;
38154 +}
38155 +
38156 +static int au_xino_set_xib(struct super_block *sb, struct path *path)
38157 +{
38158 +       int err;
38159 +       loff_t pos;
38160 +       struct au_sbinfo *sbinfo;
38161 +       struct file *file;
38162 +       struct super_block *xi_sb;
38163 +
38164 +       SiMustWriteLock(sb);
38165 +
38166 +       sbinfo = au_sbi(sb);
38167 +       file = au_xino_create2(sb, path, sbinfo->si_xib);
38168 +       err = PTR_ERR(file);
38169 +       if (IS_ERR(file))
38170 +               goto out;
38171 +       if (sbinfo->si_xib)
38172 +               fput(sbinfo->si_xib);
38173 +       sbinfo->si_xib = file;
38174 +       xi_sb = file_inode(file)->i_sb;
38175 +       sbinfo->si_ximaxent = xi_sb->s_maxbytes;
38176 +       if (unlikely(sbinfo->si_ximaxent < PAGE_SIZE)) {
38177 +               err = -EIO;
38178 +               pr_err("s_maxbytes(%llu) on %s is too small\n",
38179 +                      (u64)sbinfo->si_ximaxent, au_sbtype(xi_sb));
38180 +               goto out_unset;
38181 +       }
38182 +       sbinfo->si_ximaxent /= sizeof(ino_t);
38183 +
38184 +       err = -ENOMEM;
38185 +       if (!sbinfo->si_xib_buf)
38186 +               sbinfo->si_xib_buf = (void *)get_zeroed_page(GFP_NOFS);
38187 +       if (unlikely(!sbinfo->si_xib_buf))
38188 +               goto out_unset;
38189 +
38190 +       sbinfo->si_xib_last_pindex = 0;
38191 +       sbinfo->si_xib_next_bit = 0;
38192 +       if (vfsub_f_size_read(file) < PAGE_SIZE) {
38193 +               pos = 0;
38194 +               err = xino_fwrite(file, sbinfo->si_xib_buf, PAGE_SIZE, &pos);
38195 +               if (unlikely(err != PAGE_SIZE))
38196 +                       goto out_free;
38197 +       }
38198 +       err = 0;
38199 +       goto out; /* success */
38200 +
38201 +out_free:
38202 +       if (sbinfo->si_xib_buf)
38203 +               free_page((unsigned long)sbinfo->si_xib_buf);
38204 +       sbinfo->si_xib_buf = NULL;
38205 +       if (err >= 0)
38206 +               err = -EIO;
38207 +out_unset:
38208 +       fput(sbinfo->si_xib);
38209 +       sbinfo->si_xib = NULL;
38210 +out:
38211 +       AuTraceErr(err);
38212 +       return err;
38213 +}
38214 +
38215 +/* xino for each branch */
38216 +static void xino_clear_br(struct super_block *sb)
38217 +{
38218 +       aufs_bindex_t bindex, bbot;
38219 +       struct au_branch *br;
38220 +
38221 +       bbot = au_sbbot(sb);
38222 +       for (bindex = 0; bindex <= bbot; bindex++) {
38223 +               br = au_sbr(sb, bindex);
38224 +               AuDebugOn(!br);
38225 +               au_xino_put(br);
38226 +       }
38227 +}
38228 +
38229 +static void au_xino_set_br_shared(struct super_block *sb, struct au_branch *br,
38230 +                                 aufs_bindex_t bshared)
38231 +{
38232 +       struct au_branch *brshared;
38233 +
38234 +       brshared = au_sbr(sb, bshared);
38235 +       AuDebugOn(!brshared->br_xino);
38236 +       AuDebugOn(!brshared->br_xino->xi_file);
38237 +       if (br->br_xino != brshared->br_xino) {
38238 +               au_xino_get(brshared);
38239 +               au_xino_put(br);
38240 +               br->br_xino = brshared->br_xino;
38241 +       }
38242 +}
38243 +
38244 +struct au_xino_do_set_br {
38245 +       struct au_branch *br;
38246 +       ino_t h_ino;
38247 +       aufs_bindex_t bshared;
38248 +};
38249 +
38250 +static int au_xino_do_set_br(struct super_block *sb, struct path *path,
38251 +                            struct au_xino_do_set_br *args)
38252 +{
38253 +       int err;
38254 +       struct au_xi_calc calc;
38255 +       struct file *file;
38256 +       struct au_branch *br;
38257 +       struct au_xi_new xinew = {
38258 +               .base = path
38259 +       };
38260 +
38261 +       br = args->br;
38262 +       xinew.xi = br->br_xino;
38263 +       au_xi_calc(sb, args->h_ino, &calc);
38264 +       xinew.copy_src = au_xino_file(xinew.xi, calc.idx);
38265 +       if (args->bshared >= 0)
38266 +               /* shared xino */
38267 +               au_xino_set_br_shared(sb, br, args->bshared);
38268 +       else if (!xinew.xi) {
38269 +               /* new xino */
38270 +               err = au_xino_init(br, calc.idx, xinew.copy_src);
38271 +               if (unlikely(err))
38272 +                       goto out;
38273 +       }
38274 +
38275 +       /* force re-creating */
38276 +       xinew.xi = br->br_xino;
38277 +       xinew.idx = calc.idx;
38278 +       mutex_lock(&xinew.xi->xi_mtx);
38279 +       file = au_xi_new(sb, &xinew);
38280 +       mutex_unlock(&xinew.xi->xi_mtx);
38281 +       err = PTR_ERR(file);
38282 +       if (IS_ERR(file))
38283 +               goto out;
38284 +       AuDebugOn(!file);
38285 +
38286 +       err = au_xino_do_write(file, &calc, AUFS_ROOT_INO);
38287 +       if (unlikely(err))
38288 +               au_xino_put(br);
38289 +
38290 +out:
38291 +       AuTraceErr(err);
38292 +       return err;
38293 +}
38294 +
38295 +static int au_xino_set_br(struct super_block *sb, struct path *path)
38296 +{
38297 +       int err;
38298 +       aufs_bindex_t bindex, bbot;
38299 +       struct au_xino_do_set_br args;
38300 +       struct inode *inode;
38301 +
38302 +       SiMustWriteLock(sb);
38303 +
38304 +       bbot = au_sbbot(sb);
38305 +       inode = d_inode(sb->s_root);
38306 +       for (bindex = 0; bindex <= bbot; bindex++) {
38307 +               args.h_ino = au_h_iptr(inode, bindex)->i_ino;
38308 +               args.br = au_sbr(sb, bindex);
38309 +               args.bshared = is_sb_shared(sb, bindex, bindex - 1);
38310 +               err = au_xino_do_set_br(sb, path, &args);
38311 +               if (unlikely(err))
38312 +                       break;
38313 +       }
38314 +
38315 +       AuTraceErr(err);
38316 +       return err;
38317 +}
38318 +
38319 +void au_xino_clr(struct super_block *sb)
38320 +{
38321 +       struct au_sbinfo *sbinfo;
38322 +
38323 +       au_xigen_clr(sb);
38324 +       xino_clear_xib(sb);
38325 +       xino_clear_br(sb);
38326 +       dbgaufs_brs_del(sb, 0);
38327 +       sbinfo = au_sbi(sb);
38328 +       /* lvalue, do not call au_mntflags() */
38329 +       au_opt_clr(sbinfo->si_mntflags, XINO);
38330 +}
38331 +
38332 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xiopt, int remount)
38333 +{
38334 +       int err, skip;
38335 +       struct dentry *dentry, *parent, *cur_dentry, *cur_parent;
38336 +       struct qstr *dname, *cur_name;
38337 +       struct file *cur_xino;
38338 +       struct au_sbinfo *sbinfo;
38339 +       struct path *path, *cur_path;
38340 +
38341 +       SiMustWriteLock(sb);
38342 +
38343 +       err = 0;
38344 +       sbinfo = au_sbi(sb);
38345 +       path = &xiopt->file->f_path;
38346 +       dentry = path->dentry;
38347 +       parent = dget_parent(dentry);
38348 +       if (remount) {
38349 +               skip = 0;
38350 +               cur_xino = sbinfo->si_xib;
38351 +               if (cur_xino) {
38352 +                       cur_path = &cur_xino->f_path;
38353 +                       cur_dentry = cur_path->dentry;
38354 +                       cur_parent = dget_parent(cur_dentry);
38355 +                       cur_name = &cur_dentry->d_name;
38356 +                       dname = &dentry->d_name;
38357 +                       skip = (cur_parent == parent
38358 +                               && au_qstreq(dname, cur_name));
38359 +                       dput(cur_parent);
38360 +               }
38361 +               if (skip)
38362 +                       goto out;
38363 +       }
38364 +
38365 +       au_opt_set(sbinfo->si_mntflags, XINO);
38366 +       err = au_xino_set_xib(sb, path);
38367 +       /* si_x{read,write} are set */
38368 +       if (!err)
38369 +               err = au_xigen_set(sb, path);
38370 +       if (!err)
38371 +               err = au_xino_set_br(sb, path);
38372 +       if (!err) {
38373 +               dbgaufs_brs_add(sb, 0, /*topdown*/1);
38374 +               goto out; /* success */
38375 +       }
38376 +
38377 +       /* reset all */
38378 +       AuIOErr("failed setting xino(%d).\n", err);
38379 +       au_xino_clr(sb);
38380 +
38381 +out:
38382 +       dput(parent);
38383 +       return err;
38384 +}
38385 +
38386 +/*
38387 + * create a xinofile at the default place/path.
38388 + */
38389 +struct file *au_xino_def(struct super_block *sb)
38390 +{
38391 +       struct file *file;
38392 +       char *page, *p;
38393 +       struct au_branch *br;
38394 +       struct super_block *h_sb;
38395 +       struct path path;
38396 +       aufs_bindex_t bbot, bindex, bwr;
38397 +
38398 +       br = NULL;
38399 +       bbot = au_sbbot(sb);
38400 +       bwr = -1;
38401 +       for (bindex = 0; bindex <= bbot; bindex++) {
38402 +               br = au_sbr(sb, bindex);
38403 +               if (au_br_writable(br->br_perm)
38404 +                   && !au_test_fs_bad_xino(au_br_sb(br))) {
38405 +                       bwr = bindex;
38406 +                       break;
38407 +               }
38408 +       }
38409 +
38410 +       if (bwr >= 0) {
38411 +               file = ERR_PTR(-ENOMEM);
38412 +               page = (void *)__get_free_page(GFP_NOFS);
38413 +               if (unlikely(!page))
38414 +                       goto out;
38415 +               path.mnt = au_br_mnt(br);
38416 +               path.dentry = au_h_dptr(sb->s_root, bwr);
38417 +               p = d_path(&path, page, PATH_MAX - sizeof(AUFS_XINO_FNAME));
38418 +               file = (void *)p;
38419 +               if (!IS_ERR(p)) {
38420 +                       strcat(p, "/" AUFS_XINO_FNAME);
38421 +                       AuDbg("%s\n", p);
38422 +                       file = au_xino_create(sb, p, /*silent*/0, /*wbrtop*/1);
38423 +               }
38424 +               free_page((unsigned long)page);
38425 +       } else {
38426 +               file = au_xino_create(sb, AUFS_XINO_DEFPATH, /*silent*/0,
38427 +                                     /*wbrtop*/0);
38428 +               if (IS_ERR(file))
38429 +                       goto out;
38430 +               h_sb = file->f_path.dentry->d_sb;
38431 +               if (unlikely(au_test_fs_bad_xino(h_sb))) {
38432 +                       pr_err("xino doesn't support %s(%s)\n",
38433 +                              AUFS_XINO_DEFPATH, au_sbtype(h_sb));
38434 +                       fput(file);
38435 +                       file = ERR_PTR(-EINVAL);
38436 +               }
38437 +       }
38438 +
38439 +out:
38440 +       return file;
38441 +}
38442 +
38443 +/* ---------------------------------------------------------------------- */
38444 +
38445 +/*
38446 + * initialize the xinofile for the specified branch @br
38447 + * at the place/path where @base_file indicates.
38448 + * test whether another branch is on the same filesystem or not,
38449 + * if found then share the xinofile with another branch.
38450 + */
38451 +int au_xino_init_br(struct super_block *sb, struct au_branch *br, ino_t h_ino,
38452 +                   struct path *base)
38453 +{
38454 +       int err;
38455 +       struct au_xino_do_set_br args = {
38456 +               .h_ino  = h_ino,
38457 +               .br     = br
38458 +       };
38459 +
38460 +       args.bshared = sbr_find_shared(sb, /*btop*/0, au_sbbot(sb),
38461 +                                      au_br_sb(br));
38462 +       err = au_xino_do_set_br(sb, base, &args);
38463 +       if (unlikely(err))
38464 +               au_xino_put(br);
38465 +
38466 +       return err;
38467 +}
38468 +
38469 +/* ---------------------------------------------------------------------- */
38470 +
38471 +/*
38472 + * get an unused inode number from bitmap
38473 + */
38474 +ino_t au_xino_new_ino(struct super_block *sb)
38475 +{
38476 +       ino_t ino;
38477 +       unsigned long *p, pindex, ul, pend;
38478 +       struct au_sbinfo *sbinfo;
38479 +       struct file *file;
38480 +       int free_bit, err;
38481 +
38482 +       if (!au_opt_test(au_mntflags(sb), XINO))
38483 +               return iunique(sb, AUFS_FIRST_INO);
38484 +
38485 +       sbinfo = au_sbi(sb);
38486 +       mutex_lock(&sbinfo->si_xib_mtx);
38487 +       p = sbinfo->si_xib_buf;
38488 +       free_bit = sbinfo->si_xib_next_bit;
38489 +       if (free_bit < page_bits && !test_bit(free_bit, p))
38490 +               goto out; /* success */
38491 +       free_bit = find_first_zero_bit(p, page_bits);
38492 +       if (free_bit < page_bits)
38493 +               goto out; /* success */
38494 +
38495 +       pindex = sbinfo->si_xib_last_pindex;
38496 +       for (ul = pindex - 1; ul < ULONG_MAX; ul--) {
38497 +               err = xib_pindex(sb, ul);
38498 +               if (unlikely(err))
38499 +                       goto out_err;
38500 +               free_bit = find_first_zero_bit(p, page_bits);
38501 +               if (free_bit < page_bits)
38502 +                       goto out; /* success */
38503 +       }
38504 +
38505 +       file = sbinfo->si_xib;
38506 +       pend = vfsub_f_size_read(file) / PAGE_SIZE;
38507 +       for (ul = pindex + 1; ul <= pend; ul++) {
38508 +               err = xib_pindex(sb, ul);
38509 +               if (unlikely(err))
38510 +                       goto out_err;
38511 +               free_bit = find_first_zero_bit(p, page_bits);
38512 +               if (free_bit < page_bits)
38513 +                       goto out; /* success */
38514 +       }
38515 +       BUG();
38516 +
38517 +out:
38518 +       set_bit(free_bit, p);
38519 +       sbinfo->si_xib_next_bit = free_bit + 1;
38520 +       pindex = sbinfo->si_xib_last_pindex;
38521 +       mutex_unlock(&sbinfo->si_xib_mtx);
38522 +       ino = xib_calc_ino(pindex, free_bit);
38523 +       AuDbg("i%lu\n", (unsigned long)ino);
38524 +       return ino;
38525 +out_err:
38526 +       mutex_unlock(&sbinfo->si_xib_mtx);
38527 +       AuDbg("i0\n");
38528 +       return 0;
38529 +}
38530 +
38531 +/* for s_op->delete_inode() */
38532 +void au_xino_delete_inode(struct inode *inode, const int unlinked)
38533 +{
38534 +       int err;
38535 +       unsigned int mnt_flags;
38536 +       aufs_bindex_t bindex, bbot, bi;
38537 +       unsigned char try_trunc;
38538 +       struct au_iinfo *iinfo;
38539 +       struct super_block *sb;
38540 +       struct au_hinode *hi;
38541 +       struct inode *h_inode;
38542 +       struct au_branch *br;
38543 +       struct au_xi_calc calc;
38544 +       struct file *file;
38545 +
38546 +       AuDebugOn(au_is_bad_inode(inode));
38547 +
38548 +       sb = inode->i_sb;
38549 +       mnt_flags = au_mntflags(sb);
38550 +       if (!au_opt_test(mnt_flags, XINO)
38551 +           || inode->i_ino == AUFS_ROOT_INO)
38552 +               return;
38553 +
38554 +       if (unlinked) {
38555 +               au_xigen_inc(inode);
38556 +               au_xib_clear_bit(inode);
38557 +       }
38558 +
38559 +       iinfo = au_ii(inode);
38560 +       bindex = iinfo->ii_btop;
38561 +       if (bindex < 0)
38562 +               return;
38563 +
38564 +       try_trunc = !!au_opt_test(mnt_flags, TRUNC_XINO);
38565 +       hi = au_hinode(iinfo, bindex);
38566 +       bbot = iinfo->ii_bbot;
38567 +       for (; bindex <= bbot; bindex++, hi++) {
38568 +               h_inode = hi->hi_inode;
38569 +               if (!h_inode
38570 +                   || (!unlinked && h_inode->i_nlink))
38571 +                       continue;
38572 +
38573 +               /* inode may not be revalidated */
38574 +               bi = au_br_index(sb, hi->hi_id);
38575 +               if (bi < 0)
38576 +                       continue;
38577 +
38578 +               br = au_sbr(sb, bi);
38579 +               au_xi_calc(sb, h_inode->i_ino, &calc);
38580 +               file = au_xino_file(br->br_xino, calc.idx);
38581 +               if (IS_ERR_OR_NULL(file))
38582 +                       continue;
38583 +
38584 +               err = au_xino_do_write(file, &calc, /*ino*/0);
38585 +               if (!err && try_trunc
38586 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
38587 +                       xino_try_trunc(sb, br);
38588 +       }
38589 +}
38590 +
38591 +/* ---------------------------------------------------------------------- */
38592 +
38593 +static int au_xinondir_find(struct au_xino *xi, ino_t h_ino)
38594 +{
38595 +       int found, total, i;
38596 +
38597 +       found = -1;
38598 +       total = xi->xi_nondir.total;
38599 +       for (i = 0; i < total; i++) {
38600 +               if (xi->xi_nondir.array[i] != h_ino)
38601 +                       continue;
38602 +               found = i;
38603 +               break;
38604 +       }
38605 +
38606 +       return found;
38607 +}
38608 +
38609 +static int au_xinondir_expand(struct au_xino *xi)
38610 +{
38611 +       int err, sz;
38612 +       ino_t *p;
38613 +
38614 +       BUILD_BUG_ON(KMALLOC_MAX_SIZE > INT_MAX);
38615 +
38616 +       err = -ENOMEM;
38617 +       sz = xi->xi_nondir.total * sizeof(ino_t);
38618 +       if (unlikely(sz > KMALLOC_MAX_SIZE / 2))
38619 +               goto out;
38620 +       p = au_kzrealloc(xi->xi_nondir.array, sz, sz << 1, GFP_ATOMIC,
38621 +                        /*may_shrink*/0);
38622 +       if (p) {
38623 +               xi->xi_nondir.array = p;
38624 +               xi->xi_nondir.total <<= 1;
38625 +               AuDbg("xi_nondir.total %d\n", xi->xi_nondir.total);
38626 +               err = 0;
38627 +       }
38628 +
38629 +out:
38630 +       return err;
38631 +}
38632 +
38633 +void au_xinondir_leave(struct super_block *sb, aufs_bindex_t bindex,
38634 +                      ino_t h_ino, int idx)
38635 +{
38636 +       struct au_xino *xi;
38637 +
38638 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
38639 +       xi = au_sbr(sb, bindex)->br_xino;
38640 +       AuDebugOn(idx < 0 || xi->xi_nondir.total <= idx);
38641 +
38642 +       spin_lock(&xi->xi_nondir.spin);
38643 +       AuDebugOn(xi->xi_nondir.array[idx] != h_ino);
38644 +       xi->xi_nondir.array[idx] = 0;
38645 +       spin_unlock(&xi->xi_nondir.spin);
38646 +       wake_up_all(&xi->xi_nondir.wqh);
38647 +}
38648 +
38649 +int au_xinondir_enter(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
38650 +                     int *idx)
38651 +{
38652 +       int err, found, empty;
38653 +       struct au_xino *xi;
38654 +
38655 +       err = 0;
38656 +       *idx = -1;
38657 +       if (!au_opt_test(au_mntflags(sb), XINO))
38658 +               goto out; /* no xino */
38659 +
38660 +       xi = au_sbr(sb, bindex)->br_xino;
38661 +
38662 +again:
38663 +       spin_lock(&xi->xi_nondir.spin);
38664 +       found = au_xinondir_find(xi, h_ino);
38665 +       if (found == -1) {
38666 +               empty = au_xinondir_find(xi, /*h_ino*/0);
38667 +               if (empty == -1) {
38668 +                       empty = xi->xi_nondir.total;
38669 +                       err = au_xinondir_expand(xi);
38670 +                       if (unlikely(err))
38671 +                               goto out_unlock;
38672 +               }
38673 +               xi->xi_nondir.array[empty] = h_ino;
38674 +               *idx = empty;
38675 +       } else {
38676 +               spin_unlock(&xi->xi_nondir.spin);
38677 +               wait_event(xi->xi_nondir.wqh,
38678 +                          xi->xi_nondir.array[found] != h_ino);
38679 +               goto again;
38680 +       }
38681 +
38682 +out_unlock:
38683 +       spin_unlock(&xi->xi_nondir.spin);
38684 +out:
38685 +       return err;
38686 +}
38687 +
38688 +/* ---------------------------------------------------------------------- */
38689 +
38690 +int au_xino_path(struct seq_file *seq, struct file *file)
38691 +{
38692 +       int err;
38693 +
38694 +       err = au_seq_path(seq, &file->f_path);
38695 +       if (unlikely(err))
38696 +               goto out;
38697 +
38698 +#define Deleted "\\040(deleted)"
38699 +       seq->count -= sizeof(Deleted) - 1;
38700 +       AuDebugOn(memcmp(seq->buf + seq->count, Deleted,
38701 +                        sizeof(Deleted) - 1));
38702 +#undef Deleted
38703 +
38704 +out:
38705 +       return err;
38706 +}
38707 diff -urN /usr/share/empty/include/uapi/linux/aufs_type.h linux/include/uapi/linux/aufs_type.h
38708 --- /usr/share/empty/include/uapi/linux/aufs_type.h     1970-01-01 01:00:00.000000000 +0100
38709 +++ linux/include/uapi/linux/aufs_type.h        2021-12-03 15:40:58.239980630 +0100
38710 @@ -0,0 +1,452 @@
38711 +/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
38712 +/*
38713 + * Copyright (C) 2005-2021 Junjiro R. Okajima
38714 + *
38715 + * This program, aufs is free software; you can redistribute it and/or modify
38716 + * it under the terms of the GNU General Public License as published by
38717 + * the Free Software Foundation; either version 2 of the License, or
38718 + * (at your option) any later version.
38719 + *
38720 + * This program is distributed in the hope that it will be useful,
38721 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
38722 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
38723 + * GNU General Public License for more details.
38724 + *
38725 + * You should have received a copy of the GNU General Public License
38726 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
38727 + */
38728 +
38729 +#ifndef __AUFS_TYPE_H__
38730 +#define __AUFS_TYPE_H__
38731 +
38732 +#define AUFS_NAME      "aufs"
38733 +
38734 +#ifdef __KERNEL__
38735 +/*
38736 + * define it before including all other headers.
38737 + * sched.h may use pr_* macros before defining "current", so define the
38738 + * no-current version first, and re-define later.
38739 + */
38740 +#define pr_fmt(fmt)    AUFS_NAME " %s:%d: " fmt, __func__, __LINE__
38741 +#include <linux/sched.h>
38742 +#undef pr_fmt
38743 +#define pr_fmt(fmt) \
38744 +               AUFS_NAME " %s:%d:%.*s[%d]: " fmt, __func__, __LINE__, \
38745 +               (int)sizeof(current->comm), current->comm, current->pid
38746 +#include <linux/limits.h>
38747 +#else
38748 +#include <stdint.h>
38749 +#include <sys/types.h>
38750 +#include <limits.h>
38751 +#endif /* __KERNEL__ */
38752 +
38753 +#define AUFS_VERSION   "5.15.5-20211129"
38754 +
38755 +/* todo? move this to linux-2.6.19/include/magic.h */
38756 +#define AUFS_SUPER_MAGIC       ('a' << 24 | 'u' << 16 | 'f' << 8 | 's')
38757 +
38758 +/* ---------------------------------------------------------------------- */
38759 +
38760 +#ifdef __KERNEL__
38761 +#ifdef CONFIG_AUFS_BRANCH_MAX_127
38762 +typedef int8_t aufs_bindex_t;
38763 +#define AUFS_BRANCH_MAX 127
38764 +#else
38765 +typedef int16_t aufs_bindex_t;
38766 +#ifdef CONFIG_AUFS_BRANCH_MAX_511
38767 +#define AUFS_BRANCH_MAX 511
38768 +#elif defined(CONFIG_AUFS_BRANCH_MAX_1023)
38769 +#define AUFS_BRANCH_MAX 1023
38770 +#elif defined(CONFIG_AUFS_BRANCH_MAX_32767)
38771 +#define AUFS_BRANCH_MAX 32767
38772 +#endif
38773 +#endif
38774 +
38775 +#ifndef AUFS_BRANCH_MAX
38776 +#error unknown CONFIG_AUFS_BRANCH_MAX value
38777 +#endif
38778 +#endif /* __KERNEL__ */
38779 +
38780 +/* ---------------------------------------------------------------------- */
38781 +
38782 +#define AUFS_FSTYPE            AUFS_NAME
38783 +
38784 +#define AUFS_ROOT_INO          2
38785 +#define AUFS_FIRST_INO         11
38786 +
38787 +#define AUFS_WH_PFX            ".wh."
38788 +#define AUFS_WH_PFX_LEN                ((int)sizeof(AUFS_WH_PFX) - 1)
38789 +#define AUFS_WH_TMP_LEN                4
38790 +/* a limit for rmdir/rename a dir and copyup */
38791 +#define AUFS_MAX_NAMELEN       (NAME_MAX \
38792 +                               - AUFS_WH_PFX_LEN * 2   /* doubly whiteouted */\
38793 +                               - 1                     /* dot */\
38794 +                               - AUFS_WH_TMP_LEN)      /* hex */
38795 +#define AUFS_XINO_FNAME                "." AUFS_NAME ".xino"
38796 +#define AUFS_XINO_DEFPATH      "/tmp/" AUFS_XINO_FNAME
38797 +#define AUFS_XINO_DEF_SEC      30 /* seconds */
38798 +#define AUFS_XINO_DEF_TRUNC    45 /* percentage */
38799 +#define AUFS_DIRWH_DEF         3
38800 +#define AUFS_RDCACHE_DEF       10 /* seconds */
38801 +#define AUFS_RDCACHE_MAX       3600 /* seconds */
38802 +#define AUFS_RDBLK_DEF         512 /* bytes */
38803 +#define AUFS_RDHASH_DEF                32
38804 +#define AUFS_WKQ_NAME          AUFS_NAME "d"
38805 +#define AUFS_MFS_DEF_SEC       30 /* seconds */
38806 +#define AUFS_MFS_MAX_SEC       3600 /* seconds */
38807 +#define AUFS_FHSM_CACHE_DEF_SEC        30 /* seconds */
38808 +#define AUFS_PLINK_WARN                50 /* number of plinks in a single bucket */
38809 +
38810 +/* pseudo-link maintenace under /proc */
38811 +#define AUFS_PLINK_MAINT_NAME  "plink_maint"
38812 +#define AUFS_PLINK_MAINT_DIR   "fs/" AUFS_NAME
38813 +#define AUFS_PLINK_MAINT_PATH  AUFS_PLINK_MAINT_DIR "/" AUFS_PLINK_MAINT_NAME
38814 +
38815 +/* dirren, renamed dir */
38816 +#define AUFS_DR_INFO_PFX       AUFS_WH_PFX ".dr."
38817 +#define AUFS_DR_BRHINO_NAME    AUFS_WH_PFX "hino"
38818 +/* whiteouted doubly */
38819 +#define AUFS_WH_DR_INFO_PFX    AUFS_WH_PFX AUFS_DR_INFO_PFX
38820 +#define AUFS_WH_DR_BRHINO      AUFS_WH_PFX AUFS_DR_BRHINO_NAME
38821 +
38822 +#define AUFS_DIROPQ_NAME       AUFS_WH_PFX ".opq" /* whiteouted doubly */
38823 +#define AUFS_WH_DIROPQ         AUFS_WH_PFX AUFS_DIROPQ_NAME
38824 +
38825 +#define AUFS_BASE_NAME         AUFS_WH_PFX AUFS_NAME
38826 +#define AUFS_PLINKDIR_NAME     AUFS_WH_PFX "plnk"
38827 +#define AUFS_ORPHDIR_NAME      AUFS_WH_PFX "orph"
38828 +
38829 +/* doubly whiteouted */
38830 +#define AUFS_WH_BASE           AUFS_WH_PFX AUFS_BASE_NAME
38831 +#define AUFS_WH_PLINKDIR       AUFS_WH_PFX AUFS_PLINKDIR_NAME
38832 +#define AUFS_WH_ORPHDIR                AUFS_WH_PFX AUFS_ORPHDIR_NAME
38833 +
38834 +/* branch permissions and attributes */
38835 +#define AUFS_BRPERM_RW         "rw"
38836 +#define AUFS_BRPERM_RO         "ro"
38837 +#define AUFS_BRPERM_RR         "rr"
38838 +#define AUFS_BRATTR_COO_REG    "coo_reg"
38839 +#define AUFS_BRATTR_COO_ALL    "coo_all"
38840 +#define AUFS_BRATTR_FHSM       "fhsm"
38841 +#define AUFS_BRATTR_UNPIN      "unpin"
38842 +#define AUFS_BRATTR_ICEX       "icex"
38843 +#define AUFS_BRATTR_ICEX_SEC   "icexsec"
38844 +#define AUFS_BRATTR_ICEX_SYS   "icexsys"
38845 +#define AUFS_BRATTR_ICEX_TR    "icextr"
38846 +#define AUFS_BRATTR_ICEX_USR   "icexusr"
38847 +#define AUFS_BRATTR_ICEX_OTH   "icexoth"
38848 +#define AUFS_BRRATTR_WH                "wh"
38849 +#define AUFS_BRWATTR_NLWH      "nolwh"
38850 +#define AUFS_BRWATTR_MOO       "moo"
38851 +
38852 +#define AuBrPerm_RW            1               /* writable, hardlinkable wh */
38853 +#define AuBrPerm_RO            (1 << 1)        /* readonly */
38854 +#define AuBrPerm_RR            (1 << 2)        /* natively readonly */
38855 +#define AuBrPerm_Mask          (AuBrPerm_RW | AuBrPerm_RO | AuBrPerm_RR)
38856 +
38857 +#define AuBrAttr_COO_REG       (1 << 3)        /* copy-up on open */
38858 +#define AuBrAttr_COO_ALL       (1 << 4)
38859 +#define AuBrAttr_COO_Mask      (AuBrAttr_COO_REG | AuBrAttr_COO_ALL)
38860 +
38861 +#define AuBrAttr_FHSM          (1 << 5)        /* file-based hsm */
38862 +#define AuBrAttr_UNPIN         (1 << 6)        /* rename-able top dir of
38863 +                                                  branch. meaningless since
38864 +                                                  linux-3.18-rc1 */
38865 +
38866 +/* ignore error in copying XATTR */
38867 +#define AuBrAttr_ICEX_SEC      (1 << 7)
38868 +#define AuBrAttr_ICEX_SYS      (1 << 8)
38869 +#define AuBrAttr_ICEX_TR       (1 << 9)
38870 +#define AuBrAttr_ICEX_USR      (1 << 10)
38871 +#define AuBrAttr_ICEX_OTH      (1 << 11)
38872 +#define AuBrAttr_ICEX          (AuBrAttr_ICEX_SEC      \
38873 +                                | AuBrAttr_ICEX_SYS    \
38874 +                                | AuBrAttr_ICEX_TR     \
38875 +                                | AuBrAttr_ICEX_USR    \
38876 +                                | AuBrAttr_ICEX_OTH)
38877 +
38878 +#define AuBrRAttr_WH           (1 << 12)       /* whiteout-able */
38879 +#define AuBrRAttr_Mask         AuBrRAttr_WH
38880 +
38881 +#define AuBrWAttr_NoLinkWH     (1 << 13)       /* un-hardlinkable whiteouts */
38882 +#define AuBrWAttr_MOO          (1 << 14)       /* move-up on open */
38883 +#define AuBrWAttr_Mask         (AuBrWAttr_NoLinkWH | AuBrWAttr_MOO)
38884 +
38885 +#define AuBrAttr_CMOO_Mask     (AuBrAttr_COO_Mask | AuBrWAttr_MOO)
38886 +
38887 +/* #warning test userspace */
38888 +#ifdef __KERNEL__
38889 +#ifndef CONFIG_AUFS_FHSM
38890 +#undef AuBrAttr_FHSM
38891 +#define AuBrAttr_FHSM          0
38892 +#endif
38893 +#ifndef CONFIG_AUFS_XATTR
38894 +#undef AuBrAttr_ICEX
38895 +#define AuBrAttr_ICEX          0
38896 +#undef AuBrAttr_ICEX_SEC
38897 +#define AuBrAttr_ICEX_SEC      0
38898 +#undef AuBrAttr_ICEX_SYS
38899 +#define AuBrAttr_ICEX_SYS      0
38900 +#undef AuBrAttr_ICEX_TR
38901 +#define AuBrAttr_ICEX_TR       0
38902 +#undef AuBrAttr_ICEX_USR
38903 +#define AuBrAttr_ICEX_USR      0
38904 +#undef AuBrAttr_ICEX_OTH
38905 +#define AuBrAttr_ICEX_OTH      0
38906 +#endif
38907 +#endif
38908 +
38909 +/* the longest combination */
38910 +/* AUFS_BRATTR_ICEX and AUFS_BRATTR_ICEX_TR don't affect here */
38911 +#define AuBrPermStrSz  sizeof(AUFS_BRPERM_RW                   \
38912 +                              "+" AUFS_BRATTR_COO_REG          \
38913 +                              "+" AUFS_BRATTR_FHSM             \
38914 +                              "+" AUFS_BRATTR_UNPIN            \
38915 +                              "+" AUFS_BRATTR_ICEX_SEC         \
38916 +                              "+" AUFS_BRATTR_ICEX_SYS         \
38917 +                              "+" AUFS_BRATTR_ICEX_USR         \
38918 +                              "+" AUFS_BRATTR_ICEX_OTH         \
38919 +                              "+" AUFS_BRWATTR_NLWH)
38920 +
38921 +typedef struct {
38922 +       char a[AuBrPermStrSz];
38923 +} au_br_perm_str_t;
38924 +
38925 +static inline int au_br_writable(int brperm)
38926 +{
38927 +       return brperm & AuBrPerm_RW;
38928 +}
38929 +
38930 +static inline int au_br_whable(int brperm)
38931 +{
38932 +       return brperm & (AuBrPerm_RW | AuBrRAttr_WH);
38933 +}
38934 +
38935 +static inline int au_br_wh_linkable(int brperm)
38936 +{
38937 +       return !(brperm & AuBrWAttr_NoLinkWH);
38938 +}
38939 +
38940 +static inline int au_br_cmoo(int brperm)
38941 +{
38942 +       return brperm & AuBrAttr_CMOO_Mask;
38943 +}
38944 +
38945 +static inline int au_br_fhsm(int brperm)
38946 +{
38947 +       return brperm & AuBrAttr_FHSM;
38948 +}
38949 +
38950 +/* ---------------------------------------------------------------------- */
38951 +
38952 +/* ioctl */
38953 +enum {
38954 +       /* readdir in userspace */
38955 +       AuCtl_RDU,
38956 +       AuCtl_RDU_INO,
38957 +
38958 +       AuCtl_WBR_FD,   /* pathconf wrapper */
38959 +       AuCtl_IBUSY,    /* busy inode */
38960 +       AuCtl_MVDOWN,   /* move-down */
38961 +       AuCtl_BR,       /* info about branches */
38962 +       AuCtl_FHSM_FD   /* connection for fhsm */
38963 +};
38964 +
38965 +/* borrowed from linux/include/linux/kernel.h */
38966 +#ifndef ALIGN
38967 +#ifdef _GNU_SOURCE
38968 +#define ALIGN(x, a)            __ALIGN_MASK(x, (typeof(x))(a)-1)
38969 +#else
38970 +#define ALIGN(x, a)            (((x) + (a) - 1) & ~((a) - 1))
38971 +#endif
38972 +#define __ALIGN_MASK(x, mask)  (((x)+(mask))&~(mask))
38973 +#endif
38974 +
38975 +/* borrowed from linux/include/linux/compiler-gcc3.h */
38976 +#ifndef __aligned
38977 +#define __aligned(x)                   __attribute__((aligned(x)))
38978 +#endif
38979 +
38980 +#ifdef __KERNEL__
38981 +#ifndef __packed
38982 +#define __packed                       __attribute__((packed))
38983 +#endif
38984 +#endif
38985 +
38986 +struct au_rdu_cookie {
38987 +       uint64_t        h_pos;
38988 +       int16_t         bindex;
38989 +       uint8_t         flags;
38990 +       uint8_t         pad;
38991 +       uint32_t        generation;
38992 +} __aligned(8);
38993 +
38994 +struct au_rdu_ent {
38995 +       uint64_t        ino;
38996 +       int16_t         bindex;
38997 +       uint8_t         type;
38998 +       uint8_t         nlen;
38999 +       uint8_t         wh;
39000 +       char            name[];
39001 +} __aligned(8);
39002 +
39003 +static inline int au_rdu_len(int nlen)
39004 +{
39005 +       /* include the terminating NULL */
39006 +       return ALIGN(sizeof(struct au_rdu_ent) + nlen + 1,
39007 +                    sizeof(uint64_t));
39008 +}
39009 +
39010 +union au_rdu_ent_ul {
39011 +       struct au_rdu_ent __user        *e;
39012 +       uint64_t                        ul;
39013 +};
39014 +
39015 +enum {
39016 +       AufsCtlRduV_SZ,
39017 +       AufsCtlRduV_End
39018 +};
39019 +
39020 +struct aufs_rdu {
39021 +       /* input */
39022 +       union {
39023 +               uint64_t        sz;     /* AuCtl_RDU */
39024 +               uint64_t        nent;   /* AuCtl_RDU_INO */
39025 +       };
39026 +       union au_rdu_ent_ul     ent;
39027 +       uint16_t                verify[AufsCtlRduV_End];
39028 +
39029 +       /* input/output */
39030 +       uint32_t                blk;
39031 +
39032 +       /* output */
39033 +       union au_rdu_ent_ul     tail;
39034 +       /* number of entries which were added in a single call */
39035 +       uint64_t                rent;
39036 +       uint8_t                 full;
39037 +       uint8_t                 shwh;
39038 +
39039 +       struct au_rdu_cookie    cookie;
39040 +} __aligned(8);
39041 +
39042 +/* ---------------------------------------------------------------------- */
39043 +
39044 +/* dirren. the branch is identified by the filename who contains this */
39045 +struct au_drinfo {
39046 +       uint64_t ino;
39047 +       union {
39048 +               uint8_t oldnamelen;
39049 +               uint64_t _padding;
39050 +       };
39051 +       uint8_t oldname[];
39052 +} __aligned(8);
39053 +
39054 +struct au_drinfo_fdata {
39055 +       uint32_t magic;
39056 +       struct au_drinfo drinfo;
39057 +} __aligned(8);
39058 +
39059 +#define AUFS_DRINFO_MAGIC_V1   ('a' << 24 | 'd' << 16 | 'r' << 8 | 0x01)
39060 +/* future */
39061 +#define AUFS_DRINFO_MAGIC_V2   ('a' << 24 | 'd' << 16 | 'r' << 8 | 0x02)
39062 +
39063 +/* ---------------------------------------------------------------------- */
39064 +
39065 +struct aufs_wbr_fd {
39066 +       uint32_t        oflags;
39067 +       int16_t         brid;
39068 +} __aligned(8);
39069 +
39070 +/* ---------------------------------------------------------------------- */
39071 +
39072 +struct aufs_ibusy {
39073 +       uint64_t        ino, h_ino;
39074 +       int16_t         bindex;
39075 +} __aligned(8);
39076 +
39077 +/* ---------------------------------------------------------------------- */
39078 +
39079 +/* error code for move-down */
39080 +/* the actual message strings are implemented in aufs-util.git */
39081 +enum {
39082 +       EAU_MVDOWN_OPAQUE = 1,
39083 +       EAU_MVDOWN_WHITEOUT,
39084 +       EAU_MVDOWN_UPPER,
39085 +       EAU_MVDOWN_BOTTOM,
39086 +       EAU_MVDOWN_NOUPPER,
39087 +       EAU_MVDOWN_NOLOWERBR,
39088 +       EAU_Last
39089 +};
39090 +
39091 +/* flags for move-down */
39092 +#define AUFS_MVDOWN_DMSG       1
39093 +#define AUFS_MVDOWN_OWLOWER    (1 << 1)        /* overwrite lower */
39094 +#define AUFS_MVDOWN_KUPPER     (1 << 2)        /* keep upper */
39095 +#define AUFS_MVDOWN_ROLOWER    (1 << 3)        /* do even if lower is RO */
39096 +#define AUFS_MVDOWN_ROLOWER_R  (1 << 4)        /* did on lower RO */
39097 +#define AUFS_MVDOWN_ROUPPER    (1 << 5)        /* do even if upper is RO */
39098 +#define AUFS_MVDOWN_ROUPPER_R  (1 << 6)        /* did on upper RO */
39099 +#define AUFS_MVDOWN_BRID_UPPER (1 << 7)        /* upper brid */
39100 +#define AUFS_MVDOWN_BRID_LOWER (1 << 8)        /* lower brid */
39101 +#define AUFS_MVDOWN_FHSM_LOWER (1 << 9)        /* find fhsm attr for lower */
39102 +#define AUFS_MVDOWN_STFS       (1 << 10)       /* req. stfs */
39103 +#define AUFS_MVDOWN_STFS_FAILED        (1 << 11)       /* output: stfs is unusable */
39104 +#define AUFS_MVDOWN_BOTTOM     (1 << 12)       /* output: no more lowers */
39105 +
39106 +/* index for move-down */
39107 +enum {
39108 +       AUFS_MVDOWN_UPPER,
39109 +       AUFS_MVDOWN_LOWER,
39110 +       AUFS_MVDOWN_NARRAY
39111 +};
39112 +
39113 +/*
39114 + * additional info of move-down
39115 + * number of free blocks and inodes.
39116 + * subset of struct kstatfs, but smaller and always 64bit.
39117 + */
39118 +struct aufs_stfs {
39119 +       uint64_t        f_blocks;
39120 +       uint64_t        f_bavail;
39121 +       uint64_t        f_files;
39122 +       uint64_t        f_ffree;
39123 +};
39124 +
39125 +struct aufs_stbr {
39126 +       int16_t                 brid;   /* optional input */
39127 +       int16_t                 bindex; /* output */
39128 +       struct aufs_stfs        stfs;   /* output when AUFS_MVDOWN_STFS set */
39129 +} __aligned(8);
39130 +
39131 +struct aufs_mvdown {
39132 +       uint32_t                flags;                  /* input/output */
39133 +       struct aufs_stbr        stbr[AUFS_MVDOWN_NARRAY]; /* input/output */
39134 +       int8_t                  au_errno;               /* output */
39135 +} __aligned(8);
39136 +
39137 +/* ---------------------------------------------------------------------- */
39138 +
39139 +union aufs_brinfo {
39140 +       /* PATH_MAX may differ between kernel-space and user-space */
39141 +       char    _spacer[4096];
39142 +       struct {
39143 +               int16_t id;
39144 +               int     perm;
39145 +               char    path[];
39146 +       };
39147 +} __aligned(8);
39148 +
39149 +/* ---------------------------------------------------------------------- */
39150 +
39151 +#define AuCtlType              'A'
39152 +#define AUFS_CTL_RDU           _IOWR(AuCtlType, AuCtl_RDU, struct aufs_rdu)
39153 +#define AUFS_CTL_RDU_INO       _IOWR(AuCtlType, AuCtl_RDU_INO, struct aufs_rdu)
39154 +#define AUFS_CTL_WBR_FD                _IOW(AuCtlType, AuCtl_WBR_FD, \
39155 +                                    struct aufs_wbr_fd)
39156 +#define AUFS_CTL_IBUSY         _IOWR(AuCtlType, AuCtl_IBUSY, struct aufs_ibusy)
39157 +#define AUFS_CTL_MVDOWN                _IOWR(AuCtlType, AuCtl_MVDOWN, \
39158 +                                     struct aufs_mvdown)
39159 +#define AUFS_CTL_BRINFO                _IOW(AuCtlType, AuCtl_BR, union aufs_brinfo)
39160 +#define AUFS_CTL_FHSM_FD       _IOW(AuCtlType, AuCtl_FHSM_FD, int)
39161 +
39162 +#endif /* __AUFS_TYPE_H__ */
39163 SPDX-License-Identifier: GPL-2.0
39164 aufs5.15.5 loopback patch
39165
39166 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
39167 index c9d1ef703673c..ca8b0b1146a60 100644
39168 --- a/drivers/block/loop.c
39169 +++ b/drivers/block/loop.c
39170 @@ -672,6 +672,15 @@ static inline void loop_update_dio(struct loop_device *lo)
39171                                 lo->use_dio);
39172  }
39173
39174 +static struct file *loop_real_file(struct file *file)
39175 +{
39176 +       struct file *f = NULL;
39177 +
39178 +       if (file->f_path.dentry->d_sb->s_op->real_loop)
39179 +               f = file->f_path.dentry->d_sb->s_op->real_loop(file);
39180 +       return f;
39181 +}
39182 +
39183  static void loop_reread_partitions(struct loop_device *lo)
39184  {
39185         int rc;
39186 @@ -729,6 +738,7 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
39187  {
39188         struct file *file = fget(arg);
39189         struct file *old_file;
39190 +       struct file *f, *virt_file = NULL, *old_virt_file;
39191         int error;
39192         bool partscan;
39193         bool is_loop;
39194 @@ -748,11 +758,19 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
39195         if (!(lo->lo_flags & LO_FLAGS_READ_ONLY))
39196                 goto out_err;
39197
39198 +       f = loop_real_file(file);
39199 +       if (f) {
39200 +               virt_file = file;
39201 +               file = f;
39202 +               get_file(file);
39203 +       }
39204 +
39205         error = loop_validate_file(file, bdev);
39206         if (error)
39207                 goto out_err;
39208
39209         old_file = lo->lo_backing_file;
39210 +       old_virt_file = lo->lo_backing_virt_file;
39211
39212         error = -EINVAL;
39213
39214 @@ -765,6 +783,7 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
39215         blk_mq_freeze_queue(lo->lo_queue);
39216         mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
39217         lo->lo_backing_file = file;
39218 +       lo->lo_backing_virt_file = virt_file;
39219         lo->old_gfp_mask = mapping_gfp_mask(file->f_mapping);
39220         mapping_set_gfp_mask(file->f_mapping,
39221                              lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
39222 @@ -787,6 +806,8 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
39223          * dependency.
39224          */
39225         fput(old_file);
39226 +       if (old_virt_file)
39227 +               fput(old_virt_file);
39228         if (partscan)
39229                 loop_reread_partitions(lo);
39230         return 0;
39231 @@ -795,6 +816,8 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
39232         loop_global_unlock(lo, is_loop);
39233  out_putf:
39234         fput(file);
39235 +       if (virt_file)
39236 +               fput(virt_file);
39237         return error;
39238  }
39239
39240 @@ -1195,6 +1218,7 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
39241                           const struct loop_config *config)
39242  {
39243         struct file *file = fget(config->fd);
39244 +       struct file *f, *virt_file = NULL;
39245         struct inode *inode;
39246         struct address_space *mapping;
39247         int error;
39248 @@ -1210,6 +1234,13 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
39249         /* This is safe, since we have a reference from open(). */
39250         __module_get(THIS_MODULE);
39251
39252 +       f = loop_real_file(file);
39253 +       if (f) {
39254 +               virt_file = file;
39255 +               file = f;
39256 +               get_file(file);
39257 +       }
39258 +
39259         /*
39260          * If we don't hold exclusive handle for the device, upgrade to it
39261          * here to avoid changing device under exclusive owner.
39262 @@ -1275,6 +1306,7 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
39263         lo->use_dio = lo->lo_flags & LO_FLAGS_DIRECT_IO;
39264         lo->lo_device = bdev;
39265         lo->lo_backing_file = file;
39266 +       lo->lo_backing_virt_file = virt_file;
39267         lo->old_gfp_mask = mapping_gfp_mask(mapping);
39268         mapping_set_gfp_mask(mapping, lo->old_gfp_mask & ~(__GFP_IO|__GFP_FS));
39269
39270 @@ -1325,6 +1357,8 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
39271                 bd_abort_claiming(bdev, loop_configure);
39272  out_putf:
39273         fput(file);
39274 +       if (virt_file)
39275 +               fput(virt_file);
39276         /* This is safe: open() is still holding a reference. */
39277         module_put(THIS_MODULE);
39278         return error;
39279 @@ -1333,6 +1367,7 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
39280  static int __loop_clr_fd(struct loop_device *lo, bool release)
39281  {
39282         struct file *filp = NULL;
39283 +       struct file *virt_filp = lo->lo_backing_virt_file;
39284         gfp_t gfp = lo->old_gfp_mask;
39285         struct block_device *bdev = lo->lo_device;
39286         int err = 0;
39287 @@ -1384,6 +1419,7 @@ static int __loop_clr_fd(struct loop_device *lo, bool release)
39288
39289         spin_lock_irq(&lo->lo_lock);
39290         lo->lo_backing_file = NULL;
39291 +       lo->lo_backing_virt_file = NULL;
39292         spin_unlock_irq(&lo->lo_lock);
39293
39294         loop_release_xfer(lo);
39295 @@ -1464,6 +1500,8 @@ static int __loop_clr_fd(struct loop_device *lo, bool release)
39296          */
39297         if (filp)
39298                 fput(filp);
39299 +       if (virt_filp)
39300 +               fput(virt_filp);
39301         return err;
39302  }
39303
39304 diff --git a/drivers/block/loop.h b/drivers/block/loop.h
39305 index 04c88dd6eabd6..0ff3ba22ee170 100644
39306 --- a/drivers/block/loop.h
39307 +++ b/drivers/block/loop.h
39308 @@ -46,7 +46,7 @@ struct loop_device {
39309         int             (*ioctl)(struct loop_device *, int cmd,
39310                                  unsigned long arg);
39311
39312 -       struct file *   lo_backing_file;
39313 +       struct file     *lo_backing_file, *lo_backing_virt_file;
39314         struct block_device *lo_device;
39315         void            *key_data;
39316
39317 diff --git a/fs/aufs/f_op.c b/fs/aufs/f_op.c
39318 index 99ceca1440449..e49dfe8550329 100644
39319 --- a/fs/aufs/f_op.c
39320 +++ b/fs/aufs/f_op.c
39321 @@ -304,7 +304,7 @@ static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
39322         if (IS_ERR(h_file))
39323                 goto out;
39324
39325 -       if (au_test_loopback_kthread()) {
39326 +       if (0 && au_test_loopback_kthread()) {
39327                 au_warn_loopback(h_file->f_path.dentry->d_sb);
39328                 if (file->f_mapping != h_file->f_mapping) {
39329                         file->f_mapping = h_file->f_mapping;
39330 diff --git a/fs/aufs/loop.c b/fs/aufs/loop.c
39331 index 74347bd75b380..5ef888a1d53f4 100644
39332 --- a/fs/aufs/loop.c
39333 +++ b/fs/aufs/loop.c
39334 @@ -133,3 +133,19 @@ void au_loopback_fin(void)
39335                 symbol_put(loop_backing_file);
39336         au_kfree_try_rcu(au_warn_loopback_array);
39337  }
39338 +
39339 +/* ---------------------------------------------------------------------- */
39340 +
39341 +/* support the loopback block device insude aufs */
39342 +
39343 +struct file *aufs_real_loop(struct file *file)
39344 +{
39345 +       struct file *f;
39346 +
39347 +       BUG_ON(!au_test_aufs(file->f_path.dentry->d_sb));
39348 +       fi_read_lock(file);
39349 +       f = au_hf_top(file);
39350 +       fi_read_unlock(file);
39351 +       AuDebugOn(!f);
39352 +       return f;
39353 +}
39354 diff --git a/fs/aufs/loop.h b/fs/aufs/loop.h
39355 index 7293bee427f96..3345c098d0d47 100644
39356 --- a/fs/aufs/loop.h
39357 +++ b/fs/aufs/loop.h
39358 @@ -26,6 +26,8 @@ void au_warn_loopback(struct super_block *h_sb);
39359
39360  int au_loopback_init(void);
39361  void au_loopback_fin(void);
39362 +
39363 +struct file *aufs_real_loop(struct file *file);
39364  #else
39365  AuStub(struct file *, loop_backing_file, return NULL, struct super_block *sb)
39366
39367 @@ -36,6 +38,8 @@ AuStubVoid(au_warn_loopback, struct super_block *h_sb)
39368
39369  AuStubInt0(au_loopback_init, void)
39370  AuStubVoid(au_loopback_fin, void)
39371 +
39372 +AuStub(struct file *, aufs_real_loop, return NULL, struct file *file)
39373  #endif /* BLK_DEV_LOOP */
39374
39375  #endif /* __KERNEL__ */
39376 diff --git a/fs/aufs/super.c b/fs/aufs/super.c
39377 index ae38857541533..af32f0960e595 100644
39378 --- a/fs/aufs/super.c
39379 +++ b/fs/aufs/super.c
39380 @@ -844,7 +844,10 @@ static const struct super_operations aufs_sop = {
39381         .statfs         = aufs_statfs,
39382         .put_super      = aufs_put_super,
39383         .sync_fs        = aufs_sync_fs,
39384 -       .remount_fs     = aufs_remount_fs
39385 +       .remount_fs     = aufs_remount_fs,
39386 +#ifdef CONFIG_AUFS_BDEV_LOOP
39387 +       .real_loop      = aufs_real_loop
39388 +#endif
39389  };
39390
39391  /* ---------------------------------------------------------------------- */
39392 diff --git a/include/linux/fs.h b/include/linux/fs.h
39393 index e60d8ad854004..2ac5317f9b795 100644
39394 --- a/include/linux/fs.h
39395 +++ b/include/linux/fs.h
39396 @@ -2226,6 +2226,10 @@ struct super_operations {
39397                                   struct shrink_control *);
39398         long (*free_cached_objects)(struct super_block *,
39399                                     struct shrink_control *);
39400 +#if IS_ENABLED(CONFIG_BLK_DEV_LOOP) || IS_ENABLED(CONFIG_BLK_DEV_LOOP_MODULE)
39401 +       /* and aufs */
39402 +       struct file *(*real_loop)(struct file *);
39403 +#endif
39404  };
39405
39406  /*