]> git.pld-linux.org Git - packages/kernel.git/blob - kernel-aufs4.patch
- initial update for 4.9.0
[packages/kernel.git] / kernel-aufs4.patch
1 aufs4.x-rcN kbuild patch
2
3 diff --git a/fs/Kconfig b/fs/Kconfig
4 index 2bc7ad7..3049386 100644
5 --- a/fs/Kconfig
6 +++ b/fs/Kconfig
7 @@ -245,6 +245,7 @@ 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 ed2b632..aa6d14b 100644
17 --- a/fs/Makefile
18 +++ b/fs/Makefile
19 @@ -129,3 +129,4 @@ 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 185f8ea..5deb0d1 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 f593300..8a17054 100644
40 --- a/MAINTAINERS
41 +++ b/MAINTAINERS
42 @@ -2256,6 +2256,19 @@ 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 c9f2107..005e292 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 5c7cc95..df0268c 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 350a2c8..6f42279 100644
106 --- a/fs/fcntl.c
107 +++ b/fs/fcntl.c
108 @@ -29,7 +29,7 @@
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 @@ -60,6 +60,8 @@ 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 7e3ef3a..675fe84 100644
128 --- a/fs/inode.c
129 +++ b/fs/inode.c
130 @@ -1593,7 +1593,7 @@ 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 66215a7..a1da117 100644
141 --- a/fs/read_write.c
142 +++ b/fs/read_write.c
143 @@ -515,6 +515,28 @@ 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 dd9bf7e..9326c2a 100644
174 --- a/fs/splice.c
175 +++ b/fs/splice.c
176 @@ -1111,8 +1111,8 @@ 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 @@ -1128,9 +1128,9 @@ 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/include/linux/file.h b/include/linux/file.h
201 index 7444f5f..bdac0be 100644
202 --- a/include/linux/file.h
203 +++ b/include/linux/file.h
204 @@ -19,6 +19,7 @@ struct dentry;
205  struct path;
206  extern struct file *alloc_file(struct path *, fmode_t mode,
207         const struct file_operations *fop);
208 +extern struct file *get_empty_filp(void);
209  
210  static inline void fput_light(struct file *file, int fput_needed)
211  {
212 diff --git a/include/linux/fs.h b/include/linux/fs.h
213 index 901e25d..a71aa9e 100644
214 --- a/include/linux/fs.h
215 +++ b/include/linux/fs.h
216 @@ -1275,6 +1275,7 @@ extern void fasync_free(struct fasync_struct *);
217  /* can be called from interrupts */
218  extern void kill_fasync(struct fasync_struct **, int, int);
219  
220 +extern int setfl(int fd, struct file * filp, unsigned long arg);
221  extern void __f_setown(struct file *filp, struct pid *, enum pid_type, int force);
222  extern void f_setown(struct file *filp, unsigned long arg, int force);
223  extern void f_delown(struct file *filp);
224 @@ -1699,6 +1700,7 @@ struct file_operations {
225         ssize_t (*sendpage) (struct file *, struct page *, int, size_t, loff_t *, int);
226         unsigned long (*get_unmapped_area)(struct file *, unsigned long, unsigned long, unsigned long, unsigned long);
227         int (*check_flags)(int);
228 +       int (*setfl)(struct file *, unsigned long);
229         int (*flock) (struct file *, int, struct file_lock *);
230         ssize_t (*splice_write)(struct pipe_inode_info *, struct file *, loff_t *, size_t, unsigned int);
231         ssize_t (*splice_read)(struct file *, loff_t *, struct pipe_inode_info *, size_t, unsigned int);
232 @@ -1759,6 +1761,12 @@ ssize_t rw_copy_check_uvector(int type, const struct iovec __user * uvector,
233                               struct iovec *fast_pointer,
234                               struct iovec **ret_pointer);
235  
236 +typedef ssize_t (*vfs_readf_t)(struct file *, char __user *, size_t, loff_t *);
237 +typedef ssize_t (*vfs_writef_t)(struct file *, const char __user *, size_t,
238 +                               loff_t *);
239 +vfs_readf_t vfs_readf(struct file *file);
240 +vfs_writef_t vfs_writef(struct file *file);
241 +
242  extern ssize_t __vfs_read(struct file *, char __user *, size_t, loff_t *);
243  extern ssize_t __vfs_write(struct file *, const char __user *, size_t, loff_t *);
244  extern ssize_t vfs_read(struct file *, char __user *, size_t, loff_t *);
245 @@ -2123,6 +2131,7 @@ extern int current_umask(void);
246  extern void ihold(struct inode * inode);
247  extern void iput(struct inode *);
248  extern int generic_update_time(struct inode *, struct timespec *, int);
249 +extern int update_time(struct inode *, struct timespec *, int);
250  
251  /* /sys/fs */
252  extern struct kobject *fs_kobj;
253 diff --git a/include/linux/splice.h b/include/linux/splice.h
254 index da2751d..2e0fca6 100644
255 --- a/include/linux/splice.h
256 +++ b/include/linux/splice.h
257 @@ -83,4 +83,10 @@ extern void splice_shrink_spd(struct splice_pipe_desc *);
258  
259  extern const struct pipe_buf_operations page_cache_pipe_buf_ops;
260  extern const struct pipe_buf_operations default_pipe_buf_ops;
261 +
262 +extern long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
263 +                          loff_t *ppos, size_t len, unsigned int flags);
264 +extern long do_splice_to(struct file *in, loff_t *ppos,
265 +                        struct pipe_inode_info *pipe, size_t len,
266 +                        unsigned int flags);
267  #endif
268 aufs4.x-rcN mmap patch
269
270 diff --git a/fs/proc/base.c b/fs/proc/base.c
271 index ac0df4d..42255e5 100644
272 --- a/fs/proc/base.c
273 +++ b/fs/proc/base.c
274 @@ -1938,7 +1938,7 @@ static int map_files_get_link(struct dentry *dentry, struct path *path)
275         down_read(&mm->mmap_sem);
276         vma = find_exact_vma(mm, vm_start, vm_end);
277         if (vma && vma->vm_file) {
278 -               *path = vma->vm_file->f_path;
279 +               *path = vma_pr_or_file(vma)->f_path;
280                 path_get(path);
281                 rc = 0;
282         }
283 diff --git a/fs/proc/nommu.c b/fs/proc/nommu.c
284 index f8595e8..cb8eda0 100644
285 --- a/fs/proc/nommu.c
286 +++ b/fs/proc/nommu.c
287 @@ -45,7 +45,10 @@ static int nommu_region_show(struct seq_file *m, struct vm_region *region)
288         file = region->vm_file;
289  
290         if (file) {
291 -               struct inode *inode = file_inode(region->vm_file);
292 +               struct inode *inode;
293 +
294 +               file = vmr_pr_or_file(region);
295 +               inode = file_inode(file);
296                 dev = inode->i_sb->s_dev;
297                 ino = inode->i_ino;
298         }
299 diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
300 index f6fa99e..2750949 100644
301 --- a/fs/proc/task_mmu.c
302 +++ b/fs/proc/task_mmu.c
303 @@ -298,7 +298,10 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
304         const char *name = NULL;
305  
306         if (file) {
307 -               struct inode *inode = file_inode(vma->vm_file);
308 +               struct inode *inode;
309 +
310 +               file = vma_pr_or_file(vma);
311 +               inode = file_inode(file);
312                 dev = inode->i_sb->s_dev;
313                 ino = inode->i_ino;
314                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
315 @@ -1634,7 +1637,7 @@ static int show_numa_map(struct seq_file *m, void *v, int is_pid)
316         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
317         struct vm_area_struct *vma = v;
318         struct numa_maps *md = &numa_priv->md;
319 -       struct file *file = vma->vm_file;
320 +       struct file *file = vma_pr_or_file(vma);
321         struct mm_struct *mm = vma->vm_mm;
322         struct mm_walk walk = {
323                 .hugetlb_entry = gather_hugetlb_stats,
324 diff --git a/fs/proc/task_nommu.c b/fs/proc/task_nommu.c
325 index faacb0c..17b43be 100644
326 --- a/fs/proc/task_nommu.c
327 +++ b/fs/proc/task_nommu.c
328 @@ -163,7 +163,10 @@ static int nommu_vma_show(struct seq_file *m, struct vm_area_struct *vma,
329         file = vma->vm_file;
330  
331         if (file) {
332 -               struct inode *inode = file_inode(vma->vm_file);
333 +               struct inode *inode;
334 +
335 +               file = vma_pr_or_file(vma);
336 +               inode = file_inode(file);
337                 dev = inode->i_sb->s_dev;
338                 ino = inode->i_ino;
339                 pgoff = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
340 diff --git a/include/linux/mm.h b/include/linux/mm.h
341 index ef815b9..a772481 100644
342 --- a/include/linux/mm.h
343 +++ b/include/linux/mm.h
344 @@ -1266,6 +1266,28 @@ static inline int fixup_user_fault(struc
345  }
346  #endif
347  
348 +extern void vma_do_file_update_time(struct vm_area_struct *, const char[], int);
349 +extern struct file *vma_do_pr_or_file(struct vm_area_struct *, const char[],
350 +               int);
351 +extern void vma_do_get_file(struct vm_area_struct *, const char[], int);
352 +extern void vma_do_fput(struct vm_area_struct *, const char[], int);
353 +
354 +#define vma_file_update_time(vma)      vma_do_file_update_time(vma, __func__, \
355 +               __LINE__)
356 +#define vma_pr_or_file(vma)            vma_do_pr_or_file(vma, __func__, \
357 +               __LINE__)
358 +#define vma_get_file(vma)              vma_do_get_file(vma, __func__, __LINE__)
359 +#define vma_fput(vma)                  vma_do_fput(vma, __func__, __LINE__)
360 +
361 +#ifndef CONFIG_MMU
362 +extern struct file *vmr_do_pr_or_file(struct vm_region *, const char[], int);
363 +extern void vmr_do_fput(struct vm_region *, const char[], int);
364 +
365 +#define vmr_pr_or_file(region)         vmr_do_pr_or_file(region, __func__, \
366 +               __LINE__)
367 +#define vmr_fput(region)               vmr_do_fput(region, __func__, __LINE__)
368 +#endif /* !CONFIG_MMU */
369 +
370  extern int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len,
371                 unsigned int gup_flags);
372  extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
373 diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
374 index 903200f..55fc528 100644
375 --- a/include/linux/mm_types.h
376 +++ b/include/linux/mm_types.h
377 @@ -275,6 +275,7 @@ struct vm_region {
378         unsigned long   vm_top;         /* region allocated to here */
379         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
380         struct file     *vm_file;       /* the backing file or NULL */
381 +       struct file     *vm_prfile;     /* the virtual backing file or NULL */
382  
383         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
384         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
385 @@ -349,6 +350,7 @@ struct vm_area_struct {
386         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
387                                            units */
388         struct file * vm_file;          /* File we map to (can be NULL). */
389 +       struct file *vm_prfile;         /* shadow of vm_file */
390         void * vm_private_data;         /* was vm_pte (shared mem) */
391  
392  #ifndef CONFIG_MMU
393 diff --git a/kernel/fork.c b/kernel/fork.c
394 index beb3172..ad4cfa8 100644
395 --- a/kernel/fork.c
396 +++ b/kernel/fork.c
397 @@ -477,7 +477,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
398                         struct inode *inode = file_inode(file);
399                         struct address_space *mapping = file->f_mapping;
400  
401 -                       get_file(file);
402 +                       vma_get_file(tmp);
403                         if (tmp->vm_flags & VM_DENYWRITE)
404                                 atomic_dec(&inode->i_writecount);
405                         i_mmap_lock_write(mapping);
406 diff --git a/mm/Makefile b/mm/Makefile
407 index 2ca1faf..6b9da3f 100644
408 --- a/mm/Makefile
409 +++ b/mm/Makefile
410 @@ -40,7 +40,7 @@ obj-y                 := filemap.o mempool.o oom_kill.o \
411                            mm_init.o mmu_context.o percpu.o slab_common.o \
412                            compaction.o vmacache.o \
413                            interval_tree.o list_lru.o workingset.o \
414 -                          debug.o $(mmu-y)
415 +                          prfile.o debug.o $(mmu-y)
416  
417  obj-y += init-mm.o
418  
419 diff --git a/mm/filemap.c b/mm/filemap.c
420 index 2d0986a..4a31bad 100644
421 --- a/mm/filemap.c
422 +++ b/mm/filemap.c
423 @@ -2284,7 +2284,7 @@ int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
424         int ret = VM_FAULT_LOCKED;
425  
426         sb_start_pagefault(inode->i_sb);
427 -       file_update_time(vma->vm_file);
428 +       vma_file_update_time(vma);
429         lock_page(page);
430         if (page->mapping != inode->i_mapping) {
431                 unlock_page(page);
432 diff --git a/mm/memory.c b/mm/memory.c
433 index 793fe0f..45f39f3 100644
434 --- a/mm/memory.c
435 +++ b/mm/memory.c
436 @@ -2113,7 +2113,7 @@ static inline int wp_page_reuse(struct fault_env *fe, pte_t orig_pte,
437                 }
438  
439                 if (!page_mkwrite)
440 -                       file_update_time(vma->vm_file);
441 +                       vma_file_update_time(vma);
442         }
443  
444         return VM_FAULT_WRITE;
445 diff --git a/mm/mmap.c b/mm/mmap.c
446 index ca9d91b..f3ebc5a 100644
447 --- a/mm/mmap.c
448 +++ b/mm/mmap.c
449 @@ -163,7 +163,7 @@ static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
450         if (vma->vm_ops && vma->vm_ops->close)
451                 vma->vm_ops->close(vma);
452         if (vma->vm_file)
453 -               fput(vma->vm_file);
454 +               vma_fput(vma);
455         mpol_put(vma_policy(vma));
456         kmem_cache_free(vm_area_cachep, vma);
457         return next;
458 @@ -790,7 +790,7 @@ again:
459         if (remove_next) {
460                 if (file) {
461                         uprobe_munmap(next, next->vm_start, next->vm_end);
462 -                       fput(file);
463 +                       vma_fput(vma);
464                 }
465                 if (next->anon_vma)
466                         anon_vma_merge(vma, next);
467 @@ -1574,8 +1574,8 @@ out:
468         return addr;
469  
470  unmap_and_free_vma:
471 +       vma_fput(vma);
472         vma->vm_file = NULL;
473 -       fput(file);
474  
475         /* Undo any partial mapping done by a device driver. */
476         unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
477 @@ -2380,7 +2380,7 @@ static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
478                 goto out_free_mpol;
479  
480         if (new->vm_file)
481 -               get_file(new->vm_file);
482 +               vma_get_file(new);
483  
484         if (new->vm_ops && new->vm_ops->open)
485                 new->vm_ops->open(new);
486 @@ -2399,7 +2399,7 @@ static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
487         if (new->vm_ops && new->vm_ops->close)
488                 new->vm_ops->close(new);
489         if (new->vm_file)
490 -               fput(new->vm_file);
491 +               vma_fput(new);
492         unlink_anon_vmas(new);
493   out_free_mpol:
494         mpol_put(vma_policy(new));
495 @@ -2550,7 +2550,7 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
496         struct vm_area_struct *vma;
497         unsigned long populate = 0;
498         unsigned long ret = -EINVAL;
499 -       struct file *file;
500 +       struct file *file, *prfile;
501  
502         pr_warn_once("%s (%d) uses deprecated remap_file_pages() syscall. See Documentation/vm/remap_file_pages.txt.\n",
503                      current->comm, current->pid);
504 @@ -2625,10 +2625,27 @@ SYSCALL_DEFINE5(remap_file_pages, unsigned long, start, unsigned long, size,
505                 }
506         }
507  
508 -       file = get_file(vma->vm_file);
509 +       vma_get_file(vma);
510 +       file = vma->vm_file;
511 +       prfile = vma->vm_prfile;
512         ret = do_mmap_pgoff(vma->vm_file, start, size,
513                         prot, flags, pgoff, &populate);
514 +       if (!IS_ERR_VALUE(ret) && file && prfile) {
515 +               struct vm_area_struct *new_vma;
516 +
517 +               new_vma = find_vma(mm, ret);
518 +               if (!new_vma->vm_prfile)
519 +                       new_vma->vm_prfile = prfile;
520 +               if (new_vma != vma)
521 +                       get_file(prfile);
522 +       }
523 +       /*
524 +        * two fput()s instead of vma_fput(vma),
525 +        * coz vma may not be available anymore.
526 +        */
527         fput(file);
528 +       if (prfile)
529 +               fput(prfile);
530  out:
531         up_write(&mm->mmap_sem);
532         if (populate)
533 @@ -2903,7 +2920,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
534                 if (anon_vma_clone(new_vma, vma))
535                         goto out_free_mempol;
536                 if (new_vma->vm_file)
537 -                       get_file(new_vma->vm_file);
538 +                       vma_get_file(new_vma);
539                 if (new_vma->vm_ops && new_vma->vm_ops->open)
540                         new_vma->vm_ops->open(new_vma);
541                 vma_link(mm, new_vma, prev, rb_link, rb_parent);
542 diff --git a/mm/nommu.c b/mm/nommu.c
543 index 95daf81..5086a29 100644
544 --- a/mm/nommu.c
545 +++ b/mm/nommu.c
546 @@ -644,7 +644,7 @@ static void __put_nommu_region(struct vm_region *region)
547                 up_write(&nommu_region_sem);
548  
549                 if (region->vm_file)
550 -                       fput(region->vm_file);
551 +                       vmr_fput(region);
552  
553                 /* IO memory and memory shared directly out of the pagecache
554                  * from ramfs/tmpfs mustn't be released here */
555 @@ -802,7 +802,7 @@ static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
556         if (vma->vm_ops && vma->vm_ops->close)
557                 vma->vm_ops->close(vma);
558         if (vma->vm_file)
559 -               fput(vma->vm_file);
560 +               vma_fput(vma);
561         put_nommu_region(vma->vm_region);
562         kmem_cache_free(vm_area_cachep, vma);
563  }
564 @@ -1328,7 +1328,7 @@ unsigned long do_mmap(struct file *file,
565                                         goto error_just_free;
566                                 }
567                         }
568 -                       fput(region->vm_file);
569 +                       vmr_fput(region);
570                         kmem_cache_free(vm_region_jar, region);
571                         region = pregion;
572                         result = start;
573 @@ -1403,10 +1403,10 @@ error_just_free:
574         up_write(&nommu_region_sem);
575  error:
576         if (region->vm_file)
577 -               fput(region->vm_file);
578 +               vmr_fput(region);
579         kmem_cache_free(vm_region_jar, region);
580         if (vma->vm_file)
581 -               fput(vma->vm_file);
582 +               vma_fput(vma);
583         kmem_cache_free(vm_area_cachep, vma);
584         return ret;
585  
586 diff --git a/mm/prfile.c b/mm/prfile.c
587 new file mode 100644
588 index 0000000..b323b8a
589 --- /dev/null
590 +++ b/mm/prfile.c
591 @@ -0,0 +1,86 @@
592 +/*
593 + * Mainly for aufs which mmap(2) diffrent file and wants to print different path
594 + * in /proc/PID/maps.
595 + * Call these functions via macros defined in linux/mm.h.
596 + *
597 + * See Documentation/filesystems/aufs/design/06mmap.txt
598 + *
599 + * Copyright (c) 2014 Junjro R. Okajima
600 + * Copyright (c) 2014 Ian Campbell
601 + */
602 +
603 +#include <linux/mm.h>
604 +#include <linux/file.h>
605 +#include <linux/fs.h>
606 +
607 +/* #define PRFILE_TRACE */
608 +static inline void prfile_trace(struct file *f, struct file *pr,
609 +                             const char func[], int line, const char func2[])
610 +{
611 +#ifdef PRFILE_TRACE
612 +       if (pr)
613 +               pr_info("%s:%d: %s, %s\n", func, line, func2,
614 +                       f ? (char *)f->f_path.dentry->d_name.name : "(null)");
615 +#endif
616 +}
617 +
618 +void vma_do_file_update_time(struct vm_area_struct *vma, const char func[],
619 +                            int line)
620 +{
621 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
622 +
623 +       prfile_trace(f, pr, func, line, __func__);
624 +       file_update_time(f);
625 +       if (f && pr)
626 +               file_update_time(pr);
627 +}
628 +
629 +struct file *vma_do_pr_or_file(struct vm_area_struct *vma, const char func[],
630 +                              int line)
631 +{
632 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
633 +
634 +       prfile_trace(f, pr, func, line, __func__);
635 +       return (f && pr) ? pr : f;
636 +}
637 +
638 +void vma_do_get_file(struct vm_area_struct *vma, const char func[], int line)
639 +{
640 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
641 +
642 +       prfile_trace(f, pr, func, line, __func__);
643 +       get_file(f);
644 +       if (f && pr)
645 +               get_file(pr);
646 +}
647 +
648 +void vma_do_fput(struct vm_area_struct *vma, const char func[], int line)
649 +{
650 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
651 +
652 +       prfile_trace(f, pr, func, line, __func__);
653 +       fput(f);
654 +       if (f && pr)
655 +               fput(pr);
656 +}
657 +
658 +#ifndef CONFIG_MMU
659 +struct file *vmr_do_pr_or_file(struct vm_region *region, const char func[],
660 +                              int line)
661 +{
662 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
663 +
664 +       prfile_trace(f, pr, func, line, __func__);
665 +       return (f && pr) ? pr : f;
666 +}
667 +
668 +void vmr_do_fput(struct vm_region *region, const char func[], int line)
669 +{
670 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
671 +
672 +       prfile_trace(f, pr, func, line, __func__);
673 +       fput(f);
674 +       if (f && pr)
675 +               fput(pr);
676 +}
677 +#endif /* !CONFIG_MMU */
678 aufs4.x-rcN standalone patch
679
680 diff --git a/fs/dcache.c b/fs/dcache.c
681 index df0268c..76280ee 100644
682 --- a/fs/dcache.c
683 +++ b/fs/dcache.c
684 @@ -1272,6 +1272,7 @@ rename_retry:
685         seq = 1;
686         goto again;
687  }
688 +EXPORT_SYMBOL_GPL(d_walk);
689  
690  /*
691   * Search for at least 1 mount point in the dentry's subdirs.
692 diff --git a/fs/exec.c b/fs/exec.c
693 index 6fcfb3f..ed9d646 100644
694 --- a/fs/exec.c
695 +++ b/fs/exec.c
696 @@ -104,6 +104,7 @@ bool path_noexec(const struct path *path)
697         return (path->mnt->mnt_flags & MNT_NOEXEC) ||
698                (path->mnt->mnt_sb->s_iflags & SB_I_NOEXEC);
699  }
700 +EXPORT_SYMBOL_GPL(path_noexec);
701  
702  #ifdef CONFIG_USELIB
703  /*
704 diff --git a/fs/fcntl.c b/fs/fcntl.c
705 index 6f42279..04fd33c 100644
706 --- a/fs/fcntl.c
707 +++ b/fs/fcntl.c
708 @@ -82,6 +82,7 @@ int setfl(int fd, struct file * filp, unsigned long arg)
709   out:
710         return error;
711  }
712 +EXPORT_SYMBOL_GPL(setfl);
713  
714  static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
715                       int force)
716 diff --git a/fs/file_table.c b/fs/file_table.c
717 index ad17e05..ae9f267 100644
718 --- a/fs/file_table.c
719 +++ b/fs/file_table.c
720 @@ -147,6 +147,7 @@ over:
721         }
722         return ERR_PTR(-ENFILE);
723  }
724 +EXPORT_SYMBOL_GPL(get_empty_filp);
725  
726  /**
727   * alloc_file - allocate and initialize a 'struct file'
728 @@ -258,6 +259,7 @@ void flush_delayed_fput(void)
729  {
730         delayed_fput(NULL);
731  }
732 +EXPORT_SYMBOL_GPL(flush_delayed_fput);
733  
734  static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
735  
736 @@ -300,6 +302,7 @@ void __fput_sync(struct file *file)
737  }
738  
739  EXPORT_SYMBOL(fput);
740 +EXPORT_SYMBOL_GPL(__fput_sync);
741  
742  void put_filp(struct file *file)
743  {
744 @@ -308,6 +311,7 @@ void put_filp(struct file *file)
745                 file_free(file);
746         }
747  }
748 +EXPORT_SYMBOL_GPL(put_filp);
749  
750  void __init files_init(void)
751  { 
752 diff --git a/fs/inode.c b/fs/inode.c
753 index 675fe84..da063d3 100644
754 --- a/fs/inode.c
755 +++ b/fs/inode.c
756 @@ -1602,6 +1602,7 @@ int update_time(struct inode *inode, struct timespec *time, int flags)
757  
758         return update_time(inode, time, flags);
759  }
760 +EXPORT_SYMBOL_GPL(update_time);
761  
762  /**
763   *     touch_atime     -       update the access time
764 diff --git a/fs/namespace.c b/fs/namespace.c
765 index 7bb2cda..88ec098 100644
766 --- a/fs/namespace.c
767 +++ b/fs/namespace.c
768 @@ -463,6 +463,7 @@ void __mnt_drop_write(struct vfsmount *mnt)
769         mnt_dec_writers(real_mount(mnt));
770         preempt_enable();
771  }
772 +EXPORT_SYMBOL_GPL(__mnt_drop_write);
773  
774  /**
775   * mnt_drop_write - give up write access to a mount
776 @@ -1812,6 +1813,7 @@ int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
777         }
778         return 0;
779  }
780 +EXPORT_SYMBOL_GPL(iterate_mounts);
781  
782  static void cleanup_group_ids(struct mount *mnt, struct mount *end)
783  {
784 diff --git a/fs/notify/group.c b/fs/notify/group.c
785 index b47f7cf..618bc9e 100644
786 --- a/fs/notify/group.c
787 +++ b/fs/notify/group.c
788 @@ -22,6 +22,7 @@
789  #include <linux/srcu.h>
790  #include <linux/rculist.h>
791  #include <linux/wait.h>
792 +#include <linux/module.h>
793  
794  #include <linux/fsnotify_backend.h>
795  #include "fsnotify.h"
796 @@ -100,6 +101,7 @@ void fsnotify_get_group(struct fsnotify_group *group)
797  {
798         atomic_inc(&group->refcnt);
799  }
800 +EXPORT_SYMBOL_GPL(fsnotify_get_group);
801  
802  /*
803   * Drop a reference to a group.  Free it if it's through.
804 @@ -109,6 +111,7 @@ void fsnotify_put_group(struct fsnotify_group *group)
805         if (atomic_dec_and_test(&group->refcnt))
806                 fsnotify_final_destroy_group(group);
807  }
808 +EXPORT_SYMBOL_GPL(fsnotify_put_group);
809  
810  /*
811   * Create a new fsnotify_group and hold a reference for the group returned.
812 @@ -137,6 +140,7 @@ struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops)
813  
814         return group;
815  }
816 +EXPORT_SYMBOL_GPL(fsnotify_alloc_group);
817  
818  int fsnotify_fasync(int fd, struct file *file, int on)
819  {
820 diff --git a/fs/notify/mark.c b/fs/notify/mark.c
821 index d3fea0b..5fc06ad 100644
822 --- a/fs/notify/mark.c
823 +++ b/fs/notify/mark.c
824 @@ -113,6 +113,7 @@ void fsnotify_put_mark(struct fsnotify_mark *mark)
825                 mark->free_mark(mark);
826         }
827  }
828 +EXPORT_SYMBOL_GPL(fsnotify_put_mark);
829  
830  /* Calculate mask of events for a list of marks */
831  u32 fsnotify_recalc_mask(struct hlist_head *head)
832 @@ -230,6 +231,7 @@ void fsnotify_destroy_mark(struct fsnotify_mark *mark,
833         mutex_unlock(&group->mark_mutex);
834         fsnotify_free_mark(mark);
835  }
836 +EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
837  
838  void fsnotify_destroy_marks(struct hlist_head *head, spinlock_t *lock)
839  {
840 @@ -415,6 +417,7 @@ err:
841  
842         return ret;
843  }
844 +EXPORT_SYMBOL_GPL(fsnotify_add_mark);
845  
846  int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
847                       struct inode *inode, struct vfsmount *mnt, int allow_dups)
848 @@ -533,6 +536,7 @@ void fsnotify_init_mark(struct fsnotify_mark *mark,
849         atomic_set(&mark->refcnt, 1);
850         mark->free_mark = free_mark;
851  }
852 +EXPORT_SYMBOL_GPL(fsnotify_init_mark);
853  
854  /*
855   * Destroy all marks in destroy_list, waits for SRCU period to finish before
856 diff --git a/fs/open.c b/fs/open.c
857 index 4fd6e25..ec6f532 100644
858 --- a/fs/open.c
859 +++ b/fs/open.c
860 @@ -64,6 +64,7 @@ int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
861         inode_unlock(dentry->d_inode);
862         return ret;
863  }
864 +EXPORT_SYMBOL_GPL(do_truncate);
865  
866  long vfs_truncate(const struct path *path, loff_t length)
867  {
868 @@ -678,6 +679,7 @@ int open_check_o_direct(struct file *f)
869         }
870         return 0;
871  }
872 +EXPORT_SYMBOL_GPL(open_check_o_direct);
873  
874  static int do_dentry_open(struct file *f,
875                           struct inode *inode,
876 diff --git a/fs/read_write.c b/fs/read_write.c
877 index a1da117..c643215 100644
878 --- a/fs/read_write.c
879 +++ b/fs/read_write.c
880 @@ -525,6 +525,7 @@ vfs_readf_t vfs_readf(struct file *file)
881                 return new_sync_read;
882         return ERR_PTR(-ENOSYS);
883  }
884 +EXPORT_SYMBOL_GPL(vfs_readf);
885  
886  vfs_writef_t vfs_writef(struct file *file)
887  {
888 @@ -536,6 +537,7 @@ vfs_writef_t vfs_writef(struct file *file)
889                 return new_sync_write;
890         return ERR_PTR(-ENOSYS);
891  }
892 +EXPORT_SYMBOL_GPL(vfs_writef);
893  
894  ssize_t __kernel_write(struct file *file, const char *buf, size_t count, loff_t *pos)
895  {
896 diff --git a/fs/splice.c b/fs/splice.c
897 index 9326c2a..0606690 100644
898 --- a/fs/splice.c
899 +++ b/fs/splice.c
900 @@ -1124,6 +1124,7 @@ long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
901  
902         return splice_write(pipe, out, ppos, len, flags);
903  }
904 +EXPORT_SYMBOL_GPL(do_splice_from);
905  
906  /*
907   * Attempt to initiate a splice from a file to a pipe.
908 @@ -1153,6 +1154,7 @@ long do_splice_to(struct file *in, loff_t *ppos,
909  
910         return splice_read(in, ppos, pipe, len, flags);
911  }
912 +EXPORT_SYMBOL_GPL(do_splice_to);
913  
914  /**
915   * splice_direct_to_actor - splices data directly between two non-pipes
916 diff --git a/fs/xattr.c b/fs/xattr.c
917 index c243905..b60dc60 100644
918 --- a/fs/xattr.c
919 +++ b/fs/xattr.c
920 @@ -214,6 +214,7 @@ vfs_getxattr_alloc(struct dentry *dentry, const char *name, char **xattr_value,
921         *xattr_value = value;
922         return error;
923  }
924 +EXPORT_SYMBOL_GPL(vfs_getxattr_alloc);
925  
926  ssize_t
927  vfs_getxattr(struct dentry *dentry, const char *name, void *value, size_t size)
928 diff --git a/kernel/task_work.c b/kernel/task_work.c
929 index d513051..e056d54 100644
930 --- a/kernel/task_work.c
931 +++ b/kernel/task_work.c
932 @@ -119,3 +119,4 @@ void task_work_run(void)
933                 } while (work);
934         }
935  }
936 +EXPORT_SYMBOL_GPL(task_work_run);
937 diff --git a/security/commoncap.c b/security/commoncap.c
938 index 14540bd..4e3b242 100644
939 --- a/security/commoncap.c
940 +++ b/security/commoncap.c
941 @@ -1066,12 +1066,14 @@ int cap_mmap_addr(unsigned long addr)
942         }
943         return ret;
944  }
945 +EXPORT_SYMBOL_GPL(cap_mmap_addr);
946  
947  int cap_mmap_file(struct file *file, unsigned long reqprot,
948                   unsigned long prot, unsigned long flags)
949  {
950         return 0;
951  }
952 +EXPORT_SYMBOL_GPL(cap_mmap_file);
953  
954  #ifdef CONFIG_SECURITY
955  
956 diff --git a/security/device_cgroup.c b/security/device_cgroup.c
957 index 03c1652..f88c84b 100644
958 --- a/security/device_cgroup.c
959 +++ b/security/device_cgroup.c
960 @@ -7,6 +7,7 @@
961  #include <linux/device_cgroup.h>
962  #include <linux/cgroup.h>
963  #include <linux/ctype.h>
964 +#include <linux/export.h>
965  #include <linux/list.h>
966  #include <linux/uaccess.h>
967  #include <linux/seq_file.h>
968 @@ -849,6 +850,7 @@ int __devcgroup_inode_permission(struct inode *inode, int mask)
969         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
970                         access);
971  }
972 +EXPORT_SYMBOL_GPL(__devcgroup_inode_permission);
973  
974  int devcgroup_inode_mknod(int mode, dev_t dev)
975  {
976 diff --git a/security/security.c b/security/security.c
977 index 4838e7f..36c741e 100644
978 --- a/security/security.c
979 +++ b/security/security.c
980 @@ -434,6 +434,7 @@ int security_path_rmdir(const struct path *dir, struct dentry *dentry)
981                 return 0;
982         return call_int_hook(path_rmdir, 0, dir, dentry);
983  }
984 +EXPORT_SYMBOL_GPL(security_path_rmdir);
985  
986  int security_path_unlink(const struct path *dir, struct dentry *dentry)
987  {
988 @@ -450,6 +451,7 @@ int security_path_symlink(const struct path *dir, struct dentry *dentry,
989                 return 0;
990         return call_int_hook(path_symlink, 0, dir, dentry, old_name);
991  }
992 +EXPORT_SYMBOL_GPL(security_path_symlink);
993  
994  int security_path_link(struct dentry *old_dentry, const struct path *new_dir,
995                        struct dentry *new_dentry)
996 @@ -458,6 +460,7 @@ int security_path_link(struct dentry *old_dentry, const struct path *new_dir,
997                 return 0;
998         return call_int_hook(path_link, 0, old_dentry, new_dir, new_dentry);
999  }
1000 +EXPORT_SYMBOL_GPL(security_path_link);
1001  
1002  int security_path_rename(const struct path *old_dir, struct dentry *old_dentry,
1003                          const struct path *new_dir, struct dentry *new_dentry,
1004 @@ -485,6 +488,7 @@ int security_path_truncate(const struct path *path)
1005                 return 0;
1006         return call_int_hook(path_truncate, 0, path);
1007  }
1008 +EXPORT_SYMBOL_GPL(security_path_truncate);
1009  
1010  int security_path_chmod(const struct path *path, umode_t mode)
1011  {
1012 @@ -492,6 +496,7 @@ int security_path_chmod(const struct path *path, umode_t mode)
1013                 return 0;
1014         return call_int_hook(path_chmod, 0, path, mode);
1015  }
1016 +EXPORT_SYMBOL_GPL(security_path_chmod);
1017  
1018  int security_path_chown(const struct path *path, kuid_t uid, kgid_t gid)
1019  {
1020 @@ -499,6 +504,7 @@ int security_path_chown(const struct path *path, kuid_t uid, kgid_t gid)
1021                 return 0;
1022         return call_int_hook(path_chown, 0, path, uid, gid);
1023  }
1024 +EXPORT_SYMBOL_GPL(security_path_chown);
1025  
1026  int security_path_chroot(const struct path *path)
1027  {
1028 @@ -584,6 +590,7 @@ int security_inode_readlink(struct dentry *dentry)
1029                 return 0;
1030         return call_int_hook(inode_readlink, 0, dentry);
1031  }
1032 +EXPORT_SYMBOL_GPL(security_inode_readlink);
1033  
1034  int security_inode_follow_link(struct dentry *dentry, struct inode *inode,
1035                                bool rcu)
1036 @@ -599,6 +606,7 @@ int security_inode_permission(struct inode *inode, int mask)
1037                 return 0;
1038         return call_int_hook(inode_permission, 0, inode, mask);
1039  }
1040 +EXPORT_SYMBOL_GPL(security_inode_permission);
1041  
1042  int security_inode_setattr(struct dentry *dentry, struct iattr *attr)
1043  {
1044 @@ -758,6 +766,7 @@ int security_file_permission(struct file *file, int mask)
1045  
1046         return fsnotify_perm(file, mask);
1047  }
1048 +EXPORT_SYMBOL_GPL(security_file_permission);
1049  
1050  int security_file_alloc(struct file *file)
1051  {
1052 @@ -817,6 +826,7 @@ int security_mmap_file(struct file *file, unsigned long prot,
1053                 return ret;
1054         return ima_file_mmap(file, prot);
1055  }
1056 +EXPORT_SYMBOL_GPL(security_mmap_file);
1057  
1058  int security_mmap_addr(unsigned long addr)
1059  {
1060 diff -urN /usr/share/empty/Documentation/ABI/testing/debugfs-aufs linux/Documentation/ABI/testing/debugfs-aufs
1061 --- /usr/share/empty/Documentation/ABI/testing/debugfs-aufs     1970-01-01 01:00:00.000000000 +0100
1062 +++ linux/Documentation/ABI/testing/debugfs-aufs        2016-10-09 16:55:36.476034536 +0200
1063 @@ -0,0 +1,50 @@
1064 +What:          /debug/aufs/si_<id>/
1065 +Date:          March 2009
1066 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1067 +Description:
1068 +               Under /debug/aufs, a directory named si_<id> is created
1069 +               per aufs mount, where <id> is a unique id generated
1070 +               internally.
1071 +
1072 +What:          /debug/aufs/si_<id>/plink
1073 +Date:          Apr 2013
1074 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1075 +Description:
1076 +               It has three lines and shows the information about the
1077 +               pseudo-link. The first line is a single number
1078 +               representing a number of buckets. The second line is a
1079 +               number of pseudo-links per buckets (separated by a
1080 +               blank). The last line is a single number representing a
1081 +               total number of psedo-links.
1082 +               When the aufs mount option 'noplink' is specified, it
1083 +               will show "1\n0\n0\n".
1084 +
1085 +What:          /debug/aufs/si_<id>/xib
1086 +Date:          March 2009
1087 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1088 +Description:
1089 +               It shows the consumed blocks by xib (External Inode Number
1090 +               Bitmap), its block size and file size.
1091 +               When the aufs mount option 'noxino' is specified, it
1092 +               will be empty. About XINO files, see the aufs manual.
1093 +
1094 +What:          /debug/aufs/si_<id>/xino0, xino1 ... xinoN
1095 +Date:          March 2009
1096 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1097 +Description:
1098 +               It shows the consumed blocks by xino (External Inode Number
1099 +               Translation Table), its link count, block size and file
1100 +               size.
1101 +               When the aufs mount option 'noxino' is specified, it
1102 +               will be empty. About XINO files, see the aufs manual.
1103 +
1104 +What:          /debug/aufs/si_<id>/xigen
1105 +Date:          March 2009
1106 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1107 +Description:
1108 +               It shows the consumed blocks by xigen (External Inode
1109 +               Generation Table), its block size and file size.
1110 +               If CONFIG_AUFS_EXPORT is disabled, this entry will not
1111 +               be created.
1112 +               When the aufs mount option 'noxino' is specified, it
1113 +               will be empty. About XINO files, see the aufs manual.
1114 diff -urN /usr/share/empty/Documentation/ABI/testing/sysfs-aufs linux/Documentation/ABI/testing/sysfs-aufs
1115 --- /usr/share/empty/Documentation/ABI/testing/sysfs-aufs       1970-01-01 01:00:00.000000000 +0100
1116 +++ linux/Documentation/ABI/testing/sysfs-aufs  2016-10-09 16:55:36.476034536 +0200
1117 @@ -0,0 +1,31 @@
1118 +What:          /sys/fs/aufs/si_<id>/
1119 +Date:          March 2009
1120 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1121 +Description:
1122 +               Under /sys/fs/aufs, a directory named si_<id> is created
1123 +               per aufs mount, where <id> is a unique id generated
1124 +               internally.
1125 +
1126 +What:          /sys/fs/aufs/si_<id>/br0, br1 ... brN
1127 +Date:          March 2009
1128 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1129 +Description:
1130 +               It shows the abolute path of a member directory (which
1131 +               is called branch) in aufs, and its permission.
1132 +
1133 +What:          /sys/fs/aufs/si_<id>/brid0, brid1 ... bridN
1134 +Date:          July 2013
1135 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1136 +Description:
1137 +               It shows the id of a member directory (which is called
1138 +               branch) in aufs.
1139 +
1140 +What:          /sys/fs/aufs/si_<id>/xi_path
1141 +Date:          March 2009
1142 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1143 +Description:
1144 +               It shows the abolute path of XINO (External Inode Number
1145 +               Bitmap, Translation Table and Generation Table) file
1146 +               even if it is the default path.
1147 +               When the aufs mount option 'noxino' is specified, it
1148 +               will be empty. About XINO files, see the aufs manual.
1149 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt linux/Documentation/filesystems/aufs/design/01intro.txt
1150 --- /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt  1970-01-01 01:00:00.000000000 +0100
1151 +++ linux/Documentation/filesystems/aufs/design/01intro.txt     2016-10-09 16:55:36.479367956 +0200
1152 @@ -0,0 +1,170 @@
1153 +
1154 +# Copyright (C) 2005-2016 Junjiro R. Okajima
1155 +# 
1156 +# This program is free software; you can redistribute it and/or modify
1157 +# it under the terms of the GNU General Public License as published by
1158 +# the Free Software Foundation; either version 2 of the License, or
1159 +# (at your option) any later version.
1160 +# 
1161 +# This program is distributed in the hope that it will be useful,
1162 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1163 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1164 +# GNU General Public License for more details.
1165 +# 
1166 +# You should have received a copy of the GNU General Public License
1167 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1168 +
1169 +Introduction
1170 +----------------------------------------
1171 +
1172 +aufs [ei ju: ef es] | [a u f s]
1173 +1. abbrev. for "advanced multi-layered unification filesystem".
1174 +2. abbrev. for "another unionfs".
1175 +3. abbrev. for "auf das" in German which means "on the" in English.
1176 +   Ex. "Butter aufs Brot"(G) means "butter onto bread"(E).
1177 +   But "Filesystem aufs Filesystem" is hard to understand.
1178 +
1179 +AUFS is a filesystem with features:
1180 +- multi layered stackable unification filesystem, the member directory
1181 +  is called as a branch.
1182 +- branch permission and attribute, 'readonly', 'real-readonly',
1183 +  'readwrite', 'whiteout-able', 'link-able whiteout', etc. and their
1184 +  combination.
1185 +- internal "file copy-on-write".
1186 +- logical deletion, whiteout.
1187 +- dynamic branch manipulation, adding, deleting and changing permission.
1188 +- allow bypassing aufs, user's direct branch access.
1189 +- external inode number translation table and bitmap which maintains the
1190 +  persistent aufs inode number.
1191 +- seekable directory, including NFS readdir.
1192 +- file mapping, mmap and sharing pages.
1193 +- pseudo-link, hardlink over branches.
1194 +- loopback mounted filesystem as a branch.
1195 +- several policies to select one among multiple writable branches.
1196 +- revert a single systemcall when an error occurs in aufs.
1197 +- and more...
1198 +
1199 +
1200 +Multi Layered Stackable Unification Filesystem
1201 +----------------------------------------------------------------------
1202 +Most people already knows what it is.
1203 +It is a filesystem which unifies several directories and provides a
1204 +merged single directory. When users access a file, the access will be
1205 +passed/re-directed/converted (sorry, I am not sure which English word is
1206 +correct) to the real file on the member filesystem. The member
1207 +filesystem is called 'lower filesystem' or 'branch' and has a mode
1208 +'readonly' and 'readwrite.' And the deletion for a file on the lower
1209 +readonly branch is handled by creating 'whiteout' on the upper writable
1210 +branch.
1211 +
1212 +On LKML, there have been discussions about UnionMount (Jan Blunck,
1213 +Bharata B Rao and Valerie Aurora) and Unionfs (Erez Zadok). They took
1214 +different approaches to implement the merged-view.
1215 +The former tries putting it into VFS, and the latter implements as a
1216 +separate filesystem.
1217 +(If I misunderstand about these implementations, please let me know and
1218 +I shall correct it. Because it is a long time ago when I read their
1219 +source files last time).
1220 +
1221 +UnionMount's approach will be able to small, but may be hard to share
1222 +branches between several UnionMount since the whiteout in it is
1223 +implemented in the inode on branch filesystem and always
1224 +shared. According to Bharata's post, readdir does not seems to be
1225 +finished yet.
1226 +There are several missing features known in this implementations such as
1227 +- for users, the inode number may change silently. eg. copy-up.
1228 +- link(2) may break by copy-up.
1229 +- read(2) may get an obsoleted filedata (fstat(2) too).
1230 +- fcntl(F_SETLK) may be broken by copy-up.
1231 +- unnecessary copy-up may happen, for example mmap(MAP_PRIVATE) after
1232 +  open(O_RDWR).
1233 +
1234 +In linux-3.18, "overlay" filesystem (formerly known as "overlayfs") was
1235 +merged into mainline. This is another implementation of UnionMount as a
1236 +separated filesystem. All the limitations and known problems which
1237 +UnionMount are equally inherited to "overlay" filesystem.
1238 +
1239 +Unionfs has a longer history. When I started implementing a stackable
1240 +filesystem (Aug 2005), it already existed. It has virtual super_block,
1241 +inode, dentry and file objects and they have an array pointing lower
1242 +same kind objects. After contributing many patches for Unionfs, I
1243 +re-started my project AUFS (Jun 2006).
1244 +
1245 +In AUFS, the structure of filesystem resembles to Unionfs, but I
1246 +implemented my own ideas, approaches and enhancements and it became
1247 +totally different one.
1248 +
1249 +Comparing DM snapshot and fs based implementation
1250 +- the number of bytes to be copied between devices is much smaller.
1251 +- the type of filesystem must be one and only.
1252 +- the fs must be writable, no readonly fs, even for the lower original
1253 +  device. so the compression fs will not be usable. but if we use
1254 +  loopback mount, we may address this issue.
1255 +  for instance,
1256 +       mount /cdrom/squashfs.img /sq
1257 +       losetup /sq/ext2.img
1258 +       losetup /somewhere/cow
1259 +       dmsetup "snapshot /dev/loop0 /dev/loop1 ..."
1260 +- it will be difficult (or needs more operations) to extract the
1261 +  difference between the original device and COW.
1262 +- DM snapshot-merge may help a lot when users try merging. in the
1263 +  fs-layer union, users will use rsync(1).
1264 +
1265 +You may want to read my old paper "Filesystems in LiveCD"
1266 +(http://aufs.sourceforge.net/aufs2/report/sq/sq.pdf).
1267 +
1268 +
1269 +Several characters/aspects/persona of aufs
1270 +----------------------------------------------------------------------
1271 +
1272 +Aufs has several characters, aspects or persona.
1273 +1. a filesystem, callee of VFS helper
1274 +2. sub-VFS, caller of VFS helper for branches
1275 +3. a virtual filesystem which maintains persistent inode number
1276 +4. reader/writer of files on branches such like an application
1277 +
1278 +1. Callee of VFS Helper
1279 +As an ordinary linux filesystem, aufs is a callee of VFS. For instance,
1280 +unlink(2) from an application reaches sys_unlink() kernel function and
1281 +then vfs_unlink() is called. vfs_unlink() is one of VFS helper and it
1282 +calls filesystem specific unlink operation. Actually aufs implements the
1283 +unlink operation but it behaves like a redirector.
1284 +
1285 +2. Caller of VFS Helper for Branches
1286 +aufs_unlink() passes the unlink request to the branch filesystem as if
1287 +it were called from VFS. So the called unlink operation of the branch
1288 +filesystem acts as usual. As a caller of VFS helper, aufs should handle
1289 +every necessary pre/post operation for the branch filesystem.
1290 +- acquire the lock for the parent dir on a branch
1291 +- lookup in a branch
1292 +- revalidate dentry on a branch
1293 +- mnt_want_write() for a branch
1294 +- vfs_unlink() for a branch
1295 +- mnt_drop_write() for a branch
1296 +- release the lock on a branch
1297 +
1298 +3. Persistent Inode Number
1299 +One of the most important issue for a filesystem is to maintain inode
1300 +numbers. This is particularly important to support exporting a
1301 +filesystem via NFS. Aufs is a virtual filesystem which doesn't have a
1302 +backend block device for its own. But some storage is necessary to
1303 +keep and maintain the inode numbers. It may be a large space and may not
1304 +suit to keep in memory. Aufs rents some space from its first writable
1305 +branch filesystem (by default) and creates file(s) on it. These files
1306 +are created by aufs internally and removed soon (currently) keeping
1307 +opened.
1308 +Note: Because these files are removed, they are totally gone after
1309 +      unmounting aufs. It means the inode numbers are not persistent
1310 +      across unmount or reboot. I have a plan to make them really
1311 +      persistent which will be important for aufs on NFS server.
1312 +
1313 +4. Read/Write Files Internally (copy-on-write)
1314 +Because a branch can be readonly, when you write a file on it, aufs will
1315 +"copy-up" it to the upper writable branch internally. And then write the
1316 +originally requested thing to the file. Generally kernel doesn't
1317 +open/read/write file actively. In aufs, even a single write may cause a
1318 +internal "file copy". This behaviour is very similar to cp(1) command.
1319 +
1320 +Some people may think it is better to pass such work to user space
1321 +helper, instead of doing in kernel space. Actually I am still thinking
1322 +about it. But currently I have implemented it in kernel space.
1323 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt linux/Documentation/filesystems/aufs/design/02struct.txt
1324 --- /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt 1970-01-01 01:00:00.000000000 +0100
1325 +++ linux/Documentation/filesystems/aufs/design/02struct.txt    2016-10-09 16:55:36.479367956 +0200
1326 @@ -0,0 +1,258 @@
1327 +
1328 +# Copyright (C) 2005-2016 Junjiro R. Okajima
1329 +# 
1330 +# This program is free software; you can redistribute it and/or modify
1331 +# it under the terms of the GNU General Public License as published by
1332 +# the Free Software Foundation; either version 2 of the License, or
1333 +# (at your option) any later version.
1334 +# 
1335 +# This program is distributed in the hope that it will be useful,
1336 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1337 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1338 +# GNU General Public License for more details.
1339 +# 
1340 +# You should have received a copy of the GNU General Public License
1341 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1342 +
1343 +Basic Aufs Internal Structure
1344 +
1345 +Superblock/Inode/Dentry/File Objects
1346 +----------------------------------------------------------------------
1347 +As like an ordinary filesystem, aufs has its own
1348 +superblock/inode/dentry/file objects. All these objects have a
1349 +dynamically allocated array and store the same kind of pointers to the
1350 +lower filesystem, branch.
1351 +For example, when you build a union with one readwrite branch and one
1352 +readonly, mounted /au, /rw and /ro respectively.
1353 +- /au = /rw + /ro
1354 +- /ro/fileA exists but /rw/fileA
1355 +
1356 +Aufs lookup operation finds /ro/fileA and gets dentry for that. These
1357 +pointers are stored in a aufs dentry. The array in aufs dentry will be,
1358 +- [0] = NULL (because /rw/fileA doesn't exist)
1359 +- [1] = /ro/fileA
1360 +
1361 +This style of an array is essentially same to the aufs
1362 +superblock/inode/dentry/file objects.
1363 +
1364 +Because aufs supports manipulating branches, ie. add/delete/change
1365 +branches dynamically, these objects has its own generation. When
1366 +branches are changed, the generation in aufs superblock is
1367 +incremented. And a generation in other object are compared when it is
1368 +accessed. When a generation in other objects are obsoleted, aufs
1369 +refreshes the internal array.
1370 +
1371 +
1372 +Superblock
1373 +----------------------------------------------------------------------
1374 +Additionally aufs superblock has some data for policies to select one
1375 +among multiple writable branches, XIB files, pseudo-links and kobject.
1376 +See below in detail.
1377 +About the policies which supports copy-down a directory, see
1378 +wbr_policy.txt too.
1379 +
1380 +
1381 +Branch and XINO(External Inode Number Translation Table)
1382 +----------------------------------------------------------------------
1383 +Every branch has its own xino (external inode number translation table)
1384 +file. The xino file is created and unlinked by aufs internally. When two
1385 +members of a union exist on the same filesystem, they share the single
1386 +xino file.
1387 +The struct of a xino file is simple, just a sequence of aufs inode
1388 +numbers which is indexed by the lower inode number.
1389 +In the above sample, assume the inode number of /ro/fileA is i111 and
1390 +aufs assigns the inode number i999 for fileA. Then aufs writes 999 as
1391 +4(8) bytes at 111 * 4(8) bytes offset in the xino file.
1392 +
1393 +When the inode numbers are not contiguous, the xino file will be sparse
1394 +which has a hole in it and doesn't consume as much disk space as it
1395 +might appear. If your branch filesystem consumes disk space for such
1396 +holes, then you should specify 'xino=' option at mounting aufs.
1397 +
1398 +Aufs has a mount option to free the disk blocks for such holes in XINO
1399 +files on tmpfs or ramdisk. But it is not so effective actually. If you
1400 +meet a problem of disk shortage due to XINO files, then you should try
1401 +"tmpfs-ino.patch" (and "vfs-ino.patch" too) in aufs4-standalone.git.
1402 +The patch localizes the assignment inumbers per tmpfs-mount and avoid
1403 +the holes in XINO files.
1404 +
1405 +Also a writable branch has three kinds of "whiteout bases". All these
1406 +are existed when the branch is joined to aufs, and their names are
1407 +whiteout-ed doubly, so that users will never see their names in aufs
1408 +hierarchy.
1409 +1. a regular file which will be hardlinked to all whiteouts.
1410 +2. a directory to store a pseudo-link.
1411 +3. a directory to store an "orphan"-ed file temporary.
1412 +
1413 +1. Whiteout Base
1414 +   When you remove a file on a readonly branch, aufs handles it as a
1415 +   logical deletion and creates a whiteout on the upper writable branch
1416 +   as a hardlink of this file in order not to consume inode on the
1417 +   writable branch.
1418 +2. Pseudo-link Dir
1419 +   See below, Pseudo-link.
1420 +3. Step-Parent Dir
1421 +   When "fileC" exists on the lower readonly branch only and it is
1422 +   opened and removed with its parent dir, and then user writes
1423 +   something into it, then aufs copies-up fileC to this
1424 +   directory. Because there is no other dir to store fileC. After
1425 +   creating a file under this dir, the file is unlinked.
1426 +
1427 +Because aufs supports manipulating branches, ie. add/delete/change
1428 +dynamically, a branch has its own id. When the branch order changes,
1429 +aufs finds the new index by searching the branch id.
1430 +
1431 +
1432 +Pseudo-link
1433 +----------------------------------------------------------------------
1434 +Assume "fileA" exists on the lower readonly branch only and it is
1435 +hardlinked to "fileB" on the branch. When you write something to fileA,
1436 +aufs copies-up it to the upper writable branch. Additionally aufs
1437 +creates a hardlink under the Pseudo-link Directory of the writable
1438 +branch. The inode of a pseudo-link is kept in aufs super_block as a
1439 +simple list. If fileB is read after unlinking fileA, aufs returns
1440 +filedata from the pseudo-link instead of the lower readonly
1441 +branch. Because the pseudo-link is based upon the inode, to keep the
1442 +inode number by xino (see above) is essentially necessary.
1443 +
1444 +All the hardlinks under the Pseudo-link Directory of the writable branch
1445 +should be restored in a proper location later. Aufs provides a utility
1446 +to do this. The userspace helpers executed at remounting and unmounting
1447 +aufs by default.
1448 +During this utility is running, it puts aufs into the pseudo-link
1449 +maintenance mode. In this mode, only the process which began the
1450 +maintenance mode (and its child processes) is allowed to operate in
1451 +aufs. Some other processes which are not related to the pseudo-link will
1452 +be allowed to run too, but the rest have to return an error or wait
1453 +until the maintenance mode ends. If a process already acquires an inode
1454 +mutex (in VFS), it has to return an error.
1455 +
1456 +
1457 +XIB(external inode number bitmap)
1458 +----------------------------------------------------------------------
1459 +Addition to the xino file per a branch, aufs has an external inode number
1460 +bitmap in a superblock object. It is also an internal file such like a
1461 +xino file.
1462 +It is a simple bitmap to mark whether the aufs inode number is in-use or
1463 +not.
1464 +To reduce the file I/O, aufs prepares a single memory page to cache xib.
1465 +
1466 +As well as XINO files, aufs has a feature to truncate/refresh XIB to
1467 +reduce the number of consumed disk blocks for these files.
1468 +
1469 +
1470 +Virtual or Vertical Dir, and Readdir in Userspace
1471 +----------------------------------------------------------------------
1472 +In order to support multiple layers (branches), aufs readdir operation
1473 +constructs a virtual dir block on memory. For readdir, aufs calls
1474 +vfs_readdir() internally for each dir on branches, merges their entries
1475 +with eliminating the whiteout-ed ones, and sets it to file (dir)
1476 +object. So the file object has its entry list until it is closed. The
1477 +entry list will be updated when the file position is zero and becomes
1478 +obsoleted. This decision is made in aufs automatically.
1479 +
1480 +The dynamically allocated memory block for the name of entries has a
1481 +unit of 512 bytes (by default) and stores the names contiguously (no
1482 +padding). Another block for each entry is handled by kmem_cache too.
1483 +During building dir blocks, aufs creates hash list and judging whether
1484 +the entry is whiteouted by its upper branch or already listed.
1485 +The merged result is cached in the corresponding inode object and
1486 +maintained by a customizable life-time option.
1487 +
1488 +Some people may call it can be a security hole or invite DoS attack
1489 +since the opened and once readdir-ed dir (file object) holds its entry
1490 +list and becomes a pressure for system memory. But I'd say it is similar
1491 +to files under /proc or /sys. The virtual files in them also holds a
1492 +memory page (generally) while they are opened. When an idea to reduce
1493 +memory for them is introduced, it will be applied to aufs too.
1494 +For those who really hate this situation, I've developed readdir(3)
1495 +library which operates this merging in userspace. You just need to set
1496 +LD_PRELOAD environment variable, and aufs will not consume no memory in
1497 +kernel space for readdir(3).
1498 +
1499 +
1500 +Workqueue
1501 +----------------------------------------------------------------------
1502 +Aufs sometimes requires privilege access to a branch. For instance,
1503 +in copy-up/down operation. When a user process is going to make changes
1504 +to a file which exists in the lower readonly branch only, and the mode
1505 +of one of ancestor directories may not be writable by a user
1506 +process. Here aufs copy-up the file with its ancestors and they may
1507 +require privilege to set its owner/group/mode/etc.
1508 +This is a typical case of a application character of aufs (see
1509 +Introduction).
1510 +
1511 +Aufs uses workqueue synchronously for this case. It creates its own
1512 +workqueue. The workqueue is a kernel thread and has privilege. Aufs
1513 +passes the request to call mkdir or write (for example), and wait for
1514 +its completion. This approach solves a problem of a signal handler
1515 +simply.
1516 +If aufs didn't adopt the workqueue and changed the privilege of the
1517 +process, then the process may receive the unexpected SIGXFSZ or other
1518 +signals.
1519 +
1520 +Also aufs uses the system global workqueue ("events" kernel thread) too
1521 +for asynchronous tasks, such like handling inotify/fsnotify, re-creating a
1522 +whiteout base and etc. This is unrelated to a privilege.
1523 +Most of aufs operation tries acquiring a rw_semaphore for aufs
1524 +superblock at the beginning, at the same time waits for the completion
1525 +of all queued asynchronous tasks.
1526 +
1527 +
1528 +Whiteout
1529 +----------------------------------------------------------------------
1530 +The whiteout in aufs is very similar to Unionfs's. That is represented
1531 +by its filename. UnionMount takes an approach of a file mode, but I am
1532 +afraid several utilities (find(1) or something) will have to support it.
1533 +
1534 +Basically the whiteout represents "logical deletion" which stops aufs to
1535 +lookup further, but also it represents "dir is opaque" which also stop
1536 +further lookup.
1537 +
1538 +In aufs, rmdir(2) and rename(2) for dir uses whiteout alternatively.
1539 +In order to make several functions in a single systemcall to be
1540 +revertible, aufs adopts an approach to rename a directory to a temporary
1541 +unique whiteouted name.
1542 +For example, in rename(2) dir where the target dir already existed, aufs
1543 +renames the target dir to a temporary unique whiteouted name before the
1544 +actual rename on a branch, and then handles other actions (make it opaque,
1545 +update the attributes, etc). If an error happens in these actions, aufs
1546 +simply renames the whiteouted name back and returns an error. If all are
1547 +succeeded, aufs registers a function to remove the whiteouted unique
1548 +temporary name completely and asynchronously to the system global
1549 +workqueue.
1550 +
1551 +
1552 +Copy-up
1553 +----------------------------------------------------------------------
1554 +It is a well-known feature or concept.
1555 +When user modifies a file on a readonly branch, aufs operate "copy-up"
1556 +internally and makes change to the new file on the upper writable branch.
1557 +When the trigger systemcall does not update the timestamps of the parent
1558 +dir, aufs reverts it after copy-up.
1559 +
1560 +
1561 +Move-down (aufs3.9 and later)
1562 +----------------------------------------------------------------------
1563 +"Copy-up" is one of the essential feature in aufs. It copies a file from
1564 +the lower readonly branch to the upper writable branch when a user
1565 +changes something about the file.
1566 +"Move-down" is an opposite action of copy-up. Basically this action is
1567 +ran manually instead of automatically and internally.
1568 +For desgin and implementation, aufs has to consider these issues.
1569 +- whiteout for the file may exist on the lower branch.
1570 +- ancestor directories may not exist on the lower branch.
1571 +- diropq for the ancestor directories may exist on the upper branch.
1572 +- free space on the lower branch will reduce.
1573 +- another access to the file may happen during moving-down, including
1574 +  UDBA (see "Revalidate Dentry and UDBA").
1575 +- the file should not be hard-linked nor pseudo-linked. they should be
1576 +  handled by auplink utility later.
1577 +
1578 +Sometimes users want to move-down a file from the upper writable branch
1579 +to the lower readonly or writable branch. For instance,
1580 +- the free space of the upper writable branch is going to run out.
1581 +- create a new intermediate branch between the upper and lower branch.
1582 +- etc.
1583 +
1584 +For this purpose, use "aumvdown" command in aufs-util.git.
1585 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03atomic_open.txt linux/Documentation/filesystems/aufs/design/03atomic_open.txt
1586 --- /usr/share/empty/Documentation/filesystems/aufs/design/03atomic_open.txt    1970-01-01 01:00:00.000000000 +0100
1587 +++ linux/Documentation/filesystems/aufs/design/03atomic_open.txt       2016-10-09 16:55:36.479367956 +0200
1588 @@ -0,0 +1,85 @@
1589 +
1590 +# Copyright (C) 2015-2016 Junjiro R. Okajima
1591 +# 
1592 +# This program is free software; you can redistribute it and/or modify
1593 +# it under the terms of the GNU General Public License as published by
1594 +# the Free Software Foundation; either version 2 of the License, or
1595 +# (at your option) any later version.
1596 +# 
1597 +# This program is distributed in the hope that it will be useful,
1598 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1599 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1600 +# GNU General Public License for more details.
1601 +# 
1602 +# You should have received a copy of the GNU General Public License
1603 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1604 +
1605 +Support for a branch who has its ->atomic_open()
1606 +----------------------------------------------------------------------
1607 +The filesystems who implement its ->atomic_open() are not majority. For
1608 +example NFSv4 does, and aufs should call NFSv4 ->atomic_open,
1609 +particularly for open(O_CREAT|O_EXCL, 0400) case. Other than
1610 +->atomic_open(), NFSv4 returns an error for this open(2). While I am not
1611 +sure whether all filesystems who have ->atomic_open() behave like this,
1612 +but NFSv4 surely returns the error.
1613 +
1614 +In order to support ->atomic_open() for aufs, there are a few
1615 +approaches.
1616 +
1617 +A. Introduce aufs_atomic_open()
1618 +   - calls one of VFS:do_last(), lookup_open() or atomic_open() for
1619 +     branch fs.
1620 +B. Introduce aufs_atomic_open() calling create, open and chmod. this is
1621 +   an aufs user Pip Cet's approach
1622 +   - calls aufs_create(), VFS finish_open() and notify_change().
1623 +   - pass fake-mode to finish_open(), and then correct the mode by
1624 +     notify_change().
1625 +C. Extend aufs_open() to call branch fs's ->atomic_open()
1626 +   - no aufs_atomic_open().
1627 +   - aufs_lookup() registers the TID to an aufs internal object.
1628 +   - aufs_create() does nothing when the matching TID is registered, but
1629 +     registers the mode.
1630 +   - aufs_open() calls branch fs's ->atomic_open() when the matching
1631 +     TID is registered.
1632 +D. Extend aufs_open() to re-try branch fs's ->open() with superuser's
1633 +   credential
1634 +   - no aufs_atomic_open().
1635 +   - aufs_create() registers the TID to an internal object. this info
1636 +     represents "this process created this file just now."
1637 +   - when aufs gets EACCES from branch fs's ->open(), then confirm the
1638 +     registered TID and re-try open() with superuser's credential.
1639 +
1640 +Pros and cons for each approach.
1641 +
1642 +A.
1643 +   - straightforward but highly depends upon VFS internal.
1644 +   - the atomic behavaiour is kept.
1645 +   - some of parameters such as nameidata are hard to reproduce for
1646 +     branch fs.
1647 +   - large overhead.
1648 +B.
1649 +   - easy to implement.
1650 +   - the atomic behavaiour is lost.
1651 +C.
1652 +   - the atomic behavaiour is kept.
1653 +   - dirty and tricky.
1654 +   - VFS checks whether the file is created correctly after calling
1655 +     ->create(), which means this approach doesn't work.
1656 +D.
1657 +   - easy to implement.
1658 +   - the atomic behavaiour is lost.
1659 +   - to open a file with superuser's credential and give it to a user
1660 +     process is a bad idea, since the file object keeps the credential
1661 +     in it. It may affect LSM or something. This approach doesn't work
1662 +     either.
1663 +
1664 +The approach A is ideal, but it hard to implement. So here is a
1665 +variation of A, which is to be implemented.
1666 +
1667 +A-1. Introduce aufs_atomic_open()
1668 +     - calls branch fs ->atomic_open() if exists. otherwise calls
1669 +       vfs_create() and finish_open().
1670 +     - the demerit is that the several checks after branch fs
1671 +       ->atomic_open() are lost. in the ordinary case, the checks are
1672 +       done by VFS:do_last(), lookup_open() and atomic_open(). some can
1673 +       be implemented in aufs, but not all I am afraid.
1674 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt linux/Documentation/filesystems/aufs/design/03lookup.txt
1675 --- /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt 1970-01-01 01:00:00.000000000 +0100
1676 +++ linux/Documentation/filesystems/aufs/design/03lookup.txt    2016-10-09 16:55:36.479367956 +0200
1677 @@ -0,0 +1,113 @@
1678 +
1679 +# Copyright (C) 2005-2016 Junjiro R. Okajima
1680 +# 
1681 +# This program is free software; you can redistribute it and/or modify
1682 +# it under the terms of the GNU General Public License as published by
1683 +# the Free Software Foundation; either version 2 of the License, or
1684 +# (at your option) any later version.
1685 +# 
1686 +# This program is distributed in the hope that it will be useful,
1687 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1688 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1689 +# GNU General Public License for more details.
1690 +# 
1691 +# You should have received a copy of the GNU General Public License
1692 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1693 +
1694 +Lookup in a Branch
1695 +----------------------------------------------------------------------
1696 +Since aufs has a character of sub-VFS (see Introduction), it operates
1697 +lookup for branches as VFS does. It may be a heavy work. But almost all
1698 +lookup operation in aufs is the simplest case, ie. lookup only an entry
1699 +directly connected to its parent. Digging down the directory hierarchy
1700 +is unnecessary. VFS has a function lookup_one_len() for that use, and
1701 +aufs calls it.
1702 +
1703 +When a branch is a remote filesystem, aufs basically relies upon its
1704 +->d_revalidate(), also aufs forces the hardest revalidate tests for
1705 +them.
1706 +For d_revalidate, aufs implements three levels of revalidate tests. See
1707 +"Revalidate Dentry and UDBA" in detail.
1708 +
1709 +
1710 +Test Only the Highest One for the Directory Permission (dirperm1 option)
1711 +----------------------------------------------------------------------
1712 +Let's try case study.
1713 +- aufs has two branches, upper readwrite and lower readonly.
1714 +  /au = /rw + /ro
1715 +- "dirA" exists under /ro, but /rw. and its mode is 0700.
1716 +- user invoked "chmod a+rx /au/dirA"
1717 +- the internal copy-up is activated and "/rw/dirA" is created and its
1718 +  permission bits are set to world readable.
1719 +- then "/au/dirA" becomes world readable?
1720 +
1721 +In this case, /ro/dirA is still 0700 since it exists in readonly branch,
1722 +or it may be a natively readonly filesystem. If aufs respects the lower
1723 +branch, it should not respond readdir request from other users. But user
1724 +allowed it by chmod. Should really aufs rejects showing the entries
1725 +under /ro/dirA?
1726 +
1727 +To be honest, I don't have a good solution for this case. So aufs
1728 +implements 'dirperm1' and 'nodirperm1' mount options, and leave it to
1729 +users.
1730 +When dirperm1 is specified, aufs checks only the highest one for the
1731 +directory permission, and shows the entries. Otherwise, as usual, checks
1732 +every dir existing on all branches and rejects the request.
1733 +
1734 +As a side effect, dirperm1 option improves the performance of aufs
1735 +because the number of permission check is reduced when the number of
1736 +branch is many.
1737 +
1738 +
1739 +Revalidate Dentry and UDBA (User's Direct Branch Access)
1740 +----------------------------------------------------------------------
1741 +Generally VFS helpers re-validate a dentry as a part of lookup.
1742 +0. digging down the directory hierarchy.
1743 +1. lock the parent dir by its i_mutex.
1744 +2. lookup the final (child) entry.
1745 +3. revalidate it.
1746 +4. call the actual operation (create, unlink, etc.)
1747 +5. unlock the parent dir
1748 +
1749 +If the filesystem implements its ->d_revalidate() (step 3), then it is
1750 +called. Actually aufs implements it and checks the dentry on a branch is
1751 +still valid.
1752 +But it is not enough. Because aufs has to release the lock for the
1753 +parent dir on a branch at the end of ->lookup() (step 2) and
1754 +->d_revalidate() (step 3) while the i_mutex of the aufs dir is still
1755 +held by VFS.
1756 +If the file on a branch is changed directly, eg. bypassing aufs, after
1757 +aufs released the lock, then the subsequent operation may cause
1758 +something unpleasant result.
1759 +
1760 +This situation is a result of VFS architecture, ->lookup() and
1761 +->d_revalidate() is separated. But I never say it is wrong. It is a good
1762 +design from VFS's point of view. It is just not suitable for sub-VFS
1763 +character in aufs.
1764 +
1765 +Aufs supports such case by three level of revalidation which is
1766 +selectable by user.
1767 +1. Simple Revalidate
1768 +   Addition to the native flow in VFS's, confirm the child-parent
1769 +   relationship on the branch just after locking the parent dir on the
1770 +   branch in the "actual operation" (step 4). When this validation
1771 +   fails, aufs returns EBUSY. ->d_revalidate() (step 3) in aufs still
1772 +   checks the validation of the dentry on branches.
1773 +2. Monitor Changes Internally by Inotify/Fsnotify
1774 +   Addition to above, in the "actual operation" (step 4) aufs re-lookup
1775 +   the dentry on the branch, and returns EBUSY if it finds different
1776 +   dentry.
1777 +   Additionally, aufs sets the inotify/fsnotify watch for every dir on branches
1778 +   during it is in cache. When the event is notified, aufs registers a
1779 +   function to kernel 'events' thread by schedule_work(). And the
1780 +   function sets some special status to the cached aufs dentry and inode
1781 +   private data. If they are not cached, then aufs has nothing to
1782 +   do. When the same file is accessed through aufs (step 0-3) later,
1783 +   aufs will detect the status and refresh all necessary data.
1784 +   In this mode, aufs has to ignore the event which is fired by aufs
1785 +   itself.
1786 +3. No Extra Validation
1787 +   This is the simplest test and doesn't add any additional revalidation
1788 +   test, and skip the revalidation in step 4. It is useful and improves
1789 +   aufs performance when system surely hide the aufs branches from user,
1790 +   by over-mounting something (or another method).
1791 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt linux/Documentation/filesystems/aufs/design/04branch.txt
1792 --- /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt 1970-01-01 01:00:00.000000000 +0100
1793 +++ linux/Documentation/filesystems/aufs/design/04branch.txt    2016-10-09 16:55:36.482701377 +0200
1794 @@ -0,0 +1,74 @@
1795 +
1796 +# Copyright (C) 2005-2016 Junjiro R. Okajima
1797 +# 
1798 +# This program is free software; you can redistribute it and/or modify
1799 +# it under the terms of the GNU General Public License as published by
1800 +# the Free Software Foundation; either version 2 of the License, or
1801 +# (at your option) any later version.
1802 +# 
1803 +# This program is distributed in the hope that it will be useful,
1804 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1805 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1806 +# GNU General Public License for more details.
1807 +# 
1808 +# You should have received a copy of the GNU General Public License
1809 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1810 +
1811 +Branch Manipulation
1812 +
1813 +Since aufs supports dynamic branch manipulation, ie. add/remove a branch
1814 +and changing its permission/attribute, there are a lot of works to do.
1815 +
1816 +
1817 +Add a Branch
1818 +----------------------------------------------------------------------
1819 +o Confirm the adding dir exists outside of aufs, including loopback
1820 +  mount, and its various attributes.
1821 +o Initialize the xino file and whiteout bases if necessary.
1822 +  See struct.txt.
1823 +
1824 +o Check the owner/group/mode of the directory
1825 +  When the owner/group/mode of the adding directory differs from the
1826 +  existing branch, aufs issues a warning because it may impose a
1827 +  security risk.
1828 +  For example, when a upper writable branch has a world writable empty
1829 +  top directory, a malicious user can create any files on the writable
1830 +  branch directly, like copy-up and modify manually. If something like
1831 +  /etc/{passwd,shadow} exists on the lower readonly branch but the upper
1832 +  writable branch, and the writable branch is world-writable, then a
1833 +  malicious guy may create /etc/passwd on the writable branch directly
1834 +  and the infected file will be valid in aufs.
1835 +  I am afraid it can be a security issue, but aufs can do nothing except
1836 +  producing a warning.
1837 +
1838 +
1839 +Delete a Branch
1840 +----------------------------------------------------------------------
1841 +o Confirm the deleting branch is not busy
1842 +  To be general, there is one merit to adopt "remount" interface to
1843 +  manipulate branches. It is to discard caches. At deleting a branch,
1844 +  aufs checks the still cached (and connected) dentries and inodes. If
1845 +  there are any, then they are all in-use. An inode without its
1846 +  corresponding dentry can be alive alone (for example, inotify/fsnotify case).
1847 +
1848 +  For the cached one, aufs checks whether the same named entry exists on
1849 +  other branches.
1850 +  If the cached one is a directory, because aufs provides a merged view
1851 +  to users, as long as one dir is left on any branch aufs can show the
1852 +  dir to users. In this case, the branch can be removed from aufs.
1853 +  Otherwise aufs rejects deleting the branch.
1854 +
1855 +  If any file on the deleting branch is opened by aufs, then aufs
1856 +  rejects deleting.
1857 +
1858 +
1859 +Modify the Permission of a Branch
1860 +----------------------------------------------------------------------
1861 +o Re-initialize or remove the xino file and whiteout bases if necessary.
1862 +  See struct.txt.
1863 +
1864 +o rw --> ro: Confirm the modifying branch is not busy
1865 +  Aufs rejects the request if any of these conditions are true.
1866 +  - a file on the branch is mmap-ed.
1867 +  - a regular file on the branch is opened for write and there is no
1868 +    same named entry on the upper branch.
1869 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt linux/Documentation/filesystems/aufs/design/05wbr_policy.txt
1870 --- /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt     1970-01-01 01:00:00.000000000 +0100
1871 +++ linux/Documentation/filesystems/aufs/design/05wbr_policy.txt        2016-10-09 16:55:36.482701377 +0200
1872 @@ -0,0 +1,64 @@
1873 +
1874 +# Copyright (C) 2005-2016 Junjiro R. Okajima
1875 +# 
1876 +# This program is free software; you can redistribute it and/or modify
1877 +# it under the terms of the GNU General Public License as published by
1878 +# the Free Software Foundation; either version 2 of the License, or
1879 +# (at your option) any later version.
1880 +# 
1881 +# This program is distributed in the hope that it will be useful,
1882 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1883 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1884 +# GNU General Public License for more details.
1885 +# 
1886 +# You should have received a copy of the GNU General Public License
1887 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1888 +
1889 +Policies to Select One among Multiple Writable Branches
1890 +----------------------------------------------------------------------
1891 +When the number of writable branch is more than one, aufs has to decide
1892 +the target branch for file creation or copy-up. By default, the highest
1893 +writable branch which has the parent (or ancestor) dir of the target
1894 +file is chosen (top-down-parent policy).
1895 +By user's request, aufs implements some other policies to select the
1896 +writable branch, for file creation several policies, round-robin,
1897 +most-free-space, and other policies. For copy-up, top-down-parent,
1898 +bottom-up-parent, bottom-up and others.
1899 +
1900 +As expected, the round-robin policy selects the branch in circular. When
1901 +you have two writable branches and creates 10 new files, 5 files will be
1902 +created for each branch. mkdir(2) systemcall is an exception. When you
1903 +create 10 new directories, all will be created on the same branch.
1904 +And the most-free-space policy selects the one which has most free
1905 +space among the writable branches. The amount of free space will be
1906 +checked by aufs internally, and users can specify its time interval.
1907 +
1908 +The policies for copy-up is more simple,
1909 +top-down-parent is equivalent to the same named on in create policy,
1910 +bottom-up-parent selects the writable branch where the parent dir
1911 +exists and the nearest upper one from the copyup-source,
1912 +bottom-up selects the nearest upper writable branch from the
1913 +copyup-source, regardless the existence of the parent dir.
1914 +
1915 +There are some rules or exceptions to apply these policies.
1916 +- If there is a readonly branch above the policy-selected branch and
1917 +  the parent dir is marked as opaque (a variation of whiteout), or the
1918 +  target (creating) file is whiteout-ed on the upper readonly branch,
1919 +  then the result of the policy is ignored and the target file will be
1920 +  created on the nearest upper writable branch than the readonly branch.
1921 +- If there is a writable branch above the policy-selected branch and
1922 +  the parent dir is marked as opaque or the target file is whiteouted
1923 +  on the branch, then the result of the policy is ignored and the target
1924 +  file will be created on the highest one among the upper writable
1925 +  branches who has diropq or whiteout. In case of whiteout, aufs removes
1926 +  it as usual.
1927 +- link(2) and rename(2) systemcalls are exceptions in every policy.
1928 +  They try selecting the branch where the source exists as possible
1929 +  since copyup a large file will take long time. If it can't be,
1930 +  ie. the branch where the source exists is readonly, then they will
1931 +  follow the copyup policy.
1932 +- There is an exception for rename(2) when the target exists.
1933 +  If the rename target exists, aufs compares the index of the branches
1934 +  where the source and the target exists and selects the higher
1935 +  one. If the selected branch is readonly, then aufs follows the
1936 +  copyup policy.
1937 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt linux/Documentation/filesystems/aufs/design/06fhsm.txt
1938 --- /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt   1970-01-01 01:00:00.000000000 +0100
1939 +++ linux/Documentation/filesystems/aufs/design/06fhsm.txt      2016-10-09 16:55:36.482701377 +0200
1940 @@ -0,0 +1,120 @@
1941 +
1942 +# Copyright (C) 2011-2016 Junjiro R. Okajima
1943 +# 
1944 +# This program is free software; you can redistribute it and/or modify
1945 +# it under the terms of the GNU General Public License as published by
1946 +# the Free Software Foundation; either version 2 of the License, or
1947 +# (at your option) any later version.
1948 +# 
1949 +# This program is distributed in the hope that it will be useful,
1950 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1951 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1952 +# GNU General Public License for more details.
1953 +# 
1954 +# You should have received a copy of the GNU General Public License
1955 +# along with this program; if not, write to the Free Software
1956 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
1957 +
1958 +
1959 +File-based Hierarchical Storage Management (FHSM)
1960 +----------------------------------------------------------------------
1961 +Hierarchical Storage Management (or HSM) is a well-known feature in the
1962 +storage world. Aufs provides this feature as file-based with multiple
1963 +writable branches, based upon the principle of "Colder, the Lower".
1964 +Here the word "colder" means that the less used files, and "lower" means
1965 +that the position in the order of the stacked branches vertically.
1966 +These multiple writable branches are prioritized, ie. the topmost one
1967 +should be the fastest drive and be used heavily.
1968 +
1969 +o Characters in aufs FHSM story
1970 +- aufs itself and a new branch attribute.
1971 +- a new ioctl interface to move-down and to establish a connection with
1972 +  the daemon ("move-down" is a converse of "copy-up").
1973 +- userspace tool and daemon.
1974 +
1975 +The userspace daemon establishes a connection with aufs and waits for
1976 +the notification. The notified information is very similar to struct
1977 +statfs containing the number of consumed blocks and inodes.
1978 +When the consumed blocks/inodes of a branch exceeds the user-specified
1979 +upper watermark, the daemon activates its move-down process until the
1980 +consumed blocks/inodes reaches the user-specified lower watermark.
1981 +
1982 +The actual move-down is done by aufs based upon the request from
1983 +user-space since we need to maintain the inode number and the internal
1984 +pointer arrays in aufs.
1985 +
1986 +Currently aufs FHSM handles the regular files only. Additionally they
1987 +must not be hard-linked nor pseudo-linked.
1988 +
1989 +
1990 +o Cowork of aufs and the user-space daemon
1991 +  During the userspace daemon established the connection, aufs sends a
1992 +  small notification to it whenever aufs writes something into the
1993 +  writable branch. But it may cost high since aufs issues statfs(2)
1994 +  internally. So user can specify a new option to cache the
1995 +  info. Actually the notification is controlled by these factors.
1996 +  + the specified cache time.
1997 +  + classified as "force" by aufs internally.
1998 +  Until the specified time expires, aufs doesn't send the info
1999 +  except the forced cases. When aufs decide forcing, the info is always
2000 +  notified to userspace.
2001 +  For example, the number of free inodes is generally large enough and
2002 +  the shortage of it happens rarely. So aufs doesn't force the
2003 +  notification when creating a new file, directory and others. This is
2004 +  the typical case which aufs doesn't force.
2005 +  When aufs writes the actual filedata and the files consumes any of new
2006 +  blocks, the aufs forces notifying.
2007 +
2008 +
2009 +o Interfaces in aufs
2010 +- New branch attribute.
2011 +  + fhsm
2012 +    Specifies that the branch is managed by FHSM feature. In other word,
2013 +    participant in the FHSM.
2014 +    When nofhsm is set to the branch, it will not be the source/target
2015 +    branch of the move-down operation. This attribute is set
2016 +    independently from coo and moo attributes, and if you want full
2017 +    FHSM, you should specify them as well.
2018 +- New mount option.
2019 +  + fhsm_sec
2020 +    Specifies a second to suppress many less important info to be
2021 +    notified.
2022 +- New ioctl.
2023 +  + AUFS_CTL_FHSM_FD
2024 +    create a new file descriptor which userspace can read the notification
2025 +    (a subset of struct statfs) from aufs.
2026 +- Module parameter 'brs'
2027 +  It has to be set to 1. Otherwise the new mount option 'fhsm' will not
2028 +  be set.
2029 +- mount helpers /sbin/mount.aufs and /sbin/umount.aufs
2030 +  When there are two or more branches with fhsm attributes,
2031 +  /sbin/mount.aufs invokes the user-space daemon and /sbin/umount.aufs
2032 +  terminates it. As a result of remounting and branch-manipulation, the
2033 +  number of branches with fhsm attribute can be one. In this case,
2034 +  /sbin/mount.aufs will terminate the user-space daemon.
2035 +
2036 +
2037 +Finally the operation is done as these steps in kernel-space.
2038 +- make sure that,
2039 +  + no one else is using the file.
2040 +  + the file is not hard-linked.
2041 +  + the file is not pseudo-linked.
2042 +  + the file is a regular file.
2043 +  + the parent dir is not opaqued.
2044 +- find the target writable branch.
2045 +- make sure the file is not whiteout-ed by the upper (than the target)
2046 +  branch.
2047 +- make the parent dir on the target branch.
2048 +- mutex lock the inode on the branch.
2049 +- unlink the whiteout on the target branch (if exists).
2050 +- lookup and create the whiteout-ed temporary name on the target branch.
2051 +- copy the file as the whiteout-ed temporary name on the target branch.
2052 +- rename the whiteout-ed temporary name to the original name.
2053 +- unlink the file on the source branch.
2054 +- maintain the internal pointer array and the external inode number
2055 +  table (XINO).
2056 +- maintain the timestamps and other attributes of the parent dir and the
2057 +  file.
2058 +
2059 +And of course, in every step, an error may happen. So the operation
2060 +should restore the original file state after an error happens.
2061 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt linux/Documentation/filesystems/aufs/design/06mmap.txt
2062 --- /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt   1970-01-01 01:00:00.000000000 +0100
2063 +++ linux/Documentation/filesystems/aufs/design/06mmap.txt      2016-10-09 16:55:36.482701377 +0200
2064 @@ -0,0 +1,72 @@
2065 +
2066 +# Copyright (C) 2005-2016 Junjiro R. Okajima
2067 +# 
2068 +# This program is free software; you can redistribute it and/or modify
2069 +# it under the terms of the GNU General Public License as published by
2070 +# the Free Software Foundation; either version 2 of the License, or
2071 +# (at your option) any later version.
2072 +# 
2073 +# This program is distributed in the hope that it will be useful,
2074 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2075 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2076 +# GNU General Public License for more details.
2077 +# 
2078 +# You should have received a copy of the GNU General Public License
2079 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2080 +
2081 +mmap(2) -- File Memory Mapping
2082 +----------------------------------------------------------------------
2083 +In aufs, the file-mapped pages are handled by a branch fs directly, no
2084 +interaction with aufs. It means aufs_mmap() calls the branch fs's
2085 +->mmap().
2086 +This approach is simple and good, but there is one problem.
2087 +Under /proc, several entries show the mmapped files by its path (with
2088 +device and inode number), and the printed path will be the path on the
2089 +branch fs's instead of virtual aufs's.
2090 +This is not a problem in most cases, but some utilities lsof(1) (and its
2091 +user) may expect the path on aufs.
2092 +
2093 +To address this issue, aufs adds a new member called vm_prfile in struct
2094 +vm_area_struct (and struct vm_region). The original vm_file points to
2095 +the file on the branch fs in order to handle everything correctly as
2096 +usual. The new vm_prfile points to a virtual file in aufs, and the
2097 +show-functions in procfs refers to vm_prfile if it is set.
2098 +Also we need to maintain several other places where touching vm_file
2099 +such like
2100 +- fork()/clone() copies vma and the reference count of vm_file is
2101 +  incremented.
2102 +- merging vma maintains the ref count too.
2103 +
2104 +This is not a good approach. It just fakes the printed path. But it
2105 +leaves all behaviour around f_mapping unchanged. This is surely an
2106 +advantage.
2107 +Actually aufs had adopted another complicated approach which calls
2108 +generic_file_mmap() and handles struct vm_operations_struct. In this
2109 +approach, aufs met a hard problem and I could not solve it without
2110 +switching the approach.
2111 +
2112 +There may be one more another approach which is
2113 +- bind-mount the branch-root onto the aufs-root internally
2114 +- grab the new vfsmount (ie. struct mount)
2115 +- lazy-umount the branch-root internally
2116 +- in open(2) the aufs-file, open the branch-file with the hidden
2117 +  vfsmount (instead of the original branch's vfsmount)
2118 +- ideally this "bind-mount and lazy-umount" should be done atomically,
2119 +  but it may be possible from userspace by the mount helper.
2120 +
2121 +Adding the internal hidden vfsmount and using it in opening a file, the
2122 +file path under /proc will be printed correctly. This approach looks
2123 +smarter, but is not possible I am afraid.
2124 +- aufs-root may be bind-mount later. when it happens, another hidden
2125 +  vfsmount will be required.
2126 +- it is hard to get the chance to bind-mount and lazy-umount
2127 +  + in kernel-space, FS can have vfsmount in open(2) via
2128 +    file->f_path, and aufs can know its vfsmount. But several locks are
2129 +    already acquired, and if aufs tries to bind-mount and lazy-umount
2130 +    here, then it may cause a deadlock.
2131 +  + in user-space, bind-mount doesn't invoke the mount helper.
2132 +- since /proc shows dev and ino, aufs has to give vma these info. it
2133 +  means a new member vm_prinode will be necessary. this is essentially
2134 +  equivalent to vm_prfile described above.
2135 +
2136 +I have to give up this "looks-smater" approach.
2137 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt linux/Documentation/filesystems/aufs/design/06xattr.txt
2138 --- /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt  1970-01-01 01:00:00.000000000 +0100
2139 +++ linux/Documentation/filesystems/aufs/design/06xattr.txt     2016-10-09 16:55:36.482701377 +0200
2140 @@ -0,0 +1,96 @@
2141 +
2142 +# Copyright (C) 2014-2016 Junjiro R. Okajima
2143 +#
2144 +# This program is free software; you can redistribute it and/or modify
2145 +# it under the terms of the GNU General Public License as published by
2146 +# the Free Software Foundation; either version 2 of the License, or
2147 +# (at your option) any later version.
2148 +#
2149 +# This program is distributed in the hope that it will be useful,
2150 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2151 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2152 +# GNU General Public License for more details.
2153 +#
2154 +# You should have received a copy of the GNU General Public License
2155 +# along with this program; if not, write to the Free Software
2156 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
2157 +
2158 +
2159 +Listing XATTR/EA and getting the value
2160 +----------------------------------------------------------------------
2161 +For the inode standard attributes (owner, group, timestamps, etc.), aufs
2162 +shows the values from the topmost existing file. This behaviour is good
2163 +for the non-dir entries since the bahaviour exactly matches the shown
2164 +information. But for the directories, aufs considers all the same named
2165 +entries on the lower branches. Which means, if one of the lower entry
2166 +rejects readdir call, then aufs returns an error even if the topmost
2167 +entry allows it. This behaviour is necessary to respect the branch fs's
2168 +security, but can make users confused since the user-visible standard
2169 +attributes don't match the behaviour.
2170 +To address this issue, aufs has a mount option called dirperm1 which
2171 +checks the permission for the topmost entry only, and ignores the lower
2172 +entry's permission.
2173 +
2174 +A similar issue can happen around XATTR.
2175 +getxattr(2) and listxattr(2) families behave as if dirperm1 option is
2176 +always set. Otherwise these very unpleasant situation would happen.
2177 +- listxattr(2) may return the duplicated entries.
2178 +- users may not be able to remove or reset the XATTR forever,
2179 +
2180 +
2181 +XATTR/EA support in the internal (copy,move)-(up,down)
2182 +----------------------------------------------------------------------
2183 +Generally the extended attributes of inode are categorized as these.
2184 +- "security" for LSM and capability.
2185 +- "system" for posix ACL, 'acl' mount option is required for the branch
2186 +  fs generally.
2187 +- "trusted" for userspace, CAP_SYS_ADMIN is required.
2188 +- "user" for userspace, 'user_xattr' mount option is required for the
2189 +  branch fs generally.
2190 +
2191 +Moreover there are some other categories. Aufs handles these rather
2192 +unpopular categories as the ordinary ones, ie. there is no special
2193 +condition nor exception.
2194 +
2195 +In copy-up, the support for XATTR on the dst branch may differ from the
2196 +src branch. In this case, the copy-up operation will get an error and
2197 +the original user operation which triggered the copy-up will fail. It
2198 +can happen that even all copy-up will fail.
2199 +When both of src and dst branches support XATTR and if an error occurs
2200 +during copying XATTR, then the copy-up should fail obviously. That is a
2201 +good reason and aufs should return an error to userspace. But when only
2202 +the src branch support that XATTR, aufs should not return an error.
2203 +For example, the src branch supports ACL but the dst branch doesn't
2204 +because the dst branch may natively un-support it or temporary
2205 +un-support it due to "noacl" mount option. Of course, the dst branch fs
2206 +may NOT return an error even if the XATTR is not supported. It is
2207 +totally up to the branch fs.
2208 +
2209 +Anyway when the aufs internal copy-up gets an error from the dst branch
2210 +fs, then aufs tries removing the just copied entry and returns the error
2211 +to the userspace. The worst case of this situation will be all copy-up
2212 +will fail.
2213 +
2214 +For the copy-up operation, there two basic approaches.
2215 +- copy the specified XATTR only (by category above), and return the
2216 +  error unconditionally if it happens.
2217 +- copy all XATTR, and ignore the error on the specified category only.
2218 +
2219 +In order to support XATTR and to implement the correct behaviour, aufs
2220 +chooses the latter approach and introduces some new branch attributes,
2221 +"icexsec", "icexsys", "icextr", "icexusr", and "icexoth".
2222 +They correspond to the XATTR namespaces (see above). Additionally, to be
2223 +convenient, "icex" is also provided which means all "icex*" attributes
2224 +are set (here the word "icex" stands for "ignore copy-error on XATTR").
2225 +
2226 +The meaning of these attributes is to ignore the error from setting
2227 +XATTR on that branch.
2228 +Note that aufs tries copying all XATTR unconditionally, and ignores the
2229 +error from the dst branch according to the specified attributes.
2230 +
2231 +Some XATTR may have its default value. The default value may come from
2232 +the parent dir or the environment. If the default value is set at the
2233 +file creating-time, it will be overwritten by copy-up.
2234 +Some contradiction may happen I am afraid.
2235 +Do we need another attribute to stop copying XATTR? I am unsure. For
2236 +now, aufs implements the branch attributes to ignore the error.
2237 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt linux/Documentation/filesystems/aufs/design/07export.txt
2238 --- /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt 1970-01-01 01:00:00.000000000 +0100
2239 +++ linux/Documentation/filesystems/aufs/design/07export.txt    2016-10-09 16:55:36.482701377 +0200
2240 @@ -0,0 +1,58 @@
2241 +
2242 +# Copyright (C) 2005-2016 Junjiro R. Okajima
2243 +# 
2244 +# This program is free software; you can redistribute it and/or modify
2245 +# it under the terms of the GNU General Public License as published by
2246 +# the Free Software Foundation; either version 2 of the License, or
2247 +# (at your option) any later version.
2248 +# 
2249 +# This program is distributed in the hope that it will be useful,
2250 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2251 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2252 +# GNU General Public License for more details.
2253 +# 
2254 +# You should have received a copy of the GNU General Public License
2255 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2256 +
2257 +Export Aufs via NFS
2258 +----------------------------------------------------------------------
2259 +Here is an approach.
2260 +- like xino/xib, add a new file 'xigen' which stores aufs inode
2261 +  generation.
2262 +- iget_locked(): initialize aufs inode generation for a new inode, and
2263 +  store it in xigen file.
2264 +- destroy_inode(): increment aufs inode generation and store it in xigen
2265 +  file. it is necessary even if it is not unlinked, because any data of
2266 +  inode may be changed by UDBA.
2267 +- encode_fh(): for a root dir, simply return FILEID_ROOT. otherwise
2268 +  build file handle by
2269 +  + branch id (4 bytes)
2270 +  + superblock generation (4 bytes)
2271 +  + inode number (4 or 8 bytes)
2272 +  + parent dir inode number (4 or 8 bytes)
2273 +  + inode generation (4 bytes))
2274 +  + return value of exportfs_encode_fh() for the parent on a branch (4
2275 +    bytes)
2276 +  + file handle for a branch (by exportfs_encode_fh())
2277 +- fh_to_dentry():
2278 +  + find the index of a branch from its id in handle, and check it is
2279 +    still exist in aufs.
2280 +  + 1st level: get the inode number from handle and search it in cache.
2281 +  + 2nd level: if not found in cache, get the parent inode number from
2282 +    the handle and search it in cache. and then open the found parent
2283 +    dir, find the matching inode number by vfs_readdir() and get its
2284 +    name, and call lookup_one_len() for the target dentry.
2285 +  + 3rd level: if the parent dir is not cached, call
2286 +    exportfs_decode_fh() for a branch and get the parent on a branch,
2287 +    build a pathname of it, convert it a pathname in aufs, call
2288 +    path_lookup(). now aufs gets a parent dir dentry, then handle it as
2289 +    the 2nd level.
2290 +  + to open the dir, aufs needs struct vfsmount. aufs keeps vfsmount
2291 +    for every branch, but not itself. to get this, (currently) aufs
2292 +    searches in current->nsproxy->mnt_ns list. it may not be a good
2293 +    idea, but I didn't get other approach.
2294 +  + test the generation of the gotten inode.
2295 +- every inode operation: they may get EBUSY due to UDBA. in this case,
2296 +  convert it into ESTALE for NFSD.
2297 +- readdir(): call lockdep_on/off() because filldir in NFSD calls
2298 +  lookup_one_len(), vfs_getattr(), encode_fh() and others.
2299 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt linux/Documentation/filesystems/aufs/design/08shwh.txt
2300 --- /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt   1970-01-01 01:00:00.000000000 +0100
2301 +++ linux/Documentation/filesystems/aufs/design/08shwh.txt      2016-10-09 16:55:36.482701377 +0200
2302 @@ -0,0 +1,52 @@
2303 +
2304 +# Copyright (C) 2005-2016 Junjiro R. Okajima
2305 +# 
2306 +# This program is free software; you can redistribute it and/or modify
2307 +# it under the terms of the GNU General Public License as published by
2308 +# the Free Software Foundation; either version 2 of the License, or
2309 +# (at your option) any later version.
2310 +# 
2311 +# This program is distributed in the hope that it will be useful,
2312 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2313 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2314 +# GNU General Public License for more details.
2315 +# 
2316 +# You should have received a copy of the GNU General Public License
2317 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2318 +
2319 +Show Whiteout Mode (shwh)
2320 +----------------------------------------------------------------------
2321 +Generally aufs hides the name of whiteouts. But in some cases, to show
2322 +them is very useful for users. For instance, creating a new middle layer
2323 +(branch) by merging existing layers.
2324 +
2325 +(borrowing aufs1 HOW-TO from a user, Michael Towers)
2326 +When you have three branches,
2327 +- Bottom: 'system', squashfs (underlying base system), read-only
2328 +- Middle: 'mods', squashfs, read-only
2329 +- Top: 'overlay', ram (tmpfs), read-write
2330 +
2331 +The top layer is loaded at boot time and saved at shutdown, to preserve
2332 +the changes made to the system during the session.
2333 +When larger changes have been made, or smaller changes have accumulated,
2334 +the size of the saved top layer data grows. At this point, it would be
2335 +nice to be able to merge the two overlay branches ('mods' and 'overlay')
2336 +and rewrite the 'mods' squashfs, clearing the top layer and thus
2337 +restoring save and load speed.
2338 +
2339 +This merging is simplified by the use of another aufs mount, of just the
2340 +two overlay branches using the 'shwh' option.
2341 +# mount -t aufs -o ro,shwh,br:/livesys/overlay=ro+wh:/livesys/mods=rr+wh \
2342 +       aufs /livesys/merge_union
2343 +
2344 +A merged view of these two branches is then available at
2345 +/livesys/merge_union, and the new feature is that the whiteouts are
2346 +visible!
2347 +Note that in 'shwh' mode the aufs mount must be 'ro', which will disable
2348 +writing to all branches. Also the default mode for all branches is 'ro'.
2349 +It is now possible to save the combined contents of the two overlay
2350 +branches to a new squashfs, e.g.:
2351 +# mksquashfs /livesys/merge_union /path/to/newmods.squash
2352 +
2353 +This new squashfs archive can be stored on the boot device and the
2354 +initramfs will use it to replace the old one at the next boot.
2355 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt linux/Documentation/filesystems/aufs/design/10dynop.txt
2356 --- /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt  1970-01-01 01:00:00.000000000 +0100
2357 +++ linux/Documentation/filesystems/aufs/design/10dynop.txt     2016-10-09 16:55:36.482701377 +0200
2358 @@ -0,0 +1,47 @@
2359 +
2360 +# Copyright (C) 2010-2016 Junjiro R. Okajima
2361 +#
2362 +# This program is free software; you can redistribute it and/or modify
2363 +# it under the terms of the GNU General Public License as published by
2364 +# the Free Software Foundation; either version 2 of the License, or
2365 +# (at your option) any later version.
2366 +#
2367 +# This program is distributed in the hope that it will be useful,
2368 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2369 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2370 +# GNU General Public License for more details.
2371 +#
2372 +# You should have received a copy of the GNU General Public License
2373 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2374 +
2375 +Dynamically customizable FS operations
2376 +----------------------------------------------------------------------
2377 +Generally FS operations (struct inode_operations, struct
2378 +address_space_operations, struct file_operations, etc.) are defined as
2379 +"static const", but it never means that FS have only one set of
2380 +operation. Some FS have multiple sets of them. For instance, ext2 has
2381 +three sets, one for XIP, for NOBH, and for normal.
2382 +Since aufs overrides and redirects these operations, sometimes aufs has
2383 +to change its behaviour according to the branch FS type. More importantly
2384 +VFS acts differently if a function (member in the struct) is set or
2385 +not. It means aufs should have several sets of operations and select one
2386 +among them according to the branch FS definition.
2387 +
2388 +In order to solve this problem and not to affect the behaviour of VFS,
2389 +aufs defines these operations dynamically. For instance, aufs defines
2390 +dummy direct_IO function for struct address_space_operations, but it may
2391 +not be set to the address_space_operations actually. When the branch FS
2392 +doesn't have it, aufs doesn't set it to its address_space_operations
2393 +while the function definition itself is still alive. So the behaviour
2394 +itself will not change, and it will return an error when direct_IO is
2395 +not set.
2396 +
2397 +The lifetime of these dynamically generated operation object is
2398 +maintained by aufs branch object. When the branch is removed from aufs,
2399 +the reference counter of the object is decremented. When it reaches
2400 +zero, the dynamically generated operation object will be freed.
2401 +
2402 +This approach is designed to support AIO (io_submit), Direct I/O and
2403 +XIP (DAX) mainly.
2404 +Currently this approach is applied to address_space_operations for
2405 +regular files only.
2406 diff -urN /usr/share/empty/Documentation/filesystems/aufs/README linux/Documentation/filesystems/aufs/README
2407 --- /usr/share/empty/Documentation/filesystems/aufs/README      1970-01-01 01:00:00.000000000 +0100
2408 +++ linux/Documentation/filesystems/aufs/README 2016-10-09 16:55:36.479367956 +0200
2409 @@ -0,0 +1,392 @@
2410 +
2411 +Aufs4 -- advanced multi layered unification filesystem version 4.x
2412 +http://aufs.sf.net
2413 +Junjiro R. Okajima
2414 +
2415 +
2416 +0. Introduction
2417 +----------------------------------------
2418 +In the early days, aufs was entirely re-designed and re-implemented
2419 +Unionfs Version 1.x series. Adding many original ideas, approaches,
2420 +improvements and implementations, it becomes totally different from
2421 +Unionfs while keeping the basic features.
2422 +Recently, Unionfs Version 2.x series begin taking some of the same
2423 +approaches to aufs1's.
2424 +Unionfs is being developed by Professor Erez Zadok at Stony Brook
2425 +University and his team.
2426 +
2427 +Aufs4 supports linux-4.0 and later, and for linux-3.x series try aufs3.
2428 +If you want older kernel version support, try aufs2-2.6.git or
2429 +aufs2-standalone.git repository, aufs1 from CVS on SourceForge.
2430 +
2431 +Note: it becomes clear that "Aufs was rejected. Let's give it up."
2432 +      According to Christoph Hellwig, linux rejects all union-type
2433 +      filesystems but UnionMount.
2434 +<http://marc.info/?l=linux-kernel&m=123938533724484&w=2>
2435 +
2436 +PS. Al Viro seems have a plan to merge aufs as well as overlayfs and
2437 +    UnionMount, and he pointed out an issue around a directory mutex
2438 +    lock and aufs addressed it. But it is still unsure whether aufs will
2439 +    be merged (or any other union solution).
2440 +<http://marc.info/?l=linux-kernel&m=136312705029295&w=1>
2441 +
2442 +
2443 +1. Features
2444 +----------------------------------------
2445 +- unite several directories into a single virtual filesystem. The member
2446 +  directory is called as a branch.
2447 +- you can specify the permission flags to the branch, which are 'readonly',
2448 +  'readwrite' and 'whiteout-able.'
2449 +- by upper writable branch, internal copyup and whiteout, files/dirs on
2450 +  readonly branch are modifiable logically.
2451 +- dynamic branch manipulation, add, del.
2452 +- etc...
2453 +
2454 +Also there are many enhancements in aufs, such as:
2455 +- test only the highest one for the directory permission (dirperm1)
2456 +- copyup on open (coo=)
2457 +- 'move' policy for copy-up between two writable branches, after
2458 +  checking free space.
2459 +- xattr, acl
2460 +- readdir(3) in userspace.
2461 +- keep inode number by external inode number table
2462 +- keep the timestamps of file/dir in internal copyup operation
2463 +- seekable directory, supporting NFS readdir.
2464 +- whiteout is hardlinked in order to reduce the consumption of inodes
2465 +  on branch
2466 +- do not copyup, nor create a whiteout when it is unnecessary
2467 +- revert a single systemcall when an error occurs in aufs
2468 +- remount interface instead of ioctl
2469 +- maintain /etc/mtab by an external command, /sbin/mount.aufs.
2470 +- loopback mounted filesystem as a branch
2471 +- kernel thread for removing the dir who has a plenty of whiteouts
2472 +- support copyup sparse file (a file which has a 'hole' in it)
2473 +- default permission flags for branches
2474 +- selectable permission flags for ro branch, whether whiteout can
2475 +  exist or not
2476 +- export via NFS.
2477 +- support <sysfs>/fs/aufs and <debugfs>/aufs.
2478 +- support multiple writable branches, some policies to select one
2479 +  among multiple writable branches.
2480 +- a new semantics for link(2) and rename(2) to support multiple
2481 +  writable branches.
2482 +- no glibc changes are required.
2483 +- pseudo hardlink (hardlink over branches)
2484 +- allow a direct access manually to a file on branch, e.g. bypassing aufs.
2485 +  including NFS or remote filesystem branch.
2486 +- userspace wrapper for pathconf(3)/fpathconf(3) with _PC_LINK_MAX.
2487 +- and more...
2488 +
2489 +Currently these features are dropped temporary from aufs4.
2490 +See design/08plan.txt in detail.
2491 +- nested mount, i.e. aufs as readonly no-whiteout branch of another aufs
2492 +  (robr)
2493 +- statistics of aufs thread (/sys/fs/aufs/stat)
2494 +
2495 +Features or just an idea in the future (see also design/*.txt),
2496 +- reorder the branch index without del/re-add.
2497 +- permanent xino files for NFSD
2498 +- an option for refreshing the opened files after add/del branches
2499 +- light version, without branch manipulation. (unnecessary?)
2500 +- copyup in userspace
2501 +- inotify in userspace
2502 +- readv/writev
2503 +
2504 +
2505 +2. Download
2506 +----------------------------------------
2507 +There are three GIT trees for aufs4, aufs4-linux.git,
2508 +aufs4-standalone.git, and aufs-util.git. Note that there is no "4" in
2509 +"aufs-util.git."
2510 +While the aufs-util is always necessary, you need either of aufs4-linux
2511 +or aufs4-standalone.
2512 +
2513 +The aufs4-linux tree includes the whole linux mainline GIT tree,
2514 +git://git.kernel.org/.../torvalds/linux.git.
2515 +And you cannot select CONFIG_AUFS_FS=m for this version, eg. you cannot
2516 +build aufs4 as an external kernel module.
2517 +Several extra patches are not included in this tree. Only
2518 +aufs4-standalone tree contains them. They are described in the later
2519 +section "Configuration and Compilation."
2520 +
2521 +On the other hand, the aufs4-standalone tree has only aufs source files
2522 +and necessary patches, and you can select CONFIG_AUFS_FS=m.
2523 +But you need to apply all aufs patches manually.
2524 +
2525 +You will find GIT branches whose name is in form of "aufs4.x" where "x"
2526 +represents the linux kernel version, "linux-4.x". For instance,
2527 +"aufs4.0" is for linux-4.0. For latest "linux-4.x-rcN", use
2528 +"aufs4.x-rcN" branch.
2529 +
2530 +o aufs4-linux tree
2531 +$ git clone --reference /your/linux/git/tree \
2532 +       git://github.com/sfjro/aufs4-linux.git aufs4-linux.git
2533 +- if you don't have linux GIT tree, then remove "--reference ..."
2534 +$ cd aufs4-linux.git
2535 +$ git checkout origin/aufs4.0
2536 +
2537 +Or You may want to directly git-pull aufs into your linux GIT tree, and
2538 +leave the patch-work to GIT.
2539 +$ cd /your/linux/git/tree
2540 +$ git remote add aufs4 git://github.com/sfjro/aufs4-linux.git
2541 +$ git fetch aufs4
2542 +$ git checkout -b my4.0 v4.0
2543 +$ (add your local change...)
2544 +$ git pull aufs4 aufs4.0
2545 +- now you have v4.0 + your_changes + aufs4.0 in you my4.0 branch.
2546 +- you may need to solve some conflicts between your_changes and
2547 +  aufs4.0. in this case, git-rerere is recommended so that you can
2548 +  solve the similar conflicts automatically when you upgrade to 4.1 or
2549 +  later in the future.
2550 +
2551 +o aufs4-standalone tree
2552 +$ git clone git://github.com/sfjro/aufs4-standalone.git aufs4-standalone.git
2553 +$ cd aufs4-standalone.git
2554 +$ git checkout origin/aufs4.0
2555 +
2556 +o aufs-util tree
2557 +$ git clone git://git.code.sf.net/p/aufs/aufs-util aufs-util.git
2558 +- note that the public aufs-util.git is on SourceForge instead of
2559 +  GitHUB.
2560 +$ cd aufs-util.git
2561 +$ git checkout origin/aufs4.0
2562 +
2563 +Note: The 4.x-rcN branch is to be used with `rc' kernel versions ONLY.
2564 +The minor version number, 'x' in '4.x', of aufs may not always
2565 +follow the minor version number of the kernel.
2566 +Because changes in the kernel that cause the use of a new
2567 +minor version number do not always require changes to aufs-util.
2568 +
2569 +Since aufs-util has its own minor version number, you may not be
2570 +able to find a GIT branch in aufs-util for your kernel's
2571 +exact minor version number.
2572 +In this case, you should git-checkout the branch for the
2573 +nearest lower number.
2574 +
2575 +For (an unreleased) example:
2576 +If you are using "linux-4.10" and the "aufs4.10" branch
2577 +does not exist in aufs-util repository, then "aufs4.9", "aufs4.8"
2578 +or something numerically smaller is the branch for your kernel.
2579 +
2580 +Also you can view all branches by
2581 +       $ git branch -a
2582 +
2583 +
2584 +3. Configuration and Compilation
2585 +----------------------------------------
2586 +Make sure you have git-checkout'ed the correct branch.
2587 +
2588 +For aufs4-linux tree,
2589 +- enable CONFIG_AUFS_FS.
2590 +- set other aufs configurations if necessary.
2591 +
2592 +For aufs4-standalone tree,
2593 +There are several ways to build.
2594 +
2595 +1.
2596 +- apply ./aufs4-kbuild.patch to your kernel source files.
2597 +- apply ./aufs4-base.patch too.
2598 +- apply ./aufs4-mmap.patch too.
2599 +- apply ./aufs4-standalone.patch too, if you have a plan to set
2600 +  CONFIG_AUFS_FS=m. otherwise you don't need ./aufs4-standalone.patch.
2601 +- copy ./{Documentation,fs,include/uapi/linux/aufs_type.h} files to your
2602 +  kernel source tree. Never copy $PWD/include/uapi/linux/Kbuild.
2603 +- enable CONFIG_AUFS_FS, you can select either
2604 +  =m or =y.
2605 +- and build your kernel as usual.
2606 +- install the built kernel.
2607 +  Note: Since linux-3.9, every filesystem module requires an alias
2608 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2609 +  modules.aliases file if you set CONFIG_AUFS_FS=m.
2610 +- install the header files too by "make headers_install" to the
2611 +  directory where you specify. By default, it is $PWD/usr.
2612 +  "make help" shows a brief note for headers_install.
2613 +- and reboot your system.
2614 +
2615 +2.
2616 +- module only (CONFIG_AUFS_FS=m).
2617 +- apply ./aufs4-base.patch to your kernel source files.
2618 +- apply ./aufs4-mmap.patch too.
2619 +- apply ./aufs4-standalone.patch too.
2620 +- build your kernel, don't forget "make headers_install", and reboot.
2621 +- edit ./config.mk and set other aufs configurations if necessary.
2622 +  Note: You should read $PWD/fs/aufs/Kconfig carefully which describes
2623 +  every aufs configurations.
2624 +- build the module by simple "make".
2625 +  Note: Since linux-3.9, every filesystem module requires an alias
2626 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2627 +  modules.aliases file.
2628 +- you can specify ${KDIR} make variable which points to your kernel
2629 +  source tree.
2630 +- install the files
2631 +  + run "make install" to install the aufs module, or copy the built
2632 +    $PWD/aufs.ko to /lib/modules/... and run depmod -a (or reboot simply).
2633 +  + run "make install_headers" (instead of headers_install) to install
2634 +    the modified aufs header file (you can specify DESTDIR which is
2635 +    available in aufs standalone version's Makefile only), or copy
2636 +    $PWD/usr/include/linux/aufs_type.h to /usr/include/linux or wherever
2637 +    you like manually. By default, the target directory is $PWD/usr.
2638 +- no need to apply aufs4-kbuild.patch, nor copying source files to your
2639 +  kernel source tree.
2640 +
2641 +Note: The header file aufs_type.h is necessary to build aufs-util
2642 +      as well as "make headers_install" in the kernel source tree.
2643 +      headers_install is subject to be forgotten, but it is essentially
2644 +      necessary, not only for building aufs-util.
2645 +      You may not meet problems without headers_install in some older
2646 +      version though.
2647 +
2648 +And then,
2649 +- read README in aufs-util, build and install it
2650 +- note that your distribution may contain an obsoleted version of
2651 +  aufs_type.h in /usr/include/linux or something. When you build aufs
2652 +  utilities, make sure that your compiler refers the correct aufs header
2653 +  file which is built by "make headers_install."
2654 +- if you want to use readdir(3) in userspace or pathconf(3) wrapper,
2655 +  then run "make install_ulib" too. And refer to the aufs manual in
2656 +  detail.
2657 +
2658 +There several other patches in aufs4-standalone.git. They are all
2659 +optional. When you meet some problems, they will help you.
2660 +- aufs4-loopback.patch
2661 +  Supports a nested loopback mount in a branch-fs. This patch is
2662 +  unnecessary until aufs produces a message like "you may want to try
2663 +  another patch for loopback file".
2664 +- vfs-ino.patch
2665 +  Modifies a system global kernel internal function get_next_ino() in
2666 +  order to stop assigning 0 for an inode-number. Not directly related to
2667 +  aufs, but recommended generally.
2668 +- tmpfs-idr.patch
2669 +  Keeps the tmpfs inode number as the lowest value. Effective to reduce
2670 +  the size of aufs XINO files for tmpfs branch. Also it prevents the
2671 +  duplication of inode number, which is important for backup tools and
2672 +  other utilities. When you find aufs XINO files for tmpfs branch
2673 +  growing too much, try this patch.
2674 +- lockdep-debug.patch
2675 +  Because aufs is not only an ordinary filesystem (callee of VFS), but
2676 +  also a caller of VFS functions for branch filesystems, subclassing of
2677 +  the internal locks for LOCKDEP is necessary. LOCKDEP is a debugging
2678 +  feature of linux kernel. If you enable CONFIG_LOCKDEP, then you will
2679 +  need to apply this debug patch to expand several constant values.
2680 +  If don't know what LOCKDEP, then you don't have apply this patch.
2681 +
2682 +
2683 +4. Usage
2684 +----------------------------------------
2685 +At first, make sure aufs-util are installed, and please read the aufs
2686 +manual, aufs.5 in aufs-util.git tree.
2687 +$ man -l aufs.5
2688 +
2689 +And then,
2690 +$ mkdir /tmp/rw /tmp/aufs
2691 +# mount -t aufs -o br=/tmp/rw:${HOME} none /tmp/aufs
2692 +
2693 +Here is another example. The result is equivalent.
2694 +# mount -t aufs -o br=/tmp/rw=rw:${HOME}=ro none /tmp/aufs
2695 +  Or
2696 +# mount -t aufs -o br:/tmp/rw none /tmp/aufs
2697 +# mount -o remount,append:${HOME} /tmp/aufs
2698 +
2699 +Then, you can see whole tree of your home dir through /tmp/aufs. If
2700 +you modify a file under /tmp/aufs, the one on your home directory is
2701 +not affected, instead the same named file will be newly created under
2702 +/tmp/rw. And all of your modification to a file will be applied to
2703 +the one under /tmp/rw. This is called the file based Copy on Write
2704 +(COW) method.
2705 +Aufs mount options are described in aufs.5.
2706 +If you run chroot or something and make your aufs as a root directory,
2707 +then you need to customize the shutdown script. See the aufs manual in
2708 +detail.
2709 +
2710 +Additionally, there are some sample usages of aufs which are a
2711 +diskless system with network booting, and LiveCD over NFS.
2712 +See sample dir in CVS tree on SourceForge.
2713 +
2714 +
2715 +5. Contact
2716 +----------------------------------------
2717 +When you have any problems or strange behaviour in aufs, please let me
2718 +know with:
2719 +- /proc/mounts (instead of the output of mount(8))
2720 +- /sys/module/aufs/*
2721 +- /sys/fs/aufs/* (if you have them)
2722 +- /debug/aufs/* (if you have them)
2723 +- linux kernel version
2724 +  if your kernel is not plain, for example modified by distributor,
2725 +  the url where i can download its source is necessary too.
2726 +- aufs version which was printed at loading the module or booting the
2727 +  system, instead of the date you downloaded.
2728 +- configuration (define/undefine CONFIG_AUFS_xxx)
2729 +- kernel configuration or /proc/config.gz (if you have it)
2730 +- behaviour which you think to be incorrect
2731 +- actual operation, reproducible one is better
2732 +- mailto: aufs-users at lists.sourceforge.net
2733 +
2734 +Usually, I don't watch the Public Areas(Bugs, Support Requests, Patches,
2735 +and Feature Requests) on SourceForge. Please join and write to
2736 +aufs-users ML.
2737 +
2738 +
2739 +6. Acknowledgements
2740 +----------------------------------------
2741 +Thanks to everyone who have tried and are using aufs, whoever
2742 +have reported a bug or any feedback.
2743 +
2744 +Especially donators:
2745 +Tomas Matejicek(slax.org) made a donation (much more than once).
2746 +       Since Apr 2010, Tomas M (the author of Slax and Linux Live
2747 +       scripts) is making "doubling" donations.
2748 +       Unfortunately I cannot list all of the donators, but I really
2749 +       appreciate.
2750 +       It ends Aug 2010, but the ordinary donation URL is still available.
2751 +       <http://sourceforge.net/donate/index.php?group_id=167503>
2752 +Dai Itasaka made a donation (2007/8).
2753 +Chuck Smith made a donation (2008/4, 10 and 12).
2754 +Henk Schoneveld made a donation (2008/9).
2755 +Chih-Wei Huang, ASUS, CTC donated Eee PC 4G (2008/10).
2756 +Francois Dupoux made a donation (2008/11).
2757 +Bruno Cesar Ribas and Luis Carlos Erpen de Bona, C3SL serves public
2758 +       aufs2 GIT tree (2009/2).
2759 +William Grant made a donation (2009/3).
2760 +Patrick Lane made a donation (2009/4).
2761 +The Mail Archive (mail-archive.com) made donations (2009/5).
2762 +Nippy Networks (Ed Wildgoose) made a donation (2009/7).
2763 +New Dream Network, LLC (www.dreamhost.com) made a donation (2009/11).
2764 +Pavel Pronskiy made a donation (2011/2).
2765 +Iridium and Inmarsat satellite phone retailer (www.mailasail.com), Nippy
2766 +       Networks (Ed Wildgoose) made a donation for hardware (2011/3).
2767 +Max Lekomcev (DOM-TV project) made a donation (2011/7, 12, 2012/3, 6 and
2768 +11).
2769 +Sam Liddicott made a donation (2011/9).
2770 +Era Scarecrow made a donation (2013/4).
2771 +Bor Ratajc made a donation (2013/4).
2772 +Alessandro Gorreta made a donation (2013/4).
2773 +POIRETTE Marc made a donation (2013/4).
2774 +Alessandro Gorreta made a donation (2013/4).
2775 +lauri kasvandik made a donation (2013/5).
2776 +"pemasu from Finland" made a donation (2013/7).
2777 +The Parted Magic Project made a donation (2013/9 and 11).
2778 +Pavel Barta made a donation (2013/10).
2779 +Nikolay Pertsev made a donation (2014/5).
2780 +James B made a donation (2014/7 and 2015/7).
2781 +Stefano Di Biase made a donation (2014/8).
2782 +Daniel Epellei made a donation (2015/1).
2783 +OmegaPhil made a donation (2016/1).
2784 +Tomasz Szewczyk made a donation (2016/4).
2785 +
2786 +Thank you very much.
2787 +Donations are always, including future donations, very important and
2788 +helpful for me to keep on developing aufs.
2789 +
2790 +
2791 +7.
2792 +----------------------------------------
2793 +If you are an experienced user, no explanation is needed. Aufs is
2794 +just a linux filesystem.
2795 +
2796 +
2797 +Enjoy!
2798 +
2799 +# Local variables: ;
2800 +# mode: text;
2801 +# End: ;
2802 diff -urN /usr/share/empty/fs/aufs/aufs.h linux/fs/aufs/aufs.h
2803 --- /usr/share/empty/fs/aufs/aufs.h     1970-01-01 01:00:00.000000000 +0100
2804 +++ linux/fs/aufs/aufs.h        2016-10-09 16:55:36.486034798 +0200
2805 @@ -0,0 +1,59 @@
2806 +/*
2807 + * Copyright (C) 2005-2016 Junjiro R. Okajima
2808 + *
2809 + * This program, aufs is free software; you can redistribute it and/or modify
2810 + * it under the terms of the GNU General Public License as published by
2811 + * the Free Software Foundation; either version 2 of the License, or
2812 + * (at your option) any later version.
2813 + *
2814 + * This program is distributed in the hope that it will be useful,
2815 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2816 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2817 + * GNU General Public License for more details.
2818 + *
2819 + * You should have received a copy of the GNU General Public License
2820 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2821 + */
2822 +
2823 +/*
2824 + * all header files
2825 + */
2826 +
2827 +#ifndef __AUFS_H__
2828 +#define __AUFS_H__
2829 +
2830 +#ifdef __KERNEL__
2831 +
2832 +#define AuStub(type, name, body, ...) \
2833 +       static inline type name(__VA_ARGS__) { body; }
2834 +
2835 +#define AuStubVoid(name, ...) \
2836 +       AuStub(void, name, , __VA_ARGS__)
2837 +#define AuStubInt0(name, ...) \
2838 +       AuStub(int, name, return 0, __VA_ARGS__)
2839 +
2840 +#include "debug.h"
2841 +
2842 +#include "branch.h"
2843 +#include "cpup.h"
2844 +#include "dcsub.h"
2845 +#include "dbgaufs.h"
2846 +#include "dentry.h"
2847 +#include "dir.h"
2848 +#include "dynop.h"
2849 +#include "file.h"
2850 +#include "fstype.h"
2851 +#include "inode.h"
2852 +#include "loop.h"
2853 +#include "module.h"
2854 +#include "opts.h"
2855 +#include "rwsem.h"
2856 +#include "spl.h"
2857 +#include "super.h"
2858 +#include "sysaufs.h"
2859 +#include "vfsub.h"
2860 +#include "whout.h"
2861 +#include "wkq.h"
2862 +
2863 +#endif /* __KERNEL__ */
2864 +#endif /* __AUFS_H__ */
2865 diff -urN /usr/share/empty/fs/aufs/branch.c linux/fs/aufs/branch.c
2866 --- /usr/share/empty/fs/aufs/branch.c   1970-01-01 01:00:00.000000000 +0100
2867 +++ linux/fs/aufs/branch.c      2016-10-09 16:55:38.886097714 +0200
2868 @@ -0,0 +1,1412 @@
2869 +/*
2870 + * Copyright (C) 2005-2016 Junjiro R. Okajima
2871 + *
2872 + * This program, aufs is free software; you can redistribute it and/or modify
2873 + * it under the terms of the GNU General Public License as published by
2874 + * the Free Software Foundation; either version 2 of the License, or
2875 + * (at your option) any later version.
2876 + *
2877 + * This program is distributed in the hope that it will be useful,
2878 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2879 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2880 + * GNU General Public License for more details.
2881 + *
2882 + * You should have received a copy of the GNU General Public License
2883 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2884 + */
2885 +
2886 +/*
2887 + * branch management
2888 + */
2889 +
2890 +#include <linux/compat.h>
2891 +#include <linux/statfs.h>
2892 +#include "aufs.h"
2893 +
2894 +/*
2895 + * free a single branch
2896 + */
2897 +static void au_br_do_free(struct au_branch *br)
2898 +{
2899 +       int i;
2900 +       struct au_wbr *wbr;
2901 +       struct au_dykey **key;
2902 +
2903 +       au_hnotify_fin_br(br);
2904 +
2905 +       if (br->br_xino.xi_file)
2906 +               fput(br->br_xino.xi_file);
2907 +       mutex_destroy(&br->br_xino.xi_nondir_mtx);
2908 +
2909 +       AuDebugOn(au_br_count(br));
2910 +       au_br_count_fin(br);
2911 +
2912 +       wbr = br->br_wbr;
2913 +       if (wbr) {
2914 +               for (i = 0; i < AuBrWh_Last; i++)
2915 +                       dput(wbr->wbr_wh[i]);
2916 +               AuDebugOn(atomic_read(&wbr->wbr_wh_running));
2917 +               AuRwDestroy(&wbr->wbr_wh_rwsem);
2918 +       }
2919 +
2920 +       if (br->br_fhsm) {
2921 +               au_br_fhsm_fin(br->br_fhsm);
2922 +               au_delayed_kfree(br->br_fhsm);
2923 +       }
2924 +
2925 +       key = br->br_dykey;
2926 +       for (i = 0; i < AuBrDynOp; i++, key++)
2927 +               if (*key)
2928 +                       au_dy_put(*key);
2929 +               else
2930 +                       break;
2931 +
2932 +       /* recursive lock, s_umount of branch's */
2933 +       lockdep_off();
2934 +       path_put(&br->br_path);
2935 +       lockdep_on();
2936 +       if (wbr)
2937 +               au_delayed_kfree(wbr);
2938 +       au_delayed_kfree(br);
2939 +}
2940 +
2941 +/*
2942 + * frees all branches
2943 + */
2944 +void au_br_free(struct au_sbinfo *sbinfo)
2945 +{
2946 +       aufs_bindex_t bmax;
2947 +       struct au_branch **br;
2948 +
2949 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
2950 +
2951 +       bmax = sbinfo->si_bbot + 1;
2952 +       br = sbinfo->si_branch;
2953 +       while (bmax--)
2954 +               au_br_do_free(*br++);
2955 +}
2956 +
2957 +/*
2958 + * find the index of a branch which is specified by @br_id.
2959 + */
2960 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id)
2961 +{
2962 +       aufs_bindex_t bindex, bbot;
2963 +
2964 +       bbot = au_sbbot(sb);
2965 +       for (bindex = 0; bindex <= bbot; bindex++)
2966 +               if (au_sbr_id(sb, bindex) == br_id)
2967 +                       return bindex;
2968 +       return -1;
2969 +}
2970 +
2971 +/* ---------------------------------------------------------------------- */
2972 +
2973 +/*
2974 + * add a branch
2975 + */
2976 +
2977 +static int test_overlap(struct super_block *sb, struct dentry *h_adding,
2978 +                       struct dentry *h_root)
2979 +{
2980 +       if (unlikely(h_adding == h_root
2981 +                    || au_test_loopback_overlap(sb, h_adding)))
2982 +               return 1;
2983 +       if (h_adding->d_sb != h_root->d_sb)
2984 +               return 0;
2985 +       return au_test_subdir(h_adding, h_root)
2986 +               || au_test_subdir(h_root, h_adding);
2987 +}
2988 +
2989 +/*
2990 + * returns a newly allocated branch. @new_nbranch is a number of branches
2991 + * after adding a branch.
2992 + */
2993 +static struct au_branch *au_br_alloc(struct super_block *sb, int new_nbranch,
2994 +                                    int perm)
2995 +{
2996 +       struct au_branch *add_branch;
2997 +       struct dentry *root;
2998 +       struct inode *inode;
2999 +       int err;
3000 +
3001 +       err = -ENOMEM;
3002 +       root = sb->s_root;
3003 +       add_branch = kzalloc(sizeof(*add_branch), GFP_NOFS);
3004 +       if (unlikely(!add_branch))
3005 +               goto out;
3006 +
3007 +       err = au_hnotify_init_br(add_branch, perm);
3008 +       if (unlikely(err))
3009 +               goto out_br;
3010 +
3011 +       if (au_br_writable(perm)) {
3012 +               /* may be freed separately at changing the branch permission */
3013 +               add_branch->br_wbr = kzalloc(sizeof(*add_branch->br_wbr),
3014 +                                            GFP_NOFS);
3015 +               if (unlikely(!add_branch->br_wbr))
3016 +                       goto out_hnotify;
3017 +       }
3018 +
3019 +       if (au_br_fhsm(perm)) {
3020 +               err = au_fhsm_br_alloc(add_branch);
3021 +               if (unlikely(err))
3022 +                       goto out_wbr;
3023 +       }
3024 +
3025 +       err = au_sbr_realloc(au_sbi(sb), new_nbranch, /*may_shrink*/0);
3026 +       if (!err)
3027 +               err = au_di_realloc(au_di(root), new_nbranch, /*may_shrink*/0);
3028 +       if (!err) {
3029 +               inode = d_inode(root);
3030 +               err = au_hinode_realloc(au_ii(inode), new_nbranch, /*may_shrink*/0);
3031 +       }
3032 +       if (!err)
3033 +               return add_branch; /* success */
3034 +
3035 +out_wbr:
3036 +       if (add_branch->br_wbr)
3037 +               au_delayed_kfree(add_branch->br_wbr);
3038 +out_hnotify:
3039 +       au_hnotify_fin_br(add_branch);
3040 +out_br:
3041 +       au_delayed_kfree(add_branch);
3042 +out:
3043 +       return ERR_PTR(err);
3044 +}
3045 +
3046 +/*
3047 + * test if the branch permission is legal or not.
3048 + */
3049 +static int test_br(struct inode *inode, int brperm, char *path)
3050 +{
3051 +       int err;
3052 +
3053 +       err = (au_br_writable(brperm) && IS_RDONLY(inode));
3054 +       if (!err)
3055 +               goto out;
3056 +
3057 +       err = -EINVAL;
3058 +       pr_err("write permission for readonly mount or inode, %s\n", path);
3059 +
3060 +out:
3061 +       return err;
3062 +}
3063 +
3064 +/*
3065 + * returns:
3066 + * 0: success, the caller will add it
3067 + * plus: success, it is already unified, the caller should ignore it
3068 + * minus: error
3069 + */
3070 +static int test_add(struct super_block *sb, struct au_opt_add *add, int remount)
3071 +{
3072 +       int err;
3073 +       aufs_bindex_t bbot, bindex;
3074 +       struct dentry *root, *h_dentry;
3075 +       struct inode *inode, *h_inode;
3076 +
3077 +       root = sb->s_root;
3078 +       bbot = au_sbbot(sb);
3079 +       if (unlikely(bbot >= 0
3080 +                    && au_find_dbindex(root, add->path.dentry) >= 0)) {
3081 +               err = 1;
3082 +               if (!remount) {
3083 +                       err = -EINVAL;
3084 +                       pr_err("%s duplicated\n", add->pathname);
3085 +               }
3086 +               goto out;
3087 +       }
3088 +
3089 +       err = -ENOSPC; /* -E2BIG; */
3090 +       if (unlikely(AUFS_BRANCH_MAX <= add->bindex
3091 +                    || AUFS_BRANCH_MAX - 1 <= bbot)) {
3092 +               pr_err("number of branches exceeded %s\n", add->pathname);
3093 +               goto out;
3094 +       }
3095 +
3096 +       err = -EDOM;
3097 +       if (unlikely(add->bindex < 0 || bbot + 1 < add->bindex)) {
3098 +               pr_err("bad index %d\n", add->bindex);
3099 +               goto out;
3100 +       }
3101 +
3102 +       inode = d_inode(add->path.dentry);
3103 +       err = -ENOENT;
3104 +       if (unlikely(!inode->i_nlink)) {
3105 +               pr_err("no existence %s\n", add->pathname);
3106 +               goto out;
3107 +       }
3108 +
3109 +       err = -EINVAL;
3110 +       if (unlikely(inode->i_sb == sb)) {
3111 +               pr_err("%s must be outside\n", add->pathname);
3112 +               goto out;
3113 +       }
3114 +
3115 +       if (unlikely(au_test_fs_unsuppoted(inode->i_sb))) {
3116 +               pr_err("unsupported filesystem, %s (%s)\n",
3117 +                      add->pathname, au_sbtype(inode->i_sb));
3118 +               goto out;
3119 +       }
3120 +
3121 +       if (unlikely(inode->i_sb->s_stack_depth)) {
3122 +               pr_err("already stacked, %s (%s)\n",
3123 +                      add->pathname, au_sbtype(inode->i_sb));
3124 +               goto out;
3125 +       }
3126 +
3127 +       err = test_br(d_inode(add->path.dentry), add->perm, add->pathname);
3128 +       if (unlikely(err))
3129 +               goto out;
3130 +
3131 +       if (bbot < 0)
3132 +               return 0; /* success */
3133 +
3134 +       err = -EINVAL;
3135 +       for (bindex = 0; bindex <= bbot; bindex++)
3136 +               if (unlikely(test_overlap(sb, add->path.dentry,
3137 +                                         au_h_dptr(root, bindex)))) {
3138 +                       pr_err("%s is overlapped\n", add->pathname);
3139 +                       goto out;
3140 +               }
3141 +
3142 +       err = 0;
3143 +       if (au_opt_test(au_mntflags(sb), WARN_PERM)) {
3144 +               h_dentry = au_h_dptr(root, 0);
3145 +               h_inode = d_inode(h_dentry);
3146 +               if ((h_inode->i_mode & S_IALLUGO) != (inode->i_mode & S_IALLUGO)
3147 +                   || !uid_eq(h_inode->i_uid, inode->i_uid)
3148 +                   || !gid_eq(h_inode->i_gid, inode->i_gid))
3149 +                       pr_warn("uid/gid/perm %s %u/%u/0%o, %u/%u/0%o\n",
3150 +                               add->pathname,
3151 +                               i_uid_read(inode), i_gid_read(inode),
3152 +                               (inode->i_mode & S_IALLUGO),
3153 +                               i_uid_read(h_inode), i_gid_read(h_inode),
3154 +                               (h_inode->i_mode & S_IALLUGO));
3155 +       }
3156 +
3157 +out:
3158 +       return err;
3159 +}
3160 +
3161 +/*
3162 + * initialize or clean the whiteouts for an adding branch
3163 + */
3164 +static int au_br_init_wh(struct super_block *sb, struct au_branch *br,
3165 +                        int new_perm)
3166 +{
3167 +       int err, old_perm;
3168 +       aufs_bindex_t bindex;
3169 +       struct inode *h_inode;
3170 +       struct au_wbr *wbr;
3171 +       struct au_hinode *hdir;
3172 +       struct dentry *h_dentry;
3173 +
3174 +       err = vfsub_mnt_want_write(au_br_mnt(br));
3175 +       if (unlikely(err))
3176 +               goto out;
3177 +
3178 +       wbr = br->br_wbr;
3179 +       old_perm = br->br_perm;
3180 +       br->br_perm = new_perm;
3181 +       hdir = NULL;
3182 +       h_inode = NULL;
3183 +       bindex = au_br_index(sb, br->br_id);
3184 +       if (0 <= bindex) {
3185 +               hdir = au_hi(d_inode(sb->s_root), bindex);
3186 +               au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
3187 +       } else {
3188 +               h_dentry = au_br_dentry(br);
3189 +               h_inode = d_inode(h_dentry);
3190 +               inode_lock_nested(h_inode, AuLsc_I_PARENT);
3191 +       }
3192 +       if (!wbr)
3193 +               err = au_wh_init(br, sb);
3194 +       else {
3195 +               wbr_wh_write_lock(wbr);
3196 +               err = au_wh_init(br, sb);
3197 +               wbr_wh_write_unlock(wbr);
3198 +       }
3199 +       if (hdir)
3200 +               au_hn_inode_unlock(hdir);
3201 +       else
3202 +               inode_unlock(h_inode);
3203 +       vfsub_mnt_drop_write(au_br_mnt(br));
3204 +       br->br_perm = old_perm;
3205 +
3206 +       if (!err && wbr && !au_br_writable(new_perm)) {
3207 +               au_delayed_kfree(wbr);
3208 +               br->br_wbr = NULL;
3209 +       }
3210 +
3211 +out:
3212 +       return err;
3213 +}
3214 +
3215 +static int au_wbr_init(struct au_branch *br, struct super_block *sb,
3216 +                      int perm)
3217 +{
3218 +       int err;
3219 +       struct kstatfs kst;
3220 +       struct au_wbr *wbr;
3221 +
3222 +       wbr = br->br_wbr;
3223 +       au_rw_init(&wbr->wbr_wh_rwsem);
3224 +       atomic_set(&wbr->wbr_wh_running, 0);
3225 +
3226 +       /*
3227 +        * a limit for rmdir/rename a dir
3228 +        * cf. AUFS_MAX_NAMELEN in include/uapi/linux/aufs_type.h
3229 +        */
3230 +       err = vfs_statfs(&br->br_path, &kst);
3231 +       if (unlikely(err))
3232 +               goto out;
3233 +       err = -EINVAL;
3234 +       if (kst.f_namelen >= NAME_MAX)
3235 +               err = au_br_init_wh(sb, br, perm);
3236 +       else
3237 +               pr_err("%pd(%s), unsupported namelen %ld\n",
3238 +                      au_br_dentry(br),
3239 +                      au_sbtype(au_br_dentry(br)->d_sb), kst.f_namelen);
3240 +
3241 +out:
3242 +       return err;
3243 +}
3244 +
3245 +/* initialize a new branch */
3246 +static int au_br_init(struct au_branch *br, struct super_block *sb,
3247 +                     struct au_opt_add *add)
3248 +{
3249 +       int err;
3250 +       struct inode *h_inode;
3251 +
3252 +       err = 0;
3253 +       mutex_init(&br->br_xino.xi_nondir_mtx);
3254 +       br->br_perm = add->perm;
3255 +       br->br_path = add->path; /* set first, path_get() later */
3256 +       spin_lock_init(&br->br_dykey_lock);
3257 +       au_br_count_init(br);
3258 +       atomic_set(&br->br_xino_running, 0);
3259 +       br->br_id = au_new_br_id(sb);
3260 +       AuDebugOn(br->br_id < 0);
3261 +
3262 +       if (au_br_writable(add->perm)) {
3263 +               err = au_wbr_init(br, sb, add->perm);
3264 +               if (unlikely(err))
3265 +                       goto out_err;
3266 +       }
3267 +
3268 +       if (au_opt_test(au_mntflags(sb), XINO)) {
3269 +               h_inode = d_inode(add->path.dentry);
3270 +               err = au_xino_br(sb, br, h_inode->i_ino,
3271 +                                au_sbr(sb, 0)->br_xino.xi_file, /*do_test*/1);
3272 +               if (unlikely(err)) {
3273 +                       AuDebugOn(br->br_xino.xi_file);
3274 +                       goto out_err;
3275 +               }
3276 +       }
3277 +
3278 +       sysaufs_br_init(br);
3279 +       path_get(&br->br_path);
3280 +       goto out; /* success */
3281 +
3282 +out_err:
3283 +       memset(&br->br_path, 0, sizeof(br->br_path));
3284 +out:
3285 +       return err;
3286 +}
3287 +
3288 +static void au_br_do_add_brp(struct au_sbinfo *sbinfo, aufs_bindex_t bindex,
3289 +                            struct au_branch *br, aufs_bindex_t bbot,
3290 +                            aufs_bindex_t amount)
3291 +{
3292 +       struct au_branch **brp;
3293 +
3294 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3295 +
3296 +       brp = sbinfo->si_branch + bindex;
3297 +       memmove(brp + 1, brp, sizeof(*brp) * amount);
3298 +       *brp = br;
3299 +       sbinfo->si_bbot++;
3300 +       if (unlikely(bbot < 0))
3301 +               sbinfo->si_bbot = 0;
3302 +}
3303 +
3304 +static void au_br_do_add_hdp(struct au_dinfo *dinfo, aufs_bindex_t bindex,
3305 +                            aufs_bindex_t bbot, aufs_bindex_t amount)
3306 +{
3307 +       struct au_hdentry *hdp;
3308 +
3309 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3310 +
3311 +       hdp = au_hdentry(dinfo, bindex);
3312 +       memmove(hdp + 1, hdp, sizeof(*hdp) * amount);
3313 +       au_h_dentry_init(hdp);
3314 +       dinfo->di_bbot++;
3315 +       if (unlikely(bbot < 0))
3316 +               dinfo->di_btop = 0;
3317 +}
3318 +
3319 +static void au_br_do_add_hip(struct au_iinfo *iinfo, aufs_bindex_t bindex,
3320 +                            aufs_bindex_t bbot, aufs_bindex_t amount)
3321 +{
3322 +       struct au_hinode *hip;
3323 +
3324 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3325 +
3326 +       hip = au_hinode(iinfo, bindex);
3327 +       memmove(hip + 1, hip, sizeof(*hip) * amount);
3328 +       au_hinode_init(hip);
3329 +       iinfo->ii_bbot++;
3330 +       if (unlikely(bbot < 0))
3331 +               iinfo->ii_btop = 0;
3332 +}
3333 +
3334 +static void au_br_do_add(struct super_block *sb, struct au_branch *br,
3335 +                        aufs_bindex_t bindex)
3336 +{
3337 +       struct dentry *root, *h_dentry;
3338 +       struct inode *root_inode, *h_inode;
3339 +       aufs_bindex_t bbot, amount;
3340 +
3341 +       root = sb->s_root;
3342 +       root_inode = d_inode(root);
3343 +       bbot = au_sbbot(sb);
3344 +       amount = bbot + 1 - bindex;
3345 +       h_dentry = au_br_dentry(br);
3346 +       au_sbilist_lock();
3347 +       au_br_do_add_brp(au_sbi(sb), bindex, br, bbot, amount);
3348 +       au_br_do_add_hdp(au_di(root), bindex, bbot, amount);
3349 +       au_br_do_add_hip(au_ii(root_inode), bindex, bbot, amount);
3350 +       au_set_h_dptr(root, bindex, dget(h_dentry));
3351 +       h_inode = d_inode(h_dentry);
3352 +       au_set_h_iptr(root_inode, bindex, au_igrab(h_inode), /*flags*/0);
3353 +       au_sbilist_unlock();
3354 +}
3355 +
3356 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount)
3357 +{
3358 +       int err;
3359 +       aufs_bindex_t bbot, add_bindex;
3360 +       struct dentry *root, *h_dentry;
3361 +       struct inode *root_inode;
3362 +       struct au_branch *add_branch;
3363 +
3364 +       root = sb->s_root;
3365 +       root_inode = d_inode(root);
3366 +       IMustLock(root_inode);
3367 +       IiMustWriteLock(root_inode);
3368 +       err = test_add(sb, add, remount);
3369 +       if (unlikely(err < 0))
3370 +               goto out;
3371 +       if (err) {
3372 +               err = 0;
3373 +               goto out; /* success */
3374 +       }
3375 +
3376 +       bbot = au_sbbot(sb);
3377 +       add_branch = au_br_alloc(sb, bbot + 2, add->perm);
3378 +       err = PTR_ERR(add_branch);
3379 +       if (IS_ERR(add_branch))
3380 +               goto out;
3381 +
3382 +       err = au_br_init(add_branch, sb, add);
3383 +       if (unlikely(err)) {
3384 +               au_br_do_free(add_branch);
3385 +               goto out;
3386 +       }
3387 +
3388 +       add_bindex = add->bindex;
3389 +       if (!remount)
3390 +               au_br_do_add(sb, add_branch, add_bindex);
3391 +       else {
3392 +               sysaufs_brs_del(sb, add_bindex);
3393 +               au_br_do_add(sb, add_branch, add_bindex);
3394 +               sysaufs_brs_add(sb, add_bindex);
3395 +       }
3396 +
3397 +       h_dentry = add->path.dentry;
3398 +       if (!add_bindex) {
3399 +               au_cpup_attr_all(root_inode, /*force*/1);
3400 +               sb->s_maxbytes = h_dentry->d_sb->s_maxbytes;
3401 +       } else
3402 +               au_add_nlink(root_inode, d_inode(h_dentry));
3403 +
3404 +       /*
3405 +        * this test/set prevents aufs from handling unnecesary notify events
3406 +        * of xino files, in case of re-adding a writable branch which was
3407 +        * once detached from aufs.
3408 +        */
3409 +       if (au_xino_brid(sb) < 0
3410 +           && au_br_writable(add_branch->br_perm)
3411 +           && !au_test_fs_bad_xino(h_dentry->d_sb)
3412 +           && add_branch->br_xino.xi_file
3413 +           && add_branch->br_xino.xi_file->f_path.dentry->d_parent == h_dentry)
3414 +               au_xino_brid_set(sb, add_branch->br_id);
3415 +
3416 +out:
3417 +       return err;
3418 +}
3419 +
3420 +/* ---------------------------------------------------------------------- */
3421 +
3422 +static unsigned long long au_farray_cb(struct super_block *sb, void *a,
3423 +                                      unsigned long long max __maybe_unused,
3424 +                                      void *arg)
3425 +{
3426 +       unsigned long long n;
3427 +       struct file **p, *f;
3428 +       struct au_sphlhead *files;
3429 +       struct au_finfo *finfo;
3430 +
3431 +       n = 0;
3432 +       p = a;
3433 +       files = &au_sbi(sb)->si_files;
3434 +       spin_lock(&files->spin);
3435 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
3436 +               f = finfo->fi_file;
3437 +               if (file_count(f)
3438 +                   && !special_file(file_inode(f)->i_mode)) {
3439 +                       get_file(f);
3440 +                       *p++ = f;
3441 +                       n++;
3442 +                       AuDebugOn(n > max);
3443 +               }
3444 +       }
3445 +       spin_unlock(&files->spin);
3446 +
3447 +       return n;
3448 +}
3449 +
3450 +static struct file **au_farray_alloc(struct super_block *sb,
3451 +                                    unsigned long long *max)
3452 +{
3453 +       *max = au_nfiles(sb);
3454 +       return au_array_alloc(max, au_farray_cb, sb, /*arg*/NULL);
3455 +}
3456 +
3457 +static void au_farray_free(struct file **a, unsigned long long max)
3458 +{
3459 +       unsigned long long ull;
3460 +
3461 +       for (ull = 0; ull < max; ull++)
3462 +               if (a[ull])
3463 +                       fput(a[ull]);
3464 +       kvfree(a);
3465 +}
3466 +
3467 +/* ---------------------------------------------------------------------- */
3468 +
3469 +/*
3470 + * delete a branch
3471 + */
3472 +
3473 +/* to show the line number, do not make it inlined function */
3474 +#define AuVerbose(do_info, fmt, ...) do { \
3475 +       if (do_info) \
3476 +               pr_info(fmt, ##__VA_ARGS__); \
3477 +} while (0)
3478 +
3479 +static int au_test_ibusy(struct inode *inode, aufs_bindex_t btop,
3480 +                        aufs_bindex_t bbot)
3481 +{
3482 +       return (inode && !S_ISDIR(inode->i_mode)) || btop == bbot;
3483 +}
3484 +
3485 +static int au_test_dbusy(struct dentry *dentry, aufs_bindex_t btop,
3486 +                        aufs_bindex_t bbot)
3487 +{
3488 +       return au_test_ibusy(d_inode(dentry), btop, bbot);
3489 +}
3490 +
3491 +/*
3492 + * test if the branch is deletable or not.
3493 + */
3494 +static int test_dentry_busy(struct dentry *root, aufs_bindex_t bindex,
3495 +                           unsigned int sigen, const unsigned int verbose)
3496 +{
3497 +       int err, i, j, ndentry;
3498 +       aufs_bindex_t btop, bbot;
3499 +       struct au_dcsub_pages dpages;
3500 +       struct au_dpage *dpage;
3501 +       struct dentry *d;
3502 +
3503 +       err = au_dpages_init(&dpages, GFP_NOFS);
3504 +       if (unlikely(err))
3505 +               goto out;
3506 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
3507 +       if (unlikely(err))
3508 +               goto out_dpages;
3509 +
3510 +       for (i = 0; !err && i < dpages.ndpage; i++) {
3511 +               dpage = dpages.dpages + i;
3512 +               ndentry = dpage->ndentry;
3513 +               for (j = 0; !err && j < ndentry; j++) {
3514 +                       d = dpage->dentries[j];
3515 +                       AuDebugOn(au_dcount(d) <= 0);
3516 +                       if (!au_digen_test(d, sigen)) {
3517 +                               di_read_lock_child(d, AuLock_IR);
3518 +                               if (unlikely(au_dbrange_test(d))) {
3519 +                                       di_read_unlock(d, AuLock_IR);
3520 +                                       continue;
3521 +                               }
3522 +                       } else {
3523 +                               di_write_lock_child(d);
3524 +                               if (unlikely(au_dbrange_test(d))) {
3525 +                                       di_write_unlock(d);
3526 +                                       continue;
3527 +                               }
3528 +                               err = au_reval_dpath(d, sigen);
3529 +                               if (!err)
3530 +                                       di_downgrade_lock(d, AuLock_IR);
3531 +                               else {
3532 +                                       di_write_unlock(d);
3533 +                                       break;
3534 +                               }
3535 +                       }
3536 +
3537 +                       /* AuDbgDentry(d); */
3538 +                       btop = au_dbtop(d);
3539 +                       bbot = au_dbbot(d);
3540 +                       if (btop <= bindex
3541 +                           && bindex <= bbot
3542 +                           && au_h_dptr(d, bindex)
3543 +                           && au_test_dbusy(d, btop, bbot)) {
3544 +                               err = -EBUSY;
3545 +                               AuVerbose(verbose, "busy %pd\n", d);
3546 +                               AuDbgDentry(d);
3547 +                       }
3548 +                       di_read_unlock(d, AuLock_IR);
3549 +               }
3550 +       }
3551 +
3552 +out_dpages:
3553 +       au_dpages_free(&dpages);
3554 +out:
3555 +       return err;
3556 +}
3557 +
3558 +static int test_inode_busy(struct super_block *sb, aufs_bindex_t bindex,
3559 +                          unsigned int sigen, const unsigned int verbose)
3560 +{
3561 +       int err;
3562 +       unsigned long long max, ull;
3563 +       struct inode *i, **array;
3564 +       aufs_bindex_t btop, bbot;
3565 +
3566 +       array = au_iarray_alloc(sb, &max);
3567 +       err = PTR_ERR(array);
3568 +       if (IS_ERR(array))
3569 +               goto out;
3570 +
3571 +       err = 0;
3572 +       AuDbg("b%d\n", bindex);
3573 +       for (ull = 0; !err && ull < max; ull++) {
3574 +               i = array[ull];
3575 +               if (unlikely(!i))
3576 +                       break;
3577 +               if (i->i_ino == AUFS_ROOT_INO)
3578 +                       continue;
3579 +
3580 +               /* AuDbgInode(i); */
3581 +               if (au_iigen(i, NULL) == sigen)
3582 +                       ii_read_lock_child(i);
3583 +               else {
3584 +                       ii_write_lock_child(i);
3585 +                       err = au_refresh_hinode_self(i);
3586 +                       au_iigen_dec(i);
3587 +                       if (!err)
3588 +                               ii_downgrade_lock(i);
3589 +                       else {
3590 +                               ii_write_unlock(i);
3591 +                               break;
3592 +                       }
3593 +               }
3594 +
3595 +               btop = au_ibtop(i);
3596 +               bbot = au_ibbot(i);
3597 +               if (btop <= bindex
3598 +                   && bindex <= bbot
3599 +                   && au_h_iptr(i, bindex)
3600 +                   && au_test_ibusy(i, btop, bbot)) {
3601 +                       err = -EBUSY;
3602 +                       AuVerbose(verbose, "busy i%lu\n", i->i_ino);
3603 +                       AuDbgInode(i);
3604 +               }
3605 +               ii_read_unlock(i);
3606 +       }
3607 +       au_iarray_free(array, max);
3608 +
3609 +out:
3610 +       return err;
3611 +}
3612 +
3613 +static int test_children_busy(struct dentry *root, aufs_bindex_t bindex,
3614 +                             const unsigned int verbose)
3615 +{
3616 +       int err;
3617 +       unsigned int sigen;
3618 +
3619 +       sigen = au_sigen(root->d_sb);
3620 +       DiMustNoWaiters(root);
3621 +       IiMustNoWaiters(d_inode(root));
3622 +       di_write_unlock(root);
3623 +       err = test_dentry_busy(root, bindex, sigen, verbose);
3624 +       if (!err)
3625 +               err = test_inode_busy(root->d_sb, bindex, sigen, verbose);
3626 +       di_write_lock_child(root); /* aufs_write_lock() calls ..._child() */
3627 +
3628 +       return err;
3629 +}
3630 +
3631 +static int test_dir_busy(struct file *file, aufs_bindex_t br_id,
3632 +                        struct file **to_free, int *idx)
3633 +{
3634 +       int err;
3635 +       unsigned char matched, root;
3636 +       aufs_bindex_t bindex, bbot;
3637 +       struct au_fidir *fidir;
3638 +       struct au_hfile *hfile;
3639 +
3640 +       err = 0;
3641 +       root = IS_ROOT(file->f_path.dentry);
3642 +       if (root) {
3643 +               get_file(file);
3644 +               to_free[*idx] = file;
3645 +               (*idx)++;
3646 +               goto out;
3647 +       }
3648 +
3649 +       matched = 0;
3650 +       fidir = au_fi(file)->fi_hdir;
3651 +       AuDebugOn(!fidir);
3652 +       bbot = au_fbbot_dir(file);
3653 +       for (bindex = au_fbtop(file); bindex <= bbot; bindex++) {
3654 +               hfile = fidir->fd_hfile + bindex;
3655 +               if (!hfile->hf_file)
3656 +                       continue;
3657 +
3658 +               if (hfile->hf_br->br_id == br_id) {
3659 +                       matched = 1;
3660 +                       break;
3661 +               }
3662 +       }
3663 +       if (matched)
3664 +               err = -EBUSY;
3665 +
3666 +out:
3667 +       return err;
3668 +}
3669 +
3670 +static int test_file_busy(struct super_block *sb, aufs_bindex_t br_id,
3671 +                         struct file **to_free, int opened)
3672 +{
3673 +       int err, idx;
3674 +       unsigned long long ull, max;
3675 +       aufs_bindex_t btop;
3676 +       struct file *file, **array;
3677 +       struct dentry *root;
3678 +       struct au_hfile *hfile;
3679 +
3680 +       array = au_farray_alloc(sb, &max);
3681 +       err = PTR_ERR(array);
3682 +       if (IS_ERR(array))
3683 +               goto out;
3684 +
3685 +       err = 0;
3686 +       idx = 0;
3687 +       root = sb->s_root;
3688 +       di_write_unlock(root);
3689 +       for (ull = 0; ull < max; ull++) {
3690 +               file = array[ull];
3691 +               if (unlikely(!file))
3692 +                       break;
3693 +
3694 +               /* AuDbg("%pD\n", file); */
3695 +               fi_read_lock(file);
3696 +               btop = au_fbtop(file);
3697 +               if (!d_is_dir(file->f_path.dentry)) {
3698 +                       hfile = &au_fi(file)->fi_htop;
3699 +                       if (hfile->hf_br->br_id == br_id)
3700 +                               err = -EBUSY;
3701 +               } else
3702 +                       err = test_dir_busy(file, br_id, to_free, &idx);
3703 +               fi_read_unlock(file);
3704 +               if (unlikely(err))
3705 +                       break;
3706 +       }
3707 +       di_write_lock_child(root);
3708 +       au_farray_free(array, max);
3709 +       AuDebugOn(idx > opened);
3710 +
3711 +out:
3712 +       return err;
3713 +}
3714 +
3715 +static void br_del_file(struct file **to_free, unsigned long long opened,
3716 +                         aufs_bindex_t br_id)
3717 +{
3718 +       unsigned long long ull;
3719 +       aufs_bindex_t bindex, btop, bbot, bfound;
3720 +       struct file *file;
3721 +       struct au_fidir *fidir;
3722 +       struct au_hfile *hfile;
3723 +
3724 +       for (ull = 0; ull < opened; ull++) {
3725 +               file = to_free[ull];
3726 +               if (unlikely(!file))
3727 +                       break;
3728 +
3729 +               /* AuDbg("%pD\n", file); */
3730 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
3731 +               bfound = -1;
3732 +               fidir = au_fi(file)->fi_hdir;
3733 +               AuDebugOn(!fidir);
3734 +               fi_write_lock(file);
3735 +               btop = au_fbtop(file);
3736 +               bbot = au_fbbot_dir(file);
3737 +               for (bindex = btop; bindex <= bbot; bindex++) {
3738 +                       hfile = fidir->fd_hfile + bindex;
3739 +                       if (!hfile->hf_file)
3740 +                               continue;
3741 +
3742 +                       if (hfile->hf_br->br_id == br_id) {
3743 +                               bfound = bindex;
3744 +                               break;
3745 +                       }
3746 +               }
3747 +               AuDebugOn(bfound < 0);
3748 +               au_set_h_fptr(file, bfound, NULL);
3749 +               if (bfound == btop) {
3750 +                       for (btop++; btop <= bbot; btop++)
3751 +                               if (au_hf_dir(file, btop)) {
3752 +                                       au_set_fbtop(file, btop);
3753 +                                       break;
3754 +                               }
3755 +               }
3756 +               fi_write_unlock(file);
3757 +       }
3758 +}
3759 +
3760 +static void au_br_do_del_brp(struct au_sbinfo *sbinfo,
3761 +                            const aufs_bindex_t bindex,
3762 +                            const aufs_bindex_t bbot)
3763 +{
3764 +       struct au_branch **brp, **p;
3765 +
3766 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3767 +
3768 +       brp = sbinfo->si_branch + bindex;
3769 +       if (bindex < bbot)
3770 +               memmove(brp, brp + 1, sizeof(*brp) * (bbot - bindex));
3771 +       sbinfo->si_branch[0 + bbot] = NULL;
3772 +       sbinfo->si_bbot--;
3773 +
3774 +       p = au_krealloc(sbinfo->si_branch, sizeof(*p) * bbot, AuGFP_SBILIST,
3775 +                       /*may_shrink*/1);
3776 +       if (p)
3777 +               sbinfo->si_branch = p;
3778 +       /* harmless error */
3779 +}
3780 +
3781 +static void au_br_do_del_hdp(struct au_dinfo *dinfo, const aufs_bindex_t bindex,
3782 +                            const aufs_bindex_t bbot)
3783 +{
3784 +       struct au_hdentry *hdp, *p;
3785 +
3786 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3787 +
3788 +       hdp = au_hdentry(dinfo, bindex);
3789 +       if (bindex < bbot)
3790 +               memmove(hdp, hdp + 1, sizeof(*hdp) * (bbot - bindex));
3791 +       /* au_h_dentry_init(au_hdentry(dinfo, bbot); */
3792 +       dinfo->di_bbot--;
3793 +
3794 +       p = au_krealloc(dinfo->di_hdentry, sizeof(*p) * bbot, AuGFP_SBILIST,
3795 +                       /*may_shrink*/1);
3796 +       if (p)
3797 +               dinfo->di_hdentry = p;
3798 +       /* harmless error */
3799 +}
3800 +
3801 +static void au_br_do_del_hip(struct au_iinfo *iinfo, const aufs_bindex_t bindex,
3802 +                            const aufs_bindex_t bbot)
3803 +{
3804 +       struct au_hinode *hip, *p;
3805 +
3806 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3807 +
3808 +       hip = au_hinode(iinfo, bindex);
3809 +       if (bindex < bbot)
3810 +               memmove(hip, hip + 1, sizeof(*hip) * (bbot - bindex));
3811 +       /* au_hinode_init(au_hinode(iinfo, bbot)); */
3812 +       iinfo->ii_bbot--;
3813 +
3814 +       p = au_krealloc(iinfo->ii_hinode, sizeof(*p) * bbot, AuGFP_SBILIST,
3815 +                       /*may_shrink*/1);
3816 +       if (p)
3817 +               iinfo->ii_hinode = p;
3818 +       /* harmless error */
3819 +}
3820 +
3821 +static void au_br_do_del(struct super_block *sb, aufs_bindex_t bindex,
3822 +                        struct au_branch *br)
3823 +{
3824 +       aufs_bindex_t bbot;
3825 +       struct au_sbinfo *sbinfo;
3826 +       struct dentry *root, *h_root;
3827 +       struct inode *inode, *h_inode;
3828 +       struct au_hinode *hinode;
3829 +
3830 +       SiMustWriteLock(sb);
3831 +
3832 +       root = sb->s_root;
3833 +       inode = d_inode(root);
3834 +       sbinfo = au_sbi(sb);
3835 +       bbot = sbinfo->si_bbot;
3836 +
3837 +       h_root = au_h_dptr(root, bindex);
3838 +       hinode = au_hi(inode, bindex);
3839 +       h_inode = au_igrab(hinode->hi_inode);
3840 +       au_hiput(hinode);
3841 +
3842 +       au_sbilist_lock();
3843 +       au_br_do_del_brp(sbinfo, bindex, bbot);
3844 +       au_br_do_del_hdp(au_di(root), bindex, bbot);
3845 +       au_br_do_del_hip(au_ii(inode), bindex, bbot);
3846 +       au_sbilist_unlock();
3847 +
3848 +       dput(h_root);
3849 +       iput(h_inode);
3850 +       au_br_do_free(br);
3851 +}
3852 +
3853 +static unsigned long long empty_cb(struct super_block *sb, void *array,
3854 +                                  unsigned long long max, void *arg)
3855 +{
3856 +       return max;
3857 +}
3858 +
3859 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount)
3860 +{
3861 +       int err, rerr, i;
3862 +       unsigned long long opened;
3863 +       unsigned int mnt_flags;
3864 +       aufs_bindex_t bindex, bbot, br_id;
3865 +       unsigned char do_wh, verbose;
3866 +       struct au_branch *br;
3867 +       struct au_wbr *wbr;
3868 +       struct dentry *root;
3869 +       struct file **to_free;
3870 +
3871 +       err = 0;
3872 +       opened = 0;
3873 +       to_free = NULL;
3874 +       root = sb->s_root;
3875 +       bindex = au_find_dbindex(root, del->h_path.dentry);
3876 +       if (bindex < 0) {
3877 +               if (remount)
3878 +                       goto out; /* success */
3879 +               err = -ENOENT;
3880 +               pr_err("%s no such branch\n", del->pathname);
3881 +               goto out;
3882 +       }
3883 +       AuDbg("bindex b%d\n", bindex);
3884 +
3885 +       err = -EBUSY;
3886 +       mnt_flags = au_mntflags(sb);
3887 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3888 +       bbot = au_sbbot(sb);
3889 +       if (unlikely(!bbot)) {
3890 +               AuVerbose(verbose, "no more branches left\n");
3891 +               goto out;
3892 +       }
3893 +       br = au_sbr(sb, bindex);
3894 +       AuDebugOn(!path_equal(&br->br_path, &del->h_path));
3895 +
3896 +       br_id = br->br_id;
3897 +       opened = au_br_count(br);
3898 +       if (unlikely(opened)) {
3899 +               to_free = au_array_alloc(&opened, empty_cb, sb, NULL);
3900 +               err = PTR_ERR(to_free);
3901 +               if (IS_ERR(to_free))
3902 +                       goto out;
3903 +
3904 +               err = test_file_busy(sb, br_id, to_free, opened);
3905 +               if (unlikely(err)) {
3906 +                       AuVerbose(verbose, "%llu file(s) opened\n", opened);
3907 +                       goto out;
3908 +               }
3909 +       }
3910 +
3911 +       wbr = br->br_wbr;
3912 +       do_wh = wbr && (wbr->wbr_whbase || wbr->wbr_plink || wbr->wbr_orph);
3913 +       if (do_wh) {
3914 +               /* instead of WbrWhMustWriteLock(wbr) */
3915 +               SiMustWriteLock(sb);
3916 +               for (i = 0; i < AuBrWh_Last; i++) {
3917 +                       dput(wbr->wbr_wh[i]);
3918 +                       wbr->wbr_wh[i] = NULL;
3919 +               }
3920 +       }
3921 +
3922 +       err = test_children_busy(root, bindex, verbose);
3923 +       if (unlikely(err)) {
3924 +               if (do_wh)
3925 +                       goto out_wh;
3926 +               goto out;
3927 +       }
3928 +
3929 +       err = 0;
3930 +       if (to_free) {
3931 +               /*
3932 +                * now we confirmed the branch is deletable.
3933 +                * let's free the remaining opened dirs on the branch.
3934 +                */
3935 +               di_write_unlock(root);
3936 +               br_del_file(to_free, opened, br_id);
3937 +               di_write_lock_child(root);
3938 +       }
3939 +
3940 +       if (!remount)
3941 +               au_br_do_del(sb, bindex, br);
3942 +       else {
3943 +               sysaufs_brs_del(sb, bindex);
3944 +               au_br_do_del(sb, bindex, br);
3945 +               sysaufs_brs_add(sb, bindex);
3946 +       }
3947 +
3948 +       if (!bindex) {
3949 +               au_cpup_attr_all(d_inode(root), /*force*/1);
3950 +               sb->s_maxbytes = au_sbr_sb(sb, 0)->s_maxbytes;
3951 +       } else
3952 +               au_sub_nlink(d_inode(root), d_inode(del->h_path.dentry));
3953 +       if (au_opt_test(mnt_flags, PLINK))
3954 +               au_plink_half_refresh(sb, br_id);
3955 +
3956 +       if (au_xino_brid(sb) == br_id)
3957 +               au_xino_brid_set(sb, -1);
3958 +       goto out; /* success */
3959 +
3960 +out_wh:
3961 +       /* revert */
3962 +       rerr = au_br_init_wh(sb, br, br->br_perm);
3963 +       if (rerr)
3964 +               pr_warn("failed re-creating base whiteout, %s. (%d)\n",
3965 +                       del->pathname, rerr);
3966 +out:
3967 +       if (to_free)
3968 +               au_farray_free(to_free, opened);
3969 +       return err;
3970 +}
3971 +
3972 +/* ---------------------------------------------------------------------- */
3973 +
3974 +static int au_ibusy(struct super_block *sb, struct aufs_ibusy __user *arg)
3975 +{
3976 +       int err;
3977 +       aufs_bindex_t btop, bbot;
3978 +       struct aufs_ibusy ibusy;
3979 +       struct inode *inode, *h_inode;
3980 +
3981 +       err = -EPERM;
3982 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
3983 +               goto out;
3984 +
3985 +       err = copy_from_user(&ibusy, arg, sizeof(ibusy));
3986 +       if (!err)
3987 +               err = !access_ok(VERIFY_WRITE, &arg->h_ino, sizeof(arg->h_ino));
3988 +       if (unlikely(err)) {
3989 +               err = -EFAULT;
3990 +               AuTraceErr(err);
3991 +               goto out;
3992 +       }
3993 +
3994 +       err = -EINVAL;
3995 +       si_read_lock(sb, AuLock_FLUSH);
3996 +       if (unlikely(ibusy.bindex < 0 || ibusy.bindex > au_sbbot(sb)))
3997 +               goto out_unlock;
3998 +
3999 +       err = 0;
4000 +       ibusy.h_ino = 0; /* invalid */
4001 +       inode = ilookup(sb, ibusy.ino);
4002 +       if (!inode
4003 +           || inode->i_ino == AUFS_ROOT_INO
4004 +           || au_is_bad_inode(inode))
4005 +               goto out_unlock;
4006 +
4007 +       ii_read_lock_child(inode);
4008 +       btop = au_ibtop(inode);
4009 +       bbot = au_ibbot(inode);
4010 +       if (btop <= ibusy.bindex && ibusy.bindex <= bbot) {
4011 +               h_inode = au_h_iptr(inode, ibusy.bindex);
4012 +               if (h_inode && au_test_ibusy(inode, btop, bbot))
4013 +                       ibusy.h_ino = h_inode->i_ino;
4014 +       }
4015 +       ii_read_unlock(inode);
4016 +       iput(inode);
4017 +
4018 +out_unlock:
4019 +       si_read_unlock(sb);
4020 +       if (!err) {
4021 +               err = __put_user(ibusy.h_ino, &arg->h_ino);
4022 +               if (unlikely(err)) {
4023 +                       err = -EFAULT;
4024 +                       AuTraceErr(err);
4025 +               }
4026 +       }
4027 +out:
4028 +       return err;
4029 +}
4030 +
4031 +long au_ibusy_ioctl(struct file *file, unsigned long arg)
4032 +{
4033 +       return au_ibusy(file->f_path.dentry->d_sb, (void __user *)arg);
4034 +}
4035 +
4036 +#ifdef CONFIG_COMPAT
4037 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg)
4038 +{
4039 +       return au_ibusy(file->f_path.dentry->d_sb, compat_ptr(arg));
4040 +}
4041 +#endif
4042 +
4043 +/* ---------------------------------------------------------------------- */
4044 +
4045 +/*
4046 + * change a branch permission
4047 + */
4048 +
4049 +static void au_warn_ima(void)
4050 +{
4051 +#ifdef CONFIG_IMA
4052 +       /* since it doesn't support mark_files_ro() */
4053 +       AuWarn1("RW -> RO makes IMA to produce wrong message\n");
4054 +#endif
4055 +}
4056 +
4057 +static int do_need_sigen_inc(int a, int b)
4058 +{
4059 +       return au_br_whable(a) && !au_br_whable(b);
4060 +}
4061 +
4062 +static int need_sigen_inc(int old, int new)
4063 +{
4064 +       return do_need_sigen_inc(old, new)
4065 +               || do_need_sigen_inc(new, old);
4066 +}
4067 +
4068 +static int au_br_mod_files_ro(struct super_block *sb, aufs_bindex_t bindex)
4069 +{
4070 +       int err, do_warn;
4071 +       unsigned int mnt_flags;
4072 +       unsigned long long ull, max;
4073 +       aufs_bindex_t br_id;
4074 +       unsigned char verbose, writer;
4075 +       struct file *file, *hf, **array;
4076 +       struct au_hfile *hfile;
4077 +
4078 +       mnt_flags = au_mntflags(sb);
4079 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
4080 +
4081 +       array = au_farray_alloc(sb, &max);
4082 +       err = PTR_ERR(array);
4083 +       if (IS_ERR(array))
4084 +               goto out;
4085 +
4086 +       do_warn = 0;
4087 +       br_id = au_sbr_id(sb, bindex);
4088 +       for (ull = 0; ull < max; ull++) {
4089 +               file = array[ull];
4090 +               if (unlikely(!file))
4091 +                       break;
4092 +
4093 +               /* AuDbg("%pD\n", file); */
4094 +               fi_read_lock(file);
4095 +               if (unlikely(au_test_mmapped(file))) {
4096 +                       err = -EBUSY;
4097 +                       AuVerbose(verbose, "mmapped %pD\n", file);
4098 +                       AuDbgFile(file);
4099 +                       FiMustNoWaiters(file);
4100 +                       fi_read_unlock(file);
4101 +                       goto out_array;
4102 +               }
4103 +
4104 +               hfile = &au_fi(file)->fi_htop;
4105 +               hf = hfile->hf_file;
4106 +               if (!d_is_reg(file->f_path.dentry)
4107 +                   || !(file->f_mode & FMODE_WRITE)
4108 +                   || hfile->hf_br->br_id != br_id
4109 +                   || !(hf->f_mode & FMODE_WRITE))
4110 +                       array[ull] = NULL;
4111 +               else {
4112 +                       do_warn = 1;
4113 +                       get_file(file);
4114 +               }
4115 +
4116 +               FiMustNoWaiters(file);
4117 +               fi_read_unlock(file);
4118 +               fput(file);
4119 +       }
4120 +
4121 +       err = 0;
4122 +       if (do_warn)
4123 +               au_warn_ima();
4124 +
4125 +       for (ull = 0; ull < max; ull++) {
4126 +               file = array[ull];
4127 +               if (!file)
4128 +                       continue;
4129 +
4130 +               /* todo: already flushed? */
4131 +               /*
4132 +                * fs/super.c:mark_files_ro() is gone, but aufs keeps its
4133 +                * approach which resets f_mode and calls mnt_drop_write() and
4134 +                * file_release_write() for each file, because the branch
4135 +                * attribute in aufs world is totally different from the native
4136 +                * fs rw/ro mode.
4137 +               */
4138 +               /* fi_read_lock(file); */
4139 +               hfile = &au_fi(file)->fi_htop;
4140 +               hf = hfile->hf_file;
4141 +               /* fi_read_unlock(file); */
4142 +               spin_lock(&hf->f_lock);
4143 +               writer = !!(hf->f_mode & FMODE_WRITER);
4144 +               hf->f_mode &= ~(FMODE_WRITE | FMODE_WRITER);
4145 +               spin_unlock(&hf->f_lock);
4146 +               if (writer) {
4147 +                       put_write_access(file_inode(hf));
4148 +                       __mnt_drop_write(hf->f_path.mnt);
4149 +               }
4150 +       }
4151 +
4152 +out_array:
4153 +       au_farray_free(array, max);
4154 +out:
4155 +       AuTraceErr(err);
4156 +       return err;
4157 +}
4158 +
4159 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4160 +             int *do_refresh)
4161 +{
4162 +       int err, rerr;
4163 +       aufs_bindex_t bindex;
4164 +       struct dentry *root;
4165 +       struct au_branch *br;
4166 +       struct au_br_fhsm *bf;
4167 +
4168 +       root = sb->s_root;
4169 +       bindex = au_find_dbindex(root, mod->h_root);
4170 +       if (bindex < 0) {
4171 +               if (remount)
4172 +                       return 0; /* success */
4173 +               err = -ENOENT;
4174 +               pr_err("%s no such branch\n", mod->path);
4175 +               goto out;
4176 +       }
4177 +       AuDbg("bindex b%d\n", bindex);
4178 +
4179 +       err = test_br(d_inode(mod->h_root), mod->perm, mod->path);
4180 +       if (unlikely(err))
4181 +               goto out;
4182 +
4183 +       br = au_sbr(sb, bindex);
4184 +       AuDebugOn(mod->h_root != au_br_dentry(br));
4185 +       if (br->br_perm == mod->perm)
4186 +               return 0; /* success */
4187 +
4188 +       /* pre-allocate for non-fhsm --> fhsm */
4189 +       bf = NULL;
4190 +       if (!au_br_fhsm(br->br_perm) && au_br_fhsm(mod->perm)) {
4191 +               err = au_fhsm_br_alloc(br);
4192 +               if (unlikely(err))
4193 +                       goto out;
4194 +               bf = br->br_fhsm;
4195 +               br->br_fhsm = NULL;
4196 +       }
4197 +
4198 +       if (au_br_writable(br->br_perm)) {
4199 +               /* remove whiteout base */
4200 +               err = au_br_init_wh(sb, br, mod->perm);
4201 +               if (unlikely(err))
4202 +                       goto out_bf;
4203 +
4204 +               if (!au_br_writable(mod->perm)) {
4205 +                       /* rw --> ro, file might be mmapped */
4206 +                       DiMustNoWaiters(root);
4207 +                       IiMustNoWaiters(d_inode(root));
4208 +                       di_write_unlock(root);
4209 +                       err = au_br_mod_files_ro(sb, bindex);
4210 +                       /* aufs_write_lock() calls ..._child() */
4211 +                       di_write_lock_child(root);
4212 +
4213 +                       if (unlikely(err)) {
4214 +                               rerr = -ENOMEM;
4215 +                               br->br_wbr = kzalloc(sizeof(*br->br_wbr),
4216 +                                                    GFP_NOFS);
4217 +                               if (br->br_wbr)
4218 +                                       rerr = au_wbr_init(br, sb, br->br_perm);
4219 +                               if (unlikely(rerr)) {
4220 +                                       AuIOErr("nested error %d (%d)\n",
4221 +                                               rerr, err);
4222 +                                       br->br_perm = mod->perm;
4223 +                               }
4224 +                       }
4225 +               }
4226 +       } else if (au_br_writable(mod->perm)) {
4227 +               /* ro --> rw */
4228 +               err = -ENOMEM;
4229 +               br->br_wbr = kzalloc(sizeof(*br->br_wbr), GFP_NOFS);
4230 +               if (br->br_wbr) {
4231 +                       err = au_wbr_init(br, sb, mod->perm);
4232 +                       if (unlikely(err)) {
4233 +                               au_delayed_kfree(br->br_wbr);
4234 +                               br->br_wbr = NULL;
4235 +                       }
4236 +               }
4237 +       }
4238 +       if (unlikely(err))
4239 +               goto out_bf;
4240 +
4241 +       if (au_br_fhsm(br->br_perm)) {
4242 +               if (!au_br_fhsm(mod->perm)) {
4243 +                       /* fhsm --> non-fhsm */
4244 +                       au_br_fhsm_fin(br->br_fhsm);
4245 +                       au_delayed_kfree(br->br_fhsm);
4246 +                       br->br_fhsm = NULL;
4247 +               }
4248 +       } else if (au_br_fhsm(mod->perm))
4249 +               /* non-fhsm --> fhsm */
4250 +               br->br_fhsm = bf;
4251 +
4252 +       *do_refresh |= need_sigen_inc(br->br_perm, mod->perm);
4253 +       br->br_perm = mod->perm;
4254 +       goto out; /* success */
4255 +
4256 +out_bf:
4257 +       if (bf)
4258 +               au_delayed_kfree(bf);
4259 +out:
4260 +       AuTraceErr(err);
4261 +       return err;
4262 +}
4263 +
4264 +/* ---------------------------------------------------------------------- */
4265 +
4266 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs)
4267 +{
4268 +       int err;
4269 +       struct kstatfs kstfs;
4270 +
4271 +       err = vfs_statfs(&br->br_path, &kstfs);
4272 +       if (!err) {
4273 +               stfs->f_blocks = kstfs.f_blocks;
4274 +               stfs->f_bavail = kstfs.f_bavail;
4275 +               stfs->f_files = kstfs.f_files;
4276 +               stfs->f_ffree = kstfs.f_ffree;
4277 +       }
4278 +
4279 +       return err;
4280 +}
4281 diff -urN /usr/share/empty/fs/aufs/branch.h linux/fs/aufs/branch.h
4282 --- /usr/share/empty/fs/aufs/branch.h   1970-01-01 01:00:00.000000000 +0100
4283 +++ linux/fs/aufs/branch.h      2016-10-09 16:55:36.486034798 +0200
4284 @@ -0,0 +1,309 @@
4285 +/*
4286 + * Copyright (C) 2005-2016 Junjiro R. Okajima
4287 + *
4288 + * This program, aufs is free software; you can redistribute it and/or modify
4289 + * it under the terms of the GNU General Public License as published by
4290 + * the Free Software Foundation; either version 2 of the License, or
4291 + * (at your option) any later version.
4292 + *
4293 + * This program is distributed in the hope that it will be useful,
4294 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4295 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4296 + * GNU General Public License for more details.
4297 + *
4298 + * You should have received a copy of the GNU General Public License
4299 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4300 + */
4301 +
4302 +/*
4303 + * branch filesystems and xino for them
4304 + */
4305 +
4306 +#ifndef __AUFS_BRANCH_H__
4307 +#define __AUFS_BRANCH_H__
4308 +
4309 +#ifdef __KERNEL__
4310 +
4311 +#include <linux/mount.h>
4312 +#include "dynop.h"
4313 +#include "rwsem.h"
4314 +#include "super.h"
4315 +
4316 +/* ---------------------------------------------------------------------- */
4317 +
4318 +/* a xino file */
4319 +struct au_xino_file {
4320 +       struct file             *xi_file;
4321 +       struct mutex            xi_nondir_mtx;
4322 +
4323 +       /* todo: make xino files an array to support huge inode number */
4324 +
4325 +#ifdef CONFIG_DEBUG_FS
4326 +       struct dentry            *xi_dbgaufs;
4327 +#endif
4328 +};
4329 +
4330 +/* File-based Hierarchical Storage Management */
4331 +struct au_br_fhsm {
4332 +#ifdef CONFIG_AUFS_FHSM
4333 +       struct mutex            bf_lock;
4334 +       unsigned long           bf_jiffy;
4335 +       struct aufs_stfs        bf_stfs;
4336 +       int                     bf_readable;
4337 +#endif
4338 +};
4339 +
4340 +/* members for writable branch only */
4341 +enum {AuBrWh_BASE, AuBrWh_PLINK, AuBrWh_ORPH, AuBrWh_Last};
4342 +struct au_wbr {
4343 +       struct au_rwsem         wbr_wh_rwsem;
4344 +       struct dentry           *wbr_wh[AuBrWh_Last];
4345 +       atomic_t                wbr_wh_running;
4346 +#define wbr_whbase             wbr_wh[AuBrWh_BASE]     /* whiteout base */
4347 +#define wbr_plink              wbr_wh[AuBrWh_PLINK]    /* pseudo-link dir */
4348 +#define wbr_orph               wbr_wh[AuBrWh_ORPH]     /* dir for orphans */
4349 +
4350 +       /* mfs mode */
4351 +       unsigned long long      wbr_bytes;
4352 +};
4353 +
4354 +/* ext2 has 3 types of operations at least, ext3 has 4 */
4355 +#define AuBrDynOp (AuDyLast * 4)
4356 +
4357 +#ifdef CONFIG_AUFS_HFSNOTIFY
4358 +/* support for asynchronous destruction */
4359 +struct au_br_hfsnotify {
4360 +       struct fsnotify_group   *hfsn_group;
4361 +};
4362 +#endif
4363 +
4364 +/* sysfs entries */
4365 +struct au_brsysfs {
4366 +       char                    name[16];
4367 +       struct attribute        attr;
4368 +};
4369 +
4370 +enum {
4371 +       AuBrSysfs_BR,
4372 +       AuBrSysfs_BRID,
4373 +       AuBrSysfs_Last
4374 +};
4375 +
4376 +/* protected by superblock rwsem */
4377 +struct au_branch {
4378 +       struct au_xino_file     br_xino;
4379 +
4380 +       aufs_bindex_t           br_id;
4381 +
4382 +       int                     br_perm;
4383 +       struct path             br_path;
4384 +       spinlock_t              br_dykey_lock;
4385 +       struct au_dykey         *br_dykey[AuBrDynOp];
4386 +       struct percpu_counter   br_count;
4387 +
4388 +       struct au_wbr           *br_wbr;
4389 +       struct au_br_fhsm       *br_fhsm;
4390 +
4391 +       /* xino truncation */
4392 +       atomic_t                br_xino_running;
4393 +
4394 +#ifdef CONFIG_AUFS_HFSNOTIFY
4395 +       struct au_br_hfsnotify  *br_hfsn;
4396 +#endif
4397 +
4398 +#ifdef CONFIG_SYSFS
4399 +       /* entries under sysfs per mount-point */
4400 +       struct au_brsysfs       br_sysfs[AuBrSysfs_Last];
4401 +#endif
4402 +};
4403 +
4404 +/* ---------------------------------------------------------------------- */
4405 +
4406 +static inline struct vfsmount *au_br_mnt(struct au_branch *br)
4407 +{
4408 +       return br->br_path.mnt;
4409 +}
4410 +
4411 +static inline struct dentry *au_br_dentry(struct au_branch *br)
4412 +{
4413 +       return br->br_path.dentry;
4414 +}
4415 +
4416 +static inline struct super_block *au_br_sb(struct au_branch *br)
4417 +{
4418 +       return au_br_mnt(br)->mnt_sb;
4419 +}
4420 +
4421 +static inline void au_br_get(struct au_branch *br)
4422 +{
4423 +       percpu_counter_inc(&br->br_count);
4424 +}
4425 +
4426 +static inline void au_br_put(struct au_branch *br)
4427 +{
4428 +       percpu_counter_dec(&br->br_count);
4429 +}
4430 +
4431 +static inline s64 au_br_count(struct au_branch *br)
4432 +{
4433 +       return percpu_counter_sum(&br->br_count);
4434 +}
4435 +
4436 +static inline void au_br_count_init(struct au_branch *br)
4437 +{
4438 +       percpu_counter_init(&br->br_count, 0, GFP_NOFS);
4439 +}
4440 +
4441 +static inline void au_br_count_fin(struct au_branch *br)
4442 +{
4443 +       percpu_counter_destroy(&br->br_count);
4444 +}
4445 +
4446 +static inline int au_br_rdonly(struct au_branch *br)
4447 +{
4448 +       return ((au_br_sb(br)->s_flags & MS_RDONLY)
4449 +               || !au_br_writable(br->br_perm))
4450 +               ? -EROFS : 0;
4451 +}
4452 +
4453 +static inline int au_br_hnotifyable(int brperm __maybe_unused)
4454 +{
4455 +#ifdef CONFIG_AUFS_HNOTIFY
4456 +       return !(brperm & AuBrPerm_RR);
4457 +#else
4458 +       return 0;
4459 +#endif
4460 +}
4461 +
4462 +static inline int au_br_test_oflag(int oflag, struct au_branch *br)
4463 +{
4464 +       int err, exec_flag;
4465 +
4466 +       err = 0;
4467 +       exec_flag = oflag & __FMODE_EXEC;
4468 +       if (unlikely(exec_flag && path_noexec(&br->br_path)))
4469 +               err = -EACCES;
4470 +
4471 +       return err;
4472 +}
4473 +
4474 +/* ---------------------------------------------------------------------- */
4475 +
4476 +/* branch.c */
4477 +struct au_sbinfo;
4478 +void au_br_free(struct au_sbinfo *sinfo);
4479 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id);
4480 +struct au_opt_add;
4481 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount);
4482 +struct au_opt_del;
4483 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount);
4484 +long au_ibusy_ioctl(struct file *file, unsigned long arg);
4485 +#ifdef CONFIG_COMPAT
4486 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg);
4487 +#endif
4488 +struct au_opt_mod;
4489 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4490 +             int *do_refresh);
4491 +struct aufs_stfs;
4492 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs);
4493 +
4494 +/* xino.c */
4495 +static const loff_t au_loff_max = LLONG_MAX;
4496 +
4497 +int au_xib_trunc(struct super_block *sb);
4498 +ssize_t xino_fread(vfs_readf_t func, struct file *file, void *buf, size_t size,
4499 +                  loff_t *pos);
4500 +ssize_t xino_fwrite(vfs_writef_t func, struct file *file, void *buf,
4501 +                   size_t size, loff_t *pos);
4502 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src);
4503 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent);
4504 +ino_t au_xino_new_ino(struct super_block *sb);
4505 +void au_xino_delete_inode(struct inode *inode, const int unlinked);
4506 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4507 +                 ino_t ino);
4508 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4509 +                ino_t *ino);
4510 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t hino,
4511 +              struct file *base_file, int do_test);
4512 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex);
4513 +
4514 +struct au_opt_xino;
4515 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount);
4516 +void au_xino_clr(struct super_block *sb);
4517 +struct file *au_xino_def(struct super_block *sb);
4518 +int au_xino_path(struct seq_file *seq, struct file *file);
4519 +
4520 +/* ---------------------------------------------------------------------- */
4521 +
4522 +/* Superblock to branch */
4523 +static inline
4524 +aufs_bindex_t au_sbr_id(struct super_block *sb, aufs_bindex_t bindex)
4525 +{
4526 +       return au_sbr(sb, bindex)->br_id;
4527 +}
4528 +
4529 +static inline
4530 +struct vfsmount *au_sbr_mnt(struct super_block *sb, aufs_bindex_t bindex)
4531 +{
4532 +       return au_br_mnt(au_sbr(sb, bindex));
4533 +}
4534 +
4535 +static inline
4536 +struct super_block *au_sbr_sb(struct super_block *sb, aufs_bindex_t bindex)
4537 +{
4538 +       return au_br_sb(au_sbr(sb, bindex));
4539 +}
4540 +
4541 +static inline void au_sbr_get(struct super_block *sb, aufs_bindex_t bindex)
4542 +{
4543 +       au_br_get(au_sbr(sb, bindex));
4544 +}
4545 +
4546 +static inline void au_sbr_put(struct super_block *sb, aufs_bindex_t bindex)
4547 +{
4548 +       au_br_put(au_sbr(sb, bindex));
4549 +}
4550 +
4551 +static inline int au_sbr_perm(struct super_block *sb, aufs_bindex_t bindex)
4552 +{
4553 +       return au_sbr(sb, bindex)->br_perm;
4554 +}
4555 +
4556 +static inline int au_sbr_whable(struct super_block *sb, aufs_bindex_t bindex)
4557 +{
4558 +       return au_br_whable(au_sbr_perm(sb, bindex));
4559 +}
4560 +
4561 +/* ---------------------------------------------------------------------- */
4562 +
4563 +/*
4564 + * wbr_wh_read_lock, wbr_wh_write_lock
4565 + * wbr_wh_read_unlock, wbr_wh_write_unlock, wbr_wh_downgrade_lock
4566 + */
4567 +AuSimpleRwsemFuncs(wbr_wh, struct au_wbr *wbr, &wbr->wbr_wh_rwsem);
4568 +
4569 +#define WbrWhMustNoWaiters(wbr)        AuRwMustNoWaiters(&wbr->wbr_wh_rwsem)
4570 +#define WbrWhMustAnyLock(wbr)  AuRwMustAnyLock(&wbr->wbr_wh_rwsem)
4571 +#define WbrWhMustWriteLock(wbr)        AuRwMustWriteLock(&wbr->wbr_wh_rwsem)
4572 +
4573 +/* ---------------------------------------------------------------------- */
4574 +
4575 +#ifdef CONFIG_AUFS_FHSM
4576 +static inline void au_br_fhsm_init(struct au_br_fhsm *brfhsm)
4577 +{
4578 +       mutex_init(&brfhsm->bf_lock);
4579 +       brfhsm->bf_jiffy = 0;
4580 +       brfhsm->bf_readable = 0;
4581 +}
4582 +
4583 +static inline void au_br_fhsm_fin(struct au_br_fhsm *brfhsm)
4584 +{
4585 +       mutex_destroy(&brfhsm->bf_lock);
4586 +}
4587 +#else
4588 +AuStubVoid(au_br_fhsm_init, struct au_br_fhsm *brfhsm)
4589 +AuStubVoid(au_br_fhsm_fin, struct au_br_fhsm *brfhsm)
4590 +#endif
4591 +
4592 +#endif /* __KERNEL__ */
4593 +#endif /* __AUFS_BRANCH_H__ */
4594 diff -urN /usr/share/empty/fs/aufs/conf.mk linux/fs/aufs/conf.mk
4595 --- /usr/share/empty/fs/aufs/conf.mk    1970-01-01 01:00:00.000000000 +0100
4596 +++ linux/fs/aufs/conf.mk       2016-10-09 16:55:36.486034798 +0200
4597 @@ -0,0 +1,38 @@
4598 +
4599 +AuConfStr = CONFIG_AUFS_FS=${CONFIG_AUFS_FS}
4600 +
4601 +define AuConf
4602 +ifdef ${1}
4603 +AuConfStr += ${1}=${${1}}
4604 +endif
4605 +endef
4606 +
4607 +AuConfAll = BRANCH_MAX_127 BRANCH_MAX_511 BRANCH_MAX_1023 BRANCH_MAX_32767 \
4608 +       SBILIST \
4609 +       HNOTIFY HFSNOTIFY \
4610 +       EXPORT INO_T_64 \
4611 +       XATTR \
4612 +       FHSM \
4613 +       RDU \
4614 +       SHWH \
4615 +       BR_RAMFS \
4616 +       BR_FUSE POLL \
4617 +       BR_HFSPLUS \
4618 +       BDEV_LOOP \
4619 +       DEBUG MAGIC_SYSRQ
4620 +$(foreach i, ${AuConfAll}, \
4621 +       $(eval $(call AuConf,CONFIG_AUFS_${i})))
4622 +
4623 +AuConfName = ${obj}/conf.str
4624 +${AuConfName}.tmp: FORCE
4625 +       @echo ${AuConfStr} | tr ' ' '\n' | sed -e 's/^/"/' -e 's/$$/\\n"/' > $@
4626 +${AuConfName}: ${AuConfName}.tmp
4627 +       @diff -q $< $@ > /dev/null 2>&1 || { \
4628 +       echo '  GEN    ' $@; \
4629 +       cp -p $< $@; \
4630 +       }
4631 +FORCE:
4632 +clean-files += ${AuConfName} ${AuConfName}.tmp
4633 +${obj}/sysfs.o: ${AuConfName}
4634 +
4635 +-include ${srctree}/${src}/conf_priv.mk
4636 diff -urN /usr/share/empty/fs/aufs/cpup.c linux/fs/aufs/cpup.c
4637 --- /usr/share/empty/fs/aufs/cpup.c     1970-01-01 01:00:00.000000000 +0100
4638 +++ linux/fs/aufs/cpup.c        2016-10-09 16:55:38.886097714 +0200
4639 @@ -0,0 +1,1391 @@
4640 +/*
4641 + * Copyright (C) 2005-2016 Junjiro R. Okajima
4642 + *
4643 + * This program, aufs is free software; you can redistribute it and/or modify
4644 + * it under the terms of the GNU General Public License as published by
4645 + * the Free Software Foundation; either version 2 of the License, or
4646 + * (at your option) any later version.
4647 + *
4648 + * This program is distributed in the hope that it will be useful,
4649 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4650 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4651 + * GNU General Public License for more details.
4652 + *
4653 + * You should have received a copy of the GNU General Public License
4654 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4655 + */
4656 +
4657 +/*
4658 + * copy-up functions, see wbr_policy.c for copy-down
4659 + */
4660 +
4661 +#include <linux/fs_stack.h>
4662 +#include <linux/mm.h>
4663 +#include <linux/task_work.h>
4664 +#include "aufs.h"
4665 +
4666 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags)
4667 +{
4668 +       const unsigned int mask = S_DEAD | S_SWAPFILE | S_PRIVATE
4669 +               | S_NOATIME | S_NOCMTIME | S_AUTOMOUNT;
4670 +
4671 +       BUILD_BUG_ON(sizeof(iflags) != sizeof(dst->i_flags));
4672 +
4673 +       dst->i_flags |= iflags & ~mask;
4674 +       if (au_test_fs_notime(dst->i_sb))
4675 +               dst->i_flags |= S_NOATIME | S_NOCMTIME;
4676 +}
4677 +
4678 +void au_cpup_attr_timesizes(struct inode *inode)
4679 +{
4680 +       struct inode *h_inode;
4681 +
4682 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4683 +       fsstack_copy_attr_times(inode, h_inode);
4684 +       fsstack_copy_inode_size(inode, h_inode);
4685 +}
4686 +
4687 +void au_cpup_attr_nlink(struct inode *inode, int force)
4688 +{
4689 +       struct inode *h_inode;
4690 +       struct super_block *sb;
4691 +       aufs_bindex_t bindex, bbot;
4692 +
4693 +       sb = inode->i_sb;
4694 +       bindex = au_ibtop(inode);
4695 +       h_inode = au_h_iptr(inode, bindex);
4696 +       if (!force
4697 +           && !S_ISDIR(h_inode->i_mode)
4698 +           && au_opt_test(au_mntflags(sb), PLINK)
4699 +           && au_plink_test(inode))
4700 +               return;
4701 +
4702 +       /*
4703 +        * 0 can happen in revalidating.
4704 +        * h_inode->i_mutex may not be held here, but it is harmless since once
4705 +        * i_nlink reaches 0, it will never become positive except O_TMPFILE
4706 +        * case.
4707 +        * todo: O_TMPFILE+linkat(AT_SYMLINK_FOLLOW) bypassing aufs may cause
4708 +        *       the incorrect link count.
4709 +        */
4710 +       set_nlink(inode, h_inode->i_nlink);
4711 +
4712 +       /*
4713 +        * fewer nlink makes find(1) noisy, but larger nlink doesn't.
4714 +        * it may includes whplink directory.
4715 +        */
4716 +       if (S_ISDIR(h_inode->i_mode)) {
4717 +               bbot = au_ibbot(inode);
4718 +               for (bindex++; bindex <= bbot; bindex++) {
4719 +                       h_inode = au_h_iptr(inode, bindex);
4720 +                       if (h_inode)
4721 +                               au_add_nlink(inode, h_inode);
4722 +               }
4723 +       }
4724 +}
4725 +
4726 +void au_cpup_attr_changeable(struct inode *inode)
4727 +{
4728 +       struct inode *h_inode;
4729 +
4730 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4731 +       inode->i_mode = h_inode->i_mode;
4732 +       inode->i_uid = h_inode->i_uid;
4733 +       inode->i_gid = h_inode->i_gid;
4734 +       au_cpup_attr_timesizes(inode);
4735 +       au_cpup_attr_flags(inode, h_inode->i_flags);
4736 +}
4737 +
4738 +void au_cpup_igen(struct inode *inode, struct inode *h_inode)
4739 +{
4740 +       struct au_iinfo *iinfo = au_ii(inode);
4741 +
4742 +       IiMustWriteLock(inode);
4743 +
4744 +       iinfo->ii_higen = h_inode->i_generation;
4745 +       iinfo->ii_hsb1 = h_inode->i_sb;
4746 +}
4747 +
4748 +void au_cpup_attr_all(struct inode *inode, int force)
4749 +{
4750 +       struct inode *h_inode;
4751 +
4752 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
4753 +       au_cpup_attr_changeable(inode);
4754 +       if (inode->i_nlink > 0)
4755 +               au_cpup_attr_nlink(inode, force);
4756 +       inode->i_rdev = h_inode->i_rdev;
4757 +       inode->i_blkbits = h_inode->i_blkbits;
4758 +       au_cpup_igen(inode, h_inode);
4759 +}
4760 +
4761 +/* ---------------------------------------------------------------------- */
4762 +
4763 +/* Note: dt_dentry and dt_h_dentry are not dget/dput-ed */
4764 +
4765 +/* keep the timestamps of the parent dir when cpup */
4766 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
4767 +                   struct path *h_path)
4768 +{
4769 +       struct inode *h_inode;
4770 +
4771 +       dt->dt_dentry = dentry;
4772 +       dt->dt_h_path = *h_path;
4773 +       h_inode = d_inode(h_path->dentry);
4774 +       dt->dt_atime = h_inode->i_atime;
4775 +       dt->dt_mtime = h_inode->i_mtime;
4776 +       /* smp_mb(); */
4777 +}
4778 +
4779 +void au_dtime_revert(struct au_dtime *dt)
4780 +{
4781 +       struct iattr attr;
4782 +       int err;
4783 +
4784 +       attr.ia_atime = dt->dt_atime;
4785 +       attr.ia_mtime = dt->dt_mtime;
4786 +       attr.ia_valid = ATTR_FORCE | ATTR_MTIME | ATTR_MTIME_SET
4787 +               | ATTR_ATIME | ATTR_ATIME_SET;
4788 +
4789 +       /* no delegation since this is a directory */
4790 +       err = vfsub_notify_change(&dt->dt_h_path, &attr, /*delegated*/NULL);
4791 +       if (unlikely(err))
4792 +               pr_warn("restoring timestamps failed(%d). ignored\n", err);
4793 +}
4794 +
4795 +/* ---------------------------------------------------------------------- */
4796 +
4797 +/* internal use only */
4798 +struct au_cpup_reg_attr {
4799 +       int             valid;
4800 +       struct kstat    st;
4801 +       unsigned int    iflags; /* inode->i_flags */
4802 +};
4803 +
4804 +static noinline_for_stack
4805 +int cpup_iattr(struct dentry *dst, aufs_bindex_t bindex, struct dentry *h_src,
4806 +              struct au_cpup_reg_attr *h_src_attr)
4807 +{
4808 +       int err, sbits, icex;
4809 +       unsigned int mnt_flags;
4810 +       unsigned char verbose;
4811 +       struct iattr ia;
4812 +       struct path h_path;
4813 +       struct inode *h_isrc, *h_idst;
4814 +       struct kstat *h_st;
4815 +       struct au_branch *br;
4816 +
4817 +       h_path.dentry = au_h_dptr(dst, bindex);
4818 +       h_idst = d_inode(h_path.dentry);
4819 +       br = au_sbr(dst->d_sb, bindex);
4820 +       h_path.mnt = au_br_mnt(br);
4821 +       h_isrc = d_inode(h_src);
4822 +       ia.ia_valid = ATTR_FORCE | ATTR_UID | ATTR_GID
4823 +               | ATTR_ATIME | ATTR_MTIME
4824 +               | ATTR_ATIME_SET | ATTR_MTIME_SET;
4825 +       if (h_src_attr && h_src_attr->valid) {
4826 +               h_st = &h_src_attr->st;
4827 +               ia.ia_uid = h_st->uid;
4828 +               ia.ia_gid = h_st->gid;
4829 +               ia.ia_atime = h_st->atime;
4830 +               ia.ia_mtime = h_st->mtime;
4831 +               if (h_idst->i_mode != h_st->mode
4832 +                   && !S_ISLNK(h_idst->i_mode)) {
4833 +                       ia.ia_valid |= ATTR_MODE;
4834 +                       ia.ia_mode = h_st->mode;
4835 +               }
4836 +               sbits = !!(h_st->mode & (S_ISUID | S_ISGID));
4837 +               au_cpup_attr_flags(h_idst, h_src_attr->iflags);
4838 +       } else {
4839 +               ia.ia_uid = h_isrc->i_uid;
4840 +               ia.ia_gid = h_isrc->i_gid;
4841 +               ia.ia_atime = h_isrc->i_atime;
4842 +               ia.ia_mtime = h_isrc->i_mtime;
4843 +               if (h_idst->i_mode != h_isrc->i_mode
4844 +                   && !S_ISLNK(h_idst->i_mode)) {
4845 +                       ia.ia_valid |= ATTR_MODE;
4846 +                       ia.ia_mode = h_isrc->i_mode;
4847 +               }
4848 +               sbits = !!(h_isrc->i_mode & (S_ISUID | S_ISGID));
4849 +               au_cpup_attr_flags(h_idst, h_isrc->i_flags);
4850 +       }
4851 +       /* no delegation since it is just created */
4852 +       err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4853 +
4854 +       /* is this nfs only? */
4855 +       if (!err && sbits && au_test_nfs(h_path.dentry->d_sb)) {
4856 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
4857 +               ia.ia_mode = h_isrc->i_mode;
4858 +               err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4859 +       }
4860 +
4861 +       icex = br->br_perm & AuBrAttr_ICEX;
4862 +       if (!err) {
4863 +               mnt_flags = au_mntflags(dst->d_sb);
4864 +               verbose = !!au_opt_test(mnt_flags, VERBOSE);
4865 +               err = au_cpup_xattr(h_path.dentry, h_src, icex, verbose);
4866 +       }
4867 +
4868 +       return err;
4869 +}
4870 +
4871 +/* ---------------------------------------------------------------------- */
4872 +
4873 +static int au_do_copy_file(struct file *dst, struct file *src, loff_t len,
4874 +                          char *buf, unsigned long blksize)
4875 +{
4876 +       int err;
4877 +       size_t sz, rbytes, wbytes;
4878 +       unsigned char all_zero;
4879 +       char *p, *zp;
4880 +       struct inode *h_inode;
4881 +       /* reduce stack usage */
4882 +       struct iattr *ia;
4883 +
4884 +       zp = page_address(ZERO_PAGE(0));
4885 +       if (unlikely(!zp))
4886 +               return -ENOMEM; /* possible? */
4887 +
4888 +       err = 0;
4889 +       all_zero = 0;
4890 +       while (len) {
4891 +               AuDbg("len %lld\n", len);
4892 +               sz = blksize;
4893 +               if (len < blksize)
4894 +                       sz = len;
4895 +
4896 +               rbytes = 0;
4897 +               /* todo: signal_pending? */
4898 +               while (!rbytes || err == -EAGAIN || err == -EINTR) {
4899 +                       rbytes = vfsub_read_k(src, buf, sz, &src->f_pos);
4900 +                       err = rbytes;
4901 +               }
4902 +               if (unlikely(err < 0))
4903 +                       break;
4904 +
4905 +               all_zero = 0;
4906 +               if (len >= rbytes && rbytes == blksize)
4907 +                       all_zero = !memcmp(buf, zp, rbytes);
4908 +               if (!all_zero) {
4909 +                       wbytes = rbytes;
4910 +                       p = buf;
4911 +                       while (wbytes) {
4912 +                               size_t b;
4913 +
4914 +                               b = vfsub_write_k(dst, p, wbytes, &dst->f_pos);
4915 +                               err = b;
4916 +                               /* todo: signal_pending? */
4917 +                               if (unlikely(err == -EAGAIN || err == -EINTR))
4918 +                                       continue;
4919 +                               if (unlikely(err < 0))
4920 +                                       break;
4921 +                               wbytes -= b;
4922 +                               p += b;
4923 +                       }
4924 +                       if (unlikely(err < 0))
4925 +                               break;
4926 +               } else {
4927 +                       loff_t res;
4928 +
4929 +                       AuLabel(hole);
4930 +                       res = vfsub_llseek(dst, rbytes, SEEK_CUR);
4931 +                       err = res;
4932 +                       if (unlikely(res < 0))
4933 +                               break;
4934 +               }
4935 +               len -= rbytes;
4936 +               err = 0;
4937 +       }
4938 +
4939 +       /* the last block may be a hole */
4940 +       if (!err && all_zero) {
4941 +               AuLabel(last hole);
4942 +
4943 +               err = 1;
4944 +               if (au_test_nfs(dst->f_path.dentry->d_sb)) {
4945 +                       /* nfs requires this step to make last hole */
4946 +                       /* is this only nfs? */
4947 +                       do {
4948 +                               /* todo: signal_pending? */
4949 +                               err = vfsub_write_k(dst, "\0", 1, &dst->f_pos);
4950 +                       } while (err == -EAGAIN || err == -EINTR);
4951 +                       if (err == 1)
4952 +                               dst->f_pos--;
4953 +               }
4954 +
4955 +               if (err == 1) {
4956 +                       ia = (void *)buf;
4957 +                       ia->ia_size = dst->f_pos;
4958 +                       ia->ia_valid = ATTR_SIZE | ATTR_FILE;
4959 +                       ia->ia_file = dst;
4960 +                       h_inode = file_inode(dst);
4961 +                       inode_lock_nested(h_inode, AuLsc_I_CHILD2);
4962 +                       /* no delegation since it is just created */
4963 +                       err = vfsub_notify_change(&dst->f_path, ia,
4964 +                                                 /*delegated*/NULL);
4965 +                       inode_unlock(h_inode);
4966 +               }
4967 +       }
4968 +
4969 +       return err;
4970 +}
4971 +
4972 +int au_copy_file(struct file *dst, struct file *src, loff_t len)
4973 +{
4974 +       int err;
4975 +       unsigned long blksize;
4976 +       unsigned char do_kfree;
4977 +       char *buf;
4978 +
4979 +       err = -ENOMEM;
4980 +       blksize = dst->f_path.dentry->d_sb->s_blocksize;
4981 +       if (!blksize || PAGE_SIZE < blksize)
4982 +               blksize = PAGE_SIZE;
4983 +       AuDbg("blksize %lu\n", blksize);
4984 +       do_kfree = (blksize != PAGE_SIZE && blksize >= sizeof(struct iattr *));
4985 +       if (do_kfree)
4986 +               buf = kmalloc(blksize, GFP_NOFS);
4987 +       else
4988 +               buf = (void *)__get_free_page(GFP_NOFS);
4989 +       if (unlikely(!buf))
4990 +               goto out;
4991 +
4992 +       if (len > (1 << 22))
4993 +               AuDbg("copying a large file %lld\n", (long long)len);
4994 +
4995 +       src->f_pos = 0;
4996 +       dst->f_pos = 0;
4997 +       err = au_do_copy_file(dst, src, len, buf, blksize);
4998 +       if (do_kfree)
4999 +               au_delayed_kfree(buf);
5000 +       else
5001 +               au_delayed_free_page((unsigned long)buf);
5002 +
5003 +out:
5004 +       return err;
5005 +}
5006 +
5007 +/*
5008 + * to support a sparse file which is opened with O_APPEND,
5009 + * we need to close the file.
5010 + */
5011 +static int au_cp_regular(struct au_cp_generic *cpg)
5012 +{
5013 +       int err, i;
5014 +       enum { SRC, DST };
5015 +       struct {
5016 +               aufs_bindex_t bindex;
5017 +               unsigned int flags;
5018 +               struct dentry *dentry;
5019 +               int force_wr;
5020 +               struct file *file;
5021 +               void *label;
5022 +       } *f, file[] = {
5023 +               {
5024 +                       .bindex = cpg->bsrc,
5025 +                       .flags = O_RDONLY | O_NOATIME | O_LARGEFILE,
5026 +                       .label = &&out
5027 +               },
5028 +               {
5029 +                       .bindex = cpg->bdst,
5030 +                       .flags = O_WRONLY | O_NOATIME | O_LARGEFILE,
5031 +                       .force_wr = !!au_ftest_cpup(cpg->flags, RWDST),
5032 +                       .label = &&out_src
5033 +               }
5034 +       };
5035 +       struct super_block *sb;
5036 +       struct inode *h_src_inode;
5037 +       struct task_struct *tsk = current;
5038 +
5039 +       /* bsrc branch can be ro/rw. */
5040 +       sb = cpg->dentry->d_sb;
5041 +       f = file;
5042 +       for (i = 0; i < 2; i++, f++) {
5043 +               f->dentry = au_h_dptr(cpg->dentry, f->bindex);
5044 +               f->file = au_h_open(cpg->dentry, f->bindex, f->flags,
5045 +                                   /*file*/NULL, f->force_wr);
5046 +               err = PTR_ERR(f->file);
5047 +               if (IS_ERR(f->file))
5048 +                       goto *f->label;
5049 +       }
5050 +
5051 +       /* try stopping to update while we copyup */
5052 +       h_src_inode = d_inode(file[SRC].dentry);
5053 +       if (!au_test_nfs(h_src_inode->i_sb))
5054 +               IMustLock(h_src_inode);
5055 +       err = au_copy_file(file[DST].file, file[SRC].file, cpg->len);
5056 +
5057 +       /* i wonder if we had O_NO_DELAY_FPUT flag */
5058 +       if (tsk->flags & PF_KTHREAD)
5059 +               __fput_sync(file[DST].file);
5060 +       else {
5061 +               WARN(1, "%pD\nPlease report this warning to aufs-users ML",
5062 +                    file[DST].file);
5063 +               fput(file[DST].file);
5064 +               /*
5065 +                * too bad.
5066 +                * we have to call both since we don't know which place the file
5067 +                * was added to.
5068 +                */
5069 +               task_work_run();
5070 +               flush_delayed_fput();
5071 +       }
5072 +       au_sbr_put(sb, file[DST].bindex);
5073 +
5074 +out_src:
5075 +       fput(file[SRC].file);
5076 +       au_sbr_put(sb, file[SRC].bindex);
5077 +out:
5078 +       return err;
5079 +}
5080 +
5081 +static int au_do_cpup_regular(struct au_cp_generic *cpg,
5082 +                             struct au_cpup_reg_attr *h_src_attr)
5083 +{
5084 +       int err, rerr;
5085 +       loff_t l;
5086 +       struct path h_path;
5087 +       struct inode *h_src_inode, *h_dst_inode;
5088 +
5089 +       err = 0;
5090 +       h_src_inode = au_h_iptr(d_inode(cpg->dentry), cpg->bsrc);
5091 +       l = i_size_read(h_src_inode);
5092 +       if (cpg->len == -1 || l < cpg->len)
5093 +               cpg->len = l;
5094 +       if (cpg->len) {
5095 +               /* try stopping to update while we are referencing */
5096 +               inode_lock_nested(h_src_inode, AuLsc_I_CHILD);
5097 +               au_pin_hdir_unlock(cpg->pin);
5098 +
5099 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5100 +               h_path.mnt = au_sbr_mnt(cpg->dentry->d_sb, cpg->bsrc);
5101 +               h_src_attr->iflags = h_src_inode->i_flags;
5102 +               if (!au_test_nfs(h_src_inode->i_sb))
5103 +                       err = vfs_getattr(&h_path, &h_src_attr->st);
5104 +               else {
5105 +                       inode_unlock(h_src_inode);
5106 +                       err = vfs_getattr(&h_path, &h_src_attr->st);
5107 +                       inode_lock_nested(h_src_inode, AuLsc_I_CHILD);
5108 +               }
5109 +               if (unlikely(err)) {
5110 +                       inode_unlock(h_src_inode);
5111 +                       goto out;
5112 +               }
5113 +               h_src_attr->valid = 1;
5114 +               if (!au_test_nfs(h_src_inode->i_sb)) {
5115 +                       err = au_cp_regular(cpg);
5116 +                       inode_unlock(h_src_inode);
5117 +               } else {
5118 +                       inode_unlock(h_src_inode);
5119 +                       err = au_cp_regular(cpg);
5120 +               }
5121 +               rerr = au_pin_hdir_relock(cpg->pin);
5122 +               if (!err && rerr)
5123 +                       err = rerr;
5124 +       }
5125 +       if (!err && (h_src_inode->i_state & I_LINKABLE)) {
5126 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bdst);
5127 +               h_dst_inode = d_inode(h_path.dentry);
5128 +               spin_lock(&h_dst_inode->i_lock);
5129 +               h_dst_inode->i_state |= I_LINKABLE;
5130 +               spin_unlock(&h_dst_inode->i_lock);
5131 +       }
5132 +
5133 +out:
5134 +       return err;
5135 +}
5136 +
5137 +static int au_do_cpup_symlink(struct path *h_path, struct dentry *h_src,
5138 +                             struct inode *h_dir)
5139 +{
5140 +       int err, symlen;
5141 +       mm_segment_t old_fs;
5142 +       union {
5143 +               char *k;
5144 +               char __user *u;
5145 +       } sym;
5146 +       struct inode *h_inode = d_inode(h_src);
5147 +       const struct inode_operations *h_iop = h_inode->i_op;
5148 +
5149 +       err = -ENOSYS;
5150 +       if (unlikely(!h_iop->readlink))
5151 +               goto out;
5152 +
5153 +       err = -ENOMEM;
5154 +       sym.k = (void *)__get_free_page(GFP_NOFS);
5155 +       if (unlikely(!sym.k))
5156 +               goto out;
5157 +
5158 +       /* unnecessary to support mmap_sem since symlink is not mmap-able */
5159 +       old_fs = get_fs();
5160 +       set_fs(KERNEL_DS);
5161 +       symlen = h_iop->readlink(h_src, sym.u, PATH_MAX);
5162 +       err = symlen;
5163 +       set_fs(old_fs);
5164 +
5165 +       if (symlen > 0) {
5166 +               sym.k[symlen] = 0;
5167 +               err = vfsub_symlink(h_dir, h_path, sym.k);
5168 +       }
5169 +       au_delayed_free_page((unsigned long)sym.k);
5170 +
5171 +out:
5172 +       return err;
5173 +}
5174 +
5175 +/*
5176 + * regardless 'acl' option, reset all ACL.
5177 + * All ACL will be copied up later from the original entry on the lower branch.
5178 + */
5179 +static int au_reset_acl(struct inode *h_dir, struct path *h_path, umode_t mode)
5180 +{
5181 +       int err;
5182 +       struct dentry *h_dentry;
5183 +       struct inode *h_inode;
5184 +
5185 +       h_dentry = h_path->dentry;
5186 +       h_inode = d_inode(h_dentry);
5187 +       /* forget_all_cached_acls(h_inode)); */
5188 +       err = vfsub_removexattr(h_dentry, XATTR_NAME_POSIX_ACL_ACCESS);
5189 +       AuTraceErr(err);
5190 +       if (err == -EOPNOTSUPP)
5191 +               err = 0;
5192 +       if (!err)
5193 +               err = vfsub_acl_chmod(h_inode, mode);
5194 +
5195 +       AuTraceErr(err);
5196 +       return err;
5197 +}
5198 +
5199 +static int au_do_cpup_dir(struct au_cp_generic *cpg, struct dentry *dst_parent,
5200 +                         struct inode *h_dir, struct path *h_path)
5201 +{
5202 +       int err;
5203 +       struct inode *dir, *inode;
5204 +
5205 +       err = vfsub_removexattr(h_path->dentry, XATTR_NAME_POSIX_ACL_DEFAULT);
5206 +       AuTraceErr(err);
5207 +       if (err == -EOPNOTSUPP)
5208 +               err = 0;
5209 +       if (unlikely(err))
5210 +               goto out;
5211 +
5212 +       /*
5213 +        * strange behaviour from the users view,
5214 +        * particularry setattr case
5215 +        */
5216 +       dir = d_inode(dst_parent);
5217 +       if (au_ibtop(dir) == cpg->bdst)
5218 +               au_cpup_attr_nlink(dir, /*force*/1);
5219 +       inode = d_inode(cpg->dentry);
5220 +       au_cpup_attr_nlink(inode, /*force*/1);
5221 +
5222 +out:
5223 +       return err;
5224 +}
5225 +
5226 +static noinline_for_stack
5227 +int cpup_entry(struct au_cp_generic *cpg, struct dentry *dst_parent,
5228 +              struct au_cpup_reg_attr *h_src_attr)
5229 +{
5230 +       int err;
5231 +       umode_t mode;
5232 +       unsigned int mnt_flags;
5233 +       unsigned char isdir, isreg, force;
5234 +       const unsigned char do_dt = !!au_ftest_cpup(cpg->flags, DTIME);
5235 +       struct au_dtime dt;
5236 +       struct path h_path;
5237 +       struct dentry *h_src, *h_dst, *h_parent;
5238 +       struct inode *h_inode, *h_dir;
5239 +       struct super_block *sb;
5240 +
5241 +       /* bsrc branch can be ro/rw. */
5242 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5243 +       h_inode = d_inode(h_src);
5244 +       AuDebugOn(h_inode != au_h_iptr(d_inode(cpg->dentry), cpg->bsrc));
5245 +
5246 +       /* try stopping to be referenced while we are creating */
5247 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5248 +       if (au_ftest_cpup(cpg->flags, RENAME))
5249 +               AuDebugOn(strncmp(h_dst->d_name.name, AUFS_WH_PFX,
5250 +                                 AUFS_WH_PFX_LEN));
5251 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5252 +       h_dir = d_inode(h_parent);
5253 +       IMustLock(h_dir);
5254 +       AuDebugOn(h_parent != h_dst->d_parent);
5255 +
5256 +       sb = cpg->dentry->d_sb;
5257 +       h_path.mnt = au_sbr_mnt(sb, cpg->bdst);
5258 +       if (do_dt) {
5259 +               h_path.dentry = h_parent;
5260 +               au_dtime_store(&dt, dst_parent, &h_path);
5261 +       }
5262 +       h_path.dentry = h_dst;
5263 +
5264 +       isreg = 0;
5265 +       isdir = 0;
5266 +       mode = h_inode->i_mode;
5267 +       switch (mode & S_IFMT) {
5268 +       case S_IFREG:
5269 +               isreg = 1;
5270 +               err = vfsub_create(h_dir, &h_path, S_IRUSR | S_IWUSR,
5271 +                                  /*want_excl*/true);
5272 +               if (!err)
5273 +                       err = au_do_cpup_regular(cpg, h_src_attr);
5274 +               break;
5275 +       case S_IFDIR:
5276 +               isdir = 1;
5277 +               err = vfsub_mkdir(h_dir, &h_path, mode);
5278 +               if (!err)
5279 +                       err = au_do_cpup_dir(cpg, dst_parent, h_dir, &h_path);
5280 +               break;
5281 +       case S_IFLNK:
5282 +               err = au_do_cpup_symlink(&h_path, h_src, h_dir);
5283 +               break;
5284 +       case S_IFCHR:
5285 +       case S_IFBLK:
5286 +               AuDebugOn(!capable(CAP_MKNOD));
5287 +               /*FALLTHROUGH*/
5288 +       case S_IFIFO:
5289 +       case S_IFSOCK:
5290 +               err = vfsub_mknod(h_dir, &h_path, mode, h_inode->i_rdev);
5291 +               break;
5292 +       default:
5293 +               AuIOErr("Unknown inode type 0%o\n", mode);
5294 +               err = -EIO;
5295 +       }
5296 +       if (!err)
5297 +               err = au_reset_acl(h_dir, &h_path, mode);
5298 +
5299 +       mnt_flags = au_mntflags(sb);
5300 +       if (!au_opt_test(mnt_flags, UDBA_NONE)
5301 +           && !isdir
5302 +           && au_opt_test(mnt_flags, XINO)
5303 +           && (h_inode->i_nlink == 1
5304 +               || (h_inode->i_state & I_LINKABLE))
5305 +           /* todo: unnecessary? */
5306 +           /* && d_inode(cpg->dentry)->i_nlink == 1 */
5307 +           && cpg->bdst < cpg->bsrc
5308 +           && !au_ftest_cpup(cpg->flags, KEEPLINO))
5309 +               au_xino_write(sb, cpg->bsrc, h_inode->i_ino, /*ino*/0);
5310 +               /* ignore this error */
5311 +
5312 +       if (!err) {
5313 +               force = 0;
5314 +               if (isreg) {
5315 +                       force = !!cpg->len;
5316 +                       if (cpg->len == -1)
5317 +                               force = !!i_size_read(h_inode);
5318 +               }
5319 +               au_fhsm_wrote(sb, cpg->bdst, force);
5320 +       }
5321 +
5322 +       if (do_dt)
5323 +               au_dtime_revert(&dt);
5324 +       return err;
5325 +}
5326 +
5327 +static int au_do_ren_after_cpup(struct au_cp_generic *cpg, struct path *h_path)
5328 +{
5329 +       int err;
5330 +       struct dentry *dentry, *h_dentry, *h_parent, *parent;
5331 +       struct inode *h_dir;
5332 +       aufs_bindex_t bdst;
5333 +
5334 +       dentry = cpg->dentry;
5335 +       bdst = cpg->bdst;
5336 +       h_dentry = au_h_dptr(dentry, bdst);
5337 +       if (!au_ftest_cpup(cpg->flags, OVERWRITE)) {
5338 +               dget(h_dentry);
5339 +               au_set_h_dptr(dentry, bdst, NULL);
5340 +               err = au_lkup_neg(dentry, bdst, /*wh*/0);
5341 +               if (!err)
5342 +                       h_path->dentry = dget(au_h_dptr(dentry, bdst));
5343 +               au_set_h_dptr(dentry, bdst, h_dentry);
5344 +       } else {
5345 +               err = 0;
5346 +               parent = dget_parent(dentry);
5347 +               h_parent = au_h_dptr(parent, bdst);
5348 +               dput(parent);
5349 +               h_path->dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
5350 +               if (IS_ERR(h_path->dentry))
5351 +                       err = PTR_ERR(h_path->dentry);
5352 +       }
5353 +       if (unlikely(err))
5354 +               goto out;
5355 +
5356 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
5357 +       h_dir = d_inode(h_parent);
5358 +       IMustLock(h_dir);
5359 +       AuDbg("%pd %pd\n", h_dentry, h_path->dentry);
5360 +       /* no delegation since it is just created */
5361 +       err = vfsub_rename(h_dir, h_dentry, h_dir, h_path, /*delegated*/NULL);
5362 +       dput(h_path->dentry);
5363 +
5364 +out:
5365 +       return err;
5366 +}
5367 +
5368 +/*
5369 + * copyup the @dentry from @bsrc to @bdst.
5370 + * the caller must set the both of lower dentries.
5371 + * @len is for truncating when it is -1 copyup the entire file.
5372 + * in link/rename cases, @dst_parent may be different from the real one.
5373 + * basic->bsrc can be larger than basic->bdst.
5374 + */
5375 +static int au_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5376 +{
5377 +       int err, rerr;
5378 +       aufs_bindex_t old_ibtop;
5379 +       unsigned char isdir, plink;
5380 +       struct dentry *h_src, *h_dst, *h_parent;
5381 +       struct inode *dst_inode, *h_dir, *inode, *delegated, *src_inode;
5382 +       struct super_block *sb;
5383 +       struct au_branch *br;
5384 +       /* to reuduce stack size */
5385 +       struct {
5386 +               struct au_dtime dt;
5387 +               struct path h_path;
5388 +               struct au_cpup_reg_attr h_src_attr;
5389 +       } *a;
5390 +
5391 +       err = -ENOMEM;
5392 +       a = kmalloc(sizeof(*a), GFP_NOFS);
5393 +       if (unlikely(!a))
5394 +               goto out;
5395 +       a->h_src_attr.valid = 0;
5396 +
5397 +       sb = cpg->dentry->d_sb;
5398 +       br = au_sbr(sb, cpg->bdst);
5399 +       a->h_path.mnt = au_br_mnt(br);
5400 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5401 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5402 +       h_dir = d_inode(h_parent);
5403 +       IMustLock(h_dir);
5404 +
5405 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5406 +       inode = d_inode(cpg->dentry);
5407 +
5408 +       if (!dst_parent)
5409 +               dst_parent = dget_parent(cpg->dentry);
5410 +       else
5411 +               dget(dst_parent);
5412 +
5413 +       plink = !!au_opt_test(au_mntflags(sb), PLINK);
5414 +       dst_inode = au_h_iptr(inode, cpg->bdst);
5415 +       if (dst_inode) {
5416 +               if (unlikely(!plink)) {
5417 +                       err = -EIO;
5418 +                       AuIOErr("hi%lu(i%lu) exists on b%d "
5419 +                               "but plink is disabled\n",
5420 +                               dst_inode->i_ino, inode->i_ino, cpg->bdst);
5421 +                       goto out_parent;
5422 +               }
5423 +
5424 +               if (dst_inode->i_nlink) {
5425 +                       const int do_dt = au_ftest_cpup(cpg->flags, DTIME);
5426 +
5427 +                       h_src = au_plink_lkup(inode, cpg->bdst);
5428 +                       err = PTR_ERR(h_src);
5429 +                       if (IS_ERR(h_src))
5430 +                               goto out_parent;
5431 +                       if (unlikely(d_is_negative(h_src))) {
5432 +                               err = -EIO;
5433 +                               AuIOErr("i%lu exists on b%d "
5434 +                                       "but not pseudo-linked\n",
5435 +                                       inode->i_ino, cpg->bdst);
5436 +                               dput(h_src);
5437 +                               goto out_parent;
5438 +                       }
5439 +
5440 +                       if (do_dt) {
5441 +                               a->h_path.dentry = h_parent;
5442 +                               au_dtime_store(&a->dt, dst_parent, &a->h_path);
5443 +                       }
5444 +
5445 +                       a->h_path.dentry = h_dst;
5446 +                       delegated = NULL;
5447 +                       err = vfsub_link(h_src, h_dir, &a->h_path, &delegated);
5448 +                       if (!err && au_ftest_cpup(cpg->flags, RENAME))
5449 +                               err = au_do_ren_after_cpup(cpg, &a->h_path);
5450 +                       if (do_dt)
5451 +                               au_dtime_revert(&a->dt);
5452 +                       if (unlikely(err == -EWOULDBLOCK)) {
5453 +                               pr_warn("cannot retry for NFSv4 delegation"
5454 +                                       " for an internal link\n");
5455 +                               iput(delegated);
5456 +                       }
5457 +                       dput(h_src);
5458 +                       goto out_parent;
5459 +               } else
5460 +                       /* todo: cpup_wh_file? */
5461 +                       /* udba work */
5462 +                       au_update_ibrange(inode, /*do_put_zero*/1);
5463 +       }
5464 +
5465 +       isdir = S_ISDIR(inode->i_mode);
5466 +       old_ibtop = au_ibtop(inode);
5467 +       err = cpup_entry(cpg, dst_parent, &a->h_src_attr);
5468 +       if (unlikely(err))
5469 +               goto out_rev;
5470 +       dst_inode = d_inode(h_dst);
5471 +       inode_lock_nested(dst_inode, AuLsc_I_CHILD2);
5472 +       /* todo: necessary? */
5473 +       /* au_pin_hdir_unlock(cpg->pin); */
5474 +
5475 +       err = cpup_iattr(cpg->dentry, cpg->bdst, h_src, &a->h_src_attr);
5476 +       if (unlikely(err)) {
5477 +               /* todo: necessary? */
5478 +               /* au_pin_hdir_relock(cpg->pin); */ /* ignore an error */
5479 +               inode_unlock(dst_inode);
5480 +               goto out_rev;
5481 +       }
5482 +
5483 +       if (cpg->bdst < old_ibtop) {
5484 +               if (S_ISREG(inode->i_mode)) {
5485 +                       err = au_dy_iaop(inode, cpg->bdst, dst_inode);
5486 +                       if (unlikely(err)) {
5487 +                               /* ignore an error */
5488 +                               /* au_pin_hdir_relock(cpg->pin); */
5489 +                               inode_unlock(dst_inode);
5490 +                               goto out_rev;
5491 +                       }
5492 +               }
5493 +               au_set_ibtop(inode, cpg->bdst);
5494 +       } else
5495 +               au_set_ibbot(inode, cpg->bdst);
5496 +       au_set_h_iptr(inode, cpg->bdst, au_igrab(dst_inode),
5497 +                     au_hi_flags(inode, isdir));
5498 +
5499 +       /* todo: necessary? */
5500 +       /* err = au_pin_hdir_relock(cpg->pin); */
5501 +       inode_unlock(dst_inode);
5502 +       if (unlikely(err))
5503 +               goto out_rev;
5504 +
5505 +       src_inode = d_inode(h_src);
5506 +       if (!isdir
5507 +           && (src_inode->i_nlink > 1
5508 +               || src_inode->i_state & I_LINKABLE)
5509 +           && plink)
5510 +               au_plink_append(inode, cpg->bdst, h_dst);
5511 +
5512 +       if (au_ftest_cpup(cpg->flags, RENAME)) {
5513 +               a->h_path.dentry = h_dst;
5514 +               err = au_do_ren_after_cpup(cpg, &a->h_path);
5515 +       }
5516 +       if (!err)
5517 +               goto out_parent; /* success */
5518 +
5519 +       /* revert */
5520 +out_rev:
5521 +       a->h_path.dentry = h_parent;
5522 +       au_dtime_store(&a->dt, dst_parent, &a->h_path);
5523 +       a->h_path.dentry = h_dst;
5524 +       rerr = 0;
5525 +       if (d_is_positive(h_dst)) {
5526 +               if (!isdir) {
5527 +                       /* no delegation since it is just created */
5528 +                       rerr = vfsub_unlink(h_dir, &a->h_path,
5529 +                                           /*delegated*/NULL, /*force*/0);
5530 +               } else
5531 +                       rerr = vfsub_rmdir(h_dir, &a->h_path);
5532 +       }
5533 +       au_dtime_revert(&a->dt);
5534 +       if (rerr) {
5535 +               AuIOErr("failed removing broken entry(%d, %d)\n", err, rerr);
5536 +               err = -EIO;
5537 +       }
5538 +out_parent:
5539 +       dput(dst_parent);
5540 +       au_delayed_kfree(a);
5541 +out:
5542 +       return err;
5543 +}
5544 +
5545 +#if 0 /* reserved */
5546 +struct au_cpup_single_args {
5547 +       int *errp;
5548 +       struct au_cp_generic *cpg;
5549 +       struct dentry *dst_parent;
5550 +};
5551 +
5552 +static void au_call_cpup_single(void *args)
5553 +{
5554 +       struct au_cpup_single_args *a = args;
5555 +
5556 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5557 +       *a->errp = au_cpup_single(a->cpg, a->dst_parent);
5558 +       au_pin_hdir_release(a->cpg->pin);
5559 +}
5560 +#endif
5561 +
5562 +/*
5563 + * prevent SIGXFSZ in copy-up.
5564 + * testing CAP_MKNOD is for generic fs,
5565 + * but CAP_FSETID is for xfs only, currently.
5566 + */
5567 +static int au_cpup_sio_test(struct au_pin *pin, umode_t mode)
5568 +{
5569 +       int do_sio;
5570 +       struct super_block *sb;
5571 +       struct inode *h_dir;
5572 +
5573 +       do_sio = 0;
5574 +       sb = au_pinned_parent(pin)->d_sb;
5575 +       if (!au_wkq_test()
5576 +           && (!au_sbi(sb)->si_plink_maint_pid
5577 +               || au_plink_maint(sb, AuLock_NOPLM))) {
5578 +               switch (mode & S_IFMT) {
5579 +               case S_IFREG:
5580 +                       /* no condition about RLIMIT_FSIZE and the file size */
5581 +                       do_sio = 1;
5582 +                       break;
5583 +               case S_IFCHR:
5584 +               case S_IFBLK:
5585 +                       do_sio = !capable(CAP_MKNOD);
5586 +                       break;
5587 +               }
5588 +               if (!do_sio)
5589 +                       do_sio = ((mode & (S_ISUID | S_ISGID))
5590 +                                 && !capable(CAP_FSETID));
5591 +               /* this workaround may be removed in the future */
5592 +               if (!do_sio) {
5593 +                       h_dir = au_pinned_h_dir(pin);
5594 +                       do_sio = h_dir->i_mode & S_ISVTX;
5595 +               }
5596 +       }
5597 +
5598 +       return do_sio;
5599 +}
5600 +
5601 +#if 0 /* reserved */
5602 +int au_sio_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5603 +{
5604 +       int err, wkq_err;
5605 +       struct dentry *h_dentry;
5606 +
5607 +       h_dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5608 +       if (!au_cpup_sio_test(pin, d_inode(h_dentry)->i_mode))
5609 +               err = au_cpup_single(cpg, dst_parent);
5610 +       else {
5611 +               struct au_cpup_single_args args = {
5612 +                       .errp           = &err,
5613 +                       .cpg            = cpg,
5614 +                       .dst_parent     = dst_parent
5615 +               };
5616 +               wkq_err = au_wkq_wait(au_call_cpup_single, &args);
5617 +               if (unlikely(wkq_err))
5618 +                       err = wkq_err;
5619 +       }
5620 +
5621 +       return err;
5622 +}
5623 +#endif
5624 +
5625 +/*
5626 + * copyup the @dentry from the first active lower branch to @bdst,
5627 + * using au_cpup_single().
5628 + */
5629 +static int au_cpup_simple(struct au_cp_generic *cpg)
5630 +{
5631 +       int err;
5632 +       unsigned int flags_orig;
5633 +       struct dentry *dentry;
5634 +
5635 +       AuDebugOn(cpg->bsrc < 0);
5636 +
5637 +       dentry = cpg->dentry;
5638 +       DiMustWriteLock(dentry);
5639 +
5640 +       err = au_lkup_neg(dentry, cpg->bdst, /*wh*/1);
5641 +       if (!err) {
5642 +               flags_orig = cpg->flags;
5643 +               au_fset_cpup(cpg->flags, RENAME);
5644 +               err = au_cpup_single(cpg, NULL);
5645 +               cpg->flags = flags_orig;
5646 +               if (!err)
5647 +                       return 0; /* success */
5648 +
5649 +               /* revert */
5650 +               au_set_h_dptr(dentry, cpg->bdst, NULL);
5651 +               au_set_dbtop(dentry, cpg->bsrc);
5652 +       }
5653 +
5654 +       return err;
5655 +}
5656 +
5657 +struct au_cpup_simple_args {
5658 +       int *errp;
5659 +       struct au_cp_generic *cpg;
5660 +};
5661 +
5662 +static void au_call_cpup_simple(void *args)
5663 +{
5664 +       struct au_cpup_simple_args *a = args;
5665 +
5666 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5667 +       *a->errp = au_cpup_simple(a->cpg);
5668 +       au_pin_hdir_release(a->cpg->pin);
5669 +}
5670 +
5671 +static int au_do_sio_cpup_simple(struct au_cp_generic *cpg)
5672 +{
5673 +       int err, wkq_err;
5674 +       struct dentry *dentry, *parent;
5675 +       struct file *h_file;
5676 +       struct inode *h_dir;
5677 +
5678 +       dentry = cpg->dentry;
5679 +       h_file = NULL;
5680 +       if (au_ftest_cpup(cpg->flags, HOPEN)) {
5681 +               AuDebugOn(cpg->bsrc < 0);
5682 +               h_file = au_h_open_pre(dentry, cpg->bsrc, /*force_wr*/0);
5683 +               err = PTR_ERR(h_file);
5684 +               if (IS_ERR(h_file))
5685 +                       goto out;
5686 +       }
5687 +
5688 +       parent = dget_parent(dentry);
5689 +       h_dir = au_h_iptr(d_inode(parent), cpg->bdst);
5690 +       if (!au_test_h_perm_sio(h_dir, MAY_EXEC | MAY_WRITE)
5691 +           && !au_cpup_sio_test(cpg->pin, d_inode(dentry)->i_mode))
5692 +               err = au_cpup_simple(cpg);
5693 +       else {
5694 +               struct au_cpup_simple_args args = {
5695 +                       .errp           = &err,
5696 +                       .cpg            = cpg
5697 +               };
5698 +               wkq_err = au_wkq_wait(au_call_cpup_simple, &args);
5699 +               if (unlikely(wkq_err))
5700 +                       err = wkq_err;
5701 +       }
5702 +
5703 +       dput(parent);
5704 +       if (h_file)
5705 +               au_h_open_post(dentry, cpg->bsrc, h_file);
5706 +
5707 +out:
5708 +       return err;
5709 +}
5710 +
5711 +int au_sio_cpup_simple(struct au_cp_generic *cpg)
5712 +{
5713 +       aufs_bindex_t bsrc, bbot;
5714 +       struct dentry *dentry, *h_dentry;
5715 +
5716 +       if (cpg->bsrc < 0) {
5717 +               dentry = cpg->dentry;
5718 +               bbot = au_dbbot(dentry);
5719 +               for (bsrc = cpg->bdst + 1; bsrc <= bbot; bsrc++) {
5720 +                       h_dentry = au_h_dptr(dentry, bsrc);
5721 +                       if (h_dentry) {
5722 +                               AuDebugOn(d_is_negative(h_dentry));
5723 +                               break;
5724 +                       }
5725 +               }
5726 +               AuDebugOn(bsrc > bbot);
5727 +               cpg->bsrc = bsrc;
5728 +       }
5729 +       AuDebugOn(cpg->bsrc <= cpg->bdst);
5730 +       return au_do_sio_cpup_simple(cpg);
5731 +}
5732 +
5733 +int au_sio_cpdown_simple(struct au_cp_generic *cpg)
5734 +{
5735 +       AuDebugOn(cpg->bdst <= cpg->bsrc);
5736 +       return au_do_sio_cpup_simple(cpg);
5737 +}
5738 +
5739 +/* ---------------------------------------------------------------------- */
5740 +
5741 +/*
5742 + * copyup the deleted file for writing.
5743 + */
5744 +static int au_do_cpup_wh(struct au_cp_generic *cpg, struct dentry *wh_dentry,
5745 +                        struct file *file)
5746 +{
5747 +       int err;
5748 +       unsigned int flags_orig;
5749 +       aufs_bindex_t bsrc_orig;
5750 +       struct au_dinfo *dinfo;
5751 +       struct {
5752 +               struct au_hdentry *hd;
5753 +               struct dentry *h_dentry;
5754 +       } hdst, hsrc;
5755 +
5756 +       dinfo = au_di(cpg->dentry);
5757 +       AuRwMustWriteLock(&dinfo->di_rwsem);
5758 +
5759 +       bsrc_orig = cpg->bsrc;
5760 +       cpg->bsrc = dinfo->di_btop;
5761 +       hdst.hd = au_hdentry(dinfo, cpg->bdst);
5762 +       hdst.h_dentry = hdst.hd->hd_dentry;
5763 +       hdst.hd->hd_dentry = wh_dentry;
5764 +       dinfo->di_btop = cpg->bdst;
5765 +
5766 +       hsrc.h_dentry = NULL;
5767 +       if (file) {
5768 +               hsrc.hd = au_hdentry(dinfo, cpg->bsrc);
5769 +               hsrc.h_dentry = hsrc.hd->hd_dentry;
5770 +               hsrc.hd->hd_dentry = au_hf_top(file)->f_path.dentry;
5771 +       }
5772 +       flags_orig = cpg->flags;
5773 +       cpg->flags = !AuCpup_DTIME;
5774 +       err = au_cpup_single(cpg, /*h_parent*/NULL);
5775 +       cpg->flags = flags_orig;
5776 +       if (file) {
5777 +               if (!err)
5778 +                       err = au_reopen_nondir(file);
5779 +               hsrc.hd->hd_dentry = hsrc.h_dentry;
5780 +       }
5781 +       hdst.hd->hd_dentry = hdst.h_dentry;
5782 +       dinfo->di_btop = cpg->bsrc;
5783 +       cpg->bsrc = bsrc_orig;
5784 +
5785 +       return err;
5786 +}
5787 +
5788 +static int au_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5789 +{
5790 +       int err;
5791 +       aufs_bindex_t bdst;
5792 +       struct au_dtime dt;
5793 +       struct dentry *dentry, *parent, *h_parent, *wh_dentry;
5794 +       struct au_branch *br;
5795 +       struct path h_path;
5796 +
5797 +       dentry = cpg->dentry;
5798 +       bdst = cpg->bdst;
5799 +       br = au_sbr(dentry->d_sb, bdst);
5800 +       parent = dget_parent(dentry);
5801 +       h_parent = au_h_dptr(parent, bdst);
5802 +       wh_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
5803 +       err = PTR_ERR(wh_dentry);
5804 +       if (IS_ERR(wh_dentry))
5805 +               goto out;
5806 +
5807 +       h_path.dentry = h_parent;
5808 +       h_path.mnt = au_br_mnt(br);
5809 +       au_dtime_store(&dt, parent, &h_path);
5810 +       err = au_do_cpup_wh(cpg, wh_dentry, file);
5811 +       if (unlikely(err))
5812 +               goto out_wh;
5813 +
5814 +       dget(wh_dentry);
5815 +       h_path.dentry = wh_dentry;
5816 +       if (!d_is_dir(wh_dentry)) {
5817 +               /* no delegation since it is just created */
5818 +               err = vfsub_unlink(d_inode(h_parent), &h_path,
5819 +                                  /*delegated*/NULL, /*force*/0);
5820 +       } else
5821 +               err = vfsub_rmdir(d_inode(h_parent), &h_path);
5822 +       if (unlikely(err)) {
5823 +               AuIOErr("failed remove copied-up tmp file %pd(%d)\n",
5824 +                       wh_dentry, err);
5825 +               err = -EIO;
5826 +       }
5827 +       au_dtime_revert(&dt);
5828 +       au_set_hi_wh(d_inode(dentry), bdst, wh_dentry);
5829 +
5830 +out_wh:
5831 +       dput(wh_dentry);
5832 +out:
5833 +       dput(parent);
5834 +       return err;
5835 +}
5836 +
5837 +struct au_cpup_wh_args {
5838 +       int *errp;
5839 +       struct au_cp_generic *cpg;
5840 +       struct file *file;
5841 +};
5842 +
5843 +static void au_call_cpup_wh(void *args)
5844 +{
5845 +       struct au_cpup_wh_args *a = args;
5846 +
5847 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5848 +       *a->errp = au_cpup_wh(a->cpg, a->file);
5849 +       au_pin_hdir_release(a->cpg->pin);
5850 +}
5851 +
5852 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5853 +{
5854 +       int err, wkq_err;
5855 +       aufs_bindex_t bdst;
5856 +       struct dentry *dentry, *parent, *h_orph, *h_parent;
5857 +       struct inode *dir, *h_dir, *h_tmpdir;
5858 +       struct au_wbr *wbr;
5859 +       struct au_pin wh_pin, *pin_orig;
5860 +
5861 +       dentry = cpg->dentry;
5862 +       bdst = cpg->bdst;
5863 +       parent = dget_parent(dentry);
5864 +       dir = d_inode(parent);
5865 +       h_orph = NULL;
5866 +       h_parent = NULL;
5867 +       h_dir = au_igrab(au_h_iptr(dir, bdst));
5868 +       h_tmpdir = h_dir;
5869 +       pin_orig = NULL;
5870 +       if (!h_dir->i_nlink) {
5871 +               wbr = au_sbr(dentry->d_sb, bdst)->br_wbr;
5872 +               h_orph = wbr->wbr_orph;
5873 +
5874 +               h_parent = dget(au_h_dptr(parent, bdst));
5875 +               au_set_h_dptr(parent, bdst, dget(h_orph));
5876 +               h_tmpdir = d_inode(h_orph);
5877 +               au_set_h_iptr(dir, bdst, au_igrab(h_tmpdir), /*flags*/0);
5878 +
5879 +               inode_lock_nested(h_tmpdir, AuLsc_I_PARENT3);
5880 +               /* todo: au_h_open_pre()? */
5881 +
5882 +               pin_orig = cpg->pin;
5883 +               au_pin_init(&wh_pin, dentry, bdst, AuLsc_DI_PARENT,
5884 +                           AuLsc_I_PARENT3, cpg->pin->udba, AuPin_DI_LOCKED);
5885 +               cpg->pin = &wh_pin;
5886 +       }
5887 +
5888 +       if (!au_test_h_perm_sio(h_tmpdir, MAY_EXEC | MAY_WRITE)
5889 +           && !au_cpup_sio_test(cpg->pin, d_inode(dentry)->i_mode))
5890 +               err = au_cpup_wh(cpg, file);
5891 +       else {
5892 +               struct au_cpup_wh_args args = {
5893 +                       .errp   = &err,
5894 +                       .cpg    = cpg,
5895 +                       .file   = file
5896 +               };
5897 +               wkq_err = au_wkq_wait(au_call_cpup_wh, &args);
5898 +               if (unlikely(wkq_err))
5899 +                       err = wkq_err;
5900 +       }
5901 +
5902 +       if (h_orph) {
5903 +               inode_unlock(h_tmpdir);
5904 +               /* todo: au_h_open_post()? */
5905 +               au_set_h_iptr(dir, bdst, au_igrab(h_dir), /*flags*/0);
5906 +               au_set_h_dptr(parent, bdst, h_parent);
5907 +               AuDebugOn(!pin_orig);
5908 +               cpg->pin = pin_orig;
5909 +       }
5910 +       iput(h_dir);
5911 +       dput(parent);
5912 +
5913 +       return err;
5914 +}
5915 +
5916 +/* ---------------------------------------------------------------------- */
5917 +
5918 +/*
5919 + * generic routine for both of copy-up and copy-down.
5920 + */
5921 +/* cf. revalidate function in file.c */
5922 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
5923 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
5924 +                        struct au_pin *pin,
5925 +                        struct dentry *h_parent, void *arg),
5926 +              void *arg)
5927 +{
5928 +       int err;
5929 +       struct au_pin pin;
5930 +       struct dentry *d, *parent, *h_parent, *real_parent, *h_dentry;
5931 +
5932 +       err = 0;
5933 +       parent = dget_parent(dentry);
5934 +       if (IS_ROOT(parent))
5935 +               goto out;
5936 +
5937 +       au_pin_init(&pin, dentry, bdst, AuLsc_DI_PARENT2, AuLsc_I_PARENT2,
5938 +                   au_opt_udba(dentry->d_sb), AuPin_MNT_WRITE);
5939 +
5940 +       /* do not use au_dpage */
5941 +       real_parent = parent;
5942 +       while (1) {
5943 +               dput(parent);
5944 +               parent = dget_parent(dentry);
5945 +               h_parent = au_h_dptr(parent, bdst);
5946 +               if (h_parent)
5947 +                       goto out; /* success */
5948 +
5949 +               /* find top dir which is necessary to cpup */
5950 +               do {
5951 +                       d = parent;
5952 +                       dput(parent);
5953 +                       parent = dget_parent(d);
5954 +                       di_read_lock_parent3(parent, !AuLock_IR);
5955 +                       h_parent = au_h_dptr(parent, bdst);
5956 +                       di_read_unlock(parent, !AuLock_IR);
5957 +               } while (!h_parent);
5958 +
5959 +               if (d != real_parent)
5960 +                       di_write_lock_child3(d);
5961 +
5962 +               /* somebody else might create while we were sleeping */
5963 +               h_dentry = au_h_dptr(d, bdst);
5964 +               if (!h_dentry || d_is_negative(h_dentry)) {
5965 +                       if (h_dentry)
5966 +                               au_update_dbtop(d);
5967 +
5968 +                       au_pin_set_dentry(&pin, d);
5969 +                       err = au_do_pin(&pin);
5970 +                       if (!err) {
5971 +                               err = cp(d, bdst, &pin, h_parent, arg);
5972 +                               au_unpin(&pin);
5973 +                       }
5974 +               }
5975 +
5976 +               if (d != real_parent)
5977 +                       di_write_unlock(d);
5978 +               if (unlikely(err))
5979 +                       break;
5980 +       }
5981 +
5982 +out:
5983 +       dput(parent);
5984 +       return err;
5985 +}
5986 +
5987 +static int au_cpup_dir(struct dentry *dentry, aufs_bindex_t bdst,
5988 +                      struct au_pin *pin,
5989 +                      struct dentry *h_parent __maybe_unused,
5990 +                      void *arg __maybe_unused)
5991 +{
5992 +       struct au_cp_generic cpg = {
5993 +               .dentry = dentry,
5994 +               .bdst   = bdst,
5995 +               .bsrc   = -1,
5996 +               .len    = 0,
5997 +               .pin    = pin,
5998 +               .flags  = AuCpup_DTIME
5999 +       };
6000 +       return au_sio_cpup_simple(&cpg);
6001 +}
6002 +
6003 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
6004 +{
6005 +       return au_cp_dirs(dentry, bdst, au_cpup_dir, NULL);
6006 +}
6007 +
6008 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
6009 +{
6010 +       int err;
6011 +       struct dentry *parent;
6012 +       struct inode *dir;
6013 +
6014 +       parent = dget_parent(dentry);
6015 +       dir = d_inode(parent);
6016 +       err = 0;
6017 +       if (au_h_iptr(dir, bdst))
6018 +               goto out;
6019 +
6020 +       di_read_unlock(parent, AuLock_IR);
6021 +       di_write_lock_parent(parent);
6022 +       /* someone else might change our inode while we were sleeping */
6023 +       if (!au_h_iptr(dir, bdst))
6024 +               err = au_cpup_dirs(dentry, bdst);
6025 +       di_downgrade_lock(parent, AuLock_IR);
6026 +
6027 +out:
6028 +       dput(parent);
6029 +       return err;
6030 +}
6031 diff -urN /usr/share/empty/fs/aufs/cpup.h linux/fs/aufs/cpup.h
6032 --- /usr/share/empty/fs/aufs/cpup.h     1970-01-01 01:00:00.000000000 +0100
6033 +++ linux/fs/aufs/cpup.h        2016-10-09 16:55:36.486034798 +0200
6034 @@ -0,0 +1,94 @@
6035 +/*
6036 + * Copyright (C) 2005-2016 Junjiro R. Okajima
6037 + *
6038 + * This program, aufs is free software; you can redistribute it and/or modify
6039 + * it under the terms of the GNU General Public License as published by
6040 + * the Free Software Foundation; either version 2 of the License, or
6041 + * (at your option) any later version.
6042 + *
6043 + * This program is distributed in the hope that it will be useful,
6044 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6045 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6046 + * GNU General Public License for more details.
6047 + *
6048 + * You should have received a copy of the GNU General Public License
6049 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6050 + */
6051 +
6052 +/*
6053 + * copy-up/down functions
6054 + */
6055 +
6056 +#ifndef __AUFS_CPUP_H__
6057 +#define __AUFS_CPUP_H__
6058 +
6059 +#ifdef __KERNEL__
6060 +
6061 +#include <linux/path.h>
6062 +
6063 +struct inode;
6064 +struct file;
6065 +struct au_pin;
6066 +
6067 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags);
6068 +void au_cpup_attr_timesizes(struct inode *inode);
6069 +void au_cpup_attr_nlink(struct inode *inode, int force);
6070 +void au_cpup_attr_changeable(struct inode *inode);
6071 +void au_cpup_igen(struct inode *inode, struct inode *h_inode);
6072 +void au_cpup_attr_all(struct inode *inode, int force);
6073 +
6074 +/* ---------------------------------------------------------------------- */
6075 +
6076 +struct au_cp_generic {
6077 +       struct dentry   *dentry;
6078 +       aufs_bindex_t   bdst, bsrc;
6079 +       loff_t          len;
6080 +       struct au_pin   *pin;
6081 +       unsigned int    flags;
6082 +};
6083 +
6084 +/* cpup flags */
6085 +#define AuCpup_DTIME           1               /* do dtime_store/revert */
6086 +#define AuCpup_KEEPLINO                (1 << 1)        /* do not clear the lower xino,
6087 +                                                  for link(2) */
6088 +#define AuCpup_RENAME          (1 << 2)        /* rename after cpup */
6089 +#define AuCpup_HOPEN           (1 << 3)        /* call h_open_pre/post() in
6090 +                                                  cpup */
6091 +#define AuCpup_OVERWRITE       (1 << 4)        /* allow overwriting the
6092 +                                                  existing entry */
6093 +#define AuCpup_RWDST           (1 << 5)        /* force write target even if
6094 +                                                  the branch is marked as RO */
6095 +
6096 +#define au_ftest_cpup(flags, name)     ((flags) & AuCpup_##name)
6097 +#define au_fset_cpup(flags, name) \
6098 +       do { (flags) |= AuCpup_##name; } while (0)
6099 +#define au_fclr_cpup(flags, name) \
6100 +       do { (flags) &= ~AuCpup_##name; } while (0)
6101 +
6102 +int au_copy_file(struct file *dst, struct file *src, loff_t len);
6103 +int au_sio_cpup_simple(struct au_cp_generic *cpg);
6104 +int au_sio_cpdown_simple(struct au_cp_generic *cpg);
6105 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file);
6106 +
6107 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
6108 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
6109 +                        struct au_pin *pin,
6110 +                        struct dentry *h_parent, void *arg),
6111 +              void *arg);
6112 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
6113 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
6114 +
6115 +/* ---------------------------------------------------------------------- */
6116 +
6117 +/* keep timestamps when copyup */
6118 +struct au_dtime {
6119 +       struct dentry *dt_dentry;
6120 +       struct path dt_h_path;
6121 +       struct timespec dt_atime, dt_mtime;
6122 +};
6123 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
6124 +                   struct path *h_path);
6125 +void au_dtime_revert(struct au_dtime *dt);
6126 +
6127 +#endif /* __KERNEL__ */
6128 +#endif /* __AUFS_CPUP_H__ */
6129 diff -urN /usr/share/empty/fs/aufs/dbgaufs.c linux/fs/aufs/dbgaufs.c
6130 --- /usr/share/empty/fs/aufs/dbgaufs.c  1970-01-01 01:00:00.000000000 +0100
6131 +++ linux/fs/aufs/dbgaufs.c     2016-10-09 16:55:38.886097714 +0200
6132 @@ -0,0 +1,438 @@
6133 +/*
6134 + * Copyright (C) 2005-2016 Junjiro R. Okajima
6135 + *
6136 + * This program, aufs is free software; you can redistribute it and/or modify
6137 + * it under the terms of the GNU General Public License as published by
6138 + * the Free Software Foundation; either version 2 of the License, or
6139 + * (at your option) any later version.
6140 + *
6141 + * This program is distributed in the hope that it will be useful,
6142 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6143 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6144 + * GNU General Public License for more details.
6145 + *
6146 + * You should have received a copy of the GNU General Public License
6147 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6148 + */
6149 +
6150 +/*
6151 + * debugfs interface
6152 + */
6153 +
6154 +#include <linux/debugfs.h>
6155 +#include "aufs.h"
6156 +
6157 +#ifndef CONFIG_SYSFS
6158 +#error DEBUG_FS depends upon SYSFS
6159 +#endif
6160 +
6161 +static struct dentry *dbgaufs;
6162 +static const mode_t dbgaufs_mode = S_IRUSR | S_IRGRP | S_IROTH;
6163 +
6164 +/* 20 is max digits length of ulong 64 */
6165 +struct dbgaufs_arg {
6166 +       int n;
6167 +       char a[20 * 4];
6168 +};
6169 +
6170 +/*
6171 + * common function for all XINO files
6172 + */
6173 +static int dbgaufs_xi_release(struct inode *inode __maybe_unused,
6174 +                             struct file *file)
6175 +{
6176 +       au_delayed_kfree(file->private_data);
6177 +       return 0;
6178 +}
6179 +
6180 +static int dbgaufs_xi_open(struct file *xf, struct file *file, int do_fcnt)
6181 +{
6182 +       int err;
6183 +       struct kstat st;
6184 +       struct dbgaufs_arg *p;
6185 +
6186 +       err = -ENOMEM;
6187 +       p = kmalloc(sizeof(*p), GFP_NOFS);
6188 +       if (unlikely(!p))
6189 +               goto out;
6190 +
6191 +       err = 0;
6192 +       p->n = 0;
6193 +       file->private_data = p;
6194 +       if (!xf)
6195 +               goto out;
6196 +
6197 +       err = vfs_getattr(&xf->f_path, &st);
6198 +       if (!err) {
6199 +               if (do_fcnt)
6200 +                       p->n = snprintf
6201 +                               (p->a, sizeof(p->a), "%ld, %llux%lu %lld\n",
6202 +                                (long)file_count(xf), st.blocks, st.blksize,
6203 +                                (long long)st.size);
6204 +               else
6205 +                       p->n = snprintf(p->a, sizeof(p->a), "%llux%lu %lld\n",
6206 +                                       st.blocks, st.blksize,
6207 +                                       (long long)st.size);
6208 +               AuDebugOn(p->n >= sizeof(p->a));
6209 +       } else {
6210 +               p->n = snprintf(p->a, sizeof(p->a), "err %d\n", err);
6211 +               err = 0;
6212 +       }
6213 +
6214 +out:
6215 +       return err;
6216 +
6217 +}
6218 +
6219 +static ssize_t dbgaufs_xi_read(struct file *file, char __user *buf,
6220 +                              size_t count, loff_t *ppos)
6221 +{
6222 +       struct dbgaufs_arg *p;
6223 +
6224 +       p = file->private_data;
6225 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
6226 +}
6227 +
6228 +/* ---------------------------------------------------------------------- */
6229 +
6230 +struct dbgaufs_plink_arg {
6231 +       int n;
6232 +       char a[];
6233 +};
6234 +
6235 +static int dbgaufs_plink_release(struct inode *inode __maybe_unused,
6236 +                                struct file *file)
6237 +{
6238 +       au_delayed_free_page((unsigned long)file->private_data);
6239 +       return 0;
6240 +}
6241 +
6242 +static int dbgaufs_plink_open(struct inode *inode, struct file *file)
6243 +{
6244 +       int err, i, limit;
6245 +       unsigned long n, sum;
6246 +       struct dbgaufs_plink_arg *p;
6247 +       struct au_sbinfo *sbinfo;
6248 +       struct super_block *sb;
6249 +       struct au_sphlhead *sphl;
6250 +
6251 +       err = -ENOMEM;
6252 +       p = (void *)get_zeroed_page(GFP_NOFS);
6253 +       if (unlikely(!p))
6254 +               goto out;
6255 +
6256 +       err = -EFBIG;
6257 +       sbinfo = inode->i_private;
6258 +       sb = sbinfo->si_sb;
6259 +       si_noflush_read_lock(sb);
6260 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
6261 +               limit = PAGE_SIZE - sizeof(p->n);
6262 +
6263 +               /* the number of buckets */
6264 +               n = snprintf(p->a + p->n, limit, "%d\n", AuPlink_NHASH);
6265 +               p->n += n;
6266 +               limit -= n;
6267 +
6268 +               sum = 0;
6269 +               for (i = 0, sphl = sbinfo->si_plink;
6270 +                    i < AuPlink_NHASH;
6271 +                    i++, sphl++) {
6272 +                       n = au_sphl_count(sphl);
6273 +                       sum += n;
6274 +
6275 +                       n = snprintf(p->a + p->n, limit, "%lu ", n);
6276 +                       p->n += n;
6277 +                       limit -= n;
6278 +                       if (unlikely(limit <= 0))
6279 +                               goto out_free;
6280 +               }
6281 +               p->a[p->n - 1] = '\n';
6282 +
6283 +               /* the sum of plinks */
6284 +               n = snprintf(p->a + p->n, limit, "%lu\n", sum);
6285 +               p->n += n;
6286 +               limit -= n;
6287 +               if (unlikely(limit <= 0))
6288 +                       goto out_free;
6289 +       } else {
6290 +#define str "1\n0\n0\n"
6291 +               p->n = sizeof(str) - 1;
6292 +               strcpy(p->a, str);
6293 +#undef str
6294 +       }
6295 +       si_read_unlock(sb);
6296 +
6297 +       err = 0;
6298 +       file->private_data = p;
6299 +       goto out; /* success */
6300 +
6301 +out_free:
6302 +       au_delayed_free_page((unsigned long)p);
6303 +out:
6304 +       return err;
6305 +}
6306 +
6307 +static ssize_t dbgaufs_plink_read(struct file *file, char __user *buf,
6308 +                                 size_t count, loff_t *ppos)
6309 +{
6310 +       struct dbgaufs_plink_arg *p;
6311 +
6312 +       p = file->private_data;
6313 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
6314 +}
6315 +
6316 +static const struct file_operations dbgaufs_plink_fop = {
6317 +       .owner          = THIS_MODULE,
6318 +       .open           = dbgaufs_plink_open,
6319 +       .release        = dbgaufs_plink_release,
6320 +       .read           = dbgaufs_plink_read
6321 +};
6322 +
6323 +/* ---------------------------------------------------------------------- */
6324 +
6325 +static int dbgaufs_xib_open(struct inode *inode, struct file *file)
6326 +{
6327 +       int err;
6328 +       struct au_sbinfo *sbinfo;
6329 +       struct super_block *sb;
6330 +
6331 +       sbinfo = inode->i_private;
6332 +       sb = sbinfo->si_sb;
6333 +       si_noflush_read_lock(sb);
6334 +       err = dbgaufs_xi_open(sbinfo->si_xib, file, /*do_fcnt*/0);
6335 +       si_read_unlock(sb);
6336 +       return err;
6337 +}
6338 +
6339 +static const struct file_operations dbgaufs_xib_fop = {
6340 +       .owner          = THIS_MODULE,
6341 +       .open           = dbgaufs_xib_open,
6342 +       .release        = dbgaufs_xi_release,
6343 +       .read           = dbgaufs_xi_read
6344 +};
6345 +
6346 +/* ---------------------------------------------------------------------- */
6347 +
6348 +#define DbgaufsXi_PREFIX "xi"
6349 +
6350 +static int dbgaufs_xino_open(struct inode *inode, struct file *file)
6351 +{
6352 +       int err;
6353 +       long l;
6354 +       struct au_sbinfo *sbinfo;
6355 +       struct super_block *sb;
6356 +       struct file *xf;
6357 +       struct qstr *name;
6358 +
6359 +       err = -ENOENT;
6360 +       xf = NULL;
6361 +       name = &file->f_path.dentry->d_name;
6362 +       if (unlikely(name->len < sizeof(DbgaufsXi_PREFIX)
6363 +                    || memcmp(name->name, DbgaufsXi_PREFIX,
6364 +                              sizeof(DbgaufsXi_PREFIX) - 1)))
6365 +               goto out;
6366 +       err = kstrtol(name->name + sizeof(DbgaufsXi_PREFIX) - 1, 10, &l);
6367 +       if (unlikely(err))
6368 +               goto out;
6369 +
6370 +       sbinfo = inode->i_private;
6371 +       sb = sbinfo->si_sb;
6372 +       si_noflush_read_lock(sb);
6373 +       if (l <= au_sbbot(sb)) {
6374 +               xf = au_sbr(sb, (aufs_bindex_t)l)->br_xino.xi_file;
6375 +               err = dbgaufs_xi_open(xf, file, /*do_fcnt*/1);
6376 +       } else
6377 +               err = -ENOENT;
6378 +       si_read_unlock(sb);
6379 +
6380 +out:
6381 +       return err;
6382 +}
6383 +
6384 +static const struct file_operations dbgaufs_xino_fop = {
6385 +       .owner          = THIS_MODULE,
6386 +       .open           = dbgaufs_xino_open,
6387 +       .release        = dbgaufs_xi_release,
6388 +       .read           = dbgaufs_xi_read
6389 +};
6390 +
6391 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
6392 +{
6393 +       aufs_bindex_t bbot;
6394 +       struct au_branch *br;
6395 +       struct au_xino_file *xi;
6396 +
6397 +       if (!au_sbi(sb)->si_dbgaufs)
6398 +               return;
6399 +
6400 +       bbot = au_sbbot(sb);
6401 +       for (; bindex <= bbot; bindex++) {
6402 +               br = au_sbr(sb, bindex);
6403 +               xi = &br->br_xino;
6404 +               /* debugfs acquires the parent i_mutex */
6405 +               lockdep_off();
6406 +               debugfs_remove(xi->xi_dbgaufs);
6407 +               lockdep_on();
6408 +               xi->xi_dbgaufs = NULL;
6409 +       }
6410 +}
6411 +
6412 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
6413 +{
6414 +       struct au_sbinfo *sbinfo;
6415 +       struct dentry *parent;
6416 +       struct au_branch *br;
6417 +       struct au_xino_file *xi;
6418 +       aufs_bindex_t bbot;
6419 +       char name[sizeof(DbgaufsXi_PREFIX) + 5]; /* "xi" bindex NULL */
6420 +
6421 +       sbinfo = au_sbi(sb);
6422 +       parent = sbinfo->si_dbgaufs;
6423 +       if (!parent)
6424 +               return;
6425 +
6426 +       bbot = au_sbbot(sb);
6427 +       for (; bindex <= bbot; bindex++) {
6428 +               snprintf(name, sizeof(name), DbgaufsXi_PREFIX "%d", bindex);
6429 +               br = au_sbr(sb, bindex);
6430 +               xi = &br->br_xino;
6431 +               AuDebugOn(xi->xi_dbgaufs);
6432 +               /* debugfs acquires the parent i_mutex */
6433 +               lockdep_off();
6434 +               xi->xi_dbgaufs = debugfs_create_file(name, dbgaufs_mode, parent,
6435 +                                                    sbinfo, &dbgaufs_xino_fop);
6436 +               lockdep_on();
6437 +               /* ignore an error */
6438 +               if (unlikely(!xi->xi_dbgaufs))
6439 +                       AuWarn1("failed %s under debugfs\n", name);
6440 +       }
6441 +}
6442 +
6443 +/* ---------------------------------------------------------------------- */
6444 +
6445 +#ifdef CONFIG_AUFS_EXPORT
6446 +static int dbgaufs_xigen_open(struct inode *inode, struct file *file)
6447 +{
6448 +       int err;
6449 +       struct au_sbinfo *sbinfo;
6450 +       struct super_block *sb;
6451 +
6452 +       sbinfo = inode->i_private;
6453 +       sb = sbinfo->si_sb;
6454 +       si_noflush_read_lock(sb);
6455 +       err = dbgaufs_xi_open(sbinfo->si_xigen, file, /*do_fcnt*/0);
6456 +       si_read_unlock(sb);
6457 +       return err;
6458 +}
6459 +
6460 +static const struct file_operations dbgaufs_xigen_fop = {
6461 +       .owner          = THIS_MODULE,
6462 +       .open           = dbgaufs_xigen_open,
6463 +       .release        = dbgaufs_xi_release,
6464 +       .read           = dbgaufs_xi_read
6465 +};
6466 +
6467 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6468 +{
6469 +       int err;
6470 +
6471 +       /*
6472 +        * This function is a dynamic '__init' function actually,
6473 +        * so the tiny check for si_rwsem is unnecessary.
6474 +        */
6475 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6476 +
6477 +       err = -EIO;
6478 +       sbinfo->si_dbgaufs_xigen = debugfs_create_file
6479 +               ("xigen", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6480 +                &dbgaufs_xigen_fop);
6481 +       if (sbinfo->si_dbgaufs_xigen)
6482 +               err = 0;
6483 +
6484 +       return err;
6485 +}
6486 +#else
6487 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6488 +{
6489 +       return 0;
6490 +}
6491 +#endif /* CONFIG_AUFS_EXPORT */
6492 +
6493 +/* ---------------------------------------------------------------------- */
6494 +
6495 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo)
6496 +{
6497 +       /*
6498 +        * This function is a dynamic '__fin' function actually,
6499 +        * so the tiny check for si_rwsem is unnecessary.
6500 +        */
6501 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6502 +
6503 +       debugfs_remove_recursive(sbinfo->si_dbgaufs);
6504 +       sbinfo->si_dbgaufs = NULL;
6505 +       kobject_put(&sbinfo->si_kobj);
6506 +}
6507 +
6508 +int dbgaufs_si_init(struct au_sbinfo *sbinfo)
6509 +{
6510 +       int err;
6511 +       char name[SysaufsSiNameLen];
6512 +
6513 +       /*
6514 +        * This function is a dynamic '__init' function actually,
6515 +        * so the tiny check for si_rwsem is unnecessary.
6516 +        */
6517 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6518 +
6519 +       err = -ENOENT;
6520 +       if (!dbgaufs) {
6521 +               AuErr1("/debug/aufs is uninitialized\n");
6522 +               goto out;
6523 +       }
6524 +
6525 +       err = -EIO;
6526 +       sysaufs_name(sbinfo, name);
6527 +       sbinfo->si_dbgaufs = debugfs_create_dir(name, dbgaufs);
6528 +       if (unlikely(!sbinfo->si_dbgaufs))
6529 +               goto out;
6530 +       kobject_get(&sbinfo->si_kobj);
6531 +
6532 +       sbinfo->si_dbgaufs_xib = debugfs_create_file
6533 +               ("xib", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6534 +                &dbgaufs_xib_fop);
6535 +       if (unlikely(!sbinfo->si_dbgaufs_xib))
6536 +               goto out_dir;
6537 +
6538 +       sbinfo->si_dbgaufs_plink = debugfs_create_file
6539 +               ("plink", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6540 +                &dbgaufs_plink_fop);
6541 +       if (unlikely(!sbinfo->si_dbgaufs_plink))
6542 +               goto out_dir;
6543 +
6544 +       err = dbgaufs_xigen_init(sbinfo);
6545 +       if (!err)
6546 +               goto out; /* success */
6547 +
6548 +out_dir:
6549 +       dbgaufs_si_fin(sbinfo);
6550 +out:
6551 +       return err;
6552 +}
6553 +
6554 +/* ---------------------------------------------------------------------- */
6555 +
6556 +void dbgaufs_fin(void)
6557 +{
6558 +       debugfs_remove(dbgaufs);
6559 +}
6560 +
6561 +int __init dbgaufs_init(void)
6562 +{
6563 +       int err;
6564 +
6565 +       err = -EIO;
6566 +       dbgaufs = debugfs_create_dir(AUFS_NAME, NULL);
6567 +       if (dbgaufs)
6568 +               err = 0;
6569 +       return err;
6570 +}
6571 diff -urN /usr/share/empty/fs/aufs/dbgaufs.h linux/fs/aufs/dbgaufs.h
6572 --- /usr/share/empty/fs/aufs/dbgaufs.h  1970-01-01 01:00:00.000000000 +0100
6573 +++ linux/fs/aufs/dbgaufs.h     2016-10-09 16:55:36.486034798 +0200
6574 @@ -0,0 +1,48 @@
6575 +/*
6576 + * Copyright (C) 2005-2016 Junjiro R. Okajima
6577 + *
6578 + * This program, aufs is free software; you can redistribute it and/or modify
6579 + * it under the terms of the GNU General Public License as published by
6580 + * the Free Software Foundation; either version 2 of the License, or
6581 + * (at your option) any later version.
6582 + *
6583 + * This program is distributed in the hope that it will be useful,
6584 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6585 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6586 + * GNU General Public License for more details.
6587 + *
6588 + * You should have received a copy of the GNU General Public License
6589 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6590 + */
6591 +
6592 +/*
6593 + * debugfs interface
6594 + */
6595 +
6596 +#ifndef __DBGAUFS_H__
6597 +#define __DBGAUFS_H__
6598 +
6599 +#ifdef __KERNEL__
6600 +
6601 +struct super_block;
6602 +struct au_sbinfo;
6603 +
6604 +#ifdef CONFIG_DEBUG_FS
6605 +/* dbgaufs.c */
6606 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
6607 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
6608 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo);
6609 +int dbgaufs_si_init(struct au_sbinfo *sbinfo);
6610 +void dbgaufs_fin(void);
6611 +int __init dbgaufs_init(void);
6612 +#else
6613 +AuStubVoid(dbgaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
6614 +AuStubVoid(dbgaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
6615 +AuStubVoid(dbgaufs_si_fin, struct au_sbinfo *sbinfo)
6616 +AuStubInt0(dbgaufs_si_init, struct au_sbinfo *sbinfo)
6617 +AuStubVoid(dbgaufs_fin, void)
6618 +AuStubInt0(__init dbgaufs_init, void)
6619 +#endif /* CONFIG_DEBUG_FS */
6620 +
6621 +#endif /* __KERNEL__ */
6622 +#endif /* __DBGAUFS_H__ */
6623 diff -urN /usr/share/empty/fs/aufs/dcsub.c linux/fs/aufs/dcsub.c
6624 --- /usr/share/empty/fs/aufs/dcsub.c    1970-01-01 01:00:00.000000000 +0100
6625 +++ linux/fs/aufs/dcsub.c       2016-10-09 16:55:38.886097714 +0200
6626 @@ -0,0 +1,225 @@
6627 +/*
6628 + * Copyright (C) 2005-2016 Junjiro R. Okajima
6629 + *
6630 + * This program, aufs is free software; you can redistribute it and/or modify
6631 + * it under the terms of the GNU General Public License as published by
6632 + * the Free Software Foundation; either version 2 of the License, or
6633 + * (at your option) any later version.
6634 + *
6635 + * This program is distributed in the hope that it will be useful,
6636 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6637 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6638 + * GNU General Public License for more details.
6639 + *
6640 + * You should have received a copy of the GNU General Public License
6641 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6642 + */
6643 +
6644 +/*
6645 + * sub-routines for dentry cache
6646 + */
6647 +
6648 +#include "aufs.h"
6649 +
6650 +static void au_dpage_free(struct au_dpage *dpage)
6651 +{
6652 +       int i;
6653 +       struct dentry **p;
6654 +
6655 +       p = dpage->dentries;
6656 +       for (i = 0; i < dpage->ndentry; i++)
6657 +               dput(*p++);
6658 +       au_delayed_free_page((unsigned long)dpage->dentries);
6659 +}
6660 +
6661 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp)
6662 +{
6663 +       int err;
6664 +       void *p;
6665 +
6666 +       err = -ENOMEM;
6667 +       dpages->dpages = kmalloc(sizeof(*dpages->dpages), gfp);
6668 +       if (unlikely(!dpages->dpages))
6669 +               goto out;
6670 +
6671 +       p = (void *)__get_free_page(gfp);
6672 +       if (unlikely(!p))
6673 +               goto out_dpages;
6674 +
6675 +       dpages->dpages[0].ndentry = 0;
6676 +       dpages->dpages[0].dentries = p;
6677 +       dpages->ndpage = 1;
6678 +       return 0; /* success */
6679 +
6680 +out_dpages:
6681 +       au_delayed_kfree(dpages->dpages);
6682 +out:
6683 +       return err;
6684 +}
6685 +
6686 +void au_dpages_free(struct au_dcsub_pages *dpages)
6687 +{
6688 +       int i;
6689 +       struct au_dpage *p;
6690 +
6691 +       p = dpages->dpages;
6692 +       for (i = 0; i < dpages->ndpage; i++)
6693 +               au_dpage_free(p++);
6694 +       au_delayed_kfree(dpages->dpages);
6695 +}
6696 +
6697 +static int au_dpages_append(struct au_dcsub_pages *dpages,
6698 +                           struct dentry *dentry, gfp_t gfp)
6699 +{
6700 +       int err, sz;
6701 +       struct au_dpage *dpage;
6702 +       void *p;
6703 +
6704 +       dpage = dpages->dpages + dpages->ndpage - 1;
6705 +       sz = PAGE_SIZE / sizeof(dentry);
6706 +       if (unlikely(dpage->ndentry >= sz)) {
6707 +               AuLabel(new dpage);
6708 +               err = -ENOMEM;
6709 +               sz = dpages->ndpage * sizeof(*dpages->dpages);
6710 +               p = au_kzrealloc(dpages->dpages, sz,
6711 +                                sz + sizeof(*dpages->dpages), gfp,
6712 +                                /*may_shrink*/0);
6713 +               if (unlikely(!p))
6714 +                       goto out;
6715 +
6716 +               dpages->dpages = p;
6717 +               dpage = dpages->dpages + dpages->ndpage;
6718 +               p = (void *)__get_free_page(gfp);
6719 +               if (unlikely(!p))
6720 +                       goto out;
6721 +
6722 +               dpage->ndentry = 0;
6723 +               dpage->dentries = p;
6724 +               dpages->ndpage++;
6725 +       }
6726 +
6727 +       AuDebugOn(au_dcount(dentry) <= 0);
6728 +       dpage->dentries[dpage->ndentry++] = dget_dlock(dentry);
6729 +       return 0; /* success */
6730 +
6731 +out:
6732 +       return err;
6733 +}
6734 +
6735 +/* todo: BAD approach */
6736 +/* copied from linux/fs/dcache.c */
6737 +enum d_walk_ret {
6738 +       D_WALK_CONTINUE,
6739 +       D_WALK_QUIT,
6740 +       D_WALK_NORETRY,
6741 +       D_WALK_SKIP,
6742 +};
6743 +
6744 +extern void d_walk(struct dentry *parent, void *data,
6745 +                  enum d_walk_ret (*enter)(void *, struct dentry *),
6746 +                  void (*finish)(void *));
6747 +
6748 +struct ac_dpages_arg {
6749 +       int err;
6750 +       struct au_dcsub_pages *dpages;
6751 +       struct super_block *sb;
6752 +       au_dpages_test test;
6753 +       void *arg;
6754 +};
6755 +
6756 +static enum d_walk_ret au_call_dpages_append(void *_arg, struct dentry *dentry)
6757 +{
6758 +       enum d_walk_ret ret;
6759 +       struct ac_dpages_arg *arg = _arg;
6760 +
6761 +       ret = D_WALK_CONTINUE;
6762 +       if (dentry->d_sb == arg->sb
6763 +           && !IS_ROOT(dentry)
6764 +           && au_dcount(dentry) > 0
6765 +           && au_di(dentry)
6766 +           && (!arg->test || arg->test(dentry, arg->arg))) {
6767 +               arg->err = au_dpages_append(arg->dpages, dentry, GFP_ATOMIC);
6768 +               if (unlikely(arg->err))
6769 +                       ret = D_WALK_QUIT;
6770 +       }
6771 +
6772 +       return ret;
6773 +}
6774 +
6775 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6776 +                  au_dpages_test test, void *arg)
6777 +{
6778 +       struct ac_dpages_arg args = {
6779 +               .err    = 0,
6780 +               .dpages = dpages,
6781 +               .sb     = root->d_sb,
6782 +               .test   = test,
6783 +               .arg    = arg
6784 +       };
6785 +
6786 +       d_walk(root, &args, au_call_dpages_append, NULL);
6787 +
6788 +       return args.err;
6789 +}
6790 +
6791 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6792 +                      int do_include, au_dpages_test test, void *arg)
6793 +{
6794 +       int err;
6795 +
6796 +       err = 0;
6797 +       write_seqlock(&rename_lock);
6798 +       spin_lock(&dentry->d_lock);
6799 +       if (do_include
6800 +           && au_dcount(dentry) > 0
6801 +           && (!test || test(dentry, arg)))
6802 +               err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6803 +       spin_unlock(&dentry->d_lock);
6804 +       if (unlikely(err))
6805 +               goto out;
6806 +
6807 +       /*
6808 +        * RCU for vfsmount is unnecessary since this is a traverse in a single
6809 +        * mount
6810 +        */
6811 +       while (!IS_ROOT(dentry)) {
6812 +               dentry = dentry->d_parent; /* rename_lock is locked */
6813 +               spin_lock(&dentry->d_lock);
6814 +               if (au_dcount(dentry) > 0
6815 +                   && (!test || test(dentry, arg)))
6816 +                       err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6817 +               spin_unlock(&dentry->d_lock);
6818 +               if (unlikely(err))
6819 +                       break;
6820 +       }
6821 +
6822 +out:
6823 +       write_sequnlock(&rename_lock);
6824 +       return err;
6825 +}
6826 +
6827 +static inline int au_dcsub_dpages_aufs(struct dentry *dentry, void *arg)
6828 +{
6829 +       return au_di(dentry) && dentry->d_sb == arg;
6830 +}
6831 +
6832 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6833 +                           struct dentry *dentry, int do_include)
6834 +{
6835 +       return au_dcsub_pages_rev(dpages, dentry, do_include,
6836 +                                 au_dcsub_dpages_aufs, dentry->d_sb);
6837 +}
6838 +
6839 +int au_test_subdir(struct dentry *d1, struct dentry *d2)
6840 +{
6841 +       struct path path[2] = {
6842 +               {
6843 +                       .dentry = d1
6844 +               },
6845 +               {
6846 +                       .dentry = d2
6847 +               }
6848 +       };
6849 +
6850 +       return path_is_under(path + 0, path + 1);
6851 +}
6852 diff -urN /usr/share/empty/fs/aufs/dcsub.h linux/fs/aufs/dcsub.h
6853 --- /usr/share/empty/fs/aufs/dcsub.h    1970-01-01 01:00:00.000000000 +0100
6854 +++ linux/fs/aufs/dcsub.h       2016-10-09 16:55:36.486034798 +0200
6855 @@ -0,0 +1,136 @@
6856 +/*
6857 + * Copyright (C) 2005-2016 Junjiro R. Okajima
6858 + *
6859 + * This program, aufs is free software; you can redistribute it and/or modify
6860 + * it under the terms of the GNU General Public License as published by
6861 + * the Free Software Foundation; either version 2 of the License, or
6862 + * (at your option) any later version.
6863 + *
6864 + * This program is distributed in the hope that it will be useful,
6865 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6866 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6867 + * GNU General Public License for more details.
6868 + *
6869 + * You should have received a copy of the GNU General Public License
6870 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6871 + */
6872 +
6873 +/*
6874 + * sub-routines for dentry cache
6875 + */
6876 +
6877 +#ifndef __AUFS_DCSUB_H__
6878 +#define __AUFS_DCSUB_H__
6879 +
6880 +#ifdef __KERNEL__
6881 +
6882 +#include <linux/dcache.h>
6883 +#include <linux/fs.h>
6884 +
6885 +struct au_dpage {
6886 +       int ndentry;
6887 +       struct dentry **dentries;
6888 +};
6889 +
6890 +struct au_dcsub_pages {
6891 +       int ndpage;
6892 +       struct au_dpage *dpages;
6893 +};
6894 +
6895 +/* ---------------------------------------------------------------------- */
6896 +
6897 +/* dcsub.c */
6898 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp);
6899 +void au_dpages_free(struct au_dcsub_pages *dpages);
6900 +typedef int (*au_dpages_test)(struct dentry *dentry, void *arg);
6901 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6902 +                  au_dpages_test test, void *arg);
6903 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6904 +                      int do_include, au_dpages_test test, void *arg);
6905 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6906 +                           struct dentry *dentry, int do_include);
6907 +int au_test_subdir(struct dentry *d1, struct dentry *d2);
6908 +
6909 +/* ---------------------------------------------------------------------- */
6910 +
6911 +/*
6912 + * todo: in linux-3.13, several similar (but faster) helpers are added to
6913 + * include/linux/dcache.h. Try them (in the future).
6914 + */
6915 +
6916 +static inline int au_d_hashed_positive(struct dentry *d)
6917 +{
6918 +       int err;
6919 +       struct inode *inode = d_inode(d);
6920 +
6921 +       err = 0;
6922 +       if (unlikely(d_unhashed(d)
6923 +                    || d_is_negative(d)
6924 +                    || !inode->i_nlink))
6925 +               err = -ENOENT;
6926 +       return err;
6927 +}
6928 +
6929 +static inline int au_d_linkable(struct dentry *d)
6930 +{
6931 +       int err;
6932 +       struct inode *inode = d_inode(d);
6933 +
6934 +       err = au_d_hashed_positive(d);
6935 +       if (err
6936 +           && d_is_positive(d)
6937 +           && (inode->i_state & I_LINKABLE))
6938 +               err = 0;
6939 +       return err;
6940 +}
6941 +
6942 +static inline int au_d_alive(struct dentry *d)
6943 +{
6944 +       int err;
6945 +       struct inode *inode;
6946 +
6947 +       err = 0;
6948 +       if (!IS_ROOT(d))
6949 +               err = au_d_hashed_positive(d);
6950 +       else {
6951 +               inode = d_inode(d);
6952 +               if (unlikely(d_unlinked(d)
6953 +                            || d_is_negative(d)
6954 +                            || !inode->i_nlink))
6955 +                       err = -ENOENT;
6956 +       }
6957 +       return err;
6958 +}
6959 +
6960 +static inline int au_alive_dir(struct dentry *d)
6961 +{
6962 +       int err;
6963 +
6964 +       err = au_d_alive(d);
6965 +       if (unlikely(err || IS_DEADDIR(d_inode(d))))
6966 +               err = -ENOENT;
6967 +       return err;
6968 +}
6969 +
6970 +static inline int au_qstreq(struct qstr *a, struct qstr *b)
6971 +{
6972 +       return a->len == b->len
6973 +               && !memcmp(a->name, b->name, a->len);
6974 +}
6975 +
6976 +/*
6977 + * by the commit
6978 + * 360f547 2015-01-25 dcache: let the dentry count go down to zero without
6979 + *                     taking d_lock
6980 + * the type of d_lockref.count became int, but the inlined function d_count()
6981 + * still returns unsigned int.
6982 + * I don't know why. Maybe it is for every d_count() users?
6983 + * Anyway au_dcount() lives on.
6984 + */
6985 +static inline int au_dcount(struct dentry *d)
6986 +{
6987 +       return (int)d_count(d);
6988 +}
6989 +
6990 +#endif /* __KERNEL__ */
6991 +#endif /* __AUFS_DCSUB_H__ */
6992 diff -urN /usr/share/empty/fs/aufs/debug.c linux/fs/aufs/debug.c
6993 --- /usr/share/empty/fs/aufs/debug.c    1970-01-01 01:00:00.000000000 +0100
6994 +++ linux/fs/aufs/debug.c       2016-10-09 16:55:36.486034798 +0200
6995 @@ -0,0 +1,440 @@
6996 +/*
6997 + * Copyright (C) 2005-2016 Junjiro R. Okajima
6998 + *
6999 + * This program, aufs is free software; you can redistribute it and/or modify
7000 + * it under the terms of the GNU General Public License as published by
7001 + * the Free Software Foundation; either version 2 of the License, or
7002 + * (at your option) any later version.
7003 + *
7004 + * This program is distributed in the hope that it will be useful,
7005 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7006 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7007 + * GNU General Public License for more details.
7008 + *
7009 + * You should have received a copy of the GNU General Public License
7010 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7011 + */
7012 +
7013 +/*
7014 + * debug print functions
7015 + */
7016 +
7017 +#include "aufs.h"
7018 +
7019 +/* Returns 0, or -errno.  arg is in kp->arg. */
7020 +static int param_atomic_t_set(const char *val, const struct kernel_param *kp)
7021 +{
7022 +       int err, n;
7023 +
7024 +       err = kstrtoint(val, 0, &n);
7025 +       if (!err) {
7026 +               if (n > 0)
7027 +                       au_debug_on();
7028 +               else
7029 +                       au_debug_off();
7030 +       }
7031 +       return err;
7032 +}
7033 +
7034 +/* Returns length written or -errno.  Buffer is 4k (ie. be short!) */
7035 +static int param_atomic_t_get(char *buffer, const struct kernel_param *kp)
7036 +{
7037 +       atomic_t *a;
7038 +
7039 +       a = kp->arg;
7040 +       return sprintf(buffer, "%d", atomic_read(a));
7041 +}
7042 +
7043 +static struct kernel_param_ops param_ops_atomic_t = {
7044 +       .set = param_atomic_t_set,
7045 +       .get = param_atomic_t_get
7046 +       /* void (*free)(void *arg) */
7047 +};
7048 +
7049 +atomic_t aufs_debug = ATOMIC_INIT(0);
7050 +MODULE_PARM_DESC(debug, "debug print");
7051 +module_param_named(debug, aufs_debug, atomic_t, S_IRUGO | S_IWUSR | S_IWGRP);
7052 +
7053 +DEFINE_MUTEX(au_dbg_mtx);      /* just to serialize the dbg msgs */
7054 +char *au_plevel = KERN_DEBUG;
7055 +#define dpri(fmt, ...) do {                                    \
7056 +       if ((au_plevel                                          \
7057 +            && strcmp(au_plevel, KERN_DEBUG))                  \
7058 +           || au_debug_test())                                 \
7059 +               printk("%s" fmt, au_plevel, ##__VA_ARGS__);     \
7060 +} while (0)
7061 +
7062 +/* ---------------------------------------------------------------------- */
7063 +
7064 +void au_dpri_whlist(struct au_nhash *whlist)
7065 +{
7066 +       unsigned long ul, n;
7067 +       struct hlist_head *head;
7068 +       struct au_vdir_wh *pos;
7069 +
7070 +       n = whlist->nh_num;
7071 +       head = whlist->nh_head;
7072 +       for (ul = 0; ul < n; ul++) {
7073 +               hlist_for_each_entry(pos, head, wh_hash)
7074 +                       dpri("b%d, %.*s, %d\n",
7075 +                            pos->wh_bindex,
7076 +                            pos->wh_str.len, pos->wh_str.name,
7077 +                            pos->wh_str.len);
7078 +               head++;
7079 +       }
7080 +}
7081 +
7082 +void au_dpri_vdir(struct au_vdir *vdir)
7083 +{
7084 +       unsigned long ul;
7085 +       union au_vdir_deblk_p p;
7086 +       unsigned char *o;
7087 +
7088 +       if (!vdir || IS_ERR(vdir)) {
7089 +               dpri("err %ld\n", PTR_ERR(vdir));
7090 +               return;
7091 +       }
7092 +
7093 +       dpri("deblk %u, nblk %lu, deblk %p, last{%lu, %p}, ver %lu\n",
7094 +            vdir->vd_deblk_sz, vdir->vd_nblk, vdir->vd_deblk,
7095 +            vdir->vd_last.ul, vdir->vd_last.p.deblk, vdir->vd_version);
7096 +       for (ul = 0; ul < vdir->vd_nblk; ul++) {
7097 +               p.deblk = vdir->vd_deblk[ul];
7098 +               o = p.deblk;
7099 +               dpri("[%lu]: %p\n", ul, o);
7100 +       }
7101 +}
7102 +
7103 +static int do_pri_inode(aufs_bindex_t bindex, struct inode *inode, int hn,
7104 +                       struct dentry *wh)
7105 +{
7106 +       char *n = NULL;
7107 +       int l = 0;
7108 +
7109 +       if (!inode || IS_ERR(inode)) {
7110 +               dpri("i%d: err %ld\n", bindex, PTR_ERR(inode));
7111 +               return -1;
7112 +       }
7113 +
7114 +       /* the type of i_blocks depends upon CONFIG_LBDAF */
7115 +       BUILD_BUG_ON(sizeof(inode->i_blocks) != sizeof(unsigned long)
7116 +                    && sizeof(inode->i_blocks) != sizeof(u64));
7117 +       if (wh) {
7118 +               n = (void *)wh->d_name.name;
7119 +               l = wh->d_name.len;
7120 +       }
7121 +
7122 +       dpri("i%d: %p, i%lu, %s, cnt %d, nl %u, 0%o, sz %llu, blk %llu,"
7123 +            " hn %d, ct %lld, np %lu, st 0x%lx, f 0x%x, v %llu, g %x%s%.*s\n",
7124 +            bindex, inode,
7125 +            inode->i_ino, inode->i_sb ? au_sbtype(inode->i_sb) : "??",
7126 +            atomic_read(&inode->i_count), inode->i_nlink, inode->i_mode,
7127 +            i_size_read(inode), (unsigned long long)inode->i_blocks,
7128 +            hn, (long long)timespec_to_ns(&inode->i_ctime) & 0x0ffff,
7129 +            inode->i_mapping ? inode->i_mapping->nrpages : 0,
7130 +            inode->i_state, inode->i_flags, inode->i_version,
7131 +            inode->i_generation,
7132 +            l ? ", wh " : "", l, n);
7133 +       return 0;
7134 +}
7135 +
7136 +void au_dpri_inode(struct inode *inode)
7137 +{
7138 +       struct au_iinfo *iinfo;
7139 +       struct au_hinode *hi;
7140 +       aufs_bindex_t bindex;
7141 +       int err, hn;
7142 +
7143 +       err = do_pri_inode(-1, inode, -1, NULL);
7144 +       if (err || !au_test_aufs(inode->i_sb) || au_is_bad_inode(inode))
7145 +               return;
7146 +
7147 +       iinfo = au_ii(inode);
7148 +       dpri("i-1: btop %d, bbot %d, gen %d\n",
7149 +            iinfo->ii_btop, iinfo->ii_bbot, au_iigen(inode, NULL));
7150 +       if (iinfo->ii_btop < 0)
7151 +               return;
7152 +       hn = 0;
7153 +       for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot; bindex++) {
7154 +               hi = au_hinode(iinfo, bindex);
7155 +               hn = !!au_hn(hi);
7156 +               do_pri_inode(bindex, hi->hi_inode, hn, hi->hi_whdentry);
7157 +       }
7158 +}
7159 +
7160 +void au_dpri_dalias(struct inode *inode)
7161 +{
7162 +       struct dentry *d;
7163 +
7164 +       spin_lock(&inode->i_lock);
7165 +       hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias)
7166 +               au_dpri_dentry(d);
7167 +       spin_unlock(&inode->i_lock);
7168 +}
7169 +
7170 +static int do_pri_dentry(aufs_bindex_t bindex, struct dentry *dentry)
7171 +{
7172 +       struct dentry *wh = NULL;
7173 +       int hn;
7174 +       struct inode *inode;
7175 +       struct au_iinfo *iinfo;
7176 +       struct au_hinode *hi;
7177 +
7178 +       if (!dentry || IS_ERR(dentry)) {
7179 +               dpri("d%d: err %ld\n", bindex, PTR_ERR(dentry));
7180 +               return -1;
7181 +       }
7182 +       /* do not call dget_parent() here */
7183 +       /* note: access d_xxx without d_lock */
7184 +       dpri("d%d: %p, %pd2?, %s, cnt %d, flags 0x%x, %shashed\n",
7185 +            bindex, dentry, dentry,
7186 +            dentry->d_sb ? au_sbtype(dentry->d_sb) : "??",
7187 +            au_dcount(dentry), dentry->d_flags,
7188 +            d_unhashed(dentry) ? "un" : "");
7189 +       hn = -1;
7190 +       inode = NULL;
7191 +       if (d_is_positive(dentry))
7192 +               inode = d_inode(dentry);
7193 +       if (inode
7194 +           && au_test_aufs(dentry->d_sb)
7195 +           && bindex >= 0
7196 +           && !au_is_bad_inode(inode)) {
7197 +               iinfo = au_ii(inode);
7198 +               hi = au_hinode(iinfo, bindex);
7199 +               hn = !!au_hn(hi);
7200 +               wh = hi->hi_whdentry;
7201 +       }
7202 +       do_pri_inode(bindex, inode, hn, wh);
7203 +       return 0;
7204 +}
7205 +
7206 +void au_dpri_dentry(struct dentry *dentry)
7207 +{
7208 +       struct au_dinfo *dinfo;
7209 +       aufs_bindex_t bindex;
7210 +       int err;
7211 +
7212 +       err = do_pri_dentry(-1, dentry);
7213 +       if (err || !au_test_aufs(dentry->d_sb))
7214 +               return;
7215 +
7216 +       dinfo = au_di(dentry);
7217 +       if (!dinfo)
7218 +               return;
7219 +       dpri("d-1: btop %d, bbot %d, bwh %d, bdiropq %d, gen %d, tmp %d\n",
7220 +            dinfo->di_btop, dinfo->di_bbot,
7221 +            dinfo->di_bwh, dinfo->di_bdiropq, au_digen(dentry),
7222 +            dinfo->di_tmpfile);
7223 +       if (dinfo->di_btop < 0)
7224 +               return;
7225 +       for (bindex = dinfo->di_btop; bindex <= dinfo->di_bbot; bindex++)
7226 +               do_pri_dentry(bindex, au_hdentry(dinfo, bindex)->hd_dentry);
7227 +}
7228 +
7229 +static int do_pri_file(aufs_bindex_t bindex, struct file *file)
7230 +{
7231 +       char a[32];
7232 +
7233 +       if (!file || IS_ERR(file)) {
7234 +               dpri("f%d: err %ld\n", bindex, PTR_ERR(file));
7235 +               return -1;
7236 +       }
7237 +       a[0] = 0;
7238 +       if (bindex < 0
7239 +           && !IS_ERR_OR_NULL(file->f_path.dentry)
7240 +           && au_test_aufs(file->f_path.dentry->d_sb)
7241 +           && au_fi(file))
7242 +               snprintf(a, sizeof(a), ", gen %d, mmapped %d",
7243 +                        au_figen(file), atomic_read(&au_fi(file)->fi_mmapped));
7244 +       dpri("f%d: mode 0x%x, flags 0%o, cnt %ld, v %llu, pos %llu%s\n",
7245 +            bindex, file->f_mode, file->f_flags, (long)file_count(file),
7246 +            file->f_version, file->f_pos, a);
7247 +       if (!IS_ERR_OR_NULL(file->f_path.dentry))
7248 +               do_pri_dentry(bindex, file->f_path.dentry);
7249 +       return 0;
7250 +}
7251 +
7252 +void au_dpri_file(struct file *file)
7253 +{
7254 +       struct au_finfo *finfo;
7255 +       struct au_fidir *fidir;
7256 +       struct au_hfile *hfile;
7257 +       aufs_bindex_t bindex;
7258 +       int err;
7259 +
7260 +       err = do_pri_file(-1, file);
7261 +       if (err
7262 +           || IS_ERR_OR_NULL(file->f_path.dentry)
7263 +           || !au_test_aufs(file->f_path.dentry->d_sb))
7264 +               return;
7265 +
7266 +       finfo = au_fi(file);
7267 +       if (!finfo)
7268 +               return;
7269 +       if (finfo->fi_btop < 0)
7270 +               return;
7271 +       fidir = finfo->fi_hdir;
7272 +       if (!fidir)
7273 +               do_pri_file(finfo->fi_btop, finfo->fi_htop.hf_file);
7274 +       else
7275 +               for (bindex = finfo->fi_btop;
7276 +                    bindex >= 0 && bindex <= fidir->fd_bbot;
7277 +                    bindex++) {
7278 +                       hfile = fidir->fd_hfile + bindex;
7279 +                       do_pri_file(bindex, hfile ? hfile->hf_file : NULL);
7280 +               }
7281 +}
7282 +
7283 +static int do_pri_br(aufs_bindex_t bindex, struct au_branch *br)
7284 +{
7285 +       struct vfsmount *mnt;
7286 +       struct super_block *sb;
7287 +
7288 +       if (!br || IS_ERR(br))
7289 +               goto out;
7290 +       mnt = au_br_mnt(br);
7291 +       if (!mnt || IS_ERR(mnt))
7292 +               goto out;
7293 +       sb = mnt->mnt_sb;
7294 +       if (!sb || IS_ERR(sb))
7295 +               goto out;
7296 +
7297 +       dpri("s%d: {perm 0x%x, id %d, cnt %lld, wbr %p}, "
7298 +            "%s, dev 0x%02x%02x, flags 0x%lx, cnt %d, active %d, "
7299 +            "xino %d\n",
7300 +            bindex, br->br_perm, br->br_id, au_br_count(br),
7301 +            br->br_wbr, au_sbtype(sb), MAJOR(sb->s_dev), MINOR(sb->s_dev),
7302 +            sb->s_flags, sb->s_count,
7303 +            atomic_read(&sb->s_active), !!br->br_xino.xi_file);
7304 +       return 0;
7305 +
7306 +out:
7307 +       dpri("s%d: err %ld\n", bindex, PTR_ERR(br));
7308 +       return -1;
7309 +}
7310 +
7311 +void au_dpri_sb(struct super_block *sb)
7312 +{
7313 +       struct au_sbinfo *sbinfo;
7314 +       aufs_bindex_t bindex;
7315 +       int err;
7316 +       /* to reuduce stack size */
7317 +       struct {
7318 +               struct vfsmount mnt;
7319 +               struct au_branch fake;
7320 +       } *a;
7321 +
7322 +       /* this function can be called from magic sysrq */
7323 +       a = kzalloc(sizeof(*a), GFP_ATOMIC);
7324 +       if (unlikely(!a)) {
7325 +               dpri("no memory\n");
7326 +               return;
7327 +       }
7328 +
7329 +       a->mnt.mnt_sb = sb;
7330 +       a->fake.br_path.mnt = &a->mnt;
7331 +       au_br_count_init(&a->fake);
7332 +       err = do_pri_br(-1, &a->fake);
7333 +       au_br_count_fin(&a->fake);
7334 +       au_delayed_kfree(a);
7335 +       dpri("dev 0x%x\n", sb->s_dev);
7336 +       if (err || !au_test_aufs(sb))
7337 +               return;
7338 +
7339 +       sbinfo = au_sbi(sb);
7340 +       if (!sbinfo)
7341 +               return;
7342 +       dpri("nw %d, gen %u, kobj %d\n",
7343 +            atomic_read(&sbinfo->si_nowait.nw_len), sbinfo->si_generation,
7344 +            atomic_read(&sbinfo->si_kobj.kref.refcount));
7345 +       for (bindex = 0; bindex <= sbinfo->si_bbot; bindex++)
7346 +               do_pri_br(bindex, sbinfo->si_branch[0 + bindex]);
7347 +}
7348 +
7349 +/* ---------------------------------------------------------------------- */
7350 +
7351 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line)
7352 +{
7353 +       struct inode *h_inode, *inode = d_inode(dentry);
7354 +       struct dentry *h_dentry;
7355 +       aufs_bindex_t bindex, bbot, bi;
7356 +
7357 +       if (!inode /* || au_di(dentry)->di_lsc == AuLsc_DI_TMP */)
7358 +               return;
7359 +
7360 +       bbot = au_dbbot(dentry);
7361 +       bi = au_ibbot(inode);
7362 +       if (bi < bbot)
7363 +               bbot = bi;
7364 +       bindex = au_dbtop(dentry);
7365 +       bi = au_ibtop(inode);
7366 +       if (bi > bindex)
7367 +               bindex = bi;
7368 +
7369 +       for (; bindex <= bbot; bindex++) {
7370 +               h_dentry = au_h_dptr(dentry, bindex);
7371 +               if (!h_dentry)
7372 +                       continue;
7373 +               h_inode = au_h_iptr(inode, bindex);
7374 +               if (unlikely(h_inode != d_inode(h_dentry))) {
7375 +                       au_debug_on();
7376 +                       AuDbg("b%d, %s:%d\n", bindex, func, line);
7377 +                       AuDbgDentry(dentry);
7378 +                       AuDbgInode(inode);
7379 +                       au_debug_off();
7380 +                       BUG();
7381 +               }
7382 +       }
7383 +}
7384 +
7385 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen)
7386 +{
7387 +       int err, i, j;
7388 +       struct au_dcsub_pages dpages;
7389 +       struct au_dpage *dpage;
7390 +       struct dentry **dentries;
7391 +
7392 +       err = au_dpages_init(&dpages, GFP_NOFS);
7393 +       AuDebugOn(err);
7394 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/1);
7395 +       AuDebugOn(err);
7396 +       for (i = dpages.ndpage - 1; !err && i >= 0; i--) {
7397 +               dpage = dpages.dpages + i;
7398 +               dentries = dpage->dentries;
7399 +               for (j = dpage->ndentry - 1; !err && j >= 0; j--)
7400 +                       AuDebugOn(au_digen_test(dentries[j], sigen));
7401 +       }
7402 +       au_dpages_free(&dpages);
7403 +}
7404 +
7405 +void au_dbg_verify_kthread(void)
7406 +{
7407 +       if (au_wkq_test()) {
7408 +               au_dbg_blocked();
7409 +               /*
7410 +                * It may be recursive, but udba=notify between two aufs mounts,
7411 +                * where a single ro branch is shared, is not a problem.
7412 +                */
7413 +               /* WARN_ON(1); */
7414 +       }
7415 +}
7416 +
7417 +/* ---------------------------------------------------------------------- */
7418 +
7419 +int __init au_debug_init(void)
7420 +{
7421 +       aufs_bindex_t bindex;
7422 +       struct au_vdir_destr destr;
7423 +
7424 +       bindex = -1;
7425 +       AuDebugOn(bindex >= 0);
7426 +
7427 +       destr.len = -1;
7428 +       AuDebugOn(destr.len < NAME_MAX);
7429 +
7430 +#ifdef CONFIG_4KSTACKS
7431 +       pr_warn("CONFIG_4KSTACKS is defined.\n");
7432 +#endif
7433 +
7434 +       return 0;
7435 +}
7436 diff -urN /usr/share/empty/fs/aufs/debug.h linux/fs/aufs/debug.h
7437 --- /usr/share/empty/fs/aufs/debug.h    1970-01-01 01:00:00.000000000 +0100
7438 +++ linux/fs/aufs/debug.h       2016-10-09 16:55:36.486034798 +0200
7439 @@ -0,0 +1,225 @@
7440 +/*
7441 + * Copyright (C) 2005-2016 Junjiro R. Okajima
7442 + *
7443 + * This program, aufs is free software; you can redistribute it and/or modify
7444 + * it under the terms of the GNU General Public License as published by
7445 + * the Free Software Foundation; either version 2 of the License, or
7446 + * (at your option) any later version.
7447 + *
7448 + * This program is distributed in the hope that it will be useful,
7449 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7450 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7451 + * GNU General Public License for more details.
7452 + *
7453 + * You should have received a copy of the GNU General Public License
7454 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7455 + */
7456 +
7457 +/*
7458 + * debug print functions
7459 + */
7460 +
7461 +#ifndef __AUFS_DEBUG_H__
7462 +#define __AUFS_DEBUG_H__
7463 +
7464 +#ifdef __KERNEL__
7465 +
7466 +#include <linux/atomic.h>
7467 +#include <linux/module.h>
7468 +#include <linux/kallsyms.h>
7469 +#include <linux/sysrq.h>
7470 +
7471 +#ifdef CONFIG_AUFS_DEBUG
7472 +#define AuDebugOn(a)           BUG_ON(a)
7473 +
7474 +/* module parameter */
7475 +extern atomic_t aufs_debug;
7476 +static inline void au_debug_on(void)
7477 +{
7478 +       atomic_inc(&aufs_debug);
7479 +}
7480 +static inline void au_debug_off(void)
7481 +{
7482 +       atomic_dec_if_positive(&aufs_debug);
7483 +}
7484 +
7485 +static inline int au_debug_test(void)
7486 +{
7487 +       return atomic_read(&aufs_debug) > 0;
7488 +}
7489 +#else
7490 +#define AuDebugOn(a)           do {} while (0)
7491 +AuStubVoid(au_debug_on, void)
7492 +AuStubVoid(au_debug_off, void)
7493 +AuStubInt0(au_debug_test, void)
7494 +#endif /* CONFIG_AUFS_DEBUG */
7495 +
7496 +#define param_check_atomic_t(name, p) __param_check(name, p, atomic_t)
7497 +
7498 +/* ---------------------------------------------------------------------- */
7499 +
7500 +/* debug print */
7501 +
7502 +#define AuDbg(fmt, ...) do { \
7503 +       if (au_debug_test()) \
7504 +               pr_debug("DEBUG: " fmt, ##__VA_ARGS__); \
7505 +} while (0)
7506 +#define AuLabel(l)             AuDbg(#l "\n")
7507 +#define AuIOErr(fmt, ...)      pr_err("I/O Error, " fmt, ##__VA_ARGS__)
7508 +#define AuWarn1(fmt, ...) do { \
7509 +       static unsigned char _c; \
7510 +       if (!_c++) \
7511 +               pr_warn(fmt, ##__VA_ARGS__); \
7512 +} while (0)
7513 +
7514 +#define AuErr1(fmt, ...) do { \
7515 +       static unsigned char _c; \
7516 +       if (!_c++) \
7517 +               pr_err(fmt, ##__VA_ARGS__); \
7518 +} while (0)
7519 +
7520 +#define AuIOErr1(fmt, ...) do { \
7521 +       static unsigned char _c; \
7522 +       if (!_c++) \
7523 +               AuIOErr(fmt, ##__VA_ARGS__); \
7524 +} while (0)
7525 +
7526 +#define AuUnsupportMsg "This operation is not supported." \
7527 +                       " Please report this application to aufs-users ML."
7528 +#define AuUnsupport(fmt, ...) do { \
7529 +       pr_err(AuUnsupportMsg "\n" fmt, ##__VA_ARGS__); \
7530 +       dump_stack(); \
7531 +} while (0)
7532 +
7533 +#define AuTraceErr(e) do { \
7534 +       if (unlikely((e) < 0)) \
7535 +               AuDbg("err %d\n", (int)(e)); \
7536 +} while (0)
7537 +
7538 +#define AuTraceErrPtr(p) do { \
7539 +       if (IS_ERR(p)) \
7540 +               AuDbg("err %ld\n", PTR_ERR(p)); \
7541 +} while (0)
7542 +
7543 +/* dirty macros for debug print, use with "%.*s" and caution */
7544 +#define AuLNPair(qstr)         (qstr)->len, (qstr)->name
7545 +
7546 +/* ---------------------------------------------------------------------- */
7547 +
7548 +struct dentry;
7549 +#ifdef CONFIG_AUFS_DEBUG
7550 +extern struct mutex au_dbg_mtx;
7551 +extern char *au_plevel;
7552 +struct au_nhash;
7553 +void au_dpri_whlist(struct au_nhash *whlist);
7554 +struct au_vdir;
7555 +void au_dpri_vdir(struct au_vdir *vdir);
7556 +struct inode;
7557 +void au_dpri_inode(struct inode *inode);
7558 +void au_dpri_dalias(struct inode *inode);
7559 +void au_dpri_dentry(struct dentry *dentry);
7560 +struct file;
7561 +void au_dpri_file(struct file *filp);
7562 +struct super_block;
7563 +void au_dpri_sb(struct super_block *sb);
7564 +
7565 +#define au_dbg_verify_dinode(d) __au_dbg_verify_dinode(d, __func__, __LINE__)
7566 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
7567 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen);
7568 +void au_dbg_verify_kthread(void);
7569 +
7570 +int __init au_debug_init(void);
7571 +
7572 +#define AuDbgWhlist(w) do { \
7573 +       mutex_lock(&au_dbg_mtx); \
7574 +       AuDbg(#w "\n"); \
7575 +       au_dpri_whlist(w); \
7576 +       mutex_unlock(&au_dbg_mtx); \
7577 +} while (0)
7578 +
7579 +#define AuDbgVdir(v) do { \
7580 +       mutex_lock(&au_dbg_mtx); \
7581 +       AuDbg(#v "\n"); \
7582 +       au_dpri_vdir(v); \
7583 +       mutex_unlock(&au_dbg_mtx); \
7584 +} while (0)
7585 +
7586 +#define AuDbgInode(i) do { \
7587 +       mutex_lock(&au_dbg_mtx); \
7588 +       AuDbg(#i "\n"); \
7589 +       au_dpri_inode(i); \
7590 +       mutex_unlock(&au_dbg_mtx); \
7591 +} while (0)
7592 +
7593 +#define AuDbgDAlias(i) do { \
7594 +       mutex_lock(&au_dbg_mtx); \
7595 +       AuDbg(#i "\n"); \
7596 +       au_dpri_dalias(i); \
7597 +       mutex_unlock(&au_dbg_mtx); \
7598 +} while (0)
7599 +
7600 +#define AuDbgDentry(d) do { \
7601 +       mutex_lock(&au_dbg_mtx); \
7602 +       AuDbg(#d "\n"); \
7603 +       au_dpri_dentry(d); \
7604 +       mutex_unlock(&au_dbg_mtx); \
7605 +} while (0)
7606 +
7607 +#define AuDbgFile(f) do { \
7608 +       mutex_lock(&au_dbg_mtx); \
7609 +       AuDbg(#f "\n"); \
7610 +       au_dpri_file(f); \
7611 +       mutex_unlock(&au_dbg_mtx); \
7612 +} while (0)
7613 +
7614 +#define AuDbgSb(sb) do { \
7615 +       mutex_lock(&au_dbg_mtx); \
7616 +       AuDbg(#sb "\n"); \
7617 +       au_dpri_sb(sb); \
7618 +       mutex_unlock(&au_dbg_mtx); \
7619 +} while (0)
7620 +
7621 +#define AuDbgSym(addr) do {                            \
7622 +       char sym[KSYM_SYMBOL_LEN];                      \
7623 +       sprint_symbol(sym, (unsigned long)addr);        \
7624 +       AuDbg("%s\n", sym);                             \
7625 +} while (0)
7626 +#else
7627 +AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
7628 +AuStubVoid(au_dbg_verify_gen, struct dentry *parent, unsigned int sigen)
7629 +AuStubVoid(au_dbg_verify_kthread, void)
7630 +AuStubInt0(__init au_debug_init, void)
7631 +
7632 +#define AuDbgWhlist(w)         do {} while (0)
7633 +#define AuDbgVdir(v)           do {} while (0)
7634 +#define AuDbgInode(i)          do {} while (0)
7635 +#define AuDbgDAlias(i)         do {} while (0)
7636 +#define AuDbgDentry(d)         do {} while (0)
7637 +#define AuDbgFile(f)           do {} while (0)
7638 +#define AuDbgSb(sb)            do {} while (0)
7639 +#define AuDbgSym(addr)         do {} while (0)
7640 +#endif /* CONFIG_AUFS_DEBUG */
7641 +
7642 +/* ---------------------------------------------------------------------- */
7643 +
7644 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
7645 +int __init au_sysrq_init(void);
7646 +void au_sysrq_fin(void);
7647 +
7648 +#ifdef CONFIG_HW_CONSOLE
7649 +#define au_dbg_blocked() do { \
7650 +       WARN_ON(1); \
7651 +       handle_sysrq('w'); \
7652 +} while (0)
7653 +#else
7654 +AuStubVoid(au_dbg_blocked, void)
7655 +#endif
7656 +
7657 +#else
7658 +AuStubInt0(__init au_sysrq_init, void)
7659 +AuStubVoid(au_sysrq_fin, void)
7660 +AuStubVoid(au_dbg_blocked, void)
7661 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
7662 +
7663 +#endif /* __KERNEL__ */
7664 +#endif /* __AUFS_DEBUG_H__ */
7665 diff -urN /usr/share/empty/fs/aufs/dentry.c linux/fs/aufs/dentry.c
7666 --- /usr/share/empty/fs/aufs/dentry.c   1970-01-01 01:00:00.000000000 +0100
7667 +++ linux/fs/aufs/dentry.c      2016-10-09 16:55:38.889431135 +0200
7668 @@ -0,0 +1,1130 @@
7669 +/*
7670 + * Copyright (C) 2005-2016 Junjiro R. Okajima
7671 + *
7672 + * This program, aufs is free software; you can redistribute it and/or modify
7673 + * it under the terms of the GNU General Public License as published by
7674 + * the Free Software Foundation; either version 2 of the License, or
7675 + * (at your option) any later version.
7676 + *
7677 + * This program is distributed in the hope that it will be useful,
7678 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7679 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7680 + * GNU General Public License for more details.
7681 + *
7682 + * You should have received a copy of the GNU General Public License
7683 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7684 + */
7685 +
7686 +/*
7687 + * lookup and dentry operations
7688 + */
7689 +
7690 +#include <linux/namei.h>
7691 +#include "aufs.h"
7692 +
7693 +struct au_do_lookup_args {
7694 +       unsigned int            flags;
7695 +       mode_t                  type;
7696 +};
7697 +
7698 +/*
7699 + * returns positive/negative dentry, NULL or an error.
7700 + * NULL means whiteout-ed or not-found.
7701 + */
7702 +static struct dentry*
7703 +au_do_lookup(struct dentry *h_parent, struct dentry *dentry,
7704 +            aufs_bindex_t bindex, struct qstr *wh_name,
7705 +            struct au_do_lookup_args *args)
7706 +{
7707 +       struct dentry *h_dentry;
7708 +       struct inode *h_inode;
7709 +       struct au_branch *br;
7710 +       int wh_found, opq;
7711 +       unsigned char wh_able;
7712 +       const unsigned char allow_neg = !!au_ftest_lkup(args->flags, ALLOW_NEG);
7713 +       const unsigned char ignore_perm = !!au_ftest_lkup(args->flags,
7714 +                                                         IGNORE_PERM);
7715 +
7716 +       wh_found = 0;
7717 +       br = au_sbr(dentry->d_sb, bindex);
7718 +       wh_able = !!au_br_whable(br->br_perm);
7719 +       if (wh_able)
7720 +               wh_found = au_wh_test(h_parent, wh_name, ignore_perm);
7721 +       h_dentry = ERR_PTR(wh_found);
7722 +       if (!wh_found)
7723 +               goto real_lookup;
7724 +       if (unlikely(wh_found < 0))
7725 +               goto out;
7726 +
7727 +       /* We found a whiteout */
7728 +       /* au_set_dbbot(dentry, bindex); */
7729 +       au_set_dbwh(dentry, bindex);
7730 +       if (!allow_neg)
7731 +               return NULL; /* success */
7732 +
7733 +real_lookup:
7734 +       if (!ignore_perm)
7735 +               h_dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
7736 +       else
7737 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7738 +       if (IS_ERR(h_dentry)) {
7739 +               if (PTR_ERR(h_dentry) == -ENAMETOOLONG
7740 +                   && !allow_neg)
7741 +                       h_dentry = NULL;
7742 +               goto out;
7743 +       }
7744 +
7745 +       h_inode = d_inode(h_dentry);
7746 +       if (d_is_negative(h_dentry)) {
7747 +               if (!allow_neg)
7748 +                       goto out_neg;
7749 +       } else if (wh_found
7750 +                  || (args->type && args->type != (h_inode->i_mode & S_IFMT)))
7751 +               goto out_neg;
7752 +
7753 +       if (au_dbbot(dentry) <= bindex)
7754 +               au_set_dbbot(dentry, bindex);
7755 +       if (au_dbtop(dentry) < 0 || bindex < au_dbtop(dentry))
7756 +               au_set_dbtop(dentry, bindex);
7757 +       au_set_h_dptr(dentry, bindex, h_dentry);
7758 +
7759 +       if (!d_is_dir(h_dentry)
7760 +           || !wh_able
7761 +           || (d_really_is_positive(dentry) && !d_is_dir(dentry)))
7762 +               goto out; /* success */
7763 +
7764 +       inode_lock_nested(h_inode, AuLsc_I_CHILD);
7765 +       opq = au_diropq_test(h_dentry);
7766 +       inode_unlock(h_inode);
7767 +       if (opq > 0)
7768 +               au_set_dbdiropq(dentry, bindex);
7769 +       else if (unlikely(opq < 0)) {
7770 +               au_set_h_dptr(dentry, bindex, NULL);
7771 +               h_dentry = ERR_PTR(opq);
7772 +       }
7773 +       goto out;
7774 +
7775 +out_neg:
7776 +       dput(h_dentry);
7777 +       h_dentry = NULL;
7778 +out:
7779 +       return h_dentry;
7780 +}
7781 +
7782 +static int au_test_shwh(struct super_block *sb, const struct qstr *name)
7783 +{
7784 +       if (unlikely(!au_opt_test(au_mntflags(sb), SHWH)
7785 +                    && !strncmp(name->name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)))
7786 +               return -EPERM;
7787 +       return 0;
7788 +}
7789 +
7790 +/*
7791 + * returns the number of lower positive dentries,
7792 + * otherwise an error.
7793 + * can be called at unlinking with @type is zero.
7794 + */
7795 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t btop,
7796 +                  unsigned int flags)
7797 +{
7798 +       int npositive, err;
7799 +       aufs_bindex_t bindex, btail, bdiropq;
7800 +       unsigned char isdir, dirperm1;
7801 +       struct qstr whname;
7802 +       struct au_do_lookup_args args = {
7803 +               .flags          = flags
7804 +       };
7805 +       const struct qstr *name = &dentry->d_name;
7806 +       struct dentry *parent;
7807 +       struct super_block *sb;
7808 +
7809 +       sb = dentry->d_sb;
7810 +       err = au_test_shwh(sb, name);
7811 +       if (unlikely(err))
7812 +               goto out;
7813 +
7814 +       err = au_wh_name_alloc(&whname, name);
7815 +       if (unlikely(err))
7816 +               goto out;
7817 +
7818 +       isdir = !!d_is_dir(dentry);
7819 +       dirperm1 = !!au_opt_test(au_mntflags(sb), DIRPERM1);
7820 +
7821 +       npositive = 0;
7822 +       parent = dget_parent(dentry);
7823 +       btail = au_dbtaildir(parent);
7824 +       for (bindex = btop; bindex <= btail; bindex++) {
7825 +               struct dentry *h_parent, *h_dentry;
7826 +               struct inode *h_inode, *h_dir;
7827 +
7828 +               h_dentry = au_h_dptr(dentry, bindex);
7829 +               if (h_dentry) {
7830 +                       if (d_is_positive(h_dentry))
7831 +                               npositive++;
7832 +                       break;
7833 +               }
7834 +               h_parent = au_h_dptr(parent, bindex);
7835 +               if (!h_parent || !d_is_dir(h_parent))
7836 +                       continue;
7837 +
7838 +               h_dir = d_inode(h_parent);
7839 +               inode_lock_nested(h_dir, AuLsc_I_PARENT);
7840 +               h_dentry = au_do_lookup(h_parent, dentry, bindex, &whname,
7841 +                                       &args);
7842 +               inode_unlock(h_dir);
7843 +               err = PTR_ERR(h_dentry);
7844 +               if (IS_ERR(h_dentry))
7845 +                       goto out_parent;
7846 +               if (h_dentry)
7847 +                       au_fclr_lkup(args.flags, ALLOW_NEG);
7848 +               if (dirperm1)
7849 +                       au_fset_lkup(args.flags, IGNORE_PERM);
7850 +
7851 +               if (au_dbwh(dentry) == bindex)
7852 +                       break;
7853 +               if (!h_dentry)
7854 +                       continue;
7855 +               if (d_is_negative(h_dentry))
7856 +                       continue;
7857 +               h_inode = d_inode(h_dentry);
7858 +               npositive++;
7859 +               if (!args.type)
7860 +                       args.type = h_inode->i_mode & S_IFMT;
7861 +               if (args.type != S_IFDIR)
7862 +                       break;
7863 +               else if (isdir) {
7864 +                       /* the type of lower may be different */
7865 +                       bdiropq = au_dbdiropq(dentry);
7866 +                       if (bdiropq >= 0 && bdiropq <= bindex)
7867 +                               break;
7868 +               }
7869 +       }
7870 +
7871 +       if (npositive) {
7872 +               AuLabel(positive);
7873 +               au_update_dbtop(dentry);
7874 +       }
7875 +       err = npositive;
7876 +       if (unlikely(!au_opt_test(au_mntflags(sb), UDBA_NONE)
7877 +                    && au_dbtop(dentry) < 0)) {
7878 +               err = -EIO;
7879 +               AuIOErr("both of real entry and whiteout found, %pd, err %d\n",
7880 +                       dentry, err);
7881 +       }
7882 +
7883 +out_parent:
7884 +       dput(parent);
7885 +       au_delayed_kfree(whname.name);
7886 +out:
7887 +       return err;
7888 +}
7889 +
7890 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent)
7891 +{
7892 +       struct dentry *dentry;
7893 +       int wkq_err;
7894 +
7895 +       if (!au_test_h_perm_sio(d_inode(parent), MAY_EXEC))
7896 +               dentry = vfsub_lkup_one(name, parent);
7897 +       else {
7898 +               struct vfsub_lkup_one_args args = {
7899 +                       .errp   = &dentry,
7900 +                       .name   = name,
7901 +                       .parent = parent
7902 +               };
7903 +
7904 +               wkq_err = au_wkq_wait(vfsub_call_lkup_one, &args);
7905 +               if (unlikely(wkq_err))
7906 +                       dentry = ERR_PTR(wkq_err);
7907 +       }
7908 +
7909 +       return dentry;
7910 +}
7911 +
7912 +/*
7913 + * lookup @dentry on @bindex which should be negative.
7914 + */
7915 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh)
7916 +{
7917 +       int err;
7918 +       struct dentry *parent, *h_parent, *h_dentry;
7919 +       struct au_branch *br;
7920 +
7921 +       parent = dget_parent(dentry);
7922 +       h_parent = au_h_dptr(parent, bindex);
7923 +       br = au_sbr(dentry->d_sb, bindex);
7924 +       if (wh)
7925 +               h_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
7926 +       else
7927 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7928 +       err = PTR_ERR(h_dentry);
7929 +       if (IS_ERR(h_dentry))
7930 +               goto out;
7931 +       if (unlikely(d_is_positive(h_dentry))) {
7932 +               err = -EIO;
7933 +               AuIOErr("%pd should be negative on b%d.\n", h_dentry, bindex);
7934 +               dput(h_dentry);
7935 +               goto out;
7936 +       }
7937 +
7938 +       err = 0;
7939 +       if (bindex < au_dbtop(dentry))
7940 +               au_set_dbtop(dentry, bindex);
7941 +       if (au_dbbot(dentry) < bindex)
7942 +               au_set_dbbot(dentry, bindex);
7943 +       au_set_h_dptr(dentry, bindex, h_dentry);
7944 +
7945 +out:
7946 +       dput(parent);
7947 +       return err;
7948 +}
7949 +
7950 +/* ---------------------------------------------------------------------- */
7951 +
7952 +/* subset of struct inode */
7953 +struct au_iattr {
7954 +       unsigned long           i_ino;
7955 +       /* unsigned int         i_nlink; */
7956 +       kuid_t                  i_uid;
7957 +       kgid_t                  i_gid;
7958 +       u64                     i_version;
7959 +/*
7960 +       loff_t                  i_size;
7961 +       blkcnt_t                i_blocks;
7962 +*/
7963 +       umode_t                 i_mode;
7964 +};
7965 +
7966 +static void au_iattr_save(struct au_iattr *ia, struct inode *h_inode)
7967 +{
7968 +       ia->i_ino = h_inode->i_ino;
7969 +       /* ia->i_nlink = h_inode->i_nlink; */
7970 +       ia->i_uid = h_inode->i_uid;
7971 +       ia->i_gid = h_inode->i_gid;
7972 +       ia->i_version = h_inode->i_version;
7973 +/*
7974 +       ia->i_size = h_inode->i_size;
7975 +       ia->i_blocks = h_inode->i_blocks;
7976 +*/
7977 +       ia->i_mode = (h_inode->i_mode & S_IFMT);
7978 +}
7979 +
7980 +static int au_iattr_test(struct au_iattr *ia, struct inode *h_inode)
7981 +{
7982 +       return ia->i_ino != h_inode->i_ino
7983 +               /* || ia->i_nlink != h_inode->i_nlink */
7984 +               || !uid_eq(ia->i_uid, h_inode->i_uid)
7985 +               || !gid_eq(ia->i_gid, h_inode->i_gid)
7986 +               || ia->i_version != h_inode->i_version
7987 +/*
7988 +               || ia->i_size != h_inode->i_size
7989 +               || ia->i_blocks != h_inode->i_blocks
7990 +*/
7991 +               || ia->i_mode != (h_inode->i_mode & S_IFMT);
7992 +}
7993 +
7994 +static int au_h_verify_dentry(struct dentry *h_dentry, struct dentry *h_parent,
7995 +                             struct au_branch *br)
7996 +{
7997 +       int err;
7998 +       struct au_iattr ia;
7999 +       struct inode *h_inode;
8000 +       struct dentry *h_d;
8001 +       struct super_block *h_sb;
8002 +
8003 +       err = 0;
8004 +       memset(&ia, -1, sizeof(ia));
8005 +       h_sb = h_dentry->d_sb;
8006 +       h_inode = NULL;
8007 +       if (d_is_positive(h_dentry)) {
8008 +               h_inode = d_inode(h_dentry);
8009 +               au_iattr_save(&ia, h_inode);
8010 +       } else if (au_test_nfs(h_sb) || au_test_fuse(h_sb))
8011 +               /* nfs d_revalidate may return 0 for negative dentry */
8012 +               /* fuse d_revalidate always return 0 for negative dentry */
8013 +               goto out;
8014 +
8015 +       /* main purpose is namei.c:cached_lookup() and d_revalidate */
8016 +       h_d = vfsub_lkup_one(&h_dentry->d_name, h_parent);
8017 +       err = PTR_ERR(h_d);
8018 +       if (IS_ERR(h_d))
8019 +               goto out;
8020 +
8021 +       err = 0;
8022 +       if (unlikely(h_d != h_dentry
8023 +                    || d_inode(h_d) != h_inode
8024 +                    || (h_inode && au_iattr_test(&ia, h_inode))))
8025 +               err = au_busy_or_stale();
8026 +       dput(h_d);
8027 +
8028 +out:
8029 +       AuTraceErr(err);
8030 +       return err;
8031 +}
8032 +
8033 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
8034 +               struct dentry *h_parent, struct au_branch *br)
8035 +{
8036 +       int err;
8037 +
8038 +       err = 0;
8039 +       if (udba == AuOpt_UDBA_REVAL
8040 +           && !au_test_fs_remote(h_dentry->d_sb)) {
8041 +               IMustLock(h_dir);
8042 +               err = (d_inode(h_dentry->d_parent) != h_dir);
8043 +       } else if (udba != AuOpt_UDBA_NONE)
8044 +               err = au_h_verify_dentry(h_dentry, h_parent, br);
8045 +
8046 +       return err;
8047 +}
8048 +
8049 +/* ---------------------------------------------------------------------- */
8050 +
8051 +static int au_do_refresh_hdentry(struct dentry *dentry, struct dentry *parent)
8052 +{
8053 +       int err;
8054 +       aufs_bindex_t new_bindex, bindex, bbot, bwh, bdiropq;
8055 +       struct au_hdentry tmp, *p, *q;
8056 +       struct au_dinfo *dinfo;
8057 +       struct super_block *sb;
8058 +
8059 +       DiMustWriteLock(dentry);
8060 +
8061 +       sb = dentry->d_sb;
8062 +       dinfo = au_di(dentry);
8063 +       bbot = dinfo->di_bbot;
8064 +       bwh = dinfo->di_bwh;
8065 +       bdiropq = dinfo->di_bdiropq;
8066 +       bindex = dinfo->di_btop;
8067 +       p = au_hdentry(dinfo, bindex);
8068 +       for (; bindex <= bbot; bindex++, p++) {
8069 +               if (!p->hd_dentry)
8070 +                       continue;
8071 +
8072 +               new_bindex = au_br_index(sb, p->hd_id);
8073 +               if (new_bindex == bindex)
8074 +                       continue;
8075 +
8076 +               if (dinfo->di_bwh == bindex)
8077 +                       bwh = new_bindex;
8078 +               if (dinfo->di_bdiropq == bindex)
8079 +                       bdiropq = new_bindex;
8080 +               if (new_bindex < 0) {
8081 +                       au_hdput(p);
8082 +                       p->hd_dentry = NULL;
8083 +                       continue;
8084 +               }
8085 +
8086 +               /* swap two lower dentries, and loop again */
8087 +               q = au_hdentry(dinfo, new_bindex);
8088 +               tmp = *q;
8089 +               *q = *p;
8090 +               *p = tmp;
8091 +               if (tmp.hd_dentry) {
8092 +                       bindex--;
8093 +                       p--;
8094 +               }
8095 +       }
8096 +
8097 +       dinfo->di_bwh = -1;
8098 +       if (bwh >= 0 && bwh <= au_sbbot(sb) && au_sbr_whable(sb, bwh))
8099 +               dinfo->di_bwh = bwh;
8100 +
8101 +       dinfo->di_bdiropq = -1;
8102 +       if (bdiropq >= 0
8103 +           && bdiropq <= au_sbbot(sb)
8104 +           && au_sbr_whable(sb, bdiropq))
8105 +               dinfo->di_bdiropq = bdiropq;
8106 +
8107 +       err = -EIO;
8108 +       dinfo->di_btop = -1;
8109 +       dinfo->di_bbot = -1;
8110 +       bbot = au_dbbot(parent);
8111 +       bindex = 0;
8112 +       p = au_hdentry(dinfo, bindex);
8113 +       for (; bindex <= bbot; bindex++, p++)
8114 +               if (p->hd_dentry) {
8115 +                       dinfo->di_btop = bindex;
8116 +                       break;
8117 +               }
8118 +
8119 +       if (dinfo->di_btop >= 0) {
8120 +               bindex = bbot;
8121 +               p = au_hdentry(dinfo, bindex);
8122 +               for (; bindex >= 0; bindex--, p--)
8123 +                       if (p->hd_dentry) {
8124 +                               dinfo->di_bbot = bindex;
8125 +                               err = 0;
8126 +                               break;
8127 +                       }
8128 +       }
8129 +
8130 +       return err;
8131 +}
8132 +
8133 +static void au_do_hide(struct dentry *dentry)
8134 +{
8135 +       struct inode *inode;
8136 +
8137 +       if (d_really_is_positive(dentry)) {
8138 +               inode = d_inode(dentry);
8139 +               if (!d_is_dir(dentry)) {
8140 +                       if (inode->i_nlink && !d_unhashed(dentry))
8141 +                               drop_nlink(inode);
8142 +               } else {
8143 +                       clear_nlink(inode);
8144 +                       /* stop next lookup */
8145 +                       inode->i_flags |= S_DEAD;
8146 +               }
8147 +               smp_mb(); /* necessary? */
8148 +       }
8149 +       d_drop(dentry);
8150 +}
8151 +
8152 +static int au_hide_children(struct dentry *parent)
8153 +{
8154 +       int err, i, j, ndentry;
8155 +       struct au_dcsub_pages dpages;
8156 +       struct au_dpage *dpage;
8157 +       struct dentry *dentry;
8158 +
8159 +       err = au_dpages_init(&dpages, GFP_NOFS);
8160 +       if (unlikely(err))
8161 +               goto out;
8162 +       err = au_dcsub_pages(&dpages, parent, NULL, NULL);
8163 +       if (unlikely(err))
8164 +               goto out_dpages;
8165 +
8166 +       /* in reverse order */
8167 +       for (i = dpages.ndpage - 1; i >= 0; i--) {
8168 +               dpage = dpages.dpages + i;
8169 +               ndentry = dpage->ndentry;
8170 +               for (j = ndentry - 1; j >= 0; j--) {
8171 +                       dentry = dpage->dentries[j];
8172 +                       if (dentry != parent)
8173 +                               au_do_hide(dentry);
8174 +               }
8175 +       }
8176 +
8177 +out_dpages:
8178 +       au_dpages_free(&dpages);
8179 +out:
8180 +       return err;
8181 +}
8182 +
8183 +static void au_hide(struct dentry *dentry)
8184 +{
8185 +       int err;
8186 +
8187 +       AuDbgDentry(dentry);
8188 +       if (d_is_dir(dentry)) {
8189 +               /* shrink_dcache_parent(dentry); */
8190 +               err = au_hide_children(dentry);
8191 +               if (unlikely(err))
8192 +                       AuIOErr("%pd, failed hiding children, ignored %d\n",
8193 +                               dentry, err);
8194 +       }
8195 +       au_do_hide(dentry);
8196 +}
8197 +
8198 +/*
8199 + * By adding a dirty branch, a cached dentry may be affected in various ways.
8200 + *
8201 + * a dirty branch is added
8202 + * - on the top of layers
8203 + * - in the middle of layers
8204 + * - to the bottom of layers
8205 + *
8206 + * on the added branch there exists
8207 + * - a whiteout
8208 + * - a diropq
8209 + * - a same named entry
8210 + *   + exist
8211 + *     * negative --> positive
8212 + *     * positive --> positive
8213 + *      - type is unchanged
8214 + *      - type is changed
8215 + *   + doesn't exist
8216 + *     * negative --> negative
8217 + *     * positive --> negative (rejected by au_br_del() for non-dir case)
8218 + * - none
8219 + */
8220 +static int au_refresh_by_dinfo(struct dentry *dentry, struct au_dinfo *dinfo,
8221 +                              struct au_dinfo *tmp)
8222 +{
8223 +       int err;
8224 +       aufs_bindex_t bindex, bbot;
8225 +       struct {
8226 +               struct dentry *dentry;
8227 +               struct inode *inode;
8228 +               mode_t mode;
8229 +       } orig_h, tmp_h = {
8230 +               .dentry = NULL
8231 +       };
8232 +       struct au_hdentry *hd;
8233 +       struct inode *inode, *h_inode;
8234 +       struct dentry *h_dentry;
8235 +
8236 +       err = 0;
8237 +       AuDebugOn(dinfo->di_btop < 0);
8238 +       orig_h.mode = 0;
8239 +       orig_h.dentry = au_hdentry(dinfo, dinfo->di_btop)->hd_dentry;
8240 +       orig_h.inode = NULL;
8241 +       if (d_is_positive(orig_h.dentry)) {
8242 +               orig_h.inode = d_inode(orig_h.dentry);
8243 +               orig_h.mode = orig_h.inode->i_mode & S_IFMT;
8244 +       }
8245 +       if (tmp->di_btop >= 0) {
8246 +               tmp_h.dentry = au_hdentry(tmp, tmp->di_btop)->hd_dentry;
8247 +               if (d_is_positive(tmp_h.dentry)) {
8248 +                       tmp_h.inode = d_inode(tmp_h.dentry);
8249 +                       tmp_h.mode = tmp_h.inode->i_mode & S_IFMT;
8250 +               }
8251 +       }
8252 +
8253 +       inode = NULL;
8254 +       if (d_really_is_positive(dentry))
8255 +               inode = d_inode(dentry);
8256 +       if (!orig_h.inode) {
8257 +               AuDbg("nagative originally\n");
8258 +               if (inode) {
8259 +                       au_hide(dentry);
8260 +                       goto out;
8261 +               }
8262 +               AuDebugOn(inode);
8263 +               AuDebugOn(dinfo->di_btop != dinfo->di_bbot);
8264 +               AuDebugOn(dinfo->di_bdiropq != -1);
8265 +
8266 +               if (!tmp_h.inode) {
8267 +                       AuDbg("negative --> negative\n");
8268 +                       /* should have only one negative lower */
8269 +                       if (tmp->di_btop >= 0
8270 +                           && tmp->di_btop < dinfo->di_btop) {
8271 +                               AuDebugOn(tmp->di_btop != tmp->di_bbot);
8272 +                               AuDebugOn(dinfo->di_btop != dinfo->di_bbot);
8273 +                               au_set_h_dptr(dentry, dinfo->di_btop, NULL);
8274 +                               au_di_cp(dinfo, tmp);
8275 +                               hd = au_hdentry(tmp, tmp->di_btop);
8276 +                               au_set_h_dptr(dentry, tmp->di_btop,
8277 +                                             dget(hd->hd_dentry));
8278 +                       }
8279 +                       au_dbg_verify_dinode(dentry);
8280 +               } else {
8281 +                       AuDbg("negative --> positive\n");
8282 +                       /*
8283 +                        * similar to the behaviour of creating with bypassing
8284 +                        * aufs.
8285 +                        * unhash it in order to force an error in the
8286 +                        * succeeding create operation.
8287 +                        * we should not set S_DEAD here.
8288 +                        */
8289 +                       d_drop(dentry);
8290 +                       /* au_di_swap(tmp, dinfo); */
8291 +                       au_dbg_verify_dinode(dentry);
8292 +               }
8293 +       } else {
8294 +               AuDbg("positive originally\n");
8295 +               /* inode may be NULL */
8296 +               AuDebugOn(inode && (inode->i_mode & S_IFMT) != orig_h.mode);
8297 +               if (!tmp_h.inode) {
8298 +                       AuDbg("positive --> negative\n");
8299 +                       /* or bypassing aufs */
8300 +                       au_hide(dentry);
8301 +                       if (tmp->di_bwh >= 0 && tmp->di_bwh <= dinfo->di_btop)
8302 +                               dinfo->di_bwh = tmp->di_bwh;
8303 +                       if (inode)
8304 +                               err = au_refresh_hinode_self(inode);
8305 +                       au_dbg_verify_dinode(dentry);
8306 +               } else if (orig_h.mode == tmp_h.mode) {
8307 +                       AuDbg("positive --> positive, same type\n");
8308 +                       if (!S_ISDIR(orig_h.mode)
8309 +                           && dinfo->di_btop > tmp->di_btop) {
8310 +                               /*
8311 +                                * similar to the behaviour of removing and
8312 +                                * creating.
8313 +                                */
8314 +                               au_hide(dentry);
8315 +                               if (inode)
8316 +                                       err = au_refresh_hinode_self(inode);
8317 +                               au_dbg_verify_dinode(dentry);
8318 +                       } else {
8319 +                               /* fill empty slots */
8320 +                               if (dinfo->di_btop > tmp->di_btop)
8321 +                                       dinfo->di_btop = tmp->di_btop;
8322 +                               if (dinfo->di_bbot < tmp->di_bbot)
8323 +                                       dinfo->di_bbot = tmp->di_bbot;
8324 +                               dinfo->di_bwh = tmp->di_bwh;
8325 +                               dinfo->di_bdiropq = tmp->di_bdiropq;
8326 +                               bbot = dinfo->di_bbot;
8327 +                               bindex = tmp->di_btop;
8328 +                               hd = au_hdentry(tmp, bindex);
8329 +                               for (; bindex <= bbot; bindex++, hd++) {
8330 +                                       if (au_h_dptr(dentry, bindex))
8331 +                                               continue;
8332 +                                       h_dentry = hd->hd_dentry;
8333 +                                       if (!h_dentry)
8334 +                                               continue;
8335 +                                       AuDebugOn(d_is_negative(h_dentry));
8336 +                                       h_inode = d_inode(h_dentry);
8337 +                                       AuDebugOn(orig_h.mode
8338 +                                                 != (h_inode->i_mode
8339 +                                                     & S_IFMT));
8340 +                                       au_set_h_dptr(dentry, bindex,
8341 +                                                     dget(h_dentry));
8342 +                               }
8343 +                               if (inode)
8344 +                                       err = au_refresh_hinode(inode, dentry);
8345 +                               au_dbg_verify_dinode(dentry);
8346 +                       }
8347 +               } else {
8348 +                       AuDbg("positive --> positive, different type\n");
8349 +                       /* similar to the behaviour of removing and creating */
8350 +                       au_hide(dentry);
8351 +                       if (inode)
8352 +                               err = au_refresh_hinode_self(inode);
8353 +                       au_dbg_verify_dinode(dentry);
8354 +               }
8355 +       }
8356 +
8357 +out:
8358 +       return err;
8359 +}
8360 +
8361 +void au_refresh_dop(struct dentry *dentry, int force_reval)
8362 +{
8363 +       const struct dentry_operations *dop
8364 +               = force_reval ? &aufs_dop : dentry->d_sb->s_d_op;
8365 +       static const unsigned int mask
8366 +               = DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE;
8367 +
8368 +       BUILD_BUG_ON(sizeof(mask) != sizeof(dentry->d_flags));
8369 +
8370 +       if (dentry->d_op == dop)
8371 +               return;
8372 +
8373 +       AuDbg("%pd\n", dentry);
8374 +       spin_lock(&dentry->d_lock);
8375 +       if (dop == &aufs_dop)
8376 +               dentry->d_flags |= mask;
8377 +       else
8378 +               dentry->d_flags &= ~mask;
8379 +       dentry->d_op = dop;
8380 +       spin_unlock(&dentry->d_lock);
8381 +}
8382 +
8383 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent)
8384 +{
8385 +       int err, ebrange, nbr;
8386 +       unsigned int sigen;
8387 +       struct au_dinfo *dinfo, *tmp;
8388 +       struct super_block *sb;
8389 +       struct inode *inode;
8390 +
8391 +       DiMustWriteLock(dentry);
8392 +       AuDebugOn(IS_ROOT(dentry));
8393 +       AuDebugOn(d_really_is_negative(parent));
8394 +
8395 +       sb = dentry->d_sb;
8396 +       sigen = au_sigen(sb);
8397 +       err = au_digen_test(parent, sigen);
8398 +       if (unlikely(err))
8399 +               goto out;
8400 +
8401 +       nbr = au_sbbot(sb) + 1;
8402 +       dinfo = au_di(dentry);
8403 +       err = au_di_realloc(dinfo, nbr, /*may_shrink*/0);
8404 +       if (unlikely(err))
8405 +               goto out;
8406 +       ebrange = au_dbrange_test(dentry);
8407 +       if (!ebrange)
8408 +               ebrange = au_do_refresh_hdentry(dentry, parent);
8409 +
8410 +       if (d_unhashed(dentry) || ebrange /* || dinfo->di_tmpfile */) {
8411 +               AuDebugOn(au_dbtop(dentry) < 0 && au_dbbot(dentry) >= 0);
8412 +               if (d_really_is_positive(dentry)) {
8413 +                       inode = d_inode(dentry);
8414 +                       err = au_refresh_hinode_self(inode);
8415 +               }
8416 +               au_dbg_verify_dinode(dentry);
8417 +               if (!err)
8418 +                       goto out_dgen; /* success */
8419 +               goto out;
8420 +       }
8421 +
8422 +       /* temporary dinfo */
8423 +       AuDbgDentry(dentry);
8424 +       err = -ENOMEM;
8425 +       tmp = au_di_alloc(sb, AuLsc_DI_TMP);
8426 +       if (unlikely(!tmp))
8427 +               goto out;
8428 +       au_di_swap(tmp, dinfo);
8429 +       /* returns the number of positive dentries */
8430 +       /*
8431 +        * if current working dir is removed, it returns an error.
8432 +        * but the dentry is legal.
8433 +        */
8434 +       err = au_lkup_dentry(dentry, /*btop*/0, AuLkup_ALLOW_NEG);
8435 +       AuDbgDentry(dentry);
8436 +       au_di_swap(tmp, dinfo);
8437 +       if (err == -ENOENT)
8438 +               err = 0;
8439 +       if (err >= 0) {
8440 +               /* compare/refresh by dinfo */
8441 +               AuDbgDentry(dentry);
8442 +               err = au_refresh_by_dinfo(dentry, dinfo, tmp);
8443 +               au_dbg_verify_dinode(dentry);
8444 +               AuTraceErr(err);
8445 +       }
8446 +       au_di_realloc(dinfo, nbr, /*may_shrink*/1); /* harmless if err */
8447 +       au_rw_write_unlock(&tmp->di_rwsem);
8448 +       au_di_free(tmp);
8449 +       if (unlikely(err))
8450 +               goto out;
8451 +
8452 +out_dgen:
8453 +       au_update_digen(dentry);
8454 +out:
8455 +       if (unlikely(err && !(dentry->d_flags & DCACHE_NFSFS_RENAMED))) {
8456 +               AuIOErr("failed refreshing %pd, %d\n", dentry, err);
8457 +               AuDbgDentry(dentry);
8458 +       }
8459 +       AuTraceErr(err);
8460 +       return err;
8461 +}
8462 +
8463 +static int au_do_h_d_reval(struct dentry *h_dentry, unsigned int flags,
8464 +                          struct dentry *dentry, aufs_bindex_t bindex)
8465 +{
8466 +       int err, valid;
8467 +
8468 +       err = 0;
8469 +       if (!(h_dentry->d_flags & DCACHE_OP_REVALIDATE))
8470 +               goto out;
8471 +
8472 +       AuDbg("b%d\n", bindex);
8473 +       /*
8474 +        * gave up supporting LOOKUP_CREATE/OPEN for lower fs,
8475 +        * due to whiteout and branch permission.
8476 +        */
8477 +       flags &= ~(/*LOOKUP_PARENT |*/ LOOKUP_OPEN | LOOKUP_CREATE
8478 +                  | LOOKUP_FOLLOW | LOOKUP_EXCL);
8479 +       /* it may return tri-state */
8480 +       valid = h_dentry->d_op->d_revalidate(h_dentry, flags);
8481 +
8482 +       if (unlikely(valid < 0))
8483 +               err = valid;
8484 +       else if (!valid)
8485 +               err = -EINVAL;
8486 +
8487 +out:
8488 +       AuTraceErr(err);
8489 +       return err;
8490 +}
8491 +
8492 +/* todo: remove this */
8493 +static int h_d_revalidate(struct dentry *dentry, struct inode *inode,
8494 +                         unsigned int flags, int do_udba)
8495 +{
8496 +       int err;
8497 +       umode_t mode, h_mode;
8498 +       aufs_bindex_t bindex, btail, btop, ibs, ibe;
8499 +       unsigned char plus, unhashed, is_root, h_plus, h_nfs, tmpfile;
8500 +       struct inode *h_inode, *h_cached_inode;
8501 +       struct dentry *h_dentry;
8502 +       struct qstr *name, *h_name;
8503 +
8504 +       err = 0;
8505 +       plus = 0;
8506 +       mode = 0;
8507 +       ibs = -1;
8508 +       ibe = -1;
8509 +       unhashed = !!d_unhashed(dentry);
8510 +       is_root = !!IS_ROOT(dentry);
8511 +       name = &dentry->d_name;
8512 +       tmpfile = au_di(dentry)->di_tmpfile;
8513 +
8514 +       /*
8515 +        * Theoretically, REVAL test should be unnecessary in case of
8516 +        * {FS,I}NOTIFY.
8517 +        * But {fs,i}notify doesn't fire some necessary events,
8518 +        *      IN_ATTRIB for atime/nlink/pageio
8519 +        * Let's do REVAL test too.
8520 +        */
8521 +       if (do_udba && inode) {
8522 +               mode = (inode->i_mode & S_IFMT);
8523 +               plus = (inode->i_nlink > 0);
8524 +               ibs = au_ibtop(inode);
8525 +               ibe = au_ibbot(inode);
8526 +       }
8527 +
8528 +       btop = au_dbtop(dentry);
8529 +       btail = btop;
8530 +       if (inode && S_ISDIR(inode->i_mode))
8531 +               btail = au_dbtaildir(dentry);
8532 +       for (bindex = btop; bindex <= btail; bindex++) {
8533 +               h_dentry = au_h_dptr(dentry, bindex);
8534 +               if (!h_dentry)
8535 +                       continue;
8536 +
8537 +               AuDbg("b%d, %pd\n", bindex, h_dentry);
8538 +               h_nfs = !!au_test_nfs(h_dentry->d_sb);
8539 +               spin_lock(&h_dentry->d_lock);
8540 +               h_name = &h_dentry->d_name;
8541 +               if (unlikely(do_udba
8542 +                            && !is_root
8543 +                            && ((!h_nfs
8544 +                                 && (unhashed != !!d_unhashed(h_dentry)
8545 +                                     || (!tmpfile
8546 +                                         && !au_qstreq(name, h_name))
8547 +                                         ))
8548 +                                || (h_nfs
8549 +                                    && !(flags & LOOKUP_OPEN)
8550 +                                    && (h_dentry->d_flags
8551 +                                        & DCACHE_NFSFS_RENAMED)))
8552 +                           )) {
8553 +                       int h_unhashed;
8554 +
8555 +                       h_unhashed = d_unhashed(h_dentry);
8556 +                       spin_unlock(&h_dentry->d_lock);
8557 +                       AuDbg("unhash 0x%x 0x%x, %pd %pd\n",
8558 +                             unhashed, h_unhashed, dentry, h_dentry);
8559 +                       goto err;
8560 +               }
8561 +               spin_unlock(&h_dentry->d_lock);
8562 +
8563 +               err = au_do_h_d_reval(h_dentry, flags, dentry, bindex);
8564 +               if (unlikely(err))
8565 +                       /* do not goto err, to keep the errno */
8566 +                       break;
8567 +
8568 +               /* todo: plink too? */
8569 +               if (!do_udba)
8570 +                       continue;
8571 +
8572 +               /* UDBA tests */
8573 +               if (unlikely(!!inode != d_is_positive(h_dentry)))
8574 +                       goto err;
8575 +
8576 +               h_inode = NULL;
8577 +               if (d_is_positive(h_dentry))
8578 +                       h_inode = d_inode(h_dentry);
8579 +               h_plus = plus;
8580 +               h_mode = mode;
8581 +               h_cached_inode = h_inode;
8582 +               if (h_inode) {
8583 +                       h_mode = (h_inode->i_mode & S_IFMT);
8584 +                       h_plus = (h_inode->i_nlink > 0);
8585 +               }
8586 +               if (inode && ibs <= bindex && bindex <= ibe)
8587 +                       h_cached_inode = au_h_iptr(inode, bindex);
8588 +
8589 +               if (!h_nfs) {
8590 +                       if (unlikely(plus != h_plus && !tmpfile))
8591 +                               goto err;
8592 +               } else {
8593 +                       if (unlikely(!(h_dentry->d_flags & DCACHE_NFSFS_RENAMED)
8594 +                                    && !is_root
8595 +                                    && !IS_ROOT(h_dentry)
8596 +                                    && unhashed != d_unhashed(h_dentry)))
8597 +                               goto err;
8598 +               }
8599 +               if (unlikely(mode != h_mode
8600 +                            || h_cached_inode != h_inode))
8601 +                       goto err;
8602 +               continue;
8603 +
8604 +err:
8605 +               err = -EINVAL;
8606 +               break;
8607 +       }
8608 +
8609 +       AuTraceErr(err);
8610 +       return err;
8611 +}
8612 +
8613 +/* todo: consolidate with do_refresh() and au_reval_for_attr() */
8614 +static int simple_reval_dpath(struct dentry *dentry, unsigned int sigen)
8615 +{
8616 +       int err;
8617 +       struct dentry *parent;
8618 +
8619 +       if (!au_digen_test(dentry, sigen))
8620 +               return 0;
8621 +
8622 +       parent = dget_parent(dentry);
8623 +       di_read_lock_parent(parent, AuLock_IR);
8624 +       AuDebugOn(au_digen_test(parent, sigen));
8625 +       au_dbg_verify_gen(parent, sigen);
8626 +       err = au_refresh_dentry(dentry, parent);
8627 +       di_read_unlock(parent, AuLock_IR);
8628 +       dput(parent);
8629 +       AuTraceErr(err);
8630 +       return err;
8631 +}
8632 +
8633 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen)
8634 +{
8635 +       int err;
8636 +       struct dentry *d, *parent;
8637 +
8638 +       if (!au_ftest_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR))
8639 +               return simple_reval_dpath(dentry, sigen);
8640 +
8641 +       /* slow loop, keep it simple and stupid */
8642 +       /* cf: au_cpup_dirs() */
8643 +       err = 0;
8644 +       parent = NULL;
8645 +       while (au_digen_test(dentry, sigen)) {
8646 +               d = dentry;
8647 +               while (1) {
8648 +                       dput(parent);
8649 +                       parent = dget_parent(d);
8650 +                       if (!au_digen_test(parent, sigen))
8651 +                               break;
8652 +                       d = parent;
8653 +               }
8654 +
8655 +               if (d != dentry)
8656 +                       di_write_lock_child2(d);
8657 +
8658 +               /* someone might update our dentry while we were sleeping */
8659 +               if (au_digen_test(d, sigen)) {
8660 +                       /*
8661 +                        * todo: consolidate with simple_reval_dpath(),
8662 +                        * do_refresh() and au_reval_for_attr().
8663 +                        */
8664 +                       di_read_lock_parent(parent, AuLock_IR);
8665 +                       err = au_refresh_dentry(d, parent);
8666 +                       di_read_unlock(parent, AuLock_IR);
8667 +               }
8668 +
8669 +               if (d != dentry)
8670 +                       di_write_unlock(d);
8671 +               dput(parent);
8672 +               if (unlikely(err))
8673 +                       break;
8674 +       }
8675 +
8676 +       return err;
8677 +}
8678 +
8679 +/*
8680 + * if valid returns 1, otherwise 0.
8681 + */
8682 +static int aufs_d_revalidate(struct dentry *dentry, unsigned int flags)
8683 +{
8684 +       int valid, err;
8685 +       unsigned int sigen;
8686 +       unsigned char do_udba;
8687 +       struct super_block *sb;
8688 +       struct inode *inode;
8689 +
8690 +       /* todo: support rcu-walk? */
8691 +       if (flags & LOOKUP_RCU)
8692 +               return -ECHILD;
8693 +
8694 +       valid = 0;
8695 +       if (unlikely(!au_di(dentry)))
8696 +               goto out;
8697 +
8698 +       valid = 1;
8699 +       sb = dentry->d_sb;
8700 +       /*
8701 +        * todo: very ugly
8702 +        * i_mutex of parent dir may be held,
8703 +        * but we should not return 'invalid' due to busy.
8704 +        */
8705 +       err = aufs_read_lock(dentry, AuLock_FLUSH | AuLock_DW | AuLock_NOPLM);
8706 +       if (unlikely(err)) {
8707 +               valid = err;
8708 +               AuTraceErr(err);
8709 +               goto out;
8710 +       }
8711 +       inode = NULL;
8712 +       if (d_really_is_positive(dentry))
8713 +               inode = d_inode(dentry);
8714 +       if (unlikely(inode && au_is_bad_inode(inode))) {
8715 +               err = -EINVAL;
8716 +               AuTraceErr(err);
8717 +               goto out_dgrade;
8718 +       }
8719 +       if (unlikely(au_dbrange_test(dentry))) {
8720 +               err = -EINVAL;
8721 +               AuTraceErr(err);
8722 +               goto out_dgrade;
8723 +       }
8724 +
8725 +       sigen = au_sigen(sb);
8726 +       if (au_digen_test(dentry, sigen)) {
8727 +               AuDebugOn(IS_ROOT(dentry));
8728 +               err = au_reval_dpath(dentry, sigen);
8729 +               if (unlikely(err)) {
8730 +                       AuTraceErr(err);
8731 +                       goto out_dgrade;
8732 +               }
8733 +       }
8734 +       di_downgrade_lock(dentry, AuLock_IR);
8735 +
8736 +       err = -EINVAL;
8737 +       if (!(flags & (LOOKUP_OPEN | LOOKUP_EMPTY))
8738 +           && inode
8739 +           && !(inode->i_state && I_LINKABLE)
8740 +           && (IS_DEADDIR(inode) || !inode->i_nlink)) {
8741 +               AuTraceErr(err);
8742 +               goto out_inval;
8743 +       }
8744 +
8745 +       do_udba = !au_opt_test(au_mntflags(sb), UDBA_NONE);
8746 +       if (do_udba && inode) {
8747 +               aufs_bindex_t btop = au_ibtop(inode);
8748 +               struct inode *h_inode;
8749 +
8750 +               if (btop >= 0) {
8751 +                       h_inode = au_h_iptr(inode, btop);
8752 +                       if (h_inode && au_test_higen(inode, h_inode)) {
8753 +                               AuTraceErr(err);
8754 +                               goto out_inval;
8755 +                       }
8756 +               }
8757 +       }
8758 +
8759 +       err = h_d_revalidate(dentry, inode, flags, do_udba);
8760 +       if (unlikely(!err && do_udba && au_dbtop(dentry) < 0)) {
8761 +               err = -EIO;
8762 +               AuDbg("both of real entry and whiteout found, %p, err %d\n",
8763 +                     dentry, err);
8764 +       }
8765 +       goto out_inval;
8766 +
8767 +out_dgrade:
8768 +       di_downgrade_lock(dentry, AuLock_IR);
8769 +out_inval:
8770 +       aufs_read_unlock(dentry, AuLock_IR);
8771 +       AuTraceErr(err);
8772 +       valid = !err;
8773 +out:
8774 +       if (!valid) {
8775 +               AuDbg("%pd invalid, %d\n", dentry, valid);
8776 +               d_drop(dentry);
8777 +       }
8778 +       return valid;
8779 +}
8780 +
8781 +static void aufs_d_release(struct dentry *dentry)
8782 +{
8783 +       if (au_di(dentry)) {
8784 +               au_di_fin(dentry);
8785 +               au_hn_di_reinit(dentry);
8786 +       }
8787 +}
8788 +
8789 +const struct dentry_operations aufs_dop = {
8790 +       .d_revalidate           = aufs_d_revalidate,
8791 +       .d_weak_revalidate      = aufs_d_revalidate,
8792 +       .d_release              = aufs_d_release
8793 +};
8794 +
8795 +/* aufs_dop without d_revalidate */
8796 +const struct dentry_operations aufs_dop_noreval = {
8797 +       .d_release              = aufs_d_release
8798 +};
8799 diff -urN /usr/share/empty/fs/aufs/dentry.h linux/fs/aufs/dentry.h
8800 --- /usr/share/empty/fs/aufs/dentry.h   1970-01-01 01:00:00.000000000 +0100
8801 +++ linux/fs/aufs/dentry.h      2016-10-09 16:55:38.889431135 +0200
8802 @@ -0,0 +1,255 @@
8803 +/*
8804 + * Copyright (C) 2005-2016 Junjiro R. Okajima
8805 + *
8806 + * This program, aufs is free software; you can redistribute it and/or modify
8807 + * it under the terms of the GNU General Public License as published by
8808 + * the Free Software Foundation; either version 2 of the License, or
8809 + * (at your option) any later version.
8810 + *
8811 + * This program is distributed in the hope that it will be useful,
8812 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8813 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8814 + * GNU General Public License for more details.
8815 + *
8816 + * You should have received a copy of the GNU General Public License
8817 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8818 + */
8819 +
8820 +/*
8821 + * lookup and dentry operations
8822 + */
8823 +
8824 +#ifndef __AUFS_DENTRY_H__
8825 +#define __AUFS_DENTRY_H__
8826 +
8827 +#ifdef __KERNEL__
8828 +
8829 +#include <linux/dcache.h>
8830 +#include "rwsem.h"
8831 +
8832 +struct au_hdentry {
8833 +       struct dentry           *hd_dentry;
8834 +       aufs_bindex_t           hd_id;
8835 +};
8836 +
8837 +struct au_dinfo {
8838 +       atomic_t                di_generation;
8839 +
8840 +       struct au_rwsem         di_rwsem;
8841 +       aufs_bindex_t           di_btop, di_bbot, di_bwh, di_bdiropq;
8842 +       unsigned char           di_tmpfile; /* to allow the different name */
8843 +       union {
8844 +               struct au_hdentry       *di_hdentry;
8845 +               struct llist_node       di_lnode;       /* delayed free */
8846 +       };
8847 +} ____cacheline_aligned_in_smp;
8848 +
8849 +/* ---------------------------------------------------------------------- */
8850 +
8851 +/* flags for au_lkup_dentry() */
8852 +#define AuLkup_ALLOW_NEG       1
8853 +#define AuLkup_IGNORE_PERM     (1 << 1)
8854 +#define au_ftest_lkup(flags, name)     ((flags) & AuLkup_##name)
8855 +#define au_fset_lkup(flags, name) \
8856 +       do { (flags) |= AuLkup_##name; } while (0)
8857 +#define au_fclr_lkup(flags, name) \
8858 +       do { (flags) &= ~AuLkup_##name; } while (0)
8859 +
8860 +/* ---------------------------------------------------------------------- */
8861 +
8862 +/* dentry.c */
8863 +extern const struct dentry_operations aufs_dop, aufs_dop_noreval;
8864 +struct au_branch;
8865 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent);
8866 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
8867 +               struct dentry *h_parent, struct au_branch *br);
8868 +
8869 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t btop,
8870 +                  unsigned int flags);
8871 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh);
8872 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent);
8873 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen);
8874 +void au_refresh_dop(struct dentry *dentry, int force_reval);
8875 +
8876 +/* dinfo.c */
8877 +void au_di_init_once(void *_di);
8878 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc);
8879 +void au_di_free(struct au_dinfo *dinfo);
8880 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b);
8881 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src);
8882 +int au_di_init(struct dentry *dentry);
8883 +void au_di_fin(struct dentry *dentry);
8884 +int au_di_realloc(struct au_dinfo *dinfo, int nbr, int may_shrink);
8885 +
8886 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc);
8887 +void di_read_unlock(struct dentry *d, int flags);
8888 +void di_downgrade_lock(struct dentry *d, int flags);
8889 +void di_write_lock(struct dentry *d, unsigned int lsc);
8890 +void di_write_unlock(struct dentry *d);
8891 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir);
8892 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir);
8893 +void di_write_unlock2(struct dentry *d1, struct dentry *d2);
8894 +
8895 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex);
8896 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex);
8897 +aufs_bindex_t au_dbtail(struct dentry *dentry);
8898 +aufs_bindex_t au_dbtaildir(struct dentry *dentry);
8899 +
8900 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
8901 +                  struct dentry *h_dentry);
8902 +int au_digen_test(struct dentry *dentry, unsigned int sigen);
8903 +int au_dbrange_test(struct dentry *dentry);
8904 +void au_update_digen(struct dentry *dentry);
8905 +void au_update_dbrange(struct dentry *dentry, int do_put_zero);
8906 +void au_update_dbtop(struct dentry *dentry);
8907 +void au_update_dbbot(struct dentry *dentry);
8908 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry);
8909 +
8910 +/* ---------------------------------------------------------------------- */
8911 +
8912 +static inline struct au_dinfo *au_di(struct dentry *dentry)
8913 +{
8914 +       return dentry->d_fsdata;
8915 +}
8916 +
8917 +/* ---------------------------------------------------------------------- */
8918 +
8919 +/* lock subclass for dinfo */
8920 +enum {
8921 +       AuLsc_DI_CHILD,         /* child first */
8922 +       AuLsc_DI_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
8923 +       AuLsc_DI_CHILD3,        /* copyup dirs */
8924 +       AuLsc_DI_PARENT,
8925 +       AuLsc_DI_PARENT2,
8926 +       AuLsc_DI_PARENT3,
8927 +       AuLsc_DI_TMP            /* temp for replacing dinfo */
8928 +};
8929 +
8930 +/*
8931 + * di_read_lock_child, di_write_lock_child,
8932 + * di_read_lock_child2, di_write_lock_child2,
8933 + * di_read_lock_child3, di_write_lock_child3,
8934 + * di_read_lock_parent, di_write_lock_parent,
8935 + * di_read_lock_parent2, di_write_lock_parent2,
8936 + * di_read_lock_parent3, di_write_lock_parent3,
8937 + */
8938 +#define AuReadLockFunc(name, lsc) \
8939 +static inline void di_read_lock_##name(struct dentry *d, int flags) \
8940 +{ di_read_lock(d, flags, AuLsc_DI_##lsc); }
8941 +
8942 +#define AuWriteLockFunc(name, lsc) \
8943 +static inline void di_write_lock_##name(struct dentry *d) \
8944 +{ di_write_lock(d, AuLsc_DI_##lsc); }
8945 +
8946 +#define AuRWLockFuncs(name, lsc) \
8947 +       AuReadLockFunc(name, lsc) \
8948 +       AuWriteLockFunc(name, lsc)
8949 +
8950 +AuRWLockFuncs(child, CHILD);
8951 +AuRWLockFuncs(child2, CHILD2);
8952 +AuRWLockFuncs(child3, CHILD3);
8953 +AuRWLockFuncs(parent, PARENT);
8954 +AuRWLockFuncs(parent2, PARENT2);
8955 +AuRWLockFuncs(parent3, PARENT3);
8956 +
8957 +#undef AuReadLockFunc
8958 +#undef AuWriteLockFunc
8959 +#undef AuRWLockFuncs
8960 +
8961 +#define DiMustNoWaiters(d)     AuRwMustNoWaiters(&au_di(d)->di_rwsem)
8962 +#define DiMustAnyLock(d)       AuRwMustAnyLock(&au_di(d)->di_rwsem)
8963 +#define DiMustWriteLock(d)     AuRwMustWriteLock(&au_di(d)->di_rwsem)
8964 +
8965 +/* ---------------------------------------------------------------------- */
8966 +
8967 +/* todo: memory barrier? */
8968 +static inline unsigned int au_digen(struct dentry *d)
8969 +{
8970 +       return atomic_read(&au_di(d)->di_generation);
8971 +}
8972 +
8973 +static inline void au_h_dentry_init(struct au_hdentry *hdentry)
8974 +{
8975 +       hdentry->hd_dentry = NULL;
8976 +}
8977 +
8978 +static inline struct au_hdentry *au_hdentry(struct au_dinfo *di,
8979 +                                           aufs_bindex_t bindex)
8980 +{
8981 +       return di->di_hdentry + bindex;
8982 +}
8983 +
8984 +static inline void au_hdput(struct au_hdentry *hd)
8985 +{
8986 +       if (hd)
8987 +               dput(hd->hd_dentry);
8988 +}
8989 +
8990 +static inline aufs_bindex_t au_dbtop(struct dentry *dentry)
8991 +{
8992 +       DiMustAnyLock(dentry);
8993 +       return au_di(dentry)->di_btop;
8994 +}
8995 +
8996 +static inline aufs_bindex_t au_dbbot(struct dentry *dentry)
8997 +{
8998 +       DiMustAnyLock(dentry);
8999 +       return au_di(dentry)->di_bbot;
9000 +}
9001 +
9002 +static inline aufs_bindex_t au_dbwh(struct dentry *dentry)
9003 +{
9004 +       DiMustAnyLock(dentry);
9005 +       return au_di(dentry)->di_bwh;
9006 +}
9007 +
9008 +static inline aufs_bindex_t au_dbdiropq(struct dentry *dentry)
9009 +{
9010 +       DiMustAnyLock(dentry);
9011 +       return au_di(dentry)->di_bdiropq;
9012 +}
9013 +
9014 +/* todo: hard/soft set? */
9015 +static inline void au_set_dbtop(struct dentry *dentry, aufs_bindex_t bindex)
9016 +{
9017 +       DiMustWriteLock(dentry);
9018 +       au_di(dentry)->di_btop = bindex;
9019 +}
9020 +
9021 +static inline void au_set_dbbot(struct dentry *dentry, aufs_bindex_t bindex)
9022 +{
9023 +       DiMustWriteLock(dentry);
9024 +       au_di(dentry)->di_bbot = bindex;
9025 +}
9026 +
9027 +static inline void au_set_dbwh(struct dentry *dentry, aufs_bindex_t bindex)
9028 +{
9029 +       DiMustWriteLock(dentry);
9030 +       /* dbwh can be outside of btop - bbot range */
9031 +       au_di(dentry)->di_bwh = bindex;
9032 +}
9033 +
9034 +static inline void au_set_dbdiropq(struct dentry *dentry, aufs_bindex_t bindex)
9035 +{
9036 +       DiMustWriteLock(dentry);
9037 +       au_di(dentry)->di_bdiropq = bindex;
9038 +}
9039 +
9040 +/* ---------------------------------------------------------------------- */
9041 +
9042 +#ifdef CONFIG_AUFS_HNOTIFY
9043 +static inline void au_digen_dec(struct dentry *d)
9044 +{
9045 +       atomic_dec(&au_di(d)->di_generation);
9046 +}
9047 +
9048 +static inline void au_hn_di_reinit(struct dentry *dentry)
9049 +{
9050 +       dentry->d_fsdata = NULL;
9051 +}
9052 +#else
9053 +AuStubVoid(au_hn_di_reinit, struct dentry *dentry __maybe_unused)
9054 +#endif /* CONFIG_AUFS_HNOTIFY */
9055 +
9056 +#endif /* __KERNEL__ */
9057 +#endif /* __AUFS_DENTRY_H__ */
9058 diff -urN /usr/share/empty/fs/aufs/dinfo.c linux/fs/aufs/dinfo.c
9059 --- /usr/share/empty/fs/aufs/dinfo.c    1970-01-01 01:00:00.000000000 +0100
9060 +++ linux/fs/aufs/dinfo.c       2016-10-09 16:55:38.889431135 +0200
9061 @@ -0,0 +1,553 @@
9062 +/*
9063 + * Copyright (C) 2005-2016 Junjiro R. Okajima
9064 + *
9065 + * This program, aufs is free software; you can redistribute it and/or modify
9066 + * it under the terms of the GNU General Public License as published by
9067 + * the Free Software Foundation; either version 2 of the License, or
9068 + * (at your option) any later version.
9069 + *
9070 + * This program is distributed in the hope that it will be useful,
9071 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9072 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9073 + * GNU General Public License for more details.
9074 + *
9075 + * You should have received a copy of the GNU General Public License
9076 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9077 + */
9078 +
9079 +/*
9080 + * dentry private data
9081 + */
9082 +
9083 +#include "aufs.h"
9084 +
9085 +void au_di_init_once(void *_dinfo)
9086 +{
9087 +       struct au_dinfo *dinfo = _dinfo;
9088 +
9089 +       au_rw_init(&dinfo->di_rwsem);
9090 +}
9091 +
9092 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc)
9093 +{
9094 +       struct au_dinfo *dinfo;
9095 +       int nbr, i;
9096 +
9097 +       dinfo = au_cache_alloc_dinfo();
9098 +       if (unlikely(!dinfo))
9099 +               goto out;
9100 +
9101 +       nbr = au_sbbot(sb) + 1;
9102 +       if (nbr <= 0)
9103 +               nbr = 1;
9104 +       dinfo->di_hdentry = kcalloc(nbr, sizeof(*dinfo->di_hdentry), GFP_NOFS);
9105 +       if (dinfo->di_hdentry) {
9106 +               au_rw_write_lock_nested(&dinfo->di_rwsem, lsc);
9107 +               dinfo->di_btop = -1;
9108 +               dinfo->di_bbot = -1;
9109 +               dinfo->di_bwh = -1;
9110 +               dinfo->di_bdiropq = -1;
9111 +               dinfo->di_tmpfile = 0;
9112 +               for (i = 0; i < nbr; i++)
9113 +                       dinfo->di_hdentry[i].hd_id = -1;
9114 +               goto out;
9115 +       }
9116 +
9117 +       au_cache_dfree_dinfo(dinfo);
9118 +       dinfo = NULL;
9119 +
9120 +out:
9121 +       return dinfo;
9122 +}
9123 +
9124 +void au_di_free(struct au_dinfo *dinfo)
9125 +{
9126 +       struct au_hdentry *p;
9127 +       aufs_bindex_t bbot, bindex;
9128 +
9129 +       /* dentry may not be revalidated */
9130 +       bindex = dinfo->di_btop;
9131 +       if (bindex >= 0) {
9132 +               bbot = dinfo->di_bbot;
9133 +               p = au_hdentry(dinfo, bindex);
9134 +               while (bindex++ <= bbot)
9135 +                       au_hdput(p++);
9136 +       }
9137 +       au_delayed_kfree(dinfo->di_hdentry);
9138 +       au_cache_dfree_dinfo(dinfo);
9139 +}
9140 +
9141 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b)
9142 +{
9143 +       struct au_hdentry *p;
9144 +       aufs_bindex_t bi;
9145 +
9146 +       AuRwMustWriteLock(&a->di_rwsem);
9147 +       AuRwMustWriteLock(&b->di_rwsem);
9148 +
9149 +#define DiSwap(v, name)                                \
9150 +       do {                                    \
9151 +               v = a->di_##name;               \
9152 +               a->di_##name = b->di_##name;    \
9153 +               b->di_##name = v;               \
9154 +       } while (0)
9155 +
9156 +       DiSwap(p, hdentry);
9157 +       DiSwap(bi, btop);
9158 +       DiSwap(bi, bbot);
9159 +       DiSwap(bi, bwh);
9160 +       DiSwap(bi, bdiropq);
9161 +       /* smp_mb(); */
9162 +
9163 +#undef DiSwap
9164 +}
9165 +
9166 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src)
9167 +{
9168 +       AuRwMustWriteLock(&dst->di_rwsem);
9169 +       AuRwMustWriteLock(&src->di_rwsem);
9170 +
9171 +       dst->di_btop = src->di_btop;
9172 +       dst->di_bbot = src->di_bbot;
9173 +       dst->di_bwh = src->di_bwh;
9174 +       dst->di_bdiropq = src->di_bdiropq;
9175 +       /* smp_mb(); */
9176 +}
9177 +
9178 +int au_di_init(struct dentry *dentry)
9179 +{
9180 +       int err;
9181 +       struct super_block *sb;
9182 +       struct au_dinfo *dinfo;
9183 +
9184 +       err = 0;
9185 +       sb = dentry->d_sb;
9186 +       dinfo = au_di_alloc(sb, AuLsc_DI_CHILD);
9187 +       if (dinfo) {
9188 +               atomic_set(&dinfo->di_generation, au_sigen(sb));
9189 +               /* smp_mb(); */ /* atomic_set */
9190 +               dentry->d_fsdata = dinfo;
9191 +       } else
9192 +               err = -ENOMEM;
9193 +
9194 +       return err;
9195 +}
9196 +
9197 +void au_di_fin(struct dentry *dentry)
9198 +{
9199 +       struct au_dinfo *dinfo;
9200 +
9201 +       dinfo = au_di(dentry);
9202 +       AuRwDestroy(&dinfo->di_rwsem);
9203 +       au_di_free(dinfo);
9204 +}
9205 +
9206 +int au_di_realloc(struct au_dinfo *dinfo, int nbr, int may_shrink)
9207 +{
9208 +       int err, sz;
9209 +       struct au_hdentry *hdp;
9210 +
9211 +       AuRwMustWriteLock(&dinfo->di_rwsem);
9212 +
9213 +       err = -ENOMEM;
9214 +       sz = sizeof(*hdp) * (dinfo->di_bbot + 1);
9215 +       if (!sz)
9216 +               sz = sizeof(*hdp);
9217 +       hdp = au_kzrealloc(dinfo->di_hdentry, sz, sizeof(*hdp) * nbr, GFP_NOFS,
9218 +                          may_shrink);
9219 +       if (hdp) {
9220 +               dinfo->di_hdentry = hdp;
9221 +               err = 0;
9222 +       }
9223 +
9224 +       return err;
9225 +}
9226 +
9227 +/* ---------------------------------------------------------------------- */
9228 +
9229 +static void do_ii_write_lock(struct inode *inode, unsigned int lsc)
9230 +{
9231 +       switch (lsc) {
9232 +       case AuLsc_DI_CHILD:
9233 +               ii_write_lock_child(inode);
9234 +               break;
9235 +       case AuLsc_DI_CHILD2:
9236 +               ii_write_lock_child2(inode);
9237 +               break;
9238 +       case AuLsc_DI_CHILD3:
9239 +               ii_write_lock_child3(inode);
9240 +               break;
9241 +       case AuLsc_DI_PARENT:
9242 +               ii_write_lock_parent(inode);
9243 +               break;
9244 +       case AuLsc_DI_PARENT2:
9245 +               ii_write_lock_parent2(inode);
9246 +               break;
9247 +       case AuLsc_DI_PARENT3:
9248 +               ii_write_lock_parent3(inode);
9249 +               break;
9250 +       default:
9251 +               BUG();
9252 +       }
9253 +}
9254 +
9255 +static void do_ii_read_lock(struct inode *inode, unsigned int lsc)
9256 +{
9257 +       switch (lsc) {
9258 +       case AuLsc_DI_CHILD:
9259 +               ii_read_lock_child(inode);
9260 +               break;
9261 +       case AuLsc_DI_CHILD2:
9262 +               ii_read_lock_child2(inode);
9263 +               break;
9264 +       case AuLsc_DI_CHILD3:
9265 +               ii_read_lock_child3(inode);
9266 +               break;
9267 +       case AuLsc_DI_PARENT:
9268 +               ii_read_lock_parent(inode);
9269 +               break;
9270 +       case AuLsc_DI_PARENT2:
9271 +               ii_read_lock_parent2(inode);
9272 +               break;
9273 +       case AuLsc_DI_PARENT3:
9274 +               ii_read_lock_parent3(inode);
9275 +               break;
9276 +       default:
9277 +               BUG();
9278 +       }
9279 +}
9280 +
9281 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc)
9282 +{
9283 +       struct inode *inode;
9284 +
9285 +       au_rw_read_lock_nested(&au_di(d)->di_rwsem, lsc);
9286 +       if (d_really_is_positive(d)) {
9287 +               inode = d_inode(d);
9288 +               if (au_ftest_lock(flags, IW))
9289 +                       do_ii_write_lock(inode, lsc);
9290 +               else if (au_ftest_lock(flags, IR))
9291 +                       do_ii_read_lock(inode, lsc);
9292 +       }
9293 +}
9294 +
9295 +void di_read_unlock(struct dentry *d, int flags)
9296 +{
9297 +       struct inode *inode;
9298 +
9299 +       if (d_really_is_positive(d)) {
9300 +               inode = d_inode(d);
9301 +               if (au_ftest_lock(flags, IW)) {
9302 +                       au_dbg_verify_dinode(d);
9303 +                       ii_write_unlock(inode);
9304 +               } else if (au_ftest_lock(flags, IR)) {
9305 +                       au_dbg_verify_dinode(d);
9306 +                       ii_read_unlock(inode);
9307 +               }
9308 +       }
9309 +       au_rw_read_unlock(&au_di(d)->di_rwsem);
9310 +}
9311 +
9312 +void di_downgrade_lock(struct dentry *d, int flags)
9313 +{
9314 +       if (d_really_is_positive(d) && au_ftest_lock(flags, IR))
9315 +               ii_downgrade_lock(d_inode(d));
9316 +       au_rw_dgrade_lock(&au_di(d)->di_rwsem);
9317 +}
9318 +
9319 +void di_write_lock(struct dentry *d, unsigned int lsc)
9320 +{
9321 +       au_rw_write_lock_nested(&au_di(d)->di_rwsem, lsc);
9322 +       if (d_really_is_positive(d))
9323 +               do_ii_write_lock(d_inode(d), lsc);
9324 +}
9325 +
9326 +void di_write_unlock(struct dentry *d)
9327 +{
9328 +       au_dbg_verify_dinode(d);
9329 +       if (d_really_is_positive(d))
9330 +               ii_write_unlock(d_inode(d));
9331 +       au_rw_write_unlock(&au_di(d)->di_rwsem);
9332 +}
9333 +
9334 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir)
9335 +{
9336 +       AuDebugOn(d1 == d2
9337 +                 || d_inode(d1) == d_inode(d2)
9338 +                 || d1->d_sb != d2->d_sb);
9339 +
9340 +       if (isdir && au_test_subdir(d1, d2)) {
9341 +               di_write_lock_child(d1);
9342 +               di_write_lock_child2(d2);
9343 +       } else {
9344 +               /* there should be no races */
9345 +               di_write_lock_child(d2);
9346 +               di_write_lock_child2(d1);
9347 +       }
9348 +}
9349 +
9350 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir)
9351 +{
9352 +       AuDebugOn(d1 == d2
9353 +                 || d_inode(d1) == d_inode(d2)
9354 +                 || d1->d_sb != d2->d_sb);
9355 +
9356 +       if (isdir && au_test_subdir(d1, d2)) {
9357 +               di_write_lock_parent(d1);
9358 +               di_write_lock_parent2(d2);
9359 +       } else {
9360 +               /* there should be no races */
9361 +               di_write_lock_parent(d2);
9362 +               di_write_lock_parent2(d1);
9363 +       }
9364 +}
9365 +
9366 +void di_write_unlock2(struct dentry *d1, struct dentry *d2)
9367 +{
9368 +       di_write_unlock(d1);
9369 +       if (d_inode(d1) == d_inode(d2))
9370 +               au_rw_write_unlock(&au_di(d2)->di_rwsem);
9371 +       else
9372 +               di_write_unlock(d2);
9373 +}
9374 +
9375 +/* ---------------------------------------------------------------------- */
9376 +
9377 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex)
9378 +{
9379 +       struct dentry *d;
9380 +
9381 +       DiMustAnyLock(dentry);
9382 +
9383 +       if (au_dbtop(dentry) < 0 || bindex < au_dbtop(dentry))
9384 +               return NULL;
9385 +       AuDebugOn(bindex < 0);
9386 +       d = au_hdentry(au_di(dentry), bindex)->hd_dentry;
9387 +       AuDebugOn(d && au_dcount(d) <= 0);
9388 +       return d;
9389 +}
9390 +
9391 +/*
9392 + * extended version of au_h_dptr().
9393 + * returns a hashed and positive (or linkable) h_dentry in bindex, NULL, or
9394 + * error.
9395 + */
9396 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex)
9397 +{
9398 +       struct dentry *h_dentry;
9399 +       struct inode *inode, *h_inode;
9400 +
9401 +       AuDebugOn(d_really_is_negative(dentry));
9402 +
9403 +       h_dentry = NULL;
9404 +       if (au_dbtop(dentry) <= bindex
9405 +           && bindex <= au_dbbot(dentry))
9406 +               h_dentry = au_h_dptr(dentry, bindex);
9407 +       if (h_dentry && !au_d_linkable(h_dentry)) {
9408 +               dget(h_dentry);
9409 +               goto out; /* success */
9410 +       }
9411 +
9412 +       inode = d_inode(dentry);
9413 +       AuDebugOn(bindex < au_ibtop(inode));
9414 +       AuDebugOn(au_ibbot(inode) < bindex);
9415 +       h_inode = au_h_iptr(inode, bindex);
9416 +       h_dentry = d_find_alias(h_inode);
9417 +       if (h_dentry) {
9418 +               if (!IS_ERR(h_dentry)) {
9419 +                       if (!au_d_linkable(h_dentry))
9420 +                               goto out; /* success */
9421 +                       dput(h_dentry);
9422 +               } else
9423 +                       goto out;
9424 +       }
9425 +
9426 +       if (au_opt_test(au_mntflags(dentry->d_sb), PLINK)) {
9427 +               h_dentry = au_plink_lkup(inode, bindex);
9428 +               AuDebugOn(!h_dentry);
9429 +               if (!IS_ERR(h_dentry)) {
9430 +                       if (!au_d_hashed_positive(h_dentry))
9431 +                               goto out; /* success */
9432 +                       dput(h_dentry);
9433 +                       h_dentry = NULL;
9434 +               }
9435 +       }
9436 +
9437 +out:
9438 +       AuDbgDentry(h_dentry);
9439 +       return h_dentry;
9440 +}
9441 +
9442 +aufs_bindex_t au_dbtail(struct dentry *dentry)
9443 +{
9444 +       aufs_bindex_t bbot, bwh;
9445 +
9446 +       bbot = au_dbbot(dentry);
9447 +       if (0 <= bbot) {
9448 +               bwh = au_dbwh(dentry);
9449 +               if (!bwh)
9450 +                       return bwh;
9451 +               if (0 < bwh && bwh < bbot)
9452 +                       return bwh - 1;
9453 +       }
9454 +       return bbot;
9455 +}
9456 +
9457 +aufs_bindex_t au_dbtaildir(struct dentry *dentry)
9458 +{
9459 +       aufs_bindex_t bbot, bopq;
9460 +
9461 +       bbot = au_dbtail(dentry);
9462 +       if (0 <= bbot) {
9463 +               bopq = au_dbdiropq(dentry);
9464 +               if (0 <= bopq && bopq < bbot)
9465 +                       bbot = bopq;
9466 +       }
9467 +       return bbot;
9468 +}
9469 +
9470 +/* ---------------------------------------------------------------------- */
9471 +
9472 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
9473 +                  struct dentry *h_dentry)
9474 +{
9475 +       struct au_dinfo *dinfo;
9476 +       struct au_hdentry *hd;
9477 +       struct au_branch *br;
9478 +
9479 +       DiMustWriteLock(dentry);
9480 +
9481 +       dinfo = au_di(dentry);
9482 +       hd = au_hdentry(dinfo, bindex);
9483 +       au_hdput(hd);
9484 +       hd->hd_dentry = h_dentry;
9485 +       if (h_dentry) {
9486 +               br = au_sbr(dentry->d_sb, bindex);
9487 +               hd->hd_id = br->br_id;
9488 +       }
9489 +}
9490 +
9491 +int au_dbrange_test(struct dentry *dentry)
9492 +{
9493 +       int err;
9494 +       aufs_bindex_t btop, bbot;
9495 +
9496 +       err = 0;
9497 +       btop = au_dbtop(dentry);
9498 +       bbot = au_dbbot(dentry);
9499 +       if (btop >= 0)
9500 +               AuDebugOn(bbot < 0 && btop > bbot);
9501 +       else {
9502 +               err = -EIO;
9503 +               AuDebugOn(bbot >= 0);
9504 +       }
9505 +
9506 +       return err;
9507 +}
9508 +
9509 +int au_digen_test(struct dentry *dentry, unsigned int sigen)
9510 +{
9511 +       int err;
9512 +
9513 +       err = 0;
9514 +       if (unlikely(au_digen(dentry) != sigen
9515 +                    || au_iigen_test(d_inode(dentry), sigen)))
9516 +               err = -EIO;
9517 +
9518 +       return err;
9519 +}
9520 +
9521 +void au_update_digen(struct dentry *dentry)
9522 +{
9523 +       atomic_set(&au_di(dentry)->di_generation, au_sigen(dentry->d_sb));
9524 +       /* smp_mb(); */ /* atomic_set */
9525 +}
9526 +
9527 +void au_update_dbrange(struct dentry *dentry, int do_put_zero)
9528 +{
9529 +       struct au_dinfo *dinfo;
9530 +       struct dentry *h_d;
9531 +       struct au_hdentry *hdp;
9532 +       aufs_bindex_t bindex, bbot;
9533 +
9534 +       DiMustWriteLock(dentry);
9535 +
9536 +       dinfo = au_di(dentry);
9537 +       if (!dinfo || dinfo->di_btop < 0)
9538 +               return;
9539 +
9540 +       if (do_put_zero) {
9541 +               bbot = dinfo->di_bbot;
9542 +               bindex = dinfo->di_btop;
9543 +               hdp = au_hdentry(dinfo, bindex);
9544 +               for (; bindex <= bbot; bindex++, hdp++) {
9545 +                       h_d = hdp->hd_dentry;
9546 +                       if (h_d && d_is_negative(h_d))
9547 +                               au_set_h_dptr(dentry, bindex, NULL);
9548 +               }
9549 +       }
9550 +
9551 +       dinfo->di_btop = 0;
9552 +       hdp = au_hdentry(dinfo, dinfo->di_btop);
9553 +       for (; dinfo->di_btop <= dinfo->di_bbot; dinfo->di_btop++, hdp++)
9554 +               if (hdp->hd_dentry)
9555 +                       break;
9556 +       if (dinfo->di_btop > dinfo->di_bbot) {
9557 +               dinfo->di_btop = -1;
9558 +               dinfo->di_bbot = -1;
9559 +               return;
9560 +       }
9561 +
9562 +       hdp = au_hdentry(dinfo, dinfo->di_bbot);
9563 +       for (; dinfo->di_bbot >= 0; dinfo->di_bbot--, hdp--)
9564 +               if (hdp->hd_dentry)
9565 +                       break;
9566 +       AuDebugOn(dinfo->di_btop > dinfo->di_bbot || dinfo->di_bbot < 0);
9567 +}
9568 +
9569 +void au_update_dbtop(struct dentry *dentry)
9570 +{
9571 +       aufs_bindex_t bindex, bbot;
9572 +       struct dentry *h_dentry;
9573 +
9574 +       bbot = au_dbbot(dentry);
9575 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++) {
9576 +               h_dentry = au_h_dptr(dentry, bindex);
9577 +               if (!h_dentry)
9578 +                       continue;
9579 +               if (d_is_positive(h_dentry)) {
9580 +                       au_set_dbtop(dentry, bindex);
9581 +                       return;
9582 +               }
9583 +               au_set_h_dptr(dentry, bindex, NULL);
9584 +       }
9585 +}
9586 +
9587 +void au_update_dbbot(struct dentry *dentry)
9588 +{
9589 +       aufs_bindex_t bindex, btop;
9590 +       struct dentry *h_dentry;
9591 +
9592 +       btop = au_dbtop(dentry);
9593 +       for (bindex = au_dbbot(dentry); bindex >= btop; bindex--) {
9594 +               h_dentry = au_h_dptr(dentry, bindex);
9595 +               if (!h_dentry)
9596 +                       continue;
9597 +               if (d_is_positive(h_dentry)) {
9598 +                       au_set_dbbot(dentry, bindex);
9599 +                       return;
9600 +               }
9601 +               au_set_h_dptr(dentry, bindex, NULL);
9602 +       }
9603 +}
9604 +
9605 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry)
9606 +{
9607 +       aufs_bindex_t bindex, bbot;
9608 +
9609 +       bbot = au_dbbot(dentry);
9610 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++)
9611 +               if (au_h_dptr(dentry, bindex) == h_dentry)
9612 +                       return bindex;
9613 +       return -1;
9614 +}
9615 diff -urN /usr/share/empty/fs/aufs/dir.c linux/fs/aufs/dir.c
9616 --- /usr/share/empty/fs/aufs/dir.c      1970-01-01 01:00:00.000000000 +0100
9617 +++ linux/fs/aufs/dir.c 2016-10-09 16:55:36.489368218 +0200
9618 @@ -0,0 +1,762 @@
9619 +/*
9620 + * Copyright (C) 2005-2016 Junjiro R. Okajima
9621 + *
9622 + * This program, aufs is free software; you can redistribute it and/or modify
9623 + * it under the terms of the GNU General Public License as published by
9624 + * the Free Software Foundation; either version 2 of the License, or
9625 + * (at your option) any later version.
9626 + *
9627 + * This program is distributed in the hope that it will be useful,
9628 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9629 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9630 + * GNU General Public License for more details.
9631 + *
9632 + * You should have received a copy of the GNU General Public License
9633 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9634 + */
9635 +
9636 +/*
9637 + * directory operations
9638 + */
9639 +
9640 +#include <linux/fs_stack.h>
9641 +#include "aufs.h"
9642 +
9643 +void au_add_nlink(struct inode *dir, struct inode *h_dir)
9644 +{
9645 +       unsigned int nlink;
9646 +
9647 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9648 +
9649 +       nlink = dir->i_nlink;
9650 +       nlink += h_dir->i_nlink - 2;
9651 +       if (h_dir->i_nlink < 2)
9652 +               nlink += 2;
9653 +       smp_mb(); /* for i_nlink */
9654 +       /* 0 can happen in revaliding */
9655 +       set_nlink(dir, nlink);
9656 +}
9657 +
9658 +void au_sub_nlink(struct inode *dir, struct inode *h_dir)
9659 +{
9660 +       unsigned int nlink;
9661 +
9662 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9663 +
9664 +       nlink = dir->i_nlink;
9665 +       nlink -= h_dir->i_nlink - 2;
9666 +       if (h_dir->i_nlink < 2)
9667 +               nlink -= 2;
9668 +       smp_mb(); /* for i_nlink */
9669 +       /* nlink == 0 means the branch-fs is broken */
9670 +       set_nlink(dir, nlink);
9671 +}
9672 +
9673 +loff_t au_dir_size(struct file *file, struct dentry *dentry)
9674 +{
9675 +       loff_t sz;
9676 +       aufs_bindex_t bindex, bbot;
9677 +       struct file *h_file;
9678 +       struct dentry *h_dentry;
9679 +
9680 +       sz = 0;
9681 +       if (file) {
9682 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
9683 +
9684 +               bbot = au_fbbot_dir(file);
9685 +               for (bindex = au_fbtop(file);
9686 +                    bindex <= bbot && sz < KMALLOC_MAX_SIZE;
9687 +                    bindex++) {
9688 +                       h_file = au_hf_dir(file, bindex);
9689 +                       if (h_file && file_inode(h_file))
9690 +                               sz += vfsub_f_size_read(h_file);
9691 +               }
9692 +       } else {
9693 +               AuDebugOn(!dentry);
9694 +               AuDebugOn(!d_is_dir(dentry));
9695 +
9696 +               bbot = au_dbtaildir(dentry);
9697 +               for (bindex = au_dbtop(dentry);
9698 +                    bindex <= bbot && sz < KMALLOC_MAX_SIZE;
9699 +                    bindex++) {
9700 +                       h_dentry = au_h_dptr(dentry, bindex);
9701 +                       if (h_dentry && d_is_positive(h_dentry))
9702 +                               sz += i_size_read(d_inode(h_dentry));
9703 +               }
9704 +       }
9705 +       if (sz < KMALLOC_MAX_SIZE)
9706 +               sz = roundup_pow_of_two(sz);
9707 +       if (sz > KMALLOC_MAX_SIZE)
9708 +               sz = KMALLOC_MAX_SIZE;
9709 +       else if (sz < NAME_MAX) {
9710 +               BUILD_BUG_ON(AUFS_RDBLK_DEF < NAME_MAX);
9711 +               sz = AUFS_RDBLK_DEF;
9712 +       }
9713 +       return sz;
9714 +}
9715 +
9716 +struct au_dir_ts_arg {
9717 +       struct dentry *dentry;
9718 +       aufs_bindex_t brid;
9719 +};
9720 +
9721 +static void au_do_dir_ts(void *arg)
9722 +{
9723 +       struct au_dir_ts_arg *a = arg;
9724 +       struct au_dtime dt;
9725 +       struct path h_path;
9726 +       struct inode *dir, *h_dir;
9727 +       struct super_block *sb;
9728 +       struct au_branch *br;
9729 +       struct au_hinode *hdir;
9730 +       int err;
9731 +       aufs_bindex_t btop, bindex;
9732 +
9733 +       sb = a->dentry->d_sb;
9734 +       if (d_really_is_negative(a->dentry))
9735 +               goto out;
9736 +       /* no dir->i_mutex lock */
9737 +       aufs_read_lock(a->dentry, AuLock_DW); /* noflush */
9738 +
9739 +       dir = d_inode(a->dentry);
9740 +       btop = au_ibtop(dir);
9741 +       bindex = au_br_index(sb, a->brid);
9742 +       if (bindex < btop)
9743 +               goto out_unlock;
9744 +
9745 +       br = au_sbr(sb, bindex);
9746 +       h_path.dentry = au_h_dptr(a->dentry, bindex);
9747 +       if (!h_path.dentry)
9748 +               goto out_unlock;
9749 +       h_path.mnt = au_br_mnt(br);
9750 +       au_dtime_store(&dt, a->dentry, &h_path);
9751 +
9752 +       br = au_sbr(sb, btop);
9753 +       if (!au_br_writable(br->br_perm))
9754 +               goto out_unlock;
9755 +       h_path.dentry = au_h_dptr(a->dentry, btop);
9756 +       h_path.mnt = au_br_mnt(br);
9757 +       err = vfsub_mnt_want_write(h_path.mnt);
9758 +       if (err)
9759 +               goto out_unlock;
9760 +       hdir = au_hi(dir, btop);
9761 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
9762 +       h_dir = au_h_iptr(dir, btop);
9763 +       if (h_dir->i_nlink
9764 +           && timespec_compare(&h_dir->i_mtime, &dt.dt_mtime) < 0) {
9765 +               dt.dt_h_path = h_path;
9766 +               au_dtime_revert(&dt);
9767 +       }
9768 +       au_hn_inode_unlock(hdir);
9769 +       vfsub_mnt_drop_write(h_path.mnt);
9770 +       au_cpup_attr_timesizes(dir);
9771 +
9772 +out_unlock:
9773 +       aufs_read_unlock(a->dentry, AuLock_DW);
9774 +out:
9775 +       dput(a->dentry);
9776 +       au_nwt_done(&au_sbi(sb)->si_nowait);
9777 +       au_delayed_kfree(arg);
9778 +}
9779 +
9780 +void au_dir_ts(struct inode *dir, aufs_bindex_t bindex)
9781 +{
9782 +       int perm, wkq_err;
9783 +       aufs_bindex_t btop;
9784 +       struct au_dir_ts_arg *arg;
9785 +       struct dentry *dentry;
9786 +       struct super_block *sb;
9787 +
9788 +       IMustLock(dir);
9789 +
9790 +       dentry = d_find_any_alias(dir);
9791 +       AuDebugOn(!dentry);
9792 +       sb = dentry->d_sb;
9793 +       btop = au_ibtop(dir);
9794 +       if (btop == bindex) {
9795 +               au_cpup_attr_timesizes(dir);
9796 +               goto out;
9797 +       }
9798 +
9799 +       perm = au_sbr_perm(sb, btop);
9800 +       if (!au_br_writable(perm))
9801 +               goto out;
9802 +
9803 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
9804 +       if (!arg)
9805 +               goto out;
9806 +
9807 +       arg->dentry = dget(dentry); /* will be dput-ted by au_do_dir_ts() */
9808 +       arg->brid = au_sbr_id(sb, bindex);
9809 +       wkq_err = au_wkq_nowait(au_do_dir_ts, arg, sb, /*flags*/0);
9810 +       if (unlikely(wkq_err)) {
9811 +               pr_err("wkq %d\n", wkq_err);
9812 +               dput(dentry);
9813 +               au_delayed_kfree(arg);
9814 +       }
9815 +
9816 +out:
9817 +       dput(dentry);
9818 +}
9819 +
9820 +/* ---------------------------------------------------------------------- */
9821 +
9822 +static int reopen_dir(struct file *file)
9823 +{
9824 +       int err;
9825 +       unsigned int flags;
9826 +       aufs_bindex_t bindex, btail, btop;
9827 +       struct dentry *dentry, *h_dentry;
9828 +       struct file *h_file;
9829 +
9830 +       /* open all lower dirs */
9831 +       dentry = file->f_path.dentry;
9832 +       btop = au_dbtop(dentry);
9833 +       for (bindex = au_fbtop(file); bindex < btop; bindex++)
9834 +               au_set_h_fptr(file, bindex, NULL);
9835 +       au_set_fbtop(file, btop);
9836 +
9837 +       btail = au_dbtaildir(dentry);
9838 +       for (bindex = au_fbbot_dir(file); btail < bindex; bindex--)
9839 +               au_set_h_fptr(file, bindex, NULL);
9840 +       au_set_fbbot_dir(file, btail);
9841 +
9842 +       flags = vfsub_file_flags(file);
9843 +       for (bindex = btop; bindex <= btail; bindex++) {
9844 +               h_dentry = au_h_dptr(dentry, bindex);
9845 +               if (!h_dentry)
9846 +                       continue;
9847 +               h_file = au_hf_dir(file, bindex);
9848 +               if (h_file)
9849 +                       continue;
9850 +
9851 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9852 +               err = PTR_ERR(h_file);
9853 +               if (IS_ERR(h_file))
9854 +                       goto out; /* close all? */
9855 +               au_set_h_fptr(file, bindex, h_file);
9856 +       }
9857 +       au_update_figen(file);
9858 +       /* todo: necessary? */
9859 +       /* file->f_ra = h_file->f_ra; */
9860 +       err = 0;
9861 +
9862 +out:
9863 +       return err;
9864 +}
9865 +
9866 +static int do_open_dir(struct file *file, int flags, struct file *h_file)
9867 +{
9868 +       int err;
9869 +       aufs_bindex_t bindex, btail;
9870 +       struct dentry *dentry, *h_dentry;
9871 +       struct vfsmount *mnt;
9872 +
9873 +       FiMustWriteLock(file);
9874 +       AuDebugOn(h_file);
9875 +
9876 +       err = 0;
9877 +       mnt = file->f_path.mnt;
9878 +       dentry = file->f_path.dentry;
9879 +       file->f_version = d_inode(dentry)->i_version;
9880 +       bindex = au_dbtop(dentry);
9881 +       au_set_fbtop(file, bindex);
9882 +       btail = au_dbtaildir(dentry);
9883 +       au_set_fbbot_dir(file, btail);
9884 +       for (; !err && bindex <= btail; bindex++) {
9885 +               h_dentry = au_h_dptr(dentry, bindex);
9886 +               if (!h_dentry)
9887 +                       continue;
9888 +
9889 +               err = vfsub_test_mntns(mnt, h_dentry->d_sb);
9890 +               if (unlikely(err))
9891 +                       break;
9892 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9893 +               if (IS_ERR(h_file)) {
9894 +                       err = PTR_ERR(h_file);
9895 +                       break;
9896 +               }
9897 +               au_set_h_fptr(file, bindex, h_file);
9898 +       }
9899 +       au_update_figen(file);
9900 +       /* todo: necessary? */
9901 +       /* file->f_ra = h_file->f_ra; */
9902 +       if (!err)
9903 +               return 0; /* success */
9904 +
9905 +       /* close all */
9906 +       for (bindex = au_fbtop(file); bindex <= btail; bindex++)
9907 +               au_set_h_fptr(file, bindex, NULL);
9908 +       au_set_fbtop(file, -1);
9909 +       au_set_fbbot_dir(file, -1);
9910 +
9911 +       return err;
9912 +}
9913 +
9914 +static int aufs_open_dir(struct inode *inode __maybe_unused,
9915 +                        struct file *file)
9916 +{
9917 +       int err;
9918 +       struct super_block *sb;
9919 +       struct au_fidir *fidir;
9920 +
9921 +       err = -ENOMEM;
9922 +       sb = file->f_path.dentry->d_sb;
9923 +       si_read_lock(sb, AuLock_FLUSH);
9924 +       fidir = au_fidir_alloc(sb);
9925 +       if (fidir) {
9926 +               struct au_do_open_args args = {
9927 +                       .open   = do_open_dir,
9928 +                       .fidir  = fidir
9929 +               };
9930 +               err = au_do_open(file, &args);
9931 +               if (unlikely(err))
9932 +                       au_delayed_kfree(fidir);
9933 +       }
9934 +       si_read_unlock(sb);
9935 +       return err;
9936 +}
9937 +
9938 +static int aufs_release_dir(struct inode *inode __maybe_unused,
9939 +                           struct file *file)
9940 +{
9941 +       struct au_vdir *vdir_cache;
9942 +       struct au_finfo *finfo;
9943 +       struct au_fidir *fidir;
9944 +       struct au_hfile *hf;
9945 +       aufs_bindex_t bindex, bbot;
9946 +       int execed, delayed;
9947 +
9948 +       delayed = (current->flags & PF_KTHREAD) || in_interrupt();
9949 +       finfo = au_fi(file);
9950 +       fidir = finfo->fi_hdir;
9951 +       if (fidir) {
9952 +               au_sphl_del(&finfo->fi_hlist,
9953 +                           &au_sbi(file->f_path.dentry->d_sb)->si_files);
9954 +               vdir_cache = fidir->fd_vdir_cache; /* lock-free */
9955 +               if (vdir_cache)
9956 +                       au_vdir_free(vdir_cache, delayed);
9957 +
9958 +               bindex = finfo->fi_btop;
9959 +               if (bindex >= 0) {
9960 +                       execed = vfsub_file_execed(file);
9961 +                       hf = fidir->fd_hfile + bindex;
9962 +                       /*
9963 +                        * calls fput() instead of filp_close(),
9964 +                        * since no dnotify or lock for the lower file.
9965 +                        */
9966 +                       bbot = fidir->fd_bbot;
9967 +                       for (; bindex <= bbot; bindex++, hf++)
9968 +                               if (hf->hf_file)
9969 +                                       au_hfput(hf, execed);
9970 +               }
9971 +               au_delayed_kfree(fidir);
9972 +               finfo->fi_hdir = NULL;
9973 +       }
9974 +       au_finfo_fin(file, delayed);
9975 +       return 0;
9976 +}
9977 +
9978 +/* ---------------------------------------------------------------------- */
9979 +
9980 +static int au_do_flush_dir(struct file *file, fl_owner_t id)
9981 +{
9982 +       int err;
9983 +       aufs_bindex_t bindex, bbot;
9984 +       struct file *h_file;
9985 +
9986 +       err = 0;
9987 +       bbot = au_fbbot_dir(file);
9988 +       for (bindex = au_fbtop(file); !err && bindex <= bbot; bindex++) {
9989 +               h_file = au_hf_dir(file, bindex);
9990 +               if (h_file)
9991 +                       err = vfsub_flush(h_file, id);
9992 +       }
9993 +       return err;
9994 +}
9995 +
9996 +static int aufs_flush_dir(struct file *file, fl_owner_t id)
9997 +{
9998 +       return au_do_flush(file, id, au_do_flush_dir);
9999 +}
10000 +
10001 +/* ---------------------------------------------------------------------- */
10002 +
10003 +static int au_do_fsync_dir_no_file(struct dentry *dentry, int datasync)
10004 +{
10005 +       int err;
10006 +       aufs_bindex_t bbot, bindex;
10007 +       struct inode *inode;
10008 +       struct super_block *sb;
10009 +
10010 +       err = 0;
10011 +       sb = dentry->d_sb;
10012 +       inode = d_inode(dentry);
10013 +       IMustLock(inode);
10014 +       bbot = au_dbbot(dentry);
10015 +       for (bindex = au_dbtop(dentry); !err && bindex <= bbot; bindex++) {
10016 +               struct path h_path;
10017 +
10018 +               if (au_test_ro(sb, bindex, inode))
10019 +                       continue;
10020 +               h_path.dentry = au_h_dptr(dentry, bindex);
10021 +               if (!h_path.dentry)
10022 +                       continue;
10023 +
10024 +               h_path.mnt = au_sbr_mnt(sb, bindex);
10025 +               err = vfsub_fsync(NULL, &h_path, datasync);
10026 +       }
10027 +
10028 +       return err;
10029 +}
10030 +
10031 +static int au_do_fsync_dir(struct file *file, int datasync)
10032 +{
10033 +       int err;
10034 +       aufs_bindex_t bbot, bindex;
10035 +       struct file *h_file;
10036 +       struct super_block *sb;
10037 +       struct inode *inode;
10038 +
10039 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
10040 +       if (unlikely(err))
10041 +               goto out;
10042 +
10043 +       inode = file_inode(file);
10044 +       sb = inode->i_sb;
10045 +       bbot = au_fbbot_dir(file);
10046 +       for (bindex = au_fbtop(file); !err && bindex <= bbot; bindex++) {
10047 +               h_file = au_hf_dir(file, bindex);
10048 +               if (!h_file || au_test_ro(sb, bindex, inode))
10049 +                       continue;
10050 +
10051 +               err = vfsub_fsync(h_file, &h_file->f_path, datasync);
10052 +       }
10053 +
10054 +out:
10055 +       return err;
10056 +}
10057 +
10058 +/*
10059 + * @file may be NULL
10060 + */
10061 +static int aufs_fsync_dir(struct file *file, loff_t start, loff_t end,
10062 +                         int datasync)
10063 +{
10064 +       int err;
10065 +       struct dentry *dentry;
10066 +       struct inode *inode;
10067 +       struct super_block *sb;
10068 +
10069 +       err = 0;
10070 +       dentry = file->f_path.dentry;
10071 +       inode = d_inode(dentry);
10072 +       inode_lock(inode);
10073 +       sb = dentry->d_sb;
10074 +       si_noflush_read_lock(sb);
10075 +       if (file)
10076 +               err = au_do_fsync_dir(file, datasync);
10077 +       else {
10078 +               di_write_lock_child(dentry);
10079 +               err = au_do_fsync_dir_no_file(dentry, datasync);
10080 +       }
10081 +       au_cpup_attr_timesizes(inode);
10082 +       di_write_unlock(dentry);
10083 +       if (file)
10084 +               fi_write_unlock(file);
10085 +
10086 +       si_read_unlock(sb);
10087 +       inode_unlock(inode);
10088 +       return err;
10089 +}
10090 +
10091 +/* ---------------------------------------------------------------------- */
10092 +
10093 +static int aufs_iterate_shared(struct file *file, struct dir_context *ctx)
10094 +{
10095 +       int err;
10096 +       struct dentry *dentry;
10097 +       struct inode *inode, *h_inode;
10098 +       struct super_block *sb;
10099 +
10100 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
10101 +
10102 +       dentry = file->f_path.dentry;
10103 +       inode = d_inode(dentry);
10104 +       IMustLock(inode);
10105 +
10106 +       sb = dentry->d_sb;
10107 +       si_read_lock(sb, AuLock_FLUSH);
10108 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
10109 +       if (unlikely(err))
10110 +               goto out;
10111 +       err = au_alive_dir(dentry);
10112 +       if (!err)
10113 +               err = au_vdir_init(file);
10114 +       di_downgrade_lock(dentry, AuLock_IR);
10115 +       if (unlikely(err))
10116 +               goto out_unlock;
10117 +
10118 +       h_inode = au_h_iptr(inode, au_ibtop(inode));
10119 +       if (!au_test_nfsd()) {
10120 +               err = au_vdir_fill_de(file, ctx);
10121 +               fsstack_copy_attr_atime(inode, h_inode);
10122 +       } else {
10123 +               /*
10124 +                * nfsd filldir may call lookup_one_len(), vfs_getattr(),
10125 +                * encode_fh() and others.
10126 +                */
10127 +               atomic_inc(&h_inode->i_count);
10128 +               di_read_unlock(dentry, AuLock_IR);
10129 +               si_read_unlock(sb);
10130 +               err = au_vdir_fill_de(file, ctx);
10131 +               fsstack_copy_attr_atime(inode, h_inode);
10132 +               fi_write_unlock(file);
10133 +               iput(h_inode);
10134 +
10135 +               AuTraceErr(err);
10136 +               return err;
10137 +       }
10138 +
10139 +out_unlock:
10140 +       di_read_unlock(dentry, AuLock_IR);
10141 +       fi_write_unlock(file);
10142 +out:
10143 +       si_read_unlock(sb);
10144 +       return err;
10145 +}
10146 +
10147 +/* ---------------------------------------------------------------------- */
10148 +
10149 +#define AuTestEmpty_WHONLY     1
10150 +#define AuTestEmpty_CALLED     (1 << 1)
10151 +#define AuTestEmpty_SHWH       (1 << 2)
10152 +#define au_ftest_testempty(flags, name)        ((flags) & AuTestEmpty_##name)
10153 +#define au_fset_testempty(flags, name) \
10154 +       do { (flags) |= AuTestEmpty_##name; } while (0)
10155 +#define au_fclr_testempty(flags, name) \
10156 +       do { (flags) &= ~AuTestEmpty_##name; } while (0)
10157 +
10158 +#ifndef CONFIG_AUFS_SHWH
10159 +#undef AuTestEmpty_SHWH
10160 +#define AuTestEmpty_SHWH       0
10161 +#endif
10162 +
10163 +struct test_empty_arg {
10164 +       struct dir_context ctx;
10165 +       struct au_nhash *whlist;
10166 +       unsigned int flags;
10167 +       int err;
10168 +       aufs_bindex_t bindex;
10169 +};
10170 +
10171 +static int test_empty_cb(struct dir_context *ctx, const char *__name,
10172 +                        int namelen, loff_t offset __maybe_unused, u64 ino,
10173 +                        unsigned int d_type)
10174 +{
10175 +       struct test_empty_arg *arg = container_of(ctx, struct test_empty_arg,
10176 +                                                 ctx);
10177 +       char *name = (void *)__name;
10178 +
10179 +       arg->err = 0;
10180 +       au_fset_testempty(arg->flags, CALLED);
10181 +       /* smp_mb(); */
10182 +       if (name[0] == '.'
10183 +           && (namelen == 1 || (name[1] == '.' && namelen == 2)))
10184 +               goto out; /* success */
10185 +
10186 +       if (namelen <= AUFS_WH_PFX_LEN
10187 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
10188 +               if (au_ftest_testempty(arg->flags, WHONLY)
10189 +                   && !au_nhash_test_known_wh(arg->whlist, name, namelen))
10190 +                       arg->err = -ENOTEMPTY;
10191 +               goto out;
10192 +       }
10193 +
10194 +       name += AUFS_WH_PFX_LEN;
10195 +       namelen -= AUFS_WH_PFX_LEN;
10196 +       if (!au_nhash_test_known_wh(arg->whlist, name, namelen))
10197 +               arg->err = au_nhash_append_wh
10198 +                       (arg->whlist, name, namelen, ino, d_type, arg->bindex,
10199 +                        au_ftest_testempty(arg->flags, SHWH));
10200 +
10201 +out:
10202 +       /* smp_mb(); */
10203 +       AuTraceErr(arg->err);
10204 +       return arg->err;
10205 +}
10206 +
10207 +static int do_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
10208 +{
10209 +       int err;
10210 +       struct file *h_file;
10211 +
10212 +       h_file = au_h_open(dentry, arg->bindex,
10213 +                          O_RDONLY | O_NONBLOCK | O_DIRECTORY | O_LARGEFILE,
10214 +                          /*file*/NULL, /*force_wr*/0);
10215 +       err = PTR_ERR(h_file);
10216 +       if (IS_ERR(h_file))
10217 +               goto out;
10218 +
10219 +       err = 0;
10220 +       if (!au_opt_test(au_mntflags(dentry->d_sb), UDBA_NONE)
10221 +           && !file_inode(h_file)->i_nlink)
10222 +               goto out_put;
10223 +
10224 +       do {
10225 +               arg->err = 0;
10226 +               au_fclr_testempty(arg->flags, CALLED);
10227 +               /* smp_mb(); */
10228 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
10229 +               if (err >= 0)
10230 +                       err = arg->err;
10231 +       } while (!err && au_ftest_testempty(arg->flags, CALLED));
10232 +
10233 +out_put:
10234 +       fput(h_file);
10235 +       au_sbr_put(dentry->d_sb, arg->bindex);
10236 +out:
10237 +       return err;
10238 +}
10239 +
10240 +struct do_test_empty_args {
10241 +       int *errp;
10242 +       struct dentry *dentry;
10243 +       struct test_empty_arg *arg;
10244 +};
10245 +
10246 +static void call_do_test_empty(void *args)
10247 +{
10248 +       struct do_test_empty_args *a = args;
10249 +       *a->errp = do_test_empty(a->dentry, a->arg);
10250 +}
10251 +
10252 +static int sio_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
10253 +{
10254 +       int err, wkq_err;
10255 +       struct dentry *h_dentry;
10256 +       struct inode *h_inode;
10257 +
10258 +       h_dentry = au_h_dptr(dentry, arg->bindex);
10259 +       h_inode = d_inode(h_dentry);
10260 +       /* todo: i_mode changes anytime? */
10261 +       inode_lock_nested(h_inode, AuLsc_I_CHILD);
10262 +       err = au_test_h_perm_sio(h_inode, MAY_EXEC | MAY_READ);
10263 +       inode_unlock(h_inode);
10264 +       if (!err)
10265 +               err = do_test_empty(dentry, arg);
10266 +       else {
10267 +               struct do_test_empty_args args = {
10268 +                       .errp   = &err,
10269 +                       .dentry = dentry,
10270 +                       .arg    = arg
10271 +               };
10272 +               unsigned int flags = arg->flags;
10273 +
10274 +               wkq_err = au_wkq_wait(call_do_test_empty, &args);
10275 +               if (unlikely(wkq_err))
10276 +                       err = wkq_err;
10277 +               arg->flags = flags;
10278 +       }
10279 +
10280 +       return err;
10281 +}
10282 +
10283 +int au_test_empty_lower(struct dentry *dentry)
10284 +{
10285 +       int err;
10286 +       unsigned int rdhash;
10287 +       aufs_bindex_t bindex, btop, btail;
10288 +       struct au_nhash whlist;
10289 +       struct test_empty_arg arg = {
10290 +               .ctx = {
10291 +                       .actor = test_empty_cb
10292 +               }
10293 +       };
10294 +       int (*test_empty)(struct dentry *dentry, struct test_empty_arg *arg);
10295 +
10296 +       SiMustAnyLock(dentry->d_sb);
10297 +
10298 +       rdhash = au_sbi(dentry->d_sb)->si_rdhash;
10299 +       if (!rdhash)
10300 +               rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, dentry));
10301 +       err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
10302 +       if (unlikely(err))
10303 +               goto out;
10304 +
10305 +       arg.flags = 0;
10306 +       arg.whlist = &whlist;
10307 +       btop = au_dbtop(dentry);
10308 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
10309 +               au_fset_testempty(arg.flags, SHWH);
10310 +       test_empty = do_test_empty;
10311 +       if (au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1))
10312 +               test_empty = sio_test_empty;
10313 +       arg.bindex = btop;
10314 +       err = test_empty(dentry, &arg);
10315 +       if (unlikely(err))
10316 +               goto out_whlist;
10317 +
10318 +       au_fset_testempty(arg.flags, WHONLY);
10319 +       btail = au_dbtaildir(dentry);
10320 +       for (bindex = btop + 1; !err && bindex <= btail; bindex++) {
10321 +               struct dentry *h_dentry;
10322 +
10323 +               h_dentry = au_h_dptr(dentry, bindex);
10324 +               if (h_dentry && d_is_positive(h_dentry)) {
10325 +                       arg.bindex = bindex;
10326 +                       err = test_empty(dentry, &arg);
10327 +               }
10328 +       }
10329 +
10330 +out_whlist:
10331 +       au_nhash_wh_free(&whlist);
10332 +out:
10333 +       return err;
10334 +}
10335 +
10336 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist)
10337 +{
10338 +       int err;
10339 +       struct test_empty_arg arg = {
10340 +               .ctx = {
10341 +                       .actor = test_empty_cb
10342 +               }
10343 +       };
10344 +       aufs_bindex_t bindex, btail;
10345 +
10346 +       err = 0;
10347 +       arg.whlist = whlist;
10348 +       arg.flags = AuTestEmpty_WHONLY;
10349 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
10350 +               au_fset_testempty(arg.flags, SHWH);
10351 +       btail = au_dbtaildir(dentry);
10352 +       for (bindex = au_dbtop(dentry); !err && bindex <= btail; bindex++) {
10353 +               struct dentry *h_dentry;
10354 +
10355 +               h_dentry = au_h_dptr(dentry, bindex);
10356 +               if (h_dentry && d_is_positive(h_dentry)) {
10357 +                       arg.bindex = bindex;
10358 +                       err = sio_test_empty(dentry, &arg);
10359 +               }
10360 +       }
10361 +
10362 +       return err;
10363 +}
10364 +
10365 +/* ---------------------------------------------------------------------- */
10366 +
10367 +const struct file_operations aufs_dir_fop = {
10368 +       .owner          = THIS_MODULE,
10369 +       .llseek         = default_llseek,
10370 +       .read           = generic_read_dir,
10371 +       .iterate_shared = aufs_iterate_shared,
10372 +       .unlocked_ioctl = aufs_ioctl_dir,
10373 +#ifdef CONFIG_COMPAT
10374 +       .compat_ioctl   = aufs_compat_ioctl_dir,
10375 +#endif
10376 +       .open           = aufs_open_dir,
10377 +       .release        = aufs_release_dir,
10378 +       .flush          = aufs_flush_dir,
10379 +       .fsync          = aufs_fsync_dir
10380 +};
10381 diff -urN /usr/share/empty/fs/aufs/dir.h linux/fs/aufs/dir.h
10382 --- /usr/share/empty/fs/aufs/dir.h      1970-01-01 01:00:00.000000000 +0100
10383 +++ linux/fs/aufs/dir.h 2016-10-09 16:55:36.489368218 +0200
10384 @@ -0,0 +1,137 @@
10385 +/*
10386 + * Copyright (C) 2005-2016 Junjiro R. Okajima
10387 + *
10388 + * This program, aufs is free software; you can redistribute it and/or modify
10389 + * it under the terms of the GNU General Public License as published by
10390 + * the Free Software Foundation; either version 2 of the License, or
10391 + * (at your option) any later version.
10392 + *
10393 + * This program is distributed in the hope that it will be useful,
10394 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10395 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10396 + * GNU General Public License for more details.
10397 + *
10398 + * You should have received a copy of the GNU General Public License
10399 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10400 + */
10401 +
10402 +/*
10403 + * directory operations
10404 + */
10405 +
10406 +#ifndef __AUFS_DIR_H__
10407 +#define __AUFS_DIR_H__
10408 +
10409 +#ifdef __KERNEL__
10410 +
10411 +#include <linux/fs.h>
10412 +
10413 +/* ---------------------------------------------------------------------- */
10414 +
10415 +/* need to be faster and smaller */
10416 +
10417 +struct au_nhash {
10418 +       unsigned int            nh_num;
10419 +       struct hlist_head       *nh_head;
10420 +};
10421 +
10422 +struct au_vdir_destr {
10423 +       unsigned char   len;
10424 +       unsigned char   name[0];
10425 +} __packed;
10426 +
10427 +struct au_vdir_dehstr {
10428 +       struct hlist_node       hash;
10429 +       union {
10430 +               struct au_vdir_destr    *str;
10431 +               struct llist_node       lnode;  /* delayed free */
10432 +       };
10433 +} ____cacheline_aligned_in_smp;
10434 +
10435 +struct au_vdir_de {
10436 +       ino_t                   de_ino;
10437 +       unsigned char           de_type;
10438 +       /* caution: packed */
10439 +       struct au_vdir_destr    de_str;
10440 +} __packed;
10441 +
10442 +struct au_vdir_wh {
10443 +       struct hlist_node       wh_hash;
10444 +#ifdef CONFIG_AUFS_SHWH
10445 +       ino_t                   wh_ino;
10446 +       aufs_bindex_t           wh_bindex;
10447 +       unsigned char           wh_type;
10448 +#else
10449 +       aufs_bindex_t           wh_bindex;
10450 +#endif
10451 +       /* caution: packed */
10452 +       struct au_vdir_destr    wh_str;
10453 +} __packed;
10454 +
10455 +union au_vdir_deblk_p {
10456 +       unsigned char           *deblk;
10457 +       struct au_vdir_de       *de;
10458 +};
10459 +
10460 +struct au_vdir {
10461 +       unsigned char   **vd_deblk;
10462 +       unsigned long   vd_nblk;
10463 +       struct {
10464 +               unsigned long           ul;
10465 +               union au_vdir_deblk_p   p;
10466 +       } vd_last;
10467 +
10468 +       unsigned long   vd_version;
10469 +       unsigned int    vd_deblk_sz;
10470 +       union {
10471 +               unsigned long           vd_jiffy;
10472 +               struct llist_node       vd_lnode;       /* delayed free */
10473 +       };
10474 +} ____cacheline_aligned_in_smp;
10475 +
10476 +/* ---------------------------------------------------------------------- */
10477 +
10478 +/* dir.c */
10479 +extern const struct file_operations aufs_dir_fop;
10480 +void au_add_nlink(struct inode *dir, struct inode *h_dir);
10481 +void au_sub_nlink(struct inode *dir, struct inode *h_dir);
10482 +loff_t au_dir_size(struct file *file, struct dentry *dentry);
10483 +void au_dir_ts(struct inode *dir, aufs_bindex_t bsrc);
10484 +int au_test_empty_lower(struct dentry *dentry);
10485 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist);
10486 +
10487 +/* vdir.c */
10488 +unsigned int au_rdhash_est(loff_t sz);
10489 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp);
10490 +void au_nhash_wh_free(struct au_nhash *whlist);
10491 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
10492 +                           int limit);
10493 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen);
10494 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
10495 +                      unsigned int d_type, aufs_bindex_t bindex,
10496 +                      unsigned char shwh);
10497 +void au_vdir_free(struct au_vdir *vdir, int atonce);
10498 +int au_vdir_init(struct file *file);
10499 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx);
10500 +
10501 +/* ioctl.c */
10502 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg);
10503 +
10504 +#ifdef CONFIG_AUFS_RDU
10505 +/* rdu.c */
10506 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
10507 +#ifdef CONFIG_COMPAT
10508 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd,
10509 +                        unsigned long arg);
10510 +#endif
10511 +#else
10512 +AuStub(long, au_rdu_ioctl, return -EINVAL, struct file *file,
10513 +       unsigned int cmd, unsigned long arg)
10514 +#ifdef CONFIG_COMPAT
10515 +AuStub(long, au_rdu_compat_ioctl, return -EINVAL, struct file *file,
10516 +       unsigned int cmd, unsigned long arg)
10517 +#endif
10518 +#endif
10519 +
10520 +#endif /* __KERNEL__ */
10521 +#endif /* __AUFS_DIR_H__ */
10522 diff -urN /usr/share/empty/fs/aufs/dynop.c linux/fs/aufs/dynop.c
10523 --- /usr/share/empty/fs/aufs/dynop.c    1970-01-01 01:00:00.000000000 +0100
10524 +++ linux/fs/aufs/dynop.c       2016-10-09 16:55:36.489368218 +0200
10525 @@ -0,0 +1,371 @@
10526 +/*
10527 + * Copyright (C) 2010-2016 Junjiro R. Okajima
10528 + *
10529 + * This program, aufs is free software; you can redistribute it and/or modify
10530 + * it under the terms of the GNU General Public License as published by
10531 + * the Free Software Foundation; either version 2 of the License, or
10532 + * (at your option) any later version.
10533 + *
10534 + * This program is distributed in the hope that it will be useful,
10535 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10536 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10537 + * GNU General Public License for more details.
10538 + *
10539 + * You should have received a copy of the GNU General Public License
10540 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10541 + */
10542 +
10543 +/*
10544 + * dynamically customizable operations for regular files
10545 + */
10546 +
10547 +#include "aufs.h"
10548 +
10549 +#define DyPrSym(key)   AuDbgSym(key->dk_op.dy_hop)
10550 +
10551 +/*
10552 + * How large will these lists be?
10553 + * Usually just a few elements, 20-30 at most for each, I guess.
10554 + */
10555 +static struct au_sphlhead dynop[AuDyLast];
10556 +
10557 +static struct au_dykey *dy_gfind_get(struct au_sphlhead *sphl, const void *h_op)
10558 +{
10559 +       struct au_dykey *key, *tmp;
10560 +       struct hlist_head *head;
10561 +
10562 +       key = NULL;
10563 +       head = &sphl->head;
10564 +       rcu_read_lock();
10565 +       hlist_for_each_entry_rcu(tmp, head, dk_hnode)
10566 +               if (tmp->dk_op.dy_hop == h_op) {
10567 +                       key = tmp;
10568 +                       kref_get(&key->dk_kref);
10569 +                       break;
10570 +               }
10571 +       rcu_read_unlock();
10572 +
10573 +       return key;
10574 +}
10575 +
10576 +static struct au_dykey *dy_bradd(struct au_branch *br, struct au_dykey *key)
10577 +{
10578 +       struct au_dykey **k, *found;
10579 +       const void *h_op = key->dk_op.dy_hop;
10580 +       int i;
10581 +
10582 +       found = NULL;
10583 +       k = br->br_dykey;
10584 +       for (i = 0; i < AuBrDynOp; i++)
10585 +               if (k[i]) {
10586 +                       if (k[i]->dk_op.dy_hop == h_op) {
10587 +                               found = k[i];
10588 +                               break;
10589 +                       }
10590 +               } else
10591 +                       break;
10592 +       if (!found) {
10593 +               spin_lock(&br->br_dykey_lock);
10594 +               for (; i < AuBrDynOp; i++)
10595 +                       if (k[i]) {
10596 +                               if (k[i]->dk_op.dy_hop == h_op) {
10597 +                                       found = k[i];
10598 +                                       break;
10599 +                               }
10600 +                       } else {
10601 +                               k[i] = key;
10602 +                               break;
10603 +                       }
10604 +               spin_unlock(&br->br_dykey_lock);
10605 +               BUG_ON(i == AuBrDynOp); /* expand the array */
10606 +       }
10607 +
10608 +       return found;
10609 +}
10610 +
10611 +/* kref_get() if @key is already added */
10612 +static struct au_dykey *dy_gadd(struct au_sphlhead *sphl, struct au_dykey *key)
10613 +{
10614 +       struct au_dykey *tmp, *found;
10615 +       struct hlist_head *head;
10616 +       const void *h_op = key->dk_op.dy_hop;
10617 +
10618 +       found = NULL;
10619 +       head = &sphl->head;
10620 +       spin_lock(&sphl->spin);
10621 +       hlist_for_each_entry(tmp, head, dk_hnode)
10622 +               if (tmp->dk_op.dy_hop == h_op) {
10623 +                       kref_get(&tmp->dk_kref);
10624 +                       found = tmp;
10625 +                       break;
10626 +               }
10627 +       if (!found)
10628 +               hlist_add_head_rcu(&key->dk_hnode, head);
10629 +       spin_unlock(&sphl->spin);
10630 +
10631 +       if (!found)
10632 +               DyPrSym(key);
10633 +       return found;
10634 +}
10635 +
10636 +static void dy_free_rcu(struct rcu_head *rcu)
10637 +{
10638 +       struct au_dykey *key;
10639 +
10640 +       key = container_of(rcu, struct au_dykey, dk_rcu);
10641 +       DyPrSym(key);
10642 +       kfree(key);     /* not delayed */
10643 +}
10644 +
10645 +static void dy_free(struct kref *kref)
10646 +{
10647 +       struct au_dykey *key;
10648 +       struct au_sphlhead *sphl;
10649 +
10650 +       key = container_of(kref, struct au_dykey, dk_kref);
10651 +       sphl = dynop + key->dk_op.dy_type;
10652 +       au_sphl_del_rcu(&key->dk_hnode, sphl);
10653 +       call_rcu(&key->dk_rcu, dy_free_rcu);
10654 +}
10655 +
10656 +void au_dy_put(struct au_dykey *key)
10657 +{
10658 +       kref_put(&key->dk_kref, dy_free);
10659 +}
10660 +
10661 +/* ---------------------------------------------------------------------- */
10662 +
10663 +#define DyDbgSize(cnt, op)     AuDebugOn(cnt != sizeof(op)/sizeof(void *))
10664 +
10665 +#ifdef CONFIG_AUFS_DEBUG
10666 +#define DyDbgDeclare(cnt)      unsigned int cnt = 0
10667 +#define DyDbgInc(cnt)          do { cnt++; } while (0)
10668 +#else
10669 +#define DyDbgDeclare(cnt)      do {} while (0)
10670 +#define DyDbgInc(cnt)          do {} while (0)
10671 +#endif
10672 +
10673 +#define DySet(func, dst, src, h_op, h_sb) do {                         \
10674 +       DyDbgInc(cnt);                                                  \
10675 +       if (h_op->func) {                                               \
10676 +               if (src.func)                                           \
10677 +                       dst.func = src.func;                            \
10678 +               else                                                    \
10679 +                       AuDbg("%s %s\n", au_sbtype(h_sb), #func);       \
10680 +       }                                                               \
10681 +} while (0)
10682 +
10683 +#define DySetForce(func, dst, src) do {                \
10684 +       AuDebugOn(!src.func);                   \
10685 +       DyDbgInc(cnt);                          \
10686 +       dst.func = src.func;                    \
10687 +} while (0)
10688 +
10689 +#define DySetAop(func) \
10690 +       DySet(func, dyaop->da_op, aufs_aop, h_aop, h_sb)
10691 +#define DySetAopForce(func) \
10692 +       DySetForce(func, dyaop->da_op, aufs_aop)
10693 +
10694 +static void dy_aop(struct au_dykey *key, const void *h_op,
10695 +                  struct super_block *h_sb __maybe_unused)
10696 +{
10697 +       struct au_dyaop *dyaop = (void *)key;
10698 +       const struct address_space_operations *h_aop = h_op;
10699 +       DyDbgDeclare(cnt);
10700 +
10701 +       AuDbg("%s\n", au_sbtype(h_sb));
10702 +
10703 +       DySetAop(writepage);
10704 +       DySetAopForce(readpage);        /* force */
10705 +       DySetAop(writepages);
10706 +       DySetAop(set_page_dirty);
10707 +       DySetAop(readpages);
10708 +       DySetAop(write_begin);
10709 +       DySetAop(write_end);
10710 +       DySetAop(bmap);
10711 +       DySetAop(invalidatepage);
10712 +       DySetAop(releasepage);
10713 +       DySetAop(freepage);
10714 +       /* this one will be changed according to an aufs mount option */
10715 +       DySetAop(direct_IO);
10716 +       DySetAop(migratepage);
10717 +       DySetAop(isolate_page);
10718 +       DySetAop(putback_page);
10719 +       DySetAop(launder_page);
10720 +       DySetAop(is_partially_uptodate);
10721 +       DySetAop(is_dirty_writeback);
10722 +       DySetAop(error_remove_page);
10723 +       DySetAop(swap_activate);
10724 +       DySetAop(swap_deactivate);
10725 +
10726 +       DyDbgSize(cnt, *h_aop);
10727 +}
10728 +
10729 +/* ---------------------------------------------------------------------- */
10730 +
10731 +static void dy_bug(struct kref *kref)
10732 +{
10733 +       BUG();
10734 +}
10735 +
10736 +static struct au_dykey *dy_get(struct au_dynop *op, struct au_branch *br)
10737 +{
10738 +       struct au_dykey *key, *old;
10739 +       struct au_sphlhead *sphl;
10740 +       struct op {
10741 +               unsigned int sz;
10742 +               void (*set)(struct au_dykey *key, const void *h_op,
10743 +                           struct super_block *h_sb __maybe_unused);
10744 +       };
10745 +       static const struct op a[] = {
10746 +               [AuDy_AOP] = {
10747 +                       .sz     = sizeof(struct au_dyaop),
10748 +                       .set    = dy_aop
10749 +               }
10750 +       };
10751 +       const struct op *p;
10752 +
10753 +       sphl = dynop + op->dy_type;
10754 +       key = dy_gfind_get(sphl, op->dy_hop);
10755 +       if (key)
10756 +               goto out_add; /* success */
10757 +
10758 +       p = a + op->dy_type;
10759 +       key = kzalloc(p->sz, GFP_NOFS);
10760 +       if (unlikely(!key)) {
10761 +               key = ERR_PTR(-ENOMEM);
10762 +               goto out;
10763 +       }
10764 +
10765 +       key->dk_op.dy_hop = op->dy_hop;
10766 +       kref_init(&key->dk_kref);
10767 +       p->set(key, op->dy_hop, au_br_sb(br));
10768 +       old = dy_gadd(sphl, key);
10769 +       if (old) {
10770 +               au_delayed_kfree(key);
10771 +               key = old;
10772 +       }
10773 +
10774 +out_add:
10775 +       old = dy_bradd(br, key);
10776 +       if (old)
10777 +               /* its ref-count should never be zero here */
10778 +               kref_put(&key->dk_kref, dy_bug);
10779 +out:
10780 +       return key;
10781 +}
10782 +
10783 +/* ---------------------------------------------------------------------- */
10784 +/*
10785 + * Aufs prohibits O_DIRECT by defaut even if the branch supports it.
10786 + * This behaviour is necessary to return an error from open(O_DIRECT) instead
10787 + * of the succeeding I/O. The dio mount option enables O_DIRECT and makes
10788 + * open(O_DIRECT) always succeed, but the succeeding I/O may return an error.
10789 + * See the aufs manual in detail.
10790 + */
10791 +static void dy_adx(struct au_dyaop *dyaop, int do_dx)
10792 +{
10793 +       if (!do_dx)
10794 +               dyaop->da_op.direct_IO = NULL;
10795 +       else
10796 +               dyaop->da_op.direct_IO = aufs_aop.direct_IO;
10797 +}
10798 +
10799 +static struct au_dyaop *dy_aget(struct au_branch *br,
10800 +                               const struct address_space_operations *h_aop,
10801 +                               int do_dx)
10802 +{
10803 +       struct au_dyaop *dyaop;
10804 +       struct au_dynop op;
10805 +
10806 +       op.dy_type = AuDy_AOP;
10807 +       op.dy_haop = h_aop;
10808 +       dyaop = (void *)dy_get(&op, br);
10809 +       if (IS_ERR(dyaop))
10810 +               goto out;
10811 +       dy_adx(dyaop, do_dx);
10812 +
10813 +out:
10814 +       return dyaop;
10815 +}
10816 +
10817 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
10818 +               struct inode *h_inode)
10819 +{
10820 +       int err, do_dx;
10821 +       struct super_block *sb;
10822 +       struct au_branch *br;
10823 +       struct au_dyaop *dyaop;
10824 +
10825 +       AuDebugOn(!S_ISREG(h_inode->i_mode));
10826 +       IiMustWriteLock(inode);
10827 +
10828 +       sb = inode->i_sb;
10829 +       br = au_sbr(sb, bindex);
10830 +       do_dx = !!au_opt_test(au_mntflags(sb), DIO);
10831 +       dyaop = dy_aget(br, h_inode->i_mapping->a_ops, do_dx);
10832 +       err = PTR_ERR(dyaop);
10833 +       if (IS_ERR(dyaop))
10834 +               /* unnecessary to call dy_fput() */
10835 +               goto out;
10836 +
10837 +       err = 0;
10838 +       inode->i_mapping->a_ops = &dyaop->da_op;
10839 +
10840 +out:
10841 +       return err;
10842 +}
10843 +
10844 +/*
10845 + * Is it safe to replace a_ops during the inode/file is in operation?
10846 + * Yes, I hope so.
10847 + */
10848 +int au_dy_irefresh(struct inode *inode)
10849 +{
10850 +       int err;
10851 +       aufs_bindex_t btop;
10852 +       struct inode *h_inode;
10853 +
10854 +       err = 0;
10855 +       if (S_ISREG(inode->i_mode)) {
10856 +               btop = au_ibtop(inode);
10857 +               h_inode = au_h_iptr(inode, btop);
10858 +               err = au_dy_iaop(inode, btop, h_inode);
10859 +       }
10860 +       return err;
10861 +}
10862 +
10863 +void au_dy_arefresh(int do_dx)
10864 +{
10865 +       struct au_sphlhead *sphl;
10866 +       struct hlist_head *head;
10867 +       struct au_dykey *key;
10868 +
10869 +       sphl = dynop + AuDy_AOP;
10870 +       head = &sphl->head;
10871 +       spin_lock(&sphl->spin);
10872 +       hlist_for_each_entry(key, head, dk_hnode)
10873 +               dy_adx((void *)key, do_dx);
10874 +       spin_unlock(&sphl->spin);
10875 +}
10876 +
10877 +/* ---------------------------------------------------------------------- */
10878 +
10879 +void __init au_dy_init(void)
10880 +{
10881 +       int i;
10882 +
10883 +       /* make sure that 'struct au_dykey *' can be any type */
10884 +       BUILD_BUG_ON(offsetof(struct au_dyaop, da_key));
10885 +
10886 +       for (i = 0; i < AuDyLast; i++)
10887 +               au_sphl_init(dynop + i);
10888 +}
10889 +
10890 +void au_dy_fin(void)
10891 +{
10892 +       int i;
10893 +
10894 +       for (i = 0; i < AuDyLast; i++)
10895 +               WARN_ON(!hlist_empty(&dynop[i].head));
10896 +}
10897 diff -urN /usr/share/empty/fs/aufs/dynop.h linux/fs/aufs/dynop.h
10898 --- /usr/share/empty/fs/aufs/dynop.h    1970-01-01 01:00:00.000000000 +0100
10899 +++ linux/fs/aufs/dynop.h       2016-10-09 16:55:36.489368218 +0200
10900 @@ -0,0 +1,74 @@
10901 +/*
10902 + * Copyright (C) 2010-2016 Junjiro R. Okajima
10903 + *
10904 + * This program, aufs is free software; you can redistribute it and/or modify
10905 + * it under the terms of the GNU General Public License as published by
10906 + * the Free Software Foundation; either version 2 of the License, or
10907 + * (at your option) any later version.
10908 + *
10909 + * This program is distributed in the hope that it will be useful,
10910 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10911 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10912 + * GNU General Public License for more details.
10913 + *
10914 + * You should have received a copy of the GNU General Public License
10915 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10916 + */
10917 +
10918 +/*
10919 + * dynamically customizable operations (for regular files only)
10920 + */
10921 +
10922 +#ifndef __AUFS_DYNOP_H__
10923 +#define __AUFS_DYNOP_H__
10924 +
10925 +#ifdef __KERNEL__
10926 +
10927 +#include <linux/fs.h>
10928 +#include <linux/kref.h>
10929 +
10930 +enum {AuDy_AOP, AuDyLast};
10931 +
10932 +struct au_dynop {
10933 +       int                                             dy_type;
10934 +       union {
10935 +               const void                              *dy_hop;
10936 +               const struct address_space_operations   *dy_haop;
10937 +       };
10938 +};
10939 +
10940 +struct au_dykey {
10941 +       union {
10942 +               struct hlist_node       dk_hnode;
10943 +               struct rcu_head         dk_rcu;
10944 +       };
10945 +       struct au_dynop         dk_op;
10946 +
10947 +       /*
10948 +        * during I am in the branch local array, kref is gotten. when the
10949 +        * branch is removed, kref is put.
10950 +        */
10951 +       struct kref             dk_kref;
10952 +};
10953 +
10954 +/* stop unioning since their sizes are very different from each other */
10955 +struct au_dyaop {
10956 +       struct au_dykey                 da_key;
10957 +       struct address_space_operations da_op; /* not const */
10958 +};
10959 +
10960 +/* ---------------------------------------------------------------------- */
10961 +
10962 +/* dynop.c */
10963 +struct au_branch;
10964 +void au_dy_put(struct au_dykey *key);
10965 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
10966 +               struct inode *h_inode);
10967 +int au_dy_irefresh(struct inode *inode);
10968 +void au_dy_arefresh(int do_dio);
10969 +
10970 +void __init au_dy_init(void);
10971 +void au_dy_fin(void);
10972 +
10973 +#endif /* __KERNEL__ */
10974 +#endif /* __AUFS_DYNOP_H__ */
10975 diff -urN /usr/share/empty/fs/aufs/export.c linux/fs/aufs/export.c
10976 --- /usr/share/empty/fs/aufs/export.c   1970-01-01 01:00:00.000000000 +0100
10977 +++ linux/fs/aufs/export.c      2016-10-09 16:55:36.489368218 +0200
10978 @@ -0,0 +1,837 @@
10979 +/*
10980 + * Copyright (C) 2005-2016 Junjiro R. Okajima
10981 + *
10982 + * This program, aufs is free software; you can redistribute it and/or modify
10983 + * it under the terms of the GNU General Public License as published by
10984 + * the Free Software Foundation; either version 2 of the License, or
10985 + * (at your option) any later version.
10986 + *
10987 + * This program is distributed in the hope that it will be useful,
10988 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10989 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10990 + * GNU General Public License for more details.
10991 + *
10992 + * You should have received a copy of the GNU General Public License
10993 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10994 + */
10995 +
10996 +/*
10997 + * export via nfs
10998 + */
10999 +
11000 +#include <linux/exportfs.h>
11001 +#include <linux/fs_struct.h>
11002 +#include <linux/namei.h>
11003 +#include <linux/nsproxy.h>
11004 +#include <linux/random.h>
11005 +#include <linux/writeback.h>
11006 +#include "../fs/mount.h"
11007 +#include "aufs.h"
11008 +
11009 +union conv {
11010 +#ifdef CONFIG_AUFS_INO_T_64
11011 +       __u32 a[2];
11012 +#else
11013 +       __u32 a[1];
11014 +#endif
11015 +       ino_t ino;
11016 +};
11017 +
11018 +static ino_t decode_ino(__u32 *a)
11019 +{
11020 +       union conv u;
11021 +
11022 +       BUILD_BUG_ON(sizeof(u.ino) != sizeof(u.a));
11023 +       u.a[0] = a[0];
11024 +#ifdef CONFIG_AUFS_INO_T_64
11025 +       u.a[1] = a[1];
11026 +#endif
11027 +       return u.ino;
11028 +}
11029 +
11030 +static void encode_ino(__u32 *a, ino_t ino)
11031 +{
11032 +       union conv u;
11033 +
11034 +       u.ino = ino;
11035 +       a[0] = u.a[0];
11036 +#ifdef CONFIG_AUFS_INO_T_64
11037 +       a[1] = u.a[1];
11038 +#endif
11039 +}
11040 +
11041 +/* NFS file handle */
11042 +enum {
11043 +       Fh_br_id,
11044 +       Fh_sigen,
11045 +#ifdef CONFIG_AUFS_INO_T_64
11046 +       /* support 64bit inode number */
11047 +       Fh_ino1,
11048 +       Fh_ino2,
11049 +       Fh_dir_ino1,
11050 +       Fh_dir_ino2,
11051 +#else
11052 +       Fh_ino1,
11053 +       Fh_dir_ino1,
11054 +#endif
11055 +       Fh_igen,
11056 +       Fh_h_type,
11057 +       Fh_tail,
11058 +
11059 +       Fh_ino = Fh_ino1,
11060 +       Fh_dir_ino = Fh_dir_ino1
11061 +};
11062 +
11063 +static int au_test_anon(struct dentry *dentry)
11064 +{
11065 +       /* note: read d_flags without d_lock */
11066 +       return !!(dentry->d_flags & DCACHE_DISCONNECTED);
11067 +}
11068 +
11069 +int au_test_nfsd(void)
11070 +{
11071 +       int ret;
11072 +       struct task_struct *tsk = current;
11073 +       char comm[sizeof(tsk->comm)];
11074 +
11075 +       ret = 0;
11076 +       if (tsk->flags & PF_KTHREAD) {
11077 +               get_task_comm(comm, tsk);
11078 +               ret = !strcmp(comm, "nfsd");
11079 +       }
11080 +
11081 +       return ret;
11082 +}
11083 +
11084 +/* ---------------------------------------------------------------------- */
11085 +/* inode generation external table */
11086 +
11087 +void au_xigen_inc(struct inode *inode)
11088 +{
11089 +       loff_t pos;
11090 +       ssize_t sz;
11091 +       __u32 igen;
11092 +       struct super_block *sb;
11093 +       struct au_sbinfo *sbinfo;
11094 +
11095 +       sb = inode->i_sb;
11096 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
11097 +
11098 +       sbinfo = au_sbi(sb);
11099 +       pos = inode->i_ino;
11100 +       pos *= sizeof(igen);
11101 +       igen = inode->i_generation + 1;
11102 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xigen, &igen,
11103 +                        sizeof(igen), &pos);
11104 +       if (sz == sizeof(igen))
11105 +               return; /* success */
11106 +
11107 +       if (unlikely(sz >= 0))
11108 +               AuIOErr("xigen error (%zd)\n", sz);
11109 +}
11110 +
11111 +int au_xigen_new(struct inode *inode)
11112 +{
11113 +       int err;
11114 +       loff_t pos;
11115 +       ssize_t sz;
11116 +       struct super_block *sb;
11117 +       struct au_sbinfo *sbinfo;
11118 +       struct file *file;
11119 +
11120 +       err = 0;
11121 +       /* todo: dirty, at mount time */
11122 +       if (inode->i_ino == AUFS_ROOT_INO)
11123 +               goto out;
11124 +       sb = inode->i_sb;
11125 +       SiMustAnyLock(sb);
11126 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
11127 +               goto out;
11128 +
11129 +       err = -EFBIG;
11130 +       pos = inode->i_ino;
11131 +       if (unlikely(au_loff_max / sizeof(inode->i_generation) - 1 < pos)) {
11132 +               AuIOErr1("too large i%lld\n", pos);
11133 +               goto out;
11134 +       }
11135 +       pos *= sizeof(inode->i_generation);
11136 +
11137 +       err = 0;
11138 +       sbinfo = au_sbi(sb);
11139 +       file = sbinfo->si_xigen;
11140 +       BUG_ON(!file);
11141 +
11142 +       if (vfsub_f_size_read(file)
11143 +           < pos + sizeof(inode->i_generation)) {
11144 +               inode->i_generation = atomic_inc_return(&sbinfo->si_xigen_next);
11145 +               sz = xino_fwrite(sbinfo->si_xwrite, file, &inode->i_generation,
11146 +                                sizeof(inode->i_generation), &pos);
11147 +       } else
11148 +               sz = xino_fread(sbinfo->si_xread, file, &inode->i_generation,
11149 +                               sizeof(inode->i_generation), &pos);
11150 +       if (sz == sizeof(inode->i_generation))
11151 +               goto out; /* success */
11152 +
11153 +       err = sz;
11154 +       if (unlikely(sz >= 0)) {
11155 +               err = -EIO;
11156 +               AuIOErr("xigen error (%zd)\n", sz);
11157 +       }
11158 +
11159 +out:
11160 +       return err;
11161 +}
11162 +
11163 +int au_xigen_set(struct super_block *sb, struct file *base)
11164 +{
11165 +       int err;
11166 +       struct au_sbinfo *sbinfo;
11167 +       struct file *file;
11168 +
11169 +       SiMustWriteLock(sb);
11170 +
11171 +       sbinfo = au_sbi(sb);
11172 +       file = au_xino_create2(base, sbinfo->si_xigen);
11173 +       err = PTR_ERR(file);
11174 +       if (IS_ERR(file))
11175 +               goto out;
11176 +       err = 0;
11177 +       if (sbinfo->si_xigen)
11178 +               fput(sbinfo->si_xigen);
11179 +       sbinfo->si_xigen = file;
11180 +
11181 +out:
11182 +       return err;
11183 +}
11184 +
11185 +void au_xigen_clr(struct super_block *sb)
11186 +{
11187 +       struct au_sbinfo *sbinfo;
11188 +
11189 +       SiMustWriteLock(sb);
11190 +
11191 +       sbinfo = au_sbi(sb);
11192 +       if (sbinfo->si_xigen) {
11193 +               fput(sbinfo->si_xigen);
11194 +               sbinfo->si_xigen = NULL;
11195 +       }
11196 +}
11197 +
11198 +/* ---------------------------------------------------------------------- */
11199 +
11200 +static struct dentry *decode_by_ino(struct super_block *sb, ino_t ino,
11201 +                                   ino_t dir_ino)
11202 +{
11203 +       struct dentry *dentry, *d;
11204 +       struct inode *inode;
11205 +       unsigned int sigen;
11206 +
11207 +       dentry = NULL;
11208 +       inode = ilookup(sb, ino);
11209 +       if (!inode)
11210 +               goto out;
11211 +
11212 +       dentry = ERR_PTR(-ESTALE);
11213 +       sigen = au_sigen(sb);
11214 +       if (unlikely(au_is_bad_inode(inode)
11215 +                    || IS_DEADDIR(inode)
11216 +                    || sigen != au_iigen(inode, NULL)))
11217 +               goto out_iput;
11218 +
11219 +       dentry = NULL;
11220 +       if (!dir_ino || S_ISDIR(inode->i_mode))
11221 +               dentry = d_find_alias(inode);
11222 +       else {
11223 +               spin_lock(&inode->i_lock);
11224 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
11225 +                       spin_lock(&d->d_lock);
11226 +                       if (!au_test_anon(d)
11227 +                           && d_inode(d->d_parent)->i_ino == dir_ino) {
11228 +                               dentry = dget_dlock(d);
11229 +                               spin_unlock(&d->d_lock);
11230 +                               break;
11231 +                       }
11232 +                       spin_unlock(&d->d_lock);
11233 +               }
11234 +               spin_unlock(&inode->i_lock);
11235 +       }
11236 +       if (unlikely(dentry && au_digen_test(dentry, sigen))) {
11237 +               /* need to refresh */
11238 +               dput(dentry);
11239 +               dentry = NULL;
11240 +       }
11241 +
11242 +out_iput:
11243 +       iput(inode);
11244 +out:
11245 +       AuTraceErrPtr(dentry);
11246 +       return dentry;
11247 +}
11248 +
11249 +/* ---------------------------------------------------------------------- */
11250 +
11251 +/* todo: dirty? */
11252 +/* if exportfs_decode_fh() passed vfsmount*, we could be happy */
11253 +
11254 +struct au_compare_mnt_args {
11255 +       /* input */
11256 +       struct super_block *sb;
11257 +
11258 +       /* output */
11259 +       struct vfsmount *mnt;
11260 +};
11261 +
11262 +static int au_compare_mnt(struct vfsmount *mnt, void *arg)
11263 +{
11264 +       struct au_compare_mnt_args *a = arg;
11265 +
11266 +       if (mnt->mnt_sb != a->sb)
11267 +               return 0;
11268 +       a->mnt = mntget(mnt);
11269 +       return 1;
11270 +}
11271 +
11272 +static struct vfsmount *au_mnt_get(struct super_block *sb)
11273 +{
11274 +       int err;
11275 +       struct path root;
11276 +       struct au_compare_mnt_args args = {
11277 +               .sb = sb
11278 +       };
11279 +
11280 +       get_fs_root(current->fs, &root);
11281 +       rcu_read_lock();
11282 +       err = iterate_mounts(au_compare_mnt, &args, root.mnt);
11283 +       rcu_read_unlock();
11284 +       path_put(&root);
11285 +       AuDebugOn(!err);
11286 +       AuDebugOn(!args.mnt);
11287 +       return args.mnt;
11288 +}
11289 +
11290 +struct au_nfsd_si_lock {
11291 +       unsigned int sigen;
11292 +       aufs_bindex_t bindex, br_id;
11293 +       unsigned char force_lock;
11294 +};
11295 +
11296 +static int si_nfsd_read_lock(struct super_block *sb,
11297 +                            struct au_nfsd_si_lock *nsi_lock)
11298 +{
11299 +       int err;
11300 +       aufs_bindex_t bindex;
11301 +
11302 +       si_read_lock(sb, AuLock_FLUSH);
11303 +
11304 +       /* branch id may be wrapped around */
11305 +       err = 0;
11306 +       bindex = au_br_index(sb, nsi_lock->br_id);
11307 +       if (bindex >= 0 && nsi_lock->sigen + AUFS_BRANCH_MAX > au_sigen(sb))
11308 +               goto out; /* success */
11309 +
11310 +       err = -ESTALE;
11311 +       bindex = -1;
11312 +       if (!nsi_lock->force_lock)
11313 +               si_read_unlock(sb);
11314 +
11315 +out:
11316 +       nsi_lock->bindex = bindex;
11317 +       return err;
11318 +}
11319 +
11320 +struct find_name_by_ino {
11321 +       struct dir_context ctx;
11322 +       int called, found;
11323 +       ino_t ino;
11324 +       char *name;
11325 +       int namelen;
11326 +};
11327 +
11328 +static int
11329 +find_name_by_ino(struct dir_context *ctx, const char *name, int namelen,
11330 +                loff_t offset, u64 ino, unsigned int d_type)
11331 +{
11332 +       struct find_name_by_ino *a = container_of(ctx, struct find_name_by_ino,
11333 +                                                 ctx);
11334 +
11335 +       a->called++;
11336 +       if (a->ino != ino)
11337 +               return 0;
11338 +
11339 +       memcpy(a->name, name, namelen);
11340 +       a->namelen = namelen;
11341 +       a->found = 1;
11342 +       return 1;
11343 +}
11344 +
11345 +static struct dentry *au_lkup_by_ino(struct path *path, ino_t ino,
11346 +                                    struct au_nfsd_si_lock *nsi_lock)
11347 +{
11348 +       struct dentry *dentry, *parent;
11349 +       struct file *file;
11350 +       struct inode *dir;
11351 +       struct find_name_by_ino arg = {
11352 +               .ctx = {
11353 +                       .actor = find_name_by_ino
11354 +               }
11355 +       };
11356 +       int err;
11357 +
11358 +       parent = path->dentry;
11359 +       if (nsi_lock)
11360 +               si_read_unlock(parent->d_sb);
11361 +       file = vfsub_dentry_open(path, au_dir_roflags);
11362 +       dentry = (void *)file;
11363 +       if (IS_ERR(file))
11364 +               goto out;
11365 +
11366 +       dentry = ERR_PTR(-ENOMEM);
11367 +       arg.name = (void *)__get_free_page(GFP_NOFS);
11368 +       if (unlikely(!arg.name))
11369 +               goto out_file;
11370 +       arg.ino = ino;
11371 +       arg.found = 0;
11372 +       do {
11373 +               arg.called = 0;
11374 +               /* smp_mb(); */
11375 +               err = vfsub_iterate_dir(file, &arg.ctx);
11376 +       } while (!err && !arg.found && arg.called);
11377 +       dentry = ERR_PTR(err);
11378 +       if (unlikely(err))
11379 +               goto out_name;
11380 +       /* instead of ENOENT */
11381 +       dentry = ERR_PTR(-ESTALE);
11382 +       if (!arg.found)
11383 +               goto out_name;
11384 +
11385 +       /* do not call vfsub_lkup_one() */
11386 +       dir = d_inode(parent);
11387 +       dentry = vfsub_lookup_one_len_unlocked(arg.name, parent, arg.namelen);
11388 +       AuTraceErrPtr(dentry);
11389 +       if (IS_ERR(dentry))
11390 +               goto out_name;
11391 +       AuDebugOn(au_test_anon(dentry));
11392 +       if (unlikely(d_really_is_negative(dentry))) {
11393 +               dput(dentry);
11394 +               dentry = ERR_PTR(-ENOENT);
11395 +       }
11396 +
11397 +out_name:
11398 +       au_delayed_free_page((unsigned long)arg.name);
11399 +out_file:
11400 +       fput(file);
11401 +out:
11402 +       if (unlikely(nsi_lock
11403 +                    && si_nfsd_read_lock(parent->d_sb, nsi_lock) < 0))
11404 +               if (!IS_ERR(dentry)) {
11405 +                       dput(dentry);
11406 +                       dentry = ERR_PTR(-ESTALE);
11407 +               }
11408 +       AuTraceErrPtr(dentry);
11409 +       return dentry;
11410 +}
11411 +
11412 +static struct dentry *decode_by_dir_ino(struct super_block *sb, ino_t ino,
11413 +                                       ino_t dir_ino,
11414 +                                       struct au_nfsd_si_lock *nsi_lock)
11415 +{
11416 +       struct dentry *dentry;
11417 +       struct path path;
11418 +
11419 +       if (dir_ino != AUFS_ROOT_INO) {
11420 +               path.dentry = decode_by_ino(sb, dir_ino, 0);
11421 +               dentry = path.dentry;
11422 +               if (!path.dentry || IS_ERR(path.dentry))
11423 +                       goto out;
11424 +               AuDebugOn(au_test_anon(path.dentry));
11425 +       } else
11426 +               path.dentry = dget(sb->s_root);
11427 +
11428 +       path.mnt = au_mnt_get(sb);
11429 +       dentry = au_lkup_by_ino(&path, ino, nsi_lock);
11430 +       path_put(&path);
11431 +
11432 +out:
11433 +       AuTraceErrPtr(dentry);
11434 +       return dentry;
11435 +}
11436 +
11437 +/* ---------------------------------------------------------------------- */
11438 +
11439 +static int h_acceptable(void *expv, struct dentry *dentry)
11440 +{
11441 +       return 1;
11442 +}
11443 +
11444 +static char *au_build_path(struct dentry *h_parent, struct path *h_rootpath,
11445 +                          char *buf, int len, struct super_block *sb)
11446 +{
11447 +       char *p;
11448 +       int n;
11449 +       struct path path;
11450 +
11451 +       p = d_path(h_rootpath, buf, len);
11452 +       if (IS_ERR(p))
11453 +               goto out;
11454 +       n = strlen(p);
11455 +
11456 +       path.mnt = h_rootpath->mnt;
11457 +       path.dentry = h_parent;
11458 +       p = d_path(&path, buf, len);
11459 +       if (IS_ERR(p))
11460 +               goto out;
11461 +       if (n != 1)
11462 +               p += n;
11463 +
11464 +       path.mnt = au_mnt_get(sb);
11465 +       path.dentry = sb->s_root;
11466 +       p = d_path(&path, buf, len - strlen(p));
11467 +       mntput(path.mnt);
11468 +       if (IS_ERR(p))
11469 +               goto out;
11470 +       if (n != 1)
11471 +               p[strlen(p)] = '/';
11472 +
11473 +out:
11474 +       AuTraceErrPtr(p);
11475 +       return p;
11476 +}
11477 +
11478 +static
11479 +struct dentry *decode_by_path(struct super_block *sb, ino_t ino, __u32 *fh,
11480 +                             int fh_len, struct au_nfsd_si_lock *nsi_lock)
11481 +{
11482 +       struct dentry *dentry, *h_parent, *root;
11483 +       struct super_block *h_sb;
11484 +       char *pathname, *p;
11485 +       struct vfsmount *h_mnt;
11486 +       struct au_branch *br;
11487 +       int err;
11488 +       struct path path;
11489 +
11490 +       br = au_sbr(sb, nsi_lock->bindex);
11491 +       h_mnt = au_br_mnt(br);
11492 +       h_sb = h_mnt->mnt_sb;
11493 +       /* todo: call lower fh_to_dentry()? fh_to_parent()? */
11494 +       lockdep_off();
11495 +       h_parent = exportfs_decode_fh(h_mnt, (void *)(fh + Fh_tail),
11496 +                                     fh_len - Fh_tail, fh[Fh_h_type],
11497 +                                     h_acceptable, /*context*/NULL);
11498 +       lockdep_on();
11499 +       dentry = h_parent;
11500 +       if (unlikely(!h_parent || IS_ERR(h_parent))) {
11501 +               AuWarn1("%s decode_fh failed, %ld\n",
11502 +                       au_sbtype(h_sb), PTR_ERR(h_parent));
11503 +               goto out;
11504 +       }
11505 +       dentry = NULL;
11506 +       if (unlikely(au_test_anon(h_parent))) {
11507 +               AuWarn1("%s decode_fh returned a disconnected dentry\n",
11508 +                       au_sbtype(h_sb));
11509 +               goto out_h_parent;
11510 +       }
11511 +
11512 +       dentry = ERR_PTR(-ENOMEM);
11513 +       pathname = (void *)__get_free_page(GFP_NOFS);
11514 +       if (unlikely(!pathname))
11515 +               goto out_h_parent;
11516 +
11517 +       root = sb->s_root;
11518 +       path.mnt = h_mnt;
11519 +       di_read_lock_parent(root, !AuLock_IR);
11520 +       path.dentry = au_h_dptr(root, nsi_lock->bindex);
11521 +       di_read_unlock(root, !AuLock_IR);
11522 +       p = au_build_path(h_parent, &path, pathname, PAGE_SIZE, sb);
11523 +       dentry = (void *)p;
11524 +       if (IS_ERR(p))
11525 +               goto out_pathname;
11526 +
11527 +       si_read_unlock(sb);
11528 +       err = vfsub_kern_path(p, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &path);
11529 +       dentry = ERR_PTR(err);
11530 +       if (unlikely(err))
11531 +               goto out_relock;
11532 +
11533 +       dentry = ERR_PTR(-ENOENT);
11534 +       AuDebugOn(au_test_anon(path.dentry));
11535 +       if (unlikely(d_really_is_negative(path.dentry)))
11536 +               goto out_path;
11537 +
11538 +       if (ino != d_inode(path.dentry)->i_ino)
11539 +               dentry = au_lkup_by_ino(&path, ino, /*nsi_lock*/NULL);
11540 +       else
11541 +               dentry = dget(path.dentry);
11542 +
11543 +out_path:
11544 +       path_put(&path);
11545 +out_relock:
11546 +       if (unlikely(si_nfsd_read_lock(sb, nsi_lock) < 0))
11547 +               if (!IS_ERR(dentry)) {
11548 +                       dput(dentry);
11549 +                       dentry = ERR_PTR(-ESTALE);
11550 +               }
11551 +out_pathname:
11552 +       au_delayed_free_page((unsigned long)pathname);
11553 +out_h_parent:
11554 +       dput(h_parent);
11555 +out:
11556 +       AuTraceErrPtr(dentry);
11557 +       return dentry;
11558 +}
11559 +
11560 +/* ---------------------------------------------------------------------- */
11561 +
11562 +static struct dentry *
11563 +aufs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
11564 +                 int fh_type)
11565 +{
11566 +       struct dentry *dentry;
11567 +       __u32 *fh = fid->raw;
11568 +       struct au_branch *br;
11569 +       ino_t ino, dir_ino;
11570 +       struct au_nfsd_si_lock nsi_lock = {
11571 +               .force_lock     = 0
11572 +       };
11573 +
11574 +       dentry = ERR_PTR(-ESTALE);
11575 +       /* it should never happen, but the file handle is unreliable */
11576 +       if (unlikely(fh_len < Fh_tail))
11577 +               goto out;
11578 +       nsi_lock.sigen = fh[Fh_sigen];
11579 +       nsi_lock.br_id = fh[Fh_br_id];
11580 +
11581 +       /* branch id may be wrapped around */
11582 +       br = NULL;
11583 +       if (unlikely(si_nfsd_read_lock(sb, &nsi_lock)))
11584 +               goto out;
11585 +       nsi_lock.force_lock = 1;
11586 +
11587 +       /* is this inode still cached? */
11588 +       ino = decode_ino(fh + Fh_ino);
11589 +       /* it should never happen */
11590 +       if (unlikely(ino == AUFS_ROOT_INO))
11591 +               goto out_unlock;
11592 +
11593 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11594 +       dentry = decode_by_ino(sb, ino, dir_ino);
11595 +       if (IS_ERR(dentry))
11596 +               goto out_unlock;
11597 +       if (dentry)
11598 +               goto accept;
11599 +
11600 +       /* is the parent dir cached? */
11601 +       br = au_sbr(sb, nsi_lock.bindex);
11602 +       au_br_get(br);
11603 +       dentry = decode_by_dir_ino(sb, ino, dir_ino, &nsi_lock);
11604 +       if (IS_ERR(dentry))
11605 +               goto out_unlock;
11606 +       if (dentry)
11607 +               goto accept;
11608 +
11609 +       /* lookup path */
11610 +       dentry = decode_by_path(sb, ino, fh, fh_len, &nsi_lock);
11611 +       if (IS_ERR(dentry))
11612 +               goto out_unlock;
11613 +       if (unlikely(!dentry))
11614 +               /* todo?: make it ESTALE */
11615 +               goto out_unlock;
11616 +
11617 +accept:
11618 +       if (!au_digen_test(dentry, au_sigen(sb))
11619 +           && d_inode(dentry)->i_generation == fh[Fh_igen])
11620 +               goto out_unlock; /* success */
11621 +
11622 +       dput(dentry);
11623 +       dentry = ERR_PTR(-ESTALE);
11624 +out_unlock:
11625 +       if (br)
11626 +               au_br_put(br);
11627 +       si_read_unlock(sb);
11628 +out:
11629 +       AuTraceErrPtr(dentry);
11630 +       return dentry;
11631 +}
11632 +
11633 +#if 0 /* reserved for future use */
11634 +/* support subtreecheck option */
11635 +static struct dentry *aufs_fh_to_parent(struct super_block *sb, struct fid *fid,
11636 +                                       int fh_len, int fh_type)
11637 +{
11638 +       struct dentry *parent;
11639 +       __u32 *fh = fid->raw;
11640 +       ino_t dir_ino;
11641 +
11642 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11643 +       parent = decode_by_ino(sb, dir_ino, 0);
11644 +       if (IS_ERR(parent))
11645 +               goto out;
11646 +       if (!parent)
11647 +               parent = decode_by_path(sb, au_br_index(sb, fh[Fh_br_id]),
11648 +                                       dir_ino, fh, fh_len);
11649 +
11650 +out:
11651 +       AuTraceErrPtr(parent);
11652 +       return parent;
11653 +}
11654 +#endif
11655 +
11656 +/* ---------------------------------------------------------------------- */
11657 +
11658 +static int aufs_encode_fh(struct inode *inode, __u32 *fh, int *max_len,
11659 +                         struct inode *dir)
11660 +{
11661 +       int err;
11662 +       aufs_bindex_t bindex;
11663 +       struct super_block *sb, *h_sb;
11664 +       struct dentry *dentry, *parent, *h_parent;
11665 +       struct inode *h_dir;
11666 +       struct au_branch *br;
11667 +
11668 +       err = -ENOSPC;
11669 +       if (unlikely(*max_len <= Fh_tail)) {
11670 +               AuWarn1("NFSv2 client (max_len %d)?\n", *max_len);
11671 +               goto out;
11672 +       }
11673 +
11674 +       err = FILEID_ROOT;
11675 +       if (inode->i_ino == AUFS_ROOT_INO) {
11676 +               AuDebugOn(inode->i_ino != AUFS_ROOT_INO);
11677 +               goto out;
11678 +       }
11679 +
11680 +       h_parent = NULL;
11681 +       sb = inode->i_sb;
11682 +       err = si_read_lock(sb, AuLock_FLUSH);
11683 +       if (unlikely(err))
11684 +               goto out;
11685 +
11686 +#ifdef CONFIG_AUFS_DEBUG
11687 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
11688 +               AuWarn1("NFS-exporting requires xino\n");
11689 +#endif
11690 +       err = -EIO;
11691 +       parent = NULL;
11692 +       ii_read_lock_child(inode);
11693 +       bindex = au_ibtop(inode);
11694 +       if (!dir) {
11695 +               dentry = d_find_any_alias(inode);
11696 +               if (unlikely(!dentry))
11697 +                       goto out_unlock;
11698 +               AuDebugOn(au_test_anon(dentry));
11699 +               parent = dget_parent(dentry);
11700 +               dput(dentry);
11701 +               if (unlikely(!parent))
11702 +                       goto out_unlock;
11703 +               if (d_really_is_positive(parent))
11704 +                       dir = d_inode(parent);
11705 +       }
11706 +
11707 +       ii_read_lock_parent(dir);
11708 +       h_dir = au_h_iptr(dir, bindex);
11709 +       ii_read_unlock(dir);
11710 +       if (unlikely(!h_dir))
11711 +               goto out_parent;
11712 +       h_parent = d_find_any_alias(h_dir);
11713 +       if (unlikely(!h_parent))
11714 +               goto out_hparent;
11715 +
11716 +       err = -EPERM;
11717 +       br = au_sbr(sb, bindex);
11718 +       h_sb = au_br_sb(br);
11719 +       if (unlikely(!h_sb->s_export_op)) {
11720 +               AuErr1("%s branch is not exportable\n", au_sbtype(h_sb));
11721 +               goto out_hparent;
11722 +       }
11723 +
11724 +       fh[Fh_br_id] = br->br_id;
11725 +       fh[Fh_sigen] = au_sigen(sb);
11726 +       encode_ino(fh + Fh_ino, inode->i_ino);
11727 +       encode_ino(fh + Fh_dir_ino, dir->i_ino);
11728 +       fh[Fh_igen] = inode->i_generation;
11729 +
11730 +       *max_len -= Fh_tail;
11731 +       fh[Fh_h_type] = exportfs_encode_fh(h_parent, (void *)(fh + Fh_tail),
11732 +                                          max_len,
11733 +                                          /*connectable or subtreecheck*/0);
11734 +       err = fh[Fh_h_type];
11735 +       *max_len += Fh_tail;
11736 +       /* todo: macros? */
11737 +       if (err != FILEID_INVALID)
11738 +               err = 99;
11739 +       else
11740 +               AuWarn1("%s encode_fh failed\n", au_sbtype(h_sb));
11741 +
11742 +out_hparent:
11743 +       dput(h_parent);
11744 +out_parent:
11745 +       dput(parent);
11746 +out_unlock:
11747 +       ii_read_unlock(inode);
11748 +       si_read_unlock(sb);
11749 +out:
11750 +       if (unlikely(err < 0))
11751 +               err = FILEID_INVALID;
11752 +       return err;
11753 +}
11754 +
11755 +/* ---------------------------------------------------------------------- */
11756 +
11757 +static int aufs_commit_metadata(struct inode *inode)
11758 +{
11759 +       int err;
11760 +       aufs_bindex_t bindex;
11761 +       struct super_block *sb;
11762 +       struct inode *h_inode;
11763 +       int (*f)(struct inode *inode);
11764 +
11765 +       sb = inode->i_sb;
11766 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
11767 +       ii_write_lock_child(inode);
11768 +       bindex = au_ibtop(inode);
11769 +       AuDebugOn(bindex < 0);
11770 +       h_inode = au_h_iptr(inode, bindex);
11771 +
11772 +       f = h_inode->i_sb->s_export_op->commit_metadata;
11773 +       if (f)
11774 +               err = f(h_inode);
11775 +       else {
11776 +               struct writeback_control wbc = {
11777 +                       .sync_mode      = WB_SYNC_ALL,
11778 +                       .nr_to_write    = 0 /* metadata only */
11779 +               };
11780 +
11781 +               err = sync_inode(h_inode, &wbc);
11782 +       }
11783 +
11784 +       au_cpup_attr_timesizes(inode);
11785 +       ii_write_unlock(inode);
11786 +       si_read_unlock(sb);
11787 +       return err;
11788 +}
11789 +
11790 +/* ---------------------------------------------------------------------- */
11791 +
11792 +static struct export_operations aufs_export_op = {
11793 +       .fh_to_dentry           = aufs_fh_to_dentry,
11794 +       /* .fh_to_parent        = aufs_fh_to_parent, */
11795 +       .encode_fh              = aufs_encode_fh,
11796 +       .commit_metadata        = aufs_commit_metadata
11797 +};
11798 +
11799 +void au_export_init(struct super_block *sb)
11800 +{
11801 +       struct au_sbinfo *sbinfo;
11802 +       __u32 u;
11803 +
11804 +       BUILD_BUG_ON_MSG(IS_BUILTIN(CONFIG_AUFS_FS)
11805 +                        && IS_MODULE(CONFIG_EXPORTFS),
11806 +                        AUFS_NAME ": unsupported configuration "
11807 +                        "CONFIG_EXPORTFS=m and CONFIG_AUFS_FS=y");
11808 +
11809 +       sb->s_export_op = &aufs_export_op;
11810 +       sbinfo = au_sbi(sb);
11811 +       sbinfo->si_xigen = NULL;
11812 +       get_random_bytes(&u, sizeof(u));
11813 +       BUILD_BUG_ON(sizeof(u) != sizeof(int));
11814 +       atomic_set(&sbinfo->si_xigen_next, u);
11815 +}
11816 diff -urN /usr/share/empty/fs/aufs/fhsm.c linux/fs/aufs/fhsm.c
11817 --- /usr/share/empty/fs/aufs/fhsm.c     1970-01-01 01:00:00.000000000 +0100
11818 +++ linux/fs/aufs/fhsm.c        2016-10-09 16:55:36.489368218 +0200
11819 @@ -0,0 +1,426 @@
11820 +/*
11821 + * Copyright (C) 2011-2016 Junjiro R. Okajima
11822 + *
11823 + * This program, aufs is free software; you can redistribute it and/or modify
11824 + * it under the terms of the GNU General Public License as published by
11825 + * the Free Software Foundation; either version 2 of the License, or
11826 + * (at your option) any later version.
11827 + *
11828 + * This program is distributed in the hope that it will be useful,
11829 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11830 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11831 + * GNU General Public License for more details.
11832 + *
11833 + * You should have received a copy of the GNU General Public License
11834 + * along with this program; if not, write to the Free Software
11835 + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
11836 + */
11837 +
11838 +/*
11839 + * File-based Hierarchy Storage Management
11840 + */
11841 +
11842 +#include <linux/anon_inodes.h>
11843 +#include <linux/poll.h>
11844 +#include <linux/seq_file.h>
11845 +#include <linux/statfs.h>
11846 +#include "aufs.h"
11847 +
11848 +static aufs_bindex_t au_fhsm_bottom(struct super_block *sb)
11849 +{
11850 +       struct au_sbinfo *sbinfo;
11851 +       struct au_fhsm *fhsm;
11852 +
11853 +       SiMustAnyLock(sb);
11854 +
11855 +       sbinfo = au_sbi(sb);
11856 +       fhsm = &sbinfo->si_fhsm;
11857 +       AuDebugOn(!fhsm);
11858 +       return fhsm->fhsm_bottom;
11859 +}
11860 +
11861 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex)
11862 +{
11863 +       struct au_sbinfo *sbinfo;
11864 +       struct au_fhsm *fhsm;
11865 +
11866 +       SiMustWriteLock(sb);
11867 +
11868 +       sbinfo = au_sbi(sb);
11869 +       fhsm = &sbinfo->si_fhsm;
11870 +       AuDebugOn(!fhsm);
11871 +       fhsm->fhsm_bottom = bindex;
11872 +}
11873 +
11874 +/* ---------------------------------------------------------------------- */
11875 +
11876 +static int au_fhsm_test_jiffy(struct au_sbinfo *sbinfo, struct au_branch *br)
11877 +{
11878 +       struct au_br_fhsm *bf;
11879 +
11880 +       bf = br->br_fhsm;
11881 +       MtxMustLock(&bf->bf_lock);
11882 +
11883 +       return !bf->bf_readable
11884 +               || time_after(jiffies,
11885 +                             bf->bf_jiffy + sbinfo->si_fhsm.fhsm_expire);
11886 +}
11887 +
11888 +/* ---------------------------------------------------------------------- */
11889 +
11890 +static void au_fhsm_notify(struct super_block *sb, int val)
11891 +{
11892 +       struct au_sbinfo *sbinfo;
11893 +       struct au_fhsm *fhsm;
11894 +
11895 +       SiMustAnyLock(sb);
11896 +
11897 +       sbinfo = au_sbi(sb);
11898 +       fhsm = &sbinfo->si_fhsm;
11899 +       if (au_fhsm_pid(fhsm)
11900 +           && atomic_read(&fhsm->fhsm_readable) != -1) {
11901 +               atomic_set(&fhsm->fhsm_readable, val);
11902 +               if (val)
11903 +                       wake_up(&fhsm->fhsm_wqh);
11904 +       }
11905 +}
11906 +
11907 +static int au_fhsm_stfs(struct super_block *sb, aufs_bindex_t bindex,
11908 +                       struct aufs_stfs *rstfs, int do_lock, int do_notify)
11909 +{
11910 +       int err;
11911 +       struct au_branch *br;
11912 +       struct au_br_fhsm *bf;
11913 +
11914 +       br = au_sbr(sb, bindex);
11915 +       AuDebugOn(au_br_rdonly(br));
11916 +       bf = br->br_fhsm;
11917 +       AuDebugOn(!bf);
11918 +
11919 +       if (do_lock)
11920 +               mutex_lock(&bf->bf_lock);
11921 +       else
11922 +               MtxMustLock(&bf->bf_lock);
11923 +
11924 +       /* sb->s_root for NFS is unreliable */
11925 +       err = au_br_stfs(br, &bf->bf_stfs);
11926 +       if (unlikely(err)) {
11927 +               AuErr1("FHSM failed (%d), b%d, ignored.\n", bindex, err);
11928 +               goto out;
11929 +       }
11930 +
11931 +       bf->bf_jiffy = jiffies;
11932 +       bf->bf_readable = 1;
11933 +       if (do_notify)
11934 +               au_fhsm_notify(sb, /*val*/1);
11935 +       if (rstfs)
11936 +               *rstfs = bf->bf_stfs;
11937 +
11938 +out:
11939 +       if (do_lock)
11940 +               mutex_unlock(&bf->bf_lock);
11941 +       au_fhsm_notify(sb, /*val*/1);
11942 +
11943 +       return err;
11944 +}
11945 +
11946 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force)
11947 +{
11948 +       int err;
11949 +       struct au_sbinfo *sbinfo;
11950 +       struct au_fhsm *fhsm;
11951 +       struct au_branch *br;
11952 +       struct au_br_fhsm *bf;
11953 +
11954 +       AuDbg("b%d, force %d\n", bindex, force);
11955 +       SiMustAnyLock(sb);
11956 +
11957 +       sbinfo = au_sbi(sb);
11958 +       fhsm = &sbinfo->si_fhsm;
11959 +       if (!au_ftest_si(sbinfo, FHSM)
11960 +           || fhsm->fhsm_bottom == bindex)
11961 +               return;
11962 +
11963 +       br = au_sbr(sb, bindex);
11964 +       bf = br->br_fhsm;
11965 +       AuDebugOn(!bf);
11966 +       mutex_lock(&bf->bf_lock);
11967 +       if (force
11968 +           || au_fhsm_pid(fhsm)
11969 +           || au_fhsm_test_jiffy(sbinfo, br))
11970 +               err = au_fhsm_stfs(sb, bindex, /*rstfs*/NULL, /*do_lock*/0,
11971 +                                 /*do_notify*/1);
11972 +       mutex_unlock(&bf->bf_lock);
11973 +}
11974 +
11975 +void au_fhsm_wrote_all(struct super_block *sb, int force)
11976 +{
11977 +       aufs_bindex_t bindex, bbot;
11978 +       struct au_branch *br;
11979 +
11980 +       /* exclude the bottom */
11981 +       bbot = au_fhsm_bottom(sb);
11982 +       for (bindex = 0; bindex < bbot; bindex++) {
11983 +               br = au_sbr(sb, bindex);
11984 +               if (au_br_fhsm(br->br_perm))
11985 +                       au_fhsm_wrote(sb, bindex, force);
11986 +       }
11987 +}
11988 +
11989 +/* ---------------------------------------------------------------------- */
11990 +
11991 +static unsigned int au_fhsm_poll(struct file *file,
11992 +                                struct poll_table_struct *wait)
11993 +{
11994 +       unsigned int mask;
11995 +       struct au_sbinfo *sbinfo;
11996 +       struct au_fhsm *fhsm;
11997 +
11998 +       mask = 0;
11999 +       sbinfo = file->private_data;
12000 +       fhsm = &sbinfo->si_fhsm;
12001 +       poll_wait(file, &fhsm->fhsm_wqh, wait);
12002 +       if (atomic_read(&fhsm->fhsm_readable))
12003 +               mask = POLLIN /* | POLLRDNORM */;
12004 +
12005 +       AuTraceErr((int)mask);
12006 +       return mask;
12007 +}
12008 +
12009 +static int au_fhsm_do_read_one(struct aufs_stbr __user *stbr,
12010 +                             struct aufs_stfs *stfs, __s16 brid)
12011 +{
12012 +       int err;
12013 +
12014 +       err = copy_to_user(&stbr->stfs, stfs, sizeof(*stfs));
12015 +       if (!err)
12016 +               err = __put_user(brid, &stbr->brid);
12017 +       if (unlikely(err))
12018 +               err = -EFAULT;
12019 +
12020 +       return err;
12021 +}
12022 +
12023 +static ssize_t au_fhsm_do_read(struct super_block *sb,
12024 +                              struct aufs_stbr __user *stbr, size_t count)
12025 +{
12026 +       ssize_t err;
12027 +       int nstbr;
12028 +       aufs_bindex_t bindex, bbot;
12029 +       struct au_branch *br;
12030 +       struct au_br_fhsm *bf;
12031 +
12032 +       /* except the bottom branch */
12033 +       err = 0;
12034 +       nstbr = 0;
12035 +       bbot = au_fhsm_bottom(sb);
12036 +       for (bindex = 0; !err && bindex < bbot; bindex++) {
12037 +               br = au_sbr(sb, bindex);
12038 +               if (!au_br_fhsm(br->br_perm))
12039 +                       continue;
12040 +
12041 +               bf = br->br_fhsm;
12042 +               mutex_lock(&bf->bf_lock);
12043 +               if (bf->bf_readable) {
12044 +                       err = -EFAULT;
12045 +                       if (count >= sizeof(*stbr))
12046 +                               err = au_fhsm_do_read_one(stbr++, &bf->bf_stfs,
12047 +                                                         br->br_id);
12048 +                       if (!err) {
12049 +                               bf->bf_readable = 0;
12050 +                               count -= sizeof(*stbr);
12051 +                               nstbr++;
12052 +                       }
12053 +               }
12054 +               mutex_unlock(&bf->bf_lock);
12055 +       }
12056 +       if (!err)
12057 +               err = sizeof(*stbr) * nstbr;
12058 +
12059 +       return err;
12060 +}
12061 +
12062 +static ssize_t au_fhsm_read(struct file *file, char __user *buf, size_t count,
12063 +                          loff_t *pos)
12064 +{
12065 +       ssize_t err;
12066 +       int readable;
12067 +       aufs_bindex_t nfhsm, bindex, bbot;
12068 +       struct au_sbinfo *sbinfo;
12069 +       struct au_fhsm *fhsm;
12070 +       struct au_branch *br;
12071 +       struct super_block *sb;
12072 +
12073 +       err = 0;
12074 +       sbinfo = file->private_data;
12075 +       fhsm = &sbinfo->si_fhsm;
12076 +need_data:
12077 +       spin_lock_irq(&fhsm->fhsm_wqh.lock);
12078 +       if (!atomic_read(&fhsm->fhsm_readable)) {
12079 +               if (vfsub_file_flags(file) & O_NONBLOCK)
12080 +                       err = -EAGAIN;
12081 +               else
12082 +                       err = wait_event_interruptible_locked_irq
12083 +                               (fhsm->fhsm_wqh,
12084 +                                atomic_read(&fhsm->fhsm_readable));
12085 +       }
12086 +       spin_unlock_irq(&fhsm->fhsm_wqh.lock);
12087 +       if (unlikely(err))
12088 +               goto out;
12089 +
12090 +       /* sb may already be dead */
12091 +       au_rw_read_lock(&sbinfo->si_rwsem);
12092 +       readable = atomic_read(&fhsm->fhsm_readable);
12093 +       if (readable > 0) {
12094 +               sb = sbinfo->si_sb;
12095 +               AuDebugOn(!sb);
12096 +               /* exclude the bottom branch */
12097 +               nfhsm = 0;
12098 +               bbot = au_fhsm_bottom(sb);
12099 +               for (bindex = 0; bindex < bbot; bindex++) {
12100 +                       br = au_sbr(sb, bindex);
12101 +                       if (au_br_fhsm(br->br_perm))
12102 +                               nfhsm++;
12103 +               }
12104 +               err = -EMSGSIZE;
12105 +               if (nfhsm * sizeof(struct aufs_stbr) <= count) {
12106 +                       atomic_set(&fhsm->fhsm_readable, 0);
12107 +                       err = au_fhsm_do_read(sbinfo->si_sb, (void __user *)buf,
12108 +                                            count);
12109 +               }
12110 +       }
12111 +       au_rw_read_unlock(&sbinfo->si_rwsem);
12112 +       if (!readable)
12113 +               goto need_data;
12114 +
12115 +out:
12116 +       return err;
12117 +}
12118 +
12119 +static int au_fhsm_release(struct inode *inode, struct file *file)
12120 +{
12121 +       struct au_sbinfo *sbinfo;
12122 +       struct au_fhsm *fhsm;
12123 +
12124 +       /* sb may already be dead */
12125 +       sbinfo = file->private_data;
12126 +       fhsm = &sbinfo->si_fhsm;
12127 +       spin_lock(&fhsm->fhsm_spin);
12128 +       fhsm->fhsm_pid = 0;
12129 +       spin_unlock(&fhsm->fhsm_spin);
12130 +       kobject_put(&sbinfo->si_kobj);
12131 +
12132 +       return 0;
12133 +}
12134 +
12135 +static const struct file_operations au_fhsm_fops = {
12136 +       .owner          = THIS_MODULE,
12137 +       .llseek         = noop_llseek,
12138 +       .read           = au_fhsm_read,
12139 +       .poll           = au_fhsm_poll,
12140 +       .release        = au_fhsm_release
12141 +};
12142 +
12143 +int au_fhsm_fd(struct super_block *sb, int oflags)
12144 +{
12145 +       int err, fd;
12146 +       struct au_sbinfo *sbinfo;
12147 +       struct au_fhsm *fhsm;
12148 +
12149 +       err = -EPERM;
12150 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
12151 +               goto out;
12152 +
12153 +       err = -EINVAL;
12154 +       if (unlikely(oflags & ~(O_CLOEXEC | O_NONBLOCK)))
12155 +               goto out;
12156 +
12157 +       err = 0;
12158 +       sbinfo = au_sbi(sb);
12159 +       fhsm = &sbinfo->si_fhsm;
12160 +       spin_lock(&fhsm->fhsm_spin);
12161 +       if (!fhsm->fhsm_pid)
12162 +               fhsm->fhsm_pid = current->pid;
12163 +       else
12164 +               err = -EBUSY;
12165 +       spin_unlock(&fhsm->fhsm_spin);
12166 +       if (unlikely(err))
12167 +               goto out;
12168 +
12169 +       oflags |= O_RDONLY;
12170 +       /* oflags |= FMODE_NONOTIFY; */
12171 +       fd = anon_inode_getfd("[aufs_fhsm]", &au_fhsm_fops, sbinfo, oflags);
12172 +       err = fd;
12173 +       if (unlikely(fd < 0))
12174 +               goto out_pid;
12175 +
12176 +       /* succeed reglardless 'fhsm' status */
12177 +       kobject_get(&sbinfo->si_kobj);
12178 +       si_noflush_read_lock(sb);
12179 +       if (au_ftest_si(sbinfo, FHSM))
12180 +               au_fhsm_wrote_all(sb, /*force*/0);
12181 +       si_read_unlock(sb);
12182 +       goto out; /* success */
12183 +
12184 +out_pid:
12185 +       spin_lock(&fhsm->fhsm_spin);
12186 +       fhsm->fhsm_pid = 0;
12187 +       spin_unlock(&fhsm->fhsm_spin);
12188 +out:
12189 +       AuTraceErr(err);
12190 +       return err;
12191 +}
12192 +
12193 +/* ---------------------------------------------------------------------- */
12194 +
12195 +int au_fhsm_br_alloc(struct au_branch *br)
12196 +{
12197 +       int err;
12198 +
12199 +       err = 0;
12200 +       br->br_fhsm = kmalloc(sizeof(*br->br_fhsm), GFP_NOFS);
12201 +       if (br->br_fhsm)
12202 +               au_br_fhsm_init(br->br_fhsm);
12203 +       else
12204 +               err = -ENOMEM;
12205 +
12206 +       return err;
12207 +}
12208 +
12209 +/* ---------------------------------------------------------------------- */
12210 +
12211 +void au_fhsm_fin(struct super_block *sb)
12212 +{
12213 +       au_fhsm_notify(sb, /*val*/-1);
12214 +}
12215 +
12216 +void au_fhsm_init(struct au_sbinfo *sbinfo)
12217 +{
12218 +       struct au_fhsm *fhsm;
12219 +
12220 +       fhsm = &sbinfo->si_fhsm;
12221 +       spin_lock_init(&fhsm->fhsm_spin);
12222 +       init_waitqueue_head(&fhsm->fhsm_wqh);
12223 +       atomic_set(&fhsm->fhsm_readable, 0);
12224 +       fhsm->fhsm_expire
12225 +               = msecs_to_jiffies(AUFS_FHSM_CACHE_DEF_SEC * MSEC_PER_SEC);
12226 +       fhsm->fhsm_bottom = -1;
12227 +}
12228 +
12229 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec)
12230 +{
12231 +       sbinfo->si_fhsm.fhsm_expire
12232 +               = msecs_to_jiffies(sec * MSEC_PER_SEC);
12233 +}
12234 +
12235 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo)
12236 +{
12237 +       unsigned int u;
12238 +
12239 +       if (!au_ftest_si(sbinfo, FHSM))
12240 +               return;
12241 +
12242 +       u = jiffies_to_msecs(sbinfo->si_fhsm.fhsm_expire) / MSEC_PER_SEC;
12243 +       if (u != AUFS_FHSM_CACHE_DEF_SEC)
12244 +               seq_printf(seq, ",fhsm_sec=%u", u);
12245 +}
12246 diff -urN /usr/share/empty/fs/aufs/file.c linux/fs/aufs/file.c
12247 --- /usr/share/empty/fs/aufs/file.c     1970-01-01 01:00:00.000000000 +0100
12248 +++ linux/fs/aufs/file.c        2016-10-09 16:55:38.889431135 +0200
12249 @@ -0,0 +1,857 @@
12250 +/*
12251 + * Copyright (C) 2005-2016 Junjiro R. Okajima
12252 + *
12253 + * This program, aufs is free software; you can redistribute it and/or modify
12254 + * it under the terms of the GNU General Public License as published by
12255 + * the Free Software Foundation; either version 2 of the License, or
12256 + * (at your option) any later version.
12257 + *
12258 + * This program is distributed in the hope that it will be useful,
12259 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12260 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12261 + * GNU General Public License for more details.
12262 + *
12263 + * You should have received a copy of the GNU General Public License
12264 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12265 + */
12266 +
12267 +/*
12268 + * handling file/dir, and address_space operation
12269 + */
12270 +
12271 +#ifdef CONFIG_AUFS_DEBUG
12272 +#include <linux/migrate.h>
12273 +#endif
12274 +#include <linux/pagemap.h>
12275 +#include "aufs.h"
12276 +
12277 +/* drop flags for writing */
12278 +unsigned int au_file_roflags(unsigned int flags)
12279 +{
12280 +       flags &= ~(O_WRONLY | O_RDWR | O_APPEND | O_CREAT | O_TRUNC);
12281 +       flags |= O_RDONLY | O_NOATIME;
12282 +       return flags;
12283 +}
12284 +
12285 +/* common functions to regular file and dir */
12286 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
12287 +                      struct file *file, int force_wr)
12288 +{
12289 +       struct file *h_file;
12290 +       struct dentry *h_dentry;
12291 +       struct inode *h_inode;
12292 +       struct super_block *sb;
12293 +       struct au_branch *br;
12294 +       struct path h_path;
12295 +       int err;
12296 +
12297 +       /* a race condition can happen between open and unlink/rmdir */
12298 +       h_file = ERR_PTR(-ENOENT);
12299 +       h_dentry = au_h_dptr(dentry, bindex);
12300 +       if (au_test_nfsd() && (!h_dentry || d_is_negative(h_dentry)))
12301 +               goto out;
12302 +       h_inode = d_inode(h_dentry);
12303 +       spin_lock(&h_dentry->d_lock);
12304 +       err = (!d_unhashed(dentry) && d_unlinked(h_dentry))
12305 +               /* || !d_inode(dentry)->i_nlink */
12306 +               ;
12307 +       spin_unlock(&h_dentry->d_lock);
12308 +       if (unlikely(err))
12309 +               goto out;
12310 +
12311 +       sb = dentry->d_sb;
12312 +       br = au_sbr(sb, bindex);
12313 +       err = au_br_test_oflag(flags, br);
12314 +       h_file = ERR_PTR(err);
12315 +       if (unlikely(err))
12316 +               goto out;
12317 +
12318 +       /* drop flags for writing */
12319 +       if (au_test_ro(sb, bindex, d_inode(dentry))) {
12320 +               if (force_wr && !(flags & O_WRONLY))
12321 +                       force_wr = 0;
12322 +               flags = au_file_roflags(flags);
12323 +               if (force_wr) {
12324 +                       h_file = ERR_PTR(-EROFS);
12325 +                       flags = au_file_roflags(flags);
12326 +                       if (unlikely(vfsub_native_ro(h_inode)
12327 +                                    || IS_APPEND(h_inode)))
12328 +                               goto out;
12329 +                       flags &= ~O_ACCMODE;
12330 +                       flags |= O_WRONLY;
12331 +               }
12332 +       }
12333 +       flags &= ~O_CREAT;
12334 +       au_br_get(br);
12335 +       h_path.dentry = h_dentry;
12336 +       h_path.mnt = au_br_mnt(br);
12337 +       h_file = vfsub_dentry_open(&h_path, flags);
12338 +       if (IS_ERR(h_file))
12339 +               goto out_br;
12340 +
12341 +       if (flags & __FMODE_EXEC) {
12342 +               err = deny_write_access(h_file);
12343 +               if (unlikely(err)) {
12344 +                       fput(h_file);
12345 +                       h_file = ERR_PTR(err);
12346 +                       goto out_br;
12347 +               }
12348 +       }
12349 +       fsnotify_open(h_file);
12350 +       goto out; /* success */
12351 +
12352 +out_br:
12353 +       au_br_put(br);
12354 +out:
12355 +       return h_file;
12356 +}
12357 +
12358 +static int au_cmoo(struct dentry *dentry)
12359 +{
12360 +       int err, cmoo;
12361 +       unsigned int udba;
12362 +       struct path h_path;
12363 +       struct au_pin pin;
12364 +       struct au_cp_generic cpg = {
12365 +               .dentry = dentry,
12366 +               .bdst   = -1,
12367 +               .bsrc   = -1,
12368 +               .len    = -1,
12369 +               .pin    = &pin,
12370 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
12371 +       };
12372 +       struct inode *delegated;
12373 +       struct super_block *sb;
12374 +       struct au_sbinfo *sbinfo;
12375 +       struct au_fhsm *fhsm;
12376 +       pid_t pid;
12377 +       struct au_branch *br;
12378 +       struct dentry *parent;
12379 +       struct au_hinode *hdir;
12380 +
12381 +       DiMustWriteLock(dentry);
12382 +       IiMustWriteLock(d_inode(dentry));
12383 +
12384 +       err = 0;
12385 +       if (IS_ROOT(dentry))
12386 +               goto out;
12387 +       cpg.bsrc = au_dbtop(dentry);
12388 +       if (!cpg.bsrc)
12389 +               goto out;
12390 +
12391 +       sb = dentry->d_sb;
12392 +       sbinfo = au_sbi(sb);
12393 +       fhsm = &sbinfo->si_fhsm;
12394 +       pid = au_fhsm_pid(fhsm);
12395 +       if (pid
12396 +           && (current->pid == pid
12397 +               || current->real_parent->pid == pid))
12398 +               goto out;
12399 +
12400 +       br = au_sbr(sb, cpg.bsrc);
12401 +       cmoo = au_br_cmoo(br->br_perm);
12402 +       if (!cmoo)
12403 +               goto out;
12404 +       if (!d_is_reg(dentry))
12405 +               cmoo &= AuBrAttr_COO_ALL;
12406 +       if (!cmoo)
12407 +               goto out;
12408 +
12409 +       parent = dget_parent(dentry);
12410 +       di_write_lock_parent(parent);
12411 +       err = au_wbr_do_copyup_bu(dentry, cpg.bsrc - 1);
12412 +       cpg.bdst = err;
12413 +       if (unlikely(err < 0)) {
12414 +               err = 0;        /* there is no upper writable branch */
12415 +               goto out_dgrade;
12416 +       }
12417 +       AuDbg("bsrc %d, bdst %d\n", cpg.bsrc, cpg.bdst);
12418 +
12419 +       /* do not respect the coo attrib for the target branch */
12420 +       err = au_cpup_dirs(dentry, cpg.bdst);
12421 +       if (unlikely(err))
12422 +               goto out_dgrade;
12423 +
12424 +       di_downgrade_lock(parent, AuLock_IR);
12425 +       udba = au_opt_udba(sb);
12426 +       err = au_pin(&pin, dentry, cpg.bdst, udba,
12427 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12428 +       if (unlikely(err))
12429 +               goto out_parent;
12430 +
12431 +       err = au_sio_cpup_simple(&cpg);
12432 +       au_unpin(&pin);
12433 +       if (unlikely(err))
12434 +               goto out_parent;
12435 +       if (!(cmoo & AuBrWAttr_MOO))
12436 +               goto out_parent; /* success */
12437 +
12438 +       err = au_pin(&pin, dentry, cpg.bsrc, udba,
12439 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12440 +       if (unlikely(err))
12441 +               goto out_parent;
12442 +
12443 +       h_path.mnt = au_br_mnt(br);
12444 +       h_path.dentry = au_h_dptr(dentry, cpg.bsrc);
12445 +       hdir = au_hi(d_inode(parent), cpg.bsrc);
12446 +       delegated = NULL;
12447 +       err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated, /*force*/1);
12448 +       au_unpin(&pin);
12449 +       /* todo: keep h_dentry or not? */
12450 +       if (unlikely(err == -EWOULDBLOCK)) {
12451 +               pr_warn("cannot retry for NFSv4 delegation"
12452 +                       " for an internal unlink\n");
12453 +               iput(delegated);
12454 +       }
12455 +       if (unlikely(err)) {
12456 +               pr_err("unlink %pd after coo failed (%d), ignored\n",
12457 +                      dentry, err);
12458 +               err = 0;
12459 +       }
12460 +       goto out_parent; /* success */
12461 +
12462 +out_dgrade:
12463 +       di_downgrade_lock(parent, AuLock_IR);
12464 +out_parent:
12465 +       di_read_unlock(parent, AuLock_IR);
12466 +       dput(parent);
12467 +out:
12468 +       AuTraceErr(err);
12469 +       return err;
12470 +}
12471 +
12472 +int au_do_open(struct file *file, struct au_do_open_args *args)
12473 +{
12474 +       int err, no_lock = args->no_lock;
12475 +       struct dentry *dentry;
12476 +       struct au_finfo *finfo;
12477 +
12478 +       if (!no_lock)
12479 +               err = au_finfo_init(file, args->fidir);
12480 +       else {
12481 +               lockdep_off();
12482 +               err = au_finfo_init(file, args->fidir);
12483 +               lockdep_on();
12484 +       }
12485 +       if (unlikely(err))
12486 +               goto out;
12487 +
12488 +       dentry = file->f_path.dentry;
12489 +       AuDebugOn(IS_ERR_OR_NULL(dentry));
12490 +       if (!no_lock) {
12491 +               di_write_lock_child(dentry);
12492 +               err = au_cmoo(dentry);
12493 +               di_downgrade_lock(dentry, AuLock_IR);
12494 +               if (!err)
12495 +                       err = args->open(file, vfsub_file_flags(file), NULL);
12496 +               di_read_unlock(dentry, AuLock_IR);
12497 +       } else {
12498 +               err = au_cmoo(dentry);
12499 +               if (!err)
12500 +                       err = args->open(file, vfsub_file_flags(file),
12501 +                                        args->h_file);
12502 +               if (!err && au_fbtop(file) != au_dbtop(dentry))
12503 +                       /*
12504 +                        * cmoo happens after h_file was opened.
12505 +                        * need to refresh file later.
12506 +                        */
12507 +                       atomic_dec(&au_fi(file)->fi_generation);
12508 +       }
12509 +
12510 +       finfo = au_fi(file);
12511 +       if (!err) {
12512 +               finfo->fi_file = file;
12513 +               au_sphl_add(&finfo->fi_hlist,
12514 +                           &au_sbi(file->f_path.dentry->d_sb)->si_files);
12515 +       }
12516 +       if (!no_lock)
12517 +               fi_write_unlock(file);
12518 +       else {
12519 +               lockdep_off();
12520 +               fi_write_unlock(file);
12521 +               lockdep_on();
12522 +       }
12523 +       if (unlikely(err)) {
12524 +               finfo->fi_hdir = NULL;
12525 +               au_finfo_fin(file, /*atonce*/0);
12526 +       }
12527 +
12528 +out:
12529 +       return err;
12530 +}
12531 +
12532 +int au_reopen_nondir(struct file *file)
12533 +{
12534 +       int err;
12535 +       aufs_bindex_t btop;
12536 +       struct dentry *dentry;
12537 +       struct file *h_file, *h_file_tmp;
12538 +
12539 +       dentry = file->f_path.dentry;
12540 +       btop = au_dbtop(dentry);
12541 +       h_file_tmp = NULL;
12542 +       if (au_fbtop(file) == btop) {
12543 +               h_file = au_hf_top(file);
12544 +               if (file->f_mode == h_file->f_mode)
12545 +                       return 0; /* success */
12546 +               h_file_tmp = h_file;
12547 +               get_file(h_file_tmp);
12548 +               au_set_h_fptr(file, btop, NULL);
12549 +       }
12550 +       AuDebugOn(au_fi(file)->fi_hdir);
12551 +       /*
12552 +        * it can happen
12553 +        * file exists on both of rw and ro
12554 +        * open --> dbtop and fbtop are both 0
12555 +        * prepend a branch as rw, "rw" become ro
12556 +        * remove rw/file
12557 +        * delete the top branch, "rw" becomes rw again
12558 +        *      --> dbtop is 1, fbtop is still 0
12559 +        * write --> fbtop is 0 but dbtop is 1
12560 +        */
12561 +       /* AuDebugOn(au_fbtop(file) < btop); */
12562 +
12563 +       h_file = au_h_open(dentry, btop, vfsub_file_flags(file) & ~O_TRUNC,
12564 +                          file, /*force_wr*/0);
12565 +       err = PTR_ERR(h_file);
12566 +       if (IS_ERR(h_file)) {
12567 +               if (h_file_tmp) {
12568 +                       au_sbr_get(dentry->d_sb, btop);
12569 +                       au_set_h_fptr(file, btop, h_file_tmp);
12570 +                       h_file_tmp = NULL;
12571 +               }
12572 +               goto out; /* todo: close all? */
12573 +       }
12574 +
12575 +       err = 0;
12576 +       au_set_fbtop(file, btop);
12577 +       au_set_h_fptr(file, btop, h_file);
12578 +       au_update_figen(file);
12579 +       /* todo: necessary? */
12580 +       /* file->f_ra = h_file->f_ra; */
12581 +
12582 +out:
12583 +       if (h_file_tmp)
12584 +               fput(h_file_tmp);
12585 +       return err;
12586 +}
12587 +
12588 +/* ---------------------------------------------------------------------- */
12589 +
12590 +static int au_reopen_wh(struct file *file, aufs_bindex_t btgt,
12591 +                       struct dentry *hi_wh)
12592 +{
12593 +       int err;
12594 +       aufs_bindex_t btop;
12595 +       struct au_dinfo *dinfo;
12596 +       struct dentry *h_dentry;
12597 +       struct au_hdentry *hdp;
12598 +
12599 +       dinfo = au_di(file->f_path.dentry);
12600 +       AuRwMustWriteLock(&dinfo->di_rwsem);
12601 +
12602 +       btop = dinfo->di_btop;
12603 +       dinfo->di_btop = btgt;
12604 +       hdp = au_hdentry(dinfo, btgt);
12605 +       h_dentry = hdp->hd_dentry;
12606 +       hdp->hd_dentry = hi_wh;
12607 +       err = au_reopen_nondir(file);
12608 +       hdp->hd_dentry = h_dentry;
12609 +       dinfo->di_btop = btop;
12610 +
12611 +       return err;
12612 +}
12613 +
12614 +static int au_ready_to_write_wh(struct file *file, loff_t len,
12615 +                               aufs_bindex_t bcpup, struct au_pin *pin)
12616 +{
12617 +       int err;
12618 +       struct inode *inode, *h_inode;
12619 +       struct dentry *h_dentry, *hi_wh;
12620 +       struct au_cp_generic cpg = {
12621 +               .dentry = file->f_path.dentry,
12622 +               .bdst   = bcpup,
12623 +               .bsrc   = -1,
12624 +               .len    = len,
12625 +               .pin    = pin
12626 +       };
12627 +
12628 +       au_update_dbtop(cpg.dentry);
12629 +       inode = d_inode(cpg.dentry);
12630 +       h_inode = NULL;
12631 +       if (au_dbtop(cpg.dentry) <= bcpup
12632 +           && au_dbbot(cpg.dentry) >= bcpup) {
12633 +               h_dentry = au_h_dptr(cpg.dentry, bcpup);
12634 +               if (h_dentry && d_is_positive(h_dentry))
12635 +                       h_inode = d_inode(h_dentry);
12636 +       }
12637 +       hi_wh = au_hi_wh(inode, bcpup);
12638 +       if (!hi_wh && !h_inode)
12639 +               err = au_sio_cpup_wh(&cpg, file);
12640 +       else
12641 +               /* already copied-up after unlink */
12642 +               err = au_reopen_wh(file, bcpup, hi_wh);
12643 +
12644 +       if (!err
12645 +           && (inode->i_nlink > 1
12646 +               || (inode->i_state & I_LINKABLE))
12647 +           && au_opt_test(au_mntflags(cpg.dentry->d_sb), PLINK))
12648 +               au_plink_append(inode, bcpup, au_h_dptr(cpg.dentry, bcpup));
12649 +
12650 +       return err;
12651 +}
12652 +
12653 +/*
12654 + * prepare the @file for writing.
12655 + */
12656 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin)
12657 +{
12658 +       int err;
12659 +       aufs_bindex_t dbtop;
12660 +       struct dentry *parent;
12661 +       struct inode *inode;
12662 +       struct super_block *sb;
12663 +       struct file *h_file;
12664 +       struct au_cp_generic cpg = {
12665 +               .dentry = file->f_path.dentry,
12666 +               .bdst   = -1,
12667 +               .bsrc   = -1,
12668 +               .len    = len,
12669 +               .pin    = pin,
12670 +               .flags  = AuCpup_DTIME
12671 +       };
12672 +
12673 +       sb = cpg.dentry->d_sb;
12674 +       inode = d_inode(cpg.dentry);
12675 +       cpg.bsrc = au_fbtop(file);
12676 +       err = au_test_ro(sb, cpg.bsrc, inode);
12677 +       if (!err && (au_hf_top(file)->f_mode & FMODE_WRITE)) {
12678 +               err = au_pin(pin, cpg.dentry, cpg.bsrc, AuOpt_UDBA_NONE,
12679 +                            /*flags*/0);
12680 +               goto out;
12681 +       }
12682 +
12683 +       /* need to cpup or reopen */
12684 +       parent = dget_parent(cpg.dentry);
12685 +       di_write_lock_parent(parent);
12686 +       err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12687 +       cpg.bdst = err;
12688 +       if (unlikely(err < 0))
12689 +               goto out_dgrade;
12690 +       err = 0;
12691 +
12692 +       if (!d_unhashed(cpg.dentry) && !au_h_dptr(parent, cpg.bdst)) {
12693 +               err = au_cpup_dirs(cpg.dentry, cpg.bdst);
12694 +               if (unlikely(err))
12695 +                       goto out_dgrade;
12696 +       }
12697 +
12698 +       err = au_pin(pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12699 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12700 +       if (unlikely(err))
12701 +               goto out_dgrade;
12702 +
12703 +       dbtop = au_dbtop(cpg.dentry);
12704 +       if (dbtop <= cpg.bdst)
12705 +               cpg.bsrc = cpg.bdst;
12706 +
12707 +       if (dbtop <= cpg.bdst           /* just reopen */
12708 +           || !d_unhashed(cpg.dentry)  /* copyup and reopen */
12709 +               ) {
12710 +               h_file = au_h_open_pre(cpg.dentry, cpg.bsrc, /*force_wr*/0);
12711 +               if (IS_ERR(h_file))
12712 +                       err = PTR_ERR(h_file);
12713 +               else {
12714 +                       di_downgrade_lock(parent, AuLock_IR);
12715 +                       if (dbtop > cpg.bdst)
12716 +                               err = au_sio_cpup_simple(&cpg);
12717 +                       if (!err)
12718 +                               err = au_reopen_nondir(file);
12719 +                       au_h_open_post(cpg.dentry, cpg.bsrc, h_file);
12720 +               }
12721 +       } else {                        /* copyup as wh and reopen */
12722 +               /*
12723 +                * since writable hfsplus branch is not supported,
12724 +                * h_open_pre/post() are unnecessary.
12725 +                */
12726 +               err = au_ready_to_write_wh(file, len, cpg.bdst, pin);
12727 +               di_downgrade_lock(parent, AuLock_IR);
12728 +       }
12729 +
12730 +       if (!err) {
12731 +               au_pin_set_parent_lflag(pin, /*lflag*/0);
12732 +               goto out_dput; /* success */
12733 +       }
12734 +       au_unpin(pin);
12735 +       goto out_unlock;
12736 +
12737 +out_dgrade:
12738 +       di_downgrade_lock(parent, AuLock_IR);
12739 +out_unlock:
12740 +       di_read_unlock(parent, AuLock_IR);
12741 +out_dput:
12742 +       dput(parent);
12743 +out:
12744 +       return err;
12745 +}
12746 +
12747 +/* ---------------------------------------------------------------------- */
12748 +
12749 +int au_do_flush(struct file *file, fl_owner_t id,
12750 +               int (*flush)(struct file *file, fl_owner_t id))
12751 +{
12752 +       int err;
12753 +       struct super_block *sb;
12754 +       struct inode *inode;
12755 +
12756 +       inode = file_inode(file);
12757 +       sb = inode->i_sb;
12758 +       si_noflush_read_lock(sb);
12759 +       fi_read_lock(file);
12760 +       ii_read_lock_child(inode);
12761 +
12762 +       err = flush(file, id);
12763 +       au_cpup_attr_timesizes(inode);
12764 +
12765 +       ii_read_unlock(inode);
12766 +       fi_read_unlock(file);
12767 +       si_read_unlock(sb);
12768 +       return err;
12769 +}
12770 +
12771 +/* ---------------------------------------------------------------------- */
12772 +
12773 +static int au_file_refresh_by_inode(struct file *file, int *need_reopen)
12774 +{
12775 +       int err;
12776 +       struct au_pin pin;
12777 +       struct au_finfo *finfo;
12778 +       struct dentry *parent, *hi_wh;
12779 +       struct inode *inode;
12780 +       struct super_block *sb;
12781 +       struct au_cp_generic cpg = {
12782 +               .dentry = file->f_path.dentry,
12783 +               .bdst   = -1,
12784 +               .bsrc   = -1,
12785 +               .len    = -1,
12786 +               .pin    = &pin,
12787 +               .flags  = AuCpup_DTIME
12788 +       };
12789 +
12790 +       FiMustWriteLock(file);
12791 +
12792 +       err = 0;
12793 +       finfo = au_fi(file);
12794 +       sb = cpg.dentry->d_sb;
12795 +       inode = d_inode(cpg.dentry);
12796 +       cpg.bdst = au_ibtop(inode);
12797 +       if (cpg.bdst == finfo->fi_btop || IS_ROOT(cpg.dentry))
12798 +               goto out;
12799 +
12800 +       parent = dget_parent(cpg.dentry);
12801 +       if (au_test_ro(sb, cpg.bdst, inode)) {
12802 +               di_read_lock_parent(parent, !AuLock_IR);
12803 +               err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12804 +               cpg.bdst = err;
12805 +               di_read_unlock(parent, !AuLock_IR);
12806 +               if (unlikely(err < 0))
12807 +                       goto out_parent;
12808 +               err = 0;
12809 +       }
12810 +
12811 +       di_read_lock_parent(parent, AuLock_IR);
12812 +       hi_wh = au_hi_wh(inode, cpg.bdst);
12813 +       if (!S_ISDIR(inode->i_mode)
12814 +           && au_opt_test(au_mntflags(sb), PLINK)
12815 +           && au_plink_test(inode)
12816 +           && !d_unhashed(cpg.dentry)
12817 +           && cpg.bdst < au_dbtop(cpg.dentry)) {
12818 +               err = au_test_and_cpup_dirs(cpg.dentry, cpg.bdst);
12819 +               if (unlikely(err))
12820 +                       goto out_unlock;
12821 +
12822 +               /* always superio. */
12823 +               err = au_pin(&pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12824 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12825 +               if (!err) {
12826 +                       err = au_sio_cpup_simple(&cpg);
12827 +                       au_unpin(&pin);
12828 +               }
12829 +       } else if (hi_wh) {
12830 +               /* already copied-up after unlink */
12831 +               err = au_reopen_wh(file, cpg.bdst, hi_wh);
12832 +               *need_reopen = 0;
12833 +       }
12834 +
12835 +out_unlock:
12836 +       di_read_unlock(parent, AuLock_IR);
12837 +out_parent:
12838 +       dput(parent);
12839 +out:
12840 +       return err;
12841 +}
12842 +
12843 +static void au_do_refresh_dir(struct file *file)
12844 +{
12845 +       int execed;
12846 +       aufs_bindex_t bindex, bbot, new_bindex, brid;
12847 +       struct au_hfile *p, tmp, *q;
12848 +       struct au_finfo *finfo;
12849 +       struct super_block *sb;
12850 +       struct au_fidir *fidir;
12851 +
12852 +       FiMustWriteLock(file);
12853 +
12854 +       sb = file->f_path.dentry->d_sb;
12855 +       finfo = au_fi(file);
12856 +       fidir = finfo->fi_hdir;
12857 +       AuDebugOn(!fidir);
12858 +       p = fidir->fd_hfile + finfo->fi_btop;
12859 +       brid = p->hf_br->br_id;
12860 +       bbot = fidir->fd_bbot;
12861 +       for (bindex = finfo->fi_btop; bindex <= bbot; bindex++, p++) {
12862 +               if (!p->hf_file)
12863 +                       continue;
12864 +
12865 +               new_bindex = au_br_index(sb, p->hf_br->br_id);
12866 +               if (new_bindex == bindex)
12867 +                       continue;
12868 +               if (new_bindex < 0) {
12869 +                       au_set_h_fptr(file, bindex, NULL);
12870 +                       continue;
12871 +               }
12872 +
12873 +               /* swap two lower inode, and loop again */
12874 +               q = fidir->fd_hfile + new_bindex;
12875 +               tmp = *q;
12876 +               *q = *p;
12877 +               *p = tmp;
12878 +               if (tmp.hf_file) {
12879 +                       bindex--;
12880 +                       p--;
12881 +               }
12882 +       }
12883 +
12884 +       execed = vfsub_file_execed(file);
12885 +       p = fidir->fd_hfile;
12886 +       if (!au_test_mmapped(file) && !d_unlinked(file->f_path.dentry)) {
12887 +               bbot = au_sbbot(sb);
12888 +               for (finfo->fi_btop = 0; finfo->fi_btop <= bbot;
12889 +                    finfo->fi_btop++, p++)
12890 +                       if (p->hf_file) {
12891 +                               if (file_inode(p->hf_file))
12892 +                                       break;
12893 +                               au_hfput(p, execed);
12894 +                       }
12895 +       } else {
12896 +               bbot = au_br_index(sb, brid);
12897 +               for (finfo->fi_btop = 0; finfo->fi_btop < bbot;
12898 +                    finfo->fi_btop++, p++)
12899 +                       if (p->hf_file)
12900 +                               au_hfput(p, execed);
12901 +               bbot = au_sbbot(sb);
12902 +       }
12903 +
12904 +       p = fidir->fd_hfile + bbot;
12905 +       for (fidir->fd_bbot = bbot; fidir->fd_bbot >= finfo->fi_btop;
12906 +            fidir->fd_bbot--, p--)
12907 +               if (p->hf_file) {
12908 +                       if (file_inode(p->hf_file))
12909 +                               break;
12910 +                       au_hfput(p, execed);
12911 +               }
12912 +       AuDebugOn(fidir->fd_bbot < finfo->fi_btop);
12913 +}
12914 +
12915 +/*
12916 + * after branch manipulating, refresh the file.
12917 + */
12918 +static int refresh_file(struct file *file, int (*reopen)(struct file *file))
12919 +{
12920 +       int err, need_reopen, nbr;
12921 +       aufs_bindex_t bbot, bindex;
12922 +       struct dentry *dentry;
12923 +       struct super_block *sb;
12924 +       struct au_finfo *finfo;
12925 +       struct au_hfile *hfile;
12926 +
12927 +       dentry = file->f_path.dentry;
12928 +       sb = dentry->d_sb;
12929 +       nbr = au_sbbot(sb) + 1;
12930 +       finfo = au_fi(file);
12931 +       if (!finfo->fi_hdir) {
12932 +               hfile = &finfo->fi_htop;
12933 +               AuDebugOn(!hfile->hf_file);
12934 +               bindex = au_br_index(sb, hfile->hf_br->br_id);
12935 +               AuDebugOn(bindex < 0);
12936 +               if (bindex != finfo->fi_btop)
12937 +                       au_set_fbtop(file, bindex);
12938 +       } else {
12939 +               err = au_fidir_realloc(finfo, nbr, /*may_shrink*/0);
12940 +               if (unlikely(err))
12941 +                       goto out;
12942 +               au_do_refresh_dir(file);
12943 +       }
12944 +
12945 +       err = 0;
12946 +       need_reopen = 1;
12947 +       if (!au_test_mmapped(file))
12948 +               err = au_file_refresh_by_inode(file, &need_reopen);
12949 +       if (finfo->fi_hdir)
12950 +               /* harmless if err */
12951 +               au_fidir_realloc(finfo, nbr, /*may_shrink*/1);
12952 +       if (!err && need_reopen && !d_unlinked(dentry))
12953 +               err = reopen(file);
12954 +       if (!err) {
12955 +               au_update_figen(file);
12956 +               goto out; /* success */
12957 +       }
12958 +
12959 +       /* error, close all lower files */
12960 +       if (finfo->fi_hdir) {
12961 +               bbot = au_fbbot_dir(file);
12962 +               for (bindex = au_fbtop(file); bindex <= bbot; bindex++)
12963 +                       au_set_h_fptr(file, bindex, NULL);
12964 +       }
12965 +
12966 +out:
12967 +       return err;
12968 +}
12969 +
12970 +/* common function to regular file and dir */
12971 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
12972 +                         int wlock)
12973 +{
12974 +       int err;
12975 +       unsigned int sigen, figen;
12976 +       aufs_bindex_t btop;
12977 +       unsigned char pseudo_link;
12978 +       struct dentry *dentry;
12979 +       struct inode *inode;
12980 +
12981 +       err = 0;
12982 +       dentry = file->f_path.dentry;
12983 +       inode = d_inode(dentry);
12984 +       sigen = au_sigen(dentry->d_sb);
12985 +       fi_write_lock(file);
12986 +       figen = au_figen(file);
12987 +       di_write_lock_child(dentry);
12988 +       btop = au_dbtop(dentry);
12989 +       pseudo_link = (btop != au_ibtop(inode));
12990 +       if (sigen == figen && !pseudo_link && au_fbtop(file) == btop) {
12991 +               if (!wlock) {
12992 +                       di_downgrade_lock(dentry, AuLock_IR);
12993 +                       fi_downgrade_lock(file);
12994 +               }
12995 +               goto out; /* success */
12996 +       }
12997 +
12998 +       AuDbg("sigen %d, figen %d\n", sigen, figen);
12999 +       if (au_digen_test(dentry, sigen)) {
13000 +               err = au_reval_dpath(dentry, sigen);
13001 +               AuDebugOn(!err && au_digen_test(dentry, sigen));
13002 +       }
13003 +
13004 +       if (!err)
13005 +               err = refresh_file(file, reopen);
13006 +       if (!err) {
13007 +               if (!wlock) {
13008 +                       di_downgrade_lock(dentry, AuLock_IR);
13009 +                       fi_downgrade_lock(file);
13010 +               }
13011 +       } else {
13012 +               di_write_unlock(dentry);
13013 +               fi_write_unlock(file);
13014 +       }
13015 +
13016 +out:
13017 +       return err;
13018 +}
13019 +
13020 +/* ---------------------------------------------------------------------- */
13021 +
13022 +/* cf. aufs_nopage() */
13023 +/* for madvise(2) */
13024 +static int aufs_readpage(struct file *file __maybe_unused, struct page *page)
13025 +{
13026 +       unlock_page(page);
13027 +       return 0;
13028 +}
13029 +
13030 +/* it will never be called, but necessary to support O_DIRECT */
13031 +static ssize_t aufs_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
13032 +{ BUG(); return 0; }
13033 +
13034 +/* they will never be called. */
13035 +#ifdef CONFIG_AUFS_DEBUG
13036 +static int aufs_write_begin(struct file *file, struct address_space *mapping,
13037 +                           loff_t pos, unsigned len, unsigned flags,
13038 +                           struct page **pagep, void **fsdata)
13039 +{ AuUnsupport(); return 0; }
13040 +static int aufs_write_end(struct file *file, struct address_space *mapping,
13041 +                         loff_t pos, unsigned len, unsigned copied,
13042 +                         struct page *page, void *fsdata)
13043 +{ AuUnsupport(); return 0; }
13044 +static int aufs_writepage(struct page *page, struct writeback_control *wbc)
13045 +{ AuUnsupport(); return 0; }
13046 +
13047 +static int aufs_set_page_dirty(struct page *page)
13048 +{ AuUnsupport(); return 0; }
13049 +static void aufs_invalidatepage(struct page *page, unsigned int offset,
13050 +                               unsigned int length)
13051 +{ AuUnsupport(); }
13052 +static int aufs_releasepage(struct page *page, gfp_t gfp)
13053 +{ AuUnsupport(); return 0; }
13054 +#if 0 /* called by memory compaction regardless file */
13055 +static int aufs_migratepage(struct address_space *mapping, struct page *newpage,
13056 +                           struct page *page, enum migrate_mode mode)
13057 +{ AuUnsupport(); return 0; }
13058 +#endif
13059 +static bool aufs_isolate_page(struct page *page, isolate_mode_t mode)
13060 +{ AuUnsupport(); return true; }
13061 +static void aufs_putback_page(struct page *page)
13062 +{ AuUnsupport(); }
13063 +static int aufs_launder_page(struct page *page)
13064 +{ AuUnsupport(); return 0; }
13065 +static int aufs_is_partially_uptodate(struct page *page,
13066 +                                     unsigned long from,
13067 +                                     unsigned long count)
13068 +{ AuUnsupport(); return 0; }
13069 +static void aufs_is_dirty_writeback(struct page *page, bool *dirty,
13070 +                                   bool *writeback)
13071 +{ AuUnsupport(); }
13072 +static int aufs_error_remove_page(struct address_space *mapping,
13073 +                                 struct page *page)
13074 +{ AuUnsupport(); return 0; }
13075 +static int aufs_swap_activate(struct swap_info_struct *sis, struct file *file,
13076 +                             sector_t *span)
13077 +{ AuUnsupport(); return 0; }
13078 +static void aufs_swap_deactivate(struct file *file)
13079 +{ AuUnsupport(); }
13080 +#endif /* CONFIG_AUFS_DEBUG */
13081 +
13082 +const struct address_space_operations aufs_aop = {
13083 +       .readpage               = aufs_readpage,
13084 +       .direct_IO              = aufs_direct_IO,
13085 +#ifdef CONFIG_AUFS_DEBUG
13086 +       .writepage              = aufs_writepage,
13087 +       /* no writepages, because of writepage */
13088 +       .set_page_dirty         = aufs_set_page_dirty,
13089 +       /* no readpages, because of readpage */
13090 +       .write_begin            = aufs_write_begin,
13091 +       .write_end              = aufs_write_end,
13092 +       /* no bmap, no block device */
13093 +       .invalidatepage         = aufs_invalidatepage,
13094 +       .releasepage            = aufs_releasepage,
13095 +       /* is fallback_migrate_page ok? */
13096 +       /* .migratepage         = aufs_migratepage, */
13097 +       .isolate_page           = aufs_isolate_page,
13098 +       .putback_page           = aufs_putback_page,
13099 +       .launder_page           = aufs_launder_page,
13100 +       .is_partially_uptodate  = aufs_is_partially_uptodate,
13101 +       .is_dirty_writeback     = aufs_is_dirty_writeback,
13102 +       .error_remove_page      = aufs_error_remove_page,
13103 +       .swap_activate          = aufs_swap_activate,
13104 +       .swap_deactivate        = aufs_swap_deactivate
13105 +#endif /* CONFIG_AUFS_DEBUG */
13106 +};
13107 diff -urN /usr/share/empty/fs/aufs/file.h linux/fs/aufs/file.h
13108 --- /usr/share/empty/fs/aufs/file.h     1970-01-01 01:00:00.000000000 +0100
13109 +++ linux/fs/aufs/file.h        2016-10-09 16:55:38.889431135 +0200
13110 @@ -0,0 +1,294 @@
13111 +/*
13112 + * Copyright (C) 2005-2016 Junjiro R. Okajima
13113 + *
13114 + * This program, aufs is free software; you can redistribute it and/or modify
13115 + * it under the terms of the GNU General Public License as published by
13116 + * the Free Software Foundation; either version 2 of the License, or
13117 + * (at your option) any later version.
13118 + *
13119 + * This program is distributed in the hope that it will be useful,
13120 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13121 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13122 + * GNU General Public License for more details.
13123 + *
13124 + * You should have received a copy of the GNU General Public License
13125 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13126 + */
13127 +
13128 +/*
13129 + * file operations
13130 + */
13131 +
13132 +#ifndef __AUFS_FILE_H__
13133 +#define __AUFS_FILE_H__
13134 +
13135 +#ifdef __KERNEL__
13136 +
13137 +#include <linux/file.h>
13138 +#include <linux/fs.h>
13139 +#include <linux/poll.h>
13140 +#include "rwsem.h"
13141 +
13142 +struct au_branch;
13143 +struct au_hfile {
13144 +       struct file             *hf_file;
13145 +       struct au_branch        *hf_br;
13146 +};
13147 +
13148 +struct au_vdir;
13149 +struct au_fidir {
13150 +       aufs_bindex_t           fd_bbot;
13151 +       aufs_bindex_t           fd_nent;
13152 +       struct au_vdir          *fd_vdir_cache;
13153 +       struct au_hfile         fd_hfile[];
13154 +};
13155 +
13156 +static inline int au_fidir_sz(int nent)
13157 +{
13158 +       AuDebugOn(nent < 0);
13159 +       return sizeof(struct au_fidir) + sizeof(struct au_hfile) * nent;
13160 +}
13161 +
13162 +struct au_finfo {
13163 +       atomic_t                fi_generation;
13164 +
13165 +       struct au_rwsem         fi_rwsem;
13166 +       aufs_bindex_t           fi_btop;
13167 +
13168 +       /* do not union them */
13169 +       struct {                                /* for non-dir */
13170 +               struct au_hfile                 fi_htop;
13171 +               atomic_t                        fi_mmapped;
13172 +       };
13173 +       struct au_fidir         *fi_hdir;       /* for dir only */
13174 +
13175 +       struct hlist_node       fi_hlist;
13176 +       union {
13177 +               struct file             *fi_file;       /* very ugly */
13178 +               struct llist_node       fi_lnode;       /* delayed free */
13179 +       };
13180 +} ____cacheline_aligned_in_smp;
13181 +
13182 +/* ---------------------------------------------------------------------- */
13183 +
13184 +/* file.c */
13185 +extern const struct address_space_operations aufs_aop;
13186 +unsigned int au_file_roflags(unsigned int flags);
13187 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
13188 +                      struct file *file, int force_wr);
13189 +struct au_do_open_args {
13190 +       int             no_lock;
13191 +       int             (*open)(struct file *file, int flags,
13192 +                               struct file *h_file);
13193 +       struct au_fidir *fidir;
13194 +       struct file     *h_file;
13195 +};
13196 +int au_do_open(struct file *file, struct au_do_open_args *args);
13197 +int au_reopen_nondir(struct file *file);
13198 +struct au_pin;
13199 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin);
13200 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
13201 +                         int wlock);
13202 +int au_do_flush(struct file *file, fl_owner_t id,
13203 +               int (*flush)(struct file *file, fl_owner_t id));
13204 +
13205 +/* poll.c */
13206 +#ifdef CONFIG_AUFS_POLL
13207 +unsigned int aufs_poll(struct file *file, poll_table *wait);
13208 +#endif
13209 +
13210 +#ifdef CONFIG_AUFS_BR_HFSPLUS
13211 +/* hfsplus.c */
13212 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
13213 +                          int force_wr);
13214 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
13215 +                   struct file *h_file);
13216 +#else
13217 +AuStub(struct file *, au_h_open_pre, return NULL, struct dentry *dentry,
13218 +       aufs_bindex_t bindex, int force_wr)
13219 +AuStubVoid(au_h_open_post, struct dentry *dentry, aufs_bindex_t bindex,
13220 +          struct file *h_file);
13221 +#endif
13222 +
13223 +/* f_op.c */
13224 +extern const struct file_operations aufs_file_fop;
13225 +int au_do_open_nondir(struct file *file, int flags, struct file *h_file);
13226 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file);
13227 +struct file *au_read_pre(struct file *file, int keep_fi);
13228 +
13229 +/* finfo.c */
13230 +void au_hfput(struct au_hfile *hf, int execed);
13231 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex,
13232 +                  struct file *h_file);
13233 +
13234 +void au_update_figen(struct file *file);
13235 +struct au_fidir *au_fidir_alloc(struct super_block *sb);
13236 +int au_fidir_realloc(struct au_finfo *finfo, int nbr, int may_shrink);
13237 +
13238 +void au_fi_init_once(void *_fi);
13239 +void au_finfo_fin(struct file *file, int atonce);
13240 +int au_finfo_init(struct file *file, struct au_fidir *fidir);
13241 +
13242 +/* ioctl.c */
13243 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg);
13244 +#ifdef CONFIG_COMPAT
13245 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
13246 +                          unsigned long arg);
13247 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
13248 +                             unsigned long arg);
13249 +#endif
13250 +
13251 +/* ---------------------------------------------------------------------- */
13252 +
13253 +static inline struct au_finfo *au_fi(struct file *file)
13254 +{
13255 +       return file->private_data;
13256 +}
13257 +
13258 +/* ---------------------------------------------------------------------- */
13259 +
13260 +/*
13261 + * fi_read_lock, fi_write_lock,
13262 + * fi_read_unlock, fi_write_unlock, fi_downgrade_lock
13263 + */
13264 +AuSimpleRwsemFuncs(fi, struct file *f, &au_fi(f)->fi_rwsem);
13265 +
13266 +#define FiMustNoWaiters(f)     AuRwMustNoWaiters(&au_fi(f)->fi_rwsem)
13267 +#define FiMustAnyLock(f)       AuRwMustAnyLock(&au_fi(f)->fi_rwsem)
13268 +#define FiMustWriteLock(f)     AuRwMustWriteLock(&au_fi(f)->fi_rwsem)
13269 +
13270 +/* ---------------------------------------------------------------------- */
13271 +
13272 +/* todo: hard/soft set? */
13273 +static inline aufs_bindex_t au_fbtop(struct file *file)
13274 +{
13275 +       FiMustAnyLock(file);
13276 +       return au_fi(file)->fi_btop;
13277 +}
13278 +
13279 +static inline aufs_bindex_t au_fbbot_dir(struct file *file)
13280 +{
13281 +       FiMustAnyLock(file);
13282 +       AuDebugOn(!au_fi(file)->fi_hdir);
13283 +       return au_fi(file)->fi_hdir->fd_bbot;
13284 +}
13285 +
13286 +static inline struct au_vdir *au_fvdir_cache(struct file *file)
13287 +{
13288 +       FiMustAnyLock(file);
13289 +       AuDebugOn(!au_fi(file)->fi_hdir);
13290 +       return au_fi(file)->fi_hdir->fd_vdir_cache;
13291 +}
13292 +
13293 +static inline void au_set_fbtop(struct file *file, aufs_bindex_t bindex)
13294 +{
13295 +       FiMustWriteLock(file);
13296 +       au_fi(file)->fi_btop = bindex;
13297 +}
13298 +
13299 +static inline void au_set_fbbot_dir(struct file *file, aufs_bindex_t bindex)
13300 +{
13301 +       FiMustWriteLock(file);
13302 +       AuDebugOn(!au_fi(file)->fi_hdir);
13303 +       au_fi(file)->fi_hdir->fd_bbot = bindex;
13304 +}
13305 +
13306 +static inline void au_set_fvdir_cache(struct file *file,
13307 +                                     struct au_vdir *vdir_cache)
13308 +{
13309 +       FiMustWriteLock(file);
13310 +       AuDebugOn(!au_fi(file)->fi_hdir);
13311 +       au_fi(file)->fi_hdir->fd_vdir_cache = vdir_cache;
13312 +}
13313 +
13314 +static inline struct file *au_hf_top(struct file *file)
13315 +{
13316 +       FiMustAnyLock(file);
13317 +       AuDebugOn(au_fi(file)->fi_hdir);
13318 +       return au_fi(file)->fi_htop.hf_file;
13319 +}
13320 +
13321 +static inline struct file *au_hf_dir(struct file *file, aufs_bindex_t bindex)
13322 +{
13323 +       FiMustAnyLock(file);
13324 +       AuDebugOn(!au_fi(file)->fi_hdir);
13325 +       return au_fi(file)->fi_hdir->fd_hfile[0 + bindex].hf_file;
13326 +}
13327 +
13328 +/* todo: memory barrier? */
13329 +static inline unsigned int au_figen(struct file *f)
13330 +{
13331 +       return atomic_read(&au_fi(f)->fi_generation);
13332 +}
13333 +
13334 +static inline void au_set_mmapped(struct file *f)
13335 +{
13336 +       if (atomic_inc_return(&au_fi(f)->fi_mmapped))
13337 +               return;
13338 +       pr_warn("fi_mmapped wrapped around\n");
13339 +       while (!atomic_inc_return(&au_fi(f)->fi_mmapped))
13340 +               ;
13341 +}
13342 +
13343 +static inline void au_unset_mmapped(struct file *f)
13344 +{
13345 +       atomic_dec(&au_fi(f)->fi_mmapped);
13346 +}
13347 +
13348 +static inline int au_test_mmapped(struct file *f)
13349 +{
13350 +       return atomic_read(&au_fi(f)->fi_mmapped);
13351 +}
13352 +
13353 +/* customize vma->vm_file */
13354 +
13355 +static inline void au_do_vm_file_reset(struct vm_area_struct *vma,
13356 +                                      struct file *file)
13357 +{
13358 +       struct file *f;
13359 +
13360 +       f = vma->vm_file;
13361 +       get_file(file);
13362 +       vma->vm_file = file;
13363 +       fput(f);
13364 +}
13365 +
13366 +#ifdef CONFIG_MMU
13367 +#define AuDbgVmRegion(file, vma) do {} while (0)
13368 +
13369 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
13370 +                                   struct file *file)
13371 +{
13372 +       au_do_vm_file_reset(vma, file);
13373 +}
13374 +#else
13375 +#define AuDbgVmRegion(file, vma) \
13376 +       AuDebugOn((vma)->vm_region && (vma)->vm_region->vm_file != (file))
13377 +
13378 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
13379 +                                   struct file *file)
13380 +{
13381 +       struct file *f;
13382 +
13383 +       au_do_vm_file_reset(vma, file);
13384 +       f = vma->vm_region->vm_file;
13385 +       get_file(file);
13386 +       vma->vm_region->vm_file = file;
13387 +       fput(f);
13388 +}
13389 +#endif /* CONFIG_MMU */
13390 +
13391 +/* handle vma->vm_prfile */
13392 +static inline void au_vm_prfile_set(struct vm_area_struct *vma,
13393 +                                   struct file *file)
13394 +{
13395 +       get_file(file);
13396 +       vma->vm_prfile = file;
13397 +#ifndef CONFIG_MMU
13398 +       get_file(file);
13399 +       vma->vm_region->vm_prfile = file;
13400 +#endif
13401 +}
13402 +
13403 +#endif /* __KERNEL__ */
13404 +#endif /* __AUFS_FILE_H__ */
13405 diff -urN /usr/share/empty/fs/aufs/finfo.c linux/fs/aufs/finfo.c
13406 --- /usr/share/empty/fs/aufs/finfo.c    1970-01-01 01:00:00.000000000 +0100
13407 +++ linux/fs/aufs/finfo.c       2016-10-09 16:55:38.889431135 +0200
13408 @@ -0,0 +1,151 @@
13409 +/*
13410 + * Copyright (C) 2005-2016 Junjiro R. Okajima
13411 + *
13412 + * This program, aufs is free software; you can redistribute it and/or modify
13413 + * it under the terms of the GNU General Public License as published by
13414 + * the Free Software Foundation; either version 2 of the License, or
13415 + * (at your option) any later version.
13416 + *
13417 + * This program is distributed in the hope that it will be useful,
13418 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13419 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13420 + * GNU General Public License for more details.
13421 + *
13422 + * You should have received a copy of the GNU General Public License
13423 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13424 + */
13425 +
13426 +/*
13427 + * file private data
13428 + */
13429 +
13430 +#include "aufs.h"
13431 +
13432 +void au_hfput(struct au_hfile *hf, int execed)
13433 +{
13434 +       if (execed)
13435 +               allow_write_access(hf->hf_file);
13436 +       fput(hf->hf_file);
13437 +       hf->hf_file = NULL;
13438 +       au_br_put(hf->hf_br);
13439 +       hf->hf_br = NULL;
13440 +}
13441 +
13442 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex, struct file *val)
13443 +{
13444 +       struct au_finfo *finfo = au_fi(file);
13445 +       struct au_hfile *hf;
13446 +       struct au_fidir *fidir;
13447 +
13448 +       fidir = finfo->fi_hdir;
13449 +       if (!fidir) {
13450 +               AuDebugOn(finfo->fi_btop != bindex);
13451 +               hf = &finfo->fi_htop;
13452 +       } else
13453 +               hf = fidir->fd_hfile + bindex;
13454 +
13455 +       if (hf && hf->hf_file)
13456 +               au_hfput(hf, vfsub_file_execed(file));
13457 +       if (val) {
13458 +               FiMustWriteLock(file);
13459 +               AuDebugOn(IS_ERR_OR_NULL(file->f_path.dentry));
13460 +               hf->hf_file = val;
13461 +               hf->hf_br = au_sbr(file->f_path.dentry->d_sb, bindex);
13462 +       }
13463 +}
13464 +
13465 +void au_update_figen(struct file *file)
13466 +{
13467 +       atomic_set(&au_fi(file)->fi_generation, au_digen(file->f_path.dentry));
13468 +       /* smp_mb(); */ /* atomic_set */
13469 +}
13470 +
13471 +/* ---------------------------------------------------------------------- */
13472 +
13473 +struct au_fidir *au_fidir_alloc(struct super_block *sb)
13474 +{
13475 +       struct au_fidir *fidir;
13476 +       int nbr;
13477 +
13478 +       nbr = au_sbbot(sb) + 1;
13479 +       if (nbr < 2)
13480 +               nbr = 2; /* initial allocate for 2 branches */
13481 +       fidir = kzalloc(au_fidir_sz(nbr), GFP_NOFS);
13482 +       if (fidir) {
13483 +               fidir->fd_bbot = -1;
13484 +               fidir->fd_nent = nbr;
13485 +       }
13486 +
13487 +       return fidir;
13488 +}
13489 +
13490 +int au_fidir_realloc(struct au_finfo *finfo, int nbr, int may_shrink)
13491 +{
13492 +       int err;
13493 +       struct au_fidir *fidir, *p;
13494 +
13495 +       AuRwMustWriteLock(&finfo->fi_rwsem);
13496 +       fidir = finfo->fi_hdir;
13497 +       AuDebugOn(!fidir);
13498 +
13499 +       err = -ENOMEM;
13500 +       p = au_kzrealloc(fidir, au_fidir_sz(fidir->fd_nent), au_fidir_sz(nbr),
13501 +                        GFP_NOFS, may_shrink);
13502 +       if (p) {
13503 +               p->fd_nent = nbr;
13504 +               finfo->fi_hdir = p;
13505 +               err = 0;
13506 +       }
13507 +
13508 +       return err;
13509 +}
13510 +
13511 +/* ---------------------------------------------------------------------- */
13512 +
13513 +void au_finfo_fin(struct file *file, int atonce)
13514 +{
13515 +       struct au_finfo *finfo;
13516 +
13517 +       au_nfiles_dec(file->f_path.dentry->d_sb);
13518 +
13519 +       finfo = au_fi(file);
13520 +       AuDebugOn(finfo->fi_hdir);
13521 +       AuRwDestroy(&finfo->fi_rwsem);
13522 +       if (!atonce)
13523 +               au_cache_dfree_finfo(finfo);
13524 +       else
13525 +               au_cache_free_finfo(finfo);
13526 +}
13527 +
13528 +void au_fi_init_once(void *_finfo)
13529 +{
13530 +       struct au_finfo *finfo = _finfo;
13531 +
13532 +       au_rw_init(&finfo->fi_rwsem);
13533 +}
13534 +
13535 +int au_finfo_init(struct file *file, struct au_fidir *fidir)
13536 +{
13537 +       int err;
13538 +       struct au_finfo *finfo;
13539 +       struct dentry *dentry;
13540 +
13541 +       err = -ENOMEM;
13542 +       dentry = file->f_path.dentry;
13543 +       finfo = au_cache_alloc_finfo();
13544 +       if (unlikely(!finfo))
13545 +               goto out;
13546 +
13547 +       err = 0;
13548 +       au_nfiles_inc(dentry->d_sb);
13549 +       au_rw_write_lock(&finfo->fi_rwsem);
13550 +       finfo->fi_btop = -1;
13551 +       finfo->fi_hdir = fidir;
13552 +       atomic_set(&finfo->fi_generation, au_digen(dentry));
13553 +       /* smp_mb(); */ /* atomic_set */
13554 +
13555 +       file->private_data = finfo;
13556 +
13557 +out:
13558 +       return err;
13559 +}
13560 diff -urN /usr/share/empty/fs/aufs/f_op.c linux/fs/aufs/f_op.c
13561 --- /usr/share/empty/fs/aufs/f_op.c     1970-01-01 01:00:00.000000000 +0100
13562 +++ linux/fs/aufs/f_op.c        2016-10-09 16:55:36.489368218 +0200
13563 @@ -0,0 +1,772 @@
13564 +/*
13565 + * Copyright (C) 2005-2016 Junjiro R. Okajima
13566 + *
13567 + * This program, aufs is free software; you can redistribute it and/or modify
13568 + * it under the terms of the GNU General Public License as published by
13569 + * the Free Software Foundation; either version 2 of the License, or
13570 + * (at your option) any later version.
13571 + *
13572 + * This program is distributed in the hope that it will be useful,
13573 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13574 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13575 + * GNU General Public License for more details.
13576 + *
13577 + * You should have received a copy of the GNU General Public License
13578 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13579 + */
13580 +
13581 +/*
13582 + * file and vm operations
13583 + */
13584 +
13585 +#include <linux/aio.h>
13586 +#include <linux/fs_stack.h>
13587 +#include <linux/mman.h>
13588 +#include <linux/security.h>
13589 +#include "aufs.h"
13590 +
13591 +int au_do_open_nondir(struct file *file, int flags, struct file *h_file)
13592 +{
13593 +       int err;
13594 +       aufs_bindex_t bindex;
13595 +       struct dentry *dentry, *h_dentry;
13596 +       struct au_finfo *finfo;
13597 +       struct inode *h_inode;
13598 +
13599 +       FiMustWriteLock(file);
13600 +
13601 +       err = 0;
13602 +       dentry = file->f_path.dentry;
13603 +       AuDebugOn(IS_ERR_OR_NULL(dentry));
13604 +       finfo = au_fi(file);
13605 +       memset(&finfo->fi_htop, 0, sizeof(finfo->fi_htop));
13606 +       atomic_set(&finfo->fi_mmapped, 0);
13607 +       bindex = au_dbtop(dentry);
13608 +       if (!h_file) {
13609 +               h_dentry = au_h_dptr(dentry, bindex);
13610 +               err = vfsub_test_mntns(file->f_path.mnt, h_dentry->d_sb);
13611 +               if (unlikely(err))
13612 +                       goto out;
13613 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
13614 +       } else {
13615 +               h_dentry = h_file->f_path.dentry;
13616 +               err = vfsub_test_mntns(file->f_path.mnt, h_dentry->d_sb);
13617 +               if (unlikely(err))
13618 +                       goto out;
13619 +               get_file(h_file);
13620 +       }
13621 +       if (IS_ERR(h_file))
13622 +               err = PTR_ERR(h_file);
13623 +       else {
13624 +               if ((flags & __O_TMPFILE)
13625 +                   && !(flags & O_EXCL)) {
13626 +                       h_inode = file_inode(h_file);
13627 +                       spin_lock(&h_inode->i_lock);
13628 +                       h_inode->i_state |= I_LINKABLE;
13629 +                       spin_unlock(&h_inode->i_lock);
13630 +               }
13631 +               au_set_fbtop(file, bindex);
13632 +               au_set_h_fptr(file, bindex, h_file);
13633 +               au_update_figen(file);
13634 +               /* todo: necessary? */
13635 +               /* file->f_ra = h_file->f_ra; */
13636 +       }
13637 +
13638 +out:
13639 +       return err;
13640 +}
13641 +
13642 +static int aufs_open_nondir(struct inode *inode __maybe_unused,
13643 +                           struct file *file)
13644 +{
13645 +       int err;
13646 +       struct super_block *sb;
13647 +       struct au_do_open_args args = {
13648 +               .open   = au_do_open_nondir
13649 +       };
13650 +
13651 +       AuDbg("%pD, f_flags 0x%x, f_mode 0x%x\n",
13652 +             file, vfsub_file_flags(file), file->f_mode);
13653 +
13654 +       sb = file->f_path.dentry->d_sb;
13655 +       si_read_lock(sb, AuLock_FLUSH);
13656 +       err = au_do_open(file, &args);
13657 +       si_read_unlock(sb);
13658 +       return err;
13659 +}
13660 +
13661 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file)
13662 +{
13663 +       struct au_finfo *finfo;
13664 +       aufs_bindex_t bindex;
13665 +       int delayed;
13666 +
13667 +       finfo = au_fi(file);
13668 +       au_sphl_del(&finfo->fi_hlist,
13669 +                   &au_sbi(file->f_path.dentry->d_sb)->si_files);
13670 +       bindex = finfo->fi_btop;
13671 +       if (bindex >= 0)
13672 +               au_set_h_fptr(file, bindex, NULL);
13673 +
13674 +       delayed = (current->flags & PF_KTHREAD) || in_interrupt();
13675 +       au_finfo_fin(file, delayed);
13676 +       return 0;
13677 +}
13678 +
13679 +/* ---------------------------------------------------------------------- */
13680 +
13681 +static int au_do_flush_nondir(struct file *file, fl_owner_t id)
13682 +{
13683 +       int err;
13684 +       struct file *h_file;
13685 +
13686 +       err = 0;
13687 +       h_file = au_hf_top(file);
13688 +       if (h_file)
13689 +               err = vfsub_flush(h_file, id);
13690 +       return err;
13691 +}
13692 +
13693 +static int aufs_flush_nondir(struct file *file, fl_owner_t id)
13694 +{
13695 +       return au_do_flush(file, id, au_do_flush_nondir);
13696 +}
13697 +
13698 +/* ---------------------------------------------------------------------- */
13699 +/*
13700 + * read and write functions acquire [fdi]_rwsem once, but release before
13701 + * mmap_sem. This is because to stop a race condition between mmap(2).
13702 + * Releasing these aufs-rwsem should be safe, no branch-mamagement (by keeping
13703 + * si_rwsem), no harmful copy-up should happen. Actually copy-up may happen in
13704 + * read functions after [fdi]_rwsem are released, but it should be harmless.
13705 + */
13706 +
13707 +/* Callers should call au_read_post() or fput() in the end */
13708 +struct file *au_read_pre(struct file *file, int keep_fi)
13709 +{
13710 +       struct file *h_file;
13711 +       int err;
13712 +
13713 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13714 +       if (!err) {
13715 +               di_read_unlock(file->f_path.dentry, AuLock_IR);
13716 +               h_file = au_hf_top(file);
13717 +               get_file(h_file);
13718 +               if (!keep_fi)
13719 +                       fi_read_unlock(file);
13720 +       } else
13721 +               h_file = ERR_PTR(err);
13722 +
13723 +       return h_file;
13724 +}
13725 +
13726 +static void au_read_post(struct inode *inode, struct file *h_file)
13727 +{
13728 +       /* update without lock, I don't think it a problem */
13729 +       fsstack_copy_attr_atime(inode, file_inode(h_file));
13730 +       fput(h_file);
13731 +}
13732 +
13733 +struct au_write_pre {
13734 +       blkcnt_t blks;
13735 +       aufs_bindex_t btop;
13736 +};
13737 +
13738 +/*
13739 + * return with iinfo is write-locked
13740 + * callers should call au_write_post() or iinfo_write_unlock() + fput() in the
13741 + * end
13742 + */
13743 +static struct file *au_write_pre(struct file *file, int do_ready,
13744 +                                struct au_write_pre *wpre)
13745 +{
13746 +       struct file *h_file;
13747 +       struct dentry *dentry;
13748 +       int err;
13749 +       struct au_pin pin;
13750 +
13751 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13752 +       h_file = ERR_PTR(err);
13753 +       if (unlikely(err))
13754 +               goto out;
13755 +
13756 +       dentry = file->f_path.dentry;
13757 +       if (do_ready) {
13758 +               err = au_ready_to_write(file, -1, &pin);
13759 +               if (unlikely(err)) {
13760 +                       h_file = ERR_PTR(err);
13761 +                       di_write_unlock(dentry);
13762 +                       goto out_fi;
13763 +               }
13764 +       }
13765 +
13766 +       di_downgrade_lock(dentry, /*flags*/0);
13767 +       if (wpre)
13768 +               wpre->btop = au_fbtop(file);
13769 +       h_file = au_hf_top(file);
13770 +       get_file(h_file);
13771 +       if (wpre)
13772 +               wpre->blks = file_inode(h_file)->i_blocks;
13773 +       if (do_ready)
13774 +               au_unpin(&pin);
13775 +       di_read_unlock(dentry, /*flags*/0);
13776 +
13777 +out_fi:
13778 +       fi_write_unlock(file);
13779 +out:
13780 +       return h_file;
13781 +}
13782 +
13783 +static void au_write_post(struct inode *inode, struct file *h_file,
13784 +                         struct au_write_pre *wpre, ssize_t written)
13785 +{
13786 +       struct inode *h_inode;
13787 +
13788 +       au_cpup_attr_timesizes(inode);
13789 +       AuDebugOn(au_ibtop(inode) != wpre->btop);
13790 +       h_inode = file_inode(h_file);
13791 +       inode->i_mode = h_inode->i_mode;
13792 +       ii_write_unlock(inode);
13793 +       fput(h_file);
13794 +
13795 +       /* AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks); */
13796 +       if (written > 0)
13797 +               au_fhsm_wrote(inode->i_sb, wpre->btop,
13798 +                             /*force*/h_inode->i_blocks > wpre->blks);
13799 +}
13800 +
13801 +static ssize_t aufs_read(struct file *file, char __user *buf, size_t count,
13802 +                        loff_t *ppos)
13803 +{
13804 +       ssize_t err;
13805 +       struct inode *inode;
13806 +       struct file *h_file;
13807 +       struct super_block *sb;
13808 +
13809 +       inode = file_inode(file);
13810 +       sb = inode->i_sb;
13811 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13812 +
13813 +       h_file = au_read_pre(file, /*keep_fi*/0);
13814 +       err = PTR_ERR(h_file);
13815 +       if (IS_ERR(h_file))
13816 +               goto out;
13817 +
13818 +       /* filedata may be obsoleted by concurrent copyup, but no problem */
13819 +       err = vfsub_read_u(h_file, buf, count, ppos);
13820 +       /* todo: necessary? */
13821 +       /* file->f_ra = h_file->f_ra; */
13822 +       au_read_post(inode, h_file);
13823 +
13824 +out:
13825 +       si_read_unlock(sb);
13826 +       return err;
13827 +}
13828 +
13829 +/*
13830 + * todo: very ugly
13831 + * it locks both of i_mutex and si_rwsem for read in safe.
13832 + * if the plink maintenance mode continues forever (that is the problem),
13833 + * may loop forever.
13834 + */
13835 +static void au_mtx_and_read_lock(struct inode *inode)
13836 +{
13837 +       int err;
13838 +       struct super_block *sb = inode->i_sb;
13839 +
13840 +       while (1) {
13841 +               inode_lock(inode);
13842 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
13843 +               if (!err)
13844 +                       break;
13845 +               inode_unlock(inode);
13846 +               si_read_lock(sb, AuLock_NOPLMW);
13847 +               si_read_unlock(sb);
13848 +       }
13849 +}
13850 +
13851 +static ssize_t aufs_write(struct file *file, const char __user *ubuf,
13852 +                         size_t count, loff_t *ppos)
13853 +{
13854 +       ssize_t err;
13855 +       struct au_write_pre wpre;
13856 +       struct inode *inode;
13857 +       struct file *h_file;
13858 +       char __user *buf = (char __user *)ubuf;
13859 +
13860 +       inode = file_inode(file);
13861 +       au_mtx_and_read_lock(inode);
13862 +
13863 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
13864 +       err = PTR_ERR(h_file);
13865 +       if (IS_ERR(h_file))
13866 +               goto out;
13867 +
13868 +       err = vfsub_write_u(h_file, buf, count, ppos);
13869 +       au_write_post(inode, h_file, &wpre, err);
13870 +
13871 +out:
13872 +       si_read_unlock(inode->i_sb);
13873 +       inode_unlock(inode);
13874 +       return err;
13875 +}
13876 +
13877 +static ssize_t au_do_iter(struct file *h_file, int rw, struct kiocb *kio,
13878 +                         struct iov_iter *iov_iter)
13879 +{
13880 +       ssize_t err;
13881 +       struct file *file;
13882 +       ssize_t (*iter)(struct kiocb *, struct iov_iter *);
13883 +
13884 +       err = security_file_permission(h_file, rw);
13885 +       if (unlikely(err))
13886 +               goto out;
13887 +
13888 +       err = -ENOSYS;
13889 +       iter = NULL;
13890 +       if (rw == MAY_READ)
13891 +               iter = h_file->f_op->read_iter;
13892 +       else if (rw == MAY_WRITE)
13893 +               iter = h_file->f_op->write_iter;
13894 +
13895 +       file = kio->ki_filp;
13896 +       kio->ki_filp = h_file;
13897 +       if (iter) {
13898 +               lockdep_off();
13899 +               err = iter(kio, iov_iter);
13900 +               lockdep_on();
13901 +       } else
13902 +               /* currently there is no such fs */
13903 +               WARN_ON_ONCE(1);
13904 +       kio->ki_filp = file;
13905 +
13906 +out:
13907 +       return err;
13908 +}
13909 +
13910 +static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
13911 +{
13912 +       ssize_t err;
13913 +       struct file *file, *h_file;
13914 +       struct inode *inode;
13915 +       struct super_block *sb;
13916 +
13917 +       file = kio->ki_filp;
13918 +       inode = file_inode(file);
13919 +       sb = inode->i_sb;
13920 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13921 +
13922 +       h_file = au_read_pre(file, /*keep_fi*/1);
13923 +       err = PTR_ERR(h_file);
13924 +       if (IS_ERR(h_file))
13925 +               goto out;
13926 +
13927 +       if (au_test_loopback_kthread()) {
13928 +               au_warn_loopback(h_file->f_path.dentry->d_sb);
13929 +               if (file->f_mapping != h_file->f_mapping) {
13930 +                       file->f_mapping = h_file->f_mapping;
13931 +                       smp_mb(); /* unnecessary? */
13932 +               }
13933 +       }
13934 +       fi_read_unlock(file);
13935 +
13936 +       err = au_do_iter(h_file, MAY_READ, kio, iov_iter);
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 +static ssize_t aufs_write_iter(struct kiocb *kio, struct iov_iter *iov_iter)
13947 +{
13948 +       ssize_t err;
13949 +       struct au_write_pre wpre;
13950 +       struct inode *inode;
13951 +       struct file *file, *h_file;
13952 +
13953 +       file = kio->ki_filp;
13954 +       inode = file_inode(file);
13955 +       au_mtx_and_read_lock(inode);
13956 +
13957 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
13958 +       err = PTR_ERR(h_file);
13959 +       if (IS_ERR(h_file))
13960 +               goto out;
13961 +
13962 +       err = au_do_iter(h_file, MAY_WRITE, kio, iov_iter);
13963 +       au_write_post(inode, h_file, &wpre, err);
13964 +
13965 +out:
13966 +       si_read_unlock(inode->i_sb);
13967 +       inode_unlock(inode);
13968 +       return err;
13969 +}
13970 +
13971 +static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
13972 +                               struct pipe_inode_info *pipe, size_t len,
13973 +                               unsigned int flags)
13974 +{
13975 +       ssize_t err;
13976 +       struct file *h_file;
13977 +       struct inode *inode;
13978 +       struct super_block *sb;
13979 +
13980 +       inode = file_inode(file);
13981 +       sb = inode->i_sb;
13982 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13983 +
13984 +       h_file = au_read_pre(file, /*keep_fi*/0);
13985 +       err = PTR_ERR(h_file);
13986 +       if (IS_ERR(h_file))
13987 +               goto out;
13988 +
13989 +       err = vfsub_splice_to(h_file, ppos, pipe, len, flags);
13990 +       /* todo: necessasry? */
13991 +       /* file->f_ra = h_file->f_ra; */
13992 +       au_read_post(inode, h_file);
13993 +
13994 +out:
13995 +       si_read_unlock(sb);
13996 +       return err;
13997 +}
13998 +
13999 +static ssize_t
14000 +aufs_splice_write(struct pipe_inode_info *pipe, struct file *file, loff_t *ppos,
14001 +                 size_t len, unsigned int flags)
14002 +{
14003 +       ssize_t err;
14004 +       struct au_write_pre wpre;
14005 +       struct inode *inode;
14006 +       struct file *h_file;
14007 +
14008 +       inode = file_inode(file);
14009 +       au_mtx_and_read_lock(inode);
14010 +
14011 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14012 +       err = PTR_ERR(h_file);
14013 +       if (IS_ERR(h_file))
14014 +               goto out;
14015 +
14016 +       err = vfsub_splice_from(pipe, h_file, ppos, len, flags);
14017 +       au_write_post(inode, h_file, &wpre, err);
14018 +
14019 +out:
14020 +       si_read_unlock(inode->i_sb);
14021 +       inode_unlock(inode);
14022 +       return err;
14023 +}
14024 +
14025 +static long aufs_fallocate(struct file *file, int mode, loff_t offset,
14026 +                          loff_t len)
14027 +{
14028 +       long err;
14029 +       struct au_write_pre wpre;
14030 +       struct inode *inode;
14031 +       struct file *h_file;
14032 +
14033 +       inode = file_inode(file);
14034 +       au_mtx_and_read_lock(inode);
14035 +
14036 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14037 +       err = PTR_ERR(h_file);
14038 +       if (IS_ERR(h_file))
14039 +               goto out;
14040 +
14041 +       lockdep_off();
14042 +       err = vfs_fallocate(h_file, mode, offset, len);
14043 +       lockdep_on();
14044 +       au_write_post(inode, h_file, &wpre, /*written*/1);
14045 +
14046 +out:
14047 +       si_read_unlock(inode->i_sb);
14048 +       inode_unlock(inode);
14049 +       return err;
14050 +}
14051 +
14052 +/* ---------------------------------------------------------------------- */
14053 +
14054 +/*
14055 + * The locking order around current->mmap_sem.
14056 + * - in most and regular cases
14057 + *   file I/O syscall -- aufs_read() or something
14058 + *     -- si_rwsem for read -- mmap_sem
14059 + *     (Note that [fdi]i_rwsem are released before mmap_sem).
14060 + * - in mmap case
14061 + *   mmap(2) -- mmap_sem -- aufs_mmap() -- si_rwsem for read -- [fdi]i_rwsem
14062 + * This AB-BA order is definitly bad, but is not a problem since "si_rwsem for
14063 + * read" allows muliple processes to acquire it and [fdi]i_rwsem are not held in
14064 + * file I/O. Aufs needs to stop lockdep in aufs_mmap() though.
14065 + * It means that when aufs acquires si_rwsem for write, the process should never
14066 + * acquire mmap_sem.
14067 + *
14068 + * Actually aufs_iterate() holds [fdi]i_rwsem before mmap_sem, but this is not a
14069 + * problem either since any directory is not able to be mmap-ed.
14070 + * The similar scenario is applied to aufs_readlink() too.
14071 + */
14072 +
14073 +#if 0 /* stop calling security_file_mmap() */
14074 +/* cf. linux/include/linux/mman.h: calc_vm_prot_bits() */
14075 +#define AuConv_VM_PROT(f, b)   _calc_vm_trans(f, VM_##b, PROT_##b)
14076 +
14077 +static unsigned long au_arch_prot_conv(unsigned long flags)
14078 +{
14079 +       /* currently ppc64 only */
14080 +#ifdef CONFIG_PPC64
14081 +       /* cf. linux/arch/powerpc/include/asm/mman.h */
14082 +       AuDebugOn(arch_calc_vm_prot_bits(-1) != VM_SAO);
14083 +       return AuConv_VM_PROT(flags, SAO);
14084 +#else
14085 +       AuDebugOn(arch_calc_vm_prot_bits(-1));
14086 +       return 0;
14087 +#endif
14088 +}
14089 +
14090 +static unsigned long au_prot_conv(unsigned long flags)
14091 +{
14092 +       return AuConv_VM_PROT(flags, READ)
14093 +               | AuConv_VM_PROT(flags, WRITE)
14094 +               | AuConv_VM_PROT(flags, EXEC)
14095 +               | au_arch_prot_conv(flags);
14096 +}
14097 +
14098 +/* cf. linux/include/linux/mman.h: calc_vm_flag_bits() */
14099 +#define AuConv_VM_MAP(f, b)    _calc_vm_trans(f, VM_##b, MAP_##b)
14100 +
14101 +static unsigned long au_flag_conv(unsigned long flags)
14102 +{
14103 +       return AuConv_VM_MAP(flags, GROWSDOWN)
14104 +               | AuConv_VM_MAP(flags, DENYWRITE)
14105 +               | AuConv_VM_MAP(flags, LOCKED);
14106 +}
14107 +#endif
14108 +
14109 +static int aufs_mmap(struct file *file, struct vm_area_struct *vma)
14110 +{
14111 +       int err;
14112 +       const unsigned char wlock
14113 +               = (file->f_mode & FMODE_WRITE) && (vma->vm_flags & VM_SHARED);
14114 +       struct super_block *sb;
14115 +       struct file *h_file;
14116 +       struct inode *inode;
14117 +
14118 +       AuDbgVmRegion(file, vma);
14119 +
14120 +       inode = file_inode(file);
14121 +       sb = inode->i_sb;
14122 +       lockdep_off();
14123 +       si_read_lock(sb, AuLock_NOPLMW);
14124 +
14125 +       h_file = au_write_pre(file, wlock, /*wpre*/NULL);
14126 +       lockdep_on();
14127 +       err = PTR_ERR(h_file);
14128 +       if (IS_ERR(h_file))
14129 +               goto out;
14130 +
14131 +       err = 0;
14132 +       au_set_mmapped(file);
14133 +       au_vm_file_reset(vma, h_file);
14134 +       /*
14135 +        * we cannot call security_mmap_file() here since it may acquire
14136 +        * mmap_sem or i_mutex.
14137 +        *
14138 +        * err = security_mmap_file(h_file, au_prot_conv(vma->vm_flags),
14139 +        *                       au_flag_conv(vma->vm_flags));
14140 +        */
14141 +       if (!err)
14142 +               err = h_file->f_op->mmap(h_file, vma);
14143 +       if (!err) {
14144 +               au_vm_prfile_set(vma, file);
14145 +               fsstack_copy_attr_atime(inode, file_inode(h_file));
14146 +               goto out_fput; /* success */
14147 +       }
14148 +       au_unset_mmapped(file);
14149 +       au_vm_file_reset(vma, file);
14150 +
14151 +out_fput:
14152 +       lockdep_off();
14153 +       ii_write_unlock(inode);
14154 +       lockdep_on();
14155 +       fput(h_file);
14156 +out:
14157 +       lockdep_off();
14158 +       si_read_unlock(sb);
14159 +       lockdep_on();
14160 +       AuTraceErr(err);
14161 +       return err;
14162 +}
14163 +
14164 +/* ---------------------------------------------------------------------- */
14165 +
14166 +static int aufs_fsync_nondir(struct file *file, loff_t start, loff_t end,
14167 +                            int datasync)
14168 +{
14169 +       int err;
14170 +       struct au_write_pre wpre;
14171 +       struct inode *inode;
14172 +       struct file *h_file;
14173 +
14174 +       err = 0; /* -EBADF; */ /* posix? */
14175 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
14176 +               goto out;
14177 +
14178 +       inode = file_inode(file);
14179 +       au_mtx_and_read_lock(inode);
14180 +
14181 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14182 +       err = PTR_ERR(h_file);
14183 +       if (IS_ERR(h_file))
14184 +               goto out_unlock;
14185 +
14186 +       err = vfsub_fsync(h_file, &h_file->f_path, datasync);
14187 +       au_write_post(inode, h_file, &wpre, /*written*/0);
14188 +
14189 +out_unlock:
14190 +       si_read_unlock(inode->i_sb);
14191 +       inode_unlock(inode);
14192 +out:
14193 +       return err;
14194 +}
14195 +
14196 +/* no one supports this operation, currently */
14197 +#if 0
14198 +static int aufs_aio_fsync_nondir(struct kiocb *kio, int datasync)
14199 +{
14200 +       int err;
14201 +       struct au_write_pre wpre;
14202 +       struct inode *inode, *h_inode;
14203 +       struct file *file, *h_file;
14204 +
14205 +       err = 0; /* -EBADF; */ /* posix? */
14206 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
14207 +               goto out;
14208 +
14209 +       file = kio->ki_filp;
14210 +       inode = file_inode(file);
14211 +       au_mtx_and_read_lock(inode);
14212 +
14213 +       h_file = au_write_pre(file, /*do_ready*/1, &wpre);
14214 +       err = PTR_ERR(h_file);
14215 +       if (IS_ERR(h_file))
14216 +               goto out_unlock;
14217 +
14218 +       err = -ENOSYS;
14219 +       h_file = au_hf_top(file);
14220 +       if (h_file->f_op->aio_fsync) {
14221 +               h_inode = file_inode(h_file);
14222 +               if (!is_sync_kiocb(kio)) {
14223 +                       get_file(h_file);
14224 +                       fput(file);
14225 +               }
14226 +               kio->ki_filp = h_file;
14227 +               err = h_file->f_op->aio_fsync(kio, datasync);
14228 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
14229 +               if (!err)
14230 +                       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL);
14231 +               /*ignore*/
14232 +               inode_unlock(h_inode);
14233 +       }
14234 +       au_write_post(inode, h_file, &wpre, /*written*/0);
14235 +
14236 +out_unlock:
14237 +       si_read_unlock(inode->sb);
14238 +       inode_unlock(inode);
14239 +out:
14240 +       return err;
14241 +}
14242 +#endif
14243 +
14244 +static int aufs_fasync(int fd, struct file *file, int flag)
14245 +{
14246 +       int err;
14247 +       struct file *h_file;
14248 +       struct super_block *sb;
14249 +
14250 +       sb = file->f_path.dentry->d_sb;
14251 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
14252 +
14253 +       h_file = au_read_pre(file, /*keep_fi*/0);
14254 +       err = PTR_ERR(h_file);
14255 +       if (IS_ERR(h_file))
14256 +               goto out;
14257 +
14258 +       if (h_file->f_op->fasync)
14259 +               err = h_file->f_op->fasync(fd, h_file, flag);
14260 +       fput(h_file); /* instead of au_read_post() */
14261 +
14262 +out:
14263 +       si_read_unlock(sb);
14264 +       return err;
14265 +}
14266 +
14267 +static int aufs_setfl(struct file *file, unsigned long arg)
14268 +{
14269 +       int err;
14270 +       struct file *h_file;
14271 +       struct super_block *sb;
14272 +
14273 +       sb = file->f_path.dentry->d_sb;
14274 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
14275 +
14276 +       h_file = au_read_pre(file, /*keep_fi*/0);
14277 +       err = PTR_ERR(h_file);
14278 +       if (IS_ERR(h_file))
14279 +               goto out;
14280 +
14281 +       arg |= vfsub_file_flags(file) & FASYNC; /* stop calling h_file->fasync */
14282 +       err = setfl(/*unused fd*/-1, h_file, arg);
14283 +       fput(h_file); /* instead of au_read_post() */
14284 +
14285 +out:
14286 +       si_read_unlock(sb);
14287 +       return err;
14288 +}
14289 +
14290 +/* ---------------------------------------------------------------------- */
14291 +
14292 +/* no one supports this operation, currently */
14293 +#if 0
14294 +static ssize_t aufs_sendpage(struct file *file, struct page *page, int offset,
14295 +                            size_t len, loff_t *pos, int more)
14296 +{
14297 +}
14298 +#endif
14299 +
14300 +/* ---------------------------------------------------------------------- */
14301 +
14302 +const struct file_operations aufs_file_fop = {
14303 +       .owner          = THIS_MODULE,
14304 +
14305 +       .llseek         = default_llseek,
14306 +
14307 +       .read           = aufs_read,
14308 +       .write          = aufs_write,
14309 +       .read_iter      = aufs_read_iter,
14310 +       .write_iter     = aufs_write_iter,
14311 +
14312 +#ifdef CONFIG_AUFS_POLL
14313 +       .poll           = aufs_poll,
14314 +#endif
14315 +       .unlocked_ioctl = aufs_ioctl_nondir,
14316 +#ifdef CONFIG_COMPAT
14317 +       .compat_ioctl   = aufs_compat_ioctl_nondir,
14318 +#endif
14319 +       .mmap           = aufs_mmap,
14320 +       .open           = aufs_open_nondir,
14321 +       .flush          = aufs_flush_nondir,
14322 +       .release        = aufs_release_nondir,
14323 +       .fsync          = aufs_fsync_nondir,
14324 +       /* .aio_fsync   = aufs_aio_fsync_nondir, */
14325 +       .fasync         = aufs_fasync,
14326 +       /* .sendpage    = aufs_sendpage, */
14327 +       .setfl          = aufs_setfl,
14328 +       .splice_write   = aufs_splice_write,
14329 +       .splice_read    = aufs_splice_read,
14330 +#if 0
14331 +       .aio_splice_write = aufs_aio_splice_write,
14332 +       .aio_splice_read  = aufs_aio_splice_read,
14333 +#endif
14334 +       .fallocate      = aufs_fallocate
14335 +};
14336 diff -urN /usr/share/empty/fs/aufs/fstype.h linux/fs/aufs/fstype.h
14337 --- /usr/share/empty/fs/aufs/fstype.h   1970-01-01 01:00:00.000000000 +0100
14338 +++ linux/fs/aufs/fstype.h      2016-10-09 16:55:36.492701639 +0200
14339 @@ -0,0 +1,400 @@
14340 +/*
14341 + * Copyright (C) 2005-2016 Junjiro R. Okajima
14342 + *
14343 + * This program, aufs is free software; you can redistribute it and/or modify
14344 + * it under the terms of the GNU General Public License as published by
14345 + * the Free Software Foundation; either version 2 of the License, or
14346 + * (at your option) any later version.
14347 + *
14348 + * This program is distributed in the hope that it will be useful,
14349 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14350 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14351 + * GNU General Public License for more details.
14352 + *
14353 + * You should have received a copy of the GNU General Public License
14354 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14355 + */
14356 +
14357 +/*
14358 + * judging filesystem type
14359 + */
14360 +
14361 +#ifndef __AUFS_FSTYPE_H__
14362 +#define __AUFS_FSTYPE_H__
14363 +
14364 +#ifdef __KERNEL__
14365 +
14366 +#include <linux/fs.h>
14367 +#include <linux/magic.h>
14368 +#include <linux/nfs_fs.h>
14369 +#include <linux/romfs_fs.h>
14370 +
14371 +static inline int au_test_aufs(struct super_block *sb)
14372 +{
14373 +       return sb->s_magic == AUFS_SUPER_MAGIC;
14374 +}
14375 +
14376 +static inline const char *au_sbtype(struct super_block *sb)
14377 +{
14378 +       return sb->s_type->name;
14379 +}
14380 +
14381 +static inline int au_test_iso9660(struct super_block *sb __maybe_unused)
14382 +{
14383 +#if IS_ENABLED(CONFIG_ISO9660_FS)
14384 +       return sb->s_magic == ISOFS_SUPER_MAGIC;
14385 +#else
14386 +       return 0;
14387 +#endif
14388 +}
14389 +
14390 +static inline int au_test_romfs(struct super_block *sb __maybe_unused)
14391 +{
14392 +#if IS_ENABLED(CONFIG_ROMFS_FS)
14393 +       return sb->s_magic == ROMFS_MAGIC;
14394 +#else
14395 +       return 0;
14396 +#endif
14397 +}
14398 +
14399 +static inline int au_test_cramfs(struct super_block *sb __maybe_unused)
14400 +{
14401 +#if IS_ENABLED(CONFIG_CRAMFS)
14402 +       return sb->s_magic == CRAMFS_MAGIC;
14403 +#endif
14404 +       return 0;
14405 +}
14406 +
14407 +static inline int au_test_nfs(struct super_block *sb __maybe_unused)
14408 +{
14409 +#if IS_ENABLED(CONFIG_NFS_FS)
14410 +       return sb->s_magic == NFS_SUPER_MAGIC;
14411 +#else
14412 +       return 0;
14413 +#endif
14414 +}
14415 +
14416 +static inline int au_test_fuse(struct super_block *sb __maybe_unused)
14417 +{
14418 +#if IS_ENABLED(CONFIG_FUSE_FS)
14419 +       return sb->s_magic == FUSE_SUPER_MAGIC;
14420 +#else
14421 +       return 0;
14422 +#endif
14423 +}
14424 +
14425 +static inline int au_test_xfs(struct super_block *sb __maybe_unused)
14426 +{
14427 +#if IS_ENABLED(CONFIG_XFS_FS)
14428 +       return sb->s_magic == XFS_SB_MAGIC;
14429 +#else
14430 +       return 0;
14431 +#endif
14432 +}
14433 +
14434 +static inline int au_test_tmpfs(struct super_block *sb __maybe_unused)
14435 +{
14436 +#ifdef CONFIG_TMPFS
14437 +       return sb->s_magic == TMPFS_MAGIC;
14438 +#else
14439 +       return 0;
14440 +#endif
14441 +}
14442 +
14443 +static inline int au_test_ecryptfs(struct super_block *sb __maybe_unused)
14444 +{
14445 +#if IS_ENABLED(CONFIG_ECRYPT_FS)
14446 +       return !strcmp(au_sbtype(sb), "ecryptfs");
14447 +#else
14448 +       return 0;
14449 +#endif
14450 +}
14451 +
14452 +static inline int au_test_ramfs(struct super_block *sb)
14453 +{
14454 +       return sb->s_magic == RAMFS_MAGIC;
14455 +}
14456 +
14457 +static inline int au_test_ubifs(struct super_block *sb __maybe_unused)
14458 +{
14459 +#if IS_ENABLED(CONFIG_UBIFS_FS)
14460 +       return sb->s_magic == UBIFS_SUPER_MAGIC;
14461 +#else
14462 +       return 0;
14463 +#endif
14464 +}
14465 +
14466 +static inline int au_test_procfs(struct super_block *sb __maybe_unused)
14467 +{
14468 +#ifdef CONFIG_PROC_FS
14469 +       return sb->s_magic == PROC_SUPER_MAGIC;
14470 +#else
14471 +       return 0;
14472 +#endif
14473 +}
14474 +
14475 +static inline int au_test_sysfs(struct super_block *sb __maybe_unused)
14476 +{
14477 +#ifdef CONFIG_SYSFS
14478 +       return sb->s_magic == SYSFS_MAGIC;
14479 +#else
14480 +       return 0;
14481 +#endif
14482 +}
14483 +
14484 +static inline int au_test_configfs(struct super_block *sb __maybe_unused)
14485 +{
14486 +#if IS_ENABLED(CONFIG_CONFIGFS_FS)
14487 +       return sb->s_magic == CONFIGFS_MAGIC;
14488 +#else
14489 +       return 0;
14490 +#endif
14491 +}
14492 +
14493 +static inline int au_test_minix(struct super_block *sb __maybe_unused)
14494 +{
14495 +#if IS_ENABLED(CONFIG_MINIX_FS)
14496 +       return sb->s_magic == MINIX3_SUPER_MAGIC
14497 +               || sb->s_magic == MINIX2_SUPER_MAGIC
14498 +               || sb->s_magic == MINIX2_SUPER_MAGIC2
14499 +               || sb->s_magic == MINIX_SUPER_MAGIC
14500 +               || sb->s_magic == MINIX_SUPER_MAGIC2;
14501 +#else
14502 +       return 0;
14503 +#endif
14504 +}
14505 +
14506 +static inline int au_test_fat(struct super_block *sb __maybe_unused)
14507 +{
14508 +#if IS_ENABLED(CONFIG_FAT_FS)
14509 +       return sb->s_magic == MSDOS_SUPER_MAGIC;
14510 +#else
14511 +       return 0;
14512 +#endif
14513 +}
14514 +
14515 +static inline int au_test_msdos(struct super_block *sb)
14516 +{
14517 +       return au_test_fat(sb);
14518 +}
14519 +
14520 +static inline int au_test_vfat(struct super_block *sb)
14521 +{
14522 +       return au_test_fat(sb);
14523 +}
14524 +
14525 +static inline int au_test_securityfs(struct super_block *sb __maybe_unused)
14526 +{
14527 +#ifdef CONFIG_SECURITYFS
14528 +       return sb->s_magic == SECURITYFS_MAGIC;
14529 +#else
14530 +       return 0;
14531 +#endif
14532 +}
14533 +
14534 +static inline int au_test_squashfs(struct super_block *sb __maybe_unused)
14535 +{
14536 +#if IS_ENABLED(CONFIG_SQUASHFS)
14537 +       return sb->s_magic == SQUASHFS_MAGIC;
14538 +#else
14539 +       return 0;
14540 +#endif
14541 +}
14542 +
14543 +static inline int au_test_btrfs(struct super_block *sb __maybe_unused)
14544 +{
14545 +#if IS_ENABLED(CONFIG_BTRFS_FS)
14546 +       return sb->s_magic == BTRFS_SUPER_MAGIC;
14547 +#else
14548 +       return 0;
14549 +#endif
14550 +}
14551 +
14552 +static inline int au_test_xenfs(struct super_block *sb __maybe_unused)
14553 +{
14554 +#if IS_ENABLED(CONFIG_XENFS)
14555 +       return sb->s_magic == XENFS_SUPER_MAGIC;
14556 +#else
14557 +       return 0;
14558 +#endif
14559 +}
14560 +
14561 +static inline int au_test_debugfs(struct super_block *sb __maybe_unused)
14562 +{
14563 +#ifdef CONFIG_DEBUG_FS
14564 +       return sb->s_magic == DEBUGFS_MAGIC;
14565 +#else
14566 +       return 0;
14567 +#endif
14568 +}
14569 +
14570 +static inline int au_test_nilfs(struct super_block *sb __maybe_unused)
14571 +{
14572 +#if IS_ENABLED(CONFIG_NILFS)
14573 +       return sb->s_magic == NILFS_SUPER_MAGIC;
14574 +#else
14575 +       return 0;
14576 +#endif
14577 +}
14578 +
14579 +static inline int au_test_hfsplus(struct super_block *sb __maybe_unused)
14580 +{
14581 +#if IS_ENABLED(CONFIG_HFSPLUS_FS)
14582 +       return sb->s_magic == HFSPLUS_SUPER_MAGIC;
14583 +#else
14584 +       return 0;
14585 +#endif
14586 +}
14587 +
14588 +/* ---------------------------------------------------------------------- */
14589 +/*
14590 + * they can't be an aufs branch.
14591 + */
14592 +static inline int au_test_fs_unsuppoted(struct super_block *sb)
14593 +{
14594 +       return
14595 +#ifndef CONFIG_AUFS_BR_RAMFS
14596 +               au_test_ramfs(sb) ||
14597 +#endif
14598 +               au_test_procfs(sb)
14599 +               || au_test_sysfs(sb)
14600 +               || au_test_configfs(sb)
14601 +               || au_test_debugfs(sb)
14602 +               || au_test_securityfs(sb)
14603 +               || au_test_xenfs(sb)
14604 +               || au_test_ecryptfs(sb)
14605 +               /* || !strcmp(au_sbtype(sb), "unionfs") */
14606 +               || au_test_aufs(sb); /* will be supported in next version */
14607 +}
14608 +
14609 +static inline int au_test_fs_remote(struct super_block *sb)
14610 +{
14611 +       return !au_test_tmpfs(sb)
14612 +#ifdef CONFIG_AUFS_BR_RAMFS
14613 +               && !au_test_ramfs(sb)
14614 +#endif
14615 +               && !(sb->s_type->fs_flags & FS_REQUIRES_DEV);
14616 +}
14617 +
14618 +/* ---------------------------------------------------------------------- */
14619 +
14620 +/*
14621 + * Note: these functions (below) are created after reading ->getattr() in all
14622 + * filesystems under linux/fs. it means we have to do so in every update...
14623 + */
14624 +
14625 +/*
14626 + * some filesystems require getattr to refresh the inode attributes before
14627 + * referencing.
14628 + * in most cases, we can rely on the inode attribute in NFS (or every remote fs)
14629 + * and leave the work for d_revalidate()
14630 + */
14631 +static inline int au_test_fs_refresh_iattr(struct super_block *sb)
14632 +{
14633 +       return au_test_nfs(sb)
14634 +               || au_test_fuse(sb)
14635 +               /* || au_test_btrfs(sb) */      /* untested */
14636 +               ;
14637 +}
14638 +
14639 +/*
14640 + * filesystems which don't maintain i_size or i_blocks.
14641 + */
14642 +static inline int au_test_fs_bad_iattr_size(struct super_block *sb)
14643 +{
14644 +       return au_test_xfs(sb)
14645 +               || au_test_btrfs(sb)
14646 +               || au_test_ubifs(sb)
14647 +               || au_test_hfsplus(sb)  /* maintained, but incorrect */
14648 +               /* || au_test_minix(sb) */      /* untested */
14649 +               ;
14650 +}
14651 +
14652 +/*
14653 + * filesystems which don't store the correct value in some of their inode
14654 + * attributes.
14655 + */
14656 +static inline int au_test_fs_bad_iattr(struct super_block *sb)
14657 +{
14658 +       return au_test_fs_bad_iattr_size(sb)
14659 +               || au_test_fat(sb)
14660 +               || au_test_msdos(sb)
14661 +               || au_test_vfat(sb);
14662 +}
14663 +
14664 +/* they don't check i_nlink in link(2) */
14665 +static inline int au_test_fs_no_limit_nlink(struct super_block *sb)
14666 +{
14667 +       return au_test_tmpfs(sb)
14668 +#ifdef CONFIG_AUFS_BR_RAMFS
14669 +               || au_test_ramfs(sb)
14670 +#endif
14671 +               || au_test_ubifs(sb)
14672 +               || au_test_hfsplus(sb);
14673 +}
14674 +
14675 +/*
14676 + * filesystems which sets S_NOATIME and S_NOCMTIME.
14677 + */
14678 +static inline int au_test_fs_notime(struct super_block *sb)
14679 +{
14680 +       return au_test_nfs(sb)
14681 +               || au_test_fuse(sb)
14682 +               || au_test_ubifs(sb)
14683 +               ;
14684 +}
14685 +
14686 +/* temporary support for i#1 in cramfs */
14687 +static inline int au_test_fs_unique_ino(struct inode *inode)
14688 +{
14689 +       if (au_test_cramfs(inode->i_sb))
14690 +               return inode->i_ino != 1;
14691 +       return 1;
14692 +}
14693 +
14694 +/* ---------------------------------------------------------------------- */
14695 +
14696 +/*
14697 + * the filesystem where the xino files placed must support i/o after unlink and
14698 + * maintain i_size and i_blocks.
14699 + */
14700 +static inline int au_test_fs_bad_xino(struct super_block *sb)
14701 +{
14702 +       return au_test_fs_remote(sb)
14703 +               || au_test_fs_bad_iattr_size(sb)
14704 +               /* don't want unnecessary work for xino */
14705 +               || au_test_aufs(sb)
14706 +               || au_test_ecryptfs(sb)
14707 +               || au_test_nilfs(sb);
14708 +}
14709 +
14710 +static inline int au_test_fs_trunc_xino(struct super_block *sb)
14711 +{
14712 +       return au_test_tmpfs(sb)
14713 +               || au_test_ramfs(sb);
14714 +}
14715 +
14716 +/*
14717 + * test if the @sb is real-readonly.
14718 + */
14719 +static inline int au_test_fs_rr(struct super_block *sb)
14720 +{
14721 +       return au_test_squashfs(sb)
14722 +               || au_test_iso9660(sb)
14723 +               || au_test_cramfs(sb)
14724 +               || au_test_romfs(sb);
14725 +}
14726 +
14727 +/*
14728 + * test if the @inode is nfs with 'noacl' option
14729 + * NFS always sets MS_POSIXACL regardless its mount option 'noacl.'
14730 + */
14731 +static inline int au_test_nfs_noacl(struct inode *inode)
14732 +{
14733 +       return au_test_nfs(inode->i_sb)
14734 +               /* && IS_POSIXACL(inode) */
14735 +               && !nfs_server_capable(inode, NFS_CAP_ACLS);
14736 +}
14737 +
14738 +#endif /* __KERNEL__ */
14739 +#endif /* __AUFS_FSTYPE_H__ */
14740 diff -urN /usr/share/empty/fs/aufs/hfsnotify.c linux/fs/aufs/hfsnotify.c
14741 --- /usr/share/empty/fs/aufs/hfsnotify.c        1970-01-01 01:00:00.000000000 +0100
14742 +++ linux/fs/aufs/hfsnotify.c   2016-10-09 16:55:36.492701639 +0200
14743 @@ -0,0 +1,287 @@
14744 +/*
14745 + * Copyright (C) 2005-2016 Junjiro R. Okajima
14746 + *
14747 + * This program, aufs is free software; you can redistribute it and/or modify
14748 + * it under the terms of the GNU General Public License as published by
14749 + * the Free Software Foundation; either version 2 of the License, or
14750 + * (at your option) any later version.
14751 + *
14752 + * This program is distributed in the hope that it will be useful,
14753 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14754 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14755 + * GNU General Public License for more details.
14756 + *
14757 + * You should have received a copy of the GNU General Public License
14758 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14759 + */
14760 +
14761 +/*
14762 + * fsnotify for the lower directories
14763 + */
14764 +
14765 +#include "aufs.h"
14766 +
14767 +/* FS_IN_IGNORED is unnecessary */
14768 +static const __u32 AuHfsnMask = (FS_MOVED_TO | FS_MOVED_FROM | FS_DELETE
14769 +                                | FS_CREATE | FS_EVENT_ON_CHILD);
14770 +static DECLARE_WAIT_QUEUE_HEAD(au_hfsn_wq);
14771 +static __cacheline_aligned_in_smp atomic64_t au_hfsn_ifree = ATOMIC64_INIT(0);
14772 +
14773 +static void au_hfsn_free_mark(struct fsnotify_mark *mark)
14774 +{
14775 +       struct au_hnotify *hn = container_of(mark, struct au_hnotify,
14776 +                                            hn_mark);
14777 +       /* AuDbg("here\n"); */
14778 +       au_cache_dfree_hnotify(hn);
14779 +       smp_mb__before_atomic();
14780 +       if (atomic64_dec_and_test(&au_hfsn_ifree))
14781 +               wake_up(&au_hfsn_wq);
14782 +}
14783 +
14784 +static int au_hfsn_alloc(struct au_hinode *hinode)
14785 +{
14786 +       int err;
14787 +       struct au_hnotify *hn;
14788 +       struct super_block *sb;
14789 +       struct au_branch *br;
14790 +       struct fsnotify_mark *mark;
14791 +       aufs_bindex_t bindex;
14792 +
14793 +       hn = hinode->hi_notify;
14794 +       sb = hn->hn_aufs_inode->i_sb;
14795 +       bindex = au_br_index(sb, hinode->hi_id);
14796 +       br = au_sbr(sb, bindex);
14797 +       AuDebugOn(!br->br_hfsn);
14798 +
14799 +       mark = &hn->hn_mark;
14800 +       fsnotify_init_mark(mark, au_hfsn_free_mark);
14801 +       mark->mask = AuHfsnMask;
14802 +       /*
14803 +        * by udba rename or rmdir, aufs assign a new inode to the known
14804 +        * h_inode, so specify 1 to allow dups.
14805 +        */
14806 +       lockdep_off();
14807 +       err = fsnotify_add_mark(mark, br->br_hfsn->hfsn_group, hinode->hi_inode,
14808 +                                /*mnt*/NULL, /*allow_dups*/1);
14809 +       lockdep_on();
14810 +
14811 +       return err;
14812 +}
14813 +
14814 +static int au_hfsn_free(struct au_hinode *hinode, struct au_hnotify *hn)
14815 +{
14816 +       struct fsnotify_mark *mark;
14817 +       unsigned long long ull;
14818 +       struct fsnotify_group *group;
14819 +
14820 +       ull = atomic64_inc_return(&au_hfsn_ifree);
14821 +       BUG_ON(!ull);
14822 +
14823 +       mark = &hn->hn_mark;
14824 +       spin_lock(&mark->lock);
14825 +       group = mark->group;
14826 +       fsnotify_get_group(group);
14827 +       spin_unlock(&mark->lock);
14828 +       lockdep_off();
14829 +       fsnotify_destroy_mark(mark, group);
14830 +       fsnotify_put_mark(mark);
14831 +       fsnotify_put_group(group);
14832 +       lockdep_on();
14833 +
14834 +       /* free hn by myself */
14835 +       return 0;
14836 +}
14837 +
14838 +/* ---------------------------------------------------------------------- */
14839 +
14840 +static void au_hfsn_ctl(struct au_hinode *hinode, int do_set)
14841 +{
14842 +       struct fsnotify_mark *mark;
14843 +
14844 +       mark = &hinode->hi_notify->hn_mark;
14845 +       spin_lock(&mark->lock);
14846 +       if (do_set) {
14847 +               AuDebugOn(mark->mask & AuHfsnMask);
14848 +               mark->mask |= AuHfsnMask;
14849 +       } else {
14850 +               AuDebugOn(!(mark->mask & AuHfsnMask));
14851 +               mark->mask &= ~AuHfsnMask;
14852 +       }
14853 +       spin_unlock(&mark->lock);
14854 +       /* fsnotify_recalc_inode_mask(hinode->hi_inode); */
14855 +}
14856 +
14857 +/* ---------------------------------------------------------------------- */
14858 +
14859 +/* #define AuDbgHnotify */
14860 +#ifdef AuDbgHnotify
14861 +static char *au_hfsn_name(u32 mask)
14862 +{
14863 +#ifdef CONFIG_AUFS_DEBUG
14864 +#define test_ret(flag)                         \
14865 +       do {                                    \
14866 +               if (mask & flag)                \
14867 +                       return #flag;           \
14868 +       } while (0)
14869 +       test_ret(FS_ACCESS);
14870 +       test_ret(FS_MODIFY);
14871 +       test_ret(FS_ATTRIB);
14872 +       test_ret(FS_CLOSE_WRITE);
14873 +       test_ret(FS_CLOSE_NOWRITE);
14874 +       test_ret(FS_OPEN);
14875 +       test_ret(FS_MOVED_FROM);
14876 +       test_ret(FS_MOVED_TO);
14877 +       test_ret(FS_CREATE);
14878 +       test_ret(FS_DELETE);
14879 +       test_ret(FS_DELETE_SELF);
14880 +       test_ret(FS_MOVE_SELF);
14881 +       test_ret(FS_UNMOUNT);
14882 +       test_ret(FS_Q_OVERFLOW);
14883 +       test_ret(FS_IN_IGNORED);
14884 +       test_ret(FS_ISDIR);
14885 +       test_ret(FS_IN_ONESHOT);
14886 +       test_ret(FS_EVENT_ON_CHILD);
14887 +       return "";
14888 +#undef test_ret
14889 +#else
14890 +       return "??";
14891 +#endif
14892 +}
14893 +#endif
14894 +
14895 +/* ---------------------------------------------------------------------- */
14896 +
14897 +static void au_hfsn_free_group(struct fsnotify_group *group)
14898 +{
14899 +       struct au_br_hfsnotify *hfsn = group->private;
14900 +
14901 +       /* AuDbg("here\n"); */
14902 +       au_delayed_kfree(hfsn);
14903 +}
14904 +
14905 +static int au_hfsn_handle_event(struct fsnotify_group *group,
14906 +                               struct inode *inode,
14907 +                               struct fsnotify_mark *inode_mark,
14908 +                               struct fsnotify_mark *vfsmount_mark,
14909 +                               u32 mask, void *data, int data_type,
14910 +                               const unsigned char *file_name, u32 cookie)
14911 +{
14912 +       int err;
14913 +       struct au_hnotify *hnotify;
14914 +       struct inode *h_dir, *h_inode;
14915 +       struct qstr h_child_qstr = QSTR_INIT(file_name, strlen(file_name));
14916 +
14917 +       AuDebugOn(data_type != FSNOTIFY_EVENT_INODE);
14918 +
14919 +       err = 0;
14920 +       /* if FS_UNMOUNT happens, there must be another bug */
14921 +       AuDebugOn(mask & FS_UNMOUNT);
14922 +       if (mask & (FS_IN_IGNORED | FS_UNMOUNT))
14923 +               goto out;
14924 +
14925 +       h_dir = inode;
14926 +       h_inode = NULL;
14927 +#ifdef AuDbgHnotify
14928 +       au_debug_on();
14929 +       if (1 || h_child_qstr.len != sizeof(AUFS_XINO_FNAME) - 1
14930 +           || strncmp(h_child_qstr.name, AUFS_XINO_FNAME, h_child_qstr.len)) {
14931 +               AuDbg("i%lu, mask 0x%x %s, hcname %.*s, hi%lu\n",
14932 +                     h_dir->i_ino, mask, au_hfsn_name(mask),
14933 +                     AuLNPair(&h_child_qstr), h_inode ? h_inode->i_ino : 0);
14934 +               /* WARN_ON(1); */
14935 +       }
14936 +       au_debug_off();
14937 +#endif
14938 +
14939 +       AuDebugOn(!inode_mark);
14940 +       hnotify = container_of(inode_mark, struct au_hnotify, hn_mark);
14941 +       err = au_hnotify(h_dir, hnotify, mask, &h_child_qstr, h_inode);
14942 +
14943 +out:
14944 +       return err;
14945 +}
14946 +
14947 +static struct fsnotify_ops au_hfsn_ops = {
14948 +       .handle_event           = au_hfsn_handle_event,
14949 +       .free_group_priv        = au_hfsn_free_group
14950 +};
14951 +
14952 +/* ---------------------------------------------------------------------- */
14953 +
14954 +static void au_hfsn_fin_br(struct au_branch *br)
14955 +{
14956 +       struct au_br_hfsnotify *hfsn;
14957 +
14958 +       hfsn = br->br_hfsn;
14959 +       if (hfsn) {
14960 +               lockdep_off();
14961 +               fsnotify_put_group(hfsn->hfsn_group);
14962 +               lockdep_on();
14963 +       }
14964 +}
14965 +
14966 +static int au_hfsn_init_br(struct au_branch *br, int perm)
14967 +{
14968 +       int err;
14969 +       struct fsnotify_group *group;
14970 +       struct au_br_hfsnotify *hfsn;
14971 +
14972 +       err = 0;
14973 +       br->br_hfsn = NULL;
14974 +       if (!au_br_hnotifyable(perm))
14975 +               goto out;
14976 +
14977 +       err = -ENOMEM;
14978 +       hfsn = kmalloc(sizeof(*hfsn), GFP_NOFS);
14979 +       if (unlikely(!hfsn))
14980 +               goto out;
14981 +
14982 +       err = 0;
14983 +       group = fsnotify_alloc_group(&au_hfsn_ops);
14984 +       if (IS_ERR(group)) {
14985 +               err = PTR_ERR(group);
14986 +               pr_err("fsnotify_alloc_group() failed, %d\n", err);
14987 +               goto out_hfsn;
14988 +       }
14989 +
14990 +       group->private = hfsn;
14991 +       hfsn->hfsn_group = group;
14992 +       br->br_hfsn = hfsn;
14993 +       goto out; /* success */
14994 +
14995 +out_hfsn:
14996 +       au_delayed_kfree(hfsn);
14997 +out:
14998 +       return err;
14999 +}
15000 +
15001 +static int au_hfsn_reset_br(unsigned int udba, struct au_branch *br, int perm)
15002 +{
15003 +       int err;
15004 +
15005 +       err = 0;
15006 +       if (!br->br_hfsn)
15007 +               err = au_hfsn_init_br(br, perm);
15008 +
15009 +       return err;
15010 +}
15011 +
15012 +/* ---------------------------------------------------------------------- */
15013 +
15014 +static void au_hfsn_fin(void)
15015 +{
15016 +       AuDbg("au_hfsn_ifree %lld\n", (long long)atomic64_read(&au_hfsn_ifree));
15017 +       wait_event(au_hfsn_wq, !atomic64_read(&au_hfsn_ifree));
15018 +}
15019 +
15020 +const struct au_hnotify_op au_hnotify_op = {
15021 +       .ctl            = au_hfsn_ctl,
15022 +       .alloc          = au_hfsn_alloc,
15023 +       .free           = au_hfsn_free,
15024 +
15025 +       .fin            = au_hfsn_fin,
15026 +
15027 +       .reset_br       = au_hfsn_reset_br,
15028 +       .fin_br         = au_hfsn_fin_br,
15029 +       .init_br        = au_hfsn_init_br
15030 +};
15031 diff -urN /usr/share/empty/fs/aufs/hfsplus.c linux/fs/aufs/hfsplus.c
15032 --- /usr/share/empty/fs/aufs/hfsplus.c  1970-01-01 01:00:00.000000000 +0100
15033 +++ linux/fs/aufs/hfsplus.c     2016-10-09 16:55:36.492701639 +0200
15034 @@ -0,0 +1,56 @@
15035 +/*
15036 + * Copyright (C) 2010-2016 Junjiro R. Okajima
15037 + *
15038 + * This program, aufs is free software; you can redistribute it and/or modify
15039 + * it under the terms of the GNU General Public License as published by
15040 + * the Free Software Foundation; either version 2 of the License, or
15041 + * (at your option) any later version.
15042 + *
15043 + * This program is distributed in the hope that it will be useful,
15044 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15045 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15046 + * GNU General Public License for more details.
15047 + *
15048 + * You should have received a copy of the GNU General Public License
15049 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15050 + */
15051 +
15052 +/*
15053 + * special support for filesystems which aqucires an inode mutex
15054 + * at final closing a file, eg, hfsplus.
15055 + *
15056 + * This trick is very simple and stupid, just to open the file before really
15057 + * neceeary open to tell hfsplus that this is not the final closing.
15058 + * The caller should call au_h_open_pre() after acquiring the inode mutex,
15059 + * and au_h_open_post() after releasing it.
15060 + */
15061 +
15062 +#include "aufs.h"
15063 +
15064 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
15065 +                          int force_wr)
15066 +{
15067 +       struct file *h_file;
15068 +       struct dentry *h_dentry;
15069 +
15070 +       h_dentry = au_h_dptr(dentry, bindex);
15071 +       AuDebugOn(!h_dentry);
15072 +       AuDebugOn(d_is_negative(h_dentry));
15073 +
15074 +       h_file = NULL;
15075 +       if (au_test_hfsplus(h_dentry->d_sb)
15076 +           && d_is_reg(h_dentry))
15077 +               h_file = au_h_open(dentry, bindex,
15078 +                                  O_RDONLY | O_NOATIME | O_LARGEFILE,
15079 +                                  /*file*/NULL, force_wr);
15080 +       return h_file;
15081 +}
15082 +
15083 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
15084 +                   struct file *h_file)
15085 +{
15086 +       if (h_file) {
15087 +               fput(h_file);
15088 +               au_sbr_put(dentry->d_sb, bindex);
15089 +       }
15090 +}
15091 diff -urN /usr/share/empty/fs/aufs/hnotify.c linux/fs/aufs/hnotify.c
15092 --- /usr/share/empty/fs/aufs/hnotify.c  1970-01-01 01:00:00.000000000 +0100
15093 +++ linux/fs/aufs/hnotify.c     2016-10-09 16:55:36.492701639 +0200
15094 @@ -0,0 +1,723 @@
15095 +/*
15096 + * Copyright (C) 2005-2016 Junjiro R. Okajima
15097 + *
15098 + * This program, aufs is free software; you can redistribute it and/or modify
15099 + * it under the terms of the GNU General Public License as published by
15100 + * the Free Software Foundation; either version 2 of the License, or
15101 + * (at your option) any later version.
15102 + *
15103 + * This program is distributed in the hope that it will be useful,
15104 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15105 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15106 + * GNU General Public License for more details.
15107 + *
15108 + * You should have received a copy of the GNU General Public License
15109 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15110 + */
15111 +
15112 +/*
15113 + * abstraction to notify the direct changes on lower directories
15114 + */
15115 +
15116 +#include "aufs.h"
15117 +
15118 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode)
15119 +{
15120 +       int err;
15121 +       struct au_hnotify *hn;
15122 +
15123 +       err = -ENOMEM;
15124 +       hn = au_cache_alloc_hnotify();
15125 +       if (hn) {
15126 +               hn->hn_aufs_inode = inode;
15127 +               hinode->hi_notify = hn;
15128 +               err = au_hnotify_op.alloc(hinode);
15129 +               AuTraceErr(err);
15130 +               if (unlikely(err)) {
15131 +                       hinode->hi_notify = NULL;
15132 +                       au_cache_dfree_hnotify(hn);
15133 +                       /*
15134 +                        * The upper dir was removed by udba, but the same named
15135 +                        * dir left. In this case, aufs assignes a new inode
15136 +                        * number and set the monitor again.
15137 +                        * For the lower dir, the old monitnor is still left.
15138 +                        */
15139 +                       if (err == -EEXIST)
15140 +                               err = 0;
15141 +               }
15142 +       }
15143 +
15144 +       AuTraceErr(err);
15145 +       return err;
15146 +}
15147 +
15148 +void au_hn_free(struct au_hinode *hinode)
15149 +{
15150 +       struct au_hnotify *hn;
15151 +
15152 +       hn = hinode->hi_notify;
15153 +       if (hn) {
15154 +               hinode->hi_notify = NULL;
15155 +               if (au_hnotify_op.free(hinode, hn))
15156 +                       au_cache_dfree_hnotify(hn);
15157 +       }
15158 +}
15159 +
15160 +/* ---------------------------------------------------------------------- */
15161 +
15162 +void au_hn_ctl(struct au_hinode *hinode, int do_set)
15163 +{
15164 +       if (hinode->hi_notify)
15165 +               au_hnotify_op.ctl(hinode, do_set);
15166 +}
15167 +
15168 +void au_hn_reset(struct inode *inode, unsigned int flags)
15169 +{
15170 +       aufs_bindex_t bindex, bbot;
15171 +       struct inode *hi;
15172 +       struct dentry *iwhdentry;
15173 +
15174 +       bbot = au_ibbot(inode);
15175 +       for (bindex = au_ibtop(inode); bindex <= bbot; bindex++) {
15176 +               hi = au_h_iptr(inode, bindex);
15177 +               if (!hi)
15178 +                       continue;
15179 +
15180 +               /* inode_lock_nested(hi, AuLsc_I_CHILD); */
15181 +               iwhdentry = au_hi_wh(inode, bindex);
15182 +               if (iwhdentry)
15183 +                       dget(iwhdentry);
15184 +               au_igrab(hi);
15185 +               au_set_h_iptr(inode, bindex, NULL, 0);
15186 +               au_set_h_iptr(inode, bindex, au_igrab(hi),
15187 +                             flags & ~AuHi_XINO);
15188 +               iput(hi);
15189 +               dput(iwhdentry);
15190 +               /* inode_unlock(hi); */
15191 +       }
15192 +}
15193 +
15194 +/* ---------------------------------------------------------------------- */
15195 +
15196 +static int hn_xino(struct inode *inode, struct inode *h_inode)
15197 +{
15198 +       int err;
15199 +       aufs_bindex_t bindex, bbot, bfound, btop;
15200 +       struct inode *h_i;
15201 +
15202 +       err = 0;
15203 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15204 +               pr_warn("branch root dir was changed\n");
15205 +               goto out;
15206 +       }
15207 +
15208 +       bfound = -1;
15209 +       bbot = au_ibbot(inode);
15210 +       btop = au_ibtop(inode);
15211 +#if 0 /* reserved for future use */
15212 +       if (bindex == bbot) {
15213 +               /* keep this ino in rename case */
15214 +               goto out;
15215 +       }
15216 +#endif
15217 +       for (bindex = btop; bindex <= bbot; bindex++)
15218 +               if (au_h_iptr(inode, bindex) == h_inode) {
15219 +                       bfound = bindex;
15220 +                       break;
15221 +               }
15222 +       if (bfound < 0)
15223 +               goto out;
15224 +
15225 +       for (bindex = btop; bindex <= bbot; bindex++) {
15226 +               h_i = au_h_iptr(inode, bindex);
15227 +               if (!h_i)
15228 +                       continue;
15229 +
15230 +               err = au_xino_write(inode->i_sb, bindex, h_i->i_ino, /*ino*/0);
15231 +               /* ignore this error */
15232 +               /* bad action? */
15233 +       }
15234 +
15235 +       /* children inode number will be broken */
15236 +
15237 +out:
15238 +       AuTraceErr(err);
15239 +       return err;
15240 +}
15241 +
15242 +static int hn_gen_tree(struct dentry *dentry)
15243 +{
15244 +       int err, i, j, ndentry;
15245 +       struct au_dcsub_pages dpages;
15246 +       struct au_dpage *dpage;
15247 +       struct dentry **dentries;
15248 +
15249 +       err = au_dpages_init(&dpages, GFP_NOFS);
15250 +       if (unlikely(err))
15251 +               goto out;
15252 +       err = au_dcsub_pages(&dpages, dentry, NULL, NULL);
15253 +       if (unlikely(err))
15254 +               goto out_dpages;
15255 +
15256 +       for (i = 0; i < dpages.ndpage; i++) {
15257 +               dpage = dpages.dpages + i;
15258 +               dentries = dpage->dentries;
15259 +               ndentry = dpage->ndentry;
15260 +               for (j = 0; j < ndentry; j++) {
15261 +                       struct dentry *d;
15262 +
15263 +                       d = dentries[j];
15264 +                       if (IS_ROOT(d))
15265 +                               continue;
15266 +
15267 +                       au_digen_dec(d);
15268 +                       if (d_really_is_positive(d))
15269 +                               /* todo: reset children xino?
15270 +                                  cached children only? */
15271 +                               au_iigen_dec(d_inode(d));
15272 +               }
15273 +       }
15274 +
15275 +out_dpages:
15276 +       au_dpages_free(&dpages);
15277 +
15278 +#if 0
15279 +       /* discard children */
15280 +       dentry_unhash(dentry);
15281 +       dput(dentry);
15282 +#endif
15283 +out:
15284 +       return err;
15285 +}
15286 +
15287 +/*
15288 + * return 0 if processed.
15289 + */
15290 +static int hn_gen_by_inode(char *name, unsigned int nlen, struct inode *inode,
15291 +                          const unsigned int isdir)
15292 +{
15293 +       int err;
15294 +       struct dentry *d;
15295 +       struct qstr *dname;
15296 +
15297 +       err = 1;
15298 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15299 +               pr_warn("branch root dir was changed\n");
15300 +               err = 0;
15301 +               goto out;
15302 +       }
15303 +
15304 +       if (!isdir) {
15305 +               AuDebugOn(!name);
15306 +               au_iigen_dec(inode);
15307 +               spin_lock(&inode->i_lock);
15308 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
15309 +                       spin_lock(&d->d_lock);
15310 +                       dname = &d->d_name;
15311 +                       if (dname->len != nlen
15312 +                           && memcmp(dname->name, name, nlen)) {
15313 +                               spin_unlock(&d->d_lock);
15314 +                               continue;
15315 +                       }
15316 +                       err = 0;
15317 +                       au_digen_dec(d);
15318 +                       spin_unlock(&d->d_lock);
15319 +                       break;
15320 +               }
15321 +               spin_unlock(&inode->i_lock);
15322 +       } else {
15323 +               au_fset_si(au_sbi(inode->i_sb), FAILED_REFRESH_DIR);
15324 +               d = d_find_any_alias(inode);
15325 +               if (!d) {
15326 +                       au_iigen_dec(inode);
15327 +                       goto out;
15328 +               }
15329 +
15330 +               spin_lock(&d->d_lock);
15331 +               dname = &d->d_name;
15332 +               if (dname->len == nlen && !memcmp(dname->name, name, nlen)) {
15333 +                       spin_unlock(&d->d_lock);
15334 +                       err = hn_gen_tree(d);
15335 +                       spin_lock(&d->d_lock);
15336 +               }
15337 +               spin_unlock(&d->d_lock);
15338 +               dput(d);
15339 +       }
15340 +
15341 +out:
15342 +       AuTraceErr(err);
15343 +       return err;
15344 +}
15345 +
15346 +static int hn_gen_by_name(struct dentry *dentry, const unsigned int isdir)
15347 +{
15348 +       int err;
15349 +
15350 +       if (IS_ROOT(dentry)) {
15351 +               pr_warn("branch root dir was changed\n");
15352 +               return 0;
15353 +       }
15354 +
15355 +       err = 0;
15356 +       if (!isdir) {
15357 +               au_digen_dec(dentry);
15358 +               if (d_really_is_positive(dentry))
15359 +                       au_iigen_dec(d_inode(dentry));
15360 +       } else {
15361 +               au_fset_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR);
15362 +               if (d_really_is_positive(dentry))
15363 +                       err = hn_gen_tree(dentry);
15364 +       }
15365 +
15366 +       AuTraceErr(err);
15367 +       return err;
15368 +}
15369 +
15370 +/* ---------------------------------------------------------------------- */
15371 +
15372 +/* hnotify job flags */
15373 +#define AuHnJob_XINO0          1
15374 +#define AuHnJob_GEN            (1 << 1)
15375 +#define AuHnJob_DIRENT         (1 << 2)
15376 +#define AuHnJob_ISDIR          (1 << 3)
15377 +#define AuHnJob_TRYXINO0       (1 << 4)
15378 +#define AuHnJob_MNTPNT         (1 << 5)
15379 +#define au_ftest_hnjob(flags, name)    ((flags) & AuHnJob_##name)
15380 +#define au_fset_hnjob(flags, name) \
15381 +       do { (flags) |= AuHnJob_##name; } while (0)
15382 +#define au_fclr_hnjob(flags, name) \
15383 +       do { (flags) &= ~AuHnJob_##name; } while (0)
15384 +
15385 +enum {
15386 +       AuHn_CHILD,
15387 +       AuHn_PARENT,
15388 +       AuHnLast
15389 +};
15390 +
15391 +struct au_hnotify_args {
15392 +       struct inode *h_dir, *dir, *h_child_inode;
15393 +       u32 mask;
15394 +       unsigned int flags[AuHnLast];
15395 +       unsigned int h_child_nlen;
15396 +       char h_child_name[];
15397 +};
15398 +
15399 +struct hn_job_args {
15400 +       unsigned int flags;
15401 +       struct inode *inode, *h_inode, *dir, *h_dir;
15402 +       struct dentry *dentry;
15403 +       char *h_name;
15404 +       int h_nlen;
15405 +};
15406 +
15407 +static int hn_job(struct hn_job_args *a)
15408 +{
15409 +       const unsigned int isdir = au_ftest_hnjob(a->flags, ISDIR);
15410 +       int e;
15411 +
15412 +       /* reset xino */
15413 +       if (au_ftest_hnjob(a->flags, XINO0) && a->inode)
15414 +               hn_xino(a->inode, a->h_inode); /* ignore this error */
15415 +
15416 +       if (au_ftest_hnjob(a->flags, TRYXINO0)
15417 +           && a->inode
15418 +           && a->h_inode) {
15419 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
15420 +               if (!a->h_inode->i_nlink
15421 +                   && !(a->h_inode->i_state & I_LINKABLE))
15422 +                       hn_xino(a->inode, a->h_inode); /* ignore this error */
15423 +               inode_unlock(a->h_inode);
15424 +       }
15425 +
15426 +       /* make the generation obsolete */
15427 +       if (au_ftest_hnjob(a->flags, GEN)) {
15428 +               e = -1;
15429 +               if (a->inode)
15430 +                       e = hn_gen_by_inode(a->h_name, a->h_nlen, a->inode,
15431 +                                             isdir);
15432 +               if (e && a->dentry)
15433 +                       hn_gen_by_name(a->dentry, isdir);
15434 +               /* ignore this error */
15435 +       }
15436 +
15437 +       /* make dir entries obsolete */
15438 +       if (au_ftest_hnjob(a->flags, DIRENT) && a->inode) {
15439 +               struct au_vdir *vdir;
15440 +
15441 +               vdir = au_ivdir(a->inode);
15442 +               if (vdir)
15443 +                       vdir->vd_jiffy = 0;
15444 +               /* IMustLock(a->inode); */
15445 +               /* a->inode->i_version++; */
15446 +       }
15447 +
15448 +       /* can do nothing but warn */
15449 +       if (au_ftest_hnjob(a->flags, MNTPNT)
15450 +           && a->dentry
15451 +           && d_mountpoint(a->dentry))
15452 +               pr_warn("mount-point %pd is removed or renamed\n", a->dentry);
15453 +
15454 +       return 0;
15455 +}
15456 +
15457 +/* ---------------------------------------------------------------------- */
15458 +
15459 +static struct dentry *lookup_wlock_by_name(char *name, unsigned int nlen,
15460 +                                          struct inode *dir)
15461 +{
15462 +       struct dentry *dentry, *d, *parent;
15463 +       struct qstr *dname;
15464 +
15465 +       parent = d_find_any_alias(dir);
15466 +       if (!parent)
15467 +               return NULL;
15468 +
15469 +       dentry = NULL;
15470 +       spin_lock(&parent->d_lock);
15471 +       list_for_each_entry(d, &parent->d_subdirs, d_child) {
15472 +               /* AuDbg("%pd\n", d); */
15473 +               spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
15474 +               dname = &d->d_name;
15475 +               if (dname->len != nlen || memcmp(dname->name, name, nlen))
15476 +                       goto cont_unlock;
15477 +               if (au_di(d))
15478 +                       au_digen_dec(d);
15479 +               else
15480 +                       goto cont_unlock;
15481 +               if (au_dcount(d) > 0) {
15482 +                       dentry = dget_dlock(d);
15483 +                       spin_unlock(&d->d_lock);
15484 +                       break;
15485 +               }
15486 +
15487 +cont_unlock:
15488 +               spin_unlock(&d->d_lock);
15489 +       }
15490 +       spin_unlock(&parent->d_lock);
15491 +       dput(parent);
15492 +
15493 +       if (dentry)
15494 +               di_write_lock_child(dentry);
15495 +
15496 +       return dentry;
15497 +}
15498 +
15499 +static struct inode *lookup_wlock_by_ino(struct super_block *sb,
15500 +                                        aufs_bindex_t bindex, ino_t h_ino)
15501 +{
15502 +       struct inode *inode;
15503 +       ino_t ino;
15504 +       int err;
15505 +
15506 +       inode = NULL;
15507 +       err = au_xino_read(sb, bindex, h_ino, &ino);
15508 +       if (!err && ino)
15509 +               inode = ilookup(sb, ino);
15510 +       if (!inode)
15511 +               goto out;
15512 +
15513 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15514 +               pr_warn("wrong root branch\n");
15515 +               iput(inode);
15516 +               inode = NULL;
15517 +               goto out;
15518 +       }
15519 +
15520 +       ii_write_lock_child(inode);
15521 +
15522 +out:
15523 +       return inode;
15524 +}
15525 +
15526 +static void au_hn_bh(void *_args)
15527 +{
15528 +       struct au_hnotify_args *a = _args;
15529 +       struct super_block *sb;
15530 +       aufs_bindex_t bindex, bbot, bfound;
15531 +       unsigned char xino, try_iput;
15532 +       int err;
15533 +       struct inode *inode;
15534 +       ino_t h_ino;
15535 +       struct hn_job_args args;
15536 +       struct dentry *dentry;
15537 +       struct au_sbinfo *sbinfo;
15538 +
15539 +       AuDebugOn(!_args);
15540 +       AuDebugOn(!a->h_dir);
15541 +       AuDebugOn(!a->dir);
15542 +       AuDebugOn(!a->mask);
15543 +       AuDbg("mask 0x%x, i%lu, hi%lu, hci%lu\n",
15544 +             a->mask, a->dir->i_ino, a->h_dir->i_ino,
15545 +             a->h_child_inode ? a->h_child_inode->i_ino : 0);
15546 +
15547 +       inode = NULL;
15548 +       dentry = NULL;
15549 +       /*
15550 +        * do not lock a->dir->i_mutex here
15551 +        * because of d_revalidate() may cause a deadlock.
15552 +        */
15553 +       sb = a->dir->i_sb;
15554 +       AuDebugOn(!sb);
15555 +       sbinfo = au_sbi(sb);
15556 +       AuDebugOn(!sbinfo);
15557 +       si_write_lock(sb, AuLock_NOPLMW);
15558 +
15559 +       ii_read_lock_parent(a->dir);
15560 +       bfound = -1;
15561 +       bbot = au_ibbot(a->dir);
15562 +       for (bindex = au_ibtop(a->dir); bindex <= bbot; bindex++)
15563 +               if (au_h_iptr(a->dir, bindex) == a->h_dir) {
15564 +                       bfound = bindex;
15565 +                       break;
15566 +               }
15567 +       ii_read_unlock(a->dir);
15568 +       if (unlikely(bfound < 0))
15569 +               goto out;
15570 +
15571 +       xino = !!au_opt_test(au_mntflags(sb), XINO);
15572 +       h_ino = 0;
15573 +       if (a->h_child_inode)
15574 +               h_ino = a->h_child_inode->i_ino;
15575 +
15576 +       if (a->h_child_nlen
15577 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], GEN)
15578 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], MNTPNT)))
15579 +               dentry = lookup_wlock_by_name(a->h_child_name, a->h_child_nlen,
15580 +                                             a->dir);
15581 +       try_iput = 0;
15582 +       if (dentry && d_really_is_positive(dentry))
15583 +               inode = d_inode(dentry);
15584 +       if (xino && !inode && h_ino
15585 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], XINO0)
15586 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], TRYXINO0)
15587 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], GEN))) {
15588 +               inode = lookup_wlock_by_ino(sb, bfound, h_ino);
15589 +               try_iput = 1;
15590 +       }
15591 +
15592 +       args.flags = a->flags[AuHn_CHILD];
15593 +       args.dentry = dentry;
15594 +       args.inode = inode;
15595 +       args.h_inode = a->h_child_inode;
15596 +       args.dir = a->dir;
15597 +       args.h_dir = a->h_dir;
15598 +       args.h_name = a->h_child_name;
15599 +       args.h_nlen = a->h_child_nlen;
15600 +       err = hn_job(&args);
15601 +       if (dentry) {
15602 +               if (au_di(dentry))
15603 +                       di_write_unlock(dentry);
15604 +               dput(dentry);
15605 +       }
15606 +       if (inode && try_iput) {
15607 +               ii_write_unlock(inode);
15608 +               iput(inode);
15609 +       }
15610 +
15611 +       ii_write_lock_parent(a->dir);
15612 +       args.flags = a->flags[AuHn_PARENT];
15613 +       args.dentry = NULL;
15614 +       args.inode = a->dir;
15615 +       args.h_inode = a->h_dir;
15616 +       args.dir = NULL;
15617 +       args.h_dir = NULL;
15618 +       args.h_name = NULL;
15619 +       args.h_nlen = 0;
15620 +       err = hn_job(&args);
15621 +       ii_write_unlock(a->dir);
15622 +
15623 +out:
15624 +       iput(a->h_child_inode);
15625 +       iput(a->h_dir);
15626 +       iput(a->dir);
15627 +       si_write_unlock(sb);
15628 +       au_nwt_done(&sbinfo->si_nowait);
15629 +       au_delayed_kfree(a);
15630 +}
15631 +
15632 +/* ---------------------------------------------------------------------- */
15633 +
15634 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
15635 +              struct qstr *h_child_qstr, struct inode *h_child_inode)
15636 +{
15637 +       int err, len;
15638 +       unsigned int flags[AuHnLast], f;
15639 +       unsigned char isdir, isroot, wh;
15640 +       struct inode *dir;
15641 +       struct au_hnotify_args *args;
15642 +       char *p, *h_child_name;
15643 +
15644 +       err = 0;
15645 +       AuDebugOn(!hnotify || !hnotify->hn_aufs_inode);
15646 +       dir = igrab(hnotify->hn_aufs_inode);
15647 +       if (!dir)
15648 +               goto out;
15649 +
15650 +       isroot = (dir->i_ino == AUFS_ROOT_INO);
15651 +       wh = 0;
15652 +       h_child_name = (void *)h_child_qstr->name;
15653 +       len = h_child_qstr->len;
15654 +       if (h_child_name) {
15655 +               if (len > AUFS_WH_PFX_LEN
15656 +                   && !memcmp(h_child_name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
15657 +                       h_child_name += AUFS_WH_PFX_LEN;
15658 +                       len -= AUFS_WH_PFX_LEN;
15659 +                       wh = 1;
15660 +               }
15661 +       }
15662 +
15663 +       isdir = 0;
15664 +       if (h_child_inode)
15665 +               isdir = !!S_ISDIR(h_child_inode->i_mode);
15666 +       flags[AuHn_PARENT] = AuHnJob_ISDIR;
15667 +       flags[AuHn_CHILD] = 0;
15668 +       if (isdir)
15669 +               flags[AuHn_CHILD] = AuHnJob_ISDIR;
15670 +       au_fset_hnjob(flags[AuHn_PARENT], DIRENT);
15671 +       au_fset_hnjob(flags[AuHn_CHILD], GEN);
15672 +       switch (mask & FS_EVENTS_POSS_ON_CHILD) {
15673 +       case FS_MOVED_FROM:
15674 +       case FS_MOVED_TO:
15675 +               au_fset_hnjob(flags[AuHn_CHILD], XINO0);
15676 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15677 +               /*FALLTHROUGH*/
15678 +       case FS_CREATE:
15679 +               AuDebugOn(!h_child_name);
15680 +               break;
15681 +
15682 +       case FS_DELETE:
15683 +               /*
15684 +                * aufs never be able to get this child inode.
15685 +                * revalidation should be in d_revalidate()
15686 +                * by checking i_nlink, i_generation or d_unhashed().
15687 +                */
15688 +               AuDebugOn(!h_child_name);
15689 +               au_fset_hnjob(flags[AuHn_CHILD], TRYXINO0);
15690 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15691 +               break;
15692 +
15693 +       default:
15694 +               AuDebugOn(1);
15695 +       }
15696 +
15697 +       if (wh)
15698 +               h_child_inode = NULL;
15699 +
15700 +       err = -ENOMEM;
15701 +       /* iput() and kfree() will be called in au_hnotify() */
15702 +       args = kmalloc(sizeof(*args) + len + 1, GFP_NOFS);
15703 +       if (unlikely(!args)) {
15704 +               AuErr1("no memory\n");
15705 +               iput(dir);
15706 +               goto out;
15707 +       }
15708 +       args->flags[AuHn_PARENT] = flags[AuHn_PARENT];
15709 +       args->flags[AuHn_CHILD] = flags[AuHn_CHILD];
15710 +       args->mask = mask;
15711 +       args->dir = dir;
15712 +       args->h_dir = igrab(h_dir);
15713 +       if (h_child_inode)
15714 +               h_child_inode = igrab(h_child_inode); /* can be NULL */
15715 +       args->h_child_inode = h_child_inode;
15716 +       args->h_child_nlen = len;
15717 +       if (len) {
15718 +               p = (void *)args;
15719 +               p += sizeof(*args);
15720 +               memcpy(p, h_child_name, len);
15721 +               p[len] = 0;
15722 +       }
15723 +
15724 +       /* NFS fires the event for silly-renamed one from kworker */
15725 +       f = 0;
15726 +       if (!dir->i_nlink
15727 +           || (au_test_nfs(h_dir->i_sb) && (mask & FS_DELETE)))
15728 +               f = AuWkq_NEST;
15729 +       err = au_wkq_nowait(au_hn_bh, args, dir->i_sb, f);
15730 +       if (unlikely(err)) {
15731 +               pr_err("wkq %d\n", err);
15732 +               iput(args->h_child_inode);
15733 +               iput(args->h_dir);
15734 +               iput(args->dir);
15735 +               au_delayed_kfree(args);
15736 +       }
15737 +
15738 +out:
15739 +       return err;
15740 +}
15741 +
15742 +/* ---------------------------------------------------------------------- */
15743 +
15744 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm)
15745 +{
15746 +       int err;
15747 +
15748 +       AuDebugOn(!(udba & AuOptMask_UDBA));
15749 +
15750 +       err = 0;
15751 +       if (au_hnotify_op.reset_br)
15752 +               err = au_hnotify_op.reset_br(udba, br, perm);
15753 +
15754 +       return err;
15755 +}
15756 +
15757 +int au_hnotify_init_br(struct au_branch *br, int perm)
15758 +{
15759 +       int err;
15760 +
15761 +       err = 0;
15762 +       if (au_hnotify_op.init_br)
15763 +               err = au_hnotify_op.init_br(br, perm);
15764 +
15765 +       return err;
15766 +}
15767 +
15768 +void au_hnotify_fin_br(struct au_branch *br)
15769 +{
15770 +       if (au_hnotify_op.fin_br)
15771 +               au_hnotify_op.fin_br(br);
15772 +}
15773 +
15774 +static void au_hn_destroy_cache(void)
15775 +{
15776 +       struct au_cache *cp;
15777 +
15778 +       flush_delayed_work(&au_dfree.dwork);
15779 +       cp = au_dfree.cache + AuCache_HNOTIFY;
15780 +       AuDebugOn(!llist_empty(&cp->llist));
15781 +       kmem_cache_destroy(cp->cache);
15782 +       cp->cache = NULL;
15783 +}
15784 +
15785 +AU_CACHE_DFREE_FUNC(hnotify, HNOTIFY, hn_lnode);
15786 +
15787 +int __init au_hnotify_init(void)
15788 +{
15789 +       int err;
15790 +       struct au_cache *cp;
15791 +
15792 +       err = -ENOMEM;
15793 +       cp = au_dfree.cache + AuCache_HNOTIFY;
15794 +       cp->cache = AuCache(au_hnotify);
15795 +       if (cp->cache) {
15796 +               err = 0;
15797 +               if (au_hnotify_op.init)
15798 +                       err = au_hnotify_op.init();
15799 +               if (unlikely(err))
15800 +                       au_hn_destroy_cache();
15801 +       }
15802 +       AuTraceErr(err);
15803 +       return err;
15804 +}
15805 +
15806 +void au_hnotify_fin(void)
15807 +{
15808 +       struct au_cache *cp;
15809 +
15810 +       if (au_hnotify_op.fin)
15811 +               au_hnotify_op.fin();
15812 +
15813 +       /* cf. au_cache_fin() */
15814 +       cp = au_dfree.cache + AuCache_HNOTIFY;
15815 +       if (cp->cache)
15816 +               au_hn_destroy_cache();
15817 +}
15818 diff -urN /usr/share/empty/fs/aufs/iinfo.c linux/fs/aufs/iinfo.c
15819 --- /usr/share/empty/fs/aufs/iinfo.c    1970-01-01 01:00:00.000000000 +0100
15820 +++ linux/fs/aufs/iinfo.c       2016-10-09 16:55:38.889431135 +0200
15821 @@ -0,0 +1,285 @@
15822 +/*
15823 + * Copyright (C) 2005-2016 Junjiro R. Okajima
15824 + *
15825 + * This program, aufs is free software; you can redistribute it and/or modify
15826 + * it under the terms of the GNU General Public License as published by
15827 + * the Free Software Foundation; either version 2 of the License, or
15828 + * (at your option) any later version.
15829 + *
15830 + * This program is distributed in the hope that it will be useful,
15831 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15832 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15833 + * GNU General Public License for more details.
15834 + *
15835 + * You should have received a copy of the GNU General Public License
15836 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15837 + */
15838 +
15839 +/*
15840 + * inode private data
15841 + */
15842 +
15843 +#include "aufs.h"
15844 +
15845 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex)
15846 +{
15847 +       struct inode *h_inode;
15848 +       struct au_hinode *hinode;
15849 +
15850 +       IiMustAnyLock(inode);
15851 +
15852 +       hinode = au_hinode(au_ii(inode), bindex);
15853 +       h_inode = hinode->hi_inode;
15854 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
15855 +       return h_inode;
15856 +}
15857 +
15858 +/* todo: hard/soft set? */
15859 +void au_hiput(struct au_hinode *hinode)
15860 +{
15861 +       au_hn_free(hinode);
15862 +       dput(hinode->hi_whdentry);
15863 +       iput(hinode->hi_inode);
15864 +}
15865 +
15866 +unsigned int au_hi_flags(struct inode *inode, int isdir)
15867 +{
15868 +       unsigned int flags;
15869 +       const unsigned int mnt_flags = au_mntflags(inode->i_sb);
15870 +
15871 +       flags = 0;
15872 +       if (au_opt_test(mnt_flags, XINO))
15873 +               au_fset_hi(flags, XINO);
15874 +       if (isdir && au_opt_test(mnt_flags, UDBA_HNOTIFY))
15875 +               au_fset_hi(flags, HNOTIFY);
15876 +       return flags;
15877 +}
15878 +
15879 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
15880 +                  struct inode *h_inode, unsigned int flags)
15881 +{
15882 +       struct au_hinode *hinode;
15883 +       struct inode *hi;
15884 +       struct au_iinfo *iinfo = au_ii(inode);
15885 +
15886 +       IiMustWriteLock(inode);
15887 +
15888 +       hinode = au_hinode(iinfo, bindex);
15889 +       hi = hinode->hi_inode;
15890 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
15891 +
15892 +       if (hi)
15893 +               au_hiput(hinode);
15894 +       hinode->hi_inode = h_inode;
15895 +       if (h_inode) {
15896 +               int err;
15897 +               struct super_block *sb = inode->i_sb;
15898 +               struct au_branch *br;
15899 +
15900 +               AuDebugOn(inode->i_mode
15901 +                         && (h_inode->i_mode & S_IFMT)
15902 +                         != (inode->i_mode & S_IFMT));
15903 +               if (bindex == iinfo->ii_btop)
15904 +                       au_cpup_igen(inode, h_inode);
15905 +               br = au_sbr(sb, bindex);
15906 +               hinode->hi_id = br->br_id;
15907 +               if (au_ftest_hi(flags, XINO)) {
15908 +                       err = au_xino_write(sb, bindex, h_inode->i_ino,
15909 +                                           inode->i_ino);
15910 +                       if (unlikely(err))
15911 +                               AuIOErr1("failed au_xino_write() %d\n", err);
15912 +               }
15913 +
15914 +               if (au_ftest_hi(flags, HNOTIFY)
15915 +                   && au_br_hnotifyable(br->br_perm)) {
15916 +                       err = au_hn_alloc(hinode, inode);
15917 +                       if (unlikely(err))
15918 +                               AuIOErr1("au_hn_alloc() %d\n", err);
15919 +               }
15920 +       }
15921 +}
15922 +
15923 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
15924 +                 struct dentry *h_wh)
15925 +{
15926 +       struct au_hinode *hinode;
15927 +
15928 +       IiMustWriteLock(inode);
15929 +
15930 +       hinode = au_hinode(au_ii(inode), bindex);
15931 +       AuDebugOn(hinode->hi_whdentry);
15932 +       hinode->hi_whdentry = h_wh;
15933 +}
15934 +
15935 +void au_update_iigen(struct inode *inode, int half)
15936 +{
15937 +       struct au_iinfo *iinfo;
15938 +       struct au_iigen *iigen;
15939 +       unsigned int sigen;
15940 +
15941 +       sigen = au_sigen(inode->i_sb);
15942 +       iinfo = au_ii(inode);
15943 +       iigen = &iinfo->ii_generation;
15944 +       spin_lock(&iigen->ig_spin);
15945 +       iigen->ig_generation = sigen;
15946 +       if (half)
15947 +               au_ig_fset(iigen->ig_flags, HALF_REFRESHED);
15948 +       else
15949 +               au_ig_fclr(iigen->ig_flags, HALF_REFRESHED);
15950 +       spin_unlock(&iigen->ig_spin);
15951 +}
15952 +
15953 +/* it may be called at remount time, too */
15954 +void au_update_ibrange(struct inode *inode, int do_put_zero)
15955 +{
15956 +       struct au_iinfo *iinfo;
15957 +       aufs_bindex_t bindex, bbot;
15958 +
15959 +       AuDebugOn(au_is_bad_inode(inode));
15960 +       IiMustWriteLock(inode);
15961 +
15962 +       iinfo = au_ii(inode);
15963 +       if (do_put_zero && iinfo->ii_btop >= 0) {
15964 +               for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot;
15965 +                    bindex++) {
15966 +                       struct inode *h_i;
15967 +
15968 +                       h_i = au_hinode(iinfo, bindex)->hi_inode;
15969 +                       if (h_i
15970 +                           && !h_i->i_nlink
15971 +                           && !(h_i->i_state & I_LINKABLE))
15972 +                               au_set_h_iptr(inode, bindex, NULL, 0);
15973 +               }
15974 +       }
15975 +
15976 +       iinfo->ii_btop = -1;
15977 +       iinfo->ii_bbot = -1;
15978 +       bbot = au_sbbot(inode->i_sb);
15979 +       for (bindex = 0; bindex <= bbot; bindex++)
15980 +               if (au_hinode(iinfo, bindex)->hi_inode) {
15981 +                       iinfo->ii_btop = bindex;
15982 +                       break;
15983 +               }
15984 +       if (iinfo->ii_btop >= 0)
15985 +               for (bindex = bbot; bindex >= iinfo->ii_btop; bindex--)
15986 +                       if (au_hinode(iinfo, bindex)->hi_inode) {
15987 +                               iinfo->ii_bbot = bindex;
15988 +                               break;
15989 +                       }
15990 +       AuDebugOn(iinfo->ii_btop > iinfo->ii_bbot);
15991 +}
15992 +
15993 +/* ---------------------------------------------------------------------- */
15994 +
15995 +void au_icntnr_init_once(void *_c)
15996 +{
15997 +       struct au_icntnr *c = _c;
15998 +       struct au_iinfo *iinfo = &c->iinfo;
15999 +
16000 +       spin_lock_init(&iinfo->ii_generation.ig_spin);
16001 +       au_rw_init(&iinfo->ii_rwsem);
16002 +       inode_init_once(&c->vfs_inode);
16003 +}
16004 +
16005 +void au_hinode_init(struct au_hinode *hinode)
16006 +{
16007 +       hinode->hi_inode = NULL;
16008 +       hinode->hi_id = -1;
16009 +       au_hn_init(hinode);
16010 +       hinode->hi_whdentry = NULL;
16011 +}
16012 +
16013 +int au_iinfo_init(struct inode *inode)
16014 +{
16015 +       struct au_iinfo *iinfo;
16016 +       struct super_block *sb;
16017 +       struct au_hinode *hi;
16018 +       int nbr, i;
16019 +
16020 +       sb = inode->i_sb;
16021 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
16022 +       nbr = au_sbbot(sb) + 1;
16023 +       if (unlikely(nbr <= 0))
16024 +               nbr = 1;
16025 +       hi = kmalloc_array(nbr, sizeof(*iinfo->ii_hinode), GFP_NOFS);
16026 +       if (hi) {
16027 +               au_ninodes_inc(sb);
16028 +
16029 +               iinfo->ii_hinode = hi;
16030 +               for (i = 0; i < nbr; i++, hi++)
16031 +                       au_hinode_init(hi);
16032 +
16033 +               iinfo->ii_generation.ig_generation = au_sigen(sb);
16034 +               iinfo->ii_btop = -1;
16035 +               iinfo->ii_bbot = -1;
16036 +               iinfo->ii_vdir = NULL;
16037 +               return 0;
16038 +       }
16039 +       return -ENOMEM;
16040 +}
16041 +
16042 +int au_hinode_realloc(struct au_iinfo *iinfo, int nbr, int may_shrink)
16043 +{
16044 +       int err, i;
16045 +       struct au_hinode *hip;
16046 +
16047 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
16048 +
16049 +       err = -ENOMEM;
16050 +       hip = au_krealloc(iinfo->ii_hinode, sizeof(*hip) * nbr, GFP_NOFS,
16051 +                         may_shrink);
16052 +       if (hip) {
16053 +               iinfo->ii_hinode = hip;
16054 +               i = iinfo->ii_bbot + 1;
16055 +               hip += i;
16056 +               for (; i < nbr; i++, hip++)
16057 +                       au_hinode_init(hip);
16058 +               err = 0;
16059 +       }
16060 +
16061 +       return err;
16062 +}
16063 +
16064 +void au_iinfo_fin(struct inode *inode)
16065 +{
16066 +       struct au_iinfo *iinfo;
16067 +       struct au_hinode *hi;
16068 +       struct super_block *sb;
16069 +       aufs_bindex_t bindex, bbot;
16070 +       const unsigned char unlinked = !inode->i_nlink;
16071 +
16072 +       AuDebugOn(au_is_bad_inode(inode));
16073 +
16074 +       sb = inode->i_sb;
16075 +       au_ninodes_dec(sb);
16076 +       if (si_pid_test(sb))
16077 +               au_xino_delete_inode(inode, unlinked);
16078 +       else {
16079 +               /*
16080 +                * it is safe to hide the dependency between sbinfo and
16081 +                * sb->s_umount.
16082 +                */
16083 +               lockdep_off();
16084 +               si_noflush_read_lock(sb);
16085 +               au_xino_delete_inode(inode, unlinked);
16086 +               si_read_unlock(sb);
16087 +               lockdep_on();
16088 +       }
16089 +
16090 +       iinfo = au_ii(inode);
16091 +       if (iinfo->ii_vdir)
16092 +               au_vdir_free(iinfo->ii_vdir, /*atonce*/0);
16093 +
16094 +       bindex = iinfo->ii_btop;
16095 +       if (bindex >= 0) {
16096 +               hi = au_hinode(iinfo, bindex);
16097 +               bbot = iinfo->ii_bbot;
16098 +               while (bindex++ <= bbot) {
16099 +                       if (hi->hi_inode)
16100 +                               au_hiput(hi);
16101 +                       hi++;
16102 +               }
16103 +       }
16104 +       au_delayed_kfree(iinfo->ii_hinode);
16105 +       AuRwDestroy(&iinfo->ii_rwsem);
16106 +}
16107 diff -urN /usr/share/empty/fs/aufs/inode.c linux/fs/aufs/inode.c
16108 --- /usr/share/empty/fs/aufs/inode.c    1970-01-01 01:00:00.000000000 +0100
16109 +++ linux/fs/aufs/inode.c       2016-10-09 16:55:38.889431135 +0200
16110 @@ -0,0 +1,519 @@
16111 +/*
16112 + * Copyright (C) 2005-2016 Junjiro R. Okajima
16113 + *
16114 + * This program, aufs is free software; you can redistribute it and/or modify
16115 + * it under the terms of the GNU General Public License as published by
16116 + * the Free Software Foundation; either version 2 of the License, or
16117 + * (at your option) any later version.
16118 + *
16119 + * This program is distributed in the hope that it will be useful,
16120 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16121 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16122 + * GNU General Public License for more details.
16123 + *
16124 + * You should have received a copy of the GNU General Public License
16125 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16126 + */
16127 +
16128 +/*
16129 + * inode functions
16130 + */
16131 +
16132 +#include "aufs.h"
16133 +
16134 +struct inode *au_igrab(struct inode *inode)
16135 +{
16136 +       if (inode) {
16137 +               AuDebugOn(!atomic_read(&inode->i_count));
16138 +               ihold(inode);
16139 +       }
16140 +       return inode;
16141 +}
16142 +
16143 +static void au_refresh_hinode_attr(struct inode *inode, int do_version)
16144 +{
16145 +       au_cpup_attr_all(inode, /*force*/0);
16146 +       au_update_iigen(inode, /*half*/1);
16147 +       if (do_version)
16148 +               inode->i_version++;
16149 +}
16150 +
16151 +static int au_ii_refresh(struct inode *inode, int *update)
16152 +{
16153 +       int err, e, nbr;
16154 +       umode_t type;
16155 +       aufs_bindex_t bindex, new_bindex;
16156 +       struct super_block *sb;
16157 +       struct au_iinfo *iinfo;
16158 +       struct au_hinode *p, *q, tmp;
16159 +
16160 +       AuDebugOn(au_is_bad_inode(inode));
16161 +       IiMustWriteLock(inode);
16162 +
16163 +       *update = 0;
16164 +       sb = inode->i_sb;
16165 +       nbr = au_sbbot(sb) + 1;
16166 +       type = inode->i_mode & S_IFMT;
16167 +       iinfo = au_ii(inode);
16168 +       err = au_hinode_realloc(iinfo, nbr, /*may_shrink*/0);
16169 +       if (unlikely(err))
16170 +               goto out;
16171 +
16172 +       AuDebugOn(iinfo->ii_btop < 0);
16173 +       p = au_hinode(iinfo, iinfo->ii_btop);
16174 +       for (bindex = iinfo->ii_btop; bindex <= iinfo->ii_bbot;
16175 +            bindex++, p++) {
16176 +               if (!p->hi_inode)
16177 +                       continue;
16178 +
16179 +               AuDebugOn(type != (p->hi_inode->i_mode & S_IFMT));
16180 +               new_bindex = au_br_index(sb, p->hi_id);
16181 +               if (new_bindex == bindex)
16182 +                       continue;
16183 +
16184 +               if (new_bindex < 0) {
16185 +                       *update = 1;
16186 +                       au_hiput(p);
16187 +                       p->hi_inode = NULL;
16188 +                       continue;
16189 +               }
16190 +
16191 +               if (new_bindex < iinfo->ii_btop)
16192 +                       iinfo->ii_btop = new_bindex;
16193 +               if (iinfo->ii_bbot < new_bindex)
16194 +                       iinfo->ii_bbot = new_bindex;
16195 +               /* swap two lower inode, and loop again */
16196 +               q = au_hinode(iinfo, new_bindex);
16197 +               tmp = *q;
16198 +               *q = *p;
16199 +               *p = tmp;
16200 +               if (tmp.hi_inode) {
16201 +                       bindex--;
16202 +                       p--;
16203 +               }
16204 +       }
16205 +       au_update_ibrange(inode, /*do_put_zero*/0);
16206 +       au_hinode_realloc(iinfo, nbr, /*may_shrink*/1); /* harmless if err */
16207 +       e = au_dy_irefresh(inode);
16208 +       if (unlikely(e && !err))
16209 +               err = e;
16210 +
16211 +out:
16212 +       AuTraceErr(err);
16213 +       return err;
16214 +}
16215 +
16216 +void au_refresh_iop(struct inode *inode, int force_getattr)
16217 +{
16218 +       int type;
16219 +       struct au_sbinfo *sbi = au_sbi(inode->i_sb);
16220 +       const struct inode_operations *iop
16221 +               = force_getattr ? aufs_iop : sbi->si_iop_array;
16222 +
16223 +       if (inode->i_op == iop)
16224 +               return;
16225 +
16226 +       switch (inode->i_mode & S_IFMT) {
16227 +       case S_IFDIR:
16228 +               type = AuIop_DIR;
16229 +               break;
16230 +       case S_IFLNK:
16231 +               type = AuIop_SYMLINK;
16232 +               break;
16233 +       default:
16234 +               type = AuIop_OTHER;
16235 +               break;
16236 +       }
16237 +
16238 +       inode->i_op = iop + type;
16239 +       /* unnecessary smp_wmb() */
16240 +}
16241 +
16242 +int au_refresh_hinode_self(struct inode *inode)
16243 +{
16244 +       int err, update;
16245 +
16246 +       err = au_ii_refresh(inode, &update);
16247 +       if (!err)
16248 +               au_refresh_hinode_attr(inode, update && S_ISDIR(inode->i_mode));
16249 +
16250 +       AuTraceErr(err);
16251 +       return err;
16252 +}
16253 +
16254 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry)
16255 +{
16256 +       int err, e, update;
16257 +       unsigned int flags;
16258 +       umode_t mode;
16259 +       aufs_bindex_t bindex, bbot;
16260 +       unsigned char isdir;
16261 +       struct au_hinode *p;
16262 +       struct au_iinfo *iinfo;
16263 +
16264 +       err = au_ii_refresh(inode, &update);
16265 +       if (unlikely(err))
16266 +               goto out;
16267 +
16268 +       update = 0;
16269 +       iinfo = au_ii(inode);
16270 +       p = au_hinode(iinfo, iinfo->ii_btop);
16271 +       mode = (inode->i_mode & S_IFMT);
16272 +       isdir = S_ISDIR(mode);
16273 +       flags = au_hi_flags(inode, isdir);
16274 +       bbot = au_dbbot(dentry);
16275 +       for (bindex = au_dbtop(dentry); bindex <= bbot; bindex++) {
16276 +               struct inode *h_i, *h_inode;
16277 +               struct dentry *h_d;
16278 +
16279 +               h_d = au_h_dptr(dentry, bindex);
16280 +               if (!h_d || d_is_negative(h_d))
16281 +                       continue;
16282 +
16283 +               h_inode = d_inode(h_d);
16284 +               AuDebugOn(mode != (h_inode->i_mode & S_IFMT));
16285 +               if (iinfo->ii_btop <= bindex && bindex <= iinfo->ii_bbot) {
16286 +                       h_i = au_h_iptr(inode, bindex);
16287 +                       if (h_i) {
16288 +                               if (h_i == h_inode)
16289 +                                       continue;
16290 +                               err = -EIO;
16291 +                               break;
16292 +                       }
16293 +               }
16294 +               if (bindex < iinfo->ii_btop)
16295 +                       iinfo->ii_btop = bindex;
16296 +               if (iinfo->ii_bbot < bindex)
16297 +                       iinfo->ii_bbot = bindex;
16298 +               au_set_h_iptr(inode, bindex, au_igrab(h_inode), flags);
16299 +               update = 1;
16300 +       }
16301 +       au_update_ibrange(inode, /*do_put_zero*/0);
16302 +       e = au_dy_irefresh(inode);
16303 +       if (unlikely(e && !err))
16304 +               err = e;
16305 +       if (!err)
16306 +               au_refresh_hinode_attr(inode, update && isdir);
16307 +
16308 +out:
16309 +       AuTraceErr(err);
16310 +       return err;
16311 +}
16312 +
16313 +static int set_inode(struct inode *inode, struct dentry *dentry)
16314 +{
16315 +       int err;
16316 +       unsigned int flags;
16317 +       umode_t mode;
16318 +       aufs_bindex_t bindex, btop, btail;
16319 +       unsigned char isdir;
16320 +       struct dentry *h_dentry;
16321 +       struct inode *h_inode;
16322 +       struct au_iinfo *iinfo;
16323 +       struct inode_operations *iop;
16324 +
16325 +       IiMustWriteLock(inode);
16326 +
16327 +       err = 0;
16328 +       isdir = 0;
16329 +       iop = au_sbi(inode->i_sb)->si_iop_array;
16330 +       btop = au_dbtop(dentry);
16331 +       h_dentry = au_h_dptr(dentry, btop);
16332 +       h_inode = d_inode(h_dentry);
16333 +       mode = h_inode->i_mode;
16334 +       switch (mode & S_IFMT) {
16335 +       case S_IFREG:
16336 +               btail = au_dbtail(dentry);
16337 +               inode->i_op = iop + AuIop_OTHER;
16338 +               inode->i_fop = &aufs_file_fop;
16339 +               err = au_dy_iaop(inode, btop, h_inode);
16340 +               if (unlikely(err))
16341 +                       goto out;
16342 +               break;
16343 +       case S_IFDIR:
16344 +               isdir = 1;
16345 +               btail = au_dbtaildir(dentry);
16346 +               inode->i_op = iop + AuIop_DIR;
16347 +               inode->i_fop = &aufs_dir_fop;
16348 +               break;
16349 +       case S_IFLNK:
16350 +               btail = au_dbtail(dentry);
16351 +               inode->i_op = iop + AuIop_SYMLINK;
16352 +               break;
16353 +       case S_IFBLK:
16354 +       case S_IFCHR:
16355 +       case S_IFIFO:
16356 +       case S_IFSOCK:
16357 +               btail = au_dbtail(dentry);
16358 +               inode->i_op = iop + AuIop_OTHER;
16359 +               init_special_inode(inode, mode, h_inode->i_rdev);
16360 +               break;
16361 +       default:
16362 +               AuIOErr("Unknown file type 0%o\n", mode);
16363 +               err = -EIO;
16364 +               goto out;
16365 +       }
16366 +
16367 +       /* do not set hnotify for whiteouted dirs (SHWH mode) */
16368 +       flags = au_hi_flags(inode, isdir);
16369 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH)
16370 +           && au_ftest_hi(flags, HNOTIFY)
16371 +           && dentry->d_name.len > AUFS_WH_PFX_LEN
16372 +           && !memcmp(dentry->d_name.name, AUFS_WH_PFX, AUFS_WH_PFX_LEN))
16373 +               au_fclr_hi(flags, HNOTIFY);
16374 +       iinfo = au_ii(inode);
16375 +       iinfo->ii_btop = btop;
16376 +       iinfo->ii_bbot = btail;
16377 +       for (bindex = btop; bindex <= btail; bindex++) {
16378 +               h_dentry = au_h_dptr(dentry, bindex);
16379 +               if (h_dentry)
16380 +                       au_set_h_iptr(inode, bindex,
16381 +                                     au_igrab(d_inode(h_dentry)), flags);
16382 +       }
16383 +       au_cpup_attr_all(inode, /*force*/1);
16384 +       /*
16385 +        * to force calling aufs_get_acl() every time,
16386 +        * do not call cache_no_acl() for aufs inode.
16387 +        */
16388 +
16389 +out:
16390 +       return err;
16391 +}
16392 +
16393 +/*
16394 + * successful returns with iinfo write_locked
16395 + * minus: errno
16396 + * zero: success, matched
16397 + * plus: no error, but unmatched
16398 + */
16399 +static int reval_inode(struct inode *inode, struct dentry *dentry)
16400 +{
16401 +       int err;
16402 +       unsigned int gen, igflags;
16403 +       aufs_bindex_t bindex, bbot;
16404 +       struct inode *h_inode, *h_dinode;
16405 +       struct dentry *h_dentry;
16406 +
16407 +       /*
16408 +        * before this function, if aufs got any iinfo lock, it must be only
16409 +        * one, the parent dir.
16410 +        * it can happen by UDBA and the obsoleted inode number.
16411 +        */
16412 +       err = -EIO;
16413 +       if (unlikely(inode->i_ino == parent_ino(dentry)))
16414 +               goto out;
16415 +
16416 +       err = 1;
16417 +       ii_write_lock_new_child(inode);
16418 +       h_dentry = au_h_dptr(dentry, au_dbtop(dentry));
16419 +       h_dinode = d_inode(h_dentry);
16420 +       bbot = au_ibbot(inode);
16421 +       for (bindex = au_ibtop(inode); bindex <= bbot; bindex++) {
16422 +               h_inode = au_h_iptr(inode, bindex);
16423 +               if (!h_inode || h_inode != h_dinode)
16424 +                       continue;
16425 +
16426 +               err = 0;
16427 +               gen = au_iigen(inode, &igflags);
16428 +               if (gen == au_digen(dentry)
16429 +                   && !au_ig_ftest(igflags, HALF_REFRESHED))
16430 +                       break;
16431 +
16432 +               /* fully refresh inode using dentry */
16433 +               err = au_refresh_hinode(inode, dentry);
16434 +               if (!err)
16435 +                       au_update_iigen(inode, /*half*/0);
16436 +               break;
16437 +       }
16438 +
16439 +       if (unlikely(err))
16440 +               ii_write_unlock(inode);
16441 +out:
16442 +       return err;
16443 +}
16444 +
16445 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16446 +          unsigned int d_type, ino_t *ino)
16447 +{
16448 +       int err;
16449 +       struct mutex *mtx;
16450 +
16451 +       /* prevent hardlinked inode number from race condition */
16452 +       mtx = NULL;
16453 +       if (d_type != DT_DIR) {
16454 +               mtx = &au_sbr(sb, bindex)->br_xino.xi_nondir_mtx;
16455 +               mutex_lock(mtx);
16456 +       }
16457 +       err = au_xino_read(sb, bindex, h_ino, ino);
16458 +       if (unlikely(err))
16459 +               goto out;
16460 +
16461 +       if (!*ino) {
16462 +               err = -EIO;
16463 +               *ino = au_xino_new_ino(sb);
16464 +               if (unlikely(!*ino))
16465 +                       goto out;
16466 +               err = au_xino_write(sb, bindex, h_ino, *ino);
16467 +               if (unlikely(err))
16468 +                       goto out;
16469 +       }
16470 +
16471 +out:
16472 +       if (mtx)
16473 +               mutex_unlock(mtx);
16474 +       return err;
16475 +}
16476 +
16477 +/* successful returns with iinfo write_locked */
16478 +/* todo: return with unlocked? */
16479 +struct inode *au_new_inode(struct dentry *dentry, int must_new)
16480 +{
16481 +       struct inode *inode, *h_inode;
16482 +       struct dentry *h_dentry;
16483 +       struct super_block *sb;
16484 +       struct mutex *mtx;
16485 +       ino_t h_ino, ino;
16486 +       int err;
16487 +       aufs_bindex_t btop;
16488 +
16489 +       sb = dentry->d_sb;
16490 +       btop = au_dbtop(dentry);
16491 +       h_dentry = au_h_dptr(dentry, btop);
16492 +       h_inode = d_inode(h_dentry);
16493 +       h_ino = h_inode->i_ino;
16494 +
16495 +       /*
16496 +        * stop 'race'-ing between hardlinks under different
16497 +        * parents.
16498 +        */
16499 +       mtx = NULL;
16500 +       if (!d_is_dir(h_dentry))
16501 +               mtx = &au_sbr(sb, btop)->br_xino.xi_nondir_mtx;
16502 +
16503 +new_ino:
16504 +       if (mtx)
16505 +               mutex_lock(mtx);
16506 +       err = au_xino_read(sb, btop, h_ino, &ino);
16507 +       inode = ERR_PTR(err);
16508 +       if (unlikely(err))
16509 +               goto out;
16510 +
16511 +       if (!ino) {
16512 +               ino = au_xino_new_ino(sb);
16513 +               if (unlikely(!ino)) {
16514 +                       inode = ERR_PTR(-EIO);
16515 +                       goto out;
16516 +               }
16517 +       }
16518 +
16519 +       AuDbg("i%lu\n", (unsigned long)ino);
16520 +       inode = au_iget_locked(sb, ino);
16521 +       err = PTR_ERR(inode);
16522 +       if (IS_ERR(inode))
16523 +               goto out;
16524 +
16525 +       AuDbg("%lx, new %d\n", inode->i_state, !!(inode->i_state & I_NEW));
16526 +       if (inode->i_state & I_NEW) {
16527 +               ii_write_lock_new_child(inode);
16528 +               err = set_inode(inode, dentry);
16529 +               if (!err) {
16530 +                       unlock_new_inode(inode);
16531 +                       goto out; /* success */
16532 +               }
16533 +
16534 +               /*
16535 +                * iget_failed() calls iput(), but we need to call
16536 +                * ii_write_unlock() after iget_failed(). so dirty hack for
16537 +                * i_count.
16538 +                */
16539 +               atomic_inc(&inode->i_count);
16540 +               iget_failed(inode);
16541 +               ii_write_unlock(inode);
16542 +               au_xino_write(sb, btop, h_ino, /*ino*/0);
16543 +               /* ignore this error */
16544 +               goto out_iput;
16545 +       } else if (!must_new && !IS_DEADDIR(inode) && inode->i_nlink) {
16546 +               /*
16547 +                * horrible race condition between lookup, readdir and copyup
16548 +                * (or something).
16549 +                */
16550 +               if (mtx)
16551 +                       mutex_unlock(mtx);
16552 +               err = reval_inode(inode, dentry);
16553 +               if (unlikely(err < 0)) {
16554 +                       mtx = NULL;
16555 +                       goto out_iput;
16556 +               }
16557 +
16558 +               if (!err) {
16559 +                       mtx = NULL;
16560 +                       goto out; /* success */
16561 +               } else if (mtx)
16562 +                       mutex_lock(mtx);
16563 +       }
16564 +
16565 +       if (unlikely(au_test_fs_unique_ino(h_inode)))
16566 +               AuWarn1("Warning: Un-notified UDBA or repeatedly renamed dir,"
16567 +                       " b%d, %s, %pd, hi%lu, i%lu.\n",
16568 +                       btop, au_sbtype(h_dentry->d_sb), dentry,
16569 +                       (unsigned long)h_ino, (unsigned long)ino);
16570 +       ino = 0;
16571 +       err = au_xino_write(sb, btop, h_ino, /*ino*/0);
16572 +       if (!err) {
16573 +               iput(inode);
16574 +               if (mtx)
16575 +                       mutex_unlock(mtx);
16576 +               goto new_ino;
16577 +       }
16578 +
16579 +out_iput:
16580 +       iput(inode);
16581 +       inode = ERR_PTR(err);
16582 +out:
16583 +       if (mtx)
16584 +               mutex_unlock(mtx);
16585 +       return inode;
16586 +}
16587 +
16588 +/* ---------------------------------------------------------------------- */
16589 +
16590 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16591 +              struct inode *inode)
16592 +{
16593 +       int err;
16594 +       struct inode *hi;
16595 +
16596 +       err = au_br_rdonly(au_sbr(sb, bindex));
16597 +
16598 +       /* pseudo-link after flushed may happen out of bounds */
16599 +       if (!err
16600 +           && inode
16601 +           && au_ibtop(inode) <= bindex
16602 +           && bindex <= au_ibbot(inode)) {
16603 +               /*
16604 +                * permission check is unnecessary since vfsub routine
16605 +                * will be called later
16606 +                */
16607 +               hi = au_h_iptr(inode, bindex);
16608 +               if (hi)
16609 +                       err = IS_IMMUTABLE(hi) ? -EROFS : 0;
16610 +       }
16611 +
16612 +       return err;
16613 +}
16614 +
16615 +int au_test_h_perm(struct inode *h_inode, int mask)
16616 +{
16617 +       if (uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
16618 +               return 0;
16619 +       return inode_permission(h_inode, mask);
16620 +}
16621 +
16622 +int au_test_h_perm_sio(struct inode *h_inode, int mask)
16623 +{
16624 +       if (au_test_nfs(h_inode->i_sb)
16625 +           && (mask & MAY_WRITE)
16626 +           && S_ISDIR(h_inode->i_mode))
16627 +               mask |= MAY_READ; /* force permission check */
16628 +       return au_test_h_perm(h_inode, mask);
16629 +}
16630 diff -urN /usr/share/empty/fs/aufs/inode.h linux/fs/aufs/inode.h
16631 --- /usr/share/empty/fs/aufs/inode.h    1970-01-01 01:00:00.000000000 +0100
16632 +++ linux/fs/aufs/inode.h       2016-10-09 16:55:38.889431135 +0200
16633 @@ -0,0 +1,700 @@
16634 +/*
16635 + * Copyright (C) 2005-2016 Junjiro R. Okajima
16636 + *
16637 + * This program, aufs is free software; you can redistribute it and/or modify
16638 + * it under the terms of the GNU General Public License as published by
16639 + * the Free Software Foundation; either version 2 of the License, or
16640 + * (at your option) any later version.
16641 + *
16642 + * This program is distributed in the hope that it will be useful,
16643 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16644 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16645 + * GNU General Public License for more details.
16646 + *
16647 + * You should have received a copy of the GNU General Public License
16648 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16649 + */
16650 +
16651 +/*
16652 + * inode operations
16653 + */
16654 +
16655 +#ifndef __AUFS_INODE_H__
16656 +#define __AUFS_INODE_H__
16657 +
16658 +#ifdef __KERNEL__
16659 +
16660 +#include <linux/fsnotify.h>
16661 +#include "rwsem.h"
16662 +
16663 +struct vfsmount;
16664 +
16665 +struct au_hnotify {
16666 +#ifdef CONFIG_AUFS_HNOTIFY
16667 +#ifdef CONFIG_AUFS_HFSNOTIFY
16668 +       /* never use fsnotify_add_vfsmount_mark() */
16669 +       struct fsnotify_mark            hn_mark;
16670 +#endif
16671 +       union {
16672 +               struct inode            *hn_aufs_inode; /* no get/put */
16673 +               struct llist_node       hn_lnode;       /* delayed free */
16674 +       };
16675 +#endif
16676 +} ____cacheline_aligned_in_smp;
16677 +
16678 +struct au_hinode {
16679 +       struct inode            *hi_inode;
16680 +       aufs_bindex_t           hi_id;
16681 +#ifdef CONFIG_AUFS_HNOTIFY
16682 +       struct au_hnotify       *hi_notify;
16683 +#endif
16684 +
16685 +       /* reference to the copied-up whiteout with get/put */
16686 +       struct dentry           *hi_whdentry;
16687 +};
16688 +
16689 +/* ig_flags */
16690 +#define AuIG_HALF_REFRESHED            1
16691 +#define au_ig_ftest(flags, name)       ((flags) & AuIG_##name)
16692 +#define au_ig_fset(flags, name) \
16693 +       do { (flags) |= AuIG_##name; } while (0)
16694 +#define au_ig_fclr(flags, name) \
16695 +       do { (flags) &= ~AuIG_##name; } while (0)
16696 +
16697 +struct au_iigen {
16698 +       spinlock_t      ig_spin;
16699 +       __u32           ig_generation, ig_flags;
16700 +};
16701 +
16702 +struct au_vdir;
16703 +struct au_iinfo {
16704 +       struct au_iigen         ii_generation;
16705 +       struct super_block      *ii_hsb1;       /* no get/put */
16706 +
16707 +       struct au_rwsem         ii_rwsem;
16708 +       aufs_bindex_t           ii_btop, ii_bbot;
16709 +       __u32                   ii_higen;
16710 +       struct au_hinode        *ii_hinode;
16711 +       struct au_vdir          *ii_vdir;
16712 +};
16713 +
16714 +struct au_icntnr {
16715 +       struct au_iinfo iinfo;
16716 +       struct inode vfs_inode;
16717 +       union {
16718 +               struct hlist_node       plink;
16719 +               struct llist_node       lnode;  /* delayed free */
16720 +       };
16721 +} ____cacheline_aligned_in_smp;
16722 +
16723 +/* au_pin flags */
16724 +#define AuPin_DI_LOCKED                1
16725 +#define AuPin_MNT_WRITE                (1 << 1)
16726 +#define au_ftest_pin(flags, name)      ((flags) & AuPin_##name)
16727 +#define au_fset_pin(flags, name) \
16728 +       do { (flags) |= AuPin_##name; } while (0)
16729 +#define au_fclr_pin(flags, name) \
16730 +       do { (flags) &= ~AuPin_##name; } while (0)
16731 +
16732 +struct au_pin {
16733 +       /* input */
16734 +       struct dentry *dentry;
16735 +       unsigned int udba;
16736 +       unsigned char lsc_di, lsc_hi, flags;
16737 +       aufs_bindex_t bindex;
16738 +
16739 +       /* output */
16740 +       struct dentry *parent;
16741 +       struct au_hinode *hdir;
16742 +       struct vfsmount *h_mnt;
16743 +
16744 +       /* temporary unlock/relock for copyup */
16745 +       struct dentry *h_dentry, *h_parent;
16746 +       struct au_branch *br;
16747 +       struct task_struct *task;
16748 +};
16749 +
16750 +void au_pin_hdir_unlock(struct au_pin *p);
16751 +int au_pin_hdir_lock(struct au_pin *p);
16752 +int au_pin_hdir_relock(struct au_pin *p);
16753 +void au_pin_hdir_acquire_nest(struct au_pin *p);
16754 +void au_pin_hdir_release(struct au_pin *p);
16755 +
16756 +/* ---------------------------------------------------------------------- */
16757 +
16758 +static inline struct au_iinfo *au_ii(struct inode *inode)
16759 +{
16760 +       BUG_ON(is_bad_inode(inode));
16761 +       return &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
16762 +}
16763 +
16764 +/* ---------------------------------------------------------------------- */
16765 +
16766 +/* inode.c */
16767 +struct inode *au_igrab(struct inode *inode);
16768 +void au_refresh_iop(struct inode *inode, int force_getattr);
16769 +int au_refresh_hinode_self(struct inode *inode);
16770 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry);
16771 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16772 +          unsigned int d_type, ino_t *ino);
16773 +struct inode *au_new_inode(struct dentry *dentry, int must_new);
16774 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16775 +              struct inode *inode);
16776 +int au_test_h_perm(struct inode *h_inode, int mask);
16777 +int au_test_h_perm_sio(struct inode *h_inode, int mask);
16778 +
16779 +static inline int au_wh_ino(struct super_block *sb, aufs_bindex_t bindex,
16780 +                           ino_t h_ino, unsigned int d_type, ino_t *ino)
16781 +{
16782 +#ifdef CONFIG_AUFS_SHWH
16783 +       return au_ino(sb, bindex, h_ino, d_type, ino);
16784 +#else
16785 +       return 0;
16786 +#endif
16787 +}
16788 +
16789 +/* i_op.c */
16790 +enum {
16791 +       AuIop_SYMLINK,
16792 +       AuIop_DIR,
16793 +       AuIop_OTHER,
16794 +       AuIop_Last
16795 +};
16796 +extern struct inode_operations aufs_iop[AuIop_Last],
16797 +       aufs_iop_nogetattr[AuIop_Last];
16798 +
16799 +/* au_wr_dir flags */
16800 +#define AuWrDir_ADD_ENTRY      1
16801 +#define AuWrDir_ISDIR          (1 << 1)
16802 +#define AuWrDir_TMPFILE                (1 << 2)
16803 +#define au_ftest_wrdir(flags, name)    ((flags) & AuWrDir_##name)
16804 +#define au_fset_wrdir(flags, name) \
16805 +       do { (flags) |= AuWrDir_##name; } while (0)
16806 +#define au_fclr_wrdir(flags, name) \
16807 +       do { (flags) &= ~AuWrDir_##name; } while (0)
16808 +
16809 +struct au_wr_dir_args {
16810 +       aufs_bindex_t force_btgt;
16811 +       unsigned char flags;
16812 +};
16813 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
16814 +             struct au_wr_dir_args *args);
16815 +
16816 +struct dentry *au_pinned_h_parent(struct au_pin *pin);
16817 +void au_pin_init(struct au_pin *pin, struct dentry *dentry,
16818 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
16819 +                unsigned int udba, unsigned char flags);
16820 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
16821 +          unsigned int udba, unsigned char flags) __must_check;
16822 +int au_do_pin(struct au_pin *pin) __must_check;
16823 +void au_unpin(struct au_pin *pin);
16824 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen);
16825 +
16826 +#define AuIcpup_DID_CPUP       1
16827 +#define au_ftest_icpup(flags, name)    ((flags) & AuIcpup_##name)
16828 +#define au_fset_icpup(flags, name) \
16829 +       do { (flags) |= AuIcpup_##name; } while (0)
16830 +#define au_fclr_icpup(flags, name) \
16831 +       do { (flags) &= ~AuIcpup_##name; } while (0)
16832 +
16833 +struct au_icpup_args {
16834 +       unsigned char flags;
16835 +       unsigned char pin_flags;
16836 +       aufs_bindex_t btgt;
16837 +       unsigned int udba;
16838 +       struct au_pin pin;
16839 +       struct path h_path;
16840 +       struct inode *h_inode;
16841 +};
16842 +
16843 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
16844 +                    struct au_icpup_args *a);
16845 +
16846 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path);
16847 +
16848 +/* i_op_add.c */
16849 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
16850 +              struct dentry *h_parent, int isdir);
16851 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
16852 +              dev_t dev);
16853 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname);
16854 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
16855 +               bool want_excl);
16856 +struct vfsub_aopen_args;
16857 +int au_aopen_or_create(struct inode *dir, struct dentry *dentry,
16858 +                      struct vfsub_aopen_args *args);
16859 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode);
16860 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
16861 +             struct dentry *dentry);
16862 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
16863 +
16864 +/* i_op_del.c */
16865 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup);
16866 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
16867 +              struct dentry *h_parent, int isdir);
16868 +int aufs_unlink(struct inode *dir, struct dentry *dentry);
16869 +int aufs_rmdir(struct inode *dir, struct dentry *dentry);
16870 +
16871 +/* i_op_ren.c */
16872 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt);
16873 +int aufs_rename(struct inode *src_dir, struct dentry *src_dentry,
16874 +               struct inode *dir, struct dentry *dentry);
16875 +
16876 +/* iinfo.c */
16877 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex);
16878 +void au_hiput(struct au_hinode *hinode);
16879 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
16880 +                 struct dentry *h_wh);
16881 +unsigned int au_hi_flags(struct inode *inode, int isdir);
16882 +
16883 +/* hinode flags */
16884 +#define AuHi_XINO      1
16885 +#define AuHi_HNOTIFY   (1 << 1)
16886 +#define au_ftest_hi(flags, name)       ((flags) & AuHi_##name)
16887 +#define au_fset_hi(flags, name) \
16888 +       do { (flags) |= AuHi_##name; } while (0)
16889 +#define au_fclr_hi(flags, name) \
16890 +       do { (flags) &= ~AuHi_##name; } while (0)
16891 +
16892 +#ifndef CONFIG_AUFS_HNOTIFY
16893 +#undef AuHi_HNOTIFY
16894 +#define AuHi_HNOTIFY   0
16895 +#endif
16896 +
16897 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
16898 +                  struct inode *h_inode, unsigned int flags);
16899 +
16900 +void au_update_iigen(struct inode *inode, int half);
16901 +void au_update_ibrange(struct inode *inode, int do_put_zero);
16902 +
16903 +void au_icntnr_init_once(void *_c);
16904 +void au_hinode_init(struct au_hinode *hinode);
16905 +int au_iinfo_init(struct inode *inode);
16906 +void au_iinfo_fin(struct inode *inode);
16907 +int au_hinode_realloc(struct au_iinfo *iinfo, int nbr, int may_shrink);
16908 +
16909 +#ifdef CONFIG_PROC_FS
16910 +/* plink.c */
16911 +int au_plink_maint(struct super_block *sb, int flags);
16912 +struct au_sbinfo;
16913 +void au_plink_maint_leave(struct au_sbinfo *sbinfo);
16914 +int au_plink_maint_enter(struct super_block *sb);
16915 +#ifdef CONFIG_AUFS_DEBUG
16916 +void au_plink_list(struct super_block *sb);
16917 +#else
16918 +AuStubVoid(au_plink_list, struct super_block *sb)
16919 +#endif
16920 +int au_plink_test(struct inode *inode);
16921 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex);
16922 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
16923 +                    struct dentry *h_dentry);
16924 +void au_plink_put(struct super_block *sb, int verbose);
16925 +void au_plink_clean(struct super_block *sb, int verbose);
16926 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id);
16927 +#else
16928 +AuStubInt0(au_plink_maint, struct super_block *sb, int flags);
16929 +AuStubVoid(au_plink_maint_leave, struct au_sbinfo *sbinfo);
16930 +AuStubInt0(au_plink_maint_enter, struct super_block *sb);
16931 +AuStubVoid(au_plink_list, struct super_block *sb);
16932 +AuStubInt0(au_plink_test, struct inode *inode);
16933 +AuStub(struct dentry *, au_plink_lkup, return NULL,
16934 +       struct inode *inode, aufs_bindex_t bindex);
16935 +AuStubVoid(au_plink_append, struct inode *inode, aufs_bindex_t bindex,
16936 +          struct dentry *h_dentry);
16937 +AuStubVoid(au_plink_put, struct super_block *sb, int verbose);
16938 +AuStubVoid(au_plink_clean, struct super_block *sb, int verbose);
16939 +AuStubVoid(au_plink_half_refresh, struct super_block *sb, aufs_bindex_t br_id);
16940 +#endif /* CONFIG_PROC_FS */
16941 +
16942 +#ifdef CONFIG_AUFS_XATTR
16943 +/* xattr.c */
16944 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags,
16945 +                 unsigned int verbose);
16946 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size);
16947 +ssize_t aufs_getxattr(struct dentry *dentry, struct inode *inode,
16948 +                     const char *name, void *value, size_t size);
16949 +int aufs_setxattr(struct dentry *dentry, struct inode *inode, const char *name,
16950 +                 const void *value, size_t size, int flags);
16951 +int aufs_removexattr(struct dentry *dentry, const char *name);
16952 +
16953 +/* void au_xattr_init(struct super_block *sb); */
16954 +#else
16955 +AuStubInt0(au_cpup_xattr, struct dentry *h_dst, struct dentry *h_src,
16956 +          int ignore_flags, unsigned int verbose);
16957 +/* AuStubVoid(au_xattr_init, struct super_block *sb); */
16958 +#endif
16959 +
16960 +#ifdef CONFIG_FS_POSIX_ACL
16961 +struct posix_acl *aufs_get_acl(struct inode *inode, int type);
16962 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
16963 +#endif
16964 +
16965 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
16966 +enum {
16967 +       AU_XATTR_SET,
16968 +       AU_XATTR_REMOVE,
16969 +       AU_ACL_SET
16970 +};
16971 +
16972 +struct au_srxattr {
16973 +       int type;
16974 +       union {
16975 +               struct {
16976 +                       const char      *name;
16977 +                       const void      *value;
16978 +                       size_t          size;
16979 +                       int             flags;
16980 +               } set;
16981 +               struct {
16982 +                       const char      *name;
16983 +               } remove;
16984 +               struct {
16985 +                       struct posix_acl *acl;
16986 +                       int             type;
16987 +               } acl_set;
16988 +       } u;
16989 +};
16990 +ssize_t au_srxattr(struct dentry *dentry, struct inode *inode,
16991 +                  struct au_srxattr *arg);
16992 +#endif
16993 +
16994 +/* ---------------------------------------------------------------------- */
16995 +
16996 +/* lock subclass for iinfo */
16997 +enum {
16998 +       AuLsc_II_CHILD,         /* child first */
16999 +       AuLsc_II_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
17000 +       AuLsc_II_CHILD3,        /* copyup dirs */
17001 +       AuLsc_II_PARENT,        /* see AuLsc_I_PARENT in vfsub.h */
17002 +       AuLsc_II_PARENT2,
17003 +       AuLsc_II_PARENT3,       /* copyup dirs */
17004 +       AuLsc_II_NEW_CHILD
17005 +};
17006 +
17007 +/*
17008 + * ii_read_lock_child, ii_write_lock_child,
17009 + * ii_read_lock_child2, ii_write_lock_child2,
17010 + * ii_read_lock_child3, ii_write_lock_child3,
17011 + * ii_read_lock_parent, ii_write_lock_parent,
17012 + * ii_read_lock_parent2, ii_write_lock_parent2,
17013 + * ii_read_lock_parent3, ii_write_lock_parent3,
17014 + * ii_read_lock_new_child, ii_write_lock_new_child,
17015 + */
17016 +#define AuReadLockFunc(name, lsc) \
17017 +static inline void ii_read_lock_##name(struct inode *i) \
17018 +{ \
17019 +       au_rw_read_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
17020 +}
17021 +
17022 +#define AuWriteLockFunc(name, lsc) \
17023 +static inline void ii_write_lock_##name(struct inode *i) \
17024 +{ \
17025 +       au_rw_write_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
17026 +}
17027 +
17028 +#define AuRWLockFuncs(name, lsc) \
17029 +       AuReadLockFunc(name, lsc) \
17030 +       AuWriteLockFunc(name, lsc)
17031 +
17032 +AuRWLockFuncs(child, CHILD);
17033 +AuRWLockFuncs(child2, CHILD2);
17034 +AuRWLockFuncs(child3, CHILD3);
17035 +AuRWLockFuncs(parent, PARENT);
17036 +AuRWLockFuncs(parent2, PARENT2);
17037 +AuRWLockFuncs(parent3, PARENT3);
17038 +AuRWLockFuncs(new_child, NEW_CHILD);
17039 +
17040 +#undef AuReadLockFunc
17041 +#undef AuWriteLockFunc
17042 +#undef AuRWLockFuncs
17043 +
17044 +/*
17045 + * ii_read_unlock, ii_write_unlock, ii_downgrade_lock
17046 + */
17047 +AuSimpleUnlockRwsemFuncs(ii, struct inode *i, &au_ii(i)->ii_rwsem);
17048 +
17049 +#define IiMustNoWaiters(i)     AuRwMustNoWaiters(&au_ii(i)->ii_rwsem)
17050 +#define IiMustAnyLock(i)       AuRwMustAnyLock(&au_ii(i)->ii_rwsem)
17051 +#define IiMustWriteLock(i)     AuRwMustWriteLock(&au_ii(i)->ii_rwsem)
17052 +
17053 +/* ---------------------------------------------------------------------- */
17054 +
17055 +static inline void au_icntnr_init(struct au_icntnr *c)
17056 +{
17057 +#ifdef CONFIG_AUFS_DEBUG
17058 +       c->vfs_inode.i_mode = 0;
17059 +#endif
17060 +}
17061 +
17062 +static inline unsigned int au_iigen(struct inode *inode, unsigned int *igflags)
17063 +{
17064 +       unsigned int gen;
17065 +       struct au_iinfo *iinfo;
17066 +       struct au_iigen *iigen;
17067 +
17068 +       iinfo = au_ii(inode);
17069 +       iigen = &iinfo->ii_generation;
17070 +       spin_lock(&iigen->ig_spin);
17071 +       if (igflags)
17072 +               *igflags = iigen->ig_flags;
17073 +       gen = iigen->ig_generation;
17074 +       spin_unlock(&iigen->ig_spin);
17075 +
17076 +       return gen;
17077 +}
17078 +
17079 +/* tiny test for inode number */
17080 +/* tmpfs generation is too rough */
17081 +static inline int au_test_higen(struct inode *inode, struct inode *h_inode)
17082 +{
17083 +       struct au_iinfo *iinfo;
17084 +
17085 +       iinfo = au_ii(inode);
17086 +       AuRwMustAnyLock(&iinfo->ii_rwsem);
17087 +       return !(iinfo->ii_hsb1 == h_inode->i_sb
17088 +                && iinfo->ii_higen == h_inode->i_generation);
17089 +}
17090 +
17091 +static inline void au_iigen_dec(struct inode *inode)
17092 +{
17093 +       struct au_iinfo *iinfo;
17094 +       struct au_iigen *iigen;
17095 +
17096 +       iinfo = au_ii(inode);
17097 +       iigen = &iinfo->ii_generation;
17098 +       spin_lock(&iigen->ig_spin);
17099 +       iigen->ig_generation--;
17100 +       spin_unlock(&iigen->ig_spin);
17101 +}
17102 +
17103 +static inline int au_iigen_test(struct inode *inode, unsigned int sigen)
17104 +{
17105 +       int err;
17106 +
17107 +       err = 0;
17108 +       if (unlikely(inode && au_iigen(inode, NULL) != sigen))
17109 +               err = -EIO;
17110 +
17111 +       return err;
17112 +}
17113 +
17114 +/* ---------------------------------------------------------------------- */
17115 +
17116 +static inline struct au_hinode *au_hinode(struct au_iinfo *iinfo,
17117 +                                         aufs_bindex_t bindex)
17118 +{
17119 +       return iinfo->ii_hinode + bindex;
17120 +}
17121 +
17122 +static inline int au_is_bad_inode(struct inode *inode)
17123 +{
17124 +       return !!(is_bad_inode(inode) || !au_hinode(au_ii(inode), 0));
17125 +}
17126 +
17127 +static inline aufs_bindex_t au_ii_br_id(struct inode *inode,
17128 +                                       aufs_bindex_t bindex)
17129 +{
17130 +       IiMustAnyLock(inode);
17131 +       return au_hinode(au_ii(inode), bindex)->hi_id;
17132 +}
17133 +
17134 +static inline aufs_bindex_t au_ibtop(struct inode *inode)
17135 +{
17136 +       IiMustAnyLock(inode);
17137 +       return au_ii(inode)->ii_btop;
17138 +}
17139 +
17140 +static inline aufs_bindex_t au_ibbot(struct inode *inode)
17141 +{
17142 +       IiMustAnyLock(inode);
17143 +       return au_ii(inode)->ii_bbot;
17144 +}
17145 +
17146 +static inline struct au_vdir *au_ivdir(struct inode *inode)
17147 +{
17148 +       IiMustAnyLock(inode);
17149 +       return au_ii(inode)->ii_vdir;
17150 +}
17151 +
17152 +static inline struct dentry *au_hi_wh(struct inode *inode, aufs_bindex_t bindex)
17153 +{
17154 +       IiMustAnyLock(inode);
17155 +       return au_hinode(au_ii(inode), bindex)->hi_whdentry;
17156 +}
17157 +
17158 +static inline void au_set_ibtop(struct inode *inode, aufs_bindex_t bindex)
17159 +{
17160 +       IiMustWriteLock(inode);
17161 +       au_ii(inode)->ii_btop = bindex;
17162 +}
17163 +
17164 +static inline void au_set_ibbot(struct inode *inode, aufs_bindex_t bindex)
17165 +{
17166 +       IiMustWriteLock(inode);
17167 +       au_ii(inode)->ii_bbot = bindex;
17168 +}
17169 +
17170 +static inline void au_set_ivdir(struct inode *inode, struct au_vdir *vdir)
17171 +{
17172 +       IiMustWriteLock(inode);
17173 +       au_ii(inode)->ii_vdir = vdir;
17174 +}
17175 +
17176 +static inline struct au_hinode *au_hi(struct inode *inode, aufs_bindex_t bindex)
17177 +{
17178 +       IiMustAnyLock(inode);
17179 +       return au_hinode(au_ii(inode), bindex);
17180 +}
17181 +
17182 +/* ---------------------------------------------------------------------- */
17183 +
17184 +static inline struct dentry *au_pinned_parent(struct au_pin *pin)
17185 +{
17186 +       if (pin)
17187 +               return pin->parent;
17188 +       return NULL;
17189 +}
17190 +
17191 +static inline struct inode *au_pinned_h_dir(struct au_pin *pin)
17192 +{
17193 +       if (pin && pin->hdir)
17194 +               return pin->hdir->hi_inode;
17195 +       return NULL;
17196 +}
17197 +
17198 +static inline struct au_hinode *au_pinned_hdir(struct au_pin *pin)
17199 +{
17200 +       if (pin)
17201 +               return pin->hdir;
17202 +       return NULL;
17203 +}
17204 +
17205 +static inline void au_pin_set_dentry(struct au_pin *pin, struct dentry *dentry)
17206 +{
17207 +       if (pin)
17208 +               pin->dentry = dentry;
17209 +}
17210 +
17211 +static inline void au_pin_set_parent_lflag(struct au_pin *pin,
17212 +                                          unsigned char lflag)
17213 +{
17214 +       if (pin) {
17215 +               if (lflag)
17216 +                       au_fset_pin(pin->flags, DI_LOCKED);
17217 +               else
17218 +                       au_fclr_pin(pin->flags, DI_LOCKED);
17219 +       }
17220 +}
17221 +
17222 +#if 0 /* reserved */
17223 +static inline void au_pin_set_parent(struct au_pin *pin, struct dentry *parent)
17224 +{
17225 +       if (pin) {
17226 +               dput(pin->parent);
17227 +               pin->parent = dget(parent);
17228 +       }
17229 +}
17230 +#endif
17231 +
17232 +/* ---------------------------------------------------------------------- */
17233 +
17234 +struct au_branch;
17235 +#ifdef CONFIG_AUFS_HNOTIFY
17236 +struct au_hnotify_op {
17237 +       void (*ctl)(struct au_hinode *hinode, int do_set);
17238 +       int (*alloc)(struct au_hinode *hinode);
17239 +
17240 +       /*
17241 +        * if it returns true, the the caller should free hinode->hi_notify,
17242 +        * otherwise ->free() frees it.
17243 +        */
17244 +       int (*free)(struct au_hinode *hinode,
17245 +                   struct au_hnotify *hn) __must_check;
17246 +
17247 +       void (*fin)(void);
17248 +       int (*init)(void);
17249 +
17250 +       int (*reset_br)(unsigned int udba, struct au_branch *br, int perm);
17251 +       void (*fin_br)(struct au_branch *br);
17252 +       int (*init_br)(struct au_branch *br, int perm);
17253 +};
17254 +
17255 +/* hnotify.c */
17256 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode);
17257 +void au_hn_free(struct au_hinode *hinode);
17258 +void au_hn_ctl(struct au_hinode *hinode, int do_set);
17259 +void au_hn_reset(struct inode *inode, unsigned int flags);
17260 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
17261 +              struct qstr *h_child_qstr, struct inode *h_child_inode);
17262 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm);
17263 +int au_hnotify_init_br(struct au_branch *br, int perm);
17264 +void au_hnotify_fin_br(struct au_branch *br);
17265 +int __init au_hnotify_init(void);
17266 +void au_hnotify_fin(void);
17267 +
17268 +/* hfsnotify.c */
17269 +extern const struct au_hnotify_op au_hnotify_op;
17270 +
17271 +static inline
17272 +void au_hn_init(struct au_hinode *hinode)
17273 +{
17274 +       hinode->hi_notify = NULL;
17275 +}
17276 +
17277 +static inline struct au_hnotify *au_hn(struct au_hinode *hinode)
17278 +{
17279 +       return hinode->hi_notify;
17280 +}
17281 +
17282 +#else
17283 +AuStub(int, au_hn_alloc, return -EOPNOTSUPP,
17284 +       struct au_hinode *hinode __maybe_unused,
17285 +       struct inode *inode __maybe_unused)
17286 +AuStub(struct au_hnotify *, au_hn, return NULL, struct au_hinode *hinode)
17287 +AuStubVoid(au_hn_free, struct au_hinode *hinode __maybe_unused)
17288 +AuStubVoid(au_hn_ctl, struct au_hinode *hinode __maybe_unused,
17289 +          int do_set __maybe_unused)
17290 +AuStubVoid(au_hn_reset, struct inode *inode __maybe_unused,
17291 +          unsigned int flags __maybe_unused)
17292 +AuStubInt0(au_hnotify_reset_br, unsigned int udba __maybe_unused,
17293 +          struct au_branch *br __maybe_unused,
17294 +          int perm __maybe_unused)
17295 +AuStubInt0(au_hnotify_init_br, struct au_branch *br __maybe_unused,
17296 +          int perm __maybe_unused)
17297 +AuStubVoid(au_hnotify_fin_br, struct au_branch *br __maybe_unused)
17298 +AuStubInt0(__init au_hnotify_init, void)
17299 +AuStubVoid(au_hnotify_fin, void)
17300 +AuStubVoid(au_hn_init, struct au_hinode *hinode __maybe_unused)
17301 +#endif /* CONFIG_AUFS_HNOTIFY */
17302 +
17303 +static inline void au_hn_suspend(struct au_hinode *hdir)
17304 +{
17305 +       au_hn_ctl(hdir, /*do_set*/0);
17306 +}
17307 +
17308 +static inline void au_hn_resume(struct au_hinode *hdir)
17309 +{
17310 +       au_hn_ctl(hdir, /*do_set*/1);
17311 +}
17312 +
17313 +static inline void au_hn_inode_lock(struct au_hinode *hdir)
17314 +{
17315 +       inode_lock(hdir->hi_inode);
17316 +       au_hn_suspend(hdir);
17317 +}
17318 +
17319 +static inline void au_hn_inode_lock_nested(struct au_hinode *hdir,
17320 +                                         unsigned int sc __maybe_unused)
17321 +{
17322 +       inode_lock_nested(hdir->hi_inode, sc);
17323 +       au_hn_suspend(hdir);
17324 +}
17325 +
17326 +static inline void au_hn_inode_unlock(struct au_hinode *hdir)
17327 +{
17328 +       au_hn_resume(hdir);
17329 +       inode_unlock(hdir->hi_inode);
17330 +}
17331 +
17332 +#endif /* __KERNEL__ */
17333 +#endif /* __AUFS_INODE_H__ */
17334 diff -urN /usr/share/empty/fs/aufs/ioctl.c linux/fs/aufs/ioctl.c
17335 --- /usr/share/empty/fs/aufs/ioctl.c    1970-01-01 01:00:00.000000000 +0100
17336 +++ linux/fs/aufs/ioctl.c       2016-10-09 16:55:36.492701639 +0200
17337 @@ -0,0 +1,219 @@
17338 +/*
17339 + * Copyright (C) 2005-2016 Junjiro R. Okajima
17340 + *
17341 + * This program, aufs is free software; you can redistribute it and/or modify
17342 + * it under the terms of the GNU General Public License as published by
17343 + * the Free Software Foundation; either version 2 of the License, or
17344 + * (at your option) any later version.
17345 + *
17346 + * This program is distributed in the hope that it will be useful,
17347 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17348 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17349 + * GNU General Public License for more details.
17350 + *
17351 + * You should have received a copy of the GNU General Public License
17352 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17353 + */
17354 +
17355 +/*
17356 + * ioctl
17357 + * plink-management and readdir in userspace.
17358 + * assist the pathconf(3) wrapper library.
17359 + * move-down
17360 + * File-based Hierarchical Storage Management.
17361 + */
17362 +
17363 +#include <linux/compat.h>
17364 +#include <linux/file.h>
17365 +#include "aufs.h"
17366 +
17367 +static int au_wbr_fd(struct path *path, struct aufs_wbr_fd __user *arg)
17368 +{
17369 +       int err, fd;
17370 +       aufs_bindex_t wbi, bindex, bbot;
17371 +       struct file *h_file;
17372 +       struct super_block *sb;
17373 +       struct dentry *root;
17374 +       struct au_branch *br;
17375 +       struct aufs_wbr_fd wbrfd = {
17376 +               .oflags = au_dir_roflags,
17377 +               .brid   = -1
17378 +       };
17379 +       const int valid = O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_DIRECTORY
17380 +               | O_NOATIME | O_CLOEXEC;
17381 +
17382 +       AuDebugOn(wbrfd.oflags & ~valid);
17383 +
17384 +       if (arg) {
17385 +               err = copy_from_user(&wbrfd, arg, sizeof(wbrfd));
17386 +               if (unlikely(err)) {
17387 +                       err = -EFAULT;
17388 +                       goto out;
17389 +               }
17390 +
17391 +               err = -EINVAL;
17392 +               AuDbg("wbrfd{0%o, %d}\n", wbrfd.oflags, wbrfd.brid);
17393 +               wbrfd.oflags |= au_dir_roflags;
17394 +               AuDbg("0%o\n", wbrfd.oflags);
17395 +               if (unlikely(wbrfd.oflags & ~valid))
17396 +                       goto out;
17397 +       }
17398 +
17399 +       fd = get_unused_fd_flags(0);
17400 +       err = fd;
17401 +       if (unlikely(fd < 0))
17402 +               goto out;
17403 +
17404 +       h_file = ERR_PTR(-EINVAL);
17405 +       wbi = 0;
17406 +       br = NULL;
17407 +       sb = path->dentry->d_sb;
17408 +       root = sb->s_root;
17409 +       aufs_read_lock(root, AuLock_IR);
17410 +       bbot = au_sbbot(sb);
17411 +       if (wbrfd.brid >= 0) {
17412 +               wbi = au_br_index(sb, wbrfd.brid);
17413 +               if (unlikely(wbi < 0 || wbi > bbot))
17414 +                       goto out_unlock;
17415 +       }
17416 +
17417 +       h_file = ERR_PTR(-ENOENT);
17418 +       br = au_sbr(sb, wbi);
17419 +       if (!au_br_writable(br->br_perm)) {
17420 +               if (arg)
17421 +                       goto out_unlock;
17422 +
17423 +               bindex = wbi + 1;
17424 +               wbi = -1;
17425 +               for (; bindex <= bbot; bindex++) {
17426 +                       br = au_sbr(sb, bindex);
17427 +                       if (au_br_writable(br->br_perm)) {
17428 +                               wbi = bindex;
17429 +                               br = au_sbr(sb, wbi);
17430 +                               break;
17431 +                       }
17432 +               }
17433 +       }
17434 +       AuDbg("wbi %d\n", wbi);
17435 +       if (wbi >= 0)
17436 +               h_file = au_h_open(root, wbi, wbrfd.oflags, NULL,
17437 +                                  /*force_wr*/0);
17438 +
17439 +out_unlock:
17440 +       aufs_read_unlock(root, AuLock_IR);
17441 +       err = PTR_ERR(h_file);
17442 +       if (IS_ERR(h_file))
17443 +               goto out_fd;
17444 +
17445 +       au_br_put(br); /* cf. au_h_open() */
17446 +       fd_install(fd, h_file);
17447 +       err = fd;
17448 +       goto out; /* success */
17449 +
17450 +out_fd:
17451 +       put_unused_fd(fd);
17452 +out:
17453 +       AuTraceErr(err);
17454 +       return err;
17455 +}
17456 +
17457 +/* ---------------------------------------------------------------------- */
17458 +
17459 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg)
17460 +{
17461 +       long err;
17462 +       struct dentry *dentry;
17463 +
17464 +       switch (cmd) {
17465 +       case AUFS_CTL_RDU:
17466 +       case AUFS_CTL_RDU_INO:
17467 +               err = au_rdu_ioctl(file, cmd, arg);
17468 +               break;
17469 +
17470 +       case AUFS_CTL_WBR_FD:
17471 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17472 +               break;
17473 +
17474 +       case AUFS_CTL_IBUSY:
17475 +               err = au_ibusy_ioctl(file, arg);
17476 +               break;
17477 +
17478 +       case AUFS_CTL_BRINFO:
17479 +               err = au_brinfo_ioctl(file, arg);
17480 +               break;
17481 +
17482 +       case AUFS_CTL_FHSM_FD:
17483 +               dentry = file->f_path.dentry;
17484 +               if (IS_ROOT(dentry))
17485 +                       err = au_fhsm_fd(dentry->d_sb, arg);
17486 +               else
17487 +                       err = -ENOTTY;
17488 +               break;
17489 +
17490 +       default:
17491 +               /* do not call the lower */
17492 +               AuDbg("0x%x\n", cmd);
17493 +               err = -ENOTTY;
17494 +       }
17495 +
17496 +       AuTraceErr(err);
17497 +       return err;
17498 +}
17499 +
17500 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg)
17501 +{
17502 +       long err;
17503 +
17504 +       switch (cmd) {
17505 +       case AUFS_CTL_MVDOWN:
17506 +               err = au_mvdown(file->f_path.dentry, (void __user *)arg);
17507 +               break;
17508 +
17509 +       case AUFS_CTL_WBR_FD:
17510 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17511 +               break;
17512 +
17513 +       default:
17514 +               /* do not call the lower */
17515 +               AuDbg("0x%x\n", cmd);
17516 +               err = -ENOTTY;
17517 +       }
17518 +
17519 +       AuTraceErr(err);
17520 +       return err;
17521 +}
17522 +
17523 +#ifdef CONFIG_COMPAT
17524 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
17525 +                          unsigned long arg)
17526 +{
17527 +       long err;
17528 +
17529 +       switch (cmd) {
17530 +       case AUFS_CTL_RDU:
17531 +       case AUFS_CTL_RDU_INO:
17532 +               err = au_rdu_compat_ioctl(file, cmd, arg);
17533 +               break;
17534 +
17535 +       case AUFS_CTL_IBUSY:
17536 +               err = au_ibusy_compat_ioctl(file, arg);
17537 +               break;
17538 +
17539 +       case AUFS_CTL_BRINFO:
17540 +               err = au_brinfo_compat_ioctl(file, arg);
17541 +               break;
17542 +
17543 +       default:
17544 +               err = aufs_ioctl_dir(file, cmd, arg);
17545 +       }
17546 +
17547 +       AuTraceErr(err);
17548 +       return err;
17549 +}
17550 +
17551 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
17552 +                             unsigned long arg)
17553 +{
17554 +       return aufs_ioctl_nondir(file, cmd, (unsigned long)compat_ptr(arg));
17555 +}
17556 +#endif
17557 diff -urN /usr/share/empty/fs/aufs/i_op_add.c linux/fs/aufs/i_op_add.c
17558 --- /usr/share/empty/fs/aufs/i_op_add.c 1970-01-01 01:00:00.000000000 +0100
17559 +++ linux/fs/aufs/i_op_add.c    2016-10-09 16:55:36.492701639 +0200
17560 @@ -0,0 +1,924 @@
17561 +/*
17562 + * Copyright (C) 2005-2016 Junjiro R. Okajima
17563 + *
17564 + * This program, aufs is free software; you can redistribute it and/or modify
17565 + * it under the terms of the GNU General Public License as published by
17566 + * the Free Software Foundation; either version 2 of the License, or
17567 + * (at your option) any later version.
17568 + *
17569 + * This program is distributed in the hope that it will be useful,
17570 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17571 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17572 + * GNU General Public License for more details.
17573 + *
17574 + * You should have received a copy of the GNU General Public License
17575 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17576 + */
17577 +
17578 +/*
17579 + * inode operations (add entry)
17580 + */
17581 +
17582 +#include "aufs.h"
17583 +
17584 +/*
17585 + * final procedure of adding a new entry, except link(2).
17586 + * remove whiteout, instantiate, copyup the parent dir's times and size
17587 + * and update version.
17588 + * if it failed, re-create the removed whiteout.
17589 + */
17590 +static int epilog(struct inode *dir, aufs_bindex_t bindex,
17591 +                 struct dentry *wh_dentry, struct dentry *dentry)
17592 +{
17593 +       int err, rerr;
17594 +       aufs_bindex_t bwh;
17595 +       struct path h_path;
17596 +       struct super_block *sb;
17597 +       struct inode *inode, *h_dir;
17598 +       struct dentry *wh;
17599 +
17600 +       bwh = -1;
17601 +       sb = dir->i_sb;
17602 +       if (wh_dentry) {
17603 +               h_dir = d_inode(wh_dentry->d_parent); /* dir inode is locked */
17604 +               IMustLock(h_dir);
17605 +               AuDebugOn(au_h_iptr(dir, bindex) != h_dir);
17606 +               bwh = au_dbwh(dentry);
17607 +               h_path.dentry = wh_dentry;
17608 +               h_path.mnt = au_sbr_mnt(sb, bindex);
17609 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path,
17610 +                                         dentry);
17611 +               if (unlikely(err))
17612 +                       goto out;
17613 +       }
17614 +
17615 +       inode = au_new_inode(dentry, /*must_new*/1);
17616 +       if (!IS_ERR(inode)) {
17617 +               d_instantiate(dentry, inode);
17618 +               dir = d_inode(dentry->d_parent); /* dir inode is locked */
17619 +               IMustLock(dir);
17620 +               au_dir_ts(dir, bindex);
17621 +               dir->i_version++;
17622 +               au_fhsm_wrote(sb, bindex, /*force*/0);
17623 +               return 0; /* success */
17624 +       }
17625 +
17626 +       err = PTR_ERR(inode);
17627 +       if (!wh_dentry)
17628 +               goto out;
17629 +
17630 +       /* revert */
17631 +       /* dir inode is locked */
17632 +       wh = au_wh_create(dentry, bwh, wh_dentry->d_parent);
17633 +       rerr = PTR_ERR(wh);
17634 +       if (IS_ERR(wh)) {
17635 +               AuIOErr("%pd reverting whiteout failed(%d, %d)\n",
17636 +                       dentry, err, rerr);
17637 +               err = -EIO;
17638 +       } else
17639 +               dput(wh);
17640 +
17641 +out:
17642 +       return err;
17643 +}
17644 +
17645 +static int au_d_may_add(struct dentry *dentry)
17646 +{
17647 +       int err;
17648 +
17649 +       err = 0;
17650 +       if (unlikely(d_unhashed(dentry)))
17651 +               err = -ENOENT;
17652 +       if (unlikely(d_really_is_positive(dentry)))
17653 +               err = -EEXIST;
17654 +       return err;
17655 +}
17656 +
17657 +/*
17658 + * simple tests for the adding inode operations.
17659 + * following the checks in vfs, plus the parent-child relationship.
17660 + */
17661 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
17662 +              struct dentry *h_parent, int isdir)
17663 +{
17664 +       int err;
17665 +       umode_t h_mode;
17666 +       struct dentry *h_dentry;
17667 +       struct inode *h_inode;
17668 +
17669 +       err = -ENAMETOOLONG;
17670 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17671 +               goto out;
17672 +
17673 +       h_dentry = au_h_dptr(dentry, bindex);
17674 +       if (d_really_is_negative(dentry)) {
17675 +               err = -EEXIST;
17676 +               if (unlikely(d_is_positive(h_dentry)))
17677 +                       goto out;
17678 +       } else {
17679 +               /* rename(2) case */
17680 +               err = -EIO;
17681 +               if (unlikely(d_is_negative(h_dentry)))
17682 +                       goto out;
17683 +               h_inode = d_inode(h_dentry);
17684 +               if (unlikely(!h_inode->i_nlink))
17685 +                       goto out;
17686 +
17687 +               h_mode = h_inode->i_mode;
17688 +               if (!isdir) {
17689 +                       err = -EISDIR;
17690 +                       if (unlikely(S_ISDIR(h_mode)))
17691 +                               goto out;
17692 +               } else if (unlikely(!S_ISDIR(h_mode))) {
17693 +                       err = -ENOTDIR;
17694 +                       goto out;
17695 +               }
17696 +       }
17697 +
17698 +       err = 0;
17699 +       /* expected parent dir is locked */
17700 +       if (unlikely(h_parent != h_dentry->d_parent))
17701 +               err = -EIO;
17702 +
17703 +out:
17704 +       AuTraceErr(err);
17705 +       return err;
17706 +}
17707 +
17708 +/*
17709 + * initial procedure of adding a new entry.
17710 + * prepare writable branch and the parent dir, lock it,
17711 + * and lookup whiteout for the new entry.
17712 + */
17713 +static struct dentry*
17714 +lock_hdir_lkup_wh(struct dentry *dentry, struct au_dtime *dt,
17715 +                 struct dentry *src_dentry, struct au_pin *pin,
17716 +                 struct au_wr_dir_args *wr_dir_args)
17717 +{
17718 +       struct dentry *wh_dentry, *h_parent;
17719 +       struct super_block *sb;
17720 +       struct au_branch *br;
17721 +       int err;
17722 +       unsigned int udba;
17723 +       aufs_bindex_t bcpup;
17724 +
17725 +       AuDbg("%pd\n", dentry);
17726 +
17727 +       err = au_wr_dir(dentry, src_dentry, wr_dir_args);
17728 +       bcpup = err;
17729 +       wh_dentry = ERR_PTR(err);
17730 +       if (unlikely(err < 0))
17731 +               goto out;
17732 +
17733 +       sb = dentry->d_sb;
17734 +       udba = au_opt_udba(sb);
17735 +       err = au_pin(pin, dentry, bcpup, udba,
17736 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17737 +       wh_dentry = ERR_PTR(err);
17738 +       if (unlikely(err))
17739 +               goto out;
17740 +
17741 +       h_parent = au_pinned_h_parent(pin);
17742 +       if (udba != AuOpt_UDBA_NONE
17743 +           && au_dbtop(dentry) == bcpup)
17744 +               err = au_may_add(dentry, bcpup, h_parent,
17745 +                                au_ftest_wrdir(wr_dir_args->flags, ISDIR));
17746 +       else if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17747 +               err = -ENAMETOOLONG;
17748 +       wh_dentry = ERR_PTR(err);
17749 +       if (unlikely(err))
17750 +               goto out_unpin;
17751 +
17752 +       br = au_sbr(sb, bcpup);
17753 +       if (dt) {
17754 +               struct path tmp = {
17755 +                       .dentry = h_parent,
17756 +                       .mnt    = au_br_mnt(br)
17757 +               };
17758 +               au_dtime_store(dt, au_pinned_parent(pin), &tmp);
17759 +       }
17760 +
17761 +       wh_dentry = NULL;
17762 +       if (bcpup != au_dbwh(dentry))
17763 +               goto out; /* success */
17764 +
17765 +       /*
17766 +        * ENAMETOOLONG here means that if we allowed create such name, then it
17767 +        * would not be able to removed in the future. So we don't allow such
17768 +        * name here and we don't handle ENAMETOOLONG differently here.
17769 +        */
17770 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
17771 +
17772 +out_unpin:
17773 +       if (IS_ERR(wh_dentry))
17774 +               au_unpin(pin);
17775 +out:
17776 +       return wh_dentry;
17777 +}
17778 +
17779 +/* ---------------------------------------------------------------------- */
17780 +
17781 +enum { Mknod, Symlink, Creat };
17782 +struct simple_arg {
17783 +       int type;
17784 +       union {
17785 +               struct {
17786 +                       umode_t                 mode;
17787 +                       bool                    want_excl;
17788 +                       bool                    try_aopen;
17789 +                       struct vfsub_aopen_args *aopen;
17790 +               } c;
17791 +               struct {
17792 +                       const char *symname;
17793 +               } s;
17794 +               struct {
17795 +                       umode_t mode;
17796 +                       dev_t dev;
17797 +               } m;
17798 +       } u;
17799 +};
17800 +
17801 +static int add_simple(struct inode *dir, struct dentry *dentry,
17802 +                     struct simple_arg *arg)
17803 +{
17804 +       int err, rerr;
17805 +       aufs_bindex_t btop;
17806 +       unsigned char created;
17807 +       const unsigned char try_aopen
17808 +               = (arg->type == Creat && arg->u.c.try_aopen);
17809 +       struct dentry *wh_dentry, *parent;
17810 +       struct inode *h_dir;
17811 +       struct super_block *sb;
17812 +       struct au_branch *br;
17813 +       /* to reuduce stack size */
17814 +       struct {
17815 +               struct au_dtime dt;
17816 +               struct au_pin pin;
17817 +               struct path h_path;
17818 +               struct au_wr_dir_args wr_dir_args;
17819 +       } *a;
17820 +
17821 +       AuDbg("%pd\n", dentry);
17822 +       IMustLock(dir);
17823 +
17824 +       err = -ENOMEM;
17825 +       a = kmalloc(sizeof(*a), GFP_NOFS);
17826 +       if (unlikely(!a))
17827 +               goto out;
17828 +       a->wr_dir_args.force_btgt = -1;
17829 +       a->wr_dir_args.flags = AuWrDir_ADD_ENTRY;
17830 +
17831 +       parent = dentry->d_parent; /* dir inode is locked */
17832 +       if (!try_aopen) {
17833 +               err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
17834 +               if (unlikely(err))
17835 +                       goto out_free;
17836 +       }
17837 +       err = au_d_may_add(dentry);
17838 +       if (unlikely(err))
17839 +               goto out_unlock;
17840 +       if (!try_aopen)
17841 +               di_write_lock_parent(parent);
17842 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
17843 +                                     &a->pin, &a->wr_dir_args);
17844 +       err = PTR_ERR(wh_dentry);
17845 +       if (IS_ERR(wh_dentry))
17846 +               goto out_parent;
17847 +
17848 +       btop = au_dbtop(dentry);
17849 +       sb = dentry->d_sb;
17850 +       br = au_sbr(sb, btop);
17851 +       a->h_path.dentry = au_h_dptr(dentry, btop);
17852 +       a->h_path.mnt = au_br_mnt(br);
17853 +       h_dir = au_pinned_h_dir(&a->pin);
17854 +       switch (arg->type) {
17855 +       case Creat:
17856 +               err = 0;
17857 +               if (!try_aopen || !h_dir->i_op->atomic_open)
17858 +                       err = vfsub_create(h_dir, &a->h_path, arg->u.c.mode,
17859 +                                          arg->u.c.want_excl);
17860 +               else
17861 +                       err = vfsub_atomic_open(h_dir, a->h_path.dentry,
17862 +                                               arg->u.c.aopen, br);
17863 +               break;
17864 +       case Symlink:
17865 +               err = vfsub_symlink(h_dir, &a->h_path, arg->u.s.symname);
17866 +               break;
17867 +       case Mknod:
17868 +               err = vfsub_mknod(h_dir, &a->h_path, arg->u.m.mode,
17869 +                                 arg->u.m.dev);
17870 +               break;
17871 +       default:
17872 +               BUG();
17873 +       }
17874 +       created = !err;
17875 +       if (!err)
17876 +               err = epilog(dir, btop, wh_dentry, dentry);
17877 +
17878 +       /* revert */
17879 +       if (unlikely(created && err && d_is_positive(a->h_path.dentry))) {
17880 +               /* no delegation since it is just created */
17881 +               rerr = vfsub_unlink(h_dir, &a->h_path, /*delegated*/NULL,
17882 +                                   /*force*/0);
17883 +               if (rerr) {
17884 +                       AuIOErr("%pd revert failure(%d, %d)\n",
17885 +                               dentry, err, rerr);
17886 +                       err = -EIO;
17887 +               }
17888 +               au_dtime_revert(&a->dt);
17889 +       }
17890 +
17891 +       if (!err && try_aopen && !h_dir->i_op->atomic_open)
17892 +               *arg->u.c.aopen->opened |= FILE_CREATED;
17893 +
17894 +       au_unpin(&a->pin);
17895 +       dput(wh_dentry);
17896 +
17897 +out_parent:
17898 +       if (!try_aopen)
17899 +               di_write_unlock(parent);
17900 +out_unlock:
17901 +       if (unlikely(err)) {
17902 +               au_update_dbtop(dentry);
17903 +               d_drop(dentry);
17904 +       }
17905 +       if (!try_aopen)
17906 +               aufs_read_unlock(dentry, AuLock_DW);
17907 +out_free:
17908 +       au_delayed_kfree(a);
17909 +out:
17910 +       return err;
17911 +}
17912 +
17913 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
17914 +              dev_t dev)
17915 +{
17916 +       struct simple_arg arg = {
17917 +               .type = Mknod,
17918 +               .u.m = {
17919 +                       .mode   = mode,
17920 +                       .dev    = dev
17921 +               }
17922 +       };
17923 +       return add_simple(dir, dentry, &arg);
17924 +}
17925 +
17926 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
17927 +{
17928 +       struct simple_arg arg = {
17929 +               .type = Symlink,
17930 +               .u.s.symname = symname
17931 +       };
17932 +       return add_simple(dir, dentry, &arg);
17933 +}
17934 +
17935 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
17936 +               bool want_excl)
17937 +{
17938 +       struct simple_arg arg = {
17939 +               .type = Creat,
17940 +               .u.c = {
17941 +                       .mode           = mode,
17942 +                       .want_excl      = want_excl
17943 +               }
17944 +       };
17945 +       return add_simple(dir, dentry, &arg);
17946 +}
17947 +
17948 +int au_aopen_or_create(struct inode *dir, struct dentry *dentry,
17949 +                      struct vfsub_aopen_args *aopen_args)
17950 +{
17951 +       struct simple_arg arg = {
17952 +               .type = Creat,
17953 +               .u.c = {
17954 +                       .mode           = aopen_args->create_mode,
17955 +                       .want_excl      = aopen_args->open_flag & O_EXCL,
17956 +                       .try_aopen      = true,
17957 +                       .aopen          = aopen_args
17958 +               }
17959 +       };
17960 +       return add_simple(dir, dentry, &arg);
17961 +}
17962 +
17963 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
17964 +{
17965 +       int err;
17966 +       aufs_bindex_t bindex;
17967 +       struct super_block *sb;
17968 +       struct dentry *parent, *h_parent, *h_dentry;
17969 +       struct inode *h_dir, *inode;
17970 +       struct vfsmount *h_mnt;
17971 +       struct au_wr_dir_args wr_dir_args = {
17972 +               .force_btgt     = -1,
17973 +               .flags          = AuWrDir_TMPFILE
17974 +       };
17975 +
17976 +       /* copy-up may happen */
17977 +       inode_lock(dir);
17978 +
17979 +       sb = dir->i_sb;
17980 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
17981 +       if (unlikely(err))
17982 +               goto out;
17983 +
17984 +       err = au_di_init(dentry);
17985 +       if (unlikely(err))
17986 +               goto out_si;
17987 +
17988 +       err = -EBUSY;
17989 +       parent = d_find_any_alias(dir);
17990 +       AuDebugOn(!parent);
17991 +       di_write_lock_parent(parent);
17992 +       if (unlikely(d_inode(parent) != dir))
17993 +               goto out_parent;
17994 +
17995 +       err = au_digen_test(parent, au_sigen(sb));
17996 +       if (unlikely(err))
17997 +               goto out_parent;
17998 +
17999 +       bindex = au_dbtop(parent);
18000 +       au_set_dbtop(dentry, bindex);
18001 +       au_set_dbbot(dentry, bindex);
18002 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
18003 +       bindex = err;
18004 +       if (unlikely(err < 0))
18005 +               goto out_parent;
18006 +
18007 +       err = -EOPNOTSUPP;
18008 +       h_dir = au_h_iptr(dir, bindex);
18009 +       if (unlikely(!h_dir->i_op->tmpfile))
18010 +               goto out_parent;
18011 +
18012 +       h_mnt = au_sbr_mnt(sb, bindex);
18013 +       err = vfsub_mnt_want_write(h_mnt);
18014 +       if (unlikely(err))
18015 +               goto out_parent;
18016 +
18017 +       h_parent = au_h_dptr(parent, bindex);
18018 +       err = inode_permission(d_inode(h_parent), MAY_WRITE | MAY_EXEC);
18019 +       if (unlikely(err))
18020 +               goto out_mnt;
18021 +
18022 +       err = -ENOMEM;
18023 +       h_dentry = d_alloc(h_parent, &dentry->d_name);
18024 +       if (unlikely(!h_dentry))
18025 +               goto out_mnt;
18026 +
18027 +       err = h_dir->i_op->tmpfile(h_dir, h_dentry, mode);
18028 +       if (unlikely(err))
18029 +               goto out_dentry;
18030 +
18031 +       au_set_dbtop(dentry, bindex);
18032 +       au_set_dbbot(dentry, bindex);
18033 +       au_set_h_dptr(dentry, bindex, dget(h_dentry));
18034 +       inode = au_new_inode(dentry, /*must_new*/1);
18035 +       if (IS_ERR(inode)) {
18036 +               err = PTR_ERR(inode);
18037 +               au_set_h_dptr(dentry, bindex, NULL);
18038 +               au_set_dbtop(dentry, -1);
18039 +               au_set_dbbot(dentry, -1);
18040 +       } else {
18041 +               if (!inode->i_nlink)
18042 +                       set_nlink(inode, 1);
18043 +               d_tmpfile(dentry, inode);
18044 +               au_di(dentry)->di_tmpfile = 1;
18045 +
18046 +               /* update without i_mutex */
18047 +               if (au_ibtop(dir) == au_dbtop(dentry))
18048 +                       au_cpup_attr_timesizes(dir);
18049 +       }
18050 +
18051 +out_dentry:
18052 +       dput(h_dentry);
18053 +out_mnt:
18054 +       vfsub_mnt_drop_write(h_mnt);
18055 +out_parent:
18056 +       di_write_unlock(parent);
18057 +       dput(parent);
18058 +       di_write_unlock(dentry);
18059 +       if (unlikely(err)) {
18060 +               au_di_fin(dentry);
18061 +               dentry->d_fsdata = NULL;
18062 +       }
18063 +out_si:
18064 +       si_read_unlock(sb);
18065 +out:
18066 +       inode_unlock(dir);
18067 +       return err;
18068 +}
18069 +
18070 +/* ---------------------------------------------------------------------- */
18071 +
18072 +struct au_link_args {
18073 +       aufs_bindex_t bdst, bsrc;
18074 +       struct au_pin pin;
18075 +       struct path h_path;
18076 +       struct dentry *src_parent, *parent;
18077 +};
18078 +
18079 +static int au_cpup_before_link(struct dentry *src_dentry,
18080 +                              struct au_link_args *a)
18081 +{
18082 +       int err;
18083 +       struct dentry *h_src_dentry;
18084 +       struct au_cp_generic cpg = {
18085 +               .dentry = src_dentry,
18086 +               .bdst   = a->bdst,
18087 +               .bsrc   = a->bsrc,
18088 +               .len    = -1,
18089 +               .pin    = &a->pin,
18090 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN /* | AuCpup_KEEPLINO */
18091 +       };
18092 +
18093 +       di_read_lock_parent(a->src_parent, AuLock_IR);
18094 +       err = au_test_and_cpup_dirs(src_dentry, a->bdst);
18095 +       if (unlikely(err))
18096 +               goto out;
18097 +
18098 +       h_src_dentry = au_h_dptr(src_dentry, a->bsrc);
18099 +       err = au_pin(&a->pin, src_dentry, a->bdst,
18100 +                    au_opt_udba(src_dentry->d_sb),
18101 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
18102 +       if (unlikely(err))
18103 +               goto out;
18104 +
18105 +       err = au_sio_cpup_simple(&cpg);
18106 +       au_unpin(&a->pin);
18107 +
18108 +out:
18109 +       di_read_unlock(a->src_parent, AuLock_IR);
18110 +       return err;
18111 +}
18112 +
18113 +static int au_cpup_or_link(struct dentry *src_dentry, struct dentry *dentry,
18114 +                          struct au_link_args *a)
18115 +{
18116 +       int err;
18117 +       unsigned char plink;
18118 +       aufs_bindex_t bbot;
18119 +       struct dentry *h_src_dentry;
18120 +       struct inode *h_inode, *inode, *delegated;
18121 +       struct super_block *sb;
18122 +       struct file *h_file;
18123 +
18124 +       plink = 0;
18125 +       h_inode = NULL;
18126 +       sb = src_dentry->d_sb;
18127 +       inode = d_inode(src_dentry);
18128 +       if (au_ibtop(inode) <= a->bdst)
18129 +               h_inode = au_h_iptr(inode, a->bdst);
18130 +       if (!h_inode || !h_inode->i_nlink) {
18131 +               /* copyup src_dentry as the name of dentry. */
18132 +               bbot = au_dbbot(dentry);
18133 +               if (bbot < a->bsrc)
18134 +                       au_set_dbbot(dentry, a->bsrc);
18135 +               au_set_h_dptr(dentry, a->bsrc,
18136 +                             dget(au_h_dptr(src_dentry, a->bsrc)));
18137 +               dget(a->h_path.dentry);
18138 +               au_set_h_dptr(dentry, a->bdst, NULL);
18139 +               AuDbg("temporary d_inode...\n");
18140 +               spin_lock(&dentry->d_lock);
18141 +               dentry->d_inode = d_inode(src_dentry); /* tmp */
18142 +               spin_unlock(&dentry->d_lock);
18143 +               h_file = au_h_open_pre(dentry, a->bsrc, /*force_wr*/0);
18144 +               if (IS_ERR(h_file))
18145 +                       err = PTR_ERR(h_file);
18146 +               else {
18147 +                       struct au_cp_generic cpg = {
18148 +                               .dentry = dentry,
18149 +                               .bdst   = a->bdst,
18150 +                               .bsrc   = -1,
18151 +                               .len    = -1,
18152 +                               .pin    = &a->pin,
18153 +                               .flags  = AuCpup_KEEPLINO
18154 +                       };
18155 +                       err = au_sio_cpup_simple(&cpg);
18156 +                       au_h_open_post(dentry, a->bsrc, h_file);
18157 +                       if (!err) {
18158 +                               dput(a->h_path.dentry);
18159 +                               a->h_path.dentry = au_h_dptr(dentry, a->bdst);
18160 +                       } else
18161 +                               au_set_h_dptr(dentry, a->bdst,
18162 +                                             a->h_path.dentry);
18163 +               }
18164 +               spin_lock(&dentry->d_lock);
18165 +               dentry->d_inode = NULL; /* restore */
18166 +               spin_unlock(&dentry->d_lock);
18167 +               AuDbg("temporary d_inode...done\n");
18168 +               au_set_h_dptr(dentry, a->bsrc, NULL);
18169 +               au_set_dbbot(dentry, bbot);
18170 +       } else {
18171 +               /* the inode of src_dentry already exists on a.bdst branch */
18172 +               h_src_dentry = d_find_alias(h_inode);
18173 +               if (!h_src_dentry && au_plink_test(inode)) {
18174 +                       plink = 1;
18175 +                       h_src_dentry = au_plink_lkup(inode, a->bdst);
18176 +                       err = PTR_ERR(h_src_dentry);
18177 +                       if (IS_ERR(h_src_dentry))
18178 +                               goto out;
18179 +
18180 +                       if (unlikely(d_is_negative(h_src_dentry))) {
18181 +                               dput(h_src_dentry);
18182 +                               h_src_dentry = NULL;
18183 +                       }
18184 +
18185 +               }
18186 +               if (h_src_dentry) {
18187 +                       delegated = NULL;
18188 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
18189 +                                        &a->h_path, &delegated);
18190 +                       if (unlikely(err == -EWOULDBLOCK)) {
18191 +                               pr_warn("cannot retry for NFSv4 delegation"
18192 +                                       " for an internal link\n");
18193 +                               iput(delegated);
18194 +                       }
18195 +                       dput(h_src_dentry);
18196 +               } else {
18197 +                       AuIOErr("no dentry found for hi%lu on b%d\n",
18198 +                               h_inode->i_ino, a->bdst);
18199 +                       err = -EIO;
18200 +               }
18201 +       }
18202 +
18203 +       if (!err && !plink)
18204 +               au_plink_append(inode, a->bdst, a->h_path.dentry);
18205 +
18206 +out:
18207 +       AuTraceErr(err);
18208 +       return err;
18209 +}
18210 +
18211 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
18212 +             struct dentry *dentry)
18213 +{
18214 +       int err, rerr;
18215 +       struct au_dtime dt;
18216 +       struct au_link_args *a;
18217 +       struct dentry *wh_dentry, *h_src_dentry;
18218 +       struct inode *inode, *delegated;
18219 +       struct super_block *sb;
18220 +       struct au_wr_dir_args wr_dir_args = {
18221 +               /* .force_btgt  = -1, */
18222 +               .flags          = AuWrDir_ADD_ENTRY
18223 +       };
18224 +
18225 +       IMustLock(dir);
18226 +       inode = d_inode(src_dentry);
18227 +       IMustLock(inode);
18228 +
18229 +       err = -ENOMEM;
18230 +       a = kzalloc(sizeof(*a), GFP_NOFS);
18231 +       if (unlikely(!a))
18232 +               goto out;
18233 +
18234 +       a->parent = dentry->d_parent; /* dir inode is locked */
18235 +       err = aufs_read_and_write_lock2(dentry, src_dentry,
18236 +                                       AuLock_NOPLM | AuLock_GEN);
18237 +       if (unlikely(err))
18238 +               goto out_kfree;
18239 +       err = au_d_linkable(src_dentry);
18240 +       if (unlikely(err))
18241 +               goto out_unlock;
18242 +       err = au_d_may_add(dentry);
18243 +       if (unlikely(err))
18244 +               goto out_unlock;
18245 +
18246 +       a->src_parent = dget_parent(src_dentry);
18247 +       wr_dir_args.force_btgt = au_ibtop(inode);
18248 +
18249 +       di_write_lock_parent(a->parent);
18250 +       wr_dir_args.force_btgt = au_wbr(dentry, wr_dir_args.force_btgt);
18251 +       wh_dentry = lock_hdir_lkup_wh(dentry, &dt, src_dentry, &a->pin,
18252 +                                     &wr_dir_args);
18253 +       err = PTR_ERR(wh_dentry);
18254 +       if (IS_ERR(wh_dentry))
18255 +               goto out_parent;
18256 +
18257 +       err = 0;
18258 +       sb = dentry->d_sb;
18259 +       a->bdst = au_dbtop(dentry);
18260 +       a->h_path.dentry = au_h_dptr(dentry, a->bdst);
18261 +       a->h_path.mnt = au_sbr_mnt(sb, a->bdst);
18262 +       a->bsrc = au_ibtop(inode);
18263 +       h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
18264 +       if (!h_src_dentry && au_di(src_dentry)->di_tmpfile)
18265 +               h_src_dentry = dget(au_hi_wh(inode, a->bsrc));
18266 +       if (!h_src_dentry) {
18267 +               a->bsrc = au_dbtop(src_dentry);
18268 +               h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
18269 +               AuDebugOn(!h_src_dentry);
18270 +       } else if (IS_ERR(h_src_dentry)) {
18271 +               err = PTR_ERR(h_src_dentry);
18272 +               goto out_parent;
18273 +       }
18274 +
18275 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
18276 +               if (a->bdst < a->bsrc
18277 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */)
18278 +                       err = au_cpup_or_link(src_dentry, dentry, a);
18279 +               else {
18280 +                       delegated = NULL;
18281 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
18282 +                                        &a->h_path, &delegated);
18283 +                       if (unlikely(err == -EWOULDBLOCK)) {
18284 +                               pr_warn("cannot retry for NFSv4 delegation"
18285 +                                       " for an internal link\n");
18286 +                               iput(delegated);
18287 +                       }
18288 +               }
18289 +               dput(h_src_dentry);
18290 +       } else {
18291 +               /*
18292 +                * copyup src_dentry to the branch we process,
18293 +                * and then link(2) to it.
18294 +                */
18295 +               dput(h_src_dentry);
18296 +               if (a->bdst < a->bsrc
18297 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */) {
18298 +                       au_unpin(&a->pin);
18299 +                       di_write_unlock(a->parent);
18300 +                       err = au_cpup_before_link(src_dentry, a);
18301 +                       di_write_lock_parent(a->parent);
18302 +                       if (!err)
18303 +                               err = au_pin(&a->pin, dentry, a->bdst,
18304 +                                            au_opt_udba(sb),
18305 +                                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
18306 +                       if (unlikely(err))
18307 +                               goto out_wh;
18308 +               }
18309 +               if (!err) {
18310 +                       h_src_dentry = au_h_dptr(src_dentry, a->bdst);
18311 +                       err = -ENOENT;
18312 +                       if (h_src_dentry && d_is_positive(h_src_dentry)) {
18313 +                               delegated = NULL;
18314 +                               err = vfsub_link(h_src_dentry,
18315 +                                                au_pinned_h_dir(&a->pin),
18316 +                                                &a->h_path, &delegated);
18317 +                               if (unlikely(err == -EWOULDBLOCK)) {
18318 +                                       pr_warn("cannot retry"
18319 +                                               " for NFSv4 delegation"
18320 +                                               " for an internal link\n");
18321 +                                       iput(delegated);
18322 +                               }
18323 +                       }
18324 +               }
18325 +       }
18326 +       if (unlikely(err))
18327 +               goto out_unpin;
18328 +
18329 +       if (wh_dentry) {
18330 +               a->h_path.dentry = wh_dentry;
18331 +               err = au_wh_unlink_dentry(au_pinned_h_dir(&a->pin), &a->h_path,
18332 +                                         dentry);
18333 +               if (unlikely(err))
18334 +                       goto out_revert;
18335 +       }
18336 +
18337 +       au_dir_ts(dir, a->bdst);
18338 +       dir->i_version++;
18339 +       inc_nlink(inode);
18340 +       inode->i_ctime = dir->i_ctime;
18341 +       d_instantiate(dentry, au_igrab(inode));
18342 +       if (d_unhashed(a->h_path.dentry))
18343 +               /* some filesystem calls d_drop() */
18344 +               d_drop(dentry);
18345 +       /* some filesystems consume an inode even hardlink */
18346 +       au_fhsm_wrote(sb, a->bdst, /*force*/0);
18347 +       goto out_unpin; /* success */
18348 +
18349 +out_revert:
18350 +       /* no delegation since it is just created */
18351 +       rerr = vfsub_unlink(au_pinned_h_dir(&a->pin), &a->h_path,
18352 +                           /*delegated*/NULL, /*force*/0);
18353 +       if (unlikely(rerr)) {
18354 +               AuIOErr("%pd reverting failed(%d, %d)\n", dentry, err, rerr);
18355 +               err = -EIO;
18356 +       }
18357 +       au_dtime_revert(&dt);
18358 +out_unpin:
18359 +       au_unpin(&a->pin);
18360 +out_wh:
18361 +       dput(wh_dentry);
18362 +out_parent:
18363 +       di_write_unlock(a->parent);
18364 +       dput(a->src_parent);
18365 +out_unlock:
18366 +       if (unlikely(err)) {
18367 +               au_update_dbtop(dentry);
18368 +               d_drop(dentry);
18369 +       }
18370 +       aufs_read_and_write_unlock2(dentry, src_dentry);
18371 +out_kfree:
18372 +       au_delayed_kfree(a);
18373 +out:
18374 +       AuTraceErr(err);
18375 +       return err;
18376 +}
18377 +
18378 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
18379 +{
18380 +       int err, rerr;
18381 +       aufs_bindex_t bindex;
18382 +       unsigned char diropq;
18383 +       struct path h_path;
18384 +       struct dentry *wh_dentry, *parent, *opq_dentry;
18385 +       struct inode *h_inode;
18386 +       struct super_block *sb;
18387 +       struct {
18388 +               struct au_pin pin;
18389 +               struct au_dtime dt;
18390 +       } *a; /* reduce the stack usage */
18391 +       struct au_wr_dir_args wr_dir_args = {
18392 +               .force_btgt     = -1,
18393 +               .flags          = AuWrDir_ADD_ENTRY | AuWrDir_ISDIR
18394 +       };
18395 +
18396 +       IMustLock(dir);
18397 +
18398 +       err = -ENOMEM;
18399 +       a = kmalloc(sizeof(*a), GFP_NOFS);
18400 +       if (unlikely(!a))
18401 +               goto out;
18402 +
18403 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
18404 +       if (unlikely(err))
18405 +               goto out_free;
18406 +       err = au_d_may_add(dentry);
18407 +       if (unlikely(err))
18408 +               goto out_unlock;
18409 +
18410 +       parent = dentry->d_parent; /* dir inode is locked */
18411 +       di_write_lock_parent(parent);
18412 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
18413 +                                     &a->pin, &wr_dir_args);
18414 +       err = PTR_ERR(wh_dentry);
18415 +       if (IS_ERR(wh_dentry))
18416 +               goto out_parent;
18417 +
18418 +       sb = dentry->d_sb;
18419 +       bindex = au_dbtop(dentry);
18420 +       h_path.dentry = au_h_dptr(dentry, bindex);
18421 +       h_path.mnt = au_sbr_mnt(sb, bindex);
18422 +       err = vfsub_mkdir(au_pinned_h_dir(&a->pin), &h_path, mode);
18423 +       if (unlikely(err))
18424 +               goto out_unpin;
18425 +
18426 +       /* make the dir opaque */
18427 +       diropq = 0;
18428 +       h_inode = d_inode(h_path.dentry);
18429 +       if (wh_dentry
18430 +           || au_opt_test(au_mntflags(sb), ALWAYS_DIROPQ)) {
18431 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
18432 +               opq_dentry = au_diropq_create(dentry, bindex);
18433 +               inode_unlock(h_inode);
18434 +               err = PTR_ERR(opq_dentry);
18435 +               if (IS_ERR(opq_dentry))
18436 +                       goto out_dir;
18437 +               dput(opq_dentry);
18438 +               diropq = 1;
18439 +       }
18440 +
18441 +       err = epilog(dir, bindex, wh_dentry, dentry);
18442 +       if (!err) {
18443 +               inc_nlink(dir);
18444 +               goto out_unpin; /* success */
18445 +       }
18446 +
18447 +       /* revert */
18448 +       if (diropq) {
18449 +               AuLabel(revert opq);
18450 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
18451 +               rerr = au_diropq_remove(dentry, bindex);
18452 +               inode_unlock(h_inode);
18453 +               if (rerr) {
18454 +                       AuIOErr("%pd reverting diropq failed(%d, %d)\n",
18455 +                               dentry, err, rerr);
18456 +                       err = -EIO;
18457 +               }
18458 +       }
18459 +
18460 +out_dir:
18461 +       AuLabel(revert dir);
18462 +       rerr = vfsub_rmdir(au_pinned_h_dir(&a->pin), &h_path);
18463 +       if (rerr) {
18464 +               AuIOErr("%pd reverting dir failed(%d, %d)\n",
18465 +                       dentry, err, rerr);
18466 +               err = -EIO;
18467 +       }
18468 +       au_dtime_revert(&a->dt);
18469 +out_unpin:
18470 +       au_unpin(&a->pin);
18471 +       dput(wh_dentry);
18472 +out_parent:
18473 +       di_write_unlock(parent);
18474 +out_unlock:
18475 +       if (unlikely(err)) {
18476 +               au_update_dbtop(dentry);
18477 +               d_drop(dentry);
18478 +       }
18479 +       aufs_read_unlock(dentry, AuLock_DW);
18480 +out_free:
18481 +       au_delayed_kfree(a);
18482 +out:
18483 +       return err;
18484 +}
18485 diff -urN /usr/share/empty/fs/aufs/i_op.c linux/fs/aufs/i_op.c
18486 --- /usr/share/empty/fs/aufs/i_op.c     1970-01-01 01:00:00.000000000 +0100
18487 +++ linux/fs/aufs/i_op.c        2016-10-09 16:55:38.889431135 +0200
18488 @@ -0,0 +1,1451 @@
18489 +/*
18490 + * Copyright (C) 2005-2016 Junjiro R. Okajima
18491 + *
18492 + * This program, aufs is free software; you can redistribute it and/or modify
18493 + * it under the terms of the GNU General Public License as published by
18494 + * the Free Software Foundation; either version 2 of the License, or
18495 + * (at your option) any later version.
18496 + *
18497 + * This program is distributed in the hope that it will be useful,
18498 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18499 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18500 + * GNU General Public License for more details.
18501 + *
18502 + * You should have received a copy of the GNU General Public License
18503 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18504 + */
18505 +
18506 +/*
18507 + * inode operations (except add/del/rename)
18508 + */
18509 +
18510 +#include <linux/device_cgroup.h>
18511 +#include <linux/fs_stack.h>
18512 +#include <linux/namei.h>
18513 +#include <linux/security.h>
18514 +#include "aufs.h"
18515 +
18516 +static int h_permission(struct inode *h_inode, int mask,
18517 +                       struct path *h_path, int brperm)
18518 +{
18519 +       int err;
18520 +       const unsigned char write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18521 +
18522 +       err = -EPERM;
18523 +       if (write_mask && IS_IMMUTABLE(h_inode))
18524 +               goto out;
18525 +
18526 +       err = -EACCES;
18527 +       if (((mask & MAY_EXEC)
18528 +            && S_ISREG(h_inode->i_mode)
18529 +            && (path_noexec(h_path)
18530 +                || !(h_inode->i_mode & S_IXUGO))))
18531 +               goto out;
18532 +
18533 +       /*
18534 +        * - skip the lower fs test in the case of write to ro branch.
18535 +        * - nfs dir permission write check is optimized, but a policy for
18536 +        *   link/rename requires a real check.
18537 +        * - nfs always sets MS_POSIXACL regardless its mount option 'noacl.'
18538 +        *   in this case, generic_permission() returns -EOPNOTSUPP.
18539 +        */
18540 +       if ((write_mask && !au_br_writable(brperm))
18541 +           || (au_test_nfs(h_inode->i_sb) && S_ISDIR(h_inode->i_mode)
18542 +               && write_mask && !(mask & MAY_READ))
18543 +           || !h_inode->i_op->permission) {
18544 +               /* AuLabel(generic_permission); */
18545 +               /* AuDbg("get_acl %pf\n", h_inode->i_op->get_acl); */
18546 +               err = generic_permission(h_inode, mask);
18547 +               if (err == -EOPNOTSUPP && au_test_nfs_noacl(h_inode))
18548 +                       err = h_inode->i_op->permission(h_inode, mask);
18549 +               AuTraceErr(err);
18550 +       } else {
18551 +               /* AuLabel(h_inode->permission); */
18552 +               err = h_inode->i_op->permission(h_inode, mask);
18553 +               AuTraceErr(err);
18554 +       }
18555 +
18556 +       if (!err)
18557 +               err = devcgroup_inode_permission(h_inode, mask);
18558 +       if (!err)
18559 +               err = security_inode_permission(h_inode, mask);
18560 +
18561 +#if 0
18562 +       if (!err) {
18563 +               /* todo: do we need to call ima_path_check()? */
18564 +               struct path h_path = {
18565 +                       .dentry =
18566 +                       .mnt    = h_mnt
18567 +               };
18568 +               err = ima_path_check(&h_path,
18569 +                                    mask & (MAY_READ | MAY_WRITE | MAY_EXEC),
18570 +                                    IMA_COUNT_LEAVE);
18571 +       }
18572 +#endif
18573 +
18574 +out:
18575 +       return err;
18576 +}
18577 +
18578 +static int aufs_permission(struct inode *inode, int mask)
18579 +{
18580 +       int err;
18581 +       aufs_bindex_t bindex, bbot;
18582 +       const unsigned char isdir = !!S_ISDIR(inode->i_mode),
18583 +               write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18584 +       struct inode *h_inode;
18585 +       struct super_block *sb;
18586 +       struct au_branch *br;
18587 +
18588 +       /* todo: support rcu-walk? */
18589 +       if (mask & MAY_NOT_BLOCK)
18590 +               return -ECHILD;
18591 +
18592 +       sb = inode->i_sb;
18593 +       si_read_lock(sb, AuLock_FLUSH);
18594 +       ii_read_lock_child(inode);
18595 +#if 0
18596 +       err = au_iigen_test(inode, au_sigen(sb));
18597 +       if (unlikely(err))
18598 +               goto out;
18599 +#endif
18600 +
18601 +       if (!isdir
18602 +           || write_mask
18603 +           || au_opt_test(au_mntflags(sb), DIRPERM1)) {
18604 +               err = au_busy_or_stale();
18605 +               h_inode = au_h_iptr(inode, au_ibtop(inode));
18606 +               if (unlikely(!h_inode
18607 +                            || (h_inode->i_mode & S_IFMT)
18608 +                            != (inode->i_mode & S_IFMT)))
18609 +                       goto out;
18610 +
18611 +               err = 0;
18612 +               bindex = au_ibtop(inode);
18613 +               br = au_sbr(sb, bindex);
18614 +               err = h_permission(h_inode, mask, &br->br_path, br->br_perm);
18615 +               if (write_mask
18616 +                   && !err
18617 +                   && !special_file(h_inode->i_mode)) {
18618 +                       /* test whether the upper writable branch exists */
18619 +                       err = -EROFS;
18620 +                       for (; bindex >= 0; bindex--)
18621 +                               if (!au_br_rdonly(au_sbr(sb, bindex))) {
18622 +                                       err = 0;
18623 +                                       break;
18624 +                               }
18625 +               }
18626 +               goto out;
18627 +       }
18628 +
18629 +       /* non-write to dir */
18630 +       err = 0;
18631 +       bbot = au_ibbot(inode);
18632 +       for (bindex = au_ibtop(inode); !err && bindex <= bbot; bindex++) {
18633 +               h_inode = au_h_iptr(inode, bindex);
18634 +               if (h_inode) {
18635 +                       err = au_busy_or_stale();
18636 +                       if (unlikely(!S_ISDIR(h_inode->i_mode)))
18637 +                               break;
18638 +
18639 +                       br = au_sbr(sb, bindex);
18640 +                       err = h_permission(h_inode, mask, &br->br_path,
18641 +                                          br->br_perm);
18642 +               }
18643 +       }
18644 +
18645 +out:
18646 +       ii_read_unlock(inode);
18647 +       si_read_unlock(sb);
18648 +       return err;
18649 +}
18650 +
18651 +/* ---------------------------------------------------------------------- */
18652 +
18653 +static struct dentry *aufs_lookup(struct inode *dir, struct dentry *dentry,
18654 +                                 unsigned int flags)
18655 +{
18656 +       struct dentry *ret, *parent;
18657 +       struct inode *inode;
18658 +       struct super_block *sb;
18659 +       int err, npositive;
18660 +
18661 +       IMustLock(dir);
18662 +
18663 +       /* todo: support rcu-walk? */
18664 +       ret = ERR_PTR(-ECHILD);
18665 +       if (flags & LOOKUP_RCU)
18666 +               goto out;
18667 +
18668 +       ret = ERR_PTR(-ENAMETOOLONG);
18669 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
18670 +               goto out;
18671 +
18672 +       sb = dir->i_sb;
18673 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18674 +       ret = ERR_PTR(err);
18675 +       if (unlikely(err))
18676 +               goto out;
18677 +
18678 +       err = au_di_init(dentry);
18679 +       ret = ERR_PTR(err);
18680 +       if (unlikely(err))
18681 +               goto out_si;
18682 +
18683 +       inode = NULL;
18684 +       npositive = 0; /* suppress a warning */
18685 +       parent = dentry->d_parent; /* dir inode is locked */
18686 +       di_read_lock_parent(parent, AuLock_IR);
18687 +       err = au_alive_dir(parent);
18688 +       if (!err)
18689 +               err = au_digen_test(parent, au_sigen(sb));
18690 +       if (!err) {
18691 +               /* regardless LOOKUP_CREATE, always ALLOW_NEG */
18692 +               npositive = au_lkup_dentry(dentry, au_dbtop(parent),
18693 +                                          AuLkup_ALLOW_NEG);
18694 +               err = npositive;
18695 +       }
18696 +       di_read_unlock(parent, AuLock_IR);
18697 +       ret = ERR_PTR(err);
18698 +       if (unlikely(err < 0))
18699 +               goto out_unlock;
18700 +
18701 +       if (npositive) {
18702 +               inode = au_new_inode(dentry, /*must_new*/0);
18703 +               if (IS_ERR(inode)) {
18704 +                       ret = (void *)inode;
18705 +                       inode = NULL;
18706 +                       goto out_unlock;
18707 +               }
18708 +       }
18709 +
18710 +       if (inode)
18711 +               atomic_inc(&inode->i_count);
18712 +       ret = d_splice_alias(inode, dentry);
18713 +#if 0
18714 +       if (unlikely(d_need_lookup(dentry))) {
18715 +               spin_lock(&dentry->d_lock);
18716 +               dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
18717 +               spin_unlock(&dentry->d_lock);
18718 +       } else
18719 +#endif
18720 +       if (inode) {
18721 +               if (!IS_ERR(ret)) {
18722 +                       iput(inode);
18723 +                       if (ret && ret != dentry)
18724 +                               ii_write_unlock(inode);
18725 +               } else {
18726 +                       ii_write_unlock(inode);
18727 +                       iput(inode);
18728 +                       inode = NULL;
18729 +               }
18730 +       }
18731 +
18732 +out_unlock:
18733 +       di_write_unlock(dentry);
18734 +out_si:
18735 +       si_read_unlock(sb);
18736 +out:
18737 +       return ret;
18738 +}
18739 +
18740 +/* ---------------------------------------------------------------------- */
18741 +
18742 +struct aopen_node {
18743 +       struct hlist_node hlist;
18744 +       struct file *file, *h_file;
18745 +};
18746 +
18747 +static int au_do_aopen(struct inode *inode, struct file *file)
18748 +{
18749 +       struct au_sphlhead *aopen;
18750 +       struct aopen_node *node;
18751 +       struct au_do_open_args args = {
18752 +               .no_lock        = 1,
18753 +               .open           = au_do_open_nondir
18754 +       };
18755 +
18756 +       aopen = &au_sbi(inode->i_sb)->si_aopen;
18757 +       spin_lock(&aopen->spin);
18758 +       hlist_for_each_entry(node, &aopen->head, hlist)
18759 +               if (node->file == file) {
18760 +                       args.h_file = node->h_file;
18761 +                       break;
18762 +               }
18763 +       spin_unlock(&aopen->spin);
18764 +       /* AuDebugOn(!args.h_file); */
18765 +
18766 +       return au_do_open(file, &args);
18767 +}
18768 +
18769 +static int aufs_atomic_open(struct inode *dir, struct dentry *dentry,
18770 +                           struct file *file, unsigned int open_flag,
18771 +                           umode_t create_mode, int *opened)
18772 +{
18773 +       int err, h_opened = *opened;
18774 +       unsigned int lkup_flags;
18775 +       struct dentry *parent, *d;
18776 +       struct au_sphlhead *aopen;
18777 +       struct vfsub_aopen_args args = {
18778 +               .open_flag      = open_flag,
18779 +               .create_mode    = create_mode,
18780 +               .opened         = &h_opened
18781 +       };
18782 +       struct aopen_node aopen_node = {
18783 +               .file   = file
18784 +       };
18785 +
18786 +       IMustLock(dir);
18787 +       AuDbg("open_flag 0%o\n", open_flag);
18788 +       AuDbgDentry(dentry);
18789 +
18790 +       err = 0;
18791 +       if (!au_di(dentry)) {
18792 +               lkup_flags = LOOKUP_OPEN;
18793 +               if (open_flag & O_CREAT)
18794 +                       lkup_flags |= LOOKUP_CREATE;
18795 +               d = aufs_lookup(dir, dentry, lkup_flags);
18796 +               if (IS_ERR(d)) {
18797 +                       err = PTR_ERR(d);
18798 +                       AuTraceErr(err);
18799 +                       goto out;
18800 +               } else if (d) {
18801 +                       /*
18802 +                        * obsoleted dentry found.
18803 +                        * another error will be returned later.
18804 +                        */
18805 +                       d_drop(d);
18806 +                       AuDbgDentry(d);
18807 +                       dput(d);
18808 +               }
18809 +               AuDbgDentry(dentry);
18810 +       }
18811 +
18812 +       if (d_is_positive(dentry)
18813 +           || d_unhashed(dentry)
18814 +           || d_unlinked(dentry)
18815 +           || !(open_flag & O_CREAT))
18816 +               goto out_no_open;
18817 +
18818 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
18819 +       if (unlikely(err))
18820 +               goto out;
18821 +
18822 +       parent = dentry->d_parent;      /* dir is locked */
18823 +       di_write_lock_parent(parent);
18824 +       err = au_lkup_dentry(dentry, /*btop*/0, AuLkup_ALLOW_NEG);
18825 +       if (unlikely(err))
18826 +               goto out_unlock;
18827 +
18828 +       AuDbgDentry(dentry);
18829 +       if (d_is_positive(dentry))
18830 +               goto out_unlock;
18831 +
18832 +       args.file = get_empty_filp();
18833 +       err = PTR_ERR(args.file);
18834 +       if (IS_ERR(args.file))
18835 +               goto out_unlock;
18836 +
18837 +       args.file->f_flags = file->f_flags;
18838 +       err = au_aopen_or_create(dir, dentry, &args);
18839 +       AuTraceErr(err);
18840 +       AuDbgFile(args.file);
18841 +       if (unlikely(err < 0)) {
18842 +               if (h_opened & FILE_OPENED)
18843 +                       fput(args.file);
18844 +               else
18845 +                       put_filp(args.file);
18846 +               goto out_unlock;
18847 +       }
18848 +
18849 +       /* some filesystems don't set FILE_CREATED while succeeded? */
18850 +       *opened |= FILE_CREATED;
18851 +       if (h_opened & FILE_OPENED)
18852 +               aopen_node.h_file = args.file;
18853 +       else {
18854 +               put_filp(args.file);
18855 +               args.file = NULL;
18856 +       }
18857 +       aopen = &au_sbi(dir->i_sb)->si_aopen;
18858 +       au_sphl_add(&aopen_node.hlist, aopen);
18859 +       err = finish_open(file, dentry, au_do_aopen, opened);
18860 +       au_sphl_del(&aopen_node.hlist, aopen);
18861 +       AuTraceErr(err);
18862 +       AuDbgFile(file);
18863 +       if (aopen_node.h_file)
18864 +               fput(aopen_node.h_file);
18865 +
18866 +out_unlock:
18867 +       di_write_unlock(parent);
18868 +       aufs_read_unlock(dentry, AuLock_DW);
18869 +       AuDbgDentry(dentry);
18870 +       if (unlikely(err < 0))
18871 +               goto out;
18872 +out_no_open:
18873 +       if (err >= 0 && !(*opened & FILE_CREATED)) {
18874 +               AuLabel(out_no_open);
18875 +               dget(dentry);
18876 +               err = finish_no_open(file, dentry);
18877 +       }
18878 +out:
18879 +       AuDbg("%pd%s%s\n", dentry,
18880 +             (*opened & FILE_CREATED) ? " created" : "",
18881 +             (*opened & FILE_OPENED) ? " opened" : "");
18882 +       AuTraceErr(err);
18883 +       return err;
18884 +}
18885 +
18886 +
18887 +/* ---------------------------------------------------------------------- */
18888 +
18889 +static int au_wr_dir_cpup(struct dentry *dentry, struct dentry *parent,
18890 +                         const unsigned char add_entry, aufs_bindex_t bcpup,
18891 +                         aufs_bindex_t btop)
18892 +{
18893 +       int err;
18894 +       struct dentry *h_parent;
18895 +       struct inode *h_dir;
18896 +
18897 +       if (add_entry)
18898 +               IMustLock(d_inode(parent));
18899 +       else
18900 +               di_write_lock_parent(parent);
18901 +
18902 +       err = 0;
18903 +       if (!au_h_dptr(parent, bcpup)) {
18904 +               if (btop > bcpup)
18905 +                       err = au_cpup_dirs(dentry, bcpup);
18906 +               else if (btop < bcpup)
18907 +                       err = au_cpdown_dirs(dentry, bcpup);
18908 +               else
18909 +                       BUG();
18910 +       }
18911 +       if (!err && add_entry && !au_ftest_wrdir(add_entry, TMPFILE)) {
18912 +               h_parent = au_h_dptr(parent, bcpup);
18913 +               h_dir = d_inode(h_parent);
18914 +               inode_lock_nested(h_dir, AuLsc_I_PARENT);
18915 +               err = au_lkup_neg(dentry, bcpup, /*wh*/0);
18916 +               /* todo: no unlock here */
18917 +               inode_unlock(h_dir);
18918 +
18919 +               AuDbg("bcpup %d\n", bcpup);
18920 +               if (!err) {
18921 +                       if (d_really_is_negative(dentry))
18922 +                               au_set_h_dptr(dentry, btop, NULL);
18923 +                       au_update_dbrange(dentry, /*do_put_zero*/0);
18924 +               }
18925 +       }
18926 +
18927 +       if (!add_entry)
18928 +               di_write_unlock(parent);
18929 +       if (!err)
18930 +               err = bcpup; /* success */
18931 +
18932 +       AuTraceErr(err);
18933 +       return err;
18934 +}
18935 +
18936 +/*
18937 + * decide the branch and the parent dir where we will create a new entry.
18938 + * returns new bindex or an error.
18939 + * copyup the parent dir if needed.
18940 + */
18941 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
18942 +             struct au_wr_dir_args *args)
18943 +{
18944 +       int err;
18945 +       unsigned int flags;
18946 +       aufs_bindex_t bcpup, btop, src_btop;
18947 +       const unsigned char add_entry
18948 +               = au_ftest_wrdir(args->flags, ADD_ENTRY)
18949 +               | au_ftest_wrdir(args->flags, TMPFILE);
18950 +       struct super_block *sb;
18951 +       struct dentry *parent;
18952 +       struct au_sbinfo *sbinfo;
18953 +
18954 +       sb = dentry->d_sb;
18955 +       sbinfo = au_sbi(sb);
18956 +       parent = dget_parent(dentry);
18957 +       btop = au_dbtop(dentry);
18958 +       bcpup = btop;
18959 +       if (args->force_btgt < 0) {
18960 +               if (src_dentry) {
18961 +                       src_btop = au_dbtop(src_dentry);
18962 +                       if (src_btop < btop)
18963 +                               bcpup = src_btop;
18964 +               } else if (add_entry) {
18965 +                       flags = 0;
18966 +                       if (au_ftest_wrdir(args->flags, ISDIR))
18967 +                               au_fset_wbr(flags, DIR);
18968 +                       err = AuWbrCreate(sbinfo, dentry, flags);
18969 +                       bcpup = err;
18970 +               }
18971 +
18972 +               if (bcpup < 0 || au_test_ro(sb, bcpup, d_inode(dentry))) {
18973 +                       if (add_entry)
18974 +                               err = AuWbrCopyup(sbinfo, dentry);
18975 +                       else {
18976 +                               if (!IS_ROOT(dentry)) {
18977 +                                       di_read_lock_parent(parent, !AuLock_IR);
18978 +                                       err = AuWbrCopyup(sbinfo, dentry);
18979 +                                       di_read_unlock(parent, !AuLock_IR);
18980 +                               } else
18981 +                                       err = AuWbrCopyup(sbinfo, dentry);
18982 +                       }
18983 +                       bcpup = err;
18984 +                       if (unlikely(err < 0))
18985 +                               goto out;
18986 +               }
18987 +       } else {
18988 +               bcpup = args->force_btgt;
18989 +               AuDebugOn(au_test_ro(sb, bcpup, d_inode(dentry)));
18990 +       }
18991 +
18992 +       AuDbg("btop %d, bcpup %d\n", btop, bcpup);
18993 +       err = bcpup;
18994 +       if (bcpup == btop)
18995 +               goto out; /* success */
18996 +
18997 +       /* copyup the new parent into the branch we process */
18998 +       err = au_wr_dir_cpup(dentry, parent, add_entry, bcpup, btop);
18999 +       if (err >= 0) {
19000 +               if (d_really_is_negative(dentry)) {
19001 +                       au_set_h_dptr(dentry, btop, NULL);
19002 +                       au_set_dbtop(dentry, bcpup);
19003 +                       au_set_dbbot(dentry, bcpup);
19004 +               }
19005 +               AuDebugOn(add_entry
19006 +                         && !au_ftest_wrdir(args->flags, TMPFILE)
19007 +                         && !au_h_dptr(dentry, bcpup));
19008 +       }
19009 +
19010 +out:
19011 +       dput(parent);
19012 +       return err;
19013 +}
19014 +
19015 +/* ---------------------------------------------------------------------- */
19016 +
19017 +void au_pin_hdir_unlock(struct au_pin *p)
19018 +{
19019 +       if (p->hdir)
19020 +               au_hn_inode_unlock(p->hdir);
19021 +}
19022 +
19023 +int au_pin_hdir_lock(struct au_pin *p)
19024 +{
19025 +       int err;
19026 +
19027 +       err = 0;
19028 +       if (!p->hdir)
19029 +               goto out;
19030 +
19031 +       /* even if an error happens later, keep this lock */
19032 +       au_hn_inode_lock_nested(p->hdir, p->lsc_hi);
19033 +
19034 +       err = -EBUSY;
19035 +       if (unlikely(p->hdir->hi_inode != d_inode(p->h_parent)))
19036 +               goto out;
19037 +
19038 +       err = 0;
19039 +       if (p->h_dentry)
19040 +               err = au_h_verify(p->h_dentry, p->udba, p->hdir->hi_inode,
19041 +                                 p->h_parent, p->br);
19042 +
19043 +out:
19044 +       return err;
19045 +}
19046 +
19047 +int au_pin_hdir_relock(struct au_pin *p)
19048 +{
19049 +       int err, i;
19050 +       struct inode *h_i;
19051 +       struct dentry *h_d[] = {
19052 +               p->h_dentry,
19053 +               p->h_parent
19054 +       };
19055 +
19056 +       err = au_pin_hdir_lock(p);
19057 +       if (unlikely(err))
19058 +               goto out;
19059 +
19060 +       for (i = 0; !err && i < sizeof(h_d)/sizeof(*h_d); i++) {
19061 +               if (!h_d[i])
19062 +                       continue;
19063 +               if (d_is_positive(h_d[i])) {
19064 +                       h_i = d_inode(h_d[i]);
19065 +                       err = !h_i->i_nlink;
19066 +               }
19067 +       }
19068 +
19069 +out:
19070 +       return err;
19071 +}
19072 +
19073 +static void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task)
19074 +{
19075 +#if !defined(CONFIG_RWSEM_GENERIC_SPINLOCK) && defined(CONFIG_RWSEM_SPIN_ON_OWNER)
19076 +       p->hdir->hi_inode->i_rwsem.owner = task;
19077 +#endif
19078 +}
19079 +
19080 +void au_pin_hdir_acquire_nest(struct au_pin *p)
19081 +{
19082 +       if (p->hdir) {
19083 +               rwsem_acquire_nest(&p->hdir->hi_inode->i_rwsem.dep_map,
19084 +                                  p->lsc_hi, 0, NULL, _RET_IP_);
19085 +               au_pin_hdir_set_owner(p, current);
19086 +       }
19087 +}
19088 +
19089 +void au_pin_hdir_release(struct au_pin *p)
19090 +{
19091 +       if (p->hdir) {
19092 +               au_pin_hdir_set_owner(p, p->task);
19093 +               rwsem_release(&p->hdir->hi_inode->i_rwsem.dep_map, 1, _RET_IP_);
19094 +       }
19095 +}
19096 +
19097 +struct dentry *au_pinned_h_parent(struct au_pin *pin)
19098 +{
19099 +       if (pin && pin->parent)
19100 +               return au_h_dptr(pin->parent, pin->bindex);
19101 +       return NULL;
19102 +}
19103 +
19104 +void au_unpin(struct au_pin *p)
19105 +{
19106 +       if (p->hdir)
19107 +               au_pin_hdir_unlock(p);
19108 +       if (p->h_mnt && au_ftest_pin(p->flags, MNT_WRITE))
19109 +               vfsub_mnt_drop_write(p->h_mnt);
19110 +       if (!p->hdir)
19111 +               return;
19112 +
19113 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
19114 +               di_read_unlock(p->parent, AuLock_IR);
19115 +       iput(p->hdir->hi_inode);
19116 +       dput(p->parent);
19117 +       p->parent = NULL;
19118 +       p->hdir = NULL;
19119 +       p->h_mnt = NULL;
19120 +       /* do not clear p->task */
19121 +}
19122 +
19123 +int au_do_pin(struct au_pin *p)
19124 +{
19125 +       int err;
19126 +       struct super_block *sb;
19127 +       struct inode *h_dir;
19128 +
19129 +       err = 0;
19130 +       sb = p->dentry->d_sb;
19131 +       p->br = au_sbr(sb, p->bindex);
19132 +       if (IS_ROOT(p->dentry)) {
19133 +               if (au_ftest_pin(p->flags, MNT_WRITE)) {
19134 +                       p->h_mnt = au_br_mnt(p->br);
19135 +                       err = vfsub_mnt_want_write(p->h_mnt);
19136 +                       if (unlikely(err)) {
19137 +                               au_fclr_pin(p->flags, MNT_WRITE);
19138 +                               goto out_err;
19139 +                       }
19140 +               }
19141 +               goto out;
19142 +       }
19143 +
19144 +       p->h_dentry = NULL;
19145 +       if (p->bindex <= au_dbbot(p->dentry))
19146 +               p->h_dentry = au_h_dptr(p->dentry, p->bindex);
19147 +
19148 +       p->parent = dget_parent(p->dentry);
19149 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
19150 +               di_read_lock(p->parent, AuLock_IR, p->lsc_di);
19151 +
19152 +       h_dir = NULL;
19153 +       p->h_parent = au_h_dptr(p->parent, p->bindex);
19154 +       p->hdir = au_hi(d_inode(p->parent), p->bindex);
19155 +       if (p->hdir)
19156 +               h_dir = p->hdir->hi_inode;
19157 +
19158 +       /*
19159 +        * udba case, or
19160 +        * if DI_LOCKED is not set, then p->parent may be different
19161 +        * and h_parent can be NULL.
19162 +        */
19163 +       if (unlikely(!p->hdir || !h_dir || !p->h_parent)) {
19164 +               err = -EBUSY;
19165 +               if (!au_ftest_pin(p->flags, DI_LOCKED))
19166 +                       di_read_unlock(p->parent, AuLock_IR);
19167 +               dput(p->parent);
19168 +               p->parent = NULL;
19169 +               goto out_err;
19170 +       }
19171 +
19172 +       if (au_ftest_pin(p->flags, MNT_WRITE)) {
19173 +               p->h_mnt = au_br_mnt(p->br);
19174 +               err = vfsub_mnt_want_write(p->h_mnt);
19175 +               if (unlikely(err)) {
19176 +                       au_fclr_pin(p->flags, MNT_WRITE);
19177 +                       if (!au_ftest_pin(p->flags, DI_LOCKED))
19178 +                               di_read_unlock(p->parent, AuLock_IR);
19179 +                       dput(p->parent);
19180 +                       p->parent = NULL;
19181 +                       goto out_err;
19182 +               }
19183 +       }
19184 +
19185 +       au_igrab(h_dir);
19186 +       err = au_pin_hdir_lock(p);
19187 +       if (!err)
19188 +               goto out; /* success */
19189 +
19190 +       au_unpin(p);
19191 +
19192 +out_err:
19193 +       pr_err("err %d\n", err);
19194 +       err = au_busy_or_stale();
19195 +out:
19196 +       return err;
19197 +}
19198 +
19199 +void au_pin_init(struct au_pin *p, struct dentry *dentry,
19200 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
19201 +                unsigned int udba, unsigned char flags)
19202 +{
19203 +       p->dentry = dentry;
19204 +       p->udba = udba;
19205 +       p->lsc_di = lsc_di;
19206 +       p->lsc_hi = lsc_hi;
19207 +       p->flags = flags;
19208 +       p->bindex = bindex;
19209 +
19210 +       p->parent = NULL;
19211 +       p->hdir = NULL;
19212 +       p->h_mnt = NULL;
19213 +
19214 +       p->h_dentry = NULL;
19215 +       p->h_parent = NULL;
19216 +       p->br = NULL;
19217 +       p->task = current;
19218 +}
19219 +
19220 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
19221 +          unsigned int udba, unsigned char flags)
19222 +{
19223 +       au_pin_init(pin, dentry, bindex, AuLsc_DI_PARENT, AuLsc_I_PARENT2,
19224 +                   udba, flags);
19225 +       return au_do_pin(pin);
19226 +}
19227 +
19228 +/* ---------------------------------------------------------------------- */
19229 +
19230 +/*
19231 + * ->setattr() and ->getattr() are called in various cases.
19232 + * chmod, stat: dentry is revalidated.
19233 + * fchmod, fstat: file and dentry are not revalidated, additionally they may be
19234 + *               unhashed.
19235 + * for ->setattr(), ia->ia_file is passed from ftruncate only.
19236 + */
19237 +/* todo: consolidate with do_refresh() and simple_reval_dpath() */
19238 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen)
19239 +{
19240 +       int err;
19241 +       struct dentry *parent;
19242 +
19243 +       err = 0;
19244 +       if (au_digen_test(dentry, sigen)) {
19245 +               parent = dget_parent(dentry);
19246 +               di_read_lock_parent(parent, AuLock_IR);
19247 +               err = au_refresh_dentry(dentry, parent);
19248 +               di_read_unlock(parent, AuLock_IR);
19249 +               dput(parent);
19250 +       }
19251 +
19252 +       AuTraceErr(err);
19253 +       return err;
19254 +}
19255 +
19256 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
19257 +                    struct au_icpup_args *a)
19258 +{
19259 +       int err;
19260 +       loff_t sz;
19261 +       aufs_bindex_t btop, ibtop;
19262 +       struct dentry *hi_wh, *parent;
19263 +       struct inode *inode;
19264 +       struct au_wr_dir_args wr_dir_args = {
19265 +               .force_btgt     = -1,
19266 +               .flags          = 0
19267 +       };
19268 +
19269 +       if (d_is_dir(dentry))
19270 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
19271 +       /* plink or hi_wh() case */
19272 +       btop = au_dbtop(dentry);
19273 +       inode = d_inode(dentry);
19274 +       ibtop = au_ibtop(inode);
19275 +       if (btop != ibtop && !au_test_ro(inode->i_sb, ibtop, inode))
19276 +               wr_dir_args.force_btgt = ibtop;
19277 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
19278 +       if (unlikely(err < 0))
19279 +               goto out;
19280 +       a->btgt = err;
19281 +       if (err != btop)
19282 +               au_fset_icpup(a->flags, DID_CPUP);
19283 +
19284 +       err = 0;
19285 +       a->pin_flags = AuPin_MNT_WRITE;
19286 +       parent = NULL;
19287 +       if (!IS_ROOT(dentry)) {
19288 +               au_fset_pin(a->pin_flags, DI_LOCKED);
19289 +               parent = dget_parent(dentry);
19290 +               di_write_lock_parent(parent);
19291 +       }
19292 +
19293 +       err = au_pin(&a->pin, dentry, a->btgt, a->udba, a->pin_flags);
19294 +       if (unlikely(err))
19295 +               goto out_parent;
19296 +
19297 +       sz = -1;
19298 +       a->h_path.dentry = au_h_dptr(dentry, btop);
19299 +       a->h_inode = d_inode(a->h_path.dentry);
19300 +       if (ia && (ia->ia_valid & ATTR_SIZE)) {
19301 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
19302 +               if (ia->ia_size < i_size_read(a->h_inode))
19303 +                       sz = ia->ia_size;
19304 +               inode_unlock(a->h_inode);
19305 +       }
19306 +
19307 +       hi_wh = NULL;
19308 +       if (au_ftest_icpup(a->flags, DID_CPUP) && d_unlinked(dentry)) {
19309 +               hi_wh = au_hi_wh(inode, a->btgt);
19310 +               if (!hi_wh) {
19311 +                       struct au_cp_generic cpg = {
19312 +                               .dentry = dentry,
19313 +                               .bdst   = a->btgt,
19314 +                               .bsrc   = -1,
19315 +                               .len    = sz,
19316 +                               .pin    = &a->pin
19317 +                       };
19318 +                       err = au_sio_cpup_wh(&cpg, /*file*/NULL);
19319 +                       if (unlikely(err))
19320 +                               goto out_unlock;
19321 +                       hi_wh = au_hi_wh(inode, a->btgt);
19322 +                       /* todo: revalidate hi_wh? */
19323 +               }
19324 +       }
19325 +
19326 +       if (parent) {
19327 +               au_pin_set_parent_lflag(&a->pin, /*lflag*/0);
19328 +               di_downgrade_lock(parent, AuLock_IR);
19329 +               dput(parent);
19330 +               parent = NULL;
19331 +       }
19332 +       if (!au_ftest_icpup(a->flags, DID_CPUP))
19333 +               goto out; /* success */
19334 +
19335 +       if (!d_unhashed(dentry)) {
19336 +               struct au_cp_generic cpg = {
19337 +                       .dentry = dentry,
19338 +                       .bdst   = a->btgt,
19339 +                       .bsrc   = btop,
19340 +                       .len    = sz,
19341 +                       .pin    = &a->pin,
19342 +                       .flags  = AuCpup_DTIME | AuCpup_HOPEN
19343 +               };
19344 +               err = au_sio_cpup_simple(&cpg);
19345 +               if (!err)
19346 +                       a->h_path.dentry = au_h_dptr(dentry, a->btgt);
19347 +       } else if (!hi_wh)
19348 +               a->h_path.dentry = au_h_dptr(dentry, a->btgt);
19349 +       else
19350 +               a->h_path.dentry = hi_wh; /* do not dget here */
19351 +
19352 +out_unlock:
19353 +       a->h_inode = d_inode(a->h_path.dentry);
19354 +       if (!err)
19355 +               goto out; /* success */
19356 +       au_unpin(&a->pin);
19357 +out_parent:
19358 +       if (parent) {
19359 +               di_write_unlock(parent);
19360 +               dput(parent);
19361 +       }
19362 +out:
19363 +       if (!err)
19364 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
19365 +       return err;
19366 +}
19367 +
19368 +static int aufs_setattr(struct dentry *dentry, struct iattr *ia)
19369 +{
19370 +       int err;
19371 +       struct inode *inode, *delegated;
19372 +       struct super_block *sb;
19373 +       struct file *file;
19374 +       struct au_icpup_args *a;
19375 +
19376 +       inode = d_inode(dentry);
19377 +       IMustLock(inode);
19378 +
19379 +       err = -ENOMEM;
19380 +       a = kzalloc(sizeof(*a), GFP_NOFS);
19381 +       if (unlikely(!a))
19382 +               goto out;
19383 +
19384 +       if (ia->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
19385 +               ia->ia_valid &= ~ATTR_MODE;
19386 +
19387 +       file = NULL;
19388 +       sb = dentry->d_sb;
19389 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19390 +       if (unlikely(err))
19391 +               goto out_kfree;
19392 +
19393 +       if (ia->ia_valid & ATTR_FILE) {
19394 +               /* currently ftruncate(2) only */
19395 +               AuDebugOn(!d_is_reg(dentry));
19396 +               file = ia->ia_file;
19397 +               err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
19398 +               if (unlikely(err))
19399 +                       goto out_si;
19400 +               ia->ia_file = au_hf_top(file);
19401 +               a->udba = AuOpt_UDBA_NONE;
19402 +       } else {
19403 +               /* fchmod() doesn't pass ia_file */
19404 +               a->udba = au_opt_udba(sb);
19405 +               di_write_lock_child(dentry);
19406 +               /* no d_unlinked(), to set UDBA_NONE for root */
19407 +               if (d_unhashed(dentry))
19408 +                       a->udba = AuOpt_UDBA_NONE;
19409 +               if (a->udba != AuOpt_UDBA_NONE) {
19410 +                       AuDebugOn(IS_ROOT(dentry));
19411 +                       err = au_reval_for_attr(dentry, au_sigen(sb));
19412 +                       if (unlikely(err))
19413 +                               goto out_dentry;
19414 +               }
19415 +       }
19416 +
19417 +       err = au_pin_and_icpup(dentry, ia, a);
19418 +       if (unlikely(err < 0))
19419 +               goto out_dentry;
19420 +       if (au_ftest_icpup(a->flags, DID_CPUP)) {
19421 +               ia->ia_file = NULL;
19422 +               ia->ia_valid &= ~ATTR_FILE;
19423 +       }
19424 +
19425 +       a->h_path.mnt = au_sbr_mnt(sb, a->btgt);
19426 +       if ((ia->ia_valid & (ATTR_MODE | ATTR_CTIME))
19427 +           == (ATTR_MODE | ATTR_CTIME)) {
19428 +               err = security_path_chmod(&a->h_path, ia->ia_mode);
19429 +               if (unlikely(err))
19430 +                       goto out_unlock;
19431 +       } else if ((ia->ia_valid & (ATTR_UID | ATTR_GID))
19432 +                  && (ia->ia_valid & ATTR_CTIME)) {
19433 +               err = security_path_chown(&a->h_path, ia->ia_uid, ia->ia_gid);
19434 +               if (unlikely(err))
19435 +                       goto out_unlock;
19436 +       }
19437 +
19438 +       if (ia->ia_valid & ATTR_SIZE) {
19439 +               struct file *f;
19440 +
19441 +               if (ia->ia_size < i_size_read(inode))
19442 +                       /* unmap only */
19443 +                       truncate_setsize(inode, ia->ia_size);
19444 +
19445 +               f = NULL;
19446 +               if (ia->ia_valid & ATTR_FILE)
19447 +                       f = ia->ia_file;
19448 +               inode_unlock(a->h_inode);
19449 +               err = vfsub_trunc(&a->h_path, ia->ia_size, ia->ia_valid, f);
19450 +               inode_lock_nested(a->h_inode, AuLsc_I_CHILD);
19451 +       } else {
19452 +               delegated = NULL;
19453 +               while (1) {
19454 +                       err = vfsub_notify_change(&a->h_path, ia, &delegated);
19455 +                       if (delegated) {
19456 +                               err = break_deleg_wait(&delegated);
19457 +                               if (!err)
19458 +                                       continue;
19459 +                       }
19460 +                       break;
19461 +               }
19462 +       }
19463 +       /*
19464 +        * regardless aufs 'acl' option setting.
19465 +        * why don't all acl-aware fs call this func from their ->setattr()?
19466 +        */
19467 +       if (!err && (ia->ia_valid & ATTR_MODE))
19468 +               err = vfsub_acl_chmod(a->h_inode, ia->ia_mode);
19469 +       if (!err)
19470 +               au_cpup_attr_changeable(inode);
19471 +
19472 +out_unlock:
19473 +       inode_unlock(a->h_inode);
19474 +       au_unpin(&a->pin);
19475 +       if (unlikely(err))
19476 +               au_update_dbtop(dentry);
19477 +out_dentry:
19478 +       di_write_unlock(dentry);
19479 +       if (file) {
19480 +               fi_write_unlock(file);
19481 +               ia->ia_file = file;
19482 +               ia->ia_valid |= ATTR_FILE;
19483 +       }
19484 +out_si:
19485 +       si_read_unlock(sb);
19486 +out_kfree:
19487 +       au_delayed_kfree(a);
19488 +out:
19489 +       AuTraceErr(err);
19490 +       return err;
19491 +}
19492 +
19493 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
19494 +static int au_h_path_to_set_attr(struct dentry *dentry,
19495 +                                struct au_icpup_args *a, struct path *h_path)
19496 +{
19497 +       int err;
19498 +       struct super_block *sb;
19499 +
19500 +       sb = dentry->d_sb;
19501 +       a->udba = au_opt_udba(sb);
19502 +       /* no d_unlinked(), to set UDBA_NONE for root */
19503 +       if (d_unhashed(dentry))
19504 +               a->udba = AuOpt_UDBA_NONE;
19505 +       if (a->udba != AuOpt_UDBA_NONE) {
19506 +               AuDebugOn(IS_ROOT(dentry));
19507 +               err = au_reval_for_attr(dentry, au_sigen(sb));
19508 +               if (unlikely(err))
19509 +                       goto out;
19510 +       }
19511 +       err = au_pin_and_icpup(dentry, /*ia*/NULL, a);
19512 +       if (unlikely(err < 0))
19513 +               goto out;
19514 +
19515 +       h_path->dentry = a->h_path.dentry;
19516 +       h_path->mnt = au_sbr_mnt(sb, a->btgt);
19517 +
19518 +out:
19519 +       return err;
19520 +}
19521 +
19522 +ssize_t au_srxattr(struct dentry *dentry, struct inode *inode,
19523 +                  struct au_srxattr *arg)
19524 +{
19525 +       int err;
19526 +       struct path h_path;
19527 +       struct super_block *sb;
19528 +       struct au_icpup_args *a;
19529 +       struct inode *h_inode;
19530 +
19531 +       IMustLock(inode);
19532 +
19533 +       err = -ENOMEM;
19534 +       a = kzalloc(sizeof(*a), GFP_NOFS);
19535 +       if (unlikely(!a))
19536 +               goto out;
19537 +
19538 +       sb = dentry->d_sb;
19539 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19540 +       if (unlikely(err))
19541 +               goto out_kfree;
19542 +
19543 +       h_path.dentry = NULL;   /* silence gcc */
19544 +       di_write_lock_child(dentry);
19545 +       err = au_h_path_to_set_attr(dentry, a, &h_path);
19546 +       if (unlikely(err))
19547 +               goto out_di;
19548 +
19549 +       inode_unlock(a->h_inode);
19550 +       switch (arg->type) {
19551 +       case AU_XATTR_SET:
19552 +               AuDebugOn(d_is_negative(h_path.dentry));
19553 +               err = vfsub_setxattr(h_path.dentry,
19554 +                                    arg->u.set.name, arg->u.set.value,
19555 +                                    arg->u.set.size, arg->u.set.flags);
19556 +               break;
19557 +       case AU_XATTR_REMOVE:
19558 +               err = vfsub_removexattr(h_path.dentry, arg->u.remove.name);
19559 +               break;
19560 +       case AU_ACL_SET:
19561 +               err = -EOPNOTSUPP;
19562 +               h_inode = d_inode(h_path.dentry);
19563 +               if (h_inode->i_op->set_acl)
19564 +                       err = h_inode->i_op->set_acl(h_inode,
19565 +                                                    arg->u.acl_set.acl,
19566 +                                                    arg->u.acl_set.type);
19567 +               break;
19568 +       }
19569 +       if (!err)
19570 +               au_cpup_attr_timesizes(inode);
19571 +
19572 +       au_unpin(&a->pin);
19573 +       if (unlikely(err))
19574 +               au_update_dbtop(dentry);
19575 +
19576 +out_di:
19577 +       di_write_unlock(dentry);
19578 +       si_read_unlock(sb);
19579 +out_kfree:
19580 +       au_delayed_kfree(a);
19581 +out:
19582 +       AuTraceErr(err);
19583 +       return err;
19584 +}
19585 +#endif
19586 +
19587 +static void au_refresh_iattr(struct inode *inode, struct kstat *st,
19588 +                            unsigned int nlink)
19589 +{
19590 +       unsigned int n;
19591 +
19592 +       inode->i_mode = st->mode;
19593 +       /* don't i_[ug]id_write() here */
19594 +       inode->i_uid = st->uid;
19595 +       inode->i_gid = st->gid;
19596 +       inode->i_atime = st->atime;
19597 +       inode->i_mtime = st->mtime;
19598 +       inode->i_ctime = st->ctime;
19599 +
19600 +       au_cpup_attr_nlink(inode, /*force*/0);
19601 +       if (S_ISDIR(inode->i_mode)) {
19602 +               n = inode->i_nlink;
19603 +               n -= nlink;
19604 +               n += st->nlink;
19605 +               smp_mb(); /* for i_nlink */
19606 +               /* 0 can happen */
19607 +               set_nlink(inode, n);
19608 +       }
19609 +
19610 +       spin_lock(&inode->i_lock);
19611 +       inode->i_blocks = st->blocks;
19612 +       i_size_write(inode, st->size);
19613 +       spin_unlock(&inode->i_lock);
19614 +}
19615 +
19616 +/*
19617 + * common routine for aufs_getattr() and aufs_getxattr().
19618 + * returns zero or negative (an error).
19619 + * @dentry will be read-locked in success.
19620 + */
19621 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path)
19622 +{
19623 +       int err;
19624 +       unsigned int mnt_flags, sigen;
19625 +       unsigned char udba_none;
19626 +       aufs_bindex_t bindex;
19627 +       struct super_block *sb, *h_sb;
19628 +       struct inode *inode;
19629 +
19630 +       h_path->mnt = NULL;
19631 +       h_path->dentry = NULL;
19632 +
19633 +       err = 0;
19634 +       sb = dentry->d_sb;
19635 +       mnt_flags = au_mntflags(sb);
19636 +       udba_none = !!au_opt_test(mnt_flags, UDBA_NONE);
19637 +
19638 +       /* support fstat(2) */
19639 +       if (!d_unlinked(dentry) && !udba_none) {
19640 +               sigen = au_sigen(sb);
19641 +               err = au_digen_test(dentry, sigen);
19642 +               if (!err) {
19643 +                       di_read_lock_child(dentry, AuLock_IR);
19644 +                       err = au_dbrange_test(dentry);
19645 +                       if (unlikely(err)) {
19646 +                               di_read_unlock(dentry, AuLock_IR);
19647 +                               goto out;
19648 +                       }
19649 +               } else {
19650 +                       AuDebugOn(IS_ROOT(dentry));
19651 +                       di_write_lock_child(dentry);
19652 +                       err = au_dbrange_test(dentry);
19653 +                       if (!err)
19654 +                               err = au_reval_for_attr(dentry, sigen);
19655 +                       if (!err)
19656 +                               di_downgrade_lock(dentry, AuLock_IR);
19657 +                       else {
19658 +                               di_write_unlock(dentry);
19659 +                               goto out;
19660 +                       }
19661 +               }
19662 +       } else
19663 +               di_read_lock_child(dentry, AuLock_IR);
19664 +
19665 +       inode = d_inode(dentry);
19666 +       bindex = au_ibtop(inode);
19667 +       h_path->mnt = au_sbr_mnt(sb, bindex);
19668 +       h_sb = h_path->mnt->mnt_sb;
19669 +       if (!force
19670 +           && !au_test_fs_bad_iattr(h_sb)
19671 +           && udba_none)
19672 +               goto out; /* success */
19673 +
19674 +       if (au_dbtop(dentry) == bindex)
19675 +               h_path->dentry = au_h_dptr(dentry, bindex);
19676 +       else if (au_opt_test(mnt_flags, PLINK) && au_plink_test(inode)) {
19677 +               h_path->dentry = au_plink_lkup(inode, bindex);
19678 +               if (IS_ERR(h_path->dentry))
19679 +                       /* pretending success */
19680 +                       h_path->dentry = NULL;
19681 +               else
19682 +                       dput(h_path->dentry);
19683 +       }
19684 +
19685 +out:
19686 +       return err;
19687 +}
19688 +
19689 +static int aufs_getattr(struct vfsmount *mnt __maybe_unused,
19690 +                       struct dentry *dentry, struct kstat *st)
19691 +{
19692 +       int err;
19693 +       unsigned char positive;
19694 +       struct path h_path;
19695 +       struct inode *inode;
19696 +       struct super_block *sb;
19697 +
19698 +       inode = d_inode(dentry);
19699 +       sb = dentry->d_sb;
19700 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19701 +       if (unlikely(err))
19702 +               goto out;
19703 +       err = au_h_path_getattr(dentry, /*force*/0, &h_path);
19704 +       if (unlikely(err))
19705 +               goto out_si;
19706 +       if (unlikely(!h_path.dentry))
19707 +               /* illegally overlapped or something */
19708 +               goto out_fill; /* pretending success */
19709 +
19710 +       positive = d_is_positive(h_path.dentry);
19711 +       if (positive)
19712 +               err = vfs_getattr(&h_path, st);
19713 +       if (!err) {
19714 +               if (positive)
19715 +                       au_refresh_iattr(inode, st,
19716 +                                        d_inode(h_path.dentry)->i_nlink);
19717 +               goto out_fill; /* success */
19718 +       }
19719 +       AuTraceErr(err);
19720 +       goto out_di;
19721 +
19722 +out_fill:
19723 +       generic_fillattr(inode, st);
19724 +out_di:
19725 +       di_read_unlock(dentry, AuLock_IR);
19726 +out_si:
19727 +       si_read_unlock(sb);
19728 +out:
19729 +       AuTraceErr(err);
19730 +       return err;
19731 +}
19732 +
19733 +/* ---------------------------------------------------------------------- */
19734 +
19735 +static const char *aufs_get_link(struct dentry *dentry, struct inode *inode,
19736 +                                struct delayed_call *done)
19737 +{
19738 +       const char *ret;
19739 +       struct dentry *h_dentry;
19740 +       struct inode *h_inode;
19741 +       int err;
19742 +       aufs_bindex_t bindex;
19743 +
19744 +       ret = NULL; /* suppress a warning */
19745 +       err = -ECHILD;
19746 +       if (!dentry)
19747 +               goto out;
19748 +
19749 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
19750 +       if (unlikely(err))
19751 +               goto out;
19752 +
19753 +       err = au_d_hashed_positive(dentry);
19754 +       if (unlikely(err))
19755 +               goto out_unlock;
19756 +
19757 +       err = -EINVAL;
19758 +       inode = d_inode(dentry);
19759 +       bindex = au_ibtop(inode);
19760 +       h_inode = au_h_iptr(inode, bindex);
19761 +       if (unlikely(!h_inode->i_op->get_link))
19762 +               goto out_unlock;
19763 +
19764 +       err = -EBUSY;
19765 +       h_dentry = NULL;
19766 +       if (au_dbtop(dentry) <= bindex) {
19767 +               h_dentry = au_h_dptr(dentry, bindex);
19768 +               if (h_dentry)
19769 +                       dget(h_dentry);
19770 +       }
19771 +       if (!h_dentry) {
19772 +               h_dentry = d_find_any_alias(h_inode);
19773 +               if (IS_ERR(h_dentry)) {
19774 +                       err = PTR_ERR(h_dentry);
19775 +                       goto out_unlock;
19776 +               }
19777 +       }
19778 +       if (unlikely(!h_dentry))
19779 +               goto out_unlock;
19780 +
19781 +       err = 0;
19782 +       AuDbg("%pf\n", h_inode->i_op->get_link);
19783 +       AuDbgDentry(h_dentry);
19784 +       ret = h_inode->i_op->get_link(h_dentry, h_inode, done);
19785 +       dput(h_dentry);
19786 +       if (IS_ERR(ret))
19787 +               err = PTR_ERR(ret);
19788 +
19789 +out_unlock:
19790 +       aufs_read_unlock(dentry, AuLock_IR);
19791 +out:
19792 +       if (unlikely(err))
19793 +               ret = ERR_PTR(err);
19794 +       AuTraceErrPtr(ret);
19795 +       return ret;
19796 +}
19797 +
19798 +/* ---------------------------------------------------------------------- */
19799 +
19800 +static int au_is_special(struct inode *inode)
19801 +{
19802 +       return (inode->i_mode & (S_IFBLK | S_IFCHR | S_IFIFO | S_IFSOCK));
19803 +}
19804 +
19805 +static int aufs_update_time(struct inode *inode, struct timespec *ts, int flags)
19806 +{
19807 +       int err;
19808 +       aufs_bindex_t bindex;
19809 +       struct super_block *sb;
19810 +       struct inode *h_inode;
19811 +       struct vfsmount *h_mnt;
19812 +
19813 +       sb = inode->i_sb;
19814 +       WARN_ONCE((flags & S_ATIME) && !IS_NOATIME(inode),
19815 +                 "unexpected s_flags 0x%lx", sb->s_flags);
19816 +
19817 +       /* mmap_sem might be acquired already, cf. aufs_mmap() */
19818 +       lockdep_off();
19819 +       si_read_lock(sb, AuLock_FLUSH);
19820 +       ii_write_lock_child(inode);
19821 +       lockdep_on();
19822 +
19823 +       err = 0;
19824 +       bindex = au_ibtop(inode);
19825 +       h_inode = au_h_iptr(inode, bindex);
19826 +       if (!au_test_ro(sb, bindex, inode)) {
19827 +               h_mnt = au_sbr_mnt(sb, bindex);
19828 +               err = vfsub_mnt_want_write(h_mnt);
19829 +               if (!err) {
19830 +                       err = vfsub_update_time(h_inode, ts, flags);
19831 +                       vfsub_mnt_drop_write(h_mnt);
19832 +               }
19833 +       } else if (au_is_special(h_inode)) {
19834 +               /*
19835 +                * Never copy-up here.
19836 +                * These special files may already be opened and used for
19837 +                * communicating. If we copied it up, then the communication
19838 +                * would be corrupted.
19839 +                */
19840 +               AuWarn1("timestamps for i%lu are ignored "
19841 +                       "since it is on readonly branch (hi%lu).\n",
19842 +                       inode->i_ino, h_inode->i_ino);
19843 +       } else if (flags & ~S_ATIME) {
19844 +               err = -EIO;
19845 +               AuIOErr1("unexpected flags 0x%x\n", flags);
19846 +               AuDebugOn(1);
19847 +       }
19848 +
19849 +       lockdep_off();
19850 +       if (!err)
19851 +               au_cpup_attr_timesizes(inode);
19852 +       ii_write_unlock(inode);
19853 +       si_read_unlock(sb);
19854 +       lockdep_on();
19855 +
19856 +       if (!err && (flags & S_VERSION))
19857 +               inode_inc_iversion(inode);
19858 +
19859 +       return err;
19860 +}
19861 +
19862 +/* ---------------------------------------------------------------------- */
19863 +
19864 +/* no getattr version will be set by module.c:aufs_init() */
19865 +struct inode_operations aufs_iop_nogetattr[AuIop_Last],
19866 +       aufs_iop[] = {
19867 +       [AuIop_SYMLINK] = {
19868 +               .permission     = aufs_permission,
19869 +#ifdef CONFIG_FS_POSIX_ACL
19870 +               .get_acl        = aufs_get_acl,
19871 +               .set_acl        = aufs_set_acl, /* unsupport for symlink? */
19872 +#endif
19873 +
19874 +               .setattr        = aufs_setattr,
19875 +               .getattr        = aufs_getattr,
19876 +
19877 +#ifdef CONFIG_AUFS_XATTR
19878 +               .setxattr       = aufs_setxattr,
19879 +               .getxattr       = aufs_getxattr,
19880 +               .listxattr      = aufs_listxattr,
19881 +               .removexattr    = aufs_removexattr,
19882 +#endif
19883 +
19884 +               .readlink       = generic_readlink,
19885 +               .get_link       = aufs_get_link,
19886 +
19887 +               /* .update_time = aufs_update_time */
19888 +       },
19889 +       [AuIop_DIR] = {
19890 +               .create         = aufs_create,
19891 +               .lookup         = aufs_lookup,
19892 +               .link           = aufs_link,
19893 +               .unlink         = aufs_unlink,
19894 +               .symlink        = aufs_symlink,
19895 +               .mkdir          = aufs_mkdir,
19896 +               .rmdir          = aufs_rmdir,
19897 +               .mknod          = aufs_mknod,
19898 +               .rename         = aufs_rename,
19899 +
19900 +               .permission     = aufs_permission,
19901 +#ifdef CONFIG_FS_POSIX_ACL
19902 +               .get_acl        = aufs_get_acl,
19903 +               .set_acl        = aufs_set_acl,
19904 +#endif
19905 +
19906 +               .setattr        = aufs_setattr,
19907 +               .getattr        = aufs_getattr,
19908 +
19909 +#ifdef CONFIG_AUFS_XATTR
19910 +               .setxattr       = aufs_setxattr,
19911 +               .getxattr       = aufs_getxattr,
19912 +               .listxattr      = aufs_listxattr,
19913 +               .removexattr    = aufs_removexattr,
19914 +#endif
19915 +
19916 +               .update_time    = aufs_update_time,
19917 +               .atomic_open    = aufs_atomic_open,
19918 +               .tmpfile        = aufs_tmpfile
19919 +       },
19920 +       [AuIop_OTHER] = {
19921 +               .permission     = aufs_permission,
19922 +#ifdef CONFIG_FS_POSIX_ACL
19923 +               .get_acl        = aufs_get_acl,
19924 +               .set_acl        = aufs_set_acl,
19925 +#endif
19926 +
19927 +               .setattr        = aufs_setattr,
19928 +               .getattr        = aufs_getattr,
19929 +
19930 +#ifdef CONFIG_AUFS_XATTR
19931 +               .setxattr       = aufs_setxattr,
19932 +               .getxattr       = aufs_getxattr,
19933 +               .listxattr      = aufs_listxattr,
19934 +               .removexattr    = aufs_removexattr,
19935 +#endif
19936 +
19937 +               .update_time    = aufs_update_time
19938 +       }
19939 +};
19940 diff -urN /usr/share/empty/fs/aufs/i_op_del.c linux/fs/aufs/i_op_del.c
19941 --- /usr/share/empty/fs/aufs/i_op_del.c 1970-01-01 01:00:00.000000000 +0100
19942 +++ linux/fs/aufs/i_op_del.c    2016-10-09 16:55:36.492701639 +0200
19943 @@ -0,0 +1,511 @@
19944 +/*
19945 + * Copyright (C) 2005-2016 Junjiro R. Okajima
19946 + *
19947 + * This program, aufs is free software; you can redistribute it and/or modify
19948 + * it under the terms of the GNU General Public License as published by
19949 + * the Free Software Foundation; either version 2 of the License, or
19950 + * (at your option) any later version.
19951 + *
19952 + * This program is distributed in the hope that it will be useful,
19953 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19954 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19955 + * GNU General Public License for more details.
19956 + *
19957 + * You should have received a copy of the GNU General Public License
19958 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19959 + */
19960 +
19961 +/*
19962 + * inode operations (del entry)
19963 + */
19964 +
19965 +#include "aufs.h"
19966 +
19967 +/*
19968 + * decide if a new whiteout for @dentry is necessary or not.
19969 + * when it is necessary, prepare the parent dir for the upper branch whose
19970 + * branch index is @bcpup for creation. the actual creation of the whiteout will
19971 + * be done by caller.
19972 + * return value:
19973 + * 0: wh is unnecessary
19974 + * plus: wh is necessary
19975 + * minus: error
19976 + */
19977 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup)
19978 +{
19979 +       int need_wh, err;
19980 +       aufs_bindex_t btop;
19981 +       struct super_block *sb;
19982 +
19983 +       sb = dentry->d_sb;
19984 +       btop = au_dbtop(dentry);
19985 +       if (*bcpup < 0) {
19986 +               *bcpup = btop;
19987 +               if (au_test_ro(sb, btop, d_inode(dentry))) {
19988 +                       err = AuWbrCopyup(au_sbi(sb), dentry);
19989 +                       *bcpup = err;
19990 +                       if (unlikely(err < 0))
19991 +                               goto out;
19992 +               }
19993 +       } else
19994 +               AuDebugOn(btop < *bcpup
19995 +                         || au_test_ro(sb, *bcpup, d_inode(dentry)));
19996 +       AuDbg("bcpup %d, btop %d\n", *bcpup, btop);
19997 +
19998 +       if (*bcpup != btop) {
19999 +               err = au_cpup_dirs(dentry, *bcpup);
20000 +               if (unlikely(err))
20001 +                       goto out;
20002 +               need_wh = 1;
20003 +       } else {
20004 +               struct au_dinfo *dinfo, *tmp;
20005 +
20006 +               need_wh = -ENOMEM;
20007 +               dinfo = au_di(dentry);
20008 +               tmp = au_di_alloc(sb, AuLsc_DI_TMP);
20009 +               if (tmp) {
20010 +                       au_di_cp(tmp, dinfo);
20011 +                       au_di_swap(tmp, dinfo);
20012 +                       /* returns the number of positive dentries */
20013 +                       need_wh = au_lkup_dentry(dentry, btop + 1,
20014 +                                                /* AuLkup_IGNORE_PERM */ 0);
20015 +                       au_di_swap(tmp, dinfo);
20016 +                       au_rw_write_unlock(&tmp->di_rwsem);
20017 +                       au_di_free(tmp);
20018 +               }
20019 +       }
20020 +       AuDbg("need_wh %d\n", need_wh);
20021 +       err = need_wh;
20022 +
20023 +out:
20024 +       return err;
20025 +}
20026 +
20027 +/*
20028 + * simple tests for the del-entry operations.
20029 + * following the checks in vfs, plus the parent-child relationship.
20030 + */
20031 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
20032 +              struct dentry *h_parent, int isdir)
20033 +{
20034 +       int err;
20035 +       umode_t h_mode;
20036 +       struct dentry *h_dentry, *h_latest;
20037 +       struct inode *h_inode;
20038 +
20039 +       h_dentry = au_h_dptr(dentry, bindex);
20040 +       if (d_really_is_positive(dentry)) {
20041 +               err = -ENOENT;
20042 +               if (unlikely(d_is_negative(h_dentry)))
20043 +                       goto out;
20044 +               h_inode = d_inode(h_dentry);
20045 +               if (unlikely(!h_inode->i_nlink))
20046 +                       goto out;
20047 +
20048 +               h_mode = h_inode->i_mode;
20049 +               if (!isdir) {
20050 +                       err = -EISDIR;
20051 +                       if (unlikely(S_ISDIR(h_mode)))
20052 +                               goto out;
20053 +               } else if (unlikely(!S_ISDIR(h_mode))) {
20054 +                       err = -ENOTDIR;
20055 +                       goto out;
20056 +               }
20057 +       } else {
20058 +               /* rename(2) case */
20059 +               err = -EIO;
20060 +               if (unlikely(d_is_positive(h_dentry)))
20061 +                       goto out;
20062 +       }
20063 +
20064 +       err = -ENOENT;
20065 +       /* expected parent dir is locked */
20066 +       if (unlikely(h_parent != h_dentry->d_parent))
20067 +               goto out;
20068 +       err = 0;
20069 +
20070 +       /*
20071 +        * rmdir a dir may break the consistency on some filesystem.
20072 +        * let's try heavy test.
20073 +        */
20074 +       err = -EACCES;
20075 +       if (unlikely(!au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1)
20076 +                    && au_test_h_perm(d_inode(h_parent),
20077 +                                      MAY_EXEC | MAY_WRITE)))
20078 +               goto out;
20079 +
20080 +       h_latest = au_sio_lkup_one(&dentry->d_name, h_parent);
20081 +       err = -EIO;
20082 +       if (IS_ERR(h_latest))
20083 +               goto out;
20084 +       if (h_latest == h_dentry)
20085 +               err = 0;
20086 +       dput(h_latest);
20087 +
20088 +out:
20089 +       return err;
20090 +}
20091 +
20092 +/*
20093 + * decide the branch where we operate for @dentry. the branch index will be set
20094 + * @rbcpup. after diciding it, 'pin' it and store the timestamps of the parent
20095 + * dir for reverting.
20096 + * when a new whiteout is necessary, create it.
20097 + */
20098 +static struct dentry*
20099 +lock_hdir_create_wh(struct dentry *dentry, int isdir, aufs_bindex_t *rbcpup,
20100 +                   struct au_dtime *dt, struct au_pin *pin)
20101 +{
20102 +       struct dentry *wh_dentry;
20103 +       struct super_block *sb;
20104 +       struct path h_path;
20105 +       int err, need_wh;
20106 +       unsigned int udba;
20107 +       aufs_bindex_t bcpup;
20108 +
20109 +       need_wh = au_wr_dir_need_wh(dentry, isdir, rbcpup);
20110 +       wh_dentry = ERR_PTR(need_wh);
20111 +       if (unlikely(need_wh < 0))
20112 +               goto out;
20113 +
20114 +       sb = dentry->d_sb;
20115 +       udba = au_opt_udba(sb);
20116 +       bcpup = *rbcpup;
20117 +       err = au_pin(pin, dentry, bcpup, udba,
20118 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
20119 +       wh_dentry = ERR_PTR(err);
20120 +       if (unlikely(err))
20121 +               goto out;
20122 +
20123 +       h_path.dentry = au_pinned_h_parent(pin);
20124 +       if (udba != AuOpt_UDBA_NONE
20125 +           && au_dbtop(dentry) == bcpup) {
20126 +               err = au_may_del(dentry, bcpup, h_path.dentry, isdir);
20127 +               wh_dentry = ERR_PTR(err);
20128 +               if (unlikely(err))
20129 +                       goto out_unpin;
20130 +       }
20131 +
20132 +       h_path.mnt = au_sbr_mnt(sb, bcpup);
20133 +       au_dtime_store(dt, au_pinned_parent(pin), &h_path);
20134 +       wh_dentry = NULL;
20135 +       if (!need_wh)
20136 +               goto out; /* success, no need to create whiteout */
20137 +
20138 +       wh_dentry = au_wh_create(dentry, bcpup, h_path.dentry);
20139 +       if (IS_ERR(wh_dentry))
20140 +               goto out_unpin;
20141 +
20142 +       /* returns with the parent is locked and wh_dentry is dget-ed */
20143 +       goto out; /* success */
20144 +
20145 +out_unpin:
20146 +       au_unpin(pin);
20147 +out:
20148 +       return wh_dentry;
20149 +}
20150 +
20151 +/*
20152 + * when removing a dir, rename it to a unique temporary whiteout-ed name first
20153 + * in order to be revertible and save time for removing many child whiteouts
20154 + * under the dir.
20155 + * returns 1 when there are too many child whiteout and caller should remove
20156 + * them asynchronously. returns 0 when the number of children is enough small to
20157 + * remove now or the branch fs is a remote fs.
20158 + * otherwise return an error.
20159 + */
20160 +static int renwh_and_rmdir(struct dentry *dentry, aufs_bindex_t bindex,
20161 +                          struct au_nhash *whlist, struct inode *dir)
20162 +{
20163 +       int rmdir_later, err, dirwh;
20164 +       struct dentry *h_dentry;
20165 +       struct super_block *sb;
20166 +       struct inode *inode;
20167 +
20168 +       sb = dentry->d_sb;
20169 +       SiMustAnyLock(sb);
20170 +       h_dentry = au_h_dptr(dentry, bindex);
20171 +       err = au_whtmp_ren(h_dentry, au_sbr(sb, bindex));
20172 +       if (unlikely(err))
20173 +               goto out;
20174 +
20175 +       /* stop monitoring */
20176 +       inode = d_inode(dentry);
20177 +       au_hn_free(au_hi(inode, bindex));
20178 +
20179 +       if (!au_test_fs_remote(h_dentry->d_sb)) {
20180 +               dirwh = au_sbi(sb)->si_dirwh;
20181 +               rmdir_later = (dirwh <= 1);
20182 +               if (!rmdir_later)
20183 +                       rmdir_later = au_nhash_test_longer_wh(whlist, bindex,
20184 +                                                             dirwh);
20185 +               if (rmdir_later)
20186 +                       return rmdir_later;
20187 +       }
20188 +
20189 +       err = au_whtmp_rmdir(dir, bindex, h_dentry, whlist);
20190 +       if (unlikely(err)) {
20191 +               AuIOErr("rmdir %pd, b%d failed, %d. ignored\n",
20192 +                       h_dentry, bindex, err);
20193 +               err = 0;
20194 +       }
20195 +
20196 +out:
20197 +       AuTraceErr(err);
20198 +       return err;
20199 +}
20200 +
20201 +/*
20202 + * final procedure for deleting a entry.
20203 + * maintain dentry and iattr.
20204 + */
20205 +static void epilog(struct inode *dir, struct dentry *dentry,
20206 +                  aufs_bindex_t bindex)
20207 +{
20208 +       struct inode *inode;
20209 +
20210 +       inode = d_inode(dentry);
20211 +       d_drop(dentry);
20212 +       inode->i_ctime = dir->i_ctime;
20213 +
20214 +       au_dir_ts(dir, bindex);
20215 +       dir->i_version++;
20216 +}
20217 +
20218 +/*
20219 + * when an error happened, remove the created whiteout and revert everything.
20220 + */
20221 +static int do_revert(int err, struct inode *dir, aufs_bindex_t bindex,
20222 +                    aufs_bindex_t bwh, struct dentry *wh_dentry,
20223 +                    struct dentry *dentry, struct au_dtime *dt)
20224 +{
20225 +       int rerr;
20226 +       struct path h_path = {
20227 +               .dentry = wh_dentry,
20228 +               .mnt    = au_sbr_mnt(dir->i_sb, bindex)
20229 +       };
20230 +
20231 +       rerr = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path, dentry);
20232 +       if (!rerr) {
20233 +               au_set_dbwh(dentry, bwh);
20234 +               au_dtime_revert(dt);
20235 +               return 0;
20236 +       }
20237 +
20238 +       AuIOErr("%pd reverting whiteout failed(%d, %d)\n", dentry, err, rerr);
20239 +       return -EIO;
20240 +}
20241 +
20242 +/* ---------------------------------------------------------------------- */
20243 +
20244 +int aufs_unlink(struct inode *dir, struct dentry *dentry)
20245 +{
20246 +       int err;
20247 +       aufs_bindex_t bwh, bindex, btop;
20248 +       struct inode *inode, *h_dir, *delegated;
20249 +       struct dentry *parent, *wh_dentry;
20250 +       /* to reuduce stack size */
20251 +       struct {
20252 +               struct au_dtime dt;
20253 +               struct au_pin pin;
20254 +               struct path h_path;
20255 +       } *a;
20256 +
20257 +       IMustLock(dir);
20258 +
20259 +       err = -ENOMEM;
20260 +       a = kmalloc(sizeof(*a), GFP_NOFS);
20261 +       if (unlikely(!a))
20262 +               goto out;
20263 +
20264 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
20265 +       if (unlikely(err))
20266 +               goto out_free;
20267 +       err = au_d_hashed_positive(dentry);
20268 +       if (unlikely(err))
20269 +               goto out_unlock;
20270 +       inode = d_inode(dentry);
20271 +       IMustLock(inode);
20272 +       err = -EISDIR;
20273 +       if (unlikely(d_is_dir(dentry)))
20274 +               goto out_unlock; /* possible? */
20275 +
20276 +       btop = au_dbtop(dentry);
20277 +       bwh = au_dbwh(dentry);
20278 +       bindex = -1;
20279 +       parent = dentry->d_parent; /* dir inode is locked */
20280 +       di_write_lock_parent(parent);
20281 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/0, &bindex, &a->dt,
20282 +                                       &a->pin);
20283 +       err = PTR_ERR(wh_dentry);
20284 +       if (IS_ERR(wh_dentry))
20285 +               goto out_parent;
20286 +
20287 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, btop);
20288 +       a->h_path.dentry = au_h_dptr(dentry, btop);
20289 +       dget(a->h_path.dentry);
20290 +       if (bindex == btop) {
20291 +               h_dir = au_pinned_h_dir(&a->pin);
20292 +               delegated = NULL;
20293 +               err = vfsub_unlink(h_dir, &a->h_path, &delegated, /*force*/0);
20294 +               if (unlikely(err == -EWOULDBLOCK)) {
20295 +                       pr_warn("cannot retry for NFSv4 delegation"
20296 +                               " for an internal unlink\n");
20297 +                       iput(delegated);
20298 +               }
20299 +       } else {
20300 +               /* dir inode is locked */
20301 +               h_dir = d_inode(wh_dentry->d_parent);
20302 +               IMustLock(h_dir);
20303 +               err = 0;
20304 +       }
20305 +
20306 +       if (!err) {
20307 +               vfsub_drop_nlink(inode);
20308 +               epilog(dir, dentry, bindex);
20309 +
20310 +               /* update target timestamps */
20311 +               if (bindex == btop) {
20312 +                       vfsub_update_h_iattr(&a->h_path, /*did*/NULL);
20313 +                       /*ignore*/
20314 +                       inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
20315 +               } else
20316 +                       /* todo: this timestamp may be reverted later */
20317 +                       inode->i_ctime = h_dir->i_ctime;
20318 +               goto out_unpin; /* success */
20319 +       }
20320 +
20321 +       /* revert */
20322 +       if (wh_dentry) {
20323 +               int rerr;
20324 +
20325 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
20326 +                                &a->dt);
20327 +               if (rerr)
20328 +                       err = rerr;
20329 +       }
20330 +
20331 +out_unpin:
20332 +       au_unpin(&a->pin);
20333 +       dput(wh_dentry);
20334 +       dput(a->h_path.dentry);
20335 +out_parent:
20336 +       di_write_unlock(parent);
20337 +out_unlock:
20338 +       aufs_read_unlock(dentry, AuLock_DW);
20339 +out_free:
20340 +       au_delayed_kfree(a);
20341 +out:
20342 +       return err;
20343 +}
20344 +
20345 +int aufs_rmdir(struct inode *dir, struct dentry *dentry)
20346 +{
20347 +       int err, rmdir_later;
20348 +       aufs_bindex_t bwh, bindex, btop;
20349 +       struct inode *inode;
20350 +       struct dentry *parent, *wh_dentry, *h_dentry;
20351 +       struct au_whtmp_rmdir *args;
20352 +       /* to reuduce stack size */
20353 +       struct {
20354 +               struct au_dtime dt;
20355 +               struct au_pin pin;
20356 +       } *a;
20357 +
20358 +       IMustLock(dir);
20359 +
20360 +       err = -ENOMEM;
20361 +       a = kmalloc(sizeof(*a), GFP_NOFS);
20362 +       if (unlikely(!a))
20363 +               goto out;
20364 +
20365 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
20366 +       if (unlikely(err))
20367 +               goto out_free;
20368 +       err = au_alive_dir(dentry);
20369 +       if (unlikely(err))
20370 +               goto out_unlock;
20371 +       inode = d_inode(dentry);
20372 +       IMustLock(inode);
20373 +       err = -ENOTDIR;
20374 +       if (unlikely(!d_is_dir(dentry)))
20375 +               goto out_unlock; /* possible? */
20376 +
20377 +       err = -ENOMEM;
20378 +       args = au_whtmp_rmdir_alloc(dir->i_sb, GFP_NOFS);
20379 +       if (unlikely(!args))
20380 +               goto out_unlock;
20381 +
20382 +       parent = dentry->d_parent; /* dir inode is locked */
20383 +       di_write_lock_parent(parent);
20384 +       err = au_test_empty(dentry, &args->whlist);
20385 +       if (unlikely(err))
20386 +               goto out_parent;
20387 +
20388 +       btop = au_dbtop(dentry);
20389 +       bwh = au_dbwh(dentry);
20390 +       bindex = -1;
20391 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/1, &bindex, &a->dt,
20392 +                                       &a->pin);
20393 +       err = PTR_ERR(wh_dentry);
20394 +       if (IS_ERR(wh_dentry))
20395 +               goto out_parent;
20396 +
20397 +       h_dentry = au_h_dptr(dentry, btop);
20398 +       dget(h_dentry);
20399 +       rmdir_later = 0;
20400 +       if (bindex == btop) {
20401 +               err = renwh_and_rmdir(dentry, btop, &args->whlist, dir);
20402 +               if (err > 0) {
20403 +                       rmdir_later = err;
20404 +                       err = 0;
20405 +               }
20406 +       } else {
20407 +               /* stop monitoring */
20408 +               au_hn_free(au_hi(inode, btop));
20409 +
20410 +               /* dir inode is locked */
20411 +               IMustLock(d_inode(wh_dentry->d_parent));
20412 +               err = 0;
20413 +       }
20414 +
20415 +       if (!err) {
20416 +               vfsub_dead_dir(inode);
20417 +               au_set_dbdiropq(dentry, -1);
20418 +               epilog(dir, dentry, bindex);
20419 +
20420 +               if (rmdir_later) {
20421 +                       au_whtmp_kick_rmdir(dir, btop, h_dentry, args);
20422 +                       args = NULL;
20423 +               }
20424 +
20425 +               goto out_unpin; /* success */
20426 +       }
20427 +
20428 +       /* revert */
20429 +       AuLabel(revert);
20430 +       if (wh_dentry) {
20431 +               int rerr;
20432 +
20433 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
20434 +                                &a->dt);
20435 +               if (rerr)
20436 +                       err = rerr;
20437 +       }
20438 +
20439 +out_unpin:
20440 +       au_unpin(&a->pin);
20441 +       dput(wh_dentry);
20442 +       dput(h_dentry);
20443 +out_parent:
20444 +       di_write_unlock(parent);
20445 +       if (args)
20446 +               au_whtmp_rmdir_free(args);
20447 +out_unlock:
20448 +       aufs_read_unlock(dentry, AuLock_DW);
20449 +out_free:
20450 +       au_delayed_kfree(a);
20451 +out:
20452 +       AuTraceErr(err);
20453 +       return err;
20454 +}
20455 diff -urN /usr/share/empty/fs/aufs/i_op_ren.c linux/fs/aufs/i_op_ren.c
20456 --- /usr/share/empty/fs/aufs/i_op_ren.c 1970-01-01 01:00:00.000000000 +0100
20457 +++ linux/fs/aufs/i_op_ren.c    2016-10-09 16:55:36.492701639 +0200
20458 @@ -0,0 +1,1015 @@
20459 +/*
20460 + * Copyright (C) 2005-2016 Junjiro R. Okajima
20461 + *
20462 + * This program, aufs is free software; you can redistribute it and/or modify
20463 + * it under the terms of the GNU General Public License as published by
20464 + * the Free Software Foundation; either version 2 of the License, or
20465 + * (at your option) any later version.
20466 + *
20467 + * This program is distributed in the hope that it will be useful,
20468 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
20469 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20470 + * GNU General Public License for more details.
20471 + *
20472 + * You should have received a copy of the GNU General Public License
20473 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20474 + */
20475 +
20476 +/*
20477 + * inode operation (rename entry)
20478 + * todo: this is crazy monster
20479 + */
20480 +
20481 +#include "aufs.h"
20482 +
20483 +enum { AuSRC, AuDST, AuSrcDst };
20484 +enum { AuPARENT, AuCHILD, AuParentChild };
20485 +
20486 +#define AuRen_ISDIR    1
20487 +#define AuRen_ISSAMEDIR        (1 << 1)
20488 +#define AuRen_WHSRC    (1 << 2)
20489 +#define AuRen_WHDST    (1 << 3)
20490 +#define AuRen_MNT_WRITE        (1 << 4)
20491 +#define AuRen_DT_DSTDIR        (1 << 5)
20492 +#define AuRen_DIROPQ   (1 << 6)
20493 +#define au_ftest_ren(flags, name)      ((flags) & AuRen_##name)
20494 +#define au_fset_ren(flags, name) \
20495 +       do { (flags) |= AuRen_##name; } while (0)
20496 +#define au_fclr_ren(flags, name) \
20497 +       do { (flags) &= ~AuRen_##name; } while (0)
20498 +
20499 +struct au_ren_args {
20500 +       struct {
20501 +               struct dentry *dentry, *h_dentry, *parent, *h_parent,
20502 +                       *wh_dentry;
20503 +               struct inode *dir, *inode;
20504 +               struct au_hinode *hdir;
20505 +               struct au_dtime dt[AuParentChild];
20506 +               aufs_bindex_t btop;
20507 +       } sd[AuSrcDst];
20508 +
20509 +#define src_dentry     sd[AuSRC].dentry
20510 +#define src_dir                sd[AuSRC].dir
20511 +#define src_inode      sd[AuSRC].inode
20512 +#define src_h_dentry   sd[AuSRC].h_dentry
20513 +#define src_parent     sd[AuSRC].parent
20514 +#define src_h_parent   sd[AuSRC].h_parent
20515 +#define src_wh_dentry  sd[AuSRC].wh_dentry
20516 +#define src_hdir       sd[AuSRC].hdir
20517 +#define src_h_dir      sd[AuSRC].hdir->hi_inode
20518 +#define src_dt         sd[AuSRC].dt
20519 +#define src_btop       sd[AuSRC].btop
20520 +
20521 +#define dst_dentry     sd[AuDST].dentry
20522 +#define dst_dir                sd[AuDST].dir
20523 +#define dst_inode      sd[AuDST].inode
20524 +#define dst_h_dentry   sd[AuDST].h_dentry
20525 +#define dst_parent     sd[AuDST].parent
20526 +#define dst_h_parent   sd[AuDST].h_parent
20527 +#define dst_wh_dentry  sd[AuDST].wh_dentry
20528 +#define dst_hdir       sd[AuDST].hdir
20529 +#define dst_h_dir      sd[AuDST].hdir->hi_inode
20530 +#define dst_dt         sd[AuDST].dt
20531 +#define dst_btop       sd[AuDST].btop
20532 +
20533 +       struct dentry *h_trap;
20534 +       struct au_branch *br;
20535 +       struct au_hinode *src_hinode;
20536 +       struct path h_path;
20537 +       struct au_nhash whlist;
20538 +       aufs_bindex_t btgt, src_bwh, src_bdiropq;
20539 +
20540 +       unsigned int flags;
20541 +
20542 +       struct au_whtmp_rmdir *thargs;
20543 +       struct dentry *h_dst;
20544 +};
20545 +
20546 +/* ---------------------------------------------------------------------- */
20547 +
20548 +/*
20549 + * functions for reverting.
20550 + * when an error happened in a single rename systemcall, we should revert
20551 + * everything as if nothing happened.
20552 + * we don't need to revert the copied-up/down the parent dir since they are
20553 + * harmless.
20554 + */
20555 +
20556 +#define RevertFailure(fmt, ...) do { \
20557 +       AuIOErr("revert failure: " fmt " (%d, %d)\n", \
20558 +               ##__VA_ARGS__, err, rerr); \
20559 +       err = -EIO; \
20560 +} while (0)
20561 +
20562 +static void au_ren_rev_diropq(int err, struct au_ren_args *a)
20563 +{
20564 +       int rerr;
20565 +
20566 +       au_hn_inode_lock_nested(a->src_hinode, AuLsc_I_CHILD);
20567 +       rerr = au_diropq_remove(a->src_dentry, a->btgt);
20568 +       au_hn_inode_unlock(a->src_hinode);
20569 +       au_set_dbdiropq(a->src_dentry, a->src_bdiropq);
20570 +       if (rerr)
20571 +               RevertFailure("remove diropq %pd", a->src_dentry);
20572 +}
20573 +
20574 +static void au_ren_rev_rename(int err, struct au_ren_args *a)
20575 +{
20576 +       int rerr;
20577 +       struct inode *delegated;
20578 +
20579 +       a->h_path.dentry = vfsub_lkup_one(&a->src_dentry->d_name,
20580 +                                         a->src_h_parent);
20581 +       rerr = PTR_ERR(a->h_path.dentry);
20582 +       if (IS_ERR(a->h_path.dentry)) {
20583 +               RevertFailure("lkup one %pd", a->src_dentry);
20584 +               return;
20585 +       }
20586 +
20587 +       delegated = NULL;
20588 +       rerr = vfsub_rename(a->dst_h_dir,
20589 +                           au_h_dptr(a->src_dentry, a->btgt),
20590 +                           a->src_h_dir, &a->h_path, &delegated);
20591 +       if (unlikely(rerr == -EWOULDBLOCK)) {
20592 +               pr_warn("cannot retry for NFSv4 delegation"
20593 +                       " for an internal rename\n");
20594 +               iput(delegated);
20595 +       }
20596 +       d_drop(a->h_path.dentry);
20597 +       dput(a->h_path.dentry);
20598 +       /* au_set_h_dptr(a->src_dentry, a->btgt, NULL); */
20599 +       if (rerr)
20600 +               RevertFailure("rename %pd", a->src_dentry);
20601 +}
20602 +
20603 +static void au_ren_rev_whtmp(int err, struct au_ren_args *a)
20604 +{
20605 +       int rerr;
20606 +       struct inode *delegated;
20607 +
20608 +       a->h_path.dentry = vfsub_lkup_one(&a->dst_dentry->d_name,
20609 +                                         a->dst_h_parent);
20610 +       rerr = PTR_ERR(a->h_path.dentry);
20611 +       if (IS_ERR(a->h_path.dentry)) {
20612 +               RevertFailure("lkup one %pd", a->dst_dentry);
20613 +               return;
20614 +       }
20615 +       if (d_is_positive(a->h_path.dentry)) {
20616 +               d_drop(a->h_path.dentry);
20617 +               dput(a->h_path.dentry);
20618 +               return;
20619 +       }
20620 +
20621 +       delegated = NULL;
20622 +       rerr = vfsub_rename(a->dst_h_dir, a->h_dst, a->dst_h_dir, &a->h_path,
20623 +                           &delegated);
20624 +       if (unlikely(rerr == -EWOULDBLOCK)) {
20625 +               pr_warn("cannot retry for NFSv4 delegation"
20626 +                       " for an internal rename\n");
20627 +               iput(delegated);
20628 +       }
20629 +       d_drop(a->h_path.dentry);
20630 +       dput(a->h_path.dentry);
20631 +       if (!rerr)
20632 +               au_set_h_dptr(a->dst_dentry, a->btgt, dget(a->h_dst));
20633 +       else
20634 +               RevertFailure("rename %pd", a->h_dst);
20635 +}
20636 +
20637 +static void au_ren_rev_whsrc(int err, struct au_ren_args *a)
20638 +{
20639 +       int rerr;
20640 +
20641 +       a->h_path.dentry = a->src_wh_dentry;
20642 +       rerr = au_wh_unlink_dentry(a->src_h_dir, &a->h_path, a->src_dentry);
20643 +       au_set_dbwh(a->src_dentry, a->src_bwh);
20644 +       if (rerr)
20645 +               RevertFailure("unlink %pd", a->src_wh_dentry);
20646 +}
20647 +#undef RevertFailure
20648 +
20649 +/* ---------------------------------------------------------------------- */
20650 +
20651 +/*
20652 + * when we have to copyup the renaming entry, do it with the rename-target name
20653 + * in order to minimize the cost (the later actual rename is unnecessary).
20654 + * otherwise rename it on the target branch.
20655 + */
20656 +static int au_ren_or_cpup(struct au_ren_args *a)
20657 +{
20658 +       int err;
20659 +       struct dentry *d;
20660 +       struct inode *delegated;
20661 +
20662 +       d = a->src_dentry;
20663 +       if (au_dbtop(d) == a->btgt) {
20664 +               a->h_path.dentry = a->dst_h_dentry;
20665 +               if (au_ftest_ren(a->flags, DIROPQ)
20666 +                   && au_dbdiropq(d) == a->btgt)
20667 +                       au_fclr_ren(a->flags, DIROPQ);
20668 +               AuDebugOn(au_dbtop(d) != a->btgt);
20669 +               delegated = NULL;
20670 +               err = vfsub_rename(a->src_h_dir, au_h_dptr(d, a->btgt),
20671 +                                  a->dst_h_dir, &a->h_path, &delegated);
20672 +               if (unlikely(err == -EWOULDBLOCK)) {
20673 +                       pr_warn("cannot retry for NFSv4 delegation"
20674 +                               " for an internal rename\n");
20675 +                       iput(delegated);
20676 +               }
20677 +       } else
20678 +               BUG();
20679 +
20680 +       if (!err && a->h_dst)
20681 +               /* it will be set to dinfo later */
20682 +               dget(a->h_dst);
20683 +
20684 +       return err;
20685 +}
20686 +
20687 +/* cf. aufs_rmdir() */
20688 +static int au_ren_del_whtmp(struct au_ren_args *a)
20689 +{
20690 +       int err;
20691 +       struct inode *dir;
20692 +
20693 +       dir = a->dst_dir;
20694 +       SiMustAnyLock(dir->i_sb);
20695 +       if (!au_nhash_test_longer_wh(&a->whlist, a->btgt,
20696 +                                    au_sbi(dir->i_sb)->si_dirwh)
20697 +           || au_test_fs_remote(a->h_dst->d_sb)) {
20698 +               err = au_whtmp_rmdir(dir, a->btgt, a->h_dst, &a->whlist);
20699 +               if (unlikely(err))
20700 +                       pr_warn("failed removing whtmp dir %pd (%d), "
20701 +                               "ignored.\n", a->h_dst, err);
20702 +       } else {
20703 +               au_nhash_wh_free(&a->thargs->whlist);
20704 +               a->thargs->whlist = a->whlist;
20705 +               a->whlist.nh_num = 0;
20706 +               au_whtmp_kick_rmdir(dir, a->btgt, a->h_dst, a->thargs);
20707 +               dput(a->h_dst);
20708 +               a->thargs = NULL;
20709 +       }
20710 +
20711 +       return 0;
20712 +}
20713 +
20714 +/* make it 'opaque' dir. */
20715 +static int au_ren_diropq(struct au_ren_args *a)
20716 +{
20717 +       int err;
20718 +       struct dentry *diropq;
20719 +
20720 +       err = 0;
20721 +       a->src_bdiropq = au_dbdiropq(a->src_dentry);
20722 +       a->src_hinode = au_hi(a->src_inode, a->btgt);
20723 +       au_hn_inode_lock_nested(a->src_hinode, AuLsc_I_CHILD);
20724 +       diropq = au_diropq_create(a->src_dentry, a->btgt);
20725 +       au_hn_inode_unlock(a->src_hinode);
20726 +       if (IS_ERR(diropq))
20727 +               err = PTR_ERR(diropq);
20728 +       else
20729 +               dput(diropq);
20730 +
20731 +       return err;
20732 +}
20733 +
20734 +static int do_rename(struct au_ren_args *a)
20735 +{
20736 +       int err;
20737 +       struct dentry *d, *h_d;
20738 +
20739 +       /* prepare workqueue args for asynchronous rmdir */
20740 +       h_d = a->dst_h_dentry;
20741 +       if (au_ftest_ren(a->flags, ISDIR) && d_is_positive(h_d)) {
20742 +               err = -ENOMEM;
20743 +               a->thargs = au_whtmp_rmdir_alloc(a->src_dentry->d_sb, GFP_NOFS);
20744 +               if (unlikely(!a->thargs))
20745 +                       goto out;
20746 +               a->h_dst = dget(h_d);
20747 +       }
20748 +
20749 +       /* create whiteout for src_dentry */
20750 +       if (au_ftest_ren(a->flags, WHSRC)) {
20751 +               a->src_bwh = au_dbwh(a->src_dentry);
20752 +               AuDebugOn(a->src_bwh >= 0);
20753 +               a->src_wh_dentry
20754 +                       = au_wh_create(a->src_dentry, a->btgt, a->src_h_parent);
20755 +               err = PTR_ERR(a->src_wh_dentry);
20756 +               if (IS_ERR(a->src_wh_dentry))
20757 +                       goto out_thargs;
20758 +       }
20759 +
20760 +       /* lookup whiteout for dentry */
20761 +       if (au_ftest_ren(a->flags, WHDST)) {
20762 +               h_d = au_wh_lkup(a->dst_h_parent, &a->dst_dentry->d_name,
20763 +                                a->br);
20764 +               err = PTR_ERR(h_d);
20765 +               if (IS_ERR(h_d))
20766 +                       goto out_whsrc;
20767 +               if (d_is_negative(h_d))
20768 +                       dput(h_d);
20769 +               else
20770 +                       a->dst_wh_dentry = h_d;
20771 +       }
20772 +
20773 +       /* rename dentry to tmpwh */
20774 +       if (a->thargs) {
20775 +               err = au_whtmp_ren(a->dst_h_dentry, a->br);
20776 +               if (unlikely(err))
20777 +                       goto out_whdst;
20778 +
20779 +               d = a->dst_dentry;
20780 +               au_set_h_dptr(d, a->btgt, NULL);
20781 +               err = au_lkup_neg(d, a->btgt, /*wh*/0);
20782 +               if (unlikely(err))
20783 +                       goto out_whtmp;
20784 +               a->dst_h_dentry = au_h_dptr(d, a->btgt);
20785 +       }
20786 +
20787 +       BUG_ON(d_is_positive(a->dst_h_dentry) && a->src_btop != a->btgt);
20788 +
20789 +       /* rename by vfs_rename or cpup */
20790 +       d = a->dst_dentry;
20791 +       if (au_ftest_ren(a->flags, ISDIR)
20792 +           && (a->dst_wh_dentry
20793 +               || au_dbdiropq(d) == a->btgt
20794 +               /* hide the lower to keep xino */
20795 +               || a->btgt < au_dbbot(d)
20796 +               || au_opt_test(au_mntflags(d->d_sb), ALWAYS_DIROPQ)))
20797 +               au_fset_ren(a->flags, DIROPQ);
20798 +       err = au_ren_or_cpup(a);
20799 +       if (unlikely(err))
20800 +               /* leave the copied-up one */
20801 +               goto out_whtmp;
20802 +
20803 +       /* make dir opaque */
20804 +       if (au_ftest_ren(a->flags, DIROPQ)) {
20805 +               err = au_ren_diropq(a);
20806 +               if (unlikely(err))
20807 +                       goto out_rename;
20808 +       }
20809 +
20810 +       /* update target timestamps */
20811 +       AuDebugOn(au_dbtop(a->src_dentry) != a->btgt);
20812 +       a->h_path.dentry = au_h_dptr(a->src_dentry, a->btgt);
20813 +       vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
20814 +       a->src_inode->i_ctime = d_inode(a->h_path.dentry)->i_ctime;
20815 +
20816 +       /* remove whiteout for dentry */
20817 +       if (a->dst_wh_dentry) {
20818 +               a->h_path.dentry = a->dst_wh_dentry;
20819 +               err = au_wh_unlink_dentry(a->dst_h_dir, &a->h_path,
20820 +                                         a->dst_dentry);
20821 +               if (unlikely(err))
20822 +                       goto out_diropq;
20823 +       }
20824 +
20825 +       /* remove whtmp */
20826 +       if (a->thargs)
20827 +               au_ren_del_whtmp(a); /* ignore this error */
20828 +
20829 +       au_fhsm_wrote(a->src_dentry->d_sb, a->btgt, /*force*/0);
20830 +       err = 0;
20831 +       goto out_success;
20832 +
20833 +out_diropq:
20834 +       if (au_ftest_ren(a->flags, DIROPQ))
20835 +               au_ren_rev_diropq(err, a);
20836 +out_rename:
20837 +       au_ren_rev_rename(err, a);
20838 +       dput(a->h_dst);
20839 +out_whtmp:
20840 +       if (a->thargs)
20841 +               au_ren_rev_whtmp(err, a);
20842 +out_whdst:
20843 +       dput(a->dst_wh_dentry);
20844 +       a->dst_wh_dentry = NULL;
20845 +out_whsrc:
20846 +       if (a->src_wh_dentry)
20847 +               au_ren_rev_whsrc(err, a);
20848 +out_success:
20849 +       dput(a->src_wh_dentry);
20850 +       dput(a->dst_wh_dentry);
20851 +out_thargs:
20852 +       if (a->thargs) {
20853 +               dput(a->h_dst);
20854 +               au_whtmp_rmdir_free(a->thargs);
20855 +               a->thargs = NULL;
20856 +       }
20857 +out:
20858 +       return err;
20859 +}
20860 +
20861 +/* ---------------------------------------------------------------------- */
20862 +
20863 +/*
20864 + * test if @dentry dir can be rename destination or not.
20865 + * success means, it is a logically empty dir.
20866 + */
20867 +static int may_rename_dstdir(struct dentry *dentry, struct au_nhash *whlist)
20868 +{
20869 +       return au_test_empty(dentry, whlist);
20870 +}
20871 +
20872 +/*
20873 + * test if @dentry dir can be rename source or not.
20874 + * if it can, return 0 and @children is filled.
20875 + * success means,
20876 + * - it is a logically empty dir.
20877 + * - or, it exists on writable branch and has no children including whiteouts
20878 + *       on the lower branch.
20879 + */
20880 +static int may_rename_srcdir(struct dentry *dentry, aufs_bindex_t btgt)
20881 +{
20882 +       int err;
20883 +       unsigned int rdhash;
20884 +       aufs_bindex_t btop;
20885 +
20886 +       btop = au_dbtop(dentry);
20887 +       if (btop != btgt) {
20888 +               struct au_nhash whlist;
20889 +
20890 +               SiMustAnyLock(dentry->d_sb);
20891 +               rdhash = au_sbi(dentry->d_sb)->si_rdhash;
20892 +               if (!rdhash)
20893 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL,
20894 +                                                          dentry));
20895 +               err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
20896 +               if (unlikely(err))
20897 +                       goto out;
20898 +               err = au_test_empty(dentry, &whlist);
20899 +               au_nhash_wh_free(&whlist);
20900 +               goto out;
20901 +       }
20902 +
20903 +       if (btop == au_dbtaildir(dentry))
20904 +               return 0; /* success */
20905 +
20906 +       err = au_test_empty_lower(dentry);
20907 +
20908 +out:
20909 +       if (err == -ENOTEMPTY) {
20910 +               AuWarn1("renaming dir who has child(ren) on multiple branches,"
20911 +                       " is not supported\n");
20912 +               err = -EXDEV;
20913 +       }
20914 +       return err;
20915 +}
20916 +
20917 +/* side effect: sets whlist and h_dentry */
20918 +static int au_ren_may_dir(struct au_ren_args *a)
20919 +{
20920 +       int err;
20921 +       unsigned int rdhash;
20922 +       struct dentry *d;
20923 +
20924 +       d = a->dst_dentry;
20925 +       SiMustAnyLock(d->d_sb);
20926 +
20927 +       err = 0;
20928 +       if (au_ftest_ren(a->flags, ISDIR) && a->dst_inode) {
20929 +               rdhash = au_sbi(d->d_sb)->si_rdhash;
20930 +               if (!rdhash)
20931 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, d));
20932 +               err = au_nhash_alloc(&a->whlist, rdhash, GFP_NOFS);
20933 +               if (unlikely(err))
20934 +                       goto out;
20935 +
20936 +               au_set_dbtop(d, a->dst_btop);
20937 +               err = may_rename_dstdir(d, &a->whlist);
20938 +               au_set_dbtop(d, a->btgt);
20939 +       }
20940 +       a->dst_h_dentry = au_h_dptr(d, au_dbtop(d));
20941 +       if (unlikely(err))
20942 +               goto out;
20943 +
20944 +       d = a->src_dentry;
20945 +       a->src_h_dentry = au_h_dptr(d, au_dbtop(d));
20946 +       if (au_ftest_ren(a->flags, ISDIR)) {
20947 +               err = may_rename_srcdir(d, a->btgt);
20948 +               if (unlikely(err)) {
20949 +                       au_nhash_wh_free(&a->whlist);
20950 +                       a->whlist.nh_num = 0;
20951 +               }
20952 +       }
20953 +out:
20954 +       return err;
20955 +}
20956 +
20957 +/* ---------------------------------------------------------------------- */
20958 +
20959 +/*
20960 + * simple tests for rename.
20961 + * following the checks in vfs, plus the parent-child relationship.
20962 + */
20963 +static int au_may_ren(struct au_ren_args *a)
20964 +{
20965 +       int err, isdir;
20966 +       struct inode *h_inode;
20967 +
20968 +       if (a->src_btop == a->btgt) {
20969 +               err = au_may_del(a->src_dentry, a->btgt, a->src_h_parent,
20970 +                                au_ftest_ren(a->flags, ISDIR));
20971 +               if (unlikely(err))
20972 +                       goto out;
20973 +               err = -EINVAL;
20974 +               if (unlikely(a->src_h_dentry == a->h_trap))
20975 +                       goto out;
20976 +       }
20977 +
20978 +       err = 0;
20979 +       if (a->dst_btop != a->btgt)
20980 +               goto out;
20981 +
20982 +       err = -ENOTEMPTY;
20983 +       if (unlikely(a->dst_h_dentry == a->h_trap))
20984 +               goto out;
20985 +
20986 +       err = -EIO;
20987 +       isdir = !!au_ftest_ren(a->flags, ISDIR);
20988 +       if (d_really_is_negative(a->dst_dentry)) {
20989 +               if (d_is_negative(a->dst_h_dentry))
20990 +                       err = au_may_add(a->dst_dentry, a->btgt,
20991 +                                        a->dst_h_parent, isdir);
20992 +       } else {
20993 +               if (unlikely(d_is_negative(a->dst_h_dentry)))
20994 +                       goto out;
20995 +               h_inode = d_inode(a->dst_h_dentry);
20996 +               if (h_inode->i_nlink)
20997 +                       err = au_may_del(a->dst_dentry, a->btgt,
20998 +                                        a->dst_h_parent, isdir);
20999 +       }
21000 +
21001 +out:
21002 +       if (unlikely(err == -ENOENT || err == -EEXIST))
21003 +               err = -EIO;
21004 +       AuTraceErr(err);
21005 +       return err;
21006 +}
21007 +
21008 +/* ---------------------------------------------------------------------- */
21009 +
21010 +/*
21011 + * locking order
21012 + * (VFS)
21013 + * - src_dir and dir by lock_rename()
21014 + * - inode if exitsts
21015 + * (aufs)
21016 + * - lock all
21017 + *   + src_dentry and dentry by aufs_read_and_write_lock2() which calls,
21018 + *     + si_read_lock
21019 + *     + di_write_lock2_child()
21020 + *       + di_write_lock_child()
21021 + *        + ii_write_lock_child()
21022 + *       + di_write_lock_child2()
21023 + *        + ii_write_lock_child2()
21024 + *     + src_parent and parent
21025 + *       + di_write_lock_parent()
21026 + *        + ii_write_lock_parent()
21027 + *       + di_write_lock_parent2()
21028 + *        + ii_write_lock_parent2()
21029 + *   + lower src_dir and dir by vfsub_lock_rename()
21030 + *   + verify the every relationships between child and parent. if any
21031 + *     of them failed, unlock all and return -EBUSY.
21032 + */
21033 +static void au_ren_unlock(struct au_ren_args *a)
21034 +{
21035 +       vfsub_unlock_rename(a->src_h_parent, a->src_hdir,
21036 +                           a->dst_h_parent, a->dst_hdir);
21037 +       if (au_ftest_ren(a->flags, MNT_WRITE))
21038 +               vfsub_mnt_drop_write(au_br_mnt(a->br));
21039 +}
21040 +
21041 +static int au_ren_lock(struct au_ren_args *a)
21042 +{
21043 +       int err;
21044 +       unsigned int udba;
21045 +
21046 +       err = 0;
21047 +       a->src_h_parent = au_h_dptr(a->src_parent, a->btgt);
21048 +       a->src_hdir = au_hi(a->src_dir, a->btgt);
21049 +       a->dst_h_parent = au_h_dptr(a->dst_parent, a->btgt);
21050 +       a->dst_hdir = au_hi(a->dst_dir, a->btgt);
21051 +
21052 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
21053 +       if (unlikely(err))
21054 +               goto out;
21055 +       au_fset_ren(a->flags, MNT_WRITE);
21056 +       a->h_trap = vfsub_lock_rename(a->src_h_parent, a->src_hdir,
21057 +                                     a->dst_h_parent, a->dst_hdir);
21058 +       udba = au_opt_udba(a->src_dentry->d_sb);
21059 +       if (unlikely(a->src_hdir->hi_inode != d_inode(a->src_h_parent)
21060 +                    || a->dst_hdir->hi_inode != d_inode(a->dst_h_parent)))
21061 +               err = au_busy_or_stale();
21062 +       if (!err && au_dbtop(a->src_dentry) == a->btgt)
21063 +               err = au_h_verify(a->src_h_dentry, udba,
21064 +                                 d_inode(a->src_h_parent), a->src_h_parent,
21065 +                                 a->br);
21066 +       if (!err && au_dbtop(a->dst_dentry) == a->btgt)
21067 +               err = au_h_verify(a->dst_h_dentry, udba,
21068 +                                 d_inode(a->dst_h_parent), a->dst_h_parent,
21069 +                                 a->br);
21070 +       if (!err)
21071 +               goto out; /* success */
21072 +
21073 +       err = au_busy_or_stale();
21074 +       au_ren_unlock(a);
21075 +
21076 +out:
21077 +       return err;
21078 +}
21079 +
21080 +/* ---------------------------------------------------------------------- */
21081 +
21082 +static void au_ren_refresh_dir(struct au_ren_args *a)
21083 +{
21084 +       struct inode *dir;
21085 +
21086 +       dir = a->dst_dir;
21087 +       dir->i_version++;
21088 +       if (au_ftest_ren(a->flags, ISDIR)) {
21089 +               /* is this updating defined in POSIX? */
21090 +               au_cpup_attr_timesizes(a->src_inode);
21091 +               au_cpup_attr_nlink(dir, /*force*/1);
21092 +       }
21093 +
21094 +       au_dir_ts(dir, a->btgt);
21095 +
21096 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
21097 +               return;
21098 +
21099 +       dir = a->src_dir;
21100 +       dir->i_version++;
21101 +       if (au_ftest_ren(a->flags, ISDIR))
21102 +               au_cpup_attr_nlink(dir, /*force*/1);
21103 +       au_dir_ts(dir, a->btgt);
21104 +}
21105 +
21106 +static void au_ren_refresh(struct au_ren_args *a)
21107 +{
21108 +       aufs_bindex_t bbot, bindex;
21109 +       struct dentry *d, *h_d;
21110 +       struct inode *i, *h_i;
21111 +       struct super_block *sb;
21112 +
21113 +       d = a->dst_dentry;
21114 +       d_drop(d);
21115 +       if (a->h_dst)
21116 +               /* already dget-ed by au_ren_or_cpup() */
21117 +               au_set_h_dptr(d, a->btgt, a->h_dst);
21118 +
21119 +       i = a->dst_inode;
21120 +       if (i) {
21121 +               if (!au_ftest_ren(a->flags, ISDIR))
21122 +                       vfsub_drop_nlink(i);
21123 +               else {
21124 +                       vfsub_dead_dir(i);
21125 +                       au_cpup_attr_timesizes(i);
21126 +               }
21127 +               au_update_dbrange(d, /*do_put_zero*/1);
21128 +       } else {
21129 +               bbot = a->btgt;
21130 +               for (bindex = au_dbtop(d); bindex < bbot; bindex++)
21131 +                       au_set_h_dptr(d, bindex, NULL);
21132 +               bbot = au_dbbot(d);
21133 +               for (bindex = a->btgt + 1; bindex <= bbot; bindex++)
21134 +                       au_set_h_dptr(d, bindex, NULL);
21135 +               au_update_dbrange(d, /*do_put_zero*/0);
21136 +       }
21137 +
21138 +       d = a->src_dentry;
21139 +       au_set_dbwh(d, -1);
21140 +       bbot = au_dbbot(d);
21141 +       for (bindex = a->btgt + 1; bindex <= bbot; bindex++) {
21142 +               h_d = au_h_dptr(d, bindex);
21143 +               if (h_d)
21144 +                       au_set_h_dptr(d, bindex, NULL);
21145 +       }
21146 +       au_set_dbbot(d, a->btgt);
21147 +
21148 +       sb = d->d_sb;
21149 +       i = a->src_inode;
21150 +       if (au_opt_test(au_mntflags(sb), PLINK) && au_plink_test(i))
21151 +               return; /* success */
21152 +
21153 +       bbot = au_ibbot(i);
21154 +       for (bindex = a->btgt + 1; bindex <= bbot; bindex++) {
21155 +               h_i = au_h_iptr(i, bindex);
21156 +               if (h_i) {
21157 +                       au_xino_write(sb, bindex, h_i->i_ino, /*ino*/0);
21158 +                       /* ignore this error */
21159 +                       au_set_h_iptr(i, bindex, NULL, 0);
21160 +               }
21161 +       }
21162 +       au_set_ibbot(i, a->btgt);
21163 +}
21164 +
21165 +/* ---------------------------------------------------------------------- */
21166 +
21167 +/* mainly for link(2) and rename(2) */
21168 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt)
21169 +{
21170 +       aufs_bindex_t bdiropq, bwh;
21171 +       struct dentry *parent;
21172 +       struct au_branch *br;
21173 +
21174 +       parent = dentry->d_parent;
21175 +       IMustLock(d_inode(parent)); /* dir is locked */
21176 +
21177 +       bdiropq = au_dbdiropq(parent);
21178 +       bwh = au_dbwh(dentry);
21179 +       br = au_sbr(dentry->d_sb, btgt);
21180 +       if (au_br_rdonly(br)
21181 +           || (0 <= bdiropq && bdiropq < btgt)
21182 +           || (0 <= bwh && bwh < btgt))
21183 +               btgt = -1;
21184 +
21185 +       AuDbg("btgt %d\n", btgt);
21186 +       return btgt;
21187 +}
21188 +
21189 +/* sets src_btop, dst_btop and btgt */
21190 +static int au_ren_wbr(struct au_ren_args *a)
21191 +{
21192 +       int err;
21193 +       struct au_wr_dir_args wr_dir_args = {
21194 +               /* .force_btgt  = -1, */
21195 +               .flags          = AuWrDir_ADD_ENTRY
21196 +       };
21197 +
21198 +       a->src_btop = au_dbtop(a->src_dentry);
21199 +       a->dst_btop = au_dbtop(a->dst_dentry);
21200 +       if (au_ftest_ren(a->flags, ISDIR))
21201 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
21202 +       wr_dir_args.force_btgt = a->src_btop;
21203 +       if (a->dst_inode && a->dst_btop < a->src_btop)
21204 +               wr_dir_args.force_btgt = a->dst_btop;
21205 +       wr_dir_args.force_btgt = au_wbr(a->dst_dentry, wr_dir_args.force_btgt);
21206 +       err = au_wr_dir(a->dst_dentry, a->src_dentry, &wr_dir_args);
21207 +       a->btgt = err;
21208 +
21209 +       return err;
21210 +}
21211 +
21212 +static void au_ren_dt(struct au_ren_args *a)
21213 +{
21214 +       a->h_path.dentry = a->src_h_parent;
21215 +       au_dtime_store(a->src_dt + AuPARENT, a->src_parent, &a->h_path);
21216 +       if (!au_ftest_ren(a->flags, ISSAMEDIR)) {
21217 +               a->h_path.dentry = a->dst_h_parent;
21218 +               au_dtime_store(a->dst_dt + AuPARENT, a->dst_parent, &a->h_path);
21219 +       }
21220 +
21221 +       au_fclr_ren(a->flags, DT_DSTDIR);
21222 +       if (!au_ftest_ren(a->flags, ISDIR))
21223 +               return;
21224 +
21225 +       a->h_path.dentry = a->src_h_dentry;
21226 +       au_dtime_store(a->src_dt + AuCHILD, a->src_dentry, &a->h_path);
21227 +       if (d_is_positive(a->dst_h_dentry)) {
21228 +               au_fset_ren(a->flags, DT_DSTDIR);
21229 +               a->h_path.dentry = a->dst_h_dentry;
21230 +               au_dtime_store(a->dst_dt + AuCHILD, a->dst_dentry, &a->h_path);
21231 +       }
21232 +}
21233 +
21234 +static void au_ren_rev_dt(int err, struct au_ren_args *a)
21235 +{
21236 +       struct dentry *h_d;
21237 +       struct inode *h_inode;
21238 +
21239 +       au_dtime_revert(a->src_dt + AuPARENT);
21240 +       if (!au_ftest_ren(a->flags, ISSAMEDIR))
21241 +               au_dtime_revert(a->dst_dt + AuPARENT);
21242 +
21243 +       if (au_ftest_ren(a->flags, ISDIR) && err != -EIO) {
21244 +               h_d = a->src_dt[AuCHILD].dt_h_path.dentry;
21245 +               h_inode = d_inode(h_d);
21246 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
21247 +               au_dtime_revert(a->src_dt + AuCHILD);
21248 +               inode_unlock(h_inode);
21249 +
21250 +               if (au_ftest_ren(a->flags, DT_DSTDIR)) {
21251 +                       h_d = a->dst_dt[AuCHILD].dt_h_path.dentry;
21252 +                       h_inode = d_inode(h_d);
21253 +                       inode_lock_nested(h_inode, AuLsc_I_CHILD);
21254 +                       au_dtime_revert(a->dst_dt + AuCHILD);
21255 +                       inode_unlock(h_inode);
21256 +               }
21257 +       }
21258 +}
21259 +
21260 +/* ---------------------------------------------------------------------- */
21261 +
21262 +int aufs_rename(struct inode *_src_dir, struct dentry *_src_dentry,
21263 +               struct inode *_dst_dir, struct dentry *_dst_dentry)
21264 +{
21265 +       int err, flags;
21266 +       /* reduce stack space */
21267 +       struct au_ren_args *a;
21268 +
21269 +       AuDbg("%pd, %pd\n", _src_dentry, _dst_dentry);
21270 +       IMustLock(_src_dir);
21271 +       IMustLock(_dst_dir);
21272 +
21273 +       err = -ENOMEM;
21274 +       BUILD_BUG_ON(sizeof(*a) > PAGE_SIZE);
21275 +       a = kzalloc(sizeof(*a), GFP_NOFS);
21276 +       if (unlikely(!a))
21277 +               goto out;
21278 +
21279 +       a->src_dir = _src_dir;
21280 +       a->src_dentry = _src_dentry;
21281 +       a->src_inode = NULL;
21282 +       if (d_really_is_positive(a->src_dentry))
21283 +               a->src_inode = d_inode(a->src_dentry);
21284 +       a->src_parent = a->src_dentry->d_parent; /* dir inode is locked */
21285 +       a->dst_dir = _dst_dir;
21286 +       a->dst_dentry = _dst_dentry;
21287 +       a->dst_inode = NULL;
21288 +       if (d_really_is_positive(a->dst_dentry))
21289 +               a->dst_inode = d_inode(a->dst_dentry);
21290 +       a->dst_parent = a->dst_dentry->d_parent; /* dir inode is locked */
21291 +       if (a->dst_inode) {
21292 +               IMustLock(a->dst_inode);
21293 +               au_igrab(a->dst_inode);
21294 +       }
21295 +
21296 +       err = -ENOTDIR;
21297 +       flags = AuLock_FLUSH | AuLock_NOPLM | AuLock_GEN;
21298 +       if (d_is_dir(a->src_dentry)) {
21299 +               au_fset_ren(a->flags, ISDIR);
21300 +               if (unlikely(d_really_is_positive(a->dst_dentry)
21301 +                            && !d_is_dir(a->dst_dentry)))
21302 +                       goto out_free;
21303 +               flags |= AuLock_DIRS;
21304 +       }
21305 +       err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry, flags);
21306 +       if (unlikely(err))
21307 +               goto out_free;
21308 +
21309 +       err = au_d_hashed_positive(a->src_dentry);
21310 +       if (unlikely(err))
21311 +               goto out_unlock;
21312 +       err = -ENOENT;
21313 +       if (a->dst_inode) {
21314 +               /*
21315 +                * If it is a dir, VFS unhash dst_dentry before this
21316 +                * function. It means we cannot rely upon d_unhashed().
21317 +                */
21318 +               if (unlikely(!a->dst_inode->i_nlink))
21319 +                       goto out_unlock;
21320 +               if (!S_ISDIR(a->dst_inode->i_mode)) {
21321 +                       err = au_d_hashed_positive(a->dst_dentry);
21322 +                       if (unlikely(err))
21323 +                               goto out_unlock;
21324 +               } else if (unlikely(IS_DEADDIR(a->dst_inode)))
21325 +                       goto out_unlock;
21326 +       } else if (unlikely(d_unhashed(a->dst_dentry)))
21327 +               goto out_unlock;
21328 +
21329 +       /*
21330 +        * is it possible?
21331 +        * yes, it happened (in linux-3.3-rcN) but I don't know why.
21332 +        * there may exist a problem somewhere else.
21333 +        */
21334 +       err = -EINVAL;
21335 +       if (unlikely(d_inode(a->dst_parent) == d_inode(a->src_dentry)))
21336 +               goto out_unlock;
21337 +
21338 +       au_fset_ren(a->flags, ISSAMEDIR); /* temporary */
21339 +       di_write_lock_parent(a->dst_parent);
21340 +
21341 +       /* which branch we process */
21342 +       err = au_ren_wbr(a);
21343 +       if (unlikely(err < 0))
21344 +               goto out_parent;
21345 +       a->br = au_sbr(a->dst_dentry->d_sb, a->btgt);
21346 +       a->h_path.mnt = au_br_mnt(a->br);
21347 +
21348 +       /* are they available to be renamed */
21349 +       err = au_ren_may_dir(a);
21350 +       if (unlikely(err))
21351 +               goto out_children;
21352 +
21353 +       /* prepare the writable parent dir on the same branch */
21354 +       if (a->dst_btop == a->btgt) {
21355 +               au_fset_ren(a->flags, WHDST);
21356 +       } else {
21357 +               err = au_cpup_dirs(a->dst_dentry, a->btgt);
21358 +               if (unlikely(err))
21359 +                       goto out_children;
21360 +       }
21361 +
21362 +       if (a->src_dir != a->dst_dir) {
21363 +               /*
21364 +                * this temporary unlock is safe,
21365 +                * because both dir->i_mutex are locked.
21366 +                */
21367 +               di_write_unlock(a->dst_parent);
21368 +               di_write_lock_parent(a->src_parent);
21369 +               err = au_wr_dir_need_wh(a->src_dentry,
21370 +                                       au_ftest_ren(a->flags, ISDIR),
21371 +                                       &a->btgt);
21372 +               di_write_unlock(a->src_parent);
21373 +               di_write_lock2_parent(a->src_parent, a->dst_parent, /*isdir*/1);
21374 +               au_fclr_ren(a->flags, ISSAMEDIR);
21375 +       } else
21376 +               err = au_wr_dir_need_wh(a->src_dentry,
21377 +                                       au_ftest_ren(a->flags, ISDIR),
21378 +                                       &a->btgt);
21379 +       if (unlikely(err < 0))
21380 +               goto out_children;
21381 +       if (err)
21382 +               au_fset_ren(a->flags, WHSRC);
21383 +
21384 +       /* cpup src */
21385 +       if (a->src_btop != a->btgt) {
21386 +               struct au_pin pin;
21387 +
21388 +               err = au_pin(&pin, a->src_dentry, a->btgt,
21389 +                            au_opt_udba(a->src_dentry->d_sb),
21390 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
21391 +               if (!err) {
21392 +                       struct au_cp_generic cpg = {
21393 +                               .dentry = a->src_dentry,
21394 +                               .bdst   = a->btgt,
21395 +                               .bsrc   = a->src_btop,
21396 +                               .len    = -1,
21397 +                               .pin    = &pin,
21398 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
21399 +                       };
21400 +                       AuDebugOn(au_dbtop(a->src_dentry) != a->src_btop);
21401 +                       err = au_sio_cpup_simple(&cpg);
21402 +                       au_unpin(&pin);
21403 +               }
21404 +               if (unlikely(err))
21405 +                       goto out_children;
21406 +               a->src_btop = a->btgt;
21407 +               a->src_h_dentry = au_h_dptr(a->src_dentry, a->btgt);
21408 +               au_fset_ren(a->flags, WHSRC);
21409 +       }
21410 +
21411 +       /* lock them all */
21412 +       err = au_ren_lock(a);
21413 +       if (unlikely(err))
21414 +               /* leave the copied-up one */
21415 +               goto out_children;
21416 +
21417 +       if (!au_opt_test(au_mntflags(a->dst_dir->i_sb), UDBA_NONE))
21418 +               err = au_may_ren(a);
21419 +       else if (unlikely(a->dst_dentry->d_name.len > AUFS_MAX_NAMELEN))
21420 +               err = -ENAMETOOLONG;
21421 +       if (unlikely(err))
21422 +               goto out_hdir;
21423 +
21424 +       /* store timestamps to be revertible */
21425 +       au_ren_dt(a);
21426 +
21427 +       /* here we go */
21428 +       err = do_rename(a);
21429 +       if (unlikely(err))
21430 +               goto out_dt;
21431 +
21432 +       /* update dir attributes */
21433 +       au_ren_refresh_dir(a);
21434 +
21435 +       /* dput/iput all lower dentries */
21436 +       au_ren_refresh(a);
21437 +
21438 +       goto out_hdir; /* success */
21439 +
21440 +out_dt:
21441 +       au_ren_rev_dt(err, a);
21442 +out_hdir:
21443 +       au_ren_unlock(a);
21444 +out_children:
21445 +       au_nhash_wh_free(&a->whlist);
21446 +       if (err && a->dst_inode && a->dst_btop != a->btgt) {
21447 +               AuDbg("btop %d, btgt %d\n", a->dst_btop, a->btgt);
21448 +               au_set_h_dptr(a->dst_dentry, a->btgt, NULL);
21449 +               au_set_dbtop(a->dst_dentry, a->dst_btop);
21450 +       }
21451 +out_parent:
21452 +       if (!err)
21453 +               d_move(a->src_dentry, a->dst_dentry);
21454 +       else {
21455 +               au_update_dbtop(a->dst_dentry);
21456 +               if (!a->dst_inode)
21457 +                       d_drop(a->dst_dentry);
21458 +       }
21459 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
21460 +               di_write_unlock(a->dst_parent);
21461 +       else
21462 +               di_write_unlock2(a->src_parent, a->dst_parent);
21463 +out_unlock:
21464 +       aufs_read_and_write_unlock2(a->dst_dentry, a->src_dentry);
21465 +out_free:
21466 +       iput(a->dst_inode);
21467 +       if (a->thargs)
21468 +               au_whtmp_rmdir_free(a->thargs);
21469 +       au_delayed_kfree(a);
21470 +out:
21471 +       AuTraceErr(err);
21472 +       return err;
21473 +}
21474 diff -urN /usr/share/empty/fs/aufs/Kconfig linux/fs/aufs/Kconfig
21475 --- /usr/share/empty/fs/aufs/Kconfig    1970-01-01 01:00:00.000000000 +0100
21476 +++ linux/fs/aufs/Kconfig       2016-10-09 16:55:36.482701377 +0200
21477 @@ -0,0 +1,185 @@
21478 +config AUFS_FS
21479 +       tristate "Aufs (Advanced multi layered unification filesystem) support"
21480 +       help
21481 +       Aufs is a stackable unification filesystem such as Unionfs,
21482 +       which unifies several directories and provides a merged single
21483 +       directory.
21484 +       In the early days, aufs was entirely re-designed and
21485 +       re-implemented Unionfs Version 1.x series. Introducing many
21486 +       original ideas, approaches and improvements, it becomes totally
21487 +       different from Unionfs while keeping the basic features.
21488 +
21489 +if AUFS_FS
21490 +choice
21491 +       prompt "Maximum number of branches"
21492 +       default AUFS_BRANCH_MAX_127
21493 +       help
21494 +       Specifies the maximum number of branches (or member directories)
21495 +       in a single aufs. The larger value consumes more system
21496 +       resources and has a minor impact to performance.
21497 +config AUFS_BRANCH_MAX_127
21498 +       bool "127"
21499 +       help
21500 +       Specifies the maximum number of branches (or member directories)
21501 +       in a single aufs. The larger value consumes more system
21502 +       resources and has a minor impact to performance.
21503 +config AUFS_BRANCH_MAX_511
21504 +       bool "511"
21505 +       help
21506 +       Specifies the maximum number of branches (or member directories)
21507 +       in a single aufs. The larger value consumes more system
21508 +       resources and has a minor impact to performance.
21509 +config AUFS_BRANCH_MAX_1023
21510 +       bool "1023"
21511 +       help
21512 +       Specifies the maximum number of branches (or member directories)
21513 +       in a single aufs. The larger value consumes more system
21514 +       resources and has a minor impact to performance.
21515 +config AUFS_BRANCH_MAX_32767
21516 +       bool "32767"
21517 +       help
21518 +       Specifies the maximum number of branches (or member directories)
21519 +       in a single aufs. The larger value consumes more system
21520 +       resources and has a minor impact to performance.
21521 +endchoice
21522 +
21523 +config AUFS_SBILIST
21524 +       bool
21525 +       depends on AUFS_MAGIC_SYSRQ || PROC_FS
21526 +       default y
21527 +       help
21528 +       Automatic configuration for internal use.
21529 +       When aufs supports Magic SysRq or /proc, enabled automatically.
21530 +
21531 +config AUFS_HNOTIFY
21532 +       bool "Detect direct branch access (bypassing aufs)"
21533 +       help
21534 +       If you want to modify files on branches directly, eg. bypassing aufs,
21535 +       and want aufs to detect the changes of them fully, then enable this
21536 +       option and use 'udba=notify' mount option.
21537 +       Currently there is only one available configuration, "fsnotify".
21538 +       It will have a negative impact to the performance.
21539 +       See detail in aufs.5.
21540 +
21541 +choice
21542 +       prompt "method" if AUFS_HNOTIFY
21543 +       default AUFS_HFSNOTIFY
21544 +config AUFS_HFSNOTIFY
21545 +       bool "fsnotify"
21546 +       select FSNOTIFY
21547 +endchoice
21548 +
21549 +config AUFS_EXPORT
21550 +       bool "NFS-exportable aufs"
21551 +       depends on EXPORTFS
21552 +       help
21553 +       If you want to export your mounted aufs via NFS, then enable this
21554 +       option. There are several requirements for this configuration.
21555 +       See detail in aufs.5.
21556 +
21557 +config AUFS_INO_T_64
21558 +       bool
21559 +       depends on AUFS_EXPORT
21560 +       depends on 64BIT && !(ALPHA || S390)
21561 +       default y
21562 +       help
21563 +       Automatic configuration for internal use.
21564 +       /* typedef unsigned long/int __kernel_ino_t */
21565 +       /* alpha and s390x are int */
21566 +
21567 +config AUFS_XATTR
21568 +       bool "support for XATTR/EA (including Security Labels)"
21569 +       help
21570 +       If your branch fs supports XATTR/EA and you want to make them
21571 +       available in aufs too, then enable this opsion and specify the
21572 +       branch attributes for EA.
21573 +       See detail in aufs.5.
21574 +
21575 +config AUFS_FHSM
21576 +       bool "File-based Hierarchical Storage Management"
21577 +       help
21578 +       Hierarchical Storage Management (or HSM) is a well-known feature
21579 +       in the storage world. Aufs provides this feature as file-based.
21580 +       with multiple branches.
21581 +       These multiple branches are prioritized, ie. the topmost one
21582 +       should be the fastest drive and be used heavily.
21583 +
21584 +config AUFS_RDU
21585 +       bool "Readdir in userspace"
21586 +       help
21587 +       Aufs has two methods to provide a merged view for a directory,
21588 +       by a user-space library and by kernel-space natively. The latter
21589 +       is always enabled but sometimes large and slow.
21590 +       If you enable this option, install the library in aufs2-util
21591 +       package, and set some environment variables for your readdir(3),
21592 +       then the work will be handled in user-space which generally
21593 +       shows better performance in most cases.
21594 +       See detail in aufs.5.
21595 +
21596 +config AUFS_SHWH
21597 +       bool "Show whiteouts"
21598 +       help
21599 +       If you want to make the whiteouts in aufs visible, then enable
21600 +       this option and specify 'shwh' mount option. Although it may
21601 +       sounds like philosophy or something, but in technically it
21602 +       simply shows the name of whiteout with keeping its behaviour.
21603 +
21604 +config AUFS_BR_RAMFS
21605 +       bool "Ramfs (initramfs/rootfs) as an aufs branch"
21606 +       help
21607 +       If you want to use ramfs as an aufs branch fs, then enable this
21608 +       option. Generally tmpfs is recommended.
21609 +       Aufs prohibited them to be a branch fs by default, because
21610 +       initramfs becomes unusable after switch_root or something
21611 +       generally. If you sets initramfs as an aufs branch and boot your
21612 +       system by switch_root, you will meet a problem easily since the
21613 +       files in initramfs may be inaccessible.
21614 +       Unless you are going to use ramfs as an aufs branch fs without
21615 +       switch_root or something, leave it N.
21616 +
21617 +config AUFS_BR_FUSE
21618 +       bool "Fuse fs as an aufs branch"
21619 +       depends on FUSE_FS
21620 +       select AUFS_POLL
21621 +       help
21622 +       If you want to use fuse-based userspace filesystem as an aufs
21623 +       branch fs, then enable this option.
21624 +       It implements the internal poll(2) operation which is
21625 +       implemented by fuse only (curretnly).
21626 +
21627 +config AUFS_POLL
21628 +       bool
21629 +       help
21630 +       Automatic configuration for internal use.
21631 +
21632 +config AUFS_BR_HFSPLUS
21633 +       bool "Hfsplus as an aufs branch"
21634 +       depends on HFSPLUS_FS
21635 +       default y
21636 +       help
21637 +       If you want to use hfsplus fs as an aufs branch fs, then enable
21638 +       this option. This option introduces a small overhead at
21639 +       copying-up a file on hfsplus.
21640 +
21641 +config AUFS_BDEV_LOOP
21642 +       bool
21643 +       depends on BLK_DEV_LOOP
21644 +       default y
21645 +       help
21646 +       Automatic configuration for internal use.
21647 +       Convert =[ym] into =y.
21648 +
21649 +config AUFS_DEBUG
21650 +       bool "Debug aufs"
21651 +       help
21652 +       Enable this to compile aufs internal debug code.
21653 +       It will have a negative impact to the performance.
21654 +
21655 +config AUFS_MAGIC_SYSRQ
21656 +       bool
21657 +       depends on AUFS_DEBUG && MAGIC_SYSRQ
21658 +       default y
21659 +       help
21660 +       Automatic configuration for internal use.
21661 +       When aufs supports Magic SysRq, enabled automatically.
21662 +endif
21663 diff -urN /usr/share/empty/fs/aufs/loop.c linux/fs/aufs/loop.c
21664 --- /usr/share/empty/fs/aufs/loop.c     1970-01-01 01:00:00.000000000 +0100
21665 +++ linux/fs/aufs/loop.c        2016-10-09 16:55:38.889431135 +0200
21666 @@ -0,0 +1,147 @@
21667 +/*
21668 + * Copyright (C) 2005-2016 Junjiro R. Okajima
21669 + *
21670 + * This program, aufs is free software; you can redistribute it and/or modify
21671 + * it under the terms of the GNU General Public License as published by
21672 + * the Free Software Foundation; either version 2 of the License, or
21673 + * (at your option) any later version.
21674 + *
21675 + * This program is distributed in the hope that it will be useful,
21676 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21677 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21678 + * GNU General Public License for more details.
21679 + *
21680 + * You should have received a copy of the GNU General Public License
21681 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21682 + */
21683 +
21684 +/*
21685 + * support for loopback block device as a branch
21686 + */
21687 +
21688 +#include "aufs.h"
21689 +
21690 +/* added into drivers/block/loop.c */
21691 +static struct file *(*backing_file_func)(struct super_block *sb);
21692 +
21693 +/*
21694 + * test if two lower dentries have overlapping branches.
21695 + */
21696 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding)
21697 +{
21698 +       struct super_block *h_sb;
21699 +       struct file *backing_file;
21700 +
21701 +       if (unlikely(!backing_file_func)) {
21702 +               /* don't load "loop" module here */
21703 +               backing_file_func = symbol_get(loop_backing_file);
21704 +               if (unlikely(!backing_file_func))
21705 +                       /* "loop" module is not loaded */
21706 +                       return 0;
21707 +       }
21708 +
21709 +       h_sb = h_adding->d_sb;
21710 +       backing_file = backing_file_func(h_sb);
21711 +       if (!backing_file)
21712 +               return 0;
21713 +
21714 +       h_adding = backing_file->f_path.dentry;
21715 +       /*
21716 +        * h_adding can be local NFS.
21717 +        * in this case aufs cannot detect the loop.
21718 +        */
21719 +       if (unlikely(h_adding->d_sb == sb))
21720 +               return 1;
21721 +       return !!au_test_subdir(h_adding, sb->s_root);
21722 +}
21723 +
21724 +/* true if a kernel thread named 'loop[0-9].*' accesses a file */
21725 +int au_test_loopback_kthread(void)
21726 +{
21727 +       int ret;
21728 +       struct task_struct *tsk = current;
21729 +       char c, comm[sizeof(tsk->comm)];
21730 +
21731 +       ret = 0;
21732 +       if (tsk->flags & PF_KTHREAD) {
21733 +               get_task_comm(comm, tsk);
21734 +               c = comm[4];
21735 +               ret = ('0' <= c && c <= '9'
21736 +                      && !strncmp(comm, "loop", 4));
21737 +       }
21738 +
21739 +       return ret;
21740 +}
21741 +
21742 +/* ---------------------------------------------------------------------- */
21743 +
21744 +#define au_warn_loopback_step  16
21745 +static int au_warn_loopback_nelem = au_warn_loopback_step;
21746 +static unsigned long *au_warn_loopback_array;
21747 +
21748 +void au_warn_loopback(struct super_block *h_sb)
21749 +{
21750 +       int i, new_nelem;
21751 +       unsigned long *a, magic;
21752 +       static DEFINE_SPINLOCK(spin);
21753 +
21754 +       magic = h_sb->s_magic;
21755 +       spin_lock(&spin);
21756 +       a = au_warn_loopback_array;
21757 +       for (i = 0; i < au_warn_loopback_nelem && *a; i++)
21758 +               if (a[i] == magic) {
21759 +                       spin_unlock(&spin);
21760 +                       return;
21761 +               }
21762 +
21763 +       /* h_sb is new to us, print it */
21764 +       if (i < au_warn_loopback_nelem) {
21765 +               a[i] = magic;
21766 +               goto pr;
21767 +       }
21768 +
21769 +       /* expand the array */
21770 +       new_nelem = au_warn_loopback_nelem + au_warn_loopback_step;
21771 +       a = au_kzrealloc(au_warn_loopback_array,
21772 +                        au_warn_loopback_nelem * sizeof(unsigned long),
21773 +                        new_nelem * sizeof(unsigned long), GFP_ATOMIC,
21774 +                        /*may_shrink*/0);
21775 +       if (a) {
21776 +               au_warn_loopback_nelem = new_nelem;
21777 +               au_warn_loopback_array = a;
21778 +               a[i] = magic;
21779 +               goto pr;
21780 +       }
21781 +
21782 +       spin_unlock(&spin);
21783 +       AuWarn1("realloc failed, ignored\n");
21784 +       return;
21785 +
21786 +pr:
21787 +       spin_unlock(&spin);
21788 +       pr_warn("you may want to try another patch for loopback file "
21789 +               "on %s(0x%lx) branch\n", au_sbtype(h_sb), magic);
21790 +}
21791 +
21792 +int au_loopback_init(void)
21793 +{
21794 +       int err;
21795 +       struct super_block *sb __maybe_unused;
21796 +
21797 +       BUILD_BUG_ON(sizeof(sb->s_magic) != sizeof(unsigned long));
21798 +
21799 +       err = 0;
21800 +       au_warn_loopback_array = kcalloc(au_warn_loopback_step,
21801 +                                        sizeof(unsigned long), GFP_NOFS);
21802 +       if (unlikely(!au_warn_loopback_array))
21803 +               err = -ENOMEM;
21804 +
21805 +       return err;
21806 +}
21807 +
21808 +void au_loopback_fin(void)
21809 +{
21810 +       if (backing_file_func)
21811 +               symbol_put(loop_backing_file);
21812 +       au_delayed_kfree(au_warn_loopback_array);
21813 +}
21814 diff -urN /usr/share/empty/fs/aufs/loop.h linux/fs/aufs/loop.h
21815 --- /usr/share/empty/fs/aufs/loop.h     1970-01-01 01:00:00.000000000 +0100
21816 +++ linux/fs/aufs/loop.h        2016-10-09 16:55:36.492701639 +0200
21817 @@ -0,0 +1,52 @@
21818 +/*
21819 + * Copyright (C) 2005-2016 Junjiro R. Okajima
21820 + *
21821 + * This program, aufs is free software; you can redistribute it and/or modify
21822 + * it under the terms of the GNU General Public License as published by
21823 + * the Free Software Foundation; either version 2 of the License, or
21824 + * (at your option) any later version.
21825 + *
21826 + * This program is distributed in the hope that it will be useful,
21827 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21828 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21829 + * GNU General Public License for more details.
21830 + *
21831 + * You should have received a copy of the GNU General Public License
21832 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21833 + */
21834 +
21835 +/*
21836 + * support for loopback mount as a branch
21837 + */
21838 +
21839 +#ifndef __AUFS_LOOP_H__
21840 +#define __AUFS_LOOP_H__
21841 +
21842 +#ifdef __KERNEL__
21843 +
21844 +struct dentry;
21845 +struct super_block;
21846 +
21847 +#ifdef CONFIG_AUFS_BDEV_LOOP
21848 +/* drivers/block/loop.c */
21849 +struct file *loop_backing_file(struct super_block *sb);
21850 +
21851 +/* loop.c */
21852 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding);
21853 +int au_test_loopback_kthread(void);
21854 +void au_warn_loopback(struct super_block *h_sb);
21855 +
21856 +int au_loopback_init(void);
21857 +void au_loopback_fin(void);
21858 +#else
21859 +AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
21860 +          struct dentry *h_adding)
21861 +AuStubInt0(au_test_loopback_kthread, void)
21862 +AuStubVoid(au_warn_loopback, struct super_block *h_sb)
21863 +
21864 +AuStubInt0(au_loopback_init, void)
21865 +AuStubVoid(au_loopback_fin, void)
21866 +#endif /* BLK_DEV_LOOP */
21867 +
21868 +#endif /* __KERNEL__ */
21869 +#endif /* __AUFS_LOOP_H__ */
21870 diff -urN /usr/share/empty/fs/aufs/magic.mk linux/fs/aufs/magic.mk
21871 --- /usr/share/empty/fs/aufs/magic.mk   1970-01-01 01:00:00.000000000 +0100
21872 +++ linux/fs/aufs/magic.mk      2016-10-09 16:55:36.492701639 +0200
21873 @@ -0,0 +1,30 @@
21874 +
21875 +# defined in ${srctree}/fs/fuse/inode.c
21876 +# tristate
21877 +ifdef CONFIG_FUSE_FS
21878 +ccflags-y += -DFUSE_SUPER_MAGIC=0x65735546
21879 +endif
21880 +
21881 +# defined in ${srctree}/fs/xfs/xfs_sb.h
21882 +# tristate
21883 +ifdef CONFIG_XFS_FS
21884 +ccflags-y += -DXFS_SB_MAGIC=0x58465342
21885 +endif
21886 +
21887 +# defined in ${srctree}/fs/configfs/mount.c
21888 +# tristate
21889 +ifdef CONFIG_CONFIGFS_FS
21890 +ccflags-y += -DCONFIGFS_MAGIC=0x62656570
21891 +endif
21892 +
21893 +# defined in ${srctree}/fs/ubifs/ubifs.h
21894 +# tristate
21895 +ifdef CONFIG_UBIFS_FS
21896 +ccflags-y += -DUBIFS_SUPER_MAGIC=0x24051905
21897 +endif
21898 +
21899 +# defined in ${srctree}/fs/hfsplus/hfsplus_raw.h
21900 +# tristate
21901 +ifdef CONFIG_HFSPLUS_FS
21902 +ccflags-y += -DHFSPLUS_SUPER_MAGIC=0x482b
21903 +endif
21904 diff -urN /usr/share/empty/fs/aufs/Makefile linux/fs/aufs/Makefile
21905 --- /usr/share/empty/fs/aufs/Makefile   1970-01-01 01:00:00.000000000 +0100
21906 +++ linux/fs/aufs/Makefile      2016-10-09 16:55:36.486034798 +0200
21907 @@ -0,0 +1,44 @@
21908 +
21909 +include ${src}/magic.mk
21910 +ifeq (${CONFIG_AUFS_FS},m)
21911 +include ${src}/conf.mk
21912 +endif
21913 +-include ${src}/priv_def.mk
21914 +
21915 +# cf. include/linux/kernel.h
21916 +# enable pr_debug
21917 +ccflags-y += -DDEBUG
21918 +# sparse requires the full pathname
21919 +ifdef M
21920 +ccflags-y += -include ${M}/../../include/uapi/linux/aufs_type.h
21921 +else
21922 +ccflags-y += -include ${srctree}/include/uapi/linux/aufs_type.h
21923 +endif
21924 +
21925 +obj-$(CONFIG_AUFS_FS) += aufs.o
21926 +aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
21927 +       wkq.o vfsub.o dcsub.o \
21928 +       cpup.o whout.o wbr_policy.o \
21929 +       dinfo.o dentry.o \
21930 +       dynop.o \
21931 +       finfo.o file.o f_op.o \
21932 +       dir.o vdir.o \
21933 +       iinfo.o inode.o i_op.o i_op_add.o i_op_del.o i_op_ren.o \
21934 +       mvdown.o ioctl.o
21935 +
21936 +# all are boolean
21937 +aufs-$(CONFIG_PROC_FS) += procfs.o plink.o
21938 +aufs-$(CONFIG_SYSFS) += sysfs.o
21939 +aufs-$(CONFIG_DEBUG_FS) += dbgaufs.o
21940 +aufs-$(CONFIG_AUFS_BDEV_LOOP) += loop.o
21941 +aufs-$(CONFIG_AUFS_HNOTIFY) += hnotify.o
21942 +aufs-$(CONFIG_AUFS_HFSNOTIFY) += hfsnotify.o
21943 +aufs-$(CONFIG_AUFS_EXPORT) += export.o
21944 +aufs-$(CONFIG_AUFS_XATTR) += xattr.o
21945 +aufs-$(CONFIG_FS_POSIX_ACL) += posix_acl.o
21946 +aufs-$(CONFIG_AUFS_FHSM) += fhsm.o
21947 +aufs-$(CONFIG_AUFS_POLL) += poll.o
21948 +aufs-$(CONFIG_AUFS_RDU) += rdu.o
21949 +aufs-$(CONFIG_AUFS_BR_HFSPLUS) += hfsplus.o
21950 +aufs-$(CONFIG_AUFS_DEBUG) += debug.o
21951 +aufs-$(CONFIG_AUFS_MAGIC_SYSRQ) += sysrq.o
21952 diff -urN /usr/share/empty/fs/aufs/module.c linux/fs/aufs/module.c
21953 --- /usr/share/empty/fs/aufs/module.c   1970-01-01 01:00:00.000000000 +0100
21954 +++ linux/fs/aufs/module.c      2016-10-09 16:55:38.889431135 +0200
21955 @@ -0,0 +1,333 @@
21956 +/*
21957 + * Copyright (C) 2005-2016 Junjiro R. Okajima
21958 + *
21959 + * This program, aufs is free software; you can redistribute it and/or modify
21960 + * it under the terms of the GNU General Public License as published by
21961 + * the Free Software Foundation; either version 2 of the License, or
21962 + * (at your option) any later version.
21963 + *
21964 + * This program is distributed in the hope that it will be useful,
21965 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21966 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21967 + * GNU General Public License for more details.
21968 + *
21969 + * You should have received a copy of the GNU General Public License
21970 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21971 + */
21972 +
21973 +/*
21974 + * module global variables and operations
21975 + */
21976 +
21977 +#include <linux/module.h>
21978 +#include <linux/seq_file.h>
21979 +#include "aufs.h"
21980 +
21981 +/* shrinkable realloc */
21982 +void *au_krealloc(void *p, unsigned int new_sz, gfp_t gfp, int may_shrink)
21983 +{
21984 +       size_t sz;
21985 +       int diff;
21986 +
21987 +       sz = 0;
21988 +       diff = -1;
21989 +       if (p) {
21990 +#if 0 /* unused */
21991 +               if (!new_sz) {
21992 +                       au_delayed_kfree(p);
21993 +                       p = NULL;
21994 +                       goto out;
21995 +               }
21996 +#else
21997 +               AuDebugOn(!new_sz);
21998 +#endif
21999 +               sz = ksize(p);
22000 +               diff = au_kmidx_sub(sz, new_sz);
22001 +       }
22002 +       if (sz && !diff)
22003 +               goto out;
22004 +
22005 +       if (sz < new_sz)
22006 +               /* expand or SLOB */
22007 +               p = krealloc(p, new_sz, gfp);
22008 +       else if (new_sz < sz && may_shrink) {
22009 +               /* shrink */
22010 +               void *q;
22011 +
22012 +               q = kmalloc(new_sz, gfp);
22013 +               if (q) {
22014 +                       if (p) {
22015 +                               memcpy(q, p, new_sz);
22016 +                               au_delayed_kfree(p);
22017 +                       }
22018 +                       p = q;
22019 +               } else
22020 +                       p = NULL;
22021 +       }
22022 +
22023 +out:
22024 +       return p;
22025 +}
22026 +
22027 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp,
22028 +                  int may_shrink)
22029 +{
22030 +       p = au_krealloc(p, new_sz, gfp, may_shrink);
22031 +       if (p && new_sz > nused)
22032 +               memset(p + nused, 0, new_sz - nused);
22033 +       return p;
22034 +}
22035 +
22036 +/* ---------------------------------------------------------------------- */
22037 +/*
22038 + * aufs caches
22039 + */
22040 +
22041 +struct au_dfree au_dfree;
22042 +
22043 +/* delayed free */
22044 +static void au_do_dfree(struct work_struct *work __maybe_unused)
22045 +{
22046 +       struct llist_head *head;
22047 +       struct llist_node *node, *next;
22048 +
22049 +#define AU_CACHE_DFREE_DO_BODY(name, idx, lnode) do {                  \
22050 +               head = &au_dfree.cache[AuCache_##idx].llist;            \
22051 +               node = llist_del_all(head);                             \
22052 +               for (; node; node = next) {                             \
22053 +                       struct au_##name *p                             \
22054 +                               = llist_entry(node, struct au_##name,   \
22055 +                                             lnode);                   \
22056 +                       next = llist_next(node);                        \
22057 +                       au_cache_free_##name(p);                        \
22058 +               }                                                       \
22059 +       } while (0)
22060 +
22061 +       AU_CACHE_DFREE_DO_BODY(dinfo, DINFO, di_lnode);
22062 +       AU_CACHE_DFREE_DO_BODY(icntnr, ICNTNR, lnode);
22063 +       AU_CACHE_DFREE_DO_BODY(finfo, FINFO, fi_lnode);
22064 +       AU_CACHE_DFREE_DO_BODY(vdir, VDIR, vd_lnode);
22065 +       AU_CACHE_DFREE_DO_BODY(vdir_dehstr, DEHSTR, lnode);
22066 +#ifdef CONFIG_AUFS_HNOTIFY
22067 +       AU_CACHE_DFREE_DO_BODY(hnotify, HNOTIFY, hn_lnode);
22068 +#endif
22069 +
22070 +#define AU_DFREE_DO_BODY(llist, func) do {             \
22071 +               node = llist_del_all(llist);            \
22072 +               for (; node; node = next) {             \
22073 +                       next = llist_next(node);        \
22074 +                       func(node);                     \
22075 +               }                                       \
22076 +       } while (0)
22077 +
22078 +       AU_DFREE_DO_BODY(au_dfree.llist + AU_DFREE_KFREE, kfree);
22079 +       AU_DFREE_DO_BODY(au_dfree.llist + AU_DFREE_FREE_PAGE, au_free_page);
22080 +
22081 +#undef AU_CACHE_DFREE_DO_BODY
22082 +#undef AU_DFREE_DO_BODY
22083 +}
22084 +
22085 +AU_CACHE_DFREE_FUNC(dinfo, DINFO, di_lnode);
22086 +AU_CACHE_DFREE_FUNC(icntnr, ICNTNR, lnode);
22087 +AU_CACHE_DFREE_FUNC(finfo, FINFO, fi_lnode);
22088 +AU_CACHE_DFREE_FUNC(vdir, VDIR, vd_lnode);
22089 +AU_CACHE_DFREE_FUNC(vdir_dehstr, DEHSTR, lnode);
22090 +
22091 +static void au_cache_fin(void)
22092 +{
22093 +       int i;
22094 +       struct au_cache *cp;
22095 +
22096 +       /*
22097 +        * Make sure all delayed rcu free inodes are flushed before we
22098 +        * destroy cache.
22099 +        */
22100 +       rcu_barrier();
22101 +
22102 +       /* excluding AuCache_HNOTIFY */
22103 +       BUILD_BUG_ON(AuCache_HNOTIFY + 1 != AuCache_Last);
22104 +       flush_delayed_work(&au_dfree.dwork);
22105 +       for (i = 0; i < AuCache_HNOTIFY; i++) {
22106 +               cp = au_dfree.cache + i;
22107 +               AuDebugOn(!llist_empty(&cp->llist));
22108 +               kmem_cache_destroy(cp->cache);
22109 +               cp->cache = NULL;
22110 +       }
22111 +}
22112 +
22113 +static int __init au_cache_init(void)
22114 +{
22115 +       struct au_cache *cp;
22116 +
22117 +       cp = au_dfree.cache;
22118 +       cp[AuCache_DINFO].cache = AuCacheCtor(au_dinfo, au_di_init_once);
22119 +       if (cp[AuCache_DINFO].cache)
22120 +               /* SLAB_DESTROY_BY_RCU */
22121 +               cp[AuCache_ICNTNR].cache = AuCacheCtor(au_icntnr,
22122 +                                                      au_icntnr_init_once);
22123 +       if (cp[AuCache_ICNTNR].cache)
22124 +               cp[AuCache_FINFO].cache = AuCacheCtor(au_finfo,
22125 +                                                     au_fi_init_once);
22126 +       if (cp[AuCache_FINFO].cache)
22127 +               cp[AuCache_VDIR].cache = AuCache(au_vdir);
22128 +       if (cp[AuCache_VDIR].cache)
22129 +               cp[AuCache_DEHSTR].cache = AuCache(au_vdir_dehstr);
22130 +       if (cp[AuCache_DEHSTR].cache)
22131 +               return 0;
22132 +
22133 +       au_cache_fin();
22134 +       return -ENOMEM;
22135 +}
22136 +
22137 +/* ---------------------------------------------------------------------- */
22138 +
22139 +int au_dir_roflags;
22140 +
22141 +#ifdef CONFIG_AUFS_SBILIST
22142 +/*
22143 + * iterate_supers_type() doesn't protect us from
22144 + * remounting (branch management)
22145 + */
22146 +struct au_sphlhead au_sbilist;
22147 +#endif
22148 +
22149 +/*
22150 + * functions for module interface.
22151 + */
22152 +MODULE_LICENSE("GPL");
22153 +/* MODULE_LICENSE("GPL v2"); */
22154 +MODULE_AUTHOR("Junjiro R. Okajima <aufs-users@lists.sourceforge.net>");
22155 +MODULE_DESCRIPTION(AUFS_NAME
22156 +       " -- Advanced multi layered unification filesystem");
22157 +MODULE_VERSION(AUFS_VERSION);
22158 +MODULE_ALIAS_FS(AUFS_NAME);
22159 +
22160 +/* this module parameter has no meaning when SYSFS is disabled */
22161 +int sysaufs_brs = 1;
22162 +MODULE_PARM_DESC(brs, "use <sysfs>/fs/aufs/si_*/brN");
22163 +module_param_named(brs, sysaufs_brs, int, S_IRUGO);
22164 +
22165 +/* this module parameter has no meaning when USER_NS is disabled */
22166 +bool au_userns;
22167 +MODULE_PARM_DESC(allow_userns, "allow unprivileged to mount under userns");
22168 +module_param_named(allow_userns, au_userns, bool, S_IRUGO);
22169 +
22170 +/* ---------------------------------------------------------------------- */
22171 +
22172 +static char au_esc_chars[0x20 + 3]; /* 0x01-0x20, backslash, del, and NULL */
22173 +
22174 +int au_seq_path(struct seq_file *seq, struct path *path)
22175 +{
22176 +       int err;
22177 +
22178 +       err = seq_path(seq, path, au_esc_chars);
22179 +       if (err > 0)
22180 +               err = 0;
22181 +       else if (err < 0)
22182 +               err = -ENOMEM;
22183 +
22184 +       return err;
22185 +}
22186 +
22187 +/* ---------------------------------------------------------------------- */
22188 +
22189 +static int __init aufs_init(void)
22190 +{
22191 +       int err, i;
22192 +       char *p;
22193 +       struct au_cache *cp;
22194 +
22195 +       p = au_esc_chars;
22196 +       for (i = 1; i <= ' '; i++)
22197 +               *p++ = i;
22198 +       *p++ = '\\';
22199 +       *p++ = '\x7f';
22200 +       *p = 0;
22201 +
22202 +       au_dir_roflags = au_file_roflags(O_DIRECTORY | O_LARGEFILE);
22203 +
22204 +       memcpy(aufs_iop_nogetattr, aufs_iop, sizeof(aufs_iop));
22205 +       for (i = 0; i < AuIop_Last; i++)
22206 +               aufs_iop_nogetattr[i].getattr = NULL;
22207 +
22208 +       /* First, initialize au_dfree */
22209 +       for (i = 0; i < AuCache_Last; i++) {    /* including hnotify */
22210 +               cp = au_dfree.cache + i;
22211 +               cp->cache = NULL;
22212 +               init_llist_head(&cp->llist);
22213 +       }
22214 +       for (i = 0; i < AU_DFREE_Last; i++)
22215 +               init_llist_head(au_dfree.llist + i);
22216 +       INIT_DELAYED_WORK(&au_dfree.dwork, au_do_dfree);
22217 +
22218 +       au_sbilist_init();
22219 +       sysaufs_brs_init();
22220 +       au_debug_init();
22221 +       au_dy_init();
22222 +       err = sysaufs_init();
22223 +       if (unlikely(err))
22224 +               goto out;
22225 +       err = au_procfs_init();
22226 +       if (unlikely(err))
22227 +               goto out_sysaufs;
22228 +       err = au_wkq_init();
22229 +       if (unlikely(err))
22230 +               goto out_procfs;
22231 +       err = au_loopback_init();
22232 +       if (unlikely(err))
22233 +               goto out_wkq;
22234 +       err = au_hnotify_init();
22235 +       if (unlikely(err))
22236 +               goto out_loopback;
22237 +       err = au_sysrq_init();
22238 +       if (unlikely(err))
22239 +               goto out_hin;
22240 +       err = au_cache_init();
22241 +       if (unlikely(err))
22242 +               goto out_sysrq;
22243 +
22244 +       aufs_fs_type.fs_flags |= au_userns ? FS_USERNS_MOUNT : 0;
22245 +       err = register_filesystem(&aufs_fs_type);
22246 +       if (unlikely(err))
22247 +               goto out_cache;
22248 +
22249 +       /* since we define pr_fmt, call printk directly */
22250 +       printk(KERN_INFO AUFS_NAME " " AUFS_VERSION "\n");
22251 +       goto out; /* success */
22252 +
22253 +out_cache:
22254 +       au_cache_fin();
22255 +out_sysrq:
22256 +       au_sysrq_fin();
22257 +out_hin:
22258 +       au_hnotify_fin();
22259 +out_loopback:
22260 +       au_loopback_fin();
22261 +out_wkq:
22262 +       au_wkq_fin();
22263 +out_procfs:
22264 +       au_procfs_fin();
22265 +out_sysaufs:
22266 +       sysaufs_fin();
22267 +       au_dy_fin();
22268 +       flush_delayed_work(&au_dfree.dwork);
22269 +out:
22270 +       return err;
22271 +}
22272 +
22273 +static void __exit aufs_exit(void)
22274 +{
22275 +       unregister_filesystem(&aufs_fs_type);
22276 +       au_cache_fin();
22277 +       au_sysrq_fin();
22278 +       au_hnotify_fin();
22279 +       au_loopback_fin();
22280 +       au_wkq_fin();
22281 +       au_procfs_fin();
22282 +       sysaufs_fin();
22283 +       au_dy_fin();
22284 +       flush_delayed_work(&au_dfree.dwork);
22285 +}
22286 +
22287 +module_init(aufs_init);
22288 +module_exit(aufs_exit);
22289 diff -urN /usr/share/empty/fs/aufs/module.h linux/fs/aufs/module.h
22290 --- /usr/share/empty/fs/aufs/module.h   1970-01-01 01:00:00.000000000 +0100
22291 +++ linux/fs/aufs/module.h      2016-10-09 16:55:38.889431135 +0200
22292 @@ -0,0 +1,156 @@
22293 +/*
22294 + * Copyright (C) 2005-2016 Junjiro R. Okajima
22295 + *
22296 + * This program, aufs is free software; you can redistribute it and/or modify
22297 + * it under the terms of the GNU General Public License as published by
22298 + * the Free Software Foundation; either version 2 of the License, or
22299 + * (at your option) any later version.
22300 + *
22301 + * This program is distributed in the hope that it will be useful,
22302 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22303 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22304 + * GNU General Public License for more details.
22305 + *
22306 + * You should have received a copy of the GNU General Public License
22307 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22308 + */
22309 +
22310 +/*
22311 + * module initialization and module-global
22312 + */
22313 +
22314 +#ifndef __AUFS_MODULE_H__
22315 +#define __AUFS_MODULE_H__
22316 +
22317 +#ifdef __KERNEL__
22318 +
22319 +#include <linux/slab.h>
22320 +#include "debug.h"
22321 +
22322 +struct path;
22323 +struct seq_file;
22324 +
22325 +/* module parameters */
22326 +extern int sysaufs_brs;
22327 +extern bool au_userns;
22328 +
22329 +/* ---------------------------------------------------------------------- */
22330 +
22331 +extern int au_dir_roflags;
22332 +
22333 +void *au_krealloc(void *p, unsigned int new_sz, gfp_t gfp, int may_shrink);
22334 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp,
22335 +                  int may_shrink);
22336 +
22337 +static inline int au_kmidx_sub(size_t sz, size_t new_sz)
22338 +{
22339 +#ifndef CONFIG_SLOB
22340 +       return kmalloc_index(sz) - kmalloc_index(new_sz);
22341 +#else
22342 +       return -1; /* SLOB is untested */
22343 +#endif
22344 +}
22345 +
22346 +int au_seq_path(struct seq_file *seq, struct path *path);
22347 +
22348 +#ifdef CONFIG_PROC_FS
22349 +/* procfs.c */
22350 +int __init au_procfs_init(void);
22351 +void au_procfs_fin(void);
22352 +#else
22353 +AuStubInt0(au_procfs_init, void);
22354 +AuStubVoid(au_procfs_fin, void);
22355 +#endif
22356 +
22357 +/* ---------------------------------------------------------------------- */
22358 +
22359 +/* kmem cache and delayed free */
22360 +enum {
22361 +       AuCache_DINFO,
22362 +       AuCache_ICNTNR,
22363 +       AuCache_FINFO,
22364 +       AuCache_VDIR,
22365 +       AuCache_DEHSTR,
22366 +       AuCache_HNOTIFY, /* must be last */
22367 +       AuCache_Last
22368 +};
22369 +
22370 +enum {
22371 +       AU_DFREE_KFREE,
22372 +       AU_DFREE_FREE_PAGE,
22373 +       AU_DFREE_Last
22374 +};
22375 +
22376 +struct au_cache {
22377 +       struct kmem_cache       *cache;
22378 +       struct llist_head       llist;  /* delayed free */
22379 +};
22380 +
22381 +/*
22382 + * in order to reduce the cost of the internal timer, consolidate all the
22383 + * delayed free works into a single delayed_work.
22384 + */
22385 +struct au_dfree {
22386 +       struct au_cache         cache[AuCache_Last];
22387 +       struct llist_head       llist[AU_DFREE_Last];
22388 +       struct delayed_work     dwork;
22389 +};
22390 +
22391 +extern struct au_dfree au_dfree;
22392 +
22393 +#define AuCacheFlags           (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD)
22394 +#define AuCache(type)          KMEM_CACHE(type, AuCacheFlags)
22395 +#define AuCacheCtor(type, ctor)        \
22396 +       kmem_cache_create(#type, sizeof(struct type), \
22397 +                         __alignof__(struct type), AuCacheFlags, ctor)
22398 +
22399 +#define AU_DFREE_DELAY         msecs_to_jiffies(10)
22400 +#define AU_DFREE_BODY(lnode, llist) do {                               \
22401 +               if (llist_add(lnode, llist))                            \
22402 +                       schedule_delayed_work(&au_dfree.dwork,          \
22403 +                                             AU_DFREE_DELAY);          \
22404 +       } while (0)
22405 +#define AU_CACHE_DFREE_FUNC(name, idx, lnode)                          \
22406 +       void au_cache_dfree_##name(struct au_##name *p)                 \
22407 +       {                                                               \
22408 +               struct au_cache *cp = au_dfree.cache + AuCache_##idx;   \
22409 +               AU_DFREE_BODY(&p->lnode, &cp->llist);                   \
22410 +       }
22411 +
22412 +#define AuCacheFuncs(name, index) \
22413 +static inline struct au_##name *au_cache_alloc_##name(void) \
22414 +{ return kmem_cache_alloc(au_dfree.cache[AuCache_##index].cache, GFP_NOFS); } \
22415 +static inline void au_cache_free_##name(struct au_##name *p) \
22416 +{ kmem_cache_free(au_dfree.cache[AuCache_##index].cache, p); } \
22417 +void au_cache_dfree_##name(struct au_##name *p)
22418 +
22419 +AuCacheFuncs(dinfo, DINFO);
22420 +AuCacheFuncs(icntnr, ICNTNR);
22421 +AuCacheFuncs(finfo, FINFO);
22422 +AuCacheFuncs(vdir, VDIR);
22423 +AuCacheFuncs(vdir_dehstr, DEHSTR);
22424 +#ifdef CONFIG_AUFS_HNOTIFY
22425 +AuCacheFuncs(hnotify, HNOTIFY);
22426 +#endif
22427 +
22428 +static inline void au_delayed_kfree(const void *p)
22429 +{
22430 +       AuDebugOn(!p);
22431 +       AuDebugOn(ksize(p) < sizeof(struct llist_node));
22432 +
22433 +       AU_DFREE_BODY((void *)p, au_dfree.llist + AU_DFREE_KFREE);
22434 +}
22435 +
22436 +/* cast only */
22437 +static inline void au_free_page(void *p)
22438 +{
22439 +       free_page((unsigned long)p);
22440 +}
22441 +
22442 +static inline void au_delayed_free_page(unsigned long addr)
22443 +{
22444 +       AU_DFREE_BODY((void *)addr, au_dfree.llist + AU_DFREE_FREE_PAGE);
22445 +}
22446 +
22447 +#endif /* __KERNEL__ */
22448 +#endif /* __AUFS_MODULE_H__ */
22449 diff -urN /usr/share/empty/fs/aufs/mvdown.c linux/fs/aufs/mvdown.c
22450 --- /usr/share/empty/fs/aufs/mvdown.c   1970-01-01 01:00:00.000000000 +0100
22451 +++ linux/fs/aufs/mvdown.c      2016-10-09 16:55:36.492701639 +0200
22452 @@ -0,0 +1,704 @@
22453 +/*
22454 + * Copyright (C) 2011-2016 Junjiro R. Okajima
22455 + *
22456 + * This program, aufs is free software; you can redistribute it and/or modify
22457 + * it under the terms of the GNU General Public License as published by
22458 + * the Free Software Foundation; either version 2 of the License, or
22459 + * (at your option) any later version.
22460 + *
22461 + * This program is distributed in the hope that it will be useful,
22462 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22463 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22464 + * GNU General Public License for more details.
22465 + *
22466 + * You should have received a copy of the GNU General Public License
22467 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22468 + */
22469 +
22470 +/*
22471 + * move-down, opposite of copy-up
22472 + */
22473 +
22474 +#include "aufs.h"
22475 +
22476 +struct au_mvd_args {
22477 +       struct {
22478 +               struct super_block *h_sb;
22479 +               struct dentry *h_parent;
22480 +               struct au_hinode *hdir;
22481 +               struct inode *h_dir, *h_inode;
22482 +               struct au_pin pin;
22483 +       } info[AUFS_MVDOWN_NARRAY];
22484 +
22485 +       struct aufs_mvdown mvdown;
22486 +       struct dentry *dentry, *parent;
22487 +       struct inode *inode, *dir;
22488 +       struct super_block *sb;
22489 +       aufs_bindex_t bopq, bwh, bfound;
22490 +       unsigned char rename_lock;
22491 +};
22492 +
22493 +#define mvd_errno              mvdown.au_errno
22494 +#define mvd_bsrc               mvdown.stbr[AUFS_MVDOWN_UPPER].bindex
22495 +#define mvd_src_brid           mvdown.stbr[AUFS_MVDOWN_UPPER].brid
22496 +#define mvd_bdst               mvdown.stbr[AUFS_MVDOWN_LOWER].bindex
22497 +#define mvd_dst_brid           mvdown.stbr[AUFS_MVDOWN_LOWER].brid
22498 +
22499 +#define mvd_h_src_sb           info[AUFS_MVDOWN_UPPER].h_sb
22500 +#define mvd_h_src_parent       info[AUFS_MVDOWN_UPPER].h_parent
22501 +#define mvd_hdir_src           info[AUFS_MVDOWN_UPPER].hdir
22502 +#define mvd_h_src_dir          info[AUFS_MVDOWN_UPPER].h_dir
22503 +#define mvd_h_src_inode                info[AUFS_MVDOWN_UPPER].h_inode
22504 +#define mvd_pin_src            info[AUFS_MVDOWN_UPPER].pin
22505 +
22506 +#define mvd_h_dst_sb           info[AUFS_MVDOWN_LOWER].h_sb
22507 +#define mvd_h_dst_parent       info[AUFS_MVDOWN_LOWER].h_parent
22508 +#define mvd_hdir_dst           info[AUFS_MVDOWN_LOWER].hdir
22509 +#define mvd_h_dst_dir          info[AUFS_MVDOWN_LOWER].h_dir
22510 +#define mvd_h_dst_inode                info[AUFS_MVDOWN_LOWER].h_inode
22511 +#define mvd_pin_dst            info[AUFS_MVDOWN_LOWER].pin
22512 +
22513 +#define AU_MVD_PR(flag, ...) do {                      \
22514 +               if (flag)                               \
22515 +                       pr_err(__VA_ARGS__);            \
22516 +       } while (0)
22517 +
22518 +static int find_lower_writable(struct au_mvd_args *a)
22519 +{
22520 +       struct super_block *sb;
22521 +       aufs_bindex_t bindex, bbot;
22522 +       struct au_branch *br;
22523 +
22524 +       sb = a->sb;
22525 +       bindex = a->mvd_bsrc;
22526 +       bbot = au_sbbot(sb);
22527 +       if (a->mvdown.flags & AUFS_MVDOWN_FHSM_LOWER)
22528 +               for (bindex++; bindex <= bbot; bindex++) {
22529 +                       br = au_sbr(sb, bindex);
22530 +                       if (au_br_fhsm(br->br_perm)
22531 +                           && (!(au_br_sb(br)->s_flags & MS_RDONLY)))
22532 +                               return bindex;
22533 +               }
22534 +       else if (!(a->mvdown.flags & AUFS_MVDOWN_ROLOWER))
22535 +               for (bindex++; bindex <= bbot; bindex++) {
22536 +                       br = au_sbr(sb, bindex);
22537 +                       if (!au_br_rdonly(br))
22538 +                               return bindex;
22539 +               }
22540 +       else
22541 +               for (bindex++; bindex <= bbot; bindex++) {
22542 +                       br = au_sbr(sb, bindex);
22543 +                       if (!(au_br_sb(br)->s_flags & MS_RDONLY)) {
22544 +                               if (au_br_rdonly(br))
22545 +                                       a->mvdown.flags
22546 +                                               |= AUFS_MVDOWN_ROLOWER_R;
22547 +                               return bindex;
22548 +                       }
22549 +               }
22550 +
22551 +       return -1;
22552 +}
22553 +
22554 +/* make the parent dir on bdst */
22555 +static int au_do_mkdir(const unsigned char dmsg, struct au_mvd_args *a)
22556 +{
22557 +       int err;
22558 +
22559 +       err = 0;
22560 +       a->mvd_hdir_src = au_hi(a->dir, a->mvd_bsrc);
22561 +       a->mvd_hdir_dst = au_hi(a->dir, a->mvd_bdst);
22562 +       a->mvd_h_src_parent = au_h_dptr(a->parent, a->mvd_bsrc);
22563 +       a->mvd_h_dst_parent = NULL;
22564 +       if (au_dbbot(a->parent) >= a->mvd_bdst)
22565 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
22566 +       if (!a->mvd_h_dst_parent) {
22567 +               err = au_cpdown_dirs(a->dentry, a->mvd_bdst);
22568 +               if (unlikely(err)) {
22569 +                       AU_MVD_PR(dmsg, "cpdown_dirs failed\n");
22570 +                       goto out;
22571 +               }
22572 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
22573 +       }
22574 +
22575 +out:
22576 +       AuTraceErr(err);
22577 +       return err;
22578 +}
22579 +
22580 +/* lock them all */
22581 +static int au_do_lock(const unsigned char dmsg, struct au_mvd_args *a)
22582 +{
22583 +       int err;
22584 +       struct dentry *h_trap;
22585 +
22586 +       a->mvd_h_src_sb = au_sbr_sb(a->sb, a->mvd_bsrc);
22587 +       a->mvd_h_dst_sb = au_sbr_sb(a->sb, a->mvd_bdst);
22588 +       err = au_pin(&a->mvd_pin_dst, a->dentry, a->mvd_bdst,
22589 +                    au_opt_udba(a->sb),
22590 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
22591 +       AuTraceErr(err);
22592 +       if (unlikely(err)) {
22593 +               AU_MVD_PR(dmsg, "pin_dst failed\n");
22594 +               goto out;
22595 +       }
22596 +
22597 +       if (a->mvd_h_src_sb != a->mvd_h_dst_sb) {
22598 +               a->rename_lock = 0;
22599 +               au_pin_init(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
22600 +                           AuLsc_DI_PARENT, AuLsc_I_PARENT3,
22601 +                           au_opt_udba(a->sb),
22602 +                           AuPin_MNT_WRITE | AuPin_DI_LOCKED);
22603 +               err = au_do_pin(&a->mvd_pin_src);
22604 +               AuTraceErr(err);
22605 +               a->mvd_h_src_dir = d_inode(a->mvd_h_src_parent);
22606 +               if (unlikely(err)) {
22607 +                       AU_MVD_PR(dmsg, "pin_src failed\n");
22608 +                       goto out_dst;
22609 +               }
22610 +               goto out; /* success */
22611 +       }
22612 +
22613 +       a->rename_lock = 1;
22614 +       au_pin_hdir_unlock(&a->mvd_pin_dst);
22615 +       err = au_pin(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
22616 +                    au_opt_udba(a->sb),
22617 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
22618 +       AuTraceErr(err);
22619 +       a->mvd_h_src_dir = d_inode(a->mvd_h_src_parent);
22620 +       if (unlikely(err)) {
22621 +               AU_MVD_PR(dmsg, "pin_src failed\n");
22622 +               au_pin_hdir_lock(&a->mvd_pin_dst);
22623 +               goto out_dst;
22624 +       }
22625 +       au_pin_hdir_unlock(&a->mvd_pin_src);
22626 +       h_trap = vfsub_lock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
22627 +                                  a->mvd_h_dst_parent, a->mvd_hdir_dst);
22628 +       if (h_trap) {
22629 +               err = (h_trap != a->mvd_h_src_parent);
22630 +               if (err)
22631 +                       err = (h_trap != a->mvd_h_dst_parent);
22632 +       }
22633 +       BUG_ON(err); /* it should never happen */
22634 +       if (unlikely(a->mvd_h_src_dir != au_pinned_h_dir(&a->mvd_pin_src))) {
22635 +               err = -EBUSY;
22636 +               AuTraceErr(err);
22637 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
22638 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
22639 +               au_pin_hdir_lock(&a->mvd_pin_src);
22640 +               au_unpin(&a->mvd_pin_src);
22641 +               au_pin_hdir_lock(&a->mvd_pin_dst);
22642 +               goto out_dst;
22643 +       }
22644 +       goto out; /* success */
22645 +
22646 +out_dst:
22647 +       au_unpin(&a->mvd_pin_dst);
22648 +out:
22649 +       AuTraceErr(err);
22650 +       return err;
22651 +}
22652 +
22653 +static void au_do_unlock(const unsigned char dmsg, struct au_mvd_args *a)
22654 +{
22655 +       if (!a->rename_lock)
22656 +               au_unpin(&a->mvd_pin_src);
22657 +       else {
22658 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
22659 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
22660 +               au_pin_hdir_lock(&a->mvd_pin_src);
22661 +               au_unpin(&a->mvd_pin_src);
22662 +               au_pin_hdir_lock(&a->mvd_pin_dst);
22663 +       }
22664 +       au_unpin(&a->mvd_pin_dst);
22665 +}
22666 +
22667 +/* copy-down the file */
22668 +static int au_do_cpdown(const unsigned char dmsg, struct au_mvd_args *a)
22669 +{
22670 +       int err;
22671 +       struct au_cp_generic cpg = {
22672 +               .dentry = a->dentry,
22673 +               .bdst   = a->mvd_bdst,
22674 +               .bsrc   = a->mvd_bsrc,
22675 +               .len    = -1,
22676 +               .pin    = &a->mvd_pin_dst,
22677 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
22678 +       };
22679 +
22680 +       AuDbg("b%d, b%d\n", cpg.bsrc, cpg.bdst);
22681 +       if (a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
22682 +               au_fset_cpup(cpg.flags, OVERWRITE);
22683 +       if (a->mvdown.flags & AUFS_MVDOWN_ROLOWER)
22684 +               au_fset_cpup(cpg.flags, RWDST);
22685 +       err = au_sio_cpdown_simple(&cpg);
22686 +       if (unlikely(err))
22687 +               AU_MVD_PR(dmsg, "cpdown failed\n");
22688 +
22689 +       AuTraceErr(err);
22690 +       return err;
22691 +}
22692 +
22693 +/*
22694 + * unlink the whiteout on bdst if exist which may be created by UDBA while we
22695 + * were sleeping
22696 + */
22697 +static int au_do_unlink_wh(const unsigned char dmsg, struct au_mvd_args *a)
22698 +{
22699 +       int err;
22700 +       struct path h_path;
22701 +       struct au_branch *br;
22702 +       struct inode *delegated;
22703 +
22704 +       br = au_sbr(a->sb, a->mvd_bdst);
22705 +       h_path.dentry = au_wh_lkup(a->mvd_h_dst_parent, &a->dentry->d_name, br);
22706 +       err = PTR_ERR(h_path.dentry);
22707 +       if (IS_ERR(h_path.dentry)) {
22708 +               AU_MVD_PR(dmsg, "wh_lkup failed\n");
22709 +               goto out;
22710 +       }
22711 +
22712 +       err = 0;
22713 +       if (d_is_positive(h_path.dentry)) {
22714 +               h_path.mnt = au_br_mnt(br);
22715 +               delegated = NULL;
22716 +               err = vfsub_unlink(d_inode(a->mvd_h_dst_parent), &h_path,
22717 +                                  &delegated, /*force*/0);
22718 +               if (unlikely(err == -EWOULDBLOCK)) {
22719 +                       pr_warn("cannot retry for NFSv4 delegation"
22720 +                               " for an internal unlink\n");
22721 +                       iput(delegated);
22722 +               }
22723 +               if (unlikely(err))
22724 +                       AU_MVD_PR(dmsg, "wh_unlink failed\n");
22725 +       }
22726 +       dput(h_path.dentry);
22727 +
22728 +out:
22729 +       AuTraceErr(err);
22730 +       return err;
22731 +}
22732 +
22733 +/*
22734 + * unlink the topmost h_dentry
22735 + */
22736 +static int au_do_unlink(const unsigned char dmsg, struct au_mvd_args *a)
22737 +{
22738 +       int err;
22739 +       struct path h_path;
22740 +       struct inode *delegated;
22741 +
22742 +       h_path.mnt = au_sbr_mnt(a->sb, a->mvd_bsrc);
22743 +       h_path.dentry = au_h_dptr(a->dentry, a->mvd_bsrc);
22744 +       delegated = NULL;
22745 +       err = vfsub_unlink(a->mvd_h_src_dir, &h_path, &delegated, /*force*/0);
22746 +       if (unlikely(err == -EWOULDBLOCK)) {
22747 +               pr_warn("cannot retry for NFSv4 delegation"
22748 +                       " for an internal unlink\n");
22749 +               iput(delegated);
22750 +       }
22751 +       if (unlikely(err))
22752 +               AU_MVD_PR(dmsg, "unlink failed\n");
22753 +
22754 +       AuTraceErr(err);
22755 +       return err;
22756 +}
22757 +
22758 +/* Since mvdown succeeded, we ignore an error of this function */
22759 +static void au_do_stfs(const unsigned char dmsg, struct au_mvd_args *a)
22760 +{
22761 +       int err;
22762 +       struct au_branch *br;
22763 +
22764 +       a->mvdown.flags |= AUFS_MVDOWN_STFS_FAILED;
22765 +       br = au_sbr(a->sb, a->mvd_bsrc);
22766 +       err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_UPPER].stfs);
22767 +       if (!err) {
22768 +               br = au_sbr(a->sb, a->mvd_bdst);
22769 +               a->mvdown.stbr[AUFS_MVDOWN_LOWER].brid = br->br_id;
22770 +               err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_LOWER].stfs);
22771 +       }
22772 +       if (!err)
22773 +               a->mvdown.flags &= ~AUFS_MVDOWN_STFS_FAILED;
22774 +       else
22775 +               AU_MVD_PR(dmsg, "statfs failed (%d), ignored\n", err);
22776 +}
22777 +
22778 +/*
22779 + * copy-down the file and unlink the bsrc file.
22780 + * - unlink the bdst whout if exist
22781 + * - copy-down the file (with whtmp name and rename)
22782 + * - unlink the bsrc file
22783 + */
22784 +static int au_do_mvdown(const unsigned char dmsg, struct au_mvd_args *a)
22785 +{
22786 +       int err;
22787 +
22788 +       err = au_do_mkdir(dmsg, a);
22789 +       if (!err)
22790 +               err = au_do_lock(dmsg, a);
22791 +       if (unlikely(err))
22792 +               goto out;
22793 +
22794 +       /*
22795 +        * do not revert the activities we made on bdst since they should be
22796 +        * harmless in aufs.
22797 +        */
22798 +
22799 +       err = au_do_cpdown(dmsg, a);
22800 +       if (!err)
22801 +               err = au_do_unlink_wh(dmsg, a);
22802 +       if (!err && !(a->mvdown.flags & AUFS_MVDOWN_KUPPER))
22803 +               err = au_do_unlink(dmsg, a);
22804 +       if (unlikely(err))
22805 +               goto out_unlock;
22806 +
22807 +       AuDbg("%pd2, 0x%x, %d --> %d\n",
22808 +             a->dentry, a->mvdown.flags, a->mvd_bsrc, a->mvd_bdst);
22809 +       if (find_lower_writable(a) < 0)
22810 +               a->mvdown.flags |= AUFS_MVDOWN_BOTTOM;
22811 +
22812 +       if (a->mvdown.flags & AUFS_MVDOWN_STFS)
22813 +               au_do_stfs(dmsg, a);
22814 +
22815 +       /* maintain internal array */
22816 +       if (!(a->mvdown.flags & AUFS_MVDOWN_KUPPER)) {
22817 +               au_set_h_dptr(a->dentry, a->mvd_bsrc, NULL);
22818 +               au_set_dbtop(a->dentry, a->mvd_bdst);
22819 +               au_set_h_iptr(a->inode, a->mvd_bsrc, NULL, /*flags*/0);
22820 +               au_set_ibtop(a->inode, a->mvd_bdst);
22821 +       } else {
22822 +               /* hide the lower */
22823 +               au_set_h_dptr(a->dentry, a->mvd_bdst, NULL);
22824 +               au_set_dbbot(a->dentry, a->mvd_bsrc);
22825 +               au_set_h_iptr(a->inode, a->mvd_bdst, NULL, /*flags*/0);
22826 +               au_set_ibbot(a->inode, a->mvd_bsrc);
22827 +       }
22828 +       if (au_dbbot(a->dentry) < a->mvd_bdst)
22829 +               au_set_dbbot(a->dentry, a->mvd_bdst);
22830 +       if (au_ibbot(a->inode) < a->mvd_bdst)
22831 +               au_set_ibbot(a->inode, a->mvd_bdst);
22832 +
22833 +out_unlock:
22834 +       au_do_unlock(dmsg, a);
22835 +out:
22836 +       AuTraceErr(err);
22837 +       return err;
22838 +}
22839 +
22840 +/* ---------------------------------------------------------------------- */
22841 +
22842 +/* make sure the file is idle */
22843 +static int au_mvd_args_busy(const unsigned char dmsg, struct au_mvd_args *a)
22844 +{
22845 +       int err, plinked;
22846 +
22847 +       err = 0;
22848 +       plinked = !!au_opt_test(au_mntflags(a->sb), PLINK);
22849 +       if (au_dbtop(a->dentry) == a->mvd_bsrc
22850 +           && au_dcount(a->dentry) == 1
22851 +           && atomic_read(&a->inode->i_count) == 1
22852 +           /* && a->mvd_h_src_inode->i_nlink == 1 */
22853 +           && (!plinked || !au_plink_test(a->inode))
22854 +           && a->inode->i_nlink == 1)
22855 +               goto out;
22856 +
22857 +       err = -EBUSY;
22858 +       AU_MVD_PR(dmsg,
22859 +                 "b%d, d{b%d, c%d?}, i{c%d?, l%u}, hi{l%u}, p{%d, %d}\n",
22860 +                 a->mvd_bsrc, au_dbtop(a->dentry), au_dcount(a->dentry),
22861 +                 atomic_read(&a->inode->i_count), a->inode->i_nlink,
22862 +                 a->mvd_h_src_inode->i_nlink,
22863 +                 plinked, plinked ? au_plink_test(a->inode) : 0);
22864 +
22865 +out:
22866 +       AuTraceErr(err);
22867 +       return err;
22868 +}
22869 +
22870 +/* make sure the parent dir is fine */
22871 +static int au_mvd_args_parent(const unsigned char dmsg,
22872 +                             struct au_mvd_args *a)
22873 +{
22874 +       int err;
22875 +       aufs_bindex_t bindex;
22876 +
22877 +       err = 0;
22878 +       if (unlikely(au_alive_dir(a->parent))) {
22879 +               err = -ENOENT;
22880 +               AU_MVD_PR(dmsg, "parent dir is dead\n");
22881 +               goto out;
22882 +       }
22883 +
22884 +       a->bopq = au_dbdiropq(a->parent);
22885 +       bindex = au_wbr_nonopq(a->dentry, a->mvd_bdst);
22886 +       AuDbg("b%d\n", bindex);
22887 +       if (unlikely((bindex >= 0 && bindex < a->mvd_bdst)
22888 +                    || (a->bopq != -1 && a->bopq < a->mvd_bdst))) {
22889 +               err = -EINVAL;
22890 +               a->mvd_errno = EAU_MVDOWN_OPAQUE;
22891 +               AU_MVD_PR(dmsg, "ancestor is opaque b%d, b%d\n",
22892 +                         a->bopq, a->mvd_bdst);
22893 +       }
22894 +
22895 +out:
22896 +       AuTraceErr(err);
22897 +       return err;
22898 +}
22899 +
22900 +static int au_mvd_args_intermediate(const unsigned char dmsg,
22901 +                                   struct au_mvd_args *a)
22902 +{
22903 +       int err;
22904 +       struct au_dinfo *dinfo, *tmp;
22905 +
22906 +       /* lookup the next lower positive entry */
22907 +       err = -ENOMEM;
22908 +       tmp = au_di_alloc(a->sb, AuLsc_DI_TMP);
22909 +       if (unlikely(!tmp))
22910 +               goto out;
22911 +
22912 +       a->bfound = -1;
22913 +       a->bwh = -1;
22914 +       dinfo = au_di(a->dentry);
22915 +       au_di_cp(tmp, dinfo);
22916 +       au_di_swap(tmp, dinfo);
22917 +
22918 +       /* returns the number of positive dentries */
22919 +       err = au_lkup_dentry(a->dentry, a->mvd_bsrc + 1,
22920 +                            /* AuLkup_IGNORE_PERM */ 0);
22921 +       if (!err)
22922 +               a->bwh = au_dbwh(a->dentry);
22923 +       else if (err > 0)
22924 +               a->bfound = au_dbtop(a->dentry);
22925 +
22926 +       au_di_swap(tmp, dinfo);
22927 +       au_rw_write_unlock(&tmp->di_rwsem);
22928 +       au_di_free(tmp);
22929 +       if (unlikely(err < 0))
22930 +               AU_MVD_PR(dmsg, "failed look-up lower\n");
22931 +
22932 +       /*
22933 +        * here, we have these cases.
22934 +        * bfound == -1
22935 +        *      no positive dentry under bsrc. there are more sub-cases.
22936 +        *      bwh < 0
22937 +        *              there no whiteout, we can safely move-down.
22938 +        *      bwh <= bsrc
22939 +        *              impossible
22940 +        *      bsrc < bwh && bwh < bdst
22941 +        *              there is a whiteout on RO branch. cannot proceed.
22942 +        *      bwh == bdst
22943 +        *              there is a whiteout on the RW target branch. it should
22944 +        *              be removed.
22945 +        *      bdst < bwh
22946 +        *              there is a whiteout somewhere unrelated branch.
22947 +        * -1 < bfound && bfound <= bsrc
22948 +        *      impossible.
22949 +        * bfound < bdst
22950 +        *      found, but it is on RO branch between bsrc and bdst. cannot
22951 +        *      proceed.
22952 +        * bfound == bdst
22953 +        *      found, replace it if AUFS_MVDOWN_FORCE is set. otherwise return
22954 +        *      error.
22955 +        * bdst < bfound
22956 +        *      found, after we create the file on bdst, it will be hidden.
22957 +        */
22958 +
22959 +       AuDebugOn(a->bfound == -1
22960 +                 && a->bwh != -1
22961 +                 && a->bwh <= a->mvd_bsrc);
22962 +       AuDebugOn(-1 < a->bfound
22963 +                 && a->bfound <= a->mvd_bsrc);
22964 +
22965 +       err = -EINVAL;
22966 +       if (a->bfound == -1
22967 +           && a->mvd_bsrc < a->bwh
22968 +           && a->bwh != -1
22969 +           && a->bwh < a->mvd_bdst) {
22970 +               a->mvd_errno = EAU_MVDOWN_WHITEOUT;
22971 +               AU_MVD_PR(dmsg, "bsrc %d, bdst %d, bfound %d, bwh %d\n",
22972 +                         a->mvd_bsrc, a->mvd_bdst, a->bfound, a->bwh);
22973 +               goto out;
22974 +       } else if (a->bfound != -1 && a->bfound < a->mvd_bdst) {
22975 +               a->mvd_errno = EAU_MVDOWN_UPPER;
22976 +               AU_MVD_PR(dmsg, "bdst %d, bfound %d\n",
22977 +                         a->mvd_bdst, a->bfound);
22978 +               goto out;
22979 +       }
22980 +
22981 +       err = 0; /* success */
22982 +
22983 +out:
22984 +       AuTraceErr(err);
22985 +       return err;
22986 +}
22987 +
22988 +static int au_mvd_args_exist(const unsigned char dmsg, struct au_mvd_args *a)
22989 +{
22990 +       int err;
22991 +
22992 +       err = 0;
22993 +       if (!(a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
22994 +           && a->bfound == a->mvd_bdst)
22995 +               err = -EEXIST;
22996 +       AuTraceErr(err);
22997 +       return err;
22998 +}
22999 +
23000 +static int au_mvd_args(const unsigned char dmsg, struct au_mvd_args *a)
23001 +{
23002 +       int err;
23003 +       struct au_branch *br;
23004 +
23005 +       err = -EISDIR;
23006 +       if (unlikely(S_ISDIR(a->inode->i_mode)))
23007 +               goto out;
23008 +
23009 +       err = -EINVAL;
23010 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_UPPER))
23011 +               a->mvd_bsrc = au_ibtop(a->inode);
23012 +       else {
23013 +               a->mvd_bsrc = au_br_index(a->sb, a->mvd_src_brid);
23014 +               if (unlikely(a->mvd_bsrc < 0
23015 +                            || (a->mvd_bsrc < au_dbtop(a->dentry)
23016 +                                || au_dbbot(a->dentry) < a->mvd_bsrc
23017 +                                || !au_h_dptr(a->dentry, a->mvd_bsrc))
23018 +                            || (a->mvd_bsrc < au_ibtop(a->inode)
23019 +                                || au_ibbot(a->inode) < a->mvd_bsrc
23020 +                                || !au_h_iptr(a->inode, a->mvd_bsrc)))) {
23021 +                       a->mvd_errno = EAU_MVDOWN_NOUPPER;
23022 +                       AU_MVD_PR(dmsg, "no upper\n");
23023 +                       goto out;
23024 +               }
23025 +       }
23026 +       if (unlikely(a->mvd_bsrc == au_sbbot(a->sb))) {
23027 +               a->mvd_errno = EAU_MVDOWN_BOTTOM;
23028 +               AU_MVD_PR(dmsg, "on the bottom\n");
23029 +               goto out;
23030 +       }
23031 +       a->mvd_h_src_inode = au_h_iptr(a->inode, a->mvd_bsrc);
23032 +       br = au_sbr(a->sb, a->mvd_bsrc);
23033 +       err = au_br_rdonly(br);
23034 +       if (!(a->mvdown.flags & AUFS_MVDOWN_ROUPPER)) {
23035 +               if (unlikely(err))
23036 +                       goto out;
23037 +       } else if (!(vfsub_native_ro(a->mvd_h_src_inode)
23038 +                    || IS_APPEND(a->mvd_h_src_inode))) {
23039 +               if (err)
23040 +                       a->mvdown.flags |= AUFS_MVDOWN_ROUPPER_R;
23041 +               /* go on */
23042 +       } else
23043 +               goto out;
23044 +
23045 +       err = -EINVAL;
23046 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_LOWER)) {
23047 +               a->mvd_bdst = find_lower_writable(a);
23048 +               if (unlikely(a->mvd_bdst < 0)) {
23049 +                       a->mvd_errno = EAU_MVDOWN_BOTTOM;
23050 +                       AU_MVD_PR(dmsg, "no writable lower branch\n");
23051 +                       goto out;
23052 +               }
23053 +       } else {
23054 +               a->mvd_bdst = au_br_index(a->sb, a->mvd_dst_brid);
23055 +               if (unlikely(a->mvd_bdst < 0
23056 +                            || au_sbbot(a->sb) < a->mvd_bdst)) {
23057 +                       a->mvd_errno = EAU_MVDOWN_NOLOWERBR;
23058 +                       AU_MVD_PR(dmsg, "no lower brid\n");
23059 +                       goto out;
23060 +               }
23061 +       }
23062 +
23063 +       err = au_mvd_args_busy(dmsg, a);
23064 +       if (!err)
23065 +               err = au_mvd_args_parent(dmsg, a);
23066 +       if (!err)
23067 +               err = au_mvd_args_intermediate(dmsg, a);
23068 +       if (!err)
23069 +               err = au_mvd_args_exist(dmsg, a);
23070 +       if (!err)
23071 +               AuDbg("b%d, b%d\n", a->mvd_bsrc, a->mvd_bdst);
23072 +
23073 +out:
23074 +       AuTraceErr(err);
23075 +       return err;
23076 +}
23077 +
23078 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *uarg)
23079 +{
23080 +       int err, e;
23081 +       unsigned char dmsg;
23082 +       struct au_mvd_args *args;
23083 +       struct inode *inode;
23084 +
23085 +       inode = d_inode(dentry);
23086 +       err = -EPERM;
23087 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
23088 +               goto out;
23089 +
23090 +       err = -ENOMEM;
23091 +       args = kmalloc(sizeof(*args), GFP_NOFS);
23092 +       if (unlikely(!args))
23093 +               goto out;
23094 +
23095 +       err = copy_from_user(&args->mvdown, uarg, sizeof(args->mvdown));
23096 +       if (!err)
23097 +               err = !access_ok(VERIFY_WRITE, uarg, sizeof(*uarg));
23098 +       if (unlikely(err)) {
23099 +               err = -EFAULT;
23100 +               AuTraceErr(err);
23101 +               goto out_free;
23102 +       }
23103 +       AuDbg("flags 0x%x\n", args->mvdown.flags);
23104 +       args->mvdown.flags &= ~(AUFS_MVDOWN_ROLOWER_R | AUFS_MVDOWN_ROUPPER_R);
23105 +       args->mvdown.au_errno = 0;
23106 +       args->dentry = dentry;
23107 +       args->inode = inode;
23108 +       args->sb = dentry->d_sb;
23109 +
23110 +       err = -ENOENT;
23111 +       dmsg = !!(args->mvdown.flags & AUFS_MVDOWN_DMSG);
23112 +       args->parent = dget_parent(dentry);
23113 +       args->dir = d_inode(args->parent);
23114 +       inode_lock_nested(args->dir, I_MUTEX_PARENT);
23115 +       dput(args->parent);
23116 +       if (unlikely(args->parent != dentry->d_parent)) {
23117 +               AU_MVD_PR(dmsg, "parent dir is moved\n");
23118 +               goto out_dir;
23119 +       }
23120 +
23121 +       inode_lock_nested(inode, I_MUTEX_CHILD);
23122 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_NOPLMW);
23123 +       if (unlikely(err))
23124 +               goto out_inode;
23125 +
23126 +       di_write_lock_parent(args->parent);
23127 +       err = au_mvd_args(dmsg, args);
23128 +       if (unlikely(err))
23129 +               goto out_parent;
23130 +
23131 +       err = au_do_mvdown(dmsg, args);
23132 +       if (unlikely(err))
23133 +               goto out_parent;
23134 +
23135 +       au_cpup_attr_timesizes(args->dir);
23136 +       au_cpup_attr_timesizes(inode);
23137 +       if (!(args->mvdown.flags & AUFS_MVDOWN_KUPPER))
23138 +               au_cpup_igen(inode, au_h_iptr(inode, args->mvd_bdst));
23139 +       /* au_digen_dec(dentry); */
23140 +
23141 +out_parent:
23142 +       di_write_unlock(args->parent);
23143 +       aufs_read_unlock(dentry, AuLock_DW);
23144 +out_inode:
23145 +       inode_unlock(inode);
23146 +out_dir:
23147 +       inode_unlock(args->dir);
23148 +out_free:
23149 +       e = copy_to_user(uarg, &args->mvdown, sizeof(args->mvdown));
23150 +       if (unlikely(e))
23151 +               err = -EFAULT;
23152 +       au_delayed_kfree(args);
23153 +out:
23154 +       AuTraceErr(err);
23155 +       return err;
23156 +}
23157 diff -urN /usr/share/empty/fs/aufs/opts.c linux/fs/aufs/opts.c
23158 --- /usr/share/empty/fs/aufs/opts.c     1970-01-01 01:00:00.000000000 +0100
23159 +++ linux/fs/aufs/opts.c        2016-10-09 16:55:36.496035060 +0200
23160 @@ -0,0 +1,1860 @@
23161 +/*
23162 + * Copyright (C) 2005-2016 Junjiro R. Okajima
23163 + *
23164 + * This program, aufs is free software; you can redistribute it and/or modify
23165 + * it under the terms of the GNU General Public License as published by
23166 + * the Free Software Foundation; either version 2 of the License, or
23167 + * (at your option) any later version.
23168 + *
23169 + * This program is distributed in the hope that it will be useful,
23170 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
23171 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
23172 + * GNU General Public License for more details.
23173 + *
23174 + * You should have received a copy of the GNU General Public License
23175 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
23176 + */
23177 +
23178 +/*
23179 + * mount options/flags
23180 + */
23181 +
23182 +#include <linux/namei.h>
23183 +#include <linux/types.h> /* a distribution requires */
23184 +#include <linux/parser.h>
23185 +#include "aufs.h"
23186 +
23187 +/* ---------------------------------------------------------------------- */
23188 +
23189 +enum {
23190 +       Opt_br,
23191 +       Opt_add, Opt_del, Opt_mod, Opt_append, Opt_prepend,
23192 +       Opt_idel, Opt_imod,
23193 +       Opt_dirwh, Opt_rdcache, Opt_rdblk, Opt_rdhash,
23194 +       Opt_rdblk_def, Opt_rdhash_def,
23195 +       Opt_xino, Opt_noxino,
23196 +       Opt_trunc_xino, Opt_trunc_xino_v, Opt_notrunc_xino,
23197 +       Opt_trunc_xino_path, Opt_itrunc_xino,
23198 +       Opt_trunc_xib, Opt_notrunc_xib,
23199 +       Opt_shwh, Opt_noshwh,
23200 +       Opt_plink, Opt_noplink, Opt_list_plink,
23201 +       Opt_udba,
23202 +       Opt_dio, Opt_nodio,
23203 +       Opt_diropq_a, Opt_diropq_w,
23204 +       Opt_warn_perm, Opt_nowarn_perm,
23205 +       Opt_wbr_copyup, Opt_wbr_create,
23206 +       Opt_fhsm_sec,
23207 +       Opt_verbose, Opt_noverbose,
23208 +       Opt_sum, Opt_nosum, Opt_wsum,
23209 +       Opt_dirperm1, Opt_nodirperm1,
23210 +       Opt_acl, Opt_noacl,
23211 +       Opt_tail, Opt_ignore, Opt_ignore_silent, Opt_err
23212 +};
23213 +
23214 +static match_table_t options = {
23215 +       {Opt_br, "br=%s"},
23216 +       {Opt_br, "br:%s"},
23217 +
23218 +       {Opt_add, "add=%d:%s"},
23219 +       {Opt_add, "add:%d:%s"},
23220 +       {Opt_add, "ins=%d:%s"},
23221 +       {Opt_add, "ins:%d:%s"},
23222 +       {Opt_append, "append=%s"},
23223 +       {Opt_append, "append:%s"},
23224 +       {Opt_prepend, "prepend=%s"},
23225 +       {Opt_prepend, "prepend:%s"},
23226 +
23227 +       {Opt_del, "del=%s"},
23228 +       {Opt_del, "del:%s"},
23229 +       /* {Opt_idel, "idel:%d"}, */
23230 +       {Opt_mod, "mod=%s"},
23231 +       {Opt_mod, "mod:%s"},
23232 +       /* {Opt_imod, "imod:%d:%s"}, */
23233 +
23234 +       {Opt_dirwh, "dirwh=%d"},
23235 +
23236 +       {Opt_xino, "xino=%s"},
23237 +       {Opt_noxino, "noxino"},
23238 +       {Opt_trunc_xino, "trunc_xino"},
23239 +       {Opt_trunc_xino_v, "trunc_xino_v=%d:%d"},
23240 +       {Opt_notrunc_xino, "notrunc_xino"},
23241 +       {Opt_trunc_xino_path, "trunc_xino=%s"},
23242 +       {Opt_itrunc_xino, "itrunc_xino=%d"},
23243 +       /* {Opt_zxino, "zxino=%s"}, */
23244 +       {Opt_trunc_xib, "trunc_xib"},
23245 +       {Opt_notrunc_xib, "notrunc_xib"},
23246 +
23247 +#ifdef CONFIG_PROC_FS
23248 +       {Opt_plink, "plink"},
23249 +#else
23250 +       {Opt_ignore_silent, "plink"},
23251 +#endif
23252 +
23253 +       {Opt_noplink, "noplink"},
23254 +
23255 +#ifdef CONFIG_AUFS_DEBUG
23256 +       {Opt_list_plink, "list_plink"},
23257 +#endif
23258 +
23259 +       {Opt_udba, "udba=%s"},
23260 +
23261 +       {Opt_dio, "dio"},
23262 +       {Opt_nodio, "nodio"},
23263 +
23264 +#ifdef CONFIG_AUFS_FHSM
23265 +       {Opt_fhsm_sec, "fhsm_sec=%d"},
23266 +#else
23267 +       {Opt_ignore_silent, "fhsm_sec=%d"},
23268 +#endif
23269 +
23270 +       {Opt_diropq_a, "diropq=always"},
23271 +       {Opt_diropq_a, "diropq=a"},
23272 +       {Opt_diropq_w, "diropq=whiteouted"},
23273 +       {Opt_diropq_w, "diropq=w"},
23274 +
23275 +       {Opt_warn_perm, "warn_perm"},
23276 +       {Opt_nowarn_perm, "nowarn_perm"},
23277 +
23278 +       /* keep them temporary */
23279 +       {Opt_ignore_silent, "nodlgt"},
23280 +       {Opt_ignore_silent, "clean_plink"},
23281 +
23282 +#ifdef CONFIG_AUFS_SHWH
23283 +       {Opt_shwh, "shwh"},
23284 +#endif
23285 +       {Opt_noshwh, "noshwh"},
23286 +
23287 +       {Opt_dirperm1, "dirperm1"},
23288 +       {Opt_nodirperm1, "nodirperm1"},
23289 +
23290 +       {Opt_verbose, "verbose"},
23291 +       {Opt_verbose, "v"},
23292 +       {Opt_noverbose, "noverbose"},
23293 +       {Opt_noverbose, "quiet"},
23294 +       {Opt_noverbose, "q"},
23295 +       {Opt_noverbose, "silent"},
23296 +
23297 +       {Opt_sum, "sum"},
23298 +       {Opt_nosum, "nosum"},
23299 +       {Opt_wsum, "wsum"},
23300 +
23301 +       {Opt_rdcache, "rdcache=%d"},
23302 +       {Opt_rdblk, "rdblk=%d"},
23303 +       {Opt_rdblk_def, "rdblk=def"},
23304 +       {Opt_rdhash, "rdhash=%d"},
23305 +       {Opt_rdhash_def, "rdhash=def"},
23306 +
23307 +       {Opt_wbr_create, "create=%s"},
23308 +       {Opt_wbr_create, "create_policy=%s"},
23309 +       {Opt_wbr_copyup, "cpup=%s"},
23310 +       {Opt_wbr_copyup, "copyup=%s"},
23311 +       {Opt_wbr_copyup, "copyup_policy=%s"},
23312 +
23313 +       /* generic VFS flag */
23314 +#ifdef CONFIG_FS_POSIX_ACL
23315 +       {Opt_acl, "acl"},
23316 +       {Opt_noacl, "noacl"},
23317 +#else
23318 +       {Opt_ignore_silent, "acl"},
23319 +       {Opt_ignore_silent, "noacl"},
23320 +#endif
23321 +
23322 +       /* internal use for the scripts */
23323 +       {Opt_ignore_silent, "si=%s"},
23324 +
23325 +       {Opt_br, "dirs=%s"},
23326 +       {Opt_ignore, "debug=%d"},
23327 +       {Opt_ignore, "delete=whiteout"},
23328 +       {Opt_ignore, "delete=all"},
23329 +       {Opt_ignore, "imap=%s"},
23330 +
23331 +       /* temporary workaround, due to old mount(8)? */
23332 +       {Opt_ignore_silent, "relatime"},
23333 +
23334 +       {Opt_err, NULL}
23335 +};
23336 +
23337 +/* ---------------------------------------------------------------------- */
23338 +
23339 +static const char *au_parser_pattern(int val, match_table_t tbl)
23340 +{
23341 +       struct match_token *p;
23342 +
23343 +       p = tbl;
23344 +       while (p->pattern) {
23345 +               if (p->token == val)
23346 +                       return p->pattern;
23347 +               p++;
23348 +       }
23349 +       BUG();
23350 +       return "??";
23351 +}
23352 +
23353 +static const char *au_optstr(int *val, match_table_t tbl)
23354 +{
23355 +       struct match_token *p;
23356 +       int v;
23357 +
23358 +       v = *val;
23359 +       if (!v)
23360 +               goto out;
23361 +       p = tbl;
23362 +       while (p->pattern) {
23363 +               if (p->token
23364 +                   && (v & p->token) == p->token) {
23365 +                       *val &= ~p->token;
23366 +                       return p->pattern;
23367 +               }
23368 +               p++;
23369 +       }
23370 +
23371 +out:
23372 +       return NULL;
23373 +}
23374 +
23375 +/* ---------------------------------------------------------------------- */
23376 +
23377 +static match_table_t brperm = {
23378 +       {AuBrPerm_RO, AUFS_BRPERM_RO},
23379 +       {AuBrPerm_RR, AUFS_BRPERM_RR},
23380 +       {AuBrPerm_RW, AUFS_BRPERM_RW},
23381 +       {0, NULL}
23382 +};
23383 +
23384 +static match_table_t brattr = {
23385 +       /* general */
23386 +       {AuBrAttr_COO_REG, AUFS_BRATTR_COO_REG},
23387 +       {AuBrAttr_COO_ALL, AUFS_BRATTR_COO_ALL},
23388 +       /* 'unpin' attrib is meaningless since linux-3.18-rc1 */
23389 +       {AuBrAttr_UNPIN, AUFS_BRATTR_UNPIN},
23390 +#ifdef CONFIG_AUFS_FHSM
23391 +       {AuBrAttr_FHSM, AUFS_BRATTR_FHSM},
23392 +#endif
23393 +#ifdef CONFIG_AUFS_XATTR
23394 +       {AuBrAttr_ICEX, AUFS_BRATTR_ICEX},
23395 +       {AuBrAttr_ICEX_SEC, AUFS_BRATTR_ICEX_SEC},
23396 +       {AuBrAttr_ICEX_SYS, AUFS_BRATTR_ICEX_SYS},
23397 +       {AuBrAttr_ICEX_TR, AUFS_BRATTR_ICEX_TR},
23398 +       {AuBrAttr_ICEX_USR, AUFS_BRATTR_ICEX_USR},
23399 +       {AuBrAttr_ICEX_OTH, AUFS_BRATTR_ICEX_OTH},
23400 +#endif
23401 +
23402 +       /* ro/rr branch */
23403 +       {AuBrRAttr_WH, AUFS_BRRATTR_WH},
23404 +
23405 +       /* rw branch */
23406 +       {AuBrWAttr_MOO, AUFS_BRWATTR_MOO},
23407 +       {AuBrWAttr_NoLinkWH, AUFS_BRWATTR_NLWH},
23408 +
23409 +       {0, NULL}
23410 +};
23411 +
23412 +static int br_attr_val(char *str, match_table_t table, substring_t args[])
23413 +{
23414 +       int attr, v;
23415 +       char *p;
23416 +
23417 +       attr = 0;
23418 +       do {
23419 +               p = strchr(str, '+');
23420 +               if (p)
23421 +                       *p = 0;
23422 +               v = match_token(str, table, args);
23423 +               if (v) {
23424 +                       if (v & AuBrAttr_CMOO_Mask)
23425 +                               attr &= ~AuBrAttr_CMOO_Mask;
23426 +                       attr |= v;
23427 +               } else {
23428 +                       if (p)
23429 +                               *p = '+';
23430 +                       pr_warn("ignored branch attribute %s\n", str);
23431 +                       break;
23432 +               }
23433 +               if (p)
23434 +                       str = p + 1;
23435 +       } while (p);
23436 +
23437 +       return attr;
23438 +}
23439 +
23440 +static int au_do_optstr_br_attr(au_br_perm_str_t *str, int perm)
23441 +{
23442 +       int sz;
23443 +       const char *p;
23444 +       char *q;
23445 +
23446 +       q = str->a;
23447 +       *q = 0;
23448 +       p = au_optstr(&perm, brattr);
23449 +       if (p) {
23450 +               sz = strlen(p);
23451 +               memcpy(q, p, sz + 1);
23452 +               q += sz;
23453 +       } else
23454 +               goto out;
23455 +
23456 +       do {
23457 +               p = au_optstr(&perm, brattr);
23458 +               if (p) {
23459 +                       *q++ = '+';
23460 +                       sz = strlen(p);
23461 +                       memcpy(q, p, sz + 1);
23462 +                       q += sz;
23463 +               }
23464 +       } while (p);
23465 +
23466 +out:
23467 +       return q - str->a;
23468 +}
23469 +
23470 +static int noinline_for_stack br_perm_val(char *perm)
23471 +{
23472 +       int val, bad, sz;
23473 +       char *p;
23474 +       substring_t args[MAX_OPT_ARGS];
23475 +       au_br_perm_str_t attr;
23476 +
23477 +       p = strchr(perm, '+');
23478 +       if (p)
23479 +               *p = 0;
23480 +       val = match_token(perm, brperm, args);
23481 +       if (!val) {
23482 +               if (p)
23483 +                       *p = '+';
23484 +               pr_warn("ignored branch permission %s\n", perm);
23485 +               val = AuBrPerm_RO;
23486 +               goto out;
23487 +       }
23488 +       if (!p)
23489 +               goto out;
23490 +
23491 +       val |= br_attr_val(p + 1, brattr, args);
23492 +
23493 +       bad = 0;
23494 +       switch (val & AuBrPerm_Mask) {
23495 +       case AuBrPerm_RO:
23496 +       case AuBrPerm_RR:
23497 +               bad = val & AuBrWAttr_Mask;
23498 +               val &= ~AuBrWAttr_Mask;
23499 +               break;
23500 +       case AuBrPerm_RW:
23501 +               bad = val & AuBrRAttr_Mask;
23502 +               val &= ~AuBrRAttr_Mask;
23503 +               break;
23504 +       }
23505 +
23506 +       /*
23507 +        * 'unpin' attrib becomes meaningless since linux-3.18-rc1, but aufs
23508 +        * does not treat it as an error, just warning.
23509 +        * this is a tiny guard for the user operation.
23510 +        */
23511 +       if (val & AuBrAttr_UNPIN) {
23512 +               bad |= AuBrAttr_UNPIN;
23513 +               val &= ~AuBrAttr_UNPIN;
23514 +       }
23515 +
23516 +       if (unlikely(bad)) {
23517 +               sz = au_do_optstr_br_attr(&attr, bad);
23518 +               AuDebugOn(!sz);
23519 +               pr_warn("ignored branch attribute %s\n", attr.a);
23520 +       }
23521 +
23522 +out:
23523 +       return val;
23524 +}
23525 +
23526 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm)
23527 +{
23528 +       au_br_perm_str_t attr;
23529 +       const char *p;
23530 +       char *q;
23531 +       int sz;
23532 +
23533 +       q = str->a;
23534 +       p = au_optstr(&perm, brperm);
23535 +       AuDebugOn(!p || !*p);
23536 +       sz = strlen(p);
23537 +       memcpy(q, p, sz + 1);
23538 +       q += sz;
23539 +
23540 +       sz = au_do_optstr_br_attr(&attr, perm);
23541 +       if (sz) {
23542 +               *q++ = '+';
23543 +               memcpy(q, attr.a, sz + 1);
23544 +       }
23545 +
23546 +       AuDebugOn(strlen(str->a) >= sizeof(str->a));
23547 +}
23548 +
23549 +/* ---------------------------------------------------------------------- */
23550 +
23551 +static match_table_t udbalevel = {
23552 +       {AuOpt_UDBA_REVAL, "reval"},
23553 +       {AuOpt_UDBA_NONE, "none"},
23554 +#ifdef CONFIG_AUFS_HNOTIFY
23555 +       {AuOpt_UDBA_HNOTIFY, "notify"}, /* abstraction */
23556 +#ifdef CONFIG_AUFS_HFSNOTIFY
23557 +       {AuOpt_UDBA_HNOTIFY, "fsnotify"},
23558 +#endif
23559 +#endif
23560 +       {-1, NULL}
23561 +};
23562 +
23563 +static int noinline_for_stack udba_val(char *str)
23564 +{
23565 +       substring_t args[MAX_OPT_ARGS];
23566 +
23567 +       return match_token(str, udbalevel, args);
23568 +}
23569 +
23570 +const char *au_optstr_udba(int udba)
23571 +{
23572 +       return au_parser_pattern(udba, udbalevel);
23573 +}
23574 +
23575 +/* ---------------------------------------------------------------------- */
23576 +
23577 +static match_table_t au_wbr_create_policy = {
23578 +       {AuWbrCreate_TDP, "tdp"},
23579 +       {AuWbrCreate_TDP, "top-down-parent"},
23580 +       {AuWbrCreate_RR, "rr"},
23581 +       {AuWbrCreate_RR, "round-robin"},
23582 +       {AuWbrCreate_MFS, "mfs"},
23583 +       {AuWbrCreate_MFS, "most-free-space"},
23584 +       {AuWbrCreate_MFSV, "mfs:%d"},
23585 +       {AuWbrCreate_MFSV, "most-free-space:%d"},
23586 +
23587 +       {AuWbrCreate_MFSRR, "mfsrr:%d"},
23588 +       {AuWbrCreate_MFSRRV, "mfsrr:%d:%d"},
23589 +       {AuWbrCreate_PMFS, "pmfs"},
23590 +       {AuWbrCreate_PMFSV, "pmfs:%d"},
23591 +       {AuWbrCreate_PMFSRR, "pmfsrr:%d"},
23592 +       {AuWbrCreate_PMFSRRV, "pmfsrr:%d:%d"},
23593 +
23594 +       {-1, NULL}
23595 +};
23596 +
23597 +/*
23598 + * cf. linux/lib/parser.c and cmdline.c
23599 + * gave up calling memparse() since it uses simple_strtoull() instead of
23600 + * kstrto...().
23601 + */
23602 +static int noinline_for_stack
23603 +au_match_ull(substring_t *s, unsigned long long *result)
23604 +{
23605 +       int err;
23606 +       unsigned int len;
23607 +       char a[32];
23608 +
23609 +       err = -ERANGE;
23610 +       len = s->to - s->from;
23611 +       if (len + 1 <= sizeof(a)) {
23612 +               memcpy(a, s->from, len);
23613 +               a[len] = '\0';
23614 +               err = kstrtoull(a, 0, result);
23615 +       }
23616 +       return err;
23617 +}
23618 +
23619 +static int au_wbr_mfs_wmark(substring_t *arg, char *str,
23620 +                           struct au_opt_wbr_create *create)
23621 +{
23622 +       int err;
23623 +       unsigned long long ull;
23624 +
23625 +       err = 0;
23626 +       if (!au_match_ull(arg, &ull))
23627 +               create->mfsrr_watermark = ull;
23628 +       else {
23629 +               pr_err("bad integer in %s\n", str);
23630 +               err = -EINVAL;
23631 +       }
23632 +
23633 +       return err;
23634 +}
23635 +
23636 +static int au_wbr_mfs_sec(substring_t *arg, char *str,
23637 +                         struct au_opt_wbr_create *create)
23638 +{
23639 +       int n, err;
23640 +
23641 +       err = 0;
23642 +       if (!match_int(arg, &n) && 0 <= n && n <= AUFS_MFS_MAX_SEC)
23643 +               create->mfs_second = n;
23644 +       else {
23645 +               pr_err("bad integer in %s\n", str);
23646 +               err = -EINVAL;
23647 +       }
23648 +
23649 +       return err;
23650 +}
23651 +
23652 +static int noinline_for_stack
23653 +au_wbr_create_val(char *str, struct au_opt_wbr_create *create)
23654 +{
23655 +       int err, e;
23656 +       substring_t args[MAX_OPT_ARGS];
23657 +
23658 +       err = match_token(str, au_wbr_create_policy, args);
23659 +       create->wbr_create = err;
23660 +       switch (err) {
23661 +       case AuWbrCreate_MFSRRV:
23662 +       case AuWbrCreate_PMFSRRV:
23663 +               e = au_wbr_mfs_wmark(&args[0], str, create);
23664 +               if (!e)
23665 +                       e = au_wbr_mfs_sec(&args[1], str, create);
23666 +               if (unlikely(e))
23667 +                       err = e;
23668 +               break;
23669 +       case AuWbrCreate_MFSRR:
23670 +       case AuWbrCreate_PMFSRR:
23671 +               e = au_wbr_mfs_wmark(&args[0], str, create);
23672 +               if (unlikely(e)) {
23673 +                       err = e;
23674 +                       break;
23675 +               }
23676 +               /*FALLTHROUGH*/
23677 +       case AuWbrCreate_MFS:
23678 +       case AuWbrCreate_PMFS:
23679 +               create->mfs_second = AUFS_MFS_DEF_SEC;
23680 +               break;
23681 +       case AuWbrCreate_MFSV:
23682 +       case AuWbrCreate_PMFSV:
23683 +               e = au_wbr_mfs_sec(&args[0], str, create);
23684 +               if (unlikely(e))
23685 +                       err = e;
23686 +               break;
23687 +       }
23688 +
23689 +       return err;
23690 +}
23691 +
23692 +const char *au_optstr_wbr_create(int wbr_create)
23693 +{
23694 +       return au_parser_pattern(wbr_create, au_wbr_create_policy);
23695 +}
23696 +
23697 +static match_table_t au_wbr_copyup_policy = {
23698 +       {AuWbrCopyup_TDP, "tdp"},
23699 +       {AuWbrCopyup_TDP, "top-down-parent"},
23700 +       {AuWbrCopyup_BUP, "bup"},
23701 +       {AuWbrCopyup_BUP, "bottom-up-parent"},
23702 +       {AuWbrCopyup_BU, "bu"},
23703 +       {AuWbrCopyup_BU, "bottom-up"},
23704 +       {-1, NULL}
23705 +};
23706 +
23707 +static int noinline_for_stack au_wbr_copyup_val(char *str)
23708 +{
23709 +       substring_t args[MAX_OPT_ARGS];
23710 +
23711 +       return match_token(str, au_wbr_copyup_policy, args);
23712 +}
23713 +
23714 +const char *au_optstr_wbr_copyup(int wbr_copyup)
23715 +{
23716 +       return au_parser_pattern(wbr_copyup, au_wbr_copyup_policy);
23717 +}
23718 +
23719 +/* ---------------------------------------------------------------------- */
23720 +
23721 +static const int lkup_dirflags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
23722 +
23723 +static void dump_opts(struct au_opts *opts)
23724 +{
23725 +#ifdef CONFIG_AUFS_DEBUG
23726 +       /* reduce stack space */
23727 +       union {
23728 +               struct au_opt_add *add;
23729 +               struct au_opt_del *del;
23730 +               struct au_opt_mod *mod;
23731 +               struct au_opt_xino *xino;
23732 +               struct au_opt_xino_itrunc *xino_itrunc;
23733 +               struct au_opt_wbr_create *create;
23734 +       } u;
23735 +       struct au_opt *opt;
23736 +
23737 +       opt = opts->opt;
23738 +       while (opt->type != Opt_tail) {
23739 +               switch (opt->type) {
23740 +               case Opt_add:
23741 +                       u.add = &opt->add;
23742 +                       AuDbg("add {b%d, %s, 0x%x, %p}\n",
23743 +                                 u.add->bindex, u.add->pathname, u.add->perm,
23744 +                                 u.add->path.dentry);
23745 +                       break;
23746 +               case Opt_del:
23747 +               case Opt_idel:
23748 +                       u.del = &opt->del;
23749 +                       AuDbg("del {%s, %p}\n",
23750 +                             u.del->pathname, u.del->h_path.dentry);
23751 +                       break;
23752 +               case Opt_mod:
23753 +               case Opt_imod:
23754 +                       u.mod = &opt->mod;
23755 +                       AuDbg("mod {%s, 0x%x, %p}\n",
23756 +                                 u.mod->path, u.mod->perm, u.mod->h_root);
23757 +                       break;
23758 +               case Opt_append:
23759 +                       u.add = &opt->add;
23760 +                       AuDbg("append {b%d, %s, 0x%x, %p}\n",
23761 +                                 u.add->bindex, u.add->pathname, u.add->perm,
23762 +                                 u.add->path.dentry);
23763 +                       break;
23764 +               case Opt_prepend:
23765 +                       u.add = &opt->add;
23766 +                       AuDbg("prepend {b%d, %s, 0x%x, %p}\n",
23767 +                                 u.add->bindex, u.add->pathname, u.add->perm,
23768 +                                 u.add->path.dentry);
23769 +                       break;
23770 +               case Opt_dirwh:
23771 +                       AuDbg("dirwh %d\n", opt->dirwh);
23772 +                       break;
23773 +               case Opt_rdcache:
23774 +                       AuDbg("rdcache %d\n", opt->rdcache);
23775 +                       break;
23776 +               case Opt_rdblk:
23777 +                       AuDbg("rdblk %u\n", opt->rdblk);
23778 +                       break;
23779 +               case Opt_rdblk_def:
23780 +                       AuDbg("rdblk_def\n");
23781 +                       break;
23782 +               case Opt_rdhash:
23783 +                       AuDbg("rdhash %u\n", opt->rdhash);
23784 +                       break;
23785 +               case Opt_rdhash_def:
23786 +                       AuDbg("rdhash_def\n");
23787 +                       break;
23788 +               case Opt_xino:
23789 +                       u.xino = &opt->xino;
23790 +                       AuDbg("xino {%s %pD}\n", u.xino->path, u.xino->file);
23791 +                       break;
23792 +               case Opt_trunc_xino:
23793 +                       AuLabel(trunc_xino);
23794 +                       break;
23795 +               case Opt_notrunc_xino:
23796 +                       AuLabel(notrunc_xino);
23797 +                       break;
23798 +               case Opt_trunc_xino_path:
23799 +               case Opt_itrunc_xino:
23800 +                       u.xino_itrunc = &opt->xino_itrunc;
23801 +                       AuDbg("trunc_xino %d\n", u.xino_itrunc->bindex);
23802 +                       break;
23803 +               case Opt_noxino:
23804 +                       AuLabel(noxino);
23805 +                       break;
23806 +               case Opt_trunc_xib:
23807 +                       AuLabel(trunc_xib);
23808 +                       break;
23809 +               case Opt_notrunc_xib:
23810 +                       AuLabel(notrunc_xib);
23811 +                       break;
23812 +               case Opt_shwh:
23813 +                       AuLabel(shwh);
23814 +                       break;
23815 +               case Opt_noshwh:
23816 +                       AuLabel(noshwh);
23817 +                       break;
23818 +               case Opt_dirperm1:
23819 +                       AuLabel(dirperm1);
23820 +                       break;
23821 +               case Opt_nodirperm1:
23822 +                       AuLabel(nodirperm1);
23823 +                       break;
23824 +               case Opt_plink:
23825 +                       AuLabel(plink);
23826 +                       break;
23827 +               case Opt_noplink:
23828 +                       AuLabel(noplink);
23829 +                       break;
23830 +               case Opt_list_plink:
23831 +                       AuLabel(list_plink);
23832 +                       break;
23833 +               case Opt_udba:
23834 +                       AuDbg("udba %d, %s\n",
23835 +                                 opt->udba, au_optstr_udba(opt->udba));
23836 +                       break;
23837 +               case Opt_dio:
23838 +                       AuLabel(dio);
23839 +                       break;
23840 +               case Opt_nodio:
23841 +                       AuLabel(nodio);
23842 +                       break;
23843 +               case Opt_diropq_a:
23844 +                       AuLabel(diropq_a);
23845 +                       break;
23846 +               case Opt_diropq_w:
23847 +                       AuLabel(diropq_w);
23848 +                       break;
23849 +               case Opt_warn_perm:
23850 +                       AuLabel(warn_perm);
23851 +                       break;
23852 +               case Opt_nowarn_perm:
23853 +                       AuLabel(nowarn_perm);
23854 +                       break;
23855 +               case Opt_verbose:
23856 +                       AuLabel(verbose);
23857 +                       break;
23858 +               case Opt_noverbose:
23859 +                       AuLabel(noverbose);
23860 +                       break;
23861 +               case Opt_sum:
23862 +                       AuLabel(sum);
23863 +                       break;
23864 +               case Opt_nosum:
23865 +                       AuLabel(nosum);
23866 +                       break;
23867 +               case Opt_wsum:
23868 +                       AuLabel(wsum);
23869 +                       break;
23870 +               case Opt_wbr_create:
23871 +                       u.create = &opt->wbr_create;
23872 +                       AuDbg("create %d, %s\n", u.create->wbr_create,
23873 +                                 au_optstr_wbr_create(u.create->wbr_create));
23874 +                       switch (u.create->wbr_create) {
23875 +                       case AuWbrCreate_MFSV:
23876 +                       case AuWbrCreate_PMFSV:
23877 +                               AuDbg("%d sec\n", u.create->mfs_second);
23878 +                               break;
23879 +                       case AuWbrCreate_MFSRR:
23880 +                               AuDbg("%llu watermark\n",
23881 +                                         u.create->mfsrr_watermark);
23882 +                               break;
23883 +                       case AuWbrCreate_MFSRRV:
23884 +                       case AuWbrCreate_PMFSRRV:
23885 +                               AuDbg("%llu watermark, %d sec\n",
23886 +                                         u.create->mfsrr_watermark,
23887 +                                         u.create->mfs_second);
23888 +                               break;
23889 +                       }
23890 +                       break;
23891 +               case Opt_wbr_copyup:
23892 +                       AuDbg("copyup %d, %s\n", opt->wbr_copyup,
23893 +                                 au_optstr_wbr_copyup(opt->wbr_copyup));
23894 +                       break;
23895 +               case Opt_fhsm_sec:
23896 +                       AuDbg("fhsm_sec %u\n", opt->fhsm_second);
23897 +                       break;
23898 +               case Opt_acl:
23899 +                       AuLabel(acl);
23900 +                       break;
23901 +               case Opt_noacl:
23902 +                       AuLabel(noacl);
23903 +                       break;
23904 +               default:
23905 +                       BUG();
23906 +               }
23907 +               opt++;
23908 +       }
23909 +#endif
23910 +}
23911 +
23912 +void au_opts_free(struct au_opts *opts)
23913 +{
23914 +       struct au_opt *opt;
23915 +
23916 +       opt = opts->opt;
23917 +       while (opt->type != Opt_tail) {
23918 +               switch (opt->type) {
23919 +               case Opt_add:
23920 +               case Opt_append:
23921 +               case Opt_prepend:
23922 +                       path_put(&opt->add.path);
23923 +                       break;
23924 +               case Opt_del:
23925 +               case Opt_idel:
23926 +                       path_put(&opt->del.h_path);
23927 +                       break;
23928 +               case Opt_mod:
23929 +               case Opt_imod:
23930 +                       dput(opt->mod.h_root);
23931 +                       break;
23932 +               case Opt_xino:
23933 +                       fput(opt->xino.file);
23934 +                       break;
23935 +               }
23936 +               opt++;
23937 +       }
23938 +}
23939 +
23940 +static int opt_add(struct au_opt *opt, char *opt_str, unsigned long sb_flags,
23941 +                  aufs_bindex_t bindex)
23942 +{
23943 +       int err;
23944 +       struct au_opt_add *add = &opt->add;
23945 +       char *p;
23946 +
23947 +       add->bindex = bindex;
23948 +       add->perm = AuBrPerm_RO;
23949 +       add->pathname = opt_str;
23950 +       p = strchr(opt_str, '=');
23951 +       if (p) {
23952 +               *p++ = 0;
23953 +               if (*p)
23954 +                       add->perm = br_perm_val(p);
23955 +       }
23956 +
23957 +       err = vfsub_kern_path(add->pathname, lkup_dirflags, &add->path);
23958 +       if (!err) {
23959 +               if (!p) {
23960 +                       add->perm = AuBrPerm_RO;
23961 +                       if (au_test_fs_rr(add->path.dentry->d_sb))
23962 +                               add->perm = AuBrPerm_RR;
23963 +                       else if (!bindex && !(sb_flags & MS_RDONLY))
23964 +                               add->perm = AuBrPerm_RW;
23965 +               }
23966 +               opt->type = Opt_add;
23967 +               goto out;
23968 +       }
23969 +       pr_err("lookup failed %s (%d)\n", add->pathname, err);
23970 +       err = -EINVAL;
23971 +
23972 +out:
23973 +       return err;
23974 +}
23975 +
23976 +static int au_opts_parse_del(struct au_opt_del *del, substring_t args[])
23977 +{
23978 +       int err;
23979 +
23980 +       del->pathname = args[0].from;
23981 +       AuDbg("del path %s\n", del->pathname);
23982 +
23983 +       err = vfsub_kern_path(del->pathname, lkup_dirflags, &del->h_path);
23984 +       if (unlikely(err))
23985 +               pr_err("lookup failed %s (%d)\n", del->pathname, err);
23986 +
23987 +       return err;
23988 +}
23989 +
23990 +#if 0 /* reserved for future use */
23991 +static int au_opts_parse_idel(struct super_block *sb, aufs_bindex_t bindex,
23992 +                             struct au_opt_del *del, substring_t args[])
23993 +{
23994 +       int err;
23995 +       struct dentry *root;
23996 +
23997 +       err = -EINVAL;
23998 +       root = sb->s_root;
23999 +       aufs_read_lock(root, AuLock_FLUSH);
24000 +       if (bindex < 0 || au_sbbot(sb) < bindex) {
24001 +               pr_err("out of bounds, %d\n", bindex);
24002 +               goto out;
24003 +       }
24004 +
24005 +       err = 0;
24006 +       del->h_path.dentry = dget(au_h_dptr(root, bindex));
24007 +       del->h_path.mnt = mntget(au_sbr_mnt(sb, bindex));
24008 +
24009 +out:
24010 +       aufs_read_unlock(root, !AuLock_IR);
24011 +       return err;
24012 +}
24013 +#endif
24014 +
24015 +static int noinline_for_stack
24016 +au_opts_parse_mod(struct au_opt_mod *mod, substring_t args[])
24017 +{
24018 +       int err;
24019 +       struct path path;
24020 +       char *p;
24021 +
24022 +       err = -EINVAL;
24023 +       mod->path = args[0].from;
24024 +       p = strchr(mod->path, '=');
24025 +       if (unlikely(!p)) {
24026 +               pr_err("no permssion %s\n", args[0].from);
24027 +               goto out;
24028 +       }
24029 +
24030 +       *p++ = 0;
24031 +       err = vfsub_kern_path(mod->path, lkup_dirflags, &path);
24032 +       if (unlikely(err)) {
24033 +               pr_err("lookup failed %s (%d)\n", mod->path, err);
24034 +               goto out;
24035 +       }
24036 +
24037 +       mod->perm = br_perm_val(p);
24038 +       AuDbg("mod path %s, perm 0x%x, %s\n", mod->path, mod->perm, p);
24039 +       mod->h_root = dget(path.dentry);
24040 +       path_put(&path);
24041 +
24042 +out:
24043 +       return err;
24044 +}
24045 +
24046 +#if 0 /* reserved for future use */
24047 +static int au_opts_parse_imod(struct super_block *sb, aufs_bindex_t bindex,
24048 +                             struct au_opt_mod *mod, substring_t args[])
24049 +{
24050 +       int err;
24051 +       struct dentry *root;
24052 +
24053 +       err = -EINVAL;
24054 +       root = sb->s_root;
24055 +       aufs_read_lock(root, AuLock_FLUSH);
24056 +       if (bindex < 0 || au_sbbot(sb) < bindex) {
24057 +               pr_err("out of bounds, %d\n", bindex);
24058 +               goto out;
24059 +       }
24060 +
24061 +       err = 0;
24062 +       mod->perm = br_perm_val(args[1].from);
24063 +       AuDbg("mod path %s, perm 0x%x, %s\n",
24064 +             mod->path, mod->perm, args[1].from);
24065 +       mod->h_root = dget(au_h_dptr(root, bindex));
24066 +
24067 +out:
24068 +       aufs_read_unlock(root, !AuLock_IR);
24069 +       return err;
24070 +}
24071 +#endif
24072 +
24073 +static int au_opts_parse_xino(struct super_block *sb, struct au_opt_xino *xino,
24074 +                             substring_t args[])
24075 +{
24076 +       int err;
24077 +       struct file *file;
24078 +
24079 +       file = au_xino_create(sb, args[0].from, /*silent*/0);
24080 +       err = PTR_ERR(file);
24081 +       if (IS_ERR(file))
24082 +               goto out;
24083 +
24084 +       err = -EINVAL;
24085 +       if (unlikely(file->f_path.dentry->d_sb == sb)) {
24086 +               fput(file);
24087 +               pr_err("%s must be outside\n", args[0].from);
24088 +               goto out;
24089 +       }
24090 +
24091 +       err = 0;
24092 +       xino->file = file;
24093 +       xino->path = args[0].from;
24094 +
24095 +out:
24096 +       return err;
24097 +}
24098 +
24099 +static int noinline_for_stack
24100 +au_opts_parse_xino_itrunc_path(struct super_block *sb,
24101 +                              struct au_opt_xino_itrunc *xino_itrunc,
24102 +                              substring_t args[])
24103 +{
24104 +       int err;
24105 +       aufs_bindex_t bbot, bindex;
24106 +       struct path path;
24107 +       struct dentry *root;
24108 +
24109 +       err = vfsub_kern_path(args[0].from, lkup_dirflags, &path);
24110 +       if (unlikely(err)) {
24111 +               pr_err("lookup failed %s (%d)\n", args[0].from, err);
24112 +               goto out;
24113 +       }
24114 +
24115 +       xino_itrunc->bindex = -1;
24116 +       root = sb->s_root;
24117 +       aufs_read_lock(root, AuLock_FLUSH);
24118 +       bbot = au_sbbot(sb);
24119 +       for (bindex = 0; bindex <= bbot; bindex++) {
24120 +               if (au_h_dptr(root, bindex) == path.dentry) {
24121 +                       xino_itrunc->bindex = bindex;
24122 +                       break;
24123 +               }
24124 +       }
24125 +       aufs_read_unlock(root, !AuLock_IR);
24126 +       path_put(&path);
24127 +
24128 +       if (unlikely(xino_itrunc->bindex < 0)) {
24129 +               pr_err("no such branch %s\n", args[0].from);
24130 +               err = -EINVAL;
24131 +       }
24132 +
24133 +out:
24134 +       return err;
24135 +}
24136 +
24137 +/* called without aufs lock */
24138 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
24139 +{
24140 +       int err, n, token;
24141 +       aufs_bindex_t bindex;
24142 +       unsigned char skipped;
24143 +       struct dentry *root;
24144 +       struct au_opt *opt, *opt_tail;
24145 +       char *opt_str;
24146 +       /* reduce the stack space */
24147 +       union {
24148 +               struct au_opt_xino_itrunc *xino_itrunc;
24149 +               struct au_opt_wbr_create *create;
24150 +       } u;
24151 +       struct {
24152 +               substring_t args[MAX_OPT_ARGS];
24153 +       } *a;
24154 +
24155 +       err = -ENOMEM;
24156 +       a = kmalloc(sizeof(*a), GFP_NOFS);
24157 +       if (unlikely(!a))
24158 +               goto out;
24159 +
24160 +       root = sb->s_root;
24161 +       err = 0;
24162 +       bindex = 0;
24163 +       opt = opts->opt;
24164 +       opt_tail = opt + opts->max_opt - 1;
24165 +       opt->type = Opt_tail;
24166 +       while (!err && (opt_str = strsep(&str, ",")) && *opt_str) {
24167 +               err = -EINVAL;
24168 +               skipped = 0;
24169 +               token = match_token(opt_str, options, a->args);
24170 +               switch (token) {
24171 +               case Opt_br:
24172 +                       err = 0;
24173 +                       while (!err && (opt_str = strsep(&a->args[0].from, ":"))
24174 +                              && *opt_str) {
24175 +                               err = opt_add(opt, opt_str, opts->sb_flags,
24176 +                                             bindex++);
24177 +                               if (unlikely(!err && ++opt > opt_tail)) {
24178 +                                       err = -E2BIG;
24179 +                                       break;
24180 +                               }
24181 +                               opt->type = Opt_tail;
24182 +                               skipped = 1;
24183 +                       }
24184 +                       break;
24185 +               case Opt_add:
24186 +                       if (unlikely(match_int(&a->args[0], &n))) {
24187 +                               pr_err("bad integer in %s\n", opt_str);
24188 +                               break;
24189 +                       }
24190 +                       bindex = n;
24191 +                       err = opt_add(opt, a->args[1].from, opts->sb_flags,
24192 +                                     bindex);
24193 +                       if (!err)
24194 +                               opt->type = token;
24195 +                       break;
24196 +               case Opt_append:
24197 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
24198 +                                     /*dummy bindex*/1);
24199 +                       if (!err)
24200 +                               opt->type = token;
24201 +                       break;
24202 +               case Opt_prepend:
24203 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
24204 +                                     /*bindex*/0);
24205 +                       if (!err)
24206 +                               opt->type = token;
24207 +                       break;
24208 +               case Opt_del:
24209 +                       err = au_opts_parse_del(&opt->del, a->args);
24210 +                       if (!err)
24211 +                               opt->type = token;
24212 +                       break;
24213 +#if 0 /* reserved for future use */
24214 +               case Opt_idel:
24215 +                       del->pathname = "(indexed)";
24216 +                       if (unlikely(match_int(&args[0], &n))) {
24217 +                               pr_err("bad integer in %s\n", opt_str);
24218 +                               break;
24219 +                       }
24220 +                       err = au_opts_parse_idel(sb, n, &opt->del, a->args);
24221 +                       if (!err)
24222 +                               opt->type = token;
24223 +                       break;
24224 +#endif
24225 +               case Opt_mod:
24226 +                       err = au_opts_parse_mod(&opt->mod, a->args);
24227 +                       if (!err)
24228 +                               opt->type = token;
24229 +                       break;
24230 +#ifdef IMOD /* reserved for future use */
24231 +               case Opt_imod:
24232 +                       u.mod->path = "(indexed)";
24233 +                       if (unlikely(match_int(&a->args[0], &n))) {
24234 +                               pr_err("bad integer in %s\n", opt_str);
24235 +                               break;
24236 +                       }
24237 +                       err = au_opts_parse_imod(sb, n, &opt->mod, a->args);
24238 +                       if (!err)
24239 +                               opt->type = token;
24240 +                       break;
24241 +#endif
24242 +               case Opt_xino:
24243 +                       err = au_opts_parse_xino(sb, &opt->xino, a->args);
24244 +                       if (!err)
24245 +                               opt->type = token;
24246 +                       break;
24247 +
24248 +               case Opt_trunc_xino_path:
24249 +                       err = au_opts_parse_xino_itrunc_path
24250 +                               (sb, &opt->xino_itrunc, a->args);
24251 +                       if (!err)
24252 +                               opt->type = token;
24253 +                       break;
24254 +
24255 +               case Opt_itrunc_xino:
24256 +                       u.xino_itrunc = &opt->xino_itrunc;
24257 +                       if (unlikely(match_int(&a->args[0], &n))) {
24258 +                               pr_err("bad integer in %s\n", opt_str);
24259 +                               break;
24260 +                       }
24261 +                       u.xino_itrunc->bindex = n;
24262 +                       aufs_read_lock(root, AuLock_FLUSH);
24263 +                       if (n < 0 || au_sbbot(sb) < n) {
24264 +                               pr_err("out of bounds, %d\n", n);
24265 +                               aufs_read_unlock(root, !AuLock_IR);
24266 +                               break;
24267 +                       }
24268 +                       aufs_read_unlock(root, !AuLock_IR);
24269 +                       err = 0;
24270 +                       opt->type = token;
24271 +                       break;
24272 +
24273 +               case Opt_dirwh:
24274 +                       if (unlikely(match_int(&a->args[0], &opt->dirwh)))
24275 +                               break;
24276 +                       err = 0;
24277 +                       opt->type = token;
24278 +                       break;
24279 +
24280 +               case Opt_rdcache:
24281 +                       if (unlikely(match_int(&a->args[0], &n))) {
24282 +                               pr_err("bad integer in %s\n", opt_str);
24283 +                               break;
24284 +                       }
24285 +                       if (unlikely(n > AUFS_RDCACHE_MAX)) {
24286 +                               pr_err("rdcache must be smaller than %d\n",
24287 +                                      AUFS_RDCACHE_MAX);
24288 +                               break;
24289 +                       }
24290 +                       opt->rdcache = n;
24291 +                       err = 0;
24292 +                       opt->type = token;
24293 +                       break;
24294 +               case Opt_rdblk:
24295 +                       if (unlikely(match_int(&a->args[0], &n)
24296 +                                    || n < 0
24297 +                                    || n > KMALLOC_MAX_SIZE)) {
24298 +                               pr_err("bad integer in %s\n", opt_str);
24299 +                               break;
24300 +                       }
24301 +                       if (unlikely(n && n < NAME_MAX)) {
24302 +                               pr_err("rdblk must be larger than %d\n",
24303 +                                      NAME_MAX);
24304 +                               break;
24305 +                       }
24306 +                       opt->rdblk = n;
24307 +                       err = 0;
24308 +                       opt->type = token;
24309 +                       break;
24310 +               case Opt_rdhash:
24311 +                       if (unlikely(match_int(&a->args[0], &n)
24312 +                                    || n < 0
24313 +                                    || n * sizeof(struct hlist_head)
24314 +                                    > KMALLOC_MAX_SIZE)) {
24315 +                               pr_err("bad integer in %s\n", opt_str);
24316 +                               break;
24317 +                       }
24318 +                       opt->rdhash = n;
24319 +                       err = 0;
24320 +                       opt->type = token;
24321 +                       break;
24322 +
24323 +               case Opt_trunc_xino:
24324 +               case Opt_notrunc_xino:
24325 +               case Opt_noxino:
24326 +               case Opt_trunc_xib:
24327 +               case Opt_notrunc_xib:
24328 +               case Opt_shwh:
24329 +               case Opt_noshwh:
24330 +               case Opt_dirperm1:
24331 +               case Opt_nodirperm1:
24332 +               case Opt_plink:
24333 +               case Opt_noplink:
24334 +               case Opt_list_plink:
24335 +               case Opt_dio:
24336 +               case Opt_nodio:
24337 +               case Opt_diropq_a:
24338 +               case Opt_diropq_w:
24339 +               case Opt_warn_perm:
24340 +               case Opt_nowarn_perm:
24341 +               case Opt_verbose:
24342 +               case Opt_noverbose:
24343 +               case Opt_sum:
24344 +               case Opt_nosum:
24345 +               case Opt_wsum:
24346 +               case Opt_rdblk_def:
24347 +               case Opt_rdhash_def:
24348 +               case Opt_acl:
24349 +               case Opt_noacl:
24350 +                       err = 0;
24351 +                       opt->type = token;
24352 +                       break;
24353 +
24354 +               case Opt_udba:
24355 +                       opt->udba = udba_val(a->args[0].from);
24356 +                       if (opt->udba >= 0) {
24357 +                               err = 0;
24358 +                               opt->type = token;
24359 +                       } else
24360 +                               pr_err("wrong value, %s\n", opt_str);
24361 +                       break;
24362 +
24363 +               case Opt_wbr_create:
24364 +                       u.create = &opt->wbr_create;
24365 +                       u.create->wbr_create
24366 +                               = au_wbr_create_val(a->args[0].from, u.create);
24367 +                       if (u.create->wbr_create >= 0) {
24368 +                               err = 0;
24369 +                               opt->type = token;
24370 +                       } else
24371 +                               pr_err("wrong value, %s\n", opt_str);
24372 +                       break;
24373 +               case Opt_wbr_copyup:
24374 +                       opt->wbr_copyup = au_wbr_copyup_val(a->args[0].from);
24375 +                       if (opt->wbr_copyup >= 0) {
24376 +                               err = 0;
24377 +                               opt->type = token;
24378 +                       } else
24379 +                               pr_err("wrong value, %s\n", opt_str);
24380 +                       break;
24381 +
24382 +               case Opt_fhsm_sec:
24383 +                       if (unlikely(match_int(&a->args[0], &n)
24384 +                                    || n < 0)) {
24385 +                               pr_err("bad integer in %s\n", opt_str);
24386 +                               break;
24387 +                       }
24388 +                       if (sysaufs_brs) {
24389 +                               opt->fhsm_second = n;
24390 +                               opt->type = token;
24391 +                       } else
24392 +                               pr_warn("ignored %s\n", opt_str);
24393 +                       err = 0;
24394 +                       break;
24395 +
24396 +               case Opt_ignore:
24397 +                       pr_warn("ignored %s\n", opt_str);
24398 +                       /*FALLTHROUGH*/
24399 +               case Opt_ignore_silent:
24400 +                       skipped = 1;
24401 +                       err = 0;
24402 +                       break;
24403 +               case Opt_err:
24404 +                       pr_err("unknown option %s\n", opt_str);
24405 +                       break;
24406 +               }
24407 +
24408 +               if (!err && !skipped) {
24409 +                       if (unlikely(++opt > opt_tail)) {
24410 +                               err = -E2BIG;
24411 +                               opt--;
24412 +                               opt->type = Opt_tail;
24413 +                               break;
24414 +                       }
24415 +                       opt->type = Opt_tail;
24416 +               }
24417 +       }
24418 +
24419 +       au_delayed_kfree(a);
24420 +       dump_opts(opts);
24421 +       if (unlikely(err))
24422 +               au_opts_free(opts);
24423 +
24424 +out:
24425 +       return err;
24426 +}
24427 +
24428 +static int au_opt_wbr_create(struct super_block *sb,
24429 +                            struct au_opt_wbr_create *create)
24430 +{
24431 +       int err;
24432 +       struct au_sbinfo *sbinfo;
24433 +
24434 +       SiMustWriteLock(sb);
24435 +
24436 +       err = 1; /* handled */
24437 +       sbinfo = au_sbi(sb);
24438 +       if (sbinfo->si_wbr_create_ops->fin) {
24439 +               err = sbinfo->si_wbr_create_ops->fin(sb);
24440 +               if (!err)
24441 +                       err = 1;
24442 +       }
24443 +
24444 +       sbinfo->si_wbr_create = create->wbr_create;
24445 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + create->wbr_create;
24446 +       switch (create->wbr_create) {
24447 +       case AuWbrCreate_MFSRRV:
24448 +       case AuWbrCreate_MFSRR:
24449 +       case AuWbrCreate_PMFSRR:
24450 +       case AuWbrCreate_PMFSRRV:
24451 +               sbinfo->si_wbr_mfs.mfsrr_watermark = create->mfsrr_watermark;
24452 +               /*FALLTHROUGH*/
24453 +       case AuWbrCreate_MFS:
24454 +       case AuWbrCreate_MFSV:
24455 +       case AuWbrCreate_PMFS:
24456 +       case AuWbrCreate_PMFSV:
24457 +               sbinfo->si_wbr_mfs.mfs_expire
24458 +                       = msecs_to_jiffies(create->mfs_second * MSEC_PER_SEC);
24459 +               break;
24460 +       }
24461 +
24462 +       if (sbinfo->si_wbr_create_ops->init)
24463 +               sbinfo->si_wbr_create_ops->init(sb); /* ignore */
24464 +
24465 +       return err;
24466 +}
24467 +
24468 +/*
24469 + * returns,
24470 + * plus: processed without an error
24471 + * zero: unprocessed
24472 + */
24473 +static int au_opt_simple(struct super_block *sb, struct au_opt *opt,
24474 +                        struct au_opts *opts)
24475 +{
24476 +       int err;
24477 +       struct au_sbinfo *sbinfo;
24478 +
24479 +       SiMustWriteLock(sb);
24480 +
24481 +       err = 1; /* handled */
24482 +       sbinfo = au_sbi(sb);
24483 +       switch (opt->type) {
24484 +       case Opt_udba:
24485 +               sbinfo->si_mntflags &= ~AuOptMask_UDBA;
24486 +               sbinfo->si_mntflags |= opt->udba;
24487 +               opts->given_udba |= opt->udba;
24488 +               break;
24489 +
24490 +       case Opt_plink:
24491 +               au_opt_set(sbinfo->si_mntflags, PLINK);
24492 +               break;
24493 +       case Opt_noplink:
24494 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
24495 +                       au_plink_put(sb, /*verbose*/1);
24496 +               au_opt_clr(sbinfo->si_mntflags, PLINK);
24497 +               break;
24498 +       case Opt_list_plink:
24499 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
24500 +                       au_plink_list(sb);
24501 +               break;
24502 +
24503 +       case Opt_dio:
24504 +               au_opt_set(sbinfo->si_mntflags, DIO);
24505 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
24506 +               break;
24507 +       case Opt_nodio:
24508 +               au_opt_clr(sbinfo->si_mntflags, DIO);
24509 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
24510 +               break;
24511 +
24512 +       case Opt_fhsm_sec:
24513 +               au_fhsm_set(sbinfo, opt->fhsm_second);
24514 +               break;
24515 +
24516 +       case Opt_diropq_a:
24517 +               au_opt_set(sbinfo->si_mntflags, ALWAYS_DIROPQ);
24518 +               break;
24519 +       case Opt_diropq_w:
24520 +               au_opt_clr(sbinfo->si_mntflags, ALWAYS_DIROPQ);
24521 +               break;
24522 +
24523 +       case Opt_warn_perm:
24524 +               au_opt_set(sbinfo->si_mntflags, WARN_PERM);
24525 +               break;
24526 +       case Opt_nowarn_perm:
24527 +               au_opt_clr(sbinfo->si_mntflags, WARN_PERM);
24528 +               break;
24529 +
24530 +       case Opt_verbose:
24531 +               au_opt_set(sbinfo->si_mntflags, VERBOSE);
24532 +               break;
24533 +       case Opt_noverbose:
24534 +               au_opt_clr(sbinfo->si_mntflags, VERBOSE);
24535 +               break;
24536 +
24537 +       case Opt_sum:
24538 +               au_opt_set(sbinfo->si_mntflags, SUM);
24539 +               break;
24540 +       case Opt_wsum:
24541 +               au_opt_clr(sbinfo->si_mntflags, SUM);
24542 +               au_opt_set(sbinfo->si_mntflags, SUM_W);
24543 +       case Opt_nosum:
24544 +               au_opt_clr(sbinfo->si_mntflags, SUM);
24545 +               au_opt_clr(sbinfo->si_mntflags, SUM_W);
24546 +               break;
24547 +
24548 +       case Opt_wbr_create:
24549 +               err = au_opt_wbr_create(sb, &opt->wbr_create);
24550 +               break;
24551 +       case Opt_wbr_copyup:
24552 +               sbinfo->si_wbr_copyup = opt->wbr_copyup;
24553 +               sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + opt->wbr_copyup;
24554 +               break;
24555 +
24556 +       case Opt_dirwh:
24557 +               sbinfo->si_dirwh = opt->dirwh;
24558 +               break;
24559 +
24560 +       case Opt_rdcache:
24561 +               sbinfo->si_rdcache
24562 +                       = msecs_to_jiffies(opt->rdcache * MSEC_PER_SEC);
24563 +               break;
24564 +       case Opt_rdblk:
24565 +               sbinfo->si_rdblk = opt->rdblk;
24566 +               break;
24567 +       case Opt_rdblk_def:
24568 +               sbinfo->si_rdblk = AUFS_RDBLK_DEF;
24569 +               break;
24570 +       case Opt_rdhash:
24571 +               sbinfo->si_rdhash = opt->rdhash;
24572 +               break;
24573 +       case Opt_rdhash_def:
24574 +               sbinfo->si_rdhash = AUFS_RDHASH_DEF;
24575 +               break;
24576 +
24577 +       case Opt_shwh:
24578 +               au_opt_set(sbinfo->si_mntflags, SHWH);
24579 +               break;
24580 +       case Opt_noshwh:
24581 +               au_opt_clr(sbinfo->si_mntflags, SHWH);
24582 +               break;
24583 +
24584 +       case Opt_dirperm1:
24585 +               au_opt_set(sbinfo->si_mntflags, DIRPERM1);
24586 +               break;
24587 +       case Opt_nodirperm1:
24588 +               au_opt_clr(sbinfo->si_mntflags, DIRPERM1);
24589 +               break;
24590 +
24591 +       case Opt_trunc_xino:
24592 +               au_opt_set(sbinfo->si_mntflags, TRUNC_XINO);
24593 +               break;
24594 +       case Opt_notrunc_xino:
24595 +               au_opt_clr(sbinfo->si_mntflags, TRUNC_XINO);
24596 +               break;
24597 +
24598 +       case Opt_trunc_xino_path:
24599 +       case Opt_itrunc_xino:
24600 +               err = au_xino_trunc(sb, opt->xino_itrunc.bindex);
24601 +               if (!err)
24602 +                       err = 1;
24603 +               break;
24604 +
24605 +       case Opt_trunc_xib:
24606 +               au_fset_opts(opts->flags, TRUNC_XIB);
24607 +               break;
24608 +       case Opt_notrunc_xib:
24609 +               au_fclr_opts(opts->flags, TRUNC_XIB);
24610 +               break;
24611 +
24612 +       case Opt_acl:
24613 +               sb->s_flags |= MS_POSIXACL;
24614 +               break;
24615 +       case Opt_noacl:
24616 +               sb->s_flags &= ~MS_POSIXACL;
24617 +               break;
24618 +
24619 +       default:
24620 +               err = 0;
24621 +               break;
24622 +       }
24623 +
24624 +       return err;
24625 +}
24626 +
24627 +/*
24628 + * returns tri-state.
24629 + * plus: processed without an error
24630 + * zero: unprocessed
24631 + * minus: error
24632 + */
24633 +static int au_opt_br(struct super_block *sb, struct au_opt *opt,
24634 +                    struct au_opts *opts)
24635 +{
24636 +       int err, do_refresh;
24637 +
24638 +       err = 0;
24639 +       switch (opt->type) {
24640 +       case Opt_append:
24641 +               opt->add.bindex = au_sbbot(sb) + 1;
24642 +               if (opt->add.bindex < 0)
24643 +                       opt->add.bindex = 0;
24644 +               goto add;
24645 +       case Opt_prepend:
24646 +               opt->add.bindex = 0;
24647 +       add: /* indented label */
24648 +       case Opt_add:
24649 +               err = au_br_add(sb, &opt->add,
24650 +                               au_ftest_opts(opts->flags, REMOUNT));
24651 +               if (!err) {
24652 +                       err = 1;
24653 +                       au_fset_opts(opts->flags, REFRESH);
24654 +               }
24655 +               break;
24656 +
24657 +       case Opt_del:
24658 +       case Opt_idel:
24659 +               err = au_br_del(sb, &opt->del,
24660 +                               au_ftest_opts(opts->flags, REMOUNT));
24661 +               if (!err) {
24662 +                       err = 1;
24663 +                       au_fset_opts(opts->flags, TRUNC_XIB);
24664 +                       au_fset_opts(opts->flags, REFRESH);
24665 +               }
24666 +               break;
24667 +
24668 +       case Opt_mod:
24669 +       case Opt_imod:
24670 +               err = au_br_mod(sb, &opt->mod,
24671 +                               au_ftest_opts(opts->flags, REMOUNT),
24672 +                               &do_refresh);
24673 +               if (!err) {
24674 +                       err = 1;
24675 +                       if (do_refresh)
24676 +                               au_fset_opts(opts->flags, REFRESH);
24677 +               }
24678 +               break;
24679 +       }
24680 +
24681 +       return err;
24682 +}
24683 +
24684 +static int au_opt_xino(struct super_block *sb, struct au_opt *opt,
24685 +                      struct au_opt_xino **opt_xino,
24686 +                      struct au_opts *opts)
24687 +{
24688 +       int err;
24689 +       aufs_bindex_t bbot, bindex;
24690 +       struct dentry *root, *parent, *h_root;
24691 +
24692 +       err = 0;
24693 +       switch (opt->type) {
24694 +       case Opt_xino:
24695 +               err = au_xino_set(sb, &opt->xino,
24696 +                                 !!au_ftest_opts(opts->flags, REMOUNT));
24697 +               if (unlikely(err))
24698 +                       break;
24699 +
24700 +               *opt_xino = &opt->xino;
24701 +               au_xino_brid_set(sb, -1);
24702 +
24703 +               /* safe d_parent access */
24704 +               parent = opt->xino.file->f_path.dentry->d_parent;
24705 +               root = sb->s_root;
24706 +               bbot = au_sbbot(sb);
24707 +               for (bindex = 0; bindex <= bbot; bindex++) {
24708 +                       h_root = au_h_dptr(root, bindex);
24709 +                       if (h_root == parent) {
24710 +                               au_xino_brid_set(sb, au_sbr_id(sb, bindex));
24711 +                               break;
24712 +                       }
24713 +               }
24714 +               break;
24715 +
24716 +       case Opt_noxino:
24717 +               au_xino_clr(sb);
24718 +               au_xino_brid_set(sb, -1);
24719 +               *opt_xino = (void *)-1;
24720 +               break;
24721 +       }
24722 +
24723 +       return err;
24724 +}
24725 +
24726 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
24727 +                  unsigned int pending)
24728 +{
24729 +       int err, fhsm;
24730 +       aufs_bindex_t bindex, bbot;
24731 +       unsigned char do_plink, skip, do_free, can_no_dreval;
24732 +       struct au_branch *br;
24733 +       struct au_wbr *wbr;
24734 +       struct dentry *root, *dentry;
24735 +       struct inode *dir, *h_dir;
24736 +       struct au_sbinfo *sbinfo;
24737 +       struct au_hinode *hdir;
24738 +
24739 +       SiMustAnyLock(sb);
24740 +
24741 +       sbinfo = au_sbi(sb);
24742 +       AuDebugOn(!(sbinfo->si_mntflags & AuOptMask_UDBA));
24743 +
24744 +       if (!(sb_flags & MS_RDONLY)) {
24745 +               if (unlikely(!au_br_writable(au_sbr_perm(sb, 0))))
24746 +                       pr_warn("first branch should be rw\n");
24747 +               if (unlikely(au_opt_test(sbinfo->si_mntflags, SHWH)))
24748 +                       pr_warn_once("shwh should be used with ro\n");
24749 +       }
24750 +
24751 +       if (au_opt_test((sbinfo->si_mntflags | pending), UDBA_HNOTIFY)
24752 +           && !au_opt_test(sbinfo->si_mntflags, XINO))
24753 +               pr_warn_once("udba=*notify requires xino\n");
24754 +
24755 +       if (au_opt_test(sbinfo->si_mntflags, DIRPERM1))
24756 +               pr_warn_once("dirperm1 breaks the protection"
24757 +                            " by the permission bits on the lower branch\n");
24758 +
24759 +       err = 0;
24760 +       fhsm = 0;
24761 +       root = sb->s_root;
24762 +       dir = d_inode(root);
24763 +       do_plink = !!au_opt_test(sbinfo->si_mntflags, PLINK);
24764 +       can_no_dreval = !!au_opt_test((sbinfo->si_mntflags | pending),
24765 +                                     UDBA_NONE);
24766 +       bbot = au_sbbot(sb);
24767 +       for (bindex = 0; !err && bindex <= bbot; bindex++) {
24768 +               skip = 0;
24769 +               h_dir = au_h_iptr(dir, bindex);
24770 +               br = au_sbr(sb, bindex);
24771 +
24772 +               if ((br->br_perm & AuBrAttr_ICEX)
24773 +                   && !h_dir->i_op->listxattr)
24774 +                       br->br_perm &= ~AuBrAttr_ICEX;
24775 +#if 0
24776 +               if ((br->br_perm & AuBrAttr_ICEX_SEC)
24777 +                   && (au_br_sb(br)->s_flags & MS_NOSEC))
24778 +                       br->br_perm &= ~AuBrAttr_ICEX_SEC;
24779 +#endif
24780 +
24781 +               do_free = 0;
24782 +               wbr = br->br_wbr;
24783 +               if (wbr)
24784 +                       wbr_wh_read_lock(wbr);
24785 +
24786 +               if (!au_br_writable(br->br_perm)) {
24787 +                       do_free = !!wbr;
24788 +                       skip = (!wbr
24789 +                               || (!wbr->wbr_whbase
24790 +                                   && !wbr->wbr_plink
24791 +                                   && !wbr->wbr_orph));
24792 +               } else if (!au_br_wh_linkable(br->br_perm)) {
24793 +                       /* skip = (!br->br_whbase && !br->br_orph); */
24794 +                       skip = (!wbr || !wbr->wbr_whbase);
24795 +                       if (skip && wbr) {
24796 +                               if (do_plink)
24797 +                                       skip = !!wbr->wbr_plink;
24798 +                               else
24799 +                                       skip = !wbr->wbr_plink;
24800 +                       }
24801 +               } else {
24802 +                       /* skip = (br->br_whbase && br->br_ohph); */
24803 +                       skip = (wbr && wbr->wbr_whbase);
24804 +                       if (skip) {
24805 +                               if (do_plink)
24806 +                                       skip = !!wbr->wbr_plink;
24807 +                               else
24808 +                                       skip = !wbr->wbr_plink;
24809 +                       }
24810 +               }
24811 +               if (wbr)
24812 +                       wbr_wh_read_unlock(wbr);
24813 +
24814 +               if (can_no_dreval) {
24815 +                       dentry = br->br_path.dentry;
24816 +                       spin_lock(&dentry->d_lock);
24817 +                       if (dentry->d_flags &
24818 +                           (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE))
24819 +                               can_no_dreval = 0;
24820 +                       spin_unlock(&dentry->d_lock);
24821 +               }
24822 +
24823 +               if (au_br_fhsm(br->br_perm)) {
24824 +                       fhsm++;
24825 +                       AuDebugOn(!br->br_fhsm);
24826 +               }
24827 +
24828 +               if (skip)
24829 +                       continue;
24830 +
24831 +               hdir = au_hi(dir, bindex);
24832 +               au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
24833 +               if (wbr)
24834 +                       wbr_wh_write_lock(wbr);
24835 +               err = au_wh_init(br, sb);
24836 +               if (wbr)
24837 +                       wbr_wh_write_unlock(wbr);
24838 +               au_hn_inode_unlock(hdir);
24839 +
24840 +               if (!err && do_free) {
24841 +                       if (wbr)
24842 +                               au_delayed_kfree(wbr);
24843 +                       br->br_wbr = NULL;
24844 +               }
24845 +       }
24846 +
24847 +       if (can_no_dreval)
24848 +               au_fset_si(sbinfo, NO_DREVAL);
24849 +       else
24850 +               au_fclr_si(sbinfo, NO_DREVAL);
24851 +
24852 +       if (fhsm >= 2) {
24853 +               au_fset_si(sbinfo, FHSM);
24854 +               for (bindex = bbot; bindex >= 0; bindex--) {
24855 +                       br = au_sbr(sb, bindex);
24856 +                       if (au_br_fhsm(br->br_perm)) {
24857 +                               au_fhsm_set_bottom(sb, bindex);
24858 +                               break;
24859 +                       }
24860 +               }
24861 +       } else {
24862 +               au_fclr_si(sbinfo, FHSM);
24863 +               au_fhsm_set_bottom(sb, -1);
24864 +       }
24865 +
24866 +       return err;
24867 +}
24868 +
24869 +int au_opts_mount(struct super_block *sb, struct au_opts *opts)
24870 +{
24871 +       int err;
24872 +       unsigned int tmp;
24873 +       aufs_bindex_t bindex, bbot;
24874 +       struct au_opt *opt;
24875 +       struct au_opt_xino *opt_xino, xino;
24876 +       struct au_sbinfo *sbinfo;
24877 +       struct au_branch *br;
24878 +       struct inode *dir;
24879 +
24880 +       SiMustWriteLock(sb);
24881 +
24882 +       err = 0;
24883 +       opt_xino = NULL;
24884 +       opt = opts->opt;
24885 +       while (err >= 0 && opt->type != Opt_tail)
24886 +               err = au_opt_simple(sb, opt++, opts);
24887 +       if (err > 0)
24888 +               err = 0;
24889 +       else if (unlikely(err < 0))
24890 +               goto out;
24891 +
24892 +       /* disable xino and udba temporary */
24893 +       sbinfo = au_sbi(sb);
24894 +       tmp = sbinfo->si_mntflags;
24895 +       au_opt_clr(sbinfo->si_mntflags, XINO);
24896 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_REVAL);
24897 +
24898 +       opt = opts->opt;
24899 +       while (err >= 0 && opt->type != Opt_tail)
24900 +               err = au_opt_br(sb, opt++, opts);
24901 +       if (err > 0)
24902 +               err = 0;
24903 +       else if (unlikely(err < 0))
24904 +               goto out;
24905 +
24906 +       bbot = au_sbbot(sb);
24907 +       if (unlikely(bbot < 0)) {
24908 +               err = -EINVAL;
24909 +               pr_err("no branches\n");
24910 +               goto out;
24911 +       }
24912 +
24913 +       if (au_opt_test(tmp, XINO))
24914 +               au_opt_set(sbinfo->si_mntflags, XINO);
24915 +       opt = opts->opt;
24916 +       while (!err && opt->type != Opt_tail)
24917 +               err = au_opt_xino(sb, opt++, &opt_xino, opts);
24918 +       if (unlikely(err))
24919 +               goto out;
24920 +
24921 +       err = au_opts_verify(sb, sb->s_flags, tmp);
24922 +       if (unlikely(err))
24923 +               goto out;
24924 +
24925 +       /* restore xino */
24926 +       if (au_opt_test(tmp, XINO) && !opt_xino) {
24927 +               xino.file = au_xino_def(sb);
24928 +               err = PTR_ERR(xino.file);
24929 +               if (IS_ERR(xino.file))
24930 +                       goto out;
24931 +
24932 +               err = au_xino_set(sb, &xino, /*remount*/0);
24933 +               fput(xino.file);
24934 +               if (unlikely(err))
24935 +                       goto out;
24936 +       }
24937 +
24938 +       /* restore udba */
24939 +       tmp &= AuOptMask_UDBA;
24940 +       sbinfo->si_mntflags &= ~AuOptMask_UDBA;
24941 +       sbinfo->si_mntflags |= tmp;
24942 +       bbot = au_sbbot(sb);
24943 +       for (bindex = 0; bindex <= bbot; bindex++) {
24944 +               br = au_sbr(sb, bindex);
24945 +               err = au_hnotify_reset_br(tmp, br, br->br_perm);
24946 +               if (unlikely(err))
24947 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
24948 +                               bindex, err);
24949 +               /* go on even if err */
24950 +       }
24951 +       if (au_opt_test(tmp, UDBA_HNOTIFY)) {
24952 +               dir = d_inode(sb->s_root);
24953 +               au_hn_reset(dir, au_hi_flags(dir, /*isdir*/1) & ~AuHi_XINO);
24954 +       }
24955 +
24956 +out:
24957 +       return err;
24958 +}
24959 +
24960 +int au_opts_remount(struct super_block *sb, struct au_opts *opts)
24961 +{
24962 +       int err, rerr;
24963 +       unsigned char no_dreval;
24964 +       struct inode *dir;
24965 +       struct au_opt_xino *opt_xino;
24966 +       struct au_opt *opt;
24967 +       struct au_sbinfo *sbinfo;
24968 +
24969 +       SiMustWriteLock(sb);
24970 +
24971 +       err = 0;
24972 +       dir = d_inode(sb->s_root);
24973 +       sbinfo = au_sbi(sb);
24974 +       opt_xino = NULL;
24975 +       opt = opts->opt;
24976 +       while (err >= 0 && opt->type != Opt_tail) {
24977 +               err = au_opt_simple(sb, opt, opts);
24978 +               if (!err)
24979 +                       err = au_opt_br(sb, opt, opts);
24980 +               if (!err)
24981 +                       err = au_opt_xino(sb, opt, &opt_xino, opts);
24982 +               opt++;
24983 +       }
24984 +       if (err > 0)
24985 +               err = 0;
24986 +       AuTraceErr(err);
24987 +       /* go on even err */
24988 +
24989 +       no_dreval = !!au_ftest_si(sbinfo, NO_DREVAL);
24990 +       rerr = au_opts_verify(sb, opts->sb_flags, /*pending*/0);
24991 +       if (unlikely(rerr && !err))
24992 +               err = rerr;
24993 +
24994 +       if (no_dreval != !!au_ftest_si(sbinfo, NO_DREVAL))
24995 +               au_fset_opts(opts->flags, REFRESH_IDOP);
24996 +
24997 +       if (au_ftest_opts(opts->flags, TRUNC_XIB)) {
24998 +               rerr = au_xib_trunc(sb);
24999 +               if (unlikely(rerr && !err))
25000 +                       err = rerr;
25001 +       }
25002 +
25003 +       /* will be handled by the caller */
25004 +       if (!au_ftest_opts(opts->flags, REFRESH)
25005 +           && (opts->given_udba
25006 +               || au_opt_test(sbinfo->si_mntflags, XINO)
25007 +               || au_ftest_opts(opts->flags, REFRESH_IDOP)
25008 +                   ))
25009 +               au_fset_opts(opts->flags, REFRESH);
25010 +
25011 +       AuDbg("status 0x%x\n", opts->flags);
25012 +       return err;
25013 +}
25014 +
25015 +/* ---------------------------------------------------------------------- */
25016 +
25017 +unsigned int au_opt_udba(struct super_block *sb)
25018 +{
25019 +       return au_mntflags(sb) & AuOptMask_UDBA;
25020 +}
25021 diff -urN /usr/share/empty/fs/aufs/opts.h linux/fs/aufs/opts.h
25022 --- /usr/share/empty/fs/aufs/opts.h     1970-01-01 01:00:00.000000000 +0100
25023 +++ linux/fs/aufs/opts.h        2016-10-09 16:55:36.496035060 +0200
25024 @@ -0,0 +1,211 @@
25025 +/*
25026 + * Copyright (C) 2005-2016 Junjiro R. Okajima
25027 + *
25028 + * This program, aufs is free software; you can redistribute it and/or modify
25029 + * it under the terms of the GNU General Public License as published by
25030 + * the Free Software Foundation; either version 2 of the License, or
25031 + * (at your option) any later version.
25032 + *
25033 + * This program is distributed in the hope that it will be useful,
25034 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25035 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25036 + * GNU General Public License for more details.
25037 + *
25038 + * You should have received a copy of the GNU General Public License
25039 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25040 + */
25041 +
25042 +/*
25043 + * mount options/flags
25044 + */
25045 +
25046 +#ifndef __AUFS_OPTS_H__
25047 +#define __AUFS_OPTS_H__
25048 +
25049 +#ifdef __KERNEL__
25050 +
25051 +#include <linux/path.h>
25052 +
25053 +struct file;
25054 +struct super_block;
25055 +
25056 +/* ---------------------------------------------------------------------- */
25057 +
25058 +/* mount flags */
25059 +#define AuOpt_XINO             1               /* external inode number bitmap
25060 +                                                  and translation table */
25061 +#define AuOpt_TRUNC_XINO       (1 << 1)        /* truncate xino files */
25062 +#define AuOpt_UDBA_NONE                (1 << 2)        /* users direct branch access */
25063 +#define AuOpt_UDBA_REVAL       (1 << 3)
25064 +#define AuOpt_UDBA_HNOTIFY     (1 << 4)
25065 +#define AuOpt_SHWH             (1 << 5)        /* show whiteout */
25066 +#define AuOpt_PLINK            (1 << 6)        /* pseudo-link */
25067 +#define AuOpt_DIRPERM1         (1 << 7)        /* ignore the lower dir's perm
25068 +                                                  bits */
25069 +#define AuOpt_ALWAYS_DIROPQ    (1 << 9)        /* policy to creating diropq */
25070 +#define AuOpt_SUM              (1 << 10)       /* summation for statfs(2) */
25071 +#define AuOpt_SUM_W            (1 << 11)       /* unimplemented */
25072 +#define AuOpt_WARN_PERM                (1 << 12)       /* warn when add-branch */
25073 +#define AuOpt_VERBOSE          (1 << 13)       /* busy inode when del-branch */
25074 +#define AuOpt_DIO              (1 << 14)       /* direct io */
25075 +
25076 +#ifndef CONFIG_AUFS_HNOTIFY
25077 +#undef AuOpt_UDBA_HNOTIFY
25078 +#define AuOpt_UDBA_HNOTIFY     0
25079 +#endif
25080 +#ifndef CONFIG_AUFS_SHWH
25081 +#undef AuOpt_SHWH
25082 +#define AuOpt_SHWH             0
25083 +#endif
25084 +
25085 +#define AuOpt_Def      (AuOpt_XINO \
25086 +                        | AuOpt_UDBA_REVAL \
25087 +                        | AuOpt_PLINK \
25088 +                        /* | AuOpt_DIRPERM1 */ \
25089 +                        | AuOpt_WARN_PERM)
25090 +#define AuOptMask_UDBA (AuOpt_UDBA_NONE \
25091 +                        | AuOpt_UDBA_REVAL \
25092 +                        | AuOpt_UDBA_HNOTIFY)
25093 +
25094 +#define au_opt_test(flags, name)       (flags & AuOpt_##name)
25095 +#define au_opt_set(flags, name) do { \
25096 +       BUILD_BUG_ON(AuOpt_##name & AuOptMask_UDBA); \
25097 +       ((flags) |= AuOpt_##name); \
25098 +} while (0)
25099 +#define au_opt_set_udba(flags, name) do { \
25100 +       (flags) &= ~AuOptMask_UDBA; \
25101 +       ((flags) |= AuOpt_##name); \
25102 +} while (0)
25103 +#define au_opt_clr(flags, name) do { \
25104 +       ((flags) &= ~AuOpt_##name); \
25105 +} while (0)
25106 +
25107 +static inline unsigned int au_opts_plink(unsigned int mntflags)
25108 +{
25109 +#ifdef CONFIG_PROC_FS
25110 +       return mntflags;
25111 +#else
25112 +       return mntflags & ~AuOpt_PLINK;
25113 +#endif
25114 +}
25115 +
25116 +/* ---------------------------------------------------------------------- */
25117 +
25118 +/* policies to select one among multiple writable branches */
25119 +enum {
25120 +       AuWbrCreate_TDP,        /* top down parent */
25121 +       AuWbrCreate_RR,         /* round robin */
25122 +       AuWbrCreate_MFS,        /* most free space */
25123 +       AuWbrCreate_MFSV,       /* mfs with seconds */
25124 +       AuWbrCreate_MFSRR,      /* mfs then rr */
25125 +       AuWbrCreate_MFSRRV,     /* mfs then rr with seconds */
25126 +       AuWbrCreate_PMFS,       /* parent and mfs */
25127 +       AuWbrCreate_PMFSV,      /* parent and mfs with seconds */
25128 +       AuWbrCreate_PMFSRR,     /* parent, mfs and round-robin */
25129 +       AuWbrCreate_PMFSRRV,    /* plus seconds */
25130 +
25131 +       AuWbrCreate_Def = AuWbrCreate_TDP
25132 +};
25133 +
25134 +enum {
25135 +       AuWbrCopyup_TDP,        /* top down parent */
25136 +       AuWbrCopyup_BUP,        /* bottom up parent */
25137 +       AuWbrCopyup_BU,         /* bottom up */
25138 +
25139 +       AuWbrCopyup_Def = AuWbrCopyup_TDP
25140 +};
25141 +
25142 +/* ---------------------------------------------------------------------- */
25143 +
25144 +struct au_opt_add {
25145 +       aufs_bindex_t   bindex;
25146 +       char            *pathname;
25147 +       int             perm;
25148 +       struct path     path;
25149 +};
25150 +
25151 +struct au_opt_del {
25152 +       char            *pathname;
25153 +       struct path     h_path;
25154 +};
25155 +
25156 +struct au_opt_mod {
25157 +       char            *path;
25158 +       int             perm;
25159 +       struct dentry   *h_root;
25160 +};
25161 +
25162 +struct au_opt_xino {
25163 +       char            *path;
25164 +       struct file     *file;
25165 +};
25166 +
25167 +struct au_opt_xino_itrunc {
25168 +       aufs_bindex_t   bindex;
25169 +};
25170 +
25171 +struct au_opt_wbr_create {
25172 +       int                     wbr_create;
25173 +       int                     mfs_second;
25174 +       unsigned long long      mfsrr_watermark;
25175 +};
25176 +
25177 +struct au_opt {
25178 +       int type;
25179 +       union {
25180 +               struct au_opt_xino      xino;
25181 +               struct au_opt_xino_itrunc xino_itrunc;
25182 +               struct au_opt_add       add;
25183 +               struct au_opt_del       del;
25184 +               struct au_opt_mod       mod;
25185 +               int                     dirwh;
25186 +               int                     rdcache;
25187 +               unsigned int            rdblk;
25188 +               unsigned int            rdhash;
25189 +               int                     udba;
25190 +               struct au_opt_wbr_create wbr_create;
25191 +               int                     wbr_copyup;
25192 +               unsigned int            fhsm_second;
25193 +       };
25194 +};
25195 +
25196 +/* opts flags */
25197 +#define AuOpts_REMOUNT         1
25198 +#define AuOpts_REFRESH         (1 << 1)
25199 +#define AuOpts_TRUNC_XIB       (1 << 2)
25200 +#define AuOpts_REFRESH_DYAOP   (1 << 3)
25201 +#define AuOpts_REFRESH_IDOP    (1 << 4)
25202 +#define au_ftest_opts(flags, name)     ((flags) & AuOpts_##name)
25203 +#define au_fset_opts(flags, name) \
25204 +       do { (flags) |= AuOpts_##name; } while (0)
25205 +#define au_fclr_opts(flags, name) \
25206 +       do { (flags) &= ~AuOpts_##name; } while (0)
25207 +
25208 +struct au_opts {
25209 +       struct au_opt   *opt;
25210 +       int             max_opt;
25211 +
25212 +       unsigned int    given_udba;
25213 +       unsigned int    flags;
25214 +       unsigned long   sb_flags;
25215 +};
25216 +
25217 +/* ---------------------------------------------------------------------- */
25218 +
25219 +/* opts.c */
25220 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm);
25221 +const char *au_optstr_udba(int udba);
25222 +const char *au_optstr_wbr_copyup(int wbr_copyup);
25223 +const char *au_optstr_wbr_create(int wbr_create);
25224 +
25225 +void au_opts_free(struct au_opts *opts);
25226 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts);
25227 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
25228 +                  unsigned int pending);
25229 +int au_opts_mount(struct super_block *sb, struct au_opts *opts);
25230 +int au_opts_remount(struct super_block *sb, struct au_opts *opts);
25231 +
25232 +unsigned int au_opt_udba(struct super_block *sb);
25233 +
25234 +#endif /* __KERNEL__ */
25235 +#endif /* __AUFS_OPTS_H__ */
25236 diff -urN /usr/share/empty/fs/aufs/plink.c linux/fs/aufs/plink.c
25237 --- /usr/share/empty/fs/aufs/plink.c    1970-01-01 01:00:00.000000000 +0100
25238 +++ linux/fs/aufs/plink.c       2016-10-09 16:55:36.496035060 +0200
25239 @@ -0,0 +1,514 @@
25240 +/*
25241 + * Copyright (C) 2005-2016 Junjiro R. Okajima
25242 + *
25243 + * This program, aufs is free software; you can redistribute it and/or modify
25244 + * it under the terms of the GNU General Public License as published by
25245 + * the Free Software Foundation; either version 2 of the License, or
25246 + * (at your option) any later version.
25247 + *
25248 + * This program is distributed in the hope that it will be useful,
25249 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25250 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25251 + * GNU General Public License for more details.
25252 + *
25253 + * You should have received a copy of the GNU General Public License
25254 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25255 + */
25256 +
25257 +/*
25258 + * pseudo-link
25259 + */
25260 +
25261 +#include "aufs.h"
25262 +
25263 +/*
25264 + * the pseudo-link maintenance mode.
25265 + * during a user process maintains the pseudo-links,
25266 + * prohibit adding a new plink and branch manipulation.
25267 + *
25268 + * Flags
25269 + * NOPLM:
25270 + *     For entry functions which will handle plink, and i_mutex is already held
25271 + *     in VFS.
25272 + *     They cannot wait and should return an error at once.
25273 + *     Callers has to check the error.
25274 + * NOPLMW:
25275 + *     For entry functions which will handle plink, but i_mutex is not held
25276 + *     in VFS.
25277 + *     They can wait the plink maintenance mode to finish.
25278 + *
25279 + * They behave like F_SETLK and F_SETLKW.
25280 + * If the caller never handle plink, then both flags are unnecessary.
25281 + */
25282 +
25283 +int au_plink_maint(struct super_block *sb, int flags)
25284 +{
25285 +       int err;
25286 +       pid_t pid, ppid;
25287 +       struct task_struct *parent, *prev;
25288 +       struct au_sbinfo *sbi;
25289 +
25290 +       SiMustAnyLock(sb);
25291 +
25292 +       err = 0;
25293 +       if (!au_opt_test(au_mntflags(sb), PLINK))
25294 +               goto out;
25295 +
25296 +       sbi = au_sbi(sb);
25297 +       pid = sbi->si_plink_maint_pid;
25298 +       if (!pid || pid == current->pid)
25299 +               goto out;
25300 +
25301 +       /* todo: it highly depends upon /sbin/mount.aufs */
25302 +       prev = NULL;
25303 +       parent = current;
25304 +       ppid = 0;
25305 +       rcu_read_lock();
25306 +       while (1) {
25307 +               parent = rcu_dereference(parent->real_parent);
25308 +               if (parent == prev)
25309 +                       break;
25310 +               ppid = task_pid_vnr(parent);
25311 +               if (pid == ppid) {
25312 +                       rcu_read_unlock();
25313 +                       goto out;
25314 +               }
25315 +               prev = parent;
25316 +       }
25317 +       rcu_read_unlock();
25318 +
25319 +       if (au_ftest_lock(flags, NOPLMW)) {
25320 +               /* if there is no i_mutex lock in VFS, we don't need to wait */
25321 +               /* AuDebugOn(!lockdep_depth(current)); */
25322 +               while (sbi->si_plink_maint_pid) {
25323 +                       si_read_unlock(sb);
25324 +                       /* gave up wake_up_bit() */
25325 +                       wait_event(sbi->si_plink_wq, !sbi->si_plink_maint_pid);
25326 +
25327 +                       if (au_ftest_lock(flags, FLUSH))
25328 +                               au_nwt_flush(&sbi->si_nowait);
25329 +                       si_noflush_read_lock(sb);
25330 +               }
25331 +       } else if (au_ftest_lock(flags, NOPLM)) {
25332 +               AuDbg("ppid %d, pid %d\n", ppid, pid);
25333 +               err = -EAGAIN;
25334 +       }
25335 +
25336 +out:
25337 +       return err;
25338 +}
25339 +
25340 +void au_plink_maint_leave(struct au_sbinfo *sbinfo)
25341 +{
25342 +       spin_lock(&sbinfo->si_plink_maint_lock);
25343 +       sbinfo->si_plink_maint_pid = 0;
25344 +       spin_unlock(&sbinfo->si_plink_maint_lock);
25345 +       wake_up_all(&sbinfo->si_plink_wq);
25346 +}
25347 +
25348 +int au_plink_maint_enter(struct super_block *sb)
25349 +{
25350 +       int err;
25351 +       struct au_sbinfo *sbinfo;
25352 +
25353 +       err = 0;
25354 +       sbinfo = au_sbi(sb);
25355 +       /* make sure i am the only one in this fs */
25356 +       si_write_lock(sb, AuLock_FLUSH);
25357 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
25358 +               spin_lock(&sbinfo->si_plink_maint_lock);
25359 +               if (!sbinfo->si_plink_maint_pid)
25360 +                       sbinfo->si_plink_maint_pid = current->pid;
25361 +               else
25362 +                       err = -EBUSY;
25363 +               spin_unlock(&sbinfo->si_plink_maint_lock);
25364 +       }
25365 +       si_write_unlock(sb);
25366 +
25367 +       return err;
25368 +}
25369 +
25370 +/* ---------------------------------------------------------------------- */
25371 +
25372 +#ifdef CONFIG_AUFS_DEBUG
25373 +void au_plink_list(struct super_block *sb)
25374 +{
25375 +       int i;
25376 +       struct au_sbinfo *sbinfo;
25377 +       struct hlist_head *plink_hlist;
25378 +       struct au_icntnr *icntnr;
25379 +
25380 +       SiMustAnyLock(sb);
25381 +
25382 +       sbinfo = au_sbi(sb);
25383 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25384 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25385 +
25386 +       for (i = 0; i < AuPlink_NHASH; i++) {
25387 +               plink_hlist = &sbinfo->si_plink[i].head;
25388 +               rcu_read_lock();
25389 +               hlist_for_each_entry_rcu(icntnr, plink_hlist, plink)
25390 +                       AuDbg("%lu\n", icntnr->vfs_inode.i_ino);
25391 +               rcu_read_unlock();
25392 +       }
25393 +}
25394 +#endif
25395 +
25396 +/* is the inode pseudo-linked? */
25397 +int au_plink_test(struct inode *inode)
25398 +{
25399 +       int found, i;
25400 +       struct au_sbinfo *sbinfo;
25401 +       struct hlist_head *plink_hlist;
25402 +       struct au_icntnr *icntnr;
25403 +
25404 +       sbinfo = au_sbi(inode->i_sb);
25405 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
25406 +       AuDebugOn(!au_opt_test(au_mntflags(inode->i_sb), PLINK));
25407 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
25408 +
25409 +       found = 0;
25410 +       i = au_plink_hash(inode->i_ino);
25411 +       plink_hlist = &sbinfo->si_plink[i].head;
25412 +       rcu_read_lock();
25413 +       hlist_for_each_entry_rcu(icntnr, plink_hlist, plink)
25414 +               if (&icntnr->vfs_inode == inode) {
25415 +                       found = 1;
25416 +                       break;
25417 +               }
25418 +       rcu_read_unlock();
25419 +       return found;
25420 +}
25421 +
25422 +/* ---------------------------------------------------------------------- */
25423 +
25424 +/*
25425 + * generate a name for plink.
25426 + * the file will be stored under AUFS_WH_PLINKDIR.
25427 + */
25428 +/* 20 is max digits length of ulong 64 */
25429 +#define PLINK_NAME_LEN ((20 + 1) * 2)
25430 +
25431 +static int plink_name(char *name, int len, struct inode *inode,
25432 +                     aufs_bindex_t bindex)
25433 +{
25434 +       int rlen;
25435 +       struct inode *h_inode;
25436 +
25437 +       h_inode = au_h_iptr(inode, bindex);
25438 +       rlen = snprintf(name, len, "%lu.%lu", inode->i_ino, h_inode->i_ino);
25439 +       return rlen;
25440 +}
25441 +
25442 +struct au_do_plink_lkup_args {
25443 +       struct dentry **errp;
25444 +       struct qstr *tgtname;
25445 +       struct dentry *h_parent;
25446 +       struct au_branch *br;
25447 +};
25448 +
25449 +static struct dentry *au_do_plink_lkup(struct qstr *tgtname,
25450 +                                      struct dentry *h_parent,
25451 +                                      struct au_branch *br)
25452 +{
25453 +       struct dentry *h_dentry;
25454 +       struct inode *h_inode;
25455 +
25456 +       h_inode = d_inode(h_parent);
25457 +       inode_lock_nested(h_inode, AuLsc_I_CHILD2);
25458 +       h_dentry = vfsub_lkup_one(tgtname, h_parent);
25459 +       inode_unlock(h_inode);
25460 +       return h_dentry;
25461 +}
25462 +
25463 +static void au_call_do_plink_lkup(void *args)
25464 +{
25465 +       struct au_do_plink_lkup_args *a = args;
25466 +       *a->errp = au_do_plink_lkup(a->tgtname, a->h_parent, a->br);
25467 +}
25468 +
25469 +/* lookup the plink-ed @inode under the branch at @bindex */
25470 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex)
25471 +{
25472 +       struct dentry *h_dentry, *h_parent;
25473 +       struct au_branch *br;
25474 +       int wkq_err;
25475 +       char a[PLINK_NAME_LEN];
25476 +       struct qstr tgtname = QSTR_INIT(a, 0);
25477 +
25478 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
25479 +
25480 +       br = au_sbr(inode->i_sb, bindex);
25481 +       h_parent = br->br_wbr->wbr_plink;
25482 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
25483 +
25484 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
25485 +               struct au_do_plink_lkup_args args = {
25486 +                       .errp           = &h_dentry,
25487 +                       .tgtname        = &tgtname,
25488 +                       .h_parent       = h_parent,
25489 +                       .br             = br
25490 +               };
25491 +
25492 +               wkq_err = au_wkq_wait(au_call_do_plink_lkup, &args);
25493 +               if (unlikely(wkq_err))
25494 +                       h_dentry = ERR_PTR(wkq_err);
25495 +       } else
25496 +               h_dentry = au_do_plink_lkup(&tgtname, h_parent, br);
25497 +
25498 +       return h_dentry;
25499 +}
25500 +
25501 +/* create a pseudo-link */
25502 +static int do_whplink(struct qstr *tgt, struct dentry *h_parent,
25503 +                     struct dentry *h_dentry, struct au_branch *br)
25504 +{
25505 +       int err;
25506 +       struct path h_path = {
25507 +               .mnt = au_br_mnt(br)
25508 +       };
25509 +       struct inode *h_dir, *delegated;
25510 +
25511 +       h_dir = d_inode(h_parent);
25512 +       inode_lock_nested(h_dir, AuLsc_I_CHILD2);
25513 +again:
25514 +       h_path.dentry = vfsub_lkup_one(tgt, h_parent);
25515 +       err = PTR_ERR(h_path.dentry);
25516 +       if (IS_ERR(h_path.dentry))
25517 +               goto out;
25518 +
25519 +       err = 0;
25520 +       /* wh.plink dir is not monitored */
25521 +       /* todo: is it really safe? */
25522 +       if (d_is_positive(h_path.dentry)
25523 +           && d_inode(h_path.dentry) != d_inode(h_dentry)) {
25524 +               delegated = NULL;
25525 +               err = vfsub_unlink(h_dir, &h_path, &delegated, /*force*/0);
25526 +               if (unlikely(err == -EWOULDBLOCK)) {
25527 +                       pr_warn("cannot retry for NFSv4 delegation"
25528 +                               " for an internal unlink\n");
25529 +                       iput(delegated);
25530 +               }
25531 +               dput(h_path.dentry);
25532 +               h_path.dentry = NULL;
25533 +               if (!err)
25534 +                       goto again;
25535 +       }
25536 +       if (!err && d_is_negative(h_path.dentry)) {
25537 +               delegated = NULL;
25538 +               err = vfsub_link(h_dentry, h_dir, &h_path, &delegated);
25539 +               if (unlikely(err == -EWOULDBLOCK)) {
25540 +                       pr_warn("cannot retry for NFSv4 delegation"
25541 +                               " for an internal link\n");
25542 +                       iput(delegated);
25543 +               }
25544 +       }
25545 +       dput(h_path.dentry);
25546 +
25547 +out:
25548 +       inode_unlock(h_dir);
25549 +       return err;
25550 +}
25551 +
25552 +struct do_whplink_args {
25553 +       int *errp;
25554 +       struct qstr *tgt;
25555 +       struct dentry *h_parent;
25556 +       struct dentry *h_dentry;
25557 +       struct au_branch *br;
25558 +};
25559 +
25560 +static void call_do_whplink(void *args)
25561 +{
25562 +       struct do_whplink_args *a = args;
25563 +       *a->errp = do_whplink(a->tgt, a->h_parent, a->h_dentry, a->br);
25564 +}
25565 +
25566 +static int whplink(struct dentry *h_dentry, struct inode *inode,
25567 +                  aufs_bindex_t bindex, struct au_branch *br)
25568 +{
25569 +       int err, wkq_err;
25570 +       struct au_wbr *wbr;
25571 +       struct dentry *h_parent;
25572 +       char a[PLINK_NAME_LEN];
25573 +       struct qstr tgtname = QSTR_INIT(a, 0);
25574 +
25575 +       wbr = au_sbr(inode->i_sb, bindex)->br_wbr;
25576 +       h_parent = wbr->wbr_plink;
25577 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
25578 +
25579 +       /* always superio. */
25580 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
25581 +               struct do_whplink_args args = {
25582 +                       .errp           = &err,
25583 +                       .tgt            = &tgtname,
25584 +                       .h_parent       = h_parent,
25585 +                       .h_dentry       = h_dentry,
25586 +                       .br             = br
25587 +               };
25588 +               wkq_err = au_wkq_wait(call_do_whplink, &args);
25589 +               if (unlikely(wkq_err))
25590 +                       err = wkq_err;
25591 +       } else
25592 +               err = do_whplink(&tgtname, h_parent, h_dentry, br);
25593 +
25594 +       return err;
25595 +}
25596 +
25597 +/*
25598 + * create a new pseudo-link for @h_dentry on @bindex.
25599 + * the linked inode is held in aufs @inode.
25600 + */
25601 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
25602 +                    struct dentry *h_dentry)
25603 +{
25604 +       struct super_block *sb;
25605 +       struct au_sbinfo *sbinfo;
25606 +       struct hlist_head *plink_hlist;
25607 +       struct au_icntnr *icntnr;
25608 +       struct au_sphlhead *sphl;
25609 +       int found, err, cnt, i;
25610 +
25611 +       sb = inode->i_sb;
25612 +       sbinfo = au_sbi(sb);
25613 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25614 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25615 +
25616 +       found = au_plink_test(inode);
25617 +       if (found)
25618 +               return;
25619 +
25620 +       i = au_plink_hash(inode->i_ino);
25621 +       sphl = sbinfo->si_plink + i;
25622 +       plink_hlist = &sphl->head;
25623 +       au_igrab(inode);
25624 +
25625 +       spin_lock(&sphl->spin);
25626 +       hlist_for_each_entry(icntnr, plink_hlist, plink) {
25627 +               if (&icntnr->vfs_inode == inode) {
25628 +                       found = 1;
25629 +                       break;
25630 +               }
25631 +       }
25632 +       if (!found) {
25633 +               icntnr = container_of(inode, struct au_icntnr, vfs_inode);
25634 +               hlist_add_head_rcu(&icntnr->plink, plink_hlist);
25635 +       }
25636 +       spin_unlock(&sphl->spin);
25637 +       if (!found) {
25638 +               cnt = au_sphl_count(sphl);
25639 +#define msg "unexpectedly unblanced or too many pseudo-links"
25640 +               if (cnt > AUFS_PLINK_WARN)
25641 +                       AuWarn1(msg ", %d\n", cnt);
25642 +#undef msg
25643 +               err = whplink(h_dentry, inode, bindex, au_sbr(sb, bindex));
25644 +               if (unlikely(err)) {
25645 +                       pr_warn("err %d, damaged pseudo link.\n", err);
25646 +                       au_sphl_del_rcu(&icntnr->plink, sphl);
25647 +                       iput(&icntnr->vfs_inode);
25648 +               }
25649 +       } else
25650 +               iput(&icntnr->vfs_inode);
25651 +}
25652 +
25653 +/* free all plinks */
25654 +void au_plink_put(struct super_block *sb, int verbose)
25655 +{
25656 +       int i, warned;
25657 +       struct au_sbinfo *sbinfo;
25658 +       struct hlist_head *plink_hlist;
25659 +       struct hlist_node *tmp;
25660 +       struct au_icntnr *icntnr;
25661 +
25662 +       SiMustWriteLock(sb);
25663 +
25664 +       sbinfo = au_sbi(sb);
25665 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25666 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25667 +
25668 +       /* no spin_lock since sbinfo is write-locked */
25669 +       warned = 0;
25670 +       for (i = 0; i < AuPlink_NHASH; i++) {
25671 +               plink_hlist = &sbinfo->si_plink[i].head;
25672 +               if (!warned && verbose && !hlist_empty(plink_hlist)) {
25673 +                       pr_warn("pseudo-link is not flushed");
25674 +                       warned = 1;
25675 +               }
25676 +               hlist_for_each_entry_safe(icntnr, tmp, plink_hlist, plink)
25677 +                       iput(&icntnr->vfs_inode);
25678 +               INIT_HLIST_HEAD(plink_hlist);
25679 +       }
25680 +}
25681 +
25682 +void au_plink_clean(struct super_block *sb, int verbose)
25683 +{
25684 +       struct dentry *root;
25685 +
25686 +       root = sb->s_root;
25687 +       aufs_write_lock(root);
25688 +       if (au_opt_test(au_mntflags(sb), PLINK))
25689 +               au_plink_put(sb, verbose);
25690 +       aufs_write_unlock(root);
25691 +}
25692 +
25693 +static int au_plink_do_half_refresh(struct inode *inode, aufs_bindex_t br_id)
25694 +{
25695 +       int do_put;
25696 +       aufs_bindex_t btop, bbot, bindex;
25697 +
25698 +       do_put = 0;
25699 +       btop = au_ibtop(inode);
25700 +       bbot = au_ibbot(inode);
25701 +       if (btop >= 0) {
25702 +               for (bindex = btop; bindex <= bbot; bindex++) {
25703 +                       if (!au_h_iptr(inode, bindex)
25704 +                           || au_ii_br_id(inode, bindex) != br_id)
25705 +                               continue;
25706 +                       au_set_h_iptr(inode, bindex, NULL, 0);
25707 +                       do_put = 1;
25708 +                       break;
25709 +               }
25710 +               if (do_put)
25711 +                       for (bindex = btop; bindex <= bbot; bindex++)
25712 +                               if (au_h_iptr(inode, bindex)) {
25713 +                                       do_put = 0;
25714 +                                       break;
25715 +                               }
25716 +       } else
25717 +               do_put = 1;
25718 +
25719 +       return do_put;
25720 +}
25721 +
25722 +/* free the plinks on a branch specified by @br_id */
25723 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id)
25724 +{
25725 +       struct au_sbinfo *sbinfo;
25726 +       struct hlist_head *plink_hlist;
25727 +       struct hlist_node *tmp;
25728 +       struct au_icntnr *icntnr;
25729 +       struct inode *inode;
25730 +       int i, do_put;
25731 +
25732 +       SiMustWriteLock(sb);
25733 +
25734 +       sbinfo = au_sbi(sb);
25735 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25736 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25737 +
25738 +       /* no spin_lock since sbinfo is write-locked */
25739 +       for (i = 0; i < AuPlink_NHASH; i++) {
25740 +               plink_hlist = &sbinfo->si_plink[i].head;
25741 +               hlist_for_each_entry_safe(icntnr, tmp, plink_hlist, plink) {
25742 +                       inode = au_igrab(&icntnr->vfs_inode);
25743 +                       ii_write_lock_child(inode);
25744 +                       do_put = au_plink_do_half_refresh(inode, br_id);
25745 +                       if (do_put) {
25746 +                               hlist_del(&icntnr->plink);
25747 +                               iput(inode);
25748 +                       }
25749 +                       ii_write_unlock(inode);
25750 +                       iput(inode);
25751 +               }
25752 +       }
25753 +}
25754 diff -urN /usr/share/empty/fs/aufs/poll.c linux/fs/aufs/poll.c
25755 --- /usr/share/empty/fs/aufs/poll.c     1970-01-01 01:00:00.000000000 +0100
25756 +++ linux/fs/aufs/poll.c        2016-10-09 16:55:36.496035060 +0200
25757 @@ -0,0 +1,52 @@
25758 +/*
25759 + * Copyright (C) 2005-2016 Junjiro R. Okajima
25760 + *
25761 + * This program, aufs is free software; you can redistribute it and/or modify
25762 + * it under the terms of the GNU General Public License as published by
25763 + * the Free Software Foundation; either version 2 of the License, or
25764 + * (at your option) any later version.
25765 + *
25766 + * This program is distributed in the hope that it will be useful,
25767 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25768 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25769 + * GNU General Public License for more details.
25770 + *
25771 + * You should have received a copy of the GNU General Public License
25772 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25773 + */
25774 +
25775 +/*
25776 + * poll operation
25777 + * There is only one filesystem which implements ->poll operation, currently.
25778 + */
25779 +
25780 +#include "aufs.h"
25781 +
25782 +unsigned int aufs_poll(struct file *file, poll_table *wait)
25783 +{
25784 +       unsigned int mask;
25785 +       int err;
25786 +       struct file *h_file;
25787 +       struct super_block *sb;
25788 +
25789 +       /* We should pretend an error happened. */
25790 +       mask = POLLERR /* | POLLIN | POLLOUT */;
25791 +       sb = file->f_path.dentry->d_sb;
25792 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
25793 +
25794 +       h_file = au_read_pre(file, /*keep_fi*/0);
25795 +       err = PTR_ERR(h_file);
25796 +       if (IS_ERR(h_file))
25797 +               goto out;
25798 +
25799 +       /* it is not an error if h_file has no operation */
25800 +       mask = DEFAULT_POLLMASK;
25801 +       if (h_file->f_op->poll)
25802 +               mask = h_file->f_op->poll(h_file, wait);
25803 +       fput(h_file); /* instead of au_read_post() */
25804 +
25805 +out:
25806 +       si_read_unlock(sb);
25807 +       AuTraceErr((int)mask);
25808 +       return mask;
25809 +}
25810 diff -urN /usr/share/empty/fs/aufs/posix_acl.c linux/fs/aufs/posix_acl.c
25811 --- /usr/share/empty/fs/aufs/posix_acl.c        1970-01-01 01:00:00.000000000 +0100
25812 +++ linux/fs/aufs/posix_acl.c   2016-10-09 16:55:36.496035060 +0200
25813 @@ -0,0 +1,98 @@
25814 +/*
25815 + * Copyright (C) 2014-2016 Junjiro R. Okajima
25816 + *
25817 + * This program, aufs is free software; you can redistribute it and/or modify
25818 + * it under the terms of the GNU General Public License as published by
25819 + * the Free Software Foundation; either version 2 of the License, or
25820 + * (at your option) any later version.
25821 + *
25822 + * This program is distributed in the hope that it will be useful,
25823 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25824 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25825 + * GNU General Public License for more details.
25826 + *
25827 + * You should have received a copy of the GNU General Public License
25828 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25829 + */
25830 +
25831 +/*
25832 + * posix acl operations
25833 + */
25834 +
25835 +#include <linux/fs.h>
25836 +#include "aufs.h"
25837 +
25838 +struct posix_acl *aufs_get_acl(struct inode *inode, int type)
25839 +{
25840 +       struct posix_acl *acl;
25841 +       int err;
25842 +       aufs_bindex_t bindex;
25843 +       struct inode *h_inode;
25844 +       struct super_block *sb;
25845 +
25846 +       acl = NULL;
25847 +       sb = inode->i_sb;
25848 +       si_read_lock(sb, AuLock_FLUSH);
25849 +       ii_read_lock_child(inode);
25850 +       if (!(sb->s_flags & MS_POSIXACL))
25851 +               goto out;
25852 +
25853 +       bindex = au_ibtop(inode);
25854 +       h_inode = au_h_iptr(inode, bindex);
25855 +       if (unlikely(!h_inode
25856 +                    || ((h_inode->i_mode & S_IFMT)
25857 +                        != (inode->i_mode & S_IFMT)))) {
25858 +               err = au_busy_or_stale();
25859 +               acl = ERR_PTR(err);
25860 +               goto out;
25861 +       }
25862 +
25863 +       /* always topmost only */
25864 +       acl = get_acl(h_inode, type);
25865 +
25866 +out:
25867 +       ii_read_unlock(inode);
25868 +       si_read_unlock(sb);
25869 +
25870 +       AuTraceErrPtr(acl);
25871 +       return acl;
25872 +}
25873 +
25874 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
25875 +{
25876 +       int err;
25877 +       ssize_t ssz;
25878 +       struct dentry *dentry;
25879 +       struct au_srxattr arg = {
25880 +               .type = AU_ACL_SET,
25881 +               .u.acl_set = {
25882 +                       .acl    = acl,
25883 +                       .type   = type
25884 +               },
25885 +       };
25886 +
25887 +       IMustLock(inode);
25888 +
25889 +       if (inode->i_ino == AUFS_ROOT_INO)
25890 +               dentry = dget(inode->i_sb->s_root);
25891 +       else {
25892 +               dentry = d_find_alias(inode);
25893 +               if (!dentry)
25894 +                       dentry = d_find_any_alias(inode);
25895 +               if (!dentry) {
25896 +                       pr_warn("cannot handle this inode, "
25897 +                               "please report to aufs-users ML\n");
25898 +                       err = -ENOENT;
25899 +                       goto out;
25900 +               }
25901 +       }
25902 +
25903 +       ssz = au_srxattr(dentry, inode, &arg);
25904 +       dput(dentry);
25905 +       err = ssz;
25906 +       if (ssz >= 0)
25907 +               err = 0;
25908 +
25909 +out:
25910 +       return err;
25911 +}
25912 diff -urN /usr/share/empty/fs/aufs/procfs.c linux/fs/aufs/procfs.c
25913 --- /usr/share/empty/fs/aufs/procfs.c   1970-01-01 01:00:00.000000000 +0100
25914 +++ linux/fs/aufs/procfs.c      2016-10-09 16:55:36.496035060 +0200
25915 @@ -0,0 +1,169 @@
25916 +/*
25917 + * Copyright (C) 2010-2016 Junjiro R. Okajima
25918 + *
25919 + * This program, aufs is free software; you can redistribute it and/or modify
25920 + * it under the terms of the GNU General Public License as published by
25921 + * the Free Software Foundation; either version 2 of the License, or
25922 + * (at your option) any later version.
25923 + *
25924 + * This program is distributed in the hope that it will be useful,
25925 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25926 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25927 + * GNU General Public License for more details.
25928 + *
25929 + * You should have received a copy of the GNU General Public License
25930 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25931 + */
25932 +
25933 +/*
25934 + * procfs interfaces
25935 + */
25936 +
25937 +#include <linux/proc_fs.h>
25938 +#include "aufs.h"
25939 +
25940 +static int au_procfs_plm_release(struct inode *inode, struct file *file)
25941 +{
25942 +       struct au_sbinfo *sbinfo;
25943 +
25944 +       sbinfo = file->private_data;
25945 +       if (sbinfo) {
25946 +               au_plink_maint_leave(sbinfo);
25947 +               kobject_put(&sbinfo->si_kobj);
25948 +       }
25949 +
25950 +       return 0;
25951 +}
25952 +
25953 +static void au_procfs_plm_write_clean(struct file *file)
25954 +{
25955 +       struct au_sbinfo *sbinfo;
25956 +
25957 +       sbinfo = file->private_data;
25958 +       if (sbinfo)
25959 +               au_plink_clean(sbinfo->si_sb, /*verbose*/0);
25960 +}
25961 +
25962 +static int au_procfs_plm_write_si(struct file *file, unsigned long id)
25963 +{
25964 +       int err;
25965 +       struct super_block *sb;
25966 +       struct au_sbinfo *sbinfo;
25967 +
25968 +       err = -EBUSY;
25969 +       if (unlikely(file->private_data))
25970 +               goto out;
25971 +
25972 +       sb = NULL;
25973 +       /* don't use au_sbilist_lock() here */
25974 +       spin_lock(&au_sbilist.spin);
25975 +       hlist_for_each_entry(sbinfo, &au_sbilist.head, si_list)
25976 +               if (id == sysaufs_si_id(sbinfo)) {
25977 +                       kobject_get(&sbinfo->si_kobj);
25978 +                       sb = sbinfo->si_sb;
25979 +                       break;
25980 +               }
25981 +       spin_unlock(&au_sbilist.spin);
25982 +
25983 +       err = -EINVAL;
25984 +       if (unlikely(!sb))
25985 +               goto out;
25986 +
25987 +       err = au_plink_maint_enter(sb);
25988 +       if (!err)
25989 +               /* keep kobject_get() */
25990 +               file->private_data = sbinfo;
25991 +       else
25992 +               kobject_put(&sbinfo->si_kobj);
25993 +out:
25994 +       return err;
25995 +}
25996 +
25997 +/*
25998 + * Accept a valid "si=xxxx" only.
25999 + * Once it is accepted successfully, accept "clean" too.
26000 + */
26001 +static ssize_t au_procfs_plm_write(struct file *file, const char __user *ubuf,
26002 +                                  size_t count, loff_t *ppos)
26003 +{
26004 +       ssize_t err;
26005 +       unsigned long id;
26006 +       /* last newline is allowed */
26007 +       char buf[3 + sizeof(unsigned long) * 2 + 1];
26008 +
26009 +       err = -EACCES;
26010 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
26011 +               goto out;
26012 +
26013 +       err = -EINVAL;
26014 +       if (unlikely(count > sizeof(buf)))
26015 +               goto out;
26016 +
26017 +       err = copy_from_user(buf, ubuf, count);
26018 +       if (unlikely(err)) {
26019 +               err = -EFAULT;
26020 +               goto out;
26021 +       }
26022 +       buf[count] = 0;
26023 +
26024 +       err = -EINVAL;
26025 +       if (!strcmp("clean", buf)) {
26026 +               au_procfs_plm_write_clean(file);
26027 +               goto out_success;
26028 +       } else if (unlikely(strncmp("si=", buf, 3)))
26029 +               goto out;
26030 +
26031 +       err = kstrtoul(buf + 3, 16, &id);
26032 +       if (unlikely(err))
26033 +               goto out;
26034 +
26035 +       err = au_procfs_plm_write_si(file, id);
26036 +       if (unlikely(err))
26037 +               goto out;
26038 +
26039 +out_success:
26040 +       err = count; /* success */
26041 +out:
26042 +       return err;
26043 +}
26044 +
26045 +static const struct file_operations au_procfs_plm_fop = {
26046 +       .write          = au_procfs_plm_write,
26047 +       .release        = au_procfs_plm_release,
26048 +       .owner          = THIS_MODULE
26049 +};
26050 +
26051 +/* ---------------------------------------------------------------------- */
26052 +
26053 +static struct proc_dir_entry *au_procfs_dir;
26054 +
26055 +void au_procfs_fin(void)
26056 +{
26057 +       remove_proc_entry(AUFS_PLINK_MAINT_NAME, au_procfs_dir);
26058 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
26059 +}
26060 +
26061 +int __init au_procfs_init(void)
26062 +{
26063 +       int err;
26064 +       struct proc_dir_entry *entry;
26065 +
26066 +       err = -ENOMEM;
26067 +       au_procfs_dir = proc_mkdir(AUFS_PLINK_MAINT_DIR, NULL);
26068 +       if (unlikely(!au_procfs_dir))
26069 +               goto out;
26070 +
26071 +       entry = proc_create(AUFS_PLINK_MAINT_NAME, S_IFREG | S_IWUSR,
26072 +                           au_procfs_dir, &au_procfs_plm_fop);
26073 +       if (unlikely(!entry))
26074 +               goto out_dir;
26075 +
26076 +       err = 0;
26077 +       goto out; /* success */
26078 +
26079 +
26080 +out_dir:
26081 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
26082 +out:
26083 +       return err;
26084 +}
26085 diff -urN /usr/share/empty/fs/aufs/rdu.c linux/fs/aufs/rdu.c
26086 --- /usr/share/empty/fs/aufs/rdu.c      1970-01-01 01:00:00.000000000 +0100
26087 +++ linux/fs/aufs/rdu.c 2016-10-09 16:55:36.496035060 +0200
26088 @@ -0,0 +1,381 @@
26089 +/*
26090 + * Copyright (C) 2005-2016 Junjiro R. Okajima
26091 + *
26092 + * This program, aufs is free software; you can redistribute it and/or modify
26093 + * it under the terms of the GNU General Public License as published by
26094 + * the Free Software Foundation; either version 2 of the License, or
26095 + * (at your option) any later version.
26096 + *
26097 + * This program is distributed in the hope that it will be useful,
26098 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26099 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26100 + * GNU General Public License for more details.
26101 + *
26102 + * You should have received a copy of the GNU General Public License
26103 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26104 + */
26105 +
26106 +/*
26107 + * readdir in userspace.
26108 + */
26109 +
26110 +#include <linux/compat.h>
26111 +#include <linux/fs_stack.h>
26112 +#include <linux/security.h>
26113 +#include "aufs.h"
26114 +
26115 +/* bits for struct aufs_rdu.flags */
26116 +#define        AuRdu_CALLED    1
26117 +#define        AuRdu_CONT      (1 << 1)
26118 +#define        AuRdu_FULL      (1 << 2)
26119 +#define au_ftest_rdu(flags, name)      ((flags) & AuRdu_##name)
26120 +#define au_fset_rdu(flags, name) \
26121 +       do { (flags) |= AuRdu_##name; } while (0)
26122 +#define au_fclr_rdu(flags, name) \
26123 +       do { (flags) &= ~AuRdu_##name; } while (0)
26124 +
26125 +struct au_rdu_arg {
26126 +       struct dir_context              ctx;
26127 +       struct aufs_rdu                 *rdu;
26128 +       union au_rdu_ent_ul             ent;
26129 +       unsigned long                   end;
26130 +
26131 +       struct super_block              *sb;
26132 +       int                             err;
26133 +};
26134 +
26135 +static int au_rdu_fill(struct dir_context *ctx, const char *name, int nlen,
26136 +                      loff_t offset, u64 h_ino, unsigned int d_type)
26137 +{
26138 +       int err, len;
26139 +       struct au_rdu_arg *arg = container_of(ctx, struct au_rdu_arg, ctx);
26140 +       struct aufs_rdu *rdu = arg->rdu;
26141 +       struct au_rdu_ent ent;
26142 +
26143 +       err = 0;
26144 +       arg->err = 0;
26145 +       au_fset_rdu(rdu->cookie.flags, CALLED);
26146 +       len = au_rdu_len(nlen);
26147 +       if (arg->ent.ul + len  < arg->end) {
26148 +               ent.ino = h_ino;
26149 +               ent.bindex = rdu->cookie.bindex;
26150 +               ent.type = d_type;
26151 +               ent.nlen = nlen;
26152 +               if (unlikely(nlen > AUFS_MAX_NAMELEN))
26153 +                       ent.type = DT_UNKNOWN;
26154 +
26155 +               /* unnecessary to support mmap_sem since this is a dir */
26156 +               err = -EFAULT;
26157 +               if (copy_to_user(arg->ent.e, &ent, sizeof(ent)))
26158 +                       goto out;
26159 +               if (copy_to_user(arg->ent.e->name, name, nlen))
26160 +                       goto out;
26161 +               /* the terminating NULL */
26162 +               if (__put_user(0, arg->ent.e->name + nlen))
26163 +                       goto out;
26164 +               err = 0;
26165 +               /* AuDbg("%p, %.*s\n", arg->ent.p, nlen, name); */
26166 +               arg->ent.ul += len;
26167 +               rdu->rent++;
26168 +       } else {
26169 +               err = -EFAULT;
26170 +               au_fset_rdu(rdu->cookie.flags, FULL);
26171 +               rdu->full = 1;
26172 +               rdu->tail = arg->ent;
26173 +       }
26174 +
26175 +out:
26176 +       /* AuTraceErr(err); */
26177 +       return err;
26178 +}
26179 +
26180 +static int au_rdu_do(struct file *h_file, struct au_rdu_arg *arg)
26181 +{
26182 +       int err;
26183 +       loff_t offset;
26184 +       struct au_rdu_cookie *cookie = &arg->rdu->cookie;
26185 +
26186 +       /* we don't have to care (FMODE_32BITHASH | FMODE_64BITHASH) for ext4 */
26187 +       offset = vfsub_llseek(h_file, cookie->h_pos, SEEK_SET);
26188 +       err = offset;
26189 +       if (unlikely(offset != cookie->h_pos))
26190 +               goto out;
26191 +
26192 +       err = 0;
26193 +       do {
26194 +               arg->err = 0;
26195 +               au_fclr_rdu(cookie->flags, CALLED);
26196 +               /* smp_mb(); */
26197 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
26198 +               if (err >= 0)
26199 +                       err = arg->err;
26200 +       } while (!err
26201 +                && au_ftest_rdu(cookie->flags, CALLED)
26202 +                && !au_ftest_rdu(cookie->flags, FULL));
26203 +       cookie->h_pos = h_file->f_pos;
26204 +
26205 +out:
26206 +       AuTraceErr(err);
26207 +       return err;
26208 +}
26209 +
26210 +static int au_rdu(struct file *file, struct aufs_rdu *rdu)
26211 +{
26212 +       int err;
26213 +       aufs_bindex_t bbot;
26214 +       struct au_rdu_arg arg = {
26215 +               .ctx = {
26216 +                       .actor = au_rdu_fill
26217 +               }
26218 +       };
26219 +       struct dentry *dentry;
26220 +       struct inode *inode;
26221 +       struct file *h_file;
26222 +       struct au_rdu_cookie *cookie = &rdu->cookie;
26223 +
26224 +       err = !access_ok(VERIFY_WRITE, rdu->ent.e, rdu->sz);
26225 +       if (unlikely(err)) {
26226 +               err = -EFAULT;
26227 +               AuTraceErr(err);
26228 +               goto out;
26229 +       }
26230 +       rdu->rent = 0;
26231 +       rdu->tail = rdu->ent;
26232 +       rdu->full = 0;
26233 +       arg.rdu = rdu;
26234 +       arg.ent = rdu->ent;
26235 +       arg.end = arg.ent.ul;
26236 +       arg.end += rdu->sz;
26237 +
26238 +       err = -ENOTDIR;
26239 +       if (unlikely(!file->f_op->iterate && !file->f_op->iterate_shared))
26240 +               goto out;
26241 +
26242 +       err = security_file_permission(file, MAY_READ);
26243 +       AuTraceErr(err);
26244 +       if (unlikely(err))
26245 +               goto out;
26246 +
26247 +       dentry = file->f_path.dentry;
26248 +       inode = d_inode(dentry);
26249 +       inode_lock_shared(inode);
26250 +
26251 +       arg.sb = inode->i_sb;
26252 +       err = si_read_lock(arg.sb, AuLock_FLUSH | AuLock_NOPLM);
26253 +       if (unlikely(err))
26254 +               goto out_mtx;
26255 +       err = au_alive_dir(dentry);
26256 +       if (unlikely(err))
26257 +               goto out_si;
26258 +       /* todo: reval? */
26259 +       fi_read_lock(file);
26260 +
26261 +       err = -EAGAIN;
26262 +       if (unlikely(au_ftest_rdu(cookie->flags, CONT)
26263 +                    && cookie->generation != au_figen(file)))
26264 +               goto out_unlock;
26265 +
26266 +       err = 0;
26267 +       if (!rdu->blk) {
26268 +               rdu->blk = au_sbi(arg.sb)->si_rdblk;
26269 +               if (!rdu->blk)
26270 +                       rdu->blk = au_dir_size(file, /*dentry*/NULL);
26271 +       }
26272 +       bbot = au_fbtop(file);
26273 +       if (cookie->bindex < bbot)
26274 +               cookie->bindex = bbot;
26275 +       bbot = au_fbbot_dir(file);
26276 +       /* AuDbg("b%d, b%d\n", cookie->bindex, bbot); */
26277 +       for (; !err && cookie->bindex <= bbot;
26278 +            cookie->bindex++, cookie->h_pos = 0) {
26279 +               h_file = au_hf_dir(file, cookie->bindex);
26280 +               if (!h_file)
26281 +                       continue;
26282 +
26283 +               au_fclr_rdu(cookie->flags, FULL);
26284 +               err = au_rdu_do(h_file, &arg);
26285 +               AuTraceErr(err);
26286 +               if (unlikely(au_ftest_rdu(cookie->flags, FULL) || err))
26287 +                       break;
26288 +       }
26289 +       AuDbg("rent %llu\n", rdu->rent);
26290 +
26291 +       if (!err && !au_ftest_rdu(cookie->flags, CONT)) {
26292 +               rdu->shwh = !!au_opt_test(au_sbi(arg.sb)->si_mntflags, SHWH);
26293 +               au_fset_rdu(cookie->flags, CONT);
26294 +               cookie->generation = au_figen(file);
26295 +       }
26296 +
26297 +       ii_read_lock_child(inode);
26298 +       fsstack_copy_attr_atime(inode, au_h_iptr(inode, au_ibtop(inode)));
26299 +       ii_read_unlock(inode);
26300 +
26301 +out_unlock:
26302 +       fi_read_unlock(file);
26303 +out_si:
26304 +       si_read_unlock(arg.sb);
26305 +out_mtx:
26306 +       inode_unlock_shared(inode);
26307 +out:
26308 +       AuTraceErr(err);
26309 +       return err;
26310 +}
26311 +
26312 +static int au_rdu_ino(struct file *file, struct aufs_rdu *rdu)
26313 +{
26314 +       int err;
26315 +       ino_t ino;
26316 +       unsigned long long nent;
26317 +       union au_rdu_ent_ul *u;
26318 +       struct au_rdu_ent ent;
26319 +       struct super_block *sb;
26320 +
26321 +       err = 0;
26322 +       nent = rdu->nent;
26323 +       u = &rdu->ent;
26324 +       sb = file->f_path.dentry->d_sb;
26325 +       si_read_lock(sb, AuLock_FLUSH);
26326 +       while (nent-- > 0) {
26327 +               /* unnecessary to support mmap_sem since this is a dir */
26328 +               err = copy_from_user(&ent, u->e, sizeof(ent));
26329 +               if (!err)
26330 +                       err = !access_ok(VERIFY_WRITE, &u->e->ino, sizeof(ino));
26331 +               if (unlikely(err)) {
26332 +                       err = -EFAULT;
26333 +                       AuTraceErr(err);
26334 +                       break;
26335 +               }
26336 +
26337 +               /* AuDbg("b%d, i%llu\n", ent.bindex, ent.ino); */
26338 +               if (!ent.wh)
26339 +                       err = au_ino(sb, ent.bindex, ent.ino, ent.type, &ino);
26340 +               else
26341 +                       err = au_wh_ino(sb, ent.bindex, ent.ino, ent.type,
26342 +                                       &ino);
26343 +               if (unlikely(err)) {
26344 +                       AuTraceErr(err);
26345 +                       break;
26346 +               }
26347 +
26348 +               err = __put_user(ino, &u->e->ino);
26349 +               if (unlikely(err)) {
26350 +                       err = -EFAULT;
26351 +                       AuTraceErr(err);
26352 +                       break;
26353 +               }
26354 +               u->ul += au_rdu_len(ent.nlen);
26355 +       }
26356 +       si_read_unlock(sb);
26357 +
26358 +       return err;
26359 +}
26360 +
26361 +/* ---------------------------------------------------------------------- */
26362 +
26363 +static int au_rdu_verify(struct aufs_rdu *rdu)
26364 +{
26365 +       AuDbg("rdu{%llu, %p, %u | %u | %llu, %u, %u | "
26366 +             "%llu, b%d, 0x%x, g%u}\n",
26367 +             rdu->sz, rdu->ent.e, rdu->verify[AufsCtlRduV_SZ],
26368 +             rdu->blk,
26369 +             rdu->rent, rdu->shwh, rdu->full,
26370 +             rdu->cookie.h_pos, rdu->cookie.bindex, rdu->cookie.flags,
26371 +             rdu->cookie.generation);
26372 +
26373 +       if (rdu->verify[AufsCtlRduV_SZ] == sizeof(*rdu))
26374 +               return 0;
26375 +
26376 +       AuDbg("%u:%u\n",
26377 +             rdu->verify[AufsCtlRduV_SZ], (unsigned int)sizeof(*rdu));
26378 +       return -EINVAL;
26379 +}
26380 +
26381 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
26382 +{
26383 +       long err, e;
26384 +       struct aufs_rdu rdu;
26385 +       void __user *p = (void __user *)arg;
26386 +
26387 +       err = copy_from_user(&rdu, p, sizeof(rdu));
26388 +       if (unlikely(err)) {
26389 +               err = -EFAULT;
26390 +               AuTraceErr(err);
26391 +               goto out;
26392 +       }
26393 +       err = au_rdu_verify(&rdu);
26394 +       if (unlikely(err))
26395 +               goto out;
26396 +
26397 +       switch (cmd) {
26398 +       case AUFS_CTL_RDU:
26399 +               err = au_rdu(file, &rdu);
26400 +               if (unlikely(err))
26401 +                       break;
26402 +
26403 +               e = copy_to_user(p, &rdu, sizeof(rdu));
26404 +               if (unlikely(e)) {
26405 +                       err = -EFAULT;
26406 +                       AuTraceErr(err);
26407 +               }
26408 +               break;
26409 +       case AUFS_CTL_RDU_INO:
26410 +               err = au_rdu_ino(file, &rdu);
26411 +               break;
26412 +
26413 +       default:
26414 +               /* err = -ENOTTY; */
26415 +               err = -EINVAL;
26416 +       }
26417 +
26418 +out:
26419 +       AuTraceErr(err);
26420 +       return err;
26421 +}
26422 +
26423 +#ifdef CONFIG_COMPAT
26424 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
26425 +{
26426 +       long err, e;
26427 +       struct aufs_rdu rdu;
26428 +       void __user *p = compat_ptr(arg);
26429 +
26430 +       /* todo: get_user()? */
26431 +       err = copy_from_user(&rdu, p, sizeof(rdu));
26432 +       if (unlikely(err)) {
26433 +               err = -EFAULT;
26434 +               AuTraceErr(err);
26435 +               goto out;
26436 +       }
26437 +       rdu.ent.e = compat_ptr(rdu.ent.ul);
26438 +       err = au_rdu_verify(&rdu);
26439 +       if (unlikely(err))
26440 +               goto out;
26441 +
26442 +       switch (cmd) {
26443 +       case AUFS_CTL_RDU:
26444 +               err = au_rdu(file, &rdu);
26445 +               if (unlikely(err))
26446 +                       break;
26447 +
26448 +               rdu.ent.ul = ptr_to_compat(rdu.ent.e);
26449 +               rdu.tail.ul = ptr_to_compat(rdu.tail.e);
26450 +               e = copy_to_user(p, &rdu, sizeof(rdu));
26451 +               if (unlikely(e)) {
26452 +                       err = -EFAULT;
26453 +                       AuTraceErr(err);
26454 +               }
26455 +               break;
26456 +       case AUFS_CTL_RDU_INO:
26457 +               err = au_rdu_ino(file, &rdu);
26458 +               break;
26459 +
26460 +       default:
26461 +               /* err = -ENOTTY; */
26462 +               err = -EINVAL;
26463 +       }
26464 +
26465 +out:
26466 +       AuTraceErr(err);
26467 +       return err;
26468 +}
26469 +#endif
26470 diff -urN /usr/share/empty/fs/aufs/rwsem.h linux/fs/aufs/rwsem.h
26471 --- /usr/share/empty/fs/aufs/rwsem.h    1970-01-01 01:00:00.000000000 +0100
26472 +++ linux/fs/aufs/rwsem.h       2016-10-09 16:55:36.496035060 +0200
26473 @@ -0,0 +1,198 @@
26474 +/*
26475 + * Copyright (C) 2005-2016 Junjiro R. Okajima
26476 + *
26477 + * This program, aufs is free software; you can redistribute it and/or modify
26478 + * it under the terms of the GNU General Public License as published by
26479 + * the Free Software Foundation; either version 2 of the License, or
26480 + * (at your option) any later version.
26481 + *
26482 + * This program is distributed in the hope that it will be useful,
26483 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26484 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26485 + * GNU General Public License for more details.
26486 + *
26487 + * You should have received a copy of the GNU General Public License
26488 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26489 + */
26490 +
26491 +/*
26492 + * simple read-write semaphore wrappers
26493 + */
26494 +
26495 +#ifndef __AUFS_RWSEM_H__
26496 +#define __AUFS_RWSEM_H__
26497 +
26498 +#ifdef __KERNEL__
26499 +
26500 +#include "debug.h"
26501 +
26502 +struct au_rwsem {
26503 +       struct rw_semaphore     rwsem;
26504 +#ifdef CONFIG_AUFS_DEBUG
26505 +       /* just for debugging, not almighty counter */
26506 +       atomic_t                rcnt, wcnt;
26507 +#endif
26508 +};
26509 +
26510 +#ifdef CONFIG_LOCKDEP
26511 +#define au_lockdep_set_name(rw)                                                \
26512 +       lockdep_set_class_and_name(&(rw)->rwsem,                        \
26513 +                                  /*original key*/(rw)->rwsem.dep_map.key, \
26514 +                                  /*name*/#rw)
26515 +#else
26516 +#define au_lockdep_set_name(rw) do {} while (0)
26517 +#endif
26518 +
26519 +#ifdef CONFIG_AUFS_DEBUG
26520 +#define AuDbgCntInit(rw) do { \
26521 +       atomic_set(&(rw)->rcnt, 0); \
26522 +       atomic_set(&(rw)->wcnt, 0); \
26523 +       smp_mb(); /* atomic set */ \
26524 +} while (0)
26525 +
26526 +#define AuDbgCnt(rw, cnt)      atomic_read(&(rw)->cnt)
26527 +#define AuDbgCntInc(rw, cnt)   atomic_inc(&(rw)->cnt)
26528 +#define AuDbgCntDec(rw, cnt)   WARN_ON(atomic_dec_return(&(rw)->cnt) < 0)
26529 +#define AuDbgRcntInc(rw)       AuDbgCntInc(rw, rcnt)
26530 +#define AuDbgRcntDec(rw)       AuDbgCntDec(rw, rcnt)
26531 +#define AuDbgWcntInc(rw)       AuDbgCntInc(rw, wcnt)
26532 +#define AuDbgWcntDec(rw)       AuDbgCntDec(rw, wcnt)
26533 +#else
26534 +#define AuDbgCnt(rw, cnt)      0
26535 +#define AuDbgCntInit(rw)       do {} while (0)
26536 +#define AuDbgRcntInc(rw)       do {} while (0)
26537 +#define AuDbgRcntDec(rw)       do {} while (0)
26538 +#define AuDbgWcntInc(rw)       do {} while (0)
26539 +#define AuDbgWcntDec(rw)       do {} while (0)
26540 +#endif /* CONFIG_AUFS_DEBUG */
26541 +
26542 +/* to debug easier, do not make them inlined functions */
26543 +#define AuRwMustNoWaiters(rw)  AuDebugOn(rwsem_is_contended(&(rw)->rwsem))
26544 +/* rwsem_is_locked() is unusable */
26545 +#define AuRwMustReadLock(rw)   AuDebugOn(AuDbgCnt(rw, rcnt) <= 0)
26546 +#define AuRwMustWriteLock(rw)  AuDebugOn(AuDbgCnt(rw, wcnt) <= 0)
26547 +#define AuRwMustAnyLock(rw)    AuDebugOn(AuDbgCnt(rw, rcnt) <= 0       \
26548 +                                         && AuDbgCnt(rw, wcnt) <= 0)
26549 +#define AuRwDestroy(rw)                AuDebugOn(AuDbgCnt(rw, rcnt)            \
26550 +                                         || AuDbgCnt(rw, wcnt))
26551 +
26552 +#define au_rw_init(rw) do {                    \
26553 +               AuDbgCntInit(rw);               \
26554 +               init_rwsem(&(rw)->rwsem);       \
26555 +               au_lockdep_set_name(rw);        \
26556 +       } while (0)
26557 +
26558 +#define au_rw_init_wlock(rw) do {              \
26559 +               au_rw_init(rw);                 \
26560 +               down_write(&(rw)->rwsem);       \
26561 +               AuDbgWcntInc(rw);               \
26562 +       } while (0)
26563 +
26564 +#define au_rw_init_wlock_nested(rw, lsc) do { \
26565 +               au_rw_init(rw);                         \
26566 +               down_write_nested(&(rw)->rwsem, lsc);   \
26567 +               AuDbgWcntInc(rw);                       \
26568 +       } while (0)
26569 +
26570 +static inline void au_rw_read_lock(struct au_rwsem *rw)
26571 +{
26572 +       down_read(&rw->rwsem);
26573 +       AuDbgRcntInc(rw);
26574 +}
26575 +
26576 +static inline void au_rw_read_lock_nested(struct au_rwsem *rw, unsigned int lsc)
26577 +{
26578 +       down_read_nested(&rw->rwsem, lsc);
26579 +       AuDbgRcntInc(rw);
26580 +}
26581 +
26582 +static inline void au_rw_read_unlock(struct au_rwsem *rw)
26583 +{
26584 +       AuRwMustReadLock(rw);
26585 +       AuDbgRcntDec(rw);
26586 +       up_read(&rw->rwsem);
26587 +}
26588 +
26589 +static inline void au_rw_dgrade_lock(struct au_rwsem *rw)
26590 +{
26591 +       AuRwMustWriteLock(rw);
26592 +       AuDbgRcntInc(rw);
26593 +       AuDbgWcntDec(rw);
26594 +       downgrade_write(&rw->rwsem);
26595 +}
26596 +
26597 +static inline void au_rw_write_lock(struct au_rwsem *rw)
26598 +{
26599 +       down_write(&rw->rwsem);
26600 +       AuDbgWcntInc(rw);
26601 +}
26602 +
26603 +static inline void au_rw_write_lock_nested(struct au_rwsem *rw,
26604 +                                          unsigned int lsc)
26605 +{
26606 +       down_write_nested(&rw->rwsem, lsc);
26607 +       AuDbgWcntInc(rw);
26608 +}
26609 +
26610 +static inline void au_rw_write_unlock(struct au_rwsem *rw)
26611 +{
26612 +       AuRwMustWriteLock(rw);
26613 +       AuDbgWcntDec(rw);
26614 +       up_write(&rw->rwsem);
26615 +}
26616 +
26617 +/* why is not _nested version defined */
26618 +static inline int au_rw_read_trylock(struct au_rwsem *rw)
26619 +{
26620 +       int ret;
26621 +
26622 +       ret = down_read_trylock(&rw->rwsem);
26623 +       if (ret)
26624 +               AuDbgRcntInc(rw);
26625 +       return ret;
26626 +}
26627 +
26628 +static inline int au_rw_write_trylock(struct au_rwsem *rw)
26629 +{
26630 +       int ret;
26631 +
26632 +       ret = down_write_trylock(&rw->rwsem);
26633 +       if (ret)
26634 +               AuDbgWcntInc(rw);
26635 +       return ret;
26636 +}
26637 +
26638 +#undef AuDbgCntDec
26639 +#undef AuDbgRcntInc
26640 +#undef AuDbgRcntDec
26641 +#undef AuDbgWcntDec
26642 +
26643 +#define AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
26644 +static inline void prefix##_read_lock(param) \
26645 +{ au_rw_read_lock(rwsem); } \
26646 +static inline void prefix##_write_lock(param) \
26647 +{ au_rw_write_lock(rwsem); } \
26648 +static inline int prefix##_read_trylock(param) \
26649 +{ return au_rw_read_trylock(rwsem); } \
26650 +static inline int prefix##_write_trylock(param) \
26651 +{ return au_rw_write_trylock(rwsem); }
26652 +/* why is not _nested version defined */
26653 +/* static inline void prefix##_read_trylock_nested(param, lsc)
26654 +{ au_rw_read_trylock_nested(rwsem, lsc)); }
26655 +static inline void prefix##_write_trylock_nestd(param, lsc)
26656 +{ au_rw_write_trylock_nested(rwsem, lsc); } */
26657 +
26658 +#define AuSimpleUnlockRwsemFuncs(prefix, param, rwsem) \
26659 +static inline void prefix##_read_unlock(param) \
26660 +{ au_rw_read_unlock(rwsem); } \
26661 +static inline void prefix##_write_unlock(param) \
26662 +{ au_rw_write_unlock(rwsem); } \
26663 +static inline void prefix##_downgrade_lock(param) \
26664 +{ au_rw_dgrade_lock(rwsem); }
26665 +
26666 +#define AuSimpleRwsemFuncs(prefix, param, rwsem) \
26667 +       AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
26668 +       AuSimpleUnlockRwsemFuncs(prefix, param, rwsem)
26669 +
26670 +#endif /* __KERNEL__ */
26671 +#endif /* __AUFS_RWSEM_H__ */
26672 diff -urN /usr/share/empty/fs/aufs/sbinfo.c linux/fs/aufs/sbinfo.c
26673 --- /usr/share/empty/fs/aufs/sbinfo.c   1970-01-01 01:00:00.000000000 +0100
26674 +++ linux/fs/aufs/sbinfo.c      2016-10-09 16:55:38.889431135 +0200
26675 @@ -0,0 +1,355 @@
26676 +/*
26677 + * Copyright (C) 2005-2016 Junjiro R. Okajima
26678 + *
26679 + * This program, aufs is free software; you can redistribute it and/or modify
26680 + * it under the terms of the GNU General Public License as published by
26681 + * the Free Software Foundation; either version 2 of the License, or
26682 + * (at your option) any later version.
26683 + *
26684 + * This program is distributed in the hope that it will be useful,
26685 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26686 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26687 + * GNU General Public License for more details.
26688 + *
26689 + * You should have received a copy of the GNU General Public License
26690 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26691 + */
26692 +
26693 +/*
26694 + * superblock private data
26695 + */
26696 +
26697 +#include "aufs.h"
26698 +
26699 +/*
26700 + * they are necessary regardless sysfs is disabled.
26701 + */
26702 +void au_si_free(struct kobject *kobj)
26703 +{
26704 +       int i;
26705 +       struct au_sbinfo *sbinfo;
26706 +       char *locked __maybe_unused; /* debug only */
26707 +
26708 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
26709 +       for (i = 0; i < AuPlink_NHASH; i++)
26710 +               AuDebugOn(!hlist_empty(&sbinfo->si_plink[i].head));
26711 +       AuDebugOn(atomic_read(&sbinfo->si_nowait.nw_len));
26712 +
26713 +       AuDebugOn(percpu_counter_sum(&sbinfo->si_ninodes));
26714 +       percpu_counter_destroy(&sbinfo->si_ninodes);
26715 +       AuDebugOn(percpu_counter_sum(&sbinfo->si_nfiles));
26716 +       percpu_counter_destroy(&sbinfo->si_nfiles);
26717 +
26718 +       au_rw_write_lock(&sbinfo->si_rwsem);
26719 +       au_br_free(sbinfo);
26720 +       au_rw_write_unlock(&sbinfo->si_rwsem);
26721 +
26722 +       au_delayed_kfree(sbinfo->si_branch);
26723 +       for (i = 0; i < AU_NPIDMAP; i++)
26724 +               if (sbinfo->au_si_pid.pid_bitmap[i])
26725 +                       au_delayed_kfree(sbinfo->au_si_pid.pid_bitmap[i]);
26726 +       mutex_destroy(&sbinfo->au_si_pid.pid_mtx);
26727 +       mutex_destroy(&sbinfo->si_xib_mtx);
26728 +       AuRwDestroy(&sbinfo->si_rwsem);
26729 +
26730 +       au_delayed_kfree(sbinfo);
26731 +}
26732 +
26733 +int au_si_alloc(struct super_block *sb)
26734 +{
26735 +       int err, i;
26736 +       struct au_sbinfo *sbinfo;
26737 +
26738 +       err = -ENOMEM;
26739 +       sbinfo = kzalloc(sizeof(*sbinfo), GFP_NOFS);
26740 +       if (unlikely(!sbinfo))
26741 +               goto out;
26742 +
26743 +       /* will be reallocated separately */
26744 +       sbinfo->si_branch = kzalloc(sizeof(*sbinfo->si_branch), GFP_NOFS);
26745 +       if (unlikely(!sbinfo->si_branch))
26746 +               goto out_sbinfo;
26747 +
26748 +       err = sysaufs_si_init(sbinfo);
26749 +       if (unlikely(err))
26750 +               goto out_br;
26751 +
26752 +       au_nwt_init(&sbinfo->si_nowait);
26753 +       au_rw_init_wlock(&sbinfo->si_rwsem);
26754 +       mutex_init(&sbinfo->au_si_pid.pid_mtx);
26755 +
26756 +       percpu_counter_init(&sbinfo->si_ninodes, 0, GFP_NOFS);
26757 +       percpu_counter_init(&sbinfo->si_nfiles, 0, GFP_NOFS);
26758 +
26759 +       sbinfo->si_bbot = -1;
26760 +       sbinfo->si_last_br_id = AUFS_BRANCH_MAX / 2;
26761 +
26762 +       sbinfo->si_wbr_copyup = AuWbrCopyup_Def;
26763 +       sbinfo->si_wbr_create = AuWbrCreate_Def;
26764 +       sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + sbinfo->si_wbr_copyup;
26765 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + sbinfo->si_wbr_create;
26766 +
26767 +       au_fhsm_init(sbinfo);
26768 +
26769 +       sbinfo->si_mntflags = au_opts_plink(AuOpt_Def);
26770 +
26771 +       sbinfo->si_xino_jiffy = jiffies;
26772 +       sbinfo->si_xino_expire
26773 +               = msecs_to_jiffies(AUFS_XINO_DEF_SEC * MSEC_PER_SEC);
26774 +       mutex_init(&sbinfo->si_xib_mtx);
26775 +       sbinfo->si_xino_brid = -1;
26776 +       /* leave si_xib_last_pindex and si_xib_next_bit */
26777 +
26778 +       au_sphl_init(&sbinfo->si_aopen);
26779 +
26780 +       sbinfo->si_rdcache = msecs_to_jiffies(AUFS_RDCACHE_DEF * MSEC_PER_SEC);
26781 +       sbinfo->si_rdblk = AUFS_RDBLK_DEF;
26782 +       sbinfo->si_rdhash = AUFS_RDHASH_DEF;
26783 +       sbinfo->si_dirwh = AUFS_DIRWH_DEF;
26784 +
26785 +       for (i = 0; i < AuPlink_NHASH; i++)
26786 +               au_sphl_init(sbinfo->si_plink + i);
26787 +       init_waitqueue_head(&sbinfo->si_plink_wq);
26788 +       spin_lock_init(&sbinfo->si_plink_maint_lock);
26789 +
26790 +       au_sphl_init(&sbinfo->si_files);
26791 +
26792 +       /* with getattr by default */
26793 +       sbinfo->si_iop_array = aufs_iop;
26794 +
26795 +       /* leave other members for sysaufs and si_mnt. */
26796 +       sbinfo->si_sb = sb;
26797 +       sb->s_fs_info = sbinfo;
26798 +       si_pid_set(sb);
26799 +       return 0; /* success */
26800 +
26801 +out_br:
26802 +       au_delayed_kfree(sbinfo->si_branch);
26803 +out_sbinfo:
26804 +       au_delayed_kfree(sbinfo);
26805 +out:
26806 +       return err;
26807 +}
26808 +
26809 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr, int may_shrink)
26810 +{
26811 +       int err, sz;
26812 +       struct au_branch **brp;
26813 +
26814 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
26815 +
26816 +       err = -ENOMEM;
26817 +       sz = sizeof(*brp) * (sbinfo->si_bbot + 1);
26818 +       if (unlikely(!sz))
26819 +               sz = sizeof(*brp);
26820 +       brp = au_kzrealloc(sbinfo->si_branch, sz, sizeof(*brp) * nbr, GFP_NOFS,
26821 +                          may_shrink);
26822 +       if (brp) {
26823 +               sbinfo->si_branch = brp;
26824 +               err = 0;
26825 +       }
26826 +
26827 +       return err;
26828 +}
26829 +
26830 +/* ---------------------------------------------------------------------- */
26831 +
26832 +unsigned int au_sigen_inc(struct super_block *sb)
26833 +{
26834 +       unsigned int gen;
26835 +       struct inode *inode;
26836 +
26837 +       SiMustWriteLock(sb);
26838 +
26839 +       gen = ++au_sbi(sb)->si_generation;
26840 +       au_update_digen(sb->s_root);
26841 +       inode = d_inode(sb->s_root);
26842 +       au_update_iigen(inode, /*half*/0);
26843 +       inode->i_version++;
26844 +       return gen;
26845 +}
26846 +
26847 +aufs_bindex_t au_new_br_id(struct super_block *sb)
26848 +{
26849 +       aufs_bindex_t br_id;
26850 +       int i;
26851 +       struct au_sbinfo *sbinfo;
26852 +
26853 +       SiMustWriteLock(sb);
26854 +
26855 +       sbinfo = au_sbi(sb);
26856 +       for (i = 0; i <= AUFS_BRANCH_MAX; i++) {
26857 +               br_id = ++sbinfo->si_last_br_id;
26858 +               AuDebugOn(br_id < 0);
26859 +               if (br_id && au_br_index(sb, br_id) < 0)
26860 +                       return br_id;
26861 +       }
26862 +
26863 +       return -1;
26864 +}
26865 +
26866 +/* ---------------------------------------------------------------------- */
26867 +
26868 +/* it is ok that new 'nwt' tasks are appended while we are sleeping */
26869 +int si_read_lock(struct super_block *sb, int flags)
26870 +{
26871 +       int err;
26872 +
26873 +       err = 0;
26874 +       if (au_ftest_lock(flags, FLUSH))
26875 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26876 +
26877 +       si_noflush_read_lock(sb);
26878 +       err = au_plink_maint(sb, flags);
26879 +       if (unlikely(err))
26880 +               si_read_unlock(sb);
26881 +
26882 +       return err;
26883 +}
26884 +
26885 +int si_write_lock(struct super_block *sb, int flags)
26886 +{
26887 +       int err;
26888 +
26889 +       if (au_ftest_lock(flags, FLUSH))
26890 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26891 +
26892 +       si_noflush_write_lock(sb);
26893 +       err = au_plink_maint(sb, flags);
26894 +       if (unlikely(err))
26895 +               si_write_unlock(sb);
26896 +
26897 +       return err;
26898 +}
26899 +
26900 +/* dentry and super_block lock. call at entry point */
26901 +int aufs_read_lock(struct dentry *dentry, int flags)
26902 +{
26903 +       int err;
26904 +       struct super_block *sb;
26905 +
26906 +       sb = dentry->d_sb;
26907 +       err = si_read_lock(sb, flags);
26908 +       if (unlikely(err))
26909 +               goto out;
26910 +
26911 +       if (au_ftest_lock(flags, DW))
26912 +               di_write_lock_child(dentry);
26913 +       else
26914 +               di_read_lock_child(dentry, flags);
26915 +
26916 +       if (au_ftest_lock(flags, GEN)) {
26917 +               err = au_digen_test(dentry, au_sigen(sb));
26918 +               if (!au_opt_test(au_mntflags(sb), UDBA_NONE))
26919 +                       AuDebugOn(!err && au_dbrange_test(dentry));
26920 +               else if (!err)
26921 +                       err = au_dbrange_test(dentry);
26922 +               if (unlikely(err))
26923 +                       aufs_read_unlock(dentry, flags);
26924 +       }
26925 +
26926 +out:
26927 +       return err;
26928 +}
26929 +
26930 +void aufs_read_unlock(struct dentry *dentry, int flags)
26931 +{
26932 +       if (au_ftest_lock(flags, DW))
26933 +               di_write_unlock(dentry);
26934 +       else
26935 +               di_read_unlock(dentry, flags);
26936 +       si_read_unlock(dentry->d_sb);
26937 +}
26938 +
26939 +void aufs_write_lock(struct dentry *dentry)
26940 +{
26941 +       si_write_lock(dentry->d_sb, AuLock_FLUSH | AuLock_NOPLMW);
26942 +       di_write_lock_child(dentry);
26943 +}
26944 +
26945 +void aufs_write_unlock(struct dentry *dentry)
26946 +{
26947 +       di_write_unlock(dentry);
26948 +       si_write_unlock(dentry->d_sb);
26949 +}
26950 +
26951 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags)
26952 +{
26953 +       int err;
26954 +       unsigned int sigen;
26955 +       struct super_block *sb;
26956 +
26957 +       sb = d1->d_sb;
26958 +       err = si_read_lock(sb, flags);
26959 +       if (unlikely(err))
26960 +               goto out;
26961 +
26962 +       di_write_lock2_child(d1, d2, au_ftest_lock(flags, DIRS));
26963 +
26964 +       if (au_ftest_lock(flags, GEN)) {
26965 +               sigen = au_sigen(sb);
26966 +               err = au_digen_test(d1, sigen);
26967 +               AuDebugOn(!err && au_dbrange_test(d1));
26968 +               if (!err) {
26969 +                       err = au_digen_test(d2, sigen);
26970 +                       AuDebugOn(!err && au_dbrange_test(d2));
26971 +               }
26972 +               if (unlikely(err))
26973 +                       aufs_read_and_write_unlock2(d1, d2);
26974 +       }
26975 +
26976 +out:
26977 +       return err;
26978 +}
26979 +
26980 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2)
26981 +{
26982 +       di_write_unlock2(d1, d2);
26983 +       si_read_unlock(d1->d_sb);
26984 +}
26985 +
26986 +/* ---------------------------------------------------------------------- */
26987 +
26988 +static void si_pid_alloc(struct au_si_pid *au_si_pid, int idx)
26989 +{
26990 +       unsigned long *p;
26991 +
26992 +       BUILD_BUG_ON(sizeof(unsigned long) !=
26993 +                    sizeof(*au_si_pid->pid_bitmap));
26994 +
26995 +       mutex_lock(&au_si_pid->pid_mtx);
26996 +       p = au_si_pid->pid_bitmap[idx];
26997 +       while (!p) {
26998 +               /*
26999 +                * bad approach.
27000 +                * but keeping 'si_pid_set()' void is more important.
27001 +                */
27002 +               p = kcalloc(BITS_TO_LONGS(AU_PIDSTEP),
27003 +                           sizeof(*au_si_pid->pid_bitmap),
27004 +                           GFP_NOFS);
27005 +               if (p)
27006 +                       break;
27007 +               cond_resched();
27008 +       }
27009 +       au_si_pid->pid_bitmap[idx] = p;
27010 +       mutex_unlock(&au_si_pid->pid_mtx);
27011 +}
27012 +
27013 +void si_pid_set(struct super_block *sb)
27014 +{
27015 +       pid_t bit;
27016 +       int idx;
27017 +       unsigned long *bitmap;
27018 +       struct au_si_pid *au_si_pid;
27019 +
27020 +       si_pid_idx_bit(&idx, &bit);
27021 +       au_si_pid = &au_sbi(sb)->au_si_pid;
27022 +       bitmap = au_si_pid->pid_bitmap[idx];
27023 +       if (!bitmap) {
27024 +               si_pid_alloc(au_si_pid, idx);
27025 +               bitmap = au_si_pid->pid_bitmap[idx];
27026 +       }
27027 +       AuDebugOn(test_bit(bit, bitmap));
27028 +       set_bit(bit, bitmap);
27029 +       /* smp_mb(); */
27030 +}
27031 diff -urN /usr/share/empty/fs/aufs/spl.h linux/fs/aufs/spl.h
27032 --- /usr/share/empty/fs/aufs/spl.h      1970-01-01 01:00:00.000000000 +0100
27033 +++ linux/fs/aufs/spl.h 2016-10-09 16:55:36.496035060 +0200
27034 @@ -0,0 +1,113 @@
27035 +/*
27036 + * Copyright (C) 2005-2016 Junjiro R. Okajima
27037 + *
27038 + * This program, aufs is free software; you can redistribute it and/or modify
27039 + * it under the terms of the GNU General Public License as published by
27040 + * the Free Software Foundation; either version 2 of the License, or
27041 + * (at your option) any later version.
27042 + *
27043 + * This program is distributed in the hope that it will be useful,
27044 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27045 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27046 + * GNU General Public License for more details.
27047 + *
27048 + * You should have received a copy of the GNU General Public License
27049 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27050 + */
27051 +
27052 +/*
27053 + * simple list protected by a spinlock
27054 + */
27055 +
27056 +#ifndef __AUFS_SPL_H__
27057 +#define __AUFS_SPL_H__
27058 +
27059 +#ifdef __KERNEL__
27060 +
27061 +#if 0
27062 +struct au_splhead {
27063 +       spinlock_t              spin;
27064 +       struct list_head        head;
27065 +};
27066 +
27067 +static inline void au_spl_init(struct au_splhead *spl)
27068 +{
27069 +       spin_lock_init(&spl->spin);
27070 +       INIT_LIST_HEAD(&spl->head);
27071 +}
27072 +
27073 +static inline void au_spl_add(struct list_head *list, struct au_splhead *spl)
27074 +{
27075 +       spin_lock(&spl->spin);
27076 +       list_add(list, &spl->head);
27077 +       spin_unlock(&spl->spin);
27078 +}
27079 +
27080 +static inline void au_spl_del(struct list_head *list, struct au_splhead *spl)
27081 +{
27082 +       spin_lock(&spl->spin);
27083 +       list_del(list);
27084 +       spin_unlock(&spl->spin);
27085 +}
27086 +
27087 +static inline void au_spl_del_rcu(struct list_head *list,
27088 +                                 struct au_splhead *spl)
27089 +{
27090 +       spin_lock(&spl->spin);
27091 +       list_del_rcu(list);
27092 +       spin_unlock(&spl->spin);
27093 +}
27094 +#endif
27095 +
27096 +/* ---------------------------------------------------------------------- */
27097 +
27098 +struct au_sphlhead {
27099 +       spinlock_t              spin;
27100 +       struct hlist_head       head;
27101 +};
27102 +
27103 +static inline void au_sphl_init(struct au_sphlhead *sphl)
27104 +{
27105 +       spin_lock_init(&sphl->spin);
27106 +       INIT_HLIST_HEAD(&sphl->head);
27107 +}
27108 +
27109 +static inline void au_sphl_add(struct hlist_node *hlist,
27110 +                              struct au_sphlhead *sphl)
27111 +{
27112 +       spin_lock(&sphl->spin);
27113 +       hlist_add_head(hlist, &sphl->head);
27114 +       spin_unlock(&sphl->spin);
27115 +}
27116 +
27117 +static inline void au_sphl_del(struct hlist_node *hlist,
27118 +                              struct au_sphlhead *sphl)
27119 +{
27120 +       spin_lock(&sphl->spin);
27121 +       hlist_del(hlist);
27122 +       spin_unlock(&sphl->spin);
27123 +}
27124 +
27125 +static inline void au_sphl_del_rcu(struct hlist_node *hlist,
27126 +                                  struct au_sphlhead *sphl)
27127 +{
27128 +       spin_lock(&sphl->spin);
27129 +       hlist_del_rcu(hlist);
27130 +       spin_unlock(&sphl->spin);
27131 +}
27132 +
27133 +static inline unsigned long au_sphl_count(struct au_sphlhead *sphl)
27134 +{
27135 +       unsigned long cnt;
27136 +       struct hlist_node *pos;
27137 +
27138 +       cnt = 0;
27139 +       spin_lock(&sphl->spin);
27140 +       hlist_for_each(pos, &sphl->head)
27141 +               cnt++;
27142 +       spin_unlock(&sphl->spin);
27143 +       return cnt;
27144 +}
27145 +
27146 +#endif /* __KERNEL__ */
27147 +#endif /* __AUFS_SPL_H__ */
27148 diff -urN /usr/share/empty/fs/aufs/super.c linux/fs/aufs/super.c
27149 --- /usr/share/empty/fs/aufs/super.c    1970-01-01 01:00:00.000000000 +0100
27150 +++ linux/fs/aufs/super.c       2016-10-09 16:55:36.496035060 +0200
27151 @@ -0,0 +1,1038 @@
27152 +/*
27153 + * Copyright (C) 2005-2016 Junjiro R. Okajima
27154 + *
27155 + * This program, aufs is free software; you can redistribute it and/or modify
27156 + * it under the terms of the GNU General Public License as published by
27157 + * the Free Software Foundation; either version 2 of the License, or
27158 + * (at your option) any later version.
27159 + *
27160 + * This program is distributed in the hope that it will be useful,
27161 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27162 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27163 + * GNU General Public License for more details.
27164 + *
27165 + * You should have received a copy of the GNU General Public License
27166 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27167 + */
27168 +
27169 +/*
27170 + * mount and super_block operations
27171 + */
27172 +
27173 +#include <linux/mm.h>
27174 +#include <linux/seq_file.h>
27175 +#include <linux/statfs.h>
27176 +#include <linux/vmalloc.h>
27177 +#include "aufs.h"
27178 +
27179 +/*
27180 + * super_operations
27181 + */
27182 +static struct inode *aufs_alloc_inode(struct super_block *sb __maybe_unused)
27183 +{
27184 +       struct au_icntnr *c;
27185 +
27186 +       c = au_cache_alloc_icntnr();
27187 +       if (c) {
27188 +               au_icntnr_init(c);
27189 +               c->vfs_inode.i_version = 1; /* sigen(sb); */
27190 +               c->iinfo.ii_hinode = NULL;
27191 +               return &c->vfs_inode;
27192 +       }
27193 +       return NULL;
27194 +}
27195 +
27196 +static void aufs_destroy_inode_cb(struct rcu_head *head)
27197 +{
27198 +       struct inode *inode = container_of(head, struct inode, i_rcu);
27199 +
27200 +       au_cache_dfree_icntnr(container_of(inode, struct au_icntnr, vfs_inode));
27201 +}
27202 +
27203 +static void aufs_destroy_inode(struct inode *inode)
27204 +{
27205 +       if (!au_is_bad_inode(inode))
27206 +               au_iinfo_fin(inode);
27207 +       call_rcu(&inode->i_rcu, aufs_destroy_inode_cb);
27208 +}
27209 +
27210 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino)
27211 +{
27212 +       struct inode *inode;
27213 +       int err;
27214 +
27215 +       inode = iget_locked(sb, ino);
27216 +       if (unlikely(!inode)) {
27217 +               inode = ERR_PTR(-ENOMEM);
27218 +               goto out;
27219 +       }
27220 +       if (!(inode->i_state & I_NEW))
27221 +               goto out;
27222 +
27223 +       err = au_xigen_new(inode);
27224 +       if (!err)
27225 +               err = au_iinfo_init(inode);
27226 +       if (!err)
27227 +               inode->i_version++;
27228 +       else {
27229 +               iget_failed(inode);
27230 +               inode = ERR_PTR(err);
27231 +       }
27232 +
27233 +out:
27234 +       /* never return NULL */
27235 +       AuDebugOn(!inode);
27236 +       AuTraceErrPtr(inode);
27237 +       return inode;
27238 +}
27239 +
27240 +/* lock free root dinfo */
27241 +static int au_show_brs(struct seq_file *seq, struct super_block *sb)
27242 +{
27243 +       int err;
27244 +       aufs_bindex_t bindex, bbot;
27245 +       struct path path;
27246 +       struct au_hdentry *hdp;
27247 +       struct au_branch *br;
27248 +       au_br_perm_str_t perm;
27249 +
27250 +       err = 0;
27251 +       bbot = au_sbbot(sb);
27252 +       bindex = 0;
27253 +       hdp = au_hdentry(au_di(sb->s_root), bindex);
27254 +       for (; !err && bindex <= bbot; bindex++, hdp++) {
27255 +               br = au_sbr(sb, bindex);
27256 +               path.mnt = au_br_mnt(br);
27257 +               path.dentry = hdp->hd_dentry;
27258 +               err = au_seq_path(seq, &path);
27259 +               if (!err) {
27260 +                       au_optstr_br_perm(&perm, br->br_perm);
27261 +                       seq_printf(seq, "=%s", perm.a);
27262 +                       if (bindex != bbot)
27263 +                               seq_putc(seq, ':');
27264 +               }
27265 +       }
27266 +       if (unlikely(err || seq_has_overflowed(seq)))
27267 +               err = -E2BIG;
27268 +
27269 +       return err;
27270 +}
27271 +
27272 +static void au_show_wbr_create(struct seq_file *m, int v,
27273 +                              struct au_sbinfo *sbinfo)
27274 +{
27275 +       const char *pat;
27276 +
27277 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
27278 +
27279 +       seq_puts(m, ",create=");
27280 +       pat = au_optstr_wbr_create(v);
27281 +       switch (v) {
27282 +       case AuWbrCreate_TDP:
27283 +       case AuWbrCreate_RR:
27284 +       case AuWbrCreate_MFS:
27285 +       case AuWbrCreate_PMFS:
27286 +               seq_puts(m, pat);
27287 +               break;
27288 +       case AuWbrCreate_MFSV:
27289 +               seq_printf(m, /*pat*/"mfs:%lu",
27290 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
27291 +                          / MSEC_PER_SEC);
27292 +               break;
27293 +       case AuWbrCreate_PMFSV:
27294 +               seq_printf(m, /*pat*/"pmfs:%lu",
27295 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
27296 +                          / MSEC_PER_SEC);
27297 +               break;
27298 +       case AuWbrCreate_MFSRR:
27299 +               seq_printf(m, /*pat*/"mfsrr:%llu",
27300 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
27301 +               break;
27302 +       case AuWbrCreate_MFSRRV:
27303 +               seq_printf(m, /*pat*/"mfsrr:%llu:%lu",
27304 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
27305 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
27306 +                          / MSEC_PER_SEC);
27307 +               break;
27308 +       case AuWbrCreate_PMFSRR:
27309 +               seq_printf(m, /*pat*/"pmfsrr:%llu",
27310 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
27311 +               break;
27312 +       case AuWbrCreate_PMFSRRV:
27313 +               seq_printf(m, /*pat*/"pmfsrr:%llu:%lu",
27314 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
27315 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
27316 +                          / MSEC_PER_SEC);
27317 +               break;
27318 +       }
27319 +}
27320 +
27321 +static int au_show_xino(struct seq_file *seq, struct super_block *sb)
27322 +{
27323 +#ifdef CONFIG_SYSFS
27324 +       return 0;
27325 +#else
27326 +       int err;
27327 +       const int len = sizeof(AUFS_XINO_FNAME) - 1;
27328 +       aufs_bindex_t bindex, brid;
27329 +       struct qstr *name;
27330 +       struct file *f;
27331 +       struct dentry *d, *h_root;
27332 +
27333 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
27334 +
27335 +       err = 0;
27336 +       f = au_sbi(sb)->si_xib;
27337 +       if (!f)
27338 +               goto out;
27339 +
27340 +       /* stop printing the default xino path on the first writable branch */
27341 +       h_root = NULL;
27342 +       brid = au_xino_brid(sb);
27343 +       if (brid >= 0) {
27344 +               bindex = au_br_index(sb, brid);
27345 +               h_root = au_hdentry(au_di(sb->s_root), bindex)->hd_dentry;
27346 +       }
27347 +       d = f->f_path.dentry;
27348 +       name = &d->d_name;
27349 +       /* safe ->d_parent because the file is unlinked */
27350 +       if (d->d_parent == h_root
27351 +           && name->len == len
27352 +           && !memcmp(name->name, AUFS_XINO_FNAME, len))
27353 +               goto out;
27354 +
27355 +       seq_puts(seq, ",xino=");
27356 +       err = au_xino_path(seq, f);
27357 +
27358 +out:
27359 +       return err;
27360 +#endif
27361 +}
27362 +
27363 +/* seq_file will re-call me in case of too long string */
27364 +static int aufs_show_options(struct seq_file *m, struct dentry *dentry)
27365 +{
27366 +       int err;
27367 +       unsigned int mnt_flags, v;
27368 +       struct super_block *sb;
27369 +       struct au_sbinfo *sbinfo;
27370 +
27371 +#define AuBool(name, str) do { \
27372 +       v = au_opt_test(mnt_flags, name); \
27373 +       if (v != au_opt_test(AuOpt_Def, name)) \
27374 +               seq_printf(m, ",%s" #str, v ? "" : "no"); \
27375 +} while (0)
27376 +
27377 +#define AuStr(name, str) do { \
27378 +       v = mnt_flags & AuOptMask_##name; \
27379 +       if (v != (AuOpt_Def & AuOptMask_##name)) \
27380 +               seq_printf(m, "," #str "=%s", au_optstr_##str(v)); \
27381 +} while (0)
27382 +
27383 +#define AuUInt(name, str, val) do { \
27384 +       if (val != AUFS_##name##_DEF) \
27385 +               seq_printf(m, "," #str "=%u", val); \
27386 +} while (0)
27387 +
27388 +       sb = dentry->d_sb;
27389 +       if (sb->s_flags & MS_POSIXACL)
27390 +               seq_puts(m, ",acl");
27391 +
27392 +       /* lock free root dinfo */
27393 +       si_noflush_read_lock(sb);
27394 +       sbinfo = au_sbi(sb);
27395 +       seq_printf(m, ",si=%lx", sysaufs_si_id(sbinfo));
27396 +
27397 +       mnt_flags = au_mntflags(sb);
27398 +       if (au_opt_test(mnt_flags, XINO)) {
27399 +               err = au_show_xino(m, sb);
27400 +               if (unlikely(err))
27401 +                       goto out;
27402 +       } else
27403 +               seq_puts(m, ",noxino");
27404 +
27405 +       AuBool(TRUNC_XINO, trunc_xino);
27406 +       AuStr(UDBA, udba);
27407 +       AuBool(SHWH, shwh);
27408 +       AuBool(PLINK, plink);
27409 +       AuBool(DIO, dio);
27410 +       AuBool(DIRPERM1, dirperm1);
27411 +
27412 +       v = sbinfo->si_wbr_create;
27413 +       if (v != AuWbrCreate_Def)
27414 +               au_show_wbr_create(m, v, sbinfo);
27415 +
27416 +       v = sbinfo->si_wbr_copyup;
27417 +       if (v != AuWbrCopyup_Def)
27418 +               seq_printf(m, ",cpup=%s", au_optstr_wbr_copyup(v));
27419 +
27420 +       v = au_opt_test(mnt_flags, ALWAYS_DIROPQ);
27421 +       if (v != au_opt_test(AuOpt_Def, ALWAYS_DIROPQ))
27422 +               seq_printf(m, ",diropq=%c", v ? 'a' : 'w');
27423 +
27424 +       AuUInt(DIRWH, dirwh, sbinfo->si_dirwh);
27425 +
27426 +       v = jiffies_to_msecs(sbinfo->si_rdcache) / MSEC_PER_SEC;
27427 +       AuUInt(RDCACHE, rdcache, v);
27428 +
27429 +       AuUInt(RDBLK, rdblk, sbinfo->si_rdblk);
27430 +       AuUInt(RDHASH, rdhash, sbinfo->si_rdhash);
27431 +
27432 +       au_fhsm_show(m, sbinfo);
27433 +
27434 +       AuBool(SUM, sum);
27435 +       /* AuBool(SUM_W, wsum); */
27436 +       AuBool(WARN_PERM, warn_perm);
27437 +       AuBool(VERBOSE, verbose);
27438 +
27439 +out:
27440 +       /* be sure to print "br:" last */
27441 +       if (!sysaufs_brs) {
27442 +               seq_puts(m, ",br:");
27443 +               au_show_brs(m, sb);
27444 +       }
27445 +       si_read_unlock(sb);
27446 +       return 0;
27447 +
27448 +#undef AuBool
27449 +#undef AuStr
27450 +#undef AuUInt
27451 +}
27452 +
27453 +/* ---------------------------------------------------------------------- */
27454 +
27455 +/* sum mode which returns the summation for statfs(2) */
27456 +
27457 +static u64 au_add_till_max(u64 a, u64 b)
27458 +{
27459 +       u64 old;
27460 +
27461 +       old = a;
27462 +       a += b;
27463 +       if (old <= a)
27464 +               return a;
27465 +       return ULLONG_MAX;
27466 +}
27467 +
27468 +static u64 au_mul_till_max(u64 a, long mul)
27469 +{
27470 +       u64 old;
27471 +
27472 +       old = a;
27473 +       a *= mul;
27474 +       if (old <= a)
27475 +               return a;
27476 +       return ULLONG_MAX;
27477 +}
27478 +
27479 +static int au_statfs_sum(struct super_block *sb, struct kstatfs *buf)
27480 +{
27481 +       int err;
27482 +       long bsize, factor;
27483 +       u64 blocks, bfree, bavail, files, ffree;
27484 +       aufs_bindex_t bbot, bindex, i;
27485 +       unsigned char shared;
27486 +       struct path h_path;
27487 +       struct super_block *h_sb;
27488 +
27489 +       err = 0;
27490 +       bsize = LONG_MAX;
27491 +       files = 0;
27492 +       ffree = 0;
27493 +       blocks = 0;
27494 +       bfree = 0;
27495 +       bavail = 0;
27496 +       bbot = au_sbbot(sb);
27497 +       for (bindex = 0; bindex <= bbot; bindex++) {
27498 +               h_path.mnt = au_sbr_mnt(sb, bindex);
27499 +               h_sb = h_path.mnt->mnt_sb;
27500 +               shared = 0;
27501 +               for (i = 0; !shared && i < bindex; i++)
27502 +                       shared = (au_sbr_sb(sb, i) == h_sb);
27503 +               if (shared)
27504 +                       continue;
27505 +
27506 +               /* sb->s_root for NFS is unreliable */
27507 +               h_path.dentry = h_path.mnt->mnt_root;
27508 +               err = vfs_statfs(&h_path, buf);
27509 +               if (unlikely(err))
27510 +                       goto out;
27511 +
27512 +               if (bsize > buf->f_bsize) {
27513 +                       /*
27514 +                        * we will reduce bsize, so we have to expand blocks
27515 +                        * etc. to match them again
27516 +                        */
27517 +                       factor = (bsize / buf->f_bsize);
27518 +                       blocks = au_mul_till_max(blocks, factor);
27519 +                       bfree = au_mul_till_max(bfree, factor);
27520 +                       bavail = au_mul_till_max(bavail, factor);
27521 +                       bsize = buf->f_bsize;
27522 +               }
27523 +
27524 +               factor = (buf->f_bsize / bsize);
27525 +               blocks = au_add_till_max(blocks,
27526 +                               au_mul_till_max(buf->f_blocks, factor));
27527 +               bfree = au_add_till_max(bfree,
27528 +                               au_mul_till_max(buf->f_bfree, factor));
27529 +               bavail = au_add_till_max(bavail,
27530 +                               au_mul_till_max(buf->f_bavail, factor));
27531 +               files = au_add_till_max(files, buf->f_files);
27532 +               ffree = au_add_till_max(ffree, buf->f_ffree);
27533 +       }
27534 +
27535 +       buf->f_bsize = bsize;
27536 +       buf->f_blocks = blocks;
27537 +       buf->f_bfree = bfree;
27538 +       buf->f_bavail = bavail;
27539 +       buf->f_files = files;
27540 +       buf->f_ffree = ffree;
27541 +       buf->f_frsize = 0;
27542 +
27543 +out:
27544 +       return err;
27545 +}
27546 +
27547 +static int aufs_statfs(struct dentry *dentry, struct kstatfs *buf)
27548 +{
27549 +       int err;
27550 +       struct path h_path;
27551 +       struct super_block *sb;
27552 +
27553 +       /* lock free root dinfo */
27554 +       sb = dentry->d_sb;
27555 +       si_noflush_read_lock(sb);
27556 +       if (!au_opt_test(au_mntflags(sb), SUM)) {
27557 +               /* sb->s_root for NFS is unreliable */
27558 +               h_path.mnt = au_sbr_mnt(sb, 0);
27559 +               h_path.dentry = h_path.mnt->mnt_root;
27560 +               err = vfs_statfs(&h_path, buf);
27561 +       } else
27562 +               err = au_statfs_sum(sb, buf);
27563 +       si_read_unlock(sb);
27564 +
27565 +       if (!err) {
27566 +               buf->f_type = AUFS_SUPER_MAGIC;
27567 +               buf->f_namelen = AUFS_MAX_NAMELEN;
27568 +               memset(&buf->f_fsid, 0, sizeof(buf->f_fsid));
27569 +       }
27570 +       /* buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1; */
27571 +
27572 +       return err;
27573 +}
27574 +
27575 +/* ---------------------------------------------------------------------- */
27576 +
27577 +static int aufs_sync_fs(struct super_block *sb, int wait)
27578 +{
27579 +       int err, e;
27580 +       aufs_bindex_t bbot, bindex;
27581 +       struct au_branch *br;
27582 +       struct super_block *h_sb;
27583 +
27584 +       err = 0;
27585 +       si_noflush_read_lock(sb);
27586 +       bbot = au_sbbot(sb);
27587 +       for (bindex = 0; bindex <= bbot; bindex++) {
27588 +               br = au_sbr(sb, bindex);
27589 +               if (!au_br_writable(br->br_perm))
27590 +                       continue;
27591 +
27592 +               h_sb = au_sbr_sb(sb, bindex);
27593 +               if (h_sb->s_op->sync_fs) {
27594 +                       e = h_sb->s_op->sync_fs(h_sb, wait);
27595 +                       if (unlikely(e && !err))
27596 +                               err = e;
27597 +                       /* go on even if an error happens */
27598 +               }
27599 +       }
27600 +       si_read_unlock(sb);
27601 +
27602 +       return err;
27603 +}
27604 +
27605 +/* ---------------------------------------------------------------------- */
27606 +
27607 +/* final actions when unmounting a file system */
27608 +static void aufs_put_super(struct super_block *sb)
27609 +{
27610 +       struct au_sbinfo *sbinfo;
27611 +
27612 +       sbinfo = au_sbi(sb);
27613 +       if (!sbinfo)
27614 +               return;
27615 +
27616 +       dbgaufs_si_fin(sbinfo);
27617 +       kobject_put(&sbinfo->si_kobj);
27618 +}
27619 +
27620 +/* ---------------------------------------------------------------------- */
27621 +
27622 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb,
27623 +                    struct super_block *sb, void *arg)
27624 +{
27625 +       void *array;
27626 +       unsigned long long n, sz;
27627 +
27628 +       array = NULL;
27629 +       n = 0;
27630 +       if (!*hint)
27631 +               goto out;
27632 +
27633 +       if (*hint > ULLONG_MAX / sizeof(array)) {
27634 +               array = ERR_PTR(-EMFILE);
27635 +               pr_err("hint %llu\n", *hint);
27636 +               goto out;
27637 +       }
27638 +
27639 +       sz = sizeof(array) * *hint;
27640 +       array = kzalloc(sz, GFP_NOFS);
27641 +       if (unlikely(!array))
27642 +               array = vzalloc(sz);
27643 +       if (unlikely(!array)) {
27644 +               array = ERR_PTR(-ENOMEM);
27645 +               goto out;
27646 +       }
27647 +
27648 +       n = cb(sb, array, *hint, arg);
27649 +       AuDebugOn(n > *hint);
27650 +
27651 +out:
27652 +       *hint = n;
27653 +       return array;
27654 +}
27655 +
27656 +static unsigned long long au_iarray_cb(struct super_block *sb, void *a,
27657 +                                      unsigned long long max __maybe_unused,
27658 +                                      void *arg)
27659 +{
27660 +       unsigned long long n;
27661 +       struct inode **p, *inode;
27662 +       struct list_head *head;
27663 +
27664 +       n = 0;
27665 +       p = a;
27666 +       head = arg;
27667 +       spin_lock(&sb->s_inode_list_lock);
27668 +       list_for_each_entry(inode, head, i_sb_list) {
27669 +               if (!au_is_bad_inode(inode)
27670 +                   && au_ii(inode)->ii_btop >= 0) {
27671 +                       spin_lock(&inode->i_lock);
27672 +                       if (atomic_read(&inode->i_count)) {
27673 +                               au_igrab(inode);
27674 +                               *p++ = inode;
27675 +                               n++;
27676 +                               AuDebugOn(n > max);
27677 +                       }
27678 +                       spin_unlock(&inode->i_lock);
27679 +               }
27680 +       }
27681 +       spin_unlock(&sb->s_inode_list_lock);
27682 +
27683 +       return n;
27684 +}
27685 +
27686 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max)
27687 +{
27688 +       *max = au_ninodes(sb);
27689 +       return au_array_alloc(max, au_iarray_cb, sb, &sb->s_inodes);
27690 +}
27691 +
27692 +void au_iarray_free(struct inode **a, unsigned long long max)
27693 +{
27694 +       unsigned long long ull;
27695 +
27696 +       for (ull = 0; ull < max; ull++)
27697 +               iput(a[ull]);
27698 +       kvfree(a);
27699 +}
27700 +
27701 +/* ---------------------------------------------------------------------- */
27702 +
27703 +/*
27704 + * refresh dentry and inode at remount time.
27705 + */
27706 +/* todo: consolidate with simple_reval_dpath() and au_reval_for_attr() */
27707 +static int au_do_refresh(struct dentry *dentry, unsigned int dir_flags,
27708 +                     struct dentry *parent)
27709 +{
27710 +       int err;
27711 +
27712 +       di_write_lock_child(dentry);
27713 +       di_read_lock_parent(parent, AuLock_IR);
27714 +       err = au_refresh_dentry(dentry, parent);
27715 +       if (!err && dir_flags)
27716 +               au_hn_reset(d_inode(dentry), dir_flags);
27717 +       di_read_unlock(parent, AuLock_IR);
27718 +       di_write_unlock(dentry);
27719 +
27720 +       return err;
27721 +}
27722 +
27723 +static int au_do_refresh_d(struct dentry *dentry, unsigned int sigen,
27724 +                          struct au_sbinfo *sbinfo,
27725 +                          const unsigned int dir_flags, unsigned int do_idop)
27726 +{
27727 +       int err;
27728 +       struct dentry *parent;
27729 +
27730 +       err = 0;
27731 +       parent = dget_parent(dentry);
27732 +       if (!au_digen_test(parent, sigen) && au_digen_test(dentry, sigen)) {
27733 +               if (d_really_is_positive(dentry)) {
27734 +                       if (!d_is_dir(dentry))
27735 +                               err = au_do_refresh(dentry, /*dir_flags*/0,
27736 +                                                parent);
27737 +                       else {
27738 +                               err = au_do_refresh(dentry, dir_flags, parent);
27739 +                               if (unlikely(err))
27740 +                                       au_fset_si(sbinfo, FAILED_REFRESH_DIR);
27741 +                       }
27742 +               } else
27743 +                       err = au_do_refresh(dentry, /*dir_flags*/0, parent);
27744 +               AuDbgDentry(dentry);
27745 +       }
27746 +       dput(parent);
27747 +
27748 +       if (!err) {
27749 +               if (do_idop)
27750 +                       au_refresh_dop(dentry, /*force_reval*/0);
27751 +       } else
27752 +               au_refresh_dop(dentry, /*force_reval*/1);
27753 +
27754 +       AuTraceErr(err);
27755 +       return err;
27756 +}
27757 +
27758 +static int au_refresh_d(struct super_block *sb, unsigned int do_idop)
27759 +{
27760 +       int err, i, j, ndentry, e;
27761 +       unsigned int sigen;
27762 +       struct au_dcsub_pages dpages;
27763 +       struct au_dpage *dpage;
27764 +       struct dentry **dentries, *d;
27765 +       struct au_sbinfo *sbinfo;
27766 +       struct dentry *root = sb->s_root;
27767 +       const unsigned int dir_flags = au_hi_flags(d_inode(root), /*isdir*/1);
27768 +
27769 +       if (do_idop)
27770 +               au_refresh_dop(root, /*force_reval*/0);
27771 +
27772 +       err = au_dpages_init(&dpages, GFP_NOFS);
27773 +       if (unlikely(err))
27774 +               goto out;
27775 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
27776 +       if (unlikely(err))
27777 +               goto out_dpages;
27778 +
27779 +       sigen = au_sigen(sb);
27780 +       sbinfo = au_sbi(sb);
27781 +       for (i = 0; i < dpages.ndpage; i++) {
27782 +               dpage = dpages.dpages + i;
27783 +               dentries = dpage->dentries;
27784 +               ndentry = dpage->ndentry;
27785 +               for (j = 0; j < ndentry; j++) {
27786 +                       d = dentries[j];
27787 +                       e = au_do_refresh_d(d, sigen, sbinfo, dir_flags,
27788 +                                           do_idop);
27789 +                       if (unlikely(e && !err))
27790 +                               err = e;
27791 +                       /* go on even err */
27792 +               }
27793 +       }
27794 +
27795 +out_dpages:
27796 +       au_dpages_free(&dpages);
27797 +out:
27798 +       return err;
27799 +}
27800 +
27801 +static int au_refresh_i(struct super_block *sb, unsigned int do_idop)
27802 +{
27803 +       int err, e;
27804 +       unsigned int sigen;
27805 +       unsigned long long max, ull;
27806 +       struct inode *inode, **array;
27807 +
27808 +       array = au_iarray_alloc(sb, &max);
27809 +       err = PTR_ERR(array);
27810 +       if (IS_ERR(array))
27811 +               goto out;
27812 +
27813 +       err = 0;
27814 +       sigen = au_sigen(sb);
27815 +       for (ull = 0; ull < max; ull++) {
27816 +               inode = array[ull];
27817 +               if (unlikely(!inode))
27818 +                       break;
27819 +
27820 +               e = 0;
27821 +               ii_write_lock_child(inode);
27822 +               if (au_iigen(inode, NULL) != sigen) {
27823 +                       e = au_refresh_hinode_self(inode);
27824 +                       if (unlikely(e)) {
27825 +                               au_refresh_iop(inode, /*force_getattr*/1);
27826 +                               pr_err("error %d, i%lu\n", e, inode->i_ino);
27827 +                               if (!err)
27828 +                                       err = e;
27829 +                               /* go on even if err */
27830 +                       }
27831 +               }
27832 +               if (!e && do_idop)
27833 +                       au_refresh_iop(inode, /*force_getattr*/0);
27834 +               ii_write_unlock(inode);
27835 +       }
27836 +
27837 +       au_iarray_free(array, max);
27838 +
27839 +out:
27840 +       return err;
27841 +}
27842 +
27843 +static void au_remount_refresh(struct super_block *sb, unsigned int do_idop)
27844 +{
27845 +       int err, e;
27846 +       unsigned int udba;
27847 +       aufs_bindex_t bindex, bbot;
27848 +       struct dentry *root;
27849 +       struct inode *inode;
27850 +       struct au_branch *br;
27851 +       struct au_sbinfo *sbi;
27852 +
27853 +       au_sigen_inc(sb);
27854 +       sbi = au_sbi(sb);
27855 +       au_fclr_si(sbi, FAILED_REFRESH_DIR);
27856 +
27857 +       root = sb->s_root;
27858 +       DiMustNoWaiters(root);
27859 +       inode = d_inode(root);
27860 +       IiMustNoWaiters(inode);
27861 +
27862 +       udba = au_opt_udba(sb);
27863 +       bbot = au_sbbot(sb);
27864 +       for (bindex = 0; bindex <= bbot; bindex++) {
27865 +               br = au_sbr(sb, bindex);
27866 +               err = au_hnotify_reset_br(udba, br, br->br_perm);
27867 +               if (unlikely(err))
27868 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
27869 +                               bindex, err);
27870 +               /* go on even if err */
27871 +       }
27872 +       au_hn_reset(inode, au_hi_flags(inode, /*isdir*/1));
27873 +
27874 +       if (do_idop) {
27875 +               if (au_ftest_si(sbi, NO_DREVAL)) {
27876 +                       AuDebugOn(sb->s_d_op == &aufs_dop_noreval);
27877 +                       sb->s_d_op = &aufs_dop_noreval;
27878 +                       AuDebugOn(sbi->si_iop_array == aufs_iop_nogetattr);
27879 +                       sbi->si_iop_array = aufs_iop_nogetattr;
27880 +               } else {
27881 +                       AuDebugOn(sb->s_d_op == &aufs_dop);
27882 +                       sb->s_d_op = &aufs_dop;
27883 +                       AuDebugOn(sbi->si_iop_array == aufs_iop);
27884 +                       sbi->si_iop_array = aufs_iop;
27885 +               }
27886 +               pr_info("reset to %pf and %pf\n",
27887 +                       sb->s_d_op, sbi->si_iop_array);
27888 +       }
27889 +
27890 +       di_write_unlock(root);
27891 +       err = au_refresh_d(sb, do_idop);
27892 +       e = au_refresh_i(sb, do_idop);
27893 +       if (unlikely(e && !err))
27894 +               err = e;
27895 +       /* aufs_write_lock() calls ..._child() */
27896 +       di_write_lock_child(root);
27897 +
27898 +       au_cpup_attr_all(inode, /*force*/1);
27899 +
27900 +       if (unlikely(err))
27901 +               AuIOErr("refresh failed, ignored, %d\n", err);
27902 +}
27903 +
27904 +/* stop extra interpretation of errno in mount(8), and strange error messages */
27905 +static int cvt_err(int err)
27906 +{
27907 +       AuTraceErr(err);
27908 +
27909 +       switch (err) {
27910 +       case -ENOENT:
27911 +       case -ENOTDIR:
27912 +       case -EEXIST:
27913 +       case -EIO:
27914 +               err = -EINVAL;
27915 +       }
27916 +       return err;
27917 +}
27918 +
27919 +static int aufs_remount_fs(struct super_block *sb, int *flags, char *data)
27920 +{
27921 +       int err, do_dx;
27922 +       unsigned int mntflags;
27923 +       struct au_opts opts = {
27924 +               .opt = NULL
27925 +       };
27926 +       struct dentry *root;
27927 +       struct inode *inode;
27928 +       struct au_sbinfo *sbinfo;
27929 +
27930 +       err = 0;
27931 +       root = sb->s_root;
27932 +       if (!data || !*data) {
27933 +               err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
27934 +               if (!err) {
27935 +                       di_write_lock_child(root);
27936 +                       err = au_opts_verify(sb, *flags, /*pending*/0);
27937 +                       aufs_write_unlock(root);
27938 +               }
27939 +               goto out;
27940 +       }
27941 +
27942 +       err = -ENOMEM;
27943 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
27944 +       if (unlikely(!opts.opt))
27945 +               goto out;
27946 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
27947 +       opts.flags = AuOpts_REMOUNT;
27948 +       opts.sb_flags = *flags;
27949 +
27950 +       /* parse it before aufs lock */
27951 +       err = au_opts_parse(sb, data, &opts);
27952 +       if (unlikely(err))
27953 +               goto out_opts;
27954 +
27955 +       sbinfo = au_sbi(sb);
27956 +       inode = d_inode(root);
27957 +       inode_lock(inode);
27958 +       err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
27959 +       if (unlikely(err))
27960 +               goto out_mtx;
27961 +       di_write_lock_child(root);
27962 +
27963 +       /* au_opts_remount() may return an error */
27964 +       err = au_opts_remount(sb, &opts);
27965 +       au_opts_free(&opts);
27966 +
27967 +       if (au_ftest_opts(opts.flags, REFRESH))
27968 +               au_remount_refresh(sb, au_ftest_opts(opts.flags, REFRESH_IDOP));
27969 +
27970 +       if (au_ftest_opts(opts.flags, REFRESH_DYAOP)) {
27971 +               mntflags = au_mntflags(sb);
27972 +               do_dx = !!au_opt_test(mntflags, DIO);
27973 +               au_dy_arefresh(do_dx);
27974 +       }
27975 +
27976 +       au_fhsm_wrote_all(sb, /*force*/1); /* ?? */
27977 +       aufs_write_unlock(root);
27978 +
27979 +out_mtx:
27980 +       inode_unlock(inode);
27981 +out_opts:
27982 +       au_delayed_free_page((unsigned long)opts.opt);
27983 +out:
27984 +       err = cvt_err(err);
27985 +       AuTraceErr(err);
27986 +       return err;
27987 +}
27988 +
27989 +static const struct super_operations aufs_sop = {
27990 +       .alloc_inode    = aufs_alloc_inode,
27991 +       .destroy_inode  = aufs_destroy_inode,
27992 +       /* always deleting, no clearing */
27993 +       .drop_inode     = generic_delete_inode,
27994 +       .show_options   = aufs_show_options,
27995 +       .statfs         = aufs_statfs,
27996 +       .put_super      = aufs_put_super,
27997 +       .sync_fs        = aufs_sync_fs,
27998 +       .remount_fs     = aufs_remount_fs
27999 +};
28000 +
28001 +/* ---------------------------------------------------------------------- */
28002 +
28003 +static int alloc_root(struct super_block *sb)
28004 +{
28005 +       int err;
28006 +       struct inode *inode;
28007 +       struct dentry *root;
28008 +
28009 +       err = -ENOMEM;
28010 +       inode = au_iget_locked(sb, AUFS_ROOT_INO);
28011 +       err = PTR_ERR(inode);
28012 +       if (IS_ERR(inode))
28013 +               goto out;
28014 +
28015 +       inode->i_op = aufs_iop + AuIop_DIR; /* with getattr by default */
28016 +       inode->i_fop = &aufs_dir_fop;
28017 +       inode->i_mode = S_IFDIR;
28018 +       set_nlink(inode, 2);
28019 +       unlock_new_inode(inode);
28020 +
28021 +       root = d_make_root(inode);
28022 +       if (unlikely(!root))
28023 +               goto out;
28024 +       err = PTR_ERR(root);
28025 +       if (IS_ERR(root))
28026 +               goto out;
28027 +
28028 +       err = au_di_init(root);
28029 +       if (!err) {
28030 +               sb->s_root = root;
28031 +               return 0; /* success */
28032 +       }
28033 +       dput(root);
28034 +
28035 +out:
28036 +       return err;
28037 +}
28038 +
28039 +static int aufs_fill_super(struct super_block *sb, void *raw_data,
28040 +                          int silent __maybe_unused)
28041 +{
28042 +       int err;
28043 +       struct au_opts opts = {
28044 +               .opt = NULL
28045 +       };
28046 +       struct au_sbinfo *sbinfo;
28047 +       struct dentry *root;
28048 +       struct inode *inode;
28049 +       char *arg = raw_data;
28050 +
28051 +       if (unlikely(!arg || !*arg)) {
28052 +               err = -EINVAL;
28053 +               pr_err("no arg\n");
28054 +               goto out;
28055 +       }
28056 +
28057 +       err = -ENOMEM;
28058 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
28059 +       if (unlikely(!opts.opt))
28060 +               goto out;
28061 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
28062 +       opts.sb_flags = sb->s_flags;
28063 +
28064 +       err = au_si_alloc(sb);
28065 +       if (unlikely(err))
28066 +               goto out_opts;
28067 +       sbinfo = au_sbi(sb);
28068 +
28069 +       /* all timestamps always follow the ones on the branch */
28070 +       sb->s_flags |= MS_NOATIME | MS_NODIRATIME;
28071 +       sb->s_op = &aufs_sop;
28072 +       sb->s_d_op = &aufs_dop;
28073 +       sb->s_magic = AUFS_SUPER_MAGIC;
28074 +       sb->s_maxbytes = 0;
28075 +       sb->s_stack_depth = 1;
28076 +       au_export_init(sb);
28077 +       /* au_xattr_init(sb); */
28078 +
28079 +       err = alloc_root(sb);
28080 +       if (unlikely(err)) {
28081 +               si_write_unlock(sb);
28082 +               goto out_info;
28083 +       }
28084 +       root = sb->s_root;
28085 +       inode = d_inode(root);
28086 +
28087 +       /*
28088 +        * actually we can parse options regardless aufs lock here.
28089 +        * but at remount time, parsing must be done before aufs lock.
28090 +        * so we follow the same rule.
28091 +        */
28092 +       ii_write_lock_parent(inode);
28093 +       aufs_write_unlock(root);
28094 +       err = au_opts_parse(sb, arg, &opts);
28095 +       if (unlikely(err))
28096 +               goto out_root;
28097 +
28098 +       /* lock vfs_inode first, then aufs. */
28099 +       inode_lock(inode);
28100 +       aufs_write_lock(root);
28101 +       err = au_opts_mount(sb, &opts);
28102 +       au_opts_free(&opts);
28103 +       if (!err && au_ftest_si(sbinfo, NO_DREVAL)) {
28104 +               sb->s_d_op = &aufs_dop_noreval;
28105 +               pr_info("%pf\n", sb->s_d_op);
28106 +               au_refresh_dop(root, /*force_reval*/0);
28107 +               sbinfo->si_iop_array = aufs_iop_nogetattr;
28108 +               au_refresh_iop(inode, /*force_getattr*/0);
28109 +       }
28110 +       aufs_write_unlock(root);
28111 +       inode_unlock(inode);
28112 +       if (!err)
28113 +               goto out_opts; /* success */
28114 +
28115 +out_root:
28116 +       dput(root);
28117 +       sb->s_root = NULL;
28118 +out_info:
28119 +       dbgaufs_si_fin(sbinfo);
28120 +       kobject_put(&sbinfo->si_kobj);
28121 +       sb->s_fs_info = NULL;
28122 +out_opts:
28123 +       au_delayed_free_page((unsigned long)opts.opt);
28124 +out:
28125 +       AuTraceErr(err);
28126 +       err = cvt_err(err);
28127 +       AuTraceErr(err);
28128 +       return err;
28129 +}
28130 +
28131 +/* ---------------------------------------------------------------------- */
28132 +
28133 +static struct dentry *aufs_mount(struct file_system_type *fs_type, int flags,
28134 +                                const char *dev_name __maybe_unused,
28135 +                                void *raw_data)
28136 +{
28137 +       struct dentry *root;
28138 +       struct super_block *sb;
28139 +
28140 +       /* all timestamps always follow the ones on the branch */
28141 +       /* mnt->mnt_flags |= MNT_NOATIME | MNT_NODIRATIME; */
28142 +       root = mount_nodev(fs_type, flags, raw_data, aufs_fill_super);
28143 +       if (IS_ERR(root))
28144 +               goto out;
28145 +
28146 +       sb = root->d_sb;
28147 +       si_write_lock(sb, !AuLock_FLUSH);
28148 +       sysaufs_brs_add(sb, 0);
28149 +       si_write_unlock(sb);
28150 +       au_sbilist_add(sb);
28151 +
28152 +out:
28153 +       return root;
28154 +}
28155 +
28156 +static void aufs_kill_sb(struct super_block *sb)
28157 +{
28158 +       struct au_sbinfo *sbinfo;
28159 +
28160 +       sbinfo = au_sbi(sb);
28161 +       if (sbinfo) {
28162 +               au_sbilist_del(sb);
28163 +               aufs_write_lock(sb->s_root);
28164 +               au_fhsm_fin(sb);
28165 +               if (sbinfo->si_wbr_create_ops->fin)
28166 +                       sbinfo->si_wbr_create_ops->fin(sb);
28167 +               if (au_opt_test(sbinfo->si_mntflags, UDBA_HNOTIFY)) {
28168 +                       au_opt_set_udba(sbinfo->si_mntflags, UDBA_NONE);
28169 +                       au_remount_refresh(sb, /*do_idop*/0);
28170 +               }
28171 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
28172 +                       au_plink_put(sb, /*verbose*/1);
28173 +               au_xino_clr(sb);
28174 +               sbinfo->si_sb = NULL;
28175 +               aufs_write_unlock(sb->s_root);
28176 +               au_nwt_flush(&sbinfo->si_nowait);
28177 +       }
28178 +       kill_anon_super(sb);
28179 +}
28180 +
28181 +struct file_system_type aufs_fs_type = {
28182 +       .name           = AUFS_FSTYPE,
28183 +       /* a race between rename and others */
28184 +       .fs_flags       = FS_RENAME_DOES_D_MOVE,
28185 +       .mount          = aufs_mount,
28186 +       .kill_sb        = aufs_kill_sb,
28187 +       /* no need to __module_get() and module_put(). */
28188 +       .owner          = THIS_MODULE,
28189 +};
28190 diff -urN /usr/share/empty/fs/aufs/super.h linux/fs/aufs/super.h
28191 --- /usr/share/empty/fs/aufs/super.h    1970-01-01 01:00:00.000000000 +0100
28192 +++ linux/fs/aufs/super.h       2016-10-09 16:55:38.889431135 +0200
28193 @@ -0,0 +1,638 @@
28194 +/*
28195 + * Copyright (C) 2005-2016 Junjiro R. Okajima
28196 + *
28197 + * This program, aufs is free software; you can redistribute it and/or modify
28198 + * it under the terms of the GNU General Public License as published by
28199 + * the Free Software Foundation; either version 2 of the License, or
28200 + * (at your option) any later version.
28201 + *
28202 + * This program is distributed in the hope that it will be useful,
28203 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28204 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28205 + * GNU General Public License for more details.
28206 + *
28207 + * You should have received a copy of the GNU General Public License
28208 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28209 + */
28210 +
28211 +/*
28212 + * super_block operations
28213 + */
28214 +
28215 +#ifndef __AUFS_SUPER_H__
28216 +#define __AUFS_SUPER_H__
28217 +
28218 +#ifdef __KERNEL__
28219 +
28220 +#include <linux/fs.h>
28221 +#include <linux/kobject.h>
28222 +#include "rwsem.h"
28223 +#include "spl.h"
28224 +#include "wkq.h"
28225 +
28226 +/* policies to select one among multiple writable branches */
28227 +struct au_wbr_copyup_operations {
28228 +       int (*copyup)(struct dentry *dentry);
28229 +};
28230 +
28231 +#define AuWbr_DIR      1               /* target is a dir */
28232 +#define AuWbr_PARENT   (1 << 1)        /* always require a parent */
28233 +
28234 +#define au_ftest_wbr(flags, name)      ((flags) & AuWbr_##name)
28235 +#define au_fset_wbr(flags, name)       { (flags) |= AuWbr_##name; }
28236 +#define au_fclr_wbr(flags, name)       { (flags) &= ~AuWbr_##name; }
28237 +
28238 +struct au_wbr_create_operations {
28239 +       int (*create)(struct dentry *dentry, unsigned int flags);
28240 +       int (*init)(struct super_block *sb);
28241 +       int (*fin)(struct super_block *sb);
28242 +};
28243 +
28244 +struct au_wbr_mfs {
28245 +       struct mutex    mfs_lock; /* protect this structure */
28246 +       unsigned long   mfs_jiffy;
28247 +       unsigned long   mfs_expire;
28248 +       aufs_bindex_t   mfs_bindex;
28249 +
28250 +       unsigned long long      mfsrr_bytes;
28251 +       unsigned long long      mfsrr_watermark;
28252 +};
28253 +
28254 +#define AuPlink_NHASH 100
28255 +static inline int au_plink_hash(ino_t ino)
28256 +{
28257 +       return ino % AuPlink_NHASH;
28258 +}
28259 +
28260 +/* File-based Hierarchical Storage Management */
28261 +struct au_fhsm {
28262 +#ifdef CONFIG_AUFS_FHSM
28263 +       /* allow only one process who can receive the notification */
28264 +       spinlock_t              fhsm_spin;
28265 +       pid_t                   fhsm_pid;
28266 +       wait_queue_head_t       fhsm_wqh;
28267 +       atomic_t                fhsm_readable;
28268 +
28269 +       /* these are protected by si_rwsem */
28270 +       unsigned long           fhsm_expire;
28271 +       aufs_bindex_t           fhsm_bottom;
28272 +#endif
28273 +};
28274 +
28275 +#define AU_PIDSTEP     (int)(BITS_TO_LONGS(PID_MAX_DEFAULT) * BITS_PER_LONG)
28276 +#define AU_NPIDMAP     (int)DIV_ROUND_UP(PID_MAX_LIMIT, AU_PIDSTEP)
28277 +struct au_si_pid {
28278 +       unsigned long   *pid_bitmap[AU_NPIDMAP];
28279 +       struct mutex    pid_mtx;
28280 +};
28281 +
28282 +struct au_branch;
28283 +struct au_sbinfo {
28284 +       /* nowait tasks in the system-wide workqueue */
28285 +       struct au_nowait_tasks  si_nowait;
28286 +
28287 +       /*
28288 +        * tried sb->s_umount, but failed due to the dependecy between i_mutex.
28289 +        * rwsem for au_sbinfo is necessary.
28290 +        */
28291 +       struct au_rwsem         si_rwsem;
28292 +
28293 +       /* prevent recursive locking in deleting inode */
28294 +       struct au_si_pid        au_si_pid;
28295 +
28296 +       /*
28297 +        * dirty approach to protect sb->sb_inodes and ->s_files (gone) from
28298 +        * remount.
28299 +        */
28300 +       struct percpu_counter   si_ninodes, si_nfiles;
28301 +
28302 +       /* branch management */
28303 +       unsigned int            si_generation;
28304 +
28305 +       /* see AuSi_ flags */
28306 +       unsigned char           au_si_status;
28307 +
28308 +       aufs_bindex_t           si_bbot;
28309 +
28310 +       /* dirty trick to keep br_id plus */
28311 +       unsigned int            si_last_br_id :
28312 +                               sizeof(aufs_bindex_t) * BITS_PER_BYTE - 1;
28313 +       struct au_branch        **si_branch;
28314 +
28315 +       /* policy to select a writable branch */
28316 +       unsigned char           si_wbr_copyup;
28317 +       unsigned char           si_wbr_create;
28318 +       struct au_wbr_copyup_operations *si_wbr_copyup_ops;
28319 +       struct au_wbr_create_operations *si_wbr_create_ops;
28320 +
28321 +       /* round robin */
28322 +       atomic_t                si_wbr_rr_next;
28323 +
28324 +       /* most free space */
28325 +       struct au_wbr_mfs       si_wbr_mfs;
28326 +
28327 +       /* File-based Hierarchical Storage Management */
28328 +       struct au_fhsm          si_fhsm;
28329 +
28330 +       /* mount flags */
28331 +       /* include/asm-ia64/siginfo.h defines a macro named si_flags */
28332 +       unsigned int            si_mntflags;
28333 +
28334 +       /* external inode number (bitmap and translation table) */
28335 +       vfs_readf_t             si_xread;
28336 +       vfs_writef_t            si_xwrite;
28337 +       struct file             *si_xib;
28338 +       struct mutex            si_xib_mtx; /* protect xib members */
28339 +       unsigned long           *si_xib_buf;
28340 +       unsigned long           si_xib_last_pindex;
28341 +       int                     si_xib_next_bit;
28342 +       aufs_bindex_t           si_xino_brid;
28343 +       unsigned long           si_xino_jiffy;
28344 +       unsigned long           si_xino_expire;
28345 +       /* reserved for future use */
28346 +       /* unsigned long long   si_xib_limit; */        /* Max xib file size */
28347 +
28348 +#ifdef CONFIG_AUFS_EXPORT
28349 +       /* i_generation */
28350 +       struct file             *si_xigen;
28351 +       atomic_t                si_xigen_next;
28352 +#endif
28353 +
28354 +       /* dirty trick to suppoer atomic_open */
28355 +       struct au_sphlhead      si_aopen;
28356 +
28357 +       /* vdir parameters */
28358 +       unsigned long           si_rdcache;     /* max cache time in jiffies */
28359 +       unsigned int            si_rdblk;       /* deblk size */
28360 +       unsigned int            si_rdhash;      /* hash size */
28361 +
28362 +       /*
28363 +        * If the number of whiteouts are larger than si_dirwh, leave all of
28364 +        * them after au_whtmp_ren to reduce the cost of rmdir(2).
28365 +        * future fsck.aufs or kernel thread will remove them later.
28366 +        * Otherwise, remove all whiteouts and the dir in rmdir(2).
28367 +        */
28368 +       unsigned int            si_dirwh;
28369 +
28370 +       /* pseudo_link list */
28371 +       struct au_sphlhead      si_plink[AuPlink_NHASH];
28372 +       wait_queue_head_t       si_plink_wq;
28373 +       spinlock_t              si_plink_maint_lock;
28374 +       pid_t                   si_plink_maint_pid;
28375 +
28376 +       /* file list */
28377 +       struct au_sphlhead      si_files;
28378 +
28379 +       /* with/without getattr, brother of sb->s_d_op */
28380 +       struct inode_operations *si_iop_array;
28381 +
28382 +       /*
28383 +        * sysfs and lifetime management.
28384 +        * this is not a small structure and it may be a waste of memory in case
28385 +        * of sysfs is disabled, particulary when many aufs-es are mounted.
28386 +        * but using sysfs is majority.
28387 +        */
28388 +       struct kobject          si_kobj;
28389 +#ifdef CONFIG_DEBUG_FS
28390 +       struct dentry            *si_dbgaufs;
28391 +       struct dentry            *si_dbgaufs_plink;
28392 +       struct dentry            *si_dbgaufs_xib;
28393 +#ifdef CONFIG_AUFS_EXPORT
28394 +       struct dentry            *si_dbgaufs_xigen;
28395 +#endif
28396 +#endif
28397 +
28398 +#ifdef CONFIG_AUFS_SBILIST
28399 +       struct hlist_node       si_list;
28400 +#endif
28401 +
28402 +       /* dirty, necessary for unmounting, sysfs and sysrq */
28403 +       struct super_block      *si_sb;
28404 +};
28405 +
28406 +/* sbinfo status flags */
28407 +/*
28408 + * set true when refresh_dirs() failed at remount time.
28409 + * then try refreshing dirs at access time again.
28410 + * if it is false, refreshing dirs at access time is unnecesary
28411 + */
28412 +#define AuSi_FAILED_REFRESH_DIR        1
28413 +#define AuSi_FHSM              (1 << 1)        /* fhsm is active now */
28414 +#define AuSi_NO_DREVAL         (1 << 2)        /* disable all d_revalidate */
28415 +
28416 +#ifndef CONFIG_AUFS_FHSM
28417 +#undef AuSi_FHSM
28418 +#define AuSi_FHSM              0
28419 +#endif
28420 +
28421 +static inline unsigned char au_do_ftest_si(struct au_sbinfo *sbi,
28422 +                                          unsigned int flag)
28423 +{
28424 +       AuRwMustAnyLock(&sbi->si_rwsem);
28425 +       return sbi->au_si_status & flag;
28426 +}
28427 +#define au_ftest_si(sbinfo, name)      au_do_ftest_si(sbinfo, AuSi_##name)
28428 +#define au_fset_si(sbinfo, name) do { \
28429 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
28430 +       (sbinfo)->au_si_status |= AuSi_##name; \
28431 +} while (0)
28432 +#define au_fclr_si(sbinfo, name) do { \
28433 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
28434 +       (sbinfo)->au_si_status &= ~AuSi_##name; \
28435 +} while (0)
28436 +
28437 +/* ---------------------------------------------------------------------- */
28438 +
28439 +/* policy to select one among writable branches */
28440 +#define AuWbrCopyup(sbinfo, ...) \
28441 +       ((sbinfo)->si_wbr_copyup_ops->copyup(__VA_ARGS__))
28442 +#define AuWbrCreate(sbinfo, ...) \
28443 +       ((sbinfo)->si_wbr_create_ops->create(__VA_ARGS__))
28444 +
28445 +/* flags for si_read_lock()/aufs_read_lock()/di_read_lock() */
28446 +#define AuLock_DW              1               /* write-lock dentry */
28447 +#define AuLock_IR              (1 << 1)        /* read-lock inode */
28448 +#define AuLock_IW              (1 << 2)        /* write-lock inode */
28449 +#define AuLock_FLUSH           (1 << 3)        /* wait for 'nowait' tasks */
28450 +#define AuLock_DIRS            (1 << 4)        /* target is a pair of dirs */
28451 +#define AuLock_NOPLM           (1 << 5)        /* return err in plm mode */
28452 +#define AuLock_NOPLMW          (1 << 6)        /* wait for plm mode ends */
28453 +#define AuLock_GEN             (1 << 7)        /* test digen/iigen */
28454 +#define au_ftest_lock(flags, name)     ((flags) & AuLock_##name)
28455 +#define au_fset_lock(flags, name) \
28456 +       do { (flags) |= AuLock_##name; } while (0)
28457 +#define au_fclr_lock(flags, name) \
28458 +       do { (flags) &= ~AuLock_##name; } while (0)
28459 +
28460 +/* ---------------------------------------------------------------------- */
28461 +
28462 +/* super.c */
28463 +extern struct file_system_type aufs_fs_type;
28464 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino);
28465 +typedef unsigned long long (*au_arraycb_t)(struct super_block *sb, void *array,
28466 +                                          unsigned long long max, void *arg);
28467 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb,
28468 +                    struct super_block *sb, void *arg);
28469 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max);
28470 +void au_iarray_free(struct inode **a, unsigned long long max);
28471 +
28472 +/* sbinfo.c */
28473 +void au_si_free(struct kobject *kobj);
28474 +int au_si_alloc(struct super_block *sb);
28475 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr, int may_shrink);
28476 +
28477 +unsigned int au_sigen_inc(struct super_block *sb);
28478 +aufs_bindex_t au_new_br_id(struct super_block *sb);
28479 +
28480 +int si_read_lock(struct super_block *sb, int flags);
28481 +int si_write_lock(struct super_block *sb, int flags);
28482 +int aufs_read_lock(struct dentry *dentry, int flags);
28483 +void aufs_read_unlock(struct dentry *dentry, int flags);
28484 +void aufs_write_lock(struct dentry *dentry);
28485 +void aufs_write_unlock(struct dentry *dentry);
28486 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags);
28487 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2);
28488 +
28489 +/* wbr_policy.c */
28490 +extern struct au_wbr_copyup_operations au_wbr_copyup_ops[];
28491 +extern struct au_wbr_create_operations au_wbr_create_ops[];
28492 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst);
28493 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex);
28494 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t btop);
28495 +
28496 +/* mvdown.c */
28497 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *arg);
28498 +
28499 +#ifdef CONFIG_AUFS_FHSM
28500 +/* fhsm.c */
28501 +
28502 +static inline pid_t au_fhsm_pid(struct au_fhsm *fhsm)
28503 +{
28504 +       pid_t pid;
28505 +
28506 +       spin_lock(&fhsm->fhsm_spin);
28507 +       pid = fhsm->fhsm_pid;
28508 +       spin_unlock(&fhsm->fhsm_spin);
28509 +
28510 +       return pid;
28511 +}
28512 +
28513 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force);
28514 +void au_fhsm_wrote_all(struct super_block *sb, int force);
28515 +int au_fhsm_fd(struct super_block *sb, int oflags);
28516 +int au_fhsm_br_alloc(struct au_branch *br);
28517 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex);
28518 +void au_fhsm_fin(struct super_block *sb);
28519 +void au_fhsm_init(struct au_sbinfo *sbinfo);
28520 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec);
28521 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo);
28522 +#else
28523 +AuStubVoid(au_fhsm_wrote, struct super_block *sb, aufs_bindex_t bindex,
28524 +          int force)
28525 +AuStubVoid(au_fhsm_wrote_all, struct super_block *sb, int force)
28526 +AuStub(int, au_fhsm_fd, return -EOPNOTSUPP, struct super_block *sb, int oflags)
28527 +AuStub(pid_t, au_fhsm_pid, return 0, struct au_fhsm *fhsm)
28528 +AuStubInt0(au_fhsm_br_alloc, struct au_branch *br)
28529 +AuStubVoid(au_fhsm_set_bottom, struct super_block *sb, aufs_bindex_t bindex)
28530 +AuStubVoid(au_fhsm_fin, struct super_block *sb)
28531 +AuStubVoid(au_fhsm_init, struct au_sbinfo *sbinfo)
28532 +AuStubVoid(au_fhsm_set, struct au_sbinfo *sbinfo, unsigned int sec)
28533 +AuStubVoid(au_fhsm_show, struct seq_file *seq, struct au_sbinfo *sbinfo)
28534 +#endif
28535 +
28536 +/* ---------------------------------------------------------------------- */
28537 +
28538 +static inline struct au_sbinfo *au_sbi(struct super_block *sb)
28539 +{
28540 +       return sb->s_fs_info;
28541 +}
28542 +
28543 +/* ---------------------------------------------------------------------- */
28544 +
28545 +#ifdef CONFIG_AUFS_EXPORT
28546 +int au_test_nfsd(void);
28547 +void au_export_init(struct super_block *sb);
28548 +void au_xigen_inc(struct inode *inode);
28549 +int au_xigen_new(struct inode *inode);
28550 +int au_xigen_set(struct super_block *sb, struct file *base);
28551 +void au_xigen_clr(struct super_block *sb);
28552 +
28553 +static inline int au_busy_or_stale(void)
28554 +{
28555 +       if (!au_test_nfsd())
28556 +               return -EBUSY;
28557 +       return -ESTALE;
28558 +}
28559 +#else
28560 +AuStubInt0(au_test_nfsd, void)
28561 +AuStubVoid(au_export_init, struct super_block *sb)
28562 +AuStubVoid(au_xigen_inc, struct inode *inode)
28563 +AuStubInt0(au_xigen_new, struct inode *inode)
28564 +AuStubInt0(au_xigen_set, struct super_block *sb, struct file *base)
28565 +AuStubVoid(au_xigen_clr, struct super_block *sb)
28566 +AuStub(int, au_busy_or_stale, return -EBUSY, void)
28567 +#endif /* CONFIG_AUFS_EXPORT */
28568 +
28569 +/* ---------------------------------------------------------------------- */
28570 +
28571 +#ifdef CONFIG_AUFS_SBILIST
28572 +/* module.c */
28573 +extern struct au_sphlhead au_sbilist;
28574 +
28575 +static inline void au_sbilist_init(void)
28576 +{
28577 +       au_sphl_init(&au_sbilist);
28578 +}
28579 +
28580 +static inline void au_sbilist_add(struct super_block *sb)
28581 +{
28582 +       au_sphl_add(&au_sbi(sb)->si_list, &au_sbilist);
28583 +}
28584 +
28585 +static inline void au_sbilist_del(struct super_block *sb)
28586 +{
28587 +       au_sphl_del(&au_sbi(sb)->si_list, &au_sbilist);
28588 +}
28589 +
28590 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
28591 +static inline void au_sbilist_lock(void)
28592 +{
28593 +       spin_lock(&au_sbilist.spin);
28594 +}
28595 +
28596 +static inline void au_sbilist_unlock(void)
28597 +{
28598 +       spin_unlock(&au_sbilist.spin);
28599 +}
28600 +#define AuGFP_SBILIST  GFP_ATOMIC
28601 +#else
28602 +AuStubVoid(au_sbilist_lock, void)
28603 +AuStubVoid(au_sbilist_unlock, void)
28604 +#define AuGFP_SBILIST  GFP_NOFS
28605 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
28606 +#else
28607 +AuStubVoid(au_sbilist_init, void)
28608 +AuStubVoid(au_sbilist_add, struct super_block *sb)
28609 +AuStubVoid(au_sbilist_del, struct super_block *sb)
28610 +AuStubVoid(au_sbilist_lock, void)
28611 +AuStubVoid(au_sbilist_unlock, void)
28612 +#define AuGFP_SBILIST  GFP_NOFS
28613 +#endif
28614 +
28615 +/* ---------------------------------------------------------------------- */
28616 +
28617 +static inline void dbgaufs_si_null(struct au_sbinfo *sbinfo)
28618 +{
28619 +       /*
28620 +        * This function is a dynamic '__init' function actually,
28621 +        * so the tiny check for si_rwsem is unnecessary.
28622 +        */
28623 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
28624 +#ifdef CONFIG_DEBUG_FS
28625 +       sbinfo->si_dbgaufs = NULL;
28626 +       sbinfo->si_dbgaufs_plink = NULL;
28627 +       sbinfo->si_dbgaufs_xib = NULL;
28628 +#ifdef CONFIG_AUFS_EXPORT
28629 +       sbinfo->si_dbgaufs_xigen = NULL;
28630 +#endif
28631 +#endif
28632 +}
28633 +
28634 +/* ---------------------------------------------------------------------- */
28635 +
28636 +static inline void si_pid_idx_bit(int *idx, pid_t *bit)
28637 +{
28638 +       /* the origin of pid is 1, but the bitmap's is 0 */
28639 +       *bit = current->pid - 1;
28640 +       *idx = *bit / AU_PIDSTEP;
28641 +       *bit %= AU_PIDSTEP;
28642 +}
28643 +
28644 +static inline int si_pid_test(struct super_block *sb)
28645 +{
28646 +       pid_t bit;
28647 +       int idx;
28648 +       unsigned long *bitmap;
28649 +
28650 +       si_pid_idx_bit(&idx, &bit);
28651 +       bitmap = au_sbi(sb)->au_si_pid.pid_bitmap[idx];
28652 +       if (bitmap)
28653 +               return test_bit(bit, bitmap);
28654 +       return 0;
28655 +}
28656 +
28657 +static inline void si_pid_clr(struct super_block *sb)
28658 +{
28659 +       pid_t bit;
28660 +       int idx;
28661 +       unsigned long *bitmap;
28662 +
28663 +       si_pid_idx_bit(&idx, &bit);
28664 +       bitmap = au_sbi(sb)->au_si_pid.pid_bitmap[idx];
28665 +       BUG_ON(!bitmap);
28666 +       AuDebugOn(!test_bit(bit, bitmap));
28667 +       clear_bit(bit, bitmap);
28668 +       /* smp_mb(); */
28669 +}
28670 +
28671 +void si_pid_set(struct super_block *sb);
28672 +
28673 +/* ---------------------------------------------------------------------- */
28674 +
28675 +/* lock superblock. mainly for entry point functions */
28676 +/*
28677 + * __si_read_lock, __si_write_lock,
28678 + * __si_read_unlock, __si_write_unlock, __si_downgrade_lock
28679 + */
28680 +AuSimpleRwsemFuncs(__si, struct super_block *sb, &au_sbi(sb)->si_rwsem);
28681 +
28682 +#define SiMustNoWaiters(sb)    AuRwMustNoWaiters(&au_sbi(sb)->si_rwsem)
28683 +#define SiMustAnyLock(sb)      AuRwMustAnyLock(&au_sbi(sb)->si_rwsem)
28684 +#define SiMustWriteLock(sb)    AuRwMustWriteLock(&au_sbi(sb)->si_rwsem)
28685 +
28686 +static inline void si_noflush_read_lock(struct super_block *sb)
28687 +{
28688 +       __si_read_lock(sb);
28689 +       si_pid_set(sb);
28690 +}
28691 +
28692 +static inline int si_noflush_read_trylock(struct super_block *sb)
28693 +{
28694 +       int locked;
28695 +
28696 +       locked = __si_read_trylock(sb);
28697 +       if (locked)
28698 +               si_pid_set(sb);
28699 +       return locked;
28700 +}
28701 +
28702 +static inline void si_noflush_write_lock(struct super_block *sb)
28703 +{
28704 +       __si_write_lock(sb);
28705 +       si_pid_set(sb);
28706 +}
28707 +
28708 +static inline int si_noflush_write_trylock(struct super_block *sb)
28709 +{
28710 +       int locked;
28711 +
28712 +       locked = __si_write_trylock(sb);
28713 +       if (locked)
28714 +               si_pid_set(sb);
28715 +       return locked;
28716 +}
28717 +
28718 +#if 0 /* reserved */
28719 +static inline int si_read_trylock(struct super_block *sb, int flags)
28720 +{
28721 +       if (au_ftest_lock(flags, FLUSH))
28722 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
28723 +       return si_noflush_read_trylock(sb);
28724 +}
28725 +#endif
28726 +
28727 +static inline void si_read_unlock(struct super_block *sb)
28728 +{
28729 +       si_pid_clr(sb);
28730 +       __si_read_unlock(sb);
28731 +}
28732 +
28733 +#if 0 /* reserved */
28734 +static inline int si_write_trylock(struct super_block *sb, int flags)
28735 +{
28736 +       if (au_ftest_lock(flags, FLUSH))
28737 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
28738 +       return si_noflush_write_trylock(sb);
28739 +}
28740 +#endif
28741 +
28742 +static inline void si_write_unlock(struct super_block *sb)
28743 +{
28744 +       si_pid_clr(sb);
28745 +       __si_write_unlock(sb);
28746 +}
28747 +
28748 +#if 0 /* reserved */
28749 +static inline void si_downgrade_lock(struct super_block *sb)
28750 +{
28751 +       __si_downgrade_lock(sb);
28752 +}
28753 +#endif
28754 +
28755 +/* ---------------------------------------------------------------------- */
28756 +
28757 +static inline aufs_bindex_t au_sbbot(struct super_block *sb)
28758 +{
28759 +       SiMustAnyLock(sb);
28760 +       return au_sbi(sb)->si_bbot;
28761 +}
28762 +
28763 +static inline unsigned int au_mntflags(struct super_block *sb)
28764 +{
28765 +       SiMustAnyLock(sb);
28766 +       return au_sbi(sb)->si_mntflags;
28767 +}
28768 +
28769 +static inline unsigned int au_sigen(struct super_block *sb)
28770 +{
28771 +       SiMustAnyLock(sb);
28772 +       return au_sbi(sb)->si_generation;
28773 +}
28774 +
28775 +static inline unsigned long long au_ninodes(struct super_block *sb)
28776 +{
28777 +       s64 n = percpu_counter_sum(&au_sbi(sb)->si_ninodes);
28778 +
28779 +       BUG_ON(n < 0);
28780 +       return n;
28781 +}
28782 +
28783 +static inline void au_ninodes_inc(struct super_block *sb)
28784 +{
28785 +       percpu_counter_inc(&au_sbi(sb)->si_ninodes);
28786 +}
28787 +
28788 +static inline void au_ninodes_dec(struct super_block *sb)
28789 +{
28790 +       percpu_counter_dec(&au_sbi(sb)->si_ninodes);
28791 +}
28792 +
28793 +static inline unsigned long long au_nfiles(struct super_block *sb)
28794 +{
28795 +       s64 n = percpu_counter_sum(&au_sbi(sb)->si_nfiles);
28796 +
28797 +       BUG_ON(n < 0);
28798 +       return n;
28799 +}
28800 +
28801 +static inline void au_nfiles_inc(struct super_block *sb)
28802 +{
28803 +       percpu_counter_inc(&au_sbi(sb)->si_nfiles);
28804 +}
28805 +
28806 +static inline void au_nfiles_dec(struct super_block *sb)
28807 +{
28808 +       percpu_counter_dec(&au_sbi(sb)->si_nfiles);
28809 +}
28810 +
28811 +static inline struct au_branch *au_sbr(struct super_block *sb,
28812 +                                      aufs_bindex_t bindex)
28813 +{
28814 +       SiMustAnyLock(sb);
28815 +       return au_sbi(sb)->si_branch[0 + bindex];
28816 +}
28817 +
28818 +static inline void au_xino_brid_set(struct super_block *sb, aufs_bindex_t brid)
28819 +{
28820 +       SiMustWriteLock(sb);
28821 +       au_sbi(sb)->si_xino_brid = brid;
28822 +}
28823 +
28824 +static inline aufs_bindex_t au_xino_brid(struct super_block *sb)
28825 +{
28826 +       SiMustAnyLock(sb);
28827 +       return au_sbi(sb)->si_xino_brid;
28828 +}
28829 +
28830 +#endif /* __KERNEL__ */
28831 +#endif /* __AUFS_SUPER_H__ */
28832 diff -urN /usr/share/empty/fs/aufs/sysaufs.c linux/fs/aufs/sysaufs.c
28833 --- /usr/share/empty/fs/aufs/sysaufs.c  1970-01-01 01:00:00.000000000 +0100
28834 +++ linux/fs/aufs/sysaufs.c     2016-10-09 16:55:36.496035060 +0200
28835 @@ -0,0 +1,104 @@
28836 +/*
28837 + * Copyright (C) 2005-2016 Junjiro R. Okajima
28838 + *
28839 + * This program, aufs is free software; you can redistribute it and/or modify
28840 + * it under the terms of the GNU General Public License as published by
28841 + * the Free Software Foundation; either version 2 of the License, or
28842 + * (at your option) any later version.
28843 + *
28844 + * This program is distributed in the hope that it will be useful,
28845 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28846 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28847 + * GNU General Public License for more details.
28848 + *
28849 + * You should have received a copy of the GNU General Public License
28850 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28851 + */
28852 +
28853 +/*
28854 + * sysfs interface and lifetime management
28855 + * they are necessary regardless sysfs is disabled.
28856 + */
28857 +
28858 +#include <linux/random.h>
28859 +#include "aufs.h"
28860 +
28861 +unsigned long sysaufs_si_mask;
28862 +struct kset *sysaufs_kset;
28863 +
28864 +#define AuSiAttr(_name) { \
28865 +       .attr   = { .name = __stringify(_name), .mode = 0444 }, \
28866 +       .show   = sysaufs_si_##_name,                           \
28867 +}
28868 +
28869 +static struct sysaufs_si_attr sysaufs_si_attr_xi_path = AuSiAttr(xi_path);
28870 +struct attribute *sysaufs_si_attrs[] = {
28871 +       &sysaufs_si_attr_xi_path.attr,
28872 +       NULL,
28873 +};
28874 +
28875 +static const struct sysfs_ops au_sbi_ops = {
28876 +       .show   = sysaufs_si_show
28877 +};
28878 +
28879 +static struct kobj_type au_sbi_ktype = {
28880 +       .release        = au_si_free,
28881 +       .sysfs_ops      = &au_sbi_ops,
28882 +       .default_attrs  = sysaufs_si_attrs
28883 +};
28884 +
28885 +/* ---------------------------------------------------------------------- */
28886 +
28887 +int sysaufs_si_init(struct au_sbinfo *sbinfo)
28888 +{
28889 +       int err;
28890 +
28891 +       sbinfo->si_kobj.kset = sysaufs_kset;
28892 +       /* cf. sysaufs_name() */
28893 +       err = kobject_init_and_add
28894 +               (&sbinfo->si_kobj, &au_sbi_ktype, /*&sysaufs_kset->kobj*/NULL,
28895 +                SysaufsSiNamePrefix "%lx", sysaufs_si_id(sbinfo));
28896 +
28897 +       dbgaufs_si_null(sbinfo);
28898 +       if (!err) {
28899 +               err = dbgaufs_si_init(sbinfo);
28900 +               if (unlikely(err))
28901 +                       kobject_put(&sbinfo->si_kobj);
28902 +       }
28903 +       return err;
28904 +}
28905 +
28906 +void sysaufs_fin(void)
28907 +{
28908 +       dbgaufs_fin();
28909 +       sysfs_remove_group(&sysaufs_kset->kobj, sysaufs_attr_group);
28910 +       kset_unregister(sysaufs_kset);
28911 +}
28912 +
28913 +int __init sysaufs_init(void)
28914 +{
28915 +       int err;
28916 +
28917 +       do {
28918 +               get_random_bytes(&sysaufs_si_mask, sizeof(sysaufs_si_mask));
28919 +       } while (!sysaufs_si_mask);
28920 +
28921 +       err = -EINVAL;
28922 +       sysaufs_kset = kset_create_and_add(AUFS_NAME, NULL, fs_kobj);
28923 +       if (unlikely(!sysaufs_kset))
28924 +               goto out;
28925 +       err = PTR_ERR(sysaufs_kset);
28926 +       if (IS_ERR(sysaufs_kset))
28927 +               goto out;
28928 +       err = sysfs_create_group(&sysaufs_kset->kobj, sysaufs_attr_group);
28929 +       if (unlikely(err)) {
28930 +               kset_unregister(sysaufs_kset);
28931 +               goto out;
28932 +       }
28933 +
28934 +       err = dbgaufs_init();
28935 +       if (unlikely(err))
28936 +               sysaufs_fin();
28937 +out:
28938 +       return err;
28939 +}
28940 diff -urN /usr/share/empty/fs/aufs/sysaufs.h linux/fs/aufs/sysaufs.h
28941 --- /usr/share/empty/fs/aufs/sysaufs.h  1970-01-01 01:00:00.000000000 +0100
28942 +++ linux/fs/aufs/sysaufs.h     2016-10-09 16:55:36.496035060 +0200
28943 @@ -0,0 +1,101 @@
28944 +/*
28945 + * Copyright (C) 2005-2016 Junjiro R. Okajima
28946 + *
28947 + * This program, aufs is free software; you can redistribute it and/or modify
28948 + * it under the terms of the GNU General Public License as published by
28949 + * the Free Software Foundation; either version 2 of the License, or
28950 + * (at your option) any later version.
28951 + *
28952 + * This program is distributed in the hope that it will be useful,
28953 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28954 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28955 + * GNU General Public License for more details.
28956 + *
28957 + * You should have received a copy of the GNU General Public License
28958 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28959 + */
28960 +
28961 +/*
28962 + * sysfs interface and mount lifetime management
28963 + */
28964 +
28965 +#ifndef __SYSAUFS_H__
28966 +#define __SYSAUFS_H__
28967 +
28968 +#ifdef __KERNEL__
28969 +
28970 +#include <linux/sysfs.h>
28971 +#include "module.h"
28972 +
28973 +struct super_block;
28974 +struct au_sbinfo;
28975 +
28976 +struct sysaufs_si_attr {
28977 +       struct attribute attr;
28978 +       int (*show)(struct seq_file *seq, struct super_block *sb);
28979 +};
28980 +
28981 +/* ---------------------------------------------------------------------- */
28982 +
28983 +/* sysaufs.c */
28984 +extern unsigned long sysaufs_si_mask;
28985 +extern struct kset *sysaufs_kset;
28986 +extern struct attribute *sysaufs_si_attrs[];
28987 +int sysaufs_si_init(struct au_sbinfo *sbinfo);
28988 +int __init sysaufs_init(void);
28989 +void sysaufs_fin(void);
28990 +
28991 +/* ---------------------------------------------------------------------- */
28992 +
28993 +/* some people doesn't like to show a pointer in kernel */
28994 +static inline unsigned long sysaufs_si_id(struct au_sbinfo *sbinfo)
28995 +{
28996 +       return sysaufs_si_mask ^ (unsigned long)sbinfo;
28997 +}
28998 +
28999 +#define SysaufsSiNamePrefix    "si_"
29000 +#define SysaufsSiNameLen       (sizeof(SysaufsSiNamePrefix) + 16)
29001 +static inline void sysaufs_name(struct au_sbinfo *sbinfo, char *name)
29002 +{
29003 +       snprintf(name, SysaufsSiNameLen, SysaufsSiNamePrefix "%lx",
29004 +                sysaufs_si_id(sbinfo));
29005 +}
29006 +
29007 +struct au_branch;
29008 +#ifdef CONFIG_SYSFS
29009 +/* sysfs.c */
29010 +extern struct attribute_group *sysaufs_attr_group;
29011 +
29012 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb);
29013 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
29014 +                        char *buf);
29015 +long au_brinfo_ioctl(struct file *file, unsigned long arg);
29016 +#ifdef CONFIG_COMPAT
29017 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg);
29018 +#endif
29019 +
29020 +void sysaufs_br_init(struct au_branch *br);
29021 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
29022 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
29023 +
29024 +#define sysaufs_brs_init()     do {} while (0)
29025 +
29026 +#else
29027 +#define sysaufs_attr_group     NULL
29028 +
29029 +AuStubInt0(sysaufs_si_xi_path, struct seq_file *seq, struct super_block *sb)
29030 +AuStub(ssize_t, sysaufs_si_show, return 0, struct kobject *kobj,
29031 +       struct attribute *attr, char *buf)
29032 +AuStubVoid(sysaufs_br_init, struct au_branch *br)
29033 +AuStubVoid(sysaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
29034 +AuStubVoid(sysaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
29035 +
29036 +static inline void sysaufs_brs_init(void)
29037 +{
29038 +       sysaufs_brs = 0;
29039 +}
29040 +
29041 +#endif /* CONFIG_SYSFS */
29042 +
29043 +#endif /* __KERNEL__ */
29044 +#endif /* __SYSAUFS_H__ */
29045 diff -urN /usr/share/empty/fs/aufs/sysfs.c linux/fs/aufs/sysfs.c
29046 --- /usr/share/empty/fs/aufs/sysfs.c    1970-01-01 01:00:00.000000000 +0100
29047 +++ linux/fs/aufs/sysfs.c       2016-10-09 16:55:36.496035060 +0200
29048 @@ -0,0 +1,376 @@
29049 +/*
29050 + * Copyright (C) 2005-2016 Junjiro R. Okajima
29051 + *
29052 + * This program, aufs is free software; you can redistribute it and/or modify
29053 + * it under the terms of the GNU General Public License as published by
29054 + * the Free Software Foundation; either version 2 of the License, or
29055 + * (at your option) any later version.
29056 + *
29057 + * This program is distributed in the hope that it will be useful,
29058 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29059 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29060 + * GNU General Public License for more details.
29061 + *
29062 + * You should have received a copy of the GNU General Public License
29063 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29064 + */
29065 +
29066 +/*
29067 + * sysfs interface
29068 + */
29069 +
29070 +#include <linux/compat.h>
29071 +#include <linux/seq_file.h>
29072 +#include "aufs.h"
29073 +
29074 +#ifdef CONFIG_AUFS_FS_MODULE
29075 +/* this entry violates the "one line per file" policy of sysfs */
29076 +static ssize_t config_show(struct kobject *kobj, struct kobj_attribute *attr,
29077 +                          char *buf)
29078 +{
29079 +       ssize_t err;
29080 +       static char *conf =
29081 +/* this file is generated at compiling */
29082 +#include "conf.str"
29083 +               ;
29084 +
29085 +       err = snprintf(buf, PAGE_SIZE, conf);
29086 +       if (unlikely(err >= PAGE_SIZE))
29087 +               err = -EFBIG;
29088 +       return err;
29089 +}
29090 +
29091 +static struct kobj_attribute au_config_attr = __ATTR_RO(config);
29092 +#endif
29093 +
29094 +static struct attribute *au_attr[] = {
29095 +#ifdef CONFIG_AUFS_FS_MODULE
29096 +       &au_config_attr.attr,
29097 +#endif
29098 +       NULL,   /* need to NULL terminate the list of attributes */
29099 +};
29100 +
29101 +static struct attribute_group sysaufs_attr_group_body = {
29102 +       .attrs = au_attr
29103 +};
29104 +
29105 +struct attribute_group *sysaufs_attr_group = &sysaufs_attr_group_body;
29106 +
29107 +/* ---------------------------------------------------------------------- */
29108 +
29109 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb)
29110 +{
29111 +       int err;
29112 +
29113 +       SiMustAnyLock(sb);
29114 +
29115 +       err = 0;
29116 +       if (au_opt_test(au_mntflags(sb), XINO)) {
29117 +               err = au_xino_path(seq, au_sbi(sb)->si_xib);
29118 +               seq_putc(seq, '\n');
29119 +       }
29120 +       return err;
29121 +}
29122 +
29123 +/*
29124 + * the lifetime of branch is independent from the entry under sysfs.
29125 + * sysfs handles the lifetime of the entry, and never call ->show() after it is
29126 + * unlinked.
29127 + */
29128 +static int sysaufs_si_br(struct seq_file *seq, struct super_block *sb,
29129 +                        aufs_bindex_t bindex, int idx)
29130 +{
29131 +       int err;
29132 +       struct path path;
29133 +       struct dentry *root;
29134 +       struct au_branch *br;
29135 +       au_br_perm_str_t perm;
29136 +
29137 +       AuDbg("b%d\n", bindex);
29138 +
29139 +       err = 0;
29140 +       root = sb->s_root;
29141 +       di_read_lock_parent(root, !AuLock_IR);
29142 +       br = au_sbr(sb, bindex);
29143 +
29144 +       switch (idx) {
29145 +       case AuBrSysfs_BR:
29146 +               path.mnt = au_br_mnt(br);
29147 +               path.dentry = au_h_dptr(root, bindex);
29148 +               err = au_seq_path(seq, &path);
29149 +               if (!err) {
29150 +                       au_optstr_br_perm(&perm, br->br_perm);
29151 +                       seq_printf(seq, "=%s\n", perm.a);
29152 +               }
29153 +               break;
29154 +       case AuBrSysfs_BRID:
29155 +               seq_printf(seq, "%d\n", br->br_id);
29156 +               break;
29157 +       }
29158 +       di_read_unlock(root, !AuLock_IR);
29159 +       if (unlikely(err || seq_has_overflowed(seq)))
29160 +               err = -E2BIG;
29161 +
29162 +       return err;
29163 +}
29164 +
29165 +/* ---------------------------------------------------------------------- */
29166 +
29167 +static struct seq_file *au_seq(char *p, ssize_t len)
29168 +{
29169 +       struct seq_file *seq;
29170 +
29171 +       seq = kzalloc(sizeof(*seq), GFP_NOFS);
29172 +       if (seq) {
29173 +               /* mutex_init(&seq.lock); */
29174 +               seq->buf = p;
29175 +               seq->size = len;
29176 +               return seq; /* success */
29177 +       }
29178 +
29179 +       seq = ERR_PTR(-ENOMEM);
29180 +       return seq;
29181 +}
29182 +
29183 +#define SysaufsBr_PREFIX       "br"
29184 +#define SysaufsBrid_PREFIX     "brid"
29185 +
29186 +/* todo: file size may exceed PAGE_SIZE */
29187 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
29188 +                       char *buf)
29189 +{
29190 +       ssize_t err;
29191 +       int idx;
29192 +       long l;
29193 +       aufs_bindex_t bbot;
29194 +       struct au_sbinfo *sbinfo;
29195 +       struct super_block *sb;
29196 +       struct seq_file *seq;
29197 +       char *name;
29198 +       struct attribute **cattr;
29199 +
29200 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
29201 +       sb = sbinfo->si_sb;
29202 +
29203 +       /*
29204 +        * prevent a race condition between sysfs and aufs.
29205 +        * for instance, sysfs_file_read() calls sysfs_get_active_two() which
29206 +        * prohibits maintaining the sysfs entries.
29207 +        * hew we acquire read lock after sysfs_get_active_two().
29208 +        * on the other hand, the remount process may maintain the sysfs/aufs
29209 +        * entries after acquiring write lock.
29210 +        * it can cause a deadlock.
29211 +        * simply we gave up processing read here.
29212 +        */
29213 +       err = -EBUSY;
29214 +       if (unlikely(!si_noflush_read_trylock(sb)))
29215 +               goto out;
29216 +
29217 +       seq = au_seq(buf, PAGE_SIZE);
29218 +       err = PTR_ERR(seq);
29219 +       if (IS_ERR(seq))
29220 +               goto out_unlock;
29221 +
29222 +       name = (void *)attr->name;
29223 +       cattr = sysaufs_si_attrs;
29224 +       while (*cattr) {
29225 +               if (!strcmp(name, (*cattr)->name)) {
29226 +                       err = container_of(*cattr, struct sysaufs_si_attr, attr)
29227 +                               ->show(seq, sb);
29228 +                       goto out_seq;
29229 +               }
29230 +               cattr++;
29231 +       }
29232 +
29233 +       if (!strncmp(name, SysaufsBrid_PREFIX,
29234 +                    sizeof(SysaufsBrid_PREFIX) - 1)) {
29235 +               idx = AuBrSysfs_BRID;
29236 +               name += sizeof(SysaufsBrid_PREFIX) - 1;
29237 +       } else if (!strncmp(name, SysaufsBr_PREFIX,
29238 +                           sizeof(SysaufsBr_PREFIX) - 1)) {
29239 +               idx = AuBrSysfs_BR;
29240 +               name += sizeof(SysaufsBr_PREFIX) - 1;
29241 +       } else
29242 +                 BUG();
29243 +
29244 +       err = kstrtol(name, 10, &l);
29245 +       if (!err) {
29246 +               bbot = au_sbbot(sb);
29247 +               if (l <= bbot)
29248 +                       err = sysaufs_si_br(seq, sb, (aufs_bindex_t)l, idx);
29249 +               else
29250 +                       err = -ENOENT;
29251 +       }
29252 +
29253 +out_seq:
29254 +       if (!err) {
29255 +               err = seq->count;
29256 +               /* sysfs limit */
29257 +               if (unlikely(err == PAGE_SIZE))
29258 +                       err = -EFBIG;
29259 +       }
29260 +       au_delayed_kfree(seq);
29261 +out_unlock:
29262 +       si_read_unlock(sb);
29263 +out:
29264 +       return err;
29265 +}
29266 +
29267 +/* ---------------------------------------------------------------------- */
29268 +
29269 +static int au_brinfo(struct super_block *sb, union aufs_brinfo __user *arg)
29270 +{
29271 +       int err;
29272 +       int16_t brid;
29273 +       aufs_bindex_t bindex, bbot;
29274 +       size_t sz;
29275 +       char *buf;
29276 +       struct seq_file *seq;
29277 +       struct au_branch *br;
29278 +
29279 +       si_read_lock(sb, AuLock_FLUSH);
29280 +       bbot = au_sbbot(sb);
29281 +       err = bbot + 1;
29282 +       if (!arg)
29283 +               goto out;
29284 +
29285 +       err = -ENOMEM;
29286 +       buf = (void *)__get_free_page(GFP_NOFS);
29287 +       if (unlikely(!buf))
29288 +               goto out;
29289 +
29290 +       seq = au_seq(buf, PAGE_SIZE);
29291 +       err = PTR_ERR(seq);
29292 +       if (IS_ERR(seq))
29293 +               goto out_buf;
29294 +
29295 +       sz = sizeof(*arg) - offsetof(union aufs_brinfo, path);
29296 +       for (bindex = 0; bindex <= bbot; bindex++, arg++) {
29297 +               err = !access_ok(VERIFY_WRITE, arg, sizeof(*arg));
29298 +               if (unlikely(err))
29299 +                       break;
29300 +
29301 +               br = au_sbr(sb, bindex);
29302 +               brid = br->br_id;
29303 +               BUILD_BUG_ON(sizeof(brid) != sizeof(arg->id));
29304 +               err = __put_user(brid, &arg->id);
29305 +               if (unlikely(err))
29306 +                       break;
29307 +
29308 +               BUILD_BUG_ON(sizeof(br->br_perm) != sizeof(arg->perm));
29309 +               err = __put_user(br->br_perm, &arg->perm);
29310 +               if (unlikely(err))
29311 +                       break;
29312 +
29313 +               err = au_seq_path(seq, &br->br_path);
29314 +               if (unlikely(err))
29315 +                       break;
29316 +               seq_putc(seq, '\0');
29317 +               if (!seq_has_overflowed(seq)) {
29318 +                       err = copy_to_user(arg->path, seq->buf, seq->count);
29319 +                       seq->count = 0;
29320 +                       if (unlikely(err))
29321 +                               break;
29322 +               } else {
29323 +                       err = -E2BIG;
29324 +                       goto out_seq;
29325 +               }
29326 +       }
29327 +       if (unlikely(err))
29328 +               err = -EFAULT;
29329 +
29330 +out_seq:
29331 +       au_delayed_kfree(seq);
29332 +out_buf:
29333 +       au_delayed_free_page((unsigned long)buf);
29334 +out:
29335 +       si_read_unlock(sb);
29336 +       return err;
29337 +}
29338 +
29339 +long au_brinfo_ioctl(struct file *file, unsigned long arg)
29340 +{
29341 +       return au_brinfo(file->f_path.dentry->d_sb, (void __user *)arg);
29342 +}
29343 +
29344 +#ifdef CONFIG_COMPAT
29345 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg)
29346 +{
29347 +       return au_brinfo(file->f_path.dentry->d_sb, compat_ptr(arg));
29348 +}
29349 +#endif
29350 +
29351 +/* ---------------------------------------------------------------------- */
29352 +
29353 +void sysaufs_br_init(struct au_branch *br)
29354 +{
29355 +       int i;
29356 +       struct au_brsysfs *br_sysfs;
29357 +       struct attribute *attr;
29358 +
29359 +       br_sysfs = br->br_sysfs;
29360 +       for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
29361 +               attr = &br_sysfs->attr;
29362 +               sysfs_attr_init(attr);
29363 +               attr->name = br_sysfs->name;
29364 +               attr->mode = S_IRUGO;
29365 +               br_sysfs++;
29366 +       }
29367 +}
29368 +
29369 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
29370 +{
29371 +       struct au_branch *br;
29372 +       struct kobject *kobj;
29373 +       struct au_brsysfs *br_sysfs;
29374 +       int i;
29375 +       aufs_bindex_t bbot;
29376 +
29377 +       dbgaufs_brs_del(sb, bindex);
29378 +
29379 +       if (!sysaufs_brs)
29380 +               return;
29381 +
29382 +       kobj = &au_sbi(sb)->si_kobj;
29383 +       bbot = au_sbbot(sb);
29384 +       for (; bindex <= bbot; bindex++) {
29385 +               br = au_sbr(sb, bindex);
29386 +               br_sysfs = br->br_sysfs;
29387 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
29388 +                       sysfs_remove_file(kobj, &br_sysfs->attr);
29389 +                       br_sysfs++;
29390 +               }
29391 +       }
29392 +}
29393 +
29394 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
29395 +{
29396 +       int err, i;
29397 +       aufs_bindex_t bbot;
29398 +       struct kobject *kobj;
29399 +       struct au_branch *br;
29400 +       struct au_brsysfs *br_sysfs;
29401 +
29402 +       dbgaufs_brs_add(sb, bindex);
29403 +
29404 +       if (!sysaufs_brs)
29405 +               return;
29406 +
29407 +       kobj = &au_sbi(sb)->si_kobj;
29408 +       bbot = au_sbbot(sb);
29409 +       for (; bindex <= bbot; bindex++) {
29410 +               br = au_sbr(sb, bindex);
29411 +               br_sysfs = br->br_sysfs;
29412 +               snprintf(br_sysfs[AuBrSysfs_BR].name, sizeof(br_sysfs->name),
29413 +                        SysaufsBr_PREFIX "%d", bindex);
29414 +               snprintf(br_sysfs[AuBrSysfs_BRID].name, sizeof(br_sysfs->name),
29415 +                        SysaufsBrid_PREFIX "%d", bindex);
29416 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
29417 +                       err = sysfs_create_file(kobj, &br_sysfs->attr);
29418 +                       if (unlikely(err))
29419 +                               pr_warn("failed %s under sysfs(%d)\n",
29420 +                                       br_sysfs->name, err);
29421 +                       br_sysfs++;
29422 +               }
29423 +       }
29424 +}
29425 diff -urN /usr/share/empty/fs/aufs/sysrq.c linux/fs/aufs/sysrq.c
29426 --- /usr/share/empty/fs/aufs/sysrq.c    1970-01-01 01:00:00.000000000 +0100
29427 +++ linux/fs/aufs/sysrq.c       2016-10-09 16:55:36.496035060 +0200
29428 @@ -0,0 +1,157 @@
29429 +/*
29430 + * Copyright (C) 2005-2016 Junjiro R. Okajima
29431 + *
29432 + * This program, aufs is free software; you can redistribute it and/or modify
29433 + * it under the terms of the GNU General Public License as published by
29434 + * the Free Software Foundation; either version 2 of the License, or
29435 + * (at your option) any later version.
29436 + *
29437 + * This program is distributed in the hope that it will be useful,
29438 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29439 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29440 + * GNU General Public License for more details.
29441 + *
29442 + * You should have received a copy of the GNU General Public License
29443 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29444 + */
29445 +
29446 +/*
29447 + * magic sysrq hanlder
29448 + */
29449 +
29450 +/* #include <linux/sysrq.h> */
29451 +#include <linux/writeback.h>
29452 +#include "aufs.h"
29453 +
29454 +/* ---------------------------------------------------------------------- */
29455 +
29456 +static void sysrq_sb(struct super_block *sb)
29457 +{
29458 +       char *plevel;
29459 +       struct au_sbinfo *sbinfo;
29460 +       struct file *file;
29461 +       struct au_sphlhead *files;
29462 +       struct au_finfo *finfo;
29463 +
29464 +       plevel = au_plevel;
29465 +       au_plevel = KERN_WARNING;
29466 +
29467 +       /* since we define pr_fmt, call printk directly */
29468 +#define pr(str) printk(KERN_WARNING AUFS_NAME ": " str)
29469 +
29470 +       sbinfo = au_sbi(sb);
29471 +       printk(KERN_WARNING "si=%lx\n", sysaufs_si_id(sbinfo));
29472 +       pr("superblock\n");
29473 +       au_dpri_sb(sb);
29474 +
29475 +#if 0
29476 +       pr("root dentry\n");
29477 +       au_dpri_dentry(sb->s_root);
29478 +       pr("root inode\n");
29479 +       au_dpri_inode(d_inode(sb->s_root));
29480 +#endif
29481 +
29482 +#if 0
29483 +       do {
29484 +               int err, i, j, ndentry;
29485 +               struct au_dcsub_pages dpages;
29486 +               struct au_dpage *dpage;
29487 +
29488 +               err = au_dpages_init(&dpages, GFP_ATOMIC);
29489 +               if (unlikely(err))
29490 +                       break;
29491 +               err = au_dcsub_pages(&dpages, sb->s_root, NULL, NULL);
29492 +               if (!err)
29493 +                       for (i = 0; i < dpages.ndpage; i++) {
29494 +                               dpage = dpages.dpages + i;
29495 +                               ndentry = dpage->ndentry;
29496 +                               for (j = 0; j < ndentry; j++)
29497 +                                       au_dpri_dentry(dpage->dentries[j]);
29498 +                       }
29499 +               au_dpages_free(&dpages);
29500 +       } while (0);
29501 +#endif
29502 +
29503 +#if 1
29504 +       {
29505 +               struct inode *i;
29506 +
29507 +               pr("isolated inode\n");
29508 +               spin_lock(&sb->s_inode_list_lock);
29509 +               list_for_each_entry(i, &sb->s_inodes, i_sb_list) {
29510 +                       spin_lock(&i->i_lock);
29511 +                       if (1 || hlist_empty(&i->i_dentry))
29512 +                               au_dpri_inode(i);
29513 +                       spin_unlock(&i->i_lock);
29514 +               }
29515 +               spin_unlock(&sb->s_inode_list_lock);
29516 +       }
29517 +#endif
29518 +       pr("files\n");
29519 +       files = &au_sbi(sb)->si_files;
29520 +       spin_lock(&files->spin);
29521 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
29522 +               umode_t mode;
29523 +
29524 +               file = finfo->fi_file;
29525 +               mode = file_inode(file)->i_mode;
29526 +               if (!special_file(mode))
29527 +                       au_dpri_file(file);
29528 +       }
29529 +       spin_unlock(&files->spin);
29530 +       pr("done\n");
29531 +
29532 +#undef pr
29533 +       au_plevel = plevel;
29534 +}
29535 +
29536 +/* ---------------------------------------------------------------------- */
29537 +
29538 +/* module parameter */
29539 +static char *aufs_sysrq_key = "a";
29540 +module_param_named(sysrq, aufs_sysrq_key, charp, S_IRUGO);
29541 +MODULE_PARM_DESC(sysrq, "MagicSysRq key for " AUFS_NAME);
29542 +
29543 +static void au_sysrq(int key __maybe_unused)
29544 +{
29545 +       struct au_sbinfo *sbinfo;
29546 +
29547 +       lockdep_off();
29548 +       au_sbilist_lock();
29549 +       hlist_for_each_entry(sbinfo, &au_sbilist.head, si_list)
29550 +               sysrq_sb(sbinfo->si_sb);
29551 +       au_sbilist_unlock();
29552 +       lockdep_on();
29553 +}
29554 +
29555 +static struct sysrq_key_op au_sysrq_op = {
29556 +       .handler        = au_sysrq,
29557 +       .help_msg       = "Aufs",
29558 +       .action_msg     = "Aufs",
29559 +       .enable_mask    = SYSRQ_ENABLE_DUMP
29560 +};
29561 +
29562 +/* ---------------------------------------------------------------------- */
29563 +
29564 +int __init au_sysrq_init(void)
29565 +{
29566 +       int err;
29567 +       char key;
29568 +
29569 +       err = -1;
29570 +       key = *aufs_sysrq_key;
29571 +       if ('a' <= key && key <= 'z')
29572 +               err = register_sysrq_key(key, &au_sysrq_op);
29573 +       if (unlikely(err))
29574 +               pr_err("err %d, sysrq=%c\n", err, key);
29575 +       return err;
29576 +}
29577 +
29578 +void au_sysrq_fin(void)
29579 +{
29580 +       int err;
29581 +
29582 +       err = unregister_sysrq_key(*aufs_sysrq_key, &au_sysrq_op);
29583 +       if (unlikely(err))
29584 +               pr_err("err %d (ignored)\n", err);
29585 +}
29586 diff -urN /usr/share/empty/fs/aufs/vdir.c linux/fs/aufs/vdir.c
29587 --- /usr/share/empty/fs/aufs/vdir.c     1970-01-01 01:00:00.000000000 +0100
29588 +++ linux/fs/aufs/vdir.c        2016-10-09 16:55:38.889431135 +0200
29589 @@ -0,0 +1,900 @@
29590 +/*
29591 + * Copyright (C) 2005-2016 Junjiro R. Okajima
29592 + *
29593 + * This program, aufs is free software; you can redistribute it and/or modify
29594 + * it under the terms of the GNU General Public License as published by
29595 + * the Free Software Foundation; either version 2 of the License, or
29596 + * (at your option) any later version.
29597 + *
29598 + * This program is distributed in the hope that it will be useful,
29599 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29600 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29601 + * GNU General Public License for more details.
29602 + *
29603 + * You should have received a copy of the GNU General Public License
29604 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29605 + */
29606 +
29607 +/*
29608 + * virtual or vertical directory
29609 + */
29610 +
29611 +#include "aufs.h"
29612 +
29613 +static unsigned int calc_size(int nlen)
29614 +{
29615 +       return ALIGN(sizeof(struct au_vdir_de) + nlen, sizeof(ino_t));
29616 +}
29617 +
29618 +static int set_deblk_end(union au_vdir_deblk_p *p,
29619 +                        union au_vdir_deblk_p *deblk_end)
29620 +{
29621 +       if (calc_size(0) <= deblk_end->deblk - p->deblk) {
29622 +               p->de->de_str.len = 0;
29623 +               /* smp_mb(); */
29624 +               return 0;
29625 +       }
29626 +       return -1; /* error */
29627 +}
29628 +
29629 +/* returns true or false */
29630 +static int is_deblk_end(union au_vdir_deblk_p *p,
29631 +                       union au_vdir_deblk_p *deblk_end)
29632 +{
29633 +       if (calc_size(0) <= deblk_end->deblk - p->deblk)
29634 +               return !p->de->de_str.len;
29635 +       return 1;
29636 +}
29637 +
29638 +static unsigned char *last_deblk(struct au_vdir *vdir)
29639 +{
29640 +       return vdir->vd_deblk[vdir->vd_nblk - 1];
29641 +}
29642 +
29643 +/* ---------------------------------------------------------------------- */
29644 +
29645 +/* estimate the appropriate size for name hash table */
29646 +unsigned int au_rdhash_est(loff_t sz)
29647 +{
29648 +       unsigned int n;
29649 +
29650 +       n = UINT_MAX;
29651 +       sz >>= 10;
29652 +       if (sz < n)
29653 +               n = sz;
29654 +       if (sz < AUFS_RDHASH_DEF)
29655 +               n = AUFS_RDHASH_DEF;
29656 +       /* pr_info("n %u\n", n); */
29657 +       return n;
29658 +}
29659 +
29660 +/*
29661 + * the allocated memory has to be freed by
29662 + * au_nhash_wh_free() or au_nhash_de_free().
29663 + */
29664 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp)
29665 +{
29666 +       struct hlist_head *head;
29667 +       unsigned int u;
29668 +       size_t sz;
29669 +
29670 +       sz = sizeof(*nhash->nh_head) * num_hash;
29671 +       head = kmalloc(sz, gfp);
29672 +       if (head) {
29673 +               nhash->nh_num = num_hash;
29674 +               nhash->nh_head = head;
29675 +               for (u = 0; u < num_hash; u++)
29676 +                       INIT_HLIST_HEAD(head++);
29677 +               return 0; /* success */
29678 +       }
29679 +
29680 +       return -ENOMEM;
29681 +}
29682 +
29683 +static void nhash_count(struct hlist_head *head)
29684 +{
29685 +#if 0
29686 +       unsigned long n;
29687 +       struct hlist_node *pos;
29688 +
29689 +       n = 0;
29690 +       hlist_for_each(pos, head)
29691 +               n++;
29692 +       pr_info("%lu\n", n);
29693 +#endif
29694 +}
29695 +
29696 +static void au_nhash_wh_do_free(struct hlist_head *head)
29697 +{
29698 +       struct au_vdir_wh *pos;
29699 +       struct hlist_node *node;
29700 +
29701 +       hlist_for_each_entry_safe(pos, node, head, wh_hash)
29702 +               au_delayed_kfree(pos);
29703 +}
29704 +
29705 +static void au_nhash_de_do_free(struct hlist_head *head)
29706 +{
29707 +       struct au_vdir_dehstr *pos;
29708 +       struct hlist_node *node;
29709 +
29710 +       hlist_for_each_entry_safe(pos, node, head, hash)
29711 +               au_cache_dfree_vdir_dehstr(pos);
29712 +}
29713 +
29714 +static void au_nhash_do_free(struct au_nhash *nhash,
29715 +                            void (*free)(struct hlist_head *head))
29716 +{
29717 +       unsigned int n;
29718 +       struct hlist_head *head;
29719 +
29720 +       n = nhash->nh_num;
29721 +       if (!n)
29722 +               return;
29723 +
29724 +       head = nhash->nh_head;
29725 +       while (n-- > 0) {
29726 +               nhash_count(head);
29727 +               free(head++);
29728 +       }
29729 +       au_delayed_kfree(nhash->nh_head);
29730 +}
29731 +
29732 +void au_nhash_wh_free(struct au_nhash *whlist)
29733 +{
29734 +       au_nhash_do_free(whlist, au_nhash_wh_do_free);
29735 +}
29736 +
29737 +static void au_nhash_de_free(struct au_nhash *delist)
29738 +{
29739 +       au_nhash_do_free(delist, au_nhash_de_do_free);
29740 +}
29741 +
29742 +/* ---------------------------------------------------------------------- */
29743 +
29744 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
29745 +                           int limit)
29746 +{
29747 +       int num;
29748 +       unsigned int u, n;
29749 +       struct hlist_head *head;
29750 +       struct au_vdir_wh *pos;
29751 +
29752 +       num = 0;
29753 +       n = whlist->nh_num;
29754 +       head = whlist->nh_head;
29755 +       for (u = 0; u < n; u++, head++)
29756 +               hlist_for_each_entry(pos, head, wh_hash)
29757 +                       if (pos->wh_bindex == btgt && ++num > limit)
29758 +                               return 1;
29759 +       return 0;
29760 +}
29761 +
29762 +static struct hlist_head *au_name_hash(struct au_nhash *nhash,
29763 +                                      unsigned char *name,
29764 +                                      unsigned int len)
29765 +{
29766 +       unsigned int v;
29767 +       /* const unsigned int magic_bit = 12; */
29768 +
29769 +       AuDebugOn(!nhash->nh_num || !nhash->nh_head);
29770 +
29771 +       v = 0;
29772 +       if (len > 8)
29773 +               len = 8;
29774 +       while (len--)
29775 +               v += *name++;
29776 +       /* v = hash_long(v, magic_bit); */
29777 +       v %= nhash->nh_num;
29778 +       return nhash->nh_head + v;
29779 +}
29780 +
29781 +static int au_nhash_test_name(struct au_vdir_destr *str, const char *name,
29782 +                             int nlen)
29783 +{
29784 +       return str->len == nlen && !memcmp(str->name, name, nlen);
29785 +}
29786 +
29787 +/* returns found or not */
29788 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen)
29789 +{
29790 +       struct hlist_head *head;
29791 +       struct au_vdir_wh *pos;
29792 +       struct au_vdir_destr *str;
29793 +
29794 +       head = au_name_hash(whlist, name, nlen);
29795 +       hlist_for_each_entry(pos, head, wh_hash) {
29796 +               str = &pos->wh_str;
29797 +               AuDbg("%.*s\n", str->len, str->name);
29798 +               if (au_nhash_test_name(str, name, nlen))
29799 +                       return 1;
29800 +       }
29801 +       return 0;
29802 +}
29803 +
29804 +/* returns found(true) or not */
29805 +static int test_known(struct au_nhash *delist, char *name, int nlen)
29806 +{
29807 +       struct hlist_head *head;
29808 +       struct au_vdir_dehstr *pos;
29809 +       struct au_vdir_destr *str;
29810 +
29811 +       head = au_name_hash(delist, name, nlen);
29812 +       hlist_for_each_entry(pos, head, hash) {
29813 +               str = pos->str;
29814 +               AuDbg("%.*s\n", str->len, str->name);
29815 +               if (au_nhash_test_name(str, name, nlen))
29816 +                       return 1;
29817 +       }
29818 +       return 0;
29819 +}
29820 +
29821 +static void au_shwh_init_wh(struct au_vdir_wh *wh, ino_t ino,
29822 +                           unsigned char d_type)
29823 +{
29824 +#ifdef CONFIG_AUFS_SHWH
29825 +       wh->wh_ino = ino;
29826 +       wh->wh_type = d_type;
29827 +#endif
29828 +}
29829 +
29830 +/* ---------------------------------------------------------------------- */
29831 +
29832 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
29833 +                      unsigned int d_type, aufs_bindex_t bindex,
29834 +                      unsigned char shwh)
29835 +{
29836 +       int err;
29837 +       struct au_vdir_destr *str;
29838 +       struct au_vdir_wh *wh;
29839 +
29840 +       AuDbg("%.*s\n", nlen, name);
29841 +       AuDebugOn(!whlist->nh_num || !whlist->nh_head);
29842 +
29843 +       err = -ENOMEM;
29844 +       wh = kmalloc(sizeof(*wh) + nlen, GFP_NOFS);
29845 +       if (unlikely(!wh))
29846 +               goto out;
29847 +
29848 +       err = 0;
29849 +       wh->wh_bindex = bindex;
29850 +       if (shwh)
29851 +               au_shwh_init_wh(wh, ino, d_type);
29852 +       str = &wh->wh_str;
29853 +       str->len = nlen;
29854 +       memcpy(str->name, name, nlen);
29855 +       hlist_add_head(&wh->wh_hash, au_name_hash(whlist, name, nlen));
29856 +       /* smp_mb(); */
29857 +
29858 +out:
29859 +       return err;
29860 +}
29861 +
29862 +static int append_deblk(struct au_vdir *vdir)
29863 +{
29864 +       int err;
29865 +       unsigned long ul;
29866 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
29867 +       union au_vdir_deblk_p p, deblk_end;
29868 +       unsigned char **o;
29869 +
29870 +       err = -ENOMEM;
29871 +       o = au_krealloc(vdir->vd_deblk, sizeof(*o) * (vdir->vd_nblk + 1),
29872 +                       GFP_NOFS, /*may_shrink*/0);
29873 +       if (unlikely(!o))
29874 +               goto out;
29875 +
29876 +       vdir->vd_deblk = o;
29877 +       p.deblk = kmalloc(deblk_sz, GFP_NOFS);
29878 +       if (p.deblk) {
29879 +               ul = vdir->vd_nblk++;
29880 +               vdir->vd_deblk[ul] = p.deblk;
29881 +               vdir->vd_last.ul = ul;
29882 +               vdir->vd_last.p.deblk = p.deblk;
29883 +               deblk_end.deblk = p.deblk + deblk_sz;
29884 +               err = set_deblk_end(&p, &deblk_end);
29885 +       }
29886 +
29887 +out:
29888 +       return err;
29889 +}
29890 +
29891 +static int append_de(struct au_vdir *vdir, char *name, int nlen, ino_t ino,
29892 +                    unsigned int d_type, struct au_nhash *delist)
29893 +{
29894 +       int err;
29895 +       unsigned int sz;
29896 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
29897 +       union au_vdir_deblk_p p, *room, deblk_end;
29898 +       struct au_vdir_dehstr *dehstr;
29899 +
29900 +       p.deblk = last_deblk(vdir);
29901 +       deblk_end.deblk = p.deblk + deblk_sz;
29902 +       room = &vdir->vd_last.p;
29903 +       AuDebugOn(room->deblk < p.deblk || deblk_end.deblk <= room->deblk
29904 +                 || !is_deblk_end(room, &deblk_end));
29905 +
29906 +       sz = calc_size(nlen);
29907 +       if (unlikely(sz > deblk_end.deblk - room->deblk)) {
29908 +               err = append_deblk(vdir);
29909 +               if (unlikely(err))
29910 +                       goto out;
29911 +
29912 +               p.deblk = last_deblk(vdir);
29913 +               deblk_end.deblk = p.deblk + deblk_sz;
29914 +               /* smp_mb(); */
29915 +               AuDebugOn(room->deblk != p.deblk);
29916 +       }
29917 +
29918 +       err = -ENOMEM;
29919 +       dehstr = au_cache_alloc_vdir_dehstr();
29920 +       if (unlikely(!dehstr))
29921 +               goto out;
29922 +
29923 +       dehstr->str = &room->de->de_str;
29924 +       hlist_add_head(&dehstr->hash, au_name_hash(delist, name, nlen));
29925 +       room->de->de_ino = ino;
29926 +       room->de->de_type = d_type;
29927 +       room->de->de_str.len = nlen;
29928 +       memcpy(room->de->de_str.name, name, nlen);
29929 +
29930 +       err = 0;
29931 +       room->deblk += sz;
29932 +       if (unlikely(set_deblk_end(room, &deblk_end)))
29933 +               err = append_deblk(vdir);
29934 +       /* smp_mb(); */
29935 +
29936 +out:
29937 +       return err;
29938 +}
29939 +
29940 +/* ---------------------------------------------------------------------- */
29941 +
29942 +void au_vdir_free(struct au_vdir *vdir, int atonce)
29943 +{
29944 +       unsigned char **deblk;
29945 +
29946 +       deblk = vdir->vd_deblk;
29947 +       if (!atonce) {
29948 +               while (vdir->vd_nblk--)
29949 +                       au_delayed_kfree(*deblk++);
29950 +               au_delayed_kfree(vdir->vd_deblk);
29951 +               au_cache_dfree_vdir(vdir);
29952 +       } else {
29953 +               /* not delayed */
29954 +               while (vdir->vd_nblk--)
29955 +                       kfree(*deblk++);
29956 +               kfree(vdir->vd_deblk);
29957 +               au_cache_free_vdir(vdir);
29958 +       }
29959 +}
29960 +
29961 +static struct au_vdir *alloc_vdir(struct file *file)
29962 +{
29963 +       struct au_vdir *vdir;
29964 +       struct super_block *sb;
29965 +       int err;
29966 +
29967 +       sb = file->f_path.dentry->d_sb;
29968 +       SiMustAnyLock(sb);
29969 +
29970 +       err = -ENOMEM;
29971 +       vdir = au_cache_alloc_vdir();
29972 +       if (unlikely(!vdir))
29973 +               goto out;
29974 +
29975 +       vdir->vd_deblk = kzalloc(sizeof(*vdir->vd_deblk), GFP_NOFS);
29976 +       if (unlikely(!vdir->vd_deblk))
29977 +               goto out_free;
29978 +
29979 +       vdir->vd_deblk_sz = au_sbi(sb)->si_rdblk;
29980 +       if (!vdir->vd_deblk_sz) {
29981 +               /* estimate the appropriate size for deblk */
29982 +               vdir->vd_deblk_sz = au_dir_size(file, /*dentry*/NULL);
29983 +               /* pr_info("vd_deblk_sz %u\n", vdir->vd_deblk_sz); */
29984 +       }
29985 +       vdir->vd_nblk = 0;
29986 +       vdir->vd_version = 0;
29987 +       vdir->vd_jiffy = 0;
29988 +       err = append_deblk(vdir);
29989 +       if (!err)
29990 +               return vdir; /* success */
29991 +
29992 +       au_delayed_kfree(vdir->vd_deblk);
29993 +
29994 +out_free:
29995 +       au_cache_dfree_vdir(vdir);
29996 +out:
29997 +       vdir = ERR_PTR(err);
29998 +       return vdir;
29999 +}
30000 +
30001 +static int reinit_vdir(struct au_vdir *vdir)
30002 +{
30003 +       int err;
30004 +       union au_vdir_deblk_p p, deblk_end;
30005 +
30006 +       while (vdir->vd_nblk > 1) {
30007 +               au_delayed_kfree(vdir->vd_deblk[vdir->vd_nblk - 1]);
30008 +               /* vdir->vd_deblk[vdir->vd_nblk - 1] = NULL; */
30009 +               vdir->vd_nblk--;
30010 +       }
30011 +       p.deblk = vdir->vd_deblk[0];
30012 +       deblk_end.deblk = p.deblk + vdir->vd_deblk_sz;
30013 +       err = set_deblk_end(&p, &deblk_end);
30014 +       /* keep vd_dblk_sz */
30015 +       vdir->vd_last.ul = 0;
30016 +       vdir->vd_last.p.deblk = vdir->vd_deblk[0];
30017 +       vdir->vd_version = 0;
30018 +       vdir->vd_jiffy = 0;
30019 +       /* smp_mb(); */
30020 +       return err;
30021 +}
30022 +
30023 +/* ---------------------------------------------------------------------- */
30024 +
30025 +#define AuFillVdir_CALLED      1
30026 +#define AuFillVdir_WHABLE      (1 << 1)
30027 +#define AuFillVdir_SHWH                (1 << 2)
30028 +#define au_ftest_fillvdir(flags, name) ((flags) & AuFillVdir_##name)
30029 +#define au_fset_fillvdir(flags, name) \
30030 +       do { (flags) |= AuFillVdir_##name; } while (0)
30031 +#define au_fclr_fillvdir(flags, name) \
30032 +       do { (flags) &= ~AuFillVdir_##name; } while (0)
30033 +
30034 +#ifndef CONFIG_AUFS_SHWH
30035 +#undef AuFillVdir_SHWH
30036 +#define AuFillVdir_SHWH                0
30037 +#endif
30038 +
30039 +struct fillvdir_arg {
30040 +       struct dir_context      ctx;
30041 +       struct file             *file;
30042 +       struct au_vdir          *vdir;
30043 +       struct au_nhash         delist;
30044 +       struct au_nhash         whlist;
30045 +       aufs_bindex_t           bindex;
30046 +       unsigned int            flags;
30047 +       int                     err;
30048 +};
30049 +
30050 +static int fillvdir(struct dir_context *ctx, const char *__name, int nlen,
30051 +                   loff_t offset __maybe_unused, u64 h_ino,
30052 +                   unsigned int d_type)
30053 +{
30054 +       struct fillvdir_arg *arg = container_of(ctx, struct fillvdir_arg, ctx);
30055 +       char *name = (void *)__name;
30056 +       struct super_block *sb;
30057 +       ino_t ino;
30058 +       const unsigned char shwh = !!au_ftest_fillvdir(arg->flags, SHWH);
30059 +
30060 +       arg->err = 0;
30061 +       sb = arg->file->f_path.dentry->d_sb;
30062 +       au_fset_fillvdir(arg->flags, CALLED);
30063 +       /* smp_mb(); */
30064 +       if (nlen <= AUFS_WH_PFX_LEN
30065 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
30066 +               if (test_known(&arg->delist, name, nlen)
30067 +                   || au_nhash_test_known_wh(&arg->whlist, name, nlen))
30068 +                       goto out; /* already exists or whiteouted */
30069 +
30070 +               arg->err = au_ino(sb, arg->bindex, h_ino, d_type, &ino);
30071 +               if (!arg->err) {
30072 +                       if (unlikely(nlen > AUFS_MAX_NAMELEN))
30073 +                               d_type = DT_UNKNOWN;
30074 +                       arg->err = append_de(arg->vdir, name, nlen, ino,
30075 +                                            d_type, &arg->delist);
30076 +               }
30077 +       } else if (au_ftest_fillvdir(arg->flags, WHABLE)) {
30078 +               name += AUFS_WH_PFX_LEN;
30079 +               nlen -= AUFS_WH_PFX_LEN;
30080 +               if (au_nhash_test_known_wh(&arg->whlist, name, nlen))
30081 +                       goto out; /* already whiteouted */
30082 +
30083 +               if (shwh)
30084 +                       arg->err = au_wh_ino(sb, arg->bindex, h_ino, d_type,
30085 +                                            &ino);
30086 +               if (!arg->err) {
30087 +                       if (nlen <= AUFS_MAX_NAMELEN + AUFS_WH_PFX_LEN)
30088 +                               d_type = DT_UNKNOWN;
30089 +                       arg->err = au_nhash_append_wh
30090 +                               (&arg->whlist, name, nlen, ino, d_type,
30091 +                                arg->bindex, shwh);
30092 +               }
30093 +       }
30094 +
30095 +out:
30096 +       if (!arg->err)
30097 +               arg->vdir->vd_jiffy = jiffies;
30098 +       /* smp_mb(); */
30099 +       AuTraceErr(arg->err);
30100 +       return arg->err;
30101 +}
30102 +
30103 +static int au_handle_shwh(struct super_block *sb, struct au_vdir *vdir,
30104 +                         struct au_nhash *whlist, struct au_nhash *delist)
30105 +{
30106 +#ifdef CONFIG_AUFS_SHWH
30107 +       int err;
30108 +       unsigned int nh, u;
30109 +       struct hlist_head *head;
30110 +       struct au_vdir_wh *pos;
30111 +       struct hlist_node *n;
30112 +       char *p, *o;
30113 +       struct au_vdir_destr *destr;
30114 +
30115 +       AuDebugOn(!au_opt_test(au_mntflags(sb), SHWH));
30116 +
30117 +       err = -ENOMEM;
30118 +       o = p = (void *)__get_free_page(GFP_NOFS);
30119 +       if (unlikely(!p))
30120 +               goto out;
30121 +
30122 +       err = 0;
30123 +       nh = whlist->nh_num;
30124 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
30125 +       p += AUFS_WH_PFX_LEN;
30126 +       for (u = 0; u < nh; u++) {
30127 +               head = whlist->nh_head + u;
30128 +               hlist_for_each_entry_safe(pos, n, head, wh_hash) {
30129 +                       destr = &pos->wh_str;
30130 +                       memcpy(p, destr->name, destr->len);
30131 +                       err = append_de(vdir, o, destr->len + AUFS_WH_PFX_LEN,
30132 +                                       pos->wh_ino, pos->wh_type, delist);
30133 +                       if (unlikely(err))
30134 +                               break;
30135 +               }
30136 +       }
30137 +
30138 +       au_delayed_free_page((unsigned long)o);
30139 +
30140 +out:
30141 +       AuTraceErr(err);
30142 +       return err;
30143 +#else
30144 +       return 0;
30145 +#endif
30146 +}
30147 +
30148 +static int au_do_read_vdir(struct fillvdir_arg *arg)
30149 +{
30150 +       int err;
30151 +       unsigned int rdhash;
30152 +       loff_t offset;
30153 +       aufs_bindex_t bbot, bindex, btop;
30154 +       unsigned char shwh;
30155 +       struct file *hf, *file;
30156 +       struct super_block *sb;
30157 +
30158 +       file = arg->file;
30159 +       sb = file->f_path.dentry->d_sb;
30160 +       SiMustAnyLock(sb);
30161 +
30162 +       rdhash = au_sbi(sb)->si_rdhash;
30163 +       if (!rdhash)
30164 +               rdhash = au_rdhash_est(au_dir_size(file, /*dentry*/NULL));
30165 +       err = au_nhash_alloc(&arg->delist, rdhash, GFP_NOFS);
30166 +       if (unlikely(err))
30167 +               goto out;
30168 +       err = au_nhash_alloc(&arg->whlist, rdhash, GFP_NOFS);
30169 +       if (unlikely(err))
30170 +               goto out_delist;
30171 +
30172 +       err = 0;
30173 +       arg->flags = 0;
30174 +       shwh = 0;
30175 +       if (au_opt_test(au_mntflags(sb), SHWH)) {
30176 +               shwh = 1;
30177 +               au_fset_fillvdir(arg->flags, SHWH);
30178 +       }
30179 +       btop = au_fbtop(file);
30180 +       bbot = au_fbbot_dir(file);
30181 +       for (bindex = btop; !err && bindex <= bbot; bindex++) {
30182 +               hf = au_hf_dir(file, bindex);
30183 +               if (!hf)
30184 +                       continue;
30185 +
30186 +               offset = vfsub_llseek(hf, 0, SEEK_SET);
30187 +               err = offset;
30188 +               if (unlikely(offset))
30189 +                       break;
30190 +
30191 +               arg->bindex = bindex;
30192 +               au_fclr_fillvdir(arg->flags, WHABLE);
30193 +               if (shwh
30194 +                   || (bindex != bbot
30195 +                       && au_br_whable(au_sbr_perm(sb, bindex))))
30196 +                       au_fset_fillvdir(arg->flags, WHABLE);
30197 +               do {
30198 +                       arg->err = 0;
30199 +                       au_fclr_fillvdir(arg->flags, CALLED);
30200 +                       /* smp_mb(); */
30201 +                       err = vfsub_iterate_dir(hf, &arg->ctx);
30202 +                       if (err >= 0)
30203 +                               err = arg->err;
30204 +               } while (!err && au_ftest_fillvdir(arg->flags, CALLED));
30205 +
30206 +               /*
30207 +                * dir_relax() may be good for concurrency, but aufs should not
30208 +                * use it since it will cause a lockdep problem.
30209 +                */
30210 +       }
30211 +
30212 +       if (!err && shwh)
30213 +               err = au_handle_shwh(sb, arg->vdir, &arg->whlist, &arg->delist);
30214 +
30215 +       au_nhash_wh_free(&arg->whlist);
30216 +
30217 +out_delist:
30218 +       au_nhash_de_free(&arg->delist);
30219 +out:
30220 +       return err;
30221 +}
30222 +
30223 +static int read_vdir(struct file *file, int may_read)
30224 +{
30225 +       int err;
30226 +       unsigned long expire;
30227 +       unsigned char do_read;
30228 +       struct fillvdir_arg arg = {
30229 +               .ctx = {
30230 +                       .actor = fillvdir
30231 +               }
30232 +       };
30233 +       struct inode *inode;
30234 +       struct au_vdir *vdir, *allocated;
30235 +
30236 +       err = 0;
30237 +       inode = file_inode(file);
30238 +       IMustLock(inode);
30239 +       IiMustWriteLock(inode);
30240 +       SiMustAnyLock(inode->i_sb);
30241 +
30242 +       allocated = NULL;
30243 +       do_read = 0;
30244 +       expire = au_sbi(inode->i_sb)->si_rdcache;
30245 +       vdir = au_ivdir(inode);
30246 +       if (!vdir) {
30247 +               do_read = 1;
30248 +               vdir = alloc_vdir(file);
30249 +               err = PTR_ERR(vdir);
30250 +               if (IS_ERR(vdir))
30251 +                       goto out;
30252 +               err = 0;
30253 +               allocated = vdir;
30254 +       } else if (may_read
30255 +                  && (inode->i_version != vdir->vd_version
30256 +                      || time_after(jiffies, vdir->vd_jiffy + expire))) {
30257 +               do_read = 1;
30258 +               err = reinit_vdir(vdir);
30259 +               if (unlikely(err))
30260 +                       goto out;
30261 +       }
30262 +
30263 +       if (!do_read)
30264 +               return 0; /* success */
30265 +
30266 +       arg.file = file;
30267 +       arg.vdir = vdir;
30268 +       err = au_do_read_vdir(&arg);
30269 +       if (!err) {
30270 +               /* file->f_pos = 0; */ /* todo: ctx->pos? */
30271 +               vdir->vd_version = inode->i_version;
30272 +               vdir->vd_last.ul = 0;
30273 +               vdir->vd_last.p.deblk = vdir->vd_deblk[0];
30274 +               if (allocated)
30275 +                       au_set_ivdir(inode, allocated);
30276 +       } else if (allocated)
30277 +               au_vdir_free(allocated, /*atonce*/0);
30278 +
30279 +out:
30280 +       return err;
30281 +}
30282 +
30283 +static int copy_vdir(struct au_vdir *tgt, struct au_vdir *src)
30284 +{
30285 +       int err, rerr;
30286 +       unsigned long ul, n;
30287 +       const unsigned int deblk_sz = src->vd_deblk_sz;
30288 +
30289 +       AuDebugOn(tgt->vd_nblk != 1);
30290 +
30291 +       err = -ENOMEM;
30292 +       if (tgt->vd_nblk < src->vd_nblk) {
30293 +               unsigned char **p;
30294 +
30295 +               p = au_krealloc(tgt->vd_deblk, sizeof(*p) * src->vd_nblk,
30296 +                               GFP_NOFS, /*may_shrink*/0);
30297 +               if (unlikely(!p))
30298 +                       goto out;
30299 +               tgt->vd_deblk = p;
30300 +       }
30301 +
30302 +       if (tgt->vd_deblk_sz != deblk_sz) {
30303 +               unsigned char *p;
30304 +
30305 +               tgt->vd_deblk_sz = deblk_sz;
30306 +               p = au_krealloc(tgt->vd_deblk[0], deblk_sz, GFP_NOFS,
30307 +                               /*may_shrink*/1);
30308 +               if (unlikely(!p))
30309 +                       goto out;
30310 +               tgt->vd_deblk[0] = p;
30311 +       }
30312 +       memcpy(tgt->vd_deblk[0], src->vd_deblk[0], deblk_sz);
30313 +       tgt->vd_version = src->vd_version;
30314 +       tgt->vd_jiffy = src->vd_jiffy;
30315 +
30316 +       n = src->vd_nblk;
30317 +       for (ul = 1; ul < n; ul++) {
30318 +               tgt->vd_deblk[ul] = kmemdup(src->vd_deblk[ul], deblk_sz,
30319 +                                           GFP_NOFS);
30320 +               if (unlikely(!tgt->vd_deblk[ul]))
30321 +                       goto out;
30322 +               tgt->vd_nblk++;
30323 +       }
30324 +       tgt->vd_nblk = n;
30325 +       tgt->vd_last.ul = tgt->vd_last.ul;
30326 +       tgt->vd_last.p.deblk = tgt->vd_deblk[tgt->vd_last.ul];
30327 +       tgt->vd_last.p.deblk += src->vd_last.p.deblk
30328 +               - src->vd_deblk[src->vd_last.ul];
30329 +       /* smp_mb(); */
30330 +       return 0; /* success */
30331 +
30332 +out:
30333 +       rerr = reinit_vdir(tgt);
30334 +       BUG_ON(rerr);
30335 +       return err;
30336 +}
30337 +
30338 +int au_vdir_init(struct file *file)
30339 +{
30340 +       int err;
30341 +       struct inode *inode;
30342 +       struct au_vdir *vdir_cache, *allocated;
30343 +
30344 +       /* test file->f_pos here instead of ctx->pos */
30345 +       err = read_vdir(file, !file->f_pos);
30346 +       if (unlikely(err))
30347 +               goto out;
30348 +
30349 +       allocated = NULL;
30350 +       vdir_cache = au_fvdir_cache(file);
30351 +       if (!vdir_cache) {
30352 +               vdir_cache = alloc_vdir(file);
30353 +               err = PTR_ERR(vdir_cache);
30354 +               if (IS_ERR(vdir_cache))
30355 +                       goto out;
30356 +               allocated = vdir_cache;
30357 +       } else if (!file->f_pos && vdir_cache->vd_version != file->f_version) {
30358 +               /* test file->f_pos here instead of ctx->pos */
30359 +               err = reinit_vdir(vdir_cache);
30360 +               if (unlikely(err))
30361 +                       goto out;
30362 +       } else
30363 +               return 0; /* success */
30364 +
30365 +       inode = file_inode(file);
30366 +       err = copy_vdir(vdir_cache, au_ivdir(inode));
30367 +       if (!err) {
30368 +               file->f_version = inode->i_version;
30369 +               if (allocated)
30370 +                       au_set_fvdir_cache(file, allocated);
30371 +       } else if (allocated)
30372 +               au_vdir_free(allocated, /*atonce*/0);
30373 +
30374 +out:
30375 +       return err;
30376 +}
30377 +
30378 +static loff_t calc_offset(struct au_vdir *vdir)
30379 +{
30380 +       loff_t offset;
30381 +       union au_vdir_deblk_p p;
30382 +
30383 +       p.deblk = vdir->vd_deblk[vdir->vd_last.ul];
30384 +       offset = vdir->vd_last.p.deblk - p.deblk;
30385 +       offset += vdir->vd_deblk_sz * vdir->vd_last.ul;
30386 +       return offset;
30387 +}
30388 +
30389 +/* returns true or false */
30390 +static int seek_vdir(struct file *file, struct dir_context *ctx)
30391 +{
30392 +       int valid;
30393 +       unsigned int deblk_sz;
30394 +       unsigned long ul, n;
30395 +       loff_t offset;
30396 +       union au_vdir_deblk_p p, deblk_end;
30397 +       struct au_vdir *vdir_cache;
30398 +
30399 +       valid = 1;
30400 +       vdir_cache = au_fvdir_cache(file);
30401 +       offset = calc_offset(vdir_cache);
30402 +       AuDbg("offset %lld\n", offset);
30403 +       if (ctx->pos == offset)
30404 +               goto out;
30405 +
30406 +       vdir_cache->vd_last.ul = 0;
30407 +       vdir_cache->vd_last.p.deblk = vdir_cache->vd_deblk[0];
30408 +       if (!ctx->pos)
30409 +               goto out;
30410 +
30411 +       valid = 0;
30412 +       deblk_sz = vdir_cache->vd_deblk_sz;
30413 +       ul = div64_u64(ctx->pos, deblk_sz);
30414 +       AuDbg("ul %lu\n", ul);
30415 +       if (ul >= vdir_cache->vd_nblk)
30416 +               goto out;
30417 +
30418 +       n = vdir_cache->vd_nblk;
30419 +       for (; ul < n; ul++) {
30420 +               p.deblk = vdir_cache->vd_deblk[ul];
30421 +               deblk_end.deblk = p.deblk + deblk_sz;
30422 +               offset = ul;
30423 +               offset *= deblk_sz;
30424 +               while (!is_deblk_end(&p, &deblk_end) && offset < ctx->pos) {
30425 +                       unsigned int l;
30426 +
30427 +                       l = calc_size(p.de->de_str.len);
30428 +                       offset += l;
30429 +                       p.deblk += l;
30430 +               }
30431 +               if (!is_deblk_end(&p, &deblk_end)) {
30432 +                       valid = 1;
30433 +                       vdir_cache->vd_last.ul = ul;
30434 +                       vdir_cache->vd_last.p = p;
30435 +                       break;
30436 +               }
30437 +       }
30438 +
30439 +out:
30440 +       /* smp_mb(); */
30441 +       AuTraceErr(!valid);
30442 +       return valid;
30443 +}
30444 +
30445 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx)
30446 +{
30447 +       unsigned int l, deblk_sz;
30448 +       union au_vdir_deblk_p deblk_end;
30449 +       struct au_vdir *vdir_cache;
30450 +       struct au_vdir_de *de;
30451 +
30452 +       vdir_cache = au_fvdir_cache(file);
30453 +       if (!seek_vdir(file, ctx))
30454 +               return 0;
30455 +
30456 +       deblk_sz = vdir_cache->vd_deblk_sz;
30457 +       while (1) {
30458 +               deblk_end.deblk = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
30459 +               deblk_end.deblk += deblk_sz;
30460 +               while (!is_deblk_end(&vdir_cache->vd_last.p, &deblk_end)) {
30461 +                       de = vdir_cache->vd_last.p.de;
30462 +                       AuDbg("%.*s, off%lld, i%lu, dt%d\n",
30463 +                             de->de_str.len, de->de_str.name, ctx->pos,
30464 +                             (unsigned long)de->de_ino, de->de_type);
30465 +                       if (unlikely(!dir_emit(ctx, de->de_str.name,
30466 +                                              de->de_str.len, de->de_ino,
30467 +                                              de->de_type))) {
30468 +                               /* todo: ignore the error caused by udba? */
30469 +                               /* return err; */
30470 +                               return 0;
30471 +                       }
30472 +
30473 +                       l = calc_size(de->de_str.len);
30474 +                       vdir_cache->vd_last.p.deblk += l;
30475 +                       ctx->pos += l;
30476 +               }
30477 +               if (vdir_cache->vd_last.ul < vdir_cache->vd_nblk - 1) {
30478 +                       vdir_cache->vd_last.ul++;
30479 +                       vdir_cache->vd_last.p.deblk
30480 +                               = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
30481 +                       ctx->pos = deblk_sz * vdir_cache->vd_last.ul;
30482 +                       continue;
30483 +               }
30484 +               break;
30485 +       }
30486 +
30487 +       /* smp_mb(); */
30488 +       return 0;
30489 +}
30490 diff -urN /usr/share/empty/fs/aufs/vfsub.c linux/fs/aufs/vfsub.c
30491 --- /usr/share/empty/fs/aufs/vfsub.c    1970-01-01 01:00:00.000000000 +0100
30492 +++ linux/fs/aufs/vfsub.c       2016-10-09 16:55:36.496035060 +0200
30493 @@ -0,0 +1,884 @@
30494 +/*
30495 + * Copyright (C) 2005-2016 Junjiro R. Okajima
30496 + *
30497 + * This program, aufs is free software; you can redistribute it and/or modify
30498 + * it under the terms of the GNU General Public License as published by
30499 + * the Free Software Foundation; either version 2 of the License, or
30500 + * (at your option) any later version.
30501 + *
30502 + * This program is distributed in the hope that it will be useful,
30503 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30504 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30505 + * GNU General Public License for more details.
30506 + *
30507 + * You should have received a copy of the GNU General Public License
30508 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30509 + */
30510 +
30511 +/*
30512 + * sub-routines for VFS
30513 + */
30514 +
30515 +#include <linux/namei.h>
30516 +#include <linux/nsproxy.h>
30517 +#include <linux/security.h>
30518 +#include <linux/splice.h>
30519 +#include "../fs/mount.h"
30520 +#include "aufs.h"
30521 +
30522 +#ifdef CONFIG_AUFS_BR_FUSE
30523 +int vfsub_test_mntns(struct vfsmount *mnt, struct super_block *h_sb)
30524 +{
30525 +       struct nsproxy *ns;
30526 +
30527 +       if (!au_test_fuse(h_sb) || !au_userns)
30528 +               return 0;
30529 +
30530 +       ns = current->nsproxy;
30531 +       /* no {get,put}_nsproxy(ns) */
30532 +       return real_mount(mnt)->mnt_ns == ns->mnt_ns ? 0 : -EACCES;
30533 +}
30534 +#endif
30535 +
30536 +/* ---------------------------------------------------------------------- */
30537 +
30538 +int vfsub_update_h_iattr(struct path *h_path, int *did)
30539 +{
30540 +       int err;
30541 +       struct kstat st;
30542 +       struct super_block *h_sb;
30543 +
30544 +       /* for remote fs, leave work for its getattr or d_revalidate */
30545 +       /* for bad i_attr fs, handle them in aufs_getattr() */
30546 +       /* still some fs may acquire i_mutex. we need to skip them */
30547 +       err = 0;
30548 +       if (!did)
30549 +               did = &err;
30550 +       h_sb = h_path->dentry->d_sb;
30551 +       *did = (!au_test_fs_remote(h_sb) && au_test_fs_refresh_iattr(h_sb));
30552 +       if (*did)
30553 +               err = vfs_getattr(h_path, &st);
30554 +
30555 +       return err;
30556 +}
30557 +
30558 +/* ---------------------------------------------------------------------- */
30559 +
30560 +struct file *vfsub_dentry_open(struct path *path, int flags)
30561 +{
30562 +       struct file *file;
30563 +
30564 +       file = dentry_open(path, flags /* | __FMODE_NONOTIFY */,
30565 +                          current_cred());
30566 +       if (!IS_ERR_OR_NULL(file)
30567 +           && (file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
30568 +               i_readcount_inc(d_inode(path->dentry));
30569 +
30570 +       return file;
30571 +}
30572 +
30573 +struct file *vfsub_filp_open(const char *path, int oflags, int mode)
30574 +{
30575 +       struct file *file;
30576 +
30577 +       lockdep_off();
30578 +       file = filp_open(path,
30579 +                        oflags /* | __FMODE_NONOTIFY */,
30580 +                        mode);
30581 +       lockdep_on();
30582 +       if (IS_ERR(file))
30583 +               goto out;
30584 +       vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30585 +
30586 +out:
30587 +       return file;
30588 +}
30589 +
30590 +/*
30591 + * Ideally this function should call VFS:do_last() in order to keep all its
30592 + * checkings. But it is very hard for aufs to regenerate several VFS internal
30593 + * structure such as nameidata. This is a second (or third) best approach.
30594 + * cf. linux/fs/namei.c:do_last(), lookup_open() and atomic_open().
30595 + */
30596 +int vfsub_atomic_open(struct inode *dir, struct dentry *dentry,
30597 +                     struct vfsub_aopen_args *args, struct au_branch *br)
30598 +{
30599 +       int err;
30600 +       struct file *file = args->file;
30601 +       /* copied from linux/fs/namei.c:atomic_open() */
30602 +       struct dentry *const DENTRY_NOT_SET = (void *)-1UL;
30603 +
30604 +       IMustLock(dir);
30605 +       AuDebugOn(!dir->i_op->atomic_open);
30606 +
30607 +       err = au_br_test_oflag(args->open_flag, br);
30608 +       if (unlikely(err))
30609 +               goto out;
30610 +
30611 +       args->file->f_path.dentry = DENTRY_NOT_SET;
30612 +       args->file->f_path.mnt = au_br_mnt(br);
30613 +       err = dir->i_op->atomic_open(dir, dentry, file, args->open_flag,
30614 +                                    args->create_mode, args->opened);
30615 +       if (err >= 0) {
30616 +               /* some filesystems don't set FILE_CREATED while succeeded? */
30617 +               if (*args->opened & FILE_CREATED)
30618 +                       fsnotify_create(dir, dentry);
30619 +       } else
30620 +               goto out;
30621 +
30622 +
30623 +       if (!err) {
30624 +               /* todo: call VFS:may_open() here */
30625 +               err = open_check_o_direct(file);
30626 +               /* todo: ima_file_check() too? */
30627 +               if (!err && (args->open_flag & __FMODE_EXEC))
30628 +                       err = deny_write_access(file);
30629 +               if (unlikely(err))
30630 +                       /* note that the file is created and still opened */
30631 +                       goto out;
30632 +       }
30633 +
30634 +       au_br_get(br);
30635 +       fsnotify_open(file);
30636 +
30637 +out:
30638 +       return err;
30639 +}
30640 +
30641 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path)
30642 +{
30643 +       int err;
30644 +
30645 +       err = kern_path(name, flags, path);
30646 +       if (!err && d_is_positive(path->dentry))
30647 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
30648 +       return err;
30649 +}
30650 +
30651 +struct dentry *vfsub_lookup_one_len_unlocked(const char *name,
30652 +                                            struct dentry *parent, int len)
30653 +{
30654 +       struct path path = {
30655 +               .mnt = NULL
30656 +       };
30657 +
30658 +       path.dentry = lookup_one_len_unlocked(name, parent, len);
30659 +       if (IS_ERR(path.dentry))
30660 +               goto out;
30661 +       if (d_is_positive(path.dentry))
30662 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
30663 +
30664 +out:
30665 +       AuTraceErrPtr(path.dentry);
30666 +       return path.dentry;
30667 +}
30668 +
30669 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
30670 +                                   int len)
30671 +{
30672 +       struct path path = {
30673 +               .mnt = NULL
30674 +       };
30675 +
30676 +       /* VFS checks it too, but by WARN_ON_ONCE() */
30677 +       IMustLock(d_inode(parent));
30678 +
30679 +       path.dentry = lookup_one_len(name, parent, len);
30680 +       if (IS_ERR(path.dentry))
30681 +               goto out;
30682 +       if (d_is_positive(path.dentry))
30683 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
30684 +
30685 +out:
30686 +       AuTraceErrPtr(path.dentry);
30687 +       return path.dentry;
30688 +}
30689 +
30690 +void vfsub_call_lkup_one(void *args)
30691 +{
30692 +       struct vfsub_lkup_one_args *a = args;
30693 +       *a->errp = vfsub_lkup_one(a->name, a->parent);
30694 +}
30695 +
30696 +/* ---------------------------------------------------------------------- */
30697 +
30698 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
30699 +                                struct dentry *d2, struct au_hinode *hdir2)
30700 +{
30701 +       struct dentry *d;
30702 +
30703 +       lockdep_off();
30704 +       d = lock_rename(d1, d2);
30705 +       lockdep_on();
30706 +       au_hn_suspend(hdir1);
30707 +       if (hdir1 != hdir2)
30708 +               au_hn_suspend(hdir2);
30709 +
30710 +       return d;
30711 +}
30712 +
30713 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
30714 +                        struct dentry *d2, struct au_hinode *hdir2)
30715 +{
30716 +       au_hn_resume(hdir1);
30717 +       if (hdir1 != hdir2)
30718 +               au_hn_resume(hdir2);
30719 +       lockdep_off();
30720 +       unlock_rename(d1, d2);
30721 +       lockdep_on();
30722 +}
30723 +
30724 +/* ---------------------------------------------------------------------- */
30725 +
30726 +int vfsub_create(struct inode *dir, struct path *path, int mode, bool want_excl)
30727 +{
30728 +       int err;
30729 +       struct dentry *d;
30730 +
30731 +       IMustLock(dir);
30732 +
30733 +       d = path->dentry;
30734 +       path->dentry = d->d_parent;
30735 +       err = security_path_mknod(path, d, mode, 0);
30736 +       path->dentry = d;
30737 +       if (unlikely(err))
30738 +               goto out;
30739 +
30740 +       lockdep_off();
30741 +       err = vfs_create(dir, path->dentry, mode, want_excl);
30742 +       lockdep_on();
30743 +       if (!err) {
30744 +               struct path tmp = *path;
30745 +               int did;
30746 +
30747 +               vfsub_update_h_iattr(&tmp, &did);
30748 +               if (did) {
30749 +                       tmp.dentry = path->dentry->d_parent;
30750 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30751 +               }
30752 +               /*ignore*/
30753 +       }
30754 +
30755 +out:
30756 +       return err;
30757 +}
30758 +
30759 +int vfsub_symlink(struct inode *dir, struct path *path, const char *symname)
30760 +{
30761 +       int err;
30762 +       struct dentry *d;
30763 +
30764 +       IMustLock(dir);
30765 +
30766 +       d = path->dentry;
30767 +       path->dentry = d->d_parent;
30768 +       err = security_path_symlink(path, d, symname);
30769 +       path->dentry = d;
30770 +       if (unlikely(err))
30771 +               goto out;
30772 +
30773 +       lockdep_off();
30774 +       err = vfs_symlink(dir, path->dentry, symname);
30775 +       lockdep_on();
30776 +       if (!err) {
30777 +               struct path tmp = *path;
30778 +               int did;
30779 +
30780 +               vfsub_update_h_iattr(&tmp, &did);
30781 +               if (did) {
30782 +                       tmp.dentry = path->dentry->d_parent;
30783 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30784 +               }
30785 +               /*ignore*/
30786 +       }
30787 +
30788 +out:
30789 +       return err;
30790 +}
30791 +
30792 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev)
30793 +{
30794 +       int err;
30795 +       struct dentry *d;
30796 +
30797 +       IMustLock(dir);
30798 +
30799 +       d = path->dentry;
30800 +       path->dentry = d->d_parent;
30801 +       err = security_path_mknod(path, d, mode, new_encode_dev(dev));
30802 +       path->dentry = d;
30803 +       if (unlikely(err))
30804 +               goto out;
30805 +
30806 +       lockdep_off();
30807 +       err = vfs_mknod(dir, path->dentry, mode, dev);
30808 +       lockdep_on();
30809 +       if (!err) {
30810 +               struct path tmp = *path;
30811 +               int did;
30812 +
30813 +               vfsub_update_h_iattr(&tmp, &did);
30814 +               if (did) {
30815 +                       tmp.dentry = path->dentry->d_parent;
30816 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30817 +               }
30818 +               /*ignore*/
30819 +       }
30820 +
30821 +out:
30822 +       return err;
30823 +}
30824 +
30825 +static int au_test_nlink(struct inode *inode)
30826 +{
30827 +       const unsigned int link_max = UINT_MAX >> 1; /* rough margin */
30828 +
30829 +       if (!au_test_fs_no_limit_nlink(inode->i_sb)
30830 +           || inode->i_nlink < link_max)
30831 +               return 0;
30832 +       return -EMLINK;
30833 +}
30834 +
30835 +int vfsub_link(struct dentry *src_dentry, struct inode *dir, struct path *path,
30836 +              struct inode **delegated_inode)
30837 +{
30838 +       int err;
30839 +       struct dentry *d;
30840 +
30841 +       IMustLock(dir);
30842 +
30843 +       err = au_test_nlink(d_inode(src_dentry));
30844 +       if (unlikely(err))
30845 +               return err;
30846 +
30847 +       /* we don't call may_linkat() */
30848 +       d = path->dentry;
30849 +       path->dentry = d->d_parent;
30850 +       err = security_path_link(src_dentry, path, d);
30851 +       path->dentry = d;
30852 +       if (unlikely(err))
30853 +               goto out;
30854 +
30855 +       lockdep_off();
30856 +       err = vfs_link(src_dentry, dir, path->dentry, delegated_inode);
30857 +       lockdep_on();
30858 +       if (!err) {
30859 +               struct path tmp = *path;
30860 +               int did;
30861 +
30862 +               /* fuse has different memory inode for the same inumber */
30863 +               vfsub_update_h_iattr(&tmp, &did);
30864 +               if (did) {
30865 +                       tmp.dentry = path->dentry->d_parent;
30866 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30867 +                       tmp.dentry = src_dentry;
30868 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30869 +               }
30870 +               /*ignore*/
30871 +       }
30872 +
30873 +out:
30874 +       return err;
30875 +}
30876 +
30877 +int vfsub_rename(struct inode *src_dir, struct dentry *src_dentry,
30878 +                struct inode *dir, struct path *path,
30879 +                struct inode **delegated_inode)
30880 +{
30881 +       int err;
30882 +       struct path tmp = {
30883 +               .mnt    = path->mnt
30884 +       };
30885 +       struct dentry *d;
30886 +
30887 +       IMustLock(dir);
30888 +       IMustLock(src_dir);
30889 +
30890 +       d = path->dentry;
30891 +       path->dentry = d->d_parent;
30892 +       tmp.dentry = src_dentry->d_parent;
30893 +       err = security_path_rename(&tmp, src_dentry, path, d, /*flags*/0);
30894 +       path->dentry = d;
30895 +       if (unlikely(err))
30896 +               goto out;
30897 +
30898 +       lockdep_off();
30899 +       err = vfs_rename(src_dir, src_dentry, dir, path->dentry,
30900 +                        delegated_inode, /*flags*/0);
30901 +       lockdep_on();
30902 +       if (!err) {
30903 +               int did;
30904 +
30905 +               tmp.dentry = d->d_parent;
30906 +               vfsub_update_h_iattr(&tmp, &did);
30907 +               if (did) {
30908 +                       tmp.dentry = src_dentry;
30909 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30910 +                       tmp.dentry = src_dentry->d_parent;
30911 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30912 +               }
30913 +               /*ignore*/
30914 +       }
30915 +
30916 +out:
30917 +       return err;
30918 +}
30919 +
30920 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode)
30921 +{
30922 +       int err;
30923 +       struct dentry *d;
30924 +
30925 +       IMustLock(dir);
30926 +
30927 +       d = path->dentry;
30928 +       path->dentry = d->d_parent;
30929 +       err = security_path_mkdir(path, d, mode);
30930 +       path->dentry = d;
30931 +       if (unlikely(err))
30932 +               goto out;
30933 +
30934 +       lockdep_off();
30935 +       err = vfs_mkdir(dir, path->dentry, mode);
30936 +       lockdep_on();
30937 +       if (!err) {
30938 +               struct path tmp = *path;
30939 +               int did;
30940 +
30941 +               vfsub_update_h_iattr(&tmp, &did);
30942 +               if (did) {
30943 +                       tmp.dentry = path->dentry->d_parent;
30944 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30945 +               }
30946 +               /*ignore*/
30947 +       }
30948 +
30949 +out:
30950 +       return err;
30951 +}
30952 +
30953 +int vfsub_rmdir(struct inode *dir, struct path *path)
30954 +{
30955 +       int err;
30956 +       struct dentry *d;
30957 +
30958 +       IMustLock(dir);
30959 +
30960 +       d = path->dentry;
30961 +       path->dentry = d->d_parent;
30962 +       err = security_path_rmdir(path, d);
30963 +       path->dentry = d;
30964 +       if (unlikely(err))
30965 +               goto out;
30966 +
30967 +       lockdep_off();
30968 +       err = vfs_rmdir(dir, path->dentry);
30969 +       lockdep_on();
30970 +       if (!err) {
30971 +               struct path tmp = {
30972 +                       .dentry = path->dentry->d_parent,
30973 +                       .mnt    = path->mnt
30974 +               };
30975 +
30976 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
30977 +       }
30978 +
30979 +out:
30980 +       return err;
30981 +}
30982 +
30983 +/* ---------------------------------------------------------------------- */
30984 +
30985 +/* todo: support mmap_sem? */
30986 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
30987 +                    loff_t *ppos)
30988 +{
30989 +       ssize_t err;
30990 +
30991 +       lockdep_off();
30992 +       err = vfs_read(file, ubuf, count, ppos);
30993 +       lockdep_on();
30994 +       if (err >= 0)
30995 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30996 +       return err;
30997 +}
30998 +
30999 +/* todo: kernel_read()? */
31000 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
31001 +                    loff_t *ppos)
31002 +{
31003 +       ssize_t err;
31004 +       mm_segment_t oldfs;
31005 +       union {
31006 +               void *k;
31007 +               char __user *u;
31008 +       } buf;
31009 +
31010 +       buf.k = kbuf;
31011 +       oldfs = get_fs();
31012 +       set_fs(KERNEL_DS);
31013 +       err = vfsub_read_u(file, buf.u, count, ppos);
31014 +       set_fs(oldfs);
31015 +       return err;
31016 +}
31017 +
31018 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
31019 +                     loff_t *ppos)
31020 +{
31021 +       ssize_t err;
31022 +
31023 +       lockdep_off();
31024 +       err = vfs_write(file, ubuf, count, ppos);
31025 +       lockdep_on();
31026 +       if (err >= 0)
31027 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
31028 +       return err;
31029 +}
31030 +
31031 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count, loff_t *ppos)
31032 +{
31033 +       ssize_t err;
31034 +       mm_segment_t oldfs;
31035 +       union {
31036 +               void *k;
31037 +               const char __user *u;
31038 +       } buf;
31039 +
31040 +       buf.k = kbuf;
31041 +       oldfs = get_fs();
31042 +       set_fs(KERNEL_DS);
31043 +       err = vfsub_write_u(file, buf.u, count, ppos);
31044 +       set_fs(oldfs);
31045 +       return err;
31046 +}
31047 +
31048 +int vfsub_flush(struct file *file, fl_owner_t id)
31049 +{
31050 +       int err;
31051 +
31052 +       err = 0;
31053 +       if (file->f_op->flush) {
31054 +               if (!au_test_nfs(file->f_path.dentry->d_sb))
31055 +                       err = file->f_op->flush(file, id);
31056 +               else {
31057 +                       lockdep_off();
31058 +                       err = file->f_op->flush(file, id);
31059 +                       lockdep_on();
31060 +               }
31061 +               if (!err)
31062 +                       vfsub_update_h_iattr(&file->f_path, /*did*/NULL);
31063 +               /*ignore*/
31064 +       }
31065 +       return err;
31066 +}
31067 +
31068 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx)
31069 +{
31070 +       int err;
31071 +
31072 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
31073 +
31074 +       lockdep_off();
31075 +       err = iterate_dir(file, ctx);
31076 +       lockdep_on();
31077 +       if (err >= 0)
31078 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
31079 +       return err;
31080 +}
31081 +
31082 +long vfsub_splice_to(struct file *in, loff_t *ppos,
31083 +                    struct pipe_inode_info *pipe, size_t len,
31084 +                    unsigned int flags)
31085 +{
31086 +       long err;
31087 +
31088 +       lockdep_off();
31089 +       err = do_splice_to(in, ppos, pipe, len, flags);
31090 +       lockdep_on();
31091 +       file_accessed(in);
31092 +       if (err >= 0)
31093 +               vfsub_update_h_iattr(&in->f_path, /*did*/NULL); /*ignore*/
31094 +       return err;
31095 +}
31096 +
31097 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
31098 +                      loff_t *ppos, size_t len, unsigned int flags)
31099 +{
31100 +       long err;
31101 +
31102 +       lockdep_off();
31103 +       err = do_splice_from(pipe, out, ppos, len, flags);
31104 +       lockdep_on();
31105 +       if (err >= 0)
31106 +               vfsub_update_h_iattr(&out->f_path, /*did*/NULL); /*ignore*/
31107 +       return err;
31108 +}
31109 +
31110 +int vfsub_fsync(struct file *file, struct path *path, int datasync)
31111 +{
31112 +       int err;
31113 +
31114 +       /* file can be NULL */
31115 +       lockdep_off();
31116 +       err = vfs_fsync(file, datasync);
31117 +       lockdep_on();
31118 +       if (!err) {
31119 +               if (!path) {
31120 +                       AuDebugOn(!file);
31121 +                       path = &file->f_path;
31122 +               }
31123 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
31124 +       }
31125 +       return err;
31126 +}
31127 +
31128 +/* cf. open.c:do_sys_truncate() and do_sys_ftruncate() */
31129 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
31130 +               struct file *h_file)
31131 +{
31132 +       int err;
31133 +       struct inode *h_inode;
31134 +       struct super_block *h_sb;
31135 +
31136 +       if (!h_file) {
31137 +               err = vfsub_truncate(h_path, length);
31138 +               goto out;
31139 +       }
31140 +
31141 +       h_inode = d_inode(h_path->dentry);
31142 +       h_sb = h_inode->i_sb;
31143 +       lockdep_off();
31144 +       sb_start_write(h_sb);
31145 +       lockdep_on();
31146 +       err = locks_verify_truncate(h_inode, h_file, length);
31147 +       if (!err)
31148 +               err = security_path_truncate(h_path);
31149 +       if (!err) {
31150 +               lockdep_off();
31151 +               err = do_truncate(h_path->dentry, length, attr, h_file);
31152 +               lockdep_on();
31153 +       }
31154 +       lockdep_off();
31155 +       sb_end_write(h_sb);
31156 +       lockdep_on();
31157 +
31158 +out:
31159 +       return err;
31160 +}
31161 +
31162 +/* ---------------------------------------------------------------------- */
31163 +
31164 +struct au_vfsub_mkdir_args {
31165 +       int *errp;
31166 +       struct inode *dir;
31167 +       struct path *path;
31168 +       int mode;
31169 +};
31170 +
31171 +static void au_call_vfsub_mkdir(void *args)
31172 +{
31173 +       struct au_vfsub_mkdir_args *a = args;
31174 +       *a->errp = vfsub_mkdir(a->dir, a->path, a->mode);
31175 +}
31176 +
31177 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode)
31178 +{
31179 +       int err, do_sio, wkq_err;
31180 +
31181 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
31182 +       if (!do_sio) {
31183 +               lockdep_off();
31184 +               err = vfsub_mkdir(dir, path, mode);
31185 +               lockdep_on();
31186 +       } else {
31187 +               struct au_vfsub_mkdir_args args = {
31188 +                       .errp   = &err,
31189 +                       .dir    = dir,
31190 +                       .path   = path,
31191 +                       .mode   = mode
31192 +               };
31193 +               wkq_err = au_wkq_wait(au_call_vfsub_mkdir, &args);
31194 +               if (unlikely(wkq_err))
31195 +                       err = wkq_err;
31196 +       }
31197 +
31198 +       return err;
31199 +}
31200 +
31201 +struct au_vfsub_rmdir_args {
31202 +       int *errp;
31203 +       struct inode *dir;
31204 +       struct path *path;
31205 +};
31206 +
31207 +static void au_call_vfsub_rmdir(void *args)
31208 +{
31209 +       struct au_vfsub_rmdir_args *a = args;
31210 +       *a->errp = vfsub_rmdir(a->dir, a->path);
31211 +}
31212 +
31213 +int vfsub_sio_rmdir(struct inode *dir, struct path *path)
31214 +{
31215 +       int err, do_sio, wkq_err;
31216 +
31217 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
31218 +       if (!do_sio) {
31219 +               lockdep_off();
31220 +               err = vfsub_rmdir(dir, path);
31221 +               lockdep_on();
31222 +       } else {
31223 +               struct au_vfsub_rmdir_args args = {
31224 +                       .errp   = &err,
31225 +                       .dir    = dir,
31226 +                       .path   = path
31227 +               };
31228 +               wkq_err = au_wkq_wait(au_call_vfsub_rmdir, &args);
31229 +               if (unlikely(wkq_err))
31230 +                       err = wkq_err;
31231 +       }
31232 +
31233 +       return err;
31234 +}
31235 +
31236 +/* ---------------------------------------------------------------------- */
31237 +
31238 +struct notify_change_args {
31239 +       int *errp;
31240 +       struct path *path;
31241 +       struct iattr *ia;
31242 +       struct inode **delegated_inode;
31243 +};
31244 +
31245 +static void call_notify_change(void *args)
31246 +{
31247 +       struct notify_change_args *a = args;
31248 +       struct inode *h_inode;
31249 +
31250 +       h_inode = d_inode(a->path->dentry);
31251 +       IMustLock(h_inode);
31252 +
31253 +       *a->errp = -EPERM;
31254 +       if (!IS_IMMUTABLE(h_inode) && !IS_APPEND(h_inode)) {
31255 +               lockdep_off();
31256 +               *a->errp = notify_change(a->path->dentry, a->ia,
31257 +                                        a->delegated_inode);
31258 +               lockdep_on();
31259 +               if (!*a->errp)
31260 +                       vfsub_update_h_iattr(a->path, /*did*/NULL); /*ignore*/
31261 +       }
31262 +       AuTraceErr(*a->errp);
31263 +}
31264 +
31265 +int vfsub_notify_change(struct path *path, struct iattr *ia,
31266 +                       struct inode **delegated_inode)
31267 +{
31268 +       int err;
31269 +       struct notify_change_args args = {
31270 +               .errp                   = &err,
31271 +               .path                   = path,
31272 +               .ia                     = ia,
31273 +               .delegated_inode        = delegated_inode
31274 +       };
31275 +
31276 +       call_notify_change(&args);
31277 +
31278 +       return err;
31279 +}
31280 +
31281 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
31282 +                           struct inode **delegated_inode)
31283 +{
31284 +       int err, wkq_err;
31285 +       struct notify_change_args args = {
31286 +               .errp                   = &err,
31287 +               .path                   = path,
31288 +               .ia                     = ia,
31289 +               .delegated_inode        = delegated_inode
31290 +       };
31291 +
31292 +       wkq_err = au_wkq_wait(call_notify_change, &args);
31293 +       if (unlikely(wkq_err))
31294 +               err = wkq_err;
31295 +
31296 +       return err;
31297 +}
31298 +
31299 +/* ---------------------------------------------------------------------- */
31300 +
31301 +struct unlink_args {
31302 +       int *errp;
31303 +       struct inode *dir;
31304 +       struct path *path;
31305 +       struct inode **delegated_inode;
31306 +};
31307 +
31308 +static void call_unlink(void *args)
31309 +{
31310 +       struct unlink_args *a = args;
31311 +       struct dentry *d = a->path->dentry;
31312 +       struct inode *h_inode;
31313 +       const int stop_sillyrename = (au_test_nfs(d->d_sb)
31314 +                                     && au_dcount(d) == 1);
31315 +
31316 +       IMustLock(a->dir);
31317 +
31318 +       a->path->dentry = d->d_parent;
31319 +       *a->errp = security_path_unlink(a->path, d);
31320 +       a->path->dentry = d;
31321 +       if (unlikely(*a->errp))
31322 +               return;
31323 +
31324 +       if (!stop_sillyrename)
31325 +               dget(d);
31326 +       h_inode = NULL;
31327 +       if (d_is_positive(d)) {
31328 +               h_inode = d_inode(d);
31329 +               ihold(h_inode);
31330 +       }
31331 +
31332 +       lockdep_off();
31333 +       *a->errp = vfs_unlink(a->dir, d, a->delegated_inode);
31334 +       lockdep_on();
31335 +       if (!*a->errp) {
31336 +               struct path tmp = {
31337 +                       .dentry = d->d_parent,
31338 +                       .mnt    = a->path->mnt
31339 +               };
31340 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
31341 +       }
31342 +
31343 +       if (!stop_sillyrename)
31344 +               dput(d);
31345 +       if (h_inode)
31346 +               iput(h_inode);
31347 +
31348 +       AuTraceErr(*a->errp);
31349 +}
31350 +
31351 +/*
31352 + * @dir: must be locked.
31353 + * @dentry: target dentry.
31354 + */
31355 +int vfsub_unlink(struct inode *dir, struct path *path,
31356 +                struct inode **delegated_inode, int force)
31357 +{
31358 +       int err;
31359 +       struct unlink_args args = {
31360 +               .errp                   = &err,
31361 +               .dir                    = dir,
31362 +               .path                   = path,
31363 +               .delegated_inode        = delegated_inode
31364 +       };
31365 +
31366 +       if (!force)
31367 +               call_unlink(&args);
31368 +       else {
31369 +               int wkq_err;
31370 +
31371 +               wkq_err = au_wkq_wait(call_unlink, &args);
31372 +               if (unlikely(wkq_err))
31373 +                       err = wkq_err;
31374 +       }
31375 +
31376 +       return err;
31377 +}
31378 diff -urN /usr/share/empty/fs/aufs/vfsub.h linux/fs/aufs/vfsub.h
31379 --- /usr/share/empty/fs/aufs/vfsub.h    1970-01-01 01:00:00.000000000 +0100
31380 +++ linux/fs/aufs/vfsub.h       2016-10-09 16:55:36.496035060 +0200
31381 @@ -0,0 +1,316 @@
31382 +/*
31383 + * Copyright (C) 2005-2016 Junjiro R. Okajima
31384 + *
31385 + * This program, aufs is free software; you can redistribute it and/or modify
31386 + * it under the terms of the GNU General Public License as published by
31387 + * the Free Software Foundation; either version 2 of the License, or
31388 + * (at your option) any later version.
31389 + *
31390 + * This program is distributed in the hope that it will be useful,
31391 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31392 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31393 + * GNU General Public License for more details.
31394 + *
31395 + * You should have received a copy of the GNU General Public License
31396 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31397 + */
31398 +
31399 +/*
31400 + * sub-routines for VFS
31401 + */
31402 +
31403 +#ifndef __AUFS_VFSUB_H__
31404 +#define __AUFS_VFSUB_H__
31405 +
31406 +#ifdef __KERNEL__
31407 +
31408 +#include <linux/fs.h>
31409 +#include <linux/mount.h>
31410 +#include <linux/posix_acl.h>
31411 +#include <linux/xattr.h>
31412 +#include "debug.h"
31413 +
31414 +/* copied from linux/fs/internal.h */
31415 +/* todo: BAD approach!! */
31416 +extern void __mnt_drop_write(struct vfsmount *);
31417 +extern int open_check_o_direct(struct file *f);
31418 +
31419 +/* ---------------------------------------------------------------------- */
31420 +
31421 +/* lock subclass for lower inode */
31422 +/* default MAX_LOCKDEP_SUBCLASSES(8) is not enough */
31423 +/* reduce? gave up. */
31424 +enum {
31425 +       AuLsc_I_Begin = I_MUTEX_PARENT2, /* 5 */
31426 +       AuLsc_I_PARENT,         /* lower inode, parent first */
31427 +       AuLsc_I_PARENT2,        /* copyup dirs */
31428 +       AuLsc_I_PARENT3,        /* copyup wh */
31429 +       AuLsc_I_CHILD,
31430 +       AuLsc_I_CHILD2,
31431 +       AuLsc_I_End
31432 +};
31433 +
31434 +/* to debug easier, do not make them inlined functions */
31435 +#define MtxMustLock(mtx)       AuDebugOn(!mutex_is_locked(mtx))
31436 +#define IMustLock(i)           AuDebugOn(!inode_is_locked(i))
31437 +
31438 +/* ---------------------------------------------------------------------- */
31439 +
31440 +static inline void vfsub_drop_nlink(struct inode *inode)
31441 +{
31442 +       AuDebugOn(!inode->i_nlink);
31443 +       drop_nlink(inode);
31444 +}
31445 +
31446 +static inline void vfsub_dead_dir(struct inode *inode)
31447 +{
31448 +       AuDebugOn(!S_ISDIR(inode->i_mode));
31449 +       inode->i_flags |= S_DEAD;
31450 +       clear_nlink(inode);
31451 +}
31452 +
31453 +static inline int vfsub_native_ro(struct inode *inode)
31454 +{
31455 +       return (inode->i_sb->s_flags & MS_RDONLY)
31456 +               || IS_RDONLY(inode)
31457 +               /* || IS_APPEND(inode) */
31458 +               || IS_IMMUTABLE(inode);
31459 +}
31460 +
31461 +#ifdef CONFIG_AUFS_BR_FUSE
31462 +int vfsub_test_mntns(struct vfsmount *mnt, struct super_block *h_sb);
31463 +#else
31464 +AuStubInt0(vfsub_test_mntns, struct vfsmount *mnt, struct super_block *h_sb);
31465 +#endif
31466 +
31467 +/* ---------------------------------------------------------------------- */
31468 +
31469 +int vfsub_update_h_iattr(struct path *h_path, int *did);
31470 +struct file *vfsub_dentry_open(struct path *path, int flags);
31471 +struct file *vfsub_filp_open(const char *path, int oflags, int mode);
31472 +struct vfsub_aopen_args {
31473 +       struct file     *file;
31474 +       unsigned int    open_flag;
31475 +       umode_t         create_mode;
31476 +       int             *opened;
31477 +};
31478 +struct au_branch;
31479 +int vfsub_atomic_open(struct inode *dir, struct dentry *dentry,
31480 +                     struct vfsub_aopen_args *args, struct au_branch *br);
31481 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path);
31482 +
31483 +struct dentry *vfsub_lookup_one_len_unlocked(const char *name,
31484 +                                            struct dentry *parent, int len);
31485 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
31486 +                                   int len);
31487 +
31488 +struct vfsub_lkup_one_args {
31489 +       struct dentry **errp;
31490 +       struct qstr *name;
31491 +       struct dentry *parent;
31492 +};
31493 +
31494 +static inline struct dentry *vfsub_lkup_one(struct qstr *name,
31495 +                                           struct dentry *parent)
31496 +{
31497 +       return vfsub_lookup_one_len(name->name, parent, name->len);
31498 +}
31499 +
31500 +void vfsub_call_lkup_one(void *args);
31501 +
31502 +/* ---------------------------------------------------------------------- */
31503 +
31504 +static inline int vfsub_mnt_want_write(struct vfsmount *mnt)
31505 +{
31506 +       int err;
31507 +
31508 +       lockdep_off();
31509 +       err = mnt_want_write(mnt);
31510 +       lockdep_on();
31511 +       return err;
31512 +}
31513 +
31514 +static inline void vfsub_mnt_drop_write(struct vfsmount *mnt)
31515 +{
31516 +       lockdep_off();
31517 +       mnt_drop_write(mnt);
31518 +       lockdep_on();
31519 +}
31520 +
31521 +#if 0 /* reserved */
31522 +static inline void vfsub_mnt_drop_write_file(struct file *file)
31523 +{
31524 +       lockdep_off();
31525 +       mnt_drop_write_file(file);
31526 +       lockdep_on();
31527 +}
31528 +#endif
31529 +
31530 +/* ---------------------------------------------------------------------- */
31531 +
31532 +struct au_hinode;
31533 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
31534 +                                struct dentry *d2, struct au_hinode *hdir2);
31535 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
31536 +                        struct dentry *d2, struct au_hinode *hdir2);
31537 +
31538 +int vfsub_create(struct inode *dir, struct path *path, int mode,
31539 +                bool want_excl);
31540 +int vfsub_symlink(struct inode *dir, struct path *path,
31541 +                 const char *symname);
31542 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev);
31543 +int vfsub_link(struct dentry *src_dentry, struct inode *dir,
31544 +              struct path *path, struct inode **delegated_inode);
31545 +int vfsub_rename(struct inode *src_hdir, struct dentry *src_dentry,
31546 +                struct inode *hdir, struct path *path,
31547 +                struct inode **delegated_inode);
31548 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode);
31549 +int vfsub_rmdir(struct inode *dir, struct path *path);
31550 +
31551 +/* ---------------------------------------------------------------------- */
31552 +
31553 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
31554 +                    loff_t *ppos);
31555 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
31556 +                       loff_t *ppos);
31557 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
31558 +                     loff_t *ppos);
31559 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count,
31560 +                     loff_t *ppos);
31561 +int vfsub_flush(struct file *file, fl_owner_t id);
31562 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx);
31563 +
31564 +static inline loff_t vfsub_f_size_read(struct file *file)
31565 +{
31566 +       return i_size_read(file_inode(file));
31567 +}
31568 +
31569 +static inline unsigned int vfsub_file_flags(struct file *file)
31570 +{
31571 +       unsigned int flags;
31572 +
31573 +       spin_lock(&file->f_lock);
31574 +       flags = file->f_flags;
31575 +       spin_unlock(&file->f_lock);
31576 +
31577 +       return flags;
31578 +}
31579 +
31580 +static inline int vfsub_file_execed(struct file *file)
31581 +{
31582 +       /* todo: direct access f_flags */
31583 +       return !!(vfsub_file_flags(file) & __FMODE_EXEC);
31584 +}
31585 +
31586 +#if 0 /* reserved */
31587 +static inline void vfsub_file_accessed(struct file *h_file)
31588 +{
31589 +       file_accessed(h_file);
31590 +       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL); /*ignore*/
31591 +}
31592 +#endif
31593 +
31594 +#if 0 /* reserved */
31595 +static inline void vfsub_touch_atime(struct vfsmount *h_mnt,
31596 +                                    struct dentry *h_dentry)
31597 +{
31598 +       struct path h_path = {
31599 +               .dentry = h_dentry,
31600 +               .mnt    = h_mnt
31601 +       };
31602 +       touch_atime(&h_path);
31603 +       vfsub_update_h_iattr(&h_path, /*did*/NULL); /*ignore*/
31604 +}
31605 +#endif
31606 +
31607 +static inline int vfsub_update_time(struct inode *h_inode, struct timespec *ts,
31608 +                                   int flags)
31609 +{
31610 +       return update_time(h_inode, ts, flags);
31611 +       /* no vfsub_update_h_iattr() since we don't have struct path */
31612 +}
31613 +
31614 +#ifdef CONFIG_FS_POSIX_ACL
31615 +static inline int vfsub_acl_chmod(struct inode *h_inode, umode_t h_mode)
31616 +{
31617 +       int err;
31618 +
31619 +       err = posix_acl_chmod(h_inode, h_mode);
31620 +       if (err == -EOPNOTSUPP)
31621 +               err = 0;
31622 +       return err;
31623 +}
31624 +#else
31625 +AuStubInt0(vfsub_acl_chmod, struct inode *h_inode, umode_t h_mode);
31626 +#endif
31627 +
31628 +long vfsub_splice_to(struct file *in, loff_t *ppos,
31629 +                    struct pipe_inode_info *pipe, size_t len,
31630 +                    unsigned int flags);
31631 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
31632 +                      loff_t *ppos, size_t len, unsigned int flags);
31633 +
31634 +static inline long vfsub_truncate(struct path *path, loff_t length)
31635 +{
31636 +       long err;
31637 +
31638 +       lockdep_off();
31639 +       err = vfs_truncate(path, length);
31640 +       lockdep_on();
31641 +       return err;
31642 +}
31643 +
31644 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
31645 +               struct file *h_file);
31646 +int vfsub_fsync(struct file *file, struct path *path, int datasync);
31647 +
31648 +/* ---------------------------------------------------------------------- */
31649 +
31650 +static inline loff_t vfsub_llseek(struct file *file, loff_t offset, int origin)
31651 +{
31652 +       loff_t err;
31653 +
31654 +       lockdep_off();
31655 +       err = vfs_llseek(file, offset, origin);
31656 +       lockdep_on();
31657 +       return err;
31658 +}
31659 +
31660 +/* ---------------------------------------------------------------------- */
31661 +
31662 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode);
31663 +int vfsub_sio_rmdir(struct inode *dir, struct path *path);
31664 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
31665 +                           struct inode **delegated_inode);
31666 +int vfsub_notify_change(struct path *path, struct iattr *ia,
31667 +                       struct inode **delegated_inode);
31668 +int vfsub_unlink(struct inode *dir, struct path *path,
31669 +                struct inode **delegated_inode, int force);
31670 +
31671 +/* ---------------------------------------------------------------------- */
31672 +
31673 +static inline int vfsub_setxattr(struct dentry *dentry, const char *name,
31674 +                                const void *value, size_t size, int flags)
31675 +{
31676 +       int err;
31677 +
31678 +       lockdep_off();
31679 +       err = vfs_setxattr(dentry, name, value, size, flags);
31680 +       lockdep_on();
31681 +
31682 +       return err;
31683 +}
31684 +
31685 +static inline int vfsub_removexattr(struct dentry *dentry, const char *name)
31686 +{
31687 +       int err;
31688 +
31689 +       lockdep_off();
31690 +       err = vfs_removexattr(dentry, name);
31691 +       lockdep_on();
31692 +
31693 +       return err;
31694 +}
31695 +
31696 +#endif /* __KERNEL__ */
31697 +#endif /* __AUFS_VFSUB_H__ */
31698 diff -urN /usr/share/empty/fs/aufs/wbr_policy.c linux/fs/aufs/wbr_policy.c
31699 --- /usr/share/empty/fs/aufs/wbr_policy.c       1970-01-01 01:00:00.000000000 +0100
31700 +++ linux/fs/aufs/wbr_policy.c  2016-10-09 16:55:36.496035060 +0200
31701 @@ -0,0 +1,765 @@
31702 +/*
31703 + * Copyright (C) 2005-2016 Junjiro R. Okajima
31704 + *
31705 + * This program, aufs is free software; you can redistribute it and/or modify
31706 + * it under the terms of the GNU General Public License as published by
31707 + * the Free Software Foundation; either version 2 of the License, or
31708 + * (at your option) any later version.
31709 + *
31710 + * This program is distributed in the hope that it will be useful,
31711 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31712 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31713 + * GNU General Public License for more details.
31714 + *
31715 + * You should have received a copy of the GNU General Public License
31716 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31717 + */
31718 +
31719 +/*
31720 + * policies for selecting one among multiple writable branches
31721 + */
31722 +
31723 +#include <linux/statfs.h>
31724 +#include "aufs.h"
31725 +
31726 +/* subset of cpup_attr() */
31727 +static noinline_for_stack
31728 +int au_cpdown_attr(struct path *h_path, struct dentry *h_src)
31729 +{
31730 +       int err, sbits;
31731 +       struct iattr ia;
31732 +       struct inode *h_isrc;
31733 +
31734 +       h_isrc = d_inode(h_src);
31735 +       ia.ia_valid = ATTR_FORCE | ATTR_MODE | ATTR_UID | ATTR_GID;
31736 +       ia.ia_mode = h_isrc->i_mode;
31737 +       ia.ia_uid = h_isrc->i_uid;
31738 +       ia.ia_gid = h_isrc->i_gid;
31739 +       sbits = !!(ia.ia_mode & (S_ISUID | S_ISGID));
31740 +       au_cpup_attr_flags(d_inode(h_path->dentry), h_isrc->i_flags);
31741 +       /* no delegation since it is just created */
31742 +       err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
31743 +
31744 +       /* is this nfs only? */
31745 +       if (!err && sbits && au_test_nfs(h_path->dentry->d_sb)) {
31746 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
31747 +               ia.ia_mode = h_isrc->i_mode;
31748 +               err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
31749 +       }
31750 +
31751 +       return err;
31752 +}
31753 +
31754 +#define AuCpdown_PARENT_OPQ    1
31755 +#define AuCpdown_WHED          (1 << 1)
31756 +#define AuCpdown_MADE_DIR      (1 << 2)
31757 +#define AuCpdown_DIROPQ                (1 << 3)
31758 +#define au_ftest_cpdown(flags, name)   ((flags) & AuCpdown_##name)
31759 +#define au_fset_cpdown(flags, name) \
31760 +       do { (flags) |= AuCpdown_##name; } while (0)
31761 +#define au_fclr_cpdown(flags, name) \
31762 +       do { (flags) &= ~AuCpdown_##name; } while (0)
31763 +
31764 +static int au_cpdown_dir_opq(struct dentry *dentry, aufs_bindex_t bdst,
31765 +                            unsigned int *flags)
31766 +{
31767 +       int err;
31768 +       struct dentry *opq_dentry;
31769 +
31770 +       opq_dentry = au_diropq_create(dentry, bdst);
31771 +       err = PTR_ERR(opq_dentry);
31772 +       if (IS_ERR(opq_dentry))
31773 +               goto out;
31774 +       dput(opq_dentry);
31775 +       au_fset_cpdown(*flags, DIROPQ);
31776 +
31777 +out:
31778 +       return err;
31779 +}
31780 +
31781 +static int au_cpdown_dir_wh(struct dentry *dentry, struct dentry *h_parent,
31782 +                           struct inode *dir, aufs_bindex_t bdst)
31783 +{
31784 +       int err;
31785 +       struct path h_path;
31786 +       struct au_branch *br;
31787 +
31788 +       br = au_sbr(dentry->d_sb, bdst);
31789 +       h_path.dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
31790 +       err = PTR_ERR(h_path.dentry);
31791 +       if (IS_ERR(h_path.dentry))
31792 +               goto out;
31793 +
31794 +       err = 0;
31795 +       if (d_is_positive(h_path.dentry)) {
31796 +               h_path.mnt = au_br_mnt(br);
31797 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bdst), &h_path,
31798 +                                         dentry);
31799 +       }
31800 +       dput(h_path.dentry);
31801 +
31802 +out:
31803 +       return err;
31804 +}
31805 +
31806 +static int au_cpdown_dir(struct dentry *dentry, aufs_bindex_t bdst,
31807 +                        struct au_pin *pin,
31808 +                        struct dentry *h_parent, void *arg)
31809 +{
31810 +       int err, rerr;
31811 +       aufs_bindex_t bopq, btop;
31812 +       struct path h_path;
31813 +       struct dentry *parent;
31814 +       struct inode *h_dir, *h_inode, *inode, *dir;
31815 +       unsigned int *flags = arg;
31816 +
31817 +       btop = au_dbtop(dentry);
31818 +       /* dentry is di-locked */
31819 +       parent = dget_parent(dentry);
31820 +       dir = d_inode(parent);
31821 +       h_dir = d_inode(h_parent);
31822 +       AuDebugOn(h_dir != au_h_iptr(dir, bdst));
31823 +       IMustLock(h_dir);
31824 +
31825 +       err = au_lkup_neg(dentry, bdst, /*wh*/0);
31826 +       if (unlikely(err < 0))
31827 +               goto out;
31828 +       h_path.dentry = au_h_dptr(dentry, bdst);
31829 +       h_path.mnt = au_sbr_mnt(dentry->d_sb, bdst);
31830 +       err = vfsub_sio_mkdir(au_h_iptr(dir, bdst), &h_path,
31831 +                             S_IRWXU | S_IRUGO | S_IXUGO);
31832 +       if (unlikely(err))
31833 +               goto out_put;
31834 +       au_fset_cpdown(*flags, MADE_DIR);
31835 +
31836 +       bopq = au_dbdiropq(dentry);
31837 +       au_fclr_cpdown(*flags, WHED);
31838 +       au_fclr_cpdown(*flags, DIROPQ);
31839 +       if (au_dbwh(dentry) == bdst)
31840 +               au_fset_cpdown(*flags, WHED);
31841 +       if (!au_ftest_cpdown(*flags, PARENT_OPQ) && bopq <= bdst)
31842 +               au_fset_cpdown(*flags, PARENT_OPQ);
31843 +       h_inode = d_inode(h_path.dentry);
31844 +       inode_lock_nested(h_inode, AuLsc_I_CHILD);
31845 +       if (au_ftest_cpdown(*flags, WHED)) {
31846 +               err = au_cpdown_dir_opq(dentry, bdst, flags);
31847 +               if (unlikely(err)) {
31848 +                       inode_unlock(h_inode);
31849 +                       goto out_dir;
31850 +               }
31851 +       }
31852 +
31853 +       err = au_cpdown_attr(&h_path, au_h_dptr(dentry, btop));
31854 +       inode_unlock(h_inode);
31855 +       if (unlikely(err))
31856 +               goto out_opq;
31857 +
31858 +       if (au_ftest_cpdown(*flags, WHED)) {
31859 +               err = au_cpdown_dir_wh(dentry, h_parent, dir, bdst);
31860 +               if (unlikely(err))
31861 +                       goto out_opq;
31862 +       }
31863 +
31864 +       inode = d_inode(dentry);
31865 +       if (au_ibbot(inode) < bdst)
31866 +               au_set_ibbot(inode, bdst);
31867 +       au_set_h_iptr(inode, bdst, au_igrab(h_inode),
31868 +                     au_hi_flags(inode, /*isdir*/1));
31869 +       au_fhsm_wrote(dentry->d_sb, bdst, /*force*/0);
31870 +       goto out; /* success */
31871 +
31872 +       /* revert */
31873 +out_opq:
31874 +       if (au_ftest_cpdown(*flags, DIROPQ)) {
31875 +               inode_lock_nested(h_inode, AuLsc_I_CHILD);
31876 +               rerr = au_diropq_remove(dentry, bdst);
31877 +               inode_unlock(h_inode);
31878 +               if (unlikely(rerr)) {
31879 +                       AuIOErr("failed removing diropq for %pd b%d (%d)\n",
31880 +                               dentry, bdst, rerr);
31881 +                       err = -EIO;
31882 +                       goto out;
31883 +               }
31884 +       }
31885 +out_dir:
31886 +       if (au_ftest_cpdown(*flags, MADE_DIR)) {
31887 +               rerr = vfsub_sio_rmdir(au_h_iptr(dir, bdst), &h_path);
31888 +               if (unlikely(rerr)) {
31889 +                       AuIOErr("failed removing %pd b%d (%d)\n",
31890 +                               dentry, bdst, rerr);
31891 +                       err = -EIO;
31892 +               }
31893 +       }
31894 +out_put:
31895 +       au_set_h_dptr(dentry, bdst, NULL);
31896 +       if (au_dbbot(dentry) == bdst)
31897 +               au_update_dbbot(dentry);
31898 +out:
31899 +       dput(parent);
31900 +       return err;
31901 +}
31902 +
31903 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst)
31904 +{
31905 +       int err;
31906 +       unsigned int flags;
31907 +
31908 +       flags = 0;
31909 +       err = au_cp_dirs(dentry, bdst, au_cpdown_dir, &flags);
31910 +
31911 +       return err;
31912 +}
31913 +
31914 +/* ---------------------------------------------------------------------- */
31915 +
31916 +/* policies for create */
31917 +
31918 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex)
31919 +{
31920 +       int err, i, j, ndentry;
31921 +       aufs_bindex_t bopq;
31922 +       struct au_dcsub_pages dpages;
31923 +       struct au_dpage *dpage;
31924 +       struct dentry **dentries, *parent, *d;
31925 +
31926 +       err = au_dpages_init(&dpages, GFP_NOFS);
31927 +       if (unlikely(err))
31928 +               goto out;
31929 +       parent = dget_parent(dentry);
31930 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/0);
31931 +       if (unlikely(err))
31932 +               goto out_free;
31933 +
31934 +       err = bindex;
31935 +       for (i = 0; i < dpages.ndpage; i++) {
31936 +               dpage = dpages.dpages + i;
31937 +               dentries = dpage->dentries;
31938 +               ndentry = dpage->ndentry;
31939 +               for (j = 0; j < ndentry; j++) {
31940 +                       d = dentries[j];
31941 +                       di_read_lock_parent2(d, !AuLock_IR);
31942 +                       bopq = au_dbdiropq(d);
31943 +                       di_read_unlock(d, !AuLock_IR);
31944 +                       if (bopq >= 0 && bopq < err)
31945 +                               err = bopq;
31946 +               }
31947 +       }
31948 +
31949 +out_free:
31950 +       dput(parent);
31951 +       au_dpages_free(&dpages);
31952 +out:
31953 +       return err;
31954 +}
31955 +
31956 +static int au_wbr_bu(struct super_block *sb, aufs_bindex_t bindex)
31957 +{
31958 +       for (; bindex >= 0; bindex--)
31959 +               if (!au_br_rdonly(au_sbr(sb, bindex)))
31960 +                       return bindex;
31961 +       return -EROFS;
31962 +}
31963 +
31964 +/* top down parent */
31965 +static int au_wbr_create_tdp(struct dentry *dentry,
31966 +                            unsigned int flags __maybe_unused)
31967 +{
31968 +       int err;
31969 +       aufs_bindex_t btop, bindex;
31970 +       struct super_block *sb;
31971 +       struct dentry *parent, *h_parent;
31972 +
31973 +       sb = dentry->d_sb;
31974 +       btop = au_dbtop(dentry);
31975 +       err = btop;
31976 +       if (!au_br_rdonly(au_sbr(sb, btop)))
31977 +               goto out;
31978 +
31979 +       err = -EROFS;
31980 +       parent = dget_parent(dentry);
31981 +       for (bindex = au_dbtop(parent); bindex < btop; bindex++) {
31982 +               h_parent = au_h_dptr(parent, bindex);
31983 +               if (!h_parent || d_is_negative(h_parent))
31984 +                       continue;
31985 +
31986 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
31987 +                       err = bindex;
31988 +                       break;
31989 +               }
31990 +       }
31991 +       dput(parent);
31992 +
31993 +       /* bottom up here */
31994 +       if (unlikely(err < 0)) {
31995 +               err = au_wbr_bu(sb, btop - 1);
31996 +               if (err >= 0)
31997 +                       err = au_wbr_nonopq(dentry, err);
31998 +       }
31999 +
32000 +out:
32001 +       AuDbg("b%d\n", err);
32002 +       return err;
32003 +}
32004 +
32005 +/* ---------------------------------------------------------------------- */
32006 +
32007 +/* an exception for the policy other than tdp */
32008 +static int au_wbr_create_exp(struct dentry *dentry)
32009 +{
32010 +       int err;
32011 +       aufs_bindex_t bwh, bdiropq;
32012 +       struct dentry *parent;
32013 +
32014 +       err = -1;
32015 +       bwh = au_dbwh(dentry);
32016 +       parent = dget_parent(dentry);
32017 +       bdiropq = au_dbdiropq(parent);
32018 +       if (bwh >= 0) {
32019 +               if (bdiropq >= 0)
32020 +                       err = min(bdiropq, bwh);
32021 +               else
32022 +                       err = bwh;
32023 +               AuDbg("%d\n", err);
32024 +       } else if (bdiropq >= 0) {
32025 +               err = bdiropq;
32026 +               AuDbg("%d\n", err);
32027 +       }
32028 +       dput(parent);
32029 +
32030 +       if (err >= 0)
32031 +               err = au_wbr_nonopq(dentry, err);
32032 +
32033 +       if (err >= 0 && au_br_rdonly(au_sbr(dentry->d_sb, err)))
32034 +               err = -1;
32035 +
32036 +       AuDbg("%d\n", err);
32037 +       return err;
32038 +}
32039 +
32040 +/* ---------------------------------------------------------------------- */
32041 +
32042 +/* round robin */
32043 +static int au_wbr_create_init_rr(struct super_block *sb)
32044 +{
32045 +       int err;
32046 +
32047 +       err = au_wbr_bu(sb, au_sbbot(sb));
32048 +       atomic_set(&au_sbi(sb)->si_wbr_rr_next, -err); /* less important */
32049 +       /* smp_mb(); */
32050 +
32051 +       AuDbg("b%d\n", err);
32052 +       return err;
32053 +}
32054 +
32055 +static int au_wbr_create_rr(struct dentry *dentry, unsigned int flags)
32056 +{
32057 +       int err, nbr;
32058 +       unsigned int u;
32059 +       aufs_bindex_t bindex, bbot;
32060 +       struct super_block *sb;
32061 +       atomic_t *next;
32062 +
32063 +       err = au_wbr_create_exp(dentry);
32064 +       if (err >= 0)
32065 +               goto out;
32066 +
32067 +       sb = dentry->d_sb;
32068 +       next = &au_sbi(sb)->si_wbr_rr_next;
32069 +       bbot = au_sbbot(sb);
32070 +       nbr = bbot + 1;
32071 +       for (bindex = 0; bindex <= bbot; bindex++) {
32072 +               if (!au_ftest_wbr(flags, DIR)) {
32073 +                       err = atomic_dec_return(next) + 1;
32074 +                       /* modulo for 0 is meaningless */
32075 +                       if (unlikely(!err))
32076 +                               err = atomic_dec_return(next) + 1;
32077 +               } else
32078 +                       err = atomic_read(next);
32079 +               AuDbg("%d\n", err);
32080 +               u = err;
32081 +               err = u % nbr;
32082 +               AuDbg("%d\n", err);
32083 +               if (!au_br_rdonly(au_sbr(sb, err)))
32084 +                       break;
32085 +               err = -EROFS;
32086 +       }
32087 +
32088 +       if (err >= 0)
32089 +               err = au_wbr_nonopq(dentry, err);
32090 +
32091 +out:
32092 +       AuDbg("%d\n", err);
32093 +       return err;
32094 +}
32095 +
32096 +/* ---------------------------------------------------------------------- */
32097 +
32098 +/* most free space */
32099 +static void au_mfs(struct dentry *dentry, struct dentry *parent)
32100 +{
32101 +       struct super_block *sb;
32102 +       struct au_branch *br;
32103 +       struct au_wbr_mfs *mfs;
32104 +       struct dentry *h_parent;
32105 +       aufs_bindex_t bindex, bbot;
32106 +       int err;
32107 +       unsigned long long b, bavail;
32108 +       struct path h_path;
32109 +       /* reduce the stack usage */
32110 +       struct kstatfs *st;
32111 +
32112 +       st = kmalloc(sizeof(*st), GFP_NOFS);
32113 +       if (unlikely(!st)) {
32114 +               AuWarn1("failed updating mfs(%d), ignored\n", -ENOMEM);
32115 +               return;
32116 +       }
32117 +
32118 +       bavail = 0;
32119 +       sb = dentry->d_sb;
32120 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32121 +       MtxMustLock(&mfs->mfs_lock);
32122 +       mfs->mfs_bindex = -EROFS;
32123 +       mfs->mfsrr_bytes = 0;
32124 +       if (!parent) {
32125 +               bindex = 0;
32126 +               bbot = au_sbbot(sb);
32127 +       } else {
32128 +               bindex = au_dbtop(parent);
32129 +               bbot = au_dbtaildir(parent);
32130 +       }
32131 +
32132 +       for (; bindex <= bbot; bindex++) {
32133 +               if (parent) {
32134 +                       h_parent = au_h_dptr(parent, bindex);
32135 +                       if (!h_parent || d_is_negative(h_parent))
32136 +                               continue;
32137 +               }
32138 +               br = au_sbr(sb, bindex);
32139 +               if (au_br_rdonly(br))
32140 +                       continue;
32141 +
32142 +               /* sb->s_root for NFS is unreliable */
32143 +               h_path.mnt = au_br_mnt(br);
32144 +               h_path.dentry = h_path.mnt->mnt_root;
32145 +               err = vfs_statfs(&h_path, st);
32146 +               if (unlikely(err)) {
32147 +                       AuWarn1("failed statfs, b%d, %d\n", bindex, err);
32148 +                       continue;
32149 +               }
32150 +
32151 +               /* when the available size is equal, select the lower one */
32152 +               BUILD_BUG_ON(sizeof(b) < sizeof(st->f_bavail)
32153 +                            || sizeof(b) < sizeof(st->f_bsize));
32154 +               b = st->f_bavail * st->f_bsize;
32155 +               br->br_wbr->wbr_bytes = b;
32156 +               if (b >= bavail) {
32157 +                       bavail = b;
32158 +                       mfs->mfs_bindex = bindex;
32159 +                       mfs->mfs_jiffy = jiffies;
32160 +               }
32161 +       }
32162 +
32163 +       mfs->mfsrr_bytes = bavail;
32164 +       AuDbg("b%d\n", mfs->mfs_bindex);
32165 +       au_delayed_kfree(st);
32166 +}
32167 +
32168 +static int au_wbr_create_mfs(struct dentry *dentry, unsigned int flags)
32169 +{
32170 +       int err;
32171 +       struct dentry *parent;
32172 +       struct super_block *sb;
32173 +       struct au_wbr_mfs *mfs;
32174 +
32175 +       err = au_wbr_create_exp(dentry);
32176 +       if (err >= 0)
32177 +               goto out;
32178 +
32179 +       sb = dentry->d_sb;
32180 +       parent = NULL;
32181 +       if (au_ftest_wbr(flags, PARENT))
32182 +               parent = dget_parent(dentry);
32183 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32184 +       mutex_lock(&mfs->mfs_lock);
32185 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
32186 +           || mfs->mfs_bindex < 0
32187 +           || au_br_rdonly(au_sbr(sb, mfs->mfs_bindex)))
32188 +               au_mfs(dentry, parent);
32189 +       mutex_unlock(&mfs->mfs_lock);
32190 +       err = mfs->mfs_bindex;
32191 +       dput(parent);
32192 +
32193 +       if (err >= 0)
32194 +               err = au_wbr_nonopq(dentry, err);
32195 +
32196 +out:
32197 +       AuDbg("b%d\n", err);
32198 +       return err;
32199 +}
32200 +
32201 +static int au_wbr_create_init_mfs(struct super_block *sb)
32202 +{
32203 +       struct au_wbr_mfs *mfs;
32204 +
32205 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32206 +       mutex_init(&mfs->mfs_lock);
32207 +       mfs->mfs_jiffy = 0;
32208 +       mfs->mfs_bindex = -EROFS;
32209 +
32210 +       return 0;
32211 +}
32212 +
32213 +static int au_wbr_create_fin_mfs(struct super_block *sb __maybe_unused)
32214 +{
32215 +       mutex_destroy(&au_sbi(sb)->si_wbr_mfs.mfs_lock);
32216 +       return 0;
32217 +}
32218 +
32219 +/* ---------------------------------------------------------------------- */
32220 +
32221 +/* most free space and then round robin */
32222 +static int au_wbr_create_mfsrr(struct dentry *dentry, unsigned int flags)
32223 +{
32224 +       int err;
32225 +       struct au_wbr_mfs *mfs;
32226 +
32227 +       err = au_wbr_create_mfs(dentry, flags);
32228 +       if (err >= 0) {
32229 +               mfs = &au_sbi(dentry->d_sb)->si_wbr_mfs;
32230 +               mutex_lock(&mfs->mfs_lock);
32231 +               if (mfs->mfsrr_bytes < mfs->mfsrr_watermark)
32232 +                       err = au_wbr_create_rr(dentry, flags);
32233 +               mutex_unlock(&mfs->mfs_lock);
32234 +       }
32235 +
32236 +       AuDbg("b%d\n", err);
32237 +       return err;
32238 +}
32239 +
32240 +static int au_wbr_create_init_mfsrr(struct super_block *sb)
32241 +{
32242 +       int err;
32243 +
32244 +       au_wbr_create_init_mfs(sb); /* ignore */
32245 +       err = au_wbr_create_init_rr(sb);
32246 +
32247 +       return err;
32248 +}
32249 +
32250 +/* ---------------------------------------------------------------------- */
32251 +
32252 +/* top down parent and most free space */
32253 +static int au_wbr_create_pmfs(struct dentry *dentry, unsigned int flags)
32254 +{
32255 +       int err, e2;
32256 +       unsigned long long b;
32257 +       aufs_bindex_t bindex, btop, bbot;
32258 +       struct super_block *sb;
32259 +       struct dentry *parent, *h_parent;
32260 +       struct au_branch *br;
32261 +
32262 +       err = au_wbr_create_tdp(dentry, flags);
32263 +       if (unlikely(err < 0))
32264 +               goto out;
32265 +       parent = dget_parent(dentry);
32266 +       btop = au_dbtop(parent);
32267 +       bbot = au_dbtaildir(parent);
32268 +       if (btop == bbot)
32269 +               goto out_parent; /* success */
32270 +
32271 +       e2 = au_wbr_create_mfs(dentry, flags);
32272 +       if (e2 < 0)
32273 +               goto out_parent; /* success */
32274 +
32275 +       /* when the available size is equal, select upper one */
32276 +       sb = dentry->d_sb;
32277 +       br = au_sbr(sb, err);
32278 +       b = br->br_wbr->wbr_bytes;
32279 +       AuDbg("b%d, %llu\n", err, b);
32280 +
32281 +       for (bindex = btop; bindex <= bbot; bindex++) {
32282 +               h_parent = au_h_dptr(parent, bindex);
32283 +               if (!h_parent || d_is_negative(h_parent))
32284 +                       continue;
32285 +
32286 +               br = au_sbr(sb, bindex);
32287 +               if (!au_br_rdonly(br) && br->br_wbr->wbr_bytes > b) {
32288 +                       b = br->br_wbr->wbr_bytes;
32289 +                       err = bindex;
32290 +                       AuDbg("b%d, %llu\n", err, b);
32291 +               }
32292 +       }
32293 +
32294 +       if (err >= 0)
32295 +               err = au_wbr_nonopq(dentry, err);
32296 +
32297 +out_parent:
32298 +       dput(parent);
32299 +out:
32300 +       AuDbg("b%d\n", err);
32301 +       return err;
32302 +}
32303 +
32304 +/* ---------------------------------------------------------------------- */
32305 +
32306 +/*
32307 + * - top down parent
32308 + * - most free space with parent
32309 + * - most free space round-robin regardless parent
32310 + */
32311 +static int au_wbr_create_pmfsrr(struct dentry *dentry, unsigned int flags)
32312 +{
32313 +       int err;
32314 +       unsigned long long watermark;
32315 +       struct super_block *sb;
32316 +       struct au_branch *br;
32317 +       struct au_wbr_mfs *mfs;
32318 +
32319 +       err = au_wbr_create_pmfs(dentry, flags | AuWbr_PARENT);
32320 +       if (unlikely(err < 0))
32321 +               goto out;
32322 +
32323 +       sb = dentry->d_sb;
32324 +       br = au_sbr(sb, err);
32325 +       mfs = &au_sbi(sb)->si_wbr_mfs;
32326 +       mutex_lock(&mfs->mfs_lock);
32327 +       watermark = mfs->mfsrr_watermark;
32328 +       mutex_unlock(&mfs->mfs_lock);
32329 +       if (br->br_wbr->wbr_bytes < watermark)
32330 +               /* regardless the parent dir */
32331 +               err = au_wbr_create_mfsrr(dentry, flags);
32332 +
32333 +out:
32334 +       AuDbg("b%d\n", err);
32335 +       return err;
32336 +}
32337 +
32338 +/* ---------------------------------------------------------------------- */
32339 +
32340 +/* policies for copyup */
32341 +
32342 +/* top down parent */
32343 +static int au_wbr_copyup_tdp(struct dentry *dentry)
32344 +{
32345 +       return au_wbr_create_tdp(dentry, /*flags, anything is ok*/0);
32346 +}
32347 +
32348 +/* bottom up parent */
32349 +static int au_wbr_copyup_bup(struct dentry *dentry)
32350 +{
32351 +       int err;
32352 +       aufs_bindex_t bindex, btop;
32353 +       struct dentry *parent, *h_parent;
32354 +       struct super_block *sb;
32355 +
32356 +       err = -EROFS;
32357 +       sb = dentry->d_sb;
32358 +       parent = dget_parent(dentry);
32359 +       btop = au_dbtop(parent);
32360 +       for (bindex = au_dbtop(dentry); bindex >= btop; bindex--) {
32361 +               h_parent = au_h_dptr(parent, bindex);
32362 +               if (!h_parent || d_is_negative(h_parent))
32363 +                       continue;
32364 +
32365 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
32366 +                       err = bindex;
32367 +                       break;
32368 +               }
32369 +       }
32370 +       dput(parent);
32371 +
32372 +       /* bottom up here */
32373 +       if (unlikely(err < 0))
32374 +               err = au_wbr_bu(sb, btop - 1);
32375 +
32376 +       AuDbg("b%d\n", err);
32377 +       return err;
32378 +}
32379 +
32380 +/* bottom up */
32381 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t btop)
32382 +{
32383 +       int err;
32384 +
32385 +       err = au_wbr_bu(dentry->d_sb, btop);
32386 +       AuDbg("b%d\n", err);
32387 +       if (err > btop)
32388 +               err = au_wbr_nonopq(dentry, err);
32389 +
32390 +       AuDbg("b%d\n", err);
32391 +       return err;
32392 +}
32393 +
32394 +static int au_wbr_copyup_bu(struct dentry *dentry)
32395 +{
32396 +       int err;
32397 +       aufs_bindex_t btop;
32398 +
32399 +       btop = au_dbtop(dentry);
32400 +       err = au_wbr_do_copyup_bu(dentry, btop);
32401 +       return err;
32402 +}
32403 +
32404 +/* ---------------------------------------------------------------------- */
32405 +
32406 +struct au_wbr_copyup_operations au_wbr_copyup_ops[] = {
32407 +       [AuWbrCopyup_TDP] = {
32408 +               .copyup = au_wbr_copyup_tdp
32409 +       },
32410 +       [AuWbrCopyup_BUP] = {
32411 +               .copyup = au_wbr_copyup_bup
32412 +       },
32413 +       [AuWbrCopyup_BU] = {
32414 +               .copyup = au_wbr_copyup_bu
32415 +       }
32416 +};
32417 +
32418 +struct au_wbr_create_operations au_wbr_create_ops[] = {
32419 +       [AuWbrCreate_TDP] = {
32420 +               .create = au_wbr_create_tdp
32421 +       },
32422 +       [AuWbrCreate_RR] = {
32423 +               .create = au_wbr_create_rr,
32424 +               .init   = au_wbr_create_init_rr
32425 +       },
32426 +       [AuWbrCreate_MFS] = {
32427 +               .create = au_wbr_create_mfs,
32428 +               .init   = au_wbr_create_init_mfs,
32429 +               .fin    = au_wbr_create_fin_mfs
32430 +       },
32431 +       [AuWbrCreate_MFSV] = {
32432 +               .create = au_wbr_create_mfs,
32433 +               .init   = au_wbr_create_init_mfs,
32434 +               .fin    = au_wbr_create_fin_mfs
32435 +       },
32436 +       [AuWbrCreate_MFSRR] = {
32437 +               .create = au_wbr_create_mfsrr,
32438 +               .init   = au_wbr_create_init_mfsrr,
32439 +               .fin    = au_wbr_create_fin_mfs
32440 +       },
32441 +       [AuWbrCreate_MFSRRV] = {
32442 +               .create = au_wbr_create_mfsrr,
32443 +               .init   = au_wbr_create_init_mfsrr,
32444 +               .fin    = au_wbr_create_fin_mfs
32445 +       },
32446 +       [AuWbrCreate_PMFS] = {
32447 +               .create = au_wbr_create_pmfs,
32448 +               .init   = au_wbr_create_init_mfs,
32449 +               .fin    = au_wbr_create_fin_mfs
32450 +       },
32451 +       [AuWbrCreate_PMFSV] = {
32452 +               .create = au_wbr_create_pmfs,
32453 +               .init   = au_wbr_create_init_mfs,
32454 +               .fin    = au_wbr_create_fin_mfs
32455 +       },
32456 +       [AuWbrCreate_PMFSRR] = {
32457 +               .create = au_wbr_create_pmfsrr,
32458 +               .init   = au_wbr_create_init_mfsrr,
32459 +               .fin    = au_wbr_create_fin_mfs
32460 +       },
32461 +       [AuWbrCreate_PMFSRRV] = {
32462 +               .create = au_wbr_create_pmfsrr,
32463 +               .init   = au_wbr_create_init_mfsrr,
32464 +               .fin    = au_wbr_create_fin_mfs
32465 +       }
32466 +};
32467 diff -urN /usr/share/empty/fs/aufs/whout.c linux/fs/aufs/whout.c
32468 --- /usr/share/empty/fs/aufs/whout.c    1970-01-01 01:00:00.000000000 +0100
32469 +++ linux/fs/aufs/whout.c       2016-10-09 16:55:36.496035060 +0200
32470 @@ -0,0 +1,1060 @@
32471 +/*
32472 + * Copyright (C) 2005-2016 Junjiro R. Okajima
32473 + *
32474 + * This program, aufs is free software; you can redistribute it and/or modify
32475 + * it under the terms of the GNU General Public License as published by
32476 + * the Free Software Foundation; either version 2 of the License, or
32477 + * (at your option) any later version.
32478 + *
32479 + * This program is distributed in the hope that it will be useful,
32480 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32481 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32482 + * GNU General Public License for more details.
32483 + *
32484 + * You should have received a copy of the GNU General Public License
32485 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32486 + */
32487 +
32488 +/*
32489 + * whiteout for logical deletion and opaque directory
32490 + */
32491 +
32492 +#include "aufs.h"
32493 +
32494 +#define WH_MASK                        S_IRUGO
32495 +
32496 +/*
32497 + * If a directory contains this file, then it is opaque.  We start with the
32498 + * .wh. flag so that it is blocked by lookup.
32499 + */
32500 +static struct qstr diropq_name = QSTR_INIT(AUFS_WH_DIROPQ,
32501 +                                          sizeof(AUFS_WH_DIROPQ) - 1);
32502 +
32503 +/*
32504 + * generate whiteout name, which is NOT terminated by NULL.
32505 + * @name: original d_name.name
32506 + * @len: original d_name.len
32507 + * @wh: whiteout qstr
32508 + * returns zero when succeeds, otherwise error.
32509 + * succeeded value as wh->name should be freed by kfree().
32510 + */
32511 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name)
32512 +{
32513 +       char *p;
32514 +
32515 +       if (unlikely(name->len > PATH_MAX - AUFS_WH_PFX_LEN))
32516 +               return -ENAMETOOLONG;
32517 +
32518 +       wh->len = name->len + AUFS_WH_PFX_LEN;
32519 +       p = kmalloc(wh->len, GFP_NOFS);
32520 +       wh->name = p;
32521 +       if (p) {
32522 +               memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
32523 +               memcpy(p + AUFS_WH_PFX_LEN, name->name, name->len);
32524 +               /* smp_mb(); */
32525 +               return 0;
32526 +       }
32527 +       return -ENOMEM;
32528 +}
32529 +
32530 +/* ---------------------------------------------------------------------- */
32531 +
32532 +/*
32533 + * test if the @wh_name exists under @h_parent.
32534 + * @try_sio specifies the necessary of super-io.
32535 + */
32536 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio)
32537 +{
32538 +       int err;
32539 +       struct dentry *wh_dentry;
32540 +
32541 +       if (!try_sio)
32542 +               wh_dentry = vfsub_lkup_one(wh_name, h_parent);
32543 +       else
32544 +               wh_dentry = au_sio_lkup_one(wh_name, h_parent);
32545 +       err = PTR_ERR(wh_dentry);
32546 +       if (IS_ERR(wh_dentry)) {
32547 +               if (err == -ENAMETOOLONG)
32548 +                       err = 0;
32549 +               goto out;
32550 +       }
32551 +
32552 +       err = 0;
32553 +       if (d_is_negative(wh_dentry))
32554 +               goto out_wh; /* success */
32555 +
32556 +       err = 1;
32557 +       if (d_is_reg(wh_dentry))
32558 +               goto out_wh; /* success */
32559 +
32560 +       err = -EIO;
32561 +       AuIOErr("%pd Invalid whiteout entry type 0%o.\n",
32562 +               wh_dentry, d_inode(wh_dentry)->i_mode);
32563 +
32564 +out_wh:
32565 +       dput(wh_dentry);
32566 +out:
32567 +       return err;
32568 +}
32569 +
32570 +/*
32571 + * test if the @h_dentry sets opaque or not.
32572 + */
32573 +int au_diropq_test(struct dentry *h_dentry)
32574 +{
32575 +       int err;
32576 +       struct inode *h_dir;
32577 +
32578 +       h_dir = d_inode(h_dentry);
32579 +       err = au_wh_test(h_dentry, &diropq_name,
32580 +                        au_test_h_perm_sio(h_dir, MAY_EXEC));
32581 +       return err;
32582 +}
32583 +
32584 +/*
32585 + * returns a negative dentry whose name is unique and temporary.
32586 + */
32587 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
32588 +                            struct qstr *prefix)
32589 +{
32590 +       struct dentry *dentry;
32591 +       int i;
32592 +       char defname[NAME_MAX - AUFS_MAX_NAMELEN + DNAME_INLINE_LEN + 1],
32593 +               *name, *p;
32594 +       /* strict atomic_t is unnecessary here */
32595 +       static unsigned short cnt;
32596 +       struct qstr qs;
32597 +
32598 +       BUILD_BUG_ON(sizeof(cnt) * 2 > AUFS_WH_TMP_LEN);
32599 +
32600 +       name = defname;
32601 +       qs.len = sizeof(defname) - DNAME_INLINE_LEN + prefix->len - 1;
32602 +       if (unlikely(prefix->len > DNAME_INLINE_LEN)) {
32603 +               dentry = ERR_PTR(-ENAMETOOLONG);
32604 +               if (unlikely(qs.len > NAME_MAX))
32605 +                       goto out;
32606 +               dentry = ERR_PTR(-ENOMEM);
32607 +               name = kmalloc(qs.len + 1, GFP_NOFS);
32608 +               if (unlikely(!name))
32609 +                       goto out;
32610 +       }
32611 +
32612 +       /* doubly whiteout-ed */
32613 +       memcpy(name, AUFS_WH_PFX AUFS_WH_PFX, AUFS_WH_PFX_LEN * 2);
32614 +       p = name + AUFS_WH_PFX_LEN * 2;
32615 +       memcpy(p, prefix->name, prefix->len);
32616 +       p += prefix->len;
32617 +       *p++ = '.';
32618 +       AuDebugOn(name + qs.len + 1 - p <= AUFS_WH_TMP_LEN);
32619 +
32620 +       qs.name = name;
32621 +       for (i = 0; i < 3; i++) {
32622 +               sprintf(p, "%.*x", AUFS_WH_TMP_LEN, cnt++);
32623 +               dentry = au_sio_lkup_one(&qs, h_parent);
32624 +               if (IS_ERR(dentry) || d_is_negative(dentry))
32625 +                       goto out_name;
32626 +               dput(dentry);
32627 +       }
32628 +       /* pr_warn("could not get random name\n"); */
32629 +       dentry = ERR_PTR(-EEXIST);
32630 +       AuDbg("%.*s\n", AuLNPair(&qs));
32631 +       BUG();
32632 +
32633 +out_name:
32634 +       if (name != defname)
32635 +               au_delayed_kfree(name);
32636 +out:
32637 +       AuTraceErrPtr(dentry);
32638 +       return dentry;
32639 +}
32640 +
32641 +/*
32642 + * rename the @h_dentry on @br to the whiteouted temporary name.
32643 + */
32644 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br)
32645 +{
32646 +       int err;
32647 +       struct path h_path = {
32648 +               .mnt = au_br_mnt(br)
32649 +       };
32650 +       struct inode *h_dir, *delegated;
32651 +       struct dentry *h_parent;
32652 +
32653 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
32654 +       h_dir = d_inode(h_parent);
32655 +       IMustLock(h_dir);
32656 +
32657 +       h_path.dentry = au_whtmp_lkup(h_parent, br, &h_dentry->d_name);
32658 +       err = PTR_ERR(h_path.dentry);
32659 +       if (IS_ERR(h_path.dentry))
32660 +               goto out;
32661 +
32662 +       /* under the same dir, no need to lock_rename() */
32663 +       delegated = NULL;
32664 +       err = vfsub_rename(h_dir, h_dentry, h_dir, &h_path, &delegated);
32665 +       AuTraceErr(err);
32666 +       if (unlikely(err == -EWOULDBLOCK)) {
32667 +               pr_warn("cannot retry for NFSv4 delegation"
32668 +                       " for an internal rename\n");
32669 +               iput(delegated);
32670 +       }
32671 +       dput(h_path.dentry);
32672 +
32673 +out:
32674 +       AuTraceErr(err);
32675 +       return err;
32676 +}
32677 +
32678 +/* ---------------------------------------------------------------------- */
32679 +/*
32680 + * functions for removing a whiteout
32681 + */
32682 +
32683 +static int do_unlink_wh(struct inode *h_dir, struct path *h_path)
32684 +{
32685 +       int err, force;
32686 +       struct inode *delegated;
32687 +
32688 +       /*
32689 +        * forces superio when the dir has a sticky bit.
32690 +        * this may be a violation of unix fs semantics.
32691 +        */
32692 +       force = (h_dir->i_mode & S_ISVTX)
32693 +               && !uid_eq(current_fsuid(), d_inode(h_path->dentry)->i_uid);
32694 +       delegated = NULL;
32695 +       err = vfsub_unlink(h_dir, h_path, &delegated, force);
32696 +       if (unlikely(err == -EWOULDBLOCK)) {
32697 +               pr_warn("cannot retry for NFSv4 delegation"
32698 +                       " for an internal unlink\n");
32699 +               iput(delegated);
32700 +       }
32701 +       return err;
32702 +}
32703 +
32704 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
32705 +                       struct dentry *dentry)
32706 +{
32707 +       int err;
32708 +
32709 +       err = do_unlink_wh(h_dir, h_path);
32710 +       if (!err && dentry)
32711 +               au_set_dbwh(dentry, -1);
32712 +
32713 +       return err;
32714 +}
32715 +
32716 +static int unlink_wh_name(struct dentry *h_parent, struct qstr *wh,
32717 +                         struct au_branch *br)
32718 +{
32719 +       int err;
32720 +       struct path h_path = {
32721 +               .mnt = au_br_mnt(br)
32722 +       };
32723 +
32724 +       err = 0;
32725 +       h_path.dentry = vfsub_lkup_one(wh, h_parent);
32726 +       if (IS_ERR(h_path.dentry))
32727 +               err = PTR_ERR(h_path.dentry);
32728 +       else {
32729 +               if (d_is_reg(h_path.dentry))
32730 +                       err = do_unlink_wh(d_inode(h_parent), &h_path);
32731 +               dput(h_path.dentry);
32732 +       }
32733 +
32734 +       return err;
32735 +}
32736 +
32737 +/* ---------------------------------------------------------------------- */
32738 +/*
32739 + * initialize/clean whiteout for a branch
32740 + */
32741 +
32742 +static void au_wh_clean(struct inode *h_dir, struct path *whpath,
32743 +                       const int isdir)
32744 +{
32745 +       int err;
32746 +       struct inode *delegated;
32747 +
32748 +       if (d_is_negative(whpath->dentry))
32749 +               return;
32750 +
32751 +       if (isdir)
32752 +               err = vfsub_rmdir(h_dir, whpath);
32753 +       else {
32754 +               delegated = NULL;
32755 +               err = vfsub_unlink(h_dir, whpath, &delegated, /*force*/0);
32756 +               if (unlikely(err == -EWOULDBLOCK)) {
32757 +                       pr_warn("cannot retry for NFSv4 delegation"
32758 +                               " for an internal unlink\n");
32759 +                       iput(delegated);
32760 +               }
32761 +       }
32762 +       if (unlikely(err))
32763 +               pr_warn("failed removing %pd (%d), ignored.\n",
32764 +                       whpath->dentry, err);
32765 +}
32766 +
32767 +static int test_linkable(struct dentry *h_root)
32768 +{
32769 +       struct inode *h_dir = d_inode(h_root);
32770 +
32771 +       if (h_dir->i_op->link)
32772 +               return 0;
32773 +
32774 +       pr_err("%pd (%s) doesn't support link(2), use noplink and rw+nolwh\n",
32775 +              h_root, au_sbtype(h_root->d_sb));
32776 +       return -ENOSYS;
32777 +}
32778 +
32779 +/* todo: should this mkdir be done in /sbin/mount.aufs helper? */
32780 +static int au_whdir(struct inode *h_dir, struct path *path)
32781 +{
32782 +       int err;
32783 +
32784 +       err = -EEXIST;
32785 +       if (d_is_negative(path->dentry)) {
32786 +               int mode = S_IRWXU;
32787 +
32788 +               if (au_test_nfs(path->dentry->d_sb))
32789 +                       mode |= S_IXUGO;
32790 +               err = vfsub_mkdir(h_dir, path, mode);
32791 +       } else if (d_is_dir(path->dentry))
32792 +               err = 0;
32793 +       else
32794 +               pr_err("unknown %pd exists\n", path->dentry);
32795 +
32796 +       return err;
32797 +}
32798 +
32799 +struct au_wh_base {
32800 +       const struct qstr *name;
32801 +       struct dentry *dentry;
32802 +};
32803 +
32804 +static void au_wh_init_ro(struct inode *h_dir, struct au_wh_base base[],
32805 +                         struct path *h_path)
32806 +{
32807 +       h_path->dentry = base[AuBrWh_BASE].dentry;
32808 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
32809 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
32810 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
32811 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
32812 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
32813 +}
32814 +
32815 +/*
32816 + * returns tri-state,
32817 + * minus: error, caller should print the message
32818 + * zero: succuess
32819 + * plus: error, caller should NOT print the message
32820 + */
32821 +static int au_wh_init_rw_nolink(struct dentry *h_root, struct au_wbr *wbr,
32822 +                               int do_plink, struct au_wh_base base[],
32823 +                               struct path *h_path)
32824 +{
32825 +       int err;
32826 +       struct inode *h_dir;
32827 +
32828 +       h_dir = d_inode(h_root);
32829 +       h_path->dentry = base[AuBrWh_BASE].dentry;
32830 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
32831 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
32832 +       if (do_plink) {
32833 +               err = test_linkable(h_root);
32834 +               if (unlikely(err)) {
32835 +                       err = 1;
32836 +                       goto out;
32837 +               }
32838 +
32839 +               err = au_whdir(h_dir, h_path);
32840 +               if (unlikely(err))
32841 +                       goto out;
32842 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
32843 +       } else
32844 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
32845 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
32846 +       err = au_whdir(h_dir, h_path);
32847 +       if (unlikely(err))
32848 +               goto out;
32849 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
32850 +
32851 +out:
32852 +       return err;
32853 +}
32854 +
32855 +/*
32856 + * for the moment, aufs supports the branch filesystem which does not support
32857 + * link(2). testing on FAT which does not support i_op->setattr() fully either,
32858 + * copyup failed. finally, such filesystem will not be used as the writable
32859 + * branch.
32860 + *
32861 + * returns tri-state, see above.
32862 + */
32863 +static int au_wh_init_rw(struct dentry *h_root, struct au_wbr *wbr,
32864 +                        int do_plink, struct au_wh_base base[],
32865 +                        struct path *h_path)
32866 +{
32867 +       int err;
32868 +       struct inode *h_dir;
32869 +
32870 +       WbrWhMustWriteLock(wbr);
32871 +
32872 +       err = test_linkable(h_root);
32873 +       if (unlikely(err)) {
32874 +               err = 1;
32875 +               goto out;
32876 +       }
32877 +
32878 +       /*
32879 +        * todo: should this create be done in /sbin/mount.aufs helper?
32880 +        */
32881 +       err = -EEXIST;
32882 +       h_dir = d_inode(h_root);
32883 +       if (d_is_negative(base[AuBrWh_BASE].dentry)) {
32884 +               h_path->dentry = base[AuBrWh_BASE].dentry;
32885 +               err = vfsub_create(h_dir, h_path, WH_MASK, /*want_excl*/true);
32886 +       } else if (d_is_reg(base[AuBrWh_BASE].dentry))
32887 +               err = 0;
32888 +       else
32889 +               pr_err("unknown %pd2 exists\n", base[AuBrWh_BASE].dentry);
32890 +       if (unlikely(err))
32891 +               goto out;
32892 +
32893 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
32894 +       if (do_plink) {
32895 +               err = au_whdir(h_dir, h_path);
32896 +               if (unlikely(err))
32897 +                       goto out;
32898 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
32899 +       } else
32900 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
32901 +       wbr->wbr_whbase = dget(base[AuBrWh_BASE].dentry);
32902 +
32903 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
32904 +       err = au_whdir(h_dir, h_path);
32905 +       if (unlikely(err))
32906 +               goto out;
32907 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
32908 +
32909 +out:
32910 +       return err;
32911 +}
32912 +
32913 +/*
32914 + * initialize the whiteout base file/dir for @br.
32915 + */
32916 +int au_wh_init(struct au_branch *br, struct super_block *sb)
32917 +{
32918 +       int err, i;
32919 +       const unsigned char do_plink
32920 +               = !!au_opt_test(au_mntflags(sb), PLINK);
32921 +       struct inode *h_dir;
32922 +       struct path path = br->br_path;
32923 +       struct dentry *h_root = path.dentry;
32924 +       struct au_wbr *wbr = br->br_wbr;
32925 +       static const struct qstr base_name[] = {
32926 +               [AuBrWh_BASE] = QSTR_INIT(AUFS_BASE_NAME,
32927 +                                         sizeof(AUFS_BASE_NAME) - 1),
32928 +               [AuBrWh_PLINK] = QSTR_INIT(AUFS_PLINKDIR_NAME,
32929 +                                          sizeof(AUFS_PLINKDIR_NAME) - 1),
32930 +               [AuBrWh_ORPH] = QSTR_INIT(AUFS_ORPHDIR_NAME,
32931 +                                         sizeof(AUFS_ORPHDIR_NAME) - 1)
32932 +       };
32933 +       struct au_wh_base base[] = {
32934 +               [AuBrWh_BASE] = {
32935 +                       .name   = base_name + AuBrWh_BASE,
32936 +                       .dentry = NULL
32937 +               },
32938 +               [AuBrWh_PLINK] = {
32939 +                       .name   = base_name + AuBrWh_PLINK,
32940 +                       .dentry = NULL
32941 +               },
32942 +               [AuBrWh_ORPH] = {
32943 +                       .name   = base_name + AuBrWh_ORPH,
32944 +                       .dentry = NULL
32945 +               }
32946 +       };
32947 +
32948 +       if (wbr)
32949 +               WbrWhMustWriteLock(wbr);
32950 +
32951 +       for (i = 0; i < AuBrWh_Last; i++) {
32952 +               /* doubly whiteouted */
32953 +               struct dentry *d;
32954 +
32955 +               d = au_wh_lkup(h_root, (void *)base[i].name, br);
32956 +               err = PTR_ERR(d);
32957 +               if (IS_ERR(d))
32958 +                       goto out;
32959 +
32960 +               base[i].dentry = d;
32961 +               AuDebugOn(wbr
32962 +                         && wbr->wbr_wh[i]
32963 +                         && wbr->wbr_wh[i] != base[i].dentry);
32964 +       }
32965 +
32966 +       if (wbr)
32967 +               for (i = 0; i < AuBrWh_Last; i++) {
32968 +                       dput(wbr->wbr_wh[i]);
32969 +                       wbr->wbr_wh[i] = NULL;
32970 +               }
32971 +
32972 +       err = 0;
32973 +       if (!au_br_writable(br->br_perm)) {
32974 +               h_dir = d_inode(h_root);
32975 +               au_wh_init_ro(h_dir, base, &path);
32976 +       } else if (!au_br_wh_linkable(br->br_perm)) {
32977 +               err = au_wh_init_rw_nolink(h_root, wbr, do_plink, base, &path);
32978 +               if (err > 0)
32979 +                       goto out;
32980 +               else if (err)
32981 +                       goto out_err;
32982 +       } else {
32983 +               err = au_wh_init_rw(h_root, wbr, do_plink, base, &path);
32984 +               if (err > 0)
32985 +                       goto out;
32986 +               else if (err)
32987 +                       goto out_err;
32988 +       }
32989 +       goto out; /* success */
32990 +
32991 +out_err:
32992 +       pr_err("an error(%d) on the writable branch %pd(%s)\n",
32993 +              err, h_root, au_sbtype(h_root->d_sb));
32994 +out:
32995 +       for (i = 0; i < AuBrWh_Last; i++)
32996 +               dput(base[i].dentry);
32997 +       return err;
32998 +}
32999 +
33000 +/* ---------------------------------------------------------------------- */
33001 +/*
33002 + * whiteouts are all hard-linked usually.
33003 + * when its link count reaches a ceiling, we create a new whiteout base
33004 + * asynchronously.
33005 + */
33006 +
33007 +struct reinit_br_wh {
33008 +       struct super_block *sb;
33009 +       struct au_branch *br;
33010 +};
33011 +
33012 +static void reinit_br_wh(void *arg)
33013 +{
33014 +       int err;
33015 +       aufs_bindex_t bindex;
33016 +       struct path h_path;
33017 +       struct reinit_br_wh *a = arg;
33018 +       struct au_wbr *wbr;
33019 +       struct inode *dir, *delegated;
33020 +       struct dentry *h_root;
33021 +       struct au_hinode *hdir;
33022 +
33023 +       err = 0;
33024 +       wbr = a->br->br_wbr;
33025 +       /* big aufs lock */
33026 +       si_noflush_write_lock(a->sb);
33027 +       if (!au_br_writable(a->br->br_perm))
33028 +               goto out;
33029 +       bindex = au_br_index(a->sb, a->br->br_id);
33030 +       if (unlikely(bindex < 0))
33031 +               goto out;
33032 +
33033 +       di_read_lock_parent(a->sb->s_root, AuLock_IR);
33034 +       dir = d_inode(a->sb->s_root);
33035 +       hdir = au_hi(dir, bindex);
33036 +       h_root = au_h_dptr(a->sb->s_root, bindex);
33037 +       AuDebugOn(h_root != au_br_dentry(a->br));
33038 +
33039 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
33040 +       wbr_wh_write_lock(wbr);
33041 +       err = au_h_verify(wbr->wbr_whbase, au_opt_udba(a->sb), hdir->hi_inode,
33042 +                         h_root, a->br);
33043 +       if (!err) {
33044 +               h_path.dentry = wbr->wbr_whbase;
33045 +               h_path.mnt = au_br_mnt(a->br);
33046 +               delegated = NULL;
33047 +               err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated,
33048 +                                  /*force*/0);
33049 +               if (unlikely(err == -EWOULDBLOCK)) {
33050 +                       pr_warn("cannot retry for NFSv4 delegation"
33051 +                               " for an internal unlink\n");
33052 +                       iput(delegated);
33053 +               }
33054 +       } else {
33055 +               pr_warn("%pd is moved, ignored\n", wbr->wbr_whbase);
33056 +               err = 0;
33057 +       }
33058 +       dput(wbr->wbr_whbase);
33059 +       wbr->wbr_whbase = NULL;
33060 +       if (!err)
33061 +               err = au_wh_init(a->br, a->sb);
33062 +       wbr_wh_write_unlock(wbr);
33063 +       au_hn_inode_unlock(hdir);
33064 +       di_read_unlock(a->sb->s_root, AuLock_IR);
33065 +       if (!err)
33066 +               au_fhsm_wrote(a->sb, bindex, /*force*/0);
33067 +
33068 +out:
33069 +       if (wbr)
33070 +               atomic_dec(&wbr->wbr_wh_running);
33071 +       au_br_put(a->br);
33072 +       si_write_unlock(a->sb);
33073 +       au_nwt_done(&au_sbi(a->sb)->si_nowait);
33074 +       au_delayed_kfree(arg);
33075 +       if (unlikely(err))
33076 +               AuIOErr("err %d\n", err);
33077 +}
33078 +
33079 +static void kick_reinit_br_wh(struct super_block *sb, struct au_branch *br)
33080 +{
33081 +       int do_dec, wkq_err;
33082 +       struct reinit_br_wh *arg;
33083 +
33084 +       do_dec = 1;
33085 +       if (atomic_inc_return(&br->br_wbr->wbr_wh_running) != 1)
33086 +               goto out;
33087 +
33088 +       /* ignore ENOMEM */
33089 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
33090 +       if (arg) {
33091 +               /*
33092 +                * dec(wh_running), kfree(arg) and dec(br_count)
33093 +                * in reinit function
33094 +                */
33095 +               arg->sb = sb;
33096 +               arg->br = br;
33097 +               au_br_get(br);
33098 +               wkq_err = au_wkq_nowait(reinit_br_wh, arg, sb, /*flags*/0);
33099 +               if (unlikely(wkq_err)) {
33100 +                       atomic_dec(&br->br_wbr->wbr_wh_running);
33101 +                       au_br_put(br);
33102 +                       au_delayed_kfree(arg);
33103 +               }
33104 +               do_dec = 0;
33105 +       }
33106 +
33107 +out:
33108 +       if (do_dec)
33109 +               atomic_dec(&br->br_wbr->wbr_wh_running);
33110 +}
33111 +
33112 +/* ---------------------------------------------------------------------- */
33113 +
33114 +/*
33115 + * create the whiteout @wh.
33116 + */
33117 +static int link_or_create_wh(struct super_block *sb, aufs_bindex_t bindex,
33118 +                            struct dentry *wh)
33119 +{
33120 +       int err;
33121 +       struct path h_path = {
33122 +               .dentry = wh
33123 +       };
33124 +       struct au_branch *br;
33125 +       struct au_wbr *wbr;
33126 +       struct dentry *h_parent;
33127 +       struct inode *h_dir, *delegated;
33128 +
33129 +       h_parent = wh->d_parent; /* dir inode is locked */
33130 +       h_dir = d_inode(h_parent);
33131 +       IMustLock(h_dir);
33132 +
33133 +       br = au_sbr(sb, bindex);
33134 +       h_path.mnt = au_br_mnt(br);
33135 +       wbr = br->br_wbr;
33136 +       wbr_wh_read_lock(wbr);
33137 +       if (wbr->wbr_whbase) {
33138 +               delegated = NULL;
33139 +               err = vfsub_link(wbr->wbr_whbase, h_dir, &h_path, &delegated);
33140 +               if (unlikely(err == -EWOULDBLOCK)) {
33141 +                       pr_warn("cannot retry for NFSv4 delegation"
33142 +                               " for an internal link\n");
33143 +                       iput(delegated);
33144 +               }
33145 +               if (!err || err != -EMLINK)
33146 +                       goto out;
33147 +
33148 +               /* link count full. re-initialize br_whbase. */
33149 +               kick_reinit_br_wh(sb, br);
33150 +       }
33151 +
33152 +       /* return this error in this context */
33153 +       err = vfsub_create(h_dir, &h_path, WH_MASK, /*want_excl*/true);
33154 +       if (!err)
33155 +               au_fhsm_wrote(sb, bindex, /*force*/0);
33156 +
33157 +out:
33158 +       wbr_wh_read_unlock(wbr);
33159 +       return err;
33160 +}
33161 +
33162 +/* ---------------------------------------------------------------------- */
33163 +
33164 +/*
33165 + * create or remove the diropq.
33166 + */
33167 +static struct dentry *do_diropq(struct dentry *dentry, aufs_bindex_t bindex,
33168 +                               unsigned int flags)
33169 +{
33170 +       struct dentry *opq_dentry, *h_dentry;
33171 +       struct super_block *sb;
33172 +       struct au_branch *br;
33173 +       int err;
33174 +
33175 +       sb = dentry->d_sb;
33176 +       br = au_sbr(sb, bindex);
33177 +       h_dentry = au_h_dptr(dentry, bindex);
33178 +       opq_dentry = vfsub_lkup_one(&diropq_name, h_dentry);
33179 +       if (IS_ERR(opq_dentry))
33180 +               goto out;
33181 +
33182 +       if (au_ftest_diropq(flags, CREATE)) {
33183 +               err = link_or_create_wh(sb, bindex, opq_dentry);
33184 +               if (!err) {
33185 +                       au_set_dbdiropq(dentry, bindex);
33186 +                       goto out; /* success */
33187 +               }
33188 +       } else {
33189 +               struct path tmp = {
33190 +                       .dentry = opq_dentry,
33191 +                       .mnt    = au_br_mnt(br)
33192 +               };
33193 +               err = do_unlink_wh(au_h_iptr(d_inode(dentry), bindex), &tmp);
33194 +               if (!err)
33195 +                       au_set_dbdiropq(dentry, -1);
33196 +       }
33197 +       dput(opq_dentry);
33198 +       opq_dentry = ERR_PTR(err);
33199 +
33200 +out:
33201 +       return opq_dentry;
33202 +}
33203 +
33204 +struct do_diropq_args {
33205 +       struct dentry **errp;
33206 +       struct dentry *dentry;
33207 +       aufs_bindex_t bindex;
33208 +       unsigned int flags;
33209 +};
33210 +
33211 +static void call_do_diropq(void *args)
33212 +{
33213 +       struct do_diropq_args *a = args;
33214 +       *a->errp = do_diropq(a->dentry, a->bindex, a->flags);
33215 +}
33216 +
33217 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
33218 +                            unsigned int flags)
33219 +{
33220 +       struct dentry *diropq, *h_dentry;
33221 +
33222 +       h_dentry = au_h_dptr(dentry, bindex);
33223 +       if (!au_test_h_perm_sio(d_inode(h_dentry), MAY_EXEC | MAY_WRITE))
33224 +               diropq = do_diropq(dentry, bindex, flags);
33225 +       else {
33226 +               int wkq_err;
33227 +               struct do_diropq_args args = {
33228 +                       .errp           = &diropq,
33229 +                       .dentry         = dentry,
33230 +                       .bindex         = bindex,
33231 +                       .flags          = flags
33232 +               };
33233 +
33234 +               wkq_err = au_wkq_wait(call_do_diropq, &args);
33235 +               if (unlikely(wkq_err))
33236 +                       diropq = ERR_PTR(wkq_err);
33237 +       }
33238 +
33239 +       return diropq;
33240 +}
33241 +
33242 +/* ---------------------------------------------------------------------- */
33243 +
33244 +/*
33245 + * lookup whiteout dentry.
33246 + * @h_parent: lower parent dentry which must exist and be locked
33247 + * @base_name: name of dentry which will be whiteouted
33248 + * returns dentry for whiteout.
33249 + */
33250 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
33251 +                         struct au_branch *br)
33252 +{
33253 +       int err;
33254 +       struct qstr wh_name;
33255 +       struct dentry *wh_dentry;
33256 +
33257 +       err = au_wh_name_alloc(&wh_name, base_name);
33258 +       wh_dentry = ERR_PTR(err);
33259 +       if (!err) {
33260 +               wh_dentry = vfsub_lkup_one(&wh_name, h_parent);
33261 +               au_delayed_kfree(wh_name.name);
33262 +       }
33263 +       return wh_dentry;
33264 +}
33265 +
33266 +/*
33267 + * link/create a whiteout for @dentry on @bindex.
33268 + */
33269 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
33270 +                           struct dentry *h_parent)
33271 +{
33272 +       struct dentry *wh_dentry;
33273 +       struct super_block *sb;
33274 +       int err;
33275 +
33276 +       sb = dentry->d_sb;
33277 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, au_sbr(sb, bindex));
33278 +       if (!IS_ERR(wh_dentry) && d_is_negative(wh_dentry)) {
33279 +               err = link_or_create_wh(sb, bindex, wh_dentry);
33280 +               if (!err) {
33281 +                       au_set_dbwh(dentry, bindex);
33282 +                       au_fhsm_wrote(sb, bindex, /*force*/0);
33283 +               } else {
33284 +                       dput(wh_dentry);
33285 +                       wh_dentry = ERR_PTR(err);
33286 +               }
33287 +       }
33288 +
33289 +       return wh_dentry;
33290 +}
33291 +
33292 +/* ---------------------------------------------------------------------- */
33293 +
33294 +/* Delete all whiteouts in this directory on branch bindex. */
33295 +static int del_wh_children(struct dentry *h_dentry, struct au_nhash *whlist,
33296 +                          aufs_bindex_t bindex, struct au_branch *br)
33297 +{
33298 +       int err;
33299 +       unsigned long ul, n;
33300 +       struct qstr wh_name;
33301 +       char *p;
33302 +       struct hlist_head *head;
33303 +       struct au_vdir_wh *pos;
33304 +       struct au_vdir_destr *str;
33305 +
33306 +       err = -ENOMEM;
33307 +       p = (void *)__get_free_page(GFP_NOFS);
33308 +       wh_name.name = p;
33309 +       if (unlikely(!wh_name.name))
33310 +               goto out;
33311 +
33312 +       err = 0;
33313 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
33314 +       p += AUFS_WH_PFX_LEN;
33315 +       n = whlist->nh_num;
33316 +       head = whlist->nh_head;
33317 +       for (ul = 0; !err && ul < n; ul++, head++) {
33318 +               hlist_for_each_entry(pos, head, wh_hash) {
33319 +                       if (pos->wh_bindex != bindex)
33320 +                               continue;
33321 +
33322 +                       str = &pos->wh_str;
33323 +                       if (str->len + AUFS_WH_PFX_LEN <= PATH_MAX) {
33324 +                               memcpy(p, str->name, str->len);
33325 +                               wh_name.len = AUFS_WH_PFX_LEN + str->len;
33326 +                               err = unlink_wh_name(h_dentry, &wh_name, br);
33327 +                               if (!err)
33328 +                                       continue;
33329 +                               break;
33330 +                       }
33331 +                       AuIOErr("whiteout name too long %.*s\n",
33332 +                               str->len, str->name);
33333 +                       err = -EIO;
33334 +                       break;
33335 +               }
33336 +       }
33337 +       au_delayed_free_page((unsigned long)wh_name.name);
33338 +
33339 +out:
33340 +       return err;
33341 +}
33342 +
33343 +struct del_wh_children_args {
33344 +       int *errp;
33345 +       struct dentry *h_dentry;
33346 +       struct au_nhash *whlist;
33347 +       aufs_bindex_t bindex;
33348 +       struct au_branch *br;
33349 +};
33350 +
33351 +static void call_del_wh_children(void *args)
33352 +{
33353 +       struct del_wh_children_args *a = args;
33354 +       *a->errp = del_wh_children(a->h_dentry, a->whlist, a->bindex, a->br);
33355 +}
33356 +
33357 +/* ---------------------------------------------------------------------- */
33358 +
33359 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp)
33360 +{
33361 +       struct au_whtmp_rmdir *whtmp;
33362 +       int err;
33363 +       unsigned int rdhash;
33364 +
33365 +       SiMustAnyLock(sb);
33366 +
33367 +       whtmp = kzalloc(sizeof(*whtmp), gfp);
33368 +       if (unlikely(!whtmp)) {
33369 +               whtmp = ERR_PTR(-ENOMEM);
33370 +               goto out;
33371 +       }
33372 +
33373 +       /* no estimation for dir size */
33374 +       rdhash = au_sbi(sb)->si_rdhash;
33375 +       if (!rdhash)
33376 +               rdhash = AUFS_RDHASH_DEF;
33377 +       err = au_nhash_alloc(&whtmp->whlist, rdhash, gfp);
33378 +       if (unlikely(err)) {
33379 +               au_delayed_kfree(whtmp);
33380 +               whtmp = ERR_PTR(err);
33381 +       }
33382 +
33383 +out:
33384 +       return whtmp;
33385 +}
33386 +
33387 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp)
33388 +{
33389 +       if (whtmp->br)
33390 +               au_br_put(whtmp->br);
33391 +       dput(whtmp->wh_dentry);
33392 +       iput(whtmp->dir);
33393 +       au_nhash_wh_free(&whtmp->whlist);
33394 +       au_delayed_kfree(whtmp);
33395 +}
33396 +
33397 +/*
33398 + * rmdir the whiteouted temporary named dir @h_dentry.
33399 + * @whlist: whiteouted children.
33400 + */
33401 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
33402 +                  struct dentry *wh_dentry, struct au_nhash *whlist)
33403 +{
33404 +       int err;
33405 +       unsigned int h_nlink;
33406 +       struct path h_tmp;
33407 +       struct inode *wh_inode, *h_dir;
33408 +       struct au_branch *br;
33409 +
33410 +       h_dir = d_inode(wh_dentry->d_parent); /* dir inode is locked */
33411 +       IMustLock(h_dir);
33412 +
33413 +       br = au_sbr(dir->i_sb, bindex);
33414 +       wh_inode = d_inode(wh_dentry);
33415 +       inode_lock_nested(wh_inode, AuLsc_I_CHILD);
33416 +
33417 +       /*
33418 +        * someone else might change some whiteouts while we were sleeping.
33419 +        * it means this whlist may have an obsoleted entry.
33420 +        */
33421 +       if (!au_test_h_perm_sio(wh_inode, MAY_EXEC | MAY_WRITE))
33422 +               err = del_wh_children(wh_dentry, whlist, bindex, br);
33423 +       else {
33424 +               int wkq_err;
33425 +               struct del_wh_children_args args = {
33426 +                       .errp           = &err,
33427 +                       .h_dentry       = wh_dentry,
33428 +                       .whlist         = whlist,
33429 +                       .bindex         = bindex,
33430 +                       .br             = br
33431 +               };
33432 +
33433 +               wkq_err = au_wkq_wait(call_del_wh_children, &args);
33434 +               if (unlikely(wkq_err))
33435 +                       err = wkq_err;
33436 +       }
33437 +       inode_unlock(wh_inode);
33438 +
33439 +       if (!err) {
33440 +               h_tmp.dentry = wh_dentry;
33441 +               h_tmp.mnt = au_br_mnt(br);
33442 +               h_nlink = h_dir->i_nlink;
33443 +               err = vfsub_rmdir(h_dir, &h_tmp);
33444 +               /* some fs doesn't change the parent nlink in some cases */
33445 +               h_nlink -= h_dir->i_nlink;
33446 +       }
33447 +
33448 +       if (!err) {
33449 +               if (au_ibtop(dir) == bindex) {
33450 +                       /* todo: dir->i_mutex is necessary */
33451 +                       au_cpup_attr_timesizes(dir);
33452 +                       if (h_nlink)
33453 +                               vfsub_drop_nlink(dir);
33454 +               }
33455 +               return 0; /* success */
33456 +       }
33457 +
33458 +       pr_warn("failed removing %pd(%d), ignored\n", wh_dentry, err);
33459 +       return err;
33460 +}
33461 +
33462 +static void call_rmdir_whtmp(void *args)
33463 +{
33464 +       int err;
33465 +       aufs_bindex_t bindex;
33466 +       struct au_whtmp_rmdir *a = args;
33467 +       struct super_block *sb;
33468 +       struct dentry *h_parent;
33469 +       struct inode *h_dir;
33470 +       struct au_hinode *hdir;
33471 +
33472 +       /* rmdir by nfsd may cause deadlock with this i_mutex */
33473 +       /* inode_lock(a->dir); */
33474 +       err = -EROFS;
33475 +       sb = a->dir->i_sb;
33476 +       si_read_lock(sb, !AuLock_FLUSH);
33477 +       if (!au_br_writable(a->br->br_perm))
33478 +               goto out;
33479 +       bindex = au_br_index(sb, a->br->br_id);
33480 +       if (unlikely(bindex < 0))
33481 +               goto out;
33482 +
33483 +       err = -EIO;
33484 +       ii_write_lock_parent(a->dir);
33485 +       h_parent = dget_parent(a->wh_dentry);
33486 +       h_dir = d_inode(h_parent);
33487 +       hdir = au_hi(a->dir, bindex);
33488 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
33489 +       if (unlikely(err))
33490 +               goto out_mnt;
33491 +       au_hn_inode_lock_nested(hdir, AuLsc_I_PARENT);
33492 +       err = au_h_verify(a->wh_dentry, au_opt_udba(sb), h_dir, h_parent,
33493 +                         a->br);
33494 +       if (!err)
33495 +               err = au_whtmp_rmdir(a->dir, bindex, a->wh_dentry, &a->whlist);
33496 +       au_hn_inode_unlock(hdir);
33497 +       vfsub_mnt_drop_write(au_br_mnt(a->br));
33498 +
33499 +out_mnt:
33500 +       dput(h_parent);
33501 +       ii_write_unlock(a->dir);
33502 +out:
33503 +       /* inode_unlock(a->dir); */
33504 +       au_whtmp_rmdir_free(a);
33505 +       si_read_unlock(sb);
33506 +       au_nwt_done(&au_sbi(sb)->si_nowait);
33507 +       if (unlikely(err))
33508 +               AuIOErr("err %d\n", err);
33509 +}
33510 +
33511 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
33512 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args)
33513 +{
33514 +       int wkq_err;
33515 +       struct super_block *sb;
33516 +
33517 +       IMustLock(dir);
33518 +
33519 +       /* all post-process will be done in do_rmdir_whtmp(). */
33520 +       sb = dir->i_sb;
33521 +       args->dir = au_igrab(dir);
33522 +       args->br = au_sbr(sb, bindex);
33523 +       au_br_get(args->br);
33524 +       args->wh_dentry = dget(wh_dentry);
33525 +       wkq_err = au_wkq_nowait(call_rmdir_whtmp, args, sb, /*flags*/0);
33526 +       if (unlikely(wkq_err)) {
33527 +               pr_warn("rmdir error %pd (%d), ignored\n", wh_dentry, wkq_err);
33528 +               au_whtmp_rmdir_free(args);
33529 +       }
33530 +}
33531 diff -urN /usr/share/empty/fs/aufs/whout.h linux/fs/aufs/whout.h
33532 --- /usr/share/empty/fs/aufs/whout.h    1970-01-01 01:00:00.000000000 +0100
33533 +++ linux/fs/aufs/whout.h       2016-10-09 16:55:36.496035060 +0200
33534 @@ -0,0 +1,85 @@
33535 +/*
33536 + * Copyright (C) 2005-2016 Junjiro R. Okajima
33537 + *
33538 + * This program, aufs is free software; you can redistribute it and/or modify
33539 + * it under the terms of the GNU General Public License as published by
33540 + * the Free Software Foundation; either version 2 of the License, or
33541 + * (at your option) any later version.
33542 + *
33543 + * This program is distributed in the hope that it will be useful,
33544 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33545 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33546 + * GNU General Public License for more details.
33547 + *
33548 + * You should have received a copy of the GNU General Public License
33549 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33550 + */
33551 +
33552 +/*
33553 + * whiteout for logical deletion and opaque directory
33554 + */
33555 +
33556 +#ifndef __AUFS_WHOUT_H__
33557 +#define __AUFS_WHOUT_H__
33558 +
33559 +#ifdef __KERNEL__
33560 +
33561 +#include "dir.h"
33562 +
33563 +/* whout.c */
33564 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name);
33565 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio);
33566 +int au_diropq_test(struct dentry *h_dentry);
33567 +struct au_branch;
33568 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
33569 +                            struct qstr *prefix);
33570 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br);
33571 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
33572 +                       struct dentry *dentry);
33573 +int au_wh_init(struct au_branch *br, struct super_block *sb);
33574 +
33575 +/* diropq flags */
33576 +#define AuDiropq_CREATE        1
33577 +#define au_ftest_diropq(flags, name)   ((flags) & AuDiropq_##name)
33578 +#define au_fset_diropq(flags, name) \
33579 +       do { (flags) |= AuDiropq_##name; } while (0)
33580 +#define au_fclr_diropq(flags, name) \
33581 +       do { (flags) &= ~AuDiropq_##name; } while (0)
33582 +
33583 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
33584 +                            unsigned int flags);
33585 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
33586 +                         struct au_branch *br);
33587 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
33588 +                           struct dentry *h_parent);
33589 +
33590 +/* real rmdir for the whiteout-ed dir */
33591 +struct au_whtmp_rmdir {
33592 +       struct inode *dir;
33593 +       struct au_branch *br;
33594 +       struct dentry *wh_dentry;
33595 +       struct au_nhash whlist;
33596 +};
33597 +
33598 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp);
33599 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp);
33600 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
33601 +                  struct dentry *wh_dentry, struct au_nhash *whlist);
33602 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
33603 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args);
33604 +
33605 +/* ---------------------------------------------------------------------- */
33606 +
33607 +static inline struct dentry *au_diropq_create(struct dentry *dentry,
33608 +                                             aufs_bindex_t bindex)
33609 +{
33610 +       return au_diropq_sio(dentry, bindex, AuDiropq_CREATE);
33611 +}
33612 +
33613 +static inline int au_diropq_remove(struct dentry *dentry, aufs_bindex_t bindex)
33614 +{
33615 +       return PTR_ERR(au_diropq_sio(dentry, bindex, !AuDiropq_CREATE));
33616 +}
33617 +
33618 +#endif /* __KERNEL__ */
33619 +#endif /* __AUFS_WHOUT_H__ */
33620 diff -urN /usr/share/empty/fs/aufs/wkq.c linux/fs/aufs/wkq.c
33621 --- /usr/share/empty/fs/aufs/wkq.c      1970-01-01 01:00:00.000000000 +0100
33622 +++ linux/fs/aufs/wkq.c 2016-10-09 16:55:36.496035060 +0200
33623 @@ -0,0 +1,213 @@
33624 +/*
33625 + * Copyright (C) 2005-2016 Junjiro R. Okajima
33626 + *
33627 + * This program, aufs is free software; you can redistribute it and/or modify
33628 + * it under the terms of the GNU General Public License as published by
33629 + * the Free Software Foundation; either version 2 of the License, or
33630 + * (at your option) any later version.
33631 + *
33632 + * This program is distributed in the hope that it will be useful,
33633 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33634 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33635 + * GNU General Public License for more details.
33636 + *
33637 + * You should have received a copy of the GNU General Public License
33638 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33639 + */
33640 +
33641 +/*
33642 + * workqueue for asynchronous/super-io operations
33643 + * todo: try new dredential scheme
33644 + */
33645 +
33646 +#include <linux/module.h>
33647 +#include "aufs.h"
33648 +
33649 +/* internal workqueue named AUFS_WKQ_NAME */
33650 +
33651 +static struct workqueue_struct *au_wkq;
33652 +
33653 +struct au_wkinfo {
33654 +       struct work_struct wk;
33655 +       struct kobject *kobj;
33656 +
33657 +       unsigned int flags; /* see wkq.h */
33658 +
33659 +       au_wkq_func_t func;
33660 +       void *args;
33661 +
33662 +       struct completion *comp;
33663 +};
33664 +
33665 +/* ---------------------------------------------------------------------- */
33666 +
33667 +static void wkq_func(struct work_struct *wk)
33668 +{
33669 +       struct au_wkinfo *wkinfo = container_of(wk, struct au_wkinfo, wk);
33670 +
33671 +       AuDebugOn(!uid_eq(current_fsuid(), GLOBAL_ROOT_UID));
33672 +       AuDebugOn(rlimit(RLIMIT_FSIZE) != RLIM_INFINITY);
33673 +
33674 +       wkinfo->func(wkinfo->args);
33675 +       if (au_ftest_wkq(wkinfo->flags, WAIT))
33676 +               complete(wkinfo->comp);
33677 +       else {
33678 +               kobject_put(wkinfo->kobj);
33679 +               module_put(THIS_MODULE); /* todo: ?? */
33680 +               au_delayed_kfree(wkinfo);
33681 +       }
33682 +}
33683 +
33684 +/*
33685 + * Since struct completion is large, try allocating it dynamically.
33686 + */
33687 +#if 1 /* defined(CONFIG_4KSTACKS) || defined(AuTest4KSTACKS) */
33688 +#define AuWkqCompDeclare(name) struct completion *comp = NULL
33689 +
33690 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
33691 +{
33692 +       *comp = kmalloc(sizeof(**comp), GFP_NOFS);
33693 +       if (*comp) {
33694 +               init_completion(*comp);
33695 +               wkinfo->comp = *comp;
33696 +               return 0;
33697 +       }
33698 +       return -ENOMEM;
33699 +}
33700 +
33701 +static void au_wkq_comp_free(struct completion *comp)
33702 +{
33703 +       au_delayed_kfree(comp);
33704 +}
33705 +
33706 +#else
33707 +
33708 +/* no braces */
33709 +#define AuWkqCompDeclare(name) \
33710 +       DECLARE_COMPLETION_ONSTACK(_ ## name); \
33711 +       struct completion *comp = &_ ## name
33712 +
33713 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
33714 +{
33715 +       wkinfo->comp = *comp;
33716 +       return 0;
33717 +}
33718 +
33719 +static void au_wkq_comp_free(struct completion *comp __maybe_unused)
33720 +{
33721 +       /* empty */
33722 +}
33723 +#endif /* 4KSTACKS */
33724 +
33725 +static void au_wkq_run(struct au_wkinfo *wkinfo)
33726 +{
33727 +       if (au_ftest_wkq(wkinfo->flags, NEST)) {
33728 +               if (au_wkq_test()) {
33729 +                       AuWarn1("wkq from wkq, unless silly-rename on NFS,"
33730 +                               " due to a dead dir by UDBA?\n");
33731 +                       AuDebugOn(au_ftest_wkq(wkinfo->flags, WAIT));
33732 +               }
33733 +       } else
33734 +               au_dbg_verify_kthread();
33735 +
33736 +       if (au_ftest_wkq(wkinfo->flags, WAIT)) {
33737 +               INIT_WORK_ONSTACK(&wkinfo->wk, wkq_func);
33738 +               queue_work(au_wkq, &wkinfo->wk);
33739 +       } else {
33740 +               INIT_WORK(&wkinfo->wk, wkq_func);
33741 +               schedule_work(&wkinfo->wk);
33742 +       }
33743 +}
33744 +
33745 +/*
33746 + * Be careful. It is easy to make deadlock happen.
33747 + * processA: lock, wkq and wait
33748 + * processB: wkq and wait, lock in wkq
33749 + * --> deadlock
33750 + */
33751 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args)
33752 +{
33753 +       int err;
33754 +       AuWkqCompDeclare(comp);
33755 +       struct au_wkinfo wkinfo = {
33756 +               .flags  = flags,
33757 +               .func   = func,
33758 +               .args   = args
33759 +       };
33760 +
33761 +       err = au_wkq_comp_alloc(&wkinfo, &comp);
33762 +       if (!err) {
33763 +               au_wkq_run(&wkinfo);
33764 +               /* no timeout, no interrupt */
33765 +               wait_for_completion(wkinfo.comp);
33766 +               au_wkq_comp_free(comp);
33767 +               destroy_work_on_stack(&wkinfo.wk);
33768 +       }
33769 +
33770 +       return err;
33771 +
33772 +}
33773 +
33774 +/*
33775 + * Note: dget/dput() in func for aufs dentries are not supported. It will be a
33776 + * problem in a concurrent umounting.
33777 + */
33778 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
33779 +                 unsigned int flags)
33780 +{
33781 +       int err;
33782 +       struct au_wkinfo *wkinfo;
33783 +
33784 +       atomic_inc(&au_sbi(sb)->si_nowait.nw_len);
33785 +
33786 +       /*
33787 +        * wkq_func() must free this wkinfo.
33788 +        * it highly depends upon the implementation of workqueue.
33789 +        */
33790 +       err = 0;
33791 +       wkinfo = kmalloc(sizeof(*wkinfo), GFP_NOFS);
33792 +       if (wkinfo) {
33793 +               wkinfo->kobj = &au_sbi(sb)->si_kobj;
33794 +               wkinfo->flags = flags & ~AuWkq_WAIT;
33795 +               wkinfo->func = func;
33796 +               wkinfo->args = args;
33797 +               wkinfo->comp = NULL;
33798 +               kobject_get(wkinfo->kobj);
33799 +               __module_get(THIS_MODULE); /* todo: ?? */
33800 +
33801 +               au_wkq_run(wkinfo);
33802 +       } else {
33803 +               err = -ENOMEM;
33804 +               au_nwt_done(&au_sbi(sb)->si_nowait);
33805 +       }
33806 +
33807 +       return err;
33808 +}
33809 +
33810 +/* ---------------------------------------------------------------------- */
33811 +
33812 +void au_nwt_init(struct au_nowait_tasks *nwt)
33813 +{
33814 +       atomic_set(&nwt->nw_len, 0);
33815 +       /* smp_mb(); */ /* atomic_set */
33816 +       init_waitqueue_head(&nwt->nw_wq);
33817 +}
33818 +
33819 +void au_wkq_fin(void)
33820 +{
33821 +       destroy_workqueue(au_wkq);
33822 +}
33823 +
33824 +int __init au_wkq_init(void)
33825 +{
33826 +       int err;
33827 +
33828 +       err = 0;
33829 +       au_wkq = alloc_workqueue(AUFS_WKQ_NAME, 0, WQ_DFL_ACTIVE);
33830 +       if (IS_ERR(au_wkq))
33831 +               err = PTR_ERR(au_wkq);
33832 +       else if (!au_wkq)
33833 +               err = -ENOMEM;
33834 +
33835 +       return err;
33836 +}
33837 diff -urN /usr/share/empty/fs/aufs/wkq.h linux/fs/aufs/wkq.h
33838 --- /usr/share/empty/fs/aufs/wkq.h      1970-01-01 01:00:00.000000000 +0100
33839 +++ linux/fs/aufs/wkq.h 2016-10-09 16:55:36.496035060 +0200
33840 @@ -0,0 +1,93 @@
33841 +/*
33842 + * Copyright (C) 2005-2016 Junjiro R. Okajima
33843 + *
33844 + * This program, aufs is free software; you can redistribute it and/or modify
33845 + * it under the terms of the GNU General Public License as published by
33846 + * the Free Software Foundation; either version 2 of the License, or
33847 + * (at your option) any later version.
33848 + *
33849 + * This program is distributed in the hope that it will be useful,
33850 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33851 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33852 + * GNU General Public License for more details.
33853 + *
33854 + * You should have received a copy of the GNU General Public License
33855 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33856 + */
33857 +
33858 +/*
33859 + * workqueue for asynchronous/super-io operations
33860 + * todo: try new credentials management scheme
33861 + */
33862 +
33863 +#ifndef __AUFS_WKQ_H__
33864 +#define __AUFS_WKQ_H__
33865 +
33866 +#ifdef __KERNEL__
33867 +
33868 +#include <linux/percpu_counter.h>
33869 +
33870 +struct super_block;
33871 +
33872 +/* ---------------------------------------------------------------------- */
33873 +
33874 +/*
33875 + * in the next operation, wait for the 'nowait' tasks in system-wide workqueue
33876 + */
33877 +struct au_nowait_tasks {
33878 +       atomic_t                nw_len;
33879 +       wait_queue_head_t       nw_wq;
33880 +};
33881 +
33882 +/* ---------------------------------------------------------------------- */
33883 +
33884 +typedef void (*au_wkq_func_t)(void *args);
33885 +
33886 +/* wkq flags */
33887 +#define AuWkq_WAIT     1
33888 +#define AuWkq_NEST     (1 << 1)
33889 +#define au_ftest_wkq(flags, name)      ((flags) & AuWkq_##name)
33890 +#define au_fset_wkq(flags, name) \
33891 +       do { (flags) |= AuWkq_##name; } while (0)
33892 +#define au_fclr_wkq(flags, name) \
33893 +       do { (flags) &= ~AuWkq_##name; } while (0)
33894 +
33895 +#ifndef CONFIG_AUFS_HNOTIFY
33896 +#undef AuWkq_NEST
33897 +#define AuWkq_NEST     0
33898 +#endif
33899 +
33900 +/* wkq.c */
33901 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args);
33902 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
33903 +                 unsigned int flags);
33904 +void au_nwt_init(struct au_nowait_tasks *nwt);
33905 +int __init au_wkq_init(void);
33906 +void au_wkq_fin(void);
33907 +
33908 +/* ---------------------------------------------------------------------- */
33909 +
33910 +static inline int au_wkq_test(void)
33911 +{
33912 +       return current->flags & PF_WQ_WORKER;
33913 +}
33914 +
33915 +static inline int au_wkq_wait(au_wkq_func_t func, void *args)
33916 +{
33917 +       return au_wkq_do_wait(AuWkq_WAIT, func, args);
33918 +}
33919 +
33920 +static inline void au_nwt_done(struct au_nowait_tasks *nwt)
33921 +{
33922 +       if (atomic_dec_and_test(&nwt->nw_len))
33923 +               wake_up_all(&nwt->nw_wq);
33924 +}
33925 +
33926 +static inline int au_nwt_flush(struct au_nowait_tasks *nwt)
33927 +{
33928 +       wait_event(nwt->nw_wq, !atomic_read(&nwt->nw_len));
33929 +       return 0;
33930 +}
33931 +
33932 +#endif /* __KERNEL__ */
33933 +#endif /* __AUFS_WKQ_H__ */
33934 diff -urN /usr/share/empty/fs/aufs/xattr.c linux/fs/aufs/xattr.c
33935 --- /usr/share/empty/fs/aufs/xattr.c    1970-01-01 01:00:00.000000000 +0100
33936 +++ linux/fs/aufs/xattr.c       2016-10-09 16:55:36.496035060 +0200
33937 @@ -0,0 +1,347 @@
33938 +/*
33939 + * Copyright (C) 2014-2016 Junjiro R. Okajima
33940 + *
33941 + * This program, aufs is free software; you can redistribute it and/or modify
33942 + * it under the terms of the GNU General Public License as published by
33943 + * the Free Software Foundation; either version 2 of the License, or
33944 + * (at your option) any later version.
33945 + *
33946 + * This program is distributed in the hope that it will be useful,
33947 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33948 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33949 + * GNU General Public License for more details.
33950 + *
33951 + * You should have received a copy of the GNU General Public License
33952 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33953 + */
33954 +
33955 +/*
33956 + * handling xattr functions
33957 + */
33958 +
33959 +#include <linux/xattr.h>
33960 +#include "aufs.h"
33961 +
33962 +static int au_xattr_ignore(int err, char *name, unsigned int ignore_flags)
33963 +{
33964 +       if (!ignore_flags)
33965 +               goto out;
33966 +       switch (err) {
33967 +       case -ENOMEM:
33968 +       case -EDQUOT:
33969 +               goto out;
33970 +       }
33971 +
33972 +       if ((ignore_flags & AuBrAttr_ICEX) == AuBrAttr_ICEX) {
33973 +               err = 0;
33974 +               goto out;
33975 +       }
33976 +
33977 +#define cmp(brattr, prefix) do {                                       \
33978 +               if (!strncmp(name, XATTR_##prefix##_PREFIX,             \
33979 +                            XATTR_##prefix##_PREFIX_LEN)) {            \
33980 +                       if (ignore_flags & AuBrAttr_ICEX_##brattr)      \
33981 +                               err = 0;                                \
33982 +                       goto out;                                       \
33983 +               }                                                       \
33984 +       } while (0)
33985 +
33986 +       cmp(SEC, SECURITY);
33987 +       cmp(SYS, SYSTEM);
33988 +       cmp(TR, TRUSTED);
33989 +       cmp(USR, USER);
33990 +#undef cmp
33991 +
33992 +       if (ignore_flags & AuBrAttr_ICEX_OTH)
33993 +               err = 0;
33994 +
33995 +out:
33996 +       return err;
33997 +}
33998 +
33999 +static const int au_xattr_out_of_list = AuBrAttr_ICEX_OTH << 1;
34000 +
34001 +static int au_do_cpup_xattr(struct dentry *h_dst, struct dentry *h_src,
34002 +                           char *name, char **buf, unsigned int ignore_flags,
34003 +                           unsigned int verbose)
34004 +{
34005 +       int err;
34006 +       ssize_t ssz;
34007 +       struct inode *h_idst;
34008 +
34009 +       ssz = vfs_getxattr_alloc(h_src, name, buf, 0, GFP_NOFS);
34010 +       err = ssz;
34011 +       if (unlikely(err <= 0)) {
34012 +               if (err == -ENODATA
34013 +                   || (err == -EOPNOTSUPP
34014 +                       && ((ignore_flags & au_xattr_out_of_list)
34015 +                           || (au_test_nfs_noacl(d_inode(h_src))
34016 +                               && (!strcmp(name, XATTR_NAME_POSIX_ACL_ACCESS)
34017 +                                   || !strcmp(name,
34018 +                                              XATTR_NAME_POSIX_ACL_DEFAULT))))
34019 +                           ))
34020 +                       err = 0;
34021 +               if (err && (verbose || au_debug_test()))
34022 +                       pr_err("%s, err %d\n", name, err);
34023 +               goto out;
34024 +       }
34025 +
34026 +       /* unlock it temporary */
34027 +       h_idst = d_inode(h_dst);
34028 +       inode_unlock(h_idst);
34029 +       err = vfsub_setxattr(h_dst, name, *buf, ssz, /*flags*/0);
34030 +       inode_lock_nested(h_idst, AuLsc_I_CHILD2);
34031 +       if (unlikely(err)) {
34032 +               if (verbose || au_debug_test())
34033 +                       pr_err("%s, err %d\n", name, err);
34034 +               err = au_xattr_ignore(err, name, ignore_flags);
34035 +       }
34036 +
34037 +out:
34038 +       return err;
34039 +}
34040 +
34041 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags,
34042 +                 unsigned int verbose)
34043 +{
34044 +       int err, unlocked, acl_access, acl_default;
34045 +       ssize_t ssz;
34046 +       struct inode *h_isrc, *h_idst;
34047 +       char *value, *p, *o, *e;
34048 +
34049 +       /* try stopping to update the source inode while we are referencing */
34050 +       /* there should not be the parent-child relationship between them */
34051 +       h_isrc = d_inode(h_src);
34052 +       h_idst = d_inode(h_dst);
34053 +       inode_unlock(h_idst);
34054 +       inode_lock_nested(h_isrc, AuLsc_I_CHILD);
34055 +       inode_lock_nested(h_idst, AuLsc_I_CHILD2);
34056 +       unlocked = 0;
34057 +
34058 +       /* some filesystems don't list POSIX ACL, for example tmpfs */
34059 +       ssz = vfs_listxattr(h_src, NULL, 0);
34060 +       err = ssz;
34061 +       if (unlikely(err < 0)) {
34062 +               AuTraceErr(err);
34063 +               if (err == -ENODATA
34064 +                   || err == -EOPNOTSUPP)
34065 +                       err = 0;        /* ignore */
34066 +               goto out;
34067 +       }
34068 +
34069 +       err = 0;
34070 +       p = NULL;
34071 +       o = NULL;
34072 +       if (ssz) {
34073 +               err = -ENOMEM;
34074 +               p = kmalloc(ssz, GFP_NOFS);
34075 +               o = p;
34076 +               if (unlikely(!p))
34077 +                       goto out;
34078 +               err = vfs_listxattr(h_src, p, ssz);
34079 +       }
34080 +       inode_unlock(h_isrc);
34081 +       unlocked = 1;
34082 +       AuDbg("err %d, ssz %zd\n", err, ssz);
34083 +       if (unlikely(err < 0))
34084 +               goto out_free;
34085 +
34086 +       err = 0;
34087 +       e = p + ssz;
34088 +       value = NULL;
34089 +       acl_access = 0;
34090 +       acl_default = 0;
34091 +       while (!err && p < e) {
34092 +               acl_access |= !strncmp(p, XATTR_NAME_POSIX_ACL_ACCESS,
34093 +                                      sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1);
34094 +               acl_default |= !strncmp(p, XATTR_NAME_POSIX_ACL_DEFAULT,
34095 +                                       sizeof(XATTR_NAME_POSIX_ACL_DEFAULT)
34096 +                                       - 1);
34097 +               err = au_do_cpup_xattr(h_dst, h_src, p, &value, ignore_flags,
34098 +                                      verbose);
34099 +               p += strlen(p) + 1;
34100 +       }
34101 +       AuTraceErr(err);
34102 +       ignore_flags |= au_xattr_out_of_list;
34103 +       if (!err && !acl_access) {
34104 +               err = au_do_cpup_xattr(h_dst, h_src,
34105 +                                      XATTR_NAME_POSIX_ACL_ACCESS, &value,
34106 +                                      ignore_flags, verbose);
34107 +               AuTraceErr(err);
34108 +       }
34109 +       if (!err && !acl_default) {
34110 +               err = au_do_cpup_xattr(h_dst, h_src,
34111 +                                      XATTR_NAME_POSIX_ACL_DEFAULT, &value,
34112 +                                      ignore_flags, verbose);
34113 +               AuTraceErr(err);
34114 +       }
34115 +
34116 +       if (value)
34117 +               au_delayed_kfree(value);
34118 +
34119 +out_free:
34120 +       if (o)
34121 +               au_delayed_kfree(o);
34122 +out:
34123 +       if (!unlocked)
34124 +               inode_unlock(h_isrc);
34125 +       AuTraceErr(err);
34126 +       return err;
34127 +}
34128 +
34129 +/* ---------------------------------------------------------------------- */
34130 +
34131 +enum {
34132 +       AU_XATTR_LIST,
34133 +       AU_XATTR_GET
34134 +};
34135 +
34136 +struct au_lgxattr {
34137 +       int type;
34138 +       union {
34139 +               struct {
34140 +                       char    *list;
34141 +                       size_t  size;
34142 +               } list;
34143 +               struct {
34144 +                       const char      *name;
34145 +                       void            *value;
34146 +                       size_t          size;
34147 +               } get;
34148 +       } u;
34149 +};
34150 +
34151 +static ssize_t au_lgxattr(struct dentry *dentry, struct au_lgxattr *arg)
34152 +{
34153 +       ssize_t err;
34154 +       struct path h_path;
34155 +       struct super_block *sb;
34156 +
34157 +       sb = dentry->d_sb;
34158 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
34159 +       if (unlikely(err))
34160 +               goto out;
34161 +       err = au_h_path_getattr(dentry, /*force*/1, &h_path);
34162 +       if (unlikely(err))
34163 +               goto out_si;
34164 +       if (unlikely(!h_path.dentry))
34165 +               /* illegally overlapped or something */
34166 +               goto out_di; /* pretending success */
34167 +
34168 +       /* always topmost entry only */
34169 +       switch (arg->type) {
34170 +       case AU_XATTR_LIST:
34171 +               err = vfs_listxattr(h_path.dentry,
34172 +                                   arg->u.list.list, arg->u.list.size);
34173 +               break;
34174 +       case AU_XATTR_GET:
34175 +               AuDebugOn(d_is_negative(h_path.dentry));
34176 +               err = vfs_getxattr(h_path.dentry,
34177 +                                  arg->u.get.name, arg->u.get.value,
34178 +                                  arg->u.get.size);
34179 +               break;
34180 +       }
34181 +
34182 +out_di:
34183 +       di_read_unlock(dentry, AuLock_IR);
34184 +out_si:
34185 +       si_read_unlock(sb);
34186 +out:
34187 +       AuTraceErr(err);
34188 +       return err;
34189 +}
34190 +
34191 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size)
34192 +{
34193 +       struct au_lgxattr arg = {
34194 +               .type = AU_XATTR_LIST,
34195 +               .u.list = {
34196 +                       .list   = list,
34197 +                       .size   = size
34198 +               },
34199 +       };
34200 +
34201 +       return au_lgxattr(dentry, &arg);
34202 +}
34203 +
34204 +ssize_t aufs_getxattr(struct dentry *dentry, struct inode *inode __maybe_unused,
34205 +                     const char *name, void *value, size_t size)
34206 +{
34207 +       struct au_lgxattr arg = {
34208 +               .type = AU_XATTR_GET,
34209 +               .u.get = {
34210 +                       .name   = name,
34211 +                       .value  = value,
34212 +                       .size   = size
34213 +               },
34214 +       };
34215 +
34216 +       return au_lgxattr(dentry, &arg);
34217 +}
34218 +
34219 +int aufs_setxattr(struct dentry *dentry, struct inode *inode, const char *name,
34220 +                 const void *value, size_t size, int flags)
34221 +{
34222 +       struct au_srxattr arg = {
34223 +               .type = AU_XATTR_SET,
34224 +               .u.set = {
34225 +                       .name   = name,
34226 +                       .value  = value,
34227 +                       .size   = size,
34228 +                       .flags  = flags
34229 +               },
34230 +       };
34231 +
34232 +       return au_srxattr(dentry, inode, &arg);
34233 +}
34234 +
34235 +int aufs_removexattr(struct dentry *dentry, const char *name)
34236 +{
34237 +       struct au_srxattr arg = {
34238 +               .type = AU_XATTR_REMOVE,
34239 +               .u.remove = {
34240 +                       .name   = name
34241 +               },
34242 +       };
34243 +
34244 +       return au_srxattr(dentry, d_inode(dentry), &arg);
34245 +}
34246 +
34247 +/* ---------------------------------------------------------------------- */
34248 +
34249 +#if 0
34250 +static size_t au_xattr_list(struct dentry *dentry, char *list, size_t list_size,
34251 +                           const char *name, size_t name_len, int type)
34252 +{
34253 +       return aufs_listxattr(dentry, list, list_size);
34254 +}
34255 +
34256 +static int au_xattr_get(struct dentry *dentry, const char *name, void *buffer,
34257 +                       size_t size, int type)
34258 +{
34259 +       return aufs_getxattr(dentry, name, buffer, size);
34260 +}
34261 +
34262 +static int au_xattr_set(struct dentry *dentry, const char *name,
34263 +                       const void *value, size_t size, int flags, int type)
34264 +{
34265 +       return aufs_setxattr(dentry, name, value, size, flags);
34266 +}
34267 +
34268 +static const struct xattr_handler au_xattr_handler = {
34269 +       /* no prefix, no flags */
34270 +       .list   = au_xattr_list,
34271 +       .get    = au_xattr_get,
34272 +       .set    = au_xattr_set
34273 +       /* why no remove? */
34274 +};
34275 +
34276 +static const struct xattr_handler *au_xattr_handlers[] = {
34277 +       &au_xattr_handler
34278 +};
34279 +
34280 +void au_xattr_init(struct super_block *sb)
34281 +{
34282 +       /* sb->s_xattr = au_xattr_handlers; */
34283 +}
34284 +#endif
34285 diff -urN /usr/share/empty/fs/aufs/xino.c linux/fs/aufs/xino.c
34286 --- /usr/share/empty/fs/aufs/xino.c     1970-01-01 01:00:00.000000000 +0100
34287 +++ linux/fs/aufs/xino.c        2016-10-09 16:55:36.496035060 +0200
34288 @@ -0,0 +1,1318 @@
34289 +/*
34290 + * Copyright (C) 2005-2016 Junjiro R. Okajima
34291 + *
34292 + * This program, aufs is free software; you can redistribute it and/or modify
34293 + * it under the terms of the GNU General Public License as published by
34294 + * the Free Software Foundation; either version 2 of the License, or
34295 + * (at your option) any later version.
34296 + *
34297 + * This program is distributed in the hope that it will be useful,
34298 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
34299 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
34300 + * GNU General Public License for more details.
34301 + *
34302 + * You should have received a copy of the GNU General Public License
34303 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
34304 + */
34305 +
34306 +/*
34307 + * external inode number translation table and bitmap
34308 + */
34309 +
34310 +#include <linux/seq_file.h>
34311 +#include <linux/statfs.h>
34312 +#include "aufs.h"
34313 +
34314 +/* todo: unnecessary to support mmap_sem since kernel-space? */
34315 +ssize_t xino_fread(vfs_readf_t func, struct file *file, void *kbuf, size_t size,
34316 +                  loff_t *pos)
34317 +{
34318 +       ssize_t err;
34319 +       mm_segment_t oldfs;
34320 +       union {
34321 +               void *k;
34322 +               char __user *u;
34323 +       } buf;
34324 +
34325 +       buf.k = kbuf;
34326 +       oldfs = get_fs();
34327 +       set_fs(KERNEL_DS);
34328 +       do {
34329 +               /* todo: signal_pending? */
34330 +               err = func(file, buf.u, size, pos);
34331 +       } while (err == -EAGAIN || err == -EINTR);
34332 +       set_fs(oldfs);
34333 +
34334 +#if 0 /* reserved for future use */
34335 +       if (err > 0)
34336 +               fsnotify_access(file->f_path.dentry);
34337 +#endif
34338 +
34339 +       return err;
34340 +}
34341 +
34342 +/* ---------------------------------------------------------------------- */
34343 +
34344 +static ssize_t xino_fwrite_wkq(vfs_writef_t func, struct file *file, void *buf,
34345 +                              size_t size, loff_t *pos);
34346 +
34347 +static ssize_t do_xino_fwrite(vfs_writef_t func, struct file *file, void *kbuf,
34348 +                             size_t size, loff_t *pos)
34349 +{
34350 +       ssize_t err;
34351 +       mm_segment_t oldfs;
34352 +       union {
34353 +               void *k;
34354 +               const char __user *u;
34355 +       } buf;
34356 +       int i;
34357 +       const int prevent_endless = 10;
34358 +
34359 +       i = 0;
34360 +       buf.k = kbuf;
34361 +       oldfs = get_fs();
34362 +       set_fs(KERNEL_DS);
34363 +       do {
34364 +               err = func(file, buf.u, size, pos);
34365 +               if (err == -EINTR
34366 +                   && !au_wkq_test()
34367 +                   && fatal_signal_pending(current)) {
34368 +                       set_fs(oldfs);
34369 +                       err = xino_fwrite_wkq(func, file, kbuf, size, pos);
34370 +                       BUG_ON(err == -EINTR);
34371 +                       oldfs = get_fs();
34372 +                       set_fs(KERNEL_DS);
34373 +               }
34374 +       } while (i++ < prevent_endless
34375 +                && (err == -EAGAIN || err == -EINTR));
34376 +       set_fs(oldfs);
34377 +
34378 +#if 0 /* reserved for future use */
34379 +       if (err > 0)
34380 +               fsnotify_modify(file->f_path.dentry);
34381 +#endif
34382 +
34383 +       return err;
34384 +}
34385 +
34386 +struct do_xino_fwrite_args {
34387 +       ssize_t *errp;
34388 +       vfs_writef_t func;
34389 +       struct file *file;
34390 +       void *buf;
34391 +       size_t size;
34392 +       loff_t *pos;
34393 +};
34394 +
34395 +static void call_do_xino_fwrite(void *args)
34396 +{
34397 +       struct do_xino_fwrite_args *a = args;
34398 +       *a->errp = do_xino_fwrite(a->func, a->file, a->buf, a->size, a->pos);
34399 +}
34400 +
34401 +static ssize_t xino_fwrite_wkq(vfs_writef_t func, struct file *file, void *buf,
34402 +                              size_t size, loff_t *pos)
34403 +{
34404 +       ssize_t err;
34405 +       int wkq_err;
34406 +       struct do_xino_fwrite_args args = {
34407 +               .errp   = &err,
34408 +               .func   = func,
34409 +               .file   = file,
34410 +               .buf    = buf,
34411 +               .size   = size,
34412 +               .pos    = pos
34413 +       };
34414 +
34415 +       /*
34416 +        * it breaks RLIMIT_FSIZE and normal user's limit,
34417 +        * users should care about quota and real 'filesystem full.'
34418 +        */
34419 +       wkq_err = au_wkq_wait(call_do_xino_fwrite, &args);
34420 +       if (unlikely(wkq_err))
34421 +               err = wkq_err;
34422 +
34423 +       return err;
34424 +}
34425 +
34426 +ssize_t xino_fwrite(vfs_writef_t func, struct file *file, void *buf,
34427 +                   size_t size, loff_t *pos)
34428 +{
34429 +       ssize_t err;
34430 +
34431 +       if (rlimit(RLIMIT_FSIZE) == RLIM_INFINITY) {
34432 +               lockdep_off();
34433 +               err = do_xino_fwrite(func, file, buf, size, pos);
34434 +               lockdep_on();
34435 +       } else
34436 +               err = xino_fwrite_wkq(func, file, buf, size, pos);
34437 +
34438 +       return err;
34439 +}
34440 +
34441 +/* ---------------------------------------------------------------------- */
34442 +
34443 +/*
34444 + * create a new xinofile at the same place/path as @base_file.
34445 + */
34446 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src)
34447 +{
34448 +       struct file *file;
34449 +       struct dentry *base, *parent;
34450 +       struct inode *dir, *delegated;
34451 +       struct qstr *name;
34452 +       struct path path;
34453 +       int err;
34454 +
34455 +       base = base_file->f_path.dentry;
34456 +       parent = base->d_parent; /* dir inode is locked */
34457 +       dir = d_inode(parent);
34458 +       IMustLock(dir);
34459 +
34460 +       file = ERR_PTR(-EINVAL);
34461 +       name = &base->d_name;
34462 +       path.dentry = vfsub_lookup_one_len(name->name, parent, name->len);
34463 +       if (IS_ERR(path.dentry)) {
34464 +               file = (void *)path.dentry;
34465 +               pr_err("%pd lookup err %ld\n",
34466 +                      base, PTR_ERR(path.dentry));
34467 +               goto out;
34468 +       }
34469 +
34470 +       /* no need to mnt_want_write() since we call dentry_open() later */
34471 +       err = vfs_create(dir, path.dentry, S_IRUGO | S_IWUGO, NULL);
34472 +       if (unlikely(err)) {
34473 +               file = ERR_PTR(err);
34474 +               pr_err("%pd create err %d\n", base, err);
34475 +               goto out_dput;
34476 +       }
34477 +
34478 +       path.mnt = base_file->f_path.mnt;
34479 +       file = vfsub_dentry_open(&path,
34480 +                                O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
34481 +                                /* | __FMODE_NONOTIFY */);
34482 +       if (IS_ERR(file)) {
34483 +               pr_err("%pd open err %ld\n", base, PTR_ERR(file));
34484 +               goto out_dput;
34485 +       }
34486 +
34487 +       delegated = NULL;
34488 +       err = vfsub_unlink(dir, &file->f_path, &delegated, /*force*/0);
34489 +       if (unlikely(err == -EWOULDBLOCK)) {
34490 +               pr_warn("cannot retry for NFSv4 delegation"
34491 +                       " for an internal unlink\n");
34492 +               iput(delegated);
34493 +       }
34494 +       if (unlikely(err)) {
34495 +               pr_err("%pd unlink err %d\n", base, err);
34496 +               goto out_fput;
34497 +       }
34498 +
34499 +       if (copy_src) {
34500 +               /* no one can touch copy_src xino */
34501 +               err = au_copy_file(file, copy_src, vfsub_f_size_read(copy_src));
34502 +               if (unlikely(err)) {
34503 +                       pr_err("%pd copy err %d\n", base, err);
34504 +                       goto out_fput;
34505 +               }
34506 +       }
34507 +       goto out_dput; /* success */
34508 +
34509 +out_fput:
34510 +       fput(file);
34511 +       file = ERR_PTR(err);
34512 +out_dput:
34513 +       dput(path.dentry);
34514 +out:
34515 +       return file;
34516 +}
34517 +
34518 +struct au_xino_lock_dir {
34519 +       struct au_hinode *hdir;
34520 +       struct dentry *parent;
34521 +       struct inode *dir;
34522 +};
34523 +
34524 +static void au_xino_lock_dir(struct super_block *sb, struct file *xino,
34525 +                            struct au_xino_lock_dir *ldir)
34526 +{
34527 +       aufs_bindex_t brid, bindex;
34528 +
34529 +       ldir->hdir = NULL;
34530 +       bindex = -1;
34531 +       brid = au_xino_brid(sb);
34532 +       if (brid >= 0)
34533 +               bindex = au_br_index(sb, brid);
34534 +       if (bindex >= 0) {
34535 +               ldir->hdir = au_hi(d_inode(sb->s_root), bindex);
34536 +               au_hn_inode_lock_nested(ldir->hdir, AuLsc_I_PARENT);
34537 +       } else {
34538 +               ldir->parent = dget_parent(xino->f_path.dentry);
34539 +               ldir->dir = d_inode(ldir->parent);
34540 +               inode_lock_nested(ldir->dir, AuLsc_I_PARENT);
34541 +       }
34542 +}
34543 +
34544 +static void au_xino_unlock_dir(struct au_xino_lock_dir *ldir)
34545 +{
34546 +       if (ldir->hdir)
34547 +               au_hn_inode_unlock(ldir->hdir);
34548 +       else {
34549 +               inode_unlock(ldir->dir);
34550 +               dput(ldir->parent);
34551 +       }
34552 +}
34553 +
34554 +/* ---------------------------------------------------------------------- */
34555 +
34556 +/* trucate xino files asynchronously */
34557 +
34558 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex)
34559 +{
34560 +       int err;
34561 +       unsigned long jiffy;
34562 +       blkcnt_t blocks;
34563 +       aufs_bindex_t bi, bbot;
34564 +       struct kstatfs *st;
34565 +       struct au_branch *br;
34566 +       struct file *new_xino, *file;
34567 +       struct super_block *h_sb;
34568 +       struct au_xino_lock_dir ldir;
34569 +
34570 +       err = -ENOMEM;
34571 +       st = kmalloc(sizeof(*st), GFP_NOFS);
34572 +       if (unlikely(!st))
34573 +               goto out;
34574 +
34575 +       err = -EINVAL;
34576 +       bbot = au_sbbot(sb);
34577 +       if (unlikely(bindex < 0 || bbot < bindex))
34578 +               goto out_st;
34579 +       br = au_sbr(sb, bindex);
34580 +       file = br->br_xino.xi_file;
34581 +       if (!file)
34582 +               goto out_st;
34583 +
34584 +       err = vfs_statfs(&file->f_path, st);
34585 +       if (unlikely(err))
34586 +               AuErr1("statfs err %d, ignored\n", err);
34587 +       jiffy = jiffies;
34588 +       blocks = file_inode(file)->i_blocks;
34589 +       pr_info("begin truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
34590 +               bindex, (u64)blocks, st->f_bfree, st->f_blocks);
34591 +
34592 +       au_xino_lock_dir(sb, file, &ldir);
34593 +       /* mnt_want_write() is unnecessary here */
34594 +       new_xino = au_xino_create2(file, file);
34595 +       au_xino_unlock_dir(&ldir);
34596 +       err = PTR_ERR(new_xino);
34597 +       if (IS_ERR(new_xino)) {
34598 +               pr_err("err %d, ignored\n", err);
34599 +               goto out_st;
34600 +       }
34601 +       err = 0;
34602 +       fput(file);
34603 +       br->br_xino.xi_file = new_xino;
34604 +
34605 +       h_sb = au_br_sb(br);
34606 +       for (bi = 0; bi <= bbot; bi++) {
34607 +               if (unlikely(bi == bindex))
34608 +                       continue;
34609 +               br = au_sbr(sb, bi);
34610 +               if (au_br_sb(br) != h_sb)
34611 +                       continue;
34612 +
34613 +               fput(br->br_xino.xi_file);
34614 +               br->br_xino.xi_file = new_xino;
34615 +               get_file(new_xino);
34616 +       }
34617 +
34618 +       err = vfs_statfs(&new_xino->f_path, st);
34619 +       if (!err) {
34620 +               pr_info("end truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
34621 +                       bindex, (u64)file_inode(new_xino)->i_blocks,
34622 +                       st->f_bfree, st->f_blocks);
34623 +               if (file_inode(new_xino)->i_blocks < blocks)
34624 +                       au_sbi(sb)->si_xino_jiffy = jiffy;
34625 +       } else
34626 +               AuErr1("statfs err %d, ignored\n", err);
34627 +
34628 +out_st:
34629 +       au_delayed_kfree(st);
34630 +out:
34631 +       return err;
34632 +}
34633 +
34634 +struct xino_do_trunc_args {
34635 +       struct super_block *sb;
34636 +       struct au_branch *br;
34637 +};
34638 +
34639 +static void xino_do_trunc(void *_args)
34640 +{
34641 +       struct xino_do_trunc_args *args = _args;
34642 +       struct super_block *sb;
34643 +       struct au_branch *br;
34644 +       struct inode *dir;
34645 +       int err;
34646 +       aufs_bindex_t bindex;
34647 +
34648 +       err = 0;
34649 +       sb = args->sb;
34650 +       dir = d_inode(sb->s_root);
34651 +       br = args->br;
34652 +
34653 +       si_noflush_write_lock(sb);
34654 +       ii_read_lock_parent(dir);
34655 +       bindex = au_br_index(sb, br->br_id);
34656 +       err = au_xino_trunc(sb, bindex);
34657 +       ii_read_unlock(dir);
34658 +       if (unlikely(err))
34659 +               pr_warn("err b%d, (%d)\n", bindex, err);
34660 +       atomic_dec(&br->br_xino_running);
34661 +       au_br_put(br);
34662 +       si_write_unlock(sb);
34663 +       au_nwt_done(&au_sbi(sb)->si_nowait);
34664 +       au_delayed_kfree(args);
34665 +}
34666 +
34667 +static int xino_trunc_test(struct super_block *sb, struct au_branch *br)
34668 +{
34669 +       int err;
34670 +       struct kstatfs st;
34671 +       struct au_sbinfo *sbinfo;
34672 +
34673 +       /* todo: si_xino_expire and the ratio should be customizable */
34674 +       sbinfo = au_sbi(sb);
34675 +       if (time_before(jiffies,
34676 +                       sbinfo->si_xino_jiffy + sbinfo->si_xino_expire))
34677 +               return 0;
34678 +
34679 +       /* truncation border */
34680 +       err = vfs_statfs(&br->br_xino.xi_file->f_path, &st);
34681 +       if (unlikely(err)) {
34682 +               AuErr1("statfs err %d, ignored\n", err);
34683 +               return 0;
34684 +       }
34685 +       if (div64_u64(st.f_bfree * 100, st.f_blocks) >= AUFS_XINO_DEF_TRUNC)
34686 +               return 0;
34687 +
34688 +       return 1;
34689 +}
34690 +
34691 +static void xino_try_trunc(struct super_block *sb, struct au_branch *br)
34692 +{
34693 +       struct xino_do_trunc_args *args;
34694 +       int wkq_err;
34695 +
34696 +       if (!xino_trunc_test(sb, br))
34697 +               return;
34698 +
34699 +       if (atomic_inc_return(&br->br_xino_running) > 1)
34700 +               goto out;
34701 +
34702 +       /* lock and kfree() will be called in trunc_xino() */
34703 +       args = kmalloc(sizeof(*args), GFP_NOFS);
34704 +       if (unlikely(!args)) {
34705 +               AuErr1("no memory\n");
34706 +               goto out;
34707 +       }
34708 +
34709 +       au_br_get(br);
34710 +       args->sb = sb;
34711 +       args->br = br;
34712 +       wkq_err = au_wkq_nowait(xino_do_trunc, args, sb, /*flags*/0);
34713 +       if (!wkq_err)
34714 +               return; /* success */
34715 +
34716 +       pr_err("wkq %d\n", wkq_err);
34717 +       au_br_put(br);
34718 +       au_delayed_kfree(args);
34719 +
34720 +out:
34721 +       atomic_dec(&br->br_xino_running);
34722 +}
34723 +
34724 +/* ---------------------------------------------------------------------- */
34725 +
34726 +static int au_xino_do_write(vfs_writef_t write, struct file *file,
34727 +                           ino_t h_ino, ino_t ino)
34728 +{
34729 +       loff_t pos;
34730 +       ssize_t sz;
34731 +
34732 +       pos = h_ino;
34733 +       if (unlikely(au_loff_max / sizeof(ino) - 1 < pos)) {
34734 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
34735 +               return -EFBIG;
34736 +       }
34737 +       pos *= sizeof(ino);
34738 +       sz = xino_fwrite(write, file, &ino, sizeof(ino), &pos);
34739 +       if (sz == sizeof(ino))
34740 +               return 0; /* success */
34741 +
34742 +       AuIOErr("write failed (%zd)\n", sz);
34743 +       return -EIO;
34744 +}
34745 +
34746 +/*
34747 + * write @ino to the xinofile for the specified branch{@sb, @bindex}
34748 + * at the position of @h_ino.
34749 + * even if @ino is zero, it is written to the xinofile and means no entry.
34750 + * if the size of the xino file on a specific filesystem exceeds the watermark,
34751 + * try truncating it.
34752 + */
34753 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
34754 +                 ino_t ino)
34755 +{
34756 +       int err;
34757 +       unsigned int mnt_flags;
34758 +       struct au_branch *br;
34759 +
34760 +       BUILD_BUG_ON(sizeof(long long) != sizeof(au_loff_max)
34761 +                    || ((loff_t)-1) > 0);
34762 +       SiMustAnyLock(sb);
34763 +
34764 +       mnt_flags = au_mntflags(sb);
34765 +       if (!au_opt_test(mnt_flags, XINO))
34766 +               return 0;
34767 +
34768 +       br = au_sbr(sb, bindex);
34769 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
34770 +                              h_ino, ino);
34771 +       if (!err) {
34772 +               if (au_opt_test(mnt_flags, TRUNC_XINO)
34773 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
34774 +                       xino_try_trunc(sb, br);
34775 +               return 0; /* success */
34776 +       }
34777 +
34778 +       AuIOErr("write failed (%d)\n", err);
34779 +       return -EIO;
34780 +}
34781 +
34782 +/* ---------------------------------------------------------------------- */
34783 +
34784 +/* aufs inode number bitmap */
34785 +
34786 +static const int page_bits = (int)PAGE_SIZE * BITS_PER_BYTE;
34787 +static ino_t xib_calc_ino(unsigned long pindex, int bit)
34788 +{
34789 +       ino_t ino;
34790 +
34791 +       AuDebugOn(bit < 0 || page_bits <= bit);
34792 +       ino = AUFS_FIRST_INO + pindex * page_bits + bit;
34793 +       return ino;
34794 +}
34795 +
34796 +static void xib_calc_bit(ino_t ino, unsigned long *pindex, int *bit)
34797 +{
34798 +       AuDebugOn(ino < AUFS_FIRST_INO);
34799 +       ino -= AUFS_FIRST_INO;
34800 +       *pindex = ino / page_bits;
34801 +       *bit = ino % page_bits;
34802 +}
34803 +
34804 +static int xib_pindex(struct super_block *sb, unsigned long pindex)
34805 +{
34806 +       int err;
34807 +       loff_t pos;
34808 +       ssize_t sz;
34809 +       struct au_sbinfo *sbinfo;
34810 +       struct file *xib;
34811 +       unsigned long *p;
34812 +
34813 +       sbinfo = au_sbi(sb);
34814 +       MtxMustLock(&sbinfo->si_xib_mtx);
34815 +       AuDebugOn(pindex > ULONG_MAX / PAGE_SIZE
34816 +                 || !au_opt_test(sbinfo->si_mntflags, XINO));
34817 +
34818 +       if (pindex == sbinfo->si_xib_last_pindex)
34819 +               return 0;
34820 +
34821 +       xib = sbinfo->si_xib;
34822 +       p = sbinfo->si_xib_buf;
34823 +       pos = sbinfo->si_xib_last_pindex;
34824 +       pos *= PAGE_SIZE;
34825 +       sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
34826 +       if (unlikely(sz != PAGE_SIZE))
34827 +               goto out;
34828 +
34829 +       pos = pindex;
34830 +       pos *= PAGE_SIZE;
34831 +       if (vfsub_f_size_read(xib) >= pos + PAGE_SIZE)
34832 +               sz = xino_fread(sbinfo->si_xread, xib, p, PAGE_SIZE, &pos);
34833 +       else {
34834 +               memset(p, 0, PAGE_SIZE);
34835 +               sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
34836 +       }
34837 +       if (sz == PAGE_SIZE) {
34838 +               sbinfo->si_xib_last_pindex = pindex;
34839 +               return 0; /* success */
34840 +       }
34841 +
34842 +out:
34843 +       AuIOErr1("write failed (%zd)\n", sz);
34844 +       err = sz;
34845 +       if (sz >= 0)
34846 +               err = -EIO;
34847 +       return err;
34848 +}
34849 +
34850 +/* ---------------------------------------------------------------------- */
34851 +
34852 +static void au_xib_clear_bit(struct inode *inode)
34853 +{
34854 +       int err, bit;
34855 +       unsigned long pindex;
34856 +       struct super_block *sb;
34857 +       struct au_sbinfo *sbinfo;
34858 +
34859 +       AuDebugOn(inode->i_nlink);
34860 +
34861 +       sb = inode->i_sb;
34862 +       xib_calc_bit(inode->i_ino, &pindex, &bit);
34863 +       AuDebugOn(page_bits <= bit);
34864 +       sbinfo = au_sbi(sb);
34865 +       mutex_lock(&sbinfo->si_xib_mtx);
34866 +       err = xib_pindex(sb, pindex);
34867 +       if (!err) {
34868 +               clear_bit(bit, sbinfo->si_xib_buf);
34869 +               sbinfo->si_xib_next_bit = bit;
34870 +       }
34871 +       mutex_unlock(&sbinfo->si_xib_mtx);
34872 +}
34873 +
34874 +/* for s_op->delete_inode() */
34875 +void au_xino_delete_inode(struct inode *inode, const int unlinked)
34876 +{
34877 +       int err;
34878 +       unsigned int mnt_flags;
34879 +       aufs_bindex_t bindex, bbot, bi;
34880 +       unsigned char try_trunc;
34881 +       struct au_iinfo *iinfo;
34882 +       struct super_block *sb;
34883 +       struct au_hinode *hi;
34884 +       struct inode *h_inode;
34885 +       struct au_branch *br;
34886 +       vfs_writef_t xwrite;
34887 +
34888 +       AuDebugOn(au_is_bad_inode(inode));
34889 +
34890 +       sb = inode->i_sb;
34891 +       mnt_flags = au_mntflags(sb);
34892 +       if (!au_opt_test(mnt_flags, XINO)
34893 +           || inode->i_ino == AUFS_ROOT_INO)
34894 +               return;
34895 +
34896 +       if (unlinked) {
34897 +               au_xigen_inc(inode);
34898 +               au_xib_clear_bit(inode);
34899 +       }
34900 +
34901 +       iinfo = au_ii(inode);
34902 +       bindex = iinfo->ii_btop;
34903 +       if (bindex < 0)
34904 +               return;
34905 +
34906 +       xwrite = au_sbi(sb)->si_xwrite;
34907 +       try_trunc = !!au_opt_test(mnt_flags, TRUNC_XINO);
34908 +       hi = au_hinode(iinfo, bindex);
34909 +       bbot = iinfo->ii_bbot;
34910 +       for (; bindex <= bbot; bindex++, hi++) {
34911 +               h_inode = hi->hi_inode;
34912 +               if (!h_inode
34913 +                   || (!unlinked && h_inode->i_nlink))
34914 +                       continue;
34915 +
34916 +               /* inode may not be revalidated */
34917 +               bi = au_br_index(sb, hi->hi_id);
34918 +               if (bi < 0)
34919 +                       continue;
34920 +
34921 +               br = au_sbr(sb, bi);
34922 +               err = au_xino_do_write(xwrite, br->br_xino.xi_file,
34923 +                                      h_inode->i_ino, /*ino*/0);
34924 +               if (!err && try_trunc
34925 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
34926 +                       xino_try_trunc(sb, br);
34927 +       }
34928 +}
34929 +
34930 +/* get an unused inode number from bitmap */
34931 +ino_t au_xino_new_ino(struct super_block *sb)
34932 +{
34933 +       ino_t ino;
34934 +       unsigned long *p, pindex, ul, pend;
34935 +       struct au_sbinfo *sbinfo;
34936 +       struct file *file;
34937 +       int free_bit, err;
34938 +
34939 +       if (!au_opt_test(au_mntflags(sb), XINO))
34940 +               return iunique(sb, AUFS_FIRST_INO);
34941 +
34942 +       sbinfo = au_sbi(sb);
34943 +       mutex_lock(&sbinfo->si_xib_mtx);
34944 +       p = sbinfo->si_xib_buf;
34945 +       free_bit = sbinfo->si_xib_next_bit;
34946 +       if (free_bit < page_bits && !test_bit(free_bit, p))
34947 +               goto out; /* success */
34948 +       free_bit = find_first_zero_bit(p, page_bits);
34949 +       if (free_bit < page_bits)
34950 +               goto out; /* success */
34951 +
34952 +       pindex = sbinfo->si_xib_last_pindex;
34953 +       for (ul = pindex - 1; ul < ULONG_MAX; ul--) {
34954 +               err = xib_pindex(sb, ul);
34955 +               if (unlikely(err))
34956 +                       goto out_err;
34957 +               free_bit = find_first_zero_bit(p, page_bits);
34958 +               if (free_bit < page_bits)
34959 +                       goto out; /* success */
34960 +       }
34961 +
34962 +       file = sbinfo->si_xib;
34963 +       pend = vfsub_f_size_read(file) / PAGE_SIZE;
34964 +       for (ul = pindex + 1; ul <= pend; ul++) {
34965 +               err = xib_pindex(sb, ul);
34966 +               if (unlikely(err))
34967 +                       goto out_err;
34968 +               free_bit = find_first_zero_bit(p, page_bits);
34969 +               if (free_bit < page_bits)
34970 +                       goto out; /* success */
34971 +       }
34972 +       BUG();
34973 +
34974 +out:
34975 +       set_bit(free_bit, p);
34976 +       sbinfo->si_xib_next_bit = free_bit + 1;
34977 +       pindex = sbinfo->si_xib_last_pindex;
34978 +       mutex_unlock(&sbinfo->si_xib_mtx);
34979 +       ino = xib_calc_ino(pindex, free_bit);
34980 +       AuDbg("i%lu\n", (unsigned long)ino);
34981 +       return ino;
34982 +out_err:
34983 +       mutex_unlock(&sbinfo->si_xib_mtx);
34984 +       AuDbg("i0\n");
34985 +       return 0;
34986 +}
34987 +
34988 +/*
34989 + * read @ino from xinofile for the specified branch{@sb, @bindex}
34990 + * at the position of @h_ino.
34991 + * if @ino does not exist and @do_new is true, get new one.
34992 + */
34993 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
34994 +                ino_t *ino)
34995 +{
34996 +       int err;
34997 +       ssize_t sz;
34998 +       loff_t pos;
34999 +       struct file *file;
35000 +       struct au_sbinfo *sbinfo;
35001 +
35002 +       *ino = 0;
35003 +       if (!au_opt_test(au_mntflags(sb), XINO))
35004 +               return 0; /* no xino */
35005 +
35006 +       err = 0;
35007 +       sbinfo = au_sbi(sb);
35008 +       pos = h_ino;
35009 +       if (unlikely(au_loff_max / sizeof(*ino) - 1 < pos)) {
35010 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
35011 +               return -EFBIG;
35012 +       }
35013 +       pos *= sizeof(*ino);
35014 +
35015 +       file = au_sbr(sb, bindex)->br_xino.xi_file;
35016 +       if (vfsub_f_size_read(file) < pos + sizeof(*ino))
35017 +               return 0; /* no ino */
35018 +
35019 +       sz = xino_fread(sbinfo->si_xread, file, ino, sizeof(*ino), &pos);
35020 +       if (sz == sizeof(*ino))
35021 +               return 0; /* success */
35022 +
35023 +       err = sz;
35024 +       if (unlikely(sz >= 0)) {
35025 +               err = -EIO;
35026 +               AuIOErr("xino read error (%zd)\n", sz);
35027 +       }
35028 +
35029 +       return err;
35030 +}
35031 +
35032 +/* ---------------------------------------------------------------------- */
35033 +
35034 +/* create and set a new xino file */
35035 +
35036 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent)
35037 +{
35038 +       struct file *file;
35039 +       struct dentry *h_parent, *d;
35040 +       struct inode *h_dir, *inode;
35041 +       int err;
35042 +
35043 +       /*
35044 +        * at mount-time, and the xino file is the default path,
35045 +        * hnotify is disabled so we have no notify events to ignore.
35046 +        * when a user specified the xino, we cannot get au_hdir to be ignored.
35047 +        */
35048 +       file = vfsub_filp_open(fname, O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
35049 +                              /* | __FMODE_NONOTIFY */,
35050 +                              S_IRUGO | S_IWUGO);
35051 +       if (IS_ERR(file)) {
35052 +               if (!silent)
35053 +                       pr_err("open %s(%ld)\n", fname, PTR_ERR(file));
35054 +               return file;
35055 +       }
35056 +
35057 +       /* keep file count */
35058 +       err = 0;
35059 +       inode = file_inode(file);
35060 +       h_parent = dget_parent(file->f_path.dentry);
35061 +       h_dir = d_inode(h_parent);
35062 +       inode_lock_nested(h_dir, AuLsc_I_PARENT);
35063 +       /* mnt_want_write() is unnecessary here */
35064 +       /* no delegation since it is just created */
35065 +       if (inode->i_nlink)
35066 +               err = vfsub_unlink(h_dir, &file->f_path, /*delegated*/NULL,
35067 +                                  /*force*/0);
35068 +       inode_unlock(h_dir);
35069 +       dput(h_parent);
35070 +       if (unlikely(err)) {
35071 +               if (!silent)
35072 +                       pr_err("unlink %s(%d)\n", fname, err);
35073 +               goto out;
35074 +       }
35075 +
35076 +       err = -EINVAL;
35077 +       d = file->f_path.dentry;
35078 +       if (unlikely(sb == d->d_sb)) {
35079 +               if (!silent)
35080 +                       pr_err("%s must be outside\n", fname);
35081 +               goto out;
35082 +       }
35083 +       if (unlikely(au_test_fs_bad_xino(d->d_sb))) {
35084 +               if (!silent)
35085 +                       pr_err("xino doesn't support %s(%s)\n",
35086 +                              fname, au_sbtype(d->d_sb));
35087 +               goto out;
35088 +       }
35089 +       return file; /* success */
35090 +
35091 +out:
35092 +       fput(file);
35093 +       file = ERR_PTR(err);
35094 +       return file;
35095 +}
35096 +
35097 +/*
35098 + * find another branch who is on the same filesystem of the specified
35099 + * branch{@btgt}. search until @bbot.
35100 + */
35101 +static int is_sb_shared(struct super_block *sb, aufs_bindex_t btgt,
35102 +                       aufs_bindex_t bbot)
35103 +{
35104 +       aufs_bindex_t bindex;
35105 +       struct super_block *tgt_sb = au_sbr_sb(sb, btgt);
35106 +
35107 +       for (bindex = 0; bindex < btgt; bindex++)
35108 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
35109 +                       return bindex;
35110 +       for (bindex++; bindex <= bbot; bindex++)
35111 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
35112 +                       return bindex;
35113 +       return -1;
35114 +}
35115 +
35116 +/* ---------------------------------------------------------------------- */
35117 +
35118 +/*
35119 + * initialize the xinofile for the specified branch @br
35120 + * at the place/path where @base_file indicates.
35121 + * test whether another branch is on the same filesystem or not,
35122 + * if @do_test is true.
35123 + */
35124 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t h_ino,
35125 +              struct file *base_file, int do_test)
35126 +{
35127 +       int err;
35128 +       ino_t ino;
35129 +       aufs_bindex_t bbot, bindex;
35130 +       struct au_branch *shared_br, *b;
35131 +       struct file *file;
35132 +       struct super_block *tgt_sb;
35133 +
35134 +       shared_br = NULL;
35135 +       bbot = au_sbbot(sb);
35136 +       if (do_test) {
35137 +               tgt_sb = au_br_sb(br);
35138 +               for (bindex = 0; bindex <= bbot; bindex++) {
35139 +                       b = au_sbr(sb, bindex);
35140 +                       if (tgt_sb == au_br_sb(b)) {
35141 +                               shared_br = b;
35142 +                               break;
35143 +                       }
35144 +               }
35145 +       }
35146 +
35147 +       if (!shared_br || !shared_br->br_xino.xi_file) {
35148 +               struct au_xino_lock_dir ldir;
35149 +
35150 +               au_xino_lock_dir(sb, base_file, &ldir);
35151 +               /* mnt_want_write() is unnecessary here */
35152 +               file = au_xino_create2(base_file, NULL);
35153 +               au_xino_unlock_dir(&ldir);
35154 +               err = PTR_ERR(file);
35155 +               if (IS_ERR(file))
35156 +                       goto out;
35157 +               br->br_xino.xi_file = file;
35158 +       } else {
35159 +               br->br_xino.xi_file = shared_br->br_xino.xi_file;
35160 +               get_file(br->br_xino.xi_file);
35161 +       }
35162 +
35163 +       ino = AUFS_ROOT_INO;
35164 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
35165 +                              h_ino, ino);
35166 +       if (unlikely(err)) {
35167 +               fput(br->br_xino.xi_file);
35168 +               br->br_xino.xi_file = NULL;
35169 +       }
35170 +
35171 +out:
35172 +       return err;
35173 +}
35174 +
35175 +/* ---------------------------------------------------------------------- */
35176 +
35177 +/* trucate a xino bitmap file */
35178 +
35179 +/* todo: slow */
35180 +static int do_xib_restore(struct super_block *sb, struct file *file, void *page)
35181 +{
35182 +       int err, bit;
35183 +       ssize_t sz;
35184 +       unsigned long pindex;
35185 +       loff_t pos, pend;
35186 +       struct au_sbinfo *sbinfo;
35187 +       vfs_readf_t func;
35188 +       ino_t *ino;
35189 +       unsigned long *p;
35190 +
35191 +       err = 0;
35192 +       sbinfo = au_sbi(sb);
35193 +       MtxMustLock(&sbinfo->si_xib_mtx);
35194 +       p = sbinfo->si_xib_buf;
35195 +       func = sbinfo->si_xread;
35196 +       pend = vfsub_f_size_read(file);
35197 +       pos = 0;
35198 +       while (pos < pend) {
35199 +               sz = xino_fread(func, file, page, PAGE_SIZE, &pos);
35200 +               err = sz;
35201 +               if (unlikely(sz <= 0))
35202 +                       goto out;
35203 +
35204 +               err = 0;
35205 +               for (ino = page; sz > 0; ino++, sz -= sizeof(ino)) {
35206 +                       if (unlikely(*ino < AUFS_FIRST_INO))
35207 +                               continue;
35208 +
35209 +                       xib_calc_bit(*ino, &pindex, &bit);
35210 +                       AuDebugOn(page_bits <= bit);
35211 +                       err = xib_pindex(sb, pindex);
35212 +                       if (!err)
35213 +                               set_bit(bit, p);
35214 +                       else
35215 +                               goto out;
35216 +               }
35217 +       }
35218 +
35219 +out:
35220 +       return err;
35221 +}
35222 +
35223 +static int xib_restore(struct super_block *sb)
35224 +{
35225 +       int err;
35226 +       aufs_bindex_t bindex, bbot;
35227 +       void *page;
35228 +
35229 +       err = -ENOMEM;
35230 +       page = (void *)__get_free_page(GFP_NOFS);
35231 +       if (unlikely(!page))
35232 +               goto out;
35233 +
35234 +       err = 0;
35235 +       bbot = au_sbbot(sb);
35236 +       for (bindex = 0; !err && bindex <= bbot; bindex++)
35237 +               if (!bindex || is_sb_shared(sb, bindex, bindex - 1) < 0)
35238 +                       err = do_xib_restore
35239 +                               (sb, au_sbr(sb, bindex)->br_xino.xi_file, page);
35240 +               else
35241 +                       AuDbg("b%d\n", bindex);
35242 +       au_delayed_free_page((unsigned long)page);
35243 +
35244 +out:
35245 +       return err;
35246 +}
35247 +
35248 +int au_xib_trunc(struct super_block *sb)
35249 +{
35250 +       int err;
35251 +       ssize_t sz;
35252 +       loff_t pos;
35253 +       struct au_xino_lock_dir ldir;
35254 +       struct au_sbinfo *sbinfo;
35255 +       unsigned long *p;
35256 +       struct file *file;
35257 +
35258 +       SiMustWriteLock(sb);
35259 +
35260 +       err = 0;
35261 +       sbinfo = au_sbi(sb);
35262 +       if (!au_opt_test(sbinfo->si_mntflags, XINO))
35263 +               goto out;
35264 +
35265 +       file = sbinfo->si_xib;
35266 +       if (vfsub_f_size_read(file) <= PAGE_SIZE)
35267 +               goto out;
35268 +
35269 +       au_xino_lock_dir(sb, file, &ldir);
35270 +       /* mnt_want_write() is unnecessary here */
35271 +       file = au_xino_create2(sbinfo->si_xib, NULL);
35272 +       au_xino_unlock_dir(&ldir);
35273 +       err = PTR_ERR(file);
35274 +       if (IS_ERR(file))
35275 +               goto out;
35276 +       fput(sbinfo->si_xib);
35277 +       sbinfo->si_xib = file;
35278 +
35279 +       p = sbinfo->si_xib_buf;
35280 +       memset(p, 0, PAGE_SIZE);
35281 +       pos = 0;
35282 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xib, p, PAGE_SIZE, &pos);
35283 +       if (unlikely(sz != PAGE_SIZE)) {
35284 +               err = sz;
35285 +               AuIOErr("err %d\n", err);
35286 +               if (sz >= 0)
35287 +                       err = -EIO;
35288 +               goto out;
35289 +       }
35290 +
35291 +       mutex_lock(&sbinfo->si_xib_mtx);
35292 +       /* mnt_want_write() is unnecessary here */
35293 +       err = xib_restore(sb);
35294 +       mutex_unlock(&sbinfo->si_xib_mtx);
35295 +
35296 +out:
35297 +       return err;
35298 +}
35299 +
35300 +/* ---------------------------------------------------------------------- */
35301 +
35302 +/*
35303 + * xino mount option handlers
35304 + */
35305 +
35306 +/* xino bitmap */
35307 +static void xino_clear_xib(struct super_block *sb)
35308 +{
35309 +       struct au_sbinfo *sbinfo;
35310 +
35311 +       SiMustWriteLock(sb);
35312 +
35313 +       sbinfo = au_sbi(sb);
35314 +       sbinfo->si_xread = NULL;
35315 +       sbinfo->si_xwrite = NULL;
35316 +       if (sbinfo->si_xib)
35317 +               fput(sbinfo->si_xib);
35318 +       sbinfo->si_xib = NULL;
35319 +       if (sbinfo->si_xib_buf)
35320 +               au_delayed_free_page((unsigned long)sbinfo->si_xib_buf);
35321 +       sbinfo->si_xib_buf = NULL;
35322 +}
35323 +
35324 +static int au_xino_set_xib(struct super_block *sb, struct file *base)
35325 +{
35326 +       int err;
35327 +       loff_t pos;
35328 +       struct au_sbinfo *sbinfo;
35329 +       struct file *file;
35330 +
35331 +       SiMustWriteLock(sb);
35332 +
35333 +       sbinfo = au_sbi(sb);
35334 +       file = au_xino_create2(base, sbinfo->si_xib);
35335 +       err = PTR_ERR(file);
35336 +       if (IS_ERR(file))
35337 +               goto out;
35338 +       if (sbinfo->si_xib)
35339 +               fput(sbinfo->si_xib);
35340 +       sbinfo->si_xib = file;
35341 +       sbinfo->si_xread = vfs_readf(file);
35342 +       sbinfo->si_xwrite = vfs_writef(file);
35343 +
35344 +       err = -ENOMEM;
35345 +       if (!sbinfo->si_xib_buf)
35346 +               sbinfo->si_xib_buf = (void *)get_zeroed_page(GFP_NOFS);
35347 +       if (unlikely(!sbinfo->si_xib_buf))
35348 +               goto out_unset;
35349 +
35350 +       sbinfo->si_xib_last_pindex = 0;
35351 +       sbinfo->si_xib_next_bit = 0;
35352 +       if (vfsub_f_size_read(file) < PAGE_SIZE) {
35353 +               pos = 0;
35354 +               err = xino_fwrite(sbinfo->si_xwrite, file, sbinfo->si_xib_buf,
35355 +                                 PAGE_SIZE, &pos);
35356 +               if (unlikely(err != PAGE_SIZE))
35357 +                       goto out_free;
35358 +       }
35359 +       err = 0;
35360 +       goto out; /* success */
35361 +
35362 +out_free:
35363 +       if (sbinfo->si_xib_buf)
35364 +               au_delayed_free_page((unsigned long)sbinfo->si_xib_buf);
35365 +       sbinfo->si_xib_buf = NULL;
35366 +       if (err >= 0)
35367 +               err = -EIO;
35368 +out_unset:
35369 +       fput(sbinfo->si_xib);
35370 +       sbinfo->si_xib = NULL;
35371 +       sbinfo->si_xread = NULL;
35372 +       sbinfo->si_xwrite = NULL;
35373 +out:
35374 +       return err;
35375 +}
35376 +
35377 +/* xino for each branch */
35378 +static void xino_clear_br(struct super_block *sb)
35379 +{
35380 +       aufs_bindex_t bindex, bbot;
35381 +       struct au_branch *br;
35382 +
35383 +       bbot = au_sbbot(sb);
35384 +       for (bindex = 0; bindex <= bbot; bindex++) {
35385 +               br = au_sbr(sb, bindex);
35386 +               if (!br || !br->br_xino.xi_file)
35387 +                       continue;
35388 +
35389 +               fput(br->br_xino.xi_file);
35390 +               br->br_xino.xi_file = NULL;
35391 +       }
35392 +}
35393 +
35394 +static int au_xino_set_br(struct super_block *sb, struct file *base)
35395 +{
35396 +       int err;
35397 +       ino_t ino;
35398 +       aufs_bindex_t bindex, bbot, bshared;
35399 +       struct {
35400 +               struct file *old, *new;
35401 +       } *fpair, *p;
35402 +       struct au_branch *br;
35403 +       struct inode *inode;
35404 +       vfs_writef_t writef;
35405 +
35406 +       SiMustWriteLock(sb);
35407 +
35408 +       err = -ENOMEM;
35409 +       bbot = au_sbbot(sb);
35410 +       fpair = kcalloc(bbot + 1, sizeof(*fpair), GFP_NOFS);
35411 +       if (unlikely(!fpair))
35412 +               goto out;
35413 +
35414 +       inode = d_inode(sb->s_root);
35415 +       ino = AUFS_ROOT_INO;
35416 +       writef = au_sbi(sb)->si_xwrite;
35417 +       for (bindex = 0, p = fpair; bindex <= bbot; bindex++, p++) {
35418 +               bshared = is_sb_shared(sb, bindex, bindex - 1);
35419 +               if (bshared >= 0) {
35420 +                       /* shared xino */
35421 +                       *p = fpair[bshared];
35422 +                       get_file(p->new);
35423 +               }
35424 +
35425 +               if (!p->new) {
35426 +                       /* new xino */
35427 +                       br = au_sbr(sb, bindex);
35428 +                       p->old = br->br_xino.xi_file;
35429 +                       p->new = au_xino_create2(base, br->br_xino.xi_file);
35430 +                       err = PTR_ERR(p->new);
35431 +                       if (IS_ERR(p->new)) {
35432 +                               p->new = NULL;
35433 +                               goto out_pair;
35434 +                       }
35435 +               }
35436 +
35437 +               err = au_xino_do_write(writef, p->new,
35438 +                                      au_h_iptr(inode, bindex)->i_ino, ino);
35439 +               if (unlikely(err))
35440 +                       goto out_pair;
35441 +       }
35442 +
35443 +       for (bindex = 0, p = fpair; bindex <= bbot; bindex++, p++) {
35444 +               br = au_sbr(sb, bindex);
35445 +               if (br->br_xino.xi_file)
35446 +                       fput(br->br_xino.xi_file);
35447 +               get_file(p->new);
35448 +               br->br_xino.xi_file = p->new;
35449 +       }
35450 +
35451 +out_pair:
35452 +       for (bindex = 0, p = fpair; bindex <= bbot; bindex++, p++)
35453 +               if (p->new)
35454 +                       fput(p->new);
35455 +               else
35456 +                       break;
35457 +       au_delayed_kfree(fpair);
35458 +out:
35459 +       return err;
35460 +}
35461 +
35462 +void au_xino_clr(struct super_block *sb)
35463 +{
35464 +       struct au_sbinfo *sbinfo;
35465 +
35466 +       au_xigen_clr(sb);
35467 +       xino_clear_xib(sb);
35468 +       xino_clear_br(sb);
35469 +       sbinfo = au_sbi(sb);
35470 +       /* lvalue, do not call au_mntflags() */
35471 +       au_opt_clr(sbinfo->si_mntflags, XINO);
35472 +}
35473 +
35474 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount)
35475 +{
35476 +       int err, skip;
35477 +       struct dentry *parent, *cur_parent;
35478 +       struct qstr *dname, *cur_name;
35479 +       struct file *cur_xino;
35480 +       struct inode *dir;
35481 +       struct au_sbinfo *sbinfo;
35482 +
35483 +       SiMustWriteLock(sb);
35484 +
35485 +       err = 0;
35486 +       sbinfo = au_sbi(sb);
35487 +       parent = dget_parent(xino->file->f_path.dentry);
35488 +       if (remount) {
35489 +               skip = 0;
35490 +               dname = &xino->file->f_path.dentry->d_name;
35491 +               cur_xino = sbinfo->si_xib;
35492 +               if (cur_xino) {
35493 +                       cur_parent = dget_parent(cur_xino->f_path.dentry);
35494 +                       cur_name = &cur_xino->f_path.dentry->d_name;
35495 +                       skip = (cur_parent == parent
35496 +                               && au_qstreq(dname, cur_name));
35497 +                       dput(cur_parent);
35498 +               }
35499 +               if (skip)
35500 +                       goto out;
35501 +       }
35502 +
35503 +       au_opt_set(sbinfo->si_mntflags, XINO);
35504 +       dir = d_inode(parent);
35505 +       inode_lock_nested(dir, AuLsc_I_PARENT);
35506 +       /* mnt_want_write() is unnecessary here */
35507 +       err = au_xino_set_xib(sb, xino->file);
35508 +       if (!err)
35509 +               err = au_xigen_set(sb, xino->file);
35510 +       if (!err)
35511 +               err = au_xino_set_br(sb, xino->file);
35512 +       inode_unlock(dir);
35513 +       if (!err)
35514 +               goto out; /* success */
35515 +
35516 +       /* reset all */
35517 +       AuIOErr("failed creating xino(%d).\n", err);
35518 +       au_xigen_clr(sb);
35519 +       xino_clear_xib(sb);
35520 +
35521 +out:
35522 +       dput(parent);
35523 +       return err;
35524 +}
35525 +
35526 +/* ---------------------------------------------------------------------- */
35527 +
35528 +/*
35529 + * create a xinofile at the default place/path.
35530 + */
35531 +struct file *au_xino_def(struct super_block *sb)
35532 +{
35533 +       struct file *file;
35534 +       char *page, *p;
35535 +       struct au_branch *br;
35536 +       struct super_block *h_sb;
35537 +       struct path path;
35538 +       aufs_bindex_t bbot, bindex, bwr;
35539 +
35540 +       br = NULL;
35541 +       bbot = au_sbbot(sb);
35542 +       bwr = -1;
35543 +       for (bindex = 0; bindex <= bbot; bindex++) {
35544 +               br = au_sbr(sb, bindex);
35545 +               if (au_br_writable(br->br_perm)
35546 +                   && !au_test_fs_bad_xino(au_br_sb(br))) {
35547 +                       bwr = bindex;
35548 +                       break;
35549 +               }
35550 +       }
35551 +
35552 +       if (bwr >= 0) {
35553 +               file = ERR_PTR(-ENOMEM);
35554 +               page = (void *)__get_free_page(GFP_NOFS);
35555 +               if (unlikely(!page))
35556 +                       goto out;
35557 +               path.mnt = au_br_mnt(br);
35558 +               path.dentry = au_h_dptr(sb->s_root, bwr);
35559 +               p = d_path(&path, page, PATH_MAX - sizeof(AUFS_XINO_FNAME));
35560 +               file = (void *)p;
35561 +               if (!IS_ERR(p)) {
35562 +                       strcat(p, "/" AUFS_XINO_FNAME);
35563 +                       AuDbg("%s\n", p);
35564 +                       file = au_xino_create(sb, p, /*silent*/0);
35565 +                       if (!IS_ERR(file))
35566 +                               au_xino_brid_set(sb, br->br_id);
35567 +               }
35568 +               au_delayed_free_page((unsigned long)page);
35569 +       } else {
35570 +               file = au_xino_create(sb, AUFS_XINO_DEFPATH, /*silent*/0);
35571 +               if (IS_ERR(file))
35572 +                       goto out;
35573 +               h_sb = file->f_path.dentry->d_sb;
35574 +               if (unlikely(au_test_fs_bad_xino(h_sb))) {
35575 +                       pr_err("xino doesn't support %s(%s)\n",
35576 +                              AUFS_XINO_DEFPATH, au_sbtype(h_sb));
35577 +                       fput(file);
35578 +                       file = ERR_PTR(-EINVAL);
35579 +               }
35580 +               if (!IS_ERR(file))
35581 +                       au_xino_brid_set(sb, -1);
35582 +       }
35583 +
35584 +out:
35585 +       return file;
35586 +}
35587 +
35588 +/* ---------------------------------------------------------------------- */
35589 +
35590 +int au_xino_path(struct seq_file *seq, struct file *file)
35591 +{
35592 +       int err;
35593 +
35594 +       err = au_seq_path(seq, &file->f_path);
35595 +       if (unlikely(err))
35596 +               goto out;
35597 +
35598 +#define Deleted "\\040(deleted)"
35599 +       seq->count -= sizeof(Deleted) - 1;
35600 +       AuDebugOn(memcmp(seq->buf + seq->count, Deleted,
35601 +                        sizeof(Deleted) - 1));
35602 +#undef Deleted
35603 +
35604 +out:
35605 +       return err;
35606 +}
35607 diff -urN /usr/share/empty/include/uapi/linux/aufs_type.h linux/include/uapi/linux/aufs_type.h
35608 --- /usr/share/empty/include/uapi/linux/aufs_type.h     1970-01-01 01:00:00.000000000 +0100
35609 +++ linux/include/uapi/linux/aufs_type.h        2016-12-11 21:22:05.903059512 +0100
35610 @@ -0,0 +1,419 @@
35611 +/*
35612 + * Copyright (C) 2005-2016 Junjiro R. Okajima
35613 + *
35614 + * This program, aufs is free software; you can redistribute it and/or modify
35615 + * it under the terms of the GNU General Public License as published by
35616 + * the Free Software Foundation; either version 2 of the License, or
35617 + * (at your option) any later version.
35618 + *
35619 + * This program is distributed in the hope that it will be useful,
35620 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
35621 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
35622 + * GNU General Public License for more details.
35623 + *
35624 + * You should have received a copy of the GNU General Public License
35625 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
35626 + */
35627 +
35628 +#ifndef __AUFS_TYPE_H__
35629 +#define __AUFS_TYPE_H__
35630 +
35631 +#define AUFS_NAME      "aufs"
35632 +
35633 +#ifdef __KERNEL__
35634 +/*
35635 + * define it before including all other headers.
35636 + * sched.h may use pr_* macros before defining "current", so define the
35637 + * no-current version first, and re-define later.
35638 + */
35639 +#define pr_fmt(fmt)    AUFS_NAME " %s:%d: " fmt, __func__, __LINE__
35640 +#include <linux/sched.h>
35641 +#undef pr_fmt
35642 +#define pr_fmt(fmt) \
35643 +               AUFS_NAME " %s:%d:%.*s[%d]: " fmt, __func__, __LINE__, \
35644 +               (int)sizeof(current->comm), current->comm, current->pid
35645 +#else
35646 +#include <stdint.h>
35647 +#include <sys/types.h>
35648 +#endif /* __KERNEL__ */
35649 +
35650 +#include <linux/limits.h>
35651 +
35652 +#define AUFS_VERSION   "4.x-rcN-20161010"
35653 +
35654 +/* todo? move this to linux-2.6.19/include/magic.h */
35655 +#define AUFS_SUPER_MAGIC       ('a' << 24 | 'u' << 16 | 'f' << 8 | 's')
35656 +
35657 +/* ---------------------------------------------------------------------- */
35658 +
35659 +#ifdef CONFIG_AUFS_BRANCH_MAX_127
35660 +typedef int8_t aufs_bindex_t;
35661 +#define AUFS_BRANCH_MAX 127
35662 +#else
35663 +typedef int16_t aufs_bindex_t;
35664 +#ifdef CONFIG_AUFS_BRANCH_MAX_511
35665 +#define AUFS_BRANCH_MAX 511
35666 +#elif defined(CONFIG_AUFS_BRANCH_MAX_1023)
35667 +#define AUFS_BRANCH_MAX 1023
35668 +#elif defined(CONFIG_AUFS_BRANCH_MAX_32767)
35669 +#define AUFS_BRANCH_MAX 32767
35670 +#endif
35671 +#endif
35672 +
35673 +#ifdef __KERNEL__
35674 +#ifndef AUFS_BRANCH_MAX
35675 +#error unknown CONFIG_AUFS_BRANCH_MAX value
35676 +#endif
35677 +#endif /* __KERNEL__ */
35678 +
35679 +/* ---------------------------------------------------------------------- */
35680 +
35681 +#define AUFS_FSTYPE            AUFS_NAME
35682 +
35683 +#define AUFS_ROOT_INO          2
35684 +#define AUFS_FIRST_INO         11
35685 +
35686 +#define AUFS_WH_PFX            ".wh."
35687 +#define AUFS_WH_PFX_LEN                ((int)sizeof(AUFS_WH_PFX) - 1)
35688 +#define AUFS_WH_TMP_LEN                4
35689 +/* a limit for rmdir/rename a dir and copyup */
35690 +#define AUFS_MAX_NAMELEN       (NAME_MAX \
35691 +                               - AUFS_WH_PFX_LEN * 2   /* doubly whiteouted */\
35692 +                               - 1                     /* dot */\
35693 +                               - AUFS_WH_TMP_LEN)      /* hex */
35694 +#define AUFS_XINO_FNAME                "." AUFS_NAME ".xino"
35695 +#define AUFS_XINO_DEFPATH      "/tmp/" AUFS_XINO_FNAME
35696 +#define AUFS_XINO_DEF_SEC      30 /* seconds */
35697 +#define AUFS_XINO_DEF_TRUNC    45 /* percentage */
35698 +#define AUFS_DIRWH_DEF         3
35699 +#define AUFS_RDCACHE_DEF       10 /* seconds */
35700 +#define AUFS_RDCACHE_MAX       3600 /* seconds */
35701 +#define AUFS_RDBLK_DEF         512 /* bytes */
35702 +#define AUFS_RDHASH_DEF                32
35703 +#define AUFS_WKQ_NAME          AUFS_NAME "d"
35704 +#define AUFS_MFS_DEF_SEC       30 /* seconds */
35705 +#define AUFS_MFS_MAX_SEC       3600 /* seconds */
35706 +#define AUFS_FHSM_CACHE_DEF_SEC        30 /* seconds */
35707 +#define AUFS_PLINK_WARN                50 /* number of plinks in a single bucket */
35708 +
35709 +/* pseudo-link maintenace under /proc */
35710 +#define AUFS_PLINK_MAINT_NAME  "plink_maint"
35711 +#define AUFS_PLINK_MAINT_DIR   "fs/" AUFS_NAME
35712 +#define AUFS_PLINK_MAINT_PATH  AUFS_PLINK_MAINT_DIR "/" AUFS_PLINK_MAINT_NAME
35713 +
35714 +#define AUFS_DIROPQ_NAME       AUFS_WH_PFX ".opq" /* whiteouted doubly */
35715 +#define AUFS_WH_DIROPQ         AUFS_WH_PFX AUFS_DIROPQ_NAME
35716 +
35717 +#define AUFS_BASE_NAME         AUFS_WH_PFX AUFS_NAME
35718 +#define AUFS_PLINKDIR_NAME     AUFS_WH_PFX "plnk"
35719 +#define AUFS_ORPHDIR_NAME      AUFS_WH_PFX "orph"
35720 +
35721 +/* doubly whiteouted */
35722 +#define AUFS_WH_BASE           AUFS_WH_PFX AUFS_BASE_NAME
35723 +#define AUFS_WH_PLINKDIR       AUFS_WH_PFX AUFS_PLINKDIR_NAME
35724 +#define AUFS_WH_ORPHDIR                AUFS_WH_PFX AUFS_ORPHDIR_NAME
35725 +
35726 +/* branch permissions and attributes */
35727 +#define AUFS_BRPERM_RW         "rw"
35728 +#define AUFS_BRPERM_RO         "ro"
35729 +#define AUFS_BRPERM_RR         "rr"
35730 +#define AUFS_BRATTR_COO_REG    "coo_reg"
35731 +#define AUFS_BRATTR_COO_ALL    "coo_all"
35732 +#define AUFS_BRATTR_FHSM       "fhsm"
35733 +#define AUFS_BRATTR_UNPIN      "unpin"
35734 +#define AUFS_BRATTR_ICEX       "icex"
35735 +#define AUFS_BRATTR_ICEX_SEC   "icexsec"
35736 +#define AUFS_BRATTR_ICEX_SYS   "icexsys"
35737 +#define AUFS_BRATTR_ICEX_TR    "icextr"
35738 +#define AUFS_BRATTR_ICEX_USR   "icexusr"
35739 +#define AUFS_BRATTR_ICEX_OTH   "icexoth"
35740 +#define AUFS_BRRATTR_WH                "wh"
35741 +#define AUFS_BRWATTR_NLWH      "nolwh"
35742 +#define AUFS_BRWATTR_MOO       "moo"
35743 +
35744 +#define AuBrPerm_RW            1               /* writable, hardlinkable wh */
35745 +#define AuBrPerm_RO            (1 << 1)        /* readonly */
35746 +#define AuBrPerm_RR            (1 << 2)        /* natively readonly */
35747 +#define AuBrPerm_Mask          (AuBrPerm_RW | AuBrPerm_RO | AuBrPerm_RR)
35748 +
35749 +#define AuBrAttr_COO_REG       (1 << 3)        /* copy-up on open */
35750 +#define AuBrAttr_COO_ALL       (1 << 4)
35751 +#define AuBrAttr_COO_Mask      (AuBrAttr_COO_REG | AuBrAttr_COO_ALL)
35752 +
35753 +#define AuBrAttr_FHSM          (1 << 5)        /* file-based hsm */
35754 +#define AuBrAttr_UNPIN         (1 << 6)        /* rename-able top dir of
35755 +                                                  branch. meaningless since
35756 +                                                  linux-3.18-rc1 */
35757 +
35758 +/* ignore error in copying XATTR */
35759 +#define AuBrAttr_ICEX_SEC      (1 << 7)
35760 +#define AuBrAttr_ICEX_SYS      (1 << 8)
35761 +#define AuBrAttr_ICEX_TR       (1 << 9)
35762 +#define AuBrAttr_ICEX_USR      (1 << 10)
35763 +#define AuBrAttr_ICEX_OTH      (1 << 11)
35764 +#define AuBrAttr_ICEX          (AuBrAttr_ICEX_SEC      \
35765 +                                | AuBrAttr_ICEX_SYS    \
35766 +                                | AuBrAttr_ICEX_TR     \
35767 +                                | AuBrAttr_ICEX_USR    \
35768 +                                | AuBrAttr_ICEX_OTH)
35769 +
35770 +#define AuBrRAttr_WH           (1 << 12)       /* whiteout-able */
35771 +#define AuBrRAttr_Mask         AuBrRAttr_WH
35772 +
35773 +#define AuBrWAttr_NoLinkWH     (1 << 13)       /* un-hardlinkable whiteouts */
35774 +#define AuBrWAttr_MOO          (1 << 14)       /* move-up on open */
35775 +#define AuBrWAttr_Mask         (AuBrWAttr_NoLinkWH | AuBrWAttr_MOO)
35776 +
35777 +#define AuBrAttr_CMOO_Mask     (AuBrAttr_COO_Mask | AuBrWAttr_MOO)
35778 +
35779 +/* #warning test userspace */
35780 +#ifdef __KERNEL__
35781 +#ifndef CONFIG_AUFS_FHSM
35782 +#undef AuBrAttr_FHSM
35783 +#define AuBrAttr_FHSM          0
35784 +#endif
35785 +#ifndef CONFIG_AUFS_XATTR
35786 +#undef AuBrAttr_ICEX
35787 +#define AuBrAttr_ICEX          0
35788 +#undef AuBrAttr_ICEX_SEC
35789 +#define AuBrAttr_ICEX_SEC      0
35790 +#undef AuBrAttr_ICEX_SYS
35791 +#define AuBrAttr_ICEX_SYS      0
35792 +#undef AuBrAttr_ICEX_TR
35793 +#define AuBrAttr_ICEX_TR       0
35794 +#undef AuBrAttr_ICEX_USR
35795 +#define AuBrAttr_ICEX_USR      0
35796 +#undef AuBrAttr_ICEX_OTH
35797 +#define AuBrAttr_ICEX_OTH      0
35798 +#endif
35799 +#endif
35800 +
35801 +/* the longest combination */
35802 +/* AUFS_BRATTR_ICEX and AUFS_BRATTR_ICEX_TR don't affect here */
35803 +#define AuBrPermStrSz  sizeof(AUFS_BRPERM_RW                   \
35804 +                              "+" AUFS_BRATTR_COO_REG          \
35805 +                              "+" AUFS_BRATTR_FHSM             \
35806 +                              "+" AUFS_BRATTR_UNPIN            \
35807 +                              "+" AUFS_BRATTR_ICEX_SEC         \
35808 +                              "+" AUFS_BRATTR_ICEX_SYS         \
35809 +                              "+" AUFS_BRATTR_ICEX_USR         \
35810 +                              "+" AUFS_BRATTR_ICEX_OTH         \
35811 +                              "+" AUFS_BRWATTR_NLWH)
35812 +
35813 +typedef struct {
35814 +       char a[AuBrPermStrSz];
35815 +} au_br_perm_str_t;
35816 +
35817 +static inline int au_br_writable(int brperm)
35818 +{
35819 +       return brperm & AuBrPerm_RW;
35820 +}
35821 +
35822 +static inline int au_br_whable(int brperm)
35823 +{
35824 +       return brperm & (AuBrPerm_RW | AuBrRAttr_WH);
35825 +}
35826 +
35827 +static inline int au_br_wh_linkable(int brperm)
35828 +{
35829 +       return !(brperm & AuBrWAttr_NoLinkWH);
35830 +}
35831 +
35832 +static inline int au_br_cmoo(int brperm)
35833 +{
35834 +       return brperm & AuBrAttr_CMOO_Mask;
35835 +}
35836 +
35837 +static inline int au_br_fhsm(int brperm)
35838 +{
35839 +       return brperm & AuBrAttr_FHSM;
35840 +}
35841 +
35842 +/* ---------------------------------------------------------------------- */
35843 +
35844 +/* ioctl */
35845 +enum {
35846 +       /* readdir in userspace */
35847 +       AuCtl_RDU,
35848 +       AuCtl_RDU_INO,
35849 +
35850 +       AuCtl_WBR_FD,   /* pathconf wrapper */
35851 +       AuCtl_IBUSY,    /* busy inode */
35852 +       AuCtl_MVDOWN,   /* move-down */
35853 +       AuCtl_BR,       /* info about branches */
35854 +       AuCtl_FHSM_FD   /* connection for fhsm */
35855 +};
35856 +
35857 +/* borrowed from linux/include/linux/kernel.h */
35858 +#ifndef ALIGN
35859 +#define ALIGN(x, a)            __ALIGN_MASK(x, (typeof(x))(a)-1)
35860 +#define __ALIGN_MASK(x, mask)  (((x)+(mask))&~(mask))
35861 +#endif
35862 +
35863 +/* borrowed from linux/include/linux/compiler-gcc3.h */
35864 +#ifndef __aligned
35865 +#define __aligned(x)                   __attribute__((aligned(x)))
35866 +#endif
35867 +
35868 +#ifdef __KERNEL__
35869 +#ifndef __packed
35870 +#define __packed                       __attribute__((packed))
35871 +#endif
35872 +#endif
35873 +
35874 +struct au_rdu_cookie {
35875 +       uint64_t        h_pos;
35876 +       int16_t         bindex;
35877 +       uint8_t         flags;
35878 +       uint8_t         pad;
35879 +       uint32_t        generation;
35880 +} __aligned(8);
35881 +
35882 +struct au_rdu_ent {
35883 +       uint64_t        ino;
35884 +       int16_t         bindex;
35885 +       uint8_t         type;
35886 +       uint8_t         nlen;
35887 +       uint8_t         wh;
35888 +       char            name[0];
35889 +} __aligned(8);
35890 +
35891 +static inline int au_rdu_len(int nlen)
35892 +{
35893 +       /* include the terminating NULL */
35894 +       return ALIGN(sizeof(struct au_rdu_ent) + nlen + 1,
35895 +                    sizeof(uint64_t));
35896 +}
35897 +
35898 +union au_rdu_ent_ul {
35899 +       struct au_rdu_ent __user        *e;
35900 +       uint64_t                        ul;
35901 +};
35902 +
35903 +enum {
35904 +       AufsCtlRduV_SZ,
35905 +       AufsCtlRduV_End
35906 +};
35907 +
35908 +struct aufs_rdu {
35909 +       /* input */
35910 +       union {
35911 +               uint64_t        sz;     /* AuCtl_RDU */
35912 +               uint64_t        nent;   /* AuCtl_RDU_INO */
35913 +       };
35914 +       union au_rdu_ent_ul     ent;
35915 +       uint16_t                verify[AufsCtlRduV_End];
35916 +
35917 +       /* input/output */
35918 +       uint32_t                blk;
35919 +
35920 +       /* output */
35921 +       union au_rdu_ent_ul     tail;
35922 +       /* number of entries which were added in a single call */
35923 +       uint64_t                rent;
35924 +       uint8_t                 full;
35925 +       uint8_t                 shwh;
35926 +
35927 +       struct au_rdu_cookie    cookie;
35928 +} __aligned(8);
35929 +
35930 +/* ---------------------------------------------------------------------- */
35931 +
35932 +struct aufs_wbr_fd {
35933 +       uint32_t        oflags;
35934 +       int16_t         brid;
35935 +} __aligned(8);
35936 +
35937 +/* ---------------------------------------------------------------------- */
35938 +
35939 +struct aufs_ibusy {
35940 +       uint64_t        ino, h_ino;
35941 +       int16_t         bindex;
35942 +} __aligned(8);
35943 +
35944 +/* ---------------------------------------------------------------------- */
35945 +
35946 +/* error code for move-down */
35947 +/* the actual message strings are implemented in aufs-util.git */
35948 +enum {
35949 +       EAU_MVDOWN_OPAQUE = 1,
35950 +       EAU_MVDOWN_WHITEOUT,
35951 +       EAU_MVDOWN_UPPER,
35952 +       EAU_MVDOWN_BOTTOM,
35953 +       EAU_MVDOWN_NOUPPER,
35954 +       EAU_MVDOWN_NOLOWERBR,
35955 +       EAU_Last
35956 +};
35957 +
35958 +/* flags for move-down */
35959 +#define AUFS_MVDOWN_DMSG       1
35960 +#define AUFS_MVDOWN_OWLOWER    (1 << 1)        /* overwrite lower */
35961 +#define AUFS_MVDOWN_KUPPER     (1 << 2)        /* keep upper */
35962 +#define AUFS_MVDOWN_ROLOWER    (1 << 3)        /* do even if lower is RO */
35963 +#define AUFS_MVDOWN_ROLOWER_R  (1 << 4)        /* did on lower RO */
35964 +#define AUFS_MVDOWN_ROUPPER    (1 << 5)        /* do even if upper is RO */
35965 +#define AUFS_MVDOWN_ROUPPER_R  (1 << 6)        /* did on upper RO */
35966 +#define AUFS_MVDOWN_BRID_UPPER (1 << 7)        /* upper brid */
35967 +#define AUFS_MVDOWN_BRID_LOWER (1 << 8)        /* lower brid */
35968 +#define AUFS_MVDOWN_FHSM_LOWER (1 << 9)        /* find fhsm attr for lower */
35969 +#define AUFS_MVDOWN_STFS       (1 << 10)       /* req. stfs */
35970 +#define AUFS_MVDOWN_STFS_FAILED        (1 << 11)       /* output: stfs is unusable */
35971 +#define AUFS_MVDOWN_BOTTOM     (1 << 12)       /* output: no more lowers */
35972 +
35973 +/* index for move-down */
35974 +enum {
35975 +       AUFS_MVDOWN_UPPER,
35976 +       AUFS_MVDOWN_LOWER,
35977 +       AUFS_MVDOWN_NARRAY
35978 +};
35979 +
35980 +/*
35981 + * additional info of move-down
35982 + * number of free blocks and inodes.
35983 + * subset of struct kstatfs, but smaller and always 64bit.
35984 + */
35985 +struct aufs_stfs {
35986 +       uint64_t        f_blocks;
35987 +       uint64_t        f_bavail;
35988 +       uint64_t        f_files;
35989 +       uint64_t        f_ffree;
35990 +};
35991 +
35992 +struct aufs_stbr {
35993 +       int16_t                 brid;   /* optional input */
35994 +       int16_t                 bindex; /* output */
35995 +       struct aufs_stfs        stfs;   /* output when AUFS_MVDOWN_STFS set */
35996 +} __aligned(8);
35997 +
35998 +struct aufs_mvdown {
35999 +       uint32_t                flags;                  /* input/output */
36000 +       struct aufs_stbr        stbr[AUFS_MVDOWN_NARRAY]; /* input/output */
36001 +       int8_t                  au_errno;               /* output */
36002 +} __aligned(8);
36003 +
36004 +/* ---------------------------------------------------------------------- */
36005 +
36006 +union aufs_brinfo {
36007 +       /* PATH_MAX may differ between kernel-space and user-space */
36008 +       char    _spacer[4096];
36009 +       struct {
36010 +               int16_t id;
36011 +               int     perm;
36012 +               char    path[0];
36013 +       };
36014 +} __aligned(8);
36015 +
36016 +/* ---------------------------------------------------------------------- */
36017 +
36018 +#define AuCtlType              'A'
36019 +#define AUFS_CTL_RDU           _IOWR(AuCtlType, AuCtl_RDU, struct aufs_rdu)
36020 +#define AUFS_CTL_RDU_INO       _IOWR(AuCtlType, AuCtl_RDU_INO, struct aufs_rdu)
36021 +#define AUFS_CTL_WBR_FD                _IOW(AuCtlType, AuCtl_WBR_FD, \
36022 +                                    struct aufs_wbr_fd)
36023 +#define AUFS_CTL_IBUSY         _IOWR(AuCtlType, AuCtl_IBUSY, struct aufs_ibusy)
36024 +#define AUFS_CTL_MVDOWN                _IOWR(AuCtlType, AuCtl_MVDOWN, \
36025 +                                     struct aufs_mvdown)
36026 +#define AUFS_CTL_BRINFO                _IOW(AuCtlType, AuCtl_BR, union aufs_brinfo)
36027 +#define AUFS_CTL_FHSM_FD       _IOW(AuCtlType, AuCtl_FHSM_FD, int)
36028 +
36029 +#endif /* __AUFS_TYPE_H__ */
36030 aufs4.x-rcN loopback patch
36031
36032 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
36033 index 005e292..e9e517d 100644
36034 --- a/drivers/block/loop.c
36035 +++ b/drivers/block/loop.c
36036 @@ -551,7 +551,7 @@ static int do_req_filebacked(struct loop_device *lo, struct request *rq)
36037  }
36038  
36039  struct switch_request {
36040 -       struct file *file;
36041 +       struct file *file, *virt_file;
36042         struct completion wait;
36043  };
36044  
36045 @@ -577,6 +577,7 @@ static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
36046         mapping = file->f_mapping;
36047         mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
36048         lo->lo_backing_file = file;
36049 +       lo->lo_backing_virt_file = p->virt_file;
36050         lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
36051                 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
36052         lo->old_gfp_mask = mapping_gfp_mask(mapping);
36053 @@ -589,11 +590,13 @@ static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
36054   * First it needs to flush existing IO, it does this by sending a magic
36055   * BIO down the pipe. The completion of this BIO does the actual switch.
36056   */
36057 -static int loop_switch(struct loop_device *lo, struct file *file)
36058 +static int loop_switch(struct loop_device *lo, struct file *file,
36059 +                      struct file *virt_file)
36060  {
36061         struct switch_request w;
36062  
36063         w.file = file;
36064 +       w.virt_file = virt_file;
36065  
36066         /* freeze queue and wait for completion of scheduled requests */
36067         blk_mq_freeze_queue(lo->lo_queue);
36068 @@ -612,7 +615,16 @@ static int loop_switch(struct loop_device *lo, struct file *file)
36069   */
36070  static int loop_flush(struct loop_device *lo)
36071  {
36072 -       return loop_switch(lo, NULL);
36073 +       return loop_switch(lo, NULL, NULL);
36074 +}
36075 +
36076 +static struct file *loop_real_file(struct file *file)
36077 +{
36078 +       struct file *f = NULL;
36079 +
36080 +       if (file->f_path.dentry->d_sb->s_op->real_loop)
36081 +               f = file->f_path.dentry->d_sb->s_op->real_loop(file);
36082 +       return f;
36083  }
36084  
36085  static void loop_reread_partitions(struct loop_device *lo,
36086 @@ -649,6 +661,7 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
36087                           unsigned int arg)
36088  {
36089         struct file     *file, *old_file;
36090 +       struct file     *f, *virt_file = NULL, *old_virt_file;
36091         struct inode    *inode;
36092         int             error;
36093  
36094 @@ -665,9 +678,16 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
36095         file = fget(arg);
36096         if (!file)
36097                 goto out;
36098 +       f = loop_real_file(file);
36099 +       if (f) {
36100 +               virt_file = file;
36101 +               file = f;
36102 +               get_file(file);
36103 +       }
36104  
36105         inode = file->f_mapping->host;
36106         old_file = lo->lo_backing_file;
36107 +       old_virt_file = lo->lo_backing_virt_file;
36108  
36109         error = -EINVAL;
36110  
36111 @@ -679,17 +699,21 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
36112                 goto out_putf;
36113  
36114         /* and ... switch */
36115 -       error = loop_switch(lo, file);
36116 +       error = loop_switch(lo, file, virt_file);
36117         if (error)
36118                 goto out_putf;
36119  
36120         fput(old_file);
36121 +       if (old_virt_file)
36122 +               fput(old_virt_file);
36123         if (lo->lo_flags & LO_FLAGS_PARTSCAN)
36124                 loop_reread_partitions(lo, bdev);
36125         return 0;
36126  
36127   out_putf:
36128         fput(file);
36129 +       if (virt_file)
36130 +               fput(virt_file);
36131   out:
36132         return error;
36133  }
36134 @@ -876,7 +900,7 @@ static int loop_prepare_queue(struct loop_device *lo)
36135  static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36136                        struct block_device *bdev, unsigned int arg)
36137  {
36138 -       struct file     *file, *f;
36139 +       struct file     *file, *f, *virt_file = NULL;
36140         struct inode    *inode;
36141         struct address_space *mapping;
36142         unsigned lo_blocksize;
36143 @@ -891,6 +915,12 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36144         file = fget(arg);
36145         if (!file)
36146                 goto out;
36147 +       f = loop_real_file(file);
36148 +       if (f) {
36149 +               virt_file = file;
36150 +               file = f;
36151 +               get_file(file);
36152 +       }
36153  
36154         error = -EBUSY;
36155         if (lo->lo_state != Lo_unbound)
36156 @@ -943,6 +973,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36157         lo->lo_device = bdev;
36158         lo->lo_flags = lo_flags;
36159         lo->lo_backing_file = file;
36160 +       lo->lo_backing_virt_file = virt_file;
36161         lo->transfer = NULL;
36162         lo->ioctl = NULL;
36163         lo->lo_sizelimit = 0;
36164 @@ -975,6 +1006,8 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
36165  
36166   out_putf:
36167         fput(file);
36168 +       if (virt_file)
36169 +               fput(virt_file);
36170   out:
36171         /* This is safe: open() is still holding a reference. */
36172         module_put(THIS_MODULE);
36173 @@ -1021,6 +1054,7 @@ loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
36174  static int loop_clr_fd(struct loop_device *lo)
36175  {
36176         struct file *filp = lo->lo_backing_file;
36177 +       struct file *virt_filp = lo->lo_backing_virt_file;
36178         gfp_t gfp = lo->old_gfp_mask;
36179         struct block_device *bdev = lo->lo_device;
36180  
36181 @@ -1052,6 +1086,7 @@ static int loop_clr_fd(struct loop_device *lo)
36182         spin_lock_irq(&lo->lo_lock);
36183         lo->lo_state = Lo_rundown;
36184         lo->lo_backing_file = NULL;
36185 +       lo->lo_backing_virt_file = NULL;
36186         spin_unlock_irq(&lo->lo_lock);
36187  
36188         loop_release_xfer(lo);
36189 @@ -1096,6 +1131,8 @@ static int loop_clr_fd(struct loop_device *lo)
36190          * bd_mutex which is usually taken before lo_ctl_mutex.
36191          */
36192         fput(filp);
36193 +       if (virt_filp)
36194 +               fput(virt_filp);
36195         return 0;
36196  }
36197  
36198 diff --git a/drivers/block/loop.h b/drivers/block/loop.h
36199 index fb2237c..c3888c5 100644
36200 --- a/drivers/block/loop.h
36201 +++ b/drivers/block/loop.h
36202 @@ -46,7 +46,7 @@ struct loop_device {
36203         int             (*ioctl)(struct loop_device *, int cmd, 
36204                                  unsigned long arg); 
36205  
36206 -       struct file *   lo_backing_file;
36207 +       struct file *   lo_backing_file, *lo_backing_virt_file;
36208         struct block_device *lo_device;
36209         unsigned        lo_blocksize;
36210         void            *key_data; 
36211 diff --git a/fs/aufs/f_op.c b/fs/aufs/f_op.c
36212 index 00475fb..01390e1 100644
36213 --- a/fs/aufs/f_op.c
36214 +++ b/fs/aufs/f_op.c
36215 @@ -348,7 +348,7 @@ static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
36216         if (IS_ERR(h_file))
36217                 goto out;
36218  
36219 -       if (au_test_loopback_kthread()) {
36220 +       if (0 && au_test_loopback_kthread()) {
36221                 au_warn_loopback(h_file->f_path.dentry->d_sb);
36222                 if (file->f_mapping != h_file->f_mapping) {
36223                         file->f_mapping = h_file->f_mapping;
36224 diff --git a/fs/aufs/loop.c b/fs/aufs/loop.c
36225 index c3ca50f..a3dbdaf 100644
36226 --- a/fs/aufs/loop.c
36227 +++ b/fs/aufs/loop.c
36228 @@ -132,3 +132,19 @@ void au_loopback_fin(void)
36229                 symbol_put(loop_backing_file);
36230         au_delayed_kfree(au_warn_loopback_array);
36231  }
36232 +
36233 +/* ---------------------------------------------------------------------- */
36234 +
36235 +/* support the loopback block device insude aufs */
36236 +
36237 +struct file *aufs_real_loop(struct file *file)
36238 +{
36239 +       struct file *f;
36240 +
36241 +       BUG_ON(!au_test_aufs(file->f_path.dentry->d_sb));
36242 +       fi_read_lock(file);
36243 +       f = au_hf_top(file);
36244 +       fi_read_unlock(file);
36245 +       AuDebugOn(!f);
36246 +       return f;
36247 +}
36248 diff --git a/fs/aufs/loop.h b/fs/aufs/loop.h
36249 index 48bf070..66afec7 100644
36250 --- a/fs/aufs/loop.h
36251 +++ b/fs/aufs/loop.h
36252 @@ -25,7 +25,11 @@ void au_warn_loopback(struct super_block *h_sb);
36253  
36254  int au_loopback_init(void);
36255  void au_loopback_fin(void);
36256 +
36257 +struct file *aufs_real_loop(struct file *file);
36258  #else
36259 +AuStub(struct file *, loop_backing_file, return NULL)
36260 +
36261  AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
36262            struct dentry *h_adding)
36263  AuStubInt0(au_test_loopback_kthread, void)
36264 @@ -33,6 +37,8 @@ AuStubVoid(au_warn_loopback, struct super_block *h_sb)
36265  
36266  AuStubInt0(au_loopback_init, void)
36267  AuStubVoid(au_loopback_fin, void)
36268 +
36269 +AuStub(struct file *, aufs_real_loop, return NULL, struct file *file)
36270  #endif /* BLK_DEV_LOOP */
36271  
36272  #endif /* __KERNEL__ */
36273 diff --git a/fs/aufs/super.c b/fs/aufs/super.c
36274 index 58a773c..75f212c 100644
36275 --- a/fs/aufs/super.c
36276 +++ b/fs/aufs/super.c
36277 @@ -831,7 +831,10 @@ static const struct super_operations aufs_sop = {
36278         .statfs         = aufs_statfs,
36279         .put_super      = aufs_put_super,
36280         .sync_fs        = aufs_sync_fs,
36281 -       .remount_fs     = aufs_remount_fs
36282 +       .remount_fs     = aufs_remount_fs,
36283 +#ifdef CONFIG_AUFS_BDEV_LOOP
36284 +       .real_loop      = aufs_real_loop
36285 +#endif
36286  };
36287  
36288  /* ---------------------------------------------------------------------- */
36289 diff --git a/include/linux/fs.h b/include/linux/fs.h
36290 index fc80663..0c4c6fa 100644
36291 --- a/include/linux/fs.h
36292 +++ b/include/linux/fs.h
36293 @@ -1814,6 +1814,10 @@ struct super_operations {
36294                                   struct shrink_control *);
36295         long (*free_cached_objects)(struct super_block *,
36296                                     struct shrink_control *);
36297 +#if defined(CONFIG_BLK_DEV_LOOP) ||  defined(CONFIG_BLK_DEV_LOOP_MODULE)
36298 +       /* and aufs */
36299 +       struct file *(*real_loop)(struct file *);
36300 +#endif
36301  };
36302  
36303  /*
This page took 2.642796 seconds and 4 git commands to generate.