kernel-aufs4.patch

   1 aufs4.x-rcN kbuild patch
   2
   3 diff --git a/fs/Kconfig b/fs/Kconfig
   4 index 83eab52..31f16c4 100644
   5 --- a/fs/Kconfig
   6 +++ b/fs/Kconfig
   7 @@ -248,6 +248,7 @@ source "fs/pstore/Kconfig"
   8  source "fs/sysv/Kconfig"
   9  source "fs/ufs/Kconfig"
  10  source "fs/exofs/Kconfig"
  11 +source "fs/aufs/Kconfig"
  12
  13  endif # MISC_FILESYSTEMS
  14
  15 diff --git a/fs/Makefile b/fs/Makefile
  16 index 7bbaca9..a026491 100644
  17 --- a/fs/Makefile
  18 +++ b/fs/Makefile
  19 @@ -128,3 +128,4 @@ obj-y                               += exofs/ # Multiple modules
  20  obj-$(CONFIG_CEPH_FS)          += ceph/
  21  obj-$(CONFIG_PSTORE)           += pstore/
  22  obj-$(CONFIG_EFIVAR_FS)                += efivarfs/
  23 +obj-$(CONFIG_AUFS_FS)           += aufs/
  24 diff --git a/include/uapi/linux/Kbuild b/include/uapi/linux/Kbuild
  25 index dd9820b..d0435d9 100644
  26 --- a/include/uapi/linux/Kbuild
  27 +++ b/include/uapi/linux/Kbuild
  28 @@ -59,6 +59,7 @@ header-y += atmsvc.h
  29  header-y += atm_tcp.h
  30  header-y += atm_zatm.h
  31  header-y += audit.h
  32 +header-y += aufs_type.h
  33  header-y += auto_fs4.h
  34  header-y += auto_fs.h
  35  header-y += auxvec.h
  36 aufs4.x-rcN base patch
  37
  38 diff --git a/MAINTAINERS b/MAINTAINERS
  39 index 1b0a87f..76fd7cc 100644
  40 --- a/MAINTAINERS
  41 +++ b/MAINTAINERS
  42 @@ -2319,6 +2319,19 @@ F:       include/linux/audit.h
  43  F:     include/uapi/linux/audit.h
  44  F:     kernel/audit*
  45
  46 +AUFS (advanced multi layered unification filesystem) FILESYSTEM
  47 +M:     "J. R. Okajima" <hooanon05g@gmail.com>
  48 +L:     linux-unionfs@vger.kernel.org
  49 +L:     aufs-users@lists.sourceforge.net (members only)
  50 +W:     http://aufs.sourceforge.net
  51 +T:     git://github.com/sfjro/aufs4-linux.git
  52 +S:     Supported
  53 +F:     Documentation/filesystems/aufs/
  54 +F:     Documentation/ABI/testing/debugfs-aufs
  55 +F:     Documentation/ABI/testing/sysfs-aufs
  56 +F:     fs/aufs/
  57 +F:     include/uapi/linux/aufs_type.h
  58 +
  59  AUXILIARY DISPLAY DRIVERS
  60  M:     Miguel Ojeda Sandonis <miguel.ojeda.sandonis@gmail.com>
  61  W:     http://miguelojeda.es/auxdisplay.htm
  62 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
  63 index 0ecb646..cf39138 100644
  64 --- a/drivers/block/loop.c
  65 +++ b/drivers/block/loop.c
  66 @@ -701,6 +701,24 @@ static inline int is_loop_device(struct file *file)
  67         return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
  68  }
  69
  70 +/*
  71 + * for AUFS
  72 + * no get/put for file.
  73 + */
  74 +struct file *loop_backing_file(struct super_block *sb)
  75 +{
  76 +       struct file *ret;
  77 +       struct loop_device *l;
  78 +
  79 +       ret = NULL;
  80 +       if (MAJOR(sb->s_dev) == LOOP_MAJOR) {
  81 +               l = sb->s_bdev->bd_disk->private_data;
  82 +               ret = l->lo_backing_file;
  83 +       }
  84 +       return ret;
  85 +}
  86 +EXPORT_SYMBOL_GPL(loop_backing_file);
  87 +
  88  /* loop sysfs attributes */
  89
  90  static ssize_t loop_attr_show(struct device *dev, char *page,
  91 diff --git a/fs/dcache.c b/fs/dcache.c
  92 index 95d71ed..8ca5f09 100644
  93 --- a/fs/dcache.c
  94 +++ b/fs/dcache.c
  95 @@ -1164,7 +1164,7 @@ enum d_walk_ret {
  96   *
  97   * The @enter() and @finish() callbacks are called with d_lock held.
  98   */
  99 -static void d_walk(struct dentry *parent, void *data,
 100 +void d_walk(struct dentry *parent, void *data,
 101                    enum d_walk_ret (*enter)(void *, struct dentry *),
 102                    void (*finish)(void *))
 103  {
 104 diff --git a/fs/fcntl.c b/fs/fcntl.c
 105 index be8fbe2..f5f4f94 100644
 106 --- a/fs/fcntl.c
 107 +++ b/fs/fcntl.c
 108 @@ -30,7 +30,7 @@
 109
 110  #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
 111
 112 -static int setfl(int fd, struct file * filp, unsigned long arg)
 113 +int setfl(int fd, struct file * filp, unsigned long arg)
 114  {
 115         struct inode * inode = file_inode(filp);
 116         int error = 0;
 117 @@ -61,6 +61,8 @@ static int setfl(int fd, struct file * filp, unsigned long arg)
 118
 119         if (filp->f_op->check_flags)
 120                 error = filp->f_op->check_flags(arg);
 121 +       if (!error && filp->f_op->setfl)
 122 +               error = filp->f_op->setfl(filp, arg);
 123         if (error)
 124                 return error;
 125
 126 diff --git a/fs/inode.c b/fs/inode.c
 127 index 88110fd..9a9ba3a 100644
 128 --- a/fs/inode.c
 129 +++ b/fs/inode.c
 130 @@ -1642,7 +1642,7 @@ EXPORT_SYMBOL(generic_update_time);
 131   * This does the actual work of updating an inodes time or version.  Must have
 132   * had called mnt_want_write() before calling this.
 133   */
 134 -static int update_time(struct inode *inode, struct timespec *time, int flags)
 135 +int update_time(struct inode *inode, struct timespec *time, int flags)
 136  {
 137         int (*update_time)(struct inode *, struct timespec *, int);
 138
 139 diff --git a/fs/read_write.c b/fs/read_write.c
 140 index c4f88af..ac71b96 100644
 141 --- a/fs/read_write.c
 142 +++ b/fs/read_write.c
 143 @@ -513,6 +513,28 @@ ssize_t __vfs_write(struct file *file, const char __user *p, size_t count,
 144  }
 145  EXPORT_SYMBOL(__vfs_write);
 146
 147 +vfs_readf_t vfs_readf(struct file *file)
 148 +{
 149 +       const struct file_operations *fop = file->f_op;
 150 +
 151 +       if (fop->read)
 152 +               return fop->read;
 153 +       if (fop->read_iter)
 154 +               return new_sync_read;
 155 +       return ERR_PTR(-ENOSYS);
 156 +}
 157 +
 158 +vfs_writef_t vfs_writef(struct file *file)
 159 +{
 160 +       const struct file_operations *fop = file->f_op;
 161 +
 162 +       if (fop->write)
 163 +               return fop->write;
 164 +       if (fop->write_iter)
 165 +               return new_sync_write;
 166 +       return ERR_PTR(-ENOSYS);
 167 +}
 168 +
 169  ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
 170  {
 171         mm_segment_t old_fs;
 172 diff --git a/fs/splice.c b/fs/splice.c
 173 index 006ba50..46c87af 100644
 174 --- a/fs/splice.c
 175 +++ b/fs/splice.c
 176 @@ -859,8 +859,8 @@ EXPORT_SYMBOL(generic_splice_sendpage);
 177  /*
 178   * Attempt to initiate a splice from pipe to file.
 179   */
 180 -static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 181 -                          loff_t *ppos, size_t len, unsigned int flags)
 182 +long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 183 +                   loff_t *ppos, size_t len, unsigned int flags)
 184  {
 185         ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
 186                                 loff_t *, size_t, unsigned int);
 187 @@ -876,9 +876,9 @@ static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 188  /*
 189   * Attempt to initiate a splice from a file to a pipe.
 190   */
 191 -static long do_splice_to(struct file *in, loff_t *ppos,
 192 -                        struct pipe_inode_info *pipe, size_t len,
 193 -                        unsigned int flags)
 194 +long do_splice_to(struct file *in, loff_t *ppos,
 195 +                 struct pipe_inode_info *pipe, size_t len,
 196 +                 unsigned int flags)
 197  {
 198         ssize_t (*splice_read)(struct file *, loff_t *,
 199                                struct pipe_inode_info *, size_t, unsigned int);
 200 diff --git a/fs/sync.c b/fs/sync.c
 201 index 11ba023..abf6a5d 100644
 202 --- a/fs/sync.c
 203 +++ b/fs/sync.c
 204 @@ -27,7 +27,7 @@
 205   * wait == 1 case since in that case write_inode() functions do
 206   * sync_dirty_buffer() and thus effectively write one block at a time.
 207   */
 208 -static int __sync_filesystem(struct super_block *sb, int wait)
 209 +int __sync_filesystem(struct super_block *sb, int wait)
 210  {
 211         if (wait)
 212                 sync_inodes_sb(sb);
 213 diff --git a/include/linux/file.h b/include/linux/file.h
 214 index 61eb82c..e700888 100644
 215 --- a/include/linux/file.h
 216 +++ b/include/linux/file.h
 217 @@ -19,6 +19,7 @@ struct dentry;
 218  struct path;
 219  extern struct file *alloc_file(const struct path *, fmode_t mode,
 220         const struct file_operations *fop);
 221 +extern struct file *get_empty_filp(void);
 222
 223  static inline void fput_light(struct file *file, int fput_needed)
 224  {
 225 diff --git a/include/linux/fs.h b/include/linux/fs.h
 226 index 7251f7b..3668627 100644
 227 --- a/include/linux/fs.h
 228 +++ b/include/linux/fs.h
 229 @@ -1246,6 +1246,7 @@ extern void fasync_free(struct fasync_struct *);
 230  /* can be called from interrupts */
 231  extern void kill_fasync(struct fasync_struct **, int, int);
 232
 233 +extern int setfl(int fd, struct file * filp, unsigned long arg);
 234  extern void __f_setown(struct file *filp, struct pid *, enum pid_type, int force);
 235  extern void f_setown(struct file *filp, unsigned long arg, int force);
 236  extern void f_delown(struct file *filp);
 237 @@ -1673,6 +1674,7 @@ struct file_operations {
 238         ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
 239         unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
 240         int (*check_flags)(int);
 241 +       int (*setfl)(struct file *, unsigned long);
 242         int (*flock) (struct file *, int, struct file_lock *);
 243         ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
 244         ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
 245 @@ -1749,6 +1751,12 @@ ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
 246                               struct iovec *fast_pointer,
 247                               struct iovec **ret_pointer);
 248
 249 +typedef ssize_t (*vfs_readf_t)(struct file *, char __user *, size_t, loff_t *);
 250 +typedef ssize_t (*vfs_writef_t)(struct file *, const char __user *, size_t,
 251 +                               loff_t *);
 252 +vfs_readf_t vfs_readf(struct file *file);
 253 +vfs_writef_t vfs_writef(struct file *file);
 254 +
 255  extern ssize_t __vfs_read(struct file *, char __user *, size_t, loff_t *);
 256  extern ssize_t __vfs_write(struct file *, const char __user *, size_t, loff_t *);
 257  extern ssize_t vfs_read(struct file *, char __user *, size_t, loff_t *);
 258 @@ -2127,6 +2135,7 @@ extern int current_umask(void);
 259  extern void ihold(struct inode * inode);
 260  extern void iput(struct inode *);
 261  extern int generic_update_time(struct inode *, struct timespec *, int);
 262 +extern int update_time(struct inode *, struct timespec *, int);
 263
 264  /* /sys/fs */
 265  extern struct kobject *fs_kobj;
 266 @@ -2407,6 +2416,7 @@ static inline bool sb_is_blkdev_sb(struct super_block *sb)
 267         return false;
 268  }
 269  #endif
 270 +extern int __sync_filesystem(struct super_block *, int);
 271  extern int sync_filesystem(struct super_block *);
 272  extern const struct file_operations def_blk_fops;
 273  extern const struct file_operations def_chr_fops;
 274 diff --git a/include/linux/splice.h b/include/linux/splice.h
 275 index 00a2116..1f0a4a2 100644
 276 --- a/include/linux/splice.h
 277 +++ b/include/linux/splice.h
 278 @@ -86,4 +86,10 @@ extern void spd_release_page(struct splice_pipe_desc *, unsigned int);
 279
 280  extern const struct pipe_buf_operations page_cache_pipe_buf_ops;
 281  extern const struct pipe_buf_operations default_pipe_buf_ops;
 282 +
 283 +extern long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 284 +                          loff_t *ppos, size_t len, unsigned int flags);
 285 +extern long do_splice_to(struct file *in, loff_t *ppos,
 286 +                        struct pipe_inode_info *pipe, size_t len,
 287 +                        unsigned int flags);
 288  #endif
 289 aufs4.x-rcN mmap patch
 290
 291 diff --git a/fs/proc/base.c b/fs/proc/base.c
 292 index c87b6b9..6b71643 100644
 293 --- a/fs/proc/base.c
 294 +++ b/fs/proc/base.c
 295 @@ -1946,7 +1946,7 @@ static int map_files_get_link(struct dentry *dentry, struct path *path)
 296         down_read(&mm->mmap_sem);
 297         vma = find_exact_vma(mm, vm_start, vm_end);
 298         if (vma && vma->vm_file) {
 299 -               *path = vma->vm_file->f_path;
 300 +               *path = vma_pr_or_file(vma)->f_path;
 301                 path_get(path);
 302                 rc = 0;
 303         }
 304 diff --git a/fs/proc/nommu.c b/fs/proc/nommu.c
 305 index 7563437..7c0dc0f 100644
 306 --- a/fs/proc/nommu.c
 307 +++ b/fs/proc/nommu.c
 308 @@ -45,7 +45,10 @@ static int nommu_region_show(struct seq_file *m, struct vm_region *region)
 309         file = region->vm_file;
 310
 311         if (file) {
 312 -               struct inode *inode = file_inode(region->vm_file);
 313 +               struct inode *inode;
 314 +
 315 +               file = vmr_pr_or_file(region);
 316 +               inode = file_inode(file);
 317                 dev = inode->i_sb->s_dev;
 318                 ino = inode->i_ino;
 319         }
 320 diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
 321 index f08bd31..e1b9a37 100644
 322 --- a/fs/proc/task_mmu.c
 323 +++ b/fs/proc/task_mmu.c
 324 @@ -292,7 +292,10 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
 325         const char *name = NULL;
 326
 327         if (file) {
 328 -               struct inode *inode = file_inode(vma->vm_file);
 329 +               struct inode *inode;
 330 +
 331 +               file = vma_pr_or_file(vma);
 332 +               inode = file_inode(file);
 333                 dev = inode->i_sb->s_dev;
 334                 ino = inode->i_ino;
 335                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
 336 @@ -1629,7 +1632,7 @@ static int show_numa_map(struct seq_file *m, void *v, int is_pid)
 337         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
 338         struct vm_area_struct *vma = v;
 339         struct numa_maps *md = &numa_priv->md;
 340 -       struct file *file = vma->vm_file;
 341 +       struct file *file = vma_pr_or_file(vma);
 342         struct mm_struct *mm = vma->vm_mm;
 343         struct mm_walk walk = {
 344                 .hugetlb_entry = gather_hugetlb_stats,
 345 diff --git a/fs/proc/task_nommu.c b/fs/proc/task_nommu.c
 346 index 23266694..58e59b6 100644
 347 --- a/fs/proc/task_nommu.c
 348 +++ b/fs/proc/task_nommu.c
 349 @@ -157,7 +157,10 @@ static int nommu_vma_show(struct seq_file *m, struct vm_area_struct *vma,
 350         file = vma->vm_file;
 351
 352         if (file) {
 353 -               struct inode *inode = file_inode(vma->vm_file);
 354 +               struct inode *inode;
 355 +
 356 +               file = vma_pr_or_file(vma);
 357 +               inode = file_inode(file);
 358                 dev = inode->i_sb->s_dev;
 359                 ino = inode->i_ino;
 360                 pgoff = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
 361 diff --git a/include/linux/mm.h b/include/linux/mm.h
 362 index 00a8fa7..a90742b 100644
 363 --- a/include/linux/mm.h
 364 +++ b/include/linux/mm.h
 365 @@ -1294,6 +1294,28 @@ static inline int fixup_user_fault(struct task_struct *tsk,
 366  }
 367  #endif
 368
 369 +extern void vma_do_file_update_time(struct vm_area_struct *, const char[], int);
 370 +extern struct file *vma_do_pr_or_file(struct vm_area_struct *, const char[],
 371 +                                     int);
 372 +extern void vma_do_get_file(struct vm_area_struct *, const char[], int);
 373 +extern void vma_do_fput(struct vm_area_struct *, const char[], int);
 374 +
 375 +#define vma_file_update_time(vma)      vma_do_file_update_time(vma, __func__, \
 376 +                                                               __LINE__)
 377 +#define vma_pr_or_file(vma)            vma_do_pr_or_file(vma, __func__, \
 378 +                                                         __LINE__)
 379 +#define vma_get_file(vma)              vma_do_get_file(vma, __func__, __LINE__)
 380 +#define vma_fput(vma)                  vma_do_fput(vma, __func__, __LINE__)
 381 +
 382 +#ifndef CONFIG_MMU
 383 +extern struct file *vmr_do_pr_or_file(struct vm_region *, const char[], int);
 384 +extern void vmr_do_fput(struct vm_region *, const char[], int);
 385 +
 386 +#define vmr_pr_or_file(region)         vmr_do_pr_or_file(region, __func__, \
 387 +                                                         __LINE__)
 388 +#define vmr_fput(region)               vmr_do_fput(region, __func__, __LINE__)
 389 +#endif /* !CONFIG_MMU */
 390 +
 391  extern int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
 392                 unsigned int gup_flags);
 393  extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
 394 diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
 395 index f60f45f..38c5f76 100644
 396 --- a/include/linux/mm_types.h
 397 +++ b/include/linux/mm_types.h
 398 @@ -259,6 +259,7 @@ struct vm_region {
 399         unsigned long   vm_top;         /* region allocated to here */
 400         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
 401         struct file     *vm_file;       /* the backing file or NULL */
 402 +       struct file     *vm_prfile;     /* the virtual backing file or NULL */
 403
 404         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
 405         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
 406 @@ -333,6 +334,7 @@ struct vm_area_struct {
 407         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
 408                                            units */
 409         struct file * vm_file;          /* File we map to (can be NULL). */
 410 +       struct file *vm_prfile;         /* shadow of vm_file */
 411         void * vm_private_data;         /* was vm_pte (shared mem) */
 412
 413  #ifndef CONFIG_MMU
 414 diff --git a/kernel/fork.c b/kernel/fork.c
 415 index 6c463c80..24ae37c 100644
 416 --- a/kernel/fork.c
 417 +++ b/kernel/fork.c
 418 @@ -641,7 +641,7 @@ static __latent_entropy int dup_mmap(struct mm_struct *mm,
 419                         struct inode *inode = file_inode(file);
 420                         struct address_space *mapping = file->f_mapping;
 421
 422 -                       get_file(file);
 423 +                       vma_get_file(tmp);
 424                         if (tmp->vm_flags & VM_DENYWRITE)
 425                                 atomic_dec(&inode->i_writecount);
 426                         i_mmap_lock_write(mapping);
 427 diff --git a/mm/Makefile b/mm/Makefile
 428 index 026f6a8..723da17 100644
 429 --- a/mm/Makefile
 430 +++ b/mm/Makefile
 431 @@ -39,7 +39,7 @@ obj-y                 := filemap.o mempool.o oom_kill.o \
 432                            mm_init.o mmu_context.o percpu.o slab_common.o \
 433                            compaction.o vmacache.o swap_slots.o \
 434                            interval_tree.o list_lru.o workingset.o \
 435 -                          debug.o $(mmu-y)
 436 +                          prfile.o debug.o $(mmu-y)
 437
 438  obj-y += init-mm.o
 439
 440 diff --git a/mm/filemap.c b/mm/filemap.c
 441 index 1694623..68cd758 100644
 442 --- a/mm/filemap.c
 443 +++ b/mm/filemap.c
 444 @@ -2406,7 +2406,7 @@ int filemap_page_mkwrite(struct vm_fault *vmf)
 445         int ret = VM_FAULT_LOCKED;
 446
 447         sb_start_pagefault(inode->i_sb);
 448 -       file_update_time(vmf->vma->vm_file);
 449 +       vma_file_update_time(vmf->vma);
 450         lock_page(page);
 451         if (page->mapping != inode->i_mapping) {
 452                 unlock_page(page);
 453 diff --git a/mm/mmap.c b/mm/mmap.c
 454 index bfbe885..8bd32f9 100644
 455 --- a/mm/mmap.c
 456 +++ b/mm/mmap.c
 457 @@ -170,7 +170,7 @@ static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
 458         if (vma->vm_ops && vma->vm_ops->close)
 459                 vma->vm_ops->close(vma);
 460         if (vma->vm_file)
 461 -               fput(vma->vm_file);
 462 +               vma_fput(vma);
 463         mpol_put(vma_policy(vma));
 464         kmem_cache_free(vm_area_cachep, vma);
 465         return next;
 466 @@ -881,7 +881,7 @@ int __vma_adjust(struct vm_area_struct *vma, unsigned long start,
 467         if (remove_next) {
 468                 if (file) {
 469                         uprobe_munmap(next, next->vm_start, next->vm_end);
 470 -                       fput(file);
 471 +                       vma_fput(vma);
 472                 }
 473                 if (next->anon_vma)
 474                         anon_vma_merge(vma, next);
 475 @@ -1731,8 +1731,8 @@ unsigned long mmap_region(struct file *file, unsigned long addr,
 476         return addr;
 477
 478  unmap_and_free_vma:
 479 +       vma_fput(vma);
 480         vma->vm_file = NULL;
 481 -       fput(file);
 482
 483         /* Undo any partial mapping done by a device driver. */
 484         unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
 485 @@ -2537,7 +2537,7 @@ int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
 486                 goto out_free_mpol;
 487
 488         if (new->vm_file)
 489 -               get_file(new->vm_file);
 490 +               vma_get_file(new);
 491
 492         if (new->vm_ops && new->vm_ops->open)
 493                 new->vm_ops->open(new);
 494 @@ -2556,7 +2556,7 @@ int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
 495         if (new->vm_ops && new->vm_ops->close)
 496                 new->vm_ops->close(new);
 497         if (new->vm_file)
 498 -               fput(new->vm_file);
 499 +               vma_fput(new);
 500         unlink_anon_vmas(new);
 501   out_free_mpol:
 502         mpol_put(vma_policy(new));
 503 @@ -2710,7 +2710,7 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 504         struct vm_area_struct *vma;
 505         unsigned long populate = 0;
 506         unsigned long ret = -EINVAL;
 507 -       struct file *file;
 508 +       struct file *file, *prfile;
 509
 510         pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
 511                      current->comm, current->pid);
 512 @@ -2785,10 +2785,27 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
 513                 }
 514         }
 515
 516 -       file = get_file(vma->vm_file);
 517 +       vma_get_file(vma);
 518 +       file = vma->vm_file;
 519 +       prfile = vma->vm_prfile;
 520         ret = do_mmap_pgoff(vma->vm_file, start, size,
 521                         prot, flags, pgoff, &populate, NULL);
 522 +       if (!IS_ERR_VALUE(ret) && file && prfile) {
 523 +               struct vm_area_struct *new_vma;
 524 +
 525 +               new_vma = find_vma(mm, ret);
 526 +               if (!new_vma->vm_prfile)
 527 +                       new_vma->vm_prfile = prfile;
 528 +               if (new_vma != vma)
 529 +                       get_file(prfile);
 530 +       }
 531 +       /*
 532 +        * two fput()s instead of vma_fput(vma),
 533 +        * coz vma may not be available anymore.
 534 +        */
 535         fput(file);
 536 +       if (prfile)
 537 +               fput(prfile);
 538  out:
 539         up_write(&mm->mmap_sem);
 540         if (populate)
 541 @@ -3079,7 +3096,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
 542                 if (anon_vma_clone(new_vma, vma))
 543                         goto out_free_mempol;
 544                 if (new_vma->vm_file)
 545 -                       get_file(new_vma->vm_file);
 546 +                       vma_get_file(new_vma);
 547                 if (new_vma->vm_ops && new_vma->vm_ops->open)
 548                         new_vma->vm_ops->open(new_vma);
 549                 vma_link(mm, new_vma, prev, rb_link, rb_parent);
 550 diff --git a/mm/nommu.c b/mm/nommu.c
 551 index 2d131b9..5dc9a19 100644
 552 --- a/mm/nommu.c
 553 +++ b/mm/nommu.c
 554 @@ -637,7 +637,7 @@ static void __put_nommu_region(struct vm_region *region)
 555                 up_write(&nommu_region_sem);
 556
 557                 if (region->vm_file)
 558 -                       fput(region->vm_file);
 559 +                       vmr_fput(region);
 560
 561                 /* IO memory and memory shared directly out of the pagecache
 562                  * from ramfs/tmpfs mustn't be released here */
 563 @@ -795,7 +795,7 @@ static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
 564         if (vma->vm_ops && vma->vm_ops->close)
 565                 vma->vm_ops->close(vma);
 566         if (vma->vm_file)
 567 -               fput(vma->vm_file);
 568 +               vma_fput(vma);
 569         put_nommu_region(vma->vm_region);
 570         kmem_cache_free(vm_area_cachep, vma);
 571  }
 572 @@ -1322,7 +1322,7 @@ unsigned long do_mmap(struct file *file,
 573                                         goto error_just_free;
 574                                 }
 575                         }
 576 -                       fput(region->vm_file);
 577 +                       vmr_fput(region);
 578                         kmem_cache_free(vm_region_jar, region);
 579                         region = pregion;
 580                         result = start;
 581 @@ -1397,10 +1397,10 @@ unsigned long do_mmap(struct file *file,
 582         up_write(&nommu_region_sem);
 583  error:
 584         if (region->vm_file)
 585 -               fput(region->vm_file);
 586 +               vmr_fput(region);
 587         kmem_cache_free(vm_region_jar, region);
 588         if (vma->vm_file)
 589 -               fput(vma->vm_file);
 590 +               vma_fput(vma);
 591         kmem_cache_free(vm_area_cachep, vma);
 592         return ret;
 593
 594 diff --git a/mm/prfile.c b/mm/prfile.c
 595 new file mode 100644
 596 index 0000000..b323b8a
 597 --- /dev/null
 598 +++ b/mm/prfile.c
 599 @@ -0,0 +1,86 @@
 600 +/*
 601 + * Mainly for aufs which mmap(2) diffrent file and wants to print different path
 602 + * in /proc/PID/maps.
 603 + * Call these functions via macros defined in linux/mm.h.
 604 + *
 605 + * See Documentation/filesystems/aufs/design/06mmap.txt
 606 + *
 607 + * Copyright (c) 2014 Junjro R. Okajima
 608 + * Copyright (c) 2014 Ian Campbell
 609 + */
 610 +
 611 +#include <linux/mm.h>
 612 +#include <linux/file.h>
 613 +#include <linux/fs.h>
 614 +
 615 +/* #define PRFILE_TRACE */
 616 +static inline void prfile_trace(struct file *f, struct file *pr,
 617 +                             const char func[], int line, const char func2[])
 618 +{
 619 +#ifdef PRFILE_TRACE
 620 +       if (pr)
 621 +               pr_info("%s:%d: %s, %s\n", func, line, func2,
 622 +                       f ? (char *)f->f_path.dentry->d_name.name : "(null)");
 623 +#endif
 624 +}
 625 +
 626 +void vma_do_file_update_time(struct vm_area_struct *vma, const char func[],
 627 +                            int line)
 628 +{
 629 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
 630 +
 631 +       prfile_trace(f, pr, func, line, __func__);
 632 +       file_update_time(f);
 633 +       if (f && pr)
 634 +               file_update_time(pr);
 635 +}
 636 +
 637 +struct file *vma_do_pr_or_file(struct vm_area_struct *vma, const char func[],
 638 +                              int line)
 639 +{
 640 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
 641 +
 642 +       prfile_trace(f, pr, func, line, __func__);
 643 +       return (f && pr) ? pr : f;
 644 +}
 645 +
 646 +void vma_do_get_file(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 +       get_file(f);
 652 +       if (f && pr)
 653 +               get_file(pr);
 654 +}
 655 +
 656 +void vma_do_fput(struct vm_area_struct *vma, const char func[], int line)
 657 +{
 658 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
 659 +
 660 +       prfile_trace(f, pr, func, line, __func__);
 661 +       fput(f);
 662 +       if (f && pr)
 663 +               fput(pr);
 664 +}
 665 +
 666 +#ifndef CONFIG_MMU
 667 +struct file *vmr_do_pr_or_file(struct vm_region *region, const char func[],
 668 +                              int line)
 669 +{
 670 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
 671 +
 672 +       prfile_trace(f, pr, func, line, __func__);
 673 +       return (f && pr) ? pr : f;
 674 +}
 675 +
 676 +void vmr_do_fput(struct vm_region *region, const char func[], int line)
 677 +{
 678 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
 679 +
 680 +       prfile_trace(f, pr, func, line, __func__);
 681 +       fput(f);
 682 +       if (f && pr)
 683 +               fput(pr);
 684 +}
 685 +#endif /* !CONFIG_MMU */
 686 aufs4.x-rcN standalone patch
 687
 688 diff --git a/fs/dcache.c b/fs/dcache.c
 689 index 8ca5f09..b1ff5be 100644
 690 --- a/fs/dcache.c
 691 +++ b/fs/dcache.c
 692 @@ -1272,6 +1272,7 @@ void d_walk(struct dentry *parent, void *data,
 693         seq = 1;
 694         goto again;
 695  }
 696 +EXPORT_SYMBOL_GPL(d_walk);
 697
 698  struct check_mount {
 699         struct vfsmount *mnt;
 700 @@ -2864,6 +2865,7 @@ void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
 701
 702         write_sequnlock(&rename_lock);
 703  }
 704 +EXPORT_SYMBOL_GPL(d_exchange);
 705
 706  /**
 707   * d_ancestor - search for an ancestor
 708 diff --git a/fs/exec.c b/fs/exec.c
 709 index 65145a3..8d35776 100644
 710 --- a/fs/exec.c
 711 +++ b/fs/exec.c
 712 @@ -109,6 +109,7 @@ bool path_noexec(const struct path *path)
 713         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
 714                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
 715  }
 716 +EXPORT_SYMBOL_GPL(path_noexec);
 717
 718  #ifdef CONFIG_USELIB
 719  /*
 720 diff --git a/fs/fcntl.c b/fs/fcntl.c
 721 index f5f4f94..c671660 100644
 722 --- a/fs/fcntl.c
 723 +++ b/fs/fcntl.c
 724 @@ -83,6 +83,7 @@ int setfl(int fd, struct file * filp, unsigned long arg)
 725   out:
 726         return error;
 727  }
 728 +EXPORT_SYMBOL_GPL(setfl);
 729
 730  static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
 731                       int force)
 732 diff --git a/fs/file_table.c b/fs/file_table.c
 733 index 954d510..4fb5b10 100644
 734 --- a/fs/file_table.c
 735 +++ b/fs/file_table.c
 736 @@ -148,6 +148,7 @@ struct file *get_empty_filp(void)
 737         }
 738         return ERR_PTR(-ENFILE);
 739  }
 740 +EXPORT_SYMBOL_GPL(get_empty_filp);
 741
 742  /**
 743   * alloc_file - allocate and initialize a 'struct file'
 744 @@ -259,6 +260,7 @@ void flush_delayed_fput(void)
 745  {
 746         delayed_fput(NULL);
 747  }
 748 +EXPORT_SYMBOL_GPL(flush_delayed_fput);
 749
 750  static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
 751
 752 @@ -301,6 +303,7 @@ void __fput_sync(struct file *file)
 753  }
 754
 755  EXPORT_SYMBOL(fput);
 756 +EXPORT_SYMBOL_GPL(__fput_sync);
 757
 758  void put_filp(struct file *file)
 759  {
 760 @@ -309,6 +312,7 @@ void put_filp(struct file *file)
 761                 file_free(file);
 762         }
 763  }
 764 +EXPORT_SYMBOL_GPL(put_filp);
 765
 766  void __init files_init(void)
 767  {
 768 diff --git a/fs/inode.c b/fs/inode.c
 769 index 9a9ba3a..a3a18d8 100644
 770 --- a/fs/inode.c
 771 +++ b/fs/inode.c
 772 @@ -1651,6 +1651,7 @@ int update_time(struct inode *inode, struct timespec *time, int flags)
 773
 774         return update_time(inode, time, flags);
 775  }
 776 +EXPORT_SYMBOL_GPL(update_time);
 777
 778  /**
 779   *     touch_atime     -       update the access time
 780 diff --git a/fs/namespace.c b/fs/namespace.c
 781 index cc1375ef..9b4c67c 100644
 782 --- a/fs/namespace.c
 783 +++ b/fs/namespace.c
 784 @@ -465,6 +465,7 @@ void __mnt_drop_write(struct vfsmount *mnt)
 785         mnt_dec_writers(real_mount(mnt));
 786         preempt_enable();
 787  }
 788 +EXPORT_SYMBOL_GPL(__mnt_drop_write);
 789
 790  /**
 791   * mnt_drop_write - give up write access to a mount
 792 @@ -1884,6 +1885,7 @@ int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
 793         }
 794         return 0;
 795  }
 796 +EXPORT_SYMBOL_GPL(iterate_mounts);
 797
 798  static void cleanup_group_ids(struct mount *mnt, struct mount *end)
 799  {
 800 diff --git a/fs/notify/group.c b/fs/notify/group.c
 801 index fbe3cbe..bdfc61e 100644
 802 --- a/fs/notify/group.c
 803 +++ b/fs/notify/group.c
 804 @@ -22,6 +22,7 @@
 805  #include <linux/srcu.h>
 806  #include <linux/rculist.h>
 807  #include <linux/wait.h>
 808 +#include <linux/module.h>
 809
 810  #include <linux/fsnotify_backend.h>
 811  #include "fsnotify.h"
 812 @@ -100,6 +101,7 @@ void fsnotify_get_group(struct fsnotify_group *group)
 813  {
 814         atomic_inc(&group->refcnt);
 815  }
 816 +EXPORT_SYMBOL_GPL(fsnotify_get_group);
 817
 818  /*
 819   * Drop a reference to a group.  Free it if it's through.
 820 @@ -109,6 +111,7 @@ void fsnotify_put_group(struct fsnotify_group *group)
 821         if (atomic_dec_and_test(&group->refcnt))
 822                 fsnotify_final_destroy_group(group);
 823  }
 824 +EXPORT_SYMBOL_GPL(fsnotify_put_group);
 825
 826  /*
 827   * Create a new fsnotify_group and hold a reference for the group returned.
 828 @@ -137,6 +140,7 @@ struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops)
 829
 830         return group;
 831  }
 832 +EXPORT_SYMBOL_GPL(fsnotify_alloc_group);
 833
 834  int fsnotify_fasync(int fd, struct file *file, int on)
 835  {
 836 diff --git a/fs/notify/mark.c b/fs/notify/mark.c
 837 index 6043306..fdb50e4 100644
 838 --- a/fs/notify/mark.c
 839 +++ b/fs/notify/mark.c
 840 @@ -113,6 +113,7 @@ void fsnotify_put_mark(struct fsnotify_mark *mark)
 841                 mark->free_mark(mark);
 842         }
 843  }
 844 +EXPORT_SYMBOL_GPL(fsnotify_put_mark);
 845
 846  /* Calculate mask of events for a list of marks */
 847  u32 fsnotify_recalc_mask(struct hlist_head *head)
 848 @@ -230,6 +231,7 @@ void fsnotify_destroy_mark(struct fsnotify_mark *mark,
 849         mutex_unlock(&group->mark_mutex);
 850         fsnotify_free_mark(mark);
 851  }
 852 +EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
 853
 854  void fsnotify_destroy_marks(struct hlist_head *head, spinlock_t *lock)
 855  {
 856 @@ -415,6 +417,7 @@ int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
 857
 858         return ret;
 859  }
 860 +EXPORT_SYMBOL_GPL(fsnotify_add_mark);
 861
 862  int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
 863                       struct inode *inode, struct vfsmount *mnt, int allow_dups)
 864 @@ -521,6 +524,7 @@ void fsnotify_init_mark(struct fsnotify_mark *mark,
 865         atomic_set(&mark->refcnt, 1);
 866         mark->free_mark = free_mark;
 867  }
 868 +EXPORT_SYMBOL_GPL(fsnotify_init_mark);
 869
 870  /*
 871   * Destroy all marks in destroy_list, waits for SRCU period to finish before
 872 diff --git a/fs/open.c b/fs/open.c
 873 index 949cef2..9a892fb8 100644
 874 --- a/fs/open.c
 875 +++ b/fs/open.c
 876 @@ -64,6 +64,7 @@ int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
 877         inode_unlock(dentry->d_inode);
 878         return ret;
 879  }
 880 +EXPORT_SYMBOL_GPL(do_truncate);
 881
 882  long vfs_truncate(const struct path *path, loff_t length)
 883  {
 884 @@ -693,6 +694,7 @@ int open_check_o_direct(struct file *f)
 885         }
 886         return 0;
 887  }
 888 +EXPORT_SYMBOL_GPL(open_check_o_direct);
 889
 890  static int do_dentry_open(struct file *f,
 891                           struct inode *inode,
 892 diff --git a/fs/read_write.c b/fs/read_write.c
 893 index ac71b96..b69e687 100644
 894 --- a/fs/read_write.c
 895 +++ b/fs/read_write.c
 896 @@ -523,6 +523,7 @@ vfs_readf_t vfs_readf(struct file *file)
 897                 return new_sync_read;
 898         return ERR_PTR(-ENOSYS);
 899  }
 900 +EXPORT_SYMBOL_GPL(vfs_readf);
 901
 902  vfs_writef_t vfs_writef(struct file *file)
 903  {
 904 @@ -534,6 +535,7 @@ vfs_writef_t vfs_writef(struct file *file)
 905                 return new_sync_write;
 906         return ERR_PTR(-ENOSYS);
 907  }
 908 +EXPORT_SYMBOL_GPL(vfs_writef);
 909
 910  ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
 911  {
 912 diff --git a/fs/splice.c b/fs/splice.c
 913 index 46c87af..0efa652 100644
 914 --- a/fs/splice.c
 915 +++ b/fs/splice.c
 916 @@ -872,6 +872,7 @@ long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
 917
 918         return splice_write(pipe, out, ppos, len, flags);
 919  }
 920 +EXPORT_SYMBOL_GPL(do_splice_from);
 921
 922  /*
 923   * Attempt to initiate a splice from a file to a pipe.
 924 @@ -901,6 +902,7 @@ long do_splice_to(struct file *in, loff_t *ppos,
 925
 926         return splice_read(in, ppos, pipe, len, flags);
 927  }
 928 +EXPORT_SYMBOL_GPL(do_splice_to);
 929
 930  /**
 931   * splice_direct_to_actor - splices data directly between two non-pipes
 932 diff --git a/fs/sync.c b/fs/sync.c
 933 index abf6a5d..c86fe9c 100644
 934 --- a/fs/sync.c
 935 +++ b/fs/sync.c
 936 @@ -38,6 +38,7 @@ int __sync_filesystem(struct super_block *sb, int wait)
 937                 sb->s_op->sync_fs(sb, wait);
 938         return __sync_blockdev(sb->s_bdev, wait);
 939  }
 940 +EXPORT_SYMBOL_GPL(__sync_filesystem);
 941
 942  /*
 943   * Write out and wait upon all dirty data associated with this
 944 diff --git a/fs/xattr.c b/fs/xattr.c
 945 index 7e3317c..88910fe 100644
 946 --- a/fs/xattr.c
 947 +++ b/fs/xattr.c
 948 @@ -296,6 +296,7 @@ vfs_getxattr_alloc(struct dentry *dentry, const char *name, char **xattr_value,
 949         *xattr_value = value;
 950         return error;
 951  }
 952 +EXPORT_SYMBOL_GPL(vfs_getxattr_alloc);
 953
 954  ssize_t
 955  __vfs_getxattr(struct dentry *dentry, struct inode *inode, const char *name,
 956 diff --git a/kernel/task_work.c b/kernel/task_work.c
 957 index d513051..e056d54 100644
 958 --- a/kernel/task_work.c
 959 +++ b/kernel/task_work.c
 960 @@ -119,3 +119,4 @@ void task_work_run(void)
 961                 } while (work);
 962         }
 963  }
 964 +EXPORT_SYMBOL_GPL(task_work_run);
 965 diff --git a/security/commoncap.c b/security/commoncap.c
 966 index 78b3783..c8b3e88 100644
 967 --- a/security/commoncap.c
 968 +++ b/security/commoncap.c
 969 @@ -1062,12 +1062,14 @@ int cap_mmap_addr(unsigned long addr)
 970         }
 971         return ret;
 972  }
 973 +EXPORT_SYMBOL_GPL(cap_mmap_addr);
 974
 975  int cap_mmap_file(struct file *file, unsigned long reqprot,
 976                   unsigned long prot, unsigned long flags)
 977  {
 978         return 0;
 979  }
 980 +EXPORT_SYMBOL_GPL(cap_mmap_file);
 981
 982  #ifdef CONFIG_SECURITY
 983
 984 diff --git a/security/device_cgroup.c b/security/device_cgroup.c
 985 index 03c1652..f88c84b 100644
 986 --- a/security/device_cgroup.c
 987 +++ b/security/device_cgroup.c
 988 @@ -7,6 +7,7 @@
 989  #include <linux/device_cgroup.h>
 990  #include <linux/cgroup.h>
 991  #include <linux/ctype.h>
 992 +#include <linux/export.h>
 993  #include <linux/list.h>
 994  #include <linux/uaccess.h>
 995  #include <linux/seq_file.h>
 996 @@ -849,6 +850,7 @@ int __devcgroup_inode_permission(struct inode *inode, int mask)
 997         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
 998                         access);
 999  }
1000 +EXPORT_SYMBOL_GPL(__devcgroup_inode_permission);
1001
1002  int devcgroup_inode_mknod(int mode, dev_t dev)
1003  {
1004 diff --git a/security/security.c b/security/security.c
1005 index d0e07f2..5e323b0 100644
1006 --- a/security/security.c
1007 +++ b/security/security.c
1008 @@ -481,6 +481,7 @@ int security_path_rmdir(const struct path *dir, struct dentry *dentry)
1009                 return 0;
1010         return call_int_hook(path_rmdir, 0, dir, dentry);
1011  }
1012 +EXPORT_SYMBOL_GPL(security_path_rmdir);
1013
1014  int security_path_unlink(const struct path *dir, struct dentry *dentry)
1015  {
1016 @@ -497,6 +498,7 @@ int security_path_symlink(const struct path *dir, struct dentry *dentry,
1017                 return 0;
1018         return call_int_hook(path_symlink, 0, dir, dentry, old_name);
1019  }
1020 +EXPORT_SYMBOL_GPL(security_path_symlink);
1021
1022  int security_path_link(struct dentry *old_dentry, const struct path *new_dir,
1023                        struct dentry *new_dentry)
1024 @@ -505,6 +507,7 @@ int security_path_link(struct dentry *old_dentry, const struct path *new_dir,
1025                 return 0;
1026         return call_int_hook(path_link, 0, old_dentry, new_dir, new_dentry);
1027  }
1028 +EXPORT_SYMBOL_GPL(security_path_link);
1029
1030  int security_path_rename(const struct path *old_dir, struct dentry *old_dentry,
1031                          const struct path *new_dir, struct dentry *new_dentry,
1032 @@ -532,6 +535,7 @@ int security_path_truncate(const struct path *path)
1033                 return 0;
1034         return call_int_hook(path_truncate, 0, path);
1035  }
1036 +EXPORT_SYMBOL_GPL(security_path_truncate);
1037
1038  int security_path_chmod(const struct path *path, umode_t mode)
1039  {
1040 @@ -539,6 +543,7 @@ int security_path_chmod(const struct path *path, umode_t mode)
1041                 return 0;
1042         return call_int_hook(path_chmod, 0, path, mode);
1043  }
1044 +EXPORT_SYMBOL_GPL(security_path_chmod);
1045
1046  int security_path_chown(const struct path *path, kuid_t uid, kgid_t gid)
1047  {
1048 @@ -546,6 +551,7 @@ int security_path_chown(const struct path *path, kuid_t uid, kgid_t gid)
1049                 return 0;
1050         return call_int_hook(path_chown, 0, path, uid, gid);
1051  }
1052 +EXPORT_SYMBOL_GPL(security_path_chown);
1053
1054  int security_path_chroot(const struct path *path)
1055  {
1056 @@ -631,6 +637,7 @@ int security_inode_readlink(struct dentry *dentry)
1057                 return 0;
1058         return call_int_hook(inode_readlink, 0, dentry);
1059  }
1060 +EXPORT_SYMBOL_GPL(security_inode_readlink);
1061
1062  int security_inode_follow_link(struct dentry *dentry, struct inode *inode,
1063                                bool rcu)
1064 @@ -646,6 +653,7 @@ int security_inode_permission(struct inode *inode, int mask)
1065                 return 0;
1066         return call_int_hook(inode_permission, 0, inode, mask);
1067  }
1068 +EXPORT_SYMBOL_GPL(security_inode_permission);
1069
1070  int security_inode_setattr(struct dentry *dentry, struct iattr *attr)
1071  {
1072 @@ -817,6 +825,7 @@ int security_file_permission(struct file *file, int mask)
1073
1074         return fsnotify_perm(file, mask);
1075  }
1076 +EXPORT_SYMBOL_GPL(security_file_permission);
1077
1078  int security_file_alloc(struct file *file)
1079  {
1080 @@ -876,6 +885,7 @@ int security_mmap_file(struct file *file, unsigned long prot,
1081                 return ret;
1082         return ima_file_mmap(file, prot);
1083  }
1084 +EXPORT_SYMBOL_GPL(security_mmap_file);
1085
1086  int security_mmap_addr(unsigned long addr)
1087  {
1088 diff -urN /usr/share/empty/Documentation/ABI/testing/debugfs-aufs linux/Documentation/ABI/testing/debugfs-aufs
1089 --- /usr/share/empty/Documentation/ABI/testing/debugfs-aufs     1970-01-01 01:00:00.000000000 +0100
1090 +++ linux/Documentation/ABI/testing/debugfs-aufs        2017-05-06 22:16:52.774886674 +0200
1091 @@ -0,0 +1,50 @@
1092 +What:          /debug/aufs/si_<id>/
1093 +Date:          March 2009
1094 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1095 +Description:
1096 +               Under /debug/aufs, a directory named si_<id> is created
1097 +               per aufs mount, where <id> is a unique id generated
1098 +               internally.
1099 +
1100 +What:          /debug/aufs/si_<id>/plink
1101 +Date:          Apr 2013
1102 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1103 +Description:
1104 +               It has three lines and shows the information about the
1105 +               pseudo-link. The first line is a single number
1106 +               representing a number of buckets. The second line is a
1107 +               number of pseudo-links per buckets (separated by a
1108 +               blank). The last line is a single number representing a
1109 +               total number of psedo-links.
1110 +               When the aufs mount option 'noplink' is specified, it
1111 +               will show "1\n0\n0\n".
1112 +
1113 +What:          /debug/aufs/si_<id>/xib
1114 +Date:          March 2009
1115 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1116 +Description:
1117 +               It shows the consumed blocks by xib (External Inode Number
1118 +               Bitmap), its block size and file size.
1119 +               When the aufs mount option 'noxino' is specified, it
1120 +               will be empty. About XINO files, see the aufs manual.
1121 +
1122 +What:          /debug/aufs/si_<id>/xino0, xino1 ... xinoN
1123 +Date:          March 2009
1124 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1125 +Description:
1126 +               It shows the consumed blocks by xino (External Inode Number
1127 +               Translation Table), its link count, block size and file
1128 +               size.
1129 +               When the aufs mount option 'noxino' is specified, it
1130 +               will be empty. About XINO files, see the aufs manual.
1131 +
1132 +What:          /debug/aufs/si_<id>/xigen
1133 +Date:          March 2009
1134 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1135 +Description:
1136 +               It shows the consumed blocks by xigen (External Inode
1137 +               Generation Table), its block size and file size.
1138 +               If CONFIG_AUFS_EXPORT is disabled, this entry will not
1139 +               be created.
1140 +               When the aufs mount option 'noxino' is specified, it
1141 +               will be empty. About XINO files, see the aufs manual.
1142 diff -urN /usr/share/empty/Documentation/ABI/testing/sysfs-aufs linux/Documentation/ABI/testing/sysfs-aufs
1143 --- /usr/share/empty/Documentation/ABI/testing/sysfs-aufs       1970-01-01 01:00:00.000000000 +0100
1144 +++ linux/Documentation/ABI/testing/sysfs-aufs  2017-05-06 22:16:52.774886674 +0200
1145 @@ -0,0 +1,31 @@
1146 +What:          /sys/fs/aufs/si_<id>/
1147 +Date:          March 2009
1148 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1149 +Description:
1150 +               Under /sys/fs/aufs, a directory named si_<id> is created
1151 +               per aufs mount, where <id> is a unique id generated
1152 +               internally.
1153 +
1154 +What:          /sys/fs/aufs/si_<id>/br0, br1 ... brN
1155 +Date:          March 2009
1156 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1157 +Description:
1158 +               It shows the abolute path of a member directory (which
1159 +               is called branch) in aufs, and its permission.
1160 +
1161 +What:          /sys/fs/aufs/si_<id>/brid0, brid1 ... bridN
1162 +Date:          July 2013
1163 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1164 +Description:
1165 +               It shows the id of a member directory (which is called
1166 +               branch) in aufs.
1167 +
1168 +What:          /sys/fs/aufs/si_<id>/xi_path
1169 +Date:          March 2009
1170 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1171 +Description:
1172 +               It shows the abolute path of XINO (External Inode Number
1173 +               Bitmap, Translation Table and Generation Table) file
1174 +               even if it is the default path.
1175 +               When the aufs mount option 'noxino' is specified, it
1176 +               will be empty. About XINO files, see the aufs manual.
1177 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt linux/Documentation/filesystems/aufs/design/01intro.txt
1178 --- /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt  1970-01-01 01:00:00.000000000 +0100
1179 +++ linux/Documentation/filesystems/aufs/design/01intro.txt     2017-05-06 22:16:52.774886674 +0200
1180 @@ -0,0 +1,170 @@
1181 +
1182 +# Copyright (C) 2005-2017 Junjiro R. Okajima
1183 +#
1184 +# This program is free software; you can redistribute it and/or modify
1185 +# it under the terms of the GNU General Public License as published by
1186 +# the Free Software Foundation; either version 2 of the License, or
1187 +# (at your option) any later version.
1188 +#
1189 +# This program is distributed in the hope that it will be useful,
1190 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1191 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1192 +# GNU General Public License for more details.
1193 +#
1194 +# You should have received a copy of the GNU General Public License
1195 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1196 +
1197 +Introduction
1198 +----------------------------------------
1199 +
1200 +aufs [ei ju: ef es] | [a u f s]
1201 +1. abbrev. for "advanced multi-layered unification filesystem".
1202 +2. abbrev. for "another unionfs".
1203 +3. abbrev. for "auf das" in German which means "on the" in English.
1204 +   Ex. "Butter aufs Brot"(G) means "butter onto bread"(E).
1205 +   But "Filesystem aufs Filesystem" is hard to understand.
1206 +
1207 +AUFS is a filesystem with features:
1208 +- multi layered stackable unification filesystem, the member directory
1209 +  is called as a branch.
1210 +- branch permission and attribute, 'readonly', 'real-readonly',
1211 +  'readwrite', 'whiteout-able', 'link-able whiteout', etc. and their
1212 +  combination.
1213 +- internal "file copy-on-write".
1214 +- logical deletion, whiteout.
1215 +- dynamic branch manipulation, adding, deleting and changing permission.
1216 +- allow bypassing aufs, user's direct branch access.
1217 +- external inode number translation table and bitmap which maintains the
1218 +  persistent aufs inode number.
1219 +- seekable directory, including NFS readdir.
1220 +- file mapping, mmap and sharing pages.
1221 +- pseudo-link, hardlink over branches.
1222 +- loopback mounted filesystem as a branch.
1223 +- several policies to select one among multiple writable branches.
1224 +- revert a single systemcall when an error occurs in aufs.
1225 +- and more...
1226 +
1227 +
1228 +Multi Layered Stackable Unification Filesystem
1229 +----------------------------------------------------------------------
1230 +Most people already knows what it is.
1231 +It is a filesystem which unifies several directories and provides a
1232 +merged single directory. When users access a file, the access will be
1233 +passed/re-directed/converted (sorry, I am not sure which English word is
1234 +correct) to the real file on the member filesystem. The member
1235 +filesystem is called 'lower filesystem' or 'branch' and has a mode
1236 +'readonly' and 'readwrite.' And the deletion for a file on the lower
1237 +readonly branch is handled by creating 'whiteout' on the upper writable
1238 +branch.
1239 +
1240 +On LKML, there have been discussions about UnionMount (Jan Blunck,
1241 +Bharata B Rao and Valerie Aurora) and Unionfs (Erez Zadok). They took
1242 +different approaches to implement the merged-view.
1243 +The former tries putting it into VFS, and the latter implements as a
1244 +separate filesystem.
1245 +(If I misunderstand about these implementations, please let me know and
1246 +I shall correct it. Because it is a long time ago when I read their
1247 +source files last time).
1248 +
1249 +UnionMount's approach will be able to small, but may be hard to share
1250 +branches between several UnionMount since the whiteout in it is
1251 +implemented in the inode on branch filesystem and always
1252 +shared. According to Bharata's post, readdir does not seems to be
1253 +finished yet.
1254 +There are several missing features known in this implementations such as
1255 +- for users, the inode number may change silently. eg. copy-up.
1256 +- link(2) may break by copy-up.
1257 +- read(2) may get an obsoleted filedata (fstat(2) too).
1258 +- fcntl(F_SETLK) may be broken by copy-up.
1259 +- unnecessary copy-up may happen, for example mmap(MAP_PRIVATE) after
1260 +  open(O_RDWR).
1261 +
1262 +In linux-3.18, "overlay" filesystem (formerly known as "overlayfs") was
1263 +merged into mainline. This is another implementation of UnionMount as a
1264 +separated filesystem. All the limitations and known problems which
1265 +UnionMount are equally inherited to "overlay" filesystem.
1266 +
1267 +Unionfs has a longer history. When I started implementing a stackable
1268 +filesystem (Aug 2005), it already existed. It has virtual super_block,
1269 +inode, dentry and file objects and they have an array pointing lower
1270 +same kind objects. After contributing many patches for Unionfs, I
1271 +re-started my project AUFS (Jun 2006).
1272 +
1273 +In AUFS, the structure of filesystem resembles to Unionfs, but I
1274 +implemented my own ideas, approaches and enhancements and it became
1275 +totally different one.
1276 +
1277 +Comparing DM snapshot and fs based implementation
1278 +- the number of bytes to be copied between devices is much smaller.
1279 +- the type of filesystem must be one and only.
1280 +- the fs must be writable, no readonly fs, even for the lower original
1281 +  device. so the compression fs will not be usable. but if we use
1282 +  loopback mount, we may address this issue.
1283 +  for instance,
1284 +       mount /cdrom/squashfs.img /sq
1285 +       losetup /sq/ext2.img
1286 +       losetup /somewhere/cow
1287 +       dmsetup "snapshot /dev/loop0 /dev/loop1 ..."
1288 +- it will be difficult (or needs more operations) to extract the
1289 +  difference between the original device and COW.
1290 +- DM snapshot-merge may help a lot when users try merging. in the
1291 +  fs-layer union, users will use rsync(1).
1292 +
1293 +You may want to read my old paper "Filesystems in LiveCD"
1294 +(http://aufs.sourceforge.net/aufs2/report/sq/sq.pdf).
1295 +
1296 +
1297 +Several characters/aspects/persona of aufs
1298 +----------------------------------------------------------------------
1299 +
1300 +Aufs has several characters, aspects or persona.
1301 +1. a filesystem, callee of VFS helper
1302 +2. sub-VFS, caller of VFS helper for branches
1303 +3. a virtual filesystem which maintains persistent inode number
1304 +4. reader/writer of files on branches such like an application
1305 +
1306 +1. Callee of VFS Helper
1307 +As an ordinary linux filesystem, aufs is a callee of VFS. For instance,
1308 +unlink(2) from an application reaches sys_unlink() kernel function and
1309 +then vfs_unlink() is called. vfs_unlink() is one of VFS helper and it
1310 +calls filesystem specific unlink operation. Actually aufs implements the
1311 +unlink operation but it behaves like a redirector.
1312 +
1313 +2. Caller of VFS Helper for Branches
1314 +aufs_unlink() passes the unlink request to the branch filesystem as if
1315 +it were called from VFS. So the called unlink operation of the branch
1316 +filesystem acts as usual. As a caller of VFS helper, aufs should handle
1317 +every necessary pre/post operation for the branch filesystem.
1318 +- acquire the lock for the parent dir on a branch
1319 +- lookup in a branch
1320 +- revalidate dentry on a branch
1321 +- mnt_want_write() for a branch
1322 +- vfs_unlink() for a branch
1323 +- mnt_drop_write() for a branch
1324 +- release the lock on a branch
1325 +
1326 +3. Persistent Inode Number
1327 +One of the most important issue for a filesystem is to maintain inode
1328 +numbers. This is particularly important to support exporting a
1329 +filesystem via NFS. Aufs is a virtual filesystem which doesn't have a
1330 +backend block device for its own. But some storage is necessary to
1331 +keep and maintain the inode numbers. It may be a large space and may not
1332 +suit to keep in memory. Aufs rents some space from its first writable
1333 +branch filesystem (by default) and creates file(s) on it. These files
1334 +are created by aufs internally and removed soon (currently) keeping
1335 +opened.
1336 +Note: Because these files are removed, they are totally gone after
1337 +      unmounting aufs. It means the inode numbers are not persistent
1338 +      across unmount or reboot. I have a plan to make them really
1339 +      persistent which will be important for aufs on NFS server.
1340 +
1341 +4. Read/Write Files Internally (copy-on-write)
1342 +Because a branch can be readonly, when you write a file on it, aufs will
1343 +"copy-up" it to the upper writable branch internally. And then write the
1344 +originally requested thing to the file. Generally kernel doesn't
1345 +open/read/write file actively. In aufs, even a single write may cause a
1346 +internal "file copy". This behaviour is very similar to cp(1) command.
1347 +
1348 +Some people may think it is better to pass such work to user space
1349 +helper, instead of doing in kernel space. Actually I am still thinking
1350 +about it. But currently I have implemented it in kernel space.
1351 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt linux/Documentation/filesystems/aufs/design/02struct.txt
1352 --- /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt 1970-01-01 01:00:00.000000000 +0100
1353 +++ linux/Documentation/filesystems/aufs/design/02struct.txt    2017-05-06 22:16:52.774886674 +0200
1354 @@ -0,0 +1,258 @@
1355 +
1356 +# Copyright (C) 2005-2017 Junjiro R. Okajima
1357 +#
1358 +# This program is free software; you can redistribute it and/or modify
1359 +# it under the terms of the GNU General Public License as published by
1360 +# the Free Software Foundation; either version 2 of the License, or
1361 +# (at your option) any later version.
1362 +#
1363 +# This program is distributed in the hope that it will be useful,
1364 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1365 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1366 +# GNU General Public License for more details.
1367 +#
1368 +# You should have received a copy of the GNU General Public License
1369 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1370 +
1371 +Basic Aufs Internal Structure
1372 +
1373 +Superblock/Inode/Dentry/File Objects
1374 +----------------------------------------------------------------------
1375 +As like an ordinary filesystem, aufs has its own
1376 +superblock/inode/dentry/file objects. All these objects have a
1377 +dynamically allocated array and store the same kind of pointers to the
1378 +lower filesystem, branch.
1379 +For example, when you build a union with one readwrite branch and one
1380 +readonly, mounted /au, /rw and /ro respectively.
1381 +- /au = /rw + /ro
1382 +- /ro/fileA exists but /rw/fileA
1383 +
1384 +Aufs lookup operation finds /ro/fileA and gets dentry for that. These
1385 +pointers are stored in a aufs dentry. The array in aufs dentry will be,
1386 +- [0] = NULL (because /rw/fileA doesn't exist)
1387 +- [1] = /ro/fileA
1388 +
1389 +This style of an array is essentially same to the aufs
1390 +superblock/inode/dentry/file objects.
1391 +
1392 +Because aufs supports manipulating branches, ie. add/delete/change
1393 +branches dynamically, these objects has its own generation. When
1394 +branches are changed, the generation in aufs superblock is
1395 +incremented. And a generation in other object are compared when it is
1396 +accessed. When a generation in other objects are obsoleted, aufs
1397 +refreshes the internal array.
1398 +
1399 +
1400 +Superblock
1401 +----------------------------------------------------------------------
1402 +Additionally aufs superblock has some data for policies to select one
1403 +among multiple writable branches, XIB files, pseudo-links and kobject.
1404 +See below in detail.
1405 +About the policies which supports copy-down a directory, see
1406 +wbr_policy.txt too.
1407 +
1408 +
1409 +Branch and XINO(External Inode Number Translation Table)
1410 +----------------------------------------------------------------------
1411 +Every branch has its own xino (external inode number translation table)
1412 +file. The xino file is created and unlinked by aufs internally. When two
1413 +members of a union exist on the same filesystem, they share the single
1414 +xino file.
1415 +The struct of a xino file is simple, just a sequence of aufs inode
1416 +numbers which is indexed by the lower inode number.
1417 +In the above sample, assume the inode number of /ro/fileA is i111 and
1418 +aufs assigns the inode number i999 for fileA. Then aufs writes 999 as
1419 +4(8) bytes at 111 * 4(8) bytes offset in the xino file.
1420 +
1421 +When the inode numbers are not contiguous, the xino file will be sparse
1422 +which has a hole in it and doesn't consume as much disk space as it
1423 +might appear. If your branch filesystem consumes disk space for such
1424 +holes, then you should specify 'xino=' option at mounting aufs.
1425 +
1426 +Aufs has a mount option to free the disk blocks for such holes in XINO
1427 +files on tmpfs or ramdisk. But it is not so effective actually. If you
1428 +meet a problem of disk shortage due to XINO files, then you should try
1429 +"tmpfs-ino.patch" (and "vfs-ino.patch" too) in aufs4-standalone.git.
1430 +The patch localizes the assignment inumbers per tmpfs-mount and avoid
1431 +the holes in XINO files.
1432 +
1433 +Also a writable branch has three kinds of "whiteout bases". All these
1434 +are existed when the branch is joined to aufs, and their names are
1435 +whiteout-ed doubly, so that users will never see their names in aufs
1436 +hierarchy.
1437 +1. a regular file which will be hardlinked to all whiteouts.
1438 +2. a directory to store a pseudo-link.
1439 +3. a directory to store an "orphan"-ed file temporary.
1440 +
1441 +1. Whiteout Base
1442 +   When you remove a file on a readonly branch, aufs handles it as a
1443 +   logical deletion and creates a whiteout on the upper writable branch
1444 +   as a hardlink of this file in order not to consume inode on the
1445 +   writable branch.
1446 +2. Pseudo-link Dir
1447 +   See below, Pseudo-link.
1448 +3. Step-Parent Dir
1449 +   When "fileC" exists on the lower readonly branch only and it is
1450 +   opened and removed with its parent dir, and then user writes
1451 +   something into it, then aufs copies-up fileC to this
1452 +   directory. Because there is no other dir to store fileC. After
1453 +   creating a file under this dir, the file is unlinked.
1454 +
1455 +Because aufs supports manipulating branches, ie. add/delete/change
1456 +dynamically, a branch has its own id. When the branch order changes,
1457 +aufs finds the new index by searching the branch id.
1458 +
1459 +
1460 +Pseudo-link
1461 +----------------------------------------------------------------------
1462 +Assume "fileA" exists on the lower readonly branch only and it is
1463 +hardlinked to "fileB" on the branch. When you write something to fileA,
1464 +aufs copies-up it to the upper writable branch. Additionally aufs
1465 +creates a hardlink under the Pseudo-link Directory of the writable
1466 +branch. The inode of a pseudo-link is kept in aufs super_block as a
1467 +simple list. If fileB is read after unlinking fileA, aufs returns
1468 +filedata from the pseudo-link instead of the lower readonly
1469 +branch. Because the pseudo-link is based upon the inode, to keep the
1470 +inode number by xino (see above) is essentially necessary.
1471 +
1472 +All the hardlinks under the Pseudo-link Directory of the writable branch
1473 +should be restored in a proper location later. Aufs provides a utility
1474 +to do this. The userspace helpers executed at remounting and unmounting
1475 +aufs by default.
1476 +During this utility is running, it puts aufs into the pseudo-link
1477 +maintenance mode. In this mode, only the process which began the
1478 +maintenance mode (and its child processes) is allowed to operate in
1479 +aufs. Some other processes which are not related to the pseudo-link will
1480 +be allowed to run too, but the rest have to return an error or wait
1481 +until the maintenance mode ends. If a process already acquires an inode
1482 +mutex (in VFS), it has to return an error.
1483 +
1484 +
1485 +XIB(external inode number bitmap)
1486 +----------------------------------------------------------------------
1487 +Addition to the xino file per a branch, aufs has an external inode number
1488 +bitmap in a superblock object. It is also an internal file such like a
1489 +xino file.
1490 +It is a simple bitmap to mark whether the aufs inode number is in-use or
1491 +not.
1492 +To reduce the file I/O, aufs prepares a single memory page to cache xib.
1493 +
1494 +As well as XINO files, aufs has a feature to truncate/refresh XIB to
1495 +reduce the number of consumed disk blocks for these files.
1496 +
1497 +
1498 +Virtual or Vertical Dir, and Readdir in Userspace
1499 +----------------------------------------------------------------------
1500 +In order to support multiple layers (branches), aufs readdir operation
1501 +constructs a virtual dir block on memory. For readdir, aufs calls
1502 +vfs_readdir() internally for each dir on branches, merges their entries
1503 +with eliminating the whiteout-ed ones, and sets it to file (dir)
1504 +object. So the file object has its entry list until it is closed. The
1505 +entry list will be updated when the file position is zero and becomes
1506 +obsoleted. This decision is made in aufs automatically.
1507 +
1508 +The dynamically allocated memory block for the name of entries has a
1509 +unit of 512 bytes (by default) and stores the names contiguously (no
1510 +padding). Another block for each entry is handled by kmem_cache too.
1511 +During building dir blocks, aufs creates hash list and judging whether
1512 +the entry is whiteouted by its upper branch or already listed.
1513 +The merged result is cached in the corresponding inode object and
1514 +maintained by a customizable life-time option.
1515 +
1516 +Some people may call it can be a security hole or invite DoS attack
1517 +since the opened and once readdir-ed dir (file object) holds its entry
1518 +list and becomes a pressure for system memory. But I'd say it is similar
1519 +to files under /proc or /sys. The virtual files in them also holds a
1520 +memory page (generally) while they are opened. When an idea to reduce
1521 +memory for them is introduced, it will be applied to aufs too.
1522 +For those who really hate this situation, I've developed readdir(3)
1523 +library which operates this merging in userspace. You just need to set
1524 +LD_PRELOAD environment variable, and aufs will not consume no memory in
1525 +kernel space for readdir(3).
1526 +
1527 +
1528 +Workqueue
1529 +----------------------------------------------------------------------
1530 +Aufs sometimes requires privilege access to a branch. For instance,
1531 +in copy-up/down operation. When a user process is going to make changes
1532 +to a file which exists in the lower readonly branch only, and the mode
1533 +of one of ancestor directories may not be writable by a user
1534 +process. Here aufs copy-up the file with its ancestors and they may
1535 +require privilege to set its owner/group/mode/etc.
1536 +This is a typical case of a application character of aufs (see
1537 +Introduction).
1538 +
1539 +Aufs uses workqueue synchronously for this case. It creates its own
1540 +workqueue. The workqueue is a kernel thread and has privilege. Aufs
1541 +passes the request to call mkdir or write (for example), and wait for
1542 +its completion. This approach solves a problem of a signal handler
1543 +simply.
1544 +If aufs didn't adopt the workqueue and changed the privilege of the
1545 +process, then the process may receive the unexpected SIGXFSZ or other
1546 +signals.
1547 +
1548 +Also aufs uses the system global workqueue ("events" kernel thread) too
1549 +for asynchronous tasks, such like handling inotify/fsnotify, re-creating a
1550 +whiteout base and etc. This is unrelated to a privilege.
1551 +Most of aufs operation tries acquiring a rw_semaphore for aufs
1552 +superblock at the beginning, at the same time waits for the completion
1553 +of all queued asynchronous tasks.
1554 +
1555 +
1556 +Whiteout
1557 +----------------------------------------------------------------------
1558 +The whiteout in aufs is very similar to Unionfs's. That is represented
1559 +by its filename. UnionMount takes an approach of a file mode, but I am
1560 +afraid several utilities (find(1) or something) will have to support it.
1561 +
1562 +Basically the whiteout represents "logical deletion" which stops aufs to
1563 +lookup further, but also it represents "dir is opaque" which also stop
1564 +further lookup.
1565 +
1566 +In aufs, rmdir(2) and rename(2) for dir uses whiteout alternatively.
1567 +In order to make several functions in a single systemcall to be
1568 +revertible, aufs adopts an approach to rename a directory to a temporary
1569 +unique whiteouted name.
1570 +For example, in rename(2) dir where the target dir already existed, aufs
1571 +renames the target dir to a temporary unique whiteouted name before the
1572 +actual rename on a branch, and then handles other actions (make it opaque,
1573 +update the attributes, etc). If an error happens in these actions, aufs
1574 +simply renames the whiteouted name back and returns an error. If all are
1575 +succeeded, aufs registers a function to remove the whiteouted unique
1576 +temporary name completely and asynchronously to the system global
1577 +workqueue.
1578 +
1579 +
1580 +Copy-up
1581 +----------------------------------------------------------------------
1582 +It is a well-known feature or concept.
1583 +When user modifies a file on a readonly branch, aufs operate "copy-up"
1584 +internally and makes change to the new file on the upper writable branch.
1585 +When the trigger systemcall does not update the timestamps of the parent
1586 +dir, aufs reverts it after copy-up.
1587 +
1588 +
1589 +Move-down (aufs3.9 and later)
1590 +----------------------------------------------------------------------
1591 +"Copy-up" is one of the essential feature in aufs. It copies a file from
1592 +the lower readonly branch to the upper writable branch when a user
1593 +changes something about the file.
1594 +"Move-down" is an opposite action of copy-up. Basically this action is
1595 +ran manually instead of automatically and internally.
1596 +For desgin and implementation, aufs has to consider these issues.
1597 +- whiteout for the file may exist on the lower branch.
1598 +- ancestor directories may not exist on the lower branch.
1599 +- diropq for the ancestor directories may exist on the upper branch.
1600 +- free space on the lower branch will reduce.
1601 +- another access to the file may happen during moving-down, including
1602 +  UDBA (see "Revalidate Dentry and UDBA").
1603 +- the file should not be hard-linked nor pseudo-linked. they should be
1604 +  handled by auplink utility later.
1605 +
1606 +Sometimes users want to move-down a file from the upper writable branch
1607 +to the lower readonly or writable branch. For instance,
1608 +- the free space of the upper writable branch is going to run out.
1609 +- create a new intermediate branch between the upper and lower branch.
1610 +- etc.
1611 +
1612 +For this purpose, use "aumvdown" command in aufs-util.git.
1613 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03atomic_open.txt linux/Documentation/filesystems/aufs/design/03atomic_open.txt
1614 --- /usr/share/empty/Documentation/filesystems/aufs/design/03atomic_open.txt    1970-01-01 01:00:00.000000000 +0100
1615 +++ linux/Documentation/filesystems/aufs/design/03atomic_open.txt       2017-05-06 22:16:52.774886674 +0200
1616 @@ -0,0 +1,85 @@
1617 +
1618 +# Copyright (C) 2015-2017 Junjiro R. Okajima
1619 +#
1620 +# This program is free software; you can redistribute it and/or modify
1621 +# it under the terms of the GNU General Public License as published by
1622 +# the Free Software Foundation; either version 2 of the License, or
1623 +# (at your option) any later version.
1624 +#
1625 +# This program is distributed in the hope that it will be useful,
1626 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1627 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1628 +# GNU General Public License for more details.
1629 +#
1630 +# You should have received a copy of the GNU General Public License
1631 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1632 +
1633 +Support for a branch who has its ->atomic_open()
1634 +----------------------------------------------------------------------
1635 +The filesystems who implement its ->atomic_open() are not majority. For
1636 +example NFSv4 does, and aufs should call NFSv4 ->atomic_open,
1637 +particularly for open(O_CREAT|O_EXCL, 0400) case. Other than
1638 +->atomic_open(), NFSv4 returns an error for this open(2). While I am not
1639 +sure whether all filesystems who have ->atomic_open() behave like this,
1640 +but NFSv4 surely returns the error.
1641 +
1642 +In order to support ->atomic_open() for aufs, there are a few
1643 +approaches.
1644 +
1645 +A. Introduce aufs_atomic_open()
1646 +   - calls one of VFS:do_last(), lookup_open() or atomic_open() for
1647 +     branch fs.
1648 +B. Introduce aufs_atomic_open() calling create, open and chmod. this is
1649 +   an aufs user Pip Cet's approach
1650 +   - calls aufs_create(), VFS finish_open() and notify_change().
1651 +   - pass fake-mode to finish_open(), and then correct the mode by
1652 +     notify_change().
1653 +C. Extend aufs_open() to call branch fs's ->atomic_open()
1654 +   - no aufs_atomic_open().
1655 +   - aufs_lookup() registers the TID to an aufs internal object.
1656 +   - aufs_create() does nothing when the matching TID is registered, but
1657 +     registers the mode.
1658 +   - aufs_open() calls branch fs's ->atomic_open() when the matching
1659 +     TID is registered.
1660 +D. Extend aufs_open() to re-try branch fs's ->open() with superuser's
1661 +   credential
1662 +   - no aufs_atomic_open().
1663 +   - aufs_create() registers the TID to an internal object. this info
1664 +     represents "this process created this file just now."
1665 +   - when aufs gets EACCES from branch fs's ->open(), then confirm the
1666 +     registered TID and re-try open() with superuser's credential.
1667 +
1668 +Pros and cons for each approach.
1669 +
1670 +A.
1671 +   - straightforward but highly depends upon VFS internal.
1672 +   - the atomic behavaiour is kept.
1673 +   - some of parameters such as nameidata are hard to reproduce for
1674 +     branch fs.
1675 +   - large overhead.
1676 +B.
1677 +   - easy to implement.
1678 +   - the atomic behavaiour is lost.
1679 +C.
1680 +   - the atomic behavaiour is kept.
1681 +   - dirty and tricky.
1682 +   - VFS checks whether the file is created correctly after calling
1683 +     ->create(), which means this approach doesn't work.
1684 +D.
1685 +   - easy to implement.
1686 +   - the atomic behavaiour is lost.
1687 +   - to open a file with superuser's credential and give it to a user
1688 +     process is a bad idea, since the file object keeps the credential
1689 +     in it. It may affect LSM or something. This approach doesn't work
1690 +     either.
1691 +
1692 +The approach A is ideal, but it hard to implement. So here is a
1693 +variation of A, which is to be implemented.
1694 +
1695 +A-1. Introduce aufs_atomic_open()
1696 +     - calls branch fs ->atomic_open() if exists. otherwise calls
1697 +       vfs_create() and finish_open().
1698 +     - the demerit is that the several checks after branch fs
1699 +       ->atomic_open() are lost. in the ordinary case, the checks are
1700 +       done by VFS:do_last(), lookup_open() and atomic_open(). some can
1701 +       be implemented in aufs, but not all I am afraid.
1702 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt linux/Documentation/filesystems/aufs/design/03lookup.txt
1703 --- /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt 1970-01-01 01:00:00.000000000 +0100
1704 +++ linux/Documentation/filesystems/aufs/design/03lookup.txt    2017-05-06 22:16:52.774886674 +0200
1705 @@ -0,0 +1,113 @@
1706 +
1707 +# Copyright (C) 2005-2017 Junjiro R. Okajima
1708 +#
1709 +# This program is free software; you can redistribute it and/or modify
1710 +# it under the terms of the GNU General Public License as published by
1711 +# the Free Software Foundation; either version 2 of the License, or
1712 +# (at your option) any later version.
1713 +#
1714 +# This program is distributed in the hope that it will be useful,
1715 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1716 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1717 +# GNU General Public License for more details.
1718 +#
1719 +# You should have received a copy of the GNU General Public License
1720 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1721 +
1722 +Lookup in a Branch
1723 +----------------------------------------------------------------------
1724 +Since aufs has a character of sub-VFS (see Introduction), it operates
1725 +lookup for branches as VFS does. It may be a heavy work. But almost all
1726 +lookup operation in aufs is the simplest case, ie. lookup only an entry
1727 +directly connected to its parent. Digging down the directory hierarchy
1728 +is unnecessary. VFS has a function lookup_one_len() for that use, and
1729 +aufs calls it.
1730 +
1731 +When a branch is a remote filesystem, aufs basically relies upon its
1732 +->d_revalidate(), also aufs forces the hardest revalidate tests for
1733 +them.
1734 +For d_revalidate, aufs implements three levels of revalidate tests. See
1735 +"Revalidate Dentry and UDBA" in detail.
1736 +
1737 +
1738 +Test Only the Highest One for the Directory Permission (dirperm1 option)
1739 +----------------------------------------------------------------------
1740 +Let's try case study.
1741 +- aufs has two branches, upper readwrite and lower readonly.
1742 +  /au = /rw + /ro
1743 +- "dirA" exists under /ro, but /rw. and its mode is 0700.
1744 +- user invoked "chmod a+rx /au/dirA"
1745 +- the internal copy-up is activated and "/rw/dirA" is created and its
1746 +  permission bits are set to world readable.
1747 +- then "/au/dirA" becomes world readable?
1748 +
1749 +In this case, /ro/dirA is still 0700 since it exists in readonly branch,
1750 +or it may be a natively readonly filesystem. If aufs respects the lower
1751 +branch, it should not respond readdir request from other users. But user
1752 +allowed it by chmod. Should really aufs rejects showing the entries
1753 +under /ro/dirA?
1754 +
1755 +To be honest, I don't have a good solution for this case. So aufs
1756 +implements 'dirperm1' and 'nodirperm1' mount options, and leave it to
1757 +users.
1758 +When dirperm1 is specified, aufs checks only the highest one for the
1759 +directory permission, and shows the entries. Otherwise, as usual, checks
1760 +every dir existing on all branches and rejects the request.
1761 +
1762 +As a side effect, dirperm1 option improves the performance of aufs
1763 +because the number of permission check is reduced when the number of
1764 +branch is many.
1765 +
1766 +
1767 +Revalidate Dentry and UDBA (User's Direct Branch Access)
1768 +----------------------------------------------------------------------
1769 +Generally VFS helpers re-validate a dentry as a part of lookup.
1770 +0. digging down the directory hierarchy.
1771 +1. lock the parent dir by its i_mutex.
1772 +2. lookup the final (child) entry.
1773 +3. revalidate it.
1774 +4. call the actual operation (create, unlink, etc.)
1775 +5. unlock the parent dir
1776 +
1777 +If the filesystem implements its ->d_revalidate() (step 3), then it is
1778 +called. Actually aufs implements it and checks the dentry on a branch is
1779 +still valid.
1780 +But it is not enough. Because aufs has to release the lock for the
1781 +parent dir on a branch at the end of ->lookup() (step 2) and
1782 +->d_revalidate() (step 3) while the i_mutex of the aufs dir is still
1783 +held by VFS.
1784 +If the file on a branch is changed directly, eg. bypassing aufs, after
1785 +aufs released the lock, then the subsequent operation may cause
1786 +something unpleasant result.
1787 +
1788 +This situation is a result of VFS architecture, ->lookup() and
1789 +->d_revalidate() is separated. But I never say it is wrong. It is a good
1790 +design from VFS's point of view. It is just not suitable for sub-VFS
1791 +character in aufs.
1792 +
1793 +Aufs supports such case by three level of revalidation which is
1794 +selectable by user.
1795 +1. Simple Revalidate
1796 +   Addition to the native flow in VFS's, confirm the child-parent
1797 +   relationship on the branch just after locking the parent dir on the
1798 +   branch in the "actual operation" (step 4). When this validation
1799 +   fails, aufs returns EBUSY. ->d_revalidate() (step 3) in aufs still
1800 +   checks the validation of the dentry on branches.
1801 +2. Monitor Changes Internally by Inotify/Fsnotify
1802 +   Addition to above, in the "actual operation" (step 4) aufs re-lookup
1803 +   the dentry on the branch, and returns EBUSY if it finds different
1804 +   dentry.
1805 +   Additionally, aufs sets the inotify/fsnotify watch for every dir on branches
1806 +   during it is in cache. When the event is notified, aufs registers a
1807 +   function to kernel 'events' thread by schedule_work(). And the
1808 +   function sets some special status to the cached aufs dentry and inode
1809 +   private data. If they are not cached, then aufs has nothing to
1810 +   do. When the same file is accessed through aufs (step 0-3) later,
1811 +   aufs will detect the status and refresh all necessary data.
1812 +   In this mode, aufs has to ignore the event which is fired by aufs
1813 +   itself.
1814 +3. No Extra Validation
1815 +   This is the simplest test and doesn't add any additional revalidation
1816 +   test, and skip the revalidation in step 4. It is useful and improves
1817 +   aufs performance when system surely hide the aufs branches from user,
1818 +   by over-mounting something (or another method).
1819 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt linux/Documentation/filesystems/aufs/design/04branch.txt
1820 --- /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt 1970-01-01 01:00:00.000000000 +0100
1821 +++ linux/Documentation/filesystems/aufs/design/04branch.txt    2017-05-06 22:16:52.774886674 +0200
1822 @@ -0,0 +1,74 @@
1823 +
1824 +# Copyright (C) 2005-2017 Junjiro R. Okajima
1825 +#
1826 +# This program is free software; you can redistribute it and/or modify
1827 +# it under the terms of the GNU General Public License as published by
1828 +# the Free Software Foundation; either version 2 of the License, or
1829 +# (at your option) any later version.
1830 +#
1831 +# This program is distributed in the hope that it will be useful,
1832 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1833 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1834 +# GNU General Public License for more details.
1835 +#
1836 +# You should have received a copy of the GNU General Public License
1837 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1838 +
1839 +Branch Manipulation
1840 +
1841 +Since aufs supports dynamic branch manipulation, ie. add/remove a branch
1842 +and changing its permission/attribute, there are a lot of works to do.
1843 +
1844 +
1845 +Add a Branch
1846 +----------------------------------------------------------------------
1847 +o Confirm the adding dir exists outside of aufs, including loopback
1848 +  mount, and its various attributes.
1849 +o Initialize the xino file and whiteout bases if necessary.
1850 +  See struct.txt.
1851 +
1852 +o Check the owner/group/mode of the directory
1853 +  When the owner/group/mode of the adding directory differs from the
1854 +  existing branch, aufs issues a warning because it may impose a
1855 +  security risk.
1856 +  For example, when a upper writable branch has a world writable empty
1857 +  top directory, a malicious user can create any files on the writable
1858 +  branch directly, like copy-up and modify manually. If something like
1859 +  /etc/{passwd,shadow} exists on the lower readonly branch but the upper
1860 +  writable branch, and the writable branch is world-writable, then a
1861 +  malicious guy may create /etc/passwd on the writable branch directly
1862 +  and the infected file will be valid in aufs.
1863 +  I am afraid it can be a security issue, but aufs can do nothing except
1864 +  producing a warning.
1865 +
1866 +
1867 +Delete a Branch
1868 +----------------------------------------------------------------------
1869 +o Confirm the deleting branch is not busy
1870 +  To be general, there is one merit to adopt "remount" interface to
1871 +  manipulate branches. It is to discard caches. At deleting a branch,
1872 +  aufs checks the still cached (and connected) dentries and inodes. If
1873 +  there are any, then they are all in-use. An inode without its
1874 +  corresponding dentry can be alive alone (for example, inotify/fsnotify case).
1875 +
1876 +  For the cached one, aufs checks whether the same named entry exists on
1877 +  other branches.
1878 +  If the cached one is a directory, because aufs provides a merged view
1879 +  to users, as long as one dir is left on any branch aufs can show the
1880 +  dir to users. In this case, the branch can be removed from aufs.
1881 +  Otherwise aufs rejects deleting the branch.
1882 +
1883 +  If any file on the deleting branch is opened by aufs, then aufs
1884 +  rejects deleting.
1885 +
1886 +
1887 +Modify the Permission of a Branch
1888 +----------------------------------------------------------------------
1889 +o Re-initialize or remove the xino file and whiteout bases if necessary.
1890 +  See struct.txt.
1891 +
1892 +o rw --> ro: Confirm the modifying branch is not busy
1893 +  Aufs rejects the request if any of these conditions are true.
1894 +  - a file on the branch is mmap-ed.
1895 +  - a regular file on the branch is opened for write and there is no
1896 +    same named entry on the upper branch.
1897 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt linux/Documentation/filesystems/aufs/design/05wbr_policy.txt
1898 --- /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt     1970-01-01 01:00:00.000000000 +0100
1899 +++ linux/Documentation/filesystems/aufs/design/05wbr_policy.txt        2017-05-06 22:16:52.774886674 +0200
1900 @@ -0,0 +1,64 @@
1901 +
1902 +# Copyright (C) 2005-2017 Junjiro R. Okajima
1903 +#
1904 +# This program is free software; you can redistribute it and/or modify
1905 +# it under the terms of the GNU General Public License as published by
1906 +# the Free Software Foundation; either version 2 of the License, or
1907 +# (at your option) any later version.
1908 +#
1909 +# This program is distributed in the hope that it will be useful,
1910 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1911 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1912 +# GNU General Public License for more details.
1913 +#
1914 +# You should have received a copy of the GNU General Public License
1915 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1916 +
1917 +Policies to Select One among Multiple Writable Branches
1918 +----------------------------------------------------------------------
1919 +When the number of writable branch is more than one, aufs has to decide
1920 +the target branch for file creation or copy-up. By default, the highest
1921 +writable branch which has the parent (or ancestor) dir of the target
1922 +file is chosen (top-down-parent policy).
1923 +By user's request, aufs implements some other policies to select the
1924 +writable branch, for file creation several policies, round-robin,
1925 +most-free-space, and other policies. For copy-up, top-down-parent,
1926 +bottom-up-parent, bottom-up and others.
1927 +
1928 +As expected, the round-robin policy selects the branch in circular. When
1929 +you have two writable branches and creates 10 new files, 5 files will be
1930 +created for each branch. mkdir(2) systemcall is an exception. When you
1931 +create 10 new directories, all will be created on the same branch.
1932 +And the most-free-space policy selects the one which has most free
1933 +space among the writable branches. The amount of free space will be
1934 +checked by aufs internally, and users can specify its time interval.
1935 +
1936 +The policies for copy-up is more simple,
1937 +top-down-parent is equivalent to the same named on in create policy,
1938 +bottom-up-parent selects the writable branch where the parent dir
1939 +exists and the nearest upper one from the copyup-source,
1940 +bottom-up selects the nearest upper writable branch from the
1941 +copyup-source, regardless the existence of the parent dir.
1942 +
1943 +There are some rules or exceptions to apply these policies.
1944 +- If there is a readonly branch above the policy-selected branch and
1945 +  the parent dir is marked as opaque (a variation of whiteout), or the
1946 +  target (creating) file is whiteout-ed on the upper readonly branch,
1947 +  then the result of the policy is ignored and the target file will be
1948 +  created on the nearest upper writable branch than the readonly branch.
1949 +- If there is a writable branch above the policy-selected branch and
1950 +  the parent dir is marked as opaque or the target file is whiteouted
1951 +  on the branch, then the result of the policy is ignored and the target
1952 +  file will be created on the highest one among the upper writable
1953 +  branches who has diropq or whiteout. In case of whiteout, aufs removes
1954 +  it as usual.
1955 +- link(2) and rename(2) systemcalls are exceptions in every policy.
1956 +  They try selecting the branch where the source exists as possible
1957 +  since copyup a large file will take long time. If it can't be,
1958 +  ie. the branch where the source exists is readonly, then they will
1959 +  follow the copyup policy.
1960 +- There is an exception for rename(2) when the target exists.
1961 +  If the rename target exists, aufs compares the index of the branches
1962 +  where the source and the target exists and selects the higher
1963 +  one. If the selected branch is readonly, then aufs follows the
1964 +  copyup policy.
1965 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt linux/Documentation/filesystems/aufs/design/06fhsm.txt
1966 --- /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt   1970-01-01 01:00:00.000000000 +0100
1967 +++ linux/Documentation/filesystems/aufs/design/06fhsm.txt      2017-05-06 22:16:52.774886674 +0200
1968 @@ -0,0 +1,120 @@
1969 +
1970 +# Copyright (C) 2011-2017 Junjiro R. Okajima
1971 +#
1972 +# This program is free software; you can redistribute it and/or modify
1973 +# it under the terms of the GNU General Public License as published by
1974 +# the Free Software Foundation; either version 2 of the License, or
1975 +# (at your option) any later version.
1976 +#
1977 +# This program is distributed in the hope that it will be useful,
1978 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1979 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1980 +# GNU General Public License for more details.
1981 +#
1982 +# You should have received a copy of the GNU General Public License
1983 +# along with this program; if not, write to the Free Software
1984 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
1985 +
1986 +
1987 +File-based Hierarchical Storage Management (FHSM)
1988 +----------------------------------------------------------------------
1989 +Hierarchical Storage Management (or HSM) is a well-known feature in the
1990 +storage world. Aufs provides this feature as file-based with multiple
1991 +writable branches, based upon the principle of "Colder, the Lower".
1992 +Here the word "colder" means that the less used files, and "lower" means
1993 +that the position in the order of the stacked branches vertically.
1994 +These multiple writable branches are prioritized, ie. the topmost one
1995 +should be the fastest drive and be used heavily.
1996 +
1997 +o Characters in aufs FHSM story
1998 +- aufs itself and a new branch attribute.
1999 +- a new ioctl interface to move-down and to establish a connection with
2000 +  the daemon ("move-down" is a converse of "copy-up").
2001 +- userspace tool and daemon.
2002 +
2003 +The userspace daemon establishes a connection with aufs and waits for
2004 +the notification. The notified information is very similar to struct
2005 +statfs containing the number of consumed blocks and inodes.
2006 +When the consumed blocks/inodes of a branch exceeds the user-specified
2007 +upper watermark, the daemon activates its move-down process until the
2008 +consumed blocks/inodes reaches the user-specified lower watermark.
2009 +
2010 +The actual move-down is done by aufs based upon the request from
2011 +user-space since we need to maintain the inode number and the internal
2012 +pointer arrays in aufs.
2013 +
2014 +Currently aufs FHSM handles the regular files only. Additionally they
2015 +must not be hard-linked nor pseudo-linked.
2016 +
2017 +
2018 +o Cowork of aufs and the user-space daemon
2019 +  During the userspace daemon established the connection, aufs sends a
2020 +  small notification to it whenever aufs writes something into the
2021 +  writable branch. But it may cost high since aufs issues statfs(2)
2022 +  internally. So user can specify a new option to cache the
2023 +  info. Actually the notification is controlled by these factors.
2024 +  + the specified cache time.
2025 +  + classified as "force" by aufs internally.
2026 +  Until the specified time expires, aufs doesn't send the info
2027 +  except the forced cases. When aufs decide forcing, the info is always
2028 +  notified to userspace.
2029 +  For example, the number of free inodes is generally large enough and
2030 +  the shortage of it happens rarely. So aufs doesn't force the
2031 +  notification when creating a new file, directory and others. This is
2032 +  the typical case which aufs doesn't force.
2033 +  When aufs writes the actual filedata and the files consumes any of new
2034 +  blocks, the aufs forces notifying.
2035 +
2036 +
2037 +o Interfaces in aufs
2038 +- New branch attribute.
2039 +  + fhsm
2040 +    Specifies that the branch is managed by FHSM feature. In other word,
2041 +    participant in the FHSM.
2042 +    When nofhsm is set to the branch, it will not be the source/target
2043 +    branch of the move-down operation. This attribute is set
2044 +    independently from coo and moo attributes, and if you want full
2045 +    FHSM, you should specify them as well.
2046 +- New mount option.
2047 +  + fhsm_sec
2048 +    Specifies a second to suppress many less important info to be
2049 +    notified.
2050 +- New ioctl.
2051 +  + AUFS_CTL_FHSM_FD
2052 +    create a new file descriptor which userspace can read the notification
2053 +    (a subset of struct statfs) from aufs.
2054 +- Module parameter 'brs'
2055 +  It has to be set to 1. Otherwise the new mount option 'fhsm' will not
2056 +  be set.
2057 +- mount helpers /sbin/mount.aufs and /sbin/umount.aufs
2058 +  When there are two or more branches with fhsm attributes,
2059 +  /sbin/mount.aufs invokes the user-space daemon and /sbin/umount.aufs
2060 +  terminates it. As a result of remounting and branch-manipulation, the
2061 +  number of branches with fhsm attribute can be one. In this case,
2062 +  /sbin/mount.aufs will terminate the user-space daemon.
2063 +
2064 +
2065 +Finally the operation is done as these steps in kernel-space.
2066 +- make sure that,
2067 +  + no one else is using the file.
2068 +  + the file is not hard-linked.
2069 +  + the file is not pseudo-linked.
2070 +  + the file is a regular file.
2071 +  + the parent dir is not opaqued.
2072 +- find the target writable branch.
2073 +- make sure the file is not whiteout-ed by the upper (than the target)
2074 +  branch.
2075 +- make the parent dir on the target branch.
2076 +- mutex lock the inode on the branch.
2077 +- unlink the whiteout on the target branch (if exists).
2078 +- lookup and create the whiteout-ed temporary name on the target branch.
2079 +- copy the file as the whiteout-ed temporary name on the target branch.
2080 +- rename the whiteout-ed temporary name to the original name.
2081 +- unlink the file on the source branch.
2082 +- maintain the internal pointer array and the external inode number
2083 +  table (XINO).
2084 +- maintain the timestamps and other attributes of the parent dir and the
2085 +  file.
2086 +
2087 +And of course, in every step, an error may happen. So the operation
2088 +should restore the original file state after an error happens.
2089 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt linux/Documentation/filesystems/aufs/design/06mmap.txt
2090 --- /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt   1970-01-01 01:00:00.000000000 +0100
2091 +++ linux/Documentation/filesystems/aufs/design/06mmap.txt      2017-05-06 22:16:52.774886674 +0200
2092 @@ -0,0 +1,72 @@
2093 +
2094 +# Copyright (C) 2005-2017 Junjiro R. Okajima
2095 +#
2096 +# This program is free software; you can redistribute it and/or modify
2097 +# it under the terms of the GNU General Public License as published by
2098 +# the Free Software Foundation; either version 2 of the License, or
2099 +# (at your option) any later version.
2100 +#
2101 +# This program is distributed in the hope that it will be useful,
2102 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2103 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2104 +# GNU General Public License for more details.
2105 +#
2106 +# You should have received a copy of the GNU General Public License
2107 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2108 +
2109 +mmap(2) -- File Memory Mapping
2110 +----------------------------------------------------------------------
2111 +In aufs, the file-mapped pages are handled by a branch fs directly, no
2112 +interaction with aufs. It means aufs_mmap() calls the branch fs's
2113 +->mmap().
2114 +This approach is simple and good, but there is one problem.
2115 +Under /proc, several entries show the mmapped files by its path (with
2116 +device and inode number), and the printed path will be the path on the
2117 +branch fs's instead of virtual aufs's.
2118 +This is not a problem in most cases, but some utilities lsof(1) (and its
2119 +user) may expect the path on aufs.
2120 +
2121 +To address this issue, aufs adds a new member called vm_prfile in struct
2122 +vm_area_struct (and struct vm_region). The original vm_file points to
2123 +the file on the branch fs in order to handle everything correctly as
2124 +usual. The new vm_prfile points to a virtual file in aufs, and the
2125 +show-functions in procfs refers to vm_prfile if it is set.
2126 +Also we need to maintain several other places where touching vm_file
2127 +such like
2128 +- fork()/clone() copies vma and the reference count of vm_file is
2129 +  incremented.
2130 +- merging vma maintains the ref count too.
2131 +
2132 +This is not a good approach. It just fakes the printed path. But it
2133 +leaves all behaviour around f_mapping unchanged. This is surely an
2134 +advantage.
2135 +Actually aufs had adopted another complicated approach which calls
2136 +generic_file_mmap() and handles struct vm_operations_struct. In this
2137 +approach, aufs met a hard problem and I could not solve it without
2138 +switching the approach.
2139 +
2140 +There may be one more another approach which is
2141 +- bind-mount the branch-root onto the aufs-root internally
2142 +- grab the new vfsmount (ie. struct mount)
2143 +- lazy-umount the branch-root internally
2144 +- in open(2) the aufs-file, open the branch-file with the hidden
2145 +  vfsmount (instead of the original branch's vfsmount)
2146 +- ideally this "bind-mount and lazy-umount" should be done atomically,
2147 +  but it may be possible from userspace by the mount helper.
2148 +
2149 +Adding the internal hidden vfsmount and using it in opening a file, the
2150 +file path under /proc will be printed correctly. This approach looks
2151 +smarter, but is not possible I am afraid.
2152 +- aufs-root may be bind-mount later. when it happens, another hidden
2153 +  vfsmount will be required.
2154 +- it is hard to get the chance to bind-mount and lazy-umount
2155 +  + in kernel-space, FS can have vfsmount in open(2) via
2156 +    file->f_path, and aufs can know its vfsmount. But several locks are
2157 +    already acquired, and if aufs tries to bind-mount and lazy-umount
2158 +    here, then it may cause a deadlock.
2159 +  + in user-space, bind-mount doesn't invoke the mount helper.
2160 +- since /proc shows dev and ino, aufs has to give vma these info. it
2161 +  means a new member vm_prinode will be necessary. this is essentially
2162 +  equivalent to vm_prfile described above.
2163 +
2164 +I have to give up this "looks-smater" approach.
2165 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt linux/Documentation/filesystems/aufs/design/06xattr.txt
2166 --- /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt  1970-01-01 01:00:00.000000000 +0100
2167 +++ linux/Documentation/filesystems/aufs/design/06xattr.txt     2017-05-06 22:16:52.774886674 +0200
2168 @@ -0,0 +1,96 @@
2169 +
2170 +# Copyright (C) 2014-2017 Junjiro R. Okajima
2171 +#
2172 +# This program is free software; you can redistribute it and/or modify
2173 +# it under the terms of the GNU General Public License as published by
2174 +# the Free Software Foundation; either version 2 of the License, or
2175 +# (at your option) any later version.
2176 +#
2177 +# This program is distributed in the hope that it will be useful,
2178 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2179 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2180 +# GNU General Public License for more details.
2181 +#
2182 +# You should have received a copy of the GNU General Public License
2183 +# along with this program; if not, write to the Free Software
2184 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
2185 +
2186 +
2187 +Listing XATTR/EA and getting the value
2188 +----------------------------------------------------------------------
2189 +For the inode standard attributes (owner, group, timestamps, etc.), aufs
2190 +shows the values from the topmost existing file. This behaviour is good
2191 +for the non-dir entries since the bahaviour exactly matches the shown
2192 +information. But for the directories, aufs considers all the same named
2193 +entries on the lower branches. Which means, if one of the lower entry
2194 +rejects readdir call, then aufs returns an error even if the topmost
2195 +entry allows it. This behaviour is necessary to respect the branch fs's
2196 +security, but can make users confused since the user-visible standard
2197 +attributes don't match the behaviour.
2198 +To address this issue, aufs has a mount option called dirperm1 which
2199 +checks the permission for the topmost entry only, and ignores the lower
2200 +entry's permission.
2201 +
2202 +A similar issue can happen around XATTR.
2203 +getxattr(2) and listxattr(2) families behave as if dirperm1 option is
2204 +always set. Otherwise these very unpleasant situation would happen.
2205 +- listxattr(2) may return the duplicated entries.
2206 +- users may not be able to remove or reset the XATTR forever,
2207 +
2208 +
2209 +XATTR/EA support in the internal (copy,move)-(up,down)
2210 +----------------------------------------------------------------------
2211 +Generally the extended attributes of inode are categorized as these.
2212 +- "security" for LSM and capability.
2213 +- "system" for posix ACL, 'acl' mount option is required for the branch
2214 +  fs generally.
2215 +- "trusted" for userspace, CAP_SYS_ADMIN is required.
2216 +- "user" for userspace, 'user_xattr' mount option is required for the
2217 +  branch fs generally.
2218 +
2219 +Moreover there are some other categories. Aufs handles these rather
2220 +unpopular categories as the ordinary ones, ie. there is no special
2221 +condition nor exception.
2222 +
2223 +In copy-up, the support for XATTR on the dst branch may differ from the
2224 +src branch. In this case, the copy-up operation will get an error and
2225 +the original user operation which triggered the copy-up will fail. It
2226 +can happen that even all copy-up will fail.
2227 +When both of src and dst branches support XATTR and if an error occurs
2228 +during copying XATTR, then the copy-up should fail obviously. That is a
2229 +good reason and aufs should return an error to userspace. But when only
2230 +the src branch support that XATTR, aufs should not return an error.
2231 +For example, the src branch supports ACL but the dst branch doesn't
2232 +because the dst branch may natively un-support it or temporary
2233 +un-support it due to "noacl" mount option. Of course, the dst branch fs
2234 +may NOT return an error even if the XATTR is not supported. It is
2235 +totally up to the branch fs.
2236 +
2237 +Anyway when the aufs internal copy-up gets an error from the dst branch
2238 +fs, then aufs tries removing the just copied entry and returns the error
2239 +to the userspace. The worst case of this situation will be all copy-up
2240 +will fail.
2241 +
2242 +For the copy-up operation, there two basic approaches.
2243 +- copy the specified XATTR only (by category above), and return the
2244 +  error unconditionally if it happens.
2245 +- copy all XATTR, and ignore the error on the specified category only.
2246 +
2247 +In order to support XATTR and to implement the correct behaviour, aufs
2248 +chooses the latter approach and introduces some new branch attributes,
2249 +"icexsec", "icexsys", "icextr", "icexusr", and "icexoth".
2250 +They correspond to the XATTR namespaces (see above). Additionally, to be
2251 +convenient, "icex" is also provided which means all "icex*" attributes
2252 +are set (here the word "icex" stands for "ignore copy-error on XATTR").
2253 +
2254 +The meaning of these attributes is to ignore the error from setting
2255 +XATTR on that branch.
2256 +Note that aufs tries copying all XATTR unconditionally, and ignores the
2257 +error from the dst branch according to the specified attributes.
2258 +
2259 +Some XATTR may have its default value. The default value may come from
2260 +the parent dir or the environment. If the default value is set at the
2261 +file creating-time, it will be overwritten by copy-up.
2262 +Some contradiction may happen I am afraid.
2263 +Do we need another attribute to stop copying XATTR? I am unsure. For
2264 +now, aufs implements the branch attributes to ignore the error.
2265 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt linux/Documentation/filesystems/aufs/design/07export.txt
2266 --- /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt 1970-01-01 01:00:00.000000000 +0100
2267 +++ linux/Documentation/filesystems/aufs/design/07export.txt    2017-05-06 22:16:52.774886674 +0200
2268 @@ -0,0 +1,58 @@
2269 +
2270 +# Copyright (C) 2005-2017 Junjiro R. Okajima
2271 +#
2272 +# This program is free software; you can redistribute it and/or modify
2273 +# it under the terms of the GNU General Public License as published by
2274 +# the Free Software Foundation; either version 2 of the License, or
2275 +# (at your option) any later version.
2276 +#
2277 +# This program is distributed in the hope that it will be useful,
2278 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2279 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2280 +# GNU General Public License for more details.
2281 +#
2282 +# You should have received a copy of the GNU General Public License
2283 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2284 +
2285 +Export Aufs via NFS
2286 +----------------------------------------------------------------------
2287 +Here is an approach.
2288 +- like xino/xib, add a new file 'xigen' which stores aufs inode
2289 +  generation.
2290 +- iget_locked(): initialize aufs inode generation for a new inode, and
2291 +  store it in xigen file.
2292 +- destroy_inode(): increment aufs inode generation and store it in xigen
2293 +  file. it is necessary even if it is not unlinked, because any data of
2294 +  inode may be changed by UDBA.
2295 +- encode_fh(): for a root dir, simply return FILEID_ROOT. otherwise
2296 +  build file handle by
2297 +  + branch id (4 bytes)
2298 +  + superblock generation (4 bytes)
2299 +  + inode number (4 or 8 bytes)
2300 +  + parent dir inode number (4 or 8 bytes)
2301 +  + inode generation (4 bytes))
2302 +  + return value of exportfs_encode_fh() for the parent on a branch (4
2303 +    bytes)
2304 +  + file handle for a branch (by exportfs_encode_fh())
2305 +- fh_to_dentry():
2306 +  + find the index of a branch from its id in handle, and check it is
2307 +    still exist in aufs.
2308 +  + 1st level: get the inode number from handle and search it in cache.
2309 +  + 2nd level: if not found in cache, get the parent inode number from
2310 +    the handle and search it in cache. and then open the found parent
2311 +    dir, find the matching inode number by vfs_readdir() and get its
2312 +    name, and call lookup_one_len() for the target dentry.
2313 +  + 3rd level: if the parent dir is not cached, call
2314 +    exportfs_decode_fh() for a branch and get the parent on a branch,
2315 +    build a pathname of it, convert it a pathname in aufs, call
2316 +    path_lookup(). now aufs gets a parent dir dentry, then handle it as
2317 +    the 2nd level.
2318 +  + to open the dir, aufs needs struct vfsmount. aufs keeps vfsmount
2319 +    for every branch, but not itself. to get this, (currently) aufs
2320 +    searches in current->nsproxy->mnt_ns list. it may not be a good
2321 +    idea, but I didn't get other approach.
2322 +  + test the generation of the gotten inode.
2323 +- every inode operation: they may get EBUSY due to UDBA. in this case,
2324 +  convert it into ESTALE for NFSD.
2325 +- readdir(): call lockdep_on/off() because filldir in NFSD calls
2326 +  lookup_one_len(), vfs_getattr(), encode_fh() and others.
2327 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt linux/Documentation/filesystems/aufs/design/08shwh.txt
2328 --- /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt   1970-01-01 01:00:00.000000000 +0100
2329 +++ linux/Documentation/filesystems/aufs/design/08shwh.txt      2017-05-06 22:16:52.774886674 +0200
2330 @@ -0,0 +1,52 @@
2331 +
2332 +# Copyright (C) 2005-2017 Junjiro R. Okajima
2333 +#
2334 +# This program is free software; you can redistribute it and/or modify
2335 +# it under the terms of the GNU General Public License as published by
2336 +# the Free Software Foundation; either version 2 of the License, or
2337 +# (at your option) any later version.
2338 +#
2339 +# This program is distributed in the hope that it will be useful,
2340 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2341 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2342 +# GNU General Public License for more details.
2343 +#
2344 +# You should have received a copy of the GNU General Public License
2345 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2346 +
2347 +Show Whiteout Mode (shwh)
2348 +----------------------------------------------------------------------
2349 +Generally aufs hides the name of whiteouts. But in some cases, to show
2350 +them is very useful for users. For instance, creating a new middle layer
2351 +(branch) by merging existing layers.
2352 +
2353 +(borrowing aufs1 HOW-TO from a user, Michael Towers)
2354 +When you have three branches,
2355 +- Bottom: 'system', squashfs (underlying base system), read-only
2356 +- Middle: 'mods', squashfs, read-only
2357 +- Top: 'overlay', ram (tmpfs), read-write
2358 +
2359 +The top layer is loaded at boot time and saved at shutdown, to preserve
2360 +the changes made to the system during the session.
2361 +When larger changes have been made, or smaller changes have accumulated,
2362 +the size of the saved top layer data grows. At this point, it would be
2363 +nice to be able to merge the two overlay branches ('mods' and 'overlay')
2364 +and rewrite the 'mods' squashfs, clearing the top layer and thus
2365 +restoring save and load speed.
2366 +
2367 +This merging is simplified by the use of another aufs mount, of just the
2368 +two overlay branches using the 'shwh' option.
2369 +# mount -t aufs -o ro,shwh,br:/livesys/overlay=ro+wh:/livesys/mods=rr+wh \
2370 +       aufs /livesys/merge_union
2371 +
2372 +A merged view of these two branches is then available at
2373 +/livesys/merge_union, and the new feature is that the whiteouts are
2374 +visible!
2375 +Note that in 'shwh' mode the aufs mount must be 'ro', which will disable
2376 +writing to all branches. Also the default mode for all branches is 'ro'.
2377 +It is now possible to save the combined contents of the two overlay
2378 +branches to a new squashfs, e.g.:
2379 +# mksquashfs /livesys/merge_union /path/to/newmods.squash
2380 +
2381 +This new squashfs archive can be stored on the boot device and the
2382 +initramfs will use it to replace the old one at the next boot.
2383 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt linux/Documentation/filesystems/aufs/design/10dynop.txt
2384 --- /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt  1970-01-01 01:00:00.000000000 +0100
2385 +++ linux/Documentation/filesystems/aufs/design/10dynop.txt     2017-05-06 22:16:52.774886674 +0200
2386 @@ -0,0 +1,47 @@
2387 +
2388 +# Copyright (C) 2010-2017 Junjiro R. Okajima
2389 +#
2390 +# This program is free software; you can redistribute it and/or modify
2391 +# it under the terms of the GNU General Public License as published by
2392 +# the Free Software Foundation; either version 2 of the License, or
2393 +# (at your option) any later version.
2394 +#
2395 +# This program is distributed in the hope that it will be useful,
2396 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2397 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2398 +# GNU General Public License for more details.
2399 +#
2400 +# You should have received a copy of the GNU General Public License
2401 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2402 +
2403 +Dynamically customizable FS operations
2404 +----------------------------------------------------------------------
2405 +Generally FS operations (struct inode_operations, struct
2406 +address_space_operations, struct file_operations, etc.) are defined as
2407 +"static const", but it never means that FS have only one set of
2408 +operation. Some FS have multiple sets of them. For instance, ext2 has
2409 +three sets, one for XIP, for NOBH, and for normal.
2410 +Since aufs overrides and redirects these operations, sometimes aufs has
2411 +to change its behaviour according to the branch FS type. More importantly
2412 +VFS acts differently if a function (member in the struct) is set or
2413 +not. It means aufs should have several sets of operations and select one
2414 +among them according to the branch FS definition.
2415 +
2416 +In order to solve this problem and not to affect the behaviour of VFS,
2417 +aufs defines these operations dynamically. For instance, aufs defines
2418 +dummy direct_IO function for struct address_space_operations, but it may
2419 +not be set to the address_space_operations actually. When the branch FS
2420 +doesn't have it, aufs doesn't set it to its address_space_operations
2421 +while the function definition itself is still alive. So the behaviour
2422 +itself will not change, and it will return an error when direct_IO is
2423 +not set.
2424 +
2425 +The lifetime of these dynamically generated operation object is
2426 +maintained by aufs branch object. When the branch is removed from aufs,
2427 +the reference counter of the object is decremented. When it reaches
2428 +zero, the dynamically generated operation object will be freed.
2429 +
2430 +This approach is designed to support AIO (io_submit), Direct I/O and
2431 +XIP (DAX) mainly.
2432 +Currently this approach is applied to address_space_operations for
2433 +regular files only.
2434 diff -urN /usr/share/empty/Documentation/filesystems/aufs/README linux/Documentation/filesystems/aufs/README
2435 --- /usr/share/empty/Documentation/filesystems/aufs/README      1970-01-01 01:00:00.000000000 +0100
2436 +++ linux/Documentation/filesystems/aufs/README 2017-05-06 22:16:52.774886674 +0200
2437 @@ -0,0 +1,393 @@
2438 +
2439 +Aufs4 -- advanced multi layered unification filesystem version 4.x
2440 +http://aufs.sf.net
2441 +Junjiro R. Okajima
2442 +
2443 +
2444 +0. Introduction
2445 +----------------------------------------
2446 +In the early days, aufs was entirely re-designed and re-implemented
2447 +Unionfs Version 1.x series. Adding many original ideas, approaches,
2448 +improvements and implementations, it becomes totally different from
2449 +Unionfs while keeping the basic features.
2450 +Recently, Unionfs Version 2.x series begin taking some of the same
2451 +approaches to aufs1's.
2452 +Unionfs is being developed by Professor Erez Zadok at Stony Brook
2453 +University and his team.
2454 +
2455 +Aufs4 supports linux-4.0 and later, and for linux-3.x series try aufs3.
2456 +If you want older kernel version support, try aufs2-2.6.git or
2457 +aufs2-standalone.git repository, aufs1 from CVS on SourceForge.
2458 +
2459 +Note: it becomes clear that "Aufs was rejected. Let's give it up."
2460 +      According to Christoph Hellwig, linux rejects all union-type
2461 +      filesystems but UnionMount.
2462 +<http://marc.info/?l=linux-kernel&m=123938533724484&w=2>
2463 +
2464 +PS. Al Viro seems have a plan to merge aufs as well as overlayfs and
2465 +    UnionMount, and he pointed out an issue around a directory mutex
2466 +    lock and aufs addressed it. But it is still unsure whether aufs will
2467 +    be merged (or any other union solution).
2468 +<http://marc.info/?l=linux-kernel&m=136312705029295&w=1>
2469 +
2470 +
2471 +1. Features
2472 +----------------------------------------
2473 +- unite several directories into a single virtual filesystem. The member
2474 +  directory is called as a branch.
2475 +- you can specify the permission flags to the branch, which are 'readonly',
2476 +  'readwrite' and 'whiteout-able.'
2477 +- by upper writable branch, internal copyup and whiteout, files/dirs on
2478 +  readonly branch are modifiable logically.
2479 +- dynamic branch manipulation, add, del.
2480 +- etc...
2481 +
2482 +Also there are many enhancements in aufs, such as:
2483 +- test only the highest one for the directory permission (dirperm1)
2484 +- copyup on open (coo=)
2485 +- 'move' policy for copy-up between two writable branches, after
2486 +  checking free space.
2487 +- xattr, acl
2488 +- readdir(3) in userspace.
2489 +- keep inode number by external inode number table
2490 +- keep the timestamps of file/dir in internal copyup operation
2491 +- seekable directory, supporting NFS readdir.
2492 +- whiteout is hardlinked in order to reduce the consumption of inodes
2493 +  on branch
2494 +- do not copyup, nor create a whiteout when it is unnecessary
2495 +- revert a single systemcall when an error occurs in aufs
2496 +- remount interface instead of ioctl
2497 +- maintain /etc/mtab by an external command, /sbin/mount.aufs.
2498 +- loopback mounted filesystem as a branch
2499 +- kernel thread for removing the dir who has a plenty of whiteouts
2500 +- support copyup sparse file (a file which has a 'hole' in it)
2501 +- default permission flags for branches
2502 +- selectable permission flags for ro branch, whether whiteout can
2503 +  exist or not
2504 +- export via NFS.
2505 +- support <sysfs>/fs/aufs and <debugfs>/aufs.
2506 +- support multiple writable branches, some policies to select one
2507 +  among multiple writable branches.
2508 +- a new semantics for link(2) and rename(2) to support multiple
2509 +  writable branches.
2510 +- no glibc changes are required.
2511 +- pseudo hardlink (hardlink over branches)
2512 +- allow a direct access manually to a file on branch, e.g. bypassing aufs.
2513 +  including NFS or remote filesystem branch.
2514 +- userspace wrapper for pathconf(3)/fpathconf(3) with _PC_LINK_MAX.
2515 +- and more...
2516 +
2517 +Currently these features are dropped temporary from aufs4.
2518 +See design/08plan.txt in detail.
2519 +- nested mount, i.e. aufs as readonly no-whiteout branch of another aufs
2520 +  (robr)
2521 +- statistics of aufs thread (/sys/fs/aufs/stat)
2522 +
2523 +Features or just an idea in the future (see also design/*.txt),
2524 +- reorder the branch index without del/re-add.
2525 +- permanent xino files for NFSD
2526 +- an option for refreshing the opened files after add/del branches
2527 +- light version, without branch manipulation. (unnecessary?)
2528 +- copyup in userspace
2529 +- inotify in userspace
2530 +- readv/writev
2531 +
2532 +
2533 +2. Download
2534 +----------------------------------------
2535 +There are three GIT trees for aufs4, aufs4-linux.git,
2536 +aufs4-standalone.git, and aufs-util.git. Note that there is no "4" in
2537 +"aufs-util.git."
2538 +While the aufs-util is always necessary, you need either of aufs4-linux
2539 +or aufs4-standalone.
2540 +
2541 +The aufs4-linux tree includes the whole linux mainline GIT tree,
2542 +git://git.kernel.org/.../torvalds/linux.git.
2543 +And you cannot select CONFIG_AUFS_FS=m for this version, eg. you cannot
2544 +build aufs4 as an external kernel module.
2545 +Several extra patches are not included in this tree. Only
2546 +aufs4-standalone tree contains them. They are described in the later
2547 +section "Configuration and Compilation."
2548 +
2549 +On the other hand, the aufs4-standalone tree has only aufs source files
2550 +and necessary patches, and you can select CONFIG_AUFS_FS=m.
2551 +But you need to apply all aufs patches manually.
2552 +
2553 +You will find GIT branches whose name is in form of "aufs4.x" where "x"
2554 +represents the linux kernel version, "linux-4.x". For instance,
2555 +"aufs4.0" is for linux-4.0. For latest "linux-4.x-rcN", use
2556 +"aufs4.x-rcN" branch.
2557 +
2558 +o aufs4-linux tree
2559 +$ git clone --reference /your/linux/git/tree \
2560 +       git://github.com/sfjro/aufs4-linux.git aufs4-linux.git
2561 +- if you don't have linux GIT tree, then remove "--reference ..."
2562 +$ cd aufs4-linux.git
2563 +$ git checkout origin/aufs4.0
2564 +
2565 +Or You may want to directly git-pull aufs into your linux GIT tree, and
2566 +leave the patch-work to GIT.
2567 +$ cd /your/linux/git/tree
2568 +$ git remote add aufs4 git://github.com/sfjro/aufs4-linux.git
2569 +$ git fetch aufs4
2570 +$ git checkout -b my4.0 v4.0
2571 +$ (add your local change...)
2572 +$ git pull aufs4 aufs4.0
2573 +- now you have v4.0 + your_changes + aufs4.0 in you my4.0 branch.
2574 +- you may need to solve some conflicts between your_changes and
2575 +  aufs4.0. in this case, git-rerere is recommended so that you can
2576 +  solve the similar conflicts automatically when you upgrade to 4.1 or
2577 +  later in the future.
2578 +
2579 +o aufs4-standalone tree
2580 +$ git clone git://github.com/sfjro/aufs4-standalone.git aufs4-standalone.git
2581 +$ cd aufs4-standalone.git
2582 +$ git checkout origin/aufs4.0
2583 +
2584 +o aufs-util tree
2585 +$ git clone git://git.code.sf.net/p/aufs/aufs-util aufs-util.git
2586 +- note that the public aufs-util.git is on SourceForge instead of
2587 +  GitHUB.
2588 +$ cd aufs-util.git
2589 +$ git checkout origin/aufs4.0
2590 +
2591 +Note: The 4.x-rcN branch is to be used with `rc' kernel versions ONLY.
2592 +The minor version number, 'x' in '4.x', of aufs may not always
2593 +follow the minor version number of the kernel.
2594 +Because changes in the kernel that cause the use of a new
2595 +minor version number do not always require changes to aufs-util.
2596 +
2597 +Since aufs-util has its own minor version number, you may not be
2598 +able to find a GIT branch in aufs-util for your kernel's
2599 +exact minor version number.
2600 +In this case, you should git-checkout the branch for the
2601 +nearest lower number.
2602 +
2603 +For (an unreleased) example:
2604 +If you are using "linux-4.10" and the "aufs4.10" branch
2605 +does not exist in aufs-util repository, then "aufs4.9", "aufs4.8"
2606 +or something numerically smaller is the branch for your kernel.
2607 +
2608 +Also you can view all branches by
2609 +       $ git branch -a
2610 +
2611 +
2612 +3. Configuration and Compilation
2613 +----------------------------------------
2614 +Make sure you have git-checkout'ed the correct branch.
2615 +
2616 +For aufs4-linux tree,
2617 +- enable CONFIG_AUFS_FS.
2618 +- set other aufs configurations if necessary.
2619 +
2620 +For aufs4-standalone tree,
2621 +There are several ways to build.
2622 +
2623 +1.
2624 +- apply ./aufs4-kbuild.patch to your kernel source files.
2625 +- apply ./aufs4-base.patch too.
2626 +- apply ./aufs4-mmap.patch too.
2627 +- apply ./aufs4-standalone.patch too, if you have a plan to set
2628 +  CONFIG_AUFS_FS=m. otherwise you don't need ./aufs4-standalone.patch.
2629 +- copy ./{Documentation,fs,include/uapi/linux/aufs_type.h} files to your
2630 +  kernel source tree. Never copy $PWD/include/uapi/linux/Kbuild.
2631 +- enable CONFIG_AUFS_FS, you can select either
2632 +  =m or =y.
2633 +- and build your kernel as usual.
2634 +- install the built kernel.
2635 +  Note: Since linux-3.9, every filesystem module requires an alias
2636 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2637 +  modules.aliases file if you set CONFIG_AUFS_FS=m.
2638 +- install the header files too by "make headers_install" to the
2639 +  directory where you specify. By default, it is $PWD/usr.
2640 +  "make help" shows a brief note for headers_install.
2641 +- and reboot your system.
2642 +
2643 +2.
2644 +- module only (CONFIG_AUFS_FS=m).
2645 +- apply ./aufs4-base.patch to your kernel source files.
2646 +- apply ./aufs4-mmap.patch too.
2647 +- apply ./aufs4-standalone.patch too.
2648 +- build your kernel, don't forget "make headers_install", and reboot.
2649 +- edit ./config.mk and set other aufs configurations if necessary.
2650 +  Note: You should read $PWD/fs/aufs/Kconfig carefully which describes
2651 +  every aufs configurations.
2652 +- build the module by simple "make".
2653 +  Note: Since linux-3.9, every filesystem module requires an alias
2654 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2655 +  modules.aliases file.
2656 +- you can specify ${KDIR} make variable which points to your kernel
2657 +  source tree.
2658 +- install the files
2659 +  + run "make install" to install the aufs module, or copy the built
2660 +    $PWD/aufs.ko to /lib/modules/... and run depmod -a (or reboot simply).
2661 +  + run "make install_headers" (instead of headers_install) to install
2662 +    the modified aufs header file (you can specify DESTDIR which is
2663 +    available in aufs standalone version's Makefile only), or copy
2664 +    $PWD/usr/include/linux/aufs_type.h to /usr/include/linux or wherever
2665 +    you like manually. By default, the target directory is $PWD/usr.
2666 +- no need to apply aufs4-kbuild.patch, nor copying source files to your
2667 +  kernel source tree.
2668 +
2669 +Note: The header file aufs_type.h is necessary to build aufs-util
2670 +      as well as "make headers_install" in the kernel source tree.
2671 +      headers_install is subject to be forgotten, but it is essentially
2672 +      necessary, not only for building aufs-util.
2673 +      You may not meet problems without headers_install in some older
2674 +      version though.
2675 +
2676 +And then,
2677 +- read README in aufs-util, build and install it
2678 +- note that your distribution may contain an obsoleted version of
2679 +  aufs_type.h in /usr/include/linux or something. When you build aufs
2680 +  utilities, make sure that your compiler refers the correct aufs header
2681 +  file which is built by "make headers_install."
2682 +- if you want to use readdir(3) in userspace or pathconf(3) wrapper,
2683 +  then run "make install_ulib" too. And refer to the aufs manual in
2684 +  detail.
2685 +
2686 +There several other patches in aufs4-standalone.git. They are all
2687 +optional. When you meet some problems, they will help you.
2688 +- aufs4-loopback.patch
2689 +  Supports a nested loopback mount in a branch-fs. This patch is
2690 +  unnecessary until aufs produces a message like "you may want to try
2691 +  another patch for loopback file".
2692 +- vfs-ino.patch
2693 +  Modifies a system global kernel internal function get_next_ino() in
2694 +  order to stop assigning 0 for an inode-number. Not directly related to
2695 +  aufs, but recommended generally.
2696 +- tmpfs-idr.patch
2697 +  Keeps the tmpfs inode number as the lowest value. Effective to reduce
2698 +  the size of aufs XINO files for tmpfs branch. Also it prevents the
2699 +  duplication of inode number, which is important for backup tools and
2700 +  other utilities. When you find aufs XINO files for tmpfs branch
2701 +  growing too much, try this patch.
2702 +- lockdep-debug.patch
2703 +  Because aufs is not only an ordinary filesystem (callee of VFS), but
2704 +  also a caller of VFS functions for branch filesystems, subclassing of
2705 +  the internal locks for LOCKDEP is necessary. LOCKDEP is a debugging
2706 +  feature of linux kernel. If you enable CONFIG_LOCKDEP, then you will
2707 +  need to apply this debug patch to expand several constant values.
2708 +  If don't know what LOCKDEP, then you don't have apply this patch.
2709 +
2710 +
2711 +4. Usage
2712 +----------------------------------------
2713 +At first, make sure aufs-util are installed, and please read the aufs
2714 +manual, aufs.5 in aufs-util.git tree.
2715 +$ man -l aufs.5
2716 +
2717 +And then,
2718 +$ mkdir /tmp/rw /tmp/aufs
2719 +# mount -t aufs -o br=/tmp/rw:${HOME} none /tmp/aufs
2720 +
2721 +Here is another example. The result is equivalent.
2722 +# mount -t aufs -o br=/tmp/rw=rw:${HOME}=ro none /tmp/aufs
2723 +  Or
2724 +# mount -t aufs -o br:/tmp/rw none /tmp/aufs
2725 +# mount -o remount,append:${HOME} /tmp/aufs
2726 +
2727 +Then, you can see whole tree of your home dir through /tmp/aufs. If
2728 +you modify a file under /tmp/aufs, the one on your home directory is
2729 +not affected, instead the same named file will be newly created under
2730 +/tmp/rw. And all of your modification to a file will be applied to
2731 +the one under /tmp/rw. This is called the file based Copy on Write
2732 +(COW) method.
2733 +Aufs mount options are described in aufs.5.
2734 +If you run chroot or something and make your aufs as a root directory,
2735 +then you need to customize the shutdown script. See the aufs manual in
2736 +detail.
2737 +
2738 +Additionally, there are some sample usages of aufs which are a
2739 +diskless system with network booting, and LiveCD over NFS.
2740 +See sample dir in CVS tree on SourceForge.
2741 +
2742 +
2743 +5. Contact
2744 +----------------------------------------
2745 +When you have any problems or strange behaviour in aufs, please let me
2746 +know with:
2747 +- /proc/mounts (instead of the output of mount(8))
2748 +- /sys/module/aufs/*
2749 +- /sys/fs/aufs/* (if you have them)
2750 +- /debug/aufs/* (if you have them)
2751 +- linux kernel version
2752 +  if your kernel is not plain, for example modified by distributor,
2753 +  the url where i can download its source is necessary too.
2754 +- aufs version which was printed at loading the module or booting the
2755 +  system, instead of the date you downloaded.
2756 +- configuration (define/undefine CONFIG_AUFS_xxx)
2757 +- kernel configuration or /proc/config.gz (if you have it)
2758 +- behaviour which you think to be incorrect
2759 +- actual operation, reproducible one is better
2760 +- mailto: aufs-users at lists.sourceforge.net
2761 +
2762 +Usually, I don't watch the Public Areas(Bugs, Support Requests, Patches,
2763 +and Feature Requests) on SourceForge. Please join and write to
2764 +aufs-users ML.
2765 +
2766 +
2767 +6. Acknowledgements
2768 +----------------------------------------
2769 +Thanks to everyone who have tried and are using aufs, whoever
2770 +have reported a bug or any feedback.
2771 +
2772 +Especially donators:
2773 +Tomas Matejicek(slax.org) made a donation (much more than once).
2774 +       Since Apr 2010, Tomas M (the author of Slax and Linux Live
2775 +       scripts) is making "doubling" donations.
2776 +       Unfortunately I cannot list all of the donators, but I really
2777 +       appreciate.
2778 +       It ends Aug 2010, but the ordinary donation URL is still available.
2779 +       <http://sourceforge.net/donate/index.php?group_id=167503>
2780 +Dai Itasaka made a donation (2007/8).
2781 +Chuck Smith made a donation (2008/4, 10 and 12).
2782 +Henk Schoneveld made a donation (2008/9).
2783 +Chih-Wei Huang, ASUS, CTC donated Eee PC 4G (2008/10).
2784 +Francois Dupoux made a donation (2008/11).
2785 +Bruno Cesar Ribas and Luis Carlos Erpen de Bona, C3SL serves public
2786 +       aufs2 GIT tree (2009/2).
2787 +William Grant made a donation (2009/3).
2788 +Patrick Lane made a donation (2009/4).
2789 +The Mail Archive (mail-archive.com) made donations (2009/5).
2790 +Nippy Networks (Ed Wildgoose) made a donation (2009/7).
2791 +New Dream Network, LLC (www.dreamhost.com) made a donation (2009/11).
2792 +Pavel Pronskiy made a donation (2011/2).
2793 +Iridium and Inmarsat satellite phone retailer (www.mailasail.com), Nippy
2794 +       Networks (Ed Wildgoose) made a donation for hardware (2011/3).
2795 +Max Lekomcev (DOM-TV project) made a donation (2011/7, 12, 2012/3, 6 and
2796 +11).
2797 +Sam Liddicott made a donation (2011/9).
2798 +Era Scarecrow made a donation (2013/4).
2799 +Bor Ratajc made a donation (2013/4).
2800 +Alessandro Gorreta made a donation (2013/4).
2801 +POIRETTE Marc made a donation (2013/4).
2802 +Alessandro Gorreta made a donation (2013/4).
2803 +lauri kasvandik made a donation (2013/5).
2804 +"pemasu from Finland" made a donation (2013/7).
2805 +The Parted Magic Project made a donation (2013/9 and 11).
2806 +Pavel Barta made a donation (2013/10).
2807 +Nikolay Pertsev made a donation (2014/5).
2808 +James B made a donation (2014/7 and 2015/7).
2809 +Stefano Di Biase made a donation (2014/8).
2810 +Daniel Epellei made a donation (2015/1).
2811 +OmegaPhil made a donation (2016/1).
2812 +Tomasz Szewczyk made a donation (2016/4).
2813 +James Burry made a donation (2016/12).
2814 +
2815 +Thank you very much.
2816 +Donations are always, including future donations, very important and
2817 +helpful for me to keep on developing aufs.
2818 +
2819 +
2820 +7.
2821 +----------------------------------------
2822 +If you are an experienced user, no explanation is needed. Aufs is
2823 +just a linux filesystem.
2824 +
2825 +
2826 +Enjoy!
2827 +
2828 +# Local variables: ;
2829 +# mode: text;
2830 +# End: ;
2831 diff -urN /usr/share/empty/fs/aufs/aufs.h linux/fs/aufs/aufs.h
2832 --- /usr/share/empty/fs/aufs/aufs.h     1970-01-01 01:00:00.000000000 +0100
2833 +++ linux/fs/aufs/aufs.h        2017-05-06 22:16:52.774886674 +0200
2834 @@ -0,0 +1,59 @@
2835 +/*
2836 + * Copyright (C) 2005-2017 Junjiro R. Okajima
2837 + *
2838 + * This program, aufs is free software; you can redistribute it and/or modify
2839 + * it under the terms of the GNU General Public License as published by
2840 + * the Free Software Foundation; either version 2 of the License, or
2841 + * (at your option) any later version.
2842 + *
2843 + * This program is distributed in the hope that it will be useful,
2844 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2845 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2846 + * GNU General Public License for more details.
2847 + *
2848 + * You should have received a copy of the GNU General Public License
2849 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2850 + */
2851 +
2852 +/*
2853 + * all header files
2854 + */
2855 +
2856 +#ifndef __AUFS_H__
2857 +#define __AUFS_H__
2858 +
2859 +#ifdef __KERNEL__
2860 +
2861 +#define AuStub(type, name, body, ...) \
2862 +       static inline type name(__VA_ARGS__) { body; }
2863 +
2864 +#define AuStubVoid(name, ...) \
2865 +       AuStub(void, name, , __VA_ARGS__)
2866 +#define AuStubInt0(name, ...) \
2867 +       AuStub(int, name, return 0, __VA_ARGS__)
2868 +
2869 +#include "debug.h"
2870 +
2871 +#include "branch.h"
2872 +#include "cpup.h"
2873 +#include "dcsub.h"
2874 +#include "dbgaufs.h"
2875 +#include "dentry.h"
2876 +#include "dir.h"
2877 +#include "dynop.h"
2878 +#include "file.h"
2879 +#include "fstype.h"
2880 +#include "inode.h"
2881 +#include "loop.h"
2882 +#include "module.h"
2883 +#include "opts.h"
2884 +#include "rwsem.h"
2885 +#include "spl.h"
2886 +#include "super.h"
2887 +#include "sysaufs.h"
2888 +#include "vfsub.h"
2889 +#include "whout.h"
2890 +#include "wkq.h"
2891 +
2892 +#endif /* __KERNEL__ */
2893 +#endif /* __AUFS_H__ */
2894 diff -urN /usr/share/empty/fs/aufs/branch.c linux/fs/aufs/branch.c
2895 --- /usr/share/empty/fs/aufs/branch.c   1970-01-01 01:00:00.000000000 +0100
2896 +++ linux/fs/aufs/branch.c      2017-05-06 22:16:52.774886674 +0200
2897 @@ -0,0 +1,1423 @@
2898 +/*
2899 + * Copyright (C) 2005-2017 Junjiro R. Okajima
2900 + *
2901 + * This program, aufs is free software; you can redistribute it and/or modify
2902 + * it under the terms of the GNU General Public License as published by
2903 + * the Free Software Foundation; either version 2 of the License, or
2904 + * (at your option) any later version.
2905 + *
2906 + * This program is distributed in the hope that it will be useful,
2907 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2908 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2909 + * GNU General Public License for more details.
2910 + *
2911 + * You should have received a copy of the GNU General Public License
2912 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2913 + */
2914 +
2915 +/*
2916 + * branch management
2917 + */
2918 +
2919 +#include <linux/compat.h>
2920 +#include <linux/statfs.h>
2921 +#include "aufs.h"
2922 +
2923 +/*
2924 + * free a single branch
2925 + */
2926 +static void au_br_do_free(struct au_branch *br)
2927 +{
2928 +       int i;
2929 +       struct au_wbr *wbr;
2930 +       struct au_dykey **key;
2931 +
2932 +       au_hnotify_fin_br(br);
2933 +
2934 +       if (br->br_xino.xi_file)
2935 +               fput(br->br_xino.xi_file);
2936 +       for (i = br->br_xino.xi_nondir.total - 1; i >= 0; i--)
2937 +               AuDebugOn(br->br_xino.xi_nondir.array[i]);
2938 +       au_delayed_kfree(br->br_xino.xi_nondir.array);
2939 +
2940 +       AuDebugOn(au_br_count(br));
2941 +       au_br_count_fin(br);
2942 +
2943 +       wbr = br->br_wbr;
2944 +       if (wbr) {
2945 +               for (i = 0; i < AuBrWh_Last; i++)
2946 +                       dput(wbr->wbr_wh[i]);
2947 +               AuDebugOn(atomic_read(&wbr->wbr_wh_running));
2948 +               AuRwDestroy(&wbr->wbr_wh_rwsem);
2949 +       }
2950 +
2951 +       if (br->br_fhsm) {
2952 +               au_br_fhsm_fin(br->br_fhsm);
2953 +               au_delayed_kfree(br->br_fhsm);
2954 +       }
2955 +
2956 +       key = br->br_dykey;
2957 +       for (i = 0; i < AuBrDynOp; i++, key++)
2958 +               if (*key)
2959 +                       au_dy_put(*key);
2960 +               else
2961 +                       break;
2962 +
2963 +       /* recursive lock, s_umount of branch's */
2964 +       lockdep_off();
2965 +       path_put(&br->br_path);
2966 +       lockdep_on();
2967 +       if (wbr)
2968 +               au_delayed_kfree(wbr);
2969 +       au_delayed_kfree(br);
2970 +}
2971 +
2972 +/*
2973 + * frees all branches
2974 + */
2975 +void au_br_free(struct au_sbinfo *sbinfo)
2976 +{
2977 +       aufs_bindex_t bmax;
2978 +       struct au_branch **br;
2979 +
2980 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
2981 +
2982 +       bmax = sbinfo->si_bbot + 1;
2983 +       br = sbinfo->si_branch;
2984 +       while (bmax--)
2985 +               au_br_do_free(*br++);
2986 +}
2987 +
2988 +/*
2989 + * find the index of a branch which is specified by @br_id.
2990 + */
2991 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id)
2992 +{
2993 +       aufs_bindex_t bindex, bbot;
2994 +
2995 +       bbot = au_sbbot(sb);
2996 +       for (bindex = 0; bindex <= bbot; bindex++)
2997 +               if (au_sbr_id(sb, bindex) == br_id)
2998 +                       return bindex;
2999 +       return -1;
3000 +}
3001 +
3002 +/* ---------------------------------------------------------------------- */
3003 +
3004 +/*
3005 + * add a branch
3006 + */
3007 +
3008 +static int test_overlap(struct super_block *sb, struct dentry *h_adding,
3009 +                       struct dentry *h_root)
3010 +{
3011 +       if (unlikely(h_adding == h_root
3012 +                    || au_test_loopback_overlap(sb, h_adding)))
3013 +               return 1;
3014 +       if (h_adding->d_sb != h_root->d_sb)
3015 +               return 0;
3016 +       return au_test_subdir(h_adding, h_root)
3017 +               || au_test_subdir(h_root, h_adding);
3018 +}
3019 +
3020 +/*
3021 + * returns a newly allocated branch. @new_nbranch is a number of branches
3022 + * after adding a branch.
3023 + */
3024 +static struct au_branch *au_br_alloc(struct super_block *sb, int new_nbranch,
3025 +                                    int perm)
3026 +{
3027 +       struct au_branch *add_branch;
3028 +       struct dentry *root;
3029 +       struct inode *inode;
3030 +       int err;
3031 +
3032 +       err = -ENOMEM;
3033 +       add_branch = kzalloc(sizeof(*add_branch), GFP_NOFS);
3034 +       if (unlikely(!add_branch))
3035 +               goto out;
3036 +       add_branch->br_xino.xi_nondir.total = 8; /* initial size */
3037 +       add_branch->br_xino.xi_nondir.array
3038 +               = kzalloc(sizeof(ino_t) * add_branch->br_xino.xi_nondir.total,
3039 +                         GFP_NOFS);
3040 +       if (unlikely(!add_branch->br_xino.xi_nondir.array))
3041 +               goto out_br;
3042 +
3043 +       err = au_hnotify_init_br(add_branch, perm);
3044 +       if (unlikely(err))
3045 +               goto out_xinondir;
3046 +
3047 +       if (au_br_writable(perm)) {
3048 +               /* may be freed separately at changing the branch permission */
3049 +               add_branch->br_wbr = kzalloc(sizeof(*add_branch->br_wbr),
3050 +                                            GFP_NOFS);
3051 +               if (unlikely(!add_branch->br_wbr))
3052 +                       goto out_hnotify;
3053 +       }
3054 +
3055 +       if (au_br_fhsm(perm)) {
3056 +               err = au_fhsm_br_alloc(add_branch);
3057 +               if (unlikely(err))
3058 +                       goto out_wbr;
3059 +       }
3060 +
3061 +       root = sb->s_root;
3062 +       err = au_sbr_realloc(au_sbi(sb), new_nbranch, /*may_shrink*/0);
3063 +       if (!err)
3064 +               err = au_di_realloc(au_di(root), new_nbranch, /*may_shrink*/0);
3065 +       if (!err) {
3066 +               inode = d_inode(root);
3067 +               err = au_hinode_realloc(au_ii(inode), new_nbranch, /*may_shrink*/0);
3068 +       }
3069 +       if (!err)
3070 +               return add_branch; /* success */
3071 +
3072 +out_wbr:
3073 +       if (add_branch->br_wbr)
3074 +               au_delayed_kfree(add_branch->br_wbr);
3075 +out_hnotify:
3076 +       au_hnotify_fin_br(add_branch);
3077 +out_xinondir:
3078 +       au_delayed_kfree(add_branch->br_xino.xi_nondir.array);
3079 +out_br:
3080 +       au_delayed_kfree(add_branch);
3081 +out:
3082 +       return ERR_PTR(err);
3083 +}
3084 +
3085 +/*
3086 + * test if the branch permission is legal or not.
3087 + */
3088 +static int test_br(struct inode *inode, int brperm, char *path)
3089 +{
3090 +       int err;
3091 +
3092 +       err = (au_br_writable(brperm) && IS_RDONLY(inode));
3093 +       if (!err)
3094 +               goto out;
3095 +
3096 +       err = -EINVAL;
3097 +       pr_err("write permission for readonly mount or inode, %s\n", path);
3098 +
3099 +out:
3100 +       return err;
3101 +}
3102 +
3103 +/*
3104 + * returns:
3105 + * 0: success, the caller will add it
3106 + * plus: success, it is already unified, the caller should ignore it
3107 + * minus: error
3108 + */
3109 +static int test_add(struct super_block *sb, struct au_opt_add *add, int remount)
3110 +{
3111 +       int err;
3112 +       aufs_bindex_t bbot, bindex;
3113 +       struct dentry *root, *h_dentry;
3114 +       struct inode *inode, *h_inode;
3115 +
3116 +       root = sb->s_root;
3117 +       bbot = au_sbbot(sb);
3118 +       if (unlikely(bbot >= 0
3119 +                    && au_find_dbindex(root, add->path.dentry) >= 0)) {
3120 +               err = 1;
3121 +               if (!remount) {
3122 +                       err = -EINVAL;
3123 +                       pr_err("%s duplicated\n", add->pathname);
3124 +               }
3125 +               goto out;
3126 +       }
3127 +
3128 +       err = -ENOSPC; /* -E2BIG; */
3129 +       if (unlikely(AUFS_BRANCH_MAX <= add->bindex
3130 +                    || AUFS_BRANCH_MAX - 1 <= bbot)) {
3131 +               pr_err("number of branches exceeded %s\n", add->pathname);
3132 +               goto out;
3133 +       }
3134 +
3135 +       err = -EDOM;
3136 +       if (unlikely(add->bindex < 0 || bbot + 1 < add->bindex)) {
3137 +               pr_err("bad index %d\n", add->bindex);
3138 +               goto out;
3139 +       }
3140 +
3141 +       inode = d_inode(add->path.dentry);
3142 +       err = -ENOENT;
3143 +       if (unlikely(!inode->i_nlink)) {
3144 +               pr_err("no existence %s\n", add->pathname);
3145 +               goto out;
3146 +       }
3147 +
3148 +       err = -EINVAL;
3149 +       if (unlikely(inode->i_sb == sb)) {
3150 +               pr_err("%s must be outside\n", add->pathname);
3151 +               goto out;
3152 +       }
3153 +
3154 +       if (unlikely(au_test_fs_unsuppoted(inode->i_sb))) {
3155 +               pr_err("unsupported filesystem, %s (%s)\n",
3156 +                      add->pathname, au_sbtype(inode->i_sb));
3157 +               goto out;
3158 +       }
3159 +
3160 +       if (unlikely(inode->i_sb->s_stack_depth)) {
3161 +               pr_err("already stacked, %s (%s)\n",
3162 +                      add->pathname, au_sbtype(inode->i_sb));
3163 +               goto out;
3164 +       }
3165 +
3166 +       err = test_br(d_inode(add->path.dentry), add->perm, add->pathname);
3167 +       if (unlikely(err))
3168 +               goto out;
3169 +
3170 +       if (bbot < 0)
3171 +               return 0; /* success */
3172 +
3173 +       err = -EINVAL;
3174 +       for (bindex = 0; bindex <= bbot; bindex++)
3175 +               if (unlikely(test_overlap(sb, add->path.dentry,
3176 +                                         au_h_dptr(root, bindex)))) {
3177 +                       pr_err("%s is overlapped\n", add->pathname);
3178 +                       goto out;
3179 +               }
3180 +
3181 +       err = 0;
3182 +       if (au_opt_test(au_mntflags(sb), WARN_PERM)) {
3183 +               h_dentry = au_h_dptr(root, 0);
3184 +               h_inode = d_inode(h_dentry);
3185 +               if ((h_inode->i_mode & S_IALLUGO) != (inode->i_mode & S_IALLUGO)
3186 +                   || !uid_eq(h_inode->i_uid, inode->i_uid)
3187 +                   || !gid_eq(h_inode->i_gid, inode->i_gid))
3188 +                       pr_warn("uid/gid/perm %s %u/%u/0%o, %u/%u/0%o\n",
3189 +                               add->pathname,
3190 +                               i_uid_read(inode), i_gid_read(inode),
3191 +                               (inode->i_mode & S_IALLUGO),
3192 +                               i_uid_read(h_inode), i_gid_read(h_inode),
3193 +                               (h_inode->i_mode & S_IALLUGO));
3194 +       }
3195 +
3196 +out:
3197 +       return err;
3198 +}
3199 +
3200 +/*
3201 + * initialize or clean the whiteouts for an adding branch
3202 + */
3203 +static int au_br_init_wh(struct super_block *sb, struct au_branch *br,
3204 +                        int new_perm)
3205 +{
3206 +       int err, old_perm;
3207 +       aufs_bindex_t bindex;
3208 +       struct inode *h_inode;
3209 +       struct au_wbr *wbr;
3210 +       struct au_hinode *hdir;
3211 +       struct dentry *h_dentry;
3212 +
3213 +       err = vfsub_mnt_want_write(au_br_mnt(br));
3214 +       if (unlikely(err))
3215 +               goto out;
3216 +
3217 +       wbr = br->br_wbr;
3218 +       old_perm = br->br_perm;
3219 +       br->br_perm = new_perm;
3220 +       hdir = NULL;
3221 +       h_inode = NULL;
3222 +       bindex = au_br_index(sb, br->br_id);
3223 +       if (0 <= bindex) {
3224 +               hdir = au_hi(d_inode(sb->s_root), bindex);
3225 +               au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
3226 +       } else {
3227 +               h_dentry = au_br_dentry(br);
3228 +               h_inode = d_inode(h_dentry);
3229 +               inode_lock_nested(h_inode, AuLsc_I_PARENT);
3230 +       }
3231 +       if (!wbr)
3232 +               err = au_wh_init(br, sb);
3233 +       else {
3234 +               wbr_wh_write_lock(wbr);
3235 +               err = au_wh_init(br, sb);
3236 +               wbr_wh_write_unlock(wbr);
3237 +       }
3238 +       if (hdir)
3239 +               au_hn_inode_unlock(hdir);
3240 +       else
3241 +               inode_unlock(h_inode);
3242 +       vfsub_mnt_drop_write(au_br_mnt(br));
3243 +       br->br_perm = old_perm;
3244 +
3245 +       if (!err && wbr && !au_br_writable(new_perm)) {
3246 +               au_delayed_kfree(wbr);
3247 +               br->br_wbr = NULL;
3248 +       }
3249 +
3250 +out:
3251 +       return err;
3252 +}
3253 +
3254 +static int au_wbr_init(struct au_branch *br, struct super_block *sb,
3255 +                      int perm)
3256 +{
3257 +       int err;
3258 +       struct kstatfs kst;
3259 +       struct au_wbr *wbr;
3260 +
3261 +       wbr = br->br_wbr;
3262 +       au_rw_init(&wbr->wbr_wh_rwsem);
3263 +       atomic_set(&wbr->wbr_wh_running, 0);
3264 +
3265 +       /*
3266 +        * a limit for rmdir/rename a dir
3267 +        * cf. AUFS_MAX_NAMELEN in include/uapi/linux/aufs_type.h
3268 +        */
3269 +       err = vfs_statfs(&br->br_path, &kst);
3270 +       if (unlikely(err))
3271 +               goto out;
3272 +       err = -EINVAL;
3273 +       if (kst.f_namelen >= NAME_MAX)
3274 +               err = au_br_init_wh(sb, br, perm);
3275 +       else
3276 +               pr_err("%pd(%s), unsupported namelen %ld\n",
3277 +                      au_br_dentry(br),
3278 +                      au_sbtype(au_br_dentry(br)->d_sb), kst.f_namelen);
3279 +
3280 +out:
3281 +       return err;
3282 +}
3283 +
3284 +/* initialize a new branch */
3285 +static int au_br_init(struct au_branch *br, struct super_block *sb,
3286 +                     struct au_opt_add *add)
3287 +{
3288 +       int err;
3289 +       struct inode *h_inode;
3290 +
3291 +       err = 0;
3292 +       spin_lock_init(&br->br_xino.xi_nondir.spin);
3293 +       init_waitqueue_head(&br->br_xino.xi_nondir.wqh);
3294 +       br->br_perm = add->perm;
3295 +       br->br_path = add->path; /* set first, path_get() later */
3296 +       spin_lock_init(&br->br_dykey_lock);
3297 +       au_br_count_init(br);
3298 +       atomic_set(&br->br_xino_running, 0);
3299 +       br->br_id = au_new_br_id(sb);
3300 +       AuDebugOn(br->br_id < 0);
3301 +
3302 +       if (au_br_writable(add->perm)) {
3303 +               err = au_wbr_init(br, sb, add->perm);
3304 +               if (unlikely(err))
3305 +                       goto out_err;
3306 +       }
3307 +
3308 +       if (au_opt_test(au_mntflags(sb), XINO)) {
3309 +               h_inode = d_inode(add->path.dentry);
3310 +               err = au_xino_br(sb, br, h_inode->i_ino,
3311 +                                au_sbr(sb, 0)->br_xino.xi_file, /*do_test*/1);
3312 +               if (unlikely(err)) {
3313 +                       AuDebugOn(br->br_xino.xi_file);
3314 +                       goto out_err;
3315 +               }
3316 +       }
3317 +
3318 +       sysaufs_br_init(br);
3319 +       path_get(&br->br_path);
3320 +       goto out; /* success */
3321 +
3322 +out_err:
3323 +       memset(&br->br_path, 0, sizeof(br->br_path));
3324 +out:
3325 +       return err;
3326 +}
3327 +
3328 +static void au_br_do_add_brp(struct au_sbinfo *sbinfo, aufs_bindex_t bindex,
3329 +                            struct au_branch *br, aufs_bindex_t bbot,
3330 +                            aufs_bindex_t amount)
3331 +{
3332 +       struct au_branch **brp;
3333 +
3334 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3335 +
3336 +       brp = sbinfo->si_branch + bindex;
3337 +       memmove(brp + 1, brp, sizeof(*brp) * amount);
3338 +       *brp = br;
3339 +       sbinfo->si_bbot++;
3340 +       if (unlikely(bbot < 0))
3341 +               sbinfo->si_bbot = 0;
3342 +}
3343 +
3344 +static void au_br_do_add_hdp(struct au_dinfo *dinfo, aufs_bindex_t bindex,
3345 +                            aufs_bindex_t bbot, aufs_bindex_t amount)
3346 +{
3347 +       struct au_hdentry *hdp;
3348 +
3349 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3350 +
3351 +       hdp = au_hdentry(dinfo, bindex);
3352 +       memmove(hdp + 1, hdp, sizeof(*hdp) * amount);
3353 +       au_h_dentry_init(hdp);
3354 +       dinfo->di_bbot++;
3355 +       if (unlikely(bbot < 0))
3356 +               dinfo->di_btop = 0;
3357 +}
3358 +
3359 +static void au_br_do_add_hip(struct au_iinfo *iinfo, aufs_bindex_t bindex,
3360 +                            aufs_bindex_t bbot, aufs_bindex_t amount)
3361 +{
3362 +       struct au_hinode *hip;
3363 +
3364 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3365 +
3366 +       hip = au_hinode(iinfo, bindex);
3367 +       memmove(hip + 1, hip, sizeof(*hip) * amount);
3368 +       au_hinode_init(hip);
3369 +       iinfo->ii_bbot++;
3370 +       if (unlikely(bbot < 0))
3371 +               iinfo->ii_btop = 0;
3372 +}
3373 +
3374 +static void au_br_do_add(struct super_block *sb, struct au_branch *br,
3375 +                        aufs_bindex_t bindex)
3376 +{
3377 +       struct dentry *root, *h_dentry;
3378 +       struct inode *root_inode, *h_inode;
3379 +       aufs_bindex_t bbot, amount;
3380 +
3381 +       root = sb->s_root;
3382 +       root_inode = d_inode(root);
3383 +       bbot = au_sbbot(sb);
3384 +       amount = bbot + 1 - bindex;
3385 +       h_dentry = au_br_dentry(br);
3386 +       au_sbilist_lock();
3387 +       au_br_do_add_brp(au_sbi(sb), bindex, br, bbot, amount);
3388 +       au_br_do_add_hdp(au_di(root), bindex, bbot, amount);
3389 +       au_br_do_add_hip(au_ii(root_inode), bindex, bbot, amount);
3390 +       au_set_h_dptr(root, bindex, dget(h_dentry));
3391 +       h_inode = d_inode(h_dentry);
3392 +       au_set_h_iptr(root_inode, bindex, au_igrab(h_inode), /*flags*/0);
3393 +       au_sbilist_unlock();
3394 +}
3395 +
3396 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount)
3397 +{
3398 +       int err;
3399 +       aufs_bindex_t bbot, add_bindex;
3400 +       struct dentry *root, *h_dentry;
3401 +       struct inode *root_inode;
3402 +       struct au_branch *add_branch;
3403 +
3404 +       root = sb->s_root;
3405 +       root_inode = d_inode(root);
3406 +       IMustLock(root_inode);
3407 +       IiMustWriteLock(root_inode);
3408 +       err = test_add(sb, add, remount);
3409 +       if (unlikely(err < 0))
3410 +               goto out;
3411 +       if (err) {
3412 +               err = 0;
3413 +               goto out; /* success */
3414 +       }
3415 +
3416 +       bbot = au_sbbot(sb);
3417 +       add_branch = au_br_alloc(sb, bbot + 2, add->perm);
3418 +       err = PTR_ERR(add_branch);
3419 +       if (IS_ERR(add_branch))
3420 +               goto out;
3421 +
3422 +       err = au_br_init(add_branch, sb, add);
3423 +       if (unlikely(err)) {
3424 +               au_br_do_free(add_branch);
3425 +               goto out;
3426 +       }
3427 +
3428 +       add_bindex = add->bindex;
3429 +       if (!remount)
3430 +               au_br_do_add(sb, add_branch, add_bindex);
3431 +       else {
3432 +               sysaufs_brs_del(sb, add_bindex);
3433 +               au_br_do_add(sb, add_branch, add_bindex);
3434 +               sysaufs_brs_add(sb, add_bindex);
3435 +       }
3436 +
3437 +       h_dentry = add->path.dentry;
3438 +       if (!add_bindex) {
3439 +               au_cpup_attr_all(root_inode, /*force*/1);
3440 +               sb->s_maxbytes = h_dentry->d_sb->s_maxbytes;
3441 +       } else
3442 +               au_add_nlink(root_inode, d_inode(h_dentry));
3443 +
3444 +       /*
3445 +        * this test/set prevents aufs from handling unnecesary notify events
3446 +        * of xino files, in case of re-adding a writable branch which was
3447 +        * once detached from aufs.
3448 +        */
3449 +       if (au_xino_brid(sb) < 0
3450 +           && au_br_writable(add_branch->br_perm)
3451 +           && !au_test_fs_bad_xino(h_dentry->d_sb)
3452 +           && add_branch->br_xino.xi_file
3453 +           && add_branch->br_xino.xi_file->f_path.dentry->d_parent == h_dentry)
3454 +               au_xino_brid_set(sb, add_branch->br_id);
3455 +
3456 +out:
3457 +       return err;
3458 +}
3459 +
3460 +/* ---------------------------------------------------------------------- */
3461 +
3462 +static unsigned long long au_farray_cb(struct super_block *sb, void *a,
3463 +                                      unsigned long long max __maybe_unused,
3464 +                                      void *arg)
3465 +{
3466 +       unsigned long long n;
3467 +       struct file **p, *f;
3468 +       struct au_sphlhead *files;
3469 +       struct au_finfo *finfo;
3470 +
3471 +       n = 0;
3472 +       p = a;
3473 +       files = &au_sbi(sb)->si_files;
3474 +       spin_lock(&files->spin);
3475 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
3476 +               f = finfo->fi_file;
3477 +               if (file_count(f)
3478 +                   && !special_file(file_inode(f)->i_mode)) {
3479 +                       get_file(f);
3480 +                       *p++ = f;
3481 +                       n++;
3482 +                       AuDebugOn(n > max);
3483 +               }
3484 +       }
3485 +       spin_unlock(&files->spin);
3486 +
3487 +       return n;
3488 +}
3489 +
3490 +static struct file **au_farray_alloc(struct super_block *sb,
3491 +                                    unsigned long long *max)
3492 +{
3493 +       *max = au_nfiles(sb);
3494 +       return au_array_alloc(max, au_farray_cb, sb, /*arg*/NULL);
3495 +}
3496 +
3497 +static void au_farray_free(struct file **a, unsigned long long max)
3498 +{
3499 +       unsigned long long ull;
3500 +
3501 +       for (ull = 0; ull < max; ull++)
3502 +               if (a[ull])
3503 +                       fput(a[ull]);
3504 +       kvfree(a);
3505 +}
3506 +
3507 +/* ---------------------------------------------------------------------- */
3508 +
3509 +/*
3510 + * delete a branch
3511 + */
3512 +
3513 +/* to show the line number, do not make it inlined function */
3514 +#define AuVerbose(do_info, fmt, ...) do { \
3515 +       if (do_info) \
3516 +               pr_info(fmt, ##__VA_ARGS__); \
3517 +} while (0)
3518 +
3519 +static int au_test_ibusy(struct inode *inode, aufs_bindex_t btop,
3520 +                        aufs_bindex_t bbot)
3521 +{
3522 +       return (inode && !S_ISDIR(inode->i_mode)) || btop == bbot;
3523 +}
3524 +
3525 +static int au_test_dbusy(struct dentry *dentry, aufs_bindex_t btop,
3526 +                        aufs_bindex_t bbot)
3527 +{
3528 +       return au_test_ibusy(d_inode(dentry), btop, bbot);
3529 +}
3530 +
3531 +/*
3532 + * test if the branch is deletable or not.
3533 + */
3534 +static int test_dentry_busy(struct dentry *root, aufs_bindex_t bindex,
3535 +                           unsigned int sigen, const unsigned int verbose)
3536 +{
3537 +       int err, i, j, ndentry;
3538 +       aufs_bindex_t btop, bbot;
3539 +       struct au_dcsub_pages dpages;
3540 +       struct au_dpage *dpage;
3541 +       struct dentry *d;
3542 +
3543 +       err = au_dpages_init(&dpages, GFP_NOFS);
3544 +       if (unlikely(err))
3545 +               goto out;
3546 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
3547 +       if (unlikely(err))
3548 +               goto out_dpages;
3549 +
3550 +       for (i = 0; !err && i < dpages.ndpage; i++) {
3551 +               dpage = dpages.dpages + i;
3552 +               ndentry = dpage->ndentry;
3553 +               for (j = 0; !err && j < ndentry; j++) {
3554 +                       d = dpage->dentries[j];
3555 +                       AuDebugOn(au_dcount(d) <= 0);
3556 +                       if (!au_digen_test(d, sigen)) {
3557 +                               di_read_lock_child(d, AuLock_IR);
3558 +                               if (unlikely(au_dbrange_test(d))) {
3559 +                                       di_read_unlock(d, AuLock_IR);
3560 +                                       continue;
3561 +                               }
3562 +                       } else {
3563 +                               di_write_lock_child(d);
3564 +                               if (unlikely(au_dbrange_test(d))) {
3565 +                                       di_write_unlock(d);
3566 +                                       continue;
3567 +                               }
3568 +                               err = au_reval_dpath(d, sigen);
3569 +                               if (!err)
3570 +                                       di_downgrade_lock(d, AuLock_IR);
3571 +                               else {
3572 +                                       di_write_unlock(d);
3573 +                                       break;
3574 +                               }
3575 +                       }
3576 +
3577 +                       /* AuDbgDentry(d); */
3578 +                       btop = au_dbtop(d);
3579 +                       bbot = au_dbbot(d);
3580 +                       if (btop <= bindex
3581 +                           && bindex <= bbot
3582 +                           && au_h_dptr(d, bindex)
3583 +                           && au_test_dbusy(d, btop, bbot)) {
3584 +                               err = -EBUSY;
3585 +                               AuVerbose(verbose, "busy %pd\n", d);
3586 +                               AuDbgDentry(d);
3587 +                       }
3588 +                       di_read_unlock(d, AuLock_IR);
3589 +               }
3590 +       }
3591 +
3592 +out_dpages:
3593 +       au_dpages_free(&dpages);
3594 +out:
3595 +       return err;
3596 +}
3597 +
3598 +static int test_inode_busy(struct super_block *sb, aufs_bindex_t bindex,
3599 +                          unsigned int sigen, const unsigned int verbose)
3600 +{
3601 +       int err;
3602 +       unsigned long long max, ull;
3603 +       struct inode *i, **array;
3604 +       aufs_bindex_t btop, bbot;
3605 +
3606 +       array = au_iarray_alloc(sb, &max);
3607 +       err = PTR_ERR(array);
3608 +       if (IS_ERR(array))
3609 +               goto out;
3610 +
3611 +       err = 0;
3612 +       AuDbg("b%d\n", bindex);
3613 +       for (ull = 0; !err && ull < max; ull++) {
3614 +               i = array[ull];
3615 +               if (unlikely(!i))
3616 +                       break;
3617 +               if (i->i_ino == AUFS_ROOT_INO)
3618 +                       continue;
3619 +
3620 +               /* AuDbgInode(i); */
3621 +               if (au_iigen(i, NULL) == sigen)
3622 +                       ii_read_lock_child(i);
3623 +               else {
3624 +                       ii_write_lock_child(i);
3625 +                       err = au_refresh_hinode_self(i);
3626 +                       au_iigen_dec(i);
3627 +                       if (!err)
3628 +                               ii_downgrade_lock(i);
3629 +                       else {
3630 +                               ii_write_unlock(i);
3631 +                               break;
3632 +                       }
3633 +               }
3634 +
3635 +               btop = au_ibtop(i);
3636 +               bbot = au_ibbot(i);
3637 +               if (btop <= bindex
3638 +                   && bindex <= bbot
3639 +                   && au_h_iptr(i, bindex)
3640 +                   && au_test_ibusy(i, btop, bbot)) {
3641 +                       err = -EBUSY;
3642 +                       AuVerbose(verbose, "busy i%lu\n", i->i_ino);
3643 +                       AuDbgInode(i);
3644 +               }
3645 +               ii_read_unlock(i);
3646 +       }
3647 +       au_iarray_free(array, max);
3648 +
3649 +out:
3650 +       return err;
3651 +}
3652 +
3653 +static int test_children_busy(struct dentry *root, aufs_bindex_t bindex,
3654 +                             const unsigned int verbose)
3655 +{
3656 +       int err;
3657 +       unsigned int sigen;
3658 +
3659 +       sigen = au_sigen(root->d_sb);
3660 +       DiMustNoWaiters(root);
3661 +       IiMustNoWaiters(d_inode(root));
3662 +       di_write_unlock(root);
3663 +       err = test_dentry_busy(root, bindex, sigen, verbose);
3664 +       if (!err)
3665 +               err = test_inode_busy(root->d_sb, bindex, sigen, verbose);
3666 +       di_write_lock_child(root); /* aufs_write_lock() calls ..._child() */
3667 +
3668 +       return err;
3669 +}
3670 +
3671 +static int test_dir_busy(struct file *file, aufs_bindex_t br_id,
3672 +                        struct file **to_free, int *idx)
3673 +{
3674 +       int err;
3675 +       unsigned char matched, root;
3676 +       aufs_bindex_t bindex, bbot;
3677 +       struct au_fidir *fidir;
3678 +       struct au_hfile *hfile;
3679 +
3680 +       err = 0;
3681 +       root = IS_ROOT(file->f_path.dentry);
3682 +       if (root) {
3683 +               get_file(file);
3684 +               to_free[*idx] = file;
3685 +               (*idx)++;
3686 +               goto out;
3687 +       }
3688 +
3689 +       matched = 0;
3690 +       fidir = au_fi(file)->fi_hdir;
3691 +       AuDebugOn(!fidir);
3692 +       bbot = au_fbbot_dir(file);
3693 +       for (bindex = au_fbtop(file); bindex <= bbot; bindex++) {
3694 +               hfile = fidir->fd_hfile + bindex;
3695 +               if (!hfile->hf_file)
3696 +                       continue;
3697 +
3698 +               if (hfile->hf_br->br_id == br_id) {
3699 +                       matched = 1;
3700 +                       break;
3701 +               }
3702 +       }
3703 +       if (matched)
3704 +               err = -EBUSY;
3705 +
3706 +out:
3707 +       return err;
3708 +}
3709 +
3710 +static int test_file_busy(struct super_block *sb, aufs_bindex_t br_id,
3711 +                         struct file **to_free, int opened)
3712 +{
3713 +       int err, idx;
3714 +       unsigned long long ull, max;
3715 +       aufs_bindex_t btop;
3716 +       struct file *file, **array;
3717 +       struct dentry *root;
3718 +       struct au_hfile *hfile;
3719 +
3720 +       array = au_farray_alloc(sb, &max);
3721 +       err = PTR_ERR(array);
3722 +       if (IS_ERR(array))
3723 +               goto out;
3724 +
3725 +       err = 0;
3726 +       idx = 0;
3727 +       root = sb->s_root;
3728 +       di_write_unlock(root);
3729 +       for (ull = 0; ull < max; ull++) {
3730 +               file = array[ull];
3731 +               if (unlikely(!file))
3732 +                       break;
3733 +
3734 +               /* AuDbg("%pD\n", file); */
3735 +               fi_read_lock(file);
3736 +               btop = au_fbtop(file);
3737 +               if (!d_is_dir(file->f_path.dentry)) {
3738 +                       hfile = &au_fi(file)->fi_htop;
3739 +                       if (hfile->hf_br->br_id == br_id)
3740 +                               err = -EBUSY;
3741 +               } else
3742 +                       err = test_dir_busy(file, br_id, to_free, &idx);
3743 +               fi_read_unlock(file);
3744 +               if (unlikely(err))
3745 +                       break;
3746 +       }
3747 +       di_write_lock_child(root);
3748 +       au_farray_free(array, max);
3749 +       AuDebugOn(idx > opened);
3750 +
3751 +out:
3752 +       return err;
3753 +}
3754 +
3755 +static void br_del_file(struct file **to_free, unsigned long long opened,
3756 +                         aufs_bindex_t br_id)
3757 +{
3758 +       unsigned long long ull;
3759 +       aufs_bindex_t bindex, btop, bbot, bfound;
3760 +       struct file *file;
3761 +       struct au_fidir *fidir;
3762 +       struct au_hfile *hfile;
3763 +
3764 +       for (ull = 0; ull < opened; ull++) {
3765 +               file = to_free[ull];
3766 +               if (unlikely(!file))
3767 +                       break;
3768 +
3769 +               /* AuDbg("%pD\n", file); */
3770 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
3771 +               bfound = -1;
3772 +               fidir = au_fi(file)->fi_hdir;
3773 +               AuDebugOn(!fidir);
3774 +               fi_write_lock(file);
3775 +               btop = au_fbtop(file);
3776 +               bbot = au_fbbot_dir(file);
3777 +               for (bindex = btop; bindex <= bbot; bindex++) {
3778 +                       hfile = fidir->fd_hfile + bindex;
3779 +                       if (!hfile->hf_file)
3780 +                               continue;
3781 +
3782 +                       if (hfile->hf_br->br_id == br_id) {
3783 +                               bfound = bindex;
3784 +                               break;
3785 +                       }
3786 +               }
3787 +               AuDebugOn(bfound < 0);
3788 +               au_set_h_fptr(file, bfound, NULL);
3789 +               if (bfound == btop) {
3790 +                       for (btop++; btop <= bbot; btop++)
3791 +                               if (au_hf_dir(file, btop)) {
3792 +                                       au_set_fbtop(file, btop);
3793 +                                       break;
3794 +                               }
3795 +               }
3796 +               fi_write_unlock(file);
3797 +       }
3798 +}
3799 +
3800 +static void au_br_do_del_brp(struct au_sbinfo *sbinfo,
3801 +                            const aufs_bindex_t bindex,
3802 +                            const aufs_bindex_t bbot)
3803 +{
3804 +       struct au_branch **brp, **p;
3805 +
3806 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3807 +
3808 +       brp = sbinfo->si_branch + bindex;
3809 +       if (bindex < bbot)
3810 +               memmove(brp, brp + 1, sizeof(*brp) * (bbot - bindex));
3811 +       sbinfo->si_branch[0 + bbot] = NULL;
3812 +       sbinfo->si_bbot--;
3813 +
3814 +       p = au_krealloc(sbinfo->si_branch, sizeof(*p) * bbot, AuGFP_SBILIST,
3815 +                       /*may_shrink*/1);
3816 +       if (p)
3817 +               sbinfo->si_branch = p;
3818 +       /* harmless error */
3819 +}
3820 +
3821 +static void au_br_do_del_hdp(struct au_dinfo *dinfo, const aufs_bindex_t bindex,
3822 +                            const aufs_bindex_t bbot)
3823 +{
3824 +       struct au_hdentry *hdp, *p;
3825 +
3826 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3827 +
3828 +       hdp = au_hdentry(dinfo, bindex);
3829 +       if (bindex < bbot)
3830 +               memmove(hdp, hdp + 1, sizeof(*hdp) * (bbot - bindex));
3831 +       /* au_h_dentry_init(au_hdentry(dinfo, bbot); */
3832 +       dinfo->di_bbot--;
3833 +
3834 +       p = au_krealloc(dinfo->di_hdentry, sizeof(*p) * bbot, AuGFP_SBILIST,
3835 +                       /*may_shrink*/1);
3836 +       if (p)
3837 +               dinfo->di_hdentry = p;
3838 +       /* harmless error */
3839 +}
3840 +
3841 +static void au_br_do_del_hip(struct au_iinfo *iinfo, const aufs_bindex_t bindex,
3842 +                            const aufs_bindex_t bbot)
3843 +{
3844 +       struct au_hinode *hip, *p;
3845 +
3846 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3847 +
3848 +       hip = au_hinode(iinfo, bindex);
3849 +       if (bindex < bbot)
3850 +               memmove(hip, hip + 1, sizeof(*hip) * (bbot - bindex));
3851 +       /* au_hinode_init(au_hinode(iinfo, bbot)); */
3852 +       iinfo->ii_bbot--;
3853 +
3854 +       p = au_krealloc(iinfo->ii_hinode, sizeof(*p) * bbot, AuGFP_SBILIST,
3855 +                       /*may_shrink*/1);
3856 +       if (p)
3857 +               iinfo->ii_hinode = p;
3858 +       /* harmless error */
3859 +}
3860 +
3861 +static void au_br_do_del(struct super_block *sb, aufs_bindex_t bindex,
3862 +                        struct au_branch *br)
3863 +{
3864 +       aufs_bindex_t bbot;
3865 +       struct au_sbinfo *sbinfo;
3866 +       struct dentry *root, *h_root;
3867 +       struct inode *inode, *h_inode;
3868 +       struct au_hinode *hinode;
3869 +
3870 +       SiMustWriteLock(sb);
3871 +
3872 +       root = sb->s_root;
3873 +       inode = d_inode(root);
3874 +       sbinfo = au_sbi(sb);
3875 +       bbot = sbinfo->si_bbot;
3876 +
3877 +       h_root = au_h_dptr(root, bindex);
3878 +       hinode = au_hi(inode, bindex);
3879 +       h_inode = au_igrab(hinode->hi_inode);
3880 +       au_hiput(hinode);
3881 +
3882 +       au_sbilist_lock();
3883 +       au_br_do_del_brp(sbinfo, bindex, bbot);
3884 +       au_br_do_del_hdp(au_di(root), bindex, bbot);
3885 +       au_br_do_del_hip(au_ii(inode), bindex, bbot);
3886 +       au_sbilist_unlock();
3887 +
3888 +       dput(h_root);
3889 +       iput(h_inode);
3890 +       au_br_do_free(br);
3891 +}
3892 +
3893 +static unsigned long long empty_cb(struct super_block *sb, void *array,
3894 +                                  unsigned long long max, void *arg)
3895 +{
3896 +       return max;
3897 +}
3898 +
3899 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount)
3900 +{
3901 +       int err, rerr, i;
3902 +       unsigned long long opened;
3903 +       unsigned int mnt_flags;
3904 +       aufs_bindex_t bindex, bbot, br_id;
3905 +       unsigned char do_wh, verbose;
3906 +       struct au_branch *br;
3907 +       struct au_wbr *wbr;
3908 +       struct dentry *root;
3909 +       struct file **to_free;
3910 +
3911 +       err = 0;
3912 +       opened = 0;
3913 +       to_free = NULL;
3914 +       root = sb->s_root;
3915 +       bindex = au_find_dbindex(root, del->h_path.dentry);
3916 +       if (bindex < 0) {
3917 +               if (remount)
3918 +                       goto out; /* success */
3919 +               err = -ENOENT;
3920 +               pr_err("%s no such branch\n", del->pathname);
3921 +               goto out;
3922 +       }
3923 +       AuDbg("bindex b%d\n", bindex);
3924 +
3925 +       err = -EBUSY;
3926 +       mnt_flags = au_mntflags(sb);
3927 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3928 +       bbot = au_sbbot(sb);
3929 +       if (unlikely(!bbot)) {
3930 +               AuVerbose(verbose, "no more branches left\n");
3931 +               goto out;
3932 +       }
3933 +       br = au_sbr(sb, bindex);
3934 +       AuDebugOn(!path_equal(&br->br_path, &del->h_path));
3935 +
3936 +       br_id = br->br_id;
3937 +       opened = au_br_count(br);
3938 +       if (unlikely(opened)) {
3939 +               to_free = au_array_alloc(&opened, empty_cb, sb, NULL);
3940 +               err = PTR_ERR(to_free);
3941 +               if (IS_ERR(to_free))
3942 +                       goto out;
3943 +
3944 +               err = test_file_busy(sb, br_id, to_free, opened);
3945 +               if (unlikely(err)) {
3946 +                       AuVerbose(verbose, "%llu file(s) opened\n", opened);
3947 +                       goto out;
3948 +               }
3949 +       }
3950 +
3951 +       wbr = br->br_wbr;
3952 +       do_wh = wbr && (wbr->wbr_whbase || wbr->wbr_plink || wbr->wbr_orph);
3953 +       if (do_wh) {
3954 +               /* instead of WbrWhMustWriteLock(wbr) */
3955 +               SiMustWriteLock(sb);
3956 +               for (i = 0; i < AuBrWh_Last; i++) {
3957 +                       dput(wbr->wbr_wh[i]);
3958 +                       wbr->wbr_wh[i] = NULL;
3959 +               }
3960 +       }
3961 +
3962 +       err = test_children_busy(root, bindex, verbose);
3963 +       if (unlikely(err)) {
3964 +               if (do_wh)
3965 +                       goto out_wh;
3966 +               goto out;
3967 +       }
3968 +
3969 +       err = 0;
3970 +       if (to_free) {
3971 +               /*
3972 +                * now we confirmed the branch is deletable.
3973 +                * let's free the remaining opened dirs on the branch.
3974 +                */
3975 +               di_write_unlock(root);
3976 +               br_del_file(to_free, opened, br_id);
3977 +               di_write_lock_child(root);
3978 +       }
3979 +
3980 +       if (!remount)
3981 +               au_br_do_del(sb, bindex, br);
3982 +       else {
3983 +               sysaufs_brs_del(sb, bindex);
3984 +               au_br_do_del(sb, bindex, br);
3985 +               sysaufs_brs_add(sb, bindex);
3986 +       }
3987 +
3988 +       if (!bindex) {
3989 +               au_cpup_attr_all(d_inode(root), /*force*/1);
3990 +               sb->s_maxbytes = au_sbr_sb(sb, 0)->s_maxbytes;
3991 +       } else
3992 +               au_sub_nlink(d_inode(root), d_inode(del->h_path.dentry));
3993 +       if (au_opt_test(mnt_flags, PLINK))
3994 +               au_plink_half_refresh(sb, br_id);
3995 +
3996 +       if (au_xino_brid(sb) == br_id)
3997 +               au_xino_brid_set(sb, -1);
3998 +       goto out; /* success */
3999 +
4000 +out_wh:
4001 +       /* revert */
4002 +       rerr = au_br_init_wh(sb, br, br->br_perm);
4003 +       if (rerr)
4004 +               pr_warn("failed re-creating base whiteout, %s. (%d)\n",
4005 +                       del->pathname, rerr);
4006 +out:
4007 +       if (to_free)
4008 +               au_farray_free(to_free, opened);
4009 +       return err;
4010 +}
4011 +
4012 +/* ---------------------------------------------------------------------- */
4013 +
4014 +static int au_ibusy(struct super_block *sb, struct aufs_ibusy __user *arg)
4015 +{
4016 +       int err;
4017 +       aufs_bindex_t btop, bbot;
4018 +       struct aufs_ibusy ibusy;
4019 +       struct inode *inode, *h_inode;
4020 +
4021 +       err = -EPERM;
4022 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
4023 +               goto out;
4024 +
4025 +       err = copy_from_user(&ibusy, arg, sizeof(ibusy));
4026 +       if (!err)
4027 +               err = !access_ok(VERIFY_WRITE, &arg->h_ino, sizeof(arg->h_ino));
4028 +       if (unlikely(err)) {
4029 +               err = -EFAULT;
4030 +               AuTraceErr(err);
4031 +               goto out;
4032 +       }
4033 +
4034 +       err = -EINVAL;
4035 +       si_read_lock(sb, AuLock_FLUSH);
4036 +       if (unlikely(ibusy.bindex < 0 || ibusy.bindex > au_sbbot(sb)))
4037 +               goto out_unlock;
4038 +
4039 +       err = 0;
4040 +       ibusy.h_ino = 0; /* invalid */
4041 +       inode = ilookup(sb, ibusy.ino);
4042 +       if (!inode
4043 +           || inode->i_ino == AUFS_ROOT_INO
4044 +           || au_is_bad_inode(inode))
4045 +               goto out_unlock;
4046 +
4047 +       ii_read_lock_child(inode);
4048 +       btop = au_ibtop(inode);
4049 +       bbot = au_ibbot(inode);
4050 +       if (btop <= ibusy.bindex && ibusy.bindex <= bbot) {
4051 +               h_inode = au_h_iptr(inode, ibusy.bindex);
4052 +               if (h_inode && au_test_ibusy(inode, btop, bbot))
4053 +                       ibusy.h_ino = h_inode->i_ino;
4054 +       }
4055 +       ii_read_unlock(inode);
4056 +       iput(inode);
4057 +
4058 +out_unlock:
4059 +       si_read_unlock(sb);
4060 +       if (!err) {
4061 +               err = __put_user(ibusy.h_ino, &arg->h_ino);
4062 +               if (unlikely(err)) {
4063 +                       err = -EFAULT;
4064 +                       AuTraceErr(err);
4065 +               }
4066 +       }
4067 +out:
4068 +       return err;
4069 +}
4070 +
4071 +long au_ibusy_ioctl(struct file *file, unsigned long arg)
4072 +{
4073 +       return au_ibusy(file->f_path.dentry->d_sb, (void __user *)arg);
4074 +}
4075 +
4076 +#ifdef CONFIG_COMPAT
4077 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg)
4078 +{
4079 +       return au_ibusy(file->f_path.dentry->d_sb, compat_ptr(arg));
4080 +}
4081 +#endif
4082 +
4083 +/* ---------------------------------------------------------------------- */
4084 +
4085 +/*
4086 + * change a branch permission
4087 + */
4088 +
4089 +static void au_warn_ima(void)
4090 +{
4091 +#ifdef CONFIG_IMA
4092 +       /* since it doesn't support mark_files_ro() */
4093 +       AuWarn1("RW -> RO makes IMA to produce wrong message\n");
4094 +#endif
4095 +}
4096 +
4097 +static int do_need_sigen_inc(int a, int b)
4098 +{
4099 +       return au_br_whable(a) && !au_br_whable(b);
4100 +}
4101 +
4102 +static int need_sigen_inc(int old, int new)
4103 +{
4104 +       return do_need_sigen_inc(old, new)
4105 +               || do_need_sigen_inc(new, old);
4106 +}
4107 +
4108 +static int au_br_mod_files_ro(struct super_block *sb, aufs_bindex_t bindex)
4109 +{
4110 +       int err, do_warn;
4111 +       unsigned int mnt_flags;
4112 +       unsigned long long ull, max;
4113 +       aufs_bindex_t br_id;
4114 +       unsigned char verbose, writer;
4115 +       struct file *file, *hf, **array;
4116 +       struct au_hfile *hfile;
4117 +
4118 +       mnt_flags = au_mntflags(sb);
4119 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
4120 +
4121 +       array = au_farray_alloc(sb, &max);
4122 +       err = PTR_ERR(array);
4123 +       if (IS_ERR(array))
4124 +               goto out;
4125 +
4126 +       do_warn = 0;
4127 +       br_id = au_sbr_id(sb, bindex);
4128 +       for (ull = 0; ull < max; ull++) {
4129 +               file = array[ull];
4130 +               if (unlikely(!file))
4131 +                       break;
4132 +
4133 +               /* AuDbg("%pD\n", file); */
4134 +               fi_read_lock(file);
4135 +               if (unlikely(au_test_mmapped(file))) {
4136 +                       err = -EBUSY;
4137 +                       AuVerbose(verbose, "mmapped %pD\n", file);
4138 +                       AuDbgFile(file);
4139 +                       FiMustNoWaiters(file);
4140 +                       fi_read_unlock(file);
4141 +                       goto out_array;
4142 +               }
4143 +
4144 +               hfile = &au_fi(file)->fi_htop;
4145 +               hf = hfile->hf_file;
4146 +               if (!d_is_reg(file->f_path.dentry)
4147 +                   || !(file->f_mode & FMODE_WRITE)
4148 +                   || hfile->hf_br->br_id != br_id
4149 +                   || !(hf->f_mode & FMODE_WRITE))
4150 +                       array[ull] = NULL;
4151 +               else {
4152 +                       do_warn = 1;
4153 +                       get_file(file);
4154 +               }
4155 +
4156 +               FiMustNoWaiters(file);
4157 +               fi_read_unlock(file);
4158 +               fput(file);
4159 +       }
4160 +
4161 +       err = 0;
4162 +       if (do_warn)
4163 +               au_warn_ima();
4164 +
4165 +       for (ull = 0; ull < max; ull++) {
4166 +               file = array[ull];
4167 +               if (!file)
4168 +                       continue;
4169 +
4170 +               /* todo: already flushed? */
4171 +               /*
4172 +                * fs/super.c:mark_files_ro() is gone, but aufs keeps its
4173 +                * approach which resets f_mode and calls mnt_drop_write() and
4174 +                * file_release_write() for each file, because the branch
4175 +                * attribute in aufs world is totally different from the native
4176 +                * fs rw/ro mode.
4177 +               */
4178 +               /* fi_read_lock(file); */
4179 +               hfile = &au_fi(file)->fi_htop;
4180 +               hf = hfile->hf_file;
4181 +               /* fi_read_unlock(file); */
4182 +               spin_lock(&hf->f_lock);
4183 +               writer = !!(hf->f_mode & FMODE_WRITER);
4184 +               hf->f_mode &= ~(FMODE_WRITE | FMODE_WRITER);
4185 +               spin_unlock(&hf->f_lock);
4186 +               if (writer) {
4187 +                       put_write_access(file_inode(hf));
4188 +                       __mnt_drop_write(hf->f_path.mnt);
4189 +               }
4190 +       }
4191 +
4192 +out_array:
4193 +       au_farray_free(array, max);
4194 +out:
4195 +       AuTraceErr(err);
4196 +       return err;
4197 +}
4198 +
4199 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4200 +             int *do_refresh)
4201 +{
4202 +       int err, rerr;
4203 +       aufs_bindex_t bindex;
4204 +       struct dentry *root;
4205 +       struct au_branch *br;
4206 +       struct au_br_fhsm *bf;
4207 +
4208 +       root = sb->s_root;
4209 +       bindex = au_find_dbindex(root, mod->h_root);
4210 +       if (bindex < 0) {
4211 +               if (remount)
4212 +                       return 0; /* success */
4213 +               err = -ENOENT;
4214 +               pr_err("%s no such branch\n", mod->path);
4215 +               goto out;
4216 +       }
4217 +       AuDbg("bindex b%d\n", bindex);
4218 +
4219 +       err = test_br(d_inode(mod->h_root), mod->perm, mod->path);
4220 +       if (unlikely(err))
4221 +               goto out;
4222 +
4223 +       br = au_sbr(sb, bindex);
4224 +       AuDebugOn(mod->h_root != au_br_dentry(br));
4225 +       if (br->br_perm == mod->perm)
4226 +               return 0; /* success */
4227 +
4228 +       /* pre-allocate for non-fhsm --> fhsm */
4229 +       bf = NULL;
4230 +       if (!au_br_fhsm(br->br_perm) && au_br_fhsm(mod->perm)) {
4231 +               err = au_fhsm_br_alloc(br);
4232 +               if (unlikely(err))
4233 +                       goto out;
4234 +               bf = br->br_fhsm;
4235 +               br->br_fhsm = NULL;
4236 +       }
4237 +
4238 +       if (au_br_writable(br->br_perm)) {
4239 +               /* remove whiteout base */
4240 +               err = au_br_init_wh(sb, br, mod->perm);
4241 +               if (unlikely(err))
4242 +                       goto out_bf;
4243 +
4244 +               if (!au_br_writable(mod->perm)) {
4245 +                       /* rw --> ro, file might be mmapped */
4246 +                       DiMustNoWaiters(root);
4247 +                       IiMustNoWaiters(d_inode(root));
4248 +                       di_write_unlock(root);
4249 +                       err = au_br_mod_files_ro(sb, bindex);
4250 +                       /* aufs_write_lock() calls ..._child() */
4251 +                       di_write_lock_child(root);
4252 +
4253 +                       if (unlikely(err)) {
4254 +                               rerr = -ENOMEM;
4255 +                               br->br_wbr = kzalloc(sizeof(*br->br_wbr),
4256 +                                                    GFP_NOFS);
4257 +                               if (br->br_wbr)
4258 +                                       rerr = au_wbr_init(br, sb, br->br_perm);
4259 +                               if (unlikely(rerr)) {
4260 +                                       AuIOErr("nested error %d (%d)\n",
4261 +                                               rerr, err);
4262 +                                       br->br_perm = mod->perm;
4263 +                               }
4264 +                       }
4265 +               }
4266 +       } else if (au_br_writable(mod->perm)) {
4267 +               /* ro --> rw */
4268 +               err = -ENOMEM;
4269 +               br->br_wbr = kzalloc(sizeof(*br->br_wbr), GFP_NOFS);
4270 +               if (br->br_wbr) {
4271 +                       err = au_wbr_init(br, sb, mod->perm);
4272 +                       if (unlikely(err)) {
4273 +                               au_delayed_kfree(br->br_wbr);
4274 +                               br->br_wbr = NULL;
4275 +                       }
4276 +               }
4277 +       }
4278 +       if (unlikely(err))
4279 +               goto out_bf;
4280 +
4281 +       if (au_br_fhsm(br->br_perm)) {
4282 +               if (!au_br_fhsm(mod->perm)) {
4283 +                       /* fhsm --> non-fhsm */
4284 +                       au_br_fhsm_fin(br->br_fhsm);
4285 +                       au_delayed_kfree(br->br_fhsm);
4286 +                       br->br_fhsm = NULL;
4287 +               }
4288 +       } else if (au_br_fhsm(mod->perm))
4289 +               /* non-fhsm --> fhsm */
4290 +               br->br_fhsm = bf;
4291 +
4292 +       *do_refresh |= need_sigen_inc(br->br_perm, mod->perm);
4293 +       br->br_perm = mod->perm;
4294 +       goto out; /* success */
4295 +
4296 +out_bf:
4297 +       if (bf)
4298 +               au_delayed_kfree(bf);
4299 +out:
4300 +       AuTraceErr(err);
4301 +       return err;
4302 +}
4303 +
4304 +/* ---------------------------------------------------------------------- */
4305 +
4306 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs)
4307 +{
4308 +       int err;
4309 +       struct kstatfs kstfs;
4310 +
4311 +       err = vfs_statfs(&br->br_path, &kstfs);
4312 +       if (!err) {
4313 +               stfs->f_blocks = kstfs.f_blocks;
4314 +               stfs->f_bavail = kstfs.f_bavail;
4315 +               stfs->f_files = kstfs.f_files;
4316 +               stfs->f_ffree = kstfs.f_ffree;
4317 +       }
4318 +
4319 +       return err;
4320 +}
4321 diff -urN /usr/share/empty/fs/aufs/branch.h linux/fs/aufs/branch.h
4322 --- /usr/share/empty/fs/aufs/branch.h   1970-01-01 01:00:00.000000000 +0100
4323 +++ linux/fs/aufs/branch.h      2017-05-06 22:16:52.774886674 +0200
4324 @@ -0,0 +1,321 @@
4325 +/*
4326 + * Copyright (C) 2005-2017 Junjiro R. Okajima
4327 + *
4328 + * This program, aufs is free software; you can redistribute it and/or modify
4329 + * it under the terms of the GNU General Public License as published by
4330 + * the Free Software Foundation; either version 2 of the License, or
4331 + * (at your option) any later version.
4332 + *
4333 + * This program is distributed in the hope that it will be useful,
4334 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4335 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4336 + * GNU General Public License for more details.
4337 + *
4338 + * You should have received a copy of the GNU General Public License
4339 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4340 + */
4341 +
4342 +/*
4343 + * branch filesystems and xino for them
4344 + */
4345 +
4346 +#ifndef __AUFS_BRANCH_H__
4347 +#define __AUFS_BRANCH_H__
4348 +
4349 +#ifdef __KERNEL__
4350 +
4351 +#include <linux/mount.h>
4352 +#include "dynop.h"
4353 +#include "rwsem.h"
4354 +#include "super.h"
4355 +
4356 +/* ---------------------------------------------------------------------- */
4357 +
4358 +/* a xino file */
4359 +struct au_xino_file {
4360 +       struct file             *xi_file;
4361 +       struct {
4362 +               spinlock_t              spin;
4363 +               ino_t                   *array;
4364 +               int                     total;
4365 +               /* reserved for future use */
4366 +               /* unsigned long        *bitmap; */
4367 +               wait_queue_head_t       wqh;
4368 +       } xi_nondir;
4369 +
4370 +       /* todo: make xino files an array to support huge inode number */
4371 +
4372 +#ifdef CONFIG_DEBUG_FS
4373 +       struct dentry            *xi_dbgaufs;
4374 +#endif
4375 +};
4376 +
4377 +/* File-based Hierarchical Storage Management */
4378 +struct au_br_fhsm {
4379 +#ifdef CONFIG_AUFS_FHSM
4380 +       struct mutex            bf_lock;
4381 +       unsigned long           bf_jiffy;
4382 +       struct aufs_stfs        bf_stfs;
4383 +       int                     bf_readable;
4384 +#endif
4385 +};
4386 +
4387 +/* members for writable branch only */
4388 +enum {AuBrWh_BASE, AuBrWh_PLINK, AuBrWh_ORPH, AuBrWh_Last};
4389 +struct au_wbr {
4390 +       struct au_rwsem         wbr_wh_rwsem;
4391 +       struct dentry           *wbr_wh[AuBrWh_Last];
4392 +       atomic_t                wbr_wh_running;
4393 +#define wbr_whbase             wbr_wh[AuBrWh_BASE]     /* whiteout base */
4394 +#define wbr_plink              wbr_wh[AuBrWh_PLINK]    /* pseudo-link dir */
4395 +#define wbr_orph               wbr_wh[AuBrWh_ORPH]     /* dir for orphans */
4396 +
4397 +       /* mfs mode */
4398 +       unsigned long long      wbr_bytes;
4399 +};
4400 +
4401 +/* ext2 has 3 types of operations at least, ext3 has 4 */
4402 +#define AuBrDynOp (AuDyLast * 4)
4403 +
4404 +#ifdef CONFIG_AUFS_HFSNOTIFY
4405 +/* support for asynchronous destruction */
4406 +struct au_br_hfsnotify {
4407 +       struct fsnotify_group   *hfsn_group;
4408 +};
4409 +#endif
4410 +
4411 +/* sysfs entries */
4412 +struct au_brsysfs {
4413 +       char                    name[16];
4414 +       struct attribute        attr;
4415 +};
4416 +
4417 +enum {
4418 +       AuBrSysfs_BR,
4419 +       AuBrSysfs_BRID,
4420 +       AuBrSysfs_Last
4421 +};
4422 +
4423 +/* protected by superblock rwsem */
4424 +struct au_branch {
4425 +       struct au_xino_file     br_xino;
4426 +
4427 +       aufs_bindex_t           br_id;
4428 +
4429 +       int                     br_perm;
4430 +       struct path             br_path;
4431 +       spinlock_t              br_dykey_lock;
4432 +       struct au_dykey         *br_dykey[AuBrDynOp];
4433 +       struct percpu_counter   br_count;
4434 +
4435 +       struct au_wbr           *br_wbr;
4436 +       struct au_br_fhsm       *br_fhsm;
4437 +
4438 +       /* xino truncation */
4439 +       atomic_t                br_xino_running;
4440 +
4441 +#ifdef CONFIG_AUFS_HFSNOTIFY
4442 +       struct au_br_hfsnotify  *br_hfsn;
4443 +#endif
4444 +
4445 +#ifdef CONFIG_SYSFS
4446 +       /* entries under sysfs per mount-point */
4447 +       struct au_brsysfs       br_sysfs[AuBrSysfs_Last];
4448 +#endif
4449 +};
4450 +
4451 +/* ---------------------------------------------------------------------- */
4452 +
4453 +static inline struct vfsmount *au_br_mnt(struct au_branch *br)
4454 +{
4455 +       return br->br_path.mnt;
4456 +}
4457 +
4458 +static inline struct dentry *au_br_dentry(struct au_branch *br)
4459 +{
4460 +       return br->br_path.dentry;
4461 +}
4462 +
4463 +static inline struct super_block *au_br_sb(struct au_branch *br)
4464 +{
4465 +       return au_br_mnt(br)->mnt_sb;
4466 +}
4467 +
4468 +static inline void au_br_get(struct au_branch *br)
4469 +{
4470 +       percpu_counter_inc(&br->br_count);
4471 +}
4472 +
4473 +static inline void au_br_put(struct au_branch *br)
4474 +{
4475 +       percpu_counter_dec(&br->br_count);
4476 +}
4477 +
4478 +static inline s64 au_br_count(struct au_branch *br)
4479 +{
4480 +       return percpu_counter_sum(&br->br_count);
4481 +}
4482 +
4483 +static inline void au_br_count_init(struct au_branch *br)
4484 +{
4485 +       percpu_counter_init(&br->br_count, 0, GFP_NOFS);
4486 +}
4487 +
4488 +static inline void au_br_count_fin(struct au_branch *br)
4489 +{
4490 +       percpu_counter_destroy(&br->br_count);
4491 +}
4492 +
4493 +static inline int au_br_rdonly(struct au_branch *br)
4494 +{
4495 +       return ((au_br_sb(br)->s_flags & MS_RDONLY)
4496 +               || !au_br_writable(br->br_perm))
4497 +               ? -EROFS : 0;
4498 +}
4499 +
4500 +static inline int au_br_hnotifyable(int brperm __maybe_unused)
4501 +{
4502 +#ifdef CONFIG_AUFS_HNOTIFY
4503 +       return !(brperm & AuBrPerm_RR);
4504 +#else
4505 +       return 0;
4506 +#endif
4507 +}
4508 +
4509 +static inline int au_br_test_oflag(int oflag, struct au_branch *br)
4510 +{
4511 +       int err, exec_flag;
4512 +
4513 +       err = 0;
4514 +       exec_flag = oflag & __FMODE_EXEC;
4515 +       if (unlikely(exec_flag && path_noexec(&br->br_path)))
4516 +               err = -EACCES;
4517 +
4518 +       return err;
4519 +}
4520 +
4521 +/* ---------------------------------------------------------------------- */
4522 +
4523 +/* branch.c */
4524 +struct au_sbinfo;
4525 +void au_br_free(struct au_sbinfo *sinfo);
4526 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id);
4527 +struct au_opt_add;
4528 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount);
4529 +struct au_opt_del;
4530 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount);
4531 +long au_ibusy_ioctl(struct file *file, unsigned long arg);
4532 +#ifdef CONFIG_COMPAT
4533 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg);
4534 +#endif
4535 +struct au_opt_mod;
4536 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4537 +             int *do_refresh);
4538 +struct aufs_stfs;
4539 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs);
4540 +
4541 +/* xino.c */
4542 +static const loff_t au_loff_max = LLONG_MAX;
4543 +
4544 +int au_xib_trunc(struct super_block *sb);
4545 +ssize_t xino_fread(vfs_readf_t func, struct file *file, void *buf, size_t size,
4546 +                  loff_t *pos);
4547 +ssize_t xino_fwrite(vfs_writef_t func, struct file *file, void *buf,
4548 +                   size_t size, loff_t *pos);
4549 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src);
4550 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent);
4551 +ino_t au_xino_new_ino(struct super_block *sb);
4552 +void au_xino_delete_inode(struct inode *inode, const int unlinked);
4553 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4554 +                 ino_t ino);
4555 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4556 +                ino_t *ino);
4557 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t hino,
4558 +              struct file *base_file, int do_test);
4559 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex);
4560 +
4561 +struct au_opt_xino;
4562 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount);
4563 +void au_xino_clr(struct super_block *sb);
4564 +struct file *au_xino_def(struct super_block *sb);
4565 +int au_xino_path(struct seq_file *seq, struct file *file);
4566 +
4567 +void au_xinondir_leave(struct super_block *sb, aufs_bindex_t bindex,
4568 +                      ino_t h_ino, int idx);
4569 +int au_xinondir_enter(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4570 +                     int *idx);
4571 +
4572 +/* ---------------------------------------------------------------------- */
4573 +
4574 +/* Superblock to branch */
4575 +static inline
4576 +aufs_bindex_t au_sbr_id(struct super_block *sb, aufs_bindex_t bindex)
4577 +{
4578 +       return au_sbr(sb, bindex)->br_id;
4579 +}
4580 +
4581 +static inline
4582 +struct vfsmount *au_sbr_mnt(struct super_block *sb, aufs_bindex_t bindex)
4583 +{
4584 +       return au_br_mnt(au_sbr(sb, bindex));
4585 +}
4586 +
4587 +static inline
4588 +struct super_block *au_sbr_sb(struct super_block *sb, aufs_bindex_t bindex)
4589 +{
4590 +       return au_br_sb(au_sbr(sb, bindex));
4591 +}
4592 +
4593 +static inline void au_sbr_get(struct super_block *sb, aufs_bindex_t bindex)
4594 +{
4595 +       au_br_get(au_sbr(sb, bindex));
4596 +}
4597 +
4598 +static inline void au_sbr_put(struct super_block *sb, aufs_bindex_t bindex)
4599 +{
4600 +       au_br_put(au_sbr(sb, bindex));
4601 +}
4602 +
4603 +static inline int au_sbr_perm(struct super_block *sb, aufs_bindex_t bindex)
4604 +{
4605 +       return au_sbr(sb, bindex)->br_perm;
4606 +}
4607 +
4608 +static inline int au_sbr_whable(struct super_block *sb, aufs_bindex_t bindex)
4609 +{
4610 +       return au_br_whable(au_sbr_perm(sb, bindex));
4611 +}
4612 +
4613 +/* ---------------------------------------------------------------------- */
4614 +
4615 +/*
4616 + * wbr_wh_read_lock, wbr_wh_write_lock
4617 + * wbr_wh_read_unlock, wbr_wh_write_unlock, wbr_wh_downgrade_lock
4618 + */
4619 +AuSimpleRwsemFuncs(wbr_wh, struct au_wbr *wbr, &wbr->wbr_wh_rwsem);
4620 +
4621 +#define WbrWhMustNoWaiters(wbr)        AuRwMustNoWaiters(&wbr->wbr_wh_rwsem)
4622 +#define WbrWhMustAnyLock(wbr)  AuRwMustAnyLock(&wbr->wbr_wh_rwsem)
4623 +#define WbrWhMustWriteLock(wbr)        AuRwMustWriteLock(&wbr->wbr_wh_rwsem)
4624 +
4625 +/* ---------------------------------------------------------------------- */
4626 +
4627 +#ifdef CONFIG_AUFS_FHSM
4628 +static inline void au_br_fhsm_init(struct au_br_fhsm *brfhsm)
4629 +{
4630 +       mutex_init(&brfhsm->bf_lock);
4631 +       brfhsm->bf_jiffy = 0;
4632 +       brfhsm->bf_readable = 0;
4633 +}
4634 +
4635 +static inline void au_br_fhsm_fin(struct au_br_fhsm *brfhsm)
4636 +{
4637 +       mutex_destroy(&brfhsm->bf_lock);
4638 +}
4639 +#else
4640 +AuStubVoid(au_br_fhsm_init, struct au_br_fhsm *brfhsm)
4641 +AuStubVoid(au_br_fhsm_fin, struct au_br_fhsm *brfhsm)
4642 +#endif
4643 +
4644 +#endif /* __KERNEL__ */
4645 +#endif /* __AUFS_BRANCH_H__ */
4646 diff -urN /usr/share/empty/fs/aufs/conf.mk linux/fs/aufs/conf.mk
4647 --- /usr/share/empty/fs/aufs/conf.mk    1970-01-01 01:00:00.000000000 +0100
4648 +++ linux/fs/aufs/conf.mk       2017-05-06 22:16:52.774886674 +0200
4649 @@ -0,0 +1,38 @@
4650 +
4651 +AuConfStr = CONFIG_AUFS_FS=${CONFIG_AUFS_FS}
4652 +
4653 +define AuConf
4654 +ifdef ${1}
4655 +AuConfStr += ${1}=${${1}}
4656 +endif
4657 +endef
4658 +
4659 +AuConfAll = BRANCH_MAX_127 BRANCH_MAX_511 BRANCH_MAX_1023 BRANCH_MAX_32767 \
4660 +       SBILIST \
4661 +       HNOTIFY HFSNOTIFY \
4662 +       EXPORT INO_T_64 \
4663 +       XATTR \
4664 +       FHSM \
4665 +       RDU \
4666 +       SHWH \
4667 +       BR_RAMFS \
4668 +       BR_FUSE POLL \
4669 +       BR_HFSPLUS \
4670 +       BDEV_LOOP \
4671 +       DEBUG MAGIC_SYSRQ
4672 +$(foreach i, ${AuConfAll}, \
4673 +       $(eval $(call AuConf,CONFIG_AUFS_${i})))
4674 +
4675 +AuConfName = ${obj}/conf.str
4676 +${AuConfName}.tmp: FORCE
4677 +       @echo ${AuConfStr} | tr ' ' '\n' | sed -e 's/^/"/' -e 's/$$/\\n"/' > $@
4678 +${AuConfName}: ${AuConfName}.tmp
4679 +       @diff -q $< $@ > /dev/null 2>&1 || { \
4680 +       echo '  GEN    ' $@; \
4681 +       cp -p $< $@; \
4682 +       }
4683 +FORCE:
4684 +clean-files += ${AuConfName} ${AuConfName}.tmp
4685 +${obj}/sysfs.o: ${AuConfName}
4686 +
4687 +-include ${srctree}/${src}/conf_priv.mk
4688 diff -urN /usr/share/empty/fs/aufs/cpup.c linux/fs/aufs/cpup.c
4689 --- /usr/share/empty/fs/aufs/cpup.c     1970-01-01 01:00:00.000000000 +0100
4690 +++ linux/fs/aufs/cpup.c        2017-05-06 22:16:52.774886674 +0200
4691 @@ -0,0 +1,1407 @@
4692 +/*
4693 + * Copyright (C) 2005-2017 Junjiro R. Okajima
4694 + *
4695 + * This program, aufs is free software; you can redistribute it and/or modify
4696 + * it under the terms of the GNU General Public License as published by
4697 + * the Free Software Foundation; either version 2 of the License, or
4698 + * (at your option) any later version.
4699 + *
4700 + * This program is distributed in the hope that it will be useful,
4701 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4702 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4703 + * GNU General Public License for more details.
4704 + *
4705 + * You should have received a copy of the GNU General Public License
4706 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4707 + */
4708 +
4709 +/*
4710 + * copy-up functions, see wbr_policy.c for copy-down
4711 + */
4712 +
4713 +#include <linux/fs_stack.h>
4714 +#include <linux/mm.h>
4715 +#include <linux/task_work.h>
4716 +#include "aufs.h"
4717 +
4718 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags)
4719 +{
4720 +       const unsigned int mask = S_DEAD | S_SWAPFILE | S_PRIVATE
4721 +               | S_NOATIME | S_NOCMTIME | S_AUTOMOUNT;
4722 +
4723 +       BUILD_BUG_ON(sizeof(iflags) != sizeof(dst->i_flags));
4724 +
4725 +       dst->i_flags |= iflags & ~mask;
4726 +       if (au_test_fs_notime(dst->i_sb))
4727 +               dst->i_flags |= S_NOATIME | S_NOCMTIME;
4728 +}
4729 +
4730 +void au_cpup_attr_timesizes(struct inode *inode)
4731 +{
4732 +       struct inode *h_inode;
4733 +
4734 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4735 +       fsstack_copy_attr_times(inode, h_inode);
4736 +       fsstack_copy_inode_size(inode, h_inode);
4737 +}
4738 +
4739 +void au_cpup_attr_nlink(struct inode *inode, int force)
4740 +{
4741 +       struct inode *h_inode;
4742 +       struct super_block *sb;
4743 +       aufs_bindex_t bindex, bbot;
4744 +
4745 +       sb = inode->i_sb;
4746 +       bindex = au_ibtop(inode);
4747 +       h_inode = au_h_iptr(inode, bindex);
4748 +       if (!force
4749 +           && !S_ISDIR(h_inode->i_mode)
4750 +           && au_opt_test(au_mntflags(sb), PLINK)
4751 +           && au_plink_test(inode))
4752 +               return;
4753 +
4754 +       /*
4755 +        * 0 can happen in revalidating.
4756 +        * h_inode->i_mutex may not be held here, but it is harmless since once
4757 +        * i_nlink reaches 0, it will never become positive except O_TMPFILE
4758 +        * case.
4759 +        * todo: O_TMPFILE+linkat(AT_SYMLINK_FOLLOW) bypassing aufs may cause
4760 +        *       the incorrect link count.
4761 +        */
4762 +       set_nlink(inode, h_inode->i_nlink);
4763 +
4764 +       /*
4765 +        * fewer nlink makes find(1) noisy, but larger nlink doesn't.
4766 +        * it may includes whplink directory.
4767 +        */
4768 +       if (S_ISDIR(h_inode->i_mode)) {
4769 +               bbot = au_ibbot(inode);
4770 +               for (bindex++; bindex <= bbot; bindex++) {
4771 +                       h_inode = au_h_iptr(inode, bindex);
4772 +                       if (h_inode)
4773 +                               au_add_nlink(inode, h_inode);
4774 +               }
4775 +       }
4776 +}
4777 +
4778 +void au_cpup_attr_changeable(struct inode *inode)
4779 +{
4780 +       struct inode *h_inode;
4781 +
4782 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4783 +       inode->i_mode = h_inode->i_mode;
4784 +       inode->i_uid = h_inode->i_uid;
4785 +       inode->i_gid = h_inode->i_gid;
4786 +       au_cpup_attr_timesizes(inode);
4787 +       au_cpup_attr_flags(inode, h_inode->i_flags);
4788 +}
4789 +
4790 +void au_cpup_igen(struct inode *inode, struct inode *h_inode)
4791 +{
4792 +       struct au_iinfo *iinfo = au_ii(inode);
4793 +
4794 +       IiMustWriteLock(inode);
4795 +
4796 +       iinfo->ii_higen = h_inode->i_generation;
4797 +       iinfo->ii_hsb1 = h_inode->i_sb;
4798 +}
4799 +
4800 +void au_cpup_attr_all(struct inode *inode, int force)
4801 +{
4802 +       struct inode *h_inode;
4803 +
4804 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4805 +       au_cpup_attr_changeable(inode);
4806 +       if (inode->i_nlink > 0)
4807 +               au_cpup_attr_nlink(inode, force);
4808 +       inode->i_rdev = h_inode->i_rdev;
4809 +       inode->i_blkbits = h_inode->i_blkbits;
4810 +       au_cpup_igen(inode, h_inode);
4811 +}
4812 +
4813 +/* ---------------------------------------------------------------------- */
4814 +
4815 +/* Note: dt_dentry and dt_h_dentry are not dget/dput-ed */
4816 +
4817 +/* keep the timestamps of the parent dir when cpup */
4818 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
4819 +                   struct path *h_path)
4820 +{
4821 +       struct inode *h_inode;
4822 +
4823 +       dt->dt_dentry = dentry;
4824 +       dt->dt_h_path = *h_path;
4825 +       h_inode = d_inode(h_path->dentry);
4826 +       dt->dt_atime = h_inode->i_atime;
4827 +       dt->dt_mtime = h_inode->i_mtime;
4828 +       /* smp_mb(); */
4829 +}
4830 +
4831 +void au_dtime_revert(struct au_dtime *dt)
4832 +{
4833 +       struct iattr attr;
4834 +       int err;
4835 +
4836 +       attr.ia_atime = dt->dt_atime;
4837 +       attr.ia_mtime = dt->dt_mtime;
4838 +       attr.ia_valid = ATTR_FORCE | ATTR_MTIME | ATTR_MTIME_SET
4839 +               | ATTR_ATIME | ATTR_ATIME_SET;
4840 +
4841 +       /* no delegation since this is a directory */
4842 +       err = vfsub_notify_change(&dt->dt_h_path, &attr, /*delegated*/NULL);
4843 +       if (unlikely(err))
4844 +               pr_warn("restoring timestamps failed(%d). ignored\n", err);
4845 +}
4846 +
4847 +/* ---------------------------------------------------------------------- */
4848 +
4849 +/* internal use only */
4850 +struct au_cpup_reg_attr {
4851 +       int             valid;
4852 +       struct kstat    st;
4853 +       unsigned int    iflags; /* inode->i_flags */
4854 +};
4855 +
4856 +static noinline_for_stack
4857 +int cpup_iattr(struct dentry *dst, aufs_bindex_t bindex, struct dentry *h_src,
4858 +              struct au_cpup_reg_attr *h_src_attr)
4859 +{
4860 +       int err, sbits, icex;
4861 +       unsigned int mnt_flags;
4862 +       unsigned char verbose;
4863 +       struct iattr ia;
4864 +       struct path h_path;
4865 +       struct inode *h_isrc, *h_idst;
4866 +       struct kstat *h_st;
4867 +       struct au_branch *br;
4868 +
4869 +       h_path.dentry = au_h_dptr(dst, bindex);
4870 +       h_idst = d_inode(h_path.dentry);
4871 +       br = au_sbr(dst->d_sb, bindex);
4872 +       h_path.mnt = au_br_mnt(br);
4873 +       h_isrc = d_inode(h_src);
4874 +       ia.ia_valid = ATTR_FORCE | ATTR_UID | ATTR_GID
4875 +               | ATTR_ATIME | ATTR_MTIME
4876 +               | ATTR_ATIME_SET | ATTR_MTIME_SET;
4877 +       if (h_src_attr && h_src_attr->valid) {
4878 +               h_st = &h_src_attr->st;
4879 +               ia.ia_uid = h_st->uid;
4880 +               ia.ia_gid = h_st->gid;
4881 +               ia.ia_atime = h_st->atime;
4882 +               ia.ia_mtime = h_st->mtime;
4883 +               if (h_idst->i_mode != h_st->mode
4884 +                   && !S_ISLNK(h_idst->i_mode)) {
4885 +                       ia.ia_valid |= ATTR_MODE;
4886 +                       ia.ia_mode = h_st->mode;
4887 +               }
4888 +               sbits = !!(h_st->mode & (S_ISUID | S_ISGID));
4889 +               au_cpup_attr_flags(h_idst, h_src_attr->iflags);
4890 +       } else {
4891 +               ia.ia_uid = h_isrc->i_uid;
4892 +               ia.ia_gid = h_isrc->i_gid;
4893 +               ia.ia_atime = h_isrc->i_atime;
4894 +               ia.ia_mtime = h_isrc->i_mtime;
4895 +               if (h_idst->i_mode != h_isrc->i_mode
4896 +                   && !S_ISLNK(h_idst->i_mode)) {
4897 +                       ia.ia_valid |= ATTR_MODE;
4898 +                       ia.ia_mode = h_isrc->i_mode;
4899 +               }
4900 +               sbits = !!(h_isrc->i_mode & (S_ISUID | S_ISGID));
4901 +               au_cpup_attr_flags(h_idst, h_isrc->i_flags);
4902 +       }
4903 +       /* no delegation since it is just created */
4904 +       err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4905 +
4906 +       /* is this nfs only? */
4907 +       if (!err && sbits && au_test_nfs(h_path.dentry->d_sb)) {
4908 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
4909 +               ia.ia_mode = h_isrc->i_mode;
4910 +               err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4911 +       }
4912 +
4913 +       icex = br->br_perm & AuBrAttr_ICEX;
4914 +       if (!err) {
4915 +               mnt_flags = au_mntflags(dst->d_sb);
4916 +               verbose = !!au_opt_test(mnt_flags, VERBOSE);
4917 +               err = au_cpup_xattr(h_path.dentry, h_src, icex, verbose);
4918 +       }
4919 +
4920 +       return err;
4921 +}
4922 +
4923 +/* ---------------------------------------------------------------------- */
4924 +
4925 +static int au_do_copy_file(struct file *dst, struct file *src, loff_t len,
4926 +                          char *buf, unsigned long blksize)
4927 +{
4928 +       int err;
4929 +       size_t sz, rbytes, wbytes;
4930 +       unsigned char all_zero;
4931 +       char *p, *zp;
4932 +       struct inode *h_inode;
4933 +       /* reduce stack usage */
4934 +       struct iattr *ia;
4935 +
4936 +       zp = page_address(ZERO_PAGE(0));
4937 +       if (unlikely(!zp))
4938 +               return -ENOMEM; /* possible? */
4939 +
4940 +       err = 0;
4941 +       all_zero = 0;
4942 +       while (len) {
4943 +               AuDbg("len %lld\n", len);
4944 +               sz = blksize;
4945 +               if (len < blksize)
4946 +                       sz = len;
4947 +
4948 +               rbytes = 0;
4949 +               /* todo: signal_pending? */
4950 +               while (!rbytes || err == -EAGAIN || err == -EINTR) {
4951 +                       rbytes = vfsub_read_k(src, buf, sz, &src->f_pos);
4952 +                       err = rbytes;
4953 +               }
4954 +               if (unlikely(err < 0))
4955 +                       break;
4956 +
4957 +               all_zero = 0;
4958 +               if (len >= rbytes && rbytes == blksize)
4959 +                       all_zero = !memcmp(buf, zp, rbytes);
4960 +               if (!all_zero) {
4961 +                       wbytes = rbytes;
4962 +                       p = buf;
4963 +                       while (wbytes) {
4964 +                               size_t b;
4965 +
4966 +                               b = vfsub_write_k(dst, p, wbytes, &dst->f_pos);
4967 +                               err = b;
4968 +                               /* todo: signal_pending? */
4969 +                               if (unlikely(err == -EAGAIN || err == -EINTR))
4970 +                                       continue;
4971 +                               if (unlikely(err < 0))
4972 +                                       break;
4973 +                               wbytes -= b;
4974 +                               p += b;
4975 +                       }
4976 +                       if (unlikely(err < 0))
4977 +                               break;
4978 +               } else {
4979 +                       loff_t res;
4980 +
4981 +                       AuLabel(hole);
4982 +                       res = vfsub_llseek(dst, rbytes, SEEK_CUR);
4983 +                       err = res;
4984 +                       if (unlikely(res < 0))
4985 +                               break;
4986 +               }
4987 +               len -= rbytes;
4988 +               err = 0;
4989 +       }
4990 +
4991 +       /* the last block may be a hole */
4992 +       if (!err && all_zero) {
4993 +               AuLabel(last hole);
4994 +
4995 +               err = 1;
4996 +               if (au_test_nfs(dst->f_path.dentry->d_sb)) {
4997 +                       /* nfs requires this step to make last hole */
4998 +                       /* is this only nfs? */
4999 +                       do {
5000 +                               /* todo: signal_pending? */
5001 +                               err = vfsub_write_k(dst, "\0", 1, &dst->f_pos);
5002 +                       } while (err == -EAGAIN || err == -EINTR);
5003 +                       if (err == 1)
5004 +                               dst->f_pos--;
5005 +               }
5006 +
5007 +               if (err == 1) {
5008 +                       ia = (void *)buf;
5009 +                       ia->ia_size = dst->f_pos;
5010 +                       ia->ia_valid = ATTR_SIZE | ATTR_FILE;
5011 +                       ia->ia_file = dst;
5012 +                       h_inode = file_inode(dst);
5013 +                       inode_lock_nested(h_inode, AuLsc_I_CHILD2);
5014 +                       /* no delegation since it is just created */
5015 +                       err = vfsub_notify_change(&dst->f_path, ia,
5016 +                                                 /*delegated*/NULL);
5017 +                       inode_unlock(h_inode);
5018 +               }
5019 +       }
5020 +
5021 +       return err;
5022 +}
5023 +
5024 +int au_copy_file(struct file *dst, struct file *src, loff_t len)
5025 +{
5026 +       int err;
5027 +       unsigned long blksize;
5028 +       unsigned char do_kfree;
5029 +       char *buf;
5030 +
5031 +       err = -ENOMEM;
5032 +       blksize = dst->f_path.dentry->d_sb->s_blocksize;
5033 +       if (!blksize || PAGE_SIZE < blksize)
5034 +               blksize = PAGE_SIZE;
5035 +       AuDbg("blksize %lu\n", blksize);
5036 +       do_kfree = (blksize != PAGE_SIZE && blksize >= sizeof(struct iattr *));
5037 +       if (do_kfree)
5038 +               buf = kmalloc(blksize, GFP_NOFS);
5039 +       else
5040 +               buf = (void *)__get_free_page(GFP_NOFS);
5041 +       if (unlikely(!buf))
5042 +               goto out;
5043 +
5044 +       if (len > (1 << 22))
5045 +               AuDbg("copying a large file %lld\n", (long long)len);
5046 +
5047 +       src->f_pos = 0;
5048 +       dst->f_pos = 0;
5049 +       err = au_do_copy_file(dst, src, len, buf, blksize);
5050 +       if (do_kfree)
5051 +               au_delayed_kfree(buf);
5052 +       else
5053 +               au_delayed_free_page((unsigned long)buf);
5054 +
5055 +out:
5056 +       return err;
5057 +}
5058 +
5059 +/*
5060 + * to support a sparse file which is opened with O_APPEND,
5061 + * we need to close the file.
5062 + */
5063 +static int au_cp_regular(struct au_cp_generic *cpg)
5064 +{
5065 +       int err, i;
5066 +       enum { SRC, DST };
5067 +       struct {
5068 +               aufs_bindex_t bindex;
5069 +               unsigned int flags;
5070 +               struct dentry *dentry;
5071 +               int force_wr;
5072 +               struct file *file;
5073 +               void *label;
5074 +       } *f, file[] = {
5075 +               {
5076 +                       .bindex = cpg->bsrc,
5077 +                       .flags = O_RDONLY | O_NOATIME | O_LARGEFILE,
5078 +                       .label = &&out
5079 +               },
5080 +               {
5081 +                       .bindex = cpg->bdst,
5082 +                       .flags = O_WRONLY | O_NOATIME | O_LARGEFILE,
5083 +                       .force_wr = !!au_ftest_cpup(cpg->flags, RWDST),
5084 +                       .label = &&out_src
5085 +               }
5086 +       };
5087 +       struct super_block *sb, *h_src_sb;
5088 +       struct inode *h_src_inode;
5089 +       struct task_struct *tsk = current;
5090 +
5091 +       /* bsrc branch can be ro/rw. */
5092 +       sb = cpg->dentry->d_sb;
5093 +       f = file;
5094 +       for (i = 0; i < 2; i++, f++) {
5095 +               f->dentry = au_h_dptr(cpg->dentry, f->bindex);
5096 +               f->file = au_h_open(cpg->dentry, f->bindex, f->flags,
5097 +                                   /*file*/NULL, f->force_wr);
5098 +               err = PTR_ERR(f->file);
5099 +               if (IS_ERR(f->file))
5100 +                       goto *f->label;
5101 +       }
5102 +
5103 +       /* try stopping to update while we copyup */
5104 +       h_src_inode = d_inode(file[SRC].dentry);
5105 +       h_src_sb = h_src_inode->i_sb;
5106 +       if (!au_test_nfs(h_src_sb))
5107 +               IMustLock(h_src_inode);
5108 +
5109 +       if (h_src_sb != file_inode(file[DST].file)->i_sb
5110 +           || !file[DST].file->f_op->clone_file_range)
5111 +               err = au_copy_file(file[DST].file, file[SRC].file, cpg->len);
5112 +       else {
5113 +               if (!au_test_nfs(h_src_sb)) {
5114 +                       inode_unlock(h_src_inode);
5115 +                       err = vfsub_clone_file_range(file[SRC].file,
5116 +                                                    file[DST].file, cpg->len);
5117 +                       inode_lock(h_src_inode);
5118 +               } else
5119 +                       err = vfsub_clone_file_range(file[SRC].file,
5120 +                                                    file[DST].file, cpg->len);
5121 +               if (unlikely(err == -EOPNOTSUPP && au_test_nfs(h_src_sb)))
5122 +                       /* the backend fs on NFS may not support cloning */
5123 +                       err = au_copy_file(file[DST].file, file[SRC].file,
5124 +                                          cpg->len);
5125 +               AuTraceErr(err);
5126 +       }
5127 +
5128 +       /* i wonder if we had O_NO_DELAY_FPUT flag */
5129 +       if (tsk->flags & PF_KTHREAD)
5130 +               __fput_sync(file[DST].file);
5131 +       else {
5132 +               WARN(1, "%pD\nPlease report this warning to aufs-users ML",
5133 +                    file[DST].file);
5134 +               fput(file[DST].file);
5135 +               /*
5136 +                * too bad.
5137 +                * we have to call both since we don't know which place the file
5138 +                * was added to.
5139 +                */
5140 +               task_work_run();
5141 +               flush_delayed_fput();
5142 +       }
5143 +       au_sbr_put(sb, file[DST].bindex);
5144 +
5145 +out_src:
5146 +       fput(file[SRC].file);
5147 +       au_sbr_put(sb, file[SRC].bindex);
5148 +out:
5149 +       return err;
5150 +}
5151 +
5152 +static int au_do_cpup_regular(struct au_cp_generic *cpg,
5153 +                             struct au_cpup_reg_attr *h_src_attr)
5154 +{
5155 +       int err, rerr;
5156 +       loff_t l;
5157 +       struct path h_path;
5158 +       struct inode *h_src_inode, *h_dst_inode;
5159 +
5160 +       err = 0;
5161 +       h_src_inode = au_h_iptr(d_inode(cpg->dentry), cpg->bsrc);
5162 +       l = i_size_read(h_src_inode);
5163 +       if (cpg->len == -1 || l < cpg->len)
5164 +               cpg->len = l;
5165 +       if (cpg->len) {
5166 +               /* try stopping to update while we are referencing */
5167 +               inode_lock_nested(h_src_inode, AuLsc_I_CHILD);
5168 +               au_pin_hdir_unlock(cpg->pin);
5169 +
5170 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5171 +               h_path.mnt = au_sbr_mnt(cpg->dentry->d_sb, cpg->bsrc);
5172 +               h_src_attr->iflags = h_src_inode->i_flags;
5173 +               if (!au_test_nfs(h_src_inode->i_sb))
5174 +                       err = vfsub_getattr(&h_path, &h_src_attr->st);
5175 +               else {
5176 +                       inode_unlock(h_src_inode);
5177 +                       err = vfsub_getattr(&h_path, &h_src_attr->st);
5178 +                       inode_lock_nested(h_src_inode, AuLsc_I_CHILD);
5179 +               }
5180 +               if (unlikely(err)) {
5181 +                       inode_unlock(h_src_inode);
5182 +                       goto out;
5183 +               }
5184 +               h_src_attr->valid = 1;
5185 +               if (!au_test_nfs(h_src_inode->i_sb)) {
5186 +                       err = au_cp_regular(cpg);
5187 +                       inode_unlock(h_src_inode);
5188 +               } else {
5189 +                       inode_unlock(h_src_inode);
5190 +                       err = au_cp_regular(cpg);
5191 +               }
5192 +               rerr = au_pin_hdir_relock(cpg->pin);
5193 +               if (!err && rerr)
5194 +                       err = rerr;
5195 +       }
5196 +       if (!err && (h_src_inode->i_state & I_LINKABLE)) {
5197 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bdst);
5198 +               h_dst_inode = d_inode(h_path.dentry);
5199 +               spin_lock(&h_dst_inode->i_lock);
5200 +               h_dst_inode->i_state |= I_LINKABLE;
5201 +               spin_unlock(&h_dst_inode->i_lock);
5202 +       }
5203 +
5204 +out:
5205 +       return err;
5206 +}
5207 +
5208 +static int au_do_cpup_symlink(struct path *h_path, struct dentry *h_src,
5209 +                             struct inode *h_dir)
5210 +{
5211 +       int err, symlen;
5212 +       mm_segment_t old_fs;
5213 +       union {
5214 +               char *k;
5215 +               char __user *u;
5216 +       } sym;
5217 +
5218 +       err = -ENOMEM;
5219 +       sym.k = (void *)__get_free_page(GFP_NOFS);
5220 +       if (unlikely(!sym.k))
5221 +               goto out;
5222 +
5223 +       /* unnecessary to support mmap_sem since symlink is not mmap-able */
5224 +       old_fs = get_fs();
5225 +       set_fs(KERNEL_DS);
5226 +       symlen = vfs_readlink(h_src, sym.u, PATH_MAX);
5227 +       err = symlen;
5228 +       set_fs(old_fs);
5229 +
5230 +       if (symlen > 0) {
5231 +               sym.k[symlen] = 0;
5232 +               err = vfsub_symlink(h_dir, h_path, sym.k);
5233 +       }
5234 +       au_delayed_free_page((unsigned long)sym.k);
5235 +
5236 +out:
5237 +       return err;
5238 +}
5239 +
5240 +/*
5241 + * regardless 'acl' option, reset all ACL.
5242 + * All ACL will be copied up later from the original entry on the lower branch.
5243 + */
5244 +static int au_reset_acl(struct inode *h_dir, struct path *h_path, umode_t mode)
5245 +{
5246 +       int err;
5247 +       struct dentry *h_dentry;
5248 +       struct inode *h_inode;
5249 +
5250 +       h_dentry = h_path->dentry;
5251 +       h_inode = d_inode(h_dentry);
5252 +       /* forget_all_cached_acls(h_inode)); */
5253 +       err = vfsub_removexattr(h_dentry, XATTR_NAME_POSIX_ACL_ACCESS);
5254 +       AuTraceErr(err);
5255 +       if (err == -EOPNOTSUPP)
5256 +               err = 0;
5257 +       if (!err)
5258 +               err = vfsub_acl_chmod(h_inode, mode);
5259 +
5260 +       AuTraceErr(err);
5261 +       return err;
5262 +}
5263 +
5264 +static int au_do_cpup_dir(struct au_cp_generic *cpg, struct dentry *dst_parent,
5265 +                         struct inode *h_dir, struct path *h_path)
5266 +{
5267 +       int err;
5268 +       struct inode *dir, *inode;
5269 +
5270 +       err = vfsub_removexattr(h_path->dentry, XATTR_NAME_POSIX_ACL_DEFAULT);
5271 +       AuTraceErr(err);
5272 +       if (err == -EOPNOTSUPP)
5273 +               err = 0;
5274 +       if (unlikely(err))
5275 +               goto out;
5276 +
5277 +       /*
5278 +        * strange behaviour from the users view,
5279 +        * particularry setattr case
5280 +        */
5281 +       dir = d_inode(dst_parent);
5282 +       if (au_ibtop(dir) == cpg->bdst)
5283 +               au_cpup_attr_nlink(dir, /*force*/1);
5284 +       inode = d_inode(cpg->dentry);
5285 +       au_cpup_attr_nlink(inode, /*force*/1);
5286 +
5287 +out:
5288 +       return err;
5289 +}
5290 +
5291 +static noinline_for_stack
5292 +int cpup_entry(struct au_cp_generic *cpg, struct dentry *dst_parent,
5293 +              struct au_cpup_reg_attr *h_src_attr)
5294 +{
5295 +       int err;
5296 +       umode_t mode;
5297 +       unsigned int mnt_flags;
5298 +       unsigned char isdir, isreg, force;
5299 +       const unsigned char do_dt = !!au_ftest_cpup(cpg->flags, DTIME);
5300 +       struct au_dtime dt;
5301 +       struct path h_path;
5302 +       struct dentry *h_src, *h_dst, *h_parent;
5303 +       struct inode *h_inode, *h_dir;
5304 +       struct super_block *sb;
5305 +
5306 +       /* bsrc branch can be ro/rw. */
5307 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5308 +       h_inode = d_inode(h_src);
5309 +       AuDebugOn(h_inode != au_h_iptr(d_inode(cpg->dentry), cpg->bsrc));
5310 +
5311 +       /* try stopping to be referenced while we are creating */
5312 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5313 +       if (au_ftest_cpup(cpg->flags, RENAME))
5314 +               AuDebugOn(strncmp(h_dst->d_name.name, AUFS_WH_PFX,
5315 +                                 AUFS_WH_PFX_LEN));
5316 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5317 +       h_dir = d_inode(h_parent);
5318 +       IMustLock(h_dir);
5319 +       AuDebugOn(h_parent != h_dst->d_parent);
5320 +
5321 +       sb = cpg->dentry->d_sb;
5322 +       h_path.mnt = au_sbr_mnt(sb, cpg->bdst);
5323 +       if (do_dt) {
5324 +               h_path.dentry = h_parent;
5325 +               au_dtime_store(&dt, dst_parent, &h_path);
5326 +       }
5327 +       h_path.dentry = h_dst;
5328 +
5329 +       isreg = 0;
5330 +       isdir = 0;
5331 +       mode = h_inode->i_mode;
5332 +       switch (mode & S_IFMT) {
5333 +       case S_IFREG:
5334 +               isreg = 1;
5335 +               err = vfsub_create(h_dir, &h_path, S_IRUSR | S_IWUSR,
5336 +                                  /*want_excl*/true);
5337 +               if (!err)
5338 +                       err = au_do_cpup_regular(cpg, h_src_attr);
5339 +               break;
5340 +       case S_IFDIR:
5341 +               isdir = 1;
5342 +               err = vfsub_mkdir(h_dir, &h_path, mode);
5343 +               if (!err)
5344 +                       err = au_do_cpup_dir(cpg, dst_parent, h_dir, &h_path);
5345 +               break;
5346 +       case S_IFLNK:
5347 +               err = au_do_cpup_symlink(&h_path, h_src, h_dir);
5348 +               break;
5349 +       case S_IFCHR:
5350 +       case S_IFBLK:
5351 +               AuDebugOn(!capable(CAP_MKNOD));
5352 +               /*FALLTHROUGH*/
5353 +       case S_IFIFO:
5354 +       case S_IFSOCK:
5355 +               err = vfsub_mknod(h_dir, &h_path, mode, h_inode->i_rdev);
5356 +               break;
5357 +       default:
5358 +               AuIOErr("Unknown inode type 0%o\n", mode);
5359 +               err = -EIO;
5360 +       }
5361 +       if (!err)
5362 +               err = au_reset_acl(h_dir, &h_path, mode);
5363 +
5364 +       mnt_flags = au_mntflags(sb);
5365 +       if (!au_opt_test(mnt_flags, UDBA_NONE)
5366 +           && !isdir
5367 +           && au_opt_test(mnt_flags, XINO)
5368 +           && (h_inode->i_nlink == 1
5369 +               || (h_inode->i_state & I_LINKABLE))
5370 +           /* todo: unnecessary? */
5371 +           /* && d_inode(cpg->dentry)->i_nlink == 1 */
5372 +           && cpg->bdst < cpg->bsrc
5373 +           && !au_ftest_cpup(cpg->flags, KEEPLINO))
5374 +               au_xino_write(sb, cpg->bsrc, h_inode->i_ino, /*ino*/0);
5375 +               /* ignore this error */
5376 +
5377 +       if (!err) {
5378 +               force = 0;
5379 +               if (isreg) {
5380 +                       force = !!cpg->len;
5381 +                       if (cpg->len == -1)
5382 +                               force = !!i_size_read(h_inode);
5383 +               }
5384 +               au_fhsm_wrote(sb, cpg->bdst, force);
5385 +       }
5386 +
5387 +       if (do_dt)
5388 +               au_dtime_revert(&dt);
5389 +       return err;
5390 +}
5391 +
5392 +static int au_do_ren_after_cpup(struct au_cp_generic *cpg, struct path *h_path)
5393 +{
5394 +       int err;
5395 +       struct dentry *dentry, *h_dentry, *h_parent, *parent;
5396 +       struct inode *h_dir;
5397 +       aufs_bindex_t bdst;
5398 +
5399 +       dentry = cpg->dentry;
5400 +       bdst = cpg->bdst;
5401 +       h_dentry = au_h_dptr(dentry, bdst);
5402 +       if (!au_ftest_cpup(cpg->flags, OVERWRITE)) {
5403 +               dget(h_dentry);
5404 +               au_set_h_dptr(dentry, bdst, NULL);
5405 +               err = au_lkup_neg(dentry, bdst, /*wh*/0);
5406 +               if (!err)
5407 +                       h_path->dentry = dget(au_h_dptr(dentry, bdst));
5408 +               au_set_h_dptr(dentry, bdst, h_dentry);
5409 +       } else {
5410 +               err = 0;
5411 +               parent = dget_parent(dentry);
5412 +               h_parent = au_h_dptr(parent, bdst);
5413 +               dput(parent);
5414 +               h_path->dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
5415 +               if (IS_ERR(h_path->dentry))
5416 +                       err = PTR_ERR(h_path->dentry);
5417 +       }
5418 +       if (unlikely(err))
5419 +               goto out;
5420 +
5421 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
5422 +       h_dir = d_inode(h_parent);
5423 +       IMustLock(h_dir);
5424 +       AuDbg("%pd %pd\n", h_dentry, h_path->dentry);
5425 +       /* no delegation since it is just created */
5426 +       err = vfsub_rename(h_dir, h_dentry, h_dir, h_path, /*delegated*/NULL,
5427 +                          /*flags*/0);
5428 +       dput(h_path->dentry);
5429 +
5430 +out:
5431 +       return err;
5432 +}
5433 +
5434 +/*
5435 + * copyup the @dentry from @bsrc to @bdst.
5436 + * the caller must set the both of lower dentries.
5437 + * @len is for truncating when it is -1 copyup the entire file.
5438 + * in link/rename cases, @dst_parent may be different from the real one.
5439 + * basic->bsrc can be larger than basic->bdst.
5440 + * aufs doesn't touch the credential so
5441 + * security_inode_copy_up{,_xattr}() are unnecrssary.
5442 + */
5443 +static int au_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5444 +{
5445 +       int err, rerr;
5446 +       aufs_bindex_t old_ibtop;
5447 +       unsigned char isdir, plink;
5448 +       struct dentry *h_src, *h_dst, *h_parent;
5449 +       struct inode *dst_inode, *h_dir, *inode, *delegated, *src_inode;
5450 +       struct super_block *sb;
5451 +       struct au_branch *br;
5452 +       /* to reuduce stack size */
5453 +       struct {
5454 +               struct au_dtime dt;
5455 +               struct path h_path;
5456 +               struct au_cpup_reg_attr h_src_attr;
5457 +       } *a;
5458 +
5459 +       err = -ENOMEM;
5460 +       a = kmalloc(sizeof(*a), GFP_NOFS);
5461 +       if (unlikely(!a))
5462 +               goto out;
5463 +       a->h_src_attr.valid = 0;
5464 +
5465 +       sb = cpg->dentry->d_sb;
5466 +       br = au_sbr(sb, cpg->bdst);
5467 +       a->h_path.mnt = au_br_mnt(br);
5468 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5469 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5470 +       h_dir = d_inode(h_parent);
5471 +       IMustLock(h_dir);
5472 +
5473 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5474 +       inode = d_inode(cpg->dentry);
5475 +
5476 +       if (!dst_parent)
5477 +               dst_parent = dget_parent(cpg->dentry);
5478 +       else
5479 +               dget(dst_parent);
5480 +
5481 +       plink = !!au_opt_test(au_mntflags(sb), PLINK);
5482 +       dst_inode = au_h_iptr(inode, cpg->bdst);
5483 +       if (dst_inode) {
5484 +               if (unlikely(!plink)) {
5485 +                       err = -EIO;
5486 +                       AuIOErr("hi%lu(i%lu) exists on b%d "
5487 +                               "but plink is disabled\n",
5488 +                               dst_inode->i_ino, inode->i_ino, cpg->bdst);
5489 +                       goto out_parent;
5490 +               }
5491 +
5492 +               if (dst_inode->i_nlink) {
5493 +                       const int do_dt = au_ftest_cpup(cpg->flags, DTIME);
5494 +
5495 +                       h_src = au_plink_lkup(inode, cpg->bdst);
5496 +                       err = PTR_ERR(h_src);
5497 +                       if (IS_ERR(h_src))
5498 +                               goto out_parent;
5499 +                       if (unlikely(d_is_negative(h_src))) {
5500 +                               err = -EIO;
5501 +                               AuIOErr("i%lu exists on b%d "
5502 +                                       "but not pseudo-linked\n",
5503 +                                       inode->i_ino, cpg->bdst);
5504 +                               dput(h_src);
5505 +                               goto out_parent;
5506 +                       }
5507 +
5508 +                       if (do_dt) {
5509 +                               a->h_path.dentry = h_parent;
5510 +                               au_dtime_store(&a->dt, dst_parent, &a->h_path);
5511 +                       }
5512 +
5513 +                       a->h_path.dentry = h_dst;
5514 +                       delegated = NULL;
5515 +                       err = vfsub_link(h_src, h_dir, &a->h_path, &delegated);
5516 +                       if (!err && au_ftest_cpup(cpg->flags, RENAME))
5517 +                               err = au_do_ren_after_cpup(cpg, &a->h_path);
5518 +                       if (do_dt)
5519 +                               au_dtime_revert(&a->dt);
5520 +                       if (unlikely(err == -EWOULDBLOCK)) {
5521 +                               pr_warn("cannot retry for NFSv4 delegation"
5522 +                                       " for an internal link\n");
5523 +                               iput(delegated);
5524 +                       }
5525 +                       dput(h_src);
5526 +                       goto out_parent;
5527 +               } else
5528 +                       /* todo: cpup_wh_file? */
5529 +                       /* udba work */
5530 +                       au_update_ibrange(inode, /*do_put_zero*/1);
5531 +       }
5532 +
5533 +       isdir = S_ISDIR(inode->i_mode);
5534 +       old_ibtop = au_ibtop(inode);
5535 +       err = cpup_entry(cpg, dst_parent, &a->h_src_attr);
5536 +       if (unlikely(err))
5537 +               goto out_rev;
5538 +       dst_inode = d_inode(h_dst);
5539 +       inode_lock_nested(dst_inode, AuLsc_I_CHILD2);
5540 +       /* todo: necessary? */
5541 +       /* au_pin_hdir_unlock(cpg->pin); */
5542 +
5543 +       err = cpup_iattr(cpg->dentry, cpg->bdst, h_src, &a->h_src_attr);
5544 +       if (unlikely(err)) {
5545 +               /* todo: necessary? */
5546 +               /* au_pin_hdir_relock(cpg->pin); */ /* ignore an error */
5547 +               inode_unlock(dst_inode);
5548 +               goto out_rev;
5549 +       }
5550 +
5551 +       if (cpg->bdst < old_ibtop) {
5552 +               if (S_ISREG(inode->i_mode)) {
5553 +                       err = au_dy_iaop(inode, cpg->bdst, dst_inode);
5554 +                       if (unlikely(err)) {
5555 +                               /* ignore an error */
5556 +                               /* au_pin_hdir_relock(cpg->pin); */
5557 +                               inode_unlock(dst_inode);
5558 +                               goto out_rev;
5559 +                       }
5560 +               }
5561 +               au_set_ibtop(inode, cpg->bdst);
5562 +       } else
5563 +               au_set_ibbot(inode, cpg->bdst);
5564 +       au_set_h_iptr(inode, cpg->bdst, au_igrab(dst_inode),
5565 +                     au_hi_flags(inode, isdir));
5566 +
5567 +       /* todo: necessary? */
5568 +       /* err = au_pin_hdir_relock(cpg->pin); */
5569 +       inode_unlock(dst_inode);
5570 +       if (unlikely(err))
5571 +               goto out_rev;
5572 +
5573 +       src_inode = d_inode(h_src);
5574 +       if (!isdir
5575 +           && (src_inode->i_nlink > 1
5576 +               || src_inode->i_state & I_LINKABLE)
5577 +           && plink)
5578 +               au_plink_append(inode, cpg->bdst, h_dst);
5579 +
5580 +       if (au_ftest_cpup(cpg->flags, RENAME)) {
5581 +               a->h_path.dentry = h_dst;
5582 +               err = au_do_ren_after_cpup(cpg, &a->h_path);
5583 +       }
5584 +       if (!err)
5585 +               goto out_parent; /* success */
5586 +
5587 +       /* revert */
5588 +out_rev:
5589 +       a->h_path.dentry = h_parent;
5590 +       au_dtime_store(&a->dt, dst_parent, &a->h_path);
5591 +       a->h_path.dentry = h_dst;
5592 +       rerr = 0;
5593 +       if (d_is_positive(h_dst)) {
5594 +               if (!isdir) {
5595 +                       /* no delegation since it is just created */
5596 +                       rerr = vfsub_unlink(h_dir, &a->h_path,
5597 +                                           /*delegated*/NULL, /*force*/0);
5598 +               } else
5599 +                       rerr = vfsub_rmdir(h_dir, &a->h_path);
5600 +       }
5601 +       au_dtime_revert(&a->dt);
5602 +       if (rerr) {
5603 +               AuIOErr("failed removing broken entry(%d, %d)\n", err, rerr);
5604 +               err = -EIO;
5605 +       }
5606 +out_parent:
5607 +       dput(dst_parent);
5608 +       au_delayed_kfree(a);
5609 +out:
5610 +       return err;
5611 +}
5612 +
5613 +#if 0 /* reserved */
5614 +struct au_cpup_single_args {
5615 +       int *errp;
5616 +       struct au_cp_generic *cpg;
5617 +       struct dentry *dst_parent;
5618 +};
5619 +
5620 +static void au_call_cpup_single(void *args)
5621 +{
5622 +       struct au_cpup_single_args *a = args;
5623 +
5624 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5625 +       *a->errp = au_cpup_single(a->cpg, a->dst_parent);
5626 +       au_pin_hdir_release(a->cpg->pin);
5627 +}
5628 +#endif
5629 +
5630 +/*
5631 + * prevent SIGXFSZ in copy-up.
5632 + * testing CAP_MKNOD is for generic fs,
5633 + * but CAP_FSETID is for xfs only, currently.
5634 + */
5635 +static int au_cpup_sio_test(struct au_pin *pin, umode_t mode)
5636 +{
5637 +       int do_sio;
5638 +       struct super_block *sb;
5639 +       struct inode *h_dir;
5640 +
5641 +       do_sio = 0;
5642 +       sb = au_pinned_parent(pin)->d_sb;
5643 +       if (!au_wkq_test()
5644 +           && (!au_sbi(sb)->si_plink_maint_pid
5645 +               || au_plink_maint(sb, AuLock_NOPLM))) {
5646 +               switch (mode & S_IFMT) {
5647 +               case S_IFREG:
5648 +                       /* no condition about RLIMIT_FSIZE and the file size */
5649 +                       do_sio = 1;
5650 +                       break;
5651 +               case S_IFCHR:
5652 +               case S_IFBLK:
5653 +                       do_sio = !capable(CAP_MKNOD);
5654 +                       break;
5655 +               }
5656 +               if (!do_sio)
5657 +                       do_sio = ((mode & (S_ISUID | S_ISGID))
5658 +                                 && !capable(CAP_FSETID));
5659 +               /* this workaround may be removed in the future */
5660 +               if (!do_sio) {
5661 +                       h_dir = au_pinned_h_dir(pin);
5662 +                       do_sio = h_dir->i_mode & S_ISVTX;
5663 +               }
5664 +       }
5665 +
5666 +       return do_sio;
5667 +}
5668 +
5669 +#if 0 /* reserved */
5670 +int au_sio_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5671 +{
5672 +       int err, wkq_err;
5673 +       struct dentry *h_dentry;
5674 +
5675 +       h_dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5676 +       if (!au_cpup_sio_test(pin, d_inode(h_dentry)->i_mode))
5677 +               err = au_cpup_single(cpg, dst_parent);
5678 +       else {
5679 +               struct au_cpup_single_args args = {
5680 +                       .errp           = &err,
5681 +                       .cpg            = cpg,
5682 +                       .dst_parent     = dst_parent
5683 +               };
5684 +               wkq_err = au_wkq_wait(au_call_cpup_single, &args);
5685 +               if (unlikely(wkq_err))
5686 +                       err = wkq_err;
5687 +       }
5688 +
5689 +       return err;
5690 +}
5691 +#endif
5692 +
5693 +/*
5694 + * copyup the @dentry from the first active lower branch to @bdst,
5695 + * using au_cpup_single().
5696 + */
5697 +static int au_cpup_simple(struct au_cp_generic *cpg)
5698 +{
5699 +       int err;
5700 +       unsigned int flags_orig;
5701 +       struct dentry *dentry;
5702 +
5703 +       AuDebugOn(cpg->bsrc < 0);
5704 +
5705 +       dentry = cpg->dentry;
5706 +       DiMustWriteLock(dentry);
5707 +
5708 +       err = au_lkup_neg(dentry, cpg->bdst, /*wh*/1);
5709 +       if (!err) {
5710 +               flags_orig = cpg->flags;
5711 +               au_fset_cpup(cpg->flags, RENAME);
5712 +               err = au_cpup_single(cpg, NULL);
5713 +               cpg->flags = flags_orig;
5714 +               if (!err)
5715 +                       return 0; /* success */
5716 +
5717 +               /* revert */
5718 +               au_set_h_dptr(dentry, cpg->bdst, NULL);
5719 +               au_set_dbtop(dentry, cpg->bsrc);
5720 +       }
5721 +
5722 +       return err;
5723 +}
5724 +
5725 +struct au_cpup_simple_args {
5726 +       int *errp;
5727 +       struct au_cp_generic *cpg;
5728 +};
5729 +
5730 +static void au_call_cpup_simple(void *args)
5731 +{
5732 +       struct au_cpup_simple_args *a = args;
5733 +
5734 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5735 +       *a->errp = au_cpup_simple(a->cpg);
5736 +       au_pin_hdir_release(a->cpg->pin);
5737 +}
5738 +
5739 +static int au_do_sio_cpup_simple(struct au_cp_generic *cpg)
5740 +{
5741 +       int err, wkq_err;
5742 +       struct dentry *dentry, *parent;
5743 +       struct file *h_file;
5744 +       struct inode *h_dir;
5745 +
5746 +       dentry = cpg->dentry;
5747 +       h_file = NULL;
5748 +       if (au_ftest_cpup(cpg->flags, HOPEN)) {
5749 +               AuDebugOn(cpg->bsrc < 0);
5750 +               h_file = au_h_open_pre(dentry, cpg->bsrc, /*force_wr*/0);
5751 +               err = PTR_ERR(h_file);
5752 +               if (IS_ERR(h_file))
5753 +                       goto out;
5754 +       }
5755 +
5756 +       parent = dget_parent(dentry);
5757 +       h_dir = au_h_iptr(d_inode(parent), cpg->bdst);
5758 +       if (!au_test_h_perm_sio(h_dir, MAY_EXEC | MAY_WRITE)
5759 +           && !au_cpup_sio_test(cpg->pin, d_inode(dentry)->i_mode))
5760 +               err = au_cpup_simple(cpg);
5761 +       else {
5762 +               struct au_cpup_simple_args args = {
5763 +                       .errp           = &err,
5764 +                       .cpg            = cpg
5765 +               };
5766 +               wkq_err = au_wkq_wait(au_call_cpup_simple, &args);
5767 +               if (unlikely(wkq_err))
5768 +                       err = wkq_err;
5769 +       }
5770 +
5771 +       dput(parent);
5772 +       if (h_file)
5773 +               au_h_open_post(dentry, cpg->bsrc, h_file);
5774 +
5775 +out:
5776 +       return err;
5777 +}
5778 +
5779 +int au_sio_cpup_simple(struct au_cp_generic *cpg)
5780 +{
5781 +       aufs_bindex_t bsrc, bbot;
5782 +       struct dentry *dentry, *h_dentry;
5783 +
5784 +       if (cpg->bsrc < 0) {
5785 +               dentry = cpg->dentry;
5786 +               bbot = au_dbbot(dentry);
5787 +               for (bsrc = cpg->bdst + 1; bsrc <= bbot; bsrc++) {
5788 +                       h_dentry = au_h_dptr(dentry, bsrc);
5789 +                       if (h_dentry) {
5790 +                               AuDebugOn(d_is_negative(h_dentry));
5791 +                               break;
5792 +                       }
5793 +               }
5794 +               AuDebugOn(bsrc > bbot);
5795 +               cpg->bsrc = bsrc;
5796 +       }
5797 +       AuDebugOn(cpg->bsrc <= cpg->bdst);
5798 +       return au_do_sio_cpup_simple(cpg);
5799 +}
5800 +
5801 +int au_sio_cpdown_simple(struct au_cp_generic *cpg)
5802 +{
5803 +       AuDebugOn(cpg->bdst <= cpg->bsrc);
5804 +       return au_do_sio_cpup_simple(cpg);
5805 +}
5806 +
5807 +/* ---------------------------------------------------------------------- */
5808 +
5809 +/*
5810 + * copyup the deleted file for writing.
5811 + */
5812 +static int au_do_cpup_wh(struct au_cp_generic *cpg, struct dentry *wh_dentry,
5813 +                        struct file *file)
5814 +{
5815 +       int err;
5816 +       unsigned int flags_orig;
5817 +       aufs_bindex_t bsrc_orig;
5818 +       struct au_dinfo *dinfo;
5819 +       struct {
5820 +               struct au_hdentry *hd;
5821 +               struct dentry *h_dentry;
5822 +       } hdst, hsrc;
5823 +
5824 +       dinfo = au_di(cpg->dentry);
5825 +       AuRwMustWriteLock(&dinfo->di_rwsem);
5826 +
5827 +       bsrc_orig = cpg->bsrc;
5828 +       cpg->bsrc = dinfo->di_btop;
5829 +       hdst.hd = au_hdentry(dinfo, cpg->bdst);
5830 +       hdst.h_dentry = hdst.hd->hd_dentry;
5831 +       hdst.hd->hd_dentry = wh_dentry;
5832 +       dinfo->di_btop = cpg->bdst;
5833 +
5834 +       hsrc.h_dentry = NULL;
5835 +       if (file) {
5836 +               hsrc.hd = au_hdentry(dinfo, cpg->bsrc);
5837 +               hsrc.h_dentry = hsrc.hd->hd_dentry;
5838 +               hsrc.hd->hd_dentry = au_hf_top(file)->f_path.dentry;
5839 +       }
5840 +       flags_orig = cpg->flags;
5841 +       cpg->flags = !AuCpup_DTIME;
5842 +       err = au_cpup_single(cpg, /*h_parent*/NULL);
5843 +       cpg->flags = flags_orig;
5844 +       if (file) {
5845 +               if (!err)
5846 +                       err = au_reopen_nondir(file);
5847 +               hsrc.hd->hd_dentry = hsrc.h_dentry;
5848 +       }
5849 +       hdst.hd->hd_dentry = hdst.h_dentry;
5850 +       dinfo->di_btop = cpg->bsrc;
5851 +       cpg->bsrc = bsrc_orig;
5852 +
5853 +       return err;
5854 +}
5855 +
5856 +static int au_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5857 +{
5858 +       int err;
5859 +       aufs_bindex_t bdst;
5860 +       struct au_dtime dt;
5861 +       struct dentry *dentry, *parent, *h_parent, *wh_dentry;
5862 +       struct au_branch *br;
5863 +       struct path h_path;
5864 +
5865 +       dentry = cpg->dentry;
5866 +       bdst = cpg->bdst;
5867 +       br = au_sbr(dentry->d_sb, bdst);
5868 +       parent = dget_parent(dentry);
5869 +       h_parent = au_h_dptr(parent, bdst);
5870 +       wh_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
5871 +       err = PTR_ERR(wh_dentry);
5872 +       if (IS_ERR(wh_dentry))
5873 +               goto out;
5874 +
5875 +       h_path.dentry = h_parent;
5876 +       h_path.mnt = au_br_mnt(br);
5877 +       au_dtime_store(&dt, parent, &h_path);
5878 +       err = au_do_cpup_wh(cpg, wh_dentry, file);
5879 +       if (unlikely(err))
5880 +               goto out_wh;
5881 +
5882 +       dget(wh_dentry);
5883 +       h_path.dentry = wh_dentry;
5884 +       if (!d_is_dir(wh_dentry)) {
5885 +               /* no delegation since it is just created */
5886 +               err = vfsub_unlink(d_inode(h_parent), &h_path,
5887 +                                  /*delegated*/NULL, /*force*/0);
5888 +       } else
5889 +               err = vfsub_rmdir(d_inode(h_parent), &h_path);
5890 +       if (unlikely(err)) {
5891 +               AuIOErr("failed remove copied-up tmp file %pd(%d)\n",
5892 +                       wh_dentry, err);
5893 +               err = -EIO;
5894 +       }
5895 +       au_dtime_revert(&dt);
5896 +       au_set_hi_wh(d_inode(dentry), bdst, wh_dentry);
5897 +
5898 +out_wh:
5899 +       dput(wh_dentry);
5900 +out:
5901 +       dput(parent);
5902 +       return err;
5903 +}
5904 +
5905 +struct au_cpup_wh_args {
5906 +       int *errp;
5907 +       struct au_cp_generic *cpg;
5908 +       struct file *file;
5909 +};
5910 +
5911 +static void au_call_cpup_wh(void *args)
5912 +{
5913 +       struct au_cpup_wh_args *a = args;
5914 +
5915 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5916 +       *a->errp = au_cpup_wh(a->cpg, a->file);
5917 +       au_pin_hdir_release(a->cpg->pin);
5918 +}
5919 +
5920 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5921 +{
5922 +       int err, wkq_err;
5923 +       aufs_bindex_t bdst;
5924 +       struct dentry *dentry, *parent, *h_orph, *h_parent;
5925 +       struct inode *dir, *h_dir, *h_tmpdir;
5926 +       struct au_wbr *wbr;
5927 +       struct au_pin wh_pin, *pin_orig;
5928 +
5929 +       dentry = cpg->dentry;
5930 +       bdst = cpg->bdst;
5931 +       parent = dget_parent(dentry);
5932 +       dir = d_inode(parent);
5933 +       h_orph = NULL;
5934 +       h_parent = NULL;
5935 +       h_dir = au_igrab(au_h_iptr(dir, bdst));
5936 +       h_tmpdir = h_dir;
5937 +       pin_orig = NULL;
5938 +       if (!h_dir->i_nlink) {
5939 +               wbr = au_sbr(dentry->d_sb, bdst)->br_wbr;
5940 +               h_orph = wbr->wbr_orph;
5941 +
5942 +               h_parent = dget(au_h_dptr(parent, bdst));
5943 +               au_set_h_dptr(parent, bdst, dget(h_orph));
5944 +               h_tmpdir = d_inode(h_orph);
5945 +               au_set_h_iptr(dir, bdst, au_igrab(h_tmpdir), /*flags*/0);
5946 +
5947 +               inode_lock_nested(h_tmpdir, AuLsc_I_PARENT3);
5948 +               /* todo: au_h_open_pre()? */
5949 +
5950 +               pin_orig = cpg->pin;
5951 +               au_pin_init(&wh_pin, dentry, bdst, AuLsc_DI_PARENT,
5952 +                           AuLsc_I_PARENT3, cpg->pin->udba, AuPin_DI_LOCKED);
5953 +               cpg->pin = &wh_pin;
5954 +       }
5955 +
5956 +       if (!au_test_h_perm_sio(h_tmpdir, MAY_EXEC | MAY_WRITE)
5957 +           && !au_cpup_sio_test(cpg->pin, d_inode(dentry)->i_mode))
5958 +               err = au_cpup_wh(cpg, file);
5959 +       else {
5960 +               struct au_cpup_wh_args args = {
5961 +                       .errp   = &err,
5962 +                       .cpg    = cpg,
5963 +                       .file   = file
5964 +               };
5965 +               wkq_err = au_wkq_wait(au_call_cpup_wh, &args);
5966 +               if (unlikely(wkq_err))
5967 +                       err = wkq_err;
5968 +       }
5969 +
5970 +       if (h_orph) {
5971 +               inode_unlock(h_tmpdir);
5972 +               /* todo: au_h_open_post()? */
5973 +               au_set_h_iptr(dir, bdst, au_igrab(h_dir), /*flags*/0);
5974 +               au_set_h_dptr(parent, bdst, h_parent);
5975 +               AuDebugOn(!pin_orig);
5976 +               cpg->pin = pin_orig;
5977 +       }
5978 +       iput(h_dir);
5979 +       dput(parent);
5980 +
5981 +       return err;
5982 +}
5983 +
5984 +/* ---------------------------------------------------------------------- */
5985 +
5986 +/*
5987 + * generic routine for both of copy-up and copy-down.
5988 + */
5989 +/* cf. revalidate function in file.c */
5990 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
5991 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
5992 +                        struct au_pin *pin,
5993 +                        struct dentry *h_parent, void *arg),
5994 +              void *arg)
5995 +{
5996 +       int err;
5997 +       struct au_pin pin;
5998 +       struct dentry *d, *parent, *h_parent, *real_parent, *h_dentry;
5999 +
6000 +       err = 0;
6001 +       parent = dget_parent(dentry);
6002 +       if (IS_ROOT(parent))
6003 +               goto out;
6004 +
6005 +       au_pin_init(&pin, dentry, bdst, AuLsc_DI_PARENT2, AuLsc_I_PARENT2,
6006 +                   au_opt_udba(dentry->d_sb), AuPin_MNT_WRITE);
6007 +
6008 +       /* do not use au_dpage */
6009 +       real_parent = parent;
6010 +       while (1) {
6011 +               dput(parent);
6012 +               parent = dget_parent(dentry);
6013 +               h_parent = au_h_dptr(parent, bdst);
6014 +               if (h_parent)
6015 +                       goto out; /* success */
6016 +
6017 +               /* find top dir which is necessary to cpup */
6018 +               do {
6019 +                       d = parent;
6020 +                       dput(parent);
6021 +                       parent = dget_parent(d);
6022 +                       di_read_lock_parent3(parent, !AuLock_IR);
6023 +                       h_parent = au_h_dptr(parent, bdst);
6024 +                       di_read_unlock(parent, !AuLock_IR);
6025 +               } while (!h_parent);
6026 +
6027 +               if (d != real_parent)
6028 +                       di_write_lock_child3(d);
6029 +
6030 +               /* somebody else might create while we were sleeping */
6031 +               h_dentry = au_h_dptr(d, bdst);
6032 +               if (!h_dentry || d_is_negative(h_dentry)) {
6033 +                       if (h_dentry)
6034 +                               au_update_dbtop(d);
6035 +
6036 +                       au_pin_set_dentry(&pin, d);
6037 +                       err = au_do_pin(&pin);
6038 +                       if (!err) {
6039 +                               err = cp(d, bdst, &pin, h_parent, arg);
6040 +                               au_unpin(&pin);
6041 +                       }
6042 +               }
6043 +
6044 +               if (d != real_parent)
6045 +                       di_write_unlock(d);
6046 +               if (unlikely(err))
6047 +                       break;
6048 +       }
6049 +
6050 +out:
6051 +       dput(parent);
6052 +       return err;
6053 +}
6054 +
6055 +static int au_cpup_dir(struct dentry *dentry, aufs_bindex_t bdst,
6056 +                      struct au_pin *pin,
6057 +                      struct dentry *h_parent __maybe_unused,
6058 +                      void *arg __maybe_unused)
6059 +{
6060 +       struct au_cp_generic cpg = {
6061 +               .dentry = dentry,
6062 +               .bdst   = bdst,
6063 +               .bsrc   = -1,
6064 +               .len    = 0,
6065 +               .pin    = pin,
6066 +               .flags  = AuCpup_DTIME
6067 +       };
6068 +       return au_sio_cpup_simple(&cpg);
6069 +}
6070 +
6071 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
6072 +{
6073 +       return au_cp_dirs(dentry, bdst, au_cpup_dir, NULL);
6074 +}
6075 +
6076 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
6077 +{
6078 +       int err;
6079 +       struct dentry *parent;
6080 +       struct inode *dir;
6081 +
6082 +       parent = dget_parent(dentry);
6083 +       dir = d_inode(parent);
6084 +       err = 0;
6085 +       if (au_h_iptr(dir, bdst))
6086 +               goto out;
6087 +
6088 +       di_read_unlock(parent, AuLock_IR);
6089 +       di_write_lock_parent(parent);
6090 +       /* someone else might change our inode while we were sleeping */
6091 +       if (!au_h_iptr(dir, bdst))
6092 +               err = au_cpup_dirs(dentry, bdst);
6093 +       di_downgrade_lock(parent, AuLock_IR);
6094 +
6095 +out:
6096 +       dput(parent);
6097 +       return err;
6098 +}
6099 diff -urN /usr/share/empty/fs/aufs/cpup.h linux/fs/aufs/cpup.h
6100 --- /usr/share/empty/fs/aufs/cpup.h     1970-01-01 01:00:00.000000000 +0100
6101 +++ linux/fs/aufs/cpup.h        2017-05-06 22:16:52.774886674 +0200
6102 @@ -0,0 +1,94 @@
6103 +/*
6104 + * Copyright (C) 2005-2017 Junjiro R. Okajima
6105 + *
6106 + * This program, aufs is free software; you can redistribute it and/or modify
6107 + * it under the terms of the GNU General Public License as published by
6108 + * the Free Software Foundation; either version 2 of the License, or
6109 + * (at your option) any later version.
6110 + *
6111 + * This program is distributed in the hope that it will be useful,
6112 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6113 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6114 + * GNU General Public License for more details.
6115 + *
6116 + * You should have received a copy of the GNU General Public License
6117 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6118 + */
6119 +
6120 +/*
6121 + * copy-up/down functions
6122 + */
6123 +
6124 +#ifndef __AUFS_CPUP_H__
6125 +#define __AUFS_CPUP_H__
6126 +
6127 +#ifdef __KERNEL__
6128 +
6129 +#include <linux/path.h>
6130 +
6131 +struct inode;
6132 +struct file;
6133 +struct au_pin;
6134 +
6135 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags);
6136 +void au_cpup_attr_timesizes(struct inode *inode);
6137 +void au_cpup_attr_nlink(struct inode *inode, int force);
6138 +void au_cpup_attr_changeable(struct inode *inode);
6139 +void au_cpup_igen(struct inode *inode, struct inode *h_inode);
6140 +void au_cpup_attr_all(struct inode *inode, int force);
6141 +
6142 +/* ---------------------------------------------------------------------- */
6143 +
6144 +struct au_cp_generic {
6145 +       struct dentry   *dentry;
6146 +       aufs_bindex_t   bdst, bsrc;
6147 +       loff_t          len;
6148 +       struct au_pin   *pin;
6149 +       unsigned int    flags;
6150 +};
6151 +
6152 +/* cpup flags */
6153 +#define AuCpup_DTIME           1               /* do dtime_store/revert */
6154 +#define AuCpup_KEEPLINO                (1 << 1)        /* do not clear the lower xino,
6155 +                                                  for link(2) */
6156 +#define AuCpup_RENAME          (1 << 2)        /* rename after cpup */
6157 +#define AuCpup_HOPEN           (1 << 3)        /* call h_open_pre/post() in
6158 +                                                  cpup */
6159 +#define AuCpup_OVERWRITE       (1 << 4)        /* allow overwriting the
6160 +                                                  existing entry */
6161 +#define AuCpup_RWDST           (1 << 5)        /* force write target even if
6162 +                                                  the branch is marked as RO */
6163 +
6164 +#define au_ftest_cpup(flags, name)     ((flags) & AuCpup_##name)
6165 +#define au_fset_cpup(flags, name) \
6166 +       do { (flags) |= AuCpup_##name; } while (0)
6167 +#define au_fclr_cpup(flags, name) \
6168 +       do { (flags) &= ~AuCpup_##name; } while (0)
6169 +
6170 +int au_copy_file(struct file *dst, struct file *src, loff_t len);
6171 +int au_sio_cpup_simple(struct au_cp_generic *cpg);
6172 +int au_sio_cpdown_simple(struct au_cp_generic *cpg);
6173 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file);
6174 +
6175 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
6176 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
6177 +                        struct au_pin *pin,
6178 +                        struct dentry *h_parent, void *arg),
6179 +              void *arg);
6180 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
6181 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
6182 +
6183 +/* ---------------------------------------------------------------------- */
6184 +
6185 +/* keep timestamps when copyup */
6186 +struct au_dtime {
6187 +       struct dentry *dt_dentry;
6188 +       struct path dt_h_path;
6189 +       struct timespec dt_atime, dt_mtime;
6190 +};
6191 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
6192 +                   struct path *h_path);
6193 +void au_dtime_revert(struct au_dtime *dt);
6194 +
6195 +#endif /* __KERNEL__ */
6196 +#endif /* __AUFS_CPUP_H__ */
6197 diff -urN /usr/share/empty/fs/aufs/dbgaufs.c linux/fs/aufs/dbgaufs.c
6198 --- /usr/share/empty/fs/aufs/dbgaufs.c  1970-01-01 01:00:00.000000000 +0100
6199 +++ linux/fs/aufs/dbgaufs.c     2017-05-06 22:16:52.774886674 +0200
6200 @@ -0,0 +1,438 @@
6201 +/*
6202 + * Copyright (C) 2005-2017 Junjiro R. Okajima
6203 + *
6204 + * This program, aufs is free software; you can redistribute it and/or modify
6205 + * it under the terms of the GNU General Public License as published by
6206 + * the Free Software Foundation; either version 2 of the License, or
6207 + * (at your option) any later version.
6208 + *
6209 + * This program is distributed in the hope that it will be useful,
6210 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6211 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6212 + * GNU General Public License for more details.
6213 + *
6214 + * You should have received a copy of the GNU General Public License
6215 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6216 + */
6217 +
6218 +/*
6219 + * debugfs interface
6220 + */
6221 +
6222 +#include <linux/debugfs.h>
6223 +#include "aufs.h"
6224 +
6225 +#ifndef CONFIG_SYSFS
6226 +#error DEBUG_FS depends upon SYSFS
6227 +#endif
6228 +
6229 +static struct dentry *dbgaufs;
6230 +static const mode_t dbgaufs_mode = S_IRUSR | S_IRGRP | S_IROTH;
6231 +
6232 +/* 20 is max digits length of ulong 64 */
6233 +struct dbgaufs_arg {
6234 +       int n;
6235 +       char a[20 * 4];
6236 +};
6237 +
6238 +/*
6239 + * common function for all XINO files
6240 + */
6241 +static int dbgaufs_xi_release(struct inode *inode __maybe_unused,
6242 +                             struct file *file)
6243 +{
6244 +       au_delayed_kfree(file->private_data);
6245 +       return 0;
6246 +}
6247 +
6248 +static int dbgaufs_xi_open(struct file *xf, struct file *file, int do_fcnt)
6249 +{
6250 +       int err;
6251 +       struct kstat st;
6252 +       struct dbgaufs_arg *p;
6253 +
6254 +       err = -ENOMEM;
6255 +       p = kmalloc(sizeof(*p), GFP_NOFS);
6256 +       if (unlikely(!p))
6257 +               goto out;
6258 +
6259 +       err = 0;
6260 +       p->n = 0;
6261 +       file->private_data = p;
6262 +       if (!xf)
6263 +               goto out;
6264 +
6265 +       err = vfsub_getattr(&xf->f_path, &st);
6266 +       if (!err) {
6267 +               if (do_fcnt)
6268 +                       p->n = snprintf
6269 +                               (p->a, sizeof(p->a), "%ld, %llux%u %lld\n",
6270 +                                (long)file_count(xf), st.blocks, st.blksize,
6271 +                                (long long)st.size);
6272 +               else
6273 +                       p->n = snprintf(p->a, sizeof(p->a), "%llux%u %lld\n",
6274 +                                       st.blocks, st.blksize,
6275 +                                       (long long)st.size);
6276 +               AuDebugOn(p->n >= sizeof(p->a));
6277 +       } else {
6278 +               p->n = snprintf(p->a, sizeof(p->a), "err %d\n", err);
6279 +               err = 0;
6280 +       }
6281 +
6282 +out:
6283 +       return err;
6284 +
6285 +}
6286 +
6287 +static ssize_t dbgaufs_xi_read(struct file *file, char __user *buf,
6288 +                              size_t count, loff_t *ppos)
6289 +{
6290 +       struct dbgaufs_arg *p;
6291 +
6292 +       p = file->private_data;
6293 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
6294 +}
6295 +
6296 +/* ---------------------------------------------------------------------- */
6297 +
6298 +struct dbgaufs_plink_arg {
6299 +       int n;
6300 +       char a[];
6301 +};
6302 +
6303 +static int dbgaufs_plink_release(struct inode *inode __maybe_unused,
6304 +                                struct file *file)
6305 +{
6306 +       au_delayed_free_page((unsigned long)file->private_data);
6307 +       return 0;
6308 +}
6309 +
6310 +static int dbgaufs_plink_open(struct inode *inode, struct file *file)
6311 +{
6312 +       int err, i, limit;
6313 +       unsigned long n, sum;
6314 +       struct dbgaufs_plink_arg *p;
6315 +       struct au_sbinfo *sbinfo;
6316 +       struct super_block *sb;
6317 +       struct au_sphlhead *sphl;
6318 +
6319 +       err = -ENOMEM;
6320 +       p = (void *)get_zeroed_page(GFP_NOFS);
6321 +       if (unlikely(!p))
6322 +               goto out;
6323 +
6324 +       err = -EFBIG;
6325 +       sbinfo = inode->i_private;
6326 +       sb = sbinfo->si_sb;
6327 +       si_noflush_read_lock(sb);
6328 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
6329 +               limit = PAGE_SIZE - sizeof(p->n);
6330 +
6331 +               /* the number of buckets */
6332 +               n = snprintf(p->a + p->n, limit, "%d\n", AuPlink_NHASH);
6333 +               p->n += n;
6334 +               limit -= n;
6335 +
6336 +               sum = 0;
6337 +               for (i = 0, sphl = sbinfo->si_plink;
6338 +                    i < AuPlink_NHASH;
6339 +                    i++, sphl++) {
6340 +                       n = au_sphl_count(sphl);
6341 +                       sum += n;
6342 +
6343 +                       n = snprintf(p->a + p->n, limit, "%lu ", n);
6344 +                       p->n += n;
6345 +                       limit -= n;
6346 +                       if (unlikely(limit <= 0))
6347 +                               goto out_free;
6348 +               }
6349 +               p->a[p->n - 1] = '\n';
6350 +
6351 +               /* the sum of plinks */
6352 +               n = snprintf(p->a + p->n, limit, "%lu\n", sum);
6353 +               p->n += n;
6354 +               limit -= n;
6355 +               if (unlikely(limit <= 0))
6356 +                       goto out_free;
6357 +       } else {
6358 +#define str "1\n0\n0\n"
6359 +               p->n = sizeof(str) - 1;
6360 +               strcpy(p->a, str);
6361 +#undef str
6362 +       }
6363 +       si_read_unlock(sb);
6364 +
6365 +       err = 0;
6366 +       file->private_data = p;
6367 +       goto out; /* success */
6368 +
6369 +out_free:
6370 +       au_delayed_free_page((unsigned long)p);
6371 +out:
6372 +       return err;
6373 +}
6374 +
6375 +static ssize_t dbgaufs_plink_read(struct file *file, char __user *buf,
6376 +                                 size_t count, loff_t *ppos)
6377 +{
6378 +       struct dbgaufs_plink_arg *p;
6379 +
6380 +       p = file->private_data;
6381 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
6382 +}
6383 +
6384 +static const struct file_operations dbgaufs_plink_fop = {
6385 +       .owner          = THIS_MODULE,
6386 +       .open           = dbgaufs_plink_open,
6387 +       .release        = dbgaufs_plink_release,
6388 +       .read           = dbgaufs_plink_read
6389 +};
6390 +
6391 +/* ---------------------------------------------------------------------- */
6392 +
6393 +static int dbgaufs_xib_open(struct inode *inode, struct file *file)
6394 +{
6395 +       int err;
6396 +       struct au_sbinfo *sbinfo;
6397 +       struct super_block *sb;
6398 +
6399 +       sbinfo = inode->i_private;
6400 +       sb = sbinfo->si_sb;
6401 +       si_noflush_read_lock(sb);
6402 +       err = dbgaufs_xi_open(sbinfo->si_xib, file, /*do_fcnt*/0);
6403 +       si_read_unlock(sb);
6404 +       return err;
6405 +}
6406 +
6407 +static const struct file_operations dbgaufs_xib_fop = {
6408 +       .owner          = THIS_MODULE,
6409 +       .open           = dbgaufs_xib_open,
6410 +       .release        = dbgaufs_xi_release,
6411 +       .read           = dbgaufs_xi_read
6412 +};
6413 +
6414 +/* ---------------------------------------------------------------------- */
6415 +
6416 +#define DbgaufsXi_PREFIX "xi"
6417 +
6418 +static int dbgaufs_xino_open(struct inode *inode, struct file *file)
6419 +{
6420 +       int err;
6421 +       long l;
6422 +       struct au_sbinfo *sbinfo;
6423 +       struct super_block *sb;
6424 +       struct file *xf;
6425 +       struct qstr *name;
6426 +
6427 +       err = -ENOENT;
6428 +       xf = NULL;
6429 +       name = &file->f_path.dentry->d_name;
6430 +       if (unlikely(name->len < sizeof(DbgaufsXi_PREFIX)
6431 +                    || memcmp(name->name, DbgaufsXi_PREFIX,
6432 +                              sizeof(DbgaufsXi_PREFIX) - 1)))
6433 +               goto out;
6434 +       err = kstrtol(name->name + sizeof(DbgaufsXi_PREFIX) - 1, 10, &l);
6435 +       if (unlikely(err))
6436 +               goto out;
6437 +
6438 +       sbinfo = inode->i_private;
6439 +       sb = sbinfo->si_sb;
6440 +       si_noflush_read_lock(sb);
6441 +       if (l <= au_sbbot(sb)) {
6442 +               xf = au_sbr(sb, (aufs_bindex_t)l)->br_xino.xi_file;
6443 +               err = dbgaufs_xi_open(xf, file, /*do_fcnt*/1);
6444 +       } else
6445 +               err = -ENOENT;
6446 +       si_read_unlock(sb);
6447 +
6448 +out:
6449 +       return err;
6450 +}
6451 +
6452 +static const struct file_operations dbgaufs_xino_fop = {
6453 +       .owner          = THIS_MODULE,
6454 +       .open           = dbgaufs_xino_open,
6455 +       .release        = dbgaufs_xi_release,
6456 +       .read           = dbgaufs_xi_read
6457 +};
6458 +
6459 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
6460 +{
6461 +       aufs_bindex_t bbot;
6462 +       struct au_branch *br;
6463 +       struct au_xino_file *xi;
6464 +
6465 +       if (!au_sbi(sb)->si_dbgaufs)
6466 +               return;
6467 +
6468 +       bbot = au_sbbot(sb);
6469 +       for (; bindex <= bbot; bindex++) {
6470 +               br = au_sbr(sb, bindex);
6471 +               xi = &br->br_xino;
6472 +               /* debugfs acquires the parent i_mutex */
6473 +               lockdep_off();
6474 +               debugfs_remove(xi->xi_dbgaufs);
6475 +               lockdep_on();
6476 +               xi->xi_dbgaufs = NULL;
6477 +       }
6478 +}
6479 +
6480 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
6481 +{
6482 +       struct au_sbinfo *sbinfo;
6483 +       struct dentry *parent;
6484 +       struct au_branch *br;
6485 +       struct au_xino_file *xi;
6486 +       aufs_bindex_t bbot;
6487 +       char name[sizeof(DbgaufsXi_PREFIX) + 5]; /* "xi" bindex NULL */
6488 +
6489 +       sbinfo = au_sbi(sb);
6490 +       parent = sbinfo->si_dbgaufs;
6491 +       if (!parent)
6492 +               return;
6493 +
6494 +       bbot = au_sbbot(sb);
6495 +       for (; bindex <= bbot; bindex++) {
6496 +               snprintf(name, sizeof(name), DbgaufsXi_PREFIX "%d", bindex);
6497 +               br = au_sbr(sb, bindex);
6498 +               xi = &br->br_xino;
6499 +               AuDebugOn(xi->xi_dbgaufs);
6500 +               /* debugfs acquires the parent i_mutex */
6501 +               lockdep_off();
6502 +               xi->xi_dbgaufs = debugfs_create_file(name, dbgaufs_mode, parent,
6503 +                                                    sbinfo, &dbgaufs_xino_fop);
6504 +               lockdep_on();
6505 +               /* ignore an error */
6506 +               if (unlikely(!xi->xi_dbgaufs))
6507 +                       AuWarn1("failed %s under debugfs\n", name);
6508 +       }
6509 +}
6510 +
6511 +/* ---------------------------------------------------------------------- */
6512 +
6513 +#ifdef CONFIG_AUFS_EXPORT
6514 +static int dbgaufs_xigen_open(struct inode *inode, struct file *file)
6515 +{
6516 +       int err;
6517 +       struct au_sbinfo *sbinfo;
6518 +       struct super_block *sb;
6519 +
6520 +       sbinfo = inode->i_private;
6521 +       sb = sbinfo->si_sb;
6522 +       si_noflush_read_lock(sb);
6523 +       err = dbgaufs_xi_open(sbinfo->si_xigen, file, /*do_fcnt*/0);
6524 +       si_read_unlock(sb);
6525 +       return err;
6526 +}
6527 +
6528 +static const struct file_operations dbgaufs_xigen_fop = {
6529 +       .owner          = THIS_MODULE,
6530 +       .open           = dbgaufs_xigen_open,
6531 +       .release        = dbgaufs_xi_release,
6532 +       .read           = dbgaufs_xi_read
6533 +};
6534 +
6535 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6536 +{
6537 +       int err;
6538 +
6539 +       /*
6540 +        * This function is a dynamic '__init' function actually,
6541 +        * so the tiny check for si_rwsem is unnecessary.
6542 +        */
6543 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6544 +
6545 +       err = -EIO;
6546 +       sbinfo->si_dbgaufs_xigen = debugfs_create_file
6547 +               ("xigen", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6548 +                &dbgaufs_xigen_fop);
6549 +       if (sbinfo->si_dbgaufs_xigen)
6550 +               err = 0;
6551 +
6552 +       return err;
6553 +}
6554 +#else
6555 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6556 +{
6557 +       return 0;
6558 +}
6559 +#endif /* CONFIG_AUFS_EXPORT */
6560 +
6561 +/* ---------------------------------------------------------------------- */
6562 +
6563 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo)
6564 +{
6565 +       /*
6566 +        * This function is a dynamic '__fin' function actually,
6567 +        * so the tiny check for si_rwsem is unnecessary.
6568 +        */
6569 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6570 +
6571 +       debugfs_remove_recursive(sbinfo->si_dbgaufs);
6572 +       sbinfo->si_dbgaufs = NULL;
6573 +       kobject_put(&sbinfo->si_kobj);
6574 +}
6575 +
6576 +int dbgaufs_si_init(struct au_sbinfo *sbinfo)
6577 +{
6578 +       int err;
6579 +       char name[SysaufsSiNameLen];
6580 +
6581 +       /*
6582 +        * This function is a dynamic '__init' function actually,
6583 +        * so the tiny check for si_rwsem is unnecessary.
6584 +        */
6585 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6586 +
6587 +       err = -ENOENT;
6588 +       if (!dbgaufs) {
6589 +               AuErr1("/debug/aufs is uninitialized\n");
6590 +               goto out;
6591 +       }
6592 +
6593 +       err = -EIO;
6594 +       sysaufs_name(sbinfo, name);
6595 +       sbinfo->si_dbgaufs = debugfs_create_dir(name, dbgaufs);
6596 +       if (unlikely(!sbinfo->si_dbgaufs))
6597 +               goto out;
6598 +       kobject_get(&sbinfo->si_kobj);
6599 +
6600 +       sbinfo->si_dbgaufs_xib = debugfs_create_file
6601 +               ("xib", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6602 +                &dbgaufs_xib_fop);
6603 +       if (unlikely(!sbinfo->si_dbgaufs_xib))
6604 +               goto out_dir;
6605 +
6606 +       sbinfo->si_dbgaufs_plink = debugfs_create_file
6607 +               ("plink", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6608 +                &dbgaufs_plink_fop);
6609 +       if (unlikely(!sbinfo->si_dbgaufs_plink))
6610 +               goto out_dir;
6611 +
6612 +       err = dbgaufs_xigen_init(sbinfo);
6613 +       if (!err)
6614 +               goto out; /* success */
6615 +
6616 +out_dir:
6617 +       dbgaufs_si_fin(sbinfo);
6618 +out:
6619 +       return err;
6620 +}
6621 +
6622 +/* ---------------------------------------------------------------------- */
6623 +
6624 +void dbgaufs_fin(void)
6625 +{
6626 +       debugfs_remove(dbgaufs);
6627 +}
6628 +
6629 +int __init dbgaufs_init(void)
6630 +{
6631 +       int err;
6632 +
6633 +       err = -EIO;
6634 +       dbgaufs = debugfs_create_dir(AUFS_NAME, NULL);
6635 +       if (dbgaufs)
6636 +               err = 0;
6637 +       return err;
6638 +}
6639 diff -urN /usr/share/empty/fs/aufs/dbgaufs.h linux/fs/aufs/dbgaufs.h
6640 --- /usr/share/empty/fs/aufs/dbgaufs.h  1970-01-01 01:00:00.000000000 +0100
6641 +++ linux/fs/aufs/dbgaufs.h     2017-05-06 22:16:52.774886674 +0200
6642 @@ -0,0 +1,48 @@
6643 +/*
6644 + * Copyright (C) 2005-2017 Junjiro R. Okajima
6645 + *
6646 + * This program, aufs is free software; you can redistribute it and/or modify
6647 + * it under the terms of the GNU General Public License as published by
6648 + * the Free Software Foundation; either version 2 of the License, or
6649 + * (at your option) any later version.
6650 + *
6651 + * This program is distributed in the hope that it will be useful,
6652 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6653 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6654 + * GNU General Public License for more details.
6655 + *
6656 + * You should have received a copy of the GNU General Public License
6657 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6658 + */
6659 +
6660 +/*
6661 + * debugfs interface
6662 + */
6663 +
6664 +#ifndef __DBGAUFS_H__
6665 +#define __DBGAUFS_H__
6666 +
6667 +#ifdef __KERNEL__
6668 +
6669 +struct super_block;
6670 +struct au_sbinfo;
6671 +
6672 +#ifdef CONFIG_DEBUG_FS
6673 +/* dbgaufs.c */
6674 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
6675 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
6676 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo);
6677 +int dbgaufs_si_init(struct au_sbinfo *sbinfo);
6678 +void dbgaufs_fin(void);
6679 +int __init dbgaufs_init(void);
6680 +#else
6681 +AuStubVoid(dbgaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
6682 +AuStubVoid(dbgaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
6683 +AuStubVoid(dbgaufs_si_fin, struct au_sbinfo *sbinfo)
6684 +AuStubInt0(dbgaufs_si_init, struct au_sbinfo *sbinfo)
6685 +AuStubVoid(dbgaufs_fin, void)
6686 +AuStubInt0(__init dbgaufs_init, void)
6687 +#endif /* CONFIG_DEBUG_FS */
6688 +
6689 +#endif /* __KERNEL__ */
6690 +#endif /* __DBGAUFS_H__ */
6691 diff -urN /usr/share/empty/fs/aufs/dcsub.c linux/fs/aufs/dcsub.c
6692 --- /usr/share/empty/fs/aufs/dcsub.c    1970-01-01 01:00:00.000000000 +0100
6693 +++ linux/fs/aufs/dcsub.c       2017-05-06 22:16:52.774886674 +0200
6694 @@ -0,0 +1,225 @@
6695 +/*
6696 + * Copyright (C) 2005-2017 Junjiro R. Okajima
6697 + *
6698 + * This program, aufs is free software; you can redistribute it and/or modify
6699 + * it under the terms of the GNU General Public License as published by
6700 + * the Free Software Foundation; either version 2 of the License, or
6701 + * (at your option) any later version.
6702 + *
6703 + * This program is distributed in the hope that it will be useful,
6704 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6705 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6706 + * GNU General Public License for more details.
6707 + *
6708 + * You should have received a copy of the GNU General Public License
6709 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6710 + */
6711 +
6712 +/*
6713 + * sub-routines for dentry cache
6714 + */
6715 +
6716 +#include "aufs.h"
6717 +
6718 +static void au_dpage_free(struct au_dpage *dpage)
6719 +{
6720 +       int i;
6721 +       struct dentry **p;
6722 +
6723 +       p = dpage->dentries;
6724 +       for (i = 0; i < dpage->ndentry; i++)
6725 +               dput(*p++);
6726 +       au_delayed_free_page((unsigned long)dpage->dentries);
6727 +}
6728 +
6729 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp)
6730 +{
6731 +       int err;
6732 +       void *p;
6733 +
6734 +       err = -ENOMEM;
6735 +       dpages->dpages = kmalloc(sizeof(*dpages->dpages), gfp);
6736 +       if (unlikely(!dpages->dpages))
6737 +               goto out;
6738 +
6739 +       p = (void *)__get_free_page(gfp);
6740 +       if (unlikely(!p))
6741 +               goto out_dpages;
6742 +
6743 +       dpages->dpages[0].ndentry = 0;
6744 +       dpages->dpages[0].dentries = p;
6745 +       dpages->ndpage = 1;
6746 +       return 0; /* success */
6747 +
6748 +out_dpages:
6749 +       au_delayed_kfree(dpages->dpages);
6750 +out:
6751 +       return err;
6752 +}
6753 +
6754 +void au_dpages_free(struct au_dcsub_pages *dpages)
6755 +{
6756 +       int i;
6757 +       struct au_dpage *p;
6758 +
6759 +       p = dpages->dpages;
6760 +       for (i = 0; i < dpages->ndpage; i++)
6761 +               au_dpage_free(p++);
6762 +       au_delayed_kfree(dpages->dpages);
6763 +}
6764 +
6765 +static int au_dpages_append(struct au_dcsub_pages *dpages,
6766 +                           struct dentry *dentry, gfp_t gfp)
6767 +{
6768 +       int err, sz;
6769 +       struct au_dpage *dpage;
6770 +       void *p;
6771 +
6772 +       dpage = dpages->dpages + dpages->ndpage - 1;
6773 +       sz = PAGE_SIZE / sizeof(dentry);
6774 +       if (unlikely(dpage->ndentry >= sz)) {
6775 +               AuLabel(new dpage);
6776 +               err = -ENOMEM;
6777 +               sz = dpages->ndpage * sizeof(*dpages->dpages);
6778 +               p = au_kzrealloc(dpages->dpages, sz,
6779 +                                sz + sizeof(*dpages->dpages), gfp,
6780 +                                /*may_shrink*/0);
6781 +               if (unlikely(!p))
6782 +                       goto out;
6783 +
6784 +               dpages->dpages = p;
6785 +               dpage = dpages->dpages + dpages->ndpage;
6786 +               p = (void *)__get_free_page(gfp);
6787 +               if (unlikely(!p))
6788 +                       goto out;
6789 +
6790 +               dpage->ndentry = 0;
6791 +               dpage->dentries = p;
6792 +               dpages->ndpage++;
6793 +       }
6794 +
6795 +       AuDebugOn(au_dcount(dentry) <= 0);
6796 +       dpage->dentries[dpage->ndentry++] = dget_dlock(dentry);
6797 +       return 0; /* success */
6798 +
6799 +out:
6800 +       return err;
6801 +}
6802 +
6803 +/* todo: BAD approach */
6804 +/* copied from linux/fs/dcache.c */
6805 +enum d_walk_ret {
6806 +       D_WALK_CONTINUE,
6807 +       D_WALK_QUIT,
6808 +       D_WALK_NORETRY,
6809 +       D_WALK_SKIP,
6810 +};
6811 +
6812 +extern void d_walk(struct dentry *parent, void *data,
6813 +                  enum d_walk_ret (*enter)(void *, struct dentry *),
6814 +                  void (*finish)(void *));
6815 +
6816 +struct ac_dpages_arg {
6817 +       int err;
6818 +       struct au_dcsub_pages *dpages;
6819 +       struct super_block *sb;
6820 +       au_dpages_test test;
6821 +       void *arg;
6822 +};
6823 +
6824 +static enum d_walk_ret au_call_dpages_append(void *_arg, struct dentry *dentry)
6825 +{
6826 +       enum d_walk_ret ret;
6827 +       struct ac_dpages_arg *arg = _arg;
6828 +
6829 +       ret = D_WALK_CONTINUE;
6830 +       if (dentry->d_sb == arg->sb
6831 +           && !IS_ROOT(dentry)
6832 +           && au_dcount(dentry) > 0
6833 +           && au_di(dentry)
6834 +           && (!arg->test || arg->test(dentry, arg->arg))) {
6835 +               arg->err = au_dpages_append(arg->dpages, dentry, GFP_ATOMIC);
6836 +               if (unlikely(arg->err))
6837 +                       ret = D_WALK_QUIT;
6838 +       }
6839 +
6840 +       return ret;
6841 +}
6842 +
6843 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6844 +                  au_dpages_test test, void *arg)
6845 +{
6846 +       struct ac_dpages_arg args = {
6847 +               .err    = 0,
6848 +               .dpages = dpages,
6849 +               .sb     = root->d_sb,
6850 +               .test   = test,
6851 +               .arg    = arg
6852 +       };
6853 +
6854 +       d_walk(root, &args, au_call_dpages_append, NULL);
6855 +
6856 +       return args.err;
6857 +}
6858 +
6859 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6860 +                      int do_include, au_dpages_test test, void *arg)
6861 +{
6862 +       int err;
6863 +
6864 +       err = 0;
6865 +       write_seqlock(&rename_lock);
6866 +       spin_lock(&dentry->d_lock);
6867 +       if (do_include
6868 +           && au_dcount(dentry) > 0
6869 +           && (!test || test(dentry, arg)))
6870 +               err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6871 +       spin_unlock(&dentry->d_lock);
6872 +       if (unlikely(err))
6873 +               goto out;
6874 +
6875 +       /*
6876 +        * RCU for vfsmount is unnecessary since this is a traverse in a single
6877 +        * mount
6878 +        */
6879 +       while (!IS_ROOT(dentry)) {
6880 +               dentry = dentry->d_parent; /* rename_lock is locked */
6881 +               spin_lock(&dentry->d_lock);
6882 +               if (au_dcount(dentry) > 0
6883 +                   && (!test || test(dentry, arg)))
6884 +                       err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6885 +               spin_unlock(&dentry->d_lock);
6886 +               if (unlikely(err))
6887 +                       break;
6888 +       }
6889 +
6890 +out:
6891 +       write_sequnlock(&rename_lock);
6892 +       return err;
6893 +}
6894 +
6895 +static inline int au_dcsub_dpages_aufs(struct dentry *dentry, void *arg)
6896 +{
6897 +       return au_di(dentry) && dentry->d_sb == arg;
6898 +}
6899 +
6900 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6901 +                           struct dentry *dentry, int do_include)
6902 +{
6903 +       return au_dcsub_pages_rev(dpages, dentry, do_include,
6904 +                                 au_dcsub_dpages_aufs, dentry->d_sb);
6905 +}
6906 +
6907 +int au_test_subdir(struct dentry *d1, struct dentry *d2)
6908 +{
6909 +       struct path path[2] = {
6910 +               {
6911 +                       .dentry = d1
6912 +               },
6913 +               {
6914 +                       .dentry = d2
6915 +               }
6916 +       };
6917 +
6918 +       return path_is_under(path + 0, path + 1);
6919 +}
6920 diff -urN /usr/share/empty/fs/aufs/dcsub.h linux/fs/aufs/dcsub.h
6921 --- /usr/share/empty/fs/aufs/dcsub.h    1970-01-01 01:00:00.000000000 +0100
6922 +++ linux/fs/aufs/dcsub.h       2017-05-06 22:16:52.778220093 +0200
6923 @@ -0,0 +1,136 @@
6924 +/*
6925 + * Copyright (C) 2005-2017 Junjiro R. Okajima
6926 + *
6927 + * This program, aufs is free software; you can redistribute it and/or modify
6928 + * it under the terms of the GNU General Public License as published by
6929 + * the Free Software Foundation; either version 2 of the License, or
6930 + * (at your option) any later version.
6931 + *
6932 + * This program is distributed in the hope that it will be useful,
6933 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6934 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6935 + * GNU General Public License for more details.
6936 + *
6937 + * You should have received a copy of the GNU General Public License
6938 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6939 + */
6940 +
6941 +/*
6942 + * sub-routines for dentry cache
6943 + */
6944 +
6945 +#ifndef __AUFS_DCSUB_H__
6946 +#define __AUFS_DCSUB_H__
6947 +
6948 +#ifdef __KERNEL__
6949 +
6950 +#include <linux/dcache.h>
6951 +#include <linux/fs.h>
6952 +
6953 +struct au_dpage {
6954 +       int ndentry;
6955 +       struct dentry **dentries;
6956 +};
6957 +
6958 +struct au_dcsub_pages {
6959 +       int ndpage;
6960 +       struct au_dpage *dpages;
6961 +};
6962 +
6963 +/* ---------------------------------------------------------------------- */
6964 +
6965 +/* dcsub.c */
6966 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp);
6967 +void au_dpages_free(struct au_dcsub_pages *dpages);
6968 +typedef int (*au_dpages_test)(struct dentry *dentry, void *arg);
6969 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6970 +                  au_dpages_test test, void *arg);
6971 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6972 +                      int do_include, au_dpages_test test, void *arg);
6973 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6974 +                           struct dentry *dentry, int do_include);
6975 +int au_test_subdir(struct dentry *d1, struct dentry *d2);
6976 +
6977 +/* ---------------------------------------------------------------------- */
6978 +
6979 +/*
6980 + * todo: in linux-3.13, several similar (but faster) helpers are added to
6981 + * include/linux/dcache.h. Try them (in the future).
6982 + */
6983 +
6984 +static inline int au_d_hashed_positive(struct dentry *d)
6985 +{
6986 +       int err;
6987 +       struct inode *inode = d_inode(d);
6988 +
6989 +       err = 0;
6990 +       if (unlikely(d_unhashed(d)
6991 +                    || d_is_negative(d)
6992 +                    || !inode->i_nlink))
6993 +               err = -ENOENT;
6994 +       return err;
6995 +}
6996 +
6997 +static inline int au_d_linkable(struct dentry *d)
6998 +{
6999 +       int err;
7000 +       struct inode *inode = d_inode(d);
7001 +
7002 +       err = au_d_hashed_positive(d);
7003 +       if (err
7004 +           && d_is_positive(d)
7005 +           && (inode->i_state & I_LINKABLE))
7006 +               err = 0;
7007 +       return err;
7008 +}
7009 +
7010 +static inline int au_d_alive(struct dentry *d)
7011 +{
7012 +       int err;
7013 +       struct inode *inode;
7014 +
7015 +       err = 0;
7016 +       if (!IS_ROOT(d))
7017 +               err = au_d_hashed_positive(d);
7018 +       else {
7019 +               inode = d_inode(d);
7020 +               if (unlikely(d_unlinked(d)
7021 +                            || d_is_negative(d)
7022 +                            || !inode->i_nlink))
7023 +                       err = -ENOENT;
7024 +       }
7025 +       return err;
7026 +}
7027 +
7028 +static inline int au_alive_dir(struct dentry *d)
7029 +{
7030 +       int err;
7031 +
7032 +       err = au_d_alive(d);
7033 +       if (unlikely(err || IS_DEADDIR(d_inode(d))))
7034 +               err = -ENOENT;
7035 +       return err;
7036 +}
7037 +
7038 +static inline int au_qstreq(struct qstr *a, struct qstr *b)
7039 +{
7040 +       return a->len == b->len
7041 +               && !memcmp(a->name, b->name, a->len);
7042 +}
7043 +
7044 +/*
7045 + * by the commit
7046 + * 360f547 2015-01-25 dcache: let the dentry count go down to zero without
7047 + *                     taking d_lock
7048 + * the type of d_lockref.count became int, but the inlined function d_count()
7049 + * still returns unsigned int.
7050 + * I don't know why. Maybe it is for every d_count() users?
7051 + * Anyway au_dcount() lives on.
7052 + */
7053 +static inline int au_dcount(struct dentry *d)
7054 +{
7055 +       return (int)d_count(d);
7056 +}
7057 +
7058 +#endif /* __KERNEL__ */
7059 +#endif /* __AUFS_DCSUB_H__ */
7060 diff -urN /usr/share/empty/fs/aufs/debug.c linux/fs/aufs/debug.c
7061 --- /usr/share/empty/fs/aufs/debug.c    1970-01-01 01:00:00.000000000 +0100
7062 +++ linux/fs/aufs/debug.c       2017-05-06 22:16:52.778220093 +0200
7063 @@ -0,0 +1,440 @@
7064 +/*
7065 + * Copyright (C) 2005-2017 Junjiro R. Okajima
7066 + *
7067 + * This program, aufs is free software; you can redistribute it and/or modify
7068 + * it under the terms of the GNU General Public License as published by
7069 + * the Free Software Foundation; either version 2 of the License, or
7070 + * (at your option) any later version.
7071 + *
7072 + * This program is distributed in the hope that it will be useful,
7073 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7074 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7075 + * GNU General Public License for more details.
7076 + *
7077 + * You should have received a copy of the GNU General Public License
7078 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7079 + */
7080 +
7081 +/*
7082 + * debug print functions
7083 + */
7084 +
7085 +#include "aufs.h"
7086 +
7087 +/* Returns 0, or -errno.  arg is in kp->arg. */
7088 +static int param_atomic_t_set(const char *val, const struct kernel_param *kp)
7089 +{
7090 +       int err, n;
7091 +
7092 +       err = kstrtoint(val, 0, &n);
7093 +       if (!err) {
7094 +               if (n > 0)
7095 +                       au_debug_on();
7096 +               else
7097 +                       au_debug_off();
7098 +       }
7099 +       return err;
7100 +}
7101 +
7102 +/* Returns length written or -errno.  Buffer is 4k (ie. be short!) */
7103 +static int param_atomic_t_get(char *buffer, const struct kernel_param *kp)
7104 +{
7105 +       atomic_t *a;
7106 +
7107 +       a = kp->arg;
7108 +       return sprintf(buffer, "%d", atomic_read(a));
7109 +}
7110 +
7111 +static struct kernel_param_ops param_ops_atomic_t = {
7112 +       .set = param_atomic_t_set,
7113 +       .get = param_atomic_t_get
7114 +       /* void (*free)(void *arg) */
7115 +};
7116 +
7117 +atomic_t aufs_debug = ATOMIC_INIT(0);
7118 +MODULE_PARM_DESC(debug, "debug print");
7119 +module_param_named(debug, aufs_debug, atomic_t, S_IRUGO | S_IWUSR | S_IWGRP);
7120 +
7121 +DEFINE_MUTEX(au_dbg_mtx);      /* just to serialize the dbg msgs */
7122 +char *au_plevel = KERN_DEBUG;
7123 +#define dpri(fmt, ...) do {                                    \
7124 +       if ((au_plevel                                          \
7125 +            && strcmp(au_plevel, KERN_DEBUG))                  \
7126 +           || au_debug_test())                                 \
7127 +               printk("%s" fmt, au_plevel, ##__VA_ARGS__);     \
7128 +} while (0)
7129 +
7130 +/* ---------------------------------------------------------------------- */
7131 +
7132 +void au_dpri_whlist(struct au_nhash *whlist)
7133 +{
7134 +       unsigned long ul, n;
7135 +       struct hlist_head *head;
7136 +       struct au_vdir_wh *pos;
7137 +
7138 +       n = whlist->nh_num;
7139 +       head = whlist->nh_head;
7140 +       for (ul = 0; ul < n; ul++) {
7141 +               hlist_for_each_entry(pos, head, wh_hash)
7142 +                       dpri("b%d, %.*s, %d\n",
7143 +                            pos->wh_bindex,
7144 +                            pos->wh_str.len, pos->wh_str.name,
7145 +                            pos->wh_str.len);
7146 +               head++;
7147 +       }
7148 +}
7149 +
7150 +void au_dpri_vdir(struct au_vdir *vdir)
7151 +{
7152 +       unsigned long ul;
7153 +       union au_vdir_deblk_p p;
7154 +       unsigned char *o;
7155 +
7156 +       if (!vdir || IS_ERR(vdir)) {
7157 +               dpri("err %ld\n", PTR_ERR(vdir));
7158 +               return;
7159 +       }
7160 +
7161 +       dpri("deblk %u, nblk %lu, deblk %p, last{%lu, %p}, ver %lu\n",
7162 +            vdir->vd_deblk_sz, vdir->vd_nblk, vdir->vd_deblk,
7163 +            vdir->vd_last.ul, vdir->vd_last.p.deblk, vdir->vd_version);
7164 +       for (ul = 0; ul < vdir->vd_nblk; ul++) {
7165 +               p.deblk = vdir->vd_deblk[ul];
7166 +               o = p.deblk;
7167 +               dpri("[%lu]: %p\n", ul, o);
7168 +       }
7169 +}
7170 +
7171 +static int do_pri_inode(aufs_bindex_t bindex, struct inode *inode, int hn,
7172 +                       struct dentry *wh)
7173 +{
7174 +       char *n = NULL;
7175 +       int l = 0;
7176 +
7177 +       if (!inode || IS_ERR(inode)) {
7178 +               dpri("i%d: err %ld\n", bindex, PTR_ERR(inode));
7179 +               return -1;
7180 +       }
7181 +
7182 +       /* the type of i_blocks depends upon CONFIG_LBDAF */
7183 +       BUILD_BUG_ON(sizeof(inode->i_blocks) != sizeof(unsigned long)
7184 +                    && sizeof(inode->i_blocks) != sizeof(u64));
7185 +       if (wh) {
7186 +               n = (void *)wh->d_name.name;
7187 +               l = wh->d_name.len;
7188 +       }
7189 +
7190 +       dpri("i%d: %p, i%lu, %s, cnt %d, nl %u, 0%o, sz %llu, blk %llu,"
7191 +            " hn %d, ct %lld, np %lu, st 0x%lx, f 0x%x, v %llu, g %x%s%.*s\n",
7192 +            bindex, inode,
7193 +            inode->i_ino, inode->i_sb ? au_sbtype(inode->i_sb) : "??",
7194 +            atomic_read(&inode->i_count), inode->i_nlink, inode->i_mode,
7195 +            i_size_read(inode), (unsigned long long)inode->i_blocks,
7196 +            hn, (long long)timespec_to_ns(&inode->i_ctime) & 0x0ffff,
7197 +            inode->i_mapping ? inode->i_mapping->nrpages : 0,
7198 +            inode->i_state, inode->i_flags, inode->i_version,
7199 +            inode->i_generation,
7200 +            l ? ", wh " : "", l, n);
7201 +       return 0;
7202 +}
7203 +
7204 +void au_dpri_inode(struct inode *inode)
7205 +{
7206 +       struct au_iinfo *iinfo;
7207 +       struct au_hinode *hi;
7208 +       aufs_bindex_t bindex;
7209 +       int err, hn;
7210 +
7211 +       err = do_pri_inode(-1, inode, -1, NULL);
7212 +       if (err || !au_test_aufs(inode->i_sb) || au_is_bad_inode(inode))
7213 +               return;
7214 +
7215 +       iinfo = au_ii(inode);
7216 +       dpri("i-1: btop %d, bbot %d, gen %d\n",
7217 +            iinfo->ii_btop, iinfo->ii_bbot, au_iigen(inode, NULL));
7218 +       if (iinfo->ii_btop < 0)
7219 +               return;
7220 +       hn = 0;
7221 +       for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot; bindex++) {
7222 +               hi = au_hinode(iinfo, bindex);
7223 +               hn = !!au_hn(hi);
7224 +               do_pri_inode(bindex, hi->hi_inode, hn, hi->hi_whdentry);
7225 +       }
7226 +}
7227 +
7228 +void au_dpri_dalias(struct inode *inode)
7229 +{
7230 +       struct dentry *d;
7231 +
7232 +       spin_lock(&inode->i_lock);
7233 +       hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias)
7234 +               au_dpri_dentry(d);
7235 +       spin_unlock(&inode->i_lock);
7236 +}
7237 +
7238 +static int do_pri_dentry(aufs_bindex_t bindex, struct dentry *dentry)
7239 +{
7240 +       struct dentry *wh = NULL;
7241 +       int hn;
7242 +       struct inode *inode;
7243 +       struct au_iinfo *iinfo;
7244 +       struct au_hinode *hi;
7245 +
7246 +       if (!dentry || IS_ERR(dentry)) {
7247 +               dpri("d%d: err %ld\n", bindex, PTR_ERR(dentry));
7248 +               return -1;
7249 +       }
7250 +       /* do not call dget_parent() here */
7251 +       /* note: access d_xxx without d_lock */
7252 +       dpri("d%d: %p, %pd2?, %s, cnt %d, flags 0x%x, %shashed\n",
7253 +            bindex, dentry, dentry,
7254 +            dentry->d_sb ? au_sbtype(dentry->d_sb) : "??",
7255 +            au_dcount(dentry), dentry->d_flags,
7256 +            d_unhashed(dentry) ? "un" : "");
7257 +       hn = -1;
7258 +       inode = NULL;
7259 +       if (d_is_positive(dentry))
7260 +               inode = d_inode(dentry);
7261 +       if (inode
7262 +           && au_test_aufs(dentry->d_sb)
7263 +           && bindex >= 0
7264 +           && !au_is_bad_inode(inode)) {
7265 +               iinfo = au_ii(inode);
7266 +               hi = au_hinode(iinfo, bindex);
7267 +               hn = !!au_hn(hi);
7268 +               wh = hi->hi_whdentry;
7269 +       }
7270 +       do_pri_inode(bindex, inode, hn, wh);
7271 +       return 0;
7272 +}
7273 +
7274 +void au_dpri_dentry(struct dentry *dentry)
7275 +{
7276 +       struct au_dinfo *dinfo;
7277 +       aufs_bindex_t bindex;
7278 +       int err;
7279 +
7280 +       err = do_pri_dentry(-1, dentry);
7281 +       if (err || !au_test_aufs(dentry->d_sb))
7282 +               return;
7283 +
7284 +       dinfo = au_di(dentry);
7285 +       if (!dinfo)
7286 +               return;
7287 +       dpri("d-1: btop %d, bbot %d, bwh %d, bdiropq %d, gen %d, tmp %d\n",
7288 +            dinfo->di_btop, dinfo->di_bbot,
7289 +            dinfo->di_bwh, dinfo->di_bdiropq, au_digen(dentry),
7290 +            dinfo->di_tmpfile);
7291 +       if (dinfo->di_btop < 0)
7292 +               return;
7293 +       for (bindex = dinfo->di_btop; bindex <= dinfo->di_bbot; bindex++)
7294 +               do_pri_dentry(bindex, au_hdentry(dinfo, bindex)->hd_dentry);
7295 +}
7296 +
7297 +static int do_pri_file(aufs_bindex_t bindex, struct file *file)
7298 +{
7299 +       char a[32];
7300 +
7301 +       if (!file || IS_ERR(file)) {
7302 +               dpri("f%d: err %ld\n", bindex, PTR_ERR(file));
7303 +               return -1;
7304 +       }
7305 +       a[0] = 0;
7306 +       if (bindex < 0
7307 +           && !IS_ERR_OR_NULL(file->f_path.dentry)
7308 +           && au_test_aufs(file->f_path.dentry->d_sb)
7309 +           && au_fi(file))
7310 +               snprintf(a, sizeof(a), ", gen %d, mmapped %d",
7311 +                        au_figen(file), atomic_read(&au_fi(file)->fi_mmapped));
7312 +       dpri("f%d: mode 0x%x, flags 0%o, cnt %ld, v %llu, pos %llu%s\n",
7313 +            bindex, file->f_mode, file->f_flags, (long)file_count(file),
7314 +            file->f_version, file->f_pos, a);
7315 +       if (!IS_ERR_OR_NULL(file->f_path.dentry))
7316 +               do_pri_dentry(bindex, file->f_path.dentry);
7317 +       return 0;
7318 +}
7319 +
7320 +void au_dpri_file(struct file *file)
7321 +{
7322 +       struct au_finfo *finfo;
7323 +       struct au_fidir *fidir;
7324 +       struct au_hfile *hfile;
7325 +       aufs_bindex_t bindex;
7326 +       int err;
7327 +
7328 +       err = do_pri_file(-1, file);
7329 +       if (err
7330 +           || IS_ERR_OR_NULL(file->f_path.dentry)
7331 +           || !au_test_aufs(file->f_path.dentry->d_sb))
7332 +               return;
7333 +
7334 +       finfo = au_fi(file);
7335 +       if (!finfo)
7336 +               return;
7337 +       if (finfo->fi_btop < 0)
7338 +               return;
7339 +       fidir = finfo->fi_hdir;
7340 +       if (!fidir)
7341 +               do_pri_file(finfo->fi_btop, finfo->fi_htop.hf_file);
7342 +       else
7343 +               for (bindex = finfo->fi_btop;
7344 +                    bindex >= 0 && bindex <= fidir->fd_bbot;
7345 +                    bindex++) {
7346 +                       hfile = fidir->fd_hfile + bindex;
7347 +                       do_pri_file(bindex, hfile ? hfile->hf_file : NULL);
7348 +               }
7349 +}
7350 +
7351 +static int do_pri_br(aufs_bindex_t bindex, struct au_branch *br)
7352 +{
7353 +       struct vfsmount *mnt;
7354 +       struct super_block *sb;
7355 +
7356 +       if (!br || IS_ERR(br))
7357 +               goto out;
7358 +       mnt = au_br_mnt(br);
7359 +       if (!mnt || IS_ERR(mnt))
7360 +               goto out;
7361 +       sb = mnt->mnt_sb;
7362 +       if (!sb || IS_ERR(sb))
7363 +               goto out;
7364 +
7365 +       dpri("s%d: {perm 0x%x, id %d, cnt %lld, wbr %p}, "
7366 +            "%s, dev 0x%02x%02x, flags 0x%lx, cnt %d, active %d, "
7367 +            "xino %d\n",
7368 +            bindex, br->br_perm, br->br_id, au_br_count(br),
7369 +            br->br_wbr, au_sbtype(sb), MAJOR(sb->s_dev), MINOR(sb->s_dev),
7370 +            sb->s_flags, sb->s_count,
7371 +            atomic_read(&sb->s_active), !!br->br_xino.xi_file);
7372 +       return 0;
7373 +
7374 +out:
7375 +       dpri("s%d: err %ld\n", bindex, PTR_ERR(br));
7376 +       return -1;
7377 +}
7378 +
7379 +void au_dpri_sb(struct super_block *sb)
7380 +{
7381 +       struct au_sbinfo *sbinfo;
7382 +       aufs_bindex_t bindex;
7383 +       int err;
7384 +       /* to reuduce stack size */
7385 +       struct {
7386 +               struct vfsmount mnt;
7387 +               struct au_branch fake;
7388 +       } *a;
7389 +
7390 +       /* this function can be called from magic sysrq */
7391 +       a = kzalloc(sizeof(*a), GFP_ATOMIC);
7392 +       if (unlikely(!a)) {
7393 +               dpri("no memory\n");
7394 +               return;
7395 +       }
7396 +
7397 +       a->mnt.mnt_sb = sb;
7398 +       a->fake.br_path.mnt = &a->mnt;
7399 +       au_br_count_init(&a->fake);
7400 +       err = do_pri_br(-1, &a->fake);
7401 +       au_br_count_fin(&a->fake);
7402 +       au_delayed_kfree(a);
7403 +       dpri("dev 0x%x\n", sb->s_dev);
7404 +       if (err || !au_test_aufs(sb))
7405 +               return;
7406 +
7407 +       sbinfo = au_sbi(sb);
7408 +       if (!sbinfo)
7409 +               return;
7410 +       dpri("nw %d, gen %u, kobj %d\n",
7411 +            atomic_read(&sbinfo->si_nowait.nw_len), sbinfo->si_generation,
7412 +            kref_read(&sbinfo->si_kobj.kref));
7413 +       for (bindex = 0; bindex <= sbinfo->si_bbot; bindex++)
7414 +               do_pri_br(bindex, sbinfo->si_branch[0 + bindex]);
7415 +}
7416 +
7417 +/* ---------------------------------------------------------------------- */
7418 +
7419 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line)
7420 +{
7421 +       struct inode *h_inode, *inode = d_inode(dentry);
7422 +       struct dentry *h_dentry;
7423 +       aufs_bindex_t bindex, bbot, bi;
7424 +
7425 +       if (!inode /* || au_di(dentry)->di_lsc == AuLsc_DI_TMP */)
7426 +               return;
7427 +
7428 +       bbot = au_dbbot(dentry);
7429 +       bi = au_ibbot(inode);
7430 +       if (bi < bbot)
7431 +               bbot = bi;
7432 +       bindex = au_dbtop(dentry);
7433 +       bi = au_ibtop(inode);
7434 +       if (bi > bindex)
7435 +               bindex = bi;
7436 +
7437 +       for (; bindex <= bbot; bindex++) {
7438 +               h_dentry = au_h_dptr(dentry, bindex);
7439 +               if (!h_dentry)
7440 +                       continue;
7441 +               h_inode = au_h_iptr(inode, bindex);
7442 +               if (unlikely(h_inode != d_inode(h_dentry))) {
7443 +                       au_debug_on();
7444 +                       AuDbg("b%d, %s:%d\n", bindex, func, line);
7445 +                       AuDbgDentry(dentry);
7446 +                       AuDbgInode(inode);
7447 +                       au_debug_off();
7448 +                       BUG();
7449 +               }
7450 +       }
7451 +}
7452 +
7453 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen)
7454 +{
7455 +       int err, i, j;
7456 +       struct au_dcsub_pages dpages;
7457 +       struct au_dpage *dpage;
7458 +       struct dentry **dentries;
7459 +
7460 +       err = au_dpages_init(&dpages, GFP_NOFS);
7461 +       AuDebugOn(err);
7462 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/1);
7463 +       AuDebugOn(err);
7464 +       for (i = dpages.ndpage - 1; !err && i >= 0; i--) {
7465 +               dpage = dpages.dpages + i;
7466 +               dentries = dpage->dentries;
7467 +               for (j = dpage->ndentry - 1; !err && j >= 0; j--)
7468 +                       AuDebugOn(au_digen_test(dentries[j], sigen));
7469 +       }
7470 +       au_dpages_free(&dpages);
7471 +}
7472 +
7473 +void au_dbg_verify_kthread(void)
7474 +{
7475 +       if (au_wkq_test()) {
7476 +               au_dbg_blocked();
7477 +               /*
7478 +                * It may be recursive, but udba=notify between two aufs mounts,
7479 +                * where a single ro branch is shared, is not a problem.
7480 +                */
7481 +               /* WARN_ON(1); */
7482 +       }
7483 +}
7484 +
7485 +/* ---------------------------------------------------------------------- */
7486 +
7487 +int __init au_debug_init(void)
7488 +{
7489 +       aufs_bindex_t bindex;
7490 +       struct au_vdir_destr destr;
7491 +
7492 +       bindex = -1;
7493 +       AuDebugOn(bindex >= 0);
7494 +
7495 +       destr.len = -1;
7496 +       AuDebugOn(destr.len < NAME_MAX);
7497 +
7498 +#ifdef CONFIG_4KSTACKS
7499 +       pr_warn("CONFIG_4KSTACKS is defined.\n");
7500 +#endif
7501 +
7502 +       return 0;
7503 +}
7504 diff -urN /usr/share/empty/fs/aufs/debug.h linux/fs/aufs/debug.h
7505 --- /usr/share/empty/fs/aufs/debug.h    1970-01-01 01:00:00.000000000 +0100
7506 +++ linux/fs/aufs/debug.h       2017-05-06 22:16:52.778220093 +0200
7507 @@ -0,0 +1,225 @@
7508 +/*
7509 + * Copyright (C) 2005-2017 Junjiro R. Okajima
7510 + *
7511 + * This program, aufs is free software; you can redistribute it and/or modify
7512 + * it under the terms of the GNU General Public License as published by
7513 + * the Free Software Foundation; either version 2 of the License, or
7514 + * (at your option) any later version.
7515 + *
7516 + * This program is distributed in the hope that it will be useful,
7517 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7518 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7519 + * GNU General Public License for more details.
7520 + *
7521 + * You should have received a copy of the GNU General Public License
7522 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7523 + */
7524 +
7525 +/*
7526 + * debug print functions
7527 + */
7528 +
7529 +#ifndef __AUFS_DEBUG_H__
7530 +#define __AUFS_DEBUG_H__
7531 +
7532 +#ifdef __KERNEL__
7533 +
7534 +#include <linux/atomic.h>
7535 +#include <linux/module.h>
7536 +#include <linux/kallsyms.h>
7537 +#include <linux/sysrq.h>
7538 +
7539 +#ifdef CONFIG_AUFS_DEBUG
7540 +#define AuDebugOn(a)           BUG_ON(a)
7541 +
7542 +/* module parameter */
7543 +extern atomic_t aufs_debug;
7544 +static inline void au_debug_on(void)
7545 +{
7546 +       atomic_inc(&aufs_debug);
7547 +}
7548 +static inline void au_debug_off(void)
7549 +{
7550 +       atomic_dec_if_positive(&aufs_debug);
7551 +}
7552 +
7553 +static inline int au_debug_test(void)
7554 +{
7555 +       return atomic_read(&aufs_debug) > 0;
7556 +}
7557 +#else
7558 +#define AuDebugOn(a)           do {} while (0)
7559 +AuStubVoid(au_debug_on, void)
7560 +AuStubVoid(au_debug_off, void)
7561 +AuStubInt0(au_debug_test, void)
7562 +#endif /* CONFIG_AUFS_DEBUG */
7563 +
7564 +#define param_check_atomic_t(name, p) __param_check(name, p, atomic_t)
7565 +
7566 +/* ---------------------------------------------------------------------- */
7567 +
7568 +/* debug print */
7569 +
7570 +#define AuDbg(fmt, ...) do { \
7571 +       if (au_debug_test()) \
7572 +               pr_debug("DEBUG: " fmt, ##__VA_ARGS__); \
7573 +} while (0)
7574 +#define AuLabel(l)             AuDbg(#l "\n")
7575 +#define AuIOErr(fmt, ...)      pr_err("I/O Error, " fmt, ##__VA_ARGS__)
7576 +#define AuWarn1(fmt, ...) do { \
7577 +       static unsigned char _c; \
7578 +       if (!_c++) \
7579 +               pr_warn(fmt, ##__VA_ARGS__); \
7580 +} while (0)
7581 +
7582 +#define AuErr1(fmt, ...) do { \
7583 +       static unsigned char _c; \
7584 +       if (!_c++) \
7585 +               pr_err(fmt, ##__VA_ARGS__); \
7586 +} while (0)
7587 +
7588 +#define AuIOErr1(fmt, ...) do { \
7589 +       static unsigned char _c; \
7590 +       if (!_c++) \
7591 +               AuIOErr(fmt, ##__VA_ARGS__); \
7592 +} while (0)
7593 +
7594 +#define AuUnsupportMsg "This operation is not supported." \
7595 +                       " Please report this application to aufs-users ML."
7596 +#define AuUnsupport(fmt, ...) do { \
7597 +       pr_err(AuUnsupportMsg "\n" fmt, ##__VA_ARGS__); \
7598 +       dump_stack(); \
7599 +} while (0)
7600 +
7601 +#define AuTraceErr(e) do { \
7602 +       if (unlikely((e) < 0)) \
7603 +               AuDbg("err %d\n", (int)(e)); \
7604 +} while (0)
7605 +
7606 +#define AuTraceErrPtr(p) do { \
7607 +       if (IS_ERR(p)) \
7608 +               AuDbg("err %ld\n", PTR_ERR(p)); \
7609 +} while (0)
7610 +
7611 +/* dirty macros for debug print, use with "%.*s" and caution */
7612 +#define AuLNPair(qstr)         (qstr)->len, (qstr)->name
7613 +
7614 +/* ---------------------------------------------------------------------- */
7615 +
7616 +struct dentry;
7617 +#ifdef CONFIG_AUFS_DEBUG
7618 +extern struct mutex au_dbg_mtx;
7619 +extern char *au_plevel;
7620 +struct au_nhash;
7621 +void au_dpri_whlist(struct au_nhash *whlist);
7622 +struct au_vdir;
7623 +void au_dpri_vdir(struct au_vdir *vdir);
7624 +struct inode;
7625 +void au_dpri_inode(struct inode *inode);
7626 +void au_dpri_dalias(struct inode *inode);
7627 +void au_dpri_dentry(struct dentry *dentry);
7628 +struct file;
7629 +void au_dpri_file(struct file *filp);
7630 +struct super_block;
7631 +void au_dpri_sb(struct super_block *sb);
7632 +
7633 +#define au_dbg_verify_dinode(d) __au_dbg_verify_dinode(d, __func__, __LINE__)
7634 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
7635 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen);
7636 +void au_dbg_verify_kthread(void);
7637 +
7638 +int __init au_debug_init(void);
7639 +
7640 +#define AuDbgWhlist(w) do { \
7641 +       mutex_lock(&au_dbg_mtx); \
7642 +       AuDbg(#w "\n"); \
7643 +       au_dpri_whlist(w); \
7644 +       mutex_unlock(&au_dbg_mtx); \
7645 +} while (0)
7646 +
7647 +#define AuDbgVdir(v) do { \
7648 +       mutex_lock(&au_dbg_mtx); \
7649 +       AuDbg(#v "\n"); \
7650 +       au_dpri_vdir(v); \
7651 +       mutex_unlock(&au_dbg_mtx); \
7652 +} while (0)
7653 +
7654 +#define AuDbgInode(i) do { \
7655 +       mutex_lock(&au_dbg_mtx); \
7656 +       AuDbg(#i "\n"); \
7657 +       au_dpri_inode(i); \
7658 +       mutex_unlock(&au_dbg_mtx); \
7659 +} while (0)
7660 +
7661 +#define AuDbgDAlias(i) do { \
7662 +       mutex_lock(&au_dbg_mtx); \
7663 +       AuDbg(#i "\n"); \
7664 +       au_dpri_dalias(i); \
7665 +       mutex_unlock(&au_dbg_mtx); \
7666 +} while (0)
7667 +
7668 +#define AuDbgDentry(d) do { \
7669 +       mutex_lock(&au_dbg_mtx); \
7670 +       AuDbg(#d "\n"); \
7671 +       au_dpri_dentry(d); \
7672 +       mutex_unlock(&au_dbg_mtx); \
7673 +} while (0)
7674 +
7675 +#define AuDbgFile(f) do { \
7676 +       mutex_lock(&au_dbg_mtx); \
7677 +       AuDbg(#f "\n"); \
7678 +       au_dpri_file(f); \
7679 +       mutex_unlock(&au_dbg_mtx); \
7680 +} while (0)
7681 +
7682 +#define AuDbgSb(sb) do { \
7683 +       mutex_lock(&au_dbg_mtx); \
7684 +       AuDbg(#sb "\n"); \
7685 +       au_dpri_sb(sb); \
7686 +       mutex_unlock(&au_dbg_mtx); \
7687 +} while (0)
7688 +
7689 +#define AuDbgSym(addr) do {                            \
7690 +       char sym[KSYM_SYMBOL_LEN];                      \
7691 +       sprint_symbol(sym, (unsigned long)addr);        \
7692 +       AuDbg("%s\n", sym);                             \
7693 +} while (0)
7694 +#else
7695 +AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
7696 +AuStubVoid(au_dbg_verify_gen, struct dentry *parent, unsigned int sigen)
7697 +AuStubVoid(au_dbg_verify_kthread, void)
7698 +AuStubInt0(__init au_debug_init, void)
7699 +
7700 +#define AuDbgWhlist(w)         do {} while (0)
7701 +#define AuDbgVdir(v)           do {} while (0)
7702 +#define AuDbgInode(i)          do {} while (0)
7703 +#define AuDbgDAlias(i)         do {} while (0)
7704 +#define AuDbgDentry(d)         do {} while (0)
7705 +#define AuDbgFile(f)           do {} while (0)
7706 +#define AuDbgSb(sb)            do {} while (0)
7707 +#define AuDbgSym(addr)         do {} while (0)
7708 +#endif /* CONFIG_AUFS_DEBUG */
7709 +
7710 +/* ---------------------------------------------------------------------- */
7711 +
7712 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
7713 +int __init au_sysrq_init(void);
7714 +void au_sysrq_fin(void);
7715 +
7716 +#ifdef CONFIG_HW_CONSOLE
7717 +#define au_dbg_blocked() do { \
7718 +       WARN_ON(1); \
7719 +       handle_sysrq('w'); \
7720 +} while (0)
7721 +#else
7722 +AuStubVoid(au_dbg_blocked, void)
7723 +#endif
7724 +
7725 +#else
7726 +AuStubInt0(__init au_sysrq_init, void)
7727 +AuStubVoid(au_sysrq_fin, void)
7728 +AuStubVoid(au_dbg_blocked, void)
7729 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
7730 +
7731 +#endif /* __KERNEL__ */
7732 +#endif /* __AUFS_DEBUG_H__ */
7733 diff -urN /usr/share/empty/fs/aufs/dentry.c linux/fs/aufs/dentry.c
7734 --- /usr/share/empty/fs/aufs/dentry.c   1970-01-01 01:00:00.000000000 +0100
7735 +++ linux/fs/aufs/dentry.c      2017-05-06 22:16:52.778220093 +0200
7736 @@ -0,0 +1,1130 @@
7737 +/*
7738 + * Copyright (C) 2005-2017 Junjiro R. Okajima
7739 + *
7740 + * This program, aufs is free software; you can redistribute it and/or modify
7741 + * it under the terms of the GNU General Public License as published by
7742 + * the Free Software Foundation; either version 2 of the License, or
7743 + * (at your option) any later version.
7744 + *
7745 + * This program is distributed in the hope that it will be useful,
7746 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7747 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7748 + * GNU General Public License for more details.
7749 + *
7750 + * You should have received a copy of the GNU General Public License
7751 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7752 + */
7753 +
7754 +/*
7755 + * lookup and dentry operations
7756 + */
7757 +
7758 +#include <linux/namei.h>
7759 +#include "aufs.h"
7760 +
7761 +struct au_do_lookup_args {
7762 +       unsigned int            flags;
7763 +       mode_t                  type;
7764 +};
7765 +
7766 +/*
7767 + * returns positive/negative dentry, NULL or an error.
7768 + * NULL means whiteout-ed or not-found.
7769 + */
7770 +static struct dentry*
7771 +au_do_lookup(struct dentry *h_parent, struct dentry *dentry,
7772 +            aufs_bindex_t bindex, struct qstr *wh_name,
7773 +            struct au_do_lookup_args *args)
7774 +{
7775 +       struct dentry *h_dentry;
7776 +       struct inode *h_inode;
7777 +       struct au_branch *br;
7778 +       int wh_found, opq;
7779 +       unsigned char wh_able;
7780 +       const unsigned char allow_neg = !!au_ftest_lkup(args->flags, ALLOW_NEG);
7781 +       const unsigned char ignore_perm = !!au_ftest_lkup(args->flags,
7782 +                                                         IGNORE_PERM);
7783 +
7784 +       wh_found = 0;
7785 +       br = au_sbr(dentry->d_sb, bindex);
7786 +       wh_able = !!au_br_whable(br->br_perm);
7787 +       if (wh_able)
7788 +               wh_found = au_wh_test(h_parent, wh_name, ignore_perm);
7789 +       h_dentry = ERR_PTR(wh_found);
7790 +       if (!wh_found)
7791 +               goto real_lookup;
7792 +       if (unlikely(wh_found < 0))
7793 +               goto out;
7794 +
7795 +       /* We found a whiteout */
7796 +       /* au_set_dbbot(dentry, bindex); */
7797 +       au_set_dbwh(dentry, bindex);
7798 +       if (!allow_neg)
7799 +               return NULL; /* success */
7800 +
7801 +real_lookup:
7802 +       if (!ignore_perm)
7803 +               h_dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
7804 +       else
7805 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7806 +       if (IS_ERR(h_dentry)) {
7807 +               if (PTR_ERR(h_dentry) == -ENAMETOOLONG
7808 +                   && !allow_neg)
7809 +                       h_dentry = NULL;
7810 +               goto out;
7811 +       }
7812 +
7813 +       h_inode = d_inode(h_dentry);
7814 +       if (d_is_negative(h_dentry)) {
7815 +               if (!allow_neg)
7816 +                       goto out_neg;
7817 +       } else if (wh_found
7818 +                  || (args->type && args->type != (h_inode->i_mode & S_IFMT)))
7819 +               goto out_neg;
7820 +
7821 +       if (au_dbbot(dentry) <= bindex)
7822 +               au_set_dbbot(dentry, bindex);
7823 +       if (au_dbtop(dentry) < 0 || bindex < au_dbtop(dentry))
7824 +               au_set_dbtop(dentry, bindex);
7825 +       au_set_h_dptr(dentry, bindex, h_dentry);
7826 +
7827 +       if (!d_is_dir(h_dentry)
7828 +           || !wh_able
7829 +           || (d_really_is_positive(dentry) && !d_is_dir(dentry)))
7830 +               goto out; /* success */
7831 +
7832 +       inode_lock_nested(h_inode, AuLsc_I_CHILD);
7833 +       opq = au_diropq_test(h_dentry);
7834 +       inode_unlock(h_inode);
7835 +       if (opq > 0)
7836 +               au_set_dbdiropq(dentry, bindex);
7837 +       else if (unlikely(opq < 0)) {
7838 +               au_set_h_dptr(dentry, bindex, NULL);
7839 +               h_dentry = ERR_PTR(opq);
7840 +       }
7841 +       goto out;
7842 +
7843 +out_neg:
7844 +       dput(h_dentry);
7845 +       h_dentry = NULL;
7846 +out:
7847 +       return h_dentry;
7848 +}
7849 +
7850 +static int au_test_shwh(struct super_block *sb, const struct qstr *name)
7851 +{
7852 +       if (unlikely(!au_opt_test(au_mntflags(sb), SHWH)
7853 +                    && !strncmp(name->name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)))
7854 +               return -EPERM;
7855 +       return 0;
7856 +}
7857 +
7858 +/*
7859 + * returns the number of lower positive dentries,
7860 + * otherwise an error.
7861 + * can be called at unlinking with @type is zero.
7862 + */
7863 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t btop,
7864 +                  unsigned int flags)
7865 +{
7866 +       int npositive, err;
7867 +       aufs_bindex_t bindex, btail, bdiropq;
7868 +       unsigned char isdir, dirperm1;
7869 +       struct qstr whname;
7870 +       struct au_do_lookup_args args = {
7871 +               .flags          = flags
7872 +       };
7873 +       const struct qstr *name = &dentry->d_name;
7874 +       struct dentry *parent;
7875 +       struct super_block *sb;
7876 +
7877 +       sb = dentry->d_sb;
7878 +       err = au_test_shwh(sb, name);
7879 +       if (unlikely(err))
7880 +               goto out;
7881 +
7882 +       err = au_wh_name_alloc(&whname, name);
7883 +       if (unlikely(err))
7884 +               goto out;
7885 +
7886 +       isdir = !!d_is_dir(dentry);
7887 +       dirperm1 = !!au_opt_test(au_mntflags(sb), DIRPERM1);
7888 +
7889 +       npositive = 0;
7890 +       parent = dget_parent(dentry);
7891 +       btail = au_dbtaildir(parent);
7892 +       for (bindex = btop; bindex <= btail; bindex++) {
7893 +               struct dentry *h_parent, *h_dentry;
7894 +               struct inode *h_inode, *h_dir;
7895 +
7896 +               h_dentry = au_h_dptr(dentry, bindex);
7897 +               if (h_dentry) {
7898 +                       if (d_is_positive(h_dentry))
7899 +                               npositive++;
7900 +                       break;
7901 +               }
7902 +               h_parent = au_h_dptr(parent, bindex);
7903 +               if (!h_parent || !d_is_dir(h_parent))
7904 +                       continue;
7905 +
7906 +               h_dir = d_inode(h_parent);
7907 +               inode_lock_nested(h_dir, AuLsc_I_PARENT);
7908 +               h_dentry = au_do_lookup(h_parent, dentry, bindex, &whname,
7909 +                                       &args);
7910 +               inode_unlock(h_dir);
7911 +               err = PTR_ERR(h_dentry);
7912 +               if (IS_ERR(h_dentry))
7913 +                       goto out_parent;
7914 +               if (h_dentry)
7915 +                       au_fclr_lkup(args.flags, ALLOW_NEG);
7916 +               if (dirperm1)
7917 +                       au_fset_lkup(args.flags, IGNORE_PERM);
7918 +
7919 +               if (au_dbwh(dentry) == bindex)
7920 +                       break;
7921 +               if (!h_dentry)
7922 +                       continue;
7923 +               if (d_is_negative(h_dentry))
7924 +                       continue;
7925 +               h_inode = d_inode(h_dentry);
7926 +               npositive++;
7927 +               if (!args.type)
7928 +                       args.type = h_inode->i_mode & S_IFMT;
7929 +               if (args.type != S_IFDIR)
7930 +                       break;
7931 +               else if (isdir) {
7932 +                       /* the type of lower may be different */
7933 +                       bdiropq = au_dbdiropq(dentry);
7934 +                       if (bdiropq >= 0 && bdiropq <= bindex)
7935 +                               break;
7936 +               }
7937 +       }
7938 +
7939 +       if (npositive) {
7940 +               AuLabel(positive);
7941 +               au_update_dbtop(dentry);
7942 +       }
7943 +       err = npositive;
7944 +       if (unlikely(!au_opt_test(au_mntflags(sb), UDBA_NONE)
7945 +                    && au_dbtop(dentry) < 0)) {
7946 +               err = -EIO;
7947 +               AuIOErr("both of real entry and whiteout found, %pd, err %d\n",
7948 +                       dentry, err);
7949 +       }
7950 +
7951 +out_parent:
7952 +       dput(parent);
7953 +       au_delayed_kfree(whname.name);
7954 +out:
7955 +       return err;
7956 +}
7957 +
7958 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent)
7959 +{
7960 +       struct dentry *dentry;
7961 +       int wkq_err;
7962 +
7963 +       if (!au_test_h_perm_sio(d_inode(parent), MAY_EXEC))
7964 +               dentry = vfsub_lkup_one(name, parent);
7965 +       else {
7966 +               struct vfsub_lkup_one_args args = {
7967 +                       .errp   = &dentry,
7968 +                       .name   = name,
7969 +                       .parent = parent
7970 +               };
7971 +
7972 +               wkq_err = au_wkq_wait(vfsub_call_lkup_one, &args);
7973 +               if (unlikely(wkq_err))
7974 +                       dentry = ERR_PTR(wkq_err);
7975 +       }
7976 +
7977 +       return dentry;
7978 +}
7979 +
7980 +/*
7981 + * lookup @dentry on @bindex which should be negative.
7982 + */
7983 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh)
7984 +{
7985 +       int err;
7986 +       struct dentry *parent, *h_parent, *h_dentry;
7987 +       struct au_branch *br;
7988 +
7989 +       parent = dget_parent(dentry);
7990 +       h_parent = au_h_dptr(parent, bindex);
7991 +       br = au_sbr(dentry->d_sb, bindex);
7992 +       if (wh)
7993 +               h_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
7994 +       else
7995 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7996 +       err = PTR_ERR(h_dentry);
7997 +       if (IS_ERR(h_dentry))
7998 +               goto out;
7999 +       if (unlikely(d_is_positive(h_dentry))) {
8000 +               err = -EIO;
8001 +               AuIOErr("%pd should be negative on b%d.\n", h_dentry, bindex);
8002 +               dput(h_dentry);
8003 +               goto out;
8004 +       }
8005 +
8006 +       err = 0;
8007 +       if (bindex < au_dbtop(dentry))
8008 +               au_set_dbtop(dentry, bindex);
8009 +       if (au_dbbot(dentry) < bindex)
8010 +               au_set_dbbot(dentry, bindex);
8011 +       au_set_h_dptr(dentry, bindex, h_dentry);
8012 +
8013 +out:
8014 +       dput(parent);
8015 +       return err;
8016 +}
8017 +
8018 +/* ---------------------------------------------------------------------- */
8019 +
8020 +/* subset of struct inode */
8021 +struct au_iattr {
8022 +       unsigned long           i_ino;
8023 +       /* unsigned int         i_nlink; */
8024 +       kuid_t                  i_uid;
8025 +       kgid_t                  i_gid;
8026 +       u64                     i_version;
8027 +/*
8028 +       loff_t                  i_size;
8029 +       blkcnt_t                i_blocks;
8030 +*/
8031 +       umode_t                 i_mode;
8032 +};
8033 +
8034 +static void au_iattr_save(struct au_iattr *ia, struct inode *h_inode)
8035 +{
8036 +       ia->i_ino = h_inode->i_ino;
8037 +       /* ia->i_nlink = h_inode->i_nlink; */
8038 +       ia->i_uid = h_inode->i_uid;
8039 +       ia->i_gid = h_inode->i_gid;
8040 +       ia->i_version = h_inode->i_version;
8041 +/*
8042 +       ia->i_size = h_inode->i_size;
8043 +       ia->i_blocks = h_inode->i_blocks;
8044 +*/
8045 +       ia->i_mode = (h_inode->i_mode & S_IFMT);
8046 +}
8047 +
8048 +static int au_iattr_test(struct au_iattr *ia, struct inode *h_inode)
8049 +{
8050 +       return ia->i_ino != h_inode->i_ino
8051 +               /* || ia->i_nlink != h_inode->i_nlink */
8052 +               || !uid_eq(ia->i_uid, h_inode->i_uid)
8053 +               || !gid_eq(ia->i_gid, h_inode->i_gid)
8054 +               || ia->i_version != h_inode->i_version
8055 +/*
8056 +               || ia->i_size != h_inode->i_size
8057 +               || ia->i_blocks != h_inode->i_blocks
8058 +*/
8059 +               || ia->i_mode != (h_inode->i_mode & S_IFMT);
8060 +}
8061 +
8062 +static int au_h_verify_dentry(struct dentry *h_dentry, struct dentry *h_parent,
8063 +                             struct au_branch *br)
8064 +{
8065 +       int err;
8066 +       struct au_iattr ia;
8067 +       struct inode *h_inode;
8068 +       struct dentry *h_d;
8069 +       struct super_block *h_sb;
8070 +
8071 +       err = 0;
8072 +       memset(&ia, -1, sizeof(ia));
8073 +       h_sb = h_dentry->d_sb;
8074 +       h_inode = NULL;
8075 +       if (d_is_positive(h_dentry)) {
8076 +               h_inode = d_inode(h_dentry);
8077 +               au_iattr_save(&ia, h_inode);
8078 +       } else if (au_test_nfs(h_sb) || au_test_fuse(h_sb))
8079 +               /* nfs d_revalidate may return 0 for negative dentry */
8080 +               /* fuse d_revalidate always return 0 for negative dentry */
8081 +               goto out;
8082 +
8083 +       /* main purpose is namei.c:cached_lookup() and d_revalidate */
8084 +       h_d = vfsub_lkup_one(&h_dentry->d_name, h_parent);
8085 +       err = PTR_ERR(h_d);
8086 +       if (IS_ERR(h_d))
8087 +               goto out;
8088 +
8089 +       err = 0;
8090 +       if (unlikely(h_d != h_dentry
8091 +                    || d_inode(h_d) != h_inode
8092 +                    || (h_inode && au_iattr_test(&ia, h_inode))))
8093 +               err = au_busy_or_stale();
8094 +       dput(h_d);
8095 +
8096 +out:
8097 +       AuTraceErr(err);
8098 +       return err;
8099 +}
8100 +
8101 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
8102 +               struct dentry *h_parent, struct au_branch *br)
8103 +{
8104 +       int err;
8105 +
8106 +       err = 0;
8107 +       if (udba == AuOpt_UDBA_REVAL
8108 +           && !au_test_fs_remote(h_dentry->d_sb)) {
8109 +               IMustLock(h_dir);
8110 +               err = (d_inode(h_dentry->d_parent) != h_dir);
8111 +       } else if (udba != AuOpt_UDBA_NONE)
8112 +               err = au_h_verify_dentry(h_dentry, h_parent, br);
8113 +
8114 +       return err;
8115 +}
8116 +
8117 +/* ---------------------------------------------------------------------- */
8118 +
8119 +static int au_do_refresh_hdentry(struct dentry *dentry, struct dentry *parent)
8120 +{
8121 +       int err;
8122 +       aufs_bindex_t new_bindex, bindex, bbot, bwh, bdiropq;
8123 +       struct au_hdentry tmp, *p, *q;
8124 +       struct au_dinfo *dinfo;
8125 +       struct super_block *sb;
8126 +
8127 +       DiMustWriteLock(dentry);
8128 +
8129 +       sb = dentry->d_sb;
8130 +       dinfo = au_di(dentry);
8131 +       bbot = dinfo->di_bbot;
8132 +       bwh = dinfo->di_bwh;
8133 +       bdiropq = dinfo->di_bdiropq;
8134 +       bindex = dinfo->di_btop;
8135 +       p = au_hdentry(dinfo, bindex);
8136 +       for (; bindex <= bbot; bindex++, p++) {
8137 +               if (!p->hd_dentry)
8138 +                       continue;
8139 +
8140 +               new_bindex = au_br_index(sb, p->hd_id);
8141 +               if (new_bindex == bindex)
8142 +                       continue;
8143 +
8144 +               if (dinfo->di_bwh == bindex)
8145 +                       bwh = new_bindex;
8146 +               if (dinfo->di_bdiropq == bindex)
8147 +                       bdiropq = new_bindex;
8148 +               if (new_bindex < 0) {
8149 +                       au_hdput(p);
8150 +                       p->hd_dentry = NULL;
8151 +                       continue;
8152 +               }
8153 +
8154 +               /* swap two lower dentries, and loop again */
8155 +               q = au_hdentry(dinfo, new_bindex);
8156 +               tmp = *q;
8157 +               *q = *p;
8158 +               *p = tmp;
8159 +               if (tmp.hd_dentry) {
8160 +                       bindex--;
8161 +                       p--;
8162 +               }
8163 +       }
8164 +
8165 +       dinfo->di_bwh = -1;
8166 +       if (bwh >= 0 && bwh <= au_sbbot(sb) && au_sbr_whable(sb, bwh))
8167 +               dinfo->di_bwh = bwh;
8168 +
8169 +       dinfo->di_bdiropq = -1;
8170 +       if (bdiropq >= 0
8171 +           && bdiropq <= au_sbbot(sb)
8172 +           && au_sbr_whable(sb, bdiropq))
8173 +               dinfo->di_bdiropq = bdiropq;
8174 +
8175 +       err = -EIO;
8176 +       dinfo->di_btop = -1;
8177 +       dinfo->di_bbot = -1;
8178 +       bbot = au_dbbot(parent);
8179 +       bindex = 0;
8180 +       p = au_hdentry(dinfo, bindex);
8181 +       for (; bindex <= bbot; bindex++, p++)
8182 +               if (p->hd_dentry) {
8183 +                       dinfo->di_btop = bindex;
8184 +                       break;
8185 +               }
8186 +
8187 +       if (dinfo->di_btop >= 0) {
8188 +               bindex = bbot;
8189 +               p = au_hdentry(dinfo, bindex);
8190 +               for (; bindex >= 0; bindex--, p--)
8191 +                       if (p->hd_dentry) {
8192 +                               dinfo->di_bbot = bindex;
8193 +                               err = 0;
8194 +                               break;
8195 +                       }
8196 +       }
8197 +
8198 +       return err;
8199 +}
8200 +
8201 +static void au_do_hide(struct dentry *dentry)
8202 +{
8203 +       struct inode *inode;
8204 +
8205 +       if (d_really_is_positive(dentry)) {
8206 +               inode = d_inode(dentry);
8207 +               if (!d_is_dir(dentry)) {
8208 +                       if (inode->i_nlink && !d_unhashed(dentry))
8209 +                               drop_nlink(inode);
8210 +               } else {
8211 +                       clear_nlink(inode);
8212 +                       /* stop next lookup */
8213 +                       inode->i_flags |= S_DEAD;
8214 +               }
8215 +               smp_mb(); /* necessary? */
8216 +       }
8217 +       d_drop(dentry);
8218 +}
8219 +
8220 +static int au_hide_children(struct dentry *parent)
8221 +{
8222 +       int err, i, j, ndentry;
8223 +       struct au_dcsub_pages dpages;
8224 +       struct au_dpage *dpage;
8225 +       struct dentry *dentry;
8226 +
8227 +       err = au_dpages_init(&dpages, GFP_NOFS);
8228 +       if (unlikely(err))
8229 +               goto out;
8230 +       err = au_dcsub_pages(&dpages, parent, NULL, NULL);
8231 +       if (unlikely(err))
8232 +               goto out_dpages;
8233 +
8234 +       /* in reverse order */
8235 +       for (i = dpages.ndpage - 1; i >= 0; i--) {
8236 +               dpage = dpages.dpages + i;
8237 +               ndentry = dpage->ndentry;
8238 +               for (j = ndentry - 1; j >= 0; j--) {
8239 +                       dentry = dpage->dentries[j];
8240 +                       if (dentry != parent)
8241 +                               au_do_hide(dentry);
8242 +               }
8243 +       }
8244 +
8245 +out_dpages:
8246 +       au_dpages_free(&dpages);
8247 +out:
8248 +       return err;
8249 +}
8250 +
8251 +static void au_hide(struct dentry *dentry)
8252 +{
8253 +       int err;
8254 +
8255 +       AuDbgDentry(dentry);
8256 +       if (d_is_dir(dentry)) {
8257 +               /* shrink_dcache_parent(dentry); */
8258 +               err = au_hide_children(dentry);
8259 +               if (unlikely(err))
8260 +                       AuIOErr("%pd, failed hiding children, ignored %d\n",
8261 +                               dentry, err);
8262 +       }
8263 +       au_do_hide(dentry);
8264 +}
8265 +
8266 +/*
8267 + * By adding a dirty branch, a cached dentry may be affected in various ways.
8268 + *
8269 + * a dirty branch is added
8270 + * - on the top of layers
8271 + * - in the middle of layers
8272 + * - to the bottom of layers
8273 + *
8274 + * on the added branch there exists
8275 + * - a whiteout
8276 + * - a diropq
8277 + * - a same named entry
8278 + *   + exist
8279 + *     * negative --> positive
8280 + *     * positive --> positive
8281 + *      - type is unchanged
8282 + *      - type is changed
8283 + *   + doesn't exist
8284 + *     * negative --> negative
8285 + *     * positive --> negative (rejected by au_br_del() for non-dir case)
8286 + * - none
8287 + */
8288 +static int au_refresh_by_dinfo(struct dentry *dentry, struct au_dinfo *dinfo,
8289 +                              struct au_dinfo *tmp)
8290 +{
8291 +       int err;
8292 +       aufs_bindex_t bindex, bbot;
8293 +       struct {
8294 +               struct dentry *dentry;
8295 +               struct inode *inode;
8296 +               mode_t mode;
8297 +       } orig_h, tmp_h = {
8298 +               .dentry = NULL
8299 +       };
8300 +       struct au_hdentry *hd;
8301 +       struct inode *inode, *h_inode;
8302 +       struct dentry *h_dentry;
8303 +
8304 +       err = 0;
8305 +       AuDebugOn(dinfo->di_btop < 0);
8306 +       orig_h.mode = 0;
8307 +       orig_h.dentry = au_hdentry(dinfo, dinfo->di_btop)->hd_dentry;
8308 +       orig_h.inode = NULL;
8309 +       if (d_is_positive(orig_h.dentry)) {
8310 +               orig_h.inode = d_inode(orig_h.dentry);
8311 +               orig_h.mode = orig_h.inode->i_mode & S_IFMT;
8312 +       }
8313 +       if (tmp->di_btop >= 0) {
8314 +               tmp_h.dentry = au_hdentry(tmp, tmp->di_btop)->hd_dentry;
8315 +               if (d_is_positive(tmp_h.dentry)) {
8316 +                       tmp_h.inode = d_inode(tmp_h.dentry);
8317 +                       tmp_h.mode = tmp_h.inode->i_mode & S_IFMT;
8318 +               }
8319 +       }
8320 +
8321 +       inode = NULL;
8322 +       if (d_really_is_positive(dentry))
8323 +               inode = d_inode(dentry);
8324 +       if (!orig_h.inode) {
8325 +               AuDbg("nagative originally\n");
8326 +               if (inode) {
8327 +                       au_hide(dentry);
8328 +                       goto out;
8329 +               }
8330 +               AuDebugOn(inode);
8331 +               AuDebugOn(dinfo->di_btop != dinfo->di_bbot);
8332 +               AuDebugOn(dinfo->di_bdiropq != -1);
8333 +
8334 +               if (!tmp_h.inode) {
8335 +                       AuDbg("negative --> negative\n");
8336 +                       /* should have only one negative lower */
8337 +                       if (tmp->di_btop >= 0
8338 +                           && tmp->di_btop < dinfo->di_btop) {
8339 +                               AuDebugOn(tmp->di_btop != tmp->di_bbot);
8340 +                               AuDebugOn(dinfo->di_btop != dinfo->di_bbot);
8341 +                               au_set_h_dptr(dentry, dinfo->di_btop, NULL);
8342 +                               au_di_cp(dinfo, tmp);
8343 +                               hd = au_hdentry(tmp, tmp->di_btop);
8344 +                               au_set_h_dptr(dentry, tmp->di_btop,
8345 +                                             dget(hd->hd_dentry));
8346 +                       }
8347 +                       au_dbg_verify_dinode(dentry);
8348 +               } else {
8349 +                       AuDbg("negative --> positive\n");
8350 +                       /*
8351 +                        * similar to the behaviour of creating with bypassing
8352 +                        * aufs.
8353 +                        * unhash it in order to force an error in the
8354 +                        * succeeding create operation.
8355 +                        * we should not set S_DEAD here.
8356 +                        */
8357 +                       d_drop(dentry);
8358 +                       /* au_di_swap(tmp, dinfo); */
8359 +                       au_dbg_verify_dinode(dentry);
8360 +               }
8361 +       } else {
8362 +               AuDbg("positive originally\n");
8363 +               /* inode may be NULL */
8364 +               AuDebugOn(inode && (inode->i_mode & S_IFMT) != orig_h.mode);
8365 +               if (!tmp_h.inode) {
8366 +                       AuDbg("positive --> negative\n");
8367 +                       /* or bypassing aufs */
8368 +                       au_hide(dentry);
8369 +                       if (tmp->di_bwh >= 0 && tmp->di_bwh <= dinfo->di_btop)
8370 +                               dinfo->di_bwh = tmp->di_bwh;
8371 +                       if (inode)
8372 +                               err = au_refresh_hinode_self(inode);
8373 +                       au_dbg_verify_dinode(dentry);
8374 +               } else if (orig_h.mode == tmp_h.mode) {
8375 +                       AuDbg("positive --> positive, same type\n");
8376 +                       if (!S_ISDIR(orig_h.mode)
8377 +                           && dinfo->di_btop > tmp->di_btop) {
8378 +                               /*
8379 +                                * similar to the behaviour of removing and
8380 +                                * creating.
8381 +                                */
8382 +                               au_hide(dentry);
8383 +                               if (inode)
8384 +                                       err = au_refresh_hinode_self(inode);
8385 +                               au_dbg_verify_dinode(dentry);
8386 +                       } else {
8387 +                               /* fill empty slots */
8388 +                               if (dinfo->di_btop > tmp->di_btop)
8389 +                                       dinfo->di_btop = tmp->di_btop;
8390 +                               if (dinfo->di_bbot < tmp->di_bbot)
8391 +                                       dinfo->di_bbot = tmp->di_bbot;
8392 +                               dinfo->di_bwh = tmp->di_bwh;
8393 +                               dinfo->di_bdiropq = tmp->di_bdiropq;
8394 +                               bbot = dinfo->di_bbot;
8395 +                               bindex = tmp->di_btop;
8396 +                               hd = au_hdentry(tmp, bindex);
8397 +                               for (; bindex <= bbot; bindex++, hd++) {
8398 +                                       if (au_h_dptr(dentry, bindex))
8399 +                                               continue;
8400 +                                       h_dentry = hd->hd_dentry;
8401 +                                       if (!h_dentry)
8402 +                                               continue;
8403 +                                       AuDebugOn(d_is_negative(h_dentry));
8404 +                                       h_inode = d_inode(h_dentry);
8405 +                                       AuDebugOn(orig_h.mode
8406 +                                                 != (h_inode->i_mode
8407 +                                                     & S_IFMT));
8408 +                                       au_set_h_dptr(dentry, bindex,
8409 +                                                     dget(h_dentry));
8410 +                               }
8411 +                               if (inode)
8412 +                                       err = au_refresh_hinode(inode, dentry);
8413 +                               au_dbg_verify_dinode(dentry);
8414 +                       }
8415 +               } else {
8416 +                       AuDbg("positive --> positive, different type\n");
8417 +                       /* similar to the behaviour of removing and creating */
8418 +                       au_hide(dentry);
8419 +                       if (inode)
8420 +                               err = au_refresh_hinode_self(inode);
8421 +                       au_dbg_verify_dinode(dentry);
8422 +               }
8423 +       }
8424 +
8425 +out:
8426 +       return err;
8427 +}
8428 +
8429 +void au_refresh_dop(struct dentry *dentry, int force_reval)
8430 +{
8431 +       const struct dentry_operations *dop
8432 +               = force_reval ? &aufs_dop : dentry->d_sb->s_d_op;
8433 +       static const unsigned int mask
8434 +               = DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE;
8435 +
8436 +       BUILD_BUG_ON(sizeof(mask) != sizeof(dentry->d_flags));
8437 +
8438 +       if (dentry->d_op == dop)
8439 +               return;
8440 +
8441 +       AuDbg("%pd\n", dentry);
8442 +       spin_lock(&dentry->d_lock);
8443 +       if (dop == &aufs_dop)
8444 +               dentry->d_flags |= mask;
8445 +       else
8446 +               dentry->d_flags &= ~mask;
8447 +       dentry->d_op = dop;
8448 +       spin_unlock(&dentry->d_lock);
8449 +}
8450 +
8451 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent)
8452 +{
8453 +       int err, ebrange, nbr;
8454 +       unsigned int sigen;
8455 +       struct au_dinfo *dinfo, *tmp;
8456 +       struct super_block *sb;
8457 +       struct inode *inode;
8458 +
8459 +       DiMustWriteLock(dentry);
8460 +       AuDebugOn(IS_ROOT(dentry));
8461 +       AuDebugOn(d_really_is_negative(parent));
8462 +
8463 +       sb = dentry->d_sb;
8464 +       sigen = au_sigen(sb);
8465 +       err = au_digen_test(parent, sigen);
8466 +       if (unlikely(err))
8467 +               goto out;
8468 +
8469 +       nbr = au_sbbot(sb) + 1;
8470 +       dinfo = au_di(dentry);
8471 +       err = au_di_realloc(dinfo, nbr, /*may_shrink*/0);
8472 +       if (unlikely(err))
8473 +               goto out;
8474 +       ebrange = au_dbrange_test(dentry);
8475 +       if (!ebrange)
8476 +               ebrange = au_do_refresh_hdentry(dentry, parent);
8477 +
8478 +       if (d_unhashed(dentry) || ebrange /* || dinfo->di_tmpfile */) {
8479 +               AuDebugOn(au_dbtop(dentry) < 0 && au_dbbot(dentry) >= 0);
8480 +               if (d_really_is_positive(dentry)) {
8481 +                       inode = d_inode(dentry);
8482 +                       err = au_refresh_hinode_self(inode);
8483 +               }
8484 +               au_dbg_verify_dinode(dentry);
8485 +               if (!err)
8486 +                       goto out_dgen; /* success */
8487 +               goto out;
8488 +       }
8489 +
8490 +       /* temporary dinfo */
8491 +       AuDbgDentry(dentry);
8492 +       err = -ENOMEM;
8493 +       tmp = au_di_alloc(sb, AuLsc_DI_TMP);
8494 +       if (unlikely(!tmp))
8495 +               goto out;
8496 +       au_di_swap(tmp, dinfo);
8497 +       /* returns the number of positive dentries */
8498 +       /*
8499 +        * if current working dir is removed, it returns an error.
8500 +        * but the dentry is legal.
8501 +        */
8502 +       err = au_lkup_dentry(dentry, /*btop*/0, AuLkup_ALLOW_NEG);
8503 +       AuDbgDentry(dentry);
8504 +       au_di_swap(tmp, dinfo);
8505 +       if (err == -ENOENT)
8506 +               err = 0;
8507 +       if (err >= 0) {
8508 +               /* compare/refresh by dinfo */
8509 +               AuDbgDentry(dentry);
8510 +               err = au_refresh_by_dinfo(dentry, dinfo, tmp);
8511 +               au_dbg_verify_dinode(dentry);
8512 +               AuTraceErr(err);
8513 +       }
8514 +       au_di_realloc(dinfo, nbr, /*may_shrink*/1); /* harmless if err */
8515 +       au_rw_write_unlock(&tmp->di_rwsem);
8516 +       au_di_free(tmp);
8517 +       if (unlikely(err))
8518 +               goto out;
8519 +
8520 +out_dgen:
8521 +       au_update_digen(dentry);
8522 +out:
8523 +       if (unlikely(err && !(dentry->d_flags & DCACHE_NFSFS_RENAMED))) {
8524 +               AuIOErr("failed refreshing %pd, %d\n", dentry, err);
8525 +               AuDbgDentry(dentry);
8526 +       }
8527 +       AuTraceErr(err);
8528 +       return err;
8529 +}
8530 +
8531 +static int au_do_h_d_reval(struct dentry *h_dentry, unsigned int flags,
8532 +                          struct dentry *dentry, aufs_bindex_t bindex)
8533 +{
8534 +       int err, valid;
8535 +
8536 +       err = 0;
8537 +       if (!(h_dentry->d_flags & DCACHE_OP_REVALIDATE))
8538 +               goto out;
8539 +
8540 +       AuDbg("b%d\n", bindex);
8541 +       /*
8542 +        * gave up supporting LOOKUP_CREATE/OPEN for lower fs,
8543 +        * due to whiteout and branch permission.
8544 +        */
8545 +       flags &= ~(/*LOOKUP_PARENT |*/ LOOKUP_OPEN | LOOKUP_CREATE
8546 +                  | LOOKUP_FOLLOW | LOOKUP_EXCL);
8547 +       /* it may return tri-state */
8548 +       valid = h_dentry->d_op->d_revalidate(h_dentry, flags);
8549 +
8550 +       if (unlikely(valid < 0))
8551 +               err = valid;
8552 +       else if (!valid)
8553 +               err = -EINVAL;
8554 +
8555 +out:
8556 +       AuTraceErr(err);
8557 +       return err;
8558 +}
8559 +
8560 +/* todo: remove this */
8561 +static int h_d_revalidate(struct dentry *dentry, struct inode *inode,
8562 +                         unsigned int flags, int do_udba)
8563 +{
8564 +       int err;
8565 +       umode_t mode, h_mode;
8566 +       aufs_bindex_t bindex, btail, btop, ibs, ibe;
8567 +       unsigned char plus, unhashed, is_root, h_plus, h_nfs, tmpfile;
8568 +       struct inode *h_inode, *h_cached_inode;
8569 +       struct dentry *h_dentry;
8570 +       struct qstr *name, *h_name;
8571 +
8572 +       err = 0;
8573 +       plus = 0;
8574 +       mode = 0;
8575 +       ibs = -1;
8576 +       ibe = -1;
8577 +       unhashed = !!d_unhashed(dentry);
8578 +       is_root = !!IS_ROOT(dentry);
8579 +       name = &dentry->d_name;
8580 +       tmpfile = au_di(dentry)->di_tmpfile;
8581 +
8582 +       /*
8583 +        * Theoretically, REVAL test should be unnecessary in case of
8584 +        * {FS,I}NOTIFY.
8585 +        * But {fs,i}notify doesn't fire some necessary events,
8586 +        *      IN_ATTRIB for atime/nlink/pageio
8587 +        * Let's do REVAL test too.
8588 +        */
8589 +       if (do_udba && inode) {
8590 +               mode = (inode->i_mode & S_IFMT);
8591 +               plus = (inode->i_nlink > 0);
8592 +               ibs = au_ibtop(inode);
8593 +               ibe = au_ibbot(inode);
8594 +       }
8595 +
8596 +       btop = au_dbtop(dentry);
8597 +       btail = btop;
8598 +       if (inode && S_ISDIR(inode->i_mode))
8599 +               btail = au_dbtaildir(dentry);
8600 +       for (bindex = btop; bindex <= btail; bindex++) {
8601 +               h_dentry = au_h_dptr(dentry, bindex);
8602 +               if (!h_dentry)
8603 +                       continue;
8604 +
8605 +               AuDbg("b%d, %pd\n", bindex, h_dentry);
8606 +               h_nfs = !!au_test_nfs(h_dentry->d_sb);
8607 +               spin_lock(&h_dentry->d_lock);
8608 +               h_name = &h_dentry->d_name;
8609 +               if (unlikely(do_udba
8610 +                            && !is_root
8611 +                            && ((!h_nfs
8612 +                                 && (unhashed != !!d_unhashed(h_dentry)
8613 +                                     || (!tmpfile
8614 +                                         && !au_qstreq(name, h_name))
8615 +                                         ))
8616 +                                || (h_nfs
8617 +                                    && !(flags & LOOKUP_OPEN)
8618 +                                    && (h_dentry->d_flags
8619 +                                        & DCACHE_NFSFS_RENAMED)))
8620 +                           )) {
8621 +                       int h_unhashed;
8622 +
8623 +                       h_unhashed = d_unhashed(h_dentry);
8624 +                       spin_unlock(&h_dentry->d_lock);
8625 +                       AuDbg("unhash 0x%x 0x%x, %pd %pd\n",
8626 +                             unhashed, h_unhashed, dentry, h_dentry);
8627 +                       goto err;
8628 +               }
8629 +               spin_unlock(&h_dentry->d_lock);
8630 +
8631 +               err = au_do_h_d_reval(h_dentry, flags, dentry, bindex);
8632 +               if (unlikely(err))
8633 +                       /* do not goto err, to keep the errno */
8634 +                       break;
8635 +
8636 +               /* todo: plink too? */
8637 +               if (!do_udba)
8638 +                       continue;
8639 +
8640 +               /* UDBA tests */
8641 +               if (unlikely(!!inode != d_is_positive(h_dentry)))
8642 +                       goto err;
8643 +
8644 +               h_inode = NULL;
8645 +               if (d_is_positive(h_dentry))
8646 +                       h_inode = d_inode(h_dentry);
8647 +               h_plus = plus;
8648 +               h_mode = mode;
8649 +               h_cached_inode = h_inode;
8650 +               if (h_inode) {
8651 +                       h_mode = (h_inode->i_mode & S_IFMT);
8652 +                       h_plus = (h_inode->i_nlink > 0);
8653 +               }
8654 +               if (inode && ibs <= bindex && bindex <= ibe)
8655 +                       h_cached_inode = au_h_iptr(inode, bindex);
8656 +
8657 +               if (!h_nfs) {
8658 +                       if (unlikely(plus != h_plus && !tmpfile))
8659 +                               goto err;
8660 +               } else {
8661 +                       if (unlikely(!(h_dentry->d_flags & DCACHE_NFSFS_RENAMED)
8662 +                                    && !is_root
8663 +                                    && !IS_ROOT(h_dentry)
8664 +                                    && unhashed != d_unhashed(h_dentry)))
8665 +                               goto err;
8666 +               }
8667 +               if (unlikely(mode != h_mode
8668 +                            || h_cached_inode != h_inode))
8669 +                       goto err;
8670 +               continue;
8671 +
8672 +err:
8673 +               err = -EINVAL;
8674 +               break;
8675 +       }
8676 +
8677 +       AuTraceErr(err);
8678 +       return err;
8679 +}
8680 +
8681 +/* todo: consolidate with do_refresh() and au_reval_for_attr() */
8682 +static int simple_reval_dpath(struct dentry *dentry, unsigned int sigen)
8683 +{
8684 +       int err;
8685 +       struct dentry *parent;
8686 +
8687 +       if (!au_digen_test(dentry, sigen))
8688 +               return 0;
8689 +
8690 +       parent = dget_parent(dentry);
8691 +       di_read_lock_parent(parent, AuLock_IR);
8692 +       AuDebugOn(au_digen_test(parent, sigen));
8693 +       au_dbg_verify_gen(parent, sigen);
8694 +       err = au_refresh_dentry(dentry, parent);
8695 +       di_read_unlock(parent, AuLock_IR);
8696 +       dput(parent);
8697 +       AuTraceErr(err);
8698 +       return err;
8699 +}
8700 +
8701 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen)
8702 +{
8703 +       int err;
8704 +       struct dentry *d, *parent;
8705 +
8706 +       if (!au_ftest_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR))
8707 +               return simple_reval_dpath(dentry, sigen);
8708 +
8709 +       /* slow loop, keep it simple and stupid */
8710 +       /* cf: au_cpup_dirs() */
8711 +       err = 0;
8712 +       parent = NULL;
8713 +       while (au_digen_test(dentry, sigen)) {
8714 +               d = dentry;
8715 +               while (1) {
8716 +                       dput(parent);
8717 +                       parent = dget_parent(d);
8718 +                       if (!au_digen_test(parent, sigen))
8719 +                               break;
8720 +                       d = parent;
8721 +               }
8722 +
8723 +               if (d != dentry)
8724 +                       di_write_lock_child2(d);
8725 +
8726 +               /* someone might update our dentry while we were sleeping */
8727 +               if (au_digen_test(d, sigen)) {
8728 +                       /*
8729 +                        * todo: consolidate with simple_reval_dpath(),
8730 +                        * do_refresh() and au_reval_for_attr().
8731 +                        */
8732 +                       di_read_lock_parent(parent, AuLock_IR);
8733 +                       err = au_refresh_dentry(d, parent);
8734 +                       di_read_unlock(parent, AuLock_IR);
8735 +               }
8736 +
8737 +               if (d != dentry)
8738 +                       di_write_unlock(d);
8739 +               dput(parent);
8740 +               if (unlikely(err))
8741 +                       break;
8742 +       }
8743 +
8744 +       return err;
8745 +}
8746 +
8747 +/*
8748 + * if valid returns 1, otherwise 0.
8749 + */
8750 +static int aufs_d_revalidate(struct dentry *dentry, unsigned int flags)
8751 +{
8752 +       int valid, err;
8753 +       unsigned int sigen;
8754 +       unsigned char do_udba;
8755 +       struct super_block *sb;
8756 +       struct inode *inode;
8757 +
8758 +       /* todo: support rcu-walk? */
8759 +       if (flags & LOOKUP_RCU)
8760 +               return -ECHILD;
8761 +
8762 +       valid = 0;
8763 +       if (unlikely(!au_di(dentry)))
8764 +               goto out;
8765 +
8766 +       valid = 1;
8767 +       sb = dentry->d_sb;
8768 +       /*
8769 +        * todo: very ugly
8770 +        * i_mutex of parent dir may be held,
8771 +        * but we should not return 'invalid' due to busy.
8772 +        */
8773 +       err = aufs_read_lock(dentry, AuLock_FLUSH | AuLock_DW | AuLock_NOPLM);
8774 +       if (unlikely(err)) {
8775 +               valid = err;
8776 +               AuTraceErr(err);
8777 +               goto out;
8778 +       }
8779 +       inode = NULL;
8780 +       if (d_really_is_positive(dentry))
8781 +               inode = d_inode(dentry);
8782 +       if (unlikely(inode && au_is_bad_inode(inode))) {
8783 +               err = -EINVAL;
8784 +               AuTraceErr(err);
8785 +               goto out_dgrade;
8786 +       }
8787 +       if (unlikely(au_dbrange_test(dentry))) {
8788 +               err = -EINVAL;
8789 +               AuTraceErr(err);
8790 +               goto out_dgrade;
8791 +       }
8792 +
8793 +       sigen = au_sigen(sb);
8794 +       if (au_digen_test(dentry, sigen)) {
8795 +               AuDebugOn(IS_ROOT(dentry));
8796 +               err = au_reval_dpath(dentry, sigen);
8797 +               if (unlikely(err)) {
8798 +                       AuTraceErr(err);
8799 +                       goto out_dgrade;
8800 +               }
8801 +       }
8802 +       di_downgrade_lock(dentry, AuLock_IR);
8803 +
8804 +       err = -EINVAL;
8805 +       if (!(flags & (LOOKUP_OPEN | LOOKUP_EMPTY))
8806 +           && inode
8807 +           && !(inode->i_state && I_LINKABLE)
8808 +           && (IS_DEADDIR(inode) || !inode->i_nlink)) {
8809 +               AuTraceErr(err);
8810 +               goto out_inval;
8811 +       }
8812 +
8813 +       do_udba = !au_opt_test(au_mntflags(sb), UDBA_NONE);
8814 +       if (do_udba && inode) {
8815 +               aufs_bindex_t btop = au_ibtop(inode);
8816 +               struct inode *h_inode;
8817 +
8818 +               if (btop >= 0) {
8819 +                       h_inode = au_h_iptr(inode, btop);
8820 +                       if (h_inode && au_test_higen(inode, h_inode)) {
8821 +                               AuTraceErr(err);
8822 +                               goto out_inval;
8823 +                       }
8824 +               }
8825 +       }
8826 +
8827 +       err = h_d_revalidate(dentry, inode, flags, do_udba);
8828 +       if (unlikely(!err && do_udba && au_dbtop(dentry) < 0)) {
8829 +               err = -EIO;
8830 +               AuDbg("both of real entry and whiteout found, %p, err %d\n",
8831 +                     dentry, err);
8832 +       }
8833 +       goto out_inval;
8834 +
8835 +out_dgrade:
8836 +       di_downgrade_lock(dentry, AuLock_IR);
8837 +out_inval:
8838 +       aufs_read_unlock(dentry, AuLock_IR);
8839 +       AuTraceErr(err);
8840 +       valid = !err;
8841 +out:
8842 +       if (!valid) {
8843 +               AuDbg("%pd invalid, %d\n", dentry, valid);
8844 +               d_drop(dentry);
8845 +       }
8846 +       return valid;
8847 +}
8848 +
8849 +static void aufs_d_release(struct dentry *dentry)
8850 +{
8851 +       if (au_di(dentry)) {
8852 +               au_di_fin(dentry);
8853 +               au_hn_di_reinit(dentry);
8854 +       }
8855 +}
8856 +
8857 +const struct dentry_operations aufs_dop = {
8858 +       .d_revalidate           = aufs_d_revalidate,
8859 +       .d_weak_revalidate      = aufs_d_revalidate,
8860 +       .d_release              = aufs_d_release
8861 +};
8862 +
8863 +/* aufs_dop without d_revalidate */
8864 +const struct dentry_operations aufs_dop_noreval = {
8865 +       .d_release              = aufs_d_release
8866 +};
8867 diff -urN /usr/share/empty/fs/aufs/dentry.h linux/fs/aufs/dentry.h
8868 --- /usr/share/empty/fs/aufs/dentry.h   1970-01-01 01:00:00.000000000 +0100
8869 +++ linux/fs/aufs/dentry.h      2017-05-06 22:16:52.778220093 +0200
8870 @@ -0,0 +1,255 @@
8871 +/*
8872 + * Copyright (C) 2005-2017 Junjiro R. Okajima
8873 + *
8874 + * This program, aufs is free software; you can redistribute it and/or modify
8875 + * it under the terms of the GNU General Public License as published by
8876 + * the Free Software Foundation; either version 2 of the License, or
8877 + * (at your option) any later version.
8878 + *
8879 + * This program is distributed in the hope that it will be useful,
8880 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8881 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8882 + * GNU General Public License for more details.
8883 + *
8884 + * You should have received a copy of the GNU General Public License
8885 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8886 + */
8887 +
8888 +/*
8889 + * lookup and dentry operations
8890 + */
8891 +
8892 +#ifndef __AUFS_DENTRY_H__
8893 +#define __AUFS_DENTRY_H__
8894 +
8895 +#ifdef __KERNEL__
8896 +
8897 +#include <linux/dcache.h>
8898 +#include "rwsem.h"
8899 +
8900 +struct au_hdentry {
8901 +       struct dentry           *hd_dentry;
8902 +       aufs_bindex_t           hd_id;
8903 +};
8904 +
8905 +struct au_dinfo {
8906 +       atomic_t                di_generation;
8907 +
8908 +       struct au_rwsem         di_rwsem;
8909 +       aufs_bindex_t           di_btop, di_bbot, di_bwh, di_bdiropq;
8910 +       unsigned char           di_tmpfile; /* to allow the different name */
8911 +       union {
8912 +               struct au_hdentry       *di_hdentry;
8913 +               struct llist_node       di_lnode;       /* delayed free */
8914 +       };
8915 +} ____cacheline_aligned_in_smp;
8916 +
8917 +/* ---------------------------------------------------------------------- */
8918 +
8919 +/* flags for au_lkup_dentry() */
8920 +#define AuLkup_ALLOW_NEG       1
8921 +#define AuLkup_IGNORE_PERM     (1 << 1)
8922 +#define au_ftest_lkup(flags, name)     ((flags) & AuLkup_##name)
8923 +#define au_fset_lkup(flags, name) \
8924 +       do { (flags) |= AuLkup_##name; } while (0)
8925 +#define au_fclr_lkup(flags, name) \
8926 +       do { (flags) &= ~AuLkup_##name; } while (0)
8927 +
8928 +/* ---------------------------------------------------------------------- */
8929 +
8930 +/* dentry.c */
8931 +extern const struct dentry_operations aufs_dop, aufs_dop_noreval;
8932 +struct au_branch;
8933 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent);
8934 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
8935 +               struct dentry *h_parent, struct au_branch *br);
8936 +
8937 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t btop,
8938 +                  unsigned int flags);
8939 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh);
8940 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent);
8941 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen);
8942 +void au_refresh_dop(struct dentry *dentry, int force_reval);
8943 +
8944 +/* dinfo.c */
8945 +void au_di_init_once(void *_di);
8946 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc);
8947 +void au_di_free(struct au_dinfo *dinfo);
8948 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b);
8949 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src);
8950 +int au_di_init(struct dentry *dentry);
8951 +void au_di_fin(struct dentry *dentry);
8952 +int au_di_realloc(struct au_dinfo *dinfo, int nbr, int may_shrink);
8953 +
8954 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc);
8955 +void di_read_unlock(struct dentry *d, int flags);
8956 +void di_downgrade_lock(struct dentry *d, int flags);
8957 +void di_write_lock(struct dentry *d, unsigned int lsc);
8958 +void di_write_unlock(struct dentry *d);
8959 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir);
8960 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir);
8961 +void di_write_unlock2(struct dentry *d1, struct dentry *d2);
8962 +
8963 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex);
8964 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex);
8965 +aufs_bindex_t au_dbtail(struct dentry *dentry);
8966 +aufs_bindex_t au_dbtaildir(struct dentry *dentry);
8967 +
8968 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
8969 +                  struct dentry *h_dentry);
8970 +int au_digen_test(struct dentry *dentry, unsigned int sigen);
8971 +int au_dbrange_test(struct dentry *dentry);
8972 +void au_update_digen(struct dentry *dentry);
8973 +void au_update_dbrange(struct dentry *dentry, int do_put_zero);
8974 +void au_update_dbtop(struct dentry *dentry);
8975 +void au_update_dbbot(struct dentry *dentry);
8976 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry);
8977 +
8978 +/* ---------------------------------------------------------------------- */
8979 +
8980 +static inline struct au_dinfo *au_di(struct dentry *dentry)
8981 +{
8982 +       return dentry->d_fsdata;
8983 +}
8984 +
8985 +/* ---------------------------------------------------------------------- */
8986 +
8987 +/* lock subclass for dinfo */
8988 +enum {
8989 +       AuLsc_DI_CHILD,         /* child first */
8990 +       AuLsc_DI_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
8991 +       AuLsc_DI_CHILD3,        /* copyup dirs */
8992 +       AuLsc_DI_PARENT,
8993 +       AuLsc_DI_PARENT2,
8994 +       AuLsc_DI_PARENT3,
8995 +       AuLsc_DI_TMP            /* temp for replacing dinfo */
8996 +};
8997 +
8998 +/*
8999 + * di_read_lock_child, di_write_lock_child,
9000 + * di_read_lock_child2, di_write_lock_child2,
9001 + * di_read_lock_child3, di_write_lock_child3,
9002 + * di_read_lock_parent, di_write_lock_parent,
9003 + * di_read_lock_parent2, di_write_lock_parent2,
9004 + * di_read_lock_parent3, di_write_lock_parent3,
9005 + */
9006 +#define AuReadLockFunc(name, lsc) \
9007 +static inline void di_read_lock_##name(struct dentry *d, int flags) \
9008 +{ di_read_lock(d, flags, AuLsc_DI_##lsc); }
9009 +
9010 +#define AuWriteLockFunc(name, lsc) \
9011 +static inline void di_write_lock_##name(struct dentry *d) \
9012 +{ di_write_lock(d, AuLsc_DI_##lsc); }
9013 +
9014 +#define AuRWLockFuncs(name, lsc) \
9015 +       AuReadLockFunc(name, lsc) \
9016 +       AuWriteLockFunc(name, lsc)
9017 +
9018 +AuRWLockFuncs(child, CHILD);
9019 +AuRWLockFuncs(child2, CHILD2);
9020 +AuRWLockFuncs(child3, CHILD3);
9021 +AuRWLockFuncs(parent, PARENT);
9022 +AuRWLockFuncs(parent2, PARENT2);
9023 +AuRWLockFuncs(parent3, PARENT3);
9024 +
9025 +#undef AuReadLockFunc
9026 +#undef AuWriteLockFunc
9027 +#undef AuRWLockFuncs
9028 +
9029 +#define DiMustNoWaiters(d)     AuRwMustNoWaiters(&au_di(d)->di_rwsem)
9030 +#define DiMustAnyLock(d)       AuRwMustAnyLock(&au_di(d)->di_rwsem)
9031 +#define DiMustWriteLock(d)     AuRwMustWriteLock(&au_di(d)->di_rwsem)
9032 +
9033 +/* ---------------------------------------------------------------------- */
9034 +
9035 +/* todo: memory barrier? */
9036 +static inline unsigned int au_digen(struct dentry *d)
9037 +{
9038 +       return atomic_read(&au_di(d)->di_generation);
9039 +}
9040 +
9041 +static inline void au_h_dentry_init(struct au_hdentry *hdentry)
9042 +{
9043 +       hdentry->hd_dentry = NULL;
9044 +}
9045 +
9046 +static inline struct au_hdentry *au_hdentry(struct au_dinfo *di,
9047 +                                           aufs_bindex_t bindex)
9048 +{
9049 +       return di->di_hdentry + bindex;
9050 +}
9051 +
9052 +static inline void au_hdput(struct au_hdentry *hd)
9053 +{
9054 +       if (hd)
9055 +               dput(hd->hd_dentry);
9056 +}
9057 +
9058 +static inline aufs_bindex_t au_dbtop(struct dentry *dentry)
9059 +{
9060 +       DiMustAnyLock(dentry);
9061 +       return au_di(dentry)->di_btop;
9062 +}
9063 +
9064 +static inline aufs_bindex_t au_dbbot(struct dentry *dentry)
9065 +{
9066 +       DiMustAnyLock(dentry);
9067 +       return au_di(dentry)->di_bbot;
9068 +}
9069 +
9070 +static inline aufs_bindex_t au_dbwh(struct dentry *dentry)
9071 +{
9072 +       DiMustAnyLock(dentry);
9073 +       return au_di(dentry)->di_bwh;
9074 +}
9075 +
9076 +static inline aufs_bindex_t au_dbdiropq(struct dentry *dentry)
9077 +{
9078 +       DiMustAnyLock(dentry);
9079 +       return au_di(dentry)->di_bdiropq;
9080 +}
9081 +
9082 +/* todo: hard/soft set? */
9083 +static inline void au_set_dbtop(struct dentry *dentry, aufs_bindex_t bindex)
9084 +{
9085 +       DiMustWriteLock(dentry);
9086 +       au_di(dentry)->di_btop = bindex;
9087 +}
9088 +
9089 +static inline void au_set_dbbot(struct dentry *dentry, aufs_bindex_t bindex)
9090 +{
9091 +       DiMustWriteLock(dentry);
9092 +       au_di(dentry)->di_bbot = bindex;
9093 +}
9094 +
9095 +static inline void au_set_dbwh(struct dentry *dentry, aufs_bindex_t bindex)
9096 +{
9097 +       DiMustWriteLock(dentry);
9098 +       /* dbwh can be outside of btop - bbot range */
9099 +       au_di(dentry)->di_bwh = bindex;
9100 +}
9101 +
9102 +static inline void au_set_dbdiropq(struct dentry *dentry, aufs_bindex_t bindex)
9103 +{
9104 +       DiMustWriteLock(dentry);
9105 +       au_di(dentry)->di_bdiropq = bindex;
9106 +}
9107 +
9108 +/* ---------------------------------------------------------------------- */
9109 +
9110 +#ifdef CONFIG_AUFS_HNOTIFY
9111 +static inline void au_digen_dec(struct dentry *d)
9112 +{
9113 +       atomic_dec(&au_di(d)->di_generation);
9114 +}
9115 +
9116 +static inline void au_hn_di_reinit(struct dentry *dentry)
9117 +{
9118 +       dentry->d_fsdata = NULL;
9119 +}
9120 +#else
9121 +AuStubVoid(au_hn_di_reinit, struct dentry *dentry __maybe_unused)
9122 +#endif /* CONFIG_AUFS_HNOTIFY */
9123 +
9124 +#endif /* __KERNEL__ */
9125 +#endif /* __AUFS_DENTRY_H__ */
9126 diff -urN /usr/share/empty/fs/aufs/dinfo.c linux/fs/aufs/dinfo.c
9127 --- /usr/share/empty/fs/aufs/dinfo.c    1970-01-01 01:00:00.000000000 +0100
9128 +++ linux/fs/aufs/dinfo.c       2017-05-06 22:16:52.778220093 +0200
9129 @@ -0,0 +1,553 @@
9130 +/*
9131 + * Copyright (C) 2005-2017 Junjiro R. Okajima
9132 + *
9133 + * This program, aufs is free software; you can redistribute it and/or modify
9134 + * it under the terms of the GNU General Public License as published by
9135 + * the Free Software Foundation; either version 2 of the License, or
9136 + * (at your option) any later version.
9137 + *
9138 + * This program is distributed in the hope that it will be useful,
9139 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9140 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9141 + * GNU General Public License for more details.
9142 + *
9143 + * You should have received a copy of the GNU General Public License
9144 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9145 + */
9146 +
9147 +/*
9148 + * dentry private data
9149 + */
9150 +
9151 +#include "aufs.h"
9152 +
9153 +void au_di_init_once(void *_dinfo)
9154 +{
9155 +       struct au_dinfo *dinfo = _dinfo;
9156 +
9157 +       au_rw_init(&dinfo->di_rwsem);
9158 +}
9159 +
9160 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc)
9161 +{
9162 +       struct au_dinfo *dinfo;
9163 +       int nbr, i;
9164 +
9165 +       dinfo = au_cache_alloc_dinfo();
9166 +       if (unlikely(!dinfo))
9167 +               goto out;
9168 +
9169 +       nbr = au_sbbot(sb) + 1;
9170 +       if (nbr <= 0)
9171 +               nbr = 1;
9172 +       dinfo->di_hdentry = kcalloc(nbr, sizeof(*dinfo->di_hdentry), GFP_NOFS);
9173 +       if (dinfo->di_hdentry) {
9174 +               au_rw_write_lock_nested(&dinfo->di_rwsem, lsc);
9175 +               dinfo->di_btop = -1;
9176 +               dinfo->di_bbot = -1;
9177 +               dinfo->di_bwh = -1;
9178 +               dinfo->di_bdiropq = -1;
9179 +               dinfo->di_tmpfile = 0;
9180 +               for (i = 0; i < nbr; i++)
9181 +                       dinfo->di_hdentry[i].hd_id = -1;
9182 +               goto out;
9183 +       }
9184 +
9185 +       au_cache_dfree_dinfo(dinfo);
9186 +       dinfo = NULL;
9187 +
9188 +out:
9189 +       return dinfo;
9190 +}
9191 +
9192 +void au_di_free(struct au_dinfo *dinfo)
9193 +{
9194 +       struct au_hdentry *p;
9195 +       aufs_bindex_t bbot, bindex;
9196 +
9197 +       /* dentry may not be revalidated */
9198 +       bindex = dinfo->di_btop;
9199 +       if (bindex >= 0) {
9200 +               bbot = dinfo->di_bbot;
9201 +               p = au_hdentry(dinfo, bindex);
9202 +               while (bindex++ <= bbot)
9203 +                       au_hdput(p++);
9204 +       }
9205 +       au_delayed_kfree(dinfo->di_hdentry);
9206 +       au_cache_dfree_dinfo(dinfo);
9207 +}
9208 +
9209 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b)
9210 +{
9211 +       struct au_hdentry *p;
9212 +       aufs_bindex_t bi;
9213 +
9214 +       AuRwMustWriteLock(&a->di_rwsem);
9215 +       AuRwMustWriteLock(&b->di_rwsem);
9216 +
9217 +#define DiSwap(v, name)                                \
9218 +       do {                                    \
9219 +               v = a->di_##name;               \
9220 +               a->di_##name = b->di_##name;    \
9221 +               b->di_##name = v;               \
9222 +       } while (0)
9223 +
9224 +       DiSwap(p, hdentry);
9225 +       DiSwap(bi, btop);
9226 +       DiSwap(bi, bbot);
9227 +       DiSwap(bi, bwh);
9228 +       DiSwap(bi, bdiropq);
9229 +       /* smp_mb(); */
9230 +
9231 +#undef DiSwap
9232 +}
9233 +
9234 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src)
9235 +{
9236 +       AuRwMustWriteLock(&dst->di_rwsem);
9237 +       AuRwMustWriteLock(&src->di_rwsem);
9238 +
9239 +       dst->di_btop = src->di_btop;
9240 +       dst->di_bbot = src->di_bbot;
9241 +       dst->di_bwh = src->di_bwh;
9242 +       dst->di_bdiropq = src->di_bdiropq;
9243 +       /* smp_mb(); */
9244 +}
9245 +
9246 +int au_di_init(struct dentry *dentry)
9247 +{
9248 +       int err;
9249 +       struct super_block *sb;
9250 +       struct au_dinfo *dinfo;
9251 +
9252 +       err = 0;
9253 +       sb = dentry->d_sb;
9254 +       dinfo = au_di_alloc(sb, AuLsc_DI_CHILD);
9255 +       if (dinfo) {
9256 +               atomic_set(&dinfo->di_generation, au_sigen(sb));
9257 +               /* smp_mb(); */ /* atomic_set */
9258 +               dentry->d_fsdata = dinfo;
9259 +       } else
9260 +               err = -ENOMEM;
9261 +
9262 +       return err;
9263 +}
9264 +
9265 +void au_di_fin(struct dentry *dentry)
9266 +{
9267 +       struct au_dinfo *dinfo;
9268 +
9269 +       dinfo = au_di(dentry);
9270 +       AuRwDestroy(&dinfo->di_rwsem);
9271 +       au_di_free(dinfo);
9272 +}
9273 +
9274 +int au_di_realloc(struct au_dinfo *dinfo, int nbr, int may_shrink)
9275 +{
9276 +       int err, sz;
9277 +       struct au_hdentry *hdp;
9278 +
9279 +       AuRwMustWriteLock(&dinfo->di_rwsem);
9280 +
9281 +       err = -ENOMEM;
9282 +       sz = sizeof(*hdp) * (dinfo->di_bbot + 1);
9283 +       if (!sz)
9284 +               sz = sizeof(*hdp);
9285 +       hdp = au_kzrealloc(dinfo->di_hdentry, sz, sizeof(*hdp) * nbr, GFP_NOFS,
9286 +                          may_shrink);
9287 +       if (hdp) {
9288 +               dinfo->di_hdentry = hdp;
9289 +               err = 0;
9290 +       }
9291 +
9292 +       return err;
9293 +}
9294 +
9295 +/* ---------------------------------------------------------------------- */
9296 +
9297 +static void do_ii_write_lock(struct inode *inode, unsigned int lsc)
9298 +{
9299 +       switch (lsc) {
9300 +       case AuLsc_DI_CHILD:
9301 +               ii_write_lock_child(inode);
9302 +               break;
9303 +       case AuLsc_DI_CHILD2:
9304 +               ii_write_lock_child2(inode);
9305 +               break;
9306 +       case AuLsc_DI_CHILD3:
9307 +               ii_write_lock_child3(inode);
9308 +               break;
9309 +       case AuLsc_DI_PARENT:
9310 +               ii_write_lock_parent(inode);
9311 +               break;
9312 +       case AuLsc_DI_PARENT2:
9313 +               ii_write_lock_parent2(inode);
9314 +               break;
9315 +       case AuLsc_DI_PARENT3:
9316 +               ii_write_lock_parent3(inode);
9317 +               break;
9318 +       default:
9319 +               BUG();
9320 +       }
9321 +}
9322 +
9323 +static void do_ii_read_lock(struct inode *inode, unsigned int lsc)
9324 +{
9325 +       switch (lsc) {
9326 +       case AuLsc_DI_CHILD:
9327 +               ii_read_lock_child(inode);
9328 +               break;
9329 +       case AuLsc_DI_CHILD2:
9330 +               ii_read_lock_child2(inode);
9331 +               break;
9332 +       case AuLsc_DI_CHILD3:
9333 +               ii_read_lock_child3(inode);
9334 +               break;
9335 +       case AuLsc_DI_PARENT:
9336 +               ii_read_lock_parent(inode);
9337 +               break;
9338 +       case AuLsc_DI_PARENT2:
9339 +               ii_read_lock_parent2(inode);
9340 +               break;
9341 +       case AuLsc_DI_PARENT3:
9342 +               ii_read_lock_parent3(inode);
9343 +               break;
9344 +       default:
9345 +               BUG();
9346 +       }
9347 +}
9348 +
9349 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc)
9350 +{
9351 +       struct inode *inode;
9352 +
9353 +       au_rw_read_lock_nested(&au_di(d)->di_rwsem, lsc);
9354 +       if (d_really_is_positive(d)) {
9355 +               inode = d_inode(d);
9356 +               if (au_ftest_lock(flags, IW))
9357 +                       do_ii_write_lock(inode, lsc);
9358 +               else if (au_ftest_lock(flags, IR))
9359 +                       do_ii_read_lock(inode, lsc);
9360 +       }
9361 +}
9362 +
9363 +void di_read_unlock(struct dentry *d, int flags)
9364 +{
9365 +       struct inode *inode;
9366 +
9367 +       if (d_really_is_positive(d)) {
9368 +               inode = d_inode(d);
9369 +               if (au_ftest_lock(flags, IW)) {
9370 +                       au_dbg_verify_dinode(d);
9371 +                       ii_write_unlock(inode);
9372 +               } else if (au_ftest_lock(flags, IR)) {
9373 +                       au_dbg_verify_dinode(d);
9374 +                       ii_read_unlock(inode);
9375 +               }
9376 +       }
9377 +       au_rw_read_unlock(&au_di(d)->di_rwsem);
9378 +}
9379 +
9380 +void di_downgrade_lock(struct dentry *d, int flags)
9381 +{
9382 +       if (d_really_is_positive(d) && au_ftest_lock(flags, IR))
9383 +               ii_downgrade_lock(d_inode(d));
9384 +       au_rw_dgrade_lock(&au_di(d)->di_rwsem);
9385 +}
9386 +
9387 +void di_write_lock(struct dentry *d, unsigned int lsc)
9388 +{
9389 +       au_rw_write_lock_nested(&au_di(d)->di_rwsem, lsc);
9390 +       if (d_really_is_positive(d))
9391 +               do_ii_write_lock(d_inode(d), lsc);
9392 +}
9393 +
9394 +void di_write_unlock(struct dentry *d)
9395 +{
9396 +       au_dbg_verify_dinode(d);
9397 +       if (d_really_is_positive(d))
9398 +               ii_write_unlock(d_inode(d));
9399 +       au_rw_write_unlock(&au_di(d)->di_rwsem);
9400 +}
9401 +
9402 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir)
9403 +{
9404 +       AuDebugOn(d1 == d2
9405 +                 || d_inode(d1) == d_inode(d2)
9406 +                 || d1->d_sb != d2->d_sb);
9407 +
9408 +       if ((isdir && au_test_subdir(d1, d2))
9409 +           || d1 < d2) {
9410 +               di_write_lock_child(d1);
9411 +               di_write_lock_child2(d2);
9412 +       } else {
9413 +               di_write_lock_child(d2);
9414 +               di_write_lock_child2(d1);
9415 +       }
9416 +}
9417 +
9418 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir)
9419 +{
9420 +       AuDebugOn(d1 == d2
9421 +                 || d_inode(d1) == d_inode(d2)
9422 +                 || d1->d_sb != d2->d_sb);
9423 +
9424 +       if ((isdir && au_test_subdir(d1, d2))
9425 +           || d1 < d2) {
9426 +               di_write_lock_parent(d1);
9427 +               di_write_lock_parent2(d2);
9428 +       } else {
9429 +               di_write_lock_parent(d2);
9430 +               di_write_lock_parent2(d1);
9431 +       }
9432 +}
9433 +
9434 +void di_write_unlock2(struct dentry *d1, struct dentry *d2)
9435 +{
9436 +       di_write_unlock(d1);
9437 +       if (d_inode(d1) == d_inode(d2))
9438 +               au_rw_write_unlock(&au_di(d2)->di_rwsem);
9439 +       else
9440 +               di_write_unlock(d2);
9441 +}
9442 +
9443 +/* ---------------------------------------------------------------------- */
9444 +
9445 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex)
9446 +{
9447 +       struct dentry *d;
9448 +
9449 +       DiMustAnyLock(dentry);
9450 +
9451 +       if (au_dbtop(dentry) < 0 || bindex < au_dbtop(dentry))
9452 +               return NULL;
9453 +       AuDebugOn(bindex < 0);
9454 +       d = au_hdentry(au_di(dentry), bindex)->hd_dentry;
9455 +       AuDebugOn(d && au_dcount(d) <= 0);
9456 +       return d;
9457 +}
9458 +
9459 +/*
9460 + * extended version of au_h_dptr().
9461 + * returns a hashed and positive (or linkable) h_dentry in bindex, NULL, or
9462 + * error.
9463 + */
9464 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex)
9465 +{
9466 +       struct dentry *h_dentry;
9467 +       struct inode *inode, *h_inode;
9468 +
9469 +       AuDebugOn(d_really_is_negative(dentry));
9470 +
9471 +       h_dentry = NULL;
9472 +       if (au_dbtop(dentry) <= bindex
9473 +           && bindex <= au_dbbot(dentry))
9474 +               h_dentry = au_h_dptr(dentry, bindex);
9475 +       if (h_dentry && !au_d_linkable(h_dentry)) {
9476 +               dget(h_dentry);
9477 +               goto out; /* success */
9478 +       }
9479 +
9480 +       inode = d_inode(dentry);
9481 +       AuDebugOn(bindex < au_ibtop(inode));
9482 +       AuDebugOn(au_ibbot(inode) < bindex);
9483 +       h_inode = au_h_iptr(inode, bindex);
9484 +       h_dentry = d_find_alias(h_inode);
9485 +       if (h_dentry) {
9486 +               if (!IS_ERR(h_dentry)) {
9487 +                       if (!au_d_linkable(h_dentry))
9488 +                               goto out; /* success */
9489 +                       dput(h_dentry);
9490 +               } else
9491 +                       goto out;
9492 +       }
9493 +
9494 +       if (au_opt_test(au_mntflags(dentry->d_sb), PLINK)) {
9495 +               h_dentry = au_plink_lkup(inode, bindex);
9496 +               AuDebugOn(!h_dentry);
9497 +               if (!IS_ERR(h_dentry)) {
9498 +                       if (!au_d_hashed_positive(h_dentry))
9499 +                               goto out; /* success */
9500 +                       dput(h_dentry);
9501 +                       h_dentry = NULL;
9502 +               }
9503 +       }
9504 +
9505 +out:
9506 +       AuDbgDentry(h_dentry);
9507 +       return h_dentry;
9508 +}
9509 +
9510 +aufs_bindex_t au_dbtail(struct dentry *dentry)
9511 +{
9512 +       aufs_bindex_t bbot, bwh;
9513 +
9514 +       bbot = au_dbbot(dentry);
9515 +       if (0 <= bbot) {
9516 +               bwh = au_dbwh(dentry);
9517 +               if (!bwh)
9518 +                       return bwh;
9519 +               if (0 < bwh && bwh < bbot)
9520 +                       return bwh - 1;
9521 +       }
9522 +       return bbot;
9523 +}
9524 +
9525 +aufs_bindex_t au_dbtaildir(struct dentry *dentry)
9526 +{
9527 +       aufs_bindex_t bbot, bopq;
9528 +
9529 +       bbot = au_dbtail(dentry);
9530 +       if (0 <= bbot) {
9531 +               bopq = au_dbdiropq(dentry);
9532 +               if (0 <= bopq && bopq < bbot)
9533 +                       bbot = bopq;
9534 +       }
9535 +       return bbot;
9536 +}
9537 +
9538 +/* ---------------------------------------------------------------------- */
9539 +
9540 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
9541 +                  struct dentry *h_dentry)
9542 +{
9543 +       struct au_dinfo *dinfo;
9544 +       struct au_hdentry *hd;
9545 +       struct au_branch *br;
9546 +
9547 +       DiMustWriteLock(dentry);
9548 +
9549 +       dinfo = au_di(dentry);
9550 +       hd = au_hdentry(dinfo, bindex);
9551 +       au_hdput(hd);
9552 +       hd->hd_dentry = h_dentry;
9553 +       if (h_dentry) {
9554 +               br = au_sbr(dentry->d_sb, bindex);
9555 +               hd->hd_id = br->br_id;
9556 +       }
9557 +}
9558 +
9559 +int au_dbrange_test(struct dentry *dentry)
9560 +{
9561 +       int err;
9562 +       aufs_bindex_t btop, bbot;
9563 +
9564 +       err = 0;
9565 +       btop = au_dbtop(dentry);
9566 +       bbot = au_dbbot(dentry);
9567 +       if (btop >= 0)
9568 +               AuDebugOn(bbot < 0 && btop > bbot);
9569 +       else {
9570 +               err = -EIO;
9571 +               AuDebugOn(bbot >= 0);
9572 +       }
9573 +
9574 +       return err;
9575 +}
9576 +
9577 +int au_digen_test(struct dentry *dentry, unsigned int sigen)
9578 +{
9579 +       int err;
9580 +
9581 +       err = 0;
9582 +       if (unlikely(au_digen(dentry) != sigen
9583 +                    || au_iigen_test(d_inode(dentry), sigen)))
9584 +               err = -EIO;
9585 +
9586 +       return err;
9587 +}
9588 +
9589 +void au_update_digen(struct dentry *dentry)
9590 +{
9591 +       atomic_set(&au_di(dentry)->di_generation, au_sigen(dentry->d_sb));
9592 +       /* smp_mb(); */ /* atomic_set */
9593 +}
9594 +
9595 +void au_update_dbrange(struct dentry *dentry, int do_put_zero)
9596 +{
9597 +       struct au_dinfo *dinfo;
9598 +       struct dentry *h_d;
9599 +       struct au_hdentry *hdp;
9600 +       aufs_bindex_t bindex, bbot;
9601 +
9602 +       DiMustWriteLock(dentry);
9603 +
9604 +       dinfo = au_di(dentry);
9605 +       if (!dinfo || dinfo->di_btop < 0)
9606 +               return;
9607 +
9608 +       if (do_put_zero) {
9609 +               bbot = dinfo->di_bbot;
9610 +               bindex = dinfo->di_btop;
9611 +               hdp = au_hdentry(dinfo, bindex);
9612 +               for (; bindex <= bbot; bindex++, hdp++) {
9613 +                       h_d = hdp->hd_dentry;
9614 +                       if (h_d && d_is_negative(h_d))
9615 +                               au_set_h_dptr(dentry, bindex, NULL);
9616 +               }
9617 +       }
9618 +
9619 +       dinfo->di_btop = 0;
9620 +       hdp = au_hdentry(dinfo, dinfo->di_btop);
9621 +       for (; dinfo->di_btop <= dinfo->di_bbot; dinfo->di_btop++, hdp++)
9622 +               if (hdp->hd_dentry)
9623 +                       break;
9624 +       if (dinfo->di_btop > dinfo->di_bbot) {
9625 +               dinfo->di_btop = -1;
9626 +               dinfo->di_bbot = -1;
9627 +               return;
9628 +       }
9629 +
9630 +       hdp = au_hdentry(dinfo, dinfo->di_bbot);
9631 +       for (; dinfo->di_bbot >= 0; dinfo->di_bbot--, hdp--)
9632 +               if (hdp->hd_dentry)
9633 +                       break;
9634 +       AuDebugOn(dinfo->di_btop > dinfo->di_bbot || dinfo->di_bbot < 0);
9635 +}
9636 +
9637 +void au_update_dbtop(struct dentry *dentry)
9638 +{
9639 +       aufs_bindex_t bindex, bbot;
9640 +       struct dentry *h_dentry;
9641 +
9642 +       bbot = au_dbbot(dentry);
9643 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++) {
9644 +               h_dentry = au_h_dptr(dentry, bindex);
9645 +               if (!h_dentry)
9646 +                       continue;
9647 +               if (d_is_positive(h_dentry)) {
9648 +                       au_set_dbtop(dentry, bindex);
9649 +                       return;
9650 +               }
9651 +               au_set_h_dptr(dentry, bindex, NULL);
9652 +       }
9653 +}
9654 +
9655 +void au_update_dbbot(struct dentry *dentry)
9656 +{
9657 +       aufs_bindex_t bindex, btop;
9658 +       struct dentry *h_dentry;
9659 +
9660 +       btop = au_dbtop(dentry);
9661 +       for (bindex = au_dbbot(dentry); bindex >= btop; bindex--) {
9662 +               h_dentry = au_h_dptr(dentry, bindex);
9663 +               if (!h_dentry)
9664 +                       continue;
9665 +               if (d_is_positive(h_dentry)) {
9666 +                       au_set_dbbot(dentry, bindex);
9667 +                       return;
9668 +               }
9669 +               au_set_h_dptr(dentry, bindex, NULL);
9670 +       }
9671 +}
9672 +
9673 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry)
9674 +{
9675 +       aufs_bindex_t bindex, bbot;
9676 +
9677 +       bbot = au_dbbot(dentry);
9678 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++)
9679 +               if (au_h_dptr(dentry, bindex) == h_dentry)
9680 +                       return bindex;
9681 +       return -1;
9682 +}
9683 diff -urN /usr/share/empty/fs/aufs/dir.c linux/fs/aufs/dir.c
9684 --- /usr/share/empty/fs/aufs/dir.c      1970-01-01 01:00:00.000000000 +0100
9685 +++ linux/fs/aufs/dir.c 2017-05-06 22:16:52.778220093 +0200
9686 @@ -0,0 +1,762 @@
9687 +/*
9688 + * Copyright (C) 2005-2017 Junjiro R. Okajima
9689 + *
9690 + * This program, aufs is free software; you can redistribute it and/or modify
9691 + * it under the terms of the GNU General Public License as published by
9692 + * the Free Software Foundation; either version 2 of the License, or
9693 + * (at your option) any later version.
9694 + *
9695 + * This program is distributed in the hope that it will be useful,
9696 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9697 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9698 + * GNU General Public License for more details.
9699 + *
9700 + * You should have received a copy of the GNU General Public License
9701 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9702 + */
9703 +
9704 +/*
9705 + * directory operations
9706 + */
9707 +
9708 +#include <linux/fs_stack.h>
9709 +#include "aufs.h"
9710 +
9711 +void au_add_nlink(struct inode *dir, struct inode *h_dir)
9712 +{
9713 +       unsigned int nlink;
9714 +
9715 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9716 +
9717 +       nlink = dir->i_nlink;
9718 +       nlink += h_dir->i_nlink - 2;
9719 +       if (h_dir->i_nlink < 2)
9720 +               nlink += 2;
9721 +       smp_mb(); /* for i_nlink */
9722 +       /* 0 can happen in revaliding */
9723 +       set_nlink(dir, nlink);
9724 +}
9725 +
9726 +void au_sub_nlink(struct inode *dir, struct inode *h_dir)
9727 +{
9728 +       unsigned int nlink;
9729 +
9730 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9731 +
9732 +       nlink = dir->i_nlink;
9733 +       nlink -= h_dir->i_nlink - 2;
9734 +       if (h_dir->i_nlink < 2)
9735 +               nlink -= 2;
9736 +       smp_mb(); /* for i_nlink */
9737 +       /* nlink == 0 means the branch-fs is broken */
9738 +       set_nlink(dir, nlink);
9739 +}
9740 +
9741 +loff_t au_dir_size(struct file *file, struct dentry *dentry)
9742 +{
9743 +       loff_t sz;
9744 +       aufs_bindex_t bindex, bbot;
9745 +       struct file *h_file;
9746 +       struct dentry *h_dentry;
9747 +
9748 +       sz = 0;
9749 +       if (file) {
9750 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
9751 +
9752 +               bbot = au_fbbot_dir(file);
9753 +               for (bindex = au_fbtop(file);
9754 +                    bindex <= bbot && sz < KMALLOC_MAX_SIZE;
9755 +                    bindex++) {
9756 +                       h_file = au_hf_dir(file, bindex);
9757 +                       if (h_file && file_inode(h_file))
9758 +                               sz += vfsub_f_size_read(h_file);
9759 +               }
9760 +       } else {
9761 +               AuDebugOn(!dentry);
9762 +               AuDebugOn(!d_is_dir(dentry));
9763 +
9764 +               bbot = au_dbtaildir(dentry);
9765 +               for (bindex = au_dbtop(dentry);
9766 +                    bindex <= bbot && sz < KMALLOC_MAX_SIZE;
9767 +                    bindex++) {
9768 +                       h_dentry = au_h_dptr(dentry, bindex);
9769 +                       if (h_dentry && d_is_positive(h_dentry))
9770 +                               sz += i_size_read(d_inode(h_dentry));
9771 +               }
9772 +       }
9773 +       if (sz < KMALLOC_MAX_SIZE)
9774 +               sz = roundup_pow_of_two(sz);
9775 +       if (sz > KMALLOC_MAX_SIZE)
9776 +               sz = KMALLOC_MAX_SIZE;
9777 +       else if (sz < NAME_MAX) {
9778 +               BUILD_BUG_ON(AUFS_RDBLK_DEF < NAME_MAX);
9779 +               sz = AUFS_RDBLK_DEF;
9780 +       }
9781 +       return sz;
9782 +}
9783 +
9784 +struct au_dir_ts_arg {
9785 +       struct dentry *dentry;
9786 +       aufs_bindex_t brid;
9787 +};
9788 +
9789 +static void au_do_dir_ts(void *arg)
9790 +{
9791 +       struct au_dir_ts_arg *a = arg;
9792 +       struct au_dtime dt;
9793 +       struct path h_path;
9794 +       struct inode *dir, *h_dir;
9795 +       struct super_block *sb;
9796 +       struct au_branch *br;
9797 +       struct au_hinode *hdir;
9798 +       int err;
9799 +       aufs_bindex_t btop, bindex;
9800 +
9801 +       sb = a->dentry->d_sb;
9802 +       if (d_really_is_negative(a->dentry))
9803 +               goto out;
9804 +       /* no dir->i_mutex lock */
9805 +       aufs_read_lock(a->dentry, AuLock_DW); /* noflush */
9806 +
9807 +       dir = d_inode(a->dentry);
9808 +       btop = au_ibtop(dir);
9809 +       bindex = au_br_index(sb, a->brid);
9810 +       if (bindex < btop)
9811 +               goto out_unlock;
9812 +
9813 +       br = au_sbr(sb, bindex);
9814 +       h_path.dentry = au_h_dptr(a->dentry, bindex);
9815 +       if (!h_path.dentry)
9816 +               goto out_unlock;
9817 +       h_path.mnt = au_br_mnt(br);
9818 +       au_dtime_store(&dt, a->dentry, &h_path);
9819 +
9820 +       br = au_sbr(sb, btop);
9821 +       if (!au_br_writable(br->br_perm))
9822 +               goto out_unlock;
9823 +       h_path.dentry = au_h_dptr(a->dentry, btop);
9824 +       h_path.mnt = au_br_mnt(br);
9825 +       err = vfsub_mnt_want_write(h_path.mnt);
9826 +       if (err)
9827 +               goto out_unlock;
9828 +       hdir = au_hi(dir, btop);
9829 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
9830 +       h_dir = au_h_iptr(dir, btop);
9831 +       if (h_dir->i_nlink
9832 +           && timespec_compare(&h_dir->i_mtime, &dt.dt_mtime) < 0) {
9833 +               dt.dt_h_path = h_path;
9834 +               au_dtime_revert(&dt);
9835 +       }
9836 +       au_hn_inode_unlock(hdir);
9837 +       vfsub_mnt_drop_write(h_path.mnt);
9838 +       au_cpup_attr_timesizes(dir);
9839 +
9840 +out_unlock:
9841 +       aufs_read_unlock(a->dentry, AuLock_DW);
9842 +out:
9843 +       dput(a->dentry);
9844 +       au_nwt_done(&au_sbi(sb)->si_nowait);
9845 +       au_delayed_kfree(arg);
9846 +}
9847 +
9848 +void au_dir_ts(struct inode *dir, aufs_bindex_t bindex)
9849 +{
9850 +       int perm, wkq_err;
9851 +       aufs_bindex_t btop;
9852 +       struct au_dir_ts_arg *arg;
9853 +       struct dentry *dentry;
9854 +       struct super_block *sb;
9855 +
9856 +       IMustLock(dir);
9857 +
9858 +       dentry = d_find_any_alias(dir);
9859 +       AuDebugOn(!dentry);
9860 +       sb = dentry->d_sb;
9861 +       btop = au_ibtop(dir);
9862 +       if (btop == bindex) {
9863 +               au_cpup_attr_timesizes(dir);
9864 +               goto out;
9865 +       }
9866 +
9867 +       perm = au_sbr_perm(sb, btop);
9868 +       if (!au_br_writable(perm))
9869 +               goto out;
9870 +
9871 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
9872 +       if (!arg)
9873 +               goto out;
9874 +
9875 +       arg->dentry = dget(dentry); /* will be dput-ted by au_do_dir_ts() */
9876 +       arg->brid = au_sbr_id(sb, bindex);
9877 +       wkq_err = au_wkq_nowait(au_do_dir_ts, arg, sb, /*flags*/0);
9878 +       if (unlikely(wkq_err)) {
9879 +               pr_err("wkq %d\n", wkq_err);
9880 +               dput(dentry);
9881 +               au_delayed_kfree(arg);
9882 +       }
9883 +
9884 +out:
9885 +       dput(dentry);
9886 +}
9887 +
9888 +/* ---------------------------------------------------------------------- */
9889 +
9890 +static int reopen_dir(struct file *file)
9891 +{
9892 +       int err;
9893 +       unsigned int flags;
9894 +       aufs_bindex_t bindex, btail, btop;
9895 +       struct dentry *dentry, *h_dentry;
9896 +       struct file *h_file;
9897 +
9898 +       /* open all lower dirs */
9899 +       dentry = file->f_path.dentry;
9900 +       btop = au_dbtop(dentry);
9901 +       for (bindex = au_fbtop(file); bindex < btop; bindex++)
9902 +               au_set_h_fptr(file, bindex, NULL);
9903 +       au_set_fbtop(file, btop);
9904 +
9905 +       btail = au_dbtaildir(dentry);
9906 +       for (bindex = au_fbbot_dir(file); btail < bindex; bindex--)
9907 +               au_set_h_fptr(file, bindex, NULL);
9908 +       au_set_fbbot_dir(file, btail);
9909 +
9910 +       flags = vfsub_file_flags(file);
9911 +       for (bindex = btop; bindex <= btail; bindex++) {
9912 +               h_dentry = au_h_dptr(dentry, bindex);
9913 +               if (!h_dentry)
9914 +                       continue;
9915 +               h_file = au_hf_dir(file, bindex);
9916 +               if (h_file)
9917 +                       continue;
9918 +
9919 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9920 +               err = PTR_ERR(h_file);
9921 +               if (IS_ERR(h_file))
9922 +                       goto out; /* close all? */
9923 +               au_set_h_fptr(file, bindex, h_file);
9924 +       }
9925 +       au_update_figen(file);
9926 +       /* todo: necessary? */
9927 +       /* file->f_ra = h_file->f_ra; */
9928 +       err = 0;
9929 +
9930 +out:
9931 +       return err;
9932 +}
9933 +
9934 +static int do_open_dir(struct file *file, int flags, struct file *h_file)
9935 +{
9936 +       int err;
9937 +       aufs_bindex_t bindex, btail;
9938 +       struct dentry *dentry, *h_dentry;
9939 +       struct vfsmount *mnt;
9940 +
9941 +       FiMustWriteLock(file);
9942 +       AuDebugOn(h_file);
9943 +
9944 +       err = 0;
9945 +       mnt = file->f_path.mnt;
9946 +       dentry = file->f_path.dentry;
9947 +       file->f_version = d_inode(dentry)->i_version;
9948 +       bindex = au_dbtop(dentry);
9949 +       au_set_fbtop(file, bindex);
9950 +       btail = au_dbtaildir(dentry);
9951 +       au_set_fbbot_dir(file, btail);
9952 +       for (; !err && bindex <= btail; bindex++) {
9953 +               h_dentry = au_h_dptr(dentry, bindex);
9954 +               if (!h_dentry)
9955 +                       continue;
9956 +
9957 +               err = vfsub_test_mntns(mnt, h_dentry->d_sb);
9958 +               if (unlikely(err))
9959 +                       break;
9960 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9961 +               if (IS_ERR(h_file)) {
9962 +                       err = PTR_ERR(h_file);
9963 +                       break;
9964 +               }
9965 +               au_set_h_fptr(file, bindex, h_file);
9966 +       }
9967 +       au_update_figen(file);
9968 +       /* todo: necessary? */
9969 +       /* file->f_ra = h_file->f_ra; */
9970 +       if (!err)
9971 +               return 0; /* success */
9972 +
9973 +       /* close all */
9974 +       for (bindex = au_fbtop(file); bindex <= btail; bindex++)
9975 +               au_set_h_fptr(file, bindex, NULL);
9976 +       au_set_fbtop(file, -1);
9977 +       au_set_fbbot_dir(file, -1);
9978 +
9979 +       return err;
9980 +}
9981 +
9982 +static int aufs_open_dir(struct inode *inode __maybe_unused,
9983 +                        struct file *file)
9984 +{
9985 +       int err;
9986 +       struct super_block *sb;
9987 +       struct au_fidir *fidir;
9988 +
9989 +       err = -ENOMEM;
9990 +       sb = file->f_path.dentry->d_sb;
9991 +       si_read_lock(sb, AuLock_FLUSH);
9992 +       fidir = au_fidir_alloc(sb);
9993 +       if (fidir) {
9994 +               struct au_do_open_args args = {
9995 +                       .open   = do_open_dir,
9996 +                       .fidir  = fidir
9997 +               };
9998 +               err = au_do_open(file, &args);
9999 +               if (unlikely(err))
10000 +                       au_delayed_kfree(fidir);
10001 +       }
10002 +       si_read_unlock(sb);
10003 +       return err;
10004 +}
10005 +
10006 +static int aufs_release_dir(struct inode *inode __maybe_unused,
10007 +                           struct file *file)
10008 +{
10009 +       struct au_vdir *vdir_cache;
10010 +       struct au_finfo *finfo;
10011 +       struct au_fidir *fidir;
10012 +       struct au_hfile *hf;
10013 +       aufs_bindex_t bindex, bbot;
10014 +       int execed, delayed;
10015 +
10016 +       delayed = (current->flags & PF_KTHREAD) || in_interrupt();
10017 +       finfo = au_fi(file);
10018 +       fidir = finfo->fi_hdir;
10019 +       if (fidir) {
10020 +               au_sphl_del(&finfo->fi_hlist,
10021 +                           &au_sbi(file->f_path.dentry->d_sb)->si_files);
10022 +               vdir_cache = fidir->fd_vdir_cache; /* lock-free */
10023 +               if (vdir_cache)
10024 +                       au_vdir_free(vdir_cache, delayed);
10025 +
10026 +               bindex = finfo->fi_btop;
10027 +               if (bindex >= 0) {
10028 +                       execed = vfsub_file_execed(file);
10029 +                       hf = fidir->fd_hfile + bindex;
10030 +                       /*
10031 +                        * calls fput() instead of filp_close(),
10032 +                        * since no dnotify or lock for the lower file.
10033 +                        */
10034 +                       bbot = fidir->fd_bbot;
10035 +                       for (; bindex <= bbot; bindex++, hf++)
10036 +                               if (hf->hf_file)
10037 +                                       au_hfput(hf, execed);
10038 +               }
10039 +               au_delayed_kfree(fidir);
10040 +               finfo->fi_hdir = NULL;
10041 +       }
10042 +       au_finfo_fin(file, delayed);
10043 +       return 0;
10044 +}
10045 +
10046 +/* ---------------------------------------------------------------------- */
10047 +
10048 +static int au_do_flush_dir(struct file *file, fl_owner_t id)
10049 +{
10050 +       int err;
10051 +       aufs_bindex_t bindex, bbot;
10052 +       struct file *h_file;
10053 +
10054 +       err = 0;
10055 +       bbot = au_fbbot_dir(file);
10056 +       for (bindex = au_fbtop(file); !err && bindex <= bbot; bindex++) {
10057 +               h_file = au_hf_dir(file, bindex);
10058 +               if (h_file)
10059 +                       err = vfsub_flush(h_file, id);
10060 +       }
10061 +       return err;
10062 +}
10063 +
10064 +static int aufs_flush_dir(struct file *file, fl_owner_t id)
10065 +{
10066 +       return au_do_flush(file, id, au_do_flush_dir);
10067 +}
10068 +
10069 +/* ---------------------------------------------------------------------- */
10070 +
10071 +static int au_do_fsync_dir_no_file(struct dentry *dentry, int datasync)
10072 +{
10073 +       int err;
10074 +       aufs_bindex_t bbot, bindex;
10075 +       struct inode *inode;
10076 +       struct super_block *sb;
10077 +
10078 +       err = 0;
10079 +       sb = dentry->d_sb;
10080 +       inode = d_inode(dentry);
10081 +       IMustLock(inode);
10082 +       bbot = au_dbbot(dentry);
10083 +       for (bindex = au_dbtop(dentry); !err && bindex <= bbot; bindex++) {
10084 +               struct path h_path;
10085 +
10086 +               if (au_test_ro(sb, bindex, inode))
10087 +                       continue;
10088 +               h_path.dentry = au_h_dptr(dentry, bindex);
10089 +               if (!h_path.dentry)
10090 +                       continue;
10091 +
10092 +               h_path.mnt = au_sbr_mnt(sb, bindex);
10093 +               err = vfsub_fsync(NULL, &h_path, datasync);
10094 +       }
10095 +
10096 +       return err;
10097 +}
10098 +
10099 +static int au_do_fsync_dir(struct file *file, int datasync)
10100 +{
10101 +       int err;
10102 +       aufs_bindex_t bbot, bindex;
10103 +       struct file *h_file;
10104 +       struct super_block *sb;
10105 +       struct inode *inode;
10106 +
10107 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1, /*fi_lsc*/0);
10108 +       if (unlikely(err))
10109 +               goto out;
10110 +
10111 +       inode = file_inode(file);
10112 +       sb = inode->i_sb;
10113 +       bbot = au_fbbot_dir(file);
10114 +       for (bindex = au_fbtop(file); !err && bindex <= bbot; bindex++) {
10115 +               h_file = au_hf_dir(file, bindex);
10116 +               if (!h_file || au_test_ro(sb, bindex, inode))
10117 +                       continue;
10118 +
10119 +               err = vfsub_fsync(h_file, &h_file->f_path, datasync);
10120 +       }
10121 +
10122 +out:
10123 +       return err;
10124 +}
10125 +
10126 +/*
10127 + * @file may be NULL
10128 + */
10129 +static int aufs_fsync_dir(struct file *file, loff_t start, loff_t end,
10130 +                         int datasync)
10131 +{
10132 +       int err;
10133 +       struct dentry *dentry;
10134 +       struct inode *inode;
10135 +       struct super_block *sb;
10136 +
10137 +       err = 0;
10138 +       dentry = file->f_path.dentry;
10139 +       inode = d_inode(dentry);
10140 +       inode_lock(inode);
10141 +       sb = dentry->d_sb;
10142 +       si_noflush_read_lock(sb);
10143 +       if (file)
10144 +               err = au_do_fsync_dir(file, datasync);
10145 +       else {
10146 +               di_write_lock_child(dentry);
10147 +               err = au_do_fsync_dir_no_file(dentry, datasync);
10148 +       }
10149 +       au_cpup_attr_timesizes(inode);
10150 +       di_write_unlock(dentry);
10151 +       if (file)
10152 +               fi_write_unlock(file);
10153 +
10154 +       si_read_unlock(sb);
10155 +       inode_unlock(inode);
10156 +       return err;
10157 +}
10158 +
10159 +/* ---------------------------------------------------------------------- */
10160 +
10161 +static int aufs_iterate_shared(struct file *file, struct dir_context *ctx)
10162 +{
10163 +       int err;
10164 +       struct dentry *dentry;
10165 +       struct inode *inode, *h_inode;
10166 +       struct super_block *sb;
10167 +
10168 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
10169 +
10170 +       dentry = file->f_path.dentry;
10171 +       inode = d_inode(dentry);
10172 +       IMustLock(inode);
10173 +
10174 +       sb = dentry->d_sb;
10175 +       si_read_lock(sb, AuLock_FLUSH);
10176 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1, /*fi_lsc*/0);
10177 +       if (unlikely(err))
10178 +               goto out;
10179 +       err = au_alive_dir(dentry);
10180 +       if (!err)
10181 +               err = au_vdir_init(file);
10182 +       di_downgrade_lock(dentry, AuLock_IR);
10183 +       if (unlikely(err))
10184 +               goto out_unlock;
10185 +
10186 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
10187 +       if (!au_test_nfsd()) {
10188 +               err = au_vdir_fill_de(file, ctx);
10189 +               fsstack_copy_attr_atime(inode, h_inode);
10190 +       } else {
10191 +               /*
10192 +                * nfsd filldir may call lookup_one_len(), vfs_getattr(),
10193 +                * encode_fh() and others.
10194 +                */
10195 +               atomic_inc(&h_inode->i_count);
10196 +               di_read_unlock(dentry, AuLock_IR);
10197 +               si_read_unlock(sb);
10198 +               err = au_vdir_fill_de(file, ctx);
10199 +               fsstack_copy_attr_atime(inode, h_inode);
10200 +               fi_write_unlock(file);
10201 +               iput(h_inode);
10202 +
10203 +               AuTraceErr(err);
10204 +               return err;
10205 +       }
10206 +
10207 +out_unlock:
10208 +       di_read_unlock(dentry, AuLock_IR);
10209 +       fi_write_unlock(file);
10210 +out:
10211 +       si_read_unlock(sb);
10212 +       return err;
10213 +}
10214 +
10215 +/* ---------------------------------------------------------------------- */
10216 +
10217 +#define AuTestEmpty_WHONLY     1
10218 +#define AuTestEmpty_CALLED     (1 << 1)
10219 +#define AuTestEmpty_SHWH       (1 << 2)
10220 +#define au_ftest_testempty(flags, name)        ((flags) & AuTestEmpty_##name)
10221 +#define au_fset_testempty(flags, name) \
10222 +       do { (flags) |= AuTestEmpty_##name; } while (0)
10223 +#define au_fclr_testempty(flags, name) \
10224 +       do { (flags) &= ~AuTestEmpty_##name; } while (0)
10225 +
10226 +#ifndef CONFIG_AUFS_SHWH
10227 +#undef AuTestEmpty_SHWH
10228 +#define AuTestEmpty_SHWH       0
10229 +#endif
10230 +
10231 +struct test_empty_arg {
10232 +       struct dir_context ctx;
10233 +       struct au_nhash *whlist;
10234 +       unsigned int flags;
10235 +       int err;
10236 +       aufs_bindex_t bindex;
10237 +};
10238 +
10239 +static int test_empty_cb(struct dir_context *ctx, const char *__name,
10240 +                        int namelen, loff_t offset __maybe_unused, u64 ino,
10241 +                        unsigned int d_type)
10242 +{
10243 +       struct test_empty_arg *arg = container_of(ctx, struct test_empty_arg,
10244 +                                                 ctx);
10245 +       char *name = (void *)__name;
10246 +
10247 +       arg->err = 0;
10248 +       au_fset_testempty(arg->flags, CALLED);
10249 +       /* smp_mb(); */
10250 +       if (name[0] == '.'
10251 +           && (namelen == 1 || (name[1] == '.' && namelen == 2)))
10252 +               goto out; /* success */
10253 +
10254 +       if (namelen <= AUFS_WH_PFX_LEN
10255 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
10256 +               if (au_ftest_testempty(arg->flags, WHONLY)
10257 +                   && !au_nhash_test_known_wh(arg->whlist, name, namelen))
10258 +                       arg->err = -ENOTEMPTY;
10259 +               goto out;
10260 +       }
10261 +
10262 +       name += AUFS_WH_PFX_LEN;
10263 +       namelen -= AUFS_WH_PFX_LEN;
10264 +       if (!au_nhash_test_known_wh(arg->whlist, name, namelen))
10265 +               arg->err = au_nhash_append_wh
10266 +                       (arg->whlist, name, namelen, ino, d_type, arg->bindex,
10267 +                        au_ftest_testempty(arg->flags, SHWH));
10268 +
10269 +out:
10270 +       /* smp_mb(); */
10271 +       AuTraceErr(arg->err);
10272 +       return arg->err;
10273 +}
10274 +
10275 +static int do_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
10276 +{
10277 +       int err;
10278 +       struct file *h_file;
10279 +
10280 +       h_file = au_h_open(dentry, arg->bindex,
10281 +                          O_RDONLY | O_NONBLOCK | O_DIRECTORY | O_LARGEFILE,
10282 +                          /*file*/NULL, /*force_wr*/0);
10283 +       err = PTR_ERR(h_file);
10284 +       if (IS_ERR(h_file))
10285 +               goto out;
10286 +
10287 +       err = 0;
10288 +       if (!au_opt_test(au_mntflags(dentry->d_sb), UDBA_NONE)
10289 +           && !file_inode(h_file)->i_nlink)
10290 +               goto out_put;
10291 +
10292 +       do {
10293 +               arg->err = 0;
10294 +               au_fclr_testempty(arg->flags, CALLED);
10295 +               /* smp_mb(); */
10296 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
10297 +               if (err >= 0)
10298 +                       err = arg->err;
10299 +       } while (!err && au_ftest_testempty(arg->flags, CALLED));
10300 +
10301 +out_put:
10302 +       fput(h_file);
10303 +       au_sbr_put(dentry->d_sb, arg->bindex);
10304 +out:
10305 +       return err;
10306 +}
10307 +
10308 +struct do_test_empty_args {
10309 +       int *errp;
10310 +       struct dentry *dentry;
10311 +       struct test_empty_arg *arg;
10312 +};
10313 +
10314 +static void call_do_test_empty(void *args)
10315 +{
10316 +       struct do_test_empty_args *a = args;
10317 +       *a->errp = do_test_empty(a->dentry, a->arg);
10318 +}
10319 +
10320 +static int sio_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
10321 +{
10322 +       int err, wkq_err;
10323 +       struct dentry *h_dentry;
10324 +       struct inode *h_inode;
10325 +
10326 +       h_dentry = au_h_dptr(dentry, arg->bindex);
10327 +       h_inode = d_inode(h_dentry);
10328 +       /* todo: i_mode changes anytime? */
10329 +       inode_lock_nested(h_inode, AuLsc_I_CHILD);
10330 +       err = au_test_h_perm_sio(h_inode, MAY_EXEC | MAY_READ);
10331 +       inode_unlock(h_inode);
10332 +       if (!err)
10333 +               err = do_test_empty(dentry, arg);
10334 +       else {
10335 +               struct do_test_empty_args args = {
10336 +                       .errp   = &err,
10337 +                       .dentry = dentry,
10338 +                       .arg    = arg
10339 +               };
10340 +               unsigned int flags = arg->flags;
10341 +
10342 +               wkq_err = au_wkq_wait(call_do_test_empty, &args);
10343 +               if (unlikely(wkq_err))
10344 +                       err = wkq_err;
10345 +               arg->flags = flags;
10346 +       }
10347 +
10348 +       return err;
10349 +}
10350 +
10351 +int au_test_empty_lower(struct dentry *dentry)
10352 +{
10353 +       int err;
10354 +       unsigned int rdhash;
10355 +       aufs_bindex_t bindex, btop, btail;
10356 +       struct au_nhash whlist;
10357 +       struct test_empty_arg arg = {
10358 +               .ctx = {
10359 +                       .actor = test_empty_cb
10360 +               }
10361 +       };
10362 +       int (*test_empty)(struct dentry *dentry, struct test_empty_arg *arg);
10363 +
10364 +       SiMustAnyLock(dentry->d_sb);
10365 +
10366 +       rdhash = au_sbi(dentry->d_sb)->si_rdhash;
10367 +       if (!rdhash)
10368 +               rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, dentry));
10369 +       err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
10370 +       if (unlikely(err))
10371 +               goto out;
10372 +
10373 +       arg.flags = 0;
10374 +       arg.whlist = &whlist;
10375 +       btop = au_dbtop(dentry);
10376 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
10377 +               au_fset_testempty(arg.flags, SHWH);
10378 +       test_empty = do_test_empty;
10379 +       if (au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1))
10380 +               test_empty = sio_test_empty;
10381 +       arg.bindex = btop;
10382 +       err = test_empty(dentry, &arg);
10383 +       if (unlikely(err))
10384 +               goto out_whlist;
10385 +
10386 +       au_fset_testempty(arg.flags, WHONLY);
10387 +       btail = au_dbtaildir(dentry);
10388 +       for (bindex = btop + 1; !err && bindex <= btail; bindex++) {
10389 +               struct dentry *h_dentry;
10390 +
10391 +               h_dentry = au_h_dptr(dentry, bindex);
10392 +               if (h_dentry && d_is_positive(h_dentry)) {
10393 +                       arg.bindex = bindex;
10394 +                       err = test_empty(dentry, &arg);
10395 +               }
10396 +       }
10397 +
10398 +out_whlist:
10399 +       au_nhash_wh_free(&whlist);
10400 +out:
10401 +       return err;
10402 +}
10403 +
10404 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist)
10405 +{
10406 +       int err;
10407 +       struct test_empty_arg arg = {
10408 +               .ctx = {
10409 +                       .actor = test_empty_cb
10410 +               }
10411 +       };
10412 +       aufs_bindex_t bindex, btail;
10413 +
10414 +       err = 0;
10415 +       arg.whlist = whlist;
10416 +       arg.flags = AuTestEmpty_WHONLY;
10417 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
10418 +               au_fset_testempty(arg.flags, SHWH);
10419 +       btail = au_dbtaildir(dentry);
10420 +       for (bindex = au_dbtop(dentry); !err && bindex <= btail; bindex++) {
10421 +               struct dentry *h_dentry;
10422 +
10423 +               h_dentry = au_h_dptr(dentry, bindex);
10424 +               if (h_dentry && d_is_positive(h_dentry)) {
10425 +                       arg.bindex = bindex;
10426 +                       err = sio_test_empty(dentry, &arg);
10427 +               }
10428 +       }
10429 +
10430 +       return err;
10431 +}
10432 +
10433 +/* ---------------------------------------------------------------------- */
10434 +
10435 +const struct file_operations aufs_dir_fop = {
10436 +       .owner          = THIS_MODULE,
10437 +       .llseek         = default_llseek,
10438 +       .read           = generic_read_dir,
10439 +       .iterate_shared = aufs_iterate_shared,
10440 +       .unlocked_ioctl = aufs_ioctl_dir,
10441 +#ifdef CONFIG_COMPAT
10442 +       .compat_ioctl   = aufs_compat_ioctl_dir,
10443 +#endif
10444 +       .open           = aufs_open_dir,
10445 +       .release        = aufs_release_dir,
10446 +       .flush          = aufs_flush_dir,
10447 +       .fsync          = aufs_fsync_dir
10448 +};
10449 diff -urN /usr/share/empty/fs/aufs/dir.h linux/fs/aufs/dir.h
10450 --- /usr/share/empty/fs/aufs/dir.h      1970-01-01 01:00:00.000000000 +0100
10451 +++ linux/fs/aufs/dir.h 2017-05-06 22:16:52.778220093 +0200
10452 @@ -0,0 +1,137 @@
10453 +/*
10454 + * Copyright (C) 2005-2017 Junjiro R. Okajima
10455 + *
10456 + * This program, aufs is free software; you can redistribute it and/or modify
10457 + * it under the terms of the GNU General Public License as published by
10458 + * the Free Software Foundation; either version 2 of the License, or
10459 + * (at your option) any later version.
10460 + *
10461 + * This program is distributed in the hope that it will be useful,
10462 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10463 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10464 + * GNU General Public License for more details.
10465 + *
10466 + * You should have received a copy of the GNU General Public License
10467 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10468 + */
10469 +
10470 +/*
10471 + * directory operations
10472 + */
10473 +
10474 +#ifndef __AUFS_DIR_H__
10475 +#define __AUFS_DIR_H__
10476 +
10477 +#ifdef __KERNEL__
10478 +
10479 +#include <linux/fs.h>
10480 +
10481 +/* ---------------------------------------------------------------------- */
10482 +
10483 +/* need to be faster and smaller */
10484 +
10485 +struct au_nhash {
10486 +       unsigned int            nh_num;
10487 +       struct hlist_head       *nh_head;
10488 +};
10489 +
10490 +struct au_vdir_destr {
10491 +       unsigned char   len;
10492 +       unsigned char   name[0];
10493 +} __packed;
10494 +
10495 +struct au_vdir_dehstr {
10496 +       struct hlist_node       hash;
10497 +       union {
10498 +               struct au_vdir_destr    *str;
10499 +               struct llist_node       lnode;  /* delayed free */
10500 +       };
10501 +} ____cacheline_aligned_in_smp;
10502 +
10503 +struct au_vdir_de {
10504 +       ino_t                   de_ino;
10505 +       unsigned char           de_type;
10506 +       /* caution: packed */
10507 +       struct au_vdir_destr    de_str;
10508 +} __packed;
10509 +
10510 +struct au_vdir_wh {
10511 +       struct hlist_node       wh_hash;
10512 +#ifdef CONFIG_AUFS_SHWH
10513 +       ino_t                   wh_ino;
10514 +       aufs_bindex_t           wh_bindex;
10515 +       unsigned char           wh_type;
10516 +#else
10517 +       aufs_bindex_t           wh_bindex;
10518 +#endif
10519 +       /* caution: packed */
10520 +       struct au_vdir_destr    wh_str;
10521 +} __packed;
10522 +
10523 +union au_vdir_deblk_p {
10524 +       unsigned char           *deblk;
10525 +       struct au_vdir_de       *de;
10526 +};
10527 +
10528 +struct au_vdir {
10529 +       unsigned char   **vd_deblk;
10530 +       unsigned long   vd_nblk;
10531 +       struct {
10532 +               unsigned long           ul;
10533 +               union au_vdir_deblk_p   p;
10534 +       } vd_last;
10535 +
10536 +       unsigned long   vd_version;
10537 +       unsigned int    vd_deblk_sz;
10538 +       union {
10539 +               unsigned long           vd_jiffy;
10540 +               struct llist_node       vd_lnode;       /* delayed free */
10541 +       };
10542 +} ____cacheline_aligned_in_smp;
10543 +
10544 +/* ---------------------------------------------------------------------- */
10545 +
10546 +/* dir.c */
10547 +extern const struct file_operations aufs_dir_fop;
10548 +void au_add_nlink(struct inode *dir, struct inode *h_dir);
10549 +void au_sub_nlink(struct inode *dir, struct inode *h_dir);
10550 +loff_t au_dir_size(struct file *file, struct dentry *dentry);
10551 +void au_dir_ts(struct inode *dir, aufs_bindex_t bsrc);
10552 +int au_test_empty_lower(struct dentry *dentry);
10553 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist);
10554 +
10555 +/* vdir.c */
10556 +unsigned int au_rdhash_est(loff_t sz);
10557 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp);
10558 +void au_nhash_wh_free(struct au_nhash *whlist);
10559 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
10560 +                           int limit);
10561 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen);
10562 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
10563 +                      unsigned int d_type, aufs_bindex_t bindex,
10564 +                      unsigned char shwh);
10565 +void au_vdir_free(struct au_vdir *vdir, int atonce);
10566 +int au_vdir_init(struct file *file);
10567 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx);
10568 +
10569 +/* ioctl.c */
10570 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg);
10571 +
10572 +#ifdef CONFIG_AUFS_RDU
10573 +/* rdu.c */
10574 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
10575 +#ifdef CONFIG_COMPAT
10576 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd,
10577 +                        unsigned long arg);
10578 +#endif
10579 +#else
10580 +AuStub(long, au_rdu_ioctl, return -EINVAL, struct file *file,
10581 +       unsigned int cmd, unsigned long arg)
10582 +#ifdef CONFIG_COMPAT
10583 +AuStub(long, au_rdu_compat_ioctl, return -EINVAL, struct file *file,
10584 +       unsigned int cmd, unsigned long arg)
10585 +#endif
10586 +#endif
10587 +
10588 +#endif /* __KERNEL__ */
10589 +#endif /* __AUFS_DIR_H__ */
10590 diff -urN /usr/share/empty/fs/aufs/dynop.c linux/fs/aufs/dynop.c
10591 --- /usr/share/empty/fs/aufs/dynop.c    1970-01-01 01:00:00.000000000 +0100
10592 +++ linux/fs/aufs/dynop.c       2017-05-06 22:16:52.778220093 +0200
10593 @@ -0,0 +1,371 @@
10594 +/*
10595 + * Copyright (C) 2010-2017 Junjiro R. Okajima
10596 + *
10597 + * This program, aufs is free software; you can redistribute it and/or modify
10598 + * it under the terms of the GNU General Public License as published by
10599 + * the Free Software Foundation; either version 2 of the License, or
10600 + * (at your option) any later version.
10601 + *
10602 + * This program is distributed in the hope that it will be useful,
10603 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10604 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10605 + * GNU General Public License for more details.
10606 + *
10607 + * You should have received a copy of the GNU General Public License
10608 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10609 + */
10610 +
10611 +/*
10612 + * dynamically customizable operations for regular files
10613 + */
10614 +
10615 +#include "aufs.h"
10616 +
10617 +#define DyPrSym(key)   AuDbgSym(key->dk_op.dy_hop)
10618 +
10619 +/*
10620 + * How large will these lists be?
10621 + * Usually just a few elements, 20-30 at most for each, I guess.
10622 + */
10623 +static struct au_sphlhead dynop[AuDyLast];
10624 +
10625 +static struct au_dykey *dy_gfind_get(struct au_sphlhead *sphl, const void *h_op)
10626 +{
10627 +       struct au_dykey *key, *tmp;
10628 +       struct hlist_head *head;
10629 +
10630 +       key = NULL;
10631 +       head = &sphl->head;
10632 +       rcu_read_lock();
10633 +       hlist_for_each_entry_rcu(tmp, head, dk_hnode)
10634 +               if (tmp->dk_op.dy_hop == h_op) {
10635 +                       key = tmp;
10636 +                       kref_get(&key->dk_kref);
10637 +                       break;
10638 +               }
10639 +       rcu_read_unlock();
10640 +
10641 +       return key;
10642 +}
10643 +
10644 +static struct au_dykey *dy_bradd(struct au_branch *br, struct au_dykey *key)
10645 +{
10646 +       struct au_dykey **k, *found;
10647 +       const void *h_op = key->dk_op.dy_hop;
10648 +       int i;
10649 +
10650 +       found = NULL;
10651 +       k = br->br_dykey;
10652 +       for (i = 0; i < AuBrDynOp; i++)
10653 +               if (k[i]) {
10654 +                       if (k[i]->dk_op.dy_hop == h_op) {
10655 +                               found = k[i];
10656 +                               break;
10657 +                       }
10658 +               } else
10659 +                       break;
10660 +       if (!found) {
10661 +               spin_lock(&br->br_dykey_lock);
10662 +               for (; i < AuBrDynOp; i++)
10663 +                       if (k[i]) {
10664 +                               if (k[i]->dk_op.dy_hop == h_op) {
10665 +                                       found = k[i];
10666 +                                       break;
10667 +                               }
10668 +                       } else {
10669 +                               k[i] = key;
10670 +                               break;
10671 +                       }
10672 +               spin_unlock(&br->br_dykey_lock);
10673 +               BUG_ON(i == AuBrDynOp); /* expand the array */
10674 +       }
10675 +
10676 +       return found;
10677 +}
10678 +
10679 +/* kref_get() if @key is already added */
10680 +static struct au_dykey *dy_gadd(struct au_sphlhead *sphl, struct au_dykey *key)
10681 +{
10682 +       struct au_dykey *tmp, *found;
10683 +       struct hlist_head *head;
10684 +       const void *h_op = key->dk_op.dy_hop;
10685 +
10686 +       found = NULL;
10687 +       head = &sphl->head;
10688 +       spin_lock(&sphl->spin);
10689 +       hlist_for_each_entry(tmp, head, dk_hnode)
10690 +               if (tmp->dk_op.dy_hop == h_op) {
10691 +                       kref_get(&tmp->dk_kref);
10692 +                       found = tmp;
10693 +                       break;
10694 +               }
10695 +       if (!found)
10696 +               hlist_add_head_rcu(&key->dk_hnode, head);
10697 +       spin_unlock(&sphl->spin);
10698 +
10699 +       if (!found)
10700 +               DyPrSym(key);
10701 +       return found;
10702 +}
10703 +
10704 +static void dy_free_rcu(struct rcu_head *rcu)
10705 +{
10706 +       struct au_dykey *key;
10707 +
10708 +       key = container_of(rcu, struct au_dykey, dk_rcu);
10709 +       DyPrSym(key);
10710 +       kfree(key);     /* not delayed */
10711 +}
10712 +
10713 +static void dy_free(struct kref *kref)
10714 +{
10715 +       struct au_dykey *key;
10716 +       struct au_sphlhead *sphl;
10717 +
10718 +       key = container_of(kref, struct au_dykey, dk_kref);
10719 +       sphl = dynop + key->dk_op.dy_type;
10720 +       au_sphl_del_rcu(&key->dk_hnode, sphl);
10721 +       call_rcu(&key->dk_rcu, dy_free_rcu);
10722 +}
10723 +
10724 +void au_dy_put(struct au_dykey *key)
10725 +{
10726 +       kref_put(&key->dk_kref, dy_free);
10727 +}
10728 +
10729 +/* ---------------------------------------------------------------------- */
10730 +
10731 +#define DyDbgSize(cnt, op)     AuDebugOn(cnt != sizeof(op)/sizeof(void *))
10732 +
10733 +#ifdef CONFIG_AUFS_DEBUG
10734 +#define DyDbgDeclare(cnt)      unsigned int cnt = 0
10735 +#define DyDbgInc(cnt)          do { cnt++; } while (0)
10736 +#else
10737 +#define DyDbgDeclare(cnt)      do {} while (0)
10738 +#define DyDbgInc(cnt)          do {} while (0)
10739 +#endif
10740 +
10741 +#define DySet(func, dst, src, h_op, h_sb) do {                         \
10742 +       DyDbgInc(cnt);                                                  \
10743 +       if (h_op->func) {                                               \
10744 +               if (src.func)                                           \
10745 +                       dst.func = src.func;                            \
10746 +               else                                                    \
10747 +                       AuDbg("%s %s\n", au_sbtype(h_sb), #func);       \
10748 +       }                                                               \
10749 +} while (0)
10750 +
10751 +#define DySetForce(func, dst, src) do {                \
10752 +       AuDebugOn(!src.func);                   \
10753 +       DyDbgInc(cnt);                          \
10754 +       dst.func = src.func;                    \
10755 +} while (0)
10756 +
10757 +#define DySetAop(func) \
10758 +       DySet(func, dyaop->da_op, aufs_aop, h_aop, h_sb)
10759 +#define DySetAopForce(func) \
10760 +       DySetForce(func, dyaop->da_op, aufs_aop)
10761 +
10762 +static void dy_aop(struct au_dykey *key, const void *h_op,
10763 +                  struct super_block *h_sb __maybe_unused)
10764 +{
10765 +       struct au_dyaop *dyaop = (void *)key;
10766 +       const struct address_space_operations *h_aop = h_op;
10767 +       DyDbgDeclare(cnt);
10768 +
10769 +       AuDbg("%s\n", au_sbtype(h_sb));
10770 +
10771 +       DySetAop(writepage);
10772 +       DySetAopForce(readpage);        /* force */
10773 +       DySetAop(writepages);
10774 +       DySetAop(set_page_dirty);
10775 +       DySetAop(readpages);
10776 +       DySetAop(write_begin);
10777 +       DySetAop(write_end);
10778 +       DySetAop(bmap);
10779 +       DySetAop(invalidatepage);
10780 +       DySetAop(releasepage);
10781 +       DySetAop(freepage);
10782 +       /* this one will be changed according to an aufs mount option */
10783 +       DySetAop(direct_IO);
10784 +       DySetAop(migratepage);
10785 +       DySetAop(isolate_page);
10786 +       DySetAop(putback_page);
10787 +       DySetAop(launder_page);
10788 +       DySetAop(is_partially_uptodate);
10789 +       DySetAop(is_dirty_writeback);
10790 +       DySetAop(error_remove_page);
10791 +       DySetAop(swap_activate);
10792 +       DySetAop(swap_deactivate);
10793 +
10794 +       DyDbgSize(cnt, *h_aop);
10795 +}
10796 +
10797 +/* ---------------------------------------------------------------------- */
10798 +
10799 +static void dy_bug(struct kref *kref)
10800 +{
10801 +       BUG();
10802 +}
10803 +
10804 +static struct au_dykey *dy_get(struct au_dynop *op, struct au_branch *br)
10805 +{
10806 +       struct au_dykey *key, *old;
10807 +       struct au_sphlhead *sphl;
10808 +       struct op {
10809 +               unsigned int sz;
10810 +               void (*set)(struct au_dykey *key, const void *h_op,
10811 +                           struct super_block *h_sb __maybe_unused);
10812 +       };
10813 +       static const struct op a[] = {
10814 +               [AuDy_AOP] = {
10815 +                       .sz     = sizeof(struct au_dyaop),
10816 +                       .set    = dy_aop
10817 +               }
10818 +       };
10819 +       const struct op *p;
10820 +
10821 +       sphl = dynop + op->dy_type;
10822 +       key = dy_gfind_get(sphl, op->dy_hop);
10823 +       if (key)
10824 +               goto out_add; /* success */
10825 +
10826 +       p = a + op->dy_type;
10827 +       key = kzalloc(p->sz, GFP_NOFS);
10828 +       if (unlikely(!key)) {
10829 +               key = ERR_PTR(-ENOMEM);
10830 +               goto out;
10831 +       }
10832 +
10833 +       key->dk_op.dy_hop = op->dy_hop;
10834 +       kref_init(&key->dk_kref);
10835 +       p->set(key, op->dy_hop, au_br_sb(br));
10836 +       old = dy_gadd(sphl, key);
10837 +       if (old) {
10838 +               au_delayed_kfree(key);
10839 +               key = old;
10840 +       }
10841 +
10842 +out_add:
10843 +       old = dy_bradd(br, key);
10844 +       if (old)
10845 +               /* its ref-count should never be zero here */
10846 +               kref_put(&key->dk_kref, dy_bug);
10847 +out:
10848 +       return key;
10849 +}
10850 +
10851 +/* ---------------------------------------------------------------------- */
10852 +/*
10853 + * Aufs prohibits O_DIRECT by defaut even if the branch supports it.
10854 + * This behaviour is necessary to return an error from open(O_DIRECT) instead
10855 + * of the succeeding I/O. The dio mount option enables O_DIRECT and makes
10856 + * open(O_DIRECT) always succeed, but the succeeding I/O may return an error.
10857 + * See the aufs manual in detail.
10858 + */
10859 +static void dy_adx(struct au_dyaop *dyaop, int do_dx)
10860 +{
10861 +       if (!do_dx)
10862 +               dyaop->da_op.direct_IO = NULL;
10863 +       else
10864 +               dyaop->da_op.direct_IO = aufs_aop.direct_IO;
10865 +}
10866 +
10867 +static struct au_dyaop *dy_aget(struct au_branch *br,
10868 +                               const struct address_space_operations *h_aop,
10869 +                               int do_dx)
10870 +{
10871 +       struct au_dyaop *dyaop;
10872 +       struct au_dynop op;
10873 +
10874 +       op.dy_type = AuDy_AOP;
10875 +       op.dy_haop = h_aop;
10876 +       dyaop = (void *)dy_get(&op, br);
10877 +       if (IS_ERR(dyaop))
10878 +               goto out;
10879 +       dy_adx(dyaop, do_dx);
10880 +
10881 +out:
10882 +       return dyaop;
10883 +}
10884 +
10885 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
10886 +               struct inode *h_inode)
10887 +{
10888 +       int err, do_dx;
10889 +       struct super_block *sb;
10890 +       struct au_branch *br;
10891 +       struct au_dyaop *dyaop;
10892 +
10893 +       AuDebugOn(!S_ISREG(h_inode->i_mode));
10894 +       IiMustWriteLock(inode);
10895 +
10896 +       sb = inode->i_sb;
10897 +       br = au_sbr(sb, bindex);
10898 +       do_dx = !!au_opt_test(au_mntflags(sb), DIO);
10899 +       dyaop = dy_aget(br, h_inode->i_mapping->a_ops, do_dx);
10900 +       err = PTR_ERR(dyaop);
10901 +       if (IS_ERR(dyaop))
10902 +               /* unnecessary to call dy_fput() */
10903 +               goto out;
10904 +
10905 +       err = 0;
10906 +       inode->i_mapping->a_ops = &dyaop->da_op;
10907 +
10908 +out:
10909 +       return err;
10910 +}
10911 +
10912 +/*
10913 + * Is it safe to replace a_ops during the inode/file is in operation?
10914 + * Yes, I hope so.
10915 + */
10916 +int au_dy_irefresh(struct inode *inode)
10917 +{
10918 +       int err;
10919 +       aufs_bindex_t btop;
10920 +       struct inode *h_inode;
10921 +
10922 +       err = 0;
10923 +       if (S_ISREG(inode->i_mode)) {
10924 +               btop = au_ibtop(inode);
10925 +               h_inode = au_h_iptr(inode, btop);
10926 +               err = au_dy_iaop(inode, btop, h_inode);
10927 +       }
10928 +       return err;
10929 +}
10930 +
10931 +void au_dy_arefresh(int do_dx)
10932 +{
10933 +       struct au_sphlhead *sphl;
10934 +       struct hlist_head *head;
10935 +       struct au_dykey *key;
10936 +
10937 +       sphl = dynop + AuDy_AOP;
10938 +       head = &sphl->head;
10939 +       spin_lock(&sphl->spin);
10940 +       hlist_for_each_entry(key, head, dk_hnode)
10941 +               dy_adx((void *)key, do_dx);
10942 +       spin_unlock(&sphl->spin);
10943 +}
10944 +
10945 +/* ---------------------------------------------------------------------- */
10946 +
10947 +void __init au_dy_init(void)
10948 +{
10949 +       int i;
10950 +
10951 +       /* make sure that 'struct au_dykey *' can be any type */
10952 +       BUILD_BUG_ON(offsetof(struct au_dyaop, da_key));
10953 +
10954 +       for (i = 0; i < AuDyLast; i++)
10955 +               au_sphl_init(dynop + i);
10956 +}
10957 +
10958 +void au_dy_fin(void)
10959 +{
10960 +       int i;
10961 +
10962 +       for (i = 0; i < AuDyLast; i++)
10963 +               WARN_ON(!hlist_empty(&dynop[i].head));
10964 +}
10965 diff -urN /usr/share/empty/fs/aufs/dynop.h linux/fs/aufs/dynop.h
10966 --- /usr/share/empty/fs/aufs/dynop.h    1970-01-01 01:00:00.000000000 +0100
10967 +++ linux/fs/aufs/dynop.h       2017-05-06 22:16:52.778220093 +0200
10968 @@ -0,0 +1,74 @@
10969 +/*
10970 + * Copyright (C) 2010-2017 Junjiro R. Okajima
10971 + *
10972 + * This program, aufs is free software; you can redistribute it and/or modify
10973 + * it under the terms of the GNU General Public License as published by
10974 + * the Free Software Foundation; either version 2 of the License, or
10975 + * (at your option) any later version.
10976 + *
10977 + * This program is distributed in the hope that it will be useful,
10978 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10979 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10980 + * GNU General Public License for more details.
10981 + *
10982 + * You should have received a copy of the GNU General Public License
10983 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10984 + */
10985 +
10986 +/*
10987 + * dynamically customizable operations (for regular files only)
10988 + */
10989 +
10990 +#ifndef __AUFS_DYNOP_H__
10991 +#define __AUFS_DYNOP_H__
10992 +
10993 +#ifdef __KERNEL__
10994 +
10995 +#include <linux/fs.h>
10996 +#include <linux/kref.h>
10997 +
10998 +enum {AuDy_AOP, AuDyLast};
10999 +
11000 +struct au_dynop {
11001 +       int                                             dy_type;
11002 +       union {
11003 +               const void                              *dy_hop;
11004 +               const struct address_space_operations   *dy_haop;
11005 +       };
11006 +};
11007 +
11008 +struct au_dykey {
11009 +       union {
11010 +               struct hlist_node       dk_hnode;
11011 +               struct rcu_head         dk_rcu;
11012 +       };
11013 +       struct au_dynop         dk_op;
11014 +
11015 +       /*
11016 +        * during I am in the branch local array, kref is gotten. when the
11017 +        * branch is removed, kref is put.
11018 +        */
11019 +       struct kref             dk_kref;
11020 +};
11021 +
11022 +/* stop unioning since their sizes are very different from each other */
11023 +struct au_dyaop {
11024 +       struct au_dykey                 da_key;
11025 +       struct address_space_operations da_op; /* not const */
11026 +};
11027 +
11028 +/* ---------------------------------------------------------------------- */
11029 +
11030 +/* dynop.c */
11031 +struct au_branch;
11032 +void au_dy_put(struct au_dykey *key);
11033 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
11034 +               struct inode *h_inode);
11035 +int au_dy_irefresh(struct inode *inode);
11036 +void au_dy_arefresh(int do_dio);
11037 +
11038 +void __init au_dy_init(void);
11039 +void au_dy_fin(void);
11040 +
11041 +#endif /* __KERNEL__ */
11042 +#endif /* __AUFS_DYNOP_H__ */
11043 diff -urN /usr/share/empty/fs/aufs/export.c linux/fs/aufs/export.c
11044 --- /usr/share/empty/fs/aufs/export.c   1970-01-01 01:00:00.000000000 +0100
11045 +++ linux/fs/aufs/export.c      2017-05-06 22:16:52.778220093 +0200
11046 @@ -0,0 +1,836 @@
11047 +/*
11048 + * Copyright (C) 2005-2017 Junjiro R. Okajima
11049 + *
11050 + * This program, aufs is free software; you can redistribute it and/or modify
11051 + * it under the terms of the GNU General Public License as published by
11052 + * the Free Software Foundation; either version 2 of the License, or
11053 + * (at your option) any later version.
11054 + *
11055 + * This program is distributed in the hope that it will be useful,
11056 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11057 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11058 + * GNU General Public License for more details.
11059 + *
11060 + * You should have received a copy of the GNU General Public License
11061 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
11062 + */
11063 +
11064 +/*
11065 + * export via nfs
11066 + */
11067 +
11068 +#include <linux/exportfs.h>
11069 +#include <linux/fs_struct.h>
11070 +#include <linux/namei.h>
11071 +#include <linux/nsproxy.h>
11072 +#include <linux/random.h>
11073 +#include <linux/writeback.h>
11074 +#include "aufs.h"
11075 +
11076 +union conv {
11077 +#ifdef CONFIG_AUFS_INO_T_64
11078 +       __u32 a[2];
11079 +#else
11080 +       __u32 a[1];
11081 +#endif
11082 +       ino_t ino;
11083 +};
11084 +
11085 +static ino_t decode_ino(__u32 *a)
11086 +{
11087 +       union conv u;
11088 +
11089 +       BUILD_BUG_ON(sizeof(u.ino) != sizeof(u.a));
11090 +       u.a[0] = a[0];
11091 +#ifdef CONFIG_AUFS_INO_T_64
11092 +       u.a[1] = a[1];
11093 +#endif
11094 +       return u.ino;
11095 +}
11096 +
11097 +static void encode_ino(__u32 *a, ino_t ino)
11098 +{
11099 +       union conv u;
11100 +
11101 +       u.ino = ino;
11102 +       a[0] = u.a[0];
11103 +#ifdef CONFIG_AUFS_INO_T_64
11104 +       a[1] = u.a[1];
11105 +#endif
11106 +}
11107 +
11108 +/* NFS file handle */
11109 +enum {
11110 +       Fh_br_id,
11111 +       Fh_sigen,
11112 +#ifdef CONFIG_AUFS_INO_T_64
11113 +       /* support 64bit inode number */
11114 +       Fh_ino1,
11115 +       Fh_ino2,
11116 +       Fh_dir_ino1,
11117 +       Fh_dir_ino2,
11118 +#else
11119 +       Fh_ino1,
11120 +       Fh_dir_ino1,
11121 +#endif
11122 +       Fh_igen,
11123 +       Fh_h_type,
11124 +       Fh_tail,
11125 +
11126 +       Fh_ino = Fh_ino1,
11127 +       Fh_dir_ino = Fh_dir_ino1
11128 +};
11129 +
11130 +static int au_test_anon(struct dentry *dentry)
11131 +{
11132 +       /* note: read d_flags without d_lock */
11133 +       return !!(dentry->d_flags & DCACHE_DISCONNECTED);
11134 +}
11135 +
11136 +int au_test_nfsd(void)
11137 +{
11138 +       int ret;
11139 +       struct task_struct *tsk = current;
11140 +       char comm[sizeof(tsk->comm)];
11141 +
11142 +       ret = 0;
11143 +       if (tsk->flags & PF_KTHREAD) {
11144 +               get_task_comm(comm, tsk);
11145 +               ret = !strcmp(comm, "nfsd");
11146 +       }
11147 +
11148 +       return ret;
11149 +}
11150 +
11151 +/* ---------------------------------------------------------------------- */
11152 +/* inode generation external table */
11153 +
11154 +void au_xigen_inc(struct inode *inode)
11155 +{
11156 +       loff_t pos;
11157 +       ssize_t sz;
11158 +       __u32 igen;
11159 +       struct super_block *sb;
11160 +       struct au_sbinfo *sbinfo;
11161 +
11162 +       sb = inode->i_sb;
11163 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
11164 +
11165 +       sbinfo = au_sbi(sb);
11166 +       pos = inode->i_ino;
11167 +       pos *= sizeof(igen);
11168 +       igen = inode->i_generation + 1;
11169 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xigen, &igen,
11170 +                        sizeof(igen), &pos);
11171 +       if (sz == sizeof(igen))
11172 +               return; /* success */
11173 +
11174 +       if (unlikely(sz >= 0))
11175 +               AuIOErr("xigen error (%zd)\n", sz);
11176 +}
11177 +
11178 +int au_xigen_new(struct inode *inode)
11179 +{
11180 +       int err;
11181 +       loff_t pos;
11182 +       ssize_t sz;
11183 +       struct super_block *sb;
11184 +       struct au_sbinfo *sbinfo;
11185 +       struct file *file;
11186 +
11187 +       err = 0;
11188 +       /* todo: dirty, at mount time */
11189 +       if (inode->i_ino == AUFS_ROOT_INO)
11190 +               goto out;
11191 +       sb = inode->i_sb;
11192 +       SiMustAnyLock(sb);
11193 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
11194 +               goto out;
11195 +
11196 +       err = -EFBIG;
11197 +       pos = inode->i_ino;
11198 +       if (unlikely(au_loff_max / sizeof(inode->i_generation) - 1 < pos)) {
11199 +               AuIOErr1("too large i%lld\n", pos);
11200 +               goto out;
11201 +       }
11202 +       pos *= sizeof(inode->i_generation);
11203 +
11204 +       err = 0;
11205 +       sbinfo = au_sbi(sb);
11206 +       file = sbinfo->si_xigen;
11207 +       BUG_ON(!file);
11208 +
11209 +       if (vfsub_f_size_read(file)
11210 +           < pos + sizeof(inode->i_generation)) {
11211 +               inode->i_generation = atomic_inc_return(&sbinfo->si_xigen_next);
11212 +               sz = xino_fwrite(sbinfo->si_xwrite, file, &inode->i_generation,
11213 +                                sizeof(inode->i_generation), &pos);
11214 +       } else
11215 +               sz = xino_fread(sbinfo->si_xread, file, &inode->i_generation,
11216 +                               sizeof(inode->i_generation), &pos);
11217 +       if (sz == sizeof(inode->i_generation))
11218 +               goto out; /* success */
11219 +
11220 +       err = sz;
11221 +       if (unlikely(sz >= 0)) {
11222 +               err = -EIO;
11223 +               AuIOErr("xigen error (%zd)\n", sz);
11224 +       }
11225 +
11226 +out:
11227 +       return err;
11228 +}
11229 +
11230 +int au_xigen_set(struct super_block *sb, struct file *base)
11231 +{
11232 +       int err;
11233 +       struct au_sbinfo *sbinfo;
11234 +       struct file *file;
11235 +
11236 +       SiMustWriteLock(sb);
11237 +
11238 +       sbinfo = au_sbi(sb);
11239 +       file = au_xino_create2(base, sbinfo->si_xigen);
11240 +       err = PTR_ERR(file);
11241 +       if (IS_ERR(file))
11242 +               goto out;
11243 +       err = 0;
11244 +       if (sbinfo->si_xigen)
11245 +               fput(sbinfo->si_xigen);
11246 +       sbinfo->si_xigen = file;
11247 +
11248 +out:
11249 +       return err;
11250 +}
11251 +
11252 +void au_xigen_clr(struct super_block *sb)
11253 +{
11254 +       struct au_sbinfo *sbinfo;
11255 +
11256 +       SiMustWriteLock(sb);
11257 +
11258 +       sbinfo = au_sbi(sb);
11259 +       if (sbinfo->si_xigen) {
11260 +               fput(sbinfo->si_xigen);
11261 +               sbinfo->si_xigen = NULL;
11262 +       }
11263 +}
11264 +
11265 +/* ---------------------------------------------------------------------- */
11266 +
11267 +static struct dentry *decode_by_ino(struct super_block *sb, ino_t ino,
11268 +                                   ino_t dir_ino)
11269 +{
11270 +       struct dentry *dentry, *d;
11271 +       struct inode *inode;
11272 +       unsigned int sigen;
11273 +
11274 +       dentry = NULL;
11275 +       inode = ilookup(sb, ino);
11276 +       if (!inode)
11277 +               goto out;
11278 +
11279 +       dentry = ERR_PTR(-ESTALE);
11280 +       sigen = au_sigen(sb);
11281 +       if (unlikely(au_is_bad_inode(inode)
11282 +                    || IS_DEADDIR(inode)
11283 +                    || sigen != au_iigen(inode, NULL)))
11284 +               goto out_iput;
11285 +
11286 +       dentry = NULL;
11287 +       if (!dir_ino || S_ISDIR(inode->i_mode))
11288 +               dentry = d_find_alias(inode);
11289 +       else {
11290 +               spin_lock(&inode->i_lock);
11291 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
11292 +                       spin_lock(&d->d_lock);
11293 +                       if (!au_test_anon(d)
11294 +                           && d_inode(d->d_parent)->i_ino == dir_ino) {
11295 +                               dentry = dget_dlock(d);
11296 +                               spin_unlock(&d->d_lock);
11297 +                               break;
11298 +                       }
11299 +                       spin_unlock(&d->d_lock);
11300 +               }
11301 +               spin_unlock(&inode->i_lock);
11302 +       }
11303 +       if (unlikely(dentry && au_digen_test(dentry, sigen))) {
11304 +               /* need to refresh */
11305 +               dput(dentry);
11306 +               dentry = NULL;
11307 +       }
11308 +
11309 +out_iput:
11310 +       iput(inode);
11311 +out:
11312 +       AuTraceErrPtr(dentry);
11313 +       return dentry;
11314 +}
11315 +
11316 +/* ---------------------------------------------------------------------- */
11317 +
11318 +/* todo: dirty? */
11319 +/* if exportfs_decode_fh() passed vfsmount*, we could be happy */
11320 +
11321 +struct au_compare_mnt_args {
11322 +       /* input */
11323 +       struct super_block *sb;
11324 +
11325 +       /* output */
11326 +       struct vfsmount *mnt;
11327 +};
11328 +
11329 +static int au_compare_mnt(struct vfsmount *mnt, void *arg)
11330 +{
11331 +       struct au_compare_mnt_args *a = arg;
11332 +
11333 +       if (mnt->mnt_sb != a->sb)
11334 +               return 0;
11335 +       a->mnt = mntget(mnt);
11336 +       return 1;
11337 +}
11338 +
11339 +static struct vfsmount *au_mnt_get(struct super_block *sb)
11340 +{
11341 +       int err;
11342 +       struct path root;
11343 +       struct au_compare_mnt_args args = {
11344 +               .sb = sb
11345 +       };
11346 +
11347 +       get_fs_root(current->fs, &root);
11348 +       rcu_read_lock();
11349 +       err = iterate_mounts(au_compare_mnt, &args, root.mnt);
11350 +       rcu_read_unlock();
11351 +       path_put(&root);
11352 +       AuDebugOn(!err);
11353 +       AuDebugOn(!args.mnt);
11354 +       return args.mnt;
11355 +}
11356 +
11357 +struct au_nfsd_si_lock {
11358 +       unsigned int sigen;
11359 +       aufs_bindex_t bindex, br_id;
11360 +       unsigned char force_lock;
11361 +};
11362 +
11363 +static int si_nfsd_read_lock(struct super_block *sb,
11364 +                            struct au_nfsd_si_lock *nsi_lock)
11365 +{
11366 +       int err;
11367 +       aufs_bindex_t bindex;
11368 +
11369 +       si_read_lock(sb, AuLock_FLUSH);
11370 +
11371 +       /* branch id may be wrapped around */
11372 +       err = 0;
11373 +       bindex = au_br_index(sb, nsi_lock->br_id);
11374 +       if (bindex >= 0 && nsi_lock->sigen + AUFS_BRANCH_MAX > au_sigen(sb))
11375 +               goto out; /* success */
11376 +
11377 +       err = -ESTALE;
11378 +       bindex = -1;
11379 +       if (!nsi_lock->force_lock)
11380 +               si_read_unlock(sb);
11381 +
11382 +out:
11383 +       nsi_lock->bindex = bindex;
11384 +       return err;
11385 +}
11386 +
11387 +struct find_name_by_ino {
11388 +       struct dir_context ctx;
11389 +       int called, found;
11390 +       ino_t ino;
11391 +       char *name;
11392 +       int namelen;
11393 +};
11394 +
11395 +static int
11396 +find_name_by_ino(struct dir_context *ctx, const char *name, int namelen,
11397 +                loff_t offset, u64 ino, unsigned int d_type)
11398 +{
11399 +       struct find_name_by_ino *a = container_of(ctx, struct find_name_by_ino,
11400 +                                                 ctx);
11401 +
11402 +       a->called++;
11403 +       if (a->ino != ino)
11404 +               return 0;
11405 +
11406 +       memcpy(a->name, name, namelen);
11407 +       a->namelen = namelen;
11408 +       a->found = 1;
11409 +       return 1;
11410 +}
11411 +
11412 +static struct dentry *au_lkup_by_ino(struct path *path, ino_t ino,
11413 +                                    struct au_nfsd_si_lock *nsi_lock)
11414 +{
11415 +       struct dentry *dentry, *parent;
11416 +       struct file *file;
11417 +       struct inode *dir;
11418 +       struct find_name_by_ino arg = {
11419 +               .ctx = {
11420 +                       .actor = find_name_by_ino
11421 +               }
11422 +       };
11423 +       int err;
11424 +
11425 +       parent = path->dentry;
11426 +       if (nsi_lock)
11427 +               si_read_unlock(parent->d_sb);
11428 +       file = vfsub_dentry_open(path, au_dir_roflags);
11429 +       dentry = (void *)file;
11430 +       if (IS_ERR(file))
11431 +               goto out;
11432 +
11433 +       dentry = ERR_PTR(-ENOMEM);
11434 +       arg.name = (void *)__get_free_page(GFP_NOFS);
11435 +       if (unlikely(!arg.name))
11436 +               goto out_file;
11437 +       arg.ino = ino;
11438 +       arg.found = 0;
11439 +       do {
11440 +               arg.called = 0;
11441 +               /* smp_mb(); */
11442 +               err = vfsub_iterate_dir(file, &arg.ctx);
11443 +       } while (!err && !arg.found && arg.called);
11444 +       dentry = ERR_PTR(err);
11445 +       if (unlikely(err))
11446 +               goto out_name;
11447 +       /* instead of ENOENT */
11448 +       dentry = ERR_PTR(-ESTALE);
11449 +       if (!arg.found)
11450 +               goto out_name;
11451 +
11452 +       /* do not call vfsub_lkup_one() */
11453 +       dir = d_inode(parent);
11454 +       dentry = vfsub_lookup_one_len_unlocked(arg.name, parent, arg.namelen);
11455 +       AuTraceErrPtr(dentry);
11456 +       if (IS_ERR(dentry))
11457 +               goto out_name;
11458 +       AuDebugOn(au_test_anon(dentry));
11459 +       if (unlikely(d_really_is_negative(dentry))) {
11460 +               dput(dentry);
11461 +               dentry = ERR_PTR(-ENOENT);
11462 +       }
11463 +
11464 +out_name:
11465 +       au_delayed_free_page((unsigned long)arg.name);
11466 +out_file:
11467 +       fput(file);
11468 +out:
11469 +       if (unlikely(nsi_lock
11470 +                    && si_nfsd_read_lock(parent->d_sb, nsi_lock) < 0))
11471 +               if (!IS_ERR(dentry)) {
11472 +                       dput(dentry);
11473 +                       dentry = ERR_PTR(-ESTALE);
11474 +               }
11475 +       AuTraceErrPtr(dentry);
11476 +       return dentry;
11477 +}
11478 +
11479 +static struct dentry *decode_by_dir_ino(struct super_block *sb, ino_t ino,
11480 +                                       ino_t dir_ino,
11481 +                                       struct au_nfsd_si_lock *nsi_lock)
11482 +{
11483 +       struct dentry *dentry;
11484 +       struct path path;
11485 +
11486 +       if (dir_ino != AUFS_ROOT_INO) {
11487 +               path.dentry = decode_by_ino(sb, dir_ino, 0);
11488 +               dentry = path.dentry;
11489 +               if (!path.dentry || IS_ERR(path.dentry))
11490 +                       goto out;
11491 +               AuDebugOn(au_test_anon(path.dentry));
11492 +       } else
11493 +               path.dentry = dget(sb->s_root);
11494 +
11495 +       path.mnt = au_mnt_get(sb);
11496 +       dentry = au_lkup_by_ino(&path, ino, nsi_lock);
11497 +       path_put(&path);
11498 +
11499 +out:
11500 +       AuTraceErrPtr(dentry);
11501 +       return dentry;
11502 +}
11503 +
11504 +/* ---------------------------------------------------------------------- */
11505 +
11506 +static int h_acceptable(void *expv, struct dentry *dentry)
11507 +{
11508 +       return 1;
11509 +}
11510 +
11511 +static char *au_build_path(struct dentry *h_parent, struct path *h_rootpath,
11512 +                          char *buf, int len, struct super_block *sb)
11513 +{
11514 +       char *p;
11515 +       int n;
11516 +       struct path path;
11517 +
11518 +       p = d_path(h_rootpath, buf, len);
11519 +       if (IS_ERR(p))
11520 +               goto out;
11521 +       n = strlen(p);
11522 +
11523 +       path.mnt = h_rootpath->mnt;
11524 +       path.dentry = h_parent;
11525 +       p = d_path(&path, buf, len);
11526 +       if (IS_ERR(p))
11527 +               goto out;
11528 +       if (n != 1)
11529 +               p += n;
11530 +
11531 +       path.mnt = au_mnt_get(sb);
11532 +       path.dentry = sb->s_root;
11533 +       p = d_path(&path, buf, len - strlen(p));
11534 +       mntput(path.mnt);
11535 +       if (IS_ERR(p))
11536 +               goto out;
11537 +       if (n != 1)
11538 +               p[strlen(p)] = '/';
11539 +
11540 +out:
11541 +       AuTraceErrPtr(p);
11542 +       return p;
11543 +}
11544 +
11545 +static
11546 +struct dentry *decode_by_path(struct super_block *sb, ino_t ino, __u32 *fh,
11547 +                             int fh_len, struct au_nfsd_si_lock *nsi_lock)
11548 +{
11549 +       struct dentry *dentry, *h_parent, *root;
11550 +       struct super_block *h_sb;
11551 +       char *pathname, *p;
11552 +       struct vfsmount *h_mnt;
11553 +       struct au_branch *br;
11554 +       int err;
11555 +       struct path path;
11556 +
11557 +       br = au_sbr(sb, nsi_lock->bindex);
11558 +       h_mnt = au_br_mnt(br);
11559 +       h_sb = h_mnt->mnt_sb;
11560 +       /* todo: call lower fh_to_dentry()? fh_to_parent()? */
11561 +       lockdep_off();
11562 +       h_parent = exportfs_decode_fh(h_mnt, (void *)(fh + Fh_tail),
11563 +                                     fh_len - Fh_tail, fh[Fh_h_type],
11564 +                                     h_acceptable, /*context*/NULL);
11565 +       lockdep_on();
11566 +       dentry = h_parent;
11567 +       if (unlikely(!h_parent || IS_ERR(h_parent))) {
11568 +               AuWarn1("%s decode_fh failed, %ld\n",
11569 +                       au_sbtype(h_sb), PTR_ERR(h_parent));
11570 +               goto out;
11571 +       }
11572 +       dentry = NULL;
11573 +       if (unlikely(au_test_anon(h_parent))) {
11574 +               AuWarn1("%s decode_fh returned a disconnected dentry\n",
11575 +                       au_sbtype(h_sb));
11576 +               goto out_h_parent;
11577 +       }
11578 +
11579 +       dentry = ERR_PTR(-ENOMEM);
11580 +       pathname = (void *)__get_free_page(GFP_NOFS);
11581 +       if (unlikely(!pathname))
11582 +               goto out_h_parent;
11583 +
11584 +       root = sb->s_root;
11585 +       path.mnt = h_mnt;
11586 +       di_read_lock_parent(root, !AuLock_IR);
11587 +       path.dentry = au_h_dptr(root, nsi_lock->bindex);
11588 +       di_read_unlock(root, !AuLock_IR);
11589 +       p = au_build_path(h_parent, &path, pathname, PAGE_SIZE, sb);
11590 +       dentry = (void *)p;
11591 +       if (IS_ERR(p))
11592 +               goto out_pathname;
11593 +
11594 +       si_read_unlock(sb);
11595 +       err = vfsub_kern_path(p, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &path);
11596 +       dentry = ERR_PTR(err);
11597 +       if (unlikely(err))
11598 +               goto out_relock;
11599 +
11600 +       dentry = ERR_PTR(-ENOENT);
11601 +       AuDebugOn(au_test_anon(path.dentry));
11602 +       if (unlikely(d_really_is_negative(path.dentry)))
11603 +               goto out_path;
11604 +
11605 +       if (ino != d_inode(path.dentry)->i_ino)
11606 +               dentry = au_lkup_by_ino(&path, ino, /*nsi_lock*/NULL);
11607 +       else
11608 +               dentry = dget(path.dentry);
11609 +
11610 +out_path:
11611 +       path_put(&path);
11612 +out_relock:
11613 +       if (unlikely(si_nfsd_read_lock(sb, nsi_lock) < 0))
11614 +               if (!IS_ERR(dentry)) {
11615 +                       dput(dentry);
11616 +                       dentry = ERR_PTR(-ESTALE);
11617 +               }
11618 +out_pathname:
11619 +       au_delayed_free_page((unsigned long)pathname);
11620 +out_h_parent:
11621 +       dput(h_parent);
11622 +out:
11623 +       AuTraceErrPtr(dentry);
11624 +       return dentry;
11625 +}
11626 +
11627 +/* ---------------------------------------------------------------------- */
11628 +
11629 +static struct dentry *
11630 +aufs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
11631 +                 int fh_type)
11632 +{
11633 +       struct dentry *dentry;
11634 +       __u32 *fh = fid->raw;
11635 +       struct au_branch *br;
11636 +       ino_t ino, dir_ino;
11637 +       struct au_nfsd_si_lock nsi_lock = {
11638 +               .force_lock     = 0
11639 +       };
11640 +
11641 +       dentry = ERR_PTR(-ESTALE);
11642 +       /* it should never happen, but the file handle is unreliable */
11643 +       if (unlikely(fh_len < Fh_tail))
11644 +               goto out;
11645 +       nsi_lock.sigen = fh[Fh_sigen];
11646 +       nsi_lock.br_id = fh[Fh_br_id];
11647 +
11648 +       /* branch id may be wrapped around */
11649 +       br = NULL;
11650 +       if (unlikely(si_nfsd_read_lock(sb, &nsi_lock)))
11651 +               goto out;
11652 +       nsi_lock.force_lock = 1;
11653 +
11654 +       /* is this inode still cached? */
11655 +       ino = decode_ino(fh + Fh_ino);
11656 +       /* it should never happen */
11657 +       if (unlikely(ino == AUFS_ROOT_INO))
11658 +               goto out_unlock;
11659 +
11660 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11661 +       dentry = decode_by_ino(sb, ino, dir_ino);
11662 +       if (IS_ERR(dentry))
11663 +               goto out_unlock;
11664 +       if (dentry)
11665 +               goto accept;
11666 +
11667 +       /* is the parent dir cached? */
11668 +       br = au_sbr(sb, nsi_lock.bindex);
11669 +       au_br_get(br);
11670 +       dentry = decode_by_dir_ino(sb, ino, dir_ino, &nsi_lock);
11671 +       if (IS_ERR(dentry))
11672 +               goto out_unlock;
11673 +       if (dentry)
11674 +               goto accept;
11675 +
11676 +       /* lookup path */
11677 +       dentry = decode_by_path(sb, ino, fh, fh_len, &nsi_lock);
11678 +       if (IS_ERR(dentry))
11679 +               goto out_unlock;
11680 +       if (unlikely(!dentry))
11681 +               /* todo?: make it ESTALE */
11682 +               goto out_unlock;
11683 +
11684 +accept:
11685 +       if (!au_digen_test(dentry, au_sigen(sb))
11686 +           && d_inode(dentry)->i_generation == fh[Fh_igen])
11687 +               goto out_unlock; /* success */
11688 +
11689 +       dput(dentry);
11690 +       dentry = ERR_PTR(-ESTALE);
11691 +out_unlock:
11692 +       if (br)
11693 +               au_br_put(br);
11694 +       si_read_unlock(sb);
11695 +out:
11696 +       AuTraceErrPtr(dentry);
11697 +       return dentry;
11698 +}
11699 +
11700 +#if 0 /* reserved for future use */
11701 +/* support subtreecheck option */
11702 +static struct dentry *aufs_fh_to_parent(struct super_block *sb, struct fid *fid,
11703 +                                       int fh_len, int fh_type)
11704 +{
11705 +       struct dentry *parent;
11706 +       __u32 *fh = fid->raw;
11707 +       ino_t dir_ino;
11708 +
11709 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11710 +       parent = decode_by_ino(sb, dir_ino, 0);
11711 +       if (IS_ERR(parent))
11712 +               goto out;
11713 +       if (!parent)
11714 +               parent = decode_by_path(sb, au_br_index(sb, fh[Fh_br_id]),
11715 +                                       dir_ino, fh, fh_len);
11716 +
11717 +out:
11718 +       AuTraceErrPtr(parent);
11719 +       return parent;
11720 +}
11721 +#endif
11722 +
11723 +/* ---------------------------------------------------------------------- */
11724 +
11725 +static int aufs_encode_fh(struct inode *inode, __u32 *fh, int *max_len,
11726 +                         struct inode *dir)
11727 +{
11728 +       int err;
11729 +       aufs_bindex_t bindex;
11730 +       struct super_block *sb, *h_sb;
11731 +       struct dentry *dentry, *parent, *h_parent;
11732 +       struct inode *h_dir;
11733 +       struct au_branch *br;
11734 +
11735 +       err = -ENOSPC;
11736 +       if (unlikely(*max_len <= Fh_tail)) {
11737 +               AuWarn1("NFSv2 client (max_len %d)?\n", *max_len);
11738 +               goto out;
11739 +       }
11740 +
11741 +       err = FILEID_ROOT;
11742 +       if (inode->i_ino == AUFS_ROOT_INO) {
11743 +               AuDebugOn(inode->i_ino != AUFS_ROOT_INO);
11744 +               goto out;
11745 +       }
11746 +
11747 +       h_parent = NULL;
11748 +       sb = inode->i_sb;
11749 +       err = si_read_lock(sb, AuLock_FLUSH);
11750 +       if (unlikely(err))
11751 +               goto out;
11752 +
11753 +#ifdef CONFIG_AUFS_DEBUG
11754 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
11755 +               AuWarn1("NFS-exporting requires xino\n");
11756 +#endif
11757 +       err = -EIO;
11758 +       parent = NULL;
11759 +       ii_read_lock_child(inode);
11760 +       bindex = au_ibtop(inode);
11761 +       if (!dir) {
11762 +               dentry = d_find_any_alias(inode);
11763 +               if (unlikely(!dentry))
11764 +                       goto out_unlock;
11765 +               AuDebugOn(au_test_anon(dentry));
11766 +               parent = dget_parent(dentry);
11767 +               dput(dentry);
11768 +               if (unlikely(!parent))
11769 +                       goto out_unlock;
11770 +               if (d_really_is_positive(parent))
11771 +                       dir = d_inode(parent);
11772 +       }
11773 +
11774 +       ii_read_lock_parent(dir);
11775 +       h_dir = au_h_iptr(dir, bindex);
11776 +       ii_read_unlock(dir);
11777 +       if (unlikely(!h_dir))
11778 +               goto out_parent;
11779 +       h_parent = d_find_any_alias(h_dir);
11780 +       if (unlikely(!h_parent))
11781 +               goto out_hparent;
11782 +
11783 +       err = -EPERM;
11784 +       br = au_sbr(sb, bindex);
11785 +       h_sb = au_br_sb(br);
11786 +       if (unlikely(!h_sb->s_export_op)) {
11787 +               AuErr1("%s branch is not exportable\n", au_sbtype(h_sb));
11788 +               goto out_hparent;
11789 +       }
11790 +
11791 +       fh[Fh_br_id] = br->br_id;
11792 +       fh[Fh_sigen] = au_sigen(sb);
11793 +       encode_ino(fh + Fh_ino, inode->i_ino);
11794 +       encode_ino(fh + Fh_dir_ino, dir->i_ino);
11795 +       fh[Fh_igen] = inode->i_generation;
11796 +
11797 +       *max_len -= Fh_tail;
11798 +       fh[Fh_h_type] = exportfs_encode_fh(h_parent, (void *)(fh + Fh_tail),
11799 +                                          max_len,
11800 +                                          /*connectable or subtreecheck*/0);
11801 +       err = fh[Fh_h_type];
11802 +       *max_len += Fh_tail;
11803 +       /* todo: macros? */
11804 +       if (err != FILEID_INVALID)
11805 +               err = 99;
11806 +       else
11807 +               AuWarn1("%s encode_fh failed\n", au_sbtype(h_sb));
11808 +
11809 +out_hparent:
11810 +       dput(h_parent);
11811 +out_parent:
11812 +       dput(parent);
11813 +out_unlock:
11814 +       ii_read_unlock(inode);
11815 +       si_read_unlock(sb);
11816 +out:
11817 +       if (unlikely(err < 0))
11818 +               err = FILEID_INVALID;
11819 +       return err;
11820 +}
11821 +
11822 +/* ---------------------------------------------------------------------- */
11823 +
11824 +static int aufs_commit_metadata(struct inode *inode)
11825 +{
11826 +       int err;
11827 +       aufs_bindex_t bindex;
11828 +       struct super_block *sb;
11829 +       struct inode *h_inode;
11830 +       int (*f)(struct inode *inode);
11831 +
11832 +       sb = inode->i_sb;
11833 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
11834 +       ii_write_lock_child(inode);
11835 +       bindex = au_ibtop(inode);
11836 +       AuDebugOn(bindex < 0);
11837 +       h_inode = au_h_iptr(inode, bindex);
11838 +
11839 +       f = h_inode->i_sb->s_export_op->commit_metadata;
11840 +       if (f)
11841 +               err = f(h_inode);
11842 +       else {
11843 +               struct writeback_control wbc = {
11844 +                       .sync_mode      = WB_SYNC_ALL,
11845 +                       .nr_to_write    = 0 /* metadata only */
11846 +               };
11847 +
11848 +               err = sync_inode(h_inode, &wbc);
11849 +       }
11850 +
11851 +       au_cpup_attr_timesizes(inode);
11852 +       ii_write_unlock(inode);
11853 +       si_read_unlock(sb);
11854 +       return err;
11855 +}
11856 +
11857 +/* ---------------------------------------------------------------------- */
11858 +
11859 +static struct export_operations aufs_export_op = {
11860 +       .fh_to_dentry           = aufs_fh_to_dentry,
11861 +       /* .fh_to_parent        = aufs_fh_to_parent, */
11862 +       .encode_fh              = aufs_encode_fh,
11863 +       .commit_metadata        = aufs_commit_metadata
11864 +};
11865 +
11866 +void au_export_init(struct super_block *sb)
11867 +{
11868 +       struct au_sbinfo *sbinfo;
11869 +       __u32 u;
11870 +
11871 +       BUILD_BUG_ON_MSG(IS_BUILTIN(CONFIG_AUFS_FS)
11872 +                        && IS_MODULE(CONFIG_EXPORTFS),
11873 +                        AUFS_NAME ": unsupported configuration "
11874 +                        "CONFIG_EXPORTFS=m and CONFIG_AUFS_FS=y");
11875 +
11876 +       sb->s_export_op = &aufs_export_op;
11877 +       sbinfo = au_sbi(sb);
11878 +       sbinfo->si_xigen = NULL;
11879 +       get_random_bytes(&u, sizeof(u));
11880 +       BUILD_BUG_ON(sizeof(u) != sizeof(int));
11881 +       atomic_set(&sbinfo->si_xigen_next, u);
11882 +}
11883 diff -urN /usr/share/empty/fs/aufs/fhsm.c linux/fs/aufs/fhsm.c
11884 --- /usr/share/empty/fs/aufs/fhsm.c     1970-01-01 01:00:00.000000000 +0100
11885 +++ linux/fs/aufs/fhsm.c        2017-05-06 22:16:52.778220093 +0200
11886 @@ -0,0 +1,426 @@
11887 +/*
11888 + * Copyright (C) 2011-2017 Junjiro R. Okajima
11889 + *
11890 + * This program, aufs is free software; you can redistribute it and/or modify
11891 + * it under the terms of the GNU General Public License as published by
11892 + * the Free Software Foundation; either version 2 of the License, or
11893 + * (at your option) any later version.
11894 + *
11895 + * This program is distributed in the hope that it will be useful,
11896 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11897 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11898 + * GNU General Public License for more details.
11899 + *
11900 + * You should have received a copy of the GNU General Public License
11901 + * along with this program; if not, write to the Free Software
11902 + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
11903 + */
11904 +
11905 +/*
11906 + * File-based Hierarchy Storage Management
11907 + */
11908 +
11909 +#include <linux/anon_inodes.h>
11910 +#include <linux/poll.h>
11911 +#include <linux/seq_file.h>
11912 +#include <linux/statfs.h>
11913 +#include "aufs.h"
11914 +
11915 +static aufs_bindex_t au_fhsm_bottom(struct super_block *sb)
11916 +{
11917 +       struct au_sbinfo *sbinfo;
11918 +       struct au_fhsm *fhsm;
11919 +
11920 +       SiMustAnyLock(sb);
11921 +
11922 +       sbinfo = au_sbi(sb);
11923 +       fhsm = &sbinfo->si_fhsm;
11924 +       AuDebugOn(!fhsm);
11925 +       return fhsm->fhsm_bottom;
11926 +}
11927 +
11928 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex)
11929 +{
11930 +       struct au_sbinfo *sbinfo;
11931 +       struct au_fhsm *fhsm;
11932 +
11933 +       SiMustWriteLock(sb);
11934 +
11935 +       sbinfo = au_sbi(sb);
11936 +       fhsm = &sbinfo->si_fhsm;
11937 +       AuDebugOn(!fhsm);
11938 +       fhsm->fhsm_bottom = bindex;
11939 +}
11940 +
11941 +/* ---------------------------------------------------------------------- */
11942 +
11943 +static int au_fhsm_test_jiffy(struct au_sbinfo *sbinfo, struct au_branch *br)
11944 +{
11945 +       struct au_br_fhsm *bf;
11946 +
11947 +       bf = br->br_fhsm;
11948 +       MtxMustLock(&bf->bf_lock);
11949 +
11950 +       return !bf->bf_readable
11951 +               || time_after(jiffies,
11952 +                             bf->bf_jiffy + sbinfo->si_fhsm.fhsm_expire);
11953 +}
11954 +
11955 +/* ---------------------------------------------------------------------- */
11956 +
11957 +static void au_fhsm_notify(struct super_block *sb, int val)
11958 +{
11959 +       struct au_sbinfo *sbinfo;
11960 +       struct au_fhsm *fhsm;
11961 +
11962 +       SiMustAnyLock(sb);
11963 +
11964 +       sbinfo = au_sbi(sb);
11965 +       fhsm = &sbinfo->si_fhsm;
11966 +       if (au_fhsm_pid(fhsm)
11967 +           && atomic_read(&fhsm->fhsm_readable) != -1) {
11968 +               atomic_set(&fhsm->fhsm_readable, val);
11969 +               if (val)
11970 +                       wake_up(&fhsm->fhsm_wqh);
11971 +       }
11972 +}
11973 +
11974 +static int au_fhsm_stfs(struct super_block *sb, aufs_bindex_t bindex,
11975 +                       struct aufs_stfs *rstfs, int do_lock, int do_notify)
11976 +{
11977 +       int err;
11978 +       struct au_branch *br;
11979 +       struct au_br_fhsm *bf;
11980 +
11981 +       br = au_sbr(sb, bindex);
11982 +       AuDebugOn(au_br_rdonly(br));
11983 +       bf = br->br_fhsm;
11984 +       AuDebugOn(!bf);
11985 +
11986 +       if (do_lock)
11987 +               mutex_lock(&bf->bf_lock);
11988 +       else
11989 +               MtxMustLock(&bf->bf_lock);
11990 +
11991 +       /* sb->s_root for NFS is unreliable */
11992 +       err = au_br_stfs(br, &bf->bf_stfs);
11993 +       if (unlikely(err)) {
11994 +               AuErr1("FHSM failed (%d), b%d, ignored.\n", bindex, err);
11995 +               goto out;
11996 +       }
11997 +
11998 +       bf->bf_jiffy = jiffies;
11999 +       bf->bf_readable = 1;
12000 +       if (do_notify)
12001 +               au_fhsm_notify(sb, /*val*/1);
12002 +       if (rstfs)
12003 +               *rstfs = bf->bf_stfs;
12004 +
12005 +out:
12006 +       if (do_lock)
12007 +               mutex_unlock(&bf->bf_lock);
12008 +       au_fhsm_notify(sb, /*val*/1);
12009 +
12010 +       return err;
12011 +}
12012 +
12013 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force)
12014 +{
12015 +       int err;
12016 +       struct au_sbinfo *sbinfo;
12017 +       struct au_fhsm *fhsm;
12018 +       struct au_branch *br;
12019 +       struct au_br_fhsm *bf;
12020 +
12021 +       AuDbg("b%d, force %d\n", bindex, force);
12022 +       SiMustAnyLock(sb);
12023 +
12024 +       sbinfo = au_sbi(sb);
12025 +       fhsm = &sbinfo->si_fhsm;
12026 +       if (!au_ftest_si(sbinfo, FHSM)
12027 +           || fhsm->fhsm_bottom == bindex)
12028 +               return;
12029 +
12030 +       br = au_sbr(sb, bindex);
12031 +       bf = br->br_fhsm;
12032 +       AuDebugOn(!bf);
12033 +       mutex_lock(&bf->bf_lock);
12034 +       if (force
12035 +           || au_fhsm_pid(fhsm)
12036 +           || au_fhsm_test_jiffy(sbinfo, br))
12037 +               err = au_fhsm_stfs(sb, bindex, /*rstfs*/NULL, /*do_lock*/0,
12038 +                                 /*do_notify*/1);
12039 +       mutex_unlock(&bf->bf_lock);
12040 +}
12041 +
12042 +void au_fhsm_wrote_all(struct super_block *sb, int force)
12043 +{
12044 +       aufs_bindex_t bindex, bbot;
12045 +       struct au_branch *br;
12046 +
12047 +       /* exclude the bottom */
12048 +       bbot = au_fhsm_bottom(sb);
12049 +       for (bindex = 0; bindex < bbot; bindex++) {
12050 +               br = au_sbr(sb, bindex);
12051 +               if (au_br_fhsm(br->br_perm))
12052 +                       au_fhsm_wrote(sb, bindex, force);
12053 +       }
12054 +}
12055 +
12056 +/* ---------------------------------------------------------------------- */
12057 +
12058 +static unsigned int au_fhsm_poll(struct file *file,
12059 +                                struct poll_table_struct *wait)
12060 +{
12061 +       unsigned int mask;
12062 +       struct au_sbinfo *sbinfo;
12063 +       struct au_fhsm *fhsm;
12064 +
12065 +       mask = 0;
12066 +       sbinfo = file->private_data;
12067 +       fhsm = &sbinfo->si_fhsm;
12068 +       poll_wait(file, &fhsm->fhsm_wqh, wait);
12069 +       if (atomic_read(&fhsm->fhsm_readable))
12070 +               mask = POLLIN /* | POLLRDNORM */;
12071 +
12072 +       AuTraceErr((int)mask);
12073 +       return mask;
12074 +}
12075 +
12076 +static int au_fhsm_do_read_one(struct aufs_stbr __user *stbr,
12077 +                             struct aufs_stfs *stfs, __s16 brid)
12078 +{
12079 +       int err;
12080 +
12081 +       err = copy_to_user(&stbr->stfs, stfs, sizeof(*stfs));
12082 +       if (!err)
12083 +               err = __put_user(brid, &stbr->brid);
12084 +       if (unlikely(err))
12085 +               err = -EFAULT;
12086 +
12087 +       return err;
12088 +}
12089 +
12090 +static ssize_t au_fhsm_do_read(struct super_block *sb,
12091 +                              struct aufs_stbr __user *stbr, size_t count)
12092 +{
12093 +       ssize_t err;
12094 +       int nstbr;
12095 +       aufs_bindex_t bindex, bbot;
12096 +       struct au_branch *br;
12097 +       struct au_br_fhsm *bf;
12098 +
12099 +       /* except the bottom branch */
12100 +       err = 0;
12101 +       nstbr = 0;
12102 +       bbot = au_fhsm_bottom(sb);
12103 +       for (bindex = 0; !err && bindex < bbot; bindex++) {
12104 +               br = au_sbr(sb, bindex);
12105 +               if (!au_br_fhsm(br->br_perm))
12106 +                       continue;
12107 +
12108 +               bf = br->br_fhsm;
12109 +               mutex_lock(&bf->bf_lock);
12110 +               if (bf->bf_readable) {
12111 +                       err = -EFAULT;
12112 +                       if (count >= sizeof(*stbr))
12113 +                               err = au_fhsm_do_read_one(stbr++, &bf->bf_stfs,
12114 +                                                         br->br_id);
12115 +                       if (!err) {
12116 +                               bf->bf_readable = 0;
12117 +                               count -= sizeof(*stbr);
12118 +                               nstbr++;
12119 +                       }
12120 +               }
12121 +               mutex_unlock(&bf->bf_lock);
12122 +       }
12123 +       if (!err)
12124 +               err = sizeof(*stbr) * nstbr;
12125 +
12126 +       return err;
12127 +}
12128 +
12129 +static ssize_t au_fhsm_read(struct file *file, char __user *buf, size_t count,
12130 +                          loff_t *pos)
12131 +{
12132 +       ssize_t err;
12133 +       int readable;
12134 +       aufs_bindex_t nfhsm, bindex, bbot;
12135 +       struct au_sbinfo *sbinfo;
12136 +       struct au_fhsm *fhsm;
12137 +       struct au_branch *br;
12138 +       struct super_block *sb;
12139 +
12140 +       err = 0;
12141 +       sbinfo = file->private_data;
12142 +       fhsm = &sbinfo->si_fhsm;
12143 +need_data:
12144 +       spin_lock_irq(&fhsm->fhsm_wqh.lock);
12145 +       if (!atomic_read(&fhsm->fhsm_readable)) {
12146 +               if (vfsub_file_flags(file) & O_NONBLOCK)
12147 +                       err = -EAGAIN;
12148 +               else
12149 +                       err = wait_event_interruptible_locked_irq
12150 +                               (fhsm->fhsm_wqh,
12151 +                                atomic_read(&fhsm->fhsm_readable));
12152 +       }
12153 +       spin_unlock_irq(&fhsm->fhsm_wqh.lock);
12154 +       if (unlikely(err))
12155 +               goto out;
12156 +
12157 +       /* sb may already be dead */
12158 +       au_rw_read_lock(&sbinfo->si_rwsem);
12159 +       readable = atomic_read(&fhsm->fhsm_readable);
12160 +       if (readable > 0) {
12161 +               sb = sbinfo->si_sb;
12162 +               AuDebugOn(!sb);
12163 +               /* exclude the bottom branch */
12164 +               nfhsm = 0;
12165 +               bbot = au_fhsm_bottom(sb);
12166 +               for (bindex = 0; bindex < bbot; bindex++) {
12167 +                       br = au_sbr(sb, bindex);
12168 +                       if (au_br_fhsm(br->br_perm))
12169 +                               nfhsm++;
12170 +               }
12171 +               err = -EMSGSIZE;
12172 +               if (nfhsm * sizeof(struct aufs_stbr) <= count) {
12173 +                       atomic_set(&fhsm->fhsm_readable, 0);
12174 +                       err = au_fhsm_do_read(sbinfo->si_sb, (void __user *)buf,
12175 +                                            count);
12176 +               }
12177 +       }
12178 +       au_rw_read_unlock(&sbinfo->si_rwsem);
12179 +       if (!readable)
12180 +               goto need_data;
12181 +
12182 +out:
12183 +       return err;
12184 +}
12185 +
12186 +static int au_fhsm_release(struct inode *inode, struct file *file)
12187 +{
12188 +       struct au_sbinfo *sbinfo;
12189 +       struct au_fhsm *fhsm;
12190 +
12191 +       /* sb may already be dead */
12192 +       sbinfo = file->private_data;
12193 +       fhsm = &sbinfo->si_fhsm;
12194 +       spin_lock(&fhsm->fhsm_spin);
12195 +       fhsm->fhsm_pid = 0;
12196 +       spin_unlock(&fhsm->fhsm_spin);
12197 +       kobject_put(&sbinfo->si_kobj);
12198 +
12199 +       return 0;
12200 +}
12201 +
12202 +static const struct file_operations au_fhsm_fops = {
12203 +       .owner          = THIS_MODULE,
12204 +       .llseek         = noop_llseek,
12205 +       .read           = au_fhsm_read,
12206 +       .poll           = au_fhsm_poll,
12207 +       .release        = au_fhsm_release
12208 +};
12209 +
12210 +int au_fhsm_fd(struct super_block *sb, int oflags)
12211 +{
12212 +       int err, fd;
12213 +       struct au_sbinfo *sbinfo;
12214 +       struct au_fhsm *fhsm;
12215 +
12216 +       err = -EPERM;
12217 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
12218 +               goto out;
12219 +
12220 +       err = -EINVAL;
12221 +       if (unlikely(oflags & ~(O_CLOEXEC | O_NONBLOCK)))
12222 +               goto out;
12223 +
12224 +       err = 0;
12225 +       sbinfo = au_sbi(sb);
12226 +       fhsm = &sbinfo->si_fhsm;
12227 +       spin_lock(&fhsm->fhsm_spin);
12228 +       if (!fhsm->fhsm_pid)
12229 +               fhsm->fhsm_pid = current->pid;
12230 +       else
12231 +               err = -EBUSY;
12232 +       spin_unlock(&fhsm->fhsm_spin);
12233 +       if (unlikely(err))
12234 +               goto out;
12235 +
12236 +       oflags |= O_RDONLY;
12237 +       /* oflags |= FMODE_NONOTIFY; */
12238 +       fd = anon_inode_getfd("[aufs_fhsm]", &au_fhsm_fops, sbinfo, oflags);
12239 +       err = fd;
12240 +       if (unlikely(fd < 0))
12241 +               goto out_pid;
12242 +
12243 +       /* succeed reglardless 'fhsm' status */
12244 +       kobject_get(&sbinfo->si_kobj);
12245 +       si_noflush_read_lock(sb);
12246 +       if (au_ftest_si(sbinfo, FHSM))
12247 +               au_fhsm_wrote_all(sb, /*force*/0);
12248 +       si_read_unlock(sb);
12249 +       goto out; /* success */
12250 +
12251 +out_pid:
12252 +       spin_lock(&fhsm->fhsm_spin);
12253 +       fhsm->fhsm_pid = 0;
12254 +       spin_unlock(&fhsm->fhsm_spin);
12255 +out:
12256 +       AuTraceErr(err);
12257 +       return err;
12258 +}
12259 +
12260 +/* ---------------------------------------------------------------------- */
12261 +
12262 +int au_fhsm_br_alloc(struct au_branch *br)
12263 +{
12264 +       int err;
12265 +
12266 +       err = 0;
12267 +       br->br_fhsm = kmalloc(sizeof(*br->br_fhsm), GFP_NOFS);
12268 +       if (br->br_fhsm)
12269 +               au_br_fhsm_init(br->br_fhsm);
12270 +       else
12271 +               err = -ENOMEM;
12272 +
12273 +       return err;
12274 +}
12275 +
12276 +/* ---------------------------------------------------------------------- */
12277 +
12278 +void au_fhsm_fin(struct super_block *sb)
12279 +{
12280 +       au_fhsm_notify(sb, /*val*/-1);
12281 +}
12282 +
12283 +void au_fhsm_init(struct au_sbinfo *sbinfo)
12284 +{
12285 +       struct au_fhsm *fhsm;
12286 +
12287 +       fhsm = &sbinfo->si_fhsm;
12288 +       spin_lock_init(&fhsm->fhsm_spin);
12289 +       init_waitqueue_head(&fhsm->fhsm_wqh);
12290 +       atomic_set(&fhsm->fhsm_readable, 0);
12291 +       fhsm->fhsm_expire
12292 +               = msecs_to_jiffies(AUFS_FHSM_CACHE_DEF_SEC * MSEC_PER_SEC);
12293 +       fhsm->fhsm_bottom = -1;
12294 +}
12295 +
12296 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec)
12297 +{
12298 +       sbinfo->si_fhsm.fhsm_expire
12299 +               = msecs_to_jiffies(sec * MSEC_PER_SEC);
12300 +}
12301 +
12302 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo)
12303 +{
12304 +       unsigned int u;
12305 +
12306 +       if (!au_ftest_si(sbinfo, FHSM))
12307 +               return;
12308 +
12309 +       u = jiffies_to_msecs(sbinfo->si_fhsm.fhsm_expire) / MSEC_PER_SEC;
12310 +       if (u != AUFS_FHSM_CACHE_DEF_SEC)
12311 +               seq_printf(seq, ",fhsm_sec=%u", u);
12312 +}
12313 diff -urN /usr/share/empty/fs/aufs/file.c linux/fs/aufs/file.c
12314 --- /usr/share/empty/fs/aufs/file.c     1970-01-01 01:00:00.000000000 +0100
12315 +++ linux/fs/aufs/file.c        2017-05-06 22:16:52.778220093 +0200
12316 @@ -0,0 +1,860 @@
12317 +/*
12318 + * Copyright (C) 2005-2017 Junjiro R. Okajima
12319 + *
12320 + * This program, aufs is free software; you can redistribute it and/or modify
12321 + * it under the terms of the GNU General Public License as published by
12322 + * the Free Software Foundation; either version 2 of the License, or
12323 + * (at your option) any later version.
12324 + *
12325 + * This program is distributed in the hope that it will be useful,
12326 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12327 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12328 + * GNU General Public License for more details.
12329 + *
12330 + * You should have received a copy of the GNU General Public License
12331 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12332 + */
12333 +
12334 +/*
12335 + * handling file/dir, and address_space operation
12336 + */
12337 +
12338 +#ifdef CONFIG_AUFS_DEBUG
12339 +#include <linux/migrate.h>
12340 +#endif
12341 +#include <linux/pagemap.h>
12342 +#include "aufs.h"
12343 +
12344 +/* drop flags for writing */
12345 +unsigned int au_file_roflags(unsigned int flags)
12346 +{
12347 +       flags &= ~(O_WRONLY | O_RDWR | O_APPEND | O_CREAT | O_TRUNC);
12348 +       flags |= O_RDONLY | O_NOATIME;
12349 +       return flags;
12350 +}
12351 +
12352 +/* common functions to regular file and dir */
12353 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
12354 +                      struct file *file, int force_wr)
12355 +{
12356 +       struct file *h_file;
12357 +       struct dentry *h_dentry;
12358 +       struct inode *h_inode;
12359 +       struct super_block *sb;
12360 +       struct au_branch *br;
12361 +       struct path h_path;
12362 +       int err;
12363 +
12364 +       /* a race condition can happen between open and unlink/rmdir */
12365 +       h_file = ERR_PTR(-ENOENT);
12366 +       h_dentry = au_h_dptr(dentry, bindex);
12367 +       if (au_test_nfsd() && (!h_dentry || d_is_negative(h_dentry)))
12368 +               goto out;
12369 +       h_inode = d_inode(h_dentry);
12370 +       spin_lock(&h_dentry->d_lock);
12371 +       err = (!d_unhashed(dentry) && d_unlinked(h_dentry))
12372 +               /* || !d_inode(dentry)->i_nlink */
12373 +               ;
12374 +       spin_unlock(&h_dentry->d_lock);
12375 +       if (unlikely(err))
12376 +               goto out;
12377 +
12378 +       sb = dentry->d_sb;
12379 +       br = au_sbr(sb, bindex);
12380 +       err = au_br_test_oflag(flags, br);
12381 +       h_file = ERR_PTR(err);
12382 +       if (unlikely(err))
12383 +               goto out;
12384 +
12385 +       /* drop flags for writing */
12386 +       if (au_test_ro(sb, bindex, d_inode(dentry))) {
12387 +               if (force_wr && !(flags & O_WRONLY))
12388 +                       force_wr = 0;
12389 +               flags = au_file_roflags(flags);
12390 +               if (force_wr) {
12391 +                       h_file = ERR_PTR(-EROFS);
12392 +                       flags = au_file_roflags(flags);
12393 +                       if (unlikely(vfsub_native_ro(h_inode)
12394 +                                    || IS_APPEND(h_inode)))
12395 +                               goto out;
12396 +                       flags &= ~O_ACCMODE;
12397 +                       flags |= O_WRONLY;
12398 +               }
12399 +       }
12400 +       flags &= ~O_CREAT;
12401 +       au_br_get(br);
12402 +       h_path.dentry = h_dentry;
12403 +       h_path.mnt = au_br_mnt(br);
12404 +       h_file = vfsub_dentry_open(&h_path, flags);
12405 +       if (IS_ERR(h_file))
12406 +               goto out_br;
12407 +
12408 +       if (flags & __FMODE_EXEC) {
12409 +               err = deny_write_access(h_file);
12410 +               if (unlikely(err)) {
12411 +                       fput(h_file);
12412 +                       h_file = ERR_PTR(err);
12413 +                       goto out_br;
12414 +               }
12415 +       }
12416 +       fsnotify_open(h_file);
12417 +       goto out; /* success */
12418 +
12419 +out_br:
12420 +       au_br_put(br);
12421 +out:
12422 +       return h_file;
12423 +}
12424 +
12425 +static int au_cmoo(struct dentry *dentry)
12426 +{
12427 +       int err, cmoo;
12428 +       unsigned int udba;
12429 +       struct path h_path;
12430 +       struct au_pin pin;
12431 +       struct au_cp_generic cpg = {
12432 +               .dentry = dentry,
12433 +               .bdst   = -1,
12434 +               .bsrc   = -1,
12435 +               .len    = -1,
12436 +               .pin    = &pin,
12437 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
12438 +       };
12439 +       struct inode *delegated;
12440 +       struct super_block *sb;
12441 +       struct au_sbinfo *sbinfo;
12442 +       struct au_fhsm *fhsm;
12443 +       pid_t pid;
12444 +       struct au_branch *br;
12445 +       struct dentry *parent;
12446 +       struct au_hinode *hdir;
12447 +
12448 +       DiMustWriteLock(dentry);
12449 +       IiMustWriteLock(d_inode(dentry));
12450 +
12451 +       err = 0;
12452 +       if (IS_ROOT(dentry))
12453 +               goto out;
12454 +       cpg.bsrc = au_dbtop(dentry);
12455 +       if (!cpg.bsrc)
12456 +               goto out;
12457 +
12458 +       sb = dentry->d_sb;
12459 +       sbinfo = au_sbi(sb);
12460 +       fhsm = &sbinfo->si_fhsm;
12461 +       pid = au_fhsm_pid(fhsm);
12462 +       if (pid
12463 +           && (current->pid == pid
12464 +               || current->real_parent->pid == pid))
12465 +               goto out;
12466 +
12467 +       br = au_sbr(sb, cpg.bsrc);
12468 +       cmoo = au_br_cmoo(br->br_perm);
12469 +       if (!cmoo)
12470 +               goto out;
12471 +       if (!d_is_reg(dentry))
12472 +               cmoo &= AuBrAttr_COO_ALL;
12473 +       if (!cmoo)
12474 +               goto out;
12475 +
12476 +       parent = dget_parent(dentry);
12477 +       di_write_lock_parent(parent);
12478 +       err = au_wbr_do_copyup_bu(dentry, cpg.bsrc - 1);
12479 +       cpg.bdst = err;
12480 +       if (unlikely(err < 0)) {
12481 +               err = 0;        /* there is no upper writable branch */
12482 +               goto out_dgrade;
12483 +       }
12484 +       AuDbg("bsrc %d, bdst %d\n", cpg.bsrc, cpg.bdst);
12485 +
12486 +       /* do not respect the coo attrib for the target branch */
12487 +       err = au_cpup_dirs(dentry, cpg.bdst);
12488 +       if (unlikely(err))
12489 +               goto out_dgrade;
12490 +
12491 +       di_downgrade_lock(parent, AuLock_IR);
12492 +       udba = au_opt_udba(sb);
12493 +       err = au_pin(&pin, dentry, cpg.bdst, udba,
12494 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12495 +       if (unlikely(err))
12496 +               goto out_parent;
12497 +
12498 +       err = au_sio_cpup_simple(&cpg);
12499 +       au_unpin(&pin);
12500 +       if (unlikely(err))
12501 +               goto out_parent;
12502 +       if (!(cmoo & AuBrWAttr_MOO))
12503 +               goto out_parent; /* success */
12504 +
12505 +       err = au_pin(&pin, dentry, cpg.bsrc, udba,
12506 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12507 +       if (unlikely(err))
12508 +               goto out_parent;
12509 +
12510 +       h_path.mnt = au_br_mnt(br);
12511 +       h_path.dentry = au_h_dptr(dentry, cpg.bsrc);
12512 +       hdir = au_hi(d_inode(parent), cpg.bsrc);
12513 +       delegated = NULL;
12514 +       err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated, /*force*/1);
12515 +       au_unpin(&pin);
12516 +       /* todo: keep h_dentry or not? */
12517 +       if (unlikely(err == -EWOULDBLOCK)) {
12518 +               pr_warn("cannot retry for NFSv4 delegation"
12519 +                       " for an internal unlink\n");
12520 +               iput(delegated);
12521 +       }
12522 +       if (unlikely(err)) {
12523 +               pr_err("unlink %pd after coo failed (%d), ignored\n",
12524 +                      dentry, err);
12525 +               err = 0;
12526 +       }
12527 +       goto out_parent; /* success */
12528 +
12529 +out_dgrade:
12530 +       di_downgrade_lock(parent, AuLock_IR);
12531 +out_parent:
12532 +       di_read_unlock(parent, AuLock_IR);
12533 +       dput(parent);
12534 +out:
12535 +       AuTraceErr(err);
12536 +       return err;
12537 +}
12538 +
12539 +int au_do_open(struct file *file, struct au_do_open_args *args)
12540 +{
12541 +       int err, no_lock = args->no_lock;
12542 +       struct dentry *dentry;
12543 +       struct au_finfo *finfo;
12544 +
12545 +       if (!no_lock)
12546 +               err = au_finfo_init(file, args->fidir);
12547 +       else {
12548 +               lockdep_off();
12549 +               err = au_finfo_init(file, args->fidir);
12550 +               lockdep_on();
12551 +       }
12552 +       if (unlikely(err))
12553 +               goto out;
12554 +
12555 +       dentry = file->f_path.dentry;
12556 +       AuDebugOn(IS_ERR_OR_NULL(dentry));
12557 +       if (!no_lock) {
12558 +               di_write_lock_child(dentry);
12559 +               err = au_cmoo(dentry);
12560 +               di_downgrade_lock(dentry, AuLock_IR);
12561 +               if (!err)
12562 +                       err = args->open(file, vfsub_file_flags(file), NULL);
12563 +               di_read_unlock(dentry, AuLock_IR);
12564 +       } else {
12565 +               err = au_cmoo(dentry);
12566 +               if (!err)
12567 +                       err = args->open(file, vfsub_file_flags(file),
12568 +                                        args->h_file);
12569 +               if (!err && au_fbtop(file) != au_dbtop(dentry))
12570 +                       /*
12571 +                        * cmoo happens after h_file was opened.
12572 +                        * need to refresh file later.
12573 +                        */
12574 +                       atomic_dec(&au_fi(file)->fi_generation);
12575 +       }
12576 +
12577 +       finfo = au_fi(file);
12578 +       if (!err) {
12579 +               finfo->fi_file = file;
12580 +               au_sphl_add(&finfo->fi_hlist,
12581 +                           &au_sbi(file->f_path.dentry->d_sb)->si_files);
12582 +       }
12583 +       if (!no_lock)
12584 +               fi_write_unlock(file);
12585 +       else {
12586 +               lockdep_off();
12587 +               fi_write_unlock(file);
12588 +               lockdep_on();
12589 +       }
12590 +       if (unlikely(err)) {
12591 +               finfo->fi_hdir = NULL;
12592 +               au_finfo_fin(file, /*atonce*/0);
12593 +       }
12594 +
12595 +out:
12596 +       return err;
12597 +}
12598 +
12599 +int au_reopen_nondir(struct file *file)
12600 +{
12601 +       int err;
12602 +       aufs_bindex_t btop;
12603 +       struct dentry *dentry;
12604 +       struct file *h_file, *h_file_tmp;
12605 +
12606 +       dentry = file->f_path.dentry;
12607 +       btop = au_dbtop(dentry);
12608 +       h_file_tmp = NULL;
12609 +       if (au_fbtop(file) == btop) {
12610 +               h_file = au_hf_top(file);
12611 +               if (file->f_mode == h_file->f_mode)
12612 +                       return 0; /* success */
12613 +               h_file_tmp = h_file;
12614 +               get_file(h_file_tmp);
12615 +               au_set_h_fptr(file, btop, NULL);
12616 +       }
12617 +       AuDebugOn(au_fi(file)->fi_hdir);
12618 +       /*
12619 +        * it can happen
12620 +        * file exists on both of rw and ro
12621 +        * open --> dbtop and fbtop are both 0
12622 +        * prepend a branch as rw, "rw" become ro
12623 +        * remove rw/file
12624 +        * delete the top branch, "rw" becomes rw again
12625 +        *      --> dbtop is 1, fbtop is still 0
12626 +        * write --> fbtop is 0 but dbtop is 1
12627 +        */
12628 +       /* AuDebugOn(au_fbtop(file) < btop); */
12629 +
12630 +       h_file = au_h_open(dentry, btop, vfsub_file_flags(file) & ~O_TRUNC,
12631 +                          file, /*force_wr*/0);
12632 +       err = PTR_ERR(h_file);
12633 +       if (IS_ERR(h_file)) {
12634 +               if (h_file_tmp) {
12635 +                       au_sbr_get(dentry->d_sb, btop);
12636 +                       au_set_h_fptr(file, btop, h_file_tmp);
12637 +                       h_file_tmp = NULL;
12638 +               }
12639 +               goto out; /* todo: close all? */
12640 +       }
12641 +
12642 +       err = 0;
12643 +       au_set_fbtop(file, btop);
12644 +       au_set_h_fptr(file, btop, h_file);
12645 +       au_update_figen(file);
12646 +       /* todo: necessary? */
12647 +       /* file->f_ra = h_file->f_ra; */
12648 +
12649 +out:
12650 +       if (h_file_tmp)
12651 +               fput(h_file_tmp);
12652 +       return err;
12653 +}
12654 +
12655 +/* ---------------------------------------------------------------------- */
12656 +
12657 +static int au_reopen_wh(struct file *file, aufs_bindex_t btgt,
12658 +                       struct dentry *hi_wh)
12659 +{
12660 +       int err;
12661 +       aufs_bindex_t btop;
12662 +       struct au_dinfo *dinfo;
12663 +       struct dentry *h_dentry;
12664 +       struct au_hdentry *hdp;
12665 +
12666 +       dinfo = au_di(file->f_path.dentry);
12667 +       AuRwMustWriteLock(&dinfo->di_rwsem);
12668 +
12669 +       btop = dinfo->di_btop;
12670 +       dinfo->di_btop = btgt;
12671 +       hdp = au_hdentry(dinfo, btgt);
12672 +       h_dentry = hdp->hd_dentry;
12673 +       hdp->hd_dentry = hi_wh;
12674 +       err = au_reopen_nondir(file);
12675 +       hdp->hd_dentry = h_dentry;
12676 +       dinfo->di_btop = btop;
12677 +
12678 +       return err;
12679 +}
12680 +
12681 +static int au_ready_to_write_wh(struct file *file, loff_t len,
12682 +                               aufs_bindex_t bcpup, struct au_pin *pin)
12683 +{
12684 +       int err;
12685 +       struct inode *inode, *h_inode;
12686 +       struct dentry *h_dentry, *hi_wh;
12687 +       struct au_cp_generic cpg = {
12688 +               .dentry = file->f_path.dentry,
12689 +               .bdst   = bcpup,
12690 +               .bsrc   = -1,
12691 +               .len    = len,
12692 +               .pin    = pin
12693 +       };
12694 +
12695 +       au_update_dbtop(cpg.dentry);
12696 +       inode = d_inode(cpg.dentry);
12697 +       h_inode = NULL;
12698 +       if (au_dbtop(cpg.dentry) <= bcpup
12699 +           && au_dbbot(cpg.dentry) >= bcpup) {
12700 +               h_dentry = au_h_dptr(cpg.dentry, bcpup);
12701 +               if (h_dentry && d_is_positive(h_dentry))
12702 +                       h_inode = d_inode(h_dentry);
12703 +       }
12704 +       hi_wh = au_hi_wh(inode, bcpup);
12705 +       if (!hi_wh && !h_inode)
12706 +               err = au_sio_cpup_wh(&cpg, file);
12707 +       else
12708 +               /* already copied-up after unlink */
12709 +               err = au_reopen_wh(file, bcpup, hi_wh);
12710 +
12711 +       if (!err
12712 +           && (inode->i_nlink > 1
12713 +               || (inode->i_state & I_LINKABLE))
12714 +           && au_opt_test(au_mntflags(cpg.dentry->d_sb), PLINK))
12715 +               au_plink_append(inode, bcpup, au_h_dptr(cpg.dentry, bcpup));
12716 +
12717 +       return err;
12718 +}
12719 +
12720 +/*
12721 + * prepare the @file for writing.
12722 + */
12723 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin)
12724 +{
12725 +       int err;
12726 +       aufs_bindex_t dbtop;
12727 +       struct dentry *parent;
12728 +       struct inode *inode;
12729 +       struct super_block *sb;
12730 +       struct file *h_file;
12731 +       struct au_cp_generic cpg = {
12732 +               .dentry = file->f_path.dentry,
12733 +               .bdst   = -1,
12734 +               .bsrc   = -1,
12735 +               .len    = len,
12736 +               .pin    = pin,
12737 +               .flags  = AuCpup_DTIME
12738 +       };
12739 +
12740 +       sb = cpg.dentry->d_sb;
12741 +       inode = d_inode(cpg.dentry);
12742 +       cpg.bsrc = au_fbtop(file);
12743 +       err = au_test_ro(sb, cpg.bsrc, inode);
12744 +       if (!err && (au_hf_top(file)->f_mode & FMODE_WRITE)) {
12745 +               err = au_pin(pin, cpg.dentry, cpg.bsrc, AuOpt_UDBA_NONE,
12746 +                            /*flags*/0);
12747 +               goto out;
12748 +       }
12749 +
12750 +       /* need to cpup or reopen */
12751 +       parent = dget_parent(cpg.dentry);
12752 +       di_write_lock_parent(parent);
12753 +       err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12754 +       cpg.bdst = err;
12755 +       if (unlikely(err < 0))
12756 +               goto out_dgrade;
12757 +       err = 0;
12758 +
12759 +       if (!d_unhashed(cpg.dentry) && !au_h_dptr(parent, cpg.bdst)) {
12760 +               err = au_cpup_dirs(cpg.dentry, cpg.bdst);
12761 +               if (unlikely(err))
12762 +                       goto out_dgrade;
12763 +       }
12764 +
12765 +       err = au_pin(pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12766 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12767 +       if (unlikely(err))
12768 +               goto out_dgrade;
12769 +
12770 +       dbtop = au_dbtop(cpg.dentry);
12771 +       if (dbtop <= cpg.bdst)
12772 +               cpg.bsrc = cpg.bdst;
12773 +
12774 +       if (dbtop <= cpg.bdst           /* just reopen */
12775 +           || !d_unhashed(cpg.dentry)  /* copyup and reopen */
12776 +               ) {
12777 +               h_file = au_h_open_pre(cpg.dentry, cpg.bsrc, /*force_wr*/0);
12778 +               if (IS_ERR(h_file))
12779 +                       err = PTR_ERR(h_file);
12780 +               else {
12781 +                       di_downgrade_lock(parent, AuLock_IR);
12782 +                       if (dbtop > cpg.bdst)
12783 +                               err = au_sio_cpup_simple(&cpg);
12784 +                       if (!err)
12785 +                               err = au_reopen_nondir(file);
12786 +                       au_h_open_post(cpg.dentry, cpg.bsrc, h_file);
12787 +               }
12788 +       } else {                        /* copyup as wh and reopen */
12789 +               /*
12790 +                * since writable hfsplus branch is not supported,
12791 +                * h_open_pre/post() are unnecessary.
12792 +                */
12793 +               err = au_ready_to_write_wh(file, len, cpg.bdst, pin);
12794 +               di_downgrade_lock(parent, AuLock_IR);
12795 +       }
12796 +
12797 +       if (!err) {
12798 +               au_pin_set_parent_lflag(pin, /*lflag*/0);
12799 +               goto out_dput; /* success */
12800 +       }
12801 +       au_unpin(pin);
12802 +       goto out_unlock;
12803 +
12804 +out_dgrade:
12805 +       di_downgrade_lock(parent, AuLock_IR);
12806 +out_unlock:
12807 +       di_read_unlock(parent, AuLock_IR);
12808 +out_dput:
12809 +       dput(parent);
12810 +out:
12811 +       return err;
12812 +}
12813 +
12814 +/* ---------------------------------------------------------------------- */
12815 +
12816 +int au_do_flush(struct file *file, fl_owner_t id,
12817 +               int (*flush)(struct file *file, fl_owner_t id))
12818 +{
12819 +       int err;
12820 +       struct super_block *sb;
12821 +       struct inode *inode;
12822 +
12823 +       inode = file_inode(file);
12824 +       sb = inode->i_sb;
12825 +       si_noflush_read_lock(sb);
12826 +       fi_read_lock(file);
12827 +       ii_read_lock_child(inode);
12828 +
12829 +       err = flush(file, id);
12830 +       au_cpup_attr_timesizes(inode);
12831 +
12832 +       ii_read_unlock(inode);
12833 +       fi_read_unlock(file);
12834 +       si_read_unlock(sb);
12835 +       return err;
12836 +}
12837 +
12838 +/* ---------------------------------------------------------------------- */
12839 +
12840 +static int au_file_refresh_by_inode(struct file *file, int *need_reopen)
12841 +{
12842 +       int err;
12843 +       struct au_pin pin;
12844 +       struct au_finfo *finfo;
12845 +       struct dentry *parent, *hi_wh;
12846 +       struct inode *inode;
12847 +       struct super_block *sb;
12848 +       struct au_cp_generic cpg = {
12849 +               .dentry = file->f_path.dentry,
12850 +               .bdst   = -1,
12851 +               .bsrc   = -1,
12852 +               .len    = -1,
12853 +               .pin    = &pin,
12854 +               .flags  = AuCpup_DTIME
12855 +       };
12856 +
12857 +       FiMustWriteLock(file);
12858 +
12859 +       err = 0;
12860 +       finfo = au_fi(file);
12861 +       sb = cpg.dentry->d_sb;
12862 +       inode = d_inode(cpg.dentry);
12863 +       cpg.bdst = au_ibtop(inode);
12864 +       if (cpg.bdst == finfo->fi_btop || IS_ROOT(cpg.dentry))
12865 +               goto out;
12866 +
12867 +       parent = dget_parent(cpg.dentry);
12868 +       if (au_test_ro(sb, cpg.bdst, inode)) {
12869 +               di_read_lock_parent(parent, !AuLock_IR);
12870 +               err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12871 +               cpg.bdst = err;
12872 +               di_read_unlock(parent, !AuLock_IR);
12873 +               if (unlikely(err < 0))
12874 +                       goto out_parent;
12875 +               err = 0;
12876 +       }
12877 +
12878 +       di_read_lock_parent(parent, AuLock_IR);
12879 +       hi_wh = au_hi_wh(inode, cpg.bdst);
12880 +       if (!S_ISDIR(inode->i_mode)
12881 +           && au_opt_test(au_mntflags(sb), PLINK)
12882 +           && au_plink_test(inode)
12883 +           && !d_unhashed(cpg.dentry)
12884 +           && cpg.bdst < au_dbtop(cpg.dentry)) {
12885 +               err = au_test_and_cpup_dirs(cpg.dentry, cpg.bdst);
12886 +               if (unlikely(err))
12887 +                       goto out_unlock;
12888 +
12889 +               /* always superio. */
12890 +               err = au_pin(&pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12891 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12892 +               if (!err) {
12893 +                       err = au_sio_cpup_simple(&cpg);
12894 +                       au_unpin(&pin);
12895 +               }
12896 +       } else if (hi_wh) {
12897 +               /* already copied-up after unlink */
12898 +               err = au_reopen_wh(file, cpg.bdst, hi_wh);
12899 +               *need_reopen = 0;
12900 +       }
12901 +
12902 +out_unlock:
12903 +       di_read_unlock(parent, AuLock_IR);
12904 +out_parent:
12905 +       dput(parent);
12906 +out:
12907 +       return err;
12908 +}
12909 +
12910 +static void au_do_refresh_dir(struct file *file)
12911 +{
12912 +       int execed;
12913 +       aufs_bindex_t bindex, bbot, new_bindex, brid;
12914 +       struct au_hfile *p, tmp, *q;
12915 +       struct au_finfo *finfo;
12916 +       struct super_block *sb;
12917 +       struct au_fidir *fidir;
12918 +
12919 +       FiMustWriteLock(file);
12920 +
12921 +       sb = file->f_path.dentry->d_sb;
12922 +       finfo = au_fi(file);
12923 +       fidir = finfo->fi_hdir;
12924 +       AuDebugOn(!fidir);
12925 +       p = fidir->fd_hfile + finfo->fi_btop;
12926 +       brid = p->hf_br->br_id;
12927 +       bbot = fidir->fd_bbot;
12928 +       for (bindex = finfo->fi_btop; bindex <= bbot; bindex++, p++) {
12929 +               if (!p->hf_file)
12930 +                       continue;
12931 +
12932 +               new_bindex = au_br_index(sb, p->hf_br->br_id);
12933 +               if (new_bindex == bindex)
12934 +                       continue;
12935 +               if (new_bindex < 0) {
12936 +                       au_set_h_fptr(file, bindex, NULL);
12937 +                       continue;
12938 +               }
12939 +
12940 +               /* swap two lower inode, and loop again */
12941 +               q = fidir->fd_hfile + new_bindex;
12942 +               tmp = *q;
12943 +               *q = *p;
12944 +               *p = tmp;
12945 +               if (tmp.hf_file) {
12946 +                       bindex--;
12947 +                       p--;
12948 +               }
12949 +       }
12950 +
12951 +       execed = vfsub_file_execed(file);
12952 +       p = fidir->fd_hfile;
12953 +       if (!au_test_mmapped(file) && !d_unlinked(file->f_path.dentry)) {
12954 +               bbot = au_sbbot(sb);
12955 +               for (finfo->fi_btop = 0; finfo->fi_btop <= bbot;
12956 +                    finfo->fi_btop++, p++)
12957 +                       if (p->hf_file) {
12958 +                               if (file_inode(p->hf_file))
12959 +                                       break;
12960 +                               au_hfput(p, execed);
12961 +                       }
12962 +       } else {
12963 +               bbot = au_br_index(sb, brid);
12964 +               for (finfo->fi_btop = 0; finfo->fi_btop < bbot;
12965 +                    finfo->fi_btop++, p++)
12966 +                       if (p->hf_file)
12967 +                               au_hfput(p, execed);
12968 +               bbot = au_sbbot(sb);
12969 +       }
12970 +
12971 +       p = fidir->fd_hfile + bbot;
12972 +       for (fidir->fd_bbot = bbot; fidir->fd_bbot >= finfo->fi_btop;
12973 +            fidir->fd_bbot--, p--)
12974 +               if (p->hf_file) {
12975 +                       if (file_inode(p->hf_file))
12976 +                               break;
12977 +                       au_hfput(p, execed);
12978 +               }
12979 +       AuDebugOn(fidir->fd_bbot < finfo->fi_btop);
12980 +}
12981 +
12982 +/*
12983 + * after branch manipulating, refresh the file.
12984 + */
12985 +static int refresh_file(struct file *file, int (*reopen)(struct file *file))
12986 +{
12987 +       int err, need_reopen, nbr;
12988 +       aufs_bindex_t bbot, bindex;
12989 +       struct dentry *dentry;
12990 +       struct super_block *sb;
12991 +       struct au_finfo *finfo;
12992 +       struct au_hfile *hfile;
12993 +
12994 +       dentry = file->f_path.dentry;
12995 +       sb = dentry->d_sb;
12996 +       nbr = au_sbbot(sb) + 1;
12997 +       finfo = au_fi(file);
12998 +       if (!finfo->fi_hdir) {
12999 +               hfile = &finfo->fi_htop;
13000 +               AuDebugOn(!hfile->hf_file);
13001 +               bindex = au_br_index(sb, hfile->hf_br->br_id);
13002 +               AuDebugOn(bindex < 0);
13003 +               if (bindex != finfo->fi_btop)
13004 +                       au_set_fbtop(file, bindex);
13005 +       } else {
13006 +               err = au_fidir_realloc(finfo, nbr, /*may_shrink*/0);
13007 +               if (unlikely(err))
13008 +                       goto out;
13009 +               au_do_refresh_dir(file);
13010 +       }
13011 +
13012 +       err = 0;
13013 +       need_reopen = 1;
13014 +       if (!au_test_mmapped(file))
13015 +               err = au_file_refresh_by_inode(file, &need_reopen);
13016 +       if (finfo->fi_hdir)
13017 +               /* harmless if err */
13018 +               au_fidir_realloc(finfo, nbr, /*may_shrink*/1);
13019 +       if (!err && need_reopen && !d_unlinked(dentry))
13020 +               err = reopen(file);
13021 +       if (!err) {
13022 +               au_update_figen(file);
13023 +               goto out; /* success */
13024 +       }
13025 +
13026 +       /* error, close all lower files */
13027 +       if (finfo->fi_hdir) {
13028 +               bbot = au_fbbot_dir(file);
13029 +               for (bindex = au_fbtop(file); bindex <= bbot; bindex++)
13030 +                       au_set_h_fptr(file, bindex, NULL);
13031 +       }
13032 +
13033 +out:
13034 +       return err;
13035 +}
13036 +
13037 +/* common function to regular file and dir */
13038 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
13039 +                         int wlock, unsigned int fi_lsc)
13040 +{
13041 +       int err;
13042 +       unsigned int sigen, figen;
13043 +       aufs_bindex_t btop;
13044 +       unsigned char pseudo_link;
13045 +       struct dentry *dentry;
13046 +       struct inode *inode;
13047 +
13048 +       err = 0;
13049 +       dentry = file->f_path.dentry;
13050 +       inode = d_inode(dentry);
13051 +       sigen = au_sigen(dentry->d_sb);
13052 +       fi_write_lock_nested(file, fi_lsc);
13053 +       figen = au_figen(file);
13054 +       if (!fi_lsc)
13055 +               di_write_lock_child(dentry);
13056 +       else
13057 +               di_write_lock_child2(dentry);
13058 +       btop = au_dbtop(dentry);
13059 +       pseudo_link = (btop != au_ibtop(inode));
13060 +       if (sigen == figen && !pseudo_link && au_fbtop(file) == btop) {
13061 +               if (!wlock) {
13062 +                       di_downgrade_lock(dentry, AuLock_IR);
13063 +                       fi_downgrade_lock(file);
13064 +               }
13065 +               goto out; /* success */
13066 +       }
13067 +
13068 +       AuDbg("sigen %d, figen %d\n", sigen, figen);
13069 +       if (au_digen_test(dentry, sigen)) {
13070 +               err = au_reval_dpath(dentry, sigen);
13071 +               AuDebugOn(!err && au_digen_test(dentry, sigen));
13072 +       }
13073 +
13074 +       if (!err)
13075 +               err = refresh_file(file, reopen);
13076 +       if (!err) {
13077 +               if (!wlock) {
13078 +                       di_downgrade_lock(dentry, AuLock_IR);
13079 +                       fi_downgrade_lock(file);
13080 +               }
13081 +       } else {
13082 +               di_write_unlock(dentry);
13083 +               fi_write_unlock(file);
13084 +       }
13085 +
13086 +out:
13087 +       return err;
13088 +}
13089 +
13090 +/* ---------------------------------------------------------------------- */
13091 +
13092 +/* cf. aufs_nopage() */
13093 +/* for madvise(2) */
13094 +static int aufs_readpage(struct file *file __maybe_unused, struct page *page)
13095 +{
13096 +       unlock_page(page);
13097 +       return 0;
13098 +}
13099 +
13100 +/* it will never be called, but necessary to support O_DIRECT */
13101 +static ssize_t aufs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
13102 +{ BUG(); return 0; }
13103 +
13104 +/* they will never be called. */
13105 +#ifdef CONFIG_AUFS_DEBUG
13106 +static int aufs_write_begin(struct file *file, struct address_space *mapping,
13107 +                           loff_t pos, unsigned len, unsigned flags,
13108 +                           struct page **pagep, void **fsdata)
13109 +{ AuUnsupport(); return 0; }
13110 +static int aufs_write_end(struct file *file, struct address_space *mapping,
13111 +                         loff_t pos, unsigned len, unsigned copied,
13112 +                         struct page *page, void *fsdata)
13113 +{ AuUnsupport(); return 0; }
13114 +static int aufs_writepage(struct page *page, struct writeback_control *wbc)
13115 +{ AuUnsupport(); return 0; }
13116 +
13117 +static int aufs_set_page_dirty(struct page *page)
13118 +{ AuUnsupport(); return 0; }
13119 +static void aufs_invalidatepage(struct page *page, unsigned int offset,
13120 +                               unsigned int length)
13121 +{ AuUnsupport(); }
13122 +static int aufs_releasepage(struct page *page, gfp_t gfp)
13123 +{ AuUnsupport(); return 0; }
13124 +#if 0 /* called by memory compaction regardless file */
13125 +static int aufs_migratepage(struct address_space *mapping, struct page *newpage,
13126 +                           struct page *page, enum migrate_mode mode)
13127 +{ AuUnsupport(); return 0; }
13128 +#endif
13129 +static bool aufs_isolate_page(struct page *page, isolate_mode_t mode)
13130 +{ AuUnsupport(); return true; }
13131 +static void aufs_putback_page(struct page *page)
13132 +{ AuUnsupport(); }
13133 +static int aufs_launder_page(struct page *page)
13134 +{ AuUnsupport(); return 0; }
13135 +static int aufs_is_partially_uptodate(struct page *page,
13136 +                                     unsigned long from,
13137 +                                     unsigned long count)
13138 +{ AuUnsupport(); return 0; }
13139 +static void aufs_is_dirty_writeback(struct page *page, bool *dirty,
13140 +                                   bool *writeback)
13141 +{ AuUnsupport(); }
13142 +static int aufs_error_remove_page(struct address_space *mapping,
13143 +                                 struct page *page)
13144 +{ AuUnsupport(); return 0; }
13145 +static int aufs_swap_activate(struct swap_info_struct *sis, struct file *file,
13146 +                             sector_t *span)
13147 +{ AuUnsupport(); return 0; }
13148 +static void aufs_swap_deactivate(struct file *file)
13149 +{ AuUnsupport(); }
13150 +#endif /* CONFIG_AUFS_DEBUG */
13151 +
13152 +const struct address_space_operations aufs_aop = {
13153 +       .readpage               = aufs_readpage,
13154 +       .direct_IO              = aufs_direct_IO,
13155 +#ifdef CONFIG_AUFS_DEBUG
13156 +       .writepage              = aufs_writepage,
13157 +       /* no writepages, because of writepage */
13158 +       .set_page_dirty         = aufs_set_page_dirty,
13159 +       /* no readpages, because of readpage */
13160 +       .write_begin            = aufs_write_begin,
13161 +       .write_end              = aufs_write_end,
13162 +       /* no bmap, no block device */
13163 +       .invalidatepage         = aufs_invalidatepage,
13164 +       .releasepage            = aufs_releasepage,
13165 +       /* is fallback_migrate_page ok? */
13166 +       /* .migratepage         = aufs_migratepage, */
13167 +       .isolate_page           = aufs_isolate_page,
13168 +       .putback_page           = aufs_putback_page,
13169 +       .launder_page           = aufs_launder_page,
13170 +       .is_partially_uptodate  = aufs_is_partially_uptodate,
13171 +       .is_dirty_writeback     = aufs_is_dirty_writeback,
13172 +       .error_remove_page      = aufs_error_remove_page,
13173 +       .swap_activate          = aufs_swap_activate,
13174 +       .swap_deactivate        = aufs_swap_deactivate
13175 +#endif /* CONFIG_AUFS_DEBUG */
13176 +};
13177 diff -urN /usr/share/empty/fs/aufs/file.h linux/fs/aufs/file.h
13178 --- /usr/share/empty/fs/aufs/file.h     1970-01-01 01:00:00.000000000 +0100
13179 +++ linux/fs/aufs/file.h        2017-05-06 22:16:52.778220093 +0200
13180 @@ -0,0 +1,333 @@
13181 +/*
13182 + * Copyright (C) 2005-2017 Junjiro R. Okajima
13183 + *
13184 + * This program, aufs is free software; you can redistribute it and/or modify
13185 + * it under the terms of the GNU General Public License as published by
13186 + * the Free Software Foundation; either version 2 of the License, or
13187 + * (at your option) any later version.
13188 + *
13189 + * This program is distributed in the hope that it will be useful,
13190 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13191 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13192 + * GNU General Public License for more details.
13193 + *
13194 + * You should have received a copy of the GNU General Public License
13195 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13196 + */
13197 +
13198 +/*
13199 + * file operations
13200 + */
13201 +
13202 +#ifndef __AUFS_FILE_H__
13203 +#define __AUFS_FILE_H__
13204 +
13205 +#ifdef __KERNEL__
13206 +
13207 +#include <linux/file.h>
13208 +#include <linux/fs.h>
13209 +#include <linux/poll.h>
13210 +#include "rwsem.h"
13211 +
13212 +struct au_branch;
13213 +struct au_hfile {
13214 +       struct file             *hf_file;
13215 +       struct au_branch        *hf_br;
13216 +};
13217 +
13218 +struct au_vdir;
13219 +struct au_fidir {
13220 +       aufs_bindex_t           fd_bbot;
13221 +       aufs_bindex_t           fd_nent;
13222 +       struct au_vdir          *fd_vdir_cache;
13223 +       struct au_hfile         fd_hfile[];
13224 +};
13225 +
13226 +static inline int au_fidir_sz(int nent)
13227 +{
13228 +       AuDebugOn(nent < 0);
13229 +       return sizeof(struct au_fidir) + sizeof(struct au_hfile) * nent;
13230 +}
13231 +
13232 +struct au_finfo {
13233 +       atomic_t                fi_generation;
13234 +
13235 +       struct au_rwsem         fi_rwsem;
13236 +       aufs_bindex_t           fi_btop;
13237 +
13238 +       /* do not union them */
13239 +       struct {                                /* for non-dir */
13240 +               struct au_hfile                 fi_htop;
13241 +               atomic_t                        fi_mmapped;
13242 +       };
13243 +       struct au_fidir         *fi_hdir;       /* for dir only */
13244 +
13245 +       struct hlist_node       fi_hlist;
13246 +       union {
13247 +               struct file             *fi_file;       /* very ugly */
13248 +               struct llist_node       fi_lnode;       /* delayed free */
13249 +       };
13250 +} ____cacheline_aligned_in_smp;
13251 +
13252 +/* ---------------------------------------------------------------------- */
13253 +
13254 +/* file.c */
13255 +extern const struct address_space_operations aufs_aop;
13256 +unsigned int au_file_roflags(unsigned int flags);
13257 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
13258 +                      struct file *file, int force_wr);
13259 +struct au_do_open_args {
13260 +       int             no_lock;
13261 +       int             (*open)(struct file *file, int flags,
13262 +                               struct file *h_file);
13263 +       struct au_fidir *fidir;
13264 +       struct file     *h_file;
13265 +};
13266 +int au_do_open(struct file *file, struct au_do_open_args *args);
13267 +int au_reopen_nondir(struct file *file);
13268 +struct au_pin;
13269 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin);
13270 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
13271 +                         int wlock, unsigned int fi_lsc);
13272 +int au_do_flush(struct file *file, fl_owner_t id,
13273 +               int (*flush)(struct file *file, fl_owner_t id));
13274 +
13275 +/* poll.c */
13276 +#ifdef CONFIG_AUFS_POLL
13277 +unsigned int aufs_poll(struct file *file, poll_table *wait);
13278 +#endif
13279 +
13280 +#ifdef CONFIG_AUFS_BR_HFSPLUS
13281 +/* hfsplus.c */
13282 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
13283 +                          int force_wr);
13284 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
13285 +                   struct file *h_file);
13286 +#else
13287 +AuStub(struct file *, au_h_open_pre, return NULL, struct dentry *dentry,
13288 +       aufs_bindex_t bindex, int force_wr)
13289 +AuStubVoid(au_h_open_post, struct dentry *dentry, aufs_bindex_t bindex,
13290 +          struct file *h_file);
13291 +#endif
13292 +
13293 +/* f_op.c */
13294 +extern const struct file_operations aufs_file_fop;
13295 +int au_do_open_nondir(struct file *file, int flags, struct file *h_file);
13296 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file);
13297 +struct file *au_read_pre(struct file *file, int keep_fi, unsigned int lsc);
13298 +
13299 +/* finfo.c */
13300 +void au_hfput(struct au_hfile *hf, int execed);
13301 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex,
13302 +                  struct file *h_file);
13303 +
13304 +void au_update_figen(struct file *file);
13305 +struct au_fidir *au_fidir_alloc(struct super_block *sb);
13306 +int au_fidir_realloc(struct au_finfo *finfo, int nbr, int may_shrink);
13307 +
13308 +void au_fi_init_once(void *_fi);
13309 +void au_finfo_fin(struct file *file, int atonce);
13310 +int au_finfo_init(struct file *file, struct au_fidir *fidir);
13311 +
13312 +/* ioctl.c */
13313 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg);
13314 +#ifdef CONFIG_COMPAT
13315 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
13316 +                          unsigned long arg);
13317 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
13318 +                             unsigned long arg);
13319 +#endif
13320 +
13321 +/* ---------------------------------------------------------------------- */
13322 +
13323 +static inline struct au_finfo *au_fi(struct file *file)
13324 +{
13325 +       return file->private_data;
13326 +}
13327 +
13328 +/* ---------------------------------------------------------------------- */
13329 +
13330 +/*
13331 + * fi_read_lock, fi_write_lock,
13332 + * fi_read_unlock, fi_write_unlock, fi_downgrade_lock
13333 + */
13334 +AuSimpleRwsemFuncs(fi, struct file *f, &au_fi(f)->fi_rwsem);
13335 +
13336 +/* lock subclass for finfo */
13337 +enum {
13338 +       AuLsc_FI_1,
13339 +       AuLsc_FI_2
13340 +};
13341 +
13342 +static inline void fi_read_lock_nested(struct file *f, unsigned int lsc)
13343 +{
13344 +       au_rw_read_lock_nested(&au_fi(f)->fi_rwsem, lsc);
13345 +}
13346 +
13347 +static inline void fi_write_lock_nested(struct file *f, unsigned int lsc)
13348 +{
13349 +       au_rw_write_lock_nested(&au_fi(f)->fi_rwsem, lsc);
13350 +}
13351 +
13352 +/*
13353 + * fi_read_lock_1, fi_write_lock_1,
13354 + * fi_read_lock_2, fi_write_lock_2
13355 + */
13356 +#define AuReadLockFunc(name) \
13357 +static inline void fi_read_lock_##name(struct file *f) \
13358 +{ fi_read_lock_nested(f, AuLsc_FI_##name); }
13359 +
13360 +#define AuWriteLockFunc(name) \
13361 +static inline void fi_write_lock_##name(struct file *f) \
13362 +{ fi_write_lock_nested(f, AuLsc_FI_##name); }
13363 +
13364 +#define AuRWLockFuncs(name) \
13365 +       AuReadLockFunc(name) \
13366 +       AuWriteLockFunc(name)
13367 +
13368 +AuRWLockFuncs(1);
13369 +AuRWLockFuncs(2);
13370 +
13371 +#undef AuReadLockFunc
13372 +#undef AuWriteLockFunc
13373 +#undef AuRWLockFuncs
13374 +
13375 +#define FiMustNoWaiters(f)     AuRwMustNoWaiters(&au_fi(f)->fi_rwsem)
13376 +#define FiMustAnyLock(f)       AuRwMustAnyLock(&au_fi(f)->fi_rwsem)
13377 +#define FiMustWriteLock(f)     AuRwMustWriteLock(&au_fi(f)->fi_rwsem)
13378 +
13379 +/* ---------------------------------------------------------------------- */
13380 +
13381 +/* todo: hard/soft set? */
13382 +static inline aufs_bindex_t au_fbtop(struct file *file)
13383 +{
13384 +       FiMustAnyLock(file);
13385 +       return au_fi(file)->fi_btop;
13386 +}
13387 +
13388 +static inline aufs_bindex_t au_fbbot_dir(struct file *file)
13389 +{
13390 +       FiMustAnyLock(file);
13391 +       AuDebugOn(!au_fi(file)->fi_hdir);
13392 +       return au_fi(file)->fi_hdir->fd_bbot;
13393 +}
13394 +
13395 +static inline struct au_vdir *au_fvdir_cache(struct file *file)
13396 +{
13397 +       FiMustAnyLock(file);
13398 +       AuDebugOn(!au_fi(file)->fi_hdir);
13399 +       return au_fi(file)->fi_hdir->fd_vdir_cache;
13400 +}
13401 +
13402 +static inline void au_set_fbtop(struct file *file, aufs_bindex_t bindex)
13403 +{
13404 +       FiMustWriteLock(file);
13405 +       au_fi(file)->fi_btop = bindex;
13406 +}
13407 +
13408 +static inline void au_set_fbbot_dir(struct file *file, aufs_bindex_t bindex)
13409 +{
13410 +       FiMustWriteLock(file);
13411 +       AuDebugOn(!au_fi(file)->fi_hdir);
13412 +       au_fi(file)->fi_hdir->fd_bbot = bindex;
13413 +}
13414 +
13415 +static inline void au_set_fvdir_cache(struct file *file,
13416 +                                     struct au_vdir *vdir_cache)
13417 +{
13418 +       FiMustWriteLock(file);
13419 +       AuDebugOn(!au_fi(file)->fi_hdir);
13420 +       au_fi(file)->fi_hdir->fd_vdir_cache = vdir_cache;
13421 +}
13422 +
13423 +static inline struct file *au_hf_top(struct file *file)
13424 +{
13425 +       FiMustAnyLock(file);
13426 +       AuDebugOn(au_fi(file)->fi_hdir);
13427 +       return au_fi(file)->fi_htop.hf_file;
13428 +}
13429 +
13430 +static inline struct file *au_hf_dir(struct file *file, aufs_bindex_t bindex)
13431 +{
13432 +       FiMustAnyLock(file);
13433 +       AuDebugOn(!au_fi(file)->fi_hdir);
13434 +       return au_fi(file)->fi_hdir->fd_hfile[0 + bindex].hf_file;
13435 +}
13436 +
13437 +/* todo: memory barrier? */
13438 +static inline unsigned int au_figen(struct file *f)
13439 +{
13440 +       return atomic_read(&au_fi(f)->fi_generation);
13441 +}
13442 +
13443 +static inline void au_set_mmapped(struct file *f)
13444 +{
13445 +       if (atomic_inc_return(&au_fi(f)->fi_mmapped))
13446 +               return;
13447 +       pr_warn("fi_mmapped wrapped around\n");
13448 +       while (!atomic_inc_return(&au_fi(f)->fi_mmapped))
13449 +               ;
13450 +}
13451 +
13452 +static inline void au_unset_mmapped(struct file *f)
13453 +{
13454 +       atomic_dec(&au_fi(f)->fi_mmapped);
13455 +}
13456 +
13457 +static inline int au_test_mmapped(struct file *f)
13458 +{
13459 +       return atomic_read(&au_fi(f)->fi_mmapped);
13460 +}
13461 +
13462 +/* customize vma->vm_file */
13463 +
13464 +static inline void au_do_vm_file_reset(struct vm_area_struct *vma,
13465 +                                      struct file *file)
13466 +{
13467 +       struct file *f;
13468 +
13469 +       f = vma->vm_file;
13470 +       get_file(file);
13471 +       vma->vm_file = file;
13472 +       fput(f);
13473 +}
13474 +
13475 +#ifdef CONFIG_MMU
13476 +#define AuDbgVmRegion(file, vma) do {} while (0)
13477 +
13478 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
13479 +                                   struct file *file)
13480 +{
13481 +       au_do_vm_file_reset(vma, file);
13482 +}
13483 +#else
13484 +#define AuDbgVmRegion(file, vma) \
13485 +       AuDebugOn((vma)->vm_region && (vma)->vm_region->vm_file != (file))
13486 +
13487 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
13488 +                                   struct file *file)
13489 +{
13490 +       struct file *f;
13491 +
13492 +       au_do_vm_file_reset(vma, file);
13493 +       f = vma->vm_region->vm_file;
13494 +       get_file(file);
13495 +       vma->vm_region->vm_file = file;
13496 +       fput(f);
13497 +}
13498 +#endif /* CONFIG_MMU */
13499 +
13500 +/* handle vma->vm_prfile */
13501 +static inline void au_vm_prfile_set(struct vm_area_struct *vma,
13502 +                                   struct file *file)
13503 +{
13504 +       get_file(file);
13505 +       vma->vm_prfile = file;
13506 +#ifndef CONFIG_MMU
13507 +       get_file(file);
13508 +       vma->vm_region->vm_prfile = file;
13509 +#endif
13510 +}
13511 +
13512 +#endif /* __KERNEL__ */
13513 +#endif /* __AUFS_FILE_H__ */
13514 diff -urN /usr/share/empty/fs/aufs/finfo.c linux/fs/aufs/finfo.c
13515 --- /usr/share/empty/fs/aufs/finfo.c    1970-01-01 01:00:00.000000000 +0100
13516 +++ linux/fs/aufs/finfo.c       2017-05-06 22:16:52.778220093 +0200
13517 @@ -0,0 +1,151 @@
13518 +/*
13519 + * Copyright (C) 2005-2017 Junjiro R. Okajima
13520 + *
13521 + * This program, aufs is free software; you can redistribute it and/or modify
13522 + * it under the terms of the GNU General Public License as published by
13523 + * the Free Software Foundation; either version 2 of the License, or
13524 + * (at your option) any later version.
13525 + *
13526 + * This program is distributed in the hope that it will be useful,
13527 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13528 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13529 + * GNU General Public License for more details.
13530 + *
13531 + * You should have received a copy of the GNU General Public License
13532 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13533 + */
13534 +
13535 +/*
13536 + * file private data
13537 + */
13538 +
13539 +#include "aufs.h"
13540 +
13541 +void au_hfput(struct au_hfile *hf, int execed)
13542 +{
13543 +       if (execed)
13544 +               allow_write_access(hf->hf_file);
13545 +       fput(hf->hf_file);
13546 +       hf->hf_file = NULL;
13547 +       au_br_put(hf->hf_br);
13548 +       hf->hf_br = NULL;
13549 +}
13550 +
13551 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex, struct file *val)
13552 +{
13553 +       struct au_finfo *finfo = au_fi(file);
13554 +       struct au_hfile *hf;
13555 +       struct au_fidir *fidir;
13556 +
13557 +       fidir = finfo->fi_hdir;
13558 +       if (!fidir) {
13559 +               AuDebugOn(finfo->fi_btop != bindex);
13560 +               hf = &finfo->fi_htop;
13561 +       } else
13562 +               hf = fidir->fd_hfile + bindex;
13563 +
13564 +       if (hf && hf->hf_file)
13565 +               au_hfput(hf, vfsub_file_execed(file));
13566 +       if (val) {
13567 +               FiMustWriteLock(file);
13568 +               AuDebugOn(IS_ERR_OR_NULL(file->f_path.dentry));
13569 +               hf->hf_file = val;
13570 +               hf->hf_br = au_sbr(file->f_path.dentry->d_sb, bindex);
13571 +       }
13572 +}
13573 +
13574 +void au_update_figen(struct file *file)
13575 +{
13576 +       atomic_set(&au_fi(file)->fi_generation, au_digen(file->f_path.dentry));
13577 +       /* smp_mb(); */ /* atomic_set */
13578 +}
13579 +
13580 +/* ---------------------------------------------------------------------- */
13581 +
13582 +struct au_fidir *au_fidir_alloc(struct super_block *sb)
13583 +{
13584 +       struct au_fidir *fidir;
13585 +       int nbr;
13586 +
13587 +       nbr = au_sbbot(sb) + 1;
13588 +       if (nbr < 2)
13589 +               nbr = 2; /* initial allocate for 2 branches */
13590 +       fidir = kzalloc(au_fidir_sz(nbr), GFP_NOFS);
13591 +       if (fidir) {
13592 +               fidir->fd_bbot = -1;
13593 +               fidir->fd_nent = nbr;
13594 +       }
13595 +
13596 +       return fidir;
13597 +}
13598 +
13599 +int au_fidir_realloc(struct au_finfo *finfo, int nbr, int may_shrink)
13600 +{
13601 +       int err;
13602 +       struct au_fidir *fidir, *p;
13603 +
13604 +       AuRwMustWriteLock(&finfo->fi_rwsem);
13605 +       fidir = finfo->fi_hdir;
13606 +       AuDebugOn(!fidir);
13607 +
13608 +       err = -ENOMEM;
13609 +       p = au_kzrealloc(fidir, au_fidir_sz(fidir->fd_nent), au_fidir_sz(nbr),
13610 +                        GFP_NOFS, may_shrink);
13611 +       if (p) {
13612 +               p->fd_nent = nbr;
13613 +               finfo->fi_hdir = p;
13614 +               err = 0;
13615 +       }
13616 +
13617 +       return err;
13618 +}
13619 +
13620 +/* ---------------------------------------------------------------------- */
13621 +
13622 +void au_finfo_fin(struct file *file, int atonce)
13623 +{
13624 +       struct au_finfo *finfo;
13625 +
13626 +       au_nfiles_dec(file->f_path.dentry->d_sb);
13627 +
13628 +       finfo = au_fi(file);
13629 +       AuDebugOn(finfo->fi_hdir);
13630 +       AuRwDestroy(&finfo->fi_rwsem);
13631 +       if (!atonce)
13632 +               au_cache_dfree_finfo(finfo);
13633 +       else
13634 +               au_cache_free_finfo(finfo);
13635 +}
13636 +
13637 +void au_fi_init_once(void *_finfo)
13638 +{
13639 +       struct au_finfo *finfo = _finfo;
13640 +
13641 +       au_rw_init(&finfo->fi_rwsem);
13642 +}
13643 +
13644 +int au_finfo_init(struct file *file, struct au_fidir *fidir)
13645 +{
13646 +       int err;
13647 +       struct au_finfo *finfo;
13648 +       struct dentry *dentry;
13649 +
13650 +       err = -ENOMEM;
13651 +       dentry = file->f_path.dentry;
13652 +       finfo = au_cache_alloc_finfo();
13653 +       if (unlikely(!finfo))
13654 +               goto out;
13655 +
13656 +       err = 0;
13657 +       au_nfiles_inc(dentry->d_sb);
13658 +       au_rw_write_lock(&finfo->fi_rwsem);
13659 +       finfo->fi_btop = -1;
13660 +       finfo->fi_hdir = fidir;
13661 +       atomic_set(&finfo->fi_generation, au_digen(dentry));
13662 +       /* smp_mb(); */ /* atomic_set */
13663 +
13664 +       file->private_data = finfo;
13665 +
13666 +out:
13667 +       return err;
13668 +}
13669 diff -urN /usr/share/empty/fs/aufs/f_op.c linux/fs/aufs/f_op.c
13670 --- /usr/share/empty/fs/aufs/f_op.c     1970-01-01 01:00:00.000000000 +0100
13671 +++ linux/fs/aufs/f_op.c        2017-05-06 22:16:52.778220093 +0200
13672 @@ -0,0 +1,819 @@
13673 +/*
13674 + * Copyright (C) 2005-2017 Junjiro R. Okajima
13675 + *
13676 + * This program, aufs is free software; you can redistribute it and/or modify
13677 + * it under the terms of the GNU General Public License as published by
13678 + * the Free Software Foundation; either version 2 of the License, or
13679 + * (at your option) any later version.
13680 + *
13681 + * This program is distributed in the hope that it will be useful,
13682 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13683 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13684 + * GNU General Public License for more details.
13685 + *
13686 + * You should have received a copy of the GNU General Public License
13687 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13688 + */
13689 +
13690 +/*
13691 + * file and vm operations
13692 + */
13693 +
13694 +#include <linux/aio.h>
13695 +#include <linux/fs_stack.h>
13696 +#include <linux/mman.h>
13697 +#include <linux/security.h>
13698 +#include "aufs.h"
13699 +
13700 +int au_do_open_nondir(struct file *file, int flags, struct file *h_file)
13701 +{
13702 +       int err;
13703 +       aufs_bindex_t bindex;
13704 +       struct dentry *dentry, *h_dentry;
13705 +       struct au_finfo *finfo;
13706 +       struct inode *h_inode;
13707 +
13708 +       FiMustWriteLock(file);
13709 +
13710 +       err = 0;
13711 +       dentry = file->f_path.dentry;
13712 +       AuDebugOn(IS_ERR_OR_NULL(dentry));
13713 +       finfo = au_fi(file);
13714 +       memset(&finfo->fi_htop, 0, sizeof(finfo->fi_htop));
13715 +       atomic_set(&finfo->fi_mmapped, 0);
13716 +       bindex = au_dbtop(dentry);
13717 +       if (!h_file) {
13718 +               h_dentry = au_h_dptr(dentry, bindex);
13719 +               err = vfsub_test_mntns(file->f_path.mnt, h_dentry->d_sb);
13720 +               if (unlikely(err))
13721 +                       goto out;
13722 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
13723 +       } else {
13724 +               h_dentry = h_file->f_path.dentry;
13725 +               err = vfsub_test_mntns(file->f_path.mnt, h_dentry->d_sb);
13726 +               if (unlikely(err))
13727 +                       goto out;
13728 +               get_file(h_file);
13729 +       }
13730 +       if (IS_ERR(h_file))
13731 +               err = PTR_ERR(h_file);
13732 +       else {
13733 +               if ((flags & __O_TMPFILE)
13734 +                   && !(flags & O_EXCL)) {
13735 +                       h_inode = file_inode(h_file);
13736 +                       spin_lock(&h_inode->i_lock);
13737 +                       h_inode->i_state |= I_LINKABLE;
13738 +                       spin_unlock(&h_inode->i_lock);
13739 +               }
13740 +               au_set_fbtop(file, bindex);
13741 +               au_set_h_fptr(file, bindex, h_file);
13742 +               au_update_figen(file);
13743 +               /* todo: necessary? */
13744 +               /* file->f_ra = h_file->f_ra; */
13745 +       }
13746 +
13747 +out:
13748 +       return err;
13749 +}
13750 +
13751 +static int aufs_open_nondir(struct inode *inode __maybe_unused,
13752 +                           struct file *file)
13753 +{
13754 +       int err;
13755 +       struct super_block *sb;
13756 +       struct au_do_open_args args = {
13757 +               .open   = au_do_open_nondir
13758 +       };
13759 +
13760 +       AuDbg("%pD, f_flags 0x%x, f_mode 0x%x\n",
13761 +             file, vfsub_file_flags(file), file->f_mode);
13762 +
13763 +       sb = file->f_path.dentry->d_sb;
13764 +       si_read_lock(sb, AuLock_FLUSH);
13765 +       err = au_do_open(file, &args);
13766 +       si_read_unlock(sb);
13767 +       return err;
13768 +}
13769 +
13770 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file)
13771 +{
13772 +       struct au_finfo *finfo;
13773 +       aufs_bindex_t bindex;
13774 +       int delayed;
13775 +
13776 +       finfo = au_fi(file);
13777 +       au_sphl_del(&finfo->fi_hlist,
13778 +                   &au_sbi(file->f_path.dentry->d_sb)->si_files);
13779 +       bindex = finfo->fi_btop;
13780 +       if (bindex >= 0)
13781 +               au_set_h_fptr(file, bindex, NULL);
13782 +
13783 +       delayed = (current->flags & PF_KTHREAD) || in_interrupt();
13784 +       au_finfo_fin(file, delayed);
13785 +       return 0;
13786 +}
13787 +
13788 +/* ---------------------------------------------------------------------- */
13789 +
13790 +static int au_do_flush_nondir(struct file *file, fl_owner_t id)
13791 +{
13792 +       int err;
13793 +       struct file *h_file;
13794 +
13795 +       err = 0;
13796 +       h_file = au_hf_top(file);
13797 +       if (h_file)
13798 +               err = vfsub_flush(h_file, id);
13799 +       return err;
13800 +}
13801 +
13802 +static int aufs_flush_nondir(struct file *file, fl_owner_t id)
13803 +{
13804 +       return au_do_flush(file, id, au_do_flush_nondir);
13805 +}
13806 +
13807 +/* ---------------------------------------------------------------------- */
13808 +/*
13809 + * read and write functions acquire [fdi]_rwsem once, but release before
13810 + * mmap_sem. This is because to stop a race condition between mmap(2).
13811 + * Releasing these aufs-rwsem should be safe, no branch-mamagement (by keeping
13812 + * si_rwsem), no harmful copy-up should happen. Actually copy-up may happen in
13813 + * read functions after [fdi]_rwsem are released, but it should be harmless.
13814 + */
13815 +
13816 +/* Callers should call au_read_post() or fput() in the end */
13817 +struct file *au_read_pre(struct file *file, int keep_fi, unsigned int lsc)
13818 +{
13819 +       struct file *h_file;
13820 +       int err;
13821 +
13822 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0, lsc);
13823 +       if (!err) {
13824 +               di_read_unlock(file->f_path.dentry, AuLock_IR);
13825 +               h_file = au_hf_top(file);
13826 +               get_file(h_file);
13827 +               if (!keep_fi)
13828 +                       fi_read_unlock(file);
13829 +       } else
13830 +               h_file = ERR_PTR(err);
13831 +
13832 +       return h_file;
13833 +}
13834 +
13835 +static void au_read_post(struct inode *inode, struct file *h_file)
13836 +{
13837 +       /* update without lock, I don't think it a problem */
13838 +       fsstack_copy_attr_atime(inode, file_inode(h_file));
13839 +       fput(h_file);
13840 +}
13841 +
13842 +struct au_write_pre {
13843 +       /* input */
13844 +       unsigned int lsc;
13845 +
13846 +       /* output */
13847 +       blkcnt_t blks;
13848 +       aufs_bindex_t btop;
13849 +};
13850 +
13851 +/*
13852 + * return with iinfo is write-locked
13853 + * callers should call au_write_post() or iinfo_write_unlock() + fput() in the
13854 + * end
13855 + */
13856 +static struct file *au_write_pre(struct file *file, int do_ready,
13857 +                                struct au_write_pre *wpre)
13858 +{
13859 +       struct file *h_file;
13860 +       struct dentry *dentry;
13861 +       int err;
13862 +       unsigned int lsc;
13863 +       struct au_pin pin;
13864 +
13865 +       lsc = 0;
13866 +       if (wpre)
13867 +               lsc = wpre->lsc;
13868 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1, lsc);
13869 +       h_file = ERR_PTR(err);
13870 +       if (unlikely(err))
13871 +               goto out;
13872 +
13873 +       dentry = file->f_path.dentry;
13874 +       if (do_ready) {
13875 +               err = au_ready_to_write(file, -1, &pin);
13876 +               if (unlikely(err)) {
13877 +                       h_file = ERR_PTR(err);
13878 +                       di_write_unlock(dentry);
13879 +                       goto out_fi;
13880 +               }
13881 +       }
13882 +
13883 +       di_downgrade_lock(dentry, /*flags*/0);
13884 +       if (wpre)
13885 +               wpre->btop = au_fbtop(file);
13886 +       h_file = au_hf_top(file);
13887 +       get_file(h_file);
13888 +       if (wpre)
13889 +               wpre->blks = file_inode(h_file)->i_blocks;
13890 +       if (do_ready)
13891 +               au_unpin(&pin);
13892 +       di_read_unlock(dentry, /*flags*/0);
13893 +
13894 +out_fi:
13895 +       fi_write_unlock(file);
13896 +out:
13897 +       return h_file;
13898 +}
13899 +
13900 +static void au_write_post(struct inode *inode, struct file *h_file,
13901 +                         struct au_write_pre *wpre, ssize_t written)
13902 +{
13903 +       struct inode *h_inode;
13904 +
13905 +       au_cpup_attr_timesizes(inode);
13906 +       AuDebugOn(au_ibtop(inode) != wpre->btop);
13907 +       h_inode = file_inode(h_file);
13908 +       inode->i_mode = h_inode->i_mode;
13909 +       ii_write_unlock(inode);
13910 +       fput(h_file);
13911 +
13912 +       /* AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks); */
13913 +       if (written > 0)
13914 +               au_fhsm_wrote(inode->i_sb, wpre->btop,
13915 +                             /*force*/h_inode->i_blocks > wpre->blks);
13916 +}
13917 +
13918 +static ssize_t aufs_read(struct file *file, char __user *buf, size_t count,
13919 +                        loff_t *ppos)
13920 +{
13921 +       ssize_t err;
13922 +       struct inode *inode;
13923 +       struct file *h_file;
13924 +       struct super_block *sb;
13925 +
13926 +       inode = file_inode(file);
13927 +       sb = inode->i_sb;
13928 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13929 +
13930 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
13931 +       err = PTR_ERR(h_file);
13932 +       if (IS_ERR(h_file))
13933 +               goto out;
13934 +
13935 +       /* filedata may be obsoleted by concurrent copyup, but no problem */
13936 +       err = vfsub_read_u(h_file, buf, count, ppos);
13937 +       /* todo: necessary? */
13938 +       /* file->f_ra = h_file->f_ra; */
13939 +       au_read_post(inode, h_file);
13940 +
13941 +out:
13942 +       si_read_unlock(sb);
13943 +       return err;
13944 +}
13945 +
13946 +/*
13947 + * todo: very ugly
13948 + * it locks both of i_mutex and si_rwsem for read in safe.
13949 + * if the plink maintenance mode continues forever (that is the problem),
13950 + * may loop forever.
13951 + */
13952 +static void au_mtx_and_read_lock(struct inode *inode)
13953 +{
13954 +       int err;
13955 +       struct super_block *sb = inode->i_sb;
13956 +
13957 +       while (1) {
13958 +               inode_lock(inode);
13959 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
13960 +               if (!err)
13961 +                       break;
13962 +               inode_unlock(inode);
13963 +               si_read_lock(sb, AuLock_NOPLMW);
13964 +               si_read_unlock(sb);
13965 +       }
13966 +}
13967 +
13968 +static ssize_t aufs_write(struct file *file, const char __user *ubuf,
13969 +                         size_t count, loff_t *ppos)
13970 +{
13971 +       ssize_t err;
13972 +       struct au_write_pre wpre;
13973 +       struct inode *inode;
13974 +       struct file *h_file;
13975 +       char __user *buf = (char __user *)ubuf;
13976 +
13977 +       inode = file_inode(file);
13978 +       au_mtx_and_read_lock(inode);
13979 +
13980 +       wpre.lsc = 0;
13981 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
13982 +       err = PTR_ERR(h_file);
13983 +       if (IS_ERR(h_file))
13984 +               goto out;
13985 +
13986 +       err = vfsub_write_u(h_file, buf, count, ppos);
13987 +       au_write_post(inode, h_file, &wpre, err);
13988 +
13989 +out:
13990 +       si_read_unlock(inode->i_sb);
13991 +       inode_unlock(inode);
13992 +       return err;
13993 +}
13994 +
13995 +static ssize_t au_do_iter(struct file *h_file, int rw, struct kiocb *kio,
13996 +                         struct iov_iter *iov_iter)
13997 +{
13998 +       ssize_t err;
13999 +       struct file *file;
14000 +       ssize_t (*iter)(struct kiocb *, struct iov_iter *);
14001 +
14002 +       err = security_file_permission(h_file, rw);
14003 +       if (unlikely(err))
14004 +               goto out;
14005 +
14006 +       err = -ENOSYS;
14007 +       iter = NULL;
14008 +       if (rw == MAY_READ)
14009 +               iter = h_file->f_op->read_iter;
14010 +       else if (rw == MAY_WRITE)
14011 +               iter = h_file->f_op->write_iter;
14012 +
14013 +       file = kio->ki_filp;
14014 +       kio->ki_filp = h_file;
14015 +       if (iter) {
14016 +               lockdep_off();
14017 +               err = iter(kio, iov_iter);
14018 +               lockdep_on();
14019 +       } else
14020 +               /* currently there is no such fs */
14021 +               WARN_ON_ONCE(1);
14022 +       kio->ki_filp = file;
14023 +
14024 +out:
14025 +       return err;
14026 +}
14027 +
14028 +static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
14029 +{
14030 +       ssize_t err;
14031 +       struct file *file, *h_file;
14032 +       struct inode *inode;
14033 +       struct super_block *sb;
14034 +
14035 +       file = kio->ki_filp;
14036 +       inode = file_inode(file);
14037 +       sb = inode->i_sb;
14038 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
14039 +
14040 +       h_file = au_read_pre(file, /*keep_fi*/1, /*lsc*/0);
14041 +       err = PTR_ERR(h_file);
14042 +       if (IS_ERR(h_file))
14043 +               goto out;
14044 +
14045 +       if (au_test_loopback_kthread()) {
14046 +               au_warn_loopback(h_file->f_path.dentry->d_sb);
14047 +               if (file->f_mapping != h_file->f_mapping) {
14048 +                       file->f_mapping = h_file->f_mapping;
14049 +                       smp_mb(); /* unnecessary? */
14050 +               }
14051 +       }
14052 +       fi_read_unlock(file);
14053 +
14054 +       err = au_do_iter(h_file, MAY_READ, kio, iov_iter);
14055 +       /* todo: necessary? */
14056 +       /* file->f_ra = h_file->f_ra; */
14057 +       au_read_post(inode, h_file);
14058 +
14059 +out:
14060 +       si_read_unlock(sb);
14061 +       return err;
14062 +}
14063 +
14064 +static ssize_t aufs_write_iter(struct kiocb *kio, struct iov_iter *iov_iter)
14065 +{
14066 +       ssize_t err;
14067 +       struct au_write_pre wpre;
14068 +       struct inode *inode;
14069 +       struct file *file, *h_file;
14070 +
14071 +       file = kio->ki_filp;
14072 +       inode = file_inode(file);
14073 +       au_mtx_and_read_lock(inode);
14074 +
14075 +       wpre.lsc = 0;
14076 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14077 +       err = PTR_ERR(h_file);
14078 +       if (IS_ERR(h_file))
14079 +               goto out;
14080 +
14081 +       err = au_do_iter(h_file, MAY_WRITE, kio, iov_iter);
14082 +       au_write_post(inode, h_file, &wpre, err);
14083 +
14084 +out:
14085 +       si_read_unlock(inode->i_sb);
14086 +       inode_unlock(inode);
14087 +       return err;
14088 +}
14089 +
14090 +static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
14091 +                               struct pipe_inode_info *pipe, size_t len,
14092 +                               unsigned int flags)
14093 +{
14094 +       ssize_t err;
14095 +       struct file *h_file;
14096 +       struct inode *inode;
14097 +       struct super_block *sb;
14098 +
14099 +       inode = file_inode(file);
14100 +       sb = inode->i_sb;
14101 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
14102 +
14103 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
14104 +       err = PTR_ERR(h_file);
14105 +       if (IS_ERR(h_file))
14106 +               goto out;
14107 +
14108 +       err = vfsub_splice_to(h_file, ppos, pipe, len, flags);
14109 +       /* todo: necessasry? */
14110 +       /* file->f_ra = h_file->f_ra; */
14111 +       au_read_post(inode, h_file);
14112 +
14113 +out:
14114 +       si_read_unlock(sb);
14115 +       return err;
14116 +}
14117 +
14118 +static ssize_t
14119 +aufs_splice_write(struct pipe_inode_info *pipe, struct file *file, loff_t *ppos,
14120 +                 size_t len, unsigned int flags)
14121 +{
14122 +       ssize_t err;
14123 +       struct au_write_pre wpre;
14124 +       struct inode *inode;
14125 +       struct file *h_file;
14126 +
14127 +       inode = file_inode(file);
14128 +       au_mtx_and_read_lock(inode);
14129 +
14130 +       wpre.lsc = 0;
14131 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14132 +       err = PTR_ERR(h_file);
14133 +       if (IS_ERR(h_file))
14134 +               goto out;
14135 +
14136 +       err = vfsub_splice_from(pipe, h_file, ppos, len, flags);
14137 +       au_write_post(inode, h_file, &wpre, err);
14138 +
14139 +out:
14140 +       si_read_unlock(inode->i_sb);
14141 +       inode_unlock(inode);
14142 +       return err;
14143 +}
14144 +
14145 +static long aufs_fallocate(struct file *file, int mode, loff_t offset,
14146 +                          loff_t len)
14147 +{
14148 +       long err;
14149 +       struct au_write_pre wpre;
14150 +       struct inode *inode;
14151 +       struct file *h_file;
14152 +
14153 +       inode = file_inode(file);
14154 +       au_mtx_and_read_lock(inode);
14155 +
14156 +       wpre.lsc = 0;
14157 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14158 +       err = PTR_ERR(h_file);
14159 +       if (IS_ERR(h_file))
14160 +               goto out;
14161 +
14162 +       lockdep_off();
14163 +       err = vfs_fallocate(h_file, mode, offset, len);
14164 +       lockdep_on();
14165 +       au_write_post(inode, h_file, &wpre, /*written*/1);
14166 +
14167 +out:
14168 +       si_read_unlock(inode->i_sb);
14169 +       inode_unlock(inode);
14170 +       return err;
14171 +}
14172 +
14173 +static ssize_t aufs_copy_file_range(struct file *src, loff_t src_pos,
14174 +                                   struct file *dst, loff_t dst_pos,
14175 +                                   size_t len, unsigned int flags)
14176 +{
14177 +       ssize_t err;
14178 +       struct au_write_pre wpre;
14179 +       enum { SRC, DST };
14180 +       struct {
14181 +               struct inode *inode;
14182 +               struct file *h_file;
14183 +               struct super_block *h_sb;
14184 +       } a[2];
14185 +#define a_src  a[SRC]
14186 +#define a_dst  a[DST]
14187 +
14188 +       err = -EINVAL;
14189 +       a_src.inode = file_inode(src);
14190 +       if (unlikely(!S_ISREG(a_src.inode->i_mode)))
14191 +               goto out;
14192 +       a_dst.inode = file_inode(dst);
14193 +       if (unlikely(!S_ISREG(a_dst.inode->i_mode)))
14194 +               goto out;
14195 +
14196 +       au_mtx_and_read_lock(a_dst.inode);
14197 +       /*
14198 +        * in order to match the order in di_write_lock2_{child,parent}(),
14199 +        * use f_path.dentry for this comparision.
14200 +        */
14201 +       if (src->f_path.dentry < dst->f_path.dentry) {
14202 +               a_src.h_file = au_read_pre(src, /*keep_fi*/1, AuLsc_FI_1);
14203 +               err = PTR_ERR(a_src.h_file);
14204 +               if (IS_ERR(a_src.h_file))
14205 +                       goto out_si;
14206 +
14207 +               wpre.lsc = AuLsc_FI_2;
14208 +               a_dst.h_file = au_write_pre(dst, /*do_ready*/1, &wpre);
14209 +               err = PTR_ERR(a_dst.h_file);
14210 +               if (IS_ERR(a_dst.h_file)) {
14211 +                       au_read_post(a_src.inode, a_src.h_file);
14212 +                       goto out_si;
14213 +               }
14214 +       } else {
14215 +               wpre.lsc = AuLsc_FI_1;
14216 +               a_dst.h_file = au_write_pre(dst, /*do_ready*/1, &wpre);
14217 +               err = PTR_ERR(a_dst.h_file);
14218 +               if (IS_ERR(a_dst.h_file))
14219 +                       goto out_si;
14220 +
14221 +               a_src.h_file = au_read_pre(src, /*keep_fi*/1, AuLsc_FI_2);
14222 +               err = PTR_ERR(a_src.h_file);
14223 +               if (IS_ERR(a_src.h_file)) {
14224 +                       au_write_post(a_dst.inode, a_dst.h_file, &wpre,
14225 +                                     /*written*/0);
14226 +                       goto out_si;
14227 +               }
14228 +       }
14229 +
14230 +       err = -EXDEV;
14231 +       a_src.h_sb = file_inode(a_src.h_file)->i_sb;
14232 +       a_dst.h_sb = file_inode(a_dst.h_file)->i_sb;
14233 +       if (unlikely(a_src.h_sb != a_dst.h_sb)) {
14234 +               AuDbgFile(src);
14235 +               AuDbgFile(dst);
14236 +               goto out_file;
14237 +       }
14238 +
14239 +       err = vfsub_copy_file_range(a_src.h_file, src_pos, a_dst.h_file,
14240 +                                   dst_pos, len, flags);
14241 +
14242 +out_file:
14243 +       au_write_post(a_dst.inode, a_dst.h_file, &wpre, err);
14244 +       fi_read_unlock(src);
14245 +       au_read_post(a_src.inode, a_src.h_file);
14246 +out_si:
14247 +       si_read_unlock(a_dst.inode->i_sb);
14248 +       inode_unlock(a_dst.inode);
14249 +out:
14250 +       return err;
14251 +#undef a_src
14252 +#undef a_dst
14253 +}
14254 +
14255 +/* ---------------------------------------------------------------------- */
14256 +
14257 +/*
14258 + * The locking order around current->mmap_sem.
14259 + * - in most and regular cases
14260 + *   file I/O syscall -- aufs_read() or something
14261 + *     -- si_rwsem for read -- mmap_sem
14262 + *     (Note that [fdi]i_rwsem are released before mmap_sem).
14263 + * - in mmap case
14264 + *   mmap(2) -- mmap_sem -- aufs_mmap() -- si_rwsem for read -- [fdi]i_rwsem
14265 + * This AB-BA order is definitly bad, but is not a problem since "si_rwsem for
14266 + * read" allows muliple processes to acquire it and [fdi]i_rwsem are not held in
14267 + * file I/O. Aufs needs to stop lockdep in aufs_mmap() though.
14268 + * It means that when aufs acquires si_rwsem for write, the process should never
14269 + * acquire mmap_sem.
14270 + *
14271 + * Actually aufs_iterate() holds [fdi]i_rwsem before mmap_sem, but this is not a
14272 + * problem either since any directory is not able to be mmap-ed.
14273 + * The similar scenario is applied to aufs_readlink() too.
14274 + */
14275 +
14276 +#if 0 /* stop calling security_file_mmap() */
14277 +/* cf. linux/include/linux/mman.h: calc_vm_prot_bits() */
14278 +#define AuConv_VM_PROT(f, b)   _calc_vm_trans(f, VM_##b, PROT_##b)
14279 +
14280 +static unsigned long au_arch_prot_conv(unsigned long flags)
14281 +{
14282 +       /* currently ppc64 only */
14283 +#ifdef CONFIG_PPC64
14284 +       /* cf. linux/arch/powerpc/include/asm/mman.h */
14285 +       AuDebugOn(arch_calc_vm_prot_bits(-1) != VM_SAO);
14286 +       return AuConv_VM_PROT(flags, SAO);
14287 +#else
14288 +       AuDebugOn(arch_calc_vm_prot_bits(-1));
14289 +       return 0;
14290 +#endif
14291 +}
14292 +
14293 +static unsigned long au_prot_conv(unsigned long flags)
14294 +{
14295 +       return AuConv_VM_PROT(flags, READ)
14296 +               | AuConv_VM_PROT(flags, WRITE)
14297 +               | AuConv_VM_PROT(flags, EXEC)
14298 +               | au_arch_prot_conv(flags);
14299 +}
14300 +
14301 +/* cf. linux/include/linux/mman.h: calc_vm_flag_bits() */
14302 +#define AuConv_VM_MAP(f, b)    _calc_vm_trans(f, VM_##b, MAP_##b)
14303 +
14304 +static unsigned long au_flag_conv(unsigned long flags)
14305 +{
14306 +       return AuConv_VM_MAP(flags, GROWSDOWN)
14307 +               | AuConv_VM_MAP(flags, DENYWRITE)
14308 +               | AuConv_VM_MAP(flags, LOCKED);
14309 +}
14310 +#endif
14311 +
14312 +static int aufs_mmap(struct file *file, struct vm_area_struct *vma)
14313 +{
14314 +       int err;
14315 +       const unsigned char wlock
14316 +               = (file->f_mode & FMODE_WRITE) && (vma->vm_flags & VM_SHARED);
14317 +       struct super_block *sb;
14318 +       struct file *h_file;
14319 +       struct inode *inode;
14320 +
14321 +       AuDbgVmRegion(file, vma);
14322 +
14323 +       inode = file_inode(file);
14324 +       sb = inode->i_sb;
14325 +       lockdep_off();
14326 +       si_read_lock(sb, AuLock_NOPLMW);
14327 +
14328 +       h_file = au_write_pre(file, wlock, /*wpre*/NULL);
14329 +       lockdep_on();
14330 +       err = PTR_ERR(h_file);
14331 +       if (IS_ERR(h_file))
14332 +               goto out;
14333 +
14334 +       err = 0;
14335 +       au_set_mmapped(file);
14336 +       au_vm_file_reset(vma, h_file);
14337 +       /*
14338 +        * we cannot call security_mmap_file() here since it may acquire
14339 +        * mmap_sem or i_mutex.
14340 +        *
14341 +        * err = security_mmap_file(h_file, au_prot_conv(vma->vm_flags),
14342 +        *                       au_flag_conv(vma->vm_flags));
14343 +        */
14344 +       if (!err)
14345 +               err = call_mmap(h_file, vma);
14346 +       if (!err) {
14347 +               au_vm_prfile_set(vma, file);
14348 +               fsstack_copy_attr_atime(inode, file_inode(h_file));
14349 +               goto out_fput; /* success */
14350 +       }
14351 +       au_unset_mmapped(file);
14352 +       au_vm_file_reset(vma, file);
14353 +
14354 +out_fput:
14355 +       lockdep_off();
14356 +       ii_write_unlock(inode);
14357 +       lockdep_on();
14358 +       fput(h_file);
14359 +out:
14360 +       lockdep_off();
14361 +       si_read_unlock(sb);
14362 +       lockdep_on();
14363 +       AuTraceErr(err);
14364 +       return err;
14365 +}
14366 +
14367 +/* ---------------------------------------------------------------------- */
14368 +
14369 +static int aufs_fsync_nondir(struct file *file, loff_t start, loff_t end,
14370 +                            int datasync)
14371 +{
14372 +       int err;
14373 +       struct au_write_pre wpre;
14374 +       struct inode *inode;
14375 +       struct file *h_file;
14376 +
14377 +       err = 0; /* -EBADF; */ /* posix? */
14378 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
14379 +               goto out;
14380 +
14381 +       inode = file_inode(file);
14382 +       au_mtx_and_read_lock(inode);
14383 +
14384 +       wpre.lsc = 0;
14385 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14386 +       err = PTR_ERR(h_file);
14387 +       if (IS_ERR(h_file))
14388 +               goto out_unlock;
14389 +
14390 +       err = vfsub_fsync(h_file, &h_file->f_path, datasync);
14391 +       au_write_post(inode, h_file, &wpre, /*written*/0);
14392 +
14393 +out_unlock:
14394 +       si_read_unlock(inode->i_sb);
14395 +       inode_unlock(inode);
14396 +out:
14397 +       return err;
14398 +}
14399 +
14400 +static int aufs_fasync(int fd, struct file *file, int flag)
14401 +{
14402 +       int err;
14403 +       struct file *h_file;
14404 +       struct super_block *sb;
14405 +
14406 +       sb = file->f_path.dentry->d_sb;
14407 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
14408 +
14409 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
14410 +       err = PTR_ERR(h_file);
14411 +       if (IS_ERR(h_file))
14412 +               goto out;
14413 +
14414 +       if (h_file->f_op->fasync)
14415 +               err = h_file->f_op->fasync(fd, h_file, flag);
14416 +       fput(h_file); /* instead of au_read_post() */
14417 +
14418 +out:
14419 +       si_read_unlock(sb);
14420 +       return err;
14421 +}
14422 +
14423 +static int aufs_setfl(struct file *file, unsigned long arg)
14424 +{
14425 +       int err;
14426 +       struct file *h_file;
14427 +       struct super_block *sb;
14428 +
14429 +       sb = file->f_path.dentry->d_sb;
14430 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
14431 +
14432 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
14433 +       err = PTR_ERR(h_file);
14434 +       if (IS_ERR(h_file))
14435 +               goto out;
14436 +
14437 +       arg |= vfsub_file_flags(file) & FASYNC; /* stop calling h_file->fasync */
14438 +       err = setfl(/*unused fd*/-1, h_file, arg);
14439 +       fput(h_file); /* instead of au_read_post() */
14440 +
14441 +out:
14442 +       si_read_unlock(sb);
14443 +       return err;
14444 +}
14445 +
14446 +/* ---------------------------------------------------------------------- */
14447 +
14448 +/* no one supports this operation, currently */
14449 +#if 0
14450 +static ssize_t aufs_sendpage(struct file *file, struct page *page, int offset,
14451 +                            size_t len, loff_t *pos, int more)
14452 +{
14453 +}
14454 +#endif
14455 +
14456 +/* ---------------------------------------------------------------------- */
14457 +
14458 +const struct file_operations aufs_file_fop = {
14459 +       .owner          = THIS_MODULE,
14460 +
14461 +       .llseek         = default_llseek,
14462 +
14463 +       .read           = aufs_read,
14464 +       .write          = aufs_write,
14465 +       .read_iter      = aufs_read_iter,
14466 +       .write_iter     = aufs_write_iter,
14467 +
14468 +#ifdef CONFIG_AUFS_POLL
14469 +       .poll           = aufs_poll,
14470 +#endif
14471 +       .unlocked_ioctl = aufs_ioctl_nondir,
14472 +#ifdef CONFIG_COMPAT
14473 +       .compat_ioctl   = aufs_compat_ioctl_nondir,
14474 +#endif
14475 +       .mmap           = aufs_mmap,
14476 +       .open           = aufs_open_nondir,
14477 +       .flush          = aufs_flush_nondir,
14478 +       .release        = aufs_release_nondir,
14479 +       .fsync          = aufs_fsync_nondir,
14480 +       .fasync         = aufs_fasync,
14481 +       /* .sendpage    = aufs_sendpage, */
14482 +       .setfl          = aufs_setfl,
14483 +       .splice_write   = aufs_splice_write,
14484 +       .splice_read    = aufs_splice_read,
14485 +#if 0
14486 +       .aio_splice_write = aufs_aio_splice_write,
14487 +       .aio_splice_read  = aufs_aio_splice_read,
14488 +#endif
14489 +       .fallocate      = aufs_fallocate,
14490 +       .copy_file_range = aufs_copy_file_range
14491 +};
14492 diff -urN /usr/share/empty/fs/aufs/fstype.h linux/fs/aufs/fstype.h
14493 --- /usr/share/empty/fs/aufs/fstype.h   1970-01-01 01:00:00.000000000 +0100
14494 +++ linux/fs/aufs/fstype.h      2017-05-06 22:16:52.778220093 +0200
14495 @@ -0,0 +1,400 @@
14496 +/*
14497 + * Copyright (C) 2005-2017 Junjiro R. Okajima
14498 + *
14499 + * This program, aufs is free software; you can redistribute it and/or modify
14500 + * it under the terms of the GNU General Public License as published by
14501 + * the Free Software Foundation; either version 2 of the License, or
14502 + * (at your option) any later version.
14503 + *
14504 + * This program is distributed in the hope that it will be useful,
14505 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14506 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14507 + * GNU General Public License for more details.
14508 + *
14509 + * You should have received a copy of the GNU General Public License
14510 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14511 + */
14512 +
14513 +/*
14514 + * judging filesystem type
14515 + */
14516 +
14517 +#ifndef __AUFS_FSTYPE_H__
14518 +#define __AUFS_FSTYPE_H__
14519 +
14520 +#ifdef __KERNEL__
14521 +
14522 +#include <linux/fs.h>
14523 +#include <linux/magic.h>
14524 +#include <linux/nfs_fs.h>
14525 +#include <linux/romfs_fs.h>
14526 +
14527 +static inline int au_test_aufs(struct super_block *sb)
14528 +{
14529 +       return sb->s_magic == AUFS_SUPER_MAGIC;
14530 +}
14531 +
14532 +static inline const char *au_sbtype(struct super_block *sb)
14533 +{
14534 +       return sb->s_type->name;
14535 +}
14536 +
14537 +static inline int au_test_iso9660(struct super_block *sb __maybe_unused)
14538 +{
14539 +#if IS_ENABLED(CONFIG_ISO9660_FS)
14540 +       return sb->s_magic == ISOFS_SUPER_MAGIC;
14541 +#else
14542 +       return 0;
14543 +#endif
14544 +}
14545 +
14546 +static inline int au_test_romfs(struct super_block *sb __maybe_unused)
14547 +{
14548 +#if IS_ENABLED(CONFIG_ROMFS_FS)
14549 +       return sb->s_magic == ROMFS_MAGIC;
14550 +#else
14551 +       return 0;
14552 +#endif
14553 +}
14554 +
14555 +static inline int au_test_cramfs(struct super_block *sb __maybe_unused)
14556 +{
14557 +#if IS_ENABLED(CONFIG_CRAMFS)
14558 +       return sb->s_magic == CRAMFS_MAGIC;
14559 +#endif
14560 +       return 0;
14561 +}
14562 +
14563 +static inline int au_test_nfs(struct super_block *sb __maybe_unused)
14564 +{
14565 +#if IS_ENABLED(CONFIG_NFS_FS)
14566 +       return sb->s_magic == NFS_SUPER_MAGIC;
14567 +#else
14568 +       return 0;
14569 +#endif
14570 +}
14571 +
14572 +static inline int au_test_fuse(struct super_block *sb __maybe_unused)
14573 +{
14574 +#if IS_ENABLED(CONFIG_FUSE_FS)
14575 +       return sb->s_magic == FUSE_SUPER_MAGIC;
14576 +#else
14577 +       return 0;
14578 +#endif
14579 +}
14580 +
14581 +static inline int au_test_xfs(struct super_block *sb __maybe_unused)
14582 +{
14583 +#if IS_ENABLED(CONFIG_XFS_FS)
14584 +       return sb->s_magic == XFS_SB_MAGIC;
14585 +#else
14586 +       return 0;
14587 +#endif
14588 +}
14589 +
14590 +static inline int au_test_tmpfs(struct super_block *sb __maybe_unused)
14591 +{
14592 +#ifdef CONFIG_TMPFS
14593 +       return sb->s_magic == TMPFS_MAGIC;
14594 +#else
14595 +       return 0;
14596 +#endif
14597 +}
14598 +
14599 +static inline int au_test_ecryptfs(struct super_block *sb __maybe_unused)
14600 +{
14601 +#if IS_ENABLED(CONFIG_ECRYPT_FS)
14602 +       return !strcmp(au_sbtype(sb), "ecryptfs");
14603 +#else
14604 +       return 0;
14605 +#endif
14606 +}
14607 +
14608 +static inline int au_test_ramfs(struct super_block *sb)
14609 +{
14610 +       return sb->s_magic == RAMFS_MAGIC;
14611 +}
14612 +
14613 +static inline int au_test_ubifs(struct super_block *sb __maybe_unused)
14614 +{
14615 +#if IS_ENABLED(CONFIG_UBIFS_FS)
14616 +       return sb->s_magic == UBIFS_SUPER_MAGIC;
14617 +#else
14618 +       return 0;
14619 +#endif
14620 +}
14621 +
14622 +static inline int au_test_procfs(struct super_block *sb __maybe_unused)
14623 +{
14624 +#ifdef CONFIG_PROC_FS
14625 +       return sb->s_magic == PROC_SUPER_MAGIC;
14626 +#else
14627 +       return 0;
14628 +#endif
14629 +}
14630 +
14631 +static inline int au_test_sysfs(struct super_block *sb __maybe_unused)
14632 +{
14633 +#ifdef CONFIG_SYSFS
14634 +       return sb->s_magic == SYSFS_MAGIC;
14635 +#else
14636 +       return 0;
14637 +#endif
14638 +}
14639 +
14640 +static inline int au_test_configfs(struct super_block *sb __maybe_unused)
14641 +{
14642 +#if IS_ENABLED(CONFIG_CONFIGFS_FS)
14643 +       return sb->s_magic == CONFIGFS_MAGIC;
14644 +#else
14645 +       return 0;
14646 +#endif
14647 +}
14648 +
14649 +static inline int au_test_minix(struct super_block *sb __maybe_unused)
14650 +{
14651 +#if IS_ENABLED(CONFIG_MINIX_FS)
14652 +       return sb->s_magic == MINIX3_SUPER_MAGIC
14653 +               || sb->s_magic == MINIX2_SUPER_MAGIC
14654 +               || sb->s_magic == MINIX2_SUPER_MAGIC2
14655 +               || sb->s_magic == MINIX_SUPER_MAGIC
14656 +               || sb->s_magic == MINIX_SUPER_MAGIC2;
14657 +#else
14658 +       return 0;
14659 +#endif
14660 +}
14661 +
14662 +static inline int au_test_fat(struct super_block *sb __maybe_unused)
14663 +{
14664 +#if IS_ENABLED(CONFIG_FAT_FS)
14665 +       return sb->s_magic == MSDOS_SUPER_MAGIC;
14666 +#else
14667 +       return 0;
14668 +#endif
14669 +}
14670 +
14671 +static inline int au_test_msdos(struct super_block *sb)
14672 +{
14673 +       return au_test_fat(sb);
14674 +}
14675 +
14676 +static inline int au_test_vfat(struct super_block *sb)
14677 +{
14678 +       return au_test_fat(sb);
14679 +}
14680 +
14681 +static inline int au_test_securityfs(struct super_block *sb __maybe_unused)
14682 +{
14683 +#ifdef CONFIG_SECURITYFS
14684 +       return sb->s_magic == SECURITYFS_MAGIC;
14685 +#else
14686 +       return 0;
14687 +#endif
14688 +}
14689 +
14690 +static inline int au_test_squashfs(struct super_block *sb __maybe_unused)
14691 +{
14692 +#if IS_ENABLED(CONFIG_SQUASHFS)
14693 +       return sb->s_magic == SQUASHFS_MAGIC;
14694 +#else
14695 +       return 0;
14696 +#endif
14697 +}
14698 +
14699 +static inline int au_test_btrfs(struct super_block *sb __maybe_unused)
14700 +{
14701 +#if IS_ENABLED(CONFIG_BTRFS_FS)
14702 +       return sb->s_magic == BTRFS_SUPER_MAGIC;
14703 +#else
14704 +       return 0;
14705 +#endif
14706 +}
14707 +
14708 +static inline int au_test_xenfs(struct super_block *sb __maybe_unused)
14709 +{
14710 +#if IS_ENABLED(CONFIG_XENFS)
14711 +       return sb->s_magic == XENFS_SUPER_MAGIC;
14712 +#else
14713 +       return 0;
14714 +#endif
14715 +}
14716 +
14717 +static inline int au_test_debugfs(struct super_block *sb __maybe_unused)
14718 +{
14719 +#ifdef CONFIG_DEBUG_FS
14720 +       return sb->s_magic == DEBUGFS_MAGIC;
14721 +#else
14722 +       return 0;
14723 +#endif
14724 +}
14725 +
14726 +static inline int au_test_nilfs(struct super_block *sb __maybe_unused)
14727 +{
14728 +#if IS_ENABLED(CONFIG_NILFS)
14729 +       return sb->s_magic == NILFS_SUPER_MAGIC;
14730 +#else
14731 +       return 0;
14732 +#endif
14733 +}
14734 +
14735 +static inline int au_test_hfsplus(struct super_block *sb __maybe_unused)
14736 +{
14737 +#if IS_ENABLED(CONFIG_HFSPLUS_FS)
14738 +       return sb->s_magic == HFSPLUS_SUPER_MAGIC;
14739 +#else
14740 +       return 0;
14741 +#endif
14742 +}
14743 +
14744 +/* ---------------------------------------------------------------------- */
14745 +/*
14746 + * they can't be an aufs branch.
14747 + */
14748 +static inline int au_test_fs_unsuppoted(struct super_block *sb)
14749 +{
14750 +       return
14751 +#ifndef CONFIG_AUFS_BR_RAMFS
14752 +               au_test_ramfs(sb) ||
14753 +#endif
14754 +               au_test_procfs(sb)
14755 +               || au_test_sysfs(sb)
14756 +               || au_test_configfs(sb)
14757 +               || au_test_debugfs(sb)
14758 +               || au_test_securityfs(sb)
14759 +               || au_test_xenfs(sb)
14760 +               || au_test_ecryptfs(sb)
14761 +               /* || !strcmp(au_sbtype(sb), "unionfs") */
14762 +               || au_test_aufs(sb); /* will be supported in next version */
14763 +}
14764 +
14765 +static inline int au_test_fs_remote(struct super_block *sb)
14766 +{
14767 +       return !au_test_tmpfs(sb)
14768 +#ifdef CONFIG_AUFS_BR_RAMFS
14769 +               && !au_test_ramfs(sb)
14770 +#endif
14771 +               && !(sb->s_type->fs_flags & FS_REQUIRES_DEV);
14772 +}
14773 +
14774 +/* ---------------------------------------------------------------------- */
14775 +
14776 +/*
14777 + * Note: these functions (below) are created after reading ->getattr() in all
14778 + * filesystems under linux/fs. it means we have to do so in every update...
14779 + */
14780 +
14781 +/*
14782 + * some filesystems require getattr to refresh the inode attributes before
14783 + * referencing.
14784 + * in most cases, we can rely on the inode attribute in NFS (or every remote fs)
14785 + * and leave the work for d_revalidate()
14786 + */
14787 +static inline int au_test_fs_refresh_iattr(struct super_block *sb)
14788 +{
14789 +       return au_test_nfs(sb)
14790 +               || au_test_fuse(sb)
14791 +               /* || au_test_btrfs(sb) */      /* untested */
14792 +               ;
14793 +}
14794 +
14795 +/*
14796 + * filesystems which don't maintain i_size or i_blocks.
14797 + */
14798 +static inline int au_test_fs_bad_iattr_size(struct super_block *sb)
14799 +{
14800 +       return au_test_xfs(sb)
14801 +               || au_test_btrfs(sb)
14802 +               || au_test_ubifs(sb)
14803 +               || au_test_hfsplus(sb)  /* maintained, but incorrect */
14804 +               /* || au_test_minix(sb) */      /* untested */
14805 +               ;
14806 +}
14807 +
14808 +/*
14809 + * filesystems which don't store the correct value in some of their inode
14810 + * attributes.
14811 + */
14812 +static inline int au_test_fs_bad_iattr(struct super_block *sb)
14813 +{
14814 +       return au_test_fs_bad_iattr_size(sb)
14815 +               || au_test_fat(sb)
14816 +               || au_test_msdos(sb)
14817 +               || au_test_vfat(sb);
14818 +}
14819 +
14820 +/* they don't check i_nlink in link(2) */
14821 +static inline int au_test_fs_no_limit_nlink(struct super_block *sb)
14822 +{
14823 +       return au_test_tmpfs(sb)
14824 +#ifdef CONFIG_AUFS_BR_RAMFS
14825 +               || au_test_ramfs(sb)
14826 +#endif
14827 +               || au_test_ubifs(sb)
14828 +               || au_test_hfsplus(sb);
14829 +}
14830 +
14831 +/*
14832 + * filesystems which sets S_NOATIME and S_NOCMTIME.
14833 + */
14834 +static inline int au_test_fs_notime(struct super_block *sb)
14835 +{
14836 +       return au_test_nfs(sb)
14837 +               || au_test_fuse(sb)
14838 +               || au_test_ubifs(sb)
14839 +               ;
14840 +}
14841 +
14842 +/* temporary support for i#1 in cramfs */
14843 +static inline int au_test_fs_unique_ino(struct inode *inode)
14844 +{
14845 +       if (au_test_cramfs(inode->i_sb))
14846 +               return inode->i_ino != 1;
14847 +       return 1;
14848 +}
14849 +
14850 +/* ---------------------------------------------------------------------- */
14851 +
14852 +/*
14853 + * the filesystem where the xino files placed must support i/o after unlink and
14854 + * maintain i_size and i_blocks.
14855 + */
14856 +static inline int au_test_fs_bad_xino(struct super_block *sb)
14857 +{
14858 +       return au_test_fs_remote(sb)
14859 +               || au_test_fs_bad_iattr_size(sb)
14860 +               /* don't want unnecessary work for xino */
14861 +               || au_test_aufs(sb)
14862 +               || au_test_ecryptfs(sb)
14863 +               || au_test_nilfs(sb);
14864 +}
14865 +
14866 +static inline int au_test_fs_trunc_xino(struct super_block *sb)
14867 +{
14868 +       return au_test_tmpfs(sb)
14869 +               || au_test_ramfs(sb);
14870 +}
14871 +
14872 +/*
14873 + * test if the @sb is real-readonly.
14874 + */
14875 +static inline int au_test_fs_rr(struct super_block *sb)
14876 +{
14877 +       return au_test_squashfs(sb)
14878 +               || au_test_iso9660(sb)
14879 +               || au_test_cramfs(sb)
14880 +               || au_test_romfs(sb);
14881 +}
14882 +
14883 +/*
14884 + * test if the @inode is nfs with 'noacl' option
14885 + * NFS always sets MS_POSIXACL regardless its mount option 'noacl.'
14886 + */
14887 +static inline int au_test_nfs_noacl(struct inode *inode)
14888 +{
14889 +       return au_test_nfs(inode->i_sb)
14890 +               /* && IS_POSIXACL(inode) */
14891 +               && !nfs_server_capable(inode, NFS_CAP_ACLS);
14892 +}
14893 +
14894 +#endif /* __KERNEL__ */
14895 +#endif /* __AUFS_FSTYPE_H__ */
14896 diff -urN /usr/share/empty/fs/aufs/hfsnotify.c linux/fs/aufs/hfsnotify.c
14897 --- /usr/share/empty/fs/aufs/hfsnotify.c        1970-01-01 01:00:00.000000000 +0100
14898 +++ linux/fs/aufs/hfsnotify.c   2017-05-06 22:16:52.778220093 +0200
14899 @@ -0,0 +1,287 @@
14900 +/*
14901 + * Copyright (C) 2005-2017 Junjiro R. Okajima
14902 + *
14903 + * This program, aufs is free software; you can redistribute it and/or modify
14904 + * it under the terms of the GNU General Public License as published by
14905 + * the Free Software Foundation; either version 2 of the License, or
14906 + * (at your option) any later version.
14907 + *
14908 + * This program is distributed in the hope that it will be useful,
14909 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14910 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14911 + * GNU General Public License for more details.
14912 + *
14913 + * You should have received a copy of the GNU General Public License
14914 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14915 + */
14916 +
14917 +/*
14918 + * fsnotify for the lower directories
14919 + */
14920 +
14921 +#include "aufs.h"
14922 +
14923 +/* FS_IN_IGNORED is unnecessary */
14924 +static const __u32 AuHfsnMask = (FS_MOVED_TO | FS_MOVED_FROM | FS_DELETE
14925 +                                | FS_CREATE | FS_EVENT_ON_CHILD);
14926 +static DECLARE_WAIT_QUEUE_HEAD(au_hfsn_wq);
14927 +static __cacheline_aligned_in_smp atomic64_t au_hfsn_ifree = ATOMIC64_INIT(0);
14928 +
14929 +static void au_hfsn_free_mark(struct fsnotify_mark *mark)
14930 +{
14931 +       struct au_hnotify *hn = container_of(mark, struct au_hnotify,
14932 +                                            hn_mark);
14933 +       /* AuDbg("here\n"); */
14934 +       au_cache_dfree_hnotify(hn);
14935 +       smp_mb__before_atomic();
14936 +       if (atomic64_dec_and_test(&au_hfsn_ifree))
14937 +               wake_up(&au_hfsn_wq);
14938 +}
14939 +
14940 +static int au_hfsn_alloc(struct au_hinode *hinode)
14941 +{
14942 +       int err;
14943 +       struct au_hnotify *hn;
14944 +       struct super_block *sb;
14945 +       struct au_branch *br;
14946 +       struct fsnotify_mark *mark;
14947 +       aufs_bindex_t bindex;
14948 +
14949 +       hn = hinode->hi_notify;
14950 +       sb = hn->hn_aufs_inode->i_sb;
14951 +       bindex = au_br_index(sb, hinode->hi_id);
14952 +       br = au_sbr(sb, bindex);
14953 +       AuDebugOn(!br->br_hfsn);
14954 +
14955 +       mark = &hn->hn_mark;
14956 +       fsnotify_init_mark(mark, au_hfsn_free_mark);
14957 +       mark->mask = AuHfsnMask;
14958 +       /*
14959 +        * by udba rename or rmdir, aufs assign a new inode to the known
14960 +        * h_inode, so specify 1 to allow dups.
14961 +        */
14962 +       lockdep_off();
14963 +       err = fsnotify_add_mark(mark, br->br_hfsn->hfsn_group, hinode->hi_inode,
14964 +                                /*mnt*/NULL, /*allow_dups*/1);
14965 +       lockdep_on();
14966 +
14967 +       return err;
14968 +}
14969 +
14970 +static int au_hfsn_free(struct au_hinode *hinode, struct au_hnotify *hn)
14971 +{
14972 +       struct fsnotify_mark *mark;
14973 +       unsigned long long ull;
14974 +       struct fsnotify_group *group;
14975 +
14976 +       ull = atomic64_inc_return(&au_hfsn_ifree);
14977 +       BUG_ON(!ull);
14978 +
14979 +       mark = &hn->hn_mark;
14980 +       spin_lock(&mark->lock);
14981 +       group = mark->group;
14982 +       fsnotify_get_group(group);
14983 +       spin_unlock(&mark->lock);
14984 +       lockdep_off();
14985 +       fsnotify_destroy_mark(mark, group);
14986 +       fsnotify_put_mark(mark);
14987 +       fsnotify_put_group(group);
14988 +       lockdep_on();
14989 +
14990 +       /* free hn by myself */
14991 +       return 0;
14992 +}
14993 +
14994 +/* ---------------------------------------------------------------------- */
14995 +
14996 +static void au_hfsn_ctl(struct au_hinode *hinode, int do_set)
14997 +{
14998 +       struct fsnotify_mark *mark;
14999 +
15000 +       mark = &hinode->hi_notify->hn_mark;
15001 +       spin_lock(&mark->lock);
15002 +       if (do_set) {
15003 +               AuDebugOn(mark->mask & AuHfsnMask);
15004 +               mark->mask |= AuHfsnMask;
15005 +       } else {
15006 +               AuDebugOn(!(mark->mask & AuHfsnMask));
15007 +               mark->mask &= ~AuHfsnMask;
15008 +       }
15009 +       spin_unlock(&mark->lock);
15010 +       /* fsnotify_recalc_inode_mask(hinode->hi_inode); */
15011 +}
15012 +
15013 +/* ---------------------------------------------------------------------- */
15014 +
15015 +/* #define AuDbgHnotify */
15016 +#ifdef AuDbgHnotify
15017 +static char *au_hfsn_name(u32 mask)
15018 +{
15019 +#ifdef CONFIG_AUFS_DEBUG
15020 +#define test_ret(flag)                         \
15021 +       do {                                    \
15022 +               if (mask & flag)                \
15023 +                       return #flag;           \
15024 +       } while (0)
15025 +       test_ret(FS_ACCESS);
15026 +       test_ret(FS_MODIFY);
15027 +       test_ret(FS_ATTRIB);
15028 +       test_ret(FS_CLOSE_WRITE);
15029 +       test_ret(FS_CLOSE_NOWRITE);
15030 +       test_ret(FS_OPEN);
15031 +       test_ret(FS_MOVED_FROM);
15032 +       test_ret(FS_MOVED_TO);
15033 +       test_ret(FS_CREATE);
15034 +       test_ret(FS_DELETE);
15035 +       test_ret(FS_DELETE_SELF);
15036 +       test_ret(FS_MOVE_SELF);
15037 +       test_ret(FS_UNMOUNT);
15038 +       test_ret(FS_Q_OVERFLOW);
15039 +       test_ret(FS_IN_IGNORED);
15040 +       test_ret(FS_ISDIR);
15041 +       test_ret(FS_IN_ONESHOT);
15042 +       test_ret(FS_EVENT_ON_CHILD);
15043 +       return "";
15044 +#undef test_ret
15045 +#else
15046 +       return "??";
15047 +#endif
15048 +}
15049 +#endif
15050 +
15051 +/* ---------------------------------------------------------------------- */
15052 +
15053 +static void au_hfsn_free_group(struct fsnotify_group *group)
15054 +{
15055 +       struct au_br_hfsnotify *hfsn = group->private;
15056 +
15057 +       /* AuDbg("here\n"); */
15058 +       au_delayed_kfree(hfsn);
15059 +}
15060 +
15061 +static int au_hfsn_handle_event(struct fsnotify_group *group,
15062 +                               struct inode *inode,
15063 +                               struct fsnotify_mark *inode_mark,
15064 +                               struct fsnotify_mark *vfsmount_mark,
15065 +                               u32 mask, const void *data, int data_type,
15066 +                               const unsigned char *file_name, u32 cookie)
15067 +{
15068 +       int err;
15069 +       struct au_hnotify *hnotify;
15070 +       struct inode *h_dir, *h_inode;
15071 +       struct qstr h_child_qstr = QSTR_INIT(file_name, strlen(file_name));
15072 +
15073 +       AuDebugOn(data_type != FSNOTIFY_EVENT_INODE);
15074 +
15075 +       err = 0;
15076 +       /* if FS_UNMOUNT happens, there must be another bug */
15077 +       AuDebugOn(mask & FS_UNMOUNT);
15078 +       if (mask & (FS_IN_IGNORED | FS_UNMOUNT))
15079 +               goto out;
15080 +
15081 +       h_dir = inode;
15082 +       h_inode = NULL;
15083 +#ifdef AuDbgHnotify
15084 +       au_debug_on();
15085 +       if (1 || h_child_qstr.len != sizeof(AUFS_XINO_FNAME) - 1
15086 +           || strncmp(h_child_qstr.name, AUFS_XINO_FNAME, h_child_qstr.len)) {
15087 +               AuDbg("i%lu, mask 0x%x %s, hcname %.*s, hi%lu\n",
15088 +                     h_dir->i_ino, mask, au_hfsn_name(mask),
15089 +                     AuLNPair(&h_child_qstr), h_inode ? h_inode->i_ino : 0);
15090 +               /* WARN_ON(1); */
15091 +       }
15092 +       au_debug_off();
15093 +#endif
15094 +
15095 +       AuDebugOn(!inode_mark);
15096 +       hnotify = container_of(inode_mark, struct au_hnotify, hn_mark);
15097 +       err = au_hnotify(h_dir, hnotify, mask, &h_child_qstr, h_inode);
15098 +
15099 +out:
15100 +       return err;
15101 +}
15102 +
15103 +static struct fsnotify_ops au_hfsn_ops = {
15104 +       .handle_event           = au_hfsn_handle_event,
15105 +       .free_group_priv        = au_hfsn_free_group
15106 +};
15107 +
15108 +/* ---------------------------------------------------------------------- */
15109 +
15110 +static void au_hfsn_fin_br(struct au_branch *br)
15111 +{
15112 +       struct au_br_hfsnotify *hfsn;
15113 +
15114 +       hfsn = br->br_hfsn;
15115 +       if (hfsn) {
15116 +               lockdep_off();
15117 +               fsnotify_put_group(hfsn->hfsn_group);
15118 +               lockdep_on();
15119 +       }
15120 +}
15121 +
15122 +static int au_hfsn_init_br(struct au_branch *br, int perm)
15123 +{
15124 +       int err;
15125 +       struct fsnotify_group *group;
15126 +       struct au_br_hfsnotify *hfsn;
15127 +
15128 +       err = 0;
15129 +       br->br_hfsn = NULL;
15130 +       if (!au_br_hnotifyable(perm))
15131 +               goto out;
15132 +
15133 +       err = -ENOMEM;
15134 +       hfsn = kmalloc(sizeof(*hfsn), GFP_NOFS);
15135 +       if (unlikely(!hfsn))
15136 +               goto out;
15137 +
15138 +       err = 0;
15139 +       group = fsnotify_alloc_group(&au_hfsn_ops);
15140 +       if (IS_ERR(group)) {
15141 +               err = PTR_ERR(group);
15142 +               pr_err("fsnotify_alloc_group() failed, %d\n", err);
15143 +               goto out_hfsn;
15144 +       }
15145 +
15146 +       group->private = hfsn;
15147 +       hfsn->hfsn_group = group;
15148 +       br->br_hfsn = hfsn;
15149 +       goto out; /* success */
15150 +
15151 +out_hfsn:
15152 +       au_delayed_kfree(hfsn);
15153 +out:
15154 +       return err;
15155 +}
15156 +
15157 +static int au_hfsn_reset_br(unsigned int udba, struct au_branch *br, int perm)
15158 +{
15159 +       int err;
15160 +
15161 +       err = 0;
15162 +       if (!br->br_hfsn)
15163 +               err = au_hfsn_init_br(br, perm);
15164 +
15165 +       return err;
15166 +}
15167 +
15168 +/* ---------------------------------------------------------------------- */
15169 +
15170 +static void au_hfsn_fin(void)
15171 +{
15172 +       AuDbg("au_hfsn_ifree %lld\n", (long long)atomic64_read(&au_hfsn_ifree));
15173 +       wait_event(au_hfsn_wq, !atomic64_read(&au_hfsn_ifree));
15174 +}
15175 +
15176 +const struct au_hnotify_op au_hnotify_op = {
15177 +       .ctl            = au_hfsn_ctl,
15178 +       .alloc          = au_hfsn_alloc,
15179 +       .free           = au_hfsn_free,
15180 +
15181 +       .fin            = au_hfsn_fin,
15182 +
15183 +       .reset_br       = au_hfsn_reset_br,
15184 +       .fin_br         = au_hfsn_fin_br,
15185 +       .init_br        = au_hfsn_init_br
15186 +};
15187 diff -urN /usr/share/empty/fs/aufs/hfsplus.c linux/fs/aufs/hfsplus.c
15188 --- /usr/share/empty/fs/aufs/hfsplus.c  1970-01-01 01:00:00.000000000 +0100
15189 +++ linux/fs/aufs/hfsplus.c     2017-05-06 22:16:52.778220093 +0200
15190 @@ -0,0 +1,56 @@
15191 +/*
15192 + * Copyright (C) 2010-2017 Junjiro R. Okajima
15193 + *
15194 + * This program, aufs is free software; you can redistribute it and/or modify
15195 + * it under the terms of the GNU General Public License as published by
15196 + * the Free Software Foundation; either version 2 of the License, or
15197 + * (at your option) any later version.
15198 + *
15199 + * This program is distributed in the hope that it will be useful,
15200 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15201 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15202 + * GNU General Public License for more details.
15203 + *
15204 + * You should have received a copy of the GNU General Public License
15205 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15206 + */
15207 +
15208 +/*
15209 + * special support for filesystems which aqucires an inode mutex
15210 + * at final closing a file, eg, hfsplus.
15211 + *
15212 + * This trick is very simple and stupid, just to open the file before really
15213 + * neceeary open to tell hfsplus that this is not the final closing.
15214 + * The caller should call au_h_open_pre() after acquiring the inode mutex,
15215 + * and au_h_open_post() after releasing it.
15216 + */
15217 +
15218 +#include "aufs.h"
15219 +
15220 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
15221 +                          int force_wr)
15222 +{
15223 +       struct file *h_file;
15224 +       struct dentry *h_dentry;
15225 +
15226 +       h_dentry = au_h_dptr(dentry, bindex);
15227 +       AuDebugOn(!h_dentry);
15228 +       AuDebugOn(d_is_negative(h_dentry));
15229 +
15230 +       h_file = NULL;
15231 +       if (au_test_hfsplus(h_dentry->d_sb)
15232 +           && d_is_reg(h_dentry))
15233 +               h_file = au_h_open(dentry, bindex,
15234 +                                  O_RDONLY | O_NOATIME | O_LARGEFILE,
15235 +                                  /*file*/NULL, force_wr);
15236 +       return h_file;
15237 +}
15238 +
15239 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
15240 +                   struct file *h_file)
15241 +{
15242 +       if (h_file) {
15243 +               fput(h_file);
15244 +               au_sbr_put(dentry->d_sb, bindex);
15245 +       }
15246 +}
15247 diff -urN /usr/share/empty/fs/aufs/hnotify.c linux/fs/aufs/hnotify.c
15248 --- /usr/share/empty/fs/aufs/hnotify.c  1970-01-01 01:00:00.000000000 +0100
15249 +++ linux/fs/aufs/hnotify.c     2017-05-06 22:16:52.778220093 +0200
15250 @@ -0,0 +1,723 @@
15251 +/*
15252 + * Copyright (C) 2005-2017 Junjiro R. Okajima
15253 + *
15254 + * This program, aufs is free software; you can redistribute it and/or modify
15255 + * it under the terms of the GNU General Public License as published by
15256 + * the Free Software Foundation; either version 2 of the License, or
15257 + * (at your option) any later version.
15258 + *
15259 + * This program is distributed in the hope that it will be useful,
15260 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15261 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15262 + * GNU General Public License for more details.
15263 + *
15264 + * You should have received a copy of the GNU General Public License
15265 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15266 + */
15267 +
15268 +/*
15269 + * abstraction to notify the direct changes on lower directories
15270 + */
15271 +
15272 +#include "aufs.h"
15273 +
15274 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode)
15275 +{
15276 +       int err;
15277 +       struct au_hnotify *hn;
15278 +
15279 +       err = -ENOMEM;
15280 +       hn = au_cache_alloc_hnotify();
15281 +       if (hn) {
15282 +               hn->hn_aufs_inode = inode;
15283 +               hinode->hi_notify = hn;
15284 +               err = au_hnotify_op.alloc(hinode);
15285 +               AuTraceErr(err);
15286 +               if (unlikely(err)) {
15287 +                       hinode->hi_notify = NULL;
15288 +                       au_cache_dfree_hnotify(hn);
15289 +                       /*
15290 +                        * The upper dir was removed by udba, but the same named
15291 +                        * dir left. In this case, aufs assignes a new inode
15292 +                        * number and set the monitor again.
15293 +                        * For the lower dir, the old monitnor is still left.
15294 +                        */
15295 +                       if (err == -EEXIST)
15296 +                               err = 0;
15297 +               }
15298 +       }
15299 +
15300 +       AuTraceErr(err);
15301 +       return err;
15302 +}
15303 +
15304 +void au_hn_free(struct au_hinode *hinode)
15305 +{
15306 +       struct au_hnotify *hn;
15307 +
15308 +       hn = hinode->hi_notify;
15309 +       if (hn) {
15310 +               hinode->hi_notify = NULL;
15311 +               if (au_hnotify_op.free(hinode, hn))
15312 +                       au_cache_dfree_hnotify(hn);
15313 +       }
15314 +}
15315 +
15316 +/* ---------------------------------------------------------------------- */
15317 +
15318 +void au_hn_ctl(struct au_hinode *hinode, int do_set)
15319 +{
15320 +       if (hinode->hi_notify)
15321 +               au_hnotify_op.ctl(hinode, do_set);
15322 +}
15323 +
15324 +void au_hn_reset(struct inode *inode, unsigned int flags)
15325 +{
15326 +       aufs_bindex_t bindex, bbot;
15327 +       struct inode *hi;
15328 +       struct dentry *iwhdentry;
15329 +
15330 +       bbot = au_ibbot(inode);
15331 +       for (bindex = au_ibtop(inode); bindex <= bbot; bindex++) {
15332 +               hi = au_h_iptr(inode, bindex);
15333 +               if (!hi)
15334 +                       continue;
15335 +
15336 +               /* inode_lock_nested(hi, AuLsc_I_CHILD); */
15337 +               iwhdentry = au_hi_wh(inode, bindex);
15338 +               if (iwhdentry)
15339 +                       dget(iwhdentry);
15340 +               au_igrab(hi);
15341 +               au_set_h_iptr(inode, bindex, NULL, 0);
15342 +               au_set_h_iptr(inode, bindex, au_igrab(hi),
15343 +                             flags & ~AuHi_XINO);
15344 +               iput(hi);
15345 +               dput(iwhdentry);
15346 +               /* inode_unlock(hi); */
15347 +       }
15348 +}
15349 +
15350 +/* ---------------------------------------------------------------------- */
15351 +
15352 +static int hn_xino(struct inode *inode, struct inode *h_inode)
15353 +{
15354 +       int err;
15355 +       aufs_bindex_t bindex, bbot, bfound, btop;
15356 +       struct inode *h_i;
15357 +
15358 +       err = 0;
15359 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15360 +               pr_warn("branch root dir was changed\n");
15361 +               goto out;
15362 +       }
15363 +
15364 +       bfound = -1;
15365 +       bbot = au_ibbot(inode);
15366 +       btop = au_ibtop(inode);
15367 +#if 0 /* reserved for future use */
15368 +       if (bindex == bbot) {
15369 +               /* keep this ino in rename case */
15370 +               goto out;
15371 +       }
15372 +#endif
15373 +       for (bindex = btop; bindex <= bbot; bindex++)
15374 +               if (au_h_iptr(inode, bindex) == h_inode) {
15375 +                       bfound = bindex;
15376 +                       break;
15377 +               }
15378 +       if (bfound < 0)
15379 +               goto out;
15380 +
15381 +       for (bindex = btop; bindex <= bbot; bindex++) {
15382 +               h_i = au_h_iptr(inode, bindex);
15383 +               if (!h_i)
15384 +                       continue;
15385 +
15386 +               err = au_xino_write(inode->i_sb, bindex, h_i->i_ino, /*ino*/0);
15387 +               /* ignore this error */
15388 +               /* bad action? */
15389 +       }
15390 +
15391 +       /* children inode number will be broken */
15392 +
15393 +out:
15394 +       AuTraceErr(err);
15395 +       return err;
15396 +}
15397 +
15398 +static int hn_gen_tree(struct dentry *dentry)
15399 +{
15400 +       int err, i, j, ndentry;
15401 +       struct au_dcsub_pages dpages;
15402 +       struct au_dpage *dpage;
15403 +       struct dentry **dentries;
15404 +
15405 +       err = au_dpages_init(&dpages, GFP_NOFS);
15406 +       if (unlikely(err))
15407 +               goto out;
15408 +       err = au_dcsub_pages(&dpages, dentry, NULL, NULL);
15409 +       if (unlikely(err))
15410 +               goto out_dpages;
15411 +
15412 +       for (i = 0; i < dpages.ndpage; i++) {
15413 +               dpage = dpages.dpages + i;
15414 +               dentries = dpage->dentries;
15415 +               ndentry = dpage->ndentry;
15416 +               for (j = 0; j < ndentry; j++) {
15417 +                       struct dentry *d;
15418 +
15419 +                       d = dentries[j];
15420 +                       if (IS_ROOT(d))
15421 +                               continue;
15422 +
15423 +                       au_digen_dec(d);
15424 +                       if (d_really_is_positive(d))
15425 +                               /* todo: reset children xino?
15426 +                                  cached children only? */
15427 +                               au_iigen_dec(d_inode(d));
15428 +               }
15429 +       }
15430 +
15431 +out_dpages:
15432 +       au_dpages_free(&dpages);
15433 +
15434 +#if 0
15435 +       /* discard children */
15436 +       dentry_unhash(dentry);
15437 +       dput(dentry);
15438 +#endif
15439 +out:
15440 +       return err;
15441 +}
15442 +
15443 +/*
15444 + * return 0 if processed.
15445 + */
15446 +static int hn_gen_by_inode(char *name, unsigned int nlen, struct inode *inode,
15447 +                          const unsigned int isdir)
15448 +{
15449 +       int err;
15450 +       struct dentry *d;
15451 +       struct qstr *dname;
15452 +
15453 +       err = 1;
15454 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15455 +               pr_warn("branch root dir was changed\n");
15456 +               err = 0;
15457 +               goto out;
15458 +       }
15459 +
15460 +       if (!isdir) {
15461 +               AuDebugOn(!name);
15462 +               au_iigen_dec(inode);
15463 +               spin_lock(&inode->i_lock);
15464 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
15465 +                       spin_lock(&d->d_lock);
15466 +                       dname = &d->d_name;
15467 +                       if (dname->len != nlen
15468 +                           && memcmp(dname->name, name, nlen)) {
15469 +                               spin_unlock(&d->d_lock);
15470 +                               continue;
15471 +                       }
15472 +                       err = 0;
15473 +                       au_digen_dec(d);
15474 +                       spin_unlock(&d->d_lock);
15475 +                       break;
15476 +               }
15477 +               spin_unlock(&inode->i_lock);
15478 +       } else {
15479 +               au_fset_si(au_sbi(inode->i_sb), FAILED_REFRESH_DIR);
15480 +               d = d_find_any_alias(inode);
15481 +               if (!d) {
15482 +                       au_iigen_dec(inode);
15483 +                       goto out;
15484 +               }
15485 +
15486 +               spin_lock(&d->d_lock);
15487 +               dname = &d->d_name;
15488 +               if (dname->len == nlen && !memcmp(dname->name, name, nlen)) {
15489 +                       spin_unlock(&d->d_lock);
15490 +                       err = hn_gen_tree(d);
15491 +                       spin_lock(&d->d_lock);
15492 +               }
15493 +               spin_unlock(&d->d_lock);
15494 +               dput(d);
15495 +       }
15496 +
15497 +out:
15498 +       AuTraceErr(err);
15499 +       return err;
15500 +}
15501 +
15502 +static int hn_gen_by_name(struct dentry *dentry, const unsigned int isdir)
15503 +{
15504 +       int err;
15505 +
15506 +       if (IS_ROOT(dentry)) {
15507 +               pr_warn("branch root dir was changed\n");
15508 +               return 0;
15509 +       }
15510 +
15511 +       err = 0;
15512 +       if (!isdir) {
15513 +               au_digen_dec(dentry);
15514 +               if (d_really_is_positive(dentry))
15515 +                       au_iigen_dec(d_inode(dentry));
15516 +       } else {
15517 +               au_fset_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR);
15518 +               if (d_really_is_positive(dentry))
15519 +                       err = hn_gen_tree(dentry);
15520 +       }
15521 +
15522 +       AuTraceErr(err);
15523 +       return err;
15524 +}
15525 +
15526 +/* ---------------------------------------------------------------------- */
15527 +
15528 +/* hnotify job flags */
15529 +#define AuHnJob_XINO0          1
15530 +#define AuHnJob_GEN            (1 << 1)
15531 +#define AuHnJob_DIRENT         (1 << 2)
15532 +#define AuHnJob_ISDIR          (1 << 3)
15533 +#define AuHnJob_TRYXINO0       (1 << 4)
15534 +#define AuHnJob_MNTPNT         (1 << 5)
15535 +#define au_ftest_hnjob(flags, name)    ((flags) & AuHnJob_##name)
15536 +#define au_fset_hnjob(flags, name) \
15537 +       do { (flags) |= AuHnJob_##name; } while (0)
15538 +#define au_fclr_hnjob(flags, name) \
15539 +       do { (flags) &= ~AuHnJob_##name; } while (0)
15540 +
15541 +enum {
15542 +       AuHn_CHILD,
15543 +       AuHn_PARENT,
15544 +       AuHnLast
15545 +};
15546 +
15547 +struct au_hnotify_args {
15548 +       struct inode *h_dir, *dir, *h_child_inode;
15549 +       u32 mask;
15550 +       unsigned int flags[AuHnLast];
15551 +       unsigned int h_child_nlen;
15552 +       char h_child_name[];
15553 +};
15554 +
15555 +struct hn_job_args {
15556 +       unsigned int flags;
15557 +       struct inode *inode, *h_inode, *dir, *h_dir;
15558 +       struct dentry *dentry;
15559 +       char *h_name;
15560 +       int h_nlen;
15561 +};
15562 +
15563 +static int hn_job(struct hn_job_args *a)
15564 +{
15565 +       const unsigned int isdir = au_ftest_hnjob(a->flags, ISDIR);
15566 +       int e;
15567 +
15568 +       /* reset xino */
15569 +       if (au_ftest_hnjob(a->flags, XINO0) && a->inode)
15570 +               hn_xino(a->inode, a->h_inode); /* ignore this error */
15571 +
15572 +       if (au_ftest_hnjob(a->flags, TRYXINO0)
15573 +           && a->inode
15574 +           && a->h_inode) {
15575 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
15576 +               if (!a->h_inode->i_nlink
15577 +                   && !(a->h_inode->i_state & I_LINKABLE))
15578 +                       hn_xino(a->inode, a->h_inode); /* ignore this error */
15579 +               inode_unlock(a->h_inode);
15580 +       }
15581 +
15582 +       /* make the generation obsolete */
15583 +       if (au_ftest_hnjob(a->flags, GEN)) {
15584 +               e = -1;
15585 +               if (a->inode)
15586 +                       e = hn_gen_by_inode(a->h_name, a->h_nlen, a->inode,
15587 +                                             isdir);
15588 +               if (e && a->dentry)
15589 +                       hn_gen_by_name(a->dentry, isdir);
15590 +               /* ignore this error */
15591 +       }
15592 +
15593 +       /* make dir entries obsolete */
15594 +       if (au_ftest_hnjob(a->flags, DIRENT) && a->inode) {
15595 +               struct au_vdir *vdir;
15596 +
15597 +               vdir = au_ivdir(a->inode);
15598 +               if (vdir)
15599 +                       vdir->vd_jiffy = 0;
15600 +               /* IMustLock(a->inode); */
15601 +               /* a->inode->i_version++; */
15602 +       }
15603 +
15604 +       /* can do nothing but warn */
15605 +       if (au_ftest_hnjob(a->flags, MNTPNT)
15606 +           && a->dentry
15607 +           && d_mountpoint(a->dentry))
15608 +               pr_warn("mount-point %pd is removed or renamed\n", a->dentry);
15609 +
15610 +       return 0;
15611 +}
15612 +
15613 +/* ---------------------------------------------------------------------- */
15614 +
15615 +static struct dentry *lookup_wlock_by_name(char *name, unsigned int nlen,
15616 +                                          struct inode *dir)
15617 +{
15618 +       struct dentry *dentry, *d, *parent;
15619 +       struct qstr *dname;
15620 +
15621 +       parent = d_find_any_alias(dir);
15622 +       if (!parent)
15623 +               return NULL;
15624 +
15625 +       dentry = NULL;
15626 +       spin_lock(&parent->d_lock);
15627 +       list_for_each_entry(d, &parent->d_subdirs, d_child) {
15628 +               /* AuDbg("%pd\n", d); */
15629 +               spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
15630 +               dname = &d->d_name;
15631 +               if (dname->len != nlen || memcmp(dname->name, name, nlen))
15632 +                       goto cont_unlock;
15633 +               if (au_di(d))
15634 +                       au_digen_dec(d);
15635 +               else
15636 +                       goto cont_unlock;
15637 +               if (au_dcount(d) > 0) {
15638 +                       dentry = dget_dlock(d);
15639 +                       spin_unlock(&d->d_lock);
15640 +                       break;
15641 +               }
15642 +
15643 +cont_unlock:
15644 +               spin_unlock(&d->d_lock);
15645 +       }
15646 +       spin_unlock(&parent->d_lock);
15647 +       dput(parent);
15648 +
15649 +       if (dentry)
15650 +               di_write_lock_child(dentry);
15651 +
15652 +       return dentry;
15653 +}
15654 +
15655 +static struct inode *lookup_wlock_by_ino(struct super_block *sb,
15656 +                                        aufs_bindex_t bindex, ino_t h_ino)
15657 +{
15658 +       struct inode *inode;
15659 +       ino_t ino;
15660 +       int err;
15661 +
15662 +       inode = NULL;
15663 +       err = au_xino_read(sb, bindex, h_ino, &ino);
15664 +       if (!err && ino)
15665 +               inode = ilookup(sb, ino);
15666 +       if (!inode)
15667 +               goto out;
15668 +
15669 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15670 +               pr_warn("wrong root branch\n");
15671 +               iput(inode);
15672 +               inode = NULL;
15673 +               goto out;
15674 +       }
15675 +
15676 +       ii_write_lock_child(inode);
15677 +
15678 +out:
15679 +       return inode;
15680 +}
15681 +
15682 +static void au_hn_bh(void *_args)
15683 +{
15684 +       struct au_hnotify_args *a = _args;
15685 +       struct super_block *sb;
15686 +       aufs_bindex_t bindex, bbot, bfound;
15687 +       unsigned char xino, try_iput;
15688 +       int err;
15689 +       struct inode *inode;
15690 +       ino_t h_ino;
15691 +       struct hn_job_args args;
15692 +       struct dentry *dentry;
15693 +       struct au_sbinfo *sbinfo;
15694 +
15695 +       AuDebugOn(!_args);
15696 +       AuDebugOn(!a->h_dir);
15697 +       AuDebugOn(!a->dir);
15698 +       AuDebugOn(!a->mask);
15699 +       AuDbg("mask 0x%x, i%lu, hi%lu, hci%lu\n",
15700 +             a->mask, a->dir->i_ino, a->h_dir->i_ino,
15701 +             a->h_child_inode ? a->h_child_inode->i_ino : 0);
15702 +
15703 +       inode = NULL;
15704 +       dentry = NULL;
15705 +       /*
15706 +        * do not lock a->dir->i_mutex here
15707 +        * because of d_revalidate() may cause a deadlock.
15708 +        */
15709 +       sb = a->dir->i_sb;
15710 +       AuDebugOn(!sb);
15711 +       sbinfo = au_sbi(sb);
15712 +       AuDebugOn(!sbinfo);
15713 +       si_write_lock(sb, AuLock_NOPLMW);
15714 +
15715 +       ii_read_lock_parent(a->dir);
15716 +       bfound = -1;
15717 +       bbot = au_ibbot(a->dir);
15718 +       for (bindex = au_ibtop(a->dir); bindex <= bbot; bindex++)
15719 +               if (au_h_iptr(a->dir, bindex) == a->h_dir) {
15720 +                       bfound = bindex;
15721 +                       break;
15722 +               }
15723 +       ii_read_unlock(a->dir);
15724 +       if (unlikely(bfound < 0))
15725 +               goto out;
15726 +
15727 +       xino = !!au_opt_test(au_mntflags(sb), XINO);
15728 +       h_ino = 0;
15729 +       if (a->h_child_inode)
15730 +               h_ino = a->h_child_inode->i_ino;
15731 +
15732 +       if (a->h_child_nlen
15733 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], GEN)
15734 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], MNTPNT)))
15735 +               dentry = lookup_wlock_by_name(a->h_child_name, a->h_child_nlen,
15736 +                                             a->dir);
15737 +       try_iput = 0;
15738 +       if (dentry && d_really_is_positive(dentry))
15739 +               inode = d_inode(dentry);
15740 +       if (xino && !inode && h_ino
15741 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], XINO0)
15742 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], TRYXINO0)
15743 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], GEN))) {
15744 +               inode = lookup_wlock_by_ino(sb, bfound, h_ino);
15745 +               try_iput = 1;
15746 +       }
15747 +
15748 +       args.flags = a->flags[AuHn_CHILD];
15749 +       args.dentry = dentry;
15750 +       args.inode = inode;
15751 +       args.h_inode = a->h_child_inode;
15752 +       args.dir = a->dir;
15753 +       args.h_dir = a->h_dir;
15754 +       args.h_name = a->h_child_name;
15755 +       args.h_nlen = a->h_child_nlen;
15756 +       err = hn_job(&args);
15757 +       if (dentry) {
15758 +               if (au_di(dentry))
15759 +                       di_write_unlock(dentry);
15760 +               dput(dentry);
15761 +       }
15762 +       if (inode && try_iput) {
15763 +               ii_write_unlock(inode);
15764 +               iput(inode);
15765 +       }
15766 +
15767 +       ii_write_lock_parent(a->dir);
15768 +       args.flags = a->flags[AuHn_PARENT];
15769 +       args.dentry = NULL;
15770 +       args.inode = a->dir;
15771 +       args.h_inode = a->h_dir;
15772 +       args.dir = NULL;
15773 +       args.h_dir = NULL;
15774 +       args.h_name = NULL;
15775 +       args.h_nlen = 0;
15776 +       err = hn_job(&args);
15777 +       ii_write_unlock(a->dir);
15778 +
15779 +out:
15780 +       iput(a->h_child_inode);
15781 +       iput(a->h_dir);
15782 +       iput(a->dir);
15783 +       si_write_unlock(sb);
15784 +       au_nwt_done(&sbinfo->si_nowait);
15785 +       au_delayed_kfree(a);
15786 +}
15787 +
15788 +/* ---------------------------------------------------------------------- */
15789 +
15790 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
15791 +              struct qstr *h_child_qstr, struct inode *h_child_inode)
15792 +{
15793 +       int err, len;
15794 +       unsigned int flags[AuHnLast], f;
15795 +       unsigned char isdir, isroot, wh;
15796 +       struct inode *dir;
15797 +       struct au_hnotify_args *args;
15798 +       char *p, *h_child_name;
15799 +
15800 +       err = 0;
15801 +       AuDebugOn(!hnotify || !hnotify->hn_aufs_inode);
15802 +       dir = igrab(hnotify->hn_aufs_inode);
15803 +       if (!dir)
15804 +               goto out;
15805 +
15806 +       isroot = (dir->i_ino == AUFS_ROOT_INO);
15807 +       wh = 0;
15808 +       h_child_name = (void *)h_child_qstr->name;
15809 +       len = h_child_qstr->len;
15810 +       if (h_child_name) {
15811 +               if (len > AUFS_WH_PFX_LEN
15812 +                   && !memcmp(h_child_name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
15813 +                       h_child_name += AUFS_WH_PFX_LEN;
15814 +                       len -= AUFS_WH_PFX_LEN;
15815 +                       wh = 1;
15816 +               }
15817 +       }
15818 +
15819 +       isdir = 0;
15820 +       if (h_child_inode)
15821 +               isdir = !!S_ISDIR(h_child_inode->i_mode);
15822 +       flags[AuHn_PARENT] = AuHnJob_ISDIR;
15823 +       flags[AuHn_CHILD] = 0;
15824 +       if (isdir)
15825 +               flags[AuHn_CHILD] = AuHnJob_ISDIR;
15826 +       au_fset_hnjob(flags[AuHn_PARENT], DIRENT);
15827 +       au_fset_hnjob(flags[AuHn_CHILD], GEN);
15828 +       switch (mask & FS_EVENTS_POSS_ON_CHILD) {
15829 +       case FS_MOVED_FROM:
15830 +       case FS_MOVED_TO:
15831 +               au_fset_hnjob(flags[AuHn_CHILD], XINO0);
15832 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15833 +               /*FALLTHROUGH*/
15834 +       case FS_CREATE:
15835 +               AuDebugOn(!h_child_name);
15836 +               break;
15837 +
15838 +       case FS_DELETE:
15839 +               /*
15840 +                * aufs never be able to get this child inode.
15841 +                * revalidation should be in d_revalidate()
15842 +                * by checking i_nlink, i_generation or d_unhashed().
15843 +                */
15844 +               AuDebugOn(!h_child_name);
15845 +               au_fset_hnjob(flags[AuHn_CHILD], TRYXINO0);
15846 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15847 +               break;
15848 +
15849 +       default:
15850 +               AuDebugOn(1);
15851 +       }
15852 +
15853 +       if (wh)
15854 +               h_child_inode = NULL;
15855 +
15856 +       err = -ENOMEM;
15857 +       /* iput() and kfree() will be called in au_hnotify() */
15858 +       args = kmalloc(sizeof(*args) + len + 1, GFP_NOFS);
15859 +       if (unlikely(!args)) {
15860 +               AuErr1("no memory\n");
15861 +               iput(dir);
15862 +               goto out;
15863 +       }
15864 +       args->flags[AuHn_PARENT] = flags[AuHn_PARENT];
15865 +       args->flags[AuHn_CHILD] = flags[AuHn_CHILD];
15866 +       args->mask = mask;
15867 +       args->dir = dir;
15868 +       args->h_dir = igrab(h_dir);
15869 +       if (h_child_inode)
15870 +               h_child_inode = igrab(h_child_inode); /* can be NULL */
15871 +       args->h_child_inode = h_child_inode;
15872 +       args->h_child_nlen = len;
15873 +       if (len) {
15874 +               p = (void *)args;
15875 +               p += sizeof(*args);
15876 +               memcpy(p, h_child_name, len);
15877 +               p[len] = 0;
15878 +       }
15879 +
15880 +       /* NFS fires the event for silly-renamed one from kworker */
15881 +       f = 0;
15882 +       if (!dir->i_nlink
15883 +           || (au_test_nfs(h_dir->i_sb) && (mask & FS_DELETE)))
15884 +               f = AuWkq_NEST;
15885 +       err = au_wkq_nowait(au_hn_bh, args, dir->i_sb, f);
15886 +       if (unlikely(err)) {
15887 +               pr_err("wkq %d\n", err);
15888 +               iput(args->h_child_inode);
15889 +               iput(args->h_dir);
15890 +               iput(args->dir);
15891 +               au_delayed_kfree(args);
15892 +       }
15893 +
15894 +out:
15895 +       return err;
15896 +}
15897 +
15898 +/* ---------------------------------------------------------------------- */
15899 +
15900 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm)
15901 +{
15902 +       int err;
15903 +
15904 +       AuDebugOn(!(udba & AuOptMask_UDBA));
15905 +
15906 +       err = 0;
15907 +       if (au_hnotify_op.reset_br)
15908 +               err = au_hnotify_op.reset_br(udba, br, perm);
15909 +
15910 +       return err;
15911 +}
15912 +
15913 +int au_hnotify_init_br(struct au_branch *br, int perm)
15914 +{
15915 +       int err;
15916 +
15917 +       err = 0;
15918 +       if (au_hnotify_op.init_br)
15919 +               err = au_hnotify_op.init_br(br, perm);
15920 +
15921 +       return err;
15922 +}
15923 +
15924 +void au_hnotify_fin_br(struct au_branch *br)
15925 +{
15926 +       if (au_hnotify_op.fin_br)
15927 +               au_hnotify_op.fin_br(br);
15928 +}
15929 +
15930 +static void au_hn_destroy_cache(void)
15931 +{
15932 +       struct au_cache *cp;
15933 +
15934 +       flush_delayed_work(&au_dfree.dwork);
15935 +       cp = au_dfree.cache + AuCache_HNOTIFY;
15936 +       AuDebugOn(!llist_empty(&cp->llist));
15937 +       kmem_cache_destroy(cp->cache);
15938 +       cp->cache = NULL;
15939 +}
15940 +
15941 +AU_CACHE_DFREE_FUNC(hnotify, HNOTIFY, hn_lnode);
15942 +
15943 +int __init au_hnotify_init(void)
15944 +{
15945 +       int err;
15946 +       struct au_cache *cp;
15947 +
15948 +       err = -ENOMEM;
15949 +       cp = au_dfree.cache + AuCache_HNOTIFY;
15950 +       cp->cache = AuCache(au_hnotify);
15951 +       if (cp->cache) {
15952 +               err = 0;
15953 +               if (au_hnotify_op.init)
15954 +                       err = au_hnotify_op.init();
15955 +               if (unlikely(err))
15956 +                       au_hn_destroy_cache();
15957 +       }
15958 +       AuTraceErr(err);
15959 +       return err;
15960 +}
15961 +
15962 +void au_hnotify_fin(void)
15963 +{
15964 +       struct au_cache *cp;
15965 +
15966 +       if (au_hnotify_op.fin)
15967 +               au_hnotify_op.fin();
15968 +
15969 +       /* cf. au_cache_fin() */
15970 +       cp = au_dfree.cache + AuCache_HNOTIFY;
15971 +       if (cp->cache)
15972 +               au_hn_destroy_cache();
15973 +}
15974 diff -urN /usr/share/empty/fs/aufs/iinfo.c linux/fs/aufs/iinfo.c
15975 --- /usr/share/empty/fs/aufs/iinfo.c    1970-01-01 01:00:00.000000000 +0100
15976 +++ linux/fs/aufs/iinfo.c       2017-05-06 22:16:52.778220093 +0200
15977 @@ -0,0 +1,285 @@
15978 +/*
15979 + * Copyright (C) 2005-2017 Junjiro R. Okajima
15980 + *
15981 + * This program, aufs is free software; you can redistribute it and/or modify
15982 + * it under the terms of the GNU General Public License as published by
15983 + * the Free Software Foundation; either version 2 of the License, or
15984 + * (at your option) any later version.
15985 + *
15986 + * This program is distributed in the hope that it will be useful,
15987 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15988 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15989 + * GNU General Public License for more details.
15990 + *
15991 + * You should have received a copy of the GNU General Public License
15992 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15993 + */
15994 +
15995 +/*
15996 + * inode private data
15997 + */
15998 +
15999 +#include "aufs.h"
16000 +
16001 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex)
16002 +{
16003 +       struct inode *h_inode;
16004 +       struct au_hinode *hinode;
16005 +
16006 +       IiMustAnyLock(inode);
16007 +
16008 +       hinode = au_hinode(au_ii(inode), bindex);
16009 +       h_inode = hinode->hi_inode;
16010 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
16011 +       return h_inode;
16012 +}
16013 +
16014 +/* todo: hard/soft set? */
16015 +void au_hiput(struct au_hinode *hinode)
16016 +{
16017 +       au_hn_free(hinode);
16018 +       dput(hinode->hi_whdentry);
16019 +       iput(hinode->hi_inode);
16020 +}
16021 +
16022 +unsigned int au_hi_flags(struct inode *inode, int isdir)
16023 +{
16024 +       unsigned int flags;
16025 +       const unsigned int mnt_flags = au_mntflags(inode->i_sb);
16026 +
16027 +       flags = 0;
16028 +       if (au_opt_test(mnt_flags, XINO))
16029 +               au_fset_hi(flags, XINO);
16030 +       if (isdir && au_opt_test(mnt_flags, UDBA_HNOTIFY))
16031 +               au_fset_hi(flags, HNOTIFY);
16032 +       return flags;
16033 +}
16034 +
16035 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
16036 +                  struct inode *h_inode, unsigned int flags)
16037 +{
16038 +       struct au_hinode *hinode;
16039 +       struct inode *hi;
16040 +       struct au_iinfo *iinfo = au_ii(inode);
16041 +
16042 +       IiMustWriteLock(inode);
16043 +
16044 +       hinode = au_hinode(iinfo, bindex);
16045 +       hi = hinode->hi_inode;
16046 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
16047 +
16048 +       if (hi)
16049 +               au_hiput(hinode);
16050 +       hinode->hi_inode = h_inode;
16051 +       if (h_inode) {
16052 +               int err;
16053 +               struct super_block *sb = inode->i_sb;
16054 +               struct au_branch *br;
16055 +
16056 +               AuDebugOn(inode->i_mode
16057 +                         && (h_inode->i_mode & S_IFMT)
16058 +                         != (inode->i_mode & S_IFMT));
16059 +               if (bindex == iinfo->ii_btop)
16060 +                       au_cpup_igen(inode, h_inode);
16061 +               br = au_sbr(sb, bindex);
16062 +               hinode->hi_id = br->br_id;
16063 +               if (au_ftest_hi(flags, XINO)) {
16064 +                       err = au_xino_write(sb, bindex, h_inode->i_ino,
16065 +                                           inode->i_ino);
16066 +                       if (unlikely(err))
16067 +                               AuIOErr1("failed au_xino_write() %d\n", err);
16068 +               }
16069 +
16070 +               if (au_ftest_hi(flags, HNOTIFY)
16071 +                   && au_br_hnotifyable(br->br_perm)) {
16072 +                       err = au_hn_alloc(hinode, inode);
16073 +                       if (unlikely(err))
16074 +                               AuIOErr1("au_hn_alloc() %d\n", err);
16075 +               }
16076 +       }
16077 +}
16078 +
16079 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
16080 +                 struct dentry *h_wh)
16081 +{
16082 +       struct au_hinode *hinode;
16083 +
16084 +       IiMustWriteLock(inode);
16085 +
16086 +       hinode = au_hinode(au_ii(inode), bindex);
16087 +       AuDebugOn(hinode->hi_whdentry);
16088 +       hinode->hi_whdentry = h_wh;
16089 +}
16090 +
16091 +void au_update_iigen(struct inode *inode, int half)
16092 +{
16093 +       struct au_iinfo *iinfo;
16094 +       struct au_iigen *iigen;
16095 +       unsigned int sigen;
16096 +
16097 +       sigen = au_sigen(inode->i_sb);
16098 +       iinfo = au_ii(inode);
16099 +       iigen = &iinfo->ii_generation;
16100 +       spin_lock(&iigen->ig_spin);
16101 +       iigen->ig_generation = sigen;
16102 +       if (half)
16103 +               au_ig_fset(iigen->ig_flags, HALF_REFRESHED);
16104 +       else
16105 +               au_ig_fclr(iigen->ig_flags, HALF_REFRESHED);
16106 +       spin_unlock(&iigen->ig_spin);
16107 +}
16108 +
16109 +/* it may be called at remount time, too */
16110 +void au_update_ibrange(struct inode *inode, int do_put_zero)
16111 +{
16112 +       struct au_iinfo *iinfo;
16113 +       aufs_bindex_t bindex, bbot;
16114 +
16115 +       AuDebugOn(au_is_bad_inode(inode));
16116 +       IiMustWriteLock(inode);
16117 +
16118 +       iinfo = au_ii(inode);
16119 +       if (do_put_zero && iinfo->ii_btop >= 0) {
16120 +               for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot;
16121 +                    bindex++) {
16122 +                       struct inode *h_i;
16123 +
16124 +                       h_i = au_hinode(iinfo, bindex)->hi_inode;
16125 +                       if (h_i
16126 +                           && !h_i->i_nlink
16127 +                           && !(h_i->i_state & I_LINKABLE))
16128 +                               au_set_h_iptr(inode, bindex, NULL, 0);
16129 +               }
16130 +       }
16131 +
16132 +       iinfo->ii_btop = -1;
16133 +       iinfo->ii_bbot = -1;
16134 +       bbot = au_sbbot(inode->i_sb);
16135 +       for (bindex = 0; bindex <= bbot; bindex++)
16136 +               if (au_hinode(iinfo, bindex)->hi_inode) {
16137 +                       iinfo->ii_btop = bindex;
16138 +                       break;
16139 +               }
16140 +       if (iinfo->ii_btop >= 0)
16141 +               for (bindex = bbot; bindex >= iinfo->ii_btop; bindex--)
16142 +                       if (au_hinode(iinfo, bindex)->hi_inode) {
16143 +                               iinfo->ii_bbot = bindex;
16144 +                               break;
16145 +                       }
16146 +       AuDebugOn(iinfo->ii_btop > iinfo->ii_bbot);
16147 +}
16148 +
16149 +/* ---------------------------------------------------------------------- */
16150 +
16151 +void au_icntnr_init_once(void *_c)
16152 +{
16153 +       struct au_icntnr *c = _c;
16154 +       struct au_iinfo *iinfo = &c->iinfo;
16155 +
16156 +       spin_lock_init(&iinfo->ii_generation.ig_spin);
16157 +       au_rw_init(&iinfo->ii_rwsem);
16158 +       inode_init_once(&c->vfs_inode);
16159 +}
16160 +
16161 +void au_hinode_init(struct au_hinode *hinode)
16162 +{
16163 +       hinode->hi_inode = NULL;
16164 +       hinode->hi_id = -1;
16165 +       au_hn_init(hinode);
16166 +       hinode->hi_whdentry = NULL;
16167 +}
16168 +
16169 +int au_iinfo_init(struct inode *inode)
16170 +{
16171 +       struct au_iinfo *iinfo;
16172 +       struct super_block *sb;
16173 +       struct au_hinode *hi;
16174 +       int nbr, i;
16175 +
16176 +       sb = inode->i_sb;
16177 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
16178 +       nbr = au_sbbot(sb) + 1;
16179 +       if (unlikely(nbr <= 0))
16180 +               nbr = 1;
16181 +       hi = kmalloc_array(nbr, sizeof(*iinfo->ii_hinode), GFP_NOFS);
16182 +       if (hi) {
16183 +               au_ninodes_inc(sb);
16184 +
16185 +               iinfo->ii_hinode = hi;
16186 +               for (i = 0; i < nbr; i++, hi++)
16187 +                       au_hinode_init(hi);
16188 +
16189 +               iinfo->ii_generation.ig_generation = au_sigen(sb);
16190 +               iinfo->ii_btop = -1;
16191 +               iinfo->ii_bbot = -1;
16192 +               iinfo->ii_vdir = NULL;
16193 +               return 0;
16194 +       }
16195 +       return -ENOMEM;
16196 +}
16197 +
16198 +int au_hinode_realloc(struct au_iinfo *iinfo, int nbr, int may_shrink)
16199 +{
16200 +       int err, i;
16201 +       struct au_hinode *hip;
16202 +
16203 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
16204 +
16205 +       err = -ENOMEM;
16206 +       hip = au_krealloc(iinfo->ii_hinode, sizeof(*hip) * nbr, GFP_NOFS,
16207 +                         may_shrink);
16208 +       if (hip) {
16209 +               iinfo->ii_hinode = hip;
16210 +               i = iinfo->ii_bbot + 1;
16211 +               hip += i;
16212 +               for (; i < nbr; i++, hip++)
16213 +                       au_hinode_init(hip);
16214 +               err = 0;
16215 +       }
16216 +
16217 +       return err;
16218 +}
16219 +
16220 +void au_iinfo_fin(struct inode *inode)
16221 +{
16222 +       struct au_iinfo *iinfo;
16223 +       struct au_hinode *hi;
16224 +       struct super_block *sb;
16225 +       aufs_bindex_t bindex, bbot;
16226 +       const unsigned char unlinked = !inode->i_nlink;
16227 +
16228 +       AuDebugOn(au_is_bad_inode(inode));
16229 +
16230 +       sb = inode->i_sb;
16231 +       au_ninodes_dec(sb);
16232 +       if (si_pid_test(sb))
16233 +               au_xino_delete_inode(inode, unlinked);
16234 +       else {
16235 +               /*
16236 +                * it is safe to hide the dependency between sbinfo and
16237 +                * sb->s_umount.
16238 +                */
16239 +               lockdep_off();
16240 +               si_noflush_read_lock(sb);
16241 +               au_xino_delete_inode(inode, unlinked);
16242 +               si_read_unlock(sb);
16243 +               lockdep_on();
16244 +       }
16245 +
16246 +       iinfo = au_ii(inode);
16247 +       if (iinfo->ii_vdir)
16248 +               au_vdir_free(iinfo->ii_vdir, /*atonce*/0);
16249 +
16250 +       bindex = iinfo->ii_btop;
16251 +       if (bindex >= 0) {
16252 +               hi = au_hinode(iinfo, bindex);
16253 +               bbot = iinfo->ii_bbot;
16254 +               while (bindex++ <= bbot) {
16255 +                       if (hi->hi_inode)
16256 +                               au_hiput(hi);
16257 +                       hi++;
16258 +               }
16259 +       }
16260 +       au_delayed_kfree(iinfo->ii_hinode);
16261 +       AuRwDestroy(&iinfo->ii_rwsem);
16262 +}
16263 diff -urN /usr/share/empty/fs/aufs/inode.c linux/fs/aufs/inode.c
16264 --- /usr/share/empty/fs/aufs/inode.c    1970-01-01 01:00:00.000000000 +0100
16265 +++ linux/fs/aufs/inode.c       2017-05-06 22:16:52.781553511 +0200
16266 @@ -0,0 +1,527 @@
16267 +/*
16268 + * Copyright (C) 2005-2017 Junjiro R. Okajima
16269 + *
16270 + * This program, aufs is free software; you can redistribute it and/or modify
16271 + * it under the terms of the GNU General Public License as published by
16272 + * the Free Software Foundation; either version 2 of the License, or
16273 + * (at your option) any later version.
16274 + *
16275 + * This program is distributed in the hope that it will be useful,
16276 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16277 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16278 + * GNU General Public License for more details.
16279 + *
16280 + * You should have received a copy of the GNU General Public License
16281 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16282 + */
16283 +
16284 +/*
16285 + * inode functions
16286 + */
16287 +
16288 +#include "aufs.h"
16289 +
16290 +struct inode *au_igrab(struct inode *inode)
16291 +{
16292 +       if (inode) {
16293 +               AuDebugOn(!atomic_read(&inode->i_count));
16294 +               ihold(inode);
16295 +       }
16296 +       return inode;
16297 +}
16298 +
16299 +static void au_refresh_hinode_attr(struct inode *inode, int do_version)
16300 +{
16301 +       au_cpup_attr_all(inode, /*force*/0);
16302 +       au_update_iigen(inode, /*half*/1);
16303 +       if (do_version)
16304 +               inode->i_version++;
16305 +}
16306 +
16307 +static int au_ii_refresh(struct inode *inode, int *update)
16308 +{
16309 +       int err, e, nbr;
16310 +       umode_t type;
16311 +       aufs_bindex_t bindex, new_bindex;
16312 +       struct super_block *sb;
16313 +       struct au_iinfo *iinfo;
16314 +       struct au_hinode *p, *q, tmp;
16315 +
16316 +       AuDebugOn(au_is_bad_inode(inode));
16317 +       IiMustWriteLock(inode);
16318 +
16319 +       *update = 0;
16320 +       sb = inode->i_sb;
16321 +       nbr = au_sbbot(sb) + 1;
16322 +       type = inode->i_mode & S_IFMT;
16323 +       iinfo = au_ii(inode);
16324 +       err = au_hinode_realloc(iinfo, nbr, /*may_shrink*/0);
16325 +       if (unlikely(err))
16326 +               goto out;
16327 +
16328 +       AuDebugOn(iinfo->ii_btop < 0);
16329 +       p = au_hinode(iinfo, iinfo->ii_btop);
16330 +       for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot;
16331 +            bindex++, p++) {
16332 +               if (!p->hi_inode)
16333 +                       continue;
16334 +
16335 +               AuDebugOn(type != (p->hi_inode->i_mode & S_IFMT));
16336 +               new_bindex = au_br_index(sb, p->hi_id);
16337 +               if (new_bindex == bindex)
16338 +                       continue;
16339 +
16340 +               if (new_bindex < 0) {
16341 +                       *update = 1;
16342 +                       au_hiput(p);
16343 +                       p->hi_inode = NULL;
16344 +                       continue;
16345 +               }
16346 +
16347 +               if (new_bindex < iinfo->ii_btop)
16348 +                       iinfo->ii_btop = new_bindex;
16349 +               if (iinfo->ii_bbot < new_bindex)
16350 +                       iinfo->ii_bbot = new_bindex;
16351 +               /* swap two lower inode, and loop again */
16352 +               q = au_hinode(iinfo, new_bindex);
16353 +               tmp = *q;
16354 +               *q = *p;
16355 +               *p = tmp;
16356 +               if (tmp.hi_inode) {
16357 +                       bindex--;
16358 +                       p--;
16359 +               }
16360 +       }
16361 +       au_update_ibrange(inode, /*do_put_zero*/0);
16362 +       au_hinode_realloc(iinfo, nbr, /*may_shrink*/1); /* harmless if err */
16363 +       e = au_dy_irefresh(inode);
16364 +       if (unlikely(e && !err))
16365 +               err = e;
16366 +
16367 +out:
16368 +       AuTraceErr(err);
16369 +       return err;
16370 +}
16371 +
16372 +void au_refresh_iop(struct inode *inode, int force_getattr)
16373 +{
16374 +       int type;
16375 +       struct au_sbinfo *sbi = au_sbi(inode->i_sb);
16376 +       const struct inode_operations *iop
16377 +               = force_getattr ? aufs_iop : sbi->si_iop_array;
16378 +
16379 +       if (inode->i_op == iop)
16380 +               return;
16381 +
16382 +       switch (inode->i_mode & S_IFMT) {
16383 +       case S_IFDIR:
16384 +               type = AuIop_DIR;
16385 +               break;
16386 +       case S_IFLNK:
16387 +               type = AuIop_SYMLINK;
16388 +               break;
16389 +       default:
16390 +               type = AuIop_OTHER;
16391 +               break;
16392 +       }
16393 +
16394 +       inode->i_op = iop + type;
16395 +       /* unnecessary smp_wmb() */
16396 +}
16397 +
16398 +int au_refresh_hinode_self(struct inode *inode)
16399 +{
16400 +       int err, update;
16401 +
16402 +       err = au_ii_refresh(inode, &update);
16403 +       if (!err)
16404 +               au_refresh_hinode_attr(inode, update && S_ISDIR(inode->i_mode));
16405 +
16406 +       AuTraceErr(err);
16407 +       return err;
16408 +}
16409 +
16410 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry)
16411 +{
16412 +       int err, e, update;
16413 +       unsigned int flags;
16414 +       umode_t mode;
16415 +       aufs_bindex_t bindex, bbot;
16416 +       unsigned char isdir;
16417 +       struct au_hinode *p;
16418 +       struct au_iinfo *iinfo;
16419 +
16420 +       err = au_ii_refresh(inode, &update);
16421 +       if (unlikely(err))
16422 +               goto out;
16423 +
16424 +       update = 0;
16425 +       iinfo = au_ii(inode);
16426 +       p = au_hinode(iinfo, iinfo->ii_btop);
16427 +       mode = (inode->i_mode & S_IFMT);
16428 +       isdir = S_ISDIR(mode);
16429 +       flags = au_hi_flags(inode, isdir);
16430 +       bbot = au_dbbot(dentry);
16431 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++) {
16432 +               struct inode *h_i, *h_inode;
16433 +               struct dentry *h_d;
16434 +
16435 +               h_d = au_h_dptr(dentry, bindex);
16436 +               if (!h_d || d_is_negative(h_d))
16437 +                       continue;
16438 +
16439 +               h_inode = d_inode(h_d);
16440 +               AuDebugOn(mode != (h_inode->i_mode & S_IFMT));
16441 +               if (iinfo->ii_btop <= bindex && bindex <= iinfo->ii_bbot) {
16442 +                       h_i = au_h_iptr(inode, bindex);
16443 +                       if (h_i) {
16444 +                               if (h_i == h_inode)
16445 +                                       continue;
16446 +                               err = -EIO;
16447 +                               break;
16448 +                       }
16449 +               }
16450 +               if (bindex < iinfo->ii_btop)
16451 +                       iinfo->ii_btop = bindex;
16452 +               if (iinfo->ii_bbot < bindex)
16453 +                       iinfo->ii_bbot = bindex;
16454 +               au_set_h_iptr(inode, bindex, au_igrab(h_inode), flags);
16455 +               update = 1;
16456 +       }
16457 +       au_update_ibrange(inode, /*do_put_zero*/0);
16458 +       e = au_dy_irefresh(inode);
16459 +       if (unlikely(e && !err))
16460 +               err = e;
16461 +       if (!err)
16462 +               au_refresh_hinode_attr(inode, update && isdir);
16463 +
16464 +out:
16465 +       AuTraceErr(err);
16466 +       return err;
16467 +}
16468 +
16469 +static int set_inode(struct inode *inode, struct dentry *dentry)
16470 +{
16471 +       int err;
16472 +       unsigned int flags;
16473 +       umode_t mode;
16474 +       aufs_bindex_t bindex, btop, btail;
16475 +       unsigned char isdir;
16476 +       struct dentry *h_dentry;
16477 +       struct inode *h_inode;
16478 +       struct au_iinfo *iinfo;
16479 +       struct inode_operations *iop;
16480 +
16481 +       IiMustWriteLock(inode);
16482 +
16483 +       err = 0;
16484 +       isdir = 0;
16485 +       iop = au_sbi(inode->i_sb)->si_iop_array;
16486 +       btop = au_dbtop(dentry);
16487 +       h_dentry = au_h_dptr(dentry, btop);
16488 +       h_inode = d_inode(h_dentry);
16489 +       mode = h_inode->i_mode;
16490 +       switch (mode & S_IFMT) {
16491 +       case S_IFREG:
16492 +               btail = au_dbtail(dentry);
16493 +               inode->i_op = iop + AuIop_OTHER;
16494 +               inode->i_fop = &aufs_file_fop;
16495 +               err = au_dy_iaop(inode, btop, h_inode);
16496 +               if (unlikely(err))
16497 +                       goto out;
16498 +               break;
16499 +       case S_IFDIR:
16500 +               isdir = 1;
16501 +               btail = au_dbtaildir(dentry);
16502 +               inode->i_op = iop + AuIop_DIR;
16503 +               inode->i_fop = &aufs_dir_fop;
16504 +               break;
16505 +       case S_IFLNK:
16506 +               btail = au_dbtail(dentry);
16507 +               inode->i_op = iop + AuIop_SYMLINK;
16508 +               break;
16509 +       case S_IFBLK:
16510 +       case S_IFCHR:
16511 +       case S_IFIFO:
16512 +       case S_IFSOCK:
16513 +               btail = au_dbtail(dentry);
16514 +               inode->i_op = iop + AuIop_OTHER;
16515 +               init_special_inode(inode, mode, h_inode->i_rdev);
16516 +               break;
16517 +       default:
16518 +               AuIOErr("Unknown file type 0%o\n", mode);
16519 +               err = -EIO;
16520 +               goto out;
16521 +       }
16522 +
16523 +       /* do not set hnotify for whiteouted dirs (SHWH mode) */
16524 +       flags = au_hi_flags(inode, isdir);
16525 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH)
16526 +           && au_ftest_hi(flags, HNOTIFY)
16527 +           && dentry->d_name.len > AUFS_WH_PFX_LEN
16528 +           && !memcmp(dentry->d_name.name, AUFS_WH_PFX, AUFS_WH_PFX_LEN))
16529 +               au_fclr_hi(flags, HNOTIFY);
16530 +       iinfo = au_ii(inode);
16531 +       iinfo->ii_btop = btop;
16532 +       iinfo->ii_bbot = btail;
16533 +       for (bindex = btop; bindex <= btail; bindex++) {
16534 +               h_dentry = au_h_dptr(dentry, bindex);
16535 +               if (h_dentry)
16536 +                       au_set_h_iptr(inode, bindex,
16537 +                                     au_igrab(d_inode(h_dentry)), flags);
16538 +       }
16539 +       au_cpup_attr_all(inode, /*force*/1);
16540 +       /*
16541 +        * to force calling aufs_get_acl() every time,
16542 +        * do not call cache_no_acl() for aufs inode.
16543 +        */
16544 +
16545 +out:
16546 +       return err;
16547 +}
16548 +
16549 +/*
16550 + * successful returns with iinfo write_locked
16551 + * minus: errno
16552 + * zero: success, matched
16553 + * plus: no error, but unmatched
16554 + */
16555 +static int reval_inode(struct inode *inode, struct dentry *dentry)
16556 +{
16557 +       int err;
16558 +       unsigned int gen, igflags;
16559 +       aufs_bindex_t bindex, bbot;
16560 +       struct inode *h_inode, *h_dinode;
16561 +       struct dentry *h_dentry;
16562 +
16563 +       /*
16564 +        * before this function, if aufs got any iinfo lock, it must be only
16565 +        * one, the parent dir.
16566 +        * it can happen by UDBA and the obsoleted inode number.
16567 +        */
16568 +       err = -EIO;
16569 +       if (unlikely(inode->i_ino == parent_ino(dentry)))
16570 +               goto out;
16571 +
16572 +       err = 1;
16573 +       ii_write_lock_new_child(inode);
16574 +       h_dentry = au_h_dptr(dentry, au_dbtop(dentry));
16575 +       h_dinode = d_inode(h_dentry);
16576 +       bbot = au_ibbot(inode);
16577 +       for (bindex = au_ibtop(inode); bindex <= bbot; bindex++) {
16578 +               h_inode = au_h_iptr(inode, bindex);
16579 +               if (!h_inode || h_inode != h_dinode)
16580 +                       continue;
16581 +
16582 +               err = 0;
16583 +               gen = au_iigen(inode, &igflags);
16584 +               if (gen == au_digen(dentry)
16585 +                   && !au_ig_ftest(igflags, HALF_REFRESHED))
16586 +                       break;
16587 +
16588 +               /* fully refresh inode using dentry */
16589 +               err = au_refresh_hinode(inode, dentry);
16590 +               if (!err)
16591 +                       au_update_iigen(inode, /*half*/0);
16592 +               break;
16593 +       }
16594 +
16595 +       if (unlikely(err))
16596 +               ii_write_unlock(inode);
16597 +out:
16598 +       return err;
16599 +}
16600 +
16601 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16602 +          unsigned int d_type, ino_t *ino)
16603 +{
16604 +       int err, idx;
16605 +       const int isnondir = d_type != DT_DIR;
16606 +
16607 +       /* prevent hardlinked inode number from race condition */
16608 +       if (isnondir) {
16609 +               err = au_xinondir_enter(sb, bindex, h_ino, &idx);
16610 +               if (unlikely(err))
16611 +                       goto out;
16612 +       }
16613 +
16614 +       err = au_xino_read(sb, bindex, h_ino, ino);
16615 +       if (unlikely(err))
16616 +               goto out_xinondir;
16617 +
16618 +       if (!*ino) {
16619 +               err = -EIO;
16620 +               *ino = au_xino_new_ino(sb);
16621 +               if (unlikely(!*ino))
16622 +                       goto out_xinondir;
16623 +               err = au_xino_write(sb, bindex, h_ino, *ino);
16624 +               if (unlikely(err))
16625 +                       goto out_xinondir;
16626 +       }
16627 +
16628 +out_xinondir:
16629 +       if (isnondir && idx >= 0)
16630 +               au_xinondir_leave(sb, bindex, h_ino, idx);
16631 +out:
16632 +       return err;
16633 +}
16634 +
16635 +/* successful returns with iinfo write_locked */
16636 +/* todo: return with unlocked? */
16637 +struct inode *au_new_inode(struct dentry *dentry, int must_new)
16638 +{
16639 +       struct inode *inode, *h_inode;
16640 +       struct dentry *h_dentry;
16641 +       struct super_block *sb;
16642 +       ino_t h_ino, ino;
16643 +       int err, idx, hlinked;
16644 +       aufs_bindex_t btop;
16645 +
16646 +       sb = dentry->d_sb;
16647 +       btop = au_dbtop(dentry);
16648 +       h_dentry = au_h_dptr(dentry, btop);
16649 +       h_inode = d_inode(h_dentry);
16650 +       h_ino = h_inode->i_ino;
16651 +       hlinked = !d_is_dir(h_dentry) && h_inode->i_nlink > 1;
16652 +
16653 +new_ino:
16654 +       /*
16655 +        * stop 'race'-ing between hardlinks under different
16656 +        * parents.
16657 +        */
16658 +       if (hlinked) {
16659 +               err = au_xinondir_enter(sb, btop, h_ino, &idx);
16660 +               inode = ERR_PTR(err);
16661 +               if (unlikely(err))
16662 +                       goto out;
16663 +       }
16664 +
16665 +       err = au_xino_read(sb, btop, h_ino, &ino);
16666 +       inode = ERR_PTR(err);
16667 +       if (unlikely(err))
16668 +               goto out_xinondir;
16669 +
16670 +       if (!ino) {
16671 +               ino = au_xino_new_ino(sb);
16672 +               if (unlikely(!ino)) {
16673 +                       inode = ERR_PTR(-EIO);
16674 +                       goto out_xinondir;
16675 +               }
16676 +       }
16677 +
16678 +       AuDbg("i%lu\n", (unsigned long)ino);
16679 +       inode = au_iget_locked(sb, ino);
16680 +       err = PTR_ERR(inode);
16681 +       if (IS_ERR(inode))
16682 +               goto out_xinondir;
16683 +
16684 +       AuDbg("%lx, new %d\n", inode->i_state, !!(inode->i_state & I_NEW));
16685 +       if (inode->i_state & I_NEW) {
16686 +               ii_write_lock_new_child(inode);
16687 +               err = set_inode(inode, dentry);
16688 +               if (!err) {
16689 +                       unlock_new_inode(inode);
16690 +                       goto out_xinondir; /* success */
16691 +               }
16692 +
16693 +               /*
16694 +                * iget_failed() calls iput(), but we need to call
16695 +                * ii_write_unlock() after iget_failed(). so dirty hack for
16696 +                * i_count.
16697 +                */
16698 +               atomic_inc(&inode->i_count);
16699 +               iget_failed(inode);
16700 +               ii_write_unlock(inode);
16701 +               au_xino_write(sb, btop, h_ino, /*ino*/0);
16702 +               /* ignore this error */
16703 +               goto out_iput;
16704 +       } else if (!must_new && !IS_DEADDIR(inode) && inode->i_nlink) {
16705 +               /*
16706 +                * horrible race condition between lookup, readdir and copyup
16707 +                * (or something).
16708 +                */
16709 +               if (hlinked && idx >= 0)
16710 +                       au_xinondir_leave(sb, btop, h_ino, idx);
16711 +               err = reval_inode(inode, dentry);
16712 +               if (unlikely(err < 0)) {
16713 +                       hlinked = 0;
16714 +                       goto out_iput;
16715 +               }
16716 +               if (!err)
16717 +                       goto out; /* success */
16718 +               else if (hlinked && idx >= 0) {
16719 +                       err = au_xinondir_enter(sb, btop, h_ino, &idx);
16720 +                       if (unlikely(err)) {
16721 +                               iput(inode);
16722 +                               inode = ERR_PTR(err);
16723 +                               goto out;
16724 +                       }
16725 +               }
16726 +       }
16727 +
16728 +       if (unlikely(au_test_fs_unique_ino(h_inode)))
16729 +               AuWarn1("Warning: Un-notified UDBA or repeatedly renamed dir,"
16730 +                       " b%d, %s, %pd, hi%lu, i%lu.\n",
16731 +                       btop, au_sbtype(h_dentry->d_sb), dentry,
16732 +                       (unsigned long)h_ino, (unsigned long)ino);
16733 +       ino = 0;
16734 +       err = au_xino_write(sb, btop, h_ino, /*ino*/0);
16735 +       if (!err) {
16736 +               iput(inode);
16737 +               if (hlinked && idx >= 0)
16738 +                       au_xinondir_leave(sb, btop, h_ino, idx);
16739 +               goto new_ino;
16740 +       }
16741 +
16742 +out_iput:
16743 +       iput(inode);
16744 +       inode = ERR_PTR(err);
16745 +out_xinondir:
16746 +       if (hlinked && idx >= 0)
16747 +               au_xinondir_leave(sb, btop, h_ino, idx);
16748 +out:
16749 +       return inode;
16750 +}
16751 +
16752 +/* ---------------------------------------------------------------------- */
16753 +
16754 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16755 +              struct inode *inode)
16756 +{
16757 +       int err;
16758 +       struct inode *hi;
16759 +
16760 +       err = au_br_rdonly(au_sbr(sb, bindex));
16761 +
16762 +       /* pseudo-link after flushed may happen out of bounds */
16763 +       if (!err
16764 +           && inode
16765 +           && au_ibtop(inode) <= bindex
16766 +           && bindex <= au_ibbot(inode)) {
16767 +               /*
16768 +                * permission check is unnecessary since vfsub routine
16769 +                * will be called later
16770 +                */
16771 +               hi = au_h_iptr(inode, bindex);
16772 +               if (hi)
16773 +                       err = IS_IMMUTABLE(hi) ? -EROFS : 0;
16774 +       }
16775 +
16776 +       return err;
16777 +}
16778 +
16779 +int au_test_h_perm(struct inode *h_inode, int mask)
16780 +{
16781 +       if (uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
16782 +               return 0;
16783 +       return inode_permission(h_inode, mask);
16784 +}
16785 +
16786 +int au_test_h_perm_sio(struct inode *h_inode, int mask)
16787 +{
16788 +       if (au_test_nfs(h_inode->i_sb)
16789 +           && (mask & MAY_WRITE)
16790 +           && S_ISDIR(h_inode->i_mode))
16791 +               mask |= MAY_READ; /* force permission check */
16792 +       return au_test_h_perm(h_inode, mask);
16793 +}
16794 diff -urN /usr/share/empty/fs/aufs/inode.h linux/fs/aufs/inode.h
16795 --- /usr/share/empty/fs/aufs/inode.h    1970-01-01 01:00:00.000000000 +0100
16796 +++ linux/fs/aufs/inode.h       2017-05-06 22:16:52.781553511 +0200
16797 @@ -0,0 +1,692 @@
16798 +/*
16799 + * Copyright (C) 2005-2017 Junjiro R. Okajima
16800 + *
16801 + * This program, aufs is free software; you can redistribute it and/or modify
16802 + * it under the terms of the GNU General Public License as published by
16803 + * the Free Software Foundation; either version 2 of the License, or
16804 + * (at your option) any later version.
16805 + *
16806 + * This program is distributed in the hope that it will be useful,
16807 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16808 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16809 + * GNU General Public License for more details.
16810 + *
16811 + * You should have received a copy of the GNU General Public License
16812 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16813 + */
16814 +
16815 +/*
16816 + * inode operations
16817 + */
16818 +
16819 +#ifndef __AUFS_INODE_H__
16820 +#define __AUFS_INODE_H__
16821 +
16822 +#ifdef __KERNEL__
16823 +
16824 +#include <linux/fsnotify.h>
16825 +#include "rwsem.h"
16826 +
16827 +struct vfsmount;
16828 +
16829 +struct au_hnotify {
16830 +#ifdef CONFIG_AUFS_HNOTIFY
16831 +#ifdef CONFIG_AUFS_HFSNOTIFY
16832 +       /* never use fsnotify_add_vfsmount_mark() */
16833 +       struct fsnotify_mark            hn_mark;
16834 +#endif
16835 +       union {
16836 +               struct inode            *hn_aufs_inode; /* no get/put */
16837 +               struct llist_node       hn_lnode;       /* delayed free */
16838 +       };
16839 +#endif
16840 +} ____cacheline_aligned_in_smp;
16841 +
16842 +struct au_hinode {
16843 +       struct inode            *hi_inode;
16844 +       aufs_bindex_t           hi_id;
16845 +#ifdef CONFIG_AUFS_HNOTIFY
16846 +       struct au_hnotify       *hi_notify;
16847 +#endif
16848 +
16849 +       /* reference to the copied-up whiteout with get/put */
16850 +       struct dentry           *hi_whdentry;
16851 +};
16852 +
16853 +/* ig_flags */
16854 +#define AuIG_HALF_REFRESHED            1
16855 +#define au_ig_ftest(flags, name)       ((flags) & AuIG_##name)
16856 +#define au_ig_fset(flags, name) \
16857 +       do { (flags) |= AuIG_##name; } while (0)
16858 +#define au_ig_fclr(flags, name) \
16859 +       do { (flags) &= ~AuIG_##name; } while (0)
16860 +
16861 +struct au_iigen {
16862 +       spinlock_t      ig_spin;
16863 +       __u32           ig_generation, ig_flags;
16864 +};
16865 +
16866 +struct au_vdir;
16867 +struct au_iinfo {
16868 +       struct au_iigen         ii_generation;
16869 +       struct super_block      *ii_hsb1;       /* no get/put */
16870 +
16871 +       struct au_rwsem         ii_rwsem;
16872 +       aufs_bindex_t           ii_btop, ii_bbot;
16873 +       __u32                   ii_higen;
16874 +       struct au_hinode        *ii_hinode;
16875 +       struct au_vdir          *ii_vdir;
16876 +};
16877 +
16878 +struct au_icntnr {
16879 +       struct au_iinfo iinfo;
16880 +       struct inode vfs_inode;
16881 +       union {
16882 +               struct hlist_node       plink;
16883 +               struct llist_node       lnode;  /* delayed free */
16884 +       };
16885 +} ____cacheline_aligned_in_smp;
16886 +
16887 +/* au_pin flags */
16888 +#define AuPin_DI_LOCKED                1
16889 +#define AuPin_MNT_WRITE                (1 << 1)
16890 +#define au_ftest_pin(flags, name)      ((flags) & AuPin_##name)
16891 +#define au_fset_pin(flags, name) \
16892 +       do { (flags) |= AuPin_##name; } while (0)
16893 +#define au_fclr_pin(flags, name) \
16894 +       do { (flags) &= ~AuPin_##name; } while (0)
16895 +
16896 +struct au_pin {
16897 +       /* input */
16898 +       struct dentry *dentry;
16899 +       unsigned int udba;
16900 +       unsigned char lsc_di, lsc_hi, flags;
16901 +       aufs_bindex_t bindex;
16902 +
16903 +       /* output */
16904 +       struct dentry *parent;
16905 +       struct au_hinode *hdir;
16906 +       struct vfsmount *h_mnt;
16907 +
16908 +       /* temporary unlock/relock for copyup */
16909 +       struct dentry *h_dentry, *h_parent;
16910 +       struct au_branch *br;
16911 +       struct task_struct *task;
16912 +};
16913 +
16914 +void au_pin_hdir_unlock(struct au_pin *p);
16915 +int au_pin_hdir_lock(struct au_pin *p);
16916 +int au_pin_hdir_relock(struct au_pin *p);
16917 +void au_pin_hdir_acquire_nest(struct au_pin *p);
16918 +void au_pin_hdir_release(struct au_pin *p);
16919 +
16920 +/* ---------------------------------------------------------------------- */
16921 +
16922 +static inline struct au_iinfo *au_ii(struct inode *inode)
16923 +{
16924 +       BUG_ON(is_bad_inode(inode));
16925 +       return &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
16926 +}
16927 +
16928 +/* ---------------------------------------------------------------------- */
16929 +
16930 +/* inode.c */
16931 +struct inode *au_igrab(struct inode *inode);
16932 +void au_refresh_iop(struct inode *inode, int force_getattr);
16933 +int au_refresh_hinode_self(struct inode *inode);
16934 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry);
16935 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16936 +          unsigned int d_type, ino_t *ino);
16937 +struct inode *au_new_inode(struct dentry *dentry, int must_new);
16938 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16939 +              struct inode *inode);
16940 +int au_test_h_perm(struct inode *h_inode, int mask);
16941 +int au_test_h_perm_sio(struct inode *h_inode, int mask);
16942 +
16943 +static inline int au_wh_ino(struct super_block *sb, aufs_bindex_t bindex,
16944 +                           ino_t h_ino, unsigned int d_type, ino_t *ino)
16945 +{
16946 +#ifdef CONFIG_AUFS_SHWH
16947 +       return au_ino(sb, bindex, h_ino, d_type, ino);
16948 +#else
16949 +       return 0;
16950 +#endif
16951 +}
16952 +
16953 +/* i_op.c */
16954 +enum {
16955 +       AuIop_SYMLINK,
16956 +       AuIop_DIR,
16957 +       AuIop_OTHER,
16958 +       AuIop_Last
16959 +};
16960 +extern struct inode_operations aufs_iop[AuIop_Last],
16961 +       aufs_iop_nogetattr[AuIop_Last];
16962 +
16963 +/* au_wr_dir flags */
16964 +#define AuWrDir_ADD_ENTRY      1
16965 +#define AuWrDir_ISDIR          (1 << 1)
16966 +#define AuWrDir_TMPFILE                (1 << 2)
16967 +#define au_ftest_wrdir(flags, name)    ((flags) & AuWrDir_##name)
16968 +#define au_fset_wrdir(flags, name) \
16969 +       do { (flags) |= AuWrDir_##name; } while (0)
16970 +#define au_fclr_wrdir(flags, name) \
16971 +       do { (flags) &= ~AuWrDir_##name; } while (0)
16972 +
16973 +struct au_wr_dir_args {
16974 +       aufs_bindex_t force_btgt;
16975 +       unsigned char flags;
16976 +};
16977 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
16978 +             struct au_wr_dir_args *args);
16979 +
16980 +struct dentry *au_pinned_h_parent(struct au_pin *pin);
16981 +void au_pin_init(struct au_pin *pin, struct dentry *dentry,
16982 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
16983 +                unsigned int udba, unsigned char flags);
16984 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
16985 +          unsigned int udba, unsigned char flags) __must_check;
16986 +int au_do_pin(struct au_pin *pin) __must_check;
16987 +void au_unpin(struct au_pin *pin);
16988 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen);
16989 +
16990 +#define AuIcpup_DID_CPUP       1
16991 +#define au_ftest_icpup(flags, name)    ((flags) & AuIcpup_##name)
16992 +#define au_fset_icpup(flags, name) \
16993 +       do { (flags) |= AuIcpup_##name; } while (0)
16994 +#define au_fclr_icpup(flags, name) \
16995 +       do { (flags) &= ~AuIcpup_##name; } while (0)
16996 +
16997 +struct au_icpup_args {
16998 +       unsigned char flags;
16999 +       unsigned char pin_flags;
17000 +       aufs_bindex_t btgt;
17001 +       unsigned int udba;
17002 +       struct au_pin pin;
17003 +       struct path h_path;
17004 +       struct inode *h_inode;
17005 +};
17006 +
17007 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
17008 +                    struct au_icpup_args *a);
17009 +
17010 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path,
17011 +                     int locked);
17012 +
17013 +/* i_op_add.c */
17014 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
17015 +              struct dentry *h_parent, int isdir);
17016 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
17017 +              dev_t dev);
17018 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname);
17019 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
17020 +               bool want_excl);
17021 +struct vfsub_aopen_args;
17022 +int au_aopen_or_create(struct inode *dir, struct dentry *dentry,
17023 +                      struct vfsub_aopen_args *args);
17024 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode);
17025 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
17026 +             struct dentry *dentry);
17027 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
17028 +
17029 +/* i_op_del.c */
17030 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup);
17031 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
17032 +              struct dentry *h_parent, int isdir);
17033 +int aufs_unlink(struct inode *dir, struct dentry *dentry);
17034 +int aufs_rmdir(struct inode *dir, struct dentry *dentry);
17035 +
17036 +/* i_op_ren.c */
17037 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt);
17038 +int aufs_rename(struct inode *src_dir, struct dentry *src_dentry,
17039 +               struct inode *dir, struct dentry *dentry,
17040 +               unsigned int flags);
17041 +
17042 +/* iinfo.c */
17043 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex);
17044 +void au_hiput(struct au_hinode *hinode);
17045 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
17046 +                 struct dentry *h_wh);
17047 +unsigned int au_hi_flags(struct inode *inode, int isdir);
17048 +
17049 +/* hinode flags */
17050 +#define AuHi_XINO      1
17051 +#define AuHi_HNOTIFY   (1 << 1)
17052 +#define au_ftest_hi(flags, name)       ((flags) & AuHi_##name)
17053 +#define au_fset_hi(flags, name) \
17054 +       do { (flags) |= AuHi_##name; } while (0)
17055 +#define au_fclr_hi(flags, name) \
17056 +       do { (flags) &= ~AuHi_##name; } while (0)
17057 +
17058 +#ifndef CONFIG_AUFS_HNOTIFY
17059 +#undef AuHi_HNOTIFY
17060 +#define AuHi_HNOTIFY   0
17061 +#endif
17062 +
17063 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
17064 +                  struct inode *h_inode, unsigned int flags);
17065 +
17066 +void au_update_iigen(struct inode *inode, int half);
17067 +void au_update_ibrange(struct inode *inode, int do_put_zero);
17068 +
17069 +void au_icntnr_init_once(void *_c);
17070 +void au_hinode_init(struct au_hinode *hinode);
17071 +int au_iinfo_init(struct inode *inode);
17072 +void au_iinfo_fin(struct inode *inode);
17073 +int au_hinode_realloc(struct au_iinfo *iinfo, int nbr, int may_shrink);
17074 +
17075 +#ifdef CONFIG_PROC_FS
17076 +/* plink.c */
17077 +int au_plink_maint(struct super_block *sb, int flags);
17078 +struct au_sbinfo;
17079 +void au_plink_maint_leave(struct au_sbinfo *sbinfo);
17080 +int au_plink_maint_enter(struct super_block *sb);
17081 +#ifdef CONFIG_AUFS_DEBUG
17082 +void au_plink_list(struct super_block *sb);
17083 +#else
17084 +AuStubVoid(au_plink_list, struct super_block *sb)
17085 +#endif
17086 +int au_plink_test(struct inode *inode);
17087 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex);
17088 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
17089 +                    struct dentry *h_dentry);
17090 +void au_plink_put(struct super_block *sb, int verbose);
17091 +void au_plink_clean(struct super_block *sb, int verbose);
17092 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id);
17093 +#else
17094 +AuStubInt0(au_plink_maint, struct super_block *sb, int flags);
17095 +AuStubVoid(au_plink_maint_leave, struct au_sbinfo *sbinfo);
17096 +AuStubInt0(au_plink_maint_enter, struct super_block *sb);
17097 +AuStubVoid(au_plink_list, struct super_block *sb);
17098 +AuStubInt0(au_plink_test, struct inode *inode);
17099 +AuStub(struct dentry *, au_plink_lkup, return NULL,
17100 +       struct inode *inode, aufs_bindex_t bindex);
17101 +AuStubVoid(au_plink_append, struct inode *inode, aufs_bindex_t bindex,
17102 +          struct dentry *h_dentry);
17103 +AuStubVoid(au_plink_put, struct super_block *sb, int verbose);
17104 +AuStubVoid(au_plink_clean, struct super_block *sb, int verbose);
17105 +AuStubVoid(au_plink_half_refresh, struct super_block *sb, aufs_bindex_t br_id);
17106 +#endif /* CONFIG_PROC_FS */
17107 +
17108 +#ifdef CONFIG_AUFS_XATTR
17109 +/* xattr.c */
17110 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags,
17111 +                 unsigned int verbose);
17112 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size);
17113 +void au_xattr_init(struct super_block *sb);
17114 +#else
17115 +AuStubInt0(au_cpup_xattr, struct dentry *h_dst, struct dentry *h_src,
17116 +          int ignore_flags, unsigned int verbose);
17117 +AuStubVoid(au_xattr_init, struct super_block *sb);
17118 +#endif
17119 +
17120 +#ifdef CONFIG_FS_POSIX_ACL
17121 +struct posix_acl *aufs_get_acl(struct inode *inode, int type);
17122 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
17123 +#endif
17124 +
17125 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
17126 +enum {
17127 +       AU_XATTR_SET,
17128 +       AU_ACL_SET
17129 +};
17130 +
17131 +struct au_sxattr {
17132 +       int type;
17133 +       union {
17134 +               struct {
17135 +                       const char      *name;
17136 +                       const void      *value;
17137 +                       size_t          size;
17138 +                       int             flags;
17139 +               } set;
17140 +               struct {
17141 +                       struct posix_acl *acl;
17142 +                       int             type;
17143 +               } acl_set;
17144 +       } u;
17145 +};
17146 +ssize_t au_sxattr(struct dentry *dentry, struct inode *inode,
17147 +                 struct au_sxattr *arg);
17148 +#endif
17149 +
17150 +/* ---------------------------------------------------------------------- */
17151 +
17152 +/* lock subclass for iinfo */
17153 +enum {
17154 +       AuLsc_II_CHILD,         /* child first */
17155 +       AuLsc_II_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
17156 +       AuLsc_II_CHILD3,        /* copyup dirs */
17157 +       AuLsc_II_PARENT,        /* see AuLsc_I_PARENT in vfsub.h */
17158 +       AuLsc_II_PARENT2,
17159 +       AuLsc_II_PARENT3,       /* copyup dirs */
17160 +       AuLsc_II_NEW_CHILD
17161 +};
17162 +
17163 +/*
17164 + * ii_read_lock_child, ii_write_lock_child,
17165 + * ii_read_lock_child2, ii_write_lock_child2,
17166 + * ii_read_lock_child3, ii_write_lock_child3,
17167 + * ii_read_lock_parent, ii_write_lock_parent,
17168 + * ii_read_lock_parent2, ii_write_lock_parent2,
17169 + * ii_read_lock_parent3, ii_write_lock_parent3,
17170 + * ii_read_lock_new_child, ii_write_lock_new_child,
17171 + */
17172 +#define AuReadLockFunc(name, lsc) \
17173 +static inline void ii_read_lock_##name(struct inode *i) \
17174 +{ \
17175 +       au_rw_read_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
17176 +}
17177 +
17178 +#define AuWriteLockFunc(name, lsc) \
17179 +static inline void ii_write_lock_##name(struct inode *i) \
17180 +{ \
17181 +       au_rw_write_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
17182 +}
17183 +
17184 +#define AuRWLockFuncs(name, lsc) \
17185 +       AuReadLockFunc(name, lsc) \
17186 +       AuWriteLockFunc(name, lsc)
17187 +
17188 +AuRWLockFuncs(child, CHILD);
17189 +AuRWLockFuncs(child2, CHILD2);
17190 +AuRWLockFuncs(child3, CHILD3);
17191 +AuRWLockFuncs(parent, PARENT);
17192 +AuRWLockFuncs(parent2, PARENT2);
17193 +AuRWLockFuncs(parent3, PARENT3);
17194 +AuRWLockFuncs(new_child, NEW_CHILD);
17195 +
17196 +#undef AuReadLockFunc
17197 +#undef AuWriteLockFunc
17198 +#undef AuRWLockFuncs
17199 +
17200 +/*
17201 + * ii_read_unlock, ii_write_unlock, ii_downgrade_lock
17202 + */
17203 +AuSimpleUnlockRwsemFuncs(ii, struct inode *i, &au_ii(i)->ii_rwsem);
17204 +
17205 +#define IiMustNoWaiters(i)     AuRwMustNoWaiters(&au_ii(i)->ii_rwsem)
17206 +#define IiMustAnyLock(i)       AuRwMustAnyLock(&au_ii(i)->ii_rwsem)
17207 +#define IiMustWriteLock(i)     AuRwMustWriteLock(&au_ii(i)->ii_rwsem)
17208 +
17209 +/* ---------------------------------------------------------------------- */
17210 +
17211 +static inline void au_icntnr_init(struct au_icntnr *c)
17212 +{
17213 +#ifdef CONFIG_AUFS_DEBUG
17214 +       c->vfs_inode.i_mode = 0;
17215 +#endif
17216 +}
17217 +
17218 +static inline unsigned int au_iigen(struct inode *inode, unsigned int *igflags)
17219 +{
17220 +       unsigned int gen;
17221 +       struct au_iinfo *iinfo;
17222 +       struct au_iigen *iigen;
17223 +
17224 +       iinfo = au_ii(inode);
17225 +       iigen = &iinfo->ii_generation;
17226 +       spin_lock(&iigen->ig_spin);
17227 +       if (igflags)
17228 +               *igflags = iigen->ig_flags;
17229 +       gen = iigen->ig_generation;
17230 +       spin_unlock(&iigen->ig_spin);
17231 +
17232 +       return gen;
17233 +}
17234 +
17235 +/* tiny test for inode number */
17236 +/* tmpfs generation is too rough */
17237 +static inline int au_test_higen(struct inode *inode, struct inode *h_inode)
17238 +{
17239 +       struct au_iinfo *iinfo;
17240 +
17241 +       iinfo = au_ii(inode);
17242 +       AuRwMustAnyLock(&iinfo->ii_rwsem);
17243 +       return !(iinfo->ii_hsb1 == h_inode->i_sb
17244 +                && iinfo->ii_higen == h_inode->i_generation);
17245 +}
17246 +
17247 +static inline void au_iigen_dec(struct inode *inode)
17248 +{
17249 +       struct au_iinfo *iinfo;
17250 +       struct au_iigen *iigen;
17251 +
17252 +       iinfo = au_ii(inode);
17253 +       iigen = &iinfo->ii_generation;
17254 +       spin_lock(&iigen->ig_spin);
17255 +       iigen->ig_generation--;
17256 +       spin_unlock(&iigen->ig_spin);
17257 +}
17258 +
17259 +static inline int au_iigen_test(struct inode *inode, unsigned int sigen)
17260 +{
17261 +       int err;
17262 +
17263 +       err = 0;
17264 +       if (unlikely(inode && au_iigen(inode, NULL) != sigen))
17265 +               err = -EIO;
17266 +
17267 +       return err;
17268 +}
17269 +
17270 +/* ---------------------------------------------------------------------- */
17271 +
17272 +static inline struct au_hinode *au_hinode(struct au_iinfo *iinfo,
17273 +                                         aufs_bindex_t bindex)
17274 +{
17275 +       return iinfo->ii_hinode + bindex;
17276 +}
17277 +
17278 +static inline int au_is_bad_inode(struct inode *inode)
17279 +{
17280 +       return !!(is_bad_inode(inode) || !au_hinode(au_ii(inode), 0));
17281 +}
17282 +
17283 +static inline aufs_bindex_t au_ii_br_id(struct inode *inode,
17284 +                                       aufs_bindex_t bindex)
17285 +{
17286 +       IiMustAnyLock(inode);
17287 +       return au_hinode(au_ii(inode), bindex)->hi_id;
17288 +}
17289 +
17290 +static inline aufs_bindex_t au_ibtop(struct inode *inode)
17291 +{
17292 +       IiMustAnyLock(inode);
17293 +       return au_ii(inode)->ii_btop;
17294 +}
17295 +
17296 +static inline aufs_bindex_t au_ibbot(struct inode *inode)
17297 +{
17298 +       IiMustAnyLock(inode);
17299 +       return au_ii(inode)->ii_bbot;
17300 +}
17301 +
17302 +static inline struct au_vdir *au_ivdir(struct inode *inode)
17303 +{
17304 +       IiMustAnyLock(inode);
17305 +       return au_ii(inode)->ii_vdir;
17306 +}
17307 +
17308 +static inline struct dentry *au_hi_wh(struct inode *inode, aufs_bindex_t bindex)
17309 +{
17310 +       IiMustAnyLock(inode);
17311 +       return au_hinode(au_ii(inode), bindex)->hi_whdentry;
17312 +}
17313 +
17314 +static inline void au_set_ibtop(struct inode *inode, aufs_bindex_t bindex)
17315 +{
17316 +       IiMustWriteLock(inode);
17317 +       au_ii(inode)->ii_btop = bindex;
17318 +}
17319 +
17320 +static inline void au_set_ibbot(struct inode *inode, aufs_bindex_t bindex)
17321 +{
17322 +       IiMustWriteLock(inode);
17323 +       au_ii(inode)->ii_bbot = bindex;
17324 +}
17325 +
17326 +static inline void au_set_ivdir(struct inode *inode, struct au_vdir *vdir)
17327 +{
17328 +       IiMustWriteLock(inode);
17329 +       au_ii(inode)->ii_vdir = vdir;
17330 +}
17331 +
17332 +static inline struct au_hinode *au_hi(struct inode *inode, aufs_bindex_t bindex)
17333 +{
17334 +       IiMustAnyLock(inode);
17335 +       return au_hinode(au_ii(inode), bindex);
17336 +}
17337 +
17338 +/* ---------------------------------------------------------------------- */
17339 +
17340 +static inline struct dentry *au_pinned_parent(struct au_pin *pin)
17341 +{
17342 +       if (pin)
17343 +               return pin->parent;
17344 +       return NULL;
17345 +}
17346 +
17347 +static inline struct inode *au_pinned_h_dir(struct au_pin *pin)
17348 +{
17349 +       if (pin && pin->hdir)
17350 +               return pin->hdir->hi_inode;
17351 +       return NULL;
17352 +}
17353 +
17354 +static inline struct au_hinode *au_pinned_hdir(struct au_pin *pin)
17355 +{
17356 +       if (pin)
17357 +               return pin->hdir;
17358 +       return NULL;
17359 +}
17360 +
17361 +static inline void au_pin_set_dentry(struct au_pin *pin, struct dentry *dentry)
17362 +{
17363 +       if (pin)
17364 +               pin->dentry = dentry;
17365 +}
17366 +
17367 +static inline void au_pin_set_parent_lflag(struct au_pin *pin,
17368 +                                          unsigned char lflag)
17369 +{
17370 +       if (pin) {
17371 +               if (lflag)
17372 +                       au_fset_pin(pin->flags, DI_LOCKED);
17373 +               else
17374 +                       au_fclr_pin(pin->flags, DI_LOCKED);
17375 +       }
17376 +}
17377 +
17378 +#if 0 /* reserved */
17379 +static inline void au_pin_set_parent(struct au_pin *pin, struct dentry *parent)
17380 +{
17381 +       if (pin) {
17382 +               dput(pin->parent);
17383 +               pin->parent = dget(parent);
17384 +       }
17385 +}
17386 +#endif
17387 +
17388 +/* ---------------------------------------------------------------------- */
17389 +
17390 +struct au_branch;
17391 +#ifdef CONFIG_AUFS_HNOTIFY
17392 +struct au_hnotify_op {
17393 +       void (*ctl)(struct au_hinode *hinode, int do_set);
17394 +       int (*alloc)(struct au_hinode *hinode);
17395 +
17396 +       /*
17397 +        * if it returns true, the the caller should free hinode->hi_notify,
17398 +        * otherwise ->free() frees it.
17399 +        */
17400 +       int (*free)(struct au_hinode *hinode,
17401 +                   struct au_hnotify *hn) __must_check;
17402 +
17403 +       void (*fin)(void);
17404 +       int (*init)(void);
17405 +
17406 +       int (*reset_br)(unsigned int udba, struct au_branch *br, int perm);
17407 +       void (*fin_br)(struct au_branch *br);
17408 +       int (*init_br)(struct au_branch *br, int perm);
17409 +};
17410 +
17411 +/* hnotify.c */
17412 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode);
17413 +void au_hn_free(struct au_hinode *hinode);
17414 +void au_hn_ctl(struct au_hinode *hinode, int do_set);
17415 +void au_hn_reset(struct inode *inode, unsigned int flags);
17416 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
17417 +              struct qstr *h_child_qstr, struct inode *h_child_inode);
17418 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm);
17419 +int au_hnotify_init_br(struct au_branch *br, int perm);
17420 +void au_hnotify_fin_br(struct au_branch *br);
17421 +int __init au_hnotify_init(void);
17422 +void au_hnotify_fin(void);
17423 +
17424 +/* hfsnotify.c */
17425 +extern const struct au_hnotify_op au_hnotify_op;
17426 +
17427 +static inline
17428 +void au_hn_init(struct au_hinode *hinode)
17429 +{
17430 +       hinode->hi_notify = NULL;
17431 +}
17432 +
17433 +static inline struct au_hnotify *au_hn(struct au_hinode *hinode)
17434 +{
17435 +       return hinode->hi_notify;
17436 +}
17437 +
17438 +#else
17439 +AuStub(int, au_hn_alloc, return -EOPNOTSUPP,
17440 +       struct au_hinode *hinode __maybe_unused,
17441 +       struct inode *inode __maybe_unused)
17442 +AuStub(struct au_hnotify *, au_hn, return NULL, struct au_hinode *hinode)
17443 +AuStubVoid(au_hn_free, struct au_hinode *hinode __maybe_unused)
17444 +AuStubVoid(au_hn_ctl, struct au_hinode *hinode __maybe_unused,
17445 +          int do_set __maybe_unused)
17446 +AuStubVoid(au_hn_reset, struct inode *inode __maybe_unused,
17447 +          unsigned int flags __maybe_unused)
17448 +AuStubInt0(au_hnotify_reset_br, unsigned int udba __maybe_unused,
17449 +          struct au_branch *br __maybe_unused,
17450 +          int perm __maybe_unused)
17451 +AuStubInt0(au_hnotify_init_br, struct au_branch *br __maybe_unused,
17452 +          int perm __maybe_unused)
17453 +AuStubVoid(au_hnotify_fin_br, struct au_branch *br __maybe_unused)
17454 +AuStubInt0(__init au_hnotify_init, void)
17455 +AuStubVoid(au_hnotify_fin, void)
17456 +AuStubVoid(au_hn_init, struct au_hinode *hinode __maybe_unused)
17457 +#endif /* CONFIG_AUFS_HNOTIFY */
17458 +
17459 +static inline void au_hn_suspend(struct au_hinode *hdir)
17460 +{
17461 +       au_hn_ctl(hdir, /*do_set*/0);
17462 +}
17463 +
17464 +static inline void au_hn_resume(struct au_hinode *hdir)
17465 +{
17466 +       au_hn_ctl(hdir, /*do_set*/1);
17467 +}
17468 +
17469 +static inline void au_hn_inode_lock(struct au_hinode *hdir)
17470 +{
17471 +       inode_lock(hdir->hi_inode);
17472 +       au_hn_suspend(hdir);
17473 +}
17474 +
17475 +static inline void au_hn_inode_lock_nested(struct au_hinode *hdir,
17476 +                                         unsigned int sc __maybe_unused)
17477 +{
17478 +       inode_lock_nested(hdir->hi_inode, sc);
17479 +       au_hn_suspend(hdir);
17480 +}
17481 +
17482 +static inline void au_hn_inode_unlock(struct au_hinode *hdir)
17483 +{
17484 +       au_hn_resume(hdir);
17485 +       inode_unlock(hdir->hi_inode);
17486 +}
17487 +
17488 +#endif /* __KERNEL__ */
17489 +#endif /* __AUFS_INODE_H__ */
17490 diff -urN /usr/share/empty/fs/aufs/ioctl.c linux/fs/aufs/ioctl.c
17491 --- /usr/share/empty/fs/aufs/ioctl.c    1970-01-01 01:00:00.000000000 +0100
17492 +++ linux/fs/aufs/ioctl.c       2017-05-06 22:16:52.781553511 +0200
17493 @@ -0,0 +1,219 @@
17494 +/*
17495 + * Copyright (C) 2005-2017 Junjiro R. Okajima
17496 + *
17497 + * This program, aufs is free software; you can redistribute it and/or modify
17498 + * it under the terms of the GNU General Public License as published by
17499 + * the Free Software Foundation; either version 2 of the License, or
17500 + * (at your option) any later version.
17501 + *
17502 + * This program is distributed in the hope that it will be useful,
17503 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17504 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17505 + * GNU General Public License for more details.
17506 + *
17507 + * You should have received a copy of the GNU General Public License
17508 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17509 + */
17510 +
17511 +/*
17512 + * ioctl
17513 + * plink-management and readdir in userspace.
17514 + * assist the pathconf(3) wrapper library.
17515 + * move-down
17516 + * File-based Hierarchical Storage Management.
17517 + */
17518 +
17519 +#include <linux/compat.h>
17520 +#include <linux/file.h>
17521 +#include "aufs.h"
17522 +
17523 +static int au_wbr_fd(struct path *path, struct aufs_wbr_fd __user *arg)
17524 +{
17525 +       int err, fd;
17526 +       aufs_bindex_t wbi, bindex, bbot;
17527 +       struct file *h_file;
17528 +       struct super_block *sb;
17529 +       struct dentry *root;
17530 +       struct au_branch *br;
17531 +       struct aufs_wbr_fd wbrfd = {
17532 +               .oflags = au_dir_roflags,
17533 +               .brid   = -1
17534 +       };
17535 +       const int valid = O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_DIRECTORY
17536 +               | O_NOATIME | O_CLOEXEC;
17537 +
17538 +       AuDebugOn(wbrfd.oflags & ~valid);
17539 +
17540 +       if (arg) {
17541 +               err = copy_from_user(&wbrfd, arg, sizeof(wbrfd));
17542 +               if (unlikely(err)) {
17543 +                       err = -EFAULT;
17544 +                       goto out;
17545 +               }
17546 +
17547 +               err = -EINVAL;
17548 +               AuDbg("wbrfd{0%o, %d}\n", wbrfd.oflags, wbrfd.brid);
17549 +               wbrfd.oflags |= au_dir_roflags;
17550 +               AuDbg("0%o\n", wbrfd.oflags);
17551 +               if (unlikely(wbrfd.oflags & ~valid))
17552 +                       goto out;
17553 +       }
17554 +
17555 +       fd = get_unused_fd_flags(0);
17556 +       err = fd;
17557 +       if (unlikely(fd < 0))
17558 +               goto out;
17559 +
17560 +       h_file = ERR_PTR(-EINVAL);
17561 +       wbi = 0;
17562 +       br = NULL;
17563 +       sb = path->dentry->d_sb;
17564 +       root = sb->s_root;
17565 +       aufs_read_lock(root, AuLock_IR);
17566 +       bbot = au_sbbot(sb);
17567 +       if (wbrfd.brid >= 0) {
17568 +               wbi = au_br_index(sb, wbrfd.brid);
17569 +               if (unlikely(wbi < 0 || wbi > bbot))
17570 +                       goto out_unlock;
17571 +       }
17572 +
17573 +       h_file = ERR_PTR(-ENOENT);
17574 +       br = au_sbr(sb, wbi);
17575 +       if (!au_br_writable(br->br_perm)) {
17576 +               if (arg)
17577 +                       goto out_unlock;
17578 +
17579 +               bindex = wbi + 1;
17580 +               wbi = -1;
17581 +               for (; bindex <= bbot; bindex++) {
17582 +                       br = au_sbr(sb, bindex);
17583 +                       if (au_br_writable(br->br_perm)) {
17584 +                               wbi = bindex;
17585 +                               br = au_sbr(sb, wbi);
17586 +                               break;
17587 +                       }
17588 +               }
17589 +       }
17590 +       AuDbg("wbi %d\n", wbi);
17591 +       if (wbi >= 0)
17592 +               h_file = au_h_open(root, wbi, wbrfd.oflags, NULL,
17593 +                                  /*force_wr*/0);
17594 +
17595 +out_unlock:
17596 +       aufs_read_unlock(root, AuLock_IR);
17597 +       err = PTR_ERR(h_file);
17598 +       if (IS_ERR(h_file))
17599 +               goto out_fd;
17600 +
17601 +       au_br_put(br); /* cf. au_h_open() */
17602 +       fd_install(fd, h_file);
17603 +       err = fd;
17604 +       goto out; /* success */
17605 +
17606 +out_fd:
17607 +       put_unused_fd(fd);
17608 +out:
17609 +       AuTraceErr(err);
17610 +       return err;
17611 +}
17612 +
17613 +/* ---------------------------------------------------------------------- */
17614 +
17615 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg)
17616 +{
17617 +       long err;
17618 +       struct dentry *dentry;
17619 +
17620 +       switch (cmd) {
17621 +       case AUFS_CTL_RDU:
17622 +       case AUFS_CTL_RDU_INO:
17623 +               err = au_rdu_ioctl(file, cmd, arg);
17624 +               break;
17625 +
17626 +       case AUFS_CTL_WBR_FD:
17627 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17628 +               break;
17629 +
17630 +       case AUFS_CTL_IBUSY:
17631 +               err = au_ibusy_ioctl(file, arg);
17632 +               break;
17633 +
17634 +       case AUFS_CTL_BRINFO:
17635 +               err = au_brinfo_ioctl(file, arg);
17636 +               break;
17637 +
17638 +       case AUFS_CTL_FHSM_FD:
17639 +               dentry = file->f_path.dentry;
17640 +               if (IS_ROOT(dentry))
17641 +                       err = au_fhsm_fd(dentry->d_sb, arg);
17642 +               else
17643 +                       err = -ENOTTY;
17644 +               break;
17645 +
17646 +       default:
17647 +               /* do not call the lower */
17648 +               AuDbg("0x%x\n", cmd);
17649 +               err = -ENOTTY;
17650 +       }
17651 +
17652 +       AuTraceErr(err);
17653 +       return err;
17654 +}
17655 +
17656 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg)
17657 +{
17658 +       long err;
17659 +
17660 +       switch (cmd) {
17661 +       case AUFS_CTL_MVDOWN:
17662 +               err = au_mvdown(file->f_path.dentry, (void __user *)arg);
17663 +               break;
17664 +
17665 +       case AUFS_CTL_WBR_FD:
17666 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17667 +               break;
17668 +
17669 +       default:
17670 +               /* do not call the lower */
17671 +               AuDbg("0x%x\n", cmd);
17672 +               err = -ENOTTY;
17673 +       }
17674 +
17675 +       AuTraceErr(err);
17676 +       return err;
17677 +}
17678 +
17679 +#ifdef CONFIG_COMPAT
17680 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
17681 +                          unsigned long arg)
17682 +{
17683 +       long err;
17684 +
17685 +       switch (cmd) {
17686 +       case AUFS_CTL_RDU:
17687 +       case AUFS_CTL_RDU_INO:
17688 +               err = au_rdu_compat_ioctl(file, cmd, arg);
17689 +               break;
17690 +
17691 +       case AUFS_CTL_IBUSY:
17692 +               err = au_ibusy_compat_ioctl(file, arg);
17693 +               break;
17694 +
17695 +       case AUFS_CTL_BRINFO:
17696 +               err = au_brinfo_compat_ioctl(file, arg);
17697 +               break;
17698 +
17699 +       default:
17700 +               err = aufs_ioctl_dir(file, cmd, arg);
17701 +       }
17702 +
17703 +       AuTraceErr(err);
17704 +       return err;
17705 +}
17706 +
17707 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
17708 +                             unsigned long arg)
17709 +{
17710 +       return aufs_ioctl_nondir(file, cmd, (unsigned long)compat_ptr(arg));
17711 +}
17712 +#endif
17713 diff -urN /usr/share/empty/fs/aufs/i_op_add.c linux/fs/aufs/i_op_add.c
17714 --- /usr/share/empty/fs/aufs/i_op_add.c 1970-01-01 01:00:00.000000000 +0100
17715 +++ linux/fs/aufs/i_op_add.c    2017-05-06 22:16:52.778220093 +0200
17716 @@ -0,0 +1,920 @@
17717 +/*
17718 + * Copyright (C) 2005-2017 Junjiro R. Okajima
17719 + *
17720 + * This program, aufs is free software; you can redistribute it and/or modify
17721 + * it under the terms of the GNU General Public License as published by
17722 + * the Free Software Foundation; either version 2 of the License, or
17723 + * (at your option) any later version.
17724 + *
17725 + * This program is distributed in the hope that it will be useful,
17726 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17727 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17728 + * GNU General Public License for more details.
17729 + *
17730 + * You should have received a copy of the GNU General Public License
17731 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17732 + */
17733 +
17734 +/*
17735 + * inode operations (add entry)
17736 + */
17737 +
17738 +#include "aufs.h"
17739 +
17740 +/*
17741 + * final procedure of adding a new entry, except link(2).
17742 + * remove whiteout, instantiate, copyup the parent dir's times and size
17743 + * and update version.
17744 + * if it failed, re-create the removed whiteout.
17745 + */
17746 +static int epilog(struct inode *dir, aufs_bindex_t bindex,
17747 +                 struct dentry *wh_dentry, struct dentry *dentry)
17748 +{
17749 +       int err, rerr;
17750 +       aufs_bindex_t bwh;
17751 +       struct path h_path;
17752 +       struct super_block *sb;
17753 +       struct inode *inode, *h_dir;
17754 +       struct dentry *wh;
17755 +
17756 +       bwh = -1;
17757 +       sb = dir->i_sb;
17758 +       if (wh_dentry) {
17759 +               h_dir = d_inode(wh_dentry->d_parent); /* dir inode is locked */
17760 +               IMustLock(h_dir);
17761 +               AuDebugOn(au_h_iptr(dir, bindex) != h_dir);
17762 +               bwh = au_dbwh(dentry);
17763 +               h_path.dentry = wh_dentry;
17764 +               h_path.mnt = au_sbr_mnt(sb, bindex);
17765 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path,
17766 +                                         dentry);
17767 +               if (unlikely(err))
17768 +                       goto out;
17769 +       }
17770 +
17771 +       inode = au_new_inode(dentry, /*must_new*/1);
17772 +       if (!IS_ERR(inode)) {
17773 +               d_instantiate(dentry, inode);
17774 +               dir = d_inode(dentry->d_parent); /* dir inode is locked */
17775 +               IMustLock(dir);
17776 +               au_dir_ts(dir, bindex);
17777 +               dir->i_version++;
17778 +               au_fhsm_wrote(sb, bindex, /*force*/0);
17779 +               return 0; /* success */
17780 +       }
17781 +
17782 +       err = PTR_ERR(inode);
17783 +       if (!wh_dentry)
17784 +               goto out;
17785 +
17786 +       /* revert */
17787 +       /* dir inode is locked */
17788 +       wh = au_wh_create(dentry, bwh, wh_dentry->d_parent);
17789 +       rerr = PTR_ERR(wh);
17790 +       if (IS_ERR(wh)) {
17791 +               AuIOErr("%pd reverting whiteout failed(%d, %d)\n",
17792 +                       dentry, err, rerr);
17793 +               err = -EIO;
17794 +       } else
17795 +               dput(wh);
17796 +
17797 +out:
17798 +       return err;
17799 +}
17800 +
17801 +static int au_d_may_add(struct dentry *dentry)
17802 +{
17803 +       int err;
17804 +
17805 +       err = 0;
17806 +       if (unlikely(d_unhashed(dentry)))
17807 +               err = -ENOENT;
17808 +       if (unlikely(d_really_is_positive(dentry)))
17809 +               err = -EEXIST;
17810 +       return err;
17811 +}
17812 +
17813 +/*
17814 + * simple tests for the adding inode operations.
17815 + * following the checks in vfs, plus the parent-child relationship.
17816 + */
17817 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
17818 +              struct dentry *h_parent, int isdir)
17819 +{
17820 +       int err;
17821 +       umode_t h_mode;
17822 +       struct dentry *h_dentry;
17823 +       struct inode *h_inode;
17824 +
17825 +       err = -ENAMETOOLONG;
17826 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17827 +               goto out;
17828 +
17829 +       h_dentry = au_h_dptr(dentry, bindex);
17830 +       if (d_really_is_negative(dentry)) {
17831 +               err = -EEXIST;
17832 +               if (unlikely(d_is_positive(h_dentry)))
17833 +                       goto out;
17834 +       } else {
17835 +               /* rename(2) case */
17836 +               err = -EIO;
17837 +               if (unlikely(d_is_negative(h_dentry)))
17838 +                       goto out;
17839 +               h_inode = d_inode(h_dentry);
17840 +               if (unlikely(!h_inode->i_nlink))
17841 +                       goto out;
17842 +
17843 +               h_mode = h_inode->i_mode;
17844 +               if (!isdir) {
17845 +                       err = -EISDIR;
17846 +                       if (unlikely(S_ISDIR(h_mode)))
17847 +                               goto out;
17848 +               } else if (unlikely(!S_ISDIR(h_mode))) {
17849 +                       err = -ENOTDIR;
17850 +                       goto out;
17851 +               }
17852 +       }
17853 +
17854 +       err = 0;
17855 +       /* expected parent dir is locked */
17856 +       if (unlikely(h_parent != h_dentry->d_parent))
17857 +               err = -EIO;
17858 +
17859 +out:
17860 +       AuTraceErr(err);
17861 +       return err;
17862 +}
17863 +
17864 +/*
17865 + * initial procedure of adding a new entry.
17866 + * prepare writable branch and the parent dir, lock it,
17867 + * and lookup whiteout for the new entry.
17868 + */
17869 +static struct dentry*
17870 +lock_hdir_lkup_wh(struct dentry *dentry, struct au_dtime *dt,
17871 +                 struct dentry *src_dentry, struct au_pin *pin,
17872 +                 struct au_wr_dir_args *wr_dir_args)
17873 +{
17874 +       struct dentry *wh_dentry, *h_parent;
17875 +       struct super_block *sb;
17876 +       struct au_branch *br;
17877 +       int err;
17878 +       unsigned int udba;
17879 +       aufs_bindex_t bcpup;
17880 +
17881 +       AuDbg("%pd\n", dentry);
17882 +
17883 +       err = au_wr_dir(dentry, src_dentry, wr_dir_args);
17884 +       bcpup = err;
17885 +       wh_dentry = ERR_PTR(err);
17886 +       if (unlikely(err < 0))
17887 +               goto out;
17888 +
17889 +       sb = dentry->d_sb;
17890 +       udba = au_opt_udba(sb);
17891 +       err = au_pin(pin, dentry, bcpup, udba,
17892 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17893 +       wh_dentry = ERR_PTR(err);
17894 +       if (unlikely(err))
17895 +               goto out;
17896 +
17897 +       h_parent = au_pinned_h_parent(pin);
17898 +       if (udba != AuOpt_UDBA_NONE
17899 +           && au_dbtop(dentry) == bcpup)
17900 +               err = au_may_add(dentry, bcpup, h_parent,
17901 +                                au_ftest_wrdir(wr_dir_args->flags, ISDIR));
17902 +       else if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17903 +               err = -ENAMETOOLONG;
17904 +       wh_dentry = ERR_PTR(err);
17905 +       if (unlikely(err))
17906 +               goto out_unpin;
17907 +
17908 +       br = au_sbr(sb, bcpup);
17909 +       if (dt) {
17910 +               struct path tmp = {
17911 +                       .dentry = h_parent,
17912 +                       .mnt    = au_br_mnt(br)
17913 +               };
17914 +               au_dtime_store(dt, au_pinned_parent(pin), &tmp);
17915 +       }
17916 +
17917 +       wh_dentry = NULL;
17918 +       if (bcpup != au_dbwh(dentry))
17919 +               goto out; /* success */
17920 +
17921 +       /*
17922 +        * ENAMETOOLONG here means that if we allowed create such name, then it
17923 +        * would not be able to removed in the future. So we don't allow such
17924 +        * name here and we don't handle ENAMETOOLONG differently here.
17925 +        */
17926 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
17927 +
17928 +out_unpin:
17929 +       if (IS_ERR(wh_dentry))
17930 +               au_unpin(pin);
17931 +out:
17932 +       return wh_dentry;
17933 +}
17934 +
17935 +/* ---------------------------------------------------------------------- */
17936 +
17937 +enum { Mknod, Symlink, Creat };
17938 +struct simple_arg {
17939 +       int type;
17940 +       union {
17941 +               struct {
17942 +                       umode_t                 mode;
17943 +                       bool                    want_excl;
17944 +                       bool                    try_aopen;
17945 +                       struct vfsub_aopen_args *aopen;
17946 +               } c;
17947 +               struct {
17948 +                       const char *symname;
17949 +               } s;
17950 +               struct {
17951 +                       umode_t mode;
17952 +                       dev_t dev;
17953 +               } m;
17954 +       } u;
17955 +};
17956 +
17957 +static int add_simple(struct inode *dir, struct dentry *dentry,
17958 +                     struct simple_arg *arg)
17959 +{
17960 +       int err, rerr;
17961 +       aufs_bindex_t btop;
17962 +       unsigned char created;
17963 +       const unsigned char try_aopen
17964 +               = (arg->type == Creat && arg->u.c.try_aopen);
17965 +       struct dentry *wh_dentry, *parent;
17966 +       struct inode *h_dir;
17967 +       struct super_block *sb;
17968 +       struct au_branch *br;
17969 +       /* to reuduce stack size */
17970 +       struct {
17971 +               struct au_dtime dt;
17972 +               struct au_pin pin;
17973 +               struct path h_path;
17974 +               struct au_wr_dir_args wr_dir_args;
17975 +       } *a;
17976 +
17977 +       AuDbg("%pd\n", dentry);
17978 +       IMustLock(dir);
17979 +
17980 +       err = -ENOMEM;
17981 +       a = kmalloc(sizeof(*a), GFP_NOFS);
17982 +       if (unlikely(!a))
17983 +               goto out;
17984 +       a->wr_dir_args.force_btgt = -1;
17985 +       a->wr_dir_args.flags = AuWrDir_ADD_ENTRY;
17986 +
17987 +       parent = dentry->d_parent; /* dir inode is locked */
17988 +       if (!try_aopen) {
17989 +               err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
17990 +               if (unlikely(err))
17991 +                       goto out_free;
17992 +       }
17993 +       err = au_d_may_add(dentry);
17994 +       if (unlikely(err))
17995 +               goto out_unlock;
17996 +       if (!try_aopen)
17997 +               di_write_lock_parent(parent);
17998 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
17999 +                                     &a->pin, &a->wr_dir_args);
18000 +       err = PTR_ERR(wh_dentry);
18001 +       if (IS_ERR(wh_dentry))
18002 +               goto out_parent;
18003 +
18004 +       btop = au_dbtop(dentry);
18005 +       sb = dentry->d_sb;
18006 +       br = au_sbr(sb, btop);
18007 +       a->h_path.dentry = au_h_dptr(dentry, btop);
18008 +       a->h_path.mnt = au_br_mnt(br);
18009 +       h_dir = au_pinned_h_dir(&a->pin);
18010 +       switch (arg->type) {
18011 +       case Creat:
18012 +               err = 0;
18013 +               if (!try_aopen || !h_dir->i_op->atomic_open)
18014 +                       err = vfsub_create(h_dir, &a->h_path, arg->u.c.mode,
18015 +                                          arg->u.c.want_excl);
18016 +               else
18017 +                       err = vfsub_atomic_open(h_dir, a->h_path.dentry,
18018 +                                               arg->u.c.aopen, br);
18019 +               break;
18020 +       case Symlink:
18021 +               err = vfsub_symlink(h_dir, &a->h_path, arg->u.s.symname);
18022 +               break;
18023 +       case Mknod:
18024 +               err = vfsub_mknod(h_dir, &a->h_path, arg->u.m.mode,
18025 +                                 arg->u.m.dev);
18026 +               break;
18027 +       default:
18028 +               BUG();
18029 +       }
18030 +       created = !err;
18031 +       if (!err)
18032 +               err = epilog(dir, btop, wh_dentry, dentry);
18033 +
18034 +       /* revert */
18035 +       if (unlikely(created && err && d_is_positive(a->h_path.dentry))) {
18036 +               /* no delegation since it is just created */
18037 +               rerr = vfsub_unlink(h_dir, &a->h_path, /*delegated*/NULL,
18038 +                                   /*force*/0);
18039 +               if (rerr) {
18040 +                       AuIOErr("%pd revert failure(%d, %d)\n",
18041 +                               dentry, err, rerr);
18042 +                       err = -EIO;
18043 +               }
18044 +               au_dtime_revert(&a->dt);
18045 +       }
18046 +
18047 +       if (!err && try_aopen && !h_dir->i_op->atomic_open)
18048 +               *arg->u.c.aopen->opened |= FILE_CREATED;
18049 +
18050 +       au_unpin(&a->pin);
18051 +       dput(wh_dentry);
18052 +
18053 +out_parent:
18054 +       if (!try_aopen)
18055 +               di_write_unlock(parent);
18056 +out_unlock:
18057 +       if (unlikely(err)) {
18058 +               au_update_dbtop(dentry);
18059 +               d_drop(dentry);
18060 +       }
18061 +       if (!try_aopen)
18062 +               aufs_read_unlock(dentry, AuLock_DW);
18063 +out_free:
18064 +       au_delayed_kfree(a);
18065 +out:
18066 +       return err;
18067 +}
18068 +
18069 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
18070 +              dev_t dev)
18071 +{
18072 +       struct simple_arg arg = {
18073 +               .type = Mknod,
18074 +               .u.m = {
18075 +                       .mode   = mode,
18076 +                       .dev    = dev
18077 +               }
18078 +       };
18079 +       return add_simple(dir, dentry, &arg);
18080 +}
18081 +
18082 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
18083 +{
18084 +       struct simple_arg arg = {
18085 +               .type = Symlink,
18086 +               .u.s.symname = symname
18087 +       };
18088 +       return add_simple(dir, dentry, &arg);
18089 +}
18090 +
18091 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
18092 +               bool want_excl)
18093 +{
18094 +       struct simple_arg arg = {
18095 +               .type = Creat,
18096 +               .u.c = {
18097 +                       .mode           = mode,
18098 +                       .want_excl      = want_excl
18099 +               }
18100 +       };
18101 +       return add_simple(dir, dentry, &arg);
18102 +}
18103 +
18104 +int au_aopen_or_create(struct inode *dir, struct dentry *dentry,
18105 +                      struct vfsub_aopen_args *aopen_args)
18106 +{
18107 +       struct simple_arg arg = {
18108 +               .type = Creat,
18109 +               .u.c = {
18110 +                       .mode           = aopen_args->create_mode,
18111 +                       .want_excl      = aopen_args->open_flag & O_EXCL,
18112 +                       .try_aopen      = true,
18113 +                       .aopen          = aopen_args
18114 +               }
18115 +       };
18116 +       return add_simple(dir, dentry, &arg);
18117 +}
18118 +
18119 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
18120 +{
18121 +       int err;
18122 +       aufs_bindex_t bindex;
18123 +       struct super_block *sb;
18124 +       struct dentry *parent, *h_parent, *h_dentry;
18125 +       struct inode *h_dir, *inode;
18126 +       struct vfsmount *h_mnt;
18127 +       struct au_wr_dir_args wr_dir_args = {
18128 +               .force_btgt     = -1,
18129 +               .flags          = AuWrDir_TMPFILE
18130 +       };
18131 +
18132 +       /* copy-up may happen */
18133 +       inode_lock(dir);
18134 +
18135 +       sb = dir->i_sb;
18136 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18137 +       if (unlikely(err))
18138 +               goto out;
18139 +
18140 +       err = au_di_init(dentry);
18141 +       if (unlikely(err))
18142 +               goto out_si;
18143 +
18144 +       err = -EBUSY;
18145 +       parent = d_find_any_alias(dir);
18146 +       AuDebugOn(!parent);
18147 +       di_write_lock_parent(parent);
18148 +       if (unlikely(d_inode(parent) != dir))
18149 +               goto out_parent;
18150 +
18151 +       err = au_digen_test(parent, au_sigen(sb));
18152 +       if (unlikely(err))
18153 +               goto out_parent;
18154 +
18155 +       bindex = au_dbtop(parent);
18156 +       au_set_dbtop(dentry, bindex);
18157 +       au_set_dbbot(dentry, bindex);
18158 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
18159 +       bindex = err;
18160 +       if (unlikely(err < 0))
18161 +               goto out_parent;
18162 +
18163 +       err = -EOPNOTSUPP;
18164 +       h_dir = au_h_iptr(dir, bindex);
18165 +       if (unlikely(!h_dir->i_op->tmpfile))
18166 +               goto out_parent;
18167 +
18168 +       h_mnt = au_sbr_mnt(sb, bindex);
18169 +       err = vfsub_mnt_want_write(h_mnt);
18170 +       if (unlikely(err))
18171 +               goto out_parent;
18172 +
18173 +       h_parent = au_h_dptr(parent, bindex);
18174 +       h_dentry = vfs_tmpfile(h_parent, mode, /*open_flag*/0);
18175 +       if (IS_ERR(h_dentry)) {
18176 +               err = PTR_ERR(h_dentry);
18177 +               goto out_mnt;
18178 +       }
18179 +
18180 +       au_set_dbtop(dentry, bindex);
18181 +       au_set_dbbot(dentry, bindex);
18182 +       au_set_h_dptr(dentry, bindex, dget(h_dentry));
18183 +       inode = au_new_inode(dentry, /*must_new*/1);
18184 +       if (IS_ERR(inode)) {
18185 +               err = PTR_ERR(inode);
18186 +               au_set_h_dptr(dentry, bindex, NULL);
18187 +               au_set_dbtop(dentry, -1);
18188 +               au_set_dbbot(dentry, -1);
18189 +       } else {
18190 +               if (!inode->i_nlink)
18191 +                       set_nlink(inode, 1);
18192 +               d_tmpfile(dentry, inode);
18193 +               au_di(dentry)->di_tmpfile = 1;
18194 +
18195 +               /* update without i_mutex */
18196 +               if (au_ibtop(dir) == au_dbtop(dentry))
18197 +                       au_cpup_attr_timesizes(dir);
18198 +       }
18199 +       dput(h_dentry);
18200 +
18201 +out_mnt:
18202 +       vfsub_mnt_drop_write(h_mnt);
18203 +out_parent:
18204 +       di_write_unlock(parent);
18205 +       dput(parent);
18206 +       di_write_unlock(dentry);
18207 +       if (unlikely(err)) {
18208 +               au_di_fin(dentry);
18209 +               dentry->d_fsdata = NULL;
18210 +       }
18211 +out_si:
18212 +       si_read_unlock(sb);
18213 +out:
18214 +       inode_unlock(dir);
18215 +       return err;
18216 +}
18217 +
18218 +/* ---------------------------------------------------------------------- */
18219 +
18220 +struct au_link_args {
18221 +       aufs_bindex_t bdst, bsrc;
18222 +       struct au_pin pin;
18223 +       struct path h_path;
18224 +       struct dentry *src_parent, *parent;
18225 +};
18226 +
18227 +static int au_cpup_before_link(struct dentry *src_dentry,
18228 +                              struct au_link_args *a)
18229 +{
18230 +       int err;
18231 +       struct dentry *h_src_dentry;
18232 +       struct au_cp_generic cpg = {
18233 +               .dentry = src_dentry,
18234 +               .bdst   = a->bdst,
18235 +               .bsrc   = a->bsrc,
18236 +               .len    = -1,
18237 +               .pin    = &a->pin,
18238 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN /* | AuCpup_KEEPLINO */
18239 +       };
18240 +
18241 +       di_read_lock_parent(a->src_parent, AuLock_IR);
18242 +       err = au_test_and_cpup_dirs(src_dentry, a->bdst);
18243 +       if (unlikely(err))
18244 +               goto out;
18245 +
18246 +       h_src_dentry = au_h_dptr(src_dentry, a->bsrc);
18247 +       err = au_pin(&a->pin, src_dentry, a->bdst,
18248 +                    au_opt_udba(src_dentry->d_sb),
18249 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
18250 +       if (unlikely(err))
18251 +               goto out;
18252 +
18253 +       err = au_sio_cpup_simple(&cpg);
18254 +       au_unpin(&a->pin);
18255 +
18256 +out:
18257 +       di_read_unlock(a->src_parent, AuLock_IR);
18258 +       return err;
18259 +}
18260 +
18261 +static int au_cpup_or_link(struct dentry *src_dentry, struct dentry *dentry,
18262 +                          struct au_link_args *a)
18263 +{
18264 +       int err;
18265 +       unsigned char plink;
18266 +       aufs_bindex_t bbot;
18267 +       struct dentry *h_src_dentry;
18268 +       struct inode *h_inode, *inode, *delegated;
18269 +       struct super_block *sb;
18270 +       struct file *h_file;
18271 +
18272 +       plink = 0;
18273 +       h_inode = NULL;
18274 +       sb = src_dentry->d_sb;
18275 +       inode = d_inode(src_dentry);
18276 +       if (au_ibtop(inode) <= a->bdst)
18277 +               h_inode = au_h_iptr(inode, a->bdst);
18278 +       if (!h_inode || !h_inode->i_nlink) {
18279 +               /* copyup src_dentry as the name of dentry. */
18280 +               bbot = au_dbbot(dentry);
18281 +               if (bbot < a->bsrc)
18282 +                       au_set_dbbot(dentry, a->bsrc);
18283 +               au_set_h_dptr(dentry, a->bsrc,
18284 +                             dget(au_h_dptr(src_dentry, a->bsrc)));
18285 +               dget(a->h_path.dentry);
18286 +               au_set_h_dptr(dentry, a->bdst, NULL);
18287 +               AuDbg("temporary d_inode...\n");
18288 +               spin_lock(&dentry->d_lock);
18289 +               dentry->d_inode = d_inode(src_dentry); /* tmp */
18290 +               spin_unlock(&dentry->d_lock);
18291 +               h_file = au_h_open_pre(dentry, a->bsrc, /*force_wr*/0);
18292 +               if (IS_ERR(h_file))
18293 +                       err = PTR_ERR(h_file);
18294 +               else {
18295 +                       struct au_cp_generic cpg = {
18296 +                               .dentry = dentry,
18297 +                               .bdst   = a->bdst,
18298 +                               .bsrc   = -1,
18299 +                               .len    = -1,
18300 +                               .pin    = &a->pin,
18301 +                               .flags  = AuCpup_KEEPLINO
18302 +                       };
18303 +                       err = au_sio_cpup_simple(&cpg);
18304 +                       au_h_open_post(dentry, a->bsrc, h_file);
18305 +                       if (!err) {
18306 +                               dput(a->h_path.dentry);
18307 +                               a->h_path.dentry = au_h_dptr(dentry, a->bdst);
18308 +                       } else
18309 +                               au_set_h_dptr(dentry, a->bdst,
18310 +                                             a->h_path.dentry);
18311 +               }
18312 +               spin_lock(&dentry->d_lock);
18313 +               dentry->d_inode = NULL; /* restore */
18314 +               spin_unlock(&dentry->d_lock);
18315 +               AuDbg("temporary d_inode...done\n");
18316 +               au_set_h_dptr(dentry, a->bsrc, NULL);
18317 +               au_set_dbbot(dentry, bbot);
18318 +       } else {
18319 +               /* the inode of src_dentry already exists on a.bdst branch */
18320 +               h_src_dentry = d_find_alias(h_inode);
18321 +               if (!h_src_dentry && au_plink_test(inode)) {
18322 +                       plink = 1;
18323 +                       h_src_dentry = au_plink_lkup(inode, a->bdst);
18324 +                       err = PTR_ERR(h_src_dentry);
18325 +                       if (IS_ERR(h_src_dentry))
18326 +                               goto out;
18327 +
18328 +                       if (unlikely(d_is_negative(h_src_dentry))) {
18329 +                               dput(h_src_dentry);
18330 +                               h_src_dentry = NULL;
18331 +                       }
18332 +
18333 +               }
18334 +               if (h_src_dentry) {
18335 +                       delegated = NULL;
18336 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
18337 +                                        &a->h_path, &delegated);
18338 +                       if (unlikely(err == -EWOULDBLOCK)) {
18339 +                               pr_warn("cannot retry for NFSv4 delegation"
18340 +                                       " for an internal link\n");
18341 +                               iput(delegated);
18342 +                       }
18343 +                       dput(h_src_dentry);
18344 +               } else {
18345 +                       AuIOErr("no dentry found for hi%lu on b%d\n",
18346 +                               h_inode->i_ino, a->bdst);
18347 +                       err = -EIO;
18348 +               }
18349 +       }
18350 +
18351 +       if (!err && !plink)
18352 +               au_plink_append(inode, a->bdst, a->h_path.dentry);
18353 +
18354 +out:
18355 +       AuTraceErr(err);
18356 +       return err;
18357 +}
18358 +
18359 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
18360 +             struct dentry *dentry)
18361 +{
18362 +       int err, rerr;
18363 +       struct au_dtime dt;
18364 +       struct au_link_args *a;
18365 +       struct dentry *wh_dentry, *h_src_dentry;
18366 +       struct inode *inode, *delegated;
18367 +       struct super_block *sb;
18368 +       struct au_wr_dir_args wr_dir_args = {
18369 +               /* .force_btgt  = -1, */
18370 +               .flags          = AuWrDir_ADD_ENTRY
18371 +       };
18372 +
18373 +       IMustLock(dir);
18374 +       inode = d_inode(src_dentry);
18375 +       IMustLock(inode);
18376 +
18377 +       err = -ENOMEM;
18378 +       a = kzalloc(sizeof(*a), GFP_NOFS);
18379 +       if (unlikely(!a))
18380 +               goto out;
18381 +
18382 +       a->parent = dentry->d_parent; /* dir inode is locked */
18383 +       err = aufs_read_and_write_lock2(dentry, src_dentry,
18384 +                                       AuLock_NOPLM | AuLock_GEN);
18385 +       if (unlikely(err))
18386 +               goto out_kfree;
18387 +       err = au_d_linkable(src_dentry);
18388 +       if (unlikely(err))
18389 +               goto out_unlock;
18390 +       err = au_d_may_add(dentry);
18391 +       if (unlikely(err))
18392 +               goto out_unlock;
18393 +
18394 +       a->src_parent = dget_parent(src_dentry);
18395 +       wr_dir_args.force_btgt = au_ibtop(inode);
18396 +
18397 +       di_write_lock_parent(a->parent);
18398 +       wr_dir_args.force_btgt = au_wbr(dentry, wr_dir_args.force_btgt);
18399 +       wh_dentry = lock_hdir_lkup_wh(dentry, &dt, src_dentry, &a->pin,
18400 +                                     &wr_dir_args);
18401 +       err = PTR_ERR(wh_dentry);
18402 +       if (IS_ERR(wh_dentry))
18403 +               goto out_parent;
18404 +
18405 +       err = 0;
18406 +       sb = dentry->d_sb;
18407 +       a->bdst = au_dbtop(dentry);
18408 +       a->h_path.dentry = au_h_dptr(dentry, a->bdst);
18409 +       a->h_path.mnt = au_sbr_mnt(sb, a->bdst);
18410 +       a->bsrc = au_ibtop(inode);
18411 +       h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
18412 +       if (!h_src_dentry && au_di(src_dentry)->di_tmpfile)
18413 +               h_src_dentry = dget(au_hi_wh(inode, a->bsrc));
18414 +       if (!h_src_dentry) {
18415 +               a->bsrc = au_dbtop(src_dentry);
18416 +               h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
18417 +               AuDebugOn(!h_src_dentry);
18418 +       } else if (IS_ERR(h_src_dentry)) {
18419 +               err = PTR_ERR(h_src_dentry);
18420 +               goto out_parent;
18421 +       }
18422 +
18423 +       /*
18424 +        * aufs doesn't touch the credential so
18425 +        * security_dentry_create_files_as() is unnecrssary.
18426 +        */
18427 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
18428 +               if (a->bdst < a->bsrc
18429 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */)
18430 +                       err = au_cpup_or_link(src_dentry, dentry, a);
18431 +               else {
18432 +                       delegated = NULL;
18433 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
18434 +                                        &a->h_path, &delegated);
18435 +                       if (unlikely(err == -EWOULDBLOCK)) {
18436 +                               pr_warn("cannot retry for NFSv4 delegation"
18437 +                                       " for an internal link\n");
18438 +                               iput(delegated);
18439 +                       }
18440 +               }
18441 +               dput(h_src_dentry);
18442 +       } else {
18443 +               /*
18444 +                * copyup src_dentry to the branch we process,
18445 +                * and then link(2) to it.
18446 +                */
18447 +               dput(h_src_dentry);
18448 +               if (a->bdst < a->bsrc
18449 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */) {
18450 +                       au_unpin(&a->pin);
18451 +                       di_write_unlock(a->parent);
18452 +                       err = au_cpup_before_link(src_dentry, a);
18453 +                       di_write_lock_parent(a->parent);
18454 +                       if (!err)
18455 +                               err = au_pin(&a->pin, dentry, a->bdst,
18456 +                                            au_opt_udba(sb),
18457 +                                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
18458 +                       if (unlikely(err))
18459 +                               goto out_wh;
18460 +               }
18461 +               if (!err) {
18462 +                       h_src_dentry = au_h_dptr(src_dentry, a->bdst);
18463 +                       err = -ENOENT;
18464 +                       if (h_src_dentry && d_is_positive(h_src_dentry)) {
18465 +                               delegated = NULL;
18466 +                               err = vfsub_link(h_src_dentry,
18467 +                                                au_pinned_h_dir(&a->pin),
18468 +                                                &a->h_path, &delegated);
18469 +                               if (unlikely(err == -EWOULDBLOCK)) {
18470 +                                       pr_warn("cannot retry"
18471 +                                               " for NFSv4 delegation"
18472 +                                               " for an internal link\n");
18473 +                                       iput(delegated);
18474 +                               }
18475 +                       }
18476 +               }
18477 +       }
18478 +       if (unlikely(err))
18479 +               goto out_unpin;
18480 +
18481 +       if (wh_dentry) {
18482 +               a->h_path.dentry = wh_dentry;
18483 +               err = au_wh_unlink_dentry(au_pinned_h_dir(&a->pin), &a->h_path,
18484 +                                         dentry);
18485 +               if (unlikely(err))
18486 +                       goto out_revert;
18487 +       }
18488 +
18489 +       au_dir_ts(dir, a->bdst);
18490 +       dir->i_version++;
18491 +       inc_nlink(inode);
18492 +       inode->i_ctime = dir->i_ctime;
18493 +       d_instantiate(dentry, au_igrab(inode));
18494 +       if (d_unhashed(a->h_path.dentry))
18495 +               /* some filesystem calls d_drop() */
18496 +               d_drop(dentry);
18497 +       /* some filesystems consume an inode even hardlink */
18498 +       au_fhsm_wrote(sb, a->bdst, /*force*/0);
18499 +       goto out_unpin; /* success */
18500 +
18501 +out_revert:
18502 +       /* no delegation since it is just created */
18503 +       rerr = vfsub_unlink(au_pinned_h_dir(&a->pin), &a->h_path,
18504 +                           /*delegated*/NULL, /*force*/0);
18505 +       if (unlikely(rerr)) {
18506 +               AuIOErr("%pd reverting failed(%d, %d)\n", dentry, err, rerr);
18507 +               err = -EIO;
18508 +       }
18509 +       au_dtime_revert(&dt);
18510 +out_unpin:
18511 +       au_unpin(&a->pin);
18512 +out_wh:
18513 +       dput(wh_dentry);
18514 +out_parent:
18515 +       di_write_unlock(a->parent);
18516 +       dput(a->src_parent);
18517 +out_unlock:
18518 +       if (unlikely(err)) {
18519 +               au_update_dbtop(dentry);
18520 +               d_drop(dentry);
18521 +       }
18522 +       aufs_read_and_write_unlock2(dentry, src_dentry);
18523 +out_kfree:
18524 +       au_delayed_kfree(a);
18525 +out:
18526 +       AuTraceErr(err);
18527 +       return err;
18528 +}
18529 +
18530 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
18531 +{
18532 +       int err, rerr;
18533 +       aufs_bindex_t bindex;
18534 +       unsigned char diropq;
18535 +       struct path h_path;
18536 +       struct dentry *wh_dentry, *parent, *opq_dentry;
18537 +       struct inode *h_inode;
18538 +       struct super_block *sb;
18539 +       struct {
18540 +               struct au_pin pin;
18541 +               struct au_dtime dt;
18542 +       } *a; /* reduce the stack usage */
18543 +       struct au_wr_dir_args wr_dir_args = {
18544 +               .force_btgt     = -1,
18545 +               .flags          = AuWrDir_ADD_ENTRY | AuWrDir_ISDIR
18546 +       };
18547 +
18548 +       IMustLock(dir);
18549 +
18550 +       err = -ENOMEM;
18551 +       a = kmalloc(sizeof(*a), GFP_NOFS);
18552 +       if (unlikely(!a))
18553 +               goto out;
18554 +
18555 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
18556 +       if (unlikely(err))
18557 +               goto out_free;
18558 +       err = au_d_may_add(dentry);
18559 +       if (unlikely(err))
18560 +               goto out_unlock;
18561 +
18562 +       parent = dentry->d_parent; /* dir inode is locked */
18563 +       di_write_lock_parent(parent);
18564 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
18565 +                                     &a->pin, &wr_dir_args);
18566 +       err = PTR_ERR(wh_dentry);
18567 +       if (IS_ERR(wh_dentry))
18568 +               goto out_parent;
18569 +
18570 +       sb = dentry->d_sb;
18571 +       bindex = au_dbtop(dentry);
18572 +       h_path.dentry = au_h_dptr(dentry, bindex);
18573 +       h_path.mnt = au_sbr_mnt(sb, bindex);
18574 +       err = vfsub_mkdir(au_pinned_h_dir(&a->pin), &h_path, mode);
18575 +       if (unlikely(err))
18576 +               goto out_unpin;
18577 +
18578 +       /* make the dir opaque */
18579 +       diropq = 0;
18580 +       h_inode = d_inode(h_path.dentry);
18581 +       if (wh_dentry
18582 +           || au_opt_test(au_mntflags(sb), ALWAYS_DIROPQ)) {
18583 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
18584 +               opq_dentry = au_diropq_create(dentry, bindex);
18585 +               inode_unlock(h_inode);
18586 +               err = PTR_ERR(opq_dentry);
18587 +               if (IS_ERR(opq_dentry))
18588 +                       goto out_dir;
18589 +               dput(opq_dentry);
18590 +               diropq = 1;
18591 +       }
18592 +
18593 +       err = epilog(dir, bindex, wh_dentry, dentry);
18594 +       if (!err) {
18595 +               inc_nlink(dir);
18596 +               goto out_unpin; /* success */
18597 +       }
18598 +
18599 +       /* revert */
18600 +       if (diropq) {
18601 +               AuLabel(revert opq);
18602 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
18603 +               rerr = au_diropq_remove(dentry, bindex);
18604 +               inode_unlock(h_inode);
18605 +               if (rerr) {
18606 +                       AuIOErr("%pd reverting diropq failed(%d, %d)\n",
18607 +                               dentry, err, rerr);
18608 +                       err = -EIO;
18609 +               }
18610 +       }
18611 +
18612 +out_dir:
18613 +       AuLabel(revert dir);
18614 +       rerr = vfsub_rmdir(au_pinned_h_dir(&a->pin), &h_path);
18615 +       if (rerr) {
18616 +               AuIOErr("%pd reverting dir failed(%d, %d)\n",
18617 +                       dentry, err, rerr);
18618 +               err = -EIO;
18619 +       }
18620 +       au_dtime_revert(&a->dt);
18621 +out_unpin:
18622 +       au_unpin(&a->pin);
18623 +       dput(wh_dentry);
18624 +out_parent:
18625 +       di_write_unlock(parent);
18626 +out_unlock:
18627 +       if (unlikely(err)) {
18628 +               au_update_dbtop(dentry);
18629 +               d_drop(dentry);
18630 +       }
18631 +       aufs_read_unlock(dentry, AuLock_DW);
18632 +out_free:
18633 +       au_delayed_kfree(a);
18634 +out:
18635 +       return err;
18636 +}
18637 diff -urN /usr/share/empty/fs/aufs/i_op.c linux/fs/aufs/i_op.c
18638 --- /usr/share/empty/fs/aufs/i_op.c     1970-01-01 01:00:00.000000000 +0100
18639 +++ linux/fs/aufs/i_op.c        2017-05-06 22:16:52.778220093 +0200
18640 @@ -0,0 +1,1452 @@
18641 +/*
18642 + * Copyright (C) 2005-2017 Junjiro R. Okajima
18643 + *
18644 + * This program, aufs is free software; you can redistribute it and/or modify
18645 + * it under the terms of the GNU General Public License as published by
18646 + * the Free Software Foundation; either version 2 of the License, or
18647 + * (at your option) any later version.
18648 + *
18649 + * This program is distributed in the hope that it will be useful,
18650 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18651 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18652 + * GNU General Public License for more details.
18653 + *
18654 + * You should have received a copy of the GNU General Public License
18655 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18656 + */
18657 +
18658 +/*
18659 + * inode operations (except add/del/rename)
18660 + */
18661 +
18662 +#include <linux/device_cgroup.h>
18663 +#include <linux/fs_stack.h>
18664 +#include <linux/namei.h>
18665 +#include <linux/security.h>
18666 +#include "aufs.h"
18667 +
18668 +static int h_permission(struct inode *h_inode, int mask,
18669 +                       struct path *h_path, int brperm)
18670 +{
18671 +       int err;
18672 +       const unsigned char write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18673 +
18674 +       err = -EPERM;
18675 +       if (write_mask && IS_IMMUTABLE(h_inode))
18676 +               goto out;
18677 +
18678 +       err = -EACCES;
18679 +       if (((mask & MAY_EXEC)
18680 +            && S_ISREG(h_inode->i_mode)
18681 +            && (path_noexec(h_path)
18682 +                || !(h_inode->i_mode & S_IXUGO))))
18683 +               goto out;
18684 +
18685 +       /*
18686 +        * - skip the lower fs test in the case of write to ro branch.
18687 +        * - nfs dir permission write check is optimized, but a policy for
18688 +        *   link/rename requires a real check.
18689 +        * - nfs always sets MS_POSIXACL regardless its mount option 'noacl.'
18690 +        *   in this case, generic_permission() returns -EOPNOTSUPP.
18691 +        */
18692 +       if ((write_mask && !au_br_writable(brperm))
18693 +           || (au_test_nfs(h_inode->i_sb) && S_ISDIR(h_inode->i_mode)
18694 +               && write_mask && !(mask & MAY_READ))
18695 +           || !h_inode->i_op->permission) {
18696 +               /* AuLabel(generic_permission); */
18697 +               /* AuDbg("get_acl %pf\n", h_inode->i_op->get_acl); */
18698 +               err = generic_permission(h_inode, mask);
18699 +               if (err == -EOPNOTSUPP && au_test_nfs_noacl(h_inode))
18700 +                       err = h_inode->i_op->permission(h_inode, mask);
18701 +               AuTraceErr(err);
18702 +       } else {
18703 +               /* AuLabel(h_inode->permission); */
18704 +               err = h_inode->i_op->permission(h_inode, mask);
18705 +               AuTraceErr(err);
18706 +       }
18707 +
18708 +       if (!err)
18709 +               err = devcgroup_inode_permission(h_inode, mask);
18710 +       if (!err)
18711 +               err = security_inode_permission(h_inode, mask);
18712 +
18713 +#if 0
18714 +       if (!err) {
18715 +               /* todo: do we need to call ima_path_check()? */
18716 +               struct path h_path = {
18717 +                       .dentry =
18718 +                       .mnt    = h_mnt
18719 +               };
18720 +               err = ima_path_check(&h_path,
18721 +                                    mask & (MAY_READ | MAY_WRITE | MAY_EXEC),
18722 +                                    IMA_COUNT_LEAVE);
18723 +       }
18724 +#endif
18725 +
18726 +out:
18727 +       return err;
18728 +}
18729 +
18730 +static int aufs_permission(struct inode *inode, int mask)
18731 +{
18732 +       int err;
18733 +       aufs_bindex_t bindex, bbot;
18734 +       const unsigned char isdir = !!S_ISDIR(inode->i_mode),
18735 +               write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18736 +       struct inode *h_inode;
18737 +       struct super_block *sb;
18738 +       struct au_branch *br;
18739 +
18740 +       /* todo: support rcu-walk? */
18741 +       if (mask & MAY_NOT_BLOCK)
18742 +               return -ECHILD;
18743 +
18744 +       sb = inode->i_sb;
18745 +       si_read_lock(sb, AuLock_FLUSH);
18746 +       ii_read_lock_child(inode);
18747 +#if 0
18748 +       err = au_iigen_test(inode, au_sigen(sb));
18749 +       if (unlikely(err))
18750 +               goto out;
18751 +#endif
18752 +
18753 +       if (!isdir
18754 +           || write_mask
18755 +           || au_opt_test(au_mntflags(sb), DIRPERM1)) {
18756 +               err = au_busy_or_stale();
18757 +               h_inode = au_h_iptr(inode, au_ibtop(inode));
18758 +               if (unlikely(!h_inode
18759 +                            || (h_inode->i_mode & S_IFMT)
18760 +                            != (inode->i_mode & S_IFMT)))
18761 +                       goto out;
18762 +
18763 +               err = 0;
18764 +               bindex = au_ibtop(inode);
18765 +               br = au_sbr(sb, bindex);
18766 +               err = h_permission(h_inode, mask, &br->br_path, br->br_perm);
18767 +               if (write_mask
18768 +                   && !err
18769 +                   && !special_file(h_inode->i_mode)) {
18770 +                       /* test whether the upper writable branch exists */
18771 +                       err = -EROFS;
18772 +                       for (; bindex >= 0; bindex--)
18773 +                               if (!au_br_rdonly(au_sbr(sb, bindex))) {
18774 +                                       err = 0;
18775 +                                       break;
18776 +                               }
18777 +               }
18778 +               goto out;
18779 +       }
18780 +
18781 +       /* non-write to dir */
18782 +       err = 0;
18783 +       bbot = au_ibbot(inode);
18784 +       for (bindex = au_ibtop(inode); !err && bindex <= bbot; bindex++) {
18785 +               h_inode = au_h_iptr(inode, bindex);
18786 +               if (h_inode) {
18787 +                       err = au_busy_or_stale();
18788 +                       if (unlikely(!S_ISDIR(h_inode->i_mode)))
18789 +                               break;
18790 +
18791 +                       br = au_sbr(sb, bindex);
18792 +                       err = h_permission(h_inode, mask, &br->br_path,
18793 +                                          br->br_perm);
18794 +               }
18795 +       }
18796 +
18797 +out:
18798 +       ii_read_unlock(inode);
18799 +       si_read_unlock(sb);
18800 +       return err;
18801 +}
18802 +
18803 +/* ---------------------------------------------------------------------- */
18804 +
18805 +static struct dentry *aufs_lookup(struct inode *dir, struct dentry *dentry,
18806 +                                 unsigned int flags)
18807 +{
18808 +       struct dentry *ret, *parent;
18809 +       struct inode *inode;
18810 +       struct super_block *sb;
18811 +       int err, npositive;
18812 +
18813 +       IMustLock(dir);
18814 +
18815 +       /* todo: support rcu-walk? */
18816 +       ret = ERR_PTR(-ECHILD);
18817 +       if (flags & LOOKUP_RCU)
18818 +               goto out;
18819 +
18820 +       ret = ERR_PTR(-ENAMETOOLONG);
18821 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
18822 +               goto out;
18823 +
18824 +       sb = dir->i_sb;
18825 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18826 +       ret = ERR_PTR(err);
18827 +       if (unlikely(err))
18828 +               goto out;
18829 +
18830 +       err = au_di_init(dentry);
18831 +       ret = ERR_PTR(err);
18832 +       if (unlikely(err))
18833 +               goto out_si;
18834 +
18835 +       inode = NULL;
18836 +       npositive = 0; /* suppress a warning */
18837 +       parent = dentry->d_parent; /* dir inode is locked */
18838 +       di_read_lock_parent(parent, AuLock_IR);
18839 +       err = au_alive_dir(parent);
18840 +       if (!err)
18841 +               err = au_digen_test(parent, au_sigen(sb));
18842 +       if (!err) {
18843 +               /* regardless LOOKUP_CREATE, always ALLOW_NEG */
18844 +               npositive = au_lkup_dentry(dentry, au_dbtop(parent),
18845 +                                          AuLkup_ALLOW_NEG);
18846 +               err = npositive;
18847 +       }
18848 +       di_read_unlock(parent, AuLock_IR);
18849 +       ret = ERR_PTR(err);
18850 +       if (unlikely(err < 0))
18851 +               goto out_unlock;
18852 +
18853 +       if (npositive) {
18854 +               inode = au_new_inode(dentry, /*must_new*/0);
18855 +               if (IS_ERR(inode)) {
18856 +                       ret = (void *)inode;
18857 +                       inode = NULL;
18858 +                       goto out_unlock;
18859 +               }
18860 +       }
18861 +
18862 +       if (inode)
18863 +               atomic_inc(&inode->i_count);
18864 +       ret = d_splice_alias(inode, dentry);
18865 +#if 0
18866 +       if (unlikely(d_need_lookup(dentry))) {
18867 +               spin_lock(&dentry->d_lock);
18868 +               dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
18869 +               spin_unlock(&dentry->d_lock);
18870 +       } else
18871 +#endif
18872 +       if (inode) {
18873 +               if (!IS_ERR(ret)) {
18874 +                       iput(inode);
18875 +                       if (ret && ret != dentry)
18876 +                               ii_write_unlock(inode);
18877 +               } else {
18878 +                       ii_write_unlock(inode);
18879 +                       iput(inode);
18880 +                       inode = NULL;
18881 +               }
18882 +       }
18883 +
18884 +out_unlock:
18885 +       di_write_unlock(dentry);
18886 +out_si:
18887 +       si_read_unlock(sb);
18888 +out:
18889 +       return ret;
18890 +}
18891 +
18892 +/* ---------------------------------------------------------------------- */
18893 +
18894 +struct aopen_node {
18895 +       struct hlist_node hlist;
18896 +       struct file *file, *h_file;
18897 +};
18898 +
18899 +static int au_do_aopen(struct inode *inode, struct file *file)
18900 +{
18901 +       struct au_sphlhead *aopen;
18902 +       struct aopen_node *node;
18903 +       struct au_do_open_args args = {
18904 +               .no_lock        = 1,
18905 +               .open           = au_do_open_nondir
18906 +       };
18907 +
18908 +       aopen = &au_sbi(inode->i_sb)->si_aopen;
18909 +       spin_lock(&aopen->spin);
18910 +       hlist_for_each_entry(node, &aopen->head, hlist)
18911 +               if (node->file == file) {
18912 +                       args.h_file = node->h_file;
18913 +                       break;
18914 +               }
18915 +       spin_unlock(&aopen->spin);
18916 +       /* AuDebugOn(!args.h_file); */
18917 +
18918 +       return au_do_open(file, &args);
18919 +}
18920 +
18921 +static int aufs_atomic_open(struct inode *dir, struct dentry *dentry,
18922 +                           struct file *file, unsigned int open_flag,
18923 +                           umode_t create_mode, int *opened)
18924 +{
18925 +       int err, h_opened = *opened;
18926 +       unsigned int lkup_flags;
18927 +       struct dentry *parent, *d;
18928 +       struct au_sphlhead *aopen;
18929 +       struct vfsub_aopen_args args = {
18930 +               .open_flag      = open_flag,
18931 +               .create_mode    = create_mode,
18932 +               .opened         = &h_opened
18933 +       };
18934 +       struct aopen_node aopen_node = {
18935 +               .file   = file
18936 +       };
18937 +
18938 +       IMustLock(dir);
18939 +       AuDbg("open_flag 0%o\n", open_flag);
18940 +       AuDbgDentry(dentry);
18941 +
18942 +       err = 0;
18943 +       if (!au_di(dentry)) {
18944 +               lkup_flags = LOOKUP_OPEN;
18945 +               if (open_flag & O_CREAT)
18946 +                       lkup_flags |= LOOKUP_CREATE;
18947 +               d = aufs_lookup(dir, dentry, lkup_flags);
18948 +               if (IS_ERR(d)) {
18949 +                       err = PTR_ERR(d);
18950 +                       AuTraceErr(err);
18951 +                       goto out;
18952 +               } else if (d) {
18953 +                       /*
18954 +                        * obsoleted dentry found.
18955 +                        * another error will be returned later.
18956 +                        */
18957 +                       d_drop(d);
18958 +                       AuDbgDentry(d);
18959 +                       dput(d);
18960 +               }
18961 +               AuDbgDentry(dentry);
18962 +       }
18963 +
18964 +       if (d_is_positive(dentry)
18965 +           || d_unhashed(dentry)
18966 +           || d_unlinked(dentry)
18967 +           || !(open_flag & O_CREAT))
18968 +               goto out_no_open;
18969 +
18970 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
18971 +       if (unlikely(err))
18972 +               goto out;
18973 +
18974 +       parent = dentry->d_parent;      /* dir is locked */
18975 +       di_write_lock_parent(parent);
18976 +       err = au_lkup_dentry(dentry, /*btop*/0, AuLkup_ALLOW_NEG);
18977 +       if (unlikely(err))
18978 +               goto out_unlock;
18979 +
18980 +       AuDbgDentry(dentry);
18981 +       if (d_is_positive(dentry))
18982 +               goto out_unlock;
18983 +
18984 +       args.file = get_empty_filp();
18985 +       err = PTR_ERR(args.file);
18986 +       if (IS_ERR(args.file))
18987 +               goto out_unlock;
18988 +
18989 +       args.file->f_flags = file->f_flags;
18990 +       err = au_aopen_or_create(dir, dentry, &args);
18991 +       AuTraceErr(err);
18992 +       AuDbgFile(args.file);
18993 +       if (unlikely(err < 0)) {
18994 +               if (h_opened & FILE_OPENED)
18995 +                       fput(args.file);
18996 +               else
18997 +                       put_filp(args.file);
18998 +               goto out_unlock;
18999 +       }
19000 +
19001 +       /* some filesystems don't set FILE_CREATED while succeeded? */
19002 +       *opened |= FILE_CREATED;
19003 +       if (h_opened & FILE_OPENED)
19004 +               aopen_node.h_file = args.file;
19005 +       else {
19006 +               put_filp(args.file);
19007 +               args.file = NULL;
19008 +       }
19009 +       aopen = &au_sbi(dir->i_sb)->si_aopen;
19010 +       au_sphl_add(&aopen_node.hlist, aopen);
19011 +       err = finish_open(file, dentry, au_do_aopen, opened);
19012 +       au_sphl_del(&aopen_node.hlist, aopen);
19013 +       AuTraceErr(err);
19014 +       AuDbgFile(file);
19015 +       if (aopen_node.h_file)
19016 +               fput(aopen_node.h_file);
19017 +
19018 +out_unlock:
19019 +       di_write_unlock(parent);
19020 +       aufs_read_unlock(dentry, AuLock_DW);
19021 +       AuDbgDentry(dentry);
19022 +       if (unlikely(err < 0))
19023 +               goto out;
19024 +out_no_open:
19025 +       if (err >= 0 && !(*opened & FILE_CREATED)) {
19026 +               AuLabel(out_no_open);
19027 +               dget(dentry);
19028 +               err = finish_no_open(file, dentry);
19029 +       }
19030 +out:
19031 +       AuDbg("%pd%s%s\n", dentry,
19032 +             (*opened & FILE_CREATED) ? " created" : "",
19033 +             (*opened & FILE_OPENED) ? " opened" : "");
19034 +       AuTraceErr(err);
19035 +       return err;
19036 +}
19037 +
19038 +
19039 +/* ---------------------------------------------------------------------- */
19040 +
19041 +static int au_wr_dir_cpup(struct dentry *dentry, struct dentry *parent,
19042 +                         const unsigned char add_entry, aufs_bindex_t bcpup,
19043 +                         aufs_bindex_t btop)
19044 +{
19045 +       int err;
19046 +       struct dentry *h_parent;
19047 +       struct inode *h_dir;
19048 +
19049 +       if (add_entry)
19050 +               IMustLock(d_inode(parent));
19051 +       else
19052 +               di_write_lock_parent(parent);
19053 +
19054 +       err = 0;
19055 +       if (!au_h_dptr(parent, bcpup)) {
19056 +               if (btop > bcpup)
19057 +                       err = au_cpup_dirs(dentry, bcpup);
19058 +               else if (btop < bcpup)
19059 +                       err = au_cpdown_dirs(dentry, bcpup);
19060 +               else
19061 +                       BUG();
19062 +       }
19063 +       if (!err && add_entry && !au_ftest_wrdir(add_entry, TMPFILE)) {
19064 +               h_parent = au_h_dptr(parent, bcpup);
19065 +               h_dir = d_inode(h_parent);
19066 +               inode_lock_nested(h_dir, AuLsc_I_PARENT);
19067 +               err = au_lkup_neg(dentry, bcpup, /*wh*/0);
19068 +               /* todo: no unlock here */
19069 +               inode_unlock(h_dir);
19070 +
19071 +               AuDbg("bcpup %d\n", bcpup);
19072 +               if (!err) {
19073 +                       if (d_really_is_negative(dentry))
19074 +                               au_set_h_dptr(dentry, btop, NULL);
19075 +                       au_update_dbrange(dentry, /*do_put_zero*/0);
19076 +               }
19077 +       }
19078 +
19079 +       if (!add_entry)
19080 +               di_write_unlock(parent);
19081 +       if (!err)
19082 +               err = bcpup; /* success */
19083 +
19084 +       AuTraceErr(err);
19085 +       return err;
19086 +}
19087 +
19088 +/*
19089 + * decide the branch and the parent dir where we will create a new entry.
19090 + * returns new bindex or an error.
19091 + * copyup the parent dir if needed.
19092 + */
19093 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
19094 +             struct au_wr_dir_args *args)
19095 +{
19096 +       int err;
19097 +       unsigned int flags;
19098 +       aufs_bindex_t bcpup, btop, src_btop;
19099 +       const unsigned char add_entry
19100 +               = au_ftest_wrdir(args->flags, ADD_ENTRY)
19101 +               | au_ftest_wrdir(args->flags, TMPFILE);
19102 +       struct super_block *sb;
19103 +       struct dentry *parent;
19104 +       struct au_sbinfo *sbinfo;
19105 +
19106 +       sb = dentry->d_sb;
19107 +       sbinfo = au_sbi(sb);
19108 +       parent = dget_parent(dentry);
19109 +       btop = au_dbtop(dentry);
19110 +       bcpup = btop;
19111 +       if (args->force_btgt < 0) {
19112 +               if (src_dentry) {
19113 +                       src_btop = au_dbtop(src_dentry);
19114 +                       if (src_btop < btop)
19115 +                               bcpup = src_btop;
19116 +               } else if (add_entry) {
19117 +                       flags = 0;
19118 +                       if (au_ftest_wrdir(args->flags, ISDIR))
19119 +                               au_fset_wbr(flags, DIR);
19120 +                       err = AuWbrCreate(sbinfo, dentry, flags);
19121 +                       bcpup = err;
19122 +               }
19123 +
19124 +               if (bcpup < 0 || au_test_ro(sb, bcpup, d_inode(dentry))) {
19125 +                       if (add_entry)
19126 +                               err = AuWbrCopyup(sbinfo, dentry);
19127 +                       else {
19128 +                               if (!IS_ROOT(dentry)) {
19129 +                                       di_read_lock_parent(parent, !AuLock_IR);
19130 +                                       err = AuWbrCopyup(sbinfo, dentry);
19131 +                                       di_read_unlock(parent, !AuLock_IR);
19132 +                               } else
19133 +                                       err = AuWbrCopyup(sbinfo, dentry);
19134 +                       }
19135 +                       bcpup = err;
19136 +                       if (unlikely(err < 0))
19137 +                               goto out;
19138 +               }
19139 +       } else {
19140 +               bcpup = args->force_btgt;
19141 +               AuDebugOn(au_test_ro(sb, bcpup, d_inode(dentry)));
19142 +       }
19143 +
19144 +       AuDbg("btop %d, bcpup %d\n", btop, bcpup);
19145 +       err = bcpup;
19146 +       if (bcpup == btop)
19147 +               goto out; /* success */
19148 +
19149 +       /* copyup the new parent into the branch we process */
19150 +       err = au_wr_dir_cpup(dentry, parent, add_entry, bcpup, btop);
19151 +       if (err >= 0) {
19152 +               if (d_really_is_negative(dentry)) {
19153 +                       au_set_h_dptr(dentry, btop, NULL);
19154 +                       au_set_dbtop(dentry, bcpup);
19155 +                       au_set_dbbot(dentry, bcpup);
19156 +               }
19157 +               AuDebugOn(add_entry
19158 +                         && !au_ftest_wrdir(args->flags, TMPFILE)
19159 +                         && !au_h_dptr(dentry, bcpup));
19160 +       }
19161 +
19162 +out:
19163 +       dput(parent);
19164 +       return err;
19165 +}
19166 +
19167 +/* ---------------------------------------------------------------------- */
19168 +
19169 +void au_pin_hdir_unlock(struct au_pin *p)
19170 +{
19171 +       if (p->hdir)
19172 +               au_hn_inode_unlock(p->hdir);
19173 +}
19174 +
19175 +int au_pin_hdir_lock(struct au_pin *p)
19176 +{
19177 +       int err;
19178 +
19179 +       err = 0;
19180 +       if (!p->hdir)
19181 +               goto out;
19182 +
19183 +       /* even if an error happens later, keep this lock */
19184 +       au_hn_inode_lock_nested(p->hdir, p->lsc_hi);
19185 +
19186 +       err = -EBUSY;
19187 +       if (unlikely(p->hdir->hi_inode != d_inode(p->h_parent)))
19188 +               goto out;
19189 +
19190 +       err = 0;
19191 +       if (p->h_dentry)
19192 +               err = au_h_verify(p->h_dentry, p->udba, p->hdir->hi_inode,
19193 +                                 p->h_parent, p->br);
19194 +
19195 +out:
19196 +       return err;
19197 +}
19198 +
19199 +int au_pin_hdir_relock(struct au_pin *p)
19200 +{
19201 +       int err, i;
19202 +       struct inode *h_i;
19203 +       struct dentry *h_d[] = {
19204 +               p->h_dentry,
19205 +               p->h_parent
19206 +       };
19207 +
19208 +       err = au_pin_hdir_lock(p);
19209 +       if (unlikely(err))
19210 +               goto out;
19211 +
19212 +       for (i = 0; !err && i < sizeof(h_d)/sizeof(*h_d); i++) {
19213 +               if (!h_d[i])
19214 +                       continue;
19215 +               if (d_is_positive(h_d[i])) {
19216 +                       h_i = d_inode(h_d[i]);
19217 +                       err = !h_i->i_nlink;
19218 +               }
19219 +       }
19220 +
19221 +out:
19222 +       return err;
19223 +}
19224 +
19225 +static void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task)
19226 +{
19227 +#if !defined(CONFIG_RWSEM_GENERIC_SPINLOCK) && defined(CONFIG_RWSEM_SPIN_ON_OWNER)
19228 +       p->hdir->hi_inode->i_rwsem.owner = task;
19229 +#endif
19230 +}
19231 +
19232 +void au_pin_hdir_acquire_nest(struct au_pin *p)
19233 +{
19234 +       if (p->hdir) {
19235 +               rwsem_acquire_nest(&p->hdir->hi_inode->i_rwsem.dep_map,
19236 +                                  p->lsc_hi, 0, NULL, _RET_IP_);
19237 +               au_pin_hdir_set_owner(p, current);
19238 +       }
19239 +}
19240 +
19241 +void au_pin_hdir_release(struct au_pin *p)
19242 +{
19243 +       if (p->hdir) {
19244 +               au_pin_hdir_set_owner(p, p->task);
19245 +               rwsem_release(&p->hdir->hi_inode->i_rwsem.dep_map, 1, _RET_IP_);
19246 +       }
19247 +}
19248 +
19249 +struct dentry *au_pinned_h_parent(struct au_pin *pin)
19250 +{
19251 +       if (pin && pin->parent)
19252 +               return au_h_dptr(pin->parent, pin->bindex);
19253 +       return NULL;
19254 +}
19255 +
19256 +void au_unpin(struct au_pin *p)
19257 +{
19258 +       if (p->hdir)
19259 +               au_pin_hdir_unlock(p);
19260 +       if (p->h_mnt && au_ftest_pin(p->flags, MNT_WRITE))
19261 +               vfsub_mnt_drop_write(p->h_mnt);
19262 +       if (!p->hdir)
19263 +               return;
19264 +
19265 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
19266 +               di_read_unlock(p->parent, AuLock_IR);
19267 +       iput(p->hdir->hi_inode);
19268 +       dput(p->parent);
19269 +       p->parent = NULL;
19270 +       p->hdir = NULL;
19271 +       p->h_mnt = NULL;
19272 +       /* do not clear p->task */
19273 +}
19274 +
19275 +int au_do_pin(struct au_pin *p)
19276 +{
19277 +       int err;
19278 +       struct super_block *sb;
19279 +       struct inode *h_dir;
19280 +
19281 +       err = 0;
19282 +       sb = p->dentry->d_sb;
19283 +       p->br = au_sbr(sb, p->bindex);
19284 +       if (IS_ROOT(p->dentry)) {
19285 +               if (au_ftest_pin(p->flags, MNT_WRITE)) {
19286 +                       p->h_mnt = au_br_mnt(p->br);
19287 +                       err = vfsub_mnt_want_write(p->h_mnt);
19288 +                       if (unlikely(err)) {
19289 +                               au_fclr_pin(p->flags, MNT_WRITE);
19290 +                               goto out_err;
19291 +                       }
19292 +               }
19293 +               goto out;
19294 +       }
19295 +
19296 +       p->h_dentry = NULL;
19297 +       if (p->bindex <= au_dbbot(p->dentry))
19298 +               p->h_dentry = au_h_dptr(p->dentry, p->bindex);
19299 +
19300 +       p->parent = dget_parent(p->dentry);
19301 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
19302 +               di_read_lock(p->parent, AuLock_IR, p->lsc_di);
19303 +
19304 +       h_dir = NULL;
19305 +       p->h_parent = au_h_dptr(p->parent, p->bindex);
19306 +       p->hdir = au_hi(d_inode(p->parent), p->bindex);
19307 +       if (p->hdir)
19308 +               h_dir = p->hdir->hi_inode;
19309 +
19310 +       /*
19311 +        * udba case, or
19312 +        * if DI_LOCKED is not set, then p->parent may be different
19313 +        * and h_parent can be NULL.
19314 +        */
19315 +       if (unlikely(!p->hdir || !h_dir || !p->h_parent)) {
19316 +               err = -EBUSY;
19317 +               if (!au_ftest_pin(p->flags, DI_LOCKED))
19318 +                       di_read_unlock(p->parent, AuLock_IR);
19319 +               dput(p->parent);
19320 +               p->parent = NULL;
19321 +               goto out_err;
19322 +       }
19323 +
19324 +       if (au_ftest_pin(p->flags, MNT_WRITE)) {
19325 +               p->h_mnt = au_br_mnt(p->br);
19326 +               err = vfsub_mnt_want_write(p->h_mnt);
19327 +               if (unlikely(err)) {
19328 +                       au_fclr_pin(p->flags, MNT_WRITE);
19329 +                       if (!au_ftest_pin(p->flags, DI_LOCKED))
19330 +                               di_read_unlock(p->parent, AuLock_IR);
19331 +                       dput(p->parent);
19332 +                       p->parent = NULL;
19333 +                       goto out_err;
19334 +               }
19335 +       }
19336 +
19337 +       au_igrab(h_dir);
19338 +       err = au_pin_hdir_lock(p);
19339 +       if (!err)
19340 +               goto out; /* success */
19341 +
19342 +       au_unpin(p);
19343 +
19344 +out_err:
19345 +       pr_err("err %d\n", err);
19346 +       err = au_busy_or_stale();
19347 +out:
19348 +       return err;
19349 +}
19350 +
19351 +void au_pin_init(struct au_pin *p, struct dentry *dentry,
19352 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
19353 +                unsigned int udba, unsigned char flags)
19354 +{
19355 +       p->dentry = dentry;
19356 +       p->udba = udba;
19357 +       p->lsc_di = lsc_di;
19358 +       p->lsc_hi = lsc_hi;
19359 +       p->flags = flags;
19360 +       p->bindex = bindex;
19361 +
19362 +       p->parent = NULL;
19363 +       p->hdir = NULL;
19364 +       p->h_mnt = NULL;
19365 +
19366 +       p->h_dentry = NULL;
19367 +       p->h_parent = NULL;
19368 +       p->br = NULL;
19369 +       p->task = current;
19370 +}
19371 +
19372 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
19373 +          unsigned int udba, unsigned char flags)
19374 +{
19375 +       au_pin_init(pin, dentry, bindex, AuLsc_DI_PARENT, AuLsc_I_PARENT2,
19376 +                   udba, flags);
19377 +       return au_do_pin(pin);
19378 +}
19379 +
19380 +/* ---------------------------------------------------------------------- */
19381 +
19382 +/*
19383 + * ->setattr() and ->getattr() are called in various cases.
19384 + * chmod, stat: dentry is revalidated.
19385 + * fchmod, fstat: file and dentry are not revalidated, additionally they may be
19386 + *               unhashed.
19387 + * for ->setattr(), ia->ia_file is passed from ftruncate only.
19388 + */
19389 +/* todo: consolidate with do_refresh() and simple_reval_dpath() */
19390 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen)
19391 +{
19392 +       int err;
19393 +       struct dentry *parent;
19394 +
19395 +       err = 0;
19396 +       if (au_digen_test(dentry, sigen)) {
19397 +               parent = dget_parent(dentry);
19398 +               di_read_lock_parent(parent, AuLock_IR);
19399 +               err = au_refresh_dentry(dentry, parent);
19400 +               di_read_unlock(parent, AuLock_IR);
19401 +               dput(parent);
19402 +       }
19403 +
19404 +       AuTraceErr(err);
19405 +       return err;
19406 +}
19407 +
19408 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
19409 +                    struct au_icpup_args *a)
19410 +{
19411 +       int err;
19412 +       loff_t sz;
19413 +       aufs_bindex_t btop, ibtop;
19414 +       struct dentry *hi_wh, *parent;
19415 +       struct inode *inode;
19416 +       struct au_wr_dir_args wr_dir_args = {
19417 +               .force_btgt     = -1,
19418 +               .flags          = 0
19419 +       };
19420 +
19421 +       if (d_is_dir(dentry))
19422 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
19423 +       /* plink or hi_wh() case */
19424 +       btop = au_dbtop(dentry);
19425 +       inode = d_inode(dentry);
19426 +       ibtop = au_ibtop(inode);
19427 +       if (btop != ibtop && !au_test_ro(inode->i_sb, ibtop, inode))
19428 +               wr_dir_args.force_btgt = ibtop;
19429 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
19430 +       if (unlikely(err < 0))
19431 +               goto out;
19432 +       a->btgt = err;
19433 +       if (err != btop)
19434 +               au_fset_icpup(a->flags, DID_CPUP);
19435 +
19436 +       err = 0;
19437 +       a->pin_flags = AuPin_MNT_WRITE;
19438 +       parent = NULL;
19439 +       if (!IS_ROOT(dentry)) {
19440 +               au_fset_pin(a->pin_flags, DI_LOCKED);
19441 +               parent = dget_parent(dentry);
19442 +               di_write_lock_parent(parent);
19443 +       }
19444 +
19445 +       err = au_pin(&a->pin, dentry, a->btgt, a->udba, a->pin_flags);
19446 +       if (unlikely(err))
19447 +               goto out_parent;
19448 +
19449 +       sz = -1;
19450 +       a->h_path.dentry = au_h_dptr(dentry, btop);
19451 +       a->h_inode = d_inode(a->h_path.dentry);
19452 +       if (ia && (ia->ia_valid & ATTR_SIZE)) {
19453 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
19454 +               if (ia->ia_size < i_size_read(a->h_inode))
19455 +                       sz = ia->ia_size;
19456 +               inode_unlock(a->h_inode);
19457 +       }
19458 +
19459 +       hi_wh = NULL;
19460 +       if (au_ftest_icpup(a->flags, DID_CPUP) && d_unlinked(dentry)) {
19461 +               hi_wh = au_hi_wh(inode, a->btgt);
19462 +               if (!hi_wh) {
19463 +                       struct au_cp_generic cpg = {
19464 +                               .dentry = dentry,
19465 +                               .bdst   = a->btgt,
19466 +                               .bsrc   = -1,
19467 +                               .len    = sz,
19468 +                               .pin    = &a->pin
19469 +                       };
19470 +                       err = au_sio_cpup_wh(&cpg, /*file*/NULL);
19471 +                       if (unlikely(err))
19472 +                               goto out_unlock;
19473 +                       hi_wh = au_hi_wh(inode, a->btgt);
19474 +                       /* todo: revalidate hi_wh? */
19475 +               }
19476 +       }
19477 +
19478 +       if (parent) {
19479 +               au_pin_set_parent_lflag(&a->pin, /*lflag*/0);
19480 +               di_downgrade_lock(parent, AuLock_IR);
19481 +               dput(parent);
19482 +               parent = NULL;
19483 +       }
19484 +       if (!au_ftest_icpup(a->flags, DID_CPUP))
19485 +               goto out; /* success */
19486 +
19487 +       if (!d_unhashed(dentry)) {
19488 +               struct au_cp_generic cpg = {
19489 +                       .dentry = dentry,
19490 +                       .bdst   = a->btgt,
19491 +                       .bsrc   = btop,
19492 +                       .len    = sz,
19493 +                       .pin    = &a->pin,
19494 +                       .flags  = AuCpup_DTIME | AuCpup_HOPEN
19495 +               };
19496 +               err = au_sio_cpup_simple(&cpg);
19497 +               if (!err)
19498 +                       a->h_path.dentry = au_h_dptr(dentry, a->btgt);
19499 +       } else if (!hi_wh)
19500 +               a->h_path.dentry = au_h_dptr(dentry, a->btgt);
19501 +       else
19502 +               a->h_path.dentry = hi_wh; /* do not dget here */
19503 +
19504 +out_unlock:
19505 +       a->h_inode = d_inode(a->h_path.dentry);
19506 +       if (!err)
19507 +               goto out; /* success */
19508 +       au_unpin(&a->pin);
19509 +out_parent:
19510 +       if (parent) {
19511 +               di_write_unlock(parent);
19512 +               dput(parent);
19513 +       }
19514 +out:
19515 +       if (!err)
19516 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
19517 +       return err;
19518 +}
19519 +
19520 +static int aufs_setattr(struct dentry *dentry, struct iattr *ia)
19521 +{
19522 +       int err;
19523 +       struct inode *inode, *delegated;
19524 +       struct super_block *sb;
19525 +       struct file *file;
19526 +       struct au_icpup_args *a;
19527 +
19528 +       inode = d_inode(dentry);
19529 +       IMustLock(inode);
19530 +
19531 +       err = setattr_prepare(dentry, ia);
19532 +       if (unlikely(err))
19533 +               goto out;
19534 +
19535 +       err = -ENOMEM;
19536 +       a = kzalloc(sizeof(*a), GFP_NOFS);
19537 +       if (unlikely(!a))
19538 +               goto out;
19539 +
19540 +       if (ia->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
19541 +               ia->ia_valid &= ~ATTR_MODE;
19542 +
19543 +       file = NULL;
19544 +       sb = dentry->d_sb;
19545 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19546 +       if (unlikely(err))
19547 +               goto out_kfree;
19548 +
19549 +       if (ia->ia_valid & ATTR_FILE) {
19550 +               /* currently ftruncate(2) only */
19551 +               AuDebugOn(!d_is_reg(dentry));
19552 +               file = ia->ia_file;
19553 +               err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1,
19554 +                                           /*fi_lsc*/0);
19555 +               if (unlikely(err))
19556 +                       goto out_si;
19557 +               ia->ia_file = au_hf_top(file);
19558 +               a->udba = AuOpt_UDBA_NONE;
19559 +       } else {
19560 +               /* fchmod() doesn't pass ia_file */
19561 +               a->udba = au_opt_udba(sb);
19562 +               di_write_lock_child(dentry);
19563 +               /* no d_unlinked(), to set UDBA_NONE for root */
19564 +               if (d_unhashed(dentry))
19565 +                       a->udba = AuOpt_UDBA_NONE;
19566 +               if (a->udba != AuOpt_UDBA_NONE) {
19567 +                       AuDebugOn(IS_ROOT(dentry));
19568 +                       err = au_reval_for_attr(dentry, au_sigen(sb));
19569 +                       if (unlikely(err))
19570 +                               goto out_dentry;
19571 +               }
19572 +       }
19573 +
19574 +       err = au_pin_and_icpup(dentry, ia, a);
19575 +       if (unlikely(err < 0))
19576 +               goto out_dentry;
19577 +       if (au_ftest_icpup(a->flags, DID_CPUP)) {
19578 +               ia->ia_file = NULL;
19579 +               ia->ia_valid &= ~ATTR_FILE;
19580 +       }
19581 +
19582 +       a->h_path.mnt = au_sbr_mnt(sb, a->btgt);
19583 +       if ((ia->ia_valid & (ATTR_MODE | ATTR_CTIME))
19584 +           == (ATTR_MODE | ATTR_CTIME)) {
19585 +               err = security_path_chmod(&a->h_path, ia->ia_mode);
19586 +               if (unlikely(err))
19587 +                       goto out_unlock;
19588 +       } else if ((ia->ia_valid & (ATTR_UID | ATTR_GID))
19589 +                  && (ia->ia_valid & ATTR_CTIME)) {
19590 +               err = security_path_chown(&a->h_path, ia->ia_uid, ia->ia_gid);
19591 +               if (unlikely(err))
19592 +                       goto out_unlock;
19593 +       }
19594 +
19595 +       if (ia->ia_valid & ATTR_SIZE) {
19596 +               struct file *f;
19597 +
19598 +               if (ia->ia_size < i_size_read(inode))
19599 +                       /* unmap only */
19600 +                       truncate_setsize(inode, ia->ia_size);
19601 +
19602 +               f = NULL;
19603 +               if (ia->ia_valid & ATTR_FILE)
19604 +                       f = ia->ia_file;
19605 +               inode_unlock(a->h_inode);
19606 +               err = vfsub_trunc(&a->h_path, ia->ia_size, ia->ia_valid, f);
19607 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
19608 +       } else {
19609 +               delegated = NULL;
19610 +               while (1) {
19611 +                       err = vfsub_notify_change(&a->h_path, ia, &delegated);
19612 +                       if (delegated) {
19613 +                               err = break_deleg_wait(&delegated);
19614 +                               if (!err)
19615 +                                       continue;
19616 +                       }
19617 +                       break;
19618 +               }
19619 +       }
19620 +       /*
19621 +        * regardless aufs 'acl' option setting.
19622 +        * why don't all acl-aware fs call this func from their ->setattr()?
19623 +        */
19624 +       if (!err && (ia->ia_valid & ATTR_MODE))
19625 +               err = vfsub_acl_chmod(a->h_inode, ia->ia_mode);
19626 +       if (!err)
19627 +               au_cpup_attr_changeable(inode);
19628 +
19629 +out_unlock:
19630 +       inode_unlock(a->h_inode);
19631 +       au_unpin(&a->pin);
19632 +       if (unlikely(err))
19633 +               au_update_dbtop(dentry);
19634 +out_dentry:
19635 +       di_write_unlock(dentry);
19636 +       if (file) {
19637 +               fi_write_unlock(file);
19638 +               ia->ia_file = file;
19639 +               ia->ia_valid |= ATTR_FILE;
19640 +       }
19641 +out_si:
19642 +       si_read_unlock(sb);
19643 +out_kfree:
19644 +       au_delayed_kfree(a);
19645 +out:
19646 +       AuTraceErr(err);
19647 +       return err;
19648 +}
19649 +
19650 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
19651 +static int au_h_path_to_set_attr(struct dentry *dentry,
19652 +                                struct au_icpup_args *a, struct path *h_path)
19653 +{
19654 +       int err;
19655 +       struct super_block *sb;
19656 +
19657 +       sb = dentry->d_sb;
19658 +       a->udba = au_opt_udba(sb);
19659 +       /* no d_unlinked(), to set UDBA_NONE for root */
19660 +       if (d_unhashed(dentry))
19661 +               a->udba = AuOpt_UDBA_NONE;
19662 +       if (a->udba != AuOpt_UDBA_NONE) {
19663 +               AuDebugOn(IS_ROOT(dentry));
19664 +               err = au_reval_for_attr(dentry, au_sigen(sb));
19665 +               if (unlikely(err))
19666 +                       goto out;
19667 +       }
19668 +       err = au_pin_and_icpup(dentry, /*ia*/NULL, a);
19669 +       if (unlikely(err < 0))
19670 +               goto out;
19671 +
19672 +       h_path->dentry = a->h_path.dentry;
19673 +       h_path->mnt = au_sbr_mnt(sb, a->btgt);
19674 +
19675 +out:
19676 +       return err;
19677 +}
19678 +
19679 +ssize_t au_sxattr(struct dentry *dentry, struct inode *inode,
19680 +                 struct au_sxattr *arg)
19681 +{
19682 +       int err;
19683 +       struct path h_path;
19684 +       struct super_block *sb;
19685 +       struct au_icpup_args *a;
19686 +       struct inode *h_inode;
19687 +
19688 +       IMustLock(inode);
19689 +
19690 +       err = -ENOMEM;
19691 +       a = kzalloc(sizeof(*a), GFP_NOFS);
19692 +       if (unlikely(!a))
19693 +               goto out;
19694 +
19695 +       sb = dentry->d_sb;
19696 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19697 +       if (unlikely(err))
19698 +               goto out_kfree;
19699 +
19700 +       h_path.dentry = NULL;   /* silence gcc */
19701 +       di_write_lock_child(dentry);
19702 +       err = au_h_path_to_set_attr(dentry, a, &h_path);
19703 +       if (unlikely(err))
19704 +               goto out_di;
19705 +
19706 +       inode_unlock(a->h_inode);
19707 +       switch (arg->type) {
19708 +       case AU_XATTR_SET:
19709 +               AuDebugOn(d_is_negative(h_path.dentry));
19710 +               err = vfsub_setxattr(h_path.dentry,
19711 +                                    arg->u.set.name, arg->u.set.value,
19712 +                                    arg->u.set.size, arg->u.set.flags);
19713 +               break;
19714 +       case AU_ACL_SET:
19715 +               err = -EOPNOTSUPP;
19716 +               h_inode = d_inode(h_path.dentry);
19717 +               if (h_inode->i_op->set_acl)
19718 +                       /* this will call posix_acl_update_mode */
19719 +                       err = h_inode->i_op->set_acl(h_inode,
19720 +                                                    arg->u.acl_set.acl,
19721 +                                                    arg->u.acl_set.type);
19722 +               break;
19723 +       }
19724 +       if (!err)
19725 +               au_cpup_attr_timesizes(inode);
19726 +
19727 +       au_unpin(&a->pin);
19728 +       if (unlikely(err))
19729 +               au_update_dbtop(dentry);
19730 +
19731 +out_di:
19732 +       di_write_unlock(dentry);
19733 +       si_read_unlock(sb);
19734 +out_kfree:
19735 +       au_delayed_kfree(a);
19736 +out:
19737 +       AuTraceErr(err);
19738 +       return err;
19739 +}
19740 +#endif
19741 +
19742 +static void au_refresh_iattr(struct inode *inode, struct kstat *st,
19743 +                            unsigned int nlink)
19744 +{
19745 +       unsigned int n;
19746 +
19747 +       inode->i_mode = st->mode;
19748 +       /* don't i_[ug]id_write() here */
19749 +       inode->i_uid = st->uid;
19750 +       inode->i_gid = st->gid;
19751 +       inode->i_atime = st->atime;
19752 +       inode->i_mtime = st->mtime;
19753 +       inode->i_ctime = st->ctime;
19754 +
19755 +       au_cpup_attr_nlink(inode, /*force*/0);
19756 +       if (S_ISDIR(inode->i_mode)) {
19757 +               n = inode->i_nlink;
19758 +               n -= nlink;
19759 +               n += st->nlink;
19760 +               smp_mb(); /* for i_nlink */
19761 +               /* 0 can happen */
19762 +               set_nlink(inode, n);
19763 +       }
19764 +
19765 +       spin_lock(&inode->i_lock);
19766 +       inode->i_blocks = st->blocks;
19767 +       i_size_write(inode, st->size);
19768 +       spin_unlock(&inode->i_lock);
19769 +}
19770 +
19771 +/*
19772 + * common routine for aufs_getattr() and au_getxattr().
19773 + * returns zero or negative (an error).
19774 + * @dentry will be read-locked in success.
19775 + */
19776 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path,
19777 +                     int locked)
19778 +{
19779 +       int err;
19780 +       unsigned int mnt_flags, sigen;
19781 +       unsigned char udba_none;
19782 +       aufs_bindex_t bindex;
19783 +       struct super_block *sb, *h_sb;
19784 +       struct inode *inode;
19785 +
19786 +       h_path->mnt = NULL;
19787 +       h_path->dentry = NULL;
19788 +
19789 +       err = 0;
19790 +       sb = dentry->d_sb;
19791 +       mnt_flags = au_mntflags(sb);
19792 +       udba_none = !!au_opt_test(mnt_flags, UDBA_NONE);
19793 +
19794 +       if (unlikely(locked))
19795 +               goto body; /* skip locking dinfo */
19796 +
19797 +       /* support fstat(2) */
19798 +       if (!d_unlinked(dentry) && !udba_none) {
19799 +               sigen = au_sigen(sb);
19800 +               err = au_digen_test(dentry, sigen);
19801 +               if (!err) {
19802 +                       di_read_lock_child(dentry, AuLock_IR);
19803 +                       err = au_dbrange_test(dentry);
19804 +                       if (unlikely(err)) {
19805 +                               di_read_unlock(dentry, AuLock_IR);
19806 +                               goto out;
19807 +                       }
19808 +               } else {
19809 +                       AuDebugOn(IS_ROOT(dentry));
19810 +                       di_write_lock_child(dentry);
19811 +                       err = au_dbrange_test(dentry);
19812 +                       if (!err)
19813 +                               err = au_reval_for_attr(dentry, sigen);
19814 +                       if (!err)
19815 +                               di_downgrade_lock(dentry, AuLock_IR);
19816 +                       else {
19817 +                               di_write_unlock(dentry);
19818 +                               goto out;
19819 +                       }
19820 +               }
19821 +       } else
19822 +               di_read_lock_child(dentry, AuLock_IR);
19823 +
19824 +body:
19825 +       inode = d_inode(dentry);
19826 +       bindex = au_ibtop(inode);
19827 +       h_path->mnt = au_sbr_mnt(sb, bindex);
19828 +       h_sb = h_path->mnt->mnt_sb;
19829 +       if (!force
19830 +           && !au_test_fs_bad_iattr(h_sb)
19831 +           && udba_none)
19832 +               goto out; /* success */
19833 +
19834 +       if (au_dbtop(dentry) == bindex)
19835 +               h_path->dentry = au_h_dptr(dentry, bindex);
19836 +       else if (au_opt_test(mnt_flags, PLINK) && au_plink_test(inode)) {
19837 +               h_path->dentry = au_plink_lkup(inode, bindex);
19838 +               if (IS_ERR(h_path->dentry))
19839 +                       /* pretending success */
19840 +                       h_path->dentry = NULL;
19841 +               else
19842 +                       dput(h_path->dentry);
19843 +       }
19844 +
19845 +out:
19846 +       return err;
19847 +}
19848 +
19849 +static int aufs_getattr(const struct path *path, struct kstat *st,
19850 +                       u32 request, unsigned int query)
19851 +{
19852 +       int err;
19853 +       unsigned char positive;
19854 +       struct path h_path;
19855 +       struct dentry *dentry;
19856 +       struct inode *inode;
19857 +       struct super_block *sb;
19858 +
19859 +       dentry = path->dentry;
19860 +       inode = d_inode(dentry);
19861 +       sb = dentry->d_sb;
19862 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19863 +       if (unlikely(err))
19864 +               goto out;
19865 +       err = au_h_path_getattr(dentry, /*force*/0, &h_path, /*locked*/0);
19866 +       if (unlikely(err))
19867 +               goto out_si;
19868 +       if (unlikely(!h_path.dentry))
19869 +               /* illegally overlapped or something */
19870 +               goto out_fill; /* pretending success */
19871 +
19872 +       positive = d_is_positive(h_path.dentry);
19873 +       if (positive)
19874 +               /* no vfsub version */
19875 +               err = vfs_getattr(&h_path, st, request, query);
19876 +       if (!err) {
19877 +               if (positive)
19878 +                       au_refresh_iattr(inode, st,
19879 +                                        d_inode(h_path.dentry)->i_nlink);
19880 +               goto out_fill; /* success */
19881 +       }
19882 +       AuTraceErr(err);
19883 +       goto out_di;
19884 +
19885 +out_fill:
19886 +       generic_fillattr(inode, st);
19887 +out_di:
19888 +       di_read_unlock(dentry, AuLock_IR);
19889 +out_si:
19890 +       si_read_unlock(sb);
19891 +out:
19892 +       AuTraceErr(err);
19893 +       return err;
19894 +}
19895 +
19896 +/* ---------------------------------------------------------------------- */
19897 +
19898 +static const char *aufs_get_link(struct dentry *dentry, struct inode *inode,
19899 +                                struct delayed_call *done)
19900 +{
19901 +       const char *ret;
19902 +       struct dentry *h_dentry;
19903 +       struct inode *h_inode;
19904 +       int err;
19905 +       aufs_bindex_t bindex;
19906 +
19907 +       ret = NULL; /* suppress a warning */
19908 +       err = -ECHILD;
19909 +       if (!dentry)
19910 +               goto out;
19911 +
19912 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
19913 +       if (unlikely(err))
19914 +               goto out;
19915 +
19916 +       err = au_d_hashed_positive(dentry);
19917 +       if (unlikely(err))
19918 +               goto out_unlock;
19919 +
19920 +       err = -EINVAL;
19921 +       inode = d_inode(dentry);
19922 +       bindex = au_ibtop(inode);
19923 +       h_inode = au_h_iptr(inode, bindex);
19924 +       if (unlikely(!h_inode->i_op->get_link))
19925 +               goto out_unlock;
19926 +
19927 +       err = -EBUSY;
19928 +       h_dentry = NULL;
19929 +       if (au_dbtop(dentry) <= bindex) {
19930 +               h_dentry = au_h_dptr(dentry, bindex);
19931 +               if (h_dentry)
19932 +                       dget(h_dentry);
19933 +       }
19934 +       if (!h_dentry) {
19935 +               h_dentry = d_find_any_alias(h_inode);
19936 +               if (IS_ERR(h_dentry)) {
19937 +                       err = PTR_ERR(h_dentry);
19938 +                       goto out_unlock;
19939 +               }
19940 +       }
19941 +       if (unlikely(!h_dentry))
19942 +               goto out_unlock;
19943 +
19944 +       err = 0;
19945 +       AuDbg("%pf\n", h_inode->i_op->get_link);
19946 +       AuDbgDentry(h_dentry);
19947 +       ret = vfs_get_link(h_dentry, done);
19948 +       dput(h_dentry);
19949 +       if (IS_ERR(ret))
19950 +               err = PTR_ERR(ret);
19951 +
19952 +out_unlock:
19953 +       aufs_read_unlock(dentry, AuLock_IR);
19954 +out:
19955 +       if (unlikely(err))
19956 +               ret = ERR_PTR(err);
19957 +       AuTraceErrPtr(ret);
19958 +       return ret;
19959 +}
19960 +
19961 +/* ---------------------------------------------------------------------- */
19962 +
19963 +static int au_is_special(struct inode *inode)
19964 +{
19965 +       return (inode->i_mode & (S_IFBLK | S_IFCHR | S_IFIFO | S_IFSOCK));
19966 +}
19967 +
19968 +static int aufs_update_time(struct inode *inode, struct timespec *ts, int flags)
19969 +{
19970 +       int err;
19971 +       aufs_bindex_t bindex;
19972 +       struct super_block *sb;
19973 +       struct inode *h_inode;
19974 +       struct vfsmount *h_mnt;
19975 +
19976 +       sb = inode->i_sb;
19977 +       WARN_ONCE((flags & S_ATIME) && !IS_NOATIME(inode),
19978 +                 "unexpected s_flags 0x%lx", sb->s_flags);
19979 +
19980 +       /* mmap_sem might be acquired already, cf. aufs_mmap() */
19981 +       lockdep_off();
19982 +       si_read_lock(sb, AuLock_FLUSH);
19983 +       ii_write_lock_child(inode);
19984 +       lockdep_on();
19985 +
19986 +       err = 0;
19987 +       bindex = au_ibtop(inode);
19988 +       h_inode = au_h_iptr(inode, bindex);
19989 +       if (!au_test_ro(sb, bindex, inode)) {
19990 +               h_mnt = au_sbr_mnt(sb, bindex);
19991 +               err = vfsub_mnt_want_write(h_mnt);
19992 +               if (!err) {
19993 +                       err = vfsub_update_time(h_inode, ts, flags);
19994 +                       vfsub_mnt_drop_write(h_mnt);
19995 +               }
19996 +       } else if (au_is_special(h_inode)) {
19997 +               /*
19998 +                * Never copy-up here.
19999 +                * These special files may already be opened and used for
20000 +                * communicating. If we copied it up, then the communication
20001 +                * would be corrupted.
20002 +                */
20003 +               AuWarn1("timestamps for i%lu are ignored "
20004 +                       "since it is on readonly branch (hi%lu).\n",
20005 +                       inode->i_ino, h_inode->i_ino);
20006 +       } else if (flags & ~S_ATIME) {
20007 +               err = -EIO;
20008 +               AuIOErr1("unexpected flags 0x%x\n", flags);
20009 +               AuDebugOn(1);
20010 +       }
20011 +
20012 +       lockdep_off();
20013 +       if (!err)
20014 +               au_cpup_attr_timesizes(inode);
20015 +       ii_write_unlock(inode);
20016 +       si_read_unlock(sb);
20017 +       lockdep_on();
20018 +
20019 +       if (!err && (flags & S_VERSION))
20020 +               inode_inc_iversion(inode);
20021 +
20022 +       return err;
20023 +}
20024 +
20025 +/* ---------------------------------------------------------------------- */
20026 +
20027 +/* no getattr version will be set by module.c:aufs_init() */
20028 +struct inode_operations aufs_iop_nogetattr[AuIop_Last],
20029 +       aufs_iop[] = {
20030 +       [AuIop_SYMLINK] = {
20031 +               .permission     = aufs_permission,
20032 +#ifdef CONFIG_FS_POSIX_ACL
20033 +               .get_acl        = aufs_get_acl,
20034 +               .set_acl        = aufs_set_acl, /* unsupport for symlink? */
20035 +#endif
20036 +
20037 +               .setattr        = aufs_setattr,
20038 +               .getattr        = aufs_getattr,
20039 +
20040 +#ifdef CONFIG_AUFS_XATTR
20041 +               .listxattr      = aufs_listxattr,
20042 +#endif
20043 +
20044 +               .get_link       = aufs_get_link,
20045 +
20046 +               /* .update_time = aufs_update_time */
20047 +       },
20048 +       [AuIop_DIR] = {
20049 +               .create         = aufs_create,
20050 +               .lookup         = aufs_lookup,
20051 +               .link           = aufs_link,
20052 +               .unlink         = aufs_unlink,
20053 +               .symlink        = aufs_symlink,
20054 +               .mkdir          = aufs_mkdir,
20055 +               .rmdir          = aufs_rmdir,
20056 +               .mknod          = aufs_mknod,
20057 +               .rename         = aufs_rename,
20058 +
20059 +               .permission     = aufs_permission,
20060 +#ifdef CONFIG_FS_POSIX_ACL
20061 +               .get_acl        = aufs_get_acl,
20062 +               .set_acl        = aufs_set_acl,
20063 +#endif
20064 +
20065 +               .setattr        = aufs_setattr,
20066 +               .getattr        = aufs_getattr,
20067 +
20068 +#ifdef CONFIG_AUFS_XATTR
20069 +               .listxattr      = aufs_listxattr,
20070 +#endif
20071 +
20072 +               .update_time    = aufs_update_time,
20073 +               .atomic_open    = aufs_atomic_open,
20074 +               .tmpfile        = aufs_tmpfile
20075 +       },
20076 +       [AuIop_OTHER] = {
20077 +               .permission     = aufs_permission,
20078 +#ifdef CONFIG_FS_POSIX_ACL
20079 +               .get_acl        = aufs_get_acl,
20080 +               .set_acl        = aufs_set_acl,
20081 +#endif
20082 +
20083 +               .setattr        = aufs_setattr,
20084 +               .getattr        = aufs_getattr,
20085 +
20086 +#ifdef CONFIG_AUFS_XATTR
20087 +               .listxattr      = aufs_listxattr,
20088 +#endif
20089 +
20090 +               .update_time    = aufs_update_time
20091 +       }
20092 +};
20093 diff -urN /usr/share/empty/fs/aufs/i_op_del.c linux/fs/aufs/i_op_del.c
20094 --- /usr/share/empty/fs/aufs/i_op_del.c 1970-01-01 01:00:00.000000000 +0100
20095 +++ linux/fs/aufs/i_op_del.c    2017-05-06 22:16:52.778220093 +0200
20096 @@ -0,0 +1,511 @@
20097 +/*
20098 + * Copyright (C) 2005-2017 Junjiro R. Okajima
20099 + *
20100 + * This program, aufs is free software; you can redistribute it and/or modify
20101 + * it under the terms of the GNU General Public License as published by
20102 + * the Free Software Foundation; either version 2 of the License, or
20103 + * (at your option) any later version.
20104 + *
20105 + * This program is distributed in the hope that it will be useful,
20106 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
20107 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20108 + * GNU General Public License for more details.
20109 + *
20110 + * You should have received a copy of the GNU General Public License
20111 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20112 + */
20113 +
20114 +/*
20115 + * inode operations (del entry)
20116 + */
20117 +
20118 +#include "aufs.h"
20119 +
20120 +/*
20121 + * decide if a new whiteout for @dentry is necessary or not.
20122 + * when it is necessary, prepare the parent dir for the upper branch whose
20123 + * branch index is @bcpup for creation. the actual creation of the whiteout will
20124 + * be done by caller.
20125 + * return value:
20126 + * 0: wh is unnecessary
20127 + * plus: wh is necessary
20128 + * minus: error
20129 + */
20130 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup)
20131 +{
20132 +       int need_wh, err;
20133 +       aufs_bindex_t btop;
20134 +       struct super_block *sb;
20135 +
20136 +       sb = dentry->d_sb;
20137 +       btop = au_dbtop(dentry);
20138 +       if (*bcpup < 0) {
20139 +               *bcpup = btop;
20140 +               if (au_test_ro(sb, btop, d_inode(dentry))) {
20141 +                       err = AuWbrCopyup(au_sbi(sb), dentry);
20142 +                       *bcpup = err;
20143 +                       if (unlikely(err < 0))
20144 +                               goto out;
20145 +               }
20146 +       } else
20147 +               AuDebugOn(btop < *bcpup
20148 +                         || au_test_ro(sb, *bcpup, d_inode(dentry)));
20149 +       AuDbg("bcpup %d, btop %d\n", *bcpup, btop);
20150 +
20151 +       if (*bcpup != btop) {
20152 +               err = au_cpup_dirs(dentry, *bcpup);
20153 +               if (unlikely(err))
20154 +                       goto out;
20155 +               need_wh = 1;
20156 +       } else {
20157 +               struct au_dinfo *dinfo, *tmp;
20158 +
20159 +               need_wh = -ENOMEM;
20160 +               dinfo = au_di(dentry);
20161 +               tmp = au_di_alloc(sb, AuLsc_DI_TMP);
20162 +               if (tmp) {
20163 +                       au_di_cp(tmp, dinfo);
20164 +                       au_di_swap(tmp, dinfo);
20165 +                       /* returns the number of positive dentries */
20166 +                       need_wh = au_lkup_dentry(dentry, btop + 1,
20167 +                                                /* AuLkup_IGNORE_PERM */ 0);
20168 +                       au_di_swap(tmp, dinfo);
20169 +                       au_rw_write_unlock(&tmp->di_rwsem);
20170 +                       au_di_free(tmp);
20171 +               }
20172 +       }
20173 +       AuDbg("need_wh %d\n", need_wh);
20174 +       err = need_wh;
20175 +
20176 +out:
20177 +       return err;
20178 +}
20179 +
20180 +/*
20181 + * simple tests for the del-entry operations.
20182 + * following the checks in vfs, plus the parent-child relationship.
20183 + */
20184 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
20185 +              struct dentry *h_parent, int isdir)
20186 +{
20187 +       int err;
20188 +       umode_t h_mode;
20189 +       struct dentry *h_dentry, *h_latest;
20190 +       struct inode *h_inode;
20191 +
20192 +       h_dentry = au_h_dptr(dentry, bindex);
20193 +       if (d_really_is_positive(dentry)) {
20194 +               err = -ENOENT;
20195 +               if (unlikely(d_is_negative(h_dentry)))
20196 +                       goto out;
20197 +               h_inode = d_inode(h_dentry);
20198 +               if (unlikely(!h_inode->i_nlink))
20199 +                       goto out;
20200 +
20201 +               h_mode = h_inode->i_mode;
20202 +               if (!isdir) {
20203 +                       err = -EISDIR;
20204 +                       if (unlikely(S_ISDIR(h_mode)))
20205 +                               goto out;
20206 +               } else if (unlikely(!S_ISDIR(h_mode))) {
20207 +                       err = -ENOTDIR;
20208 +                       goto out;
20209 +               }
20210 +       } else {
20211 +               /* rename(2) case */
20212 +               err = -EIO;
20213 +               if (unlikely(d_is_positive(h_dentry)))
20214 +                       goto out;
20215 +       }
20216 +
20217 +       err = -ENOENT;
20218 +       /* expected parent dir is locked */
20219 +       if (unlikely(h_parent != h_dentry->d_parent))
20220 +               goto out;
20221 +       err = 0;
20222 +
20223 +       /*
20224 +        * rmdir a dir may break the consistency on some filesystem.
20225 +        * let's try heavy test.
20226 +        */
20227 +       err = -EACCES;
20228 +       if (unlikely(!au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1)
20229 +                    && au_test_h_perm(d_inode(h_parent),
20230 +                                      MAY_EXEC | MAY_WRITE)))
20231 +               goto out;
20232 +
20233 +       h_latest = au_sio_lkup_one(&dentry->d_name, h_parent);
20234 +       err = -EIO;
20235 +       if (IS_ERR(h_latest))
20236 +               goto out;
20237 +       if (h_latest == h_dentry)
20238 +               err = 0;
20239 +       dput(h_latest);
20240 +
20241 +out:
20242 +       return err;
20243 +}
20244 +
20245 +/*
20246 + * decide the branch where we operate for @dentry. the branch index will be set
20247 + * @rbcpup. after diciding it, 'pin' it and store the timestamps of the parent
20248 + * dir for reverting.
20249 + * when a new whiteout is necessary, create it.
20250 + */
20251 +static struct dentry*
20252 +lock_hdir_create_wh(struct dentry *dentry, int isdir, aufs_bindex_t *rbcpup,
20253 +                   struct au_dtime *dt, struct au_pin *pin)
20254 +{
20255 +       struct dentry *wh_dentry;
20256 +       struct super_block *sb;
20257 +       struct path h_path;
20258 +       int err, need_wh;
20259 +       unsigned int udba;
20260 +       aufs_bindex_t bcpup;
20261 +
20262 +       need_wh = au_wr_dir_need_wh(dentry, isdir, rbcpup);
20263 +       wh_dentry = ERR_PTR(need_wh);
20264 +       if (unlikely(need_wh < 0))
20265 +               goto out;
20266 +
20267 +       sb = dentry->d_sb;
20268 +       udba = au_opt_udba(sb);
20269 +       bcpup = *rbcpup;
20270 +       err = au_pin(pin, dentry, bcpup, udba,
20271 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
20272 +       wh_dentry = ERR_PTR(err);
20273 +       if (unlikely(err))
20274 +               goto out;
20275 +
20276 +       h_path.dentry = au_pinned_h_parent(pin);
20277 +       if (udba != AuOpt_UDBA_NONE
20278 +           && au_dbtop(dentry) == bcpup) {
20279 +               err = au_may_del(dentry, bcpup, h_path.dentry, isdir);
20280 +               wh_dentry = ERR_PTR(err);
20281 +               if (unlikely(err))
20282 +                       goto out_unpin;
20283 +       }
20284 +
20285 +       h_path.mnt = au_sbr_mnt(sb, bcpup);
20286 +       au_dtime_store(dt, au_pinned_parent(pin), &h_path);
20287 +       wh_dentry = NULL;
20288 +       if (!need_wh)
20289 +               goto out; /* success, no need to create whiteout */
20290 +
20291 +       wh_dentry = au_wh_create(dentry, bcpup, h_path.dentry);
20292 +       if (IS_ERR(wh_dentry))
20293 +               goto out_unpin;
20294 +
20295 +       /* returns with the parent is locked and wh_dentry is dget-ed */
20296 +       goto out; /* success */
20297 +
20298 +out_unpin:
20299 +       au_unpin(pin);
20300 +out:
20301 +       return wh_dentry;
20302 +}
20303 +
20304 +/*
20305 + * when removing a dir, rename it to a unique temporary whiteout-ed name first
20306 + * in order to be revertible and save time for removing many child whiteouts
20307 + * under the dir.
20308 + * returns 1 when there are too many child whiteout and caller should remove
20309 + * them asynchronously. returns 0 when the number of children is enough small to
20310 + * remove now or the branch fs is a remote fs.
20311 + * otherwise return an error.
20312 + */
20313 +static int renwh_and_rmdir(struct dentry *dentry, aufs_bindex_t bindex,
20314 +                          struct au_nhash *whlist, struct inode *dir)
20315 +{
20316 +       int rmdir_later, err, dirwh;
20317 +       struct dentry *h_dentry;
20318 +       struct super_block *sb;
20319 +       struct inode *inode;
20320 +
20321 +       sb = dentry->d_sb;
20322 +       SiMustAnyLock(sb);
20323 +       h_dentry = au_h_dptr(dentry, bindex);
20324 +       err = au_whtmp_ren(h_dentry, au_sbr(sb, bindex));
20325 +       if (unlikely(err))
20326 +               goto out;
20327 +
20328 +       /* stop monitoring */
20329 +       inode = d_inode(dentry);
20330 +       au_hn_free(au_hi(inode, bindex));
20331 +
20332 +       if (!au_test_fs_remote(h_dentry->d_sb)) {
20333 +               dirwh = au_sbi(sb)->si_dirwh;
20334 +               rmdir_later = (dirwh <= 1);
20335 +               if (!rmdir_later)
20336 +                       rmdir_later = au_nhash_test_longer_wh(whlist, bindex,
20337 +                                                             dirwh);
20338 +               if (rmdir_later)
20339 +                       return rmdir_later;
20340 +       }
20341 +
20342 +       err = au_whtmp_rmdir(dir, bindex, h_dentry, whlist);
20343 +       if (unlikely(err)) {
20344 +               AuIOErr("rmdir %pd, b%d failed, %d. ignored\n",
20345 +                       h_dentry, bindex, err);
20346 +               err = 0;
20347 +       }
20348 +
20349 +out:
20350 +       AuTraceErr(err);
20351 +       return err;
20352 +}
20353 +
20354 +/*
20355 + * final procedure for deleting a entry.
20356 + * maintain dentry and iattr.
20357 + */
20358 +static void epilog(struct inode *dir, struct dentry *dentry,
20359 +                  aufs_bindex_t bindex)
20360 +{
20361 +       struct inode *inode;
20362 +
20363 +       inode = d_inode(dentry);
20364 +       d_drop(dentry);
20365 +       inode->i_ctime = dir->i_ctime;
20366 +
20367 +       au_dir_ts(dir, bindex);
20368 +       dir->i_version++;
20369 +}
20370 +
20371 +/*
20372 + * when an error happened, remove the created whiteout and revert everything.
20373 + */
20374 +static int do_revert(int err, struct inode *dir, aufs_bindex_t bindex,
20375 +                    aufs_bindex_t bwh, struct dentry *wh_dentry,
20376 +                    struct dentry *dentry, struct au_dtime *dt)
20377 +{
20378 +       int rerr;
20379 +       struct path h_path = {
20380 +               .dentry = wh_dentry,
20381 +               .mnt    = au_sbr_mnt(dir->i_sb, bindex)
20382 +       };
20383 +
20384 +       rerr = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path, dentry);
20385 +       if (!rerr) {
20386 +               au_set_dbwh(dentry, bwh);
20387 +               au_dtime_revert(dt);
20388 +               return 0;
20389 +       }
20390 +
20391 +       AuIOErr("%pd reverting whiteout failed(%d, %d)\n", dentry, err, rerr);
20392 +       return -EIO;
20393 +}
20394 +
20395 +/* ---------------------------------------------------------------------- */
20396 +
20397 +int aufs_unlink(struct inode *dir, struct dentry *dentry)
20398 +{
20399 +       int err;
20400 +       aufs_bindex_t bwh, bindex, btop;
20401 +       struct inode *inode, *h_dir, *delegated;
20402 +       struct dentry *parent, *wh_dentry;
20403 +       /* to reuduce stack size */
20404 +       struct {
20405 +               struct au_dtime dt;
20406 +               struct au_pin pin;
20407 +               struct path h_path;
20408 +       } *a;
20409 +
20410 +       IMustLock(dir);
20411 +
20412 +       err = -ENOMEM;
20413 +       a = kmalloc(sizeof(*a), GFP_NOFS);
20414 +       if (unlikely(!a))
20415 +               goto out;
20416 +
20417 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
20418 +       if (unlikely(err))
20419 +               goto out_free;
20420 +       err = au_d_hashed_positive(dentry);
20421 +       if (unlikely(err))
20422 +               goto out_unlock;
20423 +       inode = d_inode(dentry);
20424 +       IMustLock(inode);
20425 +       err = -EISDIR;
20426 +       if (unlikely(d_is_dir(dentry)))
20427 +               goto out_unlock; /* possible? */
20428 +
20429 +       btop = au_dbtop(dentry);
20430 +       bwh = au_dbwh(dentry);
20431 +       bindex = -1;
20432 +       parent = dentry->d_parent; /* dir inode is locked */
20433 +       di_write_lock_parent(parent);
20434 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/0, &bindex, &a->dt,
20435 +                                       &a->pin);
20436 +       err = PTR_ERR(wh_dentry);
20437 +       if (IS_ERR(wh_dentry))
20438 +               goto out_parent;
20439 +
20440 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, btop);
20441 +       a->h_path.dentry = au_h_dptr(dentry, btop);
20442 +       dget(a->h_path.dentry);
20443 +       if (bindex == btop) {
20444 +               h_dir = au_pinned_h_dir(&a->pin);
20445 +               delegated = NULL;
20446 +               err = vfsub_unlink(h_dir, &a->h_path, &delegated, /*force*/0);
20447 +               if (unlikely(err == -EWOULDBLOCK)) {
20448 +                       pr_warn("cannot retry for NFSv4 delegation"
20449 +                               " for an internal unlink\n");
20450 +                       iput(delegated);
20451 +               }
20452 +       } else {
20453 +               /* dir inode is locked */
20454 +               h_dir = d_inode(wh_dentry->d_parent);
20455 +               IMustLock(h_dir);
20456 +               err = 0;
20457 +       }
20458 +
20459 +       if (!err) {
20460 +               vfsub_drop_nlink(inode);
20461 +               epilog(dir, dentry, bindex);
20462 +
20463 +               /* update target timestamps */
20464 +               if (bindex == btop) {
20465 +                       vfsub_update_h_iattr(&a->h_path, /*did*/NULL);
20466 +                       /*ignore*/
20467 +                       inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
20468 +               } else
20469 +                       /* todo: this timestamp may be reverted later */
20470 +                       inode->i_ctime = h_dir->i_ctime;
20471 +               goto out_unpin; /* success */
20472 +       }
20473 +
20474 +       /* revert */
20475 +       if (wh_dentry) {
20476 +               int rerr;
20477 +
20478 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
20479 +                                &a->dt);
20480 +               if (rerr)
20481 +                       err = rerr;
20482 +       }
20483 +
20484 +out_unpin:
20485 +       au_unpin(&a->pin);
20486 +       dput(wh_dentry);
20487 +       dput(a->h_path.dentry);
20488 +out_parent:
20489 +       di_write_unlock(parent);
20490 +out_unlock:
20491 +       aufs_read_unlock(dentry, AuLock_DW);
20492 +out_free:
20493 +       au_delayed_kfree(a);
20494 +out:
20495 +       return err;
20496 +}
20497 +
20498 +int aufs_rmdir(struct inode *dir, struct dentry *dentry)
20499 +{
20500 +       int err, rmdir_later;
20501 +       aufs_bindex_t bwh, bindex, btop;
20502 +       struct inode *inode;
20503 +       struct dentry *parent, *wh_dentry, *h_dentry;
20504 +       struct au_whtmp_rmdir *args;
20505 +       /* to reuduce stack size */
20506 +       struct {
20507 +               struct au_dtime dt;
20508 +               struct au_pin pin;
20509 +       } *a;
20510 +
20511 +       IMustLock(dir);
20512 +
20513 +       err = -ENOMEM;
20514 +       a = kmalloc(sizeof(*a), GFP_NOFS);
20515 +       if (unlikely(!a))
20516 +               goto out;
20517 +
20518 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
20519 +       if (unlikely(err))
20520 +               goto out_free;
20521 +       err = au_alive_dir(dentry);
20522 +       if (unlikely(err))
20523 +               goto out_unlock;
20524 +       inode = d_inode(dentry);
20525 +       IMustLock(inode);
20526 +       err = -ENOTDIR;
20527 +       if (unlikely(!d_is_dir(dentry)))
20528 +               goto out_unlock; /* possible? */
20529 +
20530 +       err = -ENOMEM;
20531 +       args = au_whtmp_rmdir_alloc(dir->i_sb, GFP_NOFS);
20532 +       if (unlikely(!args))
20533 +               goto out_unlock;
20534 +
20535 +       parent = dentry->d_parent; /* dir inode is locked */
20536 +       di_write_lock_parent(parent);
20537 +       err = au_test_empty(dentry, &args->whlist);
20538 +       if (unlikely(err))
20539 +               goto out_parent;
20540 +
20541 +       btop = au_dbtop(dentry);
20542 +       bwh = au_dbwh(dentry);
20543 +       bindex = -1;
20544 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/1, &bindex, &a->dt,
20545 +                                       &a->pin);
20546 +       err = PTR_ERR(wh_dentry);
20547 +       if (IS_ERR(wh_dentry))
20548 +               goto out_parent;
20549 +
20550 +       h_dentry = au_h_dptr(dentry, btop);
20551 +       dget(h_dentry);
20552 +       rmdir_later = 0;
20553 +       if (bindex == btop) {
20554 +               err = renwh_and_rmdir(dentry, btop, &args->whlist, dir);
20555 +               if (err > 0) {
20556 +                       rmdir_later = err;
20557 +                       err = 0;
20558 +               }
20559 +       } else {
20560 +               /* stop monitoring */
20561 +               au_hn_free(au_hi(inode, btop));
20562 +
20563 +               /* dir inode is locked */
20564 +               IMustLock(d_inode(wh_dentry->d_parent));
20565 +               err = 0;
20566 +       }
20567 +
20568 +       if (!err) {
20569 +               vfsub_dead_dir(inode);
20570 +               au_set_dbdiropq(dentry, -1);
20571 +               epilog(dir, dentry, bindex);
20572 +
20573 +               if (rmdir_later) {
20574 +                       au_whtmp_kick_rmdir(dir, btop, h_dentry, args);
20575 +                       args = NULL;
20576 +               }
20577 +
20578 +               goto out_unpin; /* success */
20579 +       }
20580 +
20581 +       /* revert */
20582 +       AuLabel(revert);
20583 +       if (wh_dentry) {
20584 +               int rerr;
20585 +
20586 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
20587 +                                &a->dt);
20588 +               if (rerr)
20589 +                       err = rerr;
20590 +       }
20591 +
20592 +out_unpin:
20593 +       au_unpin(&a->pin);
20594 +       dput(wh_dentry);
20595 +       dput(h_dentry);
20596 +out_parent:
20597 +       di_write_unlock(parent);
20598 +       if (args)
20599 +               au_whtmp_rmdir_free(args);
20600 +out_unlock:
20601 +       aufs_read_unlock(dentry, AuLock_DW);
20602 +out_free:
20603 +       au_delayed_kfree(a);
20604 +out:
20605 +       AuTraceErr(err);
20606 +       return err;
20607 +}
20608 diff -urN /usr/share/empty/fs/aufs/i_op_ren.c linux/fs/aufs/i_op_ren.c
20609 --- /usr/share/empty/fs/aufs/i_op_ren.c 1970-01-01 01:00:00.000000000 +0100
20610 +++ linux/fs/aufs/i_op_ren.c    2017-05-06 22:16:52.778220093 +0200
20611 @@ -0,0 +1,1165 @@
20612 +/*
20613 + * Copyright (C) 2005-2017 Junjiro R. Okajima
20614 + *
20615 + * This program, aufs is free software; you can redistribute it and/or modify
20616 + * it under the terms of the GNU General Public License as published by
20617 + * the Free Software Foundation; either version 2 of the License, or
20618 + * (at your option) any later version.
20619 + *
20620 + * This program is distributed in the hope that it will be useful,
20621 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
20622 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20623 + * GNU General Public License for more details.
20624 + *
20625 + * You should have received a copy of the GNU General Public License
20626 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20627 + */
20628 +
20629 +/*
20630 + * inode operation (rename entry)
20631 + * todo: this is crazy monster
20632 + */
20633 +
20634 +#include "aufs.h"
20635 +
20636 +enum { AuSRC, AuDST, AuSrcDst };
20637 +enum { AuPARENT, AuCHILD, AuParentChild };
20638 +
20639 +#define AuRen_ISDIR_SRC                1
20640 +#define AuRen_ISDIR_DST                (1 << 1)
20641 +#define AuRen_ISSAMEDIR                (1 << 2)
20642 +#define AuRen_WHSRC            (1 << 3)
20643 +#define AuRen_WHDST            (1 << 4)
20644 +#define AuRen_MNT_WRITE                (1 << 5)
20645 +#define AuRen_DT_DSTDIR                (1 << 6)
20646 +#define AuRen_DIROPQ_SRC       (1 << 7)
20647 +#define AuRen_DIROPQ_DST       (1 << 8)
20648 +#define au_ftest_ren(flags, name)      ((flags) & AuRen_##name)
20649 +#define au_fset_ren(flags, name) \
20650 +       do { (flags) |= AuRen_##name; } while (0)
20651 +#define au_fclr_ren(flags, name) \
20652 +       do { (flags) &= ~AuRen_##name; } while (0)
20653 +
20654 +struct au_ren_args {
20655 +       struct {
20656 +               struct dentry *dentry, *h_dentry, *parent, *h_parent,
20657 +                       *wh_dentry;
20658 +               struct inode *dir, *inode;
20659 +               struct au_hinode *hdir, *hinode;
20660 +               struct au_dtime dt[AuParentChild];
20661 +               aufs_bindex_t btop, bdiropq;
20662 +       } sd[AuSrcDst];
20663 +
20664 +#define src_dentry     sd[AuSRC].dentry
20665 +#define src_dir                sd[AuSRC].dir
20666 +#define src_inode      sd[AuSRC].inode
20667 +#define src_h_dentry   sd[AuSRC].h_dentry
20668 +#define src_parent     sd[AuSRC].parent
20669 +#define src_h_parent   sd[AuSRC].h_parent
20670 +#define src_wh_dentry  sd[AuSRC].wh_dentry
20671 +#define src_hdir       sd[AuSRC].hdir
20672 +#define src_hinode     sd[AuSRC].hinode
20673 +#define src_h_dir      sd[AuSRC].hdir->hi_inode
20674 +#define src_dt         sd[AuSRC].dt
20675 +#define src_btop       sd[AuSRC].btop
20676 +#define src_bdiropq    sd[AuSRC].bdiropq
20677 +
20678 +#define dst_dentry     sd[AuDST].dentry
20679 +#define dst_dir                sd[AuDST].dir
20680 +#define dst_inode      sd[AuDST].inode
20681 +#define dst_h_dentry   sd[AuDST].h_dentry
20682 +#define dst_parent     sd[AuDST].parent
20683 +#define dst_h_parent   sd[AuDST].h_parent
20684 +#define dst_wh_dentry  sd[AuDST].wh_dentry
20685 +#define dst_hdir       sd[AuDST].hdir
20686 +#define dst_hinode     sd[AuDST].hinode
20687 +#define dst_h_dir      sd[AuDST].hdir->hi_inode
20688 +#define dst_dt         sd[AuDST].dt
20689 +#define dst_btop       sd[AuDST].btop
20690 +#define dst_bdiropq    sd[AuDST].bdiropq
20691 +
20692 +       struct dentry *h_trap;
20693 +       struct au_branch *br;
20694 +       struct path h_path;
20695 +       struct au_nhash whlist;
20696 +       aufs_bindex_t btgt, src_bwh;
20697 +
20698 +       struct {
20699 +               unsigned short auren_flags;
20700 +               unsigned char flags;    /* syscall parameter */
20701 +               unsigned char exchange;
20702 +       } __packed;
20703 +
20704 +       struct au_whtmp_rmdir *thargs;
20705 +       struct dentry *h_dst;
20706 +};
20707 +
20708 +/* ---------------------------------------------------------------------- */
20709 +
20710 +/*
20711 + * functions for reverting.
20712 + * when an error happened in a single rename systemcall, we should revert
20713 + * everything as if nothing happened.
20714 + * we don't need to revert the copied-up/down the parent dir since they are
20715 + * harmless.
20716 + */
20717 +
20718 +#define RevertFailure(fmt, ...) do { \
20719 +       AuIOErr("revert failure: " fmt " (%d, %d)\n", \
20720 +               ##__VA_ARGS__, err, rerr); \
20721 +       err = -EIO; \
20722 +} while (0)
20723 +
20724 +static void au_ren_do_rev_diropq(int err, struct au_ren_args *a, int idx)
20725 +{
20726 +       int rerr;
20727 +       struct dentry *d;
20728 +#define src_or_dst(member) a->sd[idx].member
20729 +
20730 +       d = src_or_dst(dentry); /* {src,dst}_dentry */
20731 +       au_hn_inode_lock_nested(src_or_dst(hinode), AuLsc_I_CHILD);
20732 +       rerr = au_diropq_remove(d, a->btgt);
20733 +       au_hn_inode_unlock(src_or_dst(hinode));
20734 +       au_set_dbdiropq(d, src_or_dst(bdiropq));
20735 +       if (rerr)
20736 +               RevertFailure("remove diropq %pd", d);
20737 +
20738 +#undef src_or_dst_
20739 +}
20740 +
20741 +static void au_ren_rev_diropq(int err, struct au_ren_args *a)
20742 +{
20743 +       if (au_ftest_ren(a->auren_flags, DIROPQ_SRC))
20744 +               au_ren_do_rev_diropq(err, a, AuSRC);
20745 +       if (au_ftest_ren(a->auren_flags, DIROPQ_DST))
20746 +               au_ren_do_rev_diropq(err, a, AuDST);
20747 +}
20748 +
20749 +static void au_ren_rev_rename(int err, struct au_ren_args *a)
20750 +{
20751 +       int rerr;
20752 +       struct inode *delegated;
20753 +
20754 +       a->h_path.dentry = vfsub_lkup_one(&a->src_dentry->d_name,
20755 +                                         a->src_h_parent);
20756 +       rerr = PTR_ERR(a->h_path.dentry);
20757 +       if (IS_ERR(a->h_path.dentry)) {
20758 +               RevertFailure("lkup one %pd", a->src_dentry);
20759 +               return;
20760 +       }
20761 +
20762 +       delegated = NULL;
20763 +       rerr = vfsub_rename(a->dst_h_dir,
20764 +                           au_h_dptr(a->src_dentry, a->btgt),
20765 +                           a->src_h_dir, &a->h_path, &delegated, a->flags);
20766 +       if (unlikely(rerr == -EWOULDBLOCK)) {
20767 +               pr_warn("cannot retry for NFSv4 delegation"
20768 +                       " for an internal rename\n");
20769 +               iput(delegated);
20770 +       }
20771 +       d_drop(a->h_path.dentry);
20772 +       dput(a->h_path.dentry);
20773 +       /* au_set_h_dptr(a->src_dentry, a->btgt, NULL); */
20774 +       if (rerr)
20775 +               RevertFailure("rename %pd", a->src_dentry);
20776 +}
20777 +
20778 +static void au_ren_rev_whtmp(int err, struct au_ren_args *a)
20779 +{
20780 +       int rerr;
20781 +       struct inode *delegated;
20782 +
20783 +       a->h_path.dentry = vfsub_lkup_one(&a->dst_dentry->d_name,
20784 +                                         a->dst_h_parent);
20785 +       rerr = PTR_ERR(a->h_path.dentry);
20786 +       if (IS_ERR(a->h_path.dentry)) {
20787 +               RevertFailure("lkup one %pd", a->dst_dentry);
20788 +               return;
20789 +       }
20790 +       if (d_is_positive(a->h_path.dentry)) {
20791 +               d_drop(a->h_path.dentry);
20792 +               dput(a->h_path.dentry);
20793 +               return;
20794 +       }
20795 +
20796 +       delegated = NULL;
20797 +       rerr = vfsub_rename(a->dst_h_dir, a->h_dst, a->dst_h_dir, &a->h_path,
20798 +                           &delegated, a->flags);
20799 +       if (unlikely(rerr == -EWOULDBLOCK)) {
20800 +               pr_warn("cannot retry for NFSv4 delegation"
20801 +                       " for an internal rename\n");
20802 +               iput(delegated);
20803 +       }
20804 +       d_drop(a->h_path.dentry);
20805 +       dput(a->h_path.dentry);
20806 +       if (!rerr)
20807 +               au_set_h_dptr(a->dst_dentry, a->btgt, dget(a->h_dst));
20808 +       else
20809 +               RevertFailure("rename %pd", a->h_dst);
20810 +}
20811 +
20812 +static void au_ren_rev_whsrc(int err, struct au_ren_args *a)
20813 +{
20814 +       int rerr;
20815 +
20816 +       a->h_path.dentry = a->src_wh_dentry;
20817 +       rerr = au_wh_unlink_dentry(a->src_h_dir, &a->h_path, a->src_dentry);
20818 +       au_set_dbwh(a->src_dentry, a->src_bwh);
20819 +       if (rerr)
20820 +               RevertFailure("unlink %pd", a->src_wh_dentry);
20821 +}
20822 +#undef RevertFailure
20823 +
20824 +/* ---------------------------------------------------------------------- */
20825 +
20826 +/*
20827 + * when we have to copyup the renaming entry, do it with the rename-target name
20828 + * in order to minimize the cost (the later actual rename is unnecessary).
20829 + * otherwise rename it on the target branch.
20830 + */
20831 +static int au_ren_or_cpup(struct au_ren_args *a)
20832 +{
20833 +       int err;
20834 +       struct dentry *d;
20835 +       struct inode *delegated;
20836 +
20837 +       d = a->src_dentry;
20838 +       if (au_dbtop(d) == a->btgt) {
20839 +               a->h_path.dentry = a->dst_h_dentry;
20840 +               AuDebugOn(au_dbtop(d) != a->btgt);
20841 +               delegated = NULL;
20842 +               err = vfsub_rename(a->src_h_dir, au_h_dptr(d, a->btgt),
20843 +                                  a->dst_h_dir, &a->h_path, &delegated,
20844 +                                  a->flags);
20845 +               if (unlikely(err == -EWOULDBLOCK)) {
20846 +                       pr_warn("cannot retry for NFSv4 delegation"
20847 +                               " for an internal rename\n");
20848 +                       iput(delegated);
20849 +               }
20850 +       } else
20851 +               BUG();
20852 +
20853 +       if (!err && a->h_dst)
20854 +               /* it will be set to dinfo later */
20855 +               dget(a->h_dst);
20856 +
20857 +       return err;
20858 +}
20859 +
20860 +/* cf. aufs_rmdir() */
20861 +static int au_ren_del_whtmp(struct au_ren_args *a)
20862 +{
20863 +       int err;
20864 +       struct inode *dir;
20865 +
20866 +       dir = a->dst_dir;
20867 +       SiMustAnyLock(dir->i_sb);
20868 +       if (!au_nhash_test_longer_wh(&a->whlist, a->btgt,
20869 +                                    au_sbi(dir->i_sb)->si_dirwh)
20870 +           || au_test_fs_remote(a->h_dst->d_sb)) {
20871 +               err = au_whtmp_rmdir(dir, a->btgt, a->h_dst, &a->whlist);
20872 +               if (unlikely(err))
20873 +                       pr_warn("failed removing whtmp dir %pd (%d), "
20874 +                               "ignored.\n", a->h_dst, err);
20875 +       } else {
20876 +               au_nhash_wh_free(&a->thargs->whlist);
20877 +               a->thargs->whlist = a->whlist;
20878 +               a->whlist.nh_num = 0;
20879 +               au_whtmp_kick_rmdir(dir, a->btgt, a->h_dst, a->thargs);
20880 +               dput(a->h_dst);
20881 +               a->thargs = NULL;
20882 +       }
20883 +
20884 +       return 0;
20885 +}
20886 +
20887 +/* make it 'opaque' dir. */
20888 +static int au_ren_do_diropq(struct au_ren_args *a, int idx)
20889 +{
20890 +       int err;
20891 +       struct dentry *d, *diropq;
20892 +#define src_or_dst(member) a->sd[idx].member
20893 +
20894 +       err = 0;
20895 +       d = src_or_dst(dentry); /* {src,dst}_dentry */
20896 +       src_or_dst(bdiropq) = au_dbdiropq(d);
20897 +       src_or_dst(hinode) = au_hi(src_or_dst(inode), a->btgt);
20898 +       au_hn_inode_lock_nested(src_or_dst(hinode), AuLsc_I_CHILD);
20899 +       diropq = au_diropq_create(d, a->btgt);
20900 +       au_hn_inode_unlock(src_or_dst(hinode));
20901 +       if (IS_ERR(diropq))
20902 +               err = PTR_ERR(diropq);
20903 +       else
20904 +               dput(diropq);
20905 +
20906 +#undef src_or_dst_
20907 +       return err;
20908 +}
20909 +
20910 +static int au_ren_diropq(struct au_ren_args *a)
20911 +{
20912 +       int err;
20913 +       unsigned char always;
20914 +       struct dentry *d;
20915 +
20916 +       err = 0;
20917 +       d = a->dst_dentry; /* already renamed on the branch */
20918 +       always = !!au_opt_test(au_mntflags(d->d_sb), ALWAYS_DIROPQ);
20919 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)
20920 +           && a->btgt != au_dbdiropq(a->src_dentry)
20921 +           && (a->dst_wh_dentry
20922 +               || a->btgt <= au_dbdiropq(d)
20923 +               /* hide the lower to keep xino */
20924 +               /* the lowers may not be a dir, but we hide them anyway */
20925 +               || a->btgt < au_dbbot(d)
20926 +               || always)) {
20927 +               AuDbg("here\n");
20928 +               err = au_ren_do_diropq(a, AuSRC);
20929 +               if (unlikely(err))
20930 +                       goto out;
20931 +               au_fset_ren(a->auren_flags, DIROPQ_SRC);
20932 +       }
20933 +       if (!a->exchange)
20934 +               goto out; /* success */
20935 +
20936 +       d = a->src_dentry; /* already renamed on the branch */
20937 +       if (au_ftest_ren(a->auren_flags, ISDIR_DST)
20938 +           && a->btgt != au_dbdiropq(a->dst_dentry)
20939 +           && (a->btgt < au_dbdiropq(d)
20940 +               || a->btgt < au_dbbot(d)
20941 +               || always)) {
20942 +               AuDbgDentry(a->src_dentry);
20943 +               AuDbgDentry(a->dst_dentry);
20944 +               err = au_ren_do_diropq(a, AuDST);
20945 +               if (unlikely(err))
20946 +                       goto out_rev_src;
20947 +               au_fset_ren(a->auren_flags, DIROPQ_DST);
20948 +       }
20949 +       goto out; /* success */
20950 +
20951 +out_rev_src:
20952 +       AuDbg("err %d, reverting src\n", err);
20953 +       au_ren_rev_diropq(err, a);
20954 +out:
20955 +       return err;
20956 +}
20957 +
20958 +static int do_rename(struct au_ren_args *a)
20959 +{
20960 +       int err;
20961 +       struct dentry *d, *h_d;
20962 +
20963 +       if (!a->exchange) {
20964 +               /* prepare workqueue args for asynchronous rmdir */
20965 +               h_d = a->dst_h_dentry;
20966 +               if (au_ftest_ren(a->auren_flags, ISDIR_DST)
20967 +                   && d_is_positive(h_d)) {
20968 +                       err = -ENOMEM;
20969 +                       a->thargs = au_whtmp_rmdir_alloc(a->src_dentry->d_sb,
20970 +                                                        GFP_NOFS);
20971 +                       if (unlikely(!a->thargs))
20972 +                               goto out;
20973 +                       a->h_dst = dget(h_d);
20974 +               }
20975 +
20976 +               /* create whiteout for src_dentry */
20977 +               if (au_ftest_ren(a->auren_flags, WHSRC)) {
20978 +                       a->src_bwh = au_dbwh(a->src_dentry);
20979 +                       AuDebugOn(a->src_bwh >= 0);
20980 +                       a->src_wh_dentry = au_wh_create(a->src_dentry, a->btgt,
20981 +                                                       a->src_h_parent);
20982 +                       err = PTR_ERR(a->src_wh_dentry);
20983 +                       if (IS_ERR(a->src_wh_dentry))
20984 +                               goto out_thargs;
20985 +               }
20986 +
20987 +               /* lookup whiteout for dentry */
20988 +               if (au_ftest_ren(a->auren_flags, WHDST)) {
20989 +                       h_d = au_wh_lkup(a->dst_h_parent,
20990 +                                        &a->dst_dentry->d_name, a->br);
20991 +                       err = PTR_ERR(h_d);
20992 +                       if (IS_ERR(h_d))
20993 +                               goto out_whsrc;
20994 +                       if (d_is_negative(h_d))
20995 +                               dput(h_d);
20996 +                       else
20997 +                               a->dst_wh_dentry = h_d;
20998 +               }
20999 +
21000 +               /* rename dentry to tmpwh */
21001 +               if (a->thargs) {
21002 +                       err = au_whtmp_ren(a->dst_h_dentry, a->br);
21003 +                       if (unlikely(err))
21004 +                               goto out_whdst;
21005 +
21006 +                       d = a->dst_dentry;
21007 +                       au_set_h_dptr(d, a->btgt, NULL);
21008 +                       err = au_lkup_neg(d, a->btgt, /*wh*/0);
21009 +                       if (unlikely(err))
21010 +                               goto out_whtmp;
21011 +                       a->dst_h_dentry = au_h_dptr(d, a->btgt);
21012 +               }
21013 +       }
21014 +
21015 +       BUG_ON(d_is_positive(a->dst_h_dentry) && a->src_btop != a->btgt);
21016 +
21017 +       /* rename by vfs_rename or cpup */
21018 +       err = au_ren_or_cpup(a);
21019 +       if (unlikely(err))
21020 +               /* leave the copied-up one */
21021 +               goto out_whtmp;
21022 +
21023 +       /* make dir opaque */
21024 +       err = au_ren_diropq(a);
21025 +       if (unlikely(err))
21026 +               goto out_rename;
21027 +
21028 +       /* update target timestamps */
21029 +       if (a->exchange) {
21030 +               AuDebugOn(au_dbtop(a->dst_dentry) != a->btgt);
21031 +               a->h_path.dentry = au_h_dptr(a->dst_dentry, a->btgt);
21032 +               vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
21033 +               a->dst_inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
21034 +       }
21035 +       AuDebugOn(au_dbtop(a->src_dentry) != a->btgt);
21036 +       a->h_path.dentry = au_h_dptr(a->src_dentry, a->btgt);
21037 +       vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
21038 +       a->src_inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
21039 +
21040 +       if (!a->exchange) {
21041 +               /* remove whiteout for dentry */
21042 +               if (a->dst_wh_dentry) {
21043 +                       a->h_path.dentry = a->dst_wh_dentry;
21044 +                       err = au_wh_unlink_dentry(a->dst_h_dir, &a->h_path,
21045 +                                                 a->dst_dentry);
21046 +                       if (unlikely(err))
21047 +                               goto out_diropq;
21048 +               }
21049 +
21050 +               /* remove whtmp */
21051 +               if (a->thargs)
21052 +                       au_ren_del_whtmp(a); /* ignore this error */
21053 +
21054 +               au_fhsm_wrote(a->src_dentry->d_sb, a->btgt, /*force*/0);
21055 +       }
21056 +       err = 0;
21057 +       goto out_success;
21058 +
21059 +out_diropq:
21060 +       au_ren_rev_diropq(err, a);
21061 +out_rename:
21062 +       au_ren_rev_rename(err, a);
21063 +       dput(a->h_dst);
21064 +out_whtmp:
21065 +       if (a->thargs)
21066 +               au_ren_rev_whtmp(err, a);
21067 +out_whdst:
21068 +       dput(a->dst_wh_dentry);
21069 +       a->dst_wh_dentry = NULL;
21070 +out_whsrc:
21071 +       if (a->src_wh_dentry)
21072 +               au_ren_rev_whsrc(err, a);
21073 +out_success:
21074 +       dput(a->src_wh_dentry);
21075 +       dput(a->dst_wh_dentry);
21076 +out_thargs:
21077 +       if (a->thargs) {
21078 +               dput(a->h_dst);
21079 +               au_whtmp_rmdir_free(a->thargs);
21080 +               a->thargs = NULL;
21081 +       }
21082 +out:
21083 +       return err;
21084 +}
21085 +
21086 +/* ---------------------------------------------------------------------- */
21087 +
21088 +/*
21089 + * test if @dentry dir can be rename destination or not.
21090 + * success means, it is a logically empty dir.
21091 + */
21092 +static int may_rename_dstdir(struct dentry *dentry, struct au_nhash *whlist)
21093 +{
21094 +       return au_test_empty(dentry, whlist);
21095 +}
21096 +
21097 +/*
21098 + * test if @dentry dir can be rename source or not.
21099 + * if it can, return 0 and @children is filled.
21100 + * success means,
21101 + * - it is a logically empty dir.
21102 + * - or, it exists on writable branch and has no children including whiteouts
21103 + *       on the lower branch.
21104 + */
21105 +static int may_rename_srcdir(struct dentry *dentry, aufs_bindex_t btgt)
21106 +{
21107 +       int err;
21108 +       unsigned int rdhash;
21109 +       aufs_bindex_t btop;
21110 +
21111 +       btop = au_dbtop(dentry);
21112 +       if (btop != btgt) {
21113 +               struct au_nhash whlist;
21114 +
21115 +               SiMustAnyLock(dentry->d_sb);
21116 +               rdhash = au_sbi(dentry->d_sb)->si_rdhash;
21117 +               if (!rdhash)
21118 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL,
21119 +                                                          dentry));
21120 +               err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
21121 +               if (unlikely(err))
21122 +                       goto out;
21123 +               err = au_test_empty(dentry, &whlist);
21124 +               au_nhash_wh_free(&whlist);
21125 +               goto out;
21126 +       }
21127 +
21128 +       if (btop == au_dbtaildir(dentry))
21129 +               return 0; /* success */
21130 +
21131 +       err = au_test_empty_lower(dentry);
21132 +
21133 +out:
21134 +       if (err == -ENOTEMPTY) {
21135 +               AuWarn1("renaming dir who has child(ren) on multiple branches,"
21136 +                       " is not supported\n");
21137 +               err = -EXDEV;
21138 +       }
21139 +       return err;
21140 +}
21141 +
21142 +/* side effect: sets whlist and h_dentry */
21143 +static int au_ren_may_dir(struct au_ren_args *a)
21144 +{
21145 +       int err;
21146 +       unsigned int rdhash;
21147 +       struct dentry *d;
21148 +
21149 +       d = a->dst_dentry;
21150 +       SiMustAnyLock(d->d_sb);
21151 +
21152 +       err = 0;
21153 +       if (au_ftest_ren(a->auren_flags, ISDIR_DST) && a->dst_inode) {
21154 +               rdhash = au_sbi(d->d_sb)->si_rdhash;
21155 +               if (!rdhash)
21156 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, d));
21157 +               err = au_nhash_alloc(&a->whlist, rdhash, GFP_NOFS);
21158 +               if (unlikely(err))
21159 +                       goto out;
21160 +
21161 +               if (!a->exchange) {
21162 +                       au_set_dbtop(d, a->dst_btop);
21163 +                       err = may_rename_dstdir(d, &a->whlist);
21164 +                       au_set_dbtop(d, a->btgt);
21165 +               } else
21166 +                       err = may_rename_srcdir(d, a->btgt);
21167 +       }
21168 +       a->dst_h_dentry = au_h_dptr(d, au_dbtop(d));
21169 +       if (unlikely(err))
21170 +               goto out;
21171 +
21172 +       d = a->src_dentry;
21173 +       a->src_h_dentry = au_h_dptr(d, au_dbtop(d));
21174 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)) {
21175 +               err = may_rename_srcdir(d, a->btgt);
21176 +               if (unlikely(err)) {
21177 +                       au_nhash_wh_free(&a->whlist);
21178 +                       a->whlist.nh_num = 0;
21179 +               }
21180 +       }
21181 +out:
21182 +       return err;
21183 +}
21184 +
21185 +/* ---------------------------------------------------------------------- */
21186 +
21187 +/*
21188 + * simple tests for rename.
21189 + * following the checks in vfs, plus the parent-child relationship.
21190 + */
21191 +static int au_may_ren(struct au_ren_args *a)
21192 +{
21193 +       int err, isdir;
21194 +       struct inode *h_inode;
21195 +
21196 +       if (a->src_btop == a->btgt) {
21197 +               err = au_may_del(a->src_dentry, a->btgt, a->src_h_parent,
21198 +                                au_ftest_ren(a->auren_flags, ISDIR_SRC));
21199 +               if (unlikely(err))
21200 +                       goto out;
21201 +               err = -EINVAL;
21202 +               if (unlikely(a->src_h_dentry == a->h_trap))
21203 +                       goto out;
21204 +       }
21205 +
21206 +       err = 0;
21207 +       if (a->dst_btop != a->btgt)
21208 +               goto out;
21209 +
21210 +       err = -ENOTEMPTY;
21211 +       if (unlikely(a->dst_h_dentry == a->h_trap))
21212 +               goto out;
21213 +
21214 +       err = -EIO;
21215 +       isdir = !!au_ftest_ren(a->auren_flags, ISDIR_DST);
21216 +       if (d_really_is_negative(a->dst_dentry)) {
21217 +               if (d_is_negative(a->dst_h_dentry))
21218 +                       err = au_may_add(a->dst_dentry, a->btgt,
21219 +                                        a->dst_h_parent, isdir);
21220 +       } else {
21221 +               if (unlikely(d_is_negative(a->dst_h_dentry)))
21222 +                       goto out;
21223 +               h_inode = d_inode(a->dst_h_dentry);
21224 +               if (h_inode->i_nlink)
21225 +                       err = au_may_del(a->dst_dentry, a->btgt,
21226 +                                        a->dst_h_parent, isdir);
21227 +       }
21228 +
21229 +out:
21230 +       if (unlikely(err == -ENOENT || err == -EEXIST))
21231 +               err = -EIO;
21232 +       AuTraceErr(err);
21233 +       return err;
21234 +}
21235 +
21236 +/* ---------------------------------------------------------------------- */
21237 +
21238 +/*
21239 + * locking order
21240 + * (VFS)
21241 + * - src_dir and dir by lock_rename()
21242 + * - inode if exitsts
21243 + * (aufs)
21244 + * - lock all
21245 + *   + src_dentry and dentry by aufs_read_and_write_lock2() which calls,
21246 + *     + si_read_lock
21247 + *     + di_write_lock2_child()
21248 + *       + di_write_lock_child()
21249 + *        + ii_write_lock_child()
21250 + *       + di_write_lock_child2()
21251 + *        + ii_write_lock_child2()
21252 + *     + src_parent and parent
21253 + *       + di_write_lock_parent()
21254 + *        + ii_write_lock_parent()
21255 + *       + di_write_lock_parent2()
21256 + *        + ii_write_lock_parent2()
21257 + *   + lower src_dir and dir by vfsub_lock_rename()
21258 + *   + verify the every relationships between child and parent. if any
21259 + *     of them failed, unlock all and return -EBUSY.
21260 + */
21261 +static void au_ren_unlock(struct au_ren_args *a)
21262 +{
21263 +       vfsub_unlock_rename(a->src_h_parent, a->src_hdir,
21264 +                           a->dst_h_parent, a->dst_hdir);
21265 +       if (au_ftest_ren(a->auren_flags, MNT_WRITE))
21266 +               vfsub_mnt_drop_write(au_br_mnt(a->br));
21267 +}
21268 +
21269 +static int au_ren_lock(struct au_ren_args *a)
21270 +{
21271 +       int err;
21272 +       unsigned int udba;
21273 +
21274 +       err = 0;
21275 +       a->src_h_parent = au_h_dptr(a->src_parent, a->btgt);
21276 +       a->src_hdir = au_hi(a->src_dir, a->btgt);
21277 +       a->dst_h_parent = au_h_dptr(a->dst_parent, a->btgt);
21278 +       a->dst_hdir = au_hi(a->dst_dir, a->btgt);
21279 +
21280 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
21281 +       if (unlikely(err))
21282 +               goto out;
21283 +       au_fset_ren(a->auren_flags, MNT_WRITE);
21284 +       a->h_trap = vfsub_lock_rename(a->src_h_parent, a->src_hdir,
21285 +                                     a->dst_h_parent, a->dst_hdir);
21286 +       udba = au_opt_udba(a->src_dentry->d_sb);
21287 +       if (unlikely(a->src_hdir->hi_inode != d_inode(a->src_h_parent)
21288 +                    || a->dst_hdir->hi_inode != d_inode(a->dst_h_parent)))
21289 +               err = au_busy_or_stale();
21290 +       if (!err && au_dbtop(a->src_dentry) == a->btgt)
21291 +               err = au_h_verify(a->src_h_dentry, udba,
21292 +                                 d_inode(a->src_h_parent), a->src_h_parent,
21293 +                                 a->br);
21294 +       if (!err && au_dbtop(a->dst_dentry) == a->btgt)
21295 +               err = au_h_verify(a->dst_h_dentry, udba,
21296 +                                 d_inode(a->dst_h_parent), a->dst_h_parent,
21297 +                                 a->br);
21298 +       if (!err)
21299 +               goto out; /* success */
21300 +
21301 +       err = au_busy_or_stale();
21302 +       au_ren_unlock(a);
21303 +
21304 +out:
21305 +       return err;
21306 +}
21307 +
21308 +/* ---------------------------------------------------------------------- */
21309 +
21310 +static void au_ren_refresh_dir(struct au_ren_args *a)
21311 +{
21312 +       struct inode *dir;
21313 +
21314 +       dir = a->dst_dir;
21315 +       dir->i_version++;
21316 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)) {
21317 +               /* is this updating defined in POSIX? */
21318 +               au_cpup_attr_timesizes(a->src_inode);
21319 +               au_cpup_attr_nlink(dir, /*force*/1);
21320 +       }
21321 +       au_dir_ts(dir, a->btgt);
21322 +
21323 +       if (a->exchange) {
21324 +               dir = a->src_dir;
21325 +               dir->i_version++;
21326 +               if (au_ftest_ren(a->auren_flags, ISDIR_DST)) {
21327 +                       /* is this updating defined in POSIX? */
21328 +                       au_cpup_attr_timesizes(a->dst_inode);
21329 +                       au_cpup_attr_nlink(dir, /*force*/1);
21330 +               }
21331 +               au_dir_ts(dir, a->btgt);
21332 +       }
21333 +
21334 +       if (au_ftest_ren(a->auren_flags, ISSAMEDIR))
21335 +               return;
21336 +
21337 +       dir = a->src_dir;
21338 +       dir->i_version++;
21339 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC))
21340 +               au_cpup_attr_nlink(dir, /*force*/1);
21341 +       au_dir_ts(dir, a->btgt);
21342 +}
21343 +
21344 +static void au_ren_refresh(struct au_ren_args *a)
21345 +{
21346 +       aufs_bindex_t bbot, bindex;
21347 +       struct dentry *d, *h_d;
21348 +       struct inode *i, *h_i;
21349 +       struct super_block *sb;
21350 +
21351 +       d = a->dst_dentry;
21352 +       d_drop(d);
21353 +       if (a->h_dst)
21354 +               /* already dget-ed by au_ren_or_cpup() */
21355 +               au_set_h_dptr(d, a->btgt, a->h_dst);
21356 +
21357 +       i = a->dst_inode;
21358 +       if (i) {
21359 +               if (!a->exchange) {
21360 +                       if (!au_ftest_ren(a->auren_flags, ISDIR_DST))
21361 +                               vfsub_drop_nlink(i);
21362 +                       else {
21363 +                               vfsub_dead_dir(i);
21364 +                               au_cpup_attr_timesizes(i);
21365 +                       }
21366 +                       au_update_dbrange(d, /*do_put_zero*/1);
21367 +               } else
21368 +                       au_cpup_attr_nlink(i, /*force*/1);
21369 +       } else {
21370 +               bbot = a->btgt;
21371 +               for (bindex = au_dbtop(d); bindex < bbot; bindex++)
21372 +                       au_set_h_dptr(d, bindex, NULL);
21373 +               bbot = au_dbbot(d);
21374 +               for (bindex = a->btgt + 1; bindex <= bbot; bindex++)
21375 +                       au_set_h_dptr(d, bindex, NULL);
21376 +               au_update_dbrange(d, /*do_put_zero*/0);
21377 +       }
21378 +
21379 +       d = a->src_dentry;
21380 +       if (!a->exchange) {
21381 +               au_set_dbwh(d, -1);
21382 +               bbot = au_dbbot(d);
21383 +               for (bindex = a->btgt + 1; bindex <= bbot; bindex++) {
21384 +                       h_d = au_h_dptr(d, bindex);
21385 +                       if (h_d)
21386 +                               au_set_h_dptr(d, bindex, NULL);
21387 +               }
21388 +               au_set_dbbot(d, a->btgt);
21389 +
21390 +               sb = d->d_sb;
21391 +               i = a->src_inode;
21392 +               if (au_opt_test(au_mntflags(sb), PLINK) && au_plink_test(i))
21393 +                       return; /* success */
21394 +
21395 +               bbot = au_ibbot(i);
21396 +               for (bindex = a->btgt + 1; bindex <= bbot; bindex++) {
21397 +                       h_i = au_h_iptr(i, bindex);
21398 +                       if (h_i) {
21399 +                               au_xino_write(sb, bindex, h_i->i_ino, /*ino*/0);
21400 +                               /* ignore this error */
21401 +                               au_set_h_iptr(i, bindex, NULL, 0);
21402 +                       }
21403 +               }
21404 +               au_set_ibbot(i, a->btgt);
21405 +       }
21406 +               d_drop(a->src_dentry);
21407 +}
21408 +
21409 +/* ---------------------------------------------------------------------- */
21410 +
21411 +/* mainly for link(2) and rename(2) */
21412 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt)
21413 +{
21414 +       aufs_bindex_t bdiropq, bwh;
21415 +       struct dentry *parent;
21416 +       struct au_branch *br;
21417 +
21418 +       parent = dentry->d_parent;
21419 +       IMustLock(d_inode(parent)); /* dir is locked */
21420 +
21421 +       bdiropq = au_dbdiropq(parent);
21422 +       bwh = au_dbwh(dentry);
21423 +       br = au_sbr(dentry->d_sb, btgt);
21424 +       if (au_br_rdonly(br)
21425 +           || (0 <= bdiropq && bdiropq < btgt)
21426 +           || (0 <= bwh && bwh < btgt))
21427 +               btgt = -1;
21428 +
21429 +       AuDbg("btgt %d\n", btgt);
21430 +       return btgt;
21431 +}
21432 +
21433 +/* sets src_btop, dst_btop and btgt */
21434 +static int au_ren_wbr(struct au_ren_args *a)
21435 +{
21436 +       int err;
21437 +       struct au_wr_dir_args wr_dir_args = {
21438 +               /* .force_btgt  = -1, */
21439 +               .flags          = AuWrDir_ADD_ENTRY
21440 +       };
21441 +
21442 +       a->src_btop = au_dbtop(a->src_dentry);
21443 +       a->dst_btop = au_dbtop(a->dst_dentry);
21444 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC)
21445 +           || au_ftest_ren(a->auren_flags, ISDIR_DST))
21446 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
21447 +       wr_dir_args.force_btgt = a->src_btop;
21448 +       if (a->dst_inode && a->dst_btop < a->src_btop)
21449 +               wr_dir_args.force_btgt = a->dst_btop;
21450 +       wr_dir_args.force_btgt = au_wbr(a->dst_dentry, wr_dir_args.force_btgt);
21451 +       err = au_wr_dir(a->dst_dentry, a->src_dentry, &wr_dir_args);
21452 +       a->btgt = err;
21453 +       if (a->exchange)
21454 +               au_update_dbtop(a->dst_dentry);
21455 +
21456 +       return err;
21457 +}
21458 +
21459 +static void au_ren_dt(struct au_ren_args *a)
21460 +{
21461 +       a->h_path.dentry = a->src_h_parent;
21462 +       au_dtime_store(a->src_dt + AuPARENT, a->src_parent, &a->h_path);
21463 +       if (!au_ftest_ren(a->auren_flags, ISSAMEDIR)) {
21464 +               a->h_path.dentry = a->dst_h_parent;
21465 +               au_dtime_store(a->dst_dt + AuPARENT, a->dst_parent, &a->h_path);
21466 +       }
21467 +
21468 +       au_fclr_ren(a->auren_flags, DT_DSTDIR);
21469 +       if (!au_ftest_ren(a->auren_flags, ISDIR_SRC)
21470 +           && !a->exchange)
21471 +               return;
21472 +
21473 +       a->h_path.dentry = a->src_h_dentry;
21474 +       au_dtime_store(a->src_dt + AuCHILD, a->src_dentry, &a->h_path);
21475 +       if (d_is_positive(a->dst_h_dentry)) {
21476 +               au_fset_ren(a->auren_flags, DT_DSTDIR);
21477 +               a->h_path.dentry = a->dst_h_dentry;
21478 +               au_dtime_store(a->dst_dt + AuCHILD, a->dst_dentry, &a->h_path);
21479 +       }
21480 +}
21481 +
21482 +static void au_ren_rev_dt(int err, struct au_ren_args *a)
21483 +{
21484 +       struct dentry *h_d;
21485 +       struct inode *h_inode;
21486 +
21487 +       au_dtime_revert(a->src_dt + AuPARENT);
21488 +       if (!au_ftest_ren(a->auren_flags, ISSAMEDIR))
21489 +               au_dtime_revert(a->dst_dt + AuPARENT);
21490 +
21491 +       if (au_ftest_ren(a->auren_flags, ISDIR_SRC) && err != -EIO) {
21492 +               h_d = a->src_dt[AuCHILD].dt_h_path.dentry;
21493 +               h_inode = d_inode(h_d);
21494 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
21495 +               au_dtime_revert(a->src_dt + AuCHILD);
21496 +               inode_unlock(h_inode);
21497 +
21498 +               if (au_ftest_ren(a->auren_flags, DT_DSTDIR)) {
21499 +                       h_d = a->dst_dt[AuCHILD].dt_h_path.dentry;
21500 +                       h_inode = d_inode(h_d);
21501 +                       inode_lock_nested(h_inode, AuLsc_I_CHILD);
21502 +                       au_dtime_revert(a->dst_dt + AuCHILD);
21503 +                       inode_unlock(h_inode);
21504 +               }
21505 +       }
21506 +}
21507 +
21508 +/* ---------------------------------------------------------------------- */
21509 +
21510 +int aufs_rename(struct inode *_src_dir, struct dentry *_src_dentry,
21511 +               struct inode *_dst_dir, struct dentry *_dst_dentry,
21512 +               unsigned int _flags)
21513 +{
21514 +       int err, lock_flags;
21515 +       /* reduce stack space */
21516 +       struct au_ren_args *a;
21517 +       struct au_pin pin;
21518 +
21519 +       AuDbg("%pd, %pd, 0x%x\n", _src_dentry, _dst_dentry, _flags);
21520 +       IMustLock(_src_dir);
21521 +       IMustLock(_dst_dir);
21522 +
21523 +       err = -EINVAL;
21524 +       if (unlikely(_flags & RENAME_WHITEOUT))
21525 +               goto out;
21526 +
21527 +       err = -ENOMEM;
21528 +       BUILD_BUG_ON(sizeof(*a) > PAGE_SIZE);
21529 +       a = kzalloc(sizeof(*a), GFP_NOFS);
21530 +       if (unlikely(!a))
21531 +               goto out;
21532 +
21533 +       a->flags = _flags;
21534 +       a->exchange = _flags & RENAME_EXCHANGE;
21535 +       a->src_dir = _src_dir;
21536 +       a->src_dentry = _src_dentry;
21537 +       a->src_inode = NULL;
21538 +       if (d_really_is_positive(a->src_dentry))
21539 +               a->src_inode = d_inode(a->src_dentry);
21540 +       a->src_parent = a->src_dentry->d_parent; /* dir inode is locked */
21541 +       a->dst_dir = _dst_dir;
21542 +       a->dst_dentry = _dst_dentry;
21543 +       a->dst_inode = NULL;
21544 +       if (d_really_is_positive(a->dst_dentry))
21545 +               a->dst_inode = d_inode(a->dst_dentry);
21546 +       a->dst_parent = a->dst_dentry->d_parent; /* dir inode is locked */
21547 +       if (a->dst_inode) {
21548 +               /*
21549 +                * if EXCHANGE && src is non-dir && dst is dir,
21550 +                * dst is not locked.
21551 +                */
21552 +               /* IMustLock(a->dst_inode); */
21553 +               au_igrab(a->dst_inode);
21554 +       }
21555 +
21556 +       err = -ENOTDIR;
21557 +       lock_flags = AuLock_FLUSH | AuLock_NOPLM | AuLock_GEN;
21558 +       if (d_is_dir(a->src_dentry)) {
21559 +               au_fset_ren(a->auren_flags, ISDIR_SRC);
21560 +               if (unlikely(!a->exchange
21561 +                            && d_really_is_positive(a->dst_dentry)
21562 +                            && !d_is_dir(a->dst_dentry)))
21563 +                       goto out_free;
21564 +               lock_flags |= AuLock_DIRS;
21565 +       }
21566 +       if (a->dst_inode && d_is_dir(a->dst_dentry)) {
21567 +               au_fset_ren(a->auren_flags, ISDIR_DST);
21568 +               if (unlikely(!a->exchange
21569 +                            && d_really_is_positive(a->src_dentry)
21570 +                            && !d_is_dir(a->src_dentry)))
21571 +                       goto out_free;
21572 +               lock_flags |= AuLock_DIRS;
21573 +       }
21574 +       err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry, lock_flags);
21575 +       if (unlikely(err))
21576 +               goto out_free;
21577 +
21578 +       err = au_d_hashed_positive(a->src_dentry);
21579 +       if (unlikely(err))
21580 +               goto out_unlock;
21581 +       err = -ENOENT;
21582 +       if (a->dst_inode) {
21583 +               /*
21584 +                * If it is a dir, VFS unhash it before this
21585 +                * function. It means we cannot rely upon d_unhashed().
21586 +                */
21587 +               if (unlikely(!a->dst_inode->i_nlink))
21588 +                       goto out_unlock;
21589 +               if (!au_ftest_ren(a->auren_flags, ISDIR_DST)) {
21590 +                       err = au_d_hashed_positive(a->dst_dentry);
21591 +                       if (unlikely(err && !a->exchange))
21592 +                               goto out_unlock;
21593 +               } else if (unlikely(IS_DEADDIR(a->dst_inode)))
21594 +                       goto out_unlock;
21595 +       } else if (unlikely(d_unhashed(a->dst_dentry)))
21596 +               goto out_unlock;
21597 +
21598 +       /*
21599 +        * is it possible?
21600 +        * yes, it happened (in linux-3.3-rcN) but I don't know why.
21601 +        * there may exist a problem somewhere else.
21602 +        */
21603 +       err = -EINVAL;
21604 +       if (unlikely(d_inode(a->dst_parent) == d_inode(a->src_dentry)))
21605 +               goto out_unlock;
21606 +
21607 +       au_fset_ren(a->auren_flags, ISSAMEDIR); /* temporary */
21608 +       di_write_lock_parent(a->dst_parent);
21609 +
21610 +       /* which branch we process */
21611 +       err = au_ren_wbr(a);
21612 +       if (unlikely(err < 0))
21613 +               goto out_parent;
21614 +       a->br = au_sbr(a->dst_dentry->d_sb, a->btgt);
21615 +       a->h_path.mnt = au_br_mnt(a->br);
21616 +
21617 +       /* are they available to be renamed */
21618 +       err = au_ren_may_dir(a);
21619 +       if (unlikely(err))
21620 +               goto out_children;
21621 +
21622 +       /* prepare the writable parent dir on the same branch */
21623 +       if (a->dst_btop == a->btgt) {
21624 +               au_fset_ren(a->auren_flags, WHDST);
21625 +       } else {
21626 +               err = au_cpup_dirs(a->dst_dentry, a->btgt);
21627 +               if (unlikely(err))
21628 +                       goto out_children;
21629 +       }
21630 +
21631 +       err = 0;
21632 +       if (!a->exchange) {
21633 +               if (a->src_dir != a->dst_dir) {
21634 +                       /*
21635 +                        * this temporary unlock is safe,
21636 +                        * because both dir->i_mutex are locked.
21637 +                        */
21638 +                       di_write_unlock(a->dst_parent);
21639 +                       di_write_lock_parent(a->src_parent);
21640 +                       err = au_wr_dir_need_wh(a->src_dentry,
21641 +                                               au_ftest_ren(a->auren_flags,
21642 +                                                            ISDIR_SRC),
21643 +                                               &a->btgt);
21644 +                       di_write_unlock(a->src_parent);
21645 +                       di_write_lock2_parent(a->src_parent, a->dst_parent,
21646 +                                             /*isdir*/1);
21647 +                       au_fclr_ren(a->auren_flags, ISSAMEDIR);
21648 +               } else
21649 +                       err = au_wr_dir_need_wh(a->src_dentry,
21650 +                                               au_ftest_ren(a->auren_flags,
21651 +                                                            ISDIR_SRC),
21652 +                                               &a->btgt);
21653 +       }
21654 +       if (unlikely(err < 0))
21655 +               goto out_children;
21656 +       if (err)
21657 +               au_fset_ren(a->auren_flags, WHSRC);
21658 +
21659 +       /* cpup src */
21660 +       if (a->src_btop != a->btgt) {
21661 +               err = au_pin(&pin, a->src_dentry, a->btgt,
21662 +                            au_opt_udba(a->src_dentry->d_sb),
21663 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
21664 +               if (!err) {
21665 +                       struct au_cp_generic cpg = {
21666 +                               .dentry = a->src_dentry,
21667 +                               .bdst   = a->btgt,
21668 +                               .bsrc   = a->src_btop,
21669 +                               .len    = -1,
21670 +                               .pin    = &pin,
21671 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
21672 +                       };
21673 +                       AuDebugOn(au_dbtop(a->src_dentry) != a->src_btop);
21674 +                       err = au_sio_cpup_simple(&cpg);
21675 +                       au_unpin(&pin);
21676 +               }
21677 +               if (unlikely(err))
21678 +                       goto out_children;
21679 +               a->src_btop = a->btgt;
21680 +               a->src_h_dentry = au_h_dptr(a->src_dentry, a->btgt);
21681 +               if (!a->exchange)
21682 +                       au_fset_ren(a->auren_flags, WHSRC);
21683 +       }
21684 +
21685 +       /* cpup dst */
21686 +       if (a->exchange && a->dst_inode
21687 +           && a->dst_btop != a->btgt) {
21688 +               err = au_pin(&pin, a->dst_dentry, a->btgt,
21689 +                            au_opt_udba(a->dst_dentry->d_sb),
21690 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
21691 +               if (!err) {
21692 +                       struct au_cp_generic cpg = {
21693 +                               .dentry = a->dst_dentry,
21694 +                               .bdst   = a->btgt,
21695 +                               .bsrc   = a->dst_btop,
21696 +                               .len    = -1,
21697 +                               .pin    = &pin,
21698 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
21699 +                       };
21700 +                       err = au_sio_cpup_simple(&cpg);
21701 +                       au_unpin(&pin);
21702 +               }
21703 +               if (unlikely(err))
21704 +                       goto out_children;
21705 +               a->dst_btop = a->btgt;
21706 +               a->dst_h_dentry = au_h_dptr(a->dst_dentry, a->btgt);
21707 +       }
21708 +
21709 +       /* lock them all */
21710 +       err = au_ren_lock(a);
21711 +       if (unlikely(err))
21712 +               /* leave the copied-up one */
21713 +               goto out_children;
21714 +
21715 +       if (!a->exchange) {
21716 +               if (!au_opt_test(au_mntflags(a->dst_dir->i_sb), UDBA_NONE))
21717 +                       err = au_may_ren(a);
21718 +               else if (unlikely(a->dst_dentry->d_name.len > AUFS_MAX_NAMELEN))
21719 +                       err = -ENAMETOOLONG;
21720 +               if (unlikely(err))
21721 +                       goto out_hdir;
21722 +       }
21723 +
21724 +       /* store timestamps to be revertible */
21725 +       au_ren_dt(a);
21726 +
21727 +       /* here we go */
21728 +       err = do_rename(a);
21729 +       if (unlikely(err))
21730 +               goto out_dt;
21731 +
21732 +       /* update dir attributes */
21733 +       au_ren_refresh_dir(a);
21734 +
21735 +       /* dput/iput all lower dentries */
21736 +       au_ren_refresh(a);
21737 +
21738 +       goto out_hdir; /* success */
21739 +
21740 +out_dt:
21741 +       au_ren_rev_dt(err, a);
21742 +out_hdir:
21743 +       au_ren_unlock(a);
21744 +out_children:
21745 +       au_nhash_wh_free(&a->whlist);
21746 +       if (err && a->dst_inode && a->dst_btop != a->btgt) {
21747 +               AuDbg("btop %d, btgt %d\n", a->dst_btop, a->btgt);
21748 +               au_set_h_dptr(a->dst_dentry, a->btgt, NULL);
21749 +               au_set_dbtop(a->dst_dentry, a->dst_btop);
21750 +       }
21751 +out_parent:
21752 +       if (!err) {
21753 +               if (!a->exchange)
21754 +                       d_move(a->src_dentry, a->dst_dentry);
21755 +               else
21756 +                       d_exchange(a->src_dentry, a->dst_dentry);
21757 +       } else {
21758 +               au_update_dbtop(a->dst_dentry);
21759 +               if (!a->dst_inode)
21760 +                       d_drop(a->dst_dentry);
21761 +       }
21762 +       if (au_ftest_ren(a->auren_flags, ISSAMEDIR))
21763 +               di_write_unlock(a->dst_parent);
21764 +       else
21765 +               di_write_unlock2(a->src_parent, a->dst_parent);
21766 +out_unlock:
21767 +       aufs_read_and_write_unlock2(a->dst_dentry, a->src_dentry);
21768 +out_free:
21769 +       iput(a->dst_inode);
21770 +       if (a->thargs)
21771 +               au_whtmp_rmdir_free(a->thargs);
21772 +       au_delayed_kfree(a);
21773 +out:
21774 +       AuTraceErr(err);
21775 +       return err;
21776 +}
21777 diff -urN /usr/share/empty/fs/aufs/Kconfig linux/fs/aufs/Kconfig
21778 --- /usr/share/empty/fs/aufs/Kconfig    1970-01-01 01:00:00.000000000 +0100
21779 +++ linux/fs/aufs/Kconfig       2017-05-06 22:16:52.774886674 +0200
21780 @@ -0,0 +1,185 @@
21781 +config AUFS_FS
21782 +       tristate "Aufs (Advanced multi layered unification filesystem) support"
21783 +       help
21784 +       Aufs is a stackable unification filesystem such as Unionfs,
21785 +       which unifies several directories and provides a merged single
21786 +       directory.
21787 +       In the early days, aufs was entirely re-designed and
21788 +       re-implemented Unionfs Version 1.x series. Introducing many
21789 +       original ideas, approaches and improvements, it becomes totally
21790 +       different from Unionfs while keeping the basic features.
21791 +
21792 +if AUFS_FS
21793 +choice
21794 +       prompt "Maximum number of branches"
21795 +       default AUFS_BRANCH_MAX_127
21796 +       help
21797 +       Specifies the maximum number of branches (or member directories)
21798 +       in a single aufs. The larger value consumes more system
21799 +       resources and has a minor impact to performance.
21800 +config AUFS_BRANCH_MAX_127
21801 +       bool "127"
21802 +       help
21803 +       Specifies the maximum number of branches (or member directories)
21804 +       in a single aufs. The larger value consumes more system
21805 +       resources and has a minor impact to performance.
21806 +config AUFS_BRANCH_MAX_511
21807 +       bool "511"
21808 +       help
21809 +       Specifies the maximum number of branches (or member directories)
21810 +       in a single aufs. The larger value consumes more system
21811 +       resources and has a minor impact to performance.
21812 +config AUFS_BRANCH_MAX_1023
21813 +       bool "1023"
21814 +       help
21815 +       Specifies the maximum number of branches (or member directories)
21816 +       in a single aufs. The larger value consumes more system
21817 +       resources and has a minor impact to performance.
21818 +config AUFS_BRANCH_MAX_32767
21819 +       bool "32767"
21820 +       help
21821 +       Specifies the maximum number of branches (or member directories)
21822 +       in a single aufs. The larger value consumes more system
21823 +       resources and has a minor impact to performance.
21824 +endchoice
21825 +
21826 +config AUFS_SBILIST
21827 +       bool
21828 +       depends on AUFS_MAGIC_SYSRQ || PROC_FS
21829 +       default y
21830 +       help
21831 +       Automatic configuration for internal use.
21832 +       When aufs supports Magic SysRq or /proc, enabled automatically.
21833 +
21834 +config AUFS_HNOTIFY
21835 +       bool "Detect direct branch access (bypassing aufs)"
21836 +       help
21837 +       If you want to modify files on branches directly, eg. bypassing aufs,
21838 +       and want aufs to detect the changes of them fully, then enable this
21839 +       option and use 'udba=notify' mount option.
21840 +       Currently there is only one available configuration, "fsnotify".
21841 +       It will have a negative impact to the performance.
21842 +       See detail in aufs.5.
21843 +
21844 +choice
21845 +       prompt "method" if AUFS_HNOTIFY
21846 +       default AUFS_HFSNOTIFY
21847 +config AUFS_HFSNOTIFY
21848 +       bool "fsnotify"
21849 +       select FSNOTIFY
21850 +endchoice
21851 +
21852 +config AUFS_EXPORT
21853 +       bool "NFS-exportable aufs"
21854 +       depends on EXPORTFS
21855 +       help
21856 +       If you want to export your mounted aufs via NFS, then enable this
21857 +       option. There are several requirements for this configuration.
21858 +       See detail in aufs.5.
21859 +
21860 +config AUFS_INO_T_64
21861 +       bool
21862 +       depends on AUFS_EXPORT
21863 +       depends on 64BIT && !(ALPHA || S390)
21864 +       default y
21865 +       help
21866 +       Automatic configuration for internal use.
21867 +       /* typedef unsigned long/int __kernel_ino_t */
21868 +       /* alpha and s390x are int */
21869 +
21870 +config AUFS_XATTR
21871 +       bool "support for XATTR/EA (including Security Labels)"
21872 +       help
21873 +       If your branch fs supports XATTR/EA and you want to make them
21874 +       available in aufs too, then enable this opsion and specify the
21875 +       branch attributes for EA.
21876 +       See detail in aufs.5.
21877 +
21878 +config AUFS_FHSM
21879 +       bool "File-based Hierarchical Storage Management"
21880 +       help
21881 +       Hierarchical Storage Management (or HSM) is a well-known feature
21882 +       in the storage world. Aufs provides this feature as file-based.
21883 +       with multiple branches.
21884 +       These multiple branches are prioritized, ie. the topmost one
21885 +       should be the fastest drive and be used heavily.
21886 +
21887 +config AUFS_RDU
21888 +       bool "Readdir in userspace"
21889 +       help
21890 +       Aufs has two methods to provide a merged view for a directory,
21891 +       by a user-space library and by kernel-space natively. The latter
21892 +       is always enabled but sometimes large and slow.
21893 +       If you enable this option, install the library in aufs2-util
21894 +       package, and set some environment variables for your readdir(3),
21895 +       then the work will be handled in user-space which generally
21896 +       shows better performance in most cases.
21897 +       See detail in aufs.5.
21898 +
21899 +config AUFS_SHWH
21900 +       bool "Show whiteouts"
21901 +       help
21902 +       If you want to make the whiteouts in aufs visible, then enable
21903 +       this option and specify 'shwh' mount option. Although it may
21904 +       sounds like philosophy or something, but in technically it
21905 +       simply shows the name of whiteout with keeping its behaviour.
21906 +
21907 +config AUFS_BR_RAMFS
21908 +       bool "Ramfs (initramfs/rootfs) as an aufs branch"
21909 +       help
21910 +       If you want to use ramfs as an aufs branch fs, then enable this
21911 +       option. Generally tmpfs is recommended.
21912 +       Aufs prohibited them to be a branch fs by default, because
21913 +       initramfs becomes unusable after switch_root or something
21914 +       generally. If you sets initramfs as an aufs branch and boot your
21915 +       system by switch_root, you will meet a problem easily since the
21916 +       files in initramfs may be inaccessible.
21917 +       Unless you are going to use ramfs as an aufs branch fs without
21918 +       switch_root or something, leave it N.
21919 +
21920 +config AUFS_BR_FUSE
21921 +       bool "Fuse fs as an aufs branch"
21922 +       depends on FUSE_FS
21923 +       select AUFS_POLL
21924 +       help
21925 +       If you want to use fuse-based userspace filesystem as an aufs
21926 +       branch fs, then enable this option.
21927 +       It implements the internal poll(2) operation which is
21928 +       implemented by fuse only (curretnly).
21929 +
21930 +config AUFS_POLL
21931 +       bool
21932 +       help
21933 +       Automatic configuration for internal use.
21934 +
21935 +config AUFS_BR_HFSPLUS
21936 +       bool "Hfsplus as an aufs branch"
21937 +       depends on HFSPLUS_FS
21938 +       default y
21939 +       help
21940 +       If you want to use hfsplus fs as an aufs branch fs, then enable
21941 +       this option. This option introduces a small overhead at
21942 +       copying-up a file on hfsplus.
21943 +
21944 +config AUFS_BDEV_LOOP
21945 +       bool
21946 +       depends on BLK_DEV_LOOP
21947 +       default y
21948 +       help
21949 +       Automatic configuration for internal use.
21950 +       Convert =[ym] into =y.
21951 +
21952 +config AUFS_DEBUG
21953 +       bool "Debug aufs"
21954 +       help
21955 +       Enable this to compile aufs internal debug code.
21956 +       It will have a negative impact to the performance.
21957 +
21958 +config AUFS_MAGIC_SYSRQ
21959 +       bool
21960 +       depends on AUFS_DEBUG && MAGIC_SYSRQ
21961 +       default y
21962 +       help
21963 +       Automatic configuration for internal use.
21964 +       When aufs supports Magic SysRq, enabled automatically.
21965 +endif
21966 diff -urN /usr/share/empty/fs/aufs/loop.c linux/fs/aufs/loop.c
21967 --- /usr/share/empty/fs/aufs/loop.c     1970-01-01 01:00:00.000000000 +0100
21968 +++ linux/fs/aufs/loop.c        2017-05-06 22:16:52.781553511 +0200
21969 @@ -0,0 +1,147 @@
21970 +/*
21971 + * Copyright (C) 2005-2017 Junjiro R. Okajima
21972 + *
21973 + * This program, aufs is free software; you can redistribute it and/or modify
21974 + * it under the terms of the GNU General Public License as published by
21975 + * the Free Software Foundation; either version 2 of the License, or
21976 + * (at your option) any later version.
21977 + *
21978 + * This program is distributed in the hope that it will be useful,
21979 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21980 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21981 + * GNU General Public License for more details.
21982 + *
21983 + * You should have received a copy of the GNU General Public License
21984 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21985 + */
21986 +
21987 +/*
21988 + * support for loopback block device as a branch
21989 + */
21990 +
21991 +#include "aufs.h"
21992 +
21993 +/* added into drivers/block/loop.c */
21994 +static struct file *(*backing_file_func)(struct super_block *sb);
21995 +
21996 +/*
21997 + * test if two lower dentries have overlapping branches.
21998 + */
21999 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding)
22000 +{
22001 +       struct super_block *h_sb;
22002 +       struct file *backing_file;
22003 +
22004 +       if (unlikely(!backing_file_func)) {
22005 +               /* don't load "loop" module here */
22006 +               backing_file_func = symbol_get(loop_backing_file);
22007 +               if (unlikely(!backing_file_func))
22008 +                       /* "loop" module is not loaded */
22009 +                       return 0;
22010 +       }
22011 +
22012 +       h_sb = h_adding->d_sb;
22013 +       backing_file = backing_file_func(h_sb);
22014 +       if (!backing_file)
22015 +               return 0;
22016 +
22017 +       h_adding = backing_file->f_path.dentry;
22018 +       /*
22019 +        * h_adding can be local NFS.
22020 +        * in this case aufs cannot detect the loop.
22021 +        */
22022 +       if (unlikely(h_adding->d_sb == sb))
22023 +               return 1;
22024 +       return !!au_test_subdir(h_adding, sb->s_root);
22025 +}
22026 +
22027 +/* true if a kernel thread named 'loop[0-9].*' accesses a file */
22028 +int au_test_loopback_kthread(void)
22029 +{
22030 +       int ret;
22031 +       struct task_struct *tsk = current;
22032 +       char c, comm[sizeof(tsk->comm)];
22033 +
22034 +       ret = 0;
22035 +       if (tsk->flags & PF_KTHREAD) {
22036 +               get_task_comm(comm, tsk);
22037 +               c = comm[4];
22038 +               ret = ('0' <= c && c <= '9'
22039 +                      && !strncmp(comm, "loop", 4));
22040 +       }
22041 +
22042 +       return ret;
22043 +}
22044 +
22045 +/* ---------------------------------------------------------------------- */
22046 +
22047 +#define au_warn_loopback_step  16
22048 +static int au_warn_loopback_nelem = au_warn_loopback_step;
22049 +static unsigned long *au_warn_loopback_array;
22050 +
22051 +void au_warn_loopback(struct super_block *h_sb)
22052 +{
22053 +       int i, new_nelem;
22054 +       unsigned long *a, magic;
22055 +       static DEFINE_SPINLOCK(spin);
22056 +
22057 +       magic = h_sb->s_magic;
22058 +       spin_lock(&spin);
22059 +       a = au_warn_loopback_array;
22060 +       for (i = 0; i < au_warn_loopback_nelem && *a; i++)
22061 +               if (a[i] == magic) {
22062 +                       spin_unlock(&spin);
22063 +                       return;
22064 +               }
22065 +
22066 +       /* h_sb is new to us, print it */
22067 +       if (i < au_warn_loopback_nelem) {
22068 +               a[i] = magic;
22069 +               goto pr;
22070 +       }
22071 +
22072 +       /* expand the array */
22073 +       new_nelem = au_warn_loopback_nelem + au_warn_loopback_step;
22074 +       a = au_kzrealloc(au_warn_loopback_array,
22075 +                        au_warn_loopback_nelem * sizeof(unsigned long),
22076 +                        new_nelem * sizeof(unsigned long), GFP_ATOMIC,
22077 +                        /*may_shrink*/0);
22078 +       if (a) {
22079 +               au_warn_loopback_nelem = new_nelem;
22080 +               au_warn_loopback_array = a;
22081 +               a[i] = magic;
22082 +               goto pr;
22083 +       }
22084 +
22085 +       spin_unlock(&spin);
22086 +       AuWarn1("realloc failed, ignored\n");
22087 +       return;
22088 +
22089 +pr:
22090 +       spin_unlock(&spin);
22091 +       pr_warn("you may want to try another patch for loopback file "
22092 +               "on %s(0x%lx) branch\n", au_sbtype(h_sb), magic);
22093 +}
22094 +
22095 +int au_loopback_init(void)
22096 +{
22097 +       int err;
22098 +       struct super_block *sb __maybe_unused;
22099 +
22100 +       BUILD_BUG_ON(sizeof(sb->s_magic) != sizeof(unsigned long));
22101 +
22102 +       err = 0;
22103 +       au_warn_loopback_array = kcalloc(au_warn_loopback_step,
22104 +                                        sizeof(unsigned long), GFP_NOFS);
22105 +       if (unlikely(!au_warn_loopback_array))
22106 +               err = -ENOMEM;
22107 +
22108 +       return err;
22109 +}
22110 +
22111 +void au_loopback_fin(void)
22112 +{
22113 +       if (backing_file_func)
22114 +               symbol_put(loop_backing_file);
22115 +       au_delayed_kfree(au_warn_loopback_array);
22116 +}
22117 diff -urN /usr/share/empty/fs/aufs/loop.h linux/fs/aufs/loop.h
22118 --- /usr/share/empty/fs/aufs/loop.h     1970-01-01 01:00:00.000000000 +0100
22119 +++ linux/fs/aufs/loop.h        2017-05-06 22:16:52.781553511 +0200
22120 @@ -0,0 +1,52 @@
22121 +/*
22122 + * Copyright (C) 2005-2017 Junjiro R. Okajima
22123 + *
22124 + * This program, aufs is free software; you can redistribute it and/or modify
22125 + * it under the terms of the GNU General Public License as published by
22126 + * the Free Software Foundation; either version 2 of the License, or
22127 + * (at your option) any later version.
22128 + *
22129 + * This program is distributed in the hope that it will be useful,
22130 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22131 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22132 + * GNU General Public License for more details.
22133 + *
22134 + * You should have received a copy of the GNU General Public License
22135 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22136 + */
22137 +
22138 +/*
22139 + * support for loopback mount as a branch
22140 + */
22141 +
22142 +#ifndef __AUFS_LOOP_H__
22143 +#define __AUFS_LOOP_H__
22144 +
22145 +#ifdef __KERNEL__
22146 +
22147 +struct dentry;
22148 +struct super_block;
22149 +
22150 +#ifdef CONFIG_AUFS_BDEV_LOOP
22151 +/* drivers/block/loop.c */
22152 +struct file *loop_backing_file(struct super_block *sb);
22153 +
22154 +/* loop.c */
22155 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding);
22156 +int au_test_loopback_kthread(void);
22157 +void au_warn_loopback(struct super_block *h_sb);
22158 +
22159 +int au_loopback_init(void);
22160 +void au_loopback_fin(void);
22161 +#else
22162 +AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
22163 +          struct dentry *h_adding)
22164 +AuStubInt0(au_test_loopback_kthread, void)
22165 +AuStubVoid(au_warn_loopback, struct super_block *h_sb)
22166 +
22167 +AuStubInt0(au_loopback_init, void)
22168 +AuStubVoid(au_loopback_fin, void)
22169 +#endif /* BLK_DEV_LOOP */
22170 +
22171 +#endif /* __KERNEL__ */
22172 +#endif /* __AUFS_LOOP_H__ */
22173 diff -urN /usr/share/empty/fs/aufs/magic.mk linux/fs/aufs/magic.mk
22174 --- /usr/share/empty/fs/aufs/magic.mk   1970-01-01 01:00:00.000000000 +0100
22175 +++ linux/fs/aufs/magic.mk      2017-05-06 22:16:52.781553511 +0200
22176 @@ -0,0 +1,30 @@
22177 +
22178 +# defined in ${srctree}/fs/fuse/inode.c
22179 +# tristate
22180 +ifdef CONFIG_FUSE_FS
22181 +ccflags-y += -DFUSE_SUPER_MAGIC=0x65735546
22182 +endif
22183 +
22184 +# defined in ${srctree}/fs/xfs/xfs_sb.h
22185 +# tristate
22186 +ifdef CONFIG_XFS_FS
22187 +ccflags-y += -DXFS_SB_MAGIC=0x58465342
22188 +endif
22189 +
22190 +# defined in ${srctree}/fs/configfs/mount.c
22191 +# tristate
22192 +ifdef CONFIG_CONFIGFS_FS
22193 +ccflags-y += -DCONFIGFS_MAGIC=0x62656570
22194 +endif
22195 +
22196 +# defined in ${srctree}/fs/ubifs/ubifs.h
22197 +# tristate
22198 +ifdef CONFIG_UBIFS_FS
22199 +ccflags-y += -DUBIFS_SUPER_MAGIC=0x24051905
22200 +endif
22201 +
22202 +# defined in ${srctree}/fs/hfsplus/hfsplus_raw.h
22203 +# tristate
22204 +ifdef CONFIG_HFSPLUS_FS
22205 +ccflags-y += -DHFSPLUS_SUPER_MAGIC=0x482b
22206 +endif
22207 diff -urN /usr/share/empty/fs/aufs/Makefile linux/fs/aufs/Makefile
22208 --- /usr/share/empty/fs/aufs/Makefile   1970-01-01 01:00:00.000000000 +0100
22209 +++ linux/fs/aufs/Makefile      2017-05-06 22:16:52.774886674 +0200
22210 @@ -0,0 +1,44 @@
22211 +
22212 +include ${src}/magic.mk
22213 +ifeq (${CONFIG_AUFS_FS},m)
22214 +include ${src}/conf.mk
22215 +endif
22216 +-include ${src}/priv_def.mk
22217 +
22218 +# cf. include/linux/kernel.h
22219 +# enable pr_debug
22220 +ccflags-y += -DDEBUG
22221 +# sparse requires the full pathname
22222 +ifdef M
22223 +ccflags-y += -include ${M}/../../include/uapi/linux/aufs_type.h
22224 +else
22225 +ccflags-y += -include ${srctree}/include/uapi/linux/aufs_type.h
22226 +endif
22227 +
22228 +obj-$(CONFIG_AUFS_FS) += aufs.o
22229 +aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
22230 +       wkq.o vfsub.o dcsub.o \
22231 +       cpup.o whout.o wbr_policy.o \
22232 +       dinfo.o dentry.o \
22233 +       dynop.o \
22234 +       finfo.o file.o f_op.o \
22235 +       dir.o vdir.o \
22236 +       iinfo.o inode.o i_op.o i_op_add.o i_op_del.o i_op_ren.o \
22237 +       mvdown.o ioctl.o
22238 +
22239 +# all are boolean
22240 +aufs-$(CONFIG_PROC_FS) += procfs.o plink.o
22241 +aufs-$(CONFIG_SYSFS) += sysfs.o
22242 +aufs-$(CONFIG_DEBUG_FS) += dbgaufs.o
22243 +aufs-$(CONFIG_AUFS_BDEV_LOOP) += loop.o
22244 +aufs-$(CONFIG_AUFS_HNOTIFY) += hnotify.o
22245 +aufs-$(CONFIG_AUFS_HFSNOTIFY) += hfsnotify.o
22246 +aufs-$(CONFIG_AUFS_EXPORT) += export.o
22247 +aufs-$(CONFIG_AUFS_XATTR) += xattr.o
22248 +aufs-$(CONFIG_FS_POSIX_ACL) += posix_acl.o
22249 +aufs-$(CONFIG_AUFS_FHSM) += fhsm.o
22250 +aufs-$(CONFIG_AUFS_POLL) += poll.o
22251 +aufs-$(CONFIG_AUFS_RDU) += rdu.o
22252 +aufs-$(CONFIG_AUFS_BR_HFSPLUS) += hfsplus.o
22253 +aufs-$(CONFIG_AUFS_DEBUG) += debug.o
22254 +aufs-$(CONFIG_AUFS_MAGIC_SYSRQ) += sysrq.o
22255 diff -urN /usr/share/empty/fs/aufs/module.c linux/fs/aufs/module.c
22256 --- /usr/share/empty/fs/aufs/module.c   1970-01-01 01:00:00.000000000 +0100
22257 +++ linux/fs/aufs/module.c      2017-05-06 22:16:52.781553511 +0200
22258 @@ -0,0 +1,333 @@
22259 +/*
22260 + * Copyright (C) 2005-2017 Junjiro R. Okajima
22261 + *
22262 + * This program, aufs is free software; you can redistribute it and/or modify
22263 + * it under the terms of the GNU General Public License as published by
22264 + * the Free Software Foundation; either version 2 of the License, or
22265 + * (at your option) any later version.
22266 + *
22267 + * This program is distributed in the hope that it will be useful,
22268 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22269 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22270 + * GNU General Public License for more details.
22271 + *
22272 + * You should have received a copy of the GNU General Public License
22273 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22274 + */
22275 +
22276 +/*
22277 + * module global variables and operations
22278 + */
22279 +
22280 +#include <linux/module.h>
22281 +#include <linux/seq_file.h>
22282 +#include "aufs.h"
22283 +
22284 +/* shrinkable realloc */
22285 +void *au_krealloc(void *p, unsigned int new_sz, gfp_t gfp, int may_shrink)
22286 +{
22287 +       size_t sz;
22288 +       int diff;
22289 +
22290 +       sz = 0;
22291 +       diff = -1;
22292 +       if (p) {
22293 +#if 0 /* unused */
22294 +               if (!new_sz) {
22295 +                       au_delayed_kfree(p);
22296 +                       p = NULL;
22297 +                       goto out;
22298 +               }
22299 +#else
22300 +               AuDebugOn(!new_sz);
22301 +#endif
22302 +               sz = ksize(p);
22303 +               diff = au_kmidx_sub(sz, new_sz);
22304 +       }
22305 +       if (sz && !diff)
22306 +               goto out;
22307 +
22308 +       if (sz < new_sz)
22309 +               /* expand or SLOB */
22310 +               p = krealloc(p, new_sz, gfp);
22311 +       else if (new_sz < sz && may_shrink) {
22312 +               /* shrink */
22313 +               void *q;
22314 +
22315 +               q = kmalloc(new_sz, gfp);
22316 +               if (q) {
22317 +                       if (p) {
22318 +                               memcpy(q, p, new_sz);
22319 +                               au_delayed_kfree(p);
22320 +                       }
22321 +                       p = q;
22322 +               } else
22323 +                       p = NULL;
22324 +       }
22325 +
22326 +out:
22327 +       return p;
22328 +}
22329 +
22330 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp,
22331 +                  int may_shrink)
22332 +{
22333 +       p = au_krealloc(p, new_sz, gfp, may_shrink);
22334 +       if (p && new_sz > nused)
22335 +               memset(p + nused, 0, new_sz - nused);
22336 +       return p;
22337 +}
22338 +
22339 +/* ---------------------------------------------------------------------- */
22340 +/*
22341 + * aufs caches
22342 + */
22343 +
22344 +struct au_dfree au_dfree;
22345 +
22346 +/* delayed free */
22347 +static void au_do_dfree(struct work_struct *work __maybe_unused)
22348 +{
22349 +       struct llist_head *head;
22350 +       struct llist_node *node, *next;
22351 +
22352 +#define AU_CACHE_DFREE_DO_BODY(name, idx, lnode) do {                  \
22353 +               head = &au_dfree.cache[AuCache_##idx].llist;            \
22354 +               node = llist_del_all(head);                             \
22355 +               for (; node; node = next) {                             \
22356 +                       struct au_##name *p                             \
22357 +                               = llist_entry(node, struct au_##name,   \
22358 +                                             lnode);                   \
22359 +                       next = llist_next(node);                        \
22360 +                       au_cache_free_##name(p);                        \
22361 +               }                                                       \
22362 +       } while (0)
22363 +
22364 +       AU_CACHE_DFREE_DO_BODY(dinfo, DINFO, di_lnode);
22365 +       AU_CACHE_DFREE_DO_BODY(icntnr, ICNTNR, lnode);
22366 +       AU_CACHE_DFREE_DO_BODY(finfo, FINFO, fi_lnode);
22367 +       AU_CACHE_DFREE_DO_BODY(vdir, VDIR, vd_lnode);
22368 +       AU_CACHE_DFREE_DO_BODY(vdir_dehstr, DEHSTR, lnode);
22369 +#ifdef CONFIG_AUFS_HNOTIFY
22370 +       AU_CACHE_DFREE_DO_BODY(hnotify, HNOTIFY, hn_lnode);
22371 +#endif
22372 +
22373 +#define AU_DFREE_DO_BODY(llist, func) do {             \
22374 +               node = llist_del_all(llist);            \
22375 +               for (; node; node = next) {             \
22376 +                       next = llist_next(node);        \
22377 +                       func(node);                     \
22378 +               }                                       \
22379 +       } while (0)
22380 +
22381 +       AU_DFREE_DO_BODY(au_dfree.llist + AU_DFREE_KFREE, kfree);
22382 +       AU_DFREE_DO_BODY(au_dfree.llist + AU_DFREE_FREE_PAGE, au_free_page);
22383 +
22384 +#undef AU_CACHE_DFREE_DO_BODY
22385 +#undef AU_DFREE_DO_BODY
22386 +}
22387 +
22388 +AU_CACHE_DFREE_FUNC(dinfo, DINFO, di_lnode);
22389 +AU_CACHE_DFREE_FUNC(icntnr, ICNTNR, lnode);
22390 +AU_CACHE_DFREE_FUNC(finfo, FINFO, fi_lnode);
22391 +AU_CACHE_DFREE_FUNC(vdir, VDIR, vd_lnode);
22392 +AU_CACHE_DFREE_FUNC(vdir_dehstr, DEHSTR, lnode);
22393 +
22394 +static void au_cache_fin(void)
22395 +{
22396 +       int i;
22397 +       struct au_cache *cp;
22398 +
22399 +       /*
22400 +        * Make sure all delayed rcu free inodes are flushed before we
22401 +        * destroy cache.
22402 +        */
22403 +       rcu_barrier();
22404 +
22405 +       /* excluding AuCache_HNOTIFY */
22406 +       BUILD_BUG_ON(AuCache_HNOTIFY + 1 != AuCache_Last);
22407 +       flush_delayed_work(&au_dfree.dwork);
22408 +       for (i = 0; i < AuCache_HNOTIFY; i++) {
22409 +               cp = au_dfree.cache + i;
22410 +               AuDebugOn(!llist_empty(&cp->llist));
22411 +               kmem_cache_destroy(cp->cache);
22412 +               cp->cache = NULL;
22413 +       }
22414 +}
22415 +
22416 +static int __init au_cache_init(void)
22417 +{
22418 +       struct au_cache *cp;
22419 +
22420 +       cp = au_dfree.cache;
22421 +       cp[AuCache_DINFO].cache = AuCacheCtor(au_dinfo, au_di_init_once);
22422 +       if (cp[AuCache_DINFO].cache)
22423 +               /* SLAB_DESTROY_BY_RCU */
22424 +               cp[AuCache_ICNTNR].cache = AuCacheCtor(au_icntnr,
22425 +                                                      au_icntnr_init_once);
22426 +       if (cp[AuCache_ICNTNR].cache)
22427 +               cp[AuCache_FINFO].cache = AuCacheCtor(au_finfo,
22428 +                                                     au_fi_init_once);
22429 +       if (cp[AuCache_FINFO].cache)
22430 +               cp[AuCache_VDIR].cache = AuCache(au_vdir);
22431 +       if (cp[AuCache_VDIR].cache)
22432 +               cp[AuCache_DEHSTR].cache = AuCache(au_vdir_dehstr);
22433 +       if (cp[AuCache_DEHSTR].cache)
22434 +               return 0;
22435 +
22436 +       au_cache_fin();
22437 +       return -ENOMEM;
22438 +}
22439 +
22440 +/* ---------------------------------------------------------------------- */
22441 +
22442 +int au_dir_roflags;
22443 +
22444 +#ifdef CONFIG_AUFS_SBILIST
22445 +/*
22446 + * iterate_supers_type() doesn't protect us from
22447 + * remounting (branch management)
22448 + */
22449 +struct au_sphlhead au_sbilist;
22450 +#endif
22451 +
22452 +/*
22453 + * functions for module interface.
22454 + */
22455 +MODULE_LICENSE("GPL");
22456 +/* MODULE_LICENSE("GPL v2"); */
22457 +MODULE_AUTHOR("Junjiro R. Okajima <aufs-users@lists.sourceforge.net>");
22458 +MODULE_DESCRIPTION(AUFS_NAME
22459 +       " -- Advanced multi layered unification filesystem");
22460 +MODULE_VERSION(AUFS_VERSION);
22461 +MODULE_ALIAS_FS(AUFS_NAME);
22462 +
22463 +/* this module parameter has no meaning when SYSFS is disabled */
22464 +int sysaufs_brs = 1;
22465 +MODULE_PARM_DESC(brs, "use <sysfs>/fs/aufs/si_*/brN");
22466 +module_param_named(brs, sysaufs_brs, int, S_IRUGO);
22467 +
22468 +/* this module parameter has no meaning when USER_NS is disabled */
22469 +bool au_userns;
22470 +MODULE_PARM_DESC(allow_userns, "allow unprivileged to mount under userns");
22471 +module_param_named(allow_userns, au_userns, bool, S_IRUGO);
22472 +
22473 +/* ---------------------------------------------------------------------- */
22474 +
22475 +static char au_esc_chars[0x20 + 3]; /* 0x01-0x20, backslash, del, and NULL */
22476 +
22477 +int au_seq_path(struct seq_file *seq, struct path *path)
22478 +{
22479 +       int err;
22480 +
22481 +       err = seq_path(seq, path, au_esc_chars);
22482 +       if (err > 0)
22483 +               err = 0;
22484 +       else if (err < 0)
22485 +               err = -ENOMEM;
22486 +
22487 +       return err;
22488 +}
22489 +
22490 +/* ---------------------------------------------------------------------- */
22491 +
22492 +static int __init aufs_init(void)
22493 +{
22494 +       int err, i;
22495 +       char *p;
22496 +       struct au_cache *cp;
22497 +
22498 +       p = au_esc_chars;
22499 +       for (i = 1; i <= ' '; i++)
22500 +               *p++ = i;
22501 +       *p++ = '\\';
22502 +       *p++ = '\x7f';
22503 +       *p = 0;
22504 +
22505 +       au_dir_roflags = au_file_roflags(O_DIRECTORY | O_LARGEFILE);
22506 +
22507 +       memcpy(aufs_iop_nogetattr, aufs_iop, sizeof(aufs_iop));
22508 +       for (i = 0; i < AuIop_Last; i++)
22509 +               aufs_iop_nogetattr[i].getattr = NULL;
22510 +
22511 +       /* First, initialize au_dfree */
22512 +       for (i = 0; i < AuCache_Last; i++) {    /* including hnotify */
22513 +               cp = au_dfree.cache + i;
22514 +               cp->cache = NULL;
22515 +               init_llist_head(&cp->llist);
22516 +       }
22517 +       for (i = 0; i < AU_DFREE_Last; i++)
22518 +               init_llist_head(au_dfree.llist + i);
22519 +       INIT_DELAYED_WORK(&au_dfree.dwork, au_do_dfree);
22520 +
22521 +       au_sbilist_init();
22522 +       sysaufs_brs_init();
22523 +       au_debug_init();
22524 +       au_dy_init();
22525 +       err = sysaufs_init();
22526 +       if (unlikely(err))
22527 +               goto out;
22528 +       err = au_procfs_init();
22529 +       if (unlikely(err))
22530 +               goto out_sysaufs;
22531 +       err = au_wkq_init();
22532 +       if (unlikely(err))
22533 +               goto out_procfs;
22534 +       err = au_loopback_init();
22535 +       if (unlikely(err))
22536 +               goto out_wkq;
22537 +       err = au_hnotify_init();
22538 +       if (unlikely(err))
22539 +               goto out_loopback;
22540 +       err = au_sysrq_init();
22541 +       if (unlikely(err))
22542 +               goto out_hin;
22543 +       err = au_cache_init();
22544 +       if (unlikely(err))
22545 +               goto out_sysrq;
22546 +
22547 +       aufs_fs_type.fs_flags |= au_userns ? FS_USERNS_MOUNT : 0;
22548 +       err = register_filesystem(&aufs_fs_type);
22549 +       if (unlikely(err))
22550 +               goto out_cache;
22551 +
22552 +       /* since we define pr_fmt, call printk directly */
22553 +       printk(KERN_INFO AUFS_NAME " " AUFS_VERSION "\n");
22554 +       goto out; /* success */
22555 +
22556 +out_cache:
22557 +       au_cache_fin();
22558 +out_sysrq:
22559 +       au_sysrq_fin();
22560 +out_hin:
22561 +       au_hnotify_fin();
22562 +out_loopback:
22563 +       au_loopback_fin();
22564 +out_wkq:
22565 +       au_wkq_fin();
22566 +out_procfs:
22567 +       au_procfs_fin();
22568 +out_sysaufs:
22569 +       sysaufs_fin();
22570 +       au_dy_fin();
22571 +       flush_delayed_work(&au_dfree.dwork);
22572 +out:
22573 +       return err;
22574 +}
22575 +
22576 +static void __exit aufs_exit(void)
22577 +{
22578 +       unregister_filesystem(&aufs_fs_type);
22579 +       au_cache_fin();
22580 +       au_sysrq_fin();
22581 +       au_hnotify_fin();
22582 +       au_loopback_fin();
22583 +       au_wkq_fin();
22584 +       au_procfs_fin();
22585 +       sysaufs_fin();
22586 +       au_dy_fin();
22587 +       flush_delayed_work(&au_dfree.dwork);
22588 +}
22589 +
22590 +module_init(aufs_init);
22591 +module_exit(aufs_exit);
22592 diff -urN /usr/share/empty/fs/aufs/module.h linux/fs/aufs/module.h
22593 --- /usr/share/empty/fs/aufs/module.h   1970-01-01 01:00:00.000000000 +0100
22594 +++ linux/fs/aufs/module.h      2017-05-06 22:16:52.781553511 +0200
22595 @@ -0,0 +1,156 @@
22596 +/*
22597 + * Copyright (C) 2005-2017 Junjiro R. Okajima
22598 + *
22599 + * This program, aufs is free software; you can redistribute it and/or modify
22600 + * it under the terms of the GNU General Public License as published by
22601 + * the Free Software Foundation; either version 2 of the License, or
22602 + * (at your option) any later version.
22603 + *
22604 + * This program is distributed in the hope that it will be useful,
22605 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22606 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22607 + * GNU General Public License for more details.
22608 + *
22609 + * You should have received a copy of the GNU General Public License
22610 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22611 + */
22612 +
22613 +/*
22614 + * module initialization and module-global
22615 + */
22616 +
22617 +#ifndef __AUFS_MODULE_H__
22618 +#define __AUFS_MODULE_H__
22619 +
22620 +#ifdef __KERNEL__
22621 +
22622 +#include <linux/slab.h>
22623 +#include "debug.h"
22624 +
22625 +struct path;
22626 +struct seq_file;
22627 +
22628 +/* module parameters */
22629 +extern int sysaufs_brs;
22630 +extern bool au_userns;
22631 +
22632 +/* ---------------------------------------------------------------------- */
22633 +
22634 +extern int au_dir_roflags;
22635 +
22636 +void *au_krealloc(void *p, unsigned int new_sz, gfp_t gfp, int may_shrink);
22637 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp,
22638 +                  int may_shrink);
22639 +
22640 +static inline int au_kmidx_sub(size_t sz, size_t new_sz)
22641 +{
22642 +#ifndef CONFIG_SLOB
22643 +       return kmalloc_index(sz) - kmalloc_index(new_sz);
22644 +#else
22645 +       return -1; /* SLOB is untested */
22646 +#endif
22647 +}
22648 +
22649 +int au_seq_path(struct seq_file *seq, struct path *path);
22650 +
22651 +#ifdef CONFIG_PROC_FS
22652 +/* procfs.c */
22653 +int __init au_procfs_init(void);
22654 +void au_procfs_fin(void);
22655 +#else
22656 +AuStubInt0(au_procfs_init, void);
22657 +AuStubVoid(au_procfs_fin, void);
22658 +#endif
22659 +
22660 +/* ---------------------------------------------------------------------- */
22661 +
22662 +/* kmem cache and delayed free */
22663 +enum {
22664 +       AuCache_DINFO,
22665 +       AuCache_ICNTNR,
22666 +       AuCache_FINFO,
22667 +       AuCache_VDIR,
22668 +       AuCache_DEHSTR,
22669 +       AuCache_HNOTIFY, /* must be last */
22670 +       AuCache_Last
22671 +};
22672 +
22673 +enum {
22674 +       AU_DFREE_KFREE,
22675 +       AU_DFREE_FREE_PAGE,
22676 +       AU_DFREE_Last
22677 +};
22678 +
22679 +struct au_cache {
22680 +       struct kmem_cache       *cache;
22681 +       struct llist_head       llist;  /* delayed free */
22682 +};
22683 +
22684 +/*
22685 + * in order to reduce the cost of the internal timer, consolidate all the
22686 + * delayed free works into a single delayed_work.
22687 + */
22688 +struct au_dfree {
22689 +       struct au_cache         cache[AuCache_Last];
22690 +       struct llist_head       llist[AU_DFREE_Last];
22691 +       struct delayed_work     dwork;
22692 +};
22693 +
22694 +extern struct au_dfree au_dfree;
22695 +
22696 +#define AuCacheFlags           (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD)
22697 +#define AuCache(type)          KMEM_CACHE(type, AuCacheFlags)
22698 +#define AuCacheCtor(type, ctor)        \
22699 +       kmem_cache_create(#type, sizeof(struct type), \
22700 +                         __alignof__(struct type), AuCacheFlags, ctor)
22701 +
22702 +#define AU_DFREE_DELAY         msecs_to_jiffies(10)
22703 +#define AU_DFREE_BODY(lnode, llist) do {                               \
22704 +               if (llist_add(lnode, llist))                            \
22705 +                       schedule_delayed_work(&au_dfree.dwork,          \
22706 +                                             AU_DFREE_DELAY);          \
22707 +       } while (0)
22708 +#define AU_CACHE_DFREE_FUNC(name, idx, lnode)                          \
22709 +       void au_cache_dfree_##name(struct au_##name *p)                 \
22710 +       {                                                               \
22711 +               struct au_cache *cp = au_dfree.cache + AuCache_##idx;   \
22712 +               AU_DFREE_BODY(&p->lnode, &cp->llist);                   \
22713 +       }
22714 +
22715 +#define AuCacheFuncs(name, index) \
22716 +static inline struct au_##name *au_cache_alloc_##name(void) \
22717 +{ return kmem_cache_alloc(au_dfree.cache[AuCache_##index].cache, GFP_NOFS); } \
22718 +static inline void au_cache_free_##name(struct au_##name *p) \
22719 +{ kmem_cache_free(au_dfree.cache[AuCache_##index].cache, p); } \
22720 +void au_cache_dfree_##name(struct au_##name *p)
22721 +
22722 +AuCacheFuncs(dinfo, DINFO);
22723 +AuCacheFuncs(icntnr, ICNTNR);
22724 +AuCacheFuncs(finfo, FINFO);
22725 +AuCacheFuncs(vdir, VDIR);
22726 +AuCacheFuncs(vdir_dehstr, DEHSTR);
22727 +#ifdef CONFIG_AUFS_HNOTIFY
22728 +AuCacheFuncs(hnotify, HNOTIFY);
22729 +#endif
22730 +
22731 +static inline void au_delayed_kfree(const void *p)
22732 +{
22733 +       AuDebugOn(!p);
22734 +       AuDebugOn(ksize(p) < sizeof(struct llist_node));
22735 +
22736 +       AU_DFREE_BODY((void *)p, au_dfree.llist + AU_DFREE_KFREE);
22737 +}
22738 +
22739 +/* cast only */
22740 +static inline void au_free_page(void *p)
22741 +{
22742 +       free_page((unsigned long)p);
22743 +}
22744 +
22745 +static inline void au_delayed_free_page(unsigned long addr)
22746 +{
22747 +       AU_DFREE_BODY((void *)addr, au_dfree.llist + AU_DFREE_FREE_PAGE);
22748 +}
22749 +
22750 +#endif /* __KERNEL__ */
22751 +#endif /* __AUFS_MODULE_H__ */
22752 diff -urN /usr/share/empty/fs/aufs/mvdown.c linux/fs/aufs/mvdown.c
22753 --- /usr/share/empty/fs/aufs/mvdown.c   1970-01-01 01:00:00.000000000 +0100
22754 +++ linux/fs/aufs/mvdown.c      2017-05-06 22:16:52.781553511 +0200
22755 @@ -0,0 +1,704 @@
22756 +/*
22757 + * Copyright (C) 2011-2017 Junjiro R. Okajima
22758 + *
22759 + * This program, aufs is free software; you can redistribute it and/or modify
22760 + * it under the terms of the GNU General Public License as published by
22761 + * the Free Software Foundation; either version 2 of the License, or
22762 + * (at your option) any later version.
22763 + *
22764 + * This program is distributed in the hope that it will be useful,
22765 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22766 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22767 + * GNU General Public License for more details.
22768 + *
22769 + * You should have received a copy of the GNU General Public License
22770 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22771 + */
22772 +
22773 +/*
22774 + * move-down, opposite of copy-up
22775 + */
22776 +
22777 +#include "aufs.h"
22778 +
22779 +struct au_mvd_args {
22780 +       struct {
22781 +               struct super_block *h_sb;
22782 +               struct dentry *h_parent;
22783 +               struct au_hinode *hdir;
22784 +               struct inode *h_dir, *h_inode;
22785 +               struct au_pin pin;
22786 +       } info[AUFS_MVDOWN_NARRAY];
22787 +
22788 +       struct aufs_mvdown mvdown;
22789 +       struct dentry *dentry, *parent;
22790 +       struct inode *inode, *dir;
22791 +       struct super_block *sb;
22792 +       aufs_bindex_t bopq, bwh, bfound;
22793 +       unsigned char rename_lock;
22794 +};
22795 +
22796 +#define mvd_errno              mvdown.au_errno
22797 +#define mvd_bsrc               mvdown.stbr[AUFS_MVDOWN_UPPER].bindex
22798 +#define mvd_src_brid           mvdown.stbr[AUFS_MVDOWN_UPPER].brid
22799 +#define mvd_bdst               mvdown.stbr[AUFS_MVDOWN_LOWER].bindex
22800 +#define mvd_dst_brid           mvdown.stbr[AUFS_MVDOWN_LOWER].brid
22801 +
22802 +#define mvd_h_src_sb           info[AUFS_MVDOWN_UPPER].h_sb
22803 +#define mvd_h_src_parent       info[AUFS_MVDOWN_UPPER].h_parent
22804 +#define mvd_hdir_src           info[AUFS_MVDOWN_UPPER].hdir
22805 +#define mvd_h_src_dir          info[AUFS_MVDOWN_UPPER].h_dir
22806 +#define mvd_h_src_inode                info[AUFS_MVDOWN_UPPER].h_inode
22807 +#define mvd_pin_src            info[AUFS_MVDOWN_UPPER].pin
22808 +
22809 +#define mvd_h_dst_sb           info[AUFS_MVDOWN_LOWER].h_sb
22810 +#define mvd_h_dst_parent       info[AUFS_MVDOWN_LOWER].h_parent
22811 +#define mvd_hdir_dst           info[AUFS_MVDOWN_LOWER].hdir
22812 +#define mvd_h_dst_dir          info[AUFS_MVDOWN_LOWER].h_dir
22813 +#define mvd_h_dst_inode                info[AUFS_MVDOWN_LOWER].h_inode
22814 +#define mvd_pin_dst            info[AUFS_MVDOWN_LOWER].pin
22815 +
22816 +#define AU_MVD_PR(flag, ...) do {                      \
22817 +               if (flag)                               \
22818 +                       pr_err(__VA_ARGS__);            \
22819 +       } while (0)
22820 +
22821 +static int find_lower_writable(struct au_mvd_args *a)
22822 +{
22823 +       struct super_block *sb;
22824 +       aufs_bindex_t bindex, bbot;
22825 +       struct au_branch *br;
22826 +
22827 +       sb = a->sb;
22828 +       bindex = a->mvd_bsrc;
22829 +       bbot = au_sbbot(sb);
22830 +       if (a->mvdown.flags & AUFS_MVDOWN_FHSM_LOWER)
22831 +               for (bindex++; bindex <= bbot; bindex++) {
22832 +                       br = au_sbr(sb, bindex);
22833 +                       if (au_br_fhsm(br->br_perm)
22834 +                           && (!(au_br_sb(br)->s_flags & MS_RDONLY)))
22835 +                               return bindex;
22836 +               }
22837 +       else if (!(a->mvdown.flags & AUFS_MVDOWN_ROLOWER))
22838 +               for (bindex++; bindex <= bbot; bindex++) {
22839 +                       br = au_sbr(sb, bindex);
22840 +                       if (!au_br_rdonly(br))
22841 +                               return bindex;
22842 +               }
22843 +       else
22844 +               for (bindex++; bindex <= bbot; bindex++) {
22845 +                       br = au_sbr(sb, bindex);
22846 +                       if (!(au_br_sb(br)->s_flags & MS_RDONLY)) {
22847 +                               if (au_br_rdonly(br))
22848 +                                       a->mvdown.flags
22849 +                                               |= AUFS_MVDOWN_ROLOWER_R;
22850 +                               return bindex;
22851 +                       }
22852 +               }
22853 +
22854 +       return -1;
22855 +}
22856 +
22857 +/* make the parent dir on bdst */
22858 +static int au_do_mkdir(const unsigned char dmsg, struct au_mvd_args *a)
22859 +{
22860 +       int err;
22861 +
22862 +       err = 0;
22863 +       a->mvd_hdir_src = au_hi(a->dir, a->mvd_bsrc);
22864 +       a->mvd_hdir_dst = au_hi(a->dir, a->mvd_bdst);
22865 +       a->mvd_h_src_parent = au_h_dptr(a->parent, a->mvd_bsrc);
22866 +       a->mvd_h_dst_parent = NULL;
22867 +       if (au_dbbot(a->parent) >= a->mvd_bdst)
22868 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
22869 +       if (!a->mvd_h_dst_parent) {
22870 +               err = au_cpdown_dirs(a->dentry, a->mvd_bdst);
22871 +               if (unlikely(err)) {
22872 +                       AU_MVD_PR(dmsg, "cpdown_dirs failed\n");
22873 +                       goto out;
22874 +               }
22875 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
22876 +       }
22877 +
22878 +out:
22879 +       AuTraceErr(err);
22880 +       return err;
22881 +}
22882 +
22883 +/* lock them all */
22884 +static int au_do_lock(const unsigned char dmsg, struct au_mvd_args *a)
22885 +{
22886 +       int err;
22887 +       struct dentry *h_trap;
22888 +
22889 +       a->mvd_h_src_sb = au_sbr_sb(a->sb, a->mvd_bsrc);
22890 +       a->mvd_h_dst_sb = au_sbr_sb(a->sb, a->mvd_bdst);
22891 +       err = au_pin(&a->mvd_pin_dst, a->dentry, a->mvd_bdst,
22892 +                    au_opt_udba(a->sb),
22893 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
22894 +       AuTraceErr(err);
22895 +       if (unlikely(err)) {
22896 +               AU_MVD_PR(dmsg, "pin_dst failed\n");
22897 +               goto out;
22898 +       }
22899 +
22900 +       if (a->mvd_h_src_sb != a->mvd_h_dst_sb) {
22901 +               a->rename_lock = 0;
22902 +               au_pin_init(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
22903 +                           AuLsc_DI_PARENT, AuLsc_I_PARENT3,
22904 +                           au_opt_udba(a->sb),
22905 +                           AuPin_MNT_WRITE | AuPin_DI_LOCKED);
22906 +               err = au_do_pin(&a->mvd_pin_src);
22907 +               AuTraceErr(err);
22908 +               a->mvd_h_src_dir = d_inode(a->mvd_h_src_parent);
22909 +               if (unlikely(err)) {
22910 +                       AU_MVD_PR(dmsg, "pin_src failed\n");
22911 +                       goto out_dst;
22912 +               }
22913 +               goto out; /* success */
22914 +       }
22915 +
22916 +       a->rename_lock = 1;
22917 +       au_pin_hdir_unlock(&a->mvd_pin_dst);
22918 +       err = au_pin(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
22919 +                    au_opt_udba(a->sb),
22920 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
22921 +       AuTraceErr(err);
22922 +       a->mvd_h_src_dir = d_inode(a->mvd_h_src_parent);
22923 +       if (unlikely(err)) {
22924 +               AU_MVD_PR(dmsg, "pin_src failed\n");
22925 +               au_pin_hdir_lock(&a->mvd_pin_dst);
22926 +               goto out_dst;
22927 +       }
22928 +       au_pin_hdir_unlock(&a->mvd_pin_src);
22929 +       h_trap = vfsub_lock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
22930 +                                  a->mvd_h_dst_parent, a->mvd_hdir_dst);
22931 +       if (h_trap) {
22932 +               err = (h_trap != a->mvd_h_src_parent);
22933 +               if (err)
22934 +                       err = (h_trap != a->mvd_h_dst_parent);
22935 +       }
22936 +       BUG_ON(err); /* it should never happen */
22937 +       if (unlikely(a->mvd_h_src_dir != au_pinned_h_dir(&a->mvd_pin_src))) {
22938 +               err = -EBUSY;
22939 +               AuTraceErr(err);
22940 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
22941 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
22942 +               au_pin_hdir_lock(&a->mvd_pin_src);
22943 +               au_unpin(&a->mvd_pin_src);
22944 +               au_pin_hdir_lock(&a->mvd_pin_dst);
22945 +               goto out_dst;
22946 +       }
22947 +       goto out; /* success */
22948 +
22949 +out_dst:
22950 +       au_unpin(&a->mvd_pin_dst);
22951 +out:
22952 +       AuTraceErr(err);
22953 +       return err;
22954 +}
22955 +
22956 +static void au_do_unlock(const unsigned char dmsg, struct au_mvd_args *a)
22957 +{
22958 +       if (!a->rename_lock)
22959 +               au_unpin(&a->mvd_pin_src);
22960 +       else {
22961 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
22962 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
22963 +               au_pin_hdir_lock(&a->mvd_pin_src);
22964 +               au_unpin(&a->mvd_pin_src);
22965 +               au_pin_hdir_lock(&a->mvd_pin_dst);
22966 +       }
22967 +       au_unpin(&a->mvd_pin_dst);
22968 +}
22969 +
22970 +/* copy-down the file */
22971 +static int au_do_cpdown(const unsigned char dmsg, struct au_mvd_args *a)
22972 +{
22973 +       int err;
22974 +       struct au_cp_generic cpg = {
22975 +               .dentry = a->dentry,
22976 +               .bdst   = a->mvd_bdst,
22977 +               .bsrc   = a->mvd_bsrc,
22978 +               .len    = -1,
22979 +               .pin    = &a->mvd_pin_dst,
22980 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
22981 +       };
22982 +
22983 +       AuDbg("b%d, b%d\n", cpg.bsrc, cpg.bdst);
22984 +       if (a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
22985 +               au_fset_cpup(cpg.flags, OVERWRITE);
22986 +       if (a->mvdown.flags & AUFS_MVDOWN_ROLOWER)
22987 +               au_fset_cpup(cpg.flags, RWDST);
22988 +       err = au_sio_cpdown_simple(&cpg);
22989 +       if (unlikely(err))
22990 +               AU_MVD_PR(dmsg, "cpdown failed\n");
22991 +
22992 +       AuTraceErr(err);
22993 +       return err;
22994 +}
22995 +
22996 +/*
22997 + * unlink the whiteout on bdst if exist which may be created by UDBA while we
22998 + * were sleeping
22999 + */
23000 +static int au_do_unlink_wh(const unsigned char dmsg, struct au_mvd_args *a)
23001 +{
23002 +       int err;
23003 +       struct path h_path;
23004 +       struct au_branch *br;
23005 +       struct inode *delegated;
23006 +
23007 +       br = au_sbr(a->sb, a->mvd_bdst);
23008 +       h_path.dentry = au_wh_lkup(a->mvd_h_dst_parent, &a->dentry->d_name, br);
23009 +       err = PTR_ERR(h_path.dentry);
23010 +       if (IS_ERR(h_path.dentry)) {
23011 +               AU_MVD_PR(dmsg, "wh_lkup failed\n");
23012 +               goto out;
23013 +       }
23014 +
23015 +       err = 0;
23016 +       if (d_is_positive(h_path.dentry)) {
23017 +               h_path.mnt = au_br_mnt(br);
23018 +               delegated = NULL;
23019 +               err = vfsub_unlink(d_inode(a->mvd_h_dst_parent), &h_path,
23020 +                                  &delegated, /*force*/0);
23021 +               if (unlikely(err == -EWOULDBLOCK)) {
23022 +                       pr_warn("cannot retry for NFSv4 delegation"
23023 +                               " for an internal unlink\n");
23024 +                       iput(delegated);
23025 +               }
23026 +               if (unlikely(err))
23027 +                       AU_MVD_PR(dmsg, "wh_unlink failed\n");
23028 +       }
23029 +       dput(h_path.dentry);
23030 +
23031 +out:
23032 +       AuTraceErr(err);
23033 +       return err;
23034 +}
23035 +
23036 +/*
23037 + * unlink the topmost h_dentry
23038 + */
23039 +static int au_do_unlink(const unsigned char dmsg, struct au_mvd_args *a)
23040 +{
23041 +       int err;
23042 +       struct path h_path;
23043 +       struct inode *delegated;
23044 +
23045 +       h_path.mnt = au_sbr_mnt(a->sb, a->mvd_bsrc);
23046 +       h_path.dentry = au_h_dptr(a->dentry, a->mvd_bsrc);
23047 +       delegated = NULL;
23048 +       err = vfsub_unlink(a->mvd_h_src_dir, &h_path, &delegated, /*force*/0);
23049 +       if (unlikely(err == -EWOULDBLOCK)) {
23050 +               pr_warn("cannot retry for NFSv4 delegation"
23051 +                       " for an internal unlink\n");
23052 +               iput(delegated);
23053 +       }
23054 +       if (unlikely(err))
23055 +               AU_MVD_PR(dmsg, "unlink failed\n");
23056 +
23057 +       AuTraceErr(err);
23058 +       return err;
23059 +}
23060 +
23061 +/* Since mvdown succeeded, we ignore an error of this function */
23062 +static void au_do_stfs(const unsigned char dmsg, struct au_mvd_args *a)
23063 +{
23064 +       int err;
23065 +       struct au_branch *br;
23066 +
23067 +       a->mvdown.flags |= AUFS_MVDOWN_STFS_FAILED;
23068 +       br = au_sbr(a->sb, a->mvd_bsrc);
23069 +       err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_UPPER].stfs);
23070 +       if (!err) {
23071 +               br = au_sbr(a->sb, a->mvd_bdst);
23072 +               a->mvdown.stbr[AUFS_MVDOWN_LOWER].brid = br->br_id;
23073 +               err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_LOWER].stfs);
23074 +       }
23075 +       if (!err)
23076 +               a->mvdown.flags &= ~AUFS_MVDOWN_STFS_FAILED;
23077 +       else
23078 +               AU_MVD_PR(dmsg, "statfs failed (%d), ignored\n", err);
23079 +}
23080 +
23081 +/*
23082 + * copy-down the file and unlink the bsrc file.
23083 + * - unlink the bdst whout if exist
23084 + * - copy-down the file (with whtmp name and rename)
23085 + * - unlink the bsrc file
23086 + */
23087 +static int au_do_mvdown(const unsigned char dmsg, struct au_mvd_args *a)
23088 +{
23089 +       int err;
23090 +
23091 +       err = au_do_mkdir(dmsg, a);
23092 +       if (!err)
23093 +               err = au_do_lock(dmsg, a);
23094 +       if (unlikely(err))
23095 +               goto out;
23096 +
23097 +       /*
23098 +        * do not revert the activities we made on bdst since they should be
23099 +        * harmless in aufs.
23100 +        */
23101 +
23102 +       err = au_do_cpdown(dmsg, a);
23103 +       if (!err)
23104 +               err = au_do_unlink_wh(dmsg, a);
23105 +       if (!err && !(a->mvdown.flags & AUFS_MVDOWN_KUPPER))
23106 +               err = au_do_unlink(dmsg, a);
23107 +       if (unlikely(err))
23108 +               goto out_unlock;
23109 +
23110 +       AuDbg("%pd2, 0x%x, %d --> %d\n",
23111 +             a->dentry, a->mvdown.flags, a->mvd_bsrc, a->mvd_bdst);
23112 +       if (find_lower_writable(a) < 0)
23113 +               a->mvdown.flags |= AUFS_MVDOWN_BOTTOM;
23114 +
23115 +       if (a->mvdown.flags & AUFS_MVDOWN_STFS)
23116 +               au_do_stfs(dmsg, a);
23117 +
23118 +       /* maintain internal array */
23119 +       if (!(a->mvdown.flags & AUFS_MVDOWN_KUPPER)) {
23120 +               au_set_h_dptr(a->dentry, a->mvd_bsrc, NULL);
23121 +               au_set_dbtop(a->dentry, a->mvd_bdst);
23122 +               au_set_h_iptr(a->inode, a->mvd_bsrc, NULL, /*flags*/0);
23123 +               au_set_ibtop(a->inode, a->mvd_bdst);
23124 +       } else {
23125 +               /* hide the lower */
23126 +               au_set_h_dptr(a->dentry, a->mvd_bdst, NULL);
23127 +               au_set_dbbot(a->dentry, a->mvd_bsrc);
23128 +               au_set_h_iptr(a->inode, a->mvd_bdst, NULL, /*flags*/0);
23129 +               au_set_ibbot(a->inode, a->mvd_bsrc);
23130 +       }
23131 +       if (au_dbbot(a->dentry) < a->mvd_bdst)
23132 +               au_set_dbbot(a->dentry, a->mvd_bdst);
23133 +       if (au_ibbot(a->inode) < a->mvd_bdst)
23134 +               au_set_ibbot(a->inode, a->mvd_bdst);
23135 +
23136 +out_unlock:
23137 +       au_do_unlock(dmsg, a);
23138 +out:
23139 +       AuTraceErr(err);
23140 +       return err;
23141 +}
23142 +
23143 +/* ---------------------------------------------------------------------- */
23144 +
23145 +/* make sure the file is idle */
23146 +static int au_mvd_args_busy(const unsigned char dmsg, struct au_mvd_args *a)
23147 +{
23148 +       int err, plinked;
23149 +
23150 +       err = 0;
23151 +       plinked = !!au_opt_test(au_mntflags(a->sb), PLINK);
23152 +       if (au_dbtop(a->dentry) == a->mvd_bsrc
23153 +           && au_dcount(a->dentry) == 1
23154 +           && atomic_read(&a->inode->i_count) == 1
23155 +           /* && a->mvd_h_src_inode->i_nlink == 1 */
23156 +           && (!plinked || !au_plink_test(a->inode))
23157 +           && a->inode->i_nlink == 1)
23158 +               goto out;
23159 +
23160 +       err = -EBUSY;
23161 +       AU_MVD_PR(dmsg,
23162 +                 "b%d, d{b%d, c%d?}, i{c%d?, l%u}, hi{l%u}, p{%d, %d}\n",
23163 +                 a->mvd_bsrc, au_dbtop(a->dentry), au_dcount(a->dentry),
23164 +                 atomic_read(&a->inode->i_count), a->inode->i_nlink,
23165 +                 a->mvd_h_src_inode->i_nlink,
23166 +                 plinked, plinked ? au_plink_test(a->inode) : 0);
23167 +
23168 +out:
23169 +       AuTraceErr(err);
23170 +       return err;
23171 +}
23172 +
23173 +/* make sure the parent dir is fine */
23174 +static int au_mvd_args_parent(const unsigned char dmsg,
23175 +                             struct au_mvd_args *a)
23176 +{
23177 +       int err;
23178 +       aufs_bindex_t bindex;
23179 +
23180 +       err = 0;
23181 +       if (unlikely(au_alive_dir(a->parent))) {
23182 +               err = -ENOENT;
23183 +               AU_MVD_PR(dmsg, "parent dir is dead\n");
23184 +               goto out;
23185 +       }
23186 +
23187 +       a->bopq = au_dbdiropq(a->parent);
23188 +       bindex = au_wbr_nonopq(a->dentry, a->mvd_bdst);
23189 +       AuDbg("b%d\n", bindex);
23190 +       if (unlikely((bindex >= 0 && bindex < a->mvd_bdst)
23191 +                    || (a->bopq != -1 && a->bopq < a->mvd_bdst))) {
23192 +               err = -EINVAL;
23193 +               a->mvd_errno = EAU_MVDOWN_OPAQUE;
23194 +               AU_MVD_PR(dmsg, "ancestor is opaque b%d, b%d\n",
23195 +                         a->bopq, a->mvd_bdst);
23196 +       }
23197 +
23198 +out:
23199 +       AuTraceErr(err);
23200 +       return err;
23201 +}
23202 +
23203 +static int au_mvd_args_intermediate(const unsigned char dmsg,
23204 +                                   struct au_mvd_args *a)
23205 +{
23206 +       int err;
23207 +       struct au_dinfo *dinfo, *tmp;
23208 +
23209 +       /* lookup the next lower positive entry */
23210 +       err = -ENOMEM;
23211 +       tmp = au_di_alloc(a->sb, AuLsc_DI_TMP);
23212 +       if (unlikely(!tmp))
23213 +               goto out;
23214 +
23215 +       a->bfound = -1;
23216 +       a->bwh = -1;
23217 +       dinfo = au_di(a->dentry);
23218 +       au_di_cp(tmp, dinfo);
23219 +       au_di_swap(tmp, dinfo);
23220 +
23221 +       /* returns the number of positive dentries */
23222 +       err = au_lkup_dentry(a->dentry, a->mvd_bsrc + 1,
23223 +                            /* AuLkup_IGNORE_PERM */ 0);
23224 +       if (!err)
23225 +               a->bwh = au_dbwh(a->dentry);
23226 +       else if (err > 0)
23227 +               a->bfound = au_dbtop(a->dentry);
23228 +
23229 +       au_di_swap(tmp, dinfo);
23230 +       au_rw_write_unlock(&tmp->di_rwsem);
23231 +       au_di_free(tmp);
23232 +       if (unlikely(err < 0))
23233 +               AU_MVD_PR(dmsg, "failed look-up lower\n");
23234 +
23235 +       /*
23236 +        * here, we have these cases.
23237 +        * bfound == -1
23238 +        *      no positive dentry under bsrc. there are more sub-cases.
23239 +        *      bwh < 0
23240 +        *              there no whiteout, we can safely move-down.
23241 +        *      bwh <= bsrc
23242 +        *              impossible
23243 +        *      bsrc < bwh && bwh < bdst
23244 +        *              there is a whiteout on RO branch. cannot proceed.
23245 +        *      bwh == bdst
23246 +        *              there is a whiteout on the RW target branch. it should
23247 +        *              be removed.
23248 +        *      bdst < bwh
23249 +        *              there is a whiteout somewhere unrelated branch.
23250 +        * -1 < bfound && bfound <= bsrc
23251 +        *      impossible.
23252 +        * bfound < bdst
23253 +        *      found, but it is on RO branch between bsrc and bdst. cannot
23254 +        *      proceed.
23255 +        * bfound == bdst
23256 +        *      found, replace it if AUFS_MVDOWN_FORCE is set. otherwise return
23257 +        *      error.
23258 +        * bdst < bfound
23259 +        *      found, after we create the file on bdst, it will be hidden.
23260 +        */
23261 +
23262 +       AuDebugOn(a->bfound == -1
23263 +                 && a->bwh != -1
23264 +                 && a->bwh <= a->mvd_bsrc);
23265 +       AuDebugOn(-1 < a->bfound
23266 +                 && a->bfound <= a->mvd_bsrc);
23267 +
23268 +       err = -EINVAL;
23269 +       if (a->bfound == -1
23270 +           && a->mvd_bsrc < a->bwh
23271 +           && a->bwh != -1
23272 +           && a->bwh < a->mvd_bdst) {
23273 +               a->mvd_errno = EAU_MVDOWN_WHITEOUT;
23274 +               AU_MVD_PR(dmsg, "bsrc %d, bdst %d, bfound %d, bwh %d\n",
23275 +                         a->mvd_bsrc, a->mvd_bdst, a->bfound, a->bwh);
23276 +               goto out;
23277 +       } else if (a->bfound != -1 && a->bfound < a->mvd_bdst) {
23278 +               a->mvd_errno = EAU_MVDOWN_UPPER;
23279 +               AU_MVD_PR(dmsg, "bdst %d, bfound %d\n",
23280 +                         a->mvd_bdst, a->bfound);
23281 +               goto out;
23282 +       }
23283 +
23284 +       err = 0; /* success */
23285 +
23286 +out:
23287 +       AuTraceErr(err);
23288 +       return err;
23289 +}
23290 +
23291 +static int au_mvd_args_exist(const unsigned char dmsg, struct au_mvd_args *a)
23292 +{
23293 +       int err;
23294 +
23295 +       err = 0;
23296 +       if (!(a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
23297 +           && a->bfound == a->mvd_bdst)
23298 +               err = -EEXIST;
23299 +       AuTraceErr(err);
23300 +       return err;
23301 +}
23302 +
23303 +static int au_mvd_args(const unsigned char dmsg, struct au_mvd_args *a)
23304 +{
23305 +       int err;
23306 +       struct au_branch *br;
23307 +
23308 +       err = -EISDIR;
23309 +       if (unlikely(S_ISDIR(a->inode->i_mode)))
23310 +               goto out;
23311 +
23312 +       err = -EINVAL;
23313 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_UPPER))
23314 +               a->mvd_bsrc = au_ibtop(a->inode);
23315 +       else {
23316 +               a->mvd_bsrc = au_br_index(a->sb, a->mvd_src_brid);
23317 +               if (unlikely(a->mvd_bsrc < 0
23318 +                            || (a->mvd_bsrc < au_dbtop(a->dentry)
23319 +                                || au_dbbot(a->dentry) < a->mvd_bsrc
23320 +                                || !au_h_dptr(a->dentry, a->mvd_bsrc))
23321 +                            || (a->mvd_bsrc < au_ibtop(a->inode)
23322 +                                || au_ibbot(a->inode) < a->mvd_bsrc
23323 +                                || !au_h_iptr(a->inode, a->mvd_bsrc)))) {
23324 +                       a->mvd_errno = EAU_MVDOWN_NOUPPER;
23325 +                       AU_MVD_PR(dmsg, "no upper\n");
23326 +                       goto out;
23327 +               }
23328 +       }
23329 +       if (unlikely(a->mvd_bsrc == au_sbbot(a->sb))) {
23330 +               a->mvd_errno = EAU_MVDOWN_BOTTOM;
23331 +               AU_MVD_PR(dmsg, "on the bottom\n");
23332 +               goto out;
23333 +       }
23334 +       a->mvd_h_src_inode = au_h_iptr(a->inode, a->mvd_bsrc);
23335 +       br = au_sbr(a->sb, a->mvd_bsrc);
23336 +       err = au_br_rdonly(br);
23337 +       if (!(a->mvdown.flags & AUFS_MVDOWN_ROUPPER)) {
23338 +               if (unlikely(err))
23339 +                       goto out;
23340 +       } else if (!(vfsub_native_ro(a->mvd_h_src_inode)
23341 +                    || IS_APPEND(a->mvd_h_src_inode))) {
23342 +               if (err)
23343 +                       a->mvdown.flags |= AUFS_MVDOWN_ROUPPER_R;
23344 +               /* go on */
23345 +       } else
23346 +               goto out;
23347 +
23348 +       err = -EINVAL;
23349 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_LOWER)) {
23350 +               a->mvd_bdst = find_lower_writable(a);
23351 +               if (unlikely(a->mvd_bdst < 0)) {
23352 +                       a->mvd_errno = EAU_MVDOWN_BOTTOM;
23353 +                       AU_MVD_PR(dmsg, "no writable lower branch\n");
23354 +                       goto out;
23355 +               }
23356 +       } else {
23357 +               a->mvd_bdst = au_br_index(a->sb, a->mvd_dst_brid);
23358 +               if (unlikely(a->mvd_bdst < 0
23359 +                            || au_sbbot(a->sb) < a->mvd_bdst)) {
23360 +                       a->mvd_errno = EAU_MVDOWN_NOLOWERBR;
23361 +                       AU_MVD_PR(dmsg, "no lower brid\n");
23362 +                       goto out;
23363 +               }
23364 +       }
23365 +
23366 +       err = au_mvd_args_busy(dmsg, a);
23367 +       if (!err)
23368 +               err = au_mvd_args_parent(dmsg, a);
23369 +       if (!err)
23370 +               err = au_mvd_args_intermediate(dmsg, a);
23371 +       if (!err)
23372 +               err = au_mvd_args_exist(dmsg, a);
23373 +       if (!err)
23374 +               AuDbg("b%d, b%d\n", a->mvd_bsrc, a->mvd_bdst);
23375 +
23376 +out:
23377 +       AuTraceErr(err);
23378 +       return err;
23379 +}
23380 +
23381 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *uarg)
23382 +{
23383 +       int err, e;
23384 +       unsigned char dmsg;
23385 +       struct au_mvd_args *args;
23386 +       struct inode *inode;
23387 +
23388 +       inode = d_inode(dentry);
23389 +       err = -EPERM;
23390 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
23391 +               goto out;
23392 +
23393 +       err = -ENOMEM;
23394 +       args = kmalloc(sizeof(*args), GFP_NOFS);
23395 +       if (unlikely(!args))
23396 +               goto out;
23397 +
23398 +       err = copy_from_user(&args->mvdown, uarg, sizeof(args->mvdown));
23399 +       if (!err)
23400 +               err = !access_ok(VERIFY_WRITE, uarg, sizeof(*uarg));
23401 +       if (unlikely(err)) {
23402 +               err = -EFAULT;
23403 +               AuTraceErr(err);
23404 +               goto out_free;
23405 +       }
23406 +       AuDbg("flags 0x%x\n", args->mvdown.flags);
23407 +       args->mvdown.flags &= ~(AUFS_MVDOWN_ROLOWER_R | AUFS_MVDOWN_ROUPPER_R);
23408 +       args->mvdown.au_errno = 0;
23409 +       args->dentry = dentry;
23410 +       args->inode = inode;
23411 +       args->sb = dentry->d_sb;
23412 +
23413 +       err = -ENOENT;
23414 +       dmsg = !!(args->mvdown.flags & AUFS_MVDOWN_DMSG);
23415 +       args->parent = dget_parent(dentry);
23416 +       args->dir = d_inode(args->parent);
23417 +       inode_lock_nested(args->dir, I_MUTEX_PARENT);
23418 +       dput(args->parent);
23419 +       if (unlikely(args->parent != dentry->d_parent)) {
23420 +               AU_MVD_PR(dmsg, "parent dir is moved\n");
23421 +               goto out_dir;
23422 +       }
23423 +
23424 +       inode_lock_nested(inode, I_MUTEX_CHILD);
23425 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_NOPLMW);
23426 +       if (unlikely(err))
23427 +               goto out_inode;
23428 +
23429 +       di_write_lock_parent(args->parent);
23430 +       err = au_mvd_args(dmsg, args);
23431 +       if (unlikely(err))
23432 +               goto out_parent;
23433 +
23434 +       err = au_do_mvdown(dmsg, args);
23435 +       if (unlikely(err))
23436 +               goto out_parent;
23437 +
23438 +       au_cpup_attr_timesizes(args->dir);
23439 +       au_cpup_attr_timesizes(inode);
23440 +       if (!(args->mvdown.flags & AUFS_MVDOWN_KUPPER))
23441 +               au_cpup_igen(inode, au_h_iptr(inode, args->mvd_bdst));
23442 +       /* au_digen_dec(dentry); */
23443 +
23444 +out_parent:
23445 +       di_write_unlock(args->parent);
23446 +       aufs_read_unlock(dentry, AuLock_DW);
23447 +out_inode:
23448 +       inode_unlock(inode);
23449 +out_dir:
23450 +       inode_unlock(args->dir);
23451 +out_free:
23452 +       e = copy_to_user(uarg, &args->mvdown, sizeof(args->mvdown));
23453 +       if (unlikely(e))
23454 +               err = -EFAULT;
23455 +       au_delayed_kfree(args);
23456 +out:
23457 +       AuTraceErr(err);
23458 +       return err;
23459 +}
23460 diff -urN /usr/share/empty/fs/aufs/opts.c linux/fs/aufs/opts.c
23461 --- /usr/share/empty/fs/aufs/opts.c     1970-01-01 01:00:00.000000000 +0100
23462 +++ linux/fs/aufs/opts.c        2017-05-06 22:16:52.781553511 +0200
23463 @@ -0,0 +1,1848 @@
23464 +/*
23465 + * Copyright (C) 2005-2017 Junjiro R. Okajima
23466 + *
23467 + * This program, aufs is free software; you can redistribute it and/or modify
23468 + * it under the terms of the GNU General Public License as published by
23469 + * the Free Software Foundation; either version 2 of the License, or
23470 + * (at your option) any later version.
23471 + *
23472 + * This program is distributed in the hope that it will be useful,
23473 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
23474 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
23475 + * GNU General Public License for more details.
23476 + *
23477 + * You should have received a copy of the GNU General Public License
23478 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
23479 + */
23480 +
23481 +/*
23482 + * mount options/flags
23483 + */
23484 +
23485 +#include <linux/namei.h>
23486 +#include <linux/types.h> /* a distribution requires */
23487 +#include <linux/parser.h>
23488 +#include "aufs.h"
23489 +
23490 +/* ---------------------------------------------------------------------- */
23491 +
23492 +enum {
23493 +       Opt_br,
23494 +       Opt_add, Opt_del, Opt_mod, Opt_append, Opt_prepend,
23495 +       Opt_idel, Opt_imod,
23496 +       Opt_dirwh, Opt_rdcache, Opt_rdblk, Opt_rdhash,
23497 +       Opt_rdblk_def, Opt_rdhash_def,
23498 +       Opt_xino, Opt_noxino,
23499 +       Opt_trunc_xino, Opt_trunc_xino_v, Opt_notrunc_xino,
23500 +       Opt_trunc_xino_path, Opt_itrunc_xino,
23501 +       Opt_trunc_xib, Opt_notrunc_xib,
23502 +       Opt_shwh, Opt_noshwh,
23503 +       Opt_plink, Opt_noplink, Opt_list_plink,
23504 +       Opt_udba,
23505 +       Opt_dio, Opt_nodio,
23506 +       Opt_diropq_a, Opt_diropq_w,
23507 +       Opt_warn_perm, Opt_nowarn_perm,
23508 +       Opt_wbr_copyup, Opt_wbr_create,
23509 +       Opt_fhsm_sec,
23510 +       Opt_verbose, Opt_noverbose,
23511 +       Opt_sum, Opt_nosum, Opt_wsum,
23512 +       Opt_dirperm1, Opt_nodirperm1,
23513 +       Opt_acl, Opt_noacl,
23514 +       Opt_tail, Opt_ignore, Opt_ignore_silent, Opt_err
23515 +};
23516 +
23517 +static match_table_t options = {
23518 +       {Opt_br, "br=%s"},
23519 +       {Opt_br, "br:%s"},
23520 +
23521 +       {Opt_add, "add=%d:%s"},
23522 +       {Opt_add, "add:%d:%s"},
23523 +       {Opt_add, "ins=%d:%s"},
23524 +       {Opt_add, "ins:%d:%s"},
23525 +       {Opt_append, "append=%s"},
23526 +       {Opt_append, "append:%s"},
23527 +       {Opt_prepend, "prepend=%s"},
23528 +       {Opt_prepend, "prepend:%s"},
23529 +
23530 +       {Opt_del, "del=%s"},
23531 +       {Opt_del, "del:%s"},
23532 +       /* {Opt_idel, "idel:%d"}, */
23533 +       {Opt_mod, "mod=%s"},
23534 +       {Opt_mod, "mod:%s"},
23535 +       /* {Opt_imod, "imod:%d:%s"}, */
23536 +
23537 +       {Opt_dirwh, "dirwh=%d"},
23538 +
23539 +       {Opt_xino, "xino=%s"},
23540 +       {Opt_noxino, "noxino"},
23541 +       {Opt_trunc_xino, "trunc_xino"},
23542 +       {Opt_trunc_xino_v, "trunc_xino_v=%d:%d"},
23543 +       {Opt_notrunc_xino, "notrunc_xino"},
23544 +       {Opt_trunc_xino_path, "trunc_xino=%s"},
23545 +       {Opt_itrunc_xino, "itrunc_xino=%d"},
23546 +       /* {Opt_zxino, "zxino=%s"}, */
23547 +       {Opt_trunc_xib, "trunc_xib"},
23548 +       {Opt_notrunc_xib, "notrunc_xib"},
23549 +
23550 +#ifdef CONFIG_PROC_FS
23551 +       {Opt_plink, "plink"},
23552 +#else
23553 +       {Opt_ignore_silent, "plink"},
23554 +#endif
23555 +
23556 +       {Opt_noplink, "noplink"},
23557 +
23558 +#ifdef CONFIG_AUFS_DEBUG
23559 +       {Opt_list_plink, "list_plink"},
23560 +#endif
23561 +
23562 +       {Opt_udba, "udba=%s"},
23563 +
23564 +       {Opt_dio, "dio"},
23565 +       {Opt_nodio, "nodio"},
23566 +
23567 +#ifdef CONFIG_AUFS_FHSM
23568 +       {Opt_fhsm_sec, "fhsm_sec=%d"},
23569 +#else
23570 +       {Opt_ignore_silent, "fhsm_sec=%d"},
23571 +#endif
23572 +
23573 +       {Opt_diropq_a, "diropq=always"},
23574 +       {Opt_diropq_a, "diropq=a"},
23575 +       {Opt_diropq_w, "diropq=whiteouted"},
23576 +       {Opt_diropq_w, "diropq=w"},
23577 +
23578 +       {Opt_warn_perm, "warn_perm"},
23579 +       {Opt_nowarn_perm, "nowarn_perm"},
23580 +
23581 +       /* keep them temporary */
23582 +       {Opt_ignore_silent, "nodlgt"},
23583 +       {Opt_ignore_silent, "clean_plink"},
23584 +
23585 +#ifdef CONFIG_AUFS_SHWH
23586 +       {Opt_shwh, "shwh"},
23587 +#endif
23588 +       {Opt_noshwh, "noshwh"},
23589 +
23590 +       {Opt_dirperm1, "dirperm1"},
23591 +       {Opt_nodirperm1, "nodirperm1"},
23592 +
23593 +       {Opt_verbose, "verbose"},
23594 +       {Opt_verbose, "v"},
23595 +       {Opt_noverbose, "noverbose"},
23596 +       {Opt_noverbose, "quiet"},
23597 +       {Opt_noverbose, "q"},
23598 +       {Opt_noverbose, "silent"},
23599 +
23600 +       {Opt_sum, "sum"},
23601 +       {Opt_nosum, "nosum"},
23602 +       {Opt_wsum, "wsum"},
23603 +
23604 +       {Opt_rdcache, "rdcache=%d"},
23605 +       {Opt_rdblk, "rdblk=%d"},
23606 +       {Opt_rdblk_def, "rdblk=def"},
23607 +       {Opt_rdhash, "rdhash=%d"},
23608 +       {Opt_rdhash_def, "rdhash=def"},
23609 +
23610 +       {Opt_wbr_create, "create=%s"},
23611 +       {Opt_wbr_create, "create_policy=%s"},
23612 +       {Opt_wbr_copyup, "cpup=%s"},
23613 +       {Opt_wbr_copyup, "copyup=%s"},
23614 +       {Opt_wbr_copyup, "copyup_policy=%s"},
23615 +
23616 +       /* generic VFS flag */
23617 +#ifdef CONFIG_FS_POSIX_ACL
23618 +       {Opt_acl, "acl"},
23619 +       {Opt_noacl, "noacl"},
23620 +#else
23621 +       {Opt_ignore_silent, "acl"},
23622 +       {Opt_ignore_silent, "noacl"},
23623 +#endif
23624 +
23625 +       /* internal use for the scripts */
23626 +       {Opt_ignore_silent, "si=%s"},
23627 +
23628 +       {Opt_br, "dirs=%s"},
23629 +       {Opt_ignore, "debug=%d"},
23630 +       {Opt_ignore, "delete=whiteout"},
23631 +       {Opt_ignore, "delete=all"},
23632 +       {Opt_ignore, "imap=%s"},
23633 +
23634 +       /* temporary workaround, due to old mount(8)? */
23635 +       {Opt_ignore_silent, "relatime"},
23636 +
23637 +       {Opt_err, NULL}
23638 +};
23639 +
23640 +/* ---------------------------------------------------------------------- */
23641 +
23642 +static const char *au_parser_pattern(int val, match_table_t tbl)
23643 +{
23644 +       struct match_token *p;
23645 +
23646 +       p = tbl;
23647 +       while (p->pattern) {
23648 +               if (p->token == val)
23649 +                       return p->pattern;
23650 +               p++;
23651 +       }
23652 +       BUG();
23653 +       return "??";
23654 +}
23655 +
23656 +static const char *au_optstr(int *val, match_table_t tbl)
23657 +{
23658 +       struct match_token *p;
23659 +       int v;
23660 +
23661 +       v = *val;
23662 +       if (!v)
23663 +               goto out;
23664 +       p = tbl;
23665 +       while (p->pattern) {
23666 +               if (p->token
23667 +                   && (v & p->token) == p->token) {
23668 +                       *val &= ~p->token;
23669 +                       return p->pattern;
23670 +               }
23671 +               p++;
23672 +       }
23673 +
23674 +out:
23675 +       return NULL;
23676 +}
23677 +
23678 +/* ---------------------------------------------------------------------- */
23679 +
23680 +static match_table_t brperm = {
23681 +       {AuBrPerm_RO, AUFS_BRPERM_RO},
23682 +       {AuBrPerm_RR, AUFS_BRPERM_RR},
23683 +       {AuBrPerm_RW, AUFS_BRPERM_RW},
23684 +       {0, NULL}
23685 +};
23686 +
23687 +static match_table_t brattr = {
23688 +       /* general */
23689 +       {AuBrAttr_COO_REG, AUFS_BRATTR_COO_REG},
23690 +       {AuBrAttr_COO_ALL, AUFS_BRATTR_COO_ALL},
23691 +       /* 'unpin' attrib is meaningless since linux-3.18-rc1 */
23692 +       {AuBrAttr_UNPIN, AUFS_BRATTR_UNPIN},
23693 +#ifdef CONFIG_AUFS_FHSM
23694 +       {AuBrAttr_FHSM, AUFS_BRATTR_FHSM},
23695 +#endif
23696 +#ifdef CONFIG_AUFS_XATTR
23697 +       {AuBrAttr_ICEX, AUFS_BRATTR_ICEX},
23698 +       {AuBrAttr_ICEX_SEC, AUFS_BRATTR_ICEX_SEC},
23699 +       {AuBrAttr_ICEX_SYS, AUFS_BRATTR_ICEX_SYS},
23700 +       {AuBrAttr_ICEX_TR, AUFS_BRATTR_ICEX_TR},
23701 +       {AuBrAttr_ICEX_USR, AUFS_BRATTR_ICEX_USR},
23702 +       {AuBrAttr_ICEX_OTH, AUFS_BRATTR_ICEX_OTH},
23703 +#endif
23704 +
23705 +       /* ro/rr branch */
23706 +       {AuBrRAttr_WH, AUFS_BRRATTR_WH},
23707 +
23708 +       /* rw branch */
23709 +       {AuBrWAttr_MOO, AUFS_BRWATTR_MOO},
23710 +       {AuBrWAttr_NoLinkWH, AUFS_BRWATTR_NLWH},
23711 +
23712 +       {0, NULL}
23713 +};
23714 +
23715 +static int br_attr_val(char *str, match_table_t table, substring_t args[])
23716 +{
23717 +       int attr, v;
23718 +       char *p;
23719 +
23720 +       attr = 0;
23721 +       do {
23722 +               p = strchr(str, '+');
23723 +               if (p)
23724 +                       *p = 0;
23725 +               v = match_token(str, table, args);
23726 +               if (v) {
23727 +                       if (v & AuBrAttr_CMOO_Mask)
23728 +                               attr &= ~AuBrAttr_CMOO_Mask;
23729 +                       attr |= v;
23730 +               } else {
23731 +                       if (p)
23732 +                               *p = '+';
23733 +                       pr_warn("ignored branch attribute %s\n", str);
23734 +                       break;
23735 +               }
23736 +               if (p)
23737 +                       str = p + 1;
23738 +       } while (p);
23739 +
23740 +       return attr;
23741 +}
23742 +
23743 +static int au_do_optstr_br_attr(au_br_perm_str_t *str, int perm)
23744 +{
23745 +       int sz;
23746 +       const char *p;
23747 +       char *q;
23748 +
23749 +       q = str->a;
23750 +       *q = 0;
23751 +       p = au_optstr(&perm, brattr);
23752 +       if (p) {
23753 +               sz = strlen(p);
23754 +               memcpy(q, p, sz + 1);
23755 +               q += sz;
23756 +       } else
23757 +               goto out;
23758 +
23759 +       do {
23760 +               p = au_optstr(&perm, brattr);
23761 +               if (p) {
23762 +                       *q++ = '+';
23763 +                       sz = strlen(p);
23764 +                       memcpy(q, p, sz + 1);
23765 +                       q += sz;
23766 +               }
23767 +       } while (p);
23768 +
23769 +out:
23770 +       return q - str->a;
23771 +}
23772 +
23773 +static int noinline_for_stack br_perm_val(char *perm)
23774 +{
23775 +       int val, bad, sz;
23776 +       char *p;
23777 +       substring_t args[MAX_OPT_ARGS];
23778 +       au_br_perm_str_t attr;
23779 +
23780 +       p = strchr(perm, '+');
23781 +       if (p)
23782 +               *p = 0;
23783 +       val = match_token(perm, brperm, args);
23784 +       if (!val) {
23785 +               if (p)
23786 +                       *p = '+';
23787 +               pr_warn("ignored branch permission %s\n", perm);
23788 +               val = AuBrPerm_RO;
23789 +               goto out;
23790 +       }
23791 +       if (!p)
23792 +               goto out;
23793 +
23794 +       val |= br_attr_val(p + 1, brattr, args);
23795 +
23796 +       bad = 0;
23797 +       switch (val & AuBrPerm_Mask) {
23798 +       case AuBrPerm_RO:
23799 +       case AuBrPerm_RR:
23800 +               bad = val & AuBrWAttr_Mask;
23801 +               val &= ~AuBrWAttr_Mask;
23802 +               break;
23803 +       case AuBrPerm_RW:
23804 +               bad = val & AuBrRAttr_Mask;
23805 +               val &= ~AuBrRAttr_Mask;
23806 +               break;
23807 +       }
23808 +
23809 +       /*
23810 +        * 'unpin' attrib becomes meaningless since linux-3.18-rc1, but aufs
23811 +        * does not treat it as an error, just warning.
23812 +        * this is a tiny guard for the user operation.
23813 +        */
23814 +       if (val & AuBrAttr_UNPIN) {
23815 +               bad |= AuBrAttr_UNPIN;
23816 +               val &= ~AuBrAttr_UNPIN;
23817 +       }
23818 +
23819 +       if (unlikely(bad)) {
23820 +               sz = au_do_optstr_br_attr(&attr, bad);
23821 +               AuDebugOn(!sz);
23822 +               pr_warn("ignored branch attribute %s\n", attr.a);
23823 +       }
23824 +
23825 +out:
23826 +       return val;
23827 +}
23828 +
23829 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm)
23830 +{
23831 +       au_br_perm_str_t attr;
23832 +       const char *p;
23833 +       char *q;
23834 +       int sz;
23835 +
23836 +       q = str->a;
23837 +       p = au_optstr(&perm, brperm);
23838 +       AuDebugOn(!p || !*p);
23839 +       sz = strlen(p);
23840 +       memcpy(q, p, sz + 1);
23841 +       q += sz;
23842 +
23843 +       sz = au_do_optstr_br_attr(&attr, perm);
23844 +       if (sz) {
23845 +               *q++ = '+';
23846 +               memcpy(q, attr.a, sz + 1);
23847 +       }
23848 +
23849 +       AuDebugOn(strlen(str->a) >= sizeof(str->a));
23850 +}
23851 +
23852 +/* ---------------------------------------------------------------------- */
23853 +
23854 +static match_table_t udbalevel = {
23855 +       {AuOpt_UDBA_REVAL, "reval"},
23856 +       {AuOpt_UDBA_NONE, "none"},
23857 +#ifdef CONFIG_AUFS_HNOTIFY
23858 +       {AuOpt_UDBA_HNOTIFY, "notify"}, /* abstraction */
23859 +#ifdef CONFIG_AUFS_HFSNOTIFY
23860 +       {AuOpt_UDBA_HNOTIFY, "fsnotify"},
23861 +#endif
23862 +#endif
23863 +       {-1, NULL}
23864 +};
23865 +
23866 +static int noinline_for_stack udba_val(char *str)
23867 +{
23868 +       substring_t args[MAX_OPT_ARGS];
23869 +
23870 +       return match_token(str, udbalevel, args);
23871 +}
23872 +
23873 +const char *au_optstr_udba(int udba)
23874 +{
23875 +       return au_parser_pattern(udba, udbalevel);
23876 +}
23877 +
23878 +/* ---------------------------------------------------------------------- */
23879 +
23880 +static match_table_t au_wbr_create_policy = {
23881 +       {AuWbrCreate_TDP, "tdp"},
23882 +       {AuWbrCreate_TDP, "top-down-parent"},
23883 +       {AuWbrCreate_RR, "rr"},
23884 +       {AuWbrCreate_RR, "round-robin"},
23885 +       {AuWbrCreate_MFS, "mfs"},
23886 +       {AuWbrCreate_MFS, "most-free-space"},
23887 +       {AuWbrCreate_MFSV, "mfs:%d"},
23888 +       {AuWbrCreate_MFSV, "most-free-space:%d"},
23889 +
23890 +       /* top-down regardless the parent, and then mfs */
23891 +       {AuWbrCreate_TDMFS, "tdmfs:%d"},
23892 +       {AuWbrCreate_TDMFSV, "tdmfs:%d:%d"},
23893 +
23894 +       {AuWbrCreate_MFSRR, "mfsrr:%d"},
23895 +       {AuWbrCreate_MFSRRV, "mfsrr:%d:%d"},
23896 +       {AuWbrCreate_PMFS, "pmfs"},
23897 +       {AuWbrCreate_PMFSV, "pmfs:%d"},
23898 +       {AuWbrCreate_PMFSRR, "pmfsrr:%d"},
23899 +       {AuWbrCreate_PMFSRRV, "pmfsrr:%d:%d"},
23900 +
23901 +       {-1, NULL}
23902 +};
23903 +
23904 +static int au_wbr_mfs_wmark(substring_t *arg, char *str,
23905 +                           struct au_opt_wbr_create *create)
23906 +{
23907 +       int err;
23908 +       unsigned long long ull;
23909 +
23910 +       err = 0;
23911 +       if (!match_u64(arg, &ull))
23912 +               create->mfsrr_watermark = ull;
23913 +       else {
23914 +               pr_err("bad integer in %s\n", str);
23915 +               err = -EINVAL;
23916 +       }
23917 +
23918 +       return err;
23919 +}
23920 +
23921 +static int au_wbr_mfs_sec(substring_t *arg, char *str,
23922 +                         struct au_opt_wbr_create *create)
23923 +{
23924 +       int n, err;
23925 +
23926 +       err = 0;
23927 +       if (!match_int(arg, &n) && 0 <= n && n <= AUFS_MFS_MAX_SEC)
23928 +               create->mfs_second = n;
23929 +       else {
23930 +               pr_err("bad integer in %s\n", str);
23931 +               err = -EINVAL;
23932 +       }
23933 +
23934 +       return err;
23935 +}
23936 +
23937 +static int noinline_for_stack
23938 +au_wbr_create_val(char *str, struct au_opt_wbr_create *create)
23939 +{
23940 +       int err, e;
23941 +       substring_t args[MAX_OPT_ARGS];
23942 +
23943 +       err = match_token(str, au_wbr_create_policy, args);
23944 +       create->wbr_create = err;
23945 +       switch (err) {
23946 +       case AuWbrCreate_MFSRRV:
23947 +       case AuWbrCreate_TDMFSV:
23948 +       case AuWbrCreate_PMFSRRV:
23949 +               e = au_wbr_mfs_wmark(&args[0], str, create);
23950 +               if (!e)
23951 +                       e = au_wbr_mfs_sec(&args[1], str, create);
23952 +               if (unlikely(e))
23953 +                       err = e;
23954 +               break;
23955 +       case AuWbrCreate_MFSRR:
23956 +       case AuWbrCreate_TDMFS:
23957 +       case AuWbrCreate_PMFSRR:
23958 +               e = au_wbr_mfs_wmark(&args[0], str, create);
23959 +               if (unlikely(e)) {
23960 +                       err = e;
23961 +                       break;
23962 +               }
23963 +               /*FALLTHROUGH*/
23964 +       case AuWbrCreate_MFS:
23965 +       case AuWbrCreate_PMFS:
23966 +               create->mfs_second = AUFS_MFS_DEF_SEC;
23967 +               break;
23968 +       case AuWbrCreate_MFSV:
23969 +       case AuWbrCreate_PMFSV:
23970 +               e = au_wbr_mfs_sec(&args[0], str, create);
23971 +               if (unlikely(e))
23972 +                       err = e;
23973 +               break;
23974 +       }
23975 +
23976 +       return err;
23977 +}
23978 +
23979 +const char *au_optstr_wbr_create(int wbr_create)
23980 +{
23981 +       return au_parser_pattern(wbr_create, au_wbr_create_policy);
23982 +}
23983 +
23984 +static match_table_t au_wbr_copyup_policy = {
23985 +       {AuWbrCopyup_TDP, "tdp"},
23986 +       {AuWbrCopyup_TDP, "top-down-parent"},
23987 +       {AuWbrCopyup_BUP, "bup"},
23988 +       {AuWbrCopyup_BUP, "bottom-up-parent"},
23989 +       {AuWbrCopyup_BU, "bu"},
23990 +       {AuWbrCopyup_BU, "bottom-up"},
23991 +       {-1, NULL}
23992 +};
23993 +
23994 +static int noinline_for_stack au_wbr_copyup_val(char *str)
23995 +{
23996 +       substring_t args[MAX_OPT_ARGS];
23997 +
23998 +       return match_token(str, au_wbr_copyup_policy, args);
23999 +}
24000 +
24001 +const char *au_optstr_wbr_copyup(int wbr_copyup)
24002 +{
24003 +       return au_parser_pattern(wbr_copyup, au_wbr_copyup_policy);
24004 +}
24005 +
24006 +/* ---------------------------------------------------------------------- */
24007 +
24008 +static const int lkup_dirflags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
24009 +
24010 +static void dump_opts(struct au_opts *opts)
24011 +{
24012 +#ifdef CONFIG_AUFS_DEBUG
24013 +       /* reduce stack space */
24014 +       union {
24015 +               struct au_opt_add *add;
24016 +               struct au_opt_del *del;
24017 +               struct au_opt_mod *mod;
24018 +               struct au_opt_xino *xino;
24019 +               struct au_opt_xino_itrunc *xino_itrunc;
24020 +               struct au_opt_wbr_create *create;
24021 +       } u;
24022 +       struct au_opt *opt;
24023 +
24024 +       opt = opts->opt;
24025 +       while (opt->type != Opt_tail) {
24026 +               switch (opt->type) {
24027 +               case Opt_add:
24028 +                       u.add = &opt->add;
24029 +                       AuDbg("add {b%d, %s, 0x%x, %p}\n",
24030 +                                 u.add->bindex, u.add->pathname, u.add->perm,
24031 +                                 u.add->path.dentry);
24032 +                       break;
24033 +               case Opt_del:
24034 +               case Opt_idel:
24035 +                       u.del = &opt->del;
24036 +                       AuDbg("del {%s, %p}\n",
24037 +                             u.del->pathname, u.del->h_path.dentry);
24038 +                       break;
24039 +               case Opt_mod:
24040 +               case Opt_imod:
24041 +                       u.mod = &opt->mod;
24042 +                       AuDbg("mod {%s, 0x%x, %p}\n",
24043 +                                 u.mod->path, u.mod->perm, u.mod->h_root);
24044 +                       break;
24045 +               case Opt_append:
24046 +                       u.add = &opt->add;
24047 +                       AuDbg("append {b%d, %s, 0x%x, %p}\n",
24048 +                                 u.add->bindex, u.add->pathname, u.add->perm,
24049 +                                 u.add->path.dentry);
24050 +                       break;
24051 +               case Opt_prepend:
24052 +                       u.add = &opt->add;
24053 +                       AuDbg("prepend {b%d, %s, 0x%x, %p}\n",
24054 +                                 u.add->bindex, u.add->pathname, u.add->perm,
24055 +                                 u.add->path.dentry);
24056 +                       break;
24057 +               case Opt_dirwh:
24058 +                       AuDbg("dirwh %d\n", opt->dirwh);
24059 +                       break;
24060 +               case Opt_rdcache:
24061 +                       AuDbg("rdcache %d\n", opt->rdcache);
24062 +                       break;
24063 +               case Opt_rdblk:
24064 +                       AuDbg("rdblk %u\n", opt->rdblk);
24065 +                       break;
24066 +               case Opt_rdblk_def:
24067 +                       AuDbg("rdblk_def\n");
24068 +                       break;
24069 +               case Opt_rdhash:
24070 +                       AuDbg("rdhash %u\n", opt->rdhash);
24071 +                       break;
24072 +               case Opt_rdhash_def:
24073 +                       AuDbg("rdhash_def\n");
24074 +                       break;
24075 +               case Opt_xino:
24076 +                       u.xino = &opt->xino;
24077 +                       AuDbg("xino {%s %pD}\n", u.xino->path, u.xino->file);
24078 +                       break;
24079 +               case Opt_trunc_xino:
24080 +                       AuLabel(trunc_xino);
24081 +                       break;
24082 +               case Opt_notrunc_xino:
24083 +                       AuLabel(notrunc_xino);
24084 +                       break;
24085 +               case Opt_trunc_xino_path:
24086 +               case Opt_itrunc_xino:
24087 +                       u.xino_itrunc = &opt->xino_itrunc;
24088 +                       AuDbg("trunc_xino %d\n", u.xino_itrunc->bindex);
24089 +                       break;
24090 +               case Opt_noxino:
24091 +                       AuLabel(noxino);
24092 +                       break;
24093 +               case Opt_trunc_xib:
24094 +                       AuLabel(trunc_xib);
24095 +                       break;
24096 +               case Opt_notrunc_xib:
24097 +                       AuLabel(notrunc_xib);
24098 +                       break;
24099 +               case Opt_shwh:
24100 +                       AuLabel(shwh);
24101 +                       break;
24102 +               case Opt_noshwh:
24103 +                       AuLabel(noshwh);
24104 +                       break;
24105 +               case Opt_dirperm1:
24106 +                       AuLabel(dirperm1);
24107 +                       break;
24108 +               case Opt_nodirperm1:
24109 +                       AuLabel(nodirperm1);
24110 +                       break;
24111 +               case Opt_plink:
24112 +                       AuLabel(plink);
24113 +                       break;
24114 +               case Opt_noplink:
24115 +                       AuLabel(noplink);
24116 +                       break;
24117 +               case Opt_list_plink:
24118 +                       AuLabel(list_plink);
24119 +                       break;
24120 +               case Opt_udba:
24121 +                       AuDbg("udba %d, %s\n",
24122 +                                 opt->udba, au_optstr_udba(opt->udba));
24123 +                       break;
24124 +               case Opt_dio:
24125 +                       AuLabel(dio);
24126 +                       break;
24127 +               case Opt_nodio:
24128 +                       AuLabel(nodio);
24129 +                       break;
24130 +               case Opt_diropq_a:
24131 +                       AuLabel(diropq_a);
24132 +                       break;
24133 +               case Opt_diropq_w:
24134 +                       AuLabel(diropq_w);
24135 +                       break;
24136 +               case Opt_warn_perm:
24137 +                       AuLabel(warn_perm);
24138 +                       break;
24139 +               case Opt_nowarn_perm:
24140 +                       AuLabel(nowarn_perm);
24141 +                       break;
24142 +               case Opt_verbose:
24143 +                       AuLabel(verbose);
24144 +                       break;
24145 +               case Opt_noverbose:
24146 +                       AuLabel(noverbose);
24147 +                       break;
24148 +               case Opt_sum:
24149 +                       AuLabel(sum);
24150 +                       break;
24151 +               case Opt_nosum:
24152 +                       AuLabel(nosum);
24153 +                       break;
24154 +               case Opt_wsum:
24155 +                       AuLabel(wsum);
24156 +                       break;
24157 +               case Opt_wbr_create:
24158 +                       u.create = &opt->wbr_create;
24159 +                       AuDbg("create %d, %s\n", u.create->wbr_create,
24160 +                                 au_optstr_wbr_create(u.create->wbr_create));
24161 +                       switch (u.create->wbr_create) {
24162 +                       case AuWbrCreate_MFSV:
24163 +                       case AuWbrCreate_PMFSV:
24164 +                               AuDbg("%d sec\n", u.create->mfs_second);
24165 +                               break;
24166 +                       case AuWbrCreate_MFSRR:
24167 +                       case AuWbrCreate_TDMFS:
24168 +                               AuDbg("%llu watermark\n",
24169 +                                         u.create->mfsrr_watermark);
24170 +                               break;
24171 +                       case AuWbrCreate_MFSRRV:
24172 +                       case AuWbrCreate_TDMFSV:
24173 +                       case AuWbrCreate_PMFSRRV:
24174 +                               AuDbg("%llu watermark, %d sec\n",
24175 +                                         u.create->mfsrr_watermark,
24176 +                                         u.create->mfs_second);
24177 +                               break;
24178 +                       }
24179 +                       break;
24180 +               case Opt_wbr_copyup:
24181 +                       AuDbg("copyup %d, %s\n", opt->wbr_copyup,
24182 +                                 au_optstr_wbr_copyup(opt->wbr_copyup));
24183 +                       break;
24184 +               case Opt_fhsm_sec:
24185 +                       AuDbg("fhsm_sec %u\n", opt->fhsm_second);
24186 +                       break;
24187 +               case Opt_acl:
24188 +                       AuLabel(acl);
24189 +                       break;
24190 +               case Opt_noacl:
24191 +                       AuLabel(noacl);
24192 +                       break;
24193 +               default:
24194 +                       BUG();
24195 +               }
24196 +               opt++;
24197 +       }
24198 +#endif
24199 +}
24200 +
24201 +void au_opts_free(struct au_opts *opts)
24202 +{
24203 +       struct au_opt *opt;
24204 +
24205 +       opt = opts->opt;
24206 +       while (opt->type != Opt_tail) {
24207 +               switch (opt->type) {
24208 +               case Opt_add:
24209 +               case Opt_append:
24210 +               case Opt_prepend:
24211 +                       path_put(&opt->add.path);
24212 +                       break;
24213 +               case Opt_del:
24214 +               case Opt_idel:
24215 +                       path_put(&opt->del.h_path);
24216 +                       break;
24217 +               case Opt_mod:
24218 +               case Opt_imod:
24219 +                       dput(opt->mod.h_root);
24220 +                       break;
24221 +               case Opt_xino:
24222 +                       fput(opt->xino.file);
24223 +                       break;
24224 +               }
24225 +               opt++;
24226 +       }
24227 +}
24228 +
24229 +static int opt_add(struct au_opt *opt, char *opt_str, unsigned long sb_flags,
24230 +                  aufs_bindex_t bindex)
24231 +{
24232 +       int err;
24233 +       struct au_opt_add *add = &opt->add;
24234 +       char *p;
24235 +
24236 +       add->bindex = bindex;
24237 +       add->perm = AuBrPerm_RO;
24238 +       add->pathname = opt_str;
24239 +       p = strchr(opt_str, '=');
24240 +       if (p) {
24241 +               *p++ = 0;
24242 +               if (*p)
24243 +                       add->perm = br_perm_val(p);
24244 +       }
24245 +
24246 +       err = vfsub_kern_path(add->pathname, lkup_dirflags, &add->path);
24247 +       if (!err) {
24248 +               if (!p) {
24249 +                       add->perm = AuBrPerm_RO;
24250 +                       if (au_test_fs_rr(add->path.dentry->d_sb))
24251 +                               add->perm = AuBrPerm_RR;
24252 +                       else if (!bindex && !(sb_flags & MS_RDONLY))
24253 +                               add->perm = AuBrPerm_RW;
24254 +               }
24255 +               opt->type = Opt_add;
24256 +               goto out;
24257 +       }
24258 +       pr_err("lookup failed %s (%d)\n", add->pathname, err);
24259 +       err = -EINVAL;
24260 +
24261 +out:
24262 +       return err;
24263 +}
24264 +
24265 +static int au_opts_parse_del(struct au_opt_del *del, substring_t args[])
24266 +{
24267 +       int err;
24268 +
24269 +       del->pathname = args[0].from;
24270 +       AuDbg("del path %s\n", del->pathname);
24271 +
24272 +       err = vfsub_kern_path(del->pathname, lkup_dirflags, &del->h_path);
24273 +       if (unlikely(err))
24274 +               pr_err("lookup failed %s (%d)\n", del->pathname, err);
24275 +
24276 +       return err;
24277 +}
24278 +
24279 +#if 0 /* reserved for future use */
24280 +static int au_opts_parse_idel(struct super_block *sb, aufs_bindex_t bindex,
24281 +                             struct au_opt_del *del, substring_t args[])
24282 +{
24283 +       int err;
24284 +       struct dentry *root;
24285 +
24286 +       err = -EINVAL;
24287 +       root = sb->s_root;
24288 +       aufs_read_lock(root, AuLock_FLUSH);
24289 +       if (bindex < 0 || au_sbbot(sb) < bindex) {
24290 +               pr_err("out of bounds, %d\n", bindex);
24291 +               goto out;
24292 +       }
24293 +
24294 +       err = 0;
24295 +       del->h_path.dentry = dget(au_h_dptr(root, bindex));
24296 +       del->h_path.mnt = mntget(au_sbr_mnt(sb, bindex));
24297 +
24298 +out:
24299 +       aufs_read_unlock(root, !AuLock_IR);
24300 +       return err;
24301 +}
24302 +#endif
24303 +
24304 +static int noinline_for_stack
24305 +au_opts_parse_mod(struct au_opt_mod *mod, substring_t args[])
24306 +{
24307 +       int err;
24308 +       struct path path;
24309 +       char *p;
24310 +
24311 +       err = -EINVAL;
24312 +       mod->path = args[0].from;
24313 +       p = strchr(mod->path, '=');
24314 +       if (unlikely(!p)) {
24315 +               pr_err("no permssion %s\n", args[0].from);
24316 +               goto out;
24317 +       }
24318 +
24319 +       *p++ = 0;
24320 +       err = vfsub_kern_path(mod->path, lkup_dirflags, &path);
24321 +       if (unlikely(err)) {
24322 +               pr_err("lookup failed %s (%d)\n", mod->path, err);
24323 +               goto out;
24324 +       }
24325 +
24326 +       mod->perm = br_perm_val(p);
24327 +       AuDbg("mod path %s, perm 0x%x, %s\n", mod->path, mod->perm, p);
24328 +       mod->h_root = dget(path.dentry);
24329 +       path_put(&path);
24330 +
24331 +out:
24332 +       return err;
24333 +}
24334 +
24335 +#if 0 /* reserved for future use */
24336 +static int au_opts_parse_imod(struct super_block *sb, aufs_bindex_t bindex,
24337 +                             struct au_opt_mod *mod, substring_t args[])
24338 +{
24339 +       int err;
24340 +       struct dentry *root;
24341 +
24342 +       err = -EINVAL;
24343 +       root = sb->s_root;
24344 +       aufs_read_lock(root, AuLock_FLUSH);
24345 +       if (bindex < 0 || au_sbbot(sb) < bindex) {
24346 +               pr_err("out of bounds, %d\n", bindex);
24347 +               goto out;
24348 +       }
24349 +
24350 +       err = 0;
24351 +       mod->perm = br_perm_val(args[1].from);
24352 +       AuDbg("mod path %s, perm 0x%x, %s\n",
24353 +             mod->path, mod->perm, args[1].from);
24354 +       mod->h_root = dget(au_h_dptr(root, bindex));
24355 +
24356 +out:
24357 +       aufs_read_unlock(root, !AuLock_IR);
24358 +       return err;
24359 +}
24360 +#endif
24361 +
24362 +static int au_opts_parse_xino(struct super_block *sb, struct au_opt_xino *xino,
24363 +                             substring_t args[])
24364 +{
24365 +       int err;
24366 +       struct file *file;
24367 +
24368 +       file = au_xino_create(sb, args[0].from, /*silent*/0);
24369 +       err = PTR_ERR(file);
24370 +       if (IS_ERR(file))
24371 +               goto out;
24372 +
24373 +       err = -EINVAL;
24374 +       if (unlikely(file->f_path.dentry->d_sb == sb)) {
24375 +               fput(file);
24376 +               pr_err("%s must be outside\n", args[0].from);
24377 +               goto out;
24378 +       }
24379 +
24380 +       err = 0;
24381 +       xino->file = file;
24382 +       xino->path = args[0].from;
24383 +
24384 +out:
24385 +       return err;
24386 +}
24387 +
24388 +static int noinline_for_stack
24389 +au_opts_parse_xino_itrunc_path(struct super_block *sb,
24390 +                              struct au_opt_xino_itrunc *xino_itrunc,
24391 +                              substring_t args[])
24392 +{
24393 +       int err;
24394 +       aufs_bindex_t bbot, bindex;
24395 +       struct path path;
24396 +       struct dentry *root;
24397 +
24398 +       err = vfsub_kern_path(args[0].from, lkup_dirflags, &path);
24399 +       if (unlikely(err)) {
24400 +               pr_err("lookup failed %s (%d)\n", args[0].from, err);
24401 +               goto out;
24402 +       }
24403 +
24404 +       xino_itrunc->bindex = -1;
24405 +       root = sb->s_root;
24406 +       aufs_read_lock(root, AuLock_FLUSH);
24407 +       bbot = au_sbbot(sb);
24408 +       for (bindex = 0; bindex <= bbot; bindex++) {
24409 +               if (au_h_dptr(root, bindex) == path.dentry) {
24410 +                       xino_itrunc->bindex = bindex;
24411 +                       break;
24412 +               }
24413 +       }
24414 +       aufs_read_unlock(root, !AuLock_IR);
24415 +       path_put(&path);
24416 +
24417 +       if (unlikely(xino_itrunc->bindex < 0)) {
24418 +               pr_err("no such branch %s\n", args[0].from);
24419 +               err = -EINVAL;
24420 +       }
24421 +
24422 +out:
24423 +       return err;
24424 +}
24425 +
24426 +/* called without aufs lock */
24427 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
24428 +{
24429 +       int err, n, token;
24430 +       aufs_bindex_t bindex;
24431 +       unsigned char skipped;
24432 +       struct dentry *root;
24433 +       struct au_opt *opt, *opt_tail;
24434 +       char *opt_str;
24435 +       /* reduce the stack space */
24436 +       union {
24437 +               struct au_opt_xino_itrunc *xino_itrunc;
24438 +               struct au_opt_wbr_create *create;
24439 +       } u;
24440 +       struct {
24441 +               substring_t args[MAX_OPT_ARGS];
24442 +       } *a;
24443 +
24444 +       err = -ENOMEM;
24445 +       a = kmalloc(sizeof(*a), GFP_NOFS);
24446 +       if (unlikely(!a))
24447 +               goto out;
24448 +
24449 +       root = sb->s_root;
24450 +       err = 0;
24451 +       bindex = 0;
24452 +       opt = opts->opt;
24453 +       opt_tail = opt + opts->max_opt - 1;
24454 +       opt->type = Opt_tail;
24455 +       while (!err && (opt_str = strsep(&str, ",")) && *opt_str) {
24456 +               err = -EINVAL;
24457 +               skipped = 0;
24458 +               token = match_token(opt_str, options, a->args);
24459 +               switch (token) {
24460 +               case Opt_br:
24461 +                       err = 0;
24462 +                       while (!err && (opt_str = strsep(&a->args[0].from, ":"))
24463 +                              && *opt_str) {
24464 +                               err = opt_add(opt, opt_str, opts->sb_flags,
24465 +                                             bindex++);
24466 +                               if (unlikely(!err && ++opt > opt_tail)) {
24467 +                                       err = -E2BIG;
24468 +                                       break;
24469 +                               }
24470 +                               opt->type = Opt_tail;
24471 +                               skipped = 1;
24472 +                       }
24473 +                       break;
24474 +               case Opt_add:
24475 +                       if (unlikely(match_int(&a->args[0], &n))) {
24476 +                               pr_err("bad integer in %s\n", opt_str);
24477 +                               break;
24478 +                       }
24479 +                       bindex = n;
24480 +                       err = opt_add(opt, a->args[1].from, opts->sb_flags,
24481 +                                     bindex);
24482 +                       if (!err)
24483 +                               opt->type = token;
24484 +                       break;
24485 +               case Opt_append:
24486 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
24487 +                                     /*dummy bindex*/1);
24488 +                       if (!err)
24489 +                               opt->type = token;
24490 +                       break;
24491 +               case Opt_prepend:
24492 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
24493 +                                     /*bindex*/0);
24494 +                       if (!err)
24495 +                               opt->type = token;
24496 +                       break;
24497 +               case Opt_del:
24498 +                       err = au_opts_parse_del(&opt->del, a->args);
24499 +                       if (!err)
24500 +                               opt->type = token;
24501 +                       break;
24502 +#if 0 /* reserved for future use */
24503 +               case Opt_idel:
24504 +                       del->pathname = "(indexed)";
24505 +                       if (unlikely(match_int(&args[0], &n))) {
24506 +                               pr_err("bad integer in %s\n", opt_str);
24507 +                               break;
24508 +                       }
24509 +                       err = au_opts_parse_idel(sb, n, &opt->del, a->args);
24510 +                       if (!err)
24511 +                               opt->type = token;
24512 +                       break;
24513 +#endif
24514 +               case Opt_mod:
24515 +                       err = au_opts_parse_mod(&opt->mod, a->args);
24516 +                       if (!err)
24517 +                               opt->type = token;
24518 +                       break;
24519 +#ifdef IMOD /* reserved for future use */
24520 +               case Opt_imod:
24521 +                       u.mod->path = "(indexed)";
24522 +                       if (unlikely(match_int(&a->args[0], &n))) {
24523 +                               pr_err("bad integer in %s\n", opt_str);
24524 +                               break;
24525 +                       }
24526 +                       err = au_opts_parse_imod(sb, n, &opt->mod, a->args);
24527 +                       if (!err)
24528 +                               opt->type = token;
24529 +                       break;
24530 +#endif
24531 +               case Opt_xino:
24532 +                       err = au_opts_parse_xino(sb, &opt->xino, a->args);
24533 +                       if (!err)
24534 +                               opt->type = token;
24535 +                       break;
24536 +
24537 +               case Opt_trunc_xino_path:
24538 +                       err = au_opts_parse_xino_itrunc_path
24539 +                               (sb, &opt->xino_itrunc, a->args);
24540 +                       if (!err)
24541 +                               opt->type = token;
24542 +                       break;
24543 +
24544 +               case Opt_itrunc_xino:
24545 +                       u.xino_itrunc = &opt->xino_itrunc;
24546 +                       if (unlikely(match_int(&a->args[0], &n))) {
24547 +                               pr_err("bad integer in %s\n", opt_str);
24548 +                               break;
24549 +                       }
24550 +                       u.xino_itrunc->bindex = n;
24551 +                       aufs_read_lock(root, AuLock_FLUSH);
24552 +                       if (n < 0 || au_sbbot(sb) < n) {
24553 +                               pr_err("out of bounds, %d\n", n);
24554 +                               aufs_read_unlock(root, !AuLock_IR);
24555 +                               break;
24556 +                       }
24557 +                       aufs_read_unlock(root, !AuLock_IR);
24558 +                       err = 0;
24559 +                       opt->type = token;
24560 +                       break;
24561 +
24562 +               case Opt_dirwh:
24563 +                       if (unlikely(match_int(&a->args[0], &opt->dirwh)))
24564 +                               break;
24565 +                       err = 0;
24566 +                       opt->type = token;
24567 +                       break;
24568 +
24569 +               case Opt_rdcache:
24570 +                       if (unlikely(match_int(&a->args[0], &n))) {
24571 +                               pr_err("bad integer in %s\n", opt_str);
24572 +                               break;
24573 +                       }
24574 +                       if (unlikely(n > AUFS_RDCACHE_MAX)) {
24575 +                               pr_err("rdcache must be smaller than %d\n",
24576 +                                      AUFS_RDCACHE_MAX);
24577 +                               break;
24578 +                       }
24579 +                       opt->rdcache = n;
24580 +                       err = 0;
24581 +                       opt->type = token;
24582 +                       break;
24583 +               case Opt_rdblk:
24584 +                       if (unlikely(match_int(&a->args[0], &n)
24585 +                                    || n < 0
24586 +                                    || n > KMALLOC_MAX_SIZE)) {
24587 +                               pr_err("bad integer in %s\n", opt_str);
24588 +                               break;
24589 +                       }
24590 +                       if (unlikely(n && n < NAME_MAX)) {
24591 +                               pr_err("rdblk must be larger than %d\n",
24592 +                                      NAME_MAX);
24593 +                               break;
24594 +                       }
24595 +                       opt->rdblk = n;
24596 +                       err = 0;
24597 +                       opt->type = token;
24598 +                       break;
24599 +               case Opt_rdhash:
24600 +                       if (unlikely(match_int(&a->args[0], &n)
24601 +                                    || n < 0
24602 +                                    || n * sizeof(struct hlist_head)
24603 +                                    > KMALLOC_MAX_SIZE)) {
24604 +                               pr_err("bad integer in %s\n", opt_str);
24605 +                               break;
24606 +                       }
24607 +                       opt->rdhash = n;
24608 +                       err = 0;
24609 +                       opt->type = token;
24610 +                       break;
24611 +
24612 +               case Opt_trunc_xino:
24613 +               case Opt_notrunc_xino:
24614 +               case Opt_noxino:
24615 +               case Opt_trunc_xib:
24616 +               case Opt_notrunc_xib:
24617 +               case Opt_shwh:
24618 +               case Opt_noshwh:
24619 +               case Opt_dirperm1:
24620 +               case Opt_nodirperm1:
24621 +               case Opt_plink:
24622 +               case Opt_noplink:
24623 +               case Opt_list_plink:
24624 +               case Opt_dio:
24625 +               case Opt_nodio:
24626 +               case Opt_diropq_a:
24627 +               case Opt_diropq_w:
24628 +               case Opt_warn_perm:
24629 +               case Opt_nowarn_perm:
24630 +               case Opt_verbose:
24631 +               case Opt_noverbose:
24632 +               case Opt_sum:
24633 +               case Opt_nosum:
24634 +               case Opt_wsum:
24635 +               case Opt_rdblk_def:
24636 +               case Opt_rdhash_def:
24637 +               case Opt_acl:
24638 +               case Opt_noacl:
24639 +                       err = 0;
24640 +                       opt->type = token;
24641 +                       break;
24642 +
24643 +               case Opt_udba:
24644 +                       opt->udba = udba_val(a->args[0].from);
24645 +                       if (opt->udba >= 0) {
24646 +                               err = 0;
24647 +                               opt->type = token;
24648 +                       } else
24649 +                               pr_err("wrong value, %s\n", opt_str);
24650 +                       break;
24651 +
24652 +               case Opt_wbr_create:
24653 +                       u.create = &opt->wbr_create;
24654 +                       u.create->wbr_create
24655 +                               = au_wbr_create_val(a->args[0].from, u.create);
24656 +                       if (u.create->wbr_create >= 0) {
24657 +                               err = 0;
24658 +                               opt->type = token;
24659 +                       } else
24660 +                               pr_err("wrong value, %s\n", opt_str);
24661 +                       break;
24662 +               case Opt_wbr_copyup:
24663 +                       opt->wbr_copyup = au_wbr_copyup_val(a->args[0].from);
24664 +                       if (opt->wbr_copyup >= 0) {
24665 +                               err = 0;
24666 +                               opt->type = token;
24667 +                       } else
24668 +                               pr_err("wrong value, %s\n", opt_str);
24669 +                       break;
24670 +
24671 +               case Opt_fhsm_sec:
24672 +                       if (unlikely(match_int(&a->args[0], &n)
24673 +                                    || n < 0)) {
24674 +                               pr_err("bad integer in %s\n", opt_str);
24675 +                               break;
24676 +                       }
24677 +                       if (sysaufs_brs) {
24678 +                               opt->fhsm_second = n;
24679 +                               opt->type = token;
24680 +                       } else
24681 +                               pr_warn("ignored %s\n", opt_str);
24682 +                       err = 0;
24683 +                       break;
24684 +
24685 +               case Opt_ignore:
24686 +                       pr_warn("ignored %s\n", opt_str);
24687 +                       /*FALLTHROUGH*/
24688 +               case Opt_ignore_silent:
24689 +                       skipped = 1;
24690 +                       err = 0;
24691 +                       break;
24692 +               case Opt_err:
24693 +                       pr_err("unknown option %s\n", opt_str);
24694 +                       break;
24695 +               }
24696 +
24697 +               if (!err && !skipped) {
24698 +                       if (unlikely(++opt > opt_tail)) {
24699 +                               err = -E2BIG;
24700 +                               opt--;
24701 +                               opt->type = Opt_tail;
24702 +                               break;
24703 +                       }
24704 +                       opt->type = Opt_tail;
24705 +               }
24706 +       }
24707 +
24708 +       au_delayed_kfree(a);
24709 +       dump_opts(opts);
24710 +       if (unlikely(err))
24711 +               au_opts_free(opts);
24712 +
24713 +out:
24714 +       return err;
24715 +}
24716 +
24717 +static int au_opt_wbr_create(struct super_block *sb,
24718 +                            struct au_opt_wbr_create *create)
24719 +{
24720 +       int err;
24721 +       struct au_sbinfo *sbinfo;
24722 +
24723 +       SiMustWriteLock(sb);
24724 +
24725 +       err = 1; /* handled */
24726 +       sbinfo = au_sbi(sb);
24727 +       if (sbinfo->si_wbr_create_ops->fin) {
24728 +               err = sbinfo->si_wbr_create_ops->fin(sb);
24729 +               if (!err)
24730 +                       err = 1;
24731 +       }
24732 +
24733 +       sbinfo->si_wbr_create = create->wbr_create;
24734 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + create->wbr_create;
24735 +       switch (create->wbr_create) {
24736 +       case AuWbrCreate_MFSRRV:
24737 +       case AuWbrCreate_MFSRR:
24738 +       case AuWbrCreate_TDMFS:
24739 +       case AuWbrCreate_TDMFSV:
24740 +       case AuWbrCreate_PMFSRR:
24741 +       case AuWbrCreate_PMFSRRV:
24742 +               sbinfo->si_wbr_mfs.mfsrr_watermark = create->mfsrr_watermark;
24743 +               /*FALLTHROUGH*/
24744 +       case AuWbrCreate_MFS:
24745 +       case AuWbrCreate_MFSV:
24746 +       case AuWbrCreate_PMFS:
24747 +       case AuWbrCreate_PMFSV:
24748 +               sbinfo->si_wbr_mfs.mfs_expire
24749 +                       = msecs_to_jiffies(create->mfs_second * MSEC_PER_SEC);
24750 +               break;
24751 +       }
24752 +
24753 +       if (sbinfo->si_wbr_create_ops->init)
24754 +               sbinfo->si_wbr_create_ops->init(sb); /* ignore */
24755 +
24756 +       return err;
24757 +}
24758 +
24759 +/*
24760 + * returns,
24761 + * plus: processed without an error
24762 + * zero: unprocessed
24763 + */
24764 +static int au_opt_simple(struct super_block *sb, struct au_opt *opt,
24765 +                        struct au_opts *opts)
24766 +{
24767 +       int err;
24768 +       struct au_sbinfo *sbinfo;
24769 +
24770 +       SiMustWriteLock(sb);
24771 +
24772 +       err = 1; /* handled */
24773 +       sbinfo = au_sbi(sb);
24774 +       switch (opt->type) {
24775 +       case Opt_udba:
24776 +               sbinfo->si_mntflags &= ~AuOptMask_UDBA;
24777 +               sbinfo->si_mntflags |= opt->udba;
24778 +               opts->given_udba |= opt->udba;
24779 +               break;
24780 +
24781 +       case Opt_plink:
24782 +               au_opt_set(sbinfo->si_mntflags, PLINK);
24783 +               break;
24784 +       case Opt_noplink:
24785 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
24786 +                       au_plink_put(sb, /*verbose*/1);
24787 +               au_opt_clr(sbinfo->si_mntflags, PLINK);
24788 +               break;
24789 +       case Opt_list_plink:
24790 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
24791 +                       au_plink_list(sb);
24792 +               break;
24793 +
24794 +       case Opt_dio:
24795 +               au_opt_set(sbinfo->si_mntflags, DIO);
24796 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
24797 +               break;
24798 +       case Opt_nodio:
24799 +               au_opt_clr(sbinfo->si_mntflags, DIO);
24800 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
24801 +               break;
24802 +
24803 +       case Opt_fhsm_sec:
24804 +               au_fhsm_set(sbinfo, opt->fhsm_second);
24805 +               break;
24806 +
24807 +       case Opt_diropq_a:
24808 +               au_opt_set(sbinfo->si_mntflags, ALWAYS_DIROPQ);
24809 +               break;
24810 +       case Opt_diropq_w:
24811 +               au_opt_clr(sbinfo->si_mntflags, ALWAYS_DIROPQ);
24812 +               break;
24813 +
24814 +       case Opt_warn_perm:
24815 +               au_opt_set(sbinfo->si_mntflags, WARN_PERM);
24816 +               break;
24817 +       case Opt_nowarn_perm:
24818 +               au_opt_clr(sbinfo->si_mntflags, WARN_PERM);
24819 +               break;
24820 +
24821 +       case Opt_verbose:
24822 +               au_opt_set(sbinfo->si_mntflags, VERBOSE);
24823 +               break;
24824 +       case Opt_noverbose:
24825 +               au_opt_clr(sbinfo->si_mntflags, VERBOSE);
24826 +               break;
24827 +
24828 +       case Opt_sum:
24829 +               au_opt_set(sbinfo->si_mntflags, SUM);
24830 +               break;
24831 +       case Opt_wsum:
24832 +               au_opt_clr(sbinfo->si_mntflags, SUM);
24833 +               au_opt_set(sbinfo->si_mntflags, SUM_W);
24834 +       case Opt_nosum:
24835 +               au_opt_clr(sbinfo->si_mntflags, SUM);
24836 +               au_opt_clr(sbinfo->si_mntflags, SUM_W);
24837 +               break;
24838 +
24839 +       case Opt_wbr_create:
24840 +               err = au_opt_wbr_create(sb, &opt->wbr_create);
24841 +               break;
24842 +       case Opt_wbr_copyup:
24843 +               sbinfo->si_wbr_copyup = opt->wbr_copyup;
24844 +               sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + opt->wbr_copyup;
24845 +               break;
24846 +
24847 +       case Opt_dirwh:
24848 +               sbinfo->si_dirwh = opt->dirwh;
24849 +               break;
24850 +
24851 +       case Opt_rdcache:
24852 +               sbinfo->si_rdcache
24853 +                       = msecs_to_jiffies(opt->rdcache * MSEC_PER_SEC);
24854 +               break;
24855 +       case Opt_rdblk:
24856 +               sbinfo->si_rdblk = opt->rdblk;
24857 +               break;
24858 +       case Opt_rdblk_def:
24859 +               sbinfo->si_rdblk = AUFS_RDBLK_DEF;
24860 +               break;
24861 +       case Opt_rdhash:
24862 +               sbinfo->si_rdhash = opt->rdhash;
24863 +               break;
24864 +       case Opt_rdhash_def:
24865 +               sbinfo->si_rdhash = AUFS_RDHASH_DEF;
24866 +               break;
24867 +
24868 +       case Opt_shwh:
24869 +               au_opt_set(sbinfo->si_mntflags, SHWH);
24870 +               break;
24871 +       case Opt_noshwh:
24872 +               au_opt_clr(sbinfo->si_mntflags, SHWH);
24873 +               break;
24874 +
24875 +       case Opt_dirperm1:
24876 +               au_opt_set(sbinfo->si_mntflags, DIRPERM1);
24877 +               break;
24878 +       case Opt_nodirperm1:
24879 +               au_opt_clr(sbinfo->si_mntflags, DIRPERM1);
24880 +               break;
24881 +
24882 +       case Opt_trunc_xino:
24883 +               au_opt_set(sbinfo->si_mntflags, TRUNC_XINO);
24884 +               break;
24885 +       case Opt_notrunc_xino:
24886 +               au_opt_clr(sbinfo->si_mntflags, TRUNC_XINO);
24887 +               break;
24888 +
24889 +       case Opt_trunc_xino_path:
24890 +       case Opt_itrunc_xino:
24891 +               err = au_xino_trunc(sb, opt->xino_itrunc.bindex);
24892 +               if (!err)
24893 +                       err = 1;
24894 +               break;
24895 +
24896 +       case Opt_trunc_xib:
24897 +               au_fset_opts(opts->flags, TRUNC_XIB);
24898 +               break;
24899 +       case Opt_notrunc_xib:
24900 +               au_fclr_opts(opts->flags, TRUNC_XIB);
24901 +               break;
24902 +
24903 +       case Opt_acl:
24904 +               sb->s_flags |= MS_POSIXACL;
24905 +               break;
24906 +       case Opt_noacl:
24907 +               sb->s_flags &= ~MS_POSIXACL;
24908 +               break;
24909 +
24910 +       default:
24911 +               err = 0;
24912 +               break;
24913 +       }
24914 +
24915 +       return err;
24916 +}
24917 +
24918 +/*
24919 + * returns tri-state.
24920 + * plus: processed without an error
24921 + * zero: unprocessed
24922 + * minus: error
24923 + */
24924 +static int au_opt_br(struct super_block *sb, struct au_opt *opt,
24925 +                    struct au_opts *opts)
24926 +{
24927 +       int err, do_refresh;
24928 +
24929 +       err = 0;
24930 +       switch (opt->type) {
24931 +       case Opt_append:
24932 +               opt->add.bindex = au_sbbot(sb) + 1;
24933 +               if (opt->add.bindex < 0)
24934 +                       opt->add.bindex = 0;
24935 +               goto add;
24936 +       case Opt_prepend:
24937 +               opt->add.bindex = 0;
24938 +       add: /* indented label */
24939 +       case Opt_add:
24940 +               err = au_br_add(sb, &opt->add,
24941 +                               au_ftest_opts(opts->flags, REMOUNT));
24942 +               if (!err) {
24943 +                       err = 1;
24944 +                       au_fset_opts(opts->flags, REFRESH);
24945 +               }
24946 +               break;
24947 +
24948 +       case Opt_del:
24949 +       case Opt_idel:
24950 +               err = au_br_del(sb, &opt->del,
24951 +                               au_ftest_opts(opts->flags, REMOUNT));
24952 +               if (!err) {
24953 +                       err = 1;
24954 +                       au_fset_opts(opts->flags, TRUNC_XIB);
24955 +                       au_fset_opts(opts->flags, REFRESH);
24956 +               }
24957 +               break;
24958 +
24959 +       case Opt_mod:
24960 +       case Opt_imod:
24961 +               err = au_br_mod(sb, &opt->mod,
24962 +                               au_ftest_opts(opts->flags, REMOUNT),
24963 +                               &do_refresh);
24964 +               if (!err) {
24965 +                       err = 1;
24966 +                       if (do_refresh)
24967 +                               au_fset_opts(opts->flags, REFRESH);
24968 +               }
24969 +               break;
24970 +       }
24971 +
24972 +       return err;
24973 +}
24974 +
24975 +static int au_opt_xino(struct super_block *sb, struct au_opt *opt,
24976 +                      struct au_opt_xino **opt_xino,
24977 +                      struct au_opts *opts)
24978 +{
24979 +       int err;
24980 +       aufs_bindex_t bbot, bindex;
24981 +       struct dentry *root, *parent, *h_root;
24982 +
24983 +       err = 0;
24984 +       switch (opt->type) {
24985 +       case Opt_xino:
24986 +               err = au_xino_set(sb, &opt->xino,
24987 +                                 !!au_ftest_opts(opts->flags, REMOUNT));
24988 +               if (unlikely(err))
24989 +                       break;
24990 +
24991 +               *opt_xino = &opt->xino;
24992 +               au_xino_brid_set(sb, -1);
24993 +
24994 +               /* safe d_parent access */
24995 +               parent = opt->xino.file->f_path.dentry->d_parent;
24996 +               root = sb->s_root;
24997 +               bbot = au_sbbot(sb);
24998 +               for (bindex = 0; bindex <= bbot; bindex++) {
24999 +                       h_root = au_h_dptr(root, bindex);
25000 +                       if (h_root == parent) {
25001 +                               au_xino_brid_set(sb, au_sbr_id(sb, bindex));
25002 +                               break;
25003 +                       }
25004 +               }
25005 +               break;
25006 +
25007 +       case Opt_noxino:
25008 +               au_xino_clr(sb);
25009 +               au_xino_brid_set(sb, -1);
25010 +               *opt_xino = (void *)-1;
25011 +               break;
25012 +       }
25013 +
25014 +       return err;
25015 +}
25016 +
25017 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
25018 +                  unsigned int pending)
25019 +{
25020 +       int err, fhsm;
25021 +       aufs_bindex_t bindex, bbot;
25022 +       unsigned char do_plink, skip, do_free, can_no_dreval;
25023 +       struct au_branch *br;
25024 +       struct au_wbr *wbr;
25025 +       struct dentry *root, *dentry;
25026 +       struct inode *dir, *h_dir;
25027 +       struct au_sbinfo *sbinfo;
25028 +       struct au_hinode *hdir;
25029 +
25030 +       SiMustAnyLock(sb);
25031 +
25032 +       sbinfo = au_sbi(sb);
25033 +       AuDebugOn(!(sbinfo->si_mntflags & AuOptMask_UDBA));
25034 +
25035 +       if (!(sb_flags & MS_RDONLY)) {
25036 +               if (unlikely(!au_br_writable(au_sbr_perm(sb, 0))))
25037 +                       pr_warn("first branch should be rw\n");
25038 +               if (unlikely(au_opt_test(sbinfo->si_mntflags, SHWH)))
25039 +                       pr_warn_once("shwh should be used with ro\n");
25040 +       }
25041 +
25042 +       if (au_opt_test((sbinfo->si_mntflags | pending), UDBA_HNOTIFY)
25043 +           && !au_opt_test(sbinfo->si_mntflags, XINO))
25044 +               pr_warn_once("udba=*notify requires xino\n");
25045 +
25046 +       if (au_opt_test(sbinfo->si_mntflags, DIRPERM1))
25047 +               pr_warn_once("dirperm1 breaks the protection"
25048 +                            " by the permission bits on the lower branch\n");
25049 +
25050 +       err = 0;
25051 +       fhsm = 0;
25052 +       root = sb->s_root;
25053 +       dir = d_inode(root);
25054 +       do_plink = !!au_opt_test(sbinfo->si_mntflags, PLINK);
25055 +       can_no_dreval = !!au_opt_test((sbinfo->si_mntflags | pending),
25056 +                                     UDBA_NONE);
25057 +       bbot = au_sbbot(sb);
25058 +       for (bindex = 0; !err && bindex <= bbot; bindex++) {
25059 +               skip = 0;
25060 +               h_dir = au_h_iptr(dir, bindex);
25061 +               br = au_sbr(sb, bindex);
25062 +
25063 +               if ((br->br_perm & AuBrAttr_ICEX)
25064 +                   && !h_dir->i_op->listxattr)
25065 +                       br->br_perm &= ~AuBrAttr_ICEX;
25066 +#if 0
25067 +               if ((br->br_perm & AuBrAttr_ICEX_SEC)
25068 +                   && (au_br_sb(br)->s_flags & MS_NOSEC))
25069 +                       br->br_perm &= ~AuBrAttr_ICEX_SEC;
25070 +#endif
25071 +
25072 +               do_free = 0;
25073 +               wbr = br->br_wbr;
25074 +               if (wbr)
25075 +                       wbr_wh_read_lock(wbr);
25076 +
25077 +               if (!au_br_writable(br->br_perm)) {
25078 +                       do_free = !!wbr;
25079 +                       skip = (!wbr
25080 +                               || (!wbr->wbr_whbase
25081 +                                   && !wbr->wbr_plink
25082 +                                   && !wbr->wbr_orph));
25083 +               } else if (!au_br_wh_linkable(br->br_perm)) {
25084 +                       /* skip = (!br->br_whbase && !br->br_orph); */
25085 +                       skip = (!wbr || !wbr->wbr_whbase);
25086 +                       if (skip && wbr) {
25087 +                               if (do_plink)
25088 +                                       skip = !!wbr->wbr_plink;
25089 +                               else
25090 +                                       skip = !wbr->wbr_plink;
25091 +                       }
25092 +               } else {
25093 +                       /* skip = (br->br_whbase && br->br_ohph); */
25094 +                       skip = (wbr && wbr->wbr_whbase);
25095 +                       if (skip) {
25096 +                               if (do_plink)
25097 +                                       skip = !!wbr->wbr_plink;
25098 +                               else
25099 +                                       skip = !wbr->wbr_plink;
25100 +                       }
25101 +               }
25102 +               if (wbr)
25103 +                       wbr_wh_read_unlock(wbr);
25104 +
25105 +               if (can_no_dreval) {
25106 +                       dentry = br->br_path.dentry;
25107 +                       spin_lock(&dentry->d_lock);
25108 +                       if (dentry->d_flags &
25109 +                           (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE))
25110 +                               can_no_dreval = 0;
25111 +                       spin_unlock(&dentry->d_lock);
25112 +               }
25113 +
25114 +               if (au_br_fhsm(br->br_perm)) {
25115 +                       fhsm++;
25116 +                       AuDebugOn(!br->br_fhsm);
25117 +               }
25118 +
25119 +               if (skip)
25120 +                       continue;
25121 +
25122 +               hdir = au_hi(dir, bindex);
25123 +               au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
25124 +               if (wbr)
25125 +                       wbr_wh_write_lock(wbr);
25126 +               err = au_wh_init(br, sb);
25127 +               if (wbr)
25128 +                       wbr_wh_write_unlock(wbr);
25129 +               au_hn_inode_unlock(hdir);
25130 +
25131 +               if (!err && do_free) {
25132 +                       if (wbr)
25133 +                               au_delayed_kfree(wbr);
25134 +                       br->br_wbr = NULL;
25135 +               }
25136 +       }
25137 +
25138 +       if (can_no_dreval)
25139 +               au_fset_si(sbinfo, NO_DREVAL);
25140 +       else
25141 +               au_fclr_si(sbinfo, NO_DREVAL);
25142 +
25143 +       if (fhsm >= 2) {
25144 +               au_fset_si(sbinfo, FHSM);
25145 +               for (bindex = bbot; bindex >= 0; bindex--) {
25146 +                       br = au_sbr(sb, bindex);
25147 +                       if (au_br_fhsm(br->br_perm)) {
25148 +                               au_fhsm_set_bottom(sb, bindex);
25149 +                               break;
25150 +                       }
25151 +               }
25152 +       } else {
25153 +               au_fclr_si(sbinfo, FHSM);
25154 +               au_fhsm_set_bottom(sb, -1);
25155 +       }
25156 +
25157 +       return err;
25158 +}
25159 +
25160 +int au_opts_mount(struct super_block *sb, struct au_opts *opts)
25161 +{
25162 +       int err;
25163 +       unsigned int tmp;
25164 +       aufs_bindex_t bindex, bbot;
25165 +       struct au_opt *opt;
25166 +       struct au_opt_xino *opt_xino, xino;
25167 +       struct au_sbinfo *sbinfo;
25168 +       struct au_branch *br;
25169 +       struct inode *dir;
25170 +
25171 +       SiMustWriteLock(sb);
25172 +
25173 +       err = 0;
25174 +       opt_xino = NULL;
25175 +       opt = opts->opt;
25176 +       while (err >= 0 && opt->type != Opt_tail)
25177 +               err = au_opt_simple(sb, opt++, opts);
25178 +       if (err > 0)
25179 +               err = 0;
25180 +       else if (unlikely(err < 0))
25181 +               goto out;
25182 +
25183 +       /* disable xino and udba temporary */
25184 +       sbinfo = au_sbi(sb);
25185 +       tmp = sbinfo->si_mntflags;
25186 +       au_opt_clr(sbinfo->si_mntflags, XINO);
25187 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_REVAL);
25188 +
25189 +       opt = opts->opt;
25190 +       while (err >= 0 && opt->type != Opt_tail)
25191 +               err = au_opt_br(sb, opt++, opts);
25192 +       if (err > 0)
25193 +               err = 0;
25194 +       else if (unlikely(err < 0))
25195 +               goto out;
25196 +
25197 +       bbot = au_sbbot(sb);
25198 +       if (unlikely(bbot < 0)) {
25199 +               err = -EINVAL;
25200 +               pr_err("no branches\n");
25201 +               goto out;
25202 +       }
25203 +
25204 +       if (au_opt_test(tmp, XINO))
25205 +               au_opt_set(sbinfo->si_mntflags, XINO);
25206 +       opt = opts->opt;
25207 +       while (!err && opt->type != Opt_tail)
25208 +               err = au_opt_xino(sb, opt++, &opt_xino, opts);
25209 +       if (unlikely(err))
25210 +               goto out;
25211 +
25212 +       err = au_opts_verify(sb, sb->s_flags, tmp);
25213 +       if (unlikely(err))
25214 +               goto out;
25215 +
25216 +       /* restore xino */
25217 +       if (au_opt_test(tmp, XINO) && !opt_xino) {
25218 +               xino.file = au_xino_def(sb);
25219 +               err = PTR_ERR(xino.file);
25220 +               if (IS_ERR(xino.file))
25221 +                       goto out;
25222 +
25223 +               err = au_xino_set(sb, &xino, /*remount*/0);
25224 +               fput(xino.file);
25225 +               if (unlikely(err))
25226 +                       goto out;
25227 +       }
25228 +
25229 +       /* restore udba */
25230 +       tmp &= AuOptMask_UDBA;
25231 +       sbinfo->si_mntflags &= ~AuOptMask_UDBA;
25232 +       sbinfo->si_mntflags |= tmp;
25233 +       bbot = au_sbbot(sb);
25234 +       for (bindex = 0; bindex <= bbot; bindex++) {
25235 +               br = au_sbr(sb, bindex);
25236 +               err = au_hnotify_reset_br(tmp, br, br->br_perm);
25237 +               if (unlikely(err))
25238 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
25239 +                               bindex, err);
25240 +               /* go on even if err */
25241 +       }
25242 +       if (au_opt_test(tmp, UDBA_HNOTIFY)) {
25243 +               dir = d_inode(sb->s_root);
25244 +               au_hn_reset(dir, au_hi_flags(dir, /*isdir*/1) & ~AuHi_XINO);
25245 +       }
25246 +
25247 +out:
25248 +       return err;
25249 +}
25250 +
25251 +int au_opts_remount(struct super_block *sb, struct au_opts *opts)
25252 +{
25253 +       int err, rerr;
25254 +       unsigned char no_dreval;
25255 +       struct inode *dir;
25256 +       struct au_opt_xino *opt_xino;
25257 +       struct au_opt *opt;
25258 +       struct au_sbinfo *sbinfo;
25259 +
25260 +       SiMustWriteLock(sb);
25261 +
25262 +       err = 0;
25263 +       dir = d_inode(sb->s_root);
25264 +       sbinfo = au_sbi(sb);
25265 +       opt_xino = NULL;
25266 +       opt = opts->opt;
25267 +       while (err >= 0 && opt->type != Opt_tail) {
25268 +               err = au_opt_simple(sb, opt, opts);
25269 +               if (!err)
25270 +                       err = au_opt_br(sb, opt, opts);
25271 +               if (!err)
25272 +                       err = au_opt_xino(sb, opt, &opt_xino, opts);
25273 +               opt++;
25274 +       }
25275 +       if (err > 0)
25276 +               err = 0;
25277 +       AuTraceErr(err);
25278 +       /* go on even err */
25279 +
25280 +       no_dreval = !!au_ftest_si(sbinfo, NO_DREVAL);
25281 +       rerr = au_opts_verify(sb, opts->sb_flags, /*pending*/0);
25282 +       if (unlikely(rerr && !err))
25283 +               err = rerr;
25284 +
25285 +       if (no_dreval != !!au_ftest_si(sbinfo, NO_DREVAL))
25286 +               au_fset_opts(opts->flags, REFRESH_IDOP);
25287 +
25288 +       if (au_ftest_opts(opts->flags, TRUNC_XIB)) {
25289 +               rerr = au_xib_trunc(sb);
25290 +               if (unlikely(rerr && !err))
25291 +                       err = rerr;
25292 +       }
25293 +
25294 +       /* will be handled by the caller */
25295 +       if (!au_ftest_opts(opts->flags, REFRESH)
25296 +           && (opts->given_udba
25297 +               || au_opt_test(sbinfo->si_mntflags, XINO)
25298 +               || au_ftest_opts(opts->flags, REFRESH_IDOP)
25299 +                   ))
25300 +               au_fset_opts(opts->flags, REFRESH);
25301 +
25302 +       AuDbg("status 0x%x\n", opts->flags);
25303 +       return err;
25304 +}
25305 +
25306 +/* ---------------------------------------------------------------------- */
25307 +
25308 +unsigned int au_opt_udba(struct super_block *sb)
25309 +{
25310 +       return au_mntflags(sb) & AuOptMask_UDBA;
25311 +}
25312 diff -urN /usr/share/empty/fs/aufs/opts.h linux/fs/aufs/opts.h
25313 --- /usr/share/empty/fs/aufs/opts.h     1970-01-01 01:00:00.000000000 +0100
25314 +++ linux/fs/aufs/opts.h        2017-05-06 22:16:52.781553511 +0200
25315 @@ -0,0 +1,213 @@
25316 +/*
25317 + * Copyright (C) 2005-2017 Junjiro R. Okajima
25318 + *
25319 + * This program, aufs is free software; you can redistribute it and/or modify
25320 + * it under the terms of the GNU General Public License as published by
25321 + * the Free Software Foundation; either version 2 of the License, or
25322 + * (at your option) any later version.
25323 + *
25324 + * This program is distributed in the hope that it will be useful,
25325 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25326 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25327 + * GNU General Public License for more details.
25328 + *
25329 + * You should have received a copy of the GNU General Public License
25330 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25331 + */
25332 +
25333 +/*
25334 + * mount options/flags
25335 + */
25336 +
25337 +#ifndef __AUFS_OPTS_H__
25338 +#define __AUFS_OPTS_H__
25339 +
25340 +#ifdef __KERNEL__
25341 +
25342 +#include <linux/path.h>
25343 +
25344 +struct file;
25345 +struct super_block;
25346 +
25347 +/* ---------------------------------------------------------------------- */
25348 +
25349 +/* mount flags */
25350 +#define AuOpt_XINO             1               /* external inode number bitmap
25351 +                                                  and translation table */
25352 +#define AuOpt_TRUNC_XINO       (1 << 1)        /* truncate xino files */
25353 +#define AuOpt_UDBA_NONE                (1 << 2)        /* users direct branch access */
25354 +#define AuOpt_UDBA_REVAL       (1 << 3)
25355 +#define AuOpt_UDBA_HNOTIFY     (1 << 4)
25356 +#define AuOpt_SHWH             (1 << 5)        /* show whiteout */
25357 +#define AuOpt_PLINK            (1 << 6)        /* pseudo-link */
25358 +#define AuOpt_DIRPERM1         (1 << 7)        /* ignore the lower dir's perm
25359 +                                                  bits */
25360 +#define AuOpt_ALWAYS_DIROPQ    (1 << 9)        /* policy to creating diropq */
25361 +#define AuOpt_SUM              (1 << 10)       /* summation for statfs(2) */
25362 +#define AuOpt_SUM_W            (1 << 11)       /* unimplemented */
25363 +#define AuOpt_WARN_PERM                (1 << 12)       /* warn when add-branch */
25364 +#define AuOpt_VERBOSE          (1 << 13)       /* busy inode when del-branch */
25365 +#define AuOpt_DIO              (1 << 14)       /* direct io */
25366 +
25367 +#ifndef CONFIG_AUFS_HNOTIFY
25368 +#undef AuOpt_UDBA_HNOTIFY
25369 +#define AuOpt_UDBA_HNOTIFY     0
25370 +#endif
25371 +#ifndef CONFIG_AUFS_SHWH
25372 +#undef AuOpt_SHWH
25373 +#define AuOpt_SHWH             0
25374 +#endif
25375 +
25376 +#define AuOpt_Def      (AuOpt_XINO \
25377 +                        | AuOpt_UDBA_REVAL \
25378 +                        | AuOpt_PLINK \
25379 +                        /* | AuOpt_DIRPERM1 */ \
25380 +                        | AuOpt_WARN_PERM)
25381 +#define AuOptMask_UDBA (AuOpt_UDBA_NONE \
25382 +                        | AuOpt_UDBA_REVAL \
25383 +                        | AuOpt_UDBA_HNOTIFY)
25384 +
25385 +#define au_opt_test(flags, name)       (flags & AuOpt_##name)
25386 +#define au_opt_set(flags, name) do { \
25387 +       BUILD_BUG_ON(AuOpt_##name & AuOptMask_UDBA); \
25388 +       ((flags) |= AuOpt_##name); \
25389 +} while (0)
25390 +#define au_opt_set_udba(flags, name) do { \
25391 +       (flags) &= ~AuOptMask_UDBA; \
25392 +       ((flags) |= AuOpt_##name); \
25393 +} while (0)
25394 +#define au_opt_clr(flags, name) do { \
25395 +       ((flags) &= ~AuOpt_##name); \
25396 +} while (0)
25397 +
25398 +static inline unsigned int au_opts_plink(unsigned int mntflags)
25399 +{
25400 +#ifdef CONFIG_PROC_FS
25401 +       return mntflags;
25402 +#else
25403 +       return mntflags & ~AuOpt_PLINK;
25404 +#endif
25405 +}
25406 +
25407 +/* ---------------------------------------------------------------------- */
25408 +
25409 +/* policies to select one among multiple writable branches */
25410 +enum {
25411 +       AuWbrCreate_TDP,        /* top down parent */
25412 +       AuWbrCreate_RR,         /* round robin */
25413 +       AuWbrCreate_MFS,        /* most free space */
25414 +       AuWbrCreate_MFSV,       /* mfs with seconds */
25415 +       AuWbrCreate_MFSRR,      /* mfs then rr */
25416 +       AuWbrCreate_MFSRRV,     /* mfs then rr with seconds */
25417 +       AuWbrCreate_TDMFS,      /* top down regardless parent and mfs */
25418 +       AuWbrCreate_TDMFSV,     /* top down regardless parent and mfs */
25419 +       AuWbrCreate_PMFS,       /* parent and mfs */
25420 +       AuWbrCreate_PMFSV,      /* parent and mfs with seconds */
25421 +       AuWbrCreate_PMFSRR,     /* parent, mfs and round-robin */
25422 +       AuWbrCreate_PMFSRRV,    /* plus seconds */
25423 +
25424 +       AuWbrCreate_Def = AuWbrCreate_TDP
25425 +};
25426 +
25427 +enum {
25428 +       AuWbrCopyup_TDP,        /* top down parent */
25429 +       AuWbrCopyup_BUP,        /* bottom up parent */
25430 +       AuWbrCopyup_BU,         /* bottom up */
25431 +
25432 +       AuWbrCopyup_Def = AuWbrCopyup_TDP
25433 +};
25434 +
25435 +/* ---------------------------------------------------------------------- */
25436 +
25437 +struct au_opt_add {
25438 +       aufs_bindex_t   bindex;
25439 +       char            *pathname;
25440 +       int             perm;
25441 +       struct path     path;
25442 +};
25443 +
25444 +struct au_opt_del {
25445 +       char            *pathname;
25446 +       struct path     h_path;
25447 +};
25448 +
25449 +struct au_opt_mod {
25450 +       char            *path;
25451 +       int             perm;
25452 +       struct dentry   *h_root;
25453 +};
25454 +
25455 +struct au_opt_xino {
25456 +       char            *path;
25457 +       struct file     *file;
25458 +};
25459 +
25460 +struct au_opt_xino_itrunc {
25461 +       aufs_bindex_t   bindex;
25462 +};
25463 +
25464 +struct au_opt_wbr_create {
25465 +       int                     wbr_create;
25466 +       int                     mfs_second;
25467 +       unsigned long long      mfsrr_watermark;
25468 +};
25469 +
25470 +struct au_opt {
25471 +       int type;
25472 +       union {
25473 +               struct au_opt_xino      xino;
25474 +               struct au_opt_xino_itrunc xino_itrunc;
25475 +               struct au_opt_add       add;
25476 +               struct au_opt_del       del;
25477 +               struct au_opt_mod       mod;
25478 +               int                     dirwh;
25479 +               int                     rdcache;
25480 +               unsigned int            rdblk;
25481 +               unsigned int            rdhash;
25482 +               int                     udba;
25483 +               struct au_opt_wbr_create wbr_create;
25484 +               int                     wbr_copyup;
25485 +               unsigned int            fhsm_second;
25486 +       };
25487 +};
25488 +
25489 +/* opts flags */
25490 +#define AuOpts_REMOUNT         1
25491 +#define AuOpts_REFRESH         (1 << 1)
25492 +#define AuOpts_TRUNC_XIB       (1 << 2)
25493 +#define AuOpts_REFRESH_DYAOP   (1 << 3)
25494 +#define AuOpts_REFRESH_IDOP    (1 << 4)
25495 +#define au_ftest_opts(flags, name)     ((flags) & AuOpts_##name)
25496 +#define au_fset_opts(flags, name) \
25497 +       do { (flags) |= AuOpts_##name; } while (0)
25498 +#define au_fclr_opts(flags, name) \
25499 +       do { (flags) &= ~AuOpts_##name; } while (0)
25500 +
25501 +struct au_opts {
25502 +       struct au_opt   *opt;
25503 +       int             max_opt;
25504 +
25505 +       unsigned int    given_udba;
25506 +       unsigned int    flags;
25507 +       unsigned long   sb_flags;
25508 +};
25509 +
25510 +/* ---------------------------------------------------------------------- */
25511 +
25512 +/* opts.c */
25513 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm);
25514 +const char *au_optstr_udba(int udba);
25515 +const char *au_optstr_wbr_copyup(int wbr_copyup);
25516 +const char *au_optstr_wbr_create(int wbr_create);
25517 +
25518 +void au_opts_free(struct au_opts *opts);
25519 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts);
25520 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
25521 +                  unsigned int pending);
25522 +int au_opts_mount(struct super_block *sb, struct au_opts *opts);
25523 +int au_opts_remount(struct super_block *sb, struct au_opts *opts);
25524 +
25525 +unsigned int au_opt_udba(struct super_block *sb);
25526 +
25527 +#endif /* __KERNEL__ */
25528 +#endif /* __AUFS_OPTS_H__ */
25529 diff -urN /usr/share/empty/fs/aufs/plink.c linux/fs/aufs/plink.c
25530 --- /usr/share/empty/fs/aufs/plink.c    1970-01-01 01:00:00.000000000 +0100
25531 +++ linux/fs/aufs/plink.c       2017-05-06 22:16:52.781553511 +0200
25532 @@ -0,0 +1,514 @@
25533 +/*
25534 + * Copyright (C) 2005-2017 Junjiro R. Okajima
25535 + *
25536 + * This program, aufs is free software; you can redistribute it and/or modify
25537 + * it under the terms of the GNU General Public License as published by
25538 + * the Free Software Foundation; either version 2 of the License, or
25539 + * (at your option) any later version.
25540 + *
25541 + * This program is distributed in the hope that it will be useful,
25542 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25543 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25544 + * GNU General Public License for more details.
25545 + *
25546 + * You should have received a copy of the GNU General Public License
25547 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25548 + */
25549 +
25550 +/*
25551 + * pseudo-link
25552 + */
25553 +
25554 +#include "aufs.h"
25555 +
25556 +/*
25557 + * the pseudo-link maintenance mode.
25558 + * during a user process maintains the pseudo-links,
25559 + * prohibit adding a new plink and branch manipulation.
25560 + *
25561 + * Flags
25562 + * NOPLM:
25563 + *     For entry functions which will handle plink, and i_mutex is already held
25564 + *     in VFS.
25565 + *     They cannot wait and should return an error at once.
25566 + *     Callers has to check the error.
25567 + * NOPLMW:
25568 + *     For entry functions which will handle plink, but i_mutex is not held
25569 + *     in VFS.
25570 + *     They can wait the plink maintenance mode to finish.
25571 + *
25572 + * They behave like F_SETLK and F_SETLKW.
25573 + * If the caller never handle plink, then both flags are unnecessary.
25574 + */
25575 +
25576 +int au_plink_maint(struct super_block *sb, int flags)
25577 +{
25578 +       int err;
25579 +       pid_t pid, ppid;
25580 +       struct task_struct *parent, *prev;
25581 +       struct au_sbinfo *sbi;
25582 +
25583 +       SiMustAnyLock(sb);
25584 +
25585 +       err = 0;
25586 +       if (!au_opt_test(au_mntflags(sb), PLINK))
25587 +               goto out;
25588 +
25589 +       sbi = au_sbi(sb);
25590 +       pid = sbi->si_plink_maint_pid;
25591 +       if (!pid || pid == current->pid)
25592 +               goto out;
25593 +
25594 +       /* todo: it highly depends upon /sbin/mount.aufs */
25595 +       prev = NULL;
25596 +       parent = current;
25597 +       ppid = 0;
25598 +       rcu_read_lock();
25599 +       while (1) {
25600 +               parent = rcu_dereference(parent->real_parent);
25601 +               if (parent == prev)
25602 +                       break;
25603 +               ppid = task_pid_vnr(parent);
25604 +               if (pid == ppid) {
25605 +                       rcu_read_unlock();
25606 +                       goto out;
25607 +               }
25608 +               prev = parent;
25609 +       }
25610 +       rcu_read_unlock();
25611 +
25612 +       if (au_ftest_lock(flags, NOPLMW)) {
25613 +               /* if there is no i_mutex lock in VFS, we don't need to wait */
25614 +               /* AuDebugOn(!lockdep_depth(current)); */
25615 +               while (sbi->si_plink_maint_pid) {
25616 +                       si_read_unlock(sb);
25617 +                       /* gave up wake_up_bit() */
25618 +                       wait_event(sbi->si_plink_wq, !sbi->si_plink_maint_pid);
25619 +
25620 +                       if (au_ftest_lock(flags, FLUSH))
25621 +                               au_nwt_flush(&sbi->si_nowait);
25622 +                       si_noflush_read_lock(sb);
25623 +               }
25624 +       } else if (au_ftest_lock(flags, NOPLM)) {
25625 +               AuDbg("ppid %d, pid %d\n", ppid, pid);
25626 +               err = -EAGAIN;
25627 +       }
25628 +
25629 +out:
25630 +       return err;
25631 +}
25632 +
25633 +void au_plink_maint_leave(struct au_sbinfo *sbinfo)
25634 +{
25635 +       spin_lock(&sbinfo->si_plink_maint_lock);
25636 +       sbinfo->si_plink_maint_pid = 0;
25637 +       spin_unlock(&sbinfo->si_plink_maint_lock);
25638 +       wake_up_all(&sbinfo->si_plink_wq);
25639 +}
25640 +
25641 +int au_plink_maint_enter(struct super_block *sb)
25642 +{
25643 +       int err;
25644 +       struct au_sbinfo *sbinfo;
25645 +
25646 +       err = 0;
25647 +       sbinfo = au_sbi(sb);
25648 +       /* make sure i am the only one in this fs */
25649 +       si_write_lock(sb, AuLock_FLUSH);
25650 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
25651 +               spin_lock(&sbinfo->si_plink_maint_lock);
25652 +               if (!sbinfo->si_plink_maint_pid)
25653 +                       sbinfo->si_plink_maint_pid = current->pid;
25654 +               else
25655 +                       err = -EBUSY;
25656 +               spin_unlock(&sbinfo->si_plink_maint_lock);
25657 +       }
25658 +       si_write_unlock(sb);
25659 +
25660 +       return err;
25661 +}
25662 +
25663 +/* ---------------------------------------------------------------------- */
25664 +
25665 +#ifdef CONFIG_AUFS_DEBUG
25666 +void au_plink_list(struct super_block *sb)
25667 +{
25668 +       int i;
25669 +       struct au_sbinfo *sbinfo;
25670 +       struct hlist_head *plink_hlist;
25671 +       struct au_icntnr *icntnr;
25672 +
25673 +       SiMustAnyLock(sb);
25674 +
25675 +       sbinfo = au_sbi(sb);
25676 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25677 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25678 +
25679 +       for (i = 0; i < AuPlink_NHASH; i++) {
25680 +               plink_hlist = &sbinfo->si_plink[i].head;
25681 +               rcu_read_lock();
25682 +               hlist_for_each_entry_rcu(icntnr, plink_hlist, plink)
25683 +                       AuDbg("%lu\n", icntnr->vfs_inode.i_ino);
25684 +               rcu_read_unlock();
25685 +       }
25686 +}
25687 +#endif
25688 +
25689 +/* is the inode pseudo-linked? */
25690 +int au_plink_test(struct inode *inode)
25691 +{
25692 +       int found, i;
25693 +       struct au_sbinfo *sbinfo;
25694 +       struct hlist_head *plink_hlist;
25695 +       struct au_icntnr *icntnr;
25696 +
25697 +       sbinfo = au_sbi(inode->i_sb);
25698 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
25699 +       AuDebugOn(!au_opt_test(au_mntflags(inode->i_sb), PLINK));
25700 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
25701 +
25702 +       found = 0;
25703 +       i = au_plink_hash(inode->i_ino);
25704 +       plink_hlist = &sbinfo->si_plink[i].head;
25705 +       rcu_read_lock();
25706 +       hlist_for_each_entry_rcu(icntnr, plink_hlist, plink)
25707 +               if (&icntnr->vfs_inode == inode) {
25708 +                       found = 1;
25709 +                       break;
25710 +               }
25711 +       rcu_read_unlock();
25712 +       return found;
25713 +}
25714 +
25715 +/* ---------------------------------------------------------------------- */
25716 +
25717 +/*
25718 + * generate a name for plink.
25719 + * the file will be stored under AUFS_WH_PLINKDIR.
25720 + */
25721 +/* 20 is max digits length of ulong 64 */
25722 +#define PLINK_NAME_LEN ((20 + 1) * 2)
25723 +
25724 +static int plink_name(char *name, int len, struct inode *inode,
25725 +                     aufs_bindex_t bindex)
25726 +{
25727 +       int rlen;
25728 +       struct inode *h_inode;
25729 +
25730 +       h_inode = au_h_iptr(inode, bindex);
25731 +       rlen = snprintf(name, len, "%lu.%lu", inode->i_ino, h_inode->i_ino);
25732 +       return rlen;
25733 +}
25734 +
25735 +struct au_do_plink_lkup_args {
25736 +       struct dentry **errp;
25737 +       struct qstr *tgtname;
25738 +       struct dentry *h_parent;
25739 +       struct au_branch *br;
25740 +};
25741 +
25742 +static struct dentry *au_do_plink_lkup(struct qstr *tgtname,
25743 +                                      struct dentry *h_parent,
25744 +                                      struct au_branch *br)
25745 +{
25746 +       struct dentry *h_dentry;
25747 +       struct inode *h_inode;
25748 +
25749 +       h_inode = d_inode(h_parent);
25750 +       inode_lock_nested(h_inode, AuLsc_I_CHILD2);
25751 +       h_dentry = vfsub_lkup_one(tgtname, h_parent);
25752 +       inode_unlock(h_inode);
25753 +       return h_dentry;
25754 +}
25755 +
25756 +static void au_call_do_plink_lkup(void *args)
25757 +{
25758 +       struct au_do_plink_lkup_args *a = args;
25759 +       *a->errp = au_do_plink_lkup(a->tgtname, a->h_parent, a->br);
25760 +}
25761 +
25762 +/* lookup the plink-ed @inode under the branch at @bindex */
25763 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex)
25764 +{
25765 +       struct dentry *h_dentry, *h_parent;
25766 +       struct au_branch *br;
25767 +       int wkq_err;
25768 +       char a[PLINK_NAME_LEN];
25769 +       struct qstr tgtname = QSTR_INIT(a, 0);
25770 +
25771 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
25772 +
25773 +       br = au_sbr(inode->i_sb, bindex);
25774 +       h_parent = br->br_wbr->wbr_plink;
25775 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
25776 +
25777 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
25778 +               struct au_do_plink_lkup_args args = {
25779 +                       .errp           = &h_dentry,
25780 +                       .tgtname        = &tgtname,
25781 +                       .h_parent       = h_parent,
25782 +                       .br             = br
25783 +               };
25784 +
25785 +               wkq_err = au_wkq_wait(au_call_do_plink_lkup, &args);
25786 +               if (unlikely(wkq_err))
25787 +                       h_dentry = ERR_PTR(wkq_err);
25788 +       } else
25789 +               h_dentry = au_do_plink_lkup(&tgtname, h_parent, br);
25790 +
25791 +       return h_dentry;
25792 +}
25793 +
25794 +/* create a pseudo-link */
25795 +static int do_whplink(struct qstr *tgt, struct dentry *h_parent,
25796 +                     struct dentry *h_dentry, struct au_branch *br)
25797 +{
25798 +       int err;
25799 +       struct path h_path = {
25800 +               .mnt = au_br_mnt(br)
25801 +       };
25802 +       struct inode *h_dir, *delegated;
25803 +
25804 +       h_dir = d_inode(h_parent);
25805 +       inode_lock_nested(h_dir, AuLsc_I_CHILD2);
25806 +again:
25807 +       h_path.dentry = vfsub_lkup_one(tgt, h_parent);
25808 +       err = PTR_ERR(h_path.dentry);
25809 +       if (IS_ERR(h_path.dentry))
25810 +               goto out;
25811 +
25812 +       err = 0;
25813 +       /* wh.plink dir is not monitored */
25814 +       /* todo: is it really safe? */
25815 +       if (d_is_positive(h_path.dentry)
25816 +           && d_inode(h_path.dentry) != d_inode(h_dentry)) {
25817 +               delegated = NULL;
25818 +               err = vfsub_unlink(h_dir, &h_path, &delegated, /*force*/0);
25819 +               if (unlikely(err == -EWOULDBLOCK)) {
25820 +                       pr_warn("cannot retry for NFSv4 delegation"
25821 +                               " for an internal unlink\n");
25822 +                       iput(delegated);
25823 +               }
25824 +               dput(h_path.dentry);
25825 +               h_path.dentry = NULL;
25826 +               if (!err)
25827 +                       goto again;
25828 +       }
25829 +       if (!err && d_is_negative(h_path.dentry)) {
25830 +               delegated = NULL;
25831 +               err = vfsub_link(h_dentry, h_dir, &h_path, &delegated);
25832 +               if (unlikely(err == -EWOULDBLOCK)) {
25833 +                       pr_warn("cannot retry for NFSv4 delegation"
25834 +                               " for an internal link\n");
25835 +                       iput(delegated);
25836 +               }
25837 +       }
25838 +       dput(h_path.dentry);
25839 +
25840 +out:
25841 +       inode_unlock(h_dir);
25842 +       return err;
25843 +}
25844 +
25845 +struct do_whplink_args {
25846 +       int *errp;
25847 +       struct qstr *tgt;
25848 +       struct dentry *h_parent;
25849 +       struct dentry *h_dentry;
25850 +       struct au_branch *br;
25851 +};
25852 +
25853 +static void call_do_whplink(void *args)
25854 +{
25855 +       struct do_whplink_args *a = args;
25856 +       *a->errp = do_whplink(a->tgt, a->h_parent, a->h_dentry, a->br);
25857 +}
25858 +
25859 +static int whplink(struct dentry *h_dentry, struct inode *inode,
25860 +                  aufs_bindex_t bindex, struct au_branch *br)
25861 +{
25862 +       int err, wkq_err;
25863 +       struct au_wbr *wbr;
25864 +       struct dentry *h_parent;
25865 +       char a[PLINK_NAME_LEN];
25866 +       struct qstr tgtname = QSTR_INIT(a, 0);
25867 +
25868 +       wbr = au_sbr(inode->i_sb, bindex)->br_wbr;
25869 +       h_parent = wbr->wbr_plink;
25870 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
25871 +
25872 +       /* always superio. */
25873 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
25874 +               struct do_whplink_args args = {
25875 +                       .errp           = &err,
25876 +                       .tgt            = &tgtname,
25877 +                       .h_parent       = h_parent,
25878 +                       .h_dentry       = h_dentry,
25879 +                       .br             = br
25880 +               };
25881 +               wkq_err = au_wkq_wait(call_do_whplink, &args);
25882 +               if (unlikely(wkq_err))
25883 +                       err = wkq_err;
25884 +       } else
25885 +               err = do_whplink(&tgtname, h_parent, h_dentry, br);
25886 +
25887 +       return err;
25888 +}
25889 +
25890 +/*
25891 + * create a new pseudo-link for @h_dentry on @bindex.
25892 + * the linked inode is held in aufs @inode.
25893 + */
25894 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
25895 +                    struct dentry *h_dentry)
25896 +{
25897 +       struct super_block *sb;
25898 +       struct au_sbinfo *sbinfo;
25899 +       struct hlist_head *plink_hlist;
25900 +       struct au_icntnr *icntnr;
25901 +       struct au_sphlhead *sphl;
25902 +       int found, err, cnt, i;
25903 +
25904 +       sb = inode->i_sb;
25905 +       sbinfo = au_sbi(sb);
25906 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25907 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25908 +
25909 +       found = au_plink_test(inode);
25910 +       if (found)
25911 +               return;
25912 +
25913 +       i = au_plink_hash(inode->i_ino);
25914 +       sphl = sbinfo->si_plink + i;
25915 +       plink_hlist = &sphl->head;
25916 +       au_igrab(inode);
25917 +
25918 +       spin_lock(&sphl->spin);
25919 +       hlist_for_each_entry(icntnr, plink_hlist, plink) {
25920 +               if (&icntnr->vfs_inode == inode) {
25921 +                       found = 1;
25922 +                       break;
25923 +               }
25924 +       }
25925 +       if (!found) {
25926 +               icntnr = container_of(inode, struct au_icntnr, vfs_inode);
25927 +               hlist_add_head_rcu(&icntnr->plink, plink_hlist);
25928 +       }
25929 +       spin_unlock(&sphl->spin);
25930 +       if (!found) {
25931 +               cnt = au_sphl_count(sphl);
25932 +#define msg "unexpectedly unblanced or too many pseudo-links"
25933 +               if (cnt > AUFS_PLINK_WARN)
25934 +                       AuWarn1(msg ", %d\n", cnt);
25935 +#undef msg
25936 +               err = whplink(h_dentry, inode, bindex, au_sbr(sb, bindex));
25937 +               if (unlikely(err)) {
25938 +                       pr_warn("err %d, damaged pseudo link.\n", err);
25939 +                       au_sphl_del_rcu(&icntnr->plink, sphl);
25940 +                       iput(&icntnr->vfs_inode);
25941 +               }
25942 +       } else
25943 +               iput(&icntnr->vfs_inode);
25944 +}
25945 +
25946 +/* free all plinks */
25947 +void au_plink_put(struct super_block *sb, int verbose)
25948 +{
25949 +       int i, warned;
25950 +       struct au_sbinfo *sbinfo;
25951 +       struct hlist_head *plink_hlist;
25952 +       struct hlist_node *tmp;
25953 +       struct au_icntnr *icntnr;
25954 +
25955 +       SiMustWriteLock(sb);
25956 +
25957 +       sbinfo = au_sbi(sb);
25958 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25959 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25960 +
25961 +       /* no spin_lock since sbinfo is write-locked */
25962 +       warned = 0;
25963 +       for (i = 0; i < AuPlink_NHASH; i++) {
25964 +               plink_hlist = &sbinfo->si_plink[i].head;
25965 +               if (!warned && verbose && !hlist_empty(plink_hlist)) {
25966 +                       pr_warn("pseudo-link is not flushed");
25967 +                       warned = 1;
25968 +               }
25969 +               hlist_for_each_entry_safe(icntnr, tmp, plink_hlist, plink)
25970 +                       iput(&icntnr->vfs_inode);
25971 +               INIT_HLIST_HEAD(plink_hlist);
25972 +       }
25973 +}
25974 +
25975 +void au_plink_clean(struct super_block *sb, int verbose)
25976 +{
25977 +       struct dentry *root;
25978 +
25979 +       root = sb->s_root;
25980 +       aufs_write_lock(root);
25981 +       if (au_opt_test(au_mntflags(sb), PLINK))
25982 +               au_plink_put(sb, verbose);
25983 +       aufs_write_unlock(root);
25984 +}
25985 +
25986 +static int au_plink_do_half_refresh(struct inode *inode, aufs_bindex_t br_id)
25987 +{
25988 +       int do_put;
25989 +       aufs_bindex_t btop, bbot, bindex;
25990 +
25991 +       do_put = 0;
25992 +       btop = au_ibtop(inode);
25993 +       bbot = au_ibbot(inode);
25994 +       if (btop >= 0) {
25995 +               for (bindex = btop; bindex <= bbot; bindex++) {
25996 +                       if (!au_h_iptr(inode, bindex)
25997 +                           || au_ii_br_id(inode, bindex) != br_id)
25998 +                               continue;
25999 +                       au_set_h_iptr(inode, bindex, NULL, 0);
26000 +                       do_put = 1;
26001 +                       break;
26002 +               }
26003 +               if (do_put)
26004 +                       for (bindex = btop; bindex <= bbot; bindex++)
26005 +                               if (au_h_iptr(inode, bindex)) {
26006 +                                       do_put = 0;
26007 +                                       break;
26008 +                               }
26009 +       } else
26010 +               do_put = 1;
26011 +
26012 +       return do_put;
26013 +}
26014 +
26015 +/* free the plinks on a branch specified by @br_id */
26016 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id)
26017 +{
26018 +       struct au_sbinfo *sbinfo;
26019 +       struct hlist_head *plink_hlist;
26020 +       struct hlist_node *tmp;
26021 +       struct au_icntnr *icntnr;
26022 +       struct inode *inode;
26023 +       int i, do_put;
26024 +
26025 +       SiMustWriteLock(sb);
26026 +
26027 +       sbinfo = au_sbi(sb);
26028 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
26029 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
26030 +
26031 +       /* no spin_lock since sbinfo is write-locked */
26032 +       for (i = 0; i < AuPlink_NHASH; i++) {
26033 +               plink_hlist = &sbinfo->si_plink[i].head;
26034 +               hlist_for_each_entry_safe(icntnr, tmp, plink_hlist, plink) {
26035 +                       inode = au_igrab(&icntnr->vfs_inode);
26036 +                       ii_write_lock_child(inode);
26037 +                       do_put = au_plink_do_half_refresh(inode, br_id);
26038 +                       if (do_put) {
26039 +                               hlist_del(&icntnr->plink);
26040 +                               iput(inode);
26041 +                       }
26042 +                       ii_write_unlock(inode);
26043 +                       iput(inode);
26044 +               }
26045 +       }
26046 +}
26047 diff -urN /usr/share/empty/fs/aufs/poll.c linux/fs/aufs/poll.c
26048 --- /usr/share/empty/fs/aufs/poll.c     1970-01-01 01:00:00.000000000 +0100
26049 +++ linux/fs/aufs/poll.c        2017-05-06 22:16:52.781553511 +0200
26050 @@ -0,0 +1,52 @@
26051 +/*
26052 + * Copyright (C) 2005-2017 Junjiro R. Okajima
26053 + *
26054 + * This program, aufs is free software; you can redistribute it and/or modify
26055 + * it under the terms of the GNU General Public License as published by
26056 + * the Free Software Foundation; either version 2 of the License, or
26057 + * (at your option) any later version.
26058 + *
26059 + * This program is distributed in the hope that it will be useful,
26060 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26061 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26062 + * GNU General Public License for more details.
26063 + *
26064 + * You should have received a copy of the GNU General Public License
26065 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26066 + */
26067 +
26068 +/*
26069 + * poll operation
26070 + * There is only one filesystem which implements ->poll operation, currently.
26071 + */
26072 +
26073 +#include "aufs.h"
26074 +
26075 +unsigned int aufs_poll(struct file *file, poll_table *wait)
26076 +{
26077 +       unsigned int mask;
26078 +       int err;
26079 +       struct file *h_file;
26080 +       struct super_block *sb;
26081 +
26082 +       /* We should pretend an error happened. */
26083 +       mask = POLLERR /* | POLLIN | POLLOUT */;
26084 +       sb = file->f_path.dentry->d_sb;
26085 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
26086 +
26087 +       h_file = au_read_pre(file, /*keep_fi*/0, /*lsc*/0);
26088 +       err = PTR_ERR(h_file);
26089 +       if (IS_ERR(h_file))
26090 +               goto out;
26091 +
26092 +       /* it is not an error if h_file has no operation */
26093 +       mask = DEFAULT_POLLMASK;
26094 +       if (h_file->f_op->poll)
26095 +               mask = h_file->f_op->poll(h_file, wait);
26096 +       fput(h_file); /* instead of au_read_post() */
26097 +
26098 +out:
26099 +       si_read_unlock(sb);
26100 +       AuTraceErr((int)mask);
26101 +       return mask;
26102 +}
26103 diff -urN /usr/share/empty/fs/aufs/posix_acl.c linux/fs/aufs/posix_acl.c
26104 --- /usr/share/empty/fs/aufs/posix_acl.c        1970-01-01 01:00:00.000000000 +0100
26105 +++ linux/fs/aufs/posix_acl.c   2017-05-06 22:16:52.781553511 +0200
26106 @@ -0,0 +1,102 @@
26107 +/*
26108 + * Copyright (C) 2014-2017 Junjiro R. Okajima
26109 + *
26110 + * This program, aufs is free software; you can redistribute it and/or modify
26111 + * it under the terms of the GNU General Public License as published by
26112 + * the Free Software Foundation; either version 2 of the License, or
26113 + * (at your option) any later version.
26114 + *
26115 + * This program is distributed in the hope that it will be useful,
26116 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26117 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26118 + * GNU General Public License for more details.
26119 + *
26120 + * You should have received a copy of the GNU General Public License
26121 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26122 + */
26123 +
26124 +/*
26125 + * posix acl operations
26126 + */
26127 +
26128 +#include <linux/fs.h>
26129 +#include "aufs.h"
26130 +
26131 +struct posix_acl *aufs_get_acl(struct inode *inode, int type)
26132 +{
26133 +       struct posix_acl *acl;
26134 +       int err;
26135 +       aufs_bindex_t bindex;
26136 +       struct inode *h_inode;
26137 +       struct super_block *sb;
26138 +
26139 +       acl = NULL;
26140 +       sb = inode->i_sb;
26141 +       si_read_lock(sb, AuLock_FLUSH);
26142 +       ii_read_lock_child(inode);
26143 +       if (!(sb->s_flags & MS_POSIXACL))
26144 +               goto out;
26145 +
26146 +       bindex = au_ibtop(inode);
26147 +       h_inode = au_h_iptr(inode, bindex);
26148 +       if (unlikely(!h_inode
26149 +                    || ((h_inode->i_mode & S_IFMT)
26150 +                        != (inode->i_mode & S_IFMT)))) {
26151 +               err = au_busy_or_stale();
26152 +               acl = ERR_PTR(err);
26153 +               goto out;
26154 +       }
26155 +
26156 +       /* always topmost only */
26157 +       acl = get_acl(h_inode, type);
26158 +       if (!IS_ERR_OR_NULL(acl))
26159 +               set_cached_acl(inode, type, acl);
26160 +
26161 +out:
26162 +       ii_read_unlock(inode);
26163 +       si_read_unlock(sb);
26164 +
26165 +       AuTraceErrPtr(acl);
26166 +       return acl;
26167 +}
26168 +
26169 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
26170 +{
26171 +       int err;
26172 +       ssize_t ssz;
26173 +       struct dentry *dentry;
26174 +       struct au_sxattr arg = {
26175 +               .type = AU_ACL_SET,
26176 +               .u.acl_set = {
26177 +                       .acl    = acl,
26178 +                       .type   = type
26179 +               },
26180 +       };
26181 +
26182 +       IMustLock(inode);
26183 +
26184 +       if (inode->i_ino == AUFS_ROOT_INO)
26185 +               dentry = dget(inode->i_sb->s_root);
26186 +       else {
26187 +               dentry = d_find_alias(inode);
26188 +               if (!dentry)
26189 +                       dentry = d_find_any_alias(inode);
26190 +               if (!dentry) {
26191 +                       pr_warn("cannot handle this inode, "
26192 +                               "please report to aufs-users ML\n");
26193 +                       err = -ENOENT;
26194 +                       goto out;
26195 +               }
26196 +       }
26197 +
26198 +       ssz = au_sxattr(dentry, inode, &arg);
26199 +       dput(dentry);
26200 +       err = ssz;
26201 +       if (ssz >= 0) {
26202 +               err = 0;
26203 +               set_cached_acl(inode, type, acl);
26204 +       }
26205 +
26206 +out:
26207 +       return err;
26208 +}
26209 diff -urN /usr/share/empty/fs/aufs/procfs.c linux/fs/aufs/procfs.c
26210 --- /usr/share/empty/fs/aufs/procfs.c   1970-01-01 01:00:00.000000000 +0100
26211 +++ linux/fs/aufs/procfs.c      2017-05-06 22:16:52.781553511 +0200
26212 @@ -0,0 +1,169 @@
26213 +/*
26214 + * Copyright (C) 2010-2017 Junjiro R. Okajima
26215 + *
26216 + * This program, aufs is free software; you can redistribute it and/or modify
26217 + * it under the terms of the GNU General Public License as published by
26218 + * the Free Software Foundation; either version 2 of the License, or
26219 + * (at your option) any later version.
26220 + *
26221 + * This program is distributed in the hope that it will be useful,
26222 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26223 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26224 + * GNU General Public License for more details.
26225 + *
26226 + * You should have received a copy of the GNU General Public License
26227 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26228 + */
26229 +
26230 +/*
26231 + * procfs interfaces
26232 + */
26233 +
26234 +#include <linux/proc_fs.h>
26235 +#include "aufs.h"
26236 +
26237 +static int au_procfs_plm_release(struct inode *inode, struct file *file)
26238 +{
26239 +       struct au_sbinfo *sbinfo;
26240 +
26241 +       sbinfo = file->private_data;
26242 +       if (sbinfo) {
26243 +               au_plink_maint_leave(sbinfo);
26244 +               kobject_put(&sbinfo->si_kobj);
26245 +       }
26246 +
26247 +       return 0;
26248 +}
26249 +
26250 +static void au_procfs_plm_write_clean(struct file *file)
26251 +{
26252 +       struct au_sbinfo *sbinfo;
26253 +
26254 +       sbinfo = file->private_data;
26255 +       if (sbinfo)
26256 +               au_plink_clean(sbinfo->si_sb, /*verbose*/0);
26257 +}
26258 +
26259 +static int au_procfs_plm_write_si(struct file *file, unsigned long id)
26260 +{
26261 +       int err;
26262 +       struct super_block *sb;
26263 +       struct au_sbinfo *sbinfo;
26264 +
26265 +       err = -EBUSY;
26266 +       if (unlikely(file->private_data))
26267 +               goto out;
26268 +
26269 +       sb = NULL;
26270 +       /* don't use au_sbilist_lock() here */
26271 +       spin_lock(&au_sbilist.spin);
26272 +       hlist_for_each_entry(sbinfo, &au_sbilist.head, si_list)
26273 +               if (id == sysaufs_si_id(sbinfo)) {
26274 +                       kobject_get(&sbinfo->si_kobj);
26275 +                       sb = sbinfo->si_sb;
26276 +                       break;
26277 +               }
26278 +       spin_unlock(&au_sbilist.spin);
26279 +
26280 +       err = -EINVAL;
26281 +       if (unlikely(!sb))
26282 +               goto out;
26283 +
26284 +       err = au_plink_maint_enter(sb);
26285 +       if (!err)
26286 +               /* keep kobject_get() */
26287 +               file->private_data = sbinfo;
26288 +       else
26289 +               kobject_put(&sbinfo->si_kobj);
26290 +out:
26291 +       return err;
26292 +}
26293 +
26294 +/*
26295 + * Accept a valid "si=xxxx" only.
26296 + * Once it is accepted successfully, accept "clean" too.
26297 + */
26298 +static ssize_t au_procfs_plm_write(struct file *file, const char __user *ubuf,
26299 +                                  size_t count, loff_t *ppos)
26300 +{
26301 +       ssize_t err;
26302 +       unsigned long id;
26303 +       /* last newline is allowed */
26304 +       char buf[3 + sizeof(unsigned long) * 2 + 1];
26305 +
26306 +       err = -EACCES;
26307 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
26308 +               goto out;
26309 +
26310 +       err = -EINVAL;
26311 +       if (unlikely(count > sizeof(buf)))
26312 +               goto out;
26313 +
26314 +       err = copy_from_user(buf, ubuf, count);
26315 +       if (unlikely(err)) {
26316 +               err = -EFAULT;
26317 +               goto out;
26318 +       }
26319 +       buf[count] = 0;
26320 +
26321 +       err = -EINVAL;
26322 +       if (!strcmp("clean", buf)) {
26323 +               au_procfs_plm_write_clean(file);
26324 +               goto out_success;
26325 +       } else if (unlikely(strncmp("si=", buf, 3)))
26326 +               goto out;
26327 +
26328 +       err = kstrtoul(buf + 3, 16, &id);
26329 +       if (unlikely(err))
26330 +               goto out;
26331 +
26332 +       err = au_procfs_plm_write_si(file, id);
26333 +       if (unlikely(err))
26334 +               goto out;
26335 +
26336 +out_success:
26337 +       err = count; /* success */
26338 +out:
26339 +       return err;
26340 +}
26341 +
26342 +static const struct file_operations au_procfs_plm_fop = {
26343 +       .write          = au_procfs_plm_write,
26344 +       .release        = au_procfs_plm_release,
26345 +       .owner          = THIS_MODULE
26346 +};
26347 +
26348 +/* ---------------------------------------------------------------------- */
26349 +
26350 +static struct proc_dir_entry *au_procfs_dir;
26351 +
26352 +void au_procfs_fin(void)
26353 +{
26354 +       remove_proc_entry(AUFS_PLINK_MAINT_NAME, au_procfs_dir);
26355 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
26356 +}
26357 +
26358 +int __init au_procfs_init(void)
26359 +{
26360 +       int err;
26361 +       struct proc_dir_entry *entry;
26362 +
26363 +       err = -ENOMEM;
26364 +       au_procfs_dir = proc_mkdir(AUFS_PLINK_MAINT_DIR, NULL);
26365 +       if (unlikely(!au_procfs_dir))
26366 +               goto out;
26367 +
26368 +       entry = proc_create(AUFS_PLINK_MAINT_NAME, S_IFREG | S_IWUSR,
26369 +                           au_procfs_dir, &au_procfs_plm_fop);
26370 +       if (unlikely(!entry))
26371 +               goto out_dir;
26372 +
26373 +       err = 0;
26374 +       goto out; /* success */
26375 +
26376 +
26377 +out_dir:
26378 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
26379 +out:
26380 +       return err;
26381 +}
26382 diff -urN /usr/share/empty/fs/aufs/rdu.c linux/fs/aufs/rdu.c
26383 --- /usr/share/empty/fs/aufs/rdu.c      1970-01-01 01:00:00.000000000 +0100
26384 +++ linux/fs/aufs/rdu.c 2017-05-06 22:16:52.781553511 +0200
26385 @@ -0,0 +1,381 @@
26386 +/*
26387 + * Copyright (C) 2005-2017 Junjiro R. Okajima
26388 + *
26389 + * This program, aufs is free software; you can redistribute it and/or modify
26390 + * it under the terms of the GNU General Public License as published by
26391 + * the Free Software Foundation; either version 2 of the License, or
26392 + * (at your option) any later version.
26393 + *
26394 + * This program is distributed in the hope that it will be useful,
26395 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26396 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26397 + * GNU General Public License for more details.
26398 + *
26399 + * You should have received a copy of the GNU General Public License
26400 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26401 + */
26402 +
26403 +/*
26404 + * readdir in userspace.
26405 + */
26406 +
26407 +#include <linux/compat.h>
26408 +#include <linux/fs_stack.h>
26409 +#include <linux/security.h>
26410 +#include "aufs.h"
26411 +
26412 +/* bits for struct aufs_rdu.flags */
26413 +#define        AuRdu_CALLED    1
26414 +#define        AuRdu_CONT      (1 << 1)
26415 +#define        AuRdu_FULL      (1 << 2)
26416 +#define au_ftest_rdu(flags, name)      ((flags) & AuRdu_##name)
26417 +#define au_fset_rdu(flags, name) \
26418 +       do { (flags) |= AuRdu_##name; } while (0)
26419 +#define au_fclr_rdu(flags, name) \
26420 +       do { (flags) &= ~AuRdu_##name; } while (0)
26421 +
26422 +struct au_rdu_arg {
26423 +       struct dir_context              ctx;
26424 +       struct aufs_rdu                 *rdu;
26425 +       union au_rdu_ent_ul             ent;
26426 +       unsigned long                   end;
26427 +
26428 +       struct super_block              *sb;
26429 +       int                             err;
26430 +};
26431 +
26432 +static int au_rdu_fill(struct dir_context *ctx, const char *name, int nlen,
26433 +                      loff_t offset, u64 h_ino, unsigned int d_type)
26434 +{
26435 +       int err, len;
26436 +       struct au_rdu_arg *arg = container_of(ctx, struct au_rdu_arg, ctx);
26437 +       struct aufs_rdu *rdu = arg->rdu;
26438 +       struct au_rdu_ent ent;
26439 +
26440 +       err = 0;
26441 +       arg->err = 0;
26442 +       au_fset_rdu(rdu->cookie.flags, CALLED);
26443 +       len = au_rdu_len(nlen);
26444 +       if (arg->ent.ul + len  < arg->end) {
26445 +               ent.ino = h_ino;
26446 +               ent.bindex = rdu->cookie.bindex;
26447 +               ent.type = d_type;
26448 +               ent.nlen = nlen;
26449 +               if (unlikely(nlen > AUFS_MAX_NAMELEN))
26450 +                       ent.type = DT_UNKNOWN;
26451 +
26452 +               /* unnecessary to support mmap_sem since this is a dir */
26453 +               err = -EFAULT;
26454 +               if (copy_to_user(arg->ent.e, &ent, sizeof(ent)))
26455 +                       goto out;
26456 +               if (copy_to_user(arg->ent.e->name, name, nlen))
26457 +                       goto out;
26458 +               /* the terminating NULL */
26459 +               if (__put_user(0, arg->ent.e->name + nlen))
26460 +                       goto out;
26461 +               err = 0;
26462 +               /* AuDbg("%p, %.*s\n", arg->ent.p, nlen, name); */
26463 +               arg->ent.ul += len;
26464 +               rdu->rent++;
26465 +       } else {
26466 +               err = -EFAULT;
26467 +               au_fset_rdu(rdu->cookie.flags, FULL);
26468 +               rdu->full = 1;
26469 +               rdu->tail = arg->ent;
26470 +       }
26471 +
26472 +out:
26473 +       /* AuTraceErr(err); */
26474 +       return err;
26475 +}
26476 +
26477 +static int au_rdu_do(struct file *h_file, struct au_rdu_arg *arg)
26478 +{
26479 +       int err;
26480 +       loff_t offset;
26481 +       struct au_rdu_cookie *cookie = &arg->rdu->cookie;
26482 +
26483 +       /* we don't have to care (FMODE_32BITHASH | FMODE_64BITHASH) for ext4 */
26484 +       offset = vfsub_llseek(h_file, cookie->h_pos, SEEK_SET);
26485 +       err = offset;
26486 +       if (unlikely(offset != cookie->h_pos))
26487 +               goto out;
26488 +
26489 +       err = 0;
26490 +       do {
26491 +               arg->err = 0;
26492 +               au_fclr_rdu(cookie->flags, CALLED);
26493 +               /* smp_mb(); */
26494 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
26495 +               if (err >= 0)
26496 +                       err = arg->err;
26497 +       } while (!err
26498 +                && au_ftest_rdu(cookie->flags, CALLED)
26499 +                && !au_ftest_rdu(cookie->flags, FULL));
26500 +       cookie->h_pos = h_file->f_pos;
26501 +
26502 +out:
26503 +       AuTraceErr(err);
26504 +       return err;
26505 +}
26506 +
26507 +static int au_rdu(struct file *file, struct aufs_rdu *rdu)
26508 +{
26509 +       int err;
26510 +       aufs_bindex_t bbot;
26511 +       struct au_rdu_arg arg = {
26512 +               .ctx = {
26513 +                       .actor = au_rdu_fill
26514 +               }
26515 +       };
26516 +       struct dentry *dentry;
26517 +       struct inode *inode;
26518 +       struct file *h_file;
26519 +       struct au_rdu_cookie *cookie = &rdu->cookie;
26520 +
26521 +       err = !access_ok(VERIFY_WRITE, rdu->ent.e, rdu->sz);
26522 +       if (unlikely(err)) {
26523 +               err = -EFAULT;
26524 +               AuTraceErr(err);
26525 +               goto out;
26526 +       }
26527 +       rdu->rent = 0;
26528 +       rdu->tail = rdu->ent;
26529 +       rdu->full = 0;
26530 +       arg.rdu = rdu;
26531 +       arg.ent = rdu->ent;
26532 +       arg.end = arg.ent.ul;
26533 +       arg.end += rdu->sz;
26534 +
26535 +       err = -ENOTDIR;
26536 +       if (unlikely(!file->f_op->iterate && !file->f_op->iterate_shared))
26537 +               goto out;
26538 +
26539 +       err = security_file_permission(file, MAY_READ);
26540 +       AuTraceErr(err);
26541 +       if (unlikely(err))
26542 +               goto out;
26543 +
26544 +       dentry = file->f_path.dentry;
26545 +       inode = d_inode(dentry);
26546 +       inode_lock_shared(inode);
26547 +
26548 +       arg.sb = inode->i_sb;
26549 +       err = si_read_lock(arg.sb, AuLock_FLUSH | AuLock_NOPLM);
26550 +       if (unlikely(err))
26551 +               goto out_mtx;
26552 +       err = au_alive_dir(dentry);
26553 +       if (unlikely(err))
26554 +               goto out_si;
26555 +       /* todo: reval? */
26556 +       fi_read_lock(file);
26557 +
26558 +       err = -EAGAIN;
26559 +       if (unlikely(au_ftest_rdu(cookie->flags, CONT)
26560 +                    && cookie->generation != au_figen(file)))
26561 +               goto out_unlock;
26562 +
26563 +       err = 0;
26564 +       if (!rdu->blk) {
26565 +               rdu->blk = au_sbi(arg.sb)->si_rdblk;
26566 +               if (!rdu->blk)
26567 +                       rdu->blk = au_dir_size(file, /*dentry*/NULL);
26568 +       }
26569 +       bbot = au_fbtop(file);
26570 +       if (cookie->bindex < bbot)
26571 +               cookie->bindex = bbot;
26572 +       bbot = au_fbbot_dir(file);
26573 +       /* AuDbg("b%d, b%d\n", cookie->bindex, bbot); */
26574 +       for (; !err && cookie->bindex <= bbot;
26575 +            cookie->bindex++, cookie->h_pos = 0) {
26576 +               h_file = au_hf_dir(file, cookie->bindex);
26577 +               if (!h_file)
26578 +                       continue;
26579 +
26580 +               au_fclr_rdu(cookie->flags, FULL);
26581 +               err = au_rdu_do(h_file, &arg);
26582 +               AuTraceErr(err);
26583 +               if (unlikely(au_ftest_rdu(cookie->flags, FULL) || err))
26584 +                       break;
26585 +       }
26586 +       AuDbg("rent %llu\n", rdu->rent);
26587 +
26588 +       if (!err && !au_ftest_rdu(cookie->flags, CONT)) {
26589 +               rdu->shwh = !!au_opt_test(au_sbi(arg.sb)->si_mntflags, SHWH);
26590 +               au_fset_rdu(cookie->flags, CONT);
26591 +               cookie->generation = au_figen(file);
26592 +       }
26593 +
26594 +       ii_read_lock_child(inode);
26595 +       fsstack_copy_attr_atime(inode, au_h_iptr(inode, au_ibtop(inode)));
26596 +       ii_read_unlock(inode);
26597 +
26598 +out_unlock:
26599 +       fi_read_unlock(file);
26600 +out_si:
26601 +       si_read_unlock(arg.sb);
26602 +out_mtx:
26603 +       inode_unlock_shared(inode);
26604 +out:
26605 +       AuTraceErr(err);
26606 +       return err;
26607 +}
26608 +
26609 +static int au_rdu_ino(struct file *file, struct aufs_rdu *rdu)
26610 +{
26611 +       int err;
26612 +       ino_t ino;
26613 +       unsigned long long nent;
26614 +       union au_rdu_ent_ul *u;
26615 +       struct au_rdu_ent ent;
26616 +       struct super_block *sb;
26617 +
26618 +       err = 0;
26619 +       nent = rdu->nent;
26620 +       u = &rdu->ent;
26621 +       sb = file->f_path.dentry->d_sb;
26622 +       si_read_lock(sb, AuLock_FLUSH);
26623 +       while (nent-- > 0) {
26624 +               /* unnecessary to support mmap_sem since this is a dir */
26625 +               err = copy_from_user(&ent, u->e, sizeof(ent));
26626 +               if (!err)
26627 +                       err = !access_ok(VERIFY_WRITE, &u->e->ino, sizeof(ino));
26628 +               if (unlikely(err)) {
26629 +                       err = -EFAULT;
26630 +                       AuTraceErr(err);
26631 +                       break;
26632 +               }
26633 +
26634 +               /* AuDbg("b%d, i%llu\n", ent.bindex, ent.ino); */
26635 +               if (!ent.wh)
26636 +                       err = au_ino(sb, ent.bindex, ent.ino, ent.type, &ino);
26637 +               else
26638 +                       err = au_wh_ino(sb, ent.bindex, ent.ino, ent.type,
26639 +                                       &ino);
26640 +               if (unlikely(err)) {
26641 +                       AuTraceErr(err);
26642 +                       break;
26643 +               }
26644 +
26645 +               err = __put_user(ino, &u->e->ino);
26646 +               if (unlikely(err)) {
26647 +                       err = -EFAULT;
26648 +                       AuTraceErr(err);
26649 +                       break;
26650 +               }
26651 +               u->ul += au_rdu_len(ent.nlen);
26652 +       }
26653 +       si_read_unlock(sb);
26654 +
26655 +       return err;
26656 +}
26657 +
26658 +/* ---------------------------------------------------------------------- */
26659 +
26660 +static int au_rdu_verify(struct aufs_rdu *rdu)
26661 +{
26662 +       AuDbg("rdu{%llu, %p, %u | %u | %llu, %u, %u | "
26663 +             "%llu, b%d, 0x%x, g%u}\n",
26664 +             rdu->sz, rdu->ent.e, rdu->verify[AufsCtlRduV_SZ],
26665 +             rdu->blk,
26666 +             rdu->rent, rdu->shwh, rdu->full,
26667 +             rdu->cookie.h_pos, rdu->cookie.bindex, rdu->cookie.flags,
26668 +             rdu->cookie.generation);
26669 +
26670 +       if (rdu->verify[AufsCtlRduV_SZ] == sizeof(*rdu))
26671 +               return 0;
26672 +
26673 +       AuDbg("%u:%u\n",
26674 +             rdu->verify[AufsCtlRduV_SZ], (unsigned int)sizeof(*rdu));
26675 +       return -EINVAL;
26676 +}
26677 +
26678 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
26679 +{
26680 +       long err, e;
26681 +       struct aufs_rdu rdu;
26682 +       void __user *p = (void __user *)arg;
26683 +
26684 +       err = copy_from_user(&rdu, p, sizeof(rdu));
26685 +       if (unlikely(err)) {
26686 +               err = -EFAULT;
26687 +               AuTraceErr(err);
26688 +               goto out;
26689 +       }
26690 +       err = au_rdu_verify(&rdu);
26691 +       if (unlikely(err))
26692 +               goto out;
26693 +
26694 +       switch (cmd) {
26695 +       case AUFS_CTL_RDU:
26696 +               err = au_rdu(file, &rdu);
26697 +               if (unlikely(err))
26698 +                       break;
26699 +
26700 +               e = copy_to_user(p, &rdu, sizeof(rdu));
26701 +               if (unlikely(e)) {
26702 +                       err = -EFAULT;
26703 +                       AuTraceErr(err);
26704 +               }
26705 +               break;
26706 +       case AUFS_CTL_RDU_INO:
26707 +               err = au_rdu_ino(file, &rdu);
26708 +               break;
26709 +
26710 +       default:
26711 +               /* err = -ENOTTY; */
26712 +               err = -EINVAL;
26713 +       }
26714 +
26715 +out:
26716 +       AuTraceErr(err);
26717 +       return err;
26718 +}
26719 +
26720 +#ifdef CONFIG_COMPAT
26721 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
26722 +{
26723 +       long err, e;
26724 +       struct aufs_rdu rdu;
26725 +       void __user *p = compat_ptr(arg);
26726 +
26727 +       /* todo: get_user()? */
26728 +       err = copy_from_user(&rdu, p, sizeof(rdu));
26729 +       if (unlikely(err)) {
26730 +               err = -EFAULT;
26731 +               AuTraceErr(err);
26732 +               goto out;
26733 +       }
26734 +       rdu.ent.e = compat_ptr(rdu.ent.ul);
26735 +       err = au_rdu_verify(&rdu);
26736 +       if (unlikely(err))
26737 +               goto out;
26738 +
26739 +       switch (cmd) {
26740 +       case AUFS_CTL_RDU:
26741 +               err = au_rdu(file, &rdu);
26742 +               if (unlikely(err))
26743 +                       break;
26744 +
26745 +               rdu.ent.ul = ptr_to_compat(rdu.ent.e);
26746 +               rdu.tail.ul = ptr_to_compat(rdu.tail.e);
26747 +               e = copy_to_user(p, &rdu, sizeof(rdu));
26748 +               if (unlikely(e)) {
26749 +                       err = -EFAULT;
26750 +                       AuTraceErr(err);
26751 +               }
26752 +               break;
26753 +       case AUFS_CTL_RDU_INO:
26754 +               err = au_rdu_ino(file, &rdu);
26755 +               break;
26756 +
26757 +       default:
26758 +               /* err = -ENOTTY; */
26759 +               err = -EINVAL;
26760 +       }
26761 +
26762 +out:
26763 +       AuTraceErr(err);
26764 +       return err;
26765 +}
26766 +#endif
26767 diff -urN /usr/share/empty/fs/aufs/rwsem.h linux/fs/aufs/rwsem.h
26768 --- /usr/share/empty/fs/aufs/rwsem.h    1970-01-01 01:00:00.000000000 +0100
26769 +++ linux/fs/aufs/rwsem.h       2017-05-06 22:16:52.781553511 +0200
26770 @@ -0,0 +1,198 @@
26771 +/*
26772 + * Copyright (C) 2005-2017 Junjiro R. Okajima
26773 + *
26774 + * This program, aufs is free software; you can redistribute it and/or modify
26775 + * it under the terms of the GNU General Public License as published by
26776 + * the Free Software Foundation; either version 2 of the License, or
26777 + * (at your option) any later version.
26778 + *
26779 + * This program is distributed in the hope that it will be useful,
26780 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26781 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26782 + * GNU General Public License for more details.
26783 + *
26784 + * You should have received a copy of the GNU General Public License
26785 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26786 + */
26787 +
26788 +/*
26789 + * simple read-write semaphore wrappers
26790 + */
26791 +
26792 +#ifndef __AUFS_RWSEM_H__
26793 +#define __AUFS_RWSEM_H__
26794 +
26795 +#ifdef __KERNEL__
26796 +
26797 +#include "debug.h"
26798 +
26799 +struct au_rwsem {
26800 +       struct rw_semaphore     rwsem;
26801 +#ifdef CONFIG_AUFS_DEBUG
26802 +       /* just for debugging, not almighty counter */
26803 +       atomic_t                rcnt, wcnt;
26804 +#endif
26805 +};
26806 +
26807 +#ifdef CONFIG_LOCKDEP
26808 +#define au_lockdep_set_name(rw)                                                \
26809 +       lockdep_set_class_and_name(&(rw)->rwsem,                        \
26810 +                                  /*original key*/(rw)->rwsem.dep_map.key, \
26811 +                                  /*name*/#rw)
26812 +#else
26813 +#define au_lockdep_set_name(rw) do {} while (0)
26814 +#endif
26815 +
26816 +#ifdef CONFIG_AUFS_DEBUG
26817 +#define AuDbgCntInit(rw) do { \
26818 +       atomic_set(&(rw)->rcnt, 0); \
26819 +       atomic_set(&(rw)->wcnt, 0); \
26820 +       smp_mb(); /* atomic set */ \
26821 +} while (0)
26822 +
26823 +#define AuDbgCnt(rw, cnt)      atomic_read(&(rw)->cnt)
26824 +#define AuDbgCntInc(rw, cnt)   atomic_inc(&(rw)->cnt)
26825 +#define AuDbgCntDec(rw, cnt)   WARN_ON(atomic_dec_return(&(rw)->cnt) < 0)
26826 +#define AuDbgRcntInc(rw)       AuDbgCntInc(rw, rcnt)
26827 +#define AuDbgRcntDec(rw)       AuDbgCntDec(rw, rcnt)
26828 +#define AuDbgWcntInc(rw)       AuDbgCntInc(rw, wcnt)
26829 +#define AuDbgWcntDec(rw)       AuDbgCntDec(rw, wcnt)
26830 +#else
26831 +#define AuDbgCnt(rw, cnt)      0
26832 +#define AuDbgCntInit(rw)       do {} while (0)
26833 +#define AuDbgRcntInc(rw)       do {} while (0)
26834 +#define AuDbgRcntDec(rw)       do {} while (0)
26835 +#define AuDbgWcntInc(rw)       do {} while (0)
26836 +#define AuDbgWcntDec(rw)       do {} while (0)
26837 +#endif /* CONFIG_AUFS_DEBUG */
26838 +
26839 +/* to debug easier, do not make them inlined functions */
26840 +#define AuRwMustNoWaiters(rw)  AuDebugOn(rwsem_is_contended(&(rw)->rwsem))
26841 +/* rwsem_is_locked() is unusable */
26842 +#define AuRwMustReadLock(rw)   AuDebugOn(AuDbgCnt(rw, rcnt) <= 0)
26843 +#define AuRwMustWriteLock(rw)  AuDebugOn(AuDbgCnt(rw, wcnt) <= 0)
26844 +#define AuRwMustAnyLock(rw)    AuDebugOn(AuDbgCnt(rw, rcnt) <= 0       \
26845 +                                         && AuDbgCnt(rw, wcnt) <= 0)
26846 +#define AuRwDestroy(rw)                AuDebugOn(AuDbgCnt(rw, rcnt)            \
26847 +                                         || AuDbgCnt(rw, wcnt))
26848 +
26849 +#define au_rw_init(rw) do {                    \
26850 +               AuDbgCntInit(rw);               \
26851 +               init_rwsem(&(rw)->rwsem);       \
26852 +               au_lockdep_set_name(rw);        \
26853 +       } while (0)
26854 +
26855 +#define au_rw_init_wlock(rw) do {              \
26856 +               au_rw_init(rw);                 \
26857 +               down_write(&(rw)->rwsem);       \
26858 +               AuDbgWcntInc(rw);               \
26859 +       } while (0)
26860 +
26861 +#define au_rw_init_wlock_nested(rw, lsc) do { \
26862 +               au_rw_init(rw);                         \
26863 +               down_write_nested(&(rw)->rwsem, lsc);   \
26864 +               AuDbgWcntInc(rw);                       \
26865 +       } while (0)
26866 +
26867 +static inline void au_rw_read_lock(struct au_rwsem *rw)
26868 +{
26869 +       down_read(&rw->rwsem);
26870 +       AuDbgRcntInc(rw);
26871 +}
26872 +
26873 +static inline void au_rw_read_lock_nested(struct au_rwsem *rw, unsigned int lsc)
26874 +{
26875 +       down_read_nested(&rw->rwsem, lsc);
26876 +       AuDbgRcntInc(rw);
26877 +}
26878 +
26879 +static inline void au_rw_read_unlock(struct au_rwsem *rw)
26880 +{
26881 +       AuRwMustReadLock(rw);
26882 +       AuDbgRcntDec(rw);
26883 +       up_read(&rw->rwsem);
26884 +}
26885 +
26886 +static inline void au_rw_dgrade_lock(struct au_rwsem *rw)
26887 +{
26888 +       AuRwMustWriteLock(rw);
26889 +       AuDbgRcntInc(rw);
26890 +       AuDbgWcntDec(rw);
26891 +       downgrade_write(&rw->rwsem);
26892 +}
26893 +
26894 +static inline void au_rw_write_lock(struct au_rwsem *rw)
26895 +{
26896 +       down_write(&rw->rwsem);
26897 +       AuDbgWcntInc(rw);
26898 +}
26899 +
26900 +static inline void au_rw_write_lock_nested(struct au_rwsem *rw,
26901 +                                          unsigned int lsc)
26902 +{
26903 +       down_write_nested(&rw->rwsem, lsc);
26904 +       AuDbgWcntInc(rw);
26905 +}
26906 +
26907 +static inline void au_rw_write_unlock(struct au_rwsem *rw)
26908 +{
26909 +       AuRwMustWriteLock(rw);
26910 +       AuDbgWcntDec(rw);
26911 +       up_write(&rw->rwsem);
26912 +}
26913 +
26914 +/* why is not _nested version defined */
26915 +static inline int au_rw_read_trylock(struct au_rwsem *rw)
26916 +{
26917 +       int ret;
26918 +
26919 +       ret = down_read_trylock(&rw->rwsem);
26920 +       if (ret)
26921 +               AuDbgRcntInc(rw);
26922 +       return ret;
26923 +}
26924 +
26925 +static inline int au_rw_write_trylock(struct au_rwsem *rw)
26926 +{
26927 +       int ret;
26928 +
26929 +       ret = down_write_trylock(&rw->rwsem);
26930 +       if (ret)
26931 +               AuDbgWcntInc(rw);
26932 +       return ret;
26933 +}
26934 +
26935 +#undef AuDbgCntDec
26936 +#undef AuDbgRcntInc
26937 +#undef AuDbgRcntDec
26938 +#undef AuDbgWcntDec
26939 +
26940 +#define AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
26941 +static inline void prefix##_read_lock(param) \
26942 +{ au_rw_read_lock(rwsem); } \
26943 +static inline void prefix##_write_lock(param) \
26944 +{ au_rw_write_lock(rwsem); } \
26945 +static inline int prefix##_read_trylock(param) \
26946 +{ return au_rw_read_trylock(rwsem); } \
26947 +static inline int prefix##_write_trylock(param) \
26948 +{ return au_rw_write_trylock(rwsem); }
26949 +/* why is not _nested version defined */
26950 +/* static inline void prefix##_read_trylock_nested(param, lsc)
26951 +{ au_rw_read_trylock_nested(rwsem, lsc)); }
26952 +static inline void prefix##_write_trylock_nestd(param, lsc)
26953 +{ au_rw_write_trylock_nested(rwsem, lsc); } */
26954 +
26955 +#define AuSimpleUnlockRwsemFuncs(prefix, param, rwsem) \
26956 +static inline void prefix##_read_unlock(param) \
26957 +{ au_rw_read_unlock(rwsem); } \
26958 +static inline void prefix##_write_unlock(param) \
26959 +{ au_rw_write_unlock(rwsem); } \
26960 +static inline void prefix##_downgrade_lock(param) \
26961 +{ au_rw_dgrade_lock(rwsem); }
26962 +
26963 +#define AuSimpleRwsemFuncs(prefix, param, rwsem) \
26964 +       AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
26965 +       AuSimpleUnlockRwsemFuncs(prefix, param, rwsem)
26966 +
26967 +#endif /* __KERNEL__ */
26968 +#endif /* __AUFS_RWSEM_H__ */
26969 diff -urN /usr/share/empty/fs/aufs/sbinfo.c linux/fs/aufs/sbinfo.c
26970 --- /usr/share/empty/fs/aufs/sbinfo.c   1970-01-01 01:00:00.000000000 +0100
26971 +++ linux/fs/aufs/sbinfo.c      2017-05-06 22:16:52.781553511 +0200
26972 @@ -0,0 +1,304 @@
26973 +/*
26974 + * Copyright (C) 2005-2017 Junjiro R. Okajima
26975 + *
26976 + * This program, aufs is free software; you can redistribute it and/or modify
26977 + * it under the terms of the GNU General Public License as published by
26978 + * the Free Software Foundation; either version 2 of the License, or
26979 + * (at your option) any later version.
26980 + *
26981 + * This program is distributed in the hope that it will be useful,
26982 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26983 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26984 + * GNU General Public License for more details.
26985 + *
26986 + * You should have received a copy of the GNU General Public License
26987 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26988 + */
26989 +
26990 +/*
26991 + * superblock private data
26992 + */
26993 +
26994 +#include "aufs.h"
26995 +
26996 +/*
26997 + * they are necessary regardless sysfs is disabled.
26998 + */
26999 +void au_si_free(struct kobject *kobj)
27000 +{
27001 +       int i;
27002 +       struct au_sbinfo *sbinfo;
27003 +       char *locked __maybe_unused; /* debug only */
27004 +
27005 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
27006 +       for (i = 0; i < AuPlink_NHASH; i++)
27007 +               AuDebugOn(!hlist_empty(&sbinfo->si_plink[i].head));
27008 +       AuDebugOn(atomic_read(&sbinfo->si_nowait.nw_len));
27009 +
27010 +       AuDebugOn(percpu_counter_sum(&sbinfo->si_ninodes));
27011 +       percpu_counter_destroy(&sbinfo->si_ninodes);
27012 +       AuDebugOn(percpu_counter_sum(&sbinfo->si_nfiles));
27013 +       percpu_counter_destroy(&sbinfo->si_nfiles);
27014 +
27015 +       au_rw_write_lock(&sbinfo->si_rwsem);
27016 +       au_br_free(sbinfo);
27017 +       au_rw_write_unlock(&sbinfo->si_rwsem);
27018 +
27019 +       au_delayed_kfree(sbinfo->si_branch);
27020 +       mutex_destroy(&sbinfo->si_xib_mtx);
27021 +       AuRwDestroy(&sbinfo->si_rwsem);
27022 +
27023 +       au_delayed_kfree(sbinfo);
27024 +}
27025 +
27026 +int au_si_alloc(struct super_block *sb)
27027 +{
27028 +       int err, i;
27029 +       struct au_sbinfo *sbinfo;
27030 +
27031 +       err = -ENOMEM;
27032 +       sbinfo = kzalloc(sizeof(*sbinfo), GFP_NOFS);
27033 +       if (unlikely(!sbinfo))
27034 +               goto out;
27035 +
27036 +       /* will be reallocated separately */
27037 +       sbinfo->si_branch = kzalloc(sizeof(*sbinfo->si_branch), GFP_NOFS);
27038 +       if (unlikely(!sbinfo->si_branch))
27039 +               goto out_sbinfo;
27040 +
27041 +       err = sysaufs_si_init(sbinfo);
27042 +       if (unlikely(err))
27043 +               goto out_br;
27044 +
27045 +       au_nwt_init(&sbinfo->si_nowait);
27046 +       au_rw_init_wlock(&sbinfo->si_rwsem);
27047 +
27048 +       percpu_counter_init(&sbinfo->si_ninodes, 0, GFP_NOFS);
27049 +       percpu_counter_init(&sbinfo->si_nfiles, 0, GFP_NOFS);
27050 +
27051 +       sbinfo->si_bbot = -1;
27052 +       sbinfo->si_last_br_id = AUFS_BRANCH_MAX / 2;
27053 +
27054 +       sbinfo->si_wbr_copyup = AuWbrCopyup_Def;
27055 +       sbinfo->si_wbr_create = AuWbrCreate_Def;
27056 +       sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + sbinfo->si_wbr_copyup;
27057 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + sbinfo->si_wbr_create;
27058 +
27059 +       au_fhsm_init(sbinfo);
27060 +
27061 +       sbinfo->si_mntflags = au_opts_plink(AuOpt_Def);
27062 +
27063 +       sbinfo->si_xino_jiffy = jiffies;
27064 +       sbinfo->si_xino_expire
27065 +               = msecs_to_jiffies(AUFS_XINO_DEF_SEC * MSEC_PER_SEC);
27066 +       mutex_init(&sbinfo->si_xib_mtx);
27067 +       sbinfo->si_xino_brid = -1;
27068 +       /* leave si_xib_last_pindex and si_xib_next_bit */
27069 +
27070 +       au_sphl_init(&sbinfo->si_aopen);
27071 +
27072 +       sbinfo->si_rdcache = msecs_to_jiffies(AUFS_RDCACHE_DEF * MSEC_PER_SEC);
27073 +       sbinfo->si_rdblk = AUFS_RDBLK_DEF;
27074 +       sbinfo->si_rdhash = AUFS_RDHASH_DEF;
27075 +       sbinfo->si_dirwh = AUFS_DIRWH_DEF;
27076 +
27077 +       for (i = 0; i < AuPlink_NHASH; i++)
27078 +               au_sphl_init(sbinfo->si_plink + i);
27079 +       init_waitqueue_head(&sbinfo->si_plink_wq);
27080 +       spin_lock_init(&sbinfo->si_plink_maint_lock);
27081 +
27082 +       au_sphl_init(&sbinfo->si_files);
27083 +
27084 +       /* with getattr by default */
27085 +       sbinfo->si_iop_array = aufs_iop;
27086 +
27087 +       /* leave other members for sysaufs and si_mnt. */
27088 +       sbinfo->si_sb = sb;
27089 +       sb->s_fs_info = sbinfo;
27090 +       si_pid_set(sb);
27091 +       return 0; /* success */
27092 +
27093 +out_br:
27094 +       au_delayed_kfree(sbinfo->si_branch);
27095 +out_sbinfo:
27096 +       au_delayed_kfree(sbinfo);
27097 +out:
27098 +       return err;
27099 +}
27100 +
27101 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr, int may_shrink)
27102 +{
27103 +       int err, sz;
27104 +       struct au_branch **brp;
27105 +
27106 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
27107 +
27108 +       err = -ENOMEM;
27109 +       sz = sizeof(*brp) * (sbinfo->si_bbot + 1);
27110 +       if (unlikely(!sz))
27111 +               sz = sizeof(*brp);
27112 +       brp = au_kzrealloc(sbinfo->si_branch, sz, sizeof(*brp) * nbr, GFP_NOFS,
27113 +                          may_shrink);
27114 +       if (brp) {
27115 +               sbinfo->si_branch = brp;
27116 +               err = 0;
27117 +       }
27118 +
27119 +       return err;
27120 +}
27121 +
27122 +/* ---------------------------------------------------------------------- */
27123 +
27124 +unsigned int au_sigen_inc(struct super_block *sb)
27125 +{
27126 +       unsigned int gen;
27127 +       struct inode *inode;
27128 +
27129 +       SiMustWriteLock(sb);
27130 +
27131 +       gen = ++au_sbi(sb)->si_generation;
27132 +       au_update_digen(sb->s_root);
27133 +       inode = d_inode(sb->s_root);
27134 +       au_update_iigen(inode, /*half*/0);
27135 +       inode->i_version++;
27136 +       return gen;
27137 +}
27138 +
27139 +aufs_bindex_t au_new_br_id(struct super_block *sb)
27140 +{
27141 +       aufs_bindex_t br_id;
27142 +       int i;
27143 +       struct au_sbinfo *sbinfo;
27144 +
27145 +       SiMustWriteLock(sb);
27146 +
27147 +       sbinfo = au_sbi(sb);
27148 +       for (i = 0; i <= AUFS_BRANCH_MAX; i++) {
27149 +               br_id = ++sbinfo->si_last_br_id;
27150 +               AuDebugOn(br_id < 0);
27151 +               if (br_id && au_br_index(sb, br_id) < 0)
27152 +                       return br_id;
27153 +       }
27154 +
27155 +       return -1;
27156 +}
27157 +
27158 +/* ---------------------------------------------------------------------- */
27159 +
27160 +/* it is ok that new 'nwt' tasks are appended while we are sleeping */
27161 +int si_read_lock(struct super_block *sb, int flags)
27162 +{
27163 +       int err;
27164 +
27165 +       err = 0;
27166 +       if (au_ftest_lock(flags, FLUSH))
27167 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
27168 +
27169 +       si_noflush_read_lock(sb);
27170 +       err = au_plink_maint(sb, flags);
27171 +       if (unlikely(err))
27172 +               si_read_unlock(sb);
27173 +
27174 +       return err;
27175 +}
27176 +
27177 +int si_write_lock(struct super_block *sb, int flags)
27178 +{
27179 +       int err;
27180 +
27181 +       if (au_ftest_lock(flags, FLUSH))
27182 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
27183 +
27184 +       si_noflush_write_lock(sb);
27185 +       err = au_plink_maint(sb, flags);
27186 +       if (unlikely(err))
27187 +               si_write_unlock(sb);
27188 +
27189 +       return err;
27190 +}
27191 +
27192 +/* dentry and super_block lock. call at entry point */
27193 +int aufs_read_lock(struct dentry *dentry, int flags)
27194 +{
27195 +       int err;
27196 +       struct super_block *sb;
27197 +
27198 +       sb = dentry->d_sb;
27199 +       err = si_read_lock(sb, flags);
27200 +       if (unlikely(err))
27201 +               goto out;
27202 +
27203 +       if (au_ftest_lock(flags, DW))
27204 +               di_write_lock_child(dentry);
27205 +       else
27206 +               di_read_lock_child(dentry, flags);
27207 +
27208 +       if (au_ftest_lock(flags, GEN)) {
27209 +               err = au_digen_test(dentry, au_sigen(sb));
27210 +               if (!au_opt_test(au_mntflags(sb), UDBA_NONE))
27211 +                       AuDebugOn(!err && au_dbrange_test(dentry));
27212 +               else if (!err)
27213 +                       err = au_dbrange_test(dentry);
27214 +               if (unlikely(err))
27215 +                       aufs_read_unlock(dentry, flags);
27216 +       }
27217 +
27218 +out:
27219 +       return err;
27220 +}
27221 +
27222 +void aufs_read_unlock(struct dentry *dentry, int flags)
27223 +{
27224 +       if (au_ftest_lock(flags, DW))
27225 +               di_write_unlock(dentry);
27226 +       else
27227 +               di_read_unlock(dentry, flags);
27228 +       si_read_unlock(dentry->d_sb);
27229 +}
27230 +
27231 +void aufs_write_lock(struct dentry *dentry)
27232 +{
27233 +       si_write_lock(dentry->d_sb, AuLock_FLUSH | AuLock_NOPLMW);
27234 +       di_write_lock_child(dentry);
27235 +}
27236 +
27237 +void aufs_write_unlock(struct dentry *dentry)
27238 +{
27239 +       di_write_unlock(dentry);
27240 +       si_write_unlock(dentry->d_sb);
27241 +}
27242 +
27243 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags)
27244 +{
27245 +       int err;
27246 +       unsigned int sigen;
27247 +       struct super_block *sb;
27248 +
27249 +       sb = d1->d_sb;
27250 +       err = si_read_lock(sb, flags);
27251 +       if (unlikely(err))
27252 +               goto out;
27253 +
27254 +       di_write_lock2_child(d1, d2, au_ftest_lock(flags, DIRS));
27255 +
27256 +       if (au_ftest_lock(flags, GEN)) {
27257 +               sigen = au_sigen(sb);
27258 +               err = au_digen_test(d1, sigen);
27259 +               AuDebugOn(!err && au_dbrange_test(d1));
27260 +               if (!err) {
27261 +                       err = au_digen_test(d2, sigen);
27262 +                       AuDebugOn(!err && au_dbrange_test(d2));
27263 +               }
27264 +               if (unlikely(err))
27265 +                       aufs_read_and_write_unlock2(d1, d2);
27266 +       }
27267 +
27268 +out:
27269 +       return err;
27270 +}
27271 +
27272 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2)
27273 +{
27274 +       di_write_unlock2(d1, d2);
27275 +       si_read_unlock(d1->d_sb);
27276 +}
27277 diff -urN /usr/share/empty/fs/aufs/spl.h linux/fs/aufs/spl.h
27278 --- /usr/share/empty/fs/aufs/spl.h      1970-01-01 01:00:00.000000000 +0100
27279 +++ linux/fs/aufs/spl.h 2017-05-06 22:16:52.781553511 +0200
27280 @@ -0,0 +1,113 @@
27281 +/*
27282 + * Copyright (C) 2005-2017 Junjiro R. Okajima
27283 + *
27284 + * This program, aufs is free software; you can redistribute it and/or modify
27285 + * it under the terms of the GNU General Public License as published by
27286 + * the Free Software Foundation; either version 2 of the License, or
27287 + * (at your option) any later version.
27288 + *
27289 + * This program is distributed in the hope that it will be useful,
27290 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27291 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27292 + * GNU General Public License for more details.
27293 + *
27294 + * You should have received a copy of the GNU General Public License
27295 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27296 + */
27297 +
27298 +/*
27299 + * simple list protected by a spinlock
27300 + */
27301 +
27302 +#ifndef __AUFS_SPL_H__
27303 +#define __AUFS_SPL_H__
27304 +
27305 +#ifdef __KERNEL__
27306 +
27307 +#if 0
27308 +struct au_splhead {
27309 +       spinlock_t              spin;
27310 +       struct list_head        head;
27311 +};
27312 +
27313 +static inline void au_spl_init(struct au_splhead *spl)
27314 +{
27315 +       spin_lock_init(&spl->spin);
27316 +       INIT_LIST_HEAD(&spl->head);
27317 +}
27318 +
27319 +static inline void au_spl_add(struct list_head *list, struct au_splhead *spl)
27320 +{
27321 +       spin_lock(&spl->spin);
27322 +       list_add(list, &spl->head);
27323 +       spin_unlock(&spl->spin);
27324 +}
27325 +
27326 +static inline void au_spl_del(struct list_head *list, struct au_splhead *spl)
27327 +{
27328 +       spin_lock(&spl->spin);
27329 +       list_del(list);
27330 +       spin_unlock(&spl->spin);
27331 +}
27332 +
27333 +static inline void au_spl_del_rcu(struct list_head *list,
27334 +                                 struct au_splhead *spl)
27335 +{
27336 +       spin_lock(&spl->spin);
27337 +       list_del_rcu(list);
27338 +       spin_unlock(&spl->spin);
27339 +}
27340 +#endif
27341 +
27342 +/* ---------------------------------------------------------------------- */
27343 +
27344 +struct au_sphlhead {
27345 +       spinlock_t              spin;
27346 +       struct hlist_head       head;
27347 +};
27348 +
27349 +static inline void au_sphl_init(struct au_sphlhead *sphl)
27350 +{
27351 +       spin_lock_init(&sphl->spin);
27352 +       INIT_HLIST_HEAD(&sphl->head);
27353 +}
27354 +
27355 +static inline void au_sphl_add(struct hlist_node *hlist,
27356 +                              struct au_sphlhead *sphl)
27357 +{
27358 +       spin_lock(&sphl->spin);
27359 +       hlist_add_head(hlist, &sphl->head);
27360 +       spin_unlock(&sphl->spin);
27361 +}
27362 +
27363 +static inline void au_sphl_del(struct hlist_node *hlist,
27364 +                              struct au_sphlhead *sphl)
27365 +{
27366 +       spin_lock(&sphl->spin);
27367 +       hlist_del(hlist);
27368 +       spin_unlock(&sphl->spin);
27369 +}
27370 +
27371 +static inline void au_sphl_del_rcu(struct hlist_node *hlist,
27372 +                                  struct au_sphlhead *sphl)
27373 +{
27374 +       spin_lock(&sphl->spin);
27375 +       hlist_del_rcu(hlist);
27376 +       spin_unlock(&sphl->spin);
27377 +}
27378 +
27379 +static inline unsigned long au_sphl_count(struct au_sphlhead *sphl)
27380 +{
27381 +       unsigned long cnt;
27382 +       struct hlist_node *pos;
27383 +
27384 +       cnt = 0;
27385 +       spin_lock(&sphl->spin);
27386 +       hlist_for_each(pos, &sphl->head)
27387 +               cnt++;
27388 +       spin_unlock(&sphl->spin);
27389 +       return cnt;
27390 +}
27391 +
27392 +#endif /* __KERNEL__ */
27393 +#endif /* __AUFS_SPL_H__ */
27394 diff -urN /usr/share/empty/fs/aufs/super.c linux/fs/aufs/super.c
27395 --- /usr/share/empty/fs/aufs/super.c    1970-01-01 01:00:00.000000000 +0100
27396 +++ linux/fs/aufs/super.c       2017-05-06 22:16:52.781553511 +0200
27397 @@ -0,0 +1,1044 @@
27398 +/*
27399 + * Copyright (C) 2005-2017 Junjiro R. Okajima
27400 + *
27401 + * This program, aufs is free software; you can redistribute it and/or modify
27402 + * it under the terms of the GNU General Public License as published by
27403 + * the Free Software Foundation; either version 2 of the License, or
27404 + * (at your option) any later version.
27405 + *
27406 + * This program is distributed in the hope that it will be useful,
27407 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27408 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27409 + * GNU General Public License for more details.
27410 + *
27411 + * You should have received a copy of the GNU General Public License
27412 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27413 + */
27414 +
27415 +/*
27416 + * mount and super_block operations
27417 + */
27418 +
27419 +#include <linux/mm.h>
27420 +#include <linux/seq_file.h>
27421 +#include <linux/statfs.h>
27422 +#include <linux/vmalloc.h>
27423 +#include "aufs.h"
27424 +
27425 +/*
27426 + * super_operations
27427 + */
27428 +static struct inode *aufs_alloc_inode(struct super_block *sb __maybe_unused)
27429 +{
27430 +       struct au_icntnr *c;
27431 +
27432 +       c = au_cache_alloc_icntnr();
27433 +       if (c) {
27434 +               au_icntnr_init(c);
27435 +               c->vfs_inode.i_version = 1; /* sigen(sb); */
27436 +               c->iinfo.ii_hinode = NULL;
27437 +               return &c->vfs_inode;
27438 +       }
27439 +       return NULL;
27440 +}
27441 +
27442 +static void aufs_destroy_inode_cb(struct rcu_head *head)
27443 +{
27444 +       struct inode *inode = container_of(head, struct inode, i_rcu);
27445 +
27446 +       au_cache_dfree_icntnr(container_of(inode, struct au_icntnr, vfs_inode));
27447 +}
27448 +
27449 +static void aufs_destroy_inode(struct inode *inode)
27450 +{
27451 +       if (!au_is_bad_inode(inode))
27452 +               au_iinfo_fin(inode);
27453 +       call_rcu(&inode->i_rcu, aufs_destroy_inode_cb);
27454 +}
27455 +
27456 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino)
27457 +{
27458 +       struct inode *inode;
27459 +       int err;
27460 +
27461 +       inode = iget_locked(sb, ino);
27462 +       if (unlikely(!inode)) {
27463 +               inode = ERR_PTR(-ENOMEM);
27464 +               goto out;
27465 +       }
27466 +       if (!(inode->i_state & I_NEW))
27467 +               goto out;
27468 +
27469 +       err = au_xigen_new(inode);
27470 +       if (!err)
27471 +               err = au_iinfo_init(inode);
27472 +       if (!err)
27473 +               inode->i_version++;
27474 +       else {
27475 +               iget_failed(inode);
27476 +               inode = ERR_PTR(err);
27477 +       }
27478 +
27479 +out:
27480 +       /* never return NULL */
27481 +       AuDebugOn(!inode);
27482 +       AuTraceErrPtr(inode);
27483 +       return inode;
27484 +}
27485 +
27486 +/* lock free root dinfo */
27487 +static int au_show_brs(struct seq_file *seq, struct super_block *sb)
27488 +{
27489 +       int err;
27490 +       aufs_bindex_t bindex, bbot;
27491 +       struct path path;
27492 +       struct au_hdentry *hdp;
27493 +       struct au_branch *br;
27494 +       au_br_perm_str_t perm;
27495 +
27496 +       err = 0;
27497 +       bbot = au_sbbot(sb);
27498 +       bindex = 0;
27499 +       hdp = au_hdentry(au_di(sb->s_root), bindex);
27500 +       for (; !err && bindex <= bbot; bindex++, hdp++) {
27501 +               br = au_sbr(sb, bindex);
27502 +               path.mnt = au_br_mnt(br);
27503 +               path.dentry = hdp->hd_dentry;
27504 +               err = au_seq_path(seq, &path);
27505 +               if (!err) {
27506 +                       au_optstr_br_perm(&perm, br->br_perm);
27507 +                       seq_printf(seq, "=%s", perm.a);
27508 +                       if (bindex != bbot)
27509 +                               seq_putc(seq, ':');
27510 +               }
27511 +       }
27512 +       if (unlikely(err || seq_has_overflowed(seq)))
27513 +               err = -E2BIG;
27514 +
27515 +       return err;
27516 +}
27517 +
27518 +static void au_gen_fmt(char *fmt, int len __maybe_unused, const char *pat,
27519 +                      const char *append)
27520 +{
27521 +       char *p;
27522 +
27523 +       p = fmt;
27524 +       while (*pat != ':')
27525 +               *p++ = *pat++;
27526 +       *p++ = *pat++;
27527 +       strcpy(p, append);
27528 +       AuDebugOn(strlen(fmt) >= len);
27529 +}
27530 +
27531 +static void au_show_wbr_create(struct seq_file *m, int v,
27532 +                              struct au_sbinfo *sbinfo)
27533 +{
27534 +       const char *pat;
27535 +       char fmt[32];
27536 +       struct au_wbr_mfs *mfs;
27537 +
27538 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
27539 +
27540 +       seq_puts(m, ",create=");
27541 +       pat = au_optstr_wbr_create(v);
27542 +       mfs = &sbinfo->si_wbr_mfs;
27543 +       switch (v) {
27544 +       case AuWbrCreate_TDP:
27545 +       case AuWbrCreate_RR:
27546 +       case AuWbrCreate_MFS:
27547 +       case AuWbrCreate_PMFS:
27548 +               seq_puts(m, pat);
27549 +               break;
27550 +       case AuWbrCreate_MFSRR:
27551 +       case AuWbrCreate_TDMFS:
27552 +       case AuWbrCreate_PMFSRR:
27553 +               au_gen_fmt(fmt, sizeof(fmt), pat, "%llu");
27554 +               seq_printf(m, fmt, mfs->mfsrr_watermark);
27555 +               break;
27556 +       case AuWbrCreate_MFSV:
27557 +       case AuWbrCreate_PMFSV:
27558 +               au_gen_fmt(fmt, sizeof(fmt), pat, "%lu");
27559 +               seq_printf(m, fmt,
27560 +                          jiffies_to_msecs(mfs->mfs_expire)
27561 +                          / MSEC_PER_SEC);
27562 +               break;
27563 +       case AuWbrCreate_MFSRRV:
27564 +       case AuWbrCreate_TDMFSV:
27565 +       case AuWbrCreate_PMFSRRV:
27566 +               au_gen_fmt(fmt, sizeof(fmt), pat, "%llu:%lu");
27567 +               seq_printf(m, fmt, mfs->mfsrr_watermark,
27568 +                          jiffies_to_msecs(mfs->mfs_expire) / MSEC_PER_SEC);
27569 +               break;
27570 +       default:
27571 +               BUG();
27572 +       }
27573 +}
27574 +
27575 +static int au_show_xino(struct seq_file *seq, struct super_block *sb)
27576 +{
27577 +#ifdef CONFIG_SYSFS
27578 +       return 0;
27579 +#else
27580 +       int err;
27581 +       const int len = sizeof(AUFS_XINO_FNAME) - 1;
27582 +       aufs_bindex_t bindex, brid;
27583 +       struct qstr *name;
27584 +       struct file *f;
27585 +       struct dentry *d, *h_root;
27586 +
27587 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
27588 +
27589 +       err = 0;
27590 +       f = au_sbi(sb)->si_xib;
27591 +       if (!f)
27592 +               goto out;
27593 +
27594 +       /* stop printing the default xino path on the first writable branch */
27595 +       h_root = NULL;
27596 +       brid = au_xino_brid(sb);
27597 +       if (brid >= 0) {
27598 +               bindex = au_br_index(sb, brid);
27599 +               h_root = au_hdentry(au_di(sb->s_root), bindex)->hd_dentry;
27600 +       }
27601 +       d = f->f_path.dentry;
27602 +       name = &d->d_name;
27603 +       /* safe ->d_parent because the file is unlinked */
27604 +       if (d->d_parent == h_root
27605 +           && name->len == len
27606 +           && !memcmp(name->name, AUFS_XINO_FNAME, len))
27607 +               goto out;
27608 +
27609 +       seq_puts(seq, ",xino=");
27610 +       err = au_xino_path(seq, f);
27611 +
27612 +out:
27613 +       return err;
27614 +#endif
27615 +}
27616 +
27617 +/* seq_file will re-call me in case of too long string */
27618 +static int aufs_show_options(struct seq_file *m, struct dentry *dentry)
27619 +{
27620 +       int err;
27621 +       unsigned int mnt_flags, v;
27622 +       struct super_block *sb;
27623 +       struct au_sbinfo *sbinfo;
27624 +
27625 +#define AuBool(name, str) do { \
27626 +       v = au_opt_test(mnt_flags, name); \
27627 +       if (v != au_opt_test(AuOpt_Def, name)) \
27628 +               seq_printf(m, ",%s" #str, v ? "" : "no"); \
27629 +} while (0)
27630 +
27631 +#define AuStr(name, str) do { \
27632 +       v = mnt_flags & AuOptMask_##name; \
27633 +       if (v != (AuOpt_Def & AuOptMask_##name)) \
27634 +               seq_printf(m, "," #str "=%s", au_optstr_##str(v)); \
27635 +} while (0)
27636 +
27637 +#define AuUInt(name, str, val) do { \
27638 +       if (val != AUFS_##name##_DEF) \
27639 +               seq_printf(m, "," #str "=%u", val); \
27640 +} while (0)
27641 +
27642 +       sb = dentry->d_sb;
27643 +       if (sb->s_flags & MS_POSIXACL)
27644 +               seq_puts(m, ",acl");
27645 +
27646 +       /* lock free root dinfo */
27647 +       si_noflush_read_lock(sb);
27648 +       sbinfo = au_sbi(sb);
27649 +       seq_printf(m, ",si=%lx", sysaufs_si_id(sbinfo));
27650 +
27651 +       mnt_flags = au_mntflags(sb);
27652 +       if (au_opt_test(mnt_flags, XINO)) {
27653 +               err = au_show_xino(m, sb);
27654 +               if (unlikely(err))
27655 +                       goto out;
27656 +       } else
27657 +               seq_puts(m, ",noxino");
27658 +
27659 +       AuBool(TRUNC_XINO, trunc_xino);
27660 +       AuStr(UDBA, udba);
27661 +       AuBool(SHWH, shwh);
27662 +       AuBool(PLINK, plink);
27663 +       AuBool(DIO, dio);
27664 +       AuBool(DIRPERM1, dirperm1);
27665 +
27666 +       v = sbinfo->si_wbr_create;
27667 +       if (v != AuWbrCreate_Def)
27668 +               au_show_wbr_create(m, v, sbinfo);
27669 +
27670 +       v = sbinfo->si_wbr_copyup;
27671 +       if (v != AuWbrCopyup_Def)
27672 +               seq_printf(m, ",cpup=%s", au_optstr_wbr_copyup(v));
27673 +
27674 +       v = au_opt_test(mnt_flags, ALWAYS_DIROPQ);
27675 +       if (v != au_opt_test(AuOpt_Def, ALWAYS_DIROPQ))
27676 +               seq_printf(m, ",diropq=%c", v ? 'a' : 'w');
27677 +
27678 +       AuUInt(DIRWH, dirwh, sbinfo->si_dirwh);
27679 +
27680 +       v = jiffies_to_msecs(sbinfo->si_rdcache) / MSEC_PER_SEC;
27681 +       AuUInt(RDCACHE, rdcache, v);
27682 +
27683 +       AuUInt(RDBLK, rdblk, sbinfo->si_rdblk);
27684 +       AuUInt(RDHASH, rdhash, sbinfo->si_rdhash);
27685 +
27686 +       au_fhsm_show(m, sbinfo);
27687 +
27688 +       AuBool(SUM, sum);
27689 +       /* AuBool(SUM_W, wsum); */
27690 +       AuBool(WARN_PERM, warn_perm);
27691 +       AuBool(VERBOSE, verbose);
27692 +
27693 +out:
27694 +       /* be sure to print "br:" last */
27695 +       if (!sysaufs_brs) {
27696 +               seq_puts(m, ",br:");
27697 +               au_show_brs(m, sb);
27698 +       }
27699 +       si_read_unlock(sb);
27700 +       return 0;
27701 +
27702 +#undef AuBool
27703 +#undef AuStr
27704 +#undef AuUInt
27705 +}
27706 +
27707 +/* ---------------------------------------------------------------------- */
27708 +
27709 +/* sum mode which returns the summation for statfs(2) */
27710 +
27711 +static u64 au_add_till_max(u64 a, u64 b)
27712 +{
27713 +       u64 old;
27714 +
27715 +       old = a;
27716 +       a += b;
27717 +       if (old <= a)
27718 +               return a;
27719 +       return ULLONG_MAX;
27720 +}
27721 +
27722 +static u64 au_mul_till_max(u64 a, long mul)
27723 +{
27724 +       u64 old;
27725 +
27726 +       old = a;
27727 +       a *= mul;
27728 +       if (old <= a)
27729 +               return a;
27730 +       return ULLONG_MAX;
27731 +}
27732 +
27733 +static int au_statfs_sum(struct super_block *sb, struct kstatfs *buf)
27734 +{
27735 +       int err;
27736 +       long bsize, factor;
27737 +       u64 blocks, bfree, bavail, files, ffree;
27738 +       aufs_bindex_t bbot, bindex, i;
27739 +       unsigned char shared;
27740 +       struct path h_path;
27741 +       struct super_block *h_sb;
27742 +
27743 +       err = 0;
27744 +       bsize = LONG_MAX;
27745 +       files = 0;
27746 +       ffree = 0;
27747 +       blocks = 0;
27748 +       bfree = 0;
27749 +       bavail = 0;
27750 +       bbot = au_sbbot(sb);
27751 +       for (bindex = 0; bindex <= bbot; bindex++) {
27752 +               h_path.mnt = au_sbr_mnt(sb, bindex);
27753 +               h_sb = h_path.mnt->mnt_sb;
27754 +               shared = 0;
27755 +               for (i = 0; !shared && i < bindex; i++)
27756 +                       shared = (au_sbr_sb(sb, i) == h_sb);
27757 +               if (shared)
27758 +                       continue;
27759 +
27760 +               /* sb->s_root for NFS is unreliable */
27761 +               h_path.dentry = h_path.mnt->mnt_root;
27762 +               err = vfs_statfs(&h_path, buf);
27763 +               if (unlikely(err))
27764 +                       goto out;
27765 +
27766 +               if (bsize > buf->f_bsize) {
27767 +                       /*
27768 +                        * we will reduce bsize, so we have to expand blocks
27769 +                        * etc. to match them again
27770 +                        */
27771 +                       factor = (bsize / buf->f_bsize);
27772 +                       blocks = au_mul_till_max(blocks, factor);
27773 +                       bfree = au_mul_till_max(bfree, factor);
27774 +                       bavail = au_mul_till_max(bavail, factor);
27775 +                       bsize = buf->f_bsize;
27776 +               }
27777 +
27778 +               factor = (buf->f_bsize / bsize);
27779 +               blocks = au_add_till_max(blocks,
27780 +                               au_mul_till_max(buf->f_blocks, factor));
27781 +               bfree = au_add_till_max(bfree,
27782 +                               au_mul_till_max(buf->f_bfree, factor));
27783 +               bavail = au_add_till_max(bavail,
27784 +                               au_mul_till_max(buf->f_bavail, factor));
27785 +               files = au_add_till_max(files, buf->f_files);
27786 +               ffree = au_add_till_max(ffree, buf->f_ffree);
27787 +       }
27788 +
27789 +       buf->f_bsize = bsize;
27790 +       buf->f_blocks = blocks;
27791 +       buf->f_bfree = bfree;
27792 +       buf->f_bavail = bavail;
27793 +       buf->f_files = files;
27794 +       buf->f_ffree = ffree;
27795 +       buf->f_frsize = 0;
27796 +
27797 +out:
27798 +       return err;
27799 +}
27800 +
27801 +static int aufs_statfs(struct dentry *dentry, struct kstatfs *buf)
27802 +{
27803 +       int err;
27804 +       struct path h_path;
27805 +       struct super_block *sb;
27806 +
27807 +       /* lock free root dinfo */
27808 +       sb = dentry->d_sb;
27809 +       si_noflush_read_lock(sb);
27810 +       if (!au_opt_test(au_mntflags(sb), SUM)) {
27811 +               /* sb->s_root for NFS is unreliable */
27812 +               h_path.mnt = au_sbr_mnt(sb, 0);
27813 +               h_path.dentry = h_path.mnt->mnt_root;
27814 +               err = vfs_statfs(&h_path, buf);
27815 +       } else
27816 +               err = au_statfs_sum(sb, buf);
27817 +       si_read_unlock(sb);
27818 +
27819 +       if (!err) {
27820 +               buf->f_type = AUFS_SUPER_MAGIC;
27821 +               buf->f_namelen = AUFS_MAX_NAMELEN;
27822 +               memset(&buf->f_fsid, 0, sizeof(buf->f_fsid));
27823 +       }
27824 +       /* buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1; */
27825 +
27826 +       return err;
27827 +}
27828 +
27829 +/* ---------------------------------------------------------------------- */
27830 +
27831 +static int aufs_sync_fs(struct super_block *sb, int wait)
27832 +{
27833 +       int err, e;
27834 +       aufs_bindex_t bbot, bindex;
27835 +       struct au_branch *br;
27836 +       struct super_block *h_sb;
27837 +
27838 +       err = 0;
27839 +       si_noflush_read_lock(sb);
27840 +       bbot = au_sbbot(sb);
27841 +       for (bindex = 0; bindex <= bbot; bindex++) {
27842 +               br = au_sbr(sb, bindex);
27843 +               if (!au_br_writable(br->br_perm))
27844 +                       continue;
27845 +
27846 +               h_sb = au_sbr_sb(sb, bindex);
27847 +               e = vfsub_sync_filesystem(h_sb, wait);
27848 +               if (unlikely(e && !err))
27849 +                       err = e;
27850 +               /* go on even if an error happens */
27851 +       }
27852 +       si_read_unlock(sb);
27853 +
27854 +       return err;
27855 +}
27856 +
27857 +/* ---------------------------------------------------------------------- */
27858 +
27859 +/* final actions when unmounting a file system */
27860 +static void aufs_put_super(struct super_block *sb)
27861 +{
27862 +       struct au_sbinfo *sbinfo;
27863 +
27864 +       sbinfo = au_sbi(sb);
27865 +       if (!sbinfo)
27866 +               return;
27867 +
27868 +       dbgaufs_si_fin(sbinfo);
27869 +       kobject_put(&sbinfo->si_kobj);
27870 +}
27871 +
27872 +/* ---------------------------------------------------------------------- */
27873 +
27874 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb,
27875 +                    struct super_block *sb, void *arg)
27876 +{
27877 +       void *array;
27878 +       unsigned long long n, sz;
27879 +
27880 +       array = NULL;
27881 +       n = 0;
27882 +       if (!*hint)
27883 +               goto out;
27884 +
27885 +       if (*hint > ULLONG_MAX / sizeof(array)) {
27886 +               array = ERR_PTR(-EMFILE);
27887 +               pr_err("hint %llu\n", *hint);
27888 +               goto out;
27889 +       }
27890 +
27891 +       sz = sizeof(array) * *hint;
27892 +       array = kzalloc(sz, GFP_NOFS);
27893 +       if (unlikely(!array))
27894 +               array = vzalloc(sz);
27895 +       if (unlikely(!array)) {
27896 +               array = ERR_PTR(-ENOMEM);
27897 +               goto out;
27898 +       }
27899 +
27900 +       n = cb(sb, array, *hint, arg);
27901 +       AuDebugOn(n > *hint);
27902 +
27903 +out:
27904 +       *hint = n;
27905 +       return array;
27906 +}
27907 +
27908 +static unsigned long long au_iarray_cb(struct super_block *sb, void *a,
27909 +                                      unsigned long long max __maybe_unused,
27910 +                                      void *arg)
27911 +{
27912 +       unsigned long long n;
27913 +       struct inode **p, *inode;
27914 +       struct list_head *head;
27915 +
27916 +       n = 0;
27917 +       p = a;
27918 +       head = arg;
27919 +       spin_lock(&sb->s_inode_list_lock);
27920 +       list_for_each_entry(inode, head, i_sb_list) {
27921 +               if (!au_is_bad_inode(inode)
27922 +                   && au_ii(inode)->ii_btop >= 0) {
27923 +                       spin_lock(&inode->i_lock);
27924 +                       if (atomic_read(&inode->i_count)) {
27925 +                               au_igrab(inode);
27926 +                               *p++ = inode;
27927 +                               n++;
27928 +                               AuDebugOn(n > max);
27929 +                       }
27930 +                       spin_unlock(&inode->i_lock);
27931 +               }
27932 +       }
27933 +       spin_unlock(&sb->s_inode_list_lock);
27934 +
27935 +       return n;
27936 +}
27937 +
27938 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max)
27939 +{
27940 +       *max = au_ninodes(sb);
27941 +       return au_array_alloc(max, au_iarray_cb, sb, &sb->s_inodes);
27942 +}
27943 +
27944 +void au_iarray_free(struct inode **a, unsigned long long max)
27945 +{
27946 +       unsigned long long ull;
27947 +
27948 +       for (ull = 0; ull < max; ull++)
27949 +               iput(a[ull]);
27950 +       kvfree(a);
27951 +}
27952 +
27953 +/* ---------------------------------------------------------------------- */
27954 +
27955 +/*
27956 + * refresh dentry and inode at remount time.
27957 + */
27958 +/* todo: consolidate with simple_reval_dpath() and au_reval_for_attr() */
27959 +static int au_do_refresh(struct dentry *dentry, unsigned int dir_flags,
27960 +                     struct dentry *parent)
27961 +{
27962 +       int err;
27963 +
27964 +       di_write_lock_child(dentry);
27965 +       di_read_lock_parent(parent, AuLock_IR);
27966 +       err = au_refresh_dentry(dentry, parent);
27967 +       if (!err && dir_flags)
27968 +               au_hn_reset(d_inode(dentry), dir_flags);
27969 +       di_read_unlock(parent, AuLock_IR);
27970 +       di_write_unlock(dentry);
27971 +
27972 +       return err;
27973 +}
27974 +
27975 +static int au_do_refresh_d(struct dentry *dentry, unsigned int sigen,
27976 +                          struct au_sbinfo *sbinfo,
27977 +                          const unsigned int dir_flags, unsigned int do_idop)
27978 +{
27979 +       int err;
27980 +       struct dentry *parent;
27981 +
27982 +       err = 0;
27983 +       parent = dget_parent(dentry);
27984 +       if (!au_digen_test(parent, sigen) && au_digen_test(dentry, sigen)) {
27985 +               if (d_really_is_positive(dentry)) {
27986 +                       if (!d_is_dir(dentry))
27987 +                               err = au_do_refresh(dentry, /*dir_flags*/0,
27988 +                                                parent);
27989 +                       else {
27990 +                               err = au_do_refresh(dentry, dir_flags, parent);
27991 +                               if (unlikely(err))
27992 +                                       au_fset_si(sbinfo, FAILED_REFRESH_DIR);
27993 +                       }
27994 +               } else
27995 +                       err = au_do_refresh(dentry, /*dir_flags*/0, parent);
27996 +               AuDbgDentry(dentry);
27997 +       }
27998 +       dput(parent);
27999 +
28000 +       if (!err) {
28001 +               if (do_idop)
28002 +                       au_refresh_dop(dentry, /*force_reval*/0);
28003 +       } else
28004 +               au_refresh_dop(dentry, /*force_reval*/1);
28005 +
28006 +       AuTraceErr(err);
28007 +       return err;
28008 +}
28009 +
28010 +static int au_refresh_d(struct super_block *sb, unsigned int do_idop)
28011 +{
28012 +       int err, i, j, ndentry, e;
28013 +       unsigned int sigen;
28014 +       struct au_dcsub_pages dpages;
28015 +       struct au_dpage *dpage;
28016 +       struct dentry **dentries, *d;
28017 +       struct au_sbinfo *sbinfo;
28018 +       struct dentry *root = sb->s_root;
28019 +       const unsigned int dir_flags = au_hi_flags(d_inode(root), /*isdir*/1);
28020 +
28021 +       if (do_idop)
28022 +               au_refresh_dop(root, /*force_reval*/0);
28023 +
28024 +       err = au_dpages_init(&dpages, GFP_NOFS);
28025 +       if (unlikely(err))
28026 +               goto out;
28027 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
28028 +       if (unlikely(err))
28029 +               goto out_dpages;
28030 +
28031 +       sigen = au_sigen(sb);
28032 +       sbinfo = au_sbi(sb);
28033 +       for (i = 0; i < dpages.ndpage; i++) {
28034 +               dpage = dpages.dpages + i;
28035 +               dentries = dpage->dentries;
28036 +               ndentry = dpage->ndentry;
28037 +               for (j = 0; j < ndentry; j++) {
28038 +                       d = dentries[j];
28039 +                       e = au_do_refresh_d(d, sigen, sbinfo, dir_flags,
28040 +                                           do_idop);
28041 +                       if (unlikely(e && !err))
28042 +                               err = e;
28043 +                       /* go on even err */
28044 +               }
28045 +       }
28046 +
28047 +out_dpages:
28048 +       au_dpages_free(&dpages);
28049 +out:
28050 +       return err;
28051 +}
28052 +
28053 +static int au_refresh_i(struct super_block *sb, unsigned int do_idop)
28054 +{
28055 +       int err, e;
28056 +       unsigned int sigen;
28057 +       unsigned long long max, ull;
28058 +       struct inode *inode, **array;
28059 +
28060 +       array = au_iarray_alloc(sb, &max);
28061 +       err = PTR_ERR(array);
28062 +       if (IS_ERR(array))
28063 +               goto out;
28064 +
28065 +       err = 0;
28066 +       sigen = au_sigen(sb);
28067 +       for (ull = 0; ull < max; ull++) {
28068 +               inode = array[ull];
28069 +               if (unlikely(!inode))
28070 +                       break;
28071 +
28072 +               e = 0;
28073 +               ii_write_lock_child(inode);
28074 +               if (au_iigen(inode, NULL) != sigen) {
28075 +                       e = au_refresh_hinode_self(inode);
28076 +                       if (unlikely(e)) {
28077 +                               au_refresh_iop(inode, /*force_getattr*/1);
28078 +                               pr_err("error %d, i%lu\n", e, inode->i_ino);
28079 +                               if (!err)
28080 +                                       err = e;
28081 +                               /* go on even if err */
28082 +                       }
28083 +               }
28084 +               if (!e && do_idop)
28085 +                       au_refresh_iop(inode, /*force_getattr*/0);
28086 +               ii_write_unlock(inode);
28087 +       }
28088 +
28089 +       au_iarray_free(array, max);
28090 +
28091 +out:
28092 +       return err;
28093 +}
28094 +
28095 +static void au_remount_refresh(struct super_block *sb, unsigned int do_idop)
28096 +{
28097 +       int err, e;
28098 +       unsigned int udba;
28099 +       aufs_bindex_t bindex, bbot;
28100 +       struct dentry *root;
28101 +       struct inode *inode;
28102 +       struct au_branch *br;
28103 +       struct au_sbinfo *sbi;
28104 +
28105 +       au_sigen_inc(sb);
28106 +       sbi = au_sbi(sb);
28107 +       au_fclr_si(sbi, FAILED_REFRESH_DIR);
28108 +
28109 +       root = sb->s_root;
28110 +       DiMustNoWaiters(root);
28111 +       inode = d_inode(root);
28112 +       IiMustNoWaiters(inode);
28113 +
28114 +       udba = au_opt_udba(sb);
28115 +       bbot = au_sbbot(sb);
28116 +       for (bindex = 0; bindex <= bbot; bindex++) {
28117 +               br = au_sbr(sb, bindex);
28118 +               err = au_hnotify_reset_br(udba, br, br->br_perm);
28119 +               if (unlikely(err))
28120 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
28121 +                               bindex, err);
28122 +               /* go on even if err */
28123 +       }
28124 +       au_hn_reset(inode, au_hi_flags(inode, /*isdir*/1));
28125 +
28126 +       if (do_idop) {
28127 +               if (au_ftest_si(sbi, NO_DREVAL)) {
28128 +                       AuDebugOn(sb->s_d_op == &aufs_dop_noreval);
28129 +                       sb->s_d_op = &aufs_dop_noreval;
28130 +                       AuDebugOn(sbi->si_iop_array == aufs_iop_nogetattr);
28131 +                       sbi->si_iop_array = aufs_iop_nogetattr;
28132 +               } else {
28133 +                       AuDebugOn(sb->s_d_op == &aufs_dop);
28134 +                       sb->s_d_op = &aufs_dop;
28135 +                       AuDebugOn(sbi->si_iop_array == aufs_iop);
28136 +                       sbi->si_iop_array = aufs_iop;
28137 +               }
28138 +               pr_info("reset to %pf and %pf\n",
28139 +                       sb->s_d_op, sbi->si_iop_array);
28140 +       }
28141 +
28142 +       di_write_unlock(root);
28143 +       err = au_refresh_d(sb, do_idop);
28144 +       e = au_refresh_i(sb, do_idop);
28145 +       if (unlikely(e && !err))
28146 +               err = e;
28147 +       /* aufs_write_lock() calls ..._child() */
28148 +       di_write_lock_child(root);
28149 +
28150 +       au_cpup_attr_all(inode, /*force*/1);
28151 +
28152 +       if (unlikely(err))
28153 +               AuIOErr("refresh failed, ignored, %d\n", err);
28154 +}
28155 +
28156 +/* stop extra interpretation of errno in mount(8), and strange error messages */
28157 +static int cvt_err(int err)
28158 +{
28159 +       AuTraceErr(err);
28160 +
28161 +       switch (err) {
28162 +       case -ENOENT:
28163 +       case -ENOTDIR:
28164 +       case -EEXIST:
28165 +       case -EIO:
28166 +               err = -EINVAL;
28167 +       }
28168 +       return err;
28169 +}
28170 +
28171 +static int aufs_remount_fs(struct super_block *sb, int *flags, char *data)
28172 +{
28173 +       int err, do_dx;
28174 +       unsigned int mntflags;
28175 +       struct au_opts opts = {
28176 +               .opt = NULL
28177 +       };
28178 +       struct dentry *root;
28179 +       struct inode *inode;
28180 +       struct au_sbinfo *sbinfo;
28181 +
28182 +       err = 0;
28183 +       root = sb->s_root;
28184 +       if (!data || !*data) {
28185 +               err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
28186 +               if (!err) {
28187 +                       di_write_lock_child(root);
28188 +                       err = au_opts_verify(sb, *flags, /*pending*/0);
28189 +                       aufs_write_unlock(root);
28190 +               }
28191 +               goto out;
28192 +       }
28193 +
28194 +       err = -ENOMEM;
28195 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
28196 +       if (unlikely(!opts.opt))
28197 +               goto out;
28198 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
28199 +       opts.flags = AuOpts_REMOUNT;
28200 +       opts.sb_flags = *flags;
28201 +
28202 +       /* parse it before aufs lock */
28203 +       err = au_opts_parse(sb, data, &opts);
28204 +       if (unlikely(err))
28205 +               goto out_opts;
28206 +
28207 +       sbinfo = au_sbi(sb);
28208 +       inode = d_inode(root);
28209 +       inode_lock(inode);
28210 +       err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
28211 +       if (unlikely(err))
28212 +               goto out_mtx;
28213 +       di_write_lock_child(root);
28214 +
28215 +       /* au_opts_remount() may return an error */
28216 +       err = au_opts_remount(sb, &opts);
28217 +       au_opts_free(&opts);
28218 +
28219 +       if (au_ftest_opts(opts.flags, REFRESH))
28220 +               au_remount_refresh(sb, au_ftest_opts(opts.flags, REFRESH_IDOP));
28221 +
28222 +       if (au_ftest_opts(opts.flags, REFRESH_DYAOP)) {
28223 +               mntflags = au_mntflags(sb);
28224 +               do_dx = !!au_opt_test(mntflags, DIO);
28225 +               au_dy_arefresh(do_dx);
28226 +       }
28227 +
28228 +       au_fhsm_wrote_all(sb, /*force*/1); /* ?? */
28229 +       aufs_write_unlock(root);
28230 +
28231 +out_mtx:
28232 +       inode_unlock(inode);
28233 +out_opts:
28234 +       au_delayed_free_page((unsigned long)opts.opt);
28235 +out:
28236 +       err = cvt_err(err);
28237 +       AuTraceErr(err);
28238 +       return err;
28239 +}
28240 +
28241 +static const struct super_operations aufs_sop = {
28242 +       .alloc_inode    = aufs_alloc_inode,
28243 +       .destroy_inode  = aufs_destroy_inode,
28244 +       /* always deleting, no clearing */
28245 +       .drop_inode     = generic_delete_inode,
28246 +       .show_options   = aufs_show_options,
28247 +       .statfs         = aufs_statfs,
28248 +       .put_super      = aufs_put_super,
28249 +       .sync_fs        = aufs_sync_fs,
28250 +       .remount_fs     = aufs_remount_fs
28251 +};
28252 +
28253 +/* ---------------------------------------------------------------------- */
28254 +
28255 +static int alloc_root(struct super_block *sb)
28256 +{
28257 +       int err;
28258 +       struct inode *inode;
28259 +       struct dentry *root;
28260 +
28261 +       err = -ENOMEM;
28262 +       inode = au_iget_locked(sb, AUFS_ROOT_INO);
28263 +       err = PTR_ERR(inode);
28264 +       if (IS_ERR(inode))
28265 +               goto out;
28266 +
28267 +       inode->i_op = aufs_iop + AuIop_DIR; /* with getattr by default */
28268 +       inode->i_fop = &aufs_dir_fop;
28269 +       inode->i_mode = S_IFDIR;
28270 +       set_nlink(inode, 2);
28271 +       unlock_new_inode(inode);
28272 +
28273 +       root = d_make_root(inode);
28274 +       if (unlikely(!root))
28275 +               goto out;
28276 +       err = PTR_ERR(root);
28277 +       if (IS_ERR(root))
28278 +               goto out;
28279 +
28280 +       err = au_di_init(root);
28281 +       if (!err) {
28282 +               sb->s_root = root;
28283 +               return 0; /* success */
28284 +       }
28285 +       dput(root);
28286 +
28287 +out:
28288 +       return err;
28289 +}
28290 +
28291 +static int aufs_fill_super(struct super_block *sb, void *raw_data,
28292 +                          int silent __maybe_unused)
28293 +{
28294 +       int err;
28295 +       struct au_opts opts = {
28296 +               .opt = NULL
28297 +       };
28298 +       struct au_sbinfo *sbinfo;
28299 +       struct dentry *root;
28300 +       struct inode *inode;
28301 +       char *arg = raw_data;
28302 +
28303 +       if (unlikely(!arg || !*arg)) {
28304 +               err = -EINVAL;
28305 +               pr_err("no arg\n");
28306 +               goto out;
28307 +       }
28308 +
28309 +       err = -ENOMEM;
28310 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
28311 +       if (unlikely(!opts.opt))
28312 +               goto out;
28313 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
28314 +       opts.sb_flags = sb->s_flags;
28315 +
28316 +       err = au_si_alloc(sb);
28317 +       if (unlikely(err))
28318 +               goto out_opts;
28319 +       sbinfo = au_sbi(sb);
28320 +
28321 +       /* all timestamps always follow the ones on the branch */
28322 +       sb->s_flags |= MS_NOATIME | MS_NODIRATIME;
28323 +       sb->s_op = &aufs_sop;
28324 +       sb->s_d_op = &aufs_dop;
28325 +       sb->s_magic = AUFS_SUPER_MAGIC;
28326 +       sb->s_maxbytes = 0;
28327 +       sb->s_stack_depth = 1;
28328 +       au_export_init(sb);
28329 +       au_xattr_init(sb);
28330 +
28331 +       err = alloc_root(sb);
28332 +       if (unlikely(err)) {
28333 +               si_write_unlock(sb);
28334 +               goto out_info;
28335 +       }
28336 +       root = sb->s_root;
28337 +       inode = d_inode(root);
28338 +
28339 +       /*
28340 +        * actually we can parse options regardless aufs lock here.
28341 +        * but at remount time, parsing must be done before aufs lock.
28342 +        * so we follow the same rule.
28343 +        */
28344 +       ii_write_lock_parent(inode);
28345 +       aufs_write_unlock(root);
28346 +       err = au_opts_parse(sb, arg, &opts);
28347 +       if (unlikely(err))
28348 +               goto out_root;
28349 +
28350 +       /* lock vfs_inode first, then aufs. */
28351 +       inode_lock(inode);
28352 +       aufs_write_lock(root);
28353 +       err = au_opts_mount(sb, &opts);
28354 +       au_opts_free(&opts);
28355 +       if (!err && au_ftest_si(sbinfo, NO_DREVAL)) {
28356 +               sb->s_d_op = &aufs_dop_noreval;
28357 +               pr_info("%pf\n", sb->s_d_op);
28358 +               au_refresh_dop(root, /*force_reval*/0);
28359 +               sbinfo->si_iop_array = aufs_iop_nogetattr;
28360 +               au_refresh_iop(inode, /*force_getattr*/0);
28361 +       }
28362 +       aufs_write_unlock(root);
28363 +       inode_unlock(inode);
28364 +       if (!err)
28365 +               goto out_opts; /* success */
28366 +
28367 +out_root:
28368 +       dput(root);
28369 +       sb->s_root = NULL;
28370 +out_info:
28371 +       dbgaufs_si_fin(sbinfo);
28372 +       kobject_put(&sbinfo->si_kobj);
28373 +       sb->s_fs_info = NULL;
28374 +out_opts:
28375 +       au_delayed_free_page((unsigned long)opts.opt);
28376 +out:
28377 +       AuTraceErr(err);
28378 +       err = cvt_err(err);
28379 +       AuTraceErr(err);
28380 +       return err;
28381 +}
28382 +
28383 +/* ---------------------------------------------------------------------- */
28384 +
28385 +static struct dentry *aufs_mount(struct file_system_type *fs_type, int flags,
28386 +                                const char *dev_name __maybe_unused,
28387 +                                void *raw_data)
28388 +{
28389 +       struct dentry *root;
28390 +       struct super_block *sb;
28391 +
28392 +       /* all timestamps always follow the ones on the branch */
28393 +       /* mnt->mnt_flags |= MNT_NOATIME | MNT_NODIRATIME; */
28394 +       root = mount_nodev(fs_type, flags, raw_data, aufs_fill_super);
28395 +       if (IS_ERR(root))
28396 +               goto out;
28397 +
28398 +       sb = root->d_sb;
28399 +       si_write_lock(sb, !AuLock_FLUSH);
28400 +       sysaufs_brs_add(sb, 0);
28401 +       si_write_unlock(sb);
28402 +       au_sbilist_add(sb);
28403 +
28404 +out:
28405 +       return root;
28406 +}
28407 +
28408 +static void aufs_kill_sb(struct super_block *sb)
28409 +{
28410 +       struct au_sbinfo *sbinfo;
28411 +
28412 +       sbinfo = au_sbi(sb);
28413 +       if (sbinfo) {
28414 +               au_sbilist_del(sb);
28415 +               aufs_write_lock(sb->s_root);
28416 +               au_fhsm_fin(sb);
28417 +               if (sbinfo->si_wbr_create_ops->fin)
28418 +                       sbinfo->si_wbr_create_ops->fin(sb);
28419 +               if (au_opt_test(sbinfo->si_mntflags, UDBA_HNOTIFY)) {
28420 +                       au_opt_set_udba(sbinfo->si_mntflags, UDBA_NONE);
28421 +                       au_remount_refresh(sb, /*do_idop*/0);
28422 +               }
28423 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
28424 +                       au_plink_put(sb, /*verbose*/1);
28425 +               au_xino_clr(sb);
28426 +               sbinfo->si_sb = NULL;
28427 +               aufs_write_unlock(sb->s_root);
28428 +               au_nwt_flush(&sbinfo->si_nowait);
28429 +       }
28430 +       kill_anon_super(sb);
28431 +}
28432 +
28433 +struct file_system_type aufs_fs_type = {
28434 +       .name           = AUFS_FSTYPE,
28435 +       /* a race between rename and others */
28436 +       .fs_flags       = FS_RENAME_DOES_D_MOVE,
28437 +       .mount          = aufs_mount,
28438 +       .kill_sb        = aufs_kill_sb,
28439 +       /* no need to __module_get() and module_put(). */
28440 +       .owner          = THIS_MODULE,
28441 +};
28442 diff -urN /usr/share/empty/fs/aufs/super.h linux/fs/aufs/super.h
28443 --- /usr/share/empty/fs/aufs/super.h    1970-01-01 01:00:00.000000000 +0100
28444 +++ linux/fs/aufs/super.h       2017-05-06 22:16:52.781553511 +0200
28445 @@ -0,0 +1,617 @@
28446 +/*
28447 + * Copyright (C) 2005-2017 Junjiro R. Okajima
28448 + *
28449 + * This program, aufs is free software; you can redistribute it and/or modify
28450 + * it under the terms of the GNU General Public License as published by
28451 + * the Free Software Foundation; either version 2 of the License, or
28452 + * (at your option) any later version.
28453 + *
28454 + * This program is distributed in the hope that it will be useful,
28455 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28456 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28457 + * GNU General Public License for more details.
28458 + *
28459 + * You should have received a copy of the GNU General Public License
28460 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28461 + */
28462 +
28463 +/*
28464 + * super_block operations
28465 + */
28466 +
28467 +#ifndef __AUFS_SUPER_H__
28468 +#define __AUFS_SUPER_H__
28469 +
28470 +#ifdef __KERNEL__
28471 +
28472 +#include <linux/fs.h>
28473 +#include <linux/kobject.h>
28474 +#include "rwsem.h"
28475 +#include "spl.h"
28476 +#include "wkq.h"
28477 +
28478 +/* policies to select one among multiple writable branches */
28479 +struct au_wbr_copyup_operations {
28480 +       int (*copyup)(struct dentry *dentry);
28481 +};
28482 +
28483 +#define AuWbr_DIR      1               /* target is a dir */
28484 +#define AuWbr_PARENT   (1 << 1)        /* always require a parent */
28485 +
28486 +#define au_ftest_wbr(flags, name)      ((flags) & AuWbr_##name)
28487 +#define au_fset_wbr(flags, name)       { (flags) |= AuWbr_##name; }
28488 +#define au_fclr_wbr(flags, name)       { (flags) &= ~AuWbr_##name; }
28489 +
28490 +struct au_wbr_create_operations {
28491 +       int (*create)(struct dentry *dentry, unsigned int flags);
28492 +       int (*init)(struct super_block *sb);
28493 +       int (*fin)(struct super_block *sb);
28494 +};
28495 +
28496 +struct au_wbr_mfs {
28497 +       struct mutex    mfs_lock; /* protect this structure */
28498 +       unsigned long   mfs_jiffy;
28499 +       unsigned long   mfs_expire;
28500 +       aufs_bindex_t   mfs_bindex;
28501 +
28502 +       unsigned long long      mfsrr_bytes;
28503 +       unsigned long long      mfsrr_watermark;
28504 +};
28505 +
28506 +#define AuPlink_NHASH 100
28507 +static inline int au_plink_hash(ino_t ino)
28508 +{
28509 +       return ino % AuPlink_NHASH;
28510 +}
28511 +
28512 +/* File-based Hierarchical Storage Management */
28513 +struct au_fhsm {
28514 +#ifdef CONFIG_AUFS_FHSM
28515 +       /* allow only one process who can receive the notification */
28516 +       spinlock_t              fhsm_spin;
28517 +       pid_t                   fhsm_pid;
28518 +       wait_queue_head_t       fhsm_wqh;
28519 +       atomic_t                fhsm_readable;
28520 +
28521 +       /* these are protected by si_rwsem */
28522 +       unsigned long           fhsm_expire;
28523 +       aufs_bindex_t           fhsm_bottom;
28524 +#endif
28525 +};
28526 +
28527 +struct au_branch;
28528 +struct au_sbinfo {
28529 +       /* nowait tasks in the system-wide workqueue */
28530 +       struct au_nowait_tasks  si_nowait;
28531 +
28532 +       /*
28533 +        * tried sb->s_umount, but failed due to the dependecy between i_mutex.
28534 +        * rwsem for au_sbinfo is necessary.
28535 +        */
28536 +       struct au_rwsem         si_rwsem;
28537 +
28538 +       /*
28539 +        * dirty approach to protect sb->sb_inodes and ->s_files (gone) from
28540 +        * remount.
28541 +        */
28542 +       struct percpu_counter   si_ninodes, si_nfiles;
28543 +
28544 +       /* branch management */
28545 +       unsigned int            si_generation;
28546 +
28547 +       /* see AuSi_ flags */
28548 +       unsigned char           au_si_status;
28549 +
28550 +       aufs_bindex_t           si_bbot;
28551 +
28552 +       /* dirty trick to keep br_id plus */
28553 +       unsigned int            si_last_br_id :
28554 +                               sizeof(aufs_bindex_t) * BITS_PER_BYTE - 1;
28555 +       struct au_branch        **si_branch;
28556 +
28557 +       /* policy to select a writable branch */
28558 +       unsigned char           si_wbr_copyup;
28559 +       unsigned char           si_wbr_create;
28560 +       struct au_wbr_copyup_operations *si_wbr_copyup_ops;
28561 +       struct au_wbr_create_operations *si_wbr_create_ops;
28562 +
28563 +       /* round robin */
28564 +       atomic_t                si_wbr_rr_next;
28565 +
28566 +       /* most free space */
28567 +       struct au_wbr_mfs       si_wbr_mfs;
28568 +
28569 +       /* File-based Hierarchical Storage Management */
28570 +       struct au_fhsm          si_fhsm;
28571 +
28572 +       /* mount flags */
28573 +       /* include/asm-ia64/siginfo.h defines a macro named si_flags */
28574 +       unsigned int            si_mntflags;
28575 +
28576 +       /* external inode number (bitmap and translation table) */
28577 +       vfs_readf_t             si_xread;
28578 +       vfs_writef_t            si_xwrite;
28579 +       struct file             *si_xib;
28580 +       struct mutex            si_xib_mtx; /* protect xib members */
28581 +       unsigned long           *si_xib_buf;
28582 +       unsigned long           si_xib_last_pindex;
28583 +       int                     si_xib_next_bit;
28584 +       aufs_bindex_t           si_xino_brid;
28585 +       unsigned long           si_xino_jiffy;
28586 +       unsigned long           si_xino_expire;
28587 +       /* reserved for future use */
28588 +       /* unsigned long long   si_xib_limit; */        /* Max xib file size */
28589 +
28590 +#ifdef CONFIG_AUFS_EXPORT
28591 +       /* i_generation */
28592 +       struct file             *si_xigen;
28593 +       atomic_t                si_xigen_next;
28594 +#endif
28595 +
28596 +       /* dirty trick to suppoer atomic_open */
28597 +       struct au_sphlhead      si_aopen;
28598 +
28599 +       /* vdir parameters */
28600 +       unsigned long           si_rdcache;     /* max cache time in jiffies */
28601 +       unsigned int            si_rdblk;       /* deblk size */
28602 +       unsigned int            si_rdhash;      /* hash size */
28603 +
28604 +       /*
28605 +        * If the number of whiteouts are larger than si_dirwh, leave all of
28606 +        * them after au_whtmp_ren to reduce the cost of rmdir(2).
28607 +        * future fsck.aufs or kernel thread will remove them later.
28608 +        * Otherwise, remove all whiteouts and the dir in rmdir(2).
28609 +        */
28610 +       unsigned int            si_dirwh;
28611 +
28612 +       /* pseudo_link list */
28613 +       struct au_sphlhead      si_plink[AuPlink_NHASH];
28614 +       wait_queue_head_t       si_plink_wq;
28615 +       spinlock_t              si_plink_maint_lock;
28616 +       pid_t                   si_plink_maint_pid;
28617 +
28618 +       /* file list */
28619 +       struct au_sphlhead      si_files;
28620 +
28621 +       /* with/without getattr, brother of sb->s_d_op */
28622 +       struct inode_operations *si_iop_array;
28623 +
28624 +       /*
28625 +        * sysfs and lifetime management.
28626 +        * this is not a small structure and it may be a waste of memory in case
28627 +        * of sysfs is disabled, particulary when many aufs-es are mounted.
28628 +        * but using sysfs is majority.
28629 +        */
28630 +       struct kobject          si_kobj;
28631 +#ifdef CONFIG_DEBUG_FS
28632 +       struct dentry            *si_dbgaufs;
28633 +       struct dentry            *si_dbgaufs_plink;
28634 +       struct dentry            *si_dbgaufs_xib;
28635 +#ifdef CONFIG_AUFS_EXPORT
28636 +       struct dentry            *si_dbgaufs_xigen;
28637 +#endif
28638 +#endif
28639 +
28640 +#ifdef CONFIG_AUFS_SBILIST
28641 +       struct hlist_node       si_list;
28642 +#endif
28643 +
28644 +       /* dirty, necessary for unmounting, sysfs and sysrq */
28645 +       struct super_block      *si_sb;
28646 +};
28647 +
28648 +/* sbinfo status flags */
28649 +/*
28650 + * set true when refresh_dirs() failed at remount time.
28651 + * then try refreshing dirs at access time again.
28652 + * if it is false, refreshing dirs at access time is unnecesary
28653 + */
28654 +#define AuSi_FAILED_REFRESH_DIR        1
28655 +#define AuSi_FHSM              (1 << 1)        /* fhsm is active now */
28656 +#define AuSi_NO_DREVAL         (1 << 2)        /* disable all d_revalidate */
28657 +
28658 +#ifndef CONFIG_AUFS_FHSM
28659 +#undef AuSi_FHSM
28660 +#define AuSi_FHSM              0
28661 +#endif
28662 +
28663 +static inline unsigned char au_do_ftest_si(struct au_sbinfo *sbi,
28664 +                                          unsigned int flag)
28665 +{
28666 +       AuRwMustAnyLock(&sbi->si_rwsem);
28667 +       return sbi->au_si_status & flag;
28668 +}
28669 +#define au_ftest_si(sbinfo, name)      au_do_ftest_si(sbinfo, AuSi_##name)
28670 +#define au_fset_si(sbinfo, name) do { \
28671 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
28672 +       (sbinfo)->au_si_status |= AuSi_##name; \
28673 +} while (0)
28674 +#define au_fclr_si(sbinfo, name) do { \
28675 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
28676 +       (sbinfo)->au_si_status &= ~AuSi_##name; \
28677 +} while (0)
28678 +
28679 +/* ---------------------------------------------------------------------- */
28680 +
28681 +/* policy to select one among writable branches */
28682 +#define AuWbrCopyup(sbinfo, ...) \
28683 +       ((sbinfo)->si_wbr_copyup_ops->copyup(__VA_ARGS__))
28684 +#define AuWbrCreate(sbinfo, ...) \
28685 +       ((sbinfo)->si_wbr_create_ops->create(__VA_ARGS__))
28686 +
28687 +/* flags for si_read_lock()/aufs_read_lock()/di_read_lock() */
28688 +#define AuLock_DW              1               /* write-lock dentry */
28689 +#define AuLock_IR              (1 << 1)        /* read-lock inode */
28690 +#define AuLock_IW              (1 << 2)        /* write-lock inode */
28691 +#define AuLock_FLUSH           (1 << 3)        /* wait for 'nowait' tasks */
28692 +#define AuLock_DIRS            (1 << 4)        /* target is a pair of dirs */
28693 +                                               /* except RENAME_EXCHANGE */
28694 +#define AuLock_NOPLM           (1 << 5)        /* return err in plm mode */
28695 +#define AuLock_NOPLMW          (1 << 6)        /* wait for plm mode ends */
28696 +#define AuLock_GEN             (1 << 7)        /* test digen/iigen */
28697 +#define au_ftest_lock(flags, name)     ((flags) & AuLock_##name)
28698 +#define au_fset_lock(flags, name) \
28699 +       do { (flags) |= AuLock_##name; } while (0)
28700 +#define au_fclr_lock(flags, name) \
28701 +       do { (flags) &= ~AuLock_##name; } while (0)
28702 +
28703 +/* ---------------------------------------------------------------------- */
28704 +
28705 +/* super.c */
28706 +extern struct file_system_type aufs_fs_type;
28707 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino);
28708 +typedef unsigned long long (*au_arraycb_t)(struct super_block *sb, void *array,
28709 +                                          unsigned long long max, void *arg);
28710 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb,
28711 +                    struct super_block *sb, void *arg);
28712 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max);
28713 +void au_iarray_free(struct inode **a, unsigned long long max);
28714 +
28715 +/* sbinfo.c */
28716 +void au_si_free(struct kobject *kobj);
28717 +int au_si_alloc(struct super_block *sb);
28718 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr, int may_shrink);
28719 +
28720 +unsigned int au_sigen_inc(struct super_block *sb);
28721 +aufs_bindex_t au_new_br_id(struct super_block *sb);
28722 +
28723 +int si_read_lock(struct super_block *sb, int flags);
28724 +int si_write_lock(struct super_block *sb, int flags);
28725 +int aufs_read_lock(struct dentry *dentry, int flags);
28726 +void aufs_read_unlock(struct dentry *dentry, int flags);
28727 +void aufs_write_lock(struct dentry *dentry);
28728 +void aufs_write_unlock(struct dentry *dentry);
28729 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags);
28730 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2);
28731 +
28732 +/* wbr_policy.c */
28733 +extern struct au_wbr_copyup_operations au_wbr_copyup_ops[];
28734 +extern struct au_wbr_create_operations au_wbr_create_ops[];
28735 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst);
28736 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex);
28737 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t btop);
28738 +
28739 +/* mvdown.c */
28740 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *arg);
28741 +
28742 +#ifdef CONFIG_AUFS_FHSM
28743 +/* fhsm.c */
28744 +
28745 +static inline pid_t au_fhsm_pid(struct au_fhsm *fhsm)
28746 +{
28747 +       pid_t pid;
28748 +
28749 +       spin_lock(&fhsm->fhsm_spin);
28750 +       pid = fhsm->fhsm_pid;
28751 +       spin_unlock(&fhsm->fhsm_spin);
28752 +
28753 +       return pid;
28754 +}
28755 +
28756 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force);
28757 +void au_fhsm_wrote_all(struct super_block *sb, int force);
28758 +int au_fhsm_fd(struct super_block *sb, int oflags);
28759 +int au_fhsm_br_alloc(struct au_branch *br);
28760 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex);
28761 +void au_fhsm_fin(struct super_block *sb);
28762 +void au_fhsm_init(struct au_sbinfo *sbinfo);
28763 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec);
28764 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo);
28765 +#else
28766 +AuStubVoid(au_fhsm_wrote, struct super_block *sb, aufs_bindex_t bindex,
28767 +          int force)
28768 +AuStubVoid(au_fhsm_wrote_all, struct super_block *sb, int force)
28769 +AuStub(int, au_fhsm_fd, return -EOPNOTSUPP, struct super_block *sb, int oflags)
28770 +AuStub(pid_t, au_fhsm_pid, return 0, struct au_fhsm *fhsm)
28771 +AuStubInt0(au_fhsm_br_alloc, struct au_branch *br)
28772 +AuStubVoid(au_fhsm_set_bottom, struct super_block *sb, aufs_bindex_t bindex)
28773 +AuStubVoid(au_fhsm_fin, struct super_block *sb)
28774 +AuStubVoid(au_fhsm_init, struct au_sbinfo *sbinfo)
28775 +AuStubVoid(au_fhsm_set, struct au_sbinfo *sbinfo, unsigned int sec)
28776 +AuStubVoid(au_fhsm_show, struct seq_file *seq, struct au_sbinfo *sbinfo)
28777 +#endif
28778 +
28779 +/* ---------------------------------------------------------------------- */
28780 +
28781 +static inline struct au_sbinfo *au_sbi(struct super_block *sb)
28782 +{
28783 +       return sb->s_fs_info;
28784 +}
28785 +
28786 +/* ---------------------------------------------------------------------- */
28787 +
28788 +#ifdef CONFIG_AUFS_EXPORT
28789 +int au_test_nfsd(void);
28790 +void au_export_init(struct super_block *sb);
28791 +void au_xigen_inc(struct inode *inode);
28792 +int au_xigen_new(struct inode *inode);
28793 +int au_xigen_set(struct super_block *sb, struct file *base);
28794 +void au_xigen_clr(struct super_block *sb);
28795 +
28796 +static inline int au_busy_or_stale(void)
28797 +{
28798 +       if (!au_test_nfsd())
28799 +               return -EBUSY;
28800 +       return -ESTALE;
28801 +}
28802 +#else
28803 +AuStubInt0(au_test_nfsd, void)
28804 +AuStubVoid(au_export_init, struct super_block *sb)
28805 +AuStubVoid(au_xigen_inc, struct inode *inode)
28806 +AuStubInt0(au_xigen_new, struct inode *inode)
28807 +AuStubInt0(au_xigen_set, struct super_block *sb, struct file *base)
28808 +AuStubVoid(au_xigen_clr, struct super_block *sb)
28809 +AuStub(int, au_busy_or_stale, return -EBUSY, void)
28810 +#endif /* CONFIG_AUFS_EXPORT */
28811 +
28812 +/* ---------------------------------------------------------------------- */
28813 +
28814 +#ifdef CONFIG_AUFS_SBILIST
28815 +/* module.c */
28816 +extern struct au_sphlhead au_sbilist;
28817 +
28818 +static inline void au_sbilist_init(void)
28819 +{
28820 +       au_sphl_init(&au_sbilist);
28821 +}
28822 +
28823 +static inline void au_sbilist_add(struct super_block *sb)
28824 +{
28825 +       au_sphl_add(&au_sbi(sb)->si_list, &au_sbilist);
28826 +}
28827 +
28828 +static inline void au_sbilist_del(struct super_block *sb)
28829 +{
28830 +       au_sphl_del(&au_sbi(sb)->si_list, &au_sbilist);
28831 +}
28832 +
28833 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
28834 +static inline void au_sbilist_lock(void)
28835 +{
28836 +       spin_lock(&au_sbilist.spin);
28837 +}
28838 +
28839 +static inline void au_sbilist_unlock(void)
28840 +{
28841 +       spin_unlock(&au_sbilist.spin);
28842 +}
28843 +#define AuGFP_SBILIST  GFP_ATOMIC
28844 +#else
28845 +AuStubVoid(au_sbilist_lock, void)
28846 +AuStubVoid(au_sbilist_unlock, void)
28847 +#define AuGFP_SBILIST  GFP_NOFS
28848 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
28849 +#else
28850 +AuStubVoid(au_sbilist_init, void)
28851 +AuStubVoid(au_sbilist_add, struct super_block *sb)
28852 +AuStubVoid(au_sbilist_del, struct super_block *sb)
28853 +AuStubVoid(au_sbilist_lock, void)
28854 +AuStubVoid(au_sbilist_unlock, void)
28855 +#define AuGFP_SBILIST  GFP_NOFS
28856 +#endif
28857 +
28858 +/* ---------------------------------------------------------------------- */
28859 +
28860 +static inline void dbgaufs_si_null(struct au_sbinfo *sbinfo)
28861 +{
28862 +       /*
28863 +        * This function is a dynamic '__init' function actually,
28864 +        * so the tiny check for si_rwsem is unnecessary.
28865 +        */
28866 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
28867 +#ifdef CONFIG_DEBUG_FS
28868 +       sbinfo->si_dbgaufs = NULL;
28869 +       sbinfo->si_dbgaufs_plink = NULL;
28870 +       sbinfo->si_dbgaufs_xib = NULL;
28871 +#ifdef CONFIG_AUFS_EXPORT
28872 +       sbinfo->si_dbgaufs_xigen = NULL;
28873 +#endif
28874 +#endif
28875 +}
28876 +
28877 +/* ---------------------------------------------------------------------- */
28878 +
28879 +/* current->atomic_flags */
28880 +/* this value should never corrupt the ones defined in linux/sched.h */
28881 +#define PFA_AUFS       7
28882 +
28883 +TASK_PFA_TEST(AUFS, test_aufs) /* task_test_aufs */
28884 +TASK_PFA_SET(AUFS, aufs)       /* task_set_aufs */
28885 +TASK_PFA_CLEAR(AUFS, aufs)     /* task_clear_aufs */
28886 +
28887 +static inline int si_pid_test(struct super_block *sb)
28888 +{
28889 +       return !!task_test_aufs(current);
28890 +}
28891 +
28892 +static inline void si_pid_clr(struct super_block *sb)
28893 +{
28894 +       AuDebugOn(!task_test_aufs(current));
28895 +       task_clear_aufs(current);
28896 +}
28897 +
28898 +static inline void si_pid_set(struct super_block *sb)
28899 +{
28900 +       AuDebugOn(task_test_aufs(current));
28901 +       task_set_aufs(current);
28902 +}
28903 +
28904 +/* ---------------------------------------------------------------------- */
28905 +
28906 +/* lock superblock. mainly for entry point functions */
28907 +/*
28908 + * __si_read_lock, __si_write_lock,
28909 + * __si_read_unlock, __si_write_unlock, __si_downgrade_lock
28910 + */
28911 +AuSimpleRwsemFuncs(__si, struct super_block *sb, &au_sbi(sb)->si_rwsem);
28912 +
28913 +#define SiMustNoWaiters(sb)    AuRwMustNoWaiters(&au_sbi(sb)->si_rwsem)
28914 +#define SiMustAnyLock(sb)      AuRwMustAnyLock(&au_sbi(sb)->si_rwsem)
28915 +#define SiMustWriteLock(sb)    AuRwMustWriteLock(&au_sbi(sb)->si_rwsem)
28916 +
28917 +static inline void si_noflush_read_lock(struct super_block *sb)
28918 +{
28919 +       __si_read_lock(sb);
28920 +       si_pid_set(sb);
28921 +}
28922 +
28923 +static inline int si_noflush_read_trylock(struct super_block *sb)
28924 +{
28925 +       int locked;
28926 +
28927 +       locked = __si_read_trylock(sb);
28928 +       if (locked)
28929 +               si_pid_set(sb);
28930 +       return locked;
28931 +}
28932 +
28933 +static inline void si_noflush_write_lock(struct super_block *sb)
28934 +{
28935 +       __si_write_lock(sb);
28936 +       si_pid_set(sb);
28937 +}
28938 +
28939 +static inline int si_noflush_write_trylock(struct super_block *sb)
28940 +{
28941 +       int locked;
28942 +
28943 +       locked = __si_write_trylock(sb);
28944 +       if (locked)
28945 +               si_pid_set(sb);
28946 +       return locked;
28947 +}
28948 +
28949 +#if 0 /* reserved */
28950 +static inline int si_read_trylock(struct super_block *sb, int flags)
28951 +{
28952 +       if (au_ftest_lock(flags, FLUSH))
28953 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
28954 +       return si_noflush_read_trylock(sb);
28955 +}
28956 +#endif
28957 +
28958 +static inline void si_read_unlock(struct super_block *sb)
28959 +{
28960 +       si_pid_clr(sb);
28961 +       __si_read_unlock(sb);
28962 +}
28963 +
28964 +#if 0 /* reserved */
28965 +static inline int si_write_trylock(struct super_block *sb, int flags)
28966 +{
28967 +       if (au_ftest_lock(flags, FLUSH))
28968 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
28969 +       return si_noflush_write_trylock(sb);
28970 +}
28971 +#endif
28972 +
28973 +static inline void si_write_unlock(struct super_block *sb)
28974 +{
28975 +       si_pid_clr(sb);
28976 +       __si_write_unlock(sb);
28977 +}
28978 +
28979 +#if 0 /* reserved */
28980 +static inline void si_downgrade_lock(struct super_block *sb)
28981 +{
28982 +       __si_downgrade_lock(sb);
28983 +}
28984 +#endif
28985 +
28986 +/* ---------------------------------------------------------------------- */
28987 +
28988 +static inline aufs_bindex_t au_sbbot(struct super_block *sb)
28989 +{
28990 +       SiMustAnyLock(sb);
28991 +       return au_sbi(sb)->si_bbot;
28992 +}
28993 +
28994 +static inline unsigned int au_mntflags(struct super_block *sb)
28995 +{
28996 +       SiMustAnyLock(sb);
28997 +       return au_sbi(sb)->si_mntflags;
28998 +}
28999 +
29000 +static inline unsigned int au_sigen(struct super_block *sb)
29001 +{
29002 +       SiMustAnyLock(sb);
29003 +       return au_sbi(sb)->si_generation;
29004 +}
29005 +
29006 +static inline unsigned long long au_ninodes(struct super_block *sb)
29007 +{
29008 +       s64 n = percpu_counter_sum(&au_sbi(sb)->si_ninodes);
29009 +
29010 +       BUG_ON(n < 0);
29011 +       return n;
29012 +}
29013 +
29014 +static inline void au_ninodes_inc(struct super_block *sb)
29015 +{
29016 +       percpu_counter_inc(&au_sbi(sb)->si_ninodes);
29017 +}
29018 +
29019 +static inline void au_ninodes_dec(struct super_block *sb)
29020 +{
29021 +       percpu_counter_dec(&au_sbi(sb)->si_ninodes);
29022 +}
29023 +
29024 +static inline unsigned long long au_nfiles(struct super_block *sb)
29025 +{
29026 +       s64 n = percpu_counter_sum(&au_sbi(sb)->si_nfiles);
29027 +
29028 +       BUG_ON(n < 0);
29029 +       return n;
29030 +}
29031 +
29032 +static inline void au_nfiles_inc(struct super_block *sb)
29033 +{
29034 +       percpu_counter_inc(&au_sbi(sb)->si_nfiles);
29035 +}
29036 +
29037 +static inline void au_nfiles_dec(struct super_block *sb)
29038 +{
29039 +       percpu_counter_dec(&au_sbi(sb)->si_nfiles);
29040 +}
29041 +
29042 +static inline struct au_branch *au_sbr(struct super_block *sb,
29043 +                                      aufs_bindex_t bindex)
29044 +{
29045 +       SiMustAnyLock(sb);
29046 +       return au_sbi(sb)->si_branch[0 + bindex];
29047 +}
29048 +
29049 +static inline void au_xino_brid_set(struct super_block *sb, aufs_bindex_t brid)
29050 +{
29051 +       SiMustWriteLock(sb);
29052 +       au_sbi(sb)->si_xino_brid = brid;
29053 +}
29054 +
29055 +static inline aufs_bindex_t au_xino_brid(struct super_block *sb)
29056 +{
29057 +       SiMustAnyLock(sb);
29058 +       return au_sbi(sb)->si_xino_brid;
29059 +}
29060 +
29061 +#endif /* __KERNEL__ */
29062 +#endif /* __AUFS_SUPER_H__ */
29063 diff -urN /usr/share/empty/fs/aufs/sysaufs.c linux/fs/aufs/sysaufs.c
29064 --- /usr/share/empty/fs/aufs/sysaufs.c  1970-01-01 01:00:00.000000000 +0100
29065 +++ linux/fs/aufs/sysaufs.c     2017-05-06 22:16:52.781553511 +0200
29066 @@ -0,0 +1,104 @@
29067 +/*
29068 + * Copyright (C) 2005-2017 Junjiro R. Okajima
29069 + *
29070 + * This program, aufs is free software; you can redistribute it and/or modify
29071 + * it under the terms of the GNU General Public License as published by
29072 + * the Free Software Foundation; either version 2 of the License, or
29073 + * (at your option) any later version.
29074 + *
29075 + * This program is distributed in the hope that it will be useful,
29076 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29077 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29078 + * GNU General Public License for more details.
29079 + *
29080 + * You should have received a copy of the GNU General Public License
29081 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29082 + */
29083 +
29084 +/*
29085 + * sysfs interface and lifetime management
29086 + * they are necessary regardless sysfs is disabled.
29087 + */
29088 +
29089 +#include <linux/random.h>
29090 +#include "aufs.h"
29091 +
29092 +unsigned long sysaufs_si_mask;
29093 +struct kset *sysaufs_kset;
29094 +
29095 +#define AuSiAttr(_name) { \
29096 +       .attr   = { .name = __stringify(_name), .mode = 0444 }, \
29097 +       .show   = sysaufs_si_##_name,                           \
29098 +}
29099 +
29100 +static struct sysaufs_si_attr sysaufs_si_attr_xi_path = AuSiAttr(xi_path);
29101 +struct attribute *sysaufs_si_attrs[] = {
29102 +       &sysaufs_si_attr_xi_path.attr,
29103 +       NULL,
29104 +};
29105 +
29106 +static const struct sysfs_ops au_sbi_ops = {
29107 +       .show   = sysaufs_si_show
29108 +};
29109 +
29110 +static struct kobj_type au_sbi_ktype = {
29111 +       .release        = au_si_free,
29112 +       .sysfs_ops      = &au_sbi_ops,
29113 +       .default_attrs  = sysaufs_si_attrs
29114 +};
29115 +
29116 +/* ---------------------------------------------------------------------- */
29117 +
29118 +int sysaufs_si_init(struct au_sbinfo *sbinfo)
29119 +{
29120 +       int err;
29121 +
29122 +       sbinfo->si_kobj.kset = sysaufs_kset;
29123 +       /* cf. sysaufs_name() */
29124 +       err = kobject_init_and_add
29125 +               (&sbinfo->si_kobj, &au_sbi_ktype, /*&sysaufs_kset->kobj*/NULL,
29126 +                SysaufsSiNamePrefix "%lx", sysaufs_si_id(sbinfo));
29127 +
29128 +       dbgaufs_si_null(sbinfo);
29129 +       if (!err) {
29130 +               err = dbgaufs_si_init(sbinfo);
29131 +               if (unlikely(err))
29132 +                       kobject_put(&sbinfo->si_kobj);
29133 +       }
29134 +       return err;
29135 +}
29136 +
29137 +void sysaufs_fin(void)
29138 +{
29139 +       dbgaufs_fin();
29140 +       sysfs_remove_group(&sysaufs_kset->kobj, sysaufs_attr_group);
29141 +       kset_unregister(sysaufs_kset);
29142 +}
29143 +
29144 +int __init sysaufs_init(void)
29145 +{
29146 +       int err;
29147 +
29148 +       do {
29149 +               get_random_bytes(&sysaufs_si_mask, sizeof(sysaufs_si_mask));
29150 +       } while (!sysaufs_si_mask);
29151 +
29152 +       err = -EINVAL;
29153 +       sysaufs_kset = kset_create_and_add(AUFS_NAME, NULL, fs_kobj);
29154 +       if (unlikely(!sysaufs_kset))
29155 +               goto out;
29156 +       err = PTR_ERR(sysaufs_kset);
29157 +       if (IS_ERR(sysaufs_kset))
29158 +               goto out;
29159 +       err = sysfs_create_group(&sysaufs_kset->kobj, sysaufs_attr_group);
29160 +       if (unlikely(err)) {
29161 +               kset_unregister(sysaufs_kset);
29162 +               goto out;
29163 +       }
29164 +
29165 +       err = dbgaufs_init();
29166 +       if (unlikely(err))
29167 +               sysaufs_fin();
29168 +out:
29169 +       return err;
29170 +}
29171 diff -urN /usr/share/empty/fs/aufs/sysaufs.h linux/fs/aufs/sysaufs.h
29172 --- /usr/share/empty/fs/aufs/sysaufs.h  1970-01-01 01:00:00.000000000 +0100
29173 +++ linux/fs/aufs/sysaufs.h     2017-05-06 22:16:52.781553511 +0200
29174 @@ -0,0 +1,101 @@
29175 +/*
29176 + * Copyright (C) 2005-2017 Junjiro R. Okajima
29177 + *
29178 + * This program, aufs is free software; you can redistribute it and/or modify
29179 + * it under the terms of the GNU General Public License as published by
29180 + * the Free Software Foundation; either version 2 of the License, or
29181 + * (at your option) any later version.
29182 + *
29183 + * This program is distributed in the hope that it will be useful,
29184 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29185 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29186 + * GNU General Public License for more details.
29187 + *
29188 + * You should have received a copy of the GNU General Public License
29189 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29190 + */
29191 +
29192 +/*
29193 + * sysfs interface and mount lifetime management
29194 + */
29195 +
29196 +#ifndef __SYSAUFS_H__
29197 +#define __SYSAUFS_H__
29198 +
29199 +#ifdef __KERNEL__
29200 +
29201 +#include <linux/sysfs.h>
29202 +#include "module.h"
29203 +
29204 +struct super_block;
29205 +struct au_sbinfo;
29206 +
29207 +struct sysaufs_si_attr {
29208 +       struct attribute attr;
29209 +       int (*show)(struct seq_file *seq, struct super_block *sb);
29210 +};
29211 +
29212 +/* ---------------------------------------------------------------------- */
29213 +
29214 +/* sysaufs.c */
29215 +extern unsigned long sysaufs_si_mask;
29216 +extern struct kset *sysaufs_kset;
29217 +extern struct attribute *sysaufs_si_attrs[];
29218 +int sysaufs_si_init(struct au_sbinfo *sbinfo);
29219 +int __init sysaufs_init(void);
29220 +void sysaufs_fin(void);
29221 +
29222 +/* ---------------------------------------------------------------------- */
29223 +
29224 +/* some people doesn't like to show a pointer in kernel */
29225 +static inline unsigned long sysaufs_si_id(struct au_sbinfo *sbinfo)
29226 +{
29227 +       return sysaufs_si_mask ^ (unsigned long)sbinfo;
29228 +}
29229 +
29230 +#define SysaufsSiNamePrefix    "si_"
29231 +#define SysaufsSiNameLen       (sizeof(SysaufsSiNamePrefix) + 16)
29232 +static inline void sysaufs_name(struct au_sbinfo *sbinfo, char *name)
29233 +{
29234 +       snprintf(name, SysaufsSiNameLen, SysaufsSiNamePrefix "%lx",
29235 +                sysaufs_si_id(sbinfo));
29236 +}
29237 +
29238 +struct au_branch;
29239 +#ifdef CONFIG_SYSFS
29240 +/* sysfs.c */
29241 +extern struct attribute_group *sysaufs_attr_group;
29242 +
29243 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb);
29244 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
29245 +                        char *buf);
29246 +long au_brinfo_ioctl(struct file *file, unsigned long arg);
29247 +#ifdef CONFIG_COMPAT
29248 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg);
29249 +#endif
29250 +
29251 +void sysaufs_br_init(struct au_branch *br);
29252 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
29253 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
29254 +
29255 +#define sysaufs_brs_init()     do {} while (0)
29256 +
29257 +#else
29258 +#define sysaufs_attr_group     NULL
29259 +
29260 +AuStubInt0(sysaufs_si_xi_path, struct seq_file *seq, struct super_block *sb)
29261 +AuStub(ssize_t, sysaufs_si_show, return 0, struct kobject *kobj,
29262 +       struct attribute *attr, char *buf)
29263 +AuStubVoid(sysaufs_br_init, struct au_branch *br)
29264 +AuStubVoid(sysaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
29265 +AuStubVoid(sysaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
29266 +
29267 +static inline void sysaufs_brs_init(void)
29268 +{
29269 +       sysaufs_brs = 0;
29270 +}
29271 +
29272 +#endif /* CONFIG_SYSFS */
29273 +
29274 +#endif /* __KERNEL__ */
29275 +#endif /* __SYSAUFS_H__ */
29276 diff -urN /usr/share/empty/fs/aufs/sysfs.c linux/fs/aufs/sysfs.c
29277 --- /usr/share/empty/fs/aufs/sysfs.c    1970-01-01 01:00:00.000000000 +0100
29278 +++ linux/fs/aufs/sysfs.c       2017-05-06 22:16:52.781553511 +0200
29279 @@ -0,0 +1,376 @@
29280 +/*
29281 + * Copyright (C) 2005-2017 Junjiro R. Okajima
29282 + *
29283 + * This program, aufs is free software; you can redistribute it and/or modify
29284 + * it under the terms of the GNU General Public License as published by
29285 + * the Free Software Foundation; either version 2 of the License, or
29286 + * (at your option) any later version.
29287 + *
29288 + * This program is distributed in the hope that it will be useful,
29289 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29290 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29291 + * GNU General Public License for more details.
29292 + *
29293 + * You should have received a copy of the GNU General Public License
29294 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29295 + */
29296 +
29297 +/*
29298 + * sysfs interface
29299 + */
29300 +
29301 +#include <linux/compat.h>
29302 +#include <linux/seq_file.h>
29303 +#include "aufs.h"
29304 +
29305 +#ifdef CONFIG_AUFS_FS_MODULE
29306 +/* this entry violates the "one line per file" policy of sysfs */
29307 +static ssize_t config_show(struct kobject *kobj, struct kobj_attribute *attr,
29308 +                          char *buf)
29309 +{
29310 +       ssize_t err;
29311 +       static char *conf =
29312 +/* this file is generated at compiling */
29313 +#include "conf.str"
29314 +               ;
29315 +
29316 +       err = snprintf(buf, PAGE_SIZE, conf);
29317 +       if (unlikely(err >= PAGE_SIZE))
29318 +               err = -EFBIG;
29319 +       return err;
29320 +}
29321 +
29322 +static struct kobj_attribute au_config_attr = __ATTR_RO(config);
29323 +#endif
29324 +
29325 +static struct attribute *au_attr[] = {
29326 +#ifdef CONFIG_AUFS_FS_MODULE
29327 +       &au_config_attr.attr,
29328 +#endif
29329 +       NULL,   /* need to NULL terminate the list of attributes */
29330 +};
29331 +
29332 +static struct attribute_group sysaufs_attr_group_body = {
29333 +       .attrs = au_attr
29334 +};
29335 +
29336 +struct attribute_group *sysaufs_attr_group = &sysaufs_attr_group_body;
29337 +
29338 +/* ---------------------------------------------------------------------- */
29339 +
29340 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb)
29341 +{
29342 +       int err;
29343 +
29344 +       SiMustAnyLock(sb);
29345 +
29346 +       err = 0;
29347 +       if (au_opt_test(au_mntflags(sb), XINO)) {
29348 +               err = au_xino_path(seq, au_sbi(sb)->si_xib);
29349 +               seq_putc(seq, '\n');
29350 +       }
29351 +       return err;
29352 +}
29353 +
29354 +/*
29355 + * the lifetime of branch is independent from the entry under sysfs.
29356 + * sysfs handles the lifetime of the entry, and never call ->show() after it is
29357 + * unlinked.
29358 + */
29359 +static int sysaufs_si_br(struct seq_file *seq, struct super_block *sb,
29360 +                        aufs_bindex_t bindex, int idx)
29361 +{
29362 +       int err;
29363 +       struct path path;
29364 +       struct dentry *root;
29365 +       struct au_branch *br;
29366 +       au_br_perm_str_t perm;
29367 +
29368 +       AuDbg("b%d\n", bindex);
29369 +
29370 +       err = 0;
29371 +       root = sb->s_root;
29372 +       di_read_lock_parent(root, !AuLock_IR);
29373 +       br = au_sbr(sb, bindex);
29374 +
29375 +       switch (idx) {
29376 +       case AuBrSysfs_BR:
29377 +               path.mnt = au_br_mnt(br);
29378 +               path.dentry = au_h_dptr(root, bindex);
29379 +               err = au_seq_path(seq, &path);
29380 +               if (!err) {
29381 +                       au_optstr_br_perm(&perm, br->br_perm);
29382 +                       seq_printf(seq, "=%s\n", perm.a);
29383 +               }
29384 +               break;
29385 +       case AuBrSysfs_BRID:
29386 +               seq_printf(seq, "%d\n", br->br_id);
29387 +               break;
29388 +       }
29389 +       di_read_unlock(root, !AuLock_IR);
29390 +       if (unlikely(err || seq_has_overflowed(seq)))
29391 +               err = -E2BIG;
29392 +
29393 +       return err;
29394 +}
29395 +
29396 +/* ---------------------------------------------------------------------- */
29397 +
29398 +static struct seq_file *au_seq(char *p, ssize_t len)
29399 +{
29400 +       struct seq_file *seq;
29401 +
29402 +       seq = kzalloc(sizeof(*seq), GFP_NOFS);
29403 +       if (seq) {
29404 +               /* mutex_init(&seq.lock); */
29405 +               seq->buf = p;
29406 +               seq->size = len;
29407 +               return seq; /* success */
29408 +       }
29409 +
29410 +       seq = ERR_PTR(-ENOMEM);
29411 +       return seq;
29412 +}
29413 +
29414 +#define SysaufsBr_PREFIX       "br"
29415 +#define SysaufsBrid_PREFIX     "brid"
29416 +
29417 +/* todo: file size may exceed PAGE_SIZE */
29418 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
29419 +                       char *buf)
29420 +{
29421 +       ssize_t err;
29422 +       int idx;
29423 +       long l;
29424 +       aufs_bindex_t bbot;
29425 +       struct au_sbinfo *sbinfo;
29426 +       struct super_block *sb;
29427 +       struct seq_file *seq;
29428 +       char *name;
29429 +       struct attribute **cattr;
29430 +
29431 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
29432 +       sb = sbinfo->si_sb;
29433 +
29434 +       /*
29435 +        * prevent a race condition between sysfs and aufs.
29436 +        * for instance, sysfs_file_read() calls sysfs_get_active_two() which
29437 +        * prohibits maintaining the sysfs entries.
29438 +        * hew we acquire read lock after sysfs_get_active_two().
29439 +        * on the other hand, the remount process may maintain the sysfs/aufs
29440 +        * entries after acquiring write lock.
29441 +        * it can cause a deadlock.
29442 +        * simply we gave up processing read here.
29443 +        */
29444 +       err = -EBUSY;
29445 +       if (unlikely(!si_noflush_read_trylock(sb)))
29446 +               goto out;
29447 +
29448 +       seq = au_seq(buf, PAGE_SIZE);
29449 +       err = PTR_ERR(seq);
29450 +       if (IS_ERR(seq))
29451 +               goto out_unlock;
29452 +
29453 +       name = (void *)attr->name;
29454 +       cattr = sysaufs_si_attrs;
29455 +       while (*cattr) {
29456 +               if (!strcmp(name, (*cattr)->name)) {
29457 +                       err = container_of(*cattr, struct sysaufs_si_attr, attr)
29458 +                               ->show(seq, sb);
29459 +                       goto out_seq;
29460 +               }
29461 +               cattr++;
29462 +       }
29463 +
29464 +       if (!strncmp(name, SysaufsBrid_PREFIX,
29465 +                    sizeof(SysaufsBrid_PREFIX) - 1)) {
29466 +               idx = AuBrSysfs_BRID;
29467 +               name += sizeof(SysaufsBrid_PREFIX) - 1;
29468 +       } else if (!strncmp(name, SysaufsBr_PREFIX,
29469 +                           sizeof(SysaufsBr_PREFIX) - 1)) {
29470 +               idx = AuBrSysfs_BR;
29471 +               name += sizeof(SysaufsBr_PREFIX) - 1;
29472 +       } else
29473 +                 BUG();
29474 +
29475 +       err = kstrtol(name, 10, &l);
29476 +       if (!err) {
29477 +               bbot = au_sbbot(sb);
29478 +               if (l <= bbot)
29479 +                       err = sysaufs_si_br(seq, sb, (aufs_bindex_t)l, idx);
29480 +               else
29481 +                       err = -ENOENT;
29482 +       }
29483 +
29484 +out_seq:
29485 +       if (!err) {
29486 +               err = seq->count;
29487 +               /* sysfs limit */
29488 +               if (unlikely(err == PAGE_SIZE))
29489 +                       err = -EFBIG;
29490 +       }
29491 +       au_delayed_kfree(seq);
29492 +out_unlock:
29493 +       si_read_unlock(sb);
29494 +out:
29495 +       return err;
29496 +}
29497 +
29498 +/* ---------------------------------------------------------------------- */
29499 +
29500 +static int au_brinfo(struct super_block *sb, union aufs_brinfo __user *arg)
29501 +{
29502 +       int err;
29503 +       int16_t brid;
29504 +       aufs_bindex_t bindex, bbot;
29505 +       size_t sz;
29506 +       char *buf;
29507 +       struct seq_file *seq;
29508 +       struct au_branch *br;
29509 +
29510 +       si_read_lock(sb, AuLock_FLUSH);
29511 +       bbot = au_sbbot(sb);
29512 +       err = bbot + 1;
29513 +       if (!arg)
29514 +               goto out;
29515 +
29516 +       err = -ENOMEM;
29517 +       buf = (void *)__get_free_page(GFP_NOFS);
29518 +       if (unlikely(!buf))
29519 +               goto out;
29520 +
29521 +       seq = au_seq(buf, PAGE_SIZE);
29522 +       err = PTR_ERR(seq);
29523 +       if (IS_ERR(seq))
29524 +               goto out_buf;
29525 +
29526 +       sz = sizeof(*arg) - offsetof(union aufs_brinfo, path);
29527 +       for (bindex = 0; bindex <= bbot; bindex++, arg++) {
29528 +               err = !access_ok(VERIFY_WRITE, arg, sizeof(*arg));
29529 +               if (unlikely(err))
29530 +                       break;
29531 +
29532 +               br = au_sbr(sb, bindex);
29533 +               brid = br->br_id;
29534 +               BUILD_BUG_ON(sizeof(brid) != sizeof(arg->id));
29535 +               err = __put_user(brid, &arg->id);
29536 +               if (unlikely(err))
29537 +                       break;
29538 +
29539 +               BUILD_BUG_ON(sizeof(br->br_perm) != sizeof(arg->perm));
29540 +               err = __put_user(br->br_perm, &arg->perm);
29541 +               if (unlikely(err))
29542 +                       break;
29543 +
29544 +               err = au_seq_path(seq, &br->br_path);
29545 +               if (unlikely(err))
29546 +                       break;
29547 +               seq_putc(seq, '\0');
29548 +               if (!seq_has_overflowed(seq)) {
29549 +                       err = copy_to_user(arg->path, seq->buf, seq->count);
29550 +                       seq->count = 0;
29551 +                       if (unlikely(err))
29552 +                               break;
29553 +               } else {
29554 +                       err = -E2BIG;
29555 +                       goto out_seq;
29556 +               }
29557 +       }
29558 +       if (unlikely(err))
29559 +               err = -EFAULT;
29560 +
29561 +out_seq:
29562 +       au_delayed_kfree(seq);
29563 +out_buf:
29564 +       au_delayed_free_page((unsigned long)buf);
29565 +out:
29566 +       si_read_unlock(sb);
29567 +       return err;
29568 +}
29569 +
29570 +long au_brinfo_ioctl(struct file *file, unsigned long arg)
29571 +{
29572 +       return au_brinfo(file->f_path.dentry->d_sb, (void __user *)arg);
29573 +}
29574 +
29575 +#ifdef CONFIG_COMPAT
29576 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg)
29577 +{
29578 +       return au_brinfo(file->f_path.dentry->d_sb, compat_ptr(arg));
29579 +}
29580 +#endif
29581 +
29582 +/* ---------------------------------------------------------------------- */
29583 +
29584 +void sysaufs_br_init(struct au_branch *br)
29585 +{
29586 +       int i;
29587 +       struct au_brsysfs *br_sysfs;
29588 +       struct attribute *attr;
29589 +
29590 +       br_sysfs = br->br_sysfs;
29591 +       for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
29592 +               attr = &br_sysfs->attr;
29593 +               sysfs_attr_init(attr);
29594 +               attr->name = br_sysfs->name;
29595 +               attr->mode = S_IRUGO;
29596 +               br_sysfs++;
29597 +       }
29598 +}
29599 +
29600 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
29601 +{
29602 +       struct au_branch *br;
29603 +       struct kobject *kobj;
29604 +       struct au_brsysfs *br_sysfs;
29605 +       int i;
29606 +       aufs_bindex_t bbot;
29607 +
29608 +       dbgaufs_brs_del(sb, bindex);
29609 +
29610 +       if (!sysaufs_brs)
29611 +               return;
29612 +
29613 +       kobj = &au_sbi(sb)->si_kobj;
29614 +       bbot = au_sbbot(sb);
29615 +       for (; bindex <= bbot; bindex++) {
29616 +               br = au_sbr(sb, bindex);
29617 +               br_sysfs = br->br_sysfs;
29618 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
29619 +                       sysfs_remove_file(kobj, &br_sysfs->attr);
29620 +                       br_sysfs++;
29621 +               }
29622 +       }
29623 +}
29624 +
29625 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
29626 +{
29627 +       int err, i;
29628 +       aufs_bindex_t bbot;
29629 +       struct kobject *kobj;
29630 +       struct au_branch *br;
29631 +       struct au_brsysfs *br_sysfs;
29632 +
29633 +       dbgaufs_brs_add(sb, bindex);
29634 +
29635 +       if (!sysaufs_brs)
29636 +               return;
29637 +
29638 +       kobj = &au_sbi(sb)->si_kobj;
29639 +       bbot = au_sbbot(sb);
29640 +       for (; bindex <= bbot; bindex++) {
29641 +               br = au_sbr(sb, bindex);
29642 +               br_sysfs = br->br_sysfs;
29643 +               snprintf(br_sysfs[AuBrSysfs_BR].name, sizeof(br_sysfs->name),
29644 +                        SysaufsBr_PREFIX "%d", bindex);
29645 +               snprintf(br_sysfs[AuBrSysfs_BRID].name, sizeof(br_sysfs->name),
29646 +                        SysaufsBrid_PREFIX "%d", bindex);
29647 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
29648 +                       err = sysfs_create_file(kobj, &br_sysfs->attr);
29649 +                       if (unlikely(err))
29650 +                               pr_warn("failed %s under sysfs(%d)\n",
29651 +                                       br_sysfs->name, err);
29652 +                       br_sysfs++;
29653 +               }
29654 +       }
29655 +}
29656 diff -urN /usr/share/empty/fs/aufs/sysrq.c linux/fs/aufs/sysrq.c
29657 --- /usr/share/empty/fs/aufs/sysrq.c    1970-01-01 01:00:00.000000000 +0100
29658 +++ linux/fs/aufs/sysrq.c       2017-05-06 22:16:52.781553511 +0200
29659 @@ -0,0 +1,157 @@
29660 +/*
29661 + * Copyright (C) 2005-2017 Junjiro R. Okajima
29662 + *
29663 + * This program, aufs is free software; you can redistribute it and/or modify
29664 + * it under the terms of the GNU General Public License as published by
29665 + * the Free Software Foundation; either version 2 of the License, or
29666 + * (at your option) any later version.
29667 + *
29668 + * This program is distributed in the hope that it will be useful,
29669 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29670 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29671 + * GNU General Public License for more details.
29672 + *
29673 + * You should have received a copy of the GNU General Public License
29674 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29675 + */
29676 +
29677 +/*
29678 + * magic sysrq hanlder
29679 + */
29680 +
29681 +/* #include <linux/sysrq.h> */
29682 +#include <linux/writeback.h>
29683 +#include "aufs.h"
29684 +
29685 +/* ---------------------------------------------------------------------- */
29686 +
29687 +static void sysrq_sb(struct super_block *sb)
29688 +{
29689 +       char *plevel;
29690 +       struct au_sbinfo *sbinfo;
29691 +       struct file *file;
29692 +       struct au_sphlhead *files;
29693 +       struct au_finfo *finfo;
29694 +
29695 +       plevel = au_plevel;
29696 +       au_plevel = KERN_WARNING;
29697 +
29698 +       /* since we define pr_fmt, call printk directly */
29699 +#define pr(str) printk(KERN_WARNING AUFS_NAME ": " str)
29700 +
29701 +       sbinfo = au_sbi(sb);
29702 +       printk(KERN_WARNING "si=%lx\n", sysaufs_si_id(sbinfo));
29703 +       pr("superblock\n");
29704 +       au_dpri_sb(sb);
29705 +
29706 +#if 0
29707 +       pr("root dentry\n");
29708 +       au_dpri_dentry(sb->s_root);
29709 +       pr("root inode\n");
29710 +       au_dpri_inode(d_inode(sb->s_root));
29711 +#endif
29712 +
29713 +#if 0
29714 +       do {
29715 +               int err, i, j, ndentry;
29716 +               struct au_dcsub_pages dpages;
29717 +               struct au_dpage *dpage;
29718 +
29719 +               err = au_dpages_init(&dpages, GFP_ATOMIC);
29720 +               if (unlikely(err))
29721 +                       break;
29722 +               err = au_dcsub_pages(&dpages, sb->s_root, NULL, NULL);
29723 +               if (!err)
29724 +                       for (i = 0; i < dpages.ndpage; i++) {
29725 +                               dpage = dpages.dpages + i;
29726 +                               ndentry = dpage->ndentry;
29727 +                               for (j = 0; j < ndentry; j++)
29728 +                                       au_dpri_dentry(dpage->dentries[j]);
29729 +                       }
29730 +               au_dpages_free(&dpages);
29731 +       } while (0);
29732 +#endif
29733 +
29734 +#if 1
29735 +       {
29736 +               struct inode *i;
29737 +
29738 +               pr("isolated inode\n");
29739 +               spin_lock(&sb->s_inode_list_lock);
29740 +               list_for_each_entry(i, &sb->s_inodes, i_sb_list) {
29741 +                       spin_lock(&i->i_lock);
29742 +                       if (1 || hlist_empty(&i->i_dentry))
29743 +                               au_dpri_inode(i);
29744 +                       spin_unlock(&i->i_lock);
29745 +               }
29746 +               spin_unlock(&sb->s_inode_list_lock);
29747 +       }
29748 +#endif
29749 +       pr("files\n");
29750 +       files = &au_sbi(sb)->si_files;
29751 +       spin_lock(&files->spin);
29752 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
29753 +               umode_t mode;
29754 +
29755 +               file = finfo->fi_file;
29756 +               mode = file_inode(file)->i_mode;
29757 +               if (!special_file(mode))
29758 +                       au_dpri_file(file);
29759 +       }
29760 +       spin_unlock(&files->spin);
29761 +       pr("done\n");
29762 +
29763 +#undef pr
29764 +       au_plevel = plevel;
29765 +}
29766 +
29767 +/* ---------------------------------------------------------------------- */
29768 +
29769 +/* module parameter */
29770 +static char *aufs_sysrq_key = "a";
29771 +module_param_named(sysrq, aufs_sysrq_key, charp, S_IRUGO);
29772 +MODULE_PARM_DESC(sysrq, "MagicSysRq key for " AUFS_NAME);
29773 +
29774 +static void au_sysrq(int key __maybe_unused)
29775 +{
29776 +       struct au_sbinfo *sbinfo;
29777 +
29778 +       lockdep_off();
29779 +       au_sbilist_lock();
29780 +       hlist_for_each_entry(sbinfo, &au_sbilist.head, si_list)
29781 +               sysrq_sb(sbinfo->si_sb);
29782 +       au_sbilist_unlock();
29783 +       lockdep_on();
29784 +}
29785 +
29786 +static struct sysrq_key_op au_sysrq_op = {
29787 +       .handler        = au_sysrq,
29788 +       .help_msg       = "Aufs",
29789 +       .action_msg     = "Aufs",
29790 +       .enable_mask    = SYSRQ_ENABLE_DUMP
29791 +};
29792 +
29793 +/* ---------------------------------------------------------------------- */
29794 +
29795 +int __init au_sysrq_init(void)
29796 +{
29797 +       int err;
29798 +       char key;
29799 +
29800 +       err = -1;
29801 +       key = *aufs_sysrq_key;
29802 +       if ('a' <= key && key <= 'z')
29803 +               err = register_sysrq_key(key, &au_sysrq_op);
29804 +       if (unlikely(err))
29805 +               pr_err("err %d, sysrq=%c\n", err, key);
29806 +       return err;
29807 +}
29808 +
29809 +void au_sysrq_fin(void)
29810 +{
29811 +       int err;
29812 +
29813 +       err = unregister_sysrq_key(*aufs_sysrq_key, &au_sysrq_op);
29814 +       if (unlikely(err))
29815 +               pr_err("err %d (ignored)\n", err);
29816 +}
29817 diff -urN /usr/share/empty/fs/aufs/vdir.c linux/fs/aufs/vdir.c
29818 --- /usr/share/empty/fs/aufs/vdir.c     1970-01-01 01:00:00.000000000 +0100
29819 +++ linux/fs/aufs/vdir.c        2017-05-06 22:16:52.781553511 +0200
29820 @@ -0,0 +1,900 @@
29821 +/*
29822 + * Copyright (C) 2005-2017 Junjiro R. Okajima
29823 + *
29824 + * This program, aufs is free software; you can redistribute it and/or modify
29825 + * it under the terms of the GNU General Public License as published by
29826 + * the Free Software Foundation; either version 2 of the License, or
29827 + * (at your option) any later version.
29828 + *
29829 + * This program is distributed in the hope that it will be useful,
29830 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29831 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29832 + * GNU General Public License for more details.
29833 + *
29834 + * You should have received a copy of the GNU General Public License
29835 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29836 + */
29837 +
29838 +/*
29839 + * virtual or vertical directory
29840 + */
29841 +
29842 +#include "aufs.h"
29843 +
29844 +static unsigned int calc_size(int nlen)
29845 +{
29846 +       return ALIGN(sizeof(struct au_vdir_de) + nlen, sizeof(ino_t));
29847 +}
29848 +
29849 +static int set_deblk_end(union au_vdir_deblk_p *p,
29850 +                        union au_vdir_deblk_p *deblk_end)
29851 +{
29852 +       if (calc_size(0) <= deblk_end->deblk - p->deblk) {
29853 +               p->de->de_str.len = 0;
29854 +               /* smp_mb(); */
29855 +               return 0;
29856 +       }
29857 +       return -1; /* error */
29858 +}
29859 +
29860 +/* returns true or false */
29861 +static int is_deblk_end(union au_vdir_deblk_p *p,
29862 +                       union au_vdir_deblk_p *deblk_end)
29863 +{
29864 +       if (calc_size(0) <= deblk_end->deblk - p->deblk)
29865 +               return !p->de->de_str.len;
29866 +       return 1;
29867 +}
29868 +
29869 +static unsigned char *last_deblk(struct au_vdir *vdir)
29870 +{
29871 +       return vdir->vd_deblk[vdir->vd_nblk - 1];
29872 +}
29873 +
29874 +/* ---------------------------------------------------------------------- */
29875 +
29876 +/* estimate the appropriate size for name hash table */
29877 +unsigned int au_rdhash_est(loff_t sz)
29878 +{
29879 +       unsigned int n;
29880 +
29881 +       n = UINT_MAX;
29882 +       sz >>= 10;
29883 +       if (sz < n)
29884 +               n = sz;
29885 +       if (sz < AUFS_RDHASH_DEF)
29886 +               n = AUFS_RDHASH_DEF;
29887 +       /* pr_info("n %u\n", n); */
29888 +       return n;
29889 +}
29890 +
29891 +/*
29892 + * the allocated memory has to be freed by
29893 + * au_nhash_wh_free() or au_nhash_de_free().
29894 + */
29895 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp)
29896 +{
29897 +       struct hlist_head *head;
29898 +       unsigned int u;
29899 +       size_t sz;
29900 +
29901 +       sz = sizeof(*nhash->nh_head) * num_hash;
29902 +       head = kmalloc(sz, gfp);
29903 +       if (head) {
29904 +               nhash->nh_num = num_hash;
29905 +               nhash->nh_head = head;
29906 +               for (u = 0; u < num_hash; u++)
29907 +                       INIT_HLIST_HEAD(head++);
29908 +               return 0; /* success */
29909 +       }
29910 +
29911 +       return -ENOMEM;
29912 +}
29913 +
29914 +static void nhash_count(struct hlist_head *head)
29915 +{
29916 +#if 0
29917 +       unsigned long n;
29918 +       struct hlist_node *pos;
29919 +
29920 +       n = 0;
29921 +       hlist_for_each(pos, head)
29922 +               n++;
29923 +       pr_info("%lu\n", n);
29924 +#endif
29925 +}
29926 +
29927 +static void au_nhash_wh_do_free(struct hlist_head *head)
29928 +{
29929 +       struct au_vdir_wh *pos;
29930 +       struct hlist_node *node;
29931 +
29932 +       hlist_for_each_entry_safe(pos, node, head, wh_hash)
29933 +               au_delayed_kfree(pos);
29934 +}
29935 +
29936 +static void au_nhash_de_do_free(struct hlist_head *head)
29937 +{
29938 +       struct au_vdir_dehstr *pos;
29939 +       struct hlist_node *node;
29940 +
29941 +       hlist_for_each_entry_safe(pos, node, head, hash)
29942 +               au_cache_dfree_vdir_dehstr(pos);
29943 +}
29944 +
29945 +static void au_nhash_do_free(struct au_nhash *nhash,
29946 +                            void (*free)(struct hlist_head *head))
29947 +{
29948 +       unsigned int n;
29949 +       struct hlist_head *head;
29950 +
29951 +       n = nhash->nh_num;
29952 +       if (!n)
29953 +               return;
29954 +
29955 +       head = nhash->nh_head;
29956 +       while (n-- > 0) {
29957 +               nhash_count(head);
29958 +               free(head++);
29959 +       }
29960 +       au_delayed_kfree(nhash->nh_head);
29961 +}
29962 +
29963 +void au_nhash_wh_free(struct au_nhash *whlist)
29964 +{
29965 +       au_nhash_do_free(whlist, au_nhash_wh_do_free);
29966 +}
29967 +
29968 +static void au_nhash_de_free(struct au_nhash *delist)
29969 +{
29970 +       au_nhash_do_free(delist, au_nhash_de_do_free);
29971 +}
29972 +
29973 +/* ---------------------------------------------------------------------- */
29974 +
29975 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
29976 +                           int limit)
29977 +{
29978 +       int num;
29979 +       unsigned int u, n;
29980 +       struct hlist_head *head;
29981 +       struct au_vdir_wh *pos;
29982 +
29983 +       num = 0;
29984 +       n = whlist->nh_num;
29985 +       head = whlist->nh_head;
29986 +       for (u = 0; u < n; u++, head++)
29987 +               hlist_for_each_entry(pos, head, wh_hash)
29988 +                       if (pos->wh_bindex == btgt && ++num > limit)
29989 +                               return 1;
29990 +       return 0;
29991 +}
29992 +
29993 +static struct hlist_head *au_name_hash(struct au_nhash *nhash,
29994 +                                      unsigned char *name,
29995 +                                      unsigned int len)
29996 +{
29997 +       unsigned int v;
29998 +       /* const unsigned int magic_bit = 12; */
29999 +
30000 +       AuDebugOn(!nhash->nh_num || !nhash->nh_head);
30001 +
30002 +       v = 0;
30003 +       if (len > 8)
30004 +               len = 8;
30005 +       while (len--)
30006 +               v += *name++;
30007 +       /* v = hash_long(v, magic_bit); */
30008 +       v %= nhash->nh_num;
30009 +       return nhash->nh_head + v;
30010 +}
30011 +
30012 +static int au_nhash_test_name(struct au_vdir_destr *str, const char *name,
30013 +                             int nlen)
30014 +{
30015 +       return str->len == nlen && !memcmp(str->name, name, nlen);
30016 +}
30017 +
30018 +/* returns found or not */
30019 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen)
30020 +{
30021 +       struct hlist_head *head;
30022 +       struct au_vdir_wh *pos;
30023 +       struct au_vdir_destr *str;
30024 +
30025 +       head = au_name_hash(whlist, name, nlen);
30026 +       hlist_for_each_entry(pos, head, wh_hash) {
30027 +               str = &pos->wh_str;
30028 +               AuDbg("%.*s\n", str->len, str->name);
30029 +               if (au_nhash_test_name(str, name, nlen))
30030 +                       return 1;
30031 +       }
30032 +       return 0;
30033 +}
30034 +
30035 +/* returns found(true) or not */
30036 +static int test_known(struct au_nhash *delist, char *name, int nlen)
30037 +{
30038 +       struct hlist_head *head;
30039 +       struct au_vdir_dehstr *pos;
30040 +       struct au_vdir_destr *str;
30041 +
30042 +       head = au_name_hash(delist, name, nlen);
30043 +       hlist_for_each_entry(pos, head, hash) {
30044 +               str = pos->str;
30045 +               AuDbg("%.*s\n", str->len, str->name);
30046 +               if (au_nhash_test_name(str, name, nlen))
30047 +                       return 1;
30048 +       }
30049 +       return 0;
30050 +}
30051 +
30052 +static void au_shwh_init_wh(struct au_vdir_wh *wh, ino_t ino,
30053 +                           unsigned char d_type)
30054 +{
30055 +#ifdef CONFIG_AUFS_SHWH
30056 +       wh->wh_ino = ino;
30057 +       wh->wh_type = d_type;
30058 +#endif
30059 +}
30060 +
30061 +/* ---------------------------------------------------------------------- */
30062 +
30063 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
30064 +                      unsigned int d_type, aufs_bindex_t bindex,
30065 +                      unsigned char shwh)
30066 +{
30067 +       int err;
30068 +       struct au_vdir_destr *str;
30069 +       struct au_vdir_wh *wh;
30070 +
30071 +       AuDbg("%.*s\n", nlen, name);
30072 +       AuDebugOn(!whlist->nh_num || !whlist->nh_head);
30073 +
30074 +       err = -ENOMEM;
30075 +       wh = kmalloc(sizeof(*wh) + nlen, GFP_NOFS);
30076 +       if (unlikely(!wh))
30077 +               goto out;
30078 +
30079 +       err = 0;
30080 +       wh->wh_bindex = bindex;
30081 +       if (shwh)
30082 +               au_shwh_init_wh(wh, ino, d_type);
30083 +       str = &wh->wh_str;
30084 +       str->len = nlen;
30085 +       memcpy(str->name, name, nlen);
30086 +       hlist_add_head(&wh->wh_hash, au_name_hash(whlist, name, nlen));
30087 +       /* smp_mb(); */
30088 +
30089 +out:
30090 +       return err;
30091 +}
30092 +
30093 +static int append_deblk(struct au_vdir *vdir)
30094 +{
30095 +       int err;
30096 +       unsigned long ul;
30097 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
30098 +       union au_vdir_deblk_p p, deblk_end;
30099 +       unsigned char **o;
30100 +
30101 +       err = -ENOMEM;
30102 +       o = au_krealloc(vdir->vd_deblk, sizeof(*o) * (vdir->vd_nblk + 1),
30103 +                       GFP_NOFS, /*may_shrink*/0);
30104 +       if (unlikely(!o))
30105 +               goto out;
30106 +
30107 +       vdir->vd_deblk = o;
30108 +       p.deblk = kmalloc(deblk_sz, GFP_NOFS);
30109 +       if (p.deblk) {
30110 +               ul = vdir->vd_nblk++;
30111 +               vdir->vd_deblk[ul] = p.deblk;
30112 +               vdir->vd_last.ul = ul;
30113 +               vdir->vd_last.p.deblk = p.deblk;
30114 +               deblk_end.deblk = p.deblk + deblk_sz;
30115 +               err = set_deblk_end(&p, &deblk_end);
30116 +       }
30117 +
30118 +out:
30119 +       return err;
30120 +}
30121 +
30122 +static int append_de(struct au_vdir *vdir, char *name, int nlen, ino_t ino,
30123 +                    unsigned int d_type, struct au_nhash *delist)
30124 +{
30125 +       int err;
30126 +       unsigned int sz;
30127 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
30128 +       union au_vdir_deblk_p p, *room, deblk_end;
30129 +       struct au_vdir_dehstr *dehstr;
30130 +
30131 +       p.deblk = last_deblk(vdir);
30132 +       deblk_end.deblk = p.deblk + deblk_sz;
30133 +       room = &vdir->vd_last.p;
30134 +       AuDebugOn(room->deblk < p.deblk || deblk_end.deblk <= room->deblk
30135 +                 || !is_deblk_end(room, &deblk_end));
30136 +
30137 +       sz = calc_size(nlen);
30138 +       if (unlikely(sz > deblk_end.deblk - room->deblk)) {
30139 +               err = append_deblk(vdir);
30140 +               if (unlikely(err))
30141 +                       goto out;
30142 +
30143 +               p.deblk = last_deblk(vdir);
30144 +               deblk_end.deblk = p.deblk + deblk_sz;
30145 +               /* smp_mb(); */
30146 +               AuDebugOn(room->deblk != p.deblk);
30147 +       }
30148 +
30149 +       err = -ENOMEM;
30150 +       dehstr = au_cache_alloc_vdir_dehstr();
30151 +       if (unlikely(!dehstr))
30152 +               goto out;
30153 +
30154 +       dehstr->str = &room->de->de_str;
30155 +       hlist_add_head(&dehstr->hash, au_name_hash(delist, name, nlen));
30156 +       room->de->de_ino = ino;
30157 +       room->de->de_type = d_type;
30158 +       room->de->de_str.len = nlen;
30159 +       memcpy(room->de->de_str.name, name, nlen);
30160 +
30161 +       err = 0;
30162 +       room->deblk += sz;
30163 +       if (unlikely(set_deblk_end(room, &deblk_end)))
30164 +               err = append_deblk(vdir);
30165 +       /* smp_mb(); */
30166 +
30167 +out:
30168 +       return err;
30169 +}
30170 +
30171 +/* ---------------------------------------------------------------------- */
30172 +
30173 +void au_vdir_free(struct au_vdir *vdir, int atonce)
30174 +{
30175 +       unsigned char **deblk;
30176 +
30177 +       deblk = vdir->vd_deblk;
30178 +       if (!atonce) {
30179 +               while (vdir->vd_nblk--)
30180 +                       au_delayed_kfree(*deblk++);
30181 +               au_delayed_kfree(vdir->vd_deblk);
30182 +               au_cache_dfree_vdir(vdir);
30183 +       } else {
30184 +               /* not delayed */
30185 +               while (vdir->vd_nblk--)
30186 +                       kfree(*deblk++);
30187 +               kfree(vdir->vd_deblk);
30188 +               au_cache_free_vdir(vdir);
30189 +       }
30190 +}
30191 +
30192 +static struct au_vdir *alloc_vdir(struct file *file)
30193 +{
30194 +       struct au_vdir *vdir;
30195 +       struct super_block *sb;
30196 +       int err;
30197 +
30198 +       sb = file->f_path.dentry->d_sb;
30199 +       SiMustAnyLock(sb);
30200 +
30201 +       err = -ENOMEM;
30202 +       vdir = au_cache_alloc_vdir();
30203 +       if (unlikely(!vdir))
30204 +               goto out;
30205 +
30206 +       vdir->vd_deblk = kzalloc(sizeof(*vdir->vd_deblk), GFP_NOFS);
30207 +       if (unlikely(!vdir->vd_deblk))
30208 +               goto out_free;
30209 +
30210 +       vdir->vd_deblk_sz = au_sbi(sb)->si_rdblk;
30211 +       if (!vdir->vd_deblk_sz) {
30212 +               /* estimate the appropriate size for deblk */
30213 +               vdir->vd_deblk_sz = au_dir_size(file, /*dentry*/NULL);
30214 +               /* pr_info("vd_deblk_sz %u\n", vdir->vd_deblk_sz); */
30215 +       }
30216 +       vdir->vd_nblk = 0;
30217 +       vdir->vd_version = 0;
30218 +       vdir->vd_jiffy = 0;
30219 +       err = append_deblk(vdir);
30220 +       if (!err)
30221 +               return vdir; /* success */
30222 +
30223 +       au_delayed_kfree(vdir->vd_deblk);
30224 +
30225 +out_free:
30226 +       au_cache_dfree_vdir(vdir);
30227 +out:
30228 +       vdir = ERR_PTR(err);
30229 +       return vdir;
30230 +}
30231 +
30232 +static int reinit_vdir(struct au_vdir *vdir)
30233 +{
30234 +       int err;
30235 +       union au_vdir_deblk_p p, deblk_end;
30236 +
30237 +       while (vdir->vd_nblk > 1) {
30238 +               au_delayed_kfree(vdir->vd_deblk[vdir->vd_nblk - 1]);
30239 +               /* vdir->vd_deblk[vdir->vd_nblk - 1] = NULL; */
30240 +               vdir->vd_nblk--;
30241 +       }
30242 +       p.deblk = vdir->vd_deblk[0];
30243 +       deblk_end.deblk = p.deblk + vdir->vd_deblk_sz;
30244 +       err = set_deblk_end(&p, &deblk_end);
30245 +       /* keep vd_dblk_sz */
30246 +       vdir->vd_last.ul = 0;
30247 +       vdir->vd_last.p.deblk = vdir->vd_deblk[0];
30248 +       vdir->vd_version = 0;
30249 +       vdir->vd_jiffy = 0;
30250 +       /* smp_mb(); */
30251 +       return err;
30252 +}
30253 +
30254 +/* ---------------------------------------------------------------------- */
30255 +
30256 +#define AuFillVdir_CALLED      1
30257 +#define AuFillVdir_WHABLE      (1 << 1)
30258 +#define AuFillVdir_SHWH                (1 << 2)
30259 +#define au_ftest_fillvdir(flags, name) ((flags) & AuFillVdir_##name)
30260 +#define au_fset_fillvdir(flags, name) \
30261 +       do { (flags) |= AuFillVdir_##name; } while (0)
30262 +#define au_fclr_fillvdir(flags, name) \
30263 +       do { (flags) &= ~AuFillVdir_##name; } while (0)
30264 +
30265 +#ifndef CONFIG_AUFS_SHWH
30266 +#undef AuFillVdir_SHWH
30267 +#define AuFillVdir_SHWH                0
30268 +#endif
30269 +
30270 +struct fillvdir_arg {
30271 +       struct dir_context      ctx;
30272 +       struct file             *file;
30273 +       struct au_vdir          *vdir;
30274 +       struct au_nhash         delist;
30275 +       struct au_nhash         whlist;
30276 +       aufs_bindex_t           bindex;
30277 +       unsigned int            flags;
30278 +       int                     err;
30279 +};
30280 +
30281 +static int fillvdir(struct dir_context *ctx, const char *__name, int nlen,
30282 +                   loff_t offset __maybe_unused, u64 h_ino,
30283 +                   unsigned int d_type)
30284 +{
30285 +       struct fillvdir_arg *arg = container_of(ctx, struct fillvdir_arg, ctx);
30286 +       char *name = (void *)__name;
30287 +       struct super_block *sb;
30288 +       ino_t ino;
30289 +       const unsigned char shwh = !!au_ftest_fillvdir(arg->flags, SHWH);
30290 +
30291 +       arg->err = 0;
30292 +       sb = arg->file->f_path.dentry->d_sb;
30293 +       au_fset_fillvdir(arg->flags, CALLED);
30294 +       /* smp_mb(); */
30295 +       if (nlen <= AUFS_WH_PFX_LEN
30296 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
30297 +               if (test_known(&arg->delist, name, nlen)
30298 +                   || au_nhash_test_known_wh(&arg->whlist, name, nlen))
30299 +                       goto out; /* already exists or whiteouted */
30300 +
30301 +               arg->err = au_ino(sb, arg->bindex, h_ino, d_type, &ino);
30302 +               if (!arg->err) {
30303 +                       if (unlikely(nlen > AUFS_MAX_NAMELEN))
30304 +                               d_type = DT_UNKNOWN;
30305 +                       arg->err = append_de(arg->vdir, name, nlen, ino,
30306 +                                            d_type, &arg->delist);
30307 +               }
30308 +       } else if (au_ftest_fillvdir(arg->flags, WHABLE)) {
30309 +               name += AUFS_WH_PFX_LEN;
30310 +               nlen -= AUFS_WH_PFX_LEN;
30311 +               if (au_nhash_test_known_wh(&arg->whlist, name, nlen))
30312 +                       goto out; /* already whiteouted */
30313 +
30314 +               if (shwh)
30315 +                       arg->err = au_wh_ino(sb, arg->bindex, h_ino, d_type,
30316 +                                            &ino);
30317 +               if (!arg->err) {
30318 +                       if (nlen <= AUFS_MAX_NAMELEN + AUFS_WH_PFX_LEN)
30319 +                               d_type = DT_UNKNOWN;
30320 +                       arg->err = au_nhash_append_wh
30321 +                               (&arg->whlist, name, nlen, ino, d_type,
30322 +                                arg->bindex, shwh);
30323 +               }
30324 +       }
30325 +
30326 +out:
30327 +       if (!arg->err)
30328 +               arg->vdir->vd_jiffy = jiffies;
30329 +       /* smp_mb(); */
30330 +       AuTraceErr(arg->err);
30331 +       return arg->err;
30332 +}
30333 +
30334 +static int au_handle_shwh(struct super_block *sb, struct au_vdir *vdir,
30335 +                         struct au_nhash *whlist, struct au_nhash *delist)
30336 +{
30337 +#ifdef CONFIG_AUFS_SHWH
30338 +       int err;
30339 +       unsigned int nh, u;
30340 +       struct hlist_head *head;
30341 +       struct au_vdir_wh *pos;
30342 +       struct hlist_node *n;
30343 +       char *p, *o;
30344 +       struct au_vdir_destr *destr;
30345 +
30346 +       AuDebugOn(!au_opt_test(au_mntflags(sb), SHWH));
30347 +
30348 +       err = -ENOMEM;
30349 +       o = p = (void *)__get_free_page(GFP_NOFS);
30350 +       if (unlikely(!p))
30351 +               goto out;
30352 +
30353 +       err = 0;
30354 +       nh = whlist->nh_num;
30355 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
30356 +       p += AUFS_WH_PFX_LEN;
30357 +       for (u = 0; u < nh; u++) {
30358 +               head = whlist->nh_head + u;
30359 +               hlist_for_each_entry_safe(pos, n, head, wh_hash) {
30360 +                       destr = &pos->wh_str;
30361 +                       memcpy(p, destr->name, destr->len);
30362 +                       err = append_de(vdir, o, destr->len + AUFS_WH_PFX_LEN,
30363 +                                       pos->wh_ino, pos->wh_type, delist);
30364 +                       if (unlikely(err))
30365 +                               break;
30366 +               }
30367 +       }
30368 +
30369 +       au_delayed_free_page((unsigned long)o);
30370 +
30371 +out:
30372 +       AuTraceErr(err);
30373 +       return err;
30374 +#else
30375 +       return 0;
30376 +#endif
30377 +}
30378 +
30379 +static int au_do_read_vdir(struct fillvdir_arg *arg)
30380 +{
30381 +       int err;
30382 +       unsigned int rdhash;
30383 +       loff_t offset;
30384 +       aufs_bindex_t bbot, bindex, btop;
30385 +       unsigned char shwh;
30386 +       struct file *hf, *file;
30387 +       struct super_block *sb;
30388 +
30389 +       file = arg->file;
30390 +       sb = file->f_path.dentry->d_sb;
30391 +       SiMustAnyLock(sb);
30392 +
30393 +       rdhash = au_sbi(sb)->si_rdhash;
30394 +       if (!rdhash)
30395 +               rdhash = au_rdhash_est(au_dir_size(file, /*dentry*/NULL));
30396 +       err = au_nhash_alloc(&arg->delist, rdhash, GFP_NOFS);
30397 +       if (unlikely(err))
30398 +               goto out;
30399 +       err = au_nhash_alloc(&arg->whlist, rdhash, GFP_NOFS);
30400 +       if (unlikely(err))
30401 +               goto out_delist;
30402 +
30403 +       err = 0;
30404 +       arg->flags = 0;
30405 +       shwh = 0;
30406 +       if (au_opt_test(au_mntflags(sb), SHWH)) {
30407 +               shwh = 1;
30408 +               au_fset_fillvdir(arg->flags, SHWH);
30409 +       }
30410 +       btop = au_fbtop(file);
30411 +       bbot = au_fbbot_dir(file);
30412 +       for (bindex = btop; !err && bindex <= bbot; bindex++) {
30413 +               hf = au_hf_dir(file, bindex);
30414 +               if (!hf)
30415 +                       continue;
30416 +
30417 +               offset = vfsub_llseek(hf, 0, SEEK_SET);
30418 +               err = offset;
30419 +               if (unlikely(offset))
30420 +                       break;
30421 +
30422 +               arg->bindex = bindex;
30423 +               au_fclr_fillvdir(arg->flags, WHABLE);
30424 +               if (shwh
30425 +                   || (bindex != bbot
30426 +                       && au_br_whable(au_sbr_perm(sb, bindex))))
30427 +                       au_fset_fillvdir(arg->flags, WHABLE);
30428 +               do {
30429 +                       arg->err = 0;
30430 +                       au_fclr_fillvdir(arg->flags, CALLED);
30431 +                       /* smp_mb(); */
30432 +                       err = vfsub_iterate_dir(hf, &arg->ctx);
30433 +                       if (err >= 0)
30434 +                               err = arg->err;
30435 +               } while (!err && au_ftest_fillvdir(arg->flags, CALLED));
30436 +
30437 +               /*
30438 +                * dir_relax() may be good for concurrency, but aufs should not
30439 +                * use it since it will cause a lockdep problem.
30440 +                */
30441 +       }
30442 +
30443 +       if (!err && shwh)
30444 +               err = au_handle_shwh(sb, arg->vdir, &arg->whlist, &arg->delist);
30445 +
30446 +       au_nhash_wh_free(&arg->whlist);
30447 +
30448 +out_delist:
30449 +       au_nhash_de_free(&arg->delist);
30450 +out:
30451 +       return err;
30452 +}
30453 +
30454 +static int read_vdir(struct file *file, int may_read)
30455 +{
30456 +       int err;
30457 +       unsigned long expire;
30458 +       unsigned char do_read;
30459 +       struct fillvdir_arg arg = {
30460 +               .ctx = {
30461 +                       .actor = fillvdir
30462 +               }
30463 +       };
30464 +       struct inode *inode;
30465 +       struct au_vdir *vdir, *allocated;
30466 +
30467 +       err = 0;
30468 +       inode = file_inode(file);
30469 +       IMustLock(inode);
30470 +       IiMustWriteLock(inode);
30471 +       SiMustAnyLock(inode->i_sb);
30472 +
30473 +       allocated = NULL;
30474 +       do_read = 0;
30475 +       expire = au_sbi(inode->i_sb)->si_rdcache;
30476 +       vdir = au_ivdir(inode);
30477 +       if (!vdir) {
30478 +               do_read = 1;
30479 +               vdir = alloc_vdir(file);
30480 +               err = PTR_ERR(vdir);
30481 +               if (IS_ERR(vdir))
30482 +                       goto out;
30483 +               err = 0;
30484 +               allocated = vdir;
30485 +       } else if (may_read
30486 +                  && (inode->i_version != vdir->vd_version
30487 +                      || time_after(jiffies, vdir->vd_jiffy + expire))) {
30488 +               do_read = 1;
30489 +               err = reinit_vdir(vdir);
30490 +               if (unlikely(err))
30491 +                       goto out;
30492 +       }
30493 +
30494 +       if (!do_read)
30495 +               return 0; /* success */
30496 +
30497 +       arg.file = file;
30498 +       arg.vdir = vdir;
30499 +       err = au_do_read_vdir(&arg);
30500 +       if (!err) {
30501 +               /* file->f_pos = 0; */ /* todo: ctx->pos? */
30502 +               vdir->vd_version = inode->i_version;
30503 +               vdir->vd_last.ul = 0;
30504 +               vdir->vd_last.p.deblk = vdir->vd_deblk[0];
30505 +               if (allocated)
30506 +                       au_set_ivdir(inode, allocated);
30507 +       } else if (allocated)
30508 +               au_vdir_free(allocated, /*atonce*/0);
30509 +
30510 +out:
30511 +       return err;
30512 +}
30513 +
30514 +static int copy_vdir(struct au_vdir *tgt, struct au_vdir *src)
30515 +{
30516 +       int err, rerr;
30517 +       unsigned long ul, n;
30518 +       const unsigned int deblk_sz = src->vd_deblk_sz;
30519 +
30520 +       AuDebugOn(tgt->vd_nblk != 1);
30521 +
30522 +       err = -ENOMEM;
30523 +       if (tgt->vd_nblk < src->vd_nblk) {
30524 +               unsigned char **p;
30525 +
30526 +               p = au_krealloc(tgt->vd_deblk, sizeof(*p) * src->vd_nblk,
30527 +                               GFP_NOFS, /*may_shrink*/0);
30528 +               if (unlikely(!p))
30529 +                       goto out;
30530 +               tgt->vd_deblk = p;
30531 +       }
30532 +
30533 +       if (tgt->vd_deblk_sz != deblk_sz) {
30534 +               unsigned char *p;
30535 +
30536 +               tgt->vd_deblk_sz = deblk_sz;
30537 +               p = au_krealloc(tgt->vd_deblk[0], deblk_sz, GFP_NOFS,
30538 +                               /*may_shrink*/1);
30539 +               if (unlikely(!p))
30540 +                       goto out;
30541 +               tgt->vd_deblk[0] = p;
30542 +       }
30543 +       memcpy(tgt->vd_deblk[0], src->vd_deblk[0], deblk_sz);
30544 +       tgt->vd_version = src->vd_version;
30545 +       tgt->vd_jiffy = src->vd_jiffy;
30546 +
30547 +       n = src->vd_nblk;
30548 +       for (ul = 1; ul < n; ul++) {
30549 +               tgt->vd_deblk[ul] = kmemdup(src->vd_deblk[ul], deblk_sz,
30550 +                                           GFP_NOFS);
30551 +               if (unlikely(!tgt->vd_deblk[ul]))
30552 +                       goto out;
30553 +               tgt->vd_nblk++;
30554 +       }
30555 +       tgt->vd_nblk = n;
30556 +       tgt->vd_last.ul = tgt->vd_last.ul;
30557 +       tgt->vd_last.p.deblk = tgt->vd_deblk[tgt->vd_last.ul];
30558 +       tgt->vd_last.p.deblk += src->vd_last.p.deblk
30559 +               - src->vd_deblk[src->vd_last.ul];
30560 +       /* smp_mb(); */
30561 +       return 0; /* success */
30562 +
30563 +out:
30564 +       rerr = reinit_vdir(tgt);
30565 +       BUG_ON(rerr);
30566 +       return err;
30567 +}
30568 +
30569 +int au_vdir_init(struct file *file)
30570 +{
30571 +       int err;
30572 +       struct inode *inode;
30573 +       struct au_vdir *vdir_cache, *allocated;
30574 +
30575 +       /* test file->f_pos here instead of ctx->pos */
30576 +       err = read_vdir(file, !file->f_pos);
30577 +       if (unlikely(err))
30578 +               goto out;
30579 +
30580 +       allocated = NULL;
30581 +       vdir_cache = au_fvdir_cache(file);
30582 +       if (!vdir_cache) {
30583 +               vdir_cache = alloc_vdir(file);
30584 +               err = PTR_ERR(vdir_cache);
30585 +               if (IS_ERR(vdir_cache))
30586 +                       goto out;
30587 +               allocated = vdir_cache;
30588 +       } else if (!file->f_pos && vdir_cache->vd_version != file->f_version) {
30589 +               /* test file->f_pos here instead of ctx->pos */
30590 +               err = reinit_vdir(vdir_cache);
30591 +               if (unlikely(err))
30592 +                       goto out;
30593 +       } else
30594 +               return 0; /* success */
30595 +
30596 +       inode = file_inode(file);
30597 +       err = copy_vdir(vdir_cache, au_ivdir(inode));
30598 +       if (!err) {
30599 +               file->f_version = inode->i_version;
30600 +               if (allocated)
30601 +                       au_set_fvdir_cache(file, allocated);
30602 +       } else if (allocated)
30603 +               au_vdir_free(allocated, /*atonce*/0);
30604 +
30605 +out:
30606 +       return err;
30607 +}
30608 +
30609 +static loff_t calc_offset(struct au_vdir *vdir)
30610 +{
30611 +       loff_t offset;
30612 +       union au_vdir_deblk_p p;
30613 +
30614 +       p.deblk = vdir->vd_deblk[vdir->vd_last.ul];
30615 +       offset = vdir->vd_last.p.deblk - p.deblk;
30616 +       offset += vdir->vd_deblk_sz * vdir->vd_last.ul;
30617 +       return offset;
30618 +}
30619 +
30620 +/* returns true or false */
30621 +static int seek_vdir(struct file *file, struct dir_context *ctx)
30622 +{
30623 +       int valid;
30624 +       unsigned int deblk_sz;
30625 +       unsigned long ul, n;
30626 +       loff_t offset;
30627 +       union au_vdir_deblk_p p, deblk_end;
30628 +       struct au_vdir *vdir_cache;
30629 +
30630 +       valid = 1;
30631 +       vdir_cache = au_fvdir_cache(file);
30632 +       offset = calc_offset(vdir_cache);
30633 +       AuDbg("offset %lld\n", offset);
30634 +       if (ctx->pos == offset)
30635 +               goto out;
30636 +
30637 +       vdir_cache->vd_last.ul = 0;
30638 +       vdir_cache->vd_last.p.deblk = vdir_cache->vd_deblk[0];
30639 +       if (!ctx->pos)
30640 +               goto out;
30641 +
30642 +       valid = 0;
30643 +       deblk_sz = vdir_cache->vd_deblk_sz;
30644 +       ul = div64_u64(ctx->pos, deblk_sz);
30645 +       AuDbg("ul %lu\n", ul);
30646 +       if (ul >= vdir_cache->vd_nblk)
30647 +               goto out;
30648 +
30649 +       n = vdir_cache->vd_nblk;
30650 +       for (; ul < n; ul++) {
30651 +               p.deblk = vdir_cache->vd_deblk[ul];
30652 +               deblk_end.deblk = p.deblk + deblk_sz;
30653 +               offset = ul;
30654 +               offset *= deblk_sz;
30655 +               while (!is_deblk_end(&p, &deblk_end) && offset < ctx->pos) {
30656 +                       unsigned int l;
30657 +
30658 +                       l = calc_size(p.de->de_str.len);
30659 +                       offset += l;
30660 +                       p.deblk += l;
30661 +               }
30662 +               if (!is_deblk_end(&p, &deblk_end)) {
30663 +                       valid = 1;
30664 +                       vdir_cache->vd_last.ul = ul;
30665 +                       vdir_cache->vd_last.p = p;
30666 +                       break;
30667 +               }
30668 +       }
30669 +
30670 +out:
30671 +       /* smp_mb(); */
30672 +       AuTraceErr(!valid);
30673 +       return valid;
30674 +}
30675 +
30676 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx)
30677 +{
30678 +       unsigned int l, deblk_sz;
30679 +       union au_vdir_deblk_p deblk_end;
30680 +       struct au_vdir *vdir_cache;
30681 +       struct au_vdir_de *de;
30682 +
30683 +       vdir_cache = au_fvdir_cache(file);
30684 +       if (!seek_vdir(file, ctx))
30685 +               return 0;
30686 +
30687 +       deblk_sz = vdir_cache->vd_deblk_sz;
30688 +       while (1) {
30689 +               deblk_end.deblk = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
30690 +               deblk_end.deblk += deblk_sz;
30691 +               while (!is_deblk_end(&vdir_cache->vd_last.p, &deblk_end)) {
30692 +                       de = vdir_cache->vd_last.p.de;
30693 +                       AuDbg("%.*s, off%lld, i%lu, dt%d\n",
30694 +                             de->de_str.len, de->de_str.name, ctx->pos,
30695 +                             (unsigned long)de->de_ino, de->de_type);
30696 +                       if (unlikely(!dir_emit(ctx, de->de_str.name,
30697 +                                              de->de_str.len, de->de_ino,
30698 +                                              de->de_type))) {
30699 +                               /* todo: ignore the error caused by udba? */
30700 +                               /* return err; */
30701 +                               return 0;
30702 +                       }
30703 +
30704 +                       l = calc_size(de->de_str.len);
30705 +                       vdir_cache->vd_last.p.deblk += l;
30706 +                       ctx->pos += l;
30707 +               }
30708 +               if (vdir_cache->vd_last.ul < vdir_cache->vd_nblk - 1) {
30709 +                       vdir_cache->vd_last.ul++;
30710 +                       vdir_cache->vd_last.p.deblk
30711 +                               = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
30712 +                       ctx->pos = deblk_sz * vdir_cache->vd_last.ul;
30713 +                       continue;
30714 +               }
30715 +               break;
30716 +       }
30717 +
30718 +       /* smp_mb(); */
30719 +       return 0;
30720 +}
30721 diff -urN /usr/share/empty/fs/aufs/vfsub.c linux/fs/aufs/vfsub.c
30722 --- /usr/share/empty/fs/aufs/vfsub.c    1970-01-01 01:00:00.000000000 +0100
30723 +++ linux/fs/aufs/vfsub.c       2017-05-06 22:16:52.781553511 +0200
30724 @@ -0,0 +1,899 @@
30725 +/*
30726 + * Copyright (C) 2005-2017 Junjiro R. Okajima
30727 + *
30728 + * This program, aufs is free software; you can redistribute it and/or modify
30729 + * it under the terms of the GNU General Public License as published by
30730 + * the Free Software Foundation; either version 2 of the License, or
30731 + * (at your option) any later version.
30732 + *
30733 + * This program is distributed in the hope that it will be useful,
30734 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30735 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30736 + * GNU General Public License for more details.
30737 + *
30738 + * You should have received a copy of the GNU General Public License
30739 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30740 + */
30741 +
30742 +/*
30743 + * sub-routines for VFS
30744 + */
30745 +
30746 +#include <linux/namei.h>
30747 +#include <linux/nsproxy.h>
30748 +#include <linux/security.h>
30749 +#include <linux/splice.h>
30750 +#ifdef CONFIG_AUFS_BR_FUSE
30751 +#include "../fs/mount.h"
30752 +#endif
30753 +#include "aufs.h"
30754 +
30755 +#ifdef CONFIG_AUFS_BR_FUSE
30756 +int vfsub_test_mntns(struct vfsmount *mnt, struct super_block *h_sb)
30757 +{
30758 +       struct nsproxy *ns;
30759 +
30760 +       if (!au_test_fuse(h_sb) || !au_userns)
30761 +               return 0;
30762 +
30763 +       ns = current->nsproxy;
30764 +       /* no {get,put}_nsproxy(ns) */
30765 +       return real_mount(mnt)->mnt_ns == ns->mnt_ns ? 0 : -EACCES;
30766 +}
30767 +#endif
30768 +
30769 +int vfsub_sync_filesystem(struct super_block *h_sb, int wait)
30770 +{
30771 +       int err;
30772 +
30773 +       lockdep_off();
30774 +       down_read(&h_sb->s_umount);
30775 +       err = __sync_filesystem(h_sb, wait);
30776 +       up_read(&h_sb->s_umount);
30777 +       lockdep_on();
30778 +
30779 +       return err;
30780 +}
30781 +
30782 +/* ---------------------------------------------------------------------- */
30783 +
30784 +int vfsub_update_h_iattr(struct path *h_path, int *did)
30785 +{
30786 +       int err;
30787 +       struct kstat st;
30788 +       struct super_block *h_sb;
30789 +
30790 +       /* for remote fs, leave work for its getattr or d_revalidate */
30791 +       /* for bad i_attr fs, handle them in aufs_getattr() */
30792 +       /* still some fs may acquire i_mutex. we need to skip them */
30793 +       err = 0;
30794 +       if (!did)
30795 +               did = &err;
30796 +       h_sb = h_path->dentry->d_sb;
30797 +       *did = (!au_test_fs_remote(h_sb) && au_test_fs_refresh_iattr(h_sb));
30798 +       if (*did)
30799 +               err = vfsub_getattr(h_path, &st);
30800 +
30801 +       return err;
30802 +}
30803 +
30804 +/* ---------------------------------------------------------------------- */
30805 +
30806 +struct file *vfsub_dentry_open(struct path *path, int flags)
30807 +{
30808 +       struct file *file;
30809 +
30810 +       file = dentry_open(path, flags /* | __FMODE_NONOTIFY */,
30811 +                          current_cred());
30812 +       if (!IS_ERR_OR_NULL(file)
30813 +           && (file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
30814 +               i_readcount_inc(d_inode(path->dentry));
30815 +
30816 +       return file;
30817 +}
30818 +
30819 +struct file *vfsub_filp_open(const char *path, int oflags, int mode)
30820 +{
30821 +       struct file *file;
30822 +
30823 +       lockdep_off();
30824 +       file = filp_open(path,
30825 +                        oflags /* | __FMODE_NONOTIFY */,
30826 +                        mode);
30827 +       lockdep_on();
30828 +       if (IS_ERR(file))
30829 +               goto out;
30830 +       vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30831 +
30832 +out:
30833 +       return file;
30834 +}
30835 +
30836 +/*
30837 + * Ideally this function should call VFS:do_last() in order to keep all its
30838 + * checkings. But it is very hard for aufs to regenerate several VFS internal
30839 + * structure such as nameidata. This is a second (or third) best approach.
30840 + * cf. linux/fs/namei.c:do_last(), lookup_open() and atomic_open().
30841 + */
30842 +int vfsub_atomic_open(struct inode *dir, struct dentry *dentry,
30843 +                     struct vfsub_aopen_args *args, struct au_branch *br)
30844 +{
30845 +       int err;
30846 +       struct file *file = args->file;
30847 +       /* copied from linux/fs/namei.c:atomic_open() */
30848 +       struct dentry *const DENTRY_NOT_SET = (void *)-1UL;
30849 +
30850 +       IMustLock(dir);
30851 +       AuDebugOn(!dir->i_op->atomic_open);
30852 +
30853 +       err = au_br_test_oflag(args->open_flag, br);
30854 +       if (unlikely(err))
30855 +               goto out;
30856 +
30857 +       args->file->f_path.dentry = DENTRY_NOT_SET;
30858 +       args->file->f_path.mnt = au_br_mnt(br);
30859 +       err = dir->i_op->atomic_open(dir, dentry, file, args->open_flag,
30860 +                                    args->create_mode, args->opened);
30861 +       if (err >= 0) {
30862 +               /* some filesystems don't set FILE_CREATED while succeeded? */
30863 +               if (*args->opened & FILE_CREATED)
30864 +                       fsnotify_create(dir, dentry);
30865 +       } else
30866 +               goto out;
30867 +
30868 +
30869 +       if (!err) {
30870 +               /* todo: call VFS:may_open() here */
30871 +               err = open_check_o_direct(file);
30872 +               /* todo: ima_file_check() too? */
30873 +               if (!err && (args->open_flag & __FMODE_EXEC))
30874 +                       err = deny_write_access(file);
30875 +               if (unlikely(err))
30876 +                       /* note that the file is created and still opened */
30877 +                       goto out;
30878 +       }
30879 +
30880 +       au_br_get(br);
30881 +       fsnotify_open(file);
30882 +
30883 +out:
30884 +       return err;
30885 +}
30886 +
30887 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path)
30888 +{
30889 +       int err;
30890 +
30891 +       err = kern_path(name, flags, path);
30892 +       if (!err && d_is_positive(path->dentry))
30893 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
30894 +       return err;
30895 +}
30896 +
30897 +struct dentry *vfsub_lookup_one_len_unlocked(const char *name,
30898 +                                            struct dentry *parent, int len)
30899 +{
30900 +       struct path path = {
30901 +               .mnt = NULL
30902 +       };
30903 +
30904 +       path.dentry = lookup_one_len_unlocked(name, parent, len);
30905 +       if (IS_ERR(path.dentry))
30906 +               goto out;
30907 +       if (d_is_positive(path.dentry))
30908 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
30909 +
30910 +out:
30911 +       AuTraceErrPtr(path.dentry);
30912 +       return path.dentry;
30913 +}
30914 +
30915 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
30916 +                                   int len)
30917 +{
30918 +       struct path path = {
30919 +               .mnt = NULL
30920 +       };
30921 +
30922 +       /* VFS checks it too, but by WARN_ON_ONCE() */
30923 +       IMustLock(d_inode(parent));
30924 +
30925 +       path.dentry = lookup_one_len(name, parent, len);
30926 +       if (IS_ERR(path.dentry))
30927 +               goto out;
30928 +       if (d_is_positive(path.dentry))
30929 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
30930 +
30931 +out:
30932 +       AuTraceErrPtr(path.dentry);
30933 +       return path.dentry;
30934 +}
30935 +
30936 +void vfsub_call_lkup_one(void *args)
30937 +{
30938 +       struct vfsub_lkup_one_args *a = args;
30939 +       *a->errp = vfsub_lkup_one(a->name, a->parent);
30940 +}
30941 +
30942 +/* ---------------------------------------------------------------------- */
30943 +
30944 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
30945 +                                struct dentry *d2, struct au_hinode *hdir2)
30946 +{
30947 +       struct dentry *d;
30948 +
30949 +       lockdep_off();
30950 +       d = lock_rename(d1, d2);
30951 +       lockdep_on();
30952 +       au_hn_suspend(hdir1);
30953 +       if (hdir1 != hdir2)
30954 +               au_hn_suspend(hdir2);
30955 +
30956 +       return d;
30957 +}
30958 +
30959 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
30960 +                        struct dentry *d2, struct au_hinode *hdir2)
30961 +{
30962 +       au_hn_resume(hdir1);
30963 +       if (hdir1 != hdir2)
30964 +               au_hn_resume(hdir2);
30965 +       lockdep_off();
30966 +       unlock_rename(d1, d2);
30967 +       lockdep_on();
30968 +}
30969 +
30970 +/* ---------------------------------------------------------------------- */
30971 +
30972 +int vfsub_create(struct inode *dir, struct path *path, int mode, bool want_excl)
30973 +{
30974 +       int err;
30975 +       struct dentry *d;
30976 +
30977 +       IMustLock(dir);
30978 +
30979 +       d = path->dentry;
30980 +       path->dentry = d->d_parent;
30981 +       err = security_path_mknod(path, d, mode, 0);
30982 +       path->dentry = d;
30983 +       if (unlikely(err))
30984 +               goto out;
30985 +
30986 +       lockdep_off();
30987 +       err = vfs_create(dir, path->dentry, mode, want_excl);
30988 +       lockdep_on();
30989 +       if (!err) {
30990 +               struct path tmp = *path;
30991 +               int did;
30992 +
30993 +               vfsub_update_h_iattr(&tmp, &did);
30994 +               if (did) {
30995 +                       tmp.dentry = path->dentry->d_parent;
30996 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30997 +               }
30998 +               /*ignore*/
30999 +       }
31000 +
31001 +out:
31002 +       return err;
31003 +}
31004 +
31005 +int vfsub_symlink(struct inode *dir, struct path *path, const char *symname)
31006 +{
31007 +       int err;
31008 +       struct dentry *d;
31009 +
31010 +       IMustLock(dir);
31011 +
31012 +       d = path->dentry;
31013 +       path->dentry = d->d_parent;
31014 +       err = security_path_symlink(path, d, symname);
31015 +       path->dentry = d;
31016 +       if (unlikely(err))
31017 +               goto out;
31018 +
31019 +       lockdep_off();
31020 +       err = vfs_symlink(dir, path->dentry, symname);
31021 +       lockdep_on();
31022 +       if (!err) {
31023 +               struct path tmp = *path;
31024 +               int did;
31025 +
31026 +               vfsub_update_h_iattr(&tmp, &did);
31027 +               if (did) {
31028 +                       tmp.dentry = path->dentry->d_parent;
31029 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
31030 +               }
31031 +               /*ignore*/
31032 +       }
31033 +
31034 +out:
31035 +       return err;
31036 +}
31037 +
31038 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev)
31039 +{
31040 +       int err;
31041 +       struct dentry *d;
31042 +
31043 +       IMustLock(dir);
31044 +
31045 +       d = path->dentry;
31046 +       path->dentry = d->d_parent;
31047 +       err = security_path_mknod(path, d, mode, new_encode_dev(dev));
31048 +       path->dentry = d;
31049 +       if (unlikely(err))
31050 +               goto out;
31051 +
31052 +       lockdep_off();
31053 +       err = vfs_mknod(dir, path->dentry, mode, dev);
31054 +       lockdep_on();
31055 +       if (!err) {
31056 +               struct path tmp = *path;
31057 +               int did;
31058 +
31059 +               vfsub_update_h_iattr(&tmp, &did);
31060 +               if (did) {
31061 +                       tmp.dentry = path->dentry->d_parent;
31062 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
31063 +               }
31064 +               /*ignore*/
31065 +       }
31066 +
31067 +out:
31068 +       return err;
31069 +}
31070 +
31071 +static int au_test_nlink(struct inode *inode)
31072 +{
31073 +       const unsigned int link_max = UINT_MAX >> 1; /* rough margin */
31074 +
31075 +       if (!au_test_fs_no_limit_nlink(inode->i_sb)
31076 +           || inode->i_nlink < link_max)
31077 +               return 0;
31078 +       return -EMLINK;
31079 +}
31080 +
31081 +int vfsub_link(struct dentry *src_dentry, struct inode *dir, struct path *path,
31082 +              struct inode **delegated_inode)
31083 +{
31084 +       int err;
31085 +       struct dentry *d;
31086 +
31087 +       IMustLock(dir);
31088 +
31089 +       err = au_test_nlink(d_inode(src_dentry));
31090 +       if (unlikely(err))
31091 +               return err;
31092 +
31093 +       /* we don't call may_linkat() */
31094 +       d = path->dentry;
31095 +       path->dentry = d->d_parent;
31096 +       err = security_path_link(src_dentry, path, d);
31097 +       path->dentry = d;
31098 +       if (unlikely(err))
31099 +               goto out;
31100 +
31101 +       lockdep_off();
31102 +       err = vfs_link(src_dentry, dir, path->dentry, delegated_inode);
31103 +       lockdep_on();
31104 +       if (!err) {
31105 +               struct path tmp = *path;
31106 +               int did;
31107 +
31108 +               /* fuse has different memory inode for the same inumber */
31109 +               vfsub_update_h_iattr(&tmp, &did);
31110 +               if (did) {
31111 +                       tmp.dentry = path->dentry->d_parent;
31112 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
31113 +                       tmp.dentry = src_dentry;
31114 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
31115 +               }
31116 +               /*ignore*/
31117 +       }
31118 +
31119 +out:
31120 +       return err;
31121 +}
31122 +
31123 +int vfsub_rename(struct inode *src_dir, struct dentry *src_dentry,
31124 +                struct inode *dir, struct path *path,
31125 +                struct inode **delegated_inode, unsigned int flags)
31126 +{
31127 +       int err;
31128 +       struct path tmp = {
31129 +               .mnt    = path->mnt
31130 +       };
31131 +       struct dentry *d;
31132 +
31133 +       IMustLock(dir);
31134 +       IMustLock(src_dir);
31135 +
31136 +       d = path->dentry;
31137 +       path->dentry = d->d_parent;
31138 +       tmp.dentry = src_dentry->d_parent;
31139 +       err = security_path_rename(&tmp, src_dentry, path, d, /*flags*/0);
31140 +       path->dentry = d;
31141 +       if (unlikely(err))
31142 +               goto out;
31143 +
31144 +       lockdep_off();
31145 +       err = vfs_rename(src_dir, src_dentry, dir, path->dentry,
31146 +                        delegated_inode, flags);
31147 +       lockdep_on();
31148 +       if (!err) {
31149 +               int did;
31150 +
31151 +               tmp.dentry = d->d_parent;
31152 +               vfsub_update_h_iattr(&tmp, &did);
31153 +               if (did) {
31154 +                       tmp.dentry = src_dentry;
31155 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
31156 +                       tmp.dentry = src_dentry->d_parent;
31157 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
31158 +               }
31159 +               /*ignore*/
31160 +       }
31161 +
31162 +out:
31163 +       return err;
31164 +}
31165 +
31166 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode)
31167 +{
31168 +       int err;
31169 +       struct dentry *d;
31170 +
31171 +       IMustLock(dir);
31172 +
31173 +       d = path->dentry;
31174 +       path->dentry = d->d_parent;
31175 +       err = security_path_mkdir(path, d, mode);
31176 +       path->dentry = d;
31177 +       if (unlikely(err))
31178 +               goto out;
31179 +
31180 +       lockdep_off();
31181 +       err = vfs_mkdir(dir, path->dentry, mode);
31182 +       lockdep_on();
31183 +       if (!err) {
31184 +               struct path tmp = *path;
31185 +               int did;
31186 +
31187 +               vfsub_update_h_iattr(&tmp, &did);
31188 +               if (did) {
31189 +                       tmp.dentry = path->dentry->d_parent;
31190 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
31191 +               }
31192 +               /*ignore*/
31193 +       }
31194 +
31195 +out:
31196 +       return err;
31197 +}
31198 +
31199 +int vfsub_rmdir(struct inode *dir, struct path *path)
31200 +{
31201 +       int err;
31202 +       struct dentry *d;
31203 +
31204 +       IMustLock(dir);
31205 +
31206 +       d = path->dentry;
31207 +       path->dentry = d->d_parent;
31208 +       err = security_path_rmdir(path, d);
31209 +       path->dentry = d;
31210 +       if (unlikely(err))
31211 +               goto out;
31212 +
31213 +       lockdep_off();
31214 +       err = vfs_rmdir(dir, path->dentry);
31215 +       lockdep_on();
31216 +       if (!err) {
31217 +               struct path tmp = {
31218 +                       .dentry = path->dentry->d_parent,
31219 +                       .mnt    = path->mnt
31220 +               };
31221 +
31222 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
31223 +       }
31224 +
31225 +out:
31226 +       return err;
31227 +}
31228 +
31229 +/* ---------------------------------------------------------------------- */
31230 +
31231 +/* todo: support mmap_sem? */
31232 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
31233 +                    loff_t *ppos)
31234 +{
31235 +       ssize_t err;
31236 +
31237 +       lockdep_off();
31238 +       err = vfs_read(file, ubuf, count, ppos);
31239 +       lockdep_on();
31240 +       if (err >= 0)
31241 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
31242 +       return err;
31243 +}
31244 +
31245 +/* todo: kernel_read()? */
31246 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
31247 +                    loff_t *ppos)
31248 +{
31249 +       ssize_t err;
31250 +       mm_segment_t oldfs;
31251 +       union {
31252 +               void *k;
31253 +               char __user *u;
31254 +       } buf;
31255 +
31256 +       buf.k = kbuf;
31257 +       oldfs = get_fs();
31258 +       set_fs(KERNEL_DS);
31259 +       err = vfsub_read_u(file, buf.u, count, ppos);
31260 +       set_fs(oldfs);
31261 +       return err;
31262 +}
31263 +
31264 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
31265 +                     loff_t *ppos)
31266 +{
31267 +       ssize_t err;
31268 +
31269 +       lockdep_off();
31270 +       err = vfs_write(file, ubuf, count, ppos);
31271 +       lockdep_on();
31272 +       if (err >= 0)
31273 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
31274 +       return err;
31275 +}
31276 +
31277 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count, loff_t *ppos)
31278 +{
31279 +       ssize_t err;
31280 +       mm_segment_t oldfs;
31281 +       union {
31282 +               void *k;
31283 +               const char __user *u;
31284 +       } buf;
31285 +
31286 +       buf.k = kbuf;
31287 +       oldfs = get_fs();
31288 +       set_fs(KERNEL_DS);
31289 +       err = vfsub_write_u(file, buf.u, count, ppos);
31290 +       set_fs(oldfs);
31291 +       return err;
31292 +}
31293 +
31294 +int vfsub_flush(struct file *file, fl_owner_t id)
31295 +{
31296 +       int err;
31297 +
31298 +       err = 0;
31299 +       if (file->f_op->flush) {
31300 +               if (!au_test_nfs(file->f_path.dentry->d_sb))
31301 +                       err = file->f_op->flush(file, id);
31302 +               else {
31303 +                       lockdep_off();
31304 +                       err = file->f_op->flush(file, id);
31305 +                       lockdep_on();
31306 +               }
31307 +               if (!err)
31308 +                       vfsub_update_h_iattr(&file->f_path, /*did*/NULL);
31309 +               /*ignore*/
31310 +       }
31311 +       return err;
31312 +}
31313 +
31314 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx)
31315 +{
31316 +       int err;
31317 +
31318 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
31319 +
31320 +       lockdep_off();
31321 +       err = iterate_dir(file, ctx);
31322 +       lockdep_on();
31323 +       if (err >= 0)
31324 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
31325 +       return err;
31326 +}
31327 +
31328 +long vfsub_splice_to(struct file *in, loff_t *ppos,
31329 +                    struct pipe_inode_info *pipe, size_t len,
31330 +                    unsigned int flags)
31331 +{
31332 +       long err;
31333 +
31334 +       lockdep_off();
31335 +       err = do_splice_to(in, ppos, pipe, len, flags);
31336 +       lockdep_on();
31337 +       file_accessed(in);
31338 +       if (err >= 0)
31339 +               vfsub_update_h_iattr(&in->f_path, /*did*/NULL); /*ignore*/
31340 +       return err;
31341 +}
31342 +
31343 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
31344 +                      loff_t *ppos, size_t len, unsigned int flags)
31345 +{
31346 +       long err;
31347 +
31348 +       lockdep_off();
31349 +       err = do_splice_from(pipe, out, ppos, len, flags);
31350 +       lockdep_on();
31351 +       if (err >= 0)
31352 +               vfsub_update_h_iattr(&out->f_path, /*did*/NULL); /*ignore*/
31353 +       return err;
31354 +}
31355 +
31356 +int vfsub_fsync(struct file *file, struct path *path, int datasync)
31357 +{
31358 +       int err;
31359 +
31360 +       /* file can be NULL */
31361 +       lockdep_off();
31362 +       err = vfs_fsync(file, datasync);
31363 +       lockdep_on();
31364 +       if (!err) {
31365 +               if (!path) {
31366 +                       AuDebugOn(!file);
31367 +                       path = &file->f_path;
31368 +               }
31369 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
31370 +       }
31371 +       return err;
31372 +}
31373 +
31374 +/* cf. open.c:do_sys_truncate() and do_sys_ftruncate() */
31375 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
31376 +               struct file *h_file)
31377 +{
31378 +       int err;
31379 +       struct inode *h_inode;
31380 +       struct super_block *h_sb;
31381 +
31382 +       if (!h_file) {
31383 +               err = vfsub_truncate(h_path, length);
31384 +               goto out;
31385 +       }
31386 +
31387 +       h_inode = d_inode(h_path->dentry);
31388 +       h_sb = h_inode->i_sb;
31389 +       lockdep_off();
31390 +       sb_start_write(h_sb);
31391 +       lockdep_on();
31392 +       err = locks_verify_truncate(h_inode, h_file, length);
31393 +       if (!err)
31394 +               err = security_path_truncate(h_path);
31395 +       if (!err) {
31396 +               lockdep_off();
31397 +               err = do_truncate(h_path->dentry, length, attr, h_file);
31398 +               lockdep_on();
31399 +       }
31400 +       lockdep_off();
31401 +       sb_end_write(h_sb);
31402 +       lockdep_on();
31403 +
31404 +out:
31405 +       return err;
31406 +}
31407 +
31408 +/* ---------------------------------------------------------------------- */
31409 +
31410 +struct au_vfsub_mkdir_args {
31411 +       int *errp;
31412 +       struct inode *dir;
31413 +       struct path *path;
31414 +       int mode;
31415 +};
31416 +
31417 +static void au_call_vfsub_mkdir(void *args)
31418 +{
31419 +       struct au_vfsub_mkdir_args *a = args;
31420 +       *a->errp = vfsub_mkdir(a->dir, a->path, a->mode);
31421 +}
31422 +
31423 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode)
31424 +{
31425 +       int err, do_sio, wkq_err;
31426 +
31427 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
31428 +       if (!do_sio) {
31429 +               lockdep_off();
31430 +               err = vfsub_mkdir(dir, path, mode);
31431 +               lockdep_on();
31432 +       } else {
31433 +               struct au_vfsub_mkdir_args args = {
31434 +                       .errp   = &err,
31435 +                       .dir    = dir,
31436 +                       .path   = path,
31437 +                       .mode   = mode
31438 +               };
31439 +               wkq_err = au_wkq_wait(au_call_vfsub_mkdir, &args);
31440 +               if (unlikely(wkq_err))
31441 +                       err = wkq_err;
31442 +       }
31443 +
31444 +       return err;
31445 +}
31446 +
31447 +struct au_vfsub_rmdir_args {
31448 +       int *errp;
31449 +       struct inode *dir;
31450 +       struct path *path;
31451 +};
31452 +
31453 +static void au_call_vfsub_rmdir(void *args)
31454 +{
31455 +       struct au_vfsub_rmdir_args *a = args;
31456 +       *a->errp = vfsub_rmdir(a->dir, a->path);
31457 +}
31458 +
31459 +int vfsub_sio_rmdir(struct inode *dir, struct path *path)
31460 +{
31461 +       int err, do_sio, wkq_err;
31462 +
31463 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
31464 +       if (!do_sio) {
31465 +               lockdep_off();
31466 +               err = vfsub_rmdir(dir, path);
31467 +               lockdep_on();
31468 +       } else {
31469 +               struct au_vfsub_rmdir_args args = {
31470 +                       .errp   = &err,
31471 +                       .dir    = dir,
31472 +                       .path   = path
31473 +               };
31474 +               wkq_err = au_wkq_wait(au_call_vfsub_rmdir, &args);
31475 +               if (unlikely(wkq_err))
31476 +                       err = wkq_err;
31477 +       }
31478 +
31479 +       return err;
31480 +}
31481 +
31482 +/* ---------------------------------------------------------------------- */
31483 +
31484 +struct notify_change_args {
31485 +       int *errp;
31486 +       struct path *path;
31487 +       struct iattr *ia;
31488 +       struct inode **delegated_inode;
31489 +};
31490 +
31491 +static void call_notify_change(void *args)
31492 +{
31493 +       struct notify_change_args *a = args;
31494 +       struct inode *h_inode;
31495 +
31496 +       h_inode = d_inode(a->path->dentry);
31497 +       IMustLock(h_inode);
31498 +
31499 +       *a->errp = -EPERM;
31500 +       if (!IS_IMMUTABLE(h_inode) && !IS_APPEND(h_inode)) {
31501 +               lockdep_off();
31502 +               *a->errp = notify_change(a->path->dentry, a->ia,
31503 +                                        a->delegated_inode);
31504 +               lockdep_on();
31505 +               if (!*a->errp)
31506 +                       vfsub_update_h_iattr(a->path, /*did*/NULL); /*ignore*/
31507 +       }
31508 +       AuTraceErr(*a->errp);
31509 +}
31510 +
31511 +int vfsub_notify_change(struct path *path, struct iattr *ia,
31512 +                       struct inode **delegated_inode)
31513 +{
31514 +       int err;
31515 +       struct notify_change_args args = {
31516 +               .errp                   = &err,
31517 +               .path                   = path,
31518 +               .ia                     = ia,
31519 +               .delegated_inode        = delegated_inode
31520 +       };
31521 +
31522 +       call_notify_change(&args);
31523 +
31524 +       return err;
31525 +}
31526 +
31527 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
31528 +                           struct inode **delegated_inode)
31529 +{
31530 +       int err, wkq_err;
31531 +       struct notify_change_args args = {
31532 +               .errp                   = &err,
31533 +               .path                   = path,
31534 +               .ia                     = ia,
31535 +               .delegated_inode        = delegated_inode
31536 +       };
31537 +
31538 +       wkq_err = au_wkq_wait(call_notify_change, &args);
31539 +       if (unlikely(wkq_err))
31540 +               err = wkq_err;
31541 +
31542 +       return err;
31543 +}
31544 +
31545 +/* ---------------------------------------------------------------------- */
31546 +
31547 +struct unlink_args {
31548 +       int *errp;
31549 +       struct inode *dir;
31550 +       struct path *path;
31551 +       struct inode **delegated_inode;
31552 +};
31553 +
31554 +static void call_unlink(void *args)
31555 +{
31556 +       struct unlink_args *a = args;
31557 +       struct dentry *d = a->path->dentry;
31558 +       struct inode *h_inode;
31559 +       const int stop_sillyrename = (au_test_nfs(d->d_sb)
31560 +                                     && au_dcount(d) == 1);
31561 +
31562 +       IMustLock(a->dir);
31563 +
31564 +       a->path->dentry = d->d_parent;
31565 +       *a->errp = security_path_unlink(a->path, d);
31566 +       a->path->dentry = d;
31567 +       if (unlikely(*a->errp))
31568 +               return;
31569 +
31570 +       if (!stop_sillyrename)
31571 +               dget(d);
31572 +       h_inode = NULL;
31573 +       if (d_is_positive(d)) {
31574 +               h_inode = d_inode(d);
31575 +               ihold(h_inode);
31576 +       }
31577 +
31578 +       lockdep_off();
31579 +       *a->errp = vfs_unlink(a->dir, d, a->delegated_inode);
31580 +       lockdep_on();
31581 +       if (!*a->errp) {
31582 +               struct path tmp = {
31583 +                       .dentry = d->d_parent,
31584 +                       .mnt    = a->path->mnt
31585 +               };
31586 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
31587 +       }
31588 +
31589 +       if (!stop_sillyrename)
31590 +               dput(d);
31591 +       if (h_inode)
31592 +               iput(h_inode);
31593 +
31594 +       AuTraceErr(*a->errp);
31595 +}
31596 +
31597 +/*
31598 + * @dir: must be locked.
31599 + * @dentry: target dentry.
31600 + */
31601 +int vfsub_unlink(struct inode *dir, struct path *path,
31602 +                struct inode **delegated_inode, int force)
31603 +{
31604 +       int err;
31605 +       struct unlink_args args = {
31606 +               .errp                   = &err,
31607 +               .dir                    = dir,
31608 +               .path                   = path,
31609 +               .delegated_inode        = delegated_inode
31610 +       };
31611 +
31612 +       if (!force)
31613 +               call_unlink(&args);
31614 +       else {
31615 +               int wkq_err;
31616 +
31617 +               wkq_err = au_wkq_wait(call_unlink, &args);
31618 +               if (unlikely(wkq_err))
31619 +                       err = wkq_err;
31620 +       }
31621 +
31622 +       return err;
31623 +}
31624 diff -urN /usr/share/empty/fs/aufs/vfsub.h linux/fs/aufs/vfsub.h
31625 --- /usr/share/empty/fs/aufs/vfsub.h    1970-01-01 01:00:00.000000000 +0100
31626 +++ linux/fs/aufs/vfsub.h       2017-05-06 22:16:52.781553511 +0200
31627 @@ -0,0 +1,353 @@
31628 +/*
31629 + * Copyright (C) 2005-2017 Junjiro R. Okajima
31630 + *
31631 + * This program, aufs is free software; you can redistribute it and/or modify
31632 + * it under the terms of the GNU General Public License as published by
31633 + * the Free Software Foundation; either version 2 of the License, or
31634 + * (at your option) any later version.
31635 + *
31636 + * This program is distributed in the hope that it will be useful,
31637 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31638 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31639 + * GNU General Public License for more details.
31640 + *
31641 + * You should have received a copy of the GNU General Public License
31642 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31643 + */
31644 +
31645 +/*
31646 + * sub-routines for VFS
31647 + */
31648 +
31649 +#ifndef __AUFS_VFSUB_H__
31650 +#define __AUFS_VFSUB_H__
31651 +
31652 +#ifdef __KERNEL__
31653 +
31654 +#include <linux/fs.h>
31655 +#include <linux/mount.h>
31656 +#include <linux/posix_acl.h>
31657 +#include <linux/xattr.h>
31658 +#include "debug.h"
31659 +
31660 +/* copied from linux/fs/internal.h */
31661 +/* todo: BAD approach!! */
31662 +extern void __mnt_drop_write(struct vfsmount *);
31663 +extern int open_check_o_direct(struct file *f);
31664 +
31665 +/* ---------------------------------------------------------------------- */
31666 +
31667 +/* lock subclass for lower inode */
31668 +/* default MAX_LOCKDEP_SUBCLASSES(8) is not enough */
31669 +/* reduce? gave up. */
31670 +enum {
31671 +       AuLsc_I_Begin = I_MUTEX_PARENT2, /* 5 */
31672 +       AuLsc_I_PARENT,         /* lower inode, parent first */
31673 +       AuLsc_I_PARENT2,        /* copyup dirs */
31674 +       AuLsc_I_PARENT3,        /* copyup wh */
31675 +       AuLsc_I_CHILD,
31676 +       AuLsc_I_CHILD2,
31677 +       AuLsc_I_End
31678 +};
31679 +
31680 +/* to debug easier, do not make them inlined functions */
31681 +#define MtxMustLock(mtx)       AuDebugOn(!mutex_is_locked(mtx))
31682 +#define IMustLock(i)           AuDebugOn(!inode_is_locked(i))
31683 +
31684 +/* ---------------------------------------------------------------------- */
31685 +
31686 +static inline void vfsub_drop_nlink(struct inode *inode)
31687 +{
31688 +       AuDebugOn(!inode->i_nlink);
31689 +       drop_nlink(inode);
31690 +}
31691 +
31692 +static inline void vfsub_dead_dir(struct inode *inode)
31693 +{
31694 +       AuDebugOn(!S_ISDIR(inode->i_mode));
31695 +       inode->i_flags |= S_DEAD;
31696 +       clear_nlink(inode);
31697 +}
31698 +
31699 +static inline int vfsub_native_ro(struct inode *inode)
31700 +{
31701 +       return (inode->i_sb->s_flags & MS_RDONLY)
31702 +               || IS_RDONLY(inode)
31703 +               /* || IS_APPEND(inode) */
31704 +               || IS_IMMUTABLE(inode);
31705 +}
31706 +
31707 +#ifdef CONFIG_AUFS_BR_FUSE
31708 +int vfsub_test_mntns(struct vfsmount *mnt, struct super_block *h_sb);
31709 +#else
31710 +AuStubInt0(vfsub_test_mntns, struct vfsmount *mnt, struct super_block *h_sb);
31711 +#endif
31712 +
31713 +int vfsub_sync_filesystem(struct super_block *h_sb, int wait);
31714 +
31715 +/* ---------------------------------------------------------------------- */
31716 +
31717 +int vfsub_update_h_iattr(struct path *h_path, int *did);
31718 +struct file *vfsub_dentry_open(struct path *path, int flags);
31719 +struct file *vfsub_filp_open(const char *path, int oflags, int mode);
31720 +struct vfsub_aopen_args {
31721 +       struct file     *file;
31722 +       unsigned int    open_flag;
31723 +       umode_t         create_mode;
31724 +       int             *opened;
31725 +};
31726 +struct au_branch;
31727 +int vfsub_atomic_open(struct inode *dir, struct dentry *dentry,
31728 +                     struct vfsub_aopen_args *args, struct au_branch *br);
31729 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path);
31730 +
31731 +struct dentry *vfsub_lookup_one_len_unlocked(const char *name,
31732 +                                            struct dentry *parent, int len);
31733 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
31734 +                                   int len);
31735 +
31736 +struct vfsub_lkup_one_args {
31737 +       struct dentry **errp;
31738 +       struct qstr *name;
31739 +       struct dentry *parent;
31740 +};
31741 +
31742 +static inline struct dentry *vfsub_lkup_one(struct qstr *name,
31743 +                                           struct dentry *parent)
31744 +{
31745 +       return vfsub_lookup_one_len(name->name, parent, name->len);
31746 +}
31747 +
31748 +void vfsub_call_lkup_one(void *args);
31749 +
31750 +/* ---------------------------------------------------------------------- */
31751 +
31752 +static inline int vfsub_mnt_want_write(struct vfsmount *mnt)
31753 +{
31754 +       int err;
31755 +
31756 +       lockdep_off();
31757 +       err = mnt_want_write(mnt);
31758 +       lockdep_on();
31759 +       return err;
31760 +}
31761 +
31762 +static inline void vfsub_mnt_drop_write(struct vfsmount *mnt)
31763 +{
31764 +       lockdep_off();
31765 +       mnt_drop_write(mnt);
31766 +       lockdep_on();
31767 +}
31768 +
31769 +#if 0 /* reserved */
31770 +static inline void vfsub_mnt_drop_write_file(struct file *file)
31771 +{
31772 +       lockdep_off();
31773 +       mnt_drop_write_file(file);
31774 +       lockdep_on();
31775 +}
31776 +#endif
31777 +
31778 +/* ---------------------------------------------------------------------- */
31779 +
31780 +struct au_hinode;
31781 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
31782 +                                struct dentry *d2, struct au_hinode *hdir2);
31783 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
31784 +                        struct dentry *d2, struct au_hinode *hdir2);
31785 +
31786 +int vfsub_create(struct inode *dir, struct path *path, int mode,
31787 +                bool want_excl);
31788 +int vfsub_symlink(struct inode *dir, struct path *path,
31789 +                 const char *symname);
31790 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev);
31791 +int vfsub_link(struct dentry *src_dentry, struct inode *dir,
31792 +              struct path *path, struct inode **delegated_inode);
31793 +int vfsub_rename(struct inode *src_hdir, struct dentry *src_dentry,
31794 +                struct inode *hdir, struct path *path,
31795 +                struct inode **delegated_inode, unsigned int flags);
31796 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode);
31797 +int vfsub_rmdir(struct inode *dir, struct path *path);
31798 +
31799 +/* ---------------------------------------------------------------------- */
31800 +
31801 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
31802 +                    loff_t *ppos);
31803 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
31804 +                       loff_t *ppos);
31805 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
31806 +                     loff_t *ppos);
31807 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count,
31808 +                     loff_t *ppos);
31809 +int vfsub_flush(struct file *file, fl_owner_t id);
31810 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx);
31811 +
31812 +static inline loff_t vfsub_f_size_read(struct file *file)
31813 +{
31814 +       return i_size_read(file_inode(file));
31815 +}
31816 +
31817 +static inline unsigned int vfsub_file_flags(struct file *file)
31818 +{
31819 +       unsigned int flags;
31820 +
31821 +       spin_lock(&file->f_lock);
31822 +       flags = file->f_flags;
31823 +       spin_unlock(&file->f_lock);
31824 +
31825 +       return flags;
31826 +}
31827 +
31828 +static inline int vfsub_file_execed(struct file *file)
31829 +{
31830 +       /* todo: direct access f_flags */
31831 +       return !!(vfsub_file_flags(file) & __FMODE_EXEC);
31832 +}
31833 +
31834 +#if 0 /* reserved */
31835 +static inline void vfsub_file_accessed(struct file *h_file)
31836 +{
31837 +       file_accessed(h_file);
31838 +       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL); /*ignore*/
31839 +}
31840 +#endif
31841 +
31842 +#if 0 /* reserved */
31843 +static inline void vfsub_touch_atime(struct vfsmount *h_mnt,
31844 +                                    struct dentry *h_dentry)
31845 +{
31846 +       struct path h_path = {
31847 +               .dentry = h_dentry,
31848 +               .mnt    = h_mnt
31849 +       };
31850 +       touch_atime(&h_path);
31851 +       vfsub_update_h_iattr(&h_path, /*did*/NULL); /*ignore*/
31852 +}
31853 +#endif
31854 +
31855 +static inline int vfsub_update_time(struct inode *h_inode, struct timespec *ts,
31856 +                                   int flags)
31857 +{
31858 +       return update_time(h_inode, ts, flags);
31859 +       /* no vfsub_update_h_iattr() since we don't have struct path */
31860 +}
31861 +
31862 +#ifdef CONFIG_FS_POSIX_ACL
31863 +static inline int vfsub_acl_chmod(struct inode *h_inode, umode_t h_mode)
31864 +{
31865 +       int err;
31866 +
31867 +       err = posix_acl_chmod(h_inode, h_mode);
31868 +       if (err == -EOPNOTSUPP)
31869 +               err = 0;
31870 +       return err;
31871 +}
31872 +#else
31873 +AuStubInt0(vfsub_acl_chmod, struct inode *h_inode, umode_t h_mode);
31874 +#endif
31875 +
31876 +long vfsub_splice_to(struct file *in, loff_t *ppos,
31877 +                    struct pipe_inode_info *pipe, size_t len,
31878 +                    unsigned int flags);
31879 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
31880 +                      loff_t *ppos, size_t len, unsigned int flags);
31881 +
31882 +static inline long vfsub_truncate(struct path *path, loff_t length)
31883 +{
31884 +       long err;
31885 +
31886 +       lockdep_off();
31887 +       err = vfs_truncate(path, length);
31888 +       lockdep_on();
31889 +       return err;
31890 +}
31891 +
31892 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
31893 +               struct file *h_file);
31894 +int vfsub_fsync(struct file *file, struct path *path, int datasync);
31895 +
31896 +/*
31897 + * re-use branch fs's ioctl(FICLONE) while aufs itself doesn't support such
31898 + * ioctl.
31899 + */
31900 +static inline int vfsub_clone_file_range(struct file *src, struct file *dst,
31901 +                                        u64 len)
31902 +{
31903 +       int err;
31904 +
31905 +       lockdep_off();
31906 +       err = vfs_clone_file_range(src, 0, dst, 0, len);
31907 +       lockdep_on();
31908 +
31909 +       return err;
31910 +}
31911 +
31912 +/* copy_file_range(2) is a systemcall */
31913 +static inline ssize_t vfsub_copy_file_range(struct file *src, loff_t src_pos,
31914 +                                           struct file *dst, loff_t dst_pos,
31915 +                                           size_t len, unsigned int flags)
31916 +{
31917 +       ssize_t ssz;
31918 +
31919 +       lockdep_off();
31920 +       ssz = vfs_copy_file_range(src, src_pos, dst, dst_pos, len, flags);
31921 +       lockdep_on();
31922 +
31923 +       return ssz;
31924 +}
31925 +
31926 +/* ---------------------------------------------------------------------- */
31927 +
31928 +static inline loff_t vfsub_llseek(struct file *file, loff_t offset, int origin)
31929 +{
31930 +       loff_t err;
31931 +
31932 +       lockdep_off();
31933 +       err = vfs_llseek(file, offset, origin);
31934 +       lockdep_on();
31935 +       return err;
31936 +}
31937 +
31938 +/* ---------------------------------------------------------------------- */
31939 +
31940 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode);
31941 +int vfsub_sio_rmdir(struct inode *dir, struct path *path);
31942 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
31943 +                           struct inode **delegated_inode);
31944 +int vfsub_notify_change(struct path *path, struct iattr *ia,
31945 +                       struct inode **delegated_inode);
31946 +int vfsub_unlink(struct inode *dir, struct path *path,
31947 +                struct inode **delegated_inode, int force);
31948 +
31949 +static inline int vfsub_getattr(const struct path *path, struct kstat *st)
31950 +{
31951 +       return vfs_getattr(path, st, STATX_BASIC_STATS, AT_STATX_SYNC_AS_STAT);
31952 +}
31953 +
31954 +/* ---------------------------------------------------------------------- */
31955 +
31956 +static inline int vfsub_setxattr(struct dentry *dentry, const char *name,
31957 +                                const void *value, size_t size, int flags)
31958 +{
31959 +       int err;
31960 +
31961 +       lockdep_off();
31962 +       err = vfs_setxattr(dentry, name, value, size, flags);
31963 +       lockdep_on();
31964 +
31965 +       return err;
31966 +}
31967 +
31968 +static inline int vfsub_removexattr(struct dentry *dentry, const char *name)
31969 +{
31970 +       int err;
31971 +
31972 +       lockdep_off();
31973 +       err = vfs_removexattr(dentry, name);
31974 +       lockdep_on();
31975 +
31976 +       return err;
31977 +}
31978 +
31979 +#endif /* __KERNEL__ */
31980 +#endif /* __AUFS_VFSUB_H__ */
31981 diff -urN /usr/share/empty/fs/aufs/wbr_policy.c linux/fs/aufs/wbr_policy.c
31982 --- /usr/share/empty/fs/aufs/wbr_policy.c       1970-01-01 01:00:00.000000000 +0100
31983 +++ linux/fs/aufs/wbr_policy.c  2017-05-06 22:16:52.784886929 +0200
31984 @@ -0,0 +1,830 @@
31985 +/*
31986 + * Copyright (C) 2005-2017 Junjiro R. Okajima
31987 + *
31988 + * This program, aufs is free software; you can redistribute it and/or modify
31989 + * it under the terms of the GNU General Public License as published by
31990 + * the Free Software Foundation; either version 2 of the License, or
31991 + * (at your option) any later version.
31992 + *
31993 + * This program is distributed in the hope that it will be useful,
31994 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31995 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31996 + * GNU General Public License for more details.
31997 + *
31998 + * You should have received a copy of the GNU General Public License
31999 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32000 + */
32001 +
32002 +/*
32003 + * policies for selecting one among multiple writable branches
32004 + */
32005 +
32006 +#include <linux/statfs.h>
32007 +#include "aufs.h"
32008 +
32009 +/* subset of cpup_attr() */
32010 +static noinline_for_stack
32011 +int au_cpdown_attr(struct path *h_path, struct dentry *h_src)
32012 +{
32013 +       int err, sbits;
32014 +       struct iattr ia;
32015 +       struct inode *h_isrc;
32016 +
32017 +       h_isrc = d_inode(h_src);
32018 +       ia.ia_valid = ATTR_FORCE | ATTR_MODE | ATTR_UID | ATTR_GID;
32019 +       ia.ia_mode = h_isrc->i_mode;
32020 +       ia.ia_uid = h_isrc->i_uid;
32021 +       ia.ia_gid = h_isrc->i_gid;
32022 +       sbits = !!(ia.ia_mode & (S_ISUID | S_ISGID));
32023 +       au_cpup_attr_flags(d_inode(h_path->dentry), h_isrc->i_flags);
32024 +       /* no delegation since it is just created */
32025 +       err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
32026 +
32027 +       /* is this nfs only? */
32028 +       if (!err && sbits && au_test_nfs(h_path->dentry->d_sb)) {
32029 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
32030 +               ia.ia_mode = h_isrc->i_mode;
32031 +               err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
32032 +       }
32033 +
32034 +       return err;
32035 +}
32036 +
32037 +#define AuCpdown_PARENT_OPQ    1
32038 +#define AuCpdown_WHED          (1 << 1)
32039 +#define AuCpdown_MADE_DIR      (1 << 2)
32040 +#define AuCpdown_DIROPQ                (1 << 3)
32041 +#define au_ftest_cpdown(flags, name)   ((flags) & AuCpdown_##name)
32042 +#define au_fset_cpdown(flags, name) \
32043 +       do { (flags) |= AuCpdown_##name; } while (0)
32044 +#define au_fclr_cpdown(flags, name) \
32045 +       do { (flags) &= ~AuCpdown_##name; } while (0)
32046 +
32047 +static int au_cpdown_dir_opq(struct dentry *dentry, aufs_bindex_t bdst,
32048 +                            unsigned int *flags)
32049 +{
32050 +       int err;
32051 +       struct dentry *opq_dentry;
32052 +
32053 +       opq_dentry = au_diropq_create(dentry, bdst);
32054 +       err = PTR_ERR(opq_dentry);
32055 +       if (IS_ERR(opq_dentry))
32056 +               goto out;
32057 +       dput(opq_dentry);
32058 +       au_fset_cpdown(*flags, DIROPQ);
32059 +
32060 +out:
32061 +       return err;
32062 +}
32063 +
32064 +static int au_cpdown_dir_wh(struct dentry *dentry, struct dentry *h_parent,
32065 +                           struct inode *dir, aufs_bindex_t bdst)
32066 +{
32067 +       int err;
32068 +       struct path h_path;
32069 +       struct au_branch *br;
32070 +
32071 +       br = au_sbr(dentry->d_sb, bdst);
32072 +       h_path.dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
32073 +       err = PTR_ERR(h_path.dentry);
32074 +       if (IS_ERR(h_path.dentry))
32075 +               goto out;
32076 +
32077 +       err = 0;
32078 +       if (d_is_positive(h_path.dentry)) {
32079 +               h_path.mnt = au_br_mnt(br);
32080 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bdst), &h_path,
32081 +                                         dentry);
32082 +       }
32083 +       dput(h_path.dentry);
32084 +
32085 +out:
32086 +       return err;
32087 +}
32088 +
32089 +static int au_cpdown_dir(struct dentry *dentry, aufs_bindex_t bdst,
32090 +                        struct au_pin *pin,
32091 +                        struct dentry *h_parent, void *arg)
32092 +{
32093 +       int err, rerr;
32094 +       aufs_bindex_t bopq, btop;
32095 +       struct path h_path;
32096 +       struct dentry *parent;
32097 +       struct inode *h_dir, *h_inode, *inode, *dir;
32098 +       unsigned int *flags = arg;
32099 +
32100 +       btop = au_dbtop(dentry);
32101 +       /* dentry is di-locked */
32102 +       parent = dget_parent(dentry);
32103 +       dir = d_inode(parent);
32104 +       h_dir = d_inode(h_parent);
32105 +       AuDebugOn(h_dir != au_h_iptr(dir, bdst));
32106 +       IMustLock(h_dir);
32107 +
32108 +       err = au_lkup_neg(dentry, bdst, /*wh*/0);
32109 +       if (unlikely(err < 0))
32110 +               goto out;
32111 +       h_path.dentry = au_h_dptr(dentry, bdst);
32112 +       h_path.mnt = au_sbr_mnt(dentry->d_sb, bdst);
32113 +       err = vfsub_sio_mkdir(au_h_iptr(dir, bdst), &h_path,
32114 +                             S_IRWXU | S_IRUGO | S_IXUGO);
32115 +       if (unlikely(err))
32116 +               goto out_put;
32117 +       au_fset_cpdown(*flags, MADE_DIR);
32118 +
32119 +       bopq = au_dbdiropq(dentry);
32120 +       au_fclr_cpdown(*flags, WHED);
32121 +       au_fclr_cpdown(*flags, DIROPQ);
32122 +       if (au_dbwh(dentry) == bdst)
32123 +               au_fset_cpdown(*flags, WHED);
32124 +       if (!au_ftest_cpdown(*flags, PARENT_OPQ) && bopq <= bdst)
32125 +               au_fset_cpdown(*flags, PARENT_OPQ);
32126 +       h_inode = d_inode(h_path.dentry);
32127 +       inode_lock_nested(h_inode, AuLsc_I_CHILD);
32128 +       if (au_ftest_cpdown(*flags, WHED)) {
32129 +               err = au_cpdown_dir_opq(dentry, bdst, flags);
32130 +               if (unlikely(err)) {
32131 +                       inode_unlock(h_inode);
32132 +                       goto out_dir;
32133 +               }
32134 +       }
32135 +
32136 +       err = au_cpdown_attr(&h_path, au_h_dptr(dentry, btop));
32137 +       inode_unlock(h_inode);
32138 +       if (unlikely(err))
32139 +               goto out_opq;
32140 +
32141 +       if (au_ftest_cpdown(*flags, WHED)) {
32142 +               err = au_cpdown_dir_wh(dentry, h_parent, dir, bdst);
32143 +               if (unlikely(err))
32144 +                       goto out_opq;
32145 +       }
32146 +
32147 +       inode = d_inode(dentry);
32148 +       if (au_ibbot(inode) < bdst)
32149 +               au_set_ibbot(inode, bdst);
32150 +       au_set_h_iptr(inode, bdst, au_igrab(h_inode),
32151 +                     au_hi_flags(inode, /*isdir*/1));
32152 +       au_fhsm_wrote(dentry->d_sb, bdst, /*force*/0);
32153 +       goto out; /* success */
32154 +
32155 +       /* revert */
32156 +out_opq:
32157 +       if (au_ftest_cpdown(*flags, DIROPQ)) {
32158 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
32159 +               rerr = au_diropq_remove(dentry, bdst);
32160 +               inode_unlock(h_inode);
32161 +               if (unlikely(rerr)) {
32162 +                       AuIOErr("failed removing diropq for %pd b%d (%d)\n",
32163 +                               dentry, bdst, rerr);
32164 +                       err = -EIO;
32165 +                       goto out;
32166 +               }
32167 +       }
32168 +out_dir:
32169 +       if (au_ftest_cpdown(*flags, MADE_DIR)) {
32170 +               rerr = vfsub_sio_rmdir(au_h_iptr(dir, bdst), &h_path);
32171 +               if (unlikely(rerr)) {
32172 +                       AuIOErr("failed removing %pd b%d (%d)\n",
32173 +                               dentry, bdst, rerr);
32174 +                       err = -EIO;
32175 +               }
32176 +       }
32177 +out_put:
32178 +       au_set_h_dptr(dentry, bdst, NULL);
32179 +       if (au_dbbot(dentry) == bdst)
32180 +               au_update_dbbot(dentry);
32181 +out:
32182 +       dput(parent);
32183 +       return err;
32184 +}
32185 +
32186 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst)
32187 +{
32188 +       int err;
32189 +       unsigned int flags;
32190 +
32191 +       flags = 0;
32192 +       err = au_cp_dirs(dentry, bdst, au_cpdown_dir, &flags);
32193 +
32194 +       return err;
32195 +}
32196 +
32197 +/* ---------------------------------------------------------------------- */
32198 +
32199 +/* policies for create */
32200 +
32201 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex)
32202 +{
32203 +       int err, i, j, ndentry;
32204 +       aufs_bindex_t bopq;
32205 +       struct au_dcsub_pages dpages;
32206 +       struct au_dpage *dpage;
32207 +       struct dentry **dentries, *parent, *d;
32208 +
32209 +       err = au_dpages_init(&dpages, GFP_NOFS);
32210 +       if (unlikely(err))
32211 +               goto out;
32212 +       parent = dget_parent(dentry);
32213 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/0);
32214 +       if (unlikely(err))
32215 +               goto out_free;
32216 +
32217 +       err = bindex;
32218 +       for (i = 0; i < dpages.ndpage; i++) {
32219 +               dpage = dpages.dpages + i;
32220 +               dentries = dpage->dentries;
32221 +               ndentry = dpage->ndentry;
32222 +               for (j = 0; j < ndentry; j++) {
32223 +                       d = dentries[j];
32224 +                       di_read_lock_parent2(d, !AuLock_IR);
32225 +                       bopq = au_dbdiropq(d);
32226 +                       di_read_unlock(d, !AuLock_IR);
32227 +                       if (bopq >= 0 && bopq < err)
32228 +                               err = bopq;
32229 +               }
32230 +       }
32231 +
32232 +out_free:
32233 +       dput(parent);
32234 +       au_dpages_free(&dpages);
32235 +out:
32236 +       return err;
32237 +}
32238 +
32239 +static int au_wbr_bu(struct super_block *sb, aufs_bindex_t bindex)
32240 +{
32241 +       for (; bindex >= 0; bindex--)
32242 +               if (!au_br_rdonly(au_sbr(sb, bindex)))
32243 +                       return bindex;
32244 +       return -EROFS;
32245 +}
32246 +
32247 +/* top down parent */
32248 +static int au_wbr_create_tdp(struct dentry *dentry,
32249 +                            unsigned int flags __maybe_unused)
32250 +{
32251 +       int err;
32252 +       aufs_bindex_t btop, bindex;
32253 +       struct super_block *sb;
32254 +       struct dentry *parent, *h_parent;
32255 +
32256 +       sb = dentry->d_sb;
32257 +       btop = au_dbtop(dentry);
32258 +       err = btop;
32259 +       if (!au_br_rdonly(au_sbr(sb, btop)))
32260 +               goto out;
32261 +
32262 +       err = -EROFS;
32263 +       parent = dget_parent(dentry);
32264 +       for (bindex = au_dbtop(parent); bindex < btop; bindex++) {
32265 +               h_parent = au_h_dptr(parent, bindex);
32266 +               if (!h_parent || d_is_negative(h_parent))
32267 +                       continue;
32268 +
32269 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
32270 +                       err = bindex;
32271 +                       break;
32272 +               }
32273 +       }
32274 +       dput(parent);
32275 +
32276 +       /* bottom up here */
32277 +       if (unlikely(err < 0)) {
32278 +               err = au_wbr_bu(sb, btop - 1);
32279 +               if (err >= 0)
32280 +                       err = au_wbr_nonopq(dentry, err);
32281 +       }
32282 +
32283 +out:
32284 +       AuDbg("b%d\n", err);
32285 +       return err;
32286 +}
32287 +
32288 +/* ---------------------------------------------------------------------- */
32289 +
32290 +/* an exception for the policy other than tdp */
32291 +static int au_wbr_create_exp(struct dentry *dentry)
32292 +{
32293 +       int err;
32294 +       aufs_bindex_t bwh, bdiropq;
32295 +       struct dentry *parent;
32296 +
32297 +       err = -1;
32298 +       bwh = au_dbwh(dentry);
32299 +       parent = dget_parent(dentry);
32300 +       bdiropq = au_dbdiropq(parent);
32301 +       if (bwh >= 0) {
32302 +               if (bdiropq >= 0)
32303 +                       err = min(bdiropq, bwh);
32304 +               else
32305 +                       err = bwh;
32306 +               AuDbg("%d\n", err);
32307 +       } else if (bdiropq >= 0) {
32308 +               err = bdiropq;
32309 +               AuDbg("%d\n", err);
32310 +       }
32311 +       dput(parent);
32312 +
32313 +       if (err >= 0)
32314 +               err = au_wbr_nonopq(dentry, err);
32315 +
32316 +       if (err >= 0 && au_br_rdonly(au_sbr(dentry->d_sb, err)))
32317 +               err = -1;
32318 +
32319 +       AuDbg("%d\n", err);
32320 +       return err;
32321 +}
32322 +
32323 +/* ---------------------------------------------------------------------- */
32324 +
32325 +/* round robin */
32326 +static int au_wbr_create_init_rr(struct super_block *sb)
32327 +{
32328 +       int err;
32329 +
32330 +       err = au_wbr_bu(sb, au_sbbot(sb));
32331 +       atomic_set(&au_sbi(sb)->si_wbr_rr_next, -err); /* less important */
32332 +       /* smp_mb(); */
32333 +
32334 +       AuDbg("b%d\n", err);
32335 +       return err;
32336 +}
32337 +
32338 +static int au_wbr_create_rr(struct dentry *dentry, unsigned int flags)
32339 +{
32340 +       int err, nbr;
32341 +       unsigned int u;
32342 +       aufs_bindex_t bindex, bbot;
32343 +       struct super_block *sb;
32344 +       atomic_t *next;
32345 +
32346 +       err = au_wbr_create_exp(dentry);
32347 +       if (err >= 0)
32348 +               goto out;
32349 +
32350 +       sb = dentry->d_sb;
32351 +       next = &au_sbi(sb)->si_wbr_rr_next;
32352 +       bbot = au_sbbot(sb);
32353 +       nbr = bbot + 1;
32354 +       for (bindex = 0; bindex <= bbot; bindex++) {
32355 +               if (!au_ftest_wbr(flags, DIR)) {
32356 +                       err = atomic_dec_return(next) + 1;
32357 +                       /* modulo for 0 is meaningless */
32358 +                       if (unlikely(!err))
32359 +                               err = atomic_dec_return(next) + 1;
32360 +               } else
32361 +                       err = atomic_read(next);
32362 +               AuDbg("%d\n", err);
32363 +               u = err;
32364 +               err = u % nbr;
32365 +               AuDbg("%d\n", err);
32366 +               if (!au_br_rdonly(au_sbr(sb, err)))
32367 +                       break;
32368 +               err = -EROFS;
32369 +       }
32370 +
32371 +       if (err >= 0)
32372 +               err = au_wbr_nonopq(dentry, err);
32373 +
32374 +out:
32375 +       AuDbg("%d\n", err);
32376 +       return err;
32377 +}
32378 +
32379 +/* ---------------------------------------------------------------------- */
32380 +
32381 +/* most free space */
32382 +static void au_mfs(struct dentry *dentry, struct dentry *parent)
32383 +{
32384 +       struct super_block *sb;
32385 +       struct au_branch *br;
32386 +       struct au_wbr_mfs *mfs;
32387 +       struct dentry *h_parent;
32388 +       aufs_bindex_t bindex, bbot;
32389 +       int err;
32390 +       unsigned long long b, bavail;
32391 +       struct path h_path;
32392 +       /* reduce the stack usage */
32393 +       struct kstatfs *st;
32394 +
32395 +       st = kmalloc(sizeof(*st), GFP_NOFS);
32396 +       if (unlikely(!st)) {
32397 +               AuWarn1("failed updating mfs(%d), ignored\n", -ENOMEM);
32398 +               return;
32399 +       }
32400 +
32401 +       bavail = 0;
32402 +       sb = dentry->d_sb;
32403 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32404 +       MtxMustLock(&mfs->mfs_lock);
32405 +       mfs->mfs_bindex = -EROFS;
32406 +       mfs->mfsrr_bytes = 0;
32407 +       if (!parent) {
32408 +               bindex = 0;
32409 +               bbot = au_sbbot(sb);
32410 +       } else {
32411 +               bindex = au_dbtop(parent);
32412 +               bbot = au_dbtaildir(parent);
32413 +       }
32414 +
32415 +       for (; bindex <= bbot; bindex++) {
32416 +               if (parent) {
32417 +                       h_parent = au_h_dptr(parent, bindex);
32418 +                       if (!h_parent || d_is_negative(h_parent))
32419 +                               continue;
32420 +               }
32421 +               br = au_sbr(sb, bindex);
32422 +               if (au_br_rdonly(br))
32423 +                       continue;
32424 +
32425 +               /* sb->s_root for NFS is unreliable */
32426 +               h_path.mnt = au_br_mnt(br);
32427 +               h_path.dentry = h_path.mnt->mnt_root;
32428 +               err = vfs_statfs(&h_path, st);
32429 +               if (unlikely(err)) {
32430 +                       AuWarn1("failed statfs, b%d, %d\n", bindex, err);
32431 +                       continue;
32432 +               }
32433 +
32434 +               /* when the available size is equal, select the lower one */
32435 +               BUILD_BUG_ON(sizeof(b) < sizeof(st->f_bavail)
32436 +                            || sizeof(b) < sizeof(st->f_bsize));
32437 +               b = st->f_bavail * st->f_bsize;
32438 +               br->br_wbr->wbr_bytes = b;
32439 +               if (b >= bavail) {
32440 +                       bavail = b;
32441 +                       mfs->mfs_bindex = bindex;
32442 +                       mfs->mfs_jiffy = jiffies;
32443 +               }
32444 +       }
32445 +
32446 +       mfs->mfsrr_bytes = bavail;
32447 +       AuDbg("b%d\n", mfs->mfs_bindex);
32448 +       au_delayed_kfree(st);
32449 +}
32450 +
32451 +static int au_wbr_create_mfs(struct dentry *dentry, unsigned int flags)
32452 +{
32453 +       int err;
32454 +       struct dentry *parent;
32455 +       struct super_block *sb;
32456 +       struct au_wbr_mfs *mfs;
32457 +
32458 +       err = au_wbr_create_exp(dentry);
32459 +       if (err >= 0)
32460 +               goto out;
32461 +
32462 +       sb = dentry->d_sb;
32463 +       parent = NULL;
32464 +       if (au_ftest_wbr(flags, PARENT))
32465 +               parent = dget_parent(dentry);
32466 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32467 +       mutex_lock(&mfs->mfs_lock);
32468 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
32469 +           || mfs->mfs_bindex < 0
32470 +           || au_br_rdonly(au_sbr(sb, mfs->mfs_bindex)))
32471 +               au_mfs(dentry, parent);
32472 +       mutex_unlock(&mfs->mfs_lock);
32473 +       err = mfs->mfs_bindex;
32474 +       dput(parent);
32475 +
32476 +       if (err >= 0)
32477 +               err = au_wbr_nonopq(dentry, err);
32478 +
32479 +out:
32480 +       AuDbg("b%d\n", err);
32481 +       return err;
32482 +}
32483 +
32484 +static int au_wbr_create_init_mfs(struct super_block *sb)
32485 +{
32486 +       struct au_wbr_mfs *mfs;
32487 +
32488 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32489 +       mutex_init(&mfs->mfs_lock);
32490 +       mfs->mfs_jiffy = 0;
32491 +       mfs->mfs_bindex = -EROFS;
32492 +
32493 +       return 0;
32494 +}
32495 +
32496 +static int au_wbr_create_fin_mfs(struct super_block *sb __maybe_unused)
32497 +{
32498 +       mutex_destroy(&au_sbi(sb)->si_wbr_mfs.mfs_lock);
32499 +       return 0;
32500 +}
32501 +
32502 +/* ---------------------------------------------------------------------- */
32503 +
32504 +/* top down regardless parent, and then mfs */
32505 +static int au_wbr_create_tdmfs(struct dentry *dentry,
32506 +                              unsigned int flags __maybe_unused)
32507 +{
32508 +       int err;
32509 +       aufs_bindex_t bwh, btail, bindex, bfound, bmfs;
32510 +       unsigned long long watermark;
32511 +       struct super_block *sb;
32512 +       struct au_wbr_mfs *mfs;
32513 +       struct au_branch *br;
32514 +       struct dentry *parent;
32515 +
32516 +       sb = dentry->d_sb;
32517 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32518 +       mutex_lock(&mfs->mfs_lock);
32519 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
32520 +           || mfs->mfs_bindex < 0)
32521 +               au_mfs(dentry, /*parent*/NULL);
32522 +       watermark = mfs->mfsrr_watermark;
32523 +       bmfs = mfs->mfs_bindex;
32524 +       mutex_unlock(&mfs->mfs_lock);
32525 +
32526 +       /* another style of au_wbr_create_exp() */
32527 +       bwh = au_dbwh(dentry);
32528 +       parent = dget_parent(dentry);
32529 +       btail = au_dbtaildir(parent);
32530 +       if (bwh >= 0 && bwh < btail)
32531 +               btail = bwh;
32532 +
32533 +       err = au_wbr_nonopq(dentry, btail);
32534 +       if (unlikely(err < 0))
32535 +               goto out;
32536 +       btail = err;
32537 +       bfound = -1;
32538 +       for (bindex = 0; bindex <= btail; bindex++) {
32539 +               br = au_sbr(sb, bindex);
32540 +               if (au_br_rdonly(br))
32541 +                       continue;
32542 +               if (br->br_wbr->wbr_bytes > watermark) {
32543 +                       bfound = bindex;
32544 +                       break;
32545 +               }
32546 +       }
32547 +       err = bfound;
32548 +       if (err < 0)
32549 +               err = bmfs;
32550 +
32551 +out:
32552 +       dput(parent);
32553 +       AuDbg("b%d\n", err);
32554 +       return err;
32555 +}
32556 +
32557 +/* ---------------------------------------------------------------------- */
32558 +
32559 +/* most free space and then round robin */
32560 +static int au_wbr_create_mfsrr(struct dentry *dentry, unsigned int flags)
32561 +{
32562 +       int err;
32563 +       struct au_wbr_mfs *mfs;
32564 +
32565 +       err = au_wbr_create_mfs(dentry, flags);
32566 +       if (err >= 0) {
32567 +               mfs = &au_sbi(dentry->d_sb)->si_wbr_mfs;
32568 +               mutex_lock(&mfs->mfs_lock);
32569 +               if (mfs->mfsrr_bytes < mfs->mfsrr_watermark)
32570 +                       err = au_wbr_create_rr(dentry, flags);
32571 +               mutex_unlock(&mfs->mfs_lock);
32572 +       }
32573 +
32574 +       AuDbg("b%d\n", err);
32575 +       return err;
32576 +}
32577 +
32578 +static int au_wbr_create_init_mfsrr(struct super_block *sb)
32579 +{
32580 +       int err;
32581 +
32582 +       au_wbr_create_init_mfs(sb); /* ignore */
32583 +       err = au_wbr_create_init_rr(sb);
32584 +
32585 +       return err;
32586 +}
32587 +
32588 +/* ---------------------------------------------------------------------- */
32589 +
32590 +/* top down parent and most free space */
32591 +static int au_wbr_create_pmfs(struct dentry *dentry, unsigned int flags)
32592 +{
32593 +       int err, e2;
32594 +       unsigned long long b;
32595 +       aufs_bindex_t bindex, btop, bbot;
32596 +       struct super_block *sb;
32597 +       struct dentry *parent, *h_parent;
32598 +       struct au_branch *br;
32599 +
32600 +       err = au_wbr_create_tdp(dentry, flags);
32601 +       if (unlikely(err < 0))
32602 +               goto out;
32603 +       parent = dget_parent(dentry);
32604 +       btop = au_dbtop(parent);
32605 +       bbot = au_dbtaildir(parent);
32606 +       if (btop == bbot)
32607 +               goto out_parent; /* success */
32608 +
32609 +       e2 = au_wbr_create_mfs(dentry, flags);
32610 +       if (e2 < 0)
32611 +               goto out_parent; /* success */
32612 +
32613 +       /* when the available size is equal, select upper one */
32614 +       sb = dentry->d_sb;
32615 +       br = au_sbr(sb, err);
32616 +       b = br->br_wbr->wbr_bytes;
32617 +       AuDbg("b%d, %llu\n", err, b);
32618 +
32619 +       for (bindex = btop; bindex <= bbot; bindex++) {
32620 +               h_parent = au_h_dptr(parent, bindex);
32621 +               if (!h_parent || d_is_negative(h_parent))
32622 +                       continue;
32623 +
32624 +               br = au_sbr(sb, bindex);
32625 +               if (!au_br_rdonly(br) && br->br_wbr->wbr_bytes > b) {
32626 +                       b = br->br_wbr->wbr_bytes;
32627 +                       err = bindex;
32628 +                       AuDbg("b%d, %llu\n", err, b);
32629 +               }
32630 +       }
32631 +
32632 +       if (err >= 0)
32633 +               err = au_wbr_nonopq(dentry, err);
32634 +
32635 +out_parent:
32636 +       dput(parent);
32637 +out:
32638 +       AuDbg("b%d\n", err);
32639 +       return err;
32640 +}
32641 +
32642 +/* ---------------------------------------------------------------------- */
32643 +
32644 +/*
32645 + * - top down parent
32646 + * - most free space with parent
32647 + * - most free space round-robin regardless parent
32648 + */
32649 +static int au_wbr_create_pmfsrr(struct dentry *dentry, unsigned int flags)
32650 +{
32651 +       int err;
32652 +       unsigned long long watermark;
32653 +       struct super_block *sb;
32654 +       struct au_branch *br;
32655 +       struct au_wbr_mfs *mfs;
32656 +
32657 +       err = au_wbr_create_pmfs(dentry, flags | AuWbr_PARENT);
32658 +       if (unlikely(err < 0))
32659 +               goto out;
32660 +
32661 +       sb = dentry->d_sb;
32662 +       br = au_sbr(sb, err);
32663 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32664 +       mutex_lock(&mfs->mfs_lock);
32665 +       watermark = mfs->mfsrr_watermark;
32666 +       mutex_unlock(&mfs->mfs_lock);
32667 +       if (br->br_wbr->wbr_bytes < watermark)
32668 +               /* regardless the parent dir */
32669 +               err = au_wbr_create_mfsrr(dentry, flags);
32670 +
32671 +out:
32672 +       AuDbg("b%d\n", err);
32673 +       return err;
32674 +}
32675 +
32676 +/* ---------------------------------------------------------------------- */
32677 +
32678 +/* policies for copyup */
32679 +
32680 +/* top down parent */
32681 +static int au_wbr_copyup_tdp(struct dentry *dentry)
32682 +{
32683 +       return au_wbr_create_tdp(dentry, /*flags, anything is ok*/0);
32684 +}
32685 +
32686 +/* bottom up parent */
32687 +static int au_wbr_copyup_bup(struct dentry *dentry)
32688 +{
32689 +       int err;
32690 +       aufs_bindex_t bindex, btop;
32691 +       struct dentry *parent, *h_parent;
32692 +       struct super_block *sb;
32693 +
32694 +       err = -EROFS;
32695 +       sb = dentry->d_sb;
32696 +       parent = dget_parent(dentry);
32697 +       btop = au_dbtop(parent);
32698 +       for (bindex = au_dbtop(dentry); bindex >= btop; bindex--) {
32699 +               h_parent = au_h_dptr(parent, bindex);
32700 +               if (!h_parent || d_is_negative(h_parent))
32701 +                       continue;
32702 +
32703 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
32704 +                       err = bindex;
32705 +                       break;
32706 +               }
32707 +       }
32708 +       dput(parent);
32709 +
32710 +       /* bottom up here */
32711 +       if (unlikely(err < 0))
32712 +               err = au_wbr_bu(sb, btop - 1);
32713 +
32714 +       AuDbg("b%d\n", err);
32715 +       return err;
32716 +}
32717 +
32718 +/* bottom up */
32719 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t btop)
32720 +{
32721 +       int err;
32722 +
32723 +       err = au_wbr_bu(dentry->d_sb, btop);
32724 +       AuDbg("b%d\n", err);
32725 +       if (err > btop)
32726 +               err = au_wbr_nonopq(dentry, err);
32727 +
32728 +       AuDbg("b%d\n", err);
32729 +       return err;
32730 +}
32731 +
32732 +static int au_wbr_copyup_bu(struct dentry *dentry)
32733 +{
32734 +       int err;
32735 +       aufs_bindex_t btop;
32736 +
32737 +       btop = au_dbtop(dentry);
32738 +       err = au_wbr_do_copyup_bu(dentry, btop);
32739 +       return err;
32740 +}
32741 +
32742 +/* ---------------------------------------------------------------------- */
32743 +
32744 +struct au_wbr_copyup_operations au_wbr_copyup_ops[] = {
32745 +       [AuWbrCopyup_TDP] = {
32746 +               .copyup = au_wbr_copyup_tdp
32747 +       },
32748 +       [AuWbrCopyup_BUP] = {
32749 +               .copyup = au_wbr_copyup_bup
32750 +       },
32751 +       [AuWbrCopyup_BU] = {
32752 +               .copyup = au_wbr_copyup_bu
32753 +       }
32754 +};
32755 +
32756 +struct au_wbr_create_operations au_wbr_create_ops[] = {
32757 +       [AuWbrCreate_TDP] = {
32758 +               .create = au_wbr_create_tdp
32759 +       },
32760 +       [AuWbrCreate_RR] = {
32761 +               .create = au_wbr_create_rr,
32762 +               .init   = au_wbr_create_init_rr
32763 +       },
32764 +       [AuWbrCreate_MFS] = {
32765 +               .create = au_wbr_create_mfs,
32766 +               .init   = au_wbr_create_init_mfs,
32767 +               .fin    = au_wbr_create_fin_mfs
32768 +       },
32769 +       [AuWbrCreate_MFSV] = {
32770 +               .create = au_wbr_create_mfs,
32771 +               .init   = au_wbr_create_init_mfs,
32772 +               .fin    = au_wbr_create_fin_mfs
32773 +       },
32774 +       [AuWbrCreate_MFSRR] = {
32775 +               .create = au_wbr_create_mfsrr,
32776 +               .init   = au_wbr_create_init_mfsrr,
32777 +               .fin    = au_wbr_create_fin_mfs
32778 +       },
32779 +       [AuWbrCreate_MFSRRV] = {
32780 +               .create = au_wbr_create_mfsrr,
32781 +               .init   = au_wbr_create_init_mfsrr,
32782 +               .fin    = au_wbr_create_fin_mfs
32783 +       },
32784 +       [AuWbrCreate_TDMFS] = {
32785 +               .create = au_wbr_create_tdmfs,
32786 +               .init   = au_wbr_create_init_mfs,
32787 +               .fin    = au_wbr_create_fin_mfs
32788 +       },
32789 +       [AuWbrCreate_TDMFSV] = {
32790 +               .create = au_wbr_create_tdmfs,
32791 +               .init   = au_wbr_create_init_mfs,
32792 +               .fin    = au_wbr_create_fin_mfs
32793 +       },
32794 +       [AuWbrCreate_PMFS] = {
32795 +               .create = au_wbr_create_pmfs,
32796 +               .init   = au_wbr_create_init_mfs,
32797 +               .fin    = au_wbr_create_fin_mfs
32798 +       },
32799 +       [AuWbrCreate_PMFSV] = {
32800 +               .create = au_wbr_create_pmfs,
32801 +               .init   = au_wbr_create_init_mfs,
32802 +               .fin    = au_wbr_create_fin_mfs
32803 +       },
32804 +       [AuWbrCreate_PMFSRR] = {
32805 +               .create = au_wbr_create_pmfsrr,
32806 +               .init   = au_wbr_create_init_mfsrr,
32807 +               .fin    = au_wbr_create_fin_mfs
32808 +       },
32809 +       [AuWbrCreate_PMFSRRV] = {
32810 +               .create = au_wbr_create_pmfsrr,
32811 +               .init   = au_wbr_create_init_mfsrr,
32812 +               .fin    = au_wbr_create_fin_mfs
32813 +       }
32814 +};
32815 diff -urN /usr/share/empty/fs/aufs/whout.c linux/fs/aufs/whout.c
32816 --- /usr/share/empty/fs/aufs/whout.c    1970-01-01 01:00:00.000000000 +0100
32817 +++ linux/fs/aufs/whout.c       2017-05-06 22:16:52.784886929 +0200
32818 @@ -0,0 +1,1061 @@
32819 +/*
32820 + * Copyright (C) 2005-2017 Junjiro R. Okajima
32821 + *
32822 + * This program, aufs is free software; you can redistribute it and/or modify
32823 + * it under the terms of the GNU General Public License as published by
32824 + * the Free Software Foundation; either version 2 of the License, or
32825 + * (at your option) any later version.
32826 + *
32827 + * This program is distributed in the hope that it will be useful,
32828 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32829 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32830 + * GNU General Public License for more details.
32831 + *
32832 + * You should have received a copy of the GNU General Public License
32833 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32834 + */
32835 +
32836 +/*
32837 + * whiteout for logical deletion and opaque directory
32838 + */
32839 +
32840 +#include "aufs.h"
32841 +
32842 +#define WH_MASK                        S_IRUGO
32843 +
32844 +/*
32845 + * If a directory contains this file, then it is opaque.  We start with the
32846 + * .wh. flag so that it is blocked by lookup.
32847 + */
32848 +static struct qstr diropq_name = QSTR_INIT(AUFS_WH_DIROPQ,
32849 +                                          sizeof(AUFS_WH_DIROPQ) - 1);
32850 +
32851 +/*
32852 + * generate whiteout name, which is NOT terminated by NULL.
32853 + * @name: original d_name.name
32854 + * @len: original d_name.len
32855 + * @wh: whiteout qstr
32856 + * returns zero when succeeds, otherwise error.
32857 + * succeeded value as wh->name should be freed by kfree().
32858 + */
32859 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name)
32860 +{
32861 +       char *p;
32862 +
32863 +       if (unlikely(name->len > PATH_MAX - AUFS_WH_PFX_LEN))
32864 +               return -ENAMETOOLONG;
32865 +
32866 +       wh->len = name->len + AUFS_WH_PFX_LEN;
32867 +       p = kmalloc(wh->len, GFP_NOFS);
32868 +       wh->name = p;
32869 +       if (p) {
32870 +               memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
32871 +               memcpy(p + AUFS_WH_PFX_LEN, name->name, name->len);
32872 +               /* smp_mb(); */
32873 +               return 0;
32874 +       }
32875 +       return -ENOMEM;
32876 +}
32877 +
32878 +/* ---------------------------------------------------------------------- */
32879 +
32880 +/*
32881 + * test if the @wh_name exists under @h_parent.
32882 + * @try_sio specifies the necessary of super-io.
32883 + */
32884 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio)
32885 +{
32886 +       int err;
32887 +       struct dentry *wh_dentry;
32888 +
32889 +       if (!try_sio)
32890 +               wh_dentry = vfsub_lkup_one(wh_name, h_parent);
32891 +       else
32892 +               wh_dentry = au_sio_lkup_one(wh_name, h_parent);
32893 +       err = PTR_ERR(wh_dentry);
32894 +       if (IS_ERR(wh_dentry)) {
32895 +               if (err == -ENAMETOOLONG)
32896 +                       err = 0;
32897 +               goto out;
32898 +       }
32899 +
32900 +       err = 0;
32901 +       if (d_is_negative(wh_dentry))
32902 +               goto out_wh; /* success */
32903 +
32904 +       err = 1;
32905 +       if (d_is_reg(wh_dentry))
32906 +               goto out_wh; /* success */
32907 +
32908 +       err = -EIO;
32909 +       AuIOErr("%pd Invalid whiteout entry type 0%o.\n",
32910 +               wh_dentry, d_inode(wh_dentry)->i_mode);
32911 +
32912 +out_wh:
32913 +       dput(wh_dentry);
32914 +out:
32915 +       return err;
32916 +}
32917 +
32918 +/*
32919 + * test if the @h_dentry sets opaque or not.
32920 + */
32921 +int au_diropq_test(struct dentry *h_dentry)
32922 +{
32923 +       int err;
32924 +       struct inode *h_dir;
32925 +
32926 +       h_dir = d_inode(h_dentry);
32927 +       err = au_wh_test(h_dentry, &diropq_name,
32928 +                        au_test_h_perm_sio(h_dir, MAY_EXEC));
32929 +       return err;
32930 +}
32931 +
32932 +/*
32933 + * returns a negative dentry whose name is unique and temporary.
32934 + */
32935 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
32936 +                            struct qstr *prefix)
32937 +{
32938 +       struct dentry *dentry;
32939 +       int i;
32940 +       char defname[NAME_MAX - AUFS_MAX_NAMELEN + DNAME_INLINE_LEN + 1],
32941 +               *name, *p;
32942 +       /* strict atomic_t is unnecessary here */
32943 +       static unsigned short cnt;
32944 +       struct qstr qs;
32945 +
32946 +       BUILD_BUG_ON(sizeof(cnt) * 2 > AUFS_WH_TMP_LEN);
32947 +
32948 +       name = defname;
32949 +       qs.len = sizeof(defname) - DNAME_INLINE_LEN + prefix->len - 1;
32950 +       if (unlikely(prefix->len > DNAME_INLINE_LEN)) {
32951 +               dentry = ERR_PTR(-ENAMETOOLONG);
32952 +               if (unlikely(qs.len > NAME_MAX))
32953 +                       goto out;
32954 +               dentry = ERR_PTR(-ENOMEM);
32955 +               name = kmalloc(qs.len + 1, GFP_NOFS);
32956 +               if (unlikely(!name))
32957 +                       goto out;
32958 +       }
32959 +
32960 +       /* doubly whiteout-ed */
32961 +       memcpy(name, AUFS_WH_PFX AUFS_WH_PFX, AUFS_WH_PFX_LEN * 2);
32962 +       p = name + AUFS_WH_PFX_LEN * 2;
32963 +       memcpy(p, prefix->name, prefix->len);
32964 +       p += prefix->len;
32965 +       *p++ = '.';
32966 +       AuDebugOn(name + qs.len + 1 - p <= AUFS_WH_TMP_LEN);
32967 +
32968 +       qs.name = name;
32969 +       for (i = 0; i < 3; i++) {
32970 +               sprintf(p, "%.*x", AUFS_WH_TMP_LEN, cnt++);
32971 +               dentry = au_sio_lkup_one(&qs, h_parent);
32972 +               if (IS_ERR(dentry) || d_is_negative(dentry))
32973 +                       goto out_name;
32974 +               dput(dentry);
32975 +       }
32976 +       /* pr_warn("could not get random name\n"); */
32977 +       dentry = ERR_PTR(-EEXIST);
32978 +       AuDbg("%.*s\n", AuLNPair(&qs));
32979 +       BUG();
32980 +
32981 +out_name:
32982 +       if (name != defname)
32983 +               au_delayed_kfree(name);
32984 +out:
32985 +       AuTraceErrPtr(dentry);
32986 +       return dentry;
32987 +}
32988 +
32989 +/*
32990 + * rename the @h_dentry on @br to the whiteouted temporary name.
32991 + */
32992 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br)
32993 +{
32994 +       int err;
32995 +       struct path h_path = {
32996 +               .mnt = au_br_mnt(br)
32997 +       };
32998 +       struct inode *h_dir, *delegated;
32999 +       struct dentry *h_parent;
33000 +
33001 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
33002 +       h_dir = d_inode(h_parent);
33003 +       IMustLock(h_dir);
33004 +
33005 +       h_path.dentry = au_whtmp_lkup(h_parent, br, &h_dentry->d_name);
33006 +       err = PTR_ERR(h_path.dentry);
33007 +       if (IS_ERR(h_path.dentry))
33008 +               goto out;
33009 +
33010 +       /* under the same dir, no need to lock_rename() */
33011 +       delegated = NULL;
33012 +       err = vfsub_rename(h_dir, h_dentry, h_dir, &h_path, &delegated,
33013 +                          /*flags*/0);
33014 +       AuTraceErr(err);
33015 +       if (unlikely(err == -EWOULDBLOCK)) {
33016 +               pr_warn("cannot retry for NFSv4 delegation"
33017 +                       " for an internal rename\n");
33018 +               iput(delegated);
33019 +       }
33020 +       dput(h_path.dentry);
33021 +
33022 +out:
33023 +       AuTraceErr(err);
33024 +       return err;
33025 +}
33026 +
33027 +/* ---------------------------------------------------------------------- */
33028 +/*
33029 + * functions for removing a whiteout
33030 + */
33031 +
33032 +static int do_unlink_wh(struct inode *h_dir, struct path *h_path)
33033 +{
33034 +       int err, force;
33035 +       struct inode *delegated;
33036 +
33037 +       /*
33038 +        * forces superio when the dir has a sticky bit.
33039 +        * this may be a violation of unix fs semantics.
33040 +        */
33041 +       force = (h_dir->i_mode & S_ISVTX)
33042 +               && !uid_eq(current_fsuid(), d_inode(h_path->dentry)->i_uid);
33043 +       delegated = NULL;
33044 +       err = vfsub_unlink(h_dir, h_path, &delegated, force);
33045 +       if (unlikely(err == -EWOULDBLOCK)) {
33046 +               pr_warn("cannot retry for NFSv4 delegation"
33047 +                       " for an internal unlink\n");
33048 +               iput(delegated);
33049 +       }
33050 +       return err;
33051 +}
33052 +
33053 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
33054 +                       struct dentry *dentry)
33055 +{
33056 +       int err;
33057 +
33058 +       err = do_unlink_wh(h_dir, h_path);
33059 +       if (!err && dentry)
33060 +               au_set_dbwh(dentry, -1);
33061 +
33062 +       return err;
33063 +}
33064 +
33065 +static int unlink_wh_name(struct dentry *h_parent, struct qstr *wh,
33066 +                         struct au_branch *br)
33067 +{
33068 +       int err;
33069 +       struct path h_path = {
33070 +               .mnt = au_br_mnt(br)
33071 +       };
33072 +
33073 +       err = 0;
33074 +       h_path.dentry = vfsub_lkup_one(wh, h_parent);
33075 +       if (IS_ERR(h_path.dentry))
33076 +               err = PTR_ERR(h_path.dentry);
33077 +       else {
33078 +               if (d_is_reg(h_path.dentry))
33079 +                       err = do_unlink_wh(d_inode(h_parent), &h_path);
33080 +               dput(h_path.dentry);
33081 +       }
33082 +
33083 +       return err;
33084 +}
33085 +
33086 +/* ---------------------------------------------------------------------- */
33087 +/*
33088 + * initialize/clean whiteout for a branch
33089 + */
33090 +
33091 +static void au_wh_clean(struct inode *h_dir, struct path *whpath,
33092 +                       const int isdir)
33093 +{
33094 +       int err;
33095 +       struct inode *delegated;
33096 +
33097 +       if (d_is_negative(whpath->dentry))
33098 +               return;
33099 +
33100 +       if (isdir)
33101 +               err = vfsub_rmdir(h_dir, whpath);
33102 +       else {
33103 +               delegated = NULL;
33104 +               err = vfsub_unlink(h_dir, whpath, &delegated, /*force*/0);
33105 +               if (unlikely(err == -EWOULDBLOCK)) {
33106 +                       pr_warn("cannot retry for NFSv4 delegation"
33107 +                               " for an internal unlink\n");
33108 +                       iput(delegated);
33109 +               }
33110 +       }
33111 +       if (unlikely(err))
33112 +               pr_warn("failed removing %pd (%d), ignored.\n",
33113 +                       whpath->dentry, err);
33114 +}
33115 +
33116 +static int test_linkable(struct dentry *h_root)
33117 +{
33118 +       struct inode *h_dir = d_inode(h_root);
33119 +
33120 +       if (h_dir->i_op->link)
33121 +               return 0;
33122 +
33123 +       pr_err("%pd (%s) doesn't support link(2), use noplink and rw+nolwh\n",
33124 +              h_root, au_sbtype(h_root->d_sb));
33125 +       return -ENOSYS;
33126 +}
33127 +
33128 +/* todo: should this mkdir be done in /sbin/mount.aufs helper? */
33129 +static int au_whdir(struct inode *h_dir, struct path *path)
33130 +{
33131 +       int err;
33132 +
33133 +       err = -EEXIST;
33134 +       if (d_is_negative(path->dentry)) {
33135 +               int mode = S_IRWXU;
33136 +
33137 +               if (au_test_nfs(path->dentry->d_sb))
33138 +                       mode |= S_IXUGO;
33139 +               err = vfsub_mkdir(h_dir, path, mode);
33140 +       } else if (d_is_dir(path->dentry))
33141 +               err = 0;
33142 +       else
33143 +               pr_err("unknown %pd exists\n", path->dentry);
33144 +
33145 +       return err;
33146 +}
33147 +
33148 +struct au_wh_base {
33149 +       const struct qstr *name;
33150 +       struct dentry *dentry;
33151 +};
33152 +
33153 +static void au_wh_init_ro(struct inode *h_dir, struct au_wh_base base[],
33154 +                         struct path *h_path)
33155 +{
33156 +       h_path->dentry = base[AuBrWh_BASE].dentry;
33157 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
33158 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
33159 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
33160 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
33161 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
33162 +}
33163 +
33164 +/*
33165 + * returns tri-state,
33166 + * minus: error, caller should print the message
33167 + * zero: succuess
33168 + * plus: error, caller should NOT print the message
33169 + */
33170 +static int au_wh_init_rw_nolink(struct dentry *h_root, struct au_wbr *wbr,
33171 +                               int do_plink, struct au_wh_base base[],
33172 +                               struct path *h_path)
33173 +{
33174 +       int err;
33175 +       struct inode *h_dir;
33176 +
33177 +       h_dir = d_inode(h_root);
33178 +       h_path->dentry = base[AuBrWh_BASE].dentry;
33179 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
33180 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
33181 +       if (do_plink) {
33182 +               err = test_linkable(h_root);
33183 +               if (unlikely(err)) {
33184 +                       err = 1;
33185 +                       goto out;
33186 +               }
33187 +
33188 +               err = au_whdir(h_dir, h_path);
33189 +               if (unlikely(err))
33190 +                       goto out;
33191 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
33192 +       } else
33193 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
33194 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
33195 +       err = au_whdir(h_dir, h_path);
33196 +       if (unlikely(err))
33197 +               goto out;
33198 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
33199 +
33200 +out:
33201 +       return err;
33202 +}
33203 +
33204 +/*
33205 + * for the moment, aufs supports the branch filesystem which does not support
33206 + * link(2). testing on FAT which does not support i_op->setattr() fully either,
33207 + * copyup failed. finally, such filesystem will not be used as the writable
33208 + * branch.
33209 + *
33210 + * returns tri-state, see above.
33211 + */
33212 +static int au_wh_init_rw(struct dentry *h_root, struct au_wbr *wbr,
33213 +                        int do_plink, struct au_wh_base base[],
33214 +                        struct path *h_path)
33215 +{
33216 +       int err;
33217 +       struct inode *h_dir;
33218 +
33219 +       WbrWhMustWriteLock(wbr);
33220 +
33221 +       err = test_linkable(h_root);
33222 +       if (unlikely(err)) {
33223 +               err = 1;
33224 +               goto out;
33225 +       }
33226 +
33227 +       /*
33228 +        * todo: should this create be done in /sbin/mount.aufs helper?
33229 +        */
33230 +       err = -EEXIST;
33231 +       h_dir = d_inode(h_root);
33232 +       if (d_is_negative(base[AuBrWh_BASE].dentry)) {
33233 +               h_path->dentry = base[AuBrWh_BASE].dentry;
33234 +               err = vfsub_create(h_dir, h_path, WH_MASK, /*want_excl*/true);
33235 +       } else if (d_is_reg(base[AuBrWh_BASE].dentry))
33236 +               err = 0;
33237 +       else
33238 +               pr_err("unknown %pd2 exists\n", base[AuBrWh_BASE].dentry);
33239 +       if (unlikely(err))
33240 +               goto out;
33241 +
33242 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
33243 +       if (do_plink) {
33244 +               err = au_whdir(h_dir, h_path);
33245 +               if (unlikely(err))
33246 +                       goto out;
33247 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
33248 +       } else
33249 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
33250 +       wbr->wbr_whbase = dget(base[AuBrWh_BASE].dentry);
33251 +
33252 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
33253 +       err = au_whdir(h_dir, h_path);
33254 +       if (unlikely(err))
33255 +               goto out;
33256 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
33257 +
33258 +out:
33259 +       return err;
33260 +}
33261 +
33262 +/*
33263 + * initialize the whiteout base file/dir for @br.
33264 + */
33265 +int au_wh_init(struct au_branch *br, struct super_block *sb)
33266 +{
33267 +       int err, i;
33268 +       const unsigned char do_plink
33269 +               = !!au_opt_test(au_mntflags(sb), PLINK);
33270 +       struct inode *h_dir;
33271 +       struct path path = br->br_path;
33272 +       struct dentry *h_root = path.dentry;
33273 +       struct au_wbr *wbr = br->br_wbr;
33274 +       static const struct qstr base_name[] = {
33275 +               [AuBrWh_BASE] = QSTR_INIT(AUFS_BASE_NAME,
33276 +                                         sizeof(AUFS_BASE_NAME) - 1),
33277 +               [AuBrWh_PLINK] = QSTR_INIT(AUFS_PLINKDIR_NAME,
33278 +                                          sizeof(AUFS_PLINKDIR_NAME) - 1),
33279 +               [AuBrWh_ORPH] = QSTR_INIT(AUFS_ORPHDIR_NAME,
33280 +                                         sizeof(AUFS_ORPHDIR_NAME) - 1)
33281 +       };
33282 +       struct au_wh_base base[] = {
33283 +               [AuBrWh_BASE] = {
33284 +                       .name   = base_name + AuBrWh_BASE,
33285 +                       .dentry = NULL
33286 +               },
33287 +               [AuBrWh_PLINK] = {
33288 +                       .name   = base_name + AuBrWh_PLINK,
33289 +                       .dentry = NULL
33290 +               },
33291 +               [AuBrWh_ORPH] = {
33292 +                       .name   = base_name + AuBrWh_ORPH,
33293 +                       .dentry = NULL
33294 +               }
33295 +       };
33296 +
33297 +       if (wbr)
33298 +               WbrWhMustWriteLock(wbr);
33299 +
33300 +       for (i = 0; i < AuBrWh_Last; i++) {
33301 +               /* doubly whiteouted */
33302 +               struct dentry *d;
33303 +
33304 +               d = au_wh_lkup(h_root, (void *)base[i].name, br);
33305 +               err = PTR_ERR(d);
33306 +               if (IS_ERR(d))
33307 +                       goto out;
33308 +
33309 +               base[i].dentry = d;
33310 +               AuDebugOn(wbr
33311 +                         && wbr->wbr_wh[i]
33312 +                         && wbr->wbr_wh[i] != base[i].dentry);
33313 +       }
33314 +
33315 +       if (wbr)
33316 +               for (i = 0; i < AuBrWh_Last; i++) {
33317 +                       dput(wbr->wbr_wh[i]);
33318 +                       wbr->wbr_wh[i] = NULL;
33319 +               }
33320 +
33321 +       err = 0;
33322 +       if (!au_br_writable(br->br_perm)) {
33323 +               h_dir = d_inode(h_root);
33324 +               au_wh_init_ro(h_dir, base, &path);
33325 +       } else if (!au_br_wh_linkable(br->br_perm)) {
33326 +               err = au_wh_init_rw_nolink(h_root, wbr, do_plink, base, &path);
33327 +               if (err > 0)
33328 +                       goto out;
33329 +               else if (err)
33330 +                       goto out_err;
33331 +       } else {
33332 +               err = au_wh_init_rw(h_root, wbr, do_plink, base, &path);
33333 +               if (err > 0)
33334 +                       goto out;
33335 +               else if (err)
33336 +                       goto out_err;
33337 +       }
33338 +       goto out; /* success */
33339 +
33340 +out_err:
33341 +       pr_err("an error(%d) on the writable branch %pd(%s)\n",
33342 +              err, h_root, au_sbtype(h_root->d_sb));
33343 +out:
33344 +       for (i = 0; i < AuBrWh_Last; i++)
33345 +               dput(base[i].dentry);
33346 +       return err;
33347 +}
33348 +
33349 +/* ---------------------------------------------------------------------- */
33350 +/*
33351 + * whiteouts are all hard-linked usually.
33352 + * when its link count reaches a ceiling, we create a new whiteout base
33353 + * asynchronously.
33354 + */
33355 +
33356 +struct reinit_br_wh {
33357 +       struct super_block *sb;
33358 +       struct au_branch *br;
33359 +};
33360 +
33361 +static void reinit_br_wh(void *arg)
33362 +{
33363 +       int err;
33364 +       aufs_bindex_t bindex;
33365 +       struct path h_path;
33366 +       struct reinit_br_wh *a = arg;
33367 +       struct au_wbr *wbr;
33368 +       struct inode *dir, *delegated;
33369 +       struct dentry *h_root;
33370 +       struct au_hinode *hdir;
33371 +
33372 +       err = 0;
33373 +       wbr = a->br->br_wbr;
33374 +       /* big aufs lock */
33375 +       si_noflush_write_lock(a->sb);
33376 +       if (!au_br_writable(a->br->br_perm))
33377 +               goto out;
33378 +       bindex = au_br_index(a->sb, a->br->br_id);
33379 +       if (unlikely(bindex < 0))
33380 +               goto out;
33381 +
33382 +       di_read_lock_parent(a->sb->s_root, AuLock_IR);
33383 +       dir = d_inode(a->sb->s_root);
33384 +       hdir = au_hi(dir, bindex);
33385 +       h_root = au_h_dptr(a->sb->s_root, bindex);
33386 +       AuDebugOn(h_root != au_br_dentry(a->br));
33387 +
33388 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
33389 +       wbr_wh_write_lock(wbr);
33390 +       err = au_h_verify(wbr->wbr_whbase, au_opt_udba(a->sb), hdir->hi_inode,
33391 +                         h_root, a->br);
33392 +       if (!err) {
33393 +               h_path.dentry = wbr->wbr_whbase;
33394 +               h_path.mnt = au_br_mnt(a->br);
33395 +               delegated = NULL;
33396 +               err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated,
33397 +                                  /*force*/0);
33398 +               if (unlikely(err == -EWOULDBLOCK)) {
33399 +                       pr_warn("cannot retry for NFSv4 delegation"
33400 +                               " for an internal unlink\n");
33401 +                       iput(delegated);
33402 +               }
33403 +       } else {
33404 +               pr_warn("%pd is moved, ignored\n", wbr->wbr_whbase);
33405 +               err = 0;
33406 +       }
33407 +       dput(wbr->wbr_whbase);
33408 +       wbr->wbr_whbase = NULL;
33409 +       if (!err)
33410 +               err = au_wh_init(a->br, a->sb);
33411 +       wbr_wh_write_unlock(wbr);
33412 +       au_hn_inode_unlock(hdir);
33413 +       di_read_unlock(a->sb->s_root, AuLock_IR);
33414 +       if (!err)
33415 +               au_fhsm_wrote(a->sb, bindex, /*force*/0);
33416 +
33417 +out:
33418 +       if (wbr)
33419 +               atomic_dec(&wbr->wbr_wh_running);
33420 +       au_br_put(a->br);
33421 +       si_write_unlock(a->sb);
33422 +       au_nwt_done(&au_sbi(a->sb)->si_nowait);
33423 +       au_delayed_kfree(arg);
33424 +       if (unlikely(err))
33425 +               AuIOErr("err %d\n", err);
33426 +}
33427 +
33428 +static void kick_reinit_br_wh(struct super_block *sb, struct au_branch *br)
33429 +{
33430 +       int do_dec, wkq_err;
33431 +       struct reinit_br_wh *arg;
33432 +
33433 +       do_dec = 1;
33434 +       if (atomic_inc_return(&br->br_wbr->wbr_wh_running) != 1)
33435 +               goto out;
33436 +
33437 +       /* ignore ENOMEM */
33438 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
33439 +       if (arg) {
33440 +               /*
33441 +                * dec(wh_running), kfree(arg) and dec(br_count)
33442 +                * in reinit function
33443 +                */
33444 +               arg->sb = sb;
33445 +               arg->br = br;
33446 +               au_br_get(br);
33447 +               wkq_err = au_wkq_nowait(reinit_br_wh, arg, sb, /*flags*/0);
33448 +               if (unlikely(wkq_err)) {
33449 +                       atomic_dec(&br->br_wbr->wbr_wh_running);
33450 +                       au_br_put(br);
33451 +                       au_delayed_kfree(arg);
33452 +               }
33453 +               do_dec = 0;
33454 +       }
33455 +
33456 +out:
33457 +       if (do_dec)
33458 +               atomic_dec(&br->br_wbr->wbr_wh_running);
33459 +}
33460 +
33461 +/* ---------------------------------------------------------------------- */
33462 +
33463 +/*
33464 + * create the whiteout @wh.
33465 + */
33466 +static int link_or_create_wh(struct super_block *sb, aufs_bindex_t bindex,
33467 +                            struct dentry *wh)
33468 +{
33469 +       int err;
33470 +       struct path h_path = {
33471 +               .dentry = wh
33472 +       };
33473 +       struct au_branch *br;
33474 +       struct au_wbr *wbr;
33475 +       struct dentry *h_parent;
33476 +       struct inode *h_dir, *delegated;
33477 +
33478 +       h_parent = wh->d_parent; /* dir inode is locked */
33479 +       h_dir = d_inode(h_parent);
33480 +       IMustLock(h_dir);
33481 +
33482 +       br = au_sbr(sb, bindex);
33483 +       h_path.mnt = au_br_mnt(br);
33484 +       wbr = br->br_wbr;
33485 +       wbr_wh_read_lock(wbr);
33486 +       if (wbr->wbr_whbase) {
33487 +               delegated = NULL;
33488 +               err = vfsub_link(wbr->wbr_whbase, h_dir, &h_path, &delegated);
33489 +               if (unlikely(err == -EWOULDBLOCK)) {
33490 +                       pr_warn("cannot retry for NFSv4 delegation"
33491 +                               " for an internal link\n");
33492 +                       iput(delegated);
33493 +               }
33494 +               if (!err || err != -EMLINK)
33495 +                       goto out;
33496 +
33497 +               /* link count full. re-initialize br_whbase. */
33498 +               kick_reinit_br_wh(sb, br);
33499 +       }
33500 +
33501 +       /* return this error in this context */
33502 +       err = vfsub_create(h_dir, &h_path, WH_MASK, /*want_excl*/true);
33503 +       if (!err)
33504 +               au_fhsm_wrote(sb, bindex, /*force*/0);
33505 +
33506 +out:
33507 +       wbr_wh_read_unlock(wbr);
33508 +       return err;
33509 +}
33510 +
33511 +/* ---------------------------------------------------------------------- */
33512 +
33513 +/*
33514 + * create or remove the diropq.
33515 + */
33516 +static struct dentry *do_diropq(struct dentry *dentry, aufs_bindex_t bindex,
33517 +                               unsigned int flags)
33518 +{
33519 +       struct dentry *opq_dentry, *h_dentry;
33520 +       struct super_block *sb;
33521 +       struct au_branch *br;
33522 +       int err;
33523 +
33524 +       sb = dentry->d_sb;
33525 +       br = au_sbr(sb, bindex);
33526 +       h_dentry = au_h_dptr(dentry, bindex);
33527 +       opq_dentry = vfsub_lkup_one(&diropq_name, h_dentry);
33528 +       if (IS_ERR(opq_dentry))
33529 +               goto out;
33530 +
33531 +       if (au_ftest_diropq(flags, CREATE)) {
33532 +               err = link_or_create_wh(sb, bindex, opq_dentry);
33533 +               if (!err) {
33534 +                       au_set_dbdiropq(dentry, bindex);
33535 +                       goto out; /* success */
33536 +               }
33537 +       } else {
33538 +               struct path tmp = {
33539 +                       .dentry = opq_dentry,
33540 +                       .mnt    = au_br_mnt(br)
33541 +               };
33542 +               err = do_unlink_wh(au_h_iptr(d_inode(dentry), bindex), &tmp);
33543 +               if (!err)
33544 +                       au_set_dbdiropq(dentry, -1);
33545 +       }
33546 +       dput(opq_dentry);
33547 +       opq_dentry = ERR_PTR(err);
33548 +
33549 +out:
33550 +       return opq_dentry;
33551 +}
33552 +
33553 +struct do_diropq_args {
33554 +       struct dentry **errp;
33555 +       struct dentry *dentry;
33556 +       aufs_bindex_t bindex;
33557 +       unsigned int flags;
33558 +};
33559 +
33560 +static void call_do_diropq(void *args)
33561 +{
33562 +       struct do_diropq_args *a = args;
33563 +       *a->errp = do_diropq(a->dentry, a->bindex, a->flags);
33564 +}
33565 +
33566 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
33567 +                            unsigned int flags)
33568 +{
33569 +       struct dentry *diropq, *h_dentry;
33570 +
33571 +       h_dentry = au_h_dptr(dentry, bindex);
33572 +       if (!au_test_h_perm_sio(d_inode(h_dentry), MAY_EXEC | MAY_WRITE))
33573 +               diropq = do_diropq(dentry, bindex, flags);
33574 +       else {
33575 +               int wkq_err;
33576 +               struct do_diropq_args args = {
33577 +                       .errp           = &diropq,
33578 +                       .dentry         = dentry,
33579 +                       .bindex         = bindex,
33580 +                       .flags          = flags
33581 +               };
33582 +
33583 +               wkq_err = au_wkq_wait(call_do_diropq, &args);
33584 +               if (unlikely(wkq_err))
33585 +                       diropq = ERR_PTR(wkq_err);
33586 +       }
33587 +
33588 +       return diropq;
33589 +}
33590 +
33591 +/* ---------------------------------------------------------------------- */
33592 +
33593 +/*
33594 + * lookup whiteout dentry.
33595 + * @h_parent: lower parent dentry which must exist and be locked
33596 + * @base_name: name of dentry which will be whiteouted
33597 + * returns dentry for whiteout.
33598 + */
33599 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
33600 +                         struct au_branch *br)
33601 +{
33602 +       int err;
33603 +       struct qstr wh_name;
33604 +       struct dentry *wh_dentry;
33605 +
33606 +       err = au_wh_name_alloc(&wh_name, base_name);
33607 +       wh_dentry = ERR_PTR(err);
33608 +       if (!err) {
33609 +               wh_dentry = vfsub_lkup_one(&wh_name, h_parent);
33610 +               au_delayed_kfree(wh_name.name);
33611 +       }
33612 +       return wh_dentry;
33613 +}
33614 +
33615 +/*
33616 + * link/create a whiteout for @dentry on @bindex.
33617 + */
33618 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
33619 +                           struct dentry *h_parent)
33620 +{
33621 +       struct dentry *wh_dentry;
33622 +       struct super_block *sb;
33623 +       int err;
33624 +
33625 +       sb = dentry->d_sb;
33626 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, au_sbr(sb, bindex));
33627 +       if (!IS_ERR(wh_dentry) && d_is_negative(wh_dentry)) {
33628 +               err = link_or_create_wh(sb, bindex, wh_dentry);
33629 +               if (!err) {
33630 +                       au_set_dbwh(dentry, bindex);
33631 +                       au_fhsm_wrote(sb, bindex, /*force*/0);
33632 +               } else {
33633 +                       dput(wh_dentry);
33634 +                       wh_dentry = ERR_PTR(err);
33635 +               }
33636 +       }
33637 +
33638 +       return wh_dentry;
33639 +}
33640 +
33641 +/* ---------------------------------------------------------------------- */
33642 +
33643 +/* Delete all whiteouts in this directory on branch bindex. */
33644 +static int del_wh_children(struct dentry *h_dentry, struct au_nhash *whlist,
33645 +                          aufs_bindex_t bindex, struct au_branch *br)
33646 +{
33647 +       int err;
33648 +       unsigned long ul, n;
33649 +       struct qstr wh_name;
33650 +       char *p;
33651 +       struct hlist_head *head;
33652 +       struct au_vdir_wh *pos;
33653 +       struct au_vdir_destr *str;
33654 +
33655 +       err = -ENOMEM;
33656 +       p = (void *)__get_free_page(GFP_NOFS);
33657 +       wh_name.name = p;
33658 +       if (unlikely(!wh_name.name))
33659 +               goto out;
33660 +
33661 +       err = 0;
33662 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
33663 +       p += AUFS_WH_PFX_LEN;
33664 +       n = whlist->nh_num;
33665 +       head = whlist->nh_head;
33666 +       for (ul = 0; !err && ul < n; ul++, head++) {
33667 +               hlist_for_each_entry(pos, head, wh_hash) {
33668 +                       if (pos->wh_bindex != bindex)
33669 +                               continue;
33670 +
33671 +                       str = &pos->wh_str;
33672 +                       if (str->len + AUFS_WH_PFX_LEN <= PATH_MAX) {
33673 +                               memcpy(p, str->name, str->len);
33674 +                               wh_name.len = AUFS_WH_PFX_LEN + str->len;
33675 +                               err = unlink_wh_name(h_dentry, &wh_name, br);
33676 +                               if (!err)
33677 +                                       continue;
33678 +                               break;
33679 +                       }
33680 +                       AuIOErr("whiteout name too long %.*s\n",
33681 +                               str->len, str->name);
33682 +                       err = -EIO;
33683 +                       break;
33684 +               }
33685 +       }
33686 +       au_delayed_free_page((unsigned long)wh_name.name);
33687 +
33688 +out:
33689 +       return err;
33690 +}
33691 +
33692 +struct del_wh_children_args {
33693 +       int *errp;
33694 +       struct dentry *h_dentry;
33695 +       struct au_nhash *whlist;
33696 +       aufs_bindex_t bindex;
33697 +       struct au_branch *br;
33698 +};
33699 +
33700 +static void call_del_wh_children(void *args)
33701 +{
33702 +       struct del_wh_children_args *a = args;
33703 +       *a->errp = del_wh_children(a->h_dentry, a->whlist, a->bindex, a->br);
33704 +}
33705 +
33706 +/* ---------------------------------------------------------------------- */
33707 +
33708 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp)
33709 +{
33710 +       struct au_whtmp_rmdir *whtmp;
33711 +       int err;
33712 +       unsigned int rdhash;
33713 +
33714 +       SiMustAnyLock(sb);
33715 +
33716 +       whtmp = kzalloc(sizeof(*whtmp), gfp);
33717 +       if (unlikely(!whtmp)) {
33718 +               whtmp = ERR_PTR(-ENOMEM);
33719 +               goto out;
33720 +       }
33721 +
33722 +       /* no estimation for dir size */
33723 +       rdhash = au_sbi(sb)->si_rdhash;
33724 +       if (!rdhash)
33725 +               rdhash = AUFS_RDHASH_DEF;
33726 +       err = au_nhash_alloc(&whtmp->whlist, rdhash, gfp);
33727 +       if (unlikely(err)) {
33728 +               au_delayed_kfree(whtmp);
33729 +               whtmp = ERR_PTR(err);
33730 +       }
33731 +
33732 +out:
33733 +       return whtmp;
33734 +}
33735 +
33736 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp)
33737 +{
33738 +       if (whtmp->br)
33739 +               au_br_put(whtmp->br);
33740 +       dput(whtmp->wh_dentry);
33741 +       iput(whtmp->dir);
33742 +       au_nhash_wh_free(&whtmp->whlist);
33743 +       au_delayed_kfree(whtmp);
33744 +}
33745 +
33746 +/*
33747 + * rmdir the whiteouted temporary named dir @h_dentry.
33748 + * @whlist: whiteouted children.
33749 + */
33750 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
33751 +                  struct dentry *wh_dentry, struct au_nhash *whlist)
33752 +{
33753 +       int err;
33754 +       unsigned int h_nlink;
33755 +       struct path h_tmp;
33756 +       struct inode *wh_inode, *h_dir;
33757 +       struct au_branch *br;
33758 +
33759 +       h_dir = d_inode(wh_dentry->d_parent); /* dir inode is locked */
33760 +       IMustLock(h_dir);
33761 +
33762 +       br = au_sbr(dir->i_sb, bindex);
33763 +       wh_inode = d_inode(wh_dentry);
33764 +       inode_lock_nested(wh_inode, AuLsc_I_CHILD);
33765 +
33766 +       /*
33767 +        * someone else might change some whiteouts while we were sleeping.
33768 +        * it means this whlist may have an obsoleted entry.
33769 +        */
33770 +       if (!au_test_h_perm_sio(wh_inode, MAY_EXEC | MAY_WRITE))
33771 +               err = del_wh_children(wh_dentry, whlist, bindex, br);
33772 +       else {
33773 +               int wkq_err;
33774 +               struct del_wh_children_args args = {
33775 +                       .errp           = &err,
33776 +                       .h_dentry       = wh_dentry,
33777 +                       .whlist         = whlist,
33778 +                       .bindex         = bindex,
33779 +                       .br             = br
33780 +               };
33781 +
33782 +               wkq_err = au_wkq_wait(call_del_wh_children, &args);
33783 +               if (unlikely(wkq_err))
33784 +                       err = wkq_err;
33785 +       }
33786 +       inode_unlock(wh_inode);
33787 +
33788 +       if (!err) {
33789 +               h_tmp.dentry = wh_dentry;
33790 +               h_tmp.mnt = au_br_mnt(br);
33791 +               h_nlink = h_dir->i_nlink;
33792 +               err = vfsub_rmdir(h_dir, &h_tmp);
33793 +               /* some fs doesn't change the parent nlink in some cases */
33794 +               h_nlink -= h_dir->i_nlink;
33795 +       }
33796 +
33797 +       if (!err) {
33798 +               if (au_ibtop(dir) == bindex) {
33799 +                       /* todo: dir->i_mutex is necessary */
33800 +                       au_cpup_attr_timesizes(dir);
33801 +                       if (h_nlink)
33802 +                               vfsub_drop_nlink(dir);
33803 +               }
33804 +               return 0; /* success */
33805 +       }
33806 +
33807 +       pr_warn("failed removing %pd(%d), ignored\n", wh_dentry, err);
33808 +       return err;
33809 +}
33810 +
33811 +static void call_rmdir_whtmp(void *args)
33812 +{
33813 +       int err;
33814 +       aufs_bindex_t bindex;
33815 +       struct au_whtmp_rmdir *a = args;
33816 +       struct super_block *sb;
33817 +       struct dentry *h_parent;
33818 +       struct inode *h_dir;
33819 +       struct au_hinode *hdir;
33820 +
33821 +       /* rmdir by nfsd may cause deadlock with this i_mutex */
33822 +       /* inode_lock(a->dir); */
33823 +       err = -EROFS;
33824 +       sb = a->dir->i_sb;
33825 +       si_read_lock(sb, !AuLock_FLUSH);
33826 +       if (!au_br_writable(a->br->br_perm))
33827 +               goto out;
33828 +       bindex = au_br_index(sb, a->br->br_id);
33829 +       if (unlikely(bindex < 0))
33830 +               goto out;
33831 +
33832 +       err = -EIO;
33833 +       ii_write_lock_parent(a->dir);
33834 +       h_parent = dget_parent(a->wh_dentry);
33835 +       h_dir = d_inode(h_parent);
33836 +       hdir = au_hi(a->dir, bindex);
33837 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
33838 +       if (unlikely(err))
33839 +               goto out_mnt;
33840 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
33841 +       err = au_h_verify(a->wh_dentry, au_opt_udba(sb), h_dir, h_parent,
33842 +                         a->br);
33843 +       if (!err)
33844 +               err = au_whtmp_rmdir(a->dir, bindex, a->wh_dentry, &a->whlist);
33845 +       au_hn_inode_unlock(hdir);
33846 +       vfsub_mnt_drop_write(au_br_mnt(a->br));
33847 +
33848 +out_mnt:
33849 +       dput(h_parent);
33850 +       ii_write_unlock(a->dir);
33851 +out:
33852 +       /* inode_unlock(a->dir); */
33853 +       au_whtmp_rmdir_free(a);
33854 +       si_read_unlock(sb);
33855 +       au_nwt_done(&au_sbi(sb)->si_nowait);
33856 +       if (unlikely(err))
33857 +               AuIOErr("err %d\n", err);
33858 +}
33859 +
33860 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
33861 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args)
33862 +{
33863 +       int wkq_err;
33864 +       struct super_block *sb;
33865 +
33866 +       IMustLock(dir);
33867 +
33868 +       /* all post-process will be done in do_rmdir_whtmp(). */
33869 +       sb = dir->i_sb;
33870 +       args->dir = au_igrab(dir);
33871 +       args->br = au_sbr(sb, bindex);
33872 +       au_br_get(args->br);
33873 +       args->wh_dentry = dget(wh_dentry);
33874 +       wkq_err = au_wkq_nowait(call_rmdir_whtmp, args, sb, /*flags*/0);
33875 +       if (unlikely(wkq_err)) {
33876 +               pr_warn("rmdir error %pd (%d), ignored\n", wh_dentry, wkq_err);
33877 +               au_whtmp_rmdir_free(args);
33878 +       }
33879 +}
33880 diff -urN /usr/share/empty/fs/aufs/whout.h linux/fs/aufs/whout.h
33881 --- /usr/share/empty/fs/aufs/whout.h    1970-01-01 01:00:00.000000000 +0100
33882 +++ linux/fs/aufs/whout.h       2017-05-06 22:16:52.784886929 +0200
33883 @@ -0,0 +1,85 @@
33884 +/*
33885 + * Copyright (C) 2005-2017 Junjiro R. Okajima
33886 + *
33887 + * This program, aufs is free software; you can redistribute it and/or modify
33888 + * it under the terms of the GNU General Public License as published by
33889 + * the Free Software Foundation; either version 2 of the License, or
33890 + * (at your option) any later version.
33891 + *
33892 + * This program is distributed in the hope that it will be useful,
33893 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33894 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33895 + * GNU General Public License for more details.
33896 + *
33897 + * You should have received a copy of the GNU General Public License
33898 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33899 + */
33900 +
33901 +/*
33902 + * whiteout for logical deletion and opaque directory
33903 + */
33904 +
33905 +#ifndef __AUFS_WHOUT_H__
33906 +#define __AUFS_WHOUT_H__
33907 +
33908 +#ifdef __KERNEL__
33909 +
33910 +#include "dir.h"
33911 +
33912 +/* whout.c */
33913 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name);
33914 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio);
33915 +int au_diropq_test(struct dentry *h_dentry);
33916 +struct au_branch;
33917 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
33918 +                            struct qstr *prefix);
33919 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br);
33920 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
33921 +                       struct dentry *dentry);
33922 +int au_wh_init(struct au_branch *br, struct super_block *sb);
33923 +
33924 +/* diropq flags */
33925 +#define AuDiropq_CREATE        1
33926 +#define au_ftest_diropq(flags, name)   ((flags) & AuDiropq_##name)
33927 +#define au_fset_diropq(flags, name) \
33928 +       do { (flags) |= AuDiropq_##name; } while (0)
33929 +#define au_fclr_diropq(flags, name) \
33930 +       do { (flags) &= ~AuDiropq_##name; } while (0)
33931 +
33932 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
33933 +                            unsigned int flags);
33934 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
33935 +                         struct au_branch *br);
33936 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
33937 +                           struct dentry *h_parent);
33938 +
33939 +/* real rmdir for the whiteout-ed dir */
33940 +struct au_whtmp_rmdir {
33941 +       struct inode *dir;
33942 +       struct au_branch *br;
33943 +       struct dentry *wh_dentry;
33944 +       struct au_nhash whlist;
33945 +};
33946 +
33947 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp);
33948 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp);
33949 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
33950 +                  struct dentry *wh_dentry, struct au_nhash *whlist);
33951 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
33952 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args);
33953 +
33954 +/* ---------------------------------------------------------------------- */
33955 +
33956 +static inline struct dentry *au_diropq_create(struct dentry *dentry,
33957 +                                             aufs_bindex_t bindex)
33958 +{
33959 +       return au_diropq_sio(dentry, bindex, AuDiropq_CREATE);
33960 +}
33961 +
33962 +static inline int au_diropq_remove(struct dentry *dentry, aufs_bindex_t bindex)
33963 +{
33964 +       return PTR_ERR(au_diropq_sio(dentry, bindex, !AuDiropq_CREATE));
33965 +}
33966 +
33967 +#endif /* __KERNEL__ */
33968 +#endif /* __AUFS_WHOUT_H__ */
33969 diff -urN /usr/share/empty/fs/aufs/wkq.c linux/fs/aufs/wkq.c
33970 --- /usr/share/empty/fs/aufs/wkq.c      1970-01-01 01:00:00.000000000 +0100
33971 +++ linux/fs/aufs/wkq.c 2017-05-06 22:16:52.784886929 +0200
33972 @@ -0,0 +1,213 @@
33973 +/*
33974 + * Copyright (C) 2005-2017 Junjiro R. Okajima
33975 + *
33976 + * This program, aufs is free software; you can redistribute it and/or modify
33977 + * it under the terms of the GNU General Public License as published by
33978 + * the Free Software Foundation; either version 2 of the License, or
33979 + * (at your option) any later version.
33980 + *
33981 + * This program is distributed in the hope that it will be useful,
33982 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33983 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33984 + * GNU General Public License for more details.
33985 + *
33986 + * You should have received a copy of the GNU General Public License
33987 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33988 + */
33989 +
33990 +/*
33991 + * workqueue for asynchronous/super-io operations
33992 + * todo: try new dredential scheme
33993 + */
33994 +
33995 +#include <linux/module.h>
33996 +#include "aufs.h"
33997 +
33998 +/* internal workqueue named AUFS_WKQ_NAME */
33999 +
34000 +static struct workqueue_struct *au_wkq;
34001 +
34002 +struct au_wkinfo {
34003 +       struct work_struct wk;
34004 +       struct kobject *kobj;
34005 +
34006 +       unsigned int flags; /* see wkq.h */
34007 +
34008 +       au_wkq_func_t func;
34009 +       void *args;
34010 +
34011 +       struct completion *comp;
34012 +};
34013 +
34014 +/* ---------------------------------------------------------------------- */
34015 +
34016 +static void wkq_func(struct work_struct *wk)
34017 +{
34018 +       struct au_wkinfo *wkinfo = container_of(wk, struct au_wkinfo, wk);
34019 +
34020 +       AuDebugOn(!uid_eq(current_fsuid(), GLOBAL_ROOT_UID));
34021 +       AuDebugOn(rlimit(RLIMIT_FSIZE) != RLIM_INFINITY);
34022 +
34023 +       wkinfo->func(wkinfo->args);
34024 +       if (au_ftest_wkq(wkinfo->flags, WAIT))
34025 +               complete(wkinfo->comp);
34026 +       else {
34027 +               kobject_put(wkinfo->kobj);
34028 +               module_put(THIS_MODULE); /* todo: ?? */
34029 +               au_delayed_kfree(wkinfo);
34030 +       }
34031 +}
34032 +
34033 +/*
34034 + * Since struct completion is large, try allocating it dynamically.
34035 + */
34036 +#if 1 /* defined(CONFIG_4KSTACKS) || defined(AuTest4KSTACKS) */
34037 +#define AuWkqCompDeclare(name) struct completion *comp = NULL
34038 +
34039 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
34040 +{
34041 +       *comp = kmalloc(sizeof(**comp), GFP_NOFS);
34042 +       if (*comp) {
34043 +               init_completion(*comp);
34044 +               wkinfo->comp = *comp;
34045 +               return 0;
34046 +       }
34047 +       return -ENOMEM;
34048 +}
34049 +
34050 +static void au_wkq_comp_free(struct completion *comp)
34051 +{
34052 +       au_delayed_kfree(comp);
34053 +}
34054 +
34055 +#else
34056 +
34057 +/* no braces */
34058 +#define AuWkqCompDeclare(name) \
34059 +       DECLARE_COMPLETION_ONSTACK(_ ## name); \
34060 +       struct completion *comp = &_ ## name
34061 +
34062 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
34063 +{
34064 +       wkinfo->comp = *comp;
34065 +       return 0;
34066 +}
34067 +
34068 +static void au_wkq_comp_free(struct completion *comp __maybe_unused)
34069 +{
34070 +       /* empty */
34071 +}
34072 +#endif /* 4KSTACKS */
34073 +
34074 +static void au_wkq_run(struct au_wkinfo *wkinfo)
34075 +{
34076 +       if (au_ftest_wkq(wkinfo->flags, NEST)) {
34077 +               if (au_wkq_test()) {
34078 +                       AuWarn1("wkq from wkq, unless silly-rename on NFS,"
34079 +                               " due to a dead dir by UDBA?\n");
34080 +                       AuDebugOn(au_ftest_wkq(wkinfo->flags, WAIT));
34081 +               }
34082 +       } else
34083 +               au_dbg_verify_kthread();
34084 +
34085 +       if (au_ftest_wkq(wkinfo->flags, WAIT)) {
34086 +               INIT_WORK_ONSTACK(&wkinfo->wk, wkq_func);
34087 +               queue_work(au_wkq, &wkinfo->wk);
34088 +       } else {
34089 +               INIT_WORK(&wkinfo->wk, wkq_func);
34090 +               schedule_work(&wkinfo->wk);
34091 +       }
34092 +}
34093 +
34094 +/*
34095 + * Be careful. It is easy to make deadlock happen.
34096 + * processA: lock, wkq and wait
34097 + * processB: wkq and wait, lock in wkq
34098 + * --> deadlock
34099 + */
34100 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args)
34101 +{
34102 +       int err;
34103 +       AuWkqCompDeclare(comp);
34104 +       struct au_wkinfo wkinfo = {
34105 +               .flags  = flags,
34106 +               .func   = func,
34107 +               .args   = args
34108 +       };
34109 +
34110 +       err = au_wkq_comp_alloc(&wkinfo, &comp);
34111 +       if (!err) {
34112 +               au_wkq_run(&wkinfo);
34113 +               /* no timeout, no interrupt */
34114 +               wait_for_completion(wkinfo.comp);
34115 +               au_wkq_comp_free(comp);
34116 +               destroy_work_on_stack(&wkinfo.wk);
34117 +       }
34118 +
34119 +       return err;
34120 +
34121 +}
34122 +
34123 +/*
34124 + * Note: dget/dput() in func for aufs dentries are not supported. It will be a
34125 + * problem in a concurrent umounting.
34126 + */
34127 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
34128 +                 unsigned int flags)
34129 +{
34130 +       int err;
34131 +       struct au_wkinfo *wkinfo;
34132 +
34133 +       atomic_inc(&au_sbi(sb)->si_nowait.nw_len);
34134 +
34135 +       /*
34136 +        * wkq_func() must free this wkinfo.
34137 +        * it highly depends upon the implementation of workqueue.
34138 +        */
34139 +       err = 0;
34140 +       wkinfo = kmalloc(sizeof(*wkinfo), GFP_NOFS);
34141 +       if (wkinfo) {
34142 +               wkinfo->kobj = &au_sbi(sb)->si_kobj;
34143 +               wkinfo->flags = flags & ~AuWkq_WAIT;
34144 +               wkinfo->func = func;
34145 +               wkinfo->args = args;
34146 +               wkinfo->comp = NULL;
34147 +               kobject_get(wkinfo->kobj);
34148 +               __module_get(THIS_MODULE); /* todo: ?? */
34149 +
34150 +               au_wkq_run(wkinfo);
34151 +       } else {
34152 +               err = -ENOMEM;
34153 +               au_nwt_done(&au_sbi(sb)->si_nowait);
34154 +       }
34155 +
34156 +       return err;
34157 +}
34158 +
34159 +/* ---------------------------------------------------------------------- */
34160 +
34161 +void au_nwt_init(struct au_nowait_tasks *nwt)
34162 +{
34163 +       atomic_set(&nwt->nw_len, 0);
34164 +       /* smp_mb(); */ /* atomic_set */
34165 +       init_waitqueue_head(&nwt->nw_wq);
34166 +}
34167 +
34168 +void au_wkq_fin(void)
34169 +{
34170 +       destroy_workqueue(au_wkq);
34171 +}
34172 +
34173 +int __init au_wkq_init(void)
34174 +{
34175 +       int err;
34176 +
34177 +       err = 0;
34178 +       au_wkq = alloc_workqueue(AUFS_WKQ_NAME, 0, WQ_DFL_ACTIVE);
34179 +       if (IS_ERR(au_wkq))
34180 +               err = PTR_ERR(au_wkq);
34181 +       else if (!au_wkq)
34182 +               err = -ENOMEM;
34183 +
34184 +       return err;
34185 +}
34186 diff -urN /usr/share/empty/fs/aufs/wkq.h linux/fs/aufs/wkq.h
34187 --- /usr/share/empty/fs/aufs/wkq.h      1970-01-01 01:00:00.000000000 +0100
34188 +++ linux/fs/aufs/wkq.h 2017-05-06 22:16:52.784886929 +0200
34189 @@ -0,0 +1,93 @@
34190 +/*
34191 + * Copyright (C) 2005-2017 Junjiro R. Okajima
34192 + *
34193 + * This program, aufs is free software; you can redistribute it and/or modify
34194 + * it under the terms of the GNU General Public License as published by
34195 + * the Free Software Foundation; either version 2 of the License, or
34196 + * (at your option) any later version.
34197 + *
34198 + * This program is distributed in the hope that it will be useful,
34199 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
34200 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
34201 + * GNU General Public License for more details.
34202 + *
34203 + * You should have received a copy of the GNU General Public License
34204 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
34205 + */
34206 +
34207 +/*
34208 + * workqueue for asynchronous/super-io operations
34209 + * todo: try new credentials management scheme
34210 + */
34211 +
34212 +#ifndef __AUFS_WKQ_H__
34213 +#define __AUFS_WKQ_H__
34214 +
34215 +#ifdef __KERNEL__
34216 +
34217 +#include <linux/percpu_counter.h>
34218 +
34219 +struct super_block;
34220 +
34221 +/* ---------------------------------------------------------------------- */
34222 +
34223 +/*
34224 + * in the next operation, wait for the 'nowait' tasks in system-wide workqueue
34225 + */
34226 +struct au_nowait_tasks {
34227 +       atomic_t                nw_len;
34228 +       wait_queue_head_t       nw_wq;
34229 +};
34230 +
34231 +/* ---------------------------------------------------------------------- */
34232 +
34233 +typedef void (*au_wkq_func_t)(void *args);
34234 +
34235 +/* wkq flags */
34236 +#define AuWkq_WAIT     1
34237 +#define AuWkq_NEST     (1 << 1)
34238 +#define au_ftest_wkq(flags, name)      ((flags) & AuWkq_##name)
34239 +#define au_fset_wkq(flags, name) \
34240 +       do { (flags) |= AuWkq_##name; } while (0)
34241 +#define au_fclr_wkq(flags, name) \
34242 +       do { (flags) &= ~AuWkq_##name; } while (0)
34243 +
34244 +#ifndef CONFIG_AUFS_HNOTIFY
34245 +#undef AuWkq_NEST
34246 +#define AuWkq_NEST     0
34247 +#endif
34248 +
34249 +/* wkq.c */
34250 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args);
34251 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
34252 +                 unsigned int flags);
34253 +void au_nwt_init(struct au_nowait_tasks *nwt);
34254 +int __init au_wkq_init(void);
34255 +void au_wkq_fin(void);
34256 +
34257 +/* ---------------------------------------------------------------------- */
34258 +
34259 +static inline int au_wkq_test(void)
34260 +{
34261 +       return current->flags & PF_WQ_WORKER;
34262 +}
34263 +
34264 +static inline int au_wkq_wait(au_wkq_func_t func, void *args)
34265 +{
34266 +       return au_wkq_do_wait(AuWkq_WAIT, func, args);
34267 +}
34268 +
34269 +static inline void au_nwt_done(struct au_nowait_tasks *nwt)
34270 +{
34271 +       if (atomic_dec_and_test(&nwt->nw_len))
34272 +               wake_up_all(&nwt->nw_wq);
34273 +}
34274 +
34275 +static inline int au_nwt_flush(struct au_nowait_tasks *nwt)
34276 +{
34277 +       wait_event(nwt->nw_wq, !atomic_read(&nwt->nw_len));
34278 +       return 0;
34279 +}
34280 +
34281 +#endif /* __KERNEL__ */
34282 +#endif /* __AUFS_WKQ_H__ */
34283 diff -urN /usr/share/empty/fs/aufs/xattr.c linux/fs/aufs/xattr.c
34284 --- /usr/share/empty/fs/aufs/xattr.c    1970-01-01 01:00:00.000000000 +0100
34285 +++ linux/fs/aufs/xattr.c       2017-05-06 22:16:52.784886929 +0200
34286 @@ -0,0 +1,357 @@
34287 +/*
34288 + * Copyright (C) 2014-2017 Junjiro R. Okajima
34289 + *
34290 + * This program, aufs is free software; you can redistribute it and/or modify
34291 + * it under the terms of the GNU General Public License as published by
34292 + * the Free Software Foundation; either version 2 of the License, or
34293 + * (at your option) any later version.
34294 + *
34295 + * This program is distributed in the hope that it will be useful,
34296 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
34297 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
34298 + * GNU General Public License for more details.
34299 + *
34300 + * You should have received a copy of the GNU General Public License
34301 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
34302 + */
34303 +
34304 +/*
34305 + * handling xattr functions
34306 + */
34307 +
34308 +#include <linux/fs.h>
34309 +#include <linux/posix_acl_xattr.h>
34310 +#include <linux/xattr.h>
34311 +#include "aufs.h"
34312 +
34313 +static int au_xattr_ignore(int err, char *name, unsigned int ignore_flags)
34314 +{
34315 +       if (!ignore_flags)
34316 +               goto out;
34317 +       switch (err) {
34318 +       case -ENOMEM:
34319 +       case -EDQUOT:
34320 +               goto out;
34321 +       }
34322 +
34323 +       if ((ignore_flags & AuBrAttr_ICEX) == AuBrAttr_ICEX) {
34324 +               err = 0;
34325 +               goto out;
34326 +       }
34327 +
34328 +#define cmp(brattr, prefix) do {                                       \
34329 +               if (!strncmp(name, XATTR_##prefix##_PREFIX,             \
34330 +                            XATTR_##prefix##_PREFIX_LEN)) {            \
34331 +                       if (ignore_flags & AuBrAttr_ICEX_##brattr)      \
34332 +                               err = 0;                                \
34333 +                       goto out;                                       \
34334 +               }                                                       \
34335 +       } while (0)
34336 +
34337 +       cmp(SEC, SECURITY);
34338 +       cmp(SYS, SYSTEM);
34339 +       cmp(TR, TRUSTED);
34340 +       cmp(USR, USER);
34341 +#undef cmp
34342 +
34343 +       if (ignore_flags & AuBrAttr_ICEX_OTH)
34344 +               err = 0;
34345 +
34346 +out:
34347 +       return err;
34348 +}
34349 +
34350 +static const int au_xattr_out_of_list = AuBrAttr_ICEX_OTH << 1;
34351 +
34352 +static int au_do_cpup_xattr(struct dentry *h_dst, struct dentry *h_src,
34353 +                           char *name, char **buf, unsigned int ignore_flags,
34354 +                           unsigned int verbose)
34355 +{
34356 +       int err;
34357 +       ssize_t ssz;
34358 +       struct inode *h_idst;
34359 +
34360 +       ssz = vfs_getxattr_alloc(h_src, name, buf, 0, GFP_NOFS);
34361 +       err = ssz;
34362 +       if (unlikely(err <= 0)) {
34363 +               if (err == -ENODATA
34364 +                   || (err == -EOPNOTSUPP
34365 +                       && ((ignore_flags & au_xattr_out_of_list)
34366 +                           || (au_test_nfs_noacl(d_inode(h_src))
34367 +                               && (!strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS)
34368 +                                   || !strcmp(name,
34369 +                                              XATTR_NAME_POSIX_ACL_DEFAULT))))
34370 +                           ))
34371 +                       err = 0;
34372 +               if (err && (verbose || au_debug_test()))
34373 +                       pr_err("%s, err %d\n", name, err);
34374 +               goto out;
34375 +       }
34376 +
34377 +       /* unlock it temporary */
34378 +       h_idst = d_inode(h_dst);
34379 +       inode_unlock(h_idst);
34380 +       err = vfsub_setxattr(h_dst, name, *buf, ssz, /*flags*/0);
34381 +       inode_lock_nested(h_idst, AuLsc_I_CHILD2);
34382 +       if (unlikely(err)) {
34383 +               if (verbose || au_debug_test())
34384 +                       pr_err("%s, err %d\n", name, err);
34385 +               err = au_xattr_ignore(err, name, ignore_flags);
34386 +       }
34387 +
34388 +out:
34389 +       return err;
34390 +}
34391 +
34392 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags,
34393 +                 unsigned int verbose)
34394 +{
34395 +       int err, unlocked, acl_access, acl_default;
34396 +       ssize_t ssz;
34397 +       struct inode *h_isrc, *h_idst;
34398 +       char *value, *p, *o, *e;
34399 +
34400 +       /* try stopping to update the source inode while we are referencing */
34401 +       /* there should not be the parent-child relationship between them */
34402 +       h_isrc = d_inode(h_src);
34403 +       h_idst = d_inode(h_dst);
34404 +       inode_unlock(h_idst);
34405 +       inode_lock_nested(h_isrc, AuLsc_I_CHILD);
34406 +       inode_lock_nested(h_idst, AuLsc_I_CHILD2);
34407 +       unlocked = 0;
34408 +
34409 +       /* some filesystems don't list POSIX ACL, for example tmpfs */
34410 +       ssz = vfs_listxattr(h_src, NULL, 0);
34411 +       err = ssz;
34412 +       if (unlikely(err < 0)) {
34413 +               AuTraceErr(err);
34414 +               if (err == -ENODATA
34415 +                   || err == -EOPNOTSUPP)
34416 +                       err = 0;        /* ignore */
34417 +               goto out;
34418 +       }
34419 +
34420 +       err = 0;
34421 +       p = NULL;
34422 +       o = NULL;
34423 +       if (ssz) {
34424 +               err = -ENOMEM;
34425 +               p = kmalloc(ssz, GFP_NOFS);
34426 +               o = p;
34427 +               if (unlikely(!p))
34428 +                       goto out;
34429 +               err = vfs_listxattr(h_src, p, ssz);
34430 +       }
34431 +       inode_unlock(h_isrc);
34432 +       unlocked = 1;
34433 +       AuDbg("err %d, ssz %zd\n", err, ssz);
34434 +       if (unlikely(err < 0))
34435 +               goto out_free;
34436 +
34437 +       err = 0;
34438 +       e = p + ssz;
34439 +       value = NULL;
34440 +       acl_access = 0;
34441 +       acl_default = 0;
34442 +       while (!err && p < e) {
34443 +               acl_access |= !strncmp(p, XATTR_NAME_POSIX_ACL_ACCESS,
34444 +                                      sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1);
34445 +               acl_default |= !strncmp(p, XATTR_NAME_POSIX_ACL_DEFAULT,
34446 +                                       sizeof(XATTR_NAME_POSIX_ACL_DEFAULT)
34447 +                                       - 1);
34448 +               err = au_do_cpup_xattr(h_dst, h_src, p, &value, ignore_flags,
34449 +                                      verbose);
34450 +               p += strlen(p) + 1;
34451 +       }
34452 +       AuTraceErr(err);
34453 +       ignore_flags |= au_xattr_out_of_list;
34454 +       if (!err && !acl_access) {
34455 +               err = au_do_cpup_xattr(h_dst, h_src,
34456 +                                      XATTR_NAME_POSIX_ACL_ACCESS, &value,
34457 +                                      ignore_flags, verbose);
34458 +               AuTraceErr(err);
34459 +       }
34460 +       if (!err && !acl_default) {
34461 +               err = au_do_cpup_xattr(h_dst, h_src,
34462 +                                      XATTR_NAME_POSIX_ACL_DEFAULT, &value,
34463 +                                      ignore_flags, verbose);
34464 +               AuTraceErr(err);
34465 +       }
34466 +
34467 +       if (value)
34468 +               au_delayed_kfree(value);
34469 +
34470 +out_free:
34471 +       if (o)
34472 +               au_delayed_kfree(o);
34473 +out:
34474 +       if (!unlocked)
34475 +               inode_unlock(h_isrc);
34476 +       AuTraceErr(err);
34477 +       return err;
34478 +}
34479 +
34480 +/* ---------------------------------------------------------------------- */
34481 +
34482 +static int au_smack_reentering(struct super_block *sb)
34483 +{
34484 +#if IS_ENABLED(CONFIG_SECURITY_SMACK)
34485 +       /*
34486 +        * as a part of lookup, smack_d_instantiate() is called, and it calls
34487 +        * i_op->getxattr(). ouch.
34488 +        */
34489 +       return si_pid_test(sb);
34490 +#else
34491 +       return 0;
34492 +#endif
34493 +}
34494 +
34495 +enum {
34496 +       AU_XATTR_LIST,
34497 +       AU_XATTR_GET
34498 +};
34499 +
34500 +struct au_lgxattr {
34501 +       int type;
34502 +       union {
34503 +               struct {
34504 +                       char    *list;
34505 +                       size_t  size;
34506 +               } list;
34507 +               struct {
34508 +                       const char      *name;
34509 +                       void            *value;
34510 +                       size_t          size;
34511 +               } get;
34512 +       } u;
34513 +};
34514 +
34515 +static ssize_t au_lgxattr(struct dentry *dentry, struct au_lgxattr *arg)
34516 +{
34517 +       ssize_t err;
34518 +       int reenter;
34519 +       struct path h_path;
34520 +       struct super_block *sb;
34521 +
34522 +       sb = dentry->d_sb;
34523 +       reenter = au_smack_reentering(sb);
34524 +       if (!reenter) {
34525 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
34526 +               if (unlikely(err))
34527 +                       goto out;
34528 +       }
34529 +       err = au_h_path_getattr(dentry, /*force*/1, &h_path, reenter);
34530 +       if (unlikely(err))
34531 +               goto out_si;
34532 +       if (unlikely(!h_path.dentry))
34533 +               /* illegally overlapped or something */
34534 +               goto out_di; /* pretending success */
34535 +
34536 +       /* always topmost entry only */
34537 +       switch (arg->type) {
34538 +       case AU_XATTR_LIST:
34539 +               err = vfs_listxattr(h_path.dentry,
34540 +                                   arg->u.list.list, arg->u.list.size);
34541 +               break;
34542 +       case AU_XATTR_GET:
34543 +               AuDebugOn(d_is_negative(h_path.dentry));
34544 +               err = vfs_getxattr(h_path.dentry,
34545 +                                  arg->u.get.name, arg->u.get.value,
34546 +                                  arg->u.get.size);
34547 +               break;
34548 +       }
34549 +
34550 +out_di:
34551 +       if (!reenter)
34552 +               di_read_unlock(dentry, AuLock_IR);
34553 +out_si:
34554 +       if (!reenter)
34555 +               si_read_unlock(sb);
34556 +out:
34557 +       AuTraceErr(err);
34558 +       return err;
34559 +}
34560 +
34561 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size)
34562 +{
34563 +       struct au_lgxattr arg = {
34564 +               .type = AU_XATTR_LIST,
34565 +               .u.list = {
34566 +                       .list   = list,
34567 +                       .size   = size
34568 +               },
34569 +       };
34570 +
34571 +       return au_lgxattr(dentry, &arg);
34572 +}
34573 +
34574 +static ssize_t au_getxattr(struct dentry *dentry,
34575 +                          struct inode *inode __maybe_unused,
34576 +                          const char *name, void *value, size_t size)
34577 +{
34578 +       struct au_lgxattr arg = {
34579 +               .type = AU_XATTR_GET,
34580 +               .u.get = {
34581 +                       .name   = name,
34582 +                       .value  = value,
34583 +                       .size   = size
34584 +               },
34585 +       };
34586 +
34587 +       return au_lgxattr(dentry, &arg);
34588 +}
34589 +
34590 +static int au_setxattr(struct dentry *dentry, struct inode *inode,
34591 +                      const char *name, const void *value, size_t size,
34592 +                      int flags)
34593 +{
34594 +       struct au_sxattr arg = {
34595 +               .type = AU_XATTR_SET,
34596 +               .u.set = {
34597 +                       .name   = name,
34598 +                       .value  = value,
34599 +                       .size   = size,
34600 +                       .flags  = flags
34601 +               },
34602 +       };
34603 +
34604 +       return au_sxattr(dentry, inode, &arg);
34605 +}
34606 +
34607 +/* ---------------------------------------------------------------------- */
34608 +
34609 +static int au_xattr_get(const struct xattr_handler *handler,
34610 +                       struct dentry *dentry, struct inode *inode,
34611 +                       const char *name, void *buffer, size_t size)
34612 +{
34613 +       return au_getxattr(dentry, inode, name, buffer, size);
34614 +}
34615 +
34616 +static int au_xattr_set(const struct xattr_handler *handler,
34617 +                       struct dentry *dentry, struct inode *inode,
34618 +                       const char *name, const void *value, size_t size,
34619 +                       int flags)
34620 +{
34621 +       return au_setxattr(dentry, inode, name, value, size, flags);
34622 +}
34623 +
34624 +static const struct xattr_handler au_xattr_handler = {
34625 +       .name   = "",
34626 +       .prefix = "",
34627 +       .get    = au_xattr_get,
34628 +       .set    = au_xattr_set
34629 +};
34630 +
34631 +static const struct xattr_handler *au_xattr_handlers[] = {
34632 +#ifdef CONFIG_FS_POSIX_ACL
34633 +       &posix_acl_access_xattr_handler,
34634 +       &posix_acl_default_xattr_handler,
34635 +#endif
34636 +       &au_xattr_handler, /* must be last */
34637 +       NULL
34638 +};
34639 +
34640 +void au_xattr_init(struct super_block *sb)
34641 +{
34642 +       sb->s_xattr = au_xattr_handlers;
34643 +}
34644 diff -urN /usr/share/empty/fs/aufs/xino.c linux/fs/aufs/xino.c
34645 --- /usr/share/empty/fs/aufs/xino.c     1970-01-01 01:00:00.000000000 +0100
34646 +++ linux/fs/aufs/xino.c        2017-05-06 22:16:52.784886929 +0200
34647 @@ -0,0 +1,1415 @@
34648 +/*
34649 + * Copyright (C) 2005-2017 Junjiro R. Okajima
34650 + *
34651 + * This program, aufs is free software; you can redistribute it and/or modify
34652 + * it under the terms of the GNU General Public License as published by
34653 + * the Free Software Foundation; either version 2 of the License, or
34654 + * (at your option) any later version.
34655 + *
34656 + * This program is distributed in the hope that it will be useful,
34657 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
34658 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
34659 + * GNU General Public License for more details.
34660 + *
34661 + * You should have received a copy of the GNU General Public License
34662 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
34663 + */
34664 +
34665 +/*
34666 + * external inode number translation table and bitmap
34667 + */
34668 +
34669 +#include <linux/seq_file.h>
34670 +#include <linux/statfs.h>
34671 +#include "aufs.h"
34672 +
34673 +/* todo: unnecessary to support mmap_sem since kernel-space? */
34674 +ssize_t xino_fread(vfs_readf_t func, struct file *file, void *kbuf, size_t size,
34675 +                  loff_t *pos)
34676 +{
34677 +       ssize_t err;
34678 +       mm_segment_t oldfs;
34679 +       union {
34680 +               void *k;
34681 +               char __user *u;
34682 +       } buf;
34683 +
34684 +       buf.k = kbuf;
34685 +       oldfs = get_fs();
34686 +       set_fs(KERNEL_DS);
34687 +       do {
34688 +               /* todo: signal_pending? */
34689 +               err = func(file, buf.u, size, pos);
34690 +       } while (err == -EAGAIN || err == -EINTR);
34691 +       set_fs(oldfs);
34692 +
34693 +#if 0 /* reserved for future use */
34694 +       if (err > 0)
34695 +               fsnotify_access(file->f_path.dentry);
34696 +#endif
34697 +
34698 +       return err;
34699 +}
34700 +
34701 +/* ---------------------------------------------------------------------- */
34702 +
34703 +static ssize_t xino_fwrite_wkq(vfs_writef_t func, struct file *file, void *buf,
34704 +                              size_t size, loff_t *pos);
34705 +
34706 +static ssize_t do_xino_fwrite(vfs_writef_t func, struct file *file, void *kbuf,
34707 +                             size_t size, loff_t *pos)
34708 +{
34709 +       ssize_t err;
34710 +       mm_segment_t oldfs;
34711 +       union {
34712 +               void *k;
34713 +               const char __user *u;
34714 +       } buf;
34715 +       int i;
34716 +       const int prevent_endless = 10;
34717 +
34718 +       i = 0;
34719 +       buf.k = kbuf;
34720 +       oldfs = get_fs();
34721 +       set_fs(KERNEL_DS);
34722 +       do {
34723 +               err = func(file, buf.u, size, pos);
34724 +               if (err == -EINTR
34725 +                   && !au_wkq_test()
34726 +                   && fatal_signal_pending(current)) {
34727 +                       set_fs(oldfs);
34728 +                       err = xino_fwrite_wkq(func, file, kbuf, size, pos);
34729 +                       BUG_ON(err == -EINTR);
34730 +                       oldfs = get_fs();
34731 +                       set_fs(KERNEL_DS);
34732 +               }
34733 +       } while (i++ < prevent_endless
34734 +                && (err == -EAGAIN || err == -EINTR));
34735 +       set_fs(oldfs);
34736 +
34737 +#if 0 /* reserved for future use */
34738 +       if (err > 0)
34739 +               fsnotify_modify(file->f_path.dentry);
34740 +#endif
34741 +
34742 +       return err;
34743 +}
34744 +
34745 +struct do_xino_fwrite_args {
34746 +       ssize_t *errp;
34747 +       vfs_writef_t func;
34748 +       struct file *file;
34749 +       void *buf;
34750 +       size_t size;
34751 +       loff_t *pos;
34752 +};
34753 +
34754 +static void call_do_xino_fwrite(void *args)
34755 +{
34756 +       struct do_xino_fwrite_args *a = args;
34757 +       *a->errp = do_xino_fwrite(a->func, a->file, a->buf, a->size, a->pos);
34758 +}
34759 +
34760 +static ssize_t xino_fwrite_wkq(vfs_writef_t func, struct file *file, void *buf,
34761 +                              size_t size, loff_t *pos)
34762 +{
34763 +       ssize_t err;
34764 +       int wkq_err;
34765 +       struct do_xino_fwrite_args args = {
34766 +               .errp   = &err,
34767 +               .func   = func,
34768 +               .file   = file,
34769 +               .buf    = buf,
34770 +               .size   = size,
34771 +               .pos    = pos
34772 +       };
34773 +
34774 +       /*
34775 +        * it breaks RLIMIT_FSIZE and normal user's limit,
34776 +        * users should care about quota and real 'filesystem full.'
34777 +        */
34778 +       wkq_err = au_wkq_wait(call_do_xino_fwrite, &args);
34779 +       if (unlikely(wkq_err))
34780 +               err = wkq_err;
34781 +
34782 +       return err;
34783 +}
34784 +
34785 +ssize_t xino_fwrite(vfs_writef_t func, struct file *file, void *buf,
34786 +                   size_t size, loff_t *pos)
34787 +{
34788 +       ssize_t err;
34789 +
34790 +       if (rlimit(RLIMIT_FSIZE) == RLIM_INFINITY) {
34791 +               lockdep_off();
34792 +               err = do_xino_fwrite(func, file, buf, size, pos);
34793 +               lockdep_on();
34794 +       } else
34795 +               err = xino_fwrite_wkq(func, file, buf, size, pos);
34796 +
34797 +       return err;
34798 +}
34799 +
34800 +/* ---------------------------------------------------------------------- */
34801 +
34802 +/*
34803 + * create a new xinofile at the same place/path as @base_file.
34804 + */
34805 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src)
34806 +{
34807 +       struct file *file;
34808 +       struct dentry *base, *parent;
34809 +       struct inode *dir, *delegated;
34810 +       struct qstr *name;
34811 +       struct path path;
34812 +       int err;
34813 +
34814 +       base = base_file->f_path.dentry;
34815 +       parent = base->d_parent; /* dir inode is locked */
34816 +       dir = d_inode(parent);
34817 +       IMustLock(dir);
34818 +
34819 +       file = ERR_PTR(-EINVAL);
34820 +       name = &base->d_name;
34821 +       path.dentry = vfsub_lookup_one_len(name->name, parent, name->len);
34822 +       if (IS_ERR(path.dentry)) {
34823 +               file = (void *)path.dentry;
34824 +               pr_err("%pd lookup err %ld\n",
34825 +                      base, PTR_ERR(path.dentry));
34826 +               goto out;
34827 +       }
34828 +
34829 +       /* no need to mnt_want_write() since we call dentry_open() later */
34830 +       err = vfs_create(dir, path.dentry, S_IRUGO | S_IWUGO, NULL);
34831 +       if (unlikely(err)) {
34832 +               file = ERR_PTR(err);
34833 +               pr_err("%pd create err %d\n", base, err);
34834 +               goto out_dput;
34835 +       }
34836 +
34837 +       path.mnt = base_file->f_path.mnt;
34838 +       file = vfsub_dentry_open(&path,
34839 +                                O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
34840 +                                /* | __FMODE_NONOTIFY */);
34841 +       if (IS_ERR(file)) {
34842 +               pr_err("%pd open err %ld\n", base, PTR_ERR(file));
34843 +               goto out_dput;
34844 +       }
34845 +
34846 +       delegated = NULL;
34847 +       err = vfsub_unlink(dir, &file->f_path, &delegated, /*force*/0);
34848 +       if (unlikely(err == -EWOULDBLOCK)) {
34849 +               pr_warn("cannot retry for NFSv4 delegation"
34850 +                       " for an internal unlink\n");
34851 +               iput(delegated);
34852 +       }
34853 +       if (unlikely(err)) {
34854 +               pr_err("%pd unlink err %d\n", base, err);
34855 +               goto out_fput;
34856 +       }
34857 +
34858 +       if (copy_src) {
34859 +               /* no one can touch copy_src xino */
34860 +               err = au_copy_file(file, copy_src, vfsub_f_size_read(copy_src));
34861 +               if (unlikely(err)) {
34862 +                       pr_err("%pd copy err %d\n", base, err);
34863 +                       goto out_fput;
34864 +               }
34865 +       }
34866 +       goto out_dput; /* success */
34867 +
34868 +out_fput:
34869 +       fput(file);
34870 +       file = ERR_PTR(err);
34871 +out_dput:
34872 +       dput(path.dentry);
34873 +out:
34874 +       return file;
34875 +}
34876 +
34877 +struct au_xino_lock_dir {
34878 +       struct au_hinode *hdir;
34879 +       struct dentry *parent;
34880 +       struct inode *dir;
34881 +};
34882 +
34883 +static void au_xino_lock_dir(struct super_block *sb, struct file *xino,
34884 +                            struct au_xino_lock_dir *ldir)
34885 +{
34886 +       aufs_bindex_t brid, bindex;
34887 +
34888 +       ldir->hdir = NULL;
34889 +       bindex = -1;
34890 +       brid = au_xino_brid(sb);
34891 +       if (brid >= 0)
34892 +               bindex = au_br_index(sb, brid);
34893 +       if (bindex >= 0) {
34894 +               ldir->hdir = au_hi(d_inode(sb->s_root), bindex);
34895 +               au_hn_inode_lock_nested(ldir->hdir, AuLsc_I_PARENT);
34896 +       } else {
34897 +               ldir->parent = dget_parent(xino->f_path.dentry);
34898 +               ldir->dir = d_inode(ldir->parent);
34899 +               inode_lock_nested(ldir->dir, AuLsc_I_PARENT);
34900 +       }
34901 +}
34902 +
34903 +static void au_xino_unlock_dir(struct au_xino_lock_dir *ldir)
34904 +{
34905 +       if (ldir->hdir)
34906 +               au_hn_inode_unlock(ldir->hdir);
34907 +       else {
34908 +               inode_unlock(ldir->dir);
34909 +               dput(ldir->parent);
34910 +       }
34911 +}
34912 +
34913 +/* ---------------------------------------------------------------------- */
34914 +
34915 +/* trucate xino files asynchronously */
34916 +
34917 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex)
34918 +{
34919 +       int err;
34920 +       unsigned long jiffy;
34921 +       blkcnt_t blocks;
34922 +       aufs_bindex_t bi, bbot;
34923 +       struct kstatfs *st;
34924 +       struct au_branch *br;
34925 +       struct file *new_xino, *file;
34926 +       struct super_block *h_sb;
34927 +       struct au_xino_lock_dir ldir;
34928 +
34929 +       err = -ENOMEM;
34930 +       st = kmalloc(sizeof(*st), GFP_NOFS);
34931 +       if (unlikely(!st))
34932 +               goto out;
34933 +
34934 +       err = -EINVAL;
34935 +       bbot = au_sbbot(sb);
34936 +       if (unlikely(bindex < 0 || bbot < bindex))
34937 +               goto out_st;
34938 +       br = au_sbr(sb, bindex);
34939 +       file = br->br_xino.xi_file;
34940 +       if (!file)
34941 +               goto out_st;
34942 +
34943 +       err = vfs_statfs(&file->f_path, st);
34944 +       if (unlikely(err))
34945 +               AuErr1("statfs err %d, ignored\n", err);
34946 +       jiffy = jiffies;
34947 +       blocks = file_inode(file)->i_blocks;
34948 +       pr_info("begin truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
34949 +               bindex, (u64)blocks, st->f_bfree, st->f_blocks);
34950 +
34951 +       au_xino_lock_dir(sb, file, &ldir);
34952 +       /* mnt_want_write() is unnecessary here */
34953 +       new_xino = au_xino_create2(file, file);
34954 +       au_xino_unlock_dir(&ldir);
34955 +       err = PTR_ERR(new_xino);
34956 +       if (IS_ERR(new_xino)) {
34957 +               pr_err("err %d, ignored\n", err);
34958 +               goto out_st;
34959 +       }
34960 +       err = 0;
34961 +       fput(file);
34962 +       br->br_xino.xi_file = new_xino;
34963 +
34964 +       h_sb = au_br_sb(br);
34965 +       for (bi = 0; bi <= bbot; bi++) {
34966 +               if (unlikely(bi == bindex))
34967 +                       continue;
34968 +               br = au_sbr(sb, bi);
34969 +               if (au_br_sb(br) != h_sb)
34970 +                       continue;
34971 +
34972 +               fput(br->br_xino.xi_file);
34973 +               br->br_xino.xi_file = new_xino;
34974 +               get_file(new_xino);
34975 +       }
34976 +
34977 +       err = vfs_statfs(&new_xino->f_path, st);
34978 +       if (!err) {
34979 +               pr_info("end truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
34980 +                       bindex, (u64)file_inode(new_xino)->i_blocks,
34981 +                       st->f_bfree, st->f_blocks);
34982 +               if (file_inode(new_xino)->i_blocks < blocks)
34983 +                       au_sbi(sb)->si_xino_jiffy = jiffy;
34984 +       } else
34985 +               AuErr1("statfs err %d, ignored\n", err);
34986 +
34987 +out_st:
34988 +       au_delayed_kfree(st);
34989 +out:
34990 +       return err;
34991 +}
34992 +
34993 +struct xino_do_trunc_args {
34994 +       struct super_block *sb;
34995 +       struct au_branch *br;
34996 +};
34997 +
34998 +static void xino_do_trunc(void *_args)
34999 +{
35000 +       struct xino_do_trunc_args *args = _args;
35001 +       struct super_block *sb;
35002 +       struct au_branch *br;
35003 +       struct inode *dir;
35004 +       int err;
35005 +       aufs_bindex_t bindex;
35006 +
35007 +       err = 0;
35008 +       sb = args->sb;
35009 +       dir = d_inode(sb->s_root);
35010 +       br = args->br;
35011 +
35012 +       si_noflush_write_lock(sb);
35013 +       ii_read_lock_parent(dir);
35014 +       bindex = au_br_index(sb, br->br_id);
35015 +       err = au_xino_trunc(sb, bindex);
35016 +       ii_read_unlock(dir);
35017 +       if (unlikely(err))
35018 +               pr_warn("err b%d, (%d)\n", bindex, err);
35019 +       atomic_dec(&br->br_xino_running);
35020 +       au_br_put(br);
35021 +       si_write_unlock(sb);
35022 +       au_nwt_done(&au_sbi(sb)->si_nowait);
35023 +       au_delayed_kfree(args);
35024 +}
35025 +
35026 +static int xino_trunc_test(struct super_block *sb, struct au_branch *br)
35027 +{
35028 +       int err;
35029 +       struct kstatfs st;
35030 +       struct au_sbinfo *sbinfo;
35031 +
35032 +       /* todo: si_xino_expire and the ratio should be customizable */
35033 +       sbinfo = au_sbi(sb);
35034 +       if (time_before(jiffies,
35035 +                       sbinfo->si_xino_jiffy + sbinfo->si_xino_expire))
35036 +               return 0;
35037 +
35038 +       /* truncation border */
35039 +       err = vfs_statfs(&br->br_xino.xi_file->f_path, &st);
35040 +       if (unlikely(err)) {
35041 +               AuErr1("statfs err %d, ignored\n", err);
35042 +               return 0;
35043 +       }
35044 +       if (div64_u64(st.f_bfree * 100, st.f_blocks) >= AUFS_XINO_DEF_TRUNC)
35045 +               return 0;
35046 +
35047 +       return 1;
35048 +}
35049 +
35050 +static void xino_try_trunc(struct super_block *sb, struct au_branch *br)
35051 +{
35052 +       struct xino_do_trunc_args *args;
35053 +       int wkq_err;
35054 +
35055 +       if (!xino_trunc_test(sb, br))
35056 +               return;
35057 +
35058 +       if (atomic_inc_return(&br->br_xino_running) > 1)
35059 +               goto out;
35060 +
35061 +       /* lock and kfree() will be called in trunc_xino() */
35062 +       args = kmalloc(sizeof(*args), GFP_NOFS);
35063 +       if (unlikely(!args)) {
35064 +               AuErr1("no memory\n");
35065 +               goto out;
35066 +       }
35067 +
35068 +       au_br_get(br);
35069 +       args->sb = sb;
35070 +       args->br = br;
35071 +       wkq_err = au_wkq_nowait(xino_do_trunc, args, sb, /*flags*/0);
35072 +       if (!wkq_err)
35073 +               return; /* success */
35074 +
35075 +       pr_err("wkq %d\n", wkq_err);
35076 +       au_br_put(br);
35077 +       au_delayed_kfree(args);
35078 +
35079 +out:
35080 +       atomic_dec(&br->br_xino_running);
35081 +}
35082 +
35083 +/* ---------------------------------------------------------------------- */
35084 +
35085 +static int au_xino_do_write(vfs_writef_t write, struct file *file,
35086 +                           ino_t h_ino, ino_t ino)
35087 +{
35088 +       loff_t pos;
35089 +       ssize_t sz;
35090 +
35091 +       pos = h_ino;
35092 +       if (unlikely(au_loff_max / sizeof(ino) - 1 < pos)) {
35093 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
35094 +               return -EFBIG;
35095 +       }
35096 +       pos *= sizeof(ino);
35097 +       sz = xino_fwrite(write, file, &ino, sizeof(ino), &pos);
35098 +       if (sz == sizeof(ino))
35099 +               return 0; /* success */
35100 +
35101 +       AuIOErr("write failed (%zd)\n", sz);
35102 +       return -EIO;
35103 +}
35104 +
35105 +/*
35106 + * write @ino to the xinofile for the specified branch{@sb, @bindex}
35107 + * at the position of @h_ino.
35108 + * even if @ino is zero, it is written to the xinofile and means no entry.
35109 + * if the size of the xino file on a specific filesystem exceeds the watermark,
35110 + * try truncating it.
35111 + */
35112 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
35113 +                 ino_t ino)
35114 +{
35115 +       int err;
35116 +       unsigned int mnt_flags;
35117 +       struct au_branch *br;
35118 +
35119 +       BUILD_BUG_ON(sizeof(long long) != sizeof(au_loff_max)
35120 +                    || ((loff_t)-1) > 0);
35121 +       SiMustAnyLock(sb);
35122 +
35123 +       mnt_flags = au_mntflags(sb);
35124 +       if (!au_opt_test(mnt_flags, XINO))
35125 +               return 0;
35126 +
35127 +       br = au_sbr(sb, bindex);
35128 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
35129 +                              h_ino, ino);
35130 +       if (!err) {
35131 +               if (au_opt_test(mnt_flags, TRUNC_XINO)
35132 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
35133 +                       xino_try_trunc(sb, br);
35134 +               return 0; /* success */
35135 +       }
35136 +
35137 +       AuIOErr("write failed (%d)\n", err);
35138 +       return -EIO;
35139 +}
35140 +
35141 +/* ---------------------------------------------------------------------- */
35142 +
35143 +/* aufs inode number bitmap */
35144 +
35145 +static const int page_bits = (int)PAGE_SIZE * BITS_PER_BYTE;
35146 +static ino_t xib_calc_ino(unsigned long pindex, int bit)
35147 +{
35148 +       ino_t ino;
35149 +
35150 +       AuDebugOn(bit < 0 || page_bits <= bit);
35151 +       ino = AUFS_FIRST_INO + pindex * page_bits + bit;
35152 +       return ino;
35153 +}
35154 +
35155 +static void xib_calc_bit(ino_t ino, unsigned long *pindex, int *bit)
35156 +{
35157 +       AuDebugOn(ino < AUFS_FIRST_INO);
35158 +       ino -= AUFS_FIRST_INO;
35159 +       *pindex = ino / page_bits;
35160 +       *bit = ino % page_bits;
35161 +}
35162 +
35163 +static int xib_pindex(struct super_block *sb, unsigned long pindex)
35164 +{
35165 +       int err;
35166 +       loff_t pos;
35167 +       ssize_t sz;
35168 +       struct au_sbinfo *sbinfo;
35169 +       struct file *xib;
35170 +       unsigned long *p;
35171 +
35172 +       sbinfo = au_sbi(sb);
35173 +       MtxMustLock(&sbinfo->si_xib_mtx);
35174 +       AuDebugOn(pindex > ULONG_MAX / PAGE_SIZE
35175 +                 || !au_opt_test(sbinfo->si_mntflags, XINO));
35176 +
35177 +       if (pindex == sbinfo->si_xib_last_pindex)
35178 +               return 0;
35179 +
35180 +       xib = sbinfo->si_xib;
35181 +       p = sbinfo->si_xib_buf;
35182 +       pos = sbinfo->si_xib_last_pindex;
35183 +       pos *= PAGE_SIZE;
35184 +       sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
35185 +       if (unlikely(sz != PAGE_SIZE))
35186 +               goto out;
35187 +
35188 +       pos = pindex;
35189 +       pos *= PAGE_SIZE;
35190 +       if (vfsub_f_size_read(xib) >= pos + PAGE_SIZE)
35191 +               sz = xino_fread(sbinfo->si_xread, xib, p, PAGE_SIZE, &pos);
35192 +       else {
35193 +               memset(p, 0, PAGE_SIZE);
35194 +               sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
35195 +       }
35196 +       if (sz == PAGE_SIZE) {
35197 +               sbinfo->si_xib_last_pindex = pindex;
35198 +               return 0; /* success */
35199 +       }
35200 +
35201 +out:
35202 +       AuIOErr1("write failed (%zd)\n", sz);
35203 +       err = sz;
35204 +       if (sz >= 0)
35205 +               err = -EIO;
35206 +       return err;
35207 +}
35208 +
35209 +/* ---------------------------------------------------------------------- */
35210 +
35211 +static void au_xib_clear_bit(struct inode *inode)
35212 +{
35213 +       int err, bit;
35214 +       unsigned long pindex;
35215 +       struct super_block *sb;
35216 +       struct au_sbinfo *sbinfo;
35217 +
35218 +       AuDebugOn(inode->i_nlink);
35219 +
35220 +       sb = inode->i_sb;
35221 +       xib_calc_bit(inode->i_ino, &pindex, &bit);
35222 +       AuDebugOn(page_bits <= bit);
35223 +       sbinfo = au_sbi(sb);
35224 +       mutex_lock(&sbinfo->si_xib_mtx);
35225 +       err = xib_pindex(sb, pindex);
35226 +       if (!err) {
35227 +               clear_bit(bit, sbinfo->si_xib_buf);
35228 +               sbinfo->si_xib_next_bit = bit;
35229 +       }
35230 +       mutex_unlock(&sbinfo->si_xib_mtx);
35231 +}
35232 +
35233 +/* for s_op->delete_inode() */
35234 +void au_xino_delete_inode(struct inode *inode, const int unlinked)
35235 +{
35236 +       int err;
35237 +       unsigned int mnt_flags;
35238 +       aufs_bindex_t bindex, bbot, bi;
35239 +       unsigned char try_trunc;
35240 +       struct au_iinfo *iinfo;
35241 +       struct super_block *sb;
35242 +       struct au_hinode *hi;
35243 +       struct inode *h_inode;
35244 +       struct au_branch *br;
35245 +       vfs_writef_t xwrite;
35246 +
35247 +       AuDebugOn(au_is_bad_inode(inode));
35248 +
35249 +       sb = inode->i_sb;
35250 +       mnt_flags = au_mntflags(sb);
35251 +       if (!au_opt_test(mnt_flags, XINO)
35252 +           || inode->i_ino == AUFS_ROOT_INO)
35253 +               return;
35254 +
35255 +       if (unlinked) {
35256 +               au_xigen_inc(inode);
35257 +               au_xib_clear_bit(inode);
35258 +       }
35259 +
35260 +       iinfo = au_ii(inode);
35261 +       bindex = iinfo->ii_btop;
35262 +       if (bindex < 0)
35263 +               return;
35264 +
35265 +       xwrite = au_sbi(sb)->si_xwrite;
35266 +       try_trunc = !!au_opt_test(mnt_flags, TRUNC_XINO);
35267 +       hi = au_hinode(iinfo, bindex);
35268 +       bbot = iinfo->ii_bbot;
35269 +       for (; bindex <= bbot; bindex++, hi++) {
35270 +               h_inode = hi->hi_inode;
35271 +               if (!h_inode
35272 +                   || (!unlinked && h_inode->i_nlink))
35273 +                       continue;
35274 +
35275 +               /* inode may not be revalidated */
35276 +               bi = au_br_index(sb, hi->hi_id);
35277 +               if (bi < 0)
35278 +                       continue;
35279 +
35280 +               br = au_sbr(sb, bi);
35281 +               err = au_xino_do_write(xwrite, br->br_xino.xi_file,
35282 +                                      h_inode->i_ino, /*ino*/0);
35283 +               if (!err && try_trunc
35284 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
35285 +                       xino_try_trunc(sb, br);
35286 +       }
35287 +}
35288 +
35289 +/* get an unused inode number from bitmap */
35290 +ino_t au_xino_new_ino(struct super_block *sb)
35291 +{
35292 +       ino_t ino;
35293 +       unsigned long *p, pindex, ul, pend;
35294 +       struct au_sbinfo *sbinfo;
35295 +       struct file *file;
35296 +       int free_bit, err;
35297 +
35298 +       if (!au_opt_test(au_mntflags(sb), XINO))
35299 +               return iunique(sb, AUFS_FIRST_INO);
35300 +
35301 +       sbinfo = au_sbi(sb);
35302 +       mutex_lock(&sbinfo->si_xib_mtx);
35303 +       p = sbinfo->si_xib_buf;
35304 +       free_bit = sbinfo->si_xib_next_bit;
35305 +       if (free_bit < page_bits && !test_bit(free_bit, p))
35306 +               goto out; /* success */
35307 +       free_bit = find_first_zero_bit(p, page_bits);
35308 +       if (free_bit < page_bits)
35309 +               goto out; /* success */
35310 +
35311 +       pindex = sbinfo->si_xib_last_pindex;
35312 +       for (ul = pindex - 1; ul < ULONG_MAX; ul--) {
35313 +               err = xib_pindex(sb, ul);
35314 +               if (unlikely(err))
35315 +                       goto out_err;
35316 +               free_bit = find_first_zero_bit(p, page_bits);
35317 +               if (free_bit < page_bits)
35318 +                       goto out; /* success */
35319 +       }
35320 +
35321 +       file = sbinfo->si_xib;
35322 +       pend = vfsub_f_size_read(file) / PAGE_SIZE;
35323 +       for (ul = pindex + 1; ul <= pend; ul++) {
35324 +               err = xib_pindex(sb, ul);
35325 +               if (unlikely(err))
35326 +                       goto out_err;
35327 +               free_bit = find_first_zero_bit(p, page_bits);
35328 +               if (free_bit < page_bits)
35329 +                       goto out; /* success */
35330 +       }
35331 +       BUG();
35332 +
35333 +out:
35334 +       set_bit(free_bit, p);
35335 +       sbinfo->si_xib_next_bit = free_bit + 1;
35336 +       pindex = sbinfo->si_xib_last_pindex;
35337 +       mutex_unlock(&sbinfo->si_xib_mtx);
35338 +       ino = xib_calc_ino(pindex, free_bit);
35339 +       AuDbg("i%lu\n", (unsigned long)ino);
35340 +       return ino;
35341 +out_err:
35342 +       mutex_unlock(&sbinfo->si_xib_mtx);
35343 +       AuDbg("i0\n");
35344 +       return 0;
35345 +}
35346 +
35347 +/*
35348 + * read @ino from xinofile for the specified branch{@sb, @bindex}
35349 + * at the position of @h_ino.
35350 + * if @ino does not exist and @do_new is true, get new one.
35351 + */
35352 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
35353 +                ino_t *ino)
35354 +{
35355 +       int err;
35356 +       ssize_t sz;
35357 +       loff_t pos;
35358 +       struct file *file;
35359 +       struct au_sbinfo *sbinfo;
35360 +
35361 +       *ino = 0;
35362 +       if (!au_opt_test(au_mntflags(sb), XINO))
35363 +               return 0; /* no xino */
35364 +
35365 +       err = 0;
35366 +       sbinfo = au_sbi(sb);
35367 +       pos = h_ino;
35368 +       if (unlikely(au_loff_max / sizeof(*ino) - 1 < pos)) {
35369 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
35370 +               return -EFBIG;
35371 +       }
35372 +       pos *= sizeof(*ino);
35373 +
35374 +       file = au_sbr(sb, bindex)->br_xino.xi_file;
35375 +       if (vfsub_f_size_read(file) < pos + sizeof(*ino))
35376 +               return 0; /* no ino */
35377 +
35378 +       sz = xino_fread(sbinfo->si_xread, file, ino, sizeof(*ino), &pos);
35379 +       if (sz == sizeof(*ino))
35380 +               return 0; /* success */
35381 +
35382 +       err = sz;
35383 +       if (unlikely(sz >= 0)) {
35384 +               err = -EIO;
35385 +               AuIOErr("xino read error (%zd)\n", sz);
35386 +       }
35387 +
35388 +       return err;
35389 +}
35390 +
35391 +/* ---------------------------------------------------------------------- */
35392 +
35393 +/* create and set a new xino file */
35394 +
35395 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent)
35396 +{
35397 +       struct file *file;
35398 +       struct dentry *h_parent, *d;
35399 +       struct inode *h_dir, *inode;
35400 +       int err;
35401 +
35402 +       /*
35403 +        * at mount-time, and the xino file is the default path,
35404 +        * hnotify is disabled so we have no notify events to ignore.
35405 +        * when a user specified the xino, we cannot get au_hdir to be ignored.
35406 +        */
35407 +       file = vfsub_filp_open(fname, O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
35408 +                              /* | __FMODE_NONOTIFY */,
35409 +                              S_IRUGO | S_IWUGO);
35410 +       if (IS_ERR(file)) {
35411 +               if (!silent)
35412 +                       pr_err("open %s(%ld)\n", fname, PTR_ERR(file));
35413 +               return file;
35414 +       }
35415 +
35416 +       /* keep file count */
35417 +       err = 0;
35418 +       inode = file_inode(file);
35419 +       h_parent = dget_parent(file->f_path.dentry);
35420 +       h_dir = d_inode(h_parent);
35421 +       inode_lock_nested(h_dir, AuLsc_I_PARENT);
35422 +       /* mnt_want_write() is unnecessary here */
35423 +       /* no delegation since it is just created */
35424 +       if (inode->i_nlink)
35425 +               err = vfsub_unlink(h_dir, &file->f_path, /*delegated*/NULL,
35426 +                                  /*force*/0);
35427 +       inode_unlock(h_dir);
35428 +       dput(h_parent);
35429 +       if (unlikely(err)) {
35430 +               if (!silent)
35431 +                       pr_err("unlink %s(%d)\n", fname, err);
35432 +               goto out;
35433 +       }
35434 +
35435 +       err = -EINVAL;
35436 +       d = file->f_path.dentry;
35437 +       if (unlikely(sb == d->d_sb)) {
35438 +               if (!silent)
35439 +                       pr_err("%s must be outside\n", fname);
35440 +               goto out;
35441 +       }
35442 +       if (unlikely(au_test_fs_bad_xino(d->d_sb))) {
35443 +               if (!silent)
35444 +                       pr_err("xino doesn't support %s(%s)\n",
35445 +                              fname, au_sbtype(d->d_sb));
35446 +               goto out;
35447 +       }
35448 +       return file; /* success */
35449 +
35450 +out:
35451 +       fput(file);
35452 +       file = ERR_PTR(err);
35453 +       return file;
35454 +}
35455 +
35456 +/*
35457 + * find another branch who is on the same filesystem of the specified
35458 + * branch{@btgt}. search until @bbot.
35459 + */
35460 +static int is_sb_shared(struct super_block *sb, aufs_bindex_t btgt,
35461 +                       aufs_bindex_t bbot)
35462 +{
35463 +       aufs_bindex_t bindex;
35464 +       struct super_block *tgt_sb = au_sbr_sb(sb, btgt);
35465 +
35466 +       for (bindex = 0; bindex < btgt; bindex++)
35467 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
35468 +                       return bindex;
35469 +       for (bindex++; bindex <= bbot; bindex++)
35470 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
35471 +                       return bindex;
35472 +       return -1;
35473 +}
35474 +
35475 +/* ---------------------------------------------------------------------- */
35476 +
35477 +/*
35478 + * initialize the xinofile for the specified branch @br
35479 + * at the place/path where @base_file indicates.
35480 + * test whether another branch is on the same filesystem or not,
35481 + * if @do_test is true.
35482 + */
35483 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t h_ino,
35484 +              struct file *base_file, int do_test)
35485 +{
35486 +       int err;
35487 +       ino_t ino;
35488 +       aufs_bindex_t bbot, bindex;
35489 +       struct au_branch *shared_br, *b;
35490 +       struct file *file;
35491 +       struct super_block *tgt_sb;
35492 +
35493 +       shared_br = NULL;
35494 +       bbot = au_sbbot(sb);
35495 +       if (do_test) {
35496 +               tgt_sb = au_br_sb(br);
35497 +               for (bindex = 0; bindex <= bbot; bindex++) {
35498 +                       b = au_sbr(sb, bindex);
35499 +                       if (tgt_sb == au_br_sb(b)) {
35500 +                               shared_br = b;
35501 +                               break;
35502 +                       }
35503 +               }
35504 +       }
35505 +
35506 +       if (!shared_br || !shared_br->br_xino.xi_file) {
35507 +               struct au_xino_lock_dir ldir;
35508 +
35509 +               au_xino_lock_dir(sb, base_file, &ldir);
35510 +               /* mnt_want_write() is unnecessary here */
35511 +               file = au_xino_create2(base_file, NULL);
35512 +               au_xino_unlock_dir(&ldir);
35513 +               err = PTR_ERR(file);
35514 +               if (IS_ERR(file))
35515 +                       goto out;
35516 +               br->br_xino.xi_file = file;
35517 +       } else {
35518 +               br->br_xino.xi_file = shared_br->br_xino.xi_file;
35519 +               get_file(br->br_xino.xi_file);
35520 +       }
35521 +
35522 +       ino = AUFS_ROOT_INO;
35523 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
35524 +                              h_ino, ino);
35525 +       if (unlikely(err)) {
35526 +               fput(br->br_xino.xi_file);
35527 +               br->br_xino.xi_file = NULL;
35528 +       }
35529 +
35530 +out:
35531 +       return err;
35532 +}
35533 +
35534 +/* ---------------------------------------------------------------------- */
35535 +
35536 +/* trucate a xino bitmap file */
35537 +
35538 +/* todo: slow */
35539 +static int do_xib_restore(struct super_block *sb, struct file *file, void *page)
35540 +{
35541 +       int err, bit;
35542 +       ssize_t sz;
35543 +       unsigned long pindex;
35544 +       loff_t pos, pend;
35545 +       struct au_sbinfo *sbinfo;
35546 +       vfs_readf_t func;
35547 +       ino_t *ino;
35548 +       unsigned long *p;
35549 +
35550 +       err = 0;
35551 +       sbinfo = au_sbi(sb);
35552 +       MtxMustLock(&sbinfo->si_xib_mtx);
35553 +       p = sbinfo->si_xib_buf;
35554 +       func = sbinfo->si_xread;
35555 +       pend = vfsub_f_size_read(file);
35556 +       pos = 0;
35557 +       while (pos < pend) {
35558 +               sz = xino_fread(func, file, page, PAGE_SIZE, &pos);
35559 +               err = sz;
35560 +               if (unlikely(sz <= 0))
35561 +                       goto out;
35562 +
35563 +               err = 0;
35564 +               for (ino = page; sz > 0; ino++, sz -= sizeof(ino)) {
35565 +                       if (unlikely(*ino < AUFS_FIRST_INO))
35566 +                               continue;
35567 +
35568 +                       xib_calc_bit(*ino, &pindex, &bit);
35569 +                       AuDebugOn(page_bits <= bit);
35570 +                       err = xib_pindex(sb, pindex);
35571 +                       if (!err)
35572 +                               set_bit(bit, p);
35573 +                       else
35574 +                               goto out;
35575 +               }
35576 +       }
35577 +
35578 +out:
35579 +       return err;
35580 +}
35581 +
35582 +static int xib_restore(struct super_block *sb)
35583 +{
35584 +       int err;
35585 +       aufs_bindex_t bindex, bbot;
35586 +       void *page;
35587 +
35588 +       err = -ENOMEM;
35589 +       page = (void *)__get_free_page(GFP_NOFS);
35590 +       if (unlikely(!page))
35591 +               goto out;
35592 +
35593 +       err = 0;
35594 +       bbot = au_sbbot(sb);
35595 +       for (bindex = 0; !err && bindex <= bbot; bindex++)
35596 +               if (!bindex || is_sb_shared(sb, bindex, bindex - 1) < 0)
35597 +                       err = do_xib_restore
35598 +                               (sb, au_sbr(sb, bindex)->br_xino.xi_file, page);
35599 +               else
35600 +                       AuDbg("b%d\n", bindex);
35601 +       au_delayed_free_page((unsigned long)page);
35602 +
35603 +out:
35604 +       return err;
35605 +}
35606 +
35607 +int au_xib_trunc(struct super_block *sb)
35608 +{
35609 +       int err;
35610 +       ssize_t sz;
35611 +       loff_t pos;
35612 +       struct au_xino_lock_dir ldir;
35613 +       struct au_sbinfo *sbinfo;
35614 +       unsigned long *p;
35615 +       struct file *file;
35616 +
35617 +       SiMustWriteLock(sb);
35618 +
35619 +       err = 0;
35620 +       sbinfo = au_sbi(sb);
35621 +       if (!au_opt_test(sbinfo->si_mntflags, XINO))
35622 +               goto out;
35623 +
35624 +       file = sbinfo->si_xib;
35625 +       if (vfsub_f_size_read(file) <= PAGE_SIZE)
35626 +               goto out;
35627 +
35628 +       au_xino_lock_dir(sb, file, &ldir);
35629 +       /* mnt_want_write() is unnecessary here */
35630 +       file = au_xino_create2(sbinfo->si_xib, NULL);
35631 +       au_xino_unlock_dir(&ldir);
35632 +       err = PTR_ERR(file);
35633 +       if (IS_ERR(file))
35634 +               goto out;
35635 +       fput(sbinfo->si_xib);
35636 +       sbinfo->si_xib = file;
35637 +
35638 +       p = sbinfo->si_xib_buf;
35639 +       memset(p, 0, PAGE_SIZE);
35640 +       pos = 0;
35641 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xib, p, PAGE_SIZE, &pos);
35642 +       if (unlikely(sz != PAGE_SIZE)) {
35643 +               err = sz;
35644 +               AuIOErr("err %d\n", err);
35645 +               if (sz >= 0)
35646 +                       err = -EIO;
35647 +               goto out;
35648 +       }
35649 +
35650 +       mutex_lock(&sbinfo->si_xib_mtx);
35651 +       /* mnt_want_write() is unnecessary here */
35652 +       err = xib_restore(sb);
35653 +       mutex_unlock(&sbinfo->si_xib_mtx);
35654 +
35655 +out:
35656 +       return err;
35657 +}
35658 +
35659 +/* ---------------------------------------------------------------------- */
35660 +
35661 +/*
35662 + * xino mount option handlers
35663 + */
35664 +
35665 +/* xino bitmap */
35666 +static void xino_clear_xib(struct super_block *sb)
35667 +{
35668 +       struct au_sbinfo *sbinfo;
35669 +
35670 +       SiMustWriteLock(sb);
35671 +
35672 +       sbinfo = au_sbi(sb);
35673 +       sbinfo->si_xread = NULL;
35674 +       sbinfo->si_xwrite = NULL;
35675 +       if (sbinfo->si_xib)
35676 +               fput(sbinfo->si_xib);
35677 +       sbinfo->si_xib = NULL;
35678 +       if (sbinfo->si_xib_buf)
35679 +               au_delayed_free_page((unsigned long)sbinfo->si_xib_buf);
35680 +       sbinfo->si_xib_buf = NULL;
35681 +}
35682 +
35683 +static int au_xino_set_xib(struct super_block *sb, struct file *base)
35684 +{
35685 +       int err;
35686 +       loff_t pos;
35687 +       struct au_sbinfo *sbinfo;
35688 +       struct file *file;
35689 +
35690 +       SiMustWriteLock(sb);
35691 +
35692 +       sbinfo = au_sbi(sb);
35693 +       file = au_xino_create2(base, sbinfo->si_xib);
35694 +       err = PTR_ERR(file);
35695 +       if (IS_ERR(file))
35696 +               goto out;
35697 +       if (sbinfo->si_xib)
35698 +               fput(sbinfo->si_xib);
35699 +       sbinfo->si_xib = file;
35700 +       sbinfo->si_xread = vfs_readf(file);
35701 +       sbinfo->si_xwrite = vfs_writef(file);
35702 +
35703 +       err = -ENOMEM;
35704 +       if (!sbinfo->si_xib_buf)
35705 +               sbinfo->si_xib_buf = (void *)get_zeroed_page(GFP_NOFS);
35706 +       if (unlikely(!sbinfo->si_xib_buf))
35707 +               goto out_unset;
35708 +
35709 +       sbinfo->si_xib_last_pindex = 0;
35710 +       sbinfo->si_xib_next_bit = 0;
35711 +       if (vfsub_f_size_read(file) < PAGE_SIZE) {
35712 +               pos = 0;
35713 +               err = xino_fwrite(sbinfo->si_xwrite, file, sbinfo->si_xib_buf,
35714 +                                 PAGE_SIZE, &pos);
35715 +               if (unlikely(err != PAGE_SIZE))
35716 +                       goto out_free;
35717 +       }
35718 +       err = 0;
35719 +       goto out; /* success */
35720 +
35721 +out_free:
35722 +       if (sbinfo->si_xib_buf)
35723 +               au_delayed_free_page((unsigned long)sbinfo->si_xib_buf);
35724 +       sbinfo->si_xib_buf = NULL;
35725 +       if (err >= 0)
35726 +               err = -EIO;
35727 +out_unset:
35728 +       fput(sbinfo->si_xib);
35729 +       sbinfo->si_xib = NULL;
35730 +       sbinfo->si_xread = NULL;
35731 +       sbinfo->si_xwrite = NULL;
35732 +out:
35733 +       return err;
35734 +}
35735 +
35736 +/* xino for each branch */
35737 +static void xino_clear_br(struct super_block *sb)
35738 +{
35739 +       aufs_bindex_t bindex, bbot;
35740 +       struct au_branch *br;
35741 +
35742 +       bbot = au_sbbot(sb);
35743 +       for (bindex = 0; bindex <= bbot; bindex++) {
35744 +               br = au_sbr(sb, bindex);
35745 +               if (!br || !br->br_xino.xi_file)
35746 +                       continue;
35747 +
35748 +               fput(br->br_xino.xi_file);
35749 +               br->br_xino.xi_file = NULL;
35750 +       }
35751 +}
35752 +
35753 +static int au_xino_set_br(struct super_block *sb, struct file *base)
35754 +{
35755 +       int err;
35756 +       ino_t ino;
35757 +       aufs_bindex_t bindex, bbot, bshared;
35758 +       struct {
35759 +               struct file *old, *new;
35760 +       } *fpair, *p;
35761 +       struct au_branch *br;
35762 +       struct inode *inode;
35763 +       vfs_writef_t writef;
35764 +
35765 +       SiMustWriteLock(sb);
35766 +
35767 +       err = -ENOMEM;
35768 +       bbot = au_sbbot(sb);
35769 +       fpair = kcalloc(bbot + 1, sizeof(*fpair), GFP_NOFS);
35770 +       if (unlikely(!fpair))
35771 +               goto out;
35772 +
35773 +       inode = d_inode(sb->s_root);
35774 +       ino = AUFS_ROOT_INO;
35775 +       writef = au_sbi(sb)->si_xwrite;
35776 +       for (bindex = 0, p = fpair; bindex <= bbot; bindex++, p++) {
35777 +               bshared = is_sb_shared(sb, bindex, bindex - 1);
35778 +               if (bshared >= 0) {
35779 +                       /* shared xino */
35780 +                       *p = fpair[bshared];
35781 +                       get_file(p->new);
35782 +               }
35783 +
35784 +               if (!p->new) {
35785 +                       /* new xino */
35786 +                       br = au_sbr(sb, bindex);
35787 +                       p->old = br->br_xino.xi_file;
35788 +                       p->new = au_xino_create2(base, br->br_xino.xi_file);
35789 +                       err = PTR_ERR(p->new);
35790 +                       if (IS_ERR(p->new)) {
35791 +                               p->new = NULL;
35792 +                               goto out_pair;
35793 +                       }
35794 +               }
35795 +
35796 +               err = au_xino_do_write(writef, p->new,
35797 +                                      au_h_iptr(inode, bindex)->i_ino, ino);
35798 +               if (unlikely(err))
35799 +                       goto out_pair;
35800 +       }
35801 +
35802 +       for (bindex = 0, p = fpair; bindex <= bbot; bindex++, p++) {
35803 +               br = au_sbr(sb, bindex);
35804 +               if (br->br_xino.xi_file)
35805 +                       fput(br->br_xino.xi_file);
35806 +               get_file(p->new);
35807 +               br->br_xino.xi_file = p->new;
35808 +       }
35809 +
35810 +out_pair:
35811 +       for (bindex = 0, p = fpair; bindex <= bbot; bindex++, p++)
35812 +               if (p->new)
35813 +                       fput(p->new);
35814 +               else
35815 +                       break;
35816 +       au_delayed_kfree(fpair);
35817 +out:
35818 +       return err;
35819 +}
35820 +
35821 +void au_xino_clr(struct super_block *sb)
35822 +{
35823 +       struct au_sbinfo *sbinfo;
35824 +
35825 +       au_xigen_clr(sb);
35826 +       xino_clear_xib(sb);
35827 +       xino_clear_br(sb);
35828 +       sbinfo = au_sbi(sb);
35829 +       /* lvalue, do not call au_mntflags() */
35830 +       au_opt_clr(sbinfo->si_mntflags, XINO);
35831 +}
35832 +
35833 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount)
35834 +{
35835 +       int err, skip;
35836 +       struct dentry *parent, *cur_parent;
35837 +       struct qstr *dname, *cur_name;
35838 +       struct file *cur_xino;
35839 +       struct inode *dir;
35840 +       struct au_sbinfo *sbinfo;
35841 +
35842 +       SiMustWriteLock(sb);
35843 +
35844 +       err = 0;
35845 +       sbinfo = au_sbi(sb);
35846 +       parent = dget_parent(xino->file->f_path.dentry);
35847 +       if (remount) {
35848 +               skip = 0;
35849 +               dname = &xino->file->f_path.dentry->d_name;
35850 +               cur_xino = sbinfo->si_xib;
35851 +               if (cur_xino) {
35852 +                       cur_parent = dget_parent(cur_xino->f_path.dentry);
35853 +                       cur_name = &cur_xino->f_path.dentry->d_name;
35854 +                       skip = (cur_parent == parent
35855 +                               && au_qstreq(dname, cur_name));
35856 +                       dput(cur_parent);
35857 +               }
35858 +               if (skip)
35859 +                       goto out;
35860 +       }
35861 +
35862 +       au_opt_set(sbinfo->si_mntflags, XINO);
35863 +       dir = d_inode(parent);
35864 +       inode_lock_nested(dir, AuLsc_I_PARENT);
35865 +       /* mnt_want_write() is unnecessary here */
35866 +       err = au_xino_set_xib(sb, xino->file);
35867 +       if (!err)
35868 +               err = au_xigen_set(sb, xino->file);
35869 +       if (!err)
35870 +               err = au_xino_set_br(sb, xino->file);
35871 +       inode_unlock(dir);
35872 +       if (!err)
35873 +               goto out; /* success */
35874 +
35875 +       /* reset all */
35876 +       AuIOErr("failed creating xino(%d).\n", err);
35877 +       au_xigen_clr(sb);
35878 +       xino_clear_xib(sb);
35879 +
35880 +out:
35881 +       dput(parent);
35882 +       return err;
35883 +}
35884 +
35885 +/* ---------------------------------------------------------------------- */
35886 +
35887 +/*
35888 + * create a xinofile at the default place/path.
35889 + */
35890 +struct file *au_xino_def(struct super_block *sb)
35891 +{
35892 +       struct file *file;
35893 +       char *page, *p;
35894 +       struct au_branch *br;
35895 +       struct super_block *h_sb;
35896 +       struct path path;
35897 +       aufs_bindex_t bbot, bindex, bwr;
35898 +
35899 +       br = NULL;
35900 +       bbot = au_sbbot(sb);
35901 +       bwr = -1;
35902 +       for (bindex = 0; bindex <= bbot; bindex++) {
35903 +               br = au_sbr(sb, bindex);
35904 +               if (au_br_writable(br->br_perm)
35905 +                   && !au_test_fs_bad_xino(au_br_sb(br))) {
35906 +                       bwr = bindex;
35907 +                       break;
35908 +               }
35909 +       }
35910 +
35911 +       if (bwr >= 0) {
35912 +               file = ERR_PTR(-ENOMEM);
35913 +               page = (void *)__get_free_page(GFP_NOFS);
35914 +               if (unlikely(!page))
35915 +                       goto out;
35916 +               path.mnt = au_br_mnt(br);
35917 +               path.dentry = au_h_dptr(sb->s_root, bwr);
35918 +               p = d_path(&path, page, PATH_MAX - sizeof(AUFS_XINO_FNAME));
35919 +               file = (void *)p;
35920 +               if (!IS_ERR(p)) {
35921 +                       strcat(p, "/" AUFS_XINO_FNAME);
35922 +                       AuDbg("%s\n", p);
35923 +                       file = au_xino_create(sb, p, /*silent*/0);
35924 +                       if (!IS_ERR(file))
35925 +                               au_xino_brid_set(sb, br->br_id);
35926 +               }
35927 +               au_delayed_free_page((unsigned long)page);
35928 +       } else {
35929 +               file = au_xino_create(sb, AUFS_XINO_DEFPATH, /*silent*/0);
35930 +               if (IS_ERR(file))
35931 +                       goto out;
35932 +               h_sb = file->f_path.dentry->d_sb;
35933 +               if (unlikely(au_test_fs_bad_xino(h_sb))) {
35934 +                       pr_err("xino doesn't support %s(%s)\n",
35935 +                              AUFS_XINO_DEFPATH, au_sbtype(h_sb));
35936 +                       fput(file);
35937 +                       file = ERR_PTR(-EINVAL);
35938 +               }
35939 +               if (!IS_ERR(file))
35940 +                       au_xino_brid_set(sb, -1);
35941 +       }
35942 +
35943 +out:
35944 +       return file;
35945 +}
35946 +
35947 +/* ---------------------------------------------------------------------- */
35948 +
35949 +int au_xino_path(struct seq_file *seq, struct file *file)
35950 +{
35951 +       int err;
35952 +
35953 +       err = au_seq_path(seq, &file->f_path);
35954 +       if (unlikely(err))
35955 +               goto out;
35956 +
35957 +#define Deleted "\\040(deleted)"
35958 +       seq->count -= sizeof(Deleted) - 1;
35959 +       AuDebugOn(memcmp(seq->buf + seq->count, Deleted,
35960 +                        sizeof(Deleted) - 1));
35961 +#undef Deleted
35962 +
35963 +out:
35964 +       return err;
35965 +}
35966 +
35967 +/* ---------------------------------------------------------------------- */
35968 +
35969 +void au_xinondir_leave(struct super_block *sb, aufs_bindex_t bindex,
35970 +                      ino_t h_ino, int idx)
35971 +{
35972 +       struct au_xino_file *xino;
35973 +
35974 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
35975 +       xino = &au_sbr(sb, bindex)->br_xino;
35976 +       AuDebugOn(idx < 0 || xino->xi_nondir.total <= idx);
35977 +
35978 +       spin_lock(&xino->xi_nondir.spin);
35979 +       AuDebugOn(xino->xi_nondir.array[idx] != h_ino);
35980 +       xino->xi_nondir.array[idx] = 0;
35981 +       spin_unlock(&xino->xi_nondir.spin);
35982 +       wake_up_all(&xino->xi_nondir.wqh);
35983 +}
35984 +
35985 +static int au_xinondir_find(struct au_xino_file *xino, ino_t h_ino)
35986 +{
35987 +       int found, total, i;
35988 +
35989 +       found = -1;
35990 +       total = xino->xi_nondir.total;
35991 +       for (i = 0; i < total; i++) {
35992 +               if (xino->xi_nondir.array[i] != h_ino)
35993 +                       continue;
35994 +               found = i;
35995 +               break;
35996 +       }
35997 +
35998 +       return found;
35999 +}
36000 +
36001 +static int au_xinondir_expand(struct au_xino_file *xino)
36002 +{
36003 +       int err, sz;
36004 +       ino_t *p;
36005 +
36006 +       BUILD_BUG_ON(KMALLOC_MAX_SIZE > INT_MAX);
36007 +
36008 +       err = -ENOMEM;
36009 +       sz = xino->xi_nondir.total * sizeof(ino_t);
36010 +       if (unlikely(sz > KMALLOC_MAX_SIZE / 2))
36011 +               goto out;
36012 +       p = au_kzrealloc(xino->xi_nondir.array, sz, sz << 1, GFP_ATOMIC,
36013 +                        /*may_shrink*/0);
36014 +       if (p) {
36015 +               xino->xi_nondir.array = p;
36016 +               xino->xi_nondir.total <<= 1;
36017 +               AuDbg("xi_nondir.total %d\n", xino->xi_nondir.total);
36018 +               err = 0;
36019 +       }
36020 +
36021 +out:
36022 +       return err;
36023 +}
36024 +
36025 +int au_xinondir_enter(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
36026 +                     int *idx)
36027 +{
36028 +       int err, found, empty;
36029 +       struct au_xino_file *xino;
36030 +
36031 +       err = 0;
36032 +       *idx = -1;
36033 +       if (!au_opt_test(au_mntflags(sb), XINO))
36034 +               goto out; /* no xino */
36035 +
36036 +       xino = &au_sbr(sb, bindex)->br_xino;
36037 +
36038 +again:
36039 +       spin_lock(&xino->xi_nondir.spin);
36040 +       found = au_xinondir_find(xino, h_ino);
36041 +       if (found == -1) {
36042 +               empty = au_xinondir_find(xino, /*h_ino*/0);
36043 +               if (empty == -1) {
36044 +                       empty = xino->xi_nondir.total;
36045 +                       err = au_xinondir_expand(xino);
36046 +                       if (unlikely(err))
36047 +                               goto out_unlock;
36048 +               }
36049 +               xino->xi_nondir.array[empty] = h_ino;
36050 +               *idx = empty;
36051 +       } else {
36052 +               spin_unlock(&xino->xi_nondir.spin);
36053 +               wait_event(xino->xi_nondir.wqh,
36054 +                          xino->xi_nondir.array[found] != h_ino);
36055 +               goto again;
36056 +       }
36057 +
36058 +out_unlock:
36059 +       spin_unlock(&xino->xi_nondir.spin);
36060 +out:
36061 +       return err;
36062 +}
36063 diff -urN /usr/share/empty/include/uapi/linux/aufs_type.h linux/include/uapi/linux/aufs_type.h
36064 --- /usr/share/empty/include/uapi/linux/aufs_type.h     1970-01-01 01:00:00.000000000 +0100
36065 +++ linux/include/uapi/linux/aufs_type.h        2017-05-06 22:16:52.784886929 +0200
36066 @@ -0,0 +1,419 @@
36067 +/*
36068 + * Copyright (C) 2005-2017 Junjiro R. Okajima
36069 + *
36070 + * This program, aufs is free software; you can redistribute it and/or modify
36071 + * it under the terms of the GNU General Public License as published by
36072 + * the Free Software Foundation; either version 2 of the License, or
36073 + * (at your option) any later version.
36074 + *
36075 + * This program is distributed in the hope that it will be useful,
36076 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
36077 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
36078 + * GNU General Public License for more details.
36079 + *
36080 + * You should have received a copy of the GNU General Public License
36081 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
36082 + */
36083 +
36084 +#ifndef __AUFS_TYPE_H__
36085 +#define __AUFS_TYPE_H__
36086 +
36087 +#define AUFS_NAME      "aufs"
36088 +
36089 +#ifdef __KERNEL__
36090 +/*
36091 + * define it before including all other headers.
36092 + * sched.h may use pr_* macros before defining "current", so define the
36093 + * no-current version first, and re-define later.
36094 + */
36095 +#define pr_fmt(fmt)    AUFS_NAME " %s:%d: " fmt, __func__, __LINE__
36096 +#include <linux/sched.h>
36097 +#undef pr_fmt
36098 +#define pr_fmt(fmt) \
36099 +               AUFS_NAME " %s:%d:%.*s[%d]: " fmt, __func__, __LINE__, \
36100 +               (int)sizeof(current->comm), current->comm, current->pid
36101 +#else
36102 +#include <stdint.h>
36103 +#include <sys/types.h>
36104 +#endif /* __KERNEL__ */
36105 +
36106 +#include <linux/limits.h>
36107 +
36108 +#define AUFS_VERSION   "4.x-rcN-20170410"
36109 +
36110 +/* todo? move this to linux-2.6.19/include/magic.h */
36111 +#define AUFS_SUPER_MAGIC       ('a' << 24 | 'u' << 16 | 'f' << 8 | 's')
36112 +
36113 +/* ---------------------------------------------------------------------- */
36114 +
36115 +#ifdef CONFIG_AUFS_BRANCH_MAX_127
36116 +typedef int8_t aufs_bindex_t;
36117 +#define AUFS_BRANCH_MAX 127
36118 +#else
36119 +typedef int16_t aufs_bindex_t;
36120 +#ifdef CONFIG_AUFS_BRANCH_MAX_511
36121 +#define AUFS_BRANCH_MAX 511
36122 +#elif defined(CONFIG_AUFS_BRANCH_MAX_1023)
36123 +#define AUFS_BRANCH_MAX 1023
36124 +#elif defined(CONFIG_AUFS_BRANCH_MAX_32767)
36125 +#define AUFS_BRANCH_MAX 32767
36126 +#endif
36127 +#endif
36128 +
36129 +#ifdef __KERNEL__
36130 +#ifndef AUFS_BRANCH_MAX
36131 +#error unknown CONFIG_AUFS_BRANCH_MAX value
36132 +#endif
36133 +#endif /* __KERNEL__ */
36134 +
36135 +/* ---------------------------------------------------------------------- */
36136 +
36137 +#define AUFS_FSTYPE            AUFS_NAME
36138 +
36139 +#define AUFS_ROOT_INO          2
36140 +#define AUFS_FIRST_INO         11
36141 +
36142 +#define AUFS_WH_PFX            ".wh."
36143 +#define AUFS_WH_PFX_LEN                ((int)sizeof(AUFS_WH_PFX) - 1)
36144 +#define AUFS_WH_TMP_LEN                4
36145 +/* a limit for rmdir/rename a dir and copyup */
36146 +#define AUFS_MAX_NAMELEN       (NAME_MAX \
36147 +                               - AUFS_WH_PFX_LEN * 2   /* doubly whiteouted */\
36148 +                               - 1                     /* dot */\
36149 +                               - AUFS_WH_TMP_LEN)      /* hex */
36150 +#define AUFS_XINO_FNAME                "." AUFS_NAME ".xino"
36151 +#define AUFS_XINO_DEFPATH      "/tmp/" AUFS_XINO_FNAME
36152 +#define AUFS_XINO_DEF_SEC      30 /* seconds */
36153 +#define AUFS_XINO_DEF_TRUNC    45 /* percentage */
36154 +#define AUFS_DIRWH_DEF         3
36155 +#define AUFS_RDCACHE_DEF       10 /* seconds */
36156 +#define AUFS_RDCACHE_MAX       3600 /* seconds */
36157 +#define AUFS_RDBLK_DEF         512 /* bytes */
36158 +#define AUFS_RDHASH_DEF                32
36159 +#define AUFS_WKQ_NAME          AUFS_NAME "d"
36160 +#define AUFS_MFS_DEF_SEC       30 /* seconds */
36161 +#define AUFS_MFS_MAX_SEC       3600 /* seconds */
36162 +#define AUFS_FHSM_CACHE_DEF_SEC        30 /* seconds */
36163 +#define AUFS_PLINK_WARN                50 /* number of plinks in a single bucket */
36164 +
36165 +/* pseudo-link maintenace under /proc */
36166 +#define AUFS_PLINK_MAINT_NAME  "plink_maint"
36167 +#define AUFS_PLINK_MAINT_DIR   "fs/" AUFS_NAME
36168 +#define AUFS_PLINK_MAINT_PATH  AUFS_PLINK_MAINT_DIR "/" AUFS_PLINK_MAINT_NAME
36169 +
36170 +#define AUFS_DIROPQ_NAME       AUFS_WH_PFX ".opq" /* whiteouted doubly */
36171 +#define AUFS_WH_DIROPQ         AUFS_WH_PFX AUFS_DIROPQ_NAME
36172 +
36173 +#define AUFS_BASE_NAME         AUFS_WH_PFX AUFS_NAME
36174 +#define AUFS_PLINKDIR_NAME     AUFS_WH_PFX "plnk"
36175 +#define AUFS_ORPHDIR_NAME      AUFS_WH_PFX "orph"
36176 +
36177 +/* doubly whiteouted */
36178 +#define AUFS_WH_BASE           AUFS_WH_PFX AUFS_BASE_NAME
36179 +#define AUFS_WH_PLINKDIR       AUFS_WH_PFX AUFS_PLINKDIR_NAME
36180 +#define AUFS_WH_ORPHDIR                AUFS_WH_PFX AUFS_ORPHDIR_NAME
36181 +
36182 +/* branch permissions and attributes */
36183 +#define AUFS_BRPERM_RW         "rw"
36184 +#define AUFS_BRPERM_RO         "ro"
36185 +#define AUFS_BRPERM_RR         "rr"
36186 +#define AUFS_BRATTR_COO_REG    "coo_reg"
36187 +#define AUFS_BRATTR_COO_ALL    "coo_all"
36188 +#define AUFS_BRATTR_FHSM       "fhsm"
36189 +#define AUFS_BRATTR_UNPIN      "unpin"
36190 +#define AUFS_BRATTR_ICEX       "icex"
36191 +#define AUFS_BRATTR_ICEX_SEC   "icexsec"
36192 +#define AUFS_BRATTR_ICEX_SYS   "icexsys"
36193 +#define AUFS_BRATTR_ICEX_TR    "icextr"
36194 +#define AUFS_BRATTR_ICEX_USR   "icexusr"
36195 +#define AUFS_BRATTR_ICEX_OTH   "icexoth"
36196 +#define AUFS_BRRATTR_WH                "wh"
36197 +#define AUFS_BRWATTR_NLWH      "nolwh"
36198 +#define AUFS_BRWATTR_MOO       "moo"
36199 +
36200 +#define AuBrPerm_RW            1               /* writable, hardlinkable wh */
36201 +#define AuBrPerm_RO            (1 << 1)        /* readonly */
36202 +#define AuBrPerm_RR            (1 << 2)        /* natively readonly */
36203 +#define AuBrPerm_Mask          (AuBrPerm_RW | AuBrPerm_RO | AuBrPerm_RR)
36204 +
36205 +#define AuBrAttr_COO_REG       (1 << 3)        /* copy-up on open */
36206 +#define AuBrAttr_COO_ALL       (1 << 4)
36207 +#define AuBrAttr_COO_Mask      (AuBrAttr_COO_REG | AuBrAttr_COO_ALL)
36208 +
36209 +#define AuBrAttr_FHSM          (1 << 5)        /* file-based hsm */
36210 +#define AuBrAttr_UNPIN         (1 << 6)        /* rename-able top dir of
36211 +                                                  branch. meaningless since
36212 +                                                  linux-3.18-rc1 */
36213 +
36214 +/* ignore error in copying XATTR */
36215 +#define AuBrAttr_ICEX_SEC      (1 << 7)
36216 +#define AuBrAttr_ICEX_SYS      (1 << 8)
36217 +#define AuBrAttr_ICEX_TR       (1 << 9)
36218 +#define AuBrAttr_ICEX_USR      (1 << 10)
36219 +#define AuBrAttr_ICEX_OTH      (1 << 11)
36220 +#define AuBrAttr_ICEX          (AuBrAttr_ICEX_SEC      \
36221 +                                | AuBrAttr_ICEX_SYS    \
36222 +                                | AuBrAttr_ICEX_TR     \
36223 +                                | AuBrAttr_ICEX_USR    \
36224 +                                | AuBrAttr_ICEX_OTH)
36225 +
36226 +#define AuBrRAttr_WH           (1 << 12)       /* whiteout-able */
36227 +#define AuBrRAttr_Mask         AuBrRAttr_WH
36228 +
36229 +#define AuBrWAttr_NoLinkWH     (1 << 13)       /* un-hardlinkable whiteouts */
36230 +#define AuBrWAttr_MOO          (1 << 14)       /* move-up on open */
36231 +#define AuBrWAttr_Mask         (AuBrWAttr_NoLinkWH | AuBrWAttr_MOO)
36232 +
36233 +#define AuBrAttr_CMOO_Mask     (AuBrAttr_COO_Mask | AuBrWAttr_MOO)
36234 +
36235 +/* #warning test userspace */
36236 +#ifdef __KERNEL__
36237 +#ifndef CONFIG_AUFS_FHSM
36238 +#undef AuBrAttr_FHSM
36239 +#define AuBrAttr_FHSM          0
36240 +#endif
36241 +#ifndef CONFIG_AUFS_XATTR
36242 +#undef AuBrAttr_ICEX
36243 +#define AuBrAttr_ICEX          0
36244 +#undef AuBrAttr_ICEX_SEC
36245 +#define AuBrAttr_ICEX_SEC      0
36246 +#undef AuBrAttr_ICEX_SYS
36247 +#define AuBrAttr_ICEX_SYS      0
36248 +#undef AuBrAttr_ICEX_TR
36249 +#define AuBrAttr_ICEX_TR       0
36250 +#undef AuBrAttr_ICEX_USR
36251 +#define AuBrAttr_ICEX_USR      0
36252 +#undef AuBrAttr_ICEX_OTH
36253 +#define AuBrAttr_ICEX_OTH      0
36254 +#endif
36255 +#endif
36256 +
36257 +/* the longest combination */
36258 +/* AUFS_BRATTR_ICEX and AUFS_BRATTR_ICEX_TR don't affect here */
36259 +#define AuBrPermStrSz  sizeof(AUFS_BRPERM_RW                   \
36260 +                              "+" AUFS_BRATTR_COO_REG          \
36261 +                              "+" AUFS_BRATTR_FHSM             \
36262 +                              "+" AUFS_BRATTR_UNPIN            \
36263 +                              "+" AUFS_BRATTR_ICEX_SEC         \
36264 +                              "+" AUFS_BRATTR_ICEX_SYS         \
36265 +                              "+" AUFS_BRATTR_ICEX_USR         \
36266 +                              "+" AUFS_BRATTR_ICEX_OTH         \
36267 +                              "+" AUFS_BRWATTR_NLWH)
36268 +
36269 +typedef struct {
36270 +       char a[AuBrPermStrSz];
36271 +} au_br_perm_str_t;
36272 +
36273 +static inline int au_br_writable(int brperm)
36274 +{
36275 +       return brperm & AuBrPerm_RW;
36276 +}
36277 +
36278 +static inline int au_br_whable(int brperm)
36279 +{
36280 +       return brperm & (AuBrPerm_RW | AuBrRAttr_WH);
36281 +}
36282 +
36283 +static inline int au_br_wh_linkable(int brperm)
36284 +{
36285 +       return !(brperm & AuBrWAttr_NoLinkWH);
36286 +}
36287 +
36288 +static inline int au_br_cmoo(int brperm)
36289 +{
36290 +       return brperm & AuBrAttr_CMOO_Mask;
36291 +}
36292 +
36293 +static inline int au_br_fhsm(int brperm)
36294 +{
36295 +       return brperm & AuBrAttr_FHSM;
36296 +}
36297 +
36298 +/* ---------------------------------------------------------------------- */
36299 +
36300 +/* ioctl */
36301 +enum {
36302 +       /* readdir in userspace */
36303 +       AuCtl_RDU,
36304 +       AuCtl_RDU_INO,
36305 +
36306 +       AuCtl_WBR_FD,   /* pathconf wrapper */
36307 +       AuCtl_IBUSY,    /* busy inode */
36308 +       AuCtl_MVDOWN,   /* move-down */
36309 +       AuCtl_BR,       /* info about branches */
36310 +       AuCtl_FHSM_FD   /* connection for fhsm */
36311 +};
36312 +
36313 +/* borrowed from linux/include/linux/kernel.h */
36314 +#ifndef ALIGN
36315 +#define ALIGN(x, a)            __ALIGN_MASK(x, (typeof(x))(a)-1)
36316 +#define __ALIGN_MASK(x, mask)  (((x)+(mask))&~(mask))
36317 +#endif
36318 +
36319 +/* borrowed from linux/include/linux/compiler-gcc3.h */
36320 +#ifndef __aligned
36321 +#define __aligned(x)                   __attribute__((aligned(x)))
36322 +#endif
36323 +
36324 +#ifdef __KERNEL__
36325 +#ifndef __packed
36326 +#define __packed                       __attribute__((packed))
36327 +#endif
36328 +#endif
36329 +
36330 +struct au_rdu_cookie {
36331 +       uint64_t        h_pos;
36332 +       int16_t         bindex;
36333 +       uint8_t         flags;
36334 +       uint8_t         pad;
36335 +       uint32_t        generation;
36336 +} __aligned(8);
36337 +
36338 +struct au_rdu_ent {
36339 +       uint64_t        ino;
36340 +       int16_t         bindex;
36341 +       uint8_t         type;
36342 +       uint8_t         nlen;
36343 +       uint8_t         wh;
36344 +       char            name[0];
36345 +} __aligned(8);
36346 +
36347 +static inline int au_rdu_len(int nlen)
36348 +{
36349 +       /* include the terminating NULL */
36350 +       return ALIGN(sizeof(struct au_rdu_ent) + nlen + 1,
36351 +                    sizeof(uint64_t));
36352 +}
36353 +
36354 +union au_rdu_ent_ul {
36355 +       struct au_rdu_ent __user        *e;
36356 +       uint64_t                        ul;
36357 +};
36358 +
36359 +enum {
36360 +       AufsCtlRduV_SZ,
36361 +       AufsCtlRduV_End
36362 +};
36363 +
36364 +struct aufs_rdu {
36365 +       /* input */
36366 +       union {
36367 +               uint64_t        sz;     /* AuCtl_RDU */
36368 +               uint64_t        nent;   /* AuCtl_RDU_INO */
36369 +       };
36370 +       union au_rdu_ent_ul     ent;
36371 +       uint16_t                verify[AufsCtlRduV_End];
36372 +
36373 +       /* input/output */
36374 +       uint32_t                blk;
36375 +
36376 +       /* output */
36377 +       union au_rdu_ent_ul     tail;
36378 +       /* number of entries which were added in a single call */
36379 +       uint64_t                rent;
36380 +       uint8_t                 full;
36381 +       uint8_t                 shwh;
36382 +
36383 +       struct au_rdu_cookie    cookie;
36384 +} __aligned(8);
36385 +
36386 +/* ---------------------------------------------------------------------- */
36387 +
36388 +struct aufs_wbr_fd {
36389 +       uint32_t        oflags;
36390 +       int16_t         brid;
36391 +} __aligned(8);
36392 +
36393 +/* ---------------------------------------------------------------------- */
36394 +
36395 +struct aufs_ibusy {
36396 +       uint64_t        ino, h_ino;
36397 +       int16_t         bindex;
36398 +} __aligned(8);
36399 +
36400 +/* ---------------------------------------------------------------------- */
36401 +
36402 +/* error code for move-down */
36403 +/* the actual message strings are implemented in aufs-util.git */
36404 +enum {
36405 +       EAU_MVDOWN_OPAQUE = 1,
36406 +       EAU_MVDOWN_WHITEOUT,
36407 +       EAU_MVDOWN_UPPER,
36408 +       EAU_MVDOWN_BOTTOM,
36409 +       EAU_MVDOWN_NOUPPER,
36410 +       EAU_MVDOWN_NOLOWERBR,
36411 +       EAU_Last
36412 +};
36413 +
36414 +/* flags for move-down */
36415 +#define AUFS_MVDOWN_DMSG       1
36416 +#define AUFS_MVDOWN_OWLOWER    (1 << 1)        /* overwrite lower */
36417 +#define AUFS_MVDOWN_KUPPER     (1 << 2)        /* keep upper */
36418 +#define AUFS_MVDOWN_ROLOWER    (1 << 3)        /* do even if lower is RO */
36419 +#define AUFS_MVDOWN_ROLOWER_R  (1 << 4)        /* did on lower RO */
36420 +#define AUFS_MVDOWN_ROUPPER    (1 << 5)        /* do even if upper is RO */
36421 +#define AUFS_MVDOWN_ROUPPER_R  (1 << 6)        /* did on upper RO */
36422 +#define AUFS_MVDOWN_BRID_UPPER (1 << 7)        /* upper brid */
36423 +#define AUFS_MVDOWN_BRID_LOWER (1 << 8)        /* lower brid */
36424 +#define AUFS_MVDOWN_FHSM_LOWER (1 << 9)        /* find fhsm attr for lower */
36425 +#define AUFS_MVDOWN_STFS       (1 << 10)       /* req. stfs */
36426 +#define AUFS_MVDOWN_STFS_FAILED        (1 << 11)       /* output: stfs is unusable */
36427 +#define AUFS_MVDOWN_BOTTOM     (1 << 12)       /* output: no more lowers */
36428 +
36429 +/* index for move-down */
36430 +enum {
36431 +       AUFS_MVDOWN_UPPER,
36432 +       AUFS_MVDOWN_LOWER,
36433 +       AUFS_MVDOWN_NARRAY
36434 +};
36435 +
36436 +/*
36437 + * additional info of move-down
36438 + * number of free blocks and inodes.
36439 + * subset of struct kstatfs, but smaller and always 64bit.
36440 + */
36441 +struct aufs_stfs {
36442 +       uint64_t        f_blocks;
36443 +       uint64_t        f_bavail;
36444 +       uint64_t        f_files;
36445 +       uint64_t        f_ffree;
36446 +};
36447 +
36448 +struct aufs_stbr {
36449 +       int16_t                 brid;   /* optional input */
36450 +       int16_t                 bindex; /* output */
36451 +       struct aufs_stfs        stfs;   /* output when AUFS_MVDOWN_STFS set */
36452 +} __aligned(8);
36453 +
36454 +struct aufs_mvdown {
36455 +       uint32_t                flags;                  /* input/output */
36456 +       struct aufs_stbr        stbr[AUFS_MVDOWN_NARRAY]; /* input/output */
36457 +       int8_t                  au_errno;               /* output */
36458 +} __aligned(8);
36459 +
36460 +/* ---------------------------------------------------------------------- */
36461 +
36462 +union aufs_brinfo {
36463 +       /* PATH_MAX may differ between kernel-space and user-space */
36464 +       char    _spacer[4096];
36465 +       struct {
36466 +               int16_t id;
36467 +               int     perm;
36468 +               char    path[0];
36469 +       };
36470 +} __aligned(8);
36471 +
36472 +/* ---------------------------------------------------------------------- */
36473 +
36474 +#define AuCtlType              'A'
36475 +#define AUFS_CTL_RDU           _IOWR(AuCtlType, AuCtl_RDU, struct aufs_rdu)
36476 +#define AUFS_CTL_RDU_INO       _IOWR(AuCtlType, AuCtl_RDU_INO, struct aufs_rdu)
36477 +#define AUFS_CTL_WBR_FD                _IOW(AuCtlType, AuCtl_WBR_FD, \
36478 +                                    struct aufs_wbr_fd)
36479 +#define AUFS_CTL_IBUSY         _IOWR(AuCtlType, AuCtl_IBUSY, struct aufs_ibusy)
36480 +#define AUFS_CTL_MVDOWN                _IOWR(AuCtlType, AuCtl_MVDOWN, \
36481 +                                     struct aufs_mvdown)
36482 +#define AUFS_CTL_BRINFO                _IOW(AuCtlType, AuCtl_BR, union aufs_brinfo)
36483 +#define AUFS_CTL_FHSM_FD       _IOW(AuCtlType, AuCtl_FHSM_FD, int)
36484 +
36485 +#endif /* __AUFS_TYPE_H__ */
36486 aufs4.x-rcN loopback patch
36487
36488 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
36489 index cf39138..29b33b2 100644
36490 --- a/drivers/block/loop.c
36491 +++ b/drivers/block/loop.c
36492 @@ -551,7 +551,7 @@ static int do_req_filebacked(struct loop_device *lo, struct request *rq)
36493  }
36494
36495  struct switch_request {
36496 -       struct file *file;
36497 +       struct file *file, *virt_file;
36498         struct completion wait;
36499  };
36500
36501 @@ -577,6 +577,7 @@ static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
36502         mapping = file->f_mapping;
36503         mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
36504         lo->lo_backing_file = file;
36505 +       lo->lo_backing_virt_file = p->virt_file;
36506         lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
36507                 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
36508         lo->old_gfp_mask = mapping_gfp_mask(mapping);
36509 @@ -589,11 +590,13 @@ static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
36510   * First it needs to flush existing IO, it does this by sending a magic
36511   * BIO down the pipe. The completion of this BIO does the actual switch.
36512   */
36513 -static int loop_switch(struct loop_device *lo, struct file *file)
36514 +static int loop_switch(struct loop_device *lo, struct file *file,
36515 +                      struct file *virt_file)
36516  {
36517         struct switch_request w;
36518
36519         w.file = file;
36520 +       w.virt_file = virt_file;
36521
36522         /* freeze queue and wait for completion of scheduled requests */
36523         blk_mq_freeze_queue(lo->lo_queue);
36524 @@ -612,7 +615,16 @@ static int loop_switch(struct loop_device *lo, struct file *file)
36525   */
36526  static int loop_flush(struct loop_device *lo)
36527  {
36528 -       return loop_switch(lo, NULL);
36529 +       return loop_switch(lo, NULL, NULL);
36530 +}
36531 +
36532 +static struct file *loop_real_file(struct file *file)
36533 +{
36534 +       struct file *f = NULL;
36535 +
36536 +       if (file->f_path.dentry->d_sb->s_op->real_loop)
36537 +               f = file->f_path.dentry->d_sb->s_op->real_loop(file);
36538 +       return f;
36539  }
36540
36541  static void loop_reread_partitions(struct loop_device *lo,
36542 @@ -649,6 +661,7 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
36543                           unsigned int arg)
36544  {
36545         struct file     *file, *old_file;
36546 +       struct file     *f, *virt_file = NULL, *old_virt_file;
36547         struct inode    *inode;
36548         int             error;
36549
36550 @@ -665,9 +678,16 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
36551         file = fget(arg);
36552         if (!file)
36553                 goto out;
36554 +       f = loop_real_file(file);
36555 +       if (f) {
36556 +               virt_file = file;
36557 +               file = f;
36558 +               get_file(file);
36559 +       }
36560
36561         inode = file->f_mapping->host;
36562         old_file = lo->lo_backing_file;
36563 +       old_virt_file = lo->lo_backing_virt_file;
36564
36565         error = -EINVAL;
36566
36567 @@ -679,17 +699,21 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
36568                 goto out_putf;
36569
36570         /* and ... switch */
36571 -       error = loop_switch(lo, file);
36572 +       error = loop_switch(lo, file, virt_file);
36573         if (error)
36574                 goto out_putf;
36575
36576         fput(old_file);
36577 +       if (old_virt_file)
36578 +               fput(old_virt_file);
36579         if (lo->lo_flags & LO_FLAGS_PARTSCAN)
36580                 loop_reread_partitions(lo, bdev);
36581         return 0;
36582
36583   out_putf:
36584         fput(file);
36585 +       if (virt_file)
36586 +               fput(virt_file);
36587   out:
36588         return error;
36589  }
36590 @@ -876,7 +900,7 @@ static int loop_prepare_queue(struct loop_device *lo)
36591  static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36592                        struct block_device *bdev, unsigned int arg)
36593  {
36594 -       struct file     *file, *f;
36595 +       struct file     *file, *f, *virt_file = NULL;
36596         struct inode    *inode;
36597         struct address_space *mapping;
36598         unsigned lo_blocksize;
36599 @@ -891,6 +915,12 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36600         file = fget(arg);
36601         if (!file)
36602                 goto out;
36603 +       f = loop_real_file(file);
36604 +       if (f) {
36605 +               virt_file = file;
36606 +               file = f;
36607 +               get_file(file);
36608 +       }
36609
36610         error = -EBUSY;
36611         if (lo->lo_state != Lo_unbound)
36612 @@ -943,6 +973,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36613         lo->lo_device = bdev;
36614         lo->lo_flags = lo_flags;
36615         lo->lo_backing_file = file;
36616 +       lo->lo_backing_virt_file = virt_file;
36617         lo->transfer = NULL;
36618         lo->ioctl = NULL;
36619         lo->lo_sizelimit = 0;
36620 @@ -975,6 +1006,8 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36621
36622   out_putf:
36623         fput(file);
36624 +       if (virt_file)
36625 +               fput(virt_file);
36626   out:
36627         /* This is safe: open() is still holding a reference. */
36628         module_put(THIS_MODULE);
36629 @@ -1021,6 +1054,7 @@ loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
36630  static int loop_clr_fd(struct loop_device *lo)
36631  {
36632         struct file *filp = lo->lo_backing_file;
36633 +       struct file *virt_filp = lo->lo_backing_virt_file;
36634         gfp_t gfp = lo->old_gfp_mask;
36635         struct block_device *bdev = lo->lo_device;
36636
36637 @@ -1052,6 +1086,7 @@ static int loop_clr_fd(struct loop_device *lo)
36638         spin_lock_irq(&lo->lo_lock);
36639         lo->lo_state = Lo_rundown;
36640         lo->lo_backing_file = NULL;
36641 +       lo->lo_backing_virt_file = NULL;
36642         spin_unlock_irq(&lo->lo_lock);
36643
36644         loop_release_xfer(lo);
36645 @@ -1096,6 +1131,8 @@ static int loop_clr_fd(struct loop_device *lo)
36646          * bd_mutex which is usually taken before lo_ctl_mutex.
36647          */
36648         fput(filp);
36649 +       if (virt_filp)
36650 +               fput(virt_filp);
36651         return 0;
36652  }
36653
36654 diff --git a/drivers/block/loop.h b/drivers/block/loop.h
36655 index fb2237c..c3888c5 100644
36656 --- a/drivers/block/loop.h
36657 +++ b/drivers/block/loop.h
36658 @@ -46,7 +46,7 @@ struct loop_device {
36659         int             (*ioctl)(struct loop_device *, int cmd,
36660                                  unsigned long arg);
36661
36662 -       struct file *   lo_backing_file;
36663 +       struct file *   lo_backing_file, *lo_backing_virt_file;
36664         struct block_device *lo_device;
36665         unsigned        lo_blocksize;
36666         void            *key_data;
36667 diff --git a/fs/aufs/f_op.c b/fs/aufs/f_op.c
36668 index 7e03507..f31af1d 100644
36669 --- a/fs/aufs/f_op.c
36670 +++ b/fs/aufs/f_op.c
36671 @@ -360,7 +360,7 @@ static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
36672         if (IS_ERR(h_file))
36673                 goto out;
36674
36675 -       if (au_test_loopback_kthread()) {
36676 +       if (0 && au_test_loopback_kthread()) {
36677                 au_warn_loopback(h_file->f_path.dentry->d_sb);
36678                 if (file->f_mapping != h_file->f_mapping) {
36679                         file->f_mapping = h_file->f_mapping;
36680 diff --git a/fs/aufs/loop.c b/fs/aufs/loop.c
36681 index 61e9197..9910bea 100644
36682 --- a/fs/aufs/loop.c
36683 +++ b/fs/aufs/loop.c
36684 @@ -132,3 +132,19 @@ void au_loopback_fin(void)
36685                 symbol_put(loop_backing_file);
36686         au_delayed_kfree(au_warn_loopback_array);
36687  }
36688 +
36689 +/* ---------------------------------------------------------------------- */
36690 +
36691 +/* support the loopback block device insude aufs */
36692 +
36693 +struct file *aufs_real_loop(struct file *file)
36694 +{
36695 +       struct file *f;
36696 +
36697 +       BUG_ON(!au_test_aufs(file->f_path.dentry->d_sb));
36698 +       fi_read_lock(file);
36699 +       f = au_hf_top(file);
36700 +       fi_read_unlock(file);
36701 +       AuDebugOn(!f);
36702 +       return f;
36703 +}
36704 diff --git a/fs/aufs/loop.h b/fs/aufs/loop.h
36705 index e2df495..36e5052 100644
36706 --- a/fs/aufs/loop.h
36707 +++ b/fs/aufs/loop.h
36708 @@ -25,7 +25,11 @@ void au_warn_loopback(struct super_block *h_sb);
36709
36710  int au_loopback_init(void);
36711  void au_loopback_fin(void);
36712 +
36713 +struct file *aufs_real_loop(struct file *file);
36714  #else
36715 +AuStub(struct file *, loop_backing_file, return NULL)
36716 +
36717  AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
36718            struct dentry *h_adding)
36719  AuStubInt0(au_test_loopback_kthread, void)
36720 @@ -33,6 +37,8 @@ AuStubVoid(au_warn_loopback, struct super_block *h_sb)
36721
36722  AuStubInt0(au_loopback_init, void)
36723  AuStubVoid(au_loopback_fin, void)
36724 +
36725 +AuStub(struct file *, aufs_real_loop, return NULL, struct file *file)
36726  #endif /* BLK_DEV_LOOP */
36727
36728  #endif /* __KERNEL__ */
36729 diff --git a/fs/aufs/super.c b/fs/aufs/super.c
36730 index 0e9bbc3..b90d52f 100644
36731 --- a/fs/aufs/super.c
36732 +++ b/fs/aufs/super.c
36733 @@ -837,7 +837,10 @@ static const struct super_operations aufs_sop = {
36734         .statfs         = aufs_statfs,
36735         .put_super      = aufs_put_super,
36736         .sync_fs        = aufs_sync_fs,
36737 -       .remount_fs     = aufs_remount_fs
36738 +       .remount_fs     = aufs_remount_fs,
36739 +#ifdef CONFIG_AUFS_BDEV_LOOP
36740 +       .real_loop      = aufs_real_loop
36741 +#endif
36742  };
36743
36744  /* ---------------------------------------------------------------------- */
36745 diff --git a/include/linux/fs.h b/include/linux/fs.h
36746 index 3668627..2cbc8fe 100644
36747 --- a/include/linux/fs.h
36748 +++ b/include/linux/fs.h
36749 @@ -1810,6 +1810,10 @@ struct super_operations {
36750                                   struct shrink_control *);
36751         long (*free_cached_objects)(struct super_block *,
36752                                     struct shrink_control *);
36753 +#if defined(CONFIG_BLK_DEV_LOOP) ||  defined(CONFIG_BLK_DEV_LOOP_MODULE)
36754 +       /* and aufs */
36755 +       struct file *(*real_loop)(struct file *);
36756 +#endif
36757  };
36758
36759  /*