]> git.pld-linux.org Git - packages/kernel.git/blob - kernel-aufs3.patch
- updated configs for 3,19
[packages/kernel.git] / kernel-aufs3.patch
1 aufs3.18.1+ kbuild patch
2
3 diff --git a/fs/Kconfig b/fs/Kconfig
4 index 664991a..1481093 100644
5 --- a/fs/Kconfig
6 +++ b/fs/Kconfig
7 @@ -210,6 +210,7 @@ source "fs/ufs/Kconfig"
8  source "fs/exofs/Kconfig"
9  source "fs/f2fs/Kconfig"
10  source "fs/efivarfs/Kconfig"
11 +source "fs/aufs/Kconfig"
12  
13  endif # MISC_FILESYSTEMS
14  
15 diff --git a/fs/Makefile b/fs/Makefile
16 index da0bbb4..c8bc724 100644
17 --- a/fs/Makefile
18 +++ b/fs/Makefile
19 @@ -126,3 +126,4 @@ 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 8523f9b..11f8f74 100644
26 --- a/include/uapi/linux/Kbuild
27 +++ b/include/uapi/linux/Kbuild
28 @@ -56,6 +56,7 @@ header-y += atmppp.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 aufs3.18.1+ base patch
37
38 diff --git a/MAINTAINERS b/MAINTAINERS
39 index c721042..83801d0 100644
40 --- a/MAINTAINERS
41 +++ b/MAINTAINERS
42 @@ -1795,6 +1795,20 @@ 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://git.code.sf.net/p/aufs/aufs3-linux
52 +T:     git://github.com/sfjro/aufs3-linux.git
53 +S:     Supported
54 +F:     Documentation/filesystems/aufs/
55 +F:     Documentation/ABI/testing/debugfs-aufs
56 +F:     Documentation/ABI/testing/sysfs-aufs
57 +F:     fs/aufs/
58 +F:     include/uapi/linux/aufs_type.h
59 +
60  AUXILIARY DISPLAY DRIVERS
61  M:     Miguel Ojeda Sandonis <miguel.ojeda.sandonis@gmail.com>
62  W:     http://miguelojeda.es/auxdisplay.htm
63 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
64 index 6cb1beb..30efd68 100644
65 --- a/drivers/block/loop.c
66 +++ b/drivers/block/loop.c
67 @@ -692,6 +692,24 @@ static inline int is_loop_device(struct file *file)
68         return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
69  }
70  
71 +/*
72 + * for AUFS
73 + * no get/put for file.
74 + */
75 +struct file *loop_backing_file(struct super_block *sb)
76 +{
77 +       struct file *ret;
78 +       struct loop_device *l;
79 +
80 +       ret = NULL;
81 +       if (MAJOR(sb->s_dev) == LOOP_MAJOR) {
82 +               l = sb->s_bdev->bd_disk->private_data;
83 +               ret = l->lo_backing_file;
84 +       }
85 +       return ret;
86 +}
87 +EXPORT_SYMBOL(loop_backing_file);
88 +
89  /* loop sysfs attributes */
90  
91  static ssize_t loop_attr_show(struct device *dev, char *page,
92 diff --git a/fs/dcache.c b/fs/dcache.c
93 index 71acf8d..da7342e 100644
94 --- a/fs/dcache.c
95 +++ b/fs/dcache.c
96 @@ -1019,7 +1019,7 @@ enum d_walk_ret {
97   *
98   * The @enter() and @finish() callbacks are called with d_lock held.
99   */
100 -static void d_walk(struct dentry *parent, void *data,
101 +void d_walk(struct dentry *parent, void *data,
102                    enum d_walk_ret (*enter)(void *, struct dentry *),
103                    void (*finish)(void *))
104  {
105 diff --git a/fs/inode.c b/fs/inode.c
106 index 26753ba..df21e66 100644
107 --- a/fs/inode.c
108 +++ b/fs/inode.c
109 @@ -1497,7 +1497,7 @@ static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
110   * This does the actual work of updating an inodes time or version.  Must have
111   * had called mnt_want_write() before calling this.
112   */
113 -static int update_time(struct inode *inode, struct timespec *time, int flags)
114 +int update_time(struct inode *inode, struct timespec *time, int flags)
115  {
116         if (inode->i_op->update_time)
117                 return inode->i_op->update_time(inode, time, flags);
118 diff --git a/fs/splice.c b/fs/splice.c
119 index 75c6058..619359a 100644
120 --- a/fs/splice.c
121 +++ b/fs/splice.c
122 @@ -1114,8 +1114,8 @@ EXPORT_SYMBOL(generic_splice_sendpage);
123  /*
124   * Attempt to initiate a splice from pipe to file.
125   */
126 -static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
127 -                          loff_t *ppos, size_t len, unsigned int flags)
128 +long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
129 +                   loff_t *ppos, size_t len, unsigned int flags)
130  {
131         ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
132                                 loff_t *, size_t, unsigned int);
133 @@ -1131,9 +1131,9 @@ static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
134  /*
135   * Attempt to initiate a splice from a file to a pipe.
136   */
137 -static long do_splice_to(struct file *in, loff_t *ppos,
138 -                        struct pipe_inode_info *pipe, size_t len,
139 -                        unsigned int flags)
140 +long do_splice_to(struct file *in, loff_t *ppos,
141 +                 struct pipe_inode_info *pipe, size_t len,
142 +                 unsigned int flags)
143  {
144         ssize_t (*splice_read)(struct file *, loff_t *,
145                                struct pipe_inode_info *, size_t, unsigned int);
146 diff --git a/include/linux/fs.h b/include/linux/fs.h
147 index 9ab779e..aabcbba 100644
148 --- a/include/linux/fs.h
149 +++ b/include/linux/fs.h
150 @@ -2664,6 +2664,7 @@ extern int inode_change_ok(const struct inode *, struct iattr *);
151  extern int inode_newsize_ok(const struct inode *, loff_t offset);
152  extern void setattr_copy(struct inode *inode, const struct iattr *attr);
153  
154 +extern int update_time(struct inode *, struct timespec *, int);
155  extern int file_update_time(struct file *file);
156  
157  extern int generic_show_options(struct seq_file *m, struct dentry *root);
158 diff --git a/include/linux/splice.h b/include/linux/splice.h
159 index da2751d..2e0fca6 100644
160 --- a/include/linux/splice.h
161 +++ b/include/linux/splice.h
162 @@ -83,4 +83,10 @@ extern void splice_shrink_spd(struct splice_pipe_desc *);
163  extern void spd_release_page(struct splice_pipe_desc *, unsigned int);
164  
165  extern const struct pipe_buf_operations page_cache_pipe_buf_ops;
166 +
167 +extern long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
168 +                          loff_t *ppos, size_t len, unsigned int flags);
169 +extern long do_splice_to(struct file *in, loff_t *ppos,
170 +                        struct pipe_inode_info *pipe, size_t len,
171 +                        unsigned int flags);
172  #endif
173 aufs3.18.1+ mmap patch
174
175 diff --git a/fs/buffer.c b/fs/buffer.c
176 index 20805db..363569f 100644
177 --- a/fs/buffer.c
178 +++ b/fs/buffer.c
179 @@ -2450,7 +2450,7 @@ int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
180          * Update file times before taking page lock. We may end up failing the
181          * fault so this update may be superfluous but who really cares...
182          */
183 -       file_update_time(vma->vm_file);
184 +       vma_file_update_time(vma);
185  
186         ret = __block_page_mkwrite(vma, vmf, get_block);
187         sb_end_pagefault(sb);
188 diff --git a/fs/proc/base.c b/fs/proc/base.c
189 index 772efa4..2c944de 100644
190 --- a/fs/proc/base.c
191 +++ b/fs/proc/base.c
192 @@ -1735,7 +1735,7 @@ static int proc_map_files_get_link(struct dentry *dentry, struct path *path)
193         down_read(&mm->mmap_sem);
194         vma = find_exact_vma(mm, vm_start, vm_end);
195         if (vma && vma->vm_file) {
196 -               *path = vma->vm_file->f_path;
197 +               *path = vma_pr_or_file(vma)->f_path;
198                 path_get(path);
199                 rc = 0;
200         }
201 diff --git a/fs/proc/nommu.c b/fs/proc/nommu.c
202 index d4a3574..1397181 100644
203 --- a/fs/proc/nommu.c
204 +++ b/fs/proc/nommu.c
205 @@ -45,7 +45,10 @@ static int nommu_region_show(struct seq_file *m, struct vm_region *region)
206         file = region->vm_file;
207  
208         if (file) {
209 -               struct inode *inode = file_inode(region->vm_file);
210 +               struct inode *inode;
211 +
212 +               file = vmr_pr_or_file(region);
213 +               inode = file_inode(file);
214                 dev = inode->i_sb->s_dev;
215                 ino = inode->i_ino;
216         }
217 diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
218 index 4e0388c..fc429e7 100644
219 --- a/fs/proc/task_mmu.c
220 +++ b/fs/proc/task_mmu.c
221 @@ -276,7 +276,10 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
222         const char *name = NULL;
223  
224         if (file) {
225 -               struct inode *inode = file_inode(vma->vm_file);
226 +               struct inode *inode;
227 +
228 +               file = vma_pr_or_file(vma);
229 +               inode = file_inode(file);
230                 dev = inode->i_sb->s_dev;
231                 ino = inode->i_ino;
232                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
233 @@ -1440,7 +1443,7 @@ static int show_numa_map(struct seq_file *m, void *v, int is_pid)
234         struct proc_maps_private *proc_priv = &numa_priv->proc_maps;
235         struct vm_area_struct *vma = v;
236         struct numa_maps *md = &numa_priv->md;
237 -       struct file *file = vma->vm_file;
238 +       struct file *file = vma_pr_or_file(vma);
239         struct mm_struct *mm = vma->vm_mm;
240         struct mm_walk walk = {};
241         struct mempolicy *pol;
242 diff --git a/fs/proc/task_nommu.c b/fs/proc/task_nommu.c
243 index 599ec2e..de6cd6e 100644
244 --- a/fs/proc/task_nommu.c
245 +++ b/fs/proc/task_nommu.c
246 @@ -160,7 +160,10 @@ static int nommu_vma_show(struct seq_file *m, struct vm_area_struct *vma,
247         file = vma->vm_file;
248  
249         if (file) {
250 -               struct inode *inode = file_inode(vma->vm_file);
251 +               struct inode *inode;
252 +
253 +               file = vma_pr_or_file(file);
254 +               inode = file_inode(file);
255                 dev = inode->i_sb->s_dev;
256                 ino = inode->i_ino;
257                 pgoff = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
258 diff --git a/include/linux/mm.h b/include/linux/mm.h
259 index b464611..8027d51 100644
260 --- a/include/linux/mm.h
261 +++ b/include/linux/mm.h
262 @@ -1206,6 +1206,28 @@ static inline int fixup_user_fault(struct task_struct *tsk,
263  }
264  #endif
265  
266 +#ifdef CONFIG_MMU
267 +extern void vma_do_file_update_time(struct vm_area_struct *, const char[], int);
268 +extern struct file *vma_do_pr_or_file(struct vm_area_struct *, const char[],
269 +                                     int);
270 +extern void vma_do_get_file(struct vm_area_struct *, const char[], int);
271 +extern void vma_do_fput(struct vm_area_struct *, const char[], int);
272 +
273 +#define vma_file_update_time(vma)      vma_do_file_update_time(vma, __func__, \
274 +                                                               __LINE__)
275 +#define vma_pr_or_file(vma)            vma_do_pr_or_file(vma, __func__, \
276 +                                                         __LINE__)
277 +#define vma_get_file(vma)              vma_do_get_file(vma, __func__, __LINE__)
278 +#define vma_fput(vma)                  vma_do_fput(vma, __func__, __LINE__)
279 +#else
280 +extern struct file *vmr_do_pr_or_file(struct vm_region *, const char[], int);
281 +extern void vmr_do_fput(struct vm_region *, const char[], int);
282 +
283 +#define vmr_pr_or_file(region)         vmr_do_pr_or_file(region, __func__, \
284 +                                                         __LINE__)
285 +#define vmr_fput(region)               vmr_do_fput(region, __func__, __LINE__)
286 +#endif /* CONFIG_MMU */
287 +
288  extern int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write);
289  extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
290                 void *buf, int len, int write);
291 diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
292 index 6e0b286..8f374ed 100644
293 --- a/include/linux/mm_types.h
294 +++ b/include/linux/mm_types.h
295 @@ -232,6 +232,7 @@ struct vm_region {
296         unsigned long   vm_top;         /* region allocated to here */
297         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
298         struct file     *vm_file;       /* the backing file or NULL */
299 +       struct file     *vm_prfile;     /* the virtual backing file or NULL */
300  
301         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
302         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
303 @@ -300,6 +301,7 @@ struct vm_area_struct {
304         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
305                                            units, *not* PAGE_CACHE_SIZE */
306         struct file * vm_file;          /* File we map to (can be NULL). */
307 +       struct file *vm_prfile;         /* shadow of vm_file */
308         void * vm_private_data;         /* was vm_pte (shared mem) */
309  
310  #ifndef CONFIG_MMU
311 diff --git a/kernel/fork.c b/kernel/fork.c
312 index 9b7d746..9a3b8fe 100644
313 --- a/kernel/fork.c
314 +++ b/kernel/fork.c
315 @@ -430,7 +430,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
316                         struct inode *inode = file_inode(file);
317                         struct address_space *mapping = file->f_mapping;
318  
319 -                       get_file(file);
320 +                       vma_get_file(tmp);
321                         if (tmp->vm_flags & VM_DENYWRITE)
322                                 atomic_dec(&inode->i_writecount);
323                         mutex_lock(&mapping->i_mmap_mutex);
324 diff --git a/mm/Makefile b/mm/Makefile
325 index 8405eb0..e0bda2d 100644
326 --- a/mm/Makefile
327 +++ b/mm/Makefile
328 @@ -18,7 +18,7 @@ obj-y                 := filemap.o mempool.o oom_kill.o \
329                            mm_init.o mmu_context.o percpu.o slab_common.o \
330                            compaction.o vmacache.o \
331                            interval_tree.o list_lru.o workingset.o \
332 -                          iov_iter.o debug.o $(mmu-y)
333 +                          iov_iter.o prfile.o debug.o $(mmu-y)
334  
335  obj-y += init-mm.o
336  
337 diff --git a/mm/filemap.c b/mm/filemap.c
338 index 14b4642..99bc835 100644
339 --- a/mm/filemap.c
340 +++ b/mm/filemap.c
341 @@ -2067,7 +2067,7 @@ int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
342         int ret = VM_FAULT_LOCKED;
343  
344         sb_start_pagefault(inode->i_sb);
345 -       file_update_time(vma->vm_file);
346 +       vma_file_update_time(vma);
347         lock_page(page);
348         if (page->mapping != inode->i_mapping) {
349                 unlock_page(page);
350 diff --git a/mm/fremap.c b/mm/fremap.c
351 index 72b8fa3..a00bbf0 100644
352 --- a/mm/fremap.c
353 +++ b/mm/fremap.c
354 @@ -224,16 +224,28 @@ get_write_lock:
355                  */
356                 if (mapping_cap_account_dirty(mapping)) {
357                         unsigned long addr;
358 -                       struct file *file = get_file(vma->vm_file);
359 +                       struct file *file = vma->vm_file,
360 +                               *prfile = vma->vm_prfile;
361 +
362                         /* mmap_region may free vma; grab the info now */
363                         vm_flags = vma->vm_flags;
364  
365 +                       vma_get_file(vma);
366                         addr = mmap_region(file, start, size, vm_flags, pgoff);
367 -                       fput(file);
368 +                       vma_fput(vma);
369                         if (IS_ERR_VALUE(addr)) {
370                                 err = addr;
371                         } else {
372                                 BUG_ON(addr != start);
373 +                               if (prfile) {
374 +                                       struct vm_area_struct *new_vma;
375 +
376 +                                       new_vma = find_vma(mm, addr);
377 +                                       if (!new_vma->vm_prfile)
378 +                                               new_vma->vm_prfile = prfile;
379 +                                       if (new_vma != vma)
380 +                                               get_file(prfile);
381 +                               }
382                                 err = 0;
383                         }
384                         goto out_freed;
385 diff --git a/mm/madvise.c b/mm/madvise.c
386 index 0938b30..0b66856 100644
387 --- a/mm/madvise.c
388 +++ b/mm/madvise.c
389 @@ -324,12 +324,12 @@ static long madvise_remove(struct vm_area_struct *vma,
390          * vma's reference to the file) can go away as soon as we drop
391          * mmap_sem.
392          */
393 -       get_file(f);
394 +       vma_get_file(vma);
395         up_read(&current->mm->mmap_sem);
396         error = vfs_fallocate(f,
397                                 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
398                                 offset, end - start);
399 -       fput(f);
400 +       vma_fput(vma);
401         down_read(&current->mm->mmap_sem);
402         return error;
403  }
404 diff --git a/mm/memory.c b/mm/memory.c
405 index d5f2ae9..0830a96 100644
406 --- a/mm/memory.c
407 +++ b/mm/memory.c
408 @@ -2163,7 +2163,7 @@ reuse:
409                         set_page_dirty_balance(dirty_page);
410                         /* file_update_time outside page_lock */
411                         if (vma->vm_file)
412 -                               file_update_time(vma->vm_file);
413 +                               vma_file_update_time(vma);
414                 }
415                 put_page(dirty_page);
416                 if (page_mkwrite) {
417 diff --git a/mm/mmap.c b/mm/mmap.c
418 index ae91989..e3bee5c 100644
419 --- a/mm/mmap.c
420 +++ b/mm/mmap.c
421 @@ -277,7 +277,7 @@ static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
422         if (vma->vm_ops && vma->vm_ops->close)
423                 vma->vm_ops->close(vma);
424         if (vma->vm_file)
425 -               fput(vma->vm_file);
426 +               vma_fput(vma);
427         mpol_put(vma_policy(vma));
428         kmem_cache_free(vm_area_cachep, vma);
429         return next;
430 @@ -895,7 +895,7 @@ again:                      remove_next = 1 + (end > next->vm_end);
431         if (remove_next) {
432                 if (file) {
433                         uprobe_munmap(next, next->vm_start, next->vm_end);
434 -                       fput(file);
435 +                       vma_fput(vma);
436                 }
437                 if (next->anon_vma)
438                         anon_vma_merge(vma, next);
439 @@ -1680,8 +1680,8 @@ out:
440         return addr;
441  
442  unmap_and_free_vma:
443 +       vma_fput(vma);
444         vma->vm_file = NULL;
445 -       fput(file);
446  
447         /* Undo any partial mapping done by a device driver. */
448         unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
449 @@ -2477,7 +2477,7 @@ static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
450                 goto out_free_mpol;
451  
452         if (new->vm_file)
453 -               get_file(new->vm_file);
454 +               vma_get_file(new);
455  
456         if (new->vm_ops && new->vm_ops->open)
457                 new->vm_ops->open(new);
458 @@ -2496,7 +2496,7 @@ static int __split_vma(struct mm_struct *mm, struct vm_area_struct *vma,
459         if (new->vm_ops && new->vm_ops->close)
460                 new->vm_ops->close(new);
461         if (new->vm_file)
462 -               fput(new->vm_file);
463 +               vma_fput(new);
464         unlink_anon_vmas(new);
465   out_free_mpol:
466         mpol_put(vma_policy(new));
467 @@ -2886,7 +2886,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
468                         if (anon_vma_clone(new_vma, vma))
469                                 goto out_free_mempol;
470                         if (new_vma->vm_file)
471 -                               get_file(new_vma->vm_file);
472 +                               vma_get_file(new_vma);
473                         if (new_vma->vm_ops && new_vma->vm_ops->open)
474                                 new_vma->vm_ops->open(new_vma);
475                         vma_link(mm, new_vma, prev, rb_link, rb_parent);
476 diff --git a/mm/msync.c b/mm/msync.c
477 index 992a167..ce1915b 100644
478 --- a/mm/msync.c
479 +++ b/mm/msync.c
480 @@ -84,13 +84,13 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags)
481                 start = vma->vm_end;
482                 if ((flags & MS_SYNC) && file &&
483                                 (vma->vm_flags & VM_SHARED)) {
484 -                       get_file(file);
485 +                       vma_get_file(vma);
486                         up_read(&mm->mmap_sem);
487                         if (vma->vm_flags & VM_NONLINEAR)
488                                 error = vfs_fsync(file, 1);
489                         else
490                                 error = vfs_fsync_range(file, fstart, fend, 1);
491 -                       fput(file);
492 +                       vma_fput(vma);
493                         if (error || start >= end)
494                                 goto out;
495                         down_read(&mm->mmap_sem);
496 diff --git a/mm/nommu.c b/mm/nommu.c
497 index bd1808e..c9ea035 100644
498 --- a/mm/nommu.c
499 +++ b/mm/nommu.c
500 @@ -658,7 +658,7 @@ static void __put_nommu_region(struct vm_region *region)
501                 up_write(&nommu_region_sem);
502  
503                 if (region->vm_file)
504 -                       fput(region->vm_file);
505 +                       vmr_fput(region);
506  
507                 /* IO memory and memory shared directly out of the pagecache
508                  * from ramfs/tmpfs mustn't be released here */
509 @@ -823,7 +823,7 @@ static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
510         if (vma->vm_ops && vma->vm_ops->close)
511                 vma->vm_ops->close(vma);
512         if (vma->vm_file)
513 -               fput(vma->vm_file);
514 +               vma_fput(vma);
515         put_nommu_region(vma->vm_region);
516         kmem_cache_free(vm_area_cachep, vma);
517  }
518 @@ -1385,7 +1385,7 @@ unsigned long do_mmap_pgoff(struct file *file,
519                                         goto error_just_free;
520                                 }
521                         }
522 -                       fput(region->vm_file);
523 +                       vmr_fput(region);
524                         kmem_cache_free(vm_region_jar, region);
525                         region = pregion;
526                         result = start;
527 @@ -1461,10 +1461,10 @@ error_just_free:
528         up_write(&nommu_region_sem);
529  error:
530         if (region->vm_file)
531 -               fput(region->vm_file);
532 +               vmr_fput(region);
533         kmem_cache_free(vm_region_jar, region);
534         if (vma->vm_file)
535 -               fput(vma->vm_file);
536 +               vma_fput(vma);
537         kmem_cache_free(vm_area_cachep, vma);
538         kleave(" = %d", ret);
539         return ret;
540 diff --git a/mm/prfile.c b/mm/prfile.c
541 new file mode 100644
542 index 0000000..fc708d2
543 --- /dev/null
544 +++ b/mm/prfile.c
545 @@ -0,0 +1,86 @@
546 +/*
547 + * Mainly for aufs which mmap(2) diffrent file and wants to print different path
548 + * in /proc/PID/maps.
549 + * Call these functions via macros defined in linux/mm.h.
550 + *
551 + * See Documentation/filesystems/aufs/design/06mmap.txt
552 + *
553 + * Copyright (c) 2014 Junjro R. Okajima
554 + * Copyright (c) 2014 Ian Campbell
555 + */
556 +
557 +#include <linux/mm.h>
558 +#include <linux/file.h>
559 +#include <linux/fs.h>
560 +
561 +/* #define PRFILE_TRACE */
562 +static inline void prfile_trace(struct file *f, struct file *pr,
563 +                             const char func[], int line, const char func2[])
564 +{
565 +#ifdef PRFILE_TRACE
566 +       if (pr)
567 +               pr_info("%s:%d: %s, %p\n", func, line, func2,
568 +                       f ? (char *)f->f_dentry->d_name.name : "(null)");
569 +#endif
570 +}
571 +
572 +#ifdef CONFIG_MMU
573 +void vma_do_file_update_time(struct vm_area_struct *vma, const char func[],
574 +                            int line)
575 +{
576 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
577 +
578 +       prfile_trace(f, pr, func, line, __func__);
579 +       file_update_time(f);
580 +       if (f && pr)
581 +               file_update_time(pr);
582 +}
583 +
584 +struct file *vma_do_pr_or_file(struct vm_area_struct *vma, const char func[],
585 +                              int line)
586 +{
587 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
588 +
589 +       prfile_trace(f, pr, func, line, __func__);
590 +       return (f && pr) ? pr : f;
591 +}
592 +
593 +void vma_do_get_file(struct vm_area_struct *vma, const char func[], int line)
594 +{
595 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
596 +
597 +       prfile_trace(f, pr, func, line, __func__);
598 +       get_file(f);
599 +       if (f && pr)
600 +               get_file(pr);
601 +}
602 +
603 +void vma_do_fput(struct vm_area_struct *vma, const char func[], int line)
604 +{
605 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
606 +
607 +       prfile_trace(f, pr, func, line, __func__);
608 +       fput(f);
609 +       if (f && pr)
610 +               fput(pr);
611 +}
612 +#else
613 +struct file *vmr_do_pr_or_file(struct vm_region *region, const char func[],
614 +                              int line)
615 +{
616 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
617 +
618 +       prfile_trace(f, pr, func, line, __func__);
619 +       return (f && pr) ? pr : f;
620 +}
621 +
622 +void vmr_do_fput(struct vm_region *region, const char func[], int line)
623 +{
624 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
625 +
626 +       prfile_trace(f, pr, func, line, __func__);
627 +       fput(f);
628 +       if (f && pr)
629 +               fput(pr);
630 +}
631 +#endif /* CONFIG_MMU */
632 aufs3.18.1+ standalone patch
633
634 diff --git a/fs/dcache.c b/fs/dcache.c
635 index da7342e..b147b6c 100644
636 --- a/fs/dcache.c
637 +++ b/fs/dcache.c
638 @@ -1124,6 +1124,7 @@ rename_retry:
639         seq = 1;
640         goto again;
641  }
642 +EXPORT_SYMBOL(d_walk);
643  
644  /*
645   * Search for at least 1 mount point in the dentry's subdirs.
646 diff --git a/fs/inode.c b/fs/inode.c
647 index df21e66..c8df03d 100644
648 --- a/fs/inode.c
649 +++ b/fs/inode.c
650 @@ -57,6 +57,7 @@ static struct hlist_head *inode_hashtable __read_mostly;
651  static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
652  
653  __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
654 +EXPORT_SYMBOL(inode_sb_list_lock);
655  
656  /*
657   * Empty aops. Can be used for the cases where the user does not
658 @@ -1513,6 +1514,7 @@ int update_time(struct inode *inode, struct timespec *time, int flags)
659         mark_inode_dirty_sync(inode);
660         return 0;
661  }
662 +EXPORT_SYMBOL(update_time);
663  
664  /**
665   *     touch_atime     -       update the access time
666 diff --git a/fs/namespace.c b/fs/namespace.c
667 index 5b66b2b..68ff4e4 100644
668 --- a/fs/namespace.c
669 +++ b/fs/namespace.c
670 @@ -454,6 +454,7 @@ void __mnt_drop_write(struct vfsmount *mnt)
671         mnt_dec_writers(real_mount(mnt));
672         preempt_enable();
673  }
674 +EXPORT_SYMBOL_GPL(__mnt_drop_write);
675  
676  /**
677   * mnt_drop_write - give up write access to a mount
678 @@ -1727,6 +1728,7 @@ int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
679         }
680         return 0;
681  }
682 +EXPORT_SYMBOL(iterate_mounts);
683  
684  static void cleanup_group_ids(struct mount *mnt, struct mount *end)
685  {
686 diff --git a/fs/notify/group.c b/fs/notify/group.c
687 index d16b62c..06ca6bc 100644
688 --- a/fs/notify/group.c
689 +++ b/fs/notify/group.c
690 @@ -22,6 +22,7 @@
691  #include <linux/srcu.h>
692  #include <linux/rculist.h>
693  #include <linux/wait.h>
694 +#include <linux/module.h>
695  
696  #include <linux/fsnotify_backend.h>
697  #include "fsnotify.h"
698 @@ -72,6 +73,7 @@ void fsnotify_get_group(struct fsnotify_group *group)
699  {
700         atomic_inc(&group->refcnt);
701  }
702 +EXPORT_SYMBOL(fsnotify_get_group);
703  
704  /*
705   * Drop a reference to a group.  Free it if it's through.
706 @@ -81,6 +83,7 @@ void fsnotify_put_group(struct fsnotify_group *group)
707         if (atomic_dec_and_test(&group->refcnt))
708                 fsnotify_final_destroy_group(group);
709  }
710 +EXPORT_SYMBOL(fsnotify_put_group);
711  
712  /*
713   * Create a new fsnotify_group and hold a reference for the group returned.
714 @@ -109,6 +112,7 @@ struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops)
715  
716         return group;
717  }
718 +EXPORT_SYMBOL(fsnotify_alloc_group);
719  
720  int fsnotify_fasync(int fd, struct file *file, int on)
721  {
722 diff --git a/fs/notify/mark.c b/fs/notify/mark.c
723 index 34c38fa..d40cf58 100644
724 --- a/fs/notify/mark.c
725 +++ b/fs/notify/mark.c
726 @@ -109,6 +109,7 @@ void fsnotify_put_mark(struct fsnotify_mark *mark)
727                 mark->free_mark(mark);
728         }
729  }
730 +EXPORT_SYMBOL(fsnotify_put_mark);
731  
732  /*
733   * Any time a mark is getting freed we end up here.
734 @@ -191,6 +192,7 @@ void fsnotify_destroy_mark(struct fsnotify_mark *mark,
735         fsnotify_destroy_mark_locked(mark, group);
736         mutex_unlock(&group->mark_mutex);
737  }
738 +EXPORT_SYMBOL(fsnotify_destroy_mark);
739  
740  void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
741  {
742 @@ -311,6 +313,7 @@ err:
743  
744         return ret;
745  }
746 +EXPORT_SYMBOL(fsnotify_add_mark);
747  
748  int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
749                       struct inode *inode, struct vfsmount *mnt, int allow_dups)
750 @@ -372,6 +375,7 @@ void fsnotify_init_mark(struct fsnotify_mark *mark,
751         atomic_set(&mark->refcnt, 1);
752         mark->free_mark = free_mark;
753  }
754 +EXPORT_SYMBOL(fsnotify_init_mark);
755  
756  static int fsnotify_mark_destroy(void *ignored)
757  {
758 diff --git a/fs/open.c b/fs/open.c
759 index de92c13..65d8ab0 100644
760 --- a/fs/open.c
761 +++ b/fs/open.c
762 @@ -62,6 +62,7 @@ int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
763         mutex_unlock(&dentry->d_inode->i_mutex);
764         return ret;
765  }
766 +EXPORT_SYMBOL(do_truncate);
767  
768  long vfs_truncate(struct path *path, loff_t length)
769  {
770 diff --git a/fs/splice.c b/fs/splice.c
771 index 619359a..c14f60e 100644
772 --- a/fs/splice.c
773 +++ b/fs/splice.c
774 @@ -1127,6 +1127,7 @@ long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
775  
776         return splice_write(pipe, out, ppos, len, flags);
777  }
778 +EXPORT_SYMBOL(do_splice_from);
779  
780  /*
781   * Attempt to initiate a splice from a file to a pipe.
782 @@ -1153,6 +1154,7 @@ long do_splice_to(struct file *in, loff_t *ppos,
783  
784         return splice_read(in, ppos, pipe, len, flags);
785  }
786 +EXPORT_SYMBOL(do_splice_to);
787  
788  /**
789   * splice_direct_to_actor - splices data directly between two non-pipes
790 diff --git a/fs/xattr.c b/fs/xattr.c
791 index 64e83ef..bd71e53 100644
792 --- a/fs/xattr.c
793 +++ b/fs/xattr.c
794 @@ -207,6 +207,7 @@ vfs_getxattr_alloc(struct dentry *dentry, const char *name, char **xattr_value,
795         *xattr_value = value;
796         return error;
797  }
798 +EXPORT_SYMBOL(vfs_getxattr_alloc);
799  
800  /* Compare an extended attribute value with the given value */
801  int vfs_xattr_cmp(struct dentry *dentry, const char *xattr_name,
802 diff --git a/security/commoncap.c b/security/commoncap.c
803 index bab0611..3fa2f82 100644
804 --- a/security/commoncap.c
805 +++ b/security/commoncap.c
806 @@ -979,9 +979,11 @@ int cap_mmap_addr(unsigned long addr)
807         }
808         return ret;
809  }
810 +EXPORT_SYMBOL(cap_mmap_addr);
811  
812  int cap_mmap_file(struct file *file, unsigned long reqprot,
813                   unsigned long prot, unsigned long flags)
814  {
815         return 0;
816  }
817 +EXPORT_SYMBOL(cap_mmap_file);
818 diff --git a/security/device_cgroup.c b/security/device_cgroup.c
819 index 188c1d2..426d9af 100644
820 --- a/security/device_cgroup.c
821 +++ b/security/device_cgroup.c
822 @@ -7,6 +7,7 @@
823  #include <linux/device_cgroup.h>
824  #include <linux/cgroup.h>
825  #include <linux/ctype.h>
826 +#include <linux/export.h>
827  #include <linux/list.h>
828  #include <linux/uaccess.h>
829  #include <linux/seq_file.h>
830 @@ -849,6 +850,7 @@ int __devcgroup_inode_permission(struct inode *inode, int mask)
831         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
832                         access);
833  }
834 +EXPORT_SYMBOL(__devcgroup_inode_permission);
835  
836  int devcgroup_inode_mknod(int mode, dev_t dev)
837  {
838 diff --git a/security/security.c b/security/security.c
839 index 18b35c6..12c67af 100644
840 --- a/security/security.c
841 +++ b/security/security.c
842 @@ -407,6 +407,7 @@ int security_path_rmdir(struct path *dir, struct dentry *dentry)
843                 return 0;
844         return security_ops->path_rmdir(dir, dentry);
845  }
846 +EXPORT_SYMBOL(security_path_rmdir);
847  
848  int security_path_unlink(struct path *dir, struct dentry *dentry)
849  {
850 @@ -423,6 +424,7 @@ int security_path_symlink(struct path *dir, struct dentry *dentry,
851                 return 0;
852         return security_ops->path_symlink(dir, dentry, old_name);
853  }
854 +EXPORT_SYMBOL(security_path_symlink);
855  
856  int security_path_link(struct dentry *old_dentry, struct path *new_dir,
857                        struct dentry *new_dentry)
858 @@ -431,6 +433,7 @@ int security_path_link(struct dentry *old_dentry, struct path *new_dir,
859                 return 0;
860         return security_ops->path_link(old_dentry, new_dir, new_dentry);
861  }
862 +EXPORT_SYMBOL(security_path_link);
863  
864  int security_path_rename(struct path *old_dir, struct dentry *old_dentry,
865                          struct path *new_dir, struct dentry *new_dentry,
866 @@ -458,6 +461,7 @@ int security_path_truncate(struct path *path)
867                 return 0;
868         return security_ops->path_truncate(path);
869  }
870 +EXPORT_SYMBOL(security_path_truncate);
871  
872  int security_path_chmod(struct path *path, umode_t mode)
873  {
874 @@ -465,6 +469,7 @@ int security_path_chmod(struct path *path, umode_t mode)
875                 return 0;
876         return security_ops->path_chmod(path, mode);
877  }
878 +EXPORT_SYMBOL(security_path_chmod);
879  
880  int security_path_chown(struct path *path, kuid_t uid, kgid_t gid)
881  {
882 @@ -472,6 +477,7 @@ int security_path_chown(struct path *path, kuid_t uid, kgid_t gid)
883                 return 0;
884         return security_ops->path_chown(path, uid, gid);
885  }
886 +EXPORT_SYMBOL(security_path_chown);
887  
888  int security_path_chroot(struct path *path)
889  {
890 @@ -557,6 +563,7 @@ int security_inode_readlink(struct dentry *dentry)
891                 return 0;
892         return security_ops->inode_readlink(dentry);
893  }
894 +EXPORT_SYMBOL(security_inode_readlink);
895  
896  int security_inode_follow_link(struct dentry *dentry, struct nameidata *nd)
897  {
898 @@ -571,6 +578,7 @@ int security_inode_permission(struct inode *inode, int mask)
899                 return 0;
900         return security_ops->inode_permission(inode, mask);
901  }
902 +EXPORT_SYMBOL(security_inode_permission);
903  
904  int security_inode_setattr(struct dentry *dentry, struct iattr *attr)
905  {
906 @@ -693,6 +701,7 @@ int security_file_permission(struct file *file, int mask)
907  
908         return fsnotify_perm(file, mask);
909  }
910 +EXPORT_SYMBOL(security_file_permission);
911  
912  int security_file_alloc(struct file *file)
913  {
914 @@ -753,6 +762,7 @@ int security_mmap_file(struct file *file, unsigned long prot,
915                 return ret;
916         return ima_file_mmap(file, prot);
917  }
918 +EXPORT_SYMBOL(security_mmap_file);
919  
920  int security_mmap_addr(unsigned long addr)
921  {
922 diff -urN /usr/share/empty/Documentation/ABI/testing/debugfs-aufs linux/Documentation/ABI/testing/debugfs-aufs
923 --- /usr/share/empty/Documentation/ABI/testing/debugfs-aufs     1970-01-01 01:00:00.000000000 +0100
924 +++ linux/Documentation/ABI/testing/debugfs-aufs        2015-01-25 13:00:38.627713742 +0100
925 @@ -0,0 +1,50 @@
926 +What:          /debug/aufs/si_<id>/
927 +Date:          March 2009
928 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
929 +Description:
930 +               Under /debug/aufs, a directory named si_<id> is created
931 +               per aufs mount, where <id> is a unique id generated
932 +               internally.
933 +
934 +What:          /debug/aufs/si_<id>/plink
935 +Date:          Apr 2013
936 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
937 +Description:
938 +               It has three lines and shows the information about the
939 +               pseudo-link. The first line is a single number
940 +               representing a number of buckets. The second line is a
941 +               number of pseudo-links per buckets (separated by a
942 +               blank). The last line is a single number representing a
943 +               total number of psedo-links.
944 +               When the aufs mount option 'noplink' is specified, it
945 +               will show "1\n0\n0\n".
946 +
947 +What:          /debug/aufs/si_<id>/xib
948 +Date:          March 2009
949 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
950 +Description:
951 +               It shows the consumed blocks by xib (External Inode Number
952 +               Bitmap), its block size and file size.
953 +               When the aufs mount option 'noxino' is specified, it
954 +               will be empty. About XINO files, see the aufs manual.
955 +
956 +What:          /debug/aufs/si_<id>/xino0, xino1 ... xinoN
957 +Date:          March 2009
958 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
959 +Description:
960 +               It shows the consumed blocks by xino (External Inode Number
961 +               Translation Table), its link count, block size and file
962 +               size.
963 +               When the aufs mount option 'noxino' is specified, it
964 +               will be empty. About XINO files, see the aufs manual.
965 +
966 +What:          /debug/aufs/si_<id>/xigen
967 +Date:          March 2009
968 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
969 +Description:
970 +               It shows the consumed blocks by xigen (External Inode
971 +               Generation Table), its block size and file size.
972 +               If CONFIG_AUFS_EXPORT is disabled, this entry will not
973 +               be created.
974 +               When the aufs mount option 'noxino' is specified, it
975 +               will be empty. About XINO files, see the aufs manual.
976 diff -urN /usr/share/empty/Documentation/ABI/testing/sysfs-aufs linux/Documentation/ABI/testing/sysfs-aufs
977 --- /usr/share/empty/Documentation/ABI/testing/sysfs-aufs       1970-01-01 01:00:00.000000000 +0100
978 +++ linux/Documentation/ABI/testing/sysfs-aufs  2015-01-25 13:00:38.627713742 +0100
979 @@ -0,0 +1,31 @@
980 +What:          /sys/fs/aufs/si_<id>/
981 +Date:          March 2009
982 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
983 +Description:
984 +               Under /sys/fs/aufs, a directory named si_<id> is created
985 +               per aufs mount, where <id> is a unique id generated
986 +               internally.
987 +
988 +What:          /sys/fs/aufs/si_<id>/br0, br1 ... brN
989 +Date:          March 2009
990 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
991 +Description:
992 +               It shows the abolute path of a member directory (which
993 +               is called branch) in aufs, and its permission.
994 +
995 +What:          /sys/fs/aufs/si_<id>/brid0, brid1 ... bridN
996 +Date:          July 2013
997 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
998 +Description:
999 +               It shows the id of a member directory (which is called
1000 +               branch) in aufs.
1001 +
1002 +What:          /sys/fs/aufs/si_<id>/xi_path
1003 +Date:          March 2009
1004 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
1005 +Description:
1006 +               It shows the abolute path of XINO (External Inode Number
1007 +               Bitmap, Translation Table and Generation Table) file
1008 +               even if it is the default path.
1009 +               When the aufs mount option 'noxino' is specified, it
1010 +               will be empty. About XINO files, see the aufs manual.
1011 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt linux/Documentation/filesystems/aufs/design/01intro.txt
1012 --- /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt  1970-01-01 01:00:00.000000000 +0100
1013 +++ linux/Documentation/filesystems/aufs/design/01intro.txt     2015-01-25 13:00:38.627713742 +0100
1014 @@ -0,0 +1,161 @@
1015 +
1016 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1017 +# 
1018 +# This program is free software; you can redistribute it and/or modify
1019 +# it under the terms of the GNU General Public License as published by
1020 +# the Free Software Foundation; either version 2 of the License, or
1021 +# (at your option) any later version.
1022 +# 
1023 +# This program is distributed in the hope that it will be useful,
1024 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1025 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1026 +# GNU General Public License for more details.
1027 +# 
1028 +# You should have received a copy of the GNU General Public License
1029 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1030 +
1031 +Introduction
1032 +----------------------------------------
1033 +
1034 +aufs [ei ju: ef es] | [a u f s]
1035 +1. abbrev. for "advanced multi-layered unification filesystem".
1036 +2. abbrev. for "another unionfs".
1037 +3. abbrev. for "auf das" in German which means "on the" in English.
1038 +   Ex. "Butter aufs Brot"(G) means "butter onto bread"(E).
1039 +   But "Filesystem aufs Filesystem" is hard to understand.
1040 +
1041 +AUFS is a filesystem with features:
1042 +- multi layered stackable unification filesystem, the member directory
1043 +  is called as a branch.
1044 +- branch permission and attribute, 'readonly', 'real-readonly',
1045 +  'readwrite', 'whiteout-able', 'link-able whiteout' and their
1046 +  combination.
1047 +- internal "file copy-on-write".
1048 +- logical deletion, whiteout.
1049 +- dynamic branch manipulation, adding, deleting and changing permission.
1050 +- allow bypassing aufs, user's direct branch access.
1051 +- external inode number translation table and bitmap which maintains the
1052 +  persistent aufs inode number.
1053 +- seekable directory, including NFS readdir.
1054 +- file mapping, mmap and sharing pages.
1055 +- pseudo-link, hardlink over branches.
1056 +- loopback mounted filesystem as a branch.
1057 +- several policies to select one among multiple writable branches.
1058 +- revert a single systemcall when an error occurs in aufs.
1059 +- and more...
1060 +
1061 +
1062 +Multi Layered Stackable Unification Filesystem
1063 +----------------------------------------------------------------------
1064 +Most people already knows what it is.
1065 +It is a filesystem which unifies several directories and provides a
1066 +merged single directory. When users access a file, the access will be
1067 +passed/re-directed/converted (sorry, I am not sure which English word is
1068 +correct) to the real file on the member filesystem. The member
1069 +filesystem is called 'lower filesystem' or 'branch' and has a mode
1070 +'readonly' and 'readwrite.' And the deletion for a file on the lower
1071 +readonly branch is handled by creating 'whiteout' on the upper writable
1072 +branch.
1073 +
1074 +On LKML, there have been discussions about UnionMount (Jan Blunck,
1075 +Bharata B Rao and Valerie Aurora) and Unionfs (Erez Zadok). They took
1076 +different approaches to implement the merged-view.
1077 +The former tries putting it into VFS, and the latter implements as a
1078 +separate filesystem.
1079 +(If I misunderstand about these implementations, please let me know and
1080 +I shall correct it. Because it is a long time ago when I read their
1081 +source files last time).
1082 +
1083 +UnionMount's approach will be able to small, but may be hard to share
1084 +branches between several UnionMount since the whiteout in it is
1085 +implemented in the inode on branch filesystem and always
1086 +shared. According to Bharata's post, readdir does not seems to be
1087 +finished yet.
1088 +There are several missing features known in this implementations such as
1089 +- for users, the inode number may change silently. eg. copy-up.
1090 +- link(2) may break by copy-up.
1091 +- read(2) may get an obsoleted filedata (fstat(2) too).
1092 +- fcntl(F_SETLK) may be broken by copy-up.
1093 +- unnecessary copy-up may happen, for example mmap(MAP_PRIVATE) after
1094 +  open(O_RDWR).
1095 +
1096 +Unionfs has a longer history. When I started implementing a stacking filesystem
1097 +(Aug 2005), it already existed. It has virtual super_block, inode,
1098 +dentry and file objects and they have an array pointing lower same kind
1099 +objects. After contributing many patches for Unionfs, I re-started my
1100 +project AUFS (Jun 2006).
1101 +
1102 +In AUFS, the structure of filesystem resembles to Unionfs, but I
1103 +implemented my own ideas, approaches and enhancements and it became
1104 +totally different one.
1105 +
1106 +Comparing DM snapshot and fs based implementation
1107 +- the number of bytes to be copied between devices is much smaller.
1108 +- the type of filesystem must be one and only.
1109 +- the fs must be writable, no readonly fs, even for the lower original
1110 +  device. so the compression fs will not be usable. but if we use
1111 +  loopback mount, we may address this issue.
1112 +  for instance,
1113 +       mount /cdrom/squashfs.img /sq
1114 +       losetup /sq/ext2.img
1115 +       losetup /somewhere/cow
1116 +       dmsetup "snapshot /dev/loop0 /dev/loop1 ..."
1117 +- it will be difficult (or needs more operations) to extract the
1118 +  difference between the original device and COW.
1119 +- DM snapshot-merge may help a lot when users try merging. in the
1120 +  fs-layer union, users will use rsync(1).
1121 +
1122 +
1123 +Several characters/aspects of aufs
1124 +----------------------------------------------------------------------
1125 +
1126 +Aufs has several characters or aspects.
1127 +1. a filesystem, callee of VFS helper
1128 +2. sub-VFS, caller of VFS helper for branches
1129 +3. a virtual filesystem which maintains persistent inode number
1130 +4. reader/writer of files on branches such like an application
1131 +
1132 +1. Callee of VFS Helper
1133 +As an ordinary linux filesystem, aufs is a callee of VFS. For instance,
1134 +unlink(2) from an application reaches sys_unlink() kernel function and
1135 +then vfs_unlink() is called. vfs_unlink() is one of VFS helper and it
1136 +calls filesystem specific unlink operation. Actually aufs implements the
1137 +unlink operation but it behaves like a redirector.
1138 +
1139 +2. Caller of VFS Helper for Branches
1140 +aufs_unlink() passes the unlink request to the branch filesystem as if
1141 +it were called from VFS. So the called unlink operation of the branch
1142 +filesystem acts as usual. As a caller of VFS helper, aufs should handle
1143 +every necessary pre/post operation for the branch filesystem.
1144 +- acquire the lock for the parent dir on a branch
1145 +- lookup in a branch
1146 +- revalidate dentry on a branch
1147 +- mnt_want_write() for a branch
1148 +- vfs_unlink() for a branch
1149 +- mnt_drop_write() for a branch
1150 +- release the lock on a branch
1151 +
1152 +3. Persistent Inode Number
1153 +One of the most important issue for a filesystem is to maintain inode
1154 +numbers. This is particularly important to support exporting a
1155 +filesystem via NFS. Aufs is a virtual filesystem which doesn't have a
1156 +backend block device for its own. But some storage is necessary to
1157 +maintain inode number. It may be a large space and may not suit to keep
1158 +in memory. Aufs rents some space from its first writable branch
1159 +filesystem (by default) and creates file(s) on it. These files are
1160 +created by aufs internally and removed soon (currently) keeping opened.
1161 +Note: Because these files are removed, they are totally gone after
1162 +      unmounting aufs. It means the inode numbers are not persistent
1163 +      across unmount or reboot. I have a plan to make them really
1164 +      persistent which will be important for aufs on NFS server.
1165 +
1166 +4. Read/Write Files Internally (copy-on-write)
1167 +Because a branch can be readonly, when you write a file on it, aufs will
1168 +"copy-up" it to the upper writable branch internally. And then write the
1169 +originally requested thing to the file. Generally kernel doesn't
1170 +open/read/write file actively. In aufs, even a single write may cause a
1171 +internal "file copy". This behaviour is very similar to cp(1) command.
1172 +
1173 +Some people may think it is better to pass such work to user space
1174 +helper, instead of doing in kernel space. Actually I am still thinking
1175 +about it. But currently I have implemented it in kernel space.
1176 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt linux/Documentation/filesystems/aufs/design/02struct.txt
1177 --- /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt 1970-01-01 01:00:00.000000000 +0100
1178 +++ linux/Documentation/filesystems/aufs/design/02struct.txt    2015-01-25 13:00:38.627713742 +0100
1179 @@ -0,0 +1,251 @@
1180 +
1181 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1182 +# 
1183 +# This program is free software; you can redistribute it and/or modify
1184 +# it under the terms of the GNU General Public License as published by
1185 +# the Free Software Foundation; either version 2 of the License, or
1186 +# (at your option) any later version.
1187 +# 
1188 +# This program is distributed in the hope that it will be useful,
1189 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1190 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1191 +# GNU General Public License for more details.
1192 +# 
1193 +# You should have received a copy of the GNU General Public License
1194 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1195 +
1196 +Basic Aufs Internal Structure
1197 +
1198 +Superblock/Inode/Dentry/File Objects
1199 +----------------------------------------------------------------------
1200 +As like an ordinary filesystem, aufs has its own
1201 +superblock/inode/dentry/file objects. All these objects have a
1202 +dynamically allocated array and store the same kind of pointers to the
1203 +lower filesystem, branch.
1204 +For example, when you build a union with one readwrite branch and one
1205 +readonly, mounted /au, /rw and /ro respectively.
1206 +- /au = /rw + /ro
1207 +- /ro/fileA exists but /rw/fileA
1208 +
1209 +Aufs lookup operation finds /ro/fileA and gets dentry for that. These
1210 +pointers are stored in a aufs dentry. The array in aufs dentry will be,
1211 +- [0] = NULL
1212 +- [1] = /ro/fileA
1213 +
1214 +This style of an array is essentially same to the aufs
1215 +superblock/inode/dentry/file objects.
1216 +
1217 +Because aufs supports manipulating branches, ie. add/delete/change
1218 +dynamically, these objects has its own generation. When branches are
1219 +changed, the generation in aufs superblock is incremented. And a
1220 +generation in other object are compared when it is accessed.
1221 +When a generation in other objects are obsoleted, aufs refreshes the
1222 +internal array.
1223 +
1224 +
1225 +Superblock
1226 +----------------------------------------------------------------------
1227 +Additionally aufs superblock has some data for policies to select one
1228 +among multiple writable branches, XIB files, pseudo-links and kobject.
1229 +See below in detail.
1230 +About the policies which supports copy-down a directory, see policy.txt
1231 +too.
1232 +
1233 +
1234 +Branch and XINO(External Inode Number Translation Table)
1235 +----------------------------------------------------------------------
1236 +Every branch has its own xino (external inode number translation table)
1237 +file. The xino file is created and unlinked by aufs internally. When two
1238 +members of a union exist on the same filesystem, they share the single
1239 +xino file.
1240 +The struct of a xino file is simple, just a sequence of aufs inode
1241 +numbers which is indexed by the lower inode number.
1242 +In the above sample, assume the inode number of /ro/fileA is i111 and
1243 +aufs assigns the inode number i999 for fileA. Then aufs writes 999 as
1244 +4(8) bytes at 111 * 4(8) bytes offset in the xino file.
1245 +
1246 +When the inode numbers are not contiguous, the xino file will be sparse
1247 +which has a hole in it and doesn't consume as much disk space as it
1248 +might appear. If your branch filesystem consumes disk space for such
1249 +holes, then you should specify 'xino=' option at mounting aufs.
1250 +
1251 +Also a writable branch has three kinds of "whiteout bases". All these
1252 +are existed when the branch is joined to aufs and the names are
1253 +whiteout-ed doubly, so that users will never see their names in aufs
1254 +hierarchy.
1255 +1. a regular file which will be linked to all whiteouts.
1256 +2. a directory to store a pseudo-link.
1257 +3. a directory to store an "orphan-ed" file temporary.
1258 +
1259 +1. Whiteout Base
1260 +   When you remove a file on a readonly branch, aufs handles it as a
1261 +   logical deletion and creates a whiteout on the upper writable branch
1262 +   as a hardlink of this file in order not to consume inode on the
1263 +   writable branch.
1264 +2. Pseudo-link Dir
1265 +   See below, Pseudo-link.
1266 +3. Step-Parent Dir
1267 +   When "fileC" exists on the lower readonly branch only and it is
1268 +   opened and removed with its parent dir, and then user writes
1269 +   something into it, then aufs copies-up fileC to this
1270 +   directory. Because there is no other dir to store fileC. After
1271 +   creating a file under this dir, the file is unlinked.
1272 +
1273 +Because aufs supports manipulating branches, ie. add/delete/change
1274 +dynamically, a branch has its own id. When the branch order changes, aufs
1275 +finds the new index by searching the branch id.
1276 +
1277 +
1278 +Pseudo-link
1279 +----------------------------------------------------------------------
1280 +Assume "fileA" exists on the lower readonly branch only and it is
1281 +hardlinked to "fileB" on the branch. When you write something to fileA,
1282 +aufs copies-up it to the upper writable branch. Additionally aufs
1283 +creates a hardlink under the Pseudo-link Directory of the writable
1284 +branch. The inode of a pseudo-link is kept in aufs super_block as a
1285 +simple list. If fileB is read after unlinking fileA, aufs returns
1286 +filedata from the pseudo-link instead of the lower readonly
1287 +branch. Because the pseudo-link is based upon the inode, to keep the
1288 +inode number by xino (see above) is important.
1289 +
1290 +All the hardlinks under the Pseudo-link Directory of the writable branch
1291 +should be restored in a proper location later. Aufs provides a utility
1292 +to do this. The userspace helpers executed at remounting and unmounting
1293 +aufs by default.
1294 +During this utility is running, it puts aufs into the pseudo-link
1295 +maintenance mode. In this mode, only the process which began the
1296 +maintenance mode (and its child processes) is allowed to operate in
1297 +aufs. Some other processes which are not related to the pseudo-link will
1298 +be allowed to run too, but the rest have to return an error or wait
1299 +until the maintenance mode ends. If a process already acquires an inode
1300 +mutex (in VFS), it has to return an error.
1301 +
1302 +
1303 +XIB(external inode number bitmap)
1304 +----------------------------------------------------------------------
1305 +Addition to the xino file per a branch, aufs has an external inode number
1306 +bitmap in a superblock object. It is also a file such like a xino file.
1307 +It is a simple bitmap to mark whether the aufs inode number is in-use or
1308 +not.
1309 +To reduce the file I/O, aufs prepares a single memory page to cache xib.
1310 +
1311 +Aufs implements a feature to truncate/refresh both of xino and xib to
1312 +reduce the number of consumed disk blocks for these files.
1313 +
1314 +
1315 +Virtual or Vertical Dir, and Readdir in Userspace
1316 +----------------------------------------------------------------------
1317 +In order to support multiple layers (branches), aufs readdir operation
1318 +constructs a virtual dir block on memory. For readdir, aufs calls
1319 +vfs_readdir() internally for each dir on branches, merges their entries
1320 +with eliminating the whiteout-ed ones, and sets it to file (dir)
1321 +object. So the file object has its entry list until it is closed. The
1322 +entry list will be updated when the file position is zero and becomes
1323 +old. This decision is made in aufs automatically.
1324 +
1325 +The dynamically allocated memory block for the name of entries has a
1326 +unit of 512 bytes (by default) and stores the names contiguously (no
1327 +padding). Another block for each entry is handled by kmem_cache too.
1328 +During building dir blocks, aufs creates hash list and judging whether
1329 +the entry is whiteouted by its upper branch or already listed.
1330 +The merged result is cached in the corresponding inode object and
1331 +maintained by a customizable life-time option.
1332 +
1333 +Some people may call it can be a security hole or invite DoS attack
1334 +since the opened and once readdir-ed dir (file object) holds its entry
1335 +list and becomes a pressure for system memory. But I'd say it is similar
1336 +to files under /proc or /sys. The virtual files in them also holds a
1337 +memory page (generally) while they are opened. When an idea to reduce
1338 +memory for them is introduced, it will be applied to aufs too.
1339 +For those who really hate this situation, I've developed readdir(3)
1340 +library which operates this merging in userspace. You just need to set
1341 +LD_PRELOAD environment variable, and aufs will not consume no memory in
1342 +kernel space for readdir(3).
1343 +
1344 +
1345 +Workqueue
1346 +----------------------------------------------------------------------
1347 +Aufs sometimes requires privilege access to a branch. For instance,
1348 +in copy-up/down operation. When a user process is going to make changes
1349 +to a file which exists in the lower readonly branch only, and the mode
1350 +of one of ancestor directories may not be writable by a user
1351 +process. Here aufs copy-up the file with its ancestors and they may
1352 +require privilege to set its owner/group/mode/etc.
1353 +This is a typical case of a application character of aufs (see
1354 +Introduction).
1355 +
1356 +Aufs uses workqueue synchronously for this case. It creates its own
1357 +workqueue. The workqueue is a kernel thread and has privilege. Aufs
1358 +passes the request to call mkdir or write (for example), and wait for
1359 +its completion. This approach solves a problem of a signal handler
1360 +simply.
1361 +If aufs didn't adopt the workqueue and changed the privilege of the
1362 +process, and if the mkdir/write call arises SIGXFSZ or other signal,
1363 +then the user process might gain a privilege or the generated core file
1364 +was owned by a superuser.
1365 +
1366 +Also aufs uses the system global workqueue ("events" kernel thread) too
1367 +for asynchronous tasks, such like handling inotify/fsnotify, re-creating a
1368 +whiteout base and etc. This is unrelated to a privilege.
1369 +Most of aufs operation tries acquiring a rw_semaphore for aufs
1370 +superblock at the beginning, at the same time waits for the completion
1371 +of all queued asynchronous tasks.
1372 +
1373 +
1374 +Whiteout
1375 +----------------------------------------------------------------------
1376 +The whiteout in aufs is very similar to Unionfs's. That is represented
1377 +by its filename. UnionMount takes an approach of a file mode, but I am
1378 +afraid several utilities (find(1) or something) will have to support it.
1379 +
1380 +Basically the whiteout represents "logical deletion" which stops aufs to
1381 +lookup further, but also it represents "dir is opaque" which also stop
1382 +lookup.
1383 +
1384 +In aufs, rmdir(2) and rename(2) for dir uses whiteout alternatively.
1385 +In order to make several functions in a single systemcall to be
1386 +revertible, aufs adopts an approach to rename a directory to a temporary
1387 +unique whiteouted name.
1388 +For example, in rename(2) dir where the target dir already existed, aufs
1389 +renames the target dir to a temporary unique whiteouted name before the
1390 +actual rename on a branch and then handles other actions (make it opaque,
1391 +update the attributes, etc). If an error happens in these actions, aufs
1392 +simply renames the whiteouted name back and returns an error. If all are
1393 +succeeded, aufs registers a function to remove the whiteouted unique
1394 +temporary name completely and asynchronously to the system global
1395 +workqueue.
1396 +
1397 +
1398 +Copy-up
1399 +----------------------------------------------------------------------
1400 +It is a well-known feature or concept.
1401 +When user modifies a file on a readonly branch, aufs operate "copy-up"
1402 +internally and makes change to the new file on the upper writable branch.
1403 +When the trigger systemcall does not update the timestamps of the parent
1404 +dir, aufs reverts it after copy-up.
1405 +
1406 +
1407 +Move-down (aufs3.9 and later)
1408 +----------------------------------------------------------------------
1409 +"Copy-up" is one of the essential feature in aufs. It copies a file from
1410 +the lower readonly branch to the upper writable branch when a user
1411 +changes something about the file.
1412 +"Move-down" is an opposite action of copy-up. Basically this action is
1413 +ran manually instead of automatically and internally.
1414 +For desgin and implementation, aufs has to consider these issues.
1415 +- whiteout for the file may exist on the lower branch.
1416 +- ancestor directories may not exist on the lower branch.
1417 +- diropq for the ancestor directories may exist on the upper branch.
1418 +- free space on the lower branch will reduce.
1419 +- another access to the file may happen during moving-down, including
1420 +  UDBA.
1421 +- the file should not be hard-linked nor pseudo-linked. they should be
1422 +  handled by auplink utility later.
1423 +
1424 +Sometimes users want to move-down a file from the upper writable branch
1425 +to the lower readonly or writable branch. For instance,
1426 +- the free space of the upper writable branch is going to run out.
1427 +- create a new intermediate branch between the upper and lower branch.
1428 +- etc.
1429 +
1430 +For this purpose, use "aumvdown" command in aufs-util.git.
1431 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt linux/Documentation/filesystems/aufs/design/03lookup.txt
1432 --- /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt 1970-01-01 01:00:00.000000000 +0100
1433 +++ linux/Documentation/filesystems/aufs/design/03lookup.txt    2015-01-25 13:00:38.627713742 +0100
1434 @@ -0,0 +1,133 @@
1435 +
1436 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1437 +# 
1438 +# This program is free software; you can redistribute it and/or modify
1439 +# it under the terms of the GNU General Public License as published by
1440 +# the Free Software Foundation; either version 2 of the License, or
1441 +# (at your option) any later version.
1442 +# 
1443 +# This program is distributed in the hope that it will be useful,
1444 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1445 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1446 +# GNU General Public License for more details.
1447 +# 
1448 +# You should have received a copy of the GNU General Public License
1449 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1450 +
1451 +Lookup in a Branch
1452 +----------------------------------------------------------------------
1453 +Since aufs has a character of sub-VFS (see Introduction), it operates
1454 +lookup for branches as VFS does. It may be a heavy work. Generally
1455 +speaking struct nameidata is a bigger structure and includes many
1456 +information. But almost all lookup operation in aufs is the simplest
1457 +case, ie. lookup only an entry directly connected to its parent. Digging
1458 +down the directory hierarchy is unnecessary.
1459 +
1460 +VFS has a function lookup_one_len() for that use, but it is not usable
1461 +for a branch filesystem which requires struct nameidata. So aufs
1462 +implements a simple lookup wrapper function. When a branch filesystem
1463 +allows NULL as nameidata, it calls lookup_one_len(). Otherwise it builds
1464 +a simplest nameidata and calls lookup_hash().
1465 +Here aufs applies "a principle in NFSD", ie. if the filesystem supports
1466 +NFS-export, then it has to support NULL as a nameidata parameter for
1467 +->create(), ->lookup() and ->d_revalidate(). So the lookup wrapper in
1468 +aufs tests if ->s_export_op in the branch is NULL or not.
1469 +
1470 +When a branch is a remote filesystem, aufs basically trusts its
1471 +->d_revalidate(), also aufs forces the hardest revalidate tests for
1472 +them.
1473 +For d_revalidate, aufs implements three levels of revalidate tests. See
1474 +"Revalidate Dentry and UDBA" in detail.
1475 +
1476 +
1477 +Test Only the Highest One for the Directory Permission (dirperm1 option)
1478 +----------------------------------------------------------------------
1479 +Let's try case study.
1480 +- aufs has two branches, upper readwrite and lower readonly.
1481 +  /au = /rw + /ro
1482 +- "dirA" exists under /ro, but /rw. and its mode is 0700.
1483 +- user invoked "chmod a+rx /au/dirA"
1484 +- the internal copy-up is activated and "/rw/dirA" is created and its
1485 +  permission bits are set to world readble.
1486 +- then "/au/dirA" becomes world readable?
1487 +
1488 +In this case, /ro/dirA is still 0700 since it exists in readonly branch,
1489 +or it may be a natively readonly filesystem. If aufs respects the lower
1490 +branch, it should not respond readdir request from other users. But user
1491 +allowed it by chmod. Should really aufs rejects showing the entries
1492 +under /ro/dirA?
1493 +
1494 +To be honest, I don't have a best solution for this case. So aufs
1495 +implements 'dirperm1' and 'nodirperm1' and leave it to users.
1496 +When dirperm1 is specified, aufs checks only the highest one for the
1497 +directory permission, and shows the entries. Otherwise, as usual, checks
1498 +every dir existing on all branches and rejects the request.
1499 +
1500 +As a side effect, dirperm1 option improves the performance of aufs
1501 +because the number of permission check is reduced when the number of
1502 +branch is many.
1503 +
1504 +
1505 +Loopback Mount
1506 +----------------------------------------------------------------------
1507 +Basically aufs supports any type of filesystem and block device for a
1508 +branch (actually there are some exceptions). But it is prohibited to add
1509 +a loopback mounted one whose backend file exists in a filesystem which is
1510 +already added to aufs. The reason is to protect aufs from a recursive
1511 +lookup. If it was allowed, the aufs lookup operation might re-enter a
1512 +lookup for the loopback mounted branch in the same context, and will
1513 +cause a deadlock.
1514 +
1515 +
1516 +Revalidate Dentry and UDBA (User's Direct Branch Access)
1517 +----------------------------------------------------------------------
1518 +Generally VFS helpers re-validate a dentry as a part of lookup.
1519 +0. digging down the directory hierarchy.
1520 +1. lock the parent dir by its i_mutex.
1521 +2. lookup the final (child) entry.
1522 +3. revalidate it.
1523 +4. call the actual operation (create, unlink, etc.)
1524 +5. unlock the parent dir
1525 +
1526 +If the filesystem implements its ->d_revalidate() (step 3), then it is
1527 +called. Actually aufs implements it and checks the dentry on a branch is
1528 +still valid.
1529 +But it is not enough. Because aufs has to release the lock for the
1530 +parent dir on a branch at the end of ->lookup() (step 2) and
1531 +->d_revalidate() (step 3) while the i_mutex of the aufs dir is still
1532 +held by VFS.
1533 +If the file on a branch is changed directly, eg. bypassing aufs, after
1534 +aufs released the lock, then the subsequent operation may cause
1535 +something unpleasant result.
1536 +
1537 +This situation is a result of VFS architecture, ->lookup() and
1538 +->d_revalidate() is separated. But I never say it is wrong. It is a good
1539 +design from VFS's point of view. It is just not suitable for sub-VFS
1540 +character in aufs.
1541 +
1542 +Aufs supports such case by three level of revalidation which is
1543 +selectable by user.
1544 +1. Simple Revalidate
1545 +   Addition to the native flow in VFS's, confirm the child-parent
1546 +   relationship on the branch just after locking the parent dir on the
1547 +   branch in the "actual operation" (step 4). When this validation
1548 +   fails, aufs returns EBUSY. ->d_revalidate() (step 3) in aufs still
1549 +   checks the validation of the dentry on branches.
1550 +2. Monitor Changes Internally by Inotify/Fsnotify
1551 +   Addition to above, in the "actual operation" (step 4) aufs re-lookup
1552 +   the dentry on the branch, and returns EBUSY if it finds different
1553 +   dentry.
1554 +   Additionally, aufs sets the inotify/fsnotify watch for every dir on branches
1555 +   during it is in cache. When the event is notified, aufs registers a
1556 +   function to kernel 'events' thread by schedule_work(). And the
1557 +   function sets some special status to the cached aufs dentry and inode
1558 +   private data. If they are not cached, then aufs has nothing to
1559 +   do. When the same file is accessed through aufs (step 0-3) later,
1560 +   aufs will detect the status and refresh all necessary data.
1561 +   In this mode, aufs has to ignore the event which is fired by aufs
1562 +   itself.
1563 +3. No Extra Validation
1564 +   This is the simplest test and doesn't add any additional revalidation
1565 +   test, and skip therevalidatin in step 4. It is useful and improves
1566 +   aufs performance when system surely hide the aufs branches from user,
1567 +   by over-mounting something (or another method).
1568 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt linux/Documentation/filesystems/aufs/design/04branch.txt
1569 --- /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt 1970-01-01 01:00:00.000000000 +0100
1570 +++ linux/Documentation/filesystems/aufs/design/04branch.txt    2015-01-25 13:00:38.627713742 +0100
1571 @@ -0,0 +1,75 @@
1572 +
1573 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1574 +# 
1575 +# This program is free software; you can redistribute it and/or modify
1576 +# it under the terms of the GNU General Public License as published by
1577 +# the Free Software Foundation; either version 2 of the License, or
1578 +# (at your option) any later version.
1579 +# 
1580 +# This program is distributed in the hope that it will be useful,
1581 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1582 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1583 +# GNU General Public License for more details.
1584 +# 
1585 +# You should have received a copy of the GNU General Public License
1586 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1587 +
1588 +Branch Manipulation
1589 +
1590 +Since aufs supports dynamic branch manipulation, ie. add/remove a branch
1591 +and changing its permission/attribute, there are a lot of works to do.
1592 +
1593 +
1594 +Add a Branch
1595 +----------------------------------------------------------------------
1596 +o Confirm the adding dir exists outside of aufs, including loopback
1597 +  mount.
1598 +- and other various attributes...
1599 +o Initialize the xino file and whiteout bases if necessary.
1600 +  See struct.txt.
1601 +
1602 +o Check the owner/group/mode of the directory
1603 +  When the owner/group/mode of the adding directory differs from the
1604 +  existing branch, aufs issues a warning because it may impose a
1605 +  security risk.
1606 +  For example, when a upper writable branch has a world writable empty
1607 +  top directory, a malicious user can create any files on the writable
1608 +  branch directly, like copy-up and modify manually. If something like
1609 +  /etc/{passwd,shadow} exists on the lower readonly branch but the upper
1610 +  writable branch, and the writable branch is world-writable, then a
1611 +  malicious guy may create /etc/passwd on the writable branch directly
1612 +  and the infected file will be valid in aufs.
1613 +  I am afraid it can be a security issue, but nothing to do except
1614 +  producing a warning.
1615 +
1616 +
1617 +Delete a Branch
1618 +----------------------------------------------------------------------
1619 +o Confirm the deleting branch is not busy
1620 +  To be general, there is one merit to adopt "remount" interface to
1621 +  manipulate branches. It is to discard caches. At deleting a branch,
1622 +  aufs checks the still cached (and connected) dentries and inodes. If
1623 +  there are any, then they are all in-use. An inode without its
1624 +  corresponding dentry can be alive alone (for example, inotify/fsnotify case).
1625 +
1626 +  For the cached one, aufs checks whether the same named entry exists on
1627 +  other branches.
1628 +  If the cached one is a directory, because aufs provides a merged view
1629 +  to users, as long as one dir is left on any branch aufs can show the
1630 +  dir to users. In this case, the branch can be removed from aufs.
1631 +  Otherwise aufs rejects deleting the branch.
1632 +
1633 +  If any file on the deleting branch is opened by aufs, then aufs
1634 +  rejects deleting.
1635 +
1636 +
1637 +Modify the Permission of a Branch
1638 +----------------------------------------------------------------------
1639 +o Re-initialize or remove the xino file and whiteout bases if necessary.
1640 +  See struct.txt.
1641 +
1642 +o rw --> ro: Confirm the modifying branch is not busy
1643 +  Aufs rejects the request if any of these conditions are true.
1644 +  - a file on the branch is mmap-ed.
1645 +  - a regular file on the branch is opened for write and there is no
1646 +    same named entry on the upper branch.
1647 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt linux/Documentation/filesystems/aufs/design/05wbr_policy.txt
1648 --- /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt     1970-01-01 01:00:00.000000000 +0100
1649 +++ linux/Documentation/filesystems/aufs/design/05wbr_policy.txt        2015-01-25 13:00:38.627713742 +0100
1650 @@ -0,0 +1,64 @@
1651 +
1652 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1653 +# 
1654 +# This program is free software; you can redistribute it and/or modify
1655 +# it under the terms of the GNU General Public License as published by
1656 +# the Free Software Foundation; either version 2 of the License, or
1657 +# (at your option) any later version.
1658 +# 
1659 +# This program is distributed in the hope that it will be useful,
1660 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1661 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1662 +# GNU General Public License for more details.
1663 +# 
1664 +# You should have received a copy of the GNU General Public License
1665 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1666 +
1667 +Policies to Select One among Multiple Writable Branches
1668 +----------------------------------------------------------------------
1669 +When the number of writable branch is more than one, aufs has to decide
1670 +the target branch for file creation or copy-up. By default, the highest
1671 +writable branch which has the parent (or ancestor) dir of the target
1672 +file is chosen (top-down-parent policy).
1673 +By user's request, aufs implements some other policies to select the
1674 +writable branch, for file creation two policies, round-robin and
1675 +most-free-space policies. For copy-up three policies, top-down-parent,
1676 +bottom-up-parent and bottom-up policies.
1677 +
1678 +As expected, the round-robin policy selects the branch in circular. When
1679 +you have two writable branches and creates 10 new files, 5 files will be
1680 +created for each branch. mkdir(2) systemcall is an exception. When you
1681 +create 10 new directories, all will be created on the same branch.
1682 +And the most-free-space policy selects the one which has most free
1683 +space among the writable branches. The amount of free space will be
1684 +checked by aufs internally, and users can specify its time interval.
1685 +
1686 +The policies for copy-up is more simple,
1687 +top-down-parent is equivalent to the same named on in create policy,
1688 +bottom-up-parent selects the writable branch where the parent dir
1689 +exists and the nearest upper one from the copyup-source,
1690 +bottom-up selects the nearest upper writable branch from the
1691 +copyup-source, regardless the existence of the parent dir.
1692 +
1693 +There are some rules or exceptions to apply these policies.
1694 +- If there is a readonly branch above the policy-selected branch and
1695 +  the parent dir is marked as opaque (a variation of whiteout), or the
1696 +  target (creating) file is whiteout-ed on the upper readonly branch,
1697 +  then the result of the policy is ignored and the target file will be
1698 +  created on the nearest upper writable branch than the readonly branch.
1699 +- If there is a writable branch above the policy-selected branch and
1700 +  the parent dir is marked as opaque or the target file is whiteouted
1701 +  on the branch, then the result of the policy is ignored and the target
1702 +  file will be created on the highest one among the upper writable
1703 +  branches who has diropq or whiteout. In case of whiteout, aufs removes
1704 +  it as usual.
1705 +- link(2) and rename(2) systemcalls are exceptions in every policy.
1706 +  They try selecting the branch where the source exists as possible
1707 +  since copyup a large file will take long time. If it can't be,
1708 +  ie. the branch where the source exists is readonly, then they will
1709 +  follow the copyup policy.
1710 +- There is an exception for rename(2) when the target exists.
1711 +  If the rename target exists, aufs compares the index of the branches
1712 +  where the source and the target exists and selects the higher
1713 +  one. If the selected branch is readonly, then aufs follows the
1714 +  copyup policy.
1715 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt linux/Documentation/filesystems/aufs/design/06fhsm.txt
1716 --- /usr/share/empty/Documentation/filesystems/aufs/design/06fhsm.txt   1970-01-01 01:00:00.000000000 +0100
1717 +++ linux/Documentation/filesystems/aufs/design/06fhsm.txt      2015-01-25 13:00:38.627713742 +0100
1718 @@ -0,0 +1,120 @@
1719 +
1720 +# Copyright (C) 2011-2014 Junjiro R. Okajima
1721 +# 
1722 +# This program is free software; you can redistribute it and/or modify
1723 +# it under the terms of the GNU General Public License as published by
1724 +# the Free Software Foundation; either version 2 of the License, or
1725 +# (at your option) any later version.
1726 +# 
1727 +# This program is distributed in the hope that it will be useful,
1728 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1729 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1730 +# GNU General Public License for more details.
1731 +# 
1732 +# You should have received a copy of the GNU General Public License
1733 +# along with this program; if not, write to the Free Software
1734 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
1735 +
1736 +
1737 +File-based Hierarchical Storage Management (FHSM)
1738 +----------------------------------------------------------------------
1739 +Hierarchical Storage Management (or HSM) is a well-known feature in the
1740 +storage world. Aufs provides this feature as file-based with multiple
1741 +writable branches, based upon the principle of "Colder-Lower".
1742 +Here the word "colder" means that the less used files, and "lower" means
1743 +that the position in the order of the stacked branches.
1744 +These multiple writable branches are prioritized, ie. the topmost one
1745 +should be the fastest drive and be used heavily.
1746 +
1747 +o Characters in aufs FHSM story
1748 +- aufs itself and a new branch attribute.
1749 +- a new ioctl interface to move-down and to establish a connection with
1750 +  the daemon ("move-down" is a converse of "copy-up").
1751 +- userspace tool and daemon.
1752 +
1753 +The userspace daemon establishes a connection with aufs and waits for
1754 +the notification. The notified information is very similar to struct
1755 +statfs containing the number of consumed blocks and inodes.
1756 +When the consumed blocks/inodes of a branch exceeds the user-specified
1757 +upper watermark, the daemon activates its move-down process until the
1758 +consumed blocks/inodes reaches the user-specified lower watermark.
1759 +
1760 +The actual move-down is done by aufs based upon the request from
1761 +user-space since we need to maintain the inode number and the internal
1762 +pointer arrays in aufs.
1763 +
1764 +Currently aufs FHSM handles the regular files only. Additionally they
1765 +must not be hard-linked nor pseudo-linked.
1766 +
1767 +
1768 +o Cowork of aufs and the user-space daemon
1769 +  During the userspace daemon established the connection, aufs sends a
1770 +  small notification to it whenever aufs writes something into the
1771 +  writable branch. But it may cost high since aufs issues statfs(2)
1772 +  internally. So user can specify a new option to cache the
1773 +  info. Actually the notification is controlled by these factors.
1774 +  + the specified cache time.
1775 +  + classified as "force" by aufs internally.
1776 +  Until the specified time expires, aufs doesn't send the info
1777 +  except the forced cases. When aufs decide forcing, the info is always
1778 +  notified to userspace.
1779 +  For example, the number of free inodes is generally large enough and
1780 +  the shortage of it happens rarely. So aufs doesn't force the
1781 +  notification when creating a new file, directory and others. This is
1782 +  the typical case which aufs doesn't force.
1783 +  When aufs writes the actual filedata and the files consumes any of new
1784 +  blocks, the aufs forces notifying.
1785 +
1786 +
1787 +o Interfaces in aufs
1788 +- New branch attribute.
1789 +  + fhsm
1790 +    Specifies that the branch is managed by FHSM feature. In other word,
1791 +    participant in the FHSM.
1792 +    When nofhsm is set to the branch, it will not be the source/target
1793 +    branch of the move-down operation. This attribute is set
1794 +    independently from coo and moo attributes, and if you want full
1795 +    FHSM, you should specify them as well.
1796 +- New mount option.
1797 +  + fhsm_sec
1798 +    Specifies a second to suppress many less important info to be
1799 +    notified.
1800 +- New ioctl.
1801 +  + AUFS_CTL_FHSM_FD
1802 +    create a new file descriptor which userspace can read the notification
1803 +    (a subset of struct statfs) from aufs.
1804 +- Module parameter 'brs'
1805 +  It has to be set to 1. Otherwise the new mount option 'fhsm' will not
1806 +  be set.
1807 +- mount helpers /sbin/mount.aufs and /sbin/umount.aufs
1808 +  When there are two or more branches with fhsm attributes,
1809 +  /sbin/mount.aufs invokes the user-space daemon and /sbin/umount.aufs
1810 +  terminates it. As a result of remounting and branch-manipulation, the
1811 +  number of branches with fhsm attribute can be one. In this case,
1812 +  /sbin/mount.aufs will terminate the user-space daemon.
1813 +
1814 +
1815 +Finally the operation is done as these steps in kernel-space.
1816 +- make sure that,
1817 +  + no one else is using the file.
1818 +  + the file is not hard-linked.
1819 +  + the file is not pseudo-linked.
1820 +  + the file is a regular file.
1821 +  + the parent dir is not opaqued.
1822 +- find the target writable branch.
1823 +- make sure the file is not whiteout-ed by the upper (than the target)
1824 +  branch.
1825 +- make the parent dir on the target branch.
1826 +- mutex lock the inode on the branch.
1827 +- unlink the whiteout on the target branch (if exists).
1828 +- lookup and create the whiteout-ed temporary name on the target branch.
1829 +- copy the file as the whiteout-ed temporary name on the target branch.
1830 +- rename the whiteout-ed temporary name to the original name.
1831 +- unlink the file on the source branch.
1832 +- maintain the internal pointer array and the external inode number
1833 +  table (XINO).
1834 +- maintain the timestamps and other attributes of the parent dir and the
1835 +  file.
1836 +
1837 +And of course, in every step, an error may happen. So the operation
1838 +should restore the original file state after an error happens.
1839 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt linux/Documentation/filesystems/aufs/design/06mmap.txt
1840 --- /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt   1970-01-01 01:00:00.000000000 +0100
1841 +++ linux/Documentation/filesystems/aufs/design/06mmap.txt      2015-01-25 13:00:38.627713742 +0100
1842 @@ -0,0 +1,46 @@
1843 +
1844 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1845 +# 
1846 +# This program is free software; you can redistribute it and/or modify
1847 +# it under the terms of the GNU General Public License as published by
1848 +# the Free Software Foundation; either version 2 of the License, or
1849 +# (at your option) any later version.
1850 +# 
1851 +# This program is distributed in the hope that it will be useful,
1852 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1853 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1854 +# GNU General Public License for more details.
1855 +# 
1856 +# You should have received a copy of the GNU General Public License
1857 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1858 +
1859 +mmap(2) -- File Memory Mapping
1860 +----------------------------------------------------------------------
1861 +In aufs, the file-mapped pages are handled by a branch fs directly, no
1862 +interaction with aufs. It means aufs_mmap() calls the branch fs's
1863 +->mmap().
1864 +This approach is simple and good, but there is one problem.
1865 +Under /proc, several entries show the mmap-ped files by its path (with
1866 +device and inode number), and the printed path will be the path on the
1867 +branch fs's instead of virtual aufs's.
1868 +This is not a problem in most cases, but some utilities lsof(1) (and its
1869 +user) may expect the path on aufs.
1870 +
1871 +To address this issue, aufs adds a new member called vm_prfile in struct
1872 +vm_area_struct (and struct vm_region). The original vm_file points to
1873 +the file on the branch fs in order to handle everything correctly as
1874 +usual. The new vm_prfile points to a virtual file in aufs, and the
1875 +show-functions in procfs refers to vm_prfile if it is set.
1876 +Also we need to maintain several other places where touching vm_file
1877 +such like
1878 +- fork()/clone() copies vma and the reference count of vm_file is
1879 +  incremented.
1880 +- merging vma maintains the ref count too.
1881 +
1882 +This is not a good approach. It just faking the printed path. But it
1883 +leaves all behaviour around f_mapping unchanged. This is surely an
1884 +advantage.
1885 +Actually aufs had adopted another complicated approach which calls
1886 +generic_file_mmap() and handles struct vm_operations_struct. In this
1887 +approach, aufs met a hard problem and I could not solve it without
1888 +switching the approach.
1889 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt linux/Documentation/filesystems/aufs/design/06xattr.txt
1890 --- /usr/share/empty/Documentation/filesystems/aufs/design/06xattr.txt  1970-01-01 01:00:00.000000000 +0100
1891 +++ linux/Documentation/filesystems/aufs/design/06xattr.txt     2015-01-25 13:00:38.627713742 +0100
1892 @@ -0,0 +1,96 @@
1893 +
1894 +# Copyright (C) 2014 Junjiro R. Okajima
1895 +#
1896 +# This program is free software; you can redistribute it and/or modify
1897 +# it under the terms of the GNU General Public License as published by
1898 +# the Free Software Foundation; either version 2 of the License, or
1899 +# (at your option) any later version.
1900 +#
1901 +# This program is distributed in the hope that it will be useful,
1902 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1903 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1904 +# GNU General Public License for more details.
1905 +#
1906 +# You should have received a copy of the GNU General Public License
1907 +# along with this program; if not, write to the Free Software
1908 +# Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
1909 +
1910 +
1911 +Listing XATTR/EA and getting the value
1912 +----------------------------------------------------------------------
1913 +For the inode standard attributes (owner, group, timestamps, etc.), aufs
1914 +shows the values from the topmost existing file. This behaviour is good
1915 +for the non-dir entreis since the bahaviour exactly matches the shown
1916 +information. But for the directories, aufs considers all the same named
1917 +entries on the lower branches. Which means, if one of the lower entry
1918 +rejects readdir call, then aufs returns an error even if the topmost
1919 +entry allows it. This behaviour is necessary to respect the branch fs's
1920 +security, but can make users confused since the user-visible standard
1921 +attributes don't match the behaviour.
1922 +To address this issue, aufs has a mount option called dirperm1 which
1923 +checks the permission for the topmost entry only, and ignores the lower
1924 +entry's permission.
1925 +
1926 +A similar issue can happen around XATTR.
1927 +getxattr(2) and listxattr(2) families behave as if dirperm1 option is
1928 +always set. Otherwise these very unpleasant situation can happen.
1929 +- listxattr(2) may return the duplicated entires.
1930 +- users may not be able to remove or reset the XATTR forever,
1931 +
1932 +
1933 +XATTR/EA support in the internal (copy,move)-(up,down)
1934 +----------------------------------------------------------------------
1935 +Generally the extended attributes of inode are categorazied as these.
1936 +- "security" for LSM and capability.
1937 +- "system" for posix ACL, 'acl' mount option is required for the branch
1938 +  fs generally.
1939 +- "trusted" for userspace, CAP_SYS_ADMIN is required.
1940 +- "user" for userspace, 'user_xattr' mount option is required for the
1941 +  branch fs generally.
1942 +
1943 +Moreover there are some other categories. Aufs handles these rather
1944 +unpopular categories as the ordinary ones, ie. there is no special
1945 +condition nor exception.
1946 +
1947 +In copy-up, the support for XATTR on the dst branch may differ from the
1948 +src branch. In this case, the copy-up operation will get an error and
1949 +the original user operation which triggered the copy-up fails. It can
1950 +happen that even all copy-up will fail.
1951 +When both of src and dst branches support XATTR and if an error occurs
1952 +during copying XATTR, then the copy-up should fail obviously. That is a
1953 +good reason and aufs should return an error to userspace. But when only
1954 +the src branch support XATTR, aufs should not return an error.
1955 +For example, the src branch supports ACL but the dst branch doesn't
1956 +because the dst branch may natively un-support it or temporary
1957 +un-support it due to "noacl" mount option. Of course, the dst branch fs
1958 +may NOT return an error even if the XATTR is not supported. It is
1959 +totally up to the branch fs.
1960 +
1961 +Anyway when the aufs internal copy-up gets an error from the dst branch
1962 +fs, then aufs tries removing the just copied entry and returns the error
1963 +to the userspace. The worst case of this situation will be all copy-up
1964 +will fail.
1965 +
1966 +For the copy-up operation, there two basic approaches.
1967 +- copy the specified XATTR only (by category above), and return the
1968 +  error if it happens inconditionally.
1969 +- copy all XATTR, and ignore the error on the specified category only.
1970 +
1971 +In order to support XATTR and to implement the correct behaviour, aufs
1972 +chooses the latter approach and introduces some attributes for its
1973 +branch, "icexsec", "icexsys", "icextr", "icexusr", and "icexoth".
1974 +They correspond to the XATTR namespaces (see above). Additionally, to be
1975 +convenient, "icex" is also provided which means all "ix*" attributes are
1976 +set.
1977 +
1978 +The meaning of these attributes is to ignore the error from setting
1979 +XATTR on that branch.
1980 +Note that aufs tries copying all XATTR unconditionally, and ignores the
1981 +error from the dst branch according to the specified attributes.
1982 +
1983 +Some XATTR may have its default value. The default value may come from
1984 +the parent dir or the environment. If the default value is set at the
1985 +file creating-time, it will be overwritten by copy-up.
1986 +Some contradiction may happen I am afraid.
1987 +Do we need another attribute to stop copying XATTR? I am unsure. For
1988 +now, aufs implements the branch attributes to ignore the error.
1989 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt linux/Documentation/filesystems/aufs/design/07export.txt
1990 --- /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt 1970-01-01 01:00:00.000000000 +0100
1991 +++ linux/Documentation/filesystems/aufs/design/07export.txt    2015-01-25 13:00:38.627713742 +0100
1992 @@ -0,0 +1,58 @@
1993 +
1994 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1995 +# 
1996 +# This program is free software; you can redistribute it and/or modify
1997 +# it under the terms of the GNU General Public License as published by
1998 +# the Free Software Foundation; either version 2 of the License, or
1999 +# (at your option) any later version.
2000 +# 
2001 +# This program is distributed in the hope that it will be useful,
2002 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2003 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2004 +# GNU General Public License for more details.
2005 +# 
2006 +# You should have received a copy of the GNU General Public License
2007 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2008 +
2009 +Export Aufs via NFS
2010 +----------------------------------------------------------------------
2011 +Here is an approach.
2012 +- like xino/xib, add a new file 'xigen' which stores aufs inode
2013 +  generation.
2014 +- iget_locked(): initialize aufs inode generation for a new inode, and
2015 +  store it in xigen file.
2016 +- destroy_inode(): increment aufs inode generation and store it in xigen
2017 +  file. it is necessary even if it is not unlinked, because any data of
2018 +  inode may be changed by UDBA.
2019 +- encode_fh(): for a root dir, simply return FILEID_ROOT. otherwise
2020 +  build file handle by
2021 +  + branch id (4 bytes)
2022 +  + superblock generation (4 bytes)
2023 +  + inode number (4 or 8 bytes)
2024 +  + parent dir inode number (4 or 8 bytes)
2025 +  + inode generation (4 bytes))
2026 +  + return value of exportfs_encode_fh() for the parent on a branch (4
2027 +    bytes)
2028 +  + file handle for a branch (by exportfs_encode_fh())
2029 +- fh_to_dentry():
2030 +  + find the index of a branch from its id in handle, and check it is
2031 +    still exist in aufs.
2032 +  + 1st level: get the inode number from handle and search it in cache.
2033 +  + 2nd level: if not found, get the parent inode number from handle and
2034 +    search it in cache. and then open the parent dir, find the matching
2035 +    inode number by vfs_readdir() and get its name, and call
2036 +    lookup_one_len() for the target dentry.
2037 +  + 3rd level: if the parent dir is not cached, call
2038 +    exportfs_decode_fh() for a branch and get the parent on a branch,
2039 +    build a pathname of it, convert it a pathname in aufs, call
2040 +    path_lookup(). now aufs gets a parent dir dentry, then handle it as
2041 +    the 2nd level.
2042 +  + to open the dir, aufs needs struct vfsmount. aufs keeps vfsmount
2043 +    for every branch, but not itself. to get this, (currently) aufs
2044 +    searches in current->nsproxy->mnt_ns list. it may not be a good
2045 +    idea, but I didn't get other approach.
2046 +  + test the generation of the gotten inode.
2047 +- every inode operation: they may get EBUSY due to UDBA. in this case,
2048 +  convert it into ESTALE for NFSD.
2049 +- readdir(): call lockdep_on/off() because filldir in NFSD calls
2050 +  lookup_one_len(), vfs_getattr(), encode_fh() and others.
2051 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt linux/Documentation/filesystems/aufs/design/08shwh.txt
2052 --- /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt   1970-01-01 01:00:00.000000000 +0100
2053 +++ linux/Documentation/filesystems/aufs/design/08shwh.txt      2015-01-25 13:00:38.627713742 +0100
2054 @@ -0,0 +1,52 @@
2055 +
2056 +# Copyright (C) 2005-2014 Junjiro R. Okajima
2057 +# 
2058 +# This program is free software; you can redistribute it and/or modify
2059 +# it under the terms of the GNU General Public License as published by
2060 +# the Free Software Foundation; either version 2 of the License, or
2061 +# (at your option) any later version.
2062 +# 
2063 +# This program is distributed in the hope that it will be useful,
2064 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2065 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2066 +# GNU General Public License for more details.
2067 +# 
2068 +# You should have received a copy of the GNU General Public License
2069 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2070 +
2071 +Show Whiteout Mode (shwh)
2072 +----------------------------------------------------------------------
2073 +Generally aufs hides the name of whiteouts. But in some cases, to show
2074 +them is very useful for users. For instance, creating a new middle layer
2075 +(branch) by merging existing layers.
2076 +
2077 +(borrowing aufs1 HOW-TO from a user, Michael Towers)
2078 +When you have three branches,
2079 +- Bottom: 'system', squashfs (underlying base system), read-only
2080 +- Middle: 'mods', squashfs, read-only
2081 +- Top: 'overlay', ram (tmpfs), read-write
2082 +
2083 +The top layer is loaded at boot time and saved at shutdown, to preserve
2084 +the changes made to the system during the session.
2085 +When larger changes have been made, or smaller changes have accumulated,
2086 +the size of the saved top layer data grows. At this point, it would be
2087 +nice to be able to merge the two overlay branches ('mods' and 'overlay')
2088 +and rewrite the 'mods' squashfs, clearing the top layer and thus
2089 +restoring save and load speed.
2090 +
2091 +This merging is simplified by the use of another aufs mount, of just the
2092 +two overlay branches using the 'shwh' option.
2093 +# mount -t aufs -o ro,shwh,br:/livesys/overlay=ro+wh:/livesys/mods=rr+wh \
2094 +       aufs /livesys/merge_union
2095 +
2096 +A merged view of these two branches is then available at
2097 +/livesys/merge_union, and the new feature is that the whiteouts are
2098 +visible!
2099 +Note that in 'shwh' mode the aufs mount must be 'ro', which will disable
2100 +writing to all branches. Also the default mode for all branches is 'ro'.
2101 +It is now possible to save the combined contents of the two overlay
2102 +branches to a new squashfs, e.g.:
2103 +# mksquashfs /livesys/merge_union /path/to/newmods.squash
2104 +
2105 +This new squashfs archive can be stored on the boot device and the
2106 +initramfs will use it to replace the old one at the next boot.
2107 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt linux/Documentation/filesystems/aufs/design/10dynop.txt
2108 --- /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt  1970-01-01 01:00:00.000000000 +0100
2109 +++ linux/Documentation/filesystems/aufs/design/10dynop.txt     2015-01-25 13:00:38.627713742 +0100
2110 @@ -0,0 +1,46 @@
2111 +
2112 +# Copyright (C) 2010-2014 Junjiro R. Okajima
2113 +#
2114 +# This program is free software; you can redistribute it and/or modify
2115 +# it under the terms of the GNU General Public License as published by
2116 +# the Free Software Foundation; either version 2 of the License, or
2117 +# (at your option) any later version.
2118 +#
2119 +# This program is distributed in the hope that it will be useful,
2120 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2121 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2122 +# GNU General Public License for more details.
2123 +#
2124 +# You should have received a copy of the GNU General Public License
2125 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2126 +
2127 +Dynamically customizable FS operations
2128 +----------------------------------------------------------------------
2129 +Generally FS operations (struct inode_operations, struct
2130 +address_space_operations, struct file_operations, etc.) are defined as
2131 +"static const", but it never means that FS have only one set of
2132 +operation. Some FS have multiple sets of them. For instance, ext2 has
2133 +three sets, one for XIP, for NOBH, and for normal.
2134 +Since aufs overrides and redirects these operations, sometimes aufs has
2135 +to change its behaviour according to the branch FS type. More imporantly
2136 +VFS acts differently if a function (member in the struct) is set or
2137 +not. It means aufs should have several sets of operations and select one
2138 +among them according to the branch FS definition.
2139 +
2140 +In order to solve this problem and not to affect the behavour of VFS,
2141 +aufs defines these operations dynamically. For instance, aufs defines
2142 +aio_read function for struct file_operations, but it may not be set to
2143 +the file_operations. When the branch FS doesn't have it, aufs doesn't
2144 +set it to its file_operations while the function definition itself is
2145 +still alive. So the behaviour of io_submit(2) will not change, and it
2146 +will return an error when aio_read is not defined.
2147 +
2148 +The lifetime of these dynamically generated operation object is
2149 +maintained by aufs branch object. When the branch is removed from aufs,
2150 +the reference counter of the object is decremented. When it reaches
2151 +zero, the dynamically generated operation object will be freed.
2152 +
2153 +This approach is designed to support AIO (io_submit), Direcit I/O and
2154 +XIP mainly.
2155 +Currently this approach is applied to file_operations and
2156 +vm_operations_struct for regular files only.
2157 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/99plan.txt linux/Documentation/filesystems/aufs/design/99plan.txt
2158 --- /usr/share/empty/Documentation/filesystems/aufs/design/99plan.txt   1970-01-01 01:00:00.000000000 +0100
2159 +++ linux/Documentation/filesystems/aufs/design/99plan.txt      2015-01-25 13:00:38.627713742 +0100
2160 @@ -0,0 +1,58 @@
2161 +
2162 +# Copyright (C) 2005-2014 Junjiro R. Okajima
2163 +# 
2164 +# This program is free software; you can redistribute it and/or modify
2165 +# it under the terms of the GNU General Public License as published by
2166 +# the Free Software Foundation; either version 2 of the License, or
2167 +# (at your option) any later version.
2168 +# 
2169 +# This program is distributed in the hope that it will be useful,
2170 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
2171 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2172 +# GNU General Public License for more details.
2173 +# 
2174 +# You should have received a copy of the GNU General Public License
2175 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
2176 +
2177 +Plan
2178 +
2179 +Restoring some features which was implemented in aufs1.
2180 +They were dropped in aufs2 in order to make source files simpler and
2181 +easier to be reviewed.
2182 +
2183 +
2184 +Being Another Aufs's Readonly Branch (robr)
2185 +----------------------------------------------------------------------
2186 +Aufs1 allows aufs to be another aufs's readonly branch.
2187 +This feature was developed by a user's request. But it may not be used
2188 +currecnly.
2189 +
2190 +
2191 +Refresh the Opened File (refrof)
2192 +----------------------------------------------------------------------
2193 +This option is implemented in aufs1 but incomplete.
2194 +
2195 +When user reads from a file, he expects to get its latest filedata
2196 +generally. If the file is removed and a new same named file is created,
2197 +the content he gets is unchanged, ie. the unlinked filedata.
2198 +
2199 +Let's try case study again.
2200 +- aufs has two branches.
2201 +  /au = /rw + /ro
2202 +- "fileA" exists under /ro, but /rw.
2203 +- user opened "/au/fileA".
2204 +- he or someone else inserts a branch (/new) between /rw and /ro.
2205 +  /au = /rw + /new + /ro
2206 +- the new branch has "fileA".
2207 +- user reads from the opened "fileA"
2208 +- which filedata should aufs return, from /ro or /new?
2209 +
2210 +Some people says it has to be "from /ro" and it is a semantics of Unix.
2211 +The others say it should be "from /new" because the file is not removed
2212 +and it is equivalent to the case of someone else modifies the file.
2213 +
2214 +Here again I don't have a best and final answer. I got an idea to
2215 +implement 'refrof' and 'norefrof' option. When 'refrof' (REFResh the
2216 +Opened File) is specified (by default), aufs returns the filedata from
2217 +/new.
2218 +Otherwise from /new.
2219 diff -urN /usr/share/empty/Documentation/filesystems/aufs/README linux/Documentation/filesystems/aufs/README
2220 --- /usr/share/empty/Documentation/filesystems/aufs/README      1970-01-01 01:00:00.000000000 +0100
2221 +++ linux/Documentation/filesystems/aufs/README 2015-01-25 13:00:38.627713742 +0100
2222 @@ -0,0 +1,370 @@
2223 +
2224 +Aufs3 -- advanced multi layered unification filesystem version 3.x
2225 +http://aufs.sf.net
2226 +Junjiro R. Okajima
2227 +
2228 +
2229 +0. Introduction
2230 +----------------------------------------
2231 +In the early days, aufs was entirely re-designed and re-implemented
2232 +Unionfs Version 1.x series. After many original ideas, approaches,
2233 +improvements and implementations, it becomes totally different from
2234 +Unionfs while keeping the basic features.
2235 +Recently, Unionfs Version 2.x series begin taking some of the same
2236 +approaches to aufs1's.
2237 +Unionfs is being developed by Professor Erez Zadok at Stony Brook
2238 +University and his team.
2239 +
2240 +Aufs3 supports linux-3.0 and later.
2241 +If you want older kernel version support, try aufs2-2.6.git or
2242 +aufs2-standalone.git repository, aufs1 from CVS on SourceForge.
2243 +
2244 +Note: it becomes clear that "Aufs was rejected. Let's give it up."
2245 +      According to Christoph Hellwig, linux rejects all union-type
2246 +      filesystems but UnionMount.
2247 +<http://marc.info/?l=linux-kernel&m=123938533724484&w=2>
2248 +
2249 +PS. Al Viro seems have a plan to merge aufs as well as overlayfs and
2250 +    UnionMount, and he pointed out an issue around a directory mutex
2251 +    lock and aufs addressed it. But it is still unsure whether aufs will
2252 +    be merged (or any other union solution).
2253 +<http://marc.info/?l=linux-kernel&m=136312705029295&w=1>
2254 +
2255 +
2256 +1. Features
2257 +----------------------------------------
2258 +- unite several directories into a single virtual filesystem. The member
2259 +  directory is called as a branch.
2260 +- you can specify the permission flags to the branch, which are 'readonly',
2261 +  'readwrite' and 'whiteout-able.'
2262 +- by upper writable branch, internal copyup and whiteout, files/dirs on
2263 +  readonly branch are modifiable logically.
2264 +- dynamic branch manipulation, add, del.
2265 +- etc...
2266 +
2267 +Also there are many enhancements in aufs1, such as:
2268 +- readdir(3) in userspace.
2269 +- keep inode number by external inode number table
2270 +- keep the timestamps of file/dir in internal copyup operation
2271 +- seekable directory, supporting NFS readdir.
2272 +- whiteout is hardlinked in order to reduce the consumption of inodes
2273 +  on branch
2274 +- do not copyup, nor create a whiteout when it is unnecessary
2275 +- revert a single systemcall when an error occurs in aufs
2276 +- remount interface instead of ioctl
2277 +- maintain /etc/mtab by an external command, /sbin/mount.aufs.
2278 +- loopback mounted filesystem as a branch
2279 +- kernel thread for removing the dir who has a plenty of whiteouts
2280 +- support copyup sparse file (a file which has a 'hole' in it)
2281 +- default permission flags for branches
2282 +- selectable permission flags for ro branch, whether whiteout can
2283 +  exist or not
2284 +- export via NFS.
2285 +- support <sysfs>/fs/aufs and <debugfs>/aufs.
2286 +- support multiple writable branches, some policies to select one
2287 +  among multiple writable branches.
2288 +- a new semantics for link(2) and rename(2) to support multiple
2289 +  writable branches.
2290 +- no glibc changes are required.
2291 +- pseudo hardlink (hardlink over branches)
2292 +- allow a direct access manually to a file on branch, e.g. bypassing aufs.
2293 +  including NFS or remote filesystem branch.
2294 +- userspace wrapper for pathconf(3)/fpathconf(3) with _PC_LINK_MAX.
2295 +- and more...
2296 +
2297 +Currently these features are dropped temporary from aufs3.
2298 +See design/08plan.txt in detail.
2299 +- test only the highest one for the directory permission (dirperm1)
2300 +- copyup on open (coo=)
2301 +- nested mount, i.e. aufs as readonly no-whiteout branch of another aufs
2302 +  (robr)
2303 +- statistics of aufs thread (/sys/fs/aufs/stat)
2304 +- delegation mode (dlgt)
2305 +  a delegation of the internal branch access to support task I/O
2306 +  accounting, which also supports Linux Security Modules (LSM) mainly
2307 +  for Suse AppArmor.
2308 +- intent.open/create (file open in a single lookup)
2309 +
2310 +Features or just an idea in the future (see also design/*.txt),
2311 +- reorder the branch index without del/re-add.
2312 +- permanent xino files for NFSD
2313 +- an option for refreshing the opened files after add/del branches
2314 +- 'move' policy for copy-up between two writable branches, after
2315 +  checking free space.
2316 +- light version, without branch manipulation. (unnecessary?)
2317 +- copyup in userspace
2318 +- inotify in userspace
2319 +- readv/writev
2320 +- xattr, acl
2321 +
2322 +
2323 +2. Download
2324 +----------------------------------------
2325 +There were three GIT trees for aufs3, aufs3-linux.git,
2326 +aufs3-standalone.git, and aufs-util.git. Note that there is no "3" in
2327 +"aufs-util.git."
2328 +While the aufs-util is always necessary, you need either of aufs3-linux
2329 +or aufs3-standalone.
2330 +
2331 +The aufs3-linux tree includes the whole linux mainline GIT tree,
2332 +git://git.kernel.org/.../torvalds/linux.git.
2333 +And you cannot select CONFIG_AUFS_FS=m for this version, eg. you cannot
2334 +build aufs3 as an external kernel module.
2335 +
2336 +On the other hand, the aufs3-standalone tree has only aufs source files
2337 +and necessary patches, and you can select CONFIG_AUFS_FS=m.
2338 +
2339 +You will find GIT branches whose name is in form of "aufs3.x" where "x"
2340 +represents the linux kernel version, "linux-3.x". For instance,
2341 +"aufs3.0" is for linux-3.0. For latest "linux-3.x-rcN", use
2342 +"aufs3.x-rcN" branch.
2343 +
2344 +o aufs3-linux tree
2345 +$ git clone --reference /your/linux/git/tree \
2346 +       git://git.code.sf.net/p/aufs/aufs3-linux aufs-aufs3-linux \
2347 +       aufs3-linux.git
2348 +- if you don't have linux GIT tree, then remove "--reference ..."
2349 +$ cd aufs3-linux.git
2350 +$ git checkout origin/aufs3.0
2351 +
2352 +o aufs3-standalone tree
2353 +$ git clone git://git.code.sf.net/p/aufs/aufs3-standalone \
2354 +       aufs3-standalone.git
2355 +$ cd aufs3-standalone.git
2356 +$ git checkout origin/aufs3.0
2357 +
2358 +o aufs-util tree
2359 +$ git clone git://git.code.sf.net/p/aufs/aufs-util \
2360 +       aufs-util.git
2361 +$ cd aufs-util.git
2362 +$ git checkout origin/aufs3.0
2363 +
2364 +Note: The 3.x-rcN branch is to be used with `rc' kernel versions ONLY.
2365 +The minor version number, 'x' in '3.x', of aufs may not always
2366 +follow the minor version number of the kernel.
2367 +Because changes in the kernel that cause the use of a new
2368 +minor version number do not always require changes to aufs-util.
2369 +
2370 +Since aufs-util has its own minor version number, you may not be
2371 +able to find a GIT branch in aufs-util for your kernel's
2372 +exact minor version number.
2373 +In this case, you should git-checkout the branch for the
2374 +nearest lower number.
2375 +
2376 +For (an unreleased) example:
2377 +If you are using "linux-3.10" and the "aufs3.10" branch
2378 +does not exist in aufs-util repository, then "aufs3.9", "aufs3.8"
2379 +or something numerically smaller is the branch for your kernel.
2380 +
2381 +Also you can view all branches by
2382 +       $ git branch -a
2383 +
2384 +
2385 +3. Configuration and Compilation
2386 +----------------------------------------
2387 +Make sure you have git-checkout'ed the correct branch.
2388 +
2389 +For aufs3-linux tree,
2390 +- enable CONFIG_AUFS_FS.
2391 +- set other aufs configurations if necessary.
2392 +
2393 +For aufs3-standalone tree,
2394 +There are several ways to build.
2395 +
2396 +1.
2397 +- apply ./aufs3-kbuild.patch to your kernel source files.
2398 +- apply ./aufs3-base.patch too.
2399 +- apply ./aufs3-mmap.patch too.
2400 +- apply ./aufs3-standalone.patch too, if you have a plan to set
2401 +  CONFIG_AUFS_FS=m. otherwise you don't need ./aufs3-standalone.patch.
2402 +- copy ./{Documentation,fs,include/uapi/linux/aufs_type.h} files to your
2403 +  kernel source tree. Never copy $PWD/include/uapi/linux/Kbuild.
2404 +- enable CONFIG_AUFS_FS, you can select either
2405 +  =m or =y.
2406 +- and build your kernel as usual.
2407 +- install the built kernel.
2408 +  Note: Since linux-3.9, every filesystem module requires an alias
2409 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2410 +  modules.aliases file if you set CONFIG_AUFS_FS=m.
2411 +- install the header files too by "make headers_install" to the
2412 +  directory where you specify. By default, it is $PWD/usr.
2413 +  "make help" shows a brief note for headers_install.
2414 +- and reboot your system.
2415 +
2416 +2.
2417 +- module only (CONFIG_AUFS_FS=m).
2418 +- apply ./aufs3-base.patch to your kernel source files.
2419 +- apply ./aufs3-mmap.patch too.
2420 +- apply ./aufs3-standalone.patch too.
2421 +- build your kernel, don't forget "make headers_install", and reboot.
2422 +- edit ./config.mk and set other aufs configurations if necessary.
2423 +  Note: You should read $PWD/fs/aufs/Kconfig carefully which describes
2424 +  every aufs configurations.
2425 +- build the module by simple "make".
2426 +  Note: Since linux-3.9, every filesystem module requires an alias
2427 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2428 +  modules.aliases file.
2429 +- you can specify ${KDIR} make variable which points to your kernel
2430 +  source tree.
2431 +- install the files
2432 +  + run "make install" to install the aufs module, or copy the built
2433 +    $PWD/aufs.ko to /lib/modules/... and run depmod -a (or reboot simply).
2434 +  + run "make install_headers" (instead of headers_install) to install
2435 +    the modified aufs header file (you can specify DESTDIR which is
2436 +    available in aufs standalone version's Makefile only), or copy
2437 +    $PWD/usr/include/linux/aufs_type.h to /usr/include/linux or wherever
2438 +    you like manually. By default, the target directory is $PWD/usr.
2439 +- no need to apply aufs3-kbuild.patch, nor copying source files to your
2440 +  kernel source tree.
2441 +
2442 +Note: The header file aufs_type.h is necessary to build aufs-util
2443 +      as well as "make headers_install" in the kernel source tree.
2444 +      headers_install is subject to be forgotten, but it is essentially
2445 +      necessary, not only for building aufs-util.
2446 +      You may not meet problems without headers_install in some older
2447 +      version though.
2448 +
2449 +And then,
2450 +- read README in aufs-util, build and install it
2451 +- note that your distribution may contain an obsoleted version of
2452 +  aufs_type.h in /usr/include/linux or something. When you build aufs
2453 +  utilities, make sure that your compiler refers the correct aufs header
2454 +  file which is built by "make headers_install."
2455 +- if you want to use readdir(3) in userspace or pathconf(3) wrapper,
2456 +  then run "make install_ulib" too. And refer to the aufs manual in
2457 +  detail.
2458 +
2459 +There several other patches in aufs3-standalone.git. They are all
2460 +optional. When you meet some problems, they will help you.
2461 +- aufs3-loopback.patch
2462 +  Supports a nested loopback mount in a branch-fs. This patch is
2463 +  unnecessary until aufs produces a message like "you may want to try
2464 +  another patch for loopback file".
2465 +- vfs-ino.patch
2466 +  Modifies a system global kernel internal function get_next_ino() in
2467 +  order to stop assigning 0 for an inode-number. Not directly related to
2468 +  aufs, but recommended generally.
2469 +- tmpfs-idr.patch
2470 +  Keeps the tmpfs inode number as the lowest value. Effective to reduce
2471 +  the size of aufs XINO files for tmpfs branch. Also it prevents the
2472 +  duplication of inode number, which is important for backup tools and
2473 +  other utilities. When you find aufs XINO files for tmpfs branch
2474 +  growing too much, try this patch.
2475 +
2476 +
2477 +4. Usage
2478 +----------------------------------------
2479 +At first, make sure aufs-util are installed, and please read the aufs
2480 +manual, aufs.5 in aufs-util.git tree.
2481 +$ man -l aufs.5
2482 +
2483 +And then,
2484 +$ mkdir /tmp/rw /tmp/aufs
2485 +# mount -t aufs -o br=/tmp/rw:${HOME} none /tmp/aufs
2486 +
2487 +Here is another example. The result is equivalent.
2488 +# mount -t aufs -o br=/tmp/rw=rw:${HOME}=ro none /tmp/aufs
2489 +  Or
2490 +# mount -t aufs -o br:/tmp/rw none /tmp/aufs
2491 +# mount -o remount,append:${HOME} /tmp/aufs
2492 +
2493 +Then, you can see whole tree of your home dir through /tmp/aufs. If
2494 +you modify a file under /tmp/aufs, the one on your home directory is
2495 +not affected, instead the same named file will be newly created under
2496 +/tmp/rw. And all of your modification to a file will be applied to
2497 +the one under /tmp/rw. This is called the file based Copy on Write
2498 +(COW) method.
2499 +Aufs mount options are described in aufs.5.
2500 +If you run chroot or something and make your aufs as a root directory,
2501 +then you need to customize the shutdown script. See the aufs manual in
2502 +detail.
2503 +
2504 +Additionally, there are some sample usages of aufs which are a
2505 +diskless system with network booting, and LiveCD over NFS.
2506 +See sample dir in CVS tree on SourceForge.
2507 +
2508 +
2509 +5. Contact
2510 +----------------------------------------
2511 +When you have any problems or strange behaviour in aufs, please let me
2512 +know with:
2513 +- /proc/mounts (instead of the output of mount(8))
2514 +- /sys/module/aufs/*
2515 +- /sys/fs/aufs/* (if you have them)
2516 +- /debug/aufs/* (if you have them)
2517 +- linux kernel version
2518 +  if your kernel is not plain, for example modified by distributor,
2519 +  the url where i can download its source is necessary too.
2520 +- aufs version which was printed at loading the module or booting the
2521 +  system, instead of the date you downloaded.
2522 +- configuration (define/undefine CONFIG_AUFS_xxx)
2523 +- kernel configuration or /proc/config.gz (if you have it)
2524 +- behaviour which you think to be incorrect
2525 +- actual operation, reproducible one is better
2526 +- mailto: aufs-users at lists.sourceforge.net
2527 +
2528 +Usually, I don't watch the Public Areas(Bugs, Support Requests, Patches,
2529 +and Feature Requests) on SourceForge. Please join and write to
2530 +aufs-users ML.
2531 +
2532 +
2533 +6. Acknowledgements
2534 +----------------------------------------
2535 +Thanks to everyone who have tried and are using aufs, whoever
2536 +have reported a bug or any feedback.
2537 +
2538 +Especially donators:
2539 +Tomas Matejicek(slax.org) made a donation (much more than once).
2540 +       Since Apr 2010, Tomas M (the author of Slax and Linux Live
2541 +       scripts) is making "doubling" donations.
2542 +       Unfortunately I cannot list all of the donators, but I really
2543 +       appreciate.
2544 +       It ends Aug 2010, but the ordinary donation URL is still available.
2545 +       <http://sourceforge.net/donate/index.php?group_id=167503>
2546 +Dai Itasaka made a donation (2007/8).
2547 +Chuck Smith made a donation (2008/4, 10 and 12).
2548 +Henk Schoneveld made a donation (2008/9).
2549 +Chih-Wei Huang, ASUS, CTC donated Eee PC 4G (2008/10).
2550 +Francois Dupoux made a donation (2008/11).
2551 +Bruno Cesar Ribas and Luis Carlos Erpen de Bona, C3SL serves public
2552 +       aufs2 GIT tree (2009/2).
2553 +William Grant made a donation (2009/3).
2554 +Patrick Lane made a donation (2009/4).
2555 +The Mail Archive (mail-archive.com) made donations (2009/5).
2556 +Nippy Networks (Ed Wildgoose) made a donation (2009/7).
2557 +New Dream Network, LLC (www.dreamhost.com) made a donation (2009/11).
2558 +Pavel Pronskiy made a donation (2011/2).
2559 +Iridium and Inmarsat satellite phone retailer (www.mailasail.com), Nippy
2560 +       Networks (Ed Wildgoose) made a donation for hardware (2011/3).
2561 +Max Lekomcev (DOM-TV project) made a donation (2011/7, 12, 2012/3, 6 and
2562 +11).
2563 +Sam Liddicott made a donation (2011/9).
2564 +Era Scarecrow made a donation (2013/4).
2565 +Bor Ratajc made a donation (2013/4).
2566 +Alessandro Gorreta made a donation (2013/4).
2567 +POIRETTE Marc made a donation (2013/4).
2568 +Alessandro Gorreta made a donation (2013/4).
2569 +lauri kasvandik made a donation (2013/5).
2570 +"pemasu from Finland" made a donation (2013/7).
2571 +The Parted Magic Project made a donation (2013/9 and 11).
2572 +Pavel Barta made a donation (2013/10).
2573 +Nikolay Pertsev made a donation (2014/5).
2574 +James B made a donation (2014/7).
2575 +Stefano Di Biase made a donation (2014/8).
2576 +
2577 +Thank you very much.
2578 +Donations are always, including future donations, very important and
2579 +helpful for me to keep on developing aufs.
2580 +
2581 +
2582 +7.
2583 +----------------------------------------
2584 +If you are an experienced user, no explanation is needed. Aufs is
2585 +just a linux filesystem.
2586 +
2587 +
2588 +Enjoy!
2589 +
2590 +# Local variables: ;
2591 +# mode: text;
2592 +# End: ;
2593 diff -urN /usr/share/empty/fs/aufs/aufs.h linux/fs/aufs/aufs.h
2594 --- /usr/share/empty/fs/aufs/aufs.h     1970-01-01 01:00:00.000000000 +0100
2595 +++ linux/fs/aufs/aufs.h        2015-01-25 13:00:38.627713742 +0100
2596 @@ -0,0 +1,59 @@
2597 +/*
2598 + * Copyright (C) 2005-2014 Junjiro R. Okajima
2599 + *
2600 + * This program, aufs is free software; you can redistribute it and/or modify
2601 + * it under the terms of the GNU General Public License as published by
2602 + * the Free Software Foundation; either version 2 of the License, or
2603 + * (at your option) any later version.
2604 + *
2605 + * This program is distributed in the hope that it will be useful,
2606 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2607 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2608 + * GNU General Public License for more details.
2609 + *
2610 + * You should have received a copy of the GNU General Public License
2611 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2612 + */
2613 +
2614 +/*
2615 + * all header files
2616 + */
2617 +
2618 +#ifndef __AUFS_H__
2619 +#define __AUFS_H__
2620 +
2621 +#ifdef __KERNEL__
2622 +
2623 +#define AuStub(type, name, body, ...) \
2624 +       static inline type name(__VA_ARGS__) { body; }
2625 +
2626 +#define AuStubVoid(name, ...) \
2627 +       AuStub(void, name, , __VA_ARGS__)
2628 +#define AuStubInt0(name, ...) \
2629 +       AuStub(int, name, return 0, __VA_ARGS__)
2630 +
2631 +#include "debug.h"
2632 +
2633 +#include "branch.h"
2634 +#include "cpup.h"
2635 +#include "dcsub.h"
2636 +#include "dbgaufs.h"
2637 +#include "dentry.h"
2638 +#include "dir.h"
2639 +#include "dynop.h"
2640 +#include "file.h"
2641 +#include "fstype.h"
2642 +#include "inode.h"
2643 +#include "loop.h"
2644 +#include "module.h"
2645 +#include "opts.h"
2646 +#include "rwsem.h"
2647 +#include "spl.h"
2648 +#include "super.h"
2649 +#include "sysaufs.h"
2650 +#include "vfsub.h"
2651 +#include "whout.h"
2652 +#include "wkq.h"
2653 +
2654 +#endif /* __KERNEL__ */
2655 +#endif /* __AUFS_H__ */
2656 diff -urN /usr/share/empty/fs/aufs/branch.c linux/fs/aufs/branch.c
2657 --- /usr/share/empty/fs/aufs/branch.c   1970-01-01 01:00:00.000000000 +0100
2658 +++ linux/fs/aufs/branch.c      2015-01-25 13:00:38.627713742 +0100
2659 @@ -0,0 +1,1410 @@
2660 +/*
2661 + * Copyright (C) 2005-2014 Junjiro R. Okajima
2662 + *
2663 + * This program, aufs is free software; you can redistribute it and/or modify
2664 + * it under the terms of the GNU General Public License as published by
2665 + * the Free Software Foundation; either version 2 of the License, or
2666 + * (at your option) any later version.
2667 + *
2668 + * This program is distributed in the hope that it will be useful,
2669 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2670 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2671 + * GNU General Public License for more details.
2672 + *
2673 + * You should have received a copy of the GNU General Public License
2674 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2675 + */
2676 +
2677 +/*
2678 + * branch management
2679 + */
2680 +
2681 +#include <linux/compat.h>
2682 +#include <linux/statfs.h>
2683 +#include "aufs.h"
2684 +
2685 +/*
2686 + * free a single branch
2687 + */
2688 +static void au_br_do_free(struct au_branch *br)
2689 +{
2690 +       int i;
2691 +       struct au_wbr *wbr;
2692 +       struct au_dykey **key;
2693 +
2694 +       au_hnotify_fin_br(br);
2695 +
2696 +       if (br->br_xino.xi_file)
2697 +               fput(br->br_xino.xi_file);
2698 +       mutex_destroy(&br->br_xino.xi_nondir_mtx);
2699 +
2700 +       AuDebugOn(atomic_read(&br->br_count));
2701 +
2702 +       wbr = br->br_wbr;
2703 +       if (wbr) {
2704 +               for (i = 0; i < AuBrWh_Last; i++)
2705 +                       dput(wbr->wbr_wh[i]);
2706 +               AuDebugOn(atomic_read(&wbr->wbr_wh_running));
2707 +               AuRwDestroy(&wbr->wbr_wh_rwsem);
2708 +       }
2709 +
2710 +       if (br->br_fhsm) {
2711 +               au_br_fhsm_fin(br->br_fhsm);
2712 +               kfree(br->br_fhsm);
2713 +       }
2714 +
2715 +       key = br->br_dykey;
2716 +       for (i = 0; i < AuBrDynOp; i++, key++)
2717 +               if (*key)
2718 +                       au_dy_put(*key);
2719 +               else
2720 +                       break;
2721 +
2722 +       /* recursive lock, s_umount of branch's */
2723 +       lockdep_off();
2724 +       path_put(&br->br_path);
2725 +       lockdep_on();
2726 +       kfree(wbr);
2727 +       kfree(br);
2728 +}
2729 +
2730 +/*
2731 + * frees all branches
2732 + */
2733 +void au_br_free(struct au_sbinfo *sbinfo)
2734 +{
2735 +       aufs_bindex_t bmax;
2736 +       struct au_branch **br;
2737 +
2738 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
2739 +
2740 +       bmax = sbinfo->si_bend + 1;
2741 +       br = sbinfo->si_branch;
2742 +       while (bmax--)
2743 +               au_br_do_free(*br++);
2744 +}
2745 +
2746 +/*
2747 + * find the index of a branch which is specified by @br_id.
2748 + */
2749 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id)
2750 +{
2751 +       aufs_bindex_t bindex, bend;
2752 +
2753 +       bend = au_sbend(sb);
2754 +       for (bindex = 0; bindex <= bend; bindex++)
2755 +               if (au_sbr_id(sb, bindex) == br_id)
2756 +                       return bindex;
2757 +       return -1;
2758 +}
2759 +
2760 +/* ---------------------------------------------------------------------- */
2761 +
2762 +/*
2763 + * add a branch
2764 + */
2765 +
2766 +static int test_overlap(struct super_block *sb, struct dentry *h_adding,
2767 +                       struct dentry *h_root)
2768 +{
2769 +       if (unlikely(h_adding == h_root
2770 +                    || au_test_loopback_overlap(sb, h_adding)))
2771 +               return 1;
2772 +       if (h_adding->d_sb != h_root->d_sb)
2773 +               return 0;
2774 +       return au_test_subdir(h_adding, h_root)
2775 +               || au_test_subdir(h_root, h_adding);
2776 +}
2777 +
2778 +/*
2779 + * returns a newly allocated branch. @new_nbranch is a number of branches
2780 + * after adding a branch.
2781 + */
2782 +static struct au_branch *au_br_alloc(struct super_block *sb, int new_nbranch,
2783 +                                    int perm)
2784 +{
2785 +       struct au_branch *add_branch;
2786 +       struct dentry *root;
2787 +       int err;
2788 +
2789 +       err = -ENOMEM;
2790 +       root = sb->s_root;
2791 +       add_branch = kmalloc(sizeof(*add_branch), GFP_NOFS);
2792 +       if (unlikely(!add_branch))
2793 +               goto out;
2794 +
2795 +       err = au_hnotify_init_br(add_branch, perm);
2796 +       if (unlikely(err))
2797 +               goto out_br;
2798 +
2799 +       add_branch->br_wbr = NULL;
2800 +       if (au_br_writable(perm)) {
2801 +               /* may be freed separately at changing the branch permission */
2802 +               add_branch->br_wbr = kmalloc(sizeof(*add_branch->br_wbr),
2803 +                                            GFP_NOFS);
2804 +               if (unlikely(!add_branch->br_wbr))
2805 +                       goto out_hnotify;
2806 +       }
2807 +
2808 +       add_branch->br_fhsm = NULL;
2809 +       if (au_br_fhsm(perm)) {
2810 +               err = au_fhsm_br_alloc(add_branch);
2811 +               if (unlikely(err))
2812 +                       goto out_wbr;
2813 +       }
2814 +
2815 +       err = au_sbr_realloc(au_sbi(sb), new_nbranch);
2816 +       if (!err)
2817 +               err = au_di_realloc(au_di(root), new_nbranch);
2818 +       if (!err)
2819 +               err = au_ii_realloc(au_ii(root->d_inode), new_nbranch);
2820 +       if (!err)
2821 +               return add_branch; /* success */
2822 +
2823 +out_wbr:
2824 +       kfree(add_branch->br_wbr);
2825 +out_hnotify:
2826 +       au_hnotify_fin_br(add_branch);
2827 +out_br:
2828 +       kfree(add_branch);
2829 +out:
2830 +       return ERR_PTR(err);
2831 +}
2832 +
2833 +/*
2834 + * test if the branch permission is legal or not.
2835 + */
2836 +static int test_br(struct inode *inode, int brperm, char *path)
2837 +{
2838 +       int err;
2839 +
2840 +       err = (au_br_writable(brperm) && IS_RDONLY(inode));
2841 +       if (!err)
2842 +               goto out;
2843 +
2844 +       err = -EINVAL;
2845 +       pr_err("write permission for readonly mount or inode, %s\n", path);
2846 +
2847 +out:
2848 +       return err;
2849 +}
2850 +
2851 +/*
2852 + * returns:
2853 + * 0: success, the caller will add it
2854 + * plus: success, it is already unified, the caller should ignore it
2855 + * minus: error
2856 + */
2857 +static int test_add(struct super_block *sb, struct au_opt_add *add, int remount)
2858 +{
2859 +       int err;
2860 +       aufs_bindex_t bend, bindex;
2861 +       struct dentry *root;
2862 +       struct inode *inode, *h_inode;
2863 +
2864 +       root = sb->s_root;
2865 +       bend = au_sbend(sb);
2866 +       if (unlikely(bend >= 0
2867 +                    && au_find_dbindex(root, add->path.dentry) >= 0)) {
2868 +               err = 1;
2869 +               if (!remount) {
2870 +                       err = -EINVAL;
2871 +                       pr_err("%s duplicated\n", add->pathname);
2872 +               }
2873 +               goto out;
2874 +       }
2875 +
2876 +       err = -ENOSPC; /* -E2BIG; */
2877 +       if (unlikely(AUFS_BRANCH_MAX <= add->bindex
2878 +                    || AUFS_BRANCH_MAX - 1 <= bend)) {
2879 +               pr_err("number of branches exceeded %s\n", add->pathname);
2880 +               goto out;
2881 +       }
2882 +
2883 +       err = -EDOM;
2884 +       if (unlikely(add->bindex < 0 || bend + 1 < add->bindex)) {
2885 +               pr_err("bad index %d\n", add->bindex);
2886 +               goto out;
2887 +       }
2888 +
2889 +       inode = add->path.dentry->d_inode;
2890 +       err = -ENOENT;
2891 +       if (unlikely(!inode->i_nlink)) {
2892 +               pr_err("no existence %s\n", add->pathname);
2893 +               goto out;
2894 +       }
2895 +
2896 +       err = -EINVAL;
2897 +       if (unlikely(inode->i_sb == sb)) {
2898 +               pr_err("%s must be outside\n", add->pathname);
2899 +               goto out;
2900 +       }
2901 +
2902 +       if (unlikely(au_test_fs_unsuppoted(inode->i_sb))) {
2903 +               pr_err("unsupported filesystem, %s (%s)\n",
2904 +                      add->pathname, au_sbtype(inode->i_sb));
2905 +               goto out;
2906 +       }
2907 +
2908 +       if (unlikely(inode->i_sb->s_stack_depth)) {
2909 +               pr_err("already stacked, %s (%s)\n",
2910 +                      add->pathname, au_sbtype(inode->i_sb));
2911 +               goto out;
2912 +       }
2913 +
2914 +       err = test_br(add->path.dentry->d_inode, add->perm, add->pathname);
2915 +       if (unlikely(err))
2916 +               goto out;
2917 +
2918 +       if (bend < 0)
2919 +               return 0; /* success */
2920 +
2921 +       err = -EINVAL;
2922 +       for (bindex = 0; bindex <= bend; bindex++)
2923 +               if (unlikely(test_overlap(sb, add->path.dentry,
2924 +                                         au_h_dptr(root, bindex)))) {
2925 +                       pr_err("%s is overlapped\n", add->pathname);
2926 +                       goto out;
2927 +               }
2928 +
2929 +       err = 0;
2930 +       if (au_opt_test(au_mntflags(sb), WARN_PERM)) {
2931 +               h_inode = au_h_dptr(root, 0)->d_inode;
2932 +               if ((h_inode->i_mode & S_IALLUGO) != (inode->i_mode & S_IALLUGO)
2933 +                   || !uid_eq(h_inode->i_uid, inode->i_uid)
2934 +                   || !gid_eq(h_inode->i_gid, inode->i_gid))
2935 +                       pr_warn("uid/gid/perm %s %u/%u/0%o, %u/%u/0%o\n",
2936 +                               add->pathname,
2937 +                               i_uid_read(inode), i_gid_read(inode),
2938 +                               (inode->i_mode & S_IALLUGO),
2939 +                               i_uid_read(h_inode), i_gid_read(h_inode),
2940 +                               (h_inode->i_mode & S_IALLUGO));
2941 +       }
2942 +
2943 +out:
2944 +       return err;
2945 +}
2946 +
2947 +/*
2948 + * initialize or clean the whiteouts for an adding branch
2949 + */
2950 +static int au_br_init_wh(struct super_block *sb, struct au_branch *br,
2951 +                        int new_perm)
2952 +{
2953 +       int err, old_perm;
2954 +       aufs_bindex_t bindex;
2955 +       struct mutex *h_mtx;
2956 +       struct au_wbr *wbr;
2957 +       struct au_hinode *hdir;
2958 +
2959 +       err = vfsub_mnt_want_write(au_br_mnt(br));
2960 +       if (unlikely(err))
2961 +               goto out;
2962 +
2963 +       wbr = br->br_wbr;
2964 +       old_perm = br->br_perm;
2965 +       br->br_perm = new_perm;
2966 +       hdir = NULL;
2967 +       h_mtx = NULL;
2968 +       bindex = au_br_index(sb, br->br_id);
2969 +       if (0 <= bindex) {
2970 +               hdir = au_hi(sb->s_root->d_inode, bindex);
2971 +               au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
2972 +       } else {
2973 +               h_mtx = &au_br_dentry(br)->d_inode->i_mutex;
2974 +               mutex_lock_nested(h_mtx, AuLsc_I_PARENT);
2975 +       }
2976 +       if (!wbr)
2977 +               err = au_wh_init(br, sb);
2978 +       else {
2979 +               wbr_wh_write_lock(wbr);
2980 +               err = au_wh_init(br, sb);
2981 +               wbr_wh_write_unlock(wbr);
2982 +       }
2983 +       if (hdir)
2984 +               au_hn_imtx_unlock(hdir);
2985 +       else
2986 +               mutex_unlock(h_mtx);
2987 +       vfsub_mnt_drop_write(au_br_mnt(br));
2988 +       br->br_perm = old_perm;
2989 +
2990 +       if (!err && wbr && !au_br_writable(new_perm)) {
2991 +               kfree(wbr);
2992 +               br->br_wbr = NULL;
2993 +       }
2994 +
2995 +out:
2996 +       return err;
2997 +}
2998 +
2999 +static int au_wbr_init(struct au_branch *br, struct super_block *sb,
3000 +                      int perm)
3001 +{
3002 +       int err;
3003 +       struct kstatfs kst;
3004 +       struct au_wbr *wbr;
3005 +
3006 +       wbr = br->br_wbr;
3007 +       au_rw_init(&wbr->wbr_wh_rwsem);
3008 +       memset(wbr->wbr_wh, 0, sizeof(wbr->wbr_wh));
3009 +       atomic_set(&wbr->wbr_wh_running, 0);
3010 +       wbr->wbr_bytes = 0;
3011 +
3012 +       /*
3013 +        * a limit for rmdir/rename a dir
3014 +        * cf. AUFS_MAX_NAMELEN in include/uapi/linux/aufs_type.h
3015 +        */
3016 +       err = vfs_statfs(&br->br_path, &kst);
3017 +       if (unlikely(err))
3018 +               goto out;
3019 +       err = -EINVAL;
3020 +       if (kst.f_namelen >= NAME_MAX)
3021 +               err = au_br_init_wh(sb, br, perm);
3022 +       else
3023 +               pr_err("%pd(%s), unsupported namelen %ld\n",
3024 +                      au_br_dentry(br),
3025 +                      au_sbtype(au_br_dentry(br)->d_sb), kst.f_namelen);
3026 +
3027 +out:
3028 +       return err;
3029 +}
3030 +
3031 +/* initialize a new branch */
3032 +static int au_br_init(struct au_branch *br, struct super_block *sb,
3033 +                     struct au_opt_add *add)
3034 +{
3035 +       int err;
3036 +
3037 +       err = 0;
3038 +       memset(&br->br_xino, 0, sizeof(br->br_xino));
3039 +       mutex_init(&br->br_xino.xi_nondir_mtx);
3040 +       br->br_perm = add->perm;
3041 +       br->br_path = add->path; /* set first, path_get() later */
3042 +       spin_lock_init(&br->br_dykey_lock);
3043 +       memset(br->br_dykey, 0, sizeof(br->br_dykey));
3044 +       atomic_set(&br->br_count, 0);
3045 +       atomic_set(&br->br_xino_running, 0);
3046 +       br->br_id = au_new_br_id(sb);
3047 +       AuDebugOn(br->br_id < 0);
3048 +
3049 +       if (au_br_writable(add->perm)) {
3050 +               err = au_wbr_init(br, sb, add->perm);
3051 +               if (unlikely(err))
3052 +                       goto out_err;
3053 +       }
3054 +
3055 +       if (au_opt_test(au_mntflags(sb), XINO)) {
3056 +               err = au_xino_br(sb, br, add->path.dentry->d_inode->i_ino,
3057 +                                au_sbr(sb, 0)->br_xino.xi_file, /*do_test*/1);
3058 +               if (unlikely(err)) {
3059 +                       AuDebugOn(br->br_xino.xi_file);
3060 +                       goto out_err;
3061 +               }
3062 +       }
3063 +
3064 +       sysaufs_br_init(br);
3065 +       path_get(&br->br_path);
3066 +       goto out; /* success */
3067 +
3068 +out_err:
3069 +       memset(&br->br_path, 0, sizeof(br->br_path));
3070 +out:
3071 +       return err;
3072 +}
3073 +
3074 +static void au_br_do_add_brp(struct au_sbinfo *sbinfo, aufs_bindex_t bindex,
3075 +                            struct au_branch *br, aufs_bindex_t bend,
3076 +                            aufs_bindex_t amount)
3077 +{
3078 +       struct au_branch **brp;
3079 +
3080 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3081 +
3082 +       brp = sbinfo->si_branch + bindex;
3083 +       memmove(brp + 1, brp, sizeof(*brp) * amount);
3084 +       *brp = br;
3085 +       sbinfo->si_bend++;
3086 +       if (unlikely(bend < 0))
3087 +               sbinfo->si_bend = 0;
3088 +}
3089 +
3090 +static void au_br_do_add_hdp(struct au_dinfo *dinfo, aufs_bindex_t bindex,
3091 +                            aufs_bindex_t bend, aufs_bindex_t amount)
3092 +{
3093 +       struct au_hdentry *hdp;
3094 +
3095 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3096 +
3097 +       hdp = dinfo->di_hdentry + bindex;
3098 +       memmove(hdp + 1, hdp, sizeof(*hdp) * amount);
3099 +       au_h_dentry_init(hdp);
3100 +       dinfo->di_bend++;
3101 +       if (unlikely(bend < 0))
3102 +               dinfo->di_bstart = 0;
3103 +}
3104 +
3105 +static void au_br_do_add_hip(struct au_iinfo *iinfo, aufs_bindex_t bindex,
3106 +                            aufs_bindex_t bend, aufs_bindex_t amount)
3107 +{
3108 +       struct au_hinode *hip;
3109 +
3110 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3111 +
3112 +       hip = iinfo->ii_hinode + bindex;
3113 +       memmove(hip + 1, hip, sizeof(*hip) * amount);
3114 +       hip->hi_inode = NULL;
3115 +       au_hn_init(hip);
3116 +       iinfo->ii_bend++;
3117 +       if (unlikely(bend < 0))
3118 +               iinfo->ii_bstart = 0;
3119 +}
3120 +
3121 +static void au_br_do_add(struct super_block *sb, struct au_branch *br,
3122 +                        aufs_bindex_t bindex)
3123 +{
3124 +       struct dentry *root, *h_dentry;
3125 +       struct inode *root_inode;
3126 +       aufs_bindex_t bend, amount;
3127 +
3128 +       root = sb->s_root;
3129 +       root_inode = root->d_inode;
3130 +       bend = au_sbend(sb);
3131 +       amount = bend + 1 - bindex;
3132 +       h_dentry = au_br_dentry(br);
3133 +       au_sbilist_lock();
3134 +       au_br_do_add_brp(au_sbi(sb), bindex, br, bend, amount);
3135 +       au_br_do_add_hdp(au_di(root), bindex, bend, amount);
3136 +       au_br_do_add_hip(au_ii(root_inode), bindex, bend, amount);
3137 +       au_set_h_dptr(root, bindex, dget(h_dentry));
3138 +       au_set_h_iptr(root_inode, bindex, au_igrab(h_dentry->d_inode),
3139 +                     /*flags*/0);
3140 +       au_sbilist_unlock();
3141 +}
3142 +
3143 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount)
3144 +{
3145 +       int err;
3146 +       aufs_bindex_t bend, add_bindex;
3147 +       struct dentry *root, *h_dentry;
3148 +       struct inode *root_inode;
3149 +       struct au_branch *add_branch;
3150 +
3151 +       root = sb->s_root;
3152 +       root_inode = root->d_inode;
3153 +       IMustLock(root_inode);
3154 +       err = test_add(sb, add, remount);
3155 +       if (unlikely(err < 0))
3156 +               goto out;
3157 +       if (err) {
3158 +               err = 0;
3159 +               goto out; /* success */
3160 +       }
3161 +
3162 +       bend = au_sbend(sb);
3163 +       add_branch = au_br_alloc(sb, bend + 2, add->perm);
3164 +       err = PTR_ERR(add_branch);
3165 +       if (IS_ERR(add_branch))
3166 +               goto out;
3167 +
3168 +       err = au_br_init(add_branch, sb, add);
3169 +       if (unlikely(err)) {
3170 +               au_br_do_free(add_branch);
3171 +               goto out;
3172 +       }
3173 +
3174 +       add_bindex = add->bindex;
3175 +       if (!remount)
3176 +               au_br_do_add(sb, add_branch, add_bindex);
3177 +       else {
3178 +               sysaufs_brs_del(sb, add_bindex);
3179 +               au_br_do_add(sb, add_branch, add_bindex);
3180 +               sysaufs_brs_add(sb, add_bindex);
3181 +       }
3182 +
3183 +       h_dentry = add->path.dentry;
3184 +       if (!add_bindex) {
3185 +               au_cpup_attr_all(root_inode, /*force*/1);
3186 +               sb->s_maxbytes = h_dentry->d_sb->s_maxbytes;
3187 +       } else
3188 +               au_add_nlink(root_inode, h_dentry->d_inode);
3189 +
3190 +       /*
3191 +        * this test/set prevents aufs from handling unnecesary notify events
3192 +        * of xino files, in case of re-adding a writable branch which was
3193 +        * once detached from aufs.
3194 +        */
3195 +       if (au_xino_brid(sb) < 0
3196 +           && au_br_writable(add_branch->br_perm)
3197 +           && !au_test_fs_bad_xino(h_dentry->d_sb)
3198 +           && add_branch->br_xino.xi_file
3199 +           && add_branch->br_xino.xi_file->f_dentry->d_parent == h_dentry)
3200 +               au_xino_brid_set(sb, add_branch->br_id);
3201 +
3202 +out:
3203 +       return err;
3204 +}
3205 +
3206 +/* ---------------------------------------------------------------------- */
3207 +
3208 +static unsigned long long au_farray_cb(void *a,
3209 +                                      unsigned long long max __maybe_unused,
3210 +                                      void *arg)
3211 +{
3212 +       unsigned long long n;
3213 +       struct file **p, *f;
3214 +       struct au_sphlhead *files;
3215 +       struct au_finfo *finfo;
3216 +       struct super_block *sb = arg;
3217 +
3218 +       n = 0;
3219 +       p = a;
3220 +       files = &au_sbi(sb)->si_files;
3221 +       spin_lock(&files->spin);
3222 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
3223 +               f = finfo->fi_file;
3224 +               if (file_count(f)
3225 +                   && !special_file(file_inode(f)->i_mode)) {
3226 +                       get_file(f);
3227 +                       *p++ = f;
3228 +                       n++;
3229 +                       AuDebugOn(n > max);
3230 +               }
3231 +       }
3232 +       spin_unlock(&files->spin);
3233 +
3234 +       return n;
3235 +}
3236 +
3237 +static struct file **au_farray_alloc(struct super_block *sb,
3238 +                                    unsigned long long *max)
3239 +{
3240 +       *max = atomic_long_read(&au_sbi(sb)->si_nfiles);
3241 +       return au_array_alloc(max, au_farray_cb, sb);
3242 +}
3243 +
3244 +static void au_farray_free(struct file **a, unsigned long long max)
3245 +{
3246 +       unsigned long long ull;
3247 +
3248 +       for (ull = 0; ull < max; ull++)
3249 +               if (a[ull])
3250 +                       fput(a[ull]);
3251 +       au_array_free(a);
3252 +}
3253 +
3254 +/* ---------------------------------------------------------------------- */
3255 +
3256 +/*
3257 + * delete a branch
3258 + */
3259 +
3260 +/* to show the line number, do not make it inlined function */
3261 +#define AuVerbose(do_info, fmt, ...) do { \
3262 +       if (do_info) \
3263 +               pr_info(fmt, ##__VA_ARGS__); \
3264 +} while (0)
3265 +
3266 +static int au_test_ibusy(struct inode *inode, aufs_bindex_t bstart,
3267 +                        aufs_bindex_t bend)
3268 +{
3269 +       return (inode && !S_ISDIR(inode->i_mode)) || bstart == bend;
3270 +}
3271 +
3272 +static int au_test_dbusy(struct dentry *dentry, aufs_bindex_t bstart,
3273 +                        aufs_bindex_t bend)
3274 +{
3275 +       return au_test_ibusy(dentry->d_inode, bstart, bend);
3276 +}
3277 +
3278 +/*
3279 + * test if the branch is deletable or not.
3280 + */
3281 +static int test_dentry_busy(struct dentry *root, aufs_bindex_t bindex,
3282 +                           unsigned int sigen, const unsigned int verbose)
3283 +{
3284 +       int err, i, j, ndentry;
3285 +       aufs_bindex_t bstart, bend;
3286 +       struct au_dcsub_pages dpages;
3287 +       struct au_dpage *dpage;
3288 +       struct dentry *d;
3289 +
3290 +       err = au_dpages_init(&dpages, GFP_NOFS);
3291 +       if (unlikely(err))
3292 +               goto out;
3293 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
3294 +       if (unlikely(err))
3295 +               goto out_dpages;
3296 +
3297 +       for (i = 0; !err && i < dpages.ndpage; i++) {
3298 +               dpage = dpages.dpages + i;
3299 +               ndentry = dpage->ndentry;
3300 +               for (j = 0; !err && j < ndentry; j++) {
3301 +                       d = dpage->dentries[j];
3302 +                       AuDebugOn(au_dcount(d) <= 0);
3303 +                       if (!au_digen_test(d, sigen)) {
3304 +                               di_read_lock_child(d, AuLock_IR);
3305 +                               if (unlikely(au_dbrange_test(d))) {
3306 +                                       di_read_unlock(d, AuLock_IR);
3307 +                                       continue;
3308 +                               }
3309 +                       } else {
3310 +                               di_write_lock_child(d);
3311 +                               if (unlikely(au_dbrange_test(d))) {
3312 +                                       di_write_unlock(d);
3313 +                                       continue;
3314 +                               }
3315 +                               err = au_reval_dpath(d, sigen);
3316 +                               if (!err)
3317 +                                       di_downgrade_lock(d, AuLock_IR);
3318 +                               else {
3319 +                                       di_write_unlock(d);
3320 +                                       break;
3321 +                               }
3322 +                       }
3323 +
3324 +                       /* AuDbgDentry(d); */
3325 +                       bstart = au_dbstart(d);
3326 +                       bend = au_dbend(d);
3327 +                       if (bstart <= bindex
3328 +                           && bindex <= bend
3329 +                           && au_h_dptr(d, bindex)
3330 +                           && au_test_dbusy(d, bstart, bend)) {
3331 +                               err = -EBUSY;
3332 +                               AuVerbose(verbose, "busy %pd\n", d);
3333 +                               AuDbgDentry(d);
3334 +                       }
3335 +                       di_read_unlock(d, AuLock_IR);
3336 +               }
3337 +       }
3338 +
3339 +out_dpages:
3340 +       au_dpages_free(&dpages);
3341 +out:
3342 +       return err;
3343 +}
3344 +
3345 +static int test_inode_busy(struct super_block *sb, aufs_bindex_t bindex,
3346 +                          unsigned int sigen, const unsigned int verbose)
3347 +{
3348 +       int err;
3349 +       unsigned long long max, ull;
3350 +       struct inode *i, **array;
3351 +       aufs_bindex_t bstart, bend;
3352 +
3353 +       array = au_iarray_alloc(sb, &max);
3354 +       err = PTR_ERR(array);
3355 +       if (IS_ERR(array))
3356 +               goto out;
3357 +
3358 +       err = 0;
3359 +       AuDbg("b%d\n", bindex);
3360 +       for (ull = 0; !err && ull < max; ull++) {
3361 +               i = array[ull];
3362 +               if (unlikely(!i))
3363 +                       break;
3364 +               if (i->i_ino == AUFS_ROOT_INO)
3365 +                       continue;
3366 +
3367 +               /* AuDbgInode(i); */
3368 +               if (au_iigen(i, NULL) == sigen)
3369 +                       ii_read_lock_child(i);
3370 +               else {
3371 +                       ii_write_lock_child(i);
3372 +                       err = au_refresh_hinode_self(i);
3373 +                       au_iigen_dec(i);
3374 +                       if (!err)
3375 +                               ii_downgrade_lock(i);
3376 +                       else {
3377 +                               ii_write_unlock(i);
3378 +                               break;
3379 +                       }
3380 +               }
3381 +
3382 +               bstart = au_ibstart(i);
3383 +               bend = au_ibend(i);
3384 +               if (bstart <= bindex
3385 +                   && bindex <= bend
3386 +                   && au_h_iptr(i, bindex)
3387 +                   && au_test_ibusy(i, bstart, bend)) {
3388 +                       err = -EBUSY;
3389 +                       AuVerbose(verbose, "busy i%lu\n", i->i_ino);
3390 +                       AuDbgInode(i);
3391 +               }
3392 +               ii_read_unlock(i);
3393 +       }
3394 +       au_iarray_free(array, max);
3395 +
3396 +out:
3397 +       return err;
3398 +}
3399 +
3400 +static int test_children_busy(struct dentry *root, aufs_bindex_t bindex,
3401 +                             const unsigned int verbose)
3402 +{
3403 +       int err;
3404 +       unsigned int sigen;
3405 +
3406 +       sigen = au_sigen(root->d_sb);
3407 +       DiMustNoWaiters(root);
3408 +       IiMustNoWaiters(root->d_inode);
3409 +       di_write_unlock(root);
3410 +       err = test_dentry_busy(root, bindex, sigen, verbose);
3411 +       if (!err)
3412 +               err = test_inode_busy(root->d_sb, bindex, sigen, verbose);
3413 +       di_write_lock_child(root); /* aufs_write_lock() calls ..._child() */
3414 +
3415 +       return err;
3416 +}
3417 +
3418 +static int test_dir_busy(struct file *file, aufs_bindex_t br_id,
3419 +                        struct file **to_free, int *idx)
3420 +{
3421 +       int err;
3422 +       unsigned char matched, root;
3423 +       aufs_bindex_t bindex, bend;
3424 +       struct au_fidir *fidir;
3425 +       struct au_hfile *hfile;
3426 +
3427 +       err = 0;
3428 +       root = IS_ROOT(file->f_dentry);
3429 +       if (root) {
3430 +               get_file(file);
3431 +               to_free[*idx] = file;
3432 +               (*idx)++;
3433 +               goto out;
3434 +       }
3435 +
3436 +       matched = 0;
3437 +       fidir = au_fi(file)->fi_hdir;
3438 +       AuDebugOn(!fidir);
3439 +       bend = au_fbend_dir(file);
3440 +       for (bindex = au_fbstart(file); bindex <= bend; bindex++) {
3441 +               hfile = fidir->fd_hfile + bindex;
3442 +               if (!hfile->hf_file)
3443 +                       continue;
3444 +
3445 +               if (hfile->hf_br->br_id == br_id) {
3446 +                       matched = 1;
3447 +                       break;
3448 +               }
3449 +       }
3450 +       if (matched)
3451 +               err = -EBUSY;
3452 +
3453 +out:
3454 +       return err;
3455 +}
3456 +
3457 +static int test_file_busy(struct super_block *sb, aufs_bindex_t br_id,
3458 +                         struct file **to_free, int opened)
3459 +{
3460 +       int err, idx;
3461 +       unsigned long long ull, max;
3462 +       aufs_bindex_t bstart;
3463 +       struct file *file, **array;
3464 +       struct inode *inode;
3465 +       struct dentry *root;
3466 +       struct au_hfile *hfile;
3467 +
3468 +       array = au_farray_alloc(sb, &max);
3469 +       err = PTR_ERR(array);
3470 +       if (IS_ERR(array))
3471 +               goto out;
3472 +
3473 +       err = 0;
3474 +       idx = 0;
3475 +       root = sb->s_root;
3476 +       di_write_unlock(root);
3477 +       for (ull = 0; ull < max; ull++) {
3478 +               file = array[ull];
3479 +               if (unlikely(!file))
3480 +                       break;
3481 +
3482 +               /* AuDbg("%pD\n", file); */
3483 +               fi_read_lock(file);
3484 +               bstart = au_fbstart(file);
3485 +               inode = file_inode(file);
3486 +               if (!S_ISDIR(inode->i_mode)) {
3487 +                       hfile = &au_fi(file)->fi_htop;
3488 +                       if (hfile->hf_br->br_id == br_id)
3489 +                               err = -EBUSY;
3490 +               } else
3491 +                       err = test_dir_busy(file, br_id, to_free, &idx);
3492 +               fi_read_unlock(file);
3493 +               if (unlikely(err))
3494 +                       break;
3495 +       }
3496 +       di_write_lock_child(root);
3497 +       au_farray_free(array, max);
3498 +       AuDebugOn(idx > opened);
3499 +
3500 +out:
3501 +       return err;
3502 +}
3503 +
3504 +static void br_del_file(struct file **to_free, unsigned long long opened,
3505 +                         aufs_bindex_t br_id)
3506 +{
3507 +       unsigned long long ull;
3508 +       aufs_bindex_t bindex, bstart, bend, bfound;
3509 +       struct file *file;
3510 +       struct au_fidir *fidir;
3511 +       struct au_hfile *hfile;
3512 +
3513 +       for (ull = 0; ull < opened; ull++) {
3514 +               file = to_free[ull];
3515 +               if (unlikely(!file))
3516 +                       break;
3517 +
3518 +               /* AuDbg("%pD\n", file); */
3519 +               AuDebugOn(!S_ISDIR(file_inode(file)->i_mode));
3520 +               bfound = -1;
3521 +               fidir = au_fi(file)->fi_hdir;
3522 +               AuDebugOn(!fidir);
3523 +               fi_write_lock(file);
3524 +               bstart = au_fbstart(file);
3525 +               bend = au_fbend_dir(file);
3526 +               for (bindex = bstart; bindex <= bend; bindex++) {
3527 +                       hfile = fidir->fd_hfile + bindex;
3528 +                       if (!hfile->hf_file)
3529 +                               continue;
3530 +
3531 +                       if (hfile->hf_br->br_id == br_id) {
3532 +                               bfound = bindex;
3533 +                               break;
3534 +                       }
3535 +               }
3536 +               AuDebugOn(bfound < 0);
3537 +               au_set_h_fptr(file, bfound, NULL);
3538 +               if (bfound == bstart) {
3539 +                       for (bstart++; bstart <= bend; bstart++)
3540 +                               if (au_hf_dir(file, bstart)) {
3541 +                                       au_set_fbstart(file, bstart);
3542 +                                       break;
3543 +                               }
3544 +               }
3545 +               fi_write_unlock(file);
3546 +       }
3547 +}
3548 +
3549 +static void au_br_do_del_brp(struct au_sbinfo *sbinfo,
3550 +                            const aufs_bindex_t bindex,
3551 +                            const aufs_bindex_t bend)
3552 +{
3553 +       struct au_branch **brp, **p;
3554 +
3555 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3556 +
3557 +       brp = sbinfo->si_branch + bindex;
3558 +       if (bindex < bend)
3559 +               memmove(brp, brp + 1, sizeof(*brp) * (bend - bindex));
3560 +       sbinfo->si_branch[0 + bend] = NULL;
3561 +       sbinfo->si_bend--;
3562 +
3563 +       p = krealloc(sbinfo->si_branch, sizeof(*p) * bend, AuGFP_SBILIST);
3564 +       if (p)
3565 +               sbinfo->si_branch = p;
3566 +       /* harmless error */
3567 +}
3568 +
3569 +static void au_br_do_del_hdp(struct au_dinfo *dinfo, const aufs_bindex_t bindex,
3570 +                            const aufs_bindex_t bend)
3571 +{
3572 +       struct au_hdentry *hdp, *p;
3573 +
3574 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3575 +
3576 +       hdp = dinfo->di_hdentry;
3577 +       if (bindex < bend)
3578 +               memmove(hdp + bindex, hdp + bindex + 1,
3579 +                       sizeof(*hdp) * (bend - bindex));
3580 +       hdp[0 + bend].hd_dentry = NULL;
3581 +       dinfo->di_bend--;
3582 +
3583 +       p = krealloc(hdp, sizeof(*p) * bend, AuGFP_SBILIST);
3584 +       if (p)
3585 +               dinfo->di_hdentry = p;
3586 +       /* harmless error */
3587 +}
3588 +
3589 +static void au_br_do_del_hip(struct au_iinfo *iinfo, const aufs_bindex_t bindex,
3590 +                            const aufs_bindex_t bend)
3591 +{
3592 +       struct au_hinode *hip, *p;
3593 +
3594 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3595 +
3596 +       hip = iinfo->ii_hinode + bindex;
3597 +       if (bindex < bend)
3598 +               memmove(hip, hip + 1, sizeof(*hip) * (bend - bindex));
3599 +       iinfo->ii_hinode[0 + bend].hi_inode = NULL;
3600 +       au_hn_init(iinfo->ii_hinode + bend);
3601 +       iinfo->ii_bend--;
3602 +
3603 +       p = krealloc(iinfo->ii_hinode, sizeof(*p) * bend, AuGFP_SBILIST);
3604 +       if (p)
3605 +               iinfo->ii_hinode = p;
3606 +       /* harmless error */
3607 +}
3608 +
3609 +static void au_br_do_del(struct super_block *sb, aufs_bindex_t bindex,
3610 +                        struct au_branch *br)
3611 +{
3612 +       aufs_bindex_t bend;
3613 +       struct au_sbinfo *sbinfo;
3614 +       struct dentry *root, *h_root;
3615 +       struct inode *inode, *h_inode;
3616 +       struct au_hinode *hinode;
3617 +
3618 +       SiMustWriteLock(sb);
3619 +
3620 +       root = sb->s_root;
3621 +       inode = root->d_inode;
3622 +       sbinfo = au_sbi(sb);
3623 +       bend = sbinfo->si_bend;
3624 +
3625 +       h_root = au_h_dptr(root, bindex);
3626 +       hinode = au_hi(inode, bindex);
3627 +       h_inode = au_igrab(hinode->hi_inode);
3628 +       au_hiput(hinode);
3629 +
3630 +       au_sbilist_lock();
3631 +       au_br_do_del_brp(sbinfo, bindex, bend);
3632 +       au_br_do_del_hdp(au_di(root), bindex, bend);
3633 +       au_br_do_del_hip(au_ii(inode), bindex, bend);
3634 +       au_sbilist_unlock();
3635 +
3636 +       dput(h_root);
3637 +       iput(h_inode);
3638 +       au_br_do_free(br);
3639 +}
3640 +
3641 +static unsigned long long empty_cb(void *array, unsigned long long max,
3642 +                                  void *arg)
3643 +{
3644 +       return max;
3645 +}
3646 +
3647 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount)
3648 +{
3649 +       int err, rerr, i;
3650 +       unsigned long long opened;
3651 +       unsigned int mnt_flags;
3652 +       aufs_bindex_t bindex, bend, br_id;
3653 +       unsigned char do_wh, verbose;
3654 +       struct au_branch *br;
3655 +       struct au_wbr *wbr;
3656 +       struct dentry *root;
3657 +       struct file **to_free;
3658 +
3659 +       err = 0;
3660 +       opened = 0;
3661 +       to_free = NULL;
3662 +       root = sb->s_root;
3663 +       bindex = au_find_dbindex(root, del->h_path.dentry);
3664 +       if (bindex < 0) {
3665 +               if (remount)
3666 +                       goto out; /* success */
3667 +               err = -ENOENT;
3668 +               pr_err("%s no such branch\n", del->pathname);
3669 +               goto out;
3670 +       }
3671 +       AuDbg("bindex b%d\n", bindex);
3672 +
3673 +       err = -EBUSY;
3674 +       mnt_flags = au_mntflags(sb);
3675 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3676 +       bend = au_sbend(sb);
3677 +       if (unlikely(!bend)) {
3678 +               AuVerbose(verbose, "no more branches left\n");
3679 +               goto out;
3680 +       }
3681 +       br = au_sbr(sb, bindex);
3682 +       AuDebugOn(!path_equal(&br->br_path, &del->h_path));
3683 +
3684 +       br_id = br->br_id;
3685 +       opened = atomic_read(&br->br_count);
3686 +       if (unlikely(opened)) {
3687 +               to_free = au_array_alloc(&opened, empty_cb, NULL);
3688 +               err = PTR_ERR(to_free);
3689 +               if (IS_ERR(to_free))
3690 +                       goto out;
3691 +
3692 +               err = test_file_busy(sb, br_id, to_free, opened);
3693 +               if (unlikely(err)) {
3694 +                       AuVerbose(verbose, "%llu file(s) opened\n", opened);
3695 +                       goto out;
3696 +               }
3697 +       }
3698 +
3699 +       wbr = br->br_wbr;
3700 +       do_wh = wbr && (wbr->wbr_whbase || wbr->wbr_plink || wbr->wbr_orph);
3701 +       if (do_wh) {
3702 +               /* instead of WbrWhMustWriteLock(wbr) */
3703 +               SiMustWriteLock(sb);
3704 +               for (i = 0; i < AuBrWh_Last; i++) {
3705 +                       dput(wbr->wbr_wh[i]);
3706 +                       wbr->wbr_wh[i] = NULL;
3707 +               }
3708 +       }
3709 +
3710 +       err = test_children_busy(root, bindex, verbose);
3711 +       if (unlikely(err)) {
3712 +               if (do_wh)
3713 +                       goto out_wh;
3714 +               goto out;
3715 +       }
3716 +
3717 +       err = 0;
3718 +       if (to_free) {
3719 +               /*
3720 +                * now we confirmed the branch is deletable.
3721 +                * let's free the remaining opened dirs on the branch.
3722 +                */
3723 +               di_write_unlock(root);
3724 +               br_del_file(to_free, opened, br_id);
3725 +               di_write_lock_child(root);
3726 +       }
3727 +
3728 +       if (!remount)
3729 +               au_br_do_del(sb, bindex, br);
3730 +       else {
3731 +               sysaufs_brs_del(sb, bindex);
3732 +               au_br_do_del(sb, bindex, br);
3733 +               sysaufs_brs_add(sb, bindex);
3734 +       }
3735 +
3736 +       if (!bindex) {
3737 +               au_cpup_attr_all(root->d_inode, /*force*/1);
3738 +               sb->s_maxbytes = au_sbr_sb(sb, 0)->s_maxbytes;
3739 +       } else
3740 +               au_sub_nlink(root->d_inode, del->h_path.dentry->d_inode);
3741 +       if (au_opt_test(mnt_flags, PLINK))
3742 +               au_plink_half_refresh(sb, br_id);
3743 +
3744 +       if (au_xino_brid(sb) == br_id)
3745 +               au_xino_brid_set(sb, -1);
3746 +       goto out; /* success */
3747 +
3748 +out_wh:
3749 +       /* revert */
3750 +       rerr = au_br_init_wh(sb, br, br->br_perm);
3751 +       if (rerr)
3752 +               pr_warn("failed re-creating base whiteout, %s. (%d)\n",
3753 +                       del->pathname, rerr);
3754 +out:
3755 +       if (to_free)
3756 +               au_farray_free(to_free, opened);
3757 +       return err;
3758 +}
3759 +
3760 +/* ---------------------------------------------------------------------- */
3761 +
3762 +static int au_ibusy(struct super_block *sb, struct aufs_ibusy __user *arg)
3763 +{
3764 +       int err;
3765 +       aufs_bindex_t bstart, bend;
3766 +       struct aufs_ibusy ibusy;
3767 +       struct inode *inode, *h_inode;
3768 +
3769 +       err = -EPERM;
3770 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
3771 +               goto out;
3772 +
3773 +       err = copy_from_user(&ibusy, arg, sizeof(ibusy));
3774 +       if (!err)
3775 +               err = !access_ok(VERIFY_WRITE, &arg->h_ino, sizeof(arg->h_ino));
3776 +       if (unlikely(err)) {
3777 +               err = -EFAULT;
3778 +               AuTraceErr(err);
3779 +               goto out;
3780 +       }
3781 +
3782 +       err = -EINVAL;
3783 +       si_read_lock(sb, AuLock_FLUSH);
3784 +       if (unlikely(ibusy.bindex < 0 || ibusy.bindex > au_sbend(sb)))
3785 +               goto out_unlock;
3786 +
3787 +       err = 0;
3788 +       ibusy.h_ino = 0; /* invalid */
3789 +       inode = ilookup(sb, ibusy.ino);
3790 +       if (!inode
3791 +           || inode->i_ino == AUFS_ROOT_INO
3792 +           || is_bad_inode(inode))
3793 +               goto out_unlock;
3794 +
3795 +       ii_read_lock_child(inode);
3796 +       bstart = au_ibstart(inode);
3797 +       bend = au_ibend(inode);
3798 +       if (bstart <= ibusy.bindex && ibusy.bindex <= bend) {
3799 +               h_inode = au_h_iptr(inode, ibusy.bindex);
3800 +               if (h_inode && au_test_ibusy(inode, bstart, bend))
3801 +                       ibusy.h_ino = h_inode->i_ino;
3802 +       }
3803 +       ii_read_unlock(inode);
3804 +       iput(inode);
3805 +
3806 +out_unlock:
3807 +       si_read_unlock(sb);
3808 +       if (!err) {
3809 +               err = __put_user(ibusy.h_ino, &arg->h_ino);
3810 +               if (unlikely(err)) {
3811 +                       err = -EFAULT;
3812 +                       AuTraceErr(err);
3813 +               }
3814 +       }
3815 +out:
3816 +       return err;
3817 +}
3818 +
3819 +long au_ibusy_ioctl(struct file *file, unsigned long arg)
3820 +{
3821 +       return au_ibusy(file->f_dentry->d_sb, (void __user *)arg);
3822 +}
3823 +
3824 +#ifdef CONFIG_COMPAT
3825 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg)
3826 +{
3827 +       return au_ibusy(file->f_dentry->d_sb, compat_ptr(arg));
3828 +}
3829 +#endif
3830 +
3831 +/* ---------------------------------------------------------------------- */
3832 +
3833 +/*
3834 + * change a branch permission
3835 + */
3836 +
3837 +static void au_warn_ima(void)
3838 +{
3839 +#ifdef CONFIG_IMA
3840 +       /* since it doesn't support mark_files_ro() */
3841 +       AuWarn1("RW -> RO makes IMA to produce wrong message\n");
3842 +#endif
3843 +}
3844 +
3845 +static int do_need_sigen_inc(int a, int b)
3846 +{
3847 +       return au_br_whable(a) && !au_br_whable(b);
3848 +}
3849 +
3850 +static int need_sigen_inc(int old, int new)
3851 +{
3852 +       return do_need_sigen_inc(old, new)
3853 +               || do_need_sigen_inc(new, old);
3854 +}
3855 +
3856 +static int au_br_mod_files_ro(struct super_block *sb, aufs_bindex_t bindex)
3857 +{
3858 +       int err, do_warn;
3859 +       unsigned int mnt_flags;
3860 +       unsigned long long ull, max;
3861 +       aufs_bindex_t br_id;
3862 +       unsigned char verbose, writer;
3863 +       struct file *file, *hf, **array;
3864 +       struct inode *inode;
3865 +       struct au_hfile *hfile;
3866 +
3867 +       mnt_flags = au_mntflags(sb);
3868 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3869 +
3870 +       array = au_farray_alloc(sb, &max);
3871 +       err = PTR_ERR(array);
3872 +       if (IS_ERR(array))
3873 +               goto out;
3874 +
3875 +       do_warn = 0;
3876 +       br_id = au_sbr_id(sb, bindex);
3877 +       for (ull = 0; ull < max; ull++) {
3878 +               file = array[ull];
3879 +               if (unlikely(!file))
3880 +                       break;
3881 +
3882 +               /* AuDbg("%pD\n", file); */
3883 +               fi_read_lock(file);
3884 +               if (unlikely(au_test_mmapped(file))) {
3885 +                       err = -EBUSY;
3886 +                       AuVerbose(verbose, "mmapped %pD\n", file);
3887 +                       AuDbgFile(file);
3888 +                       FiMustNoWaiters(file);
3889 +                       fi_read_unlock(file);
3890 +                       goto out_array;
3891 +               }
3892 +
3893 +               inode = file_inode(file);
3894 +               hfile = &au_fi(file)->fi_htop;
3895 +               hf = hfile->hf_file;
3896 +               if (!S_ISREG(inode->i_mode)
3897 +                   || !(file->f_mode & FMODE_WRITE)
3898 +                   || hfile->hf_br->br_id != br_id
3899 +                   || !(hf->f_mode & FMODE_WRITE))
3900 +                       array[ull] = NULL;
3901 +               else {
3902 +                       do_warn = 1;
3903 +                       get_file(file);
3904 +               }
3905 +
3906 +               FiMustNoWaiters(file);
3907 +               fi_read_unlock(file);
3908 +               fput(file);
3909 +       }
3910 +
3911 +       err = 0;
3912 +       if (do_warn)
3913 +               au_warn_ima();
3914 +
3915 +       for (ull = 0; ull < max; ull++) {
3916 +               file = array[ull];
3917 +               if (!file)
3918 +                       continue;
3919 +
3920 +               /* todo: already flushed? */
3921 +               /*
3922 +                * fs/super.c:mark_files_ro() is gone, but aufs keeps its
3923 +                * approach which resets f_mode and calls mnt_drop_write() and
3924 +                * file_release_write() for each file, because the branch
3925 +                * attribute in aufs world is totally different from the native
3926 +                * fs rw/ro mode.
3927 +               */
3928 +               /* fi_read_lock(file); */
3929 +               hfile = &au_fi(file)->fi_htop;
3930 +               hf = hfile->hf_file;
3931 +               /* fi_read_unlock(file); */
3932 +               spin_lock(&hf->f_lock);
3933 +               writer = !!(hf->f_mode & FMODE_WRITER);
3934 +               hf->f_mode &= ~(FMODE_WRITE | FMODE_WRITER);
3935 +               spin_unlock(&hf->f_lock);
3936 +               if (writer) {
3937 +                       put_write_access(file_inode(hf));
3938 +                       __mnt_drop_write(hf->f_path.mnt);
3939 +               }
3940 +       }
3941 +
3942 +out_array:
3943 +       au_farray_free(array, max);
3944 +out:
3945 +       AuTraceErr(err);
3946 +       return err;
3947 +}
3948 +
3949 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
3950 +             int *do_refresh)
3951 +{
3952 +       int err, rerr;
3953 +       aufs_bindex_t bindex;
3954 +       struct dentry *root;
3955 +       struct au_branch *br;
3956 +       struct au_br_fhsm *bf;
3957 +
3958 +       root = sb->s_root;
3959 +       bindex = au_find_dbindex(root, mod->h_root);
3960 +       if (bindex < 0) {
3961 +               if (remount)
3962 +                       return 0; /* success */
3963 +               err = -ENOENT;
3964 +               pr_err("%s no such branch\n", mod->path);
3965 +               goto out;
3966 +       }
3967 +       AuDbg("bindex b%d\n", bindex);
3968 +
3969 +       err = test_br(mod->h_root->d_inode, mod->perm, mod->path);
3970 +       if (unlikely(err))
3971 +               goto out;
3972 +
3973 +       br = au_sbr(sb, bindex);
3974 +       AuDebugOn(mod->h_root != au_br_dentry(br));
3975 +       if (br->br_perm == mod->perm)
3976 +               return 0; /* success */
3977 +
3978 +       /* pre-allocate for non-fhsm --> fhsm */
3979 +       bf = NULL;
3980 +       if (!au_br_fhsm(br->br_perm) && au_br_fhsm(mod->perm)) {
3981 +               err = au_fhsm_br_alloc(br);
3982 +               if (unlikely(err))
3983 +                       goto out;
3984 +               bf = br->br_fhsm;
3985 +               br->br_fhsm = NULL;
3986 +       }
3987 +
3988 +       if (au_br_writable(br->br_perm)) {
3989 +               /* remove whiteout base */
3990 +               err = au_br_init_wh(sb, br, mod->perm);
3991 +               if (unlikely(err))
3992 +                       goto out_bf;
3993 +
3994 +               if (!au_br_writable(mod->perm)) {
3995 +                       /* rw --> ro, file might be mmapped */
3996 +                       DiMustNoWaiters(root);
3997 +                       IiMustNoWaiters(root->d_inode);
3998 +                       di_write_unlock(root);
3999 +                       err = au_br_mod_files_ro(sb, bindex);
4000 +                       /* aufs_write_lock() calls ..._child() */
4001 +                       di_write_lock_child(root);
4002 +
4003 +                       if (unlikely(err)) {
4004 +                               rerr = -ENOMEM;
4005 +                               br->br_wbr = kmalloc(sizeof(*br->br_wbr),
4006 +                                                    GFP_NOFS);
4007 +                               if (br->br_wbr)
4008 +                                       rerr = au_wbr_init(br, sb, br->br_perm);
4009 +                               if (unlikely(rerr)) {
4010 +                                       AuIOErr("nested error %d (%d)\n",
4011 +                                               rerr, err);
4012 +                                       br->br_perm = mod->perm;
4013 +                               }
4014 +                       }
4015 +               }
4016 +       } else if (au_br_writable(mod->perm)) {
4017 +               /* ro --> rw */
4018 +               err = -ENOMEM;
4019 +               br->br_wbr = kmalloc(sizeof(*br->br_wbr), GFP_NOFS);
4020 +               if (br->br_wbr) {
4021 +                       err = au_wbr_init(br, sb, mod->perm);
4022 +                       if (unlikely(err)) {
4023 +                               kfree(br->br_wbr);
4024 +                               br->br_wbr = NULL;
4025 +                       }
4026 +               }
4027 +       }
4028 +       if (unlikely(err))
4029 +               goto out_bf;
4030 +
4031 +       if (au_br_fhsm(br->br_perm)) {
4032 +               if (!au_br_fhsm(mod->perm)) {
4033 +                       /* fhsm --> non-fhsm */
4034 +                       au_br_fhsm_fin(br->br_fhsm);
4035 +                       kfree(br->br_fhsm);
4036 +                       br->br_fhsm = NULL;
4037 +               }
4038 +       } else if (au_br_fhsm(mod->perm))
4039 +               /* non-fhsm --> fhsm */
4040 +               br->br_fhsm = bf;
4041 +
4042 +       *do_refresh |= need_sigen_inc(br->br_perm, mod->perm);
4043 +       br->br_perm = mod->perm;
4044 +       goto out; /* success */
4045 +
4046 +out_bf:
4047 +       kfree(bf);
4048 +out:
4049 +       AuTraceErr(err);
4050 +       return err;
4051 +}
4052 +
4053 +/* ---------------------------------------------------------------------- */
4054 +
4055 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs)
4056 +{
4057 +       int err;
4058 +       struct kstatfs kstfs;
4059 +
4060 +       err = vfs_statfs(&br->br_path, &kstfs);
4061 +       if (!err) {
4062 +               stfs->f_blocks = kstfs.f_blocks;
4063 +               stfs->f_bavail = kstfs.f_bavail;
4064 +               stfs->f_files = kstfs.f_files;
4065 +               stfs->f_ffree = kstfs.f_ffree;
4066 +       }
4067 +
4068 +       return err;
4069 +}
4070 diff -urN /usr/share/empty/fs/aufs/branch.h linux/fs/aufs/branch.h
4071 --- /usr/share/empty/fs/aufs/branch.h   1970-01-01 01:00:00.000000000 +0100
4072 +++ linux/fs/aufs/branch.h      2015-01-25 13:00:38.627713742 +0100
4073 @@ -0,0 +1,267 @@
4074 +/*
4075 + * Copyright (C) 2005-2014 Junjiro R. Okajima
4076 + *
4077 + * This program, aufs is free software; you can redistribute it and/or modify
4078 + * it under the terms of the GNU General Public License as published by
4079 + * the Free Software Foundation; either version 2 of the License, or
4080 + * (at your option) any later version.
4081 + *
4082 + * This program is distributed in the hope that it will be useful,
4083 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4084 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4085 + * GNU General Public License for more details.
4086 + *
4087 + * You should have received a copy of the GNU General Public License
4088 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4089 + */
4090 +
4091 +/*
4092 + * branch filesystems and xino for them
4093 + */
4094 +
4095 +#ifndef __AUFS_BRANCH_H__
4096 +#define __AUFS_BRANCH_H__
4097 +
4098 +#ifdef __KERNEL__
4099 +
4100 +#include <linux/mount.h>
4101 +#include "dynop.h"
4102 +#include "rwsem.h"
4103 +#include "super.h"
4104 +
4105 +/* ---------------------------------------------------------------------- */
4106 +
4107 +/* a xino file */
4108 +struct au_xino_file {
4109 +       struct file             *xi_file;
4110 +       struct mutex            xi_nondir_mtx;
4111 +
4112 +       /* todo: make xino files an array to support huge inode number */
4113 +
4114 +#ifdef CONFIG_DEBUG_FS
4115 +       struct dentry            *xi_dbgaufs;
4116 +#endif
4117 +};
4118 +
4119 +/* File-based Hierarchical Storage Management */
4120 +struct au_br_fhsm {
4121 +#ifdef CONFIG_AUFS_FHSM
4122 +       struct mutex            bf_lock;
4123 +       unsigned long           bf_jiffy;
4124 +       struct aufs_stfs        bf_stfs;
4125 +       int                     bf_readable;
4126 +#endif
4127 +};
4128 +
4129 +/* members for writable branch only */
4130 +enum {AuBrWh_BASE, AuBrWh_PLINK, AuBrWh_ORPH, AuBrWh_Last};
4131 +struct au_wbr {
4132 +       struct au_rwsem         wbr_wh_rwsem;
4133 +       struct dentry           *wbr_wh[AuBrWh_Last];
4134 +       atomic_t                wbr_wh_running;
4135 +#define wbr_whbase             wbr_wh[AuBrWh_BASE]     /* whiteout base */
4136 +#define wbr_plink              wbr_wh[AuBrWh_PLINK]    /* pseudo-link dir */
4137 +#define wbr_orph               wbr_wh[AuBrWh_ORPH]     /* dir for orphans */
4138 +
4139 +       /* mfs mode */
4140 +       unsigned long long      wbr_bytes;
4141 +};
4142 +
4143 +/* ext2 has 3 types of operations at least, ext3 has 4 */
4144 +#define AuBrDynOp (AuDyLast * 4)
4145 +
4146 +#ifdef CONFIG_AUFS_HFSNOTIFY
4147 +/* support for asynchronous destruction */
4148 +struct au_br_hfsnotify {
4149 +       struct fsnotify_group   *hfsn_group;
4150 +};
4151 +#endif
4152 +
4153 +/* sysfs entries */
4154 +struct au_brsysfs {
4155 +       char                    name[16];
4156 +       struct attribute        attr;
4157 +};
4158 +
4159 +enum {
4160 +       AuBrSysfs_BR,
4161 +       AuBrSysfs_BRID,
4162 +       AuBrSysfs_Last
4163 +};
4164 +
4165 +/* protected by superblock rwsem */
4166 +struct au_branch {
4167 +       struct au_xino_file     br_xino;
4168 +
4169 +       aufs_bindex_t           br_id;
4170 +
4171 +       int                     br_perm;
4172 +       struct path             br_path;
4173 +       spinlock_t              br_dykey_lock;
4174 +       struct au_dykey         *br_dykey[AuBrDynOp];
4175 +       atomic_t                br_count;
4176 +
4177 +       struct au_wbr           *br_wbr;
4178 +       struct au_br_fhsm       *br_fhsm;
4179 +
4180 +       /* xino truncation */
4181 +       atomic_t                br_xino_running;
4182 +
4183 +#ifdef CONFIG_AUFS_HFSNOTIFY
4184 +       struct au_br_hfsnotify  *br_hfsn;
4185 +#endif
4186 +
4187 +#ifdef CONFIG_SYSFS
4188 +       /* entries under sysfs per mount-point */
4189 +       struct au_brsysfs       br_sysfs[AuBrSysfs_Last];
4190 +#endif
4191 +};
4192 +
4193 +/* ---------------------------------------------------------------------- */
4194 +
4195 +static inline struct vfsmount *au_br_mnt(struct au_branch *br)
4196 +{
4197 +       return br->br_path.mnt;
4198 +}
4199 +
4200 +static inline struct dentry *au_br_dentry(struct au_branch *br)
4201 +{
4202 +       return br->br_path.dentry;
4203 +}
4204 +
4205 +static inline struct super_block *au_br_sb(struct au_branch *br)
4206 +{
4207 +       return au_br_mnt(br)->mnt_sb;
4208 +}
4209 +
4210 +static inline int au_br_rdonly(struct au_branch *br)
4211 +{
4212 +       return ((au_br_sb(br)->s_flags & MS_RDONLY)
4213 +               || !au_br_writable(br->br_perm))
4214 +               ? -EROFS : 0;
4215 +}
4216 +
4217 +static inline int au_br_hnotifyable(int brperm __maybe_unused)
4218 +{
4219 +#ifdef CONFIG_AUFS_HNOTIFY
4220 +       return !(brperm & AuBrPerm_RR);
4221 +#else
4222 +       return 0;
4223 +#endif
4224 +}
4225 +
4226 +/* ---------------------------------------------------------------------- */
4227 +
4228 +/* branch.c */
4229 +struct au_sbinfo;
4230 +void au_br_free(struct au_sbinfo *sinfo);
4231 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id);
4232 +struct au_opt_add;
4233 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount);
4234 +struct au_opt_del;
4235 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount);
4236 +long au_ibusy_ioctl(struct file *file, unsigned long arg);
4237 +#ifdef CONFIG_COMPAT
4238 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg);
4239 +#endif
4240 +struct au_opt_mod;
4241 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4242 +             int *do_refresh);
4243 +struct aufs_stfs;
4244 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs);
4245 +
4246 +/* xino.c */
4247 +static const loff_t au_loff_max = LLONG_MAX;
4248 +
4249 +int au_xib_trunc(struct super_block *sb);
4250 +ssize_t xino_fread(au_readf_t func, struct file *file, void *buf, size_t size,
4251 +                  loff_t *pos);
4252 +ssize_t xino_fwrite(au_writef_t func, struct file *file, void *buf, size_t size,
4253 +                   loff_t *pos);
4254 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src);
4255 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent);
4256 +ino_t au_xino_new_ino(struct super_block *sb);
4257 +void au_xino_delete_inode(struct inode *inode, const int unlinked);
4258 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4259 +                 ino_t ino);
4260 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4261 +                ino_t *ino);
4262 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t hino,
4263 +              struct file *base_file, int do_test);
4264 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex);
4265 +
4266 +struct au_opt_xino;
4267 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount);
4268 +void au_xino_clr(struct super_block *sb);
4269 +struct file *au_xino_def(struct super_block *sb);
4270 +int au_xino_path(struct seq_file *seq, struct file *file);
4271 +
4272 +/* ---------------------------------------------------------------------- */
4273 +
4274 +/* Superblock to branch */
4275 +static inline
4276 +aufs_bindex_t au_sbr_id(struct super_block *sb, aufs_bindex_t bindex)
4277 +{
4278 +       return au_sbr(sb, bindex)->br_id;
4279 +}
4280 +
4281 +static inline
4282 +struct vfsmount *au_sbr_mnt(struct super_block *sb, aufs_bindex_t bindex)
4283 +{
4284 +       return au_br_mnt(au_sbr(sb, bindex));
4285 +}
4286 +
4287 +static inline
4288 +struct super_block *au_sbr_sb(struct super_block *sb, aufs_bindex_t bindex)
4289 +{
4290 +       return au_br_sb(au_sbr(sb, bindex));
4291 +}
4292 +
4293 +static inline void au_sbr_put(struct super_block *sb, aufs_bindex_t bindex)
4294 +{
4295 +       atomic_dec(&au_sbr(sb, bindex)->br_count);
4296 +}
4297 +
4298 +static inline int au_sbr_perm(struct super_block *sb, aufs_bindex_t bindex)
4299 +{
4300 +       return au_sbr(sb, bindex)->br_perm;
4301 +}
4302 +
4303 +static inline int au_sbr_whable(struct super_block *sb, aufs_bindex_t bindex)
4304 +{
4305 +       return au_br_whable(au_sbr_perm(sb, bindex));
4306 +}
4307 +
4308 +/* ---------------------------------------------------------------------- */
4309 +
4310 +/*
4311 + * wbr_wh_read_lock, wbr_wh_write_lock
4312 + * wbr_wh_read_unlock, wbr_wh_write_unlock, wbr_wh_downgrade_lock
4313 + */
4314 +AuSimpleRwsemFuncs(wbr_wh, struct au_wbr *wbr, &wbr->wbr_wh_rwsem);
4315 +
4316 +#define WbrWhMustNoWaiters(wbr)        AuRwMustNoWaiters(&wbr->wbr_wh_rwsem)
4317 +#define WbrWhMustAnyLock(wbr)  AuRwMustAnyLock(&wbr->wbr_wh_rwsem)
4318 +#define WbrWhMustWriteLock(wbr)        AuRwMustWriteLock(&wbr->wbr_wh_rwsem)
4319 +
4320 +/* ---------------------------------------------------------------------- */
4321 +
4322 +#ifdef CONFIG_AUFS_FHSM
4323 +static inline void au_br_fhsm_init(struct au_br_fhsm *brfhsm)
4324 +{
4325 +       mutex_init(&brfhsm->bf_lock);
4326 +       brfhsm->bf_jiffy = 0;
4327 +       brfhsm->bf_readable = 0;
4328 +}
4329 +
4330 +static inline void au_br_fhsm_fin(struct au_br_fhsm *brfhsm)
4331 +{
4332 +       mutex_destroy(&brfhsm->bf_lock);
4333 +}
4334 +#else
4335 +AuStubVoid(au_br_fhsm_init, struct au_br_fhsm *brfhsm)
4336 +AuStubVoid(au_br_fhsm_fin, struct au_br_fhsm *brfhsm)
4337 +#endif
4338 +
4339 +#endif /* __KERNEL__ */
4340 +#endif /* __AUFS_BRANCH_H__ */
4341 diff -urN /usr/share/empty/fs/aufs/conf.mk linux/fs/aufs/conf.mk
4342 --- /usr/share/empty/fs/aufs/conf.mk    1970-01-01 01:00:00.000000000 +0100
4343 +++ linux/fs/aufs/conf.mk       2015-01-25 13:00:38.631047076 +0100
4344 @@ -0,0 +1,38 @@
4345 +
4346 +AuConfStr = CONFIG_AUFS_FS=${CONFIG_AUFS_FS}
4347 +
4348 +define AuConf
4349 +ifdef ${1}
4350 +AuConfStr += ${1}=${${1}}
4351 +endif
4352 +endef
4353 +
4354 +AuConfAll = BRANCH_MAX_127 BRANCH_MAX_511 BRANCH_MAX_1023 BRANCH_MAX_32767 \
4355 +       SBILIST \
4356 +       HNOTIFY HFSNOTIFY \
4357 +       EXPORT INO_T_64 \
4358 +       XATTR \
4359 +       FHSM \
4360 +       RDU \
4361 +       SHWH \
4362 +       BR_RAMFS \
4363 +       BR_FUSE POLL \
4364 +       BR_HFSPLUS \
4365 +       BDEV_LOOP \
4366 +       DEBUG MAGIC_SYSRQ
4367 +$(foreach i, ${AuConfAll}, \
4368 +       $(eval $(call AuConf,CONFIG_AUFS_${i})))
4369 +
4370 +AuConfName = ${obj}/conf.str
4371 +${AuConfName}.tmp: FORCE
4372 +       @echo ${AuConfStr} | tr ' ' '\n' | sed -e 's/^/"/' -e 's/$$/\\n"/' > $@
4373 +${AuConfName}: ${AuConfName}.tmp
4374 +       @diff -q $< $@ > /dev/null 2>&1 || { \
4375 +       echo '  GEN    ' $@; \
4376 +       cp -p $< $@; \
4377 +       }
4378 +FORCE:
4379 +clean-files += ${AuConfName} ${AuConfName}.tmp
4380 +${obj}/sysfs.o: ${AuConfName}
4381 +
4382 +-include ${srctree}/${src}/conf_priv.mk
4383 diff -urN /usr/share/empty/fs/aufs/cpup.c linux/fs/aufs/cpup.c
4384 --- /usr/share/empty/fs/aufs/cpup.c     1970-01-01 01:00:00.000000000 +0100
4385 +++ linux/fs/aufs/cpup.c        2015-01-25 13:00:38.631047076 +0100
4386 @@ -0,0 +1,1303 @@
4387 +/*
4388 + * Copyright (C) 2005-2014 Junjiro R. Okajima
4389 + *
4390 + * This program, aufs is free software; you can redistribute it and/or modify
4391 + * it under the terms of the GNU General Public License as published by
4392 + * the Free Software Foundation; either version 2 of the License, or
4393 + * (at your option) any later version.
4394 + *
4395 + * This program is distributed in the hope that it will be useful,
4396 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4397 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4398 + * GNU General Public License for more details.
4399 + *
4400 + * You should have received a copy of the GNU General Public License
4401 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4402 + */
4403 +
4404 +/*
4405 + * copy-up functions, see wbr_policy.c for copy-down
4406 + */
4407 +
4408 +#include <linux/fs_stack.h>
4409 +#include <linux/mm.h>
4410 +#include "aufs.h"
4411 +
4412 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags)
4413 +{
4414 +       const unsigned int mask = S_DEAD | S_SWAPFILE | S_PRIVATE
4415 +               | S_NOATIME | S_NOCMTIME | S_AUTOMOUNT;
4416 +
4417 +       BUILD_BUG_ON(sizeof(iflags) != sizeof(dst->i_flags));
4418 +
4419 +       dst->i_flags |= iflags & ~mask;
4420 +       if (au_test_fs_notime(dst->i_sb))
4421 +               dst->i_flags |= S_NOATIME | S_NOCMTIME;
4422 +}
4423 +
4424 +void au_cpup_attr_timesizes(struct inode *inode)
4425 +{
4426 +       struct inode *h_inode;
4427 +
4428 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
4429 +       fsstack_copy_attr_times(inode, h_inode);
4430 +       fsstack_copy_inode_size(inode, h_inode);
4431 +}
4432 +
4433 +void au_cpup_attr_nlink(struct inode *inode, int force)
4434 +{
4435 +       struct inode *h_inode;
4436 +       struct super_block *sb;
4437 +       aufs_bindex_t bindex, bend;
4438 +
4439 +       sb = inode->i_sb;
4440 +       bindex = au_ibstart(inode);
4441 +       h_inode = au_h_iptr(inode, bindex);
4442 +       if (!force
4443 +           && !S_ISDIR(h_inode->i_mode)
4444 +           && au_opt_test(au_mntflags(sb), PLINK)
4445 +           && au_plink_test(inode))
4446 +               return;
4447 +
4448 +       /*
4449 +        * 0 can happen in revalidating.
4450 +        * h_inode->i_mutex may not be held here, but it is harmless since once
4451 +        * i_nlink reaches 0, it will never become positive except O_TMPFILE
4452 +        * case.
4453 +        * todo: O_TMPFILE+linkat(AT_SYMLINK_FOLLOW) bypassing aufs may cause
4454 +        *       the incorrect link count.
4455 +        */
4456 +       set_nlink(inode, h_inode->i_nlink);
4457 +
4458 +       /*
4459 +        * fewer nlink makes find(1) noisy, but larger nlink doesn't.
4460 +        * it may includes whplink directory.
4461 +        */
4462 +       if (S_ISDIR(h_inode->i_mode)) {
4463 +               bend = au_ibend(inode);
4464 +               for (bindex++; bindex <= bend; bindex++) {
4465 +                       h_inode = au_h_iptr(inode, bindex);
4466 +                       if (h_inode)
4467 +                               au_add_nlink(inode, h_inode);
4468 +               }
4469 +       }
4470 +}
4471 +
4472 +void au_cpup_attr_changeable(struct inode *inode)
4473 +{
4474 +       struct inode *h_inode;
4475 +
4476 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
4477 +       inode->i_mode = h_inode->i_mode;
4478 +       inode->i_uid = h_inode->i_uid;
4479 +       inode->i_gid = h_inode->i_gid;
4480 +       au_cpup_attr_timesizes(inode);
4481 +       au_cpup_attr_flags(inode, h_inode->i_flags);
4482 +}
4483 +
4484 +void au_cpup_igen(struct inode *inode, struct inode *h_inode)
4485 +{
4486 +       struct au_iinfo *iinfo = au_ii(inode);
4487 +
4488 +       IiMustWriteLock(inode);
4489 +
4490 +       iinfo->ii_higen = h_inode->i_generation;
4491 +       iinfo->ii_hsb1 = h_inode->i_sb;
4492 +}
4493 +
4494 +void au_cpup_attr_all(struct inode *inode, int force)
4495 +{
4496 +       struct inode *h_inode;
4497 +
4498 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
4499 +       au_cpup_attr_changeable(inode);
4500 +       if (inode->i_nlink > 0)
4501 +               au_cpup_attr_nlink(inode, force);
4502 +       inode->i_rdev = h_inode->i_rdev;
4503 +       inode->i_blkbits = h_inode->i_blkbits;
4504 +       au_cpup_igen(inode, h_inode);
4505 +}
4506 +
4507 +/* ---------------------------------------------------------------------- */
4508 +
4509 +/* Note: dt_dentry and dt_h_dentry are not dget/dput-ed */
4510 +
4511 +/* keep the timestamps of the parent dir when cpup */
4512 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
4513 +                   struct path *h_path)
4514 +{
4515 +       struct inode *h_inode;
4516 +
4517 +       dt->dt_dentry = dentry;
4518 +       dt->dt_h_path = *h_path;
4519 +       h_inode = h_path->dentry->d_inode;
4520 +       dt->dt_atime = h_inode->i_atime;
4521 +       dt->dt_mtime = h_inode->i_mtime;
4522 +       /* smp_mb(); */
4523 +}
4524 +
4525 +void au_dtime_revert(struct au_dtime *dt)
4526 +{
4527 +       struct iattr attr;
4528 +       int err;
4529 +
4530 +       attr.ia_atime = dt->dt_atime;
4531 +       attr.ia_mtime = dt->dt_mtime;
4532 +       attr.ia_valid = ATTR_FORCE | ATTR_MTIME | ATTR_MTIME_SET
4533 +               | ATTR_ATIME | ATTR_ATIME_SET;
4534 +
4535 +       /* no delegation since this is a directory */
4536 +       err = vfsub_notify_change(&dt->dt_h_path, &attr, /*delegated*/NULL);
4537 +       if (unlikely(err))
4538 +               pr_warn("restoring timestamps failed(%d). ignored\n", err);
4539 +}
4540 +
4541 +/* ---------------------------------------------------------------------- */
4542 +
4543 +/* internal use only */
4544 +struct au_cpup_reg_attr {
4545 +       int             valid;
4546 +       struct kstat    st;
4547 +       unsigned int    iflags; /* inode->i_flags */
4548 +};
4549 +
4550 +static noinline_for_stack
4551 +int cpup_iattr(struct dentry *dst, aufs_bindex_t bindex, struct dentry *h_src,
4552 +              struct au_cpup_reg_attr *h_src_attr)
4553 +{
4554 +       int err, sbits, icex;
4555 +       struct iattr ia;
4556 +       struct path h_path;
4557 +       struct inode *h_isrc, *h_idst;
4558 +       struct kstat *h_st;
4559 +       struct au_branch *br;
4560 +
4561 +       h_path.dentry = au_h_dptr(dst, bindex);
4562 +       h_idst = h_path.dentry->d_inode;
4563 +       br = au_sbr(dst->d_sb, bindex);
4564 +       h_path.mnt = au_br_mnt(br);
4565 +       h_isrc = h_src->d_inode;
4566 +       ia.ia_valid = ATTR_FORCE | ATTR_UID | ATTR_GID
4567 +               | ATTR_ATIME | ATTR_MTIME
4568 +               | ATTR_ATIME_SET | ATTR_MTIME_SET;
4569 +       if (h_src_attr && h_src_attr->valid) {
4570 +               h_st = &h_src_attr->st;
4571 +               ia.ia_uid = h_st->uid;
4572 +               ia.ia_gid = h_st->gid;
4573 +               ia.ia_atime = h_st->atime;
4574 +               ia.ia_mtime = h_st->mtime;
4575 +               if (h_idst->i_mode != h_st->mode
4576 +                   && !S_ISLNK(h_idst->i_mode)) {
4577 +                       ia.ia_valid |= ATTR_MODE;
4578 +                       ia.ia_mode = h_st->mode;
4579 +               }
4580 +               sbits = !!(h_st->mode & (S_ISUID | S_ISGID));
4581 +               au_cpup_attr_flags(h_idst, h_src_attr->iflags);
4582 +       } else {
4583 +               ia.ia_uid = h_isrc->i_uid;
4584 +               ia.ia_gid = h_isrc->i_gid;
4585 +               ia.ia_atime = h_isrc->i_atime;
4586 +               ia.ia_mtime = h_isrc->i_mtime;
4587 +               if (h_idst->i_mode != h_isrc->i_mode
4588 +                   && !S_ISLNK(h_idst->i_mode)) {
4589 +                       ia.ia_valid |= ATTR_MODE;
4590 +                       ia.ia_mode = h_isrc->i_mode;
4591 +               }
4592 +               sbits = !!(h_isrc->i_mode & (S_ISUID | S_ISGID));
4593 +               au_cpup_attr_flags(h_idst, h_isrc->i_flags);
4594 +       }
4595 +       /* no delegation since it is just created */
4596 +       err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4597 +
4598 +       /* is this nfs only? */
4599 +       if (!err && sbits && au_test_nfs(h_path.dentry->d_sb)) {
4600 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
4601 +               ia.ia_mode = h_isrc->i_mode;
4602 +               err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4603 +       }
4604 +
4605 +       icex = br->br_perm & AuBrAttr_ICEX;
4606 +       if (!err)
4607 +               err = au_cpup_xattr(h_path.dentry, h_src, icex);
4608 +
4609 +       return err;
4610 +}
4611 +
4612 +/* ---------------------------------------------------------------------- */
4613 +
4614 +static int au_do_copy_file(struct file *dst, struct file *src, loff_t len,
4615 +                          char *buf, unsigned long blksize)
4616 +{
4617 +       int err;
4618 +       size_t sz, rbytes, wbytes;
4619 +       unsigned char all_zero;
4620 +       char *p, *zp;
4621 +       struct mutex *h_mtx;
4622 +       /* reduce stack usage */
4623 +       struct iattr *ia;
4624 +
4625 +       zp = page_address(ZERO_PAGE(0));
4626 +       if (unlikely(!zp))
4627 +               return -ENOMEM; /* possible? */
4628 +
4629 +       err = 0;
4630 +       all_zero = 0;
4631 +       while (len) {
4632 +               AuDbg("len %lld\n", len);
4633 +               sz = blksize;
4634 +               if (len < blksize)
4635 +                       sz = len;
4636 +
4637 +               rbytes = 0;
4638 +               /* todo: signal_pending? */
4639 +               while (!rbytes || err == -EAGAIN || err == -EINTR) {
4640 +                       rbytes = vfsub_read_k(src, buf, sz, &src->f_pos);
4641 +                       err = rbytes;
4642 +               }
4643 +               if (unlikely(err < 0))
4644 +                       break;
4645 +
4646 +               all_zero = 0;
4647 +               if (len >= rbytes && rbytes == blksize)
4648 +                       all_zero = !memcmp(buf, zp, rbytes);
4649 +               if (!all_zero) {
4650 +                       wbytes = rbytes;
4651 +                       p = buf;
4652 +                       while (wbytes) {
4653 +                               size_t b;
4654 +
4655 +                               b = vfsub_write_k(dst, p, wbytes, &dst->f_pos);
4656 +                               err = b;
4657 +                               /* todo: signal_pending? */
4658 +                               if (unlikely(err == -EAGAIN || err == -EINTR))
4659 +                                       continue;
4660 +                               if (unlikely(err < 0))
4661 +                                       break;
4662 +                               wbytes -= b;
4663 +                               p += b;
4664 +                       }
4665 +                       if (unlikely(err < 0))
4666 +                               break;
4667 +               } else {
4668 +                       loff_t res;
4669 +
4670 +                       AuLabel(hole);
4671 +                       res = vfsub_llseek(dst, rbytes, SEEK_CUR);
4672 +                       err = res;
4673 +                       if (unlikely(res < 0))
4674 +                               break;
4675 +               }
4676 +               len -= rbytes;
4677 +               err = 0;
4678 +       }
4679 +
4680 +       /* the last block may be a hole */
4681 +       if (!err && all_zero) {
4682 +               AuLabel(last hole);
4683 +
4684 +               err = 1;
4685 +               if (au_test_nfs(dst->f_dentry->d_sb)) {
4686 +                       /* nfs requires this step to make last hole */
4687 +                       /* is this only nfs? */
4688 +                       do {
4689 +                               /* todo: signal_pending? */
4690 +                               err = vfsub_write_k(dst, "\0", 1, &dst->f_pos);
4691 +                       } while (err == -EAGAIN || err == -EINTR);
4692 +                       if (err == 1)
4693 +                               dst->f_pos--;
4694 +               }
4695 +
4696 +               if (err == 1) {
4697 +                       ia = (void *)buf;
4698 +                       ia->ia_size = dst->f_pos;
4699 +                       ia->ia_valid = ATTR_SIZE | ATTR_FILE;
4700 +                       ia->ia_file = dst;
4701 +                       h_mtx = &file_inode(dst)->i_mutex;
4702 +                       mutex_lock_nested(h_mtx, AuLsc_I_CHILD2);
4703 +                       /* no delegation since it is just created */
4704 +                       err = vfsub_notify_change(&dst->f_path, ia,
4705 +                                                 /*delegated*/NULL);
4706 +                       mutex_unlock(h_mtx);
4707 +               }
4708 +       }
4709 +
4710 +       return err;
4711 +}
4712 +
4713 +int au_copy_file(struct file *dst, struct file *src, loff_t len)
4714 +{
4715 +       int err;
4716 +       unsigned long blksize;
4717 +       unsigned char do_kfree;
4718 +       char *buf;
4719 +
4720 +       err = -ENOMEM;
4721 +       blksize = dst->f_dentry->d_sb->s_blocksize;
4722 +       if (!blksize || PAGE_SIZE < blksize)
4723 +               blksize = PAGE_SIZE;
4724 +       AuDbg("blksize %lu\n", blksize);
4725 +       do_kfree = (blksize != PAGE_SIZE && blksize >= sizeof(struct iattr *));
4726 +       if (do_kfree)
4727 +               buf = kmalloc(blksize, GFP_NOFS);
4728 +       else
4729 +               buf = (void *)__get_free_page(GFP_NOFS);
4730 +       if (unlikely(!buf))
4731 +               goto out;
4732 +
4733 +       if (len > (1 << 22))
4734 +               AuDbg("copying a large file %lld\n", (long long)len);
4735 +
4736 +       src->f_pos = 0;
4737 +       dst->f_pos = 0;
4738 +       err = au_do_copy_file(dst, src, len, buf, blksize);
4739 +       if (do_kfree)
4740 +               kfree(buf);
4741 +       else
4742 +               free_page((unsigned long)buf);
4743 +
4744 +out:
4745 +       return err;
4746 +}
4747 +
4748 +/*
4749 + * to support a sparse file which is opened with O_APPEND,
4750 + * we need to close the file.
4751 + */
4752 +static int au_cp_regular(struct au_cp_generic *cpg)
4753 +{
4754 +       int err, i;
4755 +       enum { SRC, DST };
4756 +       struct {
4757 +               aufs_bindex_t bindex;
4758 +               unsigned int flags;
4759 +               struct dentry *dentry;
4760 +               int force_wr;
4761 +               struct file *file;
4762 +               void *label;
4763 +       } *f, file[] = {
4764 +               {
4765 +                       .bindex = cpg->bsrc,
4766 +                       .flags = O_RDONLY | O_NOATIME | O_LARGEFILE,
4767 +                       .label = &&out
4768 +               },
4769 +               {
4770 +                       .bindex = cpg->bdst,
4771 +                       .flags = O_WRONLY | O_NOATIME | O_LARGEFILE,
4772 +                       .force_wr = !!au_ftest_cpup(cpg->flags, RWDST),
4773 +                       .label = &&out_src
4774 +               }
4775 +       };
4776 +       struct super_block *sb;
4777 +
4778 +       /* bsrc branch can be ro/rw. */
4779 +       sb = cpg->dentry->d_sb;
4780 +       f = file;
4781 +       for (i = 0; i < 2; i++, f++) {
4782 +               f->dentry = au_h_dptr(cpg->dentry, f->bindex);
4783 +               f->file = au_h_open(cpg->dentry, f->bindex, f->flags,
4784 +                                   /*file*/NULL, f->force_wr);
4785 +               err = PTR_ERR(f->file);
4786 +               if (IS_ERR(f->file))
4787 +                       goto *f->label;
4788 +       }
4789 +
4790 +       /* try stopping to update while we copyup */
4791 +       IMustLock(file[SRC].dentry->d_inode);
4792 +       err = au_copy_file(file[DST].file, file[SRC].file, cpg->len);
4793 +
4794 +       fput(file[DST].file);
4795 +       au_sbr_put(sb, file[DST].bindex);
4796 +
4797 +out_src:
4798 +       fput(file[SRC].file);
4799 +       au_sbr_put(sb, file[SRC].bindex);
4800 +out:
4801 +       return err;
4802 +}
4803 +
4804 +static int au_do_cpup_regular(struct au_cp_generic *cpg,
4805 +                             struct au_cpup_reg_attr *h_src_attr)
4806 +{
4807 +       int err, rerr;
4808 +       loff_t l;
4809 +       struct path h_path;
4810 +       struct inode *h_src_inode, *h_dst_inode;
4811 +
4812 +       err = 0;
4813 +       h_src_inode = au_h_iptr(cpg->dentry->d_inode, cpg->bsrc);
4814 +       l = i_size_read(h_src_inode);
4815 +       if (cpg->len == -1 || l < cpg->len)
4816 +               cpg->len = l;
4817 +       if (cpg->len) {
4818 +               /* try stopping to update while we are referencing */
4819 +               mutex_lock_nested(&h_src_inode->i_mutex, AuLsc_I_CHILD);
4820 +               au_pin_hdir_unlock(cpg->pin);
4821 +
4822 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
4823 +               h_path.mnt = au_sbr_mnt(cpg->dentry->d_sb, cpg->bsrc);
4824 +               h_src_attr->iflags = h_src_inode->i_flags;
4825 +               err = vfs_getattr(&h_path, &h_src_attr->st);
4826 +               if (unlikely(err)) {
4827 +                       mutex_unlock(&h_src_inode->i_mutex);
4828 +                       goto out;
4829 +               }
4830 +               h_src_attr->valid = 1;
4831 +               err = au_cp_regular(cpg);
4832 +               mutex_unlock(&h_src_inode->i_mutex);
4833 +               rerr = au_pin_hdir_relock(cpg->pin);
4834 +               if (!err && rerr)
4835 +                       err = rerr;
4836 +       }
4837 +       if (!err && (h_src_inode->i_state & I_LINKABLE)) {
4838 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bdst);
4839 +               h_dst_inode = h_path.dentry->d_inode;
4840 +               spin_lock(&h_dst_inode->i_lock);
4841 +               h_dst_inode->i_state |= I_LINKABLE;
4842 +               spin_unlock(&h_dst_inode->i_lock);
4843 +       }
4844 +
4845 +out:
4846 +       return err;
4847 +}
4848 +
4849 +static int au_do_cpup_symlink(struct path *h_path, struct dentry *h_src,
4850 +                             struct inode *h_dir)
4851 +{
4852 +       int err, symlen;
4853 +       mm_segment_t old_fs;
4854 +       union {
4855 +               char *k;
4856 +               char __user *u;
4857 +       } sym;
4858 +
4859 +       err = -ENOSYS;
4860 +       if (unlikely(!h_src->d_inode->i_op->readlink))
4861 +               goto out;
4862 +
4863 +       err = -ENOMEM;
4864 +       sym.k = (void *)__get_free_page(GFP_NOFS);
4865 +       if (unlikely(!sym.k))
4866 +               goto out;
4867 +
4868 +       /* unnecessary to support mmap_sem since symlink is not mmap-able */
4869 +       old_fs = get_fs();
4870 +       set_fs(KERNEL_DS);
4871 +       symlen = h_src->d_inode->i_op->readlink(h_src, sym.u, PATH_MAX);
4872 +       err = symlen;
4873 +       set_fs(old_fs);
4874 +
4875 +       if (symlen > 0) {
4876 +               sym.k[symlen] = 0;
4877 +               err = vfsub_symlink(h_dir, h_path, sym.k);
4878 +       }
4879 +       free_page((unsigned long)sym.k);
4880 +
4881 +out:
4882 +       return err;
4883 +}
4884 +
4885 +static noinline_for_stack
4886 +int cpup_entry(struct au_cp_generic *cpg, struct dentry *dst_parent,
4887 +              struct au_cpup_reg_attr *h_src_attr)
4888 +{
4889 +       int err;
4890 +       umode_t mode;
4891 +       unsigned int mnt_flags;
4892 +       unsigned char isdir, isreg, force;
4893 +       const unsigned char do_dt = !!au_ftest_cpup(cpg->flags, DTIME);
4894 +       struct au_dtime dt;
4895 +       struct path h_path;
4896 +       struct dentry *h_src, *h_dst, *h_parent;
4897 +       struct inode *h_inode, *h_dir;
4898 +       struct super_block *sb;
4899 +
4900 +       /* bsrc branch can be ro/rw. */
4901 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
4902 +       h_inode = h_src->d_inode;
4903 +       AuDebugOn(h_inode != au_h_iptr(cpg->dentry->d_inode, cpg->bsrc));
4904 +
4905 +       /* try stopping to be referenced while we are creating */
4906 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
4907 +       if (au_ftest_cpup(cpg->flags, RENAME))
4908 +               AuDebugOn(strncmp(h_dst->d_name.name, AUFS_WH_PFX,
4909 +                                 AUFS_WH_PFX_LEN));
4910 +       h_parent = h_dst->d_parent; /* dir inode is locked */
4911 +       h_dir = h_parent->d_inode;
4912 +       IMustLock(h_dir);
4913 +       AuDebugOn(h_parent != h_dst->d_parent);
4914 +
4915 +       sb = cpg->dentry->d_sb;
4916 +       h_path.mnt = au_sbr_mnt(sb, cpg->bdst);
4917 +       if (do_dt) {
4918 +               h_path.dentry = h_parent;
4919 +               au_dtime_store(&dt, dst_parent, &h_path);
4920 +       }
4921 +       h_path.dentry = h_dst;
4922 +
4923 +       isreg = 0;
4924 +       isdir = 0;
4925 +       mode = h_inode->i_mode;
4926 +       switch (mode & S_IFMT) {
4927 +       case S_IFREG:
4928 +               isreg = 1;
4929 +               err = vfsub_create(h_dir, &h_path, mode | S_IWUSR,
4930 +                                  /*want_excl*/true);
4931 +               if (!err)
4932 +                       err = au_do_cpup_regular(cpg, h_src_attr);
4933 +               break;
4934 +       case S_IFDIR:
4935 +               isdir = 1;
4936 +               err = vfsub_mkdir(h_dir, &h_path, mode);
4937 +               if (!err) {
4938 +                       /*
4939 +                        * strange behaviour from the users view,
4940 +                        * particularry setattr case
4941 +                        */
4942 +                       if (au_ibstart(dst_parent->d_inode) == cpg->bdst)
4943 +                               au_cpup_attr_nlink(dst_parent->d_inode,
4944 +                                                  /*force*/1);
4945 +                       au_cpup_attr_nlink(cpg->dentry->d_inode, /*force*/1);
4946 +               }
4947 +               break;
4948 +       case S_IFLNK:
4949 +               err = au_do_cpup_symlink(&h_path, h_src, h_dir);
4950 +               break;
4951 +       case S_IFCHR:
4952 +       case S_IFBLK:
4953 +               AuDebugOn(!capable(CAP_MKNOD));
4954 +               /*FALLTHROUGH*/
4955 +       case S_IFIFO:
4956 +       case S_IFSOCK:
4957 +               err = vfsub_mknod(h_dir, &h_path, mode, h_inode->i_rdev);
4958 +               break;
4959 +       default:
4960 +               AuIOErr("Unknown inode type 0%o\n", mode);
4961 +               err = -EIO;
4962 +       }
4963 +
4964 +       mnt_flags = au_mntflags(sb);
4965 +       if (!au_opt_test(mnt_flags, UDBA_NONE)
4966 +           && !isdir
4967 +           && au_opt_test(mnt_flags, XINO)
4968 +           && (h_inode->i_nlink == 1
4969 +               || (h_inode->i_state & I_LINKABLE))
4970 +           /* todo: unnecessary? */
4971 +           /* && cpg->dentry->d_inode->i_nlink == 1 */
4972 +           && cpg->bdst < cpg->bsrc
4973 +           && !au_ftest_cpup(cpg->flags, KEEPLINO))
4974 +               au_xino_write(sb, cpg->bsrc, h_inode->i_ino, /*ino*/0);
4975 +               /* ignore this error */
4976 +
4977 +       if (!err) {
4978 +               force = 0;
4979 +               if (isreg) {
4980 +                       force = !!cpg->len;
4981 +                       if (cpg->len == -1)
4982 +                               force = !!i_size_read(h_inode);
4983 +               }
4984 +               au_fhsm_wrote(sb, cpg->bdst, force);
4985 +       }
4986 +
4987 +       if (do_dt)
4988 +               au_dtime_revert(&dt);
4989 +       return err;
4990 +}
4991 +
4992 +static int au_do_ren_after_cpup(struct au_cp_generic *cpg, struct path *h_path)
4993 +{
4994 +       int err;
4995 +       struct dentry *dentry, *h_dentry, *h_parent, *parent;
4996 +       struct inode *h_dir;
4997 +       aufs_bindex_t bdst;
4998 +
4999 +       dentry = cpg->dentry;
5000 +       bdst = cpg->bdst;
5001 +       h_dentry = au_h_dptr(dentry, bdst);
5002 +       if (!au_ftest_cpup(cpg->flags, OVERWRITE)) {
5003 +               dget(h_dentry);
5004 +               au_set_h_dptr(dentry, bdst, NULL);
5005 +               err = au_lkup_neg(dentry, bdst, /*wh*/0);
5006 +               if (!err)
5007 +                       h_path->dentry = dget(au_h_dptr(dentry, bdst));
5008 +               au_set_h_dptr(dentry, bdst, h_dentry);
5009 +       } else {
5010 +               err = 0;
5011 +               parent = dget_parent(dentry);
5012 +               h_parent = au_h_dptr(parent, bdst);
5013 +               dput(parent);
5014 +               h_path->dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
5015 +               if (IS_ERR(h_path->dentry))
5016 +                       err = PTR_ERR(h_path->dentry);
5017 +       }
5018 +       if (unlikely(err))
5019 +               goto out;
5020 +
5021 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
5022 +       h_dir = h_parent->d_inode;
5023 +       IMustLock(h_dir);
5024 +       AuDbg("%pd %pd\n", h_dentry, h_path->dentry);
5025 +       /* no delegation since it is just created */
5026 +       err = vfsub_rename(h_dir, h_dentry, h_dir, h_path, /*delegated*/NULL);
5027 +       dput(h_path->dentry);
5028 +
5029 +out:
5030 +       return err;
5031 +}
5032 +
5033 +/*
5034 + * copyup the @dentry from @bsrc to @bdst.
5035 + * the caller must set the both of lower dentries.
5036 + * @len is for truncating when it is -1 copyup the entire file.
5037 + * in link/rename cases, @dst_parent may be different from the real one.
5038 + * basic->bsrc can be larger than basic->bdst.
5039 + */
5040 +static int au_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5041 +{
5042 +       int err, rerr;
5043 +       aufs_bindex_t old_ibstart;
5044 +       unsigned char isdir, plink;
5045 +       struct dentry *h_src, *h_dst, *h_parent;
5046 +       struct inode *dst_inode, *h_dir, *inode, *delegated;
5047 +       struct super_block *sb;
5048 +       struct au_branch *br;
5049 +       /* to reuduce stack size */
5050 +       struct {
5051 +               struct au_dtime dt;
5052 +               struct path h_path;
5053 +               struct au_cpup_reg_attr h_src_attr;
5054 +       } *a;
5055 +
5056 +       err = -ENOMEM;
5057 +       a = kmalloc(sizeof(*a), GFP_NOFS);
5058 +       if (unlikely(!a))
5059 +               goto out;
5060 +       a->h_src_attr.valid = 0;
5061 +
5062 +       sb = cpg->dentry->d_sb;
5063 +       br = au_sbr(sb, cpg->bdst);
5064 +       a->h_path.mnt = au_br_mnt(br);
5065 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5066 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5067 +       h_dir = h_parent->d_inode;
5068 +       IMustLock(h_dir);
5069 +
5070 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5071 +       inode = cpg->dentry->d_inode;
5072 +
5073 +       if (!dst_parent)
5074 +               dst_parent = dget_parent(cpg->dentry);
5075 +       else
5076 +               dget(dst_parent);
5077 +
5078 +       plink = !!au_opt_test(au_mntflags(sb), PLINK);
5079 +       dst_inode = au_h_iptr(inode, cpg->bdst);
5080 +       if (dst_inode) {
5081 +               if (unlikely(!plink)) {
5082 +                       err = -EIO;
5083 +                       AuIOErr("hi%lu(i%lu) exists on b%d "
5084 +                               "but plink is disabled\n",
5085 +                               dst_inode->i_ino, inode->i_ino, cpg->bdst);
5086 +                       goto out_parent;
5087 +               }
5088 +
5089 +               if (dst_inode->i_nlink) {
5090 +                       const int do_dt = au_ftest_cpup(cpg->flags, DTIME);
5091 +
5092 +                       h_src = au_plink_lkup(inode, cpg->bdst);
5093 +                       err = PTR_ERR(h_src);
5094 +                       if (IS_ERR(h_src))
5095 +                               goto out_parent;
5096 +                       if (unlikely(!h_src->d_inode)) {
5097 +                               err = -EIO;
5098 +                               AuIOErr("i%lu exists on a upper branch "
5099 +                                       "but not pseudo-linked\n",
5100 +                                       inode->i_ino);
5101 +                               dput(h_src);
5102 +                               goto out_parent;
5103 +                       }
5104 +
5105 +                       if (do_dt) {
5106 +                               a->h_path.dentry = h_parent;
5107 +                               au_dtime_store(&a->dt, dst_parent, &a->h_path);
5108 +                       }
5109 +
5110 +                       a->h_path.dentry = h_dst;
5111 +                       delegated = NULL;
5112 +                       err = vfsub_link(h_src, h_dir, &a->h_path, &delegated);
5113 +                       if (!err && au_ftest_cpup(cpg->flags, RENAME))
5114 +                               err = au_do_ren_after_cpup(cpg, &a->h_path);
5115 +                       if (do_dt)
5116 +                               au_dtime_revert(&a->dt);
5117 +                       if (unlikely(err == -EWOULDBLOCK)) {
5118 +                               pr_warn("cannot retry for NFSv4 delegation"
5119 +                                       " for an internal link\n");
5120 +                               iput(delegated);
5121 +                       }
5122 +                       dput(h_src);
5123 +                       goto out_parent;
5124 +               } else
5125 +                       /* todo: cpup_wh_file? */
5126 +                       /* udba work */
5127 +                       au_update_ibrange(inode, /*do_put_zero*/1);
5128 +       }
5129 +
5130 +       isdir = S_ISDIR(inode->i_mode);
5131 +       old_ibstart = au_ibstart(inode);
5132 +       err = cpup_entry(cpg, dst_parent, &a->h_src_attr);
5133 +       if (unlikely(err))
5134 +               goto out_rev;
5135 +       dst_inode = h_dst->d_inode;
5136 +       mutex_lock_nested(&dst_inode->i_mutex, AuLsc_I_CHILD2);
5137 +       /* todo: necessary? */
5138 +       /* au_pin_hdir_unlock(cpg->pin); */
5139 +
5140 +       err = cpup_iattr(cpg->dentry, cpg->bdst, h_src, &a->h_src_attr);
5141 +       if (unlikely(err)) {
5142 +               /* todo: necessary? */
5143 +               /* au_pin_hdir_relock(cpg->pin); */ /* ignore an error */
5144 +               mutex_unlock(&dst_inode->i_mutex);
5145 +               goto out_rev;
5146 +       }
5147 +
5148 +       if (cpg->bdst < old_ibstart) {
5149 +               if (S_ISREG(inode->i_mode)) {
5150 +                       err = au_dy_iaop(inode, cpg->bdst, dst_inode);
5151 +                       if (unlikely(err)) {
5152 +                               /* ignore an error */
5153 +                               /* au_pin_hdir_relock(cpg->pin); */
5154 +                               mutex_unlock(&dst_inode->i_mutex);
5155 +                               goto out_rev;
5156 +                       }
5157 +               }
5158 +               au_set_ibstart(inode, cpg->bdst);
5159 +       } else
5160 +               au_set_ibend(inode, cpg->bdst);
5161 +       au_set_h_iptr(inode, cpg->bdst, au_igrab(dst_inode),
5162 +                     au_hi_flags(inode, isdir));
5163 +
5164 +       /* todo: necessary? */
5165 +       /* err = au_pin_hdir_relock(cpg->pin); */
5166 +       mutex_unlock(&dst_inode->i_mutex);
5167 +       if (unlikely(err))
5168 +               goto out_rev;
5169 +
5170 +       if (!isdir
5171 +           && (h_src->d_inode->i_nlink > 1
5172 +               || h_src->d_inode->i_state & I_LINKABLE)
5173 +           && plink)
5174 +               au_plink_append(inode, cpg->bdst, h_dst);
5175 +
5176 +       if (au_ftest_cpup(cpg->flags, RENAME)) {
5177 +               a->h_path.dentry = h_dst;
5178 +               err = au_do_ren_after_cpup(cpg, &a->h_path);
5179 +       }
5180 +       if (!err)
5181 +               goto out_parent; /* success */
5182 +
5183 +       /* revert */
5184 +out_rev:
5185 +       a->h_path.dentry = h_parent;
5186 +       au_dtime_store(&a->dt, dst_parent, &a->h_path);
5187 +       a->h_path.dentry = h_dst;
5188 +       rerr = 0;
5189 +       if (h_dst->d_inode) {
5190 +               if (!isdir) {
5191 +                       /* no delegation since it is just created */
5192 +                       rerr = vfsub_unlink(h_dir, &a->h_path,
5193 +                                           /*delegated*/NULL, /*force*/0);
5194 +               } else
5195 +                       rerr = vfsub_rmdir(h_dir, &a->h_path);
5196 +       }
5197 +       au_dtime_revert(&a->dt);
5198 +       if (rerr) {
5199 +               AuIOErr("failed removing broken entry(%d, %d)\n", err, rerr);
5200 +               err = -EIO;
5201 +       }
5202 +out_parent:
5203 +       dput(dst_parent);
5204 +       kfree(a);
5205 +out:
5206 +       return err;
5207 +}
5208 +
5209 +#if 0 /* unused */
5210 +struct au_cpup_single_args {
5211 +       int *errp;
5212 +       struct au_cp_generic *cpg;
5213 +       struct dentry *dst_parent;
5214 +};
5215 +
5216 +static void au_call_cpup_single(void *args)
5217 +{
5218 +       struct au_cpup_single_args *a = args;
5219 +
5220 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5221 +       *a->errp = au_cpup_single(a->cpg, a->dst_parent);
5222 +       au_pin_hdir_release(a->cpg->pin);
5223 +}
5224 +#endif
5225 +
5226 +/*
5227 + * prevent SIGXFSZ in copy-up.
5228 + * testing CAP_MKNOD is for generic fs,
5229 + * but CAP_FSETID is for xfs only, currently.
5230 + */
5231 +static int au_cpup_sio_test(struct au_pin *pin, umode_t mode)
5232 +{
5233 +       int do_sio;
5234 +       struct super_block *sb;
5235 +       struct inode *h_dir;
5236 +
5237 +       do_sio = 0;
5238 +       sb = au_pinned_parent(pin)->d_sb;
5239 +       if (!au_wkq_test()
5240 +           && (!au_sbi(sb)->si_plink_maint_pid
5241 +               || au_plink_maint(sb, AuLock_NOPLM))) {
5242 +               switch (mode & S_IFMT) {
5243 +               case S_IFREG:
5244 +                       /* no condition about RLIMIT_FSIZE and the file size */
5245 +                       do_sio = 1;
5246 +                       break;
5247 +               case S_IFCHR:
5248 +               case S_IFBLK:
5249 +                       do_sio = !capable(CAP_MKNOD);
5250 +                       break;
5251 +               }
5252 +               if (!do_sio)
5253 +                       do_sio = ((mode & (S_ISUID | S_ISGID))
5254 +                                 && !capable(CAP_FSETID));
5255 +               /* this workaround may be removed in the future */
5256 +               if (!do_sio) {
5257 +                       h_dir = au_pinned_h_dir(pin);
5258 +                       do_sio = h_dir->i_mode & S_ISVTX;
5259 +               }
5260 +       }
5261 +
5262 +       return do_sio;
5263 +}
5264 +
5265 +#if 0 /* unused */
5266 +int au_sio_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5267 +{
5268 +       int err, wkq_err;
5269 +       struct dentry *h_dentry;
5270 +
5271 +       h_dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5272 +       if (!au_cpup_sio_test(pin, h_dentry->d_inode->i_mode))
5273 +               err = au_cpup_single(cpg, dst_parent);
5274 +       else {
5275 +               struct au_cpup_single_args args = {
5276 +                       .errp           = &err,
5277 +                       .cpg            = cpg,
5278 +                       .dst_parent     = dst_parent
5279 +               };
5280 +               wkq_err = au_wkq_wait(au_call_cpup_single, &args);
5281 +               if (unlikely(wkq_err))
5282 +                       err = wkq_err;
5283 +       }
5284 +
5285 +       return err;
5286 +}
5287 +#endif
5288 +
5289 +/*
5290 + * copyup the @dentry from the first active lower branch to @bdst,
5291 + * using au_cpup_single().
5292 + */
5293 +static int au_cpup_simple(struct au_cp_generic *cpg)
5294 +{
5295 +       int err;
5296 +       unsigned int flags_orig;
5297 +       struct dentry *dentry;
5298 +
5299 +       AuDebugOn(cpg->bsrc < 0);
5300 +
5301 +       dentry = cpg->dentry;
5302 +       DiMustWriteLock(dentry);
5303 +
5304 +       err = au_lkup_neg(dentry, cpg->bdst, /*wh*/1);
5305 +       if (!err) {
5306 +               flags_orig = cpg->flags;
5307 +               au_fset_cpup(cpg->flags, RENAME);
5308 +               err = au_cpup_single(cpg, NULL);
5309 +               cpg->flags = flags_orig;
5310 +               if (!err)
5311 +                       return 0; /* success */
5312 +
5313 +               /* revert */
5314 +               au_set_h_dptr(dentry, cpg->bdst, NULL);
5315 +               au_set_dbstart(dentry, cpg->bsrc);
5316 +       }
5317 +
5318 +       return err;
5319 +}
5320 +
5321 +struct au_cpup_simple_args {
5322 +       int *errp;
5323 +       struct au_cp_generic *cpg;
5324 +};
5325 +
5326 +static void au_call_cpup_simple(void *args)
5327 +{
5328 +       struct au_cpup_simple_args *a = args;
5329 +
5330 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5331 +       *a->errp = au_cpup_simple(a->cpg);
5332 +       au_pin_hdir_release(a->cpg->pin);
5333 +}
5334 +
5335 +static int au_do_sio_cpup_simple(struct au_cp_generic *cpg)
5336 +{
5337 +       int err, wkq_err;
5338 +       struct dentry *dentry, *parent;
5339 +       struct file *h_file;
5340 +       struct inode *h_dir;
5341 +
5342 +       dentry = cpg->dentry;
5343 +       h_file = NULL;
5344 +       if (au_ftest_cpup(cpg->flags, HOPEN)) {
5345 +               AuDebugOn(cpg->bsrc < 0);
5346 +               h_file = au_h_open_pre(dentry, cpg->bsrc, /*force_wr*/0);
5347 +               err = PTR_ERR(h_file);
5348 +               if (IS_ERR(h_file))
5349 +                       goto out;
5350 +       }
5351 +
5352 +       parent = dget_parent(dentry);
5353 +       h_dir = au_h_iptr(parent->d_inode, cpg->bdst);
5354 +       if (!au_test_h_perm_sio(h_dir, MAY_EXEC | MAY_WRITE)
5355 +           && !au_cpup_sio_test(cpg->pin, dentry->d_inode->i_mode))
5356 +               err = au_cpup_simple(cpg);
5357 +       else {
5358 +               struct au_cpup_simple_args args = {
5359 +                       .errp           = &err,
5360 +                       .cpg            = cpg
5361 +               };
5362 +               wkq_err = au_wkq_wait(au_call_cpup_simple, &args);
5363 +               if (unlikely(wkq_err))
5364 +                       err = wkq_err;
5365 +       }
5366 +
5367 +       dput(parent);
5368 +       if (h_file)
5369 +               au_h_open_post(dentry, cpg->bsrc, h_file);
5370 +
5371 +out:
5372 +       return err;
5373 +}
5374 +
5375 +int au_sio_cpup_simple(struct au_cp_generic *cpg)
5376 +{
5377 +       aufs_bindex_t bsrc, bend;
5378 +       struct dentry *dentry, *h_dentry;
5379 +
5380 +       if (cpg->bsrc < 0) {
5381 +               dentry = cpg->dentry;
5382 +               bend = au_dbend(dentry);
5383 +               for (bsrc = cpg->bdst + 1; bsrc <= bend; bsrc++) {
5384 +                       h_dentry = au_h_dptr(dentry, bsrc);
5385 +                       if (h_dentry) {
5386 +                               AuDebugOn(!h_dentry->d_inode);
5387 +                               break;
5388 +                       }
5389 +               }
5390 +               AuDebugOn(bsrc > bend);
5391 +               cpg->bsrc = bsrc;
5392 +       }
5393 +       AuDebugOn(cpg->bsrc <= cpg->bdst);
5394 +       return au_do_sio_cpup_simple(cpg);
5395 +}
5396 +
5397 +int au_sio_cpdown_simple(struct au_cp_generic *cpg)
5398 +{
5399 +       AuDebugOn(cpg->bdst <= cpg->bsrc);
5400 +       return au_do_sio_cpup_simple(cpg);
5401 +}
5402 +
5403 +/* ---------------------------------------------------------------------- */
5404 +
5405 +/*
5406 + * copyup the deleted file for writing.
5407 + */
5408 +static int au_do_cpup_wh(struct au_cp_generic *cpg, struct dentry *wh_dentry,
5409 +                        struct file *file)
5410 +{
5411 +       int err;
5412 +       unsigned int flags_orig;
5413 +       aufs_bindex_t bsrc_orig;
5414 +       struct dentry *h_d_dst, *h_d_start;
5415 +       struct au_dinfo *dinfo;
5416 +       struct au_hdentry *hdp;
5417 +
5418 +       dinfo = au_di(cpg->dentry);
5419 +       AuRwMustWriteLock(&dinfo->di_rwsem);
5420 +
5421 +       bsrc_orig = cpg->bsrc;
5422 +       cpg->bsrc = dinfo->di_bstart;
5423 +       hdp = dinfo->di_hdentry;
5424 +       h_d_dst = hdp[0 + cpg->bdst].hd_dentry;
5425 +       dinfo->di_bstart = cpg->bdst;
5426 +       hdp[0 + cpg->bdst].hd_dentry = wh_dentry;
5427 +       h_d_start = NULL;
5428 +       if (file) {
5429 +               h_d_start = hdp[0 + cpg->bsrc].hd_dentry;
5430 +               hdp[0 + cpg->bsrc].hd_dentry = au_hf_top(file)->f_dentry;
5431 +       }
5432 +       flags_orig = cpg->flags;
5433 +       cpg->flags = !AuCpup_DTIME;
5434 +       err = au_cpup_single(cpg, /*h_parent*/NULL);
5435 +       cpg->flags = flags_orig;
5436 +       if (file) {
5437 +               if (!err)
5438 +                       err = au_reopen_nondir(file);
5439 +               hdp[0 + cpg->bsrc].hd_dentry = h_d_start;
5440 +       }
5441 +       hdp[0 + cpg->bdst].hd_dentry = h_d_dst;
5442 +       dinfo->di_bstart = cpg->bsrc;
5443 +       cpg->bsrc = bsrc_orig;
5444 +
5445 +       return err;
5446 +}
5447 +
5448 +static int au_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5449 +{
5450 +       int err;
5451 +       aufs_bindex_t bdst;
5452 +       struct au_dtime dt;
5453 +       struct dentry *dentry, *parent, *h_parent, *wh_dentry;
5454 +       struct au_branch *br;
5455 +       struct path h_path;
5456 +
5457 +       dentry = cpg->dentry;
5458 +       bdst = cpg->bdst;
5459 +       br = au_sbr(dentry->d_sb, bdst);
5460 +       parent = dget_parent(dentry);
5461 +       h_parent = au_h_dptr(parent, bdst);
5462 +       wh_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
5463 +       err = PTR_ERR(wh_dentry);
5464 +       if (IS_ERR(wh_dentry))
5465 +               goto out;
5466 +
5467 +       h_path.dentry = h_parent;
5468 +       h_path.mnt = au_br_mnt(br);
5469 +       au_dtime_store(&dt, parent, &h_path);
5470 +       err = au_do_cpup_wh(cpg, wh_dentry, file);
5471 +       if (unlikely(err))
5472 +               goto out_wh;
5473 +
5474 +       dget(wh_dentry);
5475 +       h_path.dentry = wh_dentry;
5476 +       if (!S_ISDIR(wh_dentry->d_inode->i_mode)) {
5477 +               /* no delegation since it is just created */
5478 +               err = vfsub_unlink(h_parent->d_inode, &h_path,
5479 +                                  /*delegated*/NULL, /*force*/0);
5480 +       } else
5481 +               err = vfsub_rmdir(h_parent->d_inode, &h_path);
5482 +       if (unlikely(err)) {
5483 +               AuIOErr("failed remove copied-up tmp file %pd(%d)\n",
5484 +                       wh_dentry, err);
5485 +               err = -EIO;
5486 +       }
5487 +       au_dtime_revert(&dt);
5488 +       au_set_hi_wh(dentry->d_inode, bdst, wh_dentry);
5489 +
5490 +out_wh:
5491 +       dput(wh_dentry);
5492 +out:
5493 +       dput(parent);
5494 +       return err;
5495 +}
5496 +
5497 +struct au_cpup_wh_args {
5498 +       int *errp;
5499 +       struct au_cp_generic *cpg;
5500 +       struct file *file;
5501 +};
5502 +
5503 +static void au_call_cpup_wh(void *args)
5504 +{
5505 +       struct au_cpup_wh_args *a = args;
5506 +
5507 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5508 +       *a->errp = au_cpup_wh(a->cpg, a->file);
5509 +       au_pin_hdir_release(a->cpg->pin);
5510 +}
5511 +
5512 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5513 +{
5514 +       int err, wkq_err;
5515 +       aufs_bindex_t bdst;
5516 +       struct dentry *dentry, *parent, *h_orph, *h_parent;
5517 +       struct inode *dir, *h_dir, *h_tmpdir;
5518 +       struct au_wbr *wbr;
5519 +       struct au_pin wh_pin, *pin_orig;
5520 +
5521 +       dentry = cpg->dentry;
5522 +       bdst = cpg->bdst;
5523 +       parent = dget_parent(dentry);
5524 +       dir = parent->d_inode;
5525 +       h_orph = NULL;
5526 +       h_parent = NULL;
5527 +       h_dir = au_igrab(au_h_iptr(dir, bdst));
5528 +       h_tmpdir = h_dir;
5529 +       pin_orig = NULL;
5530 +       if (!h_dir->i_nlink) {
5531 +               wbr = au_sbr(dentry->d_sb, bdst)->br_wbr;
5532 +               h_orph = wbr->wbr_orph;
5533 +
5534 +               h_parent = dget(au_h_dptr(parent, bdst));
5535 +               au_set_h_dptr(parent, bdst, dget(h_orph));
5536 +               h_tmpdir = h_orph->d_inode;
5537 +               au_set_h_iptr(dir, bdst, au_igrab(h_tmpdir), /*flags*/0);
5538 +
5539 +               mutex_lock_nested(&h_tmpdir->i_mutex, AuLsc_I_PARENT3);
5540 +               /* todo: au_h_open_pre()? */
5541 +
5542 +               pin_orig = cpg->pin;
5543 +               au_pin_init(&wh_pin, dentry, bdst, AuLsc_DI_PARENT,
5544 +                           AuLsc_I_PARENT3, cpg->pin->udba, AuPin_DI_LOCKED);
5545 +               cpg->pin = &wh_pin;
5546 +       }
5547 +
5548 +       if (!au_test_h_perm_sio(h_tmpdir, MAY_EXEC | MAY_WRITE)
5549 +           && !au_cpup_sio_test(cpg->pin, dentry->d_inode->i_mode))
5550 +               err = au_cpup_wh(cpg, file);
5551 +       else {
5552 +               struct au_cpup_wh_args args = {
5553 +                       .errp   = &err,
5554 +                       .cpg    = cpg,
5555 +                       .file   = file
5556 +               };
5557 +               wkq_err = au_wkq_wait(au_call_cpup_wh, &args);
5558 +               if (unlikely(wkq_err))
5559 +                       err = wkq_err;
5560 +       }
5561 +
5562 +       if (h_orph) {
5563 +               mutex_unlock(&h_tmpdir->i_mutex);
5564 +               /* todo: au_h_open_post()? */
5565 +               au_set_h_iptr(dir, bdst, au_igrab(h_dir), /*flags*/0);
5566 +               au_set_h_dptr(parent, bdst, h_parent);
5567 +               AuDebugOn(!pin_orig);
5568 +               cpg->pin = pin_orig;
5569 +       }
5570 +       iput(h_dir);
5571 +       dput(parent);
5572 +
5573 +       return err;
5574 +}
5575 +
5576 +/* ---------------------------------------------------------------------- */
5577 +
5578 +/*
5579 + * generic routine for both of copy-up and copy-down.
5580 + */
5581 +/* cf. revalidate function in file.c */
5582 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
5583 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
5584 +                        struct au_pin *pin,
5585 +                        struct dentry *h_parent, void *arg),
5586 +              void *arg)
5587 +{
5588 +       int err;
5589 +       struct au_pin pin;
5590 +       struct dentry *d, *parent, *h_parent, *real_parent;
5591 +
5592 +       err = 0;
5593 +       parent = dget_parent(dentry);
5594 +       if (IS_ROOT(parent))
5595 +               goto out;
5596 +
5597 +       au_pin_init(&pin, dentry, bdst, AuLsc_DI_PARENT2, AuLsc_I_PARENT2,
5598 +                   au_opt_udba(dentry->d_sb), AuPin_MNT_WRITE);
5599 +
5600 +       /* do not use au_dpage */
5601 +       real_parent = parent;
5602 +       while (1) {
5603 +               dput(parent);
5604 +               parent = dget_parent(dentry);
5605 +               h_parent = au_h_dptr(parent, bdst);
5606 +               if (h_parent)
5607 +                       goto out; /* success */
5608 +
5609 +               /* find top dir which is necessary to cpup */
5610 +               do {
5611 +                       d = parent;
5612 +                       dput(parent);
5613 +                       parent = dget_parent(d);
5614 +                       di_read_lock_parent3(parent, !AuLock_IR);
5615 +                       h_parent = au_h_dptr(parent, bdst);
5616 +                       di_read_unlock(parent, !AuLock_IR);
5617 +               } while (!h_parent);
5618 +
5619 +               if (d != real_parent)
5620 +                       di_write_lock_child3(d);
5621 +
5622 +               /* somebody else might create while we were sleeping */
5623 +               if (!au_h_dptr(d, bdst) || !au_h_dptr(d, bdst)->d_inode) {
5624 +                       if (au_h_dptr(d, bdst))
5625 +                               au_update_dbstart(d);
5626 +
5627 +                       au_pin_set_dentry(&pin, d);
5628 +                       err = au_do_pin(&pin);
5629 +                       if (!err) {
5630 +                               err = cp(d, bdst, &pin, h_parent, arg);
5631 +                               au_unpin(&pin);
5632 +                       }
5633 +               }
5634 +
5635 +               if (d != real_parent)
5636 +                       di_write_unlock(d);
5637 +               if (unlikely(err))
5638 +                       break;
5639 +       }
5640 +
5641 +out:
5642 +       dput(parent);
5643 +       return err;
5644 +}
5645 +
5646 +static int au_cpup_dir(struct dentry *dentry, aufs_bindex_t bdst,
5647 +                      struct au_pin *pin,
5648 +                      struct dentry *h_parent __maybe_unused ,
5649 +                      void *arg __maybe_unused)
5650 +{
5651 +       struct au_cp_generic cpg = {
5652 +               .dentry = dentry,
5653 +               .bdst   = bdst,
5654 +               .bsrc   = -1,
5655 +               .len    = 0,
5656 +               .pin    = pin,
5657 +               .flags  = AuCpup_DTIME
5658 +       };
5659 +       return au_sio_cpup_simple(&cpg);
5660 +}
5661 +
5662 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
5663 +{
5664 +       return au_cp_dirs(dentry, bdst, au_cpup_dir, NULL);
5665 +}
5666 +
5667 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
5668 +{
5669 +       int err;
5670 +       struct dentry *parent;
5671 +       struct inode *dir;
5672 +
5673 +       parent = dget_parent(dentry);
5674 +       dir = parent->d_inode;
5675 +       err = 0;
5676 +       if (au_h_iptr(dir, bdst))
5677 +               goto out;
5678 +
5679 +       di_read_unlock(parent, AuLock_IR);
5680 +       di_write_lock_parent(parent);
5681 +       /* someone else might change our inode while we were sleeping */
5682 +       if (!au_h_iptr(dir, bdst))
5683 +               err = au_cpup_dirs(dentry, bdst);
5684 +       di_downgrade_lock(parent, AuLock_IR);
5685 +
5686 +out:
5687 +       dput(parent);
5688 +       return err;
5689 +}
5690 diff -urN /usr/share/empty/fs/aufs/cpup.h linux/fs/aufs/cpup.h
5691 --- /usr/share/empty/fs/aufs/cpup.h     1970-01-01 01:00:00.000000000 +0100
5692 +++ linux/fs/aufs/cpup.h        2015-01-25 13:00:38.631047076 +0100
5693 @@ -0,0 +1,94 @@
5694 +/*
5695 + * Copyright (C) 2005-2014 Junjiro R. Okajima
5696 + *
5697 + * This program, aufs is free software; you can redistribute it and/or modify
5698 + * it under the terms of the GNU General Public License as published by
5699 + * the Free Software Foundation; either version 2 of the License, or
5700 + * (at your option) any later version.
5701 + *
5702 + * This program is distributed in the hope that it will be useful,
5703 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5704 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5705 + * GNU General Public License for more details.
5706 + *
5707 + * You should have received a copy of the GNU General Public License
5708 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5709 + */
5710 +
5711 +/*
5712 + * copy-up/down functions
5713 + */
5714 +
5715 +#ifndef __AUFS_CPUP_H__
5716 +#define __AUFS_CPUP_H__
5717 +
5718 +#ifdef __KERNEL__
5719 +
5720 +#include <linux/path.h>
5721 +
5722 +struct inode;
5723 +struct file;
5724 +struct au_pin;
5725 +
5726 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags);
5727 +void au_cpup_attr_timesizes(struct inode *inode);
5728 +void au_cpup_attr_nlink(struct inode *inode, int force);
5729 +void au_cpup_attr_changeable(struct inode *inode);
5730 +void au_cpup_igen(struct inode *inode, struct inode *h_inode);
5731 +void au_cpup_attr_all(struct inode *inode, int force);
5732 +
5733 +/* ---------------------------------------------------------------------- */
5734 +
5735 +struct au_cp_generic {
5736 +       struct dentry   *dentry;
5737 +       aufs_bindex_t   bdst, bsrc;
5738 +       loff_t          len;
5739 +       struct au_pin   *pin;
5740 +       unsigned int    flags;
5741 +};
5742 +
5743 +/* cpup flags */
5744 +#define AuCpup_DTIME           1               /* do dtime_store/revert */
5745 +#define AuCpup_KEEPLINO                (1 << 1)        /* do not clear the lower xino,
5746 +                                                  for link(2) */
5747 +#define AuCpup_RENAME          (1 << 2)        /* rename after cpup */
5748 +#define AuCpup_HOPEN           (1 << 3)        /* call h_open_pre/post() in
5749 +                                                  cpup */
5750 +#define AuCpup_OVERWRITE       (1 << 4)        /* allow overwriting the
5751 +                                                  existing entry */
5752 +#define AuCpup_RWDST           (1 << 5)        /* force write target even if
5753 +                                                  the branch is marked as RO */
5754 +
5755 +#define au_ftest_cpup(flags, name)     ((flags) & AuCpup_##name)
5756 +#define au_fset_cpup(flags, name) \
5757 +       do { (flags) |= AuCpup_##name; } while (0)
5758 +#define au_fclr_cpup(flags, name) \
5759 +       do { (flags) &= ~AuCpup_##name; } while (0)
5760 +
5761 +int au_copy_file(struct file *dst, struct file *src, loff_t len);
5762 +int au_sio_cpup_simple(struct au_cp_generic *cpg);
5763 +int au_sio_cpdown_simple(struct au_cp_generic *cpg);
5764 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file);
5765 +
5766 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
5767 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
5768 +                        struct au_pin *pin,
5769 +                        struct dentry *h_parent, void *arg),
5770 +              void *arg);
5771 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
5772 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
5773 +
5774 +/* ---------------------------------------------------------------------- */
5775 +
5776 +/* keep timestamps when copyup */
5777 +struct au_dtime {
5778 +       struct dentry *dt_dentry;
5779 +       struct path dt_h_path;
5780 +       struct timespec dt_atime, dt_mtime;
5781 +};
5782 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
5783 +                   struct path *h_path);
5784 +void au_dtime_revert(struct au_dtime *dt);
5785 +
5786 +#endif /* __KERNEL__ */
5787 +#endif /* __AUFS_CPUP_H__ */
5788 diff -urN /usr/share/empty/fs/aufs/dbgaufs.c linux/fs/aufs/dbgaufs.c
5789 --- /usr/share/empty/fs/aufs/dbgaufs.c  1970-01-01 01:00:00.000000000 +0100
5790 +++ linux/fs/aufs/dbgaufs.c     2015-01-25 13:00:38.631047076 +0100
5791 @@ -0,0 +1,432 @@
5792 +/*
5793 + * Copyright (C) 2005-2014 Junjiro R. Okajima
5794 + *
5795 + * This program, aufs is free software; you can redistribute it and/or modify
5796 + * it under the terms of the GNU General Public License as published by
5797 + * the Free Software Foundation; either version 2 of the License, or
5798 + * (at your option) any later version.
5799 + *
5800 + * This program is distributed in the hope that it will be useful,
5801 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5802 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5803 + * GNU General Public License for more details.
5804 + *
5805 + * You should have received a copy of the GNU General Public License
5806 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5807 + */
5808 +
5809 +/*
5810 + * debugfs interface
5811 + */
5812 +
5813 +#include <linux/debugfs.h>
5814 +#include "aufs.h"
5815 +
5816 +#ifndef CONFIG_SYSFS
5817 +#error DEBUG_FS depends upon SYSFS
5818 +#endif
5819 +
5820 +static struct dentry *dbgaufs;
5821 +static const mode_t dbgaufs_mode = S_IRUSR | S_IRGRP | S_IROTH;
5822 +
5823 +/* 20 is max digits length of ulong 64 */
5824 +struct dbgaufs_arg {
5825 +       int n;
5826 +       char a[20 * 4];
5827 +};
5828 +
5829 +/*
5830 + * common function for all XINO files
5831 + */
5832 +static int dbgaufs_xi_release(struct inode *inode __maybe_unused,
5833 +                             struct file *file)
5834 +{
5835 +       kfree(file->private_data);
5836 +       return 0;
5837 +}
5838 +
5839 +static int dbgaufs_xi_open(struct file *xf, struct file *file, int do_fcnt)
5840 +{
5841 +       int err;
5842 +       struct kstat st;
5843 +       struct dbgaufs_arg *p;
5844 +
5845 +       err = -ENOMEM;
5846 +       p = kmalloc(sizeof(*p), GFP_NOFS);
5847 +       if (unlikely(!p))
5848 +               goto out;
5849 +
5850 +       err = 0;
5851 +       p->n = 0;
5852 +       file->private_data = p;
5853 +       if (!xf)
5854 +               goto out;
5855 +
5856 +       err = vfs_getattr(&xf->f_path, &st);
5857 +       if (!err) {
5858 +               if (do_fcnt)
5859 +                       p->n = snprintf
5860 +                               (p->a, sizeof(p->a), "%ld, %llux%lu %lld\n",
5861 +                                (long)file_count(xf), st.blocks, st.blksize,
5862 +                                (long long)st.size);
5863 +               else
5864 +                       p->n = snprintf(p->a, sizeof(p->a), "%llux%lu %lld\n",
5865 +                                       st.blocks, st.blksize,
5866 +                                       (long long)st.size);
5867 +               AuDebugOn(p->n >= sizeof(p->a));
5868 +       } else {
5869 +               p->n = snprintf(p->a, sizeof(p->a), "err %d\n", err);
5870 +               err = 0;
5871 +       }
5872 +
5873 +out:
5874 +       return err;
5875 +
5876 +}
5877 +
5878 +static ssize_t dbgaufs_xi_read(struct file *file, char __user *buf,
5879 +                              size_t count, loff_t *ppos)
5880 +{
5881 +       struct dbgaufs_arg *p;
5882 +
5883 +       p = file->private_data;
5884 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
5885 +}
5886 +
5887 +/* ---------------------------------------------------------------------- */
5888 +
5889 +struct dbgaufs_plink_arg {
5890 +       int n;
5891 +       char a[];
5892 +};
5893 +
5894 +static int dbgaufs_plink_release(struct inode *inode __maybe_unused,
5895 +                                struct file *file)
5896 +{
5897 +       free_page((unsigned long)file->private_data);
5898 +       return 0;
5899 +}
5900 +
5901 +static int dbgaufs_plink_open(struct inode *inode, struct file *file)
5902 +{
5903 +       int err, i, limit;
5904 +       unsigned long n, sum;
5905 +       struct dbgaufs_plink_arg *p;
5906 +       struct au_sbinfo *sbinfo;
5907 +       struct super_block *sb;
5908 +       struct au_sphlhead *sphl;
5909 +
5910 +       err = -ENOMEM;
5911 +       p = (void *)get_zeroed_page(GFP_NOFS);
5912 +       if (unlikely(!p))
5913 +               goto out;
5914 +
5915 +       err = -EFBIG;
5916 +       sbinfo = inode->i_private;
5917 +       sb = sbinfo->si_sb;
5918 +       si_noflush_read_lock(sb);
5919 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
5920 +               limit = PAGE_SIZE - sizeof(p->n);
5921 +
5922 +               /* the number of buckets */
5923 +               n = snprintf(p->a + p->n, limit, "%d\n", AuPlink_NHASH);
5924 +               p->n += n;
5925 +               limit -= n;
5926 +
5927 +               sum = 0;
5928 +               for (i = 0, sphl = sbinfo->si_plink;
5929 +                    i < AuPlink_NHASH;
5930 +                    i++, sphl++) {
5931 +                       n = au_sphl_count(sphl);
5932 +                       sum += n;
5933 +
5934 +                       n = snprintf(p->a + p->n, limit, "%lu ", n);
5935 +                       p->n += n;
5936 +                       limit -= n;
5937 +                       if (unlikely(limit <= 0))
5938 +                               goto out_free;
5939 +               }
5940 +               p->a[p->n - 1] = '\n';
5941 +
5942 +               /* the sum of plinks */
5943 +               n = snprintf(p->a + p->n, limit, "%lu\n", sum);
5944 +               p->n += n;
5945 +               limit -= n;
5946 +               if (unlikely(limit <= 0))
5947 +                       goto out_free;
5948 +       } else {
5949 +#define str "1\n0\n0\n"
5950 +               p->n = sizeof(str) - 1;
5951 +               strcpy(p->a, str);
5952 +#undef str
5953 +       }
5954 +       si_read_unlock(sb);
5955 +
5956 +       err = 0;
5957 +       file->private_data = p;
5958 +       goto out; /* success */
5959 +
5960 +out_free:
5961 +       free_page((unsigned long)p);
5962 +out:
5963 +       return err;
5964 +}
5965 +
5966 +static ssize_t dbgaufs_plink_read(struct file *file, char __user *buf,
5967 +                                 size_t count, loff_t *ppos)
5968 +{
5969 +       struct dbgaufs_plink_arg *p;
5970 +
5971 +       p = file->private_data;
5972 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
5973 +}
5974 +
5975 +static const struct file_operations dbgaufs_plink_fop = {
5976 +       .owner          = THIS_MODULE,
5977 +       .open           = dbgaufs_plink_open,
5978 +       .release        = dbgaufs_plink_release,
5979 +       .read           = dbgaufs_plink_read
5980 +};
5981 +
5982 +/* ---------------------------------------------------------------------- */
5983 +
5984 +static int dbgaufs_xib_open(struct inode *inode, struct file *file)
5985 +{
5986 +       int err;
5987 +       struct au_sbinfo *sbinfo;
5988 +       struct super_block *sb;
5989 +
5990 +       sbinfo = inode->i_private;
5991 +       sb = sbinfo->si_sb;
5992 +       si_noflush_read_lock(sb);
5993 +       err = dbgaufs_xi_open(sbinfo->si_xib, file, /*do_fcnt*/0);
5994 +       si_read_unlock(sb);
5995 +       return err;
5996 +}
5997 +
5998 +static const struct file_operations dbgaufs_xib_fop = {
5999 +       .owner          = THIS_MODULE,
6000 +       .open           = dbgaufs_xib_open,
6001 +       .release        = dbgaufs_xi_release,
6002 +       .read           = dbgaufs_xi_read
6003 +};
6004 +
6005 +/* ---------------------------------------------------------------------- */
6006 +
6007 +#define DbgaufsXi_PREFIX "xi"
6008 +
6009 +static int dbgaufs_xino_open(struct inode *inode, struct file *file)
6010 +{
6011 +       int err;
6012 +       long l;
6013 +       struct au_sbinfo *sbinfo;
6014 +       struct super_block *sb;
6015 +       struct file *xf;
6016 +       struct qstr *name;
6017 +
6018 +       err = -ENOENT;
6019 +       xf = NULL;
6020 +       name = &file->f_dentry->d_name;
6021 +       if (unlikely(name->len < sizeof(DbgaufsXi_PREFIX)
6022 +                    || memcmp(name->name, DbgaufsXi_PREFIX,
6023 +                              sizeof(DbgaufsXi_PREFIX) - 1)))
6024 +               goto out;
6025 +       err = kstrtol(name->name + sizeof(DbgaufsXi_PREFIX) - 1, 10, &l);
6026 +       if (unlikely(err))
6027 +               goto out;
6028 +
6029 +       sbinfo = inode->i_private;
6030 +       sb = sbinfo->si_sb;
6031 +       si_noflush_read_lock(sb);
6032 +       if (l <= au_sbend(sb)) {
6033 +               xf = au_sbr(sb, (aufs_bindex_t)l)->br_xino.xi_file;
6034 +               err = dbgaufs_xi_open(xf, file, /*do_fcnt*/1);
6035 +       } else
6036 +               err = -ENOENT;
6037 +       si_read_unlock(sb);
6038 +
6039 +out:
6040 +       return err;
6041 +}
6042 +
6043 +static const struct file_operations dbgaufs_xino_fop = {
6044 +       .owner          = THIS_MODULE,
6045 +       .open           = dbgaufs_xino_open,
6046 +       .release        = dbgaufs_xi_release,
6047 +       .read           = dbgaufs_xi_read
6048 +};
6049 +
6050 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
6051 +{
6052 +       aufs_bindex_t bend;
6053 +       struct au_branch *br;
6054 +       struct au_xino_file *xi;
6055 +
6056 +       if (!au_sbi(sb)->si_dbgaufs)
6057 +               return;
6058 +
6059 +       bend = au_sbend(sb);
6060 +       for (; bindex <= bend; bindex++) {
6061 +               br = au_sbr(sb, bindex);
6062 +               xi = &br->br_xino;
6063 +               debugfs_remove(xi->xi_dbgaufs);
6064 +               xi->xi_dbgaufs = NULL;
6065 +       }
6066 +}
6067 +
6068 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
6069 +{
6070 +       struct au_sbinfo *sbinfo;
6071 +       struct dentry *parent;
6072 +       struct au_branch *br;
6073 +       struct au_xino_file *xi;
6074 +       aufs_bindex_t bend;
6075 +       char name[sizeof(DbgaufsXi_PREFIX) + 5]; /* "xi" bindex NULL */
6076 +
6077 +       sbinfo = au_sbi(sb);
6078 +       parent = sbinfo->si_dbgaufs;
6079 +       if (!parent)
6080 +               return;
6081 +
6082 +       bend = au_sbend(sb);
6083 +       for (; bindex <= bend; bindex++) {
6084 +               snprintf(name, sizeof(name), DbgaufsXi_PREFIX "%d", bindex);
6085 +               br = au_sbr(sb, bindex);
6086 +               xi = &br->br_xino;
6087 +               AuDebugOn(xi->xi_dbgaufs);
6088 +               xi->xi_dbgaufs = debugfs_create_file(name, dbgaufs_mode, parent,
6089 +                                                    sbinfo, &dbgaufs_xino_fop);
6090 +               /* ignore an error */
6091 +               if (unlikely(!xi->xi_dbgaufs))
6092 +                       AuWarn1("failed %s under debugfs\n", name);
6093 +       }
6094 +}
6095 +
6096 +/* ---------------------------------------------------------------------- */
6097 +
6098 +#ifdef CONFIG_AUFS_EXPORT
6099 +static int dbgaufs_xigen_open(struct inode *inode, struct file *file)
6100 +{
6101 +       int err;
6102 +       struct au_sbinfo *sbinfo;
6103 +       struct super_block *sb;
6104 +
6105 +       sbinfo = inode->i_private;
6106 +       sb = sbinfo->si_sb;
6107 +       si_noflush_read_lock(sb);
6108 +       err = dbgaufs_xi_open(sbinfo->si_xigen, file, /*do_fcnt*/0);
6109 +       si_read_unlock(sb);
6110 +       return err;
6111 +}
6112 +
6113 +static const struct file_operations dbgaufs_xigen_fop = {
6114 +       .owner          = THIS_MODULE,
6115 +       .open           = dbgaufs_xigen_open,
6116 +       .release        = dbgaufs_xi_release,
6117 +       .read           = dbgaufs_xi_read
6118 +};
6119 +
6120 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6121 +{
6122 +       int err;
6123 +
6124 +       /*
6125 +        * This function is a dynamic '__init' function actually,
6126 +        * so the tiny check for si_rwsem is unnecessary.
6127 +        */
6128 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6129 +
6130 +       err = -EIO;
6131 +       sbinfo->si_dbgaufs_xigen = debugfs_create_file
6132 +               ("xigen", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6133 +                &dbgaufs_xigen_fop);
6134 +       if (sbinfo->si_dbgaufs_xigen)
6135 +               err = 0;
6136 +
6137 +       return err;
6138 +}
6139 +#else
6140 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6141 +{
6142 +       return 0;
6143 +}
6144 +#endif /* CONFIG_AUFS_EXPORT */
6145 +
6146 +/* ---------------------------------------------------------------------- */
6147 +
6148 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo)
6149 +{
6150 +       /*
6151 +        * This function is a dynamic '__init' function actually,
6152 +        * so the tiny check for si_rwsem is unnecessary.
6153 +        */
6154 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6155 +
6156 +       debugfs_remove_recursive(sbinfo->si_dbgaufs);
6157 +       sbinfo->si_dbgaufs = NULL;
6158 +       kobject_put(&sbinfo->si_kobj);
6159 +}
6160 +
6161 +int dbgaufs_si_init(struct au_sbinfo *sbinfo)
6162 +{
6163 +       int err;
6164 +       char name[SysaufsSiNameLen];
6165 +
6166 +       /*
6167 +        * This function is a dynamic '__init' function actually,
6168 +        * so the tiny check for si_rwsem is unnecessary.
6169 +        */
6170 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6171 +
6172 +       err = -ENOENT;
6173 +       if (!dbgaufs) {
6174 +               AuErr1("/debug/aufs is uninitialized\n");
6175 +               goto out;
6176 +       }
6177 +
6178 +       err = -EIO;
6179 +       sysaufs_name(sbinfo, name);
6180 +       sbinfo->si_dbgaufs = debugfs_create_dir(name, dbgaufs);
6181 +       if (unlikely(!sbinfo->si_dbgaufs))
6182 +               goto out;
6183 +       kobject_get(&sbinfo->si_kobj);
6184 +
6185 +       sbinfo->si_dbgaufs_xib = debugfs_create_file
6186 +               ("xib", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6187 +                &dbgaufs_xib_fop);
6188 +       if (unlikely(!sbinfo->si_dbgaufs_xib))
6189 +               goto out_dir;
6190 +
6191 +       sbinfo->si_dbgaufs_plink = debugfs_create_file
6192 +               ("plink", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6193 +                &dbgaufs_plink_fop);
6194 +       if (unlikely(!sbinfo->si_dbgaufs_plink))
6195 +               goto out_dir;
6196 +
6197 +       err = dbgaufs_xigen_init(sbinfo);
6198 +       if (!err)
6199 +               goto out; /* success */
6200 +
6201 +out_dir:
6202 +       dbgaufs_si_fin(sbinfo);
6203 +out:
6204 +       return err;
6205 +}
6206 +
6207 +/* ---------------------------------------------------------------------- */
6208 +
6209 +void dbgaufs_fin(void)
6210 +{
6211 +       debugfs_remove(dbgaufs);
6212 +}
6213 +
6214 +int __init dbgaufs_init(void)
6215 +{
6216 +       int err;
6217 +
6218 +       err = -EIO;
6219 +       dbgaufs = debugfs_create_dir(AUFS_NAME, NULL);
6220 +       if (dbgaufs)
6221 +               err = 0;
6222 +       return err;
6223 +}
6224 diff -urN /usr/share/empty/fs/aufs/dbgaufs.h linux/fs/aufs/dbgaufs.h
6225 --- /usr/share/empty/fs/aufs/dbgaufs.h  1970-01-01 01:00:00.000000000 +0100
6226 +++ linux/fs/aufs/dbgaufs.h     2015-01-25 13:00:38.631047076 +0100
6227 @@ -0,0 +1,48 @@
6228 +/*
6229 + * Copyright (C) 2005-2014 Junjiro R. Okajima
6230 + *
6231 + * This program, aufs is free software; you can redistribute it and/or modify
6232 + * it under the terms of the GNU General Public License as published by
6233 + * the Free Software Foundation; either version 2 of the License, or
6234 + * (at your option) any later version.
6235 + *
6236 + * This program is distributed in the hope that it will be useful,
6237 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6238 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6239 + * GNU General Public License for more details.
6240 + *
6241 + * You should have received a copy of the GNU General Public License
6242 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6243 + */
6244 +
6245 +/*
6246 + * debugfs interface
6247 + */
6248 +
6249 +#ifndef __DBGAUFS_H__
6250 +#define __DBGAUFS_H__
6251 +
6252 +#ifdef __KERNEL__
6253 +
6254 +struct super_block;
6255 +struct au_sbinfo;
6256 +
6257 +#ifdef CONFIG_DEBUG_FS
6258 +/* dbgaufs.c */
6259 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
6260 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
6261 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo);
6262 +int dbgaufs_si_init(struct au_sbinfo *sbinfo);
6263 +void dbgaufs_fin(void);
6264 +int __init dbgaufs_init(void);
6265 +#else
6266 +AuStubVoid(dbgaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
6267 +AuStubVoid(dbgaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
6268 +AuStubVoid(dbgaufs_si_fin, struct au_sbinfo *sbinfo)
6269 +AuStubInt0(dbgaufs_si_init, struct au_sbinfo *sbinfo)
6270 +AuStubVoid(dbgaufs_fin, void)
6271 +AuStubInt0(__init dbgaufs_init, void)
6272 +#endif /* CONFIG_DEBUG_FS */
6273 +
6274 +#endif /* __KERNEL__ */
6275 +#endif /* __DBGAUFS_H__ */
6276 diff -urN /usr/share/empty/fs/aufs/dcsub.c linux/fs/aufs/dcsub.c
6277 --- /usr/share/empty/fs/aufs/dcsub.c    1970-01-01 01:00:00.000000000 +0100
6278 +++ linux/fs/aufs/dcsub.c       2015-01-25 13:00:38.631047076 +0100
6279 @@ -0,0 +1,224 @@
6280 +/*
6281 + * Copyright (C) 2005-2014 Junjiro R. Okajima
6282 + *
6283 + * This program, aufs is free software; you can redistribute it and/or modify
6284 + * it under the terms of the GNU General Public License as published by
6285 + * the Free Software Foundation; either version 2 of the License, or
6286 + * (at your option) any later version.
6287 + *
6288 + * This program is distributed in the hope that it will be useful,
6289 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6290 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6291 + * GNU General Public License for more details.
6292 + *
6293 + * You should have received a copy of the GNU General Public License
6294 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6295 + */
6296 +
6297 +/*
6298 + * sub-routines for dentry cache
6299 + */
6300 +
6301 +#include "aufs.h"
6302 +
6303 +static void au_dpage_free(struct au_dpage *dpage)
6304 +{
6305 +       int i;
6306 +       struct dentry **p;
6307 +
6308 +       p = dpage->dentries;
6309 +       for (i = 0; i < dpage->ndentry; i++)
6310 +               dput(*p++);
6311 +       free_page((unsigned long)dpage->dentries);
6312 +}
6313 +
6314 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp)
6315 +{
6316 +       int err;
6317 +       void *p;
6318 +
6319 +       err = -ENOMEM;
6320 +       dpages->dpages = kmalloc(sizeof(*dpages->dpages), gfp);
6321 +       if (unlikely(!dpages->dpages))
6322 +               goto out;
6323 +
6324 +       p = (void *)__get_free_page(gfp);
6325 +       if (unlikely(!p))
6326 +               goto out_dpages;
6327 +
6328 +       dpages->dpages[0].ndentry = 0;
6329 +       dpages->dpages[0].dentries = p;
6330 +       dpages->ndpage = 1;
6331 +       return 0; /* success */
6332 +
6333 +out_dpages:
6334 +       kfree(dpages->dpages);
6335 +out:
6336 +       return err;
6337 +}
6338 +
6339 +void au_dpages_free(struct au_dcsub_pages *dpages)
6340 +{
6341 +       int i;
6342 +       struct au_dpage *p;
6343 +
6344 +       p = dpages->dpages;
6345 +       for (i = 0; i < dpages->ndpage; i++)
6346 +               au_dpage_free(p++);
6347 +       kfree(dpages->dpages);
6348 +}
6349 +
6350 +static int au_dpages_append(struct au_dcsub_pages *dpages,
6351 +                           struct dentry *dentry, gfp_t gfp)
6352 +{
6353 +       int err, sz;
6354 +       struct au_dpage *dpage;
6355 +       void *p;
6356 +
6357 +       dpage = dpages->dpages + dpages->ndpage - 1;
6358 +       sz = PAGE_SIZE / sizeof(dentry);
6359 +       if (unlikely(dpage->ndentry >= sz)) {
6360 +               AuLabel(new dpage);
6361 +               err = -ENOMEM;
6362 +               sz = dpages->ndpage * sizeof(*dpages->dpages);
6363 +               p = au_kzrealloc(dpages->dpages, sz,
6364 +                                sz + sizeof(*dpages->dpages), gfp);
6365 +               if (unlikely(!p))
6366 +                       goto out;
6367 +
6368 +               dpages->dpages = p;
6369 +               dpage = dpages->dpages + dpages->ndpage;
6370 +               p = (void *)__get_free_page(gfp);
6371 +               if (unlikely(!p))
6372 +                       goto out;
6373 +
6374 +               dpage->ndentry = 0;
6375 +               dpage->dentries = p;
6376 +               dpages->ndpage++;
6377 +       }
6378 +
6379 +       AuDebugOn(au_dcount(dentry) <= 0);
6380 +       dpage->dentries[dpage->ndentry++] = dget_dlock(dentry);
6381 +       return 0; /* success */
6382 +
6383 +out:
6384 +       return err;
6385 +}
6386 +
6387 +/* todo: BAD approach */
6388 +/* copied from linux/fs/dcache.c */
6389 +enum d_walk_ret {
6390 +       D_WALK_CONTINUE,
6391 +       D_WALK_QUIT,
6392 +       D_WALK_NORETRY,
6393 +       D_WALK_SKIP,
6394 +};
6395 +
6396 +extern void d_walk(struct dentry *parent, void *data,
6397 +                  enum d_walk_ret (*enter)(void *, struct dentry *),
6398 +                  void (*finish)(void *));
6399 +
6400 +struct ac_dpages_arg {
6401 +       int err;
6402 +       struct au_dcsub_pages *dpages;
6403 +       struct super_block *sb;
6404 +       au_dpages_test test;
6405 +       void *arg;
6406 +};
6407 +
6408 +static enum d_walk_ret au_call_dpages_append(void *_arg, struct dentry *dentry)
6409 +{
6410 +       enum d_walk_ret ret;
6411 +       struct ac_dpages_arg *arg = _arg;
6412 +
6413 +       ret = D_WALK_CONTINUE;
6414 +       if (dentry->d_sb == arg->sb
6415 +           && !IS_ROOT(dentry)
6416 +           && au_dcount(dentry) > 0
6417 +           && au_di(dentry)
6418 +           && (!arg->test || arg->test(dentry, arg->arg))) {
6419 +               arg->err = au_dpages_append(arg->dpages, dentry, GFP_ATOMIC);
6420 +               if (unlikely(arg->err))
6421 +                       ret = D_WALK_QUIT;
6422 +       }
6423 +
6424 +       return ret;
6425 +}
6426 +
6427 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6428 +                  au_dpages_test test, void *arg)
6429 +{
6430 +       struct ac_dpages_arg args = {
6431 +               .err    = 0,
6432 +               .dpages = dpages,
6433 +               .sb     = root->d_sb,
6434 +               .test   = test,
6435 +               .arg    = arg
6436 +       };
6437 +
6438 +       d_walk(root, &args, au_call_dpages_append, NULL);
6439 +
6440 +       return args.err;
6441 +}
6442 +
6443 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6444 +                      int do_include, au_dpages_test test, void *arg)
6445 +{
6446 +       int err;
6447 +
6448 +       err = 0;
6449 +       write_seqlock(&rename_lock);
6450 +       spin_lock(&dentry->d_lock);
6451 +       if (do_include
6452 +           && au_dcount(dentry) > 0
6453 +           && (!test || test(dentry, arg)))
6454 +               err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6455 +       spin_unlock(&dentry->d_lock);
6456 +       if (unlikely(err))
6457 +               goto out;
6458 +
6459 +       /*
6460 +        * RCU for vfsmount is unnecessary since this is a traverse in a single
6461 +        * mount
6462 +        */
6463 +       while (!IS_ROOT(dentry)) {
6464 +               dentry = dentry->d_parent; /* rename_lock is locked */
6465 +               spin_lock(&dentry->d_lock);
6466 +               if (au_dcount(dentry) > 0
6467 +                   && (!test || test(dentry, arg)))
6468 +                       err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6469 +               spin_unlock(&dentry->d_lock);
6470 +               if (unlikely(err))
6471 +                       break;
6472 +       }
6473 +
6474 +out:
6475 +       write_sequnlock(&rename_lock);
6476 +       return err;
6477 +}
6478 +
6479 +static inline int au_dcsub_dpages_aufs(struct dentry *dentry, void *arg)
6480 +{
6481 +       return au_di(dentry) && dentry->d_sb == arg;
6482 +}
6483 +
6484 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6485 +                           struct dentry *dentry, int do_include)
6486 +{
6487 +       return au_dcsub_pages_rev(dpages, dentry, do_include,
6488 +                                 au_dcsub_dpages_aufs, dentry->d_sb);
6489 +}
6490 +
6491 +int au_test_subdir(struct dentry *d1, struct dentry *d2)
6492 +{
6493 +       struct path path[2] = {
6494 +               {
6495 +                       .dentry = d1
6496 +               },
6497 +               {
6498 +                       .dentry = d2
6499 +               }
6500 +       };
6501 +
6502 +       return path_is_under(path + 0, path + 1);
6503 +}
6504 diff -urN /usr/share/empty/fs/aufs/dcsub.h linux/fs/aufs/dcsub.h
6505 --- /usr/share/empty/fs/aufs/dcsub.h    1970-01-01 01:00:00.000000000 +0100
6506 +++ linux/fs/aufs/dcsub.h       2015-01-25 13:00:38.631047076 +0100
6507 @@ -0,0 +1,125 @@
6508 +/*
6509 + * Copyright (C) 2005-2014 Junjiro R. Okajima
6510 + *
6511 + * This program, aufs is free software; you can redistribute it and/or modify
6512 + * it under the terms of the GNU General Public License as published by
6513 + * the Free Software Foundation; either version 2 of the License, or
6514 + * (at your option) any later version.
6515 + *
6516 + * This program is distributed in the hope that it will be useful,
6517 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6518 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6519 + * GNU General Public License for more details.
6520 + *
6521 + * You should have received a copy of the GNU General Public License
6522 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6523 + */
6524 +
6525 +/*
6526 + * sub-routines for dentry cache
6527 + */
6528 +
6529 +#ifndef __AUFS_DCSUB_H__
6530 +#define __AUFS_DCSUB_H__
6531 +
6532 +#ifdef __KERNEL__
6533 +
6534 +#include <linux/dcache.h>
6535 +#include <linux/fs.h>
6536 +
6537 +struct dentry;
6538 +
6539 +struct au_dpage {
6540 +       int ndentry;
6541 +       struct dentry **dentries;
6542 +};
6543 +
6544 +struct au_dcsub_pages {
6545 +       int ndpage;
6546 +       struct au_dpage *dpages;
6547 +};
6548 +
6549 +/* ---------------------------------------------------------------------- */
6550 +
6551 +/* dcsub.c */
6552 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp);
6553 +void au_dpages_free(struct au_dcsub_pages *dpages);
6554 +typedef int (*au_dpages_test)(struct dentry *dentry, void *arg);
6555 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6556 +                  au_dpages_test test, void *arg);
6557 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6558 +                      int do_include, au_dpages_test test, void *arg);
6559 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6560 +                           struct dentry *dentry, int do_include);
6561 +int au_test_subdir(struct dentry *d1, struct dentry *d2);
6562 +
6563 +/* ---------------------------------------------------------------------- */
6564 +
6565 +/*
6566 + * todo: in linux-3.13, several similar (but faster) helpers are added to
6567 + * include/linux/dcache.h. Try them (in the future).
6568 + */
6569 +
6570 +static inline int au_d_hashed_positive(struct dentry *d)
6571 +{
6572 +       int err;
6573 +       struct inode *inode = d->d_inode;
6574 +
6575 +       err = 0;
6576 +       if (unlikely(d_unhashed(d) || !inode || !inode->i_nlink))
6577 +               err = -ENOENT;
6578 +       return err;
6579 +}
6580 +
6581 +static inline int au_d_linkable(struct dentry *d)
6582 +{
6583 +       int err;
6584 +       struct inode *inode = d->d_inode;
6585 +
6586 +       err = au_d_hashed_positive(d);
6587 +       if (err
6588 +           && inode
6589 +           && (inode->i_state & I_LINKABLE))
6590 +               err = 0;
6591 +       return err;
6592 +}
6593 +
6594 +static inline int au_d_alive(struct dentry *d)
6595 +{
6596 +       int err;
6597 +       struct inode *inode;
6598 +
6599 +       err = 0;
6600 +       if (!IS_ROOT(d))
6601 +               err = au_d_hashed_positive(d);
6602 +       else {
6603 +               inode = d->d_inode;
6604 +               if (unlikely(d_unlinked(d) || !inode || !inode->i_nlink))
6605 +                       err = -ENOENT;
6606 +       }
6607 +       return err;
6608 +}
6609 +
6610 +static inline int au_alive_dir(struct dentry *d)
6611 +{
6612 +       int err;
6613 +
6614 +       err = au_d_alive(d);
6615 +       if (unlikely(err || IS_DEADDIR(d->d_inode)))
6616 +               err = -ENOENT;
6617 +       return err;
6618 +}
6619 +
6620 +static inline int au_qstreq(struct qstr *a, struct qstr *b)
6621 +{
6622 +       return a->len == b->len
6623 +               && !memcmp(a->name, b->name, a->len);
6624 +}
6625 +
6626 +static inline int au_dcount(struct dentry *d)
6627 +{
6628 +       return (int)d_count(d);
6629 +}
6630 +
6631 +#endif /* __KERNEL__ */
6632 +#endif /* __AUFS_DCSUB_H__ */
6633 diff -urN /usr/share/empty/fs/aufs/debug.c linux/fs/aufs/debug.c
6634 --- /usr/share/empty/fs/aufs/debug.c    1970-01-01 01:00:00.000000000 +0100
6635 +++ linux/fs/aufs/debug.c       2015-01-25 13:00:38.631047076 +0100
6636 @@ -0,0 +1,520 @@
6637 +/*
6638 + * Copyright (C) 2005-2014 Junjiro R. Okajima
6639 + *
6640 + * This program, aufs is free software; you can redistribute it and/or modify
6641 + * it under the terms of the GNU General Public License as published by
6642 + * the Free Software Foundation; either version 2 of the License, or
6643 + * (at your option) any later version.
6644 + *
6645 + * This program is distributed in the hope that it will be useful,
6646 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6647 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6648 + * GNU General Public License for more details.
6649 + *
6650 + * You should have received a copy of the GNU General Public License
6651 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6652 + */
6653 +
6654 +/*
6655 + * debug print functions
6656 + */
6657 +
6658 +#include <linux/vt_kern.h>
6659 +#include "aufs.h"
6660 +
6661 +/* Returns 0, or -errno.  arg is in kp->arg. */
6662 +static int param_atomic_t_set(const char *val, const struct kernel_param *kp)
6663 +{
6664 +       int err, n;
6665 +
6666 +       err = kstrtoint(val, 0, &n);
6667 +       if (!err) {
6668 +               if (n > 0)
6669 +                       au_debug_on();
6670 +               else
6671 +                       au_debug_off();
6672 +       }
6673 +       return err;
6674 +}
6675 +
6676 +/* Returns length written or -errno.  Buffer is 4k (ie. be short!) */
6677 +static int param_atomic_t_get(char *buffer, const struct kernel_param *kp)
6678 +{
6679 +       atomic_t *a;
6680 +
6681 +       a = kp->arg;
6682 +       return sprintf(buffer, "%d", atomic_read(a));
6683 +}
6684 +
6685 +static struct kernel_param_ops param_ops_atomic_t = {
6686 +       .set = param_atomic_t_set,
6687 +       .get = param_atomic_t_get
6688 +       /* void (*free)(void *arg) */
6689 +};
6690 +
6691 +atomic_t aufs_debug = ATOMIC_INIT(0);
6692 +MODULE_PARM_DESC(debug, "debug print");
6693 +module_param_named(debug, aufs_debug, atomic_t, S_IRUGO | S_IWUSR | S_IWGRP);
6694 +
6695 +DEFINE_MUTEX(au_dbg_mtx);      /* just to serialize the dbg msgs */
6696 +char *au_plevel = KERN_DEBUG;
6697 +#define dpri(fmt, ...) do {                                    \
6698 +       if ((au_plevel                                          \
6699 +            && strcmp(au_plevel, KERN_DEBUG))                  \
6700 +           || au_debug_test())                                 \
6701 +               printk("%s" fmt, au_plevel, ##__VA_ARGS__);     \
6702 +} while (0)
6703 +
6704 +/* ---------------------------------------------------------------------- */
6705 +
6706 +void au_dpri_whlist(struct au_nhash *whlist)
6707 +{
6708 +       unsigned long ul, n;
6709 +       struct hlist_head *head;
6710 +       struct au_vdir_wh *pos;
6711 +
6712 +       n = whlist->nh_num;
6713 +       head = whlist->nh_head;
6714 +       for (ul = 0; ul < n; ul++) {
6715 +               hlist_for_each_entry(pos, head, wh_hash)
6716 +                       dpri("b%d, %.*s, %d\n",
6717 +                            pos->wh_bindex,
6718 +                            pos->wh_str.len, pos->wh_str.name,
6719 +                            pos->wh_str.len);
6720 +               head++;
6721 +       }
6722 +}
6723 +
6724 +void au_dpri_vdir(struct au_vdir *vdir)
6725 +{
6726 +       unsigned long ul;
6727 +       union au_vdir_deblk_p p;
6728 +       unsigned char *o;
6729 +
6730 +       if (!vdir || IS_ERR(vdir)) {
6731 +               dpri("err %ld\n", PTR_ERR(vdir));
6732 +               return;
6733 +       }
6734 +
6735 +       dpri("deblk %u, nblk %lu, deblk %p, last{%lu, %p}, ver %lu\n",
6736 +            vdir->vd_deblk_sz, vdir->vd_nblk, vdir->vd_deblk,
6737 +            vdir->vd_last.ul, vdir->vd_last.p.deblk, vdir->vd_version);
6738 +       for (ul = 0; ul < vdir->vd_nblk; ul++) {
6739 +               p.deblk = vdir->vd_deblk[ul];
6740 +               o = p.deblk;
6741 +               dpri("[%lu]: %p\n", ul, o);
6742 +       }
6743 +}
6744 +
6745 +static int do_pri_inode(aufs_bindex_t bindex, struct inode *inode, int hn,
6746 +                       struct dentry *wh)
6747 +{
6748 +       char *n = NULL;
6749 +       int l = 0;
6750 +
6751 +       if (!inode || IS_ERR(inode)) {
6752 +               dpri("i%d: err %ld\n", bindex, PTR_ERR(inode));
6753 +               return -1;
6754 +       }
6755 +
6756 +       /* the type of i_blocks depends upon CONFIG_LBDAF */
6757 +       BUILD_BUG_ON(sizeof(inode->i_blocks) != sizeof(unsigned long)
6758 +                    && sizeof(inode->i_blocks) != sizeof(u64));
6759 +       if (wh) {
6760 +               n = (void *)wh->d_name.name;
6761 +               l = wh->d_name.len;
6762 +       }
6763 +
6764 +       dpri("i%d: %p, i%lu, %s, cnt %d, nl %u, 0%o, sz %llu, blk %llu,"
6765 +            " hn %d, ct %lld, np %lu, st 0x%lx, f 0x%x, v %llu, g %x%s%.*s\n",
6766 +            bindex, inode,
6767 +            inode->i_ino, inode->i_sb ? au_sbtype(inode->i_sb) : "??",
6768 +            atomic_read(&inode->i_count), inode->i_nlink, inode->i_mode,
6769 +            i_size_read(inode), (unsigned long long)inode->i_blocks,
6770 +            hn, (long long)timespec_to_ns(&inode->i_ctime) & 0x0ffff,
6771 +            inode->i_mapping ? inode->i_mapping->nrpages : 0,
6772 +            inode->i_state, inode->i_flags, inode->i_version,
6773 +            inode->i_generation,
6774 +            l ? ", wh " : "", l, n);
6775 +       return 0;
6776 +}
6777 +
6778 +void au_dpri_inode(struct inode *inode)
6779 +{
6780 +       struct au_iinfo *iinfo;
6781 +       aufs_bindex_t bindex;
6782 +       int err, hn;
6783 +
6784 +       err = do_pri_inode(-1, inode, -1, NULL);
6785 +       if (err || !au_test_aufs(inode->i_sb))
6786 +               return;
6787 +
6788 +       iinfo = au_ii(inode);
6789 +       if (!iinfo)
6790 +               return;
6791 +       dpri("i-1: bstart %d, bend %d, gen %d\n",
6792 +            iinfo->ii_bstart, iinfo->ii_bend, au_iigen(inode, NULL));
6793 +       if (iinfo->ii_bstart < 0)
6794 +               return;
6795 +       hn = 0;
6796 +       for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend; bindex++) {
6797 +               hn = !!au_hn(iinfo->ii_hinode + bindex);
6798 +               do_pri_inode(bindex, iinfo->ii_hinode[0 + bindex].hi_inode, hn,
6799 +                            iinfo->ii_hinode[0 + bindex].hi_whdentry);
6800 +       }
6801 +}
6802 +
6803 +void au_dpri_dalias(struct inode *inode)
6804 +{
6805 +       struct dentry *d;
6806 +
6807 +       spin_lock(&inode->i_lock);
6808 +       hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias)
6809 +               au_dpri_dentry(d);
6810 +       spin_unlock(&inode->i_lock);
6811 +}
6812 +
6813 +static int do_pri_dentry(aufs_bindex_t bindex, struct dentry *dentry)
6814 +{
6815 +       struct dentry *wh = NULL;
6816 +       int hn;
6817 +       struct au_iinfo *iinfo;
6818 +
6819 +       if (!dentry || IS_ERR(dentry)) {
6820 +               dpri("d%d: err %ld\n", bindex, PTR_ERR(dentry));
6821 +               return -1;
6822 +       }
6823 +       /* do not call dget_parent() here */
6824 +       /* note: access d_xxx without d_lock */
6825 +       dpri("d%d: %p, %pd2?, %s, cnt %d, flags 0x%x, %shashed\n",
6826 +            bindex, dentry, dentry,
6827 +            dentry->d_sb ? au_sbtype(dentry->d_sb) : "??",
6828 +            au_dcount(dentry), dentry->d_flags,
6829 +            d_unhashed(dentry) ? "un" : "");
6830 +       hn = -1;
6831 +       if (bindex >= 0 && dentry->d_inode && au_test_aufs(dentry->d_sb)) {
6832 +               iinfo = au_ii(dentry->d_inode);
6833 +               if (iinfo) {
6834 +                       hn = !!au_hn(iinfo->ii_hinode + bindex);
6835 +                       wh = iinfo->ii_hinode[0 + bindex].hi_whdentry;
6836 +               }
6837 +       }
6838 +       do_pri_inode(bindex, dentry->d_inode, hn, wh);
6839 +       return 0;
6840 +}
6841 +
6842 +void au_dpri_dentry(struct dentry *dentry)
6843 +{
6844 +       struct au_dinfo *dinfo;
6845 +       aufs_bindex_t bindex;
6846 +       int err;
6847 +       struct au_hdentry *hdp;
6848 +
6849 +       err = do_pri_dentry(-1, dentry);
6850 +       if (err || !au_test_aufs(dentry->d_sb))
6851 +               return;
6852 +
6853 +       dinfo = au_di(dentry);
6854 +       if (!dinfo)
6855 +               return;
6856 +       dpri("d-1: bstart %d, bend %d, bwh %d, bdiropq %d, gen %d, tmp %d\n",
6857 +            dinfo->di_bstart, dinfo->di_bend,
6858 +            dinfo->di_bwh, dinfo->di_bdiropq, au_digen(dentry),
6859 +            dinfo->di_tmpfile);
6860 +       if (dinfo->di_bstart < 0)
6861 +               return;
6862 +       hdp = dinfo->di_hdentry;
6863 +       for (bindex = dinfo->di_bstart; bindex <= dinfo->di_bend; bindex++)
6864 +               do_pri_dentry(bindex, hdp[0 + bindex].hd_dentry);
6865 +}
6866 +
6867 +static int do_pri_file(aufs_bindex_t bindex, struct file *file)
6868 +{
6869 +       char a[32];
6870 +
6871 +       if (!file || IS_ERR(file)) {
6872 +               dpri("f%d: err %ld\n", bindex, PTR_ERR(file));
6873 +               return -1;
6874 +       }
6875 +       a[0] = 0;
6876 +       if (bindex < 0
6877 +           && file->f_dentry
6878 +           && au_test_aufs(file->f_dentry->d_sb)
6879 +           && au_fi(file))
6880 +               snprintf(a, sizeof(a), ", gen %d, mmapped %d",
6881 +                        au_figen(file), atomic_read(&au_fi(file)->fi_mmapped));
6882 +       dpri("f%d: mode 0x%x, flags 0%o, cnt %ld, v %llu, pos %llu%s\n",
6883 +            bindex, file->f_mode, file->f_flags, (long)file_count(file),
6884 +            file->f_version, file->f_pos, a);
6885 +       if (file->f_dentry)
6886 +               do_pri_dentry(bindex, file->f_dentry);
6887 +       return 0;
6888 +}
6889 +
6890 +void au_dpri_file(struct file *file)
6891 +{
6892 +       struct au_finfo *finfo;
6893 +       struct au_fidir *fidir;
6894 +       struct au_hfile *hfile;
6895 +       aufs_bindex_t bindex;
6896 +       int err;
6897 +
6898 +       err = do_pri_file(-1, file);
6899 +       if (err || !file->f_dentry || !au_test_aufs(file->f_dentry->d_sb))
6900 +               return;
6901 +
6902 +       finfo = au_fi(file);
6903 +       if (!finfo)
6904 +               return;
6905 +       if (finfo->fi_btop < 0)
6906 +               return;
6907 +       fidir = finfo->fi_hdir;
6908 +       if (!fidir)
6909 +               do_pri_file(finfo->fi_btop, finfo->fi_htop.hf_file);
6910 +       else
6911 +               for (bindex = finfo->fi_btop;
6912 +                    bindex >= 0 && bindex <= fidir->fd_bbot;
6913 +                    bindex++) {
6914 +                       hfile = fidir->fd_hfile + bindex;
6915 +                       do_pri_file(bindex, hfile ? hfile->hf_file : NULL);
6916 +               }
6917 +}
6918 +
6919 +static int do_pri_br(aufs_bindex_t bindex, struct au_branch *br)
6920 +{
6921 +       struct vfsmount *mnt;
6922 +       struct super_block *sb;
6923 +
6924 +       if (!br || IS_ERR(br))
6925 +               goto out;
6926 +       mnt = au_br_mnt(br);
6927 +       if (!mnt || IS_ERR(mnt))
6928 +               goto out;
6929 +       sb = mnt->mnt_sb;
6930 +       if (!sb || IS_ERR(sb))
6931 +               goto out;
6932 +
6933 +       dpri("s%d: {perm 0x%x, id %d, cnt %d, wbr %p}, "
6934 +            "%s, dev 0x%02x%02x, flags 0x%lx, cnt %d, active %d, "
6935 +            "xino %d\n",
6936 +            bindex, br->br_perm, br->br_id, atomic_read(&br->br_count),
6937 +            br->br_wbr, au_sbtype(sb), MAJOR(sb->s_dev), MINOR(sb->s_dev),
6938 +            sb->s_flags, sb->s_count,
6939 +            atomic_read(&sb->s_active), !!br->br_xino.xi_file);
6940 +       return 0;
6941 +
6942 +out:
6943 +       dpri("s%d: err %ld\n", bindex, PTR_ERR(br));
6944 +       return -1;
6945 +}
6946 +
6947 +void au_dpri_sb(struct super_block *sb)
6948 +{
6949 +       struct au_sbinfo *sbinfo;
6950 +       aufs_bindex_t bindex;
6951 +       int err;
6952 +       /* to reuduce stack size */
6953 +       struct {
6954 +               struct vfsmount mnt;
6955 +               struct au_branch fake;
6956 +       } *a;
6957 +
6958 +       /* this function can be called from magic sysrq */
6959 +       a = kzalloc(sizeof(*a), GFP_ATOMIC);
6960 +       if (unlikely(!a)) {
6961 +               dpri("no memory\n");
6962 +               return;
6963 +       }
6964 +
6965 +       a->mnt.mnt_sb = sb;
6966 +       a->fake.br_perm = 0;
6967 +       a->fake.br_path.mnt = &a->mnt;
6968 +       a->fake.br_xino.xi_file = NULL;
6969 +       atomic_set(&a->fake.br_count, 0);
6970 +       smp_mb(); /* atomic_set */
6971 +       err = do_pri_br(-1, &a->fake);
6972 +       kfree(a);
6973 +       dpri("dev 0x%x\n", sb->s_dev);
6974 +       if (err || !au_test_aufs(sb))
6975 +               return;
6976 +
6977 +       sbinfo = au_sbi(sb);
6978 +       if (!sbinfo)
6979 +               return;
6980 +       dpri("nw %d, gen %u, kobj %d\n",
6981 +            atomic_read(&sbinfo->si_nowait.nw_len), sbinfo->si_generation,
6982 +            atomic_read(&sbinfo->si_kobj.kref.refcount));
6983 +       for (bindex = 0; bindex <= sbinfo->si_bend; bindex++)
6984 +               do_pri_br(bindex, sbinfo->si_branch[0 + bindex]);
6985 +}
6986 +
6987 +/* ---------------------------------------------------------------------- */
6988 +
6989 +void au_dbg_sleep_jiffy(int jiffy)
6990 +{
6991 +       while (jiffy)
6992 +               jiffy = schedule_timeout_uninterruptible(jiffy);
6993 +}
6994 +
6995 +void au_dbg_iattr(struct iattr *ia)
6996 +{
6997 +#define AuBit(name)                                    \
6998 +       do {                                            \
6999 +               if (ia->ia_valid & ATTR_ ## name)       \
7000 +                       dpri(#name "\n");               \
7001 +       } while (0)
7002 +       AuBit(MODE);
7003 +       AuBit(UID);
7004 +       AuBit(GID);
7005 +       AuBit(SIZE);
7006 +       AuBit(ATIME);
7007 +       AuBit(MTIME);
7008 +       AuBit(CTIME);
7009 +       AuBit(ATIME_SET);
7010 +       AuBit(MTIME_SET);
7011 +       AuBit(FORCE);
7012 +       AuBit(ATTR_FLAG);
7013 +       AuBit(KILL_SUID);
7014 +       AuBit(KILL_SGID);
7015 +       AuBit(FILE);
7016 +       AuBit(KILL_PRIV);
7017 +       AuBit(OPEN);
7018 +       AuBit(TIMES_SET);
7019 +#undef AuBit
7020 +       dpri("ia_file %p\n", ia->ia_file);
7021 +}
7022 +
7023 +/* ---------------------------------------------------------------------- */
7024 +
7025 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line)
7026 +{
7027 +       struct inode *h_inode, *inode = dentry->d_inode;
7028 +       struct dentry *h_dentry;
7029 +       aufs_bindex_t bindex, bend, bi;
7030 +
7031 +       if (!inode /* || au_di(dentry)->di_lsc == AuLsc_DI_TMP */)
7032 +               return;
7033 +
7034 +       bend = au_dbend(dentry);
7035 +       bi = au_ibend(inode);
7036 +       if (bi < bend)
7037 +               bend = bi;
7038 +       bindex = au_dbstart(dentry);
7039 +       bi = au_ibstart(inode);
7040 +       if (bi > bindex)
7041 +               bindex = bi;
7042 +
7043 +       for (; bindex <= bend; bindex++) {
7044 +               h_dentry = au_h_dptr(dentry, bindex);
7045 +               if (!h_dentry)
7046 +                       continue;
7047 +               h_inode = au_h_iptr(inode, bindex);
7048 +               if (unlikely(h_inode != h_dentry->d_inode)) {
7049 +                       au_debug_on();
7050 +                       AuDbg("b%d, %s:%d\n", bindex, func, line);
7051 +                       AuDbgDentry(dentry);
7052 +                       AuDbgInode(inode);
7053 +                       au_debug_off();
7054 +                       BUG();
7055 +               }
7056 +       }
7057 +}
7058 +
7059 +void au_dbg_verify_dir_parent(struct dentry *dentry, unsigned int sigen)
7060 +{
7061 +       struct dentry *parent;
7062 +
7063 +       parent = dget_parent(dentry);
7064 +       AuDebugOn(!S_ISDIR(dentry->d_inode->i_mode));
7065 +       AuDebugOn(IS_ROOT(dentry));
7066 +       AuDebugOn(au_digen_test(parent, sigen));
7067 +       dput(parent);
7068 +}
7069 +
7070 +void au_dbg_verify_nondir_parent(struct dentry *dentry, unsigned int sigen)
7071 +{
7072 +       struct dentry *parent;
7073 +       struct inode *inode;
7074 +
7075 +       parent = dget_parent(dentry);
7076 +       inode = dentry->d_inode;
7077 +       AuDebugOn(inode && S_ISDIR(dentry->d_inode->i_mode));
7078 +       AuDebugOn(au_digen_test(parent, sigen));
7079 +       dput(parent);
7080 +}
7081 +
7082 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen)
7083 +{
7084 +       int err, i, j;
7085 +       struct au_dcsub_pages dpages;
7086 +       struct au_dpage *dpage;
7087 +       struct dentry **dentries;
7088 +
7089 +       err = au_dpages_init(&dpages, GFP_NOFS);
7090 +       AuDebugOn(err);
7091 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/1);
7092 +       AuDebugOn(err);
7093 +       for (i = dpages.ndpage - 1; !err && i >= 0; i--) {
7094 +               dpage = dpages.dpages + i;
7095 +               dentries = dpage->dentries;
7096 +               for (j = dpage->ndentry - 1; !err && j >= 0; j--)
7097 +                       AuDebugOn(au_digen_test(dentries[j], sigen));
7098 +       }
7099 +       au_dpages_free(&dpages);
7100 +}
7101 +
7102 +void au_dbg_verify_kthread(void)
7103 +{
7104 +       if (au_wkq_test()) {
7105 +               au_dbg_blocked();
7106 +               /*
7107 +                * It may be recursive, but udba=notify between two aufs mounts,
7108 +                * where a single ro branch is shared, is not a problem.
7109 +                */
7110 +               /* WARN_ON(1); */
7111 +       }
7112 +}
7113 +
7114 +/* ---------------------------------------------------------------------- */
7115 +
7116 +void au_debug_sbinfo_init(struct au_sbinfo *sbinfo __maybe_unused)
7117 +{
7118 +#ifdef AuForceNoPlink
7119 +       au_opt_clr(sbinfo->si_mntflags, PLINK);
7120 +#endif
7121 +#ifdef AuForceNoXino
7122 +       au_opt_clr(sbinfo->si_mntflags, XINO);
7123 +#endif
7124 +#ifdef AuForceNoRefrof
7125 +       au_opt_clr(sbinfo->si_mntflags, REFROF);
7126 +#endif
7127 +#ifdef AuForceHnotify
7128 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_HNOTIFY);
7129 +#endif
7130 +#ifdef AuForceRd0
7131 +       sbinfo->si_rdblk = 0;
7132 +       sbinfo->si_rdhash = 0;
7133 +#endif
7134 +}
7135 +
7136 +int __init au_debug_init(void)
7137 +{
7138 +       aufs_bindex_t bindex;
7139 +       struct au_vdir_destr destr;
7140 +
7141 +       bindex = -1;
7142 +       AuDebugOn(bindex >= 0);
7143 +
7144 +       destr.len = -1;
7145 +       AuDebugOn(destr.len < NAME_MAX);
7146 +
7147 +#ifdef CONFIG_4KSTACKS
7148 +       pr_warn("CONFIG_4KSTACKS is defined.\n");
7149 +#endif
7150 +
7151 +#ifdef AuForceNoBrs
7152 +       sysaufs_brs = 0;
7153 +#endif
7154 +
7155 +       return 0;
7156 +}
7157 diff -urN /usr/share/empty/fs/aufs/debug.h linux/fs/aufs/debug.h
7158 --- /usr/share/empty/fs/aufs/debug.h    1970-01-01 01:00:00.000000000 +0100
7159 +++ linux/fs/aufs/debug.h       2015-01-25 13:00:38.631047076 +0100
7160 @@ -0,0 +1,262 @@
7161 +/*
7162 + * Copyright (C) 2005-2014 Junjiro R. Okajima
7163 + *
7164 + * This program, aufs is free software; you can redistribute it and/or modify
7165 + * it under the terms of the GNU General Public License as published by
7166 + * the Free Software Foundation; either version 2 of the License, or
7167 + * (at your option) any later version.
7168 + *
7169 + * This program is distributed in the hope that it will be useful,
7170 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7171 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7172 + * GNU General Public License for more details.
7173 + *
7174 + * You should have received a copy of the GNU General Public License
7175 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7176 + */
7177 +
7178 +/*
7179 + * debug print functions
7180 + */
7181 +
7182 +#ifndef __AUFS_DEBUG_H__
7183 +#define __AUFS_DEBUG_H__
7184 +
7185 +#ifdef __KERNEL__
7186 +
7187 +#include <linux/atomic.h>
7188 +#include <linux/module.h>
7189 +#include <linux/kallsyms.h>
7190 +#include <linux/sysrq.h>
7191 +
7192 +#ifdef CONFIG_AUFS_DEBUG
7193 +#define AuDebugOn(a)           BUG_ON(a)
7194 +
7195 +/* module parameter */
7196 +extern atomic_t aufs_debug;
7197 +static inline void au_debug_on(void)
7198 +{
7199 +       atomic_inc(&aufs_debug);
7200 +}
7201 +static inline void au_debug_off(void)
7202 +{
7203 +       atomic_dec_if_positive(&aufs_debug);
7204 +}
7205 +
7206 +static inline int au_debug_test(void)
7207 +{
7208 +       return atomic_read(&aufs_debug) > 0;
7209 +}
7210 +#else
7211 +#define AuDebugOn(a)           do {} while (0)
7212 +AuStubVoid(au_debug_on, void)
7213 +AuStubVoid(au_debug_off, void)
7214 +AuStubInt0(au_debug_test, void)
7215 +#endif /* CONFIG_AUFS_DEBUG */
7216 +
7217 +#define param_check_atomic_t(name, p) __param_check(name, p, atomic_t)
7218 +
7219 +/* ---------------------------------------------------------------------- */
7220 +
7221 +/* debug print */
7222 +
7223 +#define AuDbg(fmt, ...) do { \
7224 +       if (au_debug_test()) \
7225 +               pr_debug("DEBUG: " fmt, ##__VA_ARGS__); \
7226 +} while (0)
7227 +#define AuLabel(l)             AuDbg(#l "\n")
7228 +#define AuIOErr(fmt, ...)      pr_err("I/O Error, " fmt, ##__VA_ARGS__)
7229 +#define AuWarn1(fmt, ...) do { \
7230 +       static unsigned char _c; \
7231 +       if (!_c++) \
7232 +               pr_warn(fmt, ##__VA_ARGS__); \
7233 +} while (0)
7234 +
7235 +#define AuErr1(fmt, ...) do { \
7236 +       static unsigned char _c; \
7237 +       if (!_c++) \
7238 +               pr_err(fmt, ##__VA_ARGS__); \
7239 +} while (0)
7240 +
7241 +#define AuIOErr1(fmt, ...) do { \
7242 +       static unsigned char _c; \
7243 +       if (!_c++) \
7244 +               AuIOErr(fmt, ##__VA_ARGS__); \
7245 +} while (0)
7246 +
7247 +#define AuUnsupportMsg "This operation is not supported." \
7248 +                       " Please report this application to aufs-users ML."
7249 +#define AuUnsupport(fmt, ...) do { \
7250 +       pr_err(AuUnsupportMsg "\n" fmt, ##__VA_ARGS__); \
7251 +       dump_stack(); \
7252 +} while (0)
7253 +
7254 +#define AuTraceErr(e) do { \
7255 +       if (unlikely((e) < 0)) \
7256 +               AuDbg("err %d\n", (int)(e)); \
7257 +} while (0)
7258 +
7259 +#define AuTraceErrPtr(p) do { \
7260 +       if (IS_ERR(p)) \
7261 +               AuDbg("err %ld\n", PTR_ERR(p)); \
7262 +} while (0)
7263 +
7264 +/* dirty macros for debug print, use with "%.*s" and caution */
7265 +#define AuLNPair(qstr)         (qstr)->len, (qstr)->name
7266 +
7267 +/* ---------------------------------------------------------------------- */
7268 +
7269 +struct au_sbinfo;
7270 +struct au_finfo;
7271 +struct dentry;
7272 +#ifdef CONFIG_AUFS_DEBUG
7273 +extern struct mutex au_dbg_mtx;
7274 +extern char *au_plevel;
7275 +struct au_nhash;
7276 +void au_dpri_whlist(struct au_nhash *whlist);
7277 +struct au_vdir;
7278 +void au_dpri_vdir(struct au_vdir *vdir);
7279 +struct inode;
7280 +void au_dpri_inode(struct inode *inode);
7281 +void au_dpri_dalias(struct inode *inode);
7282 +void au_dpri_dentry(struct dentry *dentry);
7283 +struct file;
7284 +void au_dpri_file(struct file *filp);
7285 +struct super_block;
7286 +void au_dpri_sb(struct super_block *sb);
7287 +
7288 +void au_dbg_sleep_jiffy(int jiffy);
7289 +struct iattr;
7290 +void au_dbg_iattr(struct iattr *ia);
7291 +
7292 +#define au_dbg_verify_dinode(d) __au_dbg_verify_dinode(d, __func__, __LINE__)
7293 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
7294 +void au_dbg_verify_dir_parent(struct dentry *dentry, unsigned int sigen);
7295 +void au_dbg_verify_nondir_parent(struct dentry *dentry, unsigned int sigen);
7296 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen);
7297 +void au_dbg_verify_kthread(void);
7298 +
7299 +int __init au_debug_init(void);
7300 +void au_debug_sbinfo_init(struct au_sbinfo *sbinfo);
7301 +#define AuDbgWhlist(w) do { \
7302 +       mutex_lock(&au_dbg_mtx); \
7303 +       AuDbg(#w "\n"); \
7304 +       au_dpri_whlist(w); \
7305 +       mutex_unlock(&au_dbg_mtx); \
7306 +} while (0)
7307 +
7308 +#define AuDbgVdir(v) do { \
7309 +       mutex_lock(&au_dbg_mtx); \
7310 +       AuDbg(#v "\n"); \
7311 +       au_dpri_vdir(v); \
7312 +       mutex_unlock(&au_dbg_mtx); \
7313 +} while (0)
7314 +
7315 +#define AuDbgInode(i) do { \
7316 +       mutex_lock(&au_dbg_mtx); \
7317 +       AuDbg(#i "\n"); \
7318 +       au_dpri_inode(i); \
7319 +       mutex_unlock(&au_dbg_mtx); \
7320 +} while (0)
7321 +
7322 +#define AuDbgDAlias(i) do { \
7323 +       mutex_lock(&au_dbg_mtx); \
7324 +       AuDbg(#i "\n"); \
7325 +       au_dpri_dalias(i); \
7326 +       mutex_unlock(&au_dbg_mtx); \
7327 +} while (0)
7328 +
7329 +#define AuDbgDentry(d) do { \
7330 +       mutex_lock(&au_dbg_mtx); \
7331 +       AuDbg(#d "\n"); \
7332 +       au_dpri_dentry(d); \
7333 +       mutex_unlock(&au_dbg_mtx); \
7334 +} while (0)
7335 +
7336 +#define AuDbgFile(f) do { \
7337 +       mutex_lock(&au_dbg_mtx); \
7338 +       AuDbg(#f "\n"); \
7339 +       au_dpri_file(f); \
7340 +       mutex_unlock(&au_dbg_mtx); \
7341 +} while (0)
7342 +
7343 +#define AuDbgSb(sb) do { \
7344 +       mutex_lock(&au_dbg_mtx); \
7345 +       AuDbg(#sb "\n"); \
7346 +       au_dpri_sb(sb); \
7347 +       mutex_unlock(&au_dbg_mtx); \
7348 +} while (0)
7349 +
7350 +#define AuDbgSleep(sec) do { \
7351 +       AuDbg("sleep %d sec\n", sec); \
7352 +       ssleep(sec); \
7353 +} while (0)
7354 +
7355 +#define AuDbgSleepJiffy(jiffy) do { \
7356 +       AuDbg("sleep %d jiffies\n", jiffy); \
7357 +       au_dbg_sleep_jiffy(jiffy); \
7358 +} while (0)
7359 +
7360 +#define AuDbgIAttr(ia) do { \
7361 +       AuDbg("ia_valid 0x%x\n", (ia)->ia_valid); \
7362 +       au_dbg_iattr(ia); \
7363 +} while (0)
7364 +
7365 +#define AuDbgSym(addr) do {                            \
7366 +       char sym[KSYM_SYMBOL_LEN];                      \
7367 +       sprint_symbol(sym, (unsigned long)addr);        \
7368 +       AuDbg("%s\n", sym);                             \
7369 +} while (0)
7370 +
7371 +#define AuInfoSym(addr) do {                           \
7372 +       char sym[KSYM_SYMBOL_LEN];                      \
7373 +       sprint_symbol(sym, (unsigned long)addr);        \
7374 +       AuInfo("%s\n", sym);                            \
7375 +} while (0)
7376 +#else
7377 +AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
7378 +AuStubVoid(au_dbg_verify_dir_parent, struct dentry *dentry, unsigned int sigen)
7379 +AuStubVoid(au_dbg_verify_nondir_parent, struct dentry *dentry,
7380 +          unsigned int sigen)
7381 +AuStubVoid(au_dbg_verify_gen, struct dentry *parent, unsigned int sigen)
7382 +AuStubVoid(au_dbg_verify_kthread, void)
7383 +AuStubInt0(__init au_debug_init, void)
7384 +AuStubVoid(au_debug_sbinfo_init, struct au_sbinfo *sbinfo)
7385 +
7386 +#define AuDbgWhlist(w)         do {} while (0)
7387 +#define AuDbgVdir(v)           do {} while (0)
7388 +#define AuDbgInode(i)          do {} while (0)
7389 +#define AuDbgDAlias(i)         do {} while (0)
7390 +#define AuDbgDentry(d)         do {} while (0)
7391 +#define AuDbgFile(f)           do {} while (0)
7392 +#define AuDbgSb(sb)            do {} while (0)
7393 +#define AuDbgSleep(sec)                do {} while (0)
7394 +#define AuDbgSleepJiffy(jiffy) do {} while (0)
7395 +#define AuDbgIAttr(ia)         do {} while (0)
7396 +#define AuDbgSym(addr)         do {} while (0)
7397 +#define AuInfoSym(addr)                do {} while (0)
7398 +#endif /* CONFIG_AUFS_DEBUG */
7399 +
7400 +/* ---------------------------------------------------------------------- */
7401 +
7402 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
7403 +int __init au_sysrq_init(void);
7404 +void au_sysrq_fin(void);
7405 +
7406 +#ifdef CONFIG_HW_CONSOLE
7407 +#define au_dbg_blocked() do { \
7408 +       WARN_ON(1); \
7409 +       handle_sysrq('w'); \
7410 +} while (0)
7411 +#else
7412 +AuStubVoid(au_dbg_blocked, void)
7413 +#endif
7414 +
7415 +#else
7416 +AuStubInt0(__init au_sysrq_init, void)
7417 +AuStubVoid(au_sysrq_fin, void)
7418 +AuStubVoid(au_dbg_blocked, void)
7419 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
7420 +
7421 +#endif /* __KERNEL__ */
7422 +#endif /* __AUFS_DEBUG_H__ */
7423 diff -urN /usr/share/empty/fs/aufs/dentry.c linux/fs/aufs/dentry.c
7424 --- /usr/share/empty/fs/aufs/dentry.c   1970-01-01 01:00:00.000000000 +0100
7425 +++ linux/fs/aufs/dentry.c      2015-01-25 13:00:38.631047076 +0100
7426 @@ -0,0 +1,1096 @@
7427 +/*
7428 + * Copyright (C) 2005-2014 Junjiro R. Okajima
7429 + *
7430 + * This program, aufs is free software; you can redistribute it and/or modify
7431 + * it under the terms of the GNU General Public License as published by
7432 + * the Free Software Foundation; either version 2 of the License, or
7433 + * (at your option) any later version.
7434 + *
7435 + * This program is distributed in the hope that it will be useful,
7436 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7437 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7438 + * GNU General Public License for more details.
7439 + *
7440 + * You should have received a copy of the GNU General Public License
7441 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7442 + */
7443 +
7444 +/*
7445 + * lookup and dentry operations
7446 + */
7447 +
7448 +#include <linux/namei.h>
7449 +#include "aufs.h"
7450 +
7451 +#define AuLkup_ALLOW_NEG       1
7452 +#define AuLkup_IGNORE_PERM     (1 << 1)
7453 +#define au_ftest_lkup(flags, name)     ((flags) & AuLkup_##name)
7454 +#define au_fset_lkup(flags, name) \
7455 +       do { (flags) |= AuLkup_##name; } while (0)
7456 +#define au_fclr_lkup(flags, name) \
7457 +       do { (flags) &= ~AuLkup_##name; } while (0)
7458 +
7459 +struct au_do_lookup_args {
7460 +       unsigned int            flags;
7461 +       mode_t                  type;
7462 +};
7463 +
7464 +/*
7465 + * returns positive/negative dentry, NULL or an error.
7466 + * NULL means whiteout-ed or not-found.
7467 + */
7468 +static struct dentry*
7469 +au_do_lookup(struct dentry *h_parent, struct dentry *dentry,
7470 +            aufs_bindex_t bindex, struct qstr *wh_name,
7471 +            struct au_do_lookup_args *args)
7472 +{
7473 +       struct dentry *h_dentry;
7474 +       struct inode *h_inode, *inode;
7475 +       struct au_branch *br;
7476 +       int wh_found, opq;
7477 +       unsigned char wh_able;
7478 +       const unsigned char allow_neg = !!au_ftest_lkup(args->flags, ALLOW_NEG);
7479 +       const unsigned char ignore_perm = !!au_ftest_lkup(args->flags,
7480 +                                                         IGNORE_PERM);
7481 +
7482 +       wh_found = 0;
7483 +       br = au_sbr(dentry->d_sb, bindex);
7484 +       wh_able = !!au_br_whable(br->br_perm);
7485 +       if (wh_able)
7486 +               wh_found = au_wh_test(h_parent, wh_name, /*try_sio*/0);
7487 +       h_dentry = ERR_PTR(wh_found);
7488 +       if (!wh_found)
7489 +               goto real_lookup;
7490 +       if (unlikely(wh_found < 0))
7491 +               goto out;
7492 +
7493 +       /* We found a whiteout */
7494 +       /* au_set_dbend(dentry, bindex); */
7495 +       au_set_dbwh(dentry, bindex);
7496 +       if (!allow_neg)
7497 +               return NULL; /* success */
7498 +
7499 +real_lookup:
7500 +       if (!ignore_perm)
7501 +               h_dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
7502 +       else
7503 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7504 +       if (IS_ERR(h_dentry))
7505 +               goto out;
7506 +
7507 +       h_inode = h_dentry->d_inode;
7508 +       if (!h_inode) {
7509 +               if (!allow_neg)
7510 +                       goto out_neg;
7511 +       } else if (wh_found
7512 +                  || (args->type && args->type != (h_inode->i_mode & S_IFMT)))
7513 +               goto out_neg;
7514 +
7515 +       if (au_dbend(dentry) <= bindex)
7516 +               au_set_dbend(dentry, bindex);
7517 +       if (au_dbstart(dentry) < 0 || bindex < au_dbstart(dentry))
7518 +               au_set_dbstart(dentry, bindex);
7519 +       au_set_h_dptr(dentry, bindex, h_dentry);
7520 +
7521 +       inode = dentry->d_inode;
7522 +       if (!h_inode || !S_ISDIR(h_inode->i_mode) || !wh_able
7523 +           || (inode && !S_ISDIR(inode->i_mode)))
7524 +               goto out; /* success */
7525 +
7526 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
7527 +       opq = au_diropq_test(h_dentry);
7528 +       mutex_unlock(&h_inode->i_mutex);
7529 +       if (opq > 0)
7530 +               au_set_dbdiropq(dentry, bindex);
7531 +       else if (unlikely(opq < 0)) {
7532 +               au_set_h_dptr(dentry, bindex, NULL);
7533 +               h_dentry = ERR_PTR(opq);
7534 +       }
7535 +       goto out;
7536 +
7537 +out_neg:
7538 +       dput(h_dentry);
7539 +       h_dentry = NULL;
7540 +out:
7541 +       return h_dentry;
7542 +}
7543 +
7544 +static int au_test_shwh(struct super_block *sb, const struct qstr *name)
7545 +{
7546 +       if (unlikely(!au_opt_test(au_mntflags(sb), SHWH)
7547 +                    && !strncmp(name->name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)))
7548 +               return -EPERM;
7549 +       return 0;
7550 +}
7551 +
7552 +/*
7553 + * returns the number of lower positive dentries,
7554 + * otherwise an error.
7555 + * can be called at unlinking with @type is zero.
7556 + */
7557 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t bstart, mode_t type)
7558 +{
7559 +       int npositive, err;
7560 +       aufs_bindex_t bindex, btail, bdiropq;
7561 +       unsigned char isdir, dirperm1;
7562 +       struct qstr whname;
7563 +       struct au_do_lookup_args args = {
7564 +               .flags          = 0,
7565 +               .type           = type
7566 +       };
7567 +       const struct qstr *name = &dentry->d_name;
7568 +       struct dentry *parent;
7569 +       struct inode *inode;
7570 +       struct super_block *sb;
7571 +
7572 +       sb = dentry->d_sb;
7573 +       err = au_test_shwh(sb, name);
7574 +       if (unlikely(err))
7575 +               goto out;
7576 +
7577 +       err = au_wh_name_alloc(&whname, name);
7578 +       if (unlikely(err))
7579 +               goto out;
7580 +
7581 +       inode = dentry->d_inode;
7582 +       isdir = !!(inode && S_ISDIR(inode->i_mode));
7583 +       if (!type)
7584 +               au_fset_lkup(args.flags, ALLOW_NEG);
7585 +       dirperm1 = !!au_opt_test(au_mntflags(sb), DIRPERM1);
7586 +
7587 +       npositive = 0;
7588 +       parent = dget_parent(dentry);
7589 +       btail = au_dbtaildir(parent);
7590 +       for (bindex = bstart; bindex <= btail; bindex++) {
7591 +               struct dentry *h_parent, *h_dentry;
7592 +               struct inode *h_inode, *h_dir;
7593 +
7594 +               h_dentry = au_h_dptr(dentry, bindex);
7595 +               if (h_dentry) {
7596 +                       if (h_dentry->d_inode)
7597 +                               npositive++;
7598 +                       if (type != S_IFDIR)
7599 +                               break;
7600 +                       continue;
7601 +               }
7602 +               h_parent = au_h_dptr(parent, bindex);
7603 +               if (!h_parent)
7604 +                       continue;
7605 +               h_dir = h_parent->d_inode;
7606 +               if (!h_dir || !S_ISDIR(h_dir->i_mode))
7607 +                       continue;
7608 +
7609 +               mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
7610 +               h_dentry = au_do_lookup(h_parent, dentry, bindex, &whname,
7611 +                                       &args);
7612 +               mutex_unlock(&h_dir->i_mutex);
7613 +               err = PTR_ERR(h_dentry);
7614 +               if (IS_ERR(h_dentry))
7615 +                       goto out_parent;
7616 +               au_fclr_lkup(args.flags, ALLOW_NEG);
7617 +               if (dirperm1)
7618 +                       au_fset_lkup(args.flags, IGNORE_PERM);
7619 +
7620 +               if (au_dbwh(dentry) >= 0)
7621 +                       break;
7622 +               if (!h_dentry)
7623 +                       continue;
7624 +               h_inode = h_dentry->d_inode;
7625 +               if (!h_inode)
7626 +                       continue;
7627 +               npositive++;
7628 +               if (!args.type)
7629 +                       args.type = h_inode->i_mode & S_IFMT;
7630 +               if (args.type != S_IFDIR)
7631 +                       break;
7632 +               else if (isdir) {
7633 +                       /* the type of lower may be different */
7634 +                       bdiropq = au_dbdiropq(dentry);
7635 +                       if (bdiropq >= 0 && bdiropq <= bindex)
7636 +                               break;
7637 +               }
7638 +       }
7639 +
7640 +       if (npositive) {
7641 +               AuLabel(positive);
7642 +               au_update_dbstart(dentry);
7643 +       }
7644 +       err = npositive;
7645 +       if (unlikely(!au_opt_test(au_mntflags(sb), UDBA_NONE)
7646 +                    && au_dbstart(dentry) < 0)) {
7647 +               err = -EIO;
7648 +               AuIOErr("both of real entry and whiteout found, %pd, err %d\n",
7649 +                       dentry, err);
7650 +       }
7651 +
7652 +out_parent:
7653 +       dput(parent);
7654 +       kfree(whname.name);
7655 +out:
7656 +       return err;
7657 +}
7658 +
7659 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent)
7660 +{
7661 +       struct dentry *dentry;
7662 +       int wkq_err;
7663 +
7664 +       if (!au_test_h_perm_sio(parent->d_inode, MAY_EXEC))
7665 +               dentry = vfsub_lkup_one(name, parent);
7666 +       else {
7667 +               struct vfsub_lkup_one_args args = {
7668 +                       .errp   = &dentry,
7669 +                       .name   = name,
7670 +                       .parent = parent
7671 +               };
7672 +
7673 +               wkq_err = au_wkq_wait(vfsub_call_lkup_one, &args);
7674 +               if (unlikely(wkq_err))
7675 +                       dentry = ERR_PTR(wkq_err);
7676 +       }
7677 +
7678 +       return dentry;
7679 +}
7680 +
7681 +/*
7682 + * lookup @dentry on @bindex which should be negative.
7683 + */
7684 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh)
7685 +{
7686 +       int err;
7687 +       struct dentry *parent, *h_parent, *h_dentry;
7688 +       struct au_branch *br;
7689 +
7690 +       parent = dget_parent(dentry);
7691 +       h_parent = au_h_dptr(parent, bindex);
7692 +       br = au_sbr(dentry->d_sb, bindex);
7693 +       if (wh)
7694 +               h_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
7695 +       else
7696 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7697 +       err = PTR_ERR(h_dentry);
7698 +       if (IS_ERR(h_dentry))
7699 +               goto out;
7700 +       if (unlikely(h_dentry->d_inode)) {
7701 +               err = -EIO;
7702 +               AuIOErr("%pd should be negative on b%d.\n", h_dentry, bindex);
7703 +               dput(h_dentry);
7704 +               goto out;
7705 +       }
7706 +
7707 +       err = 0;
7708 +       if (bindex < au_dbstart(dentry))
7709 +               au_set_dbstart(dentry, bindex);
7710 +       if (au_dbend(dentry) < bindex)
7711 +               au_set_dbend(dentry, bindex);
7712 +       au_set_h_dptr(dentry, bindex, h_dentry);
7713 +
7714 +out:
7715 +       dput(parent);
7716 +       return err;
7717 +}
7718 +
7719 +/* ---------------------------------------------------------------------- */
7720 +
7721 +/* subset of struct inode */
7722 +struct au_iattr {
7723 +       unsigned long           i_ino;
7724 +       /* unsigned int         i_nlink; */
7725 +       kuid_t                  i_uid;
7726 +       kgid_t                  i_gid;
7727 +       u64                     i_version;
7728 +/*
7729 +       loff_t                  i_size;
7730 +       blkcnt_t                i_blocks;
7731 +*/
7732 +       umode_t                 i_mode;
7733 +};
7734 +
7735 +static void au_iattr_save(struct au_iattr *ia, struct inode *h_inode)
7736 +{
7737 +       ia->i_ino = h_inode->i_ino;
7738 +       /* ia->i_nlink = h_inode->i_nlink; */
7739 +       ia->i_uid = h_inode->i_uid;
7740 +       ia->i_gid = h_inode->i_gid;
7741 +       ia->i_version = h_inode->i_version;
7742 +/*
7743 +       ia->i_size = h_inode->i_size;
7744 +       ia->i_blocks = h_inode->i_blocks;
7745 +*/
7746 +       ia->i_mode = (h_inode->i_mode & S_IFMT);
7747 +}
7748 +
7749 +static int au_iattr_test(struct au_iattr *ia, struct inode *h_inode)
7750 +{
7751 +       return ia->i_ino != h_inode->i_ino
7752 +               /* || ia->i_nlink != h_inode->i_nlink */
7753 +               || !uid_eq(ia->i_uid, h_inode->i_uid)
7754 +               || !gid_eq(ia->i_gid, h_inode->i_gid)
7755 +               || ia->i_version != h_inode->i_version
7756 +/*
7757 +               || ia->i_size != h_inode->i_size
7758 +               || ia->i_blocks != h_inode->i_blocks
7759 +*/
7760 +               || ia->i_mode != (h_inode->i_mode & S_IFMT);
7761 +}
7762 +
7763 +static int au_h_verify_dentry(struct dentry *h_dentry, struct dentry *h_parent,
7764 +                             struct au_branch *br)
7765 +{
7766 +       int err;
7767 +       struct au_iattr ia;
7768 +       struct inode *h_inode;
7769 +       struct dentry *h_d;
7770 +       struct super_block *h_sb;
7771 +
7772 +       err = 0;
7773 +       memset(&ia, -1, sizeof(ia));
7774 +       h_sb = h_dentry->d_sb;
7775 +       h_inode = h_dentry->d_inode;
7776 +       if (h_inode)
7777 +               au_iattr_save(&ia, h_inode);
7778 +       else if (au_test_nfs(h_sb) || au_test_fuse(h_sb))
7779 +               /* nfs d_revalidate may return 0 for negative dentry */
7780 +               /* fuse d_revalidate always return 0 for negative dentry */
7781 +               goto out;
7782 +
7783 +       /* main purpose is namei.c:cached_lookup() and d_revalidate */
7784 +       h_d = vfsub_lkup_one(&h_dentry->d_name, h_parent);
7785 +       err = PTR_ERR(h_d);
7786 +       if (IS_ERR(h_d))
7787 +               goto out;
7788 +
7789 +       err = 0;
7790 +       if (unlikely(h_d != h_dentry
7791 +                    || h_d->d_inode != h_inode
7792 +                    || (h_inode && au_iattr_test(&ia, h_inode))))
7793 +               err = au_busy_or_stale();
7794 +       dput(h_d);
7795 +
7796 +out:
7797 +       AuTraceErr(err);
7798 +       return err;
7799 +}
7800 +
7801 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
7802 +               struct dentry *h_parent, struct au_branch *br)
7803 +{
7804 +       int err;
7805 +
7806 +       err = 0;
7807 +       if (udba == AuOpt_UDBA_REVAL
7808 +           && !au_test_fs_remote(h_dentry->d_sb)) {
7809 +               IMustLock(h_dir);
7810 +               err = (h_dentry->d_parent->d_inode != h_dir);
7811 +       } else if (udba != AuOpt_UDBA_NONE)
7812 +               err = au_h_verify_dentry(h_dentry, h_parent, br);
7813 +
7814 +       return err;
7815 +}
7816 +
7817 +/* ---------------------------------------------------------------------- */
7818 +
7819 +static int au_do_refresh_hdentry(struct dentry *dentry, struct dentry *parent)
7820 +{
7821 +       int err;
7822 +       aufs_bindex_t new_bindex, bindex, bend, bwh, bdiropq;
7823 +       struct au_hdentry tmp, *p, *q;
7824 +       struct au_dinfo *dinfo;
7825 +       struct super_block *sb;
7826 +
7827 +       DiMustWriteLock(dentry);
7828 +
7829 +       sb = dentry->d_sb;
7830 +       dinfo = au_di(dentry);
7831 +       bend = dinfo->di_bend;
7832 +       bwh = dinfo->di_bwh;
7833 +       bdiropq = dinfo->di_bdiropq;
7834 +       p = dinfo->di_hdentry + dinfo->di_bstart;
7835 +       for (bindex = dinfo->di_bstart; bindex <= bend; bindex++, p++) {
7836 +               if (!p->hd_dentry)
7837 +                       continue;
7838 +
7839 +               new_bindex = au_br_index(sb, p->hd_id);
7840 +               if (new_bindex == bindex)
7841 +                       continue;
7842 +
7843 +               if (dinfo->di_bwh == bindex)
7844 +                       bwh = new_bindex;
7845 +               if (dinfo->di_bdiropq == bindex)
7846 +                       bdiropq = new_bindex;
7847 +               if (new_bindex < 0) {
7848 +                       au_hdput(p);
7849 +                       p->hd_dentry = NULL;
7850 +                       continue;
7851 +               }
7852 +
7853 +               /* swap two lower dentries, and loop again */
7854 +               q = dinfo->di_hdentry + new_bindex;
7855 +               tmp = *q;
7856 +               *q = *p;
7857 +               *p = tmp;
7858 +               if (tmp.hd_dentry) {
7859 +                       bindex--;
7860 +                       p--;
7861 +               }
7862 +       }
7863 +
7864 +       dinfo->di_bwh = -1;
7865 +       if (bwh >= 0 && bwh <= au_sbend(sb) && au_sbr_whable(sb, bwh))
7866 +               dinfo->di_bwh = bwh;
7867 +
7868 +       dinfo->di_bdiropq = -1;
7869 +       if (bdiropq >= 0
7870 +           && bdiropq <= au_sbend(sb)
7871 +           && au_sbr_whable(sb, bdiropq))
7872 +               dinfo->di_bdiropq = bdiropq;
7873 +
7874 +       err = -EIO;
7875 +       dinfo->di_bstart = -1;
7876 +       dinfo->di_bend = -1;
7877 +       bend = au_dbend(parent);
7878 +       p = dinfo->di_hdentry;
7879 +       for (bindex = 0; bindex <= bend; bindex++, p++)
7880 +               if (p->hd_dentry) {
7881 +                       dinfo->di_bstart = bindex;
7882 +                       break;
7883 +               }
7884 +
7885 +       if (dinfo->di_bstart >= 0) {
7886 +               p = dinfo->di_hdentry + bend;
7887 +               for (bindex = bend; bindex >= 0; bindex--, p--)
7888 +                       if (p->hd_dentry) {
7889 +                               dinfo->di_bend = bindex;
7890 +                               err = 0;
7891 +                               break;
7892 +                       }
7893 +       }
7894 +
7895 +       return err;
7896 +}
7897 +
7898 +static void au_do_hide(struct dentry *dentry)
7899 +{
7900 +       struct inode *inode;
7901 +
7902 +       inode = dentry->d_inode;
7903 +       if (inode) {
7904 +               if (!S_ISDIR(inode->i_mode)) {
7905 +                       if (inode->i_nlink && !d_unhashed(dentry))
7906 +                               drop_nlink(inode);
7907 +               } else {
7908 +                       clear_nlink(inode);
7909 +                       /* stop next lookup */
7910 +                       inode->i_flags |= S_DEAD;
7911 +               }
7912 +               smp_mb(); /* necessary? */
7913 +       }
7914 +       d_drop(dentry);
7915 +}
7916 +
7917 +static int au_hide_children(struct dentry *parent)
7918 +{
7919 +       int err, i, j, ndentry;
7920 +       struct au_dcsub_pages dpages;
7921 +       struct au_dpage *dpage;
7922 +       struct dentry *dentry;
7923 +
7924 +       err = au_dpages_init(&dpages, GFP_NOFS);
7925 +       if (unlikely(err))
7926 +               goto out;
7927 +       err = au_dcsub_pages(&dpages, parent, NULL, NULL);
7928 +       if (unlikely(err))
7929 +               goto out_dpages;
7930 +
7931 +       /* in reverse order */
7932 +       for (i = dpages.ndpage - 1; i >= 0; i--) {
7933 +               dpage = dpages.dpages + i;
7934 +               ndentry = dpage->ndentry;
7935 +               for (j = ndentry - 1; j >= 0; j--) {
7936 +                       dentry = dpage->dentries[j];
7937 +                       if (dentry != parent)
7938 +                               au_do_hide(dentry);
7939 +               }
7940 +       }
7941 +
7942 +out_dpages:
7943 +       au_dpages_free(&dpages);
7944 +out:
7945 +       return err;
7946 +}
7947 +
7948 +static void au_hide(struct dentry *dentry)
7949 +{
7950 +       int err;
7951 +       struct inode *inode;
7952 +
7953 +       AuDbgDentry(dentry);
7954 +       inode = dentry->d_inode;
7955 +       if (inode && S_ISDIR(inode->i_mode)) {
7956 +               /* shrink_dcache_parent(dentry); */
7957 +               err = au_hide_children(dentry);
7958 +               if (unlikely(err))
7959 +                       AuIOErr("%pd, failed hiding children, ignored %d\n",
7960 +                               dentry, err);
7961 +       }
7962 +       au_do_hide(dentry);
7963 +}
7964 +
7965 +/*
7966 + * By adding a dirty branch, a cached dentry may be affected in various ways.
7967 + *
7968 + * a dirty branch is added
7969 + * - on the top of layers
7970 + * - in the middle of layers
7971 + * - to the bottom of layers
7972 + *
7973 + * on the added branch there exists
7974 + * - a whiteout
7975 + * - a diropq
7976 + * - a same named entry
7977 + *   + exist
7978 + *     * negative --> positive
7979 + *     * positive --> positive
7980 + *      - type is unchanged
7981 + *      - type is changed
7982 + *   + doesn't exist
7983 + *     * negative --> negative
7984 + *     * positive --> negative (rejected by au_br_del() for non-dir case)
7985 + * - none
7986 + */
7987 +static int au_refresh_by_dinfo(struct dentry *dentry, struct au_dinfo *dinfo,
7988 +                              struct au_dinfo *tmp)
7989 +{
7990 +       int err;
7991 +       aufs_bindex_t bindex, bend;
7992 +       struct {
7993 +               struct dentry *dentry;
7994 +               struct inode *inode;
7995 +               mode_t mode;
7996 +       } orig_h, tmp_h;
7997 +       struct au_hdentry *hd;
7998 +       struct inode *inode, *h_inode;
7999 +       struct dentry *h_dentry;
8000 +
8001 +       err = 0;
8002 +       AuDebugOn(dinfo->di_bstart < 0);
8003 +       orig_h.dentry = dinfo->di_hdentry[dinfo->di_bstart].hd_dentry;
8004 +       orig_h.inode = orig_h.dentry->d_inode;
8005 +       orig_h.mode = 0;
8006 +       if (orig_h.inode)
8007 +               orig_h.mode = orig_h.inode->i_mode & S_IFMT;
8008 +       memset(&tmp_h, 0, sizeof(tmp_h));
8009 +       if (tmp->di_bstart >= 0) {
8010 +               tmp_h.dentry = tmp->di_hdentry[tmp->di_bstart].hd_dentry;
8011 +               tmp_h.inode = tmp_h.dentry->d_inode;
8012 +               if (tmp_h.inode)
8013 +                       tmp_h.mode = tmp_h.inode->i_mode & S_IFMT;
8014 +       }
8015 +
8016 +       inode = dentry->d_inode;
8017 +       if (!orig_h.inode) {
8018 +               AuDbg("nagative originally\n");
8019 +               if (inode) {
8020 +                       au_hide(dentry);
8021 +                       goto out;
8022 +               }
8023 +               AuDebugOn(inode);
8024 +               AuDebugOn(dinfo->di_bstart != dinfo->di_bend);
8025 +               AuDebugOn(dinfo->di_bdiropq != -1);
8026 +
8027 +               if (!tmp_h.inode) {
8028 +                       AuDbg("negative --> negative\n");
8029 +                       /* should have only one negative lower */
8030 +                       if (tmp->di_bstart >= 0
8031 +                           && tmp->di_bstart < dinfo->di_bstart) {
8032 +                               AuDebugOn(tmp->di_bstart != tmp->di_bend);
8033 +                               AuDebugOn(dinfo->di_bstart != dinfo->di_bend);
8034 +                               au_set_h_dptr(dentry, dinfo->di_bstart, NULL);
8035 +                               au_di_cp(dinfo, tmp);
8036 +                               hd = tmp->di_hdentry + tmp->di_bstart;
8037 +                               au_set_h_dptr(dentry, tmp->di_bstart,
8038 +                                             dget(hd->hd_dentry));
8039 +                       }
8040 +                       au_dbg_verify_dinode(dentry);
8041 +               } else {
8042 +                       AuDbg("negative --> positive\n");
8043 +                       /*
8044 +                        * similar to the behaviour of creating with bypassing
8045 +                        * aufs.
8046 +                        * unhash it in order to force an error in the
8047 +                        * succeeding create operation.
8048 +                        * we should not set S_DEAD here.
8049 +                        */
8050 +                       d_drop(dentry);
8051 +                       /* au_di_swap(tmp, dinfo); */
8052 +                       au_dbg_verify_dinode(dentry);
8053 +               }
8054 +       } else {
8055 +               AuDbg("positive originally\n");
8056 +               /* inode may be NULL */
8057 +               AuDebugOn(inode && (inode->i_mode & S_IFMT) != orig_h.mode);
8058 +               if (!tmp_h.inode) {
8059 +                       AuDbg("positive --> negative\n");
8060 +                       /* or bypassing aufs */
8061 +                       au_hide(dentry);
8062 +                       if (tmp->di_bwh >= 0 && tmp->di_bwh <= dinfo->di_bstart)
8063 +                               dinfo->di_bwh = tmp->di_bwh;
8064 +                       if (inode)
8065 +                               err = au_refresh_hinode_self(inode);
8066 +                       au_dbg_verify_dinode(dentry);
8067 +               } else if (orig_h.mode == tmp_h.mode) {
8068 +                       AuDbg("positive --> positive, same type\n");
8069 +                       if (!S_ISDIR(orig_h.mode)
8070 +                           && dinfo->di_bstart > tmp->di_bstart) {
8071 +                               /*
8072 +                                * similar to the behaviour of removing and
8073 +                                * creating.
8074 +                                */
8075 +                               au_hide(dentry);
8076 +                               if (inode)
8077 +                                       err = au_refresh_hinode_self(inode);
8078 +                               au_dbg_verify_dinode(dentry);
8079 +                       } else {
8080 +                               /* fill empty slots */
8081 +                               if (dinfo->di_bstart > tmp->di_bstart)
8082 +                                       dinfo->di_bstart = tmp->di_bstart;
8083 +                               if (dinfo->di_bend < tmp->di_bend)
8084 +                                       dinfo->di_bend = tmp->di_bend;
8085 +                               dinfo->di_bwh = tmp->di_bwh;
8086 +                               dinfo->di_bdiropq = tmp->di_bdiropq;
8087 +                               hd = tmp->di_hdentry;
8088 +                               bend = dinfo->di_bend;
8089 +                               for (bindex = tmp->di_bstart; bindex <= bend;
8090 +                                    bindex++) {
8091 +                                       if (au_h_dptr(dentry, bindex))
8092 +                                               continue;
8093 +                                       h_dentry = hd[bindex].hd_dentry;
8094 +                                       if (!h_dentry)
8095 +                                               continue;
8096 +                                       h_inode = h_dentry->d_inode;
8097 +                                       AuDebugOn(!h_inode);
8098 +                                       AuDebugOn(orig_h.mode
8099 +                                                 != (h_inode->i_mode
8100 +                                                     & S_IFMT));
8101 +                                       au_set_h_dptr(dentry, bindex,
8102 +                                                     dget(h_dentry));
8103 +                               }
8104 +                               err = au_refresh_hinode(inode, dentry);
8105 +                               au_dbg_verify_dinode(dentry);
8106 +                       }
8107 +               } else {
8108 +                       AuDbg("positive --> positive, different type\n");
8109 +                       /* similar to the behaviour of removing and creating */
8110 +                       au_hide(dentry);
8111 +                       if (inode)
8112 +                               err = au_refresh_hinode_self(inode);
8113 +                       au_dbg_verify_dinode(dentry);
8114 +               }
8115 +       }
8116 +
8117 +out:
8118 +       return err;
8119 +}
8120 +
8121 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent)
8122 +{
8123 +       int err, ebrange;
8124 +       unsigned int sigen;
8125 +       struct au_dinfo *dinfo, *tmp;
8126 +       struct super_block *sb;
8127 +       struct inode *inode;
8128 +
8129 +       DiMustWriteLock(dentry);
8130 +       AuDebugOn(IS_ROOT(dentry));
8131 +       AuDebugOn(!parent->d_inode);
8132 +
8133 +       sb = dentry->d_sb;
8134 +       inode = dentry->d_inode;
8135 +       sigen = au_sigen(sb);
8136 +       err = au_digen_test(parent, sigen);
8137 +       if (unlikely(err))
8138 +               goto out;
8139 +
8140 +       dinfo = au_di(dentry);
8141 +       err = au_di_realloc(dinfo, au_sbend(sb) + 1);
8142 +       if (unlikely(err))
8143 +               goto out;
8144 +       ebrange = au_dbrange_test(dentry);
8145 +       if (!ebrange)
8146 +               ebrange = au_do_refresh_hdentry(dentry, parent);
8147 +
8148 +       if (d_unhashed(dentry) || ebrange /* || dinfo->di_tmpfile */) {
8149 +               AuDebugOn(au_dbstart(dentry) < 0 && au_dbend(dentry) >= 0);
8150 +               if (inode)
8151 +                       err = au_refresh_hinode_self(inode);
8152 +               au_dbg_verify_dinode(dentry);
8153 +               if (!err)
8154 +                       goto out_dgen; /* success */
8155 +               goto out;
8156 +       }
8157 +
8158 +       /* temporary dinfo */
8159 +       AuDbgDentry(dentry);
8160 +       err = -ENOMEM;
8161 +       tmp = au_di_alloc(sb, AuLsc_DI_TMP);
8162 +       if (unlikely(!tmp))
8163 +               goto out;
8164 +       au_di_swap(tmp, dinfo);
8165 +       /* returns the number of positive dentries */
8166 +       /*
8167 +        * if current working dir is removed, it returns an error.
8168 +        * but the dentry is legal.
8169 +        */
8170 +       err = au_lkup_dentry(dentry, /*bstart*/0, /*type*/0);
8171 +       AuDbgDentry(dentry);
8172 +       au_di_swap(tmp, dinfo);
8173 +       if (err == -ENOENT)
8174 +               err = 0;
8175 +       if (err >= 0) {
8176 +               /* compare/refresh by dinfo */
8177 +               AuDbgDentry(dentry);
8178 +               err = au_refresh_by_dinfo(dentry, dinfo, tmp);
8179 +               au_dbg_verify_dinode(dentry);
8180 +               AuTraceErr(err);
8181 +       }
8182 +       au_rw_write_unlock(&tmp->di_rwsem);
8183 +       au_di_free(tmp);
8184 +       if (unlikely(err))
8185 +               goto out;
8186 +
8187 +out_dgen:
8188 +       au_update_digen(dentry);
8189 +out:
8190 +       if (unlikely(err && !(dentry->d_flags & DCACHE_NFSFS_RENAMED))) {
8191 +               AuIOErr("failed refreshing %pd, %d\n", dentry, err);
8192 +               AuDbgDentry(dentry);
8193 +       }
8194 +       AuTraceErr(err);
8195 +       return err;
8196 +}
8197 +
8198 +static int au_do_h_d_reval(struct dentry *h_dentry, unsigned int flags,
8199 +                          struct dentry *dentry, aufs_bindex_t bindex)
8200 +{
8201 +       int err, valid;
8202 +
8203 +       err = 0;
8204 +       if (!(h_dentry->d_flags & DCACHE_OP_REVALIDATE))
8205 +               goto out;
8206 +
8207 +       AuDbg("b%d\n", bindex);
8208 +       /*
8209 +        * gave up supporting LOOKUP_CREATE/OPEN for lower fs,
8210 +        * due to whiteout and branch permission.
8211 +        */
8212 +       flags &= ~(/*LOOKUP_PARENT |*/ LOOKUP_OPEN | LOOKUP_CREATE
8213 +                  | LOOKUP_FOLLOW | LOOKUP_EXCL);
8214 +       /* it may return tri-state */
8215 +       valid = h_dentry->d_op->d_revalidate(h_dentry, flags);
8216 +
8217 +       if (unlikely(valid < 0))
8218 +               err = valid;
8219 +       else if (!valid)
8220 +               err = -EINVAL;
8221 +
8222 +out:
8223 +       AuTraceErr(err);
8224 +       return err;
8225 +}
8226 +
8227 +/* todo: remove this */
8228 +static int h_d_revalidate(struct dentry *dentry, struct inode *inode,
8229 +                         unsigned int flags, int do_udba)
8230 +{
8231 +       int err;
8232 +       umode_t mode, h_mode;
8233 +       aufs_bindex_t bindex, btail, bstart, ibs, ibe;
8234 +       unsigned char plus, unhashed, is_root, h_plus, h_nfs, tmpfile;
8235 +       struct inode *h_inode, *h_cached_inode;
8236 +       struct dentry *h_dentry;
8237 +       struct qstr *name, *h_name;
8238 +
8239 +       err = 0;
8240 +       plus = 0;
8241 +       mode = 0;
8242 +       ibs = -1;
8243 +       ibe = -1;
8244 +       unhashed = !!d_unhashed(dentry);
8245 +       is_root = !!IS_ROOT(dentry);
8246 +       name = &dentry->d_name;
8247 +       tmpfile = au_di(dentry)->di_tmpfile;
8248 +
8249 +       /*
8250 +        * Theoretically, REVAL test should be unnecessary in case of
8251 +        * {FS,I}NOTIFY.
8252 +        * But {fs,i}notify doesn't fire some necessary events,
8253 +        *      IN_ATTRIB for atime/nlink/pageio
8254 +        * Let's do REVAL test too.
8255 +        */
8256 +       if (do_udba && inode) {
8257 +               mode = (inode->i_mode & S_IFMT);
8258 +               plus = (inode->i_nlink > 0);
8259 +               ibs = au_ibstart(inode);
8260 +               ibe = au_ibend(inode);
8261 +       }
8262 +
8263 +       bstart = au_dbstart(dentry);
8264 +       btail = bstart;
8265 +       if (inode && S_ISDIR(inode->i_mode))
8266 +               btail = au_dbtaildir(dentry);
8267 +       for (bindex = bstart; bindex <= btail; bindex++) {
8268 +               h_dentry = au_h_dptr(dentry, bindex);
8269 +               if (!h_dentry)
8270 +                       continue;
8271 +
8272 +               AuDbg("b%d, %pd\n", bindex, h_dentry);
8273 +               h_nfs = !!au_test_nfs(h_dentry->d_sb);
8274 +               spin_lock(&h_dentry->d_lock);
8275 +               h_name = &h_dentry->d_name;
8276 +               if (unlikely(do_udba
8277 +                            && !is_root
8278 +                            && ((!h_nfs
8279 +                                 && (unhashed != !!d_unhashed(h_dentry)
8280 +                                     || (!tmpfile
8281 +                                         && !au_qstreq(name, h_name))
8282 +                                         ))
8283 +                                || (h_nfs
8284 +                                    && !(flags & LOOKUP_OPEN)
8285 +                                    && (h_dentry->d_flags
8286 +                                        & DCACHE_NFSFS_RENAMED)))
8287 +                           )) {
8288 +                       int h_unhashed;
8289 +
8290 +                       h_unhashed = d_unhashed(h_dentry);
8291 +                       spin_unlock(&h_dentry->d_lock);
8292 +                       AuDbg("unhash 0x%x 0x%x, %pd %pd\n",
8293 +                             unhashed, h_unhashed, dentry, h_dentry);
8294 +                       goto err;
8295 +               }
8296 +               spin_unlock(&h_dentry->d_lock);
8297 +
8298 +               err = au_do_h_d_reval(h_dentry, flags, dentry, bindex);
8299 +               if (unlikely(err))
8300 +                       /* do not goto err, to keep the errno */
8301 +                       break;
8302 +
8303 +               /* todo: plink too? */
8304 +               if (!do_udba)
8305 +                       continue;
8306 +
8307 +               /* UDBA tests */
8308 +               h_inode = h_dentry->d_inode;
8309 +               if (unlikely(!!inode != !!h_inode))
8310 +                       goto err;
8311 +
8312 +               h_plus = plus;
8313 +               h_mode = mode;
8314 +               h_cached_inode = h_inode;
8315 +               if (h_inode) {
8316 +                       h_mode = (h_inode->i_mode & S_IFMT);
8317 +                       h_plus = (h_inode->i_nlink > 0);
8318 +               }
8319 +               if (inode && ibs <= bindex && bindex <= ibe)
8320 +                       h_cached_inode = au_h_iptr(inode, bindex);
8321 +
8322 +               if (!h_nfs) {
8323 +                       if (unlikely(plus != h_plus && !tmpfile))
8324 +                               goto err;
8325 +               } else {
8326 +                       if (unlikely(!(h_dentry->d_flags & DCACHE_NFSFS_RENAMED)
8327 +                                    && !is_root
8328 +                                    && !IS_ROOT(h_dentry)
8329 +                                    && unhashed != d_unhashed(h_dentry)))
8330 +                               goto err;
8331 +               }
8332 +               if (unlikely(mode != h_mode
8333 +                            || h_cached_inode != h_inode))
8334 +                       goto err;
8335 +               continue;
8336 +
8337 +err:
8338 +               err = -EINVAL;
8339 +               break;
8340 +       }
8341 +
8342 +       AuTraceErr(err);
8343 +       return err;
8344 +}
8345 +
8346 +/* todo: consolidate with do_refresh() and au_reval_for_attr() */
8347 +static int simple_reval_dpath(struct dentry *dentry, unsigned int sigen)
8348 +{
8349 +       int err;
8350 +       struct dentry *parent;
8351 +
8352 +       if (!au_digen_test(dentry, sigen))
8353 +               return 0;
8354 +
8355 +       parent = dget_parent(dentry);
8356 +       di_read_lock_parent(parent, AuLock_IR);
8357 +       AuDebugOn(au_digen_test(parent, sigen));
8358 +       au_dbg_verify_gen(parent, sigen);
8359 +       err = au_refresh_dentry(dentry, parent);
8360 +       di_read_unlock(parent, AuLock_IR);
8361 +       dput(parent);
8362 +       AuTraceErr(err);
8363 +       return err;
8364 +}
8365 +
8366 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen)
8367 +{
8368 +       int err;
8369 +       struct dentry *d, *parent;
8370 +       struct inode *inode;
8371 +
8372 +       if (!au_ftest_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR))
8373 +               return simple_reval_dpath(dentry, sigen);
8374 +
8375 +       /* slow loop, keep it simple and stupid */
8376 +       /* cf: au_cpup_dirs() */
8377 +       err = 0;
8378 +       parent = NULL;
8379 +       while (au_digen_test(dentry, sigen)) {
8380 +               d = dentry;
8381 +               while (1) {
8382 +                       dput(parent);
8383 +                       parent = dget_parent(d);
8384 +                       if (!au_digen_test(parent, sigen))
8385 +                               break;
8386 +                       d = parent;
8387 +               }
8388 +
8389 +               inode = d->d_inode;
8390 +               if (d != dentry)
8391 +                       di_write_lock_child2(d);
8392 +
8393 +               /* someone might update our dentry while we were sleeping */
8394 +               if (au_digen_test(d, sigen)) {
8395 +                       /*
8396 +                        * todo: consolidate with simple_reval_dpath(),
8397 +                        * do_refresh() and au_reval_for_attr().
8398 +                        */
8399 +                       di_read_lock_parent(parent, AuLock_IR);
8400 +                       err = au_refresh_dentry(d, parent);
8401 +                       di_read_unlock(parent, AuLock_IR);
8402 +               }
8403 +
8404 +               if (d != dentry)
8405 +                       di_write_unlock(d);
8406 +               dput(parent);
8407 +               if (unlikely(err))
8408 +                       break;
8409 +       }
8410 +
8411 +       return err;
8412 +}
8413 +
8414 +/*
8415 + * if valid returns 1, otherwise 0.
8416 + */
8417 +static int aufs_d_revalidate(struct dentry *dentry, unsigned int flags)
8418 +{
8419 +       int valid, err;
8420 +       unsigned int sigen;
8421 +       unsigned char do_udba;
8422 +       struct super_block *sb;
8423 +       struct inode *inode;
8424 +
8425 +       /* todo: support rcu-walk? */
8426 +       if (flags & LOOKUP_RCU)
8427 +               return -ECHILD;
8428 +
8429 +       valid = 0;
8430 +       if (unlikely(!au_di(dentry)))
8431 +               goto out;
8432 +
8433 +       valid = 1;
8434 +       sb = dentry->d_sb;
8435 +       /*
8436 +        * todo: very ugly
8437 +        * i_mutex of parent dir may be held,
8438 +        * but we should not return 'invalid' due to busy.
8439 +        */
8440 +       err = aufs_read_lock(dentry, AuLock_FLUSH | AuLock_DW | AuLock_NOPLM);
8441 +       if (unlikely(err)) {
8442 +               valid = err;
8443 +               AuTraceErr(err);
8444 +               goto out;
8445 +       }
8446 +       inode = dentry->d_inode;
8447 +       if (unlikely(inode && is_bad_inode(inode))) {
8448 +               err = -EINVAL;
8449 +               AuTraceErr(err);
8450 +               goto out_dgrade;
8451 +       }
8452 +       if (unlikely(au_dbrange_test(dentry))) {
8453 +               err = -EINVAL;
8454 +               AuTraceErr(err);
8455 +               goto out_dgrade;
8456 +       }
8457 +
8458 +       sigen = au_sigen(sb);
8459 +       if (au_digen_test(dentry, sigen)) {
8460 +               AuDebugOn(IS_ROOT(dentry));
8461 +               err = au_reval_dpath(dentry, sigen);
8462 +               if (unlikely(err)) {
8463 +                       AuTraceErr(err);
8464 +                       goto out_dgrade;
8465 +               }
8466 +       }
8467 +       di_downgrade_lock(dentry, AuLock_IR);
8468 +
8469 +       err = -EINVAL;
8470 +       if (!(flags & (LOOKUP_OPEN | LOOKUP_EMPTY))
8471 +           && inode
8472 +           && !(inode->i_state && I_LINKABLE)
8473 +           && (IS_DEADDIR(inode) || !inode->i_nlink))
8474 +               goto out_inval;
8475 +
8476 +       do_udba = !au_opt_test(au_mntflags(sb), UDBA_NONE);
8477 +       if (do_udba && inode) {
8478 +               aufs_bindex_t bstart = au_ibstart(inode);
8479 +               struct inode *h_inode;
8480 +
8481 +               if (bstart >= 0) {
8482 +                       h_inode = au_h_iptr(inode, bstart);
8483 +                       if (h_inode && au_test_higen(inode, h_inode))
8484 +                               goto out_inval;
8485 +               }
8486 +       }
8487 +
8488 +       err = h_d_revalidate(dentry, inode, flags, do_udba);
8489 +       if (unlikely(!err && do_udba && au_dbstart(dentry) < 0)) {
8490 +               err = -EIO;
8491 +               AuDbg("both of real entry and whiteout found, %p, err %d\n",
8492 +                     dentry, err);
8493 +       }
8494 +       goto out_inval;
8495 +
8496 +out_dgrade:
8497 +       di_downgrade_lock(dentry, AuLock_IR);
8498 +out_inval:
8499 +       aufs_read_unlock(dentry, AuLock_IR);
8500 +       AuTraceErr(err);
8501 +       valid = !err;
8502 +out:
8503 +       if (!valid) {
8504 +               AuDbg("%pd invalid, %d\n", dentry, valid);
8505 +               d_drop(dentry);
8506 +       }
8507 +       return valid;
8508 +}
8509 +
8510 +static void aufs_d_release(struct dentry *dentry)
8511 +{
8512 +       if (au_di(dentry)) {
8513 +               au_di_fin(dentry);
8514 +               au_hn_di_reinit(dentry);
8515 +       }
8516 +}
8517 +
8518 +const struct dentry_operations aufs_dop = {
8519 +       .d_revalidate           = aufs_d_revalidate,
8520 +       .d_weak_revalidate      = aufs_d_revalidate,
8521 +       .d_release              = aufs_d_release
8522 +};
8523 diff -urN /usr/share/empty/fs/aufs/dentry.h linux/fs/aufs/dentry.h
8524 --- /usr/share/empty/fs/aufs/dentry.h   1970-01-01 01:00:00.000000000 +0100
8525 +++ linux/fs/aufs/dentry.h      2015-01-25 13:00:38.631047076 +0100
8526 @@ -0,0 +1,233 @@
8527 +/*
8528 + * Copyright (C) 2005-2014 Junjiro R. Okajima
8529 + *
8530 + * This program, aufs is free software; you can redistribute it and/or modify
8531 + * it under the terms of the GNU General Public License as published by
8532 + * the Free Software Foundation; either version 2 of the License, or
8533 + * (at your option) any later version.
8534 + *
8535 + * This program is distributed in the hope that it will be useful,
8536 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8537 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8538 + * GNU General Public License for more details.
8539 + *
8540 + * You should have received a copy of the GNU General Public License
8541 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8542 + */
8543 +
8544 +/*
8545 + * lookup and dentry operations
8546 + */
8547 +
8548 +#ifndef __AUFS_DENTRY_H__
8549 +#define __AUFS_DENTRY_H__
8550 +
8551 +#ifdef __KERNEL__
8552 +
8553 +#include <linux/dcache.h>
8554 +#include "rwsem.h"
8555 +
8556 +struct au_hdentry {
8557 +       struct dentry           *hd_dentry;
8558 +       aufs_bindex_t           hd_id;
8559 +};
8560 +
8561 +struct au_dinfo {
8562 +       atomic_t                di_generation;
8563 +
8564 +       struct au_rwsem         di_rwsem;
8565 +       aufs_bindex_t           di_bstart, di_bend, di_bwh, di_bdiropq;
8566 +       unsigned char           di_tmpfile; /* to allow the different name */
8567 +       struct au_hdentry       *di_hdentry;
8568 +} ____cacheline_aligned_in_smp;
8569 +
8570 +/* ---------------------------------------------------------------------- */
8571 +
8572 +/* dentry.c */
8573 +extern const struct dentry_operations aufs_dop;
8574 +struct au_branch;
8575 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent);
8576 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
8577 +               struct dentry *h_parent, struct au_branch *br);
8578 +
8579 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t bstart, mode_t type);
8580 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh);
8581 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent);
8582 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen);
8583 +
8584 +/* dinfo.c */
8585 +void au_di_init_once(void *_di);
8586 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc);
8587 +void au_di_free(struct au_dinfo *dinfo);
8588 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b);
8589 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src);
8590 +int au_di_init(struct dentry *dentry);
8591 +void au_di_fin(struct dentry *dentry);
8592 +int au_di_realloc(struct au_dinfo *dinfo, int nbr);
8593 +
8594 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc);
8595 +void di_read_unlock(struct dentry *d, int flags);
8596 +void di_downgrade_lock(struct dentry *d, int flags);
8597 +void di_write_lock(struct dentry *d, unsigned int lsc);
8598 +void di_write_unlock(struct dentry *d);
8599 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir);
8600 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir);
8601 +void di_write_unlock2(struct dentry *d1, struct dentry *d2);
8602 +
8603 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex);
8604 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex);
8605 +aufs_bindex_t au_dbtail(struct dentry *dentry);
8606 +aufs_bindex_t au_dbtaildir(struct dentry *dentry);
8607 +
8608 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
8609 +                  struct dentry *h_dentry);
8610 +int au_digen_test(struct dentry *dentry, unsigned int sigen);
8611 +int au_dbrange_test(struct dentry *dentry);
8612 +void au_update_digen(struct dentry *dentry);
8613 +void au_update_dbrange(struct dentry *dentry, int do_put_zero);
8614 +void au_update_dbstart(struct dentry *dentry);
8615 +void au_update_dbend(struct dentry *dentry);
8616 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry);
8617 +
8618 +/* ---------------------------------------------------------------------- */
8619 +
8620 +static inline struct au_dinfo *au_di(struct dentry *dentry)
8621 +{
8622 +       return dentry->d_fsdata;
8623 +}
8624 +
8625 +/* ---------------------------------------------------------------------- */
8626 +
8627 +/* lock subclass for dinfo */
8628 +enum {
8629 +       AuLsc_DI_CHILD,         /* child first */
8630 +       AuLsc_DI_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
8631 +       AuLsc_DI_CHILD3,        /* copyup dirs */
8632 +       AuLsc_DI_PARENT,
8633 +       AuLsc_DI_PARENT2,
8634 +       AuLsc_DI_PARENT3,
8635 +       AuLsc_DI_TMP            /* temp for replacing dinfo */
8636 +};
8637 +
8638 +/*
8639 + * di_read_lock_child, di_write_lock_child,
8640 + * di_read_lock_child2, di_write_lock_child2,
8641 + * di_read_lock_child3, di_write_lock_child3,
8642 + * di_read_lock_parent, di_write_lock_parent,
8643 + * di_read_lock_parent2, di_write_lock_parent2,
8644 + * di_read_lock_parent3, di_write_lock_parent3,
8645 + */
8646 +#define AuReadLockFunc(name, lsc) \
8647 +static inline void di_read_lock_##name(struct dentry *d, int flags) \
8648 +{ di_read_lock(d, flags, AuLsc_DI_##lsc); }
8649 +
8650 +#define AuWriteLockFunc(name, lsc) \
8651 +static inline void di_write_lock_##name(struct dentry *d) \
8652 +{ di_write_lock(d, AuLsc_DI_##lsc); }
8653 +
8654 +#define AuRWLockFuncs(name, lsc) \
8655 +       AuReadLockFunc(name, lsc) \
8656 +       AuWriteLockFunc(name, lsc)
8657 +
8658 +AuRWLockFuncs(child, CHILD);
8659 +AuRWLockFuncs(child2, CHILD2);
8660 +AuRWLockFuncs(child3, CHILD3);
8661 +AuRWLockFuncs(parent, PARENT);
8662 +AuRWLockFuncs(parent2, PARENT2);
8663 +AuRWLockFuncs(parent3, PARENT3);
8664 +
8665 +#undef AuReadLockFunc
8666 +#undef AuWriteLockFunc
8667 +#undef AuRWLockFuncs
8668 +
8669 +#define DiMustNoWaiters(d)     AuRwMustNoWaiters(&au_di(d)->di_rwsem)
8670 +#define DiMustAnyLock(d)       AuRwMustAnyLock(&au_di(d)->di_rwsem)
8671 +#define DiMustWriteLock(d)     AuRwMustWriteLock(&au_di(d)->di_rwsem)
8672 +
8673 +/* ---------------------------------------------------------------------- */
8674 +
8675 +/* todo: memory barrier? */
8676 +static inline unsigned int au_digen(struct dentry *d)
8677 +{
8678 +       return atomic_read(&au_di(d)->di_generation);
8679 +}
8680 +
8681 +static inline void au_h_dentry_init(struct au_hdentry *hdentry)
8682 +{
8683 +       hdentry->hd_dentry = NULL;
8684 +}
8685 +
8686 +static inline void au_hdput(struct au_hdentry *hd)
8687 +{
8688 +       if (hd)
8689 +               dput(hd->hd_dentry);
8690 +}
8691 +
8692 +static inline aufs_bindex_t au_dbstart(struct dentry *dentry)
8693 +{
8694 +       DiMustAnyLock(dentry);
8695 +       return au_di(dentry)->di_bstart;
8696 +}
8697 +
8698 +static inline aufs_bindex_t au_dbend(struct dentry *dentry)
8699 +{
8700 +       DiMustAnyLock(dentry);
8701 +       return au_di(dentry)->di_bend;
8702 +}
8703 +
8704 +static inline aufs_bindex_t au_dbwh(struct dentry *dentry)
8705 +{
8706 +       DiMustAnyLock(dentry);
8707 +       return au_di(dentry)->di_bwh;
8708 +}
8709 +
8710 +static inline aufs_bindex_t au_dbdiropq(struct dentry *dentry)
8711 +{
8712 +       DiMustAnyLock(dentry);
8713 +       return au_di(dentry)->di_bdiropq;
8714 +}
8715 +
8716 +/* todo: hard/soft set? */
8717 +static inline void au_set_dbstart(struct dentry *dentry, aufs_bindex_t bindex)
8718 +{
8719 +       DiMustWriteLock(dentry);
8720 +       au_di(dentry)->di_bstart = bindex;
8721 +}
8722 +
8723 +static inline void au_set_dbend(struct dentry *dentry, aufs_bindex_t bindex)
8724 +{
8725 +       DiMustWriteLock(dentry);
8726 +       au_di(dentry)->di_bend = bindex;
8727 +}
8728 +
8729 +static inline void au_set_dbwh(struct dentry *dentry, aufs_bindex_t bindex)
8730 +{
8731 +       DiMustWriteLock(dentry);
8732 +       /* dbwh can be outside of bstart - bend range */
8733 +       au_di(dentry)->di_bwh = bindex;
8734 +}
8735 +
8736 +static inline void au_set_dbdiropq(struct dentry *dentry, aufs_bindex_t bindex)
8737 +{
8738 +       DiMustWriteLock(dentry);
8739 +       au_di(dentry)->di_bdiropq = bindex;
8740 +}
8741 +
8742 +/* ---------------------------------------------------------------------- */
8743 +
8744 +#ifdef CONFIG_AUFS_HNOTIFY
8745 +static inline void au_digen_dec(struct dentry *d)
8746 +{
8747 +       atomic_dec(&au_di(d)->di_generation);
8748 +}
8749 +
8750 +static inline void au_hn_di_reinit(struct dentry *dentry)
8751 +{
8752 +       dentry->d_fsdata = NULL;
8753 +}
8754 +#else
8755 +AuStubVoid(au_hn_di_reinit, struct dentry *dentry __maybe_unused)
8756 +#endif /* CONFIG_AUFS_HNOTIFY */
8757 +
8758 +#endif /* __KERNEL__ */
8759 +#endif /* __AUFS_DENTRY_H__ */
8760 diff -urN /usr/share/empty/fs/aufs/dinfo.c linux/fs/aufs/dinfo.c
8761 --- /usr/share/empty/fs/aufs/dinfo.c    1970-01-01 01:00:00.000000000 +0100
8762 +++ linux/fs/aufs/dinfo.c       2015-01-25 13:00:38.631047076 +0100
8763 @@ -0,0 +1,544 @@
8764 +/*
8765 + * Copyright (C) 2005-2014 Junjiro R. Okajima
8766 + *
8767 + * This program, aufs is free software; you can redistribute it and/or modify
8768 + * it under the terms of the GNU General Public License as published by
8769 + * the Free Software Foundation; either version 2 of the License, or
8770 + * (at your option) any later version.
8771 + *
8772 + * This program is distributed in the hope that it will be useful,
8773 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8774 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8775 + * GNU General Public License for more details.
8776 + *
8777 + * You should have received a copy of the GNU General Public License
8778 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8779 + */
8780 +
8781 +/*
8782 + * dentry private data
8783 + */
8784 +
8785 +#include "aufs.h"
8786 +
8787 +void au_di_init_once(void *_dinfo)
8788 +{
8789 +       struct au_dinfo *dinfo = _dinfo;
8790 +       static struct lock_class_key aufs_di;
8791 +
8792 +       au_rw_init(&dinfo->di_rwsem);
8793 +       au_rw_class(&dinfo->di_rwsem, &aufs_di);
8794 +}
8795 +
8796 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc)
8797 +{
8798 +       struct au_dinfo *dinfo;
8799 +       int nbr, i;
8800 +
8801 +       dinfo = au_cache_alloc_dinfo();
8802 +       if (unlikely(!dinfo))
8803 +               goto out;
8804 +
8805 +       nbr = au_sbend(sb) + 1;
8806 +       if (nbr <= 0)
8807 +               nbr = 1;
8808 +       dinfo->di_hdentry = kcalloc(nbr, sizeof(*dinfo->di_hdentry), GFP_NOFS);
8809 +       if (dinfo->di_hdentry) {
8810 +               au_rw_write_lock_nested(&dinfo->di_rwsem, lsc);
8811 +               dinfo->di_bstart = -1;
8812 +               dinfo->di_bend = -1;
8813 +               dinfo->di_bwh = -1;
8814 +               dinfo->di_bdiropq = -1;
8815 +               dinfo->di_tmpfile = 0;
8816 +               for (i = 0; i < nbr; i++)
8817 +                       dinfo->di_hdentry[i].hd_id = -1;
8818 +               goto out;
8819 +       }
8820 +
8821 +       au_cache_free_dinfo(dinfo);
8822 +       dinfo = NULL;
8823 +
8824 +out:
8825 +       return dinfo;
8826 +}
8827 +
8828 +void au_di_free(struct au_dinfo *dinfo)
8829 +{
8830 +       struct au_hdentry *p;
8831 +       aufs_bindex_t bend, bindex;
8832 +
8833 +       /* dentry may not be revalidated */
8834 +       bindex = dinfo->di_bstart;
8835 +       if (bindex >= 0) {
8836 +               bend = dinfo->di_bend;
8837 +               p = dinfo->di_hdentry + bindex;
8838 +               while (bindex++ <= bend)
8839 +                       au_hdput(p++);
8840 +       }
8841 +       kfree(dinfo->di_hdentry);
8842 +       au_cache_free_dinfo(dinfo);
8843 +}
8844 +
8845 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b)
8846 +{
8847 +       struct au_hdentry *p;
8848 +       aufs_bindex_t bi;
8849 +
8850 +       AuRwMustWriteLock(&a->di_rwsem);
8851 +       AuRwMustWriteLock(&b->di_rwsem);
8852 +
8853 +#define DiSwap(v, name)                                \
8854 +       do {                                    \
8855 +               v = a->di_##name;               \
8856 +               a->di_##name = b->di_##name;    \
8857 +               b->di_##name = v;               \
8858 +       } while (0)
8859 +
8860 +       DiSwap(p, hdentry);
8861 +       DiSwap(bi, bstart);
8862 +       DiSwap(bi, bend);
8863 +       DiSwap(bi, bwh);
8864 +       DiSwap(bi, bdiropq);
8865 +       /* smp_mb(); */
8866 +
8867 +#undef DiSwap
8868 +}
8869 +
8870 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src)
8871 +{
8872 +       AuRwMustWriteLock(&dst->di_rwsem);
8873 +       AuRwMustWriteLock(&src->di_rwsem);
8874 +
8875 +       dst->di_bstart = src->di_bstart;
8876 +       dst->di_bend = src->di_bend;
8877 +       dst->di_bwh = src->di_bwh;
8878 +       dst->di_bdiropq = src->di_bdiropq;
8879 +       /* smp_mb(); */
8880 +}
8881 +
8882 +int au_di_init(struct dentry *dentry)
8883 +{
8884 +       int err;
8885 +       struct super_block *sb;
8886 +       struct au_dinfo *dinfo;
8887 +
8888 +       err = 0;
8889 +       sb = dentry->d_sb;
8890 +       dinfo = au_di_alloc(sb, AuLsc_DI_CHILD);
8891 +       if (dinfo) {
8892 +               atomic_set(&dinfo->di_generation, au_sigen(sb));
8893 +               /* smp_mb(); */ /* atomic_set */
8894 +               dentry->d_fsdata = dinfo;
8895 +       } else
8896 +               err = -ENOMEM;
8897 +
8898 +       return err;
8899 +}
8900 +
8901 +void au_di_fin(struct dentry *dentry)
8902 +{
8903 +       struct au_dinfo *dinfo;
8904 +
8905 +       dinfo = au_di(dentry);
8906 +       AuRwDestroy(&dinfo->di_rwsem);
8907 +       au_di_free(dinfo);
8908 +}
8909 +
8910 +int au_di_realloc(struct au_dinfo *dinfo, int nbr)
8911 +{
8912 +       int err, sz;
8913 +       struct au_hdentry *hdp;
8914 +
8915 +       AuRwMustWriteLock(&dinfo->di_rwsem);
8916 +
8917 +       err = -ENOMEM;
8918 +       sz = sizeof(*hdp) * (dinfo->di_bend + 1);
8919 +       if (!sz)
8920 +               sz = sizeof(*hdp);
8921 +       hdp = au_kzrealloc(dinfo->di_hdentry, sz, sizeof(*hdp) * nbr, GFP_NOFS);
8922 +       if (hdp) {
8923 +               dinfo->di_hdentry = hdp;
8924 +               err = 0;
8925 +       }
8926 +
8927 +       return err;
8928 +}
8929 +
8930 +/* ---------------------------------------------------------------------- */
8931 +
8932 +static void do_ii_write_lock(struct inode *inode, unsigned int lsc)
8933 +{
8934 +       switch (lsc) {
8935 +       case AuLsc_DI_CHILD:
8936 +               ii_write_lock_child(inode);
8937 +               break;
8938 +       case AuLsc_DI_CHILD2:
8939 +               ii_write_lock_child2(inode);
8940 +               break;
8941 +       case AuLsc_DI_CHILD3:
8942 +               ii_write_lock_child3(inode);
8943 +               break;
8944 +       case AuLsc_DI_PARENT:
8945 +               ii_write_lock_parent(inode);
8946 +               break;
8947 +       case AuLsc_DI_PARENT2:
8948 +               ii_write_lock_parent2(inode);
8949 +               break;
8950 +       case AuLsc_DI_PARENT3:
8951 +               ii_write_lock_parent3(inode);
8952 +               break;
8953 +       default:
8954 +               BUG();
8955 +       }
8956 +}
8957 +
8958 +static void do_ii_read_lock(struct inode *inode, unsigned int lsc)
8959 +{
8960 +       switch (lsc) {
8961 +       case AuLsc_DI_CHILD:
8962 +               ii_read_lock_child(inode);
8963 +               break;
8964 +       case AuLsc_DI_CHILD2:
8965 +               ii_read_lock_child2(inode);
8966 +               break;
8967 +       case AuLsc_DI_CHILD3:
8968 +               ii_read_lock_child3(inode);
8969 +               break;
8970 +       case AuLsc_DI_PARENT:
8971 +               ii_read_lock_parent(inode);
8972 +               break;
8973 +       case AuLsc_DI_PARENT2:
8974 +               ii_read_lock_parent2(inode);
8975 +               break;
8976 +       case AuLsc_DI_PARENT3:
8977 +               ii_read_lock_parent3(inode);
8978 +               break;
8979 +       default:
8980 +               BUG();
8981 +       }
8982 +}
8983 +
8984 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc)
8985 +{
8986 +       au_rw_read_lock_nested(&au_di(d)->di_rwsem, lsc);
8987 +       if (d->d_inode) {
8988 +               if (au_ftest_lock(flags, IW))
8989 +                       do_ii_write_lock(d->d_inode, lsc);
8990 +               else if (au_ftest_lock(flags, IR))
8991 +                       do_ii_read_lock(d->d_inode, lsc);
8992 +       }
8993 +}
8994 +
8995 +void di_read_unlock(struct dentry *d, int flags)
8996 +{
8997 +       if (d->d_inode) {
8998 +               if (au_ftest_lock(flags, IW)) {
8999 +                       au_dbg_verify_dinode(d);
9000 +                       ii_write_unlock(d->d_inode);
9001 +               } else if (au_ftest_lock(flags, IR)) {
9002 +                       au_dbg_verify_dinode(d);
9003 +                       ii_read_unlock(d->d_inode);
9004 +               }
9005 +       }
9006 +       au_rw_read_unlock(&au_di(d)->di_rwsem);
9007 +}
9008 +
9009 +void di_downgrade_lock(struct dentry *d, int flags)
9010 +{
9011 +       if (d->d_inode && au_ftest_lock(flags, IR))
9012 +               ii_downgrade_lock(d->d_inode);
9013 +       au_rw_dgrade_lock(&au_di(d)->di_rwsem);
9014 +}
9015 +
9016 +void di_write_lock(struct dentry *d, unsigned int lsc)
9017 +{
9018 +       au_rw_write_lock_nested(&au_di(d)->di_rwsem, lsc);
9019 +       if (d->d_inode)
9020 +               do_ii_write_lock(d->d_inode, lsc);
9021 +}
9022 +
9023 +void di_write_unlock(struct dentry *d)
9024 +{
9025 +       au_dbg_verify_dinode(d);
9026 +       if (d->d_inode)
9027 +               ii_write_unlock(d->d_inode);
9028 +       au_rw_write_unlock(&au_di(d)->di_rwsem);
9029 +}
9030 +
9031 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir)
9032 +{
9033 +       AuDebugOn(d1 == d2
9034 +                 || d1->d_inode == d2->d_inode
9035 +                 || d1->d_sb != d2->d_sb);
9036 +
9037 +       if (isdir && au_test_subdir(d1, d2)) {
9038 +               di_write_lock_child(d1);
9039 +               di_write_lock_child2(d2);
9040 +       } else {
9041 +               /* there should be no races */
9042 +               di_write_lock_child(d2);
9043 +               di_write_lock_child2(d1);
9044 +       }
9045 +}
9046 +
9047 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir)
9048 +{
9049 +       AuDebugOn(d1 == d2
9050 +                 || d1->d_inode == d2->d_inode
9051 +                 || d1->d_sb != d2->d_sb);
9052 +
9053 +       if (isdir && au_test_subdir(d1, d2)) {
9054 +               di_write_lock_parent(d1);
9055 +               di_write_lock_parent2(d2);
9056 +       } else {
9057 +               /* there should be no races */
9058 +               di_write_lock_parent(d2);
9059 +               di_write_lock_parent2(d1);
9060 +       }
9061 +}
9062 +
9063 +void di_write_unlock2(struct dentry *d1, struct dentry *d2)
9064 +{
9065 +       di_write_unlock(d1);
9066 +       if (d1->d_inode == d2->d_inode)
9067 +               au_rw_write_unlock(&au_di(d2)->di_rwsem);
9068 +       else
9069 +               di_write_unlock(d2);
9070 +}
9071 +
9072 +/* ---------------------------------------------------------------------- */
9073 +
9074 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex)
9075 +{
9076 +       struct dentry *d;
9077 +
9078 +       DiMustAnyLock(dentry);
9079 +
9080 +       if (au_dbstart(dentry) < 0 || bindex < au_dbstart(dentry))
9081 +               return NULL;
9082 +       AuDebugOn(bindex < 0);
9083 +       d = au_di(dentry)->di_hdentry[0 + bindex].hd_dentry;
9084 +       AuDebugOn(d && au_dcount(d) <= 0);
9085 +       return d;
9086 +}
9087 +
9088 +/*
9089 + * extended version of au_h_dptr().
9090 + * returns a hashed and positive (or linkable) h_dentry in bindex, NULL, or
9091 + * error.
9092 + */
9093 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex)
9094 +{
9095 +       struct dentry *h_dentry;
9096 +       struct inode *inode, *h_inode;
9097 +
9098 +       inode = dentry->d_inode;
9099 +       AuDebugOn(!inode);
9100 +
9101 +       h_dentry = NULL;
9102 +       if (au_dbstart(dentry) <= bindex
9103 +           && bindex <= au_dbend(dentry))
9104 +               h_dentry = au_h_dptr(dentry, bindex);
9105 +       if (h_dentry && !au_d_linkable(h_dentry)) {
9106 +               dget(h_dentry);
9107 +               goto out; /* success */
9108 +       }
9109 +
9110 +       AuDebugOn(bindex < au_ibstart(inode));
9111 +       AuDebugOn(au_ibend(inode) < bindex);
9112 +       h_inode = au_h_iptr(inode, bindex);
9113 +       h_dentry = d_find_alias(h_inode);
9114 +       if (h_dentry) {
9115 +               if (!IS_ERR(h_dentry)) {
9116 +                       if (!au_d_linkable(h_dentry))
9117 +                               goto out; /* success */
9118 +                       dput(h_dentry);
9119 +               } else
9120 +                       goto out;
9121 +       }
9122 +
9123 +       if (au_opt_test(au_mntflags(dentry->d_sb), PLINK)) {
9124 +               h_dentry = au_plink_lkup(inode, bindex);
9125 +               AuDebugOn(!h_dentry);
9126 +               if (!IS_ERR(h_dentry)) {
9127 +                       if (!au_d_hashed_positive(h_dentry))
9128 +                               goto out; /* success */
9129 +                       dput(h_dentry);
9130 +                       h_dentry = NULL;
9131 +               }
9132 +       }
9133 +
9134 +out:
9135 +       AuDbgDentry(h_dentry);
9136 +       return h_dentry;
9137 +}
9138 +
9139 +aufs_bindex_t au_dbtail(struct dentry *dentry)
9140 +{
9141 +       aufs_bindex_t bend, bwh;
9142 +
9143 +       bend = au_dbend(dentry);
9144 +       if (0 <= bend) {
9145 +               bwh = au_dbwh(dentry);
9146 +               if (!bwh)
9147 +                       return bwh;
9148 +               if (0 < bwh && bwh < bend)
9149 +                       return bwh - 1;
9150 +       }
9151 +       return bend;
9152 +}
9153 +
9154 +aufs_bindex_t au_dbtaildir(struct dentry *dentry)
9155 +{
9156 +       aufs_bindex_t bend, bopq;
9157 +
9158 +       bend = au_dbtail(dentry);
9159 +       if (0 <= bend) {
9160 +               bopq = au_dbdiropq(dentry);
9161 +               if (0 <= bopq && bopq < bend)
9162 +                       bend = bopq;
9163 +       }
9164 +       return bend;
9165 +}
9166 +
9167 +/* ---------------------------------------------------------------------- */
9168 +
9169 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
9170 +                  struct dentry *h_dentry)
9171 +{
9172 +       struct au_hdentry *hd = au_di(dentry)->di_hdentry + bindex;
9173 +       struct au_branch *br;
9174 +
9175 +       DiMustWriteLock(dentry);
9176 +
9177 +       au_hdput(hd);
9178 +       hd->hd_dentry = h_dentry;
9179 +       if (h_dentry) {
9180 +               br = au_sbr(dentry->d_sb, bindex);
9181 +               hd->hd_id = br->br_id;
9182 +       }
9183 +}
9184 +
9185 +int au_dbrange_test(struct dentry *dentry)
9186 +{
9187 +       int err;
9188 +       aufs_bindex_t bstart, bend;
9189 +
9190 +       err = 0;
9191 +       bstart = au_dbstart(dentry);
9192 +       bend = au_dbend(dentry);
9193 +       if (bstart >= 0)
9194 +               AuDebugOn(bend < 0 && bstart > bend);
9195 +       else {
9196 +               err = -EIO;
9197 +               AuDebugOn(bend >= 0);
9198 +       }
9199 +
9200 +       return err;
9201 +}
9202 +
9203 +int au_digen_test(struct dentry *dentry, unsigned int sigen)
9204 +{
9205 +       int err;
9206 +
9207 +       err = 0;
9208 +       if (unlikely(au_digen(dentry) != sigen
9209 +                    || au_iigen_test(dentry->d_inode, sigen)))
9210 +               err = -EIO;
9211 +
9212 +       return err;
9213 +}
9214 +
9215 +void au_update_digen(struct dentry *dentry)
9216 +{
9217 +       atomic_set(&au_di(dentry)->di_generation, au_sigen(dentry->d_sb));
9218 +       /* smp_mb(); */ /* atomic_set */
9219 +}
9220 +
9221 +void au_update_dbrange(struct dentry *dentry, int do_put_zero)
9222 +{
9223 +       struct au_dinfo *dinfo;
9224 +       struct dentry *h_d;
9225 +       struct au_hdentry *hdp;
9226 +
9227 +       DiMustWriteLock(dentry);
9228 +
9229 +       dinfo = au_di(dentry);
9230 +       if (!dinfo || dinfo->di_bstart < 0)
9231 +               return;
9232 +
9233 +       hdp = dinfo->di_hdentry;
9234 +       if (do_put_zero) {
9235 +               aufs_bindex_t bindex, bend;
9236 +
9237 +               bend = dinfo->di_bend;
9238 +               for (bindex = dinfo->di_bstart; bindex <= bend; bindex++) {
9239 +                       h_d = hdp[0 + bindex].hd_dentry;
9240 +                       if (h_d && !h_d->d_inode)
9241 +                               au_set_h_dptr(dentry, bindex, NULL);
9242 +               }
9243 +       }
9244 +
9245 +       dinfo->di_bstart = -1;
9246 +       while (++dinfo->di_bstart <= dinfo->di_bend)
9247 +               if (hdp[0 + dinfo->di_bstart].hd_dentry)
9248 +                       break;
9249 +       if (dinfo->di_bstart > dinfo->di_bend) {
9250 +               dinfo->di_bstart = -1;
9251 +               dinfo->di_bend = -1;
9252 +               return;
9253 +       }
9254 +
9255 +       dinfo->di_bend++;
9256 +       while (0 <= --dinfo->di_bend)
9257 +               if (hdp[0 + dinfo->di_bend].hd_dentry)
9258 +                       break;
9259 +       AuDebugOn(dinfo->di_bstart > dinfo->di_bend || dinfo->di_bend < 0);
9260 +}
9261 +
9262 +void au_update_dbstart(struct dentry *dentry)
9263 +{
9264 +       aufs_bindex_t bindex, bend;
9265 +       struct dentry *h_dentry;
9266 +
9267 +       bend = au_dbend(dentry);
9268 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++) {
9269 +               h_dentry = au_h_dptr(dentry, bindex);
9270 +               if (!h_dentry)
9271 +                       continue;
9272 +               if (h_dentry->d_inode) {
9273 +                       au_set_dbstart(dentry, bindex);
9274 +                       return;
9275 +               }
9276 +               au_set_h_dptr(dentry, bindex, NULL);
9277 +       }
9278 +}
9279 +
9280 +void au_update_dbend(struct dentry *dentry)
9281 +{
9282 +       aufs_bindex_t bindex, bstart;
9283 +       struct dentry *h_dentry;
9284 +
9285 +       bstart = au_dbstart(dentry);
9286 +       for (bindex = au_dbend(dentry); bindex >= bstart; bindex--) {
9287 +               h_dentry = au_h_dptr(dentry, bindex);
9288 +               if (!h_dentry)
9289 +                       continue;
9290 +               if (h_dentry->d_inode) {
9291 +                       au_set_dbend(dentry, bindex);
9292 +                       return;
9293 +               }
9294 +               au_set_h_dptr(dentry, bindex, NULL);
9295 +       }
9296 +}
9297 +
9298 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry)
9299 +{
9300 +       aufs_bindex_t bindex, bend;
9301 +
9302 +       bend = au_dbend(dentry);
9303 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++)
9304 +               if (au_h_dptr(dentry, bindex) == h_dentry)
9305 +                       return bindex;
9306 +       return -1;
9307 +}
9308 diff -urN /usr/share/empty/fs/aufs/dir.c linux/fs/aufs/dir.c
9309 --- /usr/share/empty/fs/aufs/dir.c      1970-01-01 01:00:00.000000000 +0100
9310 +++ linux/fs/aufs/dir.c 2015-01-25 13:00:38.631047076 +0100
9311 @@ -0,0 +1,645 @@
9312 +/*
9313 + * Copyright (C) 2005-2014 Junjiro R. Okajima
9314 + *
9315 + * This program, aufs is free software; you can redistribute it and/or modify
9316 + * it under the terms of the GNU General Public License as published by
9317 + * the Free Software Foundation; either version 2 of the License, or
9318 + * (at your option) any later version.
9319 + *
9320 + * This program is distributed in the hope that it will be useful,
9321 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9322 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9323 + * GNU General Public License for more details.
9324 + *
9325 + * You should have received a copy of the GNU General Public License
9326 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9327 + */
9328 +
9329 +/*
9330 + * directory operations
9331 + */
9332 +
9333 +#include <linux/fs_stack.h>
9334 +#include "aufs.h"
9335 +
9336 +void au_add_nlink(struct inode *dir, struct inode *h_dir)
9337 +{
9338 +       unsigned int nlink;
9339 +
9340 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9341 +
9342 +       nlink = dir->i_nlink;
9343 +       nlink += h_dir->i_nlink - 2;
9344 +       if (h_dir->i_nlink < 2)
9345 +               nlink += 2;
9346 +       smp_mb(); /* for i_nlink */
9347 +       /* 0 can happen in revaliding */
9348 +       set_nlink(dir, nlink);
9349 +}
9350 +
9351 +void au_sub_nlink(struct inode *dir, struct inode *h_dir)
9352 +{
9353 +       unsigned int nlink;
9354 +
9355 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9356 +
9357 +       nlink = dir->i_nlink;
9358 +       nlink -= h_dir->i_nlink - 2;
9359 +       if (h_dir->i_nlink < 2)
9360 +               nlink -= 2;
9361 +       smp_mb(); /* for i_nlink */
9362 +       /* nlink == 0 means the branch-fs is broken */
9363 +       set_nlink(dir, nlink);
9364 +}
9365 +
9366 +loff_t au_dir_size(struct file *file, struct dentry *dentry)
9367 +{
9368 +       loff_t sz;
9369 +       aufs_bindex_t bindex, bend;
9370 +       struct file *h_file;
9371 +       struct dentry *h_dentry;
9372 +
9373 +       sz = 0;
9374 +       if (file) {
9375 +               AuDebugOn(!file_inode(file));
9376 +               AuDebugOn(!S_ISDIR(file_inode(file)->i_mode));
9377 +
9378 +               bend = au_fbend_dir(file);
9379 +               for (bindex = au_fbstart(file);
9380 +                    bindex <= bend && sz < KMALLOC_MAX_SIZE;
9381 +                    bindex++) {
9382 +                       h_file = au_hf_dir(file, bindex);
9383 +                       if (h_file && file_inode(h_file))
9384 +                               sz += vfsub_f_size_read(h_file);
9385 +               }
9386 +       } else {
9387 +               AuDebugOn(!dentry);
9388 +               AuDebugOn(!dentry->d_inode);
9389 +               AuDebugOn(!S_ISDIR(dentry->d_inode->i_mode));
9390 +
9391 +               bend = au_dbtaildir(dentry);
9392 +               for (bindex = au_dbstart(dentry);
9393 +                    bindex <= bend && sz < KMALLOC_MAX_SIZE;
9394 +                    bindex++) {
9395 +                       h_dentry = au_h_dptr(dentry, bindex);
9396 +                       if (h_dentry && h_dentry->d_inode)
9397 +                               sz += i_size_read(h_dentry->d_inode);
9398 +               }
9399 +       }
9400 +       if (sz < KMALLOC_MAX_SIZE)
9401 +               sz = roundup_pow_of_two(sz);
9402 +       if (sz > KMALLOC_MAX_SIZE)
9403 +               sz = KMALLOC_MAX_SIZE;
9404 +       else if (sz < NAME_MAX) {
9405 +               BUILD_BUG_ON(AUFS_RDBLK_DEF < NAME_MAX);
9406 +               sz = AUFS_RDBLK_DEF;
9407 +       }
9408 +       return sz;
9409 +}
9410 +
9411 +/* ---------------------------------------------------------------------- */
9412 +
9413 +static int reopen_dir(struct file *file)
9414 +{
9415 +       int err;
9416 +       unsigned int flags;
9417 +       aufs_bindex_t bindex, btail, bstart;
9418 +       struct dentry *dentry, *h_dentry;
9419 +       struct file *h_file;
9420 +
9421 +       /* open all lower dirs */
9422 +       dentry = file->f_dentry;
9423 +       bstart = au_dbstart(dentry);
9424 +       for (bindex = au_fbstart(file); bindex < bstart; bindex++)
9425 +               au_set_h_fptr(file, bindex, NULL);
9426 +       au_set_fbstart(file, bstart);
9427 +
9428 +       btail = au_dbtaildir(dentry);
9429 +       for (bindex = au_fbend_dir(file); btail < bindex; bindex--)
9430 +               au_set_h_fptr(file, bindex, NULL);
9431 +       au_set_fbend_dir(file, btail);
9432 +
9433 +       flags = vfsub_file_flags(file);
9434 +       for (bindex = bstart; bindex <= btail; bindex++) {
9435 +               h_dentry = au_h_dptr(dentry, bindex);
9436 +               if (!h_dentry)
9437 +                       continue;
9438 +               h_file = au_hf_dir(file, bindex);
9439 +               if (h_file)
9440 +                       continue;
9441 +
9442 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9443 +               err = PTR_ERR(h_file);
9444 +               if (IS_ERR(h_file))
9445 +                       goto out; /* close all? */
9446 +               au_set_h_fptr(file, bindex, h_file);
9447 +       }
9448 +       au_update_figen(file);
9449 +       /* todo: necessary? */
9450 +       /* file->f_ra = h_file->f_ra; */
9451 +       err = 0;
9452 +
9453 +out:
9454 +       return err;
9455 +}
9456 +
9457 +static int do_open_dir(struct file *file, int flags)
9458 +{
9459 +       int err;
9460 +       aufs_bindex_t bindex, btail;
9461 +       struct dentry *dentry, *h_dentry;
9462 +       struct file *h_file;
9463 +
9464 +       FiMustWriteLock(file);
9465 +
9466 +       err = 0;
9467 +       dentry = file->f_dentry;
9468 +       file->f_version = dentry->d_inode->i_version;
9469 +       bindex = au_dbstart(dentry);
9470 +       au_set_fbstart(file, bindex);
9471 +       btail = au_dbtaildir(dentry);
9472 +       au_set_fbend_dir(file, btail);
9473 +       for (; !err && bindex <= btail; bindex++) {
9474 +               h_dentry = au_h_dptr(dentry, bindex);
9475 +               if (!h_dentry)
9476 +                       continue;
9477 +
9478 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9479 +               if (IS_ERR(h_file)) {
9480 +                       err = PTR_ERR(h_file);
9481 +                       break;
9482 +               }
9483 +               au_set_h_fptr(file, bindex, h_file);
9484 +       }
9485 +       au_update_figen(file);
9486 +       /* todo: necessary? */
9487 +       /* file->f_ra = h_file->f_ra; */
9488 +       if (!err)
9489 +               return 0; /* success */
9490 +
9491 +       /* close all */
9492 +       for (bindex = au_fbstart(file); bindex <= btail; bindex++)
9493 +               au_set_h_fptr(file, bindex, NULL);
9494 +       au_set_fbstart(file, -1);
9495 +       au_set_fbend_dir(file, -1);
9496 +
9497 +       return err;
9498 +}
9499 +
9500 +static int aufs_open_dir(struct inode *inode __maybe_unused,
9501 +                        struct file *file)
9502 +{
9503 +       int err;
9504 +       struct super_block *sb;
9505 +       struct au_fidir *fidir;
9506 +
9507 +       err = -ENOMEM;
9508 +       sb = file->f_dentry->d_sb;
9509 +       si_read_lock(sb, AuLock_FLUSH);
9510 +       fidir = au_fidir_alloc(sb);
9511 +       if (fidir) {
9512 +               err = au_do_open(file, do_open_dir, fidir);
9513 +               if (unlikely(err))
9514 +                       kfree(fidir);
9515 +       }
9516 +       si_read_unlock(sb);
9517 +       return err;
9518 +}
9519 +
9520 +static int aufs_release_dir(struct inode *inode __maybe_unused,
9521 +                           struct file *file)
9522 +{
9523 +       struct au_vdir *vdir_cache;
9524 +       struct au_finfo *finfo;
9525 +       struct au_fidir *fidir;
9526 +       aufs_bindex_t bindex, bend;
9527 +
9528 +       finfo = au_fi(file);
9529 +       fidir = finfo->fi_hdir;
9530 +       if (fidir) {
9531 +               au_sphl_del(&finfo->fi_hlist,
9532 +                           &au_sbi(file->f_dentry->d_sb)->si_files);
9533 +               vdir_cache = fidir->fd_vdir_cache; /* lock-free */
9534 +               if (vdir_cache)
9535 +                       au_vdir_free(vdir_cache);
9536 +
9537 +               bindex = finfo->fi_btop;
9538 +               if (bindex >= 0) {
9539 +                       /*
9540 +                        * calls fput() instead of filp_close(),
9541 +                        * since no dnotify or lock for the lower file.
9542 +                        */
9543 +                       bend = fidir->fd_bbot;
9544 +                       for (; bindex <= bend; bindex++)
9545 +                               au_set_h_fptr(file, bindex, NULL);
9546 +               }
9547 +               kfree(fidir);
9548 +               finfo->fi_hdir = NULL;
9549 +       }
9550 +       au_finfo_fin(file);
9551 +       return 0;
9552 +}
9553 +
9554 +/* ---------------------------------------------------------------------- */
9555 +
9556 +static int au_do_flush_dir(struct file *file, fl_owner_t id)
9557 +{
9558 +       int err;
9559 +       aufs_bindex_t bindex, bend;
9560 +       struct file *h_file;
9561 +
9562 +       err = 0;
9563 +       bend = au_fbend_dir(file);
9564 +       for (bindex = au_fbstart(file); !err && bindex <= bend; bindex++) {
9565 +               h_file = au_hf_dir(file, bindex);
9566 +               if (h_file)
9567 +                       err = vfsub_flush(h_file, id);
9568 +       }
9569 +       return err;
9570 +}
9571 +
9572 +static int aufs_flush_dir(struct file *file, fl_owner_t id)
9573 +{
9574 +       return au_do_flush(file, id, au_do_flush_dir);
9575 +}
9576 +
9577 +/* ---------------------------------------------------------------------- */
9578 +
9579 +static int au_do_fsync_dir_no_file(struct dentry *dentry, int datasync)
9580 +{
9581 +       int err;
9582 +       aufs_bindex_t bend, bindex;
9583 +       struct inode *inode;
9584 +       struct super_block *sb;
9585 +
9586 +       err = 0;
9587 +       sb = dentry->d_sb;
9588 +       inode = dentry->d_inode;
9589 +       IMustLock(inode);
9590 +       bend = au_dbend(dentry);
9591 +       for (bindex = au_dbstart(dentry); !err && bindex <= bend; bindex++) {
9592 +               struct path h_path;
9593 +
9594 +               if (au_test_ro(sb, bindex, inode))
9595 +                       continue;
9596 +               h_path.dentry = au_h_dptr(dentry, bindex);
9597 +               if (!h_path.dentry)
9598 +                       continue;
9599 +
9600 +               h_path.mnt = au_sbr_mnt(sb, bindex);
9601 +               err = vfsub_fsync(NULL, &h_path, datasync);
9602 +       }
9603 +
9604 +       return err;
9605 +}
9606 +
9607 +static int au_do_fsync_dir(struct file *file, int datasync)
9608 +{
9609 +       int err;
9610 +       aufs_bindex_t bend, bindex;
9611 +       struct file *h_file;
9612 +       struct super_block *sb;
9613 +       struct inode *inode;
9614 +
9615 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
9616 +       if (unlikely(err))
9617 +               goto out;
9618 +
9619 +       sb = file->f_dentry->d_sb;
9620 +       inode = file_inode(file);
9621 +       bend = au_fbend_dir(file);
9622 +       for (bindex = au_fbstart(file); !err && bindex <= bend; bindex++) {
9623 +               h_file = au_hf_dir(file, bindex);
9624 +               if (!h_file || au_test_ro(sb, bindex, inode))
9625 +                       continue;
9626 +
9627 +               err = vfsub_fsync(h_file, &h_file->f_path, datasync);
9628 +       }
9629 +
9630 +out:
9631 +       return err;
9632 +}
9633 +
9634 +/*
9635 + * @file may be NULL
9636 + */
9637 +static int aufs_fsync_dir(struct file *file, loff_t start, loff_t end,
9638 +                         int datasync)
9639 +{
9640 +       int err;
9641 +       struct dentry *dentry;
9642 +       struct super_block *sb;
9643 +       struct mutex *mtx;
9644 +
9645 +       err = 0;
9646 +       dentry = file->f_dentry;
9647 +       mtx = &dentry->d_inode->i_mutex;
9648 +       mutex_lock(mtx);
9649 +       sb = dentry->d_sb;
9650 +       si_noflush_read_lock(sb);
9651 +       if (file)
9652 +               err = au_do_fsync_dir(file, datasync);
9653 +       else {
9654 +               di_write_lock_child(dentry);
9655 +               err = au_do_fsync_dir_no_file(dentry, datasync);
9656 +       }
9657 +       au_cpup_attr_timesizes(dentry->d_inode);
9658 +       di_write_unlock(dentry);
9659 +       if (file)
9660 +               fi_write_unlock(file);
9661 +
9662 +       si_read_unlock(sb);
9663 +       mutex_unlock(mtx);
9664 +       return err;
9665 +}
9666 +
9667 +/* ---------------------------------------------------------------------- */
9668 +
9669 +static int aufs_iterate(struct file *file, struct dir_context *ctx)
9670 +{
9671 +       int err;
9672 +       struct dentry *dentry;
9673 +       struct inode *inode, *h_inode;
9674 +       struct super_block *sb;
9675 +
9676 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
9677 +
9678 +       dentry = file->f_dentry;
9679 +       inode = dentry->d_inode;
9680 +       IMustLock(inode);
9681 +
9682 +       sb = dentry->d_sb;
9683 +       si_read_lock(sb, AuLock_FLUSH);
9684 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
9685 +       if (unlikely(err))
9686 +               goto out;
9687 +       err = au_alive_dir(dentry);
9688 +       if (!err)
9689 +               err = au_vdir_init(file);
9690 +       di_downgrade_lock(dentry, AuLock_IR);
9691 +       if (unlikely(err))
9692 +               goto out_unlock;
9693 +
9694 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
9695 +       if (!au_test_nfsd()) {
9696 +               err = au_vdir_fill_de(file, ctx);
9697 +               fsstack_copy_attr_atime(inode, h_inode);
9698 +       } else {
9699 +               /*
9700 +                * nfsd filldir may call lookup_one_len(), vfs_getattr(),
9701 +                * encode_fh() and others.
9702 +                */
9703 +               atomic_inc(&h_inode->i_count);
9704 +               di_read_unlock(dentry, AuLock_IR);
9705 +               si_read_unlock(sb);
9706 +               err = au_vdir_fill_de(file, ctx);
9707 +               fsstack_copy_attr_atime(inode, h_inode);
9708 +               fi_write_unlock(file);
9709 +               iput(h_inode);
9710 +
9711 +               AuTraceErr(err);
9712 +               return err;
9713 +       }
9714 +
9715 +out_unlock:
9716 +       di_read_unlock(dentry, AuLock_IR);
9717 +       fi_write_unlock(file);
9718 +out:
9719 +       si_read_unlock(sb);
9720 +       return err;
9721 +}
9722 +
9723 +/* ---------------------------------------------------------------------- */
9724 +
9725 +#define AuTestEmpty_WHONLY     1
9726 +#define AuTestEmpty_CALLED     (1 << 1)
9727 +#define AuTestEmpty_SHWH       (1 << 2)
9728 +#define au_ftest_testempty(flags, name)        ((flags) & AuTestEmpty_##name)
9729 +#define au_fset_testempty(flags, name) \
9730 +       do { (flags) |= AuTestEmpty_##name; } while (0)
9731 +#define au_fclr_testempty(flags, name) \
9732 +       do { (flags) &= ~AuTestEmpty_##name; } while (0)
9733 +
9734 +#ifndef CONFIG_AUFS_SHWH
9735 +#undef AuTestEmpty_SHWH
9736 +#define AuTestEmpty_SHWH       0
9737 +#endif
9738 +
9739 +struct test_empty_arg {
9740 +       struct dir_context ctx;
9741 +       struct au_nhash *whlist;
9742 +       unsigned int flags;
9743 +       int err;
9744 +       aufs_bindex_t bindex;
9745 +};
9746 +
9747 +static int test_empty_cb(struct dir_context *ctx, const char *__name,
9748 +                        int namelen, loff_t offset __maybe_unused, u64 ino,
9749 +                        unsigned int d_type)
9750 +{
9751 +       struct test_empty_arg *arg = container_of(ctx, struct test_empty_arg,
9752 +                                                 ctx);
9753 +       char *name = (void *)__name;
9754 +
9755 +       arg->err = 0;
9756 +       au_fset_testempty(arg->flags, CALLED);
9757 +       /* smp_mb(); */
9758 +       if (name[0] == '.'
9759 +           && (namelen == 1 || (name[1] == '.' && namelen == 2)))
9760 +               goto out; /* success */
9761 +
9762 +       if (namelen <= AUFS_WH_PFX_LEN
9763 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
9764 +               if (au_ftest_testempty(arg->flags, WHONLY)
9765 +                   && !au_nhash_test_known_wh(arg->whlist, name, namelen))
9766 +                       arg->err = -ENOTEMPTY;
9767 +               goto out;
9768 +       }
9769 +
9770 +       name += AUFS_WH_PFX_LEN;
9771 +       namelen -= AUFS_WH_PFX_LEN;
9772 +       if (!au_nhash_test_known_wh(arg->whlist, name, namelen))
9773 +               arg->err = au_nhash_append_wh
9774 +                       (arg->whlist, name, namelen, ino, d_type, arg->bindex,
9775 +                        au_ftest_testempty(arg->flags, SHWH));
9776 +
9777 +out:
9778 +       /* smp_mb(); */
9779 +       AuTraceErr(arg->err);
9780 +       return arg->err;
9781 +}
9782 +
9783 +static int do_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
9784 +{
9785 +       int err;
9786 +       struct file *h_file;
9787 +
9788 +       h_file = au_h_open(dentry, arg->bindex,
9789 +                          O_RDONLY | O_NONBLOCK | O_DIRECTORY | O_LARGEFILE,
9790 +                          /*file*/NULL, /*force_wr*/0);
9791 +       err = PTR_ERR(h_file);
9792 +       if (IS_ERR(h_file))
9793 +               goto out;
9794 +
9795 +       err = 0;
9796 +       if (!au_opt_test(au_mntflags(dentry->d_sb), UDBA_NONE)
9797 +           && !file_inode(h_file)->i_nlink)
9798 +               goto out_put;
9799 +
9800 +       do {
9801 +               arg->err = 0;
9802 +               au_fclr_testempty(arg->flags, CALLED);
9803 +               /* smp_mb(); */
9804 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
9805 +               if (err >= 0)
9806 +                       err = arg->err;
9807 +       } while (!err && au_ftest_testempty(arg->flags, CALLED));
9808 +
9809 +out_put:
9810 +       fput(h_file);
9811 +       au_sbr_put(dentry->d_sb, arg->bindex);
9812 +out:
9813 +       return err;
9814 +}
9815 +
9816 +struct do_test_empty_args {
9817 +       int *errp;
9818 +       struct dentry *dentry;
9819 +       struct test_empty_arg *arg;
9820 +};
9821 +
9822 +static void call_do_test_empty(void *args)
9823 +{
9824 +       struct do_test_empty_args *a = args;
9825 +       *a->errp = do_test_empty(a->dentry, a->arg);
9826 +}
9827 +
9828 +static int sio_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
9829 +{
9830 +       int err, wkq_err;
9831 +       struct dentry *h_dentry;
9832 +       struct inode *h_inode;
9833 +
9834 +       h_dentry = au_h_dptr(dentry, arg->bindex);
9835 +       h_inode = h_dentry->d_inode;
9836 +       /* todo: i_mode changes anytime? */
9837 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
9838 +       err = au_test_h_perm_sio(h_inode, MAY_EXEC | MAY_READ);
9839 +       mutex_unlock(&h_inode->i_mutex);
9840 +       if (!err)
9841 +               err = do_test_empty(dentry, arg);
9842 +       else {
9843 +               struct do_test_empty_args args = {
9844 +                       .errp   = &err,
9845 +                       .dentry = dentry,
9846 +                       .arg    = arg
9847 +               };
9848 +               unsigned int flags = arg->flags;
9849 +
9850 +               wkq_err = au_wkq_wait(call_do_test_empty, &args);
9851 +               if (unlikely(wkq_err))
9852 +                       err = wkq_err;
9853 +               arg->flags = flags;
9854 +       }
9855 +
9856 +       return err;
9857 +}
9858 +
9859 +int au_test_empty_lower(struct dentry *dentry)
9860 +{
9861 +       int err;
9862 +       unsigned int rdhash;
9863 +       aufs_bindex_t bindex, bstart, btail;
9864 +       struct au_nhash whlist;
9865 +       struct test_empty_arg arg = {
9866 +               .ctx = {
9867 +                       .actor = au_diractor(test_empty_cb)
9868 +               }
9869 +       };
9870 +       int (*test_empty)(struct dentry *dentry, struct test_empty_arg *arg);
9871 +
9872 +       SiMustAnyLock(dentry->d_sb);
9873 +
9874 +       rdhash = au_sbi(dentry->d_sb)->si_rdhash;
9875 +       if (!rdhash)
9876 +               rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, dentry));
9877 +       err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
9878 +       if (unlikely(err))
9879 +               goto out;
9880 +
9881 +       arg.flags = 0;
9882 +       arg.whlist = &whlist;
9883 +       bstart = au_dbstart(dentry);
9884 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
9885 +               au_fset_testempty(arg.flags, SHWH);
9886 +       test_empty = do_test_empty;
9887 +       if (au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1))
9888 +               test_empty = sio_test_empty;
9889 +       arg.bindex = bstart;
9890 +       err = test_empty(dentry, &arg);
9891 +       if (unlikely(err))
9892 +               goto out_whlist;
9893 +
9894 +       au_fset_testempty(arg.flags, WHONLY);
9895 +       btail = au_dbtaildir(dentry);
9896 +       for (bindex = bstart + 1; !err && bindex <= btail; bindex++) {
9897 +               struct dentry *h_dentry;
9898 +
9899 +               h_dentry = au_h_dptr(dentry, bindex);
9900 +               if (h_dentry && h_dentry->d_inode) {
9901 +                       arg.bindex = bindex;
9902 +                       err = test_empty(dentry, &arg);
9903 +               }
9904 +       }
9905 +
9906 +out_whlist:
9907 +       au_nhash_wh_free(&whlist);
9908 +out:
9909 +       return err;
9910 +}
9911 +
9912 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist)
9913 +{
9914 +       int err;
9915 +       struct test_empty_arg arg = {
9916 +               .ctx = {
9917 +                       .actor = au_diractor(test_empty_cb)
9918 +               }
9919 +       };
9920 +       aufs_bindex_t bindex, btail;
9921 +
9922 +       err = 0;
9923 +       arg.whlist = whlist;
9924 +       arg.flags = AuTestEmpty_WHONLY;
9925 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
9926 +               au_fset_testempty(arg.flags, SHWH);
9927 +       btail = au_dbtaildir(dentry);
9928 +       for (bindex = au_dbstart(dentry); !err && bindex <= btail; bindex++) {
9929 +               struct dentry *h_dentry;
9930 +
9931 +               h_dentry = au_h_dptr(dentry, bindex);
9932 +               if (h_dentry && h_dentry->d_inode) {
9933 +                       arg.bindex = bindex;
9934 +                       err = sio_test_empty(dentry, &arg);
9935 +               }
9936 +       }
9937 +
9938 +       return err;
9939 +}
9940 +
9941 +/* ---------------------------------------------------------------------- */
9942 +
9943 +const struct file_operations aufs_dir_fop = {
9944 +       .owner          = THIS_MODULE,
9945 +       .llseek         = default_llseek,
9946 +       .read           = generic_read_dir,
9947 +       .iterate        = aufs_iterate,
9948 +       .unlocked_ioctl = aufs_ioctl_dir,
9949 +#ifdef CONFIG_COMPAT
9950 +       .compat_ioctl   = aufs_compat_ioctl_dir,
9951 +#endif
9952 +       .open           = aufs_open_dir,
9953 +       .release        = aufs_release_dir,
9954 +       .flush          = aufs_flush_dir,
9955 +       .fsync          = aufs_fsync_dir
9956 +};
9957 diff -urN /usr/share/empty/fs/aufs/dir.h linux/fs/aufs/dir.h
9958 --- /usr/share/empty/fs/aufs/dir.h      1970-01-01 01:00:00.000000000 +0100
9959 +++ linux/fs/aufs/dir.h 2015-01-25 13:00:38.631047076 +0100
9960 @@ -0,0 +1,130 @@
9961 +/*
9962 + * Copyright (C) 2005-2014 Junjiro R. Okajima
9963 + *
9964 + * This program, aufs is free software; you can redistribute it and/or modify
9965 + * it under the terms of the GNU General Public License as published by
9966 + * the Free Software Foundation; either version 2 of the License, or
9967 + * (at your option) any later version.
9968 + *
9969 + * This program is distributed in the hope that it will be useful,
9970 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9971 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9972 + * GNU General Public License for more details.
9973 + *
9974 + * You should have received a copy of the GNU General Public License
9975 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9976 + */
9977 +
9978 +/*
9979 + * directory operations
9980 + */
9981 +
9982 +#ifndef __AUFS_DIR_H__
9983 +#define __AUFS_DIR_H__
9984 +
9985 +#ifdef __KERNEL__
9986 +
9987 +#include <linux/fs.h>
9988 +
9989 +/* ---------------------------------------------------------------------- */
9990 +
9991 +/* need to be faster and smaller */
9992 +
9993 +struct au_nhash {
9994 +       unsigned int            nh_num;
9995 +       struct hlist_head       *nh_head;
9996 +};
9997 +
9998 +struct au_vdir_destr {
9999 +       unsigned char   len;
10000 +       unsigned char   name[0];
10001 +} __packed;
10002 +
10003 +struct au_vdir_dehstr {
10004 +       struct hlist_node       hash;
10005 +       struct au_vdir_destr    *str;
10006 +} ____cacheline_aligned_in_smp;
10007 +
10008 +struct au_vdir_de {
10009 +       ino_t                   de_ino;
10010 +       unsigned char           de_type;
10011 +       /* caution: packed */
10012 +       struct au_vdir_destr    de_str;
10013 +} __packed;
10014 +
10015 +struct au_vdir_wh {
10016 +       struct hlist_node       wh_hash;
10017 +#ifdef CONFIG_AUFS_SHWH
10018 +       ino_t                   wh_ino;
10019 +       aufs_bindex_t           wh_bindex;
10020 +       unsigned char           wh_type;
10021 +#else
10022 +       aufs_bindex_t           wh_bindex;
10023 +#endif
10024 +       /* caution: packed */
10025 +       struct au_vdir_destr    wh_str;
10026 +} __packed;
10027 +
10028 +union au_vdir_deblk_p {
10029 +       unsigned char           *deblk;
10030 +       struct au_vdir_de       *de;
10031 +};
10032 +
10033 +struct au_vdir {
10034 +       unsigned char   **vd_deblk;
10035 +       unsigned long   vd_nblk;
10036 +       struct {
10037 +               unsigned long           ul;
10038 +               union au_vdir_deblk_p   p;
10039 +       } vd_last;
10040 +
10041 +       unsigned long   vd_version;
10042 +       unsigned int    vd_deblk_sz;
10043 +       unsigned long   vd_jiffy;
10044 +} ____cacheline_aligned_in_smp;
10045 +
10046 +/* ---------------------------------------------------------------------- */
10047 +
10048 +/* dir.c */
10049 +extern const struct file_operations aufs_dir_fop;
10050 +void au_add_nlink(struct inode *dir, struct inode *h_dir);
10051 +void au_sub_nlink(struct inode *dir, struct inode *h_dir);
10052 +loff_t au_dir_size(struct file *file, struct dentry *dentry);
10053 +int au_test_empty_lower(struct dentry *dentry);
10054 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist);
10055 +
10056 +/* vdir.c */
10057 +unsigned int au_rdhash_est(loff_t sz);
10058 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp);
10059 +void au_nhash_wh_free(struct au_nhash *whlist);
10060 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
10061 +                           int limit);
10062 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen);
10063 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
10064 +                      unsigned int d_type, aufs_bindex_t bindex,
10065 +                      unsigned char shwh);
10066 +void au_vdir_free(struct au_vdir *vdir);
10067 +int au_vdir_init(struct file *file);
10068 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx);
10069 +
10070 +/* ioctl.c */
10071 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg);
10072 +
10073 +#ifdef CONFIG_AUFS_RDU
10074 +/* rdu.c */
10075 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
10076 +#ifdef CONFIG_COMPAT
10077 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd,
10078 +                        unsigned long arg);
10079 +#endif
10080 +#else
10081 +AuStub(long, au_rdu_ioctl, return -EINVAL, struct file *file,
10082 +       unsigned int cmd, unsigned long arg)
10083 +#ifdef CONFIG_COMPAT
10084 +AuStub(long, au_rdu_compat_ioctl, return -EINVAL, struct file *file,
10085 +       unsigned int cmd, unsigned long arg)
10086 +#endif
10087 +#endif
10088 +
10089 +#endif /* __KERNEL__ */
10090 +#endif /* __AUFS_DIR_H__ */
10091 diff -urN /usr/share/empty/fs/aufs/dynop.c linux/fs/aufs/dynop.c
10092 --- /usr/share/empty/fs/aufs/dynop.c    1970-01-01 01:00:00.000000000 +0100
10093 +++ linux/fs/aufs/dynop.c       2015-01-25 13:00:38.631047076 +0100
10094 @@ -0,0 +1,379 @@
10095 +/*
10096 + * Copyright (C) 2010-2014 Junjiro R. Okajima
10097 + *
10098 + * This program, aufs is free software; you can redistribute it and/or modify
10099 + * it under the terms of the GNU General Public License as published by
10100 + * the Free Software Foundation; either version 2 of the License, or
10101 + * (at your option) any later version.
10102 + *
10103 + * This program is distributed in the hope that it will be useful,
10104 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10105 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10106 + * GNU General Public License for more details.
10107 + *
10108 + * You should have received a copy of the GNU General Public License
10109 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10110 + */
10111 +
10112 +/*
10113 + * dynamically customizable operations for regular files
10114 + */
10115 +
10116 +#include "aufs.h"
10117 +
10118 +#define DyPrSym(key)   AuDbgSym(key->dk_op.dy_hop)
10119 +
10120 +/*
10121 + * How large will these lists be?
10122 + * Usually just a few elements, 20-30 at most for each, I guess.
10123 + */
10124 +static struct au_splhead dynop[AuDyLast];
10125 +
10126 +static struct au_dykey *dy_gfind_get(struct au_splhead *spl, const void *h_op)
10127 +{
10128 +       struct au_dykey *key, *tmp;
10129 +       struct list_head *head;
10130 +
10131 +       key = NULL;
10132 +       head = &spl->head;
10133 +       rcu_read_lock();
10134 +       list_for_each_entry_rcu(tmp, head, dk_list)
10135 +               if (tmp->dk_op.dy_hop == h_op) {
10136 +                       key = tmp;
10137 +                       kref_get(&key->dk_kref);
10138 +                       break;
10139 +               }
10140 +       rcu_read_unlock();
10141 +
10142 +       return key;
10143 +}
10144 +
10145 +static struct au_dykey *dy_bradd(struct au_branch *br, struct au_dykey *key)
10146 +{
10147 +       struct au_dykey **k, *found;
10148 +       const void *h_op = key->dk_op.dy_hop;
10149 +       int i;
10150 +
10151 +       found = NULL;
10152 +       k = br->br_dykey;
10153 +       for (i = 0; i < AuBrDynOp; i++)
10154 +               if (k[i]) {
10155 +                       if (k[i]->dk_op.dy_hop == h_op) {
10156 +                               found = k[i];
10157 +                               break;
10158 +                       }
10159 +               } else
10160 +                       break;
10161 +       if (!found) {
10162 +               spin_lock(&br->br_dykey_lock);
10163 +               for (; i < AuBrDynOp; i++)
10164 +                       if (k[i]) {
10165 +                               if (k[i]->dk_op.dy_hop == h_op) {
10166 +                                       found = k[i];
10167 +                                       break;
10168 +                               }
10169 +                       } else {
10170 +                               k[i] = key;
10171 +                               break;
10172 +                       }
10173 +               spin_unlock(&br->br_dykey_lock);
10174 +               BUG_ON(i == AuBrDynOp); /* expand the array */
10175 +       }
10176 +
10177 +       return found;
10178 +}
10179 +
10180 +/* kref_get() if @key is already added */
10181 +static struct au_dykey *dy_gadd(struct au_splhead *spl, struct au_dykey *key)
10182 +{
10183 +       struct au_dykey *tmp, *found;
10184 +       struct list_head *head;
10185 +       const void *h_op = key->dk_op.dy_hop;
10186 +
10187 +       found = NULL;
10188 +       head = &spl->head;
10189 +       spin_lock(&spl->spin);
10190 +       list_for_each_entry(tmp, head, dk_list)
10191 +               if (tmp->dk_op.dy_hop == h_op) {
10192 +                       kref_get(&tmp->dk_kref);
10193 +                       found = tmp;
10194 +                       break;
10195 +               }
10196 +       if (!found)
10197 +               list_add_rcu(&key->dk_list, head);
10198 +       spin_unlock(&spl->spin);
10199 +
10200 +       if (!found)
10201 +               DyPrSym(key);
10202 +       return found;
10203 +}
10204 +
10205 +static void dy_free_rcu(struct rcu_head *rcu)
10206 +{
10207 +       struct au_dykey *key;
10208 +
10209 +       key = container_of(rcu, struct au_dykey, dk_rcu);
10210 +       DyPrSym(key);
10211 +       kfree(key);
10212 +}
10213 +
10214 +static void dy_free(struct kref *kref)
10215 +{
10216 +       struct au_dykey *key;
10217 +       struct au_splhead *spl;
10218 +
10219 +       key = container_of(kref, struct au_dykey, dk_kref);
10220 +       spl = dynop + key->dk_op.dy_type;
10221 +       au_spl_del_rcu(&key->dk_list, spl);
10222 +       call_rcu(&key->dk_rcu, dy_free_rcu);
10223 +}
10224 +
10225 +void au_dy_put(struct au_dykey *key)
10226 +{
10227 +       kref_put(&key->dk_kref, dy_free);
10228 +}
10229 +
10230 +/* ---------------------------------------------------------------------- */
10231 +
10232 +#define DyDbgSize(cnt, op)     AuDebugOn(cnt != sizeof(op)/sizeof(void *))
10233 +
10234 +#ifdef CONFIG_AUFS_DEBUG
10235 +#define DyDbgDeclare(cnt)      unsigned int cnt = 0
10236 +#define DyDbgInc(cnt)          do { cnt++; } while (0)
10237 +#else
10238 +#define DyDbgDeclare(cnt)      do {} while (0)
10239 +#define DyDbgInc(cnt)          do {} while (0)
10240 +#endif
10241 +
10242 +#define DySet(func, dst, src, h_op, h_sb) do {                         \
10243 +       DyDbgInc(cnt);                                                  \
10244 +       if (h_op->func) {                                               \
10245 +               if (src.func)                                           \
10246 +                       dst.func = src.func;                            \
10247 +               else                                                    \
10248 +                       AuDbg("%s %s\n", au_sbtype(h_sb), #func);       \
10249 +       }                                                               \
10250 +} while (0)
10251 +
10252 +#define DySetForce(func, dst, src) do {                \
10253 +       AuDebugOn(!src.func);                   \
10254 +       DyDbgInc(cnt);                          \
10255 +       dst.func = src.func;                    \
10256 +} while (0)
10257 +
10258 +#define DySetAop(func) \
10259 +       DySet(func, dyaop->da_op, aufs_aop, h_aop, h_sb)
10260 +#define DySetAopForce(func) \
10261 +       DySetForce(func, dyaop->da_op, aufs_aop)
10262 +
10263 +static void dy_aop(struct au_dykey *key, const void *h_op,
10264 +                  struct super_block *h_sb __maybe_unused)
10265 +{
10266 +       struct au_dyaop *dyaop = (void *)key;
10267 +       const struct address_space_operations *h_aop = h_op;
10268 +       DyDbgDeclare(cnt);
10269 +
10270 +       AuDbg("%s\n", au_sbtype(h_sb));
10271 +
10272 +       DySetAop(writepage);
10273 +       DySetAopForce(readpage);        /* force */
10274 +       DySetAop(writepages);
10275 +       DySetAop(set_page_dirty);
10276 +       DySetAop(readpages);
10277 +       DySetAop(write_begin);
10278 +       DySetAop(write_end);
10279 +       DySetAop(bmap);
10280 +       DySetAop(invalidatepage);
10281 +       DySetAop(releasepage);
10282 +       DySetAop(freepage);
10283 +       /* these two will be changed according to an aufs mount option */
10284 +       DySetAop(direct_IO);
10285 +       DySetAop(get_xip_mem);
10286 +       DySetAop(migratepage);
10287 +       DySetAop(launder_page);
10288 +       DySetAop(is_partially_uptodate);
10289 +       DySetAop(is_dirty_writeback);
10290 +       DySetAop(error_remove_page);
10291 +       DySetAop(swap_activate);
10292 +       DySetAop(swap_deactivate);
10293 +
10294 +       DyDbgSize(cnt, *h_aop);
10295 +       dyaop->da_get_xip_mem = h_aop->get_xip_mem;
10296 +}
10297 +
10298 +/* ---------------------------------------------------------------------- */
10299 +
10300 +static void dy_bug(struct kref *kref)
10301 +{
10302 +       BUG();
10303 +}
10304 +
10305 +static struct au_dykey *dy_get(struct au_dynop *op, struct au_branch *br)
10306 +{
10307 +       struct au_dykey *key, *old;
10308 +       struct au_splhead *spl;
10309 +       struct op {
10310 +               unsigned int sz;
10311 +               void (*set)(struct au_dykey *key, const void *h_op,
10312 +                           struct super_block *h_sb __maybe_unused);
10313 +       };
10314 +       static const struct op a[] = {
10315 +               [AuDy_AOP] = {
10316 +                       .sz     = sizeof(struct au_dyaop),
10317 +                       .set    = dy_aop
10318 +               }
10319 +       };
10320 +       const struct op *p;
10321 +
10322 +       spl = dynop + op->dy_type;
10323 +       key = dy_gfind_get(spl, op->dy_hop);
10324 +       if (key)
10325 +               goto out_add; /* success */
10326 +
10327 +       p = a + op->dy_type;
10328 +       key = kzalloc(p->sz, GFP_NOFS);
10329 +       if (unlikely(!key)) {
10330 +               key = ERR_PTR(-ENOMEM);
10331 +               goto out;
10332 +       }
10333 +
10334 +       key->dk_op.dy_hop = op->dy_hop;
10335 +       kref_init(&key->dk_kref);
10336 +       p->set(key, op->dy_hop, au_br_sb(br));
10337 +       old = dy_gadd(spl, key);
10338 +       if (old) {
10339 +               kfree(key);
10340 +               key = old;
10341 +       }
10342 +
10343 +out_add:
10344 +       old = dy_bradd(br, key);
10345 +       if (old)
10346 +               /* its ref-count should never be zero here */
10347 +               kref_put(&key->dk_kref, dy_bug);
10348 +out:
10349 +       return key;
10350 +}
10351 +
10352 +/* ---------------------------------------------------------------------- */
10353 +/*
10354 + * Aufs prohibits O_DIRECT by defaut even if the branch supports it.
10355 + * This behaviour is necessary to return an error from open(O_DIRECT) instead
10356 + * of the succeeding I/O. The dio mount option enables O_DIRECT and makes
10357 + * open(O_DIRECT) always succeed, but the succeeding I/O may return an error.
10358 + * See the aufs manual in detail.
10359 + *
10360 + * To keep this behaviour, aufs has to set NULL to ->get_xip_mem too, and the
10361 + * performance of fadvise() and madvise() may be affected.
10362 + */
10363 +static void dy_adx(struct au_dyaop *dyaop, int do_dx)
10364 +{
10365 +       if (!do_dx) {
10366 +               dyaop->da_op.direct_IO = NULL;
10367 +               dyaop->da_op.get_xip_mem = NULL;
10368 +       } else {
10369 +               dyaop->da_op.direct_IO = aufs_aop.direct_IO;
10370 +               dyaop->da_op.get_xip_mem = aufs_aop.get_xip_mem;
10371 +               if (!dyaop->da_get_xip_mem)
10372 +                       dyaop->da_op.get_xip_mem = NULL;
10373 +       }
10374 +}
10375 +
10376 +static struct au_dyaop *dy_aget(struct au_branch *br,
10377 +                               const struct address_space_operations *h_aop,
10378 +                               int do_dx)
10379 +{
10380 +       struct au_dyaop *dyaop;
10381 +       struct au_dynop op;
10382 +
10383 +       op.dy_type = AuDy_AOP;
10384 +       op.dy_haop = h_aop;
10385 +       dyaop = (void *)dy_get(&op, br);
10386 +       if (IS_ERR(dyaop))
10387 +               goto out;
10388 +       dy_adx(dyaop, do_dx);
10389 +
10390 +out:
10391 +       return dyaop;
10392 +}
10393 +
10394 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
10395 +               struct inode *h_inode)
10396 +{
10397 +       int err, do_dx;
10398 +       struct super_block *sb;
10399 +       struct au_branch *br;
10400 +       struct au_dyaop *dyaop;
10401 +
10402 +       AuDebugOn(!S_ISREG(h_inode->i_mode));
10403 +       IiMustWriteLock(inode);
10404 +
10405 +       sb = inode->i_sb;
10406 +       br = au_sbr(sb, bindex);
10407 +       do_dx = !!au_opt_test(au_mntflags(sb), DIO);
10408 +       dyaop = dy_aget(br, h_inode->i_mapping->a_ops, do_dx);
10409 +       err = PTR_ERR(dyaop);
10410 +       if (IS_ERR(dyaop))
10411 +               /* unnecessary to call dy_fput() */
10412 +               goto out;
10413 +
10414 +       err = 0;
10415 +       inode->i_mapping->a_ops = &dyaop->da_op;
10416 +
10417 +out:
10418 +       return err;
10419 +}
10420 +
10421 +/*
10422 + * Is it safe to replace a_ops during the inode/file is in operation?
10423 + * Yes, I hope so.
10424 + */
10425 +int au_dy_irefresh(struct inode *inode)
10426 +{
10427 +       int err;
10428 +       aufs_bindex_t bstart;
10429 +       struct inode *h_inode;
10430 +
10431 +       err = 0;
10432 +       if (S_ISREG(inode->i_mode)) {
10433 +               bstart = au_ibstart(inode);
10434 +               h_inode = au_h_iptr(inode, bstart);
10435 +               err = au_dy_iaop(inode, bstart, h_inode);
10436 +       }
10437 +       return err;
10438 +}
10439 +
10440 +void au_dy_arefresh(int do_dx)
10441 +{
10442 +       struct au_splhead *spl;
10443 +       struct list_head *head;
10444 +       struct au_dykey *key;
10445 +
10446 +       spl = dynop + AuDy_AOP;
10447 +       head = &spl->head;
10448 +       spin_lock(&spl->spin);
10449 +       list_for_each_entry(key, head, dk_list)
10450 +               dy_adx((void *)key, do_dx);
10451 +       spin_unlock(&spl->spin);
10452 +}
10453 +
10454 +/* ---------------------------------------------------------------------- */
10455 +
10456 +void __init au_dy_init(void)
10457 +{
10458 +       int i;
10459 +
10460 +       /* make sure that 'struct au_dykey *' can be any type */
10461 +       BUILD_BUG_ON(offsetof(struct au_dyaop, da_key));
10462 +
10463 +       for (i = 0; i < AuDyLast; i++)
10464 +               au_spl_init(dynop + i);
10465 +}
10466 +
10467 +void au_dy_fin(void)
10468 +{
10469 +       int i;
10470 +
10471 +       for (i = 0; i < AuDyLast; i++)
10472 +               WARN_ON(!list_empty(&dynop[i].head));
10473 +}
10474 diff -urN /usr/share/empty/fs/aufs/dynop.h linux/fs/aufs/dynop.h
10475 --- /usr/share/empty/fs/aufs/dynop.h    1970-01-01 01:00:00.000000000 +0100
10476 +++ linux/fs/aufs/dynop.h       2015-01-25 13:00:38.631047076 +0100
10477 @@ -0,0 +1,75 @@
10478 +/*
10479 + * Copyright (C) 2010-2014 Junjiro R. Okajima
10480 + *
10481 + * This program, aufs is free software; you can redistribute it and/or modify
10482 + * it under the terms of the GNU General Public License as published by
10483 + * the Free Software Foundation; either version 2 of the License, or
10484 + * (at your option) any later version.
10485 + *
10486 + * This program is distributed in the hope that it will be useful,
10487 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10488 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10489 + * GNU General Public License for more details.
10490 + *
10491 + * You should have received a copy of the GNU General Public License
10492 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10493 + */
10494 +
10495 +/*
10496 + * dynamically customizable operations (for regular files only)
10497 + */
10498 +
10499 +#ifndef __AUFS_DYNOP_H__
10500 +#define __AUFS_DYNOP_H__
10501 +
10502 +#ifdef __KERNEL__
10503 +
10504 +#include "inode.h"
10505 +
10506 +enum {AuDy_AOP, AuDyLast};
10507 +
10508 +struct au_dynop {
10509 +       int                                             dy_type;
10510 +       union {
10511 +               const void                              *dy_hop;
10512 +               const struct address_space_operations   *dy_haop;
10513 +       };
10514 +};
10515 +
10516 +struct au_dykey {
10517 +       union {
10518 +               struct list_head        dk_list;
10519 +               struct rcu_head         dk_rcu;
10520 +       };
10521 +       struct au_dynop         dk_op;
10522 +
10523 +       /*
10524 +        * during I am in the branch local array, kref is gotten. when the
10525 +        * branch is removed, kref is put.
10526 +        */
10527 +       struct kref             dk_kref;
10528 +};
10529 +
10530 +/* stop unioning since their sizes are very different from each other */
10531 +struct au_dyaop {
10532 +       struct au_dykey                 da_key;
10533 +       struct address_space_operations da_op; /* not const */
10534 +       int (*da_get_xip_mem)(struct address_space *, pgoff_t, int,
10535 +                             void **, unsigned long *);
10536 +};
10537 +
10538 +/* ---------------------------------------------------------------------- */
10539 +
10540 +/* dynop.c */
10541 +struct au_branch;
10542 +void au_dy_put(struct au_dykey *key);
10543 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
10544 +               struct inode *h_inode);
10545 +int au_dy_irefresh(struct inode *inode);
10546 +void au_dy_arefresh(int do_dio);
10547 +
10548 +void __init au_dy_init(void);
10549 +void au_dy_fin(void);
10550 +
10551 +#endif /* __KERNEL__ */
10552 +#endif /* __AUFS_DYNOP_H__ */
10553 diff -urN /usr/share/empty/fs/aufs/export.c linux/fs/aufs/export.c
10554 --- /usr/share/empty/fs/aufs/export.c   1970-01-01 01:00:00.000000000 +0100
10555 +++ linux/fs/aufs/export.c      2015-01-25 13:00:38.631047076 +0100
10556 @@ -0,0 +1,831 @@
10557 +/*
10558 + * Copyright (C) 2005-2014 Junjiro R. Okajima
10559 + *
10560 + * This program, aufs is free software; you can redistribute it and/or modify
10561 + * it under the terms of the GNU General Public License as published by
10562 + * the Free Software Foundation; either version 2 of the License, or
10563 + * (at your option) any later version.
10564 + *
10565 + * This program is distributed in the hope that it will be useful,
10566 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10567 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10568 + * GNU General Public License for more details.
10569 + *
10570 + * You should have received a copy of the GNU General Public License
10571 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10572 + */
10573 +
10574 +/*
10575 + * export via nfs
10576 + */
10577 +
10578 +#include <linux/exportfs.h>
10579 +#include <linux/fs_struct.h>
10580 +#include <linux/namei.h>
10581 +#include <linux/nsproxy.h>
10582 +#include <linux/random.h>
10583 +#include <linux/writeback.h>
10584 +#include "../fs/mount.h"
10585 +#include "aufs.h"
10586 +
10587 +union conv {
10588 +#ifdef CONFIG_AUFS_INO_T_64
10589 +       __u32 a[2];
10590 +#else
10591 +       __u32 a[1];
10592 +#endif
10593 +       ino_t ino;
10594 +};
10595 +
10596 +static ino_t decode_ino(__u32 *a)
10597 +{
10598 +       union conv u;
10599 +
10600 +       BUILD_BUG_ON(sizeof(u.ino) != sizeof(u.a));
10601 +       u.a[0] = a[0];
10602 +#ifdef CONFIG_AUFS_INO_T_64
10603 +       u.a[1] = a[1];
10604 +#endif
10605 +       return u.ino;
10606 +}
10607 +
10608 +static void encode_ino(__u32 *a, ino_t ino)
10609 +{
10610 +       union conv u;
10611 +
10612 +       u.ino = ino;
10613 +       a[0] = u.a[0];
10614 +#ifdef CONFIG_AUFS_INO_T_64
10615 +       a[1] = u.a[1];
10616 +#endif
10617 +}
10618 +
10619 +/* NFS file handle */
10620 +enum {
10621 +       Fh_br_id,
10622 +       Fh_sigen,
10623 +#ifdef CONFIG_AUFS_INO_T_64
10624 +       /* support 64bit inode number */
10625 +       Fh_ino1,
10626 +       Fh_ino2,
10627 +       Fh_dir_ino1,
10628 +       Fh_dir_ino2,
10629 +#else
10630 +       Fh_ino1,
10631 +       Fh_dir_ino1,
10632 +#endif
10633 +       Fh_igen,
10634 +       Fh_h_type,
10635 +       Fh_tail,
10636 +
10637 +       Fh_ino = Fh_ino1,
10638 +       Fh_dir_ino = Fh_dir_ino1
10639 +};
10640 +
10641 +static int au_test_anon(struct dentry *dentry)
10642 +{
10643 +       /* note: read d_flags without d_lock */
10644 +       return !!(dentry->d_flags & DCACHE_DISCONNECTED);
10645 +}
10646 +
10647 +int au_test_nfsd(void)
10648 +{
10649 +       int ret;
10650 +       struct task_struct *tsk = current;
10651 +       char comm[sizeof(tsk->comm)];
10652 +
10653 +       ret = 0;
10654 +       if (tsk->flags & PF_KTHREAD) {
10655 +               get_task_comm(comm, tsk);
10656 +               ret = !strcmp(comm, "nfsd");
10657 +       }
10658 +
10659 +       return ret;
10660 +}
10661 +
10662 +/* ---------------------------------------------------------------------- */
10663 +/* inode generation external table */
10664 +
10665 +void au_xigen_inc(struct inode *inode)
10666 +{
10667 +       loff_t pos;
10668 +       ssize_t sz;
10669 +       __u32 igen;
10670 +       struct super_block *sb;
10671 +       struct au_sbinfo *sbinfo;
10672 +
10673 +       sb = inode->i_sb;
10674 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
10675 +
10676 +       sbinfo = au_sbi(sb);
10677 +       pos = inode->i_ino;
10678 +       pos *= sizeof(igen);
10679 +       igen = inode->i_generation + 1;
10680 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xigen, &igen,
10681 +                        sizeof(igen), &pos);
10682 +       if (sz == sizeof(igen))
10683 +               return; /* success */
10684 +
10685 +       if (unlikely(sz >= 0))
10686 +               AuIOErr("xigen error (%zd)\n", sz);
10687 +}
10688 +
10689 +int au_xigen_new(struct inode *inode)
10690 +{
10691 +       int err;
10692 +       loff_t pos;
10693 +       ssize_t sz;
10694 +       struct super_block *sb;
10695 +       struct au_sbinfo *sbinfo;
10696 +       struct file *file;
10697 +
10698 +       err = 0;
10699 +       /* todo: dirty, at mount time */
10700 +       if (inode->i_ino == AUFS_ROOT_INO)
10701 +               goto out;
10702 +       sb = inode->i_sb;
10703 +       SiMustAnyLock(sb);
10704 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
10705 +               goto out;
10706 +
10707 +       err = -EFBIG;
10708 +       pos = inode->i_ino;
10709 +       if (unlikely(au_loff_max / sizeof(inode->i_generation) - 1 < pos)) {
10710 +               AuIOErr1("too large i%lld\n", pos);
10711 +               goto out;
10712 +       }
10713 +       pos *= sizeof(inode->i_generation);
10714 +
10715 +       err = 0;
10716 +       sbinfo = au_sbi(sb);
10717 +       file = sbinfo->si_xigen;
10718 +       BUG_ON(!file);
10719 +
10720 +       if (vfsub_f_size_read(file)
10721 +           < pos + sizeof(inode->i_generation)) {
10722 +               inode->i_generation = atomic_inc_return(&sbinfo->si_xigen_next);
10723 +               sz = xino_fwrite(sbinfo->si_xwrite, file, &inode->i_generation,
10724 +                                sizeof(inode->i_generation), &pos);
10725 +       } else
10726 +               sz = xino_fread(sbinfo->si_xread, file, &inode->i_generation,
10727 +                               sizeof(inode->i_generation), &pos);
10728 +       if (sz == sizeof(inode->i_generation))
10729 +               goto out; /* success */
10730 +
10731 +       err = sz;
10732 +       if (unlikely(sz >= 0)) {
10733 +               err = -EIO;
10734 +               AuIOErr("xigen error (%zd)\n", sz);
10735 +       }
10736 +
10737 +out:
10738 +       return err;
10739 +}
10740 +
10741 +int au_xigen_set(struct super_block *sb, struct file *base)
10742 +{
10743 +       int err;
10744 +       struct au_sbinfo *sbinfo;
10745 +       struct file *file;
10746 +
10747 +       SiMustWriteLock(sb);
10748 +
10749 +       sbinfo = au_sbi(sb);
10750 +       file = au_xino_create2(base, sbinfo->si_xigen);
10751 +       err = PTR_ERR(file);
10752 +       if (IS_ERR(file))
10753 +               goto out;
10754 +       err = 0;
10755 +       if (sbinfo->si_xigen)
10756 +               fput(sbinfo->si_xigen);
10757 +       sbinfo->si_xigen = file;
10758 +
10759 +out:
10760 +       return err;
10761 +}
10762 +
10763 +void au_xigen_clr(struct super_block *sb)
10764 +{
10765 +       struct au_sbinfo *sbinfo;
10766 +
10767 +       SiMustWriteLock(sb);
10768 +
10769 +       sbinfo = au_sbi(sb);
10770 +       if (sbinfo->si_xigen) {
10771 +               fput(sbinfo->si_xigen);
10772 +               sbinfo->si_xigen = NULL;
10773 +       }
10774 +}
10775 +
10776 +/* ---------------------------------------------------------------------- */
10777 +
10778 +static struct dentry *decode_by_ino(struct super_block *sb, ino_t ino,
10779 +                                   ino_t dir_ino)
10780 +{
10781 +       struct dentry *dentry, *d;
10782 +       struct inode *inode;
10783 +       unsigned int sigen;
10784 +
10785 +       dentry = NULL;
10786 +       inode = ilookup(sb, ino);
10787 +       if (!inode)
10788 +               goto out;
10789 +
10790 +       dentry = ERR_PTR(-ESTALE);
10791 +       sigen = au_sigen(sb);
10792 +       if (unlikely(is_bad_inode(inode)
10793 +                    || IS_DEADDIR(inode)
10794 +                    || sigen != au_iigen(inode, NULL)))
10795 +               goto out_iput;
10796 +
10797 +       dentry = NULL;
10798 +       if (!dir_ino || S_ISDIR(inode->i_mode))
10799 +               dentry = d_find_alias(inode);
10800 +       else {
10801 +               spin_lock(&inode->i_lock);
10802 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
10803 +                       spin_lock(&d->d_lock);
10804 +                       if (!au_test_anon(d)
10805 +                           && d->d_parent->d_inode->i_ino == dir_ino) {
10806 +                               dentry = dget_dlock(d);
10807 +                               spin_unlock(&d->d_lock);
10808 +                               break;
10809 +                       }
10810 +                       spin_unlock(&d->d_lock);
10811 +               }
10812 +               spin_unlock(&inode->i_lock);
10813 +       }
10814 +       if (unlikely(dentry && au_digen_test(dentry, sigen))) {
10815 +               /* need to refresh */
10816 +               dput(dentry);
10817 +               dentry = NULL;
10818 +       }
10819 +
10820 +out_iput:
10821 +       iput(inode);
10822 +out:
10823 +       AuTraceErrPtr(dentry);
10824 +       return dentry;
10825 +}
10826 +
10827 +/* ---------------------------------------------------------------------- */
10828 +
10829 +/* todo: dirty? */
10830 +/* if exportfs_decode_fh() passed vfsmount*, we could be happy */
10831 +
10832 +struct au_compare_mnt_args {
10833 +       /* input */
10834 +       struct super_block *sb;
10835 +
10836 +       /* output */
10837 +       struct vfsmount *mnt;
10838 +};
10839 +
10840 +static int au_compare_mnt(struct vfsmount *mnt, void *arg)
10841 +{
10842 +       struct au_compare_mnt_args *a = arg;
10843 +
10844 +       if (mnt->mnt_sb != a->sb)
10845 +               return 0;
10846 +       a->mnt = mntget(mnt);
10847 +       return 1;
10848 +}
10849 +
10850 +static struct vfsmount *au_mnt_get(struct super_block *sb)
10851 +{
10852 +       int err;
10853 +       struct path root;
10854 +       struct au_compare_mnt_args args = {
10855 +               .sb = sb
10856 +       };
10857 +
10858 +       get_fs_root(current->fs, &root);
10859 +       rcu_read_lock();
10860 +       err = iterate_mounts(au_compare_mnt, &args, root.mnt);
10861 +       rcu_read_unlock();
10862 +       path_put(&root);
10863 +       AuDebugOn(!err);
10864 +       AuDebugOn(!args.mnt);
10865 +       return args.mnt;
10866 +}
10867 +
10868 +struct au_nfsd_si_lock {
10869 +       unsigned int sigen;
10870 +       aufs_bindex_t bindex, br_id;
10871 +       unsigned char force_lock;
10872 +};
10873 +
10874 +static int si_nfsd_read_lock(struct super_block *sb,
10875 +                            struct au_nfsd_si_lock *nsi_lock)
10876 +{
10877 +       int err;
10878 +       aufs_bindex_t bindex;
10879 +
10880 +       si_read_lock(sb, AuLock_FLUSH);
10881 +
10882 +       /* branch id may be wrapped around */
10883 +       err = 0;
10884 +       bindex = au_br_index(sb, nsi_lock->br_id);
10885 +       if (bindex >= 0 && nsi_lock->sigen + AUFS_BRANCH_MAX > au_sigen(sb))
10886 +               goto out; /* success */
10887 +
10888 +       err = -ESTALE;
10889 +       bindex = -1;
10890 +       if (!nsi_lock->force_lock)
10891 +               si_read_unlock(sb);
10892 +
10893 +out:
10894 +       nsi_lock->bindex = bindex;
10895 +       return err;
10896 +}
10897 +
10898 +struct find_name_by_ino {
10899 +       struct dir_context ctx;
10900 +       int called, found;
10901 +       ino_t ino;
10902 +       char *name;
10903 +       int namelen;
10904 +};
10905 +
10906 +static int
10907 +find_name_by_ino(struct dir_context *ctx, const char *name, int namelen,
10908 +                loff_t offset, u64 ino, unsigned int d_type)
10909 +{
10910 +       struct find_name_by_ino *a = container_of(ctx, struct find_name_by_ino,
10911 +                                                 ctx);
10912 +
10913 +       a->called++;
10914 +       if (a->ino != ino)
10915 +               return 0;
10916 +
10917 +       memcpy(a->name, name, namelen);
10918 +       a->namelen = namelen;
10919 +       a->found = 1;
10920 +       return 1;
10921 +}
10922 +
10923 +static struct dentry *au_lkup_by_ino(struct path *path, ino_t ino,
10924 +                                    struct au_nfsd_si_lock *nsi_lock)
10925 +{
10926 +       struct dentry *dentry, *parent;
10927 +       struct file *file;
10928 +       struct inode *dir;
10929 +       struct find_name_by_ino arg = {
10930 +               .ctx = {
10931 +                       .actor = au_diractor(find_name_by_ino)
10932 +               }
10933 +       };
10934 +       int err;
10935 +
10936 +       parent = path->dentry;
10937 +       if (nsi_lock)
10938 +               si_read_unlock(parent->d_sb);
10939 +       file = vfsub_dentry_open(path, au_dir_roflags);
10940 +       dentry = (void *)file;
10941 +       if (IS_ERR(file))
10942 +               goto out;
10943 +
10944 +       dentry = ERR_PTR(-ENOMEM);
10945 +       arg.name = (void *)__get_free_page(GFP_NOFS);
10946 +       if (unlikely(!arg.name))
10947 +               goto out_file;
10948 +       arg.ino = ino;
10949 +       arg.found = 0;
10950 +       do {
10951 +               arg.called = 0;
10952 +               /* smp_mb(); */
10953 +               err = vfsub_iterate_dir(file, &arg.ctx);
10954 +       } while (!err && !arg.found && arg.called);
10955 +       dentry = ERR_PTR(err);
10956 +       if (unlikely(err))
10957 +               goto out_name;
10958 +       /* instead of ENOENT */
10959 +       dentry = ERR_PTR(-ESTALE);
10960 +       if (!arg.found)
10961 +               goto out_name;
10962 +
10963 +       /* do not call vfsub_lkup_one() */
10964 +       dir = parent->d_inode;
10965 +       mutex_lock(&dir->i_mutex);
10966 +       dentry = vfsub_lookup_one_len(arg.name, parent, arg.namelen);
10967 +       mutex_unlock(&dir->i_mutex);
10968 +       AuTraceErrPtr(dentry);
10969 +       if (IS_ERR(dentry))
10970 +               goto out_name;
10971 +       AuDebugOn(au_test_anon(dentry));
10972 +       if (unlikely(!dentry->d_inode)) {
10973 +               dput(dentry);
10974 +               dentry = ERR_PTR(-ENOENT);
10975 +       }
10976 +
10977 +out_name:
10978 +       free_page((unsigned long)arg.name);
10979 +out_file:
10980 +       fput(file);
10981 +out:
10982 +       if (unlikely(nsi_lock
10983 +                    && si_nfsd_read_lock(parent->d_sb, nsi_lock) < 0))
10984 +               if (!IS_ERR(dentry)) {
10985 +                       dput(dentry);
10986 +                       dentry = ERR_PTR(-ESTALE);
10987 +               }
10988 +       AuTraceErrPtr(dentry);
10989 +       return dentry;
10990 +}
10991 +
10992 +static struct dentry *decode_by_dir_ino(struct super_block *sb, ino_t ino,
10993 +                                       ino_t dir_ino,
10994 +                                       struct au_nfsd_si_lock *nsi_lock)
10995 +{
10996 +       struct dentry *dentry;
10997 +       struct path path;
10998 +
10999 +       if (dir_ino != AUFS_ROOT_INO) {
11000 +               path.dentry = decode_by_ino(sb, dir_ino, 0);
11001 +               dentry = path.dentry;
11002 +               if (!path.dentry || IS_ERR(path.dentry))
11003 +                       goto out;
11004 +               AuDebugOn(au_test_anon(path.dentry));
11005 +       } else
11006 +               path.dentry = dget(sb->s_root);
11007 +
11008 +       path.mnt = au_mnt_get(sb);
11009 +       dentry = au_lkup_by_ino(&path, ino, nsi_lock);
11010 +       path_put(&path);
11011 +
11012 +out:
11013 +       AuTraceErrPtr(dentry);
11014 +       return dentry;
11015 +}
11016 +
11017 +/* ---------------------------------------------------------------------- */
11018 +
11019 +static int h_acceptable(void *expv, struct dentry *dentry)
11020 +{
11021 +       return 1;
11022 +}
11023 +
11024 +static char *au_build_path(struct dentry *h_parent, struct path *h_rootpath,
11025 +                          char *buf, int len, struct super_block *sb)
11026 +{
11027 +       char *p;
11028 +       int n;
11029 +       struct path path;
11030 +
11031 +       p = d_path(h_rootpath, buf, len);
11032 +       if (IS_ERR(p))
11033 +               goto out;
11034 +       n = strlen(p);
11035 +
11036 +       path.mnt = h_rootpath->mnt;
11037 +       path.dentry = h_parent;
11038 +       p = d_path(&path, buf, len);
11039 +       if (IS_ERR(p))
11040 +               goto out;
11041 +       if (n != 1)
11042 +               p += n;
11043 +
11044 +       path.mnt = au_mnt_get(sb);
11045 +       path.dentry = sb->s_root;
11046 +       p = d_path(&path, buf, len - strlen(p));
11047 +       mntput(path.mnt);
11048 +       if (IS_ERR(p))
11049 +               goto out;
11050 +       if (n != 1)
11051 +               p[strlen(p)] = '/';
11052 +
11053 +out:
11054 +       AuTraceErrPtr(p);
11055 +       return p;
11056 +}
11057 +
11058 +static
11059 +struct dentry *decode_by_path(struct super_block *sb, ino_t ino, __u32 *fh,
11060 +                             int fh_len, struct au_nfsd_si_lock *nsi_lock)
11061 +{
11062 +       struct dentry *dentry, *h_parent, *root;
11063 +       struct super_block *h_sb;
11064 +       char *pathname, *p;
11065 +       struct vfsmount *h_mnt;
11066 +       struct au_branch *br;
11067 +       int err;
11068 +       struct path path;
11069 +
11070 +       br = au_sbr(sb, nsi_lock->bindex);
11071 +       h_mnt = au_br_mnt(br);
11072 +       h_sb = h_mnt->mnt_sb;
11073 +       /* todo: call lower fh_to_dentry()? fh_to_parent()? */
11074 +       h_parent = exportfs_decode_fh(h_mnt, (void *)(fh + Fh_tail),
11075 +                                     fh_len - Fh_tail, fh[Fh_h_type],
11076 +                                     h_acceptable, /*context*/NULL);
11077 +       dentry = h_parent;
11078 +       if (unlikely(!h_parent || IS_ERR(h_parent))) {
11079 +               AuWarn1("%s decode_fh failed, %ld\n",
11080 +                       au_sbtype(h_sb), PTR_ERR(h_parent));
11081 +               goto out;
11082 +       }
11083 +       dentry = NULL;
11084 +       if (unlikely(au_test_anon(h_parent))) {
11085 +               AuWarn1("%s decode_fh returned a disconnected dentry\n",
11086 +                       au_sbtype(h_sb));
11087 +               goto out_h_parent;
11088 +       }
11089 +
11090 +       dentry = ERR_PTR(-ENOMEM);
11091 +       pathname = (void *)__get_free_page(GFP_NOFS);
11092 +       if (unlikely(!pathname))
11093 +               goto out_h_parent;
11094 +
11095 +       root = sb->s_root;
11096 +       path.mnt = h_mnt;
11097 +       di_read_lock_parent(root, !AuLock_IR);
11098 +       path.dentry = au_h_dptr(root, nsi_lock->bindex);
11099 +       di_read_unlock(root, !AuLock_IR);
11100 +       p = au_build_path(h_parent, &path, pathname, PAGE_SIZE, sb);
11101 +       dentry = (void *)p;
11102 +       if (IS_ERR(p))
11103 +               goto out_pathname;
11104 +
11105 +       si_read_unlock(sb);
11106 +       err = vfsub_kern_path(p, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &path);
11107 +       dentry = ERR_PTR(err);
11108 +       if (unlikely(err))
11109 +               goto out_relock;
11110 +
11111 +       dentry = ERR_PTR(-ENOENT);
11112 +       AuDebugOn(au_test_anon(path.dentry));
11113 +       if (unlikely(!path.dentry->d_inode))
11114 +               goto out_path;
11115 +
11116 +       if (ino != path.dentry->d_inode->i_ino)
11117 +               dentry = au_lkup_by_ino(&path, ino, /*nsi_lock*/NULL);
11118 +       else
11119 +               dentry = dget(path.dentry);
11120 +
11121 +out_path:
11122 +       path_put(&path);
11123 +out_relock:
11124 +       if (unlikely(si_nfsd_read_lock(sb, nsi_lock) < 0))
11125 +               if (!IS_ERR(dentry)) {
11126 +                       dput(dentry);
11127 +                       dentry = ERR_PTR(-ESTALE);
11128 +               }
11129 +out_pathname:
11130 +       free_page((unsigned long)pathname);
11131 +out_h_parent:
11132 +       dput(h_parent);
11133 +out:
11134 +       AuTraceErrPtr(dentry);
11135 +       return dentry;
11136 +}
11137 +
11138 +/* ---------------------------------------------------------------------- */
11139 +
11140 +static struct dentry *
11141 +aufs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
11142 +                 int fh_type)
11143 +{
11144 +       struct dentry *dentry;
11145 +       __u32 *fh = fid->raw;
11146 +       struct au_branch *br;
11147 +       ino_t ino, dir_ino;
11148 +       struct au_nfsd_si_lock nsi_lock = {
11149 +               .force_lock     = 0
11150 +       };
11151 +
11152 +       dentry = ERR_PTR(-ESTALE);
11153 +       /* it should never happen, but the file handle is unreliable */
11154 +       if (unlikely(fh_len < Fh_tail))
11155 +               goto out;
11156 +       nsi_lock.sigen = fh[Fh_sigen];
11157 +       nsi_lock.br_id = fh[Fh_br_id];
11158 +
11159 +       /* branch id may be wrapped around */
11160 +       br = NULL;
11161 +       if (unlikely(si_nfsd_read_lock(sb, &nsi_lock)))
11162 +               goto out;
11163 +       nsi_lock.force_lock = 1;
11164 +
11165 +       /* is this inode still cached? */
11166 +       ino = decode_ino(fh + Fh_ino);
11167 +       /* it should never happen */
11168 +       if (unlikely(ino == AUFS_ROOT_INO))
11169 +               goto out;
11170 +
11171 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11172 +       dentry = decode_by_ino(sb, ino, dir_ino);
11173 +       if (IS_ERR(dentry))
11174 +               goto out_unlock;
11175 +       if (dentry)
11176 +               goto accept;
11177 +
11178 +       /* is the parent dir cached? */
11179 +       br = au_sbr(sb, nsi_lock.bindex);
11180 +       atomic_inc(&br->br_count);
11181 +       dentry = decode_by_dir_ino(sb, ino, dir_ino, &nsi_lock);
11182 +       if (IS_ERR(dentry))
11183 +               goto out_unlock;
11184 +       if (dentry)
11185 +               goto accept;
11186 +
11187 +       /* lookup path */
11188 +       dentry = decode_by_path(sb, ino, fh, fh_len, &nsi_lock);
11189 +       if (IS_ERR(dentry))
11190 +               goto out_unlock;
11191 +       if (unlikely(!dentry))
11192 +               /* todo?: make it ESTALE */
11193 +               goto out_unlock;
11194 +
11195 +accept:
11196 +       if (!au_digen_test(dentry, au_sigen(sb))
11197 +           && dentry->d_inode->i_generation == fh[Fh_igen])
11198 +               goto out_unlock; /* success */
11199 +
11200 +       dput(dentry);
11201 +       dentry = ERR_PTR(-ESTALE);
11202 +out_unlock:
11203 +       if (br)
11204 +               atomic_dec(&br->br_count);
11205 +       si_read_unlock(sb);
11206 +out:
11207 +       AuTraceErrPtr(dentry);
11208 +       return dentry;
11209 +}
11210 +
11211 +#if 0 /* reserved for future use */
11212 +/* support subtreecheck option */
11213 +static struct dentry *aufs_fh_to_parent(struct super_block *sb, struct fid *fid,
11214 +                                       int fh_len, int fh_type)
11215 +{
11216 +       struct dentry *parent;
11217 +       __u32 *fh = fid->raw;
11218 +       ino_t dir_ino;
11219 +
11220 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11221 +       parent = decode_by_ino(sb, dir_ino, 0);
11222 +       if (IS_ERR(parent))
11223 +               goto out;
11224 +       if (!parent)
11225 +               parent = decode_by_path(sb, au_br_index(sb, fh[Fh_br_id]),
11226 +                                       dir_ino, fh, fh_len);
11227 +
11228 +out:
11229 +       AuTraceErrPtr(parent);
11230 +       return parent;
11231 +}
11232 +#endif
11233 +
11234 +/* ---------------------------------------------------------------------- */
11235 +
11236 +static int aufs_encode_fh(struct inode *inode, __u32 *fh, int *max_len,
11237 +                         struct inode *dir)
11238 +{
11239 +       int err;
11240 +       aufs_bindex_t bindex;
11241 +       struct super_block *sb, *h_sb;
11242 +       struct dentry *dentry, *parent, *h_parent;
11243 +       struct inode *h_dir;
11244 +       struct au_branch *br;
11245 +
11246 +       err = -ENOSPC;
11247 +       if (unlikely(*max_len <= Fh_tail)) {
11248 +               AuWarn1("NFSv2 client (max_len %d)?\n", *max_len);
11249 +               goto out;
11250 +       }
11251 +
11252 +       err = FILEID_ROOT;
11253 +       if (inode->i_ino == AUFS_ROOT_INO) {
11254 +               AuDebugOn(inode->i_ino != AUFS_ROOT_INO);
11255 +               goto out;
11256 +       }
11257 +
11258 +       h_parent = NULL;
11259 +       sb = inode->i_sb;
11260 +       err = si_read_lock(sb, AuLock_FLUSH);
11261 +       if (unlikely(err))
11262 +               goto out;
11263 +
11264 +#ifdef CONFIG_AUFS_DEBUG
11265 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
11266 +               AuWarn1("NFS-exporting requires xino\n");
11267 +#endif
11268 +       err = -EIO;
11269 +       parent = NULL;
11270 +       ii_read_lock_child(inode);
11271 +       bindex = au_ibstart(inode);
11272 +       if (!dir) {
11273 +               dentry = d_find_any_alias(inode);
11274 +               if (unlikely(!dentry))
11275 +                       goto out_unlock;
11276 +               AuDebugOn(au_test_anon(dentry));
11277 +               parent = dget_parent(dentry);
11278 +               dput(dentry);
11279 +               if (unlikely(!parent))
11280 +                       goto out_unlock;
11281 +               dir = parent->d_inode;
11282 +       }
11283 +
11284 +       ii_read_lock_parent(dir);
11285 +       h_dir = au_h_iptr(dir, bindex);
11286 +       ii_read_unlock(dir);
11287 +       if (unlikely(!h_dir))
11288 +               goto out_parent;
11289 +       h_parent = d_find_any_alias(h_dir);
11290 +       if (unlikely(!h_parent))
11291 +               goto out_hparent;
11292 +
11293 +       err = -EPERM;
11294 +       br = au_sbr(sb, bindex);
11295 +       h_sb = au_br_sb(br);
11296 +       if (unlikely(!h_sb->s_export_op)) {
11297 +               AuErr1("%s branch is not exportable\n", au_sbtype(h_sb));
11298 +               goto out_hparent;
11299 +       }
11300 +
11301 +       fh[Fh_br_id] = br->br_id;
11302 +       fh[Fh_sigen] = au_sigen(sb);
11303 +       encode_ino(fh + Fh_ino, inode->i_ino);
11304 +       encode_ino(fh + Fh_dir_ino, dir->i_ino);
11305 +       fh[Fh_igen] = inode->i_generation;
11306 +
11307 +       *max_len -= Fh_tail;
11308 +       fh[Fh_h_type] = exportfs_encode_fh(h_parent, (void *)(fh + Fh_tail),
11309 +                                          max_len,
11310 +                                          /*connectable or subtreecheck*/0);
11311 +       err = fh[Fh_h_type];
11312 +       *max_len += Fh_tail;
11313 +       /* todo: macros? */
11314 +       if (err != FILEID_INVALID)
11315 +               err = 99;
11316 +       else
11317 +               AuWarn1("%s encode_fh failed\n", au_sbtype(h_sb));
11318 +
11319 +out_hparent:
11320 +       dput(h_parent);
11321 +out_parent:
11322 +       dput(parent);
11323 +out_unlock:
11324 +       ii_read_unlock(inode);
11325 +       si_read_unlock(sb);
11326 +out:
11327 +       if (unlikely(err < 0))
11328 +               err = FILEID_INVALID;
11329 +       return err;
11330 +}
11331 +
11332 +/* ---------------------------------------------------------------------- */
11333 +
11334 +static int aufs_commit_metadata(struct inode *inode)
11335 +{
11336 +       int err;
11337 +       aufs_bindex_t bindex;
11338 +       struct super_block *sb;
11339 +       struct inode *h_inode;
11340 +       int (*f)(struct inode *inode);
11341 +
11342 +       sb = inode->i_sb;
11343 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
11344 +       ii_write_lock_child(inode);
11345 +       bindex = au_ibstart(inode);
11346 +       AuDebugOn(bindex < 0);
11347 +       h_inode = au_h_iptr(inode, bindex);
11348 +
11349 +       f = h_inode->i_sb->s_export_op->commit_metadata;
11350 +       if (f)
11351 +               err = f(h_inode);
11352 +       else {
11353 +               struct writeback_control wbc = {
11354 +                       .sync_mode      = WB_SYNC_ALL,
11355 +                       .nr_to_write    = 0 /* metadata only */
11356 +               };
11357 +
11358 +               err = sync_inode(h_inode, &wbc);
11359 +       }
11360 +
11361 +       au_cpup_attr_timesizes(inode);
11362 +       ii_write_unlock(inode);
11363 +       si_read_unlock(sb);
11364 +       return err;
11365 +}
11366 +
11367 +/* ---------------------------------------------------------------------- */
11368 +
11369 +static struct export_operations aufs_export_op = {
11370 +       .fh_to_dentry           = aufs_fh_to_dentry,
11371 +       /* .fh_to_parent        = aufs_fh_to_parent, */
11372 +       .encode_fh              = aufs_encode_fh,
11373 +       .commit_metadata        = aufs_commit_metadata
11374 +};
11375 +
11376 +void au_export_init(struct super_block *sb)
11377 +{
11378 +       struct au_sbinfo *sbinfo;
11379 +       __u32 u;
11380 +
11381 +       sb->s_export_op = &aufs_export_op;
11382 +       sbinfo = au_sbi(sb);
11383 +       sbinfo->si_xigen = NULL;
11384 +       get_random_bytes(&u, sizeof(u));
11385 +       BUILD_BUG_ON(sizeof(u) != sizeof(int));
11386 +       atomic_set(&sbinfo->si_xigen_next, u);
11387 +}
11388 diff -urN /usr/share/empty/fs/aufs/fhsm.c linux/fs/aufs/fhsm.c
11389 --- /usr/share/empty/fs/aufs/fhsm.c     1970-01-01 01:00:00.000000000 +0100
11390 +++ linux/fs/aufs/fhsm.c        2015-01-25 13:00:38.631047076 +0100
11391 @@ -0,0 +1,426 @@
11392 +/*
11393 + * Copyright (C) 2011-2014 Junjiro R. Okajima
11394 + *
11395 + * This program, aufs is free software; you can redistribute it and/or modify
11396 + * it under the terms of the GNU General Public License as published by
11397 + * the Free Software Foundation; either version 2 of the License, or
11398 + * (at your option) any later version.
11399 + *
11400 + * This program is distributed in the hope that it will be useful,
11401 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11402 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11403 + * GNU General Public License for more details.
11404 + *
11405 + * You should have received a copy of the GNU General Public License
11406 + * along with this program; if not, write to the Free Software
11407 + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
11408 + */
11409 +
11410 +/*
11411 + * File-based Hierarchy Storage Management
11412 + */
11413 +
11414 +#include <linux/anon_inodes.h>
11415 +#include <linux/poll.h>
11416 +#include <linux/seq_file.h>
11417 +#include <linux/statfs.h>
11418 +#include "aufs.h"
11419 +
11420 +static aufs_bindex_t au_fhsm_bottom(struct super_block *sb)
11421 +{
11422 +       struct au_sbinfo *sbinfo;
11423 +       struct au_fhsm *fhsm;
11424 +
11425 +       SiMustAnyLock(sb);
11426 +
11427 +       sbinfo = au_sbi(sb);
11428 +       fhsm = &sbinfo->si_fhsm;
11429 +       AuDebugOn(!fhsm);
11430 +       return fhsm->fhsm_bottom;
11431 +}
11432 +
11433 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex)
11434 +{
11435 +       struct au_sbinfo *sbinfo;
11436 +       struct au_fhsm *fhsm;
11437 +
11438 +       SiMustWriteLock(sb);
11439 +
11440 +       sbinfo = au_sbi(sb);
11441 +       fhsm = &sbinfo->si_fhsm;
11442 +       AuDebugOn(!fhsm);
11443 +       fhsm->fhsm_bottom = bindex;
11444 +}
11445 +
11446 +/* ---------------------------------------------------------------------- */
11447 +
11448 +static int au_fhsm_test_jiffy(struct au_sbinfo *sbinfo, struct au_branch *br)
11449 +{
11450 +       struct au_br_fhsm *bf;
11451 +
11452 +       bf = br->br_fhsm;
11453 +       MtxMustLock(&bf->bf_lock);
11454 +
11455 +       return !bf->bf_readable
11456 +               || time_after(jiffies,
11457 +                             bf->bf_jiffy + sbinfo->si_fhsm.fhsm_expire);
11458 +}
11459 +
11460 +/* ---------------------------------------------------------------------- */
11461 +
11462 +static void au_fhsm_notify(struct super_block *sb, int val)
11463 +{
11464 +       struct au_sbinfo *sbinfo;
11465 +       struct au_fhsm *fhsm;
11466 +
11467 +       SiMustAnyLock(sb);
11468 +
11469 +       sbinfo = au_sbi(sb);
11470 +       fhsm = &sbinfo->si_fhsm;
11471 +       if (au_fhsm_pid(fhsm)
11472 +           && atomic_read(&fhsm->fhsm_readable) != -1) {
11473 +               atomic_set(&fhsm->fhsm_readable, val);
11474 +               if (val)
11475 +                       wake_up(&fhsm->fhsm_wqh);
11476 +       }
11477 +}
11478 +
11479 +static int au_fhsm_stfs(struct super_block *sb, aufs_bindex_t bindex,
11480 +                       struct aufs_stfs *rstfs, int do_lock, int do_notify)
11481 +{
11482 +       int err;
11483 +       struct au_branch *br;
11484 +       struct au_br_fhsm *bf;
11485 +
11486 +       br = au_sbr(sb, bindex);
11487 +       AuDebugOn(au_br_rdonly(br));
11488 +       bf = br->br_fhsm;
11489 +       AuDebugOn(!bf);
11490 +
11491 +       if (do_lock)
11492 +               mutex_lock(&bf->bf_lock);
11493 +       else
11494 +               MtxMustLock(&bf->bf_lock);
11495 +
11496 +       /* sb->s_root for NFS is unreliable */
11497 +       err = au_br_stfs(br, &bf->bf_stfs);
11498 +       if (unlikely(err)) {
11499 +               AuErr1("FHSM failed (%d), b%d, ignored.\n", bindex, err);
11500 +               goto out;
11501 +       }
11502 +
11503 +       bf->bf_jiffy = jiffies;
11504 +       bf->bf_readable = 1;
11505 +       if (do_notify)
11506 +               au_fhsm_notify(sb, /*val*/1);
11507 +       if (rstfs)
11508 +               *rstfs = bf->bf_stfs;
11509 +
11510 +out:
11511 +       if (do_lock)
11512 +               mutex_unlock(&bf->bf_lock);
11513 +       au_fhsm_notify(sb, /*val*/1);
11514 +
11515 +       return err;
11516 +}
11517 +
11518 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force)
11519 +{
11520 +       int err;
11521 +       struct au_sbinfo *sbinfo;
11522 +       struct au_fhsm *fhsm;
11523 +       struct au_branch *br;
11524 +       struct au_br_fhsm *bf;
11525 +
11526 +       AuDbg("b%d, force %d\n", bindex, force);
11527 +       SiMustAnyLock(sb);
11528 +
11529 +       sbinfo = au_sbi(sb);
11530 +       fhsm = &sbinfo->si_fhsm;
11531 +       if (!au_ftest_si(sbinfo, FHSM)
11532 +           || fhsm->fhsm_bottom == bindex)
11533 +               return;
11534 +
11535 +       br = au_sbr(sb, bindex);
11536 +       bf = br->br_fhsm;
11537 +       AuDebugOn(!bf);
11538 +       mutex_lock(&bf->bf_lock);
11539 +       if (force
11540 +           || au_fhsm_pid(fhsm)
11541 +           || au_fhsm_test_jiffy(sbinfo, br))
11542 +               err = au_fhsm_stfs(sb, bindex, /*rstfs*/NULL, /*do_lock*/0,
11543 +                                 /*do_notify*/1);
11544 +       mutex_unlock(&bf->bf_lock);
11545 +}
11546 +
11547 +void au_fhsm_wrote_all(struct super_block *sb, int force)
11548 +{
11549 +       aufs_bindex_t bindex, bend;
11550 +       struct au_branch *br;
11551 +
11552 +       /* exclude the bottom */
11553 +       bend = au_fhsm_bottom(sb);
11554 +       for (bindex = 0; bindex < bend; bindex++) {
11555 +               br = au_sbr(sb, bindex);
11556 +               if (au_br_fhsm(br->br_perm))
11557 +                       au_fhsm_wrote(sb, bindex, force);
11558 +       }
11559 +}
11560 +
11561 +/* ---------------------------------------------------------------------- */
11562 +
11563 +static unsigned int au_fhsm_poll(struct file *file,
11564 +                                struct poll_table_struct *wait)
11565 +{
11566 +       unsigned int mask;
11567 +       struct au_sbinfo *sbinfo;
11568 +       struct au_fhsm *fhsm;
11569 +
11570 +       mask = 0;
11571 +       sbinfo = file->private_data;
11572 +       fhsm = &sbinfo->si_fhsm;
11573 +       poll_wait(file, &fhsm->fhsm_wqh, wait);
11574 +       if (atomic_read(&fhsm->fhsm_readable))
11575 +               mask = POLLIN /* | POLLRDNORM */;
11576 +
11577 +       AuTraceErr((int)mask);
11578 +       return mask;
11579 +}
11580 +
11581 +static int au_fhsm_do_read_one(struct aufs_stbr __user *stbr,
11582 +                             struct aufs_stfs *stfs, __s16 brid)
11583 +{
11584 +       int err;
11585 +
11586 +       err = copy_to_user(&stbr->stfs, stfs, sizeof(*stfs));
11587 +       if (!err)
11588 +               err = __put_user(brid, &stbr->brid);
11589 +       if (unlikely(err))
11590 +               err = -EFAULT;
11591 +
11592 +       return err;
11593 +}
11594 +
11595 +static ssize_t au_fhsm_do_read(struct super_block *sb,
11596 +                              struct aufs_stbr __user *stbr, size_t count)
11597 +{
11598 +       ssize_t err;
11599 +       int nstbr;
11600 +       aufs_bindex_t bindex, bend;
11601 +       struct au_branch *br;
11602 +       struct au_br_fhsm *bf;
11603 +
11604 +       /* except the bottom branch */
11605 +       err = 0;
11606 +       nstbr = 0;
11607 +       bend = au_fhsm_bottom(sb);
11608 +       for (bindex = 0; !err && bindex < bend; bindex++) {
11609 +               br = au_sbr(sb, bindex);
11610 +               if (!au_br_fhsm(br->br_perm))
11611 +                       continue;
11612 +
11613 +               bf = br->br_fhsm;
11614 +               mutex_lock(&bf->bf_lock);
11615 +               if (bf->bf_readable) {
11616 +                       err = -EFAULT;
11617 +                       if (count >= sizeof(*stbr))
11618 +                               err = au_fhsm_do_read_one(stbr++, &bf->bf_stfs,
11619 +                                                         br->br_id);
11620 +                       if (!err) {
11621 +                               bf->bf_readable = 0;
11622 +                               count -= sizeof(*stbr);
11623 +                               nstbr++;
11624 +                       }
11625 +               }
11626 +               mutex_unlock(&bf->bf_lock);
11627 +       }
11628 +       if (!err)
11629 +               err = sizeof(*stbr) * nstbr;
11630 +
11631 +       return err;
11632 +}
11633 +
11634 +static ssize_t au_fhsm_read(struct file *file, char __user *buf, size_t count,
11635 +                          loff_t *pos)
11636 +{
11637 +       ssize_t err;
11638 +       int readable;
11639 +       aufs_bindex_t nfhsm, bindex, bend;
11640 +       struct au_sbinfo *sbinfo;
11641 +       struct au_fhsm *fhsm;
11642 +       struct au_branch *br;
11643 +       struct super_block *sb;
11644 +
11645 +       err = 0;
11646 +       sbinfo = file->private_data;
11647 +       fhsm = &sbinfo->si_fhsm;
11648 +need_data:
11649 +       spin_lock_irq(&fhsm->fhsm_wqh.lock);
11650 +       if (!atomic_read(&fhsm->fhsm_readable)) {
11651 +               if (vfsub_file_flags(file) & O_NONBLOCK)
11652 +                       err = -EAGAIN;
11653 +               else
11654 +                       err = wait_event_interruptible_locked_irq
11655 +                               (fhsm->fhsm_wqh,
11656 +                                atomic_read(&fhsm->fhsm_readable));
11657 +       }
11658 +       spin_unlock_irq(&fhsm->fhsm_wqh.lock);
11659 +       if (unlikely(err))
11660 +               goto out;
11661 +
11662 +       /* sb may already be dead */
11663 +       au_rw_read_lock(&sbinfo->si_rwsem);
11664 +       readable = atomic_read(&fhsm->fhsm_readable);
11665 +       if (readable > 0) {
11666 +               sb = sbinfo->si_sb;
11667 +               AuDebugOn(!sb);
11668 +               /* exclude the bottom branch */
11669 +               nfhsm = 0;
11670 +               bend = au_fhsm_bottom(sb);
11671 +               for (bindex = 0; bindex < bend; bindex++) {
11672 +                       br = au_sbr(sb, bindex);
11673 +                       if (au_br_fhsm(br->br_perm))
11674 +                               nfhsm++;
11675 +               }
11676 +               err = -EMSGSIZE;
11677 +               if (nfhsm * sizeof(struct aufs_stbr) <= count) {
11678 +                       atomic_set(&fhsm->fhsm_readable, 0);
11679 +                       err = au_fhsm_do_read(sbinfo->si_sb, (void __user *)buf,
11680 +                                            count);
11681 +               }
11682 +       }
11683 +       au_rw_read_unlock(&sbinfo->si_rwsem);
11684 +       if (!readable)
11685 +               goto need_data;
11686 +
11687 +out:
11688 +       return err;
11689 +}
11690 +
11691 +static int au_fhsm_release(struct inode *inode, struct file *file)
11692 +{
11693 +       struct au_sbinfo *sbinfo;
11694 +       struct au_fhsm *fhsm;
11695 +
11696 +       /* sb may already be dead */
11697 +       sbinfo = file->private_data;
11698 +       fhsm = &sbinfo->si_fhsm;
11699 +       spin_lock(&fhsm->fhsm_spin);
11700 +       fhsm->fhsm_pid = 0;
11701 +       spin_unlock(&fhsm->fhsm_spin);
11702 +       kobject_put(&sbinfo->si_kobj);
11703 +
11704 +       return 0;
11705 +}
11706 +
11707 +static const struct file_operations au_fhsm_fops = {
11708 +       .owner          = THIS_MODULE,
11709 +       .llseek         = noop_llseek,
11710 +       .read           = au_fhsm_read,
11711 +       .poll           = au_fhsm_poll,
11712 +       .release        = au_fhsm_release
11713 +};
11714 +
11715 +int au_fhsm_fd(struct super_block *sb, int oflags)
11716 +{
11717 +       int err, fd;
11718 +       struct au_sbinfo *sbinfo;
11719 +       struct au_fhsm *fhsm;
11720 +
11721 +       err = -EPERM;
11722 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
11723 +               goto out;
11724 +
11725 +       err = -EINVAL;
11726 +       if (unlikely(oflags & ~(O_CLOEXEC | O_NONBLOCK)))
11727 +               goto out;
11728 +
11729 +       err = 0;
11730 +       sbinfo = au_sbi(sb);
11731 +       fhsm = &sbinfo->si_fhsm;
11732 +       spin_lock(&fhsm->fhsm_spin);
11733 +       if (!fhsm->fhsm_pid)
11734 +               fhsm->fhsm_pid = current->pid;
11735 +       else
11736 +               err = -EBUSY;
11737 +       spin_unlock(&fhsm->fhsm_spin);
11738 +       if (unlikely(err))
11739 +               goto out;
11740 +
11741 +       oflags |= O_RDONLY;
11742 +       /* oflags |= FMODE_NONOTIFY; */
11743 +       fd = anon_inode_getfd("[aufs_fhsm]", &au_fhsm_fops, sbinfo, oflags);
11744 +       err = fd;
11745 +       if (unlikely(fd < 0))
11746 +               goto out_pid;
11747 +
11748 +       /* succeed reglardless 'fhsm' status */
11749 +       kobject_get(&sbinfo->si_kobj);
11750 +       si_noflush_read_lock(sb);
11751 +       if (au_ftest_si(sbinfo, FHSM))
11752 +               au_fhsm_wrote_all(sb, /*force*/0);
11753 +       si_read_unlock(sb);
11754 +       goto out; /* success */
11755 +
11756 +out_pid:
11757 +       spin_lock(&fhsm->fhsm_spin);
11758 +       fhsm->fhsm_pid = 0;
11759 +       spin_unlock(&fhsm->fhsm_spin);
11760 +out:
11761 +       AuTraceErr(err);
11762 +       return err;
11763 +}
11764 +
11765 +/* ---------------------------------------------------------------------- */
11766 +
11767 +int au_fhsm_br_alloc(struct au_branch *br)
11768 +{
11769 +       int err;
11770 +
11771 +       err = 0;
11772 +       br->br_fhsm = kmalloc(sizeof(*br->br_fhsm), GFP_NOFS);
11773 +       if (br->br_fhsm)
11774 +               au_br_fhsm_init(br->br_fhsm);
11775 +       else
11776 +               err = -ENOMEM;
11777 +
11778 +       return err;
11779 +}
11780 +
11781 +/* ---------------------------------------------------------------------- */
11782 +
11783 +void au_fhsm_fin(struct super_block *sb)
11784 +{
11785 +       au_fhsm_notify(sb, /*val*/-1);
11786 +}
11787 +
11788 +void au_fhsm_init(struct au_sbinfo *sbinfo)
11789 +{
11790 +       struct au_fhsm *fhsm;
11791 +
11792 +       fhsm = &sbinfo->si_fhsm;
11793 +       spin_lock_init(&fhsm->fhsm_spin);
11794 +       init_waitqueue_head(&fhsm->fhsm_wqh);
11795 +       atomic_set(&fhsm->fhsm_readable, 0);
11796 +       fhsm->fhsm_expire
11797 +               = msecs_to_jiffies(AUFS_FHSM_CACHE_DEF_SEC * MSEC_PER_SEC);
11798 +       fhsm->fhsm_bottom = -1;
11799 +}
11800 +
11801 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec)
11802 +{
11803 +       sbinfo->si_fhsm.fhsm_expire
11804 +               = msecs_to_jiffies(sec * MSEC_PER_SEC);
11805 +}
11806 +
11807 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo)
11808 +{
11809 +       unsigned int u;
11810 +
11811 +       if (!au_ftest_si(sbinfo, FHSM))
11812 +               return;
11813 +
11814 +       u = jiffies_to_msecs(sbinfo->si_fhsm.fhsm_expire) / MSEC_PER_SEC;
11815 +       if (u != AUFS_FHSM_CACHE_DEF_SEC)
11816 +               seq_printf(seq, ",fhsm_sec=%u", u);
11817 +}
11818 diff -urN /usr/share/empty/fs/aufs/file.c linux/fs/aufs/file.c
11819 --- /usr/share/empty/fs/aufs/file.c     1970-01-01 01:00:00.000000000 +0100
11820 +++ linux/fs/aufs/file.c        2015-01-25 13:00:38.631047076 +0100
11821 @@ -0,0 +1,829 @@
11822 +/*
11823 + * Copyright (C) 2005-2014 Junjiro R. Okajima
11824 + *
11825 + * This program, aufs is free software; you can redistribute it and/or modify
11826 + * it under the terms of the GNU General Public License as published by
11827 + * the Free Software Foundation; either version 2 of the License, or
11828 + * (at your option) any later version.
11829 + *
11830 + * This program is distributed in the hope that it will be useful,
11831 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11832 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11833 + * GNU General Public License for more details.
11834 + *
11835 + * You should have received a copy of the GNU General Public License
11836 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
11837 + */
11838 +
11839 +/*
11840 + * handling file/dir, and address_space operation
11841 + */
11842 +
11843 +#ifdef CONFIG_AUFS_DEBUG
11844 +#include <linux/migrate.h>
11845 +#endif
11846 +#include <linux/pagemap.h>
11847 +#include "aufs.h"
11848 +
11849 +/* drop flags for writing */
11850 +unsigned int au_file_roflags(unsigned int flags)
11851 +{
11852 +       flags &= ~(O_WRONLY | O_RDWR | O_APPEND | O_CREAT | O_TRUNC);
11853 +       flags |= O_RDONLY | O_NOATIME;
11854 +       return flags;
11855 +}
11856 +
11857 +/* common functions to regular file and dir */
11858 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
11859 +                      struct file *file, int force_wr)
11860 +{
11861 +       struct file *h_file;
11862 +       struct dentry *h_dentry;
11863 +       struct inode *h_inode;
11864 +       struct super_block *sb;
11865 +       struct au_branch *br;
11866 +       struct path h_path;
11867 +       int err, exec_flag;
11868 +
11869 +       /* a race condition can happen between open and unlink/rmdir */
11870 +       h_file = ERR_PTR(-ENOENT);
11871 +       h_dentry = au_h_dptr(dentry, bindex);
11872 +       if (au_test_nfsd() && !h_dentry)
11873 +               goto out;
11874 +       h_inode = h_dentry->d_inode;
11875 +       if (au_test_nfsd() && !h_inode)
11876 +               goto out;
11877 +       spin_lock(&h_dentry->d_lock);
11878 +       err = (!d_unhashed(dentry) && d_unlinked(h_dentry))
11879 +               || !h_inode
11880 +               /* || !dentry->d_inode->i_nlink */
11881 +               ;
11882 +       spin_unlock(&h_dentry->d_lock);
11883 +       if (unlikely(err))
11884 +               goto out;
11885 +
11886 +       sb = dentry->d_sb;
11887 +       br = au_sbr(sb, bindex);
11888 +       h_file = ERR_PTR(-EACCES);
11889 +       exec_flag = flags & __FMODE_EXEC;
11890 +       if (exec_flag && (au_br_mnt(br)->mnt_flags & MNT_NOEXEC))
11891 +               goto out;
11892 +
11893 +       /* drop flags for writing */
11894 +       if (au_test_ro(sb, bindex, dentry->d_inode)) {
11895 +               if (force_wr && !(flags & O_WRONLY))
11896 +                       force_wr = 0;
11897 +               flags = au_file_roflags(flags);
11898 +               if (force_wr) {
11899 +                       h_file = ERR_PTR(-EROFS);
11900 +                       flags = au_file_roflags(flags);
11901 +                       if (unlikely(vfsub_native_ro(h_inode)
11902 +                                    || IS_APPEND(h_inode)))
11903 +                               goto out;
11904 +                       flags &= ~O_ACCMODE;
11905 +                       flags |= O_WRONLY;
11906 +               }
11907 +       }
11908 +       flags &= ~O_CREAT;
11909 +       atomic_inc(&br->br_count);
11910 +       h_path.dentry = h_dentry;
11911 +       h_path.mnt = au_br_mnt(br);
11912 +       h_file = vfsub_dentry_open(&h_path, flags);
11913 +       if (IS_ERR(h_file))
11914 +               goto out_br;
11915 +
11916 +       if (exec_flag) {
11917 +               err = deny_write_access(h_file);
11918 +               if (unlikely(err)) {
11919 +                       fput(h_file);
11920 +                       h_file = ERR_PTR(err);
11921 +                       goto out_br;
11922 +               }
11923 +       }
11924 +       fsnotify_open(h_file);
11925 +       goto out; /* success */
11926 +
11927 +out_br:
11928 +       atomic_dec(&br->br_count);
11929 +out:
11930 +       return h_file;
11931 +}
11932 +
11933 +static int au_cmoo(struct dentry *dentry)
11934 +{
11935 +       int err, cmoo;
11936 +       unsigned int udba;
11937 +       struct path h_path;
11938 +       struct au_pin pin;
11939 +       struct au_cp_generic cpg = {
11940 +               .dentry = dentry,
11941 +               .bdst   = -1,
11942 +               .bsrc   = -1,
11943 +               .len    = -1,
11944 +               .pin    = &pin,
11945 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
11946 +       };
11947 +       struct inode *inode, *delegated;
11948 +       struct super_block *sb;
11949 +       struct au_sbinfo *sbinfo;
11950 +       struct au_fhsm *fhsm;
11951 +       pid_t pid;
11952 +       struct au_branch *br;
11953 +       struct dentry *parent;
11954 +       struct au_hinode *hdir;
11955 +
11956 +       DiMustWriteLock(dentry);
11957 +       inode = dentry->d_inode;
11958 +       IiMustWriteLock(inode);
11959 +
11960 +       err = 0;
11961 +       if (IS_ROOT(dentry))
11962 +               goto out;
11963 +       cpg.bsrc = au_dbstart(dentry);
11964 +       if (!cpg.bsrc)
11965 +               goto out;
11966 +
11967 +       sb = dentry->d_sb;
11968 +       sbinfo = au_sbi(sb);
11969 +       fhsm = &sbinfo->si_fhsm;
11970 +       pid = au_fhsm_pid(fhsm);
11971 +       if (pid
11972 +           && (current->pid == pid
11973 +               || current->real_parent->pid == pid))
11974 +               goto out;
11975 +
11976 +       br = au_sbr(sb, cpg.bsrc);
11977 +       cmoo = au_br_cmoo(br->br_perm);
11978 +       if (!cmoo)
11979 +               goto out;
11980 +       if (!S_ISREG(inode->i_mode))
11981 +               cmoo &= AuBrAttr_COO_ALL;
11982 +       if (!cmoo)
11983 +               goto out;
11984 +
11985 +       parent = dget_parent(dentry);
11986 +       di_write_lock_parent(parent);
11987 +       err = au_wbr_do_copyup_bu(dentry, cpg.bsrc - 1);
11988 +       cpg.bdst = err;
11989 +       if (unlikely(err < 0)) {
11990 +               err = 0;        /* there is no upper writable branch */
11991 +               goto out_dgrade;
11992 +       }
11993 +       AuDbg("bsrc %d, bdst %d\n", cpg.bsrc, cpg.bdst);
11994 +
11995 +       /* do not respect the coo attrib for the target branch */
11996 +       err = au_cpup_dirs(dentry, cpg.bdst);
11997 +       if (unlikely(err))
11998 +               goto out_dgrade;
11999 +
12000 +       di_downgrade_lock(parent, AuLock_IR);
12001 +       udba = au_opt_udba(sb);
12002 +       err = au_pin(&pin, dentry, cpg.bdst, udba,
12003 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12004 +       if (unlikely(err))
12005 +               goto out_parent;
12006 +
12007 +       err = au_sio_cpup_simple(&cpg);
12008 +       au_unpin(&pin);
12009 +       if (unlikely(err))
12010 +               goto out_parent;
12011 +       if (!(cmoo & AuBrWAttr_MOO))
12012 +               goto out_parent; /* success */
12013 +
12014 +       err = au_pin(&pin, dentry, cpg.bsrc, udba,
12015 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12016 +       if (unlikely(err))
12017 +               goto out_parent;
12018 +
12019 +       h_path.mnt = au_br_mnt(br);
12020 +       h_path.dentry = au_h_dptr(dentry, cpg.bsrc);
12021 +       hdir = au_hi(parent->d_inode, cpg.bsrc);
12022 +       delegated = NULL;
12023 +       err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated, /*force*/1);
12024 +       au_unpin(&pin);
12025 +       /* todo: keep h_dentry or not? */
12026 +       if (unlikely(err == -EWOULDBLOCK)) {
12027 +               pr_warn("cannot retry for NFSv4 delegation"
12028 +                       " for an internal unlink\n");
12029 +               iput(delegated);
12030 +       }
12031 +       if (unlikely(err)) {
12032 +               pr_err("unlink %pd after coo failed (%d), ignored\n",
12033 +                      dentry, err);
12034 +               err = 0;
12035 +       }
12036 +       goto out_parent; /* success */
12037 +
12038 +out_dgrade:
12039 +       di_downgrade_lock(parent, AuLock_IR);
12040 +out_parent:
12041 +       di_read_unlock(parent, AuLock_IR);
12042 +       dput(parent);
12043 +out:
12044 +       AuTraceErr(err);
12045 +       return err;
12046 +}
12047 +
12048 +int au_do_open(struct file *file, int (*open)(struct file *file, int flags),
12049 +              struct au_fidir *fidir)
12050 +{
12051 +       int err;
12052 +       struct dentry *dentry;
12053 +       struct au_finfo *finfo;
12054 +
12055 +       err = au_finfo_init(file, fidir);
12056 +       if (unlikely(err))
12057 +               goto out;
12058 +
12059 +       dentry = file->f_dentry;
12060 +       di_write_lock_child(dentry);
12061 +       err = au_cmoo(dentry);
12062 +       di_downgrade_lock(dentry, AuLock_IR);
12063 +       if (!err)
12064 +               err = open(file, vfsub_file_flags(file));
12065 +       di_read_unlock(dentry, AuLock_IR);
12066 +
12067 +       finfo = au_fi(file);
12068 +       if (!err) {
12069 +               finfo->fi_file = file;
12070 +               au_sphl_add(&finfo->fi_hlist,
12071 +                           &au_sbi(file->f_dentry->d_sb)->si_files);
12072 +       }
12073 +       fi_write_unlock(file);
12074 +       if (unlikely(err)) {
12075 +               finfo->fi_hdir = NULL;
12076 +               au_finfo_fin(file);
12077 +       }
12078 +
12079 +out:
12080 +       return err;
12081 +}
12082 +
12083 +int au_reopen_nondir(struct file *file)
12084 +{
12085 +       int err;
12086 +       aufs_bindex_t bstart;
12087 +       struct dentry *dentry;
12088 +       struct file *h_file, *h_file_tmp;
12089 +
12090 +       dentry = file->f_dentry;
12091 +       bstart = au_dbstart(dentry);
12092 +       h_file_tmp = NULL;
12093 +       if (au_fbstart(file) == bstart) {
12094 +               h_file = au_hf_top(file);
12095 +               if (file->f_mode == h_file->f_mode)
12096 +                       return 0; /* success */
12097 +               h_file_tmp = h_file;
12098 +               get_file(h_file_tmp);
12099 +               au_set_h_fptr(file, bstart, NULL);
12100 +       }
12101 +       AuDebugOn(au_fi(file)->fi_hdir);
12102 +       /*
12103 +        * it can happen
12104 +        * file exists on both of rw and ro
12105 +        * open --> dbstart and fbstart are both 0
12106 +        * prepend a branch as rw, "rw" become ro
12107 +        * remove rw/file
12108 +        * delete the top branch, "rw" becomes rw again
12109 +        *      --> dbstart is 1, fbstart is still 0
12110 +        * write --> fbstart is 0 but dbstart is 1
12111 +        */
12112 +       /* AuDebugOn(au_fbstart(file) < bstart); */
12113 +
12114 +       h_file = au_h_open(dentry, bstart, vfsub_file_flags(file) & ~O_TRUNC,
12115 +                          file, /*force_wr*/0);
12116 +       err = PTR_ERR(h_file);
12117 +       if (IS_ERR(h_file)) {
12118 +               if (h_file_tmp) {
12119 +                       atomic_inc(&au_sbr(dentry->d_sb, bstart)->br_count);
12120 +                       au_set_h_fptr(file, bstart, h_file_tmp);
12121 +                       h_file_tmp = NULL;
12122 +               }
12123 +               goto out; /* todo: close all? */
12124 +       }
12125 +
12126 +       err = 0;
12127 +       au_set_fbstart(file, bstart);
12128 +       au_set_h_fptr(file, bstart, h_file);
12129 +       au_update_figen(file);
12130 +       /* todo: necessary? */
12131 +       /* file->f_ra = h_file->f_ra; */
12132 +
12133 +out:
12134 +       if (h_file_tmp)
12135 +               fput(h_file_tmp);
12136 +       return err;
12137 +}
12138 +
12139 +/* ---------------------------------------------------------------------- */
12140 +
12141 +static int au_reopen_wh(struct file *file, aufs_bindex_t btgt,
12142 +                       struct dentry *hi_wh)
12143 +{
12144 +       int err;
12145 +       aufs_bindex_t bstart;
12146 +       struct au_dinfo *dinfo;
12147 +       struct dentry *h_dentry;
12148 +       struct au_hdentry *hdp;
12149 +
12150 +       dinfo = au_di(file->f_dentry);
12151 +       AuRwMustWriteLock(&dinfo->di_rwsem);
12152 +
12153 +       bstart = dinfo->di_bstart;
12154 +       dinfo->di_bstart = btgt;
12155 +       hdp = dinfo->di_hdentry;
12156 +       h_dentry = hdp[0 + btgt].hd_dentry;
12157 +       hdp[0 + btgt].hd_dentry = hi_wh;
12158 +       err = au_reopen_nondir(file);
12159 +       hdp[0 + btgt].hd_dentry = h_dentry;
12160 +       dinfo->di_bstart = bstart;
12161 +
12162 +       return err;
12163 +}
12164 +
12165 +static int au_ready_to_write_wh(struct file *file, loff_t len,
12166 +                               aufs_bindex_t bcpup, struct au_pin *pin)
12167 +{
12168 +       int err;
12169 +       struct inode *inode, *h_inode;
12170 +       struct dentry *h_dentry, *hi_wh;
12171 +       struct au_cp_generic cpg = {
12172 +               .dentry = file->f_dentry,
12173 +               .bdst   = bcpup,
12174 +               .bsrc   = -1,
12175 +               .len    = len,
12176 +               .pin    = pin
12177 +       };
12178 +
12179 +       au_update_dbstart(cpg.dentry);
12180 +       inode = cpg.dentry->d_inode;
12181 +       h_inode = NULL;
12182 +       if (au_dbstart(cpg.dentry) <= bcpup
12183 +           && au_dbend(cpg.dentry) >= bcpup) {
12184 +               h_dentry = au_h_dptr(cpg.dentry, bcpup);
12185 +               if (h_dentry)
12186 +                       h_inode = h_dentry->d_inode;
12187 +       }
12188 +       hi_wh = au_hi_wh(inode, bcpup);
12189 +       if (!hi_wh && !h_inode)
12190 +               err = au_sio_cpup_wh(&cpg, file);
12191 +       else
12192 +               /* already copied-up after unlink */
12193 +               err = au_reopen_wh(file, bcpup, hi_wh);
12194 +
12195 +       if (!err
12196 +           && (inode->i_nlink > 1
12197 +               || (inode->i_state & I_LINKABLE))
12198 +           && au_opt_test(au_mntflags(cpg.dentry->d_sb), PLINK))
12199 +               au_plink_append(inode, bcpup, au_h_dptr(cpg.dentry, bcpup));
12200 +
12201 +       return err;
12202 +}
12203 +
12204 +/*
12205 + * prepare the @file for writing.
12206 + */
12207 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin)
12208 +{
12209 +       int err;
12210 +       aufs_bindex_t dbstart;
12211 +       struct dentry *parent;
12212 +       struct inode *inode;
12213 +       struct super_block *sb;
12214 +       struct file *h_file;
12215 +       struct au_cp_generic cpg = {
12216 +               .dentry = file->f_dentry,
12217 +               .bdst   = -1,
12218 +               .bsrc   = -1,
12219 +               .len    = len,
12220 +               .pin    = pin,
12221 +               .flags  = AuCpup_DTIME
12222 +       };
12223 +
12224 +       sb = cpg.dentry->d_sb;
12225 +       inode = cpg.dentry->d_inode;
12226 +       cpg.bsrc = au_fbstart(file);
12227 +       err = au_test_ro(sb, cpg.bsrc, inode);
12228 +       if (!err && (au_hf_top(file)->f_mode & FMODE_WRITE)) {
12229 +               err = au_pin(pin, cpg.dentry, cpg.bsrc, AuOpt_UDBA_NONE,
12230 +                            /*flags*/0);
12231 +               goto out;
12232 +       }
12233 +
12234 +       /* need to cpup or reopen */
12235 +       parent = dget_parent(cpg.dentry);
12236 +       di_write_lock_parent(parent);
12237 +       err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12238 +       cpg.bdst = err;
12239 +       if (unlikely(err < 0))
12240 +               goto out_dgrade;
12241 +       err = 0;
12242 +
12243 +       if (!d_unhashed(cpg.dentry) && !au_h_dptr(parent, cpg.bdst)) {
12244 +               err = au_cpup_dirs(cpg.dentry, cpg.bdst);
12245 +               if (unlikely(err))
12246 +                       goto out_dgrade;
12247 +       }
12248 +
12249 +       err = au_pin(pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12250 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12251 +       if (unlikely(err))
12252 +               goto out_dgrade;
12253 +
12254 +       dbstart = au_dbstart(cpg.dentry);
12255 +       if (dbstart <= cpg.bdst)
12256 +               cpg.bsrc = cpg.bdst;
12257 +
12258 +       if (dbstart <= cpg.bdst         /* just reopen */
12259 +           || !d_unhashed(cpg.dentry)  /* copyup and reopen */
12260 +               ) {
12261 +               h_file = au_h_open_pre(cpg.dentry, cpg.bsrc, /*force_wr*/0);
12262 +               if (IS_ERR(h_file))
12263 +                       err = PTR_ERR(h_file);
12264 +               else {
12265 +                       di_downgrade_lock(parent, AuLock_IR);
12266 +                       if (dbstart > cpg.bdst)
12267 +                               err = au_sio_cpup_simple(&cpg);
12268 +                       if (!err)
12269 +                               err = au_reopen_nondir(file);
12270 +                       au_h_open_post(cpg.dentry, cpg.bsrc, h_file);
12271 +               }
12272 +       } else {                        /* copyup as wh and reopen */
12273 +               /*
12274 +                * since writable hfsplus branch is not supported,
12275 +                * h_open_pre/post() are unnecessary.
12276 +                */
12277 +               err = au_ready_to_write_wh(file, len, cpg.bdst, pin);
12278 +               di_downgrade_lock(parent, AuLock_IR);
12279 +       }
12280 +
12281 +       if (!err) {
12282 +               au_pin_set_parent_lflag(pin, /*lflag*/0);
12283 +               goto out_dput; /* success */
12284 +       }
12285 +       au_unpin(pin);
12286 +       goto out_unlock;
12287 +
12288 +out_dgrade:
12289 +       di_downgrade_lock(parent, AuLock_IR);
12290 +out_unlock:
12291 +       di_read_unlock(parent, AuLock_IR);
12292 +out_dput:
12293 +       dput(parent);
12294 +out:
12295 +       return err;
12296 +}
12297 +
12298 +/* ---------------------------------------------------------------------- */
12299 +
12300 +int au_do_flush(struct file *file, fl_owner_t id,
12301 +               int (*flush)(struct file *file, fl_owner_t id))
12302 +{
12303 +       int err;
12304 +       struct super_block *sb;
12305 +       struct inode *inode;
12306 +
12307 +       inode = file_inode(file);
12308 +       sb = inode->i_sb;
12309 +       si_noflush_read_lock(sb);
12310 +       fi_read_lock(file);
12311 +       ii_read_lock_child(inode);
12312 +
12313 +       err = flush(file, id);
12314 +       au_cpup_attr_timesizes(inode);
12315 +
12316 +       ii_read_unlock(inode);
12317 +       fi_read_unlock(file);
12318 +       si_read_unlock(sb);
12319 +       return err;
12320 +}
12321 +
12322 +/* ---------------------------------------------------------------------- */
12323 +
12324 +static int au_file_refresh_by_inode(struct file *file, int *need_reopen)
12325 +{
12326 +       int err;
12327 +       struct au_pin pin;
12328 +       struct au_finfo *finfo;
12329 +       struct dentry *parent, *hi_wh;
12330 +       struct inode *inode;
12331 +       struct super_block *sb;
12332 +       struct au_cp_generic cpg = {
12333 +               .dentry = file->f_dentry,
12334 +               .bdst   = -1,
12335 +               .bsrc   = -1,
12336 +               .len    = -1,
12337 +               .pin    = &pin,
12338 +               .flags  = AuCpup_DTIME
12339 +       };
12340 +
12341 +       FiMustWriteLock(file);
12342 +
12343 +       err = 0;
12344 +       finfo = au_fi(file);
12345 +       sb = cpg.dentry->d_sb;
12346 +       inode = cpg.dentry->d_inode;
12347 +       cpg.bdst = au_ibstart(inode);
12348 +       if (cpg.bdst == finfo->fi_btop || IS_ROOT(cpg.dentry))
12349 +               goto out;
12350 +
12351 +       parent = dget_parent(cpg.dentry);
12352 +       if (au_test_ro(sb, cpg.bdst, inode)) {
12353 +               di_read_lock_parent(parent, !AuLock_IR);
12354 +               err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12355 +               cpg.bdst = err;
12356 +               di_read_unlock(parent, !AuLock_IR);
12357 +               if (unlikely(err < 0))
12358 +                       goto out_parent;
12359 +               err = 0;
12360 +       }
12361 +
12362 +       di_read_lock_parent(parent, AuLock_IR);
12363 +       hi_wh = au_hi_wh(inode, cpg.bdst);
12364 +       if (!S_ISDIR(inode->i_mode)
12365 +           && au_opt_test(au_mntflags(sb), PLINK)
12366 +           && au_plink_test(inode)
12367 +           && !d_unhashed(cpg.dentry)
12368 +           && cpg.bdst < au_dbstart(cpg.dentry)) {
12369 +               err = au_test_and_cpup_dirs(cpg.dentry, cpg.bdst);
12370 +               if (unlikely(err))
12371 +                       goto out_unlock;
12372 +
12373 +               /* always superio. */
12374 +               err = au_pin(&pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12375 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12376 +               if (!err) {
12377 +                       err = au_sio_cpup_simple(&cpg);
12378 +                       au_unpin(&pin);
12379 +               }
12380 +       } else if (hi_wh) {
12381 +               /* already copied-up after unlink */
12382 +               err = au_reopen_wh(file, cpg.bdst, hi_wh);
12383 +               *need_reopen = 0;
12384 +       }
12385 +
12386 +out_unlock:
12387 +       di_read_unlock(parent, AuLock_IR);
12388 +out_parent:
12389 +       dput(parent);
12390 +out:
12391 +       return err;
12392 +}
12393 +
12394 +static void au_do_refresh_dir(struct file *file)
12395 +{
12396 +       aufs_bindex_t bindex, bend, new_bindex, brid;
12397 +       struct au_hfile *p, tmp, *q;
12398 +       struct au_finfo *finfo;
12399 +       struct super_block *sb;
12400 +       struct au_fidir *fidir;
12401 +
12402 +       FiMustWriteLock(file);
12403 +
12404 +       sb = file->f_dentry->d_sb;
12405 +       finfo = au_fi(file);
12406 +       fidir = finfo->fi_hdir;
12407 +       AuDebugOn(!fidir);
12408 +       p = fidir->fd_hfile + finfo->fi_btop;
12409 +       brid = p->hf_br->br_id;
12410 +       bend = fidir->fd_bbot;
12411 +       for (bindex = finfo->fi_btop; bindex <= bend; bindex++, p++) {
12412 +               if (!p->hf_file)
12413 +                       continue;
12414 +
12415 +               new_bindex = au_br_index(sb, p->hf_br->br_id);
12416 +               if (new_bindex == bindex)
12417 +                       continue;
12418 +               if (new_bindex < 0) {
12419 +                       au_set_h_fptr(file, bindex, NULL);
12420 +                       continue;
12421 +               }
12422 +
12423 +               /* swap two lower inode, and loop again */
12424 +               q = fidir->fd_hfile + new_bindex;
12425 +               tmp = *q;
12426 +               *q = *p;
12427 +               *p = tmp;
12428 +               if (tmp.hf_file) {
12429 +                       bindex--;
12430 +                       p--;
12431 +               }
12432 +       }
12433 +
12434 +       p = fidir->fd_hfile;
12435 +       if (!au_test_mmapped(file) && !d_unlinked(file->f_dentry)) {
12436 +               bend = au_sbend(sb);
12437 +               for (finfo->fi_btop = 0; finfo->fi_btop <= bend;
12438 +                    finfo->fi_btop++, p++)
12439 +                       if (p->hf_file) {
12440 +                               if (file_inode(p->hf_file))
12441 +                                       break;
12442 +                               au_hfput(p, file);
12443 +                       }
12444 +       } else {
12445 +               bend = au_br_index(sb, brid);
12446 +               for (finfo->fi_btop = 0; finfo->fi_btop < bend;
12447 +                    finfo->fi_btop++, p++)
12448 +                       if (p->hf_file)
12449 +                               au_hfput(p, file);
12450 +               bend = au_sbend(sb);
12451 +       }
12452 +
12453 +       p = fidir->fd_hfile + bend;
12454 +       for (fidir->fd_bbot = bend; fidir->fd_bbot >= finfo->fi_btop;
12455 +            fidir->fd_bbot--, p--)
12456 +               if (p->hf_file) {
12457 +                       if (file_inode(p->hf_file))
12458 +                               break;
12459 +                       au_hfput(p, file);
12460 +               }
12461 +       AuDebugOn(fidir->fd_bbot < finfo->fi_btop);
12462 +}
12463 +
12464 +/*
12465 + * after branch manipulating, refresh the file.
12466 + */
12467 +static int refresh_file(struct file *file, int (*reopen)(struct file *file))
12468 +{
12469 +       int err, need_reopen;
12470 +       aufs_bindex_t bend, bindex;
12471 +       struct dentry *dentry;
12472 +       struct au_finfo *finfo;
12473 +       struct au_hfile *hfile;
12474 +
12475 +       dentry = file->f_dentry;
12476 +       finfo = au_fi(file);
12477 +       if (!finfo->fi_hdir) {
12478 +               hfile = &finfo->fi_htop;
12479 +               AuDebugOn(!hfile->hf_file);
12480 +               bindex = au_br_index(dentry->d_sb, hfile->hf_br->br_id);
12481 +               AuDebugOn(bindex < 0);
12482 +               if (bindex != finfo->fi_btop)
12483 +                       au_set_fbstart(file, bindex);
12484 +       } else {
12485 +               err = au_fidir_realloc(finfo, au_sbend(dentry->d_sb) + 1);
12486 +               if (unlikely(err))
12487 +                       goto out;
12488 +               au_do_refresh_dir(file);
12489 +       }
12490 +
12491 +       err = 0;
12492 +       need_reopen = 1;
12493 +       if (!au_test_mmapped(file))
12494 +               err = au_file_refresh_by_inode(file, &need_reopen);
12495 +       if (!err && need_reopen && !d_unlinked(dentry))
12496 +               err = reopen(file);
12497 +       if (!err) {
12498 +               au_update_figen(file);
12499 +               goto out; /* success */
12500 +       }
12501 +
12502 +       /* error, close all lower files */
12503 +       if (finfo->fi_hdir) {
12504 +               bend = au_fbend_dir(file);
12505 +               for (bindex = au_fbstart(file); bindex <= bend; bindex++)
12506 +                       au_set_h_fptr(file, bindex, NULL);
12507 +       }
12508 +
12509 +out:
12510 +       return err;
12511 +}
12512 +
12513 +/* common function to regular file and dir */
12514 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
12515 +                         int wlock)
12516 +{
12517 +       int err;
12518 +       unsigned int sigen, figen;
12519 +       aufs_bindex_t bstart;
12520 +       unsigned char pseudo_link;
12521 +       struct dentry *dentry;
12522 +       struct inode *inode;
12523 +
12524 +       err = 0;
12525 +       dentry = file->f_dentry;
12526 +       inode = dentry->d_inode;
12527 +       sigen = au_sigen(dentry->d_sb);
12528 +       fi_write_lock(file);
12529 +       figen = au_figen(file);
12530 +       di_write_lock_child(dentry);
12531 +       bstart = au_dbstart(dentry);
12532 +       pseudo_link = (bstart != au_ibstart(inode));
12533 +       if (sigen == figen && !pseudo_link && au_fbstart(file) == bstart) {
12534 +               if (!wlock) {
12535 +                       di_downgrade_lock(dentry, AuLock_IR);
12536 +                       fi_downgrade_lock(file);
12537 +               }
12538 +               goto out; /* success */
12539 +       }
12540 +
12541 +       AuDbg("sigen %d, figen %d\n", sigen, figen);
12542 +       if (au_digen_test(dentry, sigen)) {
12543 +               err = au_reval_dpath(dentry, sigen);
12544 +               AuDebugOn(!err && au_digen_test(dentry, sigen));
12545 +       }
12546 +
12547 +       if (!err)
12548 +               err = refresh_file(file, reopen);
12549 +       if (!err) {
12550 +               if (!wlock) {
12551 +                       di_downgrade_lock(dentry, AuLock_IR);
12552 +                       fi_downgrade_lock(file);
12553 +               }
12554 +       } else {
12555 +               di_write_unlock(dentry);
12556 +               fi_write_unlock(file);
12557 +       }
12558 +
12559 +out:
12560 +       return err;
12561 +}
12562 +
12563 +/* ---------------------------------------------------------------------- */
12564 +
12565 +/* cf. aufs_nopage() */
12566 +/* for madvise(2) */
12567 +static int aufs_readpage(struct file *file __maybe_unused, struct page *page)
12568 +{
12569 +       unlock_page(page);
12570 +       return 0;
12571 +}
12572 +
12573 +/* it will never be called, but necessary to support O_DIRECT */
12574 +static ssize_t aufs_direct_IO(int rw, struct kiocb *iocb,
12575 +                             struct iov_iter *iter, loff_t offset)
12576 +{ BUG(); return 0; }
12577 +
12578 +/*
12579 + * it will never be called, but madvise and fadvise behaves differently
12580 + * when get_xip_mem is defined
12581 + */
12582 +static int aufs_get_xip_mem(struct address_space *mapping, pgoff_t pgoff,
12583 +                           int create, void **kmem, unsigned long *pfn)
12584 +{ BUG(); return 0; }
12585 +
12586 +/* they will never be called. */
12587 +#ifdef CONFIG_AUFS_DEBUG
12588 +static int aufs_write_begin(struct file *file, struct address_space *mapping,
12589 +                           loff_t pos, unsigned len, unsigned flags,
12590 +                           struct page **pagep, void **fsdata)
12591 +{ AuUnsupport(); return 0; }
12592 +static int aufs_write_end(struct file *file, struct address_space *mapping,
12593 +                         loff_t pos, unsigned len, unsigned copied,
12594 +                         struct page *page, void *fsdata)
12595 +{ AuUnsupport(); return 0; }
12596 +static int aufs_writepage(struct page *page, struct writeback_control *wbc)
12597 +{ AuUnsupport(); return 0; }
12598 +
12599 +static int aufs_set_page_dirty(struct page *page)
12600 +{ AuUnsupport(); return 0; }
12601 +static void aufs_invalidatepage(struct page *page, unsigned int offset,
12602 +                               unsigned int length)
12603 +{ AuUnsupport(); }
12604 +static int aufs_releasepage(struct page *page, gfp_t gfp)
12605 +{ AuUnsupport(); return 0; }
12606 +static int aufs_migratepage(struct address_space *mapping, struct page *newpage,
12607 +                           struct page *page, enum migrate_mode mode)
12608 +{ AuUnsupport(); return 0; }
12609 +static int aufs_launder_page(struct page *page)
12610 +{ AuUnsupport(); return 0; }
12611 +static int aufs_is_partially_uptodate(struct page *page,
12612 +                                     unsigned long from,
12613 +                                     unsigned long count)
12614 +{ AuUnsupport(); return 0; }
12615 +static void aufs_is_dirty_writeback(struct page *page, bool *dirty,
12616 +                                   bool *writeback)
12617 +{ AuUnsupport(); }
12618 +static int aufs_error_remove_page(struct address_space *mapping,
12619 +                                 struct page *page)
12620 +{ AuUnsupport(); return 0; }
12621 +static int aufs_swap_activate(struct swap_info_struct *sis, struct file *file,
12622 +                             sector_t *span)
12623 +{ AuUnsupport(); return 0; }
12624 +static void aufs_swap_deactivate(struct file *file)
12625 +{ AuUnsupport(); }
12626 +#endif /* CONFIG_AUFS_DEBUG */
12627 +
12628 +const struct address_space_operations aufs_aop = {
12629 +       .readpage               = aufs_readpage,
12630 +       .direct_IO              = aufs_direct_IO,
12631 +       .get_xip_mem            = aufs_get_xip_mem,
12632 +#ifdef CONFIG_AUFS_DEBUG
12633 +       .writepage              = aufs_writepage,
12634 +       /* no writepages, because of writepage */
12635 +       .set_page_dirty         = aufs_set_page_dirty,
12636 +       /* no readpages, because of readpage */
12637 +       .write_begin            = aufs_write_begin,
12638 +       .write_end              = aufs_write_end,
12639 +       /* no bmap, no block device */
12640 +       .invalidatepage         = aufs_invalidatepage,
12641 +       .releasepage            = aufs_releasepage,
12642 +       .migratepage            = aufs_migratepage,
12643 +       .launder_page           = aufs_launder_page,
12644 +       .is_partially_uptodate  = aufs_is_partially_uptodate,
12645 +       .is_dirty_writeback     = aufs_is_dirty_writeback,
12646 +       .error_remove_page      = aufs_error_remove_page,
12647 +       .swap_activate          = aufs_swap_activate,
12648 +       .swap_deactivate        = aufs_swap_deactivate
12649 +#endif /* CONFIG_AUFS_DEBUG */
12650 +};
12651 diff -urN /usr/share/empty/fs/aufs/file.h linux/fs/aufs/file.h
12652 --- /usr/share/empty/fs/aufs/file.h     1970-01-01 01:00:00.000000000 +0100
12653 +++ linux/fs/aufs/file.h        2015-01-25 13:00:38.631047076 +0100
12654 @@ -0,0 +1,284 @@
12655 +/*
12656 + * Copyright (C) 2005-2014 Junjiro R. Okajima
12657 + *
12658 + * This program, aufs is free software; you can redistribute it and/or modify
12659 + * it under the terms of the GNU General Public License as published by
12660 + * the Free Software Foundation; either version 2 of the License, or
12661 + * (at your option) any later version.
12662 + *
12663 + * This program is distributed in the hope that it will be useful,
12664 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12665 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12666 + * GNU General Public License for more details.
12667 + *
12668 + * You should have received a copy of the GNU General Public License
12669 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12670 + */
12671 +
12672 +/*
12673 + * file operations
12674 + */
12675 +
12676 +#ifndef __AUFS_FILE_H__
12677 +#define __AUFS_FILE_H__
12678 +
12679 +#ifdef __KERNEL__
12680 +
12681 +#include <linux/file.h>
12682 +#include <linux/fs.h>
12683 +#include <linux/poll.h>
12684 +#include "rwsem.h"
12685 +
12686 +struct au_branch;
12687 +struct au_hfile {
12688 +       struct file             *hf_file;
12689 +       struct au_branch        *hf_br;
12690 +};
12691 +
12692 +struct au_vdir;
12693 +struct au_fidir {
12694 +       aufs_bindex_t           fd_bbot;
12695 +       aufs_bindex_t           fd_nent;
12696 +       struct au_vdir          *fd_vdir_cache;
12697 +       struct au_hfile         fd_hfile[];
12698 +};
12699 +
12700 +static inline int au_fidir_sz(int nent)
12701 +{
12702 +       AuDebugOn(nent < 0);
12703 +       return sizeof(struct au_fidir) + sizeof(struct au_hfile) * nent;
12704 +}
12705 +
12706 +struct au_finfo {
12707 +       atomic_t                fi_generation;
12708 +
12709 +       struct au_rwsem         fi_rwsem;
12710 +       aufs_bindex_t           fi_btop;
12711 +
12712 +       /* do not union them */
12713 +       struct {                                /* for non-dir */
12714 +               struct au_hfile                 fi_htop;
12715 +               atomic_t                        fi_mmapped;
12716 +       };
12717 +       struct au_fidir         *fi_hdir;       /* for dir only */
12718 +
12719 +       struct hlist_node       fi_hlist;
12720 +       struct file             *fi_file;       /* very ugly */
12721 +} ____cacheline_aligned_in_smp;
12722 +
12723 +/* ---------------------------------------------------------------------- */
12724 +
12725 +/* file.c */
12726 +extern const struct address_space_operations aufs_aop;
12727 +unsigned int au_file_roflags(unsigned int flags);
12728 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
12729 +                      struct file *file, int force_wr);
12730 +int au_do_open(struct file *file, int (*open)(struct file *file, int flags),
12731 +              struct au_fidir *fidir);
12732 +int au_reopen_nondir(struct file *file);
12733 +struct au_pin;
12734 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin);
12735 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
12736 +                         int wlock);
12737 +int au_do_flush(struct file *file, fl_owner_t id,
12738 +               int (*flush)(struct file *file, fl_owner_t id));
12739 +
12740 +/* poll.c */
12741 +#ifdef CONFIG_AUFS_POLL
12742 +unsigned int aufs_poll(struct file *file, poll_table *wait);
12743 +#endif
12744 +
12745 +#ifdef CONFIG_AUFS_BR_HFSPLUS
12746 +/* hfsplus.c */
12747 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
12748 +                          int force_wr);
12749 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
12750 +                   struct file *h_file);
12751 +#else
12752 +AuStub(struct file *, au_h_open_pre, return NULL, struct dentry *dentry,
12753 +       aufs_bindex_t bindex, int force_wr)
12754 +AuStubVoid(au_h_open_post, struct dentry *dentry, aufs_bindex_t bindex,
12755 +          struct file *h_file);
12756 +#endif
12757 +
12758 +/* f_op.c */
12759 +extern const struct file_operations aufs_file_fop;
12760 +int au_do_open_nondir(struct file *file, int flags);
12761 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file);
12762 +
12763 +/* finfo.c */
12764 +void au_hfput(struct au_hfile *hf, struct file *file);
12765 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex,
12766 +                  struct file *h_file);
12767 +
12768 +void au_update_figen(struct file *file);
12769 +struct au_fidir *au_fidir_alloc(struct super_block *sb);
12770 +int au_fidir_realloc(struct au_finfo *finfo, int nbr);
12771 +
12772 +void au_fi_init_once(void *_fi);
12773 +void au_finfo_fin(struct file *file);
12774 +int au_finfo_init(struct file *file, struct au_fidir *fidir);
12775 +
12776 +/* ioctl.c */
12777 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg);
12778 +#ifdef CONFIG_COMPAT
12779 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
12780 +                          unsigned long arg);
12781 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
12782 +                             unsigned long arg);
12783 +#endif
12784 +
12785 +/* ---------------------------------------------------------------------- */
12786 +
12787 +static inline struct au_finfo *au_fi(struct file *file)
12788 +{
12789 +       return file->private_data;
12790 +}
12791 +
12792 +/* ---------------------------------------------------------------------- */
12793 +
12794 +/*
12795 + * fi_read_lock, fi_write_lock,
12796 + * fi_read_unlock, fi_write_unlock, fi_downgrade_lock
12797 + */
12798 +AuSimpleRwsemFuncs(fi, struct file *f, &au_fi(f)->fi_rwsem);
12799 +
12800 +#define FiMustNoWaiters(f)     AuRwMustNoWaiters(&au_fi(f)->fi_rwsem)
12801 +#define FiMustAnyLock(f)       AuRwMustAnyLock(&au_fi(f)->fi_rwsem)
12802 +#define FiMustWriteLock(f)     AuRwMustWriteLock(&au_fi(f)->fi_rwsem)
12803 +
12804 +/* ---------------------------------------------------------------------- */
12805 +
12806 +/* todo: hard/soft set? */
12807 +static inline aufs_bindex_t au_fbstart(struct file *file)
12808 +{
12809 +       FiMustAnyLock(file);
12810 +       return au_fi(file)->fi_btop;
12811 +}
12812 +
12813 +static inline aufs_bindex_t au_fbend_dir(struct file *file)
12814 +{
12815 +       FiMustAnyLock(file);
12816 +       AuDebugOn(!au_fi(file)->fi_hdir);
12817 +       return au_fi(file)->fi_hdir->fd_bbot;
12818 +}
12819 +
12820 +static inline struct au_vdir *au_fvdir_cache(struct file *file)
12821 +{
12822 +       FiMustAnyLock(file);
12823 +       AuDebugOn(!au_fi(file)->fi_hdir);
12824 +       return au_fi(file)->fi_hdir->fd_vdir_cache;
12825 +}
12826 +
12827 +static inline void au_set_fbstart(struct file *file, aufs_bindex_t bindex)
12828 +{
12829 +       FiMustWriteLock(file);
12830 +       au_fi(file)->fi_btop = bindex;
12831 +}
12832 +
12833 +static inline void au_set_fbend_dir(struct file *file, aufs_bindex_t bindex)
12834 +{
12835 +       FiMustWriteLock(file);
12836 +       AuDebugOn(!au_fi(file)->fi_hdir);
12837 +       au_fi(file)->fi_hdir->fd_bbot = bindex;
12838 +}
12839 +
12840 +static inline void au_set_fvdir_cache(struct file *file,
12841 +                                     struct au_vdir *vdir_cache)
12842 +{
12843 +       FiMustWriteLock(file);
12844 +       AuDebugOn(!au_fi(file)->fi_hdir);
12845 +       au_fi(file)->fi_hdir->fd_vdir_cache = vdir_cache;
12846 +}
12847 +
12848 +static inline struct file *au_hf_top(struct file *file)
12849 +{
12850 +       FiMustAnyLock(file);
12851 +       AuDebugOn(au_fi(file)->fi_hdir);
12852 +       return au_fi(file)->fi_htop.hf_file;
12853 +}
12854 +
12855 +static inline struct file *au_hf_dir(struct file *file, aufs_bindex_t bindex)
12856 +{
12857 +       FiMustAnyLock(file);
12858 +       AuDebugOn(!au_fi(file)->fi_hdir);
12859 +       return au_fi(file)->fi_hdir->fd_hfile[0 + bindex].hf_file;
12860 +}
12861 +
12862 +/* todo: memory barrier? */
12863 +static inline unsigned int au_figen(struct file *f)
12864 +{
12865 +       return atomic_read(&au_fi(f)->fi_generation);
12866 +}
12867 +
12868 +static inline void au_set_mmapped(struct file *f)
12869 +{
12870 +       if (atomic_inc_return(&au_fi(f)->fi_mmapped))
12871 +               return;
12872 +       pr_warn("fi_mmapped wrapped around\n");
12873 +       while (!atomic_inc_return(&au_fi(f)->fi_mmapped))
12874 +               ;
12875 +}
12876 +
12877 +static inline void au_unset_mmapped(struct file *f)
12878 +{
12879 +       atomic_dec(&au_fi(f)->fi_mmapped);
12880 +}
12881 +
12882 +static inline int au_test_mmapped(struct file *f)
12883 +{
12884 +       return atomic_read(&au_fi(f)->fi_mmapped);
12885 +}
12886 +
12887 +/* customize vma->vm_file */
12888 +
12889 +static inline void au_do_vm_file_reset(struct vm_area_struct *vma,
12890 +                                      struct file *file)
12891 +{
12892 +       struct file *f;
12893 +
12894 +       f = vma->vm_file;
12895 +       get_file(file);
12896 +       vma->vm_file = file;
12897 +       fput(f);
12898 +}
12899 +
12900 +#ifdef CONFIG_MMU
12901 +#define AuDbgVmRegion(file, vma) do {} while (0)
12902 +
12903 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
12904 +                                   struct file *file)
12905 +{
12906 +       au_do_vm_file_reset(vma, file);
12907 +}
12908 +#else
12909 +#define AuDbgVmRegion(file, vma) \
12910 +       AuDebugOn((vma)->vm_region && (vma)->vm_region->vm_file != (file))
12911 +
12912 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
12913 +                                   struct file *file)
12914 +{
12915 +       struct file *f;
12916 +
12917 +       au_do_vm_file_reset(vma, file);
12918 +       f = vma->vm_region->vm_file;
12919 +       get_file(file);
12920 +       vma->vm_region->vm_file = file;
12921 +       fput(f);
12922 +}
12923 +#endif /* CONFIG_MMU */
12924 +
12925 +/* handle vma->vm_prfile */
12926 +static inline void au_vm_prfile_set(struct vm_area_struct *vma,
12927 +                                   struct file *file)
12928 +{
12929 +       get_file(file);
12930 +       vma->vm_prfile = file;
12931 +#ifndef CONFIG_MMU
12932 +       get_file(file);
12933 +       vma->vm_region->vm_prfile = file;
12934 +#endif
12935 +}
12936 +
12937 +#endif /* __KERNEL__ */
12938 +#endif /* __AUFS_FILE_H__ */
12939 diff -urN /usr/share/empty/fs/aufs/finfo.c linux/fs/aufs/finfo.c
12940 --- /usr/share/empty/fs/aufs/finfo.c    1970-01-01 01:00:00.000000000 +0100
12941 +++ linux/fs/aufs/finfo.c       2015-01-25 13:00:38.631047076 +0100
12942 @@ -0,0 +1,156 @@
12943 +/*
12944 + * Copyright (C) 2005-2014 Junjiro R. Okajima
12945 + *
12946 + * This program, aufs is free software; you can redistribute it and/or modify
12947 + * it under the terms of the GNU General Public License as published by
12948 + * the Free Software Foundation; either version 2 of the License, or
12949 + * (at your option) any later version.
12950 + *
12951 + * This program is distributed in the hope that it will be useful,
12952 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12953 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12954 + * GNU General Public License for more details.
12955 + *
12956 + * You should have received a copy of the GNU General Public License
12957 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12958 + */
12959 +
12960 +/*
12961 + * file private data
12962 + */
12963 +
12964 +#include "aufs.h"
12965 +
12966 +void au_hfput(struct au_hfile *hf, struct file *file)
12967 +{
12968 +       /* todo: direct access f_flags */
12969 +       if (vfsub_file_flags(file) & __FMODE_EXEC)
12970 +               allow_write_access(hf->hf_file);
12971 +       fput(hf->hf_file);
12972 +       hf->hf_file = NULL;
12973 +       atomic_dec(&hf->hf_br->br_count);
12974 +       hf->hf_br = NULL;
12975 +}
12976 +
12977 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex, struct file *val)
12978 +{
12979 +       struct au_finfo *finfo = au_fi(file);
12980 +       struct au_hfile *hf;
12981 +       struct au_fidir *fidir;
12982 +
12983 +       fidir = finfo->fi_hdir;
12984 +       if (!fidir) {
12985 +               AuDebugOn(finfo->fi_btop != bindex);
12986 +               hf = &finfo->fi_htop;
12987 +       } else
12988 +               hf = fidir->fd_hfile + bindex;
12989 +
12990 +       if (hf && hf->hf_file)
12991 +               au_hfput(hf, file);
12992 +       if (val) {
12993 +               FiMustWriteLock(file);
12994 +               hf->hf_file = val;
12995 +               hf->hf_br = au_sbr(file->f_dentry->d_sb, bindex);
12996 +       }
12997 +}
12998 +
12999 +void au_update_figen(struct file *file)
13000 +{
13001 +       atomic_set(&au_fi(file)->fi_generation, au_digen(file->f_dentry));
13002 +       /* smp_mb(); */ /* atomic_set */
13003 +}
13004 +
13005 +/* ---------------------------------------------------------------------- */
13006 +
13007 +struct au_fidir *au_fidir_alloc(struct super_block *sb)
13008 +{
13009 +       struct au_fidir *fidir;
13010 +       int nbr;
13011 +
13012 +       nbr = au_sbend(sb) + 1;
13013 +       if (nbr < 2)
13014 +               nbr = 2; /* initial allocate for 2 branches */
13015 +       fidir = kzalloc(au_fidir_sz(nbr), GFP_NOFS);
13016 +       if (fidir) {
13017 +               fidir->fd_bbot = -1;
13018 +               fidir->fd_nent = nbr;
13019 +               fidir->fd_vdir_cache = NULL;
13020 +       }
13021 +
13022 +       return fidir;
13023 +}
13024 +
13025 +int au_fidir_realloc(struct au_finfo *finfo, int nbr)
13026 +{
13027 +       int err;
13028 +       struct au_fidir *fidir, *p;
13029 +
13030 +       AuRwMustWriteLock(&finfo->fi_rwsem);
13031 +       fidir = finfo->fi_hdir;
13032 +       AuDebugOn(!fidir);
13033 +
13034 +       err = -ENOMEM;
13035 +       p = au_kzrealloc(fidir, au_fidir_sz(fidir->fd_nent), au_fidir_sz(nbr),
13036 +                        GFP_NOFS);
13037 +       if (p) {
13038 +               p->fd_nent = nbr;
13039 +               finfo->fi_hdir = p;
13040 +               err = 0;
13041 +       }
13042 +
13043 +       return err;
13044 +}
13045 +
13046 +/* ---------------------------------------------------------------------- */
13047 +
13048 +void au_finfo_fin(struct file *file)
13049 +{
13050 +       struct au_finfo *finfo;
13051 +
13052 +       au_nfiles_dec(file->f_dentry->d_sb);
13053 +
13054 +       finfo = au_fi(file);
13055 +       AuDebugOn(finfo->fi_hdir);
13056 +       AuRwDestroy(&finfo->fi_rwsem);
13057 +       au_cache_free_finfo(finfo);
13058 +}
13059 +
13060 +void au_fi_init_once(void *_finfo)
13061 +{
13062 +       struct au_finfo *finfo = _finfo;
13063 +       static struct lock_class_key aufs_fi;
13064 +
13065 +       au_rw_init(&finfo->fi_rwsem);
13066 +       au_rw_class(&finfo->fi_rwsem, &aufs_fi);
13067 +}
13068 +
13069 +int au_finfo_init(struct file *file, struct au_fidir *fidir)
13070 +{
13071 +       int err;
13072 +       struct au_finfo *finfo;
13073 +       struct dentry *dentry;
13074 +
13075 +       err = -ENOMEM;
13076 +       dentry = file->f_dentry;
13077 +       finfo = au_cache_alloc_finfo();
13078 +       if (unlikely(!finfo))
13079 +               goto out;
13080 +
13081 +       err = 0;
13082 +       au_nfiles_inc(dentry->d_sb);
13083 +       /* verbose coding for lock class name */
13084 +       if (!fidir)
13085 +               au_rw_class(&finfo->fi_rwsem, au_lc_key + AuLcNonDir_FIINFO);
13086 +       else
13087 +               au_rw_class(&finfo->fi_rwsem, au_lc_key + AuLcDir_FIINFO);
13088 +       au_rw_write_lock(&finfo->fi_rwsem);
13089 +       finfo->fi_btop = -1;
13090 +       finfo->fi_hdir = fidir;
13091 +       atomic_set(&finfo->fi_generation, au_digen(dentry));
13092 +       /* smp_mb(); */ /* atomic_set */
13093 +
13094 +       file->private_data = finfo;
13095 +
13096 +out:
13097 +       return err;
13098 +}
13099 diff -urN /usr/share/empty/fs/aufs/f_op.c linux/fs/aufs/f_op.c
13100 --- /usr/share/empty/fs/aufs/f_op.c     1970-01-01 01:00:00.000000000 +0100
13101 +++ linux/fs/aufs/f_op.c        2015-01-25 13:00:38.631047076 +0100
13102 @@ -0,0 +1,813 @@
13103 +/*
13104 + * Copyright (C) 2005-2014 Junjiro R. Okajima
13105 + *
13106 + * This program, aufs is free software; you can redistribute it and/or modify
13107 + * it under the terms of the GNU General Public License as published by
13108 + * the Free Software Foundation; either version 2 of the License, or
13109 + * (at your option) any later version.
13110 + *
13111 + * This program is distributed in the hope that it will be useful,
13112 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13113 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13114 + * GNU General Public License for more details.
13115 + *
13116 + * You should have received a copy of the GNU General Public License
13117 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13118 + */
13119 +
13120 +/*
13121 + * file and vm operations
13122 + */
13123 +
13124 +#include <linux/aio.h>
13125 +#include <linux/fs_stack.h>
13126 +#include <linux/mman.h>
13127 +#include <linux/security.h>
13128 +#include "aufs.h"
13129 +
13130 +int au_do_open_nondir(struct file *file, int flags)
13131 +{
13132 +       int err;
13133 +       aufs_bindex_t bindex;
13134 +       struct file *h_file;
13135 +       struct dentry *dentry;
13136 +       struct au_finfo *finfo;
13137 +       struct inode *h_inode;
13138 +
13139 +       FiMustWriteLock(file);
13140 +
13141 +       err = 0;
13142 +       dentry = file->f_dentry;
13143 +       finfo = au_fi(file);
13144 +       memset(&finfo->fi_htop, 0, sizeof(finfo->fi_htop));
13145 +       atomic_set(&finfo->fi_mmapped, 0);
13146 +       bindex = au_dbstart(dentry);
13147 +       h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
13148 +       if (IS_ERR(h_file))
13149 +               err = PTR_ERR(h_file);
13150 +       else {
13151 +               if ((flags & __O_TMPFILE)
13152 +                   && !(flags & O_EXCL)) {
13153 +                       h_inode = file_inode(h_file);
13154 +                       spin_lock(&h_inode->i_lock);
13155 +                       h_inode->i_state |= I_LINKABLE;
13156 +                       spin_unlock(&h_inode->i_lock);
13157 +               }
13158 +               au_set_fbstart(file, bindex);
13159 +               au_set_h_fptr(file, bindex, h_file);
13160 +               au_update_figen(file);
13161 +               /* todo: necessary? */
13162 +               /* file->f_ra = h_file->f_ra; */
13163 +       }
13164 +
13165 +       return err;
13166 +}
13167 +
13168 +static int aufs_open_nondir(struct inode *inode __maybe_unused,
13169 +                           struct file *file)
13170 +{
13171 +       int err;
13172 +       struct super_block *sb;
13173 +
13174 +       AuDbg("%pD, f_flags 0x%x, f_mode 0x%x\n",
13175 +             file, vfsub_file_flags(file), file->f_mode);
13176 +
13177 +       sb = file->f_dentry->d_sb;
13178 +       si_read_lock(sb, AuLock_FLUSH);
13179 +       err = au_do_open(file, au_do_open_nondir, /*fidir*/NULL);
13180 +       si_read_unlock(sb);
13181 +       return err;
13182 +}
13183 +
13184 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file)
13185 +{
13186 +       struct au_finfo *finfo;
13187 +       aufs_bindex_t bindex;
13188 +
13189 +       finfo = au_fi(file);
13190 +       au_sphl_del(&finfo->fi_hlist, &au_sbi(file->f_dentry->d_sb)->si_files);
13191 +       bindex = finfo->fi_btop;
13192 +       if (bindex >= 0)
13193 +               au_set_h_fptr(file, bindex, NULL);
13194 +
13195 +       au_finfo_fin(file);
13196 +       return 0;
13197 +}
13198 +
13199 +/* ---------------------------------------------------------------------- */
13200 +
13201 +static int au_do_flush_nondir(struct file *file, fl_owner_t id)
13202 +{
13203 +       int err;
13204 +       struct file *h_file;
13205 +
13206 +       err = 0;
13207 +       h_file = au_hf_top(file);
13208 +       if (h_file)
13209 +               err = vfsub_flush(h_file, id);
13210 +       return err;
13211 +}
13212 +
13213 +static int aufs_flush_nondir(struct file *file, fl_owner_t id)
13214 +{
13215 +       return au_do_flush(file, id, au_do_flush_nondir);
13216 +}
13217 +
13218 +/* ---------------------------------------------------------------------- */
13219 +/*
13220 + * read and write functions acquire [fdi]_rwsem once, but release before
13221 + * mmap_sem. This is because to stop a race condition between mmap(2).
13222 + * Releasing these aufs-rwsem should be safe, no branch-mamagement (by keeping
13223 + * si_rwsem), no harmful copy-up should happen. Actually copy-up may happen in
13224 + * read functions after [fdi]_rwsem are released, but it should be harmless.
13225 + */
13226 +
13227 +static ssize_t aufs_read(struct file *file, char __user *buf, size_t count,
13228 +                        loff_t *ppos)
13229 +{
13230 +       ssize_t err;
13231 +       struct dentry *dentry;
13232 +       struct file *h_file;
13233 +       struct super_block *sb;
13234 +
13235 +       dentry = file->f_dentry;
13236 +       sb = dentry->d_sb;
13237 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13238 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13239 +       if (unlikely(err))
13240 +               goto out;
13241 +
13242 +       h_file = au_hf_top(file);
13243 +       get_file(h_file);
13244 +       di_read_unlock(dentry, AuLock_IR);
13245 +       fi_read_unlock(file);
13246 +
13247 +       /* filedata may be obsoleted by concurrent copyup, but no problem */
13248 +       err = vfsub_read_u(h_file, buf, count, ppos);
13249 +       /* todo: necessary? */
13250 +       /* file->f_ra = h_file->f_ra; */
13251 +       /* update without lock, I don't think it a problem */
13252 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
13253 +       fput(h_file);
13254 +
13255 +out:
13256 +       si_read_unlock(sb);
13257 +       return err;
13258 +}
13259 +
13260 +/*
13261 + * todo: very ugly
13262 + * it locks both of i_mutex and si_rwsem for read in safe.
13263 + * if the plink maintenance mode continues forever (that is the problem),
13264 + * may loop forever.
13265 + */
13266 +static void au_mtx_and_read_lock(struct inode *inode)
13267 +{
13268 +       int err;
13269 +       struct super_block *sb = inode->i_sb;
13270 +
13271 +       while (1) {
13272 +               mutex_lock(&inode->i_mutex);
13273 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
13274 +               if (!err)
13275 +                       break;
13276 +               mutex_unlock(&inode->i_mutex);
13277 +               si_read_lock(sb, AuLock_NOPLMW);
13278 +               si_read_unlock(sb);
13279 +       }
13280 +}
13281 +
13282 +static ssize_t aufs_write(struct file *file, const char __user *ubuf,
13283 +                         size_t count, loff_t *ppos)
13284 +{
13285 +       ssize_t err;
13286 +       blkcnt_t blks;
13287 +       aufs_bindex_t bstart;
13288 +       struct au_pin pin;
13289 +       struct dentry *dentry;
13290 +       struct inode *inode, *h_inode;
13291 +       struct super_block *sb;
13292 +       struct file *h_file;
13293 +       char __user *buf = (char __user *)ubuf;
13294 +
13295 +       dentry = file->f_dentry;
13296 +       sb = dentry->d_sb;
13297 +       inode = dentry->d_inode;
13298 +       au_mtx_and_read_lock(inode);
13299 +
13300 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13301 +       if (unlikely(err))
13302 +               goto out;
13303 +
13304 +       err = au_ready_to_write(file, -1, &pin);
13305 +       di_downgrade_lock(dentry, AuLock_IR);
13306 +       if (unlikely(err)) {
13307 +               di_read_unlock(dentry, AuLock_IR);
13308 +               fi_write_unlock(file);
13309 +               goto out;
13310 +       }
13311 +
13312 +       bstart = au_fbstart(file);
13313 +       h_file = au_hf_top(file);
13314 +       get_file(h_file);
13315 +       h_inode = file_inode(h_file);
13316 +       blks = h_inode->i_blocks;
13317 +       au_unpin(&pin);
13318 +       di_read_unlock(dentry, AuLock_IR);
13319 +       fi_write_unlock(file);
13320 +
13321 +       err = vfsub_write_u(h_file, buf, count, ppos);
13322 +       ii_write_lock_child(inode);
13323 +       au_cpup_attr_timesizes(inode);
13324 +       inode->i_mode = file_inode(h_file)->i_mode;
13325 +       AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks);
13326 +       if (err > 0)
13327 +               au_fhsm_wrote(sb, bstart, /*force*/h_inode->i_blocks > blks);
13328 +       ii_write_unlock(inode);
13329 +       fput(h_file);
13330 +
13331 +out:
13332 +       si_read_unlock(sb);
13333 +       mutex_unlock(&inode->i_mutex);
13334 +       return err;
13335 +}
13336 +
13337 +static ssize_t au_do_iter(struct file *h_file, int rw, struct kiocb *kio,
13338 +                         struct iov_iter *iov_iter)
13339 +{
13340 +       ssize_t err;
13341 +       struct file *file;
13342 +       ssize_t (*iter)(struct kiocb *, struct iov_iter *);
13343 +       ssize_t (*aio)(struct kiocb *, const struct iovec *, unsigned long,
13344 +                      loff_t);
13345 +
13346 +       err = security_file_permission(h_file, rw);
13347 +       if (unlikely(err))
13348 +               goto out;
13349 +
13350 +       err = -ENOSYS;
13351 +       iter = NULL;
13352 +       aio = NULL;
13353 +       if (rw == MAY_READ) {
13354 +               iter = h_file->f_op->read_iter;
13355 +               aio = h_file->f_op->aio_read;
13356 +       } else if (rw == MAY_WRITE) {
13357 +               iter = h_file->f_op->write_iter;
13358 +               aio = h_file->f_op->aio_write;
13359 +       }
13360 +
13361 +       file = kio->ki_filp;
13362 +       kio->ki_filp = h_file;
13363 +       if (iter) {
13364 +               lockdep_off();
13365 +               err = iter(kio, iov_iter);
13366 +               lockdep_on();
13367 +       } else if (aio) {
13368 +               lockdep_off();
13369 +               err = aio(kio, iov_iter->iov, iov_iter->nr_segs, kio->ki_pos);
13370 +               lockdep_on();
13371 +       } else
13372 +               /* currently there is no such fs */
13373 +               WARN_ON_ONCE(1);
13374 +       kio->ki_filp = file;
13375 +
13376 +out:
13377 +       return err;
13378 +}
13379 +
13380 +static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
13381 +{
13382 +       ssize_t err;
13383 +       struct file *file, *h_file;
13384 +       struct dentry *dentry;
13385 +       struct super_block *sb;
13386 +
13387 +       file = kio->ki_filp;
13388 +       dentry = file->f_dentry;
13389 +       sb = dentry->d_sb;
13390 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13391 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13392 +       if (unlikely(err))
13393 +               goto out;
13394 +
13395 +       h_file = au_hf_top(file);
13396 +       get_file(h_file);
13397 +       di_read_unlock(dentry, AuLock_IR);
13398 +       fi_read_unlock(file);
13399 +
13400 +       err = au_do_iter(h_file, MAY_READ, kio, iov_iter);
13401 +       /* todo: necessary? */
13402 +       /* file->f_ra = h_file->f_ra; */
13403 +       /* update without lock, I don't think it a problem */
13404 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
13405 +       fput(h_file);
13406 +
13407 +out:
13408 +       si_read_unlock(sb);
13409 +       return err;
13410 +}
13411 +
13412 +static ssize_t aufs_write_iter(struct kiocb *kio, struct iov_iter *iov_iter)
13413 +{
13414 +       ssize_t err;
13415 +       blkcnt_t blks;
13416 +       aufs_bindex_t bstart;
13417 +       struct au_pin pin;
13418 +       struct dentry *dentry;
13419 +       struct inode *inode, *h_inode;
13420 +       struct file *file, *h_file;
13421 +       struct super_block *sb;
13422 +
13423 +       file = kio->ki_filp;
13424 +       dentry = file->f_dentry;
13425 +       sb = dentry->d_sb;
13426 +       inode = dentry->d_inode;
13427 +       au_mtx_and_read_lock(inode);
13428 +
13429 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13430 +       if (unlikely(err))
13431 +               goto out;
13432 +
13433 +       err = au_ready_to_write(file, -1, &pin);
13434 +       di_downgrade_lock(dentry, AuLock_IR);
13435 +       if (unlikely(err)) {
13436 +               di_read_unlock(dentry, AuLock_IR);
13437 +               fi_write_unlock(file);
13438 +               goto out;
13439 +       }
13440 +
13441 +       bstart = au_fbstart(file);
13442 +       h_file = au_hf_top(file);
13443 +       get_file(h_file);
13444 +       h_inode = file_inode(h_file);
13445 +       blks = h_inode->i_blocks;
13446 +       au_unpin(&pin);
13447 +       di_read_unlock(dentry, AuLock_IR);
13448 +       fi_write_unlock(file);
13449 +
13450 +       err = au_do_iter(h_file, MAY_WRITE, kio, iov_iter);
13451 +       ii_write_lock_child(inode);
13452 +       au_cpup_attr_timesizes(inode);
13453 +       inode->i_mode = file_inode(h_file)->i_mode;
13454 +       AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks);
13455 +       if (err > 0)
13456 +               au_fhsm_wrote(sb, bstart, /*force*/h_inode->i_blocks > blks);
13457 +       ii_write_unlock(inode);
13458 +       fput(h_file);
13459 +
13460 +out:
13461 +       si_read_unlock(sb);
13462 +       mutex_unlock(&inode->i_mutex);
13463 +       return err;
13464 +}
13465 +
13466 +static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
13467 +                               struct pipe_inode_info *pipe, size_t len,
13468 +                               unsigned int flags)
13469 +{
13470 +       ssize_t err;
13471 +       struct file *h_file;
13472 +       struct dentry *dentry;
13473 +       struct super_block *sb;
13474 +
13475 +       dentry = file->f_dentry;
13476 +       sb = dentry->d_sb;
13477 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13478 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13479 +       if (unlikely(err))
13480 +               goto out;
13481 +
13482 +       err = -EINVAL;
13483 +       h_file = au_hf_top(file);
13484 +       get_file(h_file);
13485 +       if (au_test_loopback_kthread()) {
13486 +               au_warn_loopback(h_file->f_dentry->d_sb);
13487 +               if (file->f_mapping != h_file->f_mapping) {
13488 +                       file->f_mapping = h_file->f_mapping;
13489 +                       smp_mb(); /* unnecessary? */
13490 +               }
13491 +       }
13492 +       di_read_unlock(dentry, AuLock_IR);
13493 +       fi_read_unlock(file);
13494 +
13495 +       err = vfsub_splice_to(h_file, ppos, pipe, len, flags);
13496 +       /* todo: necessasry? */
13497 +       /* file->f_ra = h_file->f_ra; */
13498 +       /* update without lock, I don't think it a problem */
13499 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
13500 +       fput(h_file);
13501 +
13502 +out:
13503 +       si_read_unlock(sb);
13504 +       return err;
13505 +}
13506 +
13507 +static ssize_t
13508 +aufs_splice_write(struct pipe_inode_info *pipe, struct file *file, loff_t *ppos,
13509 +                 size_t len, unsigned int flags)
13510 +{
13511 +       ssize_t err;
13512 +       blkcnt_t blks;
13513 +       aufs_bindex_t bstart;
13514 +       struct au_pin pin;
13515 +       struct dentry *dentry;
13516 +       struct inode *inode, *h_inode;
13517 +       struct super_block *sb;
13518 +       struct file *h_file;
13519 +
13520 +       dentry = file->f_dentry;
13521 +       sb = dentry->d_sb;
13522 +       inode = dentry->d_inode;
13523 +       au_mtx_and_read_lock(inode);
13524 +
13525 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13526 +       if (unlikely(err))
13527 +               goto out;
13528 +
13529 +       err = au_ready_to_write(file, -1, &pin);
13530 +       di_downgrade_lock(dentry, AuLock_IR);
13531 +       if (unlikely(err)) {
13532 +               di_read_unlock(dentry, AuLock_IR);
13533 +               fi_write_unlock(file);
13534 +               goto out;
13535 +       }
13536 +
13537 +       bstart = au_fbstart(file);
13538 +       h_file = au_hf_top(file);
13539 +       get_file(h_file);
13540 +       h_inode = file_inode(h_file);
13541 +       blks = h_inode->i_blocks;
13542 +       au_unpin(&pin);
13543 +       di_read_unlock(dentry, AuLock_IR);
13544 +       fi_write_unlock(file);
13545 +
13546 +       err = vfsub_splice_from(pipe, h_file, ppos, len, flags);
13547 +       ii_write_lock_child(inode);
13548 +       au_cpup_attr_timesizes(inode);
13549 +       inode->i_mode = file_inode(h_file)->i_mode;
13550 +       AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks);
13551 +       if (err > 0)
13552 +               au_fhsm_wrote(sb, bstart, /*force*/h_inode->i_blocks > blks);
13553 +       ii_write_unlock(inode);
13554 +       fput(h_file);
13555 +
13556 +out:
13557 +       si_read_unlock(sb);
13558 +       mutex_unlock(&inode->i_mutex);
13559 +       return err;
13560 +}
13561 +
13562 +static long aufs_fallocate(struct file *file, int mode, loff_t offset,
13563 +                          loff_t len)
13564 +{
13565 +       long err;
13566 +       struct au_pin pin;
13567 +       struct dentry *dentry;
13568 +       struct super_block *sb;
13569 +       struct inode *inode;
13570 +       struct file *h_file;
13571 +
13572 +       dentry = file->f_dentry;
13573 +       sb = dentry->d_sb;
13574 +       inode = dentry->d_inode;
13575 +       au_mtx_and_read_lock(inode);
13576 +
13577 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13578 +       if (unlikely(err))
13579 +               goto out;
13580 +
13581 +       err = au_ready_to_write(file, -1, &pin);
13582 +       di_downgrade_lock(dentry, AuLock_IR);
13583 +       if (unlikely(err)) {
13584 +               di_read_unlock(dentry, AuLock_IR);
13585 +               fi_write_unlock(file);
13586 +               goto out;
13587 +       }
13588 +
13589 +       h_file = au_hf_top(file);
13590 +       get_file(h_file);
13591 +       au_unpin(&pin);
13592 +       di_read_unlock(dentry, AuLock_IR);
13593 +       fi_write_unlock(file);
13594 +
13595 +       lockdep_off();
13596 +       err = vfs_fallocate(h_file, mode, offset, len);
13597 +       lockdep_on();
13598 +       ii_write_lock_child(inode);
13599 +       au_cpup_attr_timesizes(inode);
13600 +       inode->i_mode = file_inode(h_file)->i_mode;
13601 +       ii_write_unlock(inode);
13602 +       fput(h_file);
13603 +
13604 +out:
13605 +       si_read_unlock(sb);
13606 +       mutex_unlock(&inode->i_mutex);
13607 +       return err;
13608 +}
13609 +
13610 +/* ---------------------------------------------------------------------- */
13611 +
13612 +/*
13613 + * The locking order around current->mmap_sem.
13614 + * - in most and regular cases
13615 + *   file I/O syscall -- aufs_read() or something
13616 + *     -- si_rwsem for read -- mmap_sem
13617 + *     (Note that [fdi]i_rwsem are released before mmap_sem).
13618 + * - in mmap case
13619 + *   mmap(2) -- mmap_sem -- aufs_mmap() -- si_rwsem for read -- [fdi]i_rwsem
13620 + * This AB-BA order is definitly bad, but is not a problem since "si_rwsem for
13621 + * read" allows muliple processes to acquire it and [fdi]i_rwsem are not held in
13622 + * file I/O. Aufs needs to stop lockdep in aufs_mmap() though.
13623 + * It means that when aufs acquires si_rwsem for write, the process should never
13624 + * acquire mmap_sem.
13625 + *
13626 + * Actually aufs_iterate() holds [fdi]i_rwsem before mmap_sem, but this is not a
13627 + * problem either since any directory is not able to be mmap-ed.
13628 + * The similar scenario is applied to aufs_readlink() too.
13629 + */
13630 +
13631 +#if 0 /* stop calling security_file_mmap() */
13632 +/* cf. linux/include/linux/mman.h: calc_vm_prot_bits() */
13633 +#define AuConv_VM_PROT(f, b)   _calc_vm_trans(f, VM_##b, PROT_##b)
13634 +
13635 +static unsigned long au_arch_prot_conv(unsigned long flags)
13636 +{
13637 +       /* currently ppc64 only */
13638 +#ifdef CONFIG_PPC64
13639 +       /* cf. linux/arch/powerpc/include/asm/mman.h */
13640 +       AuDebugOn(arch_calc_vm_prot_bits(-1) != VM_SAO);
13641 +       return AuConv_VM_PROT(flags, SAO);
13642 +#else
13643 +       AuDebugOn(arch_calc_vm_prot_bits(-1));
13644 +       return 0;
13645 +#endif
13646 +}
13647 +
13648 +static unsigned long au_prot_conv(unsigned long flags)
13649 +{
13650 +       return AuConv_VM_PROT(flags, READ)
13651 +               | AuConv_VM_PROT(flags, WRITE)
13652 +               | AuConv_VM_PROT(flags, EXEC)
13653 +               | au_arch_prot_conv(flags);
13654 +}
13655 +
13656 +/* cf. linux/include/linux/mman.h: calc_vm_flag_bits() */
13657 +#define AuConv_VM_MAP(f, b)    _calc_vm_trans(f, VM_##b, MAP_##b)
13658 +
13659 +static unsigned long au_flag_conv(unsigned long flags)
13660 +{
13661 +       return AuConv_VM_MAP(flags, GROWSDOWN)
13662 +               | AuConv_VM_MAP(flags, DENYWRITE)
13663 +               | AuConv_VM_MAP(flags, LOCKED);
13664 +}
13665 +#endif
13666 +
13667 +static int aufs_mmap(struct file *file, struct vm_area_struct *vma)
13668 +{
13669 +       int err;
13670 +       aufs_bindex_t bstart;
13671 +       const unsigned char wlock
13672 +               = (file->f_mode & FMODE_WRITE) && (vma->vm_flags & VM_SHARED);
13673 +       struct dentry *dentry;
13674 +       struct super_block *sb;
13675 +       struct file *h_file;
13676 +       struct au_branch *br;
13677 +       struct au_pin pin;
13678 +
13679 +       AuDbgVmRegion(file, vma);
13680 +
13681 +       dentry = file->f_dentry;
13682 +       sb = dentry->d_sb;
13683 +       lockdep_off();
13684 +       si_read_lock(sb, AuLock_NOPLMW);
13685 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13686 +       if (unlikely(err))
13687 +               goto out;
13688 +
13689 +       if (wlock) {
13690 +               err = au_ready_to_write(file, -1, &pin);
13691 +               di_write_unlock(dentry);
13692 +               if (unlikely(err)) {
13693 +                       fi_write_unlock(file);
13694 +                       goto out;
13695 +               }
13696 +               au_unpin(&pin);
13697 +       } else
13698 +               di_write_unlock(dentry);
13699 +
13700 +       bstart = au_fbstart(file);
13701 +       br = au_sbr(sb, bstart);
13702 +       h_file = au_hf_top(file);
13703 +       get_file(h_file);
13704 +       au_set_mmapped(file);
13705 +       fi_write_unlock(file);
13706 +       lockdep_on();
13707 +
13708 +       au_vm_file_reset(vma, h_file);
13709 +       /*
13710 +        * we cannot call security_mmap_file() here since it may acquire
13711 +        * mmap_sem or i_mutex.
13712 +        *
13713 +        * err = security_mmap_file(h_file, au_prot_conv(vma->vm_flags),
13714 +        *                       au_flag_conv(vma->vm_flags));
13715 +        */
13716 +       if (!err)
13717 +               err = h_file->f_op->mmap(h_file, vma);
13718 +       if (unlikely(err))
13719 +               goto out_reset;
13720 +
13721 +       au_vm_prfile_set(vma, file);
13722 +       /* update without lock, I don't think it a problem */
13723 +       fsstack_copy_attr_atime(file_inode(file), file_inode(h_file));
13724 +       goto out_fput; /* success */
13725 +
13726 +out_reset:
13727 +       au_unset_mmapped(file);
13728 +       au_vm_file_reset(vma, file);
13729 +out_fput:
13730 +       fput(h_file);
13731 +       lockdep_off();
13732 +out:
13733 +       si_read_unlock(sb);
13734 +       lockdep_on();
13735 +       AuTraceErr(err);
13736 +       return err;
13737 +}
13738 +
13739 +/* ---------------------------------------------------------------------- */
13740 +
13741 +static int aufs_fsync_nondir(struct file *file, loff_t start, loff_t end,
13742 +                            int datasync)
13743 +{
13744 +       int err;
13745 +       struct au_pin pin;
13746 +       struct dentry *dentry;
13747 +       struct inode *inode;
13748 +       struct file *h_file;
13749 +       struct super_block *sb;
13750 +
13751 +       dentry = file->f_dentry;
13752 +       inode = dentry->d_inode;
13753 +       sb = dentry->d_sb;
13754 +       mutex_lock(&inode->i_mutex);
13755 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
13756 +       if (unlikely(err))
13757 +               goto out;
13758 +
13759 +       err = 0; /* -EBADF; */ /* posix? */
13760 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
13761 +               goto out_si;
13762 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13763 +       if (unlikely(err))
13764 +               goto out_si;
13765 +
13766 +       err = au_ready_to_write(file, -1, &pin);
13767 +       di_downgrade_lock(dentry, AuLock_IR);
13768 +       if (unlikely(err))
13769 +               goto out_unlock;
13770 +       au_unpin(&pin);
13771 +
13772 +       err = -EINVAL;
13773 +       h_file = au_hf_top(file);
13774 +       err = vfsub_fsync(h_file, &h_file->f_path, datasync);
13775 +       au_cpup_attr_timesizes(inode);
13776 +
13777 +out_unlock:
13778 +       di_read_unlock(dentry, AuLock_IR);
13779 +       fi_write_unlock(file);
13780 +out_si:
13781 +       si_read_unlock(sb);
13782 +out:
13783 +       mutex_unlock(&inode->i_mutex);
13784 +       return err;
13785 +}
13786 +
13787 +/* no one supports this operation, currently */
13788 +#if 0
13789 +static int aufs_aio_fsync_nondir(struct kiocb *kio, int datasync)
13790 +{
13791 +       int err;
13792 +       struct au_pin pin;
13793 +       struct dentry *dentry;
13794 +       struct inode *inode;
13795 +       struct file *file, *h_file;
13796 +
13797 +       file = kio->ki_filp;
13798 +       dentry = file->f_dentry;
13799 +       inode = dentry->d_inode;
13800 +       au_mtx_and_read_lock(inode);
13801 +
13802 +       err = 0; /* -EBADF; */ /* posix? */
13803 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
13804 +               goto out;
13805 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13806 +       if (unlikely(err))
13807 +               goto out;
13808 +
13809 +       err = au_ready_to_write(file, -1, &pin);
13810 +       di_downgrade_lock(dentry, AuLock_IR);
13811 +       if (unlikely(err))
13812 +               goto out_unlock;
13813 +       au_unpin(&pin);
13814 +
13815 +       err = -ENOSYS;
13816 +       h_file = au_hf_top(file);
13817 +       if (h_file->f_op->aio_fsync) {
13818 +               struct mutex *h_mtx;
13819 +
13820 +               h_mtx = &file_inode(h_file)->i_mutex;
13821 +               if (!is_sync_kiocb(kio)) {
13822 +                       get_file(h_file);
13823 +                       fput(file);
13824 +               }
13825 +               kio->ki_filp = h_file;
13826 +               err = h_file->f_op->aio_fsync(kio, datasync);
13827 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
13828 +               if (!err)
13829 +                       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL);
13830 +               /*ignore*/
13831 +               au_cpup_attr_timesizes(inode);
13832 +               mutex_unlock(h_mtx);
13833 +       }
13834 +
13835 +out_unlock:
13836 +       di_read_unlock(dentry, AuLock_IR);
13837 +       fi_write_unlock(file);
13838 +out:
13839 +       si_read_unlock(inode->sb);
13840 +       mutex_unlock(&inode->i_mutex);
13841 +       return err;
13842 +}
13843 +#endif
13844 +
13845 +static int aufs_fasync(int fd, struct file *file, int flag)
13846 +{
13847 +       int err;
13848 +       struct file *h_file;
13849 +       struct dentry *dentry;
13850 +       struct super_block *sb;
13851 +
13852 +       dentry = file->f_dentry;
13853 +       sb = dentry->d_sb;
13854 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13855 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13856 +       if (unlikely(err))
13857 +               goto out;
13858 +
13859 +       h_file = au_hf_top(file);
13860 +       if (h_file->f_op->fasync)
13861 +               err = h_file->f_op->fasync(fd, h_file, flag);
13862 +
13863 +       di_read_unlock(dentry, AuLock_IR);
13864 +       fi_read_unlock(file);
13865 +
13866 +out:
13867 +       si_read_unlock(sb);
13868 +       return err;
13869 +}
13870 +
13871 +/* ---------------------------------------------------------------------- */
13872 +
13873 +/* no one supports this operation, currently */
13874 +#if 0
13875 +static ssize_t aufs_sendpage(struct file *file, struct page *page, int offset,
13876 +                            size_t len, loff_t *pos , int more)
13877 +{
13878 +}
13879 +#endif
13880 +
13881 +/* ---------------------------------------------------------------------- */
13882 +
13883 +const struct file_operations aufs_file_fop = {
13884 +       .owner          = THIS_MODULE,
13885 +
13886 +       .llseek         = default_llseek,
13887 +
13888 +       .read           = aufs_read,
13889 +       .write          = aufs_write,
13890 +       .read_iter      = aufs_read_iter,
13891 +       .write_iter     = aufs_write_iter,
13892 +
13893 +#ifdef CONFIG_AUFS_POLL
13894 +       .poll           = aufs_poll,
13895 +#endif
13896 +       .unlocked_ioctl = aufs_ioctl_nondir,
13897 +#ifdef CONFIG_COMPAT
13898 +       .compat_ioctl   = aufs_compat_ioctl_nondir,
13899 +#endif
13900 +       .mmap           = aufs_mmap,
13901 +       .open           = aufs_open_nondir,
13902 +       .flush          = aufs_flush_nondir,
13903 +       .release        = aufs_release_nondir,
13904 +       .fsync          = aufs_fsync_nondir,
13905 +       /* .aio_fsync   = aufs_aio_fsync_nondir, */
13906 +       .fasync         = aufs_fasync,
13907 +       /* .sendpage    = aufs_sendpage, */
13908 +       .splice_write   = aufs_splice_write,
13909 +       .splice_read    = aufs_splice_read,
13910 +#if 0
13911 +       .aio_splice_write = aufs_aio_splice_write,
13912 +       .aio_splice_read  = aufs_aio_splice_read,
13913 +#endif
13914 +       .fallocate      = aufs_fallocate
13915 +};
13916 diff -urN /usr/share/empty/fs/aufs/fstype.h linux/fs/aufs/fstype.h
13917 --- /usr/share/empty/fs/aufs/fstype.h   1970-01-01 01:00:00.000000000 +0100
13918 +++ linux/fs/aufs/fstype.h      2015-01-25 13:00:38.631047076 +0100
13919 @@ -0,0 +1,469 @@
13920 +/*
13921 + * Copyright (C) 2005-2014 Junjiro R. Okajima
13922 + *
13923 + * This program, aufs is free software; you can redistribute it and/or modify
13924 + * it under the terms of the GNU General Public License as published by
13925 + * the Free Software Foundation; either version 2 of the License, or
13926 + * (at your option) any later version.
13927 + *
13928 + * This program is distributed in the hope that it will be useful,
13929 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13930 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13931 + * GNU General Public License for more details.
13932 + *
13933 + * You should have received a copy of the GNU General Public License
13934 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13935 + */
13936 +
13937 +/*
13938 + * judging filesystem type
13939 + */
13940 +
13941 +#ifndef __AUFS_FSTYPE_H__
13942 +#define __AUFS_FSTYPE_H__
13943 +
13944 +#ifdef __KERNEL__
13945 +
13946 +#include <linux/fs.h>
13947 +#include <linux/magic.h>
13948 +#include <linux/romfs_fs.h>
13949 +
13950 +static inline int au_test_aufs(struct super_block *sb)
13951 +{
13952 +       return sb->s_magic == AUFS_SUPER_MAGIC;
13953 +}
13954 +
13955 +static inline const char *au_sbtype(struct super_block *sb)
13956 +{
13957 +       return sb->s_type->name;
13958 +}
13959 +
13960 +static inline int au_test_iso9660(struct super_block *sb __maybe_unused)
13961 +{
13962 +#if defined(CONFIG_ROMFS_FS) || defined(CONFIG_ROMFS_FS_MODULE)
13963 +       return sb->s_magic == ROMFS_MAGIC;
13964 +#else
13965 +       return 0;
13966 +#endif
13967 +}
13968 +
13969 +static inline int au_test_romfs(struct super_block *sb __maybe_unused)
13970 +{
13971 +#if defined(CONFIG_ISO9660_FS) || defined(CONFIG_ISO9660_FS_MODULE)
13972 +       return sb->s_magic == ISOFS_SUPER_MAGIC;
13973 +#else
13974 +       return 0;
13975 +#endif
13976 +}
13977 +
13978 +static inline int au_test_cramfs(struct super_block *sb __maybe_unused)
13979 +{
13980 +#if defined(CONFIG_CRAMFS) || defined(CONFIG_CRAMFS_MODULE)
13981 +       return sb->s_magic == CRAMFS_MAGIC;
13982 +#endif
13983 +       return 0;
13984 +}
13985 +
13986 +static inline int au_test_nfs(struct super_block *sb __maybe_unused)
13987 +{
13988 +#if defined(CONFIG_NFS_FS) || defined(CONFIG_NFS_FS_MODULE)
13989 +       return sb->s_magic == NFS_SUPER_MAGIC;
13990 +#else
13991 +       return 0;
13992 +#endif
13993 +}
13994 +
13995 +static inline int au_test_fuse(struct super_block *sb __maybe_unused)
13996 +{
13997 +#if defined(CONFIG_FUSE_FS) || defined(CONFIG_FUSE_FS_MODULE)
13998 +       return sb->s_magic == FUSE_SUPER_MAGIC;
13999 +#else
14000 +       return 0;
14001 +#endif
14002 +}
14003 +
14004 +static inline int au_test_xfs(struct super_block *sb __maybe_unused)
14005 +{
14006 +#if defined(CONFIG_XFS_FS) || defined(CONFIG_XFS_FS_MODULE)
14007 +       return sb->s_magic == XFS_SB_MAGIC;
14008 +#else
14009 +       return 0;
14010 +#endif
14011 +}
14012 +
14013 +static inline int au_test_tmpfs(struct super_block *sb __maybe_unused)
14014 +{
14015 +#ifdef CONFIG_TMPFS
14016 +       return sb->s_magic == TMPFS_MAGIC;
14017 +#else
14018 +       return 0;
14019 +#endif
14020 +}
14021 +
14022 +static inline int au_test_ecryptfs(struct super_block *sb __maybe_unused)
14023 +{
14024 +#if defined(CONFIG_ECRYPT_FS) || defined(CONFIG_ECRYPT_FS_MODULE)
14025 +       return !strcmp(au_sbtype(sb), "ecryptfs");
14026 +#else
14027 +       return 0;
14028 +#endif
14029 +}
14030 +
14031 +static inline int au_test_ocfs2(struct super_block *sb __maybe_unused)
14032 +{
14033 +#if defined(CONFIG_OCFS2_FS) || defined(CONFIG_OCFS2_FS_MODULE)
14034 +       return sb->s_magic == OCFS2_SUPER_MAGIC;
14035 +#else
14036 +       return 0;
14037 +#endif
14038 +}
14039 +
14040 +static inline int au_test_ocfs2_dlmfs(struct super_block *sb __maybe_unused)
14041 +{
14042 +#if defined(CONFIG_OCFS2_FS_O2CB) || defined(CONFIG_OCFS2_FS_O2CB_MODULE)
14043 +       return sb->s_magic == DLMFS_MAGIC;
14044 +#else
14045 +       return 0;
14046 +#endif
14047 +}
14048 +
14049 +static inline int au_test_coda(struct super_block *sb __maybe_unused)
14050 +{
14051 +#if defined(CONFIG_CODA_FS) || defined(CONFIG_CODA_FS_MODULE)
14052 +       return sb->s_magic == CODA_SUPER_MAGIC;
14053 +#else
14054 +       return 0;
14055 +#endif
14056 +}
14057 +
14058 +static inline int au_test_v9fs(struct super_block *sb __maybe_unused)
14059 +{
14060 +#if defined(CONFIG_9P_FS) || defined(CONFIG_9P_FS_MODULE)
14061 +       return sb->s_magic == V9FS_MAGIC;
14062 +#else
14063 +       return 0;
14064 +#endif
14065 +}
14066 +
14067 +static inline int au_test_ext4(struct super_block *sb __maybe_unused)
14068 +{
14069 +#if defined(CONFIG_EXT4_FS) || defined(CONFIG_EXT4_FS_MODULE)
14070 +       return sb->s_magic == EXT4_SUPER_MAGIC;
14071 +#else
14072 +       return 0;
14073 +#endif
14074 +}
14075 +
14076 +static inline int au_test_sysv(struct super_block *sb __maybe_unused)
14077 +{
14078 +#if defined(CONFIG_SYSV_FS) || defined(CONFIG_SYSV_FS_MODULE)
14079 +       return !strcmp(au_sbtype(sb), "sysv");
14080 +#else
14081 +       return 0;
14082 +#endif
14083 +}
14084 +
14085 +static inline int au_test_ramfs(struct super_block *sb)
14086 +{
14087 +       return sb->s_magic == RAMFS_MAGIC;
14088 +}
14089 +
14090 +static inline int au_test_ubifs(struct super_block *sb __maybe_unused)
14091 +{
14092 +#if defined(CONFIG_UBIFS_FS) || defined(CONFIG_UBIFS_FS_MODULE)
14093 +       return sb->s_magic == UBIFS_SUPER_MAGIC;
14094 +#else
14095 +       return 0;
14096 +#endif
14097 +}
14098 +
14099 +static inline int au_test_procfs(struct super_block *sb __maybe_unused)
14100 +{
14101 +#ifdef CONFIG_PROC_FS
14102 +       return sb->s_magic == PROC_SUPER_MAGIC;
14103 +#else
14104 +       return 0;
14105 +#endif
14106 +}
14107 +
14108 +static inline int au_test_sysfs(struct super_block *sb __maybe_unused)
14109 +{
14110 +#ifdef CONFIG_SYSFS
14111 +       return sb->s_magic == SYSFS_MAGIC;
14112 +#else
14113 +       return 0;
14114 +#endif
14115 +}
14116 +
14117 +static inline int au_test_configfs(struct super_block *sb __maybe_unused)
14118 +{
14119 +#if defined(CONFIG_CONFIGFS_FS) || defined(CONFIG_CONFIGFS_FS_MODULE)
14120 +       return sb->s_magic == CONFIGFS_MAGIC;
14121 +#else
14122 +       return 0;
14123 +#endif
14124 +}
14125 +
14126 +static inline int au_test_minix(struct super_block *sb __maybe_unused)
14127 +{
14128 +#if defined(CONFIG_MINIX_FS) || defined(CONFIG_MINIX_FS_MODULE)
14129 +       return sb->s_magic == MINIX3_SUPER_MAGIC
14130 +               || sb->s_magic == MINIX2_SUPER_MAGIC
14131 +               || sb->s_magic == MINIX2_SUPER_MAGIC2
14132 +               || sb->s_magic == MINIX_SUPER_MAGIC
14133 +               || sb->s_magic == MINIX_SUPER_MAGIC2;
14134 +#else
14135 +       return 0;
14136 +#endif
14137 +}
14138 +
14139 +static inline int au_test_cifs(struct super_block *sb __maybe_unused)
14140 +{
14141 +#if defined(CONFIG_CIFS_FS) || defined(CONFIGCIFS_FS_MODULE)
14142 +       return sb->s_magic == CIFS_MAGIC_NUMBER;
14143 +#else
14144 +       return 0;
14145 +#endif
14146 +}
14147 +
14148 +static inline int au_test_fat(struct super_block *sb __maybe_unused)
14149 +{
14150 +#if defined(CONFIG_FAT_FS) || defined(CONFIG_FAT_FS_MODULE)
14151 +       return sb->s_magic == MSDOS_SUPER_MAGIC;
14152 +#else
14153 +       return 0;
14154 +#endif
14155 +}
14156 +
14157 +static inline int au_test_msdos(struct super_block *sb)
14158 +{
14159 +       return au_test_fat(sb);
14160 +}
14161 +
14162 +static inline int au_test_vfat(struct super_block *sb)
14163 +{
14164 +       return au_test_fat(sb);
14165 +}
14166 +
14167 +static inline int au_test_securityfs(struct super_block *sb __maybe_unused)
14168 +{
14169 +#ifdef CONFIG_SECURITYFS
14170 +       return sb->s_magic == SECURITYFS_MAGIC;
14171 +#else
14172 +       return 0;
14173 +#endif
14174 +}
14175 +
14176 +static inline int au_test_squashfs(struct super_block *sb __maybe_unused)
14177 +{
14178 +#if defined(CONFIG_SQUASHFS) || defined(CONFIG_SQUASHFS_MODULE)
14179 +       return sb->s_magic == SQUASHFS_MAGIC;
14180 +#else
14181 +       return 0;
14182 +#endif
14183 +}
14184 +
14185 +static inline int au_test_btrfs(struct super_block *sb __maybe_unused)
14186 +{
14187 +#if defined(CONFIG_BTRFS_FS) || defined(CONFIG_BTRFS_FS_MODULE)
14188 +       return sb->s_magic == BTRFS_SUPER_MAGIC;
14189 +#else
14190 +       return 0;
14191 +#endif
14192 +}
14193 +
14194 +static inline int au_test_xenfs(struct super_block *sb __maybe_unused)
14195 +{
14196 +#if defined(CONFIG_XENFS) || defined(CONFIG_XENFS_MODULE)
14197 +       return sb->s_magic == XENFS_SUPER_MAGIC;
14198 +#else
14199 +       return 0;
14200 +#endif
14201 +}
14202 +
14203 +static inline int au_test_debugfs(struct super_block *sb __maybe_unused)
14204 +{
14205 +#ifdef CONFIG_DEBUG_FS
14206 +       return sb->s_magic == DEBUGFS_MAGIC;
14207 +#else
14208 +       return 0;
14209 +#endif
14210 +}
14211 +
14212 +static inline int au_test_nilfs(struct super_block *sb __maybe_unused)
14213 +{
14214 +#if defined(CONFIG_NILFS) || defined(CONFIG_NILFS_MODULE)
14215 +       return sb->s_magic == NILFS_SUPER_MAGIC;
14216 +#else
14217 +       return 0;
14218 +#endif
14219 +}
14220 +
14221 +static inline int au_test_hfsplus(struct super_block *sb __maybe_unused)
14222 +{
14223 +#if defined(CONFIG_HFSPLUS_FS) || defined(CONFIG_HFSPLUS_FS_MODULE)
14224 +       return sb->s_magic == HFSPLUS_SUPER_MAGIC;
14225 +#else
14226 +       return 0;
14227 +#endif
14228 +}
14229 +
14230 +/* ---------------------------------------------------------------------- */
14231 +/*
14232 + * they can't be an aufs branch.
14233 + */
14234 +static inline int au_test_fs_unsuppoted(struct super_block *sb)
14235 +{
14236 +       return
14237 +#ifndef CONFIG_AUFS_BR_RAMFS
14238 +               au_test_ramfs(sb) ||
14239 +#endif
14240 +               au_test_procfs(sb)
14241 +               || au_test_sysfs(sb)
14242 +               || au_test_configfs(sb)
14243 +               || au_test_debugfs(sb)
14244 +               || au_test_securityfs(sb)
14245 +               || au_test_xenfs(sb)
14246 +               || au_test_ecryptfs(sb)
14247 +               /* || !strcmp(au_sbtype(sb), "unionfs") */
14248 +               || au_test_aufs(sb); /* will be supported in next version */
14249 +}
14250 +
14251 +static inline int au_test_fs_remote(struct super_block *sb)
14252 +{
14253 +       return !au_test_tmpfs(sb)
14254 +#ifdef CONFIG_AUFS_BR_RAMFS
14255 +               && !au_test_ramfs(sb)
14256 +#endif
14257 +               && !(sb->s_type->fs_flags & FS_REQUIRES_DEV);
14258 +}
14259 +
14260 +/* ---------------------------------------------------------------------- */
14261 +
14262 +/*
14263 + * Note: these functions (below) are created after reading ->getattr() in all
14264 + * filesystems under linux/fs. it means we have to do so in every update...
14265 + */
14266 +
14267 +/*
14268 + * some filesystems require getattr to refresh the inode attributes before
14269 + * referencing.
14270 + * in most cases, we can rely on the inode attribute in NFS (or every remote fs)
14271 + * and leave the work for d_revalidate()
14272 + */
14273 +static inline int au_test_fs_refresh_iattr(struct super_block *sb)
14274 +{
14275 +       return au_test_nfs(sb)
14276 +               || au_test_fuse(sb)
14277 +               /* || au_test_ocfs2(sb) */      /* untested */
14278 +               /* || au_test_btrfs(sb) */      /* untested */
14279 +               /* || au_test_coda(sb) */       /* untested */
14280 +               /* || au_test_v9fs(sb) */       /* untested */
14281 +               ;
14282 +}
14283 +
14284 +/*
14285 + * filesystems which don't maintain i_size or i_blocks.
14286 + */
14287 +static inline int au_test_fs_bad_iattr_size(struct super_block *sb)
14288 +{
14289 +       return au_test_xfs(sb)
14290 +               || au_test_btrfs(sb)
14291 +               || au_test_ubifs(sb)
14292 +               || au_test_hfsplus(sb)  /* maintained, but incorrect */
14293 +               /* || au_test_ext4(sb) */       /* untested */
14294 +               /* || au_test_ocfs2(sb) */      /* untested */
14295 +               /* || au_test_ocfs2_dlmfs(sb) */ /* untested */
14296 +               /* || au_test_sysv(sb) */       /* untested */
14297 +               /* || au_test_minix(sb) */      /* untested */
14298 +               ;
14299 +}
14300 +
14301 +/*
14302 + * filesystems which don't store the correct value in some of their inode
14303 + * attributes.
14304 + */
14305 +static inline int au_test_fs_bad_iattr(struct super_block *sb)
14306 +{
14307 +       return au_test_fs_bad_iattr_size(sb)
14308 +               /* || au_test_cifs(sb) */       /* untested */
14309 +               || au_test_fat(sb)
14310 +               || au_test_msdos(sb)
14311 +               || au_test_vfat(sb);
14312 +}
14313 +
14314 +/* they don't check i_nlink in link(2) */
14315 +static inline int au_test_fs_no_limit_nlink(struct super_block *sb)
14316 +{
14317 +       return au_test_tmpfs(sb)
14318 +#ifdef CONFIG_AUFS_BR_RAMFS
14319 +               || au_test_ramfs(sb)
14320 +#endif
14321 +               || au_test_ubifs(sb)
14322 +               || au_test_hfsplus(sb);
14323 +}
14324 +
14325 +/*
14326 + * filesystems which sets S_NOATIME and S_NOCMTIME.
14327 + */
14328 +static inline int au_test_fs_notime(struct super_block *sb)
14329 +{
14330 +       return au_test_nfs(sb)
14331 +               || au_test_fuse(sb)
14332 +               || au_test_ubifs(sb)
14333 +               /* || au_test_cifs(sb) */       /* untested */
14334 +               ;
14335 +}
14336 +
14337 +/*
14338 + * filesystems which requires replacing i_mapping.
14339 + */
14340 +static inline int au_test_fs_bad_mapping(struct super_block *sb)
14341 +{
14342 +       return au_test_fuse(sb)
14343 +               || au_test_ubifs(sb);
14344 +}
14345 +
14346 +/* temporary support for i#1 in cramfs */
14347 +static inline int au_test_fs_unique_ino(struct inode *inode)
14348 +{
14349 +       if (au_test_cramfs(inode->i_sb))
14350 +               return inode->i_ino != 1;
14351 +       return 1;
14352 +}
14353 +
14354 +/* ---------------------------------------------------------------------- */
14355 +
14356 +/*
14357 + * the filesystem where the xino files placed must support i/o after unlink and
14358 + * maintain i_size and i_blocks.
14359 + */
14360 +static inline int au_test_fs_bad_xino(struct super_block *sb)
14361 +{
14362 +       return au_test_fs_remote(sb)
14363 +               || au_test_fs_bad_iattr_size(sb)
14364 +               /* don't want unnecessary work for xino */
14365 +               || au_test_aufs(sb)
14366 +               || au_test_ecryptfs(sb)
14367 +               || au_test_nilfs(sb);
14368 +}
14369 +
14370 +static inline int au_test_fs_trunc_xino(struct super_block *sb)
14371 +{
14372 +       return au_test_tmpfs(sb)
14373 +               || au_test_ramfs(sb);
14374 +}
14375 +
14376 +/*
14377 + * test if the @sb is real-readonly.
14378 + */
14379 +static inline int au_test_fs_rr(struct super_block *sb)
14380 +{
14381 +       return au_test_squashfs(sb)
14382 +               || au_test_iso9660(sb)
14383 +               || au_test_cramfs(sb)
14384 +               || au_test_romfs(sb);
14385 +}
14386 +
14387 +#endif /* __KERNEL__ */
14388 +#endif /* __AUFS_FSTYPE_H__ */
14389 diff -urN /usr/share/empty/fs/aufs/hfsnotify.c linux/fs/aufs/hfsnotify.c
14390 --- /usr/share/empty/fs/aufs/hfsnotify.c        1970-01-01 01:00:00.000000000 +0100
14391 +++ linux/fs/aufs/hfsnotify.c   2015-01-25 13:00:38.631047076 +0100
14392 @@ -0,0 +1,288 @@
14393 +/*
14394 + * Copyright (C) 2005-2014 Junjiro R. Okajima
14395 + *
14396 + * This program, aufs is free software; you can redistribute it and/or modify
14397 + * it under the terms of the GNU General Public License as published by
14398 + * the Free Software Foundation; either version 2 of the License, or
14399 + * (at your option) any later version.
14400 + *
14401 + * This program is distributed in the hope that it will be useful,
14402 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14403 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14404 + * GNU General Public License for more details.
14405 + *
14406 + * You should have received a copy of the GNU General Public License
14407 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14408 + */
14409 +
14410 +/*
14411 + * fsnotify for the lower directories
14412 + */
14413 +
14414 +#include "aufs.h"
14415 +
14416 +/* FS_IN_IGNORED is unnecessary */
14417 +static const __u32 AuHfsnMask = (FS_MOVED_TO | FS_MOVED_FROM | FS_DELETE
14418 +                                | FS_CREATE | FS_EVENT_ON_CHILD);
14419 +static DECLARE_WAIT_QUEUE_HEAD(au_hfsn_wq);
14420 +static __cacheline_aligned_in_smp atomic64_t au_hfsn_ifree = ATOMIC64_INIT(0);
14421 +
14422 +static void au_hfsn_free_mark(struct fsnotify_mark *mark)
14423 +{
14424 +       struct au_hnotify *hn = container_of(mark, struct au_hnotify,
14425 +                                            hn_mark);
14426 +       AuDbg("here\n");
14427 +       au_cache_free_hnotify(hn);
14428 +       smp_mb__before_atomic();
14429 +       if (atomic64_dec_and_test(&au_hfsn_ifree))
14430 +               wake_up(&au_hfsn_wq);
14431 +}
14432 +
14433 +static int au_hfsn_alloc(struct au_hinode *hinode)
14434 +{
14435 +       int err;
14436 +       struct au_hnotify *hn;
14437 +       struct super_block *sb;
14438 +       struct au_branch *br;
14439 +       struct fsnotify_mark *mark;
14440 +       aufs_bindex_t bindex;
14441 +
14442 +       hn = hinode->hi_notify;
14443 +       sb = hn->hn_aufs_inode->i_sb;
14444 +       bindex = au_br_index(sb, hinode->hi_id);
14445 +       br = au_sbr(sb, bindex);
14446 +       AuDebugOn(!br->br_hfsn);
14447 +
14448 +       mark = &hn->hn_mark;
14449 +       fsnotify_init_mark(mark, au_hfsn_free_mark);
14450 +       mark->mask = AuHfsnMask;
14451 +       /*
14452 +        * by udba rename or rmdir, aufs assign a new inode to the known
14453 +        * h_inode, so specify 1 to allow dups.
14454 +        */
14455 +       lockdep_off();
14456 +       err = fsnotify_add_mark(mark, br->br_hfsn->hfsn_group, hinode->hi_inode,
14457 +                                /*mnt*/NULL, /*allow_dups*/1);
14458 +       /* even if err */
14459 +       fsnotify_put_mark(mark);
14460 +       lockdep_on();
14461 +
14462 +       return err;
14463 +}
14464 +
14465 +static int au_hfsn_free(struct au_hinode *hinode, struct au_hnotify *hn)
14466 +{
14467 +       struct fsnotify_mark *mark;
14468 +       unsigned long long ull;
14469 +       struct fsnotify_group *group;
14470 +
14471 +       ull = atomic64_inc_return(&au_hfsn_ifree);
14472 +       BUG_ON(!ull);
14473 +
14474 +       mark = &hn->hn_mark;
14475 +       spin_lock(&mark->lock);
14476 +       group = mark->group;
14477 +       fsnotify_get_group(group);
14478 +       spin_unlock(&mark->lock);
14479 +       lockdep_off();
14480 +       fsnotify_destroy_mark(mark, group);
14481 +       fsnotify_put_group(group);
14482 +       lockdep_on();
14483 +
14484 +       /* free hn by myself */
14485 +       return 0;
14486 +}
14487 +
14488 +/* ---------------------------------------------------------------------- */
14489 +
14490 +static void au_hfsn_ctl(struct au_hinode *hinode, int do_set)
14491 +{
14492 +       struct fsnotify_mark *mark;
14493 +
14494 +       mark = &hinode->hi_notify->hn_mark;
14495 +       spin_lock(&mark->lock);
14496 +       if (do_set) {
14497 +               AuDebugOn(mark->mask & AuHfsnMask);
14498 +               mark->mask |= AuHfsnMask;
14499 +       } else {
14500 +               AuDebugOn(!(mark->mask & AuHfsnMask));
14501 +               mark->mask &= ~AuHfsnMask;
14502 +       }
14503 +       spin_unlock(&mark->lock);
14504 +       /* fsnotify_recalc_inode_mask(hinode->hi_inode); */
14505 +}
14506 +
14507 +/* ---------------------------------------------------------------------- */
14508 +
14509 +/* #define AuDbgHnotify */
14510 +#ifdef AuDbgHnotify
14511 +static char *au_hfsn_name(u32 mask)
14512 +{
14513 +#ifdef CONFIG_AUFS_DEBUG
14514 +#define test_ret(flag)                         \
14515 +       do {                                    \
14516 +               if (mask & flag)                \
14517 +                       return #flag;           \
14518 +       } while (0)
14519 +       test_ret(FS_ACCESS);
14520 +       test_ret(FS_MODIFY);
14521 +       test_ret(FS_ATTRIB);
14522 +       test_ret(FS_CLOSE_WRITE);
14523 +       test_ret(FS_CLOSE_NOWRITE);
14524 +       test_ret(FS_OPEN);
14525 +       test_ret(FS_MOVED_FROM);
14526 +       test_ret(FS_MOVED_TO);
14527 +       test_ret(FS_CREATE);
14528 +       test_ret(FS_DELETE);
14529 +       test_ret(FS_DELETE_SELF);
14530 +       test_ret(FS_MOVE_SELF);
14531 +       test_ret(FS_UNMOUNT);
14532 +       test_ret(FS_Q_OVERFLOW);
14533 +       test_ret(FS_IN_IGNORED);
14534 +       test_ret(FS_IN_ISDIR);
14535 +       test_ret(FS_IN_ONESHOT);
14536 +       test_ret(FS_EVENT_ON_CHILD);
14537 +       return "";
14538 +#undef test_ret
14539 +#else
14540 +       return "??";
14541 +#endif
14542 +}
14543 +#endif
14544 +
14545 +/* ---------------------------------------------------------------------- */
14546 +
14547 +static void au_hfsn_free_group(struct fsnotify_group *group)
14548 +{
14549 +       struct au_br_hfsnotify *hfsn = group->private;
14550 +
14551 +       AuDbg("here\n");
14552 +       kfree(hfsn);
14553 +}
14554 +
14555 +static int au_hfsn_handle_event(struct fsnotify_group *group,
14556 +                               struct inode *inode,
14557 +                               struct fsnotify_mark *inode_mark,
14558 +                               struct fsnotify_mark *vfsmount_mark,
14559 +                               u32 mask, void *data, int data_type,
14560 +                               const unsigned char *file_name, u32 cookie)
14561 +{
14562 +       int err;
14563 +       struct au_hnotify *hnotify;
14564 +       struct inode *h_dir, *h_inode;
14565 +       struct qstr h_child_qstr = QSTR_INIT(file_name, strlen(file_name));
14566 +
14567 +       AuDebugOn(data_type != FSNOTIFY_EVENT_INODE);
14568 +
14569 +       err = 0;
14570 +       /* if FS_UNMOUNT happens, there must be another bug */
14571 +       AuDebugOn(mask & FS_UNMOUNT);
14572 +       if (mask & (FS_IN_IGNORED | FS_UNMOUNT))
14573 +               goto out;
14574 +
14575 +       h_dir = inode;
14576 +       h_inode = NULL;
14577 +#ifdef AuDbgHnotify
14578 +       au_debug_on();
14579 +       if (1 || h_child_qstr.len != sizeof(AUFS_XINO_FNAME) - 1
14580 +           || strncmp(h_child_qstr.name, AUFS_XINO_FNAME, h_child_qstr.len)) {
14581 +               AuDbg("i%lu, mask 0x%x %s, hcname %.*s, hi%lu\n",
14582 +                     h_dir->i_ino, mask, au_hfsn_name(mask),
14583 +                     AuLNPair(&h_child_qstr), h_inode ? h_inode->i_ino : 0);
14584 +               /* WARN_ON(1); */
14585 +       }
14586 +       au_debug_off();
14587 +#endif
14588 +
14589 +       AuDebugOn(!inode_mark);
14590 +       hnotify = container_of(inode_mark, struct au_hnotify, hn_mark);
14591 +       err = au_hnotify(h_dir, hnotify, mask, &h_child_qstr, h_inode);
14592 +
14593 +out:
14594 +       return err;
14595 +}
14596 +
14597 +static struct fsnotify_ops au_hfsn_ops = {
14598 +       .handle_event           = au_hfsn_handle_event,
14599 +       .free_group_priv        = au_hfsn_free_group
14600 +};
14601 +
14602 +/* ---------------------------------------------------------------------- */
14603 +
14604 +static void au_hfsn_fin_br(struct au_branch *br)
14605 +{
14606 +       struct au_br_hfsnotify *hfsn;
14607 +
14608 +       hfsn = br->br_hfsn;
14609 +       if (hfsn) {
14610 +               lockdep_off();
14611 +               fsnotify_put_group(hfsn->hfsn_group);
14612 +               lockdep_on();
14613 +       }
14614 +}
14615 +
14616 +static int au_hfsn_init_br(struct au_branch *br, int perm)
14617 +{
14618 +       int err;
14619 +       struct fsnotify_group *group;
14620 +       struct au_br_hfsnotify *hfsn;
14621 +
14622 +       err = 0;
14623 +       br->br_hfsn = NULL;
14624 +       if (!au_br_hnotifyable(perm))
14625 +               goto out;
14626 +
14627 +       err = -ENOMEM;
14628 +       hfsn = kmalloc(sizeof(*hfsn), GFP_NOFS);
14629 +       if (unlikely(!hfsn))
14630 +               goto out;
14631 +
14632 +       err = 0;
14633 +       group = fsnotify_alloc_group(&au_hfsn_ops);
14634 +       if (IS_ERR(group)) {
14635 +               err = PTR_ERR(group);
14636 +               pr_err("fsnotify_alloc_group() failed, %d\n", err);
14637 +               goto out_hfsn;
14638 +       }
14639 +
14640 +       group->private = hfsn;
14641 +       hfsn->hfsn_group = group;
14642 +       br->br_hfsn = hfsn;
14643 +       goto out; /* success */
14644 +
14645 +out_hfsn:
14646 +       kfree(hfsn);
14647 +out:
14648 +       return err;
14649 +}
14650 +
14651 +static int au_hfsn_reset_br(unsigned int udba, struct au_branch *br, int perm)
14652 +{
14653 +       int err;
14654 +
14655 +       err = 0;
14656 +       if (!br->br_hfsn)
14657 +               err = au_hfsn_init_br(br, perm);
14658 +
14659 +       return err;
14660 +}
14661 +
14662 +/* ---------------------------------------------------------------------- */
14663 +
14664 +static void au_hfsn_fin(void)
14665 +{
14666 +       AuDbg("au_hfsn_ifree %lld\n", (long long)atomic64_read(&au_hfsn_ifree));
14667 +       wait_event(au_hfsn_wq, !atomic64_read(&au_hfsn_ifree));
14668 +}
14669 +
14670 +const struct au_hnotify_op au_hnotify_op = {
14671 +       .ctl            = au_hfsn_ctl,
14672 +       .alloc          = au_hfsn_alloc,
14673 +       .free           = au_hfsn_free,
14674 +
14675 +       .fin            = au_hfsn_fin,
14676 +
14677 +       .reset_br       = au_hfsn_reset_br,
14678 +       .fin_br         = au_hfsn_fin_br,
14679 +       .init_br        = au_hfsn_init_br
14680 +};
14681 diff -urN /usr/share/empty/fs/aufs/hfsplus.c linux/fs/aufs/hfsplus.c
14682 --- /usr/share/empty/fs/aufs/hfsplus.c  1970-01-01 01:00:00.000000000 +0100
14683 +++ linux/fs/aufs/hfsplus.c     2015-01-25 13:00:38.631047076 +0100
14684 @@ -0,0 +1,56 @@
14685 +/*
14686 + * Copyright (C) 2010-2014 Junjiro R. Okajima
14687 + *
14688 + * This program, aufs is free software; you can redistribute it and/or modify
14689 + * it under the terms of the GNU General Public License as published by
14690 + * the Free Software Foundation; either version 2 of the License, or
14691 + * (at your option) any later version.
14692 + *
14693 + * This program is distributed in the hope that it will be useful,
14694 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14695 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14696 + * GNU General Public License for more details.
14697 + *
14698 + * You should have received a copy of the GNU General Public License
14699 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14700 + */
14701 +
14702 +/*
14703 + * special support for filesystems which aqucires an inode mutex
14704 + * at final closing a file, eg, hfsplus.
14705 + *
14706 + * This trick is very simple and stupid, just to open the file before really
14707 + * neceeary open to tell hfsplus that this is not the final closing.
14708 + * The caller should call au_h_open_pre() after acquiring the inode mutex,
14709 + * and au_h_open_post() after releasing it.
14710 + */
14711 +
14712 +#include "aufs.h"
14713 +
14714 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
14715 +                          int force_wr)
14716 +{
14717 +       struct file *h_file;
14718 +       struct dentry *h_dentry;
14719 +
14720 +       h_dentry = au_h_dptr(dentry, bindex);
14721 +       AuDebugOn(!h_dentry);
14722 +       AuDebugOn(!h_dentry->d_inode);
14723 +
14724 +       h_file = NULL;
14725 +       if (au_test_hfsplus(h_dentry->d_sb)
14726 +           && S_ISREG(h_dentry->d_inode->i_mode))
14727 +               h_file = au_h_open(dentry, bindex,
14728 +                                  O_RDONLY | O_NOATIME | O_LARGEFILE,
14729 +                                  /*file*/NULL, force_wr);
14730 +       return h_file;
14731 +}
14732 +
14733 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
14734 +                   struct file *h_file)
14735 +{
14736 +       if (h_file) {
14737 +               fput(h_file);
14738 +               au_sbr_put(dentry->d_sb, bindex);
14739 +       }
14740 +}
14741 diff -urN /usr/share/empty/fs/aufs/hnotify.c linux/fs/aufs/hnotify.c
14742 --- /usr/share/empty/fs/aufs/hnotify.c  1970-01-01 01:00:00.000000000 +0100
14743 +++ linux/fs/aufs/hnotify.c     2015-01-25 13:00:38.631047076 +0100
14744 @@ -0,0 +1,714 @@
14745 +/*
14746 + * Copyright (C) 2005-2014 Junjiro R. Okajima
14747 + *
14748 + * This program, aufs is free software; you can redistribute it and/or modify
14749 + * it under the terms of the GNU General Public License as published by
14750 + * the Free Software Foundation; either version 2 of the License, or
14751 + * (at your option) any later version.
14752 + *
14753 + * This program is distributed in the hope that it will be useful,
14754 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14755 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14756 + * GNU General Public License for more details.
14757 + *
14758 + * You should have received a copy of the GNU General Public License
14759 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14760 + */
14761 +
14762 +/*
14763 + * abstraction to notify the direct changes on lower directories
14764 + */
14765 +
14766 +#include "aufs.h"
14767 +
14768 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode)
14769 +{
14770 +       int err;
14771 +       struct au_hnotify *hn;
14772 +
14773 +       err = -ENOMEM;
14774 +       hn = au_cache_alloc_hnotify();
14775 +       if (hn) {
14776 +               hn->hn_aufs_inode = inode;
14777 +               hinode->hi_notify = hn;
14778 +               err = au_hnotify_op.alloc(hinode);
14779 +               AuTraceErr(err);
14780 +               if (unlikely(err)) {
14781 +                       hinode->hi_notify = NULL;
14782 +                       au_cache_free_hnotify(hn);
14783 +                       /*
14784 +                        * The upper dir was removed by udba, but the same named
14785 +                        * dir left. In this case, aufs assignes a new inode
14786 +                        * number and set the monitor again.
14787 +                        * For the lower dir, the old monitnor is still left.
14788 +                        */
14789 +                       if (err == -EEXIST)
14790 +                               err = 0;
14791 +               }
14792 +       }
14793 +
14794 +       AuTraceErr(err);
14795 +       return err;
14796 +}
14797 +
14798 +void au_hn_free(struct au_hinode *hinode)
14799 +{
14800 +       struct au_hnotify *hn;
14801 +
14802 +       hn = hinode->hi_notify;
14803 +       if (hn) {
14804 +               hinode->hi_notify = NULL;
14805 +               if (au_hnotify_op.free(hinode, hn))
14806 +                       au_cache_free_hnotify(hn);
14807 +       }
14808 +}
14809 +
14810 +/* ---------------------------------------------------------------------- */
14811 +
14812 +void au_hn_ctl(struct au_hinode *hinode, int do_set)
14813 +{
14814 +       if (hinode->hi_notify)
14815 +               au_hnotify_op.ctl(hinode, do_set);
14816 +}
14817 +
14818 +void au_hn_reset(struct inode *inode, unsigned int flags)
14819 +{
14820 +       aufs_bindex_t bindex, bend;
14821 +       struct inode *hi;
14822 +       struct dentry *iwhdentry;
14823 +
14824 +       bend = au_ibend(inode);
14825 +       for (bindex = au_ibstart(inode); bindex <= bend; bindex++) {
14826 +               hi = au_h_iptr(inode, bindex);
14827 +               if (!hi)
14828 +                       continue;
14829 +
14830 +               /* mutex_lock_nested(&hi->i_mutex, AuLsc_I_CHILD); */
14831 +               iwhdentry = au_hi_wh(inode, bindex);
14832 +               if (iwhdentry)
14833 +                       dget(iwhdentry);
14834 +               au_igrab(hi);
14835 +               au_set_h_iptr(inode, bindex, NULL, 0);
14836 +               au_set_h_iptr(inode, bindex, au_igrab(hi),
14837 +                             flags & ~AuHi_XINO);
14838 +               iput(hi);
14839 +               dput(iwhdentry);
14840 +               /* mutex_unlock(&hi->i_mutex); */
14841 +       }
14842 +}
14843 +
14844 +/* ---------------------------------------------------------------------- */
14845 +
14846 +static int hn_xino(struct inode *inode, struct inode *h_inode)
14847 +{
14848 +       int err;
14849 +       aufs_bindex_t bindex, bend, bfound, bstart;
14850 +       struct inode *h_i;
14851 +
14852 +       err = 0;
14853 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
14854 +               pr_warn("branch root dir was changed\n");
14855 +               goto out;
14856 +       }
14857 +
14858 +       bfound = -1;
14859 +       bend = au_ibend(inode);
14860 +       bstart = au_ibstart(inode);
14861 +#if 0 /* reserved for future use */
14862 +       if (bindex == bend) {
14863 +               /* keep this ino in rename case */
14864 +               goto out;
14865 +       }
14866 +#endif
14867 +       for (bindex = bstart; bindex <= bend; bindex++)
14868 +               if (au_h_iptr(inode, bindex) == h_inode) {
14869 +                       bfound = bindex;
14870 +                       break;
14871 +               }
14872 +       if (bfound < 0)
14873 +               goto out;
14874 +
14875 +       for (bindex = bstart; bindex <= bend; bindex++) {
14876 +               h_i = au_h_iptr(inode, bindex);
14877 +               if (!h_i)
14878 +                       continue;
14879 +
14880 +               err = au_xino_write(inode->i_sb, bindex, h_i->i_ino, /*ino*/0);
14881 +               /* ignore this error */
14882 +               /* bad action? */
14883 +       }
14884 +
14885 +       /* children inode number will be broken */
14886 +
14887 +out:
14888 +       AuTraceErr(err);
14889 +       return err;
14890 +}
14891 +
14892 +static int hn_gen_tree(struct dentry *dentry)
14893 +{
14894 +       int err, i, j, ndentry;
14895 +       struct au_dcsub_pages dpages;
14896 +       struct au_dpage *dpage;
14897 +       struct dentry **dentries;
14898 +
14899 +       err = au_dpages_init(&dpages, GFP_NOFS);
14900 +       if (unlikely(err))
14901 +               goto out;
14902 +       err = au_dcsub_pages(&dpages, dentry, NULL, NULL);
14903 +       if (unlikely(err))
14904 +               goto out_dpages;
14905 +
14906 +       for (i = 0; i < dpages.ndpage; i++) {
14907 +               dpage = dpages.dpages + i;
14908 +               dentries = dpage->dentries;
14909 +               ndentry = dpage->ndentry;
14910 +               for (j = 0; j < ndentry; j++) {
14911 +                       struct dentry *d;
14912 +
14913 +                       d = dentries[j];
14914 +                       if (IS_ROOT(d))
14915 +                               continue;
14916 +
14917 +                       au_digen_dec(d);
14918 +                       if (d->d_inode)
14919 +                               /* todo: reset children xino?
14920 +                                  cached children only? */
14921 +                               au_iigen_dec(d->d_inode);
14922 +               }
14923 +       }
14924 +
14925 +out_dpages:
14926 +       au_dpages_free(&dpages);
14927 +
14928 +#if 0
14929 +       /* discard children */
14930 +       dentry_unhash(dentry);
14931 +       dput(dentry);
14932 +#endif
14933 +out:
14934 +       return err;
14935 +}
14936 +
14937 +/*
14938 + * return 0 if processed.
14939 + */
14940 +static int hn_gen_by_inode(char *name, unsigned int nlen, struct inode *inode,
14941 +                          const unsigned int isdir)
14942 +{
14943 +       int err;
14944 +       struct dentry *d;
14945 +       struct qstr *dname;
14946 +
14947 +       err = 1;
14948 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
14949 +               pr_warn("branch root dir was changed\n");
14950 +               err = 0;
14951 +               goto out;
14952 +       }
14953 +
14954 +       if (!isdir) {
14955 +               AuDebugOn(!name);
14956 +               au_iigen_dec(inode);
14957 +               spin_lock(&inode->i_lock);
14958 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
14959 +                       spin_lock(&d->d_lock);
14960 +                       dname = &d->d_name;
14961 +                       if (dname->len != nlen
14962 +                           && memcmp(dname->name, name, nlen)) {
14963 +                               spin_unlock(&d->d_lock);
14964 +                               continue;
14965 +                       }
14966 +                       err = 0;
14967 +                       au_digen_dec(d);
14968 +                       spin_unlock(&d->d_lock);
14969 +                       break;
14970 +               }
14971 +               spin_unlock(&inode->i_lock);
14972 +       } else {
14973 +               au_fset_si(au_sbi(inode->i_sb), FAILED_REFRESH_DIR);
14974 +               d = d_find_any_alias(inode);
14975 +               if (!d) {
14976 +                       au_iigen_dec(inode);
14977 +                       goto out;
14978 +               }
14979 +
14980 +               spin_lock(&d->d_lock);
14981 +               dname = &d->d_name;
14982 +               if (dname->len == nlen && !memcmp(dname->name, name, nlen)) {
14983 +                       spin_unlock(&d->d_lock);
14984 +                       err = hn_gen_tree(d);
14985 +                       spin_lock(&d->d_lock);
14986 +               }
14987 +               spin_unlock(&d->d_lock);
14988 +               dput(d);
14989 +       }
14990 +
14991 +out:
14992 +       AuTraceErr(err);
14993 +       return err;
14994 +}
14995 +
14996 +static int hn_gen_by_name(struct dentry *dentry, const unsigned int isdir)
14997 +{
14998 +       int err;
14999 +       struct inode *inode;
15000 +
15001 +       inode = dentry->d_inode;
15002 +       if (IS_ROOT(dentry)
15003 +           /* || (inode && inode->i_ino == AUFS_ROOT_INO) */
15004 +               ) {
15005 +               pr_warn("branch root dir was changed\n");
15006 +               return 0;
15007 +       }
15008 +
15009 +       err = 0;
15010 +       if (!isdir) {
15011 +               au_digen_dec(dentry);
15012 +               if (inode)
15013 +                       au_iigen_dec(inode);
15014 +       } else {
15015 +               au_fset_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR);
15016 +               if (inode)
15017 +                       err = hn_gen_tree(dentry);
15018 +       }
15019 +
15020 +       AuTraceErr(err);
15021 +       return err;
15022 +}
15023 +
15024 +/* ---------------------------------------------------------------------- */
15025 +
15026 +/* hnotify job flags */
15027 +#define AuHnJob_XINO0          1
15028 +#define AuHnJob_GEN            (1 << 1)
15029 +#define AuHnJob_DIRENT         (1 << 2)
15030 +#define AuHnJob_ISDIR          (1 << 3)
15031 +#define AuHnJob_TRYXINO0       (1 << 4)
15032 +#define AuHnJob_MNTPNT         (1 << 5)
15033 +#define au_ftest_hnjob(flags, name)    ((flags) & AuHnJob_##name)
15034 +#define au_fset_hnjob(flags, name) \
15035 +       do { (flags) |= AuHnJob_##name; } while (0)
15036 +#define au_fclr_hnjob(flags, name) \
15037 +       do { (flags) &= ~AuHnJob_##name; } while (0)
15038 +
15039 +enum {
15040 +       AuHn_CHILD,
15041 +       AuHn_PARENT,
15042 +       AuHnLast
15043 +};
15044 +
15045 +struct au_hnotify_args {
15046 +       struct inode *h_dir, *dir, *h_child_inode;
15047 +       u32 mask;
15048 +       unsigned int flags[AuHnLast];
15049 +       unsigned int h_child_nlen;
15050 +       char h_child_name[];
15051 +};
15052 +
15053 +struct hn_job_args {
15054 +       unsigned int flags;
15055 +       struct inode *inode, *h_inode, *dir, *h_dir;
15056 +       struct dentry *dentry;
15057 +       char *h_name;
15058 +       int h_nlen;
15059 +};
15060 +
15061 +static int hn_job(struct hn_job_args *a)
15062 +{
15063 +       const unsigned int isdir = au_ftest_hnjob(a->flags, ISDIR);
15064 +       int e;
15065 +
15066 +       /* reset xino */
15067 +       if (au_ftest_hnjob(a->flags, XINO0) && a->inode)
15068 +               hn_xino(a->inode, a->h_inode); /* ignore this error */
15069 +
15070 +       if (au_ftest_hnjob(a->flags, TRYXINO0)
15071 +           && a->inode
15072 +           && a->h_inode) {
15073 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
15074 +               if (!a->h_inode->i_nlink
15075 +                   && !(a->h_inode->i_state & I_LINKABLE))
15076 +                       hn_xino(a->inode, a->h_inode); /* ignore this error */
15077 +               mutex_unlock(&a->h_inode->i_mutex);
15078 +       }
15079 +
15080 +       /* make the generation obsolete */
15081 +       if (au_ftest_hnjob(a->flags, GEN)) {
15082 +               e = -1;
15083 +               if (a->inode)
15084 +                       e = hn_gen_by_inode(a->h_name, a->h_nlen, a->inode,
15085 +                                             isdir);
15086 +               if (e && a->dentry)
15087 +                       hn_gen_by_name(a->dentry, isdir);
15088 +               /* ignore this error */
15089 +       }
15090 +
15091 +       /* make dir entries obsolete */
15092 +       if (au_ftest_hnjob(a->flags, DIRENT) && a->inode) {
15093 +               struct au_vdir *vdir;
15094 +
15095 +               vdir = au_ivdir(a->inode);
15096 +               if (vdir)
15097 +                       vdir->vd_jiffy = 0;
15098 +               /* IMustLock(a->inode); */
15099 +               /* a->inode->i_version++; */
15100 +       }
15101 +
15102 +       /* can do nothing but warn */
15103 +       if (au_ftest_hnjob(a->flags, MNTPNT)
15104 +           && a->dentry
15105 +           && d_mountpoint(a->dentry))
15106 +               pr_warn("mount-point %pd is removed or renamed\n", a->dentry);
15107 +
15108 +       return 0;
15109 +}
15110 +
15111 +/* ---------------------------------------------------------------------- */
15112 +
15113 +static struct dentry *lookup_wlock_by_name(char *name, unsigned int nlen,
15114 +                                          struct inode *dir)
15115 +{
15116 +       struct dentry *dentry, *d, *parent;
15117 +       struct qstr *dname;
15118 +
15119 +       parent = d_find_any_alias(dir);
15120 +       if (!parent)
15121 +               return NULL;
15122 +
15123 +       dentry = NULL;
15124 +       spin_lock(&parent->d_lock);
15125 +       list_for_each_entry(d, &parent->d_subdirs, d_child) {
15126 +               /* AuDbg("%pd\n", d); */
15127 +               spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
15128 +               dname = &d->d_name;
15129 +               if (dname->len != nlen || memcmp(dname->name, name, nlen))
15130 +                       goto cont_unlock;
15131 +               if (au_di(d))
15132 +                       au_digen_dec(d);
15133 +               else
15134 +                       goto cont_unlock;
15135 +               if (au_dcount(d) > 0) {
15136 +                       dentry = dget_dlock(d);
15137 +                       spin_unlock(&d->d_lock);
15138 +                       break;
15139 +               }
15140 +
15141 +cont_unlock:
15142 +               spin_unlock(&d->d_lock);
15143 +       }
15144 +       spin_unlock(&parent->d_lock);
15145 +       dput(parent);
15146 +
15147 +       if (dentry)
15148 +               di_write_lock_child(dentry);
15149 +
15150 +       return dentry;
15151 +}
15152 +
15153 +static struct inode *lookup_wlock_by_ino(struct super_block *sb,
15154 +                                        aufs_bindex_t bindex, ino_t h_ino)
15155 +{
15156 +       struct inode *inode;
15157 +       ino_t ino;
15158 +       int err;
15159 +
15160 +       inode = NULL;
15161 +       err = au_xino_read(sb, bindex, h_ino, &ino);
15162 +       if (!err && ino)
15163 +               inode = ilookup(sb, ino);
15164 +       if (!inode)
15165 +               goto out;
15166 +
15167 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15168 +               pr_warn("wrong root branch\n");
15169 +               iput(inode);
15170 +               inode = NULL;
15171 +               goto out;
15172 +       }
15173 +
15174 +       ii_write_lock_child(inode);
15175 +
15176 +out:
15177 +       return inode;
15178 +}
15179 +
15180 +static void au_hn_bh(void *_args)
15181 +{
15182 +       struct au_hnotify_args *a = _args;
15183 +       struct super_block *sb;
15184 +       aufs_bindex_t bindex, bend, bfound;
15185 +       unsigned char xino, try_iput;
15186 +       int err;
15187 +       struct inode *inode;
15188 +       ino_t h_ino;
15189 +       struct hn_job_args args;
15190 +       struct dentry *dentry;
15191 +       struct au_sbinfo *sbinfo;
15192 +
15193 +       AuDebugOn(!_args);
15194 +       AuDebugOn(!a->h_dir);
15195 +       AuDebugOn(!a->dir);
15196 +       AuDebugOn(!a->mask);
15197 +       AuDbg("mask 0x%x, i%lu, hi%lu, hci%lu\n",
15198 +             a->mask, a->dir->i_ino, a->h_dir->i_ino,
15199 +             a->h_child_inode ? a->h_child_inode->i_ino : 0);
15200 +
15201 +       inode = NULL;
15202 +       dentry = NULL;
15203 +       /*
15204 +        * do not lock a->dir->i_mutex here
15205 +        * because of d_revalidate() may cause a deadlock.
15206 +        */
15207 +       sb = a->dir->i_sb;
15208 +       AuDebugOn(!sb);
15209 +       sbinfo = au_sbi(sb);
15210 +       AuDebugOn(!sbinfo);
15211 +       si_write_lock(sb, AuLock_NOPLMW);
15212 +
15213 +       ii_read_lock_parent(a->dir);
15214 +       bfound = -1;
15215 +       bend = au_ibend(a->dir);
15216 +       for (bindex = au_ibstart(a->dir); bindex <= bend; bindex++)
15217 +               if (au_h_iptr(a->dir, bindex) == a->h_dir) {
15218 +                       bfound = bindex;
15219 +                       break;
15220 +               }
15221 +       ii_read_unlock(a->dir);
15222 +       if (unlikely(bfound < 0))
15223 +               goto out;
15224 +
15225 +       xino = !!au_opt_test(au_mntflags(sb), XINO);
15226 +       h_ino = 0;
15227 +       if (a->h_child_inode)
15228 +               h_ino = a->h_child_inode->i_ino;
15229 +
15230 +       if (a->h_child_nlen
15231 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], GEN)
15232 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], MNTPNT)))
15233 +               dentry = lookup_wlock_by_name(a->h_child_name, a->h_child_nlen,
15234 +                                             a->dir);
15235 +       try_iput = 0;
15236 +       if (dentry)
15237 +               inode = dentry->d_inode;
15238 +       if (xino && !inode && h_ino
15239 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], XINO0)
15240 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], TRYXINO0)
15241 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], GEN))) {
15242 +               inode = lookup_wlock_by_ino(sb, bfound, h_ino);
15243 +               try_iput = 1;
15244 +           }
15245 +
15246 +       args.flags = a->flags[AuHn_CHILD];
15247 +       args.dentry = dentry;
15248 +       args.inode = inode;
15249 +       args.h_inode = a->h_child_inode;
15250 +       args.dir = a->dir;
15251 +       args.h_dir = a->h_dir;
15252 +       args.h_name = a->h_child_name;
15253 +       args.h_nlen = a->h_child_nlen;
15254 +       err = hn_job(&args);
15255 +       if (dentry) {
15256 +               if (au_di(dentry))
15257 +                       di_write_unlock(dentry);
15258 +               dput(dentry);
15259 +       }
15260 +       if (inode && try_iput) {
15261 +               ii_write_unlock(inode);
15262 +               iput(inode);
15263 +       }
15264 +
15265 +       ii_write_lock_parent(a->dir);
15266 +       args.flags = a->flags[AuHn_PARENT];
15267 +       args.dentry = NULL;
15268 +       args.inode = a->dir;
15269 +       args.h_inode = a->h_dir;
15270 +       args.dir = NULL;
15271 +       args.h_dir = NULL;
15272 +       args.h_name = NULL;
15273 +       args.h_nlen = 0;
15274 +       err = hn_job(&args);
15275 +       ii_write_unlock(a->dir);
15276 +
15277 +out:
15278 +       iput(a->h_child_inode);
15279 +       iput(a->h_dir);
15280 +       iput(a->dir);
15281 +       si_write_unlock(sb);
15282 +       au_nwt_done(&sbinfo->si_nowait);
15283 +       kfree(a);
15284 +}
15285 +
15286 +/* ---------------------------------------------------------------------- */
15287 +
15288 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
15289 +              struct qstr *h_child_qstr, struct inode *h_child_inode)
15290 +{
15291 +       int err, len;
15292 +       unsigned int flags[AuHnLast], f;
15293 +       unsigned char isdir, isroot, wh;
15294 +       struct inode *dir;
15295 +       struct au_hnotify_args *args;
15296 +       char *p, *h_child_name;
15297 +
15298 +       err = 0;
15299 +       AuDebugOn(!hnotify || !hnotify->hn_aufs_inode);
15300 +       dir = igrab(hnotify->hn_aufs_inode);
15301 +       if (!dir)
15302 +               goto out;
15303 +
15304 +       isroot = (dir->i_ino == AUFS_ROOT_INO);
15305 +       wh = 0;
15306 +       h_child_name = (void *)h_child_qstr->name;
15307 +       len = h_child_qstr->len;
15308 +       if (h_child_name) {
15309 +               if (len > AUFS_WH_PFX_LEN
15310 +                   && !memcmp(h_child_name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
15311 +                       h_child_name += AUFS_WH_PFX_LEN;
15312 +                       len -= AUFS_WH_PFX_LEN;
15313 +                       wh = 1;
15314 +               }
15315 +       }
15316 +
15317 +       isdir = 0;
15318 +       if (h_child_inode)
15319 +               isdir = !!S_ISDIR(h_child_inode->i_mode);
15320 +       flags[AuHn_PARENT] = AuHnJob_ISDIR;
15321 +       flags[AuHn_CHILD] = 0;
15322 +       if (isdir)
15323 +               flags[AuHn_CHILD] = AuHnJob_ISDIR;
15324 +       au_fset_hnjob(flags[AuHn_PARENT], DIRENT);
15325 +       au_fset_hnjob(flags[AuHn_CHILD], GEN);
15326 +       switch (mask & FS_EVENTS_POSS_ON_CHILD) {
15327 +       case FS_MOVED_FROM:
15328 +       case FS_MOVED_TO:
15329 +               au_fset_hnjob(flags[AuHn_CHILD], XINO0);
15330 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15331 +               /*FALLTHROUGH*/
15332 +       case FS_CREATE:
15333 +               AuDebugOn(!h_child_name);
15334 +               break;
15335 +
15336 +       case FS_DELETE:
15337 +               /*
15338 +                * aufs never be able to get this child inode.
15339 +                * revalidation should be in d_revalidate()
15340 +                * by checking i_nlink, i_generation or d_unhashed().
15341 +                */
15342 +               AuDebugOn(!h_child_name);
15343 +               au_fset_hnjob(flags[AuHn_CHILD], TRYXINO0);
15344 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15345 +               break;
15346 +
15347 +       default:
15348 +               AuDebugOn(1);
15349 +       }
15350 +
15351 +       if (wh)
15352 +               h_child_inode = NULL;
15353 +
15354 +       err = -ENOMEM;
15355 +       /* iput() and kfree() will be called in au_hnotify() */
15356 +       args = kmalloc(sizeof(*args) + len + 1, GFP_NOFS);
15357 +       if (unlikely(!args)) {
15358 +               AuErr1("no memory\n");
15359 +               iput(dir);
15360 +               goto out;
15361 +       }
15362 +       args->flags[AuHn_PARENT] = flags[AuHn_PARENT];
15363 +       args->flags[AuHn_CHILD] = flags[AuHn_CHILD];
15364 +       args->mask = mask;
15365 +       args->dir = dir;
15366 +       args->h_dir = igrab(h_dir);
15367 +       if (h_child_inode)
15368 +               h_child_inode = igrab(h_child_inode); /* can be NULL */
15369 +       args->h_child_inode = h_child_inode;
15370 +       args->h_child_nlen = len;
15371 +       if (len) {
15372 +               p = (void *)args;
15373 +               p += sizeof(*args);
15374 +               memcpy(p, h_child_name, len);
15375 +               p[len] = 0;
15376 +       }
15377 +
15378 +       /* NFS fires the event for silly-renamed one from kworker */
15379 +       f = 0;
15380 +       if (!dir->i_nlink
15381 +           || (au_test_nfs(h_dir->i_sb) && (mask & FS_DELETE)))
15382 +               f = AuWkq_NEST;
15383 +       err = au_wkq_nowait(au_hn_bh, args, dir->i_sb, f);
15384 +       if (unlikely(err)) {
15385 +               pr_err("wkq %d\n", err);
15386 +               iput(args->h_child_inode);
15387 +               iput(args->h_dir);
15388 +               iput(args->dir);
15389 +               kfree(args);
15390 +       }
15391 +
15392 +out:
15393 +       return err;
15394 +}
15395 +
15396 +/* ---------------------------------------------------------------------- */
15397 +
15398 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm)
15399 +{
15400 +       int err;
15401 +
15402 +       AuDebugOn(!(udba & AuOptMask_UDBA));
15403 +
15404 +       err = 0;
15405 +       if (au_hnotify_op.reset_br)
15406 +               err = au_hnotify_op.reset_br(udba, br, perm);
15407 +
15408 +       return err;
15409 +}
15410 +
15411 +int au_hnotify_init_br(struct au_branch *br, int perm)
15412 +{
15413 +       int err;
15414 +
15415 +       err = 0;
15416 +       if (au_hnotify_op.init_br)
15417 +               err = au_hnotify_op.init_br(br, perm);
15418 +
15419 +       return err;
15420 +}
15421 +
15422 +void au_hnotify_fin_br(struct au_branch *br)
15423 +{
15424 +       if (au_hnotify_op.fin_br)
15425 +               au_hnotify_op.fin_br(br);
15426 +}
15427 +
15428 +static void au_hn_destroy_cache(void)
15429 +{
15430 +       kmem_cache_destroy(au_cachep[AuCache_HNOTIFY]);
15431 +       au_cachep[AuCache_HNOTIFY] = NULL;
15432 +}
15433 +
15434 +int __init au_hnotify_init(void)
15435 +{
15436 +       int err;
15437 +
15438 +       err = -ENOMEM;
15439 +       au_cachep[AuCache_HNOTIFY] = AuCache(au_hnotify);
15440 +       if (au_cachep[AuCache_HNOTIFY]) {
15441 +               err = 0;
15442 +               if (au_hnotify_op.init)
15443 +                       err = au_hnotify_op.init();
15444 +               if (unlikely(err))
15445 +                       au_hn_destroy_cache();
15446 +       }
15447 +       AuTraceErr(err);
15448 +       return err;
15449 +}
15450 +
15451 +void au_hnotify_fin(void)
15452 +{
15453 +       if (au_hnotify_op.fin)
15454 +               au_hnotify_op.fin();
15455 +       /* cf. au_cache_fin() */
15456 +       if (au_cachep[AuCache_HNOTIFY])
15457 +               au_hn_destroy_cache();
15458 +}
15459 diff -urN /usr/share/empty/fs/aufs/iinfo.c linux/fs/aufs/iinfo.c
15460 --- /usr/share/empty/fs/aufs/iinfo.c    1970-01-01 01:00:00.000000000 +0100
15461 +++ linux/fs/aufs/iinfo.c       2015-01-25 13:00:38.631047076 +0100
15462 @@ -0,0 +1,277 @@
15463 +/*
15464 + * Copyright (C) 2005-2014 Junjiro R. Okajima
15465 + *
15466 + * This program, aufs is free software; you can redistribute it and/or modify
15467 + * it under the terms of the GNU General Public License as published by
15468 + * the Free Software Foundation; either version 2 of the License, or
15469 + * (at your option) any later version.
15470 + *
15471 + * This program is distributed in the hope that it will be useful,
15472 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15473 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15474 + * GNU General Public License for more details.
15475 + *
15476 + * You should have received a copy of the GNU General Public License
15477 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15478 + */
15479 +
15480 +/*
15481 + * inode private data
15482 + */
15483 +
15484 +#include "aufs.h"
15485 +
15486 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex)
15487 +{
15488 +       struct inode *h_inode;
15489 +
15490 +       IiMustAnyLock(inode);
15491 +
15492 +       h_inode = au_ii(inode)->ii_hinode[0 + bindex].hi_inode;
15493 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
15494 +       return h_inode;
15495 +}
15496 +
15497 +/* todo: hard/soft set? */
15498 +void au_hiput(struct au_hinode *hinode)
15499 +{
15500 +       au_hn_free(hinode);
15501 +       dput(hinode->hi_whdentry);
15502 +       iput(hinode->hi_inode);
15503 +}
15504 +
15505 +unsigned int au_hi_flags(struct inode *inode, int isdir)
15506 +{
15507 +       unsigned int flags;
15508 +       const unsigned int mnt_flags = au_mntflags(inode->i_sb);
15509 +
15510 +       flags = 0;
15511 +       if (au_opt_test(mnt_flags, XINO))
15512 +               au_fset_hi(flags, XINO);
15513 +       if (isdir && au_opt_test(mnt_flags, UDBA_HNOTIFY))
15514 +               au_fset_hi(flags, HNOTIFY);
15515 +       return flags;
15516 +}
15517 +
15518 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
15519 +                  struct inode *h_inode, unsigned int flags)
15520 +{
15521 +       struct au_hinode *hinode;
15522 +       struct inode *hi;
15523 +       struct au_iinfo *iinfo = au_ii(inode);
15524 +
15525 +       IiMustWriteLock(inode);
15526 +
15527 +       hinode = iinfo->ii_hinode + bindex;
15528 +       hi = hinode->hi_inode;
15529 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
15530 +
15531 +       if (hi)
15532 +               au_hiput(hinode);
15533 +       hinode->hi_inode = h_inode;
15534 +       if (h_inode) {
15535 +               int err;
15536 +               struct super_block *sb = inode->i_sb;
15537 +               struct au_branch *br;
15538 +
15539 +               AuDebugOn(inode->i_mode
15540 +                         && (h_inode->i_mode & S_IFMT)
15541 +                         != (inode->i_mode & S_IFMT));
15542 +               if (bindex == iinfo->ii_bstart)
15543 +                       au_cpup_igen(inode, h_inode);
15544 +               br = au_sbr(sb, bindex);
15545 +               hinode->hi_id = br->br_id;
15546 +               if (au_ftest_hi(flags, XINO)) {
15547 +                       err = au_xino_write(sb, bindex, h_inode->i_ino,
15548 +                                           inode->i_ino);
15549 +                       if (unlikely(err))
15550 +                               AuIOErr1("failed au_xino_write() %d\n", err);
15551 +               }
15552 +
15553 +               if (au_ftest_hi(flags, HNOTIFY)
15554 +                   && au_br_hnotifyable(br->br_perm)) {
15555 +                       err = au_hn_alloc(hinode, inode);
15556 +                       if (unlikely(err))
15557 +                               AuIOErr1("au_hn_alloc() %d\n", err);
15558 +               }
15559 +       }
15560 +}
15561 +
15562 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
15563 +                 struct dentry *h_wh)
15564 +{
15565 +       struct au_hinode *hinode;
15566 +
15567 +       IiMustWriteLock(inode);
15568 +
15569 +       hinode = au_ii(inode)->ii_hinode + bindex;
15570 +       AuDebugOn(hinode->hi_whdentry);
15571 +       hinode->hi_whdentry = h_wh;
15572 +}
15573 +
15574 +void au_update_iigen(struct inode *inode, int half)
15575 +{
15576 +       struct au_iinfo *iinfo;
15577 +       struct au_iigen *iigen;
15578 +       unsigned int sigen;
15579 +
15580 +       sigen = au_sigen(inode->i_sb);
15581 +       iinfo = au_ii(inode);
15582 +       iigen = &iinfo->ii_generation;
15583 +       spin_lock(&iinfo->ii_genspin);
15584 +       iigen->ig_generation = sigen;
15585 +       if (half)
15586 +               au_ig_fset(iigen->ig_flags, HALF_REFRESHED);
15587 +       else
15588 +               au_ig_fclr(iigen->ig_flags, HALF_REFRESHED);
15589 +       spin_unlock(&iinfo->ii_genspin);
15590 +}
15591 +
15592 +/* it may be called at remount time, too */
15593 +void au_update_ibrange(struct inode *inode, int do_put_zero)
15594 +{
15595 +       struct au_iinfo *iinfo;
15596 +       aufs_bindex_t bindex, bend;
15597 +
15598 +       iinfo = au_ii(inode);
15599 +       if (!iinfo)
15600 +               return;
15601 +
15602 +       IiMustWriteLock(inode);
15603 +
15604 +       if (do_put_zero && iinfo->ii_bstart >= 0) {
15605 +               for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend;
15606 +                    bindex++) {
15607 +                       struct inode *h_i;
15608 +
15609 +                       h_i = iinfo->ii_hinode[0 + bindex].hi_inode;
15610 +                       if (h_i
15611 +                           && !h_i->i_nlink
15612 +                           && !(h_i->i_state & I_LINKABLE))
15613 +                               au_set_h_iptr(inode, bindex, NULL, 0);
15614 +               }
15615 +       }
15616 +
15617 +       iinfo->ii_bstart = -1;
15618 +       iinfo->ii_bend = -1;
15619 +       bend = au_sbend(inode->i_sb);
15620 +       for (bindex = 0; bindex <= bend; bindex++)
15621 +               if (iinfo->ii_hinode[0 + bindex].hi_inode) {
15622 +                       iinfo->ii_bstart = bindex;
15623 +                       break;
15624 +               }
15625 +       if (iinfo->ii_bstart >= 0)
15626 +               for (bindex = bend; bindex >= iinfo->ii_bstart; bindex--)
15627 +                       if (iinfo->ii_hinode[0 + bindex].hi_inode) {
15628 +                               iinfo->ii_bend = bindex;
15629 +                               break;
15630 +                       }
15631 +       AuDebugOn(iinfo->ii_bstart > iinfo->ii_bend);
15632 +}
15633 +
15634 +/* ---------------------------------------------------------------------- */
15635 +
15636 +void au_icntnr_init_once(void *_c)
15637 +{
15638 +       struct au_icntnr *c = _c;
15639 +       struct au_iinfo *iinfo = &c->iinfo;
15640 +       static struct lock_class_key aufs_ii;
15641 +
15642 +       spin_lock_init(&iinfo->ii_genspin);
15643 +       au_rw_init(&iinfo->ii_rwsem);
15644 +       au_rw_class(&iinfo->ii_rwsem, &aufs_ii);
15645 +       inode_init_once(&c->vfs_inode);
15646 +}
15647 +
15648 +int au_iinfo_init(struct inode *inode)
15649 +{
15650 +       struct au_iinfo *iinfo;
15651 +       struct super_block *sb;
15652 +       int nbr, i;
15653 +
15654 +       sb = inode->i_sb;
15655 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
15656 +       nbr = au_sbend(sb) + 1;
15657 +       if (unlikely(nbr <= 0))
15658 +               nbr = 1;
15659 +       iinfo->ii_hinode = kcalloc(nbr, sizeof(*iinfo->ii_hinode), GFP_NOFS);
15660 +       if (iinfo->ii_hinode) {
15661 +               au_ninodes_inc(sb);
15662 +               for (i = 0; i < nbr; i++)
15663 +                       iinfo->ii_hinode[i].hi_id = -1;
15664 +
15665 +               iinfo->ii_generation.ig_generation = au_sigen(sb);
15666 +               iinfo->ii_bstart = -1;
15667 +               iinfo->ii_bend = -1;
15668 +               iinfo->ii_vdir = NULL;
15669 +               return 0;
15670 +       }
15671 +       return -ENOMEM;
15672 +}
15673 +
15674 +int au_ii_realloc(struct au_iinfo *iinfo, int nbr)
15675 +{
15676 +       int err, sz;
15677 +       struct au_hinode *hip;
15678 +
15679 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
15680 +
15681 +       err = -ENOMEM;
15682 +       sz = sizeof(*hip) * (iinfo->ii_bend + 1);
15683 +       if (!sz)
15684 +               sz = sizeof(*hip);
15685 +       hip = au_kzrealloc(iinfo->ii_hinode, sz, sizeof(*hip) * nbr, GFP_NOFS);
15686 +       if (hip) {
15687 +               iinfo->ii_hinode = hip;
15688 +               err = 0;
15689 +       }
15690 +
15691 +       return err;
15692 +}
15693 +
15694 +void au_iinfo_fin(struct inode *inode)
15695 +{
15696 +       struct au_iinfo *iinfo;
15697 +       struct au_hinode *hi;
15698 +       struct super_block *sb;
15699 +       aufs_bindex_t bindex, bend;
15700 +       const unsigned char unlinked = !inode->i_nlink;
15701 +
15702 +       iinfo = au_ii(inode);
15703 +       /* bad_inode case */
15704 +       if (!iinfo)
15705 +               return;
15706 +
15707 +       sb = inode->i_sb;
15708 +       au_ninodes_dec(sb);
15709 +       if (si_pid_test(sb))
15710 +               au_xino_delete_inode(inode, unlinked);
15711 +       else {
15712 +               /*
15713 +                * it is safe to hide the dependency between sbinfo and
15714 +                * sb->s_umount.
15715 +                */
15716 +               lockdep_off();
15717 +               si_noflush_read_lock(sb);
15718 +               au_xino_delete_inode(inode, unlinked);
15719 +               si_read_unlock(sb);
15720 +               lockdep_on();
15721 +       }
15722 +
15723 +       if (iinfo->ii_vdir)
15724 +               au_vdir_free(iinfo->ii_vdir);
15725 +
15726 +       bindex = iinfo->ii_bstart;
15727 +       if (bindex >= 0) {
15728 +               hi = iinfo->ii_hinode + bindex;
15729 +               bend = iinfo->ii_bend;
15730 +               while (bindex++ <= bend) {
15731 +                       if (hi->hi_inode)
15732 +                               au_hiput(hi);
15733 +                       hi++;
15734 +               }
15735 +       }
15736 +       kfree(iinfo->ii_hinode);
15737 +       iinfo->ii_hinode = NULL;
15738 +       AuRwDestroy(&iinfo->ii_rwsem);
15739 +}
15740 diff -urN /usr/share/empty/fs/aufs/inode.c linux/fs/aufs/inode.c
15741 --- /usr/share/empty/fs/aufs/inode.c    1970-01-01 01:00:00.000000000 +0100
15742 +++ linux/fs/aufs/inode.c       2015-01-25 13:00:38.631047076 +0100
15743 @@ -0,0 +1,496 @@
15744 +/*
15745 + * Copyright (C) 2005-2014 Junjiro R. Okajima
15746 + *
15747 + * This program, aufs is free software; you can redistribute it and/or modify
15748 + * it under the terms of the GNU General Public License as published by
15749 + * the Free Software Foundation; either version 2 of the License, or
15750 + * (at your option) any later version.
15751 + *
15752 + * This program is distributed in the hope that it will be useful,
15753 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15754 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15755 + * GNU General Public License for more details.
15756 + *
15757 + * You should have received a copy of the GNU General Public License
15758 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15759 + */
15760 +
15761 +/*
15762 + * inode functions
15763 + */
15764 +
15765 +#include "aufs.h"
15766 +
15767 +struct inode *au_igrab(struct inode *inode)
15768 +{
15769 +       if (inode) {
15770 +               AuDebugOn(!atomic_read(&inode->i_count));
15771 +               ihold(inode);
15772 +       }
15773 +       return inode;
15774 +}
15775 +
15776 +static void au_refresh_hinode_attr(struct inode *inode, int do_version)
15777 +{
15778 +       au_cpup_attr_all(inode, /*force*/0);
15779 +       au_update_iigen(inode, /*half*/1);
15780 +       if (do_version)
15781 +               inode->i_version++;
15782 +}
15783 +
15784 +static int au_ii_refresh(struct inode *inode, int *update)
15785 +{
15786 +       int err, e;
15787 +       umode_t type;
15788 +       aufs_bindex_t bindex, new_bindex;
15789 +       struct super_block *sb;
15790 +       struct au_iinfo *iinfo;
15791 +       struct au_hinode *p, *q, tmp;
15792 +
15793 +       IiMustWriteLock(inode);
15794 +
15795 +       *update = 0;
15796 +       sb = inode->i_sb;
15797 +       type = inode->i_mode & S_IFMT;
15798 +       iinfo = au_ii(inode);
15799 +       err = au_ii_realloc(iinfo, au_sbend(sb) + 1);
15800 +       if (unlikely(err))
15801 +               goto out;
15802 +
15803 +       AuDebugOn(iinfo->ii_bstart < 0);
15804 +       p = iinfo->ii_hinode + iinfo->ii_bstart;
15805 +       for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend;
15806 +            bindex++, p++) {
15807 +               if (!p->hi_inode)
15808 +                       continue;
15809 +
15810 +               AuDebugOn(type != (p->hi_inode->i_mode & S_IFMT));
15811 +               new_bindex = au_br_index(sb, p->hi_id);
15812 +               if (new_bindex == bindex)
15813 +                       continue;
15814 +
15815 +               if (new_bindex < 0) {
15816 +                       *update = 1;
15817 +                       au_hiput(p);
15818 +                       p->hi_inode = NULL;
15819 +                       continue;
15820 +               }
15821 +
15822 +               if (new_bindex < iinfo->ii_bstart)
15823 +                       iinfo->ii_bstart = new_bindex;
15824 +               if (iinfo->ii_bend < new_bindex)
15825 +                       iinfo->ii_bend = new_bindex;
15826 +               /* swap two lower inode, and loop again */
15827 +               q = iinfo->ii_hinode + new_bindex;
15828 +               tmp = *q;
15829 +               *q = *p;
15830 +               *p = tmp;
15831 +               if (tmp.hi_inode) {
15832 +                       bindex--;
15833 +                       p--;
15834 +               }
15835 +       }
15836 +       au_update_ibrange(inode, /*do_put_zero*/0);
15837 +       e = au_dy_irefresh(inode);
15838 +       if (unlikely(e && !err))
15839 +               err = e;
15840 +
15841 +out:
15842 +       AuTraceErr(err);
15843 +       return err;
15844 +}
15845 +
15846 +int au_refresh_hinode_self(struct inode *inode)
15847 +{
15848 +       int err, update;
15849 +
15850 +       err = au_ii_refresh(inode, &update);
15851 +       if (!err)
15852 +               au_refresh_hinode_attr(inode, update && S_ISDIR(inode->i_mode));
15853 +
15854 +       AuTraceErr(err);
15855 +       return err;
15856 +}
15857 +
15858 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry)
15859 +{
15860 +       int err, e, update;
15861 +       unsigned int flags;
15862 +       umode_t mode;
15863 +       aufs_bindex_t bindex, bend;
15864 +       unsigned char isdir;
15865 +       struct au_hinode *p;
15866 +       struct au_iinfo *iinfo;
15867 +
15868 +       err = au_ii_refresh(inode, &update);
15869 +       if (unlikely(err))
15870 +               goto out;
15871 +
15872 +       update = 0;
15873 +       iinfo = au_ii(inode);
15874 +       p = iinfo->ii_hinode + iinfo->ii_bstart;
15875 +       mode = (inode->i_mode & S_IFMT);
15876 +       isdir = S_ISDIR(mode);
15877 +       flags = au_hi_flags(inode, isdir);
15878 +       bend = au_dbend(dentry);
15879 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++) {
15880 +               struct inode *h_i;
15881 +               struct dentry *h_d;
15882 +
15883 +               h_d = au_h_dptr(dentry, bindex);
15884 +               if (!h_d || !h_d->d_inode)
15885 +                       continue;
15886 +
15887 +               AuDebugOn(mode != (h_d->d_inode->i_mode & S_IFMT));
15888 +               if (iinfo->ii_bstart <= bindex && bindex <= iinfo->ii_bend) {
15889 +                       h_i = au_h_iptr(inode, bindex);
15890 +                       if (h_i) {
15891 +                               if (h_i == h_d->d_inode)
15892 +                                       continue;
15893 +                               err = -EIO;
15894 +                               break;
15895 +                       }
15896 +               }
15897 +               if (bindex < iinfo->ii_bstart)
15898 +                       iinfo->ii_bstart = bindex;
15899 +               if (iinfo->ii_bend < bindex)
15900 +                       iinfo->ii_bend = bindex;
15901 +               au_set_h_iptr(inode, bindex, au_igrab(h_d->d_inode), flags);
15902 +               update = 1;
15903 +       }
15904 +       au_update_ibrange(inode, /*do_put_zero*/0);
15905 +       e = au_dy_irefresh(inode);
15906 +       if (unlikely(e && !err))
15907 +               err = e;
15908 +       if (!err)
15909 +               au_refresh_hinode_attr(inode, update && isdir);
15910 +
15911 +out:
15912 +       AuTraceErr(err);
15913 +       return err;
15914 +}
15915 +
15916 +static int set_inode(struct inode *inode, struct dentry *dentry)
15917 +{
15918 +       int err;
15919 +       unsigned int flags;
15920 +       umode_t mode;
15921 +       aufs_bindex_t bindex, bstart, btail;
15922 +       unsigned char isdir;
15923 +       struct dentry *h_dentry;
15924 +       struct inode *h_inode;
15925 +       struct au_iinfo *iinfo;
15926 +
15927 +       IiMustWriteLock(inode);
15928 +
15929 +       err = 0;
15930 +       isdir = 0;
15931 +       bstart = au_dbstart(dentry);
15932 +       h_inode = au_h_dptr(dentry, bstart)->d_inode;
15933 +       mode = h_inode->i_mode;
15934 +       switch (mode & S_IFMT) {
15935 +       case S_IFREG:
15936 +               btail = au_dbtail(dentry);
15937 +               inode->i_op = &aufs_iop;
15938 +               inode->i_fop = &aufs_file_fop;
15939 +               err = au_dy_iaop(inode, bstart, h_inode);
15940 +               if (unlikely(err))
15941 +                       goto out;
15942 +               break;
15943 +       case S_IFDIR:
15944 +               isdir = 1;
15945 +               btail = au_dbtaildir(dentry);
15946 +               inode->i_op = &aufs_dir_iop;
15947 +               inode->i_fop = &aufs_dir_fop;
15948 +               break;
15949 +       case S_IFLNK:
15950 +               btail = au_dbtail(dentry);
15951 +               inode->i_op = &aufs_symlink_iop;
15952 +               break;
15953 +       case S_IFBLK:
15954 +       case S_IFCHR:
15955 +       case S_IFIFO:
15956 +       case S_IFSOCK:
15957 +               btail = au_dbtail(dentry);
15958 +               inode->i_op = &aufs_iop;
15959 +               init_special_inode(inode, mode, h_inode->i_rdev);
15960 +               break;
15961 +       default:
15962 +               AuIOErr("Unknown file type 0%o\n", mode);
15963 +               err = -EIO;
15964 +               goto out;
15965 +       }
15966 +
15967 +       /* do not set hnotify for whiteouted dirs (SHWH mode) */
15968 +       flags = au_hi_flags(inode, isdir);
15969 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH)
15970 +           && au_ftest_hi(flags, HNOTIFY)
15971 +           && dentry->d_name.len > AUFS_WH_PFX_LEN
15972 +           && !memcmp(dentry->d_name.name, AUFS_WH_PFX, AUFS_WH_PFX_LEN))
15973 +               au_fclr_hi(flags, HNOTIFY);
15974 +       iinfo = au_ii(inode);
15975 +       iinfo->ii_bstart = bstart;
15976 +       iinfo->ii_bend = btail;
15977 +       for (bindex = bstart; bindex <= btail; bindex++) {
15978 +               h_dentry = au_h_dptr(dentry, bindex);
15979 +               if (h_dentry)
15980 +                       au_set_h_iptr(inode, bindex,
15981 +                                     au_igrab(h_dentry->d_inode), flags);
15982 +       }
15983 +       au_cpup_attr_all(inode, /*force*/1);
15984 +       /*
15985 +        * to force calling aufs_get_acl() every time,
15986 +        * do not call cache_no_acl() for aufs inode.
15987 +        */
15988 +
15989 +out:
15990 +       return err;
15991 +}
15992 +
15993 +/*
15994 + * successful returns with iinfo write_locked
15995 + * minus: errno
15996 + * zero: success, matched
15997 + * plus: no error, but unmatched
15998 + */
15999 +static int reval_inode(struct inode *inode, struct dentry *dentry)
16000 +{
16001 +       int err;
16002 +       unsigned int gen;
16003 +       struct au_iigen iigen;
16004 +       aufs_bindex_t bindex, bend;
16005 +       struct inode *h_inode, *h_dinode;
16006 +
16007 +       /*
16008 +        * before this function, if aufs got any iinfo lock, it must be only
16009 +        * one, the parent dir.
16010 +        * it can happen by UDBA and the obsoleted inode number.
16011 +        */
16012 +       err = -EIO;
16013 +       if (unlikely(inode->i_ino == parent_ino(dentry)))
16014 +               goto out;
16015 +
16016 +       err = 1;
16017 +       ii_write_lock_new_child(inode);
16018 +       h_dinode = au_h_dptr(dentry, au_dbstart(dentry))->d_inode;
16019 +       bend = au_ibend(inode);
16020 +       for (bindex = au_ibstart(inode); bindex <= bend; bindex++) {
16021 +               h_inode = au_h_iptr(inode, bindex);
16022 +               if (!h_inode || h_inode != h_dinode)
16023 +                       continue;
16024 +
16025 +               err = 0;
16026 +               gen = au_iigen(inode, &iigen);
16027 +               if (gen == au_digen(dentry)
16028 +                   && !au_ig_ftest(iigen.ig_flags, HALF_REFRESHED))
16029 +                       break;
16030 +
16031 +               /* fully refresh inode using dentry */
16032 +               err = au_refresh_hinode(inode, dentry);
16033 +               if (!err)
16034 +                       au_update_iigen(inode, /*half*/0);
16035 +               break;
16036 +       }
16037 +
16038 +       if (unlikely(err))
16039 +               ii_write_unlock(inode);
16040 +out:
16041 +       return err;
16042 +}
16043 +
16044 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16045 +          unsigned int d_type, ino_t *ino)
16046 +{
16047 +       int err;
16048 +       struct mutex *mtx;
16049 +
16050 +       /* prevent hardlinked inode number from race condition */
16051 +       mtx = NULL;
16052 +       if (d_type != DT_DIR) {
16053 +               mtx = &au_sbr(sb, bindex)->br_xino.xi_nondir_mtx;
16054 +               mutex_lock(mtx);
16055 +       }
16056 +       err = au_xino_read(sb, bindex, h_ino, ino);
16057 +       if (unlikely(err))
16058 +               goto out;
16059 +
16060 +       if (!*ino) {
16061 +               err = -EIO;
16062 +               *ino = au_xino_new_ino(sb);
16063 +               if (unlikely(!*ino))
16064 +                       goto out;
16065 +               err = au_xino_write(sb, bindex, h_ino, *ino);
16066 +               if (unlikely(err))
16067 +                       goto out;
16068 +       }
16069 +
16070 +out:
16071 +       if (mtx)
16072 +               mutex_unlock(mtx);
16073 +       return err;
16074 +}
16075 +
16076 +/* successful returns with iinfo write_locked */
16077 +/* todo: return with unlocked? */
16078 +struct inode *au_new_inode(struct dentry *dentry, int must_new)
16079 +{
16080 +       struct inode *inode, *h_inode;
16081 +       struct dentry *h_dentry;
16082 +       struct super_block *sb;
16083 +       struct mutex *mtx;
16084 +       ino_t h_ino, ino;
16085 +       int err;
16086 +       aufs_bindex_t bstart;
16087 +
16088 +       sb = dentry->d_sb;
16089 +       bstart = au_dbstart(dentry);
16090 +       h_dentry = au_h_dptr(dentry, bstart);
16091 +       h_inode = h_dentry->d_inode;
16092 +       h_ino = h_inode->i_ino;
16093 +
16094 +       /*
16095 +        * stop 'race'-ing between hardlinks under different
16096 +        * parents.
16097 +        */
16098 +       mtx = NULL;
16099 +       if (!S_ISDIR(h_inode->i_mode))
16100 +               mtx = &au_sbr(sb, bstart)->br_xino.xi_nondir_mtx;
16101 +
16102 +new_ino:
16103 +       if (mtx)
16104 +               mutex_lock(mtx);
16105 +       err = au_xino_read(sb, bstart, h_ino, &ino);
16106 +       inode = ERR_PTR(err);
16107 +       if (unlikely(err))
16108 +               goto out;
16109 +
16110 +       if (!ino) {
16111 +               ino = au_xino_new_ino(sb);
16112 +               if (unlikely(!ino)) {
16113 +                       inode = ERR_PTR(-EIO);
16114 +                       goto out;
16115 +               }
16116 +       }
16117 +
16118 +       AuDbg("i%lu\n", (unsigned long)ino);
16119 +       inode = au_iget_locked(sb, ino);
16120 +       err = PTR_ERR(inode);
16121 +       if (IS_ERR(inode))
16122 +               goto out;
16123 +
16124 +       AuDbg("%lx, new %d\n", inode->i_state, !!(inode->i_state & I_NEW));
16125 +       if (inode->i_state & I_NEW) {
16126 +               /* verbose coding for lock class name */
16127 +               if (unlikely(S_ISLNK(h_inode->i_mode)))
16128 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
16129 +                                   au_lc_key + AuLcSymlink_IIINFO);
16130 +               else if (unlikely(S_ISDIR(h_inode->i_mode)))
16131 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
16132 +                                   au_lc_key + AuLcDir_IIINFO);
16133 +               else /* likely */
16134 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
16135 +                                   au_lc_key + AuLcNonDir_IIINFO);
16136 +
16137 +               ii_write_lock_new_child(inode);
16138 +               err = set_inode(inode, dentry);
16139 +               if (!err) {
16140 +                       unlock_new_inode(inode);
16141 +                       goto out; /* success */
16142 +               }
16143 +
16144 +               /*
16145 +                * iget_failed() calls iput(), but we need to call
16146 +                * ii_write_unlock() after iget_failed(). so dirty hack for
16147 +                * i_count.
16148 +                */
16149 +               atomic_inc(&inode->i_count);
16150 +               iget_failed(inode);
16151 +               ii_write_unlock(inode);
16152 +               au_xino_write(sb, bstart, h_ino, /*ino*/0);
16153 +               /* ignore this error */
16154 +               goto out_iput;
16155 +       } else if (!must_new && !IS_DEADDIR(inode) && inode->i_nlink) {
16156 +               /*
16157 +                * horrible race condition between lookup, readdir and copyup
16158 +                * (or something).
16159 +                */
16160 +               if (mtx)
16161 +                       mutex_unlock(mtx);
16162 +               err = reval_inode(inode, dentry);
16163 +               if (unlikely(err < 0)) {
16164 +                       mtx = NULL;
16165 +                       goto out_iput;
16166 +               }
16167 +
16168 +               if (!err) {
16169 +                       mtx = NULL;
16170 +                       goto out; /* success */
16171 +               } else if (mtx)
16172 +                       mutex_lock(mtx);
16173 +       }
16174 +
16175 +       if (unlikely(au_test_fs_unique_ino(h_dentry->d_inode)))
16176 +               AuWarn1("Warning: Un-notified UDBA or repeatedly renamed dir,"
16177 +                       " b%d, %s, %pd, hi%lu, i%lu.\n",
16178 +                       bstart, au_sbtype(h_dentry->d_sb), dentry,
16179 +                       (unsigned long)h_ino, (unsigned long)ino);
16180 +       ino = 0;
16181 +       err = au_xino_write(sb, bstart, h_ino, /*ino*/0);
16182 +       if (!err) {
16183 +               iput(inode);
16184 +               if (mtx)
16185 +                       mutex_unlock(mtx);
16186 +               goto new_ino;
16187 +       }
16188 +
16189 +out_iput:
16190 +       iput(inode);
16191 +       inode = ERR_PTR(err);
16192 +out:
16193 +       if (mtx)
16194 +               mutex_unlock(mtx);
16195 +       return inode;
16196 +}
16197 +
16198 +/* ---------------------------------------------------------------------- */
16199 +
16200 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16201 +              struct inode *inode)
16202 +{
16203 +       int err;
16204 +       struct inode *hi;
16205 +
16206 +       err = au_br_rdonly(au_sbr(sb, bindex));
16207 +
16208 +       /* pseudo-link after flushed may happen out of bounds */
16209 +       if (!err
16210 +           && inode
16211 +           && au_ibstart(inode) <= bindex
16212 +           && bindex <= au_ibend(inode)) {
16213 +               /*
16214 +                * permission check is unnecessary since vfsub routine
16215 +                * will be called later
16216 +                */
16217 +               hi = au_h_iptr(inode, bindex);
16218 +               if (hi)
16219 +                       err = IS_IMMUTABLE(hi) ? -EROFS : 0;
16220 +       }
16221 +
16222 +       return err;
16223 +}
16224 +
16225 +int au_test_h_perm(struct inode *h_inode, int mask)
16226 +{
16227 +       if (uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
16228 +               return 0;
16229 +       return inode_permission(h_inode, mask);
16230 +}
16231 +
16232 +int au_test_h_perm_sio(struct inode *h_inode, int mask)
16233 +{
16234 +       if (au_test_nfs(h_inode->i_sb)
16235 +           && (mask & MAY_WRITE)
16236 +           && S_ISDIR(h_inode->i_mode))
16237 +               mask |= MAY_READ; /* force permission check */
16238 +       return au_test_h_perm(h_inode, mask);
16239 +}
16240 diff -urN /usr/share/empty/fs/aufs/inode.h linux/fs/aufs/inode.h
16241 --- /usr/share/empty/fs/aufs/inode.h    1970-01-01 01:00:00.000000000 +0100
16242 +++ linux/fs/aufs/inode.h       2015-01-25 13:00:38.631047076 +0100
16243 @@ -0,0 +1,667 @@
16244 +/*
16245 + * Copyright (C) 2005-2014 Junjiro R. Okajima
16246 + *
16247 + * This program, aufs is free software; you can redistribute it and/or modify
16248 + * it under the terms of the GNU General Public License as published by
16249 + * the Free Software Foundation; either version 2 of the License, or
16250 + * (at your option) any later version.
16251 + *
16252 + * This program is distributed in the hope that it will be useful,
16253 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16254 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16255 + * GNU General Public License for more details.
16256 + *
16257 + * You should have received a copy of the GNU General Public License
16258 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16259 + */
16260 +
16261 +/*
16262 + * inode operations
16263 + */
16264 +
16265 +#ifndef __AUFS_INODE_H__
16266 +#define __AUFS_INODE_H__
16267 +
16268 +#ifdef __KERNEL__
16269 +
16270 +#include <linux/fsnotify.h>
16271 +#include "rwsem.h"
16272 +
16273 +struct vfsmount;
16274 +
16275 +struct au_hnotify {
16276 +#ifdef CONFIG_AUFS_HNOTIFY
16277 +#ifdef CONFIG_AUFS_HFSNOTIFY
16278 +       /* never use fsnotify_add_vfsmount_mark() */
16279 +       struct fsnotify_mark            hn_mark;
16280 +#endif
16281 +       struct inode                    *hn_aufs_inode; /* no get/put */
16282 +#endif
16283 +} ____cacheline_aligned_in_smp;
16284 +
16285 +struct au_hinode {
16286 +       struct inode            *hi_inode;
16287 +       aufs_bindex_t           hi_id;
16288 +#ifdef CONFIG_AUFS_HNOTIFY
16289 +       struct au_hnotify       *hi_notify;
16290 +#endif
16291 +
16292 +       /* reference to the copied-up whiteout with get/put */
16293 +       struct dentry           *hi_whdentry;
16294 +};
16295 +
16296 +/* ig_flags */
16297 +#define AuIG_HALF_REFRESHED            1
16298 +#define au_ig_ftest(flags, name)       ((flags) & AuIG_##name)
16299 +#define au_ig_fset(flags, name) \
16300 +       do { (flags) |= AuIG_##name; } while (0)
16301 +#define au_ig_fclr(flags, name) \
16302 +       do { (flags) &= ~AuIG_##name; } while (0)
16303 +
16304 +struct au_iigen {
16305 +       __u32           ig_generation, ig_flags;
16306 +};
16307 +
16308 +struct au_vdir;
16309 +struct au_iinfo {
16310 +       spinlock_t              ii_genspin;
16311 +       struct au_iigen         ii_generation;
16312 +       struct super_block      *ii_hsb1;       /* no get/put */
16313 +
16314 +       struct au_rwsem         ii_rwsem;
16315 +       aufs_bindex_t           ii_bstart, ii_bend;
16316 +       __u32                   ii_higen;
16317 +       struct au_hinode        *ii_hinode;
16318 +       struct au_vdir          *ii_vdir;
16319 +};
16320 +
16321 +struct au_icntnr {
16322 +       struct au_iinfo iinfo;
16323 +       struct inode vfs_inode;
16324 +} ____cacheline_aligned_in_smp;
16325 +
16326 +/* au_pin flags */
16327 +#define AuPin_DI_LOCKED                1
16328 +#define AuPin_MNT_WRITE                (1 << 1)
16329 +#define au_ftest_pin(flags, name)      ((flags) & AuPin_##name)
16330 +#define au_fset_pin(flags, name) \
16331 +       do { (flags) |= AuPin_##name; } while (0)
16332 +#define au_fclr_pin(flags, name) \
16333 +       do { (flags) &= ~AuPin_##name; } while (0)
16334 +
16335 +struct au_pin {
16336 +       /* input */
16337 +       struct dentry *dentry;
16338 +       unsigned int udba;
16339 +       unsigned char lsc_di, lsc_hi, flags;
16340 +       aufs_bindex_t bindex;
16341 +
16342 +       /* output */
16343 +       struct dentry *parent;
16344 +       struct au_hinode *hdir;
16345 +       struct vfsmount *h_mnt;
16346 +
16347 +       /* temporary unlock/relock for copyup */
16348 +       struct dentry *h_dentry, *h_parent;
16349 +       struct au_branch *br;
16350 +       struct task_struct *task;
16351 +};
16352 +
16353 +void au_pin_hdir_unlock(struct au_pin *p);
16354 +int au_pin_hdir_lock(struct au_pin *p);
16355 +int au_pin_hdir_relock(struct au_pin *p);
16356 +void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task);
16357 +void au_pin_hdir_acquire_nest(struct au_pin *p);
16358 +void au_pin_hdir_release(struct au_pin *p);
16359 +
16360 +/* ---------------------------------------------------------------------- */
16361 +
16362 +static inline struct au_iinfo *au_ii(struct inode *inode)
16363 +{
16364 +       struct au_iinfo *iinfo;
16365 +
16366 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
16367 +       if (iinfo->ii_hinode)
16368 +               return iinfo;
16369 +       return NULL; /* debugging bad_inode case */
16370 +}
16371 +
16372 +/* ---------------------------------------------------------------------- */
16373 +
16374 +/* inode.c */
16375 +struct inode *au_igrab(struct inode *inode);
16376 +int au_refresh_hinode_self(struct inode *inode);
16377 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry);
16378 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16379 +          unsigned int d_type, ino_t *ino);
16380 +struct inode *au_new_inode(struct dentry *dentry, int must_new);
16381 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16382 +              struct inode *inode);
16383 +int au_test_h_perm(struct inode *h_inode, int mask);
16384 +int au_test_h_perm_sio(struct inode *h_inode, int mask);
16385 +
16386 +static inline int au_wh_ino(struct super_block *sb, aufs_bindex_t bindex,
16387 +                           ino_t h_ino, unsigned int d_type, ino_t *ino)
16388 +{
16389 +#ifdef CONFIG_AUFS_SHWH
16390 +       return au_ino(sb, bindex, h_ino, d_type, ino);
16391 +#else
16392 +       return 0;
16393 +#endif
16394 +}
16395 +
16396 +/* i_op.c */
16397 +extern struct inode_operations aufs_iop, aufs_symlink_iop, aufs_dir_iop;
16398 +
16399 +/* au_wr_dir flags */
16400 +#define AuWrDir_ADD_ENTRY      1
16401 +#define AuWrDir_TMP_WHENTRY    (1 << 1)
16402 +#define AuWrDir_ISDIR          (1 << 2)
16403 +#define AuWrDir_TMPFILE                (1 << 3)
16404 +#define au_ftest_wrdir(flags, name)    ((flags) & AuWrDir_##name)
16405 +#define au_fset_wrdir(flags, name) \
16406 +       do { (flags) |= AuWrDir_##name; } while (0)
16407 +#define au_fclr_wrdir(flags, name) \
16408 +       do { (flags) &= ~AuWrDir_##name; } while (0)
16409 +
16410 +struct au_wr_dir_args {
16411 +       aufs_bindex_t force_btgt;
16412 +       unsigned char flags;
16413 +};
16414 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
16415 +             struct au_wr_dir_args *args);
16416 +
16417 +struct dentry *au_pinned_h_parent(struct au_pin *pin);
16418 +void au_pin_init(struct au_pin *pin, struct dentry *dentry,
16419 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
16420 +                unsigned int udba, unsigned char flags);
16421 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
16422 +          unsigned int udba, unsigned char flags) __must_check;
16423 +int au_do_pin(struct au_pin *pin) __must_check;
16424 +void au_unpin(struct au_pin *pin);
16425 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen);
16426 +
16427 +#define AuIcpup_DID_CPUP       1
16428 +#define au_ftest_icpup(flags, name)    ((flags) & AuIcpup_##name)
16429 +#define au_fset_icpup(flags, name) \
16430 +       do { (flags) |= AuIcpup_##name; } while (0)
16431 +#define au_fclr_icpup(flags, name) \
16432 +       do { (flags) &= ~AuIcpup_##name; } while (0)
16433 +
16434 +struct au_icpup_args {
16435 +       unsigned char flags;
16436 +       unsigned char pin_flags;
16437 +       aufs_bindex_t btgt;
16438 +       unsigned int udba;
16439 +       struct au_pin pin;
16440 +       struct path h_path;
16441 +       struct inode *h_inode;
16442 +};
16443 +
16444 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
16445 +                    struct au_icpup_args *a);
16446 +
16447 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path);
16448 +
16449 +/* i_op_add.c */
16450 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
16451 +              struct dentry *h_parent, int isdir);
16452 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
16453 +              dev_t dev);
16454 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname);
16455 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
16456 +               bool want_excl);
16457 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode);
16458 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
16459 +             struct dentry *dentry);
16460 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
16461 +
16462 +/* i_op_del.c */
16463 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup);
16464 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
16465 +              struct dentry *h_parent, int isdir);
16466 +int aufs_unlink(struct inode *dir, struct dentry *dentry);
16467 +int aufs_rmdir(struct inode *dir, struct dentry *dentry);
16468 +
16469 +/* i_op_ren.c */
16470 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt);
16471 +int aufs_rename(struct inode *src_dir, struct dentry *src_dentry,
16472 +               struct inode *dir, struct dentry *dentry);
16473 +
16474 +/* iinfo.c */
16475 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex);
16476 +void au_hiput(struct au_hinode *hinode);
16477 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
16478 +                 struct dentry *h_wh);
16479 +unsigned int au_hi_flags(struct inode *inode, int isdir);
16480 +
16481 +/* hinode flags */
16482 +#define AuHi_XINO      1
16483 +#define AuHi_HNOTIFY   (1 << 1)
16484 +#define au_ftest_hi(flags, name)       ((flags) & AuHi_##name)
16485 +#define au_fset_hi(flags, name) \
16486 +       do { (flags) |= AuHi_##name; } while (0)
16487 +#define au_fclr_hi(flags, name) \
16488 +       do { (flags) &= ~AuHi_##name; } while (0)
16489 +
16490 +#ifndef CONFIG_AUFS_HNOTIFY
16491 +#undef AuHi_HNOTIFY
16492 +#define AuHi_HNOTIFY   0
16493 +#endif
16494 +
16495 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
16496 +                  struct inode *h_inode, unsigned int flags);
16497 +
16498 +void au_update_iigen(struct inode *inode, int half);
16499 +void au_update_ibrange(struct inode *inode, int do_put_zero);
16500 +
16501 +void au_icntnr_init_once(void *_c);
16502 +int au_iinfo_init(struct inode *inode);
16503 +void au_iinfo_fin(struct inode *inode);
16504 +int au_ii_realloc(struct au_iinfo *iinfo, int nbr);
16505 +
16506 +#ifdef CONFIG_PROC_FS
16507 +/* plink.c */
16508 +int au_plink_maint(struct super_block *sb, int flags);
16509 +void au_plink_maint_leave(struct au_sbinfo *sbinfo);
16510 +int au_plink_maint_enter(struct super_block *sb);
16511 +#ifdef CONFIG_AUFS_DEBUG
16512 +void au_plink_list(struct super_block *sb);
16513 +#else
16514 +AuStubVoid(au_plink_list, struct super_block *sb)
16515 +#endif
16516 +int au_plink_test(struct inode *inode);
16517 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex);
16518 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
16519 +                    struct dentry *h_dentry);
16520 +void au_plink_put(struct super_block *sb, int verbose);
16521 +void au_plink_clean(struct super_block *sb, int verbose);
16522 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id);
16523 +#else
16524 +AuStubInt0(au_plink_maint, struct super_block *sb, int flags);
16525 +AuStubVoid(au_plink_maint_leave, struct au_sbinfo *sbinfo);
16526 +AuStubInt0(au_plink_maint_enter, struct super_block *sb);
16527 +AuStubVoid(au_plink_list, struct super_block *sb);
16528 +AuStubInt0(au_plink_test, struct inode *inode);
16529 +AuStub(struct dentry *, au_plink_lkup, return NULL,
16530 +       struct inode *inode, aufs_bindex_t bindex);
16531 +AuStubVoid(au_plink_append, struct inode *inode, aufs_bindex_t bindex,
16532 +          struct dentry *h_dentry);
16533 +AuStubVoid(au_plink_put, struct super_block *sb, int verbose);
16534 +AuStubVoid(au_plink_clean, struct super_block *sb, int verbose);
16535 +AuStubVoid(au_plink_half_refresh, struct super_block *sb, aufs_bindex_t br_id);
16536 +#endif /* CONFIG_PROC_FS */
16537 +
16538 +#ifdef CONFIG_AUFS_XATTR
16539 +/* xattr.c */
16540 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags);
16541 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size);
16542 +ssize_t aufs_getxattr(struct dentry *dentry, const char *name, void *value,
16543 +                     size_t size);
16544 +int aufs_setxattr(struct dentry *dentry, const char *name, const void *value,
16545 +                 size_t size, int flags);
16546 +int aufs_removexattr(struct dentry *dentry, const char *name);
16547 +
16548 +/* void au_xattr_init(struct super_block *sb); */
16549 +#else
16550 +AuStubInt0(au_cpup_xattr, struct dentry *h_dst, struct dentry *h_src,
16551 +          int ignore_flags);
16552 +/* AuStubVoid(au_xattr_init, struct super_block *sb); */
16553 +#endif
16554 +
16555 +#ifdef CONFIG_FS_POSIX_ACL
16556 +struct posix_acl *aufs_get_acl(struct inode *inode, int type);
16557 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
16558 +#endif
16559 +
16560 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
16561 +enum {
16562 +       AU_XATTR_SET,
16563 +       AU_XATTR_REMOVE,
16564 +       AU_ACL_SET
16565 +};
16566 +
16567 +struct au_srxattr {
16568 +       int type;
16569 +       union {
16570 +               struct {
16571 +                       const char      *name;
16572 +                       const void      *value;
16573 +                       size_t          size;
16574 +                       int             flags;
16575 +               } set;
16576 +               struct {
16577 +                       const char      *name;
16578 +               } remove;
16579 +               struct {
16580 +                       struct posix_acl *acl;
16581 +                       int             type;
16582 +               } acl_set;
16583 +       } u;
16584 +};
16585 +ssize_t au_srxattr(struct dentry *dentry, struct au_srxattr *arg);
16586 +#endif
16587 +
16588 +/* ---------------------------------------------------------------------- */
16589 +
16590 +/* lock subclass for iinfo */
16591 +enum {
16592 +       AuLsc_II_CHILD,         /* child first */
16593 +       AuLsc_II_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
16594 +       AuLsc_II_CHILD3,        /* copyup dirs */
16595 +       AuLsc_II_PARENT,        /* see AuLsc_I_PARENT in vfsub.h */
16596 +       AuLsc_II_PARENT2,
16597 +       AuLsc_II_PARENT3,       /* copyup dirs */
16598 +       AuLsc_II_NEW_CHILD
16599 +};
16600 +
16601 +/*
16602 + * ii_read_lock_child, ii_write_lock_child,
16603 + * ii_read_lock_child2, ii_write_lock_child2,
16604 + * ii_read_lock_child3, ii_write_lock_child3,
16605 + * ii_read_lock_parent, ii_write_lock_parent,
16606 + * ii_read_lock_parent2, ii_write_lock_parent2,
16607 + * ii_read_lock_parent3, ii_write_lock_parent3,
16608 + * ii_read_lock_new_child, ii_write_lock_new_child,
16609 + */
16610 +#define AuReadLockFunc(name, lsc) \
16611 +static inline void ii_read_lock_##name(struct inode *i) \
16612 +{ \
16613 +       au_rw_read_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
16614 +}
16615 +
16616 +#define AuWriteLockFunc(name, lsc) \
16617 +static inline void ii_write_lock_##name(struct inode *i) \
16618 +{ \
16619 +       au_rw_write_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
16620 +}
16621 +
16622 +#define AuRWLockFuncs(name, lsc) \
16623 +       AuReadLockFunc(name, lsc) \
16624 +       AuWriteLockFunc(name, lsc)
16625 +
16626 +AuRWLockFuncs(child, CHILD);
16627 +AuRWLockFuncs(child2, CHILD2);
16628 +AuRWLockFuncs(child3, CHILD3);
16629 +AuRWLockFuncs(parent, PARENT);
16630 +AuRWLockFuncs(parent2, PARENT2);
16631 +AuRWLockFuncs(parent3, PARENT3);
16632 +AuRWLockFuncs(new_child, NEW_CHILD);
16633 +
16634 +#undef AuReadLockFunc
16635 +#undef AuWriteLockFunc
16636 +#undef AuRWLockFuncs
16637 +
16638 +/*
16639 + * ii_read_unlock, ii_write_unlock, ii_downgrade_lock
16640 + */
16641 +AuSimpleUnlockRwsemFuncs(ii, struct inode *i, &au_ii(i)->ii_rwsem);
16642 +
16643 +#define IiMustNoWaiters(i)     AuRwMustNoWaiters(&au_ii(i)->ii_rwsem)
16644 +#define IiMustAnyLock(i)       AuRwMustAnyLock(&au_ii(i)->ii_rwsem)
16645 +#define IiMustWriteLock(i)     AuRwMustWriteLock(&au_ii(i)->ii_rwsem)
16646 +
16647 +/* ---------------------------------------------------------------------- */
16648 +
16649 +static inline void au_icntnr_init(struct au_icntnr *c)
16650 +{
16651 +#ifdef CONFIG_AUFS_DEBUG
16652 +       c->vfs_inode.i_mode = 0;
16653 +#endif
16654 +}
16655 +
16656 +static inline unsigned int au_iigen(struct inode *inode, struct au_iigen *iigen)
16657 +{
16658 +       unsigned int gen;
16659 +       struct au_iinfo *iinfo;
16660 +
16661 +       iinfo = au_ii(inode);
16662 +       spin_lock(&iinfo->ii_genspin);
16663 +       if (iigen)
16664 +               *iigen = iinfo->ii_generation;
16665 +       gen = iinfo->ii_generation.ig_generation;
16666 +       spin_unlock(&iinfo->ii_genspin);
16667 +
16668 +       return gen;
16669 +}
16670 +
16671 +/* tiny test for inode number */
16672 +/* tmpfs generation is too rough */
16673 +static inline int au_test_higen(struct inode *inode, struct inode *h_inode)
16674 +{
16675 +       struct au_iinfo *iinfo;
16676 +
16677 +       iinfo = au_ii(inode);
16678 +       AuRwMustAnyLock(&iinfo->ii_rwsem);
16679 +       return !(iinfo->ii_hsb1 == h_inode->i_sb
16680 +                && iinfo->ii_higen == h_inode->i_generation);
16681 +}
16682 +
16683 +static inline void au_iigen_dec(struct inode *inode)
16684 +{
16685 +       struct au_iinfo *iinfo;
16686 +
16687 +       iinfo = au_ii(inode);
16688 +       spin_lock(&iinfo->ii_genspin);
16689 +       iinfo->ii_generation.ig_generation--;
16690 +       spin_unlock(&iinfo->ii_genspin);
16691 +}
16692 +
16693 +static inline int au_iigen_test(struct inode *inode, unsigned int sigen)
16694 +{
16695 +       int err;
16696 +
16697 +       err = 0;
16698 +       if (unlikely(inode && au_iigen(inode, NULL) != sigen))
16699 +               err = -EIO;
16700 +
16701 +       return err;
16702 +}
16703 +
16704 +/* ---------------------------------------------------------------------- */
16705 +
16706 +static inline aufs_bindex_t au_ii_br_id(struct inode *inode,
16707 +                                       aufs_bindex_t bindex)
16708 +{
16709 +       IiMustAnyLock(inode);
16710 +       return au_ii(inode)->ii_hinode[0 + bindex].hi_id;
16711 +}
16712 +
16713 +static inline aufs_bindex_t au_ibstart(struct inode *inode)
16714 +{
16715 +       IiMustAnyLock(inode);
16716 +       return au_ii(inode)->ii_bstart;
16717 +}
16718 +
16719 +static inline aufs_bindex_t au_ibend(struct inode *inode)
16720 +{
16721 +       IiMustAnyLock(inode);
16722 +       return au_ii(inode)->ii_bend;
16723 +}
16724 +
16725 +static inline struct au_vdir *au_ivdir(struct inode *inode)
16726 +{
16727 +       IiMustAnyLock(inode);
16728 +       return au_ii(inode)->ii_vdir;
16729 +}
16730 +
16731 +static inline struct dentry *au_hi_wh(struct inode *inode, aufs_bindex_t bindex)
16732 +{
16733 +       IiMustAnyLock(inode);
16734 +       return au_ii(inode)->ii_hinode[0 + bindex].hi_whdentry;
16735 +}
16736 +
16737 +static inline void au_set_ibstart(struct inode *inode, aufs_bindex_t bindex)
16738 +{
16739 +       IiMustWriteLock(inode);
16740 +       au_ii(inode)->ii_bstart = bindex;
16741 +}
16742 +
16743 +static inline void au_set_ibend(struct inode *inode, aufs_bindex_t bindex)
16744 +{
16745 +       IiMustWriteLock(inode);
16746 +       au_ii(inode)->ii_bend = bindex;
16747 +}
16748 +
16749 +static inline void au_set_ivdir(struct inode *inode, struct au_vdir *vdir)
16750 +{
16751 +       IiMustWriteLock(inode);
16752 +       au_ii(inode)->ii_vdir = vdir;
16753 +}
16754 +
16755 +static inline struct au_hinode *au_hi(struct inode *inode, aufs_bindex_t bindex)
16756 +{
16757 +       IiMustAnyLock(inode);
16758 +       return au_ii(inode)->ii_hinode + bindex;
16759 +}
16760 +
16761 +/* ---------------------------------------------------------------------- */
16762 +
16763 +static inline struct dentry *au_pinned_parent(struct au_pin *pin)
16764 +{
16765 +       if (pin)
16766 +               return pin->parent;
16767 +       return NULL;
16768 +}
16769 +
16770 +static inline struct inode *au_pinned_h_dir(struct au_pin *pin)
16771 +{
16772 +       if (pin && pin->hdir)
16773 +               return pin->hdir->hi_inode;
16774 +       return NULL;
16775 +}
16776 +
16777 +static inline struct au_hinode *au_pinned_hdir(struct au_pin *pin)
16778 +{
16779 +       if (pin)
16780 +               return pin->hdir;
16781 +       return NULL;
16782 +}
16783 +
16784 +static inline void au_pin_set_dentry(struct au_pin *pin, struct dentry *dentry)
16785 +{
16786 +       if (pin)
16787 +               pin->dentry = dentry;
16788 +}
16789 +
16790 +static inline void au_pin_set_parent_lflag(struct au_pin *pin,
16791 +                                          unsigned char lflag)
16792 +{
16793 +       if (pin) {
16794 +               if (lflag)
16795 +                       au_fset_pin(pin->flags, DI_LOCKED);
16796 +               else
16797 +                       au_fclr_pin(pin->flags, DI_LOCKED);
16798 +       }
16799 +}
16800 +
16801 +static inline void au_pin_set_parent(struct au_pin *pin, struct dentry *parent)
16802 +{
16803 +       if (pin) {
16804 +               dput(pin->parent);
16805 +               pin->parent = dget(parent);
16806 +       }
16807 +}
16808 +
16809 +/* ---------------------------------------------------------------------- */
16810 +
16811 +struct au_branch;
16812 +#ifdef CONFIG_AUFS_HNOTIFY
16813 +struct au_hnotify_op {
16814 +       void (*ctl)(struct au_hinode *hinode, int do_set);
16815 +       int (*alloc)(struct au_hinode *hinode);
16816 +
16817 +       /*
16818 +        * if it returns true, the the caller should free hinode->hi_notify,
16819 +        * otherwise ->free() frees it.
16820 +        */
16821 +       int (*free)(struct au_hinode *hinode,
16822 +                   struct au_hnotify *hn) __must_check;
16823 +
16824 +       void (*fin)(void);
16825 +       int (*init)(void);
16826 +
16827 +       int (*reset_br)(unsigned int udba, struct au_branch *br, int perm);
16828 +       void (*fin_br)(struct au_branch *br);
16829 +       int (*init_br)(struct au_branch *br, int perm);
16830 +};
16831 +
16832 +/* hnotify.c */
16833 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode);
16834 +void au_hn_free(struct au_hinode *hinode);
16835 +void au_hn_ctl(struct au_hinode *hinode, int do_set);
16836 +void au_hn_reset(struct inode *inode, unsigned int flags);
16837 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
16838 +              struct qstr *h_child_qstr, struct inode *h_child_inode);
16839 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm);
16840 +int au_hnotify_init_br(struct au_branch *br, int perm);
16841 +void au_hnotify_fin_br(struct au_branch *br);
16842 +int __init au_hnotify_init(void);
16843 +void au_hnotify_fin(void);
16844 +
16845 +/* hfsnotify.c */
16846 +extern const struct au_hnotify_op au_hnotify_op;
16847 +
16848 +static inline
16849 +void au_hn_init(struct au_hinode *hinode)
16850 +{
16851 +       hinode->hi_notify = NULL;
16852 +}
16853 +
16854 +static inline struct au_hnotify *au_hn(struct au_hinode *hinode)
16855 +{
16856 +       return hinode->hi_notify;
16857 +}
16858 +
16859 +#else
16860 +AuStub(int, au_hn_alloc, return -EOPNOTSUPP,
16861 +       struct au_hinode *hinode __maybe_unused,
16862 +       struct inode *inode __maybe_unused)
16863 +AuStub(struct au_hnotify *, au_hn, return NULL, struct au_hinode *hinode)
16864 +AuStubVoid(au_hn_free, struct au_hinode *hinode __maybe_unused)
16865 +AuStubVoid(au_hn_ctl, struct au_hinode *hinode __maybe_unused,
16866 +          int do_set __maybe_unused)
16867 +AuStubVoid(au_hn_reset, struct inode *inode __maybe_unused,
16868 +          unsigned int flags __maybe_unused)
16869 +AuStubInt0(au_hnotify_reset_br, unsigned int udba __maybe_unused,
16870 +          struct au_branch *br __maybe_unused,
16871 +          int perm __maybe_unused)
16872 +AuStubInt0(au_hnotify_init_br, struct au_branch *br __maybe_unused,
16873 +          int perm __maybe_unused)
16874 +AuStubVoid(au_hnotify_fin_br, struct au_branch *br __maybe_unused)
16875 +AuStubInt0(__init au_hnotify_init, void)
16876 +AuStubVoid(au_hnotify_fin, void)
16877 +AuStubVoid(au_hn_init, struct au_hinode *hinode __maybe_unused)
16878 +#endif /* CONFIG_AUFS_HNOTIFY */
16879 +
16880 +static inline void au_hn_suspend(struct au_hinode *hdir)
16881 +{
16882 +       au_hn_ctl(hdir, /*do_set*/0);
16883 +}
16884 +
16885 +static inline void au_hn_resume(struct au_hinode *hdir)
16886 +{
16887 +       au_hn_ctl(hdir, /*do_set*/1);
16888 +}
16889 +
16890 +static inline void au_hn_imtx_lock(struct au_hinode *hdir)
16891 +{
16892 +       mutex_lock(&hdir->hi_inode->i_mutex);
16893 +       au_hn_suspend(hdir);
16894 +}
16895 +
16896 +static inline void au_hn_imtx_lock_nested(struct au_hinode *hdir,
16897 +                                         unsigned int sc __maybe_unused)
16898 +{
16899 +       mutex_lock_nested(&hdir->hi_inode->i_mutex, sc);
16900 +       au_hn_suspend(hdir);
16901 +}
16902 +
16903 +static inline void au_hn_imtx_unlock(struct au_hinode *hdir)
16904 +{
16905 +       au_hn_resume(hdir);
16906 +       mutex_unlock(&hdir->hi_inode->i_mutex);
16907 +}
16908 +
16909 +#endif /* __KERNEL__ */
16910 +#endif /* __AUFS_INODE_H__ */
16911 diff -urN /usr/share/empty/fs/aufs/ioctl.c linux/fs/aufs/ioctl.c
16912 --- /usr/share/empty/fs/aufs/ioctl.c    1970-01-01 01:00:00.000000000 +0100
16913 +++ linux/fs/aufs/ioctl.c       2015-01-25 13:00:38.631047076 +0100
16914 @@ -0,0 +1,219 @@
16915 +/*
16916 + * Copyright (C) 2005-2014 Junjiro R. Okajima
16917 + *
16918 + * This program, aufs is free software; you can redistribute it and/or modify
16919 + * it under the terms of the GNU General Public License as published by
16920 + * the Free Software Foundation; either version 2 of the License, or
16921 + * (at your option) any later version.
16922 + *
16923 + * This program is distributed in the hope that it will be useful,
16924 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16925 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16926 + * GNU General Public License for more details.
16927 + *
16928 + * You should have received a copy of the GNU General Public License
16929 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16930 + */
16931 +
16932 +/*
16933 + * ioctl
16934 + * plink-management and readdir in userspace.
16935 + * assist the pathconf(3) wrapper library.
16936 + * move-down
16937 + * File-based Hierarchical Storage Management.
16938 + */
16939 +
16940 +#include <linux/compat.h>
16941 +#include <linux/file.h>
16942 +#include "aufs.h"
16943 +
16944 +static int au_wbr_fd(struct path *path, struct aufs_wbr_fd __user *arg)
16945 +{
16946 +       int err, fd;
16947 +       aufs_bindex_t wbi, bindex, bend;
16948 +       struct file *h_file;
16949 +       struct super_block *sb;
16950 +       struct dentry *root;
16951 +       struct au_branch *br;
16952 +       struct aufs_wbr_fd wbrfd = {
16953 +               .oflags = au_dir_roflags,
16954 +               .brid   = -1
16955 +       };
16956 +       const int valid = O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_DIRECTORY
16957 +               | O_NOATIME | O_CLOEXEC;
16958 +
16959 +       AuDebugOn(wbrfd.oflags & ~valid);
16960 +
16961 +       if (arg) {
16962 +               err = copy_from_user(&wbrfd, arg, sizeof(wbrfd));
16963 +               if (unlikely(err)) {
16964 +                       err = -EFAULT;
16965 +                       goto out;
16966 +               }
16967 +
16968 +               err = -EINVAL;
16969 +               AuDbg("wbrfd{0%o, %d}\n", wbrfd.oflags, wbrfd.brid);
16970 +               wbrfd.oflags |= au_dir_roflags;
16971 +               AuDbg("0%o\n", wbrfd.oflags);
16972 +               if (unlikely(wbrfd.oflags & ~valid))
16973 +                       goto out;
16974 +       }
16975 +
16976 +       fd = get_unused_fd();
16977 +       err = fd;
16978 +       if (unlikely(fd < 0))
16979 +               goto out;
16980 +
16981 +       h_file = ERR_PTR(-EINVAL);
16982 +       wbi = 0;
16983 +       br = NULL;
16984 +       sb = path->dentry->d_sb;
16985 +       root = sb->s_root;
16986 +       aufs_read_lock(root, AuLock_IR);
16987 +       bend = au_sbend(sb);
16988 +       if (wbrfd.brid >= 0) {
16989 +               wbi = au_br_index(sb, wbrfd.brid);
16990 +               if (unlikely(wbi < 0 || wbi > bend))
16991 +                       goto out_unlock;
16992 +       }
16993 +
16994 +       h_file = ERR_PTR(-ENOENT);
16995 +       br = au_sbr(sb, wbi);
16996 +       if (!au_br_writable(br->br_perm)) {
16997 +               if (arg)
16998 +                       goto out_unlock;
16999 +
17000 +               bindex = wbi + 1;
17001 +               wbi = -1;
17002 +               for (; bindex <= bend; bindex++) {
17003 +                       br = au_sbr(sb, bindex);
17004 +                       if (au_br_writable(br->br_perm)) {
17005 +                               wbi = bindex;
17006 +                               br = au_sbr(sb, wbi);
17007 +                               break;
17008 +                       }
17009 +               }
17010 +       }
17011 +       AuDbg("wbi %d\n", wbi);
17012 +       if (wbi >= 0)
17013 +               h_file = au_h_open(root, wbi, wbrfd.oflags, NULL,
17014 +                                  /*force_wr*/0);
17015 +
17016 +out_unlock:
17017 +       aufs_read_unlock(root, AuLock_IR);
17018 +       err = PTR_ERR(h_file);
17019 +       if (IS_ERR(h_file))
17020 +               goto out_fd;
17021 +
17022 +       atomic_dec(&br->br_count); /* cf. au_h_open() */
17023 +       fd_install(fd, h_file);
17024 +       err = fd;
17025 +       goto out; /* success */
17026 +
17027 +out_fd:
17028 +       put_unused_fd(fd);
17029 +out:
17030 +       AuTraceErr(err);
17031 +       return err;
17032 +}
17033 +
17034 +/* ---------------------------------------------------------------------- */
17035 +
17036 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg)
17037 +{
17038 +       long err;
17039 +       struct dentry *dentry;
17040 +
17041 +       switch (cmd) {
17042 +       case AUFS_CTL_RDU:
17043 +       case AUFS_CTL_RDU_INO:
17044 +               err = au_rdu_ioctl(file, cmd, arg);
17045 +               break;
17046 +
17047 +       case AUFS_CTL_WBR_FD:
17048 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17049 +               break;
17050 +
17051 +       case AUFS_CTL_IBUSY:
17052 +               err = au_ibusy_ioctl(file, arg);
17053 +               break;
17054 +
17055 +       case AUFS_CTL_BRINFO:
17056 +               err = au_brinfo_ioctl(file, arg);
17057 +               break;
17058 +
17059 +       case AUFS_CTL_FHSM_FD:
17060 +               dentry = file->f_dentry;
17061 +               if (IS_ROOT(dentry))
17062 +                       err = au_fhsm_fd(dentry->d_sb, arg);
17063 +               else
17064 +                       err = -ENOTTY;
17065 +               break;
17066 +
17067 +       default:
17068 +               /* do not call the lower */
17069 +               AuDbg("0x%x\n", cmd);
17070 +               err = -ENOTTY;
17071 +       }
17072 +
17073 +       AuTraceErr(err);
17074 +       return err;
17075 +}
17076 +
17077 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg)
17078 +{
17079 +       long err;
17080 +
17081 +       switch (cmd) {
17082 +       case AUFS_CTL_MVDOWN:
17083 +               err = au_mvdown(file->f_dentry, (void __user *)arg);
17084 +               break;
17085 +
17086 +       case AUFS_CTL_WBR_FD:
17087 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17088 +               break;
17089 +
17090 +       default:
17091 +               /* do not call the lower */
17092 +               AuDbg("0x%x\n", cmd);
17093 +               err = -ENOTTY;
17094 +       }
17095 +
17096 +       AuTraceErr(err);
17097 +       return err;
17098 +}
17099 +
17100 +#ifdef CONFIG_COMPAT
17101 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
17102 +                          unsigned long arg)
17103 +{
17104 +       long err;
17105 +
17106 +       switch (cmd) {
17107 +       case AUFS_CTL_RDU:
17108 +       case AUFS_CTL_RDU_INO:
17109 +               err = au_rdu_compat_ioctl(file, cmd, arg);
17110 +               break;
17111 +
17112 +       case AUFS_CTL_IBUSY:
17113 +               err = au_ibusy_compat_ioctl(file, arg);
17114 +               break;
17115 +
17116 +       case AUFS_CTL_BRINFO:
17117 +               err = au_brinfo_compat_ioctl(file, arg);
17118 +               break;
17119 +
17120 +       default:
17121 +               err = aufs_ioctl_dir(file, cmd, arg);
17122 +       }
17123 +
17124 +       AuTraceErr(err);
17125 +       return err;
17126 +}
17127 +
17128 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
17129 +                             unsigned long arg)
17130 +{
17131 +       return aufs_ioctl_nondir(file, cmd, (unsigned long)compat_ptr(arg));
17132 +}
17133 +#endif
17134 diff -urN /usr/share/empty/fs/aufs/i_op_add.c linux/fs/aufs/i_op_add.c
17135 --- /usr/share/empty/fs/aufs/i_op_add.c 1970-01-01 01:00:00.000000000 +0100
17136 +++ linux/fs/aufs/i_op_add.c    2015-01-25 13:00:38.631047076 +0100
17137 @@ -0,0 +1,891 @@
17138 +/*
17139 + * Copyright (C) 2005-2014 Junjiro R. Okajima
17140 + *
17141 + * This program, aufs is free software; you can redistribute it and/or modify
17142 + * it under the terms of the GNU General Public License as published by
17143 + * the Free Software Foundation; either version 2 of the License, or
17144 + * (at your option) any later version.
17145 + *
17146 + * This program is distributed in the hope that it will be useful,
17147 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17148 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17149 + * GNU General Public License for more details.
17150 + *
17151 + * You should have received a copy of the GNU General Public License
17152 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17153 + */
17154 +
17155 +/*
17156 + * inode operations (add entry)
17157 + */
17158 +
17159 +#include "aufs.h"
17160 +
17161 +/*
17162 + * final procedure of adding a new entry, except link(2).
17163 + * remove whiteout, instantiate, copyup the parent dir's times and size
17164 + * and update version.
17165 + * if it failed, re-create the removed whiteout.
17166 + */
17167 +static int epilog(struct inode *dir, aufs_bindex_t bindex,
17168 +                 struct dentry *wh_dentry, struct dentry *dentry)
17169 +{
17170 +       int err, rerr;
17171 +       aufs_bindex_t bwh;
17172 +       struct path h_path;
17173 +       struct super_block *sb;
17174 +       struct inode *inode, *h_dir;
17175 +       struct dentry *wh;
17176 +
17177 +       bwh = -1;
17178 +       sb = dir->i_sb;
17179 +       if (wh_dentry) {
17180 +               h_dir = wh_dentry->d_parent->d_inode; /* dir inode is locked */
17181 +               IMustLock(h_dir);
17182 +               AuDebugOn(au_h_iptr(dir, bindex) != h_dir);
17183 +               bwh = au_dbwh(dentry);
17184 +               h_path.dentry = wh_dentry;
17185 +               h_path.mnt = au_sbr_mnt(sb, bindex);
17186 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path,
17187 +                                         dentry);
17188 +               if (unlikely(err))
17189 +                       goto out;
17190 +       }
17191 +
17192 +       inode = au_new_inode(dentry, /*must_new*/1);
17193 +       if (!IS_ERR(inode)) {
17194 +               d_instantiate(dentry, inode);
17195 +               dir = dentry->d_parent->d_inode; /* dir inode is locked */
17196 +               IMustLock(dir);
17197 +               if (au_ibstart(dir) == au_dbstart(dentry))
17198 +                       au_cpup_attr_timesizes(dir);
17199 +               dir->i_version++;
17200 +               au_fhsm_wrote(sb, bindex, /*force*/0);
17201 +               return 0; /* success */
17202 +       }
17203 +
17204 +       err = PTR_ERR(inode);
17205 +       if (!wh_dentry)
17206 +               goto out;
17207 +
17208 +       /* revert */
17209 +       /* dir inode is locked */
17210 +       wh = au_wh_create(dentry, bwh, wh_dentry->d_parent);
17211 +       rerr = PTR_ERR(wh);
17212 +       if (IS_ERR(wh)) {
17213 +               AuIOErr("%pd reverting whiteout failed(%d, %d)\n",
17214 +                       dentry, err, rerr);
17215 +               err = -EIO;
17216 +       } else
17217 +               dput(wh);
17218 +
17219 +out:
17220 +       return err;
17221 +}
17222 +
17223 +static int au_d_may_add(struct dentry *dentry)
17224 +{
17225 +       int err;
17226 +
17227 +       err = 0;
17228 +       if (unlikely(d_unhashed(dentry)))
17229 +               err = -ENOENT;
17230 +       if (unlikely(dentry->d_inode))
17231 +               err = -EEXIST;
17232 +       return err;
17233 +}
17234 +
17235 +/*
17236 + * simple tests for the adding inode operations.
17237 + * following the checks in vfs, plus the parent-child relationship.
17238 + */
17239 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
17240 +              struct dentry *h_parent, int isdir)
17241 +{
17242 +       int err;
17243 +       umode_t h_mode;
17244 +       struct dentry *h_dentry;
17245 +       struct inode *h_inode;
17246 +
17247 +       err = -ENAMETOOLONG;
17248 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17249 +               goto out;
17250 +
17251 +       h_dentry = au_h_dptr(dentry, bindex);
17252 +       h_inode = h_dentry->d_inode;
17253 +       if (!dentry->d_inode) {
17254 +               err = -EEXIST;
17255 +               if (unlikely(h_inode))
17256 +                       goto out;
17257 +       } else {
17258 +               /* rename(2) case */
17259 +               err = -EIO;
17260 +               if (unlikely(!h_inode || !h_inode->i_nlink))
17261 +                       goto out;
17262 +
17263 +               h_mode = h_inode->i_mode;
17264 +               if (!isdir) {
17265 +                       err = -EISDIR;
17266 +                       if (unlikely(S_ISDIR(h_mode)))
17267 +                               goto out;
17268 +               } else if (unlikely(!S_ISDIR(h_mode))) {
17269 +                       err = -ENOTDIR;
17270 +                       goto out;
17271 +               }
17272 +       }
17273 +
17274 +       err = 0;
17275 +       /* expected parent dir is locked */
17276 +       if (unlikely(h_parent != h_dentry->d_parent))
17277 +               err = -EIO;
17278 +
17279 +out:
17280 +       AuTraceErr(err);
17281 +       return err;
17282 +}
17283 +
17284 +/*
17285 + * initial procedure of adding a new entry.
17286 + * prepare writable branch and the parent dir, lock it,
17287 + * and lookup whiteout for the new entry.
17288 + */
17289 +static struct dentry*
17290 +lock_hdir_lkup_wh(struct dentry *dentry, struct au_dtime *dt,
17291 +                 struct dentry *src_dentry, struct au_pin *pin,
17292 +                 struct au_wr_dir_args *wr_dir_args)
17293 +{
17294 +       struct dentry *wh_dentry, *h_parent;
17295 +       struct super_block *sb;
17296 +       struct au_branch *br;
17297 +       int err;
17298 +       unsigned int udba;
17299 +       aufs_bindex_t bcpup;
17300 +
17301 +       AuDbg("%pd\n", dentry);
17302 +
17303 +       err = au_wr_dir(dentry, src_dentry, wr_dir_args);
17304 +       bcpup = err;
17305 +       wh_dentry = ERR_PTR(err);
17306 +       if (unlikely(err < 0))
17307 +               goto out;
17308 +
17309 +       sb = dentry->d_sb;
17310 +       udba = au_opt_udba(sb);
17311 +       err = au_pin(pin, dentry, bcpup, udba,
17312 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17313 +       wh_dentry = ERR_PTR(err);
17314 +       if (unlikely(err))
17315 +               goto out;
17316 +
17317 +       h_parent = au_pinned_h_parent(pin);
17318 +       if (udba != AuOpt_UDBA_NONE
17319 +           && au_dbstart(dentry) == bcpup)
17320 +               err = au_may_add(dentry, bcpup, h_parent,
17321 +                                au_ftest_wrdir(wr_dir_args->flags, ISDIR));
17322 +       else if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17323 +               err = -ENAMETOOLONG;
17324 +       wh_dentry = ERR_PTR(err);
17325 +       if (unlikely(err))
17326 +               goto out_unpin;
17327 +
17328 +       br = au_sbr(sb, bcpup);
17329 +       if (dt) {
17330 +               struct path tmp = {
17331 +                       .dentry = h_parent,
17332 +                       .mnt    = au_br_mnt(br)
17333 +               };
17334 +               au_dtime_store(dt, au_pinned_parent(pin), &tmp);
17335 +       }
17336 +
17337 +       wh_dentry = NULL;
17338 +       if (bcpup != au_dbwh(dentry))
17339 +               goto out; /* success */
17340 +
17341 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
17342 +
17343 +out_unpin:
17344 +       if (IS_ERR(wh_dentry))
17345 +               au_unpin(pin);
17346 +out:
17347 +       return wh_dentry;
17348 +}
17349 +
17350 +/* ---------------------------------------------------------------------- */
17351 +
17352 +enum { Mknod, Symlink, Creat };
17353 +struct simple_arg {
17354 +       int type;
17355 +       union {
17356 +               struct {
17357 +                       umode_t mode;
17358 +                       bool want_excl;
17359 +               } c;
17360 +               struct {
17361 +                       const char *symname;
17362 +               } s;
17363 +               struct {
17364 +                       umode_t mode;
17365 +                       dev_t dev;
17366 +               } m;
17367 +       } u;
17368 +};
17369 +
17370 +static int add_simple(struct inode *dir, struct dentry *dentry,
17371 +                     struct simple_arg *arg)
17372 +{
17373 +       int err, rerr;
17374 +       aufs_bindex_t bstart;
17375 +       unsigned char created;
17376 +       struct dentry *wh_dentry, *parent;
17377 +       struct inode *h_dir;
17378 +       /* to reuduce stack size */
17379 +       struct {
17380 +               struct au_dtime dt;
17381 +               struct au_pin pin;
17382 +               struct path h_path;
17383 +               struct au_wr_dir_args wr_dir_args;
17384 +       } *a;
17385 +
17386 +       AuDbg("%pd\n", dentry);
17387 +       IMustLock(dir);
17388 +
17389 +       err = -ENOMEM;
17390 +       a = kmalloc(sizeof(*a), GFP_NOFS);
17391 +       if (unlikely(!a))
17392 +               goto out;
17393 +       a->wr_dir_args.force_btgt = -1;
17394 +       a->wr_dir_args.flags = AuWrDir_ADD_ENTRY;
17395 +
17396 +       parent = dentry->d_parent; /* dir inode is locked */
17397 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
17398 +       if (unlikely(err))
17399 +               goto out_free;
17400 +       err = au_d_may_add(dentry);
17401 +       if (unlikely(err))
17402 +               goto out_unlock;
17403 +       di_write_lock_parent(parent);
17404 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
17405 +                                     &a->pin, &a->wr_dir_args);
17406 +       err = PTR_ERR(wh_dentry);
17407 +       if (IS_ERR(wh_dentry))
17408 +               goto out_parent;
17409 +
17410 +       bstart = au_dbstart(dentry);
17411 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
17412 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, bstart);
17413 +       h_dir = au_pinned_h_dir(&a->pin);
17414 +       switch (arg->type) {
17415 +       case Creat:
17416 +               err = vfsub_create(h_dir, &a->h_path, arg->u.c.mode,
17417 +                                  arg->u.c.want_excl);
17418 +               break;
17419 +       case Symlink:
17420 +               err = vfsub_symlink(h_dir, &a->h_path, arg->u.s.symname);
17421 +               break;
17422 +       case Mknod:
17423 +               err = vfsub_mknod(h_dir, &a->h_path, arg->u.m.mode,
17424 +                                 arg->u.m.dev);
17425 +               break;
17426 +       default:
17427 +               BUG();
17428 +       }
17429 +       created = !err;
17430 +       if (!err)
17431 +               err = epilog(dir, bstart, wh_dentry, dentry);
17432 +
17433 +       /* revert */
17434 +       if (unlikely(created && err && a->h_path.dentry->d_inode)) {
17435 +               /* no delegation since it is just created */
17436 +               rerr = vfsub_unlink(h_dir, &a->h_path, /*delegated*/NULL,
17437 +                                   /*force*/0);
17438 +               if (rerr) {
17439 +                       AuIOErr("%pd revert failure(%d, %d)\n",
17440 +                               dentry, err, rerr);
17441 +                       err = -EIO;
17442 +               }
17443 +               au_dtime_revert(&a->dt);
17444 +       }
17445 +
17446 +       au_unpin(&a->pin);
17447 +       dput(wh_dentry);
17448 +
17449 +out_parent:
17450 +       di_write_unlock(parent);
17451 +out_unlock:
17452 +       if (unlikely(err)) {
17453 +               au_update_dbstart(dentry);
17454 +               d_drop(dentry);
17455 +       }
17456 +       aufs_read_unlock(dentry, AuLock_DW);
17457 +out_free:
17458 +       kfree(a);
17459 +out:
17460 +       return err;
17461 +}
17462 +
17463 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
17464 +              dev_t dev)
17465 +{
17466 +       struct simple_arg arg = {
17467 +               .type = Mknod,
17468 +               .u.m = {
17469 +                       .mode   = mode,
17470 +                       .dev    = dev
17471 +               }
17472 +       };
17473 +       return add_simple(dir, dentry, &arg);
17474 +}
17475 +
17476 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
17477 +{
17478 +       struct simple_arg arg = {
17479 +               .type = Symlink,
17480 +               .u.s.symname = symname
17481 +       };
17482 +       return add_simple(dir, dentry, &arg);
17483 +}
17484 +
17485 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
17486 +               bool want_excl)
17487 +{
17488 +       struct simple_arg arg = {
17489 +               .type = Creat,
17490 +               .u.c = {
17491 +                       .mode           = mode,
17492 +                       .want_excl      = want_excl
17493 +               }
17494 +       };
17495 +       return add_simple(dir, dentry, &arg);
17496 +}
17497 +
17498 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
17499 +{
17500 +       int err;
17501 +       aufs_bindex_t bindex;
17502 +       struct super_block *sb;
17503 +       struct dentry *parent, *h_parent, *h_dentry;
17504 +       struct inode *h_dir, *inode;
17505 +       struct vfsmount *h_mnt;
17506 +       struct au_wr_dir_args wr_dir_args = {
17507 +               .force_btgt     = -1,
17508 +               .flags          = AuWrDir_TMPFILE
17509 +       };
17510 +
17511 +       /* copy-up may happen */
17512 +       mutex_lock(&dir->i_mutex);
17513 +
17514 +       sb = dir->i_sb;
17515 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
17516 +       if (unlikely(err))
17517 +               goto out;
17518 +
17519 +       err = au_di_init(dentry);
17520 +       if (unlikely(err))
17521 +               goto out_si;
17522 +
17523 +       err = -EBUSY;
17524 +       parent = d_find_any_alias(dir);
17525 +       AuDebugOn(!parent);
17526 +       di_write_lock_parent(parent);
17527 +       if (unlikely(parent->d_inode != dir))
17528 +               goto out_parent;
17529 +
17530 +       err = au_digen_test(parent, au_sigen(sb));
17531 +       if (unlikely(err))
17532 +               goto out_parent;
17533 +
17534 +       bindex = au_dbstart(parent);
17535 +       au_set_dbstart(dentry, bindex);
17536 +       au_set_dbend(dentry, bindex);
17537 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
17538 +       bindex = err;
17539 +       if (unlikely(err < 0))
17540 +               goto out_parent;
17541 +
17542 +       err = -EOPNOTSUPP;
17543 +       h_dir = au_h_iptr(dir, bindex);
17544 +       if (unlikely(!h_dir->i_op->tmpfile))
17545 +               goto out_parent;
17546 +
17547 +       h_mnt = au_sbr_mnt(sb, bindex);
17548 +       err = vfsub_mnt_want_write(h_mnt);
17549 +       if (unlikely(err))
17550 +               goto out_parent;
17551 +
17552 +       h_parent = au_h_dptr(parent, bindex);
17553 +       err = inode_permission(h_parent->d_inode, MAY_WRITE | MAY_EXEC);
17554 +       if (unlikely(err))
17555 +               goto out_mnt;
17556 +
17557 +       err = -ENOMEM;
17558 +       h_dentry = d_alloc(h_parent, &dentry->d_name);
17559 +       if (unlikely(!h_dentry))
17560 +               goto out_mnt;
17561 +
17562 +       err = h_dir->i_op->tmpfile(h_dir, h_dentry, mode);
17563 +       if (unlikely(err))
17564 +               goto out_dentry;
17565 +
17566 +       au_set_dbstart(dentry, bindex);
17567 +       au_set_dbend(dentry, bindex);
17568 +       au_set_h_dptr(dentry, bindex, dget(h_dentry));
17569 +       inode = au_new_inode(dentry, /*must_new*/1);
17570 +       if (IS_ERR(inode)) {
17571 +               err = PTR_ERR(inode);
17572 +               au_set_h_dptr(dentry, bindex, NULL);
17573 +               au_set_dbstart(dentry, -1);
17574 +               au_set_dbend(dentry, -1);
17575 +       } else {
17576 +               if (!inode->i_nlink)
17577 +                       set_nlink(inode, 1);
17578 +               d_tmpfile(dentry, inode);
17579 +               au_di(dentry)->di_tmpfile = 1;
17580 +
17581 +               /* update without i_mutex */
17582 +               if (au_ibstart(dir) == au_dbstart(dentry))
17583 +                       au_cpup_attr_timesizes(dir);
17584 +       }
17585 +
17586 +out_dentry:
17587 +       dput(h_dentry);
17588 +out_mnt:
17589 +       vfsub_mnt_drop_write(h_mnt);
17590 +out_parent:
17591 +       di_write_unlock(parent);
17592 +       dput(parent);
17593 +       di_write_unlock(dentry);
17594 +       if (!err)
17595 +#if 0
17596 +               /* verbose coding for lock class name */
17597 +               au_rw_class(&au_di(dentry)->di_rwsem,
17598 +                           au_lc_key + AuLcNonDir_DIINFO);
17599 +#else
17600 +               ;
17601 +#endif
17602 +       else {
17603 +               au_di_fin(dentry);
17604 +               dentry->d_fsdata = NULL;
17605 +       }
17606 +out_si:
17607 +       si_read_unlock(sb);
17608 +out:
17609 +       mutex_unlock(&dir->i_mutex);
17610 +       return err;
17611 +}
17612 +
17613 +/* ---------------------------------------------------------------------- */
17614 +
17615 +struct au_link_args {
17616 +       aufs_bindex_t bdst, bsrc;
17617 +       struct au_pin pin;
17618 +       struct path h_path;
17619 +       struct dentry *src_parent, *parent;
17620 +};
17621 +
17622 +static int au_cpup_before_link(struct dentry *src_dentry,
17623 +                              struct au_link_args *a)
17624 +{
17625 +       int err;
17626 +       struct dentry *h_src_dentry;
17627 +       struct au_cp_generic cpg = {
17628 +               .dentry = src_dentry,
17629 +               .bdst   = a->bdst,
17630 +               .bsrc   = a->bsrc,
17631 +               .len    = -1,
17632 +               .pin    = &a->pin,
17633 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN /* | AuCpup_KEEPLINO */
17634 +       };
17635 +
17636 +       di_read_lock_parent(a->src_parent, AuLock_IR);
17637 +       err = au_test_and_cpup_dirs(src_dentry, a->bdst);
17638 +       if (unlikely(err))
17639 +               goto out;
17640 +
17641 +       h_src_dentry = au_h_dptr(src_dentry, a->bsrc);
17642 +       err = au_pin(&a->pin, src_dentry, a->bdst,
17643 +                    au_opt_udba(src_dentry->d_sb),
17644 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17645 +       if (unlikely(err))
17646 +               goto out;
17647 +
17648 +       err = au_sio_cpup_simple(&cpg);
17649 +       au_unpin(&a->pin);
17650 +
17651 +out:
17652 +       di_read_unlock(a->src_parent, AuLock_IR);
17653 +       return err;
17654 +}
17655 +
17656 +static int au_cpup_or_link(struct dentry *src_dentry, struct dentry *dentry,
17657 +                          struct au_link_args *a)
17658 +{
17659 +       int err;
17660 +       unsigned char plink;
17661 +       aufs_bindex_t bend;
17662 +       struct dentry *h_src_dentry;
17663 +       struct inode *h_inode, *inode, *delegated;
17664 +       struct super_block *sb;
17665 +       struct file *h_file;
17666 +
17667 +       plink = 0;
17668 +       h_inode = NULL;
17669 +       sb = src_dentry->d_sb;
17670 +       inode = src_dentry->d_inode;
17671 +       if (au_ibstart(inode) <= a->bdst)
17672 +               h_inode = au_h_iptr(inode, a->bdst);
17673 +       if (!h_inode || !h_inode->i_nlink) {
17674 +               /* copyup src_dentry as the name of dentry. */
17675 +               bend = au_dbend(dentry);
17676 +               if (bend < a->bsrc)
17677 +                       au_set_dbend(dentry, a->bsrc);
17678 +               au_set_h_dptr(dentry, a->bsrc,
17679 +                             dget(au_h_dptr(src_dentry, a->bsrc)));
17680 +               dget(a->h_path.dentry);
17681 +               au_set_h_dptr(dentry, a->bdst, NULL);
17682 +               AuDbg("temporary d_inode...\n");
17683 +               spin_lock(&dentry->d_lock);
17684 +               dentry->d_inode = src_dentry->d_inode; /* tmp */
17685 +               spin_unlock(&dentry->d_lock);
17686 +               h_file = au_h_open_pre(dentry, a->bsrc, /*force_wr*/0);
17687 +               if (IS_ERR(h_file))
17688 +                       err = PTR_ERR(h_file);
17689 +               else {
17690 +                       struct au_cp_generic cpg = {
17691 +                               .dentry = dentry,
17692 +                               .bdst   = a->bdst,
17693 +                               .bsrc   = -1,
17694 +                               .len    = -1,
17695 +                               .pin    = &a->pin,
17696 +                               .flags  = AuCpup_KEEPLINO
17697 +                       };
17698 +                       err = au_sio_cpup_simple(&cpg);
17699 +                       au_h_open_post(dentry, a->bsrc, h_file);
17700 +                       if (!err) {
17701 +                               dput(a->h_path.dentry);
17702 +                               a->h_path.dentry = au_h_dptr(dentry, a->bdst);
17703 +                       } else
17704 +                               au_set_h_dptr(dentry, a->bdst,
17705 +                                             a->h_path.dentry);
17706 +               }
17707 +               spin_lock(&dentry->d_lock);
17708 +               dentry->d_inode = NULL; /* restore */
17709 +               spin_unlock(&dentry->d_lock);
17710 +               AuDbg("temporary d_inode...done\n");
17711 +               au_set_h_dptr(dentry, a->bsrc, NULL);
17712 +               au_set_dbend(dentry, bend);
17713 +       } else {
17714 +               /* the inode of src_dentry already exists on a.bdst branch */
17715 +               h_src_dentry = d_find_alias(h_inode);
17716 +               if (!h_src_dentry && au_plink_test(inode)) {
17717 +                       plink = 1;
17718 +                       h_src_dentry = au_plink_lkup(inode, a->bdst);
17719 +                       err = PTR_ERR(h_src_dentry);
17720 +                       if (IS_ERR(h_src_dentry))
17721 +                               goto out;
17722 +
17723 +                       if (unlikely(!h_src_dentry->d_inode)) {
17724 +                               dput(h_src_dentry);
17725 +                               h_src_dentry = NULL;
17726 +                       }
17727 +
17728 +               }
17729 +               if (h_src_dentry) {
17730 +                       delegated = NULL;
17731 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
17732 +                                        &a->h_path, &delegated);
17733 +                       if (unlikely(err == -EWOULDBLOCK)) {
17734 +                               pr_warn("cannot retry for NFSv4 delegation"
17735 +                                       " for an internal link\n");
17736 +                               iput(delegated);
17737 +                       }
17738 +                       dput(h_src_dentry);
17739 +               } else {
17740 +                       AuIOErr("no dentry found for hi%lu on b%d\n",
17741 +                               h_inode->i_ino, a->bdst);
17742 +                       err = -EIO;
17743 +               }
17744 +       }
17745 +
17746 +       if (!err && !plink)
17747 +               au_plink_append(inode, a->bdst, a->h_path.dentry);
17748 +
17749 +out:
17750 +       AuTraceErr(err);
17751 +       return err;
17752 +}
17753 +
17754 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
17755 +             struct dentry *dentry)
17756 +{
17757 +       int err, rerr;
17758 +       struct au_dtime dt;
17759 +       struct au_link_args *a;
17760 +       struct dentry *wh_dentry, *h_src_dentry;
17761 +       struct inode *inode, *delegated;
17762 +       struct super_block *sb;
17763 +       struct au_wr_dir_args wr_dir_args = {
17764 +               /* .force_btgt  = -1, */
17765 +               .flags          = AuWrDir_ADD_ENTRY
17766 +       };
17767 +
17768 +       IMustLock(dir);
17769 +       inode = src_dentry->d_inode;
17770 +       IMustLock(inode);
17771 +
17772 +       err = -ENOMEM;
17773 +       a = kzalloc(sizeof(*a), GFP_NOFS);
17774 +       if (unlikely(!a))
17775 +               goto out;
17776 +
17777 +       a->parent = dentry->d_parent; /* dir inode is locked */
17778 +       err = aufs_read_and_write_lock2(dentry, src_dentry,
17779 +                                       AuLock_NOPLM | AuLock_GEN);
17780 +       if (unlikely(err))
17781 +               goto out_kfree;
17782 +       err = au_d_linkable(src_dentry);
17783 +       if (unlikely(err))
17784 +               goto out_unlock;
17785 +       err = au_d_may_add(dentry);
17786 +       if (unlikely(err))
17787 +               goto out_unlock;
17788 +
17789 +       a->src_parent = dget_parent(src_dentry);
17790 +       wr_dir_args.force_btgt = au_ibstart(inode);
17791 +
17792 +       di_write_lock_parent(a->parent);
17793 +       wr_dir_args.force_btgt = au_wbr(dentry, wr_dir_args.force_btgt);
17794 +       wh_dentry = lock_hdir_lkup_wh(dentry, &dt, src_dentry, &a->pin,
17795 +                                     &wr_dir_args);
17796 +       err = PTR_ERR(wh_dentry);
17797 +       if (IS_ERR(wh_dentry))
17798 +               goto out_parent;
17799 +
17800 +       err = 0;
17801 +       sb = dentry->d_sb;
17802 +       a->bdst = au_dbstart(dentry);
17803 +       a->h_path.dentry = au_h_dptr(dentry, a->bdst);
17804 +       a->h_path.mnt = au_sbr_mnt(sb, a->bdst);
17805 +       a->bsrc = au_ibstart(inode);
17806 +       h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
17807 +       if (!h_src_dentry && au_di(src_dentry)->di_tmpfile)
17808 +               h_src_dentry = dget(au_hi_wh(inode, a->bsrc));
17809 +       if (!h_src_dentry) {
17810 +               a->bsrc = au_dbstart(src_dentry);
17811 +               h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
17812 +               AuDebugOn(!h_src_dentry);
17813 +       } else if (IS_ERR(h_src_dentry)) {
17814 +               err = PTR_ERR(h_src_dentry);
17815 +               goto out_parent;
17816 +       }
17817 +
17818 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
17819 +               if (a->bdst < a->bsrc
17820 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */)
17821 +                       err = au_cpup_or_link(src_dentry, dentry, a);
17822 +               else {
17823 +                       delegated = NULL;
17824 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
17825 +                                        &a->h_path, &delegated);
17826 +                       if (unlikely(err == -EWOULDBLOCK)) {
17827 +                               pr_warn("cannot retry for NFSv4 delegation"
17828 +                                       " for an internal link\n");
17829 +                               iput(delegated);
17830 +                       }
17831 +               }
17832 +               dput(h_src_dentry);
17833 +       } else {
17834 +               /*
17835 +                * copyup src_dentry to the branch we process,
17836 +                * and then link(2) to it.
17837 +                */
17838 +               dput(h_src_dentry);
17839 +               if (a->bdst < a->bsrc
17840 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */) {
17841 +                       au_unpin(&a->pin);
17842 +                       di_write_unlock(a->parent);
17843 +                       err = au_cpup_before_link(src_dentry, a);
17844 +                       di_write_lock_parent(a->parent);
17845 +                       if (!err)
17846 +                               err = au_pin(&a->pin, dentry, a->bdst,
17847 +                                            au_opt_udba(sb),
17848 +                                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17849 +                       if (unlikely(err))
17850 +                               goto out_wh;
17851 +               }
17852 +               if (!err) {
17853 +                       h_src_dentry = au_h_dptr(src_dentry, a->bdst);
17854 +                       err = -ENOENT;
17855 +                       if (h_src_dentry && h_src_dentry->d_inode) {
17856 +                               delegated = NULL;
17857 +                               err = vfsub_link(h_src_dentry,
17858 +                                                au_pinned_h_dir(&a->pin),
17859 +                                                &a->h_path, &delegated);
17860 +                               if (unlikely(err == -EWOULDBLOCK)) {
17861 +                                       pr_warn("cannot retry"
17862 +                                               " for NFSv4 delegation"
17863 +                                               " for an internal link\n");
17864 +                                       iput(delegated);
17865 +                               }
17866 +                       }
17867 +               }
17868 +       }
17869 +       if (unlikely(err))
17870 +               goto out_unpin;
17871 +
17872 +       if (wh_dentry) {
17873 +               a->h_path.dentry = wh_dentry;
17874 +               err = au_wh_unlink_dentry(au_pinned_h_dir(&a->pin), &a->h_path,
17875 +                                         dentry);
17876 +               if (unlikely(err))
17877 +                       goto out_revert;
17878 +       }
17879 +
17880 +       dir->i_version++;
17881 +       if (au_ibstart(dir) == au_dbstart(dentry))
17882 +               au_cpup_attr_timesizes(dir);
17883 +       inc_nlink(inode);
17884 +       inode->i_ctime = dir->i_ctime;
17885 +       d_instantiate(dentry, au_igrab(inode));
17886 +       if (d_unhashed(a->h_path.dentry))
17887 +               /* some filesystem calls d_drop() */
17888 +               d_drop(dentry);
17889 +       /* some filesystems consume an inode even hardlink */
17890 +       au_fhsm_wrote(sb, a->bdst, /*force*/0);
17891 +       goto out_unpin; /* success */
17892 +
17893 +out_revert:
17894 +       /* no delegation since it is just created */
17895 +       rerr = vfsub_unlink(au_pinned_h_dir(&a->pin), &a->h_path,
17896 +                           /*delegated*/NULL, /*force*/0);
17897 +       if (unlikely(rerr)) {
17898 +               AuIOErr("%pd reverting failed(%d, %d)\n", dentry, err, rerr);
17899 +               err = -EIO;
17900 +       }
17901 +       au_dtime_revert(&dt);
17902 +out_unpin:
17903 +       au_unpin(&a->pin);
17904 +out_wh:
17905 +       dput(wh_dentry);
17906 +out_parent:
17907 +       di_write_unlock(a->parent);
17908 +       dput(a->src_parent);
17909 +out_unlock:
17910 +       if (unlikely(err)) {
17911 +               au_update_dbstart(dentry);
17912 +               d_drop(dentry);
17913 +       }
17914 +       aufs_read_and_write_unlock2(dentry, src_dentry);
17915 +out_kfree:
17916 +       kfree(a);
17917 +out:
17918 +       AuTraceErr(err);
17919 +       return err;
17920 +}
17921 +
17922 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
17923 +{
17924 +       int err, rerr;
17925 +       aufs_bindex_t bindex;
17926 +       unsigned char diropq;
17927 +       struct path h_path;
17928 +       struct dentry *wh_dentry, *parent, *opq_dentry;
17929 +       struct mutex *h_mtx;
17930 +       struct super_block *sb;
17931 +       struct {
17932 +               struct au_pin pin;
17933 +               struct au_dtime dt;
17934 +       } *a; /* reduce the stack usage */
17935 +       struct au_wr_dir_args wr_dir_args = {
17936 +               .force_btgt     = -1,
17937 +               .flags          = AuWrDir_ADD_ENTRY | AuWrDir_ISDIR
17938 +       };
17939 +
17940 +       IMustLock(dir);
17941 +
17942 +       err = -ENOMEM;
17943 +       a = kmalloc(sizeof(*a), GFP_NOFS);
17944 +       if (unlikely(!a))
17945 +               goto out;
17946 +
17947 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
17948 +       if (unlikely(err))
17949 +               goto out_free;
17950 +       err = au_d_may_add(dentry);
17951 +       if (unlikely(err))
17952 +               goto out_unlock;
17953 +
17954 +       parent = dentry->d_parent; /* dir inode is locked */
17955 +       di_write_lock_parent(parent);
17956 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
17957 +                                     &a->pin, &wr_dir_args);
17958 +       err = PTR_ERR(wh_dentry);
17959 +       if (IS_ERR(wh_dentry))
17960 +               goto out_parent;
17961 +
17962 +       sb = dentry->d_sb;
17963 +       bindex = au_dbstart(dentry);
17964 +       h_path.dentry = au_h_dptr(dentry, bindex);
17965 +       h_path.mnt = au_sbr_mnt(sb, bindex);
17966 +       err = vfsub_mkdir(au_pinned_h_dir(&a->pin), &h_path, mode);
17967 +       if (unlikely(err))
17968 +               goto out_unpin;
17969 +
17970 +       /* make the dir opaque */
17971 +       diropq = 0;
17972 +       h_mtx = &h_path.dentry->d_inode->i_mutex;
17973 +       if (wh_dentry
17974 +           || au_opt_test(au_mntflags(sb), ALWAYS_DIROPQ)) {
17975 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
17976 +               opq_dentry = au_diropq_create(dentry, bindex);
17977 +               mutex_unlock(h_mtx);
17978 +               err = PTR_ERR(opq_dentry);
17979 +               if (IS_ERR(opq_dentry))
17980 +                       goto out_dir;
17981 +               dput(opq_dentry);
17982 +               diropq = 1;
17983 +       }
17984 +
17985 +       err = epilog(dir, bindex, wh_dentry, dentry);
17986 +       if (!err) {
17987 +               inc_nlink(dir);
17988 +               goto out_unpin; /* success */
17989 +       }
17990 +
17991 +       /* revert */
17992 +       if (diropq) {
17993 +               AuLabel(revert opq);
17994 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
17995 +               rerr = au_diropq_remove(dentry, bindex);
17996 +               mutex_unlock(h_mtx);
17997 +               if (rerr) {
17998 +                       AuIOErr("%pd reverting diropq failed(%d, %d)\n",
17999 +                               dentry, err, rerr);
18000 +                       err = -EIO;
18001 +               }
18002 +       }
18003 +
18004 +out_dir:
18005 +       AuLabel(revert dir);
18006 +       rerr = vfsub_rmdir(au_pinned_h_dir(&a->pin), &h_path);
18007 +       if (rerr) {
18008 +               AuIOErr("%pd reverting dir failed(%d, %d)\n",
18009 +                       dentry, err, rerr);
18010 +               err = -EIO;
18011 +       }
18012 +       au_dtime_revert(&a->dt);
18013 +out_unpin:
18014 +       au_unpin(&a->pin);
18015 +       dput(wh_dentry);
18016 +out_parent:
18017 +       di_write_unlock(parent);
18018 +out_unlock:
18019 +       if (unlikely(err)) {
18020 +               au_update_dbstart(dentry);
18021 +               d_drop(dentry);
18022 +       }
18023 +       aufs_read_unlock(dentry, AuLock_DW);
18024 +out_free:
18025 +       kfree(a);
18026 +out:
18027 +       return err;
18028 +}
18029 diff -urN /usr/share/empty/fs/aufs/i_op.c linux/fs/aufs/i_op.c
18030 --- /usr/share/empty/fs/aufs/i_op.c     1970-01-01 01:00:00.000000000 +0100
18031 +++ linux/fs/aufs/i_op.c        2015-01-25 13:00:38.631047076 +0100
18032 @@ -0,0 +1,1306 @@
18033 +/*
18034 + * Copyright (C) 2005-2014 Junjiro R. Okajima
18035 + *
18036 + * This program, aufs is free software; you can redistribute it and/or modify
18037 + * it under the terms of the GNU General Public License as published by
18038 + * the Free Software Foundation; either version 2 of the License, or
18039 + * (at your option) any later version.
18040 + *
18041 + * This program is distributed in the hope that it will be useful,
18042 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18043 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18044 + * GNU General Public License for more details.
18045 + *
18046 + * You should have received a copy of the GNU General Public License
18047 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18048 + */
18049 +
18050 +/*
18051 + * inode operations (except add/del/rename)
18052 + */
18053 +
18054 +#include <linux/device_cgroup.h>
18055 +#include <linux/fs_stack.h>
18056 +#include <linux/mm.h>
18057 +#include <linux/namei.h>
18058 +#include <linux/security.h>
18059 +#include "aufs.h"
18060 +
18061 +static int h_permission(struct inode *h_inode, int mask,
18062 +                       struct vfsmount *h_mnt, int brperm)
18063 +{
18064 +       int err;
18065 +       const unsigned char write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18066 +
18067 +       err = -EACCES;
18068 +       if ((write_mask && IS_IMMUTABLE(h_inode))
18069 +           || ((mask & MAY_EXEC)
18070 +               && S_ISREG(h_inode->i_mode)
18071 +               && ((h_mnt->mnt_flags & MNT_NOEXEC)
18072 +                   || !(h_inode->i_mode & S_IXUGO))))
18073 +               goto out;
18074 +
18075 +       /*
18076 +        * - skip the lower fs test in the case of write to ro branch.
18077 +        * - nfs dir permission write check is optimized, but a policy for
18078 +        *   link/rename requires a real check.
18079 +        */
18080 +       if ((write_mask && !au_br_writable(brperm))
18081 +           || (au_test_nfs(h_inode->i_sb) && S_ISDIR(h_inode->i_mode)
18082 +               && write_mask && !(mask & MAY_READ))
18083 +           || !h_inode->i_op->permission) {
18084 +               /* AuLabel(generic_permission); */
18085 +               err = generic_permission(h_inode, mask);
18086 +       } else {
18087 +               /* AuLabel(h_inode->permission); */
18088 +               err = h_inode->i_op->permission(h_inode, mask);
18089 +               AuTraceErr(err);
18090 +       }
18091 +
18092 +       if (!err)
18093 +               err = devcgroup_inode_permission(h_inode, mask);
18094 +       if (!err)
18095 +               err = security_inode_permission(h_inode, mask);
18096 +
18097 +#if 0
18098 +       if (!err) {
18099 +               /* todo: do we need to call ima_path_check()? */
18100 +               struct path h_path = {
18101 +                       .dentry =
18102 +                       .mnt    = h_mnt
18103 +               };
18104 +               err = ima_path_check(&h_path,
18105 +                                    mask & (MAY_READ | MAY_WRITE | MAY_EXEC),
18106 +                                    IMA_COUNT_LEAVE);
18107 +       }
18108 +#endif
18109 +
18110 +out:
18111 +       return err;
18112 +}
18113 +
18114 +static int aufs_permission(struct inode *inode, int mask)
18115 +{
18116 +       int err;
18117 +       aufs_bindex_t bindex, bend;
18118 +       const unsigned char isdir = !!S_ISDIR(inode->i_mode),
18119 +               write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18120 +       struct inode *h_inode;
18121 +       struct super_block *sb;
18122 +       struct au_branch *br;
18123 +
18124 +       /* todo: support rcu-walk? */
18125 +       if (mask & MAY_NOT_BLOCK)
18126 +               return -ECHILD;
18127 +
18128 +       sb = inode->i_sb;
18129 +       si_read_lock(sb, AuLock_FLUSH);
18130 +       ii_read_lock_child(inode);
18131 +#if 0
18132 +       err = au_iigen_test(inode, au_sigen(sb));
18133 +       if (unlikely(err))
18134 +               goto out;
18135 +#endif
18136 +
18137 +       if (!isdir
18138 +           || write_mask
18139 +           || au_opt_test(au_mntflags(sb), DIRPERM1)) {
18140 +               err = au_busy_or_stale();
18141 +               h_inode = au_h_iptr(inode, au_ibstart(inode));
18142 +               if (unlikely(!h_inode
18143 +                            || (h_inode->i_mode & S_IFMT)
18144 +                            != (inode->i_mode & S_IFMT)))
18145 +                       goto out;
18146 +
18147 +               err = 0;
18148 +               bindex = au_ibstart(inode);
18149 +               br = au_sbr(sb, bindex);
18150 +               err = h_permission(h_inode, mask, au_br_mnt(br), br->br_perm);
18151 +               if (write_mask
18152 +                   && !err
18153 +                   && !special_file(h_inode->i_mode)) {
18154 +                       /* test whether the upper writable branch exists */
18155 +                       err = -EROFS;
18156 +                       for (; bindex >= 0; bindex--)
18157 +                               if (!au_br_rdonly(au_sbr(sb, bindex))) {
18158 +                                       err = 0;
18159 +                                       break;
18160 +                               }
18161 +               }
18162 +               goto out;
18163 +       }
18164 +
18165 +       /* non-write to dir */
18166 +       err = 0;
18167 +       bend = au_ibend(inode);
18168 +       for (bindex = au_ibstart(inode); !err && bindex <= bend; bindex++) {
18169 +               h_inode = au_h_iptr(inode, bindex);
18170 +               if (h_inode) {
18171 +                       err = au_busy_or_stale();
18172 +                       if (unlikely(!S_ISDIR(h_inode->i_mode)))
18173 +                               break;
18174 +
18175 +                       br = au_sbr(sb, bindex);
18176 +                       err = h_permission(h_inode, mask, au_br_mnt(br),
18177 +                                          br->br_perm);
18178 +               }
18179 +       }
18180 +
18181 +out:
18182 +       ii_read_unlock(inode);
18183 +       si_read_unlock(sb);
18184 +       return err;
18185 +}
18186 +
18187 +/* ---------------------------------------------------------------------- */
18188 +
18189 +static struct dentry *aufs_lookup(struct inode *dir, struct dentry *dentry,
18190 +                                 unsigned int flags)
18191 +{
18192 +       struct dentry *ret, *parent;
18193 +       struct inode *inode;
18194 +       struct super_block *sb;
18195 +       int err, npositive;
18196 +
18197 +       IMustLock(dir);
18198 +
18199 +       /* todo: support rcu-walk? */
18200 +       ret = ERR_PTR(-ECHILD);
18201 +       if (flags & LOOKUP_RCU)
18202 +               goto out;
18203 +
18204 +       ret = ERR_PTR(-ENAMETOOLONG);
18205 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
18206 +               goto out;
18207 +
18208 +       sb = dir->i_sb;
18209 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18210 +       ret = ERR_PTR(err);
18211 +       if (unlikely(err))
18212 +               goto out;
18213 +
18214 +       err = au_di_init(dentry);
18215 +       ret = ERR_PTR(err);
18216 +       if (unlikely(err))
18217 +               goto out_si;
18218 +
18219 +       inode = NULL;
18220 +       npositive = 0; /* suppress a warning */
18221 +       parent = dentry->d_parent; /* dir inode is locked */
18222 +       di_read_lock_parent(parent, AuLock_IR);
18223 +       err = au_alive_dir(parent);
18224 +       if (!err)
18225 +               err = au_digen_test(parent, au_sigen(sb));
18226 +       if (!err) {
18227 +               npositive = au_lkup_dentry(dentry, au_dbstart(parent),
18228 +                                          /*type*/0);
18229 +               err = npositive;
18230 +       }
18231 +       di_read_unlock(parent, AuLock_IR);
18232 +       ret = ERR_PTR(err);
18233 +       if (unlikely(err < 0))
18234 +               goto out_unlock;
18235 +
18236 +       if (npositive) {
18237 +               inode = au_new_inode(dentry, /*must_new*/0);
18238 +               if (IS_ERR(inode)) {
18239 +                       ret = (void *)inode;
18240 +                       inode = NULL;
18241 +                       goto out_unlock;
18242 +               }
18243 +       }
18244 +
18245 +       if (inode)
18246 +               atomic_inc(&inode->i_count);
18247 +       ret = d_splice_alias(inode, dentry);
18248 +       if (IS_ERR(ret)
18249 +           && PTR_ERR(ret) == -EIO
18250 +           && inode
18251 +           && S_ISDIR(inode->i_mode)) {
18252 +               atomic_inc(&inode->i_count);
18253 +               ret = d_materialise_unique(dentry, inode);
18254 +               if (!IS_ERR(ret))
18255 +                       ii_write_unlock(inode);
18256 +       }
18257 +#if 0
18258 +       if (unlikely(d_need_lookup(dentry))) {
18259 +               spin_lock(&dentry->d_lock);
18260 +               dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
18261 +               spin_unlock(&dentry->d_lock);
18262 +       } else
18263 +#endif
18264 +       if (inode) {
18265 +               if (!IS_ERR(ret))
18266 +                       iput(inode);
18267 +               else {
18268 +                       ii_write_unlock(inode);
18269 +                       iput(inode);
18270 +                       inode = NULL;
18271 +               }
18272 +       }
18273 +
18274 +out_unlock:
18275 +       di_write_unlock(dentry);
18276 +       if (inode) {
18277 +               /* verbose coding for lock class name */
18278 +               if (unlikely(S_ISLNK(inode->i_mode)))
18279 +                       au_rw_class(&au_di(dentry)->di_rwsem,
18280 +                                   au_lc_key + AuLcSymlink_DIINFO);
18281 +               else if (unlikely(S_ISDIR(inode->i_mode)))
18282 +                       au_rw_class(&au_di(dentry)->di_rwsem,
18283 +                                   au_lc_key + AuLcDir_DIINFO);
18284 +               else /* likely */
18285 +                       au_rw_class(&au_di(dentry)->di_rwsem,
18286 +                                   au_lc_key + AuLcNonDir_DIINFO);
18287 +       }
18288 +out_si:
18289 +       si_read_unlock(sb);
18290 +out:
18291 +       return ret;
18292 +}
18293 +
18294 +/* ---------------------------------------------------------------------- */
18295 +
18296 +static int au_wr_dir_cpup(struct dentry *dentry, struct dentry *parent,
18297 +                         const unsigned char add_entry, aufs_bindex_t bcpup,
18298 +                         aufs_bindex_t bstart)
18299 +{
18300 +       int err;
18301 +       struct dentry *h_parent;
18302 +       struct inode *h_dir;
18303 +
18304 +       if (add_entry)
18305 +               IMustLock(parent->d_inode);
18306 +       else
18307 +               di_write_lock_parent(parent);
18308 +
18309 +       err = 0;
18310 +       if (!au_h_dptr(parent, bcpup)) {
18311 +               if (bstart > bcpup)
18312 +                       err = au_cpup_dirs(dentry, bcpup);
18313 +               else if (bstart < bcpup)
18314 +                       err = au_cpdown_dirs(dentry, bcpup);
18315 +               else
18316 +                       BUG();
18317 +       }
18318 +       if (!err && add_entry && !au_ftest_wrdir(add_entry, TMPFILE)) {
18319 +               h_parent = au_h_dptr(parent, bcpup);
18320 +               h_dir = h_parent->d_inode;
18321 +               mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
18322 +               err = au_lkup_neg(dentry, bcpup,
18323 +                                 au_ftest_wrdir(add_entry, TMP_WHENTRY));
18324 +               /* todo: no unlock here */
18325 +               mutex_unlock(&h_dir->i_mutex);
18326 +
18327 +               AuDbg("bcpup %d\n", bcpup);
18328 +               if (!err) {
18329 +                       if (!dentry->d_inode)
18330 +                               au_set_h_dptr(dentry, bstart, NULL);
18331 +                       au_update_dbrange(dentry, /*do_put_zero*/0);
18332 +               }
18333 +       }
18334 +
18335 +       if (!add_entry)
18336 +               di_write_unlock(parent);
18337 +       if (!err)
18338 +               err = bcpup; /* success */
18339 +
18340 +       AuTraceErr(err);
18341 +       return err;
18342 +}
18343 +
18344 +/*
18345 + * decide the branch and the parent dir where we will create a new entry.
18346 + * returns new bindex or an error.
18347 + * copyup the parent dir if needed.
18348 + */
18349 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
18350 +             struct au_wr_dir_args *args)
18351 +{
18352 +       int err;
18353 +       unsigned int flags;
18354 +       aufs_bindex_t bcpup, bstart, src_bstart;
18355 +       const unsigned char add_entry
18356 +               = au_ftest_wrdir(args->flags, ADD_ENTRY)
18357 +               | au_ftest_wrdir(args->flags, TMP_WHENTRY)
18358 +               | au_ftest_wrdir(args->flags, TMPFILE);
18359 +       struct super_block *sb;
18360 +       struct dentry *parent;
18361 +       struct au_sbinfo *sbinfo;
18362 +
18363 +       sb = dentry->d_sb;
18364 +       sbinfo = au_sbi(sb);
18365 +       parent = dget_parent(dentry);
18366 +       bstart = au_dbstart(dentry);
18367 +       bcpup = bstart;
18368 +       if (args->force_btgt < 0) {
18369 +               if (src_dentry) {
18370 +                       src_bstart = au_dbstart(src_dentry);
18371 +                       if (src_bstart < bstart)
18372 +                               bcpup = src_bstart;
18373 +               } else if (add_entry) {
18374 +                       flags = 0;
18375 +                       if (au_ftest_wrdir(args->flags, ISDIR))
18376 +                               au_fset_wbr(flags, DIR);
18377 +                       err = AuWbrCreate(sbinfo, dentry, flags);
18378 +                       bcpup = err;
18379 +               }
18380 +
18381 +               if (bcpup < 0 || au_test_ro(sb, bcpup, dentry->d_inode)) {
18382 +                       if (add_entry)
18383 +                               err = AuWbrCopyup(sbinfo, dentry);
18384 +                       else {
18385 +                               if (!IS_ROOT(dentry)) {
18386 +                                       di_read_lock_parent(parent, !AuLock_IR);
18387 +                                       err = AuWbrCopyup(sbinfo, dentry);
18388 +                                       di_read_unlock(parent, !AuLock_IR);
18389 +                               } else
18390 +                                       err = AuWbrCopyup(sbinfo, dentry);
18391 +                       }
18392 +                       bcpup = err;
18393 +                       if (unlikely(err < 0))
18394 +                               goto out;
18395 +               }
18396 +       } else {
18397 +               bcpup = args->force_btgt;
18398 +               AuDebugOn(au_test_ro(sb, bcpup, dentry->d_inode));
18399 +       }
18400 +
18401 +       AuDbg("bstart %d, bcpup %d\n", bstart, bcpup);
18402 +       err = bcpup;
18403 +       if (bcpup == bstart)
18404 +               goto out; /* success */
18405 +
18406 +       /* copyup the new parent into the branch we process */
18407 +       err = au_wr_dir_cpup(dentry, parent, add_entry, bcpup, bstart);
18408 +       if (err >= 0) {
18409 +               if (!dentry->d_inode) {
18410 +                       au_set_h_dptr(dentry, bstart, NULL);
18411 +                       au_set_dbstart(dentry, bcpup);
18412 +                       au_set_dbend(dentry, bcpup);
18413 +               }
18414 +               AuDebugOn(add_entry
18415 +                         && !au_ftest_wrdir(args->flags, TMPFILE)
18416 +                         && !au_h_dptr(dentry, bcpup));
18417 +       }
18418 +
18419 +out:
18420 +       dput(parent);
18421 +       return err;
18422 +}
18423 +
18424 +/* ---------------------------------------------------------------------- */
18425 +
18426 +void au_pin_hdir_unlock(struct au_pin *p)
18427 +{
18428 +       if (p->hdir)
18429 +               au_hn_imtx_unlock(p->hdir);
18430 +}
18431 +
18432 +int au_pin_hdir_lock(struct au_pin *p)
18433 +{
18434 +       int err;
18435 +
18436 +       err = 0;
18437 +       if (!p->hdir)
18438 +               goto out;
18439 +
18440 +       /* even if an error happens later, keep this lock */
18441 +       au_hn_imtx_lock_nested(p->hdir, p->lsc_hi);
18442 +
18443 +       err = -EBUSY;
18444 +       if (unlikely(p->hdir->hi_inode != p->h_parent->d_inode))
18445 +               goto out;
18446 +
18447 +       err = 0;
18448 +       if (p->h_dentry)
18449 +               err = au_h_verify(p->h_dentry, p->udba, p->hdir->hi_inode,
18450 +                                 p->h_parent, p->br);
18451 +
18452 +out:
18453 +       return err;
18454 +}
18455 +
18456 +int au_pin_hdir_relock(struct au_pin *p)
18457 +{
18458 +       int err, i;
18459 +       struct inode *h_i;
18460 +       struct dentry *h_d[] = {
18461 +               p->h_dentry,
18462 +               p->h_parent
18463 +       };
18464 +
18465 +       err = au_pin_hdir_lock(p);
18466 +       if (unlikely(err))
18467 +               goto out;
18468 +
18469 +       for (i = 0; !err && i < sizeof(h_d)/sizeof(*h_d); i++) {
18470 +               if (!h_d[i])
18471 +                       continue;
18472 +               h_i = h_d[i]->d_inode;
18473 +               if (h_i)
18474 +                       err = !h_i->i_nlink;
18475 +       }
18476 +
18477 +out:
18478 +       return err;
18479 +}
18480 +
18481 +void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task)
18482 +{
18483 +#if defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_SMP)
18484 +       p->hdir->hi_inode->i_mutex.owner = task;
18485 +#endif
18486 +}
18487 +
18488 +void au_pin_hdir_acquire_nest(struct au_pin *p)
18489 +{
18490 +       if (p->hdir) {
18491 +               mutex_acquire_nest(&p->hdir->hi_inode->i_mutex.dep_map,
18492 +                                  p->lsc_hi, 0, NULL, _RET_IP_);
18493 +               au_pin_hdir_set_owner(p, current);
18494 +       }
18495 +}
18496 +
18497 +void au_pin_hdir_release(struct au_pin *p)
18498 +{
18499 +       if (p->hdir) {
18500 +               au_pin_hdir_set_owner(p, p->task);
18501 +               mutex_release(&p->hdir->hi_inode->i_mutex.dep_map, 1, _RET_IP_);
18502 +       }
18503 +}
18504 +
18505 +struct dentry *au_pinned_h_parent(struct au_pin *pin)
18506 +{
18507 +       if (pin && pin->parent)
18508 +               return au_h_dptr(pin->parent, pin->bindex);
18509 +       return NULL;
18510 +}
18511 +
18512 +void au_unpin(struct au_pin *p)
18513 +{
18514 +       if (p->hdir)
18515 +               au_pin_hdir_unlock(p);
18516 +       if (p->h_mnt && au_ftest_pin(p->flags, MNT_WRITE))
18517 +               vfsub_mnt_drop_write(p->h_mnt);
18518 +       if (!p->hdir)
18519 +               return;
18520 +
18521 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
18522 +               di_read_unlock(p->parent, AuLock_IR);
18523 +       iput(p->hdir->hi_inode);
18524 +       dput(p->parent);
18525 +       p->parent = NULL;
18526 +       p->hdir = NULL;
18527 +       p->h_mnt = NULL;
18528 +       /* do not clear p->task */
18529 +}
18530 +
18531 +int au_do_pin(struct au_pin *p)
18532 +{
18533 +       int err;
18534 +       struct super_block *sb;
18535 +       struct inode *h_dir;
18536 +
18537 +       err = 0;
18538 +       sb = p->dentry->d_sb;
18539 +       p->br = au_sbr(sb, p->bindex);
18540 +       if (IS_ROOT(p->dentry)) {
18541 +               if (au_ftest_pin(p->flags, MNT_WRITE)) {
18542 +                       p->h_mnt = au_br_mnt(p->br);
18543 +                       err = vfsub_mnt_want_write(p->h_mnt);
18544 +                       if (unlikely(err)) {
18545 +                               au_fclr_pin(p->flags, MNT_WRITE);
18546 +                               goto out_err;
18547 +                       }
18548 +               }
18549 +               goto out;
18550 +       }
18551 +
18552 +       p->h_dentry = NULL;
18553 +       if (p->bindex <= au_dbend(p->dentry))
18554 +               p->h_dentry = au_h_dptr(p->dentry, p->bindex);
18555 +
18556 +       p->parent = dget_parent(p->dentry);
18557 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
18558 +               di_read_lock(p->parent, AuLock_IR, p->lsc_di);
18559 +
18560 +       h_dir = NULL;
18561 +       p->h_parent = au_h_dptr(p->parent, p->bindex);
18562 +       p->hdir = au_hi(p->parent->d_inode, p->bindex);
18563 +       if (p->hdir)
18564 +               h_dir = p->hdir->hi_inode;
18565 +
18566 +       /*
18567 +        * udba case, or
18568 +        * if DI_LOCKED is not set, then p->parent may be different
18569 +        * and h_parent can be NULL.
18570 +        */
18571 +       if (unlikely(!p->hdir || !h_dir || !p->h_parent)) {
18572 +               err = -EBUSY;
18573 +               if (!au_ftest_pin(p->flags, DI_LOCKED))
18574 +                       di_read_unlock(p->parent, AuLock_IR);
18575 +               dput(p->parent);
18576 +               p->parent = NULL;
18577 +               goto out_err;
18578 +       }
18579 +
18580 +       if (au_ftest_pin(p->flags, MNT_WRITE)) {
18581 +               p->h_mnt = au_br_mnt(p->br);
18582 +               err = vfsub_mnt_want_write(p->h_mnt);
18583 +               if (unlikely(err)) {
18584 +                       au_fclr_pin(p->flags, MNT_WRITE);
18585 +                       if (!au_ftest_pin(p->flags, DI_LOCKED))
18586 +                               di_read_unlock(p->parent, AuLock_IR);
18587 +                       dput(p->parent);
18588 +                       p->parent = NULL;
18589 +                       goto out_err;
18590 +               }
18591 +       }
18592 +
18593 +       au_igrab(h_dir);
18594 +       err = au_pin_hdir_lock(p);
18595 +       if (!err)
18596 +               goto out; /* success */
18597 +
18598 +       au_unpin(p);
18599 +
18600 +out_err:
18601 +       pr_err("err %d\n", err);
18602 +       err = au_busy_or_stale();
18603 +out:
18604 +       return err;
18605 +}
18606 +
18607 +void au_pin_init(struct au_pin *p, struct dentry *dentry,
18608 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
18609 +                unsigned int udba, unsigned char flags)
18610 +{
18611 +       p->dentry = dentry;
18612 +       p->udba = udba;
18613 +       p->lsc_di = lsc_di;
18614 +       p->lsc_hi = lsc_hi;
18615 +       p->flags = flags;
18616 +       p->bindex = bindex;
18617 +
18618 +       p->parent = NULL;
18619 +       p->hdir = NULL;
18620 +       p->h_mnt = NULL;
18621 +
18622 +       p->h_dentry = NULL;
18623 +       p->h_parent = NULL;
18624 +       p->br = NULL;
18625 +       p->task = current;
18626 +}
18627 +
18628 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
18629 +          unsigned int udba, unsigned char flags)
18630 +{
18631 +       au_pin_init(pin, dentry, bindex, AuLsc_DI_PARENT, AuLsc_I_PARENT2,
18632 +                   udba, flags);
18633 +       return au_do_pin(pin);
18634 +}
18635 +
18636 +/* ---------------------------------------------------------------------- */
18637 +
18638 +/*
18639 + * ->setattr() and ->getattr() are called in various cases.
18640 + * chmod, stat: dentry is revalidated.
18641 + * fchmod, fstat: file and dentry are not revalidated, additionally they may be
18642 + *               unhashed.
18643 + * for ->setattr(), ia->ia_file is passed from ftruncate only.
18644 + */
18645 +/* todo: consolidate with do_refresh() and simple_reval_dpath() */
18646 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen)
18647 +{
18648 +       int err;
18649 +       struct inode *inode;
18650 +       struct dentry *parent;
18651 +
18652 +       err = 0;
18653 +       inode = dentry->d_inode;
18654 +       if (au_digen_test(dentry, sigen)) {
18655 +               parent = dget_parent(dentry);
18656 +               di_read_lock_parent(parent, AuLock_IR);
18657 +               err = au_refresh_dentry(dentry, parent);
18658 +               di_read_unlock(parent, AuLock_IR);
18659 +               dput(parent);
18660 +       }
18661 +
18662 +       AuTraceErr(err);
18663 +       return err;
18664 +}
18665 +
18666 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
18667 +                    struct au_icpup_args *a)
18668 +{
18669 +       int err;
18670 +       loff_t sz;
18671 +       aufs_bindex_t bstart, ibstart;
18672 +       struct dentry *hi_wh, *parent;
18673 +       struct inode *inode;
18674 +       struct au_wr_dir_args wr_dir_args = {
18675 +               .force_btgt     = -1,
18676 +               .flags          = 0
18677 +       };
18678 +
18679 +       bstart = au_dbstart(dentry);
18680 +       inode = dentry->d_inode;
18681 +       if (S_ISDIR(inode->i_mode))
18682 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
18683 +       /* plink or hi_wh() case */
18684 +       ibstart = au_ibstart(inode);
18685 +       if (bstart != ibstart && !au_test_ro(inode->i_sb, ibstart, inode))
18686 +               wr_dir_args.force_btgt = ibstart;
18687 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
18688 +       if (unlikely(err < 0))
18689 +               goto out;
18690 +       a->btgt = err;
18691 +       if (err != bstart)
18692 +               au_fset_icpup(a->flags, DID_CPUP);
18693 +
18694 +       err = 0;
18695 +       a->pin_flags = AuPin_MNT_WRITE;
18696 +       parent = NULL;
18697 +       if (!IS_ROOT(dentry)) {
18698 +               au_fset_pin(a->pin_flags, DI_LOCKED);
18699 +               parent = dget_parent(dentry);
18700 +               di_write_lock_parent(parent);
18701 +       }
18702 +
18703 +       err = au_pin(&a->pin, dentry, a->btgt, a->udba, a->pin_flags);
18704 +       if (unlikely(err))
18705 +               goto out_parent;
18706 +
18707 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
18708 +       a->h_inode = a->h_path.dentry->d_inode;
18709 +       sz = -1;
18710 +       if (ia && (ia->ia_valid & ATTR_SIZE)) {
18711 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
18712 +               if (ia->ia_size < i_size_read(a->h_inode))
18713 +                       sz = ia->ia_size;
18714 +               mutex_unlock(&a->h_inode->i_mutex);
18715 +       }
18716 +
18717 +       hi_wh = NULL;
18718 +       if (au_ftest_icpup(a->flags, DID_CPUP) && d_unlinked(dentry)) {
18719 +               hi_wh = au_hi_wh(inode, a->btgt);
18720 +               if (!hi_wh) {
18721 +                       struct au_cp_generic cpg = {
18722 +                               .dentry = dentry,
18723 +                               .bdst   = a->btgt,
18724 +                               .bsrc   = -1,
18725 +                               .len    = sz,
18726 +                               .pin    = &a->pin
18727 +                       };
18728 +                       err = au_sio_cpup_wh(&cpg, /*file*/NULL);
18729 +                       if (unlikely(err))
18730 +                               goto out_unlock;
18731 +                       hi_wh = au_hi_wh(inode, a->btgt);
18732 +                       /* todo: revalidate hi_wh? */
18733 +               }
18734 +       }
18735 +
18736 +       if (parent) {
18737 +               au_pin_set_parent_lflag(&a->pin, /*lflag*/0);
18738 +               di_downgrade_lock(parent, AuLock_IR);
18739 +               dput(parent);
18740 +               parent = NULL;
18741 +       }
18742 +       if (!au_ftest_icpup(a->flags, DID_CPUP))
18743 +               goto out; /* success */
18744 +
18745 +       if (!d_unhashed(dentry)) {
18746 +               struct au_cp_generic cpg = {
18747 +                       .dentry = dentry,
18748 +                       .bdst   = a->btgt,
18749 +                       .bsrc   = bstart,
18750 +                       .len    = sz,
18751 +                       .pin    = &a->pin,
18752 +                       .flags  = AuCpup_DTIME | AuCpup_HOPEN
18753 +               };
18754 +               err = au_sio_cpup_simple(&cpg);
18755 +               if (!err)
18756 +                       a->h_path.dentry = au_h_dptr(dentry, a->btgt);
18757 +       } else if (!hi_wh)
18758 +               a->h_path.dentry = au_h_dptr(dentry, a->btgt);
18759 +       else
18760 +               a->h_path.dentry = hi_wh; /* do not dget here */
18761 +
18762 +out_unlock:
18763 +       a->h_inode = a->h_path.dentry->d_inode;
18764 +       if (!err)
18765 +               goto out; /* success */
18766 +       au_unpin(&a->pin);
18767 +out_parent:
18768 +       if (parent) {
18769 +               di_write_unlock(parent);
18770 +               dput(parent);
18771 +       }
18772 +out:
18773 +       if (!err)
18774 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
18775 +       return err;
18776 +}
18777 +
18778 +static int aufs_setattr(struct dentry *dentry, struct iattr *ia)
18779 +{
18780 +       int err;
18781 +       struct inode *inode, *delegated;
18782 +       struct super_block *sb;
18783 +       struct file *file;
18784 +       struct au_icpup_args *a;
18785 +
18786 +       inode = dentry->d_inode;
18787 +       IMustLock(inode);
18788 +
18789 +       err = -ENOMEM;
18790 +       a = kzalloc(sizeof(*a), GFP_NOFS);
18791 +       if (unlikely(!a))
18792 +               goto out;
18793 +
18794 +       if (ia->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
18795 +               ia->ia_valid &= ~ATTR_MODE;
18796 +
18797 +       file = NULL;
18798 +       sb = dentry->d_sb;
18799 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18800 +       if (unlikely(err))
18801 +               goto out_kfree;
18802 +
18803 +       if (ia->ia_valid & ATTR_FILE) {
18804 +               /* currently ftruncate(2) only */
18805 +               AuDebugOn(!S_ISREG(inode->i_mode));
18806 +               file = ia->ia_file;
18807 +               err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
18808 +               if (unlikely(err))
18809 +                       goto out_si;
18810 +               ia->ia_file = au_hf_top(file);
18811 +               a->udba = AuOpt_UDBA_NONE;
18812 +       } else {
18813 +               /* fchmod() doesn't pass ia_file */
18814 +               a->udba = au_opt_udba(sb);
18815 +               di_write_lock_child(dentry);
18816 +               /* no d_unlinked(), to set UDBA_NONE for root */
18817 +               if (d_unhashed(dentry))
18818 +                       a->udba = AuOpt_UDBA_NONE;
18819 +               if (a->udba != AuOpt_UDBA_NONE) {
18820 +                       AuDebugOn(IS_ROOT(dentry));
18821 +                       err = au_reval_for_attr(dentry, au_sigen(sb));
18822 +                       if (unlikely(err))
18823 +                               goto out_dentry;
18824 +               }
18825 +       }
18826 +
18827 +       err = au_pin_and_icpup(dentry, ia, a);
18828 +       if (unlikely(err < 0))
18829 +               goto out_dentry;
18830 +       if (au_ftest_icpup(a->flags, DID_CPUP)) {
18831 +               ia->ia_file = NULL;
18832 +               ia->ia_valid &= ~ATTR_FILE;
18833 +       }
18834 +
18835 +       a->h_path.mnt = au_sbr_mnt(sb, a->btgt);
18836 +       if ((ia->ia_valid & (ATTR_MODE | ATTR_CTIME))
18837 +           == (ATTR_MODE | ATTR_CTIME)) {
18838 +               err = security_path_chmod(&a->h_path, ia->ia_mode);
18839 +               if (unlikely(err))
18840 +                       goto out_unlock;
18841 +       } else if ((ia->ia_valid & (ATTR_UID | ATTR_GID))
18842 +                  && (ia->ia_valid & ATTR_CTIME)) {
18843 +               err = security_path_chown(&a->h_path, ia->ia_uid, ia->ia_gid);
18844 +               if (unlikely(err))
18845 +                       goto out_unlock;
18846 +       }
18847 +
18848 +       if (ia->ia_valid & ATTR_SIZE) {
18849 +               struct file *f;
18850 +
18851 +               if (ia->ia_size < i_size_read(inode))
18852 +                       /* unmap only */
18853 +                       truncate_setsize(inode, ia->ia_size);
18854 +
18855 +               f = NULL;
18856 +               if (ia->ia_valid & ATTR_FILE)
18857 +                       f = ia->ia_file;
18858 +               mutex_unlock(&a->h_inode->i_mutex);
18859 +               err = vfsub_trunc(&a->h_path, ia->ia_size, ia->ia_valid, f);
18860 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
18861 +       } else {
18862 +               delegated = NULL;
18863 +               while (1) {
18864 +                       err = vfsub_notify_change(&a->h_path, ia, &delegated);
18865 +                       if (delegated) {
18866 +                               err = break_deleg_wait(&delegated);
18867 +                               if (!err)
18868 +                                       continue;
18869 +                       }
18870 +                       break;
18871 +               }
18872 +       }
18873 +       if (!err)
18874 +               au_cpup_attr_changeable(inode);
18875 +
18876 +out_unlock:
18877 +       mutex_unlock(&a->h_inode->i_mutex);
18878 +       au_unpin(&a->pin);
18879 +       if (unlikely(err))
18880 +               au_update_dbstart(dentry);
18881 +out_dentry:
18882 +       di_write_unlock(dentry);
18883 +       if (file) {
18884 +               fi_write_unlock(file);
18885 +               ia->ia_file = file;
18886 +               ia->ia_valid |= ATTR_FILE;
18887 +       }
18888 +out_si:
18889 +       si_read_unlock(sb);
18890 +out_kfree:
18891 +       kfree(a);
18892 +out:
18893 +       AuTraceErr(err);
18894 +       return err;
18895 +}
18896 +
18897 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
18898 +static int au_h_path_to_set_attr(struct dentry *dentry,
18899 +                                struct au_icpup_args *a, struct path *h_path)
18900 +{
18901 +       int err;
18902 +       struct super_block *sb;
18903 +
18904 +       sb = dentry->d_sb;
18905 +       a->udba = au_opt_udba(sb);
18906 +       /* no d_unlinked(), to set UDBA_NONE for root */
18907 +       if (d_unhashed(dentry))
18908 +               a->udba = AuOpt_UDBA_NONE;
18909 +       if (a->udba != AuOpt_UDBA_NONE) {
18910 +               AuDebugOn(IS_ROOT(dentry));
18911 +               err = au_reval_for_attr(dentry, au_sigen(sb));
18912 +               if (unlikely(err))
18913 +                       goto out;
18914 +       }
18915 +       err = au_pin_and_icpup(dentry, /*ia*/NULL, a);
18916 +       if (unlikely(err < 0))
18917 +               goto out;
18918 +
18919 +       h_path->dentry = a->h_path.dentry;
18920 +       h_path->mnt = au_sbr_mnt(sb, a->btgt);
18921 +
18922 +out:
18923 +       return err;
18924 +}
18925 +
18926 +ssize_t au_srxattr(struct dentry *dentry, struct au_srxattr *arg)
18927 +{
18928 +       int err;
18929 +       struct path h_path;
18930 +       struct super_block *sb;
18931 +       struct au_icpup_args *a;
18932 +       struct inode *inode, *h_inode;
18933 +
18934 +       inode = dentry->d_inode;
18935 +       IMustLock(inode);
18936 +
18937 +       err = -ENOMEM;
18938 +       a = kzalloc(sizeof(*a), GFP_NOFS);
18939 +       if (unlikely(!a))
18940 +               goto out;
18941 +
18942 +       sb = dentry->d_sb;
18943 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18944 +       if (unlikely(err))
18945 +               goto out_kfree;
18946 +
18947 +       h_path.dentry = NULL;   /* silence gcc */
18948 +       di_write_lock_child(dentry);
18949 +       err = au_h_path_to_set_attr(dentry, a, &h_path);
18950 +       if (unlikely(err))
18951 +               goto out_di;
18952 +
18953 +       mutex_unlock(&a->h_inode->i_mutex);
18954 +       switch (arg->type) {
18955 +       case AU_XATTR_SET:
18956 +               err = vfsub_setxattr(h_path.dentry,
18957 +                                    arg->u.set.name, arg->u.set.value,
18958 +                                    arg->u.set.size, arg->u.set.flags);
18959 +               break;
18960 +       case AU_XATTR_REMOVE:
18961 +               err = vfsub_removexattr(h_path.dentry, arg->u.remove.name);
18962 +               break;
18963 +       case AU_ACL_SET:
18964 +               err = -EOPNOTSUPP;
18965 +               h_inode = h_path.dentry->d_inode;
18966 +               if (h_inode->i_op->set_acl)
18967 +                       err = h_inode->i_op->set_acl(h_inode,
18968 +                                                    arg->u.acl_set.acl,
18969 +                                                    arg->u.acl_set.type);
18970 +               break;
18971 +       }
18972 +       if (!err)
18973 +               au_cpup_attr_timesizes(inode);
18974 +
18975 +       au_unpin(&a->pin);
18976 +       if (unlikely(err))
18977 +               au_update_dbstart(dentry);
18978 +
18979 +out_di:
18980 +       di_write_unlock(dentry);
18981 +       si_read_unlock(sb);
18982 +out_kfree:
18983 +       kfree(a);
18984 +out:
18985 +       AuTraceErr(err);
18986 +       return err;
18987 +}
18988 +#endif
18989 +
18990 +static void au_refresh_iattr(struct inode *inode, struct kstat *st,
18991 +                            unsigned int nlink)
18992 +{
18993 +       unsigned int n;
18994 +
18995 +       inode->i_mode = st->mode;
18996 +       /* don't i_[ug]id_write() here */
18997 +       inode->i_uid = st->uid;
18998 +       inode->i_gid = st->gid;
18999 +       inode->i_atime = st->atime;
19000 +       inode->i_mtime = st->mtime;
19001 +       inode->i_ctime = st->ctime;
19002 +
19003 +       au_cpup_attr_nlink(inode, /*force*/0);
19004 +       if (S_ISDIR(inode->i_mode)) {
19005 +               n = inode->i_nlink;
19006 +               n -= nlink;
19007 +               n += st->nlink;
19008 +               smp_mb(); /* for i_nlink */
19009 +               /* 0 can happen */
19010 +               set_nlink(inode, n);
19011 +       }
19012 +
19013 +       spin_lock(&inode->i_lock);
19014 +       inode->i_blocks = st->blocks;
19015 +       i_size_write(inode, st->size);
19016 +       spin_unlock(&inode->i_lock);
19017 +}
19018 +
19019 +/*
19020 + * common routine for aufs_getattr() and aufs_getxattr().
19021 + * returns zero or negative (an error).
19022 + * @dentry will be read-locked in success.
19023 + */
19024 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path)
19025 +{
19026 +       int err;
19027 +       unsigned int mnt_flags, sigen;
19028 +       unsigned char udba_none;
19029 +       aufs_bindex_t bindex;
19030 +       struct super_block *sb, *h_sb;
19031 +       struct inode *inode;
19032 +
19033 +       h_path->mnt = NULL;
19034 +       h_path->dentry = NULL;
19035 +
19036 +       err = 0;
19037 +       sb = dentry->d_sb;
19038 +       mnt_flags = au_mntflags(sb);
19039 +       udba_none = !!au_opt_test(mnt_flags, UDBA_NONE);
19040 +
19041 +       /* support fstat(2) */
19042 +       if (!d_unlinked(dentry) && !udba_none) {
19043 +               sigen = au_sigen(sb);
19044 +               err = au_digen_test(dentry, sigen);
19045 +               if (!err) {
19046 +                       di_read_lock_child(dentry, AuLock_IR);
19047 +                       err = au_dbrange_test(dentry);
19048 +                       if (unlikely(err)) {
19049 +                               di_read_unlock(dentry, AuLock_IR);
19050 +                               goto out;
19051 +                       }
19052 +               } else {
19053 +                       AuDebugOn(IS_ROOT(dentry));
19054 +                       di_write_lock_child(dentry);
19055 +                       err = au_dbrange_test(dentry);
19056 +                       if (!err)
19057 +                               err = au_reval_for_attr(dentry, sigen);
19058 +                       if (!err)
19059 +                               di_downgrade_lock(dentry, AuLock_IR);
19060 +                       else {
19061 +                               di_write_unlock(dentry);
19062 +                               goto out;
19063 +                       }
19064 +               }
19065 +       } else
19066 +               di_read_lock_child(dentry, AuLock_IR);
19067 +
19068 +       inode = dentry->d_inode;
19069 +       bindex = au_ibstart(inode);
19070 +       h_path->mnt = au_sbr_mnt(sb, bindex);
19071 +       h_sb = h_path->mnt->mnt_sb;
19072 +       if (!force
19073 +           && !au_test_fs_bad_iattr(h_sb)
19074 +           && udba_none)
19075 +               goto out; /* success */
19076 +
19077 +       if (au_dbstart(dentry) == bindex)
19078 +               h_path->dentry = au_h_dptr(dentry, bindex);
19079 +       else if (au_opt_test(mnt_flags, PLINK) && au_plink_test(inode)) {
19080 +               h_path->dentry = au_plink_lkup(inode, bindex);
19081 +               if (IS_ERR(h_path->dentry))
19082 +                       /* pretending success */
19083 +                       h_path->dentry = NULL;
19084 +               else
19085 +                       dput(h_path->dentry);
19086 +       }
19087 +
19088 +out:
19089 +       return err;
19090 +}
19091 +
19092 +static int aufs_getattr(struct vfsmount *mnt __maybe_unused,
19093 +                       struct dentry *dentry, struct kstat *st)
19094 +{
19095 +       int err;
19096 +       unsigned char positive;
19097 +       struct path h_path;
19098 +       struct inode *inode;
19099 +       struct super_block *sb;
19100 +
19101 +       inode = dentry->d_inode;
19102 +       sb = dentry->d_sb;
19103 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19104 +       if (unlikely(err))
19105 +               goto out;
19106 +       err = au_h_path_getattr(dentry, /*force*/0, &h_path);
19107 +       if (unlikely(err))
19108 +               goto out_si;
19109 +       if (unlikely(!h_path.dentry))
19110 +               /* illegally overlapped or something */
19111 +               goto out_fill; /* pretending success */
19112 +
19113 +       positive = !!h_path.dentry->d_inode;
19114 +       if (positive)
19115 +               err = vfs_getattr(&h_path, st);
19116 +       if (!err) {
19117 +               if (positive)
19118 +                       au_refresh_iattr(inode, st,
19119 +                                        h_path.dentry->d_inode->i_nlink);
19120 +               goto out_fill; /* success */
19121 +       }
19122 +       AuTraceErr(err);
19123 +       goto out_di;
19124 +
19125 +out_fill:
19126 +       generic_fillattr(inode, st);
19127 +out_di:
19128 +       di_read_unlock(dentry, AuLock_IR);
19129 +out_si:
19130 +       si_read_unlock(sb);
19131 +out:
19132 +       AuTraceErr(err);
19133 +       return err;
19134 +}
19135 +
19136 +/* ---------------------------------------------------------------------- */
19137 +
19138 +static int h_readlink(struct dentry *dentry, int bindex, char __user *buf,
19139 +                     int bufsiz)
19140 +{
19141 +       int err;
19142 +       struct super_block *sb;
19143 +       struct dentry *h_dentry;
19144 +
19145 +       err = -EINVAL;
19146 +       h_dentry = au_h_dptr(dentry, bindex);
19147 +       if (unlikely(!h_dentry->d_inode->i_op->readlink))
19148 +               goto out;
19149 +
19150 +       err = security_inode_readlink(h_dentry);
19151 +       if (unlikely(err))
19152 +               goto out;
19153 +
19154 +       sb = dentry->d_sb;
19155 +       if (!au_test_ro(sb, bindex, dentry->d_inode)) {
19156 +               vfsub_touch_atime(au_sbr_mnt(sb, bindex), h_dentry);
19157 +               fsstack_copy_attr_atime(dentry->d_inode, h_dentry->d_inode);
19158 +       }
19159 +       err = h_dentry->d_inode->i_op->readlink(h_dentry, buf, bufsiz);
19160 +
19161 +out:
19162 +       return err;
19163 +}
19164 +
19165 +static int aufs_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
19166 +{
19167 +       int err;
19168 +
19169 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
19170 +       if (unlikely(err))
19171 +               goto out;
19172 +       err = au_d_hashed_positive(dentry);
19173 +       if (!err)
19174 +               err = h_readlink(dentry, au_dbstart(dentry), buf, bufsiz);
19175 +       aufs_read_unlock(dentry, AuLock_IR);
19176 +
19177 +out:
19178 +       return err;
19179 +}
19180 +
19181 +static void *aufs_follow_link(struct dentry *dentry, struct nameidata *nd)
19182 +{
19183 +       int err;
19184 +       mm_segment_t old_fs;
19185 +       union {
19186 +               char *k;
19187 +               char __user *u;
19188 +       } buf;
19189 +
19190 +       err = -ENOMEM;
19191 +       buf.k = (void *)__get_free_page(GFP_NOFS);
19192 +       if (unlikely(!buf.k))
19193 +               goto out;
19194 +
19195 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
19196 +       if (unlikely(err))
19197 +               goto out_name;
19198 +
19199 +       err = au_d_hashed_positive(dentry);
19200 +       if (!err) {
19201 +               old_fs = get_fs();
19202 +               set_fs(KERNEL_DS);
19203 +               err = h_readlink(dentry, au_dbstart(dentry), buf.u, PATH_MAX);
19204 +               set_fs(old_fs);
19205 +       }
19206 +       aufs_read_unlock(dentry, AuLock_IR);
19207 +
19208 +       if (err >= 0) {
19209 +               buf.k[err] = 0;
19210 +               /* will be freed by put_link */
19211 +               nd_set_link(nd, buf.k);
19212 +               return NULL; /* success */
19213 +       }
19214 +
19215 +out_name:
19216 +       free_page((unsigned long)buf.k);
19217 +out:
19218 +       AuTraceErr(err);
19219 +       return ERR_PTR(err);
19220 +}
19221 +
19222 +static void aufs_put_link(struct dentry *dentry __maybe_unused,
19223 +                         struct nameidata *nd, void *cookie __maybe_unused)
19224 +{
19225 +       char *p;
19226 +
19227 +       p = nd_get_link(nd);
19228 +       if (!IS_ERR_OR_NULL(p))
19229 +               free_page((unsigned long)p);
19230 +}
19231 +
19232 +/* ---------------------------------------------------------------------- */
19233 +
19234 +static int aufs_update_time(struct inode *inode, struct timespec *ts, int flags)
19235 +{
19236 +       int err;
19237 +       struct super_block *sb;
19238 +       struct inode *h_inode;
19239 +
19240 +       sb = inode->i_sb;
19241 +       /* mmap_sem might be acquired already, cf. aufs_mmap() */
19242 +       lockdep_off();
19243 +       si_read_lock(sb, AuLock_FLUSH);
19244 +       ii_write_lock_child(inode);
19245 +       lockdep_on();
19246 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
19247 +       err = vfsub_update_time(h_inode, ts, flags);
19248 +       lockdep_off();
19249 +       if (!err)
19250 +               au_cpup_attr_timesizes(inode);
19251 +       ii_write_unlock(inode);
19252 +       si_read_unlock(sb);
19253 +       lockdep_on();
19254 +
19255 +       if (!err && (flags & S_VERSION))
19256 +               inode_inc_iversion(inode);
19257 +
19258 +       return err;
19259 +}
19260 +
19261 +/* ---------------------------------------------------------------------- */
19262 +
19263 +struct inode_operations aufs_symlink_iop = {
19264 +       .permission     = aufs_permission,
19265 +#ifdef CONFIG_FS_POSIX_ACL
19266 +       .get_acl        = aufs_get_acl,
19267 +       .set_acl        = aufs_set_acl, /* unsupport for symlink? */
19268 +#endif
19269 +
19270 +       .setattr        = aufs_setattr,
19271 +       .getattr        = aufs_getattr,
19272 +
19273 +#ifdef CONFIG_AUFS_XATTR
19274 +       .setxattr       = aufs_setxattr,
19275 +       .getxattr       = aufs_getxattr,
19276 +       .listxattr      = aufs_listxattr,
19277 +       .removexattr    = aufs_removexattr,
19278 +#endif
19279 +
19280 +       .readlink       = aufs_readlink,
19281 +       .follow_link    = aufs_follow_link,
19282 +       .put_link       = aufs_put_link,
19283 +
19284 +       /* .update_time = aufs_update_time */
19285 +};
19286 +
19287 +struct inode_operations aufs_dir_iop = {
19288 +       .create         = aufs_create,
19289 +       .lookup         = aufs_lookup,
19290 +       .link           = aufs_link,
19291 +       .unlink         = aufs_unlink,
19292 +       .symlink        = aufs_symlink,
19293 +       .mkdir          = aufs_mkdir,
19294 +       .rmdir          = aufs_rmdir,
19295 +       .mknod          = aufs_mknod,
19296 +       .rename         = aufs_rename,
19297 +
19298 +       .permission     = aufs_permission,
19299 +#ifdef CONFIG_FS_POSIX_ACL
19300 +       .get_acl        = aufs_get_acl,
19301 +       .set_acl        = aufs_set_acl,
19302 +#endif
19303 +
19304 +       .setattr        = aufs_setattr,
19305 +       .getattr        = aufs_getattr,
19306 +
19307 +#ifdef CONFIG_AUFS_XATTR
19308 +       .setxattr       = aufs_setxattr,
19309 +       .getxattr       = aufs_getxattr,
19310 +       .listxattr      = aufs_listxattr,
19311 +       .removexattr    = aufs_removexattr,
19312 +#endif
19313 +
19314 +       .update_time    = aufs_update_time,
19315 +       /* no support for atomic_open() */
19316 +
19317 +       .tmpfile        = aufs_tmpfile
19318 +};
19319 +
19320 +struct inode_operations aufs_iop = {
19321 +       .permission     = aufs_permission,
19322 +#ifdef CONFIG_FS_POSIX_ACL
19323 +       .get_acl        = aufs_get_acl,
19324 +       .set_acl        = aufs_set_acl,
19325 +#endif
19326 +
19327 +       .setattr        = aufs_setattr,
19328 +       .getattr        = aufs_getattr,
19329 +
19330 +#ifdef CONFIG_AUFS_XATTR
19331 +       .setxattr       = aufs_setxattr,
19332 +       .getxattr       = aufs_getxattr,
19333 +       .listxattr      = aufs_listxattr,
19334 +       .removexattr    = aufs_removexattr,
19335 +#endif
19336 +
19337 +       .update_time    = aufs_update_time
19338 +};
19339 diff -urN /usr/share/empty/fs/aufs/i_op_del.c linux/fs/aufs/i_op_del.c
19340 --- /usr/share/empty/fs/aufs/i_op_del.c 1970-01-01 01:00:00.000000000 +0100
19341 +++ linux/fs/aufs/i_op_del.c    2015-01-25 13:00:38.631047076 +0100
19342 @@ -0,0 +1,507 @@
19343 +/*
19344 + * Copyright (C) 2005-2014 Junjiro R. Okajima
19345 + *
19346 + * This program, aufs is free software; you can redistribute it and/or modify
19347 + * it under the terms of the GNU General Public License as published by
19348 + * the Free Software Foundation; either version 2 of the License, or
19349 + * (at your option) any later version.
19350 + *
19351 + * This program is distributed in the hope that it will be useful,
19352 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19353 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19354 + * GNU General Public License for more details.
19355 + *
19356 + * You should have received a copy of the GNU General Public License
19357 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19358 + */
19359 +
19360 +/*
19361 + * inode operations (del entry)
19362 + */
19363 +
19364 +#include "aufs.h"
19365 +
19366 +/*
19367 + * decide if a new whiteout for @dentry is necessary or not.
19368 + * when it is necessary, prepare the parent dir for the upper branch whose
19369 + * branch index is @bcpup for creation. the actual creation of the whiteout will
19370 + * be done by caller.
19371 + * return value:
19372 + * 0: wh is unnecessary
19373 + * plus: wh is necessary
19374 + * minus: error
19375 + */
19376 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup)
19377 +{
19378 +       int need_wh, err;
19379 +       aufs_bindex_t bstart;
19380 +       struct super_block *sb;
19381 +
19382 +       sb = dentry->d_sb;
19383 +       bstart = au_dbstart(dentry);
19384 +       if (*bcpup < 0) {
19385 +               *bcpup = bstart;
19386 +               if (au_test_ro(sb, bstart, dentry->d_inode)) {
19387 +                       err = AuWbrCopyup(au_sbi(sb), dentry);
19388 +                       *bcpup = err;
19389 +                       if (unlikely(err < 0))
19390 +                               goto out;
19391 +               }
19392 +       } else
19393 +               AuDebugOn(bstart < *bcpup
19394 +                         || au_test_ro(sb, *bcpup, dentry->d_inode));
19395 +       AuDbg("bcpup %d, bstart %d\n", *bcpup, bstart);
19396 +
19397 +       if (*bcpup != bstart) {
19398 +               err = au_cpup_dirs(dentry, *bcpup);
19399 +               if (unlikely(err))
19400 +                       goto out;
19401 +               need_wh = 1;
19402 +       } else {
19403 +               struct au_dinfo *dinfo, *tmp;
19404 +
19405 +               need_wh = -ENOMEM;
19406 +               dinfo = au_di(dentry);
19407 +               tmp = au_di_alloc(sb, AuLsc_DI_TMP);
19408 +               if (tmp) {
19409 +                       au_di_cp(tmp, dinfo);
19410 +                       au_di_swap(tmp, dinfo);
19411 +                       /* returns the number of positive dentries */
19412 +                       need_wh = au_lkup_dentry(dentry, bstart + 1, /*type*/0);
19413 +                       au_di_swap(tmp, dinfo);
19414 +                       au_rw_write_unlock(&tmp->di_rwsem);
19415 +                       au_di_free(tmp);
19416 +               }
19417 +       }
19418 +       AuDbg("need_wh %d\n", need_wh);
19419 +       err = need_wh;
19420 +
19421 +out:
19422 +       return err;
19423 +}
19424 +
19425 +/*
19426 + * simple tests for the del-entry operations.
19427 + * following the checks in vfs, plus the parent-child relationship.
19428 + */
19429 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
19430 +              struct dentry *h_parent, int isdir)
19431 +{
19432 +       int err;
19433 +       umode_t h_mode;
19434 +       struct dentry *h_dentry, *h_latest;
19435 +       struct inode *h_inode;
19436 +
19437 +       h_dentry = au_h_dptr(dentry, bindex);
19438 +       h_inode = h_dentry->d_inode;
19439 +       if (dentry->d_inode) {
19440 +               err = -ENOENT;
19441 +               if (unlikely(!h_inode || !h_inode->i_nlink))
19442 +                       goto out;
19443 +
19444 +               h_mode = h_inode->i_mode;
19445 +               if (!isdir) {
19446 +                       err = -EISDIR;
19447 +                       if (unlikely(S_ISDIR(h_mode)))
19448 +                               goto out;
19449 +               } else if (unlikely(!S_ISDIR(h_mode))) {
19450 +                       err = -ENOTDIR;
19451 +                       goto out;
19452 +               }
19453 +       } else {
19454 +               /* rename(2) case */
19455 +               err = -EIO;
19456 +               if (unlikely(h_inode))
19457 +                       goto out;
19458 +       }
19459 +
19460 +       err = -ENOENT;
19461 +       /* expected parent dir is locked */
19462 +       if (unlikely(h_parent != h_dentry->d_parent))
19463 +               goto out;
19464 +       err = 0;
19465 +
19466 +       /*
19467 +        * rmdir a dir may break the consistency on some filesystem.
19468 +        * let's try heavy test.
19469 +        */
19470 +       err = -EACCES;
19471 +       if (unlikely(!au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1)
19472 +                    && au_test_h_perm(h_parent->d_inode,
19473 +                                      MAY_EXEC | MAY_WRITE)))
19474 +               goto out;
19475 +
19476 +       h_latest = au_sio_lkup_one(&dentry->d_name, h_parent);
19477 +       err = -EIO;
19478 +       if (IS_ERR(h_latest))
19479 +               goto out;
19480 +       if (h_latest == h_dentry)
19481 +               err = 0;
19482 +       dput(h_latest);
19483 +
19484 +out:
19485 +       return err;
19486 +}
19487 +
19488 +/*
19489 + * decide the branch where we operate for @dentry. the branch index will be set
19490 + * @rbcpup. after diciding it, 'pin' it and store the timestamps of the parent
19491 + * dir for reverting.
19492 + * when a new whiteout is necessary, create it.
19493 + */
19494 +static struct dentry*
19495 +lock_hdir_create_wh(struct dentry *dentry, int isdir, aufs_bindex_t *rbcpup,
19496 +                   struct au_dtime *dt, struct au_pin *pin)
19497 +{
19498 +       struct dentry *wh_dentry;
19499 +       struct super_block *sb;
19500 +       struct path h_path;
19501 +       int err, need_wh;
19502 +       unsigned int udba;
19503 +       aufs_bindex_t bcpup;
19504 +
19505 +       need_wh = au_wr_dir_need_wh(dentry, isdir, rbcpup);
19506 +       wh_dentry = ERR_PTR(need_wh);
19507 +       if (unlikely(need_wh < 0))
19508 +               goto out;
19509 +
19510 +       sb = dentry->d_sb;
19511 +       udba = au_opt_udba(sb);
19512 +       bcpup = *rbcpup;
19513 +       err = au_pin(pin, dentry, bcpup, udba,
19514 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
19515 +       wh_dentry = ERR_PTR(err);
19516 +       if (unlikely(err))
19517 +               goto out;
19518 +
19519 +       h_path.dentry = au_pinned_h_parent(pin);
19520 +       if (udba != AuOpt_UDBA_NONE
19521 +           && au_dbstart(dentry) == bcpup) {
19522 +               err = au_may_del(dentry, bcpup, h_path.dentry, isdir);
19523 +               wh_dentry = ERR_PTR(err);
19524 +               if (unlikely(err))
19525 +                       goto out_unpin;
19526 +       }
19527 +
19528 +       h_path.mnt = au_sbr_mnt(sb, bcpup);
19529 +       au_dtime_store(dt, au_pinned_parent(pin), &h_path);
19530 +       wh_dentry = NULL;
19531 +       if (!need_wh)
19532 +               goto out; /* success, no need to create whiteout */
19533 +
19534 +       wh_dentry = au_wh_create(dentry, bcpup, h_path.dentry);
19535 +       if (IS_ERR(wh_dentry))
19536 +               goto out_unpin;
19537 +
19538 +       /* returns with the parent is locked and wh_dentry is dget-ed */
19539 +       goto out; /* success */
19540 +
19541 +out_unpin:
19542 +       au_unpin(pin);
19543 +out:
19544 +       return wh_dentry;
19545 +}
19546 +
19547 +/*
19548 + * when removing a dir, rename it to a unique temporary whiteout-ed name first
19549 + * in order to be revertible and save time for removing many child whiteouts
19550 + * under the dir.
19551 + * returns 1 when there are too many child whiteout and caller should remove
19552 + * them asynchronously. returns 0 when the number of children is enough small to
19553 + * remove now or the branch fs is a remote fs.
19554 + * otherwise return an error.
19555 + */
19556 +static int renwh_and_rmdir(struct dentry *dentry, aufs_bindex_t bindex,
19557 +                          struct au_nhash *whlist, struct inode *dir)
19558 +{
19559 +       int rmdir_later, err, dirwh;
19560 +       struct dentry *h_dentry;
19561 +       struct super_block *sb;
19562 +
19563 +       sb = dentry->d_sb;
19564 +       SiMustAnyLock(sb);
19565 +       h_dentry = au_h_dptr(dentry, bindex);
19566 +       err = au_whtmp_ren(h_dentry, au_sbr(sb, bindex));
19567 +       if (unlikely(err))
19568 +               goto out;
19569 +
19570 +       /* stop monitoring */
19571 +       au_hn_free(au_hi(dentry->d_inode, bindex));
19572 +
19573 +       if (!au_test_fs_remote(h_dentry->d_sb)) {
19574 +               dirwh = au_sbi(sb)->si_dirwh;
19575 +               rmdir_later = (dirwh <= 1);
19576 +               if (!rmdir_later)
19577 +                       rmdir_later = au_nhash_test_longer_wh(whlist, bindex,
19578 +                                                             dirwh);
19579 +               if (rmdir_later)
19580 +                       return rmdir_later;
19581 +       }
19582 +
19583 +       err = au_whtmp_rmdir(dir, bindex, h_dentry, whlist);
19584 +       if (unlikely(err)) {
19585 +               AuIOErr("rmdir %pd, b%d failed, %d. ignored\n",
19586 +                       h_dentry, bindex, err);
19587 +               err = 0;
19588 +       }
19589 +
19590 +out:
19591 +       AuTraceErr(err);
19592 +       return err;
19593 +}
19594 +
19595 +/*
19596 + * final procedure for deleting a entry.
19597 + * maintain dentry and iattr.
19598 + */
19599 +static void epilog(struct inode *dir, struct dentry *dentry,
19600 +                  aufs_bindex_t bindex)
19601 +{
19602 +       struct inode *inode;
19603 +
19604 +       inode = dentry->d_inode;
19605 +       d_drop(dentry);
19606 +       inode->i_ctime = dir->i_ctime;
19607 +
19608 +       if (au_ibstart(dir) == bindex)
19609 +               au_cpup_attr_timesizes(dir);
19610 +       dir->i_version++;
19611 +}
19612 +
19613 +/*
19614 + * when an error happened, remove the created whiteout and revert everything.
19615 + */
19616 +static int do_revert(int err, struct inode *dir, aufs_bindex_t bindex,
19617 +                    aufs_bindex_t bwh, struct dentry *wh_dentry,
19618 +                    struct dentry *dentry, struct au_dtime *dt)
19619 +{
19620 +       int rerr;
19621 +       struct path h_path = {
19622 +               .dentry = wh_dentry,
19623 +               .mnt    = au_sbr_mnt(dir->i_sb, bindex)
19624 +       };
19625 +
19626 +       rerr = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path, dentry);
19627 +       if (!rerr) {
19628 +               au_set_dbwh(dentry, bwh);
19629 +               au_dtime_revert(dt);
19630 +               return 0;
19631 +       }
19632 +
19633 +       AuIOErr("%pd reverting whiteout failed(%d, %d)\n", dentry, err, rerr);
19634 +       return -EIO;
19635 +}
19636 +
19637 +/* ---------------------------------------------------------------------- */
19638 +
19639 +int aufs_unlink(struct inode *dir, struct dentry *dentry)
19640 +{
19641 +       int err;
19642 +       aufs_bindex_t bwh, bindex, bstart;
19643 +       struct inode *inode, *h_dir, *delegated;
19644 +       struct dentry *parent, *wh_dentry;
19645 +       /* to reuduce stack size */
19646 +       struct {
19647 +               struct au_dtime dt;
19648 +               struct au_pin pin;
19649 +               struct path h_path;
19650 +       } *a;
19651 +
19652 +       IMustLock(dir);
19653 +
19654 +       err = -ENOMEM;
19655 +       a = kmalloc(sizeof(*a), GFP_NOFS);
19656 +       if (unlikely(!a))
19657 +               goto out;
19658 +
19659 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
19660 +       if (unlikely(err))
19661 +               goto out_free;
19662 +       err = au_d_hashed_positive(dentry);
19663 +       if (unlikely(err))
19664 +               goto out_unlock;
19665 +       inode = dentry->d_inode;
19666 +       IMustLock(inode);
19667 +       err = -EISDIR;
19668 +       if (unlikely(S_ISDIR(inode->i_mode)))
19669 +               goto out_unlock; /* possible? */
19670 +
19671 +       bstart = au_dbstart(dentry);
19672 +       bwh = au_dbwh(dentry);
19673 +       bindex = -1;
19674 +       parent = dentry->d_parent; /* dir inode is locked */
19675 +       di_write_lock_parent(parent);
19676 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/0, &bindex, &a->dt,
19677 +                                       &a->pin);
19678 +       err = PTR_ERR(wh_dentry);
19679 +       if (IS_ERR(wh_dentry))
19680 +               goto out_parent;
19681 +
19682 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, bstart);
19683 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
19684 +       dget(a->h_path.dentry);
19685 +       if (bindex == bstart) {
19686 +               h_dir = au_pinned_h_dir(&a->pin);
19687 +               delegated = NULL;
19688 +               err = vfsub_unlink(h_dir, &a->h_path, &delegated, /*force*/0);
19689 +               if (unlikely(err == -EWOULDBLOCK)) {
19690 +                       pr_warn("cannot retry for NFSv4 delegation"
19691 +                               " for an internal unlink\n");
19692 +                       iput(delegated);
19693 +               }
19694 +       } else {
19695 +               /* dir inode is locked */
19696 +               h_dir = wh_dentry->d_parent->d_inode;
19697 +               IMustLock(h_dir);
19698 +               err = 0;
19699 +       }
19700 +
19701 +       if (!err) {
19702 +               vfsub_drop_nlink(inode);
19703 +               epilog(dir, dentry, bindex);
19704 +
19705 +               /* update target timestamps */
19706 +               if (bindex == bstart) {
19707 +                       vfsub_update_h_iattr(&a->h_path, /*did*/NULL);
19708 +                       /*ignore*/
19709 +                       inode->i_ctime = a->h_path.dentry->d_inode->i_ctime;
19710 +               } else
19711 +                       /* todo: this timestamp may be reverted later */
19712 +                       inode->i_ctime = h_dir->i_ctime;
19713 +               goto out_unpin; /* success */
19714 +       }
19715 +
19716 +       /* revert */
19717 +       if (wh_dentry) {
19718 +               int rerr;
19719 +
19720 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
19721 +                                &a->dt);
19722 +               if (rerr)
19723 +                       err = rerr;
19724 +       }
19725 +
19726 +out_unpin:
19727 +       au_unpin(&a->pin);
19728 +       dput(wh_dentry);
19729 +       dput(a->h_path.dentry);
19730 +out_parent:
19731 +       di_write_unlock(parent);
19732 +out_unlock:
19733 +       aufs_read_unlock(dentry, AuLock_DW);
19734 +out_free:
19735 +       kfree(a);
19736 +out:
19737 +       return err;
19738 +}
19739 +
19740 +int aufs_rmdir(struct inode *dir, struct dentry *dentry)
19741 +{
19742 +       int err, rmdir_later;
19743 +       aufs_bindex_t bwh, bindex, bstart;
19744 +       struct inode *inode;
19745 +       struct dentry *parent, *wh_dentry, *h_dentry;
19746 +       struct au_whtmp_rmdir *args;
19747 +       /* to reuduce stack size */
19748 +       struct {
19749 +               struct au_dtime dt;
19750 +               struct au_pin pin;
19751 +       } *a;
19752 +
19753 +       IMustLock(dir);
19754 +
19755 +       err = -ENOMEM;
19756 +       a = kmalloc(sizeof(*a), GFP_NOFS);
19757 +       if (unlikely(!a))
19758 +               goto out;
19759 +
19760 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
19761 +       if (unlikely(err))
19762 +               goto out_free;
19763 +       err = au_alive_dir(dentry);
19764 +       if (unlikely(err))
19765 +               goto out_unlock;
19766 +       inode = dentry->d_inode;
19767 +       IMustLock(inode);
19768 +       err = -ENOTDIR;
19769 +       if (unlikely(!S_ISDIR(inode->i_mode)))
19770 +               goto out_unlock; /* possible? */
19771 +
19772 +       err = -ENOMEM;
19773 +       args = au_whtmp_rmdir_alloc(dir->i_sb, GFP_NOFS);
19774 +       if (unlikely(!args))
19775 +               goto out_unlock;
19776 +
19777 +       parent = dentry->d_parent; /* dir inode is locked */
19778 +       di_write_lock_parent(parent);
19779 +       err = au_test_empty(dentry, &args->whlist);
19780 +       if (unlikely(err))
19781 +               goto out_parent;
19782 +
19783 +       bstart = au_dbstart(dentry);
19784 +       bwh = au_dbwh(dentry);
19785 +       bindex = -1;
19786 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/1, &bindex, &a->dt,
19787 +                                       &a->pin);
19788 +       err = PTR_ERR(wh_dentry);
19789 +       if (IS_ERR(wh_dentry))
19790 +               goto out_parent;
19791 +
19792 +       h_dentry = au_h_dptr(dentry, bstart);
19793 +       dget(h_dentry);
19794 +       rmdir_later = 0;
19795 +       if (bindex == bstart) {
19796 +               err = renwh_and_rmdir(dentry, bstart, &args->whlist, dir);
19797 +               if (err > 0) {
19798 +                       rmdir_later = err;
19799 +                       err = 0;
19800 +               }
19801 +       } else {
19802 +               /* stop monitoring */
19803 +               au_hn_free(au_hi(inode, bstart));
19804 +
19805 +               /* dir inode is locked */
19806 +               IMustLock(wh_dentry->d_parent->d_inode);
19807 +               err = 0;
19808 +       }
19809 +
19810 +       if (!err) {
19811 +               vfsub_dead_dir(inode);
19812 +               au_set_dbdiropq(dentry, -1);
19813 +               epilog(dir, dentry, bindex);
19814 +
19815 +               if (rmdir_later) {
19816 +                       au_whtmp_kick_rmdir(dir, bstart, h_dentry, args);
19817 +                       args = NULL;
19818 +               }
19819 +
19820 +               goto out_unpin; /* success */
19821 +       }
19822 +
19823 +       /* revert */
19824 +       AuLabel(revert);
19825 +       if (wh_dentry) {
19826 +               int rerr;
19827 +
19828 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
19829 +                                &a->dt);
19830 +               if (rerr)
19831 +                       err = rerr;
19832 +       }
19833 +
19834 +out_unpin:
19835 +       au_unpin(&a->pin);
19836 +       dput(wh_dentry);
19837 +       dput(h_dentry);
19838 +out_parent:
19839 +       di_write_unlock(parent);
19840 +       if (args)
19841 +               au_whtmp_rmdir_free(args);
19842 +out_unlock:
19843 +       aufs_read_unlock(dentry, AuLock_DW);
19844 +out_free:
19845 +       kfree(a);
19846 +out:
19847 +       AuTraceErr(err);
19848 +       return err;
19849 +}
19850 diff -urN /usr/share/empty/fs/aufs/i_op_ren.c linux/fs/aufs/i_op_ren.c
19851 --- /usr/share/empty/fs/aufs/i_op_ren.c 1970-01-01 01:00:00.000000000 +0100
19852 +++ linux/fs/aufs/i_op_ren.c    2015-01-25 13:00:38.631047076 +0100
19853 @@ -0,0 +1,1034 @@
19854 +/*
19855 + * Copyright (C) 2005-2014 Junjiro R. Okajima
19856 + *
19857 + * This program, aufs is free software; you can redistribute it and/or modify
19858 + * it under the terms of the GNU General Public License as published by
19859 + * the Free Software Foundation; either version 2 of the License, or
19860 + * (at your option) any later version.
19861 + *
19862 + * This program is distributed in the hope that it will be useful,
19863 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19864 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19865 + * GNU General Public License for more details.
19866 + *
19867 + * You should have received a copy of the GNU General Public License
19868 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19869 + */
19870 +
19871 +/*
19872 + * inode operation (rename entry)
19873 + * todo: this is crazy monster
19874 + */
19875 +
19876 +#include "aufs.h"
19877 +
19878 +enum { AuSRC, AuDST, AuSrcDst };
19879 +enum { AuPARENT, AuCHILD, AuParentChild };
19880 +
19881 +#define AuRen_ISDIR    1
19882 +#define AuRen_ISSAMEDIR        (1 << 1)
19883 +#define AuRen_WHSRC    (1 << 2)
19884 +#define AuRen_WHDST    (1 << 3)
19885 +#define AuRen_MNT_WRITE        (1 << 4)
19886 +#define AuRen_DT_DSTDIR        (1 << 5)
19887 +#define AuRen_DIROPQ   (1 << 6)
19888 +#define AuRen_CPUP     (1 << 7)
19889 +#define au_ftest_ren(flags, name)      ((flags) & AuRen_##name)
19890 +#define au_fset_ren(flags, name) \
19891 +       do { (flags) |= AuRen_##name; } while (0)
19892 +#define au_fclr_ren(flags, name) \
19893 +       do { (flags) &= ~AuRen_##name; } while (0)
19894 +
19895 +struct au_ren_args {
19896 +       struct {
19897 +               struct dentry *dentry, *h_dentry, *parent, *h_parent,
19898 +                       *wh_dentry;
19899 +               struct inode *dir, *inode;
19900 +               struct au_hinode *hdir;
19901 +               struct au_dtime dt[AuParentChild];
19902 +               aufs_bindex_t bstart;
19903 +       } sd[AuSrcDst];
19904 +
19905 +#define src_dentry     sd[AuSRC].dentry
19906 +#define src_dir                sd[AuSRC].dir
19907 +#define src_inode      sd[AuSRC].inode
19908 +#define src_h_dentry   sd[AuSRC].h_dentry
19909 +#define src_parent     sd[AuSRC].parent
19910 +#define src_h_parent   sd[AuSRC].h_parent
19911 +#define src_wh_dentry  sd[AuSRC].wh_dentry
19912 +#define src_hdir       sd[AuSRC].hdir
19913 +#define src_h_dir      sd[AuSRC].hdir->hi_inode
19914 +#define src_dt         sd[AuSRC].dt
19915 +#define src_bstart     sd[AuSRC].bstart
19916 +
19917 +#define dst_dentry     sd[AuDST].dentry
19918 +#define dst_dir                sd[AuDST].dir
19919 +#define dst_inode      sd[AuDST].inode
19920 +#define dst_h_dentry   sd[AuDST].h_dentry
19921 +#define dst_parent     sd[AuDST].parent
19922 +#define dst_h_parent   sd[AuDST].h_parent
19923 +#define dst_wh_dentry  sd[AuDST].wh_dentry
19924 +#define dst_hdir       sd[AuDST].hdir
19925 +#define dst_h_dir      sd[AuDST].hdir->hi_inode
19926 +#define dst_dt         sd[AuDST].dt
19927 +#define dst_bstart     sd[AuDST].bstart
19928 +
19929 +       struct dentry *h_trap;
19930 +       struct au_branch *br;
19931 +       struct au_hinode *src_hinode;
19932 +       struct path h_path;
19933 +       struct au_nhash whlist;
19934 +       aufs_bindex_t btgt, src_bwh, src_bdiropq;
19935 +
19936 +       unsigned int flags;
19937 +
19938 +       struct au_whtmp_rmdir *thargs;
19939 +       struct dentry *h_dst;
19940 +};
19941 +
19942 +/* ---------------------------------------------------------------------- */
19943 +
19944 +/*
19945 + * functions for reverting.
19946 + * when an error happened in a single rename systemcall, we should revert
19947 + * everything as if nothing happend.
19948 + * we don't need to revert the copied-up/down the parent dir since they are
19949 + * harmless.
19950 + */
19951 +
19952 +#define RevertFailure(fmt, ...) do { \
19953 +       AuIOErr("revert failure: " fmt " (%d, %d)\n", \
19954 +               ##__VA_ARGS__, err, rerr); \
19955 +       err = -EIO; \
19956 +} while (0)
19957 +
19958 +static void au_ren_rev_diropq(int err, struct au_ren_args *a)
19959 +{
19960 +       int rerr;
19961 +
19962 +       au_hn_imtx_lock_nested(a->src_hinode, AuLsc_I_CHILD);
19963 +       rerr = au_diropq_remove(a->src_dentry, a->btgt);
19964 +       au_hn_imtx_unlock(a->src_hinode);
19965 +       au_set_dbdiropq(a->src_dentry, a->src_bdiropq);
19966 +       if (rerr)
19967 +               RevertFailure("remove diropq %pd", a->src_dentry);
19968 +}
19969 +
19970 +static void au_ren_rev_rename(int err, struct au_ren_args *a)
19971 +{
19972 +       int rerr;
19973 +       struct inode *delegated;
19974 +
19975 +       a->h_path.dentry = vfsub_lkup_one(&a->src_dentry->d_name,
19976 +                                         a->src_h_parent);
19977 +       rerr = PTR_ERR(a->h_path.dentry);
19978 +       if (IS_ERR(a->h_path.dentry)) {
19979 +               RevertFailure("lkup one %pd", a->src_dentry);
19980 +               return;
19981 +       }
19982 +
19983 +       delegated = NULL;
19984 +       rerr = vfsub_rename(a->dst_h_dir,
19985 +                           au_h_dptr(a->src_dentry, a->btgt),
19986 +                           a->src_h_dir, &a->h_path, &delegated);
19987 +       if (unlikely(rerr == -EWOULDBLOCK)) {
19988 +               pr_warn("cannot retry for NFSv4 delegation"
19989 +                       " for an internal rename\n");
19990 +               iput(delegated);
19991 +       }
19992 +       d_drop(a->h_path.dentry);
19993 +       dput(a->h_path.dentry);
19994 +       /* au_set_h_dptr(a->src_dentry, a->btgt, NULL); */
19995 +       if (rerr)
19996 +               RevertFailure("rename %pd", a->src_dentry);
19997 +}
19998 +
19999 +static void au_ren_rev_cpup(int err, struct au_ren_args *a)
20000 +{
20001 +       int rerr;
20002 +
20003 +       a->h_path.dentry = a->dst_h_dentry;
20004 +       /* no delegation since it is just created */
20005 +       rerr = vfsub_unlink(a->dst_h_dir, &a->h_path, /*delegated*/NULL,
20006 +                           /*force*/0);
20007 +       au_set_h_dptr(a->src_dentry, a->btgt, NULL);
20008 +       au_set_dbstart(a->src_dentry, a->src_bstart);
20009 +       if (rerr)
20010 +               RevertFailure("unlink %pd", a->dst_h_dentry);
20011 +}
20012 +
20013 +static void au_ren_rev_whtmp(int err, struct au_ren_args *a)
20014 +{
20015 +       int rerr;
20016 +       struct inode *delegated;
20017 +
20018 +       a->h_path.dentry = vfsub_lkup_one(&a->dst_dentry->d_name,
20019 +                                         a->dst_h_parent);
20020 +       rerr = PTR_ERR(a->h_path.dentry);
20021 +       if (IS_ERR(a->h_path.dentry)) {
20022 +               RevertFailure("lkup one %pd", a->dst_dentry);
20023 +               return;
20024 +       }
20025 +       if (a->h_path.dentry->d_inode) {
20026 +               d_drop(a->h_path.dentry);
20027 +               dput(a->h_path.dentry);
20028 +               return;
20029 +       }
20030 +
20031 +       delegated = NULL;
20032 +       rerr = vfsub_rename(a->dst_h_dir, a->h_dst, a->dst_h_dir, &a->h_path,
20033 +                           &delegated);
20034 +       if (unlikely(rerr == -EWOULDBLOCK)) {
20035 +               pr_warn("cannot retry for NFSv4 delegation"
20036 +                       " for an internal rename\n");
20037 +               iput(delegated);
20038 +       }
20039 +       d_drop(a->h_path.dentry);
20040 +       dput(a->h_path.dentry);
20041 +       if (!rerr)
20042 +               au_set_h_dptr(a->dst_dentry, a->btgt, dget(a->h_dst));
20043 +       else
20044 +               RevertFailure("rename %pd", a->h_dst);
20045 +}
20046 +
20047 +static void au_ren_rev_whsrc(int err, struct au_ren_args *a)
20048 +{
20049 +       int rerr;
20050 +
20051 +       a->h_path.dentry = a->src_wh_dentry;
20052 +       rerr = au_wh_unlink_dentry(a->src_h_dir, &a->h_path, a->src_dentry);
20053 +       au_set_dbwh(a->src_dentry, a->src_bwh);
20054 +       if (rerr)
20055 +               RevertFailure("unlink %pd", a->src_wh_dentry);
20056 +}
20057 +#undef RevertFailure
20058 +
20059 +/* ---------------------------------------------------------------------- */
20060 +
20061 +/*
20062 + * when we have to copyup the renaming entry, do it with the rename-target name
20063 + * in order to minimize the cost (the later actual rename is unnecessary).
20064 + * otherwise rename it on the target branch.
20065 + */
20066 +static int au_ren_or_cpup(struct au_ren_args *a)
20067 +{
20068 +       int err;
20069 +       struct dentry *d;
20070 +       struct inode *delegated;
20071 +
20072 +       d = a->src_dentry;
20073 +       if (au_dbstart(d) == a->btgt) {
20074 +               a->h_path.dentry = a->dst_h_dentry;
20075 +               if (au_ftest_ren(a->flags, DIROPQ)
20076 +                   && au_dbdiropq(d) == a->btgt)
20077 +                       au_fclr_ren(a->flags, DIROPQ);
20078 +               AuDebugOn(au_dbstart(d) != a->btgt);
20079 +               delegated = NULL;
20080 +               err = vfsub_rename(a->src_h_dir, au_h_dptr(d, a->btgt),
20081 +                                  a->dst_h_dir, &a->h_path, &delegated);
20082 +               if (unlikely(err == -EWOULDBLOCK)) {
20083 +                       pr_warn("cannot retry for NFSv4 delegation"
20084 +                               " for an internal rename\n");
20085 +                       iput(delegated);
20086 +               }
20087 +       } else
20088 +               BUG();
20089 +
20090 +       if (!err && a->h_dst)
20091 +               /* it will be set to dinfo later */
20092 +               dget(a->h_dst);
20093 +
20094 +       return err;
20095 +}
20096 +
20097 +/* cf. aufs_rmdir() */
20098 +static int au_ren_del_whtmp(struct au_ren_args *a)
20099 +{
20100 +       int err;
20101 +       struct inode *dir;
20102 +
20103 +       dir = a->dst_dir;
20104 +       SiMustAnyLock(dir->i_sb);
20105 +       if (!au_nhash_test_longer_wh(&a->whlist, a->btgt,
20106 +                                    au_sbi(dir->i_sb)->si_dirwh)
20107 +           || au_test_fs_remote(a->h_dst->d_sb)) {
20108 +               err = au_whtmp_rmdir(dir, a->btgt, a->h_dst, &a->whlist);
20109 +               if (unlikely(err))
20110 +                       pr_warn("failed removing whtmp dir %pd (%d), "
20111 +                               "ignored.\n", a->h_dst, err);
20112 +       } else {
20113 +               au_nhash_wh_free(&a->thargs->whlist);
20114 +               a->thargs->whlist = a->whlist;
20115 +               a->whlist.nh_num = 0;
20116 +               au_whtmp_kick_rmdir(dir, a->btgt, a->h_dst, a->thargs);
20117 +               dput(a->h_dst);
20118 +               a->thargs = NULL;
20119 +       }
20120 +
20121 +       return 0;
20122 +}
20123 +
20124 +/* make it 'opaque' dir. */
20125 +static int au_ren_diropq(struct au_ren_args *a)
20126 +{
20127 +       int err;
20128 +       struct dentry *diropq;
20129 +
20130 +       err = 0;
20131 +       a->src_bdiropq = au_dbdiropq(a->src_dentry);
20132 +       a->src_hinode = au_hi(a->src_inode, a->btgt);
20133 +       au_hn_imtx_lock_nested(a->src_hinode, AuLsc_I_CHILD);
20134 +       diropq = au_diropq_create(a->src_dentry, a->btgt);
20135 +       au_hn_imtx_unlock(a->src_hinode);
20136 +       if (IS_ERR(diropq))
20137 +               err = PTR_ERR(diropq);
20138 +       else
20139 +               dput(diropq);
20140 +
20141 +       return err;
20142 +}
20143 +
20144 +static int do_rename(struct au_ren_args *a)
20145 +{
20146 +       int err;
20147 +       struct dentry *d, *h_d;
20148 +
20149 +       /* prepare workqueue args for asynchronous rmdir */
20150 +       h_d = a->dst_h_dentry;
20151 +       if (au_ftest_ren(a->flags, ISDIR) && h_d->d_inode) {
20152 +               err = -ENOMEM;
20153 +               a->thargs = au_whtmp_rmdir_alloc(a->src_dentry->d_sb, GFP_NOFS);
20154 +               if (unlikely(!a->thargs))
20155 +                       goto out;
20156 +               a->h_dst = dget(h_d);
20157 +       }
20158 +
20159 +       /* create whiteout for src_dentry */
20160 +       if (au_ftest_ren(a->flags, WHSRC)) {
20161 +               a->src_bwh = au_dbwh(a->src_dentry);
20162 +               AuDebugOn(a->src_bwh >= 0);
20163 +               a->src_wh_dentry
20164 +                       = au_wh_create(a->src_dentry, a->btgt, a->src_h_parent);
20165 +               err = PTR_ERR(a->src_wh_dentry);
20166 +               if (IS_ERR(a->src_wh_dentry))
20167 +                       goto out_thargs;
20168 +       }
20169 +
20170 +       /* lookup whiteout for dentry */
20171 +       if (au_ftest_ren(a->flags, WHDST)) {
20172 +               h_d = au_wh_lkup(a->dst_h_parent, &a->dst_dentry->d_name,
20173 +                                a->br);
20174 +               err = PTR_ERR(h_d);
20175 +               if (IS_ERR(h_d))
20176 +                       goto out_whsrc;
20177 +               if (!h_d->d_inode)
20178 +                       dput(h_d);
20179 +               else
20180 +                       a->dst_wh_dentry = h_d;
20181 +       }
20182 +
20183 +       /* rename dentry to tmpwh */
20184 +       if (a->thargs) {
20185 +               err = au_whtmp_ren(a->dst_h_dentry, a->br);
20186 +               if (unlikely(err))
20187 +                       goto out_whdst;
20188 +
20189 +               d = a->dst_dentry;
20190 +               au_set_h_dptr(d, a->btgt, NULL);
20191 +               err = au_lkup_neg(d, a->btgt, /*wh*/0);
20192 +               if (unlikely(err))
20193 +                       goto out_whtmp;
20194 +               a->dst_h_dentry = au_h_dptr(d, a->btgt);
20195 +       }
20196 +
20197 +       BUG_ON(a->dst_h_dentry->d_inode && a->src_bstart != a->btgt);
20198 +
20199 +       /* rename by vfs_rename or cpup */
20200 +       d = a->dst_dentry;
20201 +       if (au_ftest_ren(a->flags, ISDIR)
20202 +           && (a->dst_wh_dentry
20203 +               || au_dbdiropq(d) == a->btgt
20204 +               /* hide the lower to keep xino */
20205 +               || a->btgt < au_dbend(d)
20206 +               || au_opt_test(au_mntflags(d->d_sb), ALWAYS_DIROPQ)))
20207 +               au_fset_ren(a->flags, DIROPQ);
20208 +       err = au_ren_or_cpup(a);
20209 +       if (unlikely(err))
20210 +               /* leave the copied-up one */
20211 +               goto out_whtmp;
20212 +
20213 +       /* make dir opaque */
20214 +       if (au_ftest_ren(a->flags, DIROPQ)) {
20215 +               err = au_ren_diropq(a);
20216 +               if (unlikely(err))
20217 +                       goto out_rename;
20218 +       }
20219 +
20220 +       /* update target timestamps */
20221 +       AuDebugOn(au_dbstart(a->src_dentry) != a->btgt);
20222 +       a->h_path.dentry = au_h_dptr(a->src_dentry, a->btgt);
20223 +       vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
20224 +       a->src_inode->i_ctime = a->h_path.dentry->d_inode->i_ctime;
20225 +
20226 +       /* remove whiteout for dentry */
20227 +       if (a->dst_wh_dentry) {
20228 +               a->h_path.dentry = a->dst_wh_dentry;
20229 +               err = au_wh_unlink_dentry(a->dst_h_dir, &a->h_path,
20230 +                                         a->dst_dentry);
20231 +               if (unlikely(err))
20232 +                       goto out_diropq;
20233 +       }
20234 +
20235 +       /* remove whtmp */
20236 +       if (a->thargs)
20237 +               au_ren_del_whtmp(a); /* ignore this error */
20238 +
20239 +       au_fhsm_wrote(a->src_dentry->d_sb, a->btgt, /*force*/0);
20240 +       err = 0;
20241 +       goto out_success;
20242 +
20243 +out_diropq:
20244 +       if (au_ftest_ren(a->flags, DIROPQ))
20245 +               au_ren_rev_diropq(err, a);
20246 +out_rename:
20247 +       if (!au_ftest_ren(a->flags, CPUP))
20248 +               au_ren_rev_rename(err, a);
20249 +       else
20250 +               au_ren_rev_cpup(err, a);
20251 +       dput(a->h_dst);
20252 +out_whtmp:
20253 +       if (a->thargs)
20254 +               au_ren_rev_whtmp(err, a);
20255 +out_whdst:
20256 +       dput(a->dst_wh_dentry);
20257 +       a->dst_wh_dentry = NULL;
20258 +out_whsrc:
20259 +       if (a->src_wh_dentry)
20260 +               au_ren_rev_whsrc(err, a);
20261 +out_success:
20262 +       dput(a->src_wh_dentry);
20263 +       dput(a->dst_wh_dentry);
20264 +out_thargs:
20265 +       if (a->thargs) {
20266 +               dput(a->h_dst);
20267 +               au_whtmp_rmdir_free(a->thargs);
20268 +               a->thargs = NULL;
20269 +       }
20270 +out:
20271 +       return err;
20272 +}
20273 +
20274 +/* ---------------------------------------------------------------------- */
20275 +
20276 +/*
20277 + * test if @dentry dir can be rename destination or not.
20278 + * success means, it is a logically empty dir.
20279 + */
20280 +static int may_rename_dstdir(struct dentry *dentry, struct au_nhash *whlist)
20281 +{
20282 +       return au_test_empty(dentry, whlist);
20283 +}
20284 +
20285 +/*
20286 + * test if @dentry dir can be rename source or not.
20287 + * if it can, return 0 and @children is filled.
20288 + * success means,
20289 + * - it is a logically empty dir.
20290 + * - or, it exists on writable branch and has no children including whiteouts
20291 + *       on the lower branch.
20292 + */
20293 +static int may_rename_srcdir(struct dentry *dentry, aufs_bindex_t btgt)
20294 +{
20295 +       int err;
20296 +       unsigned int rdhash;
20297 +       aufs_bindex_t bstart;
20298 +
20299 +       bstart = au_dbstart(dentry);
20300 +       if (bstart != btgt) {
20301 +               struct au_nhash whlist;
20302 +
20303 +               SiMustAnyLock(dentry->d_sb);
20304 +               rdhash = au_sbi(dentry->d_sb)->si_rdhash;
20305 +               if (!rdhash)
20306 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL,
20307 +                                                          dentry));
20308 +               err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
20309 +               if (unlikely(err))
20310 +                       goto out;
20311 +               err = au_test_empty(dentry, &whlist);
20312 +               au_nhash_wh_free(&whlist);
20313 +               goto out;
20314 +       }
20315 +
20316 +       if (bstart == au_dbtaildir(dentry))
20317 +               return 0; /* success */
20318 +
20319 +       err = au_test_empty_lower(dentry);
20320 +
20321 +out:
20322 +       if (err == -ENOTEMPTY) {
20323 +               AuWarn1("renaming dir who has child(ren) on multiple branches,"
20324 +                       " is not supported\n");
20325 +               err = -EXDEV;
20326 +       }
20327 +       return err;
20328 +}
20329 +
20330 +/* side effect: sets whlist and h_dentry */
20331 +static int au_ren_may_dir(struct au_ren_args *a)
20332 +{
20333 +       int err;
20334 +       unsigned int rdhash;
20335 +       struct dentry *d;
20336 +
20337 +       d = a->dst_dentry;
20338 +       SiMustAnyLock(d->d_sb);
20339 +
20340 +       err = 0;
20341 +       if (au_ftest_ren(a->flags, ISDIR) && a->dst_inode) {
20342 +               rdhash = au_sbi(d->d_sb)->si_rdhash;
20343 +               if (!rdhash)
20344 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, d));
20345 +               err = au_nhash_alloc(&a->whlist, rdhash, GFP_NOFS);
20346 +               if (unlikely(err))
20347 +                       goto out;
20348 +
20349 +               au_set_dbstart(d, a->dst_bstart);
20350 +               err = may_rename_dstdir(d, &a->whlist);
20351 +               au_set_dbstart(d, a->btgt);
20352 +       }
20353 +       a->dst_h_dentry = au_h_dptr(d, au_dbstart(d));
20354 +       if (unlikely(err))
20355 +               goto out;
20356 +
20357 +       d = a->src_dentry;
20358 +       a->src_h_dentry = au_h_dptr(d, au_dbstart(d));
20359 +       if (au_ftest_ren(a->flags, ISDIR)) {
20360 +               err = may_rename_srcdir(d, a->btgt);
20361 +               if (unlikely(err)) {
20362 +                       au_nhash_wh_free(&a->whlist);
20363 +                       a->whlist.nh_num = 0;
20364 +               }
20365 +       }
20366 +out:
20367 +       return err;
20368 +}
20369 +
20370 +/* ---------------------------------------------------------------------- */
20371 +
20372 +/*
20373 + * simple tests for rename.
20374 + * following the checks in vfs, plus the parent-child relationship.
20375 + */
20376 +static int au_may_ren(struct au_ren_args *a)
20377 +{
20378 +       int err, isdir;
20379 +       struct inode *h_inode;
20380 +
20381 +       if (a->src_bstart == a->btgt) {
20382 +               err = au_may_del(a->src_dentry, a->btgt, a->src_h_parent,
20383 +                                au_ftest_ren(a->flags, ISDIR));
20384 +               if (unlikely(err))
20385 +                       goto out;
20386 +               err = -EINVAL;
20387 +               if (unlikely(a->src_h_dentry == a->h_trap))
20388 +                       goto out;
20389 +       }
20390 +
20391 +       err = 0;
20392 +       if (a->dst_bstart != a->btgt)
20393 +               goto out;
20394 +
20395 +       err = -ENOTEMPTY;
20396 +       if (unlikely(a->dst_h_dentry == a->h_trap))
20397 +               goto out;
20398 +
20399 +       err = -EIO;
20400 +       h_inode = a->dst_h_dentry->d_inode;
20401 +       isdir = !!au_ftest_ren(a->flags, ISDIR);
20402 +       if (!a->dst_dentry->d_inode) {
20403 +               if (unlikely(h_inode))
20404 +                       goto out;
20405 +               err = au_may_add(a->dst_dentry, a->btgt, a->dst_h_parent,
20406 +                                isdir);
20407 +       } else {
20408 +               if (unlikely(!h_inode || !h_inode->i_nlink))
20409 +                       goto out;
20410 +               err = au_may_del(a->dst_dentry, a->btgt, a->dst_h_parent,
20411 +                                isdir);
20412 +               if (unlikely(err))
20413 +                       goto out;
20414 +       }
20415 +
20416 +out:
20417 +       if (unlikely(err == -ENOENT || err == -EEXIST))
20418 +               err = -EIO;
20419 +       AuTraceErr(err);
20420 +       return err;
20421 +}
20422 +
20423 +/* ---------------------------------------------------------------------- */
20424 +
20425 +/*
20426 + * locking order
20427 + * (VFS)
20428 + * - src_dir and dir by lock_rename()
20429 + * - inode if exitsts
20430 + * (aufs)
20431 + * - lock all
20432 + *   + src_dentry and dentry by aufs_read_and_write_lock2() which calls,
20433 + *     + si_read_lock
20434 + *     + di_write_lock2_child()
20435 + *       + di_write_lock_child()
20436 + *        + ii_write_lock_child()
20437 + *       + di_write_lock_child2()
20438 + *        + ii_write_lock_child2()
20439 + *     + src_parent and parent
20440 + *       + di_write_lock_parent()
20441 + *        + ii_write_lock_parent()
20442 + *       + di_write_lock_parent2()
20443 + *        + ii_write_lock_parent2()
20444 + *   + lower src_dir and dir by vfsub_lock_rename()
20445 + *   + verify the every relationships between child and parent. if any
20446 + *     of them failed, unlock all and return -EBUSY.
20447 + */
20448 +static void au_ren_unlock(struct au_ren_args *a)
20449 +{
20450 +       vfsub_unlock_rename(a->src_h_parent, a->src_hdir,
20451 +                           a->dst_h_parent, a->dst_hdir);
20452 +       if (au_ftest_ren(a->flags, MNT_WRITE))
20453 +               vfsub_mnt_drop_write(au_br_mnt(a->br));
20454 +}
20455 +
20456 +static int au_ren_lock(struct au_ren_args *a)
20457 +{
20458 +       int err;
20459 +       unsigned int udba;
20460 +
20461 +       err = 0;
20462 +       a->src_h_parent = au_h_dptr(a->src_parent, a->btgt);
20463 +       a->src_hdir = au_hi(a->src_dir, a->btgt);
20464 +       a->dst_h_parent = au_h_dptr(a->dst_parent, a->btgt);
20465 +       a->dst_hdir = au_hi(a->dst_dir, a->btgt);
20466 +
20467 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
20468 +       if (unlikely(err))
20469 +               goto out;
20470 +       au_fset_ren(a->flags, MNT_WRITE);
20471 +       a->h_trap = vfsub_lock_rename(a->src_h_parent, a->src_hdir,
20472 +                                     a->dst_h_parent, a->dst_hdir);
20473 +       udba = au_opt_udba(a->src_dentry->d_sb);
20474 +       if (unlikely(a->src_hdir->hi_inode != a->src_h_parent->d_inode
20475 +                    || a->dst_hdir->hi_inode != a->dst_h_parent->d_inode))
20476 +               err = au_busy_or_stale();
20477 +       if (!err && au_dbstart(a->src_dentry) == a->btgt)
20478 +               err = au_h_verify(a->src_h_dentry, udba,
20479 +                                 a->src_h_parent->d_inode, a->src_h_parent,
20480 +                                 a->br);
20481 +       if (!err && au_dbstart(a->dst_dentry) == a->btgt)
20482 +               err = au_h_verify(a->dst_h_dentry, udba,
20483 +                                 a->dst_h_parent->d_inode, a->dst_h_parent,
20484 +                                 a->br);
20485 +       if (!err)
20486 +               goto out; /* success */
20487 +
20488 +       err = au_busy_or_stale();
20489 +       au_ren_unlock(a);
20490 +
20491 +out:
20492 +       return err;
20493 +}
20494 +
20495 +/* ---------------------------------------------------------------------- */
20496 +
20497 +static void au_ren_refresh_dir(struct au_ren_args *a)
20498 +{
20499 +       struct inode *dir;
20500 +
20501 +       dir = a->dst_dir;
20502 +       dir->i_version++;
20503 +       if (au_ftest_ren(a->flags, ISDIR)) {
20504 +               /* is this updating defined in POSIX? */
20505 +               au_cpup_attr_timesizes(a->src_inode);
20506 +               au_cpup_attr_nlink(dir, /*force*/1);
20507 +       }
20508 +
20509 +       if (au_ibstart(dir) == a->btgt)
20510 +               au_cpup_attr_timesizes(dir);
20511 +
20512 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
20513 +               return;
20514 +
20515 +       dir = a->src_dir;
20516 +       dir->i_version++;
20517 +       if (au_ftest_ren(a->flags, ISDIR))
20518 +               au_cpup_attr_nlink(dir, /*force*/1);
20519 +       if (au_ibstart(dir) == a->btgt)
20520 +               au_cpup_attr_timesizes(dir);
20521 +}
20522 +
20523 +static void au_ren_refresh(struct au_ren_args *a)
20524 +{
20525 +       aufs_bindex_t bend, bindex;
20526 +       struct dentry *d, *h_d;
20527 +       struct inode *i, *h_i;
20528 +       struct super_block *sb;
20529 +
20530 +       d = a->dst_dentry;
20531 +       d_drop(d);
20532 +       if (a->h_dst)
20533 +               /* already dget-ed by au_ren_or_cpup() */
20534 +               au_set_h_dptr(d, a->btgt, a->h_dst);
20535 +
20536 +       i = a->dst_inode;
20537 +       if (i) {
20538 +               if (!au_ftest_ren(a->flags, ISDIR))
20539 +                       vfsub_drop_nlink(i);
20540 +               else {
20541 +                       vfsub_dead_dir(i);
20542 +                       au_cpup_attr_timesizes(i);
20543 +               }
20544 +               au_update_dbrange(d, /*do_put_zero*/1);
20545 +       } else {
20546 +               bend = a->btgt;
20547 +               for (bindex = au_dbstart(d); bindex < bend; bindex++)
20548 +                       au_set_h_dptr(d, bindex, NULL);
20549 +               bend = au_dbend(d);
20550 +               for (bindex = a->btgt + 1; bindex <= bend; bindex++)
20551 +                       au_set_h_dptr(d, bindex, NULL);
20552 +               au_update_dbrange(d, /*do_put_zero*/0);
20553 +       }
20554 +
20555 +       d = a->src_dentry;
20556 +       au_set_dbwh(d, -1);
20557 +       bend = au_dbend(d);
20558 +       for (bindex = a->btgt + 1; bindex <= bend; bindex++) {
20559 +               h_d = au_h_dptr(d, bindex);
20560 +               if (h_d)
20561 +                       au_set_h_dptr(d, bindex, NULL);
20562 +       }
20563 +       au_set_dbend(d, a->btgt);
20564 +
20565 +       sb = d->d_sb;
20566 +       i = a->src_inode;
20567 +       if (au_opt_test(au_mntflags(sb), PLINK) && au_plink_test(i))
20568 +               return; /* success */
20569 +
20570 +       bend = au_ibend(i);
20571 +       for (bindex = a->btgt + 1; bindex <= bend; bindex++) {
20572 +               h_i = au_h_iptr(i, bindex);
20573 +               if (h_i) {
20574 +                       au_xino_write(sb, bindex, h_i->i_ino, /*ino*/0);
20575 +                       /* ignore this error */
20576 +                       au_set_h_iptr(i, bindex, NULL, 0);
20577 +               }
20578 +       }
20579 +       au_set_ibend(i, a->btgt);
20580 +}
20581 +
20582 +/* ---------------------------------------------------------------------- */
20583 +
20584 +/* mainly for link(2) and rename(2) */
20585 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt)
20586 +{
20587 +       aufs_bindex_t bdiropq, bwh;
20588 +       struct dentry *parent;
20589 +       struct au_branch *br;
20590 +
20591 +       parent = dentry->d_parent;
20592 +       IMustLock(parent->d_inode); /* dir is locked */
20593 +
20594 +       bdiropq = au_dbdiropq(parent);
20595 +       bwh = au_dbwh(dentry);
20596 +       br = au_sbr(dentry->d_sb, btgt);
20597 +       if (au_br_rdonly(br)
20598 +           || (0 <= bdiropq && bdiropq < btgt)
20599 +           || (0 <= bwh && bwh < btgt))
20600 +               btgt = -1;
20601 +
20602 +       AuDbg("btgt %d\n", btgt);
20603 +       return btgt;
20604 +}
20605 +
20606 +/* sets src_bstart, dst_bstart and btgt */
20607 +static int au_ren_wbr(struct au_ren_args *a)
20608 +{
20609 +       int err;
20610 +       struct au_wr_dir_args wr_dir_args = {
20611 +               /* .force_btgt  = -1, */
20612 +               .flags          = AuWrDir_ADD_ENTRY
20613 +       };
20614 +
20615 +       a->src_bstart = au_dbstart(a->src_dentry);
20616 +       a->dst_bstart = au_dbstart(a->dst_dentry);
20617 +       if (au_ftest_ren(a->flags, ISDIR))
20618 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
20619 +       wr_dir_args.force_btgt = a->src_bstart;
20620 +       if (a->dst_inode && a->dst_bstart < a->src_bstart)
20621 +               wr_dir_args.force_btgt = a->dst_bstart;
20622 +       wr_dir_args.force_btgt = au_wbr(a->dst_dentry, wr_dir_args.force_btgt);
20623 +       err = au_wr_dir(a->dst_dentry, a->src_dentry, &wr_dir_args);
20624 +       a->btgt = err;
20625 +
20626 +       return err;
20627 +}
20628 +
20629 +static void au_ren_dt(struct au_ren_args *a)
20630 +{
20631 +       a->h_path.dentry = a->src_h_parent;
20632 +       au_dtime_store(a->src_dt + AuPARENT, a->src_parent, &a->h_path);
20633 +       if (!au_ftest_ren(a->flags, ISSAMEDIR)) {
20634 +               a->h_path.dentry = a->dst_h_parent;
20635 +               au_dtime_store(a->dst_dt + AuPARENT, a->dst_parent, &a->h_path);
20636 +       }
20637 +
20638 +       au_fclr_ren(a->flags, DT_DSTDIR);
20639 +       if (!au_ftest_ren(a->flags, ISDIR))
20640 +               return;
20641 +
20642 +       a->h_path.dentry = a->src_h_dentry;
20643 +       au_dtime_store(a->src_dt + AuCHILD, a->src_dentry, &a->h_path);
20644 +       if (a->dst_h_dentry->d_inode) {
20645 +               au_fset_ren(a->flags, DT_DSTDIR);
20646 +               a->h_path.dentry = a->dst_h_dentry;
20647 +               au_dtime_store(a->dst_dt + AuCHILD, a->dst_dentry, &a->h_path);
20648 +       }
20649 +}
20650 +
20651 +static void au_ren_rev_dt(int err, struct au_ren_args *a)
20652 +{
20653 +       struct dentry *h_d;
20654 +       struct mutex *h_mtx;
20655 +
20656 +       au_dtime_revert(a->src_dt + AuPARENT);
20657 +       if (!au_ftest_ren(a->flags, ISSAMEDIR))
20658 +               au_dtime_revert(a->dst_dt + AuPARENT);
20659 +
20660 +       if (au_ftest_ren(a->flags, ISDIR) && err != -EIO) {
20661 +               h_d = a->src_dt[AuCHILD].dt_h_path.dentry;
20662 +               h_mtx = &h_d->d_inode->i_mutex;
20663 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
20664 +               au_dtime_revert(a->src_dt + AuCHILD);
20665 +               mutex_unlock(h_mtx);
20666 +
20667 +               if (au_ftest_ren(a->flags, DT_DSTDIR)) {
20668 +                       h_d = a->dst_dt[AuCHILD].dt_h_path.dentry;
20669 +                       h_mtx = &h_d->d_inode->i_mutex;
20670 +                       mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
20671 +                       au_dtime_revert(a->dst_dt + AuCHILD);
20672 +                       mutex_unlock(h_mtx);
20673 +               }
20674 +       }
20675 +}
20676 +
20677 +/* ---------------------------------------------------------------------- */
20678 +
20679 +int aufs_rename(struct inode *_src_dir, struct dentry *_src_dentry,
20680 +               struct inode *_dst_dir, struct dentry *_dst_dentry)
20681 +{
20682 +       int err, flags;
20683 +       /* reduce stack space */
20684 +       struct au_ren_args *a;
20685 +
20686 +       AuDbg("%pd, %pd\n", _src_dentry, _dst_dentry);
20687 +       IMustLock(_src_dir);
20688 +       IMustLock(_dst_dir);
20689 +
20690 +       err = -ENOMEM;
20691 +       BUILD_BUG_ON(sizeof(*a) > PAGE_SIZE);
20692 +       a = kzalloc(sizeof(*a), GFP_NOFS);
20693 +       if (unlikely(!a))
20694 +               goto out;
20695 +
20696 +       a->src_dir = _src_dir;
20697 +       a->src_dentry = _src_dentry;
20698 +       a->src_inode = a->src_dentry->d_inode;
20699 +       a->src_parent = a->src_dentry->d_parent; /* dir inode is locked */
20700 +       a->dst_dir = _dst_dir;
20701 +       a->dst_dentry = _dst_dentry;
20702 +       a->dst_inode = a->dst_dentry->d_inode;
20703 +       a->dst_parent = a->dst_dentry->d_parent; /* dir inode is locked */
20704 +       if (a->dst_inode) {
20705 +               IMustLock(a->dst_inode);
20706 +               au_igrab(a->dst_inode);
20707 +       }
20708 +
20709 +       err = -ENOTDIR;
20710 +       flags = AuLock_FLUSH | AuLock_NOPLM | AuLock_GEN;
20711 +       if (S_ISDIR(a->src_inode->i_mode)) {
20712 +               au_fset_ren(a->flags, ISDIR);
20713 +               if (unlikely(a->dst_inode && !S_ISDIR(a->dst_inode->i_mode)))
20714 +                       goto out_free;
20715 +               err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry,
20716 +                                               AuLock_DIR | flags);
20717 +       } else
20718 +               err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry,
20719 +                                               flags);
20720 +       if (unlikely(err))
20721 +               goto out_free;
20722 +
20723 +       err = au_d_hashed_positive(a->src_dentry);
20724 +       if (unlikely(err))
20725 +               goto out_unlock;
20726 +       err = -ENOENT;
20727 +       if (a->dst_inode) {
20728 +               /*
20729 +                * If it is a dir, VFS unhash dst_dentry before this
20730 +                * function. It means we cannot rely upon d_unhashed().
20731 +                */
20732 +               if (unlikely(!a->dst_inode->i_nlink))
20733 +                       goto out_unlock;
20734 +               if (!S_ISDIR(a->dst_inode->i_mode)) {
20735 +                       err = au_d_hashed_positive(a->dst_dentry);
20736 +                       if (unlikely(err))
20737 +                               goto out_unlock;
20738 +               } else if (unlikely(IS_DEADDIR(a->dst_inode)))
20739 +                       goto out_unlock;
20740 +       } else if (unlikely(d_unhashed(a->dst_dentry)))
20741 +               goto out_unlock;
20742 +
20743 +       /*
20744 +        * is it possible?
20745 +        * yes, it happend (in linux-3.3-rcN) but I don't know why.
20746 +        * there may exist a problem somewhere else.
20747 +        */
20748 +       err = -EINVAL;
20749 +       if (unlikely(a->dst_parent->d_inode == a->src_dentry->d_inode))
20750 +               goto out_unlock;
20751 +
20752 +       au_fset_ren(a->flags, ISSAMEDIR); /* temporary */
20753 +       di_write_lock_parent(a->dst_parent);
20754 +
20755 +       /* which branch we process */
20756 +       err = au_ren_wbr(a);
20757 +       if (unlikely(err < 0))
20758 +               goto out_parent;
20759 +       a->br = au_sbr(a->dst_dentry->d_sb, a->btgt);
20760 +       a->h_path.mnt = au_br_mnt(a->br);
20761 +
20762 +       /* are they available to be renamed */
20763 +       err = au_ren_may_dir(a);
20764 +       if (unlikely(err))
20765 +               goto out_children;
20766 +
20767 +       /* prepare the writable parent dir on the same branch */
20768 +       if (a->dst_bstart == a->btgt) {
20769 +               au_fset_ren(a->flags, WHDST);
20770 +       } else {
20771 +               err = au_cpup_dirs(a->dst_dentry, a->btgt);
20772 +               if (unlikely(err))
20773 +                       goto out_children;
20774 +       }
20775 +
20776 +       if (a->src_dir != a->dst_dir) {
20777 +               /*
20778 +                * this temporary unlock is safe,
20779 +                * because both dir->i_mutex are locked.
20780 +                */
20781 +               di_write_unlock(a->dst_parent);
20782 +               di_write_lock_parent(a->src_parent);
20783 +               err = au_wr_dir_need_wh(a->src_dentry,
20784 +                                       au_ftest_ren(a->flags, ISDIR),
20785 +                                       &a->btgt);
20786 +               di_write_unlock(a->src_parent);
20787 +               di_write_lock2_parent(a->src_parent, a->dst_parent, /*isdir*/1);
20788 +               au_fclr_ren(a->flags, ISSAMEDIR);
20789 +       } else
20790 +               err = au_wr_dir_need_wh(a->src_dentry,
20791 +                                       au_ftest_ren(a->flags, ISDIR),
20792 +                                       &a->btgt);
20793 +       if (unlikely(err < 0))
20794 +               goto out_children;
20795 +       if (err)
20796 +               au_fset_ren(a->flags, WHSRC);
20797 +
20798 +       /* cpup src */
20799 +       if (a->src_bstart != a->btgt) {
20800 +               struct au_pin pin;
20801 +
20802 +               err = au_pin(&pin, a->src_dentry, a->btgt,
20803 +                            au_opt_udba(a->src_dentry->d_sb),
20804 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
20805 +               if (!err) {
20806 +                       struct au_cp_generic cpg = {
20807 +                               .dentry = a->src_dentry,
20808 +                               .bdst   = a->btgt,
20809 +                               .bsrc   = a->src_bstart,
20810 +                               .len    = -1,
20811 +                               .pin    = &pin,
20812 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
20813 +                       };
20814 +                       AuDebugOn(au_dbstart(a->src_dentry) != a->src_bstart);
20815 +                       err = au_sio_cpup_simple(&cpg);
20816 +                       au_unpin(&pin);
20817 +               }
20818 +               if (unlikely(err))
20819 +                       goto out_children;
20820 +               a->src_bstart = a->btgt;
20821 +               a->src_h_dentry = au_h_dptr(a->src_dentry, a->btgt);
20822 +               au_fset_ren(a->flags, WHSRC);
20823 +       }
20824 +
20825 +       /* lock them all */
20826 +       err = au_ren_lock(a);
20827 +       if (unlikely(err))
20828 +               /* leave the copied-up one */
20829 +               goto out_children;
20830 +
20831 +       if (!au_opt_test(au_mntflags(a->dst_dir->i_sb), UDBA_NONE))
20832 +               err = au_may_ren(a);
20833 +       else if (unlikely(a->dst_dentry->d_name.len > AUFS_MAX_NAMELEN))
20834 +               err = -ENAMETOOLONG;
20835 +       if (unlikely(err))
20836 +               goto out_hdir;
20837 +
20838 +       /* store timestamps to be revertible */
20839 +       au_ren_dt(a);
20840 +
20841 +       /* here we go */
20842 +       err = do_rename(a);
20843 +       if (unlikely(err))
20844 +               goto out_dt;
20845 +
20846 +       /* update dir attributes */
20847 +       au_ren_refresh_dir(a);
20848 +
20849 +       /* dput/iput all lower dentries */
20850 +       au_ren_refresh(a);
20851 +
20852 +       goto out_hdir; /* success */
20853 +
20854 +out_dt:
20855 +       au_ren_rev_dt(err, a);
20856 +out_hdir:
20857 +       au_ren_unlock(a);
20858 +out_children:
20859 +       au_nhash_wh_free(&a->whlist);
20860 +       if (err && a->dst_inode && a->dst_bstart != a->btgt) {
20861 +               AuDbg("bstart %d, btgt %d\n", a->dst_bstart, a->btgt);
20862 +               au_set_h_dptr(a->dst_dentry, a->btgt, NULL);
20863 +               au_set_dbstart(a->dst_dentry, a->dst_bstart);
20864 +       }
20865 +out_parent:
20866 +       if (!err)
20867 +               d_move(a->src_dentry, a->dst_dentry);
20868 +       else {
20869 +               au_update_dbstart(a->dst_dentry);
20870 +               if (!a->dst_inode)
20871 +                       d_drop(a->dst_dentry);
20872 +       }
20873 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
20874 +               di_write_unlock(a->dst_parent);
20875 +       else
20876 +               di_write_unlock2(a->src_parent, a->dst_parent);
20877 +out_unlock:
20878 +       aufs_read_and_write_unlock2(a->dst_dentry, a->src_dentry);
20879 +out_free:
20880 +       iput(a->dst_inode);
20881 +       if (a->thargs)
20882 +               au_whtmp_rmdir_free(a->thargs);
20883 +       kfree(a);
20884 +out:
20885 +       AuTraceErr(err);
20886 +       return err;
20887 +}
20888 diff -urN /usr/share/empty/fs/aufs/Kconfig linux/fs/aufs/Kconfig
20889 --- /usr/share/empty/fs/aufs/Kconfig    1970-01-01 01:00:00.000000000 +0100
20890 +++ linux/fs/aufs/Kconfig       2015-01-25 13:00:38.627713742 +0100
20891 @@ -0,0 +1,185 @@
20892 +config AUFS_FS
20893 +       tristate "Aufs (Advanced multi layered unification filesystem) support"
20894 +       help
20895 +       Aufs is a stackable unification filesystem such as Unionfs,
20896 +       which unifies several directories and provides a merged single
20897 +       directory.
20898 +       In the early days, aufs was entirely re-designed and
20899 +       re-implemented Unionfs Version 1.x series. Introducing many
20900 +       original ideas, approaches and improvements, it becomes totally
20901 +       different from Unionfs while keeping the basic features.
20902 +
20903 +if AUFS_FS
20904 +choice
20905 +       prompt "Maximum number of branches"
20906 +       default AUFS_BRANCH_MAX_127
20907 +       help
20908 +       Specifies the maximum number of branches (or member directories)
20909 +       in a single aufs. The larger value consumes more system
20910 +       resources and has a minor impact to performance.
20911 +config AUFS_BRANCH_MAX_127
20912 +       bool "127"
20913 +       help
20914 +       Specifies the maximum number of branches (or member directories)
20915 +       in a single aufs. The larger value consumes more system
20916 +       resources and has a minor impact to performance.
20917 +config AUFS_BRANCH_MAX_511
20918 +       bool "511"
20919 +       help
20920 +       Specifies the maximum number of branches (or member directories)
20921 +       in a single aufs. The larger value consumes more system
20922 +       resources and has a minor impact to performance.
20923 +config AUFS_BRANCH_MAX_1023
20924 +       bool "1023"
20925 +       help
20926 +       Specifies the maximum number of branches (or member directories)
20927 +       in a single aufs. The larger value consumes more system
20928 +       resources and has a minor impact to performance.
20929 +config AUFS_BRANCH_MAX_32767
20930 +       bool "32767"
20931 +       help
20932 +       Specifies the maximum number of branches (or member directories)
20933 +       in a single aufs. The larger value consumes more system
20934 +       resources and has a minor impact to performance.
20935 +endchoice
20936 +
20937 +config AUFS_SBILIST
20938 +       bool
20939 +       depends on AUFS_MAGIC_SYSRQ || PROC_FS
20940 +       default y
20941 +       help
20942 +       Automatic configuration for internal use.
20943 +       When aufs supports Magic SysRq or /proc, enabled automatically.
20944 +
20945 +config AUFS_HNOTIFY
20946 +       bool "Detect direct branch access (bypassing aufs)"
20947 +       help
20948 +       If you want to modify files on branches directly, eg. bypassing aufs,
20949 +       and want aufs to detect the changes of them fully, then enable this
20950 +       option and use 'udba=notify' mount option.
20951 +       Currently there is only one available configuration, "fsnotify".
20952 +       It will have a negative impact to the performance.
20953 +       See detail in aufs.5.
20954 +
20955 +choice
20956 +       prompt "method" if AUFS_HNOTIFY
20957 +       default AUFS_HFSNOTIFY
20958 +config AUFS_HFSNOTIFY
20959 +       bool "fsnotify"
20960 +       select FSNOTIFY
20961 +endchoice
20962 +
20963 +config AUFS_EXPORT
20964 +       bool "NFS-exportable aufs"
20965 +       depends on EXPORTFS
20966 +       help
20967 +       If you want to export your mounted aufs via NFS, then enable this
20968 +       option. There are several requirements for this configuration.
20969 +       See detail in aufs.5.
20970 +
20971 +config AUFS_INO_T_64
20972 +       bool
20973 +       depends on AUFS_EXPORT
20974 +       depends on 64BIT && !(ALPHA || S390)
20975 +       default y
20976 +       help
20977 +       Automatic configuration for internal use.
20978 +       /* typedef unsigned long/int __kernel_ino_t */
20979 +       /* alpha and s390x are int */
20980 +
20981 +config AUFS_XATTR
20982 +       bool "support for XATTR/EA (including Security Labels)"
20983 +       help
20984 +       If your branch fs supports XATTR/EA and you want to make them
20985 +       available in aufs too, then enable this opsion and specify the
20986 +       branch attributes for EA.
20987 +       See detail in aufs.5.
20988 +
20989 +config AUFS_FHSM
20990 +       bool "File-based Hierarchical Storage Management"
20991 +       help
20992 +       Hierarchical Storage Management (or HSM) is a well-known feature
20993 +       in the storage world. Aufs provides this feature as file-based.
20994 +       with multiple branches.
20995 +       These multiple branches are prioritized, ie. the topmost one
20996 +       should be the fastest drive and be used heavily.
20997 +
20998 +config AUFS_RDU
20999 +       bool "Readdir in userspace"
21000 +       help
21001 +       Aufs has two methods to provide a merged view for a directory,
21002 +       by a user-space library and by kernel-space natively. The latter
21003 +       is always enabled but sometimes large and slow.
21004 +       If you enable this option, install the library in aufs2-util
21005 +       package, and set some environment variables for your readdir(3),
21006 +       then the work will be handled in user-space which generally
21007 +       shows better performance in most cases.
21008 +       See detail in aufs.5.
21009 +
21010 +config AUFS_SHWH
21011 +       bool "Show whiteouts"
21012 +       help
21013 +       If you want to make the whiteouts in aufs visible, then enable
21014 +       this option and specify 'shwh' mount option. Although it may
21015 +       sounds like philosophy or something, but in technically it
21016 +       simply shows the name of whiteout with keeping its behaviour.
21017 +
21018 +config AUFS_BR_RAMFS
21019 +       bool "Ramfs (initramfs/rootfs) as an aufs branch"
21020 +       help
21021 +       If you want to use ramfs as an aufs branch fs, then enable this
21022 +       option. Generally tmpfs is recommended.
21023 +       Aufs prohibited them to be a branch fs by default, because
21024 +       initramfs becomes unusable after switch_root or something
21025 +       generally. If you sets initramfs as an aufs branch and boot your
21026 +       system by switch_root, you will meet a problem easily since the
21027 +       files in initramfs may be inaccessible.
21028 +       Unless you are going to use ramfs as an aufs branch fs without
21029 +       switch_root or something, leave it N.
21030 +
21031 +config AUFS_BR_FUSE
21032 +       bool "Fuse fs as an aufs branch"
21033 +       depends on FUSE_FS
21034 +       select AUFS_POLL
21035 +       help
21036 +       If you want to use fuse-based userspace filesystem as an aufs
21037 +       branch fs, then enable this option.
21038 +       It implements the internal poll(2) operation which is
21039 +       implemented by fuse only (curretnly).
21040 +
21041 +config AUFS_POLL
21042 +       bool
21043 +       help
21044 +       Automatic configuration for internal use.
21045 +
21046 +config AUFS_BR_HFSPLUS
21047 +       bool "Hfsplus as an aufs branch"
21048 +       depends on HFSPLUS_FS
21049 +       default y
21050 +       help
21051 +       If you want to use hfsplus fs as an aufs branch fs, then enable
21052 +       this option. This option introduces a small overhead at
21053 +       copying-up a file on hfsplus.
21054 +
21055 +config AUFS_BDEV_LOOP
21056 +       bool
21057 +       depends on BLK_DEV_LOOP
21058 +       default y
21059 +       help
21060 +       Automatic configuration for internal use.
21061 +       Convert =[ym] into =y.
21062 +
21063 +config AUFS_DEBUG
21064 +       bool "Debug aufs"
21065 +       help
21066 +       Enable this to compile aufs internal debug code.
21067 +       It will have a negative impact to the performance.
21068 +
21069 +config AUFS_MAGIC_SYSRQ
21070 +       bool
21071 +       depends on AUFS_DEBUG && MAGIC_SYSRQ
21072 +       default y
21073 +       help
21074 +       Automatic configuration for internal use.
21075 +       When aufs supports Magic SysRq, enabled automatically.
21076 +endif
21077 diff -urN /usr/share/empty/fs/aufs/loop.c linux/fs/aufs/loop.c
21078 --- /usr/share/empty/fs/aufs/loop.c     1970-01-01 01:00:00.000000000 +0100
21079 +++ linux/fs/aufs/loop.c        2015-01-25 13:00:38.631047076 +0100
21080 @@ -0,0 +1,145 @@
21081 +/*
21082 + * Copyright (C) 2005-2014 Junjiro R. Okajima
21083 + *
21084 + * This program, aufs is free software; you can redistribute it and/or modify
21085 + * it under the terms of the GNU General Public License as published by
21086 + * the Free Software Foundation; either version 2 of the License, or
21087 + * (at your option) any later version.
21088 + *
21089 + * This program is distributed in the hope that it will be useful,
21090 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21091 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21092 + * GNU General Public License for more details.
21093 + *
21094 + * You should have received a copy of the GNU General Public License
21095 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21096 + */
21097 +
21098 +/*
21099 + * support for loopback block device as a branch
21100 + */
21101 +
21102 +#include "aufs.h"
21103 +
21104 +/* added into drivers/block/loop.c */
21105 +static struct file *(*backing_file_func)(struct super_block *sb);
21106 +
21107 +/*
21108 + * test if two lower dentries have overlapping branches.
21109 + */
21110 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding)
21111 +{
21112 +       struct super_block *h_sb;
21113 +       struct file *backing_file;
21114 +
21115 +       if (unlikely(!backing_file_func)) {
21116 +               /* don't load "loop" module here */
21117 +               backing_file_func = symbol_get(loop_backing_file);
21118 +               if (unlikely(!backing_file_func))
21119 +                       /* "loop" module is not loaded */
21120 +                       return 0;
21121 +       }
21122 +
21123 +       h_sb = h_adding->d_sb;
21124 +       backing_file = backing_file_func(h_sb);
21125 +       if (!backing_file)
21126 +               return 0;
21127 +
21128 +       h_adding = backing_file->f_dentry;
21129 +       /*
21130 +        * h_adding can be local NFS.
21131 +        * in this case aufs cannot detect the loop.
21132 +        */
21133 +       if (unlikely(h_adding->d_sb == sb))
21134 +               return 1;
21135 +       return !!au_test_subdir(h_adding, sb->s_root);
21136 +}
21137 +
21138 +/* true if a kernel thread named 'loop[0-9].*' accesses a file */
21139 +int au_test_loopback_kthread(void)
21140 +{
21141 +       int ret;
21142 +       struct task_struct *tsk = current;
21143 +       char c, comm[sizeof(tsk->comm)];
21144 +
21145 +       ret = 0;
21146 +       if (tsk->flags & PF_KTHREAD) {
21147 +               get_task_comm(comm, tsk);
21148 +               c = comm[4];
21149 +               ret = ('0' <= c && c <= '9'
21150 +                      && !strncmp(comm, "loop", 4));
21151 +       }
21152 +
21153 +       return ret;
21154 +}
21155 +
21156 +/* ---------------------------------------------------------------------- */
21157 +
21158 +#define au_warn_loopback_step  16
21159 +static int au_warn_loopback_nelem = au_warn_loopback_step;
21160 +static unsigned long *au_warn_loopback_array;
21161 +
21162 +void au_warn_loopback(struct super_block *h_sb)
21163 +{
21164 +       int i, new_nelem;
21165 +       unsigned long *a, magic;
21166 +       static DEFINE_SPINLOCK(spin);
21167 +
21168 +       magic = h_sb->s_magic;
21169 +       spin_lock(&spin);
21170 +       a = au_warn_loopback_array;
21171 +       for (i = 0; i < au_warn_loopback_nelem && *a; i++)
21172 +               if (a[i] == magic) {
21173 +                       spin_unlock(&spin);
21174 +                       return;
21175 +               }
21176 +
21177 +       /* h_sb is new to us, print it */
21178 +       if (i < au_warn_loopback_nelem) {
21179 +               a[i] = magic;
21180 +               goto pr;
21181 +       }
21182 +
21183 +       /* expand the array */
21184 +       new_nelem = au_warn_loopback_nelem + au_warn_loopback_step;
21185 +       a = au_kzrealloc(au_warn_loopback_array,
21186 +                        au_warn_loopback_nelem * sizeof(unsigned long),
21187 +                        new_nelem * sizeof(unsigned long), GFP_ATOMIC);
21188 +       if (a) {
21189 +               au_warn_loopback_nelem = new_nelem;
21190 +               au_warn_loopback_array = a;
21191 +               a[i] = magic;
21192 +               goto pr;
21193 +       }
21194 +
21195 +       spin_unlock(&spin);
21196 +       AuWarn1("realloc failed, ignored\n");
21197 +       return;
21198 +
21199 +pr:
21200 +       spin_unlock(&spin);
21201 +       pr_warn("you may want to try another patch for loopback file "
21202 +               "on %s(0x%lx) branch\n", au_sbtype(h_sb), magic);
21203 +}
21204 +
21205 +int au_loopback_init(void)
21206 +{
21207 +       int err;
21208 +       struct super_block *sb __maybe_unused;
21209 +
21210 +       AuDebugOn(sizeof(sb->s_magic) != sizeof(unsigned long));
21211 +
21212 +       err = 0;
21213 +       au_warn_loopback_array = kcalloc(au_warn_loopback_step,
21214 +                                        sizeof(unsigned long), GFP_NOFS);
21215 +       if (unlikely(!au_warn_loopback_array))
21216 +               err = -ENOMEM;
21217 +
21218 +       return err;
21219 +}
21220 +
21221 +void au_loopback_fin(void)
21222 +{
21223 +       symbol_put(loop_backing_file);
21224 +       kfree(au_warn_loopback_array);
21225 +}
21226 diff -urN /usr/share/empty/fs/aufs/loop.h linux/fs/aufs/loop.h
21227 --- /usr/share/empty/fs/aufs/loop.h     1970-01-01 01:00:00.000000000 +0100
21228 +++ linux/fs/aufs/loop.h        2015-01-25 13:00:38.631047076 +0100
21229 @@ -0,0 +1,52 @@
21230 +/*
21231 + * Copyright (C) 2005-2014 Junjiro R. Okajima
21232 + *
21233 + * This program, aufs is free software; you can redistribute it and/or modify
21234 + * it under the terms of the GNU General Public License as published by
21235 + * the Free Software Foundation; either version 2 of the License, or
21236 + * (at your option) any later version.
21237 + *
21238 + * This program is distributed in the hope that it will be useful,
21239 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21240 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21241 + * GNU General Public License for more details.
21242 + *
21243 + * You should have received a copy of the GNU General Public License
21244 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21245 + */
21246 +
21247 +/*
21248 + * support for loopback mount as a branch
21249 + */
21250 +
21251 +#ifndef __AUFS_LOOP_H__
21252 +#define __AUFS_LOOP_H__
21253 +
21254 +#ifdef __KERNEL__
21255 +
21256 +struct dentry;
21257 +struct super_block;
21258 +
21259 +#ifdef CONFIG_AUFS_BDEV_LOOP
21260 +/* drivers/block/loop.c */
21261 +struct file *loop_backing_file(struct super_block *sb);
21262 +
21263 +/* loop.c */
21264 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding);
21265 +int au_test_loopback_kthread(void);
21266 +void au_warn_loopback(struct super_block *h_sb);
21267 +
21268 +int au_loopback_init(void);
21269 +void au_loopback_fin(void);
21270 +#else
21271 +AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
21272 +          struct dentry *h_adding)
21273 +AuStubInt0(au_test_loopback_kthread, void)
21274 +AuStubVoid(au_warn_loopback, struct super_block *h_sb)
21275 +
21276 +AuStubInt0(au_loopback_init, void)
21277 +AuStubVoid(au_loopback_fin, void)
21278 +#endif /* BLK_DEV_LOOP */
21279 +
21280 +#endif /* __KERNEL__ */
21281 +#endif /* __AUFS_LOOP_H__ */
21282 diff -urN /usr/share/empty/fs/aufs/magic.mk linux/fs/aufs/magic.mk
21283 --- /usr/share/empty/fs/aufs/magic.mk   1970-01-01 01:00:00.000000000 +0100
21284 +++ linux/fs/aufs/magic.mk      2015-01-25 13:00:38.631047076 +0100
21285 @@ -0,0 +1,54 @@
21286 +
21287 +# defined in ${srctree}/fs/fuse/inode.c
21288 +# tristate
21289 +ifdef CONFIG_FUSE_FS
21290 +ccflags-y += -DFUSE_SUPER_MAGIC=0x65735546
21291 +endif
21292 +
21293 +# defined in ${srctree}/fs/ocfs2/ocfs2_fs.h
21294 +# tristate
21295 +ifdef CONFIG_OCFS2_FS
21296 +ccflags-y += -DOCFS2_SUPER_MAGIC=0x7461636f
21297 +endif
21298 +
21299 +# defined in ${srctree}/fs/ocfs2/dlm/userdlm.h
21300 +# tristate
21301 +ifdef CONFIG_OCFS2_FS_O2CB
21302 +ccflags-y += -DDLMFS_MAGIC=0x76a9f425
21303 +endif
21304 +
21305 +# defined in ${srctree}/fs/cifs/cifsfs.c
21306 +# tristate
21307 +ifdef CONFIG_CIFS_FS
21308 +ccflags-y += -DCIFS_MAGIC_NUMBER=0xFF534D42
21309 +endif
21310 +
21311 +# defined in ${srctree}/fs/xfs/xfs_sb.h
21312 +# tristate
21313 +ifdef CONFIG_XFS_FS
21314 +ccflags-y += -DXFS_SB_MAGIC=0x58465342
21315 +endif
21316 +
21317 +# defined in ${srctree}/fs/configfs/mount.c
21318 +# tristate
21319 +ifdef CONFIG_CONFIGFS_FS
21320 +ccflags-y += -DCONFIGFS_MAGIC=0x62656570
21321 +endif
21322 +
21323 +# defined in ${srctree}/fs/9p/v9fs.h
21324 +# tristate
21325 +ifdef CONFIG_9P_FS
21326 +ccflags-y += -DV9FS_MAGIC=0x01021997
21327 +endif
21328 +
21329 +# defined in ${srctree}/fs/ubifs/ubifs.h
21330 +# tristate
21331 +ifdef CONFIG_UBIFS_FS
21332 +ccflags-y += -DUBIFS_SUPER_MAGIC=0x24051905
21333 +endif
21334 +
21335 +# defined in ${srctree}/fs/hfsplus/hfsplus_raw.h
21336 +# tristate
21337 +ifdef CONFIG_HFSPLUS_FS
21338 +ccflags-y += -DHFSPLUS_SUPER_MAGIC=0x482b
21339 +endif
21340 diff -urN /usr/share/empty/fs/aufs/Makefile linux/fs/aufs/Makefile
21341 --- /usr/share/empty/fs/aufs/Makefile   1970-01-01 01:00:00.000000000 +0100
21342 +++ linux/fs/aufs/Makefile      2015-01-25 13:00:38.627713742 +0100
21343 @@ -0,0 +1,44 @@
21344 +
21345 +include ${src}/magic.mk
21346 +ifeq (${CONFIG_AUFS_FS},m)
21347 +include ${src}/conf.mk
21348 +endif
21349 +-include ${src}/priv_def.mk
21350 +
21351 +# cf. include/linux/kernel.h
21352 +# enable pr_debug
21353 +ccflags-y += -DDEBUG
21354 +# sparse requires the full pathname
21355 +ifdef M
21356 +ccflags-y += -include ${M}/../../include/uapi/linux/aufs_type.h
21357 +else
21358 +ccflags-y += -include ${srctree}/include/uapi/linux/aufs_type.h
21359 +endif
21360 +
21361 +obj-$(CONFIG_AUFS_FS) += aufs.o
21362 +aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
21363 +       wkq.o vfsub.o dcsub.o \
21364 +       cpup.o whout.o wbr_policy.o \
21365 +       dinfo.o dentry.o \
21366 +       dynop.o \
21367 +       finfo.o file.o f_op.o \
21368 +       dir.o vdir.o \
21369 +       iinfo.o inode.o i_op.o i_op_add.o i_op_del.o i_op_ren.o \
21370 +       mvdown.o ioctl.o
21371 +
21372 +# all are boolean
21373 +aufs-$(CONFIG_PROC_FS) += procfs.o plink.o
21374 +aufs-$(CONFIG_SYSFS) += sysfs.o
21375 +aufs-$(CONFIG_DEBUG_FS) += dbgaufs.o
21376 +aufs-$(CONFIG_AUFS_BDEV_LOOP) += loop.o
21377 +aufs-$(CONFIG_AUFS_HNOTIFY) += hnotify.o
21378 +aufs-$(CONFIG_AUFS_HFSNOTIFY) += hfsnotify.o
21379 +aufs-$(CONFIG_AUFS_EXPORT) += export.o
21380 +aufs-$(CONFIG_AUFS_XATTR) += xattr.o
21381 +aufs-$(CONFIG_FS_POSIX_ACL) += posix_acl.o
21382 +aufs-$(CONFIG_AUFS_FHSM) += fhsm.o
21383 +aufs-$(CONFIG_AUFS_POLL) += poll.o
21384 +aufs-$(CONFIG_AUFS_RDU) += rdu.o
21385 +aufs-$(CONFIG_AUFS_BR_HFSPLUS) += hfsplus.o
21386 +aufs-$(CONFIG_AUFS_DEBUG) += debug.o
21387 +aufs-$(CONFIG_AUFS_MAGIC_SYSRQ) += sysrq.o
21388 diff -urN /usr/share/empty/fs/aufs/module.c linux/fs/aufs/module.c
21389 --- /usr/share/empty/fs/aufs/module.c   1970-01-01 01:00:00.000000000 +0100
21390 +++ linux/fs/aufs/module.c      2015-01-25 13:00:38.634380408 +0100
21391 @@ -0,0 +1,210 @@
21392 +/*
21393 + * Copyright (C) 2005-2014 Junjiro R. Okajima
21394 + *
21395 + * This program, aufs is free software; you can redistribute it and/or modify
21396 + * it under the terms of the GNU General Public License as published by
21397 + * the Free Software Foundation; either version 2 of the License, or
21398 + * (at your option) any later version.
21399 + *
21400 + * This program is distributed in the hope that it will be useful,
21401 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21402 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21403 + * GNU General Public License for more details.
21404 + *
21405 + * You should have received a copy of the GNU General Public License
21406 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21407 + */
21408 +
21409 +/*
21410 + * module global variables and operations
21411 + */
21412 +
21413 +#include <linux/module.h>
21414 +#include <linux/seq_file.h>
21415 +#include "aufs.h"
21416 +
21417 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp)
21418 +{
21419 +       if (new_sz <= nused)
21420 +               return p;
21421 +
21422 +       p = krealloc(p, new_sz, gfp);
21423 +       if (p)
21424 +               memset(p + nused, 0, new_sz - nused);
21425 +       return p;
21426 +}
21427 +
21428 +/* ---------------------------------------------------------------------- */
21429 +
21430 +/*
21431 + * aufs caches
21432 + */
21433 +struct kmem_cache *au_cachep[AuCache_Last];
21434 +static int __init au_cache_init(void)
21435 +{
21436 +       au_cachep[AuCache_DINFO] = AuCacheCtor(au_dinfo, au_di_init_once);
21437 +       if (au_cachep[AuCache_DINFO])
21438 +               /* SLAB_DESTROY_BY_RCU */
21439 +               au_cachep[AuCache_ICNTNR] = AuCacheCtor(au_icntnr,
21440 +                                                       au_icntnr_init_once);
21441 +       if (au_cachep[AuCache_ICNTNR])
21442 +               au_cachep[AuCache_FINFO] = AuCacheCtor(au_finfo,
21443 +                                                      au_fi_init_once);
21444 +       if (au_cachep[AuCache_FINFO])
21445 +               au_cachep[AuCache_VDIR] = AuCache(au_vdir);
21446 +       if (au_cachep[AuCache_VDIR])
21447 +               au_cachep[AuCache_DEHSTR] = AuCache(au_vdir_dehstr);
21448 +       if (au_cachep[AuCache_DEHSTR])
21449 +               return 0;
21450 +
21451 +       return -ENOMEM;
21452 +}
21453 +
21454 +static void au_cache_fin(void)
21455 +{
21456 +       int i;
21457 +
21458 +       /*
21459 +        * Make sure all delayed rcu free inodes are flushed before we
21460 +        * destroy cache.
21461 +        */
21462 +       rcu_barrier();
21463 +
21464 +       /* excluding AuCache_HNOTIFY */
21465 +       BUILD_BUG_ON(AuCache_HNOTIFY + 1 != AuCache_Last);
21466 +       for (i = 0; i < AuCache_HNOTIFY; i++)
21467 +               if (au_cachep[i]) {
21468 +                       kmem_cache_destroy(au_cachep[i]);
21469 +                       au_cachep[i] = NULL;
21470 +               }
21471 +}
21472 +
21473 +/* ---------------------------------------------------------------------- */
21474 +
21475 +int au_dir_roflags;
21476 +
21477 +#ifdef CONFIG_AUFS_SBILIST
21478 +/*
21479 + * iterate_supers_type() doesn't protect us from
21480 + * remounting (branch management)
21481 + */
21482 +struct au_splhead au_sbilist;
21483 +#endif
21484 +
21485 +struct lock_class_key au_lc_key[AuLcKey_Last];
21486 +
21487 +/*
21488 + * functions for module interface.
21489 + */
21490 +MODULE_LICENSE("GPL");
21491 +/* MODULE_LICENSE("GPL v2"); */
21492 +MODULE_AUTHOR("Junjiro R. Okajima <aufs-users@lists.sourceforge.net>");
21493 +MODULE_DESCRIPTION(AUFS_NAME
21494 +       " -- Advanced multi layered unification filesystem");
21495 +MODULE_VERSION(AUFS_VERSION);
21496 +MODULE_ALIAS_FS(AUFS_NAME);
21497 +
21498 +/* this module parameter has no meaning when SYSFS is disabled */
21499 +int sysaufs_brs = 1;
21500 +MODULE_PARM_DESC(brs, "use <sysfs>/fs/aufs/si_*/brN");
21501 +module_param_named(brs, sysaufs_brs, int, S_IRUGO);
21502 +
21503 +/* this module parameter has no meaning when USER_NS is disabled */
21504 +static bool au_userns;
21505 +MODULE_PARM_DESC(allow_userns, "allow unprivileged to mount under userns");
21506 +module_param_named(allow_userns, au_userns, bool, S_IRUGO);
21507 +
21508 +/* ---------------------------------------------------------------------- */
21509 +
21510 +static char au_esc_chars[0x20 + 3]; /* 0x01-0x20, backslash, del, and NULL */
21511 +
21512 +int au_seq_path(struct seq_file *seq, struct path *path)
21513 +{
21514 +       return seq_path(seq, path, au_esc_chars);
21515 +}
21516 +
21517 +/* ---------------------------------------------------------------------- */
21518 +
21519 +static int __init aufs_init(void)
21520 +{
21521 +       int err, i;
21522 +       char *p;
21523 +
21524 +       p = au_esc_chars;
21525 +       for (i = 1; i <= ' '; i++)
21526 +               *p++ = i;
21527 +       *p++ = '\\';
21528 +       *p++ = '\x7f';
21529 +       *p = 0;
21530 +
21531 +       au_dir_roflags = au_file_roflags(O_DIRECTORY | O_LARGEFILE);
21532 +
21533 +       au_sbilist_init();
21534 +       sysaufs_brs_init();
21535 +       au_debug_init();
21536 +       au_dy_init();
21537 +       err = sysaufs_init();
21538 +       if (unlikely(err))
21539 +               goto out;
21540 +       err = au_procfs_init();
21541 +       if (unlikely(err))
21542 +               goto out_sysaufs;
21543 +       err = au_wkq_init();
21544 +       if (unlikely(err))
21545 +               goto out_procfs;
21546 +       err = au_loopback_init();
21547 +       if (unlikely(err))
21548 +               goto out_wkq;
21549 +       err = au_hnotify_init();
21550 +       if (unlikely(err))
21551 +               goto out_loopback;
21552 +       err = au_sysrq_init();
21553 +       if (unlikely(err))
21554 +               goto out_hin;
21555 +       err = au_cache_init();
21556 +       if (unlikely(err))
21557 +               goto out_sysrq;
21558 +
21559 +       aufs_fs_type.fs_flags |= au_userns ? FS_USERNS_MOUNT : 0;
21560 +       err = register_filesystem(&aufs_fs_type);
21561 +       if (unlikely(err))
21562 +               goto out_cache;
21563 +
21564 +       /* since we define pr_fmt, call printk directly */
21565 +       printk(KERN_INFO AUFS_NAME " " AUFS_VERSION "\n");
21566 +       goto out; /* success */
21567 +
21568 +out_cache:
21569 +       au_cache_fin();
21570 +out_sysrq:
21571 +       au_sysrq_fin();
21572 +out_hin:
21573 +       au_hnotify_fin();
21574 +out_loopback:
21575 +       au_loopback_fin();
21576 +out_wkq:
21577 +       au_wkq_fin();
21578 +out_procfs:
21579 +       au_procfs_fin();
21580 +out_sysaufs:
21581 +       sysaufs_fin();
21582 +       au_dy_fin();
21583 +out:
21584 +       return err;
21585 +}
21586 +
21587 +static void __exit aufs_exit(void)
21588 +{
21589 +       unregister_filesystem(&aufs_fs_type);
21590 +       au_cache_fin();
21591 +       au_sysrq_fin();
21592 +       au_hnotify_fin();
21593 +       au_loopback_fin();
21594 +       au_wkq_fin();
21595 +       au_procfs_fin();
21596 +       sysaufs_fin();
21597 +       au_dy_fin();
21598 +}
21599 +
21600 +module_init(aufs_init);
21601 +module_exit(aufs_exit);
21602 diff -urN /usr/share/empty/fs/aufs/module.h linux/fs/aufs/module.h
21603 --- /usr/share/empty/fs/aufs/module.h   1970-01-01 01:00:00.000000000 +0100
21604 +++ linux/fs/aufs/module.h      2015-01-25 13:00:38.634380408 +0100
21605 @@ -0,0 +1,104 @@
21606 +/*
21607 + * Copyright (C) 2005-2014 Junjiro R. Okajima
21608 + *
21609 + * This program, aufs is free software; you can redistribute it and/or modify
21610 + * it under the terms of the GNU General Public License as published by
21611 + * the Free Software Foundation; either version 2 of the License, or
21612 + * (at your option) any later version.
21613 + *
21614 + * This program is distributed in the hope that it will be useful,
21615 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21616 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21617 + * GNU General Public License for more details.
21618 + *
21619 + * You should have received a copy of the GNU General Public License
21620 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21621 + */
21622 +
21623 +/*
21624 + * module initialization and module-global
21625 + */
21626 +
21627 +#ifndef __AUFS_MODULE_H__
21628 +#define __AUFS_MODULE_H__
21629 +
21630 +#ifdef __KERNEL__
21631 +
21632 +#include <linux/slab.h>
21633 +
21634 +struct path;
21635 +struct seq_file;
21636 +
21637 +/* module parameters */
21638 +extern int sysaufs_brs;
21639 +
21640 +/* ---------------------------------------------------------------------- */
21641 +
21642 +extern int au_dir_roflags;
21643 +
21644 +enum {
21645 +       AuLcNonDir_FIINFO,
21646 +       AuLcNonDir_DIINFO,
21647 +       AuLcNonDir_IIINFO,
21648 +
21649 +       AuLcDir_FIINFO,
21650 +       AuLcDir_DIINFO,
21651 +       AuLcDir_IIINFO,
21652 +
21653 +       AuLcSymlink_DIINFO,
21654 +       AuLcSymlink_IIINFO,
21655 +
21656 +       AuLcKey_Last
21657 +};
21658 +extern struct lock_class_key au_lc_key[AuLcKey_Last];
21659 +
21660 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp);
21661 +int au_seq_path(struct seq_file *seq, struct path *path);
21662 +
21663 +#ifdef CONFIG_PROC_FS
21664 +/* procfs.c */
21665 +int __init au_procfs_init(void);
21666 +void au_procfs_fin(void);
21667 +#else
21668 +AuStubInt0(au_procfs_init, void);
21669 +AuStubVoid(au_procfs_fin, void);
21670 +#endif
21671 +
21672 +/* ---------------------------------------------------------------------- */
21673 +
21674 +/* kmem cache */
21675 +enum {
21676 +       AuCache_DINFO,
21677 +       AuCache_ICNTNR,
21678 +       AuCache_FINFO,
21679 +       AuCache_VDIR,
21680 +       AuCache_DEHSTR,
21681 +       AuCache_HNOTIFY, /* must be last */
21682 +       AuCache_Last
21683 +};
21684 +
21685 +#define AuCacheFlags           (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD)
21686 +#define AuCache(type)          KMEM_CACHE(type, AuCacheFlags)
21687 +#define AuCacheCtor(type, ctor)        \
21688 +       kmem_cache_create(#type, sizeof(struct type), \
21689 +                         __alignof__(struct type), AuCacheFlags, ctor)
21690 +
21691 +extern struct kmem_cache *au_cachep[];
21692 +
21693 +#define AuCacheFuncs(name, index) \
21694 +static inline struct au_##name *au_cache_alloc_##name(void) \
21695 +{ return kmem_cache_alloc(au_cachep[AuCache_##index], GFP_NOFS); } \
21696 +static inline void au_cache_free_##name(struct au_##name *p) \
21697 +{ kmem_cache_free(au_cachep[AuCache_##index], p); }
21698 +
21699 +AuCacheFuncs(dinfo, DINFO);
21700 +AuCacheFuncs(icntnr, ICNTNR);
21701 +AuCacheFuncs(finfo, FINFO);
21702 +AuCacheFuncs(vdir, VDIR);
21703 +AuCacheFuncs(vdir_dehstr, DEHSTR);
21704 +#ifdef CONFIG_AUFS_HNOTIFY
21705 +AuCacheFuncs(hnotify, HNOTIFY);
21706 +#endif
21707 +
21708 +#endif /* __KERNEL__ */
21709 +#endif /* __AUFS_MODULE_H__ */
21710 diff -urN /usr/share/empty/fs/aufs/mvdown.c linux/fs/aufs/mvdown.c
21711 --- /usr/share/empty/fs/aufs/mvdown.c   1970-01-01 01:00:00.000000000 +0100
21712 +++ linux/fs/aufs/mvdown.c      2015-01-25 13:00:38.634380408 +0100
21713 @@ -0,0 +1,694 @@
21714 +/*
21715 + * Copyright (C) 2011-2014 Junjiro R. Okajima
21716 + *
21717 + * This program, aufs is free software; you can redistribute it and/or modify
21718 + * it under the terms of the GNU General Public License as published by
21719 + * the Free Software Foundation; either version 2 of the License, or
21720 + * (at your option) any later version.
21721 + *
21722 + * This program is distributed in the hope that it will be useful,
21723 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21724 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21725 + * GNU General Public License for more details.
21726 + *
21727 + * You should have received a copy of the GNU General Public License
21728 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21729 + */
21730 +
21731 +/*
21732 + * move-down, opposite of copy-up
21733 + */
21734 +
21735 +#include "aufs.h"
21736 +
21737 +struct au_mvd_args {
21738 +       struct {
21739 +               struct super_block *h_sb;
21740 +               struct dentry *h_parent;
21741 +               struct au_hinode *hdir;
21742 +               struct inode *h_dir, *h_inode;
21743 +               struct au_pin pin;
21744 +       } info[AUFS_MVDOWN_NARRAY];
21745 +
21746 +       struct aufs_mvdown mvdown;
21747 +       struct dentry *dentry, *parent;
21748 +       struct inode *inode, *dir;
21749 +       struct super_block *sb;
21750 +       aufs_bindex_t bopq, bwh, bfound;
21751 +       unsigned char rename_lock;
21752 +};
21753 +
21754 +#define mvd_errno              mvdown.au_errno
21755 +#define mvd_bsrc               mvdown.stbr[AUFS_MVDOWN_UPPER].bindex
21756 +#define mvd_src_brid           mvdown.stbr[AUFS_MVDOWN_UPPER].brid
21757 +#define mvd_bdst               mvdown.stbr[AUFS_MVDOWN_LOWER].bindex
21758 +#define mvd_dst_brid           mvdown.stbr[AUFS_MVDOWN_LOWER].brid
21759 +
21760 +#define mvd_h_src_sb           info[AUFS_MVDOWN_UPPER].h_sb
21761 +#define mvd_h_src_parent       info[AUFS_MVDOWN_UPPER].h_parent
21762 +#define mvd_hdir_src           info[AUFS_MVDOWN_UPPER].hdir
21763 +#define mvd_h_src_dir          info[AUFS_MVDOWN_UPPER].h_dir
21764 +#define mvd_h_src_inode                info[AUFS_MVDOWN_UPPER].h_inode
21765 +#define mvd_pin_src            info[AUFS_MVDOWN_UPPER].pin
21766 +
21767 +#define mvd_h_dst_sb           info[AUFS_MVDOWN_LOWER].h_sb
21768 +#define mvd_h_dst_parent       info[AUFS_MVDOWN_LOWER].h_parent
21769 +#define mvd_hdir_dst           info[AUFS_MVDOWN_LOWER].hdir
21770 +#define mvd_h_dst_dir          info[AUFS_MVDOWN_LOWER].h_dir
21771 +#define mvd_h_dst_inode                info[AUFS_MVDOWN_LOWER].h_inode
21772 +#define mvd_pin_dst            info[AUFS_MVDOWN_LOWER].pin
21773 +
21774 +#define AU_MVD_PR(flag, ...) do {                      \
21775 +               if (flag)                               \
21776 +                       pr_err(__VA_ARGS__);            \
21777 +       } while (0)
21778 +
21779 +static int find_lower_writable(struct au_mvd_args *a)
21780 +{
21781 +       struct super_block *sb;
21782 +       aufs_bindex_t bindex, bend;
21783 +       struct au_branch *br;
21784 +
21785 +       sb = a->sb;
21786 +       bindex = a->mvd_bsrc;
21787 +       bend = au_sbend(sb);
21788 +       if (a->mvdown.flags & AUFS_MVDOWN_FHSM_LOWER)
21789 +               for (bindex++; bindex <= bend; bindex++) {
21790 +                       br = au_sbr(sb, bindex);
21791 +                       if (au_br_fhsm(br->br_perm)
21792 +                           && (!(au_br_sb(br)->s_flags & MS_RDONLY)))
21793 +                               return bindex;
21794 +               }
21795 +       else if (!(a->mvdown.flags & AUFS_MVDOWN_ROLOWER))
21796 +               for (bindex++; bindex <= bend; bindex++) {
21797 +                       br = au_sbr(sb, bindex);
21798 +                       if (!au_br_rdonly(br))
21799 +                               return bindex;
21800 +               }
21801 +       else
21802 +               for (bindex++; bindex <= bend; bindex++) {
21803 +                       br = au_sbr(sb, bindex);
21804 +                       if (!(au_br_sb(br)->s_flags & MS_RDONLY)) {
21805 +                               if (au_br_rdonly(br))
21806 +                                       a->mvdown.flags
21807 +                                               |= AUFS_MVDOWN_ROLOWER_R;
21808 +                               return bindex;
21809 +                       }
21810 +               }
21811 +
21812 +       return -1;
21813 +}
21814 +
21815 +/* make the parent dir on bdst */
21816 +static int au_do_mkdir(const unsigned char dmsg, struct au_mvd_args *a)
21817 +{
21818 +       int err;
21819 +
21820 +       err = 0;
21821 +       a->mvd_hdir_src = au_hi(a->dir, a->mvd_bsrc);
21822 +       a->mvd_hdir_dst = au_hi(a->dir, a->mvd_bdst);
21823 +       a->mvd_h_src_parent = au_h_dptr(a->parent, a->mvd_bsrc);
21824 +       a->mvd_h_dst_parent = NULL;
21825 +       if (au_dbend(a->parent) >= a->mvd_bdst)
21826 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
21827 +       if (!a->mvd_h_dst_parent) {
21828 +               err = au_cpdown_dirs(a->dentry, a->mvd_bdst);
21829 +               if (unlikely(err)) {
21830 +                       AU_MVD_PR(dmsg, "cpdown_dirs failed\n");
21831 +                       goto out;
21832 +               }
21833 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
21834 +       }
21835 +
21836 +out:
21837 +       AuTraceErr(err);
21838 +       return err;
21839 +}
21840 +
21841 +/* lock them all */
21842 +static int au_do_lock(const unsigned char dmsg, struct au_mvd_args *a)
21843 +{
21844 +       int err;
21845 +       struct dentry *h_trap;
21846 +
21847 +       a->mvd_h_src_sb = au_sbr_sb(a->sb, a->mvd_bsrc);
21848 +       a->mvd_h_dst_sb = au_sbr_sb(a->sb, a->mvd_bdst);
21849 +       err = au_pin(&a->mvd_pin_dst, a->dentry, a->mvd_bdst,
21850 +                    au_opt_udba(a->sb),
21851 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
21852 +       AuTraceErr(err);
21853 +       if (unlikely(err)) {
21854 +               AU_MVD_PR(dmsg, "pin_dst failed\n");
21855 +               goto out;
21856 +       }
21857 +
21858 +       if (a->mvd_h_src_sb != a->mvd_h_dst_sb) {
21859 +               a->rename_lock = 0;
21860 +               au_pin_init(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
21861 +                           AuLsc_DI_PARENT, AuLsc_I_PARENT3,
21862 +                           au_opt_udba(a->sb),
21863 +                           AuPin_MNT_WRITE | AuPin_DI_LOCKED);
21864 +               err = au_do_pin(&a->mvd_pin_src);
21865 +               AuTraceErr(err);
21866 +               a->mvd_h_src_dir = a->mvd_h_src_parent->d_inode;
21867 +               if (unlikely(err)) {
21868 +                       AU_MVD_PR(dmsg, "pin_src failed\n");
21869 +                       goto out_dst;
21870 +               }
21871 +               goto out; /* success */
21872 +       }
21873 +
21874 +       a->rename_lock = 1;
21875 +       au_pin_hdir_unlock(&a->mvd_pin_dst);
21876 +       err = au_pin(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
21877 +                    au_opt_udba(a->sb),
21878 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
21879 +       AuTraceErr(err);
21880 +       a->mvd_h_src_dir = a->mvd_h_src_parent->d_inode;
21881 +       if (unlikely(err)) {
21882 +               AU_MVD_PR(dmsg, "pin_src failed\n");
21883 +               au_pin_hdir_lock(&a->mvd_pin_dst);
21884 +               goto out_dst;
21885 +       }
21886 +       au_pin_hdir_unlock(&a->mvd_pin_src);
21887 +       h_trap = vfsub_lock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
21888 +                                  a->mvd_h_dst_parent, a->mvd_hdir_dst);
21889 +       if (h_trap) {
21890 +               err = (h_trap != a->mvd_h_src_parent);
21891 +               if (err)
21892 +                       err = (h_trap != a->mvd_h_dst_parent);
21893 +       }
21894 +       BUG_ON(err); /* it should never happen */
21895 +       if (unlikely(a->mvd_h_src_dir != au_pinned_h_dir(&a->mvd_pin_src))) {
21896 +               err = -EBUSY;
21897 +               AuTraceErr(err);
21898 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
21899 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
21900 +               au_pin_hdir_lock(&a->mvd_pin_src);
21901 +               au_unpin(&a->mvd_pin_src);
21902 +               au_pin_hdir_lock(&a->mvd_pin_dst);
21903 +               goto out_dst;
21904 +       }
21905 +       goto out; /* success */
21906 +
21907 +out_dst:
21908 +       au_unpin(&a->mvd_pin_dst);
21909 +out:
21910 +       AuTraceErr(err);
21911 +       return err;
21912 +}
21913 +
21914 +static void au_do_unlock(const unsigned char dmsg, struct au_mvd_args *a)
21915 +{
21916 +       if (!a->rename_lock)
21917 +               au_unpin(&a->mvd_pin_src);
21918 +       else {
21919 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
21920 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
21921 +               au_pin_hdir_lock(&a->mvd_pin_src);
21922 +               au_unpin(&a->mvd_pin_src);
21923 +               au_pin_hdir_lock(&a->mvd_pin_dst);
21924 +       }
21925 +       au_unpin(&a->mvd_pin_dst);
21926 +}
21927 +
21928 +/* copy-down the file */
21929 +static int au_do_cpdown(const unsigned char dmsg, struct au_mvd_args *a)
21930 +{
21931 +       int err;
21932 +       struct au_cp_generic cpg = {
21933 +               .dentry = a->dentry,
21934 +               .bdst   = a->mvd_bdst,
21935 +               .bsrc   = a->mvd_bsrc,
21936 +               .len    = -1,
21937 +               .pin    = &a->mvd_pin_dst,
21938 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
21939 +       };
21940 +
21941 +       AuDbg("b%d, b%d\n", cpg.bsrc, cpg.bdst);
21942 +       if (a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
21943 +               au_fset_cpup(cpg.flags, OVERWRITE);
21944 +       if (a->mvdown.flags & AUFS_MVDOWN_ROLOWER)
21945 +               au_fset_cpup(cpg.flags, RWDST);
21946 +       err = au_sio_cpdown_simple(&cpg);
21947 +       if (unlikely(err))
21948 +               AU_MVD_PR(dmsg, "cpdown failed\n");
21949 +
21950 +       AuTraceErr(err);
21951 +       return err;
21952 +}
21953 +
21954 +/*
21955 + * unlink the whiteout on bdst if exist which may be created by UDBA while we
21956 + * were sleeping
21957 + */
21958 +static int au_do_unlink_wh(const unsigned char dmsg, struct au_mvd_args *a)
21959 +{
21960 +       int err;
21961 +       struct path h_path;
21962 +       struct au_branch *br;
21963 +       struct inode *delegated;
21964 +
21965 +       br = au_sbr(a->sb, a->mvd_bdst);
21966 +       h_path.dentry = au_wh_lkup(a->mvd_h_dst_parent, &a->dentry->d_name, br);
21967 +       err = PTR_ERR(h_path.dentry);
21968 +       if (IS_ERR(h_path.dentry)) {
21969 +               AU_MVD_PR(dmsg, "wh_lkup failed\n");
21970 +               goto out;
21971 +       }
21972 +
21973 +       err = 0;
21974 +       if (h_path.dentry->d_inode) {
21975 +               h_path.mnt = au_br_mnt(br);
21976 +               delegated = NULL;
21977 +               err = vfsub_unlink(a->mvd_h_dst_parent->d_inode, &h_path,
21978 +                                  &delegated, /*force*/0);
21979 +               if (unlikely(err == -EWOULDBLOCK)) {
21980 +                       pr_warn("cannot retry for NFSv4 delegation"
21981 +                               " for an internal unlink\n");
21982 +                       iput(delegated);
21983 +               }
21984 +               if (unlikely(err))
21985 +                       AU_MVD_PR(dmsg, "wh_unlink failed\n");
21986 +       }
21987 +       dput(h_path.dentry);
21988 +
21989 +out:
21990 +       AuTraceErr(err);
21991 +       return err;
21992 +}
21993 +
21994 +/*
21995 + * unlink the topmost h_dentry
21996 + */
21997 +static int au_do_unlink(const unsigned char dmsg, struct au_mvd_args *a)
21998 +{
21999 +       int err;
22000 +       struct path h_path;
22001 +       struct inode *delegated;
22002 +
22003 +       h_path.mnt = au_sbr_mnt(a->sb, a->mvd_bsrc);
22004 +       h_path.dentry = au_h_dptr(a->dentry, a->mvd_bsrc);
22005 +       delegated = NULL;
22006 +       err = vfsub_unlink(a->mvd_h_src_dir, &h_path, &delegated, /*force*/0);
22007 +       if (unlikely(err == -EWOULDBLOCK)) {
22008 +               pr_warn("cannot retry for NFSv4 delegation"
22009 +                       " for an internal unlink\n");
22010 +               iput(delegated);
22011 +       }
22012 +       if (unlikely(err))
22013 +               AU_MVD_PR(dmsg, "unlink failed\n");
22014 +
22015 +       AuTraceErr(err);
22016 +       return err;
22017 +}
22018 +
22019 +/* Since mvdown succeeded, we ignore an error of this function */
22020 +static void au_do_stfs(const unsigned char dmsg, struct au_mvd_args *a)
22021 +{
22022 +       int err;
22023 +       struct au_branch *br;
22024 +
22025 +       a->mvdown.flags |= AUFS_MVDOWN_STFS_FAILED;
22026 +       br = au_sbr(a->sb, a->mvd_bsrc);
22027 +       err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_UPPER].stfs);
22028 +       if (!err) {
22029 +               br = au_sbr(a->sb, a->mvd_bdst);
22030 +               a->mvdown.stbr[AUFS_MVDOWN_LOWER].brid = br->br_id;
22031 +               err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_LOWER].stfs);
22032 +       }
22033 +       if (!err)
22034 +               a->mvdown.flags &= ~AUFS_MVDOWN_STFS_FAILED;
22035 +       else
22036 +               AU_MVD_PR(dmsg, "statfs failed (%d), ignored\n", err);
22037 +}
22038 +
22039 +/*
22040 + * copy-down the file and unlink the bsrc file.
22041 + * - unlink the bdst whout if exist
22042 + * - copy-down the file (with whtmp name and rename)
22043 + * - unlink the bsrc file
22044 + */
22045 +static int au_do_mvdown(const unsigned char dmsg, struct au_mvd_args *a)
22046 +{
22047 +       int err;
22048 +
22049 +       err = au_do_mkdir(dmsg, a);
22050 +       if (!err)
22051 +               err = au_do_lock(dmsg, a);
22052 +       if (unlikely(err))
22053 +               goto out;
22054 +
22055 +       /*
22056 +        * do not revert the activities we made on bdst since they should be
22057 +        * harmless in aufs.
22058 +        */
22059 +
22060 +       err = au_do_cpdown(dmsg, a);
22061 +       if (!err)
22062 +               err = au_do_unlink_wh(dmsg, a);
22063 +       if (!err && !(a->mvdown.flags & AUFS_MVDOWN_KUPPER))
22064 +               err = au_do_unlink(dmsg, a);
22065 +       if (unlikely(err))
22066 +               goto out_unlock;
22067 +
22068 +       AuDbg("%pd2, 0x%x, %d --> %d\n",
22069 +             a->dentry, a->mvdown.flags, a->mvd_bsrc, a->mvd_bdst);
22070 +       if (find_lower_writable(a) < 0)
22071 +               a->mvdown.flags |= AUFS_MVDOWN_BOTTOM;
22072 +
22073 +       if (a->mvdown.flags & AUFS_MVDOWN_STFS)
22074 +               au_do_stfs(dmsg, a);
22075 +
22076 +       /* maintain internal array */
22077 +       if (!(a->mvdown.flags & AUFS_MVDOWN_KUPPER)) {
22078 +               au_set_h_dptr(a->dentry, a->mvd_bsrc, NULL);
22079 +               au_set_dbstart(a->dentry, a->mvd_bdst);
22080 +               au_set_h_iptr(a->inode, a->mvd_bsrc, NULL, /*flags*/0);
22081 +               au_set_ibstart(a->inode, a->mvd_bdst);
22082 +       }
22083 +       if (au_dbend(a->dentry) < a->mvd_bdst)
22084 +               au_set_dbend(a->dentry, a->mvd_bdst);
22085 +       if (au_ibend(a->inode) < a->mvd_bdst)
22086 +               au_set_ibend(a->inode, a->mvd_bdst);
22087 +
22088 +out_unlock:
22089 +       au_do_unlock(dmsg, a);
22090 +out:
22091 +       AuTraceErr(err);
22092 +       return err;
22093 +}
22094 +
22095 +/* ---------------------------------------------------------------------- */
22096 +
22097 +/* make sure the file is idle */
22098 +static int au_mvd_args_busy(const unsigned char dmsg, struct au_mvd_args *a)
22099 +{
22100 +       int err, plinked;
22101 +
22102 +       err = 0;
22103 +       plinked = !!au_opt_test(au_mntflags(a->sb), PLINK);
22104 +       if (au_dbstart(a->dentry) == a->mvd_bsrc
22105 +           && au_dcount(a->dentry) == 1
22106 +           && atomic_read(&a->inode->i_count) == 1
22107 +           /* && a->mvd_h_src_inode->i_nlink == 1 */
22108 +           && (!plinked || !au_plink_test(a->inode))
22109 +           && a->inode->i_nlink == 1)
22110 +               goto out;
22111 +
22112 +       err = -EBUSY;
22113 +       AU_MVD_PR(dmsg,
22114 +                 "b%d, d{b%d, c%d?}, i{c%d?, l%u}, hi{l%u}, p{%d, %d}\n",
22115 +                 a->mvd_bsrc, au_dbstart(a->dentry), au_dcount(a->dentry),
22116 +                 atomic_read(&a->inode->i_count), a->inode->i_nlink,
22117 +                 a->mvd_h_src_inode->i_nlink,
22118 +                 plinked, plinked ? au_plink_test(a->inode) : 0);
22119 +
22120 +out:
22121 +       AuTraceErr(err);
22122 +       return err;
22123 +}
22124 +
22125 +/* make sure the parent dir is fine */
22126 +static int au_mvd_args_parent(const unsigned char dmsg,
22127 +                             struct au_mvd_args *a)
22128 +{
22129 +       int err;
22130 +       aufs_bindex_t bindex;
22131 +
22132 +       err = 0;
22133 +       if (unlikely(au_alive_dir(a->parent))) {
22134 +               err = -ENOENT;
22135 +               AU_MVD_PR(dmsg, "parent dir is dead\n");
22136 +               goto out;
22137 +       }
22138 +
22139 +       a->bopq = au_dbdiropq(a->parent);
22140 +       bindex = au_wbr_nonopq(a->dentry, a->mvd_bdst);
22141 +       AuDbg("b%d\n", bindex);
22142 +       if (unlikely((bindex >= 0 && bindex < a->mvd_bdst)
22143 +                    || (a->bopq != -1 && a->bopq < a->mvd_bdst))) {
22144 +               err = -EINVAL;
22145 +               a->mvd_errno = EAU_MVDOWN_OPAQUE;
22146 +               AU_MVD_PR(dmsg, "ancestor is opaque b%d, b%d\n",
22147 +                         a->bopq, a->mvd_bdst);
22148 +       }
22149 +
22150 +out:
22151 +       AuTraceErr(err);
22152 +       return err;
22153 +}
22154 +
22155 +static int au_mvd_args_intermediate(const unsigned char dmsg,
22156 +                                   struct au_mvd_args *a)
22157 +{
22158 +       int err;
22159 +       struct au_dinfo *dinfo, *tmp;
22160 +
22161 +       /* lookup the next lower positive entry */
22162 +       err = -ENOMEM;
22163 +       tmp = au_di_alloc(a->sb, AuLsc_DI_TMP);
22164 +       if (unlikely(!tmp))
22165 +               goto out;
22166 +
22167 +       a->bfound = -1;
22168 +       a->bwh = -1;
22169 +       dinfo = au_di(a->dentry);
22170 +       au_di_cp(tmp, dinfo);
22171 +       au_di_swap(tmp, dinfo);
22172 +
22173 +       /* returns the number of positive dentries */
22174 +       err = au_lkup_dentry(a->dentry, a->mvd_bsrc + 1, /*type*/0);
22175 +       if (!err)
22176 +               a->bwh = au_dbwh(a->dentry);
22177 +       else if (err > 0)
22178 +               a->bfound = au_dbstart(a->dentry);
22179 +
22180 +       au_di_swap(tmp, dinfo);
22181 +       au_rw_write_unlock(&tmp->di_rwsem);
22182 +       au_di_free(tmp);
22183 +       if (unlikely(err < 0))
22184 +               AU_MVD_PR(dmsg, "failed look-up lower\n");
22185 +
22186 +       /*
22187 +        * here, we have these cases.
22188 +        * bfound == -1
22189 +        *      no positive dentry under bsrc. there are more sub-cases.
22190 +        *      bwh < 0
22191 +        *              there no whiteout, we can safely move-down.
22192 +        *      bwh <= bsrc
22193 +        *              impossible
22194 +        *      bsrc < bwh && bwh < bdst
22195 +        *              there is a whiteout on RO branch. cannot proceed.
22196 +        *      bwh == bdst
22197 +        *              there is a whiteout on the RW target branch. it should
22198 +        *              be removed.
22199 +        *      bdst < bwh
22200 +        *              there is a whiteout somewhere unrelated branch.
22201 +        * -1 < bfound && bfound <= bsrc
22202 +        *      impossible.
22203 +        * bfound < bdst
22204 +        *      found, but it is on RO branch between bsrc and bdst. cannot
22205 +        *      proceed.
22206 +        * bfound == bdst
22207 +        *      found, replace it if AUFS_MVDOWN_FORCE is set. otherwise return
22208 +        *      error.
22209 +        * bdst < bfound
22210 +        *      found, after we create the file on bdst, it will be hidden.
22211 +        */
22212 +
22213 +       AuDebugOn(a->bfound == -1
22214 +                 && a->bwh != -1
22215 +                 && a->bwh <= a->mvd_bsrc);
22216 +       AuDebugOn(-1 < a->bfound
22217 +                 && a->bfound <= a->mvd_bsrc);
22218 +
22219 +       err = -EINVAL;
22220 +       if (a->bfound == -1
22221 +           && a->mvd_bsrc < a->bwh
22222 +           && a->bwh != -1
22223 +           && a->bwh < a->mvd_bdst) {
22224 +               a->mvd_errno = EAU_MVDOWN_WHITEOUT;
22225 +               AU_MVD_PR(dmsg, "bsrc %d, bdst %d, bfound %d, bwh %d\n",
22226 +                         a->mvd_bsrc, a->mvd_bdst, a->bfound, a->bwh);
22227 +               goto out;
22228 +       } else if (a->bfound != -1 && a->bfound < a->mvd_bdst) {
22229 +               a->mvd_errno = EAU_MVDOWN_UPPER;
22230 +               AU_MVD_PR(dmsg, "bdst %d, bfound %d\n",
22231 +                         a->mvd_bdst, a->bfound);
22232 +               goto out;
22233 +       }
22234 +
22235 +       err = 0; /* success */
22236 +
22237 +out:
22238 +       AuTraceErr(err);
22239 +       return err;
22240 +}
22241 +
22242 +static int au_mvd_args_exist(const unsigned char dmsg, struct au_mvd_args *a)
22243 +{
22244 +       int err;
22245 +
22246 +       err = 0;
22247 +       if (!(a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
22248 +           && a->bfound == a->mvd_bdst)
22249 +               err = -EEXIST;
22250 +       AuTraceErr(err);
22251 +       return err;
22252 +}
22253 +
22254 +static int au_mvd_args(const unsigned char dmsg, struct au_mvd_args *a)
22255 +{
22256 +       int err;
22257 +       struct au_branch *br;
22258 +
22259 +       err = -EISDIR;
22260 +       if (unlikely(S_ISDIR(a->inode->i_mode)))
22261 +               goto out;
22262 +
22263 +       err = -EINVAL;
22264 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_UPPER))
22265 +               a->mvd_bsrc = au_ibstart(a->inode);
22266 +       else {
22267 +               a->mvd_bsrc = au_br_index(a->sb, a->mvd_src_brid);
22268 +               if (unlikely(a->mvd_bsrc < 0
22269 +                            || (a->mvd_bsrc < au_dbstart(a->dentry)
22270 +                                || au_dbend(a->dentry) < a->mvd_bsrc
22271 +                                || !au_h_dptr(a->dentry, a->mvd_bsrc))
22272 +                            || (a->mvd_bsrc < au_ibstart(a->inode)
22273 +                                || au_ibend(a->inode) < a->mvd_bsrc
22274 +                                || !au_h_iptr(a->inode, a->mvd_bsrc)))) {
22275 +                       a->mvd_errno = EAU_MVDOWN_NOUPPER;
22276 +                       AU_MVD_PR(dmsg, "no upper\n");
22277 +                       goto out;
22278 +               }
22279 +       }
22280 +       if (unlikely(a->mvd_bsrc == au_sbend(a->sb))) {
22281 +               a->mvd_errno = EAU_MVDOWN_BOTTOM;
22282 +               AU_MVD_PR(dmsg, "on the bottom\n");
22283 +               goto out;
22284 +       }
22285 +       a->mvd_h_src_inode = au_h_iptr(a->inode, a->mvd_bsrc);
22286 +       br = au_sbr(a->sb, a->mvd_bsrc);
22287 +       err = au_br_rdonly(br);
22288 +       if (!(a->mvdown.flags & AUFS_MVDOWN_ROUPPER)) {
22289 +               if (unlikely(err))
22290 +                       goto out;
22291 +       } else if (!(vfsub_native_ro(a->mvd_h_src_inode)
22292 +                    || IS_APPEND(a->mvd_h_src_inode))) {
22293 +               if (err)
22294 +                       a->mvdown.flags |= AUFS_MVDOWN_ROUPPER_R;
22295 +               /* go on */
22296 +       } else
22297 +               goto out;
22298 +
22299 +       err = -EINVAL;
22300 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_LOWER)) {
22301 +               a->mvd_bdst = find_lower_writable(a);
22302 +               if (unlikely(a->mvd_bdst < 0)) {
22303 +                       a->mvd_errno = EAU_MVDOWN_BOTTOM;
22304 +                       AU_MVD_PR(dmsg, "no writable lower branch\n");
22305 +                       goto out;
22306 +               }
22307 +       } else {
22308 +               a->mvd_bdst = au_br_index(a->sb, a->mvd_dst_brid);
22309 +               if (unlikely(a->mvd_bdst < 0
22310 +                            || au_sbend(a->sb) < a->mvd_bdst)) {
22311 +                       a->mvd_errno = EAU_MVDOWN_NOLOWERBR;
22312 +                       AU_MVD_PR(dmsg, "no lower brid\n");
22313 +                       goto out;
22314 +               }
22315 +       }
22316 +
22317 +       err = au_mvd_args_busy(dmsg, a);
22318 +       if (!err)
22319 +               err = au_mvd_args_parent(dmsg, a);
22320 +       if (!err)
22321 +               err = au_mvd_args_intermediate(dmsg, a);
22322 +       if (!err)
22323 +               err = au_mvd_args_exist(dmsg, a);
22324 +       if (!err)
22325 +               AuDbg("b%d, b%d\n", a->mvd_bsrc, a->mvd_bdst);
22326 +
22327 +out:
22328 +       AuTraceErr(err);
22329 +       return err;
22330 +}
22331 +
22332 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *uarg)
22333 +{
22334 +       int err, e;
22335 +       unsigned char dmsg;
22336 +       struct au_mvd_args *args;
22337 +
22338 +       err = -EPERM;
22339 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
22340 +               goto out;
22341 +
22342 +       err = -ENOMEM;
22343 +       args = kmalloc(sizeof(*args), GFP_NOFS);
22344 +       if (unlikely(!args))
22345 +               goto out;
22346 +
22347 +       err = copy_from_user(&args->mvdown, uarg, sizeof(args->mvdown));
22348 +       if (!err)
22349 +               err = !access_ok(VERIFY_WRITE, uarg, sizeof(*uarg));
22350 +       if (unlikely(err)) {
22351 +               err = -EFAULT;
22352 +               AuTraceErr(err);
22353 +               goto out_free;
22354 +       }
22355 +       AuDbg("flags 0x%x\n", args->mvdown.flags);
22356 +       args->mvdown.flags &= ~(AUFS_MVDOWN_ROLOWER_R | AUFS_MVDOWN_ROUPPER_R);
22357 +       args->mvdown.au_errno = 0;
22358 +       args->dentry = dentry;
22359 +       args->inode = dentry->d_inode;
22360 +       args->sb = dentry->d_sb;
22361 +
22362 +       err = -ENOENT;
22363 +       dmsg = !!(args->mvdown.flags & AUFS_MVDOWN_DMSG);
22364 +       args->parent = dget_parent(dentry);
22365 +       args->dir = args->parent->d_inode;
22366 +       mutex_lock_nested(&args->dir->i_mutex, I_MUTEX_PARENT);
22367 +       dput(args->parent);
22368 +       if (unlikely(args->parent != dentry->d_parent)) {
22369 +               AU_MVD_PR(dmsg, "parent dir is moved\n");
22370 +               goto out_dir;
22371 +       }
22372 +
22373 +       mutex_lock_nested(&args->inode->i_mutex, I_MUTEX_CHILD);
22374 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH);
22375 +       if (unlikely(err))
22376 +               goto out_inode;
22377 +
22378 +       di_write_lock_parent(args->parent);
22379 +       err = au_mvd_args(dmsg, args);
22380 +       if (unlikely(err))
22381 +               goto out_parent;
22382 +
22383 +       err = au_do_mvdown(dmsg, args);
22384 +       if (unlikely(err))
22385 +               goto out_parent;
22386 +
22387 +       au_cpup_attr_timesizes(args->dir);
22388 +       au_cpup_attr_timesizes(args->inode);
22389 +       au_cpup_igen(args->inode, au_h_iptr(args->inode, args->mvd_bdst));
22390 +       /* au_digen_dec(dentry); */
22391 +
22392 +out_parent:
22393 +       di_write_unlock(args->parent);
22394 +       aufs_read_unlock(dentry, AuLock_DW);
22395 +out_inode:
22396 +       mutex_unlock(&args->inode->i_mutex);
22397 +out_dir:
22398 +       mutex_unlock(&args->dir->i_mutex);
22399 +out_free:
22400 +       e = copy_to_user(uarg, &args->mvdown, sizeof(args->mvdown));
22401 +       if (unlikely(e))
22402 +               err = -EFAULT;
22403 +       kfree(args);
22404 +out:
22405 +       AuTraceErr(err);
22406 +       return err;
22407 +}
22408 diff -urN /usr/share/empty/fs/aufs/opts.c linux/fs/aufs/opts.c
22409 --- /usr/share/empty/fs/aufs/opts.c     1970-01-01 01:00:00.000000000 +0100
22410 +++ linux/fs/aufs/opts.c        2015-01-25 13:00:38.634380408 +0100
22411 @@ -0,0 +1,1850 @@
22412 +/*
22413 + * Copyright (C) 2005-2014 Junjiro R. Okajima
22414 + *
22415 + * This program, aufs is free software; you can redistribute it and/or modify
22416 + * it under the terms of the GNU General Public License as published by
22417 + * the Free Software Foundation; either version 2 of the License, or
22418 + * (at your option) any later version.
22419 + *
22420 + * This program is distributed in the hope that it will be useful,
22421 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22422 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22423 + * GNU General Public License for more details.
22424 + *
22425 + * You should have received a copy of the GNU General Public License
22426 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22427 + */
22428 +
22429 +/*
22430 + * mount options/flags
22431 + */
22432 +
22433 +#include <linux/namei.h>
22434 +#include <linux/types.h> /* a distribution requires */
22435 +#include <linux/parser.h>
22436 +#include "aufs.h"
22437 +
22438 +/* ---------------------------------------------------------------------- */
22439 +
22440 +enum {
22441 +       Opt_br,
22442 +       Opt_add, Opt_del, Opt_mod, Opt_reorder, Opt_append, Opt_prepend,
22443 +       Opt_idel, Opt_imod, Opt_ireorder,
22444 +       Opt_dirwh, Opt_rdcache, Opt_rdblk, Opt_rdhash, Opt_rendir,
22445 +       Opt_rdblk_def, Opt_rdhash_def,
22446 +       Opt_xino, Opt_zxino, Opt_noxino,
22447 +       Opt_trunc_xino, Opt_trunc_xino_v, Opt_notrunc_xino,
22448 +       Opt_trunc_xino_path, Opt_itrunc_xino,
22449 +       Opt_trunc_xib, Opt_notrunc_xib,
22450 +       Opt_shwh, Opt_noshwh,
22451 +       Opt_plink, Opt_noplink, Opt_list_plink,
22452 +       Opt_udba,
22453 +       Opt_dio, Opt_nodio,
22454 +       /* Opt_lock, Opt_unlock, */
22455 +       Opt_cmd, Opt_cmd_args,
22456 +       Opt_diropq_a, Opt_diropq_w,
22457 +       Opt_warn_perm, Opt_nowarn_perm,
22458 +       Opt_wbr_copyup, Opt_wbr_create,
22459 +       Opt_fhsm_sec,
22460 +       Opt_refrof, Opt_norefrof,
22461 +       Opt_verbose, Opt_noverbose,
22462 +       Opt_sum, Opt_nosum, Opt_wsum,
22463 +       Opt_dirperm1, Opt_nodirperm1,
22464 +       Opt_acl, Opt_noacl,
22465 +       Opt_tail, Opt_ignore, Opt_ignore_silent, Opt_err
22466 +};
22467 +
22468 +static match_table_t options = {
22469 +       {Opt_br, "br=%s"},
22470 +       {Opt_br, "br:%s"},
22471 +
22472 +       {Opt_add, "add=%d:%s"},
22473 +       {Opt_add, "add:%d:%s"},
22474 +       {Opt_add, "ins=%d:%s"},
22475 +       {Opt_add, "ins:%d:%s"},
22476 +       {Opt_append, "append=%s"},
22477 +       {Opt_append, "append:%s"},
22478 +       {Opt_prepend, "prepend=%s"},
22479 +       {Opt_prepend, "prepend:%s"},
22480 +
22481 +       {Opt_del, "del=%s"},
22482 +       {Opt_del, "del:%s"},
22483 +       /* {Opt_idel, "idel:%d"}, */
22484 +       {Opt_mod, "mod=%s"},
22485 +       {Opt_mod, "mod:%s"},
22486 +       /* {Opt_imod, "imod:%d:%s"}, */
22487 +
22488 +       {Opt_dirwh, "dirwh=%d"},
22489 +
22490 +       {Opt_xino, "xino=%s"},
22491 +       {Opt_noxino, "noxino"},
22492 +       {Opt_trunc_xino, "trunc_xino"},
22493 +       {Opt_trunc_xino_v, "trunc_xino_v=%d:%d"},
22494 +       {Opt_notrunc_xino, "notrunc_xino"},
22495 +       {Opt_trunc_xino_path, "trunc_xino=%s"},
22496 +       {Opt_itrunc_xino, "itrunc_xino=%d"},
22497 +       /* {Opt_zxino, "zxino=%s"}, */
22498 +       {Opt_trunc_xib, "trunc_xib"},
22499 +       {Opt_notrunc_xib, "notrunc_xib"},
22500 +
22501 +#ifdef CONFIG_PROC_FS
22502 +       {Opt_plink, "plink"},
22503 +#else
22504 +       {Opt_ignore_silent, "plink"},
22505 +#endif
22506 +
22507 +       {Opt_noplink, "noplink"},
22508 +
22509 +#ifdef CONFIG_AUFS_DEBUG
22510 +       {Opt_list_plink, "list_plink"},
22511 +#endif
22512 +
22513 +       {Opt_udba, "udba=%s"},
22514 +
22515 +       {Opt_dio, "dio"},
22516 +       {Opt_nodio, "nodio"},
22517 +
22518 +#ifdef CONFIG_AUFS_FHSM
22519 +       {Opt_fhsm_sec, "fhsm_sec=%d"},
22520 +#else
22521 +       {Opt_ignore_silent, "fhsm_sec=%d"},
22522 +#endif
22523 +
22524 +       {Opt_diropq_a, "diropq=always"},
22525 +       {Opt_diropq_a, "diropq=a"},
22526 +       {Opt_diropq_w, "diropq=whiteouted"},
22527 +       {Opt_diropq_w, "diropq=w"},
22528 +
22529 +       {Opt_warn_perm, "warn_perm"},
22530 +       {Opt_nowarn_perm, "nowarn_perm"},
22531 +
22532 +       /* keep them temporary */
22533 +       {Opt_ignore_silent, "nodlgt"},
22534 +       {Opt_ignore_silent, "clean_plink"},
22535 +
22536 +#ifdef CONFIG_AUFS_SHWH
22537 +       {Opt_shwh, "shwh"},
22538 +#endif
22539 +       {Opt_noshwh, "noshwh"},
22540 +
22541 +       {Opt_dirperm1, "dirperm1"},
22542 +       {Opt_nodirperm1, "nodirperm1"},
22543 +
22544 +       {Opt_rendir, "rendir=%d"},
22545 +
22546 +       {Opt_refrof, "refrof"},
22547 +       {Opt_norefrof, "norefrof"},
22548 +
22549 +       {Opt_verbose, "verbose"},
22550 +       {Opt_verbose, "v"},
22551 +       {Opt_noverbose, "noverbose"},
22552 +       {Opt_noverbose, "quiet"},
22553 +       {Opt_noverbose, "q"},
22554 +       {Opt_noverbose, "silent"},
22555 +
22556 +       {Opt_sum, "sum"},
22557 +       {Opt_nosum, "nosum"},
22558 +       {Opt_wsum, "wsum"},
22559 +
22560 +       {Opt_rdcache, "rdcache=%d"},
22561 +       {Opt_rdblk, "rdblk=%d"},
22562 +       {Opt_rdblk_def, "rdblk=def"},
22563 +       {Opt_rdhash, "rdhash=%d"},
22564 +       {Opt_rdhash_def, "rdhash=def"},
22565 +
22566 +       {Opt_wbr_create, "create=%s"},
22567 +       {Opt_wbr_create, "create_policy=%s"},
22568 +       {Opt_wbr_copyup, "cpup=%s"},
22569 +       {Opt_wbr_copyup, "copyup=%s"},
22570 +       {Opt_wbr_copyup, "copyup_policy=%s"},
22571 +
22572 +       /* generic VFS flag */
22573 +#ifdef CONFIG_FS_POSIX_ACL
22574 +       {Opt_acl, "acl"},
22575 +       {Opt_noacl, "noacl"},
22576 +#else
22577 +       {Opt_ignore_silent, "acl"},
22578 +       {Opt_ignore_silent, "noacl"},
22579 +#endif
22580 +
22581 +       /* internal use for the scripts */
22582 +       {Opt_ignore_silent, "si=%s"},
22583 +
22584 +       {Opt_br, "dirs=%s"},
22585 +       {Opt_ignore, "debug=%d"},
22586 +       {Opt_ignore, "delete=whiteout"},
22587 +       {Opt_ignore, "delete=all"},
22588 +       {Opt_ignore, "imap=%s"},
22589 +
22590 +       /* temporary workaround, due to old mount(8)? */
22591 +       {Opt_ignore_silent, "relatime"},
22592 +
22593 +       {Opt_err, NULL}
22594 +};
22595 +
22596 +/* ---------------------------------------------------------------------- */
22597 +
22598 +static const char *au_parser_pattern(int val, match_table_t tbl)
22599 +{
22600 +       struct match_token *p;
22601 +
22602 +       p = tbl;
22603 +       while (p->pattern) {
22604 +               if (p->token == val)
22605 +                       return p->pattern;
22606 +               p++;
22607 +       }
22608 +       BUG();
22609 +       return "??";
22610 +}
22611 +
22612 +static const char *au_optstr(int *val, match_table_t tbl)
22613 +{
22614 +       struct match_token *p;
22615 +       int v;
22616 +
22617 +       v = *val;
22618 +       p = tbl;
22619 +       while (p->token) {
22620 +               if ((v & p->token) == p->token) {
22621 +                       *val &= ~p->token;
22622 +                       return p->pattern;
22623 +               }
22624 +               p++;
22625 +       }
22626 +       return NULL;
22627 +}
22628 +
22629 +/* ---------------------------------------------------------------------- */
22630 +
22631 +static match_table_t brperm = {
22632 +       {AuBrPerm_RO, AUFS_BRPERM_RO},
22633 +       {AuBrPerm_RR, AUFS_BRPERM_RR},
22634 +       {AuBrPerm_RW, AUFS_BRPERM_RW},
22635 +       {0, NULL}
22636 +};
22637 +
22638 +static match_table_t brattr = {
22639 +       /* general */
22640 +       {AuBrAttr_COO_REG, AUFS_BRATTR_COO_REG},
22641 +       {AuBrAttr_COO_ALL, AUFS_BRATTR_COO_ALL},
22642 +       /* 'unpin' attrib is meaningless since linux-3.18-rc1 */
22643 +       {AuBrAttr_UNPIN, AUFS_BRATTR_UNPIN},
22644 +       {AuBrAttr_FHSM, AUFS_BRATTR_FHSM},
22645 +       {AuBrAttr_ICEX, AUFS_BRATTR_ICEX},
22646 +       {AuBrAttr_ICEX_SEC, AUFS_BRATTR_ICEX_SEC},
22647 +       {AuBrAttr_ICEX_SYS, AUFS_BRATTR_ICEX_SYS},
22648 +       {AuBrAttr_ICEX_TR, AUFS_BRATTR_ICEX_TR},
22649 +       {AuBrAttr_ICEX_USR, AUFS_BRATTR_ICEX_USR},
22650 +       {AuBrAttr_ICEX_OTH, AUFS_BRATTR_ICEX_OTH},
22651 +
22652 +       /* ro/rr branch */
22653 +       {AuBrRAttr_WH, AUFS_BRRATTR_WH},
22654 +
22655 +       /* rw branch */
22656 +       {AuBrWAttr_MOO, AUFS_BRWATTR_MOO},
22657 +       {AuBrWAttr_NoLinkWH, AUFS_BRWATTR_NLWH},
22658 +
22659 +       {0, NULL}
22660 +};
22661 +
22662 +static int br_attr_val(char *str, match_table_t table, substring_t args[])
22663 +{
22664 +       int attr, v;
22665 +       char *p;
22666 +
22667 +       attr = 0;
22668 +       do {
22669 +               p = strchr(str, '+');
22670 +               if (p)
22671 +                       *p = 0;
22672 +               v = match_token(str, table, args);
22673 +               if (v) {
22674 +                       if (v & AuBrAttr_CMOO_Mask)
22675 +                               attr &= ~AuBrAttr_CMOO_Mask;
22676 +                       attr |= v;
22677 +               } else {
22678 +                       if (p)
22679 +                               *p = '+';
22680 +                       pr_warn("ignored branch attribute %s\n", str);
22681 +                       break;
22682 +               }
22683 +               if (p)
22684 +                       str = p + 1;
22685 +       } while (p);
22686 +
22687 +       return attr;
22688 +}
22689 +
22690 +static int au_do_optstr_br_attr(au_br_perm_str_t *str, int perm)
22691 +{
22692 +       int sz;
22693 +       const char *p;
22694 +       char *q;
22695 +
22696 +       q = str->a;
22697 +       *q = 0;
22698 +       p = au_optstr(&perm, brattr);
22699 +       if (p) {
22700 +               sz = strlen(p);
22701 +               memcpy(q, p, sz + 1);
22702 +               q += sz;
22703 +       } else
22704 +               goto out;
22705 +
22706 +       do {
22707 +               p = au_optstr(&perm, brattr);
22708 +               if (p) {
22709 +                       *q++ = '+';
22710 +                       sz = strlen(p);
22711 +                       memcpy(q, p, sz + 1);
22712 +                       q += sz;
22713 +               }
22714 +       } while (p);
22715 +
22716 +out:
22717 +       return q - str->a;
22718 +}
22719 +
22720 +static int noinline_for_stack br_perm_val(char *perm)
22721 +{
22722 +       int val, bad, sz;
22723 +       char *p;
22724 +       substring_t args[MAX_OPT_ARGS];
22725 +       au_br_perm_str_t attr;
22726 +
22727 +       p = strchr(perm, '+');
22728 +       if (p)
22729 +               *p = 0;
22730 +       val = match_token(perm, brperm, args);
22731 +       if (!val) {
22732 +               if (p)
22733 +                       *p = '+';
22734 +               pr_warn("ignored branch permission %s\n", perm);
22735 +               val = AuBrPerm_RO;
22736 +               goto out;
22737 +       }
22738 +       if (!p)
22739 +               goto out;
22740 +
22741 +       val |= br_attr_val(p + 1, brattr, args);
22742 +
22743 +       bad = 0;
22744 +       switch (val & AuBrPerm_Mask) {
22745 +       case AuBrPerm_RO:
22746 +       case AuBrPerm_RR:
22747 +               bad = val & AuBrWAttr_Mask;
22748 +               val &= ~AuBrWAttr_Mask;
22749 +               break;
22750 +       case AuBrPerm_RW:
22751 +               bad = val & AuBrRAttr_Mask;
22752 +               val &= ~AuBrRAttr_Mask;
22753 +               break;
22754 +       }
22755 +
22756 +       /*
22757 +        * 'unpin' attrib becomes meaningless since linux-3.18-rc1, but aufs
22758 +        * does not treat it as an error, just warning.
22759 +        * this is a tiny guard for the user operation.
22760 +        */
22761 +       if (val & AuBrAttr_UNPIN) {
22762 +               bad |= AuBrAttr_UNPIN;
22763 +               val &= ~AuBrAttr_UNPIN;
22764 +       }
22765 +
22766 +       if (unlikely(bad)) {
22767 +               sz = au_do_optstr_br_attr(&attr, bad);
22768 +               AuDebugOn(!sz);
22769 +               pr_warn("ignored branch attribute %s\n", attr.a);
22770 +       }
22771 +
22772 +out:
22773 +       return val;
22774 +}
22775 +
22776 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm)
22777 +{
22778 +       au_br_perm_str_t attr;
22779 +       const char *p;
22780 +       char *q;
22781 +       int sz;
22782 +
22783 +       q = str->a;
22784 +       p = au_optstr(&perm, brperm);
22785 +       AuDebugOn(!p || !*p);
22786 +       sz = strlen(p);
22787 +       memcpy(q, p, sz + 1);
22788 +       q += sz;
22789 +
22790 +       sz = au_do_optstr_br_attr(&attr, perm);
22791 +       if (sz) {
22792 +               *q++ = '+';
22793 +               memcpy(q, attr.a, sz + 1);
22794 +       }
22795 +
22796 +       AuDebugOn(strlen(str->a) >= sizeof(str->a));
22797 +}
22798 +
22799 +/* ---------------------------------------------------------------------- */
22800 +
22801 +static match_table_t udbalevel = {
22802 +       {AuOpt_UDBA_REVAL, "reval"},
22803 +       {AuOpt_UDBA_NONE, "none"},
22804 +#ifdef CONFIG_AUFS_HNOTIFY
22805 +       {AuOpt_UDBA_HNOTIFY, "notify"}, /* abstraction */
22806 +#ifdef CONFIG_AUFS_HFSNOTIFY
22807 +       {AuOpt_UDBA_HNOTIFY, "fsnotify"},
22808 +#endif
22809 +#endif
22810 +       {-1, NULL}
22811 +};
22812 +
22813 +static int noinline_for_stack udba_val(char *str)
22814 +{
22815 +       substring_t args[MAX_OPT_ARGS];
22816 +
22817 +       return match_token(str, udbalevel, args);
22818 +}
22819 +
22820 +const char *au_optstr_udba(int udba)
22821 +{
22822 +       return au_parser_pattern(udba, udbalevel);
22823 +}
22824 +
22825 +/* ---------------------------------------------------------------------- */
22826 +
22827 +static match_table_t au_wbr_create_policy = {
22828 +       {AuWbrCreate_TDP, "tdp"},
22829 +       {AuWbrCreate_TDP, "top-down-parent"},
22830 +       {AuWbrCreate_RR, "rr"},
22831 +       {AuWbrCreate_RR, "round-robin"},
22832 +       {AuWbrCreate_MFS, "mfs"},
22833 +       {AuWbrCreate_MFS, "most-free-space"},
22834 +       {AuWbrCreate_MFSV, "mfs:%d"},
22835 +       {AuWbrCreate_MFSV, "most-free-space:%d"},
22836 +
22837 +       {AuWbrCreate_MFSRR, "mfsrr:%d"},
22838 +       {AuWbrCreate_MFSRRV, "mfsrr:%d:%d"},
22839 +       {AuWbrCreate_PMFS, "pmfs"},
22840 +       {AuWbrCreate_PMFSV, "pmfs:%d"},
22841 +       {AuWbrCreate_PMFSRR, "pmfsrr:%d"},
22842 +       {AuWbrCreate_PMFSRRV, "pmfsrr:%d:%d"},
22843 +
22844 +       {-1, NULL}
22845 +};
22846 +
22847 +/*
22848 + * cf. linux/lib/parser.c and cmdline.c
22849 + * gave up calling memparse() since it uses simple_strtoull() instead of
22850 + * kstrto...().
22851 + */
22852 +static int noinline_for_stack
22853 +au_match_ull(substring_t *s, unsigned long long *result)
22854 +{
22855 +       int err;
22856 +       unsigned int len;
22857 +       char a[32];
22858 +
22859 +       err = -ERANGE;
22860 +       len = s->to - s->from;
22861 +       if (len + 1 <= sizeof(a)) {
22862 +               memcpy(a, s->from, len);
22863 +               a[len] = '\0';
22864 +               err = kstrtoull(a, 0, result);
22865 +       }
22866 +       return err;
22867 +}
22868 +
22869 +static int au_wbr_mfs_wmark(substring_t *arg, char *str,
22870 +                           struct au_opt_wbr_create *create)
22871 +{
22872 +       int err;
22873 +       unsigned long long ull;
22874 +
22875 +       err = 0;
22876 +       if (!au_match_ull(arg, &ull))
22877 +               create->mfsrr_watermark = ull;
22878 +       else {
22879 +               pr_err("bad integer in %s\n", str);
22880 +               err = -EINVAL;
22881 +       }
22882 +
22883 +       return err;
22884 +}
22885 +
22886 +static int au_wbr_mfs_sec(substring_t *arg, char *str,
22887 +                         struct au_opt_wbr_create *create)
22888 +{
22889 +       int n, err;
22890 +
22891 +       err = 0;
22892 +       if (!match_int(arg, &n) && 0 <= n && n <= AUFS_MFS_MAX_SEC)
22893 +               create->mfs_second = n;
22894 +       else {
22895 +               pr_err("bad integer in %s\n", str);
22896 +               err = -EINVAL;
22897 +       }
22898 +
22899 +       return err;
22900 +}
22901 +
22902 +static int noinline_for_stack
22903 +au_wbr_create_val(char *str, struct au_opt_wbr_create *create)
22904 +{
22905 +       int err, e;
22906 +       substring_t args[MAX_OPT_ARGS];
22907 +
22908 +       err = match_token(str, au_wbr_create_policy, args);
22909 +       create->wbr_create = err;
22910 +       switch (err) {
22911 +       case AuWbrCreate_MFSRRV:
22912 +       case AuWbrCreate_PMFSRRV:
22913 +               e = au_wbr_mfs_wmark(&args[0], str, create);
22914 +               if (!e)
22915 +                       e = au_wbr_mfs_sec(&args[1], str, create);
22916 +               if (unlikely(e))
22917 +                       err = e;
22918 +               break;
22919 +       case AuWbrCreate_MFSRR:
22920 +       case AuWbrCreate_PMFSRR:
22921 +               e = au_wbr_mfs_wmark(&args[0], str, create);
22922 +               if (unlikely(e)) {
22923 +                       err = e;
22924 +                       break;
22925 +               }
22926 +               /*FALLTHROUGH*/
22927 +       case AuWbrCreate_MFS:
22928 +       case AuWbrCreate_PMFS:
22929 +               create->mfs_second = AUFS_MFS_DEF_SEC;
22930 +               break;
22931 +       case AuWbrCreate_MFSV:
22932 +       case AuWbrCreate_PMFSV:
22933 +               e = au_wbr_mfs_sec(&args[0], str, create);
22934 +               if (unlikely(e))
22935 +                       err = e;
22936 +               break;
22937 +       }
22938 +
22939 +       return err;
22940 +}
22941 +
22942 +const char *au_optstr_wbr_create(int wbr_create)
22943 +{
22944 +       return au_parser_pattern(wbr_create, au_wbr_create_policy);
22945 +}
22946 +
22947 +static match_table_t au_wbr_copyup_policy = {
22948 +       {AuWbrCopyup_TDP, "tdp"},
22949 +       {AuWbrCopyup_TDP, "top-down-parent"},
22950 +       {AuWbrCopyup_BUP, "bup"},
22951 +       {AuWbrCopyup_BUP, "bottom-up-parent"},
22952 +       {AuWbrCopyup_BU, "bu"},
22953 +       {AuWbrCopyup_BU, "bottom-up"},
22954 +       {-1, NULL}
22955 +};
22956 +
22957 +static int noinline_for_stack au_wbr_copyup_val(char *str)
22958 +{
22959 +       substring_t args[MAX_OPT_ARGS];
22960 +
22961 +       return match_token(str, au_wbr_copyup_policy, args);
22962 +}
22963 +
22964 +const char *au_optstr_wbr_copyup(int wbr_copyup)
22965 +{
22966 +       return au_parser_pattern(wbr_copyup, au_wbr_copyup_policy);
22967 +}
22968 +
22969 +/* ---------------------------------------------------------------------- */
22970 +
22971 +static const int lkup_dirflags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
22972 +
22973 +static void dump_opts(struct au_opts *opts)
22974 +{
22975 +#ifdef CONFIG_AUFS_DEBUG
22976 +       /* reduce stack space */
22977 +       union {
22978 +               struct au_opt_add *add;
22979 +               struct au_opt_del *del;
22980 +               struct au_opt_mod *mod;
22981 +               struct au_opt_xino *xino;
22982 +               struct au_opt_xino_itrunc *xino_itrunc;
22983 +               struct au_opt_wbr_create *create;
22984 +       } u;
22985 +       struct au_opt *opt;
22986 +
22987 +       opt = opts->opt;
22988 +       while (opt->type != Opt_tail) {
22989 +               switch (opt->type) {
22990 +               case Opt_add:
22991 +                       u.add = &opt->add;
22992 +                       AuDbg("add {b%d, %s, 0x%x, %p}\n",
22993 +                                 u.add->bindex, u.add->pathname, u.add->perm,
22994 +                                 u.add->path.dentry);
22995 +                       break;
22996 +               case Opt_del:
22997 +               case Opt_idel:
22998 +                       u.del = &opt->del;
22999 +                       AuDbg("del {%s, %p}\n",
23000 +                             u.del->pathname, u.del->h_path.dentry);
23001 +                       break;
23002 +               case Opt_mod:
23003 +               case Opt_imod:
23004 +                       u.mod = &opt->mod;
23005 +                       AuDbg("mod {%s, 0x%x, %p}\n",
23006 +                                 u.mod->path, u.mod->perm, u.mod->h_root);
23007 +                       break;
23008 +               case Opt_append:
23009 +                       u.add = &opt->add;
23010 +                       AuDbg("append {b%d, %s, 0x%x, %p}\n",
23011 +                                 u.add->bindex, u.add->pathname, u.add->perm,
23012 +                                 u.add->path.dentry);
23013 +                       break;
23014 +               case Opt_prepend:
23015 +                       u.add = &opt->add;
23016 +                       AuDbg("prepend {b%d, %s, 0x%x, %p}\n",
23017 +                                 u.add->bindex, u.add->pathname, u.add->perm,
23018 +                                 u.add->path.dentry);
23019 +                       break;
23020 +               case Opt_dirwh:
23021 +                       AuDbg("dirwh %d\n", opt->dirwh);
23022 +                       break;
23023 +               case Opt_rdcache:
23024 +                       AuDbg("rdcache %d\n", opt->rdcache);
23025 +                       break;
23026 +               case Opt_rdblk:
23027 +                       AuDbg("rdblk %u\n", opt->rdblk);
23028 +                       break;
23029 +               case Opt_rdblk_def:
23030 +                       AuDbg("rdblk_def\n");
23031 +                       break;
23032 +               case Opt_rdhash:
23033 +                       AuDbg("rdhash %u\n", opt->rdhash);
23034 +                       break;
23035 +               case Opt_rdhash_def:
23036 +                       AuDbg("rdhash_def\n");
23037 +                       break;
23038 +               case Opt_xino:
23039 +                       u.xino = &opt->xino;
23040 +                       AuDbg("xino {%s %pD}\n", u.xino->path, u.xino->file);
23041 +                       break;
23042 +               case Opt_trunc_xino:
23043 +                       AuLabel(trunc_xino);
23044 +                       break;
23045 +               case Opt_notrunc_xino:
23046 +                       AuLabel(notrunc_xino);
23047 +                       break;
23048 +               case Opt_trunc_xino_path:
23049 +               case Opt_itrunc_xino:
23050 +                       u.xino_itrunc = &opt->xino_itrunc;
23051 +                       AuDbg("trunc_xino %d\n", u.xino_itrunc->bindex);
23052 +                       break;
23053 +
23054 +               case Opt_noxino:
23055 +                       AuLabel(noxino);
23056 +                       break;
23057 +               case Opt_trunc_xib:
23058 +                       AuLabel(trunc_xib);
23059 +                       break;
23060 +               case Opt_notrunc_xib:
23061 +                       AuLabel(notrunc_xib);
23062 +                       break;
23063 +               case Opt_shwh:
23064 +                       AuLabel(shwh);
23065 +                       break;
23066 +               case Opt_noshwh:
23067 +                       AuLabel(noshwh);
23068 +                       break;
23069 +               case Opt_dirperm1:
23070 +                       AuLabel(dirperm1);
23071 +                       break;
23072 +               case Opt_nodirperm1:
23073 +                       AuLabel(nodirperm1);
23074 +                       break;
23075 +               case Opt_plink:
23076 +                       AuLabel(plink);
23077 +                       break;
23078 +               case Opt_noplink:
23079 +                       AuLabel(noplink);
23080 +                       break;
23081 +               case Opt_list_plink:
23082 +                       AuLabel(list_plink);
23083 +                       break;
23084 +               case Opt_udba:
23085 +                       AuDbg("udba %d, %s\n",
23086 +                                 opt->udba, au_optstr_udba(opt->udba));
23087 +                       break;
23088 +               case Opt_dio:
23089 +                       AuLabel(dio);
23090 +                       break;
23091 +               case Opt_nodio:
23092 +                       AuLabel(nodio);
23093 +                       break;
23094 +               case Opt_diropq_a:
23095 +                       AuLabel(diropq_a);
23096 +                       break;
23097 +               case Opt_diropq_w:
23098 +                       AuLabel(diropq_w);
23099 +                       break;
23100 +               case Opt_warn_perm:
23101 +                       AuLabel(warn_perm);
23102 +                       break;
23103 +               case Opt_nowarn_perm:
23104 +                       AuLabel(nowarn_perm);
23105 +                       break;
23106 +               case Opt_refrof:
23107 +                       AuLabel(refrof);
23108 +                       break;
23109 +               case Opt_norefrof:
23110 +                       AuLabel(norefrof);
23111 +                       break;
23112 +               case Opt_verbose:
23113 +                       AuLabel(verbose);
23114 +                       break;
23115 +               case Opt_noverbose:
23116 +                       AuLabel(noverbose);
23117 +                       break;
23118 +               case Opt_sum:
23119 +                       AuLabel(sum);
23120 +                       break;
23121 +               case Opt_nosum:
23122 +                       AuLabel(nosum);
23123 +                       break;
23124 +               case Opt_wsum:
23125 +                       AuLabel(wsum);
23126 +                       break;
23127 +               case Opt_wbr_create:
23128 +                       u.create = &opt->wbr_create;
23129 +                       AuDbg("create %d, %s\n", u.create->wbr_create,
23130 +                                 au_optstr_wbr_create(u.create->wbr_create));
23131 +                       switch (u.create->wbr_create) {
23132 +                       case AuWbrCreate_MFSV:
23133 +                       case AuWbrCreate_PMFSV:
23134 +                               AuDbg("%d sec\n", u.create->mfs_second);
23135 +                               break;
23136 +                       case AuWbrCreate_MFSRR:
23137 +                               AuDbg("%llu watermark\n",
23138 +                                         u.create->mfsrr_watermark);
23139 +                               break;
23140 +                       case AuWbrCreate_MFSRRV:
23141 +                       case AuWbrCreate_PMFSRRV:
23142 +                               AuDbg("%llu watermark, %d sec\n",
23143 +                                         u.create->mfsrr_watermark,
23144 +                                         u.create->mfs_second);
23145 +                               break;
23146 +                       }
23147 +                       break;
23148 +               case Opt_wbr_copyup:
23149 +                       AuDbg("copyup %d, %s\n", opt->wbr_copyup,
23150 +                                 au_optstr_wbr_copyup(opt->wbr_copyup));
23151 +                       break;
23152 +               case Opt_fhsm_sec:
23153 +                       AuDbg("fhsm_sec %u\n", opt->fhsm_second);
23154 +                       break;
23155 +               case Opt_acl:
23156 +                       AuLabel(acl);
23157 +                       break;
23158 +               case Opt_noacl:
23159 +                       AuLabel(noacl);
23160 +                       break;
23161 +               default:
23162 +                       BUG();
23163 +               }
23164 +               opt++;
23165 +       }
23166 +#endif
23167 +}
23168 +
23169 +void au_opts_free(struct au_opts *opts)
23170 +{
23171 +       struct au_opt *opt;
23172 +
23173 +       opt = opts->opt;
23174 +       while (opt->type != Opt_tail) {
23175 +               switch (opt->type) {
23176 +               case Opt_add:
23177 +               case Opt_append:
23178 +               case Opt_prepend:
23179 +                       path_put(&opt->add.path);
23180 +                       break;
23181 +               case Opt_del:
23182 +               case Opt_idel:
23183 +                       path_put(&opt->del.h_path);
23184 +                       break;
23185 +               case Opt_mod:
23186 +               case Opt_imod:
23187 +                       dput(opt->mod.h_root);
23188 +                       break;
23189 +               case Opt_xino:
23190 +                       fput(opt->xino.file);
23191 +                       break;
23192 +               }
23193 +               opt++;
23194 +       }
23195 +}
23196 +
23197 +static int opt_add(struct au_opt *opt, char *opt_str, unsigned long sb_flags,
23198 +                  aufs_bindex_t bindex)
23199 +{
23200 +       int err;
23201 +       struct au_opt_add *add = &opt->add;
23202 +       char *p;
23203 +
23204 +       add->bindex = bindex;
23205 +       add->perm = AuBrPerm_RO;
23206 +       add->pathname = opt_str;
23207 +       p = strchr(opt_str, '=');
23208 +       if (p) {
23209 +               *p++ = 0;
23210 +               if (*p)
23211 +                       add->perm = br_perm_val(p);
23212 +       }
23213 +
23214 +       err = vfsub_kern_path(add->pathname, lkup_dirflags, &add->path);
23215 +       if (!err) {
23216 +               if (!p) {
23217 +                       add->perm = AuBrPerm_RO;
23218 +                       if (au_test_fs_rr(add->path.dentry->d_sb))
23219 +                               add->perm = AuBrPerm_RR;
23220 +                       else if (!bindex && !(sb_flags & MS_RDONLY))
23221 +                               add->perm = AuBrPerm_RW;
23222 +               }
23223 +               opt->type = Opt_add;
23224 +               goto out;
23225 +       }
23226 +       pr_err("lookup failed %s (%d)\n", add->pathname, err);
23227 +       err = -EINVAL;
23228 +
23229 +out:
23230 +       return err;
23231 +}
23232 +
23233 +static int au_opts_parse_del(struct au_opt_del *del, substring_t args[])
23234 +{
23235 +       int err;
23236 +
23237 +       del->pathname = args[0].from;
23238 +       AuDbg("del path %s\n", del->pathname);
23239 +
23240 +       err = vfsub_kern_path(del->pathname, lkup_dirflags, &del->h_path);
23241 +       if (unlikely(err))
23242 +               pr_err("lookup failed %s (%d)\n", del->pathname, err);
23243 +
23244 +       return err;
23245 +}
23246 +
23247 +#if 0 /* reserved for future use */
23248 +static int au_opts_parse_idel(struct super_block *sb, aufs_bindex_t bindex,
23249 +                             struct au_opt_del *del, substring_t args[])
23250 +{
23251 +       int err;
23252 +       struct dentry *root;
23253 +
23254 +       err = -EINVAL;
23255 +       root = sb->s_root;
23256 +       aufs_read_lock(root, AuLock_FLUSH);
23257 +       if (bindex < 0 || au_sbend(sb) < bindex) {
23258 +               pr_err("out of bounds, %d\n", bindex);
23259 +               goto out;
23260 +       }
23261 +
23262 +       err = 0;
23263 +       del->h_path.dentry = dget(au_h_dptr(root, bindex));
23264 +       del->h_path.mnt = mntget(au_sbr_mnt(sb, bindex));
23265 +
23266 +out:
23267 +       aufs_read_unlock(root, !AuLock_IR);
23268 +       return err;
23269 +}
23270 +#endif
23271 +
23272 +static int noinline_for_stack
23273 +au_opts_parse_mod(struct au_opt_mod *mod, substring_t args[])
23274 +{
23275 +       int err;
23276 +       struct path path;
23277 +       char *p;
23278 +
23279 +       err = -EINVAL;
23280 +       mod->path = args[0].from;
23281 +       p = strchr(mod->path, '=');
23282 +       if (unlikely(!p)) {
23283 +               pr_err("no permssion %s\n", args[0].from);
23284 +               goto out;
23285 +       }
23286 +
23287 +       *p++ = 0;
23288 +       err = vfsub_kern_path(mod->path, lkup_dirflags, &path);
23289 +       if (unlikely(err)) {
23290 +               pr_err("lookup failed %s (%d)\n", mod->path, err);
23291 +               goto out;
23292 +       }
23293 +
23294 +       mod->perm = br_perm_val(p);
23295 +       AuDbg("mod path %s, perm 0x%x, %s\n", mod->path, mod->perm, p);
23296 +       mod->h_root = dget(path.dentry);
23297 +       path_put(&path);
23298 +
23299 +out:
23300 +       return err;
23301 +}
23302 +
23303 +#if 0 /* reserved for future use */
23304 +static int au_opts_parse_imod(struct super_block *sb, aufs_bindex_t bindex,
23305 +                             struct au_opt_mod *mod, substring_t args[])
23306 +{
23307 +       int err;
23308 +       struct dentry *root;
23309 +
23310 +       err = -EINVAL;
23311 +       root = sb->s_root;
23312 +       aufs_read_lock(root, AuLock_FLUSH);
23313 +       if (bindex < 0 || au_sbend(sb) < bindex) {
23314 +               pr_err("out of bounds, %d\n", bindex);
23315 +               goto out;
23316 +       }
23317 +
23318 +       err = 0;
23319 +       mod->perm = br_perm_val(args[1].from);
23320 +       AuDbg("mod path %s, perm 0x%x, %s\n",
23321 +             mod->path, mod->perm, args[1].from);
23322 +       mod->h_root = dget(au_h_dptr(root, bindex));
23323 +
23324 +out:
23325 +       aufs_read_unlock(root, !AuLock_IR);
23326 +       return err;
23327 +}
23328 +#endif
23329 +
23330 +static int au_opts_parse_xino(struct super_block *sb, struct au_opt_xino *xino,
23331 +                             substring_t args[])
23332 +{
23333 +       int err;
23334 +       struct file *file;
23335 +
23336 +       file = au_xino_create(sb, args[0].from, /*silent*/0);
23337 +       err = PTR_ERR(file);
23338 +       if (IS_ERR(file))
23339 +               goto out;
23340 +
23341 +       err = -EINVAL;
23342 +       if (unlikely(file->f_dentry->d_sb == sb)) {
23343 +               fput(file);
23344 +               pr_err("%s must be outside\n", args[0].from);
23345 +               goto out;
23346 +       }
23347 +
23348 +       err = 0;
23349 +       xino->file = file;
23350 +       xino->path = args[0].from;
23351 +
23352 +out:
23353 +       return err;
23354 +}
23355 +
23356 +static int noinline_for_stack
23357 +au_opts_parse_xino_itrunc_path(struct super_block *sb,
23358 +                              struct au_opt_xino_itrunc *xino_itrunc,
23359 +                              substring_t args[])
23360 +{
23361 +       int err;
23362 +       aufs_bindex_t bend, bindex;
23363 +       struct path path;
23364 +       struct dentry *root;
23365 +
23366 +       err = vfsub_kern_path(args[0].from, lkup_dirflags, &path);
23367 +       if (unlikely(err)) {
23368 +               pr_err("lookup failed %s (%d)\n", args[0].from, err);
23369 +               goto out;
23370 +       }
23371 +
23372 +       xino_itrunc->bindex = -1;
23373 +       root = sb->s_root;
23374 +       aufs_read_lock(root, AuLock_FLUSH);
23375 +       bend = au_sbend(sb);
23376 +       for (bindex = 0; bindex <= bend; bindex++) {
23377 +               if (au_h_dptr(root, bindex) == path.dentry) {
23378 +                       xino_itrunc->bindex = bindex;
23379 +                       break;
23380 +               }
23381 +       }
23382 +       aufs_read_unlock(root, !AuLock_IR);
23383 +       path_put(&path);
23384 +
23385 +       if (unlikely(xino_itrunc->bindex < 0)) {
23386 +               pr_err("no such branch %s\n", args[0].from);
23387 +               err = -EINVAL;
23388 +       }
23389 +
23390 +out:
23391 +       return err;
23392 +}
23393 +
23394 +/* called without aufs lock */
23395 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
23396 +{
23397 +       int err, n, token;
23398 +       aufs_bindex_t bindex;
23399 +       unsigned char skipped;
23400 +       struct dentry *root;
23401 +       struct au_opt *opt, *opt_tail;
23402 +       char *opt_str;
23403 +       /* reduce the stack space */
23404 +       union {
23405 +               struct au_opt_xino_itrunc *xino_itrunc;
23406 +               struct au_opt_wbr_create *create;
23407 +       } u;
23408 +       struct {
23409 +               substring_t args[MAX_OPT_ARGS];
23410 +       } *a;
23411 +
23412 +       err = -ENOMEM;
23413 +       a = kmalloc(sizeof(*a), GFP_NOFS);
23414 +       if (unlikely(!a))
23415 +               goto out;
23416 +
23417 +       root = sb->s_root;
23418 +       err = 0;
23419 +       bindex = 0;
23420 +       opt = opts->opt;
23421 +       opt_tail = opt + opts->max_opt - 1;
23422 +       opt->type = Opt_tail;
23423 +       while (!err && (opt_str = strsep(&str, ",")) && *opt_str) {
23424 +               err = -EINVAL;
23425 +               skipped = 0;
23426 +               token = match_token(opt_str, options, a->args);
23427 +               switch (token) {
23428 +               case Opt_br:
23429 +                       err = 0;
23430 +                       while (!err && (opt_str = strsep(&a->args[0].from, ":"))
23431 +                              && *opt_str) {
23432 +                               err = opt_add(opt, opt_str, opts->sb_flags,
23433 +                                             bindex++);
23434 +                               if (unlikely(!err && ++opt > opt_tail)) {
23435 +                                       err = -E2BIG;
23436 +                                       break;
23437 +                               }
23438 +                               opt->type = Opt_tail;
23439 +                               skipped = 1;
23440 +                       }
23441 +                       break;
23442 +               case Opt_add:
23443 +                       if (unlikely(match_int(&a->args[0], &n))) {
23444 +                               pr_err("bad integer in %s\n", opt_str);
23445 +                               break;
23446 +                       }
23447 +                       bindex = n;
23448 +                       err = opt_add(opt, a->args[1].from, opts->sb_flags,
23449 +                                     bindex);
23450 +                       if (!err)
23451 +                               opt->type = token;
23452 +                       break;
23453 +               case Opt_append:
23454 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
23455 +                                     /*dummy bindex*/1);
23456 +                       if (!err)
23457 +                               opt->type = token;
23458 +                       break;
23459 +               case Opt_prepend:
23460 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
23461 +                                     /*bindex*/0);
23462 +                       if (!err)
23463 +                               opt->type = token;
23464 +                       break;
23465 +               case Opt_del:
23466 +                       err = au_opts_parse_del(&opt->del, a->args);
23467 +                       if (!err)
23468 +                               opt->type = token;
23469 +                       break;
23470 +#if 0 /* reserved for future use */
23471 +               case Opt_idel:
23472 +                       del->pathname = "(indexed)";
23473 +                       if (unlikely(match_int(&args[0], &n))) {
23474 +                               pr_err("bad integer in %s\n", opt_str);
23475 +                               break;
23476 +                       }
23477 +                       err = au_opts_parse_idel(sb, n, &opt->del, a->args);
23478 +                       if (!err)
23479 +                               opt->type = token;
23480 +                       break;
23481 +#endif
23482 +               case Opt_mod:
23483 +                       err = au_opts_parse_mod(&opt->mod, a->args);
23484 +                       if (!err)
23485 +                               opt->type = token;
23486 +                       break;
23487 +#ifdef IMOD /* reserved for future use */
23488 +               case Opt_imod:
23489 +                       u.mod->path = "(indexed)";
23490 +                       if (unlikely(match_int(&a->args[0], &n))) {
23491 +                               pr_err("bad integer in %s\n", opt_str);
23492 +                               break;
23493 +                       }
23494 +                       err = au_opts_parse_imod(sb, n, &opt->mod, a->args);
23495 +                       if (!err)
23496 +                               opt->type = token;
23497 +                       break;
23498 +#endif
23499 +               case Opt_xino:
23500 +                       err = au_opts_parse_xino(sb, &opt->xino, a->args);
23501 +                       if (!err)
23502 +                               opt->type = token;
23503 +                       break;
23504 +
23505 +               case Opt_trunc_xino_path:
23506 +                       err = au_opts_parse_xino_itrunc_path
23507 +                               (sb, &opt->xino_itrunc, a->args);
23508 +                       if (!err)
23509 +                               opt->type = token;
23510 +                       break;
23511 +
23512 +               case Opt_itrunc_xino:
23513 +                       u.xino_itrunc = &opt->xino_itrunc;
23514 +                       if (unlikely(match_int(&a->args[0], &n))) {
23515 +                               pr_err("bad integer in %s\n", opt_str);
23516 +                               break;
23517 +                       }
23518 +                       u.xino_itrunc->bindex = n;
23519 +                       aufs_read_lock(root, AuLock_FLUSH);
23520 +                       if (n < 0 || au_sbend(sb) < n) {
23521 +                               pr_err("out of bounds, %d\n", n);
23522 +                               aufs_read_unlock(root, !AuLock_IR);
23523 +                               break;
23524 +                       }
23525 +                       aufs_read_unlock(root, !AuLock_IR);
23526 +                       err = 0;
23527 +                       opt->type = token;
23528 +                       break;
23529 +
23530 +               case Opt_dirwh:
23531 +                       if (unlikely(match_int(&a->args[0], &opt->dirwh)))
23532 +                               break;
23533 +                       err = 0;
23534 +                       opt->type = token;
23535 +                       break;
23536 +
23537 +               case Opt_rdcache:
23538 +                       if (unlikely(match_int(&a->args[0], &n))) {
23539 +                               pr_err("bad integer in %s\n", opt_str);
23540 +                               break;
23541 +                       }
23542 +                       if (unlikely(n > AUFS_RDCACHE_MAX)) {
23543 +                               pr_err("rdcache must be smaller than %d\n",
23544 +                                      AUFS_RDCACHE_MAX);
23545 +                               break;
23546 +                       }
23547 +                       opt->rdcache = n;
23548 +                       err = 0;
23549 +                       opt->type = token;
23550 +                       break;
23551 +               case Opt_rdblk:
23552 +                       if (unlikely(match_int(&a->args[0], &n)
23553 +                                    || n < 0
23554 +                                    || n > KMALLOC_MAX_SIZE)) {
23555 +                               pr_err("bad integer in %s\n", opt_str);
23556 +                               break;
23557 +                       }
23558 +                       if (unlikely(n && n < NAME_MAX)) {
23559 +                               pr_err("rdblk must be larger than %d\n",
23560 +                                      NAME_MAX);
23561 +                               break;
23562 +                       }
23563 +                       opt->rdblk = n;
23564 +                       err = 0;
23565 +                       opt->type = token;
23566 +                       break;
23567 +               case Opt_rdhash:
23568 +                       if (unlikely(match_int(&a->args[0], &n)
23569 +                                    || n < 0
23570 +                                    || n * sizeof(struct hlist_head)
23571 +                                    > KMALLOC_MAX_SIZE)) {
23572 +                               pr_err("bad integer in %s\n", opt_str);
23573 +                               break;
23574 +                       }
23575 +                       opt->rdhash = n;
23576 +                       err = 0;
23577 +                       opt->type = token;
23578 +                       break;
23579 +
23580 +               case Opt_trunc_xino:
23581 +               case Opt_notrunc_xino:
23582 +               case Opt_noxino:
23583 +               case Opt_trunc_xib:
23584 +               case Opt_notrunc_xib:
23585 +               case Opt_shwh:
23586 +               case Opt_noshwh:
23587 +               case Opt_dirperm1:
23588 +               case Opt_nodirperm1:
23589 +               case Opt_plink:
23590 +               case Opt_noplink:
23591 +               case Opt_list_plink:
23592 +               case Opt_dio:
23593 +               case Opt_nodio:
23594 +               case Opt_diropq_a:
23595 +               case Opt_diropq_w:
23596 +               case Opt_warn_perm:
23597 +               case Opt_nowarn_perm:
23598 +               case Opt_refrof:
23599 +               case Opt_norefrof:
23600 +               case Opt_verbose:
23601 +               case Opt_noverbose:
23602 +               case Opt_sum:
23603 +               case Opt_nosum:
23604 +               case Opt_wsum:
23605 +               case Opt_rdblk_def:
23606 +               case Opt_rdhash_def:
23607 +               case Opt_acl:
23608 +               case Opt_noacl:
23609 +                       err = 0;
23610 +                       opt->type = token;
23611 +                       break;
23612 +
23613 +               case Opt_udba:
23614 +                       opt->udba = udba_val(a->args[0].from);
23615 +                       if (opt->udba >= 0) {
23616 +                               err = 0;
23617 +                               opt->type = token;
23618 +                       } else
23619 +                               pr_err("wrong value, %s\n", opt_str);
23620 +                       break;
23621 +
23622 +               case Opt_wbr_create:
23623 +                       u.create = &opt->wbr_create;
23624 +                       u.create->wbr_create
23625 +                               = au_wbr_create_val(a->args[0].from, u.create);
23626 +                       if (u.create->wbr_create >= 0) {
23627 +                               err = 0;
23628 +                               opt->type = token;
23629 +                       } else
23630 +                               pr_err("wrong value, %s\n", opt_str);
23631 +                       break;
23632 +               case Opt_wbr_copyup:
23633 +                       opt->wbr_copyup = au_wbr_copyup_val(a->args[0].from);
23634 +                       if (opt->wbr_copyup >= 0) {
23635 +                               err = 0;
23636 +                               opt->type = token;
23637 +                       } else
23638 +                               pr_err("wrong value, %s\n", opt_str);
23639 +                       break;
23640 +
23641 +               case Opt_fhsm_sec:
23642 +                       if (unlikely(match_int(&a->args[0], &n)
23643 +                                    || n < 0)) {
23644 +                               pr_err("bad integer in %s\n", opt_str);
23645 +                               break;
23646 +                       }
23647 +                       if (sysaufs_brs) {
23648 +                               opt->fhsm_second = n;
23649 +                               opt->type = token;
23650 +                       } else
23651 +                               pr_warn("ignored %s\n", opt_str);
23652 +                       err = 0;
23653 +                       break;
23654 +
23655 +               case Opt_ignore:
23656 +                       pr_warn("ignored %s\n", opt_str);
23657 +                       /*FALLTHROUGH*/
23658 +               case Opt_ignore_silent:
23659 +                       skipped = 1;
23660 +                       err = 0;
23661 +                       break;
23662 +               case Opt_err:
23663 +                       pr_err("unknown option %s\n", opt_str);
23664 +                       break;
23665 +               }
23666 +
23667 +               if (!err && !skipped) {
23668 +                       if (unlikely(++opt > opt_tail)) {
23669 +                               err = -E2BIG;
23670 +                               opt--;
23671 +                               opt->type = Opt_tail;
23672 +                               break;
23673 +                       }
23674 +                       opt->type = Opt_tail;
23675 +               }
23676 +       }
23677 +
23678 +       kfree(a);
23679 +       dump_opts(opts);
23680 +       if (unlikely(err))
23681 +               au_opts_free(opts);
23682 +
23683 +out:
23684 +       return err;
23685 +}
23686 +
23687 +static int au_opt_wbr_create(struct super_block *sb,
23688 +                            struct au_opt_wbr_create *create)
23689 +{
23690 +       int err;
23691 +       struct au_sbinfo *sbinfo;
23692 +
23693 +       SiMustWriteLock(sb);
23694 +
23695 +       err = 1; /* handled */
23696 +       sbinfo = au_sbi(sb);
23697 +       if (sbinfo->si_wbr_create_ops->fin) {
23698 +               err = sbinfo->si_wbr_create_ops->fin(sb);
23699 +               if (!err)
23700 +                       err = 1;
23701 +       }
23702 +
23703 +       sbinfo->si_wbr_create = create->wbr_create;
23704 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + create->wbr_create;
23705 +       switch (create->wbr_create) {
23706 +       case AuWbrCreate_MFSRRV:
23707 +       case AuWbrCreate_MFSRR:
23708 +       case AuWbrCreate_PMFSRR:
23709 +       case AuWbrCreate_PMFSRRV:
23710 +               sbinfo->si_wbr_mfs.mfsrr_watermark = create->mfsrr_watermark;
23711 +               /*FALLTHROUGH*/
23712 +       case AuWbrCreate_MFS:
23713 +       case AuWbrCreate_MFSV:
23714 +       case AuWbrCreate_PMFS:
23715 +       case AuWbrCreate_PMFSV:
23716 +               sbinfo->si_wbr_mfs.mfs_expire
23717 +                       = msecs_to_jiffies(create->mfs_second * MSEC_PER_SEC);
23718 +               break;
23719 +       }
23720 +
23721 +       if (sbinfo->si_wbr_create_ops->init)
23722 +               sbinfo->si_wbr_create_ops->init(sb); /* ignore */
23723 +
23724 +       return err;
23725 +}
23726 +
23727 +/*
23728 + * returns,
23729 + * plus: processed without an error
23730 + * zero: unprocessed
23731 + */
23732 +static int au_opt_simple(struct super_block *sb, struct au_opt *opt,
23733 +                        struct au_opts *opts)
23734 +{
23735 +       int err;
23736 +       struct au_sbinfo *sbinfo;
23737 +
23738 +       SiMustWriteLock(sb);
23739 +
23740 +       err = 1; /* handled */
23741 +       sbinfo = au_sbi(sb);
23742 +       switch (opt->type) {
23743 +       case Opt_udba:
23744 +               sbinfo->si_mntflags &= ~AuOptMask_UDBA;
23745 +               sbinfo->si_mntflags |= opt->udba;
23746 +               opts->given_udba |= opt->udba;
23747 +               break;
23748 +
23749 +       case Opt_plink:
23750 +               au_opt_set(sbinfo->si_mntflags, PLINK);
23751 +               break;
23752 +       case Opt_noplink:
23753 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
23754 +                       au_plink_put(sb, /*verbose*/1);
23755 +               au_opt_clr(sbinfo->si_mntflags, PLINK);
23756 +               break;
23757 +       case Opt_list_plink:
23758 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
23759 +                       au_plink_list(sb);
23760 +               break;
23761 +
23762 +       case Opt_dio:
23763 +               au_opt_set(sbinfo->si_mntflags, DIO);
23764 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
23765 +               break;
23766 +       case Opt_nodio:
23767 +               au_opt_clr(sbinfo->si_mntflags, DIO);
23768 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
23769 +               break;
23770 +
23771 +       case Opt_fhsm_sec:
23772 +               au_fhsm_set(sbinfo, opt->fhsm_second);
23773 +               break;
23774 +
23775 +       case Opt_diropq_a:
23776 +               au_opt_set(sbinfo->si_mntflags, ALWAYS_DIROPQ);
23777 +               break;
23778 +       case Opt_diropq_w:
23779 +               au_opt_clr(sbinfo->si_mntflags, ALWAYS_DIROPQ);
23780 +               break;
23781 +
23782 +       case Opt_warn_perm:
23783 +               au_opt_set(sbinfo->si_mntflags, WARN_PERM);
23784 +               break;
23785 +       case Opt_nowarn_perm:
23786 +               au_opt_clr(sbinfo->si_mntflags, WARN_PERM);
23787 +               break;
23788 +
23789 +       case Opt_refrof:
23790 +               au_opt_set(sbinfo->si_mntflags, REFROF);
23791 +               break;
23792 +       case Opt_norefrof:
23793 +               au_opt_clr(sbinfo->si_mntflags, REFROF);
23794 +               break;
23795 +
23796 +       case Opt_verbose:
23797 +               au_opt_set(sbinfo->si_mntflags, VERBOSE);
23798 +               break;
23799 +       case Opt_noverbose:
23800 +               au_opt_clr(sbinfo->si_mntflags, VERBOSE);
23801 +               break;
23802 +
23803 +       case Opt_sum:
23804 +               au_opt_set(sbinfo->si_mntflags, SUM);
23805 +               break;
23806 +       case Opt_wsum:
23807 +               au_opt_clr(sbinfo->si_mntflags, SUM);
23808 +               au_opt_set(sbinfo->si_mntflags, SUM_W);
23809 +       case Opt_nosum:
23810 +               au_opt_clr(sbinfo->si_mntflags, SUM);
23811 +               au_opt_clr(sbinfo->si_mntflags, SUM_W);
23812 +               break;
23813 +
23814 +       case Opt_wbr_create:
23815 +               err = au_opt_wbr_create(sb, &opt->wbr_create);
23816 +               break;
23817 +       case Opt_wbr_copyup:
23818 +               sbinfo->si_wbr_copyup = opt->wbr_copyup;
23819 +               sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + opt->wbr_copyup;
23820 +               break;
23821 +
23822 +       case Opt_dirwh:
23823 +               sbinfo->si_dirwh = opt->dirwh;
23824 +               break;
23825 +
23826 +       case Opt_rdcache:
23827 +               sbinfo->si_rdcache
23828 +                       = msecs_to_jiffies(opt->rdcache * MSEC_PER_SEC);
23829 +               break;
23830 +       case Opt_rdblk:
23831 +               sbinfo->si_rdblk = opt->rdblk;
23832 +               break;
23833 +       case Opt_rdblk_def:
23834 +               sbinfo->si_rdblk = AUFS_RDBLK_DEF;
23835 +               break;
23836 +       case Opt_rdhash:
23837 +               sbinfo->si_rdhash = opt->rdhash;
23838 +               break;
23839 +       case Opt_rdhash_def:
23840 +               sbinfo->si_rdhash = AUFS_RDHASH_DEF;
23841 +               break;
23842 +
23843 +       case Opt_shwh:
23844 +               au_opt_set(sbinfo->si_mntflags, SHWH);
23845 +               break;
23846 +       case Opt_noshwh:
23847 +               au_opt_clr(sbinfo->si_mntflags, SHWH);
23848 +               break;
23849 +
23850 +       case Opt_dirperm1:
23851 +               au_opt_set(sbinfo->si_mntflags, DIRPERM1);
23852 +               break;
23853 +       case Opt_nodirperm1:
23854 +               au_opt_clr(sbinfo->si_mntflags, DIRPERM1);
23855 +               break;
23856 +
23857 +       case Opt_trunc_xino:
23858 +               au_opt_set(sbinfo->si_mntflags, TRUNC_XINO);
23859 +               break;
23860 +       case Opt_notrunc_xino:
23861 +               au_opt_clr(sbinfo->si_mntflags, TRUNC_XINO);
23862 +               break;
23863 +
23864 +       case Opt_trunc_xino_path:
23865 +       case Opt_itrunc_xino:
23866 +               err = au_xino_trunc(sb, opt->xino_itrunc.bindex);
23867 +               if (!err)
23868 +                       err = 1;
23869 +               break;
23870 +
23871 +       case Opt_trunc_xib:
23872 +               au_fset_opts(opts->flags, TRUNC_XIB);
23873 +               break;
23874 +       case Opt_notrunc_xib:
23875 +               au_fclr_opts(opts->flags, TRUNC_XIB);
23876 +               break;
23877 +
23878 +       case Opt_acl:
23879 +               sb->s_flags |= MS_POSIXACL;
23880 +               break;
23881 +       case Opt_noacl:
23882 +               sb->s_flags &= ~MS_POSIXACL;
23883 +               break;
23884 +
23885 +       default:
23886 +               err = 0;
23887 +               break;
23888 +       }
23889 +
23890 +       return err;
23891 +}
23892 +
23893 +/*
23894 + * returns tri-state.
23895 + * plus: processed without an error
23896 + * zero: unprocessed
23897 + * minus: error
23898 + */
23899 +static int au_opt_br(struct super_block *sb, struct au_opt *opt,
23900 +                    struct au_opts *opts)
23901 +{
23902 +       int err, do_refresh;
23903 +
23904 +       err = 0;
23905 +       switch (opt->type) {
23906 +       case Opt_append:
23907 +               opt->add.bindex = au_sbend(sb) + 1;
23908 +               if (opt->add.bindex < 0)
23909 +                       opt->add.bindex = 0;
23910 +               goto add;
23911 +       case Opt_prepend:
23912 +               opt->add.bindex = 0;
23913 +       add: /* indented label */
23914 +       case Opt_add:
23915 +               err = au_br_add(sb, &opt->add,
23916 +                               au_ftest_opts(opts->flags, REMOUNT));
23917 +               if (!err) {
23918 +                       err = 1;
23919 +                       au_fset_opts(opts->flags, REFRESH);
23920 +               }
23921 +               break;
23922 +
23923 +       case Opt_del:
23924 +       case Opt_idel:
23925 +               err = au_br_del(sb, &opt->del,
23926 +                               au_ftest_opts(opts->flags, REMOUNT));
23927 +               if (!err) {
23928 +                       err = 1;
23929 +                       au_fset_opts(opts->flags, TRUNC_XIB);
23930 +                       au_fset_opts(opts->flags, REFRESH);
23931 +               }
23932 +               break;
23933 +
23934 +       case Opt_mod:
23935 +       case Opt_imod:
23936 +               err = au_br_mod(sb, &opt->mod,
23937 +                               au_ftest_opts(opts->flags, REMOUNT),
23938 +                               &do_refresh);
23939 +               if (!err) {
23940 +                       err = 1;
23941 +                       if (do_refresh)
23942 +                               au_fset_opts(opts->flags, REFRESH);
23943 +               }
23944 +               break;
23945 +       }
23946 +
23947 +       return err;
23948 +}
23949 +
23950 +static int au_opt_xino(struct super_block *sb, struct au_opt *opt,
23951 +                      struct au_opt_xino **opt_xino,
23952 +                      struct au_opts *opts)
23953 +{
23954 +       int err;
23955 +       aufs_bindex_t bend, bindex;
23956 +       struct dentry *root, *parent, *h_root;
23957 +
23958 +       err = 0;
23959 +       switch (opt->type) {
23960 +       case Opt_xino:
23961 +               err = au_xino_set(sb, &opt->xino,
23962 +                                 !!au_ftest_opts(opts->flags, REMOUNT));
23963 +               if (unlikely(err))
23964 +                       break;
23965 +
23966 +               *opt_xino = &opt->xino;
23967 +               au_xino_brid_set(sb, -1);
23968 +
23969 +               /* safe d_parent access */
23970 +               parent = opt->xino.file->f_dentry->d_parent;
23971 +               root = sb->s_root;
23972 +               bend = au_sbend(sb);
23973 +               for (bindex = 0; bindex <= bend; bindex++) {
23974 +                       h_root = au_h_dptr(root, bindex);
23975 +                       if (h_root == parent) {
23976 +                               au_xino_brid_set(sb, au_sbr_id(sb, bindex));
23977 +                               break;
23978 +                       }
23979 +               }
23980 +               break;
23981 +
23982 +       case Opt_noxino:
23983 +               au_xino_clr(sb);
23984 +               au_xino_brid_set(sb, -1);
23985 +               *opt_xino = (void *)-1;
23986 +               break;
23987 +       }
23988 +
23989 +       return err;
23990 +}
23991 +
23992 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
23993 +                  unsigned int pending)
23994 +{
23995 +       int err, fhsm;
23996 +       aufs_bindex_t bindex, bend;
23997 +       unsigned char do_plink, skip, do_free;
23998 +       struct au_branch *br;
23999 +       struct au_wbr *wbr;
24000 +       struct dentry *root;
24001 +       struct inode *dir, *h_dir;
24002 +       struct au_sbinfo *sbinfo;
24003 +       struct au_hinode *hdir;
24004 +
24005 +       SiMustAnyLock(sb);
24006 +
24007 +       sbinfo = au_sbi(sb);
24008 +       AuDebugOn(!(sbinfo->si_mntflags & AuOptMask_UDBA));
24009 +
24010 +       if (!(sb_flags & MS_RDONLY)) {
24011 +               if (unlikely(!au_br_writable(au_sbr_perm(sb, 0))))
24012 +                       pr_warn("first branch should be rw\n");
24013 +               if (unlikely(au_opt_test(sbinfo->si_mntflags, SHWH)))
24014 +                       pr_warn("shwh should be used with ro\n");
24015 +       }
24016 +
24017 +       if (au_opt_test((sbinfo->si_mntflags | pending), UDBA_HNOTIFY)
24018 +           && !au_opt_test(sbinfo->si_mntflags, XINO))
24019 +               pr_warn("udba=*notify requires xino\n");
24020 +
24021 +       if (au_opt_test(sbinfo->si_mntflags, DIRPERM1))
24022 +               pr_warn("dirperm1 breaks the protection"
24023 +                       " by the permission bits on the lower branch\n");
24024 +
24025 +       err = 0;
24026 +       fhsm = 0;
24027 +       root = sb->s_root;
24028 +       dir = root->d_inode;
24029 +       do_plink = !!au_opt_test(sbinfo->si_mntflags, PLINK);
24030 +       bend = au_sbend(sb);
24031 +       for (bindex = 0; !err && bindex <= bend; bindex++) {
24032 +               skip = 0;
24033 +               h_dir = au_h_iptr(dir, bindex);
24034 +               br = au_sbr(sb, bindex);
24035 +
24036 +               if ((br->br_perm & AuBrAttr_ICEX)
24037 +                   && !h_dir->i_op->listxattr)
24038 +                       br->br_perm &= ~AuBrAttr_ICEX;
24039 +#if 0
24040 +               if ((br->br_perm & AuBrAttr_ICEX_SEC)
24041 +                   && (au_br_sb(br)->s_flags & MS_NOSEC))
24042 +                       br->br_perm &= ~AuBrAttr_ICEX_SEC;
24043 +#endif
24044 +
24045 +               do_free = 0;
24046 +               wbr = br->br_wbr;
24047 +               if (wbr)
24048 +                       wbr_wh_read_lock(wbr);
24049 +
24050 +               if (!au_br_writable(br->br_perm)) {
24051 +                       do_free = !!wbr;
24052 +                       skip = (!wbr
24053 +                               || (!wbr->wbr_whbase
24054 +                                   && !wbr->wbr_plink
24055 +                                   && !wbr->wbr_orph));
24056 +               } else if (!au_br_wh_linkable(br->br_perm)) {
24057 +                       /* skip = (!br->br_whbase && !br->br_orph); */
24058 +                       skip = (!wbr || !wbr->wbr_whbase);
24059 +                       if (skip && wbr) {
24060 +                               if (do_plink)
24061 +                                       skip = !!wbr->wbr_plink;
24062 +                               else
24063 +                                       skip = !wbr->wbr_plink;
24064 +                       }
24065 +               } else {
24066 +                       /* skip = (br->br_whbase && br->br_ohph); */
24067 +                       skip = (wbr && wbr->wbr_whbase);
24068 +                       if (skip) {
24069 +                               if (do_plink)
24070 +                                       skip = !!wbr->wbr_plink;
24071 +                               else
24072 +                                       skip = !wbr->wbr_plink;
24073 +                       }
24074 +               }
24075 +               if (wbr)
24076 +                       wbr_wh_read_unlock(wbr);
24077 +
24078 +               if (au_br_fhsm(br->br_perm)) {
24079 +                       fhsm++;
24080 +                       AuDebugOn(!br->br_fhsm);
24081 +               }
24082 +
24083 +               if (skip)
24084 +                       continue;
24085 +
24086 +               hdir = au_hi(dir, bindex);
24087 +               au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
24088 +               if (wbr)
24089 +                       wbr_wh_write_lock(wbr);
24090 +               err = au_wh_init(br, sb);
24091 +               if (wbr)
24092 +                       wbr_wh_write_unlock(wbr);
24093 +               au_hn_imtx_unlock(hdir);
24094 +
24095 +               if (!err && do_free) {
24096 +                       kfree(wbr);
24097 +                       br->br_wbr = NULL;
24098 +               }
24099 +       }
24100 +
24101 +       if (fhsm >= 2) {
24102 +               au_fset_si(sbinfo, FHSM);
24103 +               for (bindex = bend; bindex >= 0; bindex--) {
24104 +                       br = au_sbr(sb, bindex);
24105 +                       if (au_br_fhsm(br->br_perm)) {
24106 +                               au_fhsm_set_bottom(sb, bindex);
24107 +                               break;
24108 +                       }
24109 +               }
24110 +       } else {
24111 +               au_fclr_si(sbinfo, FHSM);
24112 +               au_fhsm_set_bottom(sb, -1);
24113 +       }
24114 +
24115 +       return err;
24116 +}
24117 +
24118 +int au_opts_mount(struct super_block *sb, struct au_opts *opts)
24119 +{
24120 +       int err;
24121 +       unsigned int tmp;
24122 +       aufs_bindex_t bindex, bend;
24123 +       struct au_opt *opt;
24124 +       struct au_opt_xino *opt_xino, xino;
24125 +       struct au_sbinfo *sbinfo;
24126 +       struct au_branch *br;
24127 +       struct inode *dir;
24128 +
24129 +       SiMustWriteLock(sb);
24130 +
24131 +       err = 0;
24132 +       opt_xino = NULL;
24133 +       opt = opts->opt;
24134 +       while (err >= 0 && opt->type != Opt_tail)
24135 +               err = au_opt_simple(sb, opt++, opts);
24136 +       if (err > 0)
24137 +               err = 0;
24138 +       else if (unlikely(err < 0))
24139 +               goto out;
24140 +
24141 +       /* disable xino and udba temporary */
24142 +       sbinfo = au_sbi(sb);
24143 +       tmp = sbinfo->si_mntflags;
24144 +       au_opt_clr(sbinfo->si_mntflags, XINO);
24145 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_REVAL);
24146 +
24147 +       opt = opts->opt;
24148 +       while (err >= 0 && opt->type != Opt_tail)
24149 +               err = au_opt_br(sb, opt++, opts);
24150 +       if (err > 0)
24151 +               err = 0;
24152 +       else if (unlikely(err < 0))
24153 +               goto out;
24154 +
24155 +       bend = au_sbend(sb);
24156 +       if (unlikely(bend < 0)) {
24157 +               err = -EINVAL;
24158 +               pr_err("no branches\n");
24159 +               goto out;
24160 +       }
24161 +
24162 +       if (au_opt_test(tmp, XINO))
24163 +               au_opt_set(sbinfo->si_mntflags, XINO);
24164 +       opt = opts->opt;
24165 +       while (!err && opt->type != Opt_tail)
24166 +               err = au_opt_xino(sb, opt++, &opt_xino, opts);
24167 +       if (unlikely(err))
24168 +               goto out;
24169 +
24170 +       err = au_opts_verify(sb, sb->s_flags, tmp);
24171 +       if (unlikely(err))
24172 +               goto out;
24173 +
24174 +       /* restore xino */
24175 +       if (au_opt_test(tmp, XINO) && !opt_xino) {
24176 +               xino.file = au_xino_def(sb);
24177 +               err = PTR_ERR(xino.file);
24178 +               if (IS_ERR(xino.file))
24179 +                       goto out;
24180 +
24181 +               err = au_xino_set(sb, &xino, /*remount*/0);
24182 +               fput(xino.file);
24183 +               if (unlikely(err))
24184 +                       goto out;
24185 +       }
24186 +
24187 +       /* restore udba */
24188 +       tmp &= AuOptMask_UDBA;
24189 +       sbinfo->si_mntflags &= ~AuOptMask_UDBA;
24190 +       sbinfo->si_mntflags |= tmp;
24191 +       bend = au_sbend(sb);
24192 +       for (bindex = 0; bindex <= bend; bindex++) {
24193 +               br = au_sbr(sb, bindex);
24194 +               err = au_hnotify_reset_br(tmp, br, br->br_perm);
24195 +               if (unlikely(err))
24196 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
24197 +                               bindex, err);
24198 +               /* go on even if err */
24199 +       }
24200 +       if (au_opt_test(tmp, UDBA_HNOTIFY)) {
24201 +               dir = sb->s_root->d_inode;
24202 +               au_hn_reset(dir, au_hi_flags(dir, /*isdir*/1) & ~AuHi_XINO);
24203 +       }
24204 +
24205 +out:
24206 +       return err;
24207 +}
24208 +
24209 +int au_opts_remount(struct super_block *sb, struct au_opts *opts)
24210 +{
24211 +       int err, rerr;
24212 +       struct inode *dir;
24213 +       struct au_opt_xino *opt_xino;
24214 +       struct au_opt *opt;
24215 +       struct au_sbinfo *sbinfo;
24216 +
24217 +       SiMustWriteLock(sb);
24218 +
24219 +       dir = sb->s_root->d_inode;
24220 +       sbinfo = au_sbi(sb);
24221 +       err = 0;
24222 +       opt_xino = NULL;
24223 +       opt = opts->opt;
24224 +       while (err >= 0 && opt->type != Opt_tail) {
24225 +               err = au_opt_simple(sb, opt, opts);
24226 +               if (!err)
24227 +                       err = au_opt_br(sb, opt, opts);
24228 +               if (!err)
24229 +                       err = au_opt_xino(sb, opt, &opt_xino, opts);
24230 +               opt++;
24231 +       }
24232 +       if (err > 0)
24233 +               err = 0;
24234 +       AuTraceErr(err);
24235 +       /* go on even err */
24236 +
24237 +       rerr = au_opts_verify(sb, opts->sb_flags, /*pending*/0);
24238 +       if (unlikely(rerr && !err))
24239 +               err = rerr;
24240 +
24241 +       if (au_ftest_opts(opts->flags, TRUNC_XIB)) {
24242 +               rerr = au_xib_trunc(sb);
24243 +               if (unlikely(rerr && !err))
24244 +                       err = rerr;
24245 +       }
24246 +
24247 +       /* will be handled by the caller */
24248 +       if (!au_ftest_opts(opts->flags, REFRESH)
24249 +           && (opts->given_udba || au_opt_test(sbinfo->si_mntflags, XINO)))
24250 +               au_fset_opts(opts->flags, REFRESH);
24251 +
24252 +       AuDbg("status 0x%x\n", opts->flags);
24253 +       return err;
24254 +}
24255 +
24256 +/* ---------------------------------------------------------------------- */
24257 +
24258 +unsigned int au_opt_udba(struct super_block *sb)
24259 +{
24260 +       return au_mntflags(sb) & AuOptMask_UDBA;
24261 +}
24262 diff -urN /usr/share/empty/fs/aufs/opts.h linux/fs/aufs/opts.h
24263 --- /usr/share/empty/fs/aufs/opts.h     1970-01-01 01:00:00.000000000 +0100
24264 +++ linux/fs/aufs/opts.h        2015-01-25 13:00:38.634380408 +0100
24265 @@ -0,0 +1,213 @@
24266 +/*
24267 + * Copyright (C) 2005-2014 Junjiro R. Okajima
24268 + *
24269 + * This program, aufs is free software; you can redistribute it and/or modify
24270 + * it under the terms of the GNU General Public License as published by
24271 + * the Free Software Foundation; either version 2 of the License, or
24272 + * (at your option) any later version.
24273 + *
24274 + * This program is distributed in the hope that it will be useful,
24275 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24276 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24277 + * GNU General Public License for more details.
24278 + *
24279 + * You should have received a copy of the GNU General Public License
24280 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24281 + */
24282 +
24283 +/*
24284 + * mount options/flags
24285 + */
24286 +
24287 +#ifndef __AUFS_OPTS_H__
24288 +#define __AUFS_OPTS_H__
24289 +
24290 +#ifdef __KERNEL__
24291 +
24292 +#include <linux/path.h>
24293 +#include "branch.h"
24294 +
24295 +struct file;
24296 +struct super_block;
24297 +
24298 +/* ---------------------------------------------------------------------- */
24299 +
24300 +/* mount flags */
24301 +#define AuOpt_XINO             1               /* external inode number bitmap
24302 +                                                  and translation table */
24303 +#define AuOpt_TRUNC_XINO       (1 << 1)        /* truncate xino files */
24304 +#define AuOpt_UDBA_NONE                (1 << 2)        /* users direct branch access */
24305 +#define AuOpt_UDBA_REVAL       (1 << 3)
24306 +#define AuOpt_UDBA_HNOTIFY     (1 << 4)
24307 +#define AuOpt_SHWH             (1 << 5)        /* show whiteout */
24308 +#define AuOpt_PLINK            (1 << 6)        /* pseudo-link */
24309 +#define AuOpt_DIRPERM1         (1 << 7)        /* ignore the lower dir's perm
24310 +                                                  bits */
24311 +#define AuOpt_REFROF           (1 << 8)        /* unimplemented */
24312 +#define AuOpt_ALWAYS_DIROPQ    (1 << 9)        /* policy to creating diropq */
24313 +#define AuOpt_SUM              (1 << 10)       /* summation for statfs(2) */
24314 +#define AuOpt_SUM_W            (1 << 11)       /* unimplemented */
24315 +#define AuOpt_WARN_PERM                (1 << 12)       /* warn when add-branch */
24316 +#define AuOpt_VERBOSE          (1 << 13)       /* busy inode when del-branch */
24317 +#define AuOpt_DIO              (1 << 14)       /* direct io */
24318 +
24319 +#ifndef CONFIG_AUFS_HNOTIFY
24320 +#undef AuOpt_UDBA_HNOTIFY
24321 +#define AuOpt_UDBA_HNOTIFY     0
24322 +#endif
24323 +#ifndef CONFIG_AUFS_SHWH
24324 +#undef AuOpt_SHWH
24325 +#define AuOpt_SHWH             0
24326 +#endif
24327 +
24328 +#define AuOpt_Def      (AuOpt_XINO \
24329 +                        | AuOpt_UDBA_REVAL \
24330 +                        | AuOpt_PLINK \
24331 +                        /* | AuOpt_DIRPERM1 */ \
24332 +                        | AuOpt_WARN_PERM)
24333 +#define AuOptMask_UDBA (AuOpt_UDBA_NONE \
24334 +                        | AuOpt_UDBA_REVAL \
24335 +                        | AuOpt_UDBA_HNOTIFY)
24336 +
24337 +#define au_opt_test(flags, name)       (flags & AuOpt_##name)
24338 +#define au_opt_set(flags, name) do { \
24339 +       BUILD_BUG_ON(AuOpt_##name & AuOptMask_UDBA); \
24340 +       ((flags) |= AuOpt_##name); \
24341 +} while (0)
24342 +#define au_opt_set_udba(flags, name) do { \
24343 +       (flags) &= ~AuOptMask_UDBA; \
24344 +       ((flags) |= AuOpt_##name); \
24345 +} while (0)
24346 +#define au_opt_clr(flags, name) do { \
24347 +       ((flags) &= ~AuOpt_##name); \
24348 +} while (0)
24349 +
24350 +static inline unsigned int au_opts_plink(unsigned int mntflags)
24351 +{
24352 +#ifdef CONFIG_PROC_FS
24353 +       return mntflags;
24354 +#else
24355 +       return mntflags & ~AuOpt_PLINK;
24356 +#endif
24357 +}
24358 +
24359 +/* ---------------------------------------------------------------------- */
24360 +
24361 +/* policies to select one among multiple writable branches */
24362 +enum {
24363 +       AuWbrCreate_TDP,        /* top down parent */
24364 +       AuWbrCreate_RR,         /* round robin */
24365 +       AuWbrCreate_MFS,        /* most free space */
24366 +       AuWbrCreate_MFSV,       /* mfs with seconds */
24367 +       AuWbrCreate_MFSRR,      /* mfs then rr */
24368 +       AuWbrCreate_MFSRRV,     /* mfs then rr with seconds */
24369 +       AuWbrCreate_PMFS,       /* parent and mfs */
24370 +       AuWbrCreate_PMFSV,      /* parent and mfs with seconds */
24371 +       AuWbrCreate_PMFSRR,     /* parent, mfs and round-robin */
24372 +       AuWbrCreate_PMFSRRV,    /* plus seconds */
24373 +
24374 +       AuWbrCreate_Def = AuWbrCreate_TDP
24375 +};
24376 +
24377 +enum {
24378 +       AuWbrCopyup_TDP,        /* top down parent */
24379 +       AuWbrCopyup_BUP,        /* bottom up parent */
24380 +       AuWbrCopyup_BU,         /* bottom up */
24381 +
24382 +       AuWbrCopyup_Def = AuWbrCopyup_TDP
24383 +};
24384 +
24385 +/* ---------------------------------------------------------------------- */
24386 +
24387 +struct au_opt_add {
24388 +       aufs_bindex_t   bindex;
24389 +       char            *pathname;
24390 +       int             perm;
24391 +       struct path     path;
24392 +};
24393 +
24394 +struct au_opt_del {
24395 +       char            *pathname;
24396 +       struct path     h_path;
24397 +};
24398 +
24399 +struct au_opt_mod {
24400 +       char            *path;
24401 +       int             perm;
24402 +       struct dentry   *h_root;
24403 +};
24404 +
24405 +struct au_opt_xino {
24406 +       char            *path;
24407 +       struct file     *file;
24408 +};
24409 +
24410 +struct au_opt_xino_itrunc {
24411 +       aufs_bindex_t   bindex;
24412 +};
24413 +
24414 +struct au_opt_wbr_create {
24415 +       int                     wbr_create;
24416 +       int                     mfs_second;
24417 +       unsigned long long      mfsrr_watermark;
24418 +};
24419 +
24420 +struct au_opt {
24421 +       int type;
24422 +       union {
24423 +               struct au_opt_xino      xino;
24424 +               struct au_opt_xino_itrunc xino_itrunc;
24425 +               struct au_opt_add       add;
24426 +               struct au_opt_del       del;
24427 +               struct au_opt_mod       mod;
24428 +               int                     dirwh;
24429 +               int                     rdcache;
24430 +               unsigned int            rdblk;
24431 +               unsigned int            rdhash;
24432 +               int                     udba;
24433 +               struct au_opt_wbr_create wbr_create;
24434 +               int                     wbr_copyup;
24435 +               unsigned int            fhsm_second;
24436 +       };
24437 +};
24438 +
24439 +/* opts flags */
24440 +#define AuOpts_REMOUNT         1
24441 +#define AuOpts_REFRESH         (1 << 1)
24442 +#define AuOpts_TRUNC_XIB       (1 << 2)
24443 +#define AuOpts_REFRESH_DYAOP   (1 << 3)
24444 +#define au_ftest_opts(flags, name)     ((flags) & AuOpts_##name)
24445 +#define au_fset_opts(flags, name) \
24446 +       do { (flags) |= AuOpts_##name; } while (0)
24447 +#define au_fclr_opts(flags, name) \
24448 +       do { (flags) &= ~AuOpts_##name; } while (0)
24449 +
24450 +struct au_opts {
24451 +       struct au_opt   *opt;
24452 +       int             max_opt;
24453 +
24454 +       unsigned int    given_udba;
24455 +       unsigned int    flags;
24456 +       unsigned long   sb_flags;
24457 +};
24458 +
24459 +/* ---------------------------------------------------------------------- */
24460 +
24461 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm);
24462 +const char *au_optstr_udba(int udba);
24463 +const char *au_optstr_wbr_copyup(int wbr_copyup);
24464 +const char *au_optstr_wbr_create(int wbr_create);
24465 +
24466 +void au_opts_free(struct au_opts *opts);
24467 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts);
24468 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
24469 +                  unsigned int pending);
24470 +int au_opts_mount(struct super_block *sb, struct au_opts *opts);
24471 +int au_opts_remount(struct super_block *sb, struct au_opts *opts);
24472 +
24473 +unsigned int au_opt_udba(struct super_block *sb);
24474 +
24475 +/* ---------------------------------------------------------------------- */
24476 +
24477 +#endif /* __KERNEL__ */
24478 +#endif /* __AUFS_OPTS_H__ */
24479 diff -urN /usr/share/empty/fs/aufs/plink.c linux/fs/aufs/plink.c
24480 --- /usr/share/empty/fs/aufs/plink.c    1970-01-01 01:00:00.000000000 +0100
24481 +++ linux/fs/aufs/plink.c       2015-01-25 13:00:38.634380408 +0100
24482 @@ -0,0 +1,532 @@
24483 +/*
24484 + * Copyright (C) 2005-2014 Junjiro R. Okajima
24485 + *
24486 + * This program, aufs is free software; you can redistribute it and/or modify
24487 + * it under the terms of the GNU General Public License as published by
24488 + * the Free Software Foundation; either version 2 of the License, or
24489 + * (at your option) any later version.
24490 + *
24491 + * This program is distributed in the hope that it will be useful,
24492 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24493 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24494 + * GNU General Public License for more details.
24495 + *
24496 + * You should have received a copy of the GNU General Public License
24497 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24498 + */
24499 +
24500 +/*
24501 + * pseudo-link
24502 + */
24503 +
24504 +#include "aufs.h"
24505 +
24506 +/*
24507 + * the pseudo-link maintenance mode.
24508 + * during a user process maintains the pseudo-links,
24509 + * prohibit adding a new plink and branch manipulation.
24510 + *
24511 + * Flags
24512 + * NOPLM:
24513 + *     For entry functions which will handle plink, and i_mutex is already held
24514 + *     in VFS.
24515 + *     They cannot wait and should return an error at once.
24516 + *     Callers has to check the error.
24517 + * NOPLMW:
24518 + *     For entry functions which will handle plink, but i_mutex is not held
24519 + *     in VFS.
24520 + *     They can wait the plink maintenance mode to finish.
24521 + *
24522 + * They behave like F_SETLK and F_SETLKW.
24523 + * If the caller never handle plink, then both flags are unnecessary.
24524 + */
24525 +
24526 +int au_plink_maint(struct super_block *sb, int flags)
24527 +{
24528 +       int err;
24529 +       pid_t pid, ppid;
24530 +       struct au_sbinfo *sbi;
24531 +
24532 +       SiMustAnyLock(sb);
24533 +
24534 +       err = 0;
24535 +       if (!au_opt_test(au_mntflags(sb), PLINK))
24536 +               goto out;
24537 +
24538 +       sbi = au_sbi(sb);
24539 +       pid = sbi->si_plink_maint_pid;
24540 +       if (!pid || pid == current->pid)
24541 +               goto out;
24542 +
24543 +       /* todo: it highly depends upon /sbin/mount.aufs */
24544 +       rcu_read_lock();
24545 +       ppid = task_pid_vnr(rcu_dereference(current->real_parent));
24546 +       rcu_read_unlock();
24547 +       if (pid == ppid)
24548 +               goto out;
24549 +
24550 +       if (au_ftest_lock(flags, NOPLMW)) {
24551 +               /* if there is no i_mutex lock in VFS, we don't need to wait */
24552 +               /* AuDebugOn(!lockdep_depth(current)); */
24553 +               while (sbi->si_plink_maint_pid) {
24554 +                       si_read_unlock(sb);
24555 +                       /* gave up wake_up_bit() */
24556 +                       wait_event(sbi->si_plink_wq, !sbi->si_plink_maint_pid);
24557 +
24558 +                       if (au_ftest_lock(flags, FLUSH))
24559 +                               au_nwt_flush(&sbi->si_nowait);
24560 +                       si_noflush_read_lock(sb);
24561 +               }
24562 +       } else if (au_ftest_lock(flags, NOPLM)) {
24563 +               AuDbg("ppid %d, pid %d\n", ppid, pid);
24564 +               err = -EAGAIN;
24565 +       }
24566 +
24567 +out:
24568 +       return err;
24569 +}
24570 +
24571 +void au_plink_maint_leave(struct au_sbinfo *sbinfo)
24572 +{
24573 +       spin_lock(&sbinfo->si_plink_maint_lock);
24574 +       sbinfo->si_plink_maint_pid = 0;
24575 +       spin_unlock(&sbinfo->si_plink_maint_lock);
24576 +       wake_up_all(&sbinfo->si_plink_wq);
24577 +}
24578 +
24579 +int au_plink_maint_enter(struct super_block *sb)
24580 +{
24581 +       int err;
24582 +       struct au_sbinfo *sbinfo;
24583 +
24584 +       err = 0;
24585 +       sbinfo = au_sbi(sb);
24586 +       /* make sure i am the only one in this fs */
24587 +       si_write_lock(sb, AuLock_FLUSH);
24588 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
24589 +               spin_lock(&sbinfo->si_plink_maint_lock);
24590 +               if (!sbinfo->si_plink_maint_pid)
24591 +                       sbinfo->si_plink_maint_pid = current->pid;
24592 +               else
24593 +                       err = -EBUSY;
24594 +               spin_unlock(&sbinfo->si_plink_maint_lock);
24595 +       }
24596 +       si_write_unlock(sb);
24597 +
24598 +       return err;
24599 +}
24600 +
24601 +/* ---------------------------------------------------------------------- */
24602 +
24603 +#ifdef CONFIG_AUFS_DEBUG
24604 +void au_plink_list(struct super_block *sb)
24605 +{
24606 +       int i;
24607 +       struct au_sbinfo *sbinfo;
24608 +       struct hlist_head *plink_hlist;
24609 +       struct pseudo_link *plink;
24610 +
24611 +       SiMustAnyLock(sb);
24612 +
24613 +       sbinfo = au_sbi(sb);
24614 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
24615 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
24616 +
24617 +       for (i = 0; i < AuPlink_NHASH; i++) {
24618 +               plink_hlist = &sbinfo->si_plink[i].head;
24619 +               rcu_read_lock();
24620 +               hlist_for_each_entry_rcu(plink, plink_hlist, hlist)
24621 +                       AuDbg("%lu\n", plink->inode->i_ino);
24622 +               rcu_read_unlock();
24623 +       }
24624 +}
24625 +#endif
24626 +
24627 +/* is the inode pseudo-linked? */
24628 +int au_plink_test(struct inode *inode)
24629 +{
24630 +       int found, i;
24631 +       struct au_sbinfo *sbinfo;
24632 +       struct hlist_head *plink_hlist;
24633 +       struct pseudo_link *plink;
24634 +
24635 +       sbinfo = au_sbi(inode->i_sb);
24636 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
24637 +       AuDebugOn(!au_opt_test(au_mntflags(inode->i_sb), PLINK));
24638 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
24639 +
24640 +       found = 0;
24641 +       i = au_plink_hash(inode->i_ino);
24642 +       plink_hlist = &sbinfo->si_plink[i].head;
24643 +       rcu_read_lock();
24644 +       hlist_for_each_entry_rcu(plink, plink_hlist, hlist)
24645 +               if (plink->inode == inode) {
24646 +                       found = 1;
24647 +                       break;
24648 +               }
24649 +       rcu_read_unlock();
24650 +       return found;
24651 +}
24652 +
24653 +/* ---------------------------------------------------------------------- */
24654 +
24655 +/*
24656 + * generate a name for plink.
24657 + * the file will be stored under AUFS_WH_PLINKDIR.
24658 + */
24659 +/* 20 is max digits length of ulong 64 */
24660 +#define PLINK_NAME_LEN ((20 + 1) * 2)
24661 +
24662 +static int plink_name(char *name, int len, struct inode *inode,
24663 +                     aufs_bindex_t bindex)
24664 +{
24665 +       int rlen;
24666 +       struct inode *h_inode;
24667 +
24668 +       h_inode = au_h_iptr(inode, bindex);
24669 +       rlen = snprintf(name, len, "%lu.%lu", inode->i_ino, h_inode->i_ino);
24670 +       return rlen;
24671 +}
24672 +
24673 +struct au_do_plink_lkup_args {
24674 +       struct dentry **errp;
24675 +       struct qstr *tgtname;
24676 +       struct dentry *h_parent;
24677 +       struct au_branch *br;
24678 +};
24679 +
24680 +static struct dentry *au_do_plink_lkup(struct qstr *tgtname,
24681 +                                      struct dentry *h_parent,
24682 +                                      struct au_branch *br)
24683 +{
24684 +       struct dentry *h_dentry;
24685 +       struct mutex *h_mtx;
24686 +
24687 +       h_mtx = &h_parent->d_inode->i_mutex;
24688 +       mutex_lock_nested(h_mtx, AuLsc_I_CHILD2);
24689 +       h_dentry = vfsub_lkup_one(tgtname, h_parent);
24690 +       mutex_unlock(h_mtx);
24691 +       return h_dentry;
24692 +}
24693 +
24694 +static void au_call_do_plink_lkup(void *args)
24695 +{
24696 +       struct au_do_plink_lkup_args *a = args;
24697 +       *a->errp = au_do_plink_lkup(a->tgtname, a->h_parent, a->br);
24698 +}
24699 +
24700 +/* lookup the plink-ed @inode under the branch at @bindex */
24701 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex)
24702 +{
24703 +       struct dentry *h_dentry, *h_parent;
24704 +       struct au_branch *br;
24705 +       struct inode *h_dir;
24706 +       int wkq_err;
24707 +       char a[PLINK_NAME_LEN];
24708 +       struct qstr tgtname = QSTR_INIT(a, 0);
24709 +
24710 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
24711 +
24712 +       br = au_sbr(inode->i_sb, bindex);
24713 +       h_parent = br->br_wbr->wbr_plink;
24714 +       h_dir = h_parent->d_inode;
24715 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
24716 +
24717 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
24718 +               struct au_do_plink_lkup_args args = {
24719 +                       .errp           = &h_dentry,
24720 +                       .tgtname        = &tgtname,
24721 +                       .h_parent       = h_parent,
24722 +                       .br             = br
24723 +               };
24724 +
24725 +               wkq_err = au_wkq_wait(au_call_do_plink_lkup, &args);
24726 +               if (unlikely(wkq_err))
24727 +                       h_dentry = ERR_PTR(wkq_err);
24728 +       } else
24729 +               h_dentry = au_do_plink_lkup(&tgtname, h_parent, br);
24730 +
24731 +       return h_dentry;
24732 +}
24733 +
24734 +/* create a pseudo-link */
24735 +static int do_whplink(struct qstr *tgt, struct dentry *h_parent,
24736 +                     struct dentry *h_dentry, struct au_branch *br)
24737 +{
24738 +       int err;
24739 +       struct path h_path = {
24740 +               .mnt = au_br_mnt(br)
24741 +       };
24742 +       struct inode *h_dir, *delegated;
24743 +
24744 +       h_dir = h_parent->d_inode;
24745 +       mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_CHILD2);
24746 +again:
24747 +       h_path.dentry = vfsub_lkup_one(tgt, h_parent);
24748 +       err = PTR_ERR(h_path.dentry);
24749 +       if (IS_ERR(h_path.dentry))
24750 +               goto out;
24751 +
24752 +       err = 0;
24753 +       /* wh.plink dir is not monitored */
24754 +       /* todo: is it really safe? */
24755 +       if (h_path.dentry->d_inode
24756 +           && h_path.dentry->d_inode != h_dentry->d_inode) {
24757 +               delegated = NULL;
24758 +               err = vfsub_unlink(h_dir, &h_path, &delegated, /*force*/0);
24759 +               if (unlikely(err == -EWOULDBLOCK)) {
24760 +                       pr_warn("cannot retry for NFSv4 delegation"
24761 +                               " for an internal unlink\n");
24762 +                       iput(delegated);
24763 +               }
24764 +               dput(h_path.dentry);
24765 +               h_path.dentry = NULL;
24766 +               if (!err)
24767 +                       goto again;
24768 +       }
24769 +       if (!err && !h_path.dentry->d_inode) {
24770 +               delegated = NULL;
24771 +               err = vfsub_link(h_dentry, h_dir, &h_path, &delegated);
24772 +               if (unlikely(err == -EWOULDBLOCK)) {
24773 +                       pr_warn("cannot retry for NFSv4 delegation"
24774 +                               " for an internal link\n");
24775 +                       iput(delegated);
24776 +               }
24777 +       }
24778 +       dput(h_path.dentry);
24779 +
24780 +out:
24781 +       mutex_unlock(&h_dir->i_mutex);
24782 +       return err;
24783 +}
24784 +
24785 +struct do_whplink_args {
24786 +       int *errp;
24787 +       struct qstr *tgt;
24788 +       struct dentry *h_parent;
24789 +       struct dentry *h_dentry;
24790 +       struct au_branch *br;
24791 +};
24792 +
24793 +static void call_do_whplink(void *args)
24794 +{
24795 +       struct do_whplink_args *a = args;
24796 +       *a->errp = do_whplink(a->tgt, a->h_parent, a->h_dentry, a->br);
24797 +}
24798 +
24799 +static int whplink(struct dentry *h_dentry, struct inode *inode,
24800 +                  aufs_bindex_t bindex, struct au_branch *br)
24801 +{
24802 +       int err, wkq_err;
24803 +       struct au_wbr *wbr;
24804 +       struct dentry *h_parent;
24805 +       struct inode *h_dir;
24806 +       char a[PLINK_NAME_LEN];
24807 +       struct qstr tgtname = QSTR_INIT(a, 0);
24808 +
24809 +       wbr = au_sbr(inode->i_sb, bindex)->br_wbr;
24810 +       h_parent = wbr->wbr_plink;
24811 +       h_dir = h_parent->d_inode;
24812 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
24813 +
24814 +       /* always superio. */
24815 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
24816 +               struct do_whplink_args args = {
24817 +                       .errp           = &err,
24818 +                       .tgt            = &tgtname,
24819 +                       .h_parent       = h_parent,
24820 +                       .h_dentry       = h_dentry,
24821 +                       .br             = br
24822 +               };
24823 +               wkq_err = au_wkq_wait(call_do_whplink, &args);
24824 +               if (unlikely(wkq_err))
24825 +                       err = wkq_err;
24826 +       } else
24827 +               err = do_whplink(&tgtname, h_parent, h_dentry, br);
24828 +
24829 +       return err;
24830 +}
24831 +
24832 +/* free a single plink */
24833 +static void do_put_plink(struct pseudo_link *plink, int do_del)
24834 +{
24835 +       if (do_del)
24836 +               hlist_del(&plink->hlist);
24837 +       iput(plink->inode);
24838 +       kfree(plink);
24839 +}
24840 +
24841 +static void do_put_plink_rcu(struct rcu_head *rcu)
24842 +{
24843 +       struct pseudo_link *plink;
24844 +
24845 +       plink = container_of(rcu, struct pseudo_link, rcu);
24846 +       iput(plink->inode);
24847 +       kfree(plink);
24848 +}
24849 +
24850 +/*
24851 + * create a new pseudo-link for @h_dentry on @bindex.
24852 + * the linked inode is held in aufs @inode.
24853 + */
24854 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
24855 +                    struct dentry *h_dentry)
24856 +{
24857 +       struct super_block *sb;
24858 +       struct au_sbinfo *sbinfo;
24859 +       struct hlist_head *plink_hlist;
24860 +       struct pseudo_link *plink, *tmp;
24861 +       struct au_sphlhead *sphl;
24862 +       int found, err, cnt, i;
24863 +
24864 +       sb = inode->i_sb;
24865 +       sbinfo = au_sbi(sb);
24866 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
24867 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
24868 +
24869 +       found = au_plink_test(inode);
24870 +       if (found)
24871 +               return;
24872 +
24873 +       i = au_plink_hash(inode->i_ino);
24874 +       sphl = sbinfo->si_plink + i;
24875 +       plink_hlist = &sphl->head;
24876 +       tmp = kmalloc(sizeof(*plink), GFP_NOFS);
24877 +       if (tmp)
24878 +               tmp->inode = au_igrab(inode);
24879 +       else {
24880 +               err = -ENOMEM;
24881 +               goto out;
24882 +       }
24883 +
24884 +       spin_lock(&sphl->spin);
24885 +       hlist_for_each_entry(plink, plink_hlist, hlist) {
24886 +               if (plink->inode == inode) {
24887 +                       found = 1;
24888 +                       break;
24889 +               }
24890 +       }
24891 +       if (!found)
24892 +               hlist_add_head_rcu(&tmp->hlist, plink_hlist);
24893 +       spin_unlock(&sphl->spin);
24894 +       if (!found) {
24895 +               cnt = au_sphl_count(sphl);
24896 +#define msg "unexpectedly unblanced or too many pseudo-links"
24897 +               if (cnt > AUFS_PLINK_WARN)
24898 +                       AuWarn1(msg ", %d\n", cnt);
24899 +#undef msg
24900 +               err = whplink(h_dentry, inode, bindex, au_sbr(sb, bindex));
24901 +       } else {
24902 +               do_put_plink(tmp, 0);
24903 +               return;
24904 +       }
24905 +
24906 +out:
24907 +       if (unlikely(err)) {
24908 +               pr_warn("err %d, damaged pseudo link.\n", err);
24909 +               if (tmp) {
24910 +                       au_sphl_del_rcu(&tmp->hlist, sphl);
24911 +                       call_rcu(&tmp->rcu, do_put_plink_rcu);
24912 +               }
24913 +       }
24914 +}
24915 +
24916 +/* free all plinks */
24917 +void au_plink_put(struct super_block *sb, int verbose)
24918 +{
24919 +       int i, warned;
24920 +       struct au_sbinfo *sbinfo;
24921 +       struct hlist_head *plink_hlist;
24922 +       struct hlist_node *tmp;
24923 +       struct pseudo_link *plink;
24924 +
24925 +       SiMustWriteLock(sb);
24926 +
24927 +       sbinfo = au_sbi(sb);
24928 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
24929 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
24930 +
24931 +       /* no spin_lock since sbinfo is write-locked */
24932 +       warned = 0;
24933 +       for (i = 0; i < AuPlink_NHASH; i++) {
24934 +               plink_hlist = &sbinfo->si_plink[i].head;
24935 +               if (!warned && verbose && !hlist_empty(plink_hlist)) {
24936 +                       pr_warn("pseudo-link is not flushed");
24937 +                       warned = 1;
24938 +               }
24939 +               hlist_for_each_entry_safe(plink, tmp, plink_hlist, hlist)
24940 +                       do_put_plink(plink, 0);
24941 +               INIT_HLIST_HEAD(plink_hlist);
24942 +       }
24943 +}
24944 +
24945 +void au_plink_clean(struct super_block *sb, int verbose)
24946 +{
24947 +       struct dentry *root;
24948 +
24949 +       root = sb->s_root;
24950 +       aufs_write_lock(root);
24951 +       if (au_opt_test(au_mntflags(sb), PLINK))
24952 +               au_plink_put(sb, verbose);
24953 +       aufs_write_unlock(root);
24954 +}
24955 +
24956 +static int au_plink_do_half_refresh(struct inode *inode, aufs_bindex_t br_id)
24957 +{
24958 +       int do_put;
24959 +       aufs_bindex_t bstart, bend, bindex;
24960 +
24961 +       do_put = 0;
24962 +       bstart = au_ibstart(inode);
24963 +       bend = au_ibend(inode);
24964 +       if (bstart >= 0) {
24965 +               for (bindex = bstart; bindex <= bend; bindex++) {
24966 +                       if (!au_h_iptr(inode, bindex)
24967 +                           || au_ii_br_id(inode, bindex) != br_id)
24968 +                               continue;
24969 +                       au_set_h_iptr(inode, bindex, NULL, 0);
24970 +                       do_put = 1;
24971 +                       break;
24972 +               }
24973 +               if (do_put)
24974 +                       for (bindex = bstart; bindex <= bend; bindex++)
24975 +                               if (au_h_iptr(inode, bindex)) {
24976 +                                       do_put = 0;
24977 +                                       break;
24978 +                               }
24979 +       } else
24980 +               do_put = 1;
24981 +
24982 +       return do_put;
24983 +}
24984 +
24985 +/* free the plinks on a branch specified by @br_id */
24986 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id)
24987 +{
24988 +       struct au_sbinfo *sbinfo;
24989 +       struct hlist_head *plink_hlist;
24990 +       struct hlist_node *tmp;
24991 +       struct pseudo_link *plink;
24992 +       struct inode *inode;
24993 +       int i, do_put;
24994 +
24995 +       SiMustWriteLock(sb);
24996 +
24997 +       sbinfo = au_sbi(sb);
24998 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
24999 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25000 +
25001 +       /* no spin_lock since sbinfo is write-locked */
25002 +       for (i = 0; i < AuPlink_NHASH; i++) {
25003 +               plink_hlist = &sbinfo->si_plink[i].head;
25004 +               hlist_for_each_entry_safe(plink, tmp, plink_hlist, hlist) {
25005 +                       inode = au_igrab(plink->inode);
25006 +                       ii_write_lock_child(inode);
25007 +                       do_put = au_plink_do_half_refresh(inode, br_id);
25008 +                       if (do_put)
25009 +                               do_put_plink(plink, 1);
25010 +                       ii_write_unlock(inode);
25011 +                       iput(inode);
25012 +               }
25013 +       }
25014 +}
25015 diff -urN /usr/share/empty/fs/aufs/poll.c linux/fs/aufs/poll.c
25016 --- /usr/share/empty/fs/aufs/poll.c     1970-01-01 01:00:00.000000000 +0100
25017 +++ linux/fs/aufs/poll.c        2015-01-25 13:00:38.634380408 +0100
25018 @@ -0,0 +1,55 @@
25019 +/*
25020 + * Copyright (C) 2005-2014 Junjiro R. Okajima
25021 + *
25022 + * This program, aufs is free software; you can redistribute it and/or modify
25023 + * it under the terms of the GNU General Public License as published by
25024 + * the Free Software Foundation; either version 2 of the License, or
25025 + * (at your option) any later version.
25026 + *
25027 + * This program is distributed in the hope that it will be useful,
25028 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25029 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25030 + * GNU General Public License for more details.
25031 + *
25032 + * You should have received a copy of the GNU General Public License
25033 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25034 + */
25035 +
25036 +/*
25037 + * poll operation
25038 + * There is only one filesystem which implements ->poll operation, currently.
25039 + */
25040 +
25041 +#include "aufs.h"
25042 +
25043 +unsigned int aufs_poll(struct file *file, poll_table *wait)
25044 +{
25045 +       unsigned int mask;
25046 +       int err;
25047 +       struct file *h_file;
25048 +       struct dentry *dentry;
25049 +       struct super_block *sb;
25050 +
25051 +       /* We should pretend an error happened. */
25052 +       mask = POLLERR /* | POLLIN | POLLOUT */;
25053 +       dentry = file->f_dentry;
25054 +       sb = dentry->d_sb;
25055 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
25056 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
25057 +       if (unlikely(err))
25058 +               goto out;
25059 +
25060 +       /* it is not an error if h_file has no operation */
25061 +       mask = DEFAULT_POLLMASK;
25062 +       h_file = au_hf_top(file);
25063 +       if (h_file->f_op->poll)
25064 +               mask = h_file->f_op->poll(h_file, wait);
25065 +
25066 +       di_read_unlock(dentry, AuLock_IR);
25067 +       fi_read_unlock(file);
25068 +
25069 +out:
25070 +       si_read_unlock(sb);
25071 +       AuTraceErr((int)mask);
25072 +       return mask;
25073 +}
25074 diff -urN /usr/share/empty/fs/aufs/posix_acl.c linux/fs/aufs/posix_acl.c
25075 --- /usr/share/empty/fs/aufs/posix_acl.c        1970-01-01 01:00:00.000000000 +0100
25076 +++ linux/fs/aufs/posix_acl.c   2015-01-25 13:00:38.634380408 +0100
25077 @@ -0,0 +1,99 @@
25078 +/*
25079 + * Copyright (C) 2014 Junjiro R. Okajima
25080 + *
25081 + * This program, aufs is free software; you can redistribute it and/or modify
25082 + * it under the terms of the GNU General Public License as published by
25083 + * the Free Software Foundation; either version 2 of the License, or
25084 + * (at your option) any later version.
25085 + *
25086 + * This program is distributed in the hope that it will be useful,
25087 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25088 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25089 + * GNU General Public License for more details.
25090 + *
25091 + * You should have received a copy of the GNU General Public License
25092 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25093 + */
25094 +
25095 +/*
25096 + * posix acl operations
25097 + */
25098 +
25099 +#include <linux/fs.h>
25100 +#include <linux/posix_acl.h>
25101 +#include "aufs.h"
25102 +
25103 +struct posix_acl *aufs_get_acl(struct inode *inode, int type)
25104 +{
25105 +       struct posix_acl *acl;
25106 +       int err;
25107 +       aufs_bindex_t bindex;
25108 +       struct inode *h_inode;
25109 +       struct super_block *sb;
25110 +
25111 +       acl = NULL;
25112 +       sb = inode->i_sb;
25113 +       si_read_lock(sb, AuLock_FLUSH);
25114 +       ii_read_lock_child(inode);
25115 +       if (!(sb->s_flags & MS_POSIXACL))
25116 +               goto out;
25117 +
25118 +       bindex = au_ibstart(inode);
25119 +       h_inode = au_h_iptr(inode, bindex);
25120 +       if (unlikely(!h_inode
25121 +                    || ((h_inode->i_mode & S_IFMT)
25122 +                        != (inode->i_mode & S_IFMT)))) {
25123 +               err = au_busy_or_stale();
25124 +               acl = ERR_PTR(err);
25125 +               goto out;
25126 +       }
25127 +
25128 +       /* always topmost only */
25129 +       acl = get_acl(h_inode, type);
25130 +
25131 +out:
25132 +       ii_read_unlock(inode);
25133 +       si_read_unlock(sb);
25134 +
25135 +       AuTraceErrPtr(acl);
25136 +       return acl;
25137 +}
25138 +
25139 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
25140 +{
25141 +       int err;
25142 +       ssize_t ssz;
25143 +       struct dentry *dentry;
25144 +       struct au_srxattr arg = {
25145 +               .type = AU_ACL_SET,
25146 +               .u.acl_set = {
25147 +                       .acl    = acl,
25148 +                       .type   = type
25149 +               },
25150 +       };
25151 +
25152 +       mutex_lock(&inode->i_mutex);
25153 +       if (inode->i_ino == AUFS_ROOT_INO)
25154 +               dentry = dget(inode->i_sb->s_root);
25155 +       else {
25156 +               dentry = d_find_alias(inode);
25157 +               if (!dentry)
25158 +                       dentry = d_find_any_alias(inode);
25159 +               if (!dentry) {
25160 +                       pr_warn("cannot handle this inode, "
25161 +                               "please report to aufs-users ML\n");
25162 +                       err = -ENOENT;
25163 +                       goto out;
25164 +               }
25165 +       }
25166 +
25167 +       ssz = au_srxattr(dentry, &arg);
25168 +       dput(dentry);
25169 +       err = ssz;
25170 +       if (ssz >= 0)
25171 +               err = 0;
25172 +
25173 +out:
25174 +       mutex_unlock(&inode->i_mutex);
25175 +       return err;
25176 +}
25177 diff -urN /usr/share/empty/fs/aufs/procfs.c linux/fs/aufs/procfs.c
25178 --- /usr/share/empty/fs/aufs/procfs.c   1970-01-01 01:00:00.000000000 +0100
25179 +++ linux/fs/aufs/procfs.c      2015-01-25 13:00:38.634380408 +0100
25180 @@ -0,0 +1,169 @@
25181 +/*
25182 + * Copyright (C) 2010-2014 Junjiro R. Okajima
25183 + *
25184 + * This program, aufs is free software; you can redistribute it and/or modify
25185 + * it under the terms of the GNU General Public License as published by
25186 + * the Free Software Foundation; either version 2 of the License, or
25187 + * (at your option) any later version.
25188 + *
25189 + * This program is distributed in the hope that it will be useful,
25190 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25191 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25192 + * GNU General Public License for more details.
25193 + *
25194 + * You should have received a copy of the GNU General Public License
25195 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25196 + */
25197 +
25198 +/*
25199 + * procfs interfaces
25200 + */
25201 +
25202 +#include <linux/proc_fs.h>
25203 +#include "aufs.h"
25204 +
25205 +static int au_procfs_plm_release(struct inode *inode, struct file *file)
25206 +{
25207 +       struct au_sbinfo *sbinfo;
25208 +
25209 +       sbinfo = file->private_data;
25210 +       if (sbinfo) {
25211 +               au_plink_maint_leave(sbinfo);
25212 +               kobject_put(&sbinfo->si_kobj);
25213 +       }
25214 +
25215 +       return 0;
25216 +}
25217 +
25218 +static void au_procfs_plm_write_clean(struct file *file)
25219 +{
25220 +       struct au_sbinfo *sbinfo;
25221 +
25222 +       sbinfo = file->private_data;
25223 +       if (sbinfo)
25224 +               au_plink_clean(sbinfo->si_sb, /*verbose*/0);
25225 +}
25226 +
25227 +static int au_procfs_plm_write_si(struct file *file, unsigned long id)
25228 +{
25229 +       int err;
25230 +       struct super_block *sb;
25231 +       struct au_sbinfo *sbinfo;
25232 +
25233 +       err = -EBUSY;
25234 +       if (unlikely(file->private_data))
25235 +               goto out;
25236 +
25237 +       sb = NULL;
25238 +       /* don't use au_sbilist_lock() here */
25239 +       spin_lock(&au_sbilist.spin);
25240 +       list_for_each_entry(sbinfo, &au_sbilist.head, si_list)
25241 +               if (id == sysaufs_si_id(sbinfo)) {
25242 +                       kobject_get(&sbinfo->si_kobj);
25243 +                       sb = sbinfo->si_sb;
25244 +                       break;
25245 +               }
25246 +       spin_unlock(&au_sbilist.spin);
25247 +
25248 +       err = -EINVAL;
25249 +       if (unlikely(!sb))
25250 +               goto out;
25251 +
25252 +       err = au_plink_maint_enter(sb);
25253 +       if (!err)
25254 +               /* keep kobject_get() */
25255 +               file->private_data = sbinfo;
25256 +       else
25257 +               kobject_put(&sbinfo->si_kobj);
25258 +out:
25259 +       return err;
25260 +}
25261 +
25262 +/*
25263 + * Accept a valid "si=xxxx" only.
25264 + * Once it is accepted successfully, accept "clean" too.
25265 + */
25266 +static ssize_t au_procfs_plm_write(struct file *file, const char __user *ubuf,
25267 +                                  size_t count, loff_t *ppos)
25268 +{
25269 +       ssize_t err;
25270 +       unsigned long id;
25271 +       /* last newline is allowed */
25272 +       char buf[3 + sizeof(unsigned long) * 2 + 1];
25273 +
25274 +       err = -EACCES;
25275 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
25276 +               goto out;
25277 +
25278 +       err = -EINVAL;
25279 +       if (unlikely(count > sizeof(buf)))
25280 +               goto out;
25281 +
25282 +       err = copy_from_user(buf, ubuf, count);
25283 +       if (unlikely(err)) {
25284 +               err = -EFAULT;
25285 +               goto out;
25286 +       }
25287 +       buf[count] = 0;
25288 +
25289 +       err = -EINVAL;
25290 +       if (!strcmp("clean", buf)) {
25291 +               au_procfs_plm_write_clean(file);
25292 +               goto out_success;
25293 +       } else if (unlikely(strncmp("si=", buf, 3)))
25294 +               goto out;
25295 +
25296 +       err = kstrtoul(buf + 3, 16, &id);
25297 +       if (unlikely(err))
25298 +               goto out;
25299 +
25300 +       err = au_procfs_plm_write_si(file, id);
25301 +       if (unlikely(err))
25302 +               goto out;
25303 +
25304 +out_success:
25305 +       err = count; /* success */
25306 +out:
25307 +       return err;
25308 +}
25309 +
25310 +static const struct file_operations au_procfs_plm_fop = {
25311 +       .write          = au_procfs_plm_write,
25312 +       .release        = au_procfs_plm_release,
25313 +       .owner          = THIS_MODULE
25314 +};
25315 +
25316 +/* ---------------------------------------------------------------------- */
25317 +
25318 +static struct proc_dir_entry *au_procfs_dir;
25319 +
25320 +void au_procfs_fin(void)
25321 +{
25322 +       remove_proc_entry(AUFS_PLINK_MAINT_NAME, au_procfs_dir);
25323 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
25324 +}
25325 +
25326 +int __init au_procfs_init(void)
25327 +{
25328 +       int err;
25329 +       struct proc_dir_entry *entry;
25330 +
25331 +       err = -ENOMEM;
25332 +       au_procfs_dir = proc_mkdir(AUFS_PLINK_MAINT_DIR, NULL);
25333 +       if (unlikely(!au_procfs_dir))
25334 +               goto out;
25335 +
25336 +       entry = proc_create(AUFS_PLINK_MAINT_NAME, S_IFREG | S_IWUSR,
25337 +                           au_procfs_dir, &au_procfs_plm_fop);
25338 +       if (unlikely(!entry))
25339 +               goto out_dir;
25340 +
25341 +       err = 0;
25342 +       goto out; /* success */
25343 +
25344 +
25345 +out_dir:
25346 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
25347 +out:
25348 +       return err;
25349 +}
25350 diff -urN /usr/share/empty/fs/aufs/rdu.c linux/fs/aufs/rdu.c
25351 --- /usr/share/empty/fs/aufs/rdu.c      1970-01-01 01:00:00.000000000 +0100
25352 +++ linux/fs/aufs/rdu.c 2015-01-25 13:00:38.634380408 +0100
25353 @@ -0,0 +1,388 @@
25354 +/*
25355 + * Copyright (C) 2005-2014 Junjiro R. Okajima
25356 + *
25357 + * This program, aufs is free software; you can redistribute it and/or modify
25358 + * it under the terms of the GNU General Public License as published by
25359 + * the Free Software Foundation; either version 2 of the License, or
25360 + * (at your option) any later version.
25361 + *
25362 + * This program is distributed in the hope that it will be useful,
25363 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25364 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25365 + * GNU General Public License for more details.
25366 + *
25367 + * You should have received a copy of the GNU General Public License
25368 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25369 + */
25370 +
25371 +/*
25372 + * readdir in userspace.
25373 + */
25374 +
25375 +#include <linux/compat.h>
25376 +#include <linux/fs_stack.h>
25377 +#include <linux/security.h>
25378 +#include "aufs.h"
25379 +
25380 +/* bits for struct aufs_rdu.flags */
25381 +#define        AuRdu_CALLED    1
25382 +#define        AuRdu_CONT      (1 << 1)
25383 +#define        AuRdu_FULL      (1 << 2)
25384 +#define au_ftest_rdu(flags, name)      ((flags) & AuRdu_##name)
25385 +#define au_fset_rdu(flags, name) \
25386 +       do { (flags) |= AuRdu_##name; } while (0)
25387 +#define au_fclr_rdu(flags, name) \
25388 +       do { (flags) &= ~AuRdu_##name; } while (0)
25389 +
25390 +struct au_rdu_arg {
25391 +       struct dir_context              ctx;
25392 +       struct aufs_rdu                 *rdu;
25393 +       union au_rdu_ent_ul             ent;
25394 +       unsigned long                   end;
25395 +
25396 +       struct super_block              *sb;
25397 +       int                             err;
25398 +};
25399 +
25400 +static int au_rdu_fill(struct dir_context *ctx, const char *name, int nlen,
25401 +                      loff_t offset, u64 h_ino, unsigned int d_type)
25402 +{
25403 +       int err, len;
25404 +       struct au_rdu_arg *arg = container_of(ctx, struct au_rdu_arg, ctx);
25405 +       struct aufs_rdu *rdu = arg->rdu;
25406 +       struct au_rdu_ent ent;
25407 +
25408 +       err = 0;
25409 +       arg->err = 0;
25410 +       au_fset_rdu(rdu->cookie.flags, CALLED);
25411 +       len = au_rdu_len(nlen);
25412 +       if (arg->ent.ul + len  < arg->end) {
25413 +               ent.ino = h_ino;
25414 +               ent.bindex = rdu->cookie.bindex;
25415 +               ent.type = d_type;
25416 +               ent.nlen = nlen;
25417 +               if (unlikely(nlen > AUFS_MAX_NAMELEN))
25418 +                       ent.type = DT_UNKNOWN;
25419 +
25420 +               /* unnecessary to support mmap_sem since this is a dir */
25421 +               err = -EFAULT;
25422 +               if (copy_to_user(arg->ent.e, &ent, sizeof(ent)))
25423 +                       goto out;
25424 +               if (copy_to_user(arg->ent.e->name, name, nlen))
25425 +                       goto out;
25426 +               /* the terminating NULL */
25427 +               if (__put_user(0, arg->ent.e->name + nlen))
25428 +                       goto out;
25429 +               err = 0;
25430 +               /* AuDbg("%p, %.*s\n", arg->ent.p, nlen, name); */
25431 +               arg->ent.ul += len;
25432 +               rdu->rent++;
25433 +       } else {
25434 +               err = -EFAULT;
25435 +               au_fset_rdu(rdu->cookie.flags, FULL);
25436 +               rdu->full = 1;
25437 +               rdu->tail = arg->ent;
25438 +       }
25439 +
25440 +out:
25441 +       /* AuTraceErr(err); */
25442 +       return err;
25443 +}
25444 +
25445 +static int au_rdu_do(struct file *h_file, struct au_rdu_arg *arg)
25446 +{
25447 +       int err;
25448 +       loff_t offset;
25449 +       struct au_rdu_cookie *cookie = &arg->rdu->cookie;
25450 +
25451 +       /* we don't have to care (FMODE_32BITHASH | FMODE_64BITHASH) for ext4 */
25452 +       offset = vfsub_llseek(h_file, cookie->h_pos, SEEK_SET);
25453 +       err = offset;
25454 +       if (unlikely(offset != cookie->h_pos))
25455 +               goto out;
25456 +
25457 +       err = 0;
25458 +       do {
25459 +               arg->err = 0;
25460 +               au_fclr_rdu(cookie->flags, CALLED);
25461 +               /* smp_mb(); */
25462 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
25463 +               if (err >= 0)
25464 +                       err = arg->err;
25465 +       } while (!err
25466 +                && au_ftest_rdu(cookie->flags, CALLED)
25467 +                && !au_ftest_rdu(cookie->flags, FULL));
25468 +       cookie->h_pos = h_file->f_pos;
25469 +
25470 +out:
25471 +       AuTraceErr(err);
25472 +       return err;
25473 +}
25474 +
25475 +static int au_rdu(struct file *file, struct aufs_rdu *rdu)
25476 +{
25477 +       int err;
25478 +       aufs_bindex_t bend;
25479 +       struct au_rdu_arg arg = {
25480 +               .ctx = {
25481 +                       .actor = au_diractor(au_rdu_fill)
25482 +               }
25483 +       };
25484 +       struct dentry *dentry;
25485 +       struct inode *inode;
25486 +       struct file *h_file;
25487 +       struct au_rdu_cookie *cookie = &rdu->cookie;
25488 +
25489 +       err = !access_ok(VERIFY_WRITE, rdu->ent.e, rdu->sz);
25490 +       if (unlikely(err)) {
25491 +               err = -EFAULT;
25492 +               AuTraceErr(err);
25493 +               goto out;
25494 +       }
25495 +       rdu->rent = 0;
25496 +       rdu->tail = rdu->ent;
25497 +       rdu->full = 0;
25498 +       arg.rdu = rdu;
25499 +       arg.ent = rdu->ent;
25500 +       arg.end = arg.ent.ul;
25501 +       arg.end += rdu->sz;
25502 +
25503 +       err = -ENOTDIR;
25504 +       if (unlikely(!file->f_op->iterate))
25505 +               goto out;
25506 +
25507 +       err = security_file_permission(file, MAY_READ);
25508 +       AuTraceErr(err);
25509 +       if (unlikely(err))
25510 +               goto out;
25511 +
25512 +       dentry = file->f_dentry;
25513 +       inode = dentry->d_inode;
25514 +#if 1
25515 +       mutex_lock(&inode->i_mutex);
25516 +#else
25517 +       err = mutex_lock_killable(&inode->i_mutex);
25518 +       AuTraceErr(err);
25519 +       if (unlikely(err))
25520 +               goto out;
25521 +#endif
25522 +
25523 +       arg.sb = inode->i_sb;
25524 +       err = si_read_lock(arg.sb, AuLock_FLUSH | AuLock_NOPLM);
25525 +       if (unlikely(err))
25526 +               goto out_mtx;
25527 +       err = au_alive_dir(dentry);
25528 +       if (unlikely(err))
25529 +               goto out_si;
25530 +       /* todo: reval? */
25531 +       fi_read_lock(file);
25532 +
25533 +       err = -EAGAIN;
25534 +       if (unlikely(au_ftest_rdu(cookie->flags, CONT)
25535 +                    && cookie->generation != au_figen(file)))
25536 +               goto out_unlock;
25537 +
25538 +       err = 0;
25539 +       if (!rdu->blk) {
25540 +               rdu->blk = au_sbi(arg.sb)->si_rdblk;
25541 +               if (!rdu->blk)
25542 +                       rdu->blk = au_dir_size(file, /*dentry*/NULL);
25543 +       }
25544 +       bend = au_fbstart(file);
25545 +       if (cookie->bindex < bend)
25546 +               cookie->bindex = bend;
25547 +       bend = au_fbend_dir(file);
25548 +       /* AuDbg("b%d, b%d\n", cookie->bindex, bend); */
25549 +       for (; !err && cookie->bindex <= bend;
25550 +            cookie->bindex++, cookie->h_pos = 0) {
25551 +               h_file = au_hf_dir(file, cookie->bindex);
25552 +               if (!h_file)
25553 +                       continue;
25554 +
25555 +               au_fclr_rdu(cookie->flags, FULL);
25556 +               err = au_rdu_do(h_file, &arg);
25557 +               AuTraceErr(err);
25558 +               if (unlikely(au_ftest_rdu(cookie->flags, FULL) || err))
25559 +                       break;
25560 +       }
25561 +       AuDbg("rent %llu\n", rdu->rent);
25562 +
25563 +       if (!err && !au_ftest_rdu(cookie->flags, CONT)) {
25564 +               rdu->shwh = !!au_opt_test(au_sbi(arg.sb)->si_mntflags, SHWH);
25565 +               au_fset_rdu(cookie->flags, CONT);
25566 +               cookie->generation = au_figen(file);
25567 +       }
25568 +
25569 +       ii_read_lock_child(inode);
25570 +       fsstack_copy_attr_atime(inode, au_h_iptr(inode, au_ibstart(inode)));
25571 +       ii_read_unlock(inode);
25572 +
25573 +out_unlock:
25574 +       fi_read_unlock(file);
25575 +out_si:
25576 +       si_read_unlock(arg.sb);
25577 +out_mtx:
25578 +       mutex_unlock(&inode->i_mutex);
25579 +out:
25580 +       AuTraceErr(err);
25581 +       return err;
25582 +}
25583 +
25584 +static int au_rdu_ino(struct file *file, struct aufs_rdu *rdu)
25585 +{
25586 +       int err;
25587 +       ino_t ino;
25588 +       unsigned long long nent;
25589 +       union au_rdu_ent_ul *u;
25590 +       struct au_rdu_ent ent;
25591 +       struct super_block *sb;
25592 +
25593 +       err = 0;
25594 +       nent = rdu->nent;
25595 +       u = &rdu->ent;
25596 +       sb = file->f_dentry->d_sb;
25597 +       si_read_lock(sb, AuLock_FLUSH);
25598 +       while (nent-- > 0) {
25599 +               /* unnecessary to support mmap_sem since this is a dir */
25600 +               err = copy_from_user(&ent, u->e, sizeof(ent));
25601 +               if (!err)
25602 +                       err = !access_ok(VERIFY_WRITE, &u->e->ino, sizeof(ino));
25603 +               if (unlikely(err)) {
25604 +                       err = -EFAULT;
25605 +                       AuTraceErr(err);
25606 +                       break;
25607 +               }
25608 +
25609 +               /* AuDbg("b%d, i%llu\n", ent.bindex, ent.ino); */
25610 +               if (!ent.wh)
25611 +                       err = au_ino(sb, ent.bindex, ent.ino, ent.type, &ino);
25612 +               else
25613 +                       err = au_wh_ino(sb, ent.bindex, ent.ino, ent.type,
25614 +                                       &ino);
25615 +               if (unlikely(err)) {
25616 +                       AuTraceErr(err);
25617 +                       break;
25618 +               }
25619 +
25620 +               err = __put_user(ino, &u->e->ino);
25621 +               if (unlikely(err)) {
25622 +                       err = -EFAULT;
25623 +                       AuTraceErr(err);
25624 +                       break;
25625 +               }
25626 +               u->ul += au_rdu_len(ent.nlen);
25627 +       }
25628 +       si_read_unlock(sb);
25629 +
25630 +       return err;
25631 +}
25632 +
25633 +/* ---------------------------------------------------------------------- */
25634 +
25635 +static int au_rdu_verify(struct aufs_rdu *rdu)
25636 +{
25637 +       AuDbg("rdu{%llu, %p, %u | %u | %llu, %u, %u | "
25638 +             "%llu, b%d, 0x%x, g%u}\n",
25639 +             rdu->sz, rdu->ent.e, rdu->verify[AufsCtlRduV_SZ],
25640 +             rdu->blk,
25641 +             rdu->rent, rdu->shwh, rdu->full,
25642 +             rdu->cookie.h_pos, rdu->cookie.bindex, rdu->cookie.flags,
25643 +             rdu->cookie.generation);
25644 +
25645 +       if (rdu->verify[AufsCtlRduV_SZ] == sizeof(*rdu))
25646 +               return 0;
25647 +
25648 +       AuDbg("%u:%u\n",
25649 +             rdu->verify[AufsCtlRduV_SZ], (unsigned int)sizeof(*rdu));
25650 +       return -EINVAL;
25651 +}
25652 +
25653 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
25654 +{
25655 +       long err, e;
25656 +       struct aufs_rdu rdu;
25657 +       void __user *p = (void __user *)arg;
25658 +
25659 +       err = copy_from_user(&rdu, p, sizeof(rdu));
25660 +       if (unlikely(err)) {
25661 +               err = -EFAULT;
25662 +               AuTraceErr(err);
25663 +               goto out;
25664 +       }
25665 +       err = au_rdu_verify(&rdu);
25666 +       if (unlikely(err))
25667 +               goto out;
25668 +
25669 +       switch (cmd) {
25670 +       case AUFS_CTL_RDU:
25671 +               err = au_rdu(file, &rdu);
25672 +               if (unlikely(err))
25673 +                       break;
25674 +
25675 +               e = copy_to_user(p, &rdu, sizeof(rdu));
25676 +               if (unlikely(e)) {
25677 +                       err = -EFAULT;
25678 +                       AuTraceErr(err);
25679 +               }
25680 +               break;
25681 +       case AUFS_CTL_RDU_INO:
25682 +               err = au_rdu_ino(file, &rdu);
25683 +               break;
25684 +
25685 +       default:
25686 +               /* err = -ENOTTY; */
25687 +               err = -EINVAL;
25688 +       }
25689 +
25690 +out:
25691 +       AuTraceErr(err);
25692 +       return err;
25693 +}
25694 +
25695 +#ifdef CONFIG_COMPAT
25696 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
25697 +{
25698 +       long err, e;
25699 +       struct aufs_rdu rdu;
25700 +       void __user *p = compat_ptr(arg);
25701 +
25702 +       /* todo: get_user()? */
25703 +       err = copy_from_user(&rdu, p, sizeof(rdu));
25704 +       if (unlikely(err)) {
25705 +               err = -EFAULT;
25706 +               AuTraceErr(err);
25707 +               goto out;
25708 +       }
25709 +       rdu.ent.e = compat_ptr(rdu.ent.ul);
25710 +       err = au_rdu_verify(&rdu);
25711 +       if (unlikely(err))
25712 +               goto out;
25713 +
25714 +       switch (cmd) {
25715 +       case AUFS_CTL_RDU:
25716 +               err = au_rdu(file, &rdu);
25717 +               if (unlikely(err))
25718 +                       break;
25719 +
25720 +               rdu.ent.ul = ptr_to_compat(rdu.ent.e);
25721 +               rdu.tail.ul = ptr_to_compat(rdu.tail.e);
25722 +               e = copy_to_user(p, &rdu, sizeof(rdu));
25723 +               if (unlikely(e)) {
25724 +                       err = -EFAULT;
25725 +                       AuTraceErr(err);
25726 +               }
25727 +               break;
25728 +       case AUFS_CTL_RDU_INO:
25729 +               err = au_rdu_ino(file, &rdu);
25730 +               break;
25731 +
25732 +       default:
25733 +               /* err = -ENOTTY; */
25734 +               err = -EINVAL;
25735 +       }
25736 +
25737 +out:
25738 +       AuTraceErr(err);
25739 +       return err;
25740 +}
25741 +#endif
25742 diff -urN /usr/share/empty/fs/aufs/rwsem.h linux/fs/aufs/rwsem.h
25743 --- /usr/share/empty/fs/aufs/rwsem.h    1970-01-01 01:00:00.000000000 +0100
25744 +++ linux/fs/aufs/rwsem.h       2015-01-25 13:00:38.634380408 +0100
25745 @@ -0,0 +1,191 @@
25746 +/*
25747 + * Copyright (C) 2005-2014 Junjiro R. Okajima
25748 + *
25749 + * This program, aufs is free software; you can redistribute it and/or modify
25750 + * it under the terms of the GNU General Public License as published by
25751 + * the Free Software Foundation; either version 2 of the License, or
25752 + * (at your option) any later version.
25753 + *
25754 + * This program is distributed in the hope that it will be useful,
25755 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25756 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25757 + * GNU General Public License for more details.
25758 + *
25759 + * You should have received a copy of the GNU General Public License
25760 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25761 + */
25762 +
25763 +/*
25764 + * simple read-write semaphore wrappers
25765 + */
25766 +
25767 +#ifndef __AUFS_RWSEM_H__
25768 +#define __AUFS_RWSEM_H__
25769 +
25770 +#ifdef __KERNEL__
25771 +
25772 +#include "debug.h"
25773 +
25774 +struct au_rwsem {
25775 +       struct rw_semaphore     rwsem;
25776 +#ifdef CONFIG_AUFS_DEBUG
25777 +       /* just for debugging, not almighty counter */
25778 +       atomic_t                rcnt, wcnt;
25779 +#endif
25780 +};
25781 +
25782 +#ifdef CONFIG_AUFS_DEBUG
25783 +#define AuDbgCntInit(rw) do { \
25784 +       atomic_set(&(rw)->rcnt, 0); \
25785 +       atomic_set(&(rw)->wcnt, 0); \
25786 +       smp_mb(); /* atomic set */ \
25787 +} while (0)
25788 +
25789 +#define AuDbgRcntInc(rw)       atomic_inc(&(rw)->rcnt)
25790 +#define AuDbgRcntDec(rw)       WARN_ON(atomic_dec_return(&(rw)->rcnt) < 0)
25791 +#define AuDbgWcntInc(rw)       atomic_inc(&(rw)->wcnt)
25792 +#define AuDbgWcntDec(rw)       WARN_ON(atomic_dec_return(&(rw)->wcnt) < 0)
25793 +#else
25794 +#define AuDbgCntInit(rw)       do {} while (0)
25795 +#define AuDbgRcntInc(rw)       do {} while (0)
25796 +#define AuDbgRcntDec(rw)       do {} while (0)
25797 +#define AuDbgWcntInc(rw)       do {} while (0)
25798 +#define AuDbgWcntDec(rw)       do {} while (0)
25799 +#endif /* CONFIG_AUFS_DEBUG */
25800 +
25801 +/* to debug easier, do not make them inlined functions */
25802 +#define AuRwMustNoWaiters(rw)  AuDebugOn(!list_empty(&(rw)->rwsem.wait_list))
25803 +/* rwsem_is_locked() is unusable */
25804 +#define AuRwMustReadLock(rw)   AuDebugOn(atomic_read(&(rw)->rcnt) <= 0)
25805 +#define AuRwMustWriteLock(rw)  AuDebugOn(atomic_read(&(rw)->wcnt) <= 0)
25806 +#define AuRwMustAnyLock(rw)    AuDebugOn(atomic_read(&(rw)->rcnt) <= 0 \
25807 +                                       && atomic_read(&(rw)->wcnt) <= 0)
25808 +#define AuRwDestroy(rw)                AuDebugOn(atomic_read(&(rw)->rcnt) \
25809 +                                       || atomic_read(&(rw)->wcnt))
25810 +
25811 +#define au_rw_class(rw, key)   lockdep_set_class(&(rw)->rwsem, key)
25812 +
25813 +static inline void au_rw_init(struct au_rwsem *rw)
25814 +{
25815 +       AuDbgCntInit(rw);
25816 +       init_rwsem(&rw->rwsem);
25817 +}
25818 +
25819 +static inline void au_rw_init_wlock(struct au_rwsem *rw)
25820 +{
25821 +       au_rw_init(rw);
25822 +       down_write(&rw->rwsem);
25823 +       AuDbgWcntInc(rw);
25824 +}
25825 +
25826 +static inline void au_rw_init_wlock_nested(struct au_rwsem *rw,
25827 +                                          unsigned int lsc)
25828 +{
25829 +       au_rw_init(rw);
25830 +       down_write_nested(&rw->rwsem, lsc);
25831 +       AuDbgWcntInc(rw);
25832 +}
25833 +
25834 +static inline void au_rw_read_lock(struct au_rwsem *rw)
25835 +{
25836 +       down_read(&rw->rwsem);
25837 +       AuDbgRcntInc(rw);
25838 +}
25839 +
25840 +static inline void au_rw_read_lock_nested(struct au_rwsem *rw, unsigned int lsc)
25841 +{
25842 +       down_read_nested(&rw->rwsem, lsc);
25843 +       AuDbgRcntInc(rw);
25844 +}
25845 +
25846 +static inline void au_rw_read_unlock(struct au_rwsem *rw)
25847 +{
25848 +       AuRwMustReadLock(rw);
25849 +       AuDbgRcntDec(rw);
25850 +       up_read(&rw->rwsem);
25851 +}
25852 +
25853 +static inline void au_rw_dgrade_lock(struct au_rwsem *rw)
25854 +{
25855 +       AuRwMustWriteLock(rw);
25856 +       AuDbgRcntInc(rw);
25857 +       AuDbgWcntDec(rw);
25858 +       downgrade_write(&rw->rwsem);
25859 +}
25860 +
25861 +static inline void au_rw_write_lock(struct au_rwsem *rw)
25862 +{
25863 +       down_write(&rw->rwsem);
25864 +       AuDbgWcntInc(rw);
25865 +}
25866 +
25867 +static inline void au_rw_write_lock_nested(struct au_rwsem *rw,
25868 +                                          unsigned int lsc)
25869 +{
25870 +       down_write_nested(&rw->rwsem, lsc);
25871 +       AuDbgWcntInc(rw);
25872 +}
25873 +
25874 +static inline void au_rw_write_unlock(struct au_rwsem *rw)
25875 +{
25876 +       AuRwMustWriteLock(rw);
25877 +       AuDbgWcntDec(rw);
25878 +       up_write(&rw->rwsem);
25879 +}
25880 +
25881 +/* why is not _nested version defined */
25882 +static inline int au_rw_read_trylock(struct au_rwsem *rw)
25883 +{
25884 +       int ret;
25885 +
25886 +       ret = down_read_trylock(&rw->rwsem);
25887 +       if (ret)
25888 +               AuDbgRcntInc(rw);
25889 +       return ret;
25890 +}
25891 +
25892 +static inline int au_rw_write_trylock(struct au_rwsem *rw)
25893 +{
25894 +       int ret;
25895 +
25896 +       ret = down_write_trylock(&rw->rwsem);
25897 +       if (ret)
25898 +               AuDbgWcntInc(rw);
25899 +       return ret;
25900 +}
25901 +
25902 +#undef AuDbgCntInit
25903 +#undef AuDbgRcntInc
25904 +#undef AuDbgRcntDec
25905 +#undef AuDbgWcntInc
25906 +#undef AuDbgWcntDec
25907 +
25908 +#define AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
25909 +static inline void prefix##_read_lock(param) \
25910 +{ au_rw_read_lock(rwsem); } \
25911 +static inline void prefix##_write_lock(param) \
25912 +{ au_rw_write_lock(rwsem); } \
25913 +static inline int prefix##_read_trylock(param) \
25914 +{ return au_rw_read_trylock(rwsem); } \
25915 +static inline int prefix##_write_trylock(param) \
25916 +{ return au_rw_write_trylock(rwsem); }
25917 +/* why is not _nested version defined */
25918 +/* static inline void prefix##_read_trylock_nested(param, lsc)
25919 +{ au_rw_read_trylock_nested(rwsem, lsc)); }
25920 +static inline void prefix##_write_trylock_nestd(param, lsc)
25921 +{ au_rw_write_trylock_nested(rwsem, lsc); } */
25922 +
25923 +#define AuSimpleUnlockRwsemFuncs(prefix, param, rwsem) \
25924 +static inline void prefix##_read_unlock(param) \
25925 +{ au_rw_read_unlock(rwsem); } \
25926 +static inline void prefix##_write_unlock(param) \
25927 +{ au_rw_write_unlock(rwsem); } \
25928 +static inline void prefix##_downgrade_lock(param) \
25929 +{ au_rw_dgrade_lock(rwsem); }
25930 +
25931 +#define AuSimpleRwsemFuncs(prefix, param, rwsem) \
25932 +       AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
25933 +       AuSimpleUnlockRwsemFuncs(prefix, param, rwsem)
25934 +
25935 +#endif /* __KERNEL__ */
25936 +#endif /* __AUFS_RWSEM_H__ */
25937 diff -urN /usr/share/empty/fs/aufs/sbinfo.c linux/fs/aufs/sbinfo.c
25938 --- /usr/share/empty/fs/aufs/sbinfo.c   1970-01-01 01:00:00.000000000 +0100
25939 +++ linux/fs/aufs/sbinfo.c      2015-01-25 13:00:38.634380408 +0100
25940 @@ -0,0 +1,353 @@
25941 +/*
25942 + * Copyright (C) 2005-2014 Junjiro R. Okajima
25943 + *
25944 + * This program, aufs is free software; you can redistribute it and/or modify
25945 + * it under the terms of the GNU General Public License as published by
25946 + * the Free Software Foundation; either version 2 of the License, or
25947 + * (at your option) any later version.
25948 + *
25949 + * This program is distributed in the hope that it will be useful,
25950 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25951 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25952 + * GNU General Public License for more details.
25953 + *
25954 + * You should have received a copy of the GNU General Public License
25955 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25956 + */
25957 +
25958 +/*
25959 + * superblock private data
25960 + */
25961 +
25962 +#include "aufs.h"
25963 +
25964 +/*
25965 + * they are necessary regardless sysfs is disabled.
25966 + */
25967 +void au_si_free(struct kobject *kobj)
25968 +{
25969 +       int i;
25970 +       struct au_sbinfo *sbinfo;
25971 +       char *locked __maybe_unused; /* debug only */
25972 +
25973 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
25974 +       for (i = 0; i < AuPlink_NHASH; i++)
25975 +               AuDebugOn(!hlist_empty(&sbinfo->si_plink[i].head));
25976 +       AuDebugOn(atomic_read(&sbinfo->si_nowait.nw_len));
25977 +
25978 +       au_rw_write_lock(&sbinfo->si_rwsem);
25979 +       au_br_free(sbinfo);
25980 +       au_rw_write_unlock(&sbinfo->si_rwsem);
25981 +
25982 +       AuDebugOn(radix_tree_gang_lookup
25983 +                 (&sbinfo->au_si_pid.tree, (void **)&locked,
25984 +                  /*first_index*/PID_MAX_DEFAULT - 1,
25985 +                  /*max_items*/sizeof(locked)/sizeof(*locked)));
25986 +
25987 +       kfree(sbinfo->si_branch);
25988 +       kfree(sbinfo->au_si_pid.bitmap);
25989 +       mutex_destroy(&sbinfo->si_xib_mtx);
25990 +       AuRwDestroy(&sbinfo->si_rwsem);
25991 +
25992 +       kfree(sbinfo);
25993 +}
25994 +
25995 +int au_si_alloc(struct super_block *sb)
25996 +{
25997 +       int err, i;
25998 +       struct au_sbinfo *sbinfo;
25999 +       static struct lock_class_key aufs_si;
26000 +
26001 +       err = -ENOMEM;
26002 +       sbinfo = kzalloc(sizeof(*sbinfo), GFP_NOFS);
26003 +       if (unlikely(!sbinfo))
26004 +               goto out;
26005 +
26006 +       BUILD_BUG_ON(sizeof(unsigned long) !=
26007 +                    sizeof(*sbinfo->au_si_pid.bitmap));
26008 +       sbinfo->au_si_pid.bitmap = kcalloc(BITS_TO_LONGS(PID_MAX_DEFAULT),
26009 +                                       sizeof(*sbinfo->au_si_pid.bitmap),
26010 +                                       GFP_NOFS);
26011 +       if (unlikely(!sbinfo->au_si_pid.bitmap))
26012 +               goto out_sbinfo;
26013 +
26014 +       /* will be reallocated separately */
26015 +       sbinfo->si_branch = kzalloc(sizeof(*sbinfo->si_branch), GFP_NOFS);
26016 +       if (unlikely(!sbinfo->si_branch))
26017 +               goto out_pidmap;
26018 +
26019 +       err = sysaufs_si_init(sbinfo);
26020 +       if (unlikely(err))
26021 +               goto out_br;
26022 +
26023 +       au_nwt_init(&sbinfo->si_nowait);
26024 +       au_rw_init_wlock(&sbinfo->si_rwsem);
26025 +       au_rw_class(&sbinfo->si_rwsem, &aufs_si);
26026 +       spin_lock_init(&sbinfo->au_si_pid.tree_lock);
26027 +       INIT_RADIX_TREE(&sbinfo->au_si_pid.tree, GFP_ATOMIC | __GFP_NOFAIL);
26028 +
26029 +       atomic_long_set(&sbinfo->si_ninodes, 0);
26030 +       atomic_long_set(&sbinfo->si_nfiles, 0);
26031 +
26032 +       sbinfo->si_bend = -1;
26033 +       sbinfo->si_last_br_id = AUFS_BRANCH_MAX / 2;
26034 +
26035 +       sbinfo->si_wbr_copyup = AuWbrCopyup_Def;
26036 +       sbinfo->si_wbr_create = AuWbrCreate_Def;
26037 +       sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + sbinfo->si_wbr_copyup;
26038 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + sbinfo->si_wbr_create;
26039 +
26040 +       au_fhsm_init(sbinfo);
26041 +
26042 +       sbinfo->si_mntflags = au_opts_plink(AuOpt_Def);
26043 +
26044 +       sbinfo->si_xino_jiffy = jiffies;
26045 +       sbinfo->si_xino_expire
26046 +               = msecs_to_jiffies(AUFS_XINO_DEF_SEC * MSEC_PER_SEC);
26047 +       mutex_init(&sbinfo->si_xib_mtx);
26048 +       sbinfo->si_xino_brid = -1;
26049 +       /* leave si_xib_last_pindex and si_xib_next_bit */
26050 +
26051 +       sbinfo->si_rdcache = msecs_to_jiffies(AUFS_RDCACHE_DEF * MSEC_PER_SEC);
26052 +       sbinfo->si_rdblk = AUFS_RDBLK_DEF;
26053 +       sbinfo->si_rdhash = AUFS_RDHASH_DEF;
26054 +       sbinfo->si_dirwh = AUFS_DIRWH_DEF;
26055 +
26056 +       for (i = 0; i < AuPlink_NHASH; i++)
26057 +               au_sphl_init(sbinfo->si_plink + i);
26058 +       init_waitqueue_head(&sbinfo->si_plink_wq);
26059 +       spin_lock_init(&sbinfo->si_plink_maint_lock);
26060 +
26061 +       au_sphl_init(&sbinfo->si_files);
26062 +
26063 +       /* leave other members for sysaufs and si_mnt. */
26064 +       sbinfo->si_sb = sb;
26065 +       sb->s_fs_info = sbinfo;
26066 +       si_pid_set(sb);
26067 +       au_debug_sbinfo_init(sbinfo);
26068 +       return 0; /* success */
26069 +
26070 +out_br:
26071 +       kfree(sbinfo->si_branch);
26072 +out_pidmap:
26073 +       kfree(sbinfo->au_si_pid.bitmap);
26074 +out_sbinfo:
26075 +       kfree(sbinfo);
26076 +out:
26077 +       return err;
26078 +}
26079 +
26080 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr)
26081 +{
26082 +       int err, sz;
26083 +       struct au_branch **brp;
26084 +
26085 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
26086 +
26087 +       err = -ENOMEM;
26088 +       sz = sizeof(*brp) * (sbinfo->si_bend + 1);
26089 +       if (unlikely(!sz))
26090 +               sz = sizeof(*brp);
26091 +       brp = au_kzrealloc(sbinfo->si_branch, sz, sizeof(*brp) * nbr, GFP_NOFS);
26092 +       if (brp) {
26093 +               sbinfo->si_branch = brp;
26094 +               err = 0;
26095 +       }
26096 +
26097 +       return err;
26098 +}
26099 +
26100 +/* ---------------------------------------------------------------------- */
26101 +
26102 +unsigned int au_sigen_inc(struct super_block *sb)
26103 +{
26104 +       unsigned int gen;
26105 +
26106 +       SiMustWriteLock(sb);
26107 +
26108 +       gen = ++au_sbi(sb)->si_generation;
26109 +       au_update_digen(sb->s_root);
26110 +       au_update_iigen(sb->s_root->d_inode, /*half*/0);
26111 +       sb->s_root->d_inode->i_version++;
26112 +       return gen;
26113 +}
26114 +
26115 +aufs_bindex_t au_new_br_id(struct super_block *sb)
26116 +{
26117 +       aufs_bindex_t br_id;
26118 +       int i;
26119 +       struct au_sbinfo *sbinfo;
26120 +
26121 +       SiMustWriteLock(sb);
26122 +
26123 +       sbinfo = au_sbi(sb);
26124 +       for (i = 0; i <= AUFS_BRANCH_MAX; i++) {
26125 +               br_id = ++sbinfo->si_last_br_id;
26126 +               AuDebugOn(br_id < 0);
26127 +               if (br_id && au_br_index(sb, br_id) < 0)
26128 +                       return br_id;
26129 +       }
26130 +
26131 +       return -1;
26132 +}
26133 +
26134 +/* ---------------------------------------------------------------------- */
26135 +
26136 +/* it is ok that new 'nwt' tasks are appended while we are sleeping */
26137 +int si_read_lock(struct super_block *sb, int flags)
26138 +{
26139 +       int err;
26140 +
26141 +       err = 0;
26142 +       if (au_ftest_lock(flags, FLUSH))
26143 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26144 +
26145 +       si_noflush_read_lock(sb);
26146 +       err = au_plink_maint(sb, flags);
26147 +       if (unlikely(err))
26148 +               si_read_unlock(sb);
26149 +
26150 +       return err;
26151 +}
26152 +
26153 +int si_write_lock(struct super_block *sb, int flags)
26154 +{
26155 +       int err;
26156 +
26157 +       if (au_ftest_lock(flags, FLUSH))
26158 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26159 +
26160 +       si_noflush_write_lock(sb);
26161 +       err = au_plink_maint(sb, flags);
26162 +       if (unlikely(err))
26163 +               si_write_unlock(sb);
26164 +
26165 +       return err;
26166 +}
26167 +
26168 +/* dentry and super_block lock. call at entry point */
26169 +int aufs_read_lock(struct dentry *dentry, int flags)
26170 +{
26171 +       int err;
26172 +       struct super_block *sb;
26173 +
26174 +       sb = dentry->d_sb;
26175 +       err = si_read_lock(sb, flags);
26176 +       if (unlikely(err))
26177 +               goto out;
26178 +
26179 +       if (au_ftest_lock(flags, DW))
26180 +               di_write_lock_child(dentry);
26181 +       else
26182 +               di_read_lock_child(dentry, flags);
26183 +
26184 +       if (au_ftest_lock(flags, GEN)) {
26185 +               err = au_digen_test(dentry, au_sigen(sb));
26186 +               AuDebugOn(!err && au_dbrange_test(dentry));
26187 +               if (unlikely(err))
26188 +                       aufs_read_unlock(dentry, flags);
26189 +       }
26190 +
26191 +out:
26192 +       return err;
26193 +}
26194 +
26195 +void aufs_read_unlock(struct dentry *dentry, int flags)
26196 +{
26197 +       if (au_ftest_lock(flags, DW))
26198 +               di_write_unlock(dentry);
26199 +       else
26200 +               di_read_unlock(dentry, flags);
26201 +       si_read_unlock(dentry->d_sb);
26202 +}
26203 +
26204 +void aufs_write_lock(struct dentry *dentry)
26205 +{
26206 +       si_write_lock(dentry->d_sb, AuLock_FLUSH | AuLock_NOPLMW);
26207 +       di_write_lock_child(dentry);
26208 +}
26209 +
26210 +void aufs_write_unlock(struct dentry *dentry)
26211 +{
26212 +       di_write_unlock(dentry);
26213 +       si_write_unlock(dentry->d_sb);
26214 +}
26215 +
26216 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags)
26217 +{
26218 +       int err;
26219 +       unsigned int sigen;
26220 +       struct super_block *sb;
26221 +
26222 +       sb = d1->d_sb;
26223 +       err = si_read_lock(sb, flags);
26224 +       if (unlikely(err))
26225 +               goto out;
26226 +
26227 +       di_write_lock2_child(d1, d2, au_ftest_lock(flags, DIR));
26228 +
26229 +       if (au_ftest_lock(flags, GEN)) {
26230 +               sigen = au_sigen(sb);
26231 +               err = au_digen_test(d1, sigen);
26232 +               AuDebugOn(!err && au_dbrange_test(d1));
26233 +               if (!err) {
26234 +                       err = au_digen_test(d2, sigen);
26235 +                       AuDebugOn(!err && au_dbrange_test(d2));
26236 +               }
26237 +               if (unlikely(err))
26238 +                       aufs_read_and_write_unlock2(d1, d2);
26239 +       }
26240 +
26241 +out:
26242 +       return err;
26243 +}
26244 +
26245 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2)
26246 +{
26247 +       di_write_unlock2(d1, d2);
26248 +       si_read_unlock(d1->d_sb);
26249 +}
26250 +
26251 +/* ---------------------------------------------------------------------- */
26252 +
26253 +int si_pid_test_slow(struct super_block *sb)
26254 +{
26255 +       void *p;
26256 +
26257 +       rcu_read_lock();
26258 +       p = radix_tree_lookup(&au_sbi(sb)->au_si_pid.tree, current->pid);
26259 +       rcu_read_unlock();
26260 +
26261 +       return (long)!!p;
26262 +}
26263 +
26264 +void si_pid_set_slow(struct super_block *sb)
26265 +{
26266 +       int err;
26267 +       struct au_sbinfo *sbinfo;
26268 +
26269 +       AuDebugOn(si_pid_test_slow(sb));
26270 +
26271 +       sbinfo = au_sbi(sb);
26272 +       err = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
26273 +       AuDebugOn(err);
26274 +       spin_lock(&sbinfo->au_si_pid.tree_lock);
26275 +       err = radix_tree_insert(&sbinfo->au_si_pid.tree, current->pid,
26276 +                               /*any valid ptr*/sb);
26277 +       spin_unlock(&sbinfo->au_si_pid.tree_lock);
26278 +       AuDebugOn(err);
26279 +       radix_tree_preload_end();
26280 +}
26281 +
26282 +void si_pid_clr_slow(struct super_block *sb)
26283 +{
26284 +       void *p;
26285 +       struct au_sbinfo *sbinfo;
26286 +
26287 +       AuDebugOn(!si_pid_test_slow(sb));
26288 +
26289 +       sbinfo = au_sbi(sb);
26290 +       spin_lock(&sbinfo->au_si_pid.tree_lock);
26291 +       p = radix_tree_delete(&sbinfo->au_si_pid.tree, current->pid);
26292 +       spin_unlock(&sbinfo->au_si_pid.tree_lock);
26293 +}
26294 diff -urN /usr/share/empty/fs/aufs/spl.h linux/fs/aufs/spl.h
26295 --- /usr/share/empty/fs/aufs/spl.h      1970-01-01 01:00:00.000000000 +0100
26296 +++ linux/fs/aufs/spl.h 2015-01-25 13:00:38.634380408 +0100
26297 @@ -0,0 +1,111 @@
26298 +/*
26299 + * Copyright (C) 2005-2014 Junjiro R. Okajima
26300 + *
26301 + * This program, aufs is free software; you can redistribute it and/or modify
26302 + * it under the terms of the GNU General Public License as published by
26303 + * the Free Software Foundation; either version 2 of the License, or
26304 + * (at your option) any later version.
26305 + *
26306 + * This program is distributed in the hope that it will be useful,
26307 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26308 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26309 + * GNU General Public License for more details.
26310 + *
26311 + * You should have received a copy of the GNU General Public License
26312 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26313 + */
26314 +
26315 +/*
26316 + * simple list protected by a spinlock
26317 + */
26318 +
26319 +#ifndef __AUFS_SPL_H__
26320 +#define __AUFS_SPL_H__
26321 +
26322 +#ifdef __KERNEL__
26323 +
26324 +struct au_splhead {
26325 +       spinlock_t              spin;
26326 +       struct list_head        head;
26327 +};
26328 +
26329 +static inline void au_spl_init(struct au_splhead *spl)
26330 +{
26331 +       spin_lock_init(&spl->spin);
26332 +       INIT_LIST_HEAD(&spl->head);
26333 +}
26334 +
26335 +static inline void au_spl_add(struct list_head *list, struct au_splhead *spl)
26336 +{
26337 +       spin_lock(&spl->spin);
26338 +       list_add(list, &spl->head);
26339 +       spin_unlock(&spl->spin);
26340 +}
26341 +
26342 +static inline void au_spl_del(struct list_head *list, struct au_splhead *spl)
26343 +{
26344 +       spin_lock(&spl->spin);
26345 +       list_del(list);
26346 +       spin_unlock(&spl->spin);
26347 +}
26348 +
26349 +static inline void au_spl_del_rcu(struct list_head *list,
26350 +                                 struct au_splhead *spl)
26351 +{
26352 +       spin_lock(&spl->spin);
26353 +       list_del_rcu(list);
26354 +       spin_unlock(&spl->spin);
26355 +}
26356 +
26357 +/* ---------------------------------------------------------------------- */
26358 +
26359 +struct au_sphlhead {
26360 +       spinlock_t              spin;
26361 +       struct hlist_head       head;
26362 +};
26363 +
26364 +static inline void au_sphl_init(struct au_sphlhead *sphl)
26365 +{
26366 +       spin_lock_init(&sphl->spin);
26367 +       INIT_HLIST_HEAD(&sphl->head);
26368 +}
26369 +
26370 +static inline void au_sphl_add(struct hlist_node *hlist,
26371 +                              struct au_sphlhead *sphl)
26372 +{
26373 +       spin_lock(&sphl->spin);
26374 +       hlist_add_head(hlist, &sphl->head);
26375 +       spin_unlock(&sphl->spin);
26376 +}
26377 +
26378 +static inline void au_sphl_del(struct hlist_node *hlist,
26379 +                              struct au_sphlhead *sphl)
26380 +{
26381 +       spin_lock(&sphl->spin);
26382 +       hlist_del(hlist);
26383 +       spin_unlock(&sphl->spin);
26384 +}
26385 +
26386 +static inline void au_sphl_del_rcu(struct hlist_node *hlist,
26387 +                                  struct au_sphlhead *sphl)
26388 +{
26389 +       spin_lock(&sphl->spin);
26390 +       hlist_del_rcu(hlist);
26391 +       spin_unlock(&sphl->spin);
26392 +}
26393 +
26394 +static inline unsigned long au_sphl_count(struct au_sphlhead *sphl)
26395 +{
26396 +       unsigned long cnt;
26397 +       struct hlist_node *pos;
26398 +
26399 +       cnt = 0;
26400 +       spin_lock(&sphl->spin);
26401 +       hlist_for_each(pos, &sphl->head)
26402 +               cnt++;
26403 +       spin_unlock(&sphl->spin);
26404 +       return cnt;
26405 +}
26406 +
26407 +#endif /* __KERNEL__ */
26408 +#endif /* __AUFS_SPL_H__ */
26409 diff -urN /usr/share/empty/fs/aufs/super.c linux/fs/aufs/super.c
26410 --- /usr/share/empty/fs/aufs/super.c    1970-01-01 01:00:00.000000000 +0100
26411 +++ linux/fs/aufs/super.c       2015-01-25 13:00:38.634380408 +0100
26412 @@ -0,0 +1,1009 @@
26413 +/*
26414 + * Copyright (C) 2005-2014 Junjiro R. Okajima
26415 + *
26416 + * This program, aufs is free software; you can redistribute it and/or modify
26417 + * it under the terms of the GNU General Public License as published by
26418 + * the Free Software Foundation; either version 2 of the License, or
26419 + * (at your option) any later version.
26420 + *
26421 + * This program is distributed in the hope that it will be useful,
26422 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26423 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26424 + * GNU General Public License for more details.
26425 + *
26426 + * You should have received a copy of the GNU General Public License
26427 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26428 + */
26429 +
26430 +/*
26431 + * mount and super_block operations
26432 + */
26433 +
26434 +#include <linux/mm.h>
26435 +#include <linux/module.h>
26436 +#include <linux/seq_file.h>
26437 +#include <linux/statfs.h>
26438 +#include <linux/vmalloc.h>
26439 +#include <linux/writeback.h>
26440 +#include "aufs.h"
26441 +
26442 +/*
26443 + * super_operations
26444 + */
26445 +static struct inode *aufs_alloc_inode(struct super_block *sb __maybe_unused)
26446 +{
26447 +       struct au_icntnr *c;
26448 +
26449 +       c = au_cache_alloc_icntnr();
26450 +       if (c) {
26451 +               au_icntnr_init(c);
26452 +               c->vfs_inode.i_version = 1; /* sigen(sb); */
26453 +               c->iinfo.ii_hinode = NULL;
26454 +               return &c->vfs_inode;
26455 +       }
26456 +       return NULL;
26457 +}
26458 +
26459 +static void aufs_destroy_inode_cb(struct rcu_head *head)
26460 +{
26461 +       struct inode *inode = container_of(head, struct inode, i_rcu);
26462 +
26463 +       INIT_HLIST_HEAD(&inode->i_dentry);
26464 +       au_cache_free_icntnr(container_of(inode, struct au_icntnr, vfs_inode));
26465 +}
26466 +
26467 +static void aufs_destroy_inode(struct inode *inode)
26468 +{
26469 +       au_iinfo_fin(inode);
26470 +       call_rcu(&inode->i_rcu, aufs_destroy_inode_cb);
26471 +}
26472 +
26473 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino)
26474 +{
26475 +       struct inode *inode;
26476 +       int err;
26477 +
26478 +       inode = iget_locked(sb, ino);
26479 +       if (unlikely(!inode)) {
26480 +               inode = ERR_PTR(-ENOMEM);
26481 +               goto out;
26482 +       }
26483 +       if (!(inode->i_state & I_NEW))
26484 +               goto out;
26485 +
26486 +       err = au_xigen_new(inode);
26487 +       if (!err)
26488 +               err = au_iinfo_init(inode);
26489 +       if (!err)
26490 +               inode->i_version++;
26491 +       else {
26492 +               iget_failed(inode);
26493 +               inode = ERR_PTR(err);
26494 +       }
26495 +
26496 +out:
26497 +       /* never return NULL */
26498 +       AuDebugOn(!inode);
26499 +       AuTraceErrPtr(inode);
26500 +       return inode;
26501 +}
26502 +
26503 +/* lock free root dinfo */
26504 +static int au_show_brs(struct seq_file *seq, struct super_block *sb)
26505 +{
26506 +       int err;
26507 +       aufs_bindex_t bindex, bend;
26508 +       struct path path;
26509 +       struct au_hdentry *hdp;
26510 +       struct au_branch *br;
26511 +       au_br_perm_str_t perm;
26512 +
26513 +       err = 0;
26514 +       bend = au_sbend(sb);
26515 +       hdp = au_di(sb->s_root)->di_hdentry;
26516 +       for (bindex = 0; !err && bindex <= bend; bindex++) {
26517 +               br = au_sbr(sb, bindex);
26518 +               path.mnt = au_br_mnt(br);
26519 +               path.dentry = hdp[bindex].hd_dentry;
26520 +               err = au_seq_path(seq, &path);
26521 +               if (err > 0) {
26522 +                       au_optstr_br_perm(&perm, br->br_perm);
26523 +                       err = seq_printf(seq, "=%s", perm.a);
26524 +                       if (err == -1)
26525 +                               err = -E2BIG;
26526 +               }
26527 +               if (!err && bindex != bend)
26528 +                       err = seq_putc(seq, ':');
26529 +       }
26530 +
26531 +       return err;
26532 +}
26533 +
26534 +static void au_show_wbr_create(struct seq_file *m, int v,
26535 +                              struct au_sbinfo *sbinfo)
26536 +{
26537 +       const char *pat;
26538 +
26539 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
26540 +
26541 +       seq_puts(m, ",create=");
26542 +       pat = au_optstr_wbr_create(v);
26543 +       switch (v) {
26544 +       case AuWbrCreate_TDP:
26545 +       case AuWbrCreate_RR:
26546 +       case AuWbrCreate_MFS:
26547 +       case AuWbrCreate_PMFS:
26548 +               seq_puts(m, pat);
26549 +               break;
26550 +       case AuWbrCreate_MFSV:
26551 +               seq_printf(m, /*pat*/"mfs:%lu",
26552 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26553 +                          / MSEC_PER_SEC);
26554 +               break;
26555 +       case AuWbrCreate_PMFSV:
26556 +               seq_printf(m, /*pat*/"pmfs:%lu",
26557 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26558 +                          / MSEC_PER_SEC);
26559 +               break;
26560 +       case AuWbrCreate_MFSRR:
26561 +               seq_printf(m, /*pat*/"mfsrr:%llu",
26562 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
26563 +               break;
26564 +       case AuWbrCreate_MFSRRV:
26565 +               seq_printf(m, /*pat*/"mfsrr:%llu:%lu",
26566 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
26567 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26568 +                          / MSEC_PER_SEC);
26569 +               break;
26570 +       case AuWbrCreate_PMFSRR:
26571 +               seq_printf(m, /*pat*/"pmfsrr:%llu",
26572 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
26573 +               break;
26574 +       case AuWbrCreate_PMFSRRV:
26575 +               seq_printf(m, /*pat*/"pmfsrr:%llu:%lu",
26576 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
26577 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26578 +                          / MSEC_PER_SEC);
26579 +               break;
26580 +       }
26581 +}
26582 +
26583 +static int au_show_xino(struct seq_file *seq, struct super_block *sb)
26584 +{
26585 +#ifdef CONFIG_SYSFS
26586 +       return 0;
26587 +#else
26588 +       int err;
26589 +       const int len = sizeof(AUFS_XINO_FNAME) - 1;
26590 +       aufs_bindex_t bindex, brid;
26591 +       struct qstr *name;
26592 +       struct file *f;
26593 +       struct dentry *d, *h_root;
26594 +       struct au_hdentry *hdp;
26595 +
26596 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
26597 +
26598 +       err = 0;
26599 +       f = au_sbi(sb)->si_xib;
26600 +       if (!f)
26601 +               goto out;
26602 +
26603 +       /* stop printing the default xino path on the first writable branch */
26604 +       h_root = NULL;
26605 +       brid = au_xino_brid(sb);
26606 +       if (brid >= 0) {
26607 +               bindex = au_br_index(sb, brid);
26608 +               hdp = au_di(sb->s_root)->di_hdentry;
26609 +               h_root = hdp[0 + bindex].hd_dentry;
26610 +       }
26611 +       d = f->f_dentry;
26612 +       name = &d->d_name;
26613 +       /* safe ->d_parent because the file is unlinked */
26614 +       if (d->d_parent == h_root
26615 +           && name->len == len
26616 +           && !memcmp(name->name, AUFS_XINO_FNAME, len))
26617 +               goto out;
26618 +
26619 +       seq_puts(seq, ",xino=");
26620 +       err = au_xino_path(seq, f);
26621 +
26622 +out:
26623 +       return err;
26624 +#endif
26625 +}
26626 +
26627 +/* seq_file will re-call me in case of too long string */
26628 +static int aufs_show_options(struct seq_file *m, struct dentry *dentry)
26629 +{
26630 +       int err;
26631 +       unsigned int mnt_flags, v;
26632 +       struct super_block *sb;
26633 +       struct au_sbinfo *sbinfo;
26634 +
26635 +#define AuBool(name, str) do { \
26636 +       v = au_opt_test(mnt_flags, name); \
26637 +       if (v != au_opt_test(AuOpt_Def, name)) \
26638 +               seq_printf(m, ",%s" #str, v ? "" : "no"); \
26639 +} while (0)
26640 +
26641 +#define AuStr(name, str) do { \
26642 +       v = mnt_flags & AuOptMask_##name; \
26643 +       if (v != (AuOpt_Def & AuOptMask_##name)) \
26644 +               seq_printf(m, "," #str "=%s", au_optstr_##str(v)); \
26645 +} while (0)
26646 +
26647 +#define AuUInt(name, str, val) do { \
26648 +       if (val != AUFS_##name##_DEF) \
26649 +               seq_printf(m, "," #str "=%u", val); \
26650 +} while (0)
26651 +
26652 +       sb = dentry->d_sb;
26653 +       if (sb->s_flags & MS_POSIXACL)
26654 +               seq_puts(m, ",acl");
26655 +
26656 +       /* lock free root dinfo */
26657 +       si_noflush_read_lock(sb);
26658 +       sbinfo = au_sbi(sb);
26659 +       seq_printf(m, ",si=%lx", sysaufs_si_id(sbinfo));
26660 +
26661 +       mnt_flags = au_mntflags(sb);
26662 +       if (au_opt_test(mnt_flags, XINO)) {
26663 +               err = au_show_xino(m, sb);
26664 +               if (unlikely(err))
26665 +                       goto out;
26666 +       } else
26667 +               seq_puts(m, ",noxino");
26668 +
26669 +       AuBool(TRUNC_XINO, trunc_xino);
26670 +       AuStr(UDBA, udba);
26671 +       AuBool(SHWH, shwh);
26672 +       AuBool(PLINK, plink);
26673 +       AuBool(DIO, dio);
26674 +       AuBool(DIRPERM1, dirperm1);
26675 +       /* AuBool(REFROF, refrof); */
26676 +
26677 +       v = sbinfo->si_wbr_create;
26678 +       if (v != AuWbrCreate_Def)
26679 +               au_show_wbr_create(m, v, sbinfo);
26680 +
26681 +       v = sbinfo->si_wbr_copyup;
26682 +       if (v != AuWbrCopyup_Def)
26683 +               seq_printf(m, ",cpup=%s", au_optstr_wbr_copyup(v));
26684 +
26685 +       v = au_opt_test(mnt_flags, ALWAYS_DIROPQ);
26686 +       if (v != au_opt_test(AuOpt_Def, ALWAYS_DIROPQ))
26687 +               seq_printf(m, ",diropq=%c", v ? 'a' : 'w');
26688 +
26689 +       AuUInt(DIRWH, dirwh, sbinfo->si_dirwh);
26690 +
26691 +       v = jiffies_to_msecs(sbinfo->si_rdcache) / MSEC_PER_SEC;
26692 +       AuUInt(RDCACHE, rdcache, v);
26693 +
26694 +       AuUInt(RDBLK, rdblk, sbinfo->si_rdblk);
26695 +       AuUInt(RDHASH, rdhash, sbinfo->si_rdhash);
26696 +
26697 +       au_fhsm_show(m, sbinfo);
26698 +
26699 +       AuBool(SUM, sum);
26700 +       /* AuBool(SUM_W, wsum); */
26701 +       AuBool(WARN_PERM, warn_perm);
26702 +       AuBool(VERBOSE, verbose);
26703 +
26704 +out:
26705 +       /* be sure to print "br:" last */
26706 +       if (!sysaufs_brs) {
26707 +               seq_puts(m, ",br:");
26708 +               au_show_brs(m, sb);
26709 +       }
26710 +       si_read_unlock(sb);
26711 +       return 0;
26712 +
26713 +#undef AuBool
26714 +#undef AuStr
26715 +#undef AuUInt
26716 +}
26717 +
26718 +/* ---------------------------------------------------------------------- */
26719 +
26720 +/* sum mode which returns the summation for statfs(2) */
26721 +
26722 +static u64 au_add_till_max(u64 a, u64 b)
26723 +{
26724 +       u64 old;
26725 +
26726 +       old = a;
26727 +       a += b;
26728 +       if (old <= a)
26729 +               return a;
26730 +       return ULLONG_MAX;
26731 +}
26732 +
26733 +static u64 au_mul_till_max(u64 a, long mul)
26734 +{
26735 +       u64 old;
26736 +
26737 +       old = a;
26738 +       a *= mul;
26739 +       if (old <= a)
26740 +               return a;
26741 +       return ULLONG_MAX;
26742 +}
26743 +
26744 +static int au_statfs_sum(struct super_block *sb, struct kstatfs *buf)
26745 +{
26746 +       int err;
26747 +       long bsize, factor;
26748 +       u64 blocks, bfree, bavail, files, ffree;
26749 +       aufs_bindex_t bend, bindex, i;
26750 +       unsigned char shared;
26751 +       struct path h_path;
26752 +       struct super_block *h_sb;
26753 +
26754 +       err = 0;
26755 +       bsize = LONG_MAX;
26756 +       files = 0;
26757 +       ffree = 0;
26758 +       blocks = 0;
26759 +       bfree = 0;
26760 +       bavail = 0;
26761 +       bend = au_sbend(sb);
26762 +       for (bindex = 0; bindex <= bend; bindex++) {
26763 +               h_path.mnt = au_sbr_mnt(sb, bindex);
26764 +               h_sb = h_path.mnt->mnt_sb;
26765 +               shared = 0;
26766 +               for (i = 0; !shared && i < bindex; i++)
26767 +                       shared = (au_sbr_sb(sb, i) == h_sb);
26768 +               if (shared)
26769 +                       continue;
26770 +
26771 +               /* sb->s_root for NFS is unreliable */
26772 +               h_path.dentry = h_path.mnt->mnt_root;
26773 +               err = vfs_statfs(&h_path, buf);
26774 +               if (unlikely(err))
26775 +                       goto out;
26776 +
26777 +               if (bsize > buf->f_bsize) {
26778 +                       /*
26779 +                        * we will reduce bsize, so we have to expand blocks
26780 +                        * etc. to match them again
26781 +                        */
26782 +                       factor = (bsize / buf->f_bsize);
26783 +                       blocks = au_mul_till_max(blocks, factor);
26784 +                       bfree = au_mul_till_max(bfree, factor);
26785 +                       bavail = au_mul_till_max(bavail, factor);
26786 +                       bsize = buf->f_bsize;
26787 +               }
26788 +
26789 +               factor = (buf->f_bsize / bsize);
26790 +               blocks = au_add_till_max(blocks,
26791 +                               au_mul_till_max(buf->f_blocks, factor));
26792 +               bfree = au_add_till_max(bfree,
26793 +                               au_mul_till_max(buf->f_bfree, factor));
26794 +               bavail = au_add_till_max(bavail,
26795 +                               au_mul_till_max(buf->f_bavail, factor));
26796 +               files = au_add_till_max(files, buf->f_files);
26797 +               ffree = au_add_till_max(ffree, buf->f_ffree);
26798 +       }
26799 +
26800 +       buf->f_bsize = bsize;
26801 +       buf->f_blocks = blocks;
26802 +       buf->f_bfree = bfree;
26803 +       buf->f_bavail = bavail;
26804 +       buf->f_files = files;
26805 +       buf->f_ffree = ffree;
26806 +       buf->f_frsize = 0;
26807 +
26808 +out:
26809 +       return err;
26810 +}
26811 +
26812 +static int aufs_statfs(struct dentry *dentry, struct kstatfs *buf)
26813 +{
26814 +       int err;
26815 +       struct path h_path;
26816 +       struct super_block *sb;
26817 +
26818 +       /* lock free root dinfo */
26819 +       sb = dentry->d_sb;
26820 +       si_noflush_read_lock(sb);
26821 +       if (!au_opt_test(au_mntflags(sb), SUM)) {
26822 +               /* sb->s_root for NFS is unreliable */
26823 +               h_path.mnt = au_sbr_mnt(sb, 0);
26824 +               h_path.dentry = h_path.mnt->mnt_root;
26825 +               err = vfs_statfs(&h_path, buf);
26826 +       } else
26827 +               err = au_statfs_sum(sb, buf);
26828 +       si_read_unlock(sb);
26829 +
26830 +       if (!err) {
26831 +               buf->f_type = AUFS_SUPER_MAGIC;
26832 +               buf->f_namelen = AUFS_MAX_NAMELEN;
26833 +               memset(&buf->f_fsid, 0, sizeof(buf->f_fsid));
26834 +       }
26835 +       /* buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1; */
26836 +
26837 +       return err;
26838 +}
26839 +
26840 +/* ---------------------------------------------------------------------- */
26841 +
26842 +static int aufs_sync_fs(struct super_block *sb, int wait)
26843 +{
26844 +       int err, e;
26845 +       aufs_bindex_t bend, bindex;
26846 +       struct au_branch *br;
26847 +       struct super_block *h_sb;
26848 +
26849 +       err = 0;
26850 +       si_noflush_read_lock(sb);
26851 +       bend = au_sbend(sb);
26852 +       for (bindex = 0; bindex <= bend; bindex++) {
26853 +               br = au_sbr(sb, bindex);
26854 +               if (!au_br_writable(br->br_perm))
26855 +                       continue;
26856 +
26857 +               h_sb = au_sbr_sb(sb, bindex);
26858 +               if (h_sb->s_op->sync_fs) {
26859 +                       e = h_sb->s_op->sync_fs(h_sb, wait);
26860 +                       if (unlikely(e && !err))
26861 +                               err = e;
26862 +                       /* go on even if an error happens */
26863 +               }
26864 +       }
26865 +       si_read_unlock(sb);
26866 +
26867 +       return err;
26868 +}
26869 +
26870 +/* ---------------------------------------------------------------------- */
26871 +
26872 +/* final actions when unmounting a file system */
26873 +static void aufs_put_super(struct super_block *sb)
26874 +{
26875 +       struct au_sbinfo *sbinfo;
26876 +
26877 +       sbinfo = au_sbi(sb);
26878 +       if (!sbinfo)
26879 +               return;
26880 +
26881 +       dbgaufs_si_fin(sbinfo);
26882 +       kobject_put(&sbinfo->si_kobj);
26883 +}
26884 +
26885 +/* ---------------------------------------------------------------------- */
26886 +
26887 +void au_array_free(void *array)
26888 +{
26889 +       if (array) {
26890 +               if (!is_vmalloc_addr(array))
26891 +                       kfree(array);
26892 +               else
26893 +                       vfree(array);
26894 +       }
26895 +}
26896 +
26897 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb, void *arg)
26898 +{
26899 +       void *array;
26900 +       unsigned long long n, sz;
26901 +
26902 +       array = NULL;
26903 +       n = 0;
26904 +       if (!*hint)
26905 +               goto out;
26906 +
26907 +       if (*hint > ULLONG_MAX / sizeof(array)) {
26908 +               array = ERR_PTR(-EMFILE);
26909 +               pr_err("hint %llu\n", *hint);
26910 +               goto out;
26911 +       }
26912 +
26913 +       sz = sizeof(array) * *hint;
26914 +       array = kzalloc(sz, GFP_NOFS);
26915 +       if (unlikely(!array))
26916 +               array = vzalloc(sz);
26917 +       if (unlikely(!array)) {
26918 +               array = ERR_PTR(-ENOMEM);
26919 +               goto out;
26920 +       }
26921 +
26922 +       n = cb(array, *hint, arg);
26923 +       AuDebugOn(n > *hint);
26924 +
26925 +out:
26926 +       *hint = n;
26927 +       return array;
26928 +}
26929 +
26930 +static unsigned long long au_iarray_cb(void *a,
26931 +                                      unsigned long long max __maybe_unused,
26932 +                                      void *arg)
26933 +{
26934 +       unsigned long long n;
26935 +       struct inode **p, *inode;
26936 +       struct list_head *head;
26937 +
26938 +       n = 0;
26939 +       p = a;
26940 +       head = arg;
26941 +       spin_lock(&inode_sb_list_lock);
26942 +       list_for_each_entry(inode, head, i_sb_list) {
26943 +               if (!is_bad_inode(inode)
26944 +                   && au_ii(inode)->ii_bstart >= 0) {
26945 +                       spin_lock(&inode->i_lock);
26946 +                       if (atomic_read(&inode->i_count)) {
26947 +                               au_igrab(inode);
26948 +                               *p++ = inode;
26949 +                               n++;
26950 +                               AuDebugOn(n > max);
26951 +                       }
26952 +                       spin_unlock(&inode->i_lock);
26953 +               }
26954 +       }
26955 +       spin_unlock(&inode_sb_list_lock);
26956 +
26957 +       return n;
26958 +}
26959 +
26960 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max)
26961 +{
26962 +       *max = atomic_long_read(&au_sbi(sb)->si_ninodes);
26963 +       return au_array_alloc(max, au_iarray_cb, &sb->s_inodes);
26964 +}
26965 +
26966 +void au_iarray_free(struct inode **a, unsigned long long max)
26967 +{
26968 +       unsigned long long ull;
26969 +
26970 +       for (ull = 0; ull < max; ull++)
26971 +               iput(a[ull]);
26972 +       au_array_free(a);
26973 +}
26974 +
26975 +/* ---------------------------------------------------------------------- */
26976 +
26977 +/*
26978 + * refresh dentry and inode at remount time.
26979 + */
26980 +/* todo: consolidate with simple_reval_dpath() and au_reval_for_attr() */
26981 +static int au_do_refresh(struct dentry *dentry, unsigned int dir_flags,
26982 +                     struct dentry *parent)
26983 +{
26984 +       int err;
26985 +
26986 +       di_write_lock_child(dentry);
26987 +       di_read_lock_parent(parent, AuLock_IR);
26988 +       err = au_refresh_dentry(dentry, parent);
26989 +       if (!err && dir_flags)
26990 +               au_hn_reset(dentry->d_inode, dir_flags);
26991 +       di_read_unlock(parent, AuLock_IR);
26992 +       di_write_unlock(dentry);
26993 +
26994 +       return err;
26995 +}
26996 +
26997 +static int au_do_refresh_d(struct dentry *dentry, unsigned int sigen,
26998 +                          struct au_sbinfo *sbinfo,
26999 +                          const unsigned int dir_flags)
27000 +{
27001 +       int err;
27002 +       struct dentry *parent;
27003 +       struct inode *inode;
27004 +
27005 +       err = 0;
27006 +       parent = dget_parent(dentry);
27007 +       if (!au_digen_test(parent, sigen) && au_digen_test(dentry, sigen)) {
27008 +               inode = dentry->d_inode;
27009 +               if (inode) {
27010 +                       if (!S_ISDIR(inode->i_mode))
27011 +                               err = au_do_refresh(dentry, /*dir_flags*/0,
27012 +                                                parent);
27013 +                       else {
27014 +                               err = au_do_refresh(dentry, dir_flags, parent);
27015 +                               if (unlikely(err))
27016 +                                       au_fset_si(sbinfo, FAILED_REFRESH_DIR);
27017 +                       }
27018 +               } else
27019 +                       err = au_do_refresh(dentry, /*dir_flags*/0, parent);
27020 +               AuDbgDentry(dentry);
27021 +       }
27022 +       dput(parent);
27023 +
27024 +       AuTraceErr(err);
27025 +       return err;
27026 +}
27027 +
27028 +static int au_refresh_d(struct super_block *sb)
27029 +{
27030 +       int err, i, j, ndentry, e;
27031 +       unsigned int sigen;
27032 +       struct au_dcsub_pages dpages;
27033 +       struct au_dpage *dpage;
27034 +       struct dentry **dentries, *d;
27035 +       struct au_sbinfo *sbinfo;
27036 +       struct dentry *root = sb->s_root;
27037 +       const unsigned int dir_flags = au_hi_flags(root->d_inode, /*isdir*/1);
27038 +
27039 +       err = au_dpages_init(&dpages, GFP_NOFS);
27040 +       if (unlikely(err))
27041 +               goto out;
27042 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
27043 +       if (unlikely(err))
27044 +               goto out_dpages;
27045 +
27046 +       sigen = au_sigen(sb);
27047 +       sbinfo = au_sbi(sb);
27048 +       for (i = 0; i < dpages.ndpage; i++) {
27049 +               dpage = dpages.dpages + i;
27050 +               dentries = dpage->dentries;
27051 +               ndentry = dpage->ndentry;
27052 +               for (j = 0; j < ndentry; j++) {
27053 +                       d = dentries[j];
27054 +                       e = au_do_refresh_d(d, sigen, sbinfo, dir_flags);
27055 +                       if (unlikely(e && !err))
27056 +                               err = e;
27057 +                       /* go on even err */
27058 +               }
27059 +       }
27060 +
27061 +out_dpages:
27062 +       au_dpages_free(&dpages);
27063 +out:
27064 +       return err;
27065 +}
27066 +
27067 +static int au_refresh_i(struct super_block *sb)
27068 +{
27069 +       int err, e;
27070 +       unsigned int sigen;
27071 +       unsigned long long max, ull;
27072 +       struct inode *inode, **array;
27073 +
27074 +       array = au_iarray_alloc(sb, &max);
27075 +       err = PTR_ERR(array);
27076 +       if (IS_ERR(array))
27077 +               goto out;
27078 +
27079 +       err = 0;
27080 +       sigen = au_sigen(sb);
27081 +       for (ull = 0; ull < max; ull++) {
27082 +               inode = array[ull];
27083 +               if (unlikely(!inode))
27084 +                       break;
27085 +               if (au_iigen(inode, NULL) != sigen) {
27086 +                       ii_write_lock_child(inode);
27087 +                       e = au_refresh_hinode_self(inode);
27088 +                       ii_write_unlock(inode);
27089 +                       if (unlikely(e)) {
27090 +                               pr_err("error %d, i%lu\n", e, inode->i_ino);
27091 +                               if (!err)
27092 +                                       err = e;
27093 +                               /* go on even if err */
27094 +                       }
27095 +               }
27096 +       }
27097 +
27098 +       au_iarray_free(array, max);
27099 +
27100 +out:
27101 +       return err;
27102 +}
27103 +
27104 +static void au_remount_refresh(struct super_block *sb)
27105 +{
27106 +       int err, e;
27107 +       unsigned int udba;
27108 +       aufs_bindex_t bindex, bend;
27109 +       struct dentry *root;
27110 +       struct inode *inode;
27111 +       struct au_branch *br;
27112 +
27113 +       au_sigen_inc(sb);
27114 +       au_fclr_si(au_sbi(sb), FAILED_REFRESH_DIR);
27115 +
27116 +       root = sb->s_root;
27117 +       DiMustNoWaiters(root);
27118 +       inode = root->d_inode;
27119 +       IiMustNoWaiters(inode);
27120 +
27121 +       udba = au_opt_udba(sb);
27122 +       bend = au_sbend(sb);
27123 +       for (bindex = 0; bindex <= bend; bindex++) {
27124 +               br = au_sbr(sb, bindex);
27125 +               err = au_hnotify_reset_br(udba, br, br->br_perm);
27126 +               if (unlikely(err))
27127 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
27128 +                               bindex, err);
27129 +               /* go on even if err */
27130 +       }
27131 +       au_hn_reset(inode, au_hi_flags(inode, /*isdir*/1));
27132 +
27133 +       di_write_unlock(root);
27134 +       err = au_refresh_d(sb);
27135 +       e = au_refresh_i(sb);
27136 +       if (unlikely(e && !err))
27137 +               err = e;
27138 +       /* aufs_write_lock() calls ..._child() */
27139 +       di_write_lock_child(root);
27140 +
27141 +       au_cpup_attr_all(inode, /*force*/1);
27142 +
27143 +       if (unlikely(err))
27144 +               AuIOErr("refresh failed, ignored, %d\n", err);
27145 +}
27146 +
27147 +/* stop extra interpretation of errno in mount(8), and strange error messages */
27148 +static int cvt_err(int err)
27149 +{
27150 +       AuTraceErr(err);
27151 +
27152 +       switch (err) {
27153 +       case -ENOENT:
27154 +       case -ENOTDIR:
27155 +       case -EEXIST:
27156 +       case -EIO:
27157 +               err = -EINVAL;
27158 +       }
27159 +       return err;
27160 +}
27161 +
27162 +static int aufs_remount_fs(struct super_block *sb, int *flags, char *data)
27163 +{
27164 +       int err, do_dx;
27165 +       unsigned int mntflags;
27166 +       struct au_opts opts;
27167 +       struct dentry *root;
27168 +       struct inode *inode;
27169 +       struct au_sbinfo *sbinfo;
27170 +
27171 +       err = 0;
27172 +       root = sb->s_root;
27173 +       if (!data || !*data) {
27174 +               err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
27175 +               if (!err) {
27176 +                       di_write_lock_child(root);
27177 +                       err = au_opts_verify(sb, *flags, /*pending*/0);
27178 +                       aufs_write_unlock(root);
27179 +               }
27180 +               goto out;
27181 +       }
27182 +
27183 +       err = -ENOMEM;
27184 +       memset(&opts, 0, sizeof(opts));
27185 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
27186 +       if (unlikely(!opts.opt))
27187 +               goto out;
27188 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
27189 +       opts.flags = AuOpts_REMOUNT;
27190 +       opts.sb_flags = *flags;
27191 +
27192 +       /* parse it before aufs lock */
27193 +       err = au_opts_parse(sb, data, &opts);
27194 +       if (unlikely(err))
27195 +               goto out_opts;
27196 +
27197 +       sbinfo = au_sbi(sb);
27198 +       inode = root->d_inode;
27199 +       mutex_lock(&inode->i_mutex);
27200 +       err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
27201 +       if (unlikely(err))
27202 +               goto out_mtx;
27203 +       di_write_lock_child(root);
27204 +
27205 +       /* au_opts_remount() may return an error */
27206 +       err = au_opts_remount(sb, &opts);
27207 +       au_opts_free(&opts);
27208 +
27209 +       if (au_ftest_opts(opts.flags, REFRESH))
27210 +               au_remount_refresh(sb);
27211 +
27212 +       if (au_ftest_opts(opts.flags, REFRESH_DYAOP)) {
27213 +               mntflags = au_mntflags(sb);
27214 +               do_dx = !!au_opt_test(mntflags, DIO);
27215 +               au_dy_arefresh(do_dx);
27216 +       }
27217 +
27218 +       au_fhsm_wrote_all(sb, /*force*/1); /* ?? */
27219 +       aufs_write_unlock(root);
27220 +
27221 +out_mtx:
27222 +       mutex_unlock(&inode->i_mutex);
27223 +out_opts:
27224 +       free_page((unsigned long)opts.opt);
27225 +out:
27226 +       err = cvt_err(err);
27227 +       AuTraceErr(err);
27228 +       return err;
27229 +}
27230 +
27231 +static const struct super_operations aufs_sop = {
27232 +       .alloc_inode    = aufs_alloc_inode,
27233 +       .destroy_inode  = aufs_destroy_inode,
27234 +       /* always deleting, no clearing */
27235 +       .drop_inode     = generic_delete_inode,
27236 +       .show_options   = aufs_show_options,
27237 +       .statfs         = aufs_statfs,
27238 +       .put_super      = aufs_put_super,
27239 +       .sync_fs        = aufs_sync_fs,
27240 +       .remount_fs     = aufs_remount_fs
27241 +};
27242 +
27243 +/* ---------------------------------------------------------------------- */
27244 +
27245 +static int alloc_root(struct super_block *sb)
27246 +{
27247 +       int err;
27248 +       struct inode *inode;
27249 +       struct dentry *root;
27250 +
27251 +       err = -ENOMEM;
27252 +       inode = au_iget_locked(sb, AUFS_ROOT_INO);
27253 +       err = PTR_ERR(inode);
27254 +       if (IS_ERR(inode))
27255 +               goto out;
27256 +
27257 +       inode->i_op = &aufs_dir_iop;
27258 +       inode->i_fop = &aufs_dir_fop;
27259 +       inode->i_mode = S_IFDIR;
27260 +       set_nlink(inode, 2);
27261 +       unlock_new_inode(inode);
27262 +
27263 +       root = d_make_root(inode);
27264 +       if (unlikely(!root))
27265 +               goto out;
27266 +       err = PTR_ERR(root);
27267 +       if (IS_ERR(root))
27268 +               goto out;
27269 +
27270 +       err = au_di_init(root);
27271 +       if (!err) {
27272 +               sb->s_root = root;
27273 +               return 0; /* success */
27274 +       }
27275 +       dput(root);
27276 +
27277 +out:
27278 +       return err;
27279 +}
27280 +
27281 +static int aufs_fill_super(struct super_block *sb, void *raw_data,
27282 +                          int silent __maybe_unused)
27283 +{
27284 +       int err;
27285 +       struct au_opts opts;
27286 +       struct dentry *root;
27287 +       struct inode *inode;
27288 +       char *arg = raw_data;
27289 +
27290 +       if (unlikely(!arg || !*arg)) {
27291 +               err = -EINVAL;
27292 +               pr_err("no arg\n");
27293 +               goto out;
27294 +       }
27295 +
27296 +       err = -ENOMEM;
27297 +       memset(&opts, 0, sizeof(opts));
27298 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
27299 +       if (unlikely(!opts.opt))
27300 +               goto out;
27301 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
27302 +       opts.sb_flags = sb->s_flags;
27303 +
27304 +       err = au_si_alloc(sb);
27305 +       if (unlikely(err))
27306 +               goto out_opts;
27307 +
27308 +       /* all timestamps always follow the ones on the branch */
27309 +       sb->s_flags |= MS_NOATIME | MS_NODIRATIME;
27310 +       sb->s_op = &aufs_sop;
27311 +       sb->s_d_op = &aufs_dop;
27312 +       sb->s_magic = AUFS_SUPER_MAGIC;
27313 +       sb->s_maxbytes = 0;
27314 +       sb->s_stack_depth = 1;
27315 +       au_export_init(sb);
27316 +       /* au_xattr_init(sb); */
27317 +
27318 +       err = alloc_root(sb);
27319 +       if (unlikely(err)) {
27320 +               si_write_unlock(sb);
27321 +               goto out_info;
27322 +       }
27323 +       root = sb->s_root;
27324 +       inode = root->d_inode;
27325 +
27326 +       /*
27327 +        * actually we can parse options regardless aufs lock here.
27328 +        * but at remount time, parsing must be done before aufs lock.
27329 +        * so we follow the same rule.
27330 +        */
27331 +       ii_write_lock_parent(inode);
27332 +       aufs_write_unlock(root);
27333 +       err = au_opts_parse(sb, arg, &opts);
27334 +       if (unlikely(err))
27335 +               goto out_root;
27336 +
27337 +       /* lock vfs_inode first, then aufs. */
27338 +       mutex_lock(&inode->i_mutex);
27339 +       aufs_write_lock(root);
27340 +       err = au_opts_mount(sb, &opts);
27341 +       au_opts_free(&opts);
27342 +       aufs_write_unlock(root);
27343 +       mutex_unlock(&inode->i_mutex);
27344 +       if (!err)
27345 +               goto out_opts; /* success */
27346 +
27347 +out_root:
27348 +       dput(root);
27349 +       sb->s_root = NULL;
27350 +out_info:
27351 +       dbgaufs_si_fin(au_sbi(sb));
27352 +       kobject_put(&au_sbi(sb)->si_kobj);
27353 +       sb->s_fs_info = NULL;
27354 +out_opts:
27355 +       free_page((unsigned long)opts.opt);
27356 +out:
27357 +       AuTraceErr(err);
27358 +       err = cvt_err(err);
27359 +       AuTraceErr(err);
27360 +       return err;
27361 +}
27362 +
27363 +/* ---------------------------------------------------------------------- */
27364 +
27365 +static struct dentry *aufs_mount(struct file_system_type *fs_type, int flags,
27366 +                                const char *dev_name __maybe_unused,
27367 +                                void *raw_data)
27368 +{
27369 +       struct dentry *root;
27370 +       struct super_block *sb;
27371 +
27372 +       /* all timestamps always follow the ones on the branch */
27373 +       /* mnt->mnt_flags |= MNT_NOATIME | MNT_NODIRATIME; */
27374 +       root = mount_nodev(fs_type, flags, raw_data, aufs_fill_super);
27375 +       if (IS_ERR(root))
27376 +               goto out;
27377 +
27378 +       sb = root->d_sb;
27379 +       si_write_lock(sb, !AuLock_FLUSH);
27380 +       sysaufs_brs_add(sb, 0);
27381 +       si_write_unlock(sb);
27382 +       au_sbilist_add(sb);
27383 +
27384 +out:
27385 +       return root;
27386 +}
27387 +
27388 +static void aufs_kill_sb(struct super_block *sb)
27389 +{
27390 +       struct au_sbinfo *sbinfo;
27391 +
27392 +       sbinfo = au_sbi(sb);
27393 +       if (sbinfo) {
27394 +               au_sbilist_del(sb);
27395 +               aufs_write_lock(sb->s_root);
27396 +               au_fhsm_fin(sb);
27397 +               if (sbinfo->si_wbr_create_ops->fin)
27398 +                       sbinfo->si_wbr_create_ops->fin(sb);
27399 +               if (au_opt_test(sbinfo->si_mntflags, UDBA_HNOTIFY)) {
27400 +                       au_opt_set_udba(sbinfo->si_mntflags, UDBA_NONE);
27401 +                       au_remount_refresh(sb);
27402 +               }
27403 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
27404 +                       au_plink_put(sb, /*verbose*/1);
27405 +               au_xino_clr(sb);
27406 +               sbinfo->si_sb = NULL;
27407 +               aufs_write_unlock(sb->s_root);
27408 +               au_nwt_flush(&sbinfo->si_nowait);
27409 +       }
27410 +       kill_anon_super(sb);
27411 +}
27412 +
27413 +struct file_system_type aufs_fs_type = {
27414 +       .name           = AUFS_FSTYPE,
27415 +       /* a race between rename and others */
27416 +       .fs_flags       = FS_RENAME_DOES_D_MOVE,
27417 +       .mount          = aufs_mount,
27418 +       .kill_sb        = aufs_kill_sb,
27419 +       /* no need to __module_get() and module_put(). */
27420 +       .owner          = THIS_MODULE,
27421 +};
27422 diff -urN /usr/share/empty/fs/aufs/super.h linux/fs/aufs/super.h
27423 --- /usr/share/empty/fs/aufs/super.h    1970-01-01 01:00:00.000000000 +0100
27424 +++ linux/fs/aufs/super.h       2015-01-25 13:00:38.634380408 +0100
27425 @@ -0,0 +1,641 @@
27426 +/*
27427 + * Copyright (C) 2005-2014 Junjiro R. Okajima
27428 + *
27429 + * This program, aufs is free software; you can redistribute it and/or modify
27430 + * it under the terms of the GNU General Public License as published by
27431 + * the Free Software Foundation; either version 2 of the License, or
27432 + * (at your option) any later version.
27433 + *
27434 + * This program is distributed in the hope that it will be useful,
27435 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27436 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27437 + * GNU General Public License for more details.
27438 + *
27439 + * You should have received a copy of the GNU General Public License
27440 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27441 + */
27442 +
27443 +/*
27444 + * super_block operations
27445 + */
27446 +
27447 +#ifndef __AUFS_SUPER_H__
27448 +#define __AUFS_SUPER_H__
27449 +
27450 +#ifdef __KERNEL__
27451 +
27452 +#include <linux/fs.h>
27453 +#include "rwsem.h"
27454 +#include "spl.h"
27455 +#include "wkq.h"
27456 +
27457 +typedef ssize_t (*au_readf_t)(struct file *, char __user *, size_t, loff_t *);
27458 +typedef ssize_t (*au_writef_t)(struct file *, const char __user *, size_t,
27459 +                              loff_t *);
27460 +
27461 +/* policies to select one among multiple writable branches */
27462 +struct au_wbr_copyup_operations {
27463 +       int (*copyup)(struct dentry *dentry);
27464 +};
27465 +
27466 +#define AuWbr_DIR      1               /* target is a dir */
27467 +#define AuWbr_PARENT   (1 << 1)        /* always require a parent */
27468 +
27469 +#define au_ftest_wbr(flags, name)      ((flags) & AuWbr_##name)
27470 +#define au_fset_wbr(flags, name)       { (flags) |= AuWbr_##name; }
27471 +#define au_fclr_wbr(flags, name)       { (flags) &= ~AuWbr_##name; }
27472 +
27473 +struct au_wbr_create_operations {
27474 +       int (*create)(struct dentry *dentry, unsigned int flags);
27475 +       int (*init)(struct super_block *sb);
27476 +       int (*fin)(struct super_block *sb);
27477 +};
27478 +
27479 +struct au_wbr_mfs {
27480 +       struct mutex    mfs_lock; /* protect this structure */
27481 +       unsigned long   mfs_jiffy;
27482 +       unsigned long   mfs_expire;
27483 +       aufs_bindex_t   mfs_bindex;
27484 +
27485 +       unsigned long long      mfsrr_bytes;
27486 +       unsigned long long      mfsrr_watermark;
27487 +};
27488 +
27489 +struct pseudo_link {
27490 +       union {
27491 +               struct hlist_node hlist;
27492 +               struct rcu_head rcu;
27493 +       };
27494 +       struct inode *inode;
27495 +};
27496 +
27497 +#define AuPlink_NHASH 100
27498 +static inline int au_plink_hash(ino_t ino)
27499 +{
27500 +       return ino % AuPlink_NHASH;
27501 +}
27502 +
27503 +/* File-based Hierarchical Storage Management */
27504 +struct au_fhsm {
27505 +#ifdef CONFIG_AUFS_FHSM
27506 +       /* allow only one process who can receive the notification */
27507 +       spinlock_t              fhsm_spin;
27508 +       pid_t                   fhsm_pid;
27509 +       wait_queue_head_t       fhsm_wqh;
27510 +       atomic_t                fhsm_readable;
27511 +
27512 +       /* these are protected by si_rwsem */
27513 +       unsigned long           fhsm_expire;
27514 +       aufs_bindex_t           fhsm_bottom;
27515 +#endif
27516 +};
27517 +
27518 +struct au_branch;
27519 +struct au_sbinfo {
27520 +       /* nowait tasks in the system-wide workqueue */
27521 +       struct au_nowait_tasks  si_nowait;
27522 +
27523 +       /*
27524 +        * tried sb->s_umount, but failed due to the dependecy between i_mutex.
27525 +        * rwsem for au_sbinfo is necessary.
27526 +        */
27527 +       struct au_rwsem         si_rwsem;
27528 +
27529 +       /* prevent recursive locking in deleting inode */
27530 +       struct {
27531 +               unsigned long           *bitmap;
27532 +               spinlock_t              tree_lock;
27533 +               struct radix_tree_root  tree;
27534 +       } au_si_pid;
27535 +
27536 +       /*
27537 +        * dirty approach to protect sb->sb_inodes and ->s_files (gone) from
27538 +        * remount.
27539 +        */
27540 +       atomic_long_t           si_ninodes, si_nfiles;
27541 +
27542 +       /* branch management */
27543 +       unsigned int            si_generation;
27544 +
27545 +       /* see above flags */
27546 +       unsigned char           au_si_status;
27547 +
27548 +       aufs_bindex_t           si_bend;
27549 +
27550 +       /* dirty trick to keep br_id plus */
27551 +       unsigned int            si_last_br_id :
27552 +                               sizeof(aufs_bindex_t) * BITS_PER_BYTE - 1;
27553 +       struct au_branch        **si_branch;
27554 +
27555 +       /* policy to select a writable branch */
27556 +       unsigned char           si_wbr_copyup;
27557 +       unsigned char           si_wbr_create;
27558 +       struct au_wbr_copyup_operations *si_wbr_copyup_ops;
27559 +       struct au_wbr_create_operations *si_wbr_create_ops;
27560 +
27561 +       /* round robin */
27562 +       atomic_t                si_wbr_rr_next;
27563 +
27564 +       /* most free space */
27565 +       struct au_wbr_mfs       si_wbr_mfs;
27566 +
27567 +       /* File-based Hierarchical Storage Management */
27568 +       struct au_fhsm          si_fhsm;
27569 +
27570 +       /* mount flags */
27571 +       /* include/asm-ia64/siginfo.h defines a macro named si_flags */
27572 +       unsigned int            si_mntflags;
27573 +
27574 +       /* external inode number (bitmap and translation table) */
27575 +       au_readf_t              si_xread;
27576 +       au_writef_t             si_xwrite;
27577 +       struct file             *si_xib;
27578 +       struct mutex            si_xib_mtx; /* protect xib members */
27579 +       unsigned long           *si_xib_buf;
27580 +       unsigned long           si_xib_last_pindex;
27581 +       int                     si_xib_next_bit;
27582 +       aufs_bindex_t           si_xino_brid;
27583 +       unsigned long           si_xino_jiffy;
27584 +       unsigned long           si_xino_expire;
27585 +       /* reserved for future use */
27586 +       /* unsigned long long   si_xib_limit; */        /* Max xib file size */
27587 +
27588 +#ifdef CONFIG_AUFS_EXPORT
27589 +       /* i_generation */
27590 +       struct file             *si_xigen;
27591 +       atomic_t                si_xigen_next;
27592 +#endif
27593 +
27594 +       /* vdir parameters */
27595 +       unsigned long           si_rdcache;     /* max cache time in jiffies */
27596 +       unsigned int            si_rdblk;       /* deblk size */
27597 +       unsigned int            si_rdhash;      /* hash size */
27598 +
27599 +       /*
27600 +        * If the number of whiteouts are larger than si_dirwh, leave all of
27601 +        * them after au_whtmp_ren to reduce the cost of rmdir(2).
27602 +        * future fsck.aufs or kernel thread will remove them later.
27603 +        * Otherwise, remove all whiteouts and the dir in rmdir(2).
27604 +        */
27605 +       unsigned int            si_dirwh;
27606 +
27607 +       /*
27608 +        * rename(2) a directory with all children.
27609 +        */
27610 +       /* reserved for future use */
27611 +       /* int                  si_rendir; */
27612 +
27613 +       /* pseudo_link list */
27614 +       struct au_sphlhead      si_plink[AuPlink_NHASH];
27615 +       wait_queue_head_t       si_plink_wq;
27616 +       spinlock_t              si_plink_maint_lock;
27617 +       pid_t                   si_plink_maint_pid;
27618 +
27619 +       /* file list */
27620 +       struct au_sphlhead      si_files;
27621 +
27622 +       /*
27623 +        * sysfs and lifetime management.
27624 +        * this is not a small structure and it may be a waste of memory in case
27625 +        * of sysfs is disabled, particulary when many aufs-es are mounted.
27626 +        * but using sysfs is majority.
27627 +        */
27628 +       struct kobject          si_kobj;
27629 +#ifdef CONFIG_DEBUG_FS
27630 +       struct dentry            *si_dbgaufs;
27631 +       struct dentry            *si_dbgaufs_plink;
27632 +       struct dentry            *si_dbgaufs_xib;
27633 +#ifdef CONFIG_AUFS_EXPORT
27634 +       struct dentry            *si_dbgaufs_xigen;
27635 +#endif
27636 +#endif
27637 +
27638 +#ifdef CONFIG_AUFS_SBILIST
27639 +       struct list_head        si_list;
27640 +#endif
27641 +
27642 +       /* dirty, necessary for unmounting, sysfs and sysrq */
27643 +       struct super_block      *si_sb;
27644 +};
27645 +
27646 +/* sbinfo status flags */
27647 +/*
27648 + * set true when refresh_dirs() failed at remount time.
27649 + * then try refreshing dirs at access time again.
27650 + * if it is false, refreshing dirs at access time is unnecesary
27651 + */
27652 +#define AuSi_FAILED_REFRESH_DIR        1
27653 +
27654 +#define AuSi_FHSM              (1 << 1)        /* fhsm is active now */
27655 +
27656 +#ifndef CONFIG_AUFS_FHSM
27657 +#undef AuSi_FHSM
27658 +#define AuSi_FHSM              0
27659 +#endif
27660 +
27661 +static inline unsigned char au_do_ftest_si(struct au_sbinfo *sbi,
27662 +                                          unsigned int flag)
27663 +{
27664 +       AuRwMustAnyLock(&sbi->si_rwsem);
27665 +       return sbi->au_si_status & flag;
27666 +}
27667 +#define au_ftest_si(sbinfo, name)      au_do_ftest_si(sbinfo, AuSi_##name)
27668 +#define au_fset_si(sbinfo, name) do { \
27669 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
27670 +       (sbinfo)->au_si_status |= AuSi_##name; \
27671 +} while (0)
27672 +#define au_fclr_si(sbinfo, name) do { \
27673 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
27674 +       (sbinfo)->au_si_status &= ~AuSi_##name; \
27675 +} while (0)
27676 +
27677 +/* ---------------------------------------------------------------------- */
27678 +
27679 +/* policy to select one among writable branches */
27680 +#define AuWbrCopyup(sbinfo, ...) \
27681 +       ((sbinfo)->si_wbr_copyup_ops->copyup(__VA_ARGS__))
27682 +#define AuWbrCreate(sbinfo, ...) \
27683 +       ((sbinfo)->si_wbr_create_ops->create(__VA_ARGS__))
27684 +
27685 +/* flags for si_read_lock()/aufs_read_lock()/di_read_lock() */
27686 +#define AuLock_DW              1               /* write-lock dentry */
27687 +#define AuLock_IR              (1 << 1)        /* read-lock inode */
27688 +#define AuLock_IW              (1 << 2)        /* write-lock inode */
27689 +#define AuLock_FLUSH           (1 << 3)        /* wait for 'nowait' tasks */
27690 +#define AuLock_DIR             (1 << 4)        /* target is a dir */
27691 +#define AuLock_NOPLM           (1 << 5)        /* return err in plm mode */
27692 +#define AuLock_NOPLMW          (1 << 6)        /* wait for plm mode ends */
27693 +#define AuLock_GEN             (1 << 7)        /* test digen/iigen */
27694 +#define au_ftest_lock(flags, name)     ((flags) & AuLock_##name)
27695 +#define au_fset_lock(flags, name) \
27696 +       do { (flags) |= AuLock_##name; } while (0)
27697 +#define au_fclr_lock(flags, name) \
27698 +       do { (flags) &= ~AuLock_##name; } while (0)
27699 +
27700 +/* ---------------------------------------------------------------------- */
27701 +
27702 +/* super.c */
27703 +extern struct file_system_type aufs_fs_type;
27704 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino);
27705 +typedef unsigned long long (*au_arraycb_t)(void *array, unsigned long long max,
27706 +                                          void *arg);
27707 +void au_array_free(void *array);
27708 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb, void *arg);
27709 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max);
27710 +void au_iarray_free(struct inode **a, unsigned long long max);
27711 +
27712 +/* sbinfo.c */
27713 +void au_si_free(struct kobject *kobj);
27714 +int au_si_alloc(struct super_block *sb);
27715 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr);
27716 +
27717 +unsigned int au_sigen_inc(struct super_block *sb);
27718 +aufs_bindex_t au_new_br_id(struct super_block *sb);
27719 +
27720 +int si_read_lock(struct super_block *sb, int flags);
27721 +int si_write_lock(struct super_block *sb, int flags);
27722 +int aufs_read_lock(struct dentry *dentry, int flags);
27723 +void aufs_read_unlock(struct dentry *dentry, int flags);
27724 +void aufs_write_lock(struct dentry *dentry);
27725 +void aufs_write_unlock(struct dentry *dentry);
27726 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags);
27727 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2);
27728 +
27729 +int si_pid_test_slow(struct super_block *sb);
27730 +void si_pid_set_slow(struct super_block *sb);
27731 +void si_pid_clr_slow(struct super_block *sb);
27732 +
27733 +/* wbr_policy.c */
27734 +extern struct au_wbr_copyup_operations au_wbr_copyup_ops[];
27735 +extern struct au_wbr_create_operations au_wbr_create_ops[];
27736 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst);
27737 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex);
27738 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t bstart);
27739 +
27740 +/* mvdown.c */
27741 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *arg);
27742 +
27743 +#ifdef CONFIG_AUFS_FHSM
27744 +/* fhsm.c */
27745 +
27746 +static inline pid_t au_fhsm_pid(struct au_fhsm *fhsm)
27747 +{
27748 +       pid_t pid;
27749 +
27750 +       spin_lock(&fhsm->fhsm_spin);
27751 +       pid = fhsm->fhsm_pid;
27752 +       spin_unlock(&fhsm->fhsm_spin);
27753 +
27754 +       return pid;
27755 +}
27756 +
27757 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force);
27758 +void au_fhsm_wrote_all(struct super_block *sb, int force);
27759 +int au_fhsm_fd(struct super_block *sb, int oflags);
27760 +int au_fhsm_br_alloc(struct au_branch *br);
27761 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex);
27762 +void au_fhsm_fin(struct super_block *sb);
27763 +void au_fhsm_init(struct au_sbinfo *sbinfo);
27764 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec);
27765 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo);
27766 +#else
27767 +AuStubVoid(au_fhsm_wrote, struct super_block *sb, aufs_bindex_t bindex,
27768 +          int force)
27769 +AuStubVoid(au_fhsm_wrote_all, struct super_block *sb, int force)
27770 +AuStub(int, au_fhsm_fd, return -EOPNOTSUPP, struct super_block *sb, int oflags)
27771 +AuStub(pid_t, au_fhsm_pid, return 0, struct au_fhsm *fhsm)
27772 +AuStubInt0(au_fhsm_br_alloc, struct au_branch *br)
27773 +AuStubVoid(au_fhsm_set_bottom, struct super_block *sb, aufs_bindex_t bindex)
27774 +AuStubVoid(au_fhsm_fin, struct super_block *sb)
27775 +AuStubVoid(au_fhsm_init, struct au_sbinfo *sbinfo)
27776 +AuStubVoid(au_fhsm_set, struct au_sbinfo *sbinfo, unsigned int sec)
27777 +AuStubVoid(au_fhsm_show, struct seq_file *seq, struct au_sbinfo *sbinfo)
27778 +#endif
27779 +
27780 +/* ---------------------------------------------------------------------- */
27781 +
27782 +static inline struct au_sbinfo *au_sbi(struct super_block *sb)
27783 +{
27784 +       return sb->s_fs_info;
27785 +}
27786 +
27787 +/* ---------------------------------------------------------------------- */
27788 +
27789 +#ifdef CONFIG_AUFS_EXPORT
27790 +int au_test_nfsd(void);
27791 +void au_export_init(struct super_block *sb);
27792 +void au_xigen_inc(struct inode *inode);
27793 +int au_xigen_new(struct inode *inode);
27794 +int au_xigen_set(struct super_block *sb, struct file *base);
27795 +void au_xigen_clr(struct super_block *sb);
27796 +
27797 +static inline int au_busy_or_stale(void)
27798 +{
27799 +       if (!au_test_nfsd())
27800 +               return -EBUSY;
27801 +       return -ESTALE;
27802 +}
27803 +#else
27804 +AuStubInt0(au_test_nfsd, void)
27805 +AuStubVoid(au_export_init, struct super_block *sb)
27806 +AuStubVoid(au_xigen_inc, struct inode *inode)
27807 +AuStubInt0(au_xigen_new, struct inode *inode)
27808 +AuStubInt0(au_xigen_set, struct super_block *sb, struct file *base)
27809 +AuStubVoid(au_xigen_clr, struct super_block *sb)
27810 +AuStub(int, au_busy_or_stale, return -EBUSY, void)
27811 +#endif /* CONFIG_AUFS_EXPORT */
27812 +
27813 +/* ---------------------------------------------------------------------- */
27814 +
27815 +#ifdef CONFIG_AUFS_SBILIST
27816 +/* module.c */
27817 +extern struct au_splhead au_sbilist;
27818 +
27819 +static inline void au_sbilist_init(void)
27820 +{
27821 +       au_spl_init(&au_sbilist);
27822 +}
27823 +
27824 +static inline void au_sbilist_add(struct super_block *sb)
27825 +{
27826 +       au_spl_add(&au_sbi(sb)->si_list, &au_sbilist);
27827 +}
27828 +
27829 +static inline void au_sbilist_del(struct super_block *sb)
27830 +{
27831 +       au_spl_del(&au_sbi(sb)->si_list, &au_sbilist);
27832 +}
27833 +
27834 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
27835 +static inline void au_sbilist_lock(void)
27836 +{
27837 +       spin_lock(&au_sbilist.spin);
27838 +}
27839 +
27840 +static inline void au_sbilist_unlock(void)
27841 +{
27842 +       spin_unlock(&au_sbilist.spin);
27843 +}
27844 +#define AuGFP_SBILIST  GFP_ATOMIC
27845 +#else
27846 +AuStubVoid(au_sbilist_lock, void)
27847 +AuStubVoid(au_sbilist_unlock, void)
27848 +#define AuGFP_SBILIST  GFP_NOFS
27849 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
27850 +#else
27851 +AuStubVoid(au_sbilist_init, void)
27852 +AuStubVoid(au_sbilist_add, struct super_block *sb)
27853 +AuStubVoid(au_sbilist_del, struct super_block *sb)
27854 +AuStubVoid(au_sbilist_lock, void)
27855 +AuStubVoid(au_sbilist_unlock, void)
27856 +#define AuGFP_SBILIST  GFP_NOFS
27857 +#endif
27858 +
27859 +/* ---------------------------------------------------------------------- */
27860 +
27861 +static inline void dbgaufs_si_null(struct au_sbinfo *sbinfo)
27862 +{
27863 +       /*
27864 +        * This function is a dynamic '__init' function actually,
27865 +        * so the tiny check for si_rwsem is unnecessary.
27866 +        */
27867 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
27868 +#ifdef CONFIG_DEBUG_FS
27869 +       sbinfo->si_dbgaufs = NULL;
27870 +       sbinfo->si_dbgaufs_plink = NULL;
27871 +       sbinfo->si_dbgaufs_xib = NULL;
27872 +#ifdef CONFIG_AUFS_EXPORT
27873 +       sbinfo->si_dbgaufs_xigen = NULL;
27874 +#endif
27875 +#endif
27876 +}
27877 +
27878 +/* ---------------------------------------------------------------------- */
27879 +
27880 +static inline pid_t si_pid_bit(void)
27881 +{
27882 +       /* the origin of pid is 1, but the bitmap's is 0 */
27883 +       return current->pid - 1;
27884 +}
27885 +
27886 +static inline int si_pid_test(struct super_block *sb)
27887 +{
27888 +       pid_t bit;
27889 +
27890 +       bit = si_pid_bit();
27891 +       if (bit < PID_MAX_DEFAULT)
27892 +               return test_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
27893 +       return si_pid_test_slow(sb);
27894 +}
27895 +
27896 +static inline void si_pid_set(struct super_block *sb)
27897 +{
27898 +       pid_t bit;
27899 +
27900 +       bit = si_pid_bit();
27901 +       if (bit < PID_MAX_DEFAULT) {
27902 +               AuDebugOn(test_bit(bit, au_sbi(sb)->au_si_pid.bitmap));
27903 +               set_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
27904 +               /* smp_mb(); */
27905 +       } else
27906 +               si_pid_set_slow(sb);
27907 +}
27908 +
27909 +static inline void si_pid_clr(struct super_block *sb)
27910 +{
27911 +       pid_t bit;
27912 +
27913 +       bit = si_pid_bit();
27914 +       if (bit < PID_MAX_DEFAULT) {
27915 +               AuDebugOn(!test_bit(bit, au_sbi(sb)->au_si_pid.bitmap));
27916 +               clear_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
27917 +               /* smp_mb(); */
27918 +       } else
27919 +               si_pid_clr_slow(sb);
27920 +}
27921 +
27922 +/* ---------------------------------------------------------------------- */
27923 +
27924 +/* lock superblock. mainly for entry point functions */
27925 +/*
27926 + * __si_read_lock, __si_write_lock,
27927 + * __si_read_unlock, __si_write_unlock, __si_downgrade_lock
27928 + */
27929 +AuSimpleRwsemFuncs(__si, struct super_block *sb, &au_sbi(sb)->si_rwsem);
27930 +
27931 +#define SiMustNoWaiters(sb)    AuRwMustNoWaiters(&au_sbi(sb)->si_rwsem)
27932 +#define SiMustAnyLock(sb)      AuRwMustAnyLock(&au_sbi(sb)->si_rwsem)
27933 +#define SiMustWriteLock(sb)    AuRwMustWriteLock(&au_sbi(sb)->si_rwsem)
27934 +
27935 +static inline void si_noflush_read_lock(struct super_block *sb)
27936 +{
27937 +       __si_read_lock(sb);
27938 +       si_pid_set(sb);
27939 +}
27940 +
27941 +static inline int si_noflush_read_trylock(struct super_block *sb)
27942 +{
27943 +       int locked;
27944 +
27945 +       locked = __si_read_trylock(sb);
27946 +       if (locked)
27947 +               si_pid_set(sb);
27948 +       return locked;
27949 +}
27950 +
27951 +static inline void si_noflush_write_lock(struct super_block *sb)
27952 +{
27953 +       __si_write_lock(sb);
27954 +       si_pid_set(sb);
27955 +}
27956 +
27957 +static inline int si_noflush_write_trylock(struct super_block *sb)
27958 +{
27959 +       int locked;
27960 +
27961 +       locked = __si_write_trylock(sb);
27962 +       if (locked)
27963 +               si_pid_set(sb);
27964 +       return locked;
27965 +}
27966 +
27967 +#if 0 /* unused */
27968 +static inline int si_read_trylock(struct super_block *sb, int flags)
27969 +{
27970 +       if (au_ftest_lock(flags, FLUSH))
27971 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
27972 +       return si_noflush_read_trylock(sb);
27973 +}
27974 +#endif
27975 +
27976 +static inline void si_read_unlock(struct super_block *sb)
27977 +{
27978 +       si_pid_clr(sb);
27979 +       __si_read_unlock(sb);
27980 +}
27981 +
27982 +#if 0 /* unused */
27983 +static inline int si_write_trylock(struct super_block *sb, int flags)
27984 +{
27985 +       if (au_ftest_lock(flags, FLUSH))
27986 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
27987 +       return si_noflush_write_trylock(sb);
27988 +}
27989 +#endif
27990 +
27991 +static inline void si_write_unlock(struct super_block *sb)
27992 +{
27993 +       si_pid_clr(sb);
27994 +       __si_write_unlock(sb);
27995 +}
27996 +
27997 +#if 0 /* unused */
27998 +static inline void si_downgrade_lock(struct super_block *sb)
27999 +{
28000 +       __si_downgrade_lock(sb);
28001 +}
28002 +#endif
28003 +
28004 +/* ---------------------------------------------------------------------- */
28005 +
28006 +static inline aufs_bindex_t au_sbend(struct super_block *sb)
28007 +{
28008 +       SiMustAnyLock(sb);
28009 +       return au_sbi(sb)->si_bend;
28010 +}
28011 +
28012 +static inline unsigned int au_mntflags(struct super_block *sb)
28013 +{
28014 +       SiMustAnyLock(sb);
28015 +       return au_sbi(sb)->si_mntflags;
28016 +}
28017 +
28018 +static inline unsigned int au_sigen(struct super_block *sb)
28019 +{
28020 +       SiMustAnyLock(sb);
28021 +       return au_sbi(sb)->si_generation;
28022 +}
28023 +
28024 +static inline void au_ninodes_inc(struct super_block *sb)
28025 +{
28026 +       atomic_long_inc(&au_sbi(sb)->si_ninodes);
28027 +}
28028 +
28029 +static inline void au_ninodes_dec(struct super_block *sb)
28030 +{
28031 +       AuDebugOn(!atomic_long_read(&au_sbi(sb)->si_ninodes));
28032 +       atomic_long_dec(&au_sbi(sb)->si_ninodes);
28033 +}
28034 +
28035 +static inline void au_nfiles_inc(struct super_block *sb)
28036 +{
28037 +       atomic_long_inc(&au_sbi(sb)->si_nfiles);
28038 +}
28039 +
28040 +static inline void au_nfiles_dec(struct super_block *sb)
28041 +{
28042 +       AuDebugOn(!atomic_long_read(&au_sbi(sb)->si_nfiles));
28043 +       atomic_long_dec(&au_sbi(sb)->si_nfiles);
28044 +}
28045 +
28046 +static inline struct au_branch *au_sbr(struct super_block *sb,
28047 +                                      aufs_bindex_t bindex)
28048 +{
28049 +       SiMustAnyLock(sb);
28050 +       return au_sbi(sb)->si_branch[0 + bindex];
28051 +}
28052 +
28053 +static inline void au_xino_brid_set(struct super_block *sb, aufs_bindex_t brid)
28054 +{
28055 +       SiMustWriteLock(sb);
28056 +       au_sbi(sb)->si_xino_brid = brid;
28057 +}
28058 +
28059 +static inline aufs_bindex_t au_xino_brid(struct super_block *sb)
28060 +{
28061 +       SiMustAnyLock(sb);
28062 +       return au_sbi(sb)->si_xino_brid;
28063 +}
28064 +
28065 +#endif /* __KERNEL__ */
28066 +#endif /* __AUFS_SUPER_H__ */
28067 diff -urN /usr/share/empty/fs/aufs/sysaufs.c linux/fs/aufs/sysaufs.c
28068 --- /usr/share/empty/fs/aufs/sysaufs.c  1970-01-01 01:00:00.000000000 +0100
28069 +++ linux/fs/aufs/sysaufs.c     2015-01-25 13:00:38.634380408 +0100
28070 @@ -0,0 +1,104 @@
28071 +/*
28072 + * Copyright (C) 2005-2014 Junjiro R. Okajima
28073 + *
28074 + * This program, aufs is free software; you can redistribute it and/or modify
28075 + * it under the terms of the GNU General Public License as published by
28076 + * the Free Software Foundation; either version 2 of the License, or
28077 + * (at your option) any later version.
28078 + *
28079 + * This program is distributed in the hope that it will be useful,
28080 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28081 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28082 + * GNU General Public License for more details.
28083 + *
28084 + * You should have received a copy of the GNU General Public License
28085 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28086 + */
28087 +
28088 +/*
28089 + * sysfs interface and lifetime management
28090 + * they are necessary regardless sysfs is disabled.
28091 + */
28092 +
28093 +#include <linux/random.h>
28094 +#include "aufs.h"
28095 +
28096 +unsigned long sysaufs_si_mask;
28097 +struct kset *sysaufs_kset;
28098 +
28099 +#define AuSiAttr(_name) { \
28100 +       .attr   = { .name = __stringify(_name), .mode = 0444 }, \
28101 +       .show   = sysaufs_si_##_name,                           \
28102 +}
28103 +
28104 +static struct sysaufs_si_attr sysaufs_si_attr_xi_path = AuSiAttr(xi_path);
28105 +struct attribute *sysaufs_si_attrs[] = {
28106 +       &sysaufs_si_attr_xi_path.attr,
28107 +       NULL,
28108 +};
28109 +
28110 +static const struct sysfs_ops au_sbi_ops = {
28111 +       .show   = sysaufs_si_show
28112 +};
28113 +
28114 +static struct kobj_type au_sbi_ktype = {
28115 +       .release        = au_si_free,
28116 +       .sysfs_ops      = &au_sbi_ops,
28117 +       .default_attrs  = sysaufs_si_attrs
28118 +};
28119 +
28120 +/* ---------------------------------------------------------------------- */
28121 +
28122 +int sysaufs_si_init(struct au_sbinfo *sbinfo)
28123 +{
28124 +       int err;
28125 +
28126 +       sbinfo->si_kobj.kset = sysaufs_kset;
28127 +       /* cf. sysaufs_name() */
28128 +       err = kobject_init_and_add
28129 +               (&sbinfo->si_kobj, &au_sbi_ktype, /*&sysaufs_kset->kobj*/NULL,
28130 +                SysaufsSiNamePrefix "%lx", sysaufs_si_id(sbinfo));
28131 +
28132 +       dbgaufs_si_null(sbinfo);
28133 +       if (!err) {
28134 +               err = dbgaufs_si_init(sbinfo);
28135 +               if (unlikely(err))
28136 +                       kobject_put(&sbinfo->si_kobj);
28137 +       }
28138 +       return err;
28139 +}
28140 +
28141 +void sysaufs_fin(void)
28142 +{
28143 +       dbgaufs_fin();
28144 +       sysfs_remove_group(&sysaufs_kset->kobj, sysaufs_attr_group);
28145 +       kset_unregister(sysaufs_kset);
28146 +}
28147 +
28148 +int __init sysaufs_init(void)
28149 +{
28150 +       int err;
28151 +
28152 +       do {
28153 +               get_random_bytes(&sysaufs_si_mask, sizeof(sysaufs_si_mask));
28154 +       } while (!sysaufs_si_mask);
28155 +
28156 +       err = -EINVAL;
28157 +       sysaufs_kset = kset_create_and_add(AUFS_NAME, NULL, fs_kobj);
28158 +       if (unlikely(!sysaufs_kset))
28159 +               goto out;
28160 +       err = PTR_ERR(sysaufs_kset);
28161 +       if (IS_ERR(sysaufs_kset))
28162 +               goto out;
28163 +       err = sysfs_create_group(&sysaufs_kset->kobj, sysaufs_attr_group);
28164 +       if (unlikely(err)) {
28165 +               kset_unregister(sysaufs_kset);
28166 +               goto out;
28167 +       }
28168 +
28169 +       err = dbgaufs_init();
28170 +       if (unlikely(err))
28171 +               sysaufs_fin();
28172 +out:
28173 +       return err;
28174 +}
28175 diff -urN /usr/share/empty/fs/aufs/sysaufs.h linux/fs/aufs/sysaufs.h
28176 --- /usr/share/empty/fs/aufs/sysaufs.h  1970-01-01 01:00:00.000000000 +0100
28177 +++ linux/fs/aufs/sysaufs.h     2015-01-25 13:00:38.634380408 +0100
28178 @@ -0,0 +1,101 @@
28179 +/*
28180 + * Copyright (C) 2005-2014 Junjiro R. Okajima
28181 + *
28182 + * This program, aufs is free software; you can redistribute it and/or modify
28183 + * it under the terms of the GNU General Public License as published by
28184 + * the Free Software Foundation; either version 2 of the License, or
28185 + * (at your option) any later version.
28186 + *
28187 + * This program is distributed in the hope that it will be useful,
28188 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28189 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28190 + * GNU General Public License for more details.
28191 + *
28192 + * You should have received a copy of the GNU General Public License
28193 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28194 + */
28195 +
28196 +/*
28197 + * sysfs interface and mount lifetime management
28198 + */
28199 +
28200 +#ifndef __SYSAUFS_H__
28201 +#define __SYSAUFS_H__
28202 +
28203 +#ifdef __KERNEL__
28204 +
28205 +#include <linux/sysfs.h>
28206 +#include "module.h"
28207 +
28208 +struct super_block;
28209 +struct au_sbinfo;
28210 +
28211 +struct sysaufs_si_attr {
28212 +       struct attribute attr;
28213 +       int (*show)(struct seq_file *seq, struct super_block *sb);
28214 +};
28215 +
28216 +/* ---------------------------------------------------------------------- */
28217 +
28218 +/* sysaufs.c */
28219 +extern unsigned long sysaufs_si_mask;
28220 +extern struct kset *sysaufs_kset;
28221 +extern struct attribute *sysaufs_si_attrs[];
28222 +int sysaufs_si_init(struct au_sbinfo *sbinfo);
28223 +int __init sysaufs_init(void);
28224 +void sysaufs_fin(void);
28225 +
28226 +/* ---------------------------------------------------------------------- */
28227 +
28228 +/* some people doesn't like to show a pointer in kernel */
28229 +static inline unsigned long sysaufs_si_id(struct au_sbinfo *sbinfo)
28230 +{
28231 +       return sysaufs_si_mask ^ (unsigned long)sbinfo;
28232 +}
28233 +
28234 +#define SysaufsSiNamePrefix    "si_"
28235 +#define SysaufsSiNameLen       (sizeof(SysaufsSiNamePrefix) + 16)
28236 +static inline void sysaufs_name(struct au_sbinfo *sbinfo, char *name)
28237 +{
28238 +       snprintf(name, SysaufsSiNameLen, SysaufsSiNamePrefix "%lx",
28239 +                sysaufs_si_id(sbinfo));
28240 +}
28241 +
28242 +struct au_branch;
28243 +#ifdef CONFIG_SYSFS
28244 +/* sysfs.c */
28245 +extern struct attribute_group *sysaufs_attr_group;
28246 +
28247 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb);
28248 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
28249 +                        char *buf);
28250 +long au_brinfo_ioctl(struct file *file, unsigned long arg);
28251 +#ifdef CONFIG_COMPAT
28252 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg);
28253 +#endif
28254 +
28255 +void sysaufs_br_init(struct au_branch *br);
28256 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
28257 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
28258 +
28259 +#define sysaufs_brs_init()     do {} while (0)
28260 +
28261 +#else
28262 +#define sysaufs_attr_group     NULL
28263 +
28264 +AuStubInt0(sysaufs_si_xi_path, struct seq_file *seq, struct super_block *sb)
28265 +AuStub(ssize_t, sysaufs_si_show, return 0, struct kobject *kobj,
28266 +       struct attribute *attr, char *buf)
28267 +AuStubVoid(sysaufs_br_init, struct au_branch *br)
28268 +AuStubVoid(sysaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
28269 +AuStubVoid(sysaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
28270 +
28271 +static inline void sysaufs_brs_init(void)
28272 +{
28273 +       sysaufs_brs = 0;
28274 +}
28275 +
28276 +#endif /* CONFIG_SYSFS */
28277 +
28278 +#endif /* __KERNEL__ */
28279 +#endif /* __SYSAUFS_H__ */
28280 diff -urN /usr/share/empty/fs/aufs/sysfs.c linux/fs/aufs/sysfs.c
28281 --- /usr/share/empty/fs/aufs/sysfs.c    1970-01-01 01:00:00.000000000 +0100
28282 +++ linux/fs/aufs/sysfs.c       2015-01-25 13:00:38.634380408 +0100
28283 @@ -0,0 +1,372 @@
28284 +/*
28285 + * Copyright (C) 2005-2014 Junjiro R. Okajima
28286 + *
28287 + * This program, aufs is free software; you can redistribute it and/or modify
28288 + * it under the terms of the GNU General Public License as published by
28289 + * the Free Software Foundation; either version 2 of the License, or
28290 + * (at your option) any later version.
28291 + *
28292 + * This program is distributed in the hope that it will be useful,
28293 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28294 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28295 + * GNU General Public License for more details.
28296 + *
28297 + * You should have received a copy of the GNU General Public License
28298 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28299 + */
28300 +
28301 +/*
28302 + * sysfs interface
28303 + */
28304 +
28305 +#include <linux/compat.h>
28306 +#include <linux/seq_file.h>
28307 +#include "aufs.h"
28308 +
28309 +#ifdef CONFIG_AUFS_FS_MODULE
28310 +/* this entry violates the "one line per file" policy of sysfs */
28311 +static ssize_t config_show(struct kobject *kobj, struct kobj_attribute *attr,
28312 +                          char *buf)
28313 +{
28314 +       ssize_t err;
28315 +       static char *conf =
28316 +/* this file is generated at compiling */
28317 +#include "conf.str"
28318 +               ;
28319 +
28320 +       err = snprintf(buf, PAGE_SIZE, conf);
28321 +       if (unlikely(err >= PAGE_SIZE))
28322 +               err = -EFBIG;
28323 +       return err;
28324 +}
28325 +
28326 +static struct kobj_attribute au_config_attr = __ATTR_RO(config);
28327 +#endif
28328 +
28329 +static struct attribute *au_attr[] = {
28330 +#ifdef CONFIG_AUFS_FS_MODULE
28331 +       &au_config_attr.attr,
28332 +#endif
28333 +       NULL,   /* need to NULL terminate the list of attributes */
28334 +};
28335 +
28336 +static struct attribute_group sysaufs_attr_group_body = {
28337 +       .attrs = au_attr
28338 +};
28339 +
28340 +struct attribute_group *sysaufs_attr_group = &sysaufs_attr_group_body;
28341 +
28342 +/* ---------------------------------------------------------------------- */
28343 +
28344 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb)
28345 +{
28346 +       int err;
28347 +
28348 +       SiMustAnyLock(sb);
28349 +
28350 +       err = 0;
28351 +       if (au_opt_test(au_mntflags(sb), XINO)) {
28352 +               err = au_xino_path(seq, au_sbi(sb)->si_xib);
28353 +               seq_putc(seq, '\n');
28354 +       }
28355 +       return err;
28356 +}
28357 +
28358 +/*
28359 + * the lifetime of branch is independent from the entry under sysfs.
28360 + * sysfs handles the lifetime of the entry, and never call ->show() after it is
28361 + * unlinked.
28362 + */
28363 +static int sysaufs_si_br(struct seq_file *seq, struct super_block *sb,
28364 +                        aufs_bindex_t bindex, int idx)
28365 +{
28366 +       int err;
28367 +       struct path path;
28368 +       struct dentry *root;
28369 +       struct au_branch *br;
28370 +       au_br_perm_str_t perm;
28371 +
28372 +       AuDbg("b%d\n", bindex);
28373 +
28374 +       err = 0;
28375 +       root = sb->s_root;
28376 +       di_read_lock_parent(root, !AuLock_IR);
28377 +       br = au_sbr(sb, bindex);
28378 +
28379 +       switch (idx) {
28380 +       case AuBrSysfs_BR:
28381 +               path.mnt = au_br_mnt(br);
28382 +               path.dentry = au_h_dptr(root, bindex);
28383 +               au_seq_path(seq, &path);
28384 +               au_optstr_br_perm(&perm, br->br_perm);
28385 +               err = seq_printf(seq, "=%s\n", perm.a);
28386 +               break;
28387 +       case AuBrSysfs_BRID:
28388 +               err = seq_printf(seq, "%d\n", br->br_id);
28389 +               break;
28390 +       }
28391 +       di_read_unlock(root, !AuLock_IR);
28392 +       if (err == -1)
28393 +               err = -E2BIG;
28394 +
28395 +       return err;
28396 +}
28397 +
28398 +/* ---------------------------------------------------------------------- */
28399 +
28400 +static struct seq_file *au_seq(char *p, ssize_t len)
28401 +{
28402 +       struct seq_file *seq;
28403 +
28404 +       seq = kzalloc(sizeof(*seq), GFP_NOFS);
28405 +       if (seq) {
28406 +               /* mutex_init(&seq.lock); */
28407 +               seq->buf = p;
28408 +               seq->size = len;
28409 +               return seq; /* success */
28410 +       }
28411 +
28412 +       seq = ERR_PTR(-ENOMEM);
28413 +       return seq;
28414 +}
28415 +
28416 +#define SysaufsBr_PREFIX       "br"
28417 +#define SysaufsBrid_PREFIX     "brid"
28418 +
28419 +/* todo: file size may exceed PAGE_SIZE */
28420 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
28421 +                       char *buf)
28422 +{
28423 +       ssize_t err;
28424 +       int idx;
28425 +       long l;
28426 +       aufs_bindex_t bend;
28427 +       struct au_sbinfo *sbinfo;
28428 +       struct super_block *sb;
28429 +       struct seq_file *seq;
28430 +       char *name;
28431 +       struct attribute **cattr;
28432 +
28433 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
28434 +       sb = sbinfo->si_sb;
28435 +
28436 +       /*
28437 +        * prevent a race condition between sysfs and aufs.
28438 +        * for instance, sysfs_file_read() calls sysfs_get_active_two() which
28439 +        * prohibits maintaining the sysfs entries.
28440 +        * hew we acquire read lock after sysfs_get_active_two().
28441 +        * on the other hand, the remount process may maintain the sysfs/aufs
28442 +        * entries after acquiring write lock.
28443 +        * it can cause a deadlock.
28444 +        * simply we gave up processing read here.
28445 +        */
28446 +       err = -EBUSY;
28447 +       if (unlikely(!si_noflush_read_trylock(sb)))
28448 +               goto out;
28449 +
28450 +       seq = au_seq(buf, PAGE_SIZE);
28451 +       err = PTR_ERR(seq);
28452 +       if (IS_ERR(seq))
28453 +               goto out_unlock;
28454 +
28455 +       name = (void *)attr->name;
28456 +       cattr = sysaufs_si_attrs;
28457 +       while (*cattr) {
28458 +               if (!strcmp(name, (*cattr)->name)) {
28459 +                       err = container_of(*cattr, struct sysaufs_si_attr, attr)
28460 +                               ->show(seq, sb);
28461 +                       goto out_seq;
28462 +               }
28463 +               cattr++;
28464 +       }
28465 +
28466 +       if (!strncmp(name, SysaufsBrid_PREFIX,
28467 +                    sizeof(SysaufsBrid_PREFIX) - 1)) {
28468 +               idx = AuBrSysfs_BRID;
28469 +               name += sizeof(SysaufsBrid_PREFIX) - 1;
28470 +       } else if (!strncmp(name, SysaufsBr_PREFIX,
28471 +                           sizeof(SysaufsBr_PREFIX) - 1)) {
28472 +               idx = AuBrSysfs_BR;
28473 +               name += sizeof(SysaufsBr_PREFIX) - 1;
28474 +       } else
28475 +                 BUG();
28476 +
28477 +       err = kstrtol(name, 10, &l);
28478 +       if (!err) {
28479 +               bend = au_sbend(sb);
28480 +               if (l <= bend)
28481 +                       err = sysaufs_si_br(seq, sb, (aufs_bindex_t)l, idx);
28482 +               else
28483 +                       err = -ENOENT;
28484 +       }
28485 +
28486 +out_seq:
28487 +       if (!err) {
28488 +               err = seq->count;
28489 +               /* sysfs limit */
28490 +               if (unlikely(err == PAGE_SIZE))
28491 +                       err = -EFBIG;
28492 +       }
28493 +       kfree(seq);
28494 +out_unlock:
28495 +       si_read_unlock(sb);
28496 +out:
28497 +       return err;
28498 +}
28499 +
28500 +/* ---------------------------------------------------------------------- */
28501 +
28502 +static int au_brinfo(struct super_block *sb, union aufs_brinfo __user *arg)
28503 +{
28504 +       int err;
28505 +       int16_t brid;
28506 +       aufs_bindex_t bindex, bend;
28507 +       size_t sz;
28508 +       char *buf;
28509 +       struct seq_file *seq;
28510 +       struct au_branch *br;
28511 +
28512 +       si_read_lock(sb, AuLock_FLUSH);
28513 +       bend = au_sbend(sb);
28514 +       err = bend + 1;
28515 +       if (!arg)
28516 +               goto out;
28517 +
28518 +       err = -ENOMEM;
28519 +       buf = (void *)__get_free_page(GFP_NOFS);
28520 +       if (unlikely(!buf))
28521 +               goto out;
28522 +
28523 +       seq = au_seq(buf, PAGE_SIZE);
28524 +       err = PTR_ERR(seq);
28525 +       if (IS_ERR(seq))
28526 +               goto out_buf;
28527 +
28528 +       sz = sizeof(*arg) - offsetof(union aufs_brinfo, path);
28529 +       for (bindex = 0; bindex <= bend; bindex++, arg++) {
28530 +               err = !access_ok(VERIFY_WRITE, arg, sizeof(*arg));
28531 +               if (unlikely(err))
28532 +                       break;
28533 +
28534 +               br = au_sbr(sb, bindex);
28535 +               brid = br->br_id;
28536 +               BUILD_BUG_ON(sizeof(brid) != sizeof(arg->id));
28537 +               err = __put_user(brid, &arg->id);
28538 +               if (unlikely(err))
28539 +                       break;
28540 +
28541 +               BUILD_BUG_ON(sizeof(br->br_perm) != sizeof(arg->perm));
28542 +               err = __put_user(br->br_perm, &arg->perm);
28543 +               if (unlikely(err))
28544 +                       break;
28545 +
28546 +               au_seq_path(seq, &br->br_path);
28547 +               err = seq_putc(seq, '\0');
28548 +               if (!err && seq->count <= sz) {
28549 +                       err = copy_to_user(arg->path, seq->buf, seq->count);
28550 +                       seq->count = 0;
28551 +                       if (unlikely(err))
28552 +                               break;
28553 +               } else {
28554 +                       err = -E2BIG;
28555 +                       goto out_seq;
28556 +               }
28557 +       }
28558 +       if (unlikely(err))
28559 +               err = -EFAULT;
28560 +
28561 +out_seq:
28562 +       kfree(seq);
28563 +out_buf:
28564 +       free_page((unsigned long)buf);
28565 +out:
28566 +       si_read_unlock(sb);
28567 +       return err;
28568 +}
28569 +
28570 +long au_brinfo_ioctl(struct file *file, unsigned long arg)
28571 +{
28572 +       return au_brinfo(file->f_dentry->d_sb, (void __user *)arg);
28573 +}
28574 +
28575 +#ifdef CONFIG_COMPAT
28576 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg)
28577 +{
28578 +       return au_brinfo(file->f_dentry->d_sb, compat_ptr(arg));
28579 +}
28580 +#endif
28581 +
28582 +/* ---------------------------------------------------------------------- */
28583 +
28584 +void sysaufs_br_init(struct au_branch *br)
28585 +{
28586 +       int i;
28587 +       struct au_brsysfs *br_sysfs;
28588 +       struct attribute *attr;
28589 +
28590 +       br_sysfs = br->br_sysfs;
28591 +       for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
28592 +               attr = &br_sysfs->attr;
28593 +               sysfs_attr_init(attr);
28594 +               attr->name = br_sysfs->name;
28595 +               attr->mode = S_IRUGO;
28596 +               br_sysfs++;
28597 +       }
28598 +}
28599 +
28600 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
28601 +{
28602 +       struct au_branch *br;
28603 +       struct kobject *kobj;
28604 +       struct au_brsysfs *br_sysfs;
28605 +       int i;
28606 +       aufs_bindex_t bend;
28607 +
28608 +       dbgaufs_brs_del(sb, bindex);
28609 +
28610 +       if (!sysaufs_brs)
28611 +               return;
28612 +
28613 +       kobj = &au_sbi(sb)->si_kobj;
28614 +       bend = au_sbend(sb);
28615 +       for (; bindex <= bend; bindex++) {
28616 +               br = au_sbr(sb, bindex);
28617 +               br_sysfs = br->br_sysfs;
28618 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
28619 +                       sysfs_remove_file(kobj, &br_sysfs->attr);
28620 +                       br_sysfs++;
28621 +               }
28622 +       }
28623 +}
28624 +
28625 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
28626 +{
28627 +       int err, i;
28628 +       aufs_bindex_t bend;
28629 +       struct kobject *kobj;
28630 +       struct au_branch *br;
28631 +       struct au_brsysfs *br_sysfs;
28632 +
28633 +       dbgaufs_brs_add(sb, bindex);
28634 +
28635 +       if (!sysaufs_brs)
28636 +               return;
28637 +
28638 +       kobj = &au_sbi(sb)->si_kobj;
28639 +       bend = au_sbend(sb);
28640 +       for (; bindex <= bend; bindex++) {
28641 +               br = au_sbr(sb, bindex);
28642 +               br_sysfs = br->br_sysfs;
28643 +               snprintf(br_sysfs[AuBrSysfs_BR].name, sizeof(br_sysfs->name),
28644 +                        SysaufsBr_PREFIX "%d", bindex);
28645 +               snprintf(br_sysfs[AuBrSysfs_BRID].name, sizeof(br_sysfs->name),
28646 +                        SysaufsBrid_PREFIX "%d", bindex);
28647 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
28648 +                       err = sysfs_create_file(kobj, &br_sysfs->attr);
28649 +                       if (unlikely(err))
28650 +                               pr_warn("failed %s under sysfs(%d)\n",
28651 +                                       br_sysfs->name, err);
28652 +                       br_sysfs++;
28653 +               }
28654 +       }
28655 +}
28656 diff -urN /usr/share/empty/fs/aufs/sysrq.c linux/fs/aufs/sysrq.c
28657 --- /usr/share/empty/fs/aufs/sysrq.c    1970-01-01 01:00:00.000000000 +0100
28658 +++ linux/fs/aufs/sysrq.c       2015-01-25 13:00:38.634380408 +0100
28659 @@ -0,0 +1,157 @@
28660 +/*
28661 + * Copyright (C) 2005-2014 Junjiro R. Okajima
28662 + *
28663 + * This program, aufs is free software; you can redistribute it and/or modify
28664 + * it under the terms of the GNU General Public License as published by
28665 + * the Free Software Foundation; either version 2 of the License, or
28666 + * (at your option) any later version.
28667 + *
28668 + * This program is distributed in the hope that it will be useful,
28669 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28670 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28671 + * GNU General Public License for more details.
28672 + *
28673 + * You should have received a copy of the GNU General Public License
28674 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28675 + */
28676 +
28677 +/*
28678 + * magic sysrq hanlder
28679 + */
28680 +
28681 +/* #include <linux/sysrq.h> */
28682 +#include <linux/writeback.h>
28683 +#include "aufs.h"
28684 +
28685 +/* ---------------------------------------------------------------------- */
28686 +
28687 +static void sysrq_sb(struct super_block *sb)
28688 +{
28689 +       char *plevel;
28690 +       struct au_sbinfo *sbinfo;
28691 +       struct file *file;
28692 +       struct au_sphlhead *files;
28693 +       struct au_finfo *finfo;
28694 +
28695 +       plevel = au_plevel;
28696 +       au_plevel = KERN_WARNING;
28697 +
28698 +       /* since we define pr_fmt, call printk directly */
28699 +#define pr(str) printk(KERN_WARNING AUFS_NAME ": " str)
28700 +
28701 +       sbinfo = au_sbi(sb);
28702 +       printk(KERN_WARNING "si=%lx\n", sysaufs_si_id(sbinfo));
28703 +       pr("superblock\n");
28704 +       au_dpri_sb(sb);
28705 +
28706 +#if 0
28707 +       pr("root dentry\n");
28708 +       au_dpri_dentry(sb->s_root);
28709 +       pr("root inode\n");
28710 +       au_dpri_inode(sb->s_root->d_inode);
28711 +#endif
28712 +
28713 +#if 0
28714 +       do {
28715 +               int err, i, j, ndentry;
28716 +               struct au_dcsub_pages dpages;
28717 +               struct au_dpage *dpage;
28718 +
28719 +               err = au_dpages_init(&dpages, GFP_ATOMIC);
28720 +               if (unlikely(err))
28721 +                       break;
28722 +               err = au_dcsub_pages(&dpages, sb->s_root, NULL, NULL);
28723 +               if (!err)
28724 +                       for (i = 0; i < dpages.ndpage; i++) {
28725 +                               dpage = dpages.dpages + i;
28726 +                               ndentry = dpage->ndentry;
28727 +                               for (j = 0; j < ndentry; j++)
28728 +                                       au_dpri_dentry(dpage->dentries[j]);
28729 +                       }
28730 +               au_dpages_free(&dpages);
28731 +       } while (0);
28732 +#endif
28733 +
28734 +#if 1
28735 +       {
28736 +               struct inode *i;
28737 +
28738 +               pr("isolated inode\n");
28739 +               spin_lock(&inode_sb_list_lock);
28740 +               list_for_each_entry(i, &sb->s_inodes, i_sb_list) {
28741 +                       spin_lock(&i->i_lock);
28742 +                       if (1 || hlist_empty(&i->i_dentry))
28743 +                               au_dpri_inode(i);
28744 +                       spin_unlock(&i->i_lock);
28745 +               }
28746 +               spin_unlock(&inode_sb_list_lock);
28747 +       }
28748 +#endif
28749 +       pr("files\n");
28750 +       files = &au_sbi(sb)->si_files;
28751 +       spin_lock(&files->spin);
28752 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
28753 +               umode_t mode;
28754 +
28755 +               file = finfo->fi_file;
28756 +               mode = file_inode(file)->i_mode;
28757 +               if (!special_file(mode))
28758 +                       au_dpri_file(file);
28759 +       }
28760 +       spin_unlock(&files->spin);
28761 +       pr("done\n");
28762 +
28763 +#undef pr
28764 +       au_plevel = plevel;
28765 +}
28766 +
28767 +/* ---------------------------------------------------------------------- */
28768 +
28769 +/* module parameter */
28770 +static char *aufs_sysrq_key = "a";
28771 +module_param_named(sysrq, aufs_sysrq_key, charp, S_IRUGO);
28772 +MODULE_PARM_DESC(sysrq, "MagicSysRq key for " AUFS_NAME);
28773 +
28774 +static void au_sysrq(int key __maybe_unused)
28775 +{
28776 +       struct au_sbinfo *sbinfo;
28777 +
28778 +       lockdep_off();
28779 +       au_sbilist_lock();
28780 +       list_for_each_entry(sbinfo, &au_sbilist.head, si_list)
28781 +               sysrq_sb(sbinfo->si_sb);
28782 +       au_sbilist_unlock();
28783 +       lockdep_on();
28784 +}
28785 +
28786 +static struct sysrq_key_op au_sysrq_op = {
28787 +       .handler        = au_sysrq,
28788 +       .help_msg       = "Aufs",
28789 +       .action_msg     = "Aufs",
28790 +       .enable_mask    = SYSRQ_ENABLE_DUMP
28791 +};
28792 +
28793 +/* ---------------------------------------------------------------------- */
28794 +
28795 +int __init au_sysrq_init(void)
28796 +{
28797 +       int err;
28798 +       char key;
28799 +
28800 +       err = -1;
28801 +       key = *aufs_sysrq_key;
28802 +       if ('a' <= key && key <= 'z')
28803 +               err = register_sysrq_key(key, &au_sysrq_op);
28804 +       if (unlikely(err))
28805 +               pr_err("err %d, sysrq=%c\n", err, key);
28806 +       return err;
28807 +}
28808 +
28809 +void au_sysrq_fin(void)
28810 +{
28811 +       int err;
28812 +
28813 +       err = unregister_sysrq_key(*aufs_sysrq_key, &au_sysrq_op);
28814 +       if (unlikely(err))
28815 +               pr_err("err %d (ignored)\n", err);
28816 +}
28817 diff -urN /usr/share/empty/fs/aufs/vdir.c linux/fs/aufs/vdir.c
28818 --- /usr/share/empty/fs/aufs/vdir.c     1970-01-01 01:00:00.000000000 +0100
28819 +++ linux/fs/aufs/vdir.c        2015-01-25 13:00:38.634380408 +0100
28820 @@ -0,0 +1,889 @@
28821 +/*
28822 + * Copyright (C) 2005-2014 Junjiro R. Okajima
28823 + *
28824 + * This program, aufs is free software; you can redistribute it and/or modify
28825 + * it under the terms of the GNU General Public License as published by
28826 + * the Free Software Foundation; either version 2 of the License, or
28827 + * (at your option) any later version.
28828 + *
28829 + * This program is distributed in the hope that it will be useful,
28830 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28831 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28832 + * GNU General Public License for more details.
28833 + *
28834 + * You should have received a copy of the GNU General Public License
28835 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28836 + */
28837 +
28838 +/*
28839 + * virtual or vertical directory
28840 + */
28841 +
28842 +#include "aufs.h"
28843 +
28844 +static unsigned int calc_size(int nlen)
28845 +{
28846 +       return ALIGN(sizeof(struct au_vdir_de) + nlen, sizeof(ino_t));
28847 +}
28848 +
28849 +static int set_deblk_end(union au_vdir_deblk_p *p,
28850 +                        union au_vdir_deblk_p *deblk_end)
28851 +{
28852 +       if (calc_size(0) <= deblk_end->deblk - p->deblk) {
28853 +               p->de->de_str.len = 0;
28854 +               /* smp_mb(); */
28855 +               return 0;
28856 +       }
28857 +       return -1; /* error */
28858 +}
28859 +
28860 +/* returns true or false */
28861 +static int is_deblk_end(union au_vdir_deblk_p *p,
28862 +                       union au_vdir_deblk_p *deblk_end)
28863 +{
28864 +       if (calc_size(0) <= deblk_end->deblk - p->deblk)
28865 +               return !p->de->de_str.len;
28866 +       return 1;
28867 +}
28868 +
28869 +static unsigned char *last_deblk(struct au_vdir *vdir)
28870 +{
28871 +       return vdir->vd_deblk[vdir->vd_nblk - 1];
28872 +}
28873 +
28874 +/* ---------------------------------------------------------------------- */
28875 +
28876 +/* estimate the apropriate size for name hash table */
28877 +unsigned int au_rdhash_est(loff_t sz)
28878 +{
28879 +       unsigned int n;
28880 +
28881 +       n = UINT_MAX;
28882 +       sz >>= 10;
28883 +       if (sz < n)
28884 +               n = sz;
28885 +       if (sz < AUFS_RDHASH_DEF)
28886 +               n = AUFS_RDHASH_DEF;
28887 +       /* pr_info("n %u\n", n); */
28888 +       return n;
28889 +}
28890 +
28891 +/*
28892 + * the allocated memory has to be freed by
28893 + * au_nhash_wh_free() or au_nhash_de_free().
28894 + */
28895 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp)
28896 +{
28897 +       struct hlist_head *head;
28898 +       unsigned int u;
28899 +       size_t sz;
28900 +
28901 +       sz = sizeof(*nhash->nh_head) * num_hash;
28902 +       head = kmalloc(sz, gfp);
28903 +       if (head) {
28904 +               nhash->nh_num = num_hash;
28905 +               nhash->nh_head = head;
28906 +               for (u = 0; u < num_hash; u++)
28907 +                       INIT_HLIST_HEAD(head++);
28908 +               return 0; /* success */
28909 +       }
28910 +
28911 +       return -ENOMEM;
28912 +}
28913 +
28914 +static void nhash_count(struct hlist_head *head)
28915 +{
28916 +#if 0
28917 +       unsigned long n;
28918 +       struct hlist_node *pos;
28919 +
28920 +       n = 0;
28921 +       hlist_for_each(pos, head)
28922 +               n++;
28923 +       pr_info("%lu\n", n);
28924 +#endif
28925 +}
28926 +
28927 +static void au_nhash_wh_do_free(struct hlist_head *head)
28928 +{
28929 +       struct au_vdir_wh *pos;
28930 +       struct hlist_node *node;
28931 +
28932 +       hlist_for_each_entry_safe(pos, node, head, wh_hash)
28933 +               kfree(pos);
28934 +}
28935 +
28936 +static void au_nhash_de_do_free(struct hlist_head *head)
28937 +{
28938 +       struct au_vdir_dehstr *pos;
28939 +       struct hlist_node *node;
28940 +
28941 +       hlist_for_each_entry_safe(pos, node, head, hash)
28942 +               au_cache_free_vdir_dehstr(pos);
28943 +}
28944 +
28945 +static void au_nhash_do_free(struct au_nhash *nhash,
28946 +                            void (*free)(struct hlist_head *head))
28947 +{
28948 +       unsigned int n;
28949 +       struct hlist_head *head;
28950 +
28951 +       n = nhash->nh_num;
28952 +       if (!n)
28953 +               return;
28954 +
28955 +       head = nhash->nh_head;
28956 +       while (n-- > 0) {
28957 +               nhash_count(head);
28958 +               free(head++);
28959 +       }
28960 +       kfree(nhash->nh_head);
28961 +}
28962 +
28963 +void au_nhash_wh_free(struct au_nhash *whlist)
28964 +{
28965 +       au_nhash_do_free(whlist, au_nhash_wh_do_free);
28966 +}
28967 +
28968 +static void au_nhash_de_free(struct au_nhash *delist)
28969 +{
28970 +       au_nhash_do_free(delist, au_nhash_de_do_free);
28971 +}
28972 +
28973 +/* ---------------------------------------------------------------------- */
28974 +
28975 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
28976 +                           int limit)
28977 +{
28978 +       int num;
28979 +       unsigned int u, n;
28980 +       struct hlist_head *head;
28981 +       struct au_vdir_wh *pos;
28982 +
28983 +       num = 0;
28984 +       n = whlist->nh_num;
28985 +       head = whlist->nh_head;
28986 +       for (u = 0; u < n; u++, head++)
28987 +               hlist_for_each_entry(pos, head, wh_hash)
28988 +                       if (pos->wh_bindex == btgt && ++num > limit)
28989 +                               return 1;
28990 +       return 0;
28991 +}
28992 +
28993 +static struct hlist_head *au_name_hash(struct au_nhash *nhash,
28994 +                                      unsigned char *name,
28995 +                                      unsigned int len)
28996 +{
28997 +       unsigned int v;
28998 +       /* const unsigned int magic_bit = 12; */
28999 +
29000 +       AuDebugOn(!nhash->nh_num || !nhash->nh_head);
29001 +
29002 +       v = 0;
29003 +       while (len--)
29004 +               v += *name++;
29005 +       /* v = hash_long(v, magic_bit); */
29006 +       v %= nhash->nh_num;
29007 +       return nhash->nh_head + v;
29008 +}
29009 +
29010 +static int au_nhash_test_name(struct au_vdir_destr *str, const char *name,
29011 +                             int nlen)
29012 +{
29013 +       return str->len == nlen && !memcmp(str->name, name, nlen);
29014 +}
29015 +
29016 +/* returns found or not */
29017 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen)
29018 +{
29019 +       struct hlist_head *head;
29020 +       struct au_vdir_wh *pos;
29021 +       struct au_vdir_destr *str;
29022 +
29023 +       head = au_name_hash(whlist, name, nlen);
29024 +       hlist_for_each_entry(pos, head, wh_hash) {
29025 +               str = &pos->wh_str;
29026 +               AuDbg("%.*s\n", str->len, str->name);
29027 +               if (au_nhash_test_name(str, name, nlen))
29028 +                       return 1;
29029 +       }
29030 +       return 0;
29031 +}
29032 +
29033 +/* returns found(true) or not */
29034 +static int test_known(struct au_nhash *delist, char *name, int nlen)
29035 +{
29036 +       struct hlist_head *head;
29037 +       struct au_vdir_dehstr *pos;
29038 +       struct au_vdir_destr *str;
29039 +
29040 +       head = au_name_hash(delist, name, nlen);
29041 +       hlist_for_each_entry(pos, head, hash) {
29042 +               str = pos->str;
29043 +               AuDbg("%.*s\n", str->len, str->name);
29044 +               if (au_nhash_test_name(str, name, nlen))
29045 +                       return 1;
29046 +       }
29047 +       return 0;
29048 +}
29049 +
29050 +static void au_shwh_init_wh(struct au_vdir_wh *wh, ino_t ino,
29051 +                           unsigned char d_type)
29052 +{
29053 +#ifdef CONFIG_AUFS_SHWH
29054 +       wh->wh_ino = ino;
29055 +       wh->wh_type = d_type;
29056 +#endif
29057 +}
29058 +
29059 +/* ---------------------------------------------------------------------- */
29060 +
29061 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
29062 +                      unsigned int d_type, aufs_bindex_t bindex,
29063 +                      unsigned char shwh)
29064 +{
29065 +       int err;
29066 +       struct au_vdir_destr *str;
29067 +       struct au_vdir_wh *wh;
29068 +
29069 +       AuDbg("%.*s\n", nlen, name);
29070 +       AuDebugOn(!whlist->nh_num || !whlist->nh_head);
29071 +
29072 +       err = -ENOMEM;
29073 +       wh = kmalloc(sizeof(*wh) + nlen, GFP_NOFS);
29074 +       if (unlikely(!wh))
29075 +               goto out;
29076 +
29077 +       err = 0;
29078 +       wh->wh_bindex = bindex;
29079 +       if (shwh)
29080 +               au_shwh_init_wh(wh, ino, d_type);
29081 +       str = &wh->wh_str;
29082 +       str->len = nlen;
29083 +       memcpy(str->name, name, nlen);
29084 +       hlist_add_head(&wh->wh_hash, au_name_hash(whlist, name, nlen));
29085 +       /* smp_mb(); */
29086 +
29087 +out:
29088 +       return err;
29089 +}
29090 +
29091 +static int append_deblk(struct au_vdir *vdir)
29092 +{
29093 +       int err;
29094 +       unsigned long ul;
29095 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
29096 +       union au_vdir_deblk_p p, deblk_end;
29097 +       unsigned char **o;
29098 +
29099 +       err = -ENOMEM;
29100 +       o = krealloc(vdir->vd_deblk, sizeof(*o) * (vdir->vd_nblk + 1),
29101 +                    GFP_NOFS);
29102 +       if (unlikely(!o))
29103 +               goto out;
29104 +
29105 +       vdir->vd_deblk = o;
29106 +       p.deblk = kmalloc(deblk_sz, GFP_NOFS);
29107 +       if (p.deblk) {
29108 +               ul = vdir->vd_nblk++;
29109 +               vdir->vd_deblk[ul] = p.deblk;
29110 +               vdir->vd_last.ul = ul;
29111 +               vdir->vd_last.p.deblk = p.deblk;
29112 +               deblk_end.deblk = p.deblk + deblk_sz;
29113 +               err = set_deblk_end(&p, &deblk_end);
29114 +       }
29115 +
29116 +out:
29117 +       return err;
29118 +}
29119 +
29120 +static int append_de(struct au_vdir *vdir, char *name, int nlen, ino_t ino,
29121 +                    unsigned int d_type, struct au_nhash *delist)
29122 +{
29123 +       int err;
29124 +       unsigned int sz;
29125 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
29126 +       union au_vdir_deblk_p p, *room, deblk_end;
29127 +       struct au_vdir_dehstr *dehstr;
29128 +
29129 +       p.deblk = last_deblk(vdir);
29130 +       deblk_end.deblk = p.deblk + deblk_sz;
29131 +       room = &vdir->vd_last.p;
29132 +       AuDebugOn(room->deblk < p.deblk || deblk_end.deblk <= room->deblk
29133 +                 || !is_deblk_end(room, &deblk_end));
29134 +
29135 +       sz = calc_size(nlen);
29136 +       if (unlikely(sz > deblk_end.deblk - room->deblk)) {
29137 +               err = append_deblk(vdir);
29138 +               if (unlikely(err))
29139 +                       goto out;
29140 +
29141 +               p.deblk = last_deblk(vdir);
29142 +               deblk_end.deblk = p.deblk + deblk_sz;
29143 +               /* smp_mb(); */
29144 +               AuDebugOn(room->deblk != p.deblk);
29145 +       }
29146 +
29147 +       err = -ENOMEM;
29148 +       dehstr = au_cache_alloc_vdir_dehstr();
29149 +       if (unlikely(!dehstr))
29150 +               goto out;
29151 +
29152 +       dehstr->str = &room->de->de_str;
29153 +       hlist_add_head(&dehstr->hash, au_name_hash(delist, name, nlen));
29154 +       room->de->de_ino = ino;
29155 +       room->de->de_type = d_type;
29156 +       room->de->de_str.len = nlen;
29157 +       memcpy(room->de->de_str.name, name, nlen);
29158 +
29159 +       err = 0;
29160 +       room->deblk += sz;
29161 +       if (unlikely(set_deblk_end(room, &deblk_end)))
29162 +               err = append_deblk(vdir);
29163 +       /* smp_mb(); */
29164 +
29165 +out:
29166 +       return err;
29167 +}
29168 +
29169 +/* ---------------------------------------------------------------------- */
29170 +
29171 +void au_vdir_free(struct au_vdir *vdir)
29172 +{
29173 +       unsigned char **deblk;
29174 +
29175 +       deblk = vdir->vd_deblk;
29176 +       while (vdir->vd_nblk--)
29177 +               kfree(*deblk++);
29178 +       kfree(vdir->vd_deblk);
29179 +       au_cache_free_vdir(vdir);
29180 +}
29181 +
29182 +static struct au_vdir *alloc_vdir(struct file *file)
29183 +{
29184 +       struct au_vdir *vdir;
29185 +       struct super_block *sb;
29186 +       int err;
29187 +
29188 +       sb = file->f_dentry->d_sb;
29189 +       SiMustAnyLock(sb);
29190 +
29191 +       err = -ENOMEM;
29192 +       vdir = au_cache_alloc_vdir();
29193 +       if (unlikely(!vdir))
29194 +               goto out;
29195 +
29196 +       vdir->vd_deblk = kzalloc(sizeof(*vdir->vd_deblk), GFP_NOFS);
29197 +       if (unlikely(!vdir->vd_deblk))
29198 +               goto out_free;
29199 +
29200 +       vdir->vd_deblk_sz = au_sbi(sb)->si_rdblk;
29201 +       if (!vdir->vd_deblk_sz) {
29202 +               /* estimate the apropriate size for deblk */
29203 +               vdir->vd_deblk_sz = au_dir_size(file, /*dentry*/NULL);
29204 +               /* pr_info("vd_deblk_sz %u\n", vdir->vd_deblk_sz); */
29205 +       }
29206 +       vdir->vd_nblk = 0;
29207 +       vdir->vd_version = 0;
29208 +       vdir->vd_jiffy = 0;
29209 +       err = append_deblk(vdir);
29210 +       if (!err)
29211 +               return vdir; /* success */
29212 +
29213 +       kfree(vdir->vd_deblk);
29214 +
29215 +out_free:
29216 +       au_cache_free_vdir(vdir);
29217 +out:
29218 +       vdir = ERR_PTR(err);
29219 +       return vdir;
29220 +}
29221 +
29222 +static int reinit_vdir(struct au_vdir *vdir)
29223 +{
29224 +       int err;
29225 +       union au_vdir_deblk_p p, deblk_end;
29226 +
29227 +       while (vdir->vd_nblk > 1) {
29228 +               kfree(vdir->vd_deblk[vdir->vd_nblk - 1]);
29229 +               /* vdir->vd_deblk[vdir->vd_nblk - 1] = NULL; */
29230 +               vdir->vd_nblk--;
29231 +       }
29232 +       p.deblk = vdir->vd_deblk[0];
29233 +       deblk_end.deblk = p.deblk + vdir->vd_deblk_sz;
29234 +       err = set_deblk_end(&p, &deblk_end);
29235 +       /* keep vd_dblk_sz */
29236 +       vdir->vd_last.ul = 0;
29237 +       vdir->vd_last.p.deblk = vdir->vd_deblk[0];
29238 +       vdir->vd_version = 0;
29239 +       vdir->vd_jiffy = 0;
29240 +       /* smp_mb(); */
29241 +       return err;
29242 +}
29243 +
29244 +/* ---------------------------------------------------------------------- */
29245 +
29246 +#define AuFillVdir_CALLED      1
29247 +#define AuFillVdir_WHABLE      (1 << 1)
29248 +#define AuFillVdir_SHWH                (1 << 2)
29249 +#define au_ftest_fillvdir(flags, name) ((flags) & AuFillVdir_##name)
29250 +#define au_fset_fillvdir(flags, name) \
29251 +       do { (flags) |= AuFillVdir_##name; } while (0)
29252 +#define au_fclr_fillvdir(flags, name) \
29253 +       do { (flags) &= ~AuFillVdir_##name; } while (0)
29254 +
29255 +#ifndef CONFIG_AUFS_SHWH
29256 +#undef AuFillVdir_SHWH
29257 +#define AuFillVdir_SHWH                0
29258 +#endif
29259 +
29260 +struct fillvdir_arg {
29261 +       struct dir_context      ctx;
29262 +       struct file             *file;
29263 +       struct au_vdir          *vdir;
29264 +       struct au_nhash         delist;
29265 +       struct au_nhash         whlist;
29266 +       aufs_bindex_t           bindex;
29267 +       unsigned int            flags;
29268 +       int                     err;
29269 +};
29270 +
29271 +static int fillvdir(struct dir_context *ctx, const char *__name, int nlen,
29272 +                   loff_t offset __maybe_unused, u64 h_ino,
29273 +                   unsigned int d_type)
29274 +{
29275 +       struct fillvdir_arg *arg = container_of(ctx, struct fillvdir_arg, ctx);
29276 +       char *name = (void *)__name;
29277 +       struct super_block *sb;
29278 +       ino_t ino;
29279 +       const unsigned char shwh = !!au_ftest_fillvdir(arg->flags, SHWH);
29280 +
29281 +       arg->err = 0;
29282 +       sb = arg->file->f_dentry->d_sb;
29283 +       au_fset_fillvdir(arg->flags, CALLED);
29284 +       /* smp_mb(); */
29285 +       if (nlen <= AUFS_WH_PFX_LEN
29286 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
29287 +               if (test_known(&arg->delist, name, nlen)
29288 +                   || au_nhash_test_known_wh(&arg->whlist, name, nlen))
29289 +                       goto out; /* already exists or whiteouted */
29290 +
29291 +               sb = arg->file->f_dentry->d_sb;
29292 +               arg->err = au_ino(sb, arg->bindex, h_ino, d_type, &ino);
29293 +               if (!arg->err) {
29294 +                       if (unlikely(nlen > AUFS_MAX_NAMELEN))
29295 +                               d_type = DT_UNKNOWN;
29296 +                       arg->err = append_de(arg->vdir, name, nlen, ino,
29297 +                                            d_type, &arg->delist);
29298 +               }
29299 +       } else if (au_ftest_fillvdir(arg->flags, WHABLE)) {
29300 +               name += AUFS_WH_PFX_LEN;
29301 +               nlen -= AUFS_WH_PFX_LEN;
29302 +               if (au_nhash_test_known_wh(&arg->whlist, name, nlen))
29303 +                       goto out; /* already whiteouted */
29304 +
29305 +               if (shwh)
29306 +                       arg->err = au_wh_ino(sb, arg->bindex, h_ino, d_type,
29307 +                                            &ino);
29308 +               if (!arg->err) {
29309 +                       if (nlen <= AUFS_MAX_NAMELEN + AUFS_WH_PFX_LEN)
29310 +                               d_type = DT_UNKNOWN;
29311 +                       arg->err = au_nhash_append_wh
29312 +                               (&arg->whlist, name, nlen, ino, d_type,
29313 +                                arg->bindex, shwh);
29314 +               }
29315 +       }
29316 +
29317 +out:
29318 +       if (!arg->err)
29319 +               arg->vdir->vd_jiffy = jiffies;
29320 +       /* smp_mb(); */
29321 +       AuTraceErr(arg->err);
29322 +       return arg->err;
29323 +}
29324 +
29325 +static int au_handle_shwh(struct super_block *sb, struct au_vdir *vdir,
29326 +                         struct au_nhash *whlist, struct au_nhash *delist)
29327 +{
29328 +#ifdef CONFIG_AUFS_SHWH
29329 +       int err;
29330 +       unsigned int nh, u;
29331 +       struct hlist_head *head;
29332 +       struct au_vdir_wh *pos;
29333 +       struct hlist_node *n;
29334 +       char *p, *o;
29335 +       struct au_vdir_destr *destr;
29336 +
29337 +       AuDebugOn(!au_opt_test(au_mntflags(sb), SHWH));
29338 +
29339 +       err = -ENOMEM;
29340 +       o = p = (void *)__get_free_page(GFP_NOFS);
29341 +       if (unlikely(!p))
29342 +               goto out;
29343 +
29344 +       err = 0;
29345 +       nh = whlist->nh_num;
29346 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
29347 +       p += AUFS_WH_PFX_LEN;
29348 +       for (u = 0; u < nh; u++) {
29349 +               head = whlist->nh_head + u;
29350 +               hlist_for_each_entry_safe(pos, n, head, wh_hash) {
29351 +                       destr = &pos->wh_str;
29352 +                       memcpy(p, destr->name, destr->len);
29353 +                       err = append_de(vdir, o, destr->len + AUFS_WH_PFX_LEN,
29354 +                                       pos->wh_ino, pos->wh_type, delist);
29355 +                       if (unlikely(err))
29356 +                               break;
29357 +               }
29358 +       }
29359 +
29360 +       free_page((unsigned long)o);
29361 +
29362 +out:
29363 +       AuTraceErr(err);
29364 +       return err;
29365 +#else
29366 +       return 0;
29367 +#endif
29368 +}
29369 +
29370 +static int au_do_read_vdir(struct fillvdir_arg *arg)
29371 +{
29372 +       int err;
29373 +       unsigned int rdhash;
29374 +       loff_t offset;
29375 +       aufs_bindex_t bend, bindex, bstart;
29376 +       unsigned char shwh;
29377 +       struct file *hf, *file;
29378 +       struct super_block *sb;
29379 +
29380 +       file = arg->file;
29381 +       sb = file->f_dentry->d_sb;
29382 +       SiMustAnyLock(sb);
29383 +
29384 +       rdhash = au_sbi(sb)->si_rdhash;
29385 +       if (!rdhash)
29386 +               rdhash = au_rdhash_est(au_dir_size(file, /*dentry*/NULL));
29387 +       err = au_nhash_alloc(&arg->delist, rdhash, GFP_NOFS);
29388 +       if (unlikely(err))
29389 +               goto out;
29390 +       err = au_nhash_alloc(&arg->whlist, rdhash, GFP_NOFS);
29391 +       if (unlikely(err))
29392 +               goto out_delist;
29393 +
29394 +       err = 0;
29395 +       arg->flags = 0;
29396 +       shwh = 0;
29397 +       if (au_opt_test(au_mntflags(sb), SHWH)) {
29398 +               shwh = 1;
29399 +               au_fset_fillvdir(arg->flags, SHWH);
29400 +       }
29401 +       bstart = au_fbstart(file);
29402 +       bend = au_fbend_dir(file);
29403 +       for (bindex = bstart; !err && bindex <= bend; bindex++) {
29404 +               hf = au_hf_dir(file, bindex);
29405 +               if (!hf)
29406 +                       continue;
29407 +
29408 +               offset = vfsub_llseek(hf, 0, SEEK_SET);
29409 +               err = offset;
29410 +               if (unlikely(offset))
29411 +                       break;
29412 +
29413 +               arg->bindex = bindex;
29414 +               au_fclr_fillvdir(arg->flags, WHABLE);
29415 +               if (shwh
29416 +                   || (bindex != bend
29417 +                       && au_br_whable(au_sbr_perm(sb, bindex))))
29418 +                       au_fset_fillvdir(arg->flags, WHABLE);
29419 +               do {
29420 +                       arg->err = 0;
29421 +                       au_fclr_fillvdir(arg->flags, CALLED);
29422 +                       /* smp_mb(); */
29423 +                       err = vfsub_iterate_dir(hf, &arg->ctx);
29424 +                       if (err >= 0)
29425 +                               err = arg->err;
29426 +               } while (!err && au_ftest_fillvdir(arg->flags, CALLED));
29427 +
29428 +               /*
29429 +                * dir_relax() may be good for concurrency, but aufs should not
29430 +                * use it since it will cause a lockdep problem.
29431 +                */
29432 +       }
29433 +
29434 +       if (!err && shwh)
29435 +               err = au_handle_shwh(sb, arg->vdir, &arg->whlist, &arg->delist);
29436 +
29437 +       au_nhash_wh_free(&arg->whlist);
29438 +
29439 +out_delist:
29440 +       au_nhash_de_free(&arg->delist);
29441 +out:
29442 +       return err;
29443 +}
29444 +
29445 +static int read_vdir(struct file *file, int may_read)
29446 +{
29447 +       int err;
29448 +       unsigned long expire;
29449 +       unsigned char do_read;
29450 +       struct fillvdir_arg arg = {
29451 +               .ctx = {
29452 +                       .actor = au_diractor(fillvdir)
29453 +               }
29454 +       };
29455 +       struct inode *inode;
29456 +       struct au_vdir *vdir, *allocated;
29457 +
29458 +       err = 0;
29459 +       inode = file_inode(file);
29460 +       IMustLock(inode);
29461 +       SiMustAnyLock(inode->i_sb);
29462 +
29463 +       allocated = NULL;
29464 +       do_read = 0;
29465 +       expire = au_sbi(inode->i_sb)->si_rdcache;
29466 +       vdir = au_ivdir(inode);
29467 +       if (!vdir) {
29468 +               do_read = 1;
29469 +               vdir = alloc_vdir(file);
29470 +               err = PTR_ERR(vdir);
29471 +               if (IS_ERR(vdir))
29472 +                       goto out;
29473 +               err = 0;
29474 +               allocated = vdir;
29475 +       } else if (may_read
29476 +                  && (inode->i_version != vdir->vd_version
29477 +                      || time_after(jiffies, vdir->vd_jiffy + expire))) {
29478 +               do_read = 1;
29479 +               err = reinit_vdir(vdir);
29480 +               if (unlikely(err))
29481 +                       goto out;
29482 +       }
29483 +
29484 +       if (!do_read)
29485 +               return 0; /* success */
29486 +
29487 +       arg.file = file;
29488 +       arg.vdir = vdir;
29489 +       err = au_do_read_vdir(&arg);
29490 +       if (!err) {
29491 +               /* file->f_pos = 0; */ /* todo: ctx->pos? */
29492 +               vdir->vd_version = inode->i_version;
29493 +               vdir->vd_last.ul = 0;
29494 +               vdir->vd_last.p.deblk = vdir->vd_deblk[0];
29495 +               if (allocated)
29496 +                       au_set_ivdir(inode, allocated);
29497 +       } else if (allocated)
29498 +               au_vdir_free(allocated);
29499 +
29500 +out:
29501 +       return err;
29502 +}
29503 +
29504 +static int copy_vdir(struct au_vdir *tgt, struct au_vdir *src)
29505 +{
29506 +       int err, rerr;
29507 +       unsigned long ul, n;
29508 +       const unsigned int deblk_sz = src->vd_deblk_sz;
29509 +
29510 +       AuDebugOn(tgt->vd_nblk != 1);
29511 +
29512 +       err = -ENOMEM;
29513 +       if (tgt->vd_nblk < src->vd_nblk) {
29514 +               unsigned char **p;
29515 +
29516 +               p = krealloc(tgt->vd_deblk, sizeof(*p) * src->vd_nblk,
29517 +                            GFP_NOFS);
29518 +               if (unlikely(!p))
29519 +                       goto out;
29520 +               tgt->vd_deblk = p;
29521 +       }
29522 +
29523 +       if (tgt->vd_deblk_sz != deblk_sz) {
29524 +               unsigned char *p;
29525 +
29526 +               tgt->vd_deblk_sz = deblk_sz;
29527 +               p = krealloc(tgt->vd_deblk[0], deblk_sz, GFP_NOFS);
29528 +               if (unlikely(!p))
29529 +                       goto out;
29530 +               tgt->vd_deblk[0] = p;
29531 +       }
29532 +       memcpy(tgt->vd_deblk[0], src->vd_deblk[0], deblk_sz);
29533 +       tgt->vd_version = src->vd_version;
29534 +       tgt->vd_jiffy = src->vd_jiffy;
29535 +
29536 +       n = src->vd_nblk;
29537 +       for (ul = 1; ul < n; ul++) {
29538 +               tgt->vd_deblk[ul] = kmemdup(src->vd_deblk[ul], deblk_sz,
29539 +                                           GFP_NOFS);
29540 +               if (unlikely(!tgt->vd_deblk[ul]))
29541 +                       goto out;
29542 +               tgt->vd_nblk++;
29543 +       }
29544 +       tgt->vd_nblk = n;
29545 +       tgt->vd_last.ul = tgt->vd_last.ul;
29546 +       tgt->vd_last.p.deblk = tgt->vd_deblk[tgt->vd_last.ul];
29547 +       tgt->vd_last.p.deblk += src->vd_last.p.deblk
29548 +               - src->vd_deblk[src->vd_last.ul];
29549 +       /* smp_mb(); */
29550 +       return 0; /* success */
29551 +
29552 +out:
29553 +       rerr = reinit_vdir(tgt);
29554 +       BUG_ON(rerr);
29555 +       return err;
29556 +}
29557 +
29558 +int au_vdir_init(struct file *file)
29559 +{
29560 +       int err;
29561 +       struct inode *inode;
29562 +       struct au_vdir *vdir_cache, *allocated;
29563 +
29564 +       /* test file->f_pos here instead of ctx->pos */
29565 +       err = read_vdir(file, !file->f_pos);
29566 +       if (unlikely(err))
29567 +               goto out;
29568 +
29569 +       allocated = NULL;
29570 +       vdir_cache = au_fvdir_cache(file);
29571 +       if (!vdir_cache) {
29572 +               vdir_cache = alloc_vdir(file);
29573 +               err = PTR_ERR(vdir_cache);
29574 +               if (IS_ERR(vdir_cache))
29575 +                       goto out;
29576 +               allocated = vdir_cache;
29577 +       } else if (!file->f_pos && vdir_cache->vd_version != file->f_version) {
29578 +               /* test file->f_pos here instead of ctx->pos */
29579 +               err = reinit_vdir(vdir_cache);
29580 +               if (unlikely(err))
29581 +                       goto out;
29582 +       } else
29583 +               return 0; /* success */
29584 +
29585 +       inode = file_inode(file);
29586 +       err = copy_vdir(vdir_cache, au_ivdir(inode));
29587 +       if (!err) {
29588 +               file->f_version = inode->i_version;
29589 +               if (allocated)
29590 +                       au_set_fvdir_cache(file, allocated);
29591 +       } else if (allocated)
29592 +               au_vdir_free(allocated);
29593 +
29594 +out:
29595 +       return err;
29596 +}
29597 +
29598 +static loff_t calc_offset(struct au_vdir *vdir)
29599 +{
29600 +       loff_t offset;
29601 +       union au_vdir_deblk_p p;
29602 +
29603 +       p.deblk = vdir->vd_deblk[vdir->vd_last.ul];
29604 +       offset = vdir->vd_last.p.deblk - p.deblk;
29605 +       offset += vdir->vd_deblk_sz * vdir->vd_last.ul;
29606 +       return offset;
29607 +}
29608 +
29609 +/* returns true or false */
29610 +static int seek_vdir(struct file *file, struct dir_context *ctx)
29611 +{
29612 +       int valid;
29613 +       unsigned int deblk_sz;
29614 +       unsigned long ul, n;
29615 +       loff_t offset;
29616 +       union au_vdir_deblk_p p, deblk_end;
29617 +       struct au_vdir *vdir_cache;
29618 +
29619 +       valid = 1;
29620 +       vdir_cache = au_fvdir_cache(file);
29621 +       offset = calc_offset(vdir_cache);
29622 +       AuDbg("offset %lld\n", offset);
29623 +       if (ctx->pos == offset)
29624 +               goto out;
29625 +
29626 +       vdir_cache->vd_last.ul = 0;
29627 +       vdir_cache->vd_last.p.deblk = vdir_cache->vd_deblk[0];
29628 +       if (!ctx->pos)
29629 +               goto out;
29630 +
29631 +       valid = 0;
29632 +       deblk_sz = vdir_cache->vd_deblk_sz;
29633 +       ul = div64_u64(ctx->pos, deblk_sz);
29634 +       AuDbg("ul %lu\n", ul);
29635 +       if (ul >= vdir_cache->vd_nblk)
29636 +               goto out;
29637 +
29638 +       n = vdir_cache->vd_nblk;
29639 +       for (; ul < n; ul++) {
29640 +               p.deblk = vdir_cache->vd_deblk[ul];
29641 +               deblk_end.deblk = p.deblk + deblk_sz;
29642 +               offset = ul;
29643 +               offset *= deblk_sz;
29644 +               while (!is_deblk_end(&p, &deblk_end) && offset < ctx->pos) {
29645 +                       unsigned int l;
29646 +
29647 +                       l = calc_size(p.de->de_str.len);
29648 +                       offset += l;
29649 +                       p.deblk += l;
29650 +               }
29651 +               if (!is_deblk_end(&p, &deblk_end)) {
29652 +                       valid = 1;
29653 +                       vdir_cache->vd_last.ul = ul;
29654 +                       vdir_cache->vd_last.p = p;
29655 +                       break;
29656 +               }
29657 +       }
29658 +
29659 +out:
29660 +       /* smp_mb(); */
29661 +       AuTraceErr(!valid);
29662 +       return valid;
29663 +}
29664 +
29665 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx)
29666 +{
29667 +       unsigned int l, deblk_sz;
29668 +       union au_vdir_deblk_p deblk_end;
29669 +       struct au_vdir *vdir_cache;
29670 +       struct au_vdir_de *de;
29671 +
29672 +       vdir_cache = au_fvdir_cache(file);
29673 +       if (!seek_vdir(file, ctx))
29674 +               return 0;
29675 +
29676 +       deblk_sz = vdir_cache->vd_deblk_sz;
29677 +       while (1) {
29678 +               deblk_end.deblk = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
29679 +               deblk_end.deblk += deblk_sz;
29680 +               while (!is_deblk_end(&vdir_cache->vd_last.p, &deblk_end)) {
29681 +                       de = vdir_cache->vd_last.p.de;
29682 +                       AuDbg("%.*s, off%lld, i%lu, dt%d\n",
29683 +                             de->de_str.len, de->de_str.name, ctx->pos,
29684 +                             (unsigned long)de->de_ino, de->de_type);
29685 +                       if (unlikely(!dir_emit(ctx, de->de_str.name,
29686 +                                              de->de_str.len, de->de_ino,
29687 +                                              de->de_type))) {
29688 +                               /* todo: ignore the error caused by udba? */
29689 +                               /* return err; */
29690 +                               return 0;
29691 +                       }
29692 +
29693 +                       l = calc_size(de->de_str.len);
29694 +                       vdir_cache->vd_last.p.deblk += l;
29695 +                       ctx->pos += l;
29696 +               }
29697 +               if (vdir_cache->vd_last.ul < vdir_cache->vd_nblk - 1) {
29698 +                       vdir_cache->vd_last.ul++;
29699 +                       vdir_cache->vd_last.p.deblk
29700 +                               = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
29701 +                       ctx->pos = deblk_sz * vdir_cache->vd_last.ul;
29702 +                       continue;
29703 +               }
29704 +               break;
29705 +       }
29706 +
29707 +       /* smp_mb(); */
29708 +       return 0;
29709 +}
29710 diff -urN /usr/share/empty/fs/aufs/vfsub.c linux/fs/aufs/vfsub.c
29711 --- /usr/share/empty/fs/aufs/vfsub.c    1970-01-01 01:00:00.000000000 +0100
29712 +++ linux/fs/aufs/vfsub.c       2015-01-25 13:00:38.634380408 +0100
29713 @@ -0,0 +1,796 @@
29714 +/*
29715 + * Copyright (C) 2005-2014 Junjiro R. Okajima
29716 + *
29717 + * This program, aufs is free software; you can redistribute it and/or modify
29718 + * it under the terms of the GNU General Public License as published by
29719 + * the Free Software Foundation; either version 2 of the License, or
29720 + * (at your option) any later version.
29721 + *
29722 + * This program is distributed in the hope that it will be useful,
29723 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29724 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29725 + * GNU General Public License for more details.
29726 + *
29727 + * You should have received a copy of the GNU General Public License
29728 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29729 + */
29730 +
29731 +/*
29732 + * sub-routines for VFS
29733 + */
29734 +
29735 +#include <linux/ima.h>
29736 +#include <linux/namei.h>
29737 +#include <linux/security.h>
29738 +#include <linux/splice.h>
29739 +#include "aufs.h"
29740 +
29741 +int vfsub_update_h_iattr(struct path *h_path, int *did)
29742 +{
29743 +       int err;
29744 +       struct kstat st;
29745 +       struct super_block *h_sb;
29746 +
29747 +       /* for remote fs, leave work for its getattr or d_revalidate */
29748 +       /* for bad i_attr fs, handle them in aufs_getattr() */
29749 +       /* still some fs may acquire i_mutex. we need to skip them */
29750 +       err = 0;
29751 +       if (!did)
29752 +               did = &err;
29753 +       h_sb = h_path->dentry->d_sb;
29754 +       *did = (!au_test_fs_remote(h_sb) && au_test_fs_refresh_iattr(h_sb));
29755 +       if (*did)
29756 +               err = vfs_getattr(h_path, &st);
29757 +
29758 +       return err;
29759 +}
29760 +
29761 +/* ---------------------------------------------------------------------- */
29762 +
29763 +struct file *vfsub_dentry_open(struct path *path, int flags)
29764 +{
29765 +       struct file *file;
29766 +
29767 +       file = dentry_open(path, flags /* | __FMODE_NONOTIFY */,
29768 +                          current_cred());
29769 +       if (!IS_ERR_OR_NULL(file)
29770 +           && (file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
29771 +               i_readcount_inc(path->dentry->d_inode);
29772 +
29773 +       return file;
29774 +}
29775 +
29776 +struct file *vfsub_filp_open(const char *path, int oflags, int mode)
29777 +{
29778 +       struct file *file;
29779 +
29780 +       lockdep_off();
29781 +       file = filp_open(path,
29782 +                        oflags /* | __FMODE_NONOTIFY */,
29783 +                        mode);
29784 +       lockdep_on();
29785 +       if (IS_ERR(file))
29786 +               goto out;
29787 +       vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
29788 +
29789 +out:
29790 +       return file;
29791 +}
29792 +
29793 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path)
29794 +{
29795 +       int err;
29796 +
29797 +       err = kern_path(name, flags, path);
29798 +       if (!err && path->dentry->d_inode)
29799 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
29800 +       return err;
29801 +}
29802 +
29803 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
29804 +                                   int len)
29805 +{
29806 +       struct path path = {
29807 +               .mnt = NULL
29808 +       };
29809 +
29810 +       /* VFS checks it too, but by WARN_ON_ONCE() */
29811 +       IMustLock(parent->d_inode);
29812 +
29813 +       path.dentry = lookup_one_len(name, parent, len);
29814 +       if (IS_ERR(path.dentry))
29815 +               goto out;
29816 +       if (path.dentry->d_inode)
29817 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
29818 +
29819 +out:
29820 +       AuTraceErrPtr(path.dentry);
29821 +       return path.dentry;
29822 +}
29823 +
29824 +void vfsub_call_lkup_one(void *args)
29825 +{
29826 +       struct vfsub_lkup_one_args *a = args;
29827 +       *a->errp = vfsub_lkup_one(a->name, a->parent);
29828 +}
29829 +
29830 +/* ---------------------------------------------------------------------- */
29831 +
29832 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
29833 +                                struct dentry *d2, struct au_hinode *hdir2)
29834 +{
29835 +       struct dentry *d;
29836 +
29837 +       lockdep_off();
29838 +       d = lock_rename(d1, d2);
29839 +       lockdep_on();
29840 +       au_hn_suspend(hdir1);
29841 +       if (hdir1 != hdir2)
29842 +               au_hn_suspend(hdir2);
29843 +
29844 +       return d;
29845 +}
29846 +
29847 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
29848 +                        struct dentry *d2, struct au_hinode *hdir2)
29849 +{
29850 +       au_hn_resume(hdir1);
29851 +       if (hdir1 != hdir2)
29852 +               au_hn_resume(hdir2);
29853 +       lockdep_off();
29854 +       unlock_rename(d1, d2);
29855 +       lockdep_on();
29856 +}
29857 +
29858 +/* ---------------------------------------------------------------------- */
29859 +
29860 +int vfsub_create(struct inode *dir, struct path *path, int mode, bool want_excl)
29861 +{
29862 +       int err;
29863 +       struct dentry *d;
29864 +
29865 +       IMustLock(dir);
29866 +
29867 +       d = path->dentry;
29868 +       path->dentry = d->d_parent;
29869 +       err = security_path_mknod(path, d, mode, 0);
29870 +       path->dentry = d;
29871 +       if (unlikely(err))
29872 +               goto out;
29873 +
29874 +       lockdep_off();
29875 +       err = vfs_create(dir, path->dentry, mode, want_excl);
29876 +       lockdep_on();
29877 +       if (!err) {
29878 +               struct path tmp = *path;
29879 +               int did;
29880 +
29881 +               vfsub_update_h_iattr(&tmp, &did);
29882 +               if (did) {
29883 +                       tmp.dentry = path->dentry->d_parent;
29884 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
29885 +               }
29886 +               /*ignore*/
29887 +       }
29888 +
29889 +out:
29890 +       return err;
29891 +}
29892 +
29893 +int vfsub_symlink(struct inode *dir, struct path *path, const char *symname)
29894 +{
29895 +       int err;
29896 +       struct dentry *d;
29897 +
29898 +       IMustLock(dir);
29899 +
29900 +       d = path->dentry;
29901 +       path->dentry = d->d_parent;
29902 +       err = security_path_symlink(path, d, symname);
29903 +       path->dentry = d;
29904 +       if (unlikely(err))
29905 +               goto out;
29906 +
29907 +       lockdep_off();
29908 +       err = vfs_symlink(dir, path->dentry, symname);
29909 +       lockdep_on();
29910 +       if (!err) {
29911 +               struct path tmp = *path;
29912 +               int did;
29913 +
29914 +               vfsub_update_h_iattr(&tmp, &did);
29915 +               if (did) {
29916 +                       tmp.dentry = path->dentry->d_parent;
29917 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
29918 +               }
29919 +               /*ignore*/
29920 +       }
29921 +
29922 +out:
29923 +       return err;
29924 +}
29925 +
29926 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev)
29927 +{
29928 +       int err;
29929 +       struct dentry *d;
29930 +
29931 +       IMustLock(dir);
29932 +
29933 +       d = path->dentry;
29934 +       path->dentry = d->d_parent;
29935 +       err = security_path_mknod(path, d, mode, new_encode_dev(dev));
29936 +       path->dentry = d;
29937 +       if (unlikely(err))
29938 +               goto out;
29939 +
29940 +       lockdep_off();
29941 +       err = vfs_mknod(dir, path->dentry, mode, dev);
29942 +       lockdep_on();
29943 +       if (!err) {
29944 +               struct path tmp = *path;
29945 +               int did;
29946 +
29947 +               vfsub_update_h_iattr(&tmp, &did);
29948 +               if (did) {
29949 +                       tmp.dentry = path->dentry->d_parent;
29950 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
29951 +               }
29952 +               /*ignore*/
29953 +       }
29954 +
29955 +out:
29956 +       return err;
29957 +}
29958 +
29959 +static int au_test_nlink(struct inode *inode)
29960 +{
29961 +       const unsigned int link_max = UINT_MAX >> 1; /* rough margin */
29962 +
29963 +       if (!au_test_fs_no_limit_nlink(inode->i_sb)
29964 +           || inode->i_nlink < link_max)
29965 +               return 0;
29966 +       return -EMLINK;
29967 +}
29968 +
29969 +int vfsub_link(struct dentry *src_dentry, struct inode *dir, struct path *path,
29970 +              struct inode **delegated_inode)
29971 +{
29972 +       int err;
29973 +       struct dentry *d;
29974 +
29975 +       IMustLock(dir);
29976 +
29977 +       err = au_test_nlink(src_dentry->d_inode);
29978 +       if (unlikely(err))
29979 +               return err;
29980 +
29981 +       /* we don't call may_linkat() */
29982 +       d = path->dentry;
29983 +       path->dentry = d->d_parent;
29984 +       err = security_path_link(src_dentry, path, d);
29985 +       path->dentry = d;
29986 +       if (unlikely(err))
29987 +               goto out;
29988 +
29989 +       lockdep_off();
29990 +       err = vfs_link(src_dentry, dir, path->dentry, delegated_inode);
29991 +       lockdep_on();
29992 +       if (!err) {
29993 +               struct path tmp = *path;
29994 +               int did;
29995 +
29996 +               /* fuse has different memory inode for the same inumber */
29997 +               vfsub_update_h_iattr(&tmp, &did);
29998 +               if (did) {
29999 +                       tmp.dentry = path->dentry->d_parent;
30000 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30001 +                       tmp.dentry = src_dentry;
30002 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30003 +               }
30004 +               /*ignore*/
30005 +       }
30006 +
30007 +out:
30008 +       return err;
30009 +}
30010 +
30011 +int vfsub_rename(struct inode *src_dir, struct dentry *src_dentry,
30012 +                struct inode *dir, struct path *path,
30013 +                struct inode **delegated_inode)
30014 +{
30015 +       int err;
30016 +       struct path tmp = {
30017 +               .mnt    = path->mnt
30018 +       };
30019 +       struct dentry *d;
30020 +
30021 +       IMustLock(dir);
30022 +       IMustLock(src_dir);
30023 +
30024 +       d = path->dentry;
30025 +       path->dentry = d->d_parent;
30026 +       tmp.dentry = src_dentry->d_parent;
30027 +       err = security_path_rename(&tmp, src_dentry, path, d, /*flags*/0);
30028 +       path->dentry = d;
30029 +       if (unlikely(err))
30030 +               goto out;
30031 +
30032 +       lockdep_off();
30033 +       err = vfs_rename(src_dir, src_dentry, dir, path->dentry,
30034 +                        delegated_inode, /*flags*/0);
30035 +       lockdep_on();
30036 +       if (!err) {
30037 +               int did;
30038 +
30039 +               tmp.dentry = d->d_parent;
30040 +               vfsub_update_h_iattr(&tmp, &did);
30041 +               if (did) {
30042 +                       tmp.dentry = src_dentry;
30043 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30044 +                       tmp.dentry = src_dentry->d_parent;
30045 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30046 +               }
30047 +               /*ignore*/
30048 +       }
30049 +
30050 +out:
30051 +       return err;
30052 +}
30053 +
30054 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode)
30055 +{
30056 +       int err;
30057 +       struct dentry *d;
30058 +
30059 +       IMustLock(dir);
30060 +
30061 +       d = path->dentry;
30062 +       path->dentry = d->d_parent;
30063 +       err = security_path_mkdir(path, d, mode);
30064 +       path->dentry = d;
30065 +       if (unlikely(err))
30066 +               goto out;
30067 +
30068 +       lockdep_off();
30069 +       err = vfs_mkdir(dir, path->dentry, mode);
30070 +       lockdep_on();
30071 +       if (!err) {
30072 +               struct path tmp = *path;
30073 +               int did;
30074 +
30075 +               vfsub_update_h_iattr(&tmp, &did);
30076 +               if (did) {
30077 +                       tmp.dentry = path->dentry->d_parent;
30078 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30079 +               }
30080 +               /*ignore*/
30081 +       }
30082 +
30083 +out:
30084 +       return err;
30085 +}
30086 +
30087 +int vfsub_rmdir(struct inode *dir, struct path *path)
30088 +{
30089 +       int err;
30090 +       struct dentry *d;
30091 +
30092 +       IMustLock(dir);
30093 +
30094 +       d = path->dentry;
30095 +       path->dentry = d->d_parent;
30096 +       err = security_path_rmdir(path, d);
30097 +       path->dentry = d;
30098 +       if (unlikely(err))
30099 +               goto out;
30100 +
30101 +       lockdep_off();
30102 +       err = vfs_rmdir(dir, path->dentry);
30103 +       lockdep_on();
30104 +       if (!err) {
30105 +               struct path tmp = {
30106 +                       .dentry = path->dentry->d_parent,
30107 +                       .mnt    = path->mnt
30108 +               };
30109 +
30110 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
30111 +       }
30112 +
30113 +out:
30114 +       return err;
30115 +}
30116 +
30117 +/* ---------------------------------------------------------------------- */
30118 +
30119 +/* todo: support mmap_sem? */
30120 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
30121 +                    loff_t *ppos)
30122 +{
30123 +       ssize_t err;
30124 +
30125 +       lockdep_off();
30126 +       err = vfs_read(file, ubuf, count, ppos);
30127 +       lockdep_on();
30128 +       if (err >= 0)
30129 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30130 +       return err;
30131 +}
30132 +
30133 +/* todo: kernel_read()? */
30134 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
30135 +                    loff_t *ppos)
30136 +{
30137 +       ssize_t err;
30138 +       mm_segment_t oldfs;
30139 +       union {
30140 +               void *k;
30141 +               char __user *u;
30142 +       } buf;
30143 +
30144 +       buf.k = kbuf;
30145 +       oldfs = get_fs();
30146 +       set_fs(KERNEL_DS);
30147 +       err = vfsub_read_u(file, buf.u, count, ppos);
30148 +       set_fs(oldfs);
30149 +       return err;
30150 +}
30151 +
30152 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
30153 +                     loff_t *ppos)
30154 +{
30155 +       ssize_t err;
30156 +
30157 +       lockdep_off();
30158 +       err = vfs_write(file, ubuf, count, ppos);
30159 +       lockdep_on();
30160 +       if (err >= 0)
30161 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30162 +       return err;
30163 +}
30164 +
30165 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count, loff_t *ppos)
30166 +{
30167 +       ssize_t err;
30168 +       mm_segment_t oldfs;
30169 +       union {
30170 +               void *k;
30171 +               const char __user *u;
30172 +       } buf;
30173 +
30174 +       buf.k = kbuf;
30175 +       oldfs = get_fs();
30176 +       set_fs(KERNEL_DS);
30177 +       err = vfsub_write_u(file, buf.u, count, ppos);
30178 +       set_fs(oldfs);
30179 +       return err;
30180 +}
30181 +
30182 +int vfsub_flush(struct file *file, fl_owner_t id)
30183 +{
30184 +       int err;
30185 +
30186 +       err = 0;
30187 +       if (file->f_op->flush) {
30188 +               if (!au_test_nfs(file->f_dentry->d_sb))
30189 +                       err = file->f_op->flush(file, id);
30190 +               else {
30191 +                       lockdep_off();
30192 +                       err = file->f_op->flush(file, id);
30193 +                       lockdep_on();
30194 +               }
30195 +               if (!err)
30196 +                       vfsub_update_h_iattr(&file->f_path, /*did*/NULL);
30197 +               /*ignore*/
30198 +       }
30199 +       return err;
30200 +}
30201 +
30202 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx)
30203 +{
30204 +       int err;
30205 +
30206 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
30207 +
30208 +       lockdep_off();
30209 +       err = iterate_dir(file, ctx);
30210 +       lockdep_on();
30211 +       if (err >= 0)
30212 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30213 +       return err;
30214 +}
30215 +
30216 +long vfsub_splice_to(struct file *in, loff_t *ppos,
30217 +                    struct pipe_inode_info *pipe, size_t len,
30218 +                    unsigned int flags)
30219 +{
30220 +       long err;
30221 +
30222 +       lockdep_off();
30223 +       err = do_splice_to(in, ppos, pipe, len, flags);
30224 +       lockdep_on();
30225 +       file_accessed(in);
30226 +       if (err >= 0)
30227 +               vfsub_update_h_iattr(&in->f_path, /*did*/NULL); /*ignore*/
30228 +       return err;
30229 +}
30230 +
30231 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
30232 +                      loff_t *ppos, size_t len, unsigned int flags)
30233 +{
30234 +       long err;
30235 +
30236 +       lockdep_off();
30237 +       err = do_splice_from(pipe, out, ppos, len, flags);
30238 +       lockdep_on();
30239 +       if (err >= 0)
30240 +               vfsub_update_h_iattr(&out->f_path, /*did*/NULL); /*ignore*/
30241 +       return err;
30242 +}
30243 +
30244 +int vfsub_fsync(struct file *file, struct path *path, int datasync)
30245 +{
30246 +       int err;
30247 +
30248 +       /* file can be NULL */
30249 +       lockdep_off();
30250 +       err = vfs_fsync(file, datasync);
30251 +       lockdep_on();
30252 +       if (!err) {
30253 +               if (!path) {
30254 +                       AuDebugOn(!file);
30255 +                       path = &file->f_path;
30256 +               }
30257 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
30258 +       }
30259 +       return err;
30260 +}
30261 +
30262 +/* cf. open.c:do_sys_truncate() and do_sys_ftruncate() */
30263 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
30264 +               struct file *h_file)
30265 +{
30266 +       int err;
30267 +       struct inode *h_inode;
30268 +       struct super_block *h_sb;
30269 +
30270 +       if (!h_file) {
30271 +               err = vfsub_truncate(h_path, length);
30272 +               goto out;
30273 +       }
30274 +
30275 +       h_inode = h_path->dentry->d_inode;
30276 +       h_sb = h_inode->i_sb;
30277 +       lockdep_off();
30278 +       sb_start_write(h_sb);
30279 +       lockdep_on();
30280 +       err = locks_verify_truncate(h_inode, h_file, length);
30281 +       if (!err)
30282 +               err = security_path_truncate(h_path);
30283 +       if (!err) {
30284 +               lockdep_off();
30285 +               err = do_truncate(h_path->dentry, length, attr, h_file);
30286 +               lockdep_on();
30287 +       }
30288 +       lockdep_off();
30289 +       sb_end_write(h_sb);
30290 +       lockdep_on();
30291 +
30292 +out:
30293 +       return err;
30294 +}
30295 +
30296 +/* ---------------------------------------------------------------------- */
30297 +
30298 +struct au_vfsub_mkdir_args {
30299 +       int *errp;
30300 +       struct inode *dir;
30301 +       struct path *path;
30302 +       int mode;
30303 +};
30304 +
30305 +static void au_call_vfsub_mkdir(void *args)
30306 +{
30307 +       struct au_vfsub_mkdir_args *a = args;
30308 +       *a->errp = vfsub_mkdir(a->dir, a->path, a->mode);
30309 +}
30310 +
30311 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode)
30312 +{
30313 +       int err, do_sio, wkq_err;
30314 +
30315 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
30316 +       if (!do_sio) {
30317 +               lockdep_off();
30318 +               err = vfsub_mkdir(dir, path, mode);
30319 +               lockdep_on();
30320 +       } else {
30321 +               struct au_vfsub_mkdir_args args = {
30322 +                       .errp   = &err,
30323 +                       .dir    = dir,
30324 +                       .path   = path,
30325 +                       .mode   = mode
30326 +               };
30327 +               wkq_err = au_wkq_wait(au_call_vfsub_mkdir, &args);
30328 +               if (unlikely(wkq_err))
30329 +                       err = wkq_err;
30330 +       }
30331 +
30332 +       return err;
30333 +}
30334 +
30335 +struct au_vfsub_rmdir_args {
30336 +       int *errp;
30337 +       struct inode *dir;
30338 +       struct path *path;
30339 +};
30340 +
30341 +static void au_call_vfsub_rmdir(void *args)
30342 +{
30343 +       struct au_vfsub_rmdir_args *a = args;
30344 +       *a->errp = vfsub_rmdir(a->dir, a->path);
30345 +}
30346 +
30347 +int vfsub_sio_rmdir(struct inode *dir, struct path *path)
30348 +{
30349 +       int err, do_sio, wkq_err;
30350 +
30351 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
30352 +       if (!do_sio) {
30353 +               lockdep_off();
30354 +               err = vfsub_rmdir(dir, path);
30355 +               lockdep_on();
30356 +       } else {
30357 +               struct au_vfsub_rmdir_args args = {
30358 +                       .errp   = &err,
30359 +                       .dir    = dir,
30360 +                       .path   = path
30361 +               };
30362 +               wkq_err = au_wkq_wait(au_call_vfsub_rmdir, &args);
30363 +               if (unlikely(wkq_err))
30364 +                       err = wkq_err;
30365 +       }
30366 +
30367 +       return err;
30368 +}
30369 +
30370 +/* ---------------------------------------------------------------------- */
30371 +
30372 +struct notify_change_args {
30373 +       int *errp;
30374 +       struct path *path;
30375 +       struct iattr *ia;
30376 +       struct inode **delegated_inode;
30377 +};
30378 +
30379 +static void call_notify_change(void *args)
30380 +{
30381 +       struct notify_change_args *a = args;
30382 +       struct inode *h_inode;
30383 +
30384 +       h_inode = a->path->dentry->d_inode;
30385 +       IMustLock(h_inode);
30386 +
30387 +       *a->errp = -EPERM;
30388 +       if (!IS_IMMUTABLE(h_inode) && !IS_APPEND(h_inode)) {
30389 +               lockdep_off();
30390 +               *a->errp = notify_change(a->path->dentry, a->ia,
30391 +                                        a->delegated_inode);
30392 +               lockdep_on();
30393 +               if (!*a->errp)
30394 +                       vfsub_update_h_iattr(a->path, /*did*/NULL); /*ignore*/
30395 +       }
30396 +       AuTraceErr(*a->errp);
30397 +}
30398 +
30399 +int vfsub_notify_change(struct path *path, struct iattr *ia,
30400 +                       struct inode **delegated_inode)
30401 +{
30402 +       int err;
30403 +       struct notify_change_args args = {
30404 +               .errp                   = &err,
30405 +               .path                   = path,
30406 +               .ia                     = ia,
30407 +               .delegated_inode        = delegated_inode
30408 +       };
30409 +
30410 +       call_notify_change(&args);
30411 +
30412 +       return err;
30413 +}
30414 +
30415 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
30416 +                           struct inode **delegated_inode)
30417 +{
30418 +       int err, wkq_err;
30419 +       struct notify_change_args args = {
30420 +               .errp                   = &err,
30421 +               .path                   = path,
30422 +               .ia                     = ia,
30423 +               .delegated_inode        = delegated_inode
30424 +       };
30425 +
30426 +       wkq_err = au_wkq_wait(call_notify_change, &args);
30427 +       if (unlikely(wkq_err))
30428 +               err = wkq_err;
30429 +
30430 +       return err;
30431 +}
30432 +
30433 +/* ---------------------------------------------------------------------- */
30434 +
30435 +struct unlink_args {
30436 +       int *errp;
30437 +       struct inode *dir;
30438 +       struct path *path;
30439 +       struct inode **delegated_inode;
30440 +};
30441 +
30442 +static void call_unlink(void *args)
30443 +{
30444 +       struct unlink_args *a = args;
30445 +       struct dentry *d = a->path->dentry;
30446 +       struct inode *h_inode;
30447 +       const int stop_sillyrename = (au_test_nfs(d->d_sb)
30448 +                                     && au_dcount(d) == 1);
30449 +
30450 +       IMustLock(a->dir);
30451 +
30452 +       a->path->dentry = d->d_parent;
30453 +       *a->errp = security_path_unlink(a->path, d);
30454 +       a->path->dentry = d;
30455 +       if (unlikely(*a->errp))
30456 +               return;
30457 +
30458 +       if (!stop_sillyrename)
30459 +               dget(d);
30460 +       h_inode = d->d_inode;
30461 +       if (h_inode)
30462 +               ihold(h_inode);
30463 +
30464 +       lockdep_off();
30465 +       *a->errp = vfs_unlink(a->dir, d, a->delegated_inode);
30466 +       lockdep_on();
30467 +       if (!*a->errp) {
30468 +               struct path tmp = {
30469 +                       .dentry = d->d_parent,
30470 +                       .mnt    = a->path->mnt
30471 +               };
30472 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
30473 +       }
30474 +
30475 +       if (!stop_sillyrename)
30476 +               dput(d);
30477 +       if (h_inode)
30478 +               iput(h_inode);
30479 +
30480 +       AuTraceErr(*a->errp);
30481 +}
30482 +
30483 +/*
30484 + * @dir: must be locked.
30485 + * @dentry: target dentry.
30486 + */
30487 +int vfsub_unlink(struct inode *dir, struct path *path,
30488 +                struct inode **delegated_inode, int force)
30489 +{
30490 +       int err;
30491 +       struct unlink_args args = {
30492 +               .errp                   = &err,
30493 +               .dir                    = dir,
30494 +               .path                   = path,
30495 +               .delegated_inode        = delegated_inode
30496 +       };
30497 +
30498 +       if (!force)
30499 +               call_unlink(&args);
30500 +       else {
30501 +               int wkq_err;
30502 +
30503 +               wkq_err = au_wkq_wait(call_unlink, &args);
30504 +               if (unlikely(wkq_err))
30505 +                       err = wkq_err;
30506 +       }
30507 +
30508 +       return err;
30509 +}
30510 diff -urN /usr/share/empty/fs/aufs/vfsub.h linux/fs/aufs/vfsub.h
30511 --- /usr/share/empty/fs/aufs/vfsub.h    1970-01-01 01:00:00.000000000 +0100
30512 +++ linux/fs/aufs/vfsub.h       2015-01-25 13:00:38.634380408 +0100
30513 @@ -0,0 +1,310 @@
30514 +/*
30515 + * Copyright (C) 2005-2014 Junjiro R. Okajima
30516 + *
30517 + * This program, aufs is free software; you can redistribute it and/or modify
30518 + * it under the terms of the GNU General Public License as published by
30519 + * the Free Software Foundation; either version 2 of the License, or
30520 + * (at your option) any later version.
30521 + *
30522 + * This program is distributed in the hope that it will be useful,
30523 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30524 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30525 + * GNU General Public License for more details.
30526 + *
30527 + * You should have received a copy of the GNU General Public License
30528 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30529 + */
30530 +
30531 +/*
30532 + * sub-routines for VFS
30533 + */
30534 +
30535 +#ifndef __AUFS_VFSUB_H__
30536 +#define __AUFS_VFSUB_H__
30537 +
30538 +#ifdef __KERNEL__
30539 +
30540 +#include <linux/fs.h>
30541 +#include <linux/lglock.h>
30542 +#include <linux/mount.h>
30543 +#include <linux/xattr.h>
30544 +#include "debug.h"
30545 +
30546 +/* copied from linux/fs/internal.h */
30547 +/* todo: BAD approach!! */
30548 +extern void __mnt_drop_write(struct vfsmount *);
30549 +extern spinlock_t inode_sb_list_lock;
30550 +
30551 +/* ---------------------------------------------------------------------- */
30552 +
30553 +/* lock subclass for lower inode */
30554 +/* default MAX_LOCKDEP_SUBCLASSES(8) is not enough */
30555 +/* reduce? gave up. */
30556 +enum {
30557 +       AuLsc_I_Begin = I_MUTEX_PARENT2, /* 5 */
30558 +       AuLsc_I_PARENT,         /* lower inode, parent first */
30559 +       AuLsc_I_PARENT2,        /* copyup dirs */
30560 +       AuLsc_I_PARENT3,        /* copyup wh */
30561 +       AuLsc_I_CHILD,
30562 +       AuLsc_I_CHILD2,
30563 +       AuLsc_I_End
30564 +};
30565 +
30566 +/* to debug easier, do not make them inlined functions */
30567 +#define MtxMustLock(mtx)       AuDebugOn(!mutex_is_locked(mtx))
30568 +#define IMustLock(i)           MtxMustLock(&(i)->i_mutex)
30569 +
30570 +/* ---------------------------------------------------------------------- */
30571 +
30572 +static inline void vfsub_drop_nlink(struct inode *inode)
30573 +{
30574 +       AuDebugOn(!inode->i_nlink);
30575 +       drop_nlink(inode);
30576 +}
30577 +
30578 +static inline void vfsub_dead_dir(struct inode *inode)
30579 +{
30580 +       AuDebugOn(!S_ISDIR(inode->i_mode));
30581 +       inode->i_flags |= S_DEAD;
30582 +       clear_nlink(inode);
30583 +}
30584 +
30585 +static inline int vfsub_native_ro(struct inode *inode)
30586 +{
30587 +       return (inode->i_sb->s_flags & MS_RDONLY)
30588 +               || IS_RDONLY(inode)
30589 +               /* || IS_APPEND(inode) */
30590 +               || IS_IMMUTABLE(inode);
30591 +}
30592 +
30593 +/* ---------------------------------------------------------------------- */
30594 +
30595 +int vfsub_update_h_iattr(struct path *h_path, int *did);
30596 +struct file *vfsub_dentry_open(struct path *path, int flags);
30597 +struct file *vfsub_filp_open(const char *path, int oflags, int mode);
30598 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path);
30599 +
30600 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
30601 +                                   int len);
30602 +
30603 +struct vfsub_lkup_one_args {
30604 +       struct dentry **errp;
30605 +       struct qstr *name;
30606 +       struct dentry *parent;
30607 +};
30608 +
30609 +static inline struct dentry *vfsub_lkup_one(struct qstr *name,
30610 +                                           struct dentry *parent)
30611 +{
30612 +       return vfsub_lookup_one_len(name->name, parent, name->len);
30613 +}
30614 +
30615 +void vfsub_call_lkup_one(void *args);
30616 +
30617 +/* ---------------------------------------------------------------------- */
30618 +
30619 +static inline int vfsub_mnt_want_write(struct vfsmount *mnt)
30620 +{
30621 +       int err;
30622 +
30623 +       lockdep_off();
30624 +       err = mnt_want_write(mnt);
30625 +       lockdep_on();
30626 +       return err;
30627 +}
30628 +
30629 +static inline void vfsub_mnt_drop_write(struct vfsmount *mnt)
30630 +{
30631 +       lockdep_off();
30632 +       mnt_drop_write(mnt);
30633 +       lockdep_on();
30634 +}
30635 +
30636 +static inline void vfsub_mnt_drop_write_file(struct file *file)
30637 +{
30638 +       lockdep_off();
30639 +       mnt_drop_write_file(file);
30640 +       lockdep_on();
30641 +}
30642 +
30643 +/* ---------------------------------------------------------------------- */
30644 +
30645 +struct au_hinode;
30646 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
30647 +                                struct dentry *d2, struct au_hinode *hdir2);
30648 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
30649 +                        struct dentry *d2, struct au_hinode *hdir2);
30650 +
30651 +int vfsub_create(struct inode *dir, struct path *path, int mode,
30652 +                bool want_excl);
30653 +int vfsub_symlink(struct inode *dir, struct path *path,
30654 +                 const char *symname);
30655 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev);
30656 +int vfsub_link(struct dentry *src_dentry, struct inode *dir,
30657 +              struct path *path, struct inode **delegated_inode);
30658 +int vfsub_rename(struct inode *src_hdir, struct dentry *src_dentry,
30659 +                struct inode *hdir, struct path *path,
30660 +                struct inode **delegated_inode);
30661 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode);
30662 +int vfsub_rmdir(struct inode *dir, struct path *path);
30663 +
30664 +/* ---------------------------------------------------------------------- */
30665 +
30666 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
30667 +                    loff_t *ppos);
30668 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
30669 +                       loff_t *ppos);
30670 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
30671 +                     loff_t *ppos);
30672 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count,
30673 +                     loff_t *ppos);
30674 +int vfsub_flush(struct file *file, fl_owner_t id);
30675 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx);
30676 +
30677 +/* just for type-check */
30678 +static inline filldir_t au_diractor(int (*func)(struct dir_context *,
30679 +                                               const char *, int, loff_t, u64,
30680 +                                               unsigned))
30681 +{
30682 +       return (filldir_t)func;
30683 +}
30684 +
30685 +
30686 +static inline loff_t vfsub_f_size_read(struct file *file)
30687 +{
30688 +       return i_size_read(file_inode(file));
30689 +}
30690 +
30691 +static inline unsigned int vfsub_file_flags(struct file *file)
30692 +{
30693 +       unsigned int flags;
30694 +
30695 +       spin_lock(&file->f_lock);
30696 +       flags = file->f_flags;
30697 +       spin_unlock(&file->f_lock);
30698 +
30699 +       return flags;
30700 +}
30701 +
30702 +static inline void vfsub_file_accessed(struct file *h_file)
30703 +{
30704 +       file_accessed(h_file);
30705 +       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL); /*ignore*/
30706 +}
30707 +
30708 +static inline void vfsub_touch_atime(struct vfsmount *h_mnt,
30709 +                                    struct dentry *h_dentry)
30710 +{
30711 +       struct path h_path = {
30712 +               .dentry = h_dentry,
30713 +               .mnt    = h_mnt
30714 +       };
30715 +       touch_atime(&h_path);
30716 +       vfsub_update_h_iattr(&h_path, /*did*/NULL); /*ignore*/
30717 +}
30718 +
30719 +static inline int vfsub_update_time(struct inode *h_inode, struct timespec *ts,
30720 +                                   int flags)
30721 +{
30722 +       return update_time(h_inode, ts, flags);
30723 +       /* no vfsub_update_h_iattr() since we don't have struct path */
30724 +}
30725 +
30726 +long vfsub_splice_to(struct file *in, loff_t *ppos,
30727 +                    struct pipe_inode_info *pipe, size_t len,
30728 +                    unsigned int flags);
30729 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
30730 +                      loff_t *ppos, size_t len, unsigned int flags);
30731 +
30732 +static inline long vfsub_truncate(struct path *path, loff_t length)
30733 +{
30734 +       long err;
30735 +
30736 +       lockdep_off();
30737 +       err = vfs_truncate(path, length);
30738 +       lockdep_on();
30739 +       return err;
30740 +}
30741 +
30742 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
30743 +               struct file *h_file);
30744 +int vfsub_fsync(struct file *file, struct path *path, int datasync);
30745 +
30746 +/* ---------------------------------------------------------------------- */
30747 +
30748 +static inline loff_t vfsub_llseek(struct file *file, loff_t offset, int origin)
30749 +{
30750 +       loff_t err;
30751 +
30752 +       lockdep_off();
30753 +       err = vfs_llseek(file, offset, origin);
30754 +       lockdep_on();
30755 +       return err;
30756 +}
30757 +
30758 +/* ---------------------------------------------------------------------- */
30759 +
30760 +/* dirty workaround for strict type of fmode_t */
30761 +union vfsub_fmu {
30762 +       fmode_t fm;
30763 +       unsigned int ui;
30764 +};
30765 +
30766 +static inline unsigned int vfsub_fmode_to_uint(fmode_t fm)
30767 +{
30768 +       union vfsub_fmu u = {
30769 +               .fm = fm
30770 +       };
30771 +
30772 +       BUILD_BUG_ON(sizeof(u.fm) != sizeof(u.ui));
30773 +
30774 +       return u.ui;
30775 +}
30776 +
30777 +static inline fmode_t vfsub_uint_to_fmode(unsigned int ui)
30778 +{
30779 +       union vfsub_fmu u = {
30780 +               .ui = ui
30781 +       };
30782 +
30783 +       return u.fm;
30784 +}
30785 +
30786 +/* ---------------------------------------------------------------------- */
30787 +
30788 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode);
30789 +int vfsub_sio_rmdir(struct inode *dir, struct path *path);
30790 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
30791 +                           struct inode **delegated_inode);
30792 +int vfsub_notify_change(struct path *path, struct iattr *ia,
30793 +                       struct inode **delegated_inode);
30794 +int vfsub_unlink(struct inode *dir, struct path *path,
30795 +                struct inode **delegated_inode, int force);
30796 +
30797 +/* ---------------------------------------------------------------------- */
30798 +
30799 +static inline int vfsub_setxattr(struct dentry *dentry, const char *name,
30800 +                                const void *value, size_t size, int flags)
30801 +{
30802 +       int err;
30803 +
30804 +       lockdep_off();
30805 +       err = vfs_setxattr(dentry, name, value, size, flags);
30806 +       lockdep_on();
30807 +
30808 +       return err;
30809 +}
30810 +
30811 +static inline int vfsub_removexattr(struct dentry *dentry, const char *name)
30812 +{
30813 +       int err;
30814 +
30815 +       lockdep_off();
30816 +       err = vfs_removexattr(dentry, name);
30817 +       lockdep_on();
30818 +
30819 +       return err;
30820 +}
30821 +
30822 +#endif /* __KERNEL__ */
30823 +#endif /* __AUFS_VFSUB_H__ */
30824 diff -urN /usr/share/empty/fs/aufs/wbr_policy.c linux/fs/aufs/wbr_policy.c
30825 --- /usr/share/empty/fs/aufs/wbr_policy.c       1970-01-01 01:00:00.000000000 +0100
30826 +++ linux/fs/aufs/wbr_policy.c  2015-01-25 13:00:38.634380408 +0100
30827 @@ -0,0 +1,765 @@
30828 +/*
30829 + * Copyright (C) 2005-2014 Junjiro R. Okajima
30830 + *
30831 + * This program, aufs is free software; you can redistribute it and/or modify
30832 + * it under the terms of the GNU General Public License as published by
30833 + * the Free Software Foundation; either version 2 of the License, or
30834 + * (at your option) any later version.
30835 + *
30836 + * This program is distributed in the hope that it will be useful,
30837 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30838 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30839 + * GNU General Public License for more details.
30840 + *
30841 + * You should have received a copy of the GNU General Public License
30842 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30843 + */
30844 +
30845 +/*
30846 + * policies for selecting one among multiple writable branches
30847 + */
30848 +
30849 +#include <linux/statfs.h>
30850 +#include "aufs.h"
30851 +
30852 +/* subset of cpup_attr() */
30853 +static noinline_for_stack
30854 +int au_cpdown_attr(struct path *h_path, struct dentry *h_src)
30855 +{
30856 +       int err, sbits;
30857 +       struct iattr ia;
30858 +       struct inode *h_isrc;
30859 +
30860 +       h_isrc = h_src->d_inode;
30861 +       ia.ia_valid = ATTR_FORCE | ATTR_MODE | ATTR_UID | ATTR_GID;
30862 +       ia.ia_mode = h_isrc->i_mode;
30863 +       ia.ia_uid = h_isrc->i_uid;
30864 +       ia.ia_gid = h_isrc->i_gid;
30865 +       sbits = !!(ia.ia_mode & (S_ISUID | S_ISGID));
30866 +       au_cpup_attr_flags(h_path->dentry->d_inode, h_isrc->i_flags);
30867 +       /* no delegation since it is just created */
30868 +       err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
30869 +
30870 +       /* is this nfs only? */
30871 +       if (!err && sbits && au_test_nfs(h_path->dentry->d_sb)) {
30872 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
30873 +               ia.ia_mode = h_isrc->i_mode;
30874 +               err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
30875 +       }
30876 +
30877 +       return err;
30878 +}
30879 +
30880 +#define AuCpdown_PARENT_OPQ    1
30881 +#define AuCpdown_WHED          (1 << 1)
30882 +#define AuCpdown_MADE_DIR      (1 << 2)
30883 +#define AuCpdown_DIROPQ                (1 << 3)
30884 +#define au_ftest_cpdown(flags, name)   ((flags) & AuCpdown_##name)
30885 +#define au_fset_cpdown(flags, name) \
30886 +       do { (flags) |= AuCpdown_##name; } while (0)
30887 +#define au_fclr_cpdown(flags, name) \
30888 +       do { (flags) &= ~AuCpdown_##name; } while (0)
30889 +
30890 +static int au_cpdown_dir_opq(struct dentry *dentry, aufs_bindex_t bdst,
30891 +                            unsigned int *flags)
30892 +{
30893 +       int err;
30894 +       struct dentry *opq_dentry;
30895 +
30896 +       opq_dentry = au_diropq_create(dentry, bdst);
30897 +       err = PTR_ERR(opq_dentry);
30898 +       if (IS_ERR(opq_dentry))
30899 +               goto out;
30900 +       dput(opq_dentry);
30901 +       au_fset_cpdown(*flags, DIROPQ);
30902 +
30903 +out:
30904 +       return err;
30905 +}
30906 +
30907 +static int au_cpdown_dir_wh(struct dentry *dentry, struct dentry *h_parent,
30908 +                           struct inode *dir, aufs_bindex_t bdst)
30909 +{
30910 +       int err;
30911 +       struct path h_path;
30912 +       struct au_branch *br;
30913 +
30914 +       br = au_sbr(dentry->d_sb, bdst);
30915 +       h_path.dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
30916 +       err = PTR_ERR(h_path.dentry);
30917 +       if (IS_ERR(h_path.dentry))
30918 +               goto out;
30919 +
30920 +       err = 0;
30921 +       if (h_path.dentry->d_inode) {
30922 +               h_path.mnt = au_br_mnt(br);
30923 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bdst), &h_path,
30924 +                                         dentry);
30925 +       }
30926 +       dput(h_path.dentry);
30927 +
30928 +out:
30929 +       return err;
30930 +}
30931 +
30932 +static int au_cpdown_dir(struct dentry *dentry, aufs_bindex_t bdst,
30933 +                        struct au_pin *pin,
30934 +                        struct dentry *h_parent, void *arg)
30935 +{
30936 +       int err, rerr;
30937 +       aufs_bindex_t bopq, bstart;
30938 +       struct path h_path;
30939 +       struct dentry *parent;
30940 +       struct inode *h_dir, *h_inode, *inode, *dir;
30941 +       unsigned int *flags = arg;
30942 +
30943 +       bstart = au_dbstart(dentry);
30944 +       /* dentry is di-locked */
30945 +       parent = dget_parent(dentry);
30946 +       dir = parent->d_inode;
30947 +       h_dir = h_parent->d_inode;
30948 +       AuDebugOn(h_dir != au_h_iptr(dir, bdst));
30949 +       IMustLock(h_dir);
30950 +
30951 +       err = au_lkup_neg(dentry, bdst, /*wh*/0);
30952 +       if (unlikely(err < 0))
30953 +               goto out;
30954 +       h_path.dentry = au_h_dptr(dentry, bdst);
30955 +       h_path.mnt = au_sbr_mnt(dentry->d_sb, bdst);
30956 +       err = vfsub_sio_mkdir(au_h_iptr(dir, bdst), &h_path,
30957 +                             S_IRWXU | S_IRUGO | S_IXUGO);
30958 +       if (unlikely(err))
30959 +               goto out_put;
30960 +       au_fset_cpdown(*flags, MADE_DIR);
30961 +
30962 +       bopq = au_dbdiropq(dentry);
30963 +       au_fclr_cpdown(*flags, WHED);
30964 +       au_fclr_cpdown(*flags, DIROPQ);
30965 +       if (au_dbwh(dentry) == bdst)
30966 +               au_fset_cpdown(*flags, WHED);
30967 +       if (!au_ftest_cpdown(*flags, PARENT_OPQ) && bopq <= bdst)
30968 +               au_fset_cpdown(*flags, PARENT_OPQ);
30969 +       h_inode = h_path.dentry->d_inode;
30970 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
30971 +       if (au_ftest_cpdown(*flags, WHED)) {
30972 +               err = au_cpdown_dir_opq(dentry, bdst, flags);
30973 +               if (unlikely(err)) {
30974 +                       mutex_unlock(&h_inode->i_mutex);
30975 +                       goto out_dir;
30976 +               }
30977 +       }
30978 +
30979 +       err = au_cpdown_attr(&h_path, au_h_dptr(dentry, bstart));
30980 +       mutex_unlock(&h_inode->i_mutex);
30981 +       if (unlikely(err))
30982 +               goto out_opq;
30983 +
30984 +       if (au_ftest_cpdown(*flags, WHED)) {
30985 +               err = au_cpdown_dir_wh(dentry, h_parent, dir, bdst);
30986 +               if (unlikely(err))
30987 +                       goto out_opq;
30988 +       }
30989 +
30990 +       inode = dentry->d_inode;
30991 +       if (au_ibend(inode) < bdst)
30992 +               au_set_ibend(inode, bdst);
30993 +       au_set_h_iptr(inode, bdst, au_igrab(h_inode),
30994 +                     au_hi_flags(inode, /*isdir*/1));
30995 +       au_fhsm_wrote(dentry->d_sb, bdst, /*force*/0);
30996 +       goto out; /* success */
30997 +
30998 +       /* revert */
30999 +out_opq:
31000 +       if (au_ftest_cpdown(*flags, DIROPQ)) {
31001 +               mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
31002 +               rerr = au_diropq_remove(dentry, bdst);
31003 +               mutex_unlock(&h_inode->i_mutex);
31004 +               if (unlikely(rerr)) {
31005 +                       AuIOErr("failed removing diropq for %pd b%d (%d)\n",
31006 +                               dentry, bdst, rerr);
31007 +                       err = -EIO;
31008 +                       goto out;
31009 +               }
31010 +       }
31011 +out_dir:
31012 +       if (au_ftest_cpdown(*flags, MADE_DIR)) {
31013 +               rerr = vfsub_sio_rmdir(au_h_iptr(dir, bdst), &h_path);
31014 +               if (unlikely(rerr)) {
31015 +                       AuIOErr("failed removing %pd b%d (%d)\n",
31016 +                               dentry, bdst, rerr);
31017 +                       err = -EIO;
31018 +               }
31019 +       }
31020 +out_put:
31021 +       au_set_h_dptr(dentry, bdst, NULL);
31022 +       if (au_dbend(dentry) == bdst)
31023 +               au_update_dbend(dentry);
31024 +out:
31025 +       dput(parent);
31026 +       return err;
31027 +}
31028 +
31029 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst)
31030 +{
31031 +       int err;
31032 +       unsigned int flags;
31033 +
31034 +       flags = 0;
31035 +       err = au_cp_dirs(dentry, bdst, au_cpdown_dir, &flags);
31036 +
31037 +       return err;
31038 +}
31039 +
31040 +/* ---------------------------------------------------------------------- */
31041 +
31042 +/* policies for create */
31043 +
31044 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex)
31045 +{
31046 +       int err, i, j, ndentry;
31047 +       aufs_bindex_t bopq;
31048 +       struct au_dcsub_pages dpages;
31049 +       struct au_dpage *dpage;
31050 +       struct dentry **dentries, *parent, *d;
31051 +
31052 +       err = au_dpages_init(&dpages, GFP_NOFS);
31053 +       if (unlikely(err))
31054 +               goto out;
31055 +       parent = dget_parent(dentry);
31056 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/0);
31057 +       if (unlikely(err))
31058 +               goto out_free;
31059 +
31060 +       err = bindex;
31061 +       for (i = 0; i < dpages.ndpage; i++) {
31062 +               dpage = dpages.dpages + i;
31063 +               dentries = dpage->dentries;
31064 +               ndentry = dpage->ndentry;
31065 +               for (j = 0; j < ndentry; j++) {
31066 +                       d = dentries[j];
31067 +                       di_read_lock_parent2(d, !AuLock_IR);
31068 +                       bopq = au_dbdiropq(d);
31069 +                       di_read_unlock(d, !AuLock_IR);
31070 +                       if (bopq >= 0 && bopq < err)
31071 +                               err = bopq;
31072 +               }
31073 +       }
31074 +
31075 +out_free:
31076 +       dput(parent);
31077 +       au_dpages_free(&dpages);
31078 +out:
31079 +       return err;
31080 +}
31081 +
31082 +static int au_wbr_bu(struct super_block *sb, aufs_bindex_t bindex)
31083 +{
31084 +       for (; bindex >= 0; bindex--)
31085 +               if (!au_br_rdonly(au_sbr(sb, bindex)))
31086 +                       return bindex;
31087 +       return -EROFS;
31088 +}
31089 +
31090 +/* top down parent */
31091 +static int au_wbr_create_tdp(struct dentry *dentry,
31092 +                            unsigned int flags __maybe_unused)
31093 +{
31094 +       int err;
31095 +       aufs_bindex_t bstart, bindex;
31096 +       struct super_block *sb;
31097 +       struct dentry *parent, *h_parent;
31098 +
31099 +       sb = dentry->d_sb;
31100 +       bstart = au_dbstart(dentry);
31101 +       err = bstart;
31102 +       if (!au_br_rdonly(au_sbr(sb, bstart)))
31103 +               goto out;
31104 +
31105 +       err = -EROFS;
31106 +       parent = dget_parent(dentry);
31107 +       for (bindex = au_dbstart(parent); bindex < bstart; bindex++) {
31108 +               h_parent = au_h_dptr(parent, bindex);
31109 +               if (!h_parent || !h_parent->d_inode)
31110 +                       continue;
31111 +
31112 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
31113 +                       err = bindex;
31114 +                       break;
31115 +               }
31116 +       }
31117 +       dput(parent);
31118 +
31119 +       /* bottom up here */
31120 +       if (unlikely(err < 0)) {
31121 +               err = au_wbr_bu(sb, bstart - 1);
31122 +               if (err >= 0)
31123 +                       err = au_wbr_nonopq(dentry, err);
31124 +       }
31125 +
31126 +out:
31127 +       AuDbg("b%d\n", err);
31128 +       return err;
31129 +}
31130 +
31131 +/* ---------------------------------------------------------------------- */
31132 +
31133 +/* an exception for the policy other than tdp */
31134 +static int au_wbr_create_exp(struct dentry *dentry)
31135 +{
31136 +       int err;
31137 +       aufs_bindex_t bwh, bdiropq;
31138 +       struct dentry *parent;
31139 +
31140 +       err = -1;
31141 +       bwh = au_dbwh(dentry);
31142 +       parent = dget_parent(dentry);
31143 +       bdiropq = au_dbdiropq(parent);
31144 +       if (bwh >= 0) {
31145 +               if (bdiropq >= 0)
31146 +                       err = min(bdiropq, bwh);
31147 +               else
31148 +                       err = bwh;
31149 +               AuDbg("%d\n", err);
31150 +       } else if (bdiropq >= 0) {
31151 +               err = bdiropq;
31152 +               AuDbg("%d\n", err);
31153 +       }
31154 +       dput(parent);
31155 +
31156 +       if (err >= 0)
31157 +               err = au_wbr_nonopq(dentry, err);
31158 +
31159 +       if (err >= 0 && au_br_rdonly(au_sbr(dentry->d_sb, err)))
31160 +               err = -1;
31161 +
31162 +       AuDbg("%d\n", err);
31163 +       return err;
31164 +}
31165 +
31166 +/* ---------------------------------------------------------------------- */
31167 +
31168 +/* round robin */
31169 +static int au_wbr_create_init_rr(struct super_block *sb)
31170 +{
31171 +       int err;
31172 +
31173 +       err = au_wbr_bu(sb, au_sbend(sb));
31174 +       atomic_set(&au_sbi(sb)->si_wbr_rr_next, -err); /* less important */
31175 +       /* smp_mb(); */
31176 +
31177 +       AuDbg("b%d\n", err);
31178 +       return err;
31179 +}
31180 +
31181 +static int au_wbr_create_rr(struct dentry *dentry, unsigned int flags)
31182 +{
31183 +       int err, nbr;
31184 +       unsigned int u;
31185 +       aufs_bindex_t bindex, bend;
31186 +       struct super_block *sb;
31187 +       atomic_t *next;
31188 +
31189 +       err = au_wbr_create_exp(dentry);
31190 +       if (err >= 0)
31191 +               goto out;
31192 +
31193 +       sb = dentry->d_sb;
31194 +       next = &au_sbi(sb)->si_wbr_rr_next;
31195 +       bend = au_sbend(sb);
31196 +       nbr = bend + 1;
31197 +       for (bindex = 0; bindex <= bend; bindex++) {
31198 +               if (!au_ftest_wbr(flags, DIR)) {
31199 +                       err = atomic_dec_return(next) + 1;
31200 +                       /* modulo for 0 is meaningless */
31201 +                       if (unlikely(!err))
31202 +                               err = atomic_dec_return(next) + 1;
31203 +               } else
31204 +                       err = atomic_read(next);
31205 +               AuDbg("%d\n", err);
31206 +               u = err;
31207 +               err = u % nbr;
31208 +               AuDbg("%d\n", err);
31209 +               if (!au_br_rdonly(au_sbr(sb, err)))
31210 +                       break;
31211 +               err = -EROFS;
31212 +       }
31213 +
31214 +       if (err >= 0)
31215 +               err = au_wbr_nonopq(dentry, err);
31216 +
31217 +out:
31218 +       AuDbg("%d\n", err);
31219 +       return err;
31220 +}
31221 +
31222 +/* ---------------------------------------------------------------------- */
31223 +
31224 +/* most free space */
31225 +static void au_mfs(struct dentry *dentry, struct dentry *parent)
31226 +{
31227 +       struct super_block *sb;
31228 +       struct au_branch *br;
31229 +       struct au_wbr_mfs *mfs;
31230 +       struct dentry *h_parent;
31231 +       aufs_bindex_t bindex, bend;
31232 +       int err;
31233 +       unsigned long long b, bavail;
31234 +       struct path h_path;
31235 +       /* reduce the stack usage */
31236 +       struct kstatfs *st;
31237 +
31238 +       st = kmalloc(sizeof(*st), GFP_NOFS);
31239 +       if (unlikely(!st)) {
31240 +               AuWarn1("failed updating mfs(%d), ignored\n", -ENOMEM);
31241 +               return;
31242 +       }
31243 +
31244 +       bavail = 0;
31245 +       sb = dentry->d_sb;
31246 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31247 +       MtxMustLock(&mfs->mfs_lock);
31248 +       mfs->mfs_bindex = -EROFS;
31249 +       mfs->mfsrr_bytes = 0;
31250 +       if (!parent) {
31251 +               bindex = 0;
31252 +               bend = au_sbend(sb);
31253 +       } else {
31254 +               bindex = au_dbstart(parent);
31255 +               bend = au_dbtaildir(parent);
31256 +       }
31257 +
31258 +       for (; bindex <= bend; bindex++) {
31259 +               if (parent) {
31260 +                       h_parent = au_h_dptr(parent, bindex);
31261 +                       if (!h_parent || !h_parent->d_inode)
31262 +                               continue;
31263 +               }
31264 +               br = au_sbr(sb, bindex);
31265 +               if (au_br_rdonly(br))
31266 +                       continue;
31267 +
31268 +               /* sb->s_root for NFS is unreliable */
31269 +               h_path.mnt = au_br_mnt(br);
31270 +               h_path.dentry = h_path.mnt->mnt_root;
31271 +               err = vfs_statfs(&h_path, st);
31272 +               if (unlikely(err)) {
31273 +                       AuWarn1("failed statfs, b%d, %d\n", bindex, err);
31274 +                       continue;
31275 +               }
31276 +
31277 +               /* when the available size is equal, select the lower one */
31278 +               BUILD_BUG_ON(sizeof(b) < sizeof(st->f_bavail)
31279 +                            || sizeof(b) < sizeof(st->f_bsize));
31280 +               b = st->f_bavail * st->f_bsize;
31281 +               br->br_wbr->wbr_bytes = b;
31282 +               if (b >= bavail) {
31283 +                       bavail = b;
31284 +                       mfs->mfs_bindex = bindex;
31285 +                       mfs->mfs_jiffy = jiffies;
31286 +               }
31287 +       }
31288 +
31289 +       mfs->mfsrr_bytes = bavail;
31290 +       AuDbg("b%d\n", mfs->mfs_bindex);
31291 +       kfree(st);
31292 +}
31293 +
31294 +static int au_wbr_create_mfs(struct dentry *dentry, unsigned int flags)
31295 +{
31296 +       int err;
31297 +       struct dentry *parent;
31298 +       struct super_block *sb;
31299 +       struct au_wbr_mfs *mfs;
31300 +
31301 +       err = au_wbr_create_exp(dentry);
31302 +       if (err >= 0)
31303 +               goto out;
31304 +
31305 +       sb = dentry->d_sb;
31306 +       parent = NULL;
31307 +       if (au_ftest_wbr(flags, PARENT))
31308 +               parent = dget_parent(dentry);
31309 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31310 +       mutex_lock(&mfs->mfs_lock);
31311 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
31312 +           || mfs->mfs_bindex < 0
31313 +           || au_br_rdonly(au_sbr(sb, mfs->mfs_bindex)))
31314 +               au_mfs(dentry, parent);
31315 +       mutex_unlock(&mfs->mfs_lock);
31316 +       err = mfs->mfs_bindex;
31317 +       dput(parent);
31318 +
31319 +       if (err >= 0)
31320 +               err = au_wbr_nonopq(dentry, err);
31321 +
31322 +out:
31323 +       AuDbg("b%d\n", err);
31324 +       return err;
31325 +}
31326 +
31327 +static int au_wbr_create_init_mfs(struct super_block *sb)
31328 +{
31329 +       struct au_wbr_mfs *mfs;
31330 +
31331 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31332 +       mutex_init(&mfs->mfs_lock);
31333 +       mfs->mfs_jiffy = 0;
31334 +       mfs->mfs_bindex = -EROFS;
31335 +
31336 +       return 0;
31337 +}
31338 +
31339 +static int au_wbr_create_fin_mfs(struct super_block *sb __maybe_unused)
31340 +{
31341 +       mutex_destroy(&au_sbi(sb)->si_wbr_mfs.mfs_lock);
31342 +       return 0;
31343 +}
31344 +
31345 +/* ---------------------------------------------------------------------- */
31346 +
31347 +/* most free space and then round robin */
31348 +static int au_wbr_create_mfsrr(struct dentry *dentry, unsigned int flags)
31349 +{
31350 +       int err;
31351 +       struct au_wbr_mfs *mfs;
31352 +
31353 +       err = au_wbr_create_mfs(dentry, flags);
31354 +       if (err >= 0) {
31355 +               mfs = &au_sbi(dentry->d_sb)->si_wbr_mfs;
31356 +               mutex_lock(&mfs->mfs_lock);
31357 +               if (mfs->mfsrr_bytes < mfs->mfsrr_watermark)
31358 +                       err = au_wbr_create_rr(dentry, flags);
31359 +               mutex_unlock(&mfs->mfs_lock);
31360 +       }
31361 +
31362 +       AuDbg("b%d\n", err);
31363 +       return err;
31364 +}
31365 +
31366 +static int au_wbr_create_init_mfsrr(struct super_block *sb)
31367 +{
31368 +       int err;
31369 +
31370 +       au_wbr_create_init_mfs(sb); /* ignore */
31371 +       err = au_wbr_create_init_rr(sb);
31372 +
31373 +       return err;
31374 +}
31375 +
31376 +/* ---------------------------------------------------------------------- */
31377 +
31378 +/* top down parent and most free space */
31379 +static int au_wbr_create_pmfs(struct dentry *dentry, unsigned int flags)
31380 +{
31381 +       int err, e2;
31382 +       unsigned long long b;
31383 +       aufs_bindex_t bindex, bstart, bend;
31384 +       struct super_block *sb;
31385 +       struct dentry *parent, *h_parent;
31386 +       struct au_branch *br;
31387 +
31388 +       err = au_wbr_create_tdp(dentry, flags);
31389 +       if (unlikely(err < 0))
31390 +               goto out;
31391 +       parent = dget_parent(dentry);
31392 +       bstart = au_dbstart(parent);
31393 +       bend = au_dbtaildir(parent);
31394 +       if (bstart == bend)
31395 +               goto out_parent; /* success */
31396 +
31397 +       e2 = au_wbr_create_mfs(dentry, flags);
31398 +       if (e2 < 0)
31399 +               goto out_parent; /* success */
31400 +
31401 +       /* when the available size is equal, select upper one */
31402 +       sb = dentry->d_sb;
31403 +       br = au_sbr(sb, err);
31404 +       b = br->br_wbr->wbr_bytes;
31405 +       AuDbg("b%d, %llu\n", err, b);
31406 +
31407 +       for (bindex = bstart; bindex <= bend; bindex++) {
31408 +               h_parent = au_h_dptr(parent, bindex);
31409 +               if (!h_parent || !h_parent->d_inode)
31410 +                       continue;
31411 +
31412 +               br = au_sbr(sb, bindex);
31413 +               if (!au_br_rdonly(br) && br->br_wbr->wbr_bytes > b) {
31414 +                       b = br->br_wbr->wbr_bytes;
31415 +                       err = bindex;
31416 +                       AuDbg("b%d, %llu\n", err, b);
31417 +               }
31418 +       }
31419 +
31420 +       if (err >= 0)
31421 +               err = au_wbr_nonopq(dentry, err);
31422 +
31423 +out_parent:
31424 +       dput(parent);
31425 +out:
31426 +       AuDbg("b%d\n", err);
31427 +       return err;
31428 +}
31429 +
31430 +/* ---------------------------------------------------------------------- */
31431 +
31432 +/*
31433 + * - top down parent
31434 + * - most free space with parent
31435 + * - most free space round-robin regardless parent
31436 + */
31437 +static int au_wbr_create_pmfsrr(struct dentry *dentry, unsigned int flags)
31438 +{
31439 +       int err;
31440 +       unsigned long long watermark;
31441 +       struct super_block *sb;
31442 +       struct au_branch *br;
31443 +       struct au_wbr_mfs *mfs;
31444 +
31445 +       err = au_wbr_create_pmfs(dentry, flags | AuWbr_PARENT);
31446 +       if (unlikely(err < 0))
31447 +               goto out;
31448 +
31449 +       sb = dentry->d_sb;
31450 +       br = au_sbr(sb, err);
31451 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31452 +       mutex_lock(&mfs->mfs_lock);
31453 +       watermark = mfs->mfsrr_watermark;
31454 +       mutex_unlock(&mfs->mfs_lock);
31455 +       if (br->br_wbr->wbr_bytes < watermark)
31456 +               /* regardless the parent dir */
31457 +               err = au_wbr_create_mfsrr(dentry, flags);
31458 +
31459 +out:
31460 +       AuDbg("b%d\n", err);
31461 +       return err;
31462 +}
31463 +
31464 +/* ---------------------------------------------------------------------- */
31465 +
31466 +/* policies for copyup */
31467 +
31468 +/* top down parent */
31469 +static int au_wbr_copyup_tdp(struct dentry *dentry)
31470 +{
31471 +       return au_wbr_create_tdp(dentry, /*flags, anything is ok*/0);
31472 +}
31473 +
31474 +/* bottom up parent */
31475 +static int au_wbr_copyup_bup(struct dentry *dentry)
31476 +{
31477 +       int err;
31478 +       aufs_bindex_t bindex, bstart;
31479 +       struct dentry *parent, *h_parent;
31480 +       struct super_block *sb;
31481 +
31482 +       err = -EROFS;
31483 +       sb = dentry->d_sb;
31484 +       parent = dget_parent(dentry);
31485 +       bstart = au_dbstart(parent);
31486 +       for (bindex = au_dbstart(dentry); bindex >= bstart; bindex--) {
31487 +               h_parent = au_h_dptr(parent, bindex);
31488 +               if (!h_parent || !h_parent->d_inode)
31489 +                       continue;
31490 +
31491 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
31492 +                       err = bindex;
31493 +                       break;
31494 +               }
31495 +       }
31496 +       dput(parent);
31497 +
31498 +       /* bottom up here */
31499 +       if (unlikely(err < 0))
31500 +               err = au_wbr_bu(sb, bstart - 1);
31501 +
31502 +       AuDbg("b%d\n", err);
31503 +       return err;
31504 +}
31505 +
31506 +/* bottom up */
31507 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t bstart)
31508 +{
31509 +       int err;
31510 +
31511 +       err = au_wbr_bu(dentry->d_sb, bstart);
31512 +       AuDbg("b%d\n", err);
31513 +       if (err > bstart)
31514 +               err = au_wbr_nonopq(dentry, err);
31515 +
31516 +       AuDbg("b%d\n", err);
31517 +       return err;
31518 +}
31519 +
31520 +static int au_wbr_copyup_bu(struct dentry *dentry)
31521 +{
31522 +       int err;
31523 +       aufs_bindex_t bstart;
31524 +
31525 +       bstart = au_dbstart(dentry);
31526 +       err = au_wbr_do_copyup_bu(dentry, bstart);
31527 +       return err;
31528 +}
31529 +
31530 +/* ---------------------------------------------------------------------- */
31531 +
31532 +struct au_wbr_copyup_operations au_wbr_copyup_ops[] = {
31533 +       [AuWbrCopyup_TDP] = {
31534 +               .copyup = au_wbr_copyup_tdp
31535 +       },
31536 +       [AuWbrCopyup_BUP] = {
31537 +               .copyup = au_wbr_copyup_bup
31538 +       },
31539 +       [AuWbrCopyup_BU] = {
31540 +               .copyup = au_wbr_copyup_bu
31541 +       }
31542 +};
31543 +
31544 +struct au_wbr_create_operations au_wbr_create_ops[] = {
31545 +       [AuWbrCreate_TDP] = {
31546 +               .create = au_wbr_create_tdp
31547 +       },
31548 +       [AuWbrCreate_RR] = {
31549 +               .create = au_wbr_create_rr,
31550 +               .init   = au_wbr_create_init_rr
31551 +       },
31552 +       [AuWbrCreate_MFS] = {
31553 +               .create = au_wbr_create_mfs,
31554 +               .init   = au_wbr_create_init_mfs,
31555 +               .fin    = au_wbr_create_fin_mfs
31556 +       },
31557 +       [AuWbrCreate_MFSV] = {
31558 +               .create = au_wbr_create_mfs,
31559 +               .init   = au_wbr_create_init_mfs,
31560 +               .fin    = au_wbr_create_fin_mfs
31561 +       },
31562 +       [AuWbrCreate_MFSRR] = {
31563 +               .create = au_wbr_create_mfsrr,
31564 +               .init   = au_wbr_create_init_mfsrr,
31565 +               .fin    = au_wbr_create_fin_mfs
31566 +       },
31567 +       [AuWbrCreate_MFSRRV] = {
31568 +               .create = au_wbr_create_mfsrr,
31569 +               .init   = au_wbr_create_init_mfsrr,
31570 +               .fin    = au_wbr_create_fin_mfs
31571 +       },
31572 +       [AuWbrCreate_PMFS] = {
31573 +               .create = au_wbr_create_pmfs,
31574 +               .init   = au_wbr_create_init_mfs,
31575 +               .fin    = au_wbr_create_fin_mfs
31576 +       },
31577 +       [AuWbrCreate_PMFSV] = {
31578 +               .create = au_wbr_create_pmfs,
31579 +               .init   = au_wbr_create_init_mfs,
31580 +               .fin    = au_wbr_create_fin_mfs
31581 +       },
31582 +       [AuWbrCreate_PMFSRR] = {
31583 +               .create = au_wbr_create_pmfsrr,
31584 +               .init   = au_wbr_create_init_mfsrr,
31585 +               .fin    = au_wbr_create_fin_mfs
31586 +       },
31587 +       [AuWbrCreate_PMFSRRV] = {
31588 +               .create = au_wbr_create_pmfsrr,
31589 +               .init   = au_wbr_create_init_mfsrr,
31590 +               .fin    = au_wbr_create_fin_mfs
31591 +       }
31592 +};
31593 diff -urN /usr/share/empty/fs/aufs/whout.c linux/fs/aufs/whout.c
31594 --- /usr/share/empty/fs/aufs/whout.c    1970-01-01 01:00:00.000000000 +0100
31595 +++ linux/fs/aufs/whout.c       2015-01-25 13:00:38.634380408 +0100
31596 @@ -0,0 +1,1056 @@
31597 +/*
31598 + * Copyright (C) 2005-2014 Junjiro R. Okajima
31599 + *
31600 + * This program, aufs is free software; you can redistribute it and/or modify
31601 + * it under the terms of the GNU General Public License as published by
31602 + * the Free Software Foundation; either version 2 of the License, or
31603 + * (at your option) any later version.
31604 + *
31605 + * This program is distributed in the hope that it will be useful,
31606 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31607 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31608 + * GNU General Public License for more details.
31609 + *
31610 + * You should have received a copy of the GNU General Public License
31611 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31612 + */
31613 +
31614 +/*
31615 + * whiteout for logical deletion and opaque directory
31616 + */
31617 +
31618 +#include "aufs.h"
31619 +
31620 +#define WH_MASK                        S_IRUGO
31621 +
31622 +/*
31623 + * If a directory contains this file, then it is opaque.  We start with the
31624 + * .wh. flag so that it is blocked by lookup.
31625 + */
31626 +static struct qstr diropq_name = QSTR_INIT(AUFS_WH_DIROPQ,
31627 +                                          sizeof(AUFS_WH_DIROPQ) - 1);
31628 +
31629 +/*
31630 + * generate whiteout name, which is NOT terminated by NULL.
31631 + * @name: original d_name.name
31632 + * @len: original d_name.len
31633 + * @wh: whiteout qstr
31634 + * returns zero when succeeds, otherwise error.
31635 + * succeeded value as wh->name should be freed by kfree().
31636 + */
31637 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name)
31638 +{
31639 +       char *p;
31640 +
31641 +       if (unlikely(name->len > PATH_MAX - AUFS_WH_PFX_LEN))
31642 +               return -ENAMETOOLONG;
31643 +
31644 +       wh->len = name->len + AUFS_WH_PFX_LEN;
31645 +       p = kmalloc(wh->len, GFP_NOFS);
31646 +       wh->name = p;
31647 +       if (p) {
31648 +               memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
31649 +               memcpy(p + AUFS_WH_PFX_LEN, name->name, name->len);
31650 +               /* smp_mb(); */
31651 +               return 0;
31652 +       }
31653 +       return -ENOMEM;
31654 +}
31655 +
31656 +/* ---------------------------------------------------------------------- */
31657 +
31658 +/*
31659 + * test if the @wh_name exists under @h_parent.
31660 + * @try_sio specifies the necessary of super-io.
31661 + */
31662 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio)
31663 +{
31664 +       int err;
31665 +       struct dentry *wh_dentry;
31666 +
31667 +       if (!try_sio)
31668 +               wh_dentry = vfsub_lkup_one(wh_name, h_parent);
31669 +       else
31670 +               wh_dentry = au_sio_lkup_one(wh_name, h_parent);
31671 +       err = PTR_ERR(wh_dentry);
31672 +       if (IS_ERR(wh_dentry))
31673 +               goto out;
31674 +
31675 +       err = 0;
31676 +       if (!wh_dentry->d_inode)
31677 +               goto out_wh; /* success */
31678 +
31679 +       err = 1;
31680 +       if (S_ISREG(wh_dentry->d_inode->i_mode))
31681 +               goto out_wh; /* success */
31682 +
31683 +       err = -EIO;
31684 +       AuIOErr("%pd Invalid whiteout entry type 0%o.\n",
31685 +               wh_dentry, wh_dentry->d_inode->i_mode);
31686 +
31687 +out_wh:
31688 +       dput(wh_dentry);
31689 +out:
31690 +       return err;
31691 +}
31692 +
31693 +/*
31694 + * test if the @h_dentry sets opaque or not.
31695 + */
31696 +int au_diropq_test(struct dentry *h_dentry)
31697 +{
31698 +       int err;
31699 +       struct inode *h_dir;
31700 +
31701 +       h_dir = h_dentry->d_inode;
31702 +       err = au_wh_test(h_dentry, &diropq_name,
31703 +                        au_test_h_perm_sio(h_dir, MAY_EXEC));
31704 +       return err;
31705 +}
31706 +
31707 +/*
31708 + * returns a negative dentry whose name is unique and temporary.
31709 + */
31710 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
31711 +                            struct qstr *prefix)
31712 +{
31713 +       struct dentry *dentry;
31714 +       int i;
31715 +       char defname[NAME_MAX - AUFS_MAX_NAMELEN + DNAME_INLINE_LEN + 1],
31716 +               *name, *p;
31717 +       /* strict atomic_t is unnecessary here */
31718 +       static unsigned short cnt;
31719 +       struct qstr qs;
31720 +
31721 +       BUILD_BUG_ON(sizeof(cnt) * 2 > AUFS_WH_TMP_LEN);
31722 +
31723 +       name = defname;
31724 +       qs.len = sizeof(defname) - DNAME_INLINE_LEN + prefix->len - 1;
31725 +       if (unlikely(prefix->len > DNAME_INLINE_LEN)) {
31726 +               dentry = ERR_PTR(-ENAMETOOLONG);
31727 +               if (unlikely(qs.len > NAME_MAX))
31728 +                       goto out;
31729 +               dentry = ERR_PTR(-ENOMEM);
31730 +               name = kmalloc(qs.len + 1, GFP_NOFS);
31731 +               if (unlikely(!name))
31732 +                       goto out;
31733 +       }
31734 +
31735 +       /* doubly whiteout-ed */
31736 +       memcpy(name, AUFS_WH_PFX AUFS_WH_PFX, AUFS_WH_PFX_LEN * 2);
31737 +       p = name + AUFS_WH_PFX_LEN * 2;
31738 +       memcpy(p, prefix->name, prefix->len);
31739 +       p += prefix->len;
31740 +       *p++ = '.';
31741 +       AuDebugOn(name + qs.len + 1 - p <= AUFS_WH_TMP_LEN);
31742 +
31743 +       qs.name = name;
31744 +       for (i = 0; i < 3; i++) {
31745 +               sprintf(p, "%.*x", AUFS_WH_TMP_LEN, cnt++);
31746 +               dentry = au_sio_lkup_one(&qs, h_parent);
31747 +               if (IS_ERR(dentry) || !dentry->d_inode)
31748 +                       goto out_name;
31749 +               dput(dentry);
31750 +       }
31751 +       /* pr_warn("could not get random name\n"); */
31752 +       dentry = ERR_PTR(-EEXIST);
31753 +       AuDbg("%.*s\n", AuLNPair(&qs));
31754 +       BUG();
31755 +
31756 +out_name:
31757 +       if (name != defname)
31758 +               kfree(name);
31759 +out:
31760 +       AuTraceErrPtr(dentry);
31761 +       return dentry;
31762 +}
31763 +
31764 +/*
31765 + * rename the @h_dentry on @br to the whiteouted temporary name.
31766 + */
31767 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br)
31768 +{
31769 +       int err;
31770 +       struct path h_path = {
31771 +               .mnt = au_br_mnt(br)
31772 +       };
31773 +       struct inode *h_dir, *delegated;
31774 +       struct dentry *h_parent;
31775 +
31776 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
31777 +       h_dir = h_parent->d_inode;
31778 +       IMustLock(h_dir);
31779 +
31780 +       h_path.dentry = au_whtmp_lkup(h_parent, br, &h_dentry->d_name);
31781 +       err = PTR_ERR(h_path.dentry);
31782 +       if (IS_ERR(h_path.dentry))
31783 +               goto out;
31784 +
31785 +       /* under the same dir, no need to lock_rename() */
31786 +       delegated = NULL;
31787 +       err = vfsub_rename(h_dir, h_dentry, h_dir, &h_path, &delegated);
31788 +       AuTraceErr(err);
31789 +       if (unlikely(err == -EWOULDBLOCK)) {
31790 +               pr_warn("cannot retry for NFSv4 delegation"
31791 +                       " for an internal rename\n");
31792 +               iput(delegated);
31793 +       }
31794 +       dput(h_path.dentry);
31795 +
31796 +out:
31797 +       AuTraceErr(err);
31798 +       return err;
31799 +}
31800 +
31801 +/* ---------------------------------------------------------------------- */
31802 +/*
31803 + * functions for removing a whiteout
31804 + */
31805 +
31806 +static int do_unlink_wh(struct inode *h_dir, struct path *h_path)
31807 +{
31808 +       int err, force;
31809 +       struct inode *delegated;
31810 +
31811 +       /*
31812 +        * forces superio when the dir has a sticky bit.
31813 +        * this may be a violation of unix fs semantics.
31814 +        */
31815 +       force = (h_dir->i_mode & S_ISVTX)
31816 +               && !uid_eq(current_fsuid(), h_path->dentry->d_inode->i_uid);
31817 +       delegated = NULL;
31818 +       err = vfsub_unlink(h_dir, h_path, &delegated, force);
31819 +       if (unlikely(err == -EWOULDBLOCK)) {
31820 +               pr_warn("cannot retry for NFSv4 delegation"
31821 +                       " for an internal unlink\n");
31822 +               iput(delegated);
31823 +       }
31824 +       return err;
31825 +}
31826 +
31827 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
31828 +                       struct dentry *dentry)
31829 +{
31830 +       int err;
31831 +
31832 +       err = do_unlink_wh(h_dir, h_path);
31833 +       if (!err && dentry)
31834 +               au_set_dbwh(dentry, -1);
31835 +
31836 +       return err;
31837 +}
31838 +
31839 +static int unlink_wh_name(struct dentry *h_parent, struct qstr *wh,
31840 +                         struct au_branch *br)
31841 +{
31842 +       int err;
31843 +       struct path h_path = {
31844 +               .mnt = au_br_mnt(br)
31845 +       };
31846 +
31847 +       err = 0;
31848 +       h_path.dentry = vfsub_lkup_one(wh, h_parent);
31849 +       if (IS_ERR(h_path.dentry))
31850 +               err = PTR_ERR(h_path.dentry);
31851 +       else {
31852 +               if (h_path.dentry->d_inode
31853 +                   && S_ISREG(h_path.dentry->d_inode->i_mode))
31854 +                       err = do_unlink_wh(h_parent->d_inode, &h_path);
31855 +               dput(h_path.dentry);
31856 +       }
31857 +
31858 +       return err;
31859 +}
31860 +
31861 +/* ---------------------------------------------------------------------- */
31862 +/*
31863 + * initialize/clean whiteout for a branch
31864 + */
31865 +
31866 +static void au_wh_clean(struct inode *h_dir, struct path *whpath,
31867 +                       const int isdir)
31868 +{
31869 +       int err;
31870 +       struct inode *delegated;
31871 +
31872 +       if (!whpath->dentry->d_inode)
31873 +               return;
31874 +
31875 +       if (isdir)
31876 +               err = vfsub_rmdir(h_dir, whpath);
31877 +       else {
31878 +               delegated = NULL;
31879 +               err = vfsub_unlink(h_dir, whpath, &delegated, /*force*/0);
31880 +               if (unlikely(err == -EWOULDBLOCK)) {
31881 +                       pr_warn("cannot retry for NFSv4 delegation"
31882 +                               " for an internal unlink\n");
31883 +                       iput(delegated);
31884 +               }
31885 +       }
31886 +       if (unlikely(err))
31887 +               pr_warn("failed removing %pd (%d), ignored.\n",
31888 +                       whpath->dentry, err);
31889 +}
31890 +
31891 +static int test_linkable(struct dentry *h_root)
31892 +{
31893 +       struct inode *h_dir = h_root->d_inode;
31894 +
31895 +       if (h_dir->i_op->link)
31896 +               return 0;
31897 +
31898 +       pr_err("%pd (%s) doesn't support link(2), use noplink and rw+nolwh\n",
31899 +              h_root, au_sbtype(h_root->d_sb));
31900 +       return -ENOSYS;
31901 +}
31902 +
31903 +/* todo: should this mkdir be done in /sbin/mount.aufs helper? */
31904 +static int au_whdir(struct inode *h_dir, struct path *path)
31905 +{
31906 +       int err;
31907 +
31908 +       err = -EEXIST;
31909 +       if (!path->dentry->d_inode) {
31910 +               int mode = S_IRWXU;
31911 +
31912 +               if (au_test_nfs(path->dentry->d_sb))
31913 +                       mode |= S_IXUGO;
31914 +               err = vfsub_mkdir(h_dir, path, mode);
31915 +       } else if (S_ISDIR(path->dentry->d_inode->i_mode))
31916 +               err = 0;
31917 +       else
31918 +               pr_err("unknown %pd exists\n", path->dentry);
31919 +
31920 +       return err;
31921 +}
31922 +
31923 +struct au_wh_base {
31924 +       const struct qstr *name;
31925 +       struct dentry *dentry;
31926 +};
31927 +
31928 +static void au_wh_init_ro(struct inode *h_dir, struct au_wh_base base[],
31929 +                         struct path *h_path)
31930 +{
31931 +       h_path->dentry = base[AuBrWh_BASE].dentry;
31932 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
31933 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
31934 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
31935 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
31936 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
31937 +}
31938 +
31939 +/*
31940 + * returns tri-state,
31941 + * minus: error, caller should print the message
31942 + * zero: succuess
31943 + * plus: error, caller should NOT print the message
31944 + */
31945 +static int au_wh_init_rw_nolink(struct dentry *h_root, struct au_wbr *wbr,
31946 +                               int do_plink, struct au_wh_base base[],
31947 +                               struct path *h_path)
31948 +{
31949 +       int err;
31950 +       struct inode *h_dir;
31951 +
31952 +       h_dir = h_root->d_inode;
31953 +       h_path->dentry = base[AuBrWh_BASE].dentry;
31954 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
31955 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
31956 +       if (do_plink) {
31957 +               err = test_linkable(h_root);
31958 +               if (unlikely(err)) {
31959 +                       err = 1;
31960 +                       goto out;
31961 +               }
31962 +
31963 +               err = au_whdir(h_dir, h_path);
31964 +               if (unlikely(err))
31965 +                       goto out;
31966 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
31967 +       } else
31968 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
31969 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
31970 +       err = au_whdir(h_dir, h_path);
31971 +       if (unlikely(err))
31972 +               goto out;
31973 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
31974 +
31975 +out:
31976 +       return err;
31977 +}
31978 +
31979 +/*
31980 + * for the moment, aufs supports the branch filesystem which does not support
31981 + * link(2). testing on FAT which does not support i_op->setattr() fully either,
31982 + * copyup failed. finally, such filesystem will not be used as the writable
31983 + * branch.
31984 + *
31985 + * returns tri-state, see above.
31986 + */
31987 +static int au_wh_init_rw(struct dentry *h_root, struct au_wbr *wbr,
31988 +                        int do_plink, struct au_wh_base base[],
31989 +                        struct path *h_path)
31990 +{
31991 +       int err;
31992 +       struct inode *h_dir;
31993 +
31994 +       WbrWhMustWriteLock(wbr);
31995 +
31996 +       err = test_linkable(h_root);
31997 +       if (unlikely(err)) {
31998 +               err = 1;
31999 +               goto out;
32000 +       }
32001 +
32002 +       /*
32003 +        * todo: should this create be done in /sbin/mount.aufs helper?
32004 +        */
32005 +       err = -EEXIST;
32006 +       h_dir = h_root->d_inode;
32007 +       if (!base[AuBrWh_BASE].dentry->d_inode) {
32008 +               h_path->dentry = base[AuBrWh_BASE].dentry;
32009 +               err = vfsub_create(h_dir, h_path, WH_MASK, /*want_excl*/true);
32010 +       } else if (S_ISREG(base[AuBrWh_BASE].dentry->d_inode->i_mode))
32011 +               err = 0;
32012 +       else
32013 +               pr_err("unknown %pd2 exists\n", base[AuBrWh_BASE].dentry);
32014 +       if (unlikely(err))
32015 +               goto out;
32016 +
32017 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
32018 +       if (do_plink) {
32019 +               err = au_whdir(h_dir, h_path);
32020 +               if (unlikely(err))
32021 +                       goto out;
32022 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
32023 +       } else
32024 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
32025 +       wbr->wbr_whbase = dget(base[AuBrWh_BASE].dentry);
32026 +
32027 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
32028 +       err = au_whdir(h_dir, h_path);
32029 +       if (unlikely(err))
32030 +               goto out;
32031 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
32032 +
32033 +out:
32034 +       return err;
32035 +}
32036 +
32037 +/*
32038 + * initialize the whiteout base file/dir for @br.
32039 + */
32040 +int au_wh_init(struct au_branch *br, struct super_block *sb)
32041 +{
32042 +       int err, i;
32043 +       const unsigned char do_plink
32044 +               = !!au_opt_test(au_mntflags(sb), PLINK);
32045 +       struct inode *h_dir;
32046 +       struct path path = br->br_path;
32047 +       struct dentry *h_root = path.dentry;
32048 +       struct au_wbr *wbr = br->br_wbr;
32049 +       static const struct qstr base_name[] = {
32050 +               [AuBrWh_BASE] = QSTR_INIT(AUFS_BASE_NAME,
32051 +                                         sizeof(AUFS_BASE_NAME) - 1),
32052 +               [AuBrWh_PLINK] = QSTR_INIT(AUFS_PLINKDIR_NAME,
32053 +                                          sizeof(AUFS_PLINKDIR_NAME) - 1),
32054 +               [AuBrWh_ORPH] = QSTR_INIT(AUFS_ORPHDIR_NAME,
32055 +                                         sizeof(AUFS_ORPHDIR_NAME) - 1)
32056 +       };
32057 +       struct au_wh_base base[] = {
32058 +               [AuBrWh_BASE] = {
32059 +                       .name   = base_name + AuBrWh_BASE,
32060 +                       .dentry = NULL
32061 +               },
32062 +               [AuBrWh_PLINK] = {
32063 +                       .name   = base_name + AuBrWh_PLINK,
32064 +                       .dentry = NULL
32065 +               },
32066 +               [AuBrWh_ORPH] = {
32067 +                       .name   = base_name + AuBrWh_ORPH,
32068 +                       .dentry = NULL
32069 +               }
32070 +       };
32071 +
32072 +       if (wbr)
32073 +               WbrWhMustWriteLock(wbr);
32074 +
32075 +       for (i = 0; i < AuBrWh_Last; i++) {
32076 +               /* doubly whiteouted */
32077 +               struct dentry *d;
32078 +
32079 +               d = au_wh_lkup(h_root, (void *)base[i].name, br);
32080 +               err = PTR_ERR(d);
32081 +               if (IS_ERR(d))
32082 +                       goto out;
32083 +
32084 +               base[i].dentry = d;
32085 +               AuDebugOn(wbr
32086 +                         && wbr->wbr_wh[i]
32087 +                         && wbr->wbr_wh[i] != base[i].dentry);
32088 +       }
32089 +
32090 +       if (wbr)
32091 +               for (i = 0; i < AuBrWh_Last; i++) {
32092 +                       dput(wbr->wbr_wh[i]);
32093 +                       wbr->wbr_wh[i] = NULL;
32094 +               }
32095 +
32096 +       err = 0;
32097 +       if (!au_br_writable(br->br_perm)) {
32098 +               h_dir = h_root->d_inode;
32099 +               au_wh_init_ro(h_dir, base, &path);
32100 +       } else if (!au_br_wh_linkable(br->br_perm)) {
32101 +               err = au_wh_init_rw_nolink(h_root, wbr, do_plink, base, &path);
32102 +               if (err > 0)
32103 +                       goto out;
32104 +               else if (err)
32105 +                       goto out_err;
32106 +       } else {
32107 +               err = au_wh_init_rw(h_root, wbr, do_plink, base, &path);
32108 +               if (err > 0)
32109 +                       goto out;
32110 +               else if (err)
32111 +                       goto out_err;
32112 +       }
32113 +       goto out; /* success */
32114 +
32115 +out_err:
32116 +       pr_err("an error(%d) on the writable branch %pd(%s)\n",
32117 +              err, h_root, au_sbtype(h_root->d_sb));
32118 +out:
32119 +       for (i = 0; i < AuBrWh_Last; i++)
32120 +               dput(base[i].dentry);
32121 +       return err;
32122 +}
32123 +
32124 +/* ---------------------------------------------------------------------- */
32125 +/*
32126 + * whiteouts are all hard-linked usually.
32127 + * when its link count reaches a ceiling, we create a new whiteout base
32128 + * asynchronously.
32129 + */
32130 +
32131 +struct reinit_br_wh {
32132 +       struct super_block *sb;
32133 +       struct au_branch *br;
32134 +};
32135 +
32136 +static void reinit_br_wh(void *arg)
32137 +{
32138 +       int err;
32139 +       aufs_bindex_t bindex;
32140 +       struct path h_path;
32141 +       struct reinit_br_wh *a = arg;
32142 +       struct au_wbr *wbr;
32143 +       struct inode *dir, *delegated;
32144 +       struct dentry *h_root;
32145 +       struct au_hinode *hdir;
32146 +
32147 +       err = 0;
32148 +       wbr = a->br->br_wbr;
32149 +       /* big aufs lock */
32150 +       si_noflush_write_lock(a->sb);
32151 +       if (!au_br_writable(a->br->br_perm))
32152 +               goto out;
32153 +       bindex = au_br_index(a->sb, a->br->br_id);
32154 +       if (unlikely(bindex < 0))
32155 +               goto out;
32156 +
32157 +       di_read_lock_parent(a->sb->s_root, AuLock_IR);
32158 +       dir = a->sb->s_root->d_inode;
32159 +       hdir = au_hi(dir, bindex);
32160 +       h_root = au_h_dptr(a->sb->s_root, bindex);
32161 +       AuDebugOn(h_root != au_br_dentry(a->br));
32162 +
32163 +       au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
32164 +       wbr_wh_write_lock(wbr);
32165 +       err = au_h_verify(wbr->wbr_whbase, au_opt_udba(a->sb), hdir->hi_inode,
32166 +                         h_root, a->br);
32167 +       if (!err) {
32168 +               h_path.dentry = wbr->wbr_whbase;
32169 +               h_path.mnt = au_br_mnt(a->br);
32170 +               delegated = NULL;
32171 +               err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated,
32172 +                                  /*force*/0);
32173 +               if (unlikely(err == -EWOULDBLOCK)) {
32174 +                       pr_warn("cannot retry for NFSv4 delegation"
32175 +                               " for an internal unlink\n");
32176 +                       iput(delegated);
32177 +               }
32178 +       } else {
32179 +               pr_warn("%pd is moved, ignored\n", wbr->wbr_whbase);
32180 +               err = 0;
32181 +       }
32182 +       dput(wbr->wbr_whbase);
32183 +       wbr->wbr_whbase = NULL;
32184 +       if (!err)
32185 +               err = au_wh_init(a->br, a->sb);
32186 +       wbr_wh_write_unlock(wbr);
32187 +       au_hn_imtx_unlock(hdir);
32188 +       di_read_unlock(a->sb->s_root, AuLock_IR);
32189 +       if (!err)
32190 +               au_fhsm_wrote(a->sb, bindex, /*force*/0);
32191 +
32192 +out:
32193 +       if (wbr)
32194 +               atomic_dec(&wbr->wbr_wh_running);
32195 +       atomic_dec(&a->br->br_count);
32196 +       si_write_unlock(a->sb);
32197 +       au_nwt_done(&au_sbi(a->sb)->si_nowait);
32198 +       kfree(arg);
32199 +       if (unlikely(err))
32200 +               AuIOErr("err %d\n", err);
32201 +}
32202 +
32203 +static void kick_reinit_br_wh(struct super_block *sb, struct au_branch *br)
32204 +{
32205 +       int do_dec, wkq_err;
32206 +       struct reinit_br_wh *arg;
32207 +
32208 +       do_dec = 1;
32209 +       if (atomic_inc_return(&br->br_wbr->wbr_wh_running) != 1)
32210 +               goto out;
32211 +
32212 +       /* ignore ENOMEM */
32213 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
32214 +       if (arg) {
32215 +               /*
32216 +                * dec(wh_running), kfree(arg) and dec(br_count)
32217 +                * in reinit function
32218 +                */
32219 +               arg->sb = sb;
32220 +               arg->br = br;
32221 +               atomic_inc(&br->br_count);
32222 +               wkq_err = au_wkq_nowait(reinit_br_wh, arg, sb, /*flags*/0);
32223 +               if (unlikely(wkq_err)) {
32224 +                       atomic_dec(&br->br_wbr->wbr_wh_running);
32225 +                       atomic_dec(&br->br_count);
32226 +                       kfree(arg);
32227 +               }
32228 +               do_dec = 0;
32229 +       }
32230 +
32231 +out:
32232 +       if (do_dec)
32233 +               atomic_dec(&br->br_wbr->wbr_wh_running);
32234 +}
32235 +
32236 +/* ---------------------------------------------------------------------- */
32237 +
32238 +/*
32239 + * create the whiteout @wh.
32240 + */
32241 +static int link_or_create_wh(struct super_block *sb, aufs_bindex_t bindex,
32242 +                            struct dentry *wh)
32243 +{
32244 +       int err;
32245 +       struct path h_path = {
32246 +               .dentry = wh
32247 +       };
32248 +       struct au_branch *br;
32249 +       struct au_wbr *wbr;
32250 +       struct dentry *h_parent;
32251 +       struct inode *h_dir, *delegated;
32252 +
32253 +       h_parent = wh->d_parent; /* dir inode is locked */
32254 +       h_dir = h_parent->d_inode;
32255 +       IMustLock(h_dir);
32256 +
32257 +       br = au_sbr(sb, bindex);
32258 +       h_path.mnt = au_br_mnt(br);
32259 +       wbr = br->br_wbr;
32260 +       wbr_wh_read_lock(wbr);
32261 +       if (wbr->wbr_whbase) {
32262 +               delegated = NULL;
32263 +               err = vfsub_link(wbr->wbr_whbase, h_dir, &h_path, &delegated);
32264 +               if (unlikely(err == -EWOULDBLOCK)) {
32265 +                       pr_warn("cannot retry for NFSv4 delegation"
32266 +                               " for an internal link\n");
32267 +                       iput(delegated);
32268 +               }
32269 +               if (!err || err != -EMLINK)
32270 +                       goto out;
32271 +
32272 +               /* link count full. re-initialize br_whbase. */
32273 +               kick_reinit_br_wh(sb, br);
32274 +       }
32275 +
32276 +       /* return this error in this context */
32277 +       err = vfsub_create(h_dir, &h_path, WH_MASK, /*want_excl*/true);
32278 +       if (!err)
32279 +               au_fhsm_wrote(sb, bindex, /*force*/0);
32280 +
32281 +out:
32282 +       wbr_wh_read_unlock(wbr);
32283 +       return err;
32284 +}
32285 +
32286 +/* ---------------------------------------------------------------------- */
32287 +
32288 +/*
32289 + * create or remove the diropq.
32290 + */
32291 +static struct dentry *do_diropq(struct dentry *dentry, aufs_bindex_t bindex,
32292 +                               unsigned int flags)
32293 +{
32294 +       struct dentry *opq_dentry, *h_dentry;
32295 +       struct super_block *sb;
32296 +       struct au_branch *br;
32297 +       int err;
32298 +
32299 +       sb = dentry->d_sb;
32300 +       br = au_sbr(sb, bindex);
32301 +       h_dentry = au_h_dptr(dentry, bindex);
32302 +       opq_dentry = vfsub_lkup_one(&diropq_name, h_dentry);
32303 +       if (IS_ERR(opq_dentry))
32304 +               goto out;
32305 +
32306 +       if (au_ftest_diropq(flags, CREATE)) {
32307 +               err = link_or_create_wh(sb, bindex, opq_dentry);
32308 +               if (!err) {
32309 +                       au_set_dbdiropq(dentry, bindex);
32310 +                       goto out; /* success */
32311 +               }
32312 +       } else {
32313 +               struct path tmp = {
32314 +                       .dentry = opq_dentry,
32315 +                       .mnt    = au_br_mnt(br)
32316 +               };
32317 +               err = do_unlink_wh(au_h_iptr(dentry->d_inode, bindex), &tmp);
32318 +               if (!err)
32319 +                       au_set_dbdiropq(dentry, -1);
32320 +       }
32321 +       dput(opq_dentry);
32322 +       opq_dentry = ERR_PTR(err);
32323 +
32324 +out:
32325 +       return opq_dentry;
32326 +}
32327 +
32328 +struct do_diropq_args {
32329 +       struct dentry **errp;
32330 +       struct dentry *dentry;
32331 +       aufs_bindex_t bindex;
32332 +       unsigned int flags;
32333 +};
32334 +
32335 +static void call_do_diropq(void *args)
32336 +{
32337 +       struct do_diropq_args *a = args;
32338 +       *a->errp = do_diropq(a->dentry, a->bindex, a->flags);
32339 +}
32340 +
32341 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
32342 +                            unsigned int flags)
32343 +{
32344 +       struct dentry *diropq, *h_dentry;
32345 +
32346 +       h_dentry = au_h_dptr(dentry, bindex);
32347 +       if (!au_test_h_perm_sio(h_dentry->d_inode, MAY_EXEC | MAY_WRITE))
32348 +               diropq = do_diropq(dentry, bindex, flags);
32349 +       else {
32350 +               int wkq_err;
32351 +               struct do_diropq_args args = {
32352 +                       .errp           = &diropq,
32353 +                       .dentry         = dentry,
32354 +                       .bindex         = bindex,
32355 +                       .flags          = flags
32356 +               };
32357 +
32358 +               wkq_err = au_wkq_wait(call_do_diropq, &args);
32359 +               if (unlikely(wkq_err))
32360 +                       diropq = ERR_PTR(wkq_err);
32361 +       }
32362 +
32363 +       return diropq;
32364 +}
32365 +
32366 +/* ---------------------------------------------------------------------- */
32367 +
32368 +/*
32369 + * lookup whiteout dentry.
32370 + * @h_parent: lower parent dentry which must exist and be locked
32371 + * @base_name: name of dentry which will be whiteouted
32372 + * returns dentry for whiteout.
32373 + */
32374 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
32375 +                         struct au_branch *br)
32376 +{
32377 +       int err;
32378 +       struct qstr wh_name;
32379 +       struct dentry *wh_dentry;
32380 +
32381 +       err = au_wh_name_alloc(&wh_name, base_name);
32382 +       wh_dentry = ERR_PTR(err);
32383 +       if (!err) {
32384 +               wh_dentry = vfsub_lkup_one(&wh_name, h_parent);
32385 +               kfree(wh_name.name);
32386 +       }
32387 +       return wh_dentry;
32388 +}
32389 +
32390 +/*
32391 + * link/create a whiteout for @dentry on @bindex.
32392 + */
32393 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
32394 +                           struct dentry *h_parent)
32395 +{
32396 +       struct dentry *wh_dentry;
32397 +       struct super_block *sb;
32398 +       int err;
32399 +
32400 +       sb = dentry->d_sb;
32401 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, au_sbr(sb, bindex));
32402 +       if (!IS_ERR(wh_dentry) && !wh_dentry->d_inode) {
32403 +               err = link_or_create_wh(sb, bindex, wh_dentry);
32404 +               if (!err) {
32405 +                       au_set_dbwh(dentry, bindex);
32406 +                       au_fhsm_wrote(sb, bindex, /*force*/0);
32407 +               } else {
32408 +                       dput(wh_dentry);
32409 +                       wh_dentry = ERR_PTR(err);
32410 +               }
32411 +       }
32412 +
32413 +       return wh_dentry;
32414 +}
32415 +
32416 +/* ---------------------------------------------------------------------- */
32417 +
32418 +/* Delete all whiteouts in this directory on branch bindex. */
32419 +static int del_wh_children(struct dentry *h_dentry, struct au_nhash *whlist,
32420 +                          aufs_bindex_t bindex, struct au_branch *br)
32421 +{
32422 +       int err;
32423 +       unsigned long ul, n;
32424 +       struct qstr wh_name;
32425 +       char *p;
32426 +       struct hlist_head *head;
32427 +       struct au_vdir_wh *pos;
32428 +       struct au_vdir_destr *str;
32429 +
32430 +       err = -ENOMEM;
32431 +       p = (void *)__get_free_page(GFP_NOFS);
32432 +       wh_name.name = p;
32433 +       if (unlikely(!wh_name.name))
32434 +               goto out;
32435 +
32436 +       err = 0;
32437 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
32438 +       p += AUFS_WH_PFX_LEN;
32439 +       n = whlist->nh_num;
32440 +       head = whlist->nh_head;
32441 +       for (ul = 0; !err && ul < n; ul++, head++) {
32442 +               hlist_for_each_entry(pos, head, wh_hash) {
32443 +                       if (pos->wh_bindex != bindex)
32444 +                               continue;
32445 +
32446 +                       str = &pos->wh_str;
32447 +                       if (str->len + AUFS_WH_PFX_LEN <= PATH_MAX) {
32448 +                               memcpy(p, str->name, str->len);
32449 +                               wh_name.len = AUFS_WH_PFX_LEN + str->len;
32450 +                               err = unlink_wh_name(h_dentry, &wh_name, br);
32451 +                               if (!err)
32452 +                                       continue;
32453 +                               break;
32454 +                       }
32455 +                       AuIOErr("whiteout name too long %.*s\n",
32456 +                               str->len, str->name);
32457 +                       err = -EIO;
32458 +                       break;
32459 +               }
32460 +       }
32461 +       free_page((unsigned long)wh_name.name);
32462 +
32463 +out:
32464 +       return err;
32465 +}
32466 +
32467 +struct del_wh_children_args {
32468 +       int *errp;
32469 +       struct dentry *h_dentry;
32470 +       struct au_nhash *whlist;
32471 +       aufs_bindex_t bindex;
32472 +       struct au_branch *br;
32473 +};
32474 +
32475 +static void call_del_wh_children(void *args)
32476 +{
32477 +       struct del_wh_children_args *a = args;
32478 +       *a->errp = del_wh_children(a->h_dentry, a->whlist, a->bindex, a->br);
32479 +}
32480 +
32481 +/* ---------------------------------------------------------------------- */
32482 +
32483 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp)
32484 +{
32485 +       struct au_whtmp_rmdir *whtmp;
32486 +       int err;
32487 +       unsigned int rdhash;
32488 +
32489 +       SiMustAnyLock(sb);
32490 +
32491 +       whtmp = kmalloc(sizeof(*whtmp), gfp);
32492 +       if (unlikely(!whtmp)) {
32493 +               whtmp = ERR_PTR(-ENOMEM);
32494 +               goto out;
32495 +       }
32496 +
32497 +       whtmp->dir = NULL;
32498 +       whtmp->br = NULL;
32499 +       whtmp->wh_dentry = NULL;
32500 +       /* no estimation for dir size */
32501 +       rdhash = au_sbi(sb)->si_rdhash;
32502 +       if (!rdhash)
32503 +               rdhash = AUFS_RDHASH_DEF;
32504 +       err = au_nhash_alloc(&whtmp->whlist, rdhash, gfp);
32505 +       if (unlikely(err)) {
32506 +               kfree(whtmp);
32507 +               whtmp = ERR_PTR(err);
32508 +       }
32509 +
32510 +out:
32511 +       return whtmp;
32512 +}
32513 +
32514 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp)
32515 +{
32516 +       if (whtmp->br)
32517 +               atomic_dec(&whtmp->br->br_count);
32518 +       dput(whtmp->wh_dentry);
32519 +       iput(whtmp->dir);
32520 +       au_nhash_wh_free(&whtmp->whlist);
32521 +       kfree(whtmp);
32522 +}
32523 +
32524 +/*
32525 + * rmdir the whiteouted temporary named dir @h_dentry.
32526 + * @whlist: whiteouted children.
32527 + */
32528 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
32529 +                  struct dentry *wh_dentry, struct au_nhash *whlist)
32530 +{
32531 +       int err;
32532 +       struct path h_tmp;
32533 +       struct inode *wh_inode, *h_dir;
32534 +       struct au_branch *br;
32535 +
32536 +       h_dir = wh_dentry->d_parent->d_inode; /* dir inode is locked */
32537 +       IMustLock(h_dir);
32538 +
32539 +       br = au_sbr(dir->i_sb, bindex);
32540 +       wh_inode = wh_dentry->d_inode;
32541 +       mutex_lock_nested(&wh_inode->i_mutex, AuLsc_I_CHILD);
32542 +
32543 +       /*
32544 +        * someone else might change some whiteouts while we were sleeping.
32545 +        * it means this whlist may have an obsoleted entry.
32546 +        */
32547 +       if (!au_test_h_perm_sio(wh_inode, MAY_EXEC | MAY_WRITE))
32548 +               err = del_wh_children(wh_dentry, whlist, bindex, br);
32549 +       else {
32550 +               int wkq_err;
32551 +               struct del_wh_children_args args = {
32552 +                       .errp           = &err,
32553 +                       .h_dentry       = wh_dentry,
32554 +                       .whlist         = whlist,
32555 +                       .bindex         = bindex,
32556 +                       .br             = br
32557 +               };
32558 +
32559 +               wkq_err = au_wkq_wait(call_del_wh_children, &args);
32560 +               if (unlikely(wkq_err))
32561 +                       err = wkq_err;
32562 +       }
32563 +       mutex_unlock(&wh_inode->i_mutex);
32564 +
32565 +       if (!err) {
32566 +               h_tmp.dentry = wh_dentry;
32567 +               h_tmp.mnt = au_br_mnt(br);
32568 +               err = vfsub_rmdir(h_dir, &h_tmp);
32569 +       }
32570 +
32571 +       if (!err) {
32572 +               if (au_ibstart(dir) == bindex) {
32573 +                       /* todo: dir->i_mutex is necessary */
32574 +                       au_cpup_attr_timesizes(dir);
32575 +                       vfsub_drop_nlink(dir);
32576 +               }
32577 +               return 0; /* success */
32578 +       }
32579 +
32580 +       pr_warn("failed removing %pd(%d), ignored\n", wh_dentry, err);
32581 +       return err;
32582 +}
32583 +
32584 +static void call_rmdir_whtmp(void *args)
32585 +{
32586 +       int err;
32587 +       aufs_bindex_t bindex;
32588 +       struct au_whtmp_rmdir *a = args;
32589 +       struct super_block *sb;
32590 +       struct dentry *h_parent;
32591 +       struct inode *h_dir;
32592 +       struct au_hinode *hdir;
32593 +
32594 +       /* rmdir by nfsd may cause deadlock with this i_mutex */
32595 +       /* mutex_lock(&a->dir->i_mutex); */
32596 +       err = -EROFS;
32597 +       sb = a->dir->i_sb;
32598 +       si_read_lock(sb, !AuLock_FLUSH);
32599 +       if (!au_br_writable(a->br->br_perm))
32600 +               goto out;
32601 +       bindex = au_br_index(sb, a->br->br_id);
32602 +       if (unlikely(bindex < 0))
32603 +               goto out;
32604 +
32605 +       err = -EIO;
32606 +       ii_write_lock_parent(a->dir);
32607 +       h_parent = dget_parent(a->wh_dentry);
32608 +       h_dir = h_parent->d_inode;
32609 +       hdir = au_hi(a->dir, bindex);
32610 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
32611 +       if (unlikely(err))
32612 +               goto out_mnt;
32613 +       au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
32614 +       err = au_h_verify(a->wh_dentry, au_opt_udba(sb), h_dir, h_parent,
32615 +                         a->br);
32616 +       if (!err)
32617 +               err = au_whtmp_rmdir(a->dir, bindex, a->wh_dentry, &a->whlist);
32618 +       au_hn_imtx_unlock(hdir);
32619 +       vfsub_mnt_drop_write(au_br_mnt(a->br));
32620 +
32621 +out_mnt:
32622 +       dput(h_parent);
32623 +       ii_write_unlock(a->dir);
32624 +out:
32625 +       /* mutex_unlock(&a->dir->i_mutex); */
32626 +       au_whtmp_rmdir_free(a);
32627 +       si_read_unlock(sb);
32628 +       au_nwt_done(&au_sbi(sb)->si_nowait);
32629 +       if (unlikely(err))
32630 +               AuIOErr("err %d\n", err);
32631 +}
32632 +
32633 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
32634 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args)
32635 +{
32636 +       int wkq_err;
32637 +       struct super_block *sb;
32638 +
32639 +       IMustLock(dir);
32640 +
32641 +       /* all post-process will be done in do_rmdir_whtmp(). */
32642 +       sb = dir->i_sb;
32643 +       args->dir = au_igrab(dir);
32644 +       args->br = au_sbr(sb, bindex);
32645 +       atomic_inc(&args->br->br_count);
32646 +       args->wh_dentry = dget(wh_dentry);
32647 +       wkq_err = au_wkq_nowait(call_rmdir_whtmp, args, sb, /*flags*/0);
32648 +       if (unlikely(wkq_err)) {
32649 +               pr_warn("rmdir error %pd (%d), ignored\n", wh_dentry, wkq_err);
32650 +               au_whtmp_rmdir_free(args);
32651 +       }
32652 +}
32653 diff -urN /usr/share/empty/fs/aufs/whout.h linux/fs/aufs/whout.h
32654 --- /usr/share/empty/fs/aufs/whout.h    1970-01-01 01:00:00.000000000 +0100
32655 +++ linux/fs/aufs/whout.h       2015-01-25 13:00:38.634380408 +0100
32656 @@ -0,0 +1,85 @@
32657 +/*
32658 + * Copyright (C) 2005-2014 Junjiro R. Okajima
32659 + *
32660 + * This program, aufs is free software; you can redistribute it and/or modify
32661 + * it under the terms of the GNU General Public License as published by
32662 + * the Free Software Foundation; either version 2 of the License, or
32663 + * (at your option) any later version.
32664 + *
32665 + * This program is distributed in the hope that it will be useful,
32666 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32667 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32668 + * GNU General Public License for more details.
32669 + *
32670 + * You should have received a copy of the GNU General Public License
32671 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32672 + */
32673 +
32674 +/*
32675 + * whiteout for logical deletion and opaque directory
32676 + */
32677 +
32678 +#ifndef __AUFS_WHOUT_H__
32679 +#define __AUFS_WHOUT_H__
32680 +
32681 +#ifdef __KERNEL__
32682 +
32683 +#include "dir.h"
32684 +
32685 +/* whout.c */
32686 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name);
32687 +struct au_branch;
32688 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio);
32689 +int au_diropq_test(struct dentry *h_dentry);
32690 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
32691 +                            struct qstr *prefix);
32692 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br);
32693 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
32694 +                       struct dentry *dentry);
32695 +int au_wh_init(struct au_branch *br, struct super_block *sb);
32696 +
32697 +/* diropq flags */
32698 +#define AuDiropq_CREATE        1
32699 +#define au_ftest_diropq(flags, name)   ((flags) & AuDiropq_##name)
32700 +#define au_fset_diropq(flags, name) \
32701 +       do { (flags) |= AuDiropq_##name; } while (0)
32702 +#define au_fclr_diropq(flags, name) \
32703 +       do { (flags) &= ~AuDiropq_##name; } while (0)
32704 +
32705 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
32706 +                            unsigned int flags);
32707 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
32708 +                         struct au_branch *br);
32709 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
32710 +                           struct dentry *h_parent);
32711 +
32712 +/* real rmdir for the whiteout-ed dir */
32713 +struct au_whtmp_rmdir {
32714 +       struct inode *dir;
32715 +       struct au_branch *br;
32716 +       struct dentry *wh_dentry;
32717 +       struct au_nhash whlist;
32718 +};
32719 +
32720 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp);
32721 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp);
32722 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
32723 +                  struct dentry *wh_dentry, struct au_nhash *whlist);
32724 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
32725 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args);
32726 +
32727 +/* ---------------------------------------------------------------------- */
32728 +
32729 +static inline struct dentry *au_diropq_create(struct dentry *dentry,
32730 +                                             aufs_bindex_t bindex)
32731 +{
32732 +       return au_diropq_sio(dentry, bindex, AuDiropq_CREATE);
32733 +}
32734 +
32735 +static inline int au_diropq_remove(struct dentry *dentry, aufs_bindex_t bindex)
32736 +{
32737 +       return PTR_ERR(au_diropq_sio(dentry, bindex, !AuDiropq_CREATE));
32738 +}
32739 +
32740 +#endif /* __KERNEL__ */
32741 +#endif /* __AUFS_WHOUT_H__ */
32742 diff -urN /usr/share/empty/fs/aufs/wkq.c linux/fs/aufs/wkq.c
32743 --- /usr/share/empty/fs/aufs/wkq.c      1970-01-01 01:00:00.000000000 +0100
32744 +++ linux/fs/aufs/wkq.c 2015-01-25 13:00:38.634380408 +0100
32745 @@ -0,0 +1,213 @@
32746 +/*
32747 + * Copyright (C) 2005-2014 Junjiro R. Okajima
32748 + *
32749 + * This program, aufs is free software; you can redistribute it and/or modify
32750 + * it under the terms of the GNU General Public License as published by
32751 + * the Free Software Foundation; either version 2 of the License, or
32752 + * (at your option) any later version.
32753 + *
32754 + * This program is distributed in the hope that it will be useful,
32755 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32756 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32757 + * GNU General Public License for more details.
32758 + *
32759 + * You should have received a copy of the GNU General Public License
32760 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32761 + */
32762 +
32763 +/*
32764 + * workqueue for asynchronous/super-io operations
32765 + * todo: try new dredential scheme
32766 + */
32767 +
32768 +#include <linux/module.h>
32769 +#include "aufs.h"
32770 +
32771 +/* internal workqueue named AUFS_WKQ_NAME */
32772 +
32773 +static struct workqueue_struct *au_wkq;
32774 +
32775 +struct au_wkinfo {
32776 +       struct work_struct wk;
32777 +       struct kobject *kobj;
32778 +
32779 +       unsigned int flags; /* see wkq.h */
32780 +
32781 +       au_wkq_func_t func;
32782 +       void *args;
32783 +
32784 +       struct completion *comp;
32785 +};
32786 +
32787 +/* ---------------------------------------------------------------------- */
32788 +
32789 +static void wkq_func(struct work_struct *wk)
32790 +{
32791 +       struct au_wkinfo *wkinfo = container_of(wk, struct au_wkinfo, wk);
32792 +
32793 +       AuDebugOn(!uid_eq(current_fsuid(), GLOBAL_ROOT_UID));
32794 +       AuDebugOn(rlimit(RLIMIT_FSIZE) != RLIM_INFINITY);
32795 +
32796 +       wkinfo->func(wkinfo->args);
32797 +       if (au_ftest_wkq(wkinfo->flags, WAIT))
32798 +               complete(wkinfo->comp);
32799 +       else {
32800 +               kobject_put(wkinfo->kobj);
32801 +               module_put(THIS_MODULE); /* todo: ?? */
32802 +               kfree(wkinfo);
32803 +       }
32804 +}
32805 +
32806 +/*
32807 + * Since struct completion is large, try allocating it dynamically.
32808 + */
32809 +#if 1 /* defined(CONFIG_4KSTACKS) || defined(AuTest4KSTACKS) */
32810 +#define AuWkqCompDeclare(name) struct completion *comp = NULL
32811 +
32812 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
32813 +{
32814 +       *comp = kmalloc(sizeof(**comp), GFP_NOFS);
32815 +       if (*comp) {
32816 +               init_completion(*comp);
32817 +               wkinfo->comp = *comp;
32818 +               return 0;
32819 +       }
32820 +       return -ENOMEM;
32821 +}
32822 +
32823 +static void au_wkq_comp_free(struct completion *comp)
32824 +{
32825 +       kfree(comp);
32826 +}
32827 +
32828 +#else
32829 +
32830 +/* no braces */
32831 +#define AuWkqCompDeclare(name) \
32832 +       DECLARE_COMPLETION_ONSTACK(_ ## name); \
32833 +       struct completion *comp = &_ ## name
32834 +
32835 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
32836 +{
32837 +       wkinfo->comp = *comp;
32838 +       return 0;
32839 +}
32840 +
32841 +static void au_wkq_comp_free(struct completion *comp __maybe_unused)
32842 +{
32843 +       /* empty */
32844 +}
32845 +#endif /* 4KSTACKS */
32846 +
32847 +static void au_wkq_run(struct au_wkinfo *wkinfo)
32848 +{
32849 +       if (au_ftest_wkq(wkinfo->flags, NEST)) {
32850 +               if (au_wkq_test()) {
32851 +                       AuWarn1("wkq from wkq, unless silly-rename on NFS,"
32852 +                               " due to a dead dir by UDBA?\n");
32853 +                       AuDebugOn(au_ftest_wkq(wkinfo->flags, WAIT));
32854 +               }
32855 +       } else
32856 +               au_dbg_verify_kthread();
32857 +
32858 +       if (au_ftest_wkq(wkinfo->flags, WAIT)) {
32859 +               INIT_WORK_ONSTACK(&wkinfo->wk, wkq_func);
32860 +               queue_work(au_wkq, &wkinfo->wk);
32861 +       } else {
32862 +               INIT_WORK(&wkinfo->wk, wkq_func);
32863 +               schedule_work(&wkinfo->wk);
32864 +       }
32865 +}
32866 +
32867 +/*
32868 + * Be careful. It is easy to make deadlock happen.
32869 + * processA: lock, wkq and wait
32870 + * processB: wkq and wait, lock in wkq
32871 + * --> deadlock
32872 + */
32873 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args)
32874 +{
32875 +       int err;
32876 +       AuWkqCompDeclare(comp);
32877 +       struct au_wkinfo wkinfo = {
32878 +               .flags  = flags,
32879 +               .func   = func,
32880 +               .args   = args
32881 +       };
32882 +
32883 +       err = au_wkq_comp_alloc(&wkinfo, &comp);
32884 +       if (!err) {
32885 +               au_wkq_run(&wkinfo);
32886 +               /* no timeout, no interrupt */
32887 +               wait_for_completion(wkinfo.comp);
32888 +               au_wkq_comp_free(comp);
32889 +               destroy_work_on_stack(&wkinfo.wk);
32890 +       }
32891 +
32892 +       return err;
32893 +
32894 +}
32895 +
32896 +/*
32897 + * Note: dget/dput() in func for aufs dentries are not supported. It will be a
32898 + * problem in a concurrent umounting.
32899 + */
32900 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
32901 +                 unsigned int flags)
32902 +{
32903 +       int err;
32904 +       struct au_wkinfo *wkinfo;
32905 +
32906 +       atomic_inc(&au_sbi(sb)->si_nowait.nw_len);
32907 +
32908 +       /*
32909 +        * wkq_func() must free this wkinfo.
32910 +        * it highly depends upon the implementation of workqueue.
32911 +        */
32912 +       err = 0;
32913 +       wkinfo = kmalloc(sizeof(*wkinfo), GFP_NOFS);
32914 +       if (wkinfo) {
32915 +               wkinfo->kobj = &au_sbi(sb)->si_kobj;
32916 +               wkinfo->flags = flags & ~AuWkq_WAIT;
32917 +               wkinfo->func = func;
32918 +               wkinfo->args = args;
32919 +               wkinfo->comp = NULL;
32920 +               kobject_get(wkinfo->kobj);
32921 +               __module_get(THIS_MODULE); /* todo: ?? */
32922 +
32923 +               au_wkq_run(wkinfo);
32924 +       } else {
32925 +               err = -ENOMEM;
32926 +               au_nwt_done(&au_sbi(sb)->si_nowait);
32927 +       }
32928 +
32929 +       return err;
32930 +}
32931 +
32932 +/* ---------------------------------------------------------------------- */
32933 +
32934 +void au_nwt_init(struct au_nowait_tasks *nwt)
32935 +{
32936 +       atomic_set(&nwt->nw_len, 0);
32937 +       /* smp_mb(); */ /* atomic_set */
32938 +       init_waitqueue_head(&nwt->nw_wq);
32939 +}
32940 +
32941 +void au_wkq_fin(void)
32942 +{
32943 +       destroy_workqueue(au_wkq);
32944 +}
32945 +
32946 +int __init au_wkq_init(void)
32947 +{
32948 +       int err;
32949 +
32950 +       err = 0;
32951 +       au_wkq = alloc_workqueue(AUFS_WKQ_NAME, 0, WQ_DFL_ACTIVE);
32952 +       if (IS_ERR(au_wkq))
32953 +               err = PTR_ERR(au_wkq);
32954 +       else if (!au_wkq)
32955 +               err = -ENOMEM;
32956 +
32957 +       return err;
32958 +}
32959 diff -urN /usr/share/empty/fs/aufs/wkq.h linux/fs/aufs/wkq.h
32960 --- /usr/share/empty/fs/aufs/wkq.h      1970-01-01 01:00:00.000000000 +0100
32961 +++ linux/fs/aufs/wkq.h 2015-01-25 13:00:38.634380408 +0100
32962 @@ -0,0 +1,91 @@
32963 +/*
32964 + * Copyright (C) 2005-2014 Junjiro R. Okajima
32965 + *
32966 + * This program, aufs is free software; you can redistribute it and/or modify
32967 + * it under the terms of the GNU General Public License as published by
32968 + * the Free Software Foundation; either version 2 of the License, or
32969 + * (at your option) any later version.
32970 + *
32971 + * This program is distributed in the hope that it will be useful,
32972 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32973 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32974 + * GNU General Public License for more details.
32975 + *
32976 + * You should have received a copy of the GNU General Public License
32977 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32978 + */
32979 +
32980 +/*
32981 + * workqueue for asynchronous/super-io operations
32982 + * todo: try new credentials management scheme
32983 + */
32984 +
32985 +#ifndef __AUFS_WKQ_H__
32986 +#define __AUFS_WKQ_H__
32987 +
32988 +#ifdef __KERNEL__
32989 +
32990 +struct super_block;
32991 +
32992 +/* ---------------------------------------------------------------------- */
32993 +
32994 +/*
32995 + * in the next operation, wait for the 'nowait' tasks in system-wide workqueue
32996 + */
32997 +struct au_nowait_tasks {
32998 +       atomic_t                nw_len;
32999 +       wait_queue_head_t       nw_wq;
33000 +};
33001 +
33002 +/* ---------------------------------------------------------------------- */
33003 +
33004 +typedef void (*au_wkq_func_t)(void *args);
33005 +
33006 +/* wkq flags */
33007 +#define AuWkq_WAIT     1
33008 +#define AuWkq_NEST     (1 << 1)
33009 +#define au_ftest_wkq(flags, name)      ((flags) & AuWkq_##name)
33010 +#define au_fset_wkq(flags, name) \
33011 +       do { (flags) |= AuWkq_##name; } while (0)
33012 +#define au_fclr_wkq(flags, name) \
33013 +       do { (flags) &= ~AuWkq_##name; } while (0)
33014 +
33015 +#ifndef CONFIG_AUFS_HNOTIFY
33016 +#undef AuWkq_NEST
33017 +#define AuWkq_NEST     0
33018 +#endif
33019 +
33020 +/* wkq.c */
33021 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args);
33022 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
33023 +                 unsigned int flags);
33024 +void au_nwt_init(struct au_nowait_tasks *nwt);
33025 +int __init au_wkq_init(void);
33026 +void au_wkq_fin(void);
33027 +
33028 +/* ---------------------------------------------------------------------- */
33029 +
33030 +static inline int au_wkq_test(void)
33031 +{
33032 +       return current->flags & PF_WQ_WORKER;
33033 +}
33034 +
33035 +static inline int au_wkq_wait(au_wkq_func_t func, void *args)
33036 +{
33037 +       return au_wkq_do_wait(AuWkq_WAIT, func, args);
33038 +}
33039 +
33040 +static inline void au_nwt_done(struct au_nowait_tasks *nwt)
33041 +{
33042 +       if (atomic_dec_and_test(&nwt->nw_len))
33043 +               wake_up_all(&nwt->nw_wq);
33044 +}
33045 +
33046 +static inline int au_nwt_flush(struct au_nowait_tasks *nwt)
33047 +{
33048 +       wait_event(nwt->nw_wq, !atomic_read(&nwt->nw_len));
33049 +       return 0;
33050 +}
33051 +
33052 +#endif /* __KERNEL__ */
33053 +#endif /* __AUFS_WKQ_H__ */
33054 diff -urN /usr/share/empty/fs/aufs/xattr.c linux/fs/aufs/xattr.c
33055 --- /usr/share/empty/fs/aufs/xattr.c    1970-01-01 01:00:00.000000000 +0100
33056 +++ linux/fs/aufs/xattr.c       2015-01-25 13:00:38.634380408 +0100
33057 @@ -0,0 +1,334 @@
33058 +/*
33059 + * Copyright (C) 2014 Junjiro R. Okajima
33060 + *
33061 + * This program, aufs is free software; you can redistribute it and/or modify
33062 + * it under the terms of the GNU General Public License as published by
33063 + * the Free Software Foundation; either version 2 of the License, or
33064 + * (at your option) any later version.
33065 + *
33066 + * This program is distributed in the hope that it will be useful,
33067 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33068 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33069 + * GNU General Public License for more details.
33070 + *
33071 + * You should have received a copy of the GNU General Public License
33072 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33073 + */
33074 +
33075 +/*
33076 + * handling xattr functions
33077 + */
33078 +
33079 +#include <linux/xattr.h>
33080 +#include "aufs.h"
33081 +
33082 +static int au_xattr_ignore(int err, char *name, unsigned int ignore_flags)
33083 +{
33084 +       if (!ignore_flags)
33085 +               goto out;
33086 +       switch (err) {
33087 +       case -ENOMEM:
33088 +       case -EDQUOT:
33089 +               goto out;
33090 +       }
33091 +
33092 +       if ((ignore_flags & AuBrAttr_ICEX) == AuBrAttr_ICEX) {
33093 +               err = 0;
33094 +               goto out;
33095 +       }
33096 +
33097 +#define cmp(brattr, prefix) do {                                       \
33098 +               if (!strncmp(name, XATTR_##prefix##_PREFIX,             \
33099 +                            XATTR_##prefix##_PREFIX_LEN)) {            \
33100 +                       if (ignore_flags & AuBrAttr_ICEX_##brattr)      \
33101 +                               err = 0;                                \
33102 +                       goto out;                                       \
33103 +               }                                                       \
33104 +       } while (0)
33105 +
33106 +       cmp(SEC, SECURITY);
33107 +       cmp(SYS, SYSTEM);
33108 +       cmp(TR, TRUSTED);
33109 +       cmp(USR, USER);
33110 +#undef cmp
33111 +
33112 +       if (ignore_flags & AuBrAttr_ICEX_OTH)
33113 +               err = 0;
33114 +
33115 +out:
33116 +       return err;
33117 +}
33118 +
33119 +static const int au_xattr_out_of_list = AuBrAttr_ICEX_OTH << 1;
33120 +
33121 +static int au_do_cpup_xattr(struct dentry *h_dst, struct dentry *h_src,
33122 +                           char *name, char **buf, unsigned int ignore_flags)
33123 +{
33124 +       int err;
33125 +       ssize_t ssz;
33126 +       struct inode *h_idst;
33127 +
33128 +       ssz = vfs_getxattr_alloc(h_src, name, buf, 0, GFP_NOFS);
33129 +       err = ssz;
33130 +       if (unlikely(err <= 0)) {
33131 +               AuTraceErr(err);
33132 +               if (err == -ENODATA
33133 +                   || (err == -EOPNOTSUPP
33134 +                       && (ignore_flags & au_xattr_out_of_list)))
33135 +                       err = 0;
33136 +               goto out;
33137 +       }
33138 +
33139 +       /* unlock it temporary */
33140 +       h_idst = h_dst->d_inode;
33141 +       mutex_unlock(&h_idst->i_mutex);
33142 +       err = vfsub_setxattr(h_dst, name, *buf, ssz, /*flags*/0);
33143 +       mutex_lock_nested(&h_idst->i_mutex, AuLsc_I_CHILD2);
33144 +       if (unlikely(err)) {
33145 +               AuDbg("%s, err %d\n", name, err);
33146 +               err = au_xattr_ignore(err, name, ignore_flags);
33147 +       }
33148 +
33149 +out:
33150 +       return err;
33151 +}
33152 +
33153 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags)
33154 +{
33155 +       int err, unlocked, acl_access, acl_default;
33156 +       ssize_t ssz;
33157 +       struct inode *h_isrc, *h_idst;
33158 +       char *value, *p, *o, *e;
33159 +
33160 +       /* try stopping to update the source inode while we are referencing */
33161 +       /* there should not be the parent-child relation ship between them */
33162 +       h_isrc = h_src->d_inode;
33163 +       h_idst = h_dst->d_inode;
33164 +       mutex_unlock(&h_idst->i_mutex);
33165 +       mutex_lock_nested(&h_isrc->i_mutex, AuLsc_I_CHILD);
33166 +       mutex_lock_nested(&h_idst->i_mutex, AuLsc_I_CHILD2);
33167 +       unlocked = 0;
33168 +
33169 +       /* some filesystems don't list POSIX ACL, for example tmpfs */
33170 +       ssz = vfs_listxattr(h_src, NULL, 0);
33171 +       err = ssz;
33172 +       if (unlikely(err < 0)) {
33173 +               AuTraceErr(err);
33174 +               if (err == -ENODATA
33175 +                   || err == -EOPNOTSUPP)
33176 +                       err = 0;        /* ignore */
33177 +               goto out;
33178 +       }
33179 +
33180 +       err = 0;
33181 +       p = NULL;
33182 +       o = NULL;
33183 +       if (ssz) {
33184 +               err = -ENOMEM;
33185 +               p = kmalloc(ssz, GFP_NOFS);
33186 +               o = p;
33187 +               if (unlikely(!p))
33188 +                       goto out;
33189 +               err = vfs_listxattr(h_src, p, ssz);
33190 +       }
33191 +       mutex_unlock(&h_isrc->i_mutex);
33192 +       unlocked = 1;
33193 +       AuDbg("err %d, ssz %zd\n", err, ssz);
33194 +       if (unlikely(err < 0))
33195 +               goto out_free;
33196 +
33197 +       err = 0;
33198 +       e = p + ssz;
33199 +       value = NULL;
33200 +       acl_access = 0;
33201 +       acl_default = 0;
33202 +       while (!err && p < e) {
33203 +               acl_access |= !strncmp(p, XATTR_NAME_POSIX_ACL_ACCESS,
33204 +                                      sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1);
33205 +               acl_default |= !strncmp(p, XATTR_NAME_POSIX_ACL_DEFAULT,
33206 +                                       sizeof(XATTR_NAME_POSIX_ACL_DEFAULT)
33207 +                                       - 1);
33208 +               err = au_do_cpup_xattr(h_dst, h_src, p, &value, ignore_flags);
33209 +               p += strlen(p) + 1;
33210 +       }
33211 +       AuTraceErr(err);
33212 +       ignore_flags |= au_xattr_out_of_list;
33213 +       if (!err && !acl_access) {
33214 +               err = au_do_cpup_xattr(h_dst, h_src,
33215 +                                      XATTR_NAME_POSIX_ACL_ACCESS, &value,
33216 +                                      ignore_flags);
33217 +               AuTraceErr(err);
33218 +       }
33219 +       if (!err && !acl_default) {
33220 +               err = au_do_cpup_xattr(h_dst, h_src,
33221 +                                      XATTR_NAME_POSIX_ACL_DEFAULT, &value,
33222 +                                      ignore_flags);
33223 +               AuTraceErr(err);
33224 +       }
33225 +
33226 +       kfree(value);
33227 +
33228 +out_free:
33229 +       kfree(o);
33230 +out:
33231 +       if (!unlocked)
33232 +               mutex_unlock(&h_isrc->i_mutex);
33233 +       AuTraceErr(err);
33234 +       return err;
33235 +}
33236 +
33237 +/* ---------------------------------------------------------------------- */
33238 +
33239 +enum {
33240 +       AU_XATTR_LIST,
33241 +       AU_XATTR_GET
33242 +};
33243 +
33244 +struct au_lgxattr {
33245 +       int type;
33246 +       union {
33247 +               struct {
33248 +                       char    *list;
33249 +                       size_t  size;
33250 +               } list;
33251 +               struct {
33252 +                       const char      *name;
33253 +                       void            *value;
33254 +                       size_t          size;
33255 +               } get;
33256 +       } u;
33257 +};
33258 +
33259 +static ssize_t au_lgxattr(struct dentry *dentry, struct au_lgxattr *arg)
33260 +{
33261 +       ssize_t err;
33262 +       struct path h_path;
33263 +       struct super_block *sb;
33264 +
33265 +       sb = dentry->d_sb;
33266 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
33267 +       if (unlikely(err))
33268 +               goto out;
33269 +       err = au_h_path_getattr(dentry, /*force*/1, &h_path);
33270 +       if (unlikely(err))
33271 +               goto out_si;
33272 +       if (unlikely(!h_path.dentry))
33273 +               /* illegally overlapped or something */
33274 +               goto out_di; /* pretending success */
33275 +
33276 +       /* always topmost entry only */
33277 +       switch (arg->type) {
33278 +       case AU_XATTR_LIST:
33279 +               err = vfs_listxattr(h_path.dentry,
33280 +                                   arg->u.list.list, arg->u.list.size);
33281 +               break;
33282 +       case AU_XATTR_GET:
33283 +               err = vfs_getxattr(h_path.dentry,
33284 +                                  arg->u.get.name, arg->u.get.value,
33285 +                                  arg->u.get.size);
33286 +               break;
33287 +       }
33288 +
33289 +out_di:
33290 +       di_read_unlock(dentry, AuLock_IR);
33291 +out_si:
33292 +       si_read_unlock(sb);
33293 +out:
33294 +       AuTraceErr(err);
33295 +       return err;
33296 +}
33297 +
33298 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size)
33299 +{
33300 +       struct au_lgxattr arg = {
33301 +               .type = AU_XATTR_LIST,
33302 +               .u.list = {
33303 +                       .list   = list,
33304 +                       .size   = size
33305 +               },
33306 +       };
33307 +
33308 +       return au_lgxattr(dentry, &arg);
33309 +}
33310 +
33311 +ssize_t aufs_getxattr(struct dentry *dentry, const char *name, void *value,
33312 +                     size_t size)
33313 +{
33314 +       struct au_lgxattr arg = {
33315 +               .type = AU_XATTR_GET,
33316 +               .u.get = {
33317 +                       .name   = name,
33318 +                       .value  = value,
33319 +                       .size   = size
33320 +               },
33321 +       };
33322 +
33323 +       return au_lgxattr(dentry, &arg);
33324 +}
33325 +
33326 +int aufs_setxattr(struct dentry *dentry, const char *name, const void *value,
33327 +                 size_t size, int flags)
33328 +{
33329 +       struct au_srxattr arg = {
33330 +               .type = AU_XATTR_SET,
33331 +               .u.set = {
33332 +                       .name   = name,
33333 +                       .value  = value,
33334 +                       .size   = size,
33335 +                       .flags  = flags
33336 +               },
33337 +       };
33338 +
33339 +       return au_srxattr(dentry, &arg);
33340 +}
33341 +
33342 +int aufs_removexattr(struct dentry *dentry, const char *name)
33343 +{
33344 +       struct au_srxattr arg = {
33345 +               .type = AU_XATTR_REMOVE,
33346 +               .u.remove = {
33347 +                       .name   = name
33348 +               },
33349 +       };
33350 +
33351 +       return au_srxattr(dentry, &arg);
33352 +}
33353 +
33354 +/* ---------------------------------------------------------------------- */
33355 +
33356 +#if 0
33357 +static size_t au_xattr_list(struct dentry *dentry, char *list, size_t list_size,
33358 +                           const char *name, size_t name_len, int type)
33359 +{
33360 +       return aufs_listxattr(dentry, list, list_size);
33361 +}
33362 +
33363 +static int au_xattr_get(struct dentry *dentry, const char *name, void *buffer,
33364 +                       size_t size, int type)
33365 +{
33366 +       return aufs_getxattr(dentry, name, buffer, size);
33367 +}
33368 +
33369 +static int au_xattr_set(struct dentry *dentry, const char *name,
33370 +                       const void *value, size_t size, int flags, int type)
33371 +{
33372 +       return aufs_setxattr(dentry, name, value, size, flags);
33373 +}
33374 +
33375 +static const struct xattr_handler au_xattr_handler = {
33376 +       /* no prefix, no flags */
33377 +       .list   = au_xattr_list,
33378 +       .get    = au_xattr_get,
33379 +       .set    = au_xattr_set
33380 +       /* why no remove? */
33381 +};
33382 +
33383 +static const struct xattr_handler *au_xattr_handlers[] = {
33384 +       &au_xattr_handler
33385 +};
33386 +
33387 +void au_xattr_init(struct super_block *sb)
33388 +{
33389 +       /* sb->s_xattr = au_xattr_handlers; */
33390 +}
33391 +#endif
33392 diff -urN /usr/share/empty/fs/aufs/xino.c linux/fs/aufs/xino.c
33393 --- /usr/share/empty/fs/aufs/xino.c     1970-01-01 01:00:00.000000000 +0100
33394 +++ linux/fs/aufs/xino.c        2015-01-25 13:00:38.634380408 +0100
33395 @@ -0,0 +1,1318 @@
33396 +/*
33397 + * Copyright (C) 2005-2014 Junjiro R. Okajima
33398 + *
33399 + * This program, aufs is free software; you can redistribute it and/or modify
33400 + * it under the terms of the GNU General Public License as published by
33401 + * the Free Software Foundation; either version 2 of the License, or
33402 + * (at your option) any later version.
33403 + *
33404 + * This program is distributed in the hope that it will be useful,
33405 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33406 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33407 + * GNU General Public License for more details.
33408 + *
33409 + * You should have received a copy of the GNU General Public License
33410 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33411 + */
33412 +
33413 +/*
33414 + * external inode number translation table and bitmap
33415 + */
33416 +
33417 +#include <linux/seq_file.h>
33418 +#include <linux/statfs.h>
33419 +#include "aufs.h"
33420 +
33421 +/* todo: unnecessary to support mmap_sem since kernel-space? */
33422 +ssize_t xino_fread(au_readf_t func, struct file *file, void *kbuf, size_t size,
33423 +                  loff_t *pos)
33424 +{
33425 +       ssize_t err;
33426 +       mm_segment_t oldfs;
33427 +       union {
33428 +               void *k;
33429 +               char __user *u;
33430 +       } buf;
33431 +
33432 +       buf.k = kbuf;
33433 +       oldfs = get_fs();
33434 +       set_fs(KERNEL_DS);
33435 +       do {
33436 +               /* todo: signal_pending? */
33437 +               err = func(file, buf.u, size, pos);
33438 +       } while (err == -EAGAIN || err == -EINTR);
33439 +       set_fs(oldfs);
33440 +
33441 +#if 0 /* reserved for future use */
33442 +       if (err > 0)
33443 +               fsnotify_access(file->f_dentry);
33444 +#endif
33445 +
33446 +       return err;
33447 +}
33448 +
33449 +/* ---------------------------------------------------------------------- */
33450 +
33451 +static ssize_t do_xino_fwrite(au_writef_t func, struct file *file, void *kbuf,
33452 +                             size_t size, loff_t *pos)
33453 +{
33454 +       ssize_t err;
33455 +       mm_segment_t oldfs;
33456 +       union {
33457 +               void *k;
33458 +               const char __user *u;
33459 +       } buf;
33460 +
33461 +       buf.k = kbuf;
33462 +       oldfs = get_fs();
33463 +       set_fs(KERNEL_DS);
33464 +       do {
33465 +               /* todo: signal_pending? */
33466 +               err = func(file, buf.u, size, pos);
33467 +       } while (err == -EAGAIN || err == -EINTR);
33468 +       set_fs(oldfs);
33469 +
33470 +#if 0 /* reserved for future use */
33471 +       if (err > 0)
33472 +               fsnotify_modify(file->f_dentry);
33473 +#endif
33474 +
33475 +       return err;
33476 +}
33477 +
33478 +struct do_xino_fwrite_args {
33479 +       ssize_t *errp;
33480 +       au_writef_t func;
33481 +       struct file *file;
33482 +       void *buf;
33483 +       size_t size;
33484 +       loff_t *pos;
33485 +};
33486 +
33487 +static void call_do_xino_fwrite(void *args)
33488 +{
33489 +       struct do_xino_fwrite_args *a = args;
33490 +       *a->errp = do_xino_fwrite(a->func, a->file, a->buf, a->size, a->pos);
33491 +}
33492 +
33493 +ssize_t xino_fwrite(au_writef_t func, struct file *file, void *buf, size_t size,
33494 +                   loff_t *pos)
33495 +{
33496 +       ssize_t err;
33497 +
33498 +       /* todo: signal block and no wkq? */
33499 +       if (rlimit(RLIMIT_FSIZE) == RLIM_INFINITY) {
33500 +               lockdep_off();
33501 +               err = do_xino_fwrite(func, file, buf, size, pos);
33502 +               lockdep_on();
33503 +       } else {
33504 +               /*
33505 +                * it breaks RLIMIT_FSIZE and normal user's limit,
33506 +                * users should care about quota and real 'filesystem full.'
33507 +                */
33508 +               int wkq_err;
33509 +               struct do_xino_fwrite_args args = {
33510 +                       .errp   = &err,
33511 +                       .func   = func,
33512 +                       .file   = file,
33513 +                       .buf    = buf,
33514 +                       .size   = size,
33515 +                       .pos    = pos
33516 +               };
33517 +
33518 +               wkq_err = au_wkq_wait(call_do_xino_fwrite, &args);
33519 +               if (unlikely(wkq_err))
33520 +                       err = wkq_err;
33521 +       }
33522 +
33523 +       return err;
33524 +}
33525 +
33526 +/* ---------------------------------------------------------------------- */
33527 +
33528 +/*
33529 + * create a new xinofile at the same place/path as @base_file.
33530 + */
33531 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src)
33532 +{
33533 +       struct file *file;
33534 +       struct dentry *base, *parent;
33535 +       struct inode *dir, *delegated;
33536 +       struct qstr *name;
33537 +       struct path path;
33538 +       int err;
33539 +
33540 +       base = base_file->f_dentry;
33541 +       parent = base->d_parent; /* dir inode is locked */
33542 +       dir = parent->d_inode;
33543 +       IMustLock(dir);
33544 +
33545 +       file = ERR_PTR(-EINVAL);
33546 +       name = &base->d_name;
33547 +       path.dentry = vfsub_lookup_one_len(name->name, parent, name->len);
33548 +       if (IS_ERR(path.dentry)) {
33549 +               file = (void *)path.dentry;
33550 +               pr_err("%pd lookup err %ld\n",
33551 +                      base, PTR_ERR(path.dentry));
33552 +               goto out;
33553 +       }
33554 +
33555 +       /* no need to mnt_want_write() since we call dentry_open() later */
33556 +       err = vfs_create(dir, path.dentry, S_IRUGO | S_IWUGO, NULL);
33557 +       if (unlikely(err)) {
33558 +               file = ERR_PTR(err);
33559 +               pr_err("%pd create err %d\n", base, err);
33560 +               goto out_dput;
33561 +       }
33562 +
33563 +       path.mnt = base_file->f_path.mnt;
33564 +       file = vfsub_dentry_open(&path,
33565 +                                O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
33566 +                                /* | __FMODE_NONOTIFY */);
33567 +       if (IS_ERR(file)) {
33568 +               pr_err("%pd open err %ld\n", base, PTR_ERR(file));
33569 +               goto out_dput;
33570 +       }
33571 +
33572 +       delegated = NULL;
33573 +       err = vfsub_unlink(dir, &file->f_path, &delegated, /*force*/0);
33574 +       if (unlikely(err == -EWOULDBLOCK)) {
33575 +               pr_warn("cannot retry for NFSv4 delegation"
33576 +                       " for an internal unlink\n");
33577 +               iput(delegated);
33578 +       }
33579 +       if (unlikely(err)) {
33580 +               pr_err("%pd unlink err %d\n", base, err);
33581 +               goto out_fput;
33582 +       }
33583 +
33584 +       if (copy_src) {
33585 +               /* no one can touch copy_src xino */
33586 +               err = au_copy_file(file, copy_src, vfsub_f_size_read(copy_src));
33587 +               if (unlikely(err)) {
33588 +                       pr_err("%pd copy err %d\n", base, err);
33589 +                       goto out_fput;
33590 +               }
33591 +       }
33592 +       goto out_dput; /* success */
33593 +
33594 +out_fput:
33595 +       fput(file);
33596 +       file = ERR_PTR(err);
33597 +out_dput:
33598 +       dput(path.dentry);
33599 +out:
33600 +       return file;
33601 +}
33602 +
33603 +struct au_xino_lock_dir {
33604 +       struct au_hinode *hdir;
33605 +       struct dentry *parent;
33606 +       struct mutex *mtx;
33607 +};
33608 +
33609 +static void au_xino_lock_dir(struct super_block *sb, struct file *xino,
33610 +                            struct au_xino_lock_dir *ldir)
33611 +{
33612 +       aufs_bindex_t brid, bindex;
33613 +
33614 +       ldir->hdir = NULL;
33615 +       bindex = -1;
33616 +       brid = au_xino_brid(sb);
33617 +       if (brid >= 0)
33618 +               bindex = au_br_index(sb, brid);
33619 +       if (bindex >= 0) {
33620 +               ldir->hdir = au_hi(sb->s_root->d_inode, bindex);
33621 +               au_hn_imtx_lock_nested(ldir->hdir, AuLsc_I_PARENT);
33622 +       } else {
33623 +               ldir->parent = dget_parent(xino->f_dentry);
33624 +               ldir->mtx = &ldir->parent->d_inode->i_mutex;
33625 +               mutex_lock_nested(ldir->mtx, AuLsc_I_PARENT);
33626 +       }
33627 +}
33628 +
33629 +static void au_xino_unlock_dir(struct au_xino_lock_dir *ldir)
33630 +{
33631 +       if (ldir->hdir)
33632 +               au_hn_imtx_unlock(ldir->hdir);
33633 +       else {
33634 +               mutex_unlock(ldir->mtx);
33635 +               dput(ldir->parent);
33636 +       }
33637 +}
33638 +
33639 +/* ---------------------------------------------------------------------- */
33640 +
33641 +/* trucate xino files asynchronously */
33642 +
33643 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex)
33644 +{
33645 +       int err;
33646 +       unsigned long jiffy;
33647 +       blkcnt_t blocks;
33648 +       aufs_bindex_t bi, bend;
33649 +       struct kstatfs *st;
33650 +       struct au_branch *br;
33651 +       struct file *new_xino, *file;
33652 +       struct super_block *h_sb;
33653 +       struct au_xino_lock_dir ldir;
33654 +
33655 +       err = -ENOMEM;
33656 +       st = kzalloc(sizeof(*st), GFP_NOFS);
33657 +       if (unlikely(!st))
33658 +               goto out;
33659 +
33660 +       err = -EINVAL;
33661 +       bend = au_sbend(sb);
33662 +       if (unlikely(bindex < 0 || bend < bindex))
33663 +               goto out_st;
33664 +       br = au_sbr(sb, bindex);
33665 +       file = br->br_xino.xi_file;
33666 +       if (!file)
33667 +               goto out_st;
33668 +
33669 +       err = vfs_statfs(&file->f_path, st);
33670 +       if (unlikely(err))
33671 +               AuErr1("statfs err %d, ignored\n", err);
33672 +       jiffy = jiffies;
33673 +       blocks = file_inode(file)->i_blocks;
33674 +       pr_info("begin truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
33675 +               bindex, (u64)blocks, st->f_bfree, st->f_blocks);
33676 +
33677 +       au_xino_lock_dir(sb, file, &ldir);
33678 +       /* mnt_want_write() is unnecessary here */
33679 +       new_xino = au_xino_create2(file, file);
33680 +       au_xino_unlock_dir(&ldir);
33681 +       err = PTR_ERR(new_xino);
33682 +       if (IS_ERR(new_xino)) {
33683 +               pr_err("err %d, ignored\n", err);
33684 +               goto out_st;
33685 +       }
33686 +       err = 0;
33687 +       fput(file);
33688 +       br->br_xino.xi_file = new_xino;
33689 +
33690 +       h_sb = au_br_sb(br);
33691 +       for (bi = 0; bi <= bend; bi++) {
33692 +               if (unlikely(bi == bindex))
33693 +                       continue;
33694 +               br = au_sbr(sb, bi);
33695 +               if (au_br_sb(br) != h_sb)
33696 +                       continue;
33697 +
33698 +               fput(br->br_xino.xi_file);
33699 +               br->br_xino.xi_file = new_xino;
33700 +               get_file(new_xino);
33701 +       }
33702 +
33703 +       err = vfs_statfs(&new_xino->f_path, st);
33704 +       if (!err) {
33705 +               pr_info("end truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
33706 +                       bindex, (u64)file_inode(new_xino)->i_blocks,
33707 +                       st->f_bfree, st->f_blocks);
33708 +               if (file_inode(new_xino)->i_blocks < blocks)
33709 +                       au_sbi(sb)->si_xino_jiffy = jiffy;
33710 +       } else
33711 +               AuErr1("statfs err %d, ignored\n", err);
33712 +
33713 +out_st:
33714 +       kfree(st);
33715 +out:
33716 +       return err;
33717 +}
33718 +
33719 +struct xino_do_trunc_args {
33720 +       struct super_block *sb;
33721 +       struct au_branch *br;
33722 +};
33723 +
33724 +static void xino_do_trunc(void *_args)
33725 +{
33726 +       struct xino_do_trunc_args *args = _args;
33727 +       struct super_block *sb;
33728 +       struct au_branch *br;
33729 +       struct inode *dir;
33730 +       int err;
33731 +       aufs_bindex_t bindex;
33732 +
33733 +       err = 0;
33734 +       sb = args->sb;
33735 +       dir = sb->s_root->d_inode;
33736 +       br = args->br;
33737 +
33738 +       si_noflush_write_lock(sb);
33739 +       ii_read_lock_parent(dir);
33740 +       bindex = au_br_index(sb, br->br_id);
33741 +       err = au_xino_trunc(sb, bindex);
33742 +       ii_read_unlock(dir);
33743 +       if (unlikely(err))
33744 +               pr_warn("err b%d, (%d)\n", bindex, err);
33745 +       atomic_dec(&br->br_xino_running);
33746 +       atomic_dec(&br->br_count);
33747 +       si_write_unlock(sb);
33748 +       au_nwt_done(&au_sbi(sb)->si_nowait);
33749 +       kfree(args);
33750 +}
33751 +
33752 +static int xino_trunc_test(struct super_block *sb, struct au_branch *br)
33753 +{
33754 +       int err;
33755 +       struct kstatfs st;
33756 +       struct au_sbinfo *sbinfo;
33757 +
33758 +       /* todo: si_xino_expire and the ratio should be customizable */
33759 +       sbinfo = au_sbi(sb);
33760 +       if (time_before(jiffies,
33761 +                       sbinfo->si_xino_jiffy + sbinfo->si_xino_expire))
33762 +               return 0;
33763 +
33764 +       /* truncation border */
33765 +       err = vfs_statfs(&br->br_xino.xi_file->f_path, &st);
33766 +       if (unlikely(err)) {
33767 +               AuErr1("statfs err %d, ignored\n", err);
33768 +               return 0;
33769 +       }
33770 +       if (div64_u64(st.f_bfree * 100, st.f_blocks) >= AUFS_XINO_DEF_TRUNC)
33771 +               return 0;
33772 +
33773 +       return 1;
33774 +}
33775 +
33776 +static void xino_try_trunc(struct super_block *sb, struct au_branch *br)
33777 +{
33778 +       struct xino_do_trunc_args *args;
33779 +       int wkq_err;
33780 +
33781 +       if (!xino_trunc_test(sb, br))
33782 +               return;
33783 +
33784 +       if (atomic_inc_return(&br->br_xino_running) > 1)
33785 +               goto out;
33786 +
33787 +       /* lock and kfree() will be called in trunc_xino() */
33788 +       args = kmalloc(sizeof(*args), GFP_NOFS);
33789 +       if (unlikely(!args)) {
33790 +               AuErr1("no memory\n");
33791 +               goto out_args;
33792 +       }
33793 +
33794 +       atomic_inc(&br->br_count);
33795 +       args->sb = sb;
33796 +       args->br = br;
33797 +       wkq_err = au_wkq_nowait(xino_do_trunc, args, sb, /*flags*/0);
33798 +       if (!wkq_err)
33799 +               return; /* success */
33800 +
33801 +       pr_err("wkq %d\n", wkq_err);
33802 +       atomic_dec(&br->br_count);
33803 +
33804 +out_args:
33805 +       kfree(args);
33806 +out:
33807 +       atomic_dec(&br->br_xino_running);
33808 +}
33809 +
33810 +/* ---------------------------------------------------------------------- */
33811 +
33812 +static int au_xino_do_write(au_writef_t write, struct file *file,
33813 +                           ino_t h_ino, ino_t ino)
33814 +{
33815 +       loff_t pos;
33816 +       ssize_t sz;
33817 +
33818 +       pos = h_ino;
33819 +       if (unlikely(au_loff_max / sizeof(ino) - 1 < pos)) {
33820 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
33821 +               return -EFBIG;
33822 +       }
33823 +       pos *= sizeof(ino);
33824 +       sz = xino_fwrite(write, file, &ino, sizeof(ino), &pos);
33825 +       if (sz == sizeof(ino))
33826 +               return 0; /* success */
33827 +
33828 +       AuIOErr("write failed (%zd)\n", sz);
33829 +       return -EIO;
33830 +}
33831 +
33832 +/*
33833 + * write @ino to the xinofile for the specified branch{@sb, @bindex}
33834 + * at the position of @h_ino.
33835 + * even if @ino is zero, it is written to the xinofile and means no entry.
33836 + * if the size of the xino file on a specific filesystem exceeds the watermark,
33837 + * try truncating it.
33838 + */
33839 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
33840 +                 ino_t ino)
33841 +{
33842 +       int err;
33843 +       unsigned int mnt_flags;
33844 +       struct au_branch *br;
33845 +
33846 +       BUILD_BUG_ON(sizeof(long long) != sizeof(au_loff_max)
33847 +                    || ((loff_t)-1) > 0);
33848 +       SiMustAnyLock(sb);
33849 +
33850 +       mnt_flags = au_mntflags(sb);
33851 +       if (!au_opt_test(mnt_flags, XINO))
33852 +               return 0;
33853 +
33854 +       br = au_sbr(sb, bindex);
33855 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
33856 +                              h_ino, ino);
33857 +       if (!err) {
33858 +               if (au_opt_test(mnt_flags, TRUNC_XINO)
33859 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
33860 +                       xino_try_trunc(sb, br);
33861 +               return 0; /* success */
33862 +       }
33863 +
33864 +       AuIOErr("write failed (%d)\n", err);
33865 +       return -EIO;
33866 +}
33867 +
33868 +/* ---------------------------------------------------------------------- */
33869 +
33870 +/* aufs inode number bitmap */
33871 +
33872 +static const int page_bits = (int)PAGE_SIZE * BITS_PER_BYTE;
33873 +static ino_t xib_calc_ino(unsigned long pindex, int bit)
33874 +{
33875 +       ino_t ino;
33876 +
33877 +       AuDebugOn(bit < 0 || page_bits <= bit);
33878 +       ino = AUFS_FIRST_INO + pindex * page_bits + bit;
33879 +       return ino;
33880 +}
33881 +
33882 +static void xib_calc_bit(ino_t ino, unsigned long *pindex, int *bit)
33883 +{
33884 +       AuDebugOn(ino < AUFS_FIRST_INO);
33885 +       ino -= AUFS_FIRST_INO;
33886 +       *pindex = ino / page_bits;
33887 +       *bit = ino % page_bits;
33888 +}
33889 +
33890 +static int xib_pindex(struct super_block *sb, unsigned long pindex)
33891 +{
33892 +       int err;
33893 +       loff_t pos;
33894 +       ssize_t sz;
33895 +       struct au_sbinfo *sbinfo;
33896 +       struct file *xib;
33897 +       unsigned long *p;
33898 +
33899 +       sbinfo = au_sbi(sb);
33900 +       MtxMustLock(&sbinfo->si_xib_mtx);
33901 +       AuDebugOn(pindex > ULONG_MAX / PAGE_SIZE
33902 +                 || !au_opt_test(sbinfo->si_mntflags, XINO));
33903 +
33904 +       if (pindex == sbinfo->si_xib_last_pindex)
33905 +               return 0;
33906 +
33907 +       xib = sbinfo->si_xib;
33908 +       p = sbinfo->si_xib_buf;
33909 +       pos = sbinfo->si_xib_last_pindex;
33910 +       pos *= PAGE_SIZE;
33911 +       sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
33912 +       if (unlikely(sz != PAGE_SIZE))
33913 +               goto out;
33914 +
33915 +       pos = pindex;
33916 +       pos *= PAGE_SIZE;
33917 +       if (vfsub_f_size_read(xib) >= pos + PAGE_SIZE)
33918 +               sz = xino_fread(sbinfo->si_xread, xib, p, PAGE_SIZE, &pos);
33919 +       else {
33920 +               memset(p, 0, PAGE_SIZE);
33921 +               sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
33922 +       }
33923 +       if (sz == PAGE_SIZE) {
33924 +               sbinfo->si_xib_last_pindex = pindex;
33925 +               return 0; /* success */
33926 +       }
33927 +
33928 +out:
33929 +       AuIOErr1("write failed (%zd)\n", sz);
33930 +       err = sz;
33931 +       if (sz >= 0)
33932 +               err = -EIO;
33933 +       return err;
33934 +}
33935 +
33936 +/* ---------------------------------------------------------------------- */
33937 +
33938 +static void au_xib_clear_bit(struct inode *inode)
33939 +{
33940 +       int err, bit;
33941 +       unsigned long pindex;
33942 +       struct super_block *sb;
33943 +       struct au_sbinfo *sbinfo;
33944 +
33945 +       AuDebugOn(inode->i_nlink);
33946 +
33947 +       sb = inode->i_sb;
33948 +       xib_calc_bit(inode->i_ino, &pindex, &bit);
33949 +       AuDebugOn(page_bits <= bit);
33950 +       sbinfo = au_sbi(sb);
33951 +       mutex_lock(&sbinfo->si_xib_mtx);
33952 +       err = xib_pindex(sb, pindex);
33953 +       if (!err) {
33954 +               clear_bit(bit, sbinfo->si_xib_buf);
33955 +               sbinfo->si_xib_next_bit = bit;
33956 +       }
33957 +       mutex_unlock(&sbinfo->si_xib_mtx);
33958 +}
33959 +
33960 +/* for s_op->delete_inode() */
33961 +void au_xino_delete_inode(struct inode *inode, const int unlinked)
33962 +{
33963 +       int err;
33964 +       unsigned int mnt_flags;
33965 +       aufs_bindex_t bindex, bend, bi;
33966 +       unsigned char try_trunc;
33967 +       struct au_iinfo *iinfo;
33968 +       struct super_block *sb;
33969 +       struct au_hinode *hi;
33970 +       struct inode *h_inode;
33971 +       struct au_branch *br;
33972 +       au_writef_t xwrite;
33973 +
33974 +       sb = inode->i_sb;
33975 +       mnt_flags = au_mntflags(sb);
33976 +       if (!au_opt_test(mnt_flags, XINO)
33977 +           || inode->i_ino == AUFS_ROOT_INO)
33978 +               return;
33979 +
33980 +       if (unlinked) {
33981 +               au_xigen_inc(inode);
33982 +               au_xib_clear_bit(inode);
33983 +       }
33984 +
33985 +       iinfo = au_ii(inode);
33986 +       if (!iinfo)
33987 +               return;
33988 +
33989 +       bindex = iinfo->ii_bstart;
33990 +       if (bindex < 0)
33991 +               return;
33992 +
33993 +       xwrite = au_sbi(sb)->si_xwrite;
33994 +       try_trunc = !!au_opt_test(mnt_flags, TRUNC_XINO);
33995 +       hi = iinfo->ii_hinode + bindex;
33996 +       bend = iinfo->ii_bend;
33997 +       for (; bindex <= bend; bindex++, hi++) {
33998 +               h_inode = hi->hi_inode;
33999 +               if (!h_inode
34000 +                   || (!unlinked && h_inode->i_nlink))
34001 +                       continue;
34002 +
34003 +               /* inode may not be revalidated */
34004 +               bi = au_br_index(sb, hi->hi_id);
34005 +               if (bi < 0)
34006 +                       continue;
34007 +
34008 +               br = au_sbr(sb, bi);
34009 +               err = au_xino_do_write(xwrite, br->br_xino.xi_file,
34010 +                                      h_inode->i_ino, /*ino*/0);
34011 +               if (!err && try_trunc
34012 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
34013 +                       xino_try_trunc(sb, br);
34014 +       }
34015 +}
34016 +
34017 +/* get an unused inode number from bitmap */
34018 +ino_t au_xino_new_ino(struct super_block *sb)
34019 +{
34020 +       ino_t ino;
34021 +       unsigned long *p, pindex, ul, pend;
34022 +       struct au_sbinfo *sbinfo;
34023 +       struct file *file;
34024 +       int free_bit, err;
34025 +
34026 +       if (!au_opt_test(au_mntflags(sb), XINO))
34027 +               return iunique(sb, AUFS_FIRST_INO);
34028 +
34029 +       sbinfo = au_sbi(sb);
34030 +       mutex_lock(&sbinfo->si_xib_mtx);
34031 +       p = sbinfo->si_xib_buf;
34032 +       free_bit = sbinfo->si_xib_next_bit;
34033 +       if (free_bit < page_bits && !test_bit(free_bit, p))
34034 +               goto out; /* success */
34035 +       free_bit = find_first_zero_bit(p, page_bits);
34036 +       if (free_bit < page_bits)
34037 +               goto out; /* success */
34038 +
34039 +       pindex = sbinfo->si_xib_last_pindex;
34040 +       for (ul = pindex - 1; ul < ULONG_MAX; ul--) {
34041 +               err = xib_pindex(sb, ul);
34042 +               if (unlikely(err))
34043 +                       goto out_err;
34044 +               free_bit = find_first_zero_bit(p, page_bits);
34045 +               if (free_bit < page_bits)
34046 +                       goto out; /* success */
34047 +       }
34048 +
34049 +       file = sbinfo->si_xib;
34050 +       pend = vfsub_f_size_read(file) / PAGE_SIZE;
34051 +       for (ul = pindex + 1; ul <= pend; ul++) {
34052 +               err = xib_pindex(sb, ul);
34053 +               if (unlikely(err))
34054 +                       goto out_err;
34055 +               free_bit = find_first_zero_bit(p, page_bits);
34056 +               if (free_bit < page_bits)
34057 +                       goto out; /* success */
34058 +       }
34059 +       BUG();
34060 +
34061 +out:
34062 +       set_bit(free_bit, p);
34063 +       sbinfo->si_xib_next_bit = free_bit + 1;
34064 +       pindex = sbinfo->si_xib_last_pindex;
34065 +       mutex_unlock(&sbinfo->si_xib_mtx);
34066 +       ino = xib_calc_ino(pindex, free_bit);
34067 +       AuDbg("i%lu\n", (unsigned long)ino);
34068 +       return ino;
34069 +out_err:
34070 +       mutex_unlock(&sbinfo->si_xib_mtx);
34071 +       AuDbg("i0\n");
34072 +       return 0;
34073 +}
34074 +
34075 +/*
34076 + * read @ino from xinofile for the specified branch{@sb, @bindex}
34077 + * at the position of @h_ino.
34078 + * if @ino does not exist and @do_new is true, get new one.
34079 + */
34080 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
34081 +                ino_t *ino)
34082 +{
34083 +       int err;
34084 +       ssize_t sz;
34085 +       loff_t pos;
34086 +       struct file *file;
34087 +       struct au_sbinfo *sbinfo;
34088 +
34089 +       *ino = 0;
34090 +       if (!au_opt_test(au_mntflags(sb), XINO))
34091 +               return 0; /* no xino */
34092 +
34093 +       err = 0;
34094 +       sbinfo = au_sbi(sb);
34095 +       pos = h_ino;
34096 +       if (unlikely(au_loff_max / sizeof(*ino) - 1 < pos)) {
34097 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
34098 +               return -EFBIG;
34099 +       }
34100 +       pos *= sizeof(*ino);
34101 +
34102 +       file = au_sbr(sb, bindex)->br_xino.xi_file;
34103 +       if (vfsub_f_size_read(file) < pos + sizeof(*ino))
34104 +               return 0; /* no ino */
34105 +
34106 +       sz = xino_fread(sbinfo->si_xread, file, ino, sizeof(*ino), &pos);
34107 +       if (sz == sizeof(*ino))
34108 +               return 0; /* success */
34109 +
34110 +       err = sz;
34111 +       if (unlikely(sz >= 0)) {
34112 +               err = -EIO;
34113 +               AuIOErr("xino read error (%zd)\n", sz);
34114 +       }
34115 +
34116 +       return err;
34117 +}
34118 +
34119 +/* ---------------------------------------------------------------------- */
34120 +
34121 +/* create and set a new xino file */
34122 +
34123 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent)
34124 +{
34125 +       struct file *file;
34126 +       struct dentry *h_parent, *d;
34127 +       struct inode *h_dir;
34128 +       int err;
34129 +
34130 +       /*
34131 +        * at mount-time, and the xino file is the default path,
34132 +        * hnotify is disabled so we have no notify events to ignore.
34133 +        * when a user specified the xino, we cannot get au_hdir to be ignored.
34134 +        */
34135 +       file = vfsub_filp_open(fname, O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
34136 +                              /* | __FMODE_NONOTIFY */,
34137 +                              S_IRUGO | S_IWUGO);
34138 +       if (IS_ERR(file)) {
34139 +               if (!silent)
34140 +                       pr_err("open %s(%ld)\n", fname, PTR_ERR(file));
34141 +               return file;
34142 +       }
34143 +
34144 +       /* keep file count */
34145 +       h_parent = dget_parent(file->f_dentry);
34146 +       h_dir = h_parent->d_inode;
34147 +       mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
34148 +       /* mnt_want_write() is unnecessary here */
34149 +       /* no delegation since it is just created */
34150 +       err = vfsub_unlink(h_dir, &file->f_path, /*delegated*/NULL, /*force*/0);
34151 +       mutex_unlock(&h_dir->i_mutex);
34152 +       dput(h_parent);
34153 +       if (unlikely(err)) {
34154 +               if (!silent)
34155 +                       pr_err("unlink %s(%d)\n", fname, err);
34156 +               goto out;
34157 +       }
34158 +
34159 +       err = -EINVAL;
34160 +       d = file->f_dentry;
34161 +       if (unlikely(sb == d->d_sb)) {
34162 +               if (!silent)
34163 +                       pr_err("%s must be outside\n", fname);
34164 +               goto out;
34165 +       }
34166 +       if (unlikely(au_test_fs_bad_xino(d->d_sb))) {
34167 +               if (!silent)
34168 +                       pr_err("xino doesn't support %s(%s)\n",
34169 +                              fname, au_sbtype(d->d_sb));
34170 +               goto out;
34171 +       }
34172 +       return file; /* success */
34173 +
34174 +out:
34175 +       fput(file);
34176 +       file = ERR_PTR(err);
34177 +       return file;
34178 +}
34179 +
34180 +/*
34181 + * find another branch who is on the same filesystem of the specified
34182 + * branch{@btgt}. search until @bend.
34183 + */
34184 +static int is_sb_shared(struct super_block *sb, aufs_bindex_t btgt,
34185 +                       aufs_bindex_t bend)
34186 +{
34187 +       aufs_bindex_t bindex;
34188 +       struct super_block *tgt_sb = au_sbr_sb(sb, btgt);
34189 +
34190 +       for (bindex = 0; bindex < btgt; bindex++)
34191 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
34192 +                       return bindex;
34193 +       for (bindex++; bindex <= bend; bindex++)
34194 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
34195 +                       return bindex;
34196 +       return -1;
34197 +}
34198 +
34199 +/* ---------------------------------------------------------------------- */
34200 +
34201 +/*
34202 + * initialize the xinofile for the specified branch @br
34203 + * at the place/path where @base_file indicates.
34204 + * test whether another branch is on the same filesystem or not,
34205 + * if @do_test is true.
34206 + */
34207 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t h_ino,
34208 +              struct file *base_file, int do_test)
34209 +{
34210 +       int err;
34211 +       ino_t ino;
34212 +       aufs_bindex_t bend, bindex;
34213 +       struct au_branch *shared_br, *b;
34214 +       struct file *file;
34215 +       struct super_block *tgt_sb;
34216 +
34217 +       shared_br = NULL;
34218 +       bend = au_sbend(sb);
34219 +       if (do_test) {
34220 +               tgt_sb = au_br_sb(br);
34221 +               for (bindex = 0; bindex <= bend; bindex++) {
34222 +                       b = au_sbr(sb, bindex);
34223 +                       if (tgt_sb == au_br_sb(b)) {
34224 +                               shared_br = b;
34225 +                               break;
34226 +                       }
34227 +               }
34228 +       }
34229 +
34230 +       if (!shared_br || !shared_br->br_xino.xi_file) {
34231 +               struct au_xino_lock_dir ldir;
34232 +
34233 +               au_xino_lock_dir(sb, base_file, &ldir);
34234 +               /* mnt_want_write() is unnecessary here */
34235 +               file = au_xino_create2(base_file, NULL);
34236 +               au_xino_unlock_dir(&ldir);
34237 +               err = PTR_ERR(file);
34238 +               if (IS_ERR(file))
34239 +                       goto out;
34240 +               br->br_xino.xi_file = file;
34241 +       } else {
34242 +               br->br_xino.xi_file = shared_br->br_xino.xi_file;
34243 +               get_file(br->br_xino.xi_file);
34244 +       }
34245 +
34246 +       ino = AUFS_ROOT_INO;
34247 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
34248 +                              h_ino, ino);
34249 +       if (unlikely(err)) {
34250 +               fput(br->br_xino.xi_file);
34251 +               br->br_xino.xi_file = NULL;
34252 +       }
34253 +
34254 +out:
34255 +       return err;
34256 +}
34257 +
34258 +/* ---------------------------------------------------------------------- */
34259 +
34260 +/* trucate a xino bitmap file */
34261 +
34262 +/* todo: slow */
34263 +static int do_xib_restore(struct super_block *sb, struct file *file, void *page)
34264 +{
34265 +       int err, bit;
34266 +       ssize_t sz;
34267 +       unsigned long pindex;
34268 +       loff_t pos, pend;
34269 +       struct au_sbinfo *sbinfo;
34270 +       au_readf_t func;
34271 +       ino_t *ino;
34272 +       unsigned long *p;
34273 +
34274 +       err = 0;
34275 +       sbinfo = au_sbi(sb);
34276 +       MtxMustLock(&sbinfo->si_xib_mtx);
34277 +       p = sbinfo->si_xib_buf;
34278 +       func = sbinfo->si_xread;
34279 +       pend = vfsub_f_size_read(file);
34280 +       pos = 0;
34281 +       while (pos < pend) {
34282 +               sz = xino_fread(func, file, page, PAGE_SIZE, &pos);
34283 +               err = sz;
34284 +               if (unlikely(sz <= 0))
34285 +                       goto out;
34286 +
34287 +               err = 0;
34288 +               for (ino = page; sz > 0; ino++, sz -= sizeof(ino)) {
34289 +                       if (unlikely(*ino < AUFS_FIRST_INO))
34290 +                               continue;
34291 +
34292 +                       xib_calc_bit(*ino, &pindex, &bit);
34293 +                       AuDebugOn(page_bits <= bit);
34294 +                       err = xib_pindex(sb, pindex);
34295 +                       if (!err)
34296 +                               set_bit(bit, p);
34297 +                       else
34298 +                               goto out;
34299 +               }
34300 +       }
34301 +
34302 +out:
34303 +       return err;
34304 +}
34305 +
34306 +static int xib_restore(struct super_block *sb)
34307 +{
34308 +       int err;
34309 +       aufs_bindex_t bindex, bend;
34310 +       void *page;
34311 +
34312 +       err = -ENOMEM;
34313 +       page = (void *)__get_free_page(GFP_NOFS);
34314 +       if (unlikely(!page))
34315 +               goto out;
34316 +
34317 +       err = 0;
34318 +       bend = au_sbend(sb);
34319 +       for (bindex = 0; !err && bindex <= bend; bindex++)
34320 +               if (!bindex || is_sb_shared(sb, bindex, bindex - 1) < 0)
34321 +                       err = do_xib_restore
34322 +                               (sb, au_sbr(sb, bindex)->br_xino.xi_file, page);
34323 +               else
34324 +                       AuDbg("b%d\n", bindex);
34325 +       free_page((unsigned long)page);
34326 +
34327 +out:
34328 +       return err;
34329 +}
34330 +
34331 +int au_xib_trunc(struct super_block *sb)
34332 +{
34333 +       int err;
34334 +       ssize_t sz;
34335 +       loff_t pos;
34336 +       struct au_xino_lock_dir ldir;
34337 +       struct au_sbinfo *sbinfo;
34338 +       unsigned long *p;
34339 +       struct file *file;
34340 +
34341 +       SiMustWriteLock(sb);
34342 +
34343 +       err = 0;
34344 +       sbinfo = au_sbi(sb);
34345 +       if (!au_opt_test(sbinfo->si_mntflags, XINO))
34346 +               goto out;
34347 +
34348 +       file = sbinfo->si_xib;
34349 +       if (vfsub_f_size_read(file) <= PAGE_SIZE)
34350 +               goto out;
34351 +
34352 +       au_xino_lock_dir(sb, file, &ldir);
34353 +       /* mnt_want_write() is unnecessary here */
34354 +       file = au_xino_create2(sbinfo->si_xib, NULL);
34355 +       au_xino_unlock_dir(&ldir);
34356 +       err = PTR_ERR(file);
34357 +       if (IS_ERR(file))
34358 +               goto out;
34359 +       fput(sbinfo->si_xib);
34360 +       sbinfo->si_xib = file;
34361 +
34362 +       p = sbinfo->si_xib_buf;
34363 +       memset(p, 0, PAGE_SIZE);
34364 +       pos = 0;
34365 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xib, p, PAGE_SIZE, &pos);
34366 +       if (unlikely(sz != PAGE_SIZE)) {
34367 +               err = sz;
34368 +               AuIOErr("err %d\n", err);
34369 +               if (sz >= 0)
34370 +                       err = -EIO;
34371 +               goto out;
34372 +       }
34373 +
34374 +       mutex_lock(&sbinfo->si_xib_mtx);
34375 +       /* mnt_want_write() is unnecessary here */
34376 +       err = xib_restore(sb);
34377 +       mutex_unlock(&sbinfo->si_xib_mtx);
34378 +
34379 +out:
34380 +       return err;
34381 +}
34382 +
34383 +/* ---------------------------------------------------------------------- */
34384 +
34385 +/*
34386 + * xino mount option handlers
34387 + */
34388 +static au_readf_t find_readf(struct file *h_file)
34389 +{
34390 +       const struct file_operations *fop = h_file->f_op;
34391 +
34392 +       if (fop->read)
34393 +               return fop->read;
34394 +       if (fop->aio_read)
34395 +               return do_sync_read;
34396 +       if (fop->read_iter)
34397 +               return new_sync_read;
34398 +       return ERR_PTR(-ENOSYS);
34399 +}
34400 +
34401 +static au_writef_t find_writef(struct file *h_file)
34402 +{
34403 +       const struct file_operations *fop = h_file->f_op;
34404 +
34405 +       if (fop->write)
34406 +               return fop->write;
34407 +       if (fop->aio_write)
34408 +               return do_sync_write;
34409 +       if (fop->write_iter)
34410 +               return new_sync_write;
34411 +       return ERR_PTR(-ENOSYS);
34412 +}
34413 +
34414 +/* xino bitmap */
34415 +static void xino_clear_xib(struct super_block *sb)
34416 +{
34417 +       struct au_sbinfo *sbinfo;
34418 +
34419 +       SiMustWriteLock(sb);
34420 +
34421 +       sbinfo = au_sbi(sb);
34422 +       sbinfo->si_xread = NULL;
34423 +       sbinfo->si_xwrite = NULL;
34424 +       if (sbinfo->si_xib)
34425 +               fput(sbinfo->si_xib);
34426 +       sbinfo->si_xib = NULL;
34427 +       free_page((unsigned long)sbinfo->si_xib_buf);
34428 +       sbinfo->si_xib_buf = NULL;
34429 +}
34430 +
34431 +static int au_xino_set_xib(struct super_block *sb, struct file *base)
34432 +{
34433 +       int err;
34434 +       loff_t pos;
34435 +       struct au_sbinfo *sbinfo;
34436 +       struct file *file;
34437 +
34438 +       SiMustWriteLock(sb);
34439 +
34440 +       sbinfo = au_sbi(sb);
34441 +       file = au_xino_create2(base, sbinfo->si_xib);
34442 +       err = PTR_ERR(file);
34443 +       if (IS_ERR(file))
34444 +               goto out;
34445 +       if (sbinfo->si_xib)
34446 +               fput(sbinfo->si_xib);
34447 +       sbinfo->si_xib = file;
34448 +       sbinfo->si_xread = find_readf(file);
34449 +       sbinfo->si_xwrite = find_writef(file);
34450 +
34451 +       err = -ENOMEM;
34452 +       if (!sbinfo->si_xib_buf)
34453 +               sbinfo->si_xib_buf = (void *)get_zeroed_page(GFP_NOFS);
34454 +       if (unlikely(!sbinfo->si_xib_buf))
34455 +               goto out_unset;
34456 +
34457 +       sbinfo->si_xib_last_pindex = 0;
34458 +       sbinfo->si_xib_next_bit = 0;
34459 +       if (vfsub_f_size_read(file) < PAGE_SIZE) {
34460 +               pos = 0;
34461 +               err = xino_fwrite(sbinfo->si_xwrite, file, sbinfo->si_xib_buf,
34462 +                                 PAGE_SIZE, &pos);
34463 +               if (unlikely(err != PAGE_SIZE))
34464 +                       goto out_free;
34465 +       }
34466 +       err = 0;
34467 +       goto out; /* success */
34468 +
34469 +out_free:
34470 +       free_page((unsigned long)sbinfo->si_xib_buf);
34471 +       sbinfo->si_xib_buf = NULL;
34472 +       if (err >= 0)
34473 +               err = -EIO;
34474 +out_unset:
34475 +       fput(sbinfo->si_xib);
34476 +       sbinfo->si_xib = NULL;
34477 +       sbinfo->si_xread = NULL;
34478 +       sbinfo->si_xwrite = NULL;
34479 +out:
34480 +       return err;
34481 +}
34482 +
34483 +/* xino for each branch */
34484 +static void xino_clear_br(struct super_block *sb)
34485 +{
34486 +       aufs_bindex_t bindex, bend;
34487 +       struct au_branch *br;
34488 +
34489 +       bend = au_sbend(sb);
34490 +       for (bindex = 0; bindex <= bend; bindex++) {
34491 +               br = au_sbr(sb, bindex);
34492 +               if (!br || !br->br_xino.xi_file)
34493 +                       continue;
34494 +
34495 +               fput(br->br_xino.xi_file);
34496 +               br->br_xino.xi_file = NULL;
34497 +       }
34498 +}
34499 +
34500 +static int au_xino_set_br(struct super_block *sb, struct file *base)
34501 +{
34502 +       int err;
34503 +       ino_t ino;
34504 +       aufs_bindex_t bindex, bend, bshared;
34505 +       struct {
34506 +               struct file *old, *new;
34507 +       } *fpair, *p;
34508 +       struct au_branch *br;
34509 +       struct inode *inode;
34510 +       au_writef_t writef;
34511 +
34512 +       SiMustWriteLock(sb);
34513 +
34514 +       err = -ENOMEM;
34515 +       bend = au_sbend(sb);
34516 +       fpair = kcalloc(bend + 1, sizeof(*fpair), GFP_NOFS);
34517 +       if (unlikely(!fpair))
34518 +               goto out;
34519 +
34520 +       inode = sb->s_root->d_inode;
34521 +       ino = AUFS_ROOT_INO;
34522 +       writef = au_sbi(sb)->si_xwrite;
34523 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++) {
34524 +               br = au_sbr(sb, bindex);
34525 +               bshared = is_sb_shared(sb, bindex, bindex - 1);
34526 +               if (bshared >= 0) {
34527 +                       /* shared xino */
34528 +                       *p = fpair[bshared];
34529 +                       get_file(p->new);
34530 +               }
34531 +
34532 +               if (!p->new) {
34533 +                       /* new xino */
34534 +                       p->old = br->br_xino.xi_file;
34535 +                       p->new = au_xino_create2(base, br->br_xino.xi_file);
34536 +                       err = PTR_ERR(p->new);
34537 +                       if (IS_ERR(p->new)) {
34538 +                               p->new = NULL;
34539 +                               goto out_pair;
34540 +                       }
34541 +               }
34542 +
34543 +               err = au_xino_do_write(writef, p->new,
34544 +                                      au_h_iptr(inode, bindex)->i_ino, ino);
34545 +               if (unlikely(err))
34546 +                       goto out_pair;
34547 +       }
34548 +
34549 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++) {
34550 +               br = au_sbr(sb, bindex);
34551 +               if (br->br_xino.xi_file)
34552 +                       fput(br->br_xino.xi_file);
34553 +               get_file(p->new);
34554 +               br->br_xino.xi_file = p->new;
34555 +       }
34556 +
34557 +out_pair:
34558 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++)
34559 +               if (p->new)
34560 +                       fput(p->new);
34561 +               else
34562 +                       break;
34563 +       kfree(fpair);
34564 +out:
34565 +       return err;
34566 +}
34567 +
34568 +void au_xino_clr(struct super_block *sb)
34569 +{
34570 +       struct au_sbinfo *sbinfo;
34571 +
34572 +       au_xigen_clr(sb);
34573 +       xino_clear_xib(sb);
34574 +       xino_clear_br(sb);
34575 +       sbinfo = au_sbi(sb);
34576 +       /* lvalue, do not call au_mntflags() */
34577 +       au_opt_clr(sbinfo->si_mntflags, XINO);
34578 +}
34579 +
34580 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount)
34581 +{
34582 +       int err, skip;
34583 +       struct dentry *parent, *cur_parent;
34584 +       struct qstr *dname, *cur_name;
34585 +       struct file *cur_xino;
34586 +       struct inode *dir;
34587 +       struct au_sbinfo *sbinfo;
34588 +
34589 +       SiMustWriteLock(sb);
34590 +
34591 +       err = 0;
34592 +       sbinfo = au_sbi(sb);
34593 +       parent = dget_parent(xino->file->f_dentry);
34594 +       if (remount) {
34595 +               skip = 0;
34596 +               dname = &xino->file->f_dentry->d_name;
34597 +               cur_xino = sbinfo->si_xib;
34598 +               if (cur_xino) {
34599 +                       cur_parent = dget_parent(cur_xino->f_dentry);
34600 +                       cur_name = &cur_xino->f_dentry->d_name;
34601 +                       skip = (cur_parent == parent
34602 +                               && au_qstreq(dname, cur_name));
34603 +                       dput(cur_parent);
34604 +               }
34605 +               if (skip)
34606 +                       goto out;
34607 +       }
34608 +
34609 +       au_opt_set(sbinfo->si_mntflags, XINO);
34610 +       dir = parent->d_inode;
34611 +       mutex_lock_nested(&dir->i_mutex, AuLsc_I_PARENT);
34612 +       /* mnt_want_write() is unnecessary here */
34613 +       err = au_xino_set_xib(sb, xino->file);
34614 +       if (!err)
34615 +               err = au_xigen_set(sb, xino->file);
34616 +       if (!err)
34617 +               err = au_xino_set_br(sb, xino->file);
34618 +       mutex_unlock(&dir->i_mutex);
34619 +       if (!err)
34620 +               goto out; /* success */
34621 +
34622 +       /* reset all */
34623 +       AuIOErr("failed creating xino(%d).\n", err);
34624 +       au_xigen_clr(sb);
34625 +       xino_clear_xib(sb);
34626 +
34627 +out:
34628 +       dput(parent);
34629 +       return err;
34630 +}
34631 +
34632 +/* ---------------------------------------------------------------------- */
34633 +
34634 +/*
34635 + * create a xinofile at the default place/path.
34636 + */
34637 +struct file *au_xino_def(struct super_block *sb)
34638 +{
34639 +       struct file *file;
34640 +       char *page, *p;
34641 +       struct au_branch *br;
34642 +       struct super_block *h_sb;
34643 +       struct path path;
34644 +       aufs_bindex_t bend, bindex, bwr;
34645 +
34646 +       br = NULL;
34647 +       bend = au_sbend(sb);
34648 +       bwr = -1;
34649 +       for (bindex = 0; bindex <= bend; bindex++) {
34650 +               br = au_sbr(sb, bindex);
34651 +               if (au_br_writable(br->br_perm)
34652 +                   && !au_test_fs_bad_xino(au_br_sb(br))) {
34653 +                       bwr = bindex;
34654 +                       break;
34655 +               }
34656 +       }
34657 +
34658 +       if (bwr >= 0) {
34659 +               file = ERR_PTR(-ENOMEM);
34660 +               page = (void *)__get_free_page(GFP_NOFS);
34661 +               if (unlikely(!page))
34662 +                       goto out;
34663 +               path.mnt = au_br_mnt(br);
34664 +               path.dentry = au_h_dptr(sb->s_root, bwr);
34665 +               p = d_path(&path, page, PATH_MAX - sizeof(AUFS_XINO_FNAME));
34666 +               file = (void *)p;
34667 +               if (!IS_ERR(p)) {
34668 +                       strcat(p, "/" AUFS_XINO_FNAME);
34669 +                       AuDbg("%s\n", p);
34670 +                       file = au_xino_create(sb, p, /*silent*/0);
34671 +                       if (!IS_ERR(file))
34672 +                               au_xino_brid_set(sb, br->br_id);
34673 +               }
34674 +               free_page((unsigned long)page);
34675 +       } else {
34676 +               file = au_xino_create(sb, AUFS_XINO_DEFPATH, /*silent*/0);
34677 +               if (IS_ERR(file))
34678 +                       goto out;
34679 +               h_sb = file->f_dentry->d_sb;
34680 +               if (unlikely(au_test_fs_bad_xino(h_sb))) {
34681 +                       pr_err("xino doesn't support %s(%s)\n",
34682 +                              AUFS_XINO_DEFPATH, au_sbtype(h_sb));
34683 +                       fput(file);
34684 +                       file = ERR_PTR(-EINVAL);
34685 +               }
34686 +               if (!IS_ERR(file))
34687 +                       au_xino_brid_set(sb, -1);
34688 +       }
34689 +
34690 +out:
34691 +       return file;
34692 +}
34693 +
34694 +/* ---------------------------------------------------------------------- */
34695 +
34696 +int au_xino_path(struct seq_file *seq, struct file *file)
34697 +{
34698 +       int err;
34699 +
34700 +       err = au_seq_path(seq, &file->f_path);
34701 +       if (unlikely(err < 0))
34702 +               goto out;
34703 +
34704 +       err = 0;
34705 +#define Deleted "\\040(deleted)"
34706 +       seq->count -= sizeof(Deleted) - 1;
34707 +       AuDebugOn(memcmp(seq->buf + seq->count, Deleted,
34708 +                        sizeof(Deleted) - 1));
34709 +#undef Deleted
34710 +
34711 +out:
34712 +       return err;
34713 +}
34714 diff -urN /usr/share/empty/include/uapi/linux/aufs_type.h linux/include/uapi/linux/aufs_type.h
34715 --- /usr/share/empty/include/uapi/linux/aufs_type.h     1970-01-01 01:00:00.000000000 +0100
34716 +++ linux/include/uapi/linux/aufs_type.h        2015-01-25 13:00:38.637713742 +0100
34717 @@ -0,0 +1,419 @@
34718 +/*
34719 + * Copyright (C) 2005-2014 Junjiro R. Okajima
34720 + *
34721 + * This program, aufs is free software; you can redistribute it and/or modify
34722 + * it under the terms of the GNU General Public License as published by
34723 + * the Free Software Foundation; either version 2 of the License, or
34724 + * (at your option) any later version.
34725 + *
34726 + * This program is distributed in the hope that it will be useful,
34727 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
34728 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
34729 + * GNU General Public License for more details.
34730 + *
34731 + * You should have received a copy of the GNU General Public License
34732 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
34733 + */
34734 +
34735 +#ifndef __AUFS_TYPE_H__
34736 +#define __AUFS_TYPE_H__
34737 +
34738 +#define AUFS_NAME      "aufs"
34739 +
34740 +#ifdef __KERNEL__
34741 +/*
34742 + * define it before including all other headers.
34743 + * sched.h may use pr_* macros before defining "current", so define the
34744 + * no-current version first, and re-define later.
34745 + */
34746 +#define pr_fmt(fmt)    AUFS_NAME " %s:%d: " fmt, __func__, __LINE__
34747 +#include <linux/sched.h>
34748 +#undef pr_fmt
34749 +#define pr_fmt(fmt) \
34750 +               AUFS_NAME " %s:%d:%.*s[%d]: " fmt, __func__, __LINE__, \
34751 +               (int)sizeof(current->comm), current->comm, current->pid
34752 +#else
34753 +#include <stdint.h>
34754 +#include <sys/types.h>
34755 +#endif /* __KERNEL__ */
34756 +
34757 +#include <linux/limits.h>
34758 +
34759 +#define AUFS_VERSION   "3.18.1+-20150119"
34760 +
34761 +/* todo? move this to linux-2.6.19/include/magic.h */
34762 +#define AUFS_SUPER_MAGIC       ('a' << 24 | 'u' << 16 | 'f' << 8 | 's')
34763 +
34764 +/* ---------------------------------------------------------------------- */
34765 +
34766 +#ifdef CONFIG_AUFS_BRANCH_MAX_127
34767 +typedef int8_t aufs_bindex_t;
34768 +#define AUFS_BRANCH_MAX 127
34769 +#else
34770 +typedef int16_t aufs_bindex_t;
34771 +#ifdef CONFIG_AUFS_BRANCH_MAX_511
34772 +#define AUFS_BRANCH_MAX 511
34773 +#elif defined(CONFIG_AUFS_BRANCH_MAX_1023)
34774 +#define AUFS_BRANCH_MAX 1023
34775 +#elif defined(CONFIG_AUFS_BRANCH_MAX_32767)
34776 +#define AUFS_BRANCH_MAX 32767
34777 +#endif
34778 +#endif
34779 +
34780 +#ifdef __KERNEL__
34781 +#ifndef AUFS_BRANCH_MAX
34782 +#error unknown CONFIG_AUFS_BRANCH_MAX value
34783 +#endif
34784 +#endif /* __KERNEL__ */
34785 +
34786 +/* ---------------------------------------------------------------------- */
34787 +
34788 +#define AUFS_FSTYPE            AUFS_NAME
34789 +
34790 +#define AUFS_ROOT_INO          2
34791 +#define AUFS_FIRST_INO         11
34792 +
34793 +#define AUFS_WH_PFX            ".wh."
34794 +#define AUFS_WH_PFX_LEN                ((int)sizeof(AUFS_WH_PFX) - 1)
34795 +#define AUFS_WH_TMP_LEN                4
34796 +/* a limit for rmdir/rename a dir and copyup */
34797 +#define AUFS_MAX_NAMELEN       (NAME_MAX \
34798 +                               - AUFS_WH_PFX_LEN * 2   /* doubly whiteouted */\
34799 +                               - 1                     /* dot */\
34800 +                               - AUFS_WH_TMP_LEN)      /* hex */
34801 +#define AUFS_XINO_FNAME                "." AUFS_NAME ".xino"
34802 +#define AUFS_XINO_DEFPATH      "/tmp/" AUFS_XINO_FNAME
34803 +#define AUFS_XINO_DEF_SEC      30 /* seconds */
34804 +#define AUFS_XINO_DEF_TRUNC    45 /* percentage */
34805 +#define AUFS_DIRWH_DEF         3
34806 +#define AUFS_RDCACHE_DEF       10 /* seconds */
34807 +#define AUFS_RDCACHE_MAX       3600 /* seconds */
34808 +#define AUFS_RDBLK_DEF         512 /* bytes */
34809 +#define AUFS_RDHASH_DEF                32
34810 +#define AUFS_WKQ_NAME          AUFS_NAME "d"
34811 +#define AUFS_MFS_DEF_SEC       30 /* seconds */
34812 +#define AUFS_MFS_MAX_SEC       3600 /* seconds */
34813 +#define AUFS_FHSM_CACHE_DEF_SEC        30 /* seconds */
34814 +#define AUFS_PLINK_WARN                50 /* number of plinks in a single bucket */
34815 +
34816 +/* pseudo-link maintenace under /proc */
34817 +#define AUFS_PLINK_MAINT_NAME  "plink_maint"
34818 +#define AUFS_PLINK_MAINT_DIR   "fs/" AUFS_NAME
34819 +#define AUFS_PLINK_MAINT_PATH  AUFS_PLINK_MAINT_DIR "/" AUFS_PLINK_MAINT_NAME
34820 +
34821 +#define AUFS_DIROPQ_NAME       AUFS_WH_PFX ".opq" /* whiteouted doubly */
34822 +#define AUFS_WH_DIROPQ         AUFS_WH_PFX AUFS_DIROPQ_NAME
34823 +
34824 +#define AUFS_BASE_NAME         AUFS_WH_PFX AUFS_NAME
34825 +#define AUFS_PLINKDIR_NAME     AUFS_WH_PFX "plnk"
34826 +#define AUFS_ORPHDIR_NAME      AUFS_WH_PFX "orph"
34827 +
34828 +/* doubly whiteouted */
34829 +#define AUFS_WH_BASE           AUFS_WH_PFX AUFS_BASE_NAME
34830 +#define AUFS_WH_PLINKDIR       AUFS_WH_PFX AUFS_PLINKDIR_NAME
34831 +#define AUFS_WH_ORPHDIR                AUFS_WH_PFX AUFS_ORPHDIR_NAME
34832 +
34833 +/* branch permissions and attributes */
34834 +#define AUFS_BRPERM_RW         "rw"
34835 +#define AUFS_BRPERM_RO         "ro"
34836 +#define AUFS_BRPERM_RR         "rr"
34837 +#define AUFS_BRATTR_COO_REG    "coo_reg"
34838 +#define AUFS_BRATTR_COO_ALL    "coo_all"
34839 +#define AUFS_BRATTR_FHSM       "fhsm"
34840 +#define AUFS_BRATTR_UNPIN      "unpin"
34841 +#define AUFS_BRATTR_ICEX       "icex"
34842 +#define AUFS_BRATTR_ICEX_SEC   "icexsec"
34843 +#define AUFS_BRATTR_ICEX_SYS   "icexsys"
34844 +#define AUFS_BRATTR_ICEX_TR    "icextr"
34845 +#define AUFS_BRATTR_ICEX_USR   "icexusr"
34846 +#define AUFS_BRATTR_ICEX_OTH   "icexoth"
34847 +#define AUFS_BRRATTR_WH                "wh"
34848 +#define AUFS_BRWATTR_NLWH      "nolwh"
34849 +#define AUFS_BRWATTR_MOO       "moo"
34850 +
34851 +#define AuBrPerm_RW            1               /* writable, hardlinkable wh */
34852 +#define AuBrPerm_RO            (1 << 1)        /* readonly */
34853 +#define AuBrPerm_RR            (1 << 2)        /* natively readonly */
34854 +#define AuBrPerm_Mask          (AuBrPerm_RW | AuBrPerm_RO | AuBrPerm_RR)
34855 +
34856 +#define AuBrAttr_COO_REG       (1 << 3)        /* copy-up on open */
34857 +#define AuBrAttr_COO_ALL       (1 << 4)
34858 +#define AuBrAttr_COO_Mask      (AuBrAttr_COO_REG | AuBrAttr_COO_ALL)
34859 +
34860 +#define AuBrAttr_FHSM          (1 << 5)        /* file-based hsm */
34861 +#define AuBrAttr_UNPIN         (1 << 6)        /* rename-able top dir of
34862 +                                                  branch. meaningless since
34863 +                                                  linux-3.18-rc1 */
34864 +
34865 +/* ignore error in copying XATTR */
34866 +#define AuBrAttr_ICEX_SEC      (1 << 7)
34867 +#define AuBrAttr_ICEX_SYS      (1 << 8)
34868 +#define AuBrAttr_ICEX_TR       (1 << 9)
34869 +#define AuBrAttr_ICEX_USR      (1 << 10)
34870 +#define AuBrAttr_ICEX_OTH      (1 << 11)
34871 +#define AuBrAttr_ICEX          (AuBrAttr_ICEX_SEC      \
34872 +                                | AuBrAttr_ICEX_SYS    \
34873 +                                | AuBrAttr_ICEX_TR     \
34874 +                                | AuBrAttr_ICEX_USR    \
34875 +                                | AuBrAttr_ICEX_OTH)
34876 +
34877 +#define AuBrRAttr_WH           (1 << 12)       /* whiteout-able */
34878 +#define AuBrRAttr_Mask         AuBrRAttr_WH
34879 +
34880 +#define AuBrWAttr_NoLinkWH     (1 << 13)       /* un-hardlinkable whiteouts */
34881 +#define AuBrWAttr_MOO          (1 << 14)       /* move-up on open */
34882 +#define AuBrWAttr_Mask         (AuBrWAttr_NoLinkWH | AuBrWAttr_MOO)
34883 +
34884 +#define AuBrAttr_CMOO_Mask     (AuBrAttr_COO_Mask | AuBrWAttr_MOO)
34885 +
34886 +/* #warning test userspace */
34887 +#ifdef __KERNEL__
34888 +#ifndef CONFIG_AUFS_FHSM
34889 +#undef AuBrAttr_FHSM
34890 +#define AuBrAttr_FHSM          0
34891 +#endif
34892 +#ifndef CONFIG_AUFS_XATTR
34893 +#undef AuBrAttr_ICEX
34894 +#define AuBrAttr_ICEX          0
34895 +#undef AuBrAttr_ICEX_SEC
34896 +#define AuBrAttr_ICEX_SEC      0
34897 +#undef AuBrAttr_ICEX_SYS
34898 +#define AuBrAttr_ICEX_SYS      0
34899 +#undef AuBrAttr_ICEX_TR
34900 +#define AuBrAttr_ICEX_TR       0
34901 +#undef AuBrAttr_ICEX_USR
34902 +#define AuBrAttr_ICEX_USR      0
34903 +#undef AuBrAttr_ICEX_OTH
34904 +#define AuBrAttr_ICEX_OTH      0
34905 +#endif
34906 +#endif
34907 +
34908 +/* the longest combination */
34909 +/* AUFS_BRATTR_ICEX and AUFS_BRATTR_ICEX_TR don't affect here */
34910 +#define AuBrPermStrSz  sizeof(AUFS_BRPERM_RW                   \
34911 +                              "+" AUFS_BRATTR_COO_REG          \
34912 +                              "+" AUFS_BRATTR_FHSM             \
34913 +                              "+" AUFS_BRATTR_UNPIN            \
34914 +                              "+" AUFS_BRATTR_ICEX_SEC \
34915 +                              "+" AUFS_BRATTR_ICEX_SYS \
34916 +                              "+" AUFS_BRATTR_ICEX_USR \
34917 +                              "+" AUFS_BRATTR_ICEX_OTH \
34918 +                              "+" AUFS_BRWATTR_NLWH)
34919 +
34920 +typedef struct {
34921 +       char a[AuBrPermStrSz];
34922 +} au_br_perm_str_t;
34923 +
34924 +static inline int au_br_writable(int brperm)
34925 +{
34926 +       return brperm & AuBrPerm_RW;
34927 +}
34928 +
34929 +static inline int au_br_whable(int brperm)
34930 +{
34931 +       return brperm & (AuBrPerm_RW | AuBrRAttr_WH);
34932 +}
34933 +
34934 +static inline int au_br_wh_linkable(int brperm)
34935 +{
34936 +       return !(brperm & AuBrWAttr_NoLinkWH);
34937 +}
34938 +
34939 +static inline int au_br_cmoo(int brperm)
34940 +{
34941 +       return brperm & AuBrAttr_CMOO_Mask;
34942 +}
34943 +
34944 +static inline int au_br_fhsm(int brperm)
34945 +{
34946 +       return brperm & AuBrAttr_FHSM;
34947 +}
34948 +
34949 +/* ---------------------------------------------------------------------- */
34950 +
34951 +/* ioctl */
34952 +enum {
34953 +       /* readdir in userspace */
34954 +       AuCtl_RDU,
34955 +       AuCtl_RDU_INO,
34956 +
34957 +       AuCtl_WBR_FD,   /* pathconf wrapper */
34958 +       AuCtl_IBUSY,    /* busy inode */
34959 +       AuCtl_MVDOWN,   /* move-down */
34960 +       AuCtl_BR,       /* info about branches */
34961 +       AuCtl_FHSM_FD   /* connection for fhsm */
34962 +};
34963 +
34964 +/* borrowed from linux/include/linux/kernel.h */
34965 +#ifndef ALIGN
34966 +#define ALIGN(x, a)            __ALIGN_MASK(x, (typeof(x))(a)-1)
34967 +#define __ALIGN_MASK(x, mask)  (((x)+(mask))&~(mask))
34968 +#endif
34969 +
34970 +/* borrowed from linux/include/linux/compiler-gcc3.h */
34971 +#ifndef __aligned
34972 +#define __aligned(x)                   __attribute__((aligned(x)))
34973 +#endif
34974 +
34975 +#ifdef __KERNEL__
34976 +#ifndef __packed
34977 +#define __packed                       __attribute__((packed))
34978 +#endif
34979 +#endif
34980 +
34981 +struct au_rdu_cookie {
34982 +       uint64_t        h_pos;
34983 +       int16_t         bindex;
34984 +       uint8_t         flags;
34985 +       uint8_t         pad;
34986 +       uint32_t        generation;
34987 +} __aligned(8);
34988 +
34989 +struct au_rdu_ent {
34990 +       uint64_t        ino;
34991 +       int16_t         bindex;
34992 +       uint8_t         type;
34993 +       uint8_t         nlen;
34994 +       uint8_t         wh;
34995 +       char            name[0];
34996 +} __aligned(8);
34997 +
34998 +static inline int au_rdu_len(int nlen)
34999 +{
35000 +       /* include the terminating NULL */
35001 +       return ALIGN(sizeof(struct au_rdu_ent) + nlen + 1,
35002 +                    sizeof(uint64_t));
35003 +}
35004 +
35005 +union au_rdu_ent_ul {
35006 +       struct au_rdu_ent __user        *e;
35007 +       uint64_t                        ul;
35008 +};
35009 +
35010 +enum {
35011 +       AufsCtlRduV_SZ,
35012 +       AufsCtlRduV_End
35013 +};
35014 +
35015 +struct aufs_rdu {
35016 +       /* input */
35017 +       union {
35018 +               uint64_t        sz;     /* AuCtl_RDU */
35019 +               uint64_t        nent;   /* AuCtl_RDU_INO */
35020 +       };
35021 +       union au_rdu_ent_ul     ent;
35022 +       uint16_t                verify[AufsCtlRduV_End];
35023 +
35024 +       /* input/output */
35025 +       uint32_t                blk;
35026 +
35027 +       /* output */
35028 +       union au_rdu_ent_ul     tail;
35029 +       /* number of entries which were added in a single call */
35030 +       uint64_t                rent;
35031 +       uint8_t                 full;
35032 +       uint8_t                 shwh;
35033 +
35034 +       struct au_rdu_cookie    cookie;
35035 +} __aligned(8);
35036 +
35037 +/* ---------------------------------------------------------------------- */
35038 +
35039 +struct aufs_wbr_fd {
35040 +       uint32_t        oflags;
35041 +       int16_t         brid;
35042 +} __aligned(8);
35043 +
35044 +/* ---------------------------------------------------------------------- */
35045 +
35046 +struct aufs_ibusy {
35047 +       uint64_t        ino, h_ino;
35048 +       int16_t         bindex;
35049 +} __aligned(8);
35050 +
35051 +/* ---------------------------------------------------------------------- */
35052 +
35053 +/* error code for move-down */
35054 +/* the actual message strings are implemented in aufs-util.git */
35055 +enum {
35056 +       EAU_MVDOWN_OPAQUE = 1,
35057 +       EAU_MVDOWN_WHITEOUT,
35058 +       EAU_MVDOWN_UPPER,
35059 +       EAU_MVDOWN_BOTTOM,
35060 +       EAU_MVDOWN_NOUPPER,
35061 +       EAU_MVDOWN_NOLOWERBR,
35062 +       EAU_Last
35063 +};
35064 +
35065 +/* flags for move-down */
35066 +#define AUFS_MVDOWN_DMSG       1
35067 +#define AUFS_MVDOWN_OWLOWER    (1 << 1)        /* overwrite lower */
35068 +#define AUFS_MVDOWN_KUPPER     (1 << 2)        /* keep upper */
35069 +#define AUFS_MVDOWN_ROLOWER    (1 << 3)        /* do even if lower is RO */
35070 +#define AUFS_MVDOWN_ROLOWER_R  (1 << 4)        /* did on lower RO */
35071 +#define AUFS_MVDOWN_ROUPPER    (1 << 5)        /* do even if upper is RO */
35072 +#define AUFS_MVDOWN_ROUPPER_R  (1 << 6)        /* did on upper RO */
35073 +#define AUFS_MVDOWN_BRID_UPPER (1 << 7)        /* upper brid */
35074 +#define AUFS_MVDOWN_BRID_LOWER (1 << 8)        /* lower brid */
35075 +#define AUFS_MVDOWN_FHSM_LOWER (1 << 9)        /* find fhsm attr for lower */
35076 +#define AUFS_MVDOWN_STFS       (1 << 10)       /* req. stfs */
35077 +#define AUFS_MVDOWN_STFS_FAILED        (1 << 11)       /* output: stfs is unusable */
35078 +#define AUFS_MVDOWN_BOTTOM     (1 << 12)       /* output: no more lowers */
35079 +
35080 +/* index for move-down */
35081 +enum {
35082 +       AUFS_MVDOWN_UPPER,
35083 +       AUFS_MVDOWN_LOWER,
35084 +       AUFS_MVDOWN_NARRAY
35085 +};
35086 +
35087 +/*
35088 + * additional info of move-down
35089 + * number of free blocks and inodes.
35090 + * subset of struct kstatfs, but smaller and always 64bit.
35091 + */
35092 +struct aufs_stfs {
35093 +       uint64_t        f_blocks;
35094 +       uint64_t        f_bavail;
35095 +       uint64_t        f_files;
35096 +       uint64_t        f_ffree;
35097 +};
35098 +
35099 +struct aufs_stbr {
35100 +       int16_t                 brid;   /* optional input */
35101 +       int16_t                 bindex; /* output */
35102 +       struct aufs_stfs        stfs;   /* output when AUFS_MVDOWN_STFS set */
35103 +} __aligned(8);
35104 +
35105 +struct aufs_mvdown {
35106 +       uint32_t                flags;                  /* input/output */
35107 +       struct aufs_stbr        stbr[AUFS_MVDOWN_NARRAY]; /* input/output */
35108 +       int8_t                  au_errno;               /* output */
35109 +} __aligned(8);
35110 +
35111 +/* ---------------------------------------------------------------------- */
35112 +
35113 +union aufs_brinfo {
35114 +       /* PATH_MAX may differ between kernel-space and user-space */
35115 +       char    _spacer[4096];
35116 +       struct {
35117 +               int16_t id;
35118 +               int     perm;
35119 +               char    path[0];
35120 +       };
35121 +} __aligned(8);
35122 +
35123 +/* ---------------------------------------------------------------------- */
35124 +
35125 +#define AuCtlType              'A'
35126 +#define AUFS_CTL_RDU           _IOWR(AuCtlType, AuCtl_RDU, struct aufs_rdu)
35127 +#define AUFS_CTL_RDU_INO       _IOWR(AuCtlType, AuCtl_RDU_INO, struct aufs_rdu)
35128 +#define AUFS_CTL_WBR_FD                _IOW(AuCtlType, AuCtl_WBR_FD, \
35129 +                                    struct aufs_wbr_fd)
35130 +#define AUFS_CTL_IBUSY         _IOWR(AuCtlType, AuCtl_IBUSY, struct aufs_ibusy)
35131 +#define AUFS_CTL_MVDOWN                _IOWR(AuCtlType, AuCtl_MVDOWN, \
35132 +                                     struct aufs_mvdown)
35133 +#define AUFS_CTL_BRINFO                _IOW(AuCtlType, AuCtl_BR, union aufs_brinfo)
35134 +#define AUFS_CTL_FHSM_FD       _IOW(AuCtlType, AuCtl_FHSM_FD, int)
35135 +
35136 +#endif /* __AUFS_TYPE_H__ */
35137 aufs3.18.1+ loopback patch
35138
35139 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
35140 index 30efd68..77b31b4 100644
35141 --- a/drivers/block/loop.c
35142 +++ b/drivers/block/loop.c
35143 @@ -514,7 +514,7 @@ out:
35144  }
35145  
35146  struct switch_request {
35147 -       struct file *file;
35148 +       struct file *file, *virt_file;
35149         struct completion wait;
35150  };
35151  
35152 @@ -576,7 +576,8 @@ static int loop_thread(void *data)
35153   * First it needs to flush existing IO, it does this by sending a magic
35154   * BIO down the pipe. The completion of this BIO does the actual switch.
35155   */
35156 -static int loop_switch(struct loop_device *lo, struct file *file)
35157 +static int loop_switch(struct loop_device *lo, struct file *file,
35158 +                      struct file *virt_file)
35159  {
35160         struct switch_request w;
35161         struct bio *bio = bio_alloc(GFP_KERNEL, 0);
35162 @@ -584,6 +585,7 @@ static int loop_switch(struct loop_device *lo, struct file *file)
35163                 return -ENOMEM;
35164         init_completion(&w.wait);
35165         w.file = file;
35166 +       w.virt_file = virt_file;
35167         bio->bi_private = &w;
35168         bio->bi_bdev = NULL;
35169         loop_make_request(lo->lo_queue, bio);
35170 @@ -600,7 +602,7 @@ static int loop_flush(struct loop_device *lo)
35171         if (!lo->lo_thread)
35172                 return 0;
35173  
35174 -       return loop_switch(lo, NULL);
35175 +       return loop_switch(lo, NULL, NULL);
35176  }
35177  
35178  /*
35179 @@ -619,6 +621,7 @@ static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
35180         mapping = file->f_mapping;
35181         mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
35182         lo->lo_backing_file = file;
35183 +       lo->lo_backing_virt_file = p->virt_file;
35184         lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
35185                 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
35186         lo->old_gfp_mask = mapping_gfp_mask(mapping);
35187 @@ -627,6 +630,13 @@ out:
35188         complete(&p->wait);
35189  }
35190  
35191 +static struct file *loop_real_file(struct file *file)
35192 +{
35193 +       struct file *f = NULL;
35194 +       if (file->f_dentry->d_sb->s_op->real_loop)
35195 +               f = file->f_dentry->d_sb->s_op->real_loop(file);
35196 +       return f;
35197 +}
35198  
35199  /*
35200   * loop_change_fd switched the backing store of a loopback device to
35201 @@ -640,6 +650,7 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
35202                           unsigned int arg)
35203  {
35204         struct file     *file, *old_file;
35205 +       struct file     *f, *virt_file = NULL, *old_virt_file;
35206         struct inode    *inode;
35207         int             error;
35208  
35209 @@ -656,9 +667,16 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
35210         file = fget(arg);
35211         if (!file)
35212                 goto out;
35213 +       f = loop_real_file(file);
35214 +       if (f) {
35215 +               virt_file = file;
35216 +               file = f;
35217 +               get_file(file);
35218 +       }
35219  
35220         inode = file->f_mapping->host;
35221         old_file = lo->lo_backing_file;
35222 +       old_virt_file = lo->lo_backing_virt_file;
35223  
35224         error = -EINVAL;
35225  
35226 @@ -670,17 +688,21 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
35227                 goto out_putf;
35228  
35229         /* and ... switch */
35230 -       error = loop_switch(lo, file);
35231 +       error = loop_switch(lo, file, virt_file);
35232         if (error)
35233                 goto out_putf;
35234  
35235         fput(old_file);
35236 +       if (old_virt_file)
35237 +               fput(old_virt_file);
35238         if (lo->lo_flags & LO_FLAGS_PARTSCAN)
35239                 ioctl_by_bdev(bdev, BLKRRPART, 0);
35240         return 0;
35241  
35242   out_putf:
35243         fput(file);
35244 +       if (virt_file)
35245 +               fput(virt_file);
35246   out:
35247         return error;
35248  }
35249 @@ -841,7 +863,7 @@ static void loop_config_discard(struct loop_device *lo)
35250  static int loop_set_fd(struct loop_device *lo, fmode_t mode,
35251                        struct block_device *bdev, unsigned int arg)
35252  {
35253 -       struct file     *file, *f;
35254 +       struct file     *file, *f, *virt_file = NULL;
35255         struct inode    *inode;
35256         struct address_space *mapping;
35257         unsigned lo_blocksize;
35258 @@ -856,6 +878,12 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
35259         file = fget(arg);
35260         if (!file)
35261                 goto out;
35262 +       f = loop_real_file(file);
35263 +       if (f) {
35264 +               virt_file = file;
35265 +               file = f;
35266 +               get_file(file);
35267 +       }
35268  
35269         error = -EBUSY;
35270         if (lo->lo_state != Lo_unbound)
35271 @@ -904,6 +932,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
35272         lo->lo_device = bdev;
35273         lo->lo_flags = lo_flags;
35274         lo->lo_backing_file = file;
35275 +       lo->lo_backing_virt_file = virt_file;
35276         lo->transfer = transfer_none;
35277         lo->ioctl = NULL;
35278         lo->lo_sizelimit = 0;
35279 @@ -948,6 +977,7 @@ out_clr:
35280         lo->lo_thread = NULL;
35281         lo->lo_device = NULL;
35282         lo->lo_backing_file = NULL;
35283 +       lo->lo_backing_virt_file = NULL;
35284         lo->lo_flags = 0;
35285         set_capacity(lo->lo_disk, 0);
35286         invalidate_bdev(bdev);
35287 @@ -957,6 +987,8 @@ out_clr:
35288         lo->lo_state = Lo_unbound;
35289   out_putf:
35290         fput(file);
35291 +       if (virt_file)
35292 +               fput(virt_file);
35293   out:
35294         /* This is safe: open() is still holding a reference. */
35295         module_put(THIS_MODULE);
35296 @@ -1003,6 +1035,7 @@ loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
35297  static int loop_clr_fd(struct loop_device *lo)
35298  {
35299         struct file *filp = lo->lo_backing_file;
35300 +       struct file *virt_filp = lo->lo_backing_virt_file;
35301         gfp_t gfp = lo->old_gfp_mask;
35302         struct block_device *bdev = lo->lo_device;
35303  
35304 @@ -1036,6 +1069,7 @@ static int loop_clr_fd(struct loop_device *lo)
35305  
35306         spin_lock_irq(&lo->lo_lock);
35307         lo->lo_backing_file = NULL;
35308 +       lo->lo_backing_virt_file = NULL;
35309         spin_unlock_irq(&lo->lo_lock);
35310  
35311         loop_release_xfer(lo);
35312 @@ -1078,6 +1112,8 @@ static int loop_clr_fd(struct loop_device *lo)
35313          * bd_mutex which is usually taken before lo_ctl_mutex.
35314          */
35315         fput(filp);
35316 +       if (virt_filp)
35317 +               fput(virt_filp);
35318         return 0;
35319  }
35320  
35321 diff --git a/drivers/block/loop.h b/drivers/block/loop.h
35322 index 90df5d6..cb91822 100644
35323 --- a/drivers/block/loop.h
35324 +++ b/drivers/block/loop.h
35325 @@ -44,7 +44,7 @@ struct loop_device {
35326         int             (*ioctl)(struct loop_device *, int cmd, 
35327                                  unsigned long arg); 
35328  
35329 -       struct file *   lo_backing_file;
35330 +       struct file *   lo_backing_file, *lo_backing_virt_file;
35331         struct block_device *lo_device;
35332         unsigned        lo_blocksize;
35333         void            *key_data; 
35334 diff --git a/fs/aufs/f_op.c b/fs/aufs/f_op.c
35335 index 3dd9138..63a424c 100644
35336 --- a/fs/aufs/f_op.c
35337 +++ b/fs/aufs/f_op.c
35338 @@ -367,7 +367,7 @@ static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
35339         err = -EINVAL;
35340         h_file = au_hf_top(file);
35341         get_file(h_file);
35342 -       if (au_test_loopback_kthread()) {
35343 +       if (0 && au_test_loopback_kthread()) {
35344                 au_warn_loopback(h_file->f_dentry->d_sb);
35345                 if (file->f_mapping != h_file->f_mapping) {
35346                         file->f_mapping = h_file->f_mapping;
35347 diff --git a/fs/aufs/loop.c b/fs/aufs/loop.c
35348 index 3b03b52..4ab749d 100644
35349 --- a/fs/aufs/loop.c
35350 +++ b/fs/aufs/loop.c
35351 @@ -130,3 +130,19 @@ void au_loopback_fin(void)
35352         symbol_put(loop_backing_file);
35353         kfree(au_warn_loopback_array);
35354  }
35355 +
35356 +/* ---------------------------------------------------------------------- */
35357 +
35358 +/* support the loopback block device insude aufs */
35359 +
35360 +struct file *aufs_real_loop(struct file *file)
35361 +{
35362 +       struct file *f;
35363 +
35364 +       BUG_ON(!au_test_aufs(file->f_dentry->d_sb));
35365 +       fi_read_lock(file);
35366 +       f = au_hf_top(file);
35367 +       fi_read_unlock(file);
35368 +       AuDebugOn(!f);
35369 +       return f;
35370 +}
35371 diff --git a/fs/aufs/loop.h b/fs/aufs/loop.h
35372 index da8b756..28cb7ea 100644
35373 --- a/fs/aufs/loop.h
35374 +++ b/fs/aufs/loop.h
35375 @@ -25,7 +25,11 @@ void au_warn_loopback(struct super_block *h_sb);
35376  
35377  int au_loopback_init(void);
35378  void au_loopback_fin(void);
35379 +
35380 +struct file *aufs_real_loop(struct file *file);
35381  #else
35382 +AuStub(struct file *, loop_backing_file, return NULL)
35383 +
35384  AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
35385            struct dentry *h_adding)
35386  AuStubInt0(au_test_loopback_kthread, void)
35387 @@ -33,6 +37,8 @@ AuStubVoid(au_warn_loopback, struct super_block *h_sb)
35388  
35389  AuStubInt0(au_loopback_init, void)
35390  AuStubVoid(au_loopback_fin, void)
35391 +
35392 +AuStub(struct file *, aufs_real_loop, return NULL, struct file *file)
35393  #endif /* BLK_DEV_LOOP */
35394  
35395  #endif /* __KERNEL__ */
35396 diff --git a/fs/aufs/super.c b/fs/aufs/super.c
35397 index 2ec0b4f..65a6781 100644
35398 --- a/fs/aufs/super.c
35399 +++ b/fs/aufs/super.c
35400 @@ -812,7 +812,10 @@ static const struct super_operations aufs_sop = {
35401         .statfs         = aufs_statfs,
35402         .put_super      = aufs_put_super,
35403         .sync_fs        = aufs_sync_fs,
35404 -       .remount_fs     = aufs_remount_fs
35405 +       .remount_fs     = aufs_remount_fs,
35406 +#ifdef CONFIG_AUFS_BDEV_LOOP
35407 +       .real_loop      = aufs_real_loop
35408 +#endif
35409  };
35410  
35411  /* ---------------------------------------------------------------------- */
35412 diff --git a/include/linux/fs.h b/include/linux/fs.h
35413 index aabcbba..1a634f5 100644
35414 --- a/include/linux/fs.h
35415 +++ b/include/linux/fs.h
35416 @@ -1594,6 +1594,10 @@ struct super_operations {
35417         int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
35418         long (*nr_cached_objects)(struct super_block *, int);
35419         long (*free_cached_objects)(struct super_block *, long, int);
35420 +#if defined(CONFIG_BLK_DEV_LOOP) ||  defined(CONFIG_BLK_DEV_LOOP_MODULE)
35421 +       /* and aufs */
35422 +       struct file *(*real_loop)(struct file *);
35423 +#endif
35424  };
35425  
35426  /*
35427 diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
35428 index 50777b5..f9c6b3d 100644
35429 --- a/include/linux/shmem_fs.h
35430 +++ b/include/linux/shmem_fs.h
35431 @@ -26,10 +26,13 @@ struct shmem_inode_info {
35432  };
35433  
35434  struct shmem_sb_info {
35435 +       struct mutex idr_lock;
35436 +       bool idr_nouse;
35437 +       struct idr idr;             /* manages inode-number */
35438         unsigned long max_blocks;   /* How many blocks are allowed */
35439         struct percpu_counter used_blocks;  /* How many are allocated */
35440 -       unsigned long max_inodes;   /* How many inodes are allowed */
35441 -       unsigned long free_inodes;  /* How many are left for allocation */
35442 +       int max_inodes;             /* How many inodes are allowed */
35443 +       int free_inodes;            /* How many are left for allocation */
35444         spinlock_t stat_lock;       /* Serialize shmem_sb_info changes */
35445         kuid_t uid;                 /* Mount uid for root directory */
35446         kgid_t gid;                 /* Mount gid for root directory */
35447 diff --git a/mm/shmem.c b/mm/shmem.c
35448 index 185836b..ac7f9fb 100644
35449 --- a/mm/shmem.c
35450 +++ b/mm/shmem.c
35451 @@ -110,9 +110,13 @@ static unsigned long shmem_default_max_blocks(void)
35452         return totalram_pages / 2;
35453  }
35454  
35455 -static unsigned long shmem_default_max_inodes(void)
35456 +static int shmem_default_max_inodes(void)
35457  {
35458 -       return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
35459 +       unsigned long ul;
35460 +
35461 +       ul = INT_MAX;
35462 +       ul = min3(ul, totalram_pages - totalhigh_pages, totalram_pages / 2);
35463 +       return ul;
35464  }
35465  #endif
35466  
35467 @@ -592,6 +596,7 @@ static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
35468  static void shmem_evict_inode(struct inode *inode)
35469  {
35470         struct shmem_inode_info *info = SHMEM_I(inode);
35471 +       struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
35472  
35473         if (inode->i_mapping->a_ops == &shmem_aops) {
35474                 shmem_unacct_size(info->flags, inode->i_size);
35475 @@ -607,6 +612,11 @@ static void shmem_evict_inode(struct inode *inode)
35476  
35477         simple_xattrs_free(&info->xattrs);
35478         WARN_ON(inode->i_blocks);
35479 +       if (!sbinfo->idr_nouse && inode->i_ino) {
35480 +               mutex_lock(&sbinfo->idr_lock);
35481 +               idr_remove(&sbinfo->idr, inode->i_ino);
35482 +               mutex_unlock(&sbinfo->idr_lock);
35483 +       }
35484         shmem_free_inode(inode->i_sb);
35485         clear_inode(inode);
35486  }
35487 @@ -1406,13 +1416,13 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
35488         struct inode *inode;
35489         struct shmem_inode_info *info;
35490         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
35491 +       int ino;
35492  
35493         if (shmem_reserve_inode(sb))
35494                 return NULL;
35495  
35496         inode = new_inode(sb);
35497         if (inode) {
35498 -               inode->i_ino = get_next_ino();
35499                 inode_init_owner(inode, dir, mode);
35500                 inode->i_blocks = 0;
35501                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
35502 @@ -1454,6 +1464,25 @@ static struct inode *shmem_get_inode(struct super_block *sb, const struct inode
35503                         mpol_shared_policy_init(&info->policy, NULL);
35504                         break;
35505                 }
35506 +
35507 +               if (!sbinfo->idr_nouse) {
35508 +                       /* inum 0 and 1 are unused */
35509 +                       mutex_lock(&sbinfo->idr_lock);
35510 +                       ino = idr_alloc(&sbinfo->idr, inode, 2, INT_MAX,
35511 +                                       GFP_NOFS);
35512 +                       if (ino > 0) {
35513 +                               inode->i_ino = ino;
35514 +                               mutex_unlock(&sbinfo->idr_lock);
35515 +                               __insert_inode_hash(inode, inode->i_ino);
35516 +                       } else {
35517 +                               inode->i_ino = 0;
35518 +                               mutex_unlock(&sbinfo->idr_lock);
35519 +                               iput(inode);
35520 +                               /* shmem_free_inode() will be called */
35521 +                               inode = NULL;
35522 +                       }
35523 +               } else
35524 +                       inode->i_ino = get_next_ino();
35525         } else
35526                 shmem_free_inode(sb);
35527         return inode;
35528 @@ -2674,8 +2703,7 @@ static struct dentry *shmem_get_parent(struct dentry *child)
35529  static int shmem_match(struct inode *ino, void *vfh)
35530  {
35531         __u32 *fh = vfh;
35532 -       __u64 inum = fh[2];
35533 -       inum = (inum << 32) | fh[1];
35534 +       __u64 inum = fh[1];
35535         return ino->i_ino == inum && fh[0] == ino->i_generation;
35536  }
35537  
35538 @@ -2686,14 +2714,11 @@ static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
35539         struct dentry *dentry = NULL;
35540         u64 inum;
35541  
35542 -       if (fh_len < 3)
35543 +       if (fh_len < 2)
35544                 return NULL;
35545  
35546 -       inum = fid->raw[2];
35547 -       inum = (inum << 32) | fid->raw[1];
35548 -
35549 -       inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
35550 -                       shmem_match, fid->raw);
35551 +       inum = fid->raw[1];
35552 +       inode = ilookup5(sb, inum, shmem_match, fid->raw);
35553         if (inode) {
35554                 dentry = d_find_alias(inode);
35555                 iput(inode);
35556 @@ -2705,30 +2730,15 @@ static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
35557  static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
35558                                 struct inode *parent)
35559  {
35560 -       if (*len < 3) {
35561 -               *len = 3;
35562 +       if (*len < 2) {
35563 +               *len = 2;
35564                 return FILEID_INVALID;
35565         }
35566  
35567 -       if (inode_unhashed(inode)) {
35568 -               /* Unfortunately insert_inode_hash is not idempotent,
35569 -                * so as we hash inodes here rather than at creation
35570 -                * time, we need a lock to ensure we only try
35571 -                * to do it once
35572 -                */
35573 -               static DEFINE_SPINLOCK(lock);
35574 -               spin_lock(&lock);
35575 -               if (inode_unhashed(inode))
35576 -                       __insert_inode_hash(inode,
35577 -                                           inode->i_ino + inode->i_generation);
35578 -               spin_unlock(&lock);
35579 -       }
35580 -
35581         fh[0] = inode->i_generation;
35582         fh[1] = inode->i_ino;
35583 -       fh[2] = ((__u64)inode->i_ino) >> 32;
35584  
35585 -       *len = 3;
35586 +       *len = 2;
35587         return 1;
35588  }
35589  
35590 @@ -2793,7 +2803,7 @@ static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
35591                                 goto bad_val;
35592                 } else if (!strcmp(this_char,"nr_inodes")) {
35593                         sbinfo->max_inodes = memparse(value, &rest);
35594 -                       if (*rest)
35595 +                       if (*rest || sbinfo->max_inodes < 2)
35596                                 goto bad_val;
35597                 } else if (!strcmp(this_char,"mode")) {
35598                         if (remount)
35599 @@ -2846,7 +2856,7 @@ static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
35600  {
35601         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
35602         struct shmem_sb_info config = *sbinfo;
35603 -       unsigned long inodes;
35604 +       int inodes;
35605         int error = -EINVAL;
35606  
35607         config.mpol = NULL;
35608 @@ -2894,7 +2904,7 @@ static int shmem_show_options(struct seq_file *seq, struct dentry *root)
35609                 seq_printf(seq, ",size=%luk",
35610                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
35611         if (sbinfo->max_inodes != shmem_default_max_inodes())
35612 -               seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
35613 +               seq_printf(seq, ",nr_inodes=%d", sbinfo->max_inodes);
35614         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
35615                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
35616         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
35617 @@ -2983,6 +2993,8 @@ static void shmem_put_super(struct super_block *sb)
35618  {
35619         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
35620  
35621 +       if (!sbinfo->idr_nouse)
35622 +               idr_destroy(&sbinfo->idr);
35623         percpu_counter_destroy(&sbinfo->used_blocks);
35624         mpol_put(sbinfo->mpol);
35625         kfree(sbinfo);
35626 @@ -3001,6 +3013,8 @@ int shmem_fill_super(struct super_block *sb, void *data, int silent)
35627         if (!sbinfo)
35628                 return -ENOMEM;
35629  
35630 +       mutex_init(&sbinfo->idr_lock);
35631 +       idr_init(&sbinfo->idr);
35632         sbinfo->mode = S_IRWXUGO | S_ISVTX;
35633         sbinfo->uid = current_fsuid();
35634         sbinfo->gid = current_fsgid();
35635 @@ -3104,6 +3118,15 @@ static void shmem_destroy_inodecache(void)
35636         kmem_cache_destroy(shmem_inode_cachep);
35637  }
35638  
35639 +static __init void shmem_no_idr(struct super_block *sb)
35640 +{
35641 +       struct shmem_sb_info *sbinfo;
35642 +
35643 +       sbinfo = SHMEM_SB(sb);
35644 +       sbinfo->idr_nouse = true;
35645 +       idr_destroy(&sbinfo->idr);
35646 +}
35647 +
35648  static const struct address_space_operations shmem_aops = {
35649         .writepage      = shmem_writepage,
35650         .set_page_dirty = __set_page_dirty_no_writeback,
35651 @@ -3246,6 +3269,7 @@ int __init shmem_init(void)
35652                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
35653                 goto out1;
35654         }
35655 +       shmem_no_idr(shm_mnt->mnt_sb);
35656         return 0;
35657  
35658  out1:
35659 diff --git a/fs/inode.c b/fs/inode.c
35660 index 26753ba..61e0af2 100644
35661 --- a/fs/inode.c
35662 +++ b/fs/inode.c
35663 @@ -840,6 +840,8 @@ unsigned int get_next_ino(void)
35664         unsigned int *p = &get_cpu_var(last_ino);
35665         unsigned int res = *p;
35666  
35667 +start:
35668 +
35669  #ifdef CONFIG_SMP
35670         if (unlikely((res & (LAST_INO_BATCH-1)) == 0)) {
35671                 static atomic_t shared_last_ino;
35672 @@ -849,7 +851,9 @@ unsigned int get_next_ino(void)
35673         }
35674  #endif
35675  
35676 -       *p = ++res;
35677 +       if (unlikely(!++res))
35678 +               goto start;     /* never zero */
35679 +       *p = res;
35680         put_cpu_var(last_ino);
35681         return res;
35682  }
This page took 3.462397 seconds and 3 git commands to generate.