]> git.pld-linux.org Git - packages/kernel.git/blob - kernel-aufs3.patch
- updated aufs3 patch from upstream
[packages/kernel.git] / kernel-aufs3.patch
1 aufs3.19 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 bedff48..9e7c0b7 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 00b10002..f4a389c 100644
26 --- a/include/uapi/linux/Kbuild
27 +++ b/include/uapi/linux/Kbuild
28 @@ -57,6 +57,7 @@ header-y += atmsvc.h
29  header-y += atm_tcp.h
30  header-y += atm_zatm.h
31  header-y += audit.h
32 +header-y += aufs_type.h
33  header-y += auto_fs4.h
34  header-y += auto_fs.h
35  header-y += auxvec.h
36 aufs3.19 base patch
37
38 diff --git a/MAINTAINERS b/MAINTAINERS
39 index d66a97d..1c113a4 100644
40 --- a/MAINTAINERS
41 +++ b/MAINTAINERS
42 @@ -1833,6 +1833,20 @@ 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 e368d4f..eaed72c 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 aa149e7..98d87de 100644
107 --- a/fs/inode.c
108 +++ b/fs/inode.c
109 @@ -1499,7 +1499,7 @@ 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 42efe13..f634198 100644
148 --- a/include/linux/fs.h
149 +++ b/include/linux/fs.h
150 @@ -2690,6 +2690,7 @@ 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 aufs3.19 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 3f3d7ae..426bcc7 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 246eae8..dfd0875 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 @@ -1475,7 +1478,7 @@ 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 dd5ea30..445d798 100644
260 --- a/include/linux/mm.h
261 +++ b/include/linux/mm.h
262 @@ -1224,6 +1224,28 @@ 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 6d34aa2..f13dd51 100644
293 --- a/include/linux/mm_types.h
294 +++ b/include/linux/mm_types.h
295 @@ -233,6 +233,7 @@ 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 @@ -301,6 +302,7 @@ 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 4dc2dda..5984c61 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                         i_mmap_lock_write(mapping);
324 diff --git a/mm/Makefile b/mm/Makefile
325 index 4bf586e..59bd276 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 673e458..a623932 100644
339 --- a/mm/filemap.c
340 +++ b/mm/filemap.c
341 @@ -2063,7 +2063,7 @@ 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 2805d71..1b011b0 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 a271adc..29a932f 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 2c3536c..444bb1d 100644
406 --- a/mm/memory.c
407 +++ b/mm/memory.c
408 @@ -2157,7 +2157,7 @@ reuse:
409  
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 7f684d5..ffa1b91 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 @@ -897,7 +897,7 @@ 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 @@ -1682,8 +1682,8 @@ 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 @@ -2484,7 +2484,7 @@ 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 @@ -2503,7 +2503,7 @@ 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 @@ -2895,7 +2895,7 @@ 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 28bd8c4..3c0ace2 100644
498 --- a/mm/nommu.c
499 +++ b/mm/nommu.c
500 @@ -659,7 +659,7 @@ 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 @@ -824,7 +824,7 @@ 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 @@ -1375,7 +1375,7 @@ 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 @@ -1451,10 +1451,10 @@ 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 aufs3.19 standalone patch
633
634 diff --git a/fs/dcache.c b/fs/dcache.c
635 index eaed72c..3b4386e 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 98d87de..084d84f 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 @@ -1515,6 +1516,7 @@ 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 cd1e968..a99a3a7 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 @@ -1728,6 +1729,7 @@ 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 92e48c7..d2c4b68 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  /* Calculate mask of events for a list of marks */
733  u32 fsnotify_recalc_mask(struct hlist_head *head)
734 @@ -202,6 +203,7 @@ 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  /*
741   * Destroy all marks in the given list. The marks must be already detached from
742 @@ -376,6 +378,7 @@ 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 @@ -455,6 +458,7 @@ 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 813be03..328f0d6 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 4ef6985..6bb6303 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 2915d85..58382dd 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        2015-03-27 21:56:35.453461667 +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  2015-03-27 21:56:35.456795001 +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     2015-03-27 21:56:35.456795001 +0100
1014 @@ -0,0 +1,161 @@
1015 +
1016 +# Copyright (C) 2005-2015 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    2015-03-27 21:56:35.456795001 +0100
1179 @@ -0,0 +1,251 @@
1180 +
1181 +# Copyright (C) 2005-2015 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    2015-03-27 21:56:35.456795001 +0100
1434 @@ -0,0 +1,133 @@
1435 +
1436 +# Copyright (C) 2005-2015 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    2015-03-27 21:56:35.456795001 +0100
1571 @@ -0,0 +1,75 @@
1572 +
1573 +# Copyright (C) 2005-2015 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        2015-03-27 21:56:35.456795001 +0100
1650 @@ -0,0 +1,64 @@
1651 +
1652 +# Copyright (C) 2005-2015 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      2015-03-27 21:56:35.456795001 +0100
1718 @@ -0,0 +1,120 @@
1719 +
1720 +# Copyright (C) 2011-2015 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      2015-03-27 21:56:35.460128334 +0100
1842 @@ -0,0 +1,46 @@
1843 +
1844 +# Copyright (C) 2005-2015 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     2015-03-27 21:56:35.460128334 +0100
1892 @@ -0,0 +1,96 @@
1893 +
1894 +# Copyright (C) 2014-2015 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    2015-03-27 21:56:35.460128334 +0100
1992 @@ -0,0 +1,58 @@
1993 +
1994 +# Copyright (C) 2005-2015 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      2015-03-27 21:56:35.460128334 +0100
2054 @@ -0,0 +1,52 @@
2055 +
2056 +# Copyright (C) 2005-2015 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     2015-03-27 21:56:35.460128334 +0100
2110 @@ -0,0 +1,46 @@
2111 +
2112 +# Copyright (C) 2010-2015 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      2015-03-27 21:56:35.460128334 +0100
2160 @@ -0,0 +1,58 @@
2161 +
2162 +# Copyright (C) 2005-2015 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 2015-03-27 21:56:35.456795001 +0100
2222 @@ -0,0 +1,389 @@
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 +Several extra patches are not included in this tree. Only
2336 +aufs3-standalong tree contains them. They are describe in the later
2337 +section "Configuration and Compilation."
2338 +
2339 +On the other hand, the aufs3-standalone tree has only aufs source files
2340 +and necessary patches, and you can select CONFIG_AUFS_FS=m.
2341 +But you need to apply all aufs patches manually.
2342 +
2343 +You will find GIT branches whose name is in form of "aufs3.x" where "x"
2344 +represents the linux kernel version, "linux-3.x". For instance,
2345 +"aufs3.0" is for linux-3.0. For latest "linux-3.x-rcN", use
2346 +"aufs3.x-rcN" branch.
2347 +
2348 +o aufs3-linux tree
2349 +$ git clone --reference /your/linux/git/tree \
2350 +       git://git.code.sf.net/p/aufs/aufs3-linux aufs3-linux.git
2351 +- if you don't have linux GIT tree, then remove "--reference ..."
2352 +$ cd aufs3-linux.git
2353 +$ git checkout origin/aufs3.0
2354 +
2355 +Or You may want to directly git-pull aufs into your linux GIT tree, and
2356 +leave the patch-work to GIT.
2357 +$ cd /your/linux/git/tree
2358 +$ git remote add aufs3 https://github.com/sfjro/aufs3-linux.git
2359 +- aufs3-linux.git tree also exists on github.
2360 +$ git fetch aufs3
2361 +$ git checkout -b my3.14 v3.14
2362 +$ (add your change...)
2363 +$ git pull aufs3 aufs3.14
2364 +- now you have v3.14 + your_changes + aufs3.14 in you my3.14 branch.
2365 +- you may need to solve some conflicts between your_changes and
2366 +  aufs3.14. in this case, git-rerere is recommended so that you can
2367 +  solve the similar confilicts automatically when you upgrade to 3.15 or
2368 +  later in the future.
2369 +
2370 +o aufs3-standalone tree
2371 +$ git clone git://git.code.sf.net/p/aufs/aufs3-standalone \
2372 +       aufs3-standalone.git
2373 +$ cd aufs3-standalone.git
2374 +$ git checkout origin/aufs3.0
2375 +
2376 +o aufs-util tree
2377 +$ git clone git://git.code.sf.net/p/aufs/aufs-util \
2378 +       aufs-util.git
2379 +$ cd aufs-util.git
2380 +$ git checkout origin/aufs3.0
2381 +
2382 +Note: The 3.x-rcN branch is to be used with `rc' kernel versions ONLY.
2383 +The minor version number, 'x' in '3.x', of aufs may not always
2384 +follow the minor version number of the kernel.
2385 +Because changes in the kernel that cause the use of a new
2386 +minor version number do not always require changes to aufs-util.
2387 +
2388 +Since aufs-util has its own minor version number, you may not be
2389 +able to find a GIT branch in aufs-util for your kernel's
2390 +exact minor version number.
2391 +In this case, you should git-checkout the branch for the
2392 +nearest lower number.
2393 +
2394 +For (an unreleased) example:
2395 +If you are using "linux-3.10" and the "aufs3.10" branch
2396 +does not exist in aufs-util repository, then "aufs3.9", "aufs3.8"
2397 +or something numerically smaller is the branch for your kernel.
2398 +
2399 +Also you can view all branches by
2400 +       $ git branch -a
2401 +
2402 +
2403 +3. Configuration and Compilation
2404 +----------------------------------------
2405 +Make sure you have git-checkout'ed the correct branch.
2406 +
2407 +For aufs3-linux tree,
2408 +- enable CONFIG_AUFS_FS.
2409 +- set other aufs configurations if necessary.
2410 +
2411 +For aufs3-standalone tree,
2412 +There are several ways to build.
2413 +
2414 +1.
2415 +- apply ./aufs3-kbuild.patch to your kernel source files.
2416 +- apply ./aufs3-base.patch too.
2417 +- apply ./aufs3-mmap.patch too.
2418 +- apply ./aufs3-standalone.patch too, if you have a plan to set
2419 +  CONFIG_AUFS_FS=m. otherwise you don't need ./aufs3-standalone.patch.
2420 +- copy ./{Documentation,fs,include/uapi/linux/aufs_type.h} files to your
2421 +  kernel source tree. Never copy $PWD/include/uapi/linux/Kbuild.
2422 +- enable CONFIG_AUFS_FS, you can select either
2423 +  =m or =y.
2424 +- and build your kernel as usual.
2425 +- install the built kernel.
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 if you set CONFIG_AUFS_FS=m.
2429 +- install the header files too by "make headers_install" to the
2430 +  directory where you specify. By default, it is $PWD/usr.
2431 +  "make help" shows a brief note for headers_install.
2432 +- and reboot your system.
2433 +
2434 +2.
2435 +- module only (CONFIG_AUFS_FS=m).
2436 +- apply ./aufs3-base.patch to your kernel source files.
2437 +- apply ./aufs3-mmap.patch too.
2438 +- apply ./aufs3-standalone.patch too.
2439 +- build your kernel, don't forget "make headers_install", and reboot.
2440 +- edit ./config.mk and set other aufs configurations if necessary.
2441 +  Note: You should read $PWD/fs/aufs/Kconfig carefully which describes
2442 +  every aufs configurations.
2443 +- build the module by simple "make".
2444 +  Note: Since linux-3.9, every filesystem module requires an alias
2445 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2446 +  modules.aliases file.
2447 +- you can specify ${KDIR} make variable which points to your kernel
2448 +  source tree.
2449 +- install the files
2450 +  + run "make install" to install the aufs module, or copy the built
2451 +    $PWD/aufs.ko to /lib/modules/... and run depmod -a (or reboot simply).
2452 +  + run "make install_headers" (instead of headers_install) to install
2453 +    the modified aufs header file (you can specify DESTDIR which is
2454 +    available in aufs standalone version's Makefile only), or copy
2455 +    $PWD/usr/include/linux/aufs_type.h to /usr/include/linux or wherever
2456 +    you like manually. By default, the target directory is $PWD/usr.
2457 +- no need to apply aufs3-kbuild.patch, nor copying source files to your
2458 +  kernel source tree.
2459 +
2460 +Note: The header file aufs_type.h is necessary to build aufs-util
2461 +      as well as "make headers_install" in the kernel source tree.
2462 +      headers_install is subject to be forgotten, but it is essentially
2463 +      necessary, not only for building aufs-util.
2464 +      You may not meet problems without headers_install in some older
2465 +      version though.
2466 +
2467 +And then,
2468 +- read README in aufs-util, build and install it
2469 +- note that your distribution may contain an obsoleted version of
2470 +  aufs_type.h in /usr/include/linux or something. When you build aufs
2471 +  utilities, make sure that your compiler refers the correct aufs header
2472 +  file which is built by "make headers_install."
2473 +- if you want to use readdir(3) in userspace or pathconf(3) wrapper,
2474 +  then run "make install_ulib" too. And refer to the aufs manual in
2475 +  detail.
2476 +
2477 +There several other patches in aufs3-standalone.git. They are all
2478 +optional. When you meet some problems, they will help you.
2479 +- aufs3-loopback.patch
2480 +  Supports a nested loopback mount in a branch-fs. This patch is
2481 +  unnecessary until aufs produces a message like "you may want to try
2482 +  another patch for loopback file".
2483 +- vfs-ino.patch
2484 +  Modifies a system global kernel internal function get_next_ino() in
2485 +  order to stop assigning 0 for an inode-number. Not directly related to
2486 +  aufs, but recommended generally.
2487 +- tmpfs-idr.patch
2488 +  Keeps the tmpfs inode number as the lowest value. Effective to reduce
2489 +  the size of aufs XINO files for tmpfs branch. Also it prevents the
2490 +  duplication of inode number, which is important for backup tools and
2491 +  other utilities. When you find aufs XINO files for tmpfs branch
2492 +  growing too much, try this patch.
2493 +
2494 +
2495 +4. Usage
2496 +----------------------------------------
2497 +At first, make sure aufs-util are installed, and please read the aufs
2498 +manual, aufs.5 in aufs-util.git tree.
2499 +$ man -l aufs.5
2500 +
2501 +And then,
2502 +$ mkdir /tmp/rw /tmp/aufs
2503 +# mount -t aufs -o br=/tmp/rw:${HOME} none /tmp/aufs
2504 +
2505 +Here is another example. The result is equivalent.
2506 +# mount -t aufs -o br=/tmp/rw=rw:${HOME}=ro none /tmp/aufs
2507 +  Or
2508 +# mount -t aufs -o br:/tmp/rw none /tmp/aufs
2509 +# mount -o remount,append:${HOME} /tmp/aufs
2510 +
2511 +Then, you can see whole tree of your home dir through /tmp/aufs. If
2512 +you modify a file under /tmp/aufs, the one on your home directory is
2513 +not affected, instead the same named file will be newly created under
2514 +/tmp/rw. And all of your modification to a file will be applied to
2515 +the one under /tmp/rw. This is called the file based Copy on Write
2516 +(COW) method.
2517 +Aufs mount options are described in aufs.5.
2518 +If you run chroot or something and make your aufs as a root directory,
2519 +then you need to customize the shutdown script. See the aufs manual in
2520 +detail.
2521 +
2522 +Additionally, there are some sample usages of aufs which are a
2523 +diskless system with network booting, and LiveCD over NFS.
2524 +See sample dir in CVS tree on SourceForge.
2525 +
2526 +
2527 +5. Contact
2528 +----------------------------------------
2529 +When you have any problems or strange behaviour in aufs, please let me
2530 +know with:
2531 +- /proc/mounts (instead of the output of mount(8))
2532 +- /sys/module/aufs/*
2533 +- /sys/fs/aufs/* (if you have them)
2534 +- /debug/aufs/* (if you have them)
2535 +- linux kernel version
2536 +  if your kernel is not plain, for example modified by distributor,
2537 +  the url where i can download its source is necessary too.
2538 +- aufs version which was printed at loading the module or booting the
2539 +  system, instead of the date you downloaded.
2540 +- configuration (define/undefine CONFIG_AUFS_xxx)
2541 +- kernel configuration or /proc/config.gz (if you have it)
2542 +- behaviour which you think to be incorrect
2543 +- actual operation, reproducible one is better
2544 +- mailto: aufs-users at lists.sourceforge.net
2545 +
2546 +Usually, I don't watch the Public Areas(Bugs, Support Requests, Patches,
2547 +and Feature Requests) on SourceForge. Please join and write to
2548 +aufs-users ML.
2549 +
2550 +
2551 +6. Acknowledgements
2552 +----------------------------------------
2553 +Thanks to everyone who have tried and are using aufs, whoever
2554 +have reported a bug or any feedback.
2555 +
2556 +Especially donators:
2557 +Tomas Matejicek(slax.org) made a donation (much more than once).
2558 +       Since Apr 2010, Tomas M (the author of Slax and Linux Live
2559 +       scripts) is making "doubling" donations.
2560 +       Unfortunately I cannot list all of the donators, but I really
2561 +       appreciate.
2562 +       It ends Aug 2010, but the ordinary donation URL is still available.
2563 +       <http://sourceforge.net/donate/index.php?group_id=167503>
2564 +Dai Itasaka made a donation (2007/8).
2565 +Chuck Smith made a donation (2008/4, 10 and 12).
2566 +Henk Schoneveld made a donation (2008/9).
2567 +Chih-Wei Huang, ASUS, CTC donated Eee PC 4G (2008/10).
2568 +Francois Dupoux made a donation (2008/11).
2569 +Bruno Cesar Ribas and Luis Carlos Erpen de Bona, C3SL serves public
2570 +       aufs2 GIT tree (2009/2).
2571 +William Grant made a donation (2009/3).
2572 +Patrick Lane made a donation (2009/4).
2573 +The Mail Archive (mail-archive.com) made donations (2009/5).
2574 +Nippy Networks (Ed Wildgoose) made a donation (2009/7).
2575 +New Dream Network, LLC (www.dreamhost.com) made a donation (2009/11).
2576 +Pavel Pronskiy made a donation (2011/2).
2577 +Iridium and Inmarsat satellite phone retailer (www.mailasail.com), Nippy
2578 +       Networks (Ed Wildgoose) made a donation for hardware (2011/3).
2579 +Max Lekomcev (DOM-TV project) made a donation (2011/7, 12, 2012/3, 6 and
2580 +11).
2581 +Sam Liddicott made a donation (2011/9).
2582 +Era Scarecrow made a donation (2013/4).
2583 +Bor Ratajc made a donation (2013/4).
2584 +Alessandro Gorreta made a donation (2013/4).
2585 +POIRETTE Marc made a donation (2013/4).
2586 +Alessandro Gorreta made a donation (2013/4).
2587 +lauri kasvandik made a donation (2013/5).
2588 +"pemasu from Finland" made a donation (2013/7).
2589 +The Parted Magic Project made a donation (2013/9 and 11).
2590 +Pavel Barta made a donation (2013/10).
2591 +Nikolay Pertsev made a donation (2014/5).
2592 +James B made a donation (2014/7).
2593 +Stefano Di Biase made a donation (2014/8).
2594 +Daniel Epellei made a donation (2015/1).
2595 +
2596 +Thank you very much.
2597 +Donations are always, including future donations, very important and
2598 +helpful for me to keep on developing aufs.
2599 +
2600 +
2601 +7.
2602 +----------------------------------------
2603 +If you are an experienced user, no explanation is needed. Aufs is
2604 +just a linux filesystem.
2605 +
2606 +
2607 +Enjoy!
2608 +
2609 +# Local variables: ;
2610 +# mode: text;
2611 +# End: ;
2612 diff -urN /usr/share/empty/fs/aufs/aufs.h linux/fs/aufs/aufs.h
2613 --- /usr/share/empty/fs/aufs/aufs.h     1970-01-01 01:00:00.000000000 +0100
2614 +++ linux/fs/aufs/aufs.h        2015-03-27 21:56:35.460128334 +0100
2615 @@ -0,0 +1,59 @@
2616 +/*
2617 + * Copyright (C) 2005-2015 Junjiro R. Okajima
2618 + *
2619 + * This program, aufs is free software; you can redistribute it and/or modify
2620 + * it under the terms of the GNU General Public License as published by
2621 + * the Free Software Foundation; either version 2 of the License, or
2622 + * (at your option) any later version.
2623 + *
2624 + * This program is distributed in the hope that it will be useful,
2625 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2626 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2627 + * GNU General Public License for more details.
2628 + *
2629 + * You should have received a copy of the GNU General Public License
2630 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2631 + */
2632 +
2633 +/*
2634 + * all header files
2635 + */
2636 +
2637 +#ifndef __AUFS_H__
2638 +#define __AUFS_H__
2639 +
2640 +#ifdef __KERNEL__
2641 +
2642 +#define AuStub(type, name, body, ...) \
2643 +       static inline type name(__VA_ARGS__) { body; }
2644 +
2645 +#define AuStubVoid(name, ...) \
2646 +       AuStub(void, name, , __VA_ARGS__)
2647 +#define AuStubInt0(name, ...) \
2648 +       AuStub(int, name, return 0, __VA_ARGS__)
2649 +
2650 +#include "debug.h"
2651 +
2652 +#include "branch.h"
2653 +#include "cpup.h"
2654 +#include "dcsub.h"
2655 +#include "dbgaufs.h"
2656 +#include "dentry.h"
2657 +#include "dir.h"
2658 +#include "dynop.h"
2659 +#include "file.h"
2660 +#include "fstype.h"
2661 +#include "inode.h"
2662 +#include "loop.h"
2663 +#include "module.h"
2664 +#include "opts.h"
2665 +#include "rwsem.h"
2666 +#include "spl.h"
2667 +#include "super.h"
2668 +#include "sysaufs.h"
2669 +#include "vfsub.h"
2670 +#include "whout.h"
2671 +#include "wkq.h"
2672 +
2673 +#endif /* __KERNEL__ */
2674 +#endif /* __AUFS_H__ */
2675 diff -urN /usr/share/empty/fs/aufs/branch.c linux/fs/aufs/branch.c
2676 --- /usr/share/empty/fs/aufs/branch.c   1970-01-01 01:00:00.000000000 +0100
2677 +++ linux/fs/aufs/branch.c      2015-03-27 21:56:35.460128334 +0100
2678 @@ -0,0 +1,1408 @@
2679 +/*
2680 + * Copyright (C) 2005-2015 Junjiro R. Okajima
2681 + *
2682 + * This program, aufs is free software; you can redistribute it and/or modify
2683 + * it under the terms of the GNU General Public License as published by
2684 + * the Free Software Foundation; either version 2 of the License, or
2685 + * (at your option) any later version.
2686 + *
2687 + * This program is distributed in the hope that it will be useful,
2688 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2689 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2690 + * GNU General Public License for more details.
2691 + *
2692 + * You should have received a copy of the GNU General Public License
2693 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2694 + */
2695 +
2696 +/*
2697 + * branch management
2698 + */
2699 +
2700 +#include <linux/compat.h>
2701 +#include <linux/statfs.h>
2702 +#include "aufs.h"
2703 +
2704 +/*
2705 + * free a single branch
2706 + */
2707 +static void au_br_do_free(struct au_branch *br)
2708 +{
2709 +       int i;
2710 +       struct au_wbr *wbr;
2711 +       struct au_dykey **key;
2712 +
2713 +       au_hnotify_fin_br(br);
2714 +
2715 +       if (br->br_xino.xi_file)
2716 +               fput(br->br_xino.xi_file);
2717 +       mutex_destroy(&br->br_xino.xi_nondir_mtx);
2718 +
2719 +       AuDebugOn(atomic_read(&br->br_count));
2720 +
2721 +       wbr = br->br_wbr;
2722 +       if (wbr) {
2723 +               for (i = 0; i < AuBrWh_Last; i++)
2724 +                       dput(wbr->wbr_wh[i]);
2725 +               AuDebugOn(atomic_read(&wbr->wbr_wh_running));
2726 +               AuRwDestroy(&wbr->wbr_wh_rwsem);
2727 +       }
2728 +
2729 +       if (br->br_fhsm) {
2730 +               au_br_fhsm_fin(br->br_fhsm);
2731 +               kfree(br->br_fhsm);
2732 +       }
2733 +
2734 +       key = br->br_dykey;
2735 +       for (i = 0; i < AuBrDynOp; i++, key++)
2736 +               if (*key)
2737 +                       au_dy_put(*key);
2738 +               else
2739 +                       break;
2740 +
2741 +       /* recursive lock, s_umount of branch's */
2742 +       lockdep_off();
2743 +       path_put(&br->br_path);
2744 +       lockdep_on();
2745 +       kfree(wbr);
2746 +       kfree(br);
2747 +}
2748 +
2749 +/*
2750 + * frees all branches
2751 + */
2752 +void au_br_free(struct au_sbinfo *sbinfo)
2753 +{
2754 +       aufs_bindex_t bmax;
2755 +       struct au_branch **br;
2756 +
2757 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
2758 +
2759 +       bmax = sbinfo->si_bend + 1;
2760 +       br = sbinfo->si_branch;
2761 +       while (bmax--)
2762 +               au_br_do_free(*br++);
2763 +}
2764 +
2765 +/*
2766 + * find the index of a branch which is specified by @br_id.
2767 + */
2768 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id)
2769 +{
2770 +       aufs_bindex_t bindex, bend;
2771 +
2772 +       bend = au_sbend(sb);
2773 +       for (bindex = 0; bindex <= bend; bindex++)
2774 +               if (au_sbr_id(sb, bindex) == br_id)
2775 +                       return bindex;
2776 +       return -1;
2777 +}
2778 +
2779 +/* ---------------------------------------------------------------------- */
2780 +
2781 +/*
2782 + * add a branch
2783 + */
2784 +
2785 +static int test_overlap(struct super_block *sb, struct dentry *h_adding,
2786 +                       struct dentry *h_root)
2787 +{
2788 +       if (unlikely(h_adding == h_root
2789 +                    || au_test_loopback_overlap(sb, h_adding)))
2790 +               return 1;
2791 +       if (h_adding->d_sb != h_root->d_sb)
2792 +               return 0;
2793 +       return au_test_subdir(h_adding, h_root)
2794 +               || au_test_subdir(h_root, h_adding);
2795 +}
2796 +
2797 +/*
2798 + * returns a newly allocated branch. @new_nbranch is a number of branches
2799 + * after adding a branch.
2800 + */
2801 +static struct au_branch *au_br_alloc(struct super_block *sb, int new_nbranch,
2802 +                                    int perm)
2803 +{
2804 +       struct au_branch *add_branch;
2805 +       struct dentry *root;
2806 +       int err;
2807 +
2808 +       err = -ENOMEM;
2809 +       root = sb->s_root;
2810 +       add_branch = kmalloc(sizeof(*add_branch), GFP_NOFS);
2811 +       if (unlikely(!add_branch))
2812 +               goto out;
2813 +
2814 +       err = au_hnotify_init_br(add_branch, perm);
2815 +       if (unlikely(err))
2816 +               goto out_br;
2817 +
2818 +       add_branch->br_wbr = NULL;
2819 +       if (au_br_writable(perm)) {
2820 +               /* may be freed separately at changing the branch permission */
2821 +               add_branch->br_wbr = kmalloc(sizeof(*add_branch->br_wbr),
2822 +                                            GFP_NOFS);
2823 +               if (unlikely(!add_branch->br_wbr))
2824 +                       goto out_hnotify;
2825 +       }
2826 +
2827 +       add_branch->br_fhsm = NULL;
2828 +       if (au_br_fhsm(perm)) {
2829 +               err = au_fhsm_br_alloc(add_branch);
2830 +               if (unlikely(err))
2831 +                       goto out_wbr;
2832 +       }
2833 +
2834 +       err = au_sbr_realloc(au_sbi(sb), new_nbranch);
2835 +       if (!err)
2836 +               err = au_di_realloc(au_di(root), new_nbranch);
2837 +       if (!err)
2838 +               err = au_ii_realloc(au_ii(root->d_inode), new_nbranch);
2839 +       if (!err)
2840 +               return add_branch; /* success */
2841 +
2842 +out_wbr:
2843 +       kfree(add_branch->br_wbr);
2844 +out_hnotify:
2845 +       au_hnotify_fin_br(add_branch);
2846 +out_br:
2847 +       kfree(add_branch);
2848 +out:
2849 +       return ERR_PTR(err);
2850 +}
2851 +
2852 +/*
2853 + * test if the branch permission is legal or not.
2854 + */
2855 +static int test_br(struct inode *inode, int brperm, char *path)
2856 +{
2857 +       int err;
2858 +
2859 +       err = (au_br_writable(brperm) && IS_RDONLY(inode));
2860 +       if (!err)
2861 +               goto out;
2862 +
2863 +       err = -EINVAL;
2864 +       pr_err("write permission for readonly mount or inode, %s\n", path);
2865 +
2866 +out:
2867 +       return err;
2868 +}
2869 +
2870 +/*
2871 + * returns:
2872 + * 0: success, the caller will add it
2873 + * plus: success, it is already unified, the caller should ignore it
2874 + * minus: error
2875 + */
2876 +static int test_add(struct super_block *sb, struct au_opt_add *add, int remount)
2877 +{
2878 +       int err;
2879 +       aufs_bindex_t bend, bindex;
2880 +       struct dentry *root;
2881 +       struct inode *inode, *h_inode;
2882 +
2883 +       root = sb->s_root;
2884 +       bend = au_sbend(sb);
2885 +       if (unlikely(bend >= 0
2886 +                    && au_find_dbindex(root, add->path.dentry) >= 0)) {
2887 +               err = 1;
2888 +               if (!remount) {
2889 +                       err = -EINVAL;
2890 +                       pr_err("%s duplicated\n", add->pathname);
2891 +               }
2892 +               goto out;
2893 +       }
2894 +
2895 +       err = -ENOSPC; /* -E2BIG; */
2896 +       if (unlikely(AUFS_BRANCH_MAX <= add->bindex
2897 +                    || AUFS_BRANCH_MAX - 1 <= bend)) {
2898 +               pr_err("number of branches exceeded %s\n", add->pathname);
2899 +               goto out;
2900 +       }
2901 +
2902 +       err = -EDOM;
2903 +       if (unlikely(add->bindex < 0 || bend + 1 < add->bindex)) {
2904 +               pr_err("bad index %d\n", add->bindex);
2905 +               goto out;
2906 +       }
2907 +
2908 +       inode = add->path.dentry->d_inode;
2909 +       err = -ENOENT;
2910 +       if (unlikely(!inode->i_nlink)) {
2911 +               pr_err("no existence %s\n", add->pathname);
2912 +               goto out;
2913 +       }
2914 +
2915 +       err = -EINVAL;
2916 +       if (unlikely(inode->i_sb == sb)) {
2917 +               pr_err("%s must be outside\n", add->pathname);
2918 +               goto out;
2919 +       }
2920 +
2921 +       if (unlikely(au_test_fs_unsuppoted(inode->i_sb))) {
2922 +               pr_err("unsupported filesystem, %s (%s)\n",
2923 +                      add->pathname, au_sbtype(inode->i_sb));
2924 +               goto out;
2925 +       }
2926 +
2927 +       if (unlikely(inode->i_sb->s_stack_depth)) {
2928 +               pr_err("already stacked, %s (%s)\n",
2929 +                      add->pathname, au_sbtype(inode->i_sb));
2930 +               goto out;
2931 +       }
2932 +
2933 +       err = test_br(add->path.dentry->d_inode, add->perm, add->pathname);
2934 +       if (unlikely(err))
2935 +               goto out;
2936 +
2937 +       if (bend < 0)
2938 +               return 0; /* success */
2939 +
2940 +       err = -EINVAL;
2941 +       for (bindex = 0; bindex <= bend; bindex++)
2942 +               if (unlikely(test_overlap(sb, add->path.dentry,
2943 +                                         au_h_dptr(root, bindex)))) {
2944 +                       pr_err("%s is overlapped\n", add->pathname);
2945 +                       goto out;
2946 +               }
2947 +
2948 +       err = 0;
2949 +       if (au_opt_test(au_mntflags(sb), WARN_PERM)) {
2950 +               h_inode = au_h_dptr(root, 0)->d_inode;
2951 +               if ((h_inode->i_mode & S_IALLUGO) != (inode->i_mode & S_IALLUGO)
2952 +                   || !uid_eq(h_inode->i_uid, inode->i_uid)
2953 +                   || !gid_eq(h_inode->i_gid, inode->i_gid))
2954 +                       pr_warn("uid/gid/perm %s %u/%u/0%o, %u/%u/0%o\n",
2955 +                               add->pathname,
2956 +                               i_uid_read(inode), i_gid_read(inode),
2957 +                               (inode->i_mode & S_IALLUGO),
2958 +                               i_uid_read(h_inode), i_gid_read(h_inode),
2959 +                               (h_inode->i_mode & S_IALLUGO));
2960 +       }
2961 +
2962 +out:
2963 +       return err;
2964 +}
2965 +
2966 +/*
2967 + * initialize or clean the whiteouts for an adding branch
2968 + */
2969 +static int au_br_init_wh(struct super_block *sb, struct au_branch *br,
2970 +                        int new_perm)
2971 +{
2972 +       int err, old_perm;
2973 +       aufs_bindex_t bindex;
2974 +       struct mutex *h_mtx;
2975 +       struct au_wbr *wbr;
2976 +       struct au_hinode *hdir;
2977 +
2978 +       err = vfsub_mnt_want_write(au_br_mnt(br));
2979 +       if (unlikely(err))
2980 +               goto out;
2981 +
2982 +       wbr = br->br_wbr;
2983 +       old_perm = br->br_perm;
2984 +       br->br_perm = new_perm;
2985 +       hdir = NULL;
2986 +       h_mtx = NULL;
2987 +       bindex = au_br_index(sb, br->br_id);
2988 +       if (0 <= bindex) {
2989 +               hdir = au_hi(sb->s_root->d_inode, bindex);
2990 +               au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
2991 +       } else {
2992 +               h_mtx = &au_br_dentry(br)->d_inode->i_mutex;
2993 +               mutex_lock_nested(h_mtx, AuLsc_I_PARENT);
2994 +       }
2995 +       if (!wbr)
2996 +               err = au_wh_init(br, sb);
2997 +       else {
2998 +               wbr_wh_write_lock(wbr);
2999 +               err = au_wh_init(br, sb);
3000 +               wbr_wh_write_unlock(wbr);
3001 +       }
3002 +       if (hdir)
3003 +               au_hn_imtx_unlock(hdir);
3004 +       else
3005 +               mutex_unlock(h_mtx);
3006 +       vfsub_mnt_drop_write(au_br_mnt(br));
3007 +       br->br_perm = old_perm;
3008 +
3009 +       if (!err && wbr && !au_br_writable(new_perm)) {
3010 +               kfree(wbr);
3011 +               br->br_wbr = NULL;
3012 +       }
3013 +
3014 +out:
3015 +       return err;
3016 +}
3017 +
3018 +static int au_wbr_init(struct au_branch *br, struct super_block *sb,
3019 +                      int perm)
3020 +{
3021 +       int err;
3022 +       struct kstatfs kst;
3023 +       struct au_wbr *wbr;
3024 +
3025 +       wbr = br->br_wbr;
3026 +       au_rw_init(&wbr->wbr_wh_rwsem);
3027 +       memset(wbr->wbr_wh, 0, sizeof(wbr->wbr_wh));
3028 +       atomic_set(&wbr->wbr_wh_running, 0);
3029 +       wbr->wbr_bytes = 0;
3030 +
3031 +       /*
3032 +        * a limit for rmdir/rename a dir
3033 +        * cf. AUFS_MAX_NAMELEN in include/uapi/linux/aufs_type.h
3034 +        */
3035 +       err = vfs_statfs(&br->br_path, &kst);
3036 +       if (unlikely(err))
3037 +               goto out;
3038 +       err = -EINVAL;
3039 +       if (kst.f_namelen >= NAME_MAX)
3040 +               err = au_br_init_wh(sb, br, perm);
3041 +       else
3042 +               pr_err("%pd(%s), unsupported namelen %ld\n",
3043 +                      au_br_dentry(br),
3044 +                      au_sbtype(au_br_dentry(br)->d_sb), kst.f_namelen);
3045 +
3046 +out:
3047 +       return err;
3048 +}
3049 +
3050 +/* initialize a new branch */
3051 +static int au_br_init(struct au_branch *br, struct super_block *sb,
3052 +                     struct au_opt_add *add)
3053 +{
3054 +       int err;
3055 +
3056 +       err = 0;
3057 +       memset(&br->br_xino, 0, sizeof(br->br_xino));
3058 +       mutex_init(&br->br_xino.xi_nondir_mtx);
3059 +       br->br_perm = add->perm;
3060 +       br->br_path = add->path; /* set first, path_get() later */
3061 +       spin_lock_init(&br->br_dykey_lock);
3062 +       memset(br->br_dykey, 0, sizeof(br->br_dykey));
3063 +       atomic_set(&br->br_count, 0);
3064 +       atomic_set(&br->br_xino_running, 0);
3065 +       br->br_id = au_new_br_id(sb);
3066 +       AuDebugOn(br->br_id < 0);
3067 +
3068 +       if (au_br_writable(add->perm)) {
3069 +               err = au_wbr_init(br, sb, add->perm);
3070 +               if (unlikely(err))
3071 +                       goto out_err;
3072 +       }
3073 +
3074 +       if (au_opt_test(au_mntflags(sb), XINO)) {
3075 +               err = au_xino_br(sb, br, add->path.dentry->d_inode->i_ino,
3076 +                                au_sbr(sb, 0)->br_xino.xi_file, /*do_test*/1);
3077 +               if (unlikely(err)) {
3078 +                       AuDebugOn(br->br_xino.xi_file);
3079 +                       goto out_err;
3080 +               }
3081 +       }
3082 +
3083 +       sysaufs_br_init(br);
3084 +       path_get(&br->br_path);
3085 +       goto out; /* success */
3086 +
3087 +out_err:
3088 +       memset(&br->br_path, 0, sizeof(br->br_path));
3089 +out:
3090 +       return err;
3091 +}
3092 +
3093 +static void au_br_do_add_brp(struct au_sbinfo *sbinfo, aufs_bindex_t bindex,
3094 +                            struct au_branch *br, aufs_bindex_t bend,
3095 +                            aufs_bindex_t amount)
3096 +{
3097 +       struct au_branch **brp;
3098 +
3099 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3100 +
3101 +       brp = sbinfo->si_branch + bindex;
3102 +       memmove(brp + 1, brp, sizeof(*brp) * amount);
3103 +       *brp = br;
3104 +       sbinfo->si_bend++;
3105 +       if (unlikely(bend < 0))
3106 +               sbinfo->si_bend = 0;
3107 +}
3108 +
3109 +static void au_br_do_add_hdp(struct au_dinfo *dinfo, aufs_bindex_t bindex,
3110 +                            aufs_bindex_t bend, aufs_bindex_t amount)
3111 +{
3112 +       struct au_hdentry *hdp;
3113 +
3114 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3115 +
3116 +       hdp = dinfo->di_hdentry + bindex;
3117 +       memmove(hdp + 1, hdp, sizeof(*hdp) * amount);
3118 +       au_h_dentry_init(hdp);
3119 +       dinfo->di_bend++;
3120 +       if (unlikely(bend < 0))
3121 +               dinfo->di_bstart = 0;
3122 +}
3123 +
3124 +static void au_br_do_add_hip(struct au_iinfo *iinfo, aufs_bindex_t bindex,
3125 +                            aufs_bindex_t bend, aufs_bindex_t amount)
3126 +{
3127 +       struct au_hinode *hip;
3128 +
3129 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3130 +
3131 +       hip = iinfo->ii_hinode + bindex;
3132 +       memmove(hip + 1, hip, sizeof(*hip) * amount);
3133 +       hip->hi_inode = NULL;
3134 +       au_hn_init(hip);
3135 +       iinfo->ii_bend++;
3136 +       if (unlikely(bend < 0))
3137 +               iinfo->ii_bstart = 0;
3138 +}
3139 +
3140 +static void au_br_do_add(struct super_block *sb, struct au_branch *br,
3141 +                        aufs_bindex_t bindex)
3142 +{
3143 +       struct dentry *root, *h_dentry;
3144 +       struct inode *root_inode;
3145 +       aufs_bindex_t bend, amount;
3146 +
3147 +       root = sb->s_root;
3148 +       root_inode = root->d_inode;
3149 +       bend = au_sbend(sb);
3150 +       amount = bend + 1 - bindex;
3151 +       h_dentry = au_br_dentry(br);
3152 +       au_sbilist_lock();
3153 +       au_br_do_add_brp(au_sbi(sb), bindex, br, bend, amount);
3154 +       au_br_do_add_hdp(au_di(root), bindex, bend, amount);
3155 +       au_br_do_add_hip(au_ii(root_inode), bindex, bend, amount);
3156 +       au_set_h_dptr(root, bindex, dget(h_dentry));
3157 +       au_set_h_iptr(root_inode, bindex, au_igrab(h_dentry->d_inode),
3158 +                     /*flags*/0);
3159 +       au_sbilist_unlock();
3160 +}
3161 +
3162 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount)
3163 +{
3164 +       int err;
3165 +       aufs_bindex_t bend, add_bindex;
3166 +       struct dentry *root, *h_dentry;
3167 +       struct inode *root_inode;
3168 +       struct au_branch *add_branch;
3169 +
3170 +       root = sb->s_root;
3171 +       root_inode = root->d_inode;
3172 +       IMustLock(root_inode);
3173 +       err = test_add(sb, add, remount);
3174 +       if (unlikely(err < 0))
3175 +               goto out;
3176 +       if (err) {
3177 +               err = 0;
3178 +               goto out; /* success */
3179 +       }
3180 +
3181 +       bend = au_sbend(sb);
3182 +       add_branch = au_br_alloc(sb, bend + 2, add->perm);
3183 +       err = PTR_ERR(add_branch);
3184 +       if (IS_ERR(add_branch))
3185 +               goto out;
3186 +
3187 +       err = au_br_init(add_branch, sb, add);
3188 +       if (unlikely(err)) {
3189 +               au_br_do_free(add_branch);
3190 +               goto out;
3191 +       }
3192 +
3193 +       add_bindex = add->bindex;
3194 +       if (!remount)
3195 +               au_br_do_add(sb, add_branch, add_bindex);
3196 +       else {
3197 +               sysaufs_brs_del(sb, add_bindex);
3198 +               au_br_do_add(sb, add_branch, add_bindex);
3199 +               sysaufs_brs_add(sb, add_bindex);
3200 +       }
3201 +
3202 +       h_dentry = add->path.dentry;
3203 +       if (!add_bindex) {
3204 +               au_cpup_attr_all(root_inode, /*force*/1);
3205 +               sb->s_maxbytes = h_dentry->d_sb->s_maxbytes;
3206 +       } else
3207 +               au_add_nlink(root_inode, h_dentry->d_inode);
3208 +
3209 +       /*
3210 +        * this test/set prevents aufs from handling unnecesary notify events
3211 +        * of xino files, in case of re-adding a writable branch which was
3212 +        * once detached from aufs.
3213 +        */
3214 +       if (au_xino_brid(sb) < 0
3215 +           && au_br_writable(add_branch->br_perm)
3216 +           && !au_test_fs_bad_xino(h_dentry->d_sb)
3217 +           && add_branch->br_xino.xi_file
3218 +           && add_branch->br_xino.xi_file->f_path.dentry->d_parent == h_dentry)
3219 +               au_xino_brid_set(sb, add_branch->br_id);
3220 +
3221 +out:
3222 +       return err;
3223 +}
3224 +
3225 +/* ---------------------------------------------------------------------- */
3226 +
3227 +static unsigned long long au_farray_cb(void *a,
3228 +                                      unsigned long long max __maybe_unused,
3229 +                                      void *arg)
3230 +{
3231 +       unsigned long long n;
3232 +       struct file **p, *f;
3233 +       struct au_sphlhead *files;
3234 +       struct au_finfo *finfo;
3235 +       struct super_block *sb = arg;
3236 +
3237 +       n = 0;
3238 +       p = a;
3239 +       files = &au_sbi(sb)->si_files;
3240 +       spin_lock(&files->spin);
3241 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
3242 +               f = finfo->fi_file;
3243 +               if (file_count(f)
3244 +                   && !special_file(file_inode(f)->i_mode)) {
3245 +                       get_file(f);
3246 +                       *p++ = f;
3247 +                       n++;
3248 +                       AuDebugOn(n > max);
3249 +               }
3250 +       }
3251 +       spin_unlock(&files->spin);
3252 +
3253 +       return n;
3254 +}
3255 +
3256 +static struct file **au_farray_alloc(struct super_block *sb,
3257 +                                    unsigned long long *max)
3258 +{
3259 +       *max = atomic_long_read(&au_sbi(sb)->si_nfiles);
3260 +       return au_array_alloc(max, au_farray_cb, sb);
3261 +}
3262 +
3263 +static void au_farray_free(struct file **a, unsigned long long max)
3264 +{
3265 +       unsigned long long ull;
3266 +
3267 +       for (ull = 0; ull < max; ull++)
3268 +               if (a[ull])
3269 +                       fput(a[ull]);
3270 +       au_array_free(a);
3271 +}
3272 +
3273 +/* ---------------------------------------------------------------------- */
3274 +
3275 +/*
3276 + * delete a branch
3277 + */
3278 +
3279 +/* to show the line number, do not make it inlined function */
3280 +#define AuVerbose(do_info, fmt, ...) do { \
3281 +       if (do_info) \
3282 +               pr_info(fmt, ##__VA_ARGS__); \
3283 +} while (0)
3284 +
3285 +static int au_test_ibusy(struct inode *inode, aufs_bindex_t bstart,
3286 +                        aufs_bindex_t bend)
3287 +{
3288 +       return (inode && !S_ISDIR(inode->i_mode)) || bstart == bend;
3289 +}
3290 +
3291 +static int au_test_dbusy(struct dentry *dentry, aufs_bindex_t bstart,
3292 +                        aufs_bindex_t bend)
3293 +{
3294 +       return au_test_ibusy(dentry->d_inode, bstart, bend);
3295 +}
3296 +
3297 +/*
3298 + * test if the branch is deletable or not.
3299 + */
3300 +static int test_dentry_busy(struct dentry *root, aufs_bindex_t bindex,
3301 +                           unsigned int sigen, const unsigned int verbose)
3302 +{
3303 +       int err, i, j, ndentry;
3304 +       aufs_bindex_t bstart, bend;
3305 +       struct au_dcsub_pages dpages;
3306 +       struct au_dpage *dpage;
3307 +       struct dentry *d;
3308 +
3309 +       err = au_dpages_init(&dpages, GFP_NOFS);
3310 +       if (unlikely(err))
3311 +               goto out;
3312 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
3313 +       if (unlikely(err))
3314 +               goto out_dpages;
3315 +
3316 +       for (i = 0; !err && i < dpages.ndpage; i++) {
3317 +               dpage = dpages.dpages + i;
3318 +               ndentry = dpage->ndentry;
3319 +               for (j = 0; !err && j < ndentry; j++) {
3320 +                       d = dpage->dentries[j];
3321 +                       AuDebugOn(au_dcount(d) <= 0);
3322 +                       if (!au_digen_test(d, sigen)) {
3323 +                               di_read_lock_child(d, AuLock_IR);
3324 +                               if (unlikely(au_dbrange_test(d))) {
3325 +                                       di_read_unlock(d, AuLock_IR);
3326 +                                       continue;
3327 +                               }
3328 +                       } else {
3329 +                               di_write_lock_child(d);
3330 +                               if (unlikely(au_dbrange_test(d))) {
3331 +                                       di_write_unlock(d);
3332 +                                       continue;
3333 +                               }
3334 +                               err = au_reval_dpath(d, sigen);
3335 +                               if (!err)
3336 +                                       di_downgrade_lock(d, AuLock_IR);
3337 +                               else {
3338 +                                       di_write_unlock(d);
3339 +                                       break;
3340 +                               }
3341 +                       }
3342 +
3343 +                       /* AuDbgDentry(d); */
3344 +                       bstart = au_dbstart(d);
3345 +                       bend = au_dbend(d);
3346 +                       if (bstart <= bindex
3347 +                           && bindex <= bend
3348 +                           && au_h_dptr(d, bindex)
3349 +                           && au_test_dbusy(d, bstart, bend)) {
3350 +                               err = -EBUSY;
3351 +                               AuVerbose(verbose, "busy %pd\n", d);
3352 +                               AuDbgDentry(d);
3353 +                       }
3354 +                       di_read_unlock(d, AuLock_IR);
3355 +               }
3356 +       }
3357 +
3358 +out_dpages:
3359 +       au_dpages_free(&dpages);
3360 +out:
3361 +       return err;
3362 +}
3363 +
3364 +static int test_inode_busy(struct super_block *sb, aufs_bindex_t bindex,
3365 +                          unsigned int sigen, const unsigned int verbose)
3366 +{
3367 +       int err;
3368 +       unsigned long long max, ull;
3369 +       struct inode *i, **array;
3370 +       aufs_bindex_t bstart, bend;
3371 +
3372 +       array = au_iarray_alloc(sb, &max);
3373 +       err = PTR_ERR(array);
3374 +       if (IS_ERR(array))
3375 +               goto out;
3376 +
3377 +       err = 0;
3378 +       AuDbg("b%d\n", bindex);
3379 +       for (ull = 0; !err && ull < max; ull++) {
3380 +               i = array[ull];
3381 +               if (unlikely(!i))
3382 +                       break;
3383 +               if (i->i_ino == AUFS_ROOT_INO)
3384 +                       continue;
3385 +
3386 +               /* AuDbgInode(i); */
3387 +               if (au_iigen(i, NULL) == sigen)
3388 +                       ii_read_lock_child(i);
3389 +               else {
3390 +                       ii_write_lock_child(i);
3391 +                       err = au_refresh_hinode_self(i);
3392 +                       au_iigen_dec(i);
3393 +                       if (!err)
3394 +                               ii_downgrade_lock(i);
3395 +                       else {
3396 +                               ii_write_unlock(i);
3397 +                               break;
3398 +                       }
3399 +               }
3400 +
3401 +               bstart = au_ibstart(i);
3402 +               bend = au_ibend(i);
3403 +               if (bstart <= bindex
3404 +                   && bindex <= bend
3405 +                   && au_h_iptr(i, bindex)
3406 +                   && au_test_ibusy(i, bstart, bend)) {
3407 +                       err = -EBUSY;
3408 +                       AuVerbose(verbose, "busy i%lu\n", i->i_ino);
3409 +                       AuDbgInode(i);
3410 +               }
3411 +               ii_read_unlock(i);
3412 +       }
3413 +       au_iarray_free(array, max);
3414 +
3415 +out:
3416 +       return err;
3417 +}
3418 +
3419 +static int test_children_busy(struct dentry *root, aufs_bindex_t bindex,
3420 +                             const unsigned int verbose)
3421 +{
3422 +       int err;
3423 +       unsigned int sigen;
3424 +
3425 +       sigen = au_sigen(root->d_sb);
3426 +       DiMustNoWaiters(root);
3427 +       IiMustNoWaiters(root->d_inode);
3428 +       di_write_unlock(root);
3429 +       err = test_dentry_busy(root, bindex, sigen, verbose);
3430 +       if (!err)
3431 +               err = test_inode_busy(root->d_sb, bindex, sigen, verbose);
3432 +       di_write_lock_child(root); /* aufs_write_lock() calls ..._child() */
3433 +
3434 +       return err;
3435 +}
3436 +
3437 +static int test_dir_busy(struct file *file, aufs_bindex_t br_id,
3438 +                        struct file **to_free, int *idx)
3439 +{
3440 +       int err;
3441 +       unsigned char matched, root;
3442 +       aufs_bindex_t bindex, bend;
3443 +       struct au_fidir *fidir;
3444 +       struct au_hfile *hfile;
3445 +
3446 +       err = 0;
3447 +       root = IS_ROOT(file->f_path.dentry);
3448 +       if (root) {
3449 +               get_file(file);
3450 +               to_free[*idx] = file;
3451 +               (*idx)++;
3452 +               goto out;
3453 +       }
3454 +
3455 +       matched = 0;
3456 +       fidir = au_fi(file)->fi_hdir;
3457 +       AuDebugOn(!fidir);
3458 +       bend = au_fbend_dir(file);
3459 +       for (bindex = au_fbstart(file); bindex <= bend; bindex++) {
3460 +               hfile = fidir->fd_hfile + bindex;
3461 +               if (!hfile->hf_file)
3462 +                       continue;
3463 +
3464 +               if (hfile->hf_br->br_id == br_id) {
3465 +                       matched = 1;
3466 +                       break;
3467 +               }
3468 +       }
3469 +       if (matched)
3470 +               err = -EBUSY;
3471 +
3472 +out:
3473 +       return err;
3474 +}
3475 +
3476 +static int test_file_busy(struct super_block *sb, aufs_bindex_t br_id,
3477 +                         struct file **to_free, int opened)
3478 +{
3479 +       int err, idx;
3480 +       unsigned long long ull, max;
3481 +       aufs_bindex_t bstart;
3482 +       struct file *file, **array;
3483 +       struct dentry *root;
3484 +       struct au_hfile *hfile;
3485 +
3486 +       array = au_farray_alloc(sb, &max);
3487 +       err = PTR_ERR(array);
3488 +       if (IS_ERR(array))
3489 +               goto out;
3490 +
3491 +       err = 0;
3492 +       idx = 0;
3493 +       root = sb->s_root;
3494 +       di_write_unlock(root);
3495 +       for (ull = 0; ull < max; ull++) {
3496 +               file = array[ull];
3497 +               if (unlikely(!file))
3498 +                       break;
3499 +
3500 +               /* AuDbg("%pD\n", file); */
3501 +               fi_read_lock(file);
3502 +               bstart = au_fbstart(file);
3503 +               if (!d_is_dir(file->f_path.dentry)) {
3504 +                       hfile = &au_fi(file)->fi_htop;
3505 +                       if (hfile->hf_br->br_id == br_id)
3506 +                               err = -EBUSY;
3507 +               } else
3508 +                       err = test_dir_busy(file, br_id, to_free, &idx);
3509 +               fi_read_unlock(file);
3510 +               if (unlikely(err))
3511 +                       break;
3512 +       }
3513 +       di_write_lock_child(root);
3514 +       au_farray_free(array, max);
3515 +       AuDebugOn(idx > opened);
3516 +
3517 +out:
3518 +       return err;
3519 +}
3520 +
3521 +static void br_del_file(struct file **to_free, unsigned long long opened,
3522 +                         aufs_bindex_t br_id)
3523 +{
3524 +       unsigned long long ull;
3525 +       aufs_bindex_t bindex, bstart, bend, bfound;
3526 +       struct file *file;
3527 +       struct au_fidir *fidir;
3528 +       struct au_hfile *hfile;
3529 +
3530 +       for (ull = 0; ull < opened; ull++) {
3531 +               file = to_free[ull];
3532 +               if (unlikely(!file))
3533 +                       break;
3534 +
3535 +               /* AuDbg("%pD\n", file); */
3536 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
3537 +               bfound = -1;
3538 +               fidir = au_fi(file)->fi_hdir;
3539 +               AuDebugOn(!fidir);
3540 +               fi_write_lock(file);
3541 +               bstart = au_fbstart(file);
3542 +               bend = au_fbend_dir(file);
3543 +               for (bindex = bstart; bindex <= bend; bindex++) {
3544 +                       hfile = fidir->fd_hfile + bindex;
3545 +                       if (!hfile->hf_file)
3546 +                               continue;
3547 +
3548 +                       if (hfile->hf_br->br_id == br_id) {
3549 +                               bfound = bindex;
3550 +                               break;
3551 +                       }
3552 +               }
3553 +               AuDebugOn(bfound < 0);
3554 +               au_set_h_fptr(file, bfound, NULL);
3555 +               if (bfound == bstart) {
3556 +                       for (bstart++; bstart <= bend; bstart++)
3557 +                               if (au_hf_dir(file, bstart)) {
3558 +                                       au_set_fbstart(file, bstart);
3559 +                                       break;
3560 +                               }
3561 +               }
3562 +               fi_write_unlock(file);
3563 +       }
3564 +}
3565 +
3566 +static void au_br_do_del_brp(struct au_sbinfo *sbinfo,
3567 +                            const aufs_bindex_t bindex,
3568 +                            const aufs_bindex_t bend)
3569 +{
3570 +       struct au_branch **brp, **p;
3571 +
3572 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3573 +
3574 +       brp = sbinfo->si_branch + bindex;
3575 +       if (bindex < bend)
3576 +               memmove(brp, brp + 1, sizeof(*brp) * (bend - bindex));
3577 +       sbinfo->si_branch[0 + bend] = NULL;
3578 +       sbinfo->si_bend--;
3579 +
3580 +       p = krealloc(sbinfo->si_branch, sizeof(*p) * bend, AuGFP_SBILIST);
3581 +       if (p)
3582 +               sbinfo->si_branch = p;
3583 +       /* harmless error */
3584 +}
3585 +
3586 +static void au_br_do_del_hdp(struct au_dinfo *dinfo, const aufs_bindex_t bindex,
3587 +                            const aufs_bindex_t bend)
3588 +{
3589 +       struct au_hdentry *hdp, *p;
3590 +
3591 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3592 +
3593 +       hdp = dinfo->di_hdentry;
3594 +       if (bindex < bend)
3595 +               memmove(hdp + bindex, hdp + bindex + 1,
3596 +                       sizeof(*hdp) * (bend - bindex));
3597 +       hdp[0 + bend].hd_dentry = NULL;
3598 +       dinfo->di_bend--;
3599 +
3600 +       p = krealloc(hdp, sizeof(*p) * bend, AuGFP_SBILIST);
3601 +       if (p)
3602 +               dinfo->di_hdentry = p;
3603 +       /* harmless error */
3604 +}
3605 +
3606 +static void au_br_do_del_hip(struct au_iinfo *iinfo, const aufs_bindex_t bindex,
3607 +                            const aufs_bindex_t bend)
3608 +{
3609 +       struct au_hinode *hip, *p;
3610 +
3611 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3612 +
3613 +       hip = iinfo->ii_hinode + bindex;
3614 +       if (bindex < bend)
3615 +               memmove(hip, hip + 1, sizeof(*hip) * (bend - bindex));
3616 +       iinfo->ii_hinode[0 + bend].hi_inode = NULL;
3617 +       au_hn_init(iinfo->ii_hinode + bend);
3618 +       iinfo->ii_bend--;
3619 +
3620 +       p = krealloc(iinfo->ii_hinode, sizeof(*p) * bend, AuGFP_SBILIST);
3621 +       if (p)
3622 +               iinfo->ii_hinode = p;
3623 +       /* harmless error */
3624 +}
3625 +
3626 +static void au_br_do_del(struct super_block *sb, aufs_bindex_t bindex,
3627 +                        struct au_branch *br)
3628 +{
3629 +       aufs_bindex_t bend;
3630 +       struct au_sbinfo *sbinfo;
3631 +       struct dentry *root, *h_root;
3632 +       struct inode *inode, *h_inode;
3633 +       struct au_hinode *hinode;
3634 +
3635 +       SiMustWriteLock(sb);
3636 +
3637 +       root = sb->s_root;
3638 +       inode = root->d_inode;
3639 +       sbinfo = au_sbi(sb);
3640 +       bend = sbinfo->si_bend;
3641 +
3642 +       h_root = au_h_dptr(root, bindex);
3643 +       hinode = au_hi(inode, bindex);
3644 +       h_inode = au_igrab(hinode->hi_inode);
3645 +       au_hiput(hinode);
3646 +
3647 +       au_sbilist_lock();
3648 +       au_br_do_del_brp(sbinfo, bindex, bend);
3649 +       au_br_do_del_hdp(au_di(root), bindex, bend);
3650 +       au_br_do_del_hip(au_ii(inode), bindex, bend);
3651 +       au_sbilist_unlock();
3652 +
3653 +       dput(h_root);
3654 +       iput(h_inode);
3655 +       au_br_do_free(br);
3656 +}
3657 +
3658 +static unsigned long long empty_cb(void *array, unsigned long long max,
3659 +                                  void *arg)
3660 +{
3661 +       return max;
3662 +}
3663 +
3664 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount)
3665 +{
3666 +       int err, rerr, i;
3667 +       unsigned long long opened;
3668 +       unsigned int mnt_flags;
3669 +       aufs_bindex_t bindex, bend, br_id;
3670 +       unsigned char do_wh, verbose;
3671 +       struct au_branch *br;
3672 +       struct au_wbr *wbr;
3673 +       struct dentry *root;
3674 +       struct file **to_free;
3675 +
3676 +       err = 0;
3677 +       opened = 0;
3678 +       to_free = NULL;
3679 +       root = sb->s_root;
3680 +       bindex = au_find_dbindex(root, del->h_path.dentry);
3681 +       if (bindex < 0) {
3682 +               if (remount)
3683 +                       goto out; /* success */
3684 +               err = -ENOENT;
3685 +               pr_err("%s no such branch\n", del->pathname);
3686 +               goto out;
3687 +       }
3688 +       AuDbg("bindex b%d\n", bindex);
3689 +
3690 +       err = -EBUSY;
3691 +       mnt_flags = au_mntflags(sb);
3692 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3693 +       bend = au_sbend(sb);
3694 +       if (unlikely(!bend)) {
3695 +               AuVerbose(verbose, "no more branches left\n");
3696 +               goto out;
3697 +       }
3698 +       br = au_sbr(sb, bindex);
3699 +       AuDebugOn(!path_equal(&br->br_path, &del->h_path));
3700 +
3701 +       br_id = br->br_id;
3702 +       opened = atomic_read(&br->br_count);
3703 +       if (unlikely(opened)) {
3704 +               to_free = au_array_alloc(&opened, empty_cb, NULL);
3705 +               err = PTR_ERR(to_free);
3706 +               if (IS_ERR(to_free))
3707 +                       goto out;
3708 +
3709 +               err = test_file_busy(sb, br_id, to_free, opened);
3710 +               if (unlikely(err)) {
3711 +                       AuVerbose(verbose, "%llu file(s) opened\n", opened);
3712 +                       goto out;
3713 +               }
3714 +       }
3715 +
3716 +       wbr = br->br_wbr;
3717 +       do_wh = wbr && (wbr->wbr_whbase || wbr->wbr_plink || wbr->wbr_orph);
3718 +       if (do_wh) {
3719 +               /* instead of WbrWhMustWriteLock(wbr) */
3720 +               SiMustWriteLock(sb);
3721 +               for (i = 0; i < AuBrWh_Last; i++) {
3722 +                       dput(wbr->wbr_wh[i]);
3723 +                       wbr->wbr_wh[i] = NULL;
3724 +               }
3725 +       }
3726 +
3727 +       err = test_children_busy(root, bindex, verbose);
3728 +       if (unlikely(err)) {
3729 +               if (do_wh)
3730 +                       goto out_wh;
3731 +               goto out;
3732 +       }
3733 +
3734 +       err = 0;
3735 +       if (to_free) {
3736 +               /*
3737 +                * now we confirmed the branch is deletable.
3738 +                * let's free the remaining opened dirs on the branch.
3739 +                */
3740 +               di_write_unlock(root);
3741 +               br_del_file(to_free, opened, br_id);
3742 +               di_write_lock_child(root);
3743 +       }
3744 +
3745 +       if (!remount)
3746 +               au_br_do_del(sb, bindex, br);
3747 +       else {
3748 +               sysaufs_brs_del(sb, bindex);
3749 +               au_br_do_del(sb, bindex, br);
3750 +               sysaufs_brs_add(sb, bindex);
3751 +       }
3752 +
3753 +       if (!bindex) {
3754 +               au_cpup_attr_all(root->d_inode, /*force*/1);
3755 +               sb->s_maxbytes = au_sbr_sb(sb, 0)->s_maxbytes;
3756 +       } else
3757 +               au_sub_nlink(root->d_inode, del->h_path.dentry->d_inode);
3758 +       if (au_opt_test(mnt_flags, PLINK))
3759 +               au_plink_half_refresh(sb, br_id);
3760 +
3761 +       if (au_xino_brid(sb) == br_id)
3762 +               au_xino_brid_set(sb, -1);
3763 +       goto out; /* success */
3764 +
3765 +out_wh:
3766 +       /* revert */
3767 +       rerr = au_br_init_wh(sb, br, br->br_perm);
3768 +       if (rerr)
3769 +               pr_warn("failed re-creating base whiteout, %s. (%d)\n",
3770 +                       del->pathname, rerr);
3771 +out:
3772 +       if (to_free)
3773 +               au_farray_free(to_free, opened);
3774 +       return err;
3775 +}
3776 +
3777 +/* ---------------------------------------------------------------------- */
3778 +
3779 +static int au_ibusy(struct super_block *sb, struct aufs_ibusy __user *arg)
3780 +{
3781 +       int err;
3782 +       aufs_bindex_t bstart, bend;
3783 +       struct aufs_ibusy ibusy;
3784 +       struct inode *inode, *h_inode;
3785 +
3786 +       err = -EPERM;
3787 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
3788 +               goto out;
3789 +
3790 +       err = copy_from_user(&ibusy, arg, sizeof(ibusy));
3791 +       if (!err)
3792 +               err = !access_ok(VERIFY_WRITE, &arg->h_ino, sizeof(arg->h_ino));
3793 +       if (unlikely(err)) {
3794 +               err = -EFAULT;
3795 +               AuTraceErr(err);
3796 +               goto out;
3797 +       }
3798 +
3799 +       err = -EINVAL;
3800 +       si_read_lock(sb, AuLock_FLUSH);
3801 +       if (unlikely(ibusy.bindex < 0 || ibusy.bindex > au_sbend(sb)))
3802 +               goto out_unlock;
3803 +
3804 +       err = 0;
3805 +       ibusy.h_ino = 0; /* invalid */
3806 +       inode = ilookup(sb, ibusy.ino);
3807 +       if (!inode
3808 +           || inode->i_ino == AUFS_ROOT_INO
3809 +           || is_bad_inode(inode))
3810 +               goto out_unlock;
3811 +
3812 +       ii_read_lock_child(inode);
3813 +       bstart = au_ibstart(inode);
3814 +       bend = au_ibend(inode);
3815 +       if (bstart <= ibusy.bindex && ibusy.bindex <= bend) {
3816 +               h_inode = au_h_iptr(inode, ibusy.bindex);
3817 +               if (h_inode && au_test_ibusy(inode, bstart, bend))
3818 +                       ibusy.h_ino = h_inode->i_ino;
3819 +       }
3820 +       ii_read_unlock(inode);
3821 +       iput(inode);
3822 +
3823 +out_unlock:
3824 +       si_read_unlock(sb);
3825 +       if (!err) {
3826 +               err = __put_user(ibusy.h_ino, &arg->h_ino);
3827 +               if (unlikely(err)) {
3828 +                       err = -EFAULT;
3829 +                       AuTraceErr(err);
3830 +               }
3831 +       }
3832 +out:
3833 +       return err;
3834 +}
3835 +
3836 +long au_ibusy_ioctl(struct file *file, unsigned long arg)
3837 +{
3838 +       return au_ibusy(file->f_path.dentry->d_sb, (void __user *)arg);
3839 +}
3840 +
3841 +#ifdef CONFIG_COMPAT
3842 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg)
3843 +{
3844 +       return au_ibusy(file->f_path.dentry->d_sb, compat_ptr(arg));
3845 +}
3846 +#endif
3847 +
3848 +/* ---------------------------------------------------------------------- */
3849 +
3850 +/*
3851 + * change a branch permission
3852 + */
3853 +
3854 +static void au_warn_ima(void)
3855 +{
3856 +#ifdef CONFIG_IMA
3857 +       /* since it doesn't support mark_files_ro() */
3858 +       AuWarn1("RW -> RO makes IMA to produce wrong message\n");
3859 +#endif
3860 +}
3861 +
3862 +static int do_need_sigen_inc(int a, int b)
3863 +{
3864 +       return au_br_whable(a) && !au_br_whable(b);
3865 +}
3866 +
3867 +static int need_sigen_inc(int old, int new)
3868 +{
3869 +       return do_need_sigen_inc(old, new)
3870 +               || do_need_sigen_inc(new, old);
3871 +}
3872 +
3873 +static int au_br_mod_files_ro(struct super_block *sb, aufs_bindex_t bindex)
3874 +{
3875 +       int err, do_warn;
3876 +       unsigned int mnt_flags;
3877 +       unsigned long long ull, max;
3878 +       aufs_bindex_t br_id;
3879 +       unsigned char verbose, writer;
3880 +       struct file *file, *hf, **array;
3881 +       struct inode *inode;
3882 +       struct au_hfile *hfile;
3883 +
3884 +       mnt_flags = au_mntflags(sb);
3885 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3886 +
3887 +       array = au_farray_alloc(sb, &max);
3888 +       err = PTR_ERR(array);
3889 +       if (IS_ERR(array))
3890 +               goto out;
3891 +
3892 +       do_warn = 0;
3893 +       br_id = au_sbr_id(sb, bindex);
3894 +       for (ull = 0; ull < max; ull++) {
3895 +               file = array[ull];
3896 +               if (unlikely(!file))
3897 +                       break;
3898 +
3899 +               /* AuDbg("%pD\n", file); */
3900 +               fi_read_lock(file);
3901 +               if (unlikely(au_test_mmapped(file))) {
3902 +                       err = -EBUSY;
3903 +                       AuVerbose(verbose, "mmapped %pD\n", file);
3904 +                       AuDbgFile(file);
3905 +                       FiMustNoWaiters(file);
3906 +                       fi_read_unlock(file);
3907 +                       goto out_array;
3908 +               }
3909 +
3910 +               inode = file_inode(file);
3911 +               hfile = &au_fi(file)->fi_htop;
3912 +               hf = hfile->hf_file;
3913 +               if (!S_ISREG(inode->i_mode)
3914 +                   || !(file->f_mode & FMODE_WRITE)
3915 +                   || hfile->hf_br->br_id != br_id
3916 +                   || !(hf->f_mode & FMODE_WRITE))
3917 +                       array[ull] = NULL;
3918 +               else {
3919 +                       do_warn = 1;
3920 +                       get_file(file);
3921 +               }
3922 +
3923 +               FiMustNoWaiters(file);
3924 +               fi_read_unlock(file);
3925 +               fput(file);
3926 +       }
3927 +
3928 +       err = 0;
3929 +       if (do_warn)
3930 +               au_warn_ima();
3931 +
3932 +       for (ull = 0; ull < max; ull++) {
3933 +               file = array[ull];
3934 +               if (!file)
3935 +                       continue;
3936 +
3937 +               /* todo: already flushed? */
3938 +               /*
3939 +                * fs/super.c:mark_files_ro() is gone, but aufs keeps its
3940 +                * approach which resets f_mode and calls mnt_drop_write() and
3941 +                * file_release_write() for each file, because the branch
3942 +                * attribute in aufs world is totally different from the native
3943 +                * fs rw/ro mode.
3944 +               */
3945 +               /* fi_read_lock(file); */
3946 +               hfile = &au_fi(file)->fi_htop;
3947 +               hf = hfile->hf_file;
3948 +               /* fi_read_unlock(file); */
3949 +               spin_lock(&hf->f_lock);
3950 +               writer = !!(hf->f_mode & FMODE_WRITER);
3951 +               hf->f_mode &= ~(FMODE_WRITE | FMODE_WRITER);
3952 +               spin_unlock(&hf->f_lock);
3953 +               if (writer) {
3954 +                       put_write_access(file_inode(hf));
3955 +                       __mnt_drop_write(hf->f_path.mnt);
3956 +               }
3957 +       }
3958 +
3959 +out_array:
3960 +       au_farray_free(array, max);
3961 +out:
3962 +       AuTraceErr(err);
3963 +       return err;
3964 +}
3965 +
3966 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
3967 +             int *do_refresh)
3968 +{
3969 +       int err, rerr;
3970 +       aufs_bindex_t bindex;
3971 +       struct dentry *root;
3972 +       struct au_branch *br;
3973 +       struct au_br_fhsm *bf;
3974 +
3975 +       root = sb->s_root;
3976 +       bindex = au_find_dbindex(root, mod->h_root);
3977 +       if (bindex < 0) {
3978 +               if (remount)
3979 +                       return 0; /* success */
3980 +               err = -ENOENT;
3981 +               pr_err("%s no such branch\n", mod->path);
3982 +               goto out;
3983 +       }
3984 +       AuDbg("bindex b%d\n", bindex);
3985 +
3986 +       err = test_br(mod->h_root->d_inode, mod->perm, mod->path);
3987 +       if (unlikely(err))
3988 +               goto out;
3989 +
3990 +       br = au_sbr(sb, bindex);
3991 +       AuDebugOn(mod->h_root != au_br_dentry(br));
3992 +       if (br->br_perm == mod->perm)
3993 +               return 0; /* success */
3994 +
3995 +       /* pre-allocate for non-fhsm --> fhsm */
3996 +       bf = NULL;
3997 +       if (!au_br_fhsm(br->br_perm) && au_br_fhsm(mod->perm)) {
3998 +               err = au_fhsm_br_alloc(br);
3999 +               if (unlikely(err))
4000 +                       goto out;
4001 +               bf = br->br_fhsm;
4002 +               br->br_fhsm = NULL;
4003 +       }
4004 +
4005 +       if (au_br_writable(br->br_perm)) {
4006 +               /* remove whiteout base */
4007 +               err = au_br_init_wh(sb, br, mod->perm);
4008 +               if (unlikely(err))
4009 +                       goto out_bf;
4010 +
4011 +               if (!au_br_writable(mod->perm)) {
4012 +                       /* rw --> ro, file might be mmapped */
4013 +                       DiMustNoWaiters(root);
4014 +                       IiMustNoWaiters(root->d_inode);
4015 +                       di_write_unlock(root);
4016 +                       err = au_br_mod_files_ro(sb, bindex);
4017 +                       /* aufs_write_lock() calls ..._child() */
4018 +                       di_write_lock_child(root);
4019 +
4020 +                       if (unlikely(err)) {
4021 +                               rerr = -ENOMEM;
4022 +                               br->br_wbr = kmalloc(sizeof(*br->br_wbr),
4023 +                                                    GFP_NOFS);
4024 +                               if (br->br_wbr)
4025 +                                       rerr = au_wbr_init(br, sb, br->br_perm);
4026 +                               if (unlikely(rerr)) {
4027 +                                       AuIOErr("nested error %d (%d)\n",
4028 +                                               rerr, err);
4029 +                                       br->br_perm = mod->perm;
4030 +                               }
4031 +                       }
4032 +               }
4033 +       } else if (au_br_writable(mod->perm)) {
4034 +               /* ro --> rw */
4035 +               err = -ENOMEM;
4036 +               br->br_wbr = kmalloc(sizeof(*br->br_wbr), GFP_NOFS);
4037 +               if (br->br_wbr) {
4038 +                       err = au_wbr_init(br, sb, mod->perm);
4039 +                       if (unlikely(err)) {
4040 +                               kfree(br->br_wbr);
4041 +                               br->br_wbr = NULL;
4042 +                       }
4043 +               }
4044 +       }
4045 +       if (unlikely(err))
4046 +               goto out_bf;
4047 +
4048 +       if (au_br_fhsm(br->br_perm)) {
4049 +               if (!au_br_fhsm(mod->perm)) {
4050 +                       /* fhsm --> non-fhsm */
4051 +                       au_br_fhsm_fin(br->br_fhsm);
4052 +                       kfree(br->br_fhsm);
4053 +                       br->br_fhsm = NULL;
4054 +               }
4055 +       } else if (au_br_fhsm(mod->perm))
4056 +               /* non-fhsm --> fhsm */
4057 +               br->br_fhsm = bf;
4058 +
4059 +       *do_refresh |= need_sigen_inc(br->br_perm, mod->perm);
4060 +       br->br_perm = mod->perm;
4061 +       goto out; /* success */
4062 +
4063 +out_bf:
4064 +       kfree(bf);
4065 +out:
4066 +       AuTraceErr(err);
4067 +       return err;
4068 +}
4069 +
4070 +/* ---------------------------------------------------------------------- */
4071 +
4072 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs)
4073 +{
4074 +       int err;
4075 +       struct kstatfs kstfs;
4076 +
4077 +       err = vfs_statfs(&br->br_path, &kstfs);
4078 +       if (!err) {
4079 +               stfs->f_blocks = kstfs.f_blocks;
4080 +               stfs->f_bavail = kstfs.f_bavail;
4081 +               stfs->f_files = kstfs.f_files;
4082 +               stfs->f_ffree = kstfs.f_ffree;
4083 +       }
4084 +
4085 +       return err;
4086 +}
4087 diff -urN /usr/share/empty/fs/aufs/branch.h linux/fs/aufs/branch.h
4088 --- /usr/share/empty/fs/aufs/branch.h   1970-01-01 01:00:00.000000000 +0100
4089 +++ linux/fs/aufs/branch.h      2015-03-27 21:56:35.460128334 +0100
4090 @@ -0,0 +1,267 @@
4091 +/*
4092 + * Copyright (C) 2005-2015 Junjiro R. Okajima
4093 + *
4094 + * This program, aufs is free software; you can redistribute it and/or modify
4095 + * it under the terms of the GNU General Public License as published by
4096 + * the Free Software Foundation; either version 2 of the License, or
4097 + * (at your option) any later version.
4098 + *
4099 + * This program is distributed in the hope that it will be useful,
4100 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4101 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4102 + * GNU General Public License for more details.
4103 + *
4104 + * You should have received a copy of the GNU General Public License
4105 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4106 + */
4107 +
4108 +/*
4109 + * branch filesystems and xino for them
4110 + */
4111 +
4112 +#ifndef __AUFS_BRANCH_H__
4113 +#define __AUFS_BRANCH_H__
4114 +
4115 +#ifdef __KERNEL__
4116 +
4117 +#include <linux/mount.h>
4118 +#include "dynop.h"
4119 +#include "rwsem.h"
4120 +#include "super.h"
4121 +
4122 +/* ---------------------------------------------------------------------- */
4123 +
4124 +/* a xino file */
4125 +struct au_xino_file {
4126 +       struct file             *xi_file;
4127 +       struct mutex            xi_nondir_mtx;
4128 +
4129 +       /* todo: make xino files an array to support huge inode number */
4130 +
4131 +#ifdef CONFIG_DEBUG_FS
4132 +       struct dentry            *xi_dbgaufs;
4133 +#endif
4134 +};
4135 +
4136 +/* File-based Hierarchical Storage Management */
4137 +struct au_br_fhsm {
4138 +#ifdef CONFIG_AUFS_FHSM
4139 +       struct mutex            bf_lock;
4140 +       unsigned long           bf_jiffy;
4141 +       struct aufs_stfs        bf_stfs;
4142 +       int                     bf_readable;
4143 +#endif
4144 +};
4145 +
4146 +/* members for writable branch only */
4147 +enum {AuBrWh_BASE, AuBrWh_PLINK, AuBrWh_ORPH, AuBrWh_Last};
4148 +struct au_wbr {
4149 +       struct au_rwsem         wbr_wh_rwsem;
4150 +       struct dentry           *wbr_wh[AuBrWh_Last];
4151 +       atomic_t                wbr_wh_running;
4152 +#define wbr_whbase             wbr_wh[AuBrWh_BASE]     /* whiteout base */
4153 +#define wbr_plink              wbr_wh[AuBrWh_PLINK]    /* pseudo-link dir */
4154 +#define wbr_orph               wbr_wh[AuBrWh_ORPH]     /* dir for orphans */
4155 +
4156 +       /* mfs mode */
4157 +       unsigned long long      wbr_bytes;
4158 +};
4159 +
4160 +/* ext2 has 3 types of operations at least, ext3 has 4 */
4161 +#define AuBrDynOp (AuDyLast * 4)
4162 +
4163 +#ifdef CONFIG_AUFS_HFSNOTIFY
4164 +/* support for asynchronous destruction */
4165 +struct au_br_hfsnotify {
4166 +       struct fsnotify_group   *hfsn_group;
4167 +};
4168 +#endif
4169 +
4170 +/* sysfs entries */
4171 +struct au_brsysfs {
4172 +       char                    name[16];
4173 +       struct attribute        attr;
4174 +};
4175 +
4176 +enum {
4177 +       AuBrSysfs_BR,
4178 +       AuBrSysfs_BRID,
4179 +       AuBrSysfs_Last
4180 +};
4181 +
4182 +/* protected by superblock rwsem */
4183 +struct au_branch {
4184 +       struct au_xino_file     br_xino;
4185 +
4186 +       aufs_bindex_t           br_id;
4187 +
4188 +       int                     br_perm;
4189 +       struct path             br_path;
4190 +       spinlock_t              br_dykey_lock;
4191 +       struct au_dykey         *br_dykey[AuBrDynOp];
4192 +       atomic_t                br_count;
4193 +
4194 +       struct au_wbr           *br_wbr;
4195 +       struct au_br_fhsm       *br_fhsm;
4196 +
4197 +       /* xino truncation */
4198 +       atomic_t                br_xino_running;
4199 +
4200 +#ifdef CONFIG_AUFS_HFSNOTIFY
4201 +       struct au_br_hfsnotify  *br_hfsn;
4202 +#endif
4203 +
4204 +#ifdef CONFIG_SYSFS
4205 +       /* entries under sysfs per mount-point */
4206 +       struct au_brsysfs       br_sysfs[AuBrSysfs_Last];
4207 +#endif
4208 +};
4209 +
4210 +/* ---------------------------------------------------------------------- */
4211 +
4212 +static inline struct vfsmount *au_br_mnt(struct au_branch *br)
4213 +{
4214 +       return br->br_path.mnt;
4215 +}
4216 +
4217 +static inline struct dentry *au_br_dentry(struct au_branch *br)
4218 +{
4219 +       return br->br_path.dentry;
4220 +}
4221 +
4222 +static inline struct super_block *au_br_sb(struct au_branch *br)
4223 +{
4224 +       return au_br_mnt(br)->mnt_sb;
4225 +}
4226 +
4227 +static inline int au_br_rdonly(struct au_branch *br)
4228 +{
4229 +       return ((au_br_sb(br)->s_flags & MS_RDONLY)
4230 +               || !au_br_writable(br->br_perm))
4231 +               ? -EROFS : 0;
4232 +}
4233 +
4234 +static inline int au_br_hnotifyable(int brperm __maybe_unused)
4235 +{
4236 +#ifdef CONFIG_AUFS_HNOTIFY
4237 +       return !(brperm & AuBrPerm_RR);
4238 +#else
4239 +       return 0;
4240 +#endif
4241 +}
4242 +
4243 +/* ---------------------------------------------------------------------- */
4244 +
4245 +/* branch.c */
4246 +struct au_sbinfo;
4247 +void au_br_free(struct au_sbinfo *sinfo);
4248 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id);
4249 +struct au_opt_add;
4250 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount);
4251 +struct au_opt_del;
4252 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount);
4253 +long au_ibusy_ioctl(struct file *file, unsigned long arg);
4254 +#ifdef CONFIG_COMPAT
4255 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg);
4256 +#endif
4257 +struct au_opt_mod;
4258 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
4259 +             int *do_refresh);
4260 +struct aufs_stfs;
4261 +int au_br_stfs(struct au_branch *br, struct aufs_stfs *stfs);
4262 +
4263 +/* xino.c */
4264 +static const loff_t au_loff_max = LLONG_MAX;
4265 +
4266 +int au_xib_trunc(struct super_block *sb);
4267 +ssize_t xino_fread(au_readf_t func, struct file *file, void *buf, size_t size,
4268 +                  loff_t *pos);
4269 +ssize_t xino_fwrite(au_writef_t func, struct file *file, void *buf, size_t size,
4270 +                   loff_t *pos);
4271 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src);
4272 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent);
4273 +ino_t au_xino_new_ino(struct super_block *sb);
4274 +void au_xino_delete_inode(struct inode *inode, const int unlinked);
4275 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4276 +                 ino_t ino);
4277 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
4278 +                ino_t *ino);
4279 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t hino,
4280 +              struct file *base_file, int do_test);
4281 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex);
4282 +
4283 +struct au_opt_xino;
4284 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount);
4285 +void au_xino_clr(struct super_block *sb);
4286 +struct file *au_xino_def(struct super_block *sb);
4287 +int au_xino_path(struct seq_file *seq, struct file *file);
4288 +
4289 +/* ---------------------------------------------------------------------- */
4290 +
4291 +/* Superblock to branch */
4292 +static inline
4293 +aufs_bindex_t au_sbr_id(struct super_block *sb, aufs_bindex_t bindex)
4294 +{
4295 +       return au_sbr(sb, bindex)->br_id;
4296 +}
4297 +
4298 +static inline
4299 +struct vfsmount *au_sbr_mnt(struct super_block *sb, aufs_bindex_t bindex)
4300 +{
4301 +       return au_br_mnt(au_sbr(sb, bindex));
4302 +}
4303 +
4304 +static inline
4305 +struct super_block *au_sbr_sb(struct super_block *sb, aufs_bindex_t bindex)
4306 +{
4307 +       return au_br_sb(au_sbr(sb, bindex));
4308 +}
4309 +
4310 +static inline void au_sbr_put(struct super_block *sb, aufs_bindex_t bindex)
4311 +{
4312 +       atomic_dec(&au_sbr(sb, bindex)->br_count);
4313 +}
4314 +
4315 +static inline int au_sbr_perm(struct super_block *sb, aufs_bindex_t bindex)
4316 +{
4317 +       return au_sbr(sb, bindex)->br_perm;
4318 +}
4319 +
4320 +static inline int au_sbr_whable(struct super_block *sb, aufs_bindex_t bindex)
4321 +{
4322 +       return au_br_whable(au_sbr_perm(sb, bindex));
4323 +}
4324 +
4325 +/* ---------------------------------------------------------------------- */
4326 +
4327 +/*
4328 + * wbr_wh_read_lock, wbr_wh_write_lock
4329 + * wbr_wh_read_unlock, wbr_wh_write_unlock, wbr_wh_downgrade_lock
4330 + */
4331 +AuSimpleRwsemFuncs(wbr_wh, struct au_wbr *wbr, &wbr->wbr_wh_rwsem);
4332 +
4333 +#define WbrWhMustNoWaiters(wbr)        AuRwMustNoWaiters(&wbr->wbr_wh_rwsem)
4334 +#define WbrWhMustAnyLock(wbr)  AuRwMustAnyLock(&wbr->wbr_wh_rwsem)
4335 +#define WbrWhMustWriteLock(wbr)        AuRwMustWriteLock(&wbr->wbr_wh_rwsem)
4336 +
4337 +/* ---------------------------------------------------------------------- */
4338 +
4339 +#ifdef CONFIG_AUFS_FHSM
4340 +static inline void au_br_fhsm_init(struct au_br_fhsm *brfhsm)
4341 +{
4342 +       mutex_init(&brfhsm->bf_lock);
4343 +       brfhsm->bf_jiffy = 0;
4344 +       brfhsm->bf_readable = 0;
4345 +}
4346 +
4347 +static inline void au_br_fhsm_fin(struct au_br_fhsm *brfhsm)
4348 +{
4349 +       mutex_destroy(&brfhsm->bf_lock);
4350 +}
4351 +#else
4352 +AuStubVoid(au_br_fhsm_init, struct au_br_fhsm *brfhsm)
4353 +AuStubVoid(au_br_fhsm_fin, struct au_br_fhsm *brfhsm)
4354 +#endif
4355 +
4356 +#endif /* __KERNEL__ */
4357 +#endif /* __AUFS_BRANCH_H__ */
4358 diff -urN /usr/share/empty/fs/aufs/conf.mk linux/fs/aufs/conf.mk
4359 --- /usr/share/empty/fs/aufs/conf.mk    1970-01-01 01:00:00.000000000 +0100
4360 +++ linux/fs/aufs/conf.mk       2015-03-27 21:56:35.460128334 +0100
4361 @@ -0,0 +1,38 @@
4362 +
4363 +AuConfStr = CONFIG_AUFS_FS=${CONFIG_AUFS_FS}
4364 +
4365 +define AuConf
4366 +ifdef ${1}
4367 +AuConfStr += ${1}=${${1}}
4368 +endif
4369 +endef
4370 +
4371 +AuConfAll = BRANCH_MAX_127 BRANCH_MAX_511 BRANCH_MAX_1023 BRANCH_MAX_32767 \
4372 +       SBILIST \
4373 +       HNOTIFY HFSNOTIFY \
4374 +       EXPORT INO_T_64 \
4375 +       XATTR \
4376 +       FHSM \
4377 +       RDU \
4378 +       SHWH \
4379 +       BR_RAMFS \
4380 +       BR_FUSE POLL \
4381 +       BR_HFSPLUS \
4382 +       BDEV_LOOP \
4383 +       DEBUG MAGIC_SYSRQ
4384 +$(foreach i, ${AuConfAll}, \
4385 +       $(eval $(call AuConf,CONFIG_AUFS_${i})))
4386 +
4387 +AuConfName = ${obj}/conf.str
4388 +${AuConfName}.tmp: FORCE
4389 +       @echo ${AuConfStr} | tr ' ' '\n' | sed -e 's/^/"/' -e 's/$$/\\n"/' > $@
4390 +${AuConfName}: ${AuConfName}.tmp
4391 +       @diff -q $< $@ > /dev/null 2>&1 || { \
4392 +       echo '  GEN    ' $@; \
4393 +       cp -p $< $@; \
4394 +       }
4395 +FORCE:
4396 +clean-files += ${AuConfName} ${AuConfName}.tmp
4397 +${obj}/sysfs.o: ${AuConfName}
4398 +
4399 +-include ${srctree}/${src}/conf_priv.mk
4400 diff -urN /usr/share/empty/fs/aufs/cpup.c linux/fs/aufs/cpup.c
4401 --- /usr/share/empty/fs/aufs/cpup.c     1970-01-01 01:00:00.000000000 +0100
4402 +++ linux/fs/aufs/cpup.c        2015-03-27 21:56:35.460128334 +0100
4403 @@ -0,0 +1,1303 @@
4404 +/*
4405 + * Copyright (C) 2005-2015 Junjiro R. Okajima
4406 + *
4407 + * This program, aufs is free software; you can redistribute it and/or modify
4408 + * it under the terms of the GNU General Public License as published by
4409 + * the Free Software Foundation; either version 2 of the License, or
4410 + * (at your option) any later version.
4411 + *
4412 + * This program is distributed in the hope that it will be useful,
4413 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
4414 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
4415 + * GNU General Public License for more details.
4416 + *
4417 + * You should have received a copy of the GNU General Public License
4418 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
4419 + */
4420 +
4421 +/*
4422 + * copy-up functions, see wbr_policy.c for copy-down
4423 + */
4424 +
4425 +#include <linux/fs_stack.h>
4426 +#include <linux/mm.h>
4427 +#include "aufs.h"
4428 +
4429 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags)
4430 +{
4431 +       const unsigned int mask = S_DEAD | S_SWAPFILE | S_PRIVATE
4432 +               | S_NOATIME | S_NOCMTIME | S_AUTOMOUNT;
4433 +
4434 +       BUILD_BUG_ON(sizeof(iflags) != sizeof(dst->i_flags));
4435 +
4436 +       dst->i_flags |= iflags & ~mask;
4437 +       if (au_test_fs_notime(dst->i_sb))
4438 +               dst->i_flags |= S_NOATIME | S_NOCMTIME;
4439 +}
4440 +
4441 +void au_cpup_attr_timesizes(struct inode *inode)
4442 +{
4443 +       struct inode *h_inode;
4444 +
4445 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
4446 +       fsstack_copy_attr_times(inode, h_inode);
4447 +       fsstack_copy_inode_size(inode, h_inode);
4448 +}
4449 +
4450 +void au_cpup_attr_nlink(struct inode *inode, int force)
4451 +{
4452 +       struct inode *h_inode;
4453 +       struct super_block *sb;
4454 +       aufs_bindex_t bindex, bend;
4455 +
4456 +       sb = inode->i_sb;
4457 +       bindex = au_ibstart(inode);
4458 +       h_inode = au_h_iptr(inode, bindex);
4459 +       if (!force
4460 +           && !S_ISDIR(h_inode->i_mode)
4461 +           && au_opt_test(au_mntflags(sb), PLINK)
4462 +           && au_plink_test(inode))
4463 +               return;
4464 +
4465 +       /*
4466 +        * 0 can happen in revalidating.
4467 +        * h_inode->i_mutex may not be held here, but it is harmless since once
4468 +        * i_nlink reaches 0, it will never become positive except O_TMPFILE
4469 +        * case.
4470 +        * todo: O_TMPFILE+linkat(AT_SYMLINK_FOLLOW) bypassing aufs may cause
4471 +        *       the incorrect link count.
4472 +        */
4473 +       set_nlink(inode, h_inode->i_nlink);
4474 +
4475 +       /*
4476 +        * fewer nlink makes find(1) noisy, but larger nlink doesn't.
4477 +        * it may includes whplink directory.
4478 +        */
4479 +       if (S_ISDIR(h_inode->i_mode)) {
4480 +               bend = au_ibend(inode);
4481 +               for (bindex++; bindex <= bend; bindex++) {
4482 +                       h_inode = au_h_iptr(inode, bindex);
4483 +                       if (h_inode)
4484 +                               au_add_nlink(inode, h_inode);
4485 +               }
4486 +       }
4487 +}
4488 +
4489 +void au_cpup_attr_changeable(struct inode *inode)
4490 +{
4491 +       struct inode *h_inode;
4492 +
4493 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
4494 +       inode->i_mode = h_inode->i_mode;
4495 +       inode->i_uid = h_inode->i_uid;
4496 +       inode->i_gid = h_inode->i_gid;
4497 +       au_cpup_attr_timesizes(inode);
4498 +       au_cpup_attr_flags(inode, h_inode->i_flags);
4499 +}
4500 +
4501 +void au_cpup_igen(struct inode *inode, struct inode *h_inode)
4502 +{
4503 +       struct au_iinfo *iinfo = au_ii(inode);
4504 +
4505 +       IiMustWriteLock(inode);
4506 +
4507 +       iinfo->ii_higen = h_inode->i_generation;
4508 +       iinfo->ii_hsb1 = h_inode->i_sb;
4509 +}
4510 +
4511 +void au_cpup_attr_all(struct inode *inode, int force)
4512 +{
4513 +       struct inode *h_inode;
4514 +
4515 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
4516 +       au_cpup_attr_changeable(inode);
4517 +       if (inode->i_nlink > 0)
4518 +               au_cpup_attr_nlink(inode, force);
4519 +       inode->i_rdev = h_inode->i_rdev;
4520 +       inode->i_blkbits = h_inode->i_blkbits;
4521 +       au_cpup_igen(inode, h_inode);
4522 +}
4523 +
4524 +/* ---------------------------------------------------------------------- */
4525 +
4526 +/* Note: dt_dentry and dt_h_dentry are not dget/dput-ed */
4527 +
4528 +/* keep the timestamps of the parent dir when cpup */
4529 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
4530 +                   struct path *h_path)
4531 +{
4532 +       struct inode *h_inode;
4533 +
4534 +       dt->dt_dentry = dentry;
4535 +       dt->dt_h_path = *h_path;
4536 +       h_inode = h_path->dentry->d_inode;
4537 +       dt->dt_atime = h_inode->i_atime;
4538 +       dt->dt_mtime = h_inode->i_mtime;
4539 +       /* smp_mb(); */
4540 +}
4541 +
4542 +void au_dtime_revert(struct au_dtime *dt)
4543 +{
4544 +       struct iattr attr;
4545 +       int err;
4546 +
4547 +       attr.ia_atime = dt->dt_atime;
4548 +       attr.ia_mtime = dt->dt_mtime;
4549 +       attr.ia_valid = ATTR_FORCE | ATTR_MTIME | ATTR_MTIME_SET
4550 +               | ATTR_ATIME | ATTR_ATIME_SET;
4551 +
4552 +       /* no delegation since this is a directory */
4553 +       err = vfsub_notify_change(&dt->dt_h_path, &attr, /*delegated*/NULL);
4554 +       if (unlikely(err))
4555 +               pr_warn("restoring timestamps failed(%d). ignored\n", err);
4556 +}
4557 +
4558 +/* ---------------------------------------------------------------------- */
4559 +
4560 +/* internal use only */
4561 +struct au_cpup_reg_attr {
4562 +       int             valid;
4563 +       struct kstat    st;
4564 +       unsigned int    iflags; /* inode->i_flags */
4565 +};
4566 +
4567 +static noinline_for_stack
4568 +int cpup_iattr(struct dentry *dst, aufs_bindex_t bindex, struct dentry *h_src,
4569 +              struct au_cpup_reg_attr *h_src_attr)
4570 +{
4571 +       int err, sbits, icex;
4572 +       struct iattr ia;
4573 +       struct path h_path;
4574 +       struct inode *h_isrc, *h_idst;
4575 +       struct kstat *h_st;
4576 +       struct au_branch *br;
4577 +
4578 +       h_path.dentry = au_h_dptr(dst, bindex);
4579 +       h_idst = h_path.dentry->d_inode;
4580 +       br = au_sbr(dst->d_sb, bindex);
4581 +       h_path.mnt = au_br_mnt(br);
4582 +       h_isrc = h_src->d_inode;
4583 +       ia.ia_valid = ATTR_FORCE | ATTR_UID | ATTR_GID
4584 +               | ATTR_ATIME | ATTR_MTIME
4585 +               | ATTR_ATIME_SET | ATTR_MTIME_SET;
4586 +       if (h_src_attr && h_src_attr->valid) {
4587 +               h_st = &h_src_attr->st;
4588 +               ia.ia_uid = h_st->uid;
4589 +               ia.ia_gid = h_st->gid;
4590 +               ia.ia_atime = h_st->atime;
4591 +               ia.ia_mtime = h_st->mtime;
4592 +               if (h_idst->i_mode != h_st->mode
4593 +                   && !S_ISLNK(h_idst->i_mode)) {
4594 +                       ia.ia_valid |= ATTR_MODE;
4595 +                       ia.ia_mode = h_st->mode;
4596 +               }
4597 +               sbits = !!(h_st->mode & (S_ISUID | S_ISGID));
4598 +               au_cpup_attr_flags(h_idst, h_src_attr->iflags);
4599 +       } else {
4600 +               ia.ia_uid = h_isrc->i_uid;
4601 +               ia.ia_gid = h_isrc->i_gid;
4602 +               ia.ia_atime = h_isrc->i_atime;
4603 +               ia.ia_mtime = h_isrc->i_mtime;
4604 +               if (h_idst->i_mode != h_isrc->i_mode
4605 +                   && !S_ISLNK(h_idst->i_mode)) {
4606 +                       ia.ia_valid |= ATTR_MODE;
4607 +                       ia.ia_mode = h_isrc->i_mode;
4608 +               }
4609 +               sbits = !!(h_isrc->i_mode & (S_ISUID | S_ISGID));
4610 +               au_cpup_attr_flags(h_idst, h_isrc->i_flags);
4611 +       }
4612 +       /* no delegation since it is just created */
4613 +       err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4614 +
4615 +       /* is this nfs only? */
4616 +       if (!err && sbits && au_test_nfs(h_path.dentry->d_sb)) {
4617 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
4618 +               ia.ia_mode = h_isrc->i_mode;
4619 +               err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4620 +       }
4621 +
4622 +       icex = br->br_perm & AuBrAttr_ICEX;
4623 +       if (!err)
4624 +               err = au_cpup_xattr(h_path.dentry, h_src, icex);
4625 +
4626 +       return err;
4627 +}
4628 +
4629 +/* ---------------------------------------------------------------------- */
4630 +
4631 +static int au_do_copy_file(struct file *dst, struct file *src, loff_t len,
4632 +                          char *buf, unsigned long blksize)
4633 +{
4634 +       int err;
4635 +       size_t sz, rbytes, wbytes;
4636 +       unsigned char all_zero;
4637 +       char *p, *zp;
4638 +       struct mutex *h_mtx;
4639 +       /* reduce stack usage */
4640 +       struct iattr *ia;
4641 +
4642 +       zp = page_address(ZERO_PAGE(0));
4643 +       if (unlikely(!zp))
4644 +               return -ENOMEM; /* possible? */
4645 +
4646 +       err = 0;
4647 +       all_zero = 0;
4648 +       while (len) {
4649 +               AuDbg("len %lld\n", len);
4650 +               sz = blksize;
4651 +               if (len < blksize)
4652 +                       sz = len;
4653 +
4654 +               rbytes = 0;
4655 +               /* todo: signal_pending? */
4656 +               while (!rbytes || err == -EAGAIN || err == -EINTR) {
4657 +                       rbytes = vfsub_read_k(src, buf, sz, &src->f_pos);
4658 +                       err = rbytes;
4659 +               }
4660 +               if (unlikely(err < 0))
4661 +                       break;
4662 +
4663 +               all_zero = 0;
4664 +               if (len >= rbytes && rbytes == blksize)
4665 +                       all_zero = !memcmp(buf, zp, rbytes);
4666 +               if (!all_zero) {
4667 +                       wbytes = rbytes;
4668 +                       p = buf;
4669 +                       while (wbytes) {
4670 +                               size_t b;
4671 +
4672 +                               b = vfsub_write_k(dst, p, wbytes, &dst->f_pos);
4673 +                               err = b;
4674 +                               /* todo: signal_pending? */
4675 +                               if (unlikely(err == -EAGAIN || err == -EINTR))
4676 +                                       continue;
4677 +                               if (unlikely(err < 0))
4678 +                                       break;
4679 +                               wbytes -= b;
4680 +                               p += b;
4681 +                       }
4682 +                       if (unlikely(err < 0))
4683 +                               break;
4684 +               } else {
4685 +                       loff_t res;
4686 +
4687 +                       AuLabel(hole);
4688 +                       res = vfsub_llseek(dst, rbytes, SEEK_CUR);
4689 +                       err = res;
4690 +                       if (unlikely(res < 0))
4691 +                               break;
4692 +               }
4693 +               len -= rbytes;
4694 +               err = 0;
4695 +       }
4696 +
4697 +       /* the last block may be a hole */
4698 +       if (!err && all_zero) {
4699 +               AuLabel(last hole);
4700 +
4701 +               err = 1;
4702 +               if (au_test_nfs(dst->f_path.dentry->d_sb)) {
4703 +                       /* nfs requires this step to make last hole */
4704 +                       /* is this only nfs? */
4705 +                       do {
4706 +                               /* todo: signal_pending? */
4707 +                               err = vfsub_write_k(dst, "\0", 1, &dst->f_pos);
4708 +                       } while (err == -EAGAIN || err == -EINTR);
4709 +                       if (err == 1)
4710 +                               dst->f_pos--;
4711 +               }
4712 +
4713 +               if (err == 1) {
4714 +                       ia = (void *)buf;
4715 +                       ia->ia_size = dst->f_pos;
4716 +                       ia->ia_valid = ATTR_SIZE | ATTR_FILE;
4717 +                       ia->ia_file = dst;
4718 +                       h_mtx = &file_inode(dst)->i_mutex;
4719 +                       mutex_lock_nested(h_mtx, AuLsc_I_CHILD2);
4720 +                       /* no delegation since it is just created */
4721 +                       err = vfsub_notify_change(&dst->f_path, ia,
4722 +                                                 /*delegated*/NULL);
4723 +                       mutex_unlock(h_mtx);
4724 +               }
4725 +       }
4726 +
4727 +       return err;
4728 +}
4729 +
4730 +int au_copy_file(struct file *dst, struct file *src, loff_t len)
4731 +{
4732 +       int err;
4733 +       unsigned long blksize;
4734 +       unsigned char do_kfree;
4735 +       char *buf;
4736 +
4737 +       err = -ENOMEM;
4738 +       blksize = dst->f_path.dentry->d_sb->s_blocksize;
4739 +       if (!blksize || PAGE_SIZE < blksize)
4740 +               blksize = PAGE_SIZE;
4741 +       AuDbg("blksize %lu\n", blksize);
4742 +       do_kfree = (blksize != PAGE_SIZE && blksize >= sizeof(struct iattr *));
4743 +       if (do_kfree)
4744 +               buf = kmalloc(blksize, GFP_NOFS);
4745 +       else
4746 +               buf = (void *)__get_free_page(GFP_NOFS);
4747 +       if (unlikely(!buf))
4748 +               goto out;
4749 +
4750 +       if (len > (1 << 22))
4751 +               AuDbg("copying a large file %lld\n", (long long)len);
4752 +
4753 +       src->f_pos = 0;
4754 +       dst->f_pos = 0;
4755 +       err = au_do_copy_file(dst, src, len, buf, blksize);
4756 +       if (do_kfree)
4757 +               kfree(buf);
4758 +       else
4759 +               free_page((unsigned long)buf);
4760 +
4761 +out:
4762 +       return err;
4763 +}
4764 +
4765 +/*
4766 + * to support a sparse file which is opened with O_APPEND,
4767 + * we need to close the file.
4768 + */
4769 +static int au_cp_regular(struct au_cp_generic *cpg)
4770 +{
4771 +       int err, i;
4772 +       enum { SRC, DST };
4773 +       struct {
4774 +               aufs_bindex_t bindex;
4775 +               unsigned int flags;
4776 +               struct dentry *dentry;
4777 +               int force_wr;
4778 +               struct file *file;
4779 +               void *label;
4780 +       } *f, file[] = {
4781 +               {
4782 +                       .bindex = cpg->bsrc,
4783 +                       .flags = O_RDONLY | O_NOATIME | O_LARGEFILE,
4784 +                       .label = &&out
4785 +               },
4786 +               {
4787 +                       .bindex = cpg->bdst,
4788 +                       .flags = O_WRONLY | O_NOATIME | O_LARGEFILE,
4789 +                       .force_wr = !!au_ftest_cpup(cpg->flags, RWDST),
4790 +                       .label = &&out_src
4791 +               }
4792 +       };
4793 +       struct super_block *sb;
4794 +
4795 +       /* bsrc branch can be ro/rw. */
4796 +       sb = cpg->dentry->d_sb;
4797 +       f = file;
4798 +       for (i = 0; i < 2; i++, f++) {
4799 +               f->dentry = au_h_dptr(cpg->dentry, f->bindex);
4800 +               f->file = au_h_open(cpg->dentry, f->bindex, f->flags,
4801 +                                   /*file*/NULL, f->force_wr);
4802 +               err = PTR_ERR(f->file);
4803 +               if (IS_ERR(f->file))
4804 +                       goto *f->label;
4805 +       }
4806 +
4807 +       /* try stopping to update while we copyup */
4808 +       IMustLock(file[SRC].dentry->d_inode);
4809 +       err = au_copy_file(file[DST].file, file[SRC].file, cpg->len);
4810 +
4811 +       fput(file[DST].file);
4812 +       au_sbr_put(sb, file[DST].bindex);
4813 +
4814 +out_src:
4815 +       fput(file[SRC].file);
4816 +       au_sbr_put(sb, file[SRC].bindex);
4817 +out:
4818 +       return err;
4819 +}
4820 +
4821 +static int au_do_cpup_regular(struct au_cp_generic *cpg,
4822 +                             struct au_cpup_reg_attr *h_src_attr)
4823 +{
4824 +       int err, rerr;
4825 +       loff_t l;
4826 +       struct path h_path;
4827 +       struct inode *h_src_inode, *h_dst_inode;
4828 +
4829 +       err = 0;
4830 +       h_src_inode = au_h_iptr(cpg->dentry->d_inode, cpg->bsrc);
4831 +       l = i_size_read(h_src_inode);
4832 +       if (cpg->len == -1 || l < cpg->len)
4833 +               cpg->len = l;
4834 +       if (cpg->len) {
4835 +               /* try stopping to update while we are referencing */
4836 +               mutex_lock_nested(&h_src_inode->i_mutex, AuLsc_I_CHILD);
4837 +               au_pin_hdir_unlock(cpg->pin);
4838 +
4839 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
4840 +               h_path.mnt = au_sbr_mnt(cpg->dentry->d_sb, cpg->bsrc);
4841 +               h_src_attr->iflags = h_src_inode->i_flags;
4842 +               err = vfs_getattr(&h_path, &h_src_attr->st);
4843 +               if (unlikely(err)) {
4844 +                       mutex_unlock(&h_src_inode->i_mutex);
4845 +                       goto out;
4846 +               }
4847 +               h_src_attr->valid = 1;
4848 +               err = au_cp_regular(cpg);
4849 +               mutex_unlock(&h_src_inode->i_mutex);
4850 +               rerr = au_pin_hdir_relock(cpg->pin);
4851 +               if (!err && rerr)
4852 +                       err = rerr;
4853 +       }
4854 +       if (!err && (h_src_inode->i_state & I_LINKABLE)) {
4855 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bdst);
4856 +               h_dst_inode = h_path.dentry->d_inode;
4857 +               spin_lock(&h_dst_inode->i_lock);
4858 +               h_dst_inode->i_state |= I_LINKABLE;
4859 +               spin_unlock(&h_dst_inode->i_lock);
4860 +       }
4861 +
4862 +out:
4863 +       return err;
4864 +}
4865 +
4866 +static int au_do_cpup_symlink(struct path *h_path, struct dentry *h_src,
4867 +                             struct inode *h_dir)
4868 +{
4869 +       int err, symlen;
4870 +       mm_segment_t old_fs;
4871 +       union {
4872 +               char *k;
4873 +               char __user *u;
4874 +       } sym;
4875 +
4876 +       err = -ENOSYS;
4877 +       if (unlikely(!h_src->d_inode->i_op->readlink))
4878 +               goto out;
4879 +
4880 +       err = -ENOMEM;
4881 +       sym.k = (void *)__get_free_page(GFP_NOFS);
4882 +       if (unlikely(!sym.k))
4883 +               goto out;
4884 +
4885 +       /* unnecessary to support mmap_sem since symlink is not mmap-able */
4886 +       old_fs = get_fs();
4887 +       set_fs(KERNEL_DS);
4888 +       symlen = h_src->d_inode->i_op->readlink(h_src, sym.u, PATH_MAX);
4889 +       err = symlen;
4890 +       set_fs(old_fs);
4891 +
4892 +       if (symlen > 0) {
4893 +               sym.k[symlen] = 0;
4894 +               err = vfsub_symlink(h_dir, h_path, sym.k);
4895 +       }
4896 +       free_page((unsigned long)sym.k);
4897 +
4898 +out:
4899 +       return err;
4900 +}
4901 +
4902 +static noinline_for_stack
4903 +int cpup_entry(struct au_cp_generic *cpg, struct dentry *dst_parent,
4904 +              struct au_cpup_reg_attr *h_src_attr)
4905 +{
4906 +       int err;
4907 +       umode_t mode;
4908 +       unsigned int mnt_flags;
4909 +       unsigned char isdir, isreg, force;
4910 +       const unsigned char do_dt = !!au_ftest_cpup(cpg->flags, DTIME);
4911 +       struct au_dtime dt;
4912 +       struct path h_path;
4913 +       struct dentry *h_src, *h_dst, *h_parent;
4914 +       struct inode *h_inode, *h_dir;
4915 +       struct super_block *sb;
4916 +
4917 +       /* bsrc branch can be ro/rw. */
4918 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
4919 +       h_inode = h_src->d_inode;
4920 +       AuDebugOn(h_inode != au_h_iptr(cpg->dentry->d_inode, cpg->bsrc));
4921 +
4922 +       /* try stopping to be referenced while we are creating */
4923 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
4924 +       if (au_ftest_cpup(cpg->flags, RENAME))
4925 +               AuDebugOn(strncmp(h_dst->d_name.name, AUFS_WH_PFX,
4926 +                                 AUFS_WH_PFX_LEN));
4927 +       h_parent = h_dst->d_parent; /* dir inode is locked */
4928 +       h_dir = h_parent->d_inode;
4929 +       IMustLock(h_dir);
4930 +       AuDebugOn(h_parent != h_dst->d_parent);
4931 +
4932 +       sb = cpg->dentry->d_sb;
4933 +       h_path.mnt = au_sbr_mnt(sb, cpg->bdst);
4934 +       if (do_dt) {
4935 +               h_path.dentry = h_parent;
4936 +               au_dtime_store(&dt, dst_parent, &h_path);
4937 +       }
4938 +       h_path.dentry = h_dst;
4939 +
4940 +       isreg = 0;
4941 +       isdir = 0;
4942 +       mode = h_inode->i_mode;
4943 +       switch (mode & S_IFMT) {
4944 +       case S_IFREG:
4945 +               isreg = 1;
4946 +               err = vfsub_create(h_dir, &h_path, mode | S_IWUSR,
4947 +                                  /*want_excl*/true);
4948 +               if (!err)
4949 +                       err = au_do_cpup_regular(cpg, h_src_attr);
4950 +               break;
4951 +       case S_IFDIR:
4952 +               isdir = 1;
4953 +               err = vfsub_mkdir(h_dir, &h_path, mode);
4954 +               if (!err) {
4955 +                       /*
4956 +                        * strange behaviour from the users view,
4957 +                        * particularry setattr case
4958 +                        */
4959 +                       if (au_ibstart(dst_parent->d_inode) == cpg->bdst)
4960 +                               au_cpup_attr_nlink(dst_parent->d_inode,
4961 +                                                  /*force*/1);
4962 +                       au_cpup_attr_nlink(cpg->dentry->d_inode, /*force*/1);
4963 +               }
4964 +               break;
4965 +       case S_IFLNK:
4966 +               err = au_do_cpup_symlink(&h_path, h_src, h_dir);
4967 +               break;
4968 +       case S_IFCHR:
4969 +       case S_IFBLK:
4970 +               AuDebugOn(!capable(CAP_MKNOD));
4971 +               /*FALLTHROUGH*/
4972 +       case S_IFIFO:
4973 +       case S_IFSOCK:
4974 +               err = vfsub_mknod(h_dir, &h_path, mode, h_inode->i_rdev);
4975 +               break;
4976 +       default:
4977 +               AuIOErr("Unknown inode type 0%o\n", mode);
4978 +               err = -EIO;
4979 +       }
4980 +
4981 +       mnt_flags = au_mntflags(sb);
4982 +       if (!au_opt_test(mnt_flags, UDBA_NONE)
4983 +           && !isdir
4984 +           && au_opt_test(mnt_flags, XINO)
4985 +           && (h_inode->i_nlink == 1
4986 +               || (h_inode->i_state & I_LINKABLE))
4987 +           /* todo: unnecessary? */
4988 +           /* && cpg->dentry->d_inode->i_nlink == 1 */
4989 +           && cpg->bdst < cpg->bsrc
4990 +           && !au_ftest_cpup(cpg->flags, KEEPLINO))
4991 +               au_xino_write(sb, cpg->bsrc, h_inode->i_ino, /*ino*/0);
4992 +               /* ignore this error */
4993 +
4994 +       if (!err) {
4995 +               force = 0;
4996 +               if (isreg) {
4997 +                       force = !!cpg->len;
4998 +                       if (cpg->len == -1)
4999 +                               force = !!i_size_read(h_inode);
5000 +               }
5001 +               au_fhsm_wrote(sb, cpg->bdst, force);
5002 +       }
5003 +
5004 +       if (do_dt)
5005 +               au_dtime_revert(&dt);
5006 +       return err;
5007 +}
5008 +
5009 +static int au_do_ren_after_cpup(struct au_cp_generic *cpg, struct path *h_path)
5010 +{
5011 +       int err;
5012 +       struct dentry *dentry, *h_dentry, *h_parent, *parent;
5013 +       struct inode *h_dir;
5014 +       aufs_bindex_t bdst;
5015 +
5016 +       dentry = cpg->dentry;
5017 +       bdst = cpg->bdst;
5018 +       h_dentry = au_h_dptr(dentry, bdst);
5019 +       if (!au_ftest_cpup(cpg->flags, OVERWRITE)) {
5020 +               dget(h_dentry);
5021 +               au_set_h_dptr(dentry, bdst, NULL);
5022 +               err = au_lkup_neg(dentry, bdst, /*wh*/0);
5023 +               if (!err)
5024 +                       h_path->dentry = dget(au_h_dptr(dentry, bdst));
5025 +               au_set_h_dptr(dentry, bdst, h_dentry);
5026 +       } else {
5027 +               err = 0;
5028 +               parent = dget_parent(dentry);
5029 +               h_parent = au_h_dptr(parent, bdst);
5030 +               dput(parent);
5031 +               h_path->dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
5032 +               if (IS_ERR(h_path->dentry))
5033 +                       err = PTR_ERR(h_path->dentry);
5034 +       }
5035 +       if (unlikely(err))
5036 +               goto out;
5037 +
5038 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
5039 +       h_dir = h_parent->d_inode;
5040 +       IMustLock(h_dir);
5041 +       AuDbg("%pd %pd\n", h_dentry, h_path->dentry);
5042 +       /* no delegation since it is just created */
5043 +       err = vfsub_rename(h_dir, h_dentry, h_dir, h_path, /*delegated*/NULL);
5044 +       dput(h_path->dentry);
5045 +
5046 +out:
5047 +       return err;
5048 +}
5049 +
5050 +/*
5051 + * copyup the @dentry from @bsrc to @bdst.
5052 + * the caller must set the both of lower dentries.
5053 + * @len is for truncating when it is -1 copyup the entire file.
5054 + * in link/rename cases, @dst_parent may be different from the real one.
5055 + * basic->bsrc can be larger than basic->bdst.
5056 + */
5057 +static int au_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5058 +{
5059 +       int err, rerr;
5060 +       aufs_bindex_t old_ibstart;
5061 +       unsigned char isdir, plink;
5062 +       struct dentry *h_src, *h_dst, *h_parent;
5063 +       struct inode *dst_inode, *h_dir, *inode, *delegated;
5064 +       struct super_block *sb;
5065 +       struct au_branch *br;
5066 +       /* to reuduce stack size */
5067 +       struct {
5068 +               struct au_dtime dt;
5069 +               struct path h_path;
5070 +               struct au_cpup_reg_attr h_src_attr;
5071 +       } *a;
5072 +
5073 +       err = -ENOMEM;
5074 +       a = kmalloc(sizeof(*a), GFP_NOFS);
5075 +       if (unlikely(!a))
5076 +               goto out;
5077 +       a->h_src_attr.valid = 0;
5078 +
5079 +       sb = cpg->dentry->d_sb;
5080 +       br = au_sbr(sb, cpg->bdst);
5081 +       a->h_path.mnt = au_br_mnt(br);
5082 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
5083 +       h_parent = h_dst->d_parent; /* dir inode is locked */
5084 +       h_dir = h_parent->d_inode;
5085 +       IMustLock(h_dir);
5086 +
5087 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
5088 +       inode = cpg->dentry->d_inode;
5089 +
5090 +       if (!dst_parent)
5091 +               dst_parent = dget_parent(cpg->dentry);
5092 +       else
5093 +               dget(dst_parent);
5094 +
5095 +       plink = !!au_opt_test(au_mntflags(sb), PLINK);
5096 +       dst_inode = au_h_iptr(inode, cpg->bdst);
5097 +       if (dst_inode) {
5098 +               if (unlikely(!plink)) {
5099 +                       err = -EIO;
5100 +                       AuIOErr("hi%lu(i%lu) exists on b%d "
5101 +                               "but plink is disabled\n",
5102 +                               dst_inode->i_ino, inode->i_ino, cpg->bdst);
5103 +                       goto out_parent;
5104 +               }
5105 +
5106 +               if (dst_inode->i_nlink) {
5107 +                       const int do_dt = au_ftest_cpup(cpg->flags, DTIME);
5108 +
5109 +                       h_src = au_plink_lkup(inode, cpg->bdst);
5110 +                       err = PTR_ERR(h_src);
5111 +                       if (IS_ERR(h_src))
5112 +                               goto out_parent;
5113 +                       if (unlikely(!h_src->d_inode)) {
5114 +                               err = -EIO;
5115 +                               AuIOErr("i%lu exists on a upper branch "
5116 +                                       "but not pseudo-linked\n",
5117 +                                       inode->i_ino);
5118 +                               dput(h_src);
5119 +                               goto out_parent;
5120 +                       }
5121 +
5122 +                       if (do_dt) {
5123 +                               a->h_path.dentry = h_parent;
5124 +                               au_dtime_store(&a->dt, dst_parent, &a->h_path);
5125 +                       }
5126 +
5127 +                       a->h_path.dentry = h_dst;
5128 +                       delegated = NULL;
5129 +                       err = vfsub_link(h_src, h_dir, &a->h_path, &delegated);
5130 +                       if (!err && au_ftest_cpup(cpg->flags, RENAME))
5131 +                               err = au_do_ren_after_cpup(cpg, &a->h_path);
5132 +                       if (do_dt)
5133 +                               au_dtime_revert(&a->dt);
5134 +                       if (unlikely(err == -EWOULDBLOCK)) {
5135 +                               pr_warn("cannot retry for NFSv4 delegation"
5136 +                                       " for an internal link\n");
5137 +                               iput(delegated);
5138 +                       }
5139 +                       dput(h_src);
5140 +                       goto out_parent;
5141 +               } else
5142 +                       /* todo: cpup_wh_file? */
5143 +                       /* udba work */
5144 +                       au_update_ibrange(inode, /*do_put_zero*/1);
5145 +       }
5146 +
5147 +       isdir = S_ISDIR(inode->i_mode);
5148 +       old_ibstart = au_ibstart(inode);
5149 +       err = cpup_entry(cpg, dst_parent, &a->h_src_attr);
5150 +       if (unlikely(err))
5151 +               goto out_rev;
5152 +       dst_inode = h_dst->d_inode;
5153 +       mutex_lock_nested(&dst_inode->i_mutex, AuLsc_I_CHILD2);
5154 +       /* todo: necessary? */
5155 +       /* au_pin_hdir_unlock(cpg->pin); */
5156 +
5157 +       err = cpup_iattr(cpg->dentry, cpg->bdst, h_src, &a->h_src_attr);
5158 +       if (unlikely(err)) {
5159 +               /* todo: necessary? */
5160 +               /* au_pin_hdir_relock(cpg->pin); */ /* ignore an error */
5161 +               mutex_unlock(&dst_inode->i_mutex);
5162 +               goto out_rev;
5163 +       }
5164 +
5165 +       if (cpg->bdst < old_ibstart) {
5166 +               if (S_ISREG(inode->i_mode)) {
5167 +                       err = au_dy_iaop(inode, cpg->bdst, dst_inode);
5168 +                       if (unlikely(err)) {
5169 +                               /* ignore an error */
5170 +                               /* au_pin_hdir_relock(cpg->pin); */
5171 +                               mutex_unlock(&dst_inode->i_mutex);
5172 +                               goto out_rev;
5173 +                       }
5174 +               }
5175 +               au_set_ibstart(inode, cpg->bdst);
5176 +       } else
5177 +               au_set_ibend(inode, cpg->bdst);
5178 +       au_set_h_iptr(inode, cpg->bdst, au_igrab(dst_inode),
5179 +                     au_hi_flags(inode, isdir));
5180 +
5181 +       /* todo: necessary? */
5182 +       /* err = au_pin_hdir_relock(cpg->pin); */
5183 +       mutex_unlock(&dst_inode->i_mutex);
5184 +       if (unlikely(err))
5185 +               goto out_rev;
5186 +
5187 +       if (!isdir
5188 +           && (h_src->d_inode->i_nlink > 1
5189 +               || h_src->d_inode->i_state & I_LINKABLE)
5190 +           && plink)
5191 +               au_plink_append(inode, cpg->bdst, h_dst);
5192 +
5193 +       if (au_ftest_cpup(cpg->flags, RENAME)) {
5194 +               a->h_path.dentry = h_dst;
5195 +               err = au_do_ren_after_cpup(cpg, &a->h_path);
5196 +       }
5197 +       if (!err)
5198 +               goto out_parent; /* success */
5199 +
5200 +       /* revert */
5201 +out_rev:
5202 +       a->h_path.dentry = h_parent;
5203 +       au_dtime_store(&a->dt, dst_parent, &a->h_path);
5204 +       a->h_path.dentry = h_dst;
5205 +       rerr = 0;
5206 +       if (h_dst->d_inode) {
5207 +               if (!isdir) {
5208 +                       /* no delegation since it is just created */
5209 +                       rerr = vfsub_unlink(h_dir, &a->h_path,
5210 +                                           /*delegated*/NULL, /*force*/0);
5211 +               } else
5212 +                       rerr = vfsub_rmdir(h_dir, &a->h_path);
5213 +       }
5214 +       au_dtime_revert(&a->dt);
5215 +       if (rerr) {
5216 +               AuIOErr("failed removing broken entry(%d, %d)\n", err, rerr);
5217 +               err = -EIO;
5218 +       }
5219 +out_parent:
5220 +       dput(dst_parent);
5221 +       kfree(a);
5222 +out:
5223 +       return err;
5224 +}
5225 +
5226 +#if 0 /* unused */
5227 +struct au_cpup_single_args {
5228 +       int *errp;
5229 +       struct au_cp_generic *cpg;
5230 +       struct dentry *dst_parent;
5231 +};
5232 +
5233 +static void au_call_cpup_single(void *args)
5234 +{
5235 +       struct au_cpup_single_args *a = args;
5236 +
5237 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5238 +       *a->errp = au_cpup_single(a->cpg, a->dst_parent);
5239 +       au_pin_hdir_release(a->cpg->pin);
5240 +}
5241 +#endif
5242 +
5243 +/*
5244 + * prevent SIGXFSZ in copy-up.
5245 + * testing CAP_MKNOD is for generic fs,
5246 + * but CAP_FSETID is for xfs only, currently.
5247 + */
5248 +static int au_cpup_sio_test(struct au_pin *pin, umode_t mode)
5249 +{
5250 +       int do_sio;
5251 +       struct super_block *sb;
5252 +       struct inode *h_dir;
5253 +
5254 +       do_sio = 0;
5255 +       sb = au_pinned_parent(pin)->d_sb;
5256 +       if (!au_wkq_test()
5257 +           && (!au_sbi(sb)->si_plink_maint_pid
5258 +               || au_plink_maint(sb, AuLock_NOPLM))) {
5259 +               switch (mode & S_IFMT) {
5260 +               case S_IFREG:
5261 +                       /* no condition about RLIMIT_FSIZE and the file size */
5262 +                       do_sio = 1;
5263 +                       break;
5264 +               case S_IFCHR:
5265 +               case S_IFBLK:
5266 +                       do_sio = !capable(CAP_MKNOD);
5267 +                       break;
5268 +               }
5269 +               if (!do_sio)
5270 +                       do_sio = ((mode & (S_ISUID | S_ISGID))
5271 +                                 && !capable(CAP_FSETID));
5272 +               /* this workaround may be removed in the future */
5273 +               if (!do_sio) {
5274 +                       h_dir = au_pinned_h_dir(pin);
5275 +                       do_sio = h_dir->i_mode & S_ISVTX;
5276 +               }
5277 +       }
5278 +
5279 +       return do_sio;
5280 +}
5281 +
5282 +#if 0 /* unused */
5283 +int au_sio_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
5284 +{
5285 +       int err, wkq_err;
5286 +       struct dentry *h_dentry;
5287 +
5288 +       h_dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
5289 +       if (!au_cpup_sio_test(pin, h_dentry->d_inode->i_mode))
5290 +               err = au_cpup_single(cpg, dst_parent);
5291 +       else {
5292 +               struct au_cpup_single_args args = {
5293 +                       .errp           = &err,
5294 +                       .cpg            = cpg,
5295 +                       .dst_parent     = dst_parent
5296 +               };
5297 +               wkq_err = au_wkq_wait(au_call_cpup_single, &args);
5298 +               if (unlikely(wkq_err))
5299 +                       err = wkq_err;
5300 +       }
5301 +
5302 +       return err;
5303 +}
5304 +#endif
5305 +
5306 +/*
5307 + * copyup the @dentry from the first active lower branch to @bdst,
5308 + * using au_cpup_single().
5309 + */
5310 +static int au_cpup_simple(struct au_cp_generic *cpg)
5311 +{
5312 +       int err;
5313 +       unsigned int flags_orig;
5314 +       struct dentry *dentry;
5315 +
5316 +       AuDebugOn(cpg->bsrc < 0);
5317 +
5318 +       dentry = cpg->dentry;
5319 +       DiMustWriteLock(dentry);
5320 +
5321 +       err = au_lkup_neg(dentry, cpg->bdst, /*wh*/1);
5322 +       if (!err) {
5323 +               flags_orig = cpg->flags;
5324 +               au_fset_cpup(cpg->flags, RENAME);
5325 +               err = au_cpup_single(cpg, NULL);
5326 +               cpg->flags = flags_orig;
5327 +               if (!err)
5328 +                       return 0; /* success */
5329 +
5330 +               /* revert */
5331 +               au_set_h_dptr(dentry, cpg->bdst, NULL);
5332 +               au_set_dbstart(dentry, cpg->bsrc);
5333 +       }
5334 +
5335 +       return err;
5336 +}
5337 +
5338 +struct au_cpup_simple_args {
5339 +       int *errp;
5340 +       struct au_cp_generic *cpg;
5341 +};
5342 +
5343 +static void au_call_cpup_simple(void *args)
5344 +{
5345 +       struct au_cpup_simple_args *a = args;
5346 +
5347 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5348 +       *a->errp = au_cpup_simple(a->cpg);
5349 +       au_pin_hdir_release(a->cpg->pin);
5350 +}
5351 +
5352 +static int au_do_sio_cpup_simple(struct au_cp_generic *cpg)
5353 +{
5354 +       int err, wkq_err;
5355 +       struct dentry *dentry, *parent;
5356 +       struct file *h_file;
5357 +       struct inode *h_dir;
5358 +
5359 +       dentry = cpg->dentry;
5360 +       h_file = NULL;
5361 +       if (au_ftest_cpup(cpg->flags, HOPEN)) {
5362 +               AuDebugOn(cpg->bsrc < 0);
5363 +               h_file = au_h_open_pre(dentry, cpg->bsrc, /*force_wr*/0);
5364 +               err = PTR_ERR(h_file);
5365 +               if (IS_ERR(h_file))
5366 +                       goto out;
5367 +       }
5368 +
5369 +       parent = dget_parent(dentry);
5370 +       h_dir = au_h_iptr(parent->d_inode, cpg->bdst);
5371 +       if (!au_test_h_perm_sio(h_dir, MAY_EXEC | MAY_WRITE)
5372 +           && !au_cpup_sio_test(cpg->pin, dentry->d_inode->i_mode))
5373 +               err = au_cpup_simple(cpg);
5374 +       else {
5375 +               struct au_cpup_simple_args args = {
5376 +                       .errp           = &err,
5377 +                       .cpg            = cpg
5378 +               };
5379 +               wkq_err = au_wkq_wait(au_call_cpup_simple, &args);
5380 +               if (unlikely(wkq_err))
5381 +                       err = wkq_err;
5382 +       }
5383 +
5384 +       dput(parent);
5385 +       if (h_file)
5386 +               au_h_open_post(dentry, cpg->bsrc, h_file);
5387 +
5388 +out:
5389 +       return err;
5390 +}
5391 +
5392 +int au_sio_cpup_simple(struct au_cp_generic *cpg)
5393 +{
5394 +       aufs_bindex_t bsrc, bend;
5395 +       struct dentry *dentry, *h_dentry;
5396 +
5397 +       if (cpg->bsrc < 0) {
5398 +               dentry = cpg->dentry;
5399 +               bend = au_dbend(dentry);
5400 +               for (bsrc = cpg->bdst + 1; bsrc <= bend; bsrc++) {
5401 +                       h_dentry = au_h_dptr(dentry, bsrc);
5402 +                       if (h_dentry) {
5403 +                               AuDebugOn(!h_dentry->d_inode);
5404 +                               break;
5405 +                       }
5406 +               }
5407 +               AuDebugOn(bsrc > bend);
5408 +               cpg->bsrc = bsrc;
5409 +       }
5410 +       AuDebugOn(cpg->bsrc <= cpg->bdst);
5411 +       return au_do_sio_cpup_simple(cpg);
5412 +}
5413 +
5414 +int au_sio_cpdown_simple(struct au_cp_generic *cpg)
5415 +{
5416 +       AuDebugOn(cpg->bdst <= cpg->bsrc);
5417 +       return au_do_sio_cpup_simple(cpg);
5418 +}
5419 +
5420 +/* ---------------------------------------------------------------------- */
5421 +
5422 +/*
5423 + * copyup the deleted file for writing.
5424 + */
5425 +static int au_do_cpup_wh(struct au_cp_generic *cpg, struct dentry *wh_dentry,
5426 +                        struct file *file)
5427 +{
5428 +       int err;
5429 +       unsigned int flags_orig;
5430 +       aufs_bindex_t bsrc_orig;
5431 +       struct dentry *h_d_dst, *h_d_start;
5432 +       struct au_dinfo *dinfo;
5433 +       struct au_hdentry *hdp;
5434 +
5435 +       dinfo = au_di(cpg->dentry);
5436 +       AuRwMustWriteLock(&dinfo->di_rwsem);
5437 +
5438 +       bsrc_orig = cpg->bsrc;
5439 +       cpg->bsrc = dinfo->di_bstart;
5440 +       hdp = dinfo->di_hdentry;
5441 +       h_d_dst = hdp[0 + cpg->bdst].hd_dentry;
5442 +       dinfo->di_bstart = cpg->bdst;
5443 +       hdp[0 + cpg->bdst].hd_dentry = wh_dentry;
5444 +       h_d_start = NULL;
5445 +       if (file) {
5446 +               h_d_start = hdp[0 + cpg->bsrc].hd_dentry;
5447 +               hdp[0 + cpg->bsrc].hd_dentry = au_hf_top(file)->f_path.dentry;
5448 +       }
5449 +       flags_orig = cpg->flags;
5450 +       cpg->flags = !AuCpup_DTIME;
5451 +       err = au_cpup_single(cpg, /*h_parent*/NULL);
5452 +       cpg->flags = flags_orig;
5453 +       if (file) {
5454 +               if (!err)
5455 +                       err = au_reopen_nondir(file);
5456 +               hdp[0 + cpg->bsrc].hd_dentry = h_d_start;
5457 +       }
5458 +       hdp[0 + cpg->bdst].hd_dentry = h_d_dst;
5459 +       dinfo->di_bstart = cpg->bsrc;
5460 +       cpg->bsrc = bsrc_orig;
5461 +
5462 +       return err;
5463 +}
5464 +
5465 +static int au_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5466 +{
5467 +       int err;
5468 +       aufs_bindex_t bdst;
5469 +       struct au_dtime dt;
5470 +       struct dentry *dentry, *parent, *h_parent, *wh_dentry;
5471 +       struct au_branch *br;
5472 +       struct path h_path;
5473 +
5474 +       dentry = cpg->dentry;
5475 +       bdst = cpg->bdst;
5476 +       br = au_sbr(dentry->d_sb, bdst);
5477 +       parent = dget_parent(dentry);
5478 +       h_parent = au_h_dptr(parent, bdst);
5479 +       wh_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
5480 +       err = PTR_ERR(wh_dentry);
5481 +       if (IS_ERR(wh_dentry))
5482 +               goto out;
5483 +
5484 +       h_path.dentry = h_parent;
5485 +       h_path.mnt = au_br_mnt(br);
5486 +       au_dtime_store(&dt, parent, &h_path);
5487 +       err = au_do_cpup_wh(cpg, wh_dentry, file);
5488 +       if (unlikely(err))
5489 +               goto out_wh;
5490 +
5491 +       dget(wh_dentry);
5492 +       h_path.dentry = wh_dentry;
5493 +       if (!d_is_dir(wh_dentry)) {
5494 +               /* no delegation since it is just created */
5495 +               err = vfsub_unlink(h_parent->d_inode, &h_path,
5496 +                                  /*delegated*/NULL, /*force*/0);
5497 +       } else
5498 +               err = vfsub_rmdir(h_parent->d_inode, &h_path);
5499 +       if (unlikely(err)) {
5500 +               AuIOErr("failed remove copied-up tmp file %pd(%d)\n",
5501 +                       wh_dentry, err);
5502 +               err = -EIO;
5503 +       }
5504 +       au_dtime_revert(&dt);
5505 +       au_set_hi_wh(dentry->d_inode, bdst, wh_dentry);
5506 +
5507 +out_wh:
5508 +       dput(wh_dentry);
5509 +out:
5510 +       dput(parent);
5511 +       return err;
5512 +}
5513 +
5514 +struct au_cpup_wh_args {
5515 +       int *errp;
5516 +       struct au_cp_generic *cpg;
5517 +       struct file *file;
5518 +};
5519 +
5520 +static void au_call_cpup_wh(void *args)
5521 +{
5522 +       struct au_cpup_wh_args *a = args;
5523 +
5524 +       au_pin_hdir_acquire_nest(a->cpg->pin);
5525 +       *a->errp = au_cpup_wh(a->cpg, a->file);
5526 +       au_pin_hdir_release(a->cpg->pin);
5527 +}
5528 +
5529 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file)
5530 +{
5531 +       int err, wkq_err;
5532 +       aufs_bindex_t bdst;
5533 +       struct dentry *dentry, *parent, *h_orph, *h_parent;
5534 +       struct inode *dir, *h_dir, *h_tmpdir;
5535 +       struct au_wbr *wbr;
5536 +       struct au_pin wh_pin, *pin_orig;
5537 +
5538 +       dentry = cpg->dentry;
5539 +       bdst = cpg->bdst;
5540 +       parent = dget_parent(dentry);
5541 +       dir = parent->d_inode;
5542 +       h_orph = NULL;
5543 +       h_parent = NULL;
5544 +       h_dir = au_igrab(au_h_iptr(dir, bdst));
5545 +       h_tmpdir = h_dir;
5546 +       pin_orig = NULL;
5547 +       if (!h_dir->i_nlink) {
5548 +               wbr = au_sbr(dentry->d_sb, bdst)->br_wbr;
5549 +               h_orph = wbr->wbr_orph;
5550 +
5551 +               h_parent = dget(au_h_dptr(parent, bdst));
5552 +               au_set_h_dptr(parent, bdst, dget(h_orph));
5553 +               h_tmpdir = h_orph->d_inode;
5554 +               au_set_h_iptr(dir, bdst, au_igrab(h_tmpdir), /*flags*/0);
5555 +
5556 +               mutex_lock_nested(&h_tmpdir->i_mutex, AuLsc_I_PARENT3);
5557 +               /* todo: au_h_open_pre()? */
5558 +
5559 +               pin_orig = cpg->pin;
5560 +               au_pin_init(&wh_pin, dentry, bdst, AuLsc_DI_PARENT,
5561 +                           AuLsc_I_PARENT3, cpg->pin->udba, AuPin_DI_LOCKED);
5562 +               cpg->pin = &wh_pin;
5563 +       }
5564 +
5565 +       if (!au_test_h_perm_sio(h_tmpdir, MAY_EXEC | MAY_WRITE)
5566 +           && !au_cpup_sio_test(cpg->pin, dentry->d_inode->i_mode))
5567 +               err = au_cpup_wh(cpg, file);
5568 +       else {
5569 +               struct au_cpup_wh_args args = {
5570 +                       .errp   = &err,
5571 +                       .cpg    = cpg,
5572 +                       .file   = file
5573 +               };
5574 +               wkq_err = au_wkq_wait(au_call_cpup_wh, &args);
5575 +               if (unlikely(wkq_err))
5576 +                       err = wkq_err;
5577 +       }
5578 +
5579 +       if (h_orph) {
5580 +               mutex_unlock(&h_tmpdir->i_mutex);
5581 +               /* todo: au_h_open_post()? */
5582 +               au_set_h_iptr(dir, bdst, au_igrab(h_dir), /*flags*/0);
5583 +               au_set_h_dptr(parent, bdst, h_parent);
5584 +               AuDebugOn(!pin_orig);
5585 +               cpg->pin = pin_orig;
5586 +       }
5587 +       iput(h_dir);
5588 +       dput(parent);
5589 +
5590 +       return err;
5591 +}
5592 +
5593 +/* ---------------------------------------------------------------------- */
5594 +
5595 +/*
5596 + * generic routine for both of copy-up and copy-down.
5597 + */
5598 +/* cf. revalidate function in file.c */
5599 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
5600 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
5601 +                        struct au_pin *pin,
5602 +                        struct dentry *h_parent, void *arg),
5603 +              void *arg)
5604 +{
5605 +       int err;
5606 +       struct au_pin pin;
5607 +       struct dentry *d, *parent, *h_parent, *real_parent;
5608 +
5609 +       err = 0;
5610 +       parent = dget_parent(dentry);
5611 +       if (IS_ROOT(parent))
5612 +               goto out;
5613 +
5614 +       au_pin_init(&pin, dentry, bdst, AuLsc_DI_PARENT2, AuLsc_I_PARENT2,
5615 +                   au_opt_udba(dentry->d_sb), AuPin_MNT_WRITE);
5616 +
5617 +       /* do not use au_dpage */
5618 +       real_parent = parent;
5619 +       while (1) {
5620 +               dput(parent);
5621 +               parent = dget_parent(dentry);
5622 +               h_parent = au_h_dptr(parent, bdst);
5623 +               if (h_parent)
5624 +                       goto out; /* success */
5625 +
5626 +               /* find top dir which is necessary to cpup */
5627 +               do {
5628 +                       d = parent;
5629 +                       dput(parent);
5630 +                       parent = dget_parent(d);
5631 +                       di_read_lock_parent3(parent, !AuLock_IR);
5632 +                       h_parent = au_h_dptr(parent, bdst);
5633 +                       di_read_unlock(parent, !AuLock_IR);
5634 +               } while (!h_parent);
5635 +
5636 +               if (d != real_parent)
5637 +                       di_write_lock_child3(d);
5638 +
5639 +               /* somebody else might create while we were sleeping */
5640 +               if (!au_h_dptr(d, bdst) || !au_h_dptr(d, bdst)->d_inode) {
5641 +                       if (au_h_dptr(d, bdst))
5642 +                               au_update_dbstart(d);
5643 +
5644 +                       au_pin_set_dentry(&pin, d);
5645 +                       err = au_do_pin(&pin);
5646 +                       if (!err) {
5647 +                               err = cp(d, bdst, &pin, h_parent, arg);
5648 +                               au_unpin(&pin);
5649 +                       }
5650 +               }
5651 +
5652 +               if (d != real_parent)
5653 +                       di_write_unlock(d);
5654 +               if (unlikely(err))
5655 +                       break;
5656 +       }
5657 +
5658 +out:
5659 +       dput(parent);
5660 +       return err;
5661 +}
5662 +
5663 +static int au_cpup_dir(struct dentry *dentry, aufs_bindex_t bdst,
5664 +                      struct au_pin *pin,
5665 +                      struct dentry *h_parent __maybe_unused,
5666 +                      void *arg __maybe_unused)
5667 +{
5668 +       struct au_cp_generic cpg = {
5669 +               .dentry = dentry,
5670 +               .bdst   = bdst,
5671 +               .bsrc   = -1,
5672 +               .len    = 0,
5673 +               .pin    = pin,
5674 +               .flags  = AuCpup_DTIME
5675 +       };
5676 +       return au_sio_cpup_simple(&cpg);
5677 +}
5678 +
5679 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
5680 +{
5681 +       return au_cp_dirs(dentry, bdst, au_cpup_dir, NULL);
5682 +}
5683 +
5684 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
5685 +{
5686 +       int err;
5687 +       struct dentry *parent;
5688 +       struct inode *dir;
5689 +
5690 +       parent = dget_parent(dentry);
5691 +       dir = parent->d_inode;
5692 +       err = 0;
5693 +       if (au_h_iptr(dir, bdst))
5694 +               goto out;
5695 +
5696 +       di_read_unlock(parent, AuLock_IR);
5697 +       di_write_lock_parent(parent);
5698 +       /* someone else might change our inode while we were sleeping */
5699 +       if (!au_h_iptr(dir, bdst))
5700 +               err = au_cpup_dirs(dentry, bdst);
5701 +       di_downgrade_lock(parent, AuLock_IR);
5702 +
5703 +out:
5704 +       dput(parent);
5705 +       return err;
5706 +}
5707 diff -urN /usr/share/empty/fs/aufs/cpup.h linux/fs/aufs/cpup.h
5708 --- /usr/share/empty/fs/aufs/cpup.h     1970-01-01 01:00:00.000000000 +0100
5709 +++ linux/fs/aufs/cpup.h        2015-03-27 21:56:35.460128334 +0100
5710 @@ -0,0 +1,94 @@
5711 +/*
5712 + * Copyright (C) 2005-2015 Junjiro R. Okajima
5713 + *
5714 + * This program, aufs is free software; you can redistribute it and/or modify
5715 + * it under the terms of the GNU General Public License as published by
5716 + * the Free Software Foundation; either version 2 of the License, or
5717 + * (at your option) any later version.
5718 + *
5719 + * This program is distributed in the hope that it will be useful,
5720 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5721 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5722 + * GNU General Public License for more details.
5723 + *
5724 + * You should have received a copy of the GNU General Public License
5725 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5726 + */
5727 +
5728 +/*
5729 + * copy-up/down functions
5730 + */
5731 +
5732 +#ifndef __AUFS_CPUP_H__
5733 +#define __AUFS_CPUP_H__
5734 +
5735 +#ifdef __KERNEL__
5736 +
5737 +#include <linux/path.h>
5738 +
5739 +struct inode;
5740 +struct file;
5741 +struct au_pin;
5742 +
5743 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags);
5744 +void au_cpup_attr_timesizes(struct inode *inode);
5745 +void au_cpup_attr_nlink(struct inode *inode, int force);
5746 +void au_cpup_attr_changeable(struct inode *inode);
5747 +void au_cpup_igen(struct inode *inode, struct inode *h_inode);
5748 +void au_cpup_attr_all(struct inode *inode, int force);
5749 +
5750 +/* ---------------------------------------------------------------------- */
5751 +
5752 +struct au_cp_generic {
5753 +       struct dentry   *dentry;
5754 +       aufs_bindex_t   bdst, bsrc;
5755 +       loff_t          len;
5756 +       struct au_pin   *pin;
5757 +       unsigned int    flags;
5758 +};
5759 +
5760 +/* cpup flags */
5761 +#define AuCpup_DTIME           1               /* do dtime_store/revert */
5762 +#define AuCpup_KEEPLINO                (1 << 1)        /* do not clear the lower xino,
5763 +                                                  for link(2) */
5764 +#define AuCpup_RENAME          (1 << 2)        /* rename after cpup */
5765 +#define AuCpup_HOPEN           (1 << 3)        /* call h_open_pre/post() in
5766 +                                                  cpup */
5767 +#define AuCpup_OVERWRITE       (1 << 4)        /* allow overwriting the
5768 +                                                  existing entry */
5769 +#define AuCpup_RWDST           (1 << 5)        /* force write target even if
5770 +                                                  the branch is marked as RO */
5771 +
5772 +#define au_ftest_cpup(flags, name)     ((flags) & AuCpup_##name)
5773 +#define au_fset_cpup(flags, name) \
5774 +       do { (flags) |= AuCpup_##name; } while (0)
5775 +#define au_fclr_cpup(flags, name) \
5776 +       do { (flags) &= ~AuCpup_##name; } while (0)
5777 +
5778 +int au_copy_file(struct file *dst, struct file *src, loff_t len);
5779 +int au_sio_cpup_simple(struct au_cp_generic *cpg);
5780 +int au_sio_cpdown_simple(struct au_cp_generic *cpg);
5781 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file);
5782 +
5783 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
5784 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
5785 +                        struct au_pin *pin,
5786 +                        struct dentry *h_parent, void *arg),
5787 +              void *arg);
5788 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
5789 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
5790 +
5791 +/* ---------------------------------------------------------------------- */
5792 +
5793 +/* keep timestamps when copyup */
5794 +struct au_dtime {
5795 +       struct dentry *dt_dentry;
5796 +       struct path dt_h_path;
5797 +       struct timespec dt_atime, dt_mtime;
5798 +};
5799 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
5800 +                   struct path *h_path);
5801 +void au_dtime_revert(struct au_dtime *dt);
5802 +
5803 +#endif /* __KERNEL__ */
5804 +#endif /* __AUFS_CPUP_H__ */
5805 diff -urN /usr/share/empty/fs/aufs/dbgaufs.c linux/fs/aufs/dbgaufs.c
5806 --- /usr/share/empty/fs/aufs/dbgaufs.c  1970-01-01 01:00:00.000000000 +0100
5807 +++ linux/fs/aufs/dbgaufs.c     2015-03-27 21:56:35.463461668 +0100
5808 @@ -0,0 +1,432 @@
5809 +/*
5810 + * Copyright (C) 2005-2015 Junjiro R. Okajima
5811 + *
5812 + * This program, aufs is free software; you can redistribute it and/or modify
5813 + * it under the terms of the GNU General Public License as published by
5814 + * the Free Software Foundation; either version 2 of the License, or
5815 + * (at your option) any later version.
5816 + *
5817 + * This program is distributed in the hope that it will be useful,
5818 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5819 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5820 + * GNU General Public License for more details.
5821 + *
5822 + * You should have received a copy of the GNU General Public License
5823 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5824 + */
5825 +
5826 +/*
5827 + * debugfs interface
5828 + */
5829 +
5830 +#include <linux/debugfs.h>
5831 +#include "aufs.h"
5832 +
5833 +#ifndef CONFIG_SYSFS
5834 +#error DEBUG_FS depends upon SYSFS
5835 +#endif
5836 +
5837 +static struct dentry *dbgaufs;
5838 +static const mode_t dbgaufs_mode = S_IRUSR | S_IRGRP | S_IROTH;
5839 +
5840 +/* 20 is max digits length of ulong 64 */
5841 +struct dbgaufs_arg {
5842 +       int n;
5843 +       char a[20 * 4];
5844 +};
5845 +
5846 +/*
5847 + * common function for all XINO files
5848 + */
5849 +static int dbgaufs_xi_release(struct inode *inode __maybe_unused,
5850 +                             struct file *file)
5851 +{
5852 +       kfree(file->private_data);
5853 +       return 0;
5854 +}
5855 +
5856 +static int dbgaufs_xi_open(struct file *xf, struct file *file, int do_fcnt)
5857 +{
5858 +       int err;
5859 +       struct kstat st;
5860 +       struct dbgaufs_arg *p;
5861 +
5862 +       err = -ENOMEM;
5863 +       p = kmalloc(sizeof(*p), GFP_NOFS);
5864 +       if (unlikely(!p))
5865 +               goto out;
5866 +
5867 +       err = 0;
5868 +       p->n = 0;
5869 +       file->private_data = p;
5870 +       if (!xf)
5871 +               goto out;
5872 +
5873 +       err = vfs_getattr(&xf->f_path, &st);
5874 +       if (!err) {
5875 +               if (do_fcnt)
5876 +                       p->n = snprintf
5877 +                               (p->a, sizeof(p->a), "%ld, %llux%lu %lld\n",
5878 +                                (long)file_count(xf), st.blocks, st.blksize,
5879 +                                (long long)st.size);
5880 +               else
5881 +                       p->n = snprintf(p->a, sizeof(p->a), "%llux%lu %lld\n",
5882 +                                       st.blocks, st.blksize,
5883 +                                       (long long)st.size);
5884 +               AuDebugOn(p->n >= sizeof(p->a));
5885 +       } else {
5886 +               p->n = snprintf(p->a, sizeof(p->a), "err %d\n", err);
5887 +               err = 0;
5888 +       }
5889 +
5890 +out:
5891 +       return err;
5892 +
5893 +}
5894 +
5895 +static ssize_t dbgaufs_xi_read(struct file *file, char __user *buf,
5896 +                              size_t count, loff_t *ppos)
5897 +{
5898 +       struct dbgaufs_arg *p;
5899 +
5900 +       p = file->private_data;
5901 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
5902 +}
5903 +
5904 +/* ---------------------------------------------------------------------- */
5905 +
5906 +struct dbgaufs_plink_arg {
5907 +       int n;
5908 +       char a[];
5909 +};
5910 +
5911 +static int dbgaufs_plink_release(struct inode *inode __maybe_unused,
5912 +                                struct file *file)
5913 +{
5914 +       free_page((unsigned long)file->private_data);
5915 +       return 0;
5916 +}
5917 +
5918 +static int dbgaufs_plink_open(struct inode *inode, struct file *file)
5919 +{
5920 +       int err, i, limit;
5921 +       unsigned long n, sum;
5922 +       struct dbgaufs_plink_arg *p;
5923 +       struct au_sbinfo *sbinfo;
5924 +       struct super_block *sb;
5925 +       struct au_sphlhead *sphl;
5926 +
5927 +       err = -ENOMEM;
5928 +       p = (void *)get_zeroed_page(GFP_NOFS);
5929 +       if (unlikely(!p))
5930 +               goto out;
5931 +
5932 +       err = -EFBIG;
5933 +       sbinfo = inode->i_private;
5934 +       sb = sbinfo->si_sb;
5935 +       si_noflush_read_lock(sb);
5936 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
5937 +               limit = PAGE_SIZE - sizeof(p->n);
5938 +
5939 +               /* the number of buckets */
5940 +               n = snprintf(p->a + p->n, limit, "%d\n", AuPlink_NHASH);
5941 +               p->n += n;
5942 +               limit -= n;
5943 +
5944 +               sum = 0;
5945 +               for (i = 0, sphl = sbinfo->si_plink;
5946 +                    i < AuPlink_NHASH;
5947 +                    i++, sphl++) {
5948 +                       n = au_sphl_count(sphl);
5949 +                       sum += n;
5950 +
5951 +                       n = snprintf(p->a + p->n, limit, "%lu ", n);
5952 +                       p->n += n;
5953 +                       limit -= n;
5954 +                       if (unlikely(limit <= 0))
5955 +                               goto out_free;
5956 +               }
5957 +               p->a[p->n - 1] = '\n';
5958 +
5959 +               /* the sum of plinks */
5960 +               n = snprintf(p->a + p->n, limit, "%lu\n", sum);
5961 +               p->n += n;
5962 +               limit -= n;
5963 +               if (unlikely(limit <= 0))
5964 +                       goto out_free;
5965 +       } else {
5966 +#define str "1\n0\n0\n"
5967 +               p->n = sizeof(str) - 1;
5968 +               strcpy(p->a, str);
5969 +#undef str
5970 +       }
5971 +       si_read_unlock(sb);
5972 +
5973 +       err = 0;
5974 +       file->private_data = p;
5975 +       goto out; /* success */
5976 +
5977 +out_free:
5978 +       free_page((unsigned long)p);
5979 +out:
5980 +       return err;
5981 +}
5982 +
5983 +static ssize_t dbgaufs_plink_read(struct file *file, char __user *buf,
5984 +                                 size_t count, loff_t *ppos)
5985 +{
5986 +       struct dbgaufs_plink_arg *p;
5987 +
5988 +       p = file->private_data;
5989 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
5990 +}
5991 +
5992 +static const struct file_operations dbgaufs_plink_fop = {
5993 +       .owner          = THIS_MODULE,
5994 +       .open           = dbgaufs_plink_open,
5995 +       .release        = dbgaufs_plink_release,
5996 +       .read           = dbgaufs_plink_read
5997 +};
5998 +
5999 +/* ---------------------------------------------------------------------- */
6000 +
6001 +static int dbgaufs_xib_open(struct inode *inode, struct file *file)
6002 +{
6003 +       int err;
6004 +       struct au_sbinfo *sbinfo;
6005 +       struct super_block *sb;
6006 +
6007 +       sbinfo = inode->i_private;
6008 +       sb = sbinfo->si_sb;
6009 +       si_noflush_read_lock(sb);
6010 +       err = dbgaufs_xi_open(sbinfo->si_xib, file, /*do_fcnt*/0);
6011 +       si_read_unlock(sb);
6012 +       return err;
6013 +}
6014 +
6015 +static const struct file_operations dbgaufs_xib_fop = {
6016 +       .owner          = THIS_MODULE,
6017 +       .open           = dbgaufs_xib_open,
6018 +       .release        = dbgaufs_xi_release,
6019 +       .read           = dbgaufs_xi_read
6020 +};
6021 +
6022 +/* ---------------------------------------------------------------------- */
6023 +
6024 +#define DbgaufsXi_PREFIX "xi"
6025 +
6026 +static int dbgaufs_xino_open(struct inode *inode, struct file *file)
6027 +{
6028 +       int err;
6029 +       long l;
6030 +       struct au_sbinfo *sbinfo;
6031 +       struct super_block *sb;
6032 +       struct file *xf;
6033 +       struct qstr *name;
6034 +
6035 +       err = -ENOENT;
6036 +       xf = NULL;
6037 +       name = &file->f_path.dentry->d_name;
6038 +       if (unlikely(name->len < sizeof(DbgaufsXi_PREFIX)
6039 +                    || memcmp(name->name, DbgaufsXi_PREFIX,
6040 +                              sizeof(DbgaufsXi_PREFIX) - 1)))
6041 +               goto out;
6042 +       err = kstrtol(name->name + sizeof(DbgaufsXi_PREFIX) - 1, 10, &l);
6043 +       if (unlikely(err))
6044 +               goto out;
6045 +
6046 +       sbinfo = inode->i_private;
6047 +       sb = sbinfo->si_sb;
6048 +       si_noflush_read_lock(sb);
6049 +       if (l <= au_sbend(sb)) {
6050 +               xf = au_sbr(sb, (aufs_bindex_t)l)->br_xino.xi_file;
6051 +               err = dbgaufs_xi_open(xf, file, /*do_fcnt*/1);
6052 +       } else
6053 +               err = -ENOENT;
6054 +       si_read_unlock(sb);
6055 +
6056 +out:
6057 +       return err;
6058 +}
6059 +
6060 +static const struct file_operations dbgaufs_xino_fop = {
6061 +       .owner          = THIS_MODULE,
6062 +       .open           = dbgaufs_xino_open,
6063 +       .release        = dbgaufs_xi_release,
6064 +       .read           = dbgaufs_xi_read
6065 +};
6066 +
6067 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
6068 +{
6069 +       aufs_bindex_t bend;
6070 +       struct au_branch *br;
6071 +       struct au_xino_file *xi;
6072 +
6073 +       if (!au_sbi(sb)->si_dbgaufs)
6074 +               return;
6075 +
6076 +       bend = au_sbend(sb);
6077 +       for (; bindex <= bend; bindex++) {
6078 +               br = au_sbr(sb, bindex);
6079 +               xi = &br->br_xino;
6080 +               debugfs_remove(xi->xi_dbgaufs);
6081 +               xi->xi_dbgaufs = NULL;
6082 +       }
6083 +}
6084 +
6085 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
6086 +{
6087 +       struct au_sbinfo *sbinfo;
6088 +       struct dentry *parent;
6089 +       struct au_branch *br;
6090 +       struct au_xino_file *xi;
6091 +       aufs_bindex_t bend;
6092 +       char name[sizeof(DbgaufsXi_PREFIX) + 5]; /* "xi" bindex NULL */
6093 +
6094 +       sbinfo = au_sbi(sb);
6095 +       parent = sbinfo->si_dbgaufs;
6096 +       if (!parent)
6097 +               return;
6098 +
6099 +       bend = au_sbend(sb);
6100 +       for (; bindex <= bend; bindex++) {
6101 +               snprintf(name, sizeof(name), DbgaufsXi_PREFIX "%d", bindex);
6102 +               br = au_sbr(sb, bindex);
6103 +               xi = &br->br_xino;
6104 +               AuDebugOn(xi->xi_dbgaufs);
6105 +               xi->xi_dbgaufs = debugfs_create_file(name, dbgaufs_mode, parent,
6106 +                                                    sbinfo, &dbgaufs_xino_fop);
6107 +               /* ignore an error */
6108 +               if (unlikely(!xi->xi_dbgaufs))
6109 +                       AuWarn1("failed %s under debugfs\n", name);
6110 +       }
6111 +}
6112 +
6113 +/* ---------------------------------------------------------------------- */
6114 +
6115 +#ifdef CONFIG_AUFS_EXPORT
6116 +static int dbgaufs_xigen_open(struct inode *inode, struct file *file)
6117 +{
6118 +       int err;
6119 +       struct au_sbinfo *sbinfo;
6120 +       struct super_block *sb;
6121 +
6122 +       sbinfo = inode->i_private;
6123 +       sb = sbinfo->si_sb;
6124 +       si_noflush_read_lock(sb);
6125 +       err = dbgaufs_xi_open(sbinfo->si_xigen, file, /*do_fcnt*/0);
6126 +       si_read_unlock(sb);
6127 +       return err;
6128 +}
6129 +
6130 +static const struct file_operations dbgaufs_xigen_fop = {
6131 +       .owner          = THIS_MODULE,
6132 +       .open           = dbgaufs_xigen_open,
6133 +       .release        = dbgaufs_xi_release,
6134 +       .read           = dbgaufs_xi_read
6135 +};
6136 +
6137 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6138 +{
6139 +       int err;
6140 +
6141 +       /*
6142 +        * This function is a dynamic '__init' function actually,
6143 +        * so the tiny check for si_rwsem is unnecessary.
6144 +        */
6145 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6146 +
6147 +       err = -EIO;
6148 +       sbinfo->si_dbgaufs_xigen = debugfs_create_file
6149 +               ("xigen", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6150 +                &dbgaufs_xigen_fop);
6151 +       if (sbinfo->si_dbgaufs_xigen)
6152 +               err = 0;
6153 +
6154 +       return err;
6155 +}
6156 +#else
6157 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
6158 +{
6159 +       return 0;
6160 +}
6161 +#endif /* CONFIG_AUFS_EXPORT */
6162 +
6163 +/* ---------------------------------------------------------------------- */
6164 +
6165 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo)
6166 +{
6167 +       /*
6168 +        * This function is a dynamic '__init' function actually,
6169 +        * so the tiny check for si_rwsem is unnecessary.
6170 +        */
6171 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6172 +
6173 +       debugfs_remove_recursive(sbinfo->si_dbgaufs);
6174 +       sbinfo->si_dbgaufs = NULL;
6175 +       kobject_put(&sbinfo->si_kobj);
6176 +}
6177 +
6178 +int dbgaufs_si_init(struct au_sbinfo *sbinfo)
6179 +{
6180 +       int err;
6181 +       char name[SysaufsSiNameLen];
6182 +
6183 +       /*
6184 +        * This function is a dynamic '__init' function actually,
6185 +        * so the tiny check for si_rwsem is unnecessary.
6186 +        */
6187 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
6188 +
6189 +       err = -ENOENT;
6190 +       if (!dbgaufs) {
6191 +               AuErr1("/debug/aufs is uninitialized\n");
6192 +               goto out;
6193 +       }
6194 +
6195 +       err = -EIO;
6196 +       sysaufs_name(sbinfo, name);
6197 +       sbinfo->si_dbgaufs = debugfs_create_dir(name, dbgaufs);
6198 +       if (unlikely(!sbinfo->si_dbgaufs))
6199 +               goto out;
6200 +       kobject_get(&sbinfo->si_kobj);
6201 +
6202 +       sbinfo->si_dbgaufs_xib = debugfs_create_file
6203 +               ("xib", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6204 +                &dbgaufs_xib_fop);
6205 +       if (unlikely(!sbinfo->si_dbgaufs_xib))
6206 +               goto out_dir;
6207 +
6208 +       sbinfo->si_dbgaufs_plink = debugfs_create_file
6209 +               ("plink", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
6210 +                &dbgaufs_plink_fop);
6211 +       if (unlikely(!sbinfo->si_dbgaufs_plink))
6212 +               goto out_dir;
6213 +
6214 +       err = dbgaufs_xigen_init(sbinfo);
6215 +       if (!err)
6216 +               goto out; /* success */
6217 +
6218 +out_dir:
6219 +       dbgaufs_si_fin(sbinfo);
6220 +out:
6221 +       return err;
6222 +}
6223 +
6224 +/* ---------------------------------------------------------------------- */
6225 +
6226 +void dbgaufs_fin(void)
6227 +{
6228 +       debugfs_remove(dbgaufs);
6229 +}
6230 +
6231 +int __init dbgaufs_init(void)
6232 +{
6233 +       int err;
6234 +
6235 +       err = -EIO;
6236 +       dbgaufs = debugfs_create_dir(AUFS_NAME, NULL);
6237 +       if (dbgaufs)
6238 +               err = 0;
6239 +       return err;
6240 +}
6241 diff -urN /usr/share/empty/fs/aufs/dbgaufs.h linux/fs/aufs/dbgaufs.h
6242 --- /usr/share/empty/fs/aufs/dbgaufs.h  1970-01-01 01:00:00.000000000 +0100
6243 +++ linux/fs/aufs/dbgaufs.h     2015-03-27 21:56:35.463461668 +0100
6244 @@ -0,0 +1,48 @@
6245 +/*
6246 + * Copyright (C) 2005-2015 Junjiro R. Okajima
6247 + *
6248 + * This program, aufs is free software; you can redistribute it and/or modify
6249 + * it under the terms of the GNU General Public License as published by
6250 + * the Free Software Foundation; either version 2 of the License, or
6251 + * (at your option) any later version.
6252 + *
6253 + * This program is distributed in the hope that it will be useful,
6254 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6255 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6256 + * GNU General Public License for more details.
6257 + *
6258 + * You should have received a copy of the GNU General Public License
6259 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6260 + */
6261 +
6262 +/*
6263 + * debugfs interface
6264 + */
6265 +
6266 +#ifndef __DBGAUFS_H__
6267 +#define __DBGAUFS_H__
6268 +
6269 +#ifdef __KERNEL__
6270 +
6271 +struct super_block;
6272 +struct au_sbinfo;
6273 +
6274 +#ifdef CONFIG_DEBUG_FS
6275 +/* dbgaufs.c */
6276 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
6277 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
6278 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo);
6279 +int dbgaufs_si_init(struct au_sbinfo *sbinfo);
6280 +void dbgaufs_fin(void);
6281 +int __init dbgaufs_init(void);
6282 +#else
6283 +AuStubVoid(dbgaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
6284 +AuStubVoid(dbgaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
6285 +AuStubVoid(dbgaufs_si_fin, struct au_sbinfo *sbinfo)
6286 +AuStubInt0(dbgaufs_si_init, struct au_sbinfo *sbinfo)
6287 +AuStubVoid(dbgaufs_fin, void)
6288 +AuStubInt0(__init dbgaufs_init, void)
6289 +#endif /* CONFIG_DEBUG_FS */
6290 +
6291 +#endif /* __KERNEL__ */
6292 +#endif /* __DBGAUFS_H__ */
6293 diff -urN /usr/share/empty/fs/aufs/dcsub.c linux/fs/aufs/dcsub.c
6294 --- /usr/share/empty/fs/aufs/dcsub.c    1970-01-01 01:00:00.000000000 +0100
6295 +++ linux/fs/aufs/dcsub.c       2015-03-27 21:56:35.463461668 +0100
6296 @@ -0,0 +1,224 @@
6297 +/*
6298 + * Copyright (C) 2005-2015 Junjiro R. Okajima
6299 + *
6300 + * This program, aufs is free software; you can redistribute it and/or modify
6301 + * it under the terms of the GNU General Public License as published by
6302 + * the Free Software Foundation; either version 2 of the License, or
6303 + * (at your option) any later version.
6304 + *
6305 + * This program is distributed in the hope that it will be useful,
6306 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6307 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6308 + * GNU General Public License for more details.
6309 + *
6310 + * You should have received a copy of the GNU General Public License
6311 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6312 + */
6313 +
6314 +/*
6315 + * sub-routines for dentry cache
6316 + */
6317 +
6318 +#include "aufs.h"
6319 +
6320 +static void au_dpage_free(struct au_dpage *dpage)
6321 +{
6322 +       int i;
6323 +       struct dentry **p;
6324 +
6325 +       p = dpage->dentries;
6326 +       for (i = 0; i < dpage->ndentry; i++)
6327 +               dput(*p++);
6328 +       free_page((unsigned long)dpage->dentries);
6329 +}
6330 +
6331 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp)
6332 +{
6333 +       int err;
6334 +       void *p;
6335 +
6336 +       err = -ENOMEM;
6337 +       dpages->dpages = kmalloc(sizeof(*dpages->dpages), gfp);
6338 +       if (unlikely(!dpages->dpages))
6339 +               goto out;
6340 +
6341 +       p = (void *)__get_free_page(gfp);
6342 +       if (unlikely(!p))
6343 +               goto out_dpages;
6344 +
6345 +       dpages->dpages[0].ndentry = 0;
6346 +       dpages->dpages[0].dentries = p;
6347 +       dpages->ndpage = 1;
6348 +       return 0; /* success */
6349 +
6350 +out_dpages:
6351 +       kfree(dpages->dpages);
6352 +out:
6353 +       return err;
6354 +}
6355 +
6356 +void au_dpages_free(struct au_dcsub_pages *dpages)
6357 +{
6358 +       int i;
6359 +       struct au_dpage *p;
6360 +
6361 +       p = dpages->dpages;
6362 +       for (i = 0; i < dpages->ndpage; i++)
6363 +               au_dpage_free(p++);
6364 +       kfree(dpages->dpages);
6365 +}
6366 +
6367 +static int au_dpages_append(struct au_dcsub_pages *dpages,
6368 +                           struct dentry *dentry, gfp_t gfp)
6369 +{
6370 +       int err, sz;
6371 +       struct au_dpage *dpage;
6372 +       void *p;
6373 +
6374 +       dpage = dpages->dpages + dpages->ndpage - 1;
6375 +       sz = PAGE_SIZE / sizeof(dentry);
6376 +       if (unlikely(dpage->ndentry >= sz)) {
6377 +               AuLabel(new dpage);
6378 +               err = -ENOMEM;
6379 +               sz = dpages->ndpage * sizeof(*dpages->dpages);
6380 +               p = au_kzrealloc(dpages->dpages, sz,
6381 +                                sz + sizeof(*dpages->dpages), gfp);
6382 +               if (unlikely(!p))
6383 +                       goto out;
6384 +
6385 +               dpages->dpages = p;
6386 +               dpage = dpages->dpages + dpages->ndpage;
6387 +               p = (void *)__get_free_page(gfp);
6388 +               if (unlikely(!p))
6389 +                       goto out;
6390 +
6391 +               dpage->ndentry = 0;
6392 +               dpage->dentries = p;
6393 +               dpages->ndpage++;
6394 +       }
6395 +
6396 +       AuDebugOn(au_dcount(dentry) <= 0);
6397 +       dpage->dentries[dpage->ndentry++] = dget_dlock(dentry);
6398 +       return 0; /* success */
6399 +
6400 +out:
6401 +       return err;
6402 +}
6403 +
6404 +/* todo: BAD approach */
6405 +/* copied from linux/fs/dcache.c */
6406 +enum d_walk_ret {
6407 +       D_WALK_CONTINUE,
6408 +       D_WALK_QUIT,
6409 +       D_WALK_NORETRY,
6410 +       D_WALK_SKIP,
6411 +};
6412 +
6413 +extern void d_walk(struct dentry *parent, void *data,
6414 +                  enum d_walk_ret (*enter)(void *, struct dentry *),
6415 +                  void (*finish)(void *));
6416 +
6417 +struct ac_dpages_arg {
6418 +       int err;
6419 +       struct au_dcsub_pages *dpages;
6420 +       struct super_block *sb;
6421 +       au_dpages_test test;
6422 +       void *arg;
6423 +};
6424 +
6425 +static enum d_walk_ret au_call_dpages_append(void *_arg, struct dentry *dentry)
6426 +{
6427 +       enum d_walk_ret ret;
6428 +       struct ac_dpages_arg *arg = _arg;
6429 +
6430 +       ret = D_WALK_CONTINUE;
6431 +       if (dentry->d_sb == arg->sb
6432 +           && !IS_ROOT(dentry)
6433 +           && au_dcount(dentry) > 0
6434 +           && au_di(dentry)
6435 +           && (!arg->test || arg->test(dentry, arg->arg))) {
6436 +               arg->err = au_dpages_append(arg->dpages, dentry, GFP_ATOMIC);
6437 +               if (unlikely(arg->err))
6438 +                       ret = D_WALK_QUIT;
6439 +       }
6440 +
6441 +       return ret;
6442 +}
6443 +
6444 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6445 +                  au_dpages_test test, void *arg)
6446 +{
6447 +       struct ac_dpages_arg args = {
6448 +               .err    = 0,
6449 +               .dpages = dpages,
6450 +               .sb     = root->d_sb,
6451 +               .test   = test,
6452 +               .arg    = arg
6453 +       };
6454 +
6455 +       d_walk(root, &args, au_call_dpages_append, NULL);
6456 +
6457 +       return args.err;
6458 +}
6459 +
6460 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6461 +                      int do_include, au_dpages_test test, void *arg)
6462 +{
6463 +       int err;
6464 +
6465 +       err = 0;
6466 +       write_seqlock(&rename_lock);
6467 +       spin_lock(&dentry->d_lock);
6468 +       if (do_include
6469 +           && au_dcount(dentry) > 0
6470 +           && (!test || test(dentry, arg)))
6471 +               err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6472 +       spin_unlock(&dentry->d_lock);
6473 +       if (unlikely(err))
6474 +               goto out;
6475 +
6476 +       /*
6477 +        * RCU for vfsmount is unnecessary since this is a traverse in a single
6478 +        * mount
6479 +        */
6480 +       while (!IS_ROOT(dentry)) {
6481 +               dentry = dentry->d_parent; /* rename_lock is locked */
6482 +               spin_lock(&dentry->d_lock);
6483 +               if (au_dcount(dentry) > 0
6484 +                   && (!test || test(dentry, arg)))
6485 +                       err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
6486 +               spin_unlock(&dentry->d_lock);
6487 +               if (unlikely(err))
6488 +                       break;
6489 +       }
6490 +
6491 +out:
6492 +       write_sequnlock(&rename_lock);
6493 +       return err;
6494 +}
6495 +
6496 +static inline int au_dcsub_dpages_aufs(struct dentry *dentry, void *arg)
6497 +{
6498 +       return au_di(dentry) && dentry->d_sb == arg;
6499 +}
6500 +
6501 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6502 +                           struct dentry *dentry, int do_include)
6503 +{
6504 +       return au_dcsub_pages_rev(dpages, dentry, do_include,
6505 +                                 au_dcsub_dpages_aufs, dentry->d_sb);
6506 +}
6507 +
6508 +int au_test_subdir(struct dentry *d1, struct dentry *d2)
6509 +{
6510 +       struct path path[2] = {
6511 +               {
6512 +                       .dentry = d1
6513 +               },
6514 +               {
6515 +                       .dentry = d2
6516 +               }
6517 +       };
6518 +
6519 +       return path_is_under(path + 0, path + 1);
6520 +}
6521 diff -urN /usr/share/empty/fs/aufs/dcsub.h linux/fs/aufs/dcsub.h
6522 --- /usr/share/empty/fs/aufs/dcsub.h    1970-01-01 01:00:00.000000000 +0100
6523 +++ linux/fs/aufs/dcsub.h       2015-03-27 21:56:35.463461668 +0100
6524 @@ -0,0 +1,125 @@
6525 +/*
6526 + * Copyright (C) 2005-2015 Junjiro R. Okajima
6527 + *
6528 + * This program, aufs is free software; you can redistribute it and/or modify
6529 + * it under the terms of the GNU General Public License as published by
6530 + * the Free Software Foundation; either version 2 of the License, or
6531 + * (at your option) any later version.
6532 + *
6533 + * This program is distributed in the hope that it will be useful,
6534 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6535 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6536 + * GNU General Public License for more details.
6537 + *
6538 + * You should have received a copy of the GNU General Public License
6539 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6540 + */
6541 +
6542 +/*
6543 + * sub-routines for dentry cache
6544 + */
6545 +
6546 +#ifndef __AUFS_DCSUB_H__
6547 +#define __AUFS_DCSUB_H__
6548 +
6549 +#ifdef __KERNEL__
6550 +
6551 +#include <linux/dcache.h>
6552 +#include <linux/fs.h>
6553 +
6554 +struct dentry;
6555 +
6556 +struct au_dpage {
6557 +       int ndentry;
6558 +       struct dentry **dentries;
6559 +};
6560 +
6561 +struct au_dcsub_pages {
6562 +       int ndpage;
6563 +       struct au_dpage *dpages;
6564 +};
6565 +
6566 +/* ---------------------------------------------------------------------- */
6567 +
6568 +/* dcsub.c */
6569 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp);
6570 +void au_dpages_free(struct au_dcsub_pages *dpages);
6571 +typedef int (*au_dpages_test)(struct dentry *dentry, void *arg);
6572 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
6573 +                  au_dpages_test test, void *arg);
6574 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
6575 +                      int do_include, au_dpages_test test, void *arg);
6576 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
6577 +                           struct dentry *dentry, int do_include);
6578 +int au_test_subdir(struct dentry *d1, struct dentry *d2);
6579 +
6580 +/* ---------------------------------------------------------------------- */
6581 +
6582 +/*
6583 + * todo: in linux-3.13, several similar (but faster) helpers are added to
6584 + * include/linux/dcache.h. Try them (in the future).
6585 + */
6586 +
6587 +static inline int au_d_hashed_positive(struct dentry *d)
6588 +{
6589 +       int err;
6590 +       struct inode *inode = d->d_inode;
6591 +
6592 +       err = 0;
6593 +       if (unlikely(d_unhashed(d) || !inode || !inode->i_nlink))
6594 +               err = -ENOENT;
6595 +       return err;
6596 +}
6597 +
6598 +static inline int au_d_linkable(struct dentry *d)
6599 +{
6600 +       int err;
6601 +       struct inode *inode = d->d_inode;
6602 +
6603 +       err = au_d_hashed_positive(d);
6604 +       if (err
6605 +           && inode
6606 +           && (inode->i_state & I_LINKABLE))
6607 +               err = 0;
6608 +       return err;
6609 +}
6610 +
6611 +static inline int au_d_alive(struct dentry *d)
6612 +{
6613 +       int err;
6614 +       struct inode *inode;
6615 +
6616 +       err = 0;
6617 +       if (!IS_ROOT(d))
6618 +               err = au_d_hashed_positive(d);
6619 +       else {
6620 +               inode = d->d_inode;
6621 +               if (unlikely(d_unlinked(d) || !inode || !inode->i_nlink))
6622 +                       err = -ENOENT;
6623 +       }
6624 +       return err;
6625 +}
6626 +
6627 +static inline int au_alive_dir(struct dentry *d)
6628 +{
6629 +       int err;
6630 +
6631 +       err = au_d_alive(d);
6632 +       if (unlikely(err || IS_DEADDIR(d->d_inode)))
6633 +               err = -ENOENT;
6634 +       return err;
6635 +}
6636 +
6637 +static inline int au_qstreq(struct qstr *a, struct qstr *b)
6638 +{
6639 +       return a->len == b->len
6640 +               && !memcmp(a->name, b->name, a->len);
6641 +}
6642 +
6643 +static inline int au_dcount(struct dentry *d)
6644 +{
6645 +       return (int)d_count(d);
6646 +}
6647 +
6648 +#endif /* __KERNEL__ */
6649 +#endif /* __AUFS_DCSUB_H__ */
6650 diff -urN /usr/share/empty/fs/aufs/debug.c linux/fs/aufs/debug.c
6651 --- /usr/share/empty/fs/aufs/debug.c    1970-01-01 01:00:00.000000000 +0100
6652 +++ linux/fs/aufs/debug.c       2015-03-27 21:56:35.463461668 +0100
6653 @@ -0,0 +1,520 @@
6654 +/*
6655 + * Copyright (C) 2005-2015 Junjiro R. Okajima
6656 + *
6657 + * This program, aufs is free software; you can redistribute it and/or modify
6658 + * it under the terms of the GNU General Public License as published by
6659 + * the Free Software Foundation; either version 2 of the License, or
6660 + * (at your option) any later version.
6661 + *
6662 + * This program is distributed in the hope that it will be useful,
6663 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6664 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6665 + * GNU General Public License for more details.
6666 + *
6667 + * You should have received a copy of the GNU General Public License
6668 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6669 + */
6670 +
6671 +/*
6672 + * debug print functions
6673 + */
6674 +
6675 +#include <linux/vt_kern.h>
6676 +#include "aufs.h"
6677 +
6678 +/* Returns 0, or -errno.  arg is in kp->arg. */
6679 +static int param_atomic_t_set(const char *val, const struct kernel_param *kp)
6680 +{
6681 +       int err, n;
6682 +
6683 +       err = kstrtoint(val, 0, &n);
6684 +       if (!err) {
6685 +               if (n > 0)
6686 +                       au_debug_on();
6687 +               else
6688 +                       au_debug_off();
6689 +       }
6690 +       return err;
6691 +}
6692 +
6693 +/* Returns length written or -errno.  Buffer is 4k (ie. be short!) */
6694 +static int param_atomic_t_get(char *buffer, const struct kernel_param *kp)
6695 +{
6696 +       atomic_t *a;
6697 +
6698 +       a = kp->arg;
6699 +       return sprintf(buffer, "%d", atomic_read(a));
6700 +}
6701 +
6702 +static struct kernel_param_ops param_ops_atomic_t = {
6703 +       .set = param_atomic_t_set,
6704 +       .get = param_atomic_t_get
6705 +       /* void (*free)(void *arg) */
6706 +};
6707 +
6708 +atomic_t aufs_debug = ATOMIC_INIT(0);
6709 +MODULE_PARM_DESC(debug, "debug print");
6710 +module_param_named(debug, aufs_debug, atomic_t, S_IRUGO | S_IWUSR | S_IWGRP);
6711 +
6712 +DEFINE_MUTEX(au_dbg_mtx);      /* just to serialize the dbg msgs */
6713 +char *au_plevel = KERN_DEBUG;
6714 +#define dpri(fmt, ...) do {                                    \
6715 +       if ((au_plevel                                          \
6716 +            && strcmp(au_plevel, KERN_DEBUG))                  \
6717 +           || au_debug_test())                                 \
6718 +               printk("%s" fmt, au_plevel, ##__VA_ARGS__);     \
6719 +} while (0)
6720 +
6721 +/* ---------------------------------------------------------------------- */
6722 +
6723 +void au_dpri_whlist(struct au_nhash *whlist)
6724 +{
6725 +       unsigned long ul, n;
6726 +       struct hlist_head *head;
6727 +       struct au_vdir_wh *pos;
6728 +
6729 +       n = whlist->nh_num;
6730 +       head = whlist->nh_head;
6731 +       for (ul = 0; ul < n; ul++) {
6732 +               hlist_for_each_entry(pos, head, wh_hash)
6733 +                       dpri("b%d, %.*s, %d\n",
6734 +                            pos->wh_bindex,
6735 +                            pos->wh_str.len, pos->wh_str.name,
6736 +                            pos->wh_str.len);
6737 +               head++;
6738 +       }
6739 +}
6740 +
6741 +void au_dpri_vdir(struct au_vdir *vdir)
6742 +{
6743 +       unsigned long ul;
6744 +       union au_vdir_deblk_p p;
6745 +       unsigned char *o;
6746 +
6747 +       if (!vdir || IS_ERR(vdir)) {
6748 +               dpri("err %ld\n", PTR_ERR(vdir));
6749 +               return;
6750 +       }
6751 +
6752 +       dpri("deblk %u, nblk %lu, deblk %p, last{%lu, %p}, ver %lu\n",
6753 +            vdir->vd_deblk_sz, vdir->vd_nblk, vdir->vd_deblk,
6754 +            vdir->vd_last.ul, vdir->vd_last.p.deblk, vdir->vd_version);
6755 +       for (ul = 0; ul < vdir->vd_nblk; ul++) {
6756 +               p.deblk = vdir->vd_deblk[ul];
6757 +               o = p.deblk;
6758 +               dpri("[%lu]: %p\n", ul, o);
6759 +       }
6760 +}
6761 +
6762 +static int do_pri_inode(aufs_bindex_t bindex, struct inode *inode, int hn,
6763 +                       struct dentry *wh)
6764 +{
6765 +       char *n = NULL;
6766 +       int l = 0;
6767 +
6768 +       if (!inode || IS_ERR(inode)) {
6769 +               dpri("i%d: err %ld\n", bindex, PTR_ERR(inode));
6770 +               return -1;
6771 +       }
6772 +
6773 +       /* the type of i_blocks depends upon CONFIG_LBDAF */
6774 +       BUILD_BUG_ON(sizeof(inode->i_blocks) != sizeof(unsigned long)
6775 +                    && sizeof(inode->i_blocks) != sizeof(u64));
6776 +       if (wh) {
6777 +               n = (void *)wh->d_name.name;
6778 +               l = wh->d_name.len;
6779 +       }
6780 +
6781 +       dpri("i%d: %p, i%lu, %s, cnt %d, nl %u, 0%o, sz %llu, blk %llu,"
6782 +            " hn %d, ct %lld, np %lu, st 0x%lx, f 0x%x, v %llu, g %x%s%.*s\n",
6783 +            bindex, inode,
6784 +            inode->i_ino, inode->i_sb ? au_sbtype(inode->i_sb) : "??",
6785 +            atomic_read(&inode->i_count), inode->i_nlink, inode->i_mode,
6786 +            i_size_read(inode), (unsigned long long)inode->i_blocks,
6787 +            hn, (long long)timespec_to_ns(&inode->i_ctime) & 0x0ffff,
6788 +            inode->i_mapping ? inode->i_mapping->nrpages : 0,
6789 +            inode->i_state, inode->i_flags, inode->i_version,
6790 +            inode->i_generation,
6791 +            l ? ", wh " : "", l, n);
6792 +       return 0;
6793 +}
6794 +
6795 +void au_dpri_inode(struct inode *inode)
6796 +{
6797 +       struct au_iinfo *iinfo;
6798 +       aufs_bindex_t bindex;
6799 +       int err, hn;
6800 +
6801 +       err = do_pri_inode(-1, inode, -1, NULL);
6802 +       if (err || !au_test_aufs(inode->i_sb))
6803 +               return;
6804 +
6805 +       iinfo = au_ii(inode);
6806 +       if (!iinfo)
6807 +               return;
6808 +       dpri("i-1: bstart %d, bend %d, gen %d\n",
6809 +            iinfo->ii_bstart, iinfo->ii_bend, au_iigen(inode, NULL));
6810 +       if (iinfo->ii_bstart < 0)
6811 +               return;
6812 +       hn = 0;
6813 +       for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend; bindex++) {
6814 +               hn = !!au_hn(iinfo->ii_hinode + bindex);
6815 +               do_pri_inode(bindex, iinfo->ii_hinode[0 + bindex].hi_inode, hn,
6816 +                            iinfo->ii_hinode[0 + bindex].hi_whdentry);
6817 +       }
6818 +}
6819 +
6820 +void au_dpri_dalias(struct inode *inode)
6821 +{
6822 +       struct dentry *d;
6823 +
6824 +       spin_lock(&inode->i_lock);
6825 +       hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias)
6826 +               au_dpri_dentry(d);
6827 +       spin_unlock(&inode->i_lock);
6828 +}
6829 +
6830 +static int do_pri_dentry(aufs_bindex_t bindex, struct dentry *dentry)
6831 +{
6832 +       struct dentry *wh = NULL;
6833 +       int hn;
6834 +       struct au_iinfo *iinfo;
6835 +
6836 +       if (!dentry || IS_ERR(dentry)) {
6837 +               dpri("d%d: err %ld\n", bindex, PTR_ERR(dentry));
6838 +               return -1;
6839 +       }
6840 +       /* do not call dget_parent() here */
6841 +       /* note: access d_xxx without d_lock */
6842 +       dpri("d%d: %p, %pd2?, %s, cnt %d, flags 0x%x, %shashed\n",
6843 +            bindex, dentry, dentry,
6844 +            dentry->d_sb ? au_sbtype(dentry->d_sb) : "??",
6845 +            au_dcount(dentry), dentry->d_flags,
6846 +            d_unhashed(dentry) ? "un" : "");
6847 +       hn = -1;
6848 +       if (bindex >= 0 && dentry->d_inode && au_test_aufs(dentry->d_sb)) {
6849 +               iinfo = au_ii(dentry->d_inode);
6850 +               if (iinfo) {
6851 +                       hn = !!au_hn(iinfo->ii_hinode + bindex);
6852 +                       wh = iinfo->ii_hinode[0 + bindex].hi_whdentry;
6853 +               }
6854 +       }
6855 +       do_pri_inode(bindex, dentry->d_inode, hn, wh);
6856 +       return 0;
6857 +}
6858 +
6859 +void au_dpri_dentry(struct dentry *dentry)
6860 +{
6861 +       struct au_dinfo *dinfo;
6862 +       aufs_bindex_t bindex;
6863 +       int err;
6864 +       struct au_hdentry *hdp;
6865 +
6866 +       err = do_pri_dentry(-1, dentry);
6867 +       if (err || !au_test_aufs(dentry->d_sb))
6868 +               return;
6869 +
6870 +       dinfo = au_di(dentry);
6871 +       if (!dinfo)
6872 +               return;
6873 +       dpri("d-1: bstart %d, bend %d, bwh %d, bdiropq %d, gen %d, tmp %d\n",
6874 +            dinfo->di_bstart, dinfo->di_bend,
6875 +            dinfo->di_bwh, dinfo->di_bdiropq, au_digen(dentry),
6876 +            dinfo->di_tmpfile);
6877 +       if (dinfo->di_bstart < 0)
6878 +               return;
6879 +       hdp = dinfo->di_hdentry;
6880 +       for (bindex = dinfo->di_bstart; bindex <= dinfo->di_bend; bindex++)
6881 +               do_pri_dentry(bindex, hdp[0 + bindex].hd_dentry);
6882 +}
6883 +
6884 +static int do_pri_file(aufs_bindex_t bindex, struct file *file)
6885 +{
6886 +       char a[32];
6887 +
6888 +       if (!file || IS_ERR(file)) {
6889 +               dpri("f%d: err %ld\n", bindex, PTR_ERR(file));
6890 +               return -1;
6891 +       }
6892 +       a[0] = 0;
6893 +       if (bindex < 0
6894 +           && file->f_path.dentry
6895 +           && au_test_aufs(file->f_path.dentry->d_sb)
6896 +           && au_fi(file))
6897 +               snprintf(a, sizeof(a), ", gen %d, mmapped %d",
6898 +                        au_figen(file), atomic_read(&au_fi(file)->fi_mmapped));
6899 +       dpri("f%d: mode 0x%x, flags 0%o, cnt %ld, v %llu, pos %llu%s\n",
6900 +            bindex, file->f_mode, file->f_flags, (long)file_count(file),
6901 +            file->f_version, file->f_pos, a);
6902 +       if (file->f_path.dentry)
6903 +               do_pri_dentry(bindex, file->f_path.dentry);
6904 +       return 0;
6905 +}
6906 +
6907 +void au_dpri_file(struct file *file)
6908 +{
6909 +       struct au_finfo *finfo;
6910 +       struct au_fidir *fidir;
6911 +       struct au_hfile *hfile;
6912 +       aufs_bindex_t bindex;
6913 +       int err;
6914 +
6915 +       err = do_pri_file(-1, file);
6916 +       if (err
6917 +           || !file->f_path.dentry
6918 +           || !au_test_aufs(file->f_path.dentry->d_sb))
6919 +               return;
6920 +
6921 +       finfo = au_fi(file);
6922 +       if (!finfo)
6923 +               return;
6924 +       if (finfo->fi_btop < 0)
6925 +               return;
6926 +       fidir = finfo->fi_hdir;
6927 +       if (!fidir)
6928 +               do_pri_file(finfo->fi_btop, finfo->fi_htop.hf_file);
6929 +       else
6930 +               for (bindex = finfo->fi_btop;
6931 +                    bindex >= 0 && bindex <= fidir->fd_bbot;
6932 +                    bindex++) {
6933 +                       hfile = fidir->fd_hfile + bindex;
6934 +                       do_pri_file(bindex, hfile ? hfile->hf_file : NULL);
6935 +               }
6936 +}
6937 +
6938 +static int do_pri_br(aufs_bindex_t bindex, struct au_branch *br)
6939 +{
6940 +       struct vfsmount *mnt;
6941 +       struct super_block *sb;
6942 +
6943 +       if (!br || IS_ERR(br))
6944 +               goto out;
6945 +       mnt = au_br_mnt(br);
6946 +       if (!mnt || IS_ERR(mnt))
6947 +               goto out;
6948 +       sb = mnt->mnt_sb;
6949 +       if (!sb || IS_ERR(sb))
6950 +               goto out;
6951 +
6952 +       dpri("s%d: {perm 0x%x, id %d, cnt %d, wbr %p}, "
6953 +            "%s, dev 0x%02x%02x, flags 0x%lx, cnt %d, active %d, "
6954 +            "xino %d\n",
6955 +            bindex, br->br_perm, br->br_id, atomic_read(&br->br_count),
6956 +            br->br_wbr, au_sbtype(sb), MAJOR(sb->s_dev), MINOR(sb->s_dev),
6957 +            sb->s_flags, sb->s_count,
6958 +            atomic_read(&sb->s_active), !!br->br_xino.xi_file);
6959 +       return 0;
6960 +
6961 +out:
6962 +       dpri("s%d: err %ld\n", bindex, PTR_ERR(br));
6963 +       return -1;
6964 +}
6965 +
6966 +void au_dpri_sb(struct super_block *sb)
6967 +{
6968 +       struct au_sbinfo *sbinfo;
6969 +       aufs_bindex_t bindex;
6970 +       int err;
6971 +       /* to reuduce stack size */
6972 +       struct {
6973 +               struct vfsmount mnt;
6974 +               struct au_branch fake;
6975 +       } *a;
6976 +
6977 +       /* this function can be called from magic sysrq */
6978 +       a = kzalloc(sizeof(*a), GFP_ATOMIC);
6979 +       if (unlikely(!a)) {
6980 +               dpri("no memory\n");
6981 +               return;
6982 +       }
6983 +
6984 +       a->mnt.mnt_sb = sb;
6985 +       a->fake.br_perm = 0;
6986 +       a->fake.br_path.mnt = &a->mnt;
6987 +       a->fake.br_xino.xi_file = NULL;
6988 +       atomic_set(&a->fake.br_count, 0);
6989 +       smp_mb(); /* atomic_set */
6990 +       err = do_pri_br(-1, &a->fake);
6991 +       kfree(a);
6992 +       dpri("dev 0x%x\n", sb->s_dev);
6993 +       if (err || !au_test_aufs(sb))
6994 +               return;
6995 +
6996 +       sbinfo = au_sbi(sb);
6997 +       if (!sbinfo)
6998 +               return;
6999 +       dpri("nw %d, gen %u, kobj %d\n",
7000 +            atomic_read(&sbinfo->si_nowait.nw_len), sbinfo->si_generation,
7001 +            atomic_read(&sbinfo->si_kobj.kref.refcount));
7002 +       for (bindex = 0; bindex <= sbinfo->si_bend; bindex++)
7003 +               do_pri_br(bindex, sbinfo->si_branch[0 + bindex]);
7004 +}
7005 +
7006 +/* ---------------------------------------------------------------------- */
7007 +
7008 +void au_dbg_sleep_jiffy(int jiffy)
7009 +{
7010 +       while (jiffy)
7011 +               jiffy = schedule_timeout_uninterruptible(jiffy);
7012 +}
7013 +
7014 +void au_dbg_iattr(struct iattr *ia)
7015 +{
7016 +#define AuBit(name)                                    \
7017 +       do {                                            \
7018 +               if (ia->ia_valid & ATTR_ ## name)       \
7019 +                       dpri(#name "\n");               \
7020 +       } while (0)
7021 +       AuBit(MODE);
7022 +       AuBit(UID);
7023 +       AuBit(GID);
7024 +       AuBit(SIZE);
7025 +       AuBit(ATIME);
7026 +       AuBit(MTIME);
7027 +       AuBit(CTIME);
7028 +       AuBit(ATIME_SET);
7029 +       AuBit(MTIME_SET);
7030 +       AuBit(FORCE);
7031 +       AuBit(ATTR_FLAG);
7032 +       AuBit(KILL_SUID);
7033 +       AuBit(KILL_SGID);
7034 +       AuBit(FILE);
7035 +       AuBit(KILL_PRIV);
7036 +       AuBit(OPEN);
7037 +       AuBit(TIMES_SET);
7038 +#undef AuBit
7039 +       dpri("ia_file %p\n", ia->ia_file);
7040 +}
7041 +
7042 +/* ---------------------------------------------------------------------- */
7043 +
7044 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line)
7045 +{
7046 +       struct inode *h_inode, *inode = dentry->d_inode;
7047 +       struct dentry *h_dentry;
7048 +       aufs_bindex_t bindex, bend, bi;
7049 +
7050 +       if (!inode /* || au_di(dentry)->di_lsc == AuLsc_DI_TMP */)
7051 +               return;
7052 +
7053 +       bend = au_dbend(dentry);
7054 +       bi = au_ibend(inode);
7055 +       if (bi < bend)
7056 +               bend = bi;
7057 +       bindex = au_dbstart(dentry);
7058 +       bi = au_ibstart(inode);
7059 +       if (bi > bindex)
7060 +               bindex = bi;
7061 +
7062 +       for (; bindex <= bend; bindex++) {
7063 +               h_dentry = au_h_dptr(dentry, bindex);
7064 +               if (!h_dentry)
7065 +                       continue;
7066 +               h_inode = au_h_iptr(inode, bindex);
7067 +               if (unlikely(h_inode != h_dentry->d_inode)) {
7068 +                       au_debug_on();
7069 +                       AuDbg("b%d, %s:%d\n", bindex, func, line);
7070 +                       AuDbgDentry(dentry);
7071 +                       AuDbgInode(inode);
7072 +                       au_debug_off();
7073 +                       BUG();
7074 +               }
7075 +       }
7076 +}
7077 +
7078 +void au_dbg_verify_dir_parent(struct dentry *dentry, unsigned int sigen)
7079 +{
7080 +       struct dentry *parent;
7081 +
7082 +       parent = dget_parent(dentry);
7083 +       AuDebugOn(!d_is_dir(dentry));
7084 +       AuDebugOn(IS_ROOT(dentry));
7085 +       AuDebugOn(au_digen_test(parent, sigen));
7086 +       dput(parent);
7087 +}
7088 +
7089 +void au_dbg_verify_nondir_parent(struct dentry *dentry, unsigned int sigen)
7090 +{
7091 +       struct dentry *parent;
7092 +
7093 +       parent = dget_parent(dentry);
7094 +       AuDebugOn(d_is_dir(dentry));
7095 +       AuDebugOn(au_digen_test(parent, sigen));
7096 +       dput(parent);
7097 +}
7098 +
7099 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen)
7100 +{
7101 +       int err, i, j;
7102 +       struct au_dcsub_pages dpages;
7103 +       struct au_dpage *dpage;
7104 +       struct dentry **dentries;
7105 +
7106 +       err = au_dpages_init(&dpages, GFP_NOFS);
7107 +       AuDebugOn(err);
7108 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/1);
7109 +       AuDebugOn(err);
7110 +       for (i = dpages.ndpage - 1; !err && i >= 0; i--) {
7111 +               dpage = dpages.dpages + i;
7112 +               dentries = dpage->dentries;
7113 +               for (j = dpage->ndentry - 1; !err && j >= 0; j--)
7114 +                       AuDebugOn(au_digen_test(dentries[j], sigen));
7115 +       }
7116 +       au_dpages_free(&dpages);
7117 +}
7118 +
7119 +void au_dbg_verify_kthread(void)
7120 +{
7121 +       if (au_wkq_test()) {
7122 +               au_dbg_blocked();
7123 +               /*
7124 +                * It may be recursive, but udba=notify between two aufs mounts,
7125 +                * where a single ro branch is shared, is not a problem.
7126 +                */
7127 +               /* WARN_ON(1); */
7128 +       }
7129 +}
7130 +
7131 +/* ---------------------------------------------------------------------- */
7132 +
7133 +void au_debug_sbinfo_init(struct au_sbinfo *sbinfo __maybe_unused)
7134 +{
7135 +#ifdef AuForceNoPlink
7136 +       au_opt_clr(sbinfo->si_mntflags, PLINK);
7137 +#endif
7138 +#ifdef AuForceNoXino
7139 +       au_opt_clr(sbinfo->si_mntflags, XINO);
7140 +#endif
7141 +#ifdef AuForceNoRefrof
7142 +       au_opt_clr(sbinfo->si_mntflags, REFROF);
7143 +#endif
7144 +#ifdef AuForceHnotify
7145 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_HNOTIFY);
7146 +#endif
7147 +#ifdef AuForceRd0
7148 +       sbinfo->si_rdblk = 0;
7149 +       sbinfo->si_rdhash = 0;
7150 +#endif
7151 +}
7152 +
7153 +int __init au_debug_init(void)
7154 +{
7155 +       aufs_bindex_t bindex;
7156 +       struct au_vdir_destr destr;
7157 +
7158 +       bindex = -1;
7159 +       AuDebugOn(bindex >= 0);
7160 +
7161 +       destr.len = -1;
7162 +       AuDebugOn(destr.len < NAME_MAX);
7163 +
7164 +#ifdef CONFIG_4KSTACKS
7165 +       pr_warn("CONFIG_4KSTACKS is defined.\n");
7166 +#endif
7167 +
7168 +#ifdef AuForceNoBrs
7169 +       sysaufs_brs = 0;
7170 +#endif
7171 +
7172 +       return 0;
7173 +}
7174 diff -urN /usr/share/empty/fs/aufs/debug.h linux/fs/aufs/debug.h
7175 --- /usr/share/empty/fs/aufs/debug.h    1970-01-01 01:00:00.000000000 +0100
7176 +++ linux/fs/aufs/debug.h       2015-03-27 21:56:35.463461668 +0100
7177 @@ -0,0 +1,262 @@
7178 +/*
7179 + * Copyright (C) 2005-2015 Junjiro R. Okajima
7180 + *
7181 + * This program, aufs is free software; you can redistribute it and/or modify
7182 + * it under the terms of the GNU General Public License as published by
7183 + * the Free Software Foundation; either version 2 of the License, or
7184 + * (at your option) any later version.
7185 + *
7186 + * This program is distributed in the hope that it will be useful,
7187 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7188 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7189 + * GNU General Public License for more details.
7190 + *
7191 + * You should have received a copy of the GNU General Public License
7192 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7193 + */
7194 +
7195 +/*
7196 + * debug print functions
7197 + */
7198 +
7199 +#ifndef __AUFS_DEBUG_H__
7200 +#define __AUFS_DEBUG_H__
7201 +
7202 +#ifdef __KERNEL__
7203 +
7204 +#include <linux/atomic.h>
7205 +#include <linux/module.h>
7206 +#include <linux/kallsyms.h>
7207 +#include <linux/sysrq.h>
7208 +
7209 +#ifdef CONFIG_AUFS_DEBUG
7210 +#define AuDebugOn(a)           BUG_ON(a)
7211 +
7212 +/* module parameter */
7213 +extern atomic_t aufs_debug;
7214 +static inline void au_debug_on(void)
7215 +{
7216 +       atomic_inc(&aufs_debug);
7217 +}
7218 +static inline void au_debug_off(void)
7219 +{
7220 +       atomic_dec_if_positive(&aufs_debug);
7221 +}
7222 +
7223 +static inline int au_debug_test(void)
7224 +{
7225 +       return atomic_read(&aufs_debug) > 0;
7226 +}
7227 +#else
7228 +#define AuDebugOn(a)           do {} while (0)
7229 +AuStubVoid(au_debug_on, void)
7230 +AuStubVoid(au_debug_off, void)
7231 +AuStubInt0(au_debug_test, void)
7232 +#endif /* CONFIG_AUFS_DEBUG */
7233 +
7234 +#define param_check_atomic_t(name, p) __param_check(name, p, atomic_t)
7235 +
7236 +/* ---------------------------------------------------------------------- */
7237 +
7238 +/* debug print */
7239 +
7240 +#define AuDbg(fmt, ...) do { \
7241 +       if (au_debug_test()) \
7242 +               pr_debug("DEBUG: " fmt, ##__VA_ARGS__); \
7243 +} while (0)
7244 +#define AuLabel(l)             AuDbg(#l "\n")
7245 +#define AuIOErr(fmt, ...)      pr_err("I/O Error, " fmt, ##__VA_ARGS__)
7246 +#define AuWarn1(fmt, ...) do { \
7247 +       static unsigned char _c; \
7248 +       if (!_c++) \
7249 +               pr_warn(fmt, ##__VA_ARGS__); \
7250 +} while (0)
7251 +
7252 +#define AuErr1(fmt, ...) do { \
7253 +       static unsigned char _c; \
7254 +       if (!_c++) \
7255 +               pr_err(fmt, ##__VA_ARGS__); \
7256 +} while (0)
7257 +
7258 +#define AuIOErr1(fmt, ...) do { \
7259 +       static unsigned char _c; \
7260 +       if (!_c++) \
7261 +               AuIOErr(fmt, ##__VA_ARGS__); \
7262 +} while (0)
7263 +
7264 +#define AuUnsupportMsg "This operation is not supported." \
7265 +                       " Please report this application to aufs-users ML."
7266 +#define AuUnsupport(fmt, ...) do { \
7267 +       pr_err(AuUnsupportMsg "\n" fmt, ##__VA_ARGS__); \
7268 +       dump_stack(); \
7269 +} while (0)
7270 +
7271 +#define AuTraceErr(e) do { \
7272 +       if (unlikely((e) < 0)) \
7273 +               AuDbg("err %d\n", (int)(e)); \
7274 +} while (0)
7275 +
7276 +#define AuTraceErrPtr(p) do { \
7277 +       if (IS_ERR(p)) \
7278 +               AuDbg("err %ld\n", PTR_ERR(p)); \
7279 +} while (0)
7280 +
7281 +/* dirty macros for debug print, use with "%.*s" and caution */
7282 +#define AuLNPair(qstr)         (qstr)->len, (qstr)->name
7283 +
7284 +/* ---------------------------------------------------------------------- */
7285 +
7286 +struct au_sbinfo;
7287 +struct au_finfo;
7288 +struct dentry;
7289 +#ifdef CONFIG_AUFS_DEBUG
7290 +extern struct mutex au_dbg_mtx;
7291 +extern char *au_plevel;
7292 +struct au_nhash;
7293 +void au_dpri_whlist(struct au_nhash *whlist);
7294 +struct au_vdir;
7295 +void au_dpri_vdir(struct au_vdir *vdir);
7296 +struct inode;
7297 +void au_dpri_inode(struct inode *inode);
7298 +void au_dpri_dalias(struct inode *inode);
7299 +void au_dpri_dentry(struct dentry *dentry);
7300 +struct file;
7301 +void au_dpri_file(struct file *filp);
7302 +struct super_block;
7303 +void au_dpri_sb(struct super_block *sb);
7304 +
7305 +void au_dbg_sleep_jiffy(int jiffy);
7306 +struct iattr;
7307 +void au_dbg_iattr(struct iattr *ia);
7308 +
7309 +#define au_dbg_verify_dinode(d) __au_dbg_verify_dinode(d, __func__, __LINE__)
7310 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
7311 +void au_dbg_verify_dir_parent(struct dentry *dentry, unsigned int sigen);
7312 +void au_dbg_verify_nondir_parent(struct dentry *dentry, unsigned int sigen);
7313 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen);
7314 +void au_dbg_verify_kthread(void);
7315 +
7316 +int __init au_debug_init(void);
7317 +void au_debug_sbinfo_init(struct au_sbinfo *sbinfo);
7318 +#define AuDbgWhlist(w) do { \
7319 +       mutex_lock(&au_dbg_mtx); \
7320 +       AuDbg(#w "\n"); \
7321 +       au_dpri_whlist(w); \
7322 +       mutex_unlock(&au_dbg_mtx); \
7323 +} while (0)
7324 +
7325 +#define AuDbgVdir(v) do { \
7326 +       mutex_lock(&au_dbg_mtx); \
7327 +       AuDbg(#v "\n"); \
7328 +       au_dpri_vdir(v); \
7329 +       mutex_unlock(&au_dbg_mtx); \
7330 +} while (0)
7331 +
7332 +#define AuDbgInode(i) do { \
7333 +       mutex_lock(&au_dbg_mtx); \
7334 +       AuDbg(#i "\n"); \
7335 +       au_dpri_inode(i); \
7336 +       mutex_unlock(&au_dbg_mtx); \
7337 +} while (0)
7338 +
7339 +#define AuDbgDAlias(i) do { \
7340 +       mutex_lock(&au_dbg_mtx); \
7341 +       AuDbg(#i "\n"); \
7342 +       au_dpri_dalias(i); \
7343 +       mutex_unlock(&au_dbg_mtx); \
7344 +} while (0)
7345 +
7346 +#define AuDbgDentry(d) do { \
7347 +       mutex_lock(&au_dbg_mtx); \
7348 +       AuDbg(#d "\n"); \
7349 +       au_dpri_dentry(d); \
7350 +       mutex_unlock(&au_dbg_mtx); \
7351 +} while (0)
7352 +
7353 +#define AuDbgFile(f) do { \
7354 +       mutex_lock(&au_dbg_mtx); \
7355 +       AuDbg(#f "\n"); \
7356 +       au_dpri_file(f); \
7357 +       mutex_unlock(&au_dbg_mtx); \
7358 +} while (0)
7359 +
7360 +#define AuDbgSb(sb) do { \
7361 +       mutex_lock(&au_dbg_mtx); \
7362 +       AuDbg(#sb "\n"); \
7363 +       au_dpri_sb(sb); \
7364 +       mutex_unlock(&au_dbg_mtx); \
7365 +} while (0)
7366 +
7367 +#define AuDbgSleep(sec) do { \
7368 +       AuDbg("sleep %d sec\n", sec); \
7369 +       ssleep(sec); \
7370 +} while (0)
7371 +
7372 +#define AuDbgSleepJiffy(jiffy) do { \
7373 +       AuDbg("sleep %d jiffies\n", jiffy); \
7374 +       au_dbg_sleep_jiffy(jiffy); \
7375 +} while (0)
7376 +
7377 +#define AuDbgIAttr(ia) do { \
7378 +       AuDbg("ia_valid 0x%x\n", (ia)->ia_valid); \
7379 +       au_dbg_iattr(ia); \
7380 +} while (0)
7381 +
7382 +#define AuDbgSym(addr) do {                            \
7383 +       char sym[KSYM_SYMBOL_LEN];                      \
7384 +       sprint_symbol(sym, (unsigned long)addr);        \
7385 +       AuDbg("%s\n", sym);                             \
7386 +} while (0)
7387 +
7388 +#define AuInfoSym(addr) do {                           \
7389 +       char sym[KSYM_SYMBOL_LEN];                      \
7390 +       sprint_symbol(sym, (unsigned long)addr);        \
7391 +       AuInfo("%s\n", sym);                            \
7392 +} while (0)
7393 +#else
7394 +AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
7395 +AuStubVoid(au_dbg_verify_dir_parent, struct dentry *dentry, unsigned int sigen)
7396 +AuStubVoid(au_dbg_verify_nondir_parent, struct dentry *dentry,
7397 +          unsigned int sigen)
7398 +AuStubVoid(au_dbg_verify_gen, struct dentry *parent, unsigned int sigen)
7399 +AuStubVoid(au_dbg_verify_kthread, void)
7400 +AuStubInt0(__init au_debug_init, void)
7401 +AuStubVoid(au_debug_sbinfo_init, struct au_sbinfo *sbinfo)
7402 +
7403 +#define AuDbgWhlist(w)         do {} while (0)
7404 +#define AuDbgVdir(v)           do {} while (0)
7405 +#define AuDbgInode(i)          do {} while (0)
7406 +#define AuDbgDAlias(i)         do {} while (0)
7407 +#define AuDbgDentry(d)         do {} while (0)
7408 +#define AuDbgFile(f)           do {} while (0)
7409 +#define AuDbgSb(sb)            do {} while (0)
7410 +#define AuDbgSleep(sec)                do {} while (0)
7411 +#define AuDbgSleepJiffy(jiffy) do {} while (0)
7412 +#define AuDbgIAttr(ia)         do {} while (0)
7413 +#define AuDbgSym(addr)         do {} while (0)
7414 +#define AuInfoSym(addr)                do {} while (0)
7415 +#endif /* CONFIG_AUFS_DEBUG */
7416 +
7417 +/* ---------------------------------------------------------------------- */
7418 +
7419 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
7420 +int __init au_sysrq_init(void);
7421 +void au_sysrq_fin(void);
7422 +
7423 +#ifdef CONFIG_HW_CONSOLE
7424 +#define au_dbg_blocked() do { \
7425 +       WARN_ON(1); \
7426 +       handle_sysrq('w'); \
7427 +} while (0)
7428 +#else
7429 +AuStubVoid(au_dbg_blocked, void)
7430 +#endif
7431 +
7432 +#else
7433 +AuStubInt0(__init au_sysrq_init, void)
7434 +AuStubVoid(au_sysrq_fin, void)
7435 +AuStubVoid(au_dbg_blocked, void)
7436 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
7437 +
7438 +#endif /* __KERNEL__ */
7439 +#endif /* __AUFS_DEBUG_H__ */
7440 diff -urN /usr/share/empty/fs/aufs/dentry.c linux/fs/aufs/dentry.c
7441 --- /usr/share/empty/fs/aufs/dentry.c   1970-01-01 01:00:00.000000000 +0100
7442 +++ linux/fs/aufs/dentry.c      2015-03-27 21:56:35.463461668 +0100
7443 @@ -0,0 +1,1097 @@
7444 +/*
7445 + * Copyright (C) 2005-2015 Junjiro R. Okajima
7446 + *
7447 + * This program, aufs is free software; you can redistribute it and/or modify
7448 + * it under the terms of the GNU General Public License as published by
7449 + * the Free Software Foundation; either version 2 of the License, or
7450 + * (at your option) any later version.
7451 + *
7452 + * This program is distributed in the hope that it will be useful,
7453 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7454 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7455 + * GNU General Public License for more details.
7456 + *
7457 + * You should have received a copy of the GNU General Public License
7458 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7459 + */
7460 +
7461 +/*
7462 + * lookup and dentry operations
7463 + */
7464 +
7465 +#include <linux/namei.h>
7466 +#include "aufs.h"
7467 +
7468 +#define AuLkup_ALLOW_NEG       1
7469 +#define AuLkup_IGNORE_PERM     (1 << 1)
7470 +#define au_ftest_lkup(flags, name)     ((flags) & AuLkup_##name)
7471 +#define au_fset_lkup(flags, name) \
7472 +       do { (flags) |= AuLkup_##name; } while (0)
7473 +#define au_fclr_lkup(flags, name) \
7474 +       do { (flags) &= ~AuLkup_##name; } while (0)
7475 +
7476 +struct au_do_lookup_args {
7477 +       unsigned int            flags;
7478 +       mode_t                  type;
7479 +};
7480 +
7481 +/*
7482 + * returns positive/negative dentry, NULL or an error.
7483 + * NULL means whiteout-ed or not-found.
7484 + */
7485 +static struct dentry*
7486 +au_do_lookup(struct dentry *h_parent, struct dentry *dentry,
7487 +            aufs_bindex_t bindex, struct qstr *wh_name,
7488 +            struct au_do_lookup_args *args)
7489 +{
7490 +       struct dentry *h_dentry;
7491 +       struct inode *h_inode;
7492 +       struct au_branch *br;
7493 +       int wh_found, opq;
7494 +       unsigned char wh_able;
7495 +       const unsigned char allow_neg = !!au_ftest_lkup(args->flags, ALLOW_NEG);
7496 +       const unsigned char ignore_perm = !!au_ftest_lkup(args->flags,
7497 +                                                         IGNORE_PERM);
7498 +
7499 +       wh_found = 0;
7500 +       br = au_sbr(dentry->d_sb, bindex);
7501 +       wh_able = !!au_br_whable(br->br_perm);
7502 +       if (wh_able)
7503 +               wh_found = au_wh_test(h_parent, wh_name, /*try_sio*/0);
7504 +       h_dentry = ERR_PTR(wh_found);
7505 +       if (!wh_found)
7506 +               goto real_lookup;
7507 +       if (unlikely(wh_found < 0))
7508 +               goto out;
7509 +
7510 +       /* We found a whiteout */
7511 +       /* au_set_dbend(dentry, bindex); */
7512 +       au_set_dbwh(dentry, bindex);
7513 +       if (!allow_neg)
7514 +               return NULL; /* success */
7515 +
7516 +real_lookup:
7517 +       if (!ignore_perm)
7518 +               h_dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
7519 +       else
7520 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7521 +       if (IS_ERR(h_dentry)) {
7522 +               if (PTR_ERR(h_dentry) == -ENAMETOOLONG
7523 +                   && !allow_neg)
7524 +                       h_dentry = NULL;
7525 +               goto out;
7526 +       }
7527 +
7528 +       h_inode = h_dentry->d_inode;
7529 +       if (!h_inode) {
7530 +               if (!allow_neg)
7531 +                       goto out_neg;
7532 +       } else if (wh_found
7533 +                  || (args->type && args->type != (h_inode->i_mode & S_IFMT)))
7534 +               goto out_neg;
7535 +
7536 +       if (au_dbend(dentry) <= bindex)
7537 +               au_set_dbend(dentry, bindex);
7538 +       if (au_dbstart(dentry) < 0 || bindex < au_dbstart(dentry))
7539 +               au_set_dbstart(dentry, bindex);
7540 +       au_set_h_dptr(dentry, bindex, h_dentry);
7541 +
7542 +       if (!d_is_dir(h_dentry)
7543 +           || !wh_able
7544 +           || (d_is_positive(dentry) && !d_is_dir(dentry)))
7545 +               goto out; /* success */
7546 +
7547 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
7548 +       opq = au_diropq_test(h_dentry);
7549 +       mutex_unlock(&h_inode->i_mutex);
7550 +       if (opq > 0)
7551 +               au_set_dbdiropq(dentry, bindex);
7552 +       else if (unlikely(opq < 0)) {
7553 +               au_set_h_dptr(dentry, bindex, NULL);
7554 +               h_dentry = ERR_PTR(opq);
7555 +       }
7556 +       goto out;
7557 +
7558 +out_neg:
7559 +       dput(h_dentry);
7560 +       h_dentry = NULL;
7561 +out:
7562 +       return h_dentry;
7563 +}
7564 +
7565 +static int au_test_shwh(struct super_block *sb, const struct qstr *name)
7566 +{
7567 +       if (unlikely(!au_opt_test(au_mntflags(sb), SHWH)
7568 +                    && !strncmp(name->name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)))
7569 +               return -EPERM;
7570 +       return 0;
7571 +}
7572 +
7573 +/*
7574 + * returns the number of lower positive dentries,
7575 + * otherwise an error.
7576 + * can be called at unlinking with @type is zero.
7577 + */
7578 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t bstart, mode_t type)
7579 +{
7580 +       int npositive, err;
7581 +       aufs_bindex_t bindex, btail, bdiropq;
7582 +       unsigned char isdir, dirperm1;
7583 +       struct qstr whname;
7584 +       struct au_do_lookup_args args = {
7585 +               .flags          = 0,
7586 +               .type           = type
7587 +       };
7588 +       const struct qstr *name = &dentry->d_name;
7589 +       struct dentry *parent;
7590 +       struct inode *inode;
7591 +       struct super_block *sb;
7592 +
7593 +       sb = dentry->d_sb;
7594 +       err = au_test_shwh(sb, name);
7595 +       if (unlikely(err))
7596 +               goto out;
7597 +
7598 +       err = au_wh_name_alloc(&whname, name);
7599 +       if (unlikely(err))
7600 +               goto out;
7601 +
7602 +       inode = dentry->d_inode;
7603 +       isdir = !!d_is_dir(dentry);
7604 +       if (!type)
7605 +               au_fset_lkup(args.flags, ALLOW_NEG);
7606 +       dirperm1 = !!au_opt_test(au_mntflags(sb), DIRPERM1);
7607 +
7608 +       npositive = 0;
7609 +       parent = dget_parent(dentry);
7610 +       btail = au_dbtaildir(parent);
7611 +       for (bindex = bstart; bindex <= btail; bindex++) {
7612 +               struct dentry *h_parent, *h_dentry;
7613 +               struct inode *h_inode, *h_dir;
7614 +
7615 +               h_dentry = au_h_dptr(dentry, bindex);
7616 +               if (h_dentry) {
7617 +                       if (h_dentry->d_inode)
7618 +                               npositive++;
7619 +                       if (type != S_IFDIR)
7620 +                               break;
7621 +                       continue;
7622 +               }
7623 +               h_parent = au_h_dptr(parent, bindex);
7624 +               if (!h_parent || !d_is_dir(h_parent))
7625 +                       continue;
7626 +
7627 +               h_dir = h_parent->d_inode;
7628 +               mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
7629 +               h_dentry = au_do_lookup(h_parent, dentry, bindex, &whname,
7630 +                                       &args);
7631 +               mutex_unlock(&h_dir->i_mutex);
7632 +               err = PTR_ERR(h_dentry);
7633 +               if (IS_ERR(h_dentry))
7634 +                       goto out_parent;
7635 +               if (h_dentry)
7636 +                       au_fclr_lkup(args.flags, ALLOW_NEG);
7637 +               if (dirperm1)
7638 +                       au_fset_lkup(args.flags, IGNORE_PERM);
7639 +
7640 +               if (au_dbwh(dentry) >= 0)
7641 +                       break;
7642 +               if (!h_dentry)
7643 +                       continue;
7644 +               h_inode = h_dentry->d_inode;
7645 +               if (!h_inode)
7646 +                       continue;
7647 +               npositive++;
7648 +               if (!args.type)
7649 +                       args.type = h_inode->i_mode & S_IFMT;
7650 +               if (args.type != S_IFDIR)
7651 +                       break;
7652 +               else if (isdir) {
7653 +                       /* the type of lower may be different */
7654 +                       bdiropq = au_dbdiropq(dentry);
7655 +                       if (bdiropq >= 0 && bdiropq <= bindex)
7656 +                               break;
7657 +               }
7658 +       }
7659 +
7660 +       if (npositive) {
7661 +               AuLabel(positive);
7662 +               au_update_dbstart(dentry);
7663 +       }
7664 +       err = npositive;
7665 +       if (unlikely(!au_opt_test(au_mntflags(sb), UDBA_NONE)
7666 +                    && au_dbstart(dentry) < 0)) {
7667 +               err = -EIO;
7668 +               AuIOErr("both of real entry and whiteout found, %pd, err %d\n",
7669 +                       dentry, err);
7670 +       }
7671 +
7672 +out_parent:
7673 +       dput(parent);
7674 +       kfree(whname.name);
7675 +out:
7676 +       return err;
7677 +}
7678 +
7679 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent)
7680 +{
7681 +       struct dentry *dentry;
7682 +       int wkq_err;
7683 +
7684 +       if (!au_test_h_perm_sio(parent->d_inode, MAY_EXEC))
7685 +               dentry = vfsub_lkup_one(name, parent);
7686 +       else {
7687 +               struct vfsub_lkup_one_args args = {
7688 +                       .errp   = &dentry,
7689 +                       .name   = name,
7690 +                       .parent = parent
7691 +               };
7692 +
7693 +               wkq_err = au_wkq_wait(vfsub_call_lkup_one, &args);
7694 +               if (unlikely(wkq_err))
7695 +                       dentry = ERR_PTR(wkq_err);
7696 +       }
7697 +
7698 +       return dentry;
7699 +}
7700 +
7701 +/*
7702 + * lookup @dentry on @bindex which should be negative.
7703 + */
7704 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh)
7705 +{
7706 +       int err;
7707 +       struct dentry *parent, *h_parent, *h_dentry;
7708 +       struct au_branch *br;
7709 +
7710 +       parent = dget_parent(dentry);
7711 +       h_parent = au_h_dptr(parent, bindex);
7712 +       br = au_sbr(dentry->d_sb, bindex);
7713 +       if (wh)
7714 +               h_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
7715 +       else
7716 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent);
7717 +       err = PTR_ERR(h_dentry);
7718 +       if (IS_ERR(h_dentry))
7719 +               goto out;
7720 +       if (unlikely(h_dentry->d_inode)) {
7721 +               err = -EIO;
7722 +               AuIOErr("%pd should be negative on b%d.\n", h_dentry, bindex);
7723 +               dput(h_dentry);
7724 +               goto out;
7725 +       }
7726 +
7727 +       err = 0;
7728 +       if (bindex < au_dbstart(dentry))
7729 +               au_set_dbstart(dentry, bindex);
7730 +       if (au_dbend(dentry) < bindex)
7731 +               au_set_dbend(dentry, bindex);
7732 +       au_set_h_dptr(dentry, bindex, h_dentry);
7733 +
7734 +out:
7735 +       dput(parent);
7736 +       return err;
7737 +}
7738 +
7739 +/* ---------------------------------------------------------------------- */
7740 +
7741 +/* subset of struct inode */
7742 +struct au_iattr {
7743 +       unsigned long           i_ino;
7744 +       /* unsigned int         i_nlink; */
7745 +       kuid_t                  i_uid;
7746 +       kgid_t                  i_gid;
7747 +       u64                     i_version;
7748 +/*
7749 +       loff_t                  i_size;
7750 +       blkcnt_t                i_blocks;
7751 +*/
7752 +       umode_t                 i_mode;
7753 +};
7754 +
7755 +static void au_iattr_save(struct au_iattr *ia, struct inode *h_inode)
7756 +{
7757 +       ia->i_ino = h_inode->i_ino;
7758 +       /* ia->i_nlink = h_inode->i_nlink; */
7759 +       ia->i_uid = h_inode->i_uid;
7760 +       ia->i_gid = h_inode->i_gid;
7761 +       ia->i_version = h_inode->i_version;
7762 +/*
7763 +       ia->i_size = h_inode->i_size;
7764 +       ia->i_blocks = h_inode->i_blocks;
7765 +*/
7766 +       ia->i_mode = (h_inode->i_mode & S_IFMT);
7767 +}
7768 +
7769 +static int au_iattr_test(struct au_iattr *ia, struct inode *h_inode)
7770 +{
7771 +       return ia->i_ino != h_inode->i_ino
7772 +               /* || ia->i_nlink != h_inode->i_nlink */
7773 +               || !uid_eq(ia->i_uid, h_inode->i_uid)
7774 +               || !gid_eq(ia->i_gid, h_inode->i_gid)
7775 +               || ia->i_version != h_inode->i_version
7776 +/*
7777 +               || ia->i_size != h_inode->i_size
7778 +               || ia->i_blocks != h_inode->i_blocks
7779 +*/
7780 +               || ia->i_mode != (h_inode->i_mode & S_IFMT);
7781 +}
7782 +
7783 +static int au_h_verify_dentry(struct dentry *h_dentry, struct dentry *h_parent,
7784 +                             struct au_branch *br)
7785 +{
7786 +       int err;
7787 +       struct au_iattr ia;
7788 +       struct inode *h_inode;
7789 +       struct dentry *h_d;
7790 +       struct super_block *h_sb;
7791 +
7792 +       err = 0;
7793 +       memset(&ia, -1, sizeof(ia));
7794 +       h_sb = h_dentry->d_sb;
7795 +       h_inode = h_dentry->d_inode;
7796 +       if (h_inode)
7797 +               au_iattr_save(&ia, h_inode);
7798 +       else if (au_test_nfs(h_sb) || au_test_fuse(h_sb))
7799 +               /* nfs d_revalidate may return 0 for negative dentry */
7800 +               /* fuse d_revalidate always return 0 for negative dentry */
7801 +               goto out;
7802 +
7803 +       /* main purpose is namei.c:cached_lookup() and d_revalidate */
7804 +       h_d = vfsub_lkup_one(&h_dentry->d_name, h_parent);
7805 +       err = PTR_ERR(h_d);
7806 +       if (IS_ERR(h_d))
7807 +               goto out;
7808 +
7809 +       err = 0;
7810 +       if (unlikely(h_d != h_dentry
7811 +                    || h_d->d_inode != h_inode
7812 +                    || (h_inode && au_iattr_test(&ia, h_inode))))
7813 +               err = au_busy_or_stale();
7814 +       dput(h_d);
7815 +
7816 +out:
7817 +       AuTraceErr(err);
7818 +       return err;
7819 +}
7820 +
7821 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
7822 +               struct dentry *h_parent, struct au_branch *br)
7823 +{
7824 +       int err;
7825 +
7826 +       err = 0;
7827 +       if (udba == AuOpt_UDBA_REVAL
7828 +           && !au_test_fs_remote(h_dentry->d_sb)) {
7829 +               IMustLock(h_dir);
7830 +               err = (h_dentry->d_parent->d_inode != h_dir);
7831 +       } else if (udba != AuOpt_UDBA_NONE)
7832 +               err = au_h_verify_dentry(h_dentry, h_parent, br);
7833 +
7834 +       return err;
7835 +}
7836 +
7837 +/* ---------------------------------------------------------------------- */
7838 +
7839 +static int au_do_refresh_hdentry(struct dentry *dentry, struct dentry *parent)
7840 +{
7841 +       int err;
7842 +       aufs_bindex_t new_bindex, bindex, bend, bwh, bdiropq;
7843 +       struct au_hdentry tmp, *p, *q;
7844 +       struct au_dinfo *dinfo;
7845 +       struct super_block *sb;
7846 +
7847 +       DiMustWriteLock(dentry);
7848 +
7849 +       sb = dentry->d_sb;
7850 +       dinfo = au_di(dentry);
7851 +       bend = dinfo->di_bend;
7852 +       bwh = dinfo->di_bwh;
7853 +       bdiropq = dinfo->di_bdiropq;
7854 +       p = dinfo->di_hdentry + dinfo->di_bstart;
7855 +       for (bindex = dinfo->di_bstart; bindex <= bend; bindex++, p++) {
7856 +               if (!p->hd_dentry)
7857 +                       continue;
7858 +
7859 +               new_bindex = au_br_index(sb, p->hd_id);
7860 +               if (new_bindex == bindex)
7861 +                       continue;
7862 +
7863 +               if (dinfo->di_bwh == bindex)
7864 +                       bwh = new_bindex;
7865 +               if (dinfo->di_bdiropq == bindex)
7866 +                       bdiropq = new_bindex;
7867 +               if (new_bindex < 0) {
7868 +                       au_hdput(p);
7869 +                       p->hd_dentry = NULL;
7870 +                       continue;
7871 +               }
7872 +
7873 +               /* swap two lower dentries, and loop again */
7874 +               q = dinfo->di_hdentry + new_bindex;
7875 +               tmp = *q;
7876 +               *q = *p;
7877 +               *p = tmp;
7878 +               if (tmp.hd_dentry) {
7879 +                       bindex--;
7880 +                       p--;
7881 +               }
7882 +       }
7883 +
7884 +       dinfo->di_bwh = -1;
7885 +       if (bwh >= 0 && bwh <= au_sbend(sb) && au_sbr_whable(sb, bwh))
7886 +               dinfo->di_bwh = bwh;
7887 +
7888 +       dinfo->di_bdiropq = -1;
7889 +       if (bdiropq >= 0
7890 +           && bdiropq <= au_sbend(sb)
7891 +           && au_sbr_whable(sb, bdiropq))
7892 +               dinfo->di_bdiropq = bdiropq;
7893 +
7894 +       err = -EIO;
7895 +       dinfo->di_bstart = -1;
7896 +       dinfo->di_bend = -1;
7897 +       bend = au_dbend(parent);
7898 +       p = dinfo->di_hdentry;
7899 +       for (bindex = 0; bindex <= bend; bindex++, p++)
7900 +               if (p->hd_dentry) {
7901 +                       dinfo->di_bstart = bindex;
7902 +                       break;
7903 +               }
7904 +
7905 +       if (dinfo->di_bstart >= 0) {
7906 +               p = dinfo->di_hdentry + bend;
7907 +               for (bindex = bend; bindex >= 0; bindex--, p--)
7908 +                       if (p->hd_dentry) {
7909 +                               dinfo->di_bend = bindex;
7910 +                               err = 0;
7911 +                               break;
7912 +                       }
7913 +       }
7914 +
7915 +       return err;
7916 +}
7917 +
7918 +static void au_do_hide(struct dentry *dentry)
7919 +{
7920 +       struct inode *inode;
7921 +
7922 +       inode = dentry->d_inode;
7923 +       if (inode) {
7924 +               if (!S_ISDIR(inode->i_mode)) {
7925 +                       if (inode->i_nlink && !d_unhashed(dentry))
7926 +                               drop_nlink(inode);
7927 +               } else {
7928 +                       clear_nlink(inode);
7929 +                       /* stop next lookup */
7930 +                       inode->i_flags |= S_DEAD;
7931 +               }
7932 +               smp_mb(); /* necessary? */
7933 +       }
7934 +       d_drop(dentry);
7935 +}
7936 +
7937 +static int au_hide_children(struct dentry *parent)
7938 +{
7939 +       int err, i, j, ndentry;
7940 +       struct au_dcsub_pages dpages;
7941 +       struct au_dpage *dpage;
7942 +       struct dentry *dentry;
7943 +
7944 +       err = au_dpages_init(&dpages, GFP_NOFS);
7945 +       if (unlikely(err))
7946 +               goto out;
7947 +       err = au_dcsub_pages(&dpages, parent, NULL, NULL);
7948 +       if (unlikely(err))
7949 +               goto out_dpages;
7950 +
7951 +       /* in reverse order */
7952 +       for (i = dpages.ndpage - 1; i >= 0; i--) {
7953 +               dpage = dpages.dpages + i;
7954 +               ndentry = dpage->ndentry;
7955 +               for (j = ndentry - 1; j >= 0; j--) {
7956 +                       dentry = dpage->dentries[j];
7957 +                       if (dentry != parent)
7958 +                               au_do_hide(dentry);
7959 +               }
7960 +       }
7961 +
7962 +out_dpages:
7963 +       au_dpages_free(&dpages);
7964 +out:
7965 +       return err;
7966 +}
7967 +
7968 +static void au_hide(struct dentry *dentry)
7969 +{
7970 +       int err;
7971 +
7972 +       AuDbgDentry(dentry);
7973 +       if (d_is_dir(dentry)) {
7974 +               /* shrink_dcache_parent(dentry); */
7975 +               err = au_hide_children(dentry);
7976 +               if (unlikely(err))
7977 +                       AuIOErr("%pd, failed hiding children, ignored %d\n",
7978 +                               dentry, err);
7979 +       }
7980 +       au_do_hide(dentry);
7981 +}
7982 +
7983 +/*
7984 + * By adding a dirty branch, a cached dentry may be affected in various ways.
7985 + *
7986 + * a dirty branch is added
7987 + * - on the top of layers
7988 + * - in the middle of layers
7989 + * - to the bottom of layers
7990 + *
7991 + * on the added branch there exists
7992 + * - a whiteout
7993 + * - a diropq
7994 + * - a same named entry
7995 + *   + exist
7996 + *     * negative --> positive
7997 + *     * positive --> positive
7998 + *      - type is unchanged
7999 + *      - type is changed
8000 + *   + doesn't exist
8001 + *     * negative --> negative
8002 + *     * positive --> negative (rejected by au_br_del() for non-dir case)
8003 + * - none
8004 + */
8005 +static int au_refresh_by_dinfo(struct dentry *dentry, struct au_dinfo *dinfo,
8006 +                              struct au_dinfo *tmp)
8007 +{
8008 +       int err;
8009 +       aufs_bindex_t bindex, bend;
8010 +       struct {
8011 +               struct dentry *dentry;
8012 +               struct inode *inode;
8013 +               mode_t mode;
8014 +       } orig_h, tmp_h;
8015 +       struct au_hdentry *hd;
8016 +       struct inode *inode, *h_inode;
8017 +       struct dentry *h_dentry;
8018 +
8019 +       err = 0;
8020 +       AuDebugOn(dinfo->di_bstart < 0);
8021 +       orig_h.dentry = dinfo->di_hdentry[dinfo->di_bstart].hd_dentry;
8022 +       orig_h.inode = orig_h.dentry->d_inode;
8023 +       orig_h.mode = 0;
8024 +       if (orig_h.inode)
8025 +               orig_h.mode = orig_h.inode->i_mode & S_IFMT;
8026 +       memset(&tmp_h, 0, sizeof(tmp_h));
8027 +       if (tmp->di_bstart >= 0) {
8028 +               tmp_h.dentry = tmp->di_hdentry[tmp->di_bstart].hd_dentry;
8029 +               tmp_h.inode = tmp_h.dentry->d_inode;
8030 +               if (tmp_h.inode)
8031 +                       tmp_h.mode = tmp_h.inode->i_mode & S_IFMT;
8032 +       }
8033 +
8034 +       inode = dentry->d_inode;
8035 +       if (!orig_h.inode) {
8036 +               AuDbg("nagative originally\n");
8037 +               if (inode) {
8038 +                       au_hide(dentry);
8039 +                       goto out;
8040 +               }
8041 +               AuDebugOn(inode);
8042 +               AuDebugOn(dinfo->di_bstart != dinfo->di_bend);
8043 +               AuDebugOn(dinfo->di_bdiropq != -1);
8044 +
8045 +               if (!tmp_h.inode) {
8046 +                       AuDbg("negative --> negative\n");
8047 +                       /* should have only one negative lower */
8048 +                       if (tmp->di_bstart >= 0
8049 +                           && tmp->di_bstart < dinfo->di_bstart) {
8050 +                               AuDebugOn(tmp->di_bstart != tmp->di_bend);
8051 +                               AuDebugOn(dinfo->di_bstart != dinfo->di_bend);
8052 +                               au_set_h_dptr(dentry, dinfo->di_bstart, NULL);
8053 +                               au_di_cp(dinfo, tmp);
8054 +                               hd = tmp->di_hdentry + tmp->di_bstart;
8055 +                               au_set_h_dptr(dentry, tmp->di_bstart,
8056 +                                             dget(hd->hd_dentry));
8057 +                       }
8058 +                       au_dbg_verify_dinode(dentry);
8059 +               } else {
8060 +                       AuDbg("negative --> positive\n");
8061 +                       /*
8062 +                        * similar to the behaviour of creating with bypassing
8063 +                        * aufs.
8064 +                        * unhash it in order to force an error in the
8065 +                        * succeeding create operation.
8066 +                        * we should not set S_DEAD here.
8067 +                        */
8068 +                       d_drop(dentry);
8069 +                       /* au_di_swap(tmp, dinfo); */
8070 +                       au_dbg_verify_dinode(dentry);
8071 +               }
8072 +       } else {
8073 +               AuDbg("positive originally\n");
8074 +               /* inode may be NULL */
8075 +               AuDebugOn(inode && (inode->i_mode & S_IFMT) != orig_h.mode);
8076 +               if (!tmp_h.inode) {
8077 +                       AuDbg("positive --> negative\n");
8078 +                       /* or bypassing aufs */
8079 +                       au_hide(dentry);
8080 +                       if (tmp->di_bwh >= 0 && tmp->di_bwh <= dinfo->di_bstart)
8081 +                               dinfo->di_bwh = tmp->di_bwh;
8082 +                       if (inode)
8083 +                               err = au_refresh_hinode_self(inode);
8084 +                       au_dbg_verify_dinode(dentry);
8085 +               } else if (orig_h.mode == tmp_h.mode) {
8086 +                       AuDbg("positive --> positive, same type\n");
8087 +                       if (!S_ISDIR(orig_h.mode)
8088 +                           && dinfo->di_bstart > tmp->di_bstart) {
8089 +                               /*
8090 +                                * similar to the behaviour of removing and
8091 +                                * creating.
8092 +                                */
8093 +                               au_hide(dentry);
8094 +                               if (inode)
8095 +                                       err = au_refresh_hinode_self(inode);
8096 +                               au_dbg_verify_dinode(dentry);
8097 +                       } else {
8098 +                               /* fill empty slots */
8099 +                               if (dinfo->di_bstart > tmp->di_bstart)
8100 +                                       dinfo->di_bstart = tmp->di_bstart;
8101 +                               if (dinfo->di_bend < tmp->di_bend)
8102 +                                       dinfo->di_bend = tmp->di_bend;
8103 +                               dinfo->di_bwh = tmp->di_bwh;
8104 +                               dinfo->di_bdiropq = tmp->di_bdiropq;
8105 +                               hd = tmp->di_hdentry;
8106 +                               bend = dinfo->di_bend;
8107 +                               for (bindex = tmp->di_bstart; bindex <= bend;
8108 +                                    bindex++) {
8109 +                                       if (au_h_dptr(dentry, bindex))
8110 +                                               continue;
8111 +                                       h_dentry = hd[bindex].hd_dentry;
8112 +                                       if (!h_dentry)
8113 +                                               continue;
8114 +                                       h_inode = h_dentry->d_inode;
8115 +                                       AuDebugOn(!h_inode);
8116 +                                       AuDebugOn(orig_h.mode
8117 +                                                 != (h_inode->i_mode
8118 +                                                     & S_IFMT));
8119 +                                       au_set_h_dptr(dentry, bindex,
8120 +                                                     dget(h_dentry));
8121 +                               }
8122 +                               err = au_refresh_hinode(inode, dentry);
8123 +                               au_dbg_verify_dinode(dentry);
8124 +                       }
8125 +               } else {
8126 +                       AuDbg("positive --> positive, different type\n");
8127 +                       /* similar to the behaviour of removing and creating */
8128 +                       au_hide(dentry);
8129 +                       if (inode)
8130 +                               err = au_refresh_hinode_self(inode);
8131 +                       au_dbg_verify_dinode(dentry);
8132 +               }
8133 +       }
8134 +
8135 +out:
8136 +       return err;
8137 +}
8138 +
8139 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent)
8140 +{
8141 +       int err, ebrange;
8142 +       unsigned int sigen;
8143 +       struct au_dinfo *dinfo, *tmp;
8144 +       struct super_block *sb;
8145 +       struct inode *inode;
8146 +
8147 +       DiMustWriteLock(dentry);
8148 +       AuDebugOn(IS_ROOT(dentry));
8149 +       AuDebugOn(!parent->d_inode);
8150 +
8151 +       sb = dentry->d_sb;
8152 +       inode = dentry->d_inode;
8153 +       sigen = au_sigen(sb);
8154 +       err = au_digen_test(parent, sigen);
8155 +       if (unlikely(err))
8156 +               goto out;
8157 +
8158 +       dinfo = au_di(dentry);
8159 +       err = au_di_realloc(dinfo, au_sbend(sb) + 1);
8160 +       if (unlikely(err))
8161 +               goto out;
8162 +       ebrange = au_dbrange_test(dentry);
8163 +       if (!ebrange)
8164 +               ebrange = au_do_refresh_hdentry(dentry, parent);
8165 +
8166 +       if (d_unhashed(dentry) || ebrange /* || dinfo->di_tmpfile */) {
8167 +               AuDebugOn(au_dbstart(dentry) < 0 && au_dbend(dentry) >= 0);
8168 +               if (inode)
8169 +                       err = au_refresh_hinode_self(inode);
8170 +               au_dbg_verify_dinode(dentry);
8171 +               if (!err)
8172 +                       goto out_dgen; /* success */
8173 +               goto out;
8174 +       }
8175 +
8176 +       /* temporary dinfo */
8177 +       AuDbgDentry(dentry);
8178 +       err = -ENOMEM;
8179 +       tmp = au_di_alloc(sb, AuLsc_DI_TMP);
8180 +       if (unlikely(!tmp))
8181 +               goto out;
8182 +       au_di_swap(tmp, dinfo);
8183 +       /* returns the number of positive dentries */
8184 +       /*
8185 +        * if current working dir is removed, it returns an error.
8186 +        * but the dentry is legal.
8187 +        */
8188 +       err = au_lkup_dentry(dentry, /*bstart*/0, /*type*/0);
8189 +       AuDbgDentry(dentry);
8190 +       au_di_swap(tmp, dinfo);
8191 +       if (err == -ENOENT)
8192 +               err = 0;
8193 +       if (err >= 0) {
8194 +               /* compare/refresh by dinfo */
8195 +               AuDbgDentry(dentry);
8196 +               err = au_refresh_by_dinfo(dentry, dinfo, tmp);
8197 +               au_dbg_verify_dinode(dentry);
8198 +               AuTraceErr(err);
8199 +       }
8200 +       au_rw_write_unlock(&tmp->di_rwsem);
8201 +       au_di_free(tmp);
8202 +       if (unlikely(err))
8203 +               goto out;
8204 +
8205 +out_dgen:
8206 +       au_update_digen(dentry);
8207 +out:
8208 +       if (unlikely(err && !(dentry->d_flags & DCACHE_NFSFS_RENAMED))) {
8209 +               AuIOErr("failed refreshing %pd, %d\n", dentry, err);
8210 +               AuDbgDentry(dentry);
8211 +       }
8212 +       AuTraceErr(err);
8213 +       return err;
8214 +}
8215 +
8216 +static int au_do_h_d_reval(struct dentry *h_dentry, unsigned int flags,
8217 +                          struct dentry *dentry, aufs_bindex_t bindex)
8218 +{
8219 +       int err, valid;
8220 +
8221 +       err = 0;
8222 +       if (!(h_dentry->d_flags & DCACHE_OP_REVALIDATE))
8223 +               goto out;
8224 +
8225 +       AuDbg("b%d\n", bindex);
8226 +       /*
8227 +        * gave up supporting LOOKUP_CREATE/OPEN for lower fs,
8228 +        * due to whiteout and branch permission.
8229 +        */
8230 +       flags &= ~(/*LOOKUP_PARENT |*/ LOOKUP_OPEN | LOOKUP_CREATE
8231 +                  | LOOKUP_FOLLOW | LOOKUP_EXCL);
8232 +       /* it may return tri-state */
8233 +       valid = h_dentry->d_op->d_revalidate(h_dentry, flags);
8234 +
8235 +       if (unlikely(valid < 0))
8236 +               err = valid;
8237 +       else if (!valid)
8238 +               err = -EINVAL;
8239 +
8240 +out:
8241 +       AuTraceErr(err);
8242 +       return err;
8243 +}
8244 +
8245 +/* todo: remove this */
8246 +static int h_d_revalidate(struct dentry *dentry, struct inode *inode,
8247 +                         unsigned int flags, int do_udba)
8248 +{
8249 +       int err;
8250 +       umode_t mode, h_mode;
8251 +       aufs_bindex_t bindex, btail, bstart, ibs, ibe;
8252 +       unsigned char plus, unhashed, is_root, h_plus, h_nfs, tmpfile;
8253 +       struct inode *h_inode, *h_cached_inode;
8254 +       struct dentry *h_dentry;
8255 +       struct qstr *name, *h_name;
8256 +
8257 +       err = 0;
8258 +       plus = 0;
8259 +       mode = 0;
8260 +       ibs = -1;
8261 +       ibe = -1;
8262 +       unhashed = !!d_unhashed(dentry);
8263 +       is_root = !!IS_ROOT(dentry);
8264 +       name = &dentry->d_name;
8265 +       tmpfile = au_di(dentry)->di_tmpfile;
8266 +
8267 +       /*
8268 +        * Theoretically, REVAL test should be unnecessary in case of
8269 +        * {FS,I}NOTIFY.
8270 +        * But {fs,i}notify doesn't fire some necessary events,
8271 +        *      IN_ATTRIB for atime/nlink/pageio
8272 +        * Let's do REVAL test too.
8273 +        */
8274 +       if (do_udba && inode) {
8275 +               mode = (inode->i_mode & S_IFMT);
8276 +               plus = (inode->i_nlink > 0);
8277 +               ibs = au_ibstart(inode);
8278 +               ibe = au_ibend(inode);
8279 +       }
8280 +
8281 +       bstart = au_dbstart(dentry);
8282 +       btail = bstart;
8283 +       if (inode && S_ISDIR(inode->i_mode))
8284 +               btail = au_dbtaildir(dentry);
8285 +       for (bindex = bstart; bindex <= btail; bindex++) {
8286 +               h_dentry = au_h_dptr(dentry, bindex);
8287 +               if (!h_dentry)
8288 +                       continue;
8289 +
8290 +               AuDbg("b%d, %pd\n", bindex, h_dentry);
8291 +               h_nfs = !!au_test_nfs(h_dentry->d_sb);
8292 +               spin_lock(&h_dentry->d_lock);
8293 +               h_name = &h_dentry->d_name;
8294 +               if (unlikely(do_udba
8295 +                            && !is_root
8296 +                            && ((!h_nfs
8297 +                                 && (unhashed != !!d_unhashed(h_dentry)
8298 +                                     || (!tmpfile
8299 +                                         && !au_qstreq(name, h_name))
8300 +                                         ))
8301 +                                || (h_nfs
8302 +                                    && !(flags & LOOKUP_OPEN)
8303 +                                    && (h_dentry->d_flags
8304 +                                        & DCACHE_NFSFS_RENAMED)))
8305 +                           )) {
8306 +                       int h_unhashed;
8307 +
8308 +                       h_unhashed = d_unhashed(h_dentry);
8309 +                       spin_unlock(&h_dentry->d_lock);
8310 +                       AuDbg("unhash 0x%x 0x%x, %pd %pd\n",
8311 +                             unhashed, h_unhashed, dentry, h_dentry);
8312 +                       goto err;
8313 +               }
8314 +               spin_unlock(&h_dentry->d_lock);
8315 +
8316 +               err = au_do_h_d_reval(h_dentry, flags, dentry, bindex);
8317 +               if (unlikely(err))
8318 +                       /* do not goto err, to keep the errno */
8319 +                       break;
8320 +
8321 +               /* todo: plink too? */
8322 +               if (!do_udba)
8323 +                       continue;
8324 +
8325 +               /* UDBA tests */
8326 +               h_inode = h_dentry->d_inode;
8327 +               if (unlikely(!!inode != !!h_inode))
8328 +                       goto err;
8329 +
8330 +               h_plus = plus;
8331 +               h_mode = mode;
8332 +               h_cached_inode = h_inode;
8333 +               if (h_inode) {
8334 +                       h_mode = (h_inode->i_mode & S_IFMT);
8335 +                       h_plus = (h_inode->i_nlink > 0);
8336 +               }
8337 +               if (inode && ibs <= bindex && bindex <= ibe)
8338 +                       h_cached_inode = au_h_iptr(inode, bindex);
8339 +
8340 +               if (!h_nfs) {
8341 +                       if (unlikely(plus != h_plus && !tmpfile))
8342 +                               goto err;
8343 +               } else {
8344 +                       if (unlikely(!(h_dentry->d_flags & DCACHE_NFSFS_RENAMED)
8345 +                                    && !is_root
8346 +                                    && !IS_ROOT(h_dentry)
8347 +                                    && unhashed != d_unhashed(h_dentry)))
8348 +                               goto err;
8349 +               }
8350 +               if (unlikely(mode != h_mode
8351 +                            || h_cached_inode != h_inode))
8352 +                       goto err;
8353 +               continue;
8354 +
8355 +err:
8356 +               err = -EINVAL;
8357 +               break;
8358 +       }
8359 +
8360 +       AuTraceErr(err);
8361 +       return err;
8362 +}
8363 +
8364 +/* todo: consolidate with do_refresh() and au_reval_for_attr() */
8365 +static int simple_reval_dpath(struct dentry *dentry, unsigned int sigen)
8366 +{
8367 +       int err;
8368 +       struct dentry *parent;
8369 +
8370 +       if (!au_digen_test(dentry, sigen))
8371 +               return 0;
8372 +
8373 +       parent = dget_parent(dentry);
8374 +       di_read_lock_parent(parent, AuLock_IR);
8375 +       AuDebugOn(au_digen_test(parent, sigen));
8376 +       au_dbg_verify_gen(parent, sigen);
8377 +       err = au_refresh_dentry(dentry, parent);
8378 +       di_read_unlock(parent, AuLock_IR);
8379 +       dput(parent);
8380 +       AuTraceErr(err);
8381 +       return err;
8382 +}
8383 +
8384 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen)
8385 +{
8386 +       int err;
8387 +       struct dentry *d, *parent;
8388 +       struct inode *inode;
8389 +
8390 +       if (!au_ftest_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR))
8391 +               return simple_reval_dpath(dentry, sigen);
8392 +
8393 +       /* slow loop, keep it simple and stupid */
8394 +       /* cf: au_cpup_dirs() */
8395 +       err = 0;
8396 +       parent = NULL;
8397 +       while (au_digen_test(dentry, sigen)) {
8398 +               d = dentry;
8399 +               while (1) {
8400 +                       dput(parent);
8401 +                       parent = dget_parent(d);
8402 +                       if (!au_digen_test(parent, sigen))
8403 +                               break;
8404 +                       d = parent;
8405 +               }
8406 +
8407 +               inode = d->d_inode;
8408 +               if (d != dentry)
8409 +                       di_write_lock_child2(d);
8410 +
8411 +               /* someone might update our dentry while we were sleeping */
8412 +               if (au_digen_test(d, sigen)) {
8413 +                       /*
8414 +                        * todo: consolidate with simple_reval_dpath(),
8415 +                        * do_refresh() and au_reval_for_attr().
8416 +                        */
8417 +                       di_read_lock_parent(parent, AuLock_IR);
8418 +                       err = au_refresh_dentry(d, parent);
8419 +                       di_read_unlock(parent, AuLock_IR);
8420 +               }
8421 +
8422 +               if (d != dentry)
8423 +                       di_write_unlock(d);
8424 +               dput(parent);
8425 +               if (unlikely(err))
8426 +                       break;
8427 +       }
8428 +
8429 +       return err;
8430 +}
8431 +
8432 +/*
8433 + * if valid returns 1, otherwise 0.
8434 + */
8435 +static int aufs_d_revalidate(struct dentry *dentry, unsigned int flags)
8436 +{
8437 +       int valid, err;
8438 +       unsigned int sigen;
8439 +       unsigned char do_udba;
8440 +       struct super_block *sb;
8441 +       struct inode *inode;
8442 +
8443 +       /* todo: support rcu-walk? */
8444 +       if (flags & LOOKUP_RCU)
8445 +               return -ECHILD;
8446 +
8447 +       valid = 0;
8448 +       if (unlikely(!au_di(dentry)))
8449 +               goto out;
8450 +
8451 +       valid = 1;
8452 +       sb = dentry->d_sb;
8453 +       /*
8454 +        * todo: very ugly
8455 +        * i_mutex of parent dir may be held,
8456 +        * but we should not return 'invalid' due to busy.
8457 +        */
8458 +       err = aufs_read_lock(dentry, AuLock_FLUSH | AuLock_DW | AuLock_NOPLM);
8459 +       if (unlikely(err)) {
8460 +               valid = err;
8461 +               AuTraceErr(err);
8462 +               goto out;
8463 +       }
8464 +       inode = dentry->d_inode;
8465 +       if (unlikely(inode && is_bad_inode(inode))) {
8466 +               err = -EINVAL;
8467 +               AuTraceErr(err);
8468 +               goto out_dgrade;
8469 +       }
8470 +       if (unlikely(au_dbrange_test(dentry))) {
8471 +               err = -EINVAL;
8472 +               AuTraceErr(err);
8473 +               goto out_dgrade;
8474 +       }
8475 +
8476 +       sigen = au_sigen(sb);
8477 +       if (au_digen_test(dentry, sigen)) {
8478 +               AuDebugOn(IS_ROOT(dentry));
8479 +               err = au_reval_dpath(dentry, sigen);
8480 +               if (unlikely(err)) {
8481 +                       AuTraceErr(err);
8482 +                       goto out_dgrade;
8483 +               }
8484 +       }
8485 +       di_downgrade_lock(dentry, AuLock_IR);
8486 +
8487 +       err = -EINVAL;
8488 +       if (!(flags & (LOOKUP_OPEN | LOOKUP_EMPTY))
8489 +           && inode
8490 +           && !(inode->i_state && I_LINKABLE)
8491 +           && (IS_DEADDIR(inode) || !inode->i_nlink))
8492 +               goto out_inval;
8493 +
8494 +       do_udba = !au_opt_test(au_mntflags(sb), UDBA_NONE);
8495 +       if (do_udba && inode) {
8496 +               aufs_bindex_t bstart = au_ibstart(inode);
8497 +               struct inode *h_inode;
8498 +
8499 +               if (bstart >= 0) {
8500 +                       h_inode = au_h_iptr(inode, bstart);
8501 +                       if (h_inode && au_test_higen(inode, h_inode))
8502 +                               goto out_inval;
8503 +               }
8504 +       }
8505 +
8506 +       err = h_d_revalidate(dentry, inode, flags, do_udba);
8507 +       if (unlikely(!err && do_udba && au_dbstart(dentry) < 0)) {
8508 +               err = -EIO;
8509 +               AuDbg("both of real entry and whiteout found, %p, err %d\n",
8510 +                     dentry, err);
8511 +       }
8512 +       goto out_inval;
8513 +
8514 +out_dgrade:
8515 +       di_downgrade_lock(dentry, AuLock_IR);
8516 +out_inval:
8517 +       aufs_read_unlock(dentry, AuLock_IR);
8518 +       AuTraceErr(err);
8519 +       valid = !err;
8520 +out:
8521 +       if (!valid) {
8522 +               AuDbg("%pd invalid, %d\n", dentry, valid);
8523 +               d_drop(dentry);
8524 +       }
8525 +       return valid;
8526 +}
8527 +
8528 +static void aufs_d_release(struct dentry *dentry)
8529 +{
8530 +       if (au_di(dentry)) {
8531 +               au_di_fin(dentry);
8532 +               au_hn_di_reinit(dentry);
8533 +       }
8534 +}
8535 +
8536 +const struct dentry_operations aufs_dop = {
8537 +       .d_revalidate           = aufs_d_revalidate,
8538 +       .d_weak_revalidate      = aufs_d_revalidate,
8539 +       .d_release              = aufs_d_release
8540 +};
8541 diff -urN /usr/share/empty/fs/aufs/dentry.h linux/fs/aufs/dentry.h
8542 --- /usr/share/empty/fs/aufs/dentry.h   1970-01-01 01:00:00.000000000 +0100
8543 +++ linux/fs/aufs/dentry.h      2015-03-27 21:56:35.463461668 +0100
8544 @@ -0,0 +1,233 @@
8545 +/*
8546 + * Copyright (C) 2005-2015 Junjiro R. Okajima
8547 + *
8548 + * This program, aufs is free software; you can redistribute it and/or modify
8549 + * it under the terms of the GNU General Public License as published by
8550 + * the Free Software Foundation; either version 2 of the License, or
8551 + * (at your option) any later version.
8552 + *
8553 + * This program is distributed in the hope that it will be useful,
8554 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8555 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8556 + * GNU General Public License for more details.
8557 + *
8558 + * You should have received a copy of the GNU General Public License
8559 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8560 + */
8561 +
8562 +/*
8563 + * lookup and dentry operations
8564 + */
8565 +
8566 +#ifndef __AUFS_DENTRY_H__
8567 +#define __AUFS_DENTRY_H__
8568 +
8569 +#ifdef __KERNEL__
8570 +
8571 +#include <linux/dcache.h>
8572 +#include "rwsem.h"
8573 +
8574 +struct au_hdentry {
8575 +       struct dentry           *hd_dentry;
8576 +       aufs_bindex_t           hd_id;
8577 +};
8578 +
8579 +struct au_dinfo {
8580 +       atomic_t                di_generation;
8581 +
8582 +       struct au_rwsem         di_rwsem;
8583 +       aufs_bindex_t           di_bstart, di_bend, di_bwh, di_bdiropq;
8584 +       unsigned char           di_tmpfile; /* to allow the different name */
8585 +       struct au_hdentry       *di_hdentry;
8586 +} ____cacheline_aligned_in_smp;
8587 +
8588 +/* ---------------------------------------------------------------------- */
8589 +
8590 +/* dentry.c */
8591 +extern const struct dentry_operations aufs_dop;
8592 +struct au_branch;
8593 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent);
8594 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
8595 +               struct dentry *h_parent, struct au_branch *br);
8596 +
8597 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t bstart, mode_t type);
8598 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh);
8599 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent);
8600 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen);
8601 +
8602 +/* dinfo.c */
8603 +void au_di_init_once(void *_di);
8604 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc);
8605 +void au_di_free(struct au_dinfo *dinfo);
8606 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b);
8607 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src);
8608 +int au_di_init(struct dentry *dentry);
8609 +void au_di_fin(struct dentry *dentry);
8610 +int au_di_realloc(struct au_dinfo *dinfo, int nbr);
8611 +
8612 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc);
8613 +void di_read_unlock(struct dentry *d, int flags);
8614 +void di_downgrade_lock(struct dentry *d, int flags);
8615 +void di_write_lock(struct dentry *d, unsigned int lsc);
8616 +void di_write_unlock(struct dentry *d);
8617 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir);
8618 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir);
8619 +void di_write_unlock2(struct dentry *d1, struct dentry *d2);
8620 +
8621 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex);
8622 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex);
8623 +aufs_bindex_t au_dbtail(struct dentry *dentry);
8624 +aufs_bindex_t au_dbtaildir(struct dentry *dentry);
8625 +
8626 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
8627 +                  struct dentry *h_dentry);
8628 +int au_digen_test(struct dentry *dentry, unsigned int sigen);
8629 +int au_dbrange_test(struct dentry *dentry);
8630 +void au_update_digen(struct dentry *dentry);
8631 +void au_update_dbrange(struct dentry *dentry, int do_put_zero);
8632 +void au_update_dbstart(struct dentry *dentry);
8633 +void au_update_dbend(struct dentry *dentry);
8634 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry);
8635 +
8636 +/* ---------------------------------------------------------------------- */
8637 +
8638 +static inline struct au_dinfo *au_di(struct dentry *dentry)
8639 +{
8640 +       return dentry->d_fsdata;
8641 +}
8642 +
8643 +/* ---------------------------------------------------------------------- */
8644 +
8645 +/* lock subclass for dinfo */
8646 +enum {
8647 +       AuLsc_DI_CHILD,         /* child first */
8648 +       AuLsc_DI_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
8649 +       AuLsc_DI_CHILD3,        /* copyup dirs */
8650 +       AuLsc_DI_PARENT,
8651 +       AuLsc_DI_PARENT2,
8652 +       AuLsc_DI_PARENT3,
8653 +       AuLsc_DI_TMP            /* temp for replacing dinfo */
8654 +};
8655 +
8656 +/*
8657 + * di_read_lock_child, di_write_lock_child,
8658 + * di_read_lock_child2, di_write_lock_child2,
8659 + * di_read_lock_child3, di_write_lock_child3,
8660 + * di_read_lock_parent, di_write_lock_parent,
8661 + * di_read_lock_parent2, di_write_lock_parent2,
8662 + * di_read_lock_parent3, di_write_lock_parent3,
8663 + */
8664 +#define AuReadLockFunc(name, lsc) \
8665 +static inline void di_read_lock_##name(struct dentry *d, int flags) \
8666 +{ di_read_lock(d, flags, AuLsc_DI_##lsc); }
8667 +
8668 +#define AuWriteLockFunc(name, lsc) \
8669 +static inline void di_write_lock_##name(struct dentry *d) \
8670 +{ di_write_lock(d, AuLsc_DI_##lsc); }
8671 +
8672 +#define AuRWLockFuncs(name, lsc) \
8673 +       AuReadLockFunc(name, lsc) \
8674 +       AuWriteLockFunc(name, lsc)
8675 +
8676 +AuRWLockFuncs(child, CHILD);
8677 +AuRWLockFuncs(child2, CHILD2);
8678 +AuRWLockFuncs(child3, CHILD3);
8679 +AuRWLockFuncs(parent, PARENT);
8680 +AuRWLockFuncs(parent2, PARENT2);
8681 +AuRWLockFuncs(parent3, PARENT3);
8682 +
8683 +#undef AuReadLockFunc
8684 +#undef AuWriteLockFunc
8685 +#undef AuRWLockFuncs
8686 +
8687 +#define DiMustNoWaiters(d)     AuRwMustNoWaiters(&au_di(d)->di_rwsem)
8688 +#define DiMustAnyLock(d)       AuRwMustAnyLock(&au_di(d)->di_rwsem)
8689 +#define DiMustWriteLock(d)     AuRwMustWriteLock(&au_di(d)->di_rwsem)
8690 +
8691 +/* ---------------------------------------------------------------------- */
8692 +
8693 +/* todo: memory barrier? */
8694 +static inline unsigned int au_digen(struct dentry *d)
8695 +{
8696 +       return atomic_read(&au_di(d)->di_generation);
8697 +}
8698 +
8699 +static inline void au_h_dentry_init(struct au_hdentry *hdentry)
8700 +{
8701 +       hdentry->hd_dentry = NULL;
8702 +}
8703 +
8704 +static inline void au_hdput(struct au_hdentry *hd)
8705 +{
8706 +       if (hd)
8707 +               dput(hd->hd_dentry);
8708 +}
8709 +
8710 +static inline aufs_bindex_t au_dbstart(struct dentry *dentry)
8711 +{
8712 +       DiMustAnyLock(dentry);
8713 +       return au_di(dentry)->di_bstart;
8714 +}
8715 +
8716 +static inline aufs_bindex_t au_dbend(struct dentry *dentry)
8717 +{
8718 +       DiMustAnyLock(dentry);
8719 +       return au_di(dentry)->di_bend;
8720 +}
8721 +
8722 +static inline aufs_bindex_t au_dbwh(struct dentry *dentry)
8723 +{
8724 +       DiMustAnyLock(dentry);
8725 +       return au_di(dentry)->di_bwh;
8726 +}
8727 +
8728 +static inline aufs_bindex_t au_dbdiropq(struct dentry *dentry)
8729 +{
8730 +       DiMustAnyLock(dentry);
8731 +       return au_di(dentry)->di_bdiropq;
8732 +}
8733 +
8734 +/* todo: hard/soft set? */
8735 +static inline void au_set_dbstart(struct dentry *dentry, aufs_bindex_t bindex)
8736 +{
8737 +       DiMustWriteLock(dentry);
8738 +       au_di(dentry)->di_bstart = bindex;
8739 +}
8740 +
8741 +static inline void au_set_dbend(struct dentry *dentry, aufs_bindex_t bindex)
8742 +{
8743 +       DiMustWriteLock(dentry);
8744 +       au_di(dentry)->di_bend = bindex;
8745 +}
8746 +
8747 +static inline void au_set_dbwh(struct dentry *dentry, aufs_bindex_t bindex)
8748 +{
8749 +       DiMustWriteLock(dentry);
8750 +       /* dbwh can be outside of bstart - bend range */
8751 +       au_di(dentry)->di_bwh = bindex;
8752 +}
8753 +
8754 +static inline void au_set_dbdiropq(struct dentry *dentry, aufs_bindex_t bindex)
8755 +{
8756 +       DiMustWriteLock(dentry);
8757 +       au_di(dentry)->di_bdiropq = bindex;
8758 +}
8759 +
8760 +/* ---------------------------------------------------------------------- */
8761 +
8762 +#ifdef CONFIG_AUFS_HNOTIFY
8763 +static inline void au_digen_dec(struct dentry *d)
8764 +{
8765 +       atomic_dec(&au_di(d)->di_generation);
8766 +}
8767 +
8768 +static inline void au_hn_di_reinit(struct dentry *dentry)
8769 +{
8770 +       dentry->d_fsdata = NULL;
8771 +}
8772 +#else
8773 +AuStubVoid(au_hn_di_reinit, struct dentry *dentry __maybe_unused)
8774 +#endif /* CONFIG_AUFS_HNOTIFY */
8775 +
8776 +#endif /* __KERNEL__ */
8777 +#endif /* __AUFS_DENTRY_H__ */
8778 diff -urN /usr/share/empty/fs/aufs/dinfo.c linux/fs/aufs/dinfo.c
8779 --- /usr/share/empty/fs/aufs/dinfo.c    1970-01-01 01:00:00.000000000 +0100
8780 +++ linux/fs/aufs/dinfo.c       2015-03-27 21:56:35.463461668 +0100
8781 @@ -0,0 +1,544 @@
8782 +/*
8783 + * Copyright (C) 2005-2015 Junjiro R. Okajima
8784 + *
8785 + * This program, aufs is free software; you can redistribute it and/or modify
8786 + * it under the terms of the GNU General Public License as published by
8787 + * the Free Software Foundation; either version 2 of the License, or
8788 + * (at your option) any later version.
8789 + *
8790 + * This program is distributed in the hope that it will be useful,
8791 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8792 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8793 + * GNU General Public License for more details.
8794 + *
8795 + * You should have received a copy of the GNU General Public License
8796 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8797 + */
8798 +
8799 +/*
8800 + * dentry private data
8801 + */
8802 +
8803 +#include "aufs.h"
8804 +
8805 +void au_di_init_once(void *_dinfo)
8806 +{
8807 +       struct au_dinfo *dinfo = _dinfo;
8808 +       static struct lock_class_key aufs_di;
8809 +
8810 +       au_rw_init(&dinfo->di_rwsem);
8811 +       au_rw_class(&dinfo->di_rwsem, &aufs_di);
8812 +}
8813 +
8814 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc)
8815 +{
8816 +       struct au_dinfo *dinfo;
8817 +       int nbr, i;
8818 +
8819 +       dinfo = au_cache_alloc_dinfo();
8820 +       if (unlikely(!dinfo))
8821 +               goto out;
8822 +
8823 +       nbr = au_sbend(sb) + 1;
8824 +       if (nbr <= 0)
8825 +               nbr = 1;
8826 +       dinfo->di_hdentry = kcalloc(nbr, sizeof(*dinfo->di_hdentry), GFP_NOFS);
8827 +       if (dinfo->di_hdentry) {
8828 +               au_rw_write_lock_nested(&dinfo->di_rwsem, lsc);
8829 +               dinfo->di_bstart = -1;
8830 +               dinfo->di_bend = -1;
8831 +               dinfo->di_bwh = -1;
8832 +               dinfo->di_bdiropq = -1;
8833 +               dinfo->di_tmpfile = 0;
8834 +               for (i = 0; i < nbr; i++)
8835 +                       dinfo->di_hdentry[i].hd_id = -1;
8836 +               goto out;
8837 +       }
8838 +
8839 +       au_cache_free_dinfo(dinfo);
8840 +       dinfo = NULL;
8841 +
8842 +out:
8843 +       return dinfo;
8844 +}
8845 +
8846 +void au_di_free(struct au_dinfo *dinfo)
8847 +{
8848 +       struct au_hdentry *p;
8849 +       aufs_bindex_t bend, bindex;
8850 +
8851 +       /* dentry may not be revalidated */
8852 +       bindex = dinfo->di_bstart;
8853 +       if (bindex >= 0) {
8854 +               bend = dinfo->di_bend;
8855 +               p = dinfo->di_hdentry + bindex;
8856 +               while (bindex++ <= bend)
8857 +                       au_hdput(p++);
8858 +       }
8859 +       kfree(dinfo->di_hdentry);
8860 +       au_cache_free_dinfo(dinfo);
8861 +}
8862 +
8863 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b)
8864 +{
8865 +       struct au_hdentry *p;
8866 +       aufs_bindex_t bi;
8867 +
8868 +       AuRwMustWriteLock(&a->di_rwsem);
8869 +       AuRwMustWriteLock(&b->di_rwsem);
8870 +
8871 +#define DiSwap(v, name)                                \
8872 +       do {                                    \
8873 +               v = a->di_##name;               \
8874 +               a->di_##name = b->di_##name;    \
8875 +               b->di_##name = v;               \
8876 +       } while (0)
8877 +
8878 +       DiSwap(p, hdentry);
8879 +       DiSwap(bi, bstart);
8880 +       DiSwap(bi, bend);
8881 +       DiSwap(bi, bwh);
8882 +       DiSwap(bi, bdiropq);
8883 +       /* smp_mb(); */
8884 +
8885 +#undef DiSwap
8886 +}
8887 +
8888 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src)
8889 +{
8890 +       AuRwMustWriteLock(&dst->di_rwsem);
8891 +       AuRwMustWriteLock(&src->di_rwsem);
8892 +
8893 +       dst->di_bstart = src->di_bstart;
8894 +       dst->di_bend = src->di_bend;
8895 +       dst->di_bwh = src->di_bwh;
8896 +       dst->di_bdiropq = src->di_bdiropq;
8897 +       /* smp_mb(); */
8898 +}
8899 +
8900 +int au_di_init(struct dentry *dentry)
8901 +{
8902 +       int err;
8903 +       struct super_block *sb;
8904 +       struct au_dinfo *dinfo;
8905 +
8906 +       err = 0;
8907 +       sb = dentry->d_sb;
8908 +       dinfo = au_di_alloc(sb, AuLsc_DI_CHILD);
8909 +       if (dinfo) {
8910 +               atomic_set(&dinfo->di_generation, au_sigen(sb));
8911 +               /* smp_mb(); */ /* atomic_set */
8912 +               dentry->d_fsdata = dinfo;
8913 +       } else
8914 +               err = -ENOMEM;
8915 +
8916 +       return err;
8917 +}
8918 +
8919 +void au_di_fin(struct dentry *dentry)
8920 +{
8921 +       struct au_dinfo *dinfo;
8922 +
8923 +       dinfo = au_di(dentry);
8924 +       AuRwDestroy(&dinfo->di_rwsem);
8925 +       au_di_free(dinfo);
8926 +}
8927 +
8928 +int au_di_realloc(struct au_dinfo *dinfo, int nbr)
8929 +{
8930 +       int err, sz;
8931 +       struct au_hdentry *hdp;
8932 +
8933 +       AuRwMustWriteLock(&dinfo->di_rwsem);
8934 +
8935 +       err = -ENOMEM;
8936 +       sz = sizeof(*hdp) * (dinfo->di_bend + 1);
8937 +       if (!sz)
8938 +               sz = sizeof(*hdp);
8939 +       hdp = au_kzrealloc(dinfo->di_hdentry, sz, sizeof(*hdp) * nbr, GFP_NOFS);
8940 +       if (hdp) {
8941 +               dinfo->di_hdentry = hdp;
8942 +               err = 0;
8943 +       }
8944 +
8945 +       return err;
8946 +}
8947 +
8948 +/* ---------------------------------------------------------------------- */
8949 +
8950 +static void do_ii_write_lock(struct inode *inode, unsigned int lsc)
8951 +{
8952 +       switch (lsc) {
8953 +       case AuLsc_DI_CHILD:
8954 +               ii_write_lock_child(inode);
8955 +               break;
8956 +       case AuLsc_DI_CHILD2:
8957 +               ii_write_lock_child2(inode);
8958 +               break;
8959 +       case AuLsc_DI_CHILD3:
8960 +               ii_write_lock_child3(inode);
8961 +               break;
8962 +       case AuLsc_DI_PARENT:
8963 +               ii_write_lock_parent(inode);
8964 +               break;
8965 +       case AuLsc_DI_PARENT2:
8966 +               ii_write_lock_parent2(inode);
8967 +               break;
8968 +       case AuLsc_DI_PARENT3:
8969 +               ii_write_lock_parent3(inode);
8970 +               break;
8971 +       default:
8972 +               BUG();
8973 +       }
8974 +}
8975 +
8976 +static void do_ii_read_lock(struct inode *inode, unsigned int lsc)
8977 +{
8978 +       switch (lsc) {
8979 +       case AuLsc_DI_CHILD:
8980 +               ii_read_lock_child(inode);
8981 +               break;
8982 +       case AuLsc_DI_CHILD2:
8983 +               ii_read_lock_child2(inode);
8984 +               break;
8985 +       case AuLsc_DI_CHILD3:
8986 +               ii_read_lock_child3(inode);
8987 +               break;
8988 +       case AuLsc_DI_PARENT:
8989 +               ii_read_lock_parent(inode);
8990 +               break;
8991 +       case AuLsc_DI_PARENT2:
8992 +               ii_read_lock_parent2(inode);
8993 +               break;
8994 +       case AuLsc_DI_PARENT3:
8995 +               ii_read_lock_parent3(inode);
8996 +               break;
8997 +       default:
8998 +               BUG();
8999 +       }
9000 +}
9001 +
9002 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc)
9003 +{
9004 +       au_rw_read_lock_nested(&au_di(d)->di_rwsem, lsc);
9005 +       if (d->d_inode) {
9006 +               if (au_ftest_lock(flags, IW))
9007 +                       do_ii_write_lock(d->d_inode, lsc);
9008 +               else if (au_ftest_lock(flags, IR))
9009 +                       do_ii_read_lock(d->d_inode, lsc);
9010 +       }
9011 +}
9012 +
9013 +void di_read_unlock(struct dentry *d, int flags)
9014 +{
9015 +       if (d->d_inode) {
9016 +               if (au_ftest_lock(flags, IW)) {
9017 +                       au_dbg_verify_dinode(d);
9018 +                       ii_write_unlock(d->d_inode);
9019 +               } else if (au_ftest_lock(flags, IR)) {
9020 +                       au_dbg_verify_dinode(d);
9021 +                       ii_read_unlock(d->d_inode);
9022 +               }
9023 +       }
9024 +       au_rw_read_unlock(&au_di(d)->di_rwsem);
9025 +}
9026 +
9027 +void di_downgrade_lock(struct dentry *d, int flags)
9028 +{
9029 +       if (d->d_inode && au_ftest_lock(flags, IR))
9030 +               ii_downgrade_lock(d->d_inode);
9031 +       au_rw_dgrade_lock(&au_di(d)->di_rwsem);
9032 +}
9033 +
9034 +void di_write_lock(struct dentry *d, unsigned int lsc)
9035 +{
9036 +       au_rw_write_lock_nested(&au_di(d)->di_rwsem, lsc);
9037 +       if (d->d_inode)
9038 +               do_ii_write_lock(d->d_inode, lsc);
9039 +}
9040 +
9041 +void di_write_unlock(struct dentry *d)
9042 +{
9043 +       au_dbg_verify_dinode(d);
9044 +       if (d->d_inode)
9045 +               ii_write_unlock(d->d_inode);
9046 +       au_rw_write_unlock(&au_di(d)->di_rwsem);
9047 +}
9048 +
9049 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir)
9050 +{
9051 +       AuDebugOn(d1 == d2
9052 +                 || d1->d_inode == d2->d_inode
9053 +                 || d1->d_sb != d2->d_sb);
9054 +
9055 +       if (isdir && au_test_subdir(d1, d2)) {
9056 +               di_write_lock_child(d1);
9057 +               di_write_lock_child2(d2);
9058 +       } else {
9059 +               /* there should be no races */
9060 +               di_write_lock_child(d2);
9061 +               di_write_lock_child2(d1);
9062 +       }
9063 +}
9064 +
9065 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir)
9066 +{
9067 +       AuDebugOn(d1 == d2
9068 +                 || d1->d_inode == d2->d_inode
9069 +                 || d1->d_sb != d2->d_sb);
9070 +
9071 +       if (isdir && au_test_subdir(d1, d2)) {
9072 +               di_write_lock_parent(d1);
9073 +               di_write_lock_parent2(d2);
9074 +       } else {
9075 +               /* there should be no races */
9076 +               di_write_lock_parent(d2);
9077 +               di_write_lock_parent2(d1);
9078 +       }
9079 +}
9080 +
9081 +void di_write_unlock2(struct dentry *d1, struct dentry *d2)
9082 +{
9083 +       di_write_unlock(d1);
9084 +       if (d1->d_inode == d2->d_inode)
9085 +               au_rw_write_unlock(&au_di(d2)->di_rwsem);
9086 +       else
9087 +               di_write_unlock(d2);
9088 +}
9089 +
9090 +/* ---------------------------------------------------------------------- */
9091 +
9092 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex)
9093 +{
9094 +       struct dentry *d;
9095 +
9096 +       DiMustAnyLock(dentry);
9097 +
9098 +       if (au_dbstart(dentry) < 0 || bindex < au_dbstart(dentry))
9099 +               return NULL;
9100 +       AuDebugOn(bindex < 0);
9101 +       d = au_di(dentry)->di_hdentry[0 + bindex].hd_dentry;
9102 +       AuDebugOn(d && au_dcount(d) <= 0);
9103 +       return d;
9104 +}
9105 +
9106 +/*
9107 + * extended version of au_h_dptr().
9108 + * returns a hashed and positive (or linkable) h_dentry in bindex, NULL, or
9109 + * error.
9110 + */
9111 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex)
9112 +{
9113 +       struct dentry *h_dentry;
9114 +       struct inode *inode, *h_inode;
9115 +
9116 +       inode = dentry->d_inode;
9117 +       AuDebugOn(!inode);
9118 +
9119 +       h_dentry = NULL;
9120 +       if (au_dbstart(dentry) <= bindex
9121 +           && bindex <= au_dbend(dentry))
9122 +               h_dentry = au_h_dptr(dentry, bindex);
9123 +       if (h_dentry && !au_d_linkable(h_dentry)) {
9124 +               dget(h_dentry);
9125 +               goto out; /* success */
9126 +       }
9127 +
9128 +       AuDebugOn(bindex < au_ibstart(inode));
9129 +       AuDebugOn(au_ibend(inode) < bindex);
9130 +       h_inode = au_h_iptr(inode, bindex);
9131 +       h_dentry = d_find_alias(h_inode);
9132 +       if (h_dentry) {
9133 +               if (!IS_ERR(h_dentry)) {
9134 +                       if (!au_d_linkable(h_dentry))
9135 +                               goto out; /* success */
9136 +                       dput(h_dentry);
9137 +               } else
9138 +                       goto out;
9139 +       }
9140 +
9141 +       if (au_opt_test(au_mntflags(dentry->d_sb), PLINK)) {
9142 +               h_dentry = au_plink_lkup(inode, bindex);
9143 +               AuDebugOn(!h_dentry);
9144 +               if (!IS_ERR(h_dentry)) {
9145 +                       if (!au_d_hashed_positive(h_dentry))
9146 +                               goto out; /* success */
9147 +                       dput(h_dentry);
9148 +                       h_dentry = NULL;
9149 +               }
9150 +       }
9151 +
9152 +out:
9153 +       AuDbgDentry(h_dentry);
9154 +       return h_dentry;
9155 +}
9156 +
9157 +aufs_bindex_t au_dbtail(struct dentry *dentry)
9158 +{
9159 +       aufs_bindex_t bend, bwh;
9160 +
9161 +       bend = au_dbend(dentry);
9162 +       if (0 <= bend) {
9163 +               bwh = au_dbwh(dentry);
9164 +               if (!bwh)
9165 +                       return bwh;
9166 +               if (0 < bwh && bwh < bend)
9167 +                       return bwh - 1;
9168 +       }
9169 +       return bend;
9170 +}
9171 +
9172 +aufs_bindex_t au_dbtaildir(struct dentry *dentry)
9173 +{
9174 +       aufs_bindex_t bend, bopq;
9175 +
9176 +       bend = au_dbtail(dentry);
9177 +       if (0 <= bend) {
9178 +               bopq = au_dbdiropq(dentry);
9179 +               if (0 <= bopq && bopq < bend)
9180 +                       bend = bopq;
9181 +       }
9182 +       return bend;
9183 +}
9184 +
9185 +/* ---------------------------------------------------------------------- */
9186 +
9187 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
9188 +                  struct dentry *h_dentry)
9189 +{
9190 +       struct au_hdentry *hd = au_di(dentry)->di_hdentry + bindex;
9191 +       struct au_branch *br;
9192 +
9193 +       DiMustWriteLock(dentry);
9194 +
9195 +       au_hdput(hd);
9196 +       hd->hd_dentry = h_dentry;
9197 +       if (h_dentry) {
9198 +               br = au_sbr(dentry->d_sb, bindex);
9199 +               hd->hd_id = br->br_id;
9200 +       }
9201 +}
9202 +
9203 +int au_dbrange_test(struct dentry *dentry)
9204 +{
9205 +       int err;
9206 +       aufs_bindex_t bstart, bend;
9207 +
9208 +       err = 0;
9209 +       bstart = au_dbstart(dentry);
9210 +       bend = au_dbend(dentry);
9211 +       if (bstart >= 0)
9212 +               AuDebugOn(bend < 0 && bstart > bend);
9213 +       else {
9214 +               err = -EIO;
9215 +               AuDebugOn(bend >= 0);
9216 +       }
9217 +
9218 +       return err;
9219 +}
9220 +
9221 +int au_digen_test(struct dentry *dentry, unsigned int sigen)
9222 +{
9223 +       int err;
9224 +
9225 +       err = 0;
9226 +       if (unlikely(au_digen(dentry) != sigen
9227 +                    || au_iigen_test(dentry->d_inode, sigen)))
9228 +               err = -EIO;
9229 +
9230 +       return err;
9231 +}
9232 +
9233 +void au_update_digen(struct dentry *dentry)
9234 +{
9235 +       atomic_set(&au_di(dentry)->di_generation, au_sigen(dentry->d_sb));
9236 +       /* smp_mb(); */ /* atomic_set */
9237 +}
9238 +
9239 +void au_update_dbrange(struct dentry *dentry, int do_put_zero)
9240 +{
9241 +       struct au_dinfo *dinfo;
9242 +       struct dentry *h_d;
9243 +       struct au_hdentry *hdp;
9244 +
9245 +       DiMustWriteLock(dentry);
9246 +
9247 +       dinfo = au_di(dentry);
9248 +       if (!dinfo || dinfo->di_bstart < 0)
9249 +               return;
9250 +
9251 +       hdp = dinfo->di_hdentry;
9252 +       if (do_put_zero) {
9253 +               aufs_bindex_t bindex, bend;
9254 +
9255 +               bend = dinfo->di_bend;
9256 +               for (bindex = dinfo->di_bstart; bindex <= bend; bindex++) {
9257 +                       h_d = hdp[0 + bindex].hd_dentry;
9258 +                       if (h_d && !h_d->d_inode)
9259 +                               au_set_h_dptr(dentry, bindex, NULL);
9260 +               }
9261 +       }
9262 +
9263 +       dinfo->di_bstart = -1;
9264 +       while (++dinfo->di_bstart <= dinfo->di_bend)
9265 +               if (hdp[0 + dinfo->di_bstart].hd_dentry)
9266 +                       break;
9267 +       if (dinfo->di_bstart > dinfo->di_bend) {
9268 +               dinfo->di_bstart = -1;
9269 +               dinfo->di_bend = -1;
9270 +               return;
9271 +       }
9272 +
9273 +       dinfo->di_bend++;
9274 +       while (0 <= --dinfo->di_bend)
9275 +               if (hdp[0 + dinfo->di_bend].hd_dentry)
9276 +                       break;
9277 +       AuDebugOn(dinfo->di_bstart > dinfo->di_bend || dinfo->di_bend < 0);
9278 +}
9279 +
9280 +void au_update_dbstart(struct dentry *dentry)
9281 +{
9282 +       aufs_bindex_t bindex, bend;
9283 +       struct dentry *h_dentry;
9284 +
9285 +       bend = au_dbend(dentry);
9286 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++) {
9287 +               h_dentry = au_h_dptr(dentry, bindex);
9288 +               if (!h_dentry)
9289 +                       continue;
9290 +               if (h_dentry->d_inode) {
9291 +                       au_set_dbstart(dentry, bindex);
9292 +                       return;
9293 +               }
9294 +               au_set_h_dptr(dentry, bindex, NULL);
9295 +       }
9296 +}
9297 +
9298 +void au_update_dbend(struct dentry *dentry)
9299 +{
9300 +       aufs_bindex_t bindex, bstart;
9301 +       struct dentry *h_dentry;
9302 +
9303 +       bstart = au_dbstart(dentry);
9304 +       for (bindex = au_dbend(dentry); bindex >= bstart; bindex--) {
9305 +               h_dentry = au_h_dptr(dentry, bindex);
9306 +               if (!h_dentry)
9307 +                       continue;
9308 +               if (h_dentry->d_inode) {
9309 +                       au_set_dbend(dentry, bindex);
9310 +                       return;
9311 +               }
9312 +               au_set_h_dptr(dentry, bindex, NULL);
9313 +       }
9314 +}
9315 +
9316 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry)
9317 +{
9318 +       aufs_bindex_t bindex, bend;
9319 +
9320 +       bend = au_dbend(dentry);
9321 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++)
9322 +               if (au_h_dptr(dentry, bindex) == h_dentry)
9323 +                       return bindex;
9324 +       return -1;
9325 +}
9326 diff -urN /usr/share/empty/fs/aufs/dir.c linux/fs/aufs/dir.c
9327 --- /usr/share/empty/fs/aufs/dir.c      1970-01-01 01:00:00.000000000 +0100
9328 +++ linux/fs/aufs/dir.c 2015-03-27 21:56:35.463461668 +0100
9329 @@ -0,0 +1,643 @@
9330 +/*
9331 + * Copyright (C) 2005-2015 Junjiro R. Okajima
9332 + *
9333 + * This program, aufs is free software; you can redistribute it and/or modify
9334 + * it under the terms of the GNU General Public License as published by
9335 + * the Free Software Foundation; either version 2 of the License, or
9336 + * (at your option) any later version.
9337 + *
9338 + * This program is distributed in the hope that it will be useful,
9339 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9340 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9341 + * GNU General Public License for more details.
9342 + *
9343 + * You should have received a copy of the GNU General Public License
9344 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9345 + */
9346 +
9347 +/*
9348 + * directory operations
9349 + */
9350 +
9351 +#include <linux/fs_stack.h>
9352 +#include "aufs.h"
9353 +
9354 +void au_add_nlink(struct inode *dir, struct inode *h_dir)
9355 +{
9356 +       unsigned int nlink;
9357 +
9358 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9359 +
9360 +       nlink = dir->i_nlink;
9361 +       nlink += h_dir->i_nlink - 2;
9362 +       if (h_dir->i_nlink < 2)
9363 +               nlink += 2;
9364 +       smp_mb(); /* for i_nlink */
9365 +       /* 0 can happen in revaliding */
9366 +       set_nlink(dir, nlink);
9367 +}
9368 +
9369 +void au_sub_nlink(struct inode *dir, struct inode *h_dir)
9370 +{
9371 +       unsigned int nlink;
9372 +
9373 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
9374 +
9375 +       nlink = dir->i_nlink;
9376 +       nlink -= h_dir->i_nlink - 2;
9377 +       if (h_dir->i_nlink < 2)
9378 +               nlink -= 2;
9379 +       smp_mb(); /* for i_nlink */
9380 +       /* nlink == 0 means the branch-fs is broken */
9381 +       set_nlink(dir, nlink);
9382 +}
9383 +
9384 +loff_t au_dir_size(struct file *file, struct dentry *dentry)
9385 +{
9386 +       loff_t sz;
9387 +       aufs_bindex_t bindex, bend;
9388 +       struct file *h_file;
9389 +       struct dentry *h_dentry;
9390 +
9391 +       sz = 0;
9392 +       if (file) {
9393 +               AuDebugOn(!d_is_dir(file->f_path.dentry));
9394 +
9395 +               bend = au_fbend_dir(file);
9396 +               for (bindex = au_fbstart(file);
9397 +                    bindex <= bend && sz < KMALLOC_MAX_SIZE;
9398 +                    bindex++) {
9399 +                       h_file = au_hf_dir(file, bindex);
9400 +                       if (h_file && file_inode(h_file))
9401 +                               sz += vfsub_f_size_read(h_file);
9402 +               }
9403 +       } else {
9404 +               AuDebugOn(!dentry);
9405 +               AuDebugOn(!d_is_dir(dentry));
9406 +
9407 +               bend = au_dbtaildir(dentry);
9408 +               for (bindex = au_dbstart(dentry);
9409 +                    bindex <= bend && sz < KMALLOC_MAX_SIZE;
9410 +                    bindex++) {
9411 +                       h_dentry = au_h_dptr(dentry, bindex);
9412 +                       if (h_dentry && h_dentry->d_inode)
9413 +                               sz += i_size_read(h_dentry->d_inode);
9414 +               }
9415 +       }
9416 +       if (sz < KMALLOC_MAX_SIZE)
9417 +               sz = roundup_pow_of_two(sz);
9418 +       if (sz > KMALLOC_MAX_SIZE)
9419 +               sz = KMALLOC_MAX_SIZE;
9420 +       else if (sz < NAME_MAX) {
9421 +               BUILD_BUG_ON(AUFS_RDBLK_DEF < NAME_MAX);
9422 +               sz = AUFS_RDBLK_DEF;
9423 +       }
9424 +       return sz;
9425 +}
9426 +
9427 +/* ---------------------------------------------------------------------- */
9428 +
9429 +static int reopen_dir(struct file *file)
9430 +{
9431 +       int err;
9432 +       unsigned int flags;
9433 +       aufs_bindex_t bindex, btail, bstart;
9434 +       struct dentry *dentry, *h_dentry;
9435 +       struct file *h_file;
9436 +
9437 +       /* open all lower dirs */
9438 +       dentry = file->f_path.dentry;
9439 +       bstart = au_dbstart(dentry);
9440 +       for (bindex = au_fbstart(file); bindex < bstart; bindex++)
9441 +               au_set_h_fptr(file, bindex, NULL);
9442 +       au_set_fbstart(file, bstart);
9443 +
9444 +       btail = au_dbtaildir(dentry);
9445 +       for (bindex = au_fbend_dir(file); btail < bindex; bindex--)
9446 +               au_set_h_fptr(file, bindex, NULL);
9447 +       au_set_fbend_dir(file, btail);
9448 +
9449 +       flags = vfsub_file_flags(file);
9450 +       for (bindex = bstart; bindex <= btail; bindex++) {
9451 +               h_dentry = au_h_dptr(dentry, bindex);
9452 +               if (!h_dentry)
9453 +                       continue;
9454 +               h_file = au_hf_dir(file, bindex);
9455 +               if (h_file)
9456 +                       continue;
9457 +
9458 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9459 +               err = PTR_ERR(h_file);
9460 +               if (IS_ERR(h_file))
9461 +                       goto out; /* close all? */
9462 +               au_set_h_fptr(file, bindex, h_file);
9463 +       }
9464 +       au_update_figen(file);
9465 +       /* todo: necessary? */
9466 +       /* file->f_ra = h_file->f_ra; */
9467 +       err = 0;
9468 +
9469 +out:
9470 +       return err;
9471 +}
9472 +
9473 +static int do_open_dir(struct file *file, int flags)
9474 +{
9475 +       int err;
9476 +       aufs_bindex_t bindex, btail;
9477 +       struct dentry *dentry, *h_dentry;
9478 +       struct file *h_file;
9479 +
9480 +       FiMustWriteLock(file);
9481 +
9482 +       err = 0;
9483 +       dentry = file->f_path.dentry;
9484 +       file->f_version = dentry->d_inode->i_version;
9485 +       bindex = au_dbstart(dentry);
9486 +       au_set_fbstart(file, bindex);
9487 +       btail = au_dbtaildir(dentry);
9488 +       au_set_fbend_dir(file, btail);
9489 +       for (; !err && bindex <= btail; bindex++) {
9490 +               h_dentry = au_h_dptr(dentry, bindex);
9491 +               if (!h_dentry)
9492 +                       continue;
9493 +
9494 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
9495 +               if (IS_ERR(h_file)) {
9496 +                       err = PTR_ERR(h_file);
9497 +                       break;
9498 +               }
9499 +               au_set_h_fptr(file, bindex, h_file);
9500 +       }
9501 +       au_update_figen(file);
9502 +       /* todo: necessary? */
9503 +       /* file->f_ra = h_file->f_ra; */
9504 +       if (!err)
9505 +               return 0; /* success */
9506 +
9507 +       /* close all */
9508 +       for (bindex = au_fbstart(file); bindex <= btail; bindex++)
9509 +               au_set_h_fptr(file, bindex, NULL);
9510 +       au_set_fbstart(file, -1);
9511 +       au_set_fbend_dir(file, -1);
9512 +
9513 +       return err;
9514 +}
9515 +
9516 +static int aufs_open_dir(struct inode *inode __maybe_unused,
9517 +                        struct file *file)
9518 +{
9519 +       int err;
9520 +       struct super_block *sb;
9521 +       struct au_fidir *fidir;
9522 +
9523 +       err = -ENOMEM;
9524 +       sb = file->f_path.dentry->d_sb;
9525 +       si_read_lock(sb, AuLock_FLUSH);
9526 +       fidir = au_fidir_alloc(sb);
9527 +       if (fidir) {
9528 +               err = au_do_open(file, do_open_dir, fidir);
9529 +               if (unlikely(err))
9530 +                       kfree(fidir);
9531 +       }
9532 +       si_read_unlock(sb);
9533 +       return err;
9534 +}
9535 +
9536 +static int aufs_release_dir(struct inode *inode __maybe_unused,
9537 +                           struct file *file)
9538 +{
9539 +       struct au_vdir *vdir_cache;
9540 +       struct au_finfo *finfo;
9541 +       struct au_fidir *fidir;
9542 +       aufs_bindex_t bindex, bend;
9543 +
9544 +       finfo = au_fi(file);
9545 +       fidir = finfo->fi_hdir;
9546 +       if (fidir) {
9547 +               au_sphl_del(&finfo->fi_hlist,
9548 +                           &au_sbi(file->f_path.dentry->d_sb)->si_files);
9549 +               vdir_cache = fidir->fd_vdir_cache; /* lock-free */
9550 +               if (vdir_cache)
9551 +                       au_vdir_free(vdir_cache);
9552 +
9553 +               bindex = finfo->fi_btop;
9554 +               if (bindex >= 0) {
9555 +                       /*
9556 +                        * calls fput() instead of filp_close(),
9557 +                        * since no dnotify or lock for the lower file.
9558 +                        */
9559 +                       bend = fidir->fd_bbot;
9560 +                       for (; bindex <= bend; bindex++)
9561 +                               au_set_h_fptr(file, bindex, NULL);
9562 +               }
9563 +               kfree(fidir);
9564 +               finfo->fi_hdir = NULL;
9565 +       }
9566 +       au_finfo_fin(file);
9567 +       return 0;
9568 +}
9569 +
9570 +/* ---------------------------------------------------------------------- */
9571 +
9572 +static int au_do_flush_dir(struct file *file, fl_owner_t id)
9573 +{
9574 +       int err;
9575 +       aufs_bindex_t bindex, bend;
9576 +       struct file *h_file;
9577 +
9578 +       err = 0;
9579 +       bend = au_fbend_dir(file);
9580 +       for (bindex = au_fbstart(file); !err && bindex <= bend; bindex++) {
9581 +               h_file = au_hf_dir(file, bindex);
9582 +               if (h_file)
9583 +                       err = vfsub_flush(h_file, id);
9584 +       }
9585 +       return err;
9586 +}
9587 +
9588 +static int aufs_flush_dir(struct file *file, fl_owner_t id)
9589 +{
9590 +       return au_do_flush(file, id, au_do_flush_dir);
9591 +}
9592 +
9593 +/* ---------------------------------------------------------------------- */
9594 +
9595 +static int au_do_fsync_dir_no_file(struct dentry *dentry, int datasync)
9596 +{
9597 +       int err;
9598 +       aufs_bindex_t bend, bindex;
9599 +       struct inode *inode;
9600 +       struct super_block *sb;
9601 +
9602 +       err = 0;
9603 +       sb = dentry->d_sb;
9604 +       inode = dentry->d_inode;
9605 +       IMustLock(inode);
9606 +       bend = au_dbend(dentry);
9607 +       for (bindex = au_dbstart(dentry); !err && bindex <= bend; bindex++) {
9608 +               struct path h_path;
9609 +
9610 +               if (au_test_ro(sb, bindex, inode))
9611 +                       continue;
9612 +               h_path.dentry = au_h_dptr(dentry, bindex);
9613 +               if (!h_path.dentry)
9614 +                       continue;
9615 +
9616 +               h_path.mnt = au_sbr_mnt(sb, bindex);
9617 +               err = vfsub_fsync(NULL, &h_path, datasync);
9618 +       }
9619 +
9620 +       return err;
9621 +}
9622 +
9623 +static int au_do_fsync_dir(struct file *file, int datasync)
9624 +{
9625 +       int err;
9626 +       aufs_bindex_t bend, bindex;
9627 +       struct file *h_file;
9628 +       struct super_block *sb;
9629 +       struct inode *inode;
9630 +
9631 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
9632 +       if (unlikely(err))
9633 +               goto out;
9634 +
9635 +       sb = file->f_path.dentry->d_sb;
9636 +       inode = file_inode(file);
9637 +       bend = au_fbend_dir(file);
9638 +       for (bindex = au_fbstart(file); !err && bindex <= bend; bindex++) {
9639 +               h_file = au_hf_dir(file, bindex);
9640 +               if (!h_file || au_test_ro(sb, bindex, inode))
9641 +                       continue;
9642 +
9643 +               err = vfsub_fsync(h_file, &h_file->f_path, datasync);
9644 +       }
9645 +
9646 +out:
9647 +       return err;
9648 +}
9649 +
9650 +/*
9651 + * @file may be NULL
9652 + */
9653 +static int aufs_fsync_dir(struct file *file, loff_t start, loff_t end,
9654 +                         int datasync)
9655 +{
9656 +       int err;
9657 +       struct dentry *dentry;
9658 +       struct super_block *sb;
9659 +       struct mutex *mtx;
9660 +
9661 +       err = 0;
9662 +       dentry = file->f_path.dentry;
9663 +       mtx = &dentry->d_inode->i_mutex;
9664 +       mutex_lock(mtx);
9665 +       sb = dentry->d_sb;
9666 +       si_noflush_read_lock(sb);
9667 +       if (file)
9668 +               err = au_do_fsync_dir(file, datasync);
9669 +       else {
9670 +               di_write_lock_child(dentry);
9671 +               err = au_do_fsync_dir_no_file(dentry, datasync);
9672 +       }
9673 +       au_cpup_attr_timesizes(dentry->d_inode);
9674 +       di_write_unlock(dentry);
9675 +       if (file)
9676 +               fi_write_unlock(file);
9677 +
9678 +       si_read_unlock(sb);
9679 +       mutex_unlock(mtx);
9680 +       return err;
9681 +}
9682 +
9683 +/* ---------------------------------------------------------------------- */
9684 +
9685 +static int aufs_iterate(struct file *file, struct dir_context *ctx)
9686 +{
9687 +       int err;
9688 +       struct dentry *dentry;
9689 +       struct inode *inode, *h_inode;
9690 +       struct super_block *sb;
9691 +
9692 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
9693 +
9694 +       dentry = file->f_path.dentry;
9695 +       inode = dentry->d_inode;
9696 +       IMustLock(inode);
9697 +
9698 +       sb = dentry->d_sb;
9699 +       si_read_lock(sb, AuLock_FLUSH);
9700 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
9701 +       if (unlikely(err))
9702 +               goto out;
9703 +       err = au_alive_dir(dentry);
9704 +       if (!err)
9705 +               err = au_vdir_init(file);
9706 +       di_downgrade_lock(dentry, AuLock_IR);
9707 +       if (unlikely(err))
9708 +               goto out_unlock;
9709 +
9710 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
9711 +       if (!au_test_nfsd()) {
9712 +               err = au_vdir_fill_de(file, ctx);
9713 +               fsstack_copy_attr_atime(inode, h_inode);
9714 +       } else {
9715 +               /*
9716 +                * nfsd filldir may call lookup_one_len(), vfs_getattr(),
9717 +                * encode_fh() and others.
9718 +                */
9719 +               atomic_inc(&h_inode->i_count);
9720 +               di_read_unlock(dentry, AuLock_IR);
9721 +               si_read_unlock(sb);
9722 +               err = au_vdir_fill_de(file, ctx);
9723 +               fsstack_copy_attr_atime(inode, h_inode);
9724 +               fi_write_unlock(file);
9725 +               iput(h_inode);
9726 +
9727 +               AuTraceErr(err);
9728 +               return err;
9729 +       }
9730 +
9731 +out_unlock:
9732 +       di_read_unlock(dentry, AuLock_IR);
9733 +       fi_write_unlock(file);
9734 +out:
9735 +       si_read_unlock(sb);
9736 +       return err;
9737 +}
9738 +
9739 +/* ---------------------------------------------------------------------- */
9740 +
9741 +#define AuTestEmpty_WHONLY     1
9742 +#define AuTestEmpty_CALLED     (1 << 1)
9743 +#define AuTestEmpty_SHWH       (1 << 2)
9744 +#define au_ftest_testempty(flags, name)        ((flags) & AuTestEmpty_##name)
9745 +#define au_fset_testempty(flags, name) \
9746 +       do { (flags) |= AuTestEmpty_##name; } while (0)
9747 +#define au_fclr_testempty(flags, name) \
9748 +       do { (flags) &= ~AuTestEmpty_##name; } while (0)
9749 +
9750 +#ifndef CONFIG_AUFS_SHWH
9751 +#undef AuTestEmpty_SHWH
9752 +#define AuTestEmpty_SHWH       0
9753 +#endif
9754 +
9755 +struct test_empty_arg {
9756 +       struct dir_context ctx;
9757 +       struct au_nhash *whlist;
9758 +       unsigned int flags;
9759 +       int err;
9760 +       aufs_bindex_t bindex;
9761 +};
9762 +
9763 +static int test_empty_cb(struct dir_context *ctx, const char *__name,
9764 +                        int namelen, loff_t offset __maybe_unused, u64 ino,
9765 +                        unsigned int d_type)
9766 +{
9767 +       struct test_empty_arg *arg = container_of(ctx, struct test_empty_arg,
9768 +                                                 ctx);
9769 +       char *name = (void *)__name;
9770 +
9771 +       arg->err = 0;
9772 +       au_fset_testempty(arg->flags, CALLED);
9773 +       /* smp_mb(); */
9774 +       if (name[0] == '.'
9775 +           && (namelen == 1 || (name[1] == '.' && namelen == 2)))
9776 +               goto out; /* success */
9777 +
9778 +       if (namelen <= AUFS_WH_PFX_LEN
9779 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
9780 +               if (au_ftest_testempty(arg->flags, WHONLY)
9781 +                   && !au_nhash_test_known_wh(arg->whlist, name, namelen))
9782 +                       arg->err = -ENOTEMPTY;
9783 +               goto out;
9784 +       }
9785 +
9786 +       name += AUFS_WH_PFX_LEN;
9787 +       namelen -= AUFS_WH_PFX_LEN;
9788 +       if (!au_nhash_test_known_wh(arg->whlist, name, namelen))
9789 +               arg->err = au_nhash_append_wh
9790 +                       (arg->whlist, name, namelen, ino, d_type, arg->bindex,
9791 +                        au_ftest_testempty(arg->flags, SHWH));
9792 +
9793 +out:
9794 +       /* smp_mb(); */
9795 +       AuTraceErr(arg->err);
9796 +       return arg->err;
9797 +}
9798 +
9799 +static int do_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
9800 +{
9801 +       int err;
9802 +       struct file *h_file;
9803 +
9804 +       h_file = au_h_open(dentry, arg->bindex,
9805 +                          O_RDONLY | O_NONBLOCK | O_DIRECTORY | O_LARGEFILE,
9806 +                          /*file*/NULL, /*force_wr*/0);
9807 +       err = PTR_ERR(h_file);
9808 +       if (IS_ERR(h_file))
9809 +               goto out;
9810 +
9811 +       err = 0;
9812 +       if (!au_opt_test(au_mntflags(dentry->d_sb), UDBA_NONE)
9813 +           && !file_inode(h_file)->i_nlink)
9814 +               goto out_put;
9815 +
9816 +       do {
9817 +               arg->err = 0;
9818 +               au_fclr_testempty(arg->flags, CALLED);
9819 +               /* smp_mb(); */
9820 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
9821 +               if (err >= 0)
9822 +                       err = arg->err;
9823 +       } while (!err && au_ftest_testempty(arg->flags, CALLED));
9824 +
9825 +out_put:
9826 +       fput(h_file);
9827 +       au_sbr_put(dentry->d_sb, arg->bindex);
9828 +out:
9829 +       return err;
9830 +}
9831 +
9832 +struct do_test_empty_args {
9833 +       int *errp;
9834 +       struct dentry *dentry;
9835 +       struct test_empty_arg *arg;
9836 +};
9837 +
9838 +static void call_do_test_empty(void *args)
9839 +{
9840 +       struct do_test_empty_args *a = args;
9841 +       *a->errp = do_test_empty(a->dentry, a->arg);
9842 +}
9843 +
9844 +static int sio_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
9845 +{
9846 +       int err, wkq_err;
9847 +       struct dentry *h_dentry;
9848 +       struct inode *h_inode;
9849 +
9850 +       h_dentry = au_h_dptr(dentry, arg->bindex);
9851 +       h_inode = h_dentry->d_inode;
9852 +       /* todo: i_mode changes anytime? */
9853 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
9854 +       err = au_test_h_perm_sio(h_inode, MAY_EXEC | MAY_READ);
9855 +       mutex_unlock(&h_inode->i_mutex);
9856 +       if (!err)
9857 +               err = do_test_empty(dentry, arg);
9858 +       else {
9859 +               struct do_test_empty_args args = {
9860 +                       .errp   = &err,
9861 +                       .dentry = dentry,
9862 +                       .arg    = arg
9863 +               };
9864 +               unsigned int flags = arg->flags;
9865 +
9866 +               wkq_err = au_wkq_wait(call_do_test_empty, &args);
9867 +               if (unlikely(wkq_err))
9868 +                       err = wkq_err;
9869 +               arg->flags = flags;
9870 +       }
9871 +
9872 +       return err;
9873 +}
9874 +
9875 +int au_test_empty_lower(struct dentry *dentry)
9876 +{
9877 +       int err;
9878 +       unsigned int rdhash;
9879 +       aufs_bindex_t bindex, bstart, btail;
9880 +       struct au_nhash whlist;
9881 +       struct test_empty_arg arg = {
9882 +               .ctx = {
9883 +                       .actor = test_empty_cb
9884 +               }
9885 +       };
9886 +       int (*test_empty)(struct dentry *dentry, struct test_empty_arg *arg);
9887 +
9888 +       SiMustAnyLock(dentry->d_sb);
9889 +
9890 +       rdhash = au_sbi(dentry->d_sb)->si_rdhash;
9891 +       if (!rdhash)
9892 +               rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, dentry));
9893 +       err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
9894 +       if (unlikely(err))
9895 +               goto out;
9896 +
9897 +       arg.flags = 0;
9898 +       arg.whlist = &whlist;
9899 +       bstart = au_dbstart(dentry);
9900 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
9901 +               au_fset_testempty(arg.flags, SHWH);
9902 +       test_empty = do_test_empty;
9903 +       if (au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1))
9904 +               test_empty = sio_test_empty;
9905 +       arg.bindex = bstart;
9906 +       err = test_empty(dentry, &arg);
9907 +       if (unlikely(err))
9908 +               goto out_whlist;
9909 +
9910 +       au_fset_testempty(arg.flags, WHONLY);
9911 +       btail = au_dbtaildir(dentry);
9912 +       for (bindex = bstart + 1; !err && bindex <= btail; bindex++) {
9913 +               struct dentry *h_dentry;
9914 +
9915 +               h_dentry = au_h_dptr(dentry, bindex);
9916 +               if (h_dentry && h_dentry->d_inode) {
9917 +                       arg.bindex = bindex;
9918 +                       err = test_empty(dentry, &arg);
9919 +               }
9920 +       }
9921 +
9922 +out_whlist:
9923 +       au_nhash_wh_free(&whlist);
9924 +out:
9925 +       return err;
9926 +}
9927 +
9928 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist)
9929 +{
9930 +       int err;
9931 +       struct test_empty_arg arg = {
9932 +               .ctx = {
9933 +                       .actor = test_empty_cb
9934 +               }
9935 +       };
9936 +       aufs_bindex_t bindex, btail;
9937 +
9938 +       err = 0;
9939 +       arg.whlist = whlist;
9940 +       arg.flags = AuTestEmpty_WHONLY;
9941 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
9942 +               au_fset_testempty(arg.flags, SHWH);
9943 +       btail = au_dbtaildir(dentry);
9944 +       for (bindex = au_dbstart(dentry); !err && bindex <= btail; bindex++) {
9945 +               struct dentry *h_dentry;
9946 +
9947 +               h_dentry = au_h_dptr(dentry, bindex);
9948 +               if (h_dentry && h_dentry->d_inode) {
9949 +                       arg.bindex = bindex;
9950 +                       err = sio_test_empty(dentry, &arg);
9951 +               }
9952 +       }
9953 +
9954 +       return err;
9955 +}
9956 +
9957 +/* ---------------------------------------------------------------------- */
9958 +
9959 +const struct file_operations aufs_dir_fop = {
9960 +       .owner          = THIS_MODULE,
9961 +       .llseek         = default_llseek,
9962 +       .read           = generic_read_dir,
9963 +       .iterate        = aufs_iterate,
9964 +       .unlocked_ioctl = aufs_ioctl_dir,
9965 +#ifdef CONFIG_COMPAT
9966 +       .compat_ioctl   = aufs_compat_ioctl_dir,
9967 +#endif
9968 +       .open           = aufs_open_dir,
9969 +       .release        = aufs_release_dir,
9970 +       .flush          = aufs_flush_dir,
9971 +       .fsync          = aufs_fsync_dir
9972 +};
9973 diff -urN /usr/share/empty/fs/aufs/dir.h linux/fs/aufs/dir.h
9974 --- /usr/share/empty/fs/aufs/dir.h      1970-01-01 01:00:00.000000000 +0100
9975 +++ linux/fs/aufs/dir.h 2015-03-27 21:56:35.463461668 +0100
9976 @@ -0,0 +1,130 @@
9977 +/*
9978 + * Copyright (C) 2005-2015 Junjiro R. Okajima
9979 + *
9980 + * This program, aufs is free software; you can redistribute it and/or modify
9981 + * it under the terms of the GNU General Public License as published by
9982 + * the Free Software Foundation; either version 2 of the License, or
9983 + * (at your option) any later version.
9984 + *
9985 + * This program is distributed in the hope that it will be useful,
9986 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9987 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9988 + * GNU General Public License for more details.
9989 + *
9990 + * You should have received a copy of the GNU General Public License
9991 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9992 + */
9993 +
9994 +/*
9995 + * directory operations
9996 + */
9997 +
9998 +#ifndef __AUFS_DIR_H__
9999 +#define __AUFS_DIR_H__
10000 +
10001 +#ifdef __KERNEL__
10002 +
10003 +#include <linux/fs.h>
10004 +
10005 +/* ---------------------------------------------------------------------- */
10006 +
10007 +/* need to be faster and smaller */
10008 +
10009 +struct au_nhash {
10010 +       unsigned int            nh_num;
10011 +       struct hlist_head       *nh_head;
10012 +};
10013 +
10014 +struct au_vdir_destr {
10015 +       unsigned char   len;
10016 +       unsigned char   name[0];
10017 +} __packed;
10018 +
10019 +struct au_vdir_dehstr {
10020 +       struct hlist_node       hash;
10021 +       struct au_vdir_destr    *str;
10022 +} ____cacheline_aligned_in_smp;
10023 +
10024 +struct au_vdir_de {
10025 +       ino_t                   de_ino;
10026 +       unsigned char           de_type;
10027 +       /* caution: packed */
10028 +       struct au_vdir_destr    de_str;
10029 +} __packed;
10030 +
10031 +struct au_vdir_wh {
10032 +       struct hlist_node       wh_hash;
10033 +#ifdef CONFIG_AUFS_SHWH
10034 +       ino_t                   wh_ino;
10035 +       aufs_bindex_t           wh_bindex;
10036 +       unsigned char           wh_type;
10037 +#else
10038 +       aufs_bindex_t           wh_bindex;
10039 +#endif
10040 +       /* caution: packed */
10041 +       struct au_vdir_destr    wh_str;
10042 +} __packed;
10043 +
10044 +union au_vdir_deblk_p {
10045 +       unsigned char           *deblk;
10046 +       struct au_vdir_de       *de;
10047 +};
10048 +
10049 +struct au_vdir {
10050 +       unsigned char   **vd_deblk;
10051 +       unsigned long   vd_nblk;
10052 +       struct {
10053 +               unsigned long           ul;
10054 +               union au_vdir_deblk_p   p;
10055 +       } vd_last;
10056 +
10057 +       unsigned long   vd_version;
10058 +       unsigned int    vd_deblk_sz;
10059 +       unsigned long   vd_jiffy;
10060 +} ____cacheline_aligned_in_smp;
10061 +
10062 +/* ---------------------------------------------------------------------- */
10063 +
10064 +/* dir.c */
10065 +extern const struct file_operations aufs_dir_fop;
10066 +void au_add_nlink(struct inode *dir, struct inode *h_dir);
10067 +void au_sub_nlink(struct inode *dir, struct inode *h_dir);
10068 +loff_t au_dir_size(struct file *file, struct dentry *dentry);
10069 +int au_test_empty_lower(struct dentry *dentry);
10070 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist);
10071 +
10072 +/* vdir.c */
10073 +unsigned int au_rdhash_est(loff_t sz);
10074 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp);
10075 +void au_nhash_wh_free(struct au_nhash *whlist);
10076 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
10077 +                           int limit);
10078 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen);
10079 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
10080 +                      unsigned int d_type, aufs_bindex_t bindex,
10081 +                      unsigned char shwh);
10082 +void au_vdir_free(struct au_vdir *vdir);
10083 +int au_vdir_init(struct file *file);
10084 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx);
10085 +
10086 +/* ioctl.c */
10087 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg);
10088 +
10089 +#ifdef CONFIG_AUFS_RDU
10090 +/* rdu.c */
10091 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
10092 +#ifdef CONFIG_COMPAT
10093 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd,
10094 +                        unsigned long arg);
10095 +#endif
10096 +#else
10097 +AuStub(long, au_rdu_ioctl, return -EINVAL, struct file *file,
10098 +       unsigned int cmd, unsigned long arg)
10099 +#ifdef CONFIG_COMPAT
10100 +AuStub(long, au_rdu_compat_ioctl, return -EINVAL, struct file *file,
10101 +       unsigned int cmd, unsigned long arg)
10102 +#endif
10103 +#endif
10104 +
10105 +#endif /* __KERNEL__ */
10106 +#endif /* __AUFS_DIR_H__ */
10107 diff -urN /usr/share/empty/fs/aufs/dynop.c linux/fs/aufs/dynop.c
10108 --- /usr/share/empty/fs/aufs/dynop.c    1970-01-01 01:00:00.000000000 +0100
10109 +++ linux/fs/aufs/dynop.c       2015-03-27 21:56:35.463461668 +0100
10110 @@ -0,0 +1,379 @@
10111 +/*
10112 + * Copyright (C) 2010-2015 Junjiro R. Okajima
10113 + *
10114 + * This program, aufs is free software; you can redistribute it and/or modify
10115 + * it under the terms of the GNU General Public License as published by
10116 + * the Free Software Foundation; either version 2 of the License, or
10117 + * (at your option) any later version.
10118 + *
10119 + * This program is distributed in the hope that it will be useful,
10120 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10121 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10122 + * GNU General Public License for more details.
10123 + *
10124 + * You should have received a copy of the GNU General Public License
10125 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10126 + */
10127 +
10128 +/*
10129 + * dynamically customizable operations for regular files
10130 + */
10131 +
10132 +#include "aufs.h"
10133 +
10134 +#define DyPrSym(key)   AuDbgSym(key->dk_op.dy_hop)
10135 +
10136 +/*
10137 + * How large will these lists be?
10138 + * Usually just a few elements, 20-30 at most for each, I guess.
10139 + */
10140 +static struct au_splhead dynop[AuDyLast];
10141 +
10142 +static struct au_dykey *dy_gfind_get(struct au_splhead *spl, const void *h_op)
10143 +{
10144 +       struct au_dykey *key, *tmp;
10145 +       struct list_head *head;
10146 +
10147 +       key = NULL;
10148 +       head = &spl->head;
10149 +       rcu_read_lock();
10150 +       list_for_each_entry_rcu(tmp, head, dk_list)
10151 +               if (tmp->dk_op.dy_hop == h_op) {
10152 +                       key = tmp;
10153 +                       kref_get(&key->dk_kref);
10154 +                       break;
10155 +               }
10156 +       rcu_read_unlock();
10157 +
10158 +       return key;
10159 +}
10160 +
10161 +static struct au_dykey *dy_bradd(struct au_branch *br, struct au_dykey *key)
10162 +{
10163 +       struct au_dykey **k, *found;
10164 +       const void *h_op = key->dk_op.dy_hop;
10165 +       int i;
10166 +
10167 +       found = NULL;
10168 +       k = br->br_dykey;
10169 +       for (i = 0; i < AuBrDynOp; i++)
10170 +               if (k[i]) {
10171 +                       if (k[i]->dk_op.dy_hop == h_op) {
10172 +                               found = k[i];
10173 +                               break;
10174 +                       }
10175 +               } else
10176 +                       break;
10177 +       if (!found) {
10178 +               spin_lock(&br->br_dykey_lock);
10179 +               for (; i < AuBrDynOp; i++)
10180 +                       if (k[i]) {
10181 +                               if (k[i]->dk_op.dy_hop == h_op) {
10182 +                                       found = k[i];
10183 +                                       break;
10184 +                               }
10185 +                       } else {
10186 +                               k[i] = key;
10187 +                               break;
10188 +                       }
10189 +               spin_unlock(&br->br_dykey_lock);
10190 +               BUG_ON(i == AuBrDynOp); /* expand the array */
10191 +       }
10192 +
10193 +       return found;
10194 +}
10195 +
10196 +/* kref_get() if @key is already added */
10197 +static struct au_dykey *dy_gadd(struct au_splhead *spl, struct au_dykey *key)
10198 +{
10199 +       struct au_dykey *tmp, *found;
10200 +       struct list_head *head;
10201 +       const void *h_op = key->dk_op.dy_hop;
10202 +
10203 +       found = NULL;
10204 +       head = &spl->head;
10205 +       spin_lock(&spl->spin);
10206 +       list_for_each_entry(tmp, head, dk_list)
10207 +               if (tmp->dk_op.dy_hop == h_op) {
10208 +                       kref_get(&tmp->dk_kref);
10209 +                       found = tmp;
10210 +                       break;
10211 +               }
10212 +       if (!found)
10213 +               list_add_rcu(&key->dk_list, head);
10214 +       spin_unlock(&spl->spin);
10215 +
10216 +       if (!found)
10217 +               DyPrSym(key);
10218 +       return found;
10219 +}
10220 +
10221 +static void dy_free_rcu(struct rcu_head *rcu)
10222 +{
10223 +       struct au_dykey *key;
10224 +
10225 +       key = container_of(rcu, struct au_dykey, dk_rcu);
10226 +       DyPrSym(key);
10227 +       kfree(key);
10228 +}
10229 +
10230 +static void dy_free(struct kref *kref)
10231 +{
10232 +       struct au_dykey *key;
10233 +       struct au_splhead *spl;
10234 +
10235 +       key = container_of(kref, struct au_dykey, dk_kref);
10236 +       spl = dynop + key->dk_op.dy_type;
10237 +       au_spl_del_rcu(&key->dk_list, spl);
10238 +       call_rcu(&key->dk_rcu, dy_free_rcu);
10239 +}
10240 +
10241 +void au_dy_put(struct au_dykey *key)
10242 +{
10243 +       kref_put(&key->dk_kref, dy_free);
10244 +}
10245 +
10246 +/* ---------------------------------------------------------------------- */
10247 +
10248 +#define DyDbgSize(cnt, op)     AuDebugOn(cnt != sizeof(op)/sizeof(void *))
10249 +
10250 +#ifdef CONFIG_AUFS_DEBUG
10251 +#define DyDbgDeclare(cnt)      unsigned int cnt = 0
10252 +#define DyDbgInc(cnt)          do { cnt++; } while (0)
10253 +#else
10254 +#define DyDbgDeclare(cnt)      do {} while (0)
10255 +#define DyDbgInc(cnt)          do {} while (0)
10256 +#endif
10257 +
10258 +#define DySet(func, dst, src, h_op, h_sb) do {                         \
10259 +       DyDbgInc(cnt);                                                  \
10260 +       if (h_op->func) {                                               \
10261 +               if (src.func)                                           \
10262 +                       dst.func = src.func;                            \
10263 +               else                                                    \
10264 +                       AuDbg("%s %s\n", au_sbtype(h_sb), #func);       \
10265 +       }                                                               \
10266 +} while (0)
10267 +
10268 +#define DySetForce(func, dst, src) do {                \
10269 +       AuDebugOn(!src.func);                   \
10270 +       DyDbgInc(cnt);                          \
10271 +       dst.func = src.func;                    \
10272 +} while (0)
10273 +
10274 +#define DySetAop(func) \
10275 +       DySet(func, dyaop->da_op, aufs_aop, h_aop, h_sb)
10276 +#define DySetAopForce(func) \
10277 +       DySetForce(func, dyaop->da_op, aufs_aop)
10278 +
10279 +static void dy_aop(struct au_dykey *key, const void *h_op,
10280 +                  struct super_block *h_sb __maybe_unused)
10281 +{
10282 +       struct au_dyaop *dyaop = (void *)key;
10283 +       const struct address_space_operations *h_aop = h_op;
10284 +       DyDbgDeclare(cnt);
10285 +
10286 +       AuDbg("%s\n", au_sbtype(h_sb));
10287 +
10288 +       DySetAop(writepage);
10289 +       DySetAopForce(readpage);        /* force */
10290 +       DySetAop(writepages);
10291 +       DySetAop(set_page_dirty);
10292 +       DySetAop(readpages);
10293 +       DySetAop(write_begin);
10294 +       DySetAop(write_end);
10295 +       DySetAop(bmap);
10296 +       DySetAop(invalidatepage);
10297 +       DySetAop(releasepage);
10298 +       DySetAop(freepage);
10299 +       /* these two will be changed according to an aufs mount option */
10300 +       DySetAop(direct_IO);
10301 +       DySetAop(get_xip_mem);
10302 +       DySetAop(migratepage);
10303 +       DySetAop(launder_page);
10304 +       DySetAop(is_partially_uptodate);
10305 +       DySetAop(is_dirty_writeback);
10306 +       DySetAop(error_remove_page);
10307 +       DySetAop(swap_activate);
10308 +       DySetAop(swap_deactivate);
10309 +
10310 +       DyDbgSize(cnt, *h_aop);
10311 +       dyaop->da_get_xip_mem = h_aop->get_xip_mem;
10312 +}
10313 +
10314 +/* ---------------------------------------------------------------------- */
10315 +
10316 +static void dy_bug(struct kref *kref)
10317 +{
10318 +       BUG();
10319 +}
10320 +
10321 +static struct au_dykey *dy_get(struct au_dynop *op, struct au_branch *br)
10322 +{
10323 +       struct au_dykey *key, *old;
10324 +       struct au_splhead *spl;
10325 +       struct op {
10326 +               unsigned int sz;
10327 +               void (*set)(struct au_dykey *key, const void *h_op,
10328 +                           struct super_block *h_sb __maybe_unused);
10329 +       };
10330 +       static const struct op a[] = {
10331 +               [AuDy_AOP] = {
10332 +                       .sz     = sizeof(struct au_dyaop),
10333 +                       .set    = dy_aop
10334 +               }
10335 +       };
10336 +       const struct op *p;
10337 +
10338 +       spl = dynop + op->dy_type;
10339 +       key = dy_gfind_get(spl, op->dy_hop);
10340 +       if (key)
10341 +               goto out_add; /* success */
10342 +
10343 +       p = a + op->dy_type;
10344 +       key = kzalloc(p->sz, GFP_NOFS);
10345 +       if (unlikely(!key)) {
10346 +               key = ERR_PTR(-ENOMEM);
10347 +               goto out;
10348 +       }
10349 +
10350 +       key->dk_op.dy_hop = op->dy_hop;
10351 +       kref_init(&key->dk_kref);
10352 +       p->set(key, op->dy_hop, au_br_sb(br));
10353 +       old = dy_gadd(spl, key);
10354 +       if (old) {
10355 +               kfree(key);
10356 +               key = old;
10357 +       }
10358 +
10359 +out_add:
10360 +       old = dy_bradd(br, key);
10361 +       if (old)
10362 +               /* its ref-count should never be zero here */
10363 +               kref_put(&key->dk_kref, dy_bug);
10364 +out:
10365 +       return key;
10366 +}
10367 +
10368 +/* ---------------------------------------------------------------------- */
10369 +/*
10370 + * Aufs prohibits O_DIRECT by defaut even if the branch supports it.
10371 + * This behaviour is necessary to return an error from open(O_DIRECT) instead
10372 + * of the succeeding I/O. The dio mount option enables O_DIRECT and makes
10373 + * open(O_DIRECT) always succeed, but the succeeding I/O may return an error.
10374 + * See the aufs manual in detail.
10375 + *
10376 + * To keep this behaviour, aufs has to set NULL to ->get_xip_mem too, and the
10377 + * performance of fadvise() and madvise() may be affected.
10378 + */
10379 +static void dy_adx(struct au_dyaop *dyaop, int do_dx)
10380 +{
10381 +       if (!do_dx) {
10382 +               dyaop->da_op.direct_IO = NULL;
10383 +               dyaop->da_op.get_xip_mem = NULL;
10384 +       } else {
10385 +               dyaop->da_op.direct_IO = aufs_aop.direct_IO;
10386 +               dyaop->da_op.get_xip_mem = aufs_aop.get_xip_mem;
10387 +               if (!dyaop->da_get_xip_mem)
10388 +                       dyaop->da_op.get_xip_mem = NULL;
10389 +       }
10390 +}
10391 +
10392 +static struct au_dyaop *dy_aget(struct au_branch *br,
10393 +                               const struct address_space_operations *h_aop,
10394 +                               int do_dx)
10395 +{
10396 +       struct au_dyaop *dyaop;
10397 +       struct au_dynop op;
10398 +
10399 +       op.dy_type = AuDy_AOP;
10400 +       op.dy_haop = h_aop;
10401 +       dyaop = (void *)dy_get(&op, br);
10402 +       if (IS_ERR(dyaop))
10403 +               goto out;
10404 +       dy_adx(dyaop, do_dx);
10405 +
10406 +out:
10407 +       return dyaop;
10408 +}
10409 +
10410 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
10411 +               struct inode *h_inode)
10412 +{
10413 +       int err, do_dx;
10414 +       struct super_block *sb;
10415 +       struct au_branch *br;
10416 +       struct au_dyaop *dyaop;
10417 +
10418 +       AuDebugOn(!S_ISREG(h_inode->i_mode));
10419 +       IiMustWriteLock(inode);
10420 +
10421 +       sb = inode->i_sb;
10422 +       br = au_sbr(sb, bindex);
10423 +       do_dx = !!au_opt_test(au_mntflags(sb), DIO);
10424 +       dyaop = dy_aget(br, h_inode->i_mapping->a_ops, do_dx);
10425 +       err = PTR_ERR(dyaop);
10426 +       if (IS_ERR(dyaop))
10427 +               /* unnecessary to call dy_fput() */
10428 +               goto out;
10429 +
10430 +       err = 0;
10431 +       inode->i_mapping->a_ops = &dyaop->da_op;
10432 +
10433 +out:
10434 +       return err;
10435 +}
10436 +
10437 +/*
10438 + * Is it safe to replace a_ops during the inode/file is in operation?
10439 + * Yes, I hope so.
10440 + */
10441 +int au_dy_irefresh(struct inode *inode)
10442 +{
10443 +       int err;
10444 +       aufs_bindex_t bstart;
10445 +       struct inode *h_inode;
10446 +
10447 +       err = 0;
10448 +       if (S_ISREG(inode->i_mode)) {
10449 +               bstart = au_ibstart(inode);
10450 +               h_inode = au_h_iptr(inode, bstart);
10451 +               err = au_dy_iaop(inode, bstart, h_inode);
10452 +       }
10453 +       return err;
10454 +}
10455 +
10456 +void au_dy_arefresh(int do_dx)
10457 +{
10458 +       struct au_splhead *spl;
10459 +       struct list_head *head;
10460 +       struct au_dykey *key;
10461 +
10462 +       spl = dynop + AuDy_AOP;
10463 +       head = &spl->head;
10464 +       spin_lock(&spl->spin);
10465 +       list_for_each_entry(key, head, dk_list)
10466 +               dy_adx((void *)key, do_dx);
10467 +       spin_unlock(&spl->spin);
10468 +}
10469 +
10470 +/* ---------------------------------------------------------------------- */
10471 +
10472 +void __init au_dy_init(void)
10473 +{
10474 +       int i;
10475 +
10476 +       /* make sure that 'struct au_dykey *' can be any type */
10477 +       BUILD_BUG_ON(offsetof(struct au_dyaop, da_key));
10478 +
10479 +       for (i = 0; i < AuDyLast; i++)
10480 +               au_spl_init(dynop + i);
10481 +}
10482 +
10483 +void au_dy_fin(void)
10484 +{
10485 +       int i;
10486 +
10487 +       for (i = 0; i < AuDyLast; i++)
10488 +               WARN_ON(!list_empty(&dynop[i].head));
10489 +}
10490 diff -urN /usr/share/empty/fs/aufs/dynop.h linux/fs/aufs/dynop.h
10491 --- /usr/share/empty/fs/aufs/dynop.h    1970-01-01 01:00:00.000000000 +0100
10492 +++ linux/fs/aufs/dynop.h       2015-03-27 21:56:35.463461668 +0100
10493 @@ -0,0 +1,75 @@
10494 +/*
10495 + * Copyright (C) 2010-2015 Junjiro R. Okajima
10496 + *
10497 + * This program, aufs is free software; you can redistribute it and/or modify
10498 + * it under the terms of the GNU General Public License as published by
10499 + * the Free Software Foundation; either version 2 of the License, or
10500 + * (at your option) any later version.
10501 + *
10502 + * This program is distributed in the hope that it will be useful,
10503 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10504 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10505 + * GNU General Public License for more details.
10506 + *
10507 + * You should have received a copy of the GNU General Public License
10508 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10509 + */
10510 +
10511 +/*
10512 + * dynamically customizable operations (for regular files only)
10513 + */
10514 +
10515 +#ifndef __AUFS_DYNOP_H__
10516 +#define __AUFS_DYNOP_H__
10517 +
10518 +#ifdef __KERNEL__
10519 +
10520 +#include "inode.h"
10521 +
10522 +enum {AuDy_AOP, AuDyLast};
10523 +
10524 +struct au_dynop {
10525 +       int                                             dy_type;
10526 +       union {
10527 +               const void                              *dy_hop;
10528 +               const struct address_space_operations   *dy_haop;
10529 +       };
10530 +};
10531 +
10532 +struct au_dykey {
10533 +       union {
10534 +               struct list_head        dk_list;
10535 +               struct rcu_head         dk_rcu;
10536 +       };
10537 +       struct au_dynop         dk_op;
10538 +
10539 +       /*
10540 +        * during I am in the branch local array, kref is gotten. when the
10541 +        * branch is removed, kref is put.
10542 +        */
10543 +       struct kref             dk_kref;
10544 +};
10545 +
10546 +/* stop unioning since their sizes are very different from each other */
10547 +struct au_dyaop {
10548 +       struct au_dykey                 da_key;
10549 +       struct address_space_operations da_op; /* not const */
10550 +       int (*da_get_xip_mem)(struct address_space *, pgoff_t, int,
10551 +                             void **, unsigned long *);
10552 +};
10553 +
10554 +/* ---------------------------------------------------------------------- */
10555 +
10556 +/* dynop.c */
10557 +struct au_branch;
10558 +void au_dy_put(struct au_dykey *key);
10559 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
10560 +               struct inode *h_inode);
10561 +int au_dy_irefresh(struct inode *inode);
10562 +void au_dy_arefresh(int do_dio);
10563 +
10564 +void __init au_dy_init(void);
10565 +void au_dy_fin(void);
10566 +
10567 +#endif /* __KERNEL__ */
10568 +#endif /* __AUFS_DYNOP_H__ */
10569 diff -urN /usr/share/empty/fs/aufs/export.c linux/fs/aufs/export.c
10570 --- /usr/share/empty/fs/aufs/export.c   1970-01-01 01:00:00.000000000 +0100
10571 +++ linux/fs/aufs/export.c      2015-03-27 21:56:35.463461668 +0100
10572 @@ -0,0 +1,831 @@
10573 +/*
10574 + * Copyright (C) 2005-2015 Junjiro R. Okajima
10575 + *
10576 + * This program, aufs is free software; you can redistribute it and/or modify
10577 + * it under the terms of the GNU General Public License as published by
10578 + * the Free Software Foundation; either version 2 of the License, or
10579 + * (at your option) any later version.
10580 + *
10581 + * This program is distributed in the hope that it will be useful,
10582 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10583 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10584 + * GNU General Public License for more details.
10585 + *
10586 + * You should have received a copy of the GNU General Public License
10587 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10588 + */
10589 +
10590 +/*
10591 + * export via nfs
10592 + */
10593 +
10594 +#include <linux/exportfs.h>
10595 +#include <linux/fs_struct.h>
10596 +#include <linux/namei.h>
10597 +#include <linux/nsproxy.h>
10598 +#include <linux/random.h>
10599 +#include <linux/writeback.h>
10600 +#include "../fs/mount.h"
10601 +#include "aufs.h"
10602 +
10603 +union conv {
10604 +#ifdef CONFIG_AUFS_INO_T_64
10605 +       __u32 a[2];
10606 +#else
10607 +       __u32 a[1];
10608 +#endif
10609 +       ino_t ino;
10610 +};
10611 +
10612 +static ino_t decode_ino(__u32 *a)
10613 +{
10614 +       union conv u;
10615 +
10616 +       BUILD_BUG_ON(sizeof(u.ino) != sizeof(u.a));
10617 +       u.a[0] = a[0];
10618 +#ifdef CONFIG_AUFS_INO_T_64
10619 +       u.a[1] = a[1];
10620 +#endif
10621 +       return u.ino;
10622 +}
10623 +
10624 +static void encode_ino(__u32 *a, ino_t ino)
10625 +{
10626 +       union conv u;
10627 +
10628 +       u.ino = ino;
10629 +       a[0] = u.a[0];
10630 +#ifdef CONFIG_AUFS_INO_T_64
10631 +       a[1] = u.a[1];
10632 +#endif
10633 +}
10634 +
10635 +/* NFS file handle */
10636 +enum {
10637 +       Fh_br_id,
10638 +       Fh_sigen,
10639 +#ifdef CONFIG_AUFS_INO_T_64
10640 +       /* support 64bit inode number */
10641 +       Fh_ino1,
10642 +       Fh_ino2,
10643 +       Fh_dir_ino1,
10644 +       Fh_dir_ino2,
10645 +#else
10646 +       Fh_ino1,
10647 +       Fh_dir_ino1,
10648 +#endif
10649 +       Fh_igen,
10650 +       Fh_h_type,
10651 +       Fh_tail,
10652 +
10653 +       Fh_ino = Fh_ino1,
10654 +       Fh_dir_ino = Fh_dir_ino1
10655 +};
10656 +
10657 +static int au_test_anon(struct dentry *dentry)
10658 +{
10659 +       /* note: read d_flags without d_lock */
10660 +       return !!(dentry->d_flags & DCACHE_DISCONNECTED);
10661 +}
10662 +
10663 +int au_test_nfsd(void)
10664 +{
10665 +       int ret;
10666 +       struct task_struct *tsk = current;
10667 +       char comm[sizeof(tsk->comm)];
10668 +
10669 +       ret = 0;
10670 +       if (tsk->flags & PF_KTHREAD) {
10671 +               get_task_comm(comm, tsk);
10672 +               ret = !strcmp(comm, "nfsd");
10673 +       }
10674 +
10675 +       return ret;
10676 +}
10677 +
10678 +/* ---------------------------------------------------------------------- */
10679 +/* inode generation external table */
10680 +
10681 +void au_xigen_inc(struct inode *inode)
10682 +{
10683 +       loff_t pos;
10684 +       ssize_t sz;
10685 +       __u32 igen;
10686 +       struct super_block *sb;
10687 +       struct au_sbinfo *sbinfo;
10688 +
10689 +       sb = inode->i_sb;
10690 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
10691 +
10692 +       sbinfo = au_sbi(sb);
10693 +       pos = inode->i_ino;
10694 +       pos *= sizeof(igen);
10695 +       igen = inode->i_generation + 1;
10696 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xigen, &igen,
10697 +                        sizeof(igen), &pos);
10698 +       if (sz == sizeof(igen))
10699 +               return; /* success */
10700 +
10701 +       if (unlikely(sz >= 0))
10702 +               AuIOErr("xigen error (%zd)\n", sz);
10703 +}
10704 +
10705 +int au_xigen_new(struct inode *inode)
10706 +{
10707 +       int err;
10708 +       loff_t pos;
10709 +       ssize_t sz;
10710 +       struct super_block *sb;
10711 +       struct au_sbinfo *sbinfo;
10712 +       struct file *file;
10713 +
10714 +       err = 0;
10715 +       /* todo: dirty, at mount time */
10716 +       if (inode->i_ino == AUFS_ROOT_INO)
10717 +               goto out;
10718 +       sb = inode->i_sb;
10719 +       SiMustAnyLock(sb);
10720 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
10721 +               goto out;
10722 +
10723 +       err = -EFBIG;
10724 +       pos = inode->i_ino;
10725 +       if (unlikely(au_loff_max / sizeof(inode->i_generation) - 1 < pos)) {
10726 +               AuIOErr1("too large i%lld\n", pos);
10727 +               goto out;
10728 +       }
10729 +       pos *= sizeof(inode->i_generation);
10730 +
10731 +       err = 0;
10732 +       sbinfo = au_sbi(sb);
10733 +       file = sbinfo->si_xigen;
10734 +       BUG_ON(!file);
10735 +
10736 +       if (vfsub_f_size_read(file)
10737 +           < pos + sizeof(inode->i_generation)) {
10738 +               inode->i_generation = atomic_inc_return(&sbinfo->si_xigen_next);
10739 +               sz = xino_fwrite(sbinfo->si_xwrite, file, &inode->i_generation,
10740 +                                sizeof(inode->i_generation), &pos);
10741 +       } else
10742 +               sz = xino_fread(sbinfo->si_xread, file, &inode->i_generation,
10743 +                               sizeof(inode->i_generation), &pos);
10744 +       if (sz == sizeof(inode->i_generation))
10745 +               goto out; /* success */
10746 +
10747 +       err = sz;
10748 +       if (unlikely(sz >= 0)) {
10749 +               err = -EIO;
10750 +               AuIOErr("xigen error (%zd)\n", sz);
10751 +       }
10752 +
10753 +out:
10754 +       return err;
10755 +}
10756 +
10757 +int au_xigen_set(struct super_block *sb, struct file *base)
10758 +{
10759 +       int err;
10760 +       struct au_sbinfo *sbinfo;
10761 +       struct file *file;
10762 +
10763 +       SiMustWriteLock(sb);
10764 +
10765 +       sbinfo = au_sbi(sb);
10766 +       file = au_xino_create2(base, sbinfo->si_xigen);
10767 +       err = PTR_ERR(file);
10768 +       if (IS_ERR(file))
10769 +               goto out;
10770 +       err = 0;
10771 +       if (sbinfo->si_xigen)
10772 +               fput(sbinfo->si_xigen);
10773 +       sbinfo->si_xigen = file;
10774 +
10775 +out:
10776 +       return err;
10777 +}
10778 +
10779 +void au_xigen_clr(struct super_block *sb)
10780 +{
10781 +       struct au_sbinfo *sbinfo;
10782 +
10783 +       SiMustWriteLock(sb);
10784 +
10785 +       sbinfo = au_sbi(sb);
10786 +       if (sbinfo->si_xigen) {
10787 +               fput(sbinfo->si_xigen);
10788 +               sbinfo->si_xigen = NULL;
10789 +       }
10790 +}
10791 +
10792 +/* ---------------------------------------------------------------------- */
10793 +
10794 +static struct dentry *decode_by_ino(struct super_block *sb, ino_t ino,
10795 +                                   ino_t dir_ino)
10796 +{
10797 +       struct dentry *dentry, *d;
10798 +       struct inode *inode;
10799 +       unsigned int sigen;
10800 +
10801 +       dentry = NULL;
10802 +       inode = ilookup(sb, ino);
10803 +       if (!inode)
10804 +               goto out;
10805 +
10806 +       dentry = ERR_PTR(-ESTALE);
10807 +       sigen = au_sigen(sb);
10808 +       if (unlikely(is_bad_inode(inode)
10809 +                    || IS_DEADDIR(inode)
10810 +                    || sigen != au_iigen(inode, NULL)))
10811 +               goto out_iput;
10812 +
10813 +       dentry = NULL;
10814 +       if (!dir_ino || S_ISDIR(inode->i_mode))
10815 +               dentry = d_find_alias(inode);
10816 +       else {
10817 +               spin_lock(&inode->i_lock);
10818 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
10819 +                       spin_lock(&d->d_lock);
10820 +                       if (!au_test_anon(d)
10821 +                           && d->d_parent->d_inode->i_ino == dir_ino) {
10822 +                               dentry = dget_dlock(d);
10823 +                               spin_unlock(&d->d_lock);
10824 +                               break;
10825 +                       }
10826 +                       spin_unlock(&d->d_lock);
10827 +               }
10828 +               spin_unlock(&inode->i_lock);
10829 +       }
10830 +       if (unlikely(dentry && au_digen_test(dentry, sigen))) {
10831 +               /* need to refresh */
10832 +               dput(dentry);
10833 +               dentry = NULL;
10834 +       }
10835 +
10836 +out_iput:
10837 +       iput(inode);
10838 +out:
10839 +       AuTraceErrPtr(dentry);
10840 +       return dentry;
10841 +}
10842 +
10843 +/* ---------------------------------------------------------------------- */
10844 +
10845 +/* todo: dirty? */
10846 +/* if exportfs_decode_fh() passed vfsmount*, we could be happy */
10847 +
10848 +struct au_compare_mnt_args {
10849 +       /* input */
10850 +       struct super_block *sb;
10851 +
10852 +       /* output */
10853 +       struct vfsmount *mnt;
10854 +};
10855 +
10856 +static int au_compare_mnt(struct vfsmount *mnt, void *arg)
10857 +{
10858 +       struct au_compare_mnt_args *a = arg;
10859 +
10860 +       if (mnt->mnt_sb != a->sb)
10861 +               return 0;
10862 +       a->mnt = mntget(mnt);
10863 +       return 1;
10864 +}
10865 +
10866 +static struct vfsmount *au_mnt_get(struct super_block *sb)
10867 +{
10868 +       int err;
10869 +       struct path root;
10870 +       struct au_compare_mnt_args args = {
10871 +               .sb = sb
10872 +       };
10873 +
10874 +       get_fs_root(current->fs, &root);
10875 +       rcu_read_lock();
10876 +       err = iterate_mounts(au_compare_mnt, &args, root.mnt);
10877 +       rcu_read_unlock();
10878 +       path_put(&root);
10879 +       AuDebugOn(!err);
10880 +       AuDebugOn(!args.mnt);
10881 +       return args.mnt;
10882 +}
10883 +
10884 +struct au_nfsd_si_lock {
10885 +       unsigned int sigen;
10886 +       aufs_bindex_t bindex, br_id;
10887 +       unsigned char force_lock;
10888 +};
10889 +
10890 +static int si_nfsd_read_lock(struct super_block *sb,
10891 +                            struct au_nfsd_si_lock *nsi_lock)
10892 +{
10893 +       int err;
10894 +       aufs_bindex_t bindex;
10895 +
10896 +       si_read_lock(sb, AuLock_FLUSH);
10897 +
10898 +       /* branch id may be wrapped around */
10899 +       err = 0;
10900 +       bindex = au_br_index(sb, nsi_lock->br_id);
10901 +       if (bindex >= 0 && nsi_lock->sigen + AUFS_BRANCH_MAX > au_sigen(sb))
10902 +               goto out; /* success */
10903 +
10904 +       err = -ESTALE;
10905 +       bindex = -1;
10906 +       if (!nsi_lock->force_lock)
10907 +               si_read_unlock(sb);
10908 +
10909 +out:
10910 +       nsi_lock->bindex = bindex;
10911 +       return err;
10912 +}
10913 +
10914 +struct find_name_by_ino {
10915 +       struct dir_context ctx;
10916 +       int called, found;
10917 +       ino_t ino;
10918 +       char *name;
10919 +       int namelen;
10920 +};
10921 +
10922 +static int
10923 +find_name_by_ino(struct dir_context *ctx, const char *name, int namelen,
10924 +                loff_t offset, u64 ino, unsigned int d_type)
10925 +{
10926 +       struct find_name_by_ino *a = container_of(ctx, struct find_name_by_ino,
10927 +                                                 ctx);
10928 +
10929 +       a->called++;
10930 +       if (a->ino != ino)
10931 +               return 0;
10932 +
10933 +       memcpy(a->name, name, namelen);
10934 +       a->namelen = namelen;
10935 +       a->found = 1;
10936 +       return 1;
10937 +}
10938 +
10939 +static struct dentry *au_lkup_by_ino(struct path *path, ino_t ino,
10940 +                                    struct au_nfsd_si_lock *nsi_lock)
10941 +{
10942 +       struct dentry *dentry, *parent;
10943 +       struct file *file;
10944 +       struct inode *dir;
10945 +       struct find_name_by_ino arg = {
10946 +               .ctx = {
10947 +                       .actor = find_name_by_ino
10948 +               }
10949 +       };
10950 +       int err;
10951 +
10952 +       parent = path->dentry;
10953 +       if (nsi_lock)
10954 +               si_read_unlock(parent->d_sb);
10955 +       file = vfsub_dentry_open(path, au_dir_roflags);
10956 +       dentry = (void *)file;
10957 +       if (IS_ERR(file))
10958 +               goto out;
10959 +
10960 +       dentry = ERR_PTR(-ENOMEM);
10961 +       arg.name = (void *)__get_free_page(GFP_NOFS);
10962 +       if (unlikely(!arg.name))
10963 +               goto out_file;
10964 +       arg.ino = ino;
10965 +       arg.found = 0;
10966 +       do {
10967 +               arg.called = 0;
10968 +               /* smp_mb(); */
10969 +               err = vfsub_iterate_dir(file, &arg.ctx);
10970 +       } while (!err && !arg.found && arg.called);
10971 +       dentry = ERR_PTR(err);
10972 +       if (unlikely(err))
10973 +               goto out_name;
10974 +       /* instead of ENOENT */
10975 +       dentry = ERR_PTR(-ESTALE);
10976 +       if (!arg.found)
10977 +               goto out_name;
10978 +
10979 +       /* do not call vfsub_lkup_one() */
10980 +       dir = parent->d_inode;
10981 +       mutex_lock(&dir->i_mutex);
10982 +       dentry = vfsub_lookup_one_len(arg.name, parent, arg.namelen);
10983 +       mutex_unlock(&dir->i_mutex);
10984 +       AuTraceErrPtr(dentry);
10985 +       if (IS_ERR(dentry))
10986 +               goto out_name;
10987 +       AuDebugOn(au_test_anon(dentry));
10988 +       if (unlikely(!dentry->d_inode)) {
10989 +               dput(dentry);
10990 +               dentry = ERR_PTR(-ENOENT);
10991 +       }
10992 +
10993 +out_name:
10994 +       free_page((unsigned long)arg.name);
10995 +out_file:
10996 +       fput(file);
10997 +out:
10998 +       if (unlikely(nsi_lock
10999 +                    && si_nfsd_read_lock(parent->d_sb, nsi_lock) < 0))
11000 +               if (!IS_ERR(dentry)) {
11001 +                       dput(dentry);
11002 +                       dentry = ERR_PTR(-ESTALE);
11003 +               }
11004 +       AuTraceErrPtr(dentry);
11005 +       return dentry;
11006 +}
11007 +
11008 +static struct dentry *decode_by_dir_ino(struct super_block *sb, ino_t ino,
11009 +                                       ino_t dir_ino,
11010 +                                       struct au_nfsd_si_lock *nsi_lock)
11011 +{
11012 +       struct dentry *dentry;
11013 +       struct path path;
11014 +
11015 +       if (dir_ino != AUFS_ROOT_INO) {
11016 +               path.dentry = decode_by_ino(sb, dir_ino, 0);
11017 +               dentry = path.dentry;
11018 +               if (!path.dentry || IS_ERR(path.dentry))
11019 +                       goto out;
11020 +               AuDebugOn(au_test_anon(path.dentry));
11021 +       } else
11022 +               path.dentry = dget(sb->s_root);
11023 +
11024 +       path.mnt = au_mnt_get(sb);
11025 +       dentry = au_lkup_by_ino(&path, ino, nsi_lock);
11026 +       path_put(&path);
11027 +
11028 +out:
11029 +       AuTraceErrPtr(dentry);
11030 +       return dentry;
11031 +}
11032 +
11033 +/* ---------------------------------------------------------------------- */
11034 +
11035 +static int h_acceptable(void *expv, struct dentry *dentry)
11036 +{
11037 +       return 1;
11038 +}
11039 +
11040 +static char *au_build_path(struct dentry *h_parent, struct path *h_rootpath,
11041 +                          char *buf, int len, struct super_block *sb)
11042 +{
11043 +       char *p;
11044 +       int n;
11045 +       struct path path;
11046 +
11047 +       p = d_path(h_rootpath, buf, len);
11048 +       if (IS_ERR(p))
11049 +               goto out;
11050 +       n = strlen(p);
11051 +
11052 +       path.mnt = h_rootpath->mnt;
11053 +       path.dentry = h_parent;
11054 +       p = d_path(&path, buf, len);
11055 +       if (IS_ERR(p))
11056 +               goto out;
11057 +       if (n != 1)
11058 +               p += n;
11059 +
11060 +       path.mnt = au_mnt_get(sb);
11061 +       path.dentry = sb->s_root;
11062 +       p = d_path(&path, buf, len - strlen(p));
11063 +       mntput(path.mnt);
11064 +       if (IS_ERR(p))
11065 +               goto out;
11066 +       if (n != 1)
11067 +               p[strlen(p)] = '/';
11068 +
11069 +out:
11070 +       AuTraceErrPtr(p);
11071 +       return p;
11072 +}
11073 +
11074 +static
11075 +struct dentry *decode_by_path(struct super_block *sb, ino_t ino, __u32 *fh,
11076 +                             int fh_len, struct au_nfsd_si_lock *nsi_lock)
11077 +{
11078 +       struct dentry *dentry, *h_parent, *root;
11079 +       struct super_block *h_sb;
11080 +       char *pathname, *p;
11081 +       struct vfsmount *h_mnt;
11082 +       struct au_branch *br;
11083 +       int err;
11084 +       struct path path;
11085 +
11086 +       br = au_sbr(sb, nsi_lock->bindex);
11087 +       h_mnt = au_br_mnt(br);
11088 +       h_sb = h_mnt->mnt_sb;
11089 +       /* todo: call lower fh_to_dentry()? fh_to_parent()? */
11090 +       h_parent = exportfs_decode_fh(h_mnt, (void *)(fh + Fh_tail),
11091 +                                     fh_len - Fh_tail, fh[Fh_h_type],
11092 +                                     h_acceptable, /*context*/NULL);
11093 +       dentry = h_parent;
11094 +       if (unlikely(!h_parent || IS_ERR(h_parent))) {
11095 +               AuWarn1("%s decode_fh failed, %ld\n",
11096 +                       au_sbtype(h_sb), PTR_ERR(h_parent));
11097 +               goto out;
11098 +       }
11099 +       dentry = NULL;
11100 +       if (unlikely(au_test_anon(h_parent))) {
11101 +               AuWarn1("%s decode_fh returned a disconnected dentry\n",
11102 +                       au_sbtype(h_sb));
11103 +               goto out_h_parent;
11104 +       }
11105 +
11106 +       dentry = ERR_PTR(-ENOMEM);
11107 +       pathname = (void *)__get_free_page(GFP_NOFS);
11108 +       if (unlikely(!pathname))
11109 +               goto out_h_parent;
11110 +
11111 +       root = sb->s_root;
11112 +       path.mnt = h_mnt;
11113 +       di_read_lock_parent(root, !AuLock_IR);
11114 +       path.dentry = au_h_dptr(root, nsi_lock->bindex);
11115 +       di_read_unlock(root, !AuLock_IR);
11116 +       p = au_build_path(h_parent, &path, pathname, PAGE_SIZE, sb);
11117 +       dentry = (void *)p;
11118 +       if (IS_ERR(p))
11119 +               goto out_pathname;
11120 +
11121 +       si_read_unlock(sb);
11122 +       err = vfsub_kern_path(p, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &path);
11123 +       dentry = ERR_PTR(err);
11124 +       if (unlikely(err))
11125 +               goto out_relock;
11126 +
11127 +       dentry = ERR_PTR(-ENOENT);
11128 +       AuDebugOn(au_test_anon(path.dentry));
11129 +       if (unlikely(!path.dentry->d_inode))
11130 +               goto out_path;
11131 +
11132 +       if (ino != path.dentry->d_inode->i_ino)
11133 +               dentry = au_lkup_by_ino(&path, ino, /*nsi_lock*/NULL);
11134 +       else
11135 +               dentry = dget(path.dentry);
11136 +
11137 +out_path:
11138 +       path_put(&path);
11139 +out_relock:
11140 +       if (unlikely(si_nfsd_read_lock(sb, nsi_lock) < 0))
11141 +               if (!IS_ERR(dentry)) {
11142 +                       dput(dentry);
11143 +                       dentry = ERR_PTR(-ESTALE);
11144 +               }
11145 +out_pathname:
11146 +       free_page((unsigned long)pathname);
11147 +out_h_parent:
11148 +       dput(h_parent);
11149 +out:
11150 +       AuTraceErrPtr(dentry);
11151 +       return dentry;
11152 +}
11153 +
11154 +/* ---------------------------------------------------------------------- */
11155 +
11156 +static struct dentry *
11157 +aufs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
11158 +                 int fh_type)
11159 +{
11160 +       struct dentry *dentry;
11161 +       __u32 *fh = fid->raw;
11162 +       struct au_branch *br;
11163 +       ino_t ino, dir_ino;
11164 +       struct au_nfsd_si_lock nsi_lock = {
11165 +               .force_lock     = 0
11166 +       };
11167 +
11168 +       dentry = ERR_PTR(-ESTALE);
11169 +       /* it should never happen, but the file handle is unreliable */
11170 +       if (unlikely(fh_len < Fh_tail))
11171 +               goto out;
11172 +       nsi_lock.sigen = fh[Fh_sigen];
11173 +       nsi_lock.br_id = fh[Fh_br_id];
11174 +
11175 +       /* branch id may be wrapped around */
11176 +       br = NULL;
11177 +       if (unlikely(si_nfsd_read_lock(sb, &nsi_lock)))
11178 +               goto out;
11179 +       nsi_lock.force_lock = 1;
11180 +
11181 +       /* is this inode still cached? */
11182 +       ino = decode_ino(fh + Fh_ino);
11183 +       /* it should never happen */
11184 +       if (unlikely(ino == AUFS_ROOT_INO))
11185 +               goto out;
11186 +
11187 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11188 +       dentry = decode_by_ino(sb, ino, dir_ino);
11189 +       if (IS_ERR(dentry))
11190 +               goto out_unlock;
11191 +       if (dentry)
11192 +               goto accept;
11193 +
11194 +       /* is the parent dir cached? */
11195 +       br = au_sbr(sb, nsi_lock.bindex);
11196 +       atomic_inc(&br->br_count);
11197 +       dentry = decode_by_dir_ino(sb, ino, dir_ino, &nsi_lock);
11198 +       if (IS_ERR(dentry))
11199 +               goto out_unlock;
11200 +       if (dentry)
11201 +               goto accept;
11202 +
11203 +       /* lookup path */
11204 +       dentry = decode_by_path(sb, ino, fh, fh_len, &nsi_lock);
11205 +       if (IS_ERR(dentry))
11206 +               goto out_unlock;
11207 +       if (unlikely(!dentry))
11208 +               /* todo?: make it ESTALE */
11209 +               goto out_unlock;
11210 +
11211 +accept:
11212 +       if (!au_digen_test(dentry, au_sigen(sb))
11213 +           && dentry->d_inode->i_generation == fh[Fh_igen])
11214 +               goto out_unlock; /* success */
11215 +
11216 +       dput(dentry);
11217 +       dentry = ERR_PTR(-ESTALE);
11218 +out_unlock:
11219 +       if (br)
11220 +               atomic_dec(&br->br_count);
11221 +       si_read_unlock(sb);
11222 +out:
11223 +       AuTraceErrPtr(dentry);
11224 +       return dentry;
11225 +}
11226 +
11227 +#if 0 /* reserved for future use */
11228 +/* support subtreecheck option */
11229 +static struct dentry *aufs_fh_to_parent(struct super_block *sb, struct fid *fid,
11230 +                                       int fh_len, int fh_type)
11231 +{
11232 +       struct dentry *parent;
11233 +       __u32 *fh = fid->raw;
11234 +       ino_t dir_ino;
11235 +
11236 +       dir_ino = decode_ino(fh + Fh_dir_ino);
11237 +       parent = decode_by_ino(sb, dir_ino, 0);
11238 +       if (IS_ERR(parent))
11239 +               goto out;
11240 +       if (!parent)
11241 +               parent = decode_by_path(sb, au_br_index(sb, fh[Fh_br_id]),
11242 +                                       dir_ino, fh, fh_len);
11243 +
11244 +out:
11245 +       AuTraceErrPtr(parent);
11246 +       return parent;
11247 +}
11248 +#endif
11249 +
11250 +/* ---------------------------------------------------------------------- */
11251 +
11252 +static int aufs_encode_fh(struct inode *inode, __u32 *fh, int *max_len,
11253 +                         struct inode *dir)
11254 +{
11255 +       int err;
11256 +       aufs_bindex_t bindex;
11257 +       struct super_block *sb, *h_sb;
11258 +       struct dentry *dentry, *parent, *h_parent;
11259 +       struct inode *h_dir;
11260 +       struct au_branch *br;
11261 +
11262 +       err = -ENOSPC;
11263 +       if (unlikely(*max_len <= Fh_tail)) {
11264 +               AuWarn1("NFSv2 client (max_len %d)?\n", *max_len);
11265 +               goto out;
11266 +       }
11267 +
11268 +       err = FILEID_ROOT;
11269 +       if (inode->i_ino == AUFS_ROOT_INO) {
11270 +               AuDebugOn(inode->i_ino != AUFS_ROOT_INO);
11271 +               goto out;
11272 +       }
11273 +
11274 +       h_parent = NULL;
11275 +       sb = inode->i_sb;
11276 +       err = si_read_lock(sb, AuLock_FLUSH);
11277 +       if (unlikely(err))
11278 +               goto out;
11279 +
11280 +#ifdef CONFIG_AUFS_DEBUG
11281 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
11282 +               AuWarn1("NFS-exporting requires xino\n");
11283 +#endif
11284 +       err = -EIO;
11285 +       parent = NULL;
11286 +       ii_read_lock_child(inode);
11287 +       bindex = au_ibstart(inode);
11288 +       if (!dir) {
11289 +               dentry = d_find_any_alias(inode);
11290 +               if (unlikely(!dentry))
11291 +                       goto out_unlock;
11292 +               AuDebugOn(au_test_anon(dentry));
11293 +               parent = dget_parent(dentry);
11294 +               dput(dentry);
11295 +               if (unlikely(!parent))
11296 +                       goto out_unlock;
11297 +               dir = parent->d_inode;
11298 +       }
11299 +
11300 +       ii_read_lock_parent(dir);
11301 +       h_dir = au_h_iptr(dir, bindex);
11302 +       ii_read_unlock(dir);
11303 +       if (unlikely(!h_dir))
11304 +               goto out_parent;
11305 +       h_parent = d_find_any_alias(h_dir);
11306 +       if (unlikely(!h_parent))
11307 +               goto out_hparent;
11308 +
11309 +       err = -EPERM;
11310 +       br = au_sbr(sb, bindex);
11311 +       h_sb = au_br_sb(br);
11312 +       if (unlikely(!h_sb->s_export_op)) {
11313 +               AuErr1("%s branch is not exportable\n", au_sbtype(h_sb));
11314 +               goto out_hparent;
11315 +       }
11316 +
11317 +       fh[Fh_br_id] = br->br_id;
11318 +       fh[Fh_sigen] = au_sigen(sb);
11319 +       encode_ino(fh + Fh_ino, inode->i_ino);
11320 +       encode_ino(fh + Fh_dir_ino, dir->i_ino);
11321 +       fh[Fh_igen] = inode->i_generation;
11322 +
11323 +       *max_len -= Fh_tail;
11324 +       fh[Fh_h_type] = exportfs_encode_fh(h_parent, (void *)(fh + Fh_tail),
11325 +                                          max_len,
11326 +                                          /*connectable or subtreecheck*/0);
11327 +       err = fh[Fh_h_type];
11328 +       *max_len += Fh_tail;
11329 +       /* todo: macros? */
11330 +       if (err != FILEID_INVALID)
11331 +               err = 99;
11332 +       else
11333 +               AuWarn1("%s encode_fh failed\n", au_sbtype(h_sb));
11334 +
11335 +out_hparent:
11336 +       dput(h_parent);
11337 +out_parent:
11338 +       dput(parent);
11339 +out_unlock:
11340 +       ii_read_unlock(inode);
11341 +       si_read_unlock(sb);
11342 +out:
11343 +       if (unlikely(err < 0))
11344 +               err = FILEID_INVALID;
11345 +       return err;
11346 +}
11347 +
11348 +/* ---------------------------------------------------------------------- */
11349 +
11350 +static int aufs_commit_metadata(struct inode *inode)
11351 +{
11352 +       int err;
11353 +       aufs_bindex_t bindex;
11354 +       struct super_block *sb;
11355 +       struct inode *h_inode;
11356 +       int (*f)(struct inode *inode);
11357 +
11358 +       sb = inode->i_sb;
11359 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
11360 +       ii_write_lock_child(inode);
11361 +       bindex = au_ibstart(inode);
11362 +       AuDebugOn(bindex < 0);
11363 +       h_inode = au_h_iptr(inode, bindex);
11364 +
11365 +       f = h_inode->i_sb->s_export_op->commit_metadata;
11366 +       if (f)
11367 +               err = f(h_inode);
11368 +       else {
11369 +               struct writeback_control wbc = {
11370 +                       .sync_mode      = WB_SYNC_ALL,
11371 +                       .nr_to_write    = 0 /* metadata only */
11372 +               };
11373 +
11374 +               err = sync_inode(h_inode, &wbc);
11375 +       }
11376 +
11377 +       au_cpup_attr_timesizes(inode);
11378 +       ii_write_unlock(inode);
11379 +       si_read_unlock(sb);
11380 +       return err;
11381 +}
11382 +
11383 +/* ---------------------------------------------------------------------- */
11384 +
11385 +static struct export_operations aufs_export_op = {
11386 +       .fh_to_dentry           = aufs_fh_to_dentry,
11387 +       /* .fh_to_parent        = aufs_fh_to_parent, */
11388 +       .encode_fh              = aufs_encode_fh,
11389 +       .commit_metadata        = aufs_commit_metadata
11390 +};
11391 +
11392 +void au_export_init(struct super_block *sb)
11393 +{
11394 +       struct au_sbinfo *sbinfo;
11395 +       __u32 u;
11396 +
11397 +       sb->s_export_op = &aufs_export_op;
11398 +       sbinfo = au_sbi(sb);
11399 +       sbinfo->si_xigen = NULL;
11400 +       get_random_bytes(&u, sizeof(u));
11401 +       BUILD_BUG_ON(sizeof(u) != sizeof(int));
11402 +       atomic_set(&sbinfo->si_xigen_next, u);
11403 +}
11404 diff -urN /usr/share/empty/fs/aufs/fhsm.c linux/fs/aufs/fhsm.c
11405 --- /usr/share/empty/fs/aufs/fhsm.c     1970-01-01 01:00:00.000000000 +0100
11406 +++ linux/fs/aufs/fhsm.c        2015-03-27 21:56:35.463461668 +0100
11407 @@ -0,0 +1,426 @@
11408 +/*
11409 + * Copyright (C) 2011-2015 Junjiro R. Okajima
11410 + *
11411 + * This program, aufs is free software; you can redistribute it and/or modify
11412 + * it under the terms of the GNU General Public License as published by
11413 + * the Free Software Foundation; either version 2 of the License, or
11414 + * (at your option) any later version.
11415 + *
11416 + * This program is distributed in the hope that it will be useful,
11417 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11418 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11419 + * GNU General Public License for more details.
11420 + *
11421 + * You should have received a copy of the GNU General Public License
11422 + * along with this program; if not, write to the Free Software
11423 + * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
11424 + */
11425 +
11426 +/*
11427 + * File-based Hierarchy Storage Management
11428 + */
11429 +
11430 +#include <linux/anon_inodes.h>
11431 +#include <linux/poll.h>
11432 +#include <linux/seq_file.h>
11433 +#include <linux/statfs.h>
11434 +#include "aufs.h"
11435 +
11436 +static aufs_bindex_t au_fhsm_bottom(struct super_block *sb)
11437 +{
11438 +       struct au_sbinfo *sbinfo;
11439 +       struct au_fhsm *fhsm;
11440 +
11441 +       SiMustAnyLock(sb);
11442 +
11443 +       sbinfo = au_sbi(sb);
11444 +       fhsm = &sbinfo->si_fhsm;
11445 +       AuDebugOn(!fhsm);
11446 +       return fhsm->fhsm_bottom;
11447 +}
11448 +
11449 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex)
11450 +{
11451 +       struct au_sbinfo *sbinfo;
11452 +       struct au_fhsm *fhsm;
11453 +
11454 +       SiMustWriteLock(sb);
11455 +
11456 +       sbinfo = au_sbi(sb);
11457 +       fhsm = &sbinfo->si_fhsm;
11458 +       AuDebugOn(!fhsm);
11459 +       fhsm->fhsm_bottom = bindex;
11460 +}
11461 +
11462 +/* ---------------------------------------------------------------------- */
11463 +
11464 +static int au_fhsm_test_jiffy(struct au_sbinfo *sbinfo, struct au_branch *br)
11465 +{
11466 +       struct au_br_fhsm *bf;
11467 +
11468 +       bf = br->br_fhsm;
11469 +       MtxMustLock(&bf->bf_lock);
11470 +
11471 +       return !bf->bf_readable
11472 +               || time_after(jiffies,
11473 +                             bf->bf_jiffy + sbinfo->si_fhsm.fhsm_expire);
11474 +}
11475 +
11476 +/* ---------------------------------------------------------------------- */
11477 +
11478 +static void au_fhsm_notify(struct super_block *sb, int val)
11479 +{
11480 +       struct au_sbinfo *sbinfo;
11481 +       struct au_fhsm *fhsm;
11482 +
11483 +       SiMustAnyLock(sb);
11484 +
11485 +       sbinfo = au_sbi(sb);
11486 +       fhsm = &sbinfo->si_fhsm;
11487 +       if (au_fhsm_pid(fhsm)
11488 +           && atomic_read(&fhsm->fhsm_readable) != -1) {
11489 +               atomic_set(&fhsm->fhsm_readable, val);
11490 +               if (val)
11491 +                       wake_up(&fhsm->fhsm_wqh);
11492 +       }
11493 +}
11494 +
11495 +static int au_fhsm_stfs(struct super_block *sb, aufs_bindex_t bindex,
11496 +                       struct aufs_stfs *rstfs, int do_lock, int do_notify)
11497 +{
11498 +       int err;
11499 +       struct au_branch *br;
11500 +       struct au_br_fhsm *bf;
11501 +
11502 +       br = au_sbr(sb, bindex);
11503 +       AuDebugOn(au_br_rdonly(br));
11504 +       bf = br->br_fhsm;
11505 +       AuDebugOn(!bf);
11506 +
11507 +       if (do_lock)
11508 +               mutex_lock(&bf->bf_lock);
11509 +       else
11510 +               MtxMustLock(&bf->bf_lock);
11511 +
11512 +       /* sb->s_root for NFS is unreliable */
11513 +       err = au_br_stfs(br, &bf->bf_stfs);
11514 +       if (unlikely(err)) {
11515 +               AuErr1("FHSM failed (%d), b%d, ignored.\n", bindex, err);
11516 +               goto out;
11517 +       }
11518 +
11519 +       bf->bf_jiffy = jiffies;
11520 +       bf->bf_readable = 1;
11521 +       if (do_notify)
11522 +               au_fhsm_notify(sb, /*val*/1);
11523 +       if (rstfs)
11524 +               *rstfs = bf->bf_stfs;
11525 +
11526 +out:
11527 +       if (do_lock)
11528 +               mutex_unlock(&bf->bf_lock);
11529 +       au_fhsm_notify(sb, /*val*/1);
11530 +
11531 +       return err;
11532 +}
11533 +
11534 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force)
11535 +{
11536 +       int err;
11537 +       struct au_sbinfo *sbinfo;
11538 +       struct au_fhsm *fhsm;
11539 +       struct au_branch *br;
11540 +       struct au_br_fhsm *bf;
11541 +
11542 +       AuDbg("b%d, force %d\n", bindex, force);
11543 +       SiMustAnyLock(sb);
11544 +
11545 +       sbinfo = au_sbi(sb);
11546 +       fhsm = &sbinfo->si_fhsm;
11547 +       if (!au_ftest_si(sbinfo, FHSM)
11548 +           || fhsm->fhsm_bottom == bindex)
11549 +               return;
11550 +
11551 +       br = au_sbr(sb, bindex);
11552 +       bf = br->br_fhsm;
11553 +       AuDebugOn(!bf);
11554 +       mutex_lock(&bf->bf_lock);
11555 +       if (force
11556 +           || au_fhsm_pid(fhsm)
11557 +           || au_fhsm_test_jiffy(sbinfo, br))
11558 +               err = au_fhsm_stfs(sb, bindex, /*rstfs*/NULL, /*do_lock*/0,
11559 +                                 /*do_notify*/1);
11560 +       mutex_unlock(&bf->bf_lock);
11561 +}
11562 +
11563 +void au_fhsm_wrote_all(struct super_block *sb, int force)
11564 +{
11565 +       aufs_bindex_t bindex, bend;
11566 +       struct au_branch *br;
11567 +
11568 +       /* exclude the bottom */
11569 +       bend = au_fhsm_bottom(sb);
11570 +       for (bindex = 0; bindex < bend; bindex++) {
11571 +               br = au_sbr(sb, bindex);
11572 +               if (au_br_fhsm(br->br_perm))
11573 +                       au_fhsm_wrote(sb, bindex, force);
11574 +       }
11575 +}
11576 +
11577 +/* ---------------------------------------------------------------------- */
11578 +
11579 +static unsigned int au_fhsm_poll(struct file *file,
11580 +                                struct poll_table_struct *wait)
11581 +{
11582 +       unsigned int mask;
11583 +       struct au_sbinfo *sbinfo;
11584 +       struct au_fhsm *fhsm;
11585 +
11586 +       mask = 0;
11587 +       sbinfo = file->private_data;
11588 +       fhsm = &sbinfo->si_fhsm;
11589 +       poll_wait(file, &fhsm->fhsm_wqh, wait);
11590 +       if (atomic_read(&fhsm->fhsm_readable))
11591 +               mask = POLLIN /* | POLLRDNORM */;
11592 +
11593 +       AuTraceErr((int)mask);
11594 +       return mask;
11595 +}
11596 +
11597 +static int au_fhsm_do_read_one(struct aufs_stbr __user *stbr,
11598 +                             struct aufs_stfs *stfs, __s16 brid)
11599 +{
11600 +       int err;
11601 +
11602 +       err = copy_to_user(&stbr->stfs, stfs, sizeof(*stfs));
11603 +       if (!err)
11604 +               err = __put_user(brid, &stbr->brid);
11605 +       if (unlikely(err))
11606 +               err = -EFAULT;
11607 +
11608 +       return err;
11609 +}
11610 +
11611 +static ssize_t au_fhsm_do_read(struct super_block *sb,
11612 +                              struct aufs_stbr __user *stbr, size_t count)
11613 +{
11614 +       ssize_t err;
11615 +       int nstbr;
11616 +       aufs_bindex_t bindex, bend;
11617 +       struct au_branch *br;
11618 +       struct au_br_fhsm *bf;
11619 +
11620 +       /* except the bottom branch */
11621 +       err = 0;
11622 +       nstbr = 0;
11623 +       bend = au_fhsm_bottom(sb);
11624 +       for (bindex = 0; !err && bindex < bend; bindex++) {
11625 +               br = au_sbr(sb, bindex);
11626 +               if (!au_br_fhsm(br->br_perm))
11627 +                       continue;
11628 +
11629 +               bf = br->br_fhsm;
11630 +               mutex_lock(&bf->bf_lock);
11631 +               if (bf->bf_readable) {
11632 +                       err = -EFAULT;
11633 +                       if (count >= sizeof(*stbr))
11634 +                               err = au_fhsm_do_read_one(stbr++, &bf->bf_stfs,
11635 +                                                         br->br_id);
11636 +                       if (!err) {
11637 +                               bf->bf_readable = 0;
11638 +                               count -= sizeof(*stbr);
11639 +                               nstbr++;
11640 +                       }
11641 +               }
11642 +               mutex_unlock(&bf->bf_lock);
11643 +       }
11644 +       if (!err)
11645 +               err = sizeof(*stbr) * nstbr;
11646 +
11647 +       return err;
11648 +}
11649 +
11650 +static ssize_t au_fhsm_read(struct file *file, char __user *buf, size_t count,
11651 +                          loff_t *pos)
11652 +{
11653 +       ssize_t err;
11654 +       int readable;
11655 +       aufs_bindex_t nfhsm, bindex, bend;
11656 +       struct au_sbinfo *sbinfo;
11657 +       struct au_fhsm *fhsm;
11658 +       struct au_branch *br;
11659 +       struct super_block *sb;
11660 +
11661 +       err = 0;
11662 +       sbinfo = file->private_data;
11663 +       fhsm = &sbinfo->si_fhsm;
11664 +need_data:
11665 +       spin_lock_irq(&fhsm->fhsm_wqh.lock);
11666 +       if (!atomic_read(&fhsm->fhsm_readable)) {
11667 +               if (vfsub_file_flags(file) & O_NONBLOCK)
11668 +                       err = -EAGAIN;
11669 +               else
11670 +                       err = wait_event_interruptible_locked_irq
11671 +                               (fhsm->fhsm_wqh,
11672 +                                atomic_read(&fhsm->fhsm_readable));
11673 +       }
11674 +       spin_unlock_irq(&fhsm->fhsm_wqh.lock);
11675 +       if (unlikely(err))
11676 +               goto out;
11677 +
11678 +       /* sb may already be dead */
11679 +       au_rw_read_lock(&sbinfo->si_rwsem);
11680 +       readable = atomic_read(&fhsm->fhsm_readable);
11681 +       if (readable > 0) {
11682 +               sb = sbinfo->si_sb;
11683 +               AuDebugOn(!sb);
11684 +               /* exclude the bottom branch */
11685 +               nfhsm = 0;
11686 +               bend = au_fhsm_bottom(sb);
11687 +               for (bindex = 0; bindex < bend; bindex++) {
11688 +                       br = au_sbr(sb, bindex);
11689 +                       if (au_br_fhsm(br->br_perm))
11690 +                               nfhsm++;
11691 +               }
11692 +               err = -EMSGSIZE;
11693 +               if (nfhsm * sizeof(struct aufs_stbr) <= count) {
11694 +                       atomic_set(&fhsm->fhsm_readable, 0);
11695 +                       err = au_fhsm_do_read(sbinfo->si_sb, (void __user *)buf,
11696 +                                            count);
11697 +               }
11698 +       }
11699 +       au_rw_read_unlock(&sbinfo->si_rwsem);
11700 +       if (!readable)
11701 +               goto need_data;
11702 +
11703 +out:
11704 +       return err;
11705 +}
11706 +
11707 +static int au_fhsm_release(struct inode *inode, struct file *file)
11708 +{
11709 +       struct au_sbinfo *sbinfo;
11710 +       struct au_fhsm *fhsm;
11711 +
11712 +       /* sb may already be dead */
11713 +       sbinfo = file->private_data;
11714 +       fhsm = &sbinfo->si_fhsm;
11715 +       spin_lock(&fhsm->fhsm_spin);
11716 +       fhsm->fhsm_pid = 0;
11717 +       spin_unlock(&fhsm->fhsm_spin);
11718 +       kobject_put(&sbinfo->si_kobj);
11719 +
11720 +       return 0;
11721 +}
11722 +
11723 +static const struct file_operations au_fhsm_fops = {
11724 +       .owner          = THIS_MODULE,
11725 +       .llseek         = noop_llseek,
11726 +       .read           = au_fhsm_read,
11727 +       .poll           = au_fhsm_poll,
11728 +       .release        = au_fhsm_release
11729 +};
11730 +
11731 +int au_fhsm_fd(struct super_block *sb, int oflags)
11732 +{
11733 +       int err, fd;
11734 +       struct au_sbinfo *sbinfo;
11735 +       struct au_fhsm *fhsm;
11736 +
11737 +       err = -EPERM;
11738 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
11739 +               goto out;
11740 +
11741 +       err = -EINVAL;
11742 +       if (unlikely(oflags & ~(O_CLOEXEC | O_NONBLOCK)))
11743 +               goto out;
11744 +
11745 +       err = 0;
11746 +       sbinfo = au_sbi(sb);
11747 +       fhsm = &sbinfo->si_fhsm;
11748 +       spin_lock(&fhsm->fhsm_spin);
11749 +       if (!fhsm->fhsm_pid)
11750 +               fhsm->fhsm_pid = current->pid;
11751 +       else
11752 +               err = -EBUSY;
11753 +       spin_unlock(&fhsm->fhsm_spin);
11754 +       if (unlikely(err))
11755 +               goto out;
11756 +
11757 +       oflags |= O_RDONLY;
11758 +       /* oflags |= FMODE_NONOTIFY; */
11759 +       fd = anon_inode_getfd("[aufs_fhsm]", &au_fhsm_fops, sbinfo, oflags);
11760 +       err = fd;
11761 +       if (unlikely(fd < 0))
11762 +               goto out_pid;
11763 +
11764 +       /* succeed reglardless 'fhsm' status */
11765 +       kobject_get(&sbinfo->si_kobj);
11766 +       si_noflush_read_lock(sb);
11767 +       if (au_ftest_si(sbinfo, FHSM))
11768 +               au_fhsm_wrote_all(sb, /*force*/0);
11769 +       si_read_unlock(sb);
11770 +       goto out; /* success */
11771 +
11772 +out_pid:
11773 +       spin_lock(&fhsm->fhsm_spin);
11774 +       fhsm->fhsm_pid = 0;
11775 +       spin_unlock(&fhsm->fhsm_spin);
11776 +out:
11777 +       AuTraceErr(err);
11778 +       return err;
11779 +}
11780 +
11781 +/* ---------------------------------------------------------------------- */
11782 +
11783 +int au_fhsm_br_alloc(struct au_branch *br)
11784 +{
11785 +       int err;
11786 +
11787 +       err = 0;
11788 +       br->br_fhsm = kmalloc(sizeof(*br->br_fhsm), GFP_NOFS);
11789 +       if (br->br_fhsm)
11790 +               au_br_fhsm_init(br->br_fhsm);
11791 +       else
11792 +               err = -ENOMEM;
11793 +
11794 +       return err;
11795 +}
11796 +
11797 +/* ---------------------------------------------------------------------- */
11798 +
11799 +void au_fhsm_fin(struct super_block *sb)
11800 +{
11801 +       au_fhsm_notify(sb, /*val*/-1);
11802 +}
11803 +
11804 +void au_fhsm_init(struct au_sbinfo *sbinfo)
11805 +{
11806 +       struct au_fhsm *fhsm;
11807 +
11808 +       fhsm = &sbinfo->si_fhsm;
11809 +       spin_lock_init(&fhsm->fhsm_spin);
11810 +       init_waitqueue_head(&fhsm->fhsm_wqh);
11811 +       atomic_set(&fhsm->fhsm_readable, 0);
11812 +       fhsm->fhsm_expire
11813 +               = msecs_to_jiffies(AUFS_FHSM_CACHE_DEF_SEC * MSEC_PER_SEC);
11814 +       fhsm->fhsm_bottom = -1;
11815 +}
11816 +
11817 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec)
11818 +{
11819 +       sbinfo->si_fhsm.fhsm_expire
11820 +               = msecs_to_jiffies(sec * MSEC_PER_SEC);
11821 +}
11822 +
11823 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo)
11824 +{
11825 +       unsigned int u;
11826 +
11827 +       if (!au_ftest_si(sbinfo, FHSM))
11828 +               return;
11829 +
11830 +       u = jiffies_to_msecs(sbinfo->si_fhsm.fhsm_expire) / MSEC_PER_SEC;
11831 +       if (u != AUFS_FHSM_CACHE_DEF_SEC)
11832 +               seq_printf(seq, ",fhsm_sec=%u", u);
11833 +}
11834 diff -urN /usr/share/empty/fs/aufs/file.c linux/fs/aufs/file.c
11835 --- /usr/share/empty/fs/aufs/file.c     1970-01-01 01:00:00.000000000 +0100
11836 +++ linux/fs/aufs/file.c        2015-03-27 21:56:35.463461668 +0100
11837 @@ -0,0 +1,829 @@
11838 +/*
11839 + * Copyright (C) 2005-2015 Junjiro R. Okajima
11840 + *
11841 + * This program, aufs is free software; you can redistribute it and/or modify
11842 + * it under the terms of the GNU General Public License as published by
11843 + * the Free Software Foundation; either version 2 of the License, or
11844 + * (at your option) any later version.
11845 + *
11846 + * This program is distributed in the hope that it will be useful,
11847 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11848 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11849 + * GNU General Public License for more details.
11850 + *
11851 + * You should have received a copy of the GNU General Public License
11852 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
11853 + */
11854 +
11855 +/*
11856 + * handling file/dir, and address_space operation
11857 + */
11858 +
11859 +#ifdef CONFIG_AUFS_DEBUG
11860 +#include <linux/migrate.h>
11861 +#endif
11862 +#include <linux/pagemap.h>
11863 +#include "aufs.h"
11864 +
11865 +/* drop flags for writing */
11866 +unsigned int au_file_roflags(unsigned int flags)
11867 +{
11868 +       flags &= ~(O_WRONLY | O_RDWR | O_APPEND | O_CREAT | O_TRUNC);
11869 +       flags |= O_RDONLY | O_NOATIME;
11870 +       return flags;
11871 +}
11872 +
11873 +/* common functions to regular file and dir */
11874 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
11875 +                      struct file *file, int force_wr)
11876 +{
11877 +       struct file *h_file;
11878 +       struct dentry *h_dentry;
11879 +       struct inode *h_inode;
11880 +       struct super_block *sb;
11881 +       struct au_branch *br;
11882 +       struct path h_path;
11883 +       int err, exec_flag;
11884 +
11885 +       /* a race condition can happen between open and unlink/rmdir */
11886 +       h_file = ERR_PTR(-ENOENT);
11887 +       h_dentry = au_h_dptr(dentry, bindex);
11888 +       if (au_test_nfsd() && !h_dentry)
11889 +               goto out;
11890 +       h_inode = h_dentry->d_inode;
11891 +       if (au_test_nfsd() && !h_inode)
11892 +               goto out;
11893 +       spin_lock(&h_dentry->d_lock);
11894 +       err = (!d_unhashed(dentry) && d_unlinked(h_dentry))
11895 +               || !h_inode
11896 +               /* || !dentry->d_inode->i_nlink */
11897 +               ;
11898 +       spin_unlock(&h_dentry->d_lock);
11899 +       if (unlikely(err))
11900 +               goto out;
11901 +
11902 +       sb = dentry->d_sb;
11903 +       br = au_sbr(sb, bindex);
11904 +       h_file = ERR_PTR(-EACCES);
11905 +       exec_flag = flags & __FMODE_EXEC;
11906 +       if (exec_flag && (au_br_mnt(br)->mnt_flags & MNT_NOEXEC))
11907 +               goto out;
11908 +
11909 +       /* drop flags for writing */
11910 +       if (au_test_ro(sb, bindex, dentry->d_inode)) {
11911 +               if (force_wr && !(flags & O_WRONLY))
11912 +                       force_wr = 0;
11913 +               flags = au_file_roflags(flags);
11914 +               if (force_wr) {
11915 +                       h_file = ERR_PTR(-EROFS);
11916 +                       flags = au_file_roflags(flags);
11917 +                       if (unlikely(vfsub_native_ro(h_inode)
11918 +                                    || IS_APPEND(h_inode)))
11919 +                               goto out;
11920 +                       flags &= ~O_ACCMODE;
11921 +                       flags |= O_WRONLY;
11922 +               }
11923 +       }
11924 +       flags &= ~O_CREAT;
11925 +       atomic_inc(&br->br_count);
11926 +       h_path.dentry = h_dentry;
11927 +       h_path.mnt = au_br_mnt(br);
11928 +       h_file = vfsub_dentry_open(&h_path, flags);
11929 +       if (IS_ERR(h_file))
11930 +               goto out_br;
11931 +
11932 +       if (exec_flag) {
11933 +               err = deny_write_access(h_file);
11934 +               if (unlikely(err)) {
11935 +                       fput(h_file);
11936 +                       h_file = ERR_PTR(err);
11937 +                       goto out_br;
11938 +               }
11939 +       }
11940 +       fsnotify_open(h_file);
11941 +       goto out; /* success */
11942 +
11943 +out_br:
11944 +       atomic_dec(&br->br_count);
11945 +out:
11946 +       return h_file;
11947 +}
11948 +
11949 +static int au_cmoo(struct dentry *dentry)
11950 +{
11951 +       int err, cmoo;
11952 +       unsigned int udba;
11953 +       struct path h_path;
11954 +       struct au_pin pin;
11955 +       struct au_cp_generic cpg = {
11956 +               .dentry = dentry,
11957 +               .bdst   = -1,
11958 +               .bsrc   = -1,
11959 +               .len    = -1,
11960 +               .pin    = &pin,
11961 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
11962 +       };
11963 +       struct inode *inode, *delegated;
11964 +       struct super_block *sb;
11965 +       struct au_sbinfo *sbinfo;
11966 +       struct au_fhsm *fhsm;
11967 +       pid_t pid;
11968 +       struct au_branch *br;
11969 +       struct dentry *parent;
11970 +       struct au_hinode *hdir;
11971 +
11972 +       DiMustWriteLock(dentry);
11973 +       inode = dentry->d_inode;
11974 +       IiMustWriteLock(inode);
11975 +
11976 +       err = 0;
11977 +       if (IS_ROOT(dentry))
11978 +               goto out;
11979 +       cpg.bsrc = au_dbstart(dentry);
11980 +       if (!cpg.bsrc)
11981 +               goto out;
11982 +
11983 +       sb = dentry->d_sb;
11984 +       sbinfo = au_sbi(sb);
11985 +       fhsm = &sbinfo->si_fhsm;
11986 +       pid = au_fhsm_pid(fhsm);
11987 +       if (pid
11988 +           && (current->pid == pid
11989 +               || current->real_parent->pid == pid))
11990 +               goto out;
11991 +
11992 +       br = au_sbr(sb, cpg.bsrc);
11993 +       cmoo = au_br_cmoo(br->br_perm);
11994 +       if (!cmoo)
11995 +               goto out;
11996 +       if (!S_ISREG(inode->i_mode))
11997 +               cmoo &= AuBrAttr_COO_ALL;
11998 +       if (!cmoo)
11999 +               goto out;
12000 +
12001 +       parent = dget_parent(dentry);
12002 +       di_write_lock_parent(parent);
12003 +       err = au_wbr_do_copyup_bu(dentry, cpg.bsrc - 1);
12004 +       cpg.bdst = err;
12005 +       if (unlikely(err < 0)) {
12006 +               err = 0;        /* there is no upper writable branch */
12007 +               goto out_dgrade;
12008 +       }
12009 +       AuDbg("bsrc %d, bdst %d\n", cpg.bsrc, cpg.bdst);
12010 +
12011 +       /* do not respect the coo attrib for the target branch */
12012 +       err = au_cpup_dirs(dentry, cpg.bdst);
12013 +       if (unlikely(err))
12014 +               goto out_dgrade;
12015 +
12016 +       di_downgrade_lock(parent, AuLock_IR);
12017 +       udba = au_opt_udba(sb);
12018 +       err = au_pin(&pin, dentry, cpg.bdst, udba,
12019 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12020 +       if (unlikely(err))
12021 +               goto out_parent;
12022 +
12023 +       err = au_sio_cpup_simple(&cpg);
12024 +       au_unpin(&pin);
12025 +       if (unlikely(err))
12026 +               goto out_parent;
12027 +       if (!(cmoo & AuBrWAttr_MOO))
12028 +               goto out_parent; /* success */
12029 +
12030 +       err = au_pin(&pin, dentry, cpg.bsrc, udba,
12031 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12032 +       if (unlikely(err))
12033 +               goto out_parent;
12034 +
12035 +       h_path.mnt = au_br_mnt(br);
12036 +       h_path.dentry = au_h_dptr(dentry, cpg.bsrc);
12037 +       hdir = au_hi(parent->d_inode, cpg.bsrc);
12038 +       delegated = NULL;
12039 +       err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated, /*force*/1);
12040 +       au_unpin(&pin);
12041 +       /* todo: keep h_dentry or not? */
12042 +       if (unlikely(err == -EWOULDBLOCK)) {
12043 +               pr_warn("cannot retry for NFSv4 delegation"
12044 +                       " for an internal unlink\n");
12045 +               iput(delegated);
12046 +       }
12047 +       if (unlikely(err)) {
12048 +               pr_err("unlink %pd after coo failed (%d), ignored\n",
12049 +                      dentry, err);
12050 +               err = 0;
12051 +       }
12052 +       goto out_parent; /* success */
12053 +
12054 +out_dgrade:
12055 +       di_downgrade_lock(parent, AuLock_IR);
12056 +out_parent:
12057 +       di_read_unlock(parent, AuLock_IR);
12058 +       dput(parent);
12059 +out:
12060 +       AuTraceErr(err);
12061 +       return err;
12062 +}
12063 +
12064 +int au_do_open(struct file *file, int (*open)(struct file *file, int flags),
12065 +              struct au_fidir *fidir)
12066 +{
12067 +       int err;
12068 +       struct dentry *dentry;
12069 +       struct au_finfo *finfo;
12070 +
12071 +       err = au_finfo_init(file, fidir);
12072 +       if (unlikely(err))
12073 +               goto out;
12074 +
12075 +       dentry = file->f_path.dentry;
12076 +       di_write_lock_child(dentry);
12077 +       err = au_cmoo(dentry);
12078 +       di_downgrade_lock(dentry, AuLock_IR);
12079 +       if (!err)
12080 +               err = open(file, vfsub_file_flags(file));
12081 +       di_read_unlock(dentry, AuLock_IR);
12082 +
12083 +       finfo = au_fi(file);
12084 +       if (!err) {
12085 +               finfo->fi_file = file;
12086 +               au_sphl_add(&finfo->fi_hlist,
12087 +                           &au_sbi(file->f_path.dentry->d_sb)->si_files);
12088 +       }
12089 +       fi_write_unlock(file);
12090 +       if (unlikely(err)) {
12091 +               finfo->fi_hdir = NULL;
12092 +               au_finfo_fin(file);
12093 +       }
12094 +
12095 +out:
12096 +       return err;
12097 +}
12098 +
12099 +int au_reopen_nondir(struct file *file)
12100 +{
12101 +       int err;
12102 +       aufs_bindex_t bstart;
12103 +       struct dentry *dentry;
12104 +       struct file *h_file, *h_file_tmp;
12105 +
12106 +       dentry = file->f_path.dentry;
12107 +       bstart = au_dbstart(dentry);
12108 +       h_file_tmp = NULL;
12109 +       if (au_fbstart(file) == bstart) {
12110 +               h_file = au_hf_top(file);
12111 +               if (file->f_mode == h_file->f_mode)
12112 +                       return 0; /* success */
12113 +               h_file_tmp = h_file;
12114 +               get_file(h_file_tmp);
12115 +               au_set_h_fptr(file, bstart, NULL);
12116 +       }
12117 +       AuDebugOn(au_fi(file)->fi_hdir);
12118 +       /*
12119 +        * it can happen
12120 +        * file exists on both of rw and ro
12121 +        * open --> dbstart and fbstart are both 0
12122 +        * prepend a branch as rw, "rw" become ro
12123 +        * remove rw/file
12124 +        * delete the top branch, "rw" becomes rw again
12125 +        *      --> dbstart is 1, fbstart is still 0
12126 +        * write --> fbstart is 0 but dbstart is 1
12127 +        */
12128 +       /* AuDebugOn(au_fbstart(file) < bstart); */
12129 +
12130 +       h_file = au_h_open(dentry, bstart, vfsub_file_flags(file) & ~O_TRUNC,
12131 +                          file, /*force_wr*/0);
12132 +       err = PTR_ERR(h_file);
12133 +       if (IS_ERR(h_file)) {
12134 +               if (h_file_tmp) {
12135 +                       atomic_inc(&au_sbr(dentry->d_sb, bstart)->br_count);
12136 +                       au_set_h_fptr(file, bstart, h_file_tmp);
12137 +                       h_file_tmp = NULL;
12138 +               }
12139 +               goto out; /* todo: close all? */
12140 +       }
12141 +
12142 +       err = 0;
12143 +       au_set_fbstart(file, bstart);
12144 +       au_set_h_fptr(file, bstart, h_file);
12145 +       au_update_figen(file);
12146 +       /* todo: necessary? */
12147 +       /* file->f_ra = h_file->f_ra; */
12148 +
12149 +out:
12150 +       if (h_file_tmp)
12151 +               fput(h_file_tmp);
12152 +       return err;
12153 +}
12154 +
12155 +/* ---------------------------------------------------------------------- */
12156 +
12157 +static int au_reopen_wh(struct file *file, aufs_bindex_t btgt,
12158 +                       struct dentry *hi_wh)
12159 +{
12160 +       int err;
12161 +       aufs_bindex_t bstart;
12162 +       struct au_dinfo *dinfo;
12163 +       struct dentry *h_dentry;
12164 +       struct au_hdentry *hdp;
12165 +
12166 +       dinfo = au_di(file->f_path.dentry);
12167 +       AuRwMustWriteLock(&dinfo->di_rwsem);
12168 +
12169 +       bstart = dinfo->di_bstart;
12170 +       dinfo->di_bstart = btgt;
12171 +       hdp = dinfo->di_hdentry;
12172 +       h_dentry = hdp[0 + btgt].hd_dentry;
12173 +       hdp[0 + btgt].hd_dentry = hi_wh;
12174 +       err = au_reopen_nondir(file);
12175 +       hdp[0 + btgt].hd_dentry = h_dentry;
12176 +       dinfo->di_bstart = bstart;
12177 +
12178 +       return err;
12179 +}
12180 +
12181 +static int au_ready_to_write_wh(struct file *file, loff_t len,
12182 +                               aufs_bindex_t bcpup, struct au_pin *pin)
12183 +{
12184 +       int err;
12185 +       struct inode *inode, *h_inode;
12186 +       struct dentry *h_dentry, *hi_wh;
12187 +       struct au_cp_generic cpg = {
12188 +               .dentry = file->f_path.dentry,
12189 +               .bdst   = bcpup,
12190 +               .bsrc   = -1,
12191 +               .len    = len,
12192 +               .pin    = pin
12193 +       };
12194 +
12195 +       au_update_dbstart(cpg.dentry);
12196 +       inode = cpg.dentry->d_inode;
12197 +       h_inode = NULL;
12198 +       if (au_dbstart(cpg.dentry) <= bcpup
12199 +           && au_dbend(cpg.dentry) >= bcpup) {
12200 +               h_dentry = au_h_dptr(cpg.dentry, bcpup);
12201 +               if (h_dentry)
12202 +                       h_inode = h_dentry->d_inode;
12203 +       }
12204 +       hi_wh = au_hi_wh(inode, bcpup);
12205 +       if (!hi_wh && !h_inode)
12206 +               err = au_sio_cpup_wh(&cpg, file);
12207 +       else
12208 +               /* already copied-up after unlink */
12209 +               err = au_reopen_wh(file, bcpup, hi_wh);
12210 +
12211 +       if (!err
12212 +           && (inode->i_nlink > 1
12213 +               || (inode->i_state & I_LINKABLE))
12214 +           && au_opt_test(au_mntflags(cpg.dentry->d_sb), PLINK))
12215 +               au_plink_append(inode, bcpup, au_h_dptr(cpg.dentry, bcpup));
12216 +
12217 +       return err;
12218 +}
12219 +
12220 +/*
12221 + * prepare the @file for writing.
12222 + */
12223 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin)
12224 +{
12225 +       int err;
12226 +       aufs_bindex_t dbstart;
12227 +       struct dentry *parent;
12228 +       struct inode *inode;
12229 +       struct super_block *sb;
12230 +       struct file *h_file;
12231 +       struct au_cp_generic cpg = {
12232 +               .dentry = file->f_path.dentry,
12233 +               .bdst   = -1,
12234 +               .bsrc   = -1,
12235 +               .len    = len,
12236 +               .pin    = pin,
12237 +               .flags  = AuCpup_DTIME
12238 +       };
12239 +
12240 +       sb = cpg.dentry->d_sb;
12241 +       inode = cpg.dentry->d_inode;
12242 +       cpg.bsrc = au_fbstart(file);
12243 +       err = au_test_ro(sb, cpg.bsrc, inode);
12244 +       if (!err && (au_hf_top(file)->f_mode & FMODE_WRITE)) {
12245 +               err = au_pin(pin, cpg.dentry, cpg.bsrc, AuOpt_UDBA_NONE,
12246 +                            /*flags*/0);
12247 +               goto out;
12248 +       }
12249 +
12250 +       /* need to cpup or reopen */
12251 +       parent = dget_parent(cpg.dentry);
12252 +       di_write_lock_parent(parent);
12253 +       err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12254 +       cpg.bdst = err;
12255 +       if (unlikely(err < 0))
12256 +               goto out_dgrade;
12257 +       err = 0;
12258 +
12259 +       if (!d_unhashed(cpg.dentry) && !au_h_dptr(parent, cpg.bdst)) {
12260 +               err = au_cpup_dirs(cpg.dentry, cpg.bdst);
12261 +               if (unlikely(err))
12262 +                       goto out_dgrade;
12263 +       }
12264 +
12265 +       err = au_pin(pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12266 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12267 +       if (unlikely(err))
12268 +               goto out_dgrade;
12269 +
12270 +       dbstart = au_dbstart(cpg.dentry);
12271 +       if (dbstart <= cpg.bdst)
12272 +               cpg.bsrc = cpg.bdst;
12273 +
12274 +       if (dbstart <= cpg.bdst         /* just reopen */
12275 +           || !d_unhashed(cpg.dentry)  /* copyup and reopen */
12276 +               ) {
12277 +               h_file = au_h_open_pre(cpg.dentry, cpg.bsrc, /*force_wr*/0);
12278 +               if (IS_ERR(h_file))
12279 +                       err = PTR_ERR(h_file);
12280 +               else {
12281 +                       di_downgrade_lock(parent, AuLock_IR);
12282 +                       if (dbstart > cpg.bdst)
12283 +                               err = au_sio_cpup_simple(&cpg);
12284 +                       if (!err)
12285 +                               err = au_reopen_nondir(file);
12286 +                       au_h_open_post(cpg.dentry, cpg.bsrc, h_file);
12287 +               }
12288 +       } else {                        /* copyup as wh and reopen */
12289 +               /*
12290 +                * since writable hfsplus branch is not supported,
12291 +                * h_open_pre/post() are unnecessary.
12292 +                */
12293 +               err = au_ready_to_write_wh(file, len, cpg.bdst, pin);
12294 +               di_downgrade_lock(parent, AuLock_IR);
12295 +       }
12296 +
12297 +       if (!err) {
12298 +               au_pin_set_parent_lflag(pin, /*lflag*/0);
12299 +               goto out_dput; /* success */
12300 +       }
12301 +       au_unpin(pin);
12302 +       goto out_unlock;
12303 +
12304 +out_dgrade:
12305 +       di_downgrade_lock(parent, AuLock_IR);
12306 +out_unlock:
12307 +       di_read_unlock(parent, AuLock_IR);
12308 +out_dput:
12309 +       dput(parent);
12310 +out:
12311 +       return err;
12312 +}
12313 +
12314 +/* ---------------------------------------------------------------------- */
12315 +
12316 +int au_do_flush(struct file *file, fl_owner_t id,
12317 +               int (*flush)(struct file *file, fl_owner_t id))
12318 +{
12319 +       int err;
12320 +       struct super_block *sb;
12321 +       struct inode *inode;
12322 +
12323 +       inode = file_inode(file);
12324 +       sb = inode->i_sb;
12325 +       si_noflush_read_lock(sb);
12326 +       fi_read_lock(file);
12327 +       ii_read_lock_child(inode);
12328 +
12329 +       err = flush(file, id);
12330 +       au_cpup_attr_timesizes(inode);
12331 +
12332 +       ii_read_unlock(inode);
12333 +       fi_read_unlock(file);
12334 +       si_read_unlock(sb);
12335 +       return err;
12336 +}
12337 +
12338 +/* ---------------------------------------------------------------------- */
12339 +
12340 +static int au_file_refresh_by_inode(struct file *file, int *need_reopen)
12341 +{
12342 +       int err;
12343 +       struct au_pin pin;
12344 +       struct au_finfo *finfo;
12345 +       struct dentry *parent, *hi_wh;
12346 +       struct inode *inode;
12347 +       struct super_block *sb;
12348 +       struct au_cp_generic cpg = {
12349 +               .dentry = file->f_path.dentry,
12350 +               .bdst   = -1,
12351 +               .bsrc   = -1,
12352 +               .len    = -1,
12353 +               .pin    = &pin,
12354 +               .flags  = AuCpup_DTIME
12355 +       };
12356 +
12357 +       FiMustWriteLock(file);
12358 +
12359 +       err = 0;
12360 +       finfo = au_fi(file);
12361 +       sb = cpg.dentry->d_sb;
12362 +       inode = cpg.dentry->d_inode;
12363 +       cpg.bdst = au_ibstart(inode);
12364 +       if (cpg.bdst == finfo->fi_btop || IS_ROOT(cpg.dentry))
12365 +               goto out;
12366 +
12367 +       parent = dget_parent(cpg.dentry);
12368 +       if (au_test_ro(sb, cpg.bdst, inode)) {
12369 +               di_read_lock_parent(parent, !AuLock_IR);
12370 +               err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
12371 +               cpg.bdst = err;
12372 +               di_read_unlock(parent, !AuLock_IR);
12373 +               if (unlikely(err < 0))
12374 +                       goto out_parent;
12375 +               err = 0;
12376 +       }
12377 +
12378 +       di_read_lock_parent(parent, AuLock_IR);
12379 +       hi_wh = au_hi_wh(inode, cpg.bdst);
12380 +       if (!S_ISDIR(inode->i_mode)
12381 +           && au_opt_test(au_mntflags(sb), PLINK)
12382 +           && au_plink_test(inode)
12383 +           && !d_unhashed(cpg.dentry)
12384 +           && cpg.bdst < au_dbstart(cpg.dentry)) {
12385 +               err = au_test_and_cpup_dirs(cpg.dentry, cpg.bdst);
12386 +               if (unlikely(err))
12387 +                       goto out_unlock;
12388 +
12389 +               /* always superio. */
12390 +               err = au_pin(&pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
12391 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
12392 +               if (!err) {
12393 +                       err = au_sio_cpup_simple(&cpg);
12394 +                       au_unpin(&pin);
12395 +               }
12396 +       } else if (hi_wh) {
12397 +               /* already copied-up after unlink */
12398 +               err = au_reopen_wh(file, cpg.bdst, hi_wh);
12399 +               *need_reopen = 0;
12400 +       }
12401 +
12402 +out_unlock:
12403 +       di_read_unlock(parent, AuLock_IR);
12404 +out_parent:
12405 +       dput(parent);
12406 +out:
12407 +       return err;
12408 +}
12409 +
12410 +static void au_do_refresh_dir(struct file *file)
12411 +{
12412 +       aufs_bindex_t bindex, bend, new_bindex, brid;
12413 +       struct au_hfile *p, tmp, *q;
12414 +       struct au_finfo *finfo;
12415 +       struct super_block *sb;
12416 +       struct au_fidir *fidir;
12417 +
12418 +       FiMustWriteLock(file);
12419 +
12420 +       sb = file->f_path.dentry->d_sb;
12421 +       finfo = au_fi(file);
12422 +       fidir = finfo->fi_hdir;
12423 +       AuDebugOn(!fidir);
12424 +       p = fidir->fd_hfile + finfo->fi_btop;
12425 +       brid = p->hf_br->br_id;
12426 +       bend = fidir->fd_bbot;
12427 +       for (bindex = finfo->fi_btop; bindex <= bend; bindex++, p++) {
12428 +               if (!p->hf_file)
12429 +                       continue;
12430 +
12431 +               new_bindex = au_br_index(sb, p->hf_br->br_id);
12432 +               if (new_bindex == bindex)
12433 +                       continue;
12434 +               if (new_bindex < 0) {
12435 +                       au_set_h_fptr(file, bindex, NULL);
12436 +                       continue;
12437 +               }
12438 +
12439 +               /* swap two lower inode, and loop again */
12440 +               q = fidir->fd_hfile + new_bindex;
12441 +               tmp = *q;
12442 +               *q = *p;
12443 +               *p = tmp;
12444 +               if (tmp.hf_file) {
12445 +                       bindex--;
12446 +                       p--;
12447 +               }
12448 +       }
12449 +
12450 +       p = fidir->fd_hfile;
12451 +       if (!au_test_mmapped(file) && !d_unlinked(file->f_path.dentry)) {
12452 +               bend = au_sbend(sb);
12453 +               for (finfo->fi_btop = 0; finfo->fi_btop <= bend;
12454 +                    finfo->fi_btop++, p++)
12455 +                       if (p->hf_file) {
12456 +                               if (file_inode(p->hf_file))
12457 +                                       break;
12458 +                               au_hfput(p, file);
12459 +                       }
12460 +       } else {
12461 +               bend = au_br_index(sb, brid);
12462 +               for (finfo->fi_btop = 0; finfo->fi_btop < bend;
12463 +                    finfo->fi_btop++, p++)
12464 +                       if (p->hf_file)
12465 +                               au_hfput(p, file);
12466 +               bend = au_sbend(sb);
12467 +       }
12468 +
12469 +       p = fidir->fd_hfile + bend;
12470 +       for (fidir->fd_bbot = bend; fidir->fd_bbot >= finfo->fi_btop;
12471 +            fidir->fd_bbot--, p--)
12472 +               if (p->hf_file) {
12473 +                       if (file_inode(p->hf_file))
12474 +                               break;
12475 +                       au_hfput(p, file);
12476 +               }
12477 +       AuDebugOn(fidir->fd_bbot < finfo->fi_btop);
12478 +}
12479 +
12480 +/*
12481 + * after branch manipulating, refresh the file.
12482 + */
12483 +static int refresh_file(struct file *file, int (*reopen)(struct file *file))
12484 +{
12485 +       int err, need_reopen;
12486 +       aufs_bindex_t bend, bindex;
12487 +       struct dentry *dentry;
12488 +       struct au_finfo *finfo;
12489 +       struct au_hfile *hfile;
12490 +
12491 +       dentry = file->f_path.dentry;
12492 +       finfo = au_fi(file);
12493 +       if (!finfo->fi_hdir) {
12494 +               hfile = &finfo->fi_htop;
12495 +               AuDebugOn(!hfile->hf_file);
12496 +               bindex = au_br_index(dentry->d_sb, hfile->hf_br->br_id);
12497 +               AuDebugOn(bindex < 0);
12498 +               if (bindex != finfo->fi_btop)
12499 +                       au_set_fbstart(file, bindex);
12500 +       } else {
12501 +               err = au_fidir_realloc(finfo, au_sbend(dentry->d_sb) + 1);
12502 +               if (unlikely(err))
12503 +                       goto out;
12504 +               au_do_refresh_dir(file);
12505 +       }
12506 +
12507 +       err = 0;
12508 +       need_reopen = 1;
12509 +       if (!au_test_mmapped(file))
12510 +               err = au_file_refresh_by_inode(file, &need_reopen);
12511 +       if (!err && need_reopen && !d_unlinked(dentry))
12512 +               err = reopen(file);
12513 +       if (!err) {
12514 +               au_update_figen(file);
12515 +               goto out; /* success */
12516 +       }
12517 +
12518 +       /* error, close all lower files */
12519 +       if (finfo->fi_hdir) {
12520 +               bend = au_fbend_dir(file);
12521 +               for (bindex = au_fbstart(file); bindex <= bend; bindex++)
12522 +                       au_set_h_fptr(file, bindex, NULL);
12523 +       }
12524 +
12525 +out:
12526 +       return err;
12527 +}
12528 +
12529 +/* common function to regular file and dir */
12530 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
12531 +                         int wlock)
12532 +{
12533 +       int err;
12534 +       unsigned int sigen, figen;
12535 +       aufs_bindex_t bstart;
12536 +       unsigned char pseudo_link;
12537 +       struct dentry *dentry;
12538 +       struct inode *inode;
12539 +
12540 +       err = 0;
12541 +       dentry = file->f_path.dentry;
12542 +       inode = dentry->d_inode;
12543 +       sigen = au_sigen(dentry->d_sb);
12544 +       fi_write_lock(file);
12545 +       figen = au_figen(file);
12546 +       di_write_lock_child(dentry);
12547 +       bstart = au_dbstart(dentry);
12548 +       pseudo_link = (bstart != au_ibstart(inode));
12549 +       if (sigen == figen && !pseudo_link && au_fbstart(file) == bstart) {
12550 +               if (!wlock) {
12551 +                       di_downgrade_lock(dentry, AuLock_IR);
12552 +                       fi_downgrade_lock(file);
12553 +               }
12554 +               goto out; /* success */
12555 +       }
12556 +
12557 +       AuDbg("sigen %d, figen %d\n", sigen, figen);
12558 +       if (au_digen_test(dentry, sigen)) {
12559 +               err = au_reval_dpath(dentry, sigen);
12560 +               AuDebugOn(!err && au_digen_test(dentry, sigen));
12561 +       }
12562 +
12563 +       if (!err)
12564 +               err = refresh_file(file, reopen);
12565 +       if (!err) {
12566 +               if (!wlock) {
12567 +                       di_downgrade_lock(dentry, AuLock_IR);
12568 +                       fi_downgrade_lock(file);
12569 +               }
12570 +       } else {
12571 +               di_write_unlock(dentry);
12572 +               fi_write_unlock(file);
12573 +       }
12574 +
12575 +out:
12576 +       return err;
12577 +}
12578 +
12579 +/* ---------------------------------------------------------------------- */
12580 +
12581 +/* cf. aufs_nopage() */
12582 +/* for madvise(2) */
12583 +static int aufs_readpage(struct file *file __maybe_unused, struct page *page)
12584 +{
12585 +       unlock_page(page);
12586 +       return 0;
12587 +}
12588 +
12589 +/* it will never be called, but necessary to support O_DIRECT */
12590 +static ssize_t aufs_direct_IO(int rw, struct kiocb *iocb,
12591 +                             struct iov_iter *iter, loff_t offset)
12592 +{ BUG(); return 0; }
12593 +
12594 +/*
12595 + * it will never be called, but madvise and fadvise behaves differently
12596 + * when get_xip_mem is defined
12597 + */
12598 +static int aufs_get_xip_mem(struct address_space *mapping, pgoff_t pgoff,
12599 +                           int create, void **kmem, unsigned long *pfn)
12600 +{ BUG(); return 0; }
12601 +
12602 +/* they will never be called. */
12603 +#ifdef CONFIG_AUFS_DEBUG
12604 +static int aufs_write_begin(struct file *file, struct address_space *mapping,
12605 +                           loff_t pos, unsigned len, unsigned flags,
12606 +                           struct page **pagep, void **fsdata)
12607 +{ AuUnsupport(); return 0; }
12608 +static int aufs_write_end(struct file *file, struct address_space *mapping,
12609 +                         loff_t pos, unsigned len, unsigned copied,
12610 +                         struct page *page, void *fsdata)
12611 +{ AuUnsupport(); return 0; }
12612 +static int aufs_writepage(struct page *page, struct writeback_control *wbc)
12613 +{ AuUnsupport(); return 0; }
12614 +
12615 +static int aufs_set_page_dirty(struct page *page)
12616 +{ AuUnsupport(); return 0; }
12617 +static void aufs_invalidatepage(struct page *page, unsigned int offset,
12618 +                               unsigned int length)
12619 +{ AuUnsupport(); }
12620 +static int aufs_releasepage(struct page *page, gfp_t gfp)
12621 +{ AuUnsupport(); return 0; }
12622 +static int aufs_migratepage(struct address_space *mapping, struct page *newpage,
12623 +                           struct page *page, enum migrate_mode mode)
12624 +{ AuUnsupport(); return 0; }
12625 +static int aufs_launder_page(struct page *page)
12626 +{ AuUnsupport(); return 0; }
12627 +static int aufs_is_partially_uptodate(struct page *page,
12628 +                                     unsigned long from,
12629 +                                     unsigned long count)
12630 +{ AuUnsupport(); return 0; }
12631 +static void aufs_is_dirty_writeback(struct page *page, bool *dirty,
12632 +                                   bool *writeback)
12633 +{ AuUnsupport(); }
12634 +static int aufs_error_remove_page(struct address_space *mapping,
12635 +                                 struct page *page)
12636 +{ AuUnsupport(); return 0; }
12637 +static int aufs_swap_activate(struct swap_info_struct *sis, struct file *file,
12638 +                             sector_t *span)
12639 +{ AuUnsupport(); return 0; }
12640 +static void aufs_swap_deactivate(struct file *file)
12641 +{ AuUnsupport(); }
12642 +#endif /* CONFIG_AUFS_DEBUG */
12643 +
12644 +const struct address_space_operations aufs_aop = {
12645 +       .readpage               = aufs_readpage,
12646 +       .direct_IO              = aufs_direct_IO,
12647 +       .get_xip_mem            = aufs_get_xip_mem,
12648 +#ifdef CONFIG_AUFS_DEBUG
12649 +       .writepage              = aufs_writepage,
12650 +       /* no writepages, because of writepage */
12651 +       .set_page_dirty         = aufs_set_page_dirty,
12652 +       /* no readpages, because of readpage */
12653 +       .write_begin            = aufs_write_begin,
12654 +       .write_end              = aufs_write_end,
12655 +       /* no bmap, no block device */
12656 +       .invalidatepage         = aufs_invalidatepage,
12657 +       .releasepage            = aufs_releasepage,
12658 +       .migratepage            = aufs_migratepage,
12659 +       .launder_page           = aufs_launder_page,
12660 +       .is_partially_uptodate  = aufs_is_partially_uptodate,
12661 +       .is_dirty_writeback     = aufs_is_dirty_writeback,
12662 +       .error_remove_page      = aufs_error_remove_page,
12663 +       .swap_activate          = aufs_swap_activate,
12664 +       .swap_deactivate        = aufs_swap_deactivate
12665 +#endif /* CONFIG_AUFS_DEBUG */
12666 +};
12667 diff -urN /usr/share/empty/fs/aufs/file.h linux/fs/aufs/file.h
12668 --- /usr/share/empty/fs/aufs/file.h     1970-01-01 01:00:00.000000000 +0100
12669 +++ linux/fs/aufs/file.h        2015-03-27 21:56:35.463461668 +0100
12670 @@ -0,0 +1,284 @@
12671 +/*
12672 + * Copyright (C) 2005-2015 Junjiro R. Okajima
12673 + *
12674 + * This program, aufs is free software; you can redistribute it and/or modify
12675 + * it under the terms of the GNU General Public License as published by
12676 + * the Free Software Foundation; either version 2 of the License, or
12677 + * (at your option) any later version.
12678 + *
12679 + * This program is distributed in the hope that it will be useful,
12680 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12681 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12682 + * GNU General Public License for more details.
12683 + *
12684 + * You should have received a copy of the GNU General Public License
12685 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12686 + */
12687 +
12688 +/*
12689 + * file operations
12690 + */
12691 +
12692 +#ifndef __AUFS_FILE_H__
12693 +#define __AUFS_FILE_H__
12694 +
12695 +#ifdef __KERNEL__
12696 +
12697 +#include <linux/file.h>
12698 +#include <linux/fs.h>
12699 +#include <linux/poll.h>
12700 +#include "rwsem.h"
12701 +
12702 +struct au_branch;
12703 +struct au_hfile {
12704 +       struct file             *hf_file;
12705 +       struct au_branch        *hf_br;
12706 +};
12707 +
12708 +struct au_vdir;
12709 +struct au_fidir {
12710 +       aufs_bindex_t           fd_bbot;
12711 +       aufs_bindex_t           fd_nent;
12712 +       struct au_vdir          *fd_vdir_cache;
12713 +       struct au_hfile         fd_hfile[];
12714 +};
12715 +
12716 +static inline int au_fidir_sz(int nent)
12717 +{
12718 +       AuDebugOn(nent < 0);
12719 +       return sizeof(struct au_fidir) + sizeof(struct au_hfile) * nent;
12720 +}
12721 +
12722 +struct au_finfo {
12723 +       atomic_t                fi_generation;
12724 +
12725 +       struct au_rwsem         fi_rwsem;
12726 +       aufs_bindex_t           fi_btop;
12727 +
12728 +       /* do not union them */
12729 +       struct {                                /* for non-dir */
12730 +               struct au_hfile                 fi_htop;
12731 +               atomic_t                        fi_mmapped;
12732 +       };
12733 +       struct au_fidir         *fi_hdir;       /* for dir only */
12734 +
12735 +       struct hlist_node       fi_hlist;
12736 +       struct file             *fi_file;       /* very ugly */
12737 +} ____cacheline_aligned_in_smp;
12738 +
12739 +/* ---------------------------------------------------------------------- */
12740 +
12741 +/* file.c */
12742 +extern const struct address_space_operations aufs_aop;
12743 +unsigned int au_file_roflags(unsigned int flags);
12744 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
12745 +                      struct file *file, int force_wr);
12746 +int au_do_open(struct file *file, int (*open)(struct file *file, int flags),
12747 +              struct au_fidir *fidir);
12748 +int au_reopen_nondir(struct file *file);
12749 +struct au_pin;
12750 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin);
12751 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
12752 +                         int wlock);
12753 +int au_do_flush(struct file *file, fl_owner_t id,
12754 +               int (*flush)(struct file *file, fl_owner_t id));
12755 +
12756 +/* poll.c */
12757 +#ifdef CONFIG_AUFS_POLL
12758 +unsigned int aufs_poll(struct file *file, poll_table *wait);
12759 +#endif
12760 +
12761 +#ifdef CONFIG_AUFS_BR_HFSPLUS
12762 +/* hfsplus.c */
12763 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
12764 +                          int force_wr);
12765 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
12766 +                   struct file *h_file);
12767 +#else
12768 +AuStub(struct file *, au_h_open_pre, return NULL, struct dentry *dentry,
12769 +       aufs_bindex_t bindex, int force_wr)
12770 +AuStubVoid(au_h_open_post, struct dentry *dentry, aufs_bindex_t bindex,
12771 +          struct file *h_file);
12772 +#endif
12773 +
12774 +/* f_op.c */
12775 +extern const struct file_operations aufs_file_fop;
12776 +int au_do_open_nondir(struct file *file, int flags);
12777 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file);
12778 +
12779 +/* finfo.c */
12780 +void au_hfput(struct au_hfile *hf, struct file *file);
12781 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex,
12782 +                  struct file *h_file);
12783 +
12784 +void au_update_figen(struct file *file);
12785 +struct au_fidir *au_fidir_alloc(struct super_block *sb);
12786 +int au_fidir_realloc(struct au_finfo *finfo, int nbr);
12787 +
12788 +void au_fi_init_once(void *_fi);
12789 +void au_finfo_fin(struct file *file);
12790 +int au_finfo_init(struct file *file, struct au_fidir *fidir);
12791 +
12792 +/* ioctl.c */
12793 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg);
12794 +#ifdef CONFIG_COMPAT
12795 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
12796 +                          unsigned long arg);
12797 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
12798 +                             unsigned long arg);
12799 +#endif
12800 +
12801 +/* ---------------------------------------------------------------------- */
12802 +
12803 +static inline struct au_finfo *au_fi(struct file *file)
12804 +{
12805 +       return file->private_data;
12806 +}
12807 +
12808 +/* ---------------------------------------------------------------------- */
12809 +
12810 +/*
12811 + * fi_read_lock, fi_write_lock,
12812 + * fi_read_unlock, fi_write_unlock, fi_downgrade_lock
12813 + */
12814 +AuSimpleRwsemFuncs(fi, struct file *f, &au_fi(f)->fi_rwsem);
12815 +
12816 +#define FiMustNoWaiters(f)     AuRwMustNoWaiters(&au_fi(f)->fi_rwsem)
12817 +#define FiMustAnyLock(f)       AuRwMustAnyLock(&au_fi(f)->fi_rwsem)
12818 +#define FiMustWriteLock(f)     AuRwMustWriteLock(&au_fi(f)->fi_rwsem)
12819 +
12820 +/* ---------------------------------------------------------------------- */
12821 +
12822 +/* todo: hard/soft set? */
12823 +static inline aufs_bindex_t au_fbstart(struct file *file)
12824 +{
12825 +       FiMustAnyLock(file);
12826 +       return au_fi(file)->fi_btop;
12827 +}
12828 +
12829 +static inline aufs_bindex_t au_fbend_dir(struct file *file)
12830 +{
12831 +       FiMustAnyLock(file);
12832 +       AuDebugOn(!au_fi(file)->fi_hdir);
12833 +       return au_fi(file)->fi_hdir->fd_bbot;
12834 +}
12835 +
12836 +static inline struct au_vdir *au_fvdir_cache(struct file *file)
12837 +{
12838 +       FiMustAnyLock(file);
12839 +       AuDebugOn(!au_fi(file)->fi_hdir);
12840 +       return au_fi(file)->fi_hdir->fd_vdir_cache;
12841 +}
12842 +
12843 +static inline void au_set_fbstart(struct file *file, aufs_bindex_t bindex)
12844 +{
12845 +       FiMustWriteLock(file);
12846 +       au_fi(file)->fi_btop = bindex;
12847 +}
12848 +
12849 +static inline void au_set_fbend_dir(struct file *file, aufs_bindex_t bindex)
12850 +{
12851 +       FiMustWriteLock(file);
12852 +       AuDebugOn(!au_fi(file)->fi_hdir);
12853 +       au_fi(file)->fi_hdir->fd_bbot = bindex;
12854 +}
12855 +
12856 +static inline void au_set_fvdir_cache(struct file *file,
12857 +                                     struct au_vdir *vdir_cache)
12858 +{
12859 +       FiMustWriteLock(file);
12860 +       AuDebugOn(!au_fi(file)->fi_hdir);
12861 +       au_fi(file)->fi_hdir->fd_vdir_cache = vdir_cache;
12862 +}
12863 +
12864 +static inline struct file *au_hf_top(struct file *file)
12865 +{
12866 +       FiMustAnyLock(file);
12867 +       AuDebugOn(au_fi(file)->fi_hdir);
12868 +       return au_fi(file)->fi_htop.hf_file;
12869 +}
12870 +
12871 +static inline struct file *au_hf_dir(struct file *file, aufs_bindex_t bindex)
12872 +{
12873 +       FiMustAnyLock(file);
12874 +       AuDebugOn(!au_fi(file)->fi_hdir);
12875 +       return au_fi(file)->fi_hdir->fd_hfile[0 + bindex].hf_file;
12876 +}
12877 +
12878 +/* todo: memory barrier? */
12879 +static inline unsigned int au_figen(struct file *f)
12880 +{
12881 +       return atomic_read(&au_fi(f)->fi_generation);
12882 +}
12883 +
12884 +static inline void au_set_mmapped(struct file *f)
12885 +{
12886 +       if (atomic_inc_return(&au_fi(f)->fi_mmapped))
12887 +               return;
12888 +       pr_warn("fi_mmapped wrapped around\n");
12889 +       while (!atomic_inc_return(&au_fi(f)->fi_mmapped))
12890 +               ;
12891 +}
12892 +
12893 +static inline void au_unset_mmapped(struct file *f)
12894 +{
12895 +       atomic_dec(&au_fi(f)->fi_mmapped);
12896 +}
12897 +
12898 +static inline int au_test_mmapped(struct file *f)
12899 +{
12900 +       return atomic_read(&au_fi(f)->fi_mmapped);
12901 +}
12902 +
12903 +/* customize vma->vm_file */
12904 +
12905 +static inline void au_do_vm_file_reset(struct vm_area_struct *vma,
12906 +                                      struct file *file)
12907 +{
12908 +       struct file *f;
12909 +
12910 +       f = vma->vm_file;
12911 +       get_file(file);
12912 +       vma->vm_file = file;
12913 +       fput(f);
12914 +}
12915 +
12916 +#ifdef CONFIG_MMU
12917 +#define AuDbgVmRegion(file, vma) do {} while (0)
12918 +
12919 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
12920 +                                   struct file *file)
12921 +{
12922 +       au_do_vm_file_reset(vma, file);
12923 +}
12924 +#else
12925 +#define AuDbgVmRegion(file, vma) \
12926 +       AuDebugOn((vma)->vm_region && (vma)->vm_region->vm_file != (file))
12927 +
12928 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
12929 +                                   struct file *file)
12930 +{
12931 +       struct file *f;
12932 +
12933 +       au_do_vm_file_reset(vma, file);
12934 +       f = vma->vm_region->vm_file;
12935 +       get_file(file);
12936 +       vma->vm_region->vm_file = file;
12937 +       fput(f);
12938 +}
12939 +#endif /* CONFIG_MMU */
12940 +
12941 +/* handle vma->vm_prfile */
12942 +static inline void au_vm_prfile_set(struct vm_area_struct *vma,
12943 +                                   struct file *file)
12944 +{
12945 +       get_file(file);
12946 +       vma->vm_prfile = file;
12947 +#ifndef CONFIG_MMU
12948 +       get_file(file);
12949 +       vma->vm_region->vm_prfile = file;
12950 +#endif
12951 +}
12952 +
12953 +#endif /* __KERNEL__ */
12954 +#endif /* __AUFS_FILE_H__ */
12955 diff -urN /usr/share/empty/fs/aufs/finfo.c linux/fs/aufs/finfo.c
12956 --- /usr/share/empty/fs/aufs/finfo.c    1970-01-01 01:00:00.000000000 +0100
12957 +++ linux/fs/aufs/finfo.c       2015-03-27 21:56:35.463461668 +0100
12958 @@ -0,0 +1,156 @@
12959 +/*
12960 + * Copyright (C) 2005-2015 Junjiro R. Okajima
12961 + *
12962 + * This program, aufs is free software; you can redistribute it and/or modify
12963 + * it under the terms of the GNU General Public License as published by
12964 + * the Free Software Foundation; either version 2 of the License, or
12965 + * (at your option) any later version.
12966 + *
12967 + * This program is distributed in the hope that it will be useful,
12968 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12969 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12970 + * GNU General Public License for more details.
12971 + *
12972 + * You should have received a copy of the GNU General Public License
12973 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12974 + */
12975 +
12976 +/*
12977 + * file private data
12978 + */
12979 +
12980 +#include "aufs.h"
12981 +
12982 +void au_hfput(struct au_hfile *hf, struct file *file)
12983 +{
12984 +       /* todo: direct access f_flags */
12985 +       if (vfsub_file_flags(file) & __FMODE_EXEC)
12986 +               allow_write_access(hf->hf_file);
12987 +       fput(hf->hf_file);
12988 +       hf->hf_file = NULL;
12989 +       atomic_dec(&hf->hf_br->br_count);
12990 +       hf->hf_br = NULL;
12991 +}
12992 +
12993 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex, struct file *val)
12994 +{
12995 +       struct au_finfo *finfo = au_fi(file);
12996 +       struct au_hfile *hf;
12997 +       struct au_fidir *fidir;
12998 +
12999 +       fidir = finfo->fi_hdir;
13000 +       if (!fidir) {
13001 +               AuDebugOn(finfo->fi_btop != bindex);
13002 +               hf = &finfo->fi_htop;
13003 +       } else
13004 +               hf = fidir->fd_hfile + bindex;
13005 +
13006 +       if (hf && hf->hf_file)
13007 +               au_hfput(hf, file);
13008 +       if (val) {
13009 +               FiMustWriteLock(file);
13010 +               hf->hf_file = val;
13011 +               hf->hf_br = au_sbr(file->f_path.dentry->d_sb, bindex);
13012 +       }
13013 +}
13014 +
13015 +void au_update_figen(struct file *file)
13016 +{
13017 +       atomic_set(&au_fi(file)->fi_generation, au_digen(file->f_path.dentry));
13018 +       /* smp_mb(); */ /* atomic_set */
13019 +}
13020 +
13021 +/* ---------------------------------------------------------------------- */
13022 +
13023 +struct au_fidir *au_fidir_alloc(struct super_block *sb)
13024 +{
13025 +       struct au_fidir *fidir;
13026 +       int nbr;
13027 +
13028 +       nbr = au_sbend(sb) + 1;
13029 +       if (nbr < 2)
13030 +               nbr = 2; /* initial allocate for 2 branches */
13031 +       fidir = kzalloc(au_fidir_sz(nbr), GFP_NOFS);
13032 +       if (fidir) {
13033 +               fidir->fd_bbot = -1;
13034 +               fidir->fd_nent = nbr;
13035 +               fidir->fd_vdir_cache = NULL;
13036 +       }
13037 +
13038 +       return fidir;
13039 +}
13040 +
13041 +int au_fidir_realloc(struct au_finfo *finfo, int nbr)
13042 +{
13043 +       int err;
13044 +       struct au_fidir *fidir, *p;
13045 +
13046 +       AuRwMustWriteLock(&finfo->fi_rwsem);
13047 +       fidir = finfo->fi_hdir;
13048 +       AuDebugOn(!fidir);
13049 +
13050 +       err = -ENOMEM;
13051 +       p = au_kzrealloc(fidir, au_fidir_sz(fidir->fd_nent), au_fidir_sz(nbr),
13052 +                        GFP_NOFS);
13053 +       if (p) {
13054 +               p->fd_nent = nbr;
13055 +               finfo->fi_hdir = p;
13056 +               err = 0;
13057 +       }
13058 +
13059 +       return err;
13060 +}
13061 +
13062 +/* ---------------------------------------------------------------------- */
13063 +
13064 +void au_finfo_fin(struct file *file)
13065 +{
13066 +       struct au_finfo *finfo;
13067 +
13068 +       au_nfiles_dec(file->f_path.dentry->d_sb);
13069 +
13070 +       finfo = au_fi(file);
13071 +       AuDebugOn(finfo->fi_hdir);
13072 +       AuRwDestroy(&finfo->fi_rwsem);
13073 +       au_cache_free_finfo(finfo);
13074 +}
13075 +
13076 +void au_fi_init_once(void *_finfo)
13077 +{
13078 +       struct au_finfo *finfo = _finfo;
13079 +       static struct lock_class_key aufs_fi;
13080 +
13081 +       au_rw_init(&finfo->fi_rwsem);
13082 +       au_rw_class(&finfo->fi_rwsem, &aufs_fi);
13083 +}
13084 +
13085 +int au_finfo_init(struct file *file, struct au_fidir *fidir)
13086 +{
13087 +       int err;
13088 +       struct au_finfo *finfo;
13089 +       struct dentry *dentry;
13090 +
13091 +       err = -ENOMEM;
13092 +       dentry = file->f_path.dentry;
13093 +       finfo = au_cache_alloc_finfo();
13094 +       if (unlikely(!finfo))
13095 +               goto out;
13096 +
13097 +       err = 0;
13098 +       au_nfiles_inc(dentry->d_sb);
13099 +       /* verbose coding for lock class name */
13100 +       if (!fidir)
13101 +               au_rw_class(&finfo->fi_rwsem, au_lc_key + AuLcNonDir_FIINFO);
13102 +       else
13103 +               au_rw_class(&finfo->fi_rwsem, au_lc_key + AuLcDir_FIINFO);
13104 +       au_rw_write_lock(&finfo->fi_rwsem);
13105 +       finfo->fi_btop = -1;
13106 +       finfo->fi_hdir = fidir;
13107 +       atomic_set(&finfo->fi_generation, au_digen(dentry));
13108 +       /* smp_mb(); */ /* atomic_set */
13109 +
13110 +       file->private_data = finfo;
13111 +
13112 +out:
13113 +       return err;
13114 +}
13115 diff -urN /usr/share/empty/fs/aufs/f_op.c linux/fs/aufs/f_op.c
13116 --- /usr/share/empty/fs/aufs/f_op.c     1970-01-01 01:00:00.000000000 +0100
13117 +++ linux/fs/aufs/f_op.c        2015-03-27 21:56:35.463461668 +0100
13118 @@ -0,0 +1,814 @@
13119 +/*
13120 + * Copyright (C) 2005-2015 Junjiro R. Okajima
13121 + *
13122 + * This program, aufs is free software; you can redistribute it and/or modify
13123 + * it under the terms of the GNU General Public License as published by
13124 + * the Free Software Foundation; either version 2 of the License, or
13125 + * (at your option) any later version.
13126 + *
13127 + * This program is distributed in the hope that it will be useful,
13128 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13129 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13130 + * GNU General Public License for more details.
13131 + *
13132 + * You should have received a copy of the GNU General Public License
13133 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13134 + */
13135 +
13136 +/*
13137 + * file and vm operations
13138 + */
13139 +
13140 +#include <linux/aio.h>
13141 +#include <linux/fs_stack.h>
13142 +#include <linux/mman.h>
13143 +#include <linux/security.h>
13144 +#include "aufs.h"
13145 +
13146 +int au_do_open_nondir(struct file *file, int flags)
13147 +{
13148 +       int err;
13149 +       aufs_bindex_t bindex;
13150 +       struct file *h_file;
13151 +       struct dentry *dentry;
13152 +       struct au_finfo *finfo;
13153 +       struct inode *h_inode;
13154 +
13155 +       FiMustWriteLock(file);
13156 +
13157 +       err = 0;
13158 +       dentry = file->f_path.dentry;
13159 +       finfo = au_fi(file);
13160 +       memset(&finfo->fi_htop, 0, sizeof(finfo->fi_htop));
13161 +       atomic_set(&finfo->fi_mmapped, 0);
13162 +       bindex = au_dbstart(dentry);
13163 +       h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
13164 +       if (IS_ERR(h_file))
13165 +               err = PTR_ERR(h_file);
13166 +       else {
13167 +               if ((flags & __O_TMPFILE)
13168 +                   && !(flags & O_EXCL)) {
13169 +                       h_inode = file_inode(h_file);
13170 +                       spin_lock(&h_inode->i_lock);
13171 +                       h_inode->i_state |= I_LINKABLE;
13172 +                       spin_unlock(&h_inode->i_lock);
13173 +               }
13174 +               au_set_fbstart(file, bindex);
13175 +               au_set_h_fptr(file, bindex, h_file);
13176 +               au_update_figen(file);
13177 +               /* todo: necessary? */
13178 +               /* file->f_ra = h_file->f_ra; */
13179 +       }
13180 +
13181 +       return err;
13182 +}
13183 +
13184 +static int aufs_open_nondir(struct inode *inode __maybe_unused,
13185 +                           struct file *file)
13186 +{
13187 +       int err;
13188 +       struct super_block *sb;
13189 +
13190 +       AuDbg("%pD, f_flags 0x%x, f_mode 0x%x\n",
13191 +             file, vfsub_file_flags(file), file->f_mode);
13192 +
13193 +       sb = file->f_path.dentry->d_sb;
13194 +       si_read_lock(sb, AuLock_FLUSH);
13195 +       err = au_do_open(file, au_do_open_nondir, /*fidir*/NULL);
13196 +       si_read_unlock(sb);
13197 +       return err;
13198 +}
13199 +
13200 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file)
13201 +{
13202 +       struct au_finfo *finfo;
13203 +       aufs_bindex_t bindex;
13204 +
13205 +       finfo = au_fi(file);
13206 +       au_sphl_del(&finfo->fi_hlist,
13207 +                   &au_sbi(file->f_path.dentry->d_sb)->si_files);
13208 +       bindex = finfo->fi_btop;
13209 +       if (bindex >= 0)
13210 +               au_set_h_fptr(file, bindex, NULL);
13211 +
13212 +       au_finfo_fin(file);
13213 +       return 0;
13214 +}
13215 +
13216 +/* ---------------------------------------------------------------------- */
13217 +
13218 +static int au_do_flush_nondir(struct file *file, fl_owner_t id)
13219 +{
13220 +       int err;
13221 +       struct file *h_file;
13222 +
13223 +       err = 0;
13224 +       h_file = au_hf_top(file);
13225 +       if (h_file)
13226 +               err = vfsub_flush(h_file, id);
13227 +       return err;
13228 +}
13229 +
13230 +static int aufs_flush_nondir(struct file *file, fl_owner_t id)
13231 +{
13232 +       return au_do_flush(file, id, au_do_flush_nondir);
13233 +}
13234 +
13235 +/* ---------------------------------------------------------------------- */
13236 +/*
13237 + * read and write functions acquire [fdi]_rwsem once, but release before
13238 + * mmap_sem. This is because to stop a race condition between mmap(2).
13239 + * Releasing these aufs-rwsem should be safe, no branch-mamagement (by keeping
13240 + * si_rwsem), no harmful copy-up should happen. Actually copy-up may happen in
13241 + * read functions after [fdi]_rwsem are released, but it should be harmless.
13242 + */
13243 +
13244 +static ssize_t aufs_read(struct file *file, char __user *buf, size_t count,
13245 +                        loff_t *ppos)
13246 +{
13247 +       ssize_t err;
13248 +       struct dentry *dentry;
13249 +       struct file *h_file;
13250 +       struct super_block *sb;
13251 +
13252 +       dentry = file->f_path.dentry;
13253 +       sb = dentry->d_sb;
13254 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13255 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13256 +       if (unlikely(err))
13257 +               goto out;
13258 +
13259 +       h_file = au_hf_top(file);
13260 +       get_file(h_file);
13261 +       di_read_unlock(dentry, AuLock_IR);
13262 +       fi_read_unlock(file);
13263 +
13264 +       /* filedata may be obsoleted by concurrent copyup, but no problem */
13265 +       err = vfsub_read_u(h_file, buf, count, ppos);
13266 +       /* todo: necessary? */
13267 +       /* file->f_ra = h_file->f_ra; */
13268 +       /* update without lock, I don't think it a problem */
13269 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
13270 +       fput(h_file);
13271 +
13272 +out:
13273 +       si_read_unlock(sb);
13274 +       return err;
13275 +}
13276 +
13277 +/*
13278 + * todo: very ugly
13279 + * it locks both of i_mutex and si_rwsem for read in safe.
13280 + * if the plink maintenance mode continues forever (that is the problem),
13281 + * may loop forever.
13282 + */
13283 +static void au_mtx_and_read_lock(struct inode *inode)
13284 +{
13285 +       int err;
13286 +       struct super_block *sb = inode->i_sb;
13287 +
13288 +       while (1) {
13289 +               mutex_lock(&inode->i_mutex);
13290 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
13291 +               if (!err)
13292 +                       break;
13293 +               mutex_unlock(&inode->i_mutex);
13294 +               si_read_lock(sb, AuLock_NOPLMW);
13295 +               si_read_unlock(sb);
13296 +       }
13297 +}
13298 +
13299 +static ssize_t aufs_write(struct file *file, const char __user *ubuf,
13300 +                         size_t count, loff_t *ppos)
13301 +{
13302 +       ssize_t err;
13303 +       blkcnt_t blks;
13304 +       aufs_bindex_t bstart;
13305 +       struct au_pin pin;
13306 +       struct dentry *dentry;
13307 +       struct inode *inode, *h_inode;
13308 +       struct super_block *sb;
13309 +       struct file *h_file;
13310 +       char __user *buf = (char __user *)ubuf;
13311 +
13312 +       dentry = file->f_path.dentry;
13313 +       sb = dentry->d_sb;
13314 +       inode = dentry->d_inode;
13315 +       au_mtx_and_read_lock(inode);
13316 +
13317 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13318 +       if (unlikely(err))
13319 +               goto out;
13320 +
13321 +       err = au_ready_to_write(file, -1, &pin);
13322 +       di_downgrade_lock(dentry, AuLock_IR);
13323 +       if (unlikely(err)) {
13324 +               di_read_unlock(dentry, AuLock_IR);
13325 +               fi_write_unlock(file);
13326 +               goto out;
13327 +       }
13328 +
13329 +       bstart = au_fbstart(file);
13330 +       h_file = au_hf_top(file);
13331 +       get_file(h_file);
13332 +       h_inode = file_inode(h_file);
13333 +       blks = h_inode->i_blocks;
13334 +       au_unpin(&pin);
13335 +       di_read_unlock(dentry, AuLock_IR);
13336 +       fi_write_unlock(file);
13337 +
13338 +       err = vfsub_write_u(h_file, buf, count, ppos);
13339 +       ii_write_lock_child(inode);
13340 +       au_cpup_attr_timesizes(inode);
13341 +       inode->i_mode = file_inode(h_file)->i_mode;
13342 +       AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks);
13343 +       if (err > 0)
13344 +               au_fhsm_wrote(sb, bstart, /*force*/h_inode->i_blocks > blks);
13345 +       ii_write_unlock(inode);
13346 +       fput(h_file);
13347 +
13348 +out:
13349 +       si_read_unlock(sb);
13350 +       mutex_unlock(&inode->i_mutex);
13351 +       return err;
13352 +}
13353 +
13354 +static ssize_t au_do_iter(struct file *h_file, int rw, struct kiocb *kio,
13355 +                         struct iov_iter *iov_iter)
13356 +{
13357 +       ssize_t err;
13358 +       struct file *file;
13359 +       ssize_t (*iter)(struct kiocb *, struct iov_iter *);
13360 +       ssize_t (*aio)(struct kiocb *, const struct iovec *, unsigned long,
13361 +                      loff_t);
13362 +
13363 +       err = security_file_permission(h_file, rw);
13364 +       if (unlikely(err))
13365 +               goto out;
13366 +
13367 +       err = -ENOSYS;
13368 +       iter = NULL;
13369 +       aio = NULL;
13370 +       if (rw == MAY_READ) {
13371 +               iter = h_file->f_op->read_iter;
13372 +               aio = h_file->f_op->aio_read;
13373 +       } else if (rw == MAY_WRITE) {
13374 +               iter = h_file->f_op->write_iter;
13375 +               aio = h_file->f_op->aio_write;
13376 +       }
13377 +
13378 +       file = kio->ki_filp;
13379 +       kio->ki_filp = h_file;
13380 +       if (iter) {
13381 +               lockdep_off();
13382 +               err = iter(kio, iov_iter);
13383 +               lockdep_on();
13384 +       } else if (aio) {
13385 +               lockdep_off();
13386 +               err = aio(kio, iov_iter->iov, iov_iter->nr_segs, kio->ki_pos);
13387 +               lockdep_on();
13388 +       } else
13389 +               /* currently there is no such fs */
13390 +               WARN_ON_ONCE(1);
13391 +       kio->ki_filp = file;
13392 +
13393 +out:
13394 +       return err;
13395 +}
13396 +
13397 +static ssize_t aufs_read_iter(struct kiocb *kio, struct iov_iter *iov_iter)
13398 +{
13399 +       ssize_t err;
13400 +       struct file *file, *h_file;
13401 +       struct dentry *dentry;
13402 +       struct super_block *sb;
13403 +
13404 +       file = kio->ki_filp;
13405 +       dentry = file->f_path.dentry;
13406 +       sb = dentry->d_sb;
13407 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13408 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13409 +       if (unlikely(err))
13410 +               goto out;
13411 +
13412 +       h_file = au_hf_top(file);
13413 +       get_file(h_file);
13414 +       di_read_unlock(dentry, AuLock_IR);
13415 +       fi_read_unlock(file);
13416 +
13417 +       err = au_do_iter(h_file, MAY_READ, kio, iov_iter);
13418 +       /* todo: necessary? */
13419 +       /* file->f_ra = h_file->f_ra; */
13420 +       /* update without lock, I don't think it a problem */
13421 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
13422 +       fput(h_file);
13423 +
13424 +out:
13425 +       si_read_unlock(sb);
13426 +       return err;
13427 +}
13428 +
13429 +static ssize_t aufs_write_iter(struct kiocb *kio, struct iov_iter *iov_iter)
13430 +{
13431 +       ssize_t err;
13432 +       blkcnt_t blks;
13433 +       aufs_bindex_t bstart;
13434 +       struct au_pin pin;
13435 +       struct dentry *dentry;
13436 +       struct inode *inode, *h_inode;
13437 +       struct file *file, *h_file;
13438 +       struct super_block *sb;
13439 +
13440 +       file = kio->ki_filp;
13441 +       dentry = file->f_path.dentry;
13442 +       sb = dentry->d_sb;
13443 +       inode = dentry->d_inode;
13444 +       au_mtx_and_read_lock(inode);
13445 +
13446 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13447 +       if (unlikely(err))
13448 +               goto out;
13449 +
13450 +       err = au_ready_to_write(file, -1, &pin);
13451 +       di_downgrade_lock(dentry, AuLock_IR);
13452 +       if (unlikely(err)) {
13453 +               di_read_unlock(dentry, AuLock_IR);
13454 +               fi_write_unlock(file);
13455 +               goto out;
13456 +       }
13457 +
13458 +       bstart = au_fbstart(file);
13459 +       h_file = au_hf_top(file);
13460 +       get_file(h_file);
13461 +       h_inode = file_inode(h_file);
13462 +       blks = h_inode->i_blocks;
13463 +       au_unpin(&pin);
13464 +       di_read_unlock(dentry, AuLock_IR);
13465 +       fi_write_unlock(file);
13466 +
13467 +       err = au_do_iter(h_file, MAY_WRITE, kio, iov_iter);
13468 +       ii_write_lock_child(inode);
13469 +       au_cpup_attr_timesizes(inode);
13470 +       inode->i_mode = file_inode(h_file)->i_mode;
13471 +       AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks);
13472 +       if (err > 0)
13473 +               au_fhsm_wrote(sb, bstart, /*force*/h_inode->i_blocks > blks);
13474 +       ii_write_unlock(inode);
13475 +       fput(h_file);
13476 +
13477 +out:
13478 +       si_read_unlock(sb);
13479 +       mutex_unlock(&inode->i_mutex);
13480 +       return err;
13481 +}
13482 +
13483 +static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
13484 +                               struct pipe_inode_info *pipe, size_t len,
13485 +                               unsigned int flags)
13486 +{
13487 +       ssize_t err;
13488 +       struct file *h_file;
13489 +       struct dentry *dentry;
13490 +       struct super_block *sb;
13491 +
13492 +       dentry = file->f_path.dentry;
13493 +       sb = dentry->d_sb;
13494 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13495 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13496 +       if (unlikely(err))
13497 +               goto out;
13498 +
13499 +       err = -EINVAL;
13500 +       h_file = au_hf_top(file);
13501 +       get_file(h_file);
13502 +       if (au_test_loopback_kthread()) {
13503 +               au_warn_loopback(h_file->f_path.dentry->d_sb);
13504 +               if (file->f_mapping != h_file->f_mapping) {
13505 +                       file->f_mapping = h_file->f_mapping;
13506 +                       smp_mb(); /* unnecessary? */
13507 +               }
13508 +       }
13509 +       di_read_unlock(dentry, AuLock_IR);
13510 +       fi_read_unlock(file);
13511 +
13512 +       err = vfsub_splice_to(h_file, ppos, pipe, len, flags);
13513 +       /* todo: necessasry? */
13514 +       /* file->f_ra = h_file->f_ra; */
13515 +       /* update without lock, I don't think it a problem */
13516 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
13517 +       fput(h_file);
13518 +
13519 +out:
13520 +       si_read_unlock(sb);
13521 +       return err;
13522 +}
13523 +
13524 +static ssize_t
13525 +aufs_splice_write(struct pipe_inode_info *pipe, struct file *file, loff_t *ppos,
13526 +                 size_t len, unsigned int flags)
13527 +{
13528 +       ssize_t err;
13529 +       blkcnt_t blks;
13530 +       aufs_bindex_t bstart;
13531 +       struct au_pin pin;
13532 +       struct dentry *dentry;
13533 +       struct inode *inode, *h_inode;
13534 +       struct super_block *sb;
13535 +       struct file *h_file;
13536 +
13537 +       dentry = file->f_path.dentry;
13538 +       sb = dentry->d_sb;
13539 +       inode = dentry->d_inode;
13540 +       au_mtx_and_read_lock(inode);
13541 +
13542 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13543 +       if (unlikely(err))
13544 +               goto out;
13545 +
13546 +       err = au_ready_to_write(file, -1, &pin);
13547 +       di_downgrade_lock(dentry, AuLock_IR);
13548 +       if (unlikely(err)) {
13549 +               di_read_unlock(dentry, AuLock_IR);
13550 +               fi_write_unlock(file);
13551 +               goto out;
13552 +       }
13553 +
13554 +       bstart = au_fbstart(file);
13555 +       h_file = au_hf_top(file);
13556 +       get_file(h_file);
13557 +       h_inode = file_inode(h_file);
13558 +       blks = h_inode->i_blocks;
13559 +       au_unpin(&pin);
13560 +       di_read_unlock(dentry, AuLock_IR);
13561 +       fi_write_unlock(file);
13562 +
13563 +       err = vfsub_splice_from(pipe, h_file, ppos, len, flags);
13564 +       ii_write_lock_child(inode);
13565 +       au_cpup_attr_timesizes(inode);
13566 +       inode->i_mode = file_inode(h_file)->i_mode;
13567 +       AuDbg("blks %llu, %llu\n", (u64)blks, (u64)h_inode->i_blocks);
13568 +       if (err > 0)
13569 +               au_fhsm_wrote(sb, bstart, /*force*/h_inode->i_blocks > blks);
13570 +       ii_write_unlock(inode);
13571 +       fput(h_file);
13572 +
13573 +out:
13574 +       si_read_unlock(sb);
13575 +       mutex_unlock(&inode->i_mutex);
13576 +       return err;
13577 +}
13578 +
13579 +static long aufs_fallocate(struct file *file, int mode, loff_t offset,
13580 +                          loff_t len)
13581 +{
13582 +       long err;
13583 +       struct au_pin pin;
13584 +       struct dentry *dentry;
13585 +       struct super_block *sb;
13586 +       struct inode *inode;
13587 +       struct file *h_file;
13588 +
13589 +       dentry = file->f_path.dentry;
13590 +       sb = dentry->d_sb;
13591 +       inode = dentry->d_inode;
13592 +       au_mtx_and_read_lock(inode);
13593 +
13594 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13595 +       if (unlikely(err))
13596 +               goto out;
13597 +
13598 +       err = au_ready_to_write(file, -1, &pin);
13599 +       di_downgrade_lock(dentry, AuLock_IR);
13600 +       if (unlikely(err)) {
13601 +               di_read_unlock(dentry, AuLock_IR);
13602 +               fi_write_unlock(file);
13603 +               goto out;
13604 +       }
13605 +
13606 +       h_file = au_hf_top(file);
13607 +       get_file(h_file);
13608 +       au_unpin(&pin);
13609 +       di_read_unlock(dentry, AuLock_IR);
13610 +       fi_write_unlock(file);
13611 +
13612 +       lockdep_off();
13613 +       err = vfs_fallocate(h_file, mode, offset, len);
13614 +       lockdep_on();
13615 +       ii_write_lock_child(inode);
13616 +       au_cpup_attr_timesizes(inode);
13617 +       inode->i_mode = file_inode(h_file)->i_mode;
13618 +       ii_write_unlock(inode);
13619 +       fput(h_file);
13620 +
13621 +out:
13622 +       si_read_unlock(sb);
13623 +       mutex_unlock(&inode->i_mutex);
13624 +       return err;
13625 +}
13626 +
13627 +/* ---------------------------------------------------------------------- */
13628 +
13629 +/*
13630 + * The locking order around current->mmap_sem.
13631 + * - in most and regular cases
13632 + *   file I/O syscall -- aufs_read() or something
13633 + *     -- si_rwsem for read -- mmap_sem
13634 + *     (Note that [fdi]i_rwsem are released before mmap_sem).
13635 + * - in mmap case
13636 + *   mmap(2) -- mmap_sem -- aufs_mmap() -- si_rwsem for read -- [fdi]i_rwsem
13637 + * This AB-BA order is definitly bad, but is not a problem since "si_rwsem for
13638 + * read" allows muliple processes to acquire it and [fdi]i_rwsem are not held in
13639 + * file I/O. Aufs needs to stop lockdep in aufs_mmap() though.
13640 + * It means that when aufs acquires si_rwsem for write, the process should never
13641 + * acquire mmap_sem.
13642 + *
13643 + * Actually aufs_iterate() holds [fdi]i_rwsem before mmap_sem, but this is not a
13644 + * problem either since any directory is not able to be mmap-ed.
13645 + * The similar scenario is applied to aufs_readlink() too.
13646 + */
13647 +
13648 +#if 0 /* stop calling security_file_mmap() */
13649 +/* cf. linux/include/linux/mman.h: calc_vm_prot_bits() */
13650 +#define AuConv_VM_PROT(f, b)   _calc_vm_trans(f, VM_##b, PROT_##b)
13651 +
13652 +static unsigned long au_arch_prot_conv(unsigned long flags)
13653 +{
13654 +       /* currently ppc64 only */
13655 +#ifdef CONFIG_PPC64
13656 +       /* cf. linux/arch/powerpc/include/asm/mman.h */
13657 +       AuDebugOn(arch_calc_vm_prot_bits(-1) != VM_SAO);
13658 +       return AuConv_VM_PROT(flags, SAO);
13659 +#else
13660 +       AuDebugOn(arch_calc_vm_prot_bits(-1));
13661 +       return 0;
13662 +#endif
13663 +}
13664 +
13665 +static unsigned long au_prot_conv(unsigned long flags)
13666 +{
13667 +       return AuConv_VM_PROT(flags, READ)
13668 +               | AuConv_VM_PROT(flags, WRITE)
13669 +               | AuConv_VM_PROT(flags, EXEC)
13670 +               | au_arch_prot_conv(flags);
13671 +}
13672 +
13673 +/* cf. linux/include/linux/mman.h: calc_vm_flag_bits() */
13674 +#define AuConv_VM_MAP(f, b)    _calc_vm_trans(f, VM_##b, MAP_##b)
13675 +
13676 +static unsigned long au_flag_conv(unsigned long flags)
13677 +{
13678 +       return AuConv_VM_MAP(flags, GROWSDOWN)
13679 +               | AuConv_VM_MAP(flags, DENYWRITE)
13680 +               | AuConv_VM_MAP(flags, LOCKED);
13681 +}
13682 +#endif
13683 +
13684 +static int aufs_mmap(struct file *file, struct vm_area_struct *vma)
13685 +{
13686 +       int err;
13687 +       aufs_bindex_t bstart;
13688 +       const unsigned char wlock
13689 +               = (file->f_mode & FMODE_WRITE) && (vma->vm_flags & VM_SHARED);
13690 +       struct dentry *dentry;
13691 +       struct super_block *sb;
13692 +       struct file *h_file;
13693 +       struct au_branch *br;
13694 +       struct au_pin pin;
13695 +
13696 +       AuDbgVmRegion(file, vma);
13697 +
13698 +       dentry = file->f_path.dentry;
13699 +       sb = dentry->d_sb;
13700 +       lockdep_off();
13701 +       si_read_lock(sb, AuLock_NOPLMW);
13702 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13703 +       if (unlikely(err))
13704 +               goto out;
13705 +
13706 +       if (wlock) {
13707 +               err = au_ready_to_write(file, -1, &pin);
13708 +               di_write_unlock(dentry);
13709 +               if (unlikely(err)) {
13710 +                       fi_write_unlock(file);
13711 +                       goto out;
13712 +               }
13713 +               au_unpin(&pin);
13714 +       } else
13715 +               di_write_unlock(dentry);
13716 +
13717 +       bstart = au_fbstart(file);
13718 +       br = au_sbr(sb, bstart);
13719 +       h_file = au_hf_top(file);
13720 +       get_file(h_file);
13721 +       au_set_mmapped(file);
13722 +       fi_write_unlock(file);
13723 +       lockdep_on();
13724 +
13725 +       au_vm_file_reset(vma, h_file);
13726 +       /*
13727 +        * we cannot call security_mmap_file() here since it may acquire
13728 +        * mmap_sem or i_mutex.
13729 +        *
13730 +        * err = security_mmap_file(h_file, au_prot_conv(vma->vm_flags),
13731 +        *                       au_flag_conv(vma->vm_flags));
13732 +        */
13733 +       if (!err)
13734 +               err = h_file->f_op->mmap(h_file, vma);
13735 +       if (unlikely(err))
13736 +               goto out_reset;
13737 +
13738 +       au_vm_prfile_set(vma, file);
13739 +       /* update without lock, I don't think it a problem */
13740 +       fsstack_copy_attr_atime(file_inode(file), file_inode(h_file));
13741 +       goto out_fput; /* success */
13742 +
13743 +out_reset:
13744 +       au_unset_mmapped(file);
13745 +       au_vm_file_reset(vma, file);
13746 +out_fput:
13747 +       fput(h_file);
13748 +       lockdep_off();
13749 +out:
13750 +       si_read_unlock(sb);
13751 +       lockdep_on();
13752 +       AuTraceErr(err);
13753 +       return err;
13754 +}
13755 +
13756 +/* ---------------------------------------------------------------------- */
13757 +
13758 +static int aufs_fsync_nondir(struct file *file, loff_t start, loff_t end,
13759 +                            int datasync)
13760 +{
13761 +       int err;
13762 +       struct au_pin pin;
13763 +       struct dentry *dentry;
13764 +       struct inode *inode;
13765 +       struct file *h_file;
13766 +       struct super_block *sb;
13767 +
13768 +       dentry = file->f_path.dentry;
13769 +       inode = dentry->d_inode;
13770 +       sb = dentry->d_sb;
13771 +       mutex_lock(&inode->i_mutex);
13772 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
13773 +       if (unlikely(err))
13774 +               goto out;
13775 +
13776 +       err = 0; /* -EBADF; */ /* posix? */
13777 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
13778 +               goto out_si;
13779 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13780 +       if (unlikely(err))
13781 +               goto out_si;
13782 +
13783 +       err = au_ready_to_write(file, -1, &pin);
13784 +       di_downgrade_lock(dentry, AuLock_IR);
13785 +       if (unlikely(err))
13786 +               goto out_unlock;
13787 +       au_unpin(&pin);
13788 +
13789 +       err = -EINVAL;
13790 +       h_file = au_hf_top(file);
13791 +       err = vfsub_fsync(h_file, &h_file->f_path, datasync);
13792 +       au_cpup_attr_timesizes(inode);
13793 +
13794 +out_unlock:
13795 +       di_read_unlock(dentry, AuLock_IR);
13796 +       fi_write_unlock(file);
13797 +out_si:
13798 +       si_read_unlock(sb);
13799 +out:
13800 +       mutex_unlock(&inode->i_mutex);
13801 +       return err;
13802 +}
13803 +
13804 +/* no one supports this operation, currently */
13805 +#if 0
13806 +static int aufs_aio_fsync_nondir(struct kiocb *kio, int datasync)
13807 +{
13808 +       int err;
13809 +       struct au_pin pin;
13810 +       struct dentry *dentry;
13811 +       struct inode *inode;
13812 +       struct file *file, *h_file;
13813 +
13814 +       file = kio->ki_filp;
13815 +       dentry = file->f_path.dentry;
13816 +       inode = dentry->d_inode;
13817 +       au_mtx_and_read_lock(inode);
13818 +
13819 +       err = 0; /* -EBADF; */ /* posix? */
13820 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
13821 +               goto out;
13822 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
13823 +       if (unlikely(err))
13824 +               goto out;
13825 +
13826 +       err = au_ready_to_write(file, -1, &pin);
13827 +       di_downgrade_lock(dentry, AuLock_IR);
13828 +       if (unlikely(err))
13829 +               goto out_unlock;
13830 +       au_unpin(&pin);
13831 +
13832 +       err = -ENOSYS;
13833 +       h_file = au_hf_top(file);
13834 +       if (h_file->f_op->aio_fsync) {
13835 +               struct mutex *h_mtx;
13836 +
13837 +               h_mtx = &file_inode(h_file)->i_mutex;
13838 +               if (!is_sync_kiocb(kio)) {
13839 +                       get_file(h_file);
13840 +                       fput(file);
13841 +               }
13842 +               kio->ki_filp = h_file;
13843 +               err = h_file->f_op->aio_fsync(kio, datasync);
13844 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
13845 +               if (!err)
13846 +                       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL);
13847 +               /*ignore*/
13848 +               au_cpup_attr_timesizes(inode);
13849 +               mutex_unlock(h_mtx);
13850 +       }
13851 +
13852 +out_unlock:
13853 +       di_read_unlock(dentry, AuLock_IR);
13854 +       fi_write_unlock(file);
13855 +out:
13856 +       si_read_unlock(inode->sb);
13857 +       mutex_unlock(&inode->i_mutex);
13858 +       return err;
13859 +}
13860 +#endif
13861 +
13862 +static int aufs_fasync(int fd, struct file *file, int flag)
13863 +{
13864 +       int err;
13865 +       struct file *h_file;
13866 +       struct dentry *dentry;
13867 +       struct super_block *sb;
13868 +
13869 +       dentry = file->f_path.dentry;
13870 +       sb = dentry->d_sb;
13871 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
13872 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
13873 +       if (unlikely(err))
13874 +               goto out;
13875 +
13876 +       h_file = au_hf_top(file);
13877 +       if (h_file->f_op->fasync)
13878 +               err = h_file->f_op->fasync(fd, h_file, flag);
13879 +
13880 +       di_read_unlock(dentry, AuLock_IR);
13881 +       fi_read_unlock(file);
13882 +
13883 +out:
13884 +       si_read_unlock(sb);
13885 +       return err;
13886 +}
13887 +
13888 +/* ---------------------------------------------------------------------- */
13889 +
13890 +/* no one supports this operation, currently */
13891 +#if 0
13892 +static ssize_t aufs_sendpage(struct file *file, struct page *page, int offset,
13893 +                            size_t len, loff_t *pos, int more)
13894 +{
13895 +}
13896 +#endif
13897 +
13898 +/* ---------------------------------------------------------------------- */
13899 +
13900 +const struct file_operations aufs_file_fop = {
13901 +       .owner          = THIS_MODULE,
13902 +
13903 +       .llseek         = default_llseek,
13904 +
13905 +       .read           = aufs_read,
13906 +       .write          = aufs_write,
13907 +       .read_iter      = aufs_read_iter,
13908 +       .write_iter     = aufs_write_iter,
13909 +
13910 +#ifdef CONFIG_AUFS_POLL
13911 +       .poll           = aufs_poll,
13912 +#endif
13913 +       .unlocked_ioctl = aufs_ioctl_nondir,
13914 +#ifdef CONFIG_COMPAT
13915 +       .compat_ioctl   = aufs_compat_ioctl_nondir,
13916 +#endif
13917 +       .mmap           = aufs_mmap,
13918 +       .open           = aufs_open_nondir,
13919 +       .flush          = aufs_flush_nondir,
13920 +       .release        = aufs_release_nondir,
13921 +       .fsync          = aufs_fsync_nondir,
13922 +       /* .aio_fsync   = aufs_aio_fsync_nondir, */
13923 +       .fasync         = aufs_fasync,
13924 +       /* .sendpage    = aufs_sendpage, */
13925 +       .splice_write   = aufs_splice_write,
13926 +       .splice_read    = aufs_splice_read,
13927 +#if 0
13928 +       .aio_splice_write = aufs_aio_splice_write,
13929 +       .aio_splice_read  = aufs_aio_splice_read,
13930 +#endif
13931 +       .fallocate      = aufs_fallocate
13932 +};
13933 diff -urN /usr/share/empty/fs/aufs/fstype.h linux/fs/aufs/fstype.h
13934 --- /usr/share/empty/fs/aufs/fstype.h   1970-01-01 01:00:00.000000000 +0100
13935 +++ linux/fs/aufs/fstype.h      2015-03-27 21:56:35.463461668 +0100
13936 @@ -0,0 +1,469 @@
13937 +/*
13938 + * Copyright (C) 2005-2015 Junjiro R. Okajima
13939 + *
13940 + * This program, aufs is free software; you can redistribute it and/or modify
13941 + * it under the terms of the GNU General Public License as published by
13942 + * the Free Software Foundation; either version 2 of the License, or
13943 + * (at your option) any later version.
13944 + *
13945 + * This program is distributed in the hope that it will be useful,
13946 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13947 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13948 + * GNU General Public License for more details.
13949 + *
13950 + * You should have received a copy of the GNU General Public License
13951 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13952 + */
13953 +
13954 +/*
13955 + * judging filesystem type
13956 + */
13957 +
13958 +#ifndef __AUFS_FSTYPE_H__
13959 +#define __AUFS_FSTYPE_H__
13960 +
13961 +#ifdef __KERNEL__
13962 +
13963 +#include <linux/fs.h>
13964 +#include <linux/magic.h>
13965 +#include <linux/romfs_fs.h>
13966 +
13967 +static inline int au_test_aufs(struct super_block *sb)
13968 +{
13969 +       return sb->s_magic == AUFS_SUPER_MAGIC;
13970 +}
13971 +
13972 +static inline const char *au_sbtype(struct super_block *sb)
13973 +{
13974 +       return sb->s_type->name;
13975 +}
13976 +
13977 +static inline int au_test_iso9660(struct super_block *sb __maybe_unused)
13978 +{
13979 +#if defined(CONFIG_ISO9660_FS) || defined(CONFIG_ISO9660_FS_MODULE)
13980 +       return sb->s_magic == ISOFS_SUPER_MAGIC;
13981 +#else
13982 +       return 0;
13983 +#endif
13984 +}
13985 +
13986 +static inline int au_test_romfs(struct super_block *sb __maybe_unused)
13987 +{
13988 +#if defined(CONFIG_ROMFS_FS) || defined(CONFIG_ROMFS_FS_MODULE)
13989 +       return sb->s_magic == ROMFS_MAGIC;
13990 +#else
13991 +       return 0;
13992 +#endif
13993 +}
13994 +
13995 +static inline int au_test_cramfs(struct super_block *sb __maybe_unused)
13996 +{
13997 +#if defined(CONFIG_CRAMFS) || defined(CONFIG_CRAMFS_MODULE)
13998 +       return sb->s_magic == CRAMFS_MAGIC;
13999 +#endif
14000 +       return 0;
14001 +}
14002 +
14003 +static inline int au_test_nfs(struct super_block *sb __maybe_unused)
14004 +{
14005 +#if defined(CONFIG_NFS_FS) || defined(CONFIG_NFS_FS_MODULE)
14006 +       return sb->s_magic == NFS_SUPER_MAGIC;
14007 +#else
14008 +       return 0;
14009 +#endif
14010 +}
14011 +
14012 +static inline int au_test_fuse(struct super_block *sb __maybe_unused)
14013 +{
14014 +#if defined(CONFIG_FUSE_FS) || defined(CONFIG_FUSE_FS_MODULE)
14015 +       return sb->s_magic == FUSE_SUPER_MAGIC;
14016 +#else
14017 +       return 0;
14018 +#endif
14019 +}
14020 +
14021 +static inline int au_test_xfs(struct super_block *sb __maybe_unused)
14022 +{
14023 +#if defined(CONFIG_XFS_FS) || defined(CONFIG_XFS_FS_MODULE)
14024 +       return sb->s_magic == XFS_SB_MAGIC;
14025 +#else
14026 +       return 0;
14027 +#endif
14028 +}
14029 +
14030 +static inline int au_test_tmpfs(struct super_block *sb __maybe_unused)
14031 +{
14032 +#ifdef CONFIG_TMPFS
14033 +       return sb->s_magic == TMPFS_MAGIC;
14034 +#else
14035 +       return 0;
14036 +#endif
14037 +}
14038 +
14039 +static inline int au_test_ecryptfs(struct super_block *sb __maybe_unused)
14040 +{
14041 +#if defined(CONFIG_ECRYPT_FS) || defined(CONFIG_ECRYPT_FS_MODULE)
14042 +       return !strcmp(au_sbtype(sb), "ecryptfs");
14043 +#else
14044 +       return 0;
14045 +#endif
14046 +}
14047 +
14048 +static inline int au_test_ocfs2(struct super_block *sb __maybe_unused)
14049 +{
14050 +#if defined(CONFIG_OCFS2_FS) || defined(CONFIG_OCFS2_FS_MODULE)
14051 +       return sb->s_magic == OCFS2_SUPER_MAGIC;
14052 +#else
14053 +       return 0;
14054 +#endif
14055 +}
14056 +
14057 +static inline int au_test_ocfs2_dlmfs(struct super_block *sb __maybe_unused)
14058 +{
14059 +#if defined(CONFIG_OCFS2_FS_O2CB) || defined(CONFIG_OCFS2_FS_O2CB_MODULE)
14060 +       return sb->s_magic == DLMFS_MAGIC;
14061 +#else
14062 +       return 0;
14063 +#endif
14064 +}
14065 +
14066 +static inline int au_test_coda(struct super_block *sb __maybe_unused)
14067 +{
14068 +#if defined(CONFIG_CODA_FS) || defined(CONFIG_CODA_FS_MODULE)
14069 +       return sb->s_magic == CODA_SUPER_MAGIC;
14070 +#else
14071 +       return 0;
14072 +#endif
14073 +}
14074 +
14075 +static inline int au_test_v9fs(struct super_block *sb __maybe_unused)
14076 +{
14077 +#if defined(CONFIG_9P_FS) || defined(CONFIG_9P_FS_MODULE)
14078 +       return sb->s_magic == V9FS_MAGIC;
14079 +#else
14080 +       return 0;
14081 +#endif
14082 +}
14083 +
14084 +static inline int au_test_ext4(struct super_block *sb __maybe_unused)
14085 +{
14086 +#if defined(CONFIG_EXT4_FS) || defined(CONFIG_EXT4_FS_MODULE)
14087 +       return sb->s_magic == EXT4_SUPER_MAGIC;
14088 +#else
14089 +       return 0;
14090 +#endif
14091 +}
14092 +
14093 +static inline int au_test_sysv(struct super_block *sb __maybe_unused)
14094 +{
14095 +#if defined(CONFIG_SYSV_FS) || defined(CONFIG_SYSV_FS_MODULE)
14096 +       return !strcmp(au_sbtype(sb), "sysv");
14097 +#else
14098 +       return 0;
14099 +#endif
14100 +}
14101 +
14102 +static inline int au_test_ramfs(struct super_block *sb)
14103 +{
14104 +       return sb->s_magic == RAMFS_MAGIC;
14105 +}
14106 +
14107 +static inline int au_test_ubifs(struct super_block *sb __maybe_unused)
14108 +{
14109 +#if defined(CONFIG_UBIFS_FS) || defined(CONFIG_UBIFS_FS_MODULE)
14110 +       return sb->s_magic == UBIFS_SUPER_MAGIC;
14111 +#else
14112 +       return 0;
14113 +#endif
14114 +}
14115 +
14116 +static inline int au_test_procfs(struct super_block *sb __maybe_unused)
14117 +{
14118 +#ifdef CONFIG_PROC_FS
14119 +       return sb->s_magic == PROC_SUPER_MAGIC;
14120 +#else
14121 +       return 0;
14122 +#endif
14123 +}
14124 +
14125 +static inline int au_test_sysfs(struct super_block *sb __maybe_unused)
14126 +{
14127 +#ifdef CONFIG_SYSFS
14128 +       return sb->s_magic == SYSFS_MAGIC;
14129 +#else
14130 +       return 0;
14131 +#endif
14132 +}
14133 +
14134 +static inline int au_test_configfs(struct super_block *sb __maybe_unused)
14135 +{
14136 +#if defined(CONFIG_CONFIGFS_FS) || defined(CONFIG_CONFIGFS_FS_MODULE)
14137 +       return sb->s_magic == CONFIGFS_MAGIC;
14138 +#else
14139 +       return 0;
14140 +#endif
14141 +}
14142 +
14143 +static inline int au_test_minix(struct super_block *sb __maybe_unused)
14144 +{
14145 +#if defined(CONFIG_MINIX_FS) || defined(CONFIG_MINIX_FS_MODULE)
14146 +       return sb->s_magic == MINIX3_SUPER_MAGIC
14147 +               || sb->s_magic == MINIX2_SUPER_MAGIC
14148 +               || sb->s_magic == MINIX2_SUPER_MAGIC2
14149 +               || sb->s_magic == MINIX_SUPER_MAGIC
14150 +               || sb->s_magic == MINIX_SUPER_MAGIC2;
14151 +#else
14152 +       return 0;
14153 +#endif
14154 +}
14155 +
14156 +static inline int au_test_cifs(struct super_block *sb __maybe_unused)
14157 +{
14158 +#if defined(CONFIG_CIFS_FS) || defined(CONFIGCIFS_FS_MODULE)
14159 +       return sb->s_magic == CIFS_MAGIC_NUMBER;
14160 +#else
14161 +       return 0;
14162 +#endif
14163 +}
14164 +
14165 +static inline int au_test_fat(struct super_block *sb __maybe_unused)
14166 +{
14167 +#if defined(CONFIG_FAT_FS) || defined(CONFIG_FAT_FS_MODULE)
14168 +       return sb->s_magic == MSDOS_SUPER_MAGIC;
14169 +#else
14170 +       return 0;
14171 +#endif
14172 +}
14173 +
14174 +static inline int au_test_msdos(struct super_block *sb)
14175 +{
14176 +       return au_test_fat(sb);
14177 +}
14178 +
14179 +static inline int au_test_vfat(struct super_block *sb)
14180 +{
14181 +       return au_test_fat(sb);
14182 +}
14183 +
14184 +static inline int au_test_securityfs(struct super_block *sb __maybe_unused)
14185 +{
14186 +#ifdef CONFIG_SECURITYFS
14187 +       return sb->s_magic == SECURITYFS_MAGIC;
14188 +#else
14189 +       return 0;
14190 +#endif
14191 +}
14192 +
14193 +static inline int au_test_squashfs(struct super_block *sb __maybe_unused)
14194 +{
14195 +#if defined(CONFIG_SQUASHFS) || defined(CONFIG_SQUASHFS_MODULE)
14196 +       return sb->s_magic == SQUASHFS_MAGIC;
14197 +#else
14198 +       return 0;
14199 +#endif
14200 +}
14201 +
14202 +static inline int au_test_btrfs(struct super_block *sb __maybe_unused)
14203 +{
14204 +#if defined(CONFIG_BTRFS_FS) || defined(CONFIG_BTRFS_FS_MODULE)
14205 +       return sb->s_magic == BTRFS_SUPER_MAGIC;
14206 +#else
14207 +       return 0;
14208 +#endif
14209 +}
14210 +
14211 +static inline int au_test_xenfs(struct super_block *sb __maybe_unused)
14212 +{
14213 +#if defined(CONFIG_XENFS) || defined(CONFIG_XENFS_MODULE)
14214 +       return sb->s_magic == XENFS_SUPER_MAGIC;
14215 +#else
14216 +       return 0;
14217 +#endif
14218 +}
14219 +
14220 +static inline int au_test_debugfs(struct super_block *sb __maybe_unused)
14221 +{
14222 +#ifdef CONFIG_DEBUG_FS
14223 +       return sb->s_magic == DEBUGFS_MAGIC;
14224 +#else
14225 +       return 0;
14226 +#endif
14227 +}
14228 +
14229 +static inline int au_test_nilfs(struct super_block *sb __maybe_unused)
14230 +{
14231 +#if defined(CONFIG_NILFS) || defined(CONFIG_NILFS_MODULE)
14232 +       return sb->s_magic == NILFS_SUPER_MAGIC;
14233 +#else
14234 +       return 0;
14235 +#endif
14236 +}
14237 +
14238 +static inline int au_test_hfsplus(struct super_block *sb __maybe_unused)
14239 +{
14240 +#if defined(CONFIG_HFSPLUS_FS) || defined(CONFIG_HFSPLUS_FS_MODULE)
14241 +       return sb->s_magic == HFSPLUS_SUPER_MAGIC;
14242 +#else
14243 +       return 0;
14244 +#endif
14245 +}
14246 +
14247 +/* ---------------------------------------------------------------------- */
14248 +/*
14249 + * they can't be an aufs branch.
14250 + */
14251 +static inline int au_test_fs_unsuppoted(struct super_block *sb)
14252 +{
14253 +       return
14254 +#ifndef CONFIG_AUFS_BR_RAMFS
14255 +               au_test_ramfs(sb) ||
14256 +#endif
14257 +               au_test_procfs(sb)
14258 +               || au_test_sysfs(sb)
14259 +               || au_test_configfs(sb)
14260 +               || au_test_debugfs(sb)
14261 +               || au_test_securityfs(sb)
14262 +               || au_test_xenfs(sb)
14263 +               || au_test_ecryptfs(sb)
14264 +               /* || !strcmp(au_sbtype(sb), "unionfs") */
14265 +               || au_test_aufs(sb); /* will be supported in next version */
14266 +}
14267 +
14268 +static inline int au_test_fs_remote(struct super_block *sb)
14269 +{
14270 +       return !au_test_tmpfs(sb)
14271 +#ifdef CONFIG_AUFS_BR_RAMFS
14272 +               && !au_test_ramfs(sb)
14273 +#endif
14274 +               && !(sb->s_type->fs_flags & FS_REQUIRES_DEV);
14275 +}
14276 +
14277 +/* ---------------------------------------------------------------------- */
14278 +
14279 +/*
14280 + * Note: these functions (below) are created after reading ->getattr() in all
14281 + * filesystems under linux/fs. it means we have to do so in every update...
14282 + */
14283 +
14284 +/*
14285 + * some filesystems require getattr to refresh the inode attributes before
14286 + * referencing.
14287 + * in most cases, we can rely on the inode attribute in NFS (or every remote fs)
14288 + * and leave the work for d_revalidate()
14289 + */
14290 +static inline int au_test_fs_refresh_iattr(struct super_block *sb)
14291 +{
14292 +       return au_test_nfs(sb)
14293 +               || au_test_fuse(sb)
14294 +               /* || au_test_ocfs2(sb) */      /* untested */
14295 +               /* || au_test_btrfs(sb) */      /* untested */
14296 +               /* || au_test_coda(sb) */       /* untested */
14297 +               /* || au_test_v9fs(sb) */       /* untested */
14298 +               ;
14299 +}
14300 +
14301 +/*
14302 + * filesystems which don't maintain i_size or i_blocks.
14303 + */
14304 +static inline int au_test_fs_bad_iattr_size(struct super_block *sb)
14305 +{
14306 +       return au_test_xfs(sb)
14307 +               || au_test_btrfs(sb)
14308 +               || au_test_ubifs(sb)
14309 +               || au_test_hfsplus(sb)  /* maintained, but incorrect */
14310 +               /* || au_test_ext4(sb) */       /* untested */
14311 +               /* || au_test_ocfs2(sb) */      /* untested */
14312 +               /* || au_test_ocfs2_dlmfs(sb) */ /* untested */
14313 +               /* || au_test_sysv(sb) */       /* untested */
14314 +               /* || au_test_minix(sb) */      /* untested */
14315 +               ;
14316 +}
14317 +
14318 +/*
14319 + * filesystems which don't store the correct value in some of their inode
14320 + * attributes.
14321 + */
14322 +static inline int au_test_fs_bad_iattr(struct super_block *sb)
14323 +{
14324 +       return au_test_fs_bad_iattr_size(sb)
14325 +               /* || au_test_cifs(sb) */       /* untested */
14326 +               || au_test_fat(sb)
14327 +               || au_test_msdos(sb)
14328 +               || au_test_vfat(sb);
14329 +}
14330 +
14331 +/* they don't check i_nlink in link(2) */
14332 +static inline int au_test_fs_no_limit_nlink(struct super_block *sb)
14333 +{
14334 +       return au_test_tmpfs(sb)
14335 +#ifdef CONFIG_AUFS_BR_RAMFS
14336 +               || au_test_ramfs(sb)
14337 +#endif
14338 +               || au_test_ubifs(sb)
14339 +               || au_test_hfsplus(sb);
14340 +}
14341 +
14342 +/*
14343 + * filesystems which sets S_NOATIME and S_NOCMTIME.
14344 + */
14345 +static inline int au_test_fs_notime(struct super_block *sb)
14346 +{
14347 +       return au_test_nfs(sb)
14348 +               || au_test_fuse(sb)
14349 +               || au_test_ubifs(sb)
14350 +               /* || au_test_cifs(sb) */       /* untested */
14351 +               ;
14352 +}
14353 +
14354 +/*
14355 + * filesystems which requires replacing i_mapping.
14356 + */
14357 +static inline int au_test_fs_bad_mapping(struct super_block *sb)
14358 +{
14359 +       return au_test_fuse(sb)
14360 +               || au_test_ubifs(sb);
14361 +}
14362 +
14363 +/* temporary support for i#1 in cramfs */
14364 +static inline int au_test_fs_unique_ino(struct inode *inode)
14365 +{
14366 +       if (au_test_cramfs(inode->i_sb))
14367 +               return inode->i_ino != 1;
14368 +       return 1;
14369 +}
14370 +
14371 +/* ---------------------------------------------------------------------- */
14372 +
14373 +/*
14374 + * the filesystem where the xino files placed must support i/o after unlink and
14375 + * maintain i_size and i_blocks.
14376 + */
14377 +static inline int au_test_fs_bad_xino(struct super_block *sb)
14378 +{
14379 +       return au_test_fs_remote(sb)
14380 +               || au_test_fs_bad_iattr_size(sb)
14381 +               /* don't want unnecessary work for xino */
14382 +               || au_test_aufs(sb)
14383 +               || au_test_ecryptfs(sb)
14384 +               || au_test_nilfs(sb);
14385 +}
14386 +
14387 +static inline int au_test_fs_trunc_xino(struct super_block *sb)
14388 +{
14389 +       return au_test_tmpfs(sb)
14390 +               || au_test_ramfs(sb);
14391 +}
14392 +
14393 +/*
14394 + * test if the @sb is real-readonly.
14395 + */
14396 +static inline int au_test_fs_rr(struct super_block *sb)
14397 +{
14398 +       return au_test_squashfs(sb)
14399 +               || au_test_iso9660(sb)
14400 +               || au_test_cramfs(sb)
14401 +               || au_test_romfs(sb);
14402 +}
14403 +
14404 +#endif /* __KERNEL__ */
14405 +#endif /* __AUFS_FSTYPE_H__ */
14406 diff -urN /usr/share/empty/fs/aufs/hfsnotify.c linux/fs/aufs/hfsnotify.c
14407 --- /usr/share/empty/fs/aufs/hfsnotify.c        1970-01-01 01:00:00.000000000 +0100
14408 +++ linux/fs/aufs/hfsnotify.c   2015-03-27 21:56:35.463461668 +0100
14409 @@ -0,0 +1,288 @@
14410 +/*
14411 + * Copyright (C) 2005-2015 Junjiro R. Okajima
14412 + *
14413 + * This program, aufs is free software; you can redistribute it and/or modify
14414 + * it under the terms of the GNU General Public License as published by
14415 + * the Free Software Foundation; either version 2 of the License, or
14416 + * (at your option) any later version.
14417 + *
14418 + * This program is distributed in the hope that it will be useful,
14419 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14420 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14421 + * GNU General Public License for more details.
14422 + *
14423 + * You should have received a copy of the GNU General Public License
14424 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14425 + */
14426 +
14427 +/*
14428 + * fsnotify for the lower directories
14429 + */
14430 +
14431 +#include "aufs.h"
14432 +
14433 +/* FS_IN_IGNORED is unnecessary */
14434 +static const __u32 AuHfsnMask = (FS_MOVED_TO | FS_MOVED_FROM | FS_DELETE
14435 +                                | FS_CREATE | FS_EVENT_ON_CHILD);
14436 +static DECLARE_WAIT_QUEUE_HEAD(au_hfsn_wq);
14437 +static __cacheline_aligned_in_smp atomic64_t au_hfsn_ifree = ATOMIC64_INIT(0);
14438 +
14439 +static void au_hfsn_free_mark(struct fsnotify_mark *mark)
14440 +{
14441 +       struct au_hnotify *hn = container_of(mark, struct au_hnotify,
14442 +                                            hn_mark);
14443 +       AuDbg("here\n");
14444 +       au_cache_free_hnotify(hn);
14445 +       smp_mb__before_atomic();
14446 +       if (atomic64_dec_and_test(&au_hfsn_ifree))
14447 +               wake_up(&au_hfsn_wq);
14448 +}
14449 +
14450 +static int au_hfsn_alloc(struct au_hinode *hinode)
14451 +{
14452 +       int err;
14453 +       struct au_hnotify *hn;
14454 +       struct super_block *sb;
14455 +       struct au_branch *br;
14456 +       struct fsnotify_mark *mark;
14457 +       aufs_bindex_t bindex;
14458 +
14459 +       hn = hinode->hi_notify;
14460 +       sb = hn->hn_aufs_inode->i_sb;
14461 +       bindex = au_br_index(sb, hinode->hi_id);
14462 +       br = au_sbr(sb, bindex);
14463 +       AuDebugOn(!br->br_hfsn);
14464 +
14465 +       mark = &hn->hn_mark;
14466 +       fsnotify_init_mark(mark, au_hfsn_free_mark);
14467 +       mark->mask = AuHfsnMask;
14468 +       /*
14469 +        * by udba rename or rmdir, aufs assign a new inode to the known
14470 +        * h_inode, so specify 1 to allow dups.
14471 +        */
14472 +       lockdep_off();
14473 +       err = fsnotify_add_mark(mark, br->br_hfsn->hfsn_group, hinode->hi_inode,
14474 +                                /*mnt*/NULL, /*allow_dups*/1);
14475 +       /* even if err */
14476 +       fsnotify_put_mark(mark);
14477 +       lockdep_on();
14478 +
14479 +       return err;
14480 +}
14481 +
14482 +static int au_hfsn_free(struct au_hinode *hinode, struct au_hnotify *hn)
14483 +{
14484 +       struct fsnotify_mark *mark;
14485 +       unsigned long long ull;
14486 +       struct fsnotify_group *group;
14487 +
14488 +       ull = atomic64_inc_return(&au_hfsn_ifree);
14489 +       BUG_ON(!ull);
14490 +
14491 +       mark = &hn->hn_mark;
14492 +       spin_lock(&mark->lock);
14493 +       group = mark->group;
14494 +       fsnotify_get_group(group);
14495 +       spin_unlock(&mark->lock);
14496 +       lockdep_off();
14497 +       fsnotify_destroy_mark(mark, group);
14498 +       fsnotify_put_group(group);
14499 +       lockdep_on();
14500 +
14501 +       /* free hn by myself */
14502 +       return 0;
14503 +}
14504 +
14505 +/* ---------------------------------------------------------------------- */
14506 +
14507 +static void au_hfsn_ctl(struct au_hinode *hinode, int do_set)
14508 +{
14509 +       struct fsnotify_mark *mark;
14510 +
14511 +       mark = &hinode->hi_notify->hn_mark;
14512 +       spin_lock(&mark->lock);
14513 +       if (do_set) {
14514 +               AuDebugOn(mark->mask & AuHfsnMask);
14515 +               mark->mask |= AuHfsnMask;
14516 +       } else {
14517 +               AuDebugOn(!(mark->mask & AuHfsnMask));
14518 +               mark->mask &= ~AuHfsnMask;
14519 +       }
14520 +       spin_unlock(&mark->lock);
14521 +       /* fsnotify_recalc_inode_mask(hinode->hi_inode); */
14522 +}
14523 +
14524 +/* ---------------------------------------------------------------------- */
14525 +
14526 +/* #define AuDbgHnotify */
14527 +#ifdef AuDbgHnotify
14528 +static char *au_hfsn_name(u32 mask)
14529 +{
14530 +#ifdef CONFIG_AUFS_DEBUG
14531 +#define test_ret(flag)                         \
14532 +       do {                                    \
14533 +               if (mask & flag)                \
14534 +                       return #flag;           \
14535 +       } while (0)
14536 +       test_ret(FS_ACCESS);
14537 +       test_ret(FS_MODIFY);
14538 +       test_ret(FS_ATTRIB);
14539 +       test_ret(FS_CLOSE_WRITE);
14540 +       test_ret(FS_CLOSE_NOWRITE);
14541 +       test_ret(FS_OPEN);
14542 +       test_ret(FS_MOVED_FROM);
14543 +       test_ret(FS_MOVED_TO);
14544 +       test_ret(FS_CREATE);
14545 +       test_ret(FS_DELETE);
14546 +       test_ret(FS_DELETE_SELF);
14547 +       test_ret(FS_MOVE_SELF);
14548 +       test_ret(FS_UNMOUNT);
14549 +       test_ret(FS_Q_OVERFLOW);
14550 +       test_ret(FS_IN_IGNORED);
14551 +       test_ret(FS_IN_ISDIR);
14552 +       test_ret(FS_IN_ONESHOT);
14553 +       test_ret(FS_EVENT_ON_CHILD);
14554 +       return "";
14555 +#undef test_ret
14556 +#else
14557 +       return "??";
14558 +#endif
14559 +}
14560 +#endif
14561 +
14562 +/* ---------------------------------------------------------------------- */
14563 +
14564 +static void au_hfsn_free_group(struct fsnotify_group *group)
14565 +{
14566 +       struct au_br_hfsnotify *hfsn = group->private;
14567 +
14568 +       AuDbg("here\n");
14569 +       kfree(hfsn);
14570 +}
14571 +
14572 +static int au_hfsn_handle_event(struct fsnotify_group *group,
14573 +                               struct inode *inode,
14574 +                               struct fsnotify_mark *inode_mark,
14575 +                               struct fsnotify_mark *vfsmount_mark,
14576 +                               u32 mask, void *data, int data_type,
14577 +                               const unsigned char *file_name, u32 cookie)
14578 +{
14579 +       int err;
14580 +       struct au_hnotify *hnotify;
14581 +       struct inode *h_dir, *h_inode;
14582 +       struct qstr h_child_qstr = QSTR_INIT(file_name, strlen(file_name));
14583 +
14584 +       AuDebugOn(data_type != FSNOTIFY_EVENT_INODE);
14585 +
14586 +       err = 0;
14587 +       /* if FS_UNMOUNT happens, there must be another bug */
14588 +       AuDebugOn(mask & FS_UNMOUNT);
14589 +       if (mask & (FS_IN_IGNORED | FS_UNMOUNT))
14590 +               goto out;
14591 +
14592 +       h_dir = inode;
14593 +       h_inode = NULL;
14594 +#ifdef AuDbgHnotify
14595 +       au_debug_on();
14596 +       if (1 || h_child_qstr.len != sizeof(AUFS_XINO_FNAME) - 1
14597 +           || strncmp(h_child_qstr.name, AUFS_XINO_FNAME, h_child_qstr.len)) {
14598 +               AuDbg("i%lu, mask 0x%x %s, hcname %.*s, hi%lu\n",
14599 +                     h_dir->i_ino, mask, au_hfsn_name(mask),
14600 +                     AuLNPair(&h_child_qstr), h_inode ? h_inode->i_ino : 0);
14601 +               /* WARN_ON(1); */
14602 +       }
14603 +       au_debug_off();
14604 +#endif
14605 +
14606 +       AuDebugOn(!inode_mark);
14607 +       hnotify = container_of(inode_mark, struct au_hnotify, hn_mark);
14608 +       err = au_hnotify(h_dir, hnotify, mask, &h_child_qstr, h_inode);
14609 +
14610 +out:
14611 +       return err;
14612 +}
14613 +
14614 +static struct fsnotify_ops au_hfsn_ops = {
14615 +       .handle_event           = au_hfsn_handle_event,
14616 +       .free_group_priv        = au_hfsn_free_group
14617 +};
14618 +
14619 +/* ---------------------------------------------------------------------- */
14620 +
14621 +static void au_hfsn_fin_br(struct au_branch *br)
14622 +{
14623 +       struct au_br_hfsnotify *hfsn;
14624 +
14625 +       hfsn = br->br_hfsn;
14626 +       if (hfsn) {
14627 +               lockdep_off();
14628 +               fsnotify_put_group(hfsn->hfsn_group);
14629 +               lockdep_on();
14630 +       }
14631 +}
14632 +
14633 +static int au_hfsn_init_br(struct au_branch *br, int perm)
14634 +{
14635 +       int err;
14636 +       struct fsnotify_group *group;
14637 +       struct au_br_hfsnotify *hfsn;
14638 +
14639 +       err = 0;
14640 +       br->br_hfsn = NULL;
14641 +       if (!au_br_hnotifyable(perm))
14642 +               goto out;
14643 +
14644 +       err = -ENOMEM;
14645 +       hfsn = kmalloc(sizeof(*hfsn), GFP_NOFS);
14646 +       if (unlikely(!hfsn))
14647 +               goto out;
14648 +
14649 +       err = 0;
14650 +       group = fsnotify_alloc_group(&au_hfsn_ops);
14651 +       if (IS_ERR(group)) {
14652 +               err = PTR_ERR(group);
14653 +               pr_err("fsnotify_alloc_group() failed, %d\n", err);
14654 +               goto out_hfsn;
14655 +       }
14656 +
14657 +       group->private = hfsn;
14658 +       hfsn->hfsn_group = group;
14659 +       br->br_hfsn = hfsn;
14660 +       goto out; /* success */
14661 +
14662 +out_hfsn:
14663 +       kfree(hfsn);
14664 +out:
14665 +       return err;
14666 +}
14667 +
14668 +static int au_hfsn_reset_br(unsigned int udba, struct au_branch *br, int perm)
14669 +{
14670 +       int err;
14671 +
14672 +       err = 0;
14673 +       if (!br->br_hfsn)
14674 +               err = au_hfsn_init_br(br, perm);
14675 +
14676 +       return err;
14677 +}
14678 +
14679 +/* ---------------------------------------------------------------------- */
14680 +
14681 +static void au_hfsn_fin(void)
14682 +{
14683 +       AuDbg("au_hfsn_ifree %lld\n", (long long)atomic64_read(&au_hfsn_ifree));
14684 +       wait_event(au_hfsn_wq, !atomic64_read(&au_hfsn_ifree));
14685 +}
14686 +
14687 +const struct au_hnotify_op au_hnotify_op = {
14688 +       .ctl            = au_hfsn_ctl,
14689 +       .alloc          = au_hfsn_alloc,
14690 +       .free           = au_hfsn_free,
14691 +
14692 +       .fin            = au_hfsn_fin,
14693 +
14694 +       .reset_br       = au_hfsn_reset_br,
14695 +       .fin_br         = au_hfsn_fin_br,
14696 +       .init_br        = au_hfsn_init_br
14697 +};
14698 diff -urN /usr/share/empty/fs/aufs/hfsplus.c linux/fs/aufs/hfsplus.c
14699 --- /usr/share/empty/fs/aufs/hfsplus.c  1970-01-01 01:00:00.000000000 +0100
14700 +++ linux/fs/aufs/hfsplus.c     2015-03-27 21:56:35.463461668 +0100
14701 @@ -0,0 +1,56 @@
14702 +/*
14703 + * Copyright (C) 2010-2015 Junjiro R. Okajima
14704 + *
14705 + * This program, aufs is free software; you can redistribute it and/or modify
14706 + * it under the terms of the GNU General Public License as published by
14707 + * the Free Software Foundation; either version 2 of the License, or
14708 + * (at your option) any later version.
14709 + *
14710 + * This program is distributed in the hope that it will be useful,
14711 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14712 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14713 + * GNU General Public License for more details.
14714 + *
14715 + * You should have received a copy of the GNU General Public License
14716 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14717 + */
14718 +
14719 +/*
14720 + * special support for filesystems which aqucires an inode mutex
14721 + * at final closing a file, eg, hfsplus.
14722 + *
14723 + * This trick is very simple and stupid, just to open the file before really
14724 + * neceeary open to tell hfsplus that this is not the final closing.
14725 + * The caller should call au_h_open_pre() after acquiring the inode mutex,
14726 + * and au_h_open_post() after releasing it.
14727 + */
14728 +
14729 +#include "aufs.h"
14730 +
14731 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
14732 +                          int force_wr)
14733 +{
14734 +       struct file *h_file;
14735 +       struct dentry *h_dentry;
14736 +
14737 +       h_dentry = au_h_dptr(dentry, bindex);
14738 +       AuDebugOn(!h_dentry);
14739 +       AuDebugOn(!h_dentry->d_inode);
14740 +
14741 +       h_file = NULL;
14742 +       if (au_test_hfsplus(h_dentry->d_sb)
14743 +           && S_ISREG(h_dentry->d_inode->i_mode))
14744 +               h_file = au_h_open(dentry, bindex,
14745 +                                  O_RDONLY | O_NOATIME | O_LARGEFILE,
14746 +                                  /*file*/NULL, force_wr);
14747 +       return h_file;
14748 +}
14749 +
14750 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
14751 +                   struct file *h_file)
14752 +{
14753 +       if (h_file) {
14754 +               fput(h_file);
14755 +               au_sbr_put(dentry->d_sb, bindex);
14756 +       }
14757 +}
14758 diff -urN /usr/share/empty/fs/aufs/hnotify.c linux/fs/aufs/hnotify.c
14759 --- /usr/share/empty/fs/aufs/hnotify.c  1970-01-01 01:00:00.000000000 +0100
14760 +++ linux/fs/aufs/hnotify.c     2015-03-27 21:56:35.463461668 +0100
14761 @@ -0,0 +1,714 @@
14762 +/*
14763 + * Copyright (C) 2005-2015 Junjiro R. Okajima
14764 + *
14765 + * This program, aufs is free software; you can redistribute it and/or modify
14766 + * it under the terms of the GNU General Public License as published by
14767 + * the Free Software Foundation; either version 2 of the License, or
14768 + * (at your option) any later version.
14769 + *
14770 + * This program is distributed in the hope that it will be useful,
14771 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14772 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14773 + * GNU General Public License for more details.
14774 + *
14775 + * You should have received a copy of the GNU General Public License
14776 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14777 + */
14778 +
14779 +/*
14780 + * abstraction to notify the direct changes on lower directories
14781 + */
14782 +
14783 +#include "aufs.h"
14784 +
14785 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode)
14786 +{
14787 +       int err;
14788 +       struct au_hnotify *hn;
14789 +
14790 +       err = -ENOMEM;
14791 +       hn = au_cache_alloc_hnotify();
14792 +       if (hn) {
14793 +               hn->hn_aufs_inode = inode;
14794 +               hinode->hi_notify = hn;
14795 +               err = au_hnotify_op.alloc(hinode);
14796 +               AuTraceErr(err);
14797 +               if (unlikely(err)) {
14798 +                       hinode->hi_notify = NULL;
14799 +                       au_cache_free_hnotify(hn);
14800 +                       /*
14801 +                        * The upper dir was removed by udba, but the same named
14802 +                        * dir left. In this case, aufs assignes a new inode
14803 +                        * number and set the monitor again.
14804 +                        * For the lower dir, the old monitnor is still left.
14805 +                        */
14806 +                       if (err == -EEXIST)
14807 +                               err = 0;
14808 +               }
14809 +       }
14810 +
14811 +       AuTraceErr(err);
14812 +       return err;
14813 +}
14814 +
14815 +void au_hn_free(struct au_hinode *hinode)
14816 +{
14817 +       struct au_hnotify *hn;
14818 +
14819 +       hn = hinode->hi_notify;
14820 +       if (hn) {
14821 +               hinode->hi_notify = NULL;
14822 +               if (au_hnotify_op.free(hinode, hn))
14823 +                       au_cache_free_hnotify(hn);
14824 +       }
14825 +}
14826 +
14827 +/* ---------------------------------------------------------------------- */
14828 +
14829 +void au_hn_ctl(struct au_hinode *hinode, int do_set)
14830 +{
14831 +       if (hinode->hi_notify)
14832 +               au_hnotify_op.ctl(hinode, do_set);
14833 +}
14834 +
14835 +void au_hn_reset(struct inode *inode, unsigned int flags)
14836 +{
14837 +       aufs_bindex_t bindex, bend;
14838 +       struct inode *hi;
14839 +       struct dentry *iwhdentry;
14840 +
14841 +       bend = au_ibend(inode);
14842 +       for (bindex = au_ibstart(inode); bindex <= bend; bindex++) {
14843 +               hi = au_h_iptr(inode, bindex);
14844 +               if (!hi)
14845 +                       continue;
14846 +
14847 +               /* mutex_lock_nested(&hi->i_mutex, AuLsc_I_CHILD); */
14848 +               iwhdentry = au_hi_wh(inode, bindex);
14849 +               if (iwhdentry)
14850 +                       dget(iwhdentry);
14851 +               au_igrab(hi);
14852 +               au_set_h_iptr(inode, bindex, NULL, 0);
14853 +               au_set_h_iptr(inode, bindex, au_igrab(hi),
14854 +                             flags & ~AuHi_XINO);
14855 +               iput(hi);
14856 +               dput(iwhdentry);
14857 +               /* mutex_unlock(&hi->i_mutex); */
14858 +       }
14859 +}
14860 +
14861 +/* ---------------------------------------------------------------------- */
14862 +
14863 +static int hn_xino(struct inode *inode, struct inode *h_inode)
14864 +{
14865 +       int err;
14866 +       aufs_bindex_t bindex, bend, bfound, bstart;
14867 +       struct inode *h_i;
14868 +
14869 +       err = 0;
14870 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
14871 +               pr_warn("branch root dir was changed\n");
14872 +               goto out;
14873 +       }
14874 +
14875 +       bfound = -1;
14876 +       bend = au_ibend(inode);
14877 +       bstart = au_ibstart(inode);
14878 +#if 0 /* reserved for future use */
14879 +       if (bindex == bend) {
14880 +               /* keep this ino in rename case */
14881 +               goto out;
14882 +       }
14883 +#endif
14884 +       for (bindex = bstart; bindex <= bend; bindex++)
14885 +               if (au_h_iptr(inode, bindex) == h_inode) {
14886 +                       bfound = bindex;
14887 +                       break;
14888 +               }
14889 +       if (bfound < 0)
14890 +               goto out;
14891 +
14892 +       for (bindex = bstart; bindex <= bend; bindex++) {
14893 +               h_i = au_h_iptr(inode, bindex);
14894 +               if (!h_i)
14895 +                       continue;
14896 +
14897 +               err = au_xino_write(inode->i_sb, bindex, h_i->i_ino, /*ino*/0);
14898 +               /* ignore this error */
14899 +               /* bad action? */
14900 +       }
14901 +
14902 +       /* children inode number will be broken */
14903 +
14904 +out:
14905 +       AuTraceErr(err);
14906 +       return err;
14907 +}
14908 +
14909 +static int hn_gen_tree(struct dentry *dentry)
14910 +{
14911 +       int err, i, j, ndentry;
14912 +       struct au_dcsub_pages dpages;
14913 +       struct au_dpage *dpage;
14914 +       struct dentry **dentries;
14915 +
14916 +       err = au_dpages_init(&dpages, GFP_NOFS);
14917 +       if (unlikely(err))
14918 +               goto out;
14919 +       err = au_dcsub_pages(&dpages, dentry, NULL, NULL);
14920 +       if (unlikely(err))
14921 +               goto out_dpages;
14922 +
14923 +       for (i = 0; i < dpages.ndpage; i++) {
14924 +               dpage = dpages.dpages + i;
14925 +               dentries = dpage->dentries;
14926 +               ndentry = dpage->ndentry;
14927 +               for (j = 0; j < ndentry; j++) {
14928 +                       struct dentry *d;
14929 +
14930 +                       d = dentries[j];
14931 +                       if (IS_ROOT(d))
14932 +                               continue;
14933 +
14934 +                       au_digen_dec(d);
14935 +                       if (d->d_inode)
14936 +                               /* todo: reset children xino?
14937 +                                  cached children only? */
14938 +                               au_iigen_dec(d->d_inode);
14939 +               }
14940 +       }
14941 +
14942 +out_dpages:
14943 +       au_dpages_free(&dpages);
14944 +
14945 +#if 0
14946 +       /* discard children */
14947 +       dentry_unhash(dentry);
14948 +       dput(dentry);
14949 +#endif
14950 +out:
14951 +       return err;
14952 +}
14953 +
14954 +/*
14955 + * return 0 if processed.
14956 + */
14957 +static int hn_gen_by_inode(char *name, unsigned int nlen, struct inode *inode,
14958 +                          const unsigned int isdir)
14959 +{
14960 +       int err;
14961 +       struct dentry *d;
14962 +       struct qstr *dname;
14963 +
14964 +       err = 1;
14965 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
14966 +               pr_warn("branch root dir was changed\n");
14967 +               err = 0;
14968 +               goto out;
14969 +       }
14970 +
14971 +       if (!isdir) {
14972 +               AuDebugOn(!name);
14973 +               au_iigen_dec(inode);
14974 +               spin_lock(&inode->i_lock);
14975 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
14976 +                       spin_lock(&d->d_lock);
14977 +                       dname = &d->d_name;
14978 +                       if (dname->len != nlen
14979 +                           && memcmp(dname->name, name, nlen)) {
14980 +                               spin_unlock(&d->d_lock);
14981 +                               continue;
14982 +                       }
14983 +                       err = 0;
14984 +                       au_digen_dec(d);
14985 +                       spin_unlock(&d->d_lock);
14986 +                       break;
14987 +               }
14988 +               spin_unlock(&inode->i_lock);
14989 +       } else {
14990 +               au_fset_si(au_sbi(inode->i_sb), FAILED_REFRESH_DIR);
14991 +               d = d_find_any_alias(inode);
14992 +               if (!d) {
14993 +                       au_iigen_dec(inode);
14994 +                       goto out;
14995 +               }
14996 +
14997 +               spin_lock(&d->d_lock);
14998 +               dname = &d->d_name;
14999 +               if (dname->len == nlen && !memcmp(dname->name, name, nlen)) {
15000 +                       spin_unlock(&d->d_lock);
15001 +                       err = hn_gen_tree(d);
15002 +                       spin_lock(&d->d_lock);
15003 +               }
15004 +               spin_unlock(&d->d_lock);
15005 +               dput(d);
15006 +       }
15007 +
15008 +out:
15009 +       AuTraceErr(err);
15010 +       return err;
15011 +}
15012 +
15013 +static int hn_gen_by_name(struct dentry *dentry, const unsigned int isdir)
15014 +{
15015 +       int err;
15016 +       struct inode *inode;
15017 +
15018 +       inode = dentry->d_inode;
15019 +       if (IS_ROOT(dentry)
15020 +           /* || (inode && inode->i_ino == AUFS_ROOT_INO) */
15021 +               ) {
15022 +               pr_warn("branch root dir was changed\n");
15023 +               return 0;
15024 +       }
15025 +
15026 +       err = 0;
15027 +       if (!isdir) {
15028 +               au_digen_dec(dentry);
15029 +               if (inode)
15030 +                       au_iigen_dec(inode);
15031 +       } else {
15032 +               au_fset_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR);
15033 +               if (inode)
15034 +                       err = hn_gen_tree(dentry);
15035 +       }
15036 +
15037 +       AuTraceErr(err);
15038 +       return err;
15039 +}
15040 +
15041 +/* ---------------------------------------------------------------------- */
15042 +
15043 +/* hnotify job flags */
15044 +#define AuHnJob_XINO0          1
15045 +#define AuHnJob_GEN            (1 << 1)
15046 +#define AuHnJob_DIRENT         (1 << 2)
15047 +#define AuHnJob_ISDIR          (1 << 3)
15048 +#define AuHnJob_TRYXINO0       (1 << 4)
15049 +#define AuHnJob_MNTPNT         (1 << 5)
15050 +#define au_ftest_hnjob(flags, name)    ((flags) & AuHnJob_##name)
15051 +#define au_fset_hnjob(flags, name) \
15052 +       do { (flags) |= AuHnJob_##name; } while (0)
15053 +#define au_fclr_hnjob(flags, name) \
15054 +       do { (flags) &= ~AuHnJob_##name; } while (0)
15055 +
15056 +enum {
15057 +       AuHn_CHILD,
15058 +       AuHn_PARENT,
15059 +       AuHnLast
15060 +};
15061 +
15062 +struct au_hnotify_args {
15063 +       struct inode *h_dir, *dir, *h_child_inode;
15064 +       u32 mask;
15065 +       unsigned int flags[AuHnLast];
15066 +       unsigned int h_child_nlen;
15067 +       char h_child_name[];
15068 +};
15069 +
15070 +struct hn_job_args {
15071 +       unsigned int flags;
15072 +       struct inode *inode, *h_inode, *dir, *h_dir;
15073 +       struct dentry *dentry;
15074 +       char *h_name;
15075 +       int h_nlen;
15076 +};
15077 +
15078 +static int hn_job(struct hn_job_args *a)
15079 +{
15080 +       const unsigned int isdir = au_ftest_hnjob(a->flags, ISDIR);
15081 +       int e;
15082 +
15083 +       /* reset xino */
15084 +       if (au_ftest_hnjob(a->flags, XINO0) && a->inode)
15085 +               hn_xino(a->inode, a->h_inode); /* ignore this error */
15086 +
15087 +       if (au_ftest_hnjob(a->flags, TRYXINO0)
15088 +           && a->inode
15089 +           && a->h_inode) {
15090 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
15091 +               if (!a->h_inode->i_nlink
15092 +                   && !(a->h_inode->i_state & I_LINKABLE))
15093 +                       hn_xino(a->inode, a->h_inode); /* ignore this error */
15094 +               mutex_unlock(&a->h_inode->i_mutex);
15095 +       }
15096 +
15097 +       /* make the generation obsolete */
15098 +       if (au_ftest_hnjob(a->flags, GEN)) {
15099 +               e = -1;
15100 +               if (a->inode)
15101 +                       e = hn_gen_by_inode(a->h_name, a->h_nlen, a->inode,
15102 +                                             isdir);
15103 +               if (e && a->dentry)
15104 +                       hn_gen_by_name(a->dentry, isdir);
15105 +               /* ignore this error */
15106 +       }
15107 +
15108 +       /* make dir entries obsolete */
15109 +       if (au_ftest_hnjob(a->flags, DIRENT) && a->inode) {
15110 +               struct au_vdir *vdir;
15111 +
15112 +               vdir = au_ivdir(a->inode);
15113 +               if (vdir)
15114 +                       vdir->vd_jiffy = 0;
15115 +               /* IMustLock(a->inode); */
15116 +               /* a->inode->i_version++; */
15117 +       }
15118 +
15119 +       /* can do nothing but warn */
15120 +       if (au_ftest_hnjob(a->flags, MNTPNT)
15121 +           && a->dentry
15122 +           && d_mountpoint(a->dentry))
15123 +               pr_warn("mount-point %pd is removed or renamed\n", a->dentry);
15124 +
15125 +       return 0;
15126 +}
15127 +
15128 +/* ---------------------------------------------------------------------- */
15129 +
15130 +static struct dentry *lookup_wlock_by_name(char *name, unsigned int nlen,
15131 +                                          struct inode *dir)
15132 +{
15133 +       struct dentry *dentry, *d, *parent;
15134 +       struct qstr *dname;
15135 +
15136 +       parent = d_find_any_alias(dir);
15137 +       if (!parent)
15138 +               return NULL;
15139 +
15140 +       dentry = NULL;
15141 +       spin_lock(&parent->d_lock);
15142 +       list_for_each_entry(d, &parent->d_subdirs, d_child) {
15143 +               /* AuDbg("%pd\n", d); */
15144 +               spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
15145 +               dname = &d->d_name;
15146 +               if (dname->len != nlen || memcmp(dname->name, name, nlen))
15147 +                       goto cont_unlock;
15148 +               if (au_di(d))
15149 +                       au_digen_dec(d);
15150 +               else
15151 +                       goto cont_unlock;
15152 +               if (au_dcount(d) > 0) {
15153 +                       dentry = dget_dlock(d);
15154 +                       spin_unlock(&d->d_lock);
15155 +                       break;
15156 +               }
15157 +
15158 +cont_unlock:
15159 +               spin_unlock(&d->d_lock);
15160 +       }
15161 +       spin_unlock(&parent->d_lock);
15162 +       dput(parent);
15163 +
15164 +       if (dentry)
15165 +               di_write_lock_child(dentry);
15166 +
15167 +       return dentry;
15168 +}
15169 +
15170 +static struct inode *lookup_wlock_by_ino(struct super_block *sb,
15171 +                                        aufs_bindex_t bindex, ino_t h_ino)
15172 +{
15173 +       struct inode *inode;
15174 +       ino_t ino;
15175 +       int err;
15176 +
15177 +       inode = NULL;
15178 +       err = au_xino_read(sb, bindex, h_ino, &ino);
15179 +       if (!err && ino)
15180 +               inode = ilookup(sb, ino);
15181 +       if (!inode)
15182 +               goto out;
15183 +
15184 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
15185 +               pr_warn("wrong root branch\n");
15186 +               iput(inode);
15187 +               inode = NULL;
15188 +               goto out;
15189 +       }
15190 +
15191 +       ii_write_lock_child(inode);
15192 +
15193 +out:
15194 +       return inode;
15195 +}
15196 +
15197 +static void au_hn_bh(void *_args)
15198 +{
15199 +       struct au_hnotify_args *a = _args;
15200 +       struct super_block *sb;
15201 +       aufs_bindex_t bindex, bend, bfound;
15202 +       unsigned char xino, try_iput;
15203 +       int err;
15204 +       struct inode *inode;
15205 +       ino_t h_ino;
15206 +       struct hn_job_args args;
15207 +       struct dentry *dentry;
15208 +       struct au_sbinfo *sbinfo;
15209 +
15210 +       AuDebugOn(!_args);
15211 +       AuDebugOn(!a->h_dir);
15212 +       AuDebugOn(!a->dir);
15213 +       AuDebugOn(!a->mask);
15214 +       AuDbg("mask 0x%x, i%lu, hi%lu, hci%lu\n",
15215 +             a->mask, a->dir->i_ino, a->h_dir->i_ino,
15216 +             a->h_child_inode ? a->h_child_inode->i_ino : 0);
15217 +
15218 +       inode = NULL;
15219 +       dentry = NULL;
15220 +       /*
15221 +        * do not lock a->dir->i_mutex here
15222 +        * because of d_revalidate() may cause a deadlock.
15223 +        */
15224 +       sb = a->dir->i_sb;
15225 +       AuDebugOn(!sb);
15226 +       sbinfo = au_sbi(sb);
15227 +       AuDebugOn(!sbinfo);
15228 +       si_write_lock(sb, AuLock_NOPLMW);
15229 +
15230 +       ii_read_lock_parent(a->dir);
15231 +       bfound = -1;
15232 +       bend = au_ibend(a->dir);
15233 +       for (bindex = au_ibstart(a->dir); bindex <= bend; bindex++)
15234 +               if (au_h_iptr(a->dir, bindex) == a->h_dir) {
15235 +                       bfound = bindex;
15236 +                       break;
15237 +               }
15238 +       ii_read_unlock(a->dir);
15239 +       if (unlikely(bfound < 0))
15240 +               goto out;
15241 +
15242 +       xino = !!au_opt_test(au_mntflags(sb), XINO);
15243 +       h_ino = 0;
15244 +       if (a->h_child_inode)
15245 +               h_ino = a->h_child_inode->i_ino;
15246 +
15247 +       if (a->h_child_nlen
15248 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], GEN)
15249 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], MNTPNT)))
15250 +               dentry = lookup_wlock_by_name(a->h_child_name, a->h_child_nlen,
15251 +                                             a->dir);
15252 +       try_iput = 0;
15253 +       if (dentry)
15254 +               inode = dentry->d_inode;
15255 +       if (xino && !inode && h_ino
15256 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], XINO0)
15257 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], TRYXINO0)
15258 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], GEN))) {
15259 +               inode = lookup_wlock_by_ino(sb, bfound, h_ino);
15260 +               try_iput = 1;
15261 +           }
15262 +
15263 +       args.flags = a->flags[AuHn_CHILD];
15264 +       args.dentry = dentry;
15265 +       args.inode = inode;
15266 +       args.h_inode = a->h_child_inode;
15267 +       args.dir = a->dir;
15268 +       args.h_dir = a->h_dir;
15269 +       args.h_name = a->h_child_name;
15270 +       args.h_nlen = a->h_child_nlen;
15271 +       err = hn_job(&args);
15272 +       if (dentry) {
15273 +               if (au_di(dentry))
15274 +                       di_write_unlock(dentry);
15275 +               dput(dentry);
15276 +       }
15277 +       if (inode && try_iput) {
15278 +               ii_write_unlock(inode);
15279 +               iput(inode);
15280 +       }
15281 +
15282 +       ii_write_lock_parent(a->dir);
15283 +       args.flags = a->flags[AuHn_PARENT];
15284 +       args.dentry = NULL;
15285 +       args.inode = a->dir;
15286 +       args.h_inode = a->h_dir;
15287 +       args.dir = NULL;
15288 +       args.h_dir = NULL;
15289 +       args.h_name = NULL;
15290 +       args.h_nlen = 0;
15291 +       err = hn_job(&args);
15292 +       ii_write_unlock(a->dir);
15293 +
15294 +out:
15295 +       iput(a->h_child_inode);
15296 +       iput(a->h_dir);
15297 +       iput(a->dir);
15298 +       si_write_unlock(sb);
15299 +       au_nwt_done(&sbinfo->si_nowait);
15300 +       kfree(a);
15301 +}
15302 +
15303 +/* ---------------------------------------------------------------------- */
15304 +
15305 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
15306 +              struct qstr *h_child_qstr, struct inode *h_child_inode)
15307 +{
15308 +       int err, len;
15309 +       unsigned int flags[AuHnLast], f;
15310 +       unsigned char isdir, isroot, wh;
15311 +       struct inode *dir;
15312 +       struct au_hnotify_args *args;
15313 +       char *p, *h_child_name;
15314 +
15315 +       err = 0;
15316 +       AuDebugOn(!hnotify || !hnotify->hn_aufs_inode);
15317 +       dir = igrab(hnotify->hn_aufs_inode);
15318 +       if (!dir)
15319 +               goto out;
15320 +
15321 +       isroot = (dir->i_ino == AUFS_ROOT_INO);
15322 +       wh = 0;
15323 +       h_child_name = (void *)h_child_qstr->name;
15324 +       len = h_child_qstr->len;
15325 +       if (h_child_name) {
15326 +               if (len > AUFS_WH_PFX_LEN
15327 +                   && !memcmp(h_child_name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
15328 +                       h_child_name += AUFS_WH_PFX_LEN;
15329 +                       len -= AUFS_WH_PFX_LEN;
15330 +                       wh = 1;
15331 +               }
15332 +       }
15333 +
15334 +       isdir = 0;
15335 +       if (h_child_inode)
15336 +               isdir = !!S_ISDIR(h_child_inode->i_mode);
15337 +       flags[AuHn_PARENT] = AuHnJob_ISDIR;
15338 +       flags[AuHn_CHILD] = 0;
15339 +       if (isdir)
15340 +               flags[AuHn_CHILD] = AuHnJob_ISDIR;
15341 +       au_fset_hnjob(flags[AuHn_PARENT], DIRENT);
15342 +       au_fset_hnjob(flags[AuHn_CHILD], GEN);
15343 +       switch (mask & FS_EVENTS_POSS_ON_CHILD) {
15344 +       case FS_MOVED_FROM:
15345 +       case FS_MOVED_TO:
15346 +               au_fset_hnjob(flags[AuHn_CHILD], XINO0);
15347 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15348 +               /*FALLTHROUGH*/
15349 +       case FS_CREATE:
15350 +               AuDebugOn(!h_child_name);
15351 +               break;
15352 +
15353 +       case FS_DELETE:
15354 +               /*
15355 +                * aufs never be able to get this child inode.
15356 +                * revalidation should be in d_revalidate()
15357 +                * by checking i_nlink, i_generation or d_unhashed().
15358 +                */
15359 +               AuDebugOn(!h_child_name);
15360 +               au_fset_hnjob(flags[AuHn_CHILD], TRYXINO0);
15361 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
15362 +               break;
15363 +
15364 +       default:
15365 +               AuDebugOn(1);
15366 +       }
15367 +
15368 +       if (wh)
15369 +               h_child_inode = NULL;
15370 +
15371 +       err = -ENOMEM;
15372 +       /* iput() and kfree() will be called in au_hnotify() */
15373 +       args = kmalloc(sizeof(*args) + len + 1, GFP_NOFS);
15374 +       if (unlikely(!args)) {
15375 +               AuErr1("no memory\n");
15376 +               iput(dir);
15377 +               goto out;
15378 +       }
15379 +       args->flags[AuHn_PARENT] = flags[AuHn_PARENT];
15380 +       args->flags[AuHn_CHILD] = flags[AuHn_CHILD];
15381 +       args->mask = mask;
15382 +       args->dir = dir;
15383 +       args->h_dir = igrab(h_dir);
15384 +       if (h_child_inode)
15385 +               h_child_inode = igrab(h_child_inode); /* can be NULL */
15386 +       args->h_child_inode = h_child_inode;
15387 +       args->h_child_nlen = len;
15388 +       if (len) {
15389 +               p = (void *)args;
15390 +               p += sizeof(*args);
15391 +               memcpy(p, h_child_name, len);
15392 +               p[len] = 0;
15393 +       }
15394 +
15395 +       /* NFS fires the event for silly-renamed one from kworker */
15396 +       f = 0;
15397 +       if (!dir->i_nlink
15398 +           || (au_test_nfs(h_dir->i_sb) && (mask & FS_DELETE)))
15399 +               f = AuWkq_NEST;
15400 +       err = au_wkq_nowait(au_hn_bh, args, dir->i_sb, f);
15401 +       if (unlikely(err)) {
15402 +               pr_err("wkq %d\n", err);
15403 +               iput(args->h_child_inode);
15404 +               iput(args->h_dir);
15405 +               iput(args->dir);
15406 +               kfree(args);
15407 +       }
15408 +
15409 +out:
15410 +       return err;
15411 +}
15412 +
15413 +/* ---------------------------------------------------------------------- */
15414 +
15415 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm)
15416 +{
15417 +       int err;
15418 +
15419 +       AuDebugOn(!(udba & AuOptMask_UDBA));
15420 +
15421 +       err = 0;
15422 +       if (au_hnotify_op.reset_br)
15423 +               err = au_hnotify_op.reset_br(udba, br, perm);
15424 +
15425 +       return err;
15426 +}
15427 +
15428 +int au_hnotify_init_br(struct au_branch *br, int perm)
15429 +{
15430 +       int err;
15431 +
15432 +       err = 0;
15433 +       if (au_hnotify_op.init_br)
15434 +               err = au_hnotify_op.init_br(br, perm);
15435 +
15436 +       return err;
15437 +}
15438 +
15439 +void au_hnotify_fin_br(struct au_branch *br)
15440 +{
15441 +       if (au_hnotify_op.fin_br)
15442 +               au_hnotify_op.fin_br(br);
15443 +}
15444 +
15445 +static void au_hn_destroy_cache(void)
15446 +{
15447 +       kmem_cache_destroy(au_cachep[AuCache_HNOTIFY]);
15448 +       au_cachep[AuCache_HNOTIFY] = NULL;
15449 +}
15450 +
15451 +int __init au_hnotify_init(void)
15452 +{
15453 +       int err;
15454 +
15455 +       err = -ENOMEM;
15456 +       au_cachep[AuCache_HNOTIFY] = AuCache(au_hnotify);
15457 +       if (au_cachep[AuCache_HNOTIFY]) {
15458 +               err = 0;
15459 +               if (au_hnotify_op.init)
15460 +                       err = au_hnotify_op.init();
15461 +               if (unlikely(err))
15462 +                       au_hn_destroy_cache();
15463 +       }
15464 +       AuTraceErr(err);
15465 +       return err;
15466 +}
15467 +
15468 +void au_hnotify_fin(void)
15469 +{
15470 +       if (au_hnotify_op.fin)
15471 +               au_hnotify_op.fin();
15472 +       /* cf. au_cache_fin() */
15473 +       if (au_cachep[AuCache_HNOTIFY])
15474 +               au_hn_destroy_cache();
15475 +}
15476 diff -urN /usr/share/empty/fs/aufs/iinfo.c linux/fs/aufs/iinfo.c
15477 --- /usr/share/empty/fs/aufs/iinfo.c    1970-01-01 01:00:00.000000000 +0100
15478 +++ linux/fs/aufs/iinfo.c       2015-03-27 21:56:35.466795000 +0100
15479 @@ -0,0 +1,277 @@
15480 +/*
15481 + * Copyright (C) 2005-2015 Junjiro R. Okajima
15482 + *
15483 + * This program, aufs is free software; you can redistribute it and/or modify
15484 + * it under the terms of the GNU General Public License as published by
15485 + * the Free Software Foundation; either version 2 of the License, or
15486 + * (at your option) any later version.
15487 + *
15488 + * This program is distributed in the hope that it will be useful,
15489 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15490 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15491 + * GNU General Public License for more details.
15492 + *
15493 + * You should have received a copy of the GNU General Public License
15494 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15495 + */
15496 +
15497 +/*
15498 + * inode private data
15499 + */
15500 +
15501 +#include "aufs.h"
15502 +
15503 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex)
15504 +{
15505 +       struct inode *h_inode;
15506 +
15507 +       IiMustAnyLock(inode);
15508 +
15509 +       h_inode = au_ii(inode)->ii_hinode[0 + bindex].hi_inode;
15510 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
15511 +       return h_inode;
15512 +}
15513 +
15514 +/* todo: hard/soft set? */
15515 +void au_hiput(struct au_hinode *hinode)
15516 +{
15517 +       au_hn_free(hinode);
15518 +       dput(hinode->hi_whdentry);
15519 +       iput(hinode->hi_inode);
15520 +}
15521 +
15522 +unsigned int au_hi_flags(struct inode *inode, int isdir)
15523 +{
15524 +       unsigned int flags;
15525 +       const unsigned int mnt_flags = au_mntflags(inode->i_sb);
15526 +
15527 +       flags = 0;
15528 +       if (au_opt_test(mnt_flags, XINO))
15529 +               au_fset_hi(flags, XINO);
15530 +       if (isdir && au_opt_test(mnt_flags, UDBA_HNOTIFY))
15531 +               au_fset_hi(flags, HNOTIFY);
15532 +       return flags;
15533 +}
15534 +
15535 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
15536 +                  struct inode *h_inode, unsigned int flags)
15537 +{
15538 +       struct au_hinode *hinode;
15539 +       struct inode *hi;
15540 +       struct au_iinfo *iinfo = au_ii(inode);
15541 +
15542 +       IiMustWriteLock(inode);
15543 +
15544 +       hinode = iinfo->ii_hinode + bindex;
15545 +       hi = hinode->hi_inode;
15546 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
15547 +
15548 +       if (hi)
15549 +               au_hiput(hinode);
15550 +       hinode->hi_inode = h_inode;
15551 +       if (h_inode) {
15552 +               int err;
15553 +               struct super_block *sb = inode->i_sb;
15554 +               struct au_branch *br;
15555 +
15556 +               AuDebugOn(inode->i_mode
15557 +                         && (h_inode->i_mode & S_IFMT)
15558 +                         != (inode->i_mode & S_IFMT));
15559 +               if (bindex == iinfo->ii_bstart)
15560 +                       au_cpup_igen(inode, h_inode);
15561 +               br = au_sbr(sb, bindex);
15562 +               hinode->hi_id = br->br_id;
15563 +               if (au_ftest_hi(flags, XINO)) {
15564 +                       err = au_xino_write(sb, bindex, h_inode->i_ino,
15565 +                                           inode->i_ino);
15566 +                       if (unlikely(err))
15567 +                               AuIOErr1("failed au_xino_write() %d\n", err);
15568 +               }
15569 +
15570 +               if (au_ftest_hi(flags, HNOTIFY)
15571 +                   && au_br_hnotifyable(br->br_perm)) {
15572 +                       err = au_hn_alloc(hinode, inode);
15573 +                       if (unlikely(err))
15574 +                               AuIOErr1("au_hn_alloc() %d\n", err);
15575 +               }
15576 +       }
15577 +}
15578 +
15579 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
15580 +                 struct dentry *h_wh)
15581 +{
15582 +       struct au_hinode *hinode;
15583 +
15584 +       IiMustWriteLock(inode);
15585 +
15586 +       hinode = au_ii(inode)->ii_hinode + bindex;
15587 +       AuDebugOn(hinode->hi_whdentry);
15588 +       hinode->hi_whdentry = h_wh;
15589 +}
15590 +
15591 +void au_update_iigen(struct inode *inode, int half)
15592 +{
15593 +       struct au_iinfo *iinfo;
15594 +       struct au_iigen *iigen;
15595 +       unsigned int sigen;
15596 +
15597 +       sigen = au_sigen(inode->i_sb);
15598 +       iinfo = au_ii(inode);
15599 +       iigen = &iinfo->ii_generation;
15600 +       spin_lock(&iinfo->ii_genspin);
15601 +       iigen->ig_generation = sigen;
15602 +       if (half)
15603 +               au_ig_fset(iigen->ig_flags, HALF_REFRESHED);
15604 +       else
15605 +               au_ig_fclr(iigen->ig_flags, HALF_REFRESHED);
15606 +       spin_unlock(&iinfo->ii_genspin);
15607 +}
15608 +
15609 +/* it may be called at remount time, too */
15610 +void au_update_ibrange(struct inode *inode, int do_put_zero)
15611 +{
15612 +       struct au_iinfo *iinfo;
15613 +       aufs_bindex_t bindex, bend;
15614 +
15615 +       iinfo = au_ii(inode);
15616 +       if (!iinfo)
15617 +               return;
15618 +
15619 +       IiMustWriteLock(inode);
15620 +
15621 +       if (do_put_zero && iinfo->ii_bstart >= 0) {
15622 +               for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend;
15623 +                    bindex++) {
15624 +                       struct inode *h_i;
15625 +
15626 +                       h_i = iinfo->ii_hinode[0 + bindex].hi_inode;
15627 +                       if (h_i
15628 +                           && !h_i->i_nlink
15629 +                           && !(h_i->i_state & I_LINKABLE))
15630 +                               au_set_h_iptr(inode, bindex, NULL, 0);
15631 +               }
15632 +       }
15633 +
15634 +       iinfo->ii_bstart = -1;
15635 +       iinfo->ii_bend = -1;
15636 +       bend = au_sbend(inode->i_sb);
15637 +       for (bindex = 0; bindex <= bend; bindex++)
15638 +               if (iinfo->ii_hinode[0 + bindex].hi_inode) {
15639 +                       iinfo->ii_bstart = bindex;
15640 +                       break;
15641 +               }
15642 +       if (iinfo->ii_bstart >= 0)
15643 +               for (bindex = bend; bindex >= iinfo->ii_bstart; bindex--)
15644 +                       if (iinfo->ii_hinode[0 + bindex].hi_inode) {
15645 +                               iinfo->ii_bend = bindex;
15646 +                               break;
15647 +                       }
15648 +       AuDebugOn(iinfo->ii_bstart > iinfo->ii_bend);
15649 +}
15650 +
15651 +/* ---------------------------------------------------------------------- */
15652 +
15653 +void au_icntnr_init_once(void *_c)
15654 +{
15655 +       struct au_icntnr *c = _c;
15656 +       struct au_iinfo *iinfo = &c->iinfo;
15657 +       static struct lock_class_key aufs_ii;
15658 +
15659 +       spin_lock_init(&iinfo->ii_genspin);
15660 +       au_rw_init(&iinfo->ii_rwsem);
15661 +       au_rw_class(&iinfo->ii_rwsem, &aufs_ii);
15662 +       inode_init_once(&c->vfs_inode);
15663 +}
15664 +
15665 +int au_iinfo_init(struct inode *inode)
15666 +{
15667 +       struct au_iinfo *iinfo;
15668 +       struct super_block *sb;
15669 +       int nbr, i;
15670 +
15671 +       sb = inode->i_sb;
15672 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
15673 +       nbr = au_sbend(sb) + 1;
15674 +       if (unlikely(nbr <= 0))
15675 +               nbr = 1;
15676 +       iinfo->ii_hinode = kcalloc(nbr, sizeof(*iinfo->ii_hinode), GFP_NOFS);
15677 +       if (iinfo->ii_hinode) {
15678 +               au_ninodes_inc(sb);
15679 +               for (i = 0; i < nbr; i++)
15680 +                       iinfo->ii_hinode[i].hi_id = -1;
15681 +
15682 +               iinfo->ii_generation.ig_generation = au_sigen(sb);
15683 +               iinfo->ii_bstart = -1;
15684 +               iinfo->ii_bend = -1;
15685 +               iinfo->ii_vdir = NULL;
15686 +               return 0;
15687 +       }
15688 +       return -ENOMEM;
15689 +}
15690 +
15691 +int au_ii_realloc(struct au_iinfo *iinfo, int nbr)
15692 +{
15693 +       int err, sz;
15694 +       struct au_hinode *hip;
15695 +
15696 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
15697 +
15698 +       err = -ENOMEM;
15699 +       sz = sizeof(*hip) * (iinfo->ii_bend + 1);
15700 +       if (!sz)
15701 +               sz = sizeof(*hip);
15702 +       hip = au_kzrealloc(iinfo->ii_hinode, sz, sizeof(*hip) * nbr, GFP_NOFS);
15703 +       if (hip) {
15704 +               iinfo->ii_hinode = hip;
15705 +               err = 0;
15706 +       }
15707 +
15708 +       return err;
15709 +}
15710 +
15711 +void au_iinfo_fin(struct inode *inode)
15712 +{
15713 +       struct au_iinfo *iinfo;
15714 +       struct au_hinode *hi;
15715 +       struct super_block *sb;
15716 +       aufs_bindex_t bindex, bend;
15717 +       const unsigned char unlinked = !inode->i_nlink;
15718 +
15719 +       iinfo = au_ii(inode);
15720 +       /* bad_inode case */
15721 +       if (!iinfo)
15722 +               return;
15723 +
15724 +       sb = inode->i_sb;
15725 +       au_ninodes_dec(sb);
15726 +       if (si_pid_test(sb))
15727 +               au_xino_delete_inode(inode, unlinked);
15728 +       else {
15729 +               /*
15730 +                * it is safe to hide the dependency between sbinfo and
15731 +                * sb->s_umount.
15732 +                */
15733 +               lockdep_off();
15734 +               si_noflush_read_lock(sb);
15735 +               au_xino_delete_inode(inode, unlinked);
15736 +               si_read_unlock(sb);
15737 +               lockdep_on();
15738 +       }
15739 +
15740 +       if (iinfo->ii_vdir)
15741 +               au_vdir_free(iinfo->ii_vdir);
15742 +
15743 +       bindex = iinfo->ii_bstart;
15744 +       if (bindex >= 0) {
15745 +               hi = iinfo->ii_hinode + bindex;
15746 +               bend = iinfo->ii_bend;
15747 +               while (bindex++ <= bend) {
15748 +                       if (hi->hi_inode)
15749 +                               au_hiput(hi);
15750 +                       hi++;
15751 +               }
15752 +       }
15753 +       kfree(iinfo->ii_hinode);
15754 +       iinfo->ii_hinode = NULL;
15755 +       AuRwDestroy(&iinfo->ii_rwsem);
15756 +}
15757 diff -urN /usr/share/empty/fs/aufs/inode.c linux/fs/aufs/inode.c
15758 --- /usr/share/empty/fs/aufs/inode.c    1970-01-01 01:00:00.000000000 +0100
15759 +++ linux/fs/aufs/inode.c       2015-03-27 21:56:35.466795000 +0100
15760 @@ -0,0 +1,495 @@
15761 +/*
15762 + * Copyright (C) 2005-2015 Junjiro R. Okajima
15763 + *
15764 + * This program, aufs is free software; you can redistribute it and/or modify
15765 + * it under the terms of the GNU General Public License as published by
15766 + * the Free Software Foundation; either version 2 of the License, or
15767 + * (at your option) any later version.
15768 + *
15769 + * This program is distributed in the hope that it will be useful,
15770 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15771 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15772 + * GNU General Public License for more details.
15773 + *
15774 + * You should have received a copy of the GNU General Public License
15775 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15776 + */
15777 +
15778 +/*
15779 + * inode functions
15780 + */
15781 +
15782 +#include "aufs.h"
15783 +
15784 +struct inode *au_igrab(struct inode *inode)
15785 +{
15786 +       if (inode) {
15787 +               AuDebugOn(!atomic_read(&inode->i_count));
15788 +               ihold(inode);
15789 +       }
15790 +       return inode;
15791 +}
15792 +
15793 +static void au_refresh_hinode_attr(struct inode *inode, int do_version)
15794 +{
15795 +       au_cpup_attr_all(inode, /*force*/0);
15796 +       au_update_iigen(inode, /*half*/1);
15797 +       if (do_version)
15798 +               inode->i_version++;
15799 +}
15800 +
15801 +static int au_ii_refresh(struct inode *inode, int *update)
15802 +{
15803 +       int err, e;
15804 +       umode_t type;
15805 +       aufs_bindex_t bindex, new_bindex;
15806 +       struct super_block *sb;
15807 +       struct au_iinfo *iinfo;
15808 +       struct au_hinode *p, *q, tmp;
15809 +
15810 +       IiMustWriteLock(inode);
15811 +
15812 +       *update = 0;
15813 +       sb = inode->i_sb;
15814 +       type = inode->i_mode & S_IFMT;
15815 +       iinfo = au_ii(inode);
15816 +       err = au_ii_realloc(iinfo, au_sbend(sb) + 1);
15817 +       if (unlikely(err))
15818 +               goto out;
15819 +
15820 +       AuDebugOn(iinfo->ii_bstart < 0);
15821 +       p = iinfo->ii_hinode + iinfo->ii_bstart;
15822 +       for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend;
15823 +            bindex++, p++) {
15824 +               if (!p->hi_inode)
15825 +                       continue;
15826 +
15827 +               AuDebugOn(type != (p->hi_inode->i_mode & S_IFMT));
15828 +               new_bindex = au_br_index(sb, p->hi_id);
15829 +               if (new_bindex == bindex)
15830 +                       continue;
15831 +
15832 +               if (new_bindex < 0) {
15833 +                       *update = 1;
15834 +                       au_hiput(p);
15835 +                       p->hi_inode = NULL;
15836 +                       continue;
15837 +               }
15838 +
15839 +               if (new_bindex < iinfo->ii_bstart)
15840 +                       iinfo->ii_bstart = new_bindex;
15841 +               if (iinfo->ii_bend < new_bindex)
15842 +                       iinfo->ii_bend = new_bindex;
15843 +               /* swap two lower inode, and loop again */
15844 +               q = iinfo->ii_hinode + new_bindex;
15845 +               tmp = *q;
15846 +               *q = *p;
15847 +               *p = tmp;
15848 +               if (tmp.hi_inode) {
15849 +                       bindex--;
15850 +                       p--;
15851 +               }
15852 +       }
15853 +       au_update_ibrange(inode, /*do_put_zero*/0);
15854 +       e = au_dy_irefresh(inode);
15855 +       if (unlikely(e && !err))
15856 +               err = e;
15857 +
15858 +out:
15859 +       AuTraceErr(err);
15860 +       return err;
15861 +}
15862 +
15863 +int au_refresh_hinode_self(struct inode *inode)
15864 +{
15865 +       int err, update;
15866 +
15867 +       err = au_ii_refresh(inode, &update);
15868 +       if (!err)
15869 +               au_refresh_hinode_attr(inode, update && S_ISDIR(inode->i_mode));
15870 +
15871 +       AuTraceErr(err);
15872 +       return err;
15873 +}
15874 +
15875 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry)
15876 +{
15877 +       int err, e, update;
15878 +       unsigned int flags;
15879 +       umode_t mode;
15880 +       aufs_bindex_t bindex, bend;
15881 +       unsigned char isdir;
15882 +       struct au_hinode *p;
15883 +       struct au_iinfo *iinfo;
15884 +
15885 +       err = au_ii_refresh(inode, &update);
15886 +       if (unlikely(err))
15887 +               goto out;
15888 +
15889 +       update = 0;
15890 +       iinfo = au_ii(inode);
15891 +       p = iinfo->ii_hinode + iinfo->ii_bstart;
15892 +       mode = (inode->i_mode & S_IFMT);
15893 +       isdir = S_ISDIR(mode);
15894 +       flags = au_hi_flags(inode, isdir);
15895 +       bend = au_dbend(dentry);
15896 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++) {
15897 +               struct inode *h_i;
15898 +               struct dentry *h_d;
15899 +
15900 +               h_d = au_h_dptr(dentry, bindex);
15901 +               if (!h_d || !h_d->d_inode)
15902 +                       continue;
15903 +
15904 +               AuDebugOn(mode != (h_d->d_inode->i_mode & S_IFMT));
15905 +               if (iinfo->ii_bstart <= bindex && bindex <= iinfo->ii_bend) {
15906 +                       h_i = au_h_iptr(inode, bindex);
15907 +                       if (h_i) {
15908 +                               if (h_i == h_d->d_inode)
15909 +                                       continue;
15910 +                               err = -EIO;
15911 +                               break;
15912 +                       }
15913 +               }
15914 +               if (bindex < iinfo->ii_bstart)
15915 +                       iinfo->ii_bstart = bindex;
15916 +               if (iinfo->ii_bend < bindex)
15917 +                       iinfo->ii_bend = bindex;
15918 +               au_set_h_iptr(inode, bindex, au_igrab(h_d->d_inode), flags);
15919 +               update = 1;
15920 +       }
15921 +       au_update_ibrange(inode, /*do_put_zero*/0);
15922 +       e = au_dy_irefresh(inode);
15923 +       if (unlikely(e && !err))
15924 +               err = e;
15925 +       if (!err)
15926 +               au_refresh_hinode_attr(inode, update && isdir);
15927 +
15928 +out:
15929 +       AuTraceErr(err);
15930 +       return err;
15931 +}
15932 +
15933 +static int set_inode(struct inode *inode, struct dentry *dentry)
15934 +{
15935 +       int err;
15936 +       unsigned int flags;
15937 +       umode_t mode;
15938 +       aufs_bindex_t bindex, bstart, btail;
15939 +       unsigned char isdir;
15940 +       struct dentry *h_dentry;
15941 +       struct inode *h_inode;
15942 +       struct au_iinfo *iinfo;
15943 +
15944 +       IiMustWriteLock(inode);
15945 +
15946 +       err = 0;
15947 +       isdir = 0;
15948 +       bstart = au_dbstart(dentry);
15949 +       h_inode = au_h_dptr(dentry, bstart)->d_inode;
15950 +       mode = h_inode->i_mode;
15951 +       switch (mode & S_IFMT) {
15952 +       case S_IFREG:
15953 +               btail = au_dbtail(dentry);
15954 +               inode->i_op = &aufs_iop;
15955 +               inode->i_fop = &aufs_file_fop;
15956 +               err = au_dy_iaop(inode, bstart, h_inode);
15957 +               if (unlikely(err))
15958 +                       goto out;
15959 +               break;
15960 +       case S_IFDIR:
15961 +               isdir = 1;
15962 +               btail = au_dbtaildir(dentry);
15963 +               inode->i_op = &aufs_dir_iop;
15964 +               inode->i_fop = &aufs_dir_fop;
15965 +               break;
15966 +       case S_IFLNK:
15967 +               btail = au_dbtail(dentry);
15968 +               inode->i_op = &aufs_symlink_iop;
15969 +               break;
15970 +       case S_IFBLK:
15971 +       case S_IFCHR:
15972 +       case S_IFIFO:
15973 +       case S_IFSOCK:
15974 +               btail = au_dbtail(dentry);
15975 +               inode->i_op = &aufs_iop;
15976 +               init_special_inode(inode, mode, h_inode->i_rdev);
15977 +               break;
15978 +       default:
15979 +               AuIOErr("Unknown file type 0%o\n", mode);
15980 +               err = -EIO;
15981 +               goto out;
15982 +       }
15983 +
15984 +       /* do not set hnotify for whiteouted dirs (SHWH mode) */
15985 +       flags = au_hi_flags(inode, isdir);
15986 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH)
15987 +           && au_ftest_hi(flags, HNOTIFY)
15988 +           && dentry->d_name.len > AUFS_WH_PFX_LEN
15989 +           && !memcmp(dentry->d_name.name, AUFS_WH_PFX, AUFS_WH_PFX_LEN))
15990 +               au_fclr_hi(flags, HNOTIFY);
15991 +       iinfo = au_ii(inode);
15992 +       iinfo->ii_bstart = bstart;
15993 +       iinfo->ii_bend = btail;
15994 +       for (bindex = bstart; bindex <= btail; bindex++) {
15995 +               h_dentry = au_h_dptr(dentry, bindex);
15996 +               if (h_dentry)
15997 +                       au_set_h_iptr(inode, bindex,
15998 +                                     au_igrab(h_dentry->d_inode), flags);
15999 +       }
16000 +       au_cpup_attr_all(inode, /*force*/1);
16001 +       /*
16002 +        * to force calling aufs_get_acl() every time,
16003 +        * do not call cache_no_acl() for aufs inode.
16004 +        */
16005 +
16006 +out:
16007 +       return err;
16008 +}
16009 +
16010 +/*
16011 + * successful returns with iinfo write_locked
16012 + * minus: errno
16013 + * zero: success, matched
16014 + * plus: no error, but unmatched
16015 + */
16016 +static int reval_inode(struct inode *inode, struct dentry *dentry)
16017 +{
16018 +       int err;
16019 +       unsigned int gen;
16020 +       struct au_iigen iigen;
16021 +       aufs_bindex_t bindex, bend;
16022 +       struct inode *h_inode, *h_dinode;
16023 +
16024 +       /*
16025 +        * before this function, if aufs got any iinfo lock, it must be only
16026 +        * one, the parent dir.
16027 +        * it can happen by UDBA and the obsoleted inode number.
16028 +        */
16029 +       err = -EIO;
16030 +       if (unlikely(inode->i_ino == parent_ino(dentry)))
16031 +               goto out;
16032 +
16033 +       err = 1;
16034 +       ii_write_lock_new_child(inode);
16035 +       h_dinode = au_h_dptr(dentry, au_dbstart(dentry))->d_inode;
16036 +       bend = au_ibend(inode);
16037 +       for (bindex = au_ibstart(inode); bindex <= bend; bindex++) {
16038 +               h_inode = au_h_iptr(inode, bindex);
16039 +               if (!h_inode || h_inode != h_dinode)
16040 +                       continue;
16041 +
16042 +               err = 0;
16043 +               gen = au_iigen(inode, &iigen);
16044 +               if (gen == au_digen(dentry)
16045 +                   && !au_ig_ftest(iigen.ig_flags, HALF_REFRESHED))
16046 +                       break;
16047 +
16048 +               /* fully refresh inode using dentry */
16049 +               err = au_refresh_hinode(inode, dentry);
16050 +               if (!err)
16051 +                       au_update_iigen(inode, /*half*/0);
16052 +               break;
16053 +       }
16054 +
16055 +       if (unlikely(err))
16056 +               ii_write_unlock(inode);
16057 +out:
16058 +       return err;
16059 +}
16060 +
16061 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16062 +          unsigned int d_type, ino_t *ino)
16063 +{
16064 +       int err;
16065 +       struct mutex *mtx;
16066 +
16067 +       /* prevent hardlinked inode number from race condition */
16068 +       mtx = NULL;
16069 +       if (d_type != DT_DIR) {
16070 +               mtx = &au_sbr(sb, bindex)->br_xino.xi_nondir_mtx;
16071 +               mutex_lock(mtx);
16072 +       }
16073 +       err = au_xino_read(sb, bindex, h_ino, ino);
16074 +       if (unlikely(err))
16075 +               goto out;
16076 +
16077 +       if (!*ino) {
16078 +               err = -EIO;
16079 +               *ino = au_xino_new_ino(sb);
16080 +               if (unlikely(!*ino))
16081 +                       goto out;
16082 +               err = au_xino_write(sb, bindex, h_ino, *ino);
16083 +               if (unlikely(err))
16084 +                       goto out;
16085 +       }
16086 +
16087 +out:
16088 +       if (mtx)
16089 +               mutex_unlock(mtx);
16090 +       return err;
16091 +}
16092 +
16093 +/* successful returns with iinfo write_locked */
16094 +/* todo: return with unlocked? */
16095 +struct inode *au_new_inode(struct dentry *dentry, int must_new)
16096 +{
16097 +       struct inode *inode;
16098 +       struct dentry *h_dentry;
16099 +       struct super_block *sb;
16100 +       struct mutex *mtx;
16101 +       ino_t h_ino, ino;
16102 +       int err;
16103 +       aufs_bindex_t bstart;
16104 +
16105 +       sb = dentry->d_sb;
16106 +       bstart = au_dbstart(dentry);
16107 +       h_dentry = au_h_dptr(dentry, bstart);
16108 +       h_ino = h_dentry->d_inode->i_ino;
16109 +
16110 +       /*
16111 +        * stop 'race'-ing between hardlinks under different
16112 +        * parents.
16113 +        */
16114 +       mtx = NULL;
16115 +       if (!d_is_dir(h_dentry))
16116 +               mtx = &au_sbr(sb, bstart)->br_xino.xi_nondir_mtx;
16117 +
16118 +new_ino:
16119 +       if (mtx)
16120 +               mutex_lock(mtx);
16121 +       err = au_xino_read(sb, bstart, h_ino, &ino);
16122 +       inode = ERR_PTR(err);
16123 +       if (unlikely(err))
16124 +               goto out;
16125 +
16126 +       if (!ino) {
16127 +               ino = au_xino_new_ino(sb);
16128 +               if (unlikely(!ino)) {
16129 +                       inode = ERR_PTR(-EIO);
16130 +                       goto out;
16131 +               }
16132 +       }
16133 +
16134 +       AuDbg("i%lu\n", (unsigned long)ino);
16135 +       inode = au_iget_locked(sb, ino);
16136 +       err = PTR_ERR(inode);
16137 +       if (IS_ERR(inode))
16138 +               goto out;
16139 +
16140 +       AuDbg("%lx, new %d\n", inode->i_state, !!(inode->i_state & I_NEW));
16141 +       if (inode->i_state & I_NEW) {
16142 +               /* verbose coding for lock class name */
16143 +               if (unlikely(d_is_symlink(h_dentry)))
16144 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
16145 +                                   au_lc_key + AuLcSymlink_IIINFO);
16146 +               else if (unlikely(d_is_dir(h_dentry)))
16147 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
16148 +                                   au_lc_key + AuLcDir_IIINFO);
16149 +               else /* likely */
16150 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
16151 +                                   au_lc_key + AuLcNonDir_IIINFO);
16152 +
16153 +               ii_write_lock_new_child(inode);
16154 +               err = set_inode(inode, dentry);
16155 +               if (!err) {
16156 +                       unlock_new_inode(inode);
16157 +                       goto out; /* success */
16158 +               }
16159 +
16160 +               /*
16161 +                * iget_failed() calls iput(), but we need to call
16162 +                * ii_write_unlock() after iget_failed(). so dirty hack for
16163 +                * i_count.
16164 +                */
16165 +               atomic_inc(&inode->i_count);
16166 +               iget_failed(inode);
16167 +               ii_write_unlock(inode);
16168 +               au_xino_write(sb, bstart, h_ino, /*ino*/0);
16169 +               /* ignore this error */
16170 +               goto out_iput;
16171 +       } else if (!must_new && !IS_DEADDIR(inode) && inode->i_nlink) {
16172 +               /*
16173 +                * horrible race condition between lookup, readdir and copyup
16174 +                * (or something).
16175 +                */
16176 +               if (mtx)
16177 +                       mutex_unlock(mtx);
16178 +               err = reval_inode(inode, dentry);
16179 +               if (unlikely(err < 0)) {
16180 +                       mtx = NULL;
16181 +                       goto out_iput;
16182 +               }
16183 +
16184 +               if (!err) {
16185 +                       mtx = NULL;
16186 +                       goto out; /* success */
16187 +               } else if (mtx)
16188 +                       mutex_lock(mtx);
16189 +       }
16190 +
16191 +       if (unlikely(au_test_fs_unique_ino(h_dentry->d_inode)))
16192 +               AuWarn1("Warning: Un-notified UDBA or repeatedly renamed dir,"
16193 +                       " b%d, %s, %pd, hi%lu, i%lu.\n",
16194 +                       bstart, au_sbtype(h_dentry->d_sb), dentry,
16195 +                       (unsigned long)h_ino, (unsigned long)ino);
16196 +       ino = 0;
16197 +       err = au_xino_write(sb, bstart, h_ino, /*ino*/0);
16198 +       if (!err) {
16199 +               iput(inode);
16200 +               if (mtx)
16201 +                       mutex_unlock(mtx);
16202 +               goto new_ino;
16203 +       }
16204 +
16205 +out_iput:
16206 +       iput(inode);
16207 +       inode = ERR_PTR(err);
16208 +out:
16209 +       if (mtx)
16210 +               mutex_unlock(mtx);
16211 +       return inode;
16212 +}
16213 +
16214 +/* ---------------------------------------------------------------------- */
16215 +
16216 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16217 +              struct inode *inode)
16218 +{
16219 +       int err;
16220 +       struct inode *hi;
16221 +
16222 +       err = au_br_rdonly(au_sbr(sb, bindex));
16223 +
16224 +       /* pseudo-link after flushed may happen out of bounds */
16225 +       if (!err
16226 +           && inode
16227 +           && au_ibstart(inode) <= bindex
16228 +           && bindex <= au_ibend(inode)) {
16229 +               /*
16230 +                * permission check is unnecessary since vfsub routine
16231 +                * will be called later
16232 +                */
16233 +               hi = au_h_iptr(inode, bindex);
16234 +               if (hi)
16235 +                       err = IS_IMMUTABLE(hi) ? -EROFS : 0;
16236 +       }
16237 +
16238 +       return err;
16239 +}
16240 +
16241 +int au_test_h_perm(struct inode *h_inode, int mask)
16242 +{
16243 +       if (uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
16244 +               return 0;
16245 +       return inode_permission(h_inode, mask);
16246 +}
16247 +
16248 +int au_test_h_perm_sio(struct inode *h_inode, int mask)
16249 +{
16250 +       if (au_test_nfs(h_inode->i_sb)
16251 +           && (mask & MAY_WRITE)
16252 +           && S_ISDIR(h_inode->i_mode))
16253 +               mask |= MAY_READ; /* force permission check */
16254 +       return au_test_h_perm(h_inode, mask);
16255 +}
16256 diff -urN /usr/share/empty/fs/aufs/inode.h linux/fs/aufs/inode.h
16257 --- /usr/share/empty/fs/aufs/inode.h    1970-01-01 01:00:00.000000000 +0100
16258 +++ linux/fs/aufs/inode.h       2015-03-27 21:56:35.466795000 +0100
16259 @@ -0,0 +1,667 @@
16260 +/*
16261 + * Copyright (C) 2005-2015 Junjiro R. Okajima
16262 + *
16263 + * This program, aufs is free software; you can redistribute it and/or modify
16264 + * it under the terms of the GNU General Public License as published by
16265 + * the Free Software Foundation; either version 2 of the License, or
16266 + * (at your option) any later version.
16267 + *
16268 + * This program is distributed in the hope that it will be useful,
16269 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16270 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16271 + * GNU General Public License for more details.
16272 + *
16273 + * You should have received a copy of the GNU General Public License
16274 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16275 + */
16276 +
16277 +/*
16278 + * inode operations
16279 + */
16280 +
16281 +#ifndef __AUFS_INODE_H__
16282 +#define __AUFS_INODE_H__
16283 +
16284 +#ifdef __KERNEL__
16285 +
16286 +#include <linux/fsnotify.h>
16287 +#include "rwsem.h"
16288 +
16289 +struct vfsmount;
16290 +
16291 +struct au_hnotify {
16292 +#ifdef CONFIG_AUFS_HNOTIFY
16293 +#ifdef CONFIG_AUFS_HFSNOTIFY
16294 +       /* never use fsnotify_add_vfsmount_mark() */
16295 +       struct fsnotify_mark            hn_mark;
16296 +#endif
16297 +       struct inode                    *hn_aufs_inode; /* no get/put */
16298 +#endif
16299 +} ____cacheline_aligned_in_smp;
16300 +
16301 +struct au_hinode {
16302 +       struct inode            *hi_inode;
16303 +       aufs_bindex_t           hi_id;
16304 +#ifdef CONFIG_AUFS_HNOTIFY
16305 +       struct au_hnotify       *hi_notify;
16306 +#endif
16307 +
16308 +       /* reference to the copied-up whiteout with get/put */
16309 +       struct dentry           *hi_whdentry;
16310 +};
16311 +
16312 +/* ig_flags */
16313 +#define AuIG_HALF_REFRESHED            1
16314 +#define au_ig_ftest(flags, name)       ((flags) & AuIG_##name)
16315 +#define au_ig_fset(flags, name) \
16316 +       do { (flags) |= AuIG_##name; } while (0)
16317 +#define au_ig_fclr(flags, name) \
16318 +       do { (flags) &= ~AuIG_##name; } while (0)
16319 +
16320 +struct au_iigen {
16321 +       __u32           ig_generation, ig_flags;
16322 +};
16323 +
16324 +struct au_vdir;
16325 +struct au_iinfo {
16326 +       spinlock_t              ii_genspin;
16327 +       struct au_iigen         ii_generation;
16328 +       struct super_block      *ii_hsb1;       /* no get/put */
16329 +
16330 +       struct au_rwsem         ii_rwsem;
16331 +       aufs_bindex_t           ii_bstart, ii_bend;
16332 +       __u32                   ii_higen;
16333 +       struct au_hinode        *ii_hinode;
16334 +       struct au_vdir          *ii_vdir;
16335 +};
16336 +
16337 +struct au_icntnr {
16338 +       struct au_iinfo iinfo;
16339 +       struct inode vfs_inode;
16340 +} ____cacheline_aligned_in_smp;
16341 +
16342 +/* au_pin flags */
16343 +#define AuPin_DI_LOCKED                1
16344 +#define AuPin_MNT_WRITE                (1 << 1)
16345 +#define au_ftest_pin(flags, name)      ((flags) & AuPin_##name)
16346 +#define au_fset_pin(flags, name) \
16347 +       do { (flags) |= AuPin_##name; } while (0)
16348 +#define au_fclr_pin(flags, name) \
16349 +       do { (flags) &= ~AuPin_##name; } while (0)
16350 +
16351 +struct au_pin {
16352 +       /* input */
16353 +       struct dentry *dentry;
16354 +       unsigned int udba;
16355 +       unsigned char lsc_di, lsc_hi, flags;
16356 +       aufs_bindex_t bindex;
16357 +
16358 +       /* output */
16359 +       struct dentry *parent;
16360 +       struct au_hinode *hdir;
16361 +       struct vfsmount *h_mnt;
16362 +
16363 +       /* temporary unlock/relock for copyup */
16364 +       struct dentry *h_dentry, *h_parent;
16365 +       struct au_branch *br;
16366 +       struct task_struct *task;
16367 +};
16368 +
16369 +void au_pin_hdir_unlock(struct au_pin *p);
16370 +int au_pin_hdir_lock(struct au_pin *p);
16371 +int au_pin_hdir_relock(struct au_pin *p);
16372 +void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task);
16373 +void au_pin_hdir_acquire_nest(struct au_pin *p);
16374 +void au_pin_hdir_release(struct au_pin *p);
16375 +
16376 +/* ---------------------------------------------------------------------- */
16377 +
16378 +static inline struct au_iinfo *au_ii(struct inode *inode)
16379 +{
16380 +       struct au_iinfo *iinfo;
16381 +
16382 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
16383 +       if (iinfo->ii_hinode)
16384 +               return iinfo;
16385 +       return NULL; /* debugging bad_inode case */
16386 +}
16387 +
16388 +/* ---------------------------------------------------------------------- */
16389 +
16390 +/* inode.c */
16391 +struct inode *au_igrab(struct inode *inode);
16392 +int au_refresh_hinode_self(struct inode *inode);
16393 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry);
16394 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
16395 +          unsigned int d_type, ino_t *ino);
16396 +struct inode *au_new_inode(struct dentry *dentry, int must_new);
16397 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
16398 +              struct inode *inode);
16399 +int au_test_h_perm(struct inode *h_inode, int mask);
16400 +int au_test_h_perm_sio(struct inode *h_inode, int mask);
16401 +
16402 +static inline int au_wh_ino(struct super_block *sb, aufs_bindex_t bindex,
16403 +                           ino_t h_ino, unsigned int d_type, ino_t *ino)
16404 +{
16405 +#ifdef CONFIG_AUFS_SHWH
16406 +       return au_ino(sb, bindex, h_ino, d_type, ino);
16407 +#else
16408 +       return 0;
16409 +#endif
16410 +}
16411 +
16412 +/* i_op.c */
16413 +extern struct inode_operations aufs_iop, aufs_symlink_iop, aufs_dir_iop;
16414 +
16415 +/* au_wr_dir flags */
16416 +#define AuWrDir_ADD_ENTRY      1
16417 +#define AuWrDir_TMP_WHENTRY    (1 << 1)
16418 +#define AuWrDir_ISDIR          (1 << 2)
16419 +#define AuWrDir_TMPFILE                (1 << 3)
16420 +#define au_ftest_wrdir(flags, name)    ((flags) & AuWrDir_##name)
16421 +#define au_fset_wrdir(flags, name) \
16422 +       do { (flags) |= AuWrDir_##name; } while (0)
16423 +#define au_fclr_wrdir(flags, name) \
16424 +       do { (flags) &= ~AuWrDir_##name; } while (0)
16425 +
16426 +struct au_wr_dir_args {
16427 +       aufs_bindex_t force_btgt;
16428 +       unsigned char flags;
16429 +};
16430 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
16431 +             struct au_wr_dir_args *args);
16432 +
16433 +struct dentry *au_pinned_h_parent(struct au_pin *pin);
16434 +void au_pin_init(struct au_pin *pin, struct dentry *dentry,
16435 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
16436 +                unsigned int udba, unsigned char flags);
16437 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
16438 +          unsigned int udba, unsigned char flags) __must_check;
16439 +int au_do_pin(struct au_pin *pin) __must_check;
16440 +void au_unpin(struct au_pin *pin);
16441 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen);
16442 +
16443 +#define AuIcpup_DID_CPUP       1
16444 +#define au_ftest_icpup(flags, name)    ((flags) & AuIcpup_##name)
16445 +#define au_fset_icpup(flags, name) \
16446 +       do { (flags) |= AuIcpup_##name; } while (0)
16447 +#define au_fclr_icpup(flags, name) \
16448 +       do { (flags) &= ~AuIcpup_##name; } while (0)
16449 +
16450 +struct au_icpup_args {
16451 +       unsigned char flags;
16452 +       unsigned char pin_flags;
16453 +       aufs_bindex_t btgt;
16454 +       unsigned int udba;
16455 +       struct au_pin pin;
16456 +       struct path h_path;
16457 +       struct inode *h_inode;
16458 +};
16459 +
16460 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
16461 +                    struct au_icpup_args *a);
16462 +
16463 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path);
16464 +
16465 +/* i_op_add.c */
16466 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
16467 +              struct dentry *h_parent, int isdir);
16468 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
16469 +              dev_t dev);
16470 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname);
16471 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
16472 +               bool want_excl);
16473 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode);
16474 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
16475 +             struct dentry *dentry);
16476 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
16477 +
16478 +/* i_op_del.c */
16479 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup);
16480 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
16481 +              struct dentry *h_parent, int isdir);
16482 +int aufs_unlink(struct inode *dir, struct dentry *dentry);
16483 +int aufs_rmdir(struct inode *dir, struct dentry *dentry);
16484 +
16485 +/* i_op_ren.c */
16486 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt);
16487 +int aufs_rename(struct inode *src_dir, struct dentry *src_dentry,
16488 +               struct inode *dir, struct dentry *dentry);
16489 +
16490 +/* iinfo.c */
16491 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex);
16492 +void au_hiput(struct au_hinode *hinode);
16493 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
16494 +                 struct dentry *h_wh);
16495 +unsigned int au_hi_flags(struct inode *inode, int isdir);
16496 +
16497 +/* hinode flags */
16498 +#define AuHi_XINO      1
16499 +#define AuHi_HNOTIFY   (1 << 1)
16500 +#define au_ftest_hi(flags, name)       ((flags) & AuHi_##name)
16501 +#define au_fset_hi(flags, name) \
16502 +       do { (flags) |= AuHi_##name; } while (0)
16503 +#define au_fclr_hi(flags, name) \
16504 +       do { (flags) &= ~AuHi_##name; } while (0)
16505 +
16506 +#ifndef CONFIG_AUFS_HNOTIFY
16507 +#undef AuHi_HNOTIFY
16508 +#define AuHi_HNOTIFY   0
16509 +#endif
16510 +
16511 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
16512 +                  struct inode *h_inode, unsigned int flags);
16513 +
16514 +void au_update_iigen(struct inode *inode, int half);
16515 +void au_update_ibrange(struct inode *inode, int do_put_zero);
16516 +
16517 +void au_icntnr_init_once(void *_c);
16518 +int au_iinfo_init(struct inode *inode);
16519 +void au_iinfo_fin(struct inode *inode);
16520 +int au_ii_realloc(struct au_iinfo *iinfo, int nbr);
16521 +
16522 +#ifdef CONFIG_PROC_FS
16523 +/* plink.c */
16524 +int au_plink_maint(struct super_block *sb, int flags);
16525 +void au_plink_maint_leave(struct au_sbinfo *sbinfo);
16526 +int au_plink_maint_enter(struct super_block *sb);
16527 +#ifdef CONFIG_AUFS_DEBUG
16528 +void au_plink_list(struct super_block *sb);
16529 +#else
16530 +AuStubVoid(au_plink_list, struct super_block *sb)
16531 +#endif
16532 +int au_plink_test(struct inode *inode);
16533 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex);
16534 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
16535 +                    struct dentry *h_dentry);
16536 +void au_plink_put(struct super_block *sb, int verbose);
16537 +void au_plink_clean(struct super_block *sb, int verbose);
16538 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id);
16539 +#else
16540 +AuStubInt0(au_plink_maint, struct super_block *sb, int flags);
16541 +AuStubVoid(au_plink_maint_leave, struct au_sbinfo *sbinfo);
16542 +AuStubInt0(au_plink_maint_enter, struct super_block *sb);
16543 +AuStubVoid(au_plink_list, struct super_block *sb);
16544 +AuStubInt0(au_plink_test, struct inode *inode);
16545 +AuStub(struct dentry *, au_plink_lkup, return NULL,
16546 +       struct inode *inode, aufs_bindex_t bindex);
16547 +AuStubVoid(au_plink_append, struct inode *inode, aufs_bindex_t bindex,
16548 +          struct dentry *h_dentry);
16549 +AuStubVoid(au_plink_put, struct super_block *sb, int verbose);
16550 +AuStubVoid(au_plink_clean, struct super_block *sb, int verbose);
16551 +AuStubVoid(au_plink_half_refresh, struct super_block *sb, aufs_bindex_t br_id);
16552 +#endif /* CONFIG_PROC_FS */
16553 +
16554 +#ifdef CONFIG_AUFS_XATTR
16555 +/* xattr.c */
16556 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags);
16557 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size);
16558 +ssize_t aufs_getxattr(struct dentry *dentry, const char *name, void *value,
16559 +                     size_t size);
16560 +int aufs_setxattr(struct dentry *dentry, const char *name, const void *value,
16561 +                 size_t size, int flags);
16562 +int aufs_removexattr(struct dentry *dentry, const char *name);
16563 +
16564 +/* void au_xattr_init(struct super_block *sb); */
16565 +#else
16566 +AuStubInt0(au_cpup_xattr, struct dentry *h_dst, struct dentry *h_src,
16567 +          int ignore_flags);
16568 +/* AuStubVoid(au_xattr_init, struct super_block *sb); */
16569 +#endif
16570 +
16571 +#ifdef CONFIG_FS_POSIX_ACL
16572 +struct posix_acl *aufs_get_acl(struct inode *inode, int type);
16573 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type);
16574 +#endif
16575 +
16576 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
16577 +enum {
16578 +       AU_XATTR_SET,
16579 +       AU_XATTR_REMOVE,
16580 +       AU_ACL_SET
16581 +};
16582 +
16583 +struct au_srxattr {
16584 +       int type;
16585 +       union {
16586 +               struct {
16587 +                       const char      *name;
16588 +                       const void      *value;
16589 +                       size_t          size;
16590 +                       int             flags;
16591 +               } set;
16592 +               struct {
16593 +                       const char      *name;
16594 +               } remove;
16595 +               struct {
16596 +                       struct posix_acl *acl;
16597 +                       int             type;
16598 +               } acl_set;
16599 +       } u;
16600 +};
16601 +ssize_t au_srxattr(struct dentry *dentry, struct au_srxattr *arg);
16602 +#endif
16603 +
16604 +/* ---------------------------------------------------------------------- */
16605 +
16606 +/* lock subclass for iinfo */
16607 +enum {
16608 +       AuLsc_II_CHILD,         /* child first */
16609 +       AuLsc_II_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
16610 +       AuLsc_II_CHILD3,        /* copyup dirs */
16611 +       AuLsc_II_PARENT,        /* see AuLsc_I_PARENT in vfsub.h */
16612 +       AuLsc_II_PARENT2,
16613 +       AuLsc_II_PARENT3,       /* copyup dirs */
16614 +       AuLsc_II_NEW_CHILD
16615 +};
16616 +
16617 +/*
16618 + * ii_read_lock_child, ii_write_lock_child,
16619 + * ii_read_lock_child2, ii_write_lock_child2,
16620 + * ii_read_lock_child3, ii_write_lock_child3,
16621 + * ii_read_lock_parent, ii_write_lock_parent,
16622 + * ii_read_lock_parent2, ii_write_lock_parent2,
16623 + * ii_read_lock_parent3, ii_write_lock_parent3,
16624 + * ii_read_lock_new_child, ii_write_lock_new_child,
16625 + */
16626 +#define AuReadLockFunc(name, lsc) \
16627 +static inline void ii_read_lock_##name(struct inode *i) \
16628 +{ \
16629 +       au_rw_read_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
16630 +}
16631 +
16632 +#define AuWriteLockFunc(name, lsc) \
16633 +static inline void ii_write_lock_##name(struct inode *i) \
16634 +{ \
16635 +       au_rw_write_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
16636 +}
16637 +
16638 +#define AuRWLockFuncs(name, lsc) \
16639 +       AuReadLockFunc(name, lsc) \
16640 +       AuWriteLockFunc(name, lsc)
16641 +
16642 +AuRWLockFuncs(child, CHILD);
16643 +AuRWLockFuncs(child2, CHILD2);
16644 +AuRWLockFuncs(child3, CHILD3);
16645 +AuRWLockFuncs(parent, PARENT);
16646 +AuRWLockFuncs(parent2, PARENT2);
16647 +AuRWLockFuncs(parent3, PARENT3);
16648 +AuRWLockFuncs(new_child, NEW_CHILD);
16649 +
16650 +#undef AuReadLockFunc
16651 +#undef AuWriteLockFunc
16652 +#undef AuRWLockFuncs
16653 +
16654 +/*
16655 + * ii_read_unlock, ii_write_unlock, ii_downgrade_lock
16656 + */
16657 +AuSimpleUnlockRwsemFuncs(ii, struct inode *i, &au_ii(i)->ii_rwsem);
16658 +
16659 +#define IiMustNoWaiters(i)     AuRwMustNoWaiters(&au_ii(i)->ii_rwsem)
16660 +#define IiMustAnyLock(i)       AuRwMustAnyLock(&au_ii(i)->ii_rwsem)
16661 +#define IiMustWriteLock(i)     AuRwMustWriteLock(&au_ii(i)->ii_rwsem)
16662 +
16663 +/* ---------------------------------------------------------------------- */
16664 +
16665 +static inline void au_icntnr_init(struct au_icntnr *c)
16666 +{
16667 +#ifdef CONFIG_AUFS_DEBUG
16668 +       c->vfs_inode.i_mode = 0;
16669 +#endif
16670 +}
16671 +
16672 +static inline unsigned int au_iigen(struct inode *inode, struct au_iigen *iigen)
16673 +{
16674 +       unsigned int gen;
16675 +       struct au_iinfo *iinfo;
16676 +
16677 +       iinfo = au_ii(inode);
16678 +       spin_lock(&iinfo->ii_genspin);
16679 +       if (iigen)
16680 +               *iigen = iinfo->ii_generation;
16681 +       gen = iinfo->ii_generation.ig_generation;
16682 +       spin_unlock(&iinfo->ii_genspin);
16683 +
16684 +       return gen;
16685 +}
16686 +
16687 +/* tiny test for inode number */
16688 +/* tmpfs generation is too rough */
16689 +static inline int au_test_higen(struct inode *inode, struct inode *h_inode)
16690 +{
16691 +       struct au_iinfo *iinfo;
16692 +
16693 +       iinfo = au_ii(inode);
16694 +       AuRwMustAnyLock(&iinfo->ii_rwsem);
16695 +       return !(iinfo->ii_hsb1 == h_inode->i_sb
16696 +                && iinfo->ii_higen == h_inode->i_generation);
16697 +}
16698 +
16699 +static inline void au_iigen_dec(struct inode *inode)
16700 +{
16701 +       struct au_iinfo *iinfo;
16702 +
16703 +       iinfo = au_ii(inode);
16704 +       spin_lock(&iinfo->ii_genspin);
16705 +       iinfo->ii_generation.ig_generation--;
16706 +       spin_unlock(&iinfo->ii_genspin);
16707 +}
16708 +
16709 +static inline int au_iigen_test(struct inode *inode, unsigned int sigen)
16710 +{
16711 +       int err;
16712 +
16713 +       err = 0;
16714 +       if (unlikely(inode && au_iigen(inode, NULL) != sigen))
16715 +               err = -EIO;
16716 +
16717 +       return err;
16718 +}
16719 +
16720 +/* ---------------------------------------------------------------------- */
16721 +
16722 +static inline aufs_bindex_t au_ii_br_id(struct inode *inode,
16723 +                                       aufs_bindex_t bindex)
16724 +{
16725 +       IiMustAnyLock(inode);
16726 +       return au_ii(inode)->ii_hinode[0 + bindex].hi_id;
16727 +}
16728 +
16729 +static inline aufs_bindex_t au_ibstart(struct inode *inode)
16730 +{
16731 +       IiMustAnyLock(inode);
16732 +       return au_ii(inode)->ii_bstart;
16733 +}
16734 +
16735 +static inline aufs_bindex_t au_ibend(struct inode *inode)
16736 +{
16737 +       IiMustAnyLock(inode);
16738 +       return au_ii(inode)->ii_bend;
16739 +}
16740 +
16741 +static inline struct au_vdir *au_ivdir(struct inode *inode)
16742 +{
16743 +       IiMustAnyLock(inode);
16744 +       return au_ii(inode)->ii_vdir;
16745 +}
16746 +
16747 +static inline struct dentry *au_hi_wh(struct inode *inode, aufs_bindex_t bindex)
16748 +{
16749 +       IiMustAnyLock(inode);
16750 +       return au_ii(inode)->ii_hinode[0 + bindex].hi_whdentry;
16751 +}
16752 +
16753 +static inline void au_set_ibstart(struct inode *inode, aufs_bindex_t bindex)
16754 +{
16755 +       IiMustWriteLock(inode);
16756 +       au_ii(inode)->ii_bstart = bindex;
16757 +}
16758 +
16759 +static inline void au_set_ibend(struct inode *inode, aufs_bindex_t bindex)
16760 +{
16761 +       IiMustWriteLock(inode);
16762 +       au_ii(inode)->ii_bend = bindex;
16763 +}
16764 +
16765 +static inline void au_set_ivdir(struct inode *inode, struct au_vdir *vdir)
16766 +{
16767 +       IiMustWriteLock(inode);
16768 +       au_ii(inode)->ii_vdir = vdir;
16769 +}
16770 +
16771 +static inline struct au_hinode *au_hi(struct inode *inode, aufs_bindex_t bindex)
16772 +{
16773 +       IiMustAnyLock(inode);
16774 +       return au_ii(inode)->ii_hinode + bindex;
16775 +}
16776 +
16777 +/* ---------------------------------------------------------------------- */
16778 +
16779 +static inline struct dentry *au_pinned_parent(struct au_pin *pin)
16780 +{
16781 +       if (pin)
16782 +               return pin->parent;
16783 +       return NULL;
16784 +}
16785 +
16786 +static inline struct inode *au_pinned_h_dir(struct au_pin *pin)
16787 +{
16788 +       if (pin && pin->hdir)
16789 +               return pin->hdir->hi_inode;
16790 +       return NULL;
16791 +}
16792 +
16793 +static inline struct au_hinode *au_pinned_hdir(struct au_pin *pin)
16794 +{
16795 +       if (pin)
16796 +               return pin->hdir;
16797 +       return NULL;
16798 +}
16799 +
16800 +static inline void au_pin_set_dentry(struct au_pin *pin, struct dentry *dentry)
16801 +{
16802 +       if (pin)
16803 +               pin->dentry = dentry;
16804 +}
16805 +
16806 +static inline void au_pin_set_parent_lflag(struct au_pin *pin,
16807 +                                          unsigned char lflag)
16808 +{
16809 +       if (pin) {
16810 +               if (lflag)
16811 +                       au_fset_pin(pin->flags, DI_LOCKED);
16812 +               else
16813 +                       au_fclr_pin(pin->flags, DI_LOCKED);
16814 +       }
16815 +}
16816 +
16817 +static inline void au_pin_set_parent(struct au_pin *pin, struct dentry *parent)
16818 +{
16819 +       if (pin) {
16820 +               dput(pin->parent);
16821 +               pin->parent = dget(parent);
16822 +       }
16823 +}
16824 +
16825 +/* ---------------------------------------------------------------------- */
16826 +
16827 +struct au_branch;
16828 +#ifdef CONFIG_AUFS_HNOTIFY
16829 +struct au_hnotify_op {
16830 +       void (*ctl)(struct au_hinode *hinode, int do_set);
16831 +       int (*alloc)(struct au_hinode *hinode);
16832 +
16833 +       /*
16834 +        * if it returns true, the the caller should free hinode->hi_notify,
16835 +        * otherwise ->free() frees it.
16836 +        */
16837 +       int (*free)(struct au_hinode *hinode,
16838 +                   struct au_hnotify *hn) __must_check;
16839 +
16840 +       void (*fin)(void);
16841 +       int (*init)(void);
16842 +
16843 +       int (*reset_br)(unsigned int udba, struct au_branch *br, int perm);
16844 +       void (*fin_br)(struct au_branch *br);
16845 +       int (*init_br)(struct au_branch *br, int perm);
16846 +};
16847 +
16848 +/* hnotify.c */
16849 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode);
16850 +void au_hn_free(struct au_hinode *hinode);
16851 +void au_hn_ctl(struct au_hinode *hinode, int do_set);
16852 +void au_hn_reset(struct inode *inode, unsigned int flags);
16853 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
16854 +              struct qstr *h_child_qstr, struct inode *h_child_inode);
16855 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm);
16856 +int au_hnotify_init_br(struct au_branch *br, int perm);
16857 +void au_hnotify_fin_br(struct au_branch *br);
16858 +int __init au_hnotify_init(void);
16859 +void au_hnotify_fin(void);
16860 +
16861 +/* hfsnotify.c */
16862 +extern const struct au_hnotify_op au_hnotify_op;
16863 +
16864 +static inline
16865 +void au_hn_init(struct au_hinode *hinode)
16866 +{
16867 +       hinode->hi_notify = NULL;
16868 +}
16869 +
16870 +static inline struct au_hnotify *au_hn(struct au_hinode *hinode)
16871 +{
16872 +       return hinode->hi_notify;
16873 +}
16874 +
16875 +#else
16876 +AuStub(int, au_hn_alloc, return -EOPNOTSUPP,
16877 +       struct au_hinode *hinode __maybe_unused,
16878 +       struct inode *inode __maybe_unused)
16879 +AuStub(struct au_hnotify *, au_hn, return NULL, struct au_hinode *hinode)
16880 +AuStubVoid(au_hn_free, struct au_hinode *hinode __maybe_unused)
16881 +AuStubVoid(au_hn_ctl, struct au_hinode *hinode __maybe_unused,
16882 +          int do_set __maybe_unused)
16883 +AuStubVoid(au_hn_reset, struct inode *inode __maybe_unused,
16884 +          unsigned int flags __maybe_unused)
16885 +AuStubInt0(au_hnotify_reset_br, unsigned int udba __maybe_unused,
16886 +          struct au_branch *br __maybe_unused,
16887 +          int perm __maybe_unused)
16888 +AuStubInt0(au_hnotify_init_br, struct au_branch *br __maybe_unused,
16889 +          int perm __maybe_unused)
16890 +AuStubVoid(au_hnotify_fin_br, struct au_branch *br __maybe_unused)
16891 +AuStubInt0(__init au_hnotify_init, void)
16892 +AuStubVoid(au_hnotify_fin, void)
16893 +AuStubVoid(au_hn_init, struct au_hinode *hinode __maybe_unused)
16894 +#endif /* CONFIG_AUFS_HNOTIFY */
16895 +
16896 +static inline void au_hn_suspend(struct au_hinode *hdir)
16897 +{
16898 +       au_hn_ctl(hdir, /*do_set*/0);
16899 +}
16900 +
16901 +static inline void au_hn_resume(struct au_hinode *hdir)
16902 +{
16903 +       au_hn_ctl(hdir, /*do_set*/1);
16904 +}
16905 +
16906 +static inline void au_hn_imtx_lock(struct au_hinode *hdir)
16907 +{
16908 +       mutex_lock(&hdir->hi_inode->i_mutex);
16909 +       au_hn_suspend(hdir);
16910 +}
16911 +
16912 +static inline void au_hn_imtx_lock_nested(struct au_hinode *hdir,
16913 +                                         unsigned int sc __maybe_unused)
16914 +{
16915 +       mutex_lock_nested(&hdir->hi_inode->i_mutex, sc);
16916 +       au_hn_suspend(hdir);
16917 +}
16918 +
16919 +static inline void au_hn_imtx_unlock(struct au_hinode *hdir)
16920 +{
16921 +       au_hn_resume(hdir);
16922 +       mutex_unlock(&hdir->hi_inode->i_mutex);
16923 +}
16924 +
16925 +#endif /* __KERNEL__ */
16926 +#endif /* __AUFS_INODE_H__ */
16927 diff -urN /usr/share/empty/fs/aufs/ioctl.c linux/fs/aufs/ioctl.c
16928 --- /usr/share/empty/fs/aufs/ioctl.c    1970-01-01 01:00:00.000000000 +0100
16929 +++ linux/fs/aufs/ioctl.c       2015-03-27 21:56:35.466795000 +0100
16930 @@ -0,0 +1,219 @@
16931 +/*
16932 + * Copyright (C) 2005-2015 Junjiro R. Okajima
16933 + *
16934 + * This program, aufs is free software; you can redistribute it and/or modify
16935 + * it under the terms of the GNU General Public License as published by
16936 + * the Free Software Foundation; either version 2 of the License, or
16937 + * (at your option) any later version.
16938 + *
16939 + * This program is distributed in the hope that it will be useful,
16940 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16941 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16942 + * GNU General Public License for more details.
16943 + *
16944 + * You should have received a copy of the GNU General Public License
16945 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16946 + */
16947 +
16948 +/*
16949 + * ioctl
16950 + * plink-management and readdir in userspace.
16951 + * assist the pathconf(3) wrapper library.
16952 + * move-down
16953 + * File-based Hierarchical Storage Management.
16954 + */
16955 +
16956 +#include <linux/compat.h>
16957 +#include <linux/file.h>
16958 +#include "aufs.h"
16959 +
16960 +static int au_wbr_fd(struct path *path, struct aufs_wbr_fd __user *arg)
16961 +{
16962 +       int err, fd;
16963 +       aufs_bindex_t wbi, bindex, bend;
16964 +       struct file *h_file;
16965 +       struct super_block *sb;
16966 +       struct dentry *root;
16967 +       struct au_branch *br;
16968 +       struct aufs_wbr_fd wbrfd = {
16969 +               .oflags = au_dir_roflags,
16970 +               .brid   = -1
16971 +       };
16972 +       const int valid = O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_DIRECTORY
16973 +               | O_NOATIME | O_CLOEXEC;
16974 +
16975 +       AuDebugOn(wbrfd.oflags & ~valid);
16976 +
16977 +       if (arg) {
16978 +               err = copy_from_user(&wbrfd, arg, sizeof(wbrfd));
16979 +               if (unlikely(err)) {
16980 +                       err = -EFAULT;
16981 +                       goto out;
16982 +               }
16983 +
16984 +               err = -EINVAL;
16985 +               AuDbg("wbrfd{0%o, %d}\n", wbrfd.oflags, wbrfd.brid);
16986 +               wbrfd.oflags |= au_dir_roflags;
16987 +               AuDbg("0%o\n", wbrfd.oflags);
16988 +               if (unlikely(wbrfd.oflags & ~valid))
16989 +                       goto out;
16990 +       }
16991 +
16992 +       fd = get_unused_fd_flags(0);
16993 +       err = fd;
16994 +       if (unlikely(fd < 0))
16995 +               goto out;
16996 +
16997 +       h_file = ERR_PTR(-EINVAL);
16998 +       wbi = 0;
16999 +       br = NULL;
17000 +       sb = path->dentry->d_sb;
17001 +       root = sb->s_root;
17002 +       aufs_read_lock(root, AuLock_IR);
17003 +       bend = au_sbend(sb);
17004 +       if (wbrfd.brid >= 0) {
17005 +               wbi = au_br_index(sb, wbrfd.brid);
17006 +               if (unlikely(wbi < 0 || wbi > bend))
17007 +                       goto out_unlock;
17008 +       }
17009 +
17010 +       h_file = ERR_PTR(-ENOENT);
17011 +       br = au_sbr(sb, wbi);
17012 +       if (!au_br_writable(br->br_perm)) {
17013 +               if (arg)
17014 +                       goto out_unlock;
17015 +
17016 +               bindex = wbi + 1;
17017 +               wbi = -1;
17018 +               for (; bindex <= bend; bindex++) {
17019 +                       br = au_sbr(sb, bindex);
17020 +                       if (au_br_writable(br->br_perm)) {
17021 +                               wbi = bindex;
17022 +                               br = au_sbr(sb, wbi);
17023 +                               break;
17024 +                       }
17025 +               }
17026 +       }
17027 +       AuDbg("wbi %d\n", wbi);
17028 +       if (wbi >= 0)
17029 +               h_file = au_h_open(root, wbi, wbrfd.oflags, NULL,
17030 +                                  /*force_wr*/0);
17031 +
17032 +out_unlock:
17033 +       aufs_read_unlock(root, AuLock_IR);
17034 +       err = PTR_ERR(h_file);
17035 +       if (IS_ERR(h_file))
17036 +               goto out_fd;
17037 +
17038 +       atomic_dec(&br->br_count); /* cf. au_h_open() */
17039 +       fd_install(fd, h_file);
17040 +       err = fd;
17041 +       goto out; /* success */
17042 +
17043 +out_fd:
17044 +       put_unused_fd(fd);
17045 +out:
17046 +       AuTraceErr(err);
17047 +       return err;
17048 +}
17049 +
17050 +/* ---------------------------------------------------------------------- */
17051 +
17052 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg)
17053 +{
17054 +       long err;
17055 +       struct dentry *dentry;
17056 +
17057 +       switch (cmd) {
17058 +       case AUFS_CTL_RDU:
17059 +       case AUFS_CTL_RDU_INO:
17060 +               err = au_rdu_ioctl(file, cmd, arg);
17061 +               break;
17062 +
17063 +       case AUFS_CTL_WBR_FD:
17064 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17065 +               break;
17066 +
17067 +       case AUFS_CTL_IBUSY:
17068 +               err = au_ibusy_ioctl(file, arg);
17069 +               break;
17070 +
17071 +       case AUFS_CTL_BRINFO:
17072 +               err = au_brinfo_ioctl(file, arg);
17073 +               break;
17074 +
17075 +       case AUFS_CTL_FHSM_FD:
17076 +               dentry = file->f_path.dentry;
17077 +               if (IS_ROOT(dentry))
17078 +                       err = au_fhsm_fd(dentry->d_sb, arg);
17079 +               else
17080 +                       err = -ENOTTY;
17081 +               break;
17082 +
17083 +       default:
17084 +               /* do not call the lower */
17085 +               AuDbg("0x%x\n", cmd);
17086 +               err = -ENOTTY;
17087 +       }
17088 +
17089 +       AuTraceErr(err);
17090 +       return err;
17091 +}
17092 +
17093 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg)
17094 +{
17095 +       long err;
17096 +
17097 +       switch (cmd) {
17098 +       case AUFS_CTL_MVDOWN:
17099 +               err = au_mvdown(file->f_path.dentry, (void __user *)arg);
17100 +               break;
17101 +
17102 +       case AUFS_CTL_WBR_FD:
17103 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
17104 +               break;
17105 +
17106 +       default:
17107 +               /* do not call the lower */
17108 +               AuDbg("0x%x\n", cmd);
17109 +               err = -ENOTTY;
17110 +       }
17111 +
17112 +       AuTraceErr(err);
17113 +       return err;
17114 +}
17115 +
17116 +#ifdef CONFIG_COMPAT
17117 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
17118 +                          unsigned long arg)
17119 +{
17120 +       long err;
17121 +
17122 +       switch (cmd) {
17123 +       case AUFS_CTL_RDU:
17124 +       case AUFS_CTL_RDU_INO:
17125 +               err = au_rdu_compat_ioctl(file, cmd, arg);
17126 +               break;
17127 +
17128 +       case AUFS_CTL_IBUSY:
17129 +               err = au_ibusy_compat_ioctl(file, arg);
17130 +               break;
17131 +
17132 +       case AUFS_CTL_BRINFO:
17133 +               err = au_brinfo_compat_ioctl(file, arg);
17134 +               break;
17135 +
17136 +       default:
17137 +               err = aufs_ioctl_dir(file, cmd, arg);
17138 +       }
17139 +
17140 +       AuTraceErr(err);
17141 +       return err;
17142 +}
17143 +
17144 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
17145 +                             unsigned long arg)
17146 +{
17147 +       return aufs_ioctl_nondir(file, cmd, (unsigned long)compat_ptr(arg));
17148 +}
17149 +#endif
17150 diff -urN /usr/share/empty/fs/aufs/i_op_add.c linux/fs/aufs/i_op_add.c
17151 --- /usr/share/empty/fs/aufs/i_op_add.c 1970-01-01 01:00:00.000000000 +0100
17152 +++ linux/fs/aufs/i_op_add.c    2015-03-27 21:56:35.463461668 +0100
17153 @@ -0,0 +1,896 @@
17154 +/*
17155 + * Copyright (C) 2005-2015 Junjiro R. Okajima
17156 + *
17157 + * This program, aufs is free software; you can redistribute it and/or modify
17158 + * it under the terms of the GNU General Public License as published by
17159 + * the Free Software Foundation; either version 2 of the License, or
17160 + * (at your option) any later version.
17161 + *
17162 + * This program is distributed in the hope that it will be useful,
17163 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
17164 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17165 + * GNU General Public License for more details.
17166 + *
17167 + * You should have received a copy of the GNU General Public License
17168 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
17169 + */
17170 +
17171 +/*
17172 + * inode operations (add entry)
17173 + */
17174 +
17175 +#include "aufs.h"
17176 +
17177 +/*
17178 + * final procedure of adding a new entry, except link(2).
17179 + * remove whiteout, instantiate, copyup the parent dir's times and size
17180 + * and update version.
17181 + * if it failed, re-create the removed whiteout.
17182 + */
17183 +static int epilog(struct inode *dir, aufs_bindex_t bindex,
17184 +                 struct dentry *wh_dentry, struct dentry *dentry)
17185 +{
17186 +       int err, rerr;
17187 +       aufs_bindex_t bwh;
17188 +       struct path h_path;
17189 +       struct super_block *sb;
17190 +       struct inode *inode, *h_dir;
17191 +       struct dentry *wh;
17192 +
17193 +       bwh = -1;
17194 +       sb = dir->i_sb;
17195 +       if (wh_dentry) {
17196 +               h_dir = wh_dentry->d_parent->d_inode; /* dir inode is locked */
17197 +               IMustLock(h_dir);
17198 +               AuDebugOn(au_h_iptr(dir, bindex) != h_dir);
17199 +               bwh = au_dbwh(dentry);
17200 +               h_path.dentry = wh_dentry;
17201 +               h_path.mnt = au_sbr_mnt(sb, bindex);
17202 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path,
17203 +                                         dentry);
17204 +               if (unlikely(err))
17205 +                       goto out;
17206 +       }
17207 +
17208 +       inode = au_new_inode(dentry, /*must_new*/1);
17209 +       if (!IS_ERR(inode)) {
17210 +               d_instantiate(dentry, inode);
17211 +               dir = dentry->d_parent->d_inode; /* dir inode is locked */
17212 +               IMustLock(dir);
17213 +               if (au_ibstart(dir) == au_dbstart(dentry))
17214 +                       au_cpup_attr_timesizes(dir);
17215 +               dir->i_version++;
17216 +               au_fhsm_wrote(sb, bindex, /*force*/0);
17217 +               return 0; /* success */
17218 +       }
17219 +
17220 +       err = PTR_ERR(inode);
17221 +       if (!wh_dentry)
17222 +               goto out;
17223 +
17224 +       /* revert */
17225 +       /* dir inode is locked */
17226 +       wh = au_wh_create(dentry, bwh, wh_dentry->d_parent);
17227 +       rerr = PTR_ERR(wh);
17228 +       if (IS_ERR(wh)) {
17229 +               AuIOErr("%pd reverting whiteout failed(%d, %d)\n",
17230 +                       dentry, err, rerr);
17231 +               err = -EIO;
17232 +       } else
17233 +               dput(wh);
17234 +
17235 +out:
17236 +       return err;
17237 +}
17238 +
17239 +static int au_d_may_add(struct dentry *dentry)
17240 +{
17241 +       int err;
17242 +
17243 +       err = 0;
17244 +       if (unlikely(d_unhashed(dentry)))
17245 +               err = -ENOENT;
17246 +       if (unlikely(dentry->d_inode))
17247 +               err = -EEXIST;
17248 +       return err;
17249 +}
17250 +
17251 +/*
17252 + * simple tests for the adding inode operations.
17253 + * following the checks in vfs, plus the parent-child relationship.
17254 + */
17255 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
17256 +              struct dentry *h_parent, int isdir)
17257 +{
17258 +       int err;
17259 +       umode_t h_mode;
17260 +       struct dentry *h_dentry;
17261 +       struct inode *h_inode;
17262 +
17263 +       err = -ENAMETOOLONG;
17264 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17265 +               goto out;
17266 +
17267 +       h_dentry = au_h_dptr(dentry, bindex);
17268 +       h_inode = h_dentry->d_inode;
17269 +       if (!dentry->d_inode) {
17270 +               err = -EEXIST;
17271 +               if (unlikely(h_inode))
17272 +                       goto out;
17273 +       } else {
17274 +               /* rename(2) case */
17275 +               err = -EIO;
17276 +               if (unlikely(!h_inode || !h_inode->i_nlink))
17277 +                       goto out;
17278 +
17279 +               h_mode = h_inode->i_mode;
17280 +               if (!isdir) {
17281 +                       err = -EISDIR;
17282 +                       if (unlikely(S_ISDIR(h_mode)))
17283 +                               goto out;
17284 +               } else if (unlikely(!S_ISDIR(h_mode))) {
17285 +                       err = -ENOTDIR;
17286 +                       goto out;
17287 +               }
17288 +       }
17289 +
17290 +       err = 0;
17291 +       /* expected parent dir is locked */
17292 +       if (unlikely(h_parent != h_dentry->d_parent))
17293 +               err = -EIO;
17294 +
17295 +out:
17296 +       AuTraceErr(err);
17297 +       return err;
17298 +}
17299 +
17300 +/*
17301 + * initial procedure of adding a new entry.
17302 + * prepare writable branch and the parent dir, lock it,
17303 + * and lookup whiteout for the new entry.
17304 + */
17305 +static struct dentry*
17306 +lock_hdir_lkup_wh(struct dentry *dentry, struct au_dtime *dt,
17307 +                 struct dentry *src_dentry, struct au_pin *pin,
17308 +                 struct au_wr_dir_args *wr_dir_args)
17309 +{
17310 +       struct dentry *wh_dentry, *h_parent;
17311 +       struct super_block *sb;
17312 +       struct au_branch *br;
17313 +       int err;
17314 +       unsigned int udba;
17315 +       aufs_bindex_t bcpup;
17316 +
17317 +       AuDbg("%pd\n", dentry);
17318 +
17319 +       err = au_wr_dir(dentry, src_dentry, wr_dir_args);
17320 +       bcpup = err;
17321 +       wh_dentry = ERR_PTR(err);
17322 +       if (unlikely(err < 0))
17323 +               goto out;
17324 +
17325 +       sb = dentry->d_sb;
17326 +       udba = au_opt_udba(sb);
17327 +       err = au_pin(pin, dentry, bcpup, udba,
17328 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17329 +       wh_dentry = ERR_PTR(err);
17330 +       if (unlikely(err))
17331 +               goto out;
17332 +
17333 +       h_parent = au_pinned_h_parent(pin);
17334 +       if (udba != AuOpt_UDBA_NONE
17335 +           && au_dbstart(dentry) == bcpup)
17336 +               err = au_may_add(dentry, bcpup, h_parent,
17337 +                                au_ftest_wrdir(wr_dir_args->flags, ISDIR));
17338 +       else if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17339 +               err = -ENAMETOOLONG;
17340 +       wh_dentry = ERR_PTR(err);
17341 +       if (unlikely(err))
17342 +               goto out_unpin;
17343 +
17344 +       br = au_sbr(sb, bcpup);
17345 +       if (dt) {
17346 +               struct path tmp = {
17347 +                       .dentry = h_parent,
17348 +                       .mnt    = au_br_mnt(br)
17349 +               };
17350 +               au_dtime_store(dt, au_pinned_parent(pin), &tmp);
17351 +       }
17352 +
17353 +       wh_dentry = NULL;
17354 +       if (bcpup != au_dbwh(dentry))
17355 +               goto out; /* success */
17356 +
17357 +       /*
17358 +        * ENAMETOOLONG here means that if we allowed create such name, then it
17359 +        * would not be able to removed in the future. So we don't allow such
17360 +        * name here and we don't handle ENAMETOOLONG differently here.
17361 +        */
17362 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
17363 +
17364 +out_unpin:
17365 +       if (IS_ERR(wh_dentry))
17366 +               au_unpin(pin);
17367 +out:
17368 +       return wh_dentry;
17369 +}
17370 +
17371 +/* ---------------------------------------------------------------------- */
17372 +
17373 +enum { Mknod, Symlink, Creat };
17374 +struct simple_arg {
17375 +       int type;
17376 +       union {
17377 +               struct {
17378 +                       umode_t mode;
17379 +                       bool want_excl;
17380 +               } c;
17381 +               struct {
17382 +                       const char *symname;
17383 +               } s;
17384 +               struct {
17385 +                       umode_t mode;
17386 +                       dev_t dev;
17387 +               } m;
17388 +       } u;
17389 +};
17390 +
17391 +static int add_simple(struct inode *dir, struct dentry *dentry,
17392 +                     struct simple_arg *arg)
17393 +{
17394 +       int err, rerr;
17395 +       aufs_bindex_t bstart;
17396 +       unsigned char created;
17397 +       struct dentry *wh_dentry, *parent;
17398 +       struct inode *h_dir;
17399 +       /* to reuduce stack size */
17400 +       struct {
17401 +               struct au_dtime dt;
17402 +               struct au_pin pin;
17403 +               struct path h_path;
17404 +               struct au_wr_dir_args wr_dir_args;
17405 +       } *a;
17406 +
17407 +       AuDbg("%pd\n", dentry);
17408 +       IMustLock(dir);
17409 +
17410 +       err = -ENOMEM;
17411 +       a = kmalloc(sizeof(*a), GFP_NOFS);
17412 +       if (unlikely(!a))
17413 +               goto out;
17414 +       a->wr_dir_args.force_btgt = -1;
17415 +       a->wr_dir_args.flags = AuWrDir_ADD_ENTRY;
17416 +
17417 +       parent = dentry->d_parent; /* dir inode is locked */
17418 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
17419 +       if (unlikely(err))
17420 +               goto out_free;
17421 +       err = au_d_may_add(dentry);
17422 +       if (unlikely(err))
17423 +               goto out_unlock;
17424 +       di_write_lock_parent(parent);
17425 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
17426 +                                     &a->pin, &a->wr_dir_args);
17427 +       err = PTR_ERR(wh_dentry);
17428 +       if (IS_ERR(wh_dentry))
17429 +               goto out_parent;
17430 +
17431 +       bstart = au_dbstart(dentry);
17432 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
17433 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, bstart);
17434 +       h_dir = au_pinned_h_dir(&a->pin);
17435 +       switch (arg->type) {
17436 +       case Creat:
17437 +               err = vfsub_create(h_dir, &a->h_path, arg->u.c.mode,
17438 +                                  arg->u.c.want_excl);
17439 +               break;
17440 +       case Symlink:
17441 +               err = vfsub_symlink(h_dir, &a->h_path, arg->u.s.symname);
17442 +               break;
17443 +       case Mknod:
17444 +               err = vfsub_mknod(h_dir, &a->h_path, arg->u.m.mode,
17445 +                                 arg->u.m.dev);
17446 +               break;
17447 +       default:
17448 +               BUG();
17449 +       }
17450 +       created = !err;
17451 +       if (!err)
17452 +               err = epilog(dir, bstart, wh_dentry, dentry);
17453 +
17454 +       /* revert */
17455 +       if (unlikely(created && err && a->h_path.dentry->d_inode)) {
17456 +               /* no delegation since it is just created */
17457 +               rerr = vfsub_unlink(h_dir, &a->h_path, /*delegated*/NULL,
17458 +                                   /*force*/0);
17459 +               if (rerr) {
17460 +                       AuIOErr("%pd revert failure(%d, %d)\n",
17461 +                               dentry, err, rerr);
17462 +                       err = -EIO;
17463 +               }
17464 +               au_dtime_revert(&a->dt);
17465 +       }
17466 +
17467 +       au_unpin(&a->pin);
17468 +       dput(wh_dentry);
17469 +
17470 +out_parent:
17471 +       di_write_unlock(parent);
17472 +out_unlock:
17473 +       if (unlikely(err)) {
17474 +               au_update_dbstart(dentry);
17475 +               d_drop(dentry);
17476 +       }
17477 +       aufs_read_unlock(dentry, AuLock_DW);
17478 +out_free:
17479 +       kfree(a);
17480 +out:
17481 +       return err;
17482 +}
17483 +
17484 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
17485 +              dev_t dev)
17486 +{
17487 +       struct simple_arg arg = {
17488 +               .type = Mknod,
17489 +               .u.m = {
17490 +                       .mode   = mode,
17491 +                       .dev    = dev
17492 +               }
17493 +       };
17494 +       return add_simple(dir, dentry, &arg);
17495 +}
17496 +
17497 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
17498 +{
17499 +       struct simple_arg arg = {
17500 +               .type = Symlink,
17501 +               .u.s.symname = symname
17502 +       };
17503 +       return add_simple(dir, dentry, &arg);
17504 +}
17505 +
17506 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
17507 +               bool want_excl)
17508 +{
17509 +       struct simple_arg arg = {
17510 +               .type = Creat,
17511 +               .u.c = {
17512 +                       .mode           = mode,
17513 +                       .want_excl      = want_excl
17514 +               }
17515 +       };
17516 +       return add_simple(dir, dentry, &arg);
17517 +}
17518 +
17519 +int aufs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
17520 +{
17521 +       int err;
17522 +       aufs_bindex_t bindex;
17523 +       struct super_block *sb;
17524 +       struct dentry *parent, *h_parent, *h_dentry;
17525 +       struct inode *h_dir, *inode;
17526 +       struct vfsmount *h_mnt;
17527 +       struct au_wr_dir_args wr_dir_args = {
17528 +               .force_btgt     = -1,
17529 +               .flags          = AuWrDir_TMPFILE
17530 +       };
17531 +
17532 +       /* copy-up may happen */
17533 +       mutex_lock(&dir->i_mutex);
17534 +
17535 +       sb = dir->i_sb;
17536 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
17537 +       if (unlikely(err))
17538 +               goto out;
17539 +
17540 +       err = au_di_init(dentry);
17541 +       if (unlikely(err))
17542 +               goto out_si;
17543 +
17544 +       err = -EBUSY;
17545 +       parent = d_find_any_alias(dir);
17546 +       AuDebugOn(!parent);
17547 +       di_write_lock_parent(parent);
17548 +       if (unlikely(parent->d_inode != dir))
17549 +               goto out_parent;
17550 +
17551 +       err = au_digen_test(parent, au_sigen(sb));
17552 +       if (unlikely(err))
17553 +               goto out_parent;
17554 +
17555 +       bindex = au_dbstart(parent);
17556 +       au_set_dbstart(dentry, bindex);
17557 +       au_set_dbend(dentry, bindex);
17558 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
17559 +       bindex = err;
17560 +       if (unlikely(err < 0))
17561 +               goto out_parent;
17562 +
17563 +       err = -EOPNOTSUPP;
17564 +       h_dir = au_h_iptr(dir, bindex);
17565 +       if (unlikely(!h_dir->i_op->tmpfile))
17566 +               goto out_parent;
17567 +
17568 +       h_mnt = au_sbr_mnt(sb, bindex);
17569 +       err = vfsub_mnt_want_write(h_mnt);
17570 +       if (unlikely(err))
17571 +               goto out_parent;
17572 +
17573 +       h_parent = au_h_dptr(parent, bindex);
17574 +       err = inode_permission(h_parent->d_inode, MAY_WRITE | MAY_EXEC);
17575 +       if (unlikely(err))
17576 +               goto out_mnt;
17577 +
17578 +       err = -ENOMEM;
17579 +       h_dentry = d_alloc(h_parent, &dentry->d_name);
17580 +       if (unlikely(!h_dentry))
17581 +               goto out_mnt;
17582 +
17583 +       err = h_dir->i_op->tmpfile(h_dir, h_dentry, mode);
17584 +       if (unlikely(err))
17585 +               goto out_dentry;
17586 +
17587 +       au_set_dbstart(dentry, bindex);
17588 +       au_set_dbend(dentry, bindex);
17589 +       au_set_h_dptr(dentry, bindex, dget(h_dentry));
17590 +       inode = au_new_inode(dentry, /*must_new*/1);
17591 +       if (IS_ERR(inode)) {
17592 +               err = PTR_ERR(inode);
17593 +               au_set_h_dptr(dentry, bindex, NULL);
17594 +               au_set_dbstart(dentry, -1);
17595 +               au_set_dbend(dentry, -1);
17596 +       } else {
17597 +               if (!inode->i_nlink)
17598 +                       set_nlink(inode, 1);
17599 +               d_tmpfile(dentry, inode);
17600 +               au_di(dentry)->di_tmpfile = 1;
17601 +
17602 +               /* update without i_mutex */
17603 +               if (au_ibstart(dir) == au_dbstart(dentry))
17604 +                       au_cpup_attr_timesizes(dir);
17605 +       }
17606 +
17607 +out_dentry:
17608 +       dput(h_dentry);
17609 +out_mnt:
17610 +       vfsub_mnt_drop_write(h_mnt);
17611 +out_parent:
17612 +       di_write_unlock(parent);
17613 +       dput(parent);
17614 +       di_write_unlock(dentry);
17615 +       if (!err)
17616 +#if 0
17617 +               /* verbose coding for lock class name */
17618 +               au_rw_class(&au_di(dentry)->di_rwsem,
17619 +                           au_lc_key + AuLcNonDir_DIINFO);
17620 +#else
17621 +               ;
17622 +#endif
17623 +       else {
17624 +               au_di_fin(dentry);
17625 +               dentry->d_fsdata = NULL;
17626 +       }
17627 +out_si:
17628 +       si_read_unlock(sb);
17629 +out:
17630 +       mutex_unlock(&dir->i_mutex);
17631 +       return err;
17632 +}
17633 +
17634 +/* ---------------------------------------------------------------------- */
17635 +
17636 +struct au_link_args {
17637 +       aufs_bindex_t bdst, bsrc;
17638 +       struct au_pin pin;
17639 +       struct path h_path;
17640 +       struct dentry *src_parent, *parent;
17641 +};
17642 +
17643 +static int au_cpup_before_link(struct dentry *src_dentry,
17644 +                              struct au_link_args *a)
17645 +{
17646 +       int err;
17647 +       struct dentry *h_src_dentry;
17648 +       struct au_cp_generic cpg = {
17649 +               .dentry = src_dentry,
17650 +               .bdst   = a->bdst,
17651 +               .bsrc   = a->bsrc,
17652 +               .len    = -1,
17653 +               .pin    = &a->pin,
17654 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN /* | AuCpup_KEEPLINO */
17655 +       };
17656 +
17657 +       di_read_lock_parent(a->src_parent, AuLock_IR);
17658 +       err = au_test_and_cpup_dirs(src_dentry, a->bdst);
17659 +       if (unlikely(err))
17660 +               goto out;
17661 +
17662 +       h_src_dentry = au_h_dptr(src_dentry, a->bsrc);
17663 +       err = au_pin(&a->pin, src_dentry, a->bdst,
17664 +                    au_opt_udba(src_dentry->d_sb),
17665 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17666 +       if (unlikely(err))
17667 +               goto out;
17668 +
17669 +       err = au_sio_cpup_simple(&cpg);
17670 +       au_unpin(&a->pin);
17671 +
17672 +out:
17673 +       di_read_unlock(a->src_parent, AuLock_IR);
17674 +       return err;
17675 +}
17676 +
17677 +static int au_cpup_or_link(struct dentry *src_dentry, struct dentry *dentry,
17678 +                          struct au_link_args *a)
17679 +{
17680 +       int err;
17681 +       unsigned char plink;
17682 +       aufs_bindex_t bend;
17683 +       struct dentry *h_src_dentry;
17684 +       struct inode *h_inode, *inode, *delegated;
17685 +       struct super_block *sb;
17686 +       struct file *h_file;
17687 +
17688 +       plink = 0;
17689 +       h_inode = NULL;
17690 +       sb = src_dentry->d_sb;
17691 +       inode = src_dentry->d_inode;
17692 +       if (au_ibstart(inode) <= a->bdst)
17693 +               h_inode = au_h_iptr(inode, a->bdst);
17694 +       if (!h_inode || !h_inode->i_nlink) {
17695 +               /* copyup src_dentry as the name of dentry. */
17696 +               bend = au_dbend(dentry);
17697 +               if (bend < a->bsrc)
17698 +                       au_set_dbend(dentry, a->bsrc);
17699 +               au_set_h_dptr(dentry, a->bsrc,
17700 +                             dget(au_h_dptr(src_dentry, a->bsrc)));
17701 +               dget(a->h_path.dentry);
17702 +               au_set_h_dptr(dentry, a->bdst, NULL);
17703 +               AuDbg("temporary d_inode...\n");
17704 +               spin_lock(&dentry->d_lock);
17705 +               dentry->d_inode = src_dentry->d_inode; /* tmp */
17706 +               spin_unlock(&dentry->d_lock);
17707 +               h_file = au_h_open_pre(dentry, a->bsrc, /*force_wr*/0);
17708 +               if (IS_ERR(h_file))
17709 +                       err = PTR_ERR(h_file);
17710 +               else {
17711 +                       struct au_cp_generic cpg = {
17712 +                               .dentry = dentry,
17713 +                               .bdst   = a->bdst,
17714 +                               .bsrc   = -1,
17715 +                               .len    = -1,
17716 +                               .pin    = &a->pin,
17717 +                               .flags  = AuCpup_KEEPLINO
17718 +                       };
17719 +                       err = au_sio_cpup_simple(&cpg);
17720 +                       au_h_open_post(dentry, a->bsrc, h_file);
17721 +                       if (!err) {
17722 +                               dput(a->h_path.dentry);
17723 +                               a->h_path.dentry = au_h_dptr(dentry, a->bdst);
17724 +                       } else
17725 +                               au_set_h_dptr(dentry, a->bdst,
17726 +                                             a->h_path.dentry);
17727 +               }
17728 +               spin_lock(&dentry->d_lock);
17729 +               dentry->d_inode = NULL; /* restore */
17730 +               spin_unlock(&dentry->d_lock);
17731 +               AuDbg("temporary d_inode...done\n");
17732 +               au_set_h_dptr(dentry, a->bsrc, NULL);
17733 +               au_set_dbend(dentry, bend);
17734 +       } else {
17735 +               /* the inode of src_dentry already exists on a.bdst branch */
17736 +               h_src_dentry = d_find_alias(h_inode);
17737 +               if (!h_src_dentry && au_plink_test(inode)) {
17738 +                       plink = 1;
17739 +                       h_src_dentry = au_plink_lkup(inode, a->bdst);
17740 +                       err = PTR_ERR(h_src_dentry);
17741 +                       if (IS_ERR(h_src_dentry))
17742 +                               goto out;
17743 +
17744 +                       if (unlikely(!h_src_dentry->d_inode)) {
17745 +                               dput(h_src_dentry);
17746 +                               h_src_dentry = NULL;
17747 +                       }
17748 +
17749 +               }
17750 +               if (h_src_dentry) {
17751 +                       delegated = NULL;
17752 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
17753 +                                        &a->h_path, &delegated);
17754 +                       if (unlikely(err == -EWOULDBLOCK)) {
17755 +                               pr_warn("cannot retry for NFSv4 delegation"
17756 +                                       " for an internal link\n");
17757 +                               iput(delegated);
17758 +                       }
17759 +                       dput(h_src_dentry);
17760 +               } else {
17761 +                       AuIOErr("no dentry found for hi%lu on b%d\n",
17762 +                               h_inode->i_ino, a->bdst);
17763 +                       err = -EIO;
17764 +               }
17765 +       }
17766 +
17767 +       if (!err && !plink)
17768 +               au_plink_append(inode, a->bdst, a->h_path.dentry);
17769 +
17770 +out:
17771 +       AuTraceErr(err);
17772 +       return err;
17773 +}
17774 +
17775 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
17776 +             struct dentry *dentry)
17777 +{
17778 +       int err, rerr;
17779 +       struct au_dtime dt;
17780 +       struct au_link_args *a;
17781 +       struct dentry *wh_dentry, *h_src_dentry;
17782 +       struct inode *inode, *delegated;
17783 +       struct super_block *sb;
17784 +       struct au_wr_dir_args wr_dir_args = {
17785 +               /* .force_btgt  = -1, */
17786 +               .flags          = AuWrDir_ADD_ENTRY
17787 +       };
17788 +
17789 +       IMustLock(dir);
17790 +       inode = src_dentry->d_inode;
17791 +       IMustLock(inode);
17792 +
17793 +       err = -ENOMEM;
17794 +       a = kzalloc(sizeof(*a), GFP_NOFS);
17795 +       if (unlikely(!a))
17796 +               goto out;
17797 +
17798 +       a->parent = dentry->d_parent; /* dir inode is locked */
17799 +       err = aufs_read_and_write_lock2(dentry, src_dentry,
17800 +                                       AuLock_NOPLM | AuLock_GEN);
17801 +       if (unlikely(err))
17802 +               goto out_kfree;
17803 +       err = au_d_linkable(src_dentry);
17804 +       if (unlikely(err))
17805 +               goto out_unlock;
17806 +       err = au_d_may_add(dentry);
17807 +       if (unlikely(err))
17808 +               goto out_unlock;
17809 +
17810 +       a->src_parent = dget_parent(src_dentry);
17811 +       wr_dir_args.force_btgt = au_ibstart(inode);
17812 +
17813 +       di_write_lock_parent(a->parent);
17814 +       wr_dir_args.force_btgt = au_wbr(dentry, wr_dir_args.force_btgt);
17815 +       wh_dentry = lock_hdir_lkup_wh(dentry, &dt, src_dentry, &a->pin,
17816 +                                     &wr_dir_args);
17817 +       err = PTR_ERR(wh_dentry);
17818 +       if (IS_ERR(wh_dentry))
17819 +               goto out_parent;
17820 +
17821 +       err = 0;
17822 +       sb = dentry->d_sb;
17823 +       a->bdst = au_dbstart(dentry);
17824 +       a->h_path.dentry = au_h_dptr(dentry, a->bdst);
17825 +       a->h_path.mnt = au_sbr_mnt(sb, a->bdst);
17826 +       a->bsrc = au_ibstart(inode);
17827 +       h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
17828 +       if (!h_src_dentry && au_di(src_dentry)->di_tmpfile)
17829 +               h_src_dentry = dget(au_hi_wh(inode, a->bsrc));
17830 +       if (!h_src_dentry) {
17831 +               a->bsrc = au_dbstart(src_dentry);
17832 +               h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
17833 +               AuDebugOn(!h_src_dentry);
17834 +       } else if (IS_ERR(h_src_dentry)) {
17835 +               err = PTR_ERR(h_src_dentry);
17836 +               goto out_parent;
17837 +       }
17838 +
17839 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
17840 +               if (a->bdst < a->bsrc
17841 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */)
17842 +                       err = au_cpup_or_link(src_dentry, dentry, a);
17843 +               else {
17844 +                       delegated = NULL;
17845 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
17846 +                                        &a->h_path, &delegated);
17847 +                       if (unlikely(err == -EWOULDBLOCK)) {
17848 +                               pr_warn("cannot retry for NFSv4 delegation"
17849 +                                       " for an internal link\n");
17850 +                               iput(delegated);
17851 +                       }
17852 +               }
17853 +               dput(h_src_dentry);
17854 +       } else {
17855 +               /*
17856 +                * copyup src_dentry to the branch we process,
17857 +                * and then link(2) to it.
17858 +                */
17859 +               dput(h_src_dentry);
17860 +               if (a->bdst < a->bsrc
17861 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */) {
17862 +                       au_unpin(&a->pin);
17863 +                       di_write_unlock(a->parent);
17864 +                       err = au_cpup_before_link(src_dentry, a);
17865 +                       di_write_lock_parent(a->parent);
17866 +                       if (!err)
17867 +                               err = au_pin(&a->pin, dentry, a->bdst,
17868 +                                            au_opt_udba(sb),
17869 +                                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
17870 +                       if (unlikely(err))
17871 +                               goto out_wh;
17872 +               }
17873 +               if (!err) {
17874 +                       h_src_dentry = au_h_dptr(src_dentry, a->bdst);
17875 +                       err = -ENOENT;
17876 +                       if (h_src_dentry && h_src_dentry->d_inode) {
17877 +                               delegated = NULL;
17878 +                               err = vfsub_link(h_src_dentry,
17879 +                                                au_pinned_h_dir(&a->pin),
17880 +                                                &a->h_path, &delegated);
17881 +                               if (unlikely(err == -EWOULDBLOCK)) {
17882 +                                       pr_warn("cannot retry"
17883 +                                               " for NFSv4 delegation"
17884 +                                               " for an internal link\n");
17885 +                                       iput(delegated);
17886 +                               }
17887 +                       }
17888 +               }
17889 +       }
17890 +       if (unlikely(err))
17891 +               goto out_unpin;
17892 +
17893 +       if (wh_dentry) {
17894 +               a->h_path.dentry = wh_dentry;
17895 +               err = au_wh_unlink_dentry(au_pinned_h_dir(&a->pin), &a->h_path,
17896 +                                         dentry);
17897 +               if (unlikely(err))
17898 +                       goto out_revert;
17899 +       }
17900 +
17901 +       dir->i_version++;
17902 +       if (au_ibstart(dir) == au_dbstart(dentry))
17903 +               au_cpup_attr_timesizes(dir);
17904 +       inc_nlink(inode);
17905 +       inode->i_ctime = dir->i_ctime;
17906 +       d_instantiate(dentry, au_igrab(inode));
17907 +       if (d_unhashed(a->h_path.dentry))
17908 +               /* some filesystem calls d_drop() */
17909 +               d_drop(dentry);
17910 +       /* some filesystems consume an inode even hardlink */
17911 +       au_fhsm_wrote(sb, a->bdst, /*force*/0);
17912 +       goto out_unpin; /* success */
17913 +
17914 +out_revert:
17915 +       /* no delegation since it is just created */
17916 +       rerr = vfsub_unlink(au_pinned_h_dir(&a->pin), &a->h_path,
17917 +                           /*delegated*/NULL, /*force*/0);
17918 +       if (unlikely(rerr)) {
17919 +               AuIOErr("%pd reverting failed(%d, %d)\n", dentry, err, rerr);
17920 +               err = -EIO;
17921 +       }
17922 +       au_dtime_revert(&dt);
17923 +out_unpin:
17924 +       au_unpin(&a->pin);
17925 +out_wh:
17926 +       dput(wh_dentry);
17927 +out_parent:
17928 +       di_write_unlock(a->parent);
17929 +       dput(a->src_parent);
17930 +out_unlock:
17931 +       if (unlikely(err)) {
17932 +               au_update_dbstart(dentry);
17933 +               d_drop(dentry);
17934 +       }
17935 +       aufs_read_and_write_unlock2(dentry, src_dentry);
17936 +out_kfree:
17937 +       kfree(a);
17938 +out:
17939 +       AuTraceErr(err);
17940 +       return err;
17941 +}
17942 +
17943 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
17944 +{
17945 +       int err, rerr;
17946 +       aufs_bindex_t bindex;
17947 +       unsigned char diropq;
17948 +       struct path h_path;
17949 +       struct dentry *wh_dentry, *parent, *opq_dentry;
17950 +       struct mutex *h_mtx;
17951 +       struct super_block *sb;
17952 +       struct {
17953 +               struct au_pin pin;
17954 +               struct au_dtime dt;
17955 +       } *a; /* reduce the stack usage */
17956 +       struct au_wr_dir_args wr_dir_args = {
17957 +               .force_btgt     = -1,
17958 +               .flags          = AuWrDir_ADD_ENTRY | AuWrDir_ISDIR
17959 +       };
17960 +
17961 +       IMustLock(dir);
17962 +
17963 +       err = -ENOMEM;
17964 +       a = kmalloc(sizeof(*a), GFP_NOFS);
17965 +       if (unlikely(!a))
17966 +               goto out;
17967 +
17968 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
17969 +       if (unlikely(err))
17970 +               goto out_free;
17971 +       err = au_d_may_add(dentry);
17972 +       if (unlikely(err))
17973 +               goto out_unlock;
17974 +
17975 +       parent = dentry->d_parent; /* dir inode is locked */
17976 +       di_write_lock_parent(parent);
17977 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
17978 +                                     &a->pin, &wr_dir_args);
17979 +       err = PTR_ERR(wh_dentry);
17980 +       if (IS_ERR(wh_dentry))
17981 +               goto out_parent;
17982 +
17983 +       sb = dentry->d_sb;
17984 +       bindex = au_dbstart(dentry);
17985 +       h_path.dentry = au_h_dptr(dentry, bindex);
17986 +       h_path.mnt = au_sbr_mnt(sb, bindex);
17987 +       err = vfsub_mkdir(au_pinned_h_dir(&a->pin), &h_path, mode);
17988 +       if (unlikely(err))
17989 +               goto out_unpin;
17990 +
17991 +       /* make the dir opaque */
17992 +       diropq = 0;
17993 +       h_mtx = &h_path.dentry->d_inode->i_mutex;
17994 +       if (wh_dentry
17995 +           || au_opt_test(au_mntflags(sb), ALWAYS_DIROPQ)) {
17996 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
17997 +               opq_dentry = au_diropq_create(dentry, bindex);
17998 +               mutex_unlock(h_mtx);
17999 +               err = PTR_ERR(opq_dentry);
18000 +               if (IS_ERR(opq_dentry))
18001 +                       goto out_dir;
18002 +               dput(opq_dentry);
18003 +               diropq = 1;
18004 +       }
18005 +
18006 +       err = epilog(dir, bindex, wh_dentry, dentry);
18007 +       if (!err) {
18008 +               inc_nlink(dir);
18009 +               goto out_unpin; /* success */
18010 +       }
18011 +
18012 +       /* revert */
18013 +       if (diropq) {
18014 +               AuLabel(revert opq);
18015 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
18016 +               rerr = au_diropq_remove(dentry, bindex);
18017 +               mutex_unlock(h_mtx);
18018 +               if (rerr) {
18019 +                       AuIOErr("%pd reverting diropq failed(%d, %d)\n",
18020 +                               dentry, err, rerr);
18021 +                       err = -EIO;
18022 +               }
18023 +       }
18024 +
18025 +out_dir:
18026 +       AuLabel(revert dir);
18027 +       rerr = vfsub_rmdir(au_pinned_h_dir(&a->pin), &h_path);
18028 +       if (rerr) {
18029 +               AuIOErr("%pd reverting dir failed(%d, %d)\n",
18030 +                       dentry, err, rerr);
18031 +               err = -EIO;
18032 +       }
18033 +       au_dtime_revert(&a->dt);
18034 +out_unpin:
18035 +       au_unpin(&a->pin);
18036 +       dput(wh_dentry);
18037 +out_parent:
18038 +       di_write_unlock(parent);
18039 +out_unlock:
18040 +       if (unlikely(err)) {
18041 +               au_update_dbstart(dentry);
18042 +               d_drop(dentry);
18043 +       }
18044 +       aufs_read_unlock(dentry, AuLock_DW);
18045 +out_free:
18046 +       kfree(a);
18047 +out:
18048 +       return err;
18049 +}
18050 diff -urN /usr/share/empty/fs/aufs/i_op.c linux/fs/aufs/i_op.c
18051 --- /usr/share/empty/fs/aufs/i_op.c     1970-01-01 01:00:00.000000000 +0100
18052 +++ linux/fs/aufs/i_op.c        2015-03-27 21:56:35.463461668 +0100
18053 @@ -0,0 +1,1299 @@
18054 +/*
18055 + * Copyright (C) 2005-2015 Junjiro R. Okajima
18056 + *
18057 + * This program, aufs is free software; you can redistribute it and/or modify
18058 + * it under the terms of the GNU General Public License as published by
18059 + * the Free Software Foundation; either version 2 of the License, or
18060 + * (at your option) any later version.
18061 + *
18062 + * This program is distributed in the hope that it will be useful,
18063 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18064 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18065 + * GNU General Public License for more details.
18066 + *
18067 + * You should have received a copy of the GNU General Public License
18068 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18069 + */
18070 +
18071 +/*
18072 + * inode operations (except add/del/rename)
18073 + */
18074 +
18075 +#include <linux/device_cgroup.h>
18076 +#include <linux/fs_stack.h>
18077 +#include <linux/mm.h>
18078 +#include <linux/namei.h>
18079 +#include <linux/security.h>
18080 +#include "aufs.h"
18081 +
18082 +static int h_permission(struct inode *h_inode, int mask,
18083 +                       struct vfsmount *h_mnt, int brperm)
18084 +{
18085 +       int err;
18086 +       const unsigned char write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18087 +
18088 +       err = -EACCES;
18089 +       if ((write_mask && IS_IMMUTABLE(h_inode))
18090 +           || ((mask & MAY_EXEC)
18091 +               && S_ISREG(h_inode->i_mode)
18092 +               && ((h_mnt->mnt_flags & MNT_NOEXEC)
18093 +                   || !(h_inode->i_mode & S_IXUGO))))
18094 +               goto out;
18095 +
18096 +       /*
18097 +        * - skip the lower fs test in the case of write to ro branch.
18098 +        * - nfs dir permission write check is optimized, but a policy for
18099 +        *   link/rename requires a real check.
18100 +        */
18101 +       if ((write_mask && !au_br_writable(brperm))
18102 +           || (au_test_nfs(h_inode->i_sb) && S_ISDIR(h_inode->i_mode)
18103 +               && write_mask && !(mask & MAY_READ))
18104 +           || !h_inode->i_op->permission) {
18105 +               /* AuLabel(generic_permission); */
18106 +               err = generic_permission(h_inode, mask);
18107 +       } else {
18108 +               /* AuLabel(h_inode->permission); */
18109 +               err = h_inode->i_op->permission(h_inode, mask);
18110 +               AuTraceErr(err);
18111 +       }
18112 +
18113 +       if (!err)
18114 +               err = devcgroup_inode_permission(h_inode, mask);
18115 +       if (!err)
18116 +               err = security_inode_permission(h_inode, mask);
18117 +
18118 +#if 0
18119 +       if (!err) {
18120 +               /* todo: do we need to call ima_path_check()? */
18121 +               struct path h_path = {
18122 +                       .dentry =
18123 +                       .mnt    = h_mnt
18124 +               };
18125 +               err = ima_path_check(&h_path,
18126 +                                    mask & (MAY_READ | MAY_WRITE | MAY_EXEC),
18127 +                                    IMA_COUNT_LEAVE);
18128 +       }
18129 +#endif
18130 +
18131 +out:
18132 +       return err;
18133 +}
18134 +
18135 +static int aufs_permission(struct inode *inode, int mask)
18136 +{
18137 +       int err;
18138 +       aufs_bindex_t bindex, bend;
18139 +       const unsigned char isdir = !!S_ISDIR(inode->i_mode),
18140 +               write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
18141 +       struct inode *h_inode;
18142 +       struct super_block *sb;
18143 +       struct au_branch *br;
18144 +
18145 +       /* todo: support rcu-walk? */
18146 +       if (mask & MAY_NOT_BLOCK)
18147 +               return -ECHILD;
18148 +
18149 +       sb = inode->i_sb;
18150 +       si_read_lock(sb, AuLock_FLUSH);
18151 +       ii_read_lock_child(inode);
18152 +#if 0
18153 +       err = au_iigen_test(inode, au_sigen(sb));
18154 +       if (unlikely(err))
18155 +               goto out;
18156 +#endif
18157 +
18158 +       if (!isdir
18159 +           || write_mask
18160 +           || au_opt_test(au_mntflags(sb), DIRPERM1)) {
18161 +               err = au_busy_or_stale();
18162 +               h_inode = au_h_iptr(inode, au_ibstart(inode));
18163 +               if (unlikely(!h_inode
18164 +                            || (h_inode->i_mode & S_IFMT)
18165 +                            != (inode->i_mode & S_IFMT)))
18166 +                       goto out;
18167 +
18168 +               err = 0;
18169 +               bindex = au_ibstart(inode);
18170 +               br = au_sbr(sb, bindex);
18171 +               err = h_permission(h_inode, mask, au_br_mnt(br), br->br_perm);
18172 +               if (write_mask
18173 +                   && !err
18174 +                   && !special_file(h_inode->i_mode)) {
18175 +                       /* test whether the upper writable branch exists */
18176 +                       err = -EROFS;
18177 +                       for (; bindex >= 0; bindex--)
18178 +                               if (!au_br_rdonly(au_sbr(sb, bindex))) {
18179 +                                       err = 0;
18180 +                                       break;
18181 +                               }
18182 +               }
18183 +               goto out;
18184 +       }
18185 +
18186 +       /* non-write to dir */
18187 +       err = 0;
18188 +       bend = au_ibend(inode);
18189 +       for (bindex = au_ibstart(inode); !err && bindex <= bend; bindex++) {
18190 +               h_inode = au_h_iptr(inode, bindex);
18191 +               if (h_inode) {
18192 +                       err = au_busy_or_stale();
18193 +                       if (unlikely(!S_ISDIR(h_inode->i_mode)))
18194 +                               break;
18195 +
18196 +                       br = au_sbr(sb, bindex);
18197 +                       err = h_permission(h_inode, mask, au_br_mnt(br),
18198 +                                          br->br_perm);
18199 +               }
18200 +       }
18201 +
18202 +out:
18203 +       ii_read_unlock(inode);
18204 +       si_read_unlock(sb);
18205 +       return err;
18206 +}
18207 +
18208 +/* ---------------------------------------------------------------------- */
18209 +
18210 +static struct dentry *aufs_lookup(struct inode *dir, struct dentry *dentry,
18211 +                                 unsigned int flags)
18212 +{
18213 +       struct dentry *ret, *parent;
18214 +       struct inode *inode;
18215 +       struct super_block *sb;
18216 +       int err, npositive;
18217 +
18218 +       IMustLock(dir);
18219 +
18220 +       /* todo: support rcu-walk? */
18221 +       ret = ERR_PTR(-ECHILD);
18222 +       if (flags & LOOKUP_RCU)
18223 +               goto out;
18224 +
18225 +       ret = ERR_PTR(-ENAMETOOLONG);
18226 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
18227 +               goto out;
18228 +
18229 +       sb = dir->i_sb;
18230 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18231 +       ret = ERR_PTR(err);
18232 +       if (unlikely(err))
18233 +               goto out;
18234 +
18235 +       err = au_di_init(dentry);
18236 +       ret = ERR_PTR(err);
18237 +       if (unlikely(err))
18238 +               goto out_si;
18239 +
18240 +       inode = NULL;
18241 +       npositive = 0; /* suppress a warning */
18242 +       parent = dentry->d_parent; /* dir inode is locked */
18243 +       di_read_lock_parent(parent, AuLock_IR);
18244 +       err = au_alive_dir(parent);
18245 +       if (!err)
18246 +               err = au_digen_test(parent, au_sigen(sb));
18247 +       if (!err) {
18248 +               npositive = au_lkup_dentry(dentry, au_dbstart(parent),
18249 +                                          /*type*/0);
18250 +               err = npositive;
18251 +       }
18252 +       di_read_unlock(parent, AuLock_IR);
18253 +       ret = ERR_PTR(err);
18254 +       if (unlikely(err < 0))
18255 +               goto out_unlock;
18256 +
18257 +       if (npositive) {
18258 +               inode = au_new_inode(dentry, /*must_new*/0);
18259 +               if (IS_ERR(inode)) {
18260 +                       ret = (void *)inode;
18261 +                       inode = NULL;
18262 +                       goto out_unlock;
18263 +               }
18264 +       }
18265 +
18266 +       if (inode)
18267 +               atomic_inc(&inode->i_count);
18268 +       ret = d_splice_alias(inode, dentry);
18269 +#if 0
18270 +       if (unlikely(d_need_lookup(dentry))) {
18271 +               spin_lock(&dentry->d_lock);
18272 +               dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
18273 +               spin_unlock(&dentry->d_lock);
18274 +       } else
18275 +#endif
18276 +       if (inode) {
18277 +               if (!IS_ERR(ret)) {
18278 +                       iput(inode);
18279 +                       if (ret && ret != dentry)
18280 +                               ii_write_unlock(inode);
18281 +               } else {
18282 +                       ii_write_unlock(inode);
18283 +                       iput(inode);
18284 +                       inode = NULL;
18285 +               }
18286 +       }
18287 +
18288 +out_unlock:
18289 +       di_write_unlock(dentry);
18290 +       if (inode) {
18291 +               /* verbose coding for lock class name */
18292 +               if (unlikely(S_ISLNK(inode->i_mode)))
18293 +                       au_rw_class(&au_di(dentry)->di_rwsem,
18294 +                                   au_lc_key + AuLcSymlink_DIINFO);
18295 +               else if (unlikely(S_ISDIR(inode->i_mode)))
18296 +                       au_rw_class(&au_di(dentry)->di_rwsem,
18297 +                                   au_lc_key + AuLcDir_DIINFO);
18298 +               else /* likely */
18299 +                       au_rw_class(&au_di(dentry)->di_rwsem,
18300 +                                   au_lc_key + AuLcNonDir_DIINFO);
18301 +       }
18302 +out_si:
18303 +       si_read_unlock(sb);
18304 +out:
18305 +       return ret;
18306 +}
18307 +
18308 +/* ---------------------------------------------------------------------- */
18309 +
18310 +static int au_wr_dir_cpup(struct dentry *dentry, struct dentry *parent,
18311 +                         const unsigned char add_entry, aufs_bindex_t bcpup,
18312 +                         aufs_bindex_t bstart)
18313 +{
18314 +       int err;
18315 +       struct dentry *h_parent;
18316 +       struct inode *h_dir;
18317 +
18318 +       if (add_entry)
18319 +               IMustLock(parent->d_inode);
18320 +       else
18321 +               di_write_lock_parent(parent);
18322 +
18323 +       err = 0;
18324 +       if (!au_h_dptr(parent, bcpup)) {
18325 +               if (bstart > bcpup)
18326 +                       err = au_cpup_dirs(dentry, bcpup);
18327 +               else if (bstart < bcpup)
18328 +                       err = au_cpdown_dirs(dentry, bcpup);
18329 +               else
18330 +                       BUG();
18331 +       }
18332 +       if (!err && add_entry && !au_ftest_wrdir(add_entry, TMPFILE)) {
18333 +               h_parent = au_h_dptr(parent, bcpup);
18334 +               h_dir = h_parent->d_inode;
18335 +               mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
18336 +               err = au_lkup_neg(dentry, bcpup,
18337 +                                 au_ftest_wrdir(add_entry, TMP_WHENTRY));
18338 +               /* todo: no unlock here */
18339 +               mutex_unlock(&h_dir->i_mutex);
18340 +
18341 +               AuDbg("bcpup %d\n", bcpup);
18342 +               if (!err) {
18343 +                       if (!dentry->d_inode)
18344 +                               au_set_h_dptr(dentry, bstart, NULL);
18345 +                       au_update_dbrange(dentry, /*do_put_zero*/0);
18346 +               }
18347 +       }
18348 +
18349 +       if (!add_entry)
18350 +               di_write_unlock(parent);
18351 +       if (!err)
18352 +               err = bcpup; /* success */
18353 +
18354 +       AuTraceErr(err);
18355 +       return err;
18356 +}
18357 +
18358 +/*
18359 + * decide the branch and the parent dir where we will create a new entry.
18360 + * returns new bindex or an error.
18361 + * copyup the parent dir if needed.
18362 + */
18363 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
18364 +             struct au_wr_dir_args *args)
18365 +{
18366 +       int err;
18367 +       unsigned int flags;
18368 +       aufs_bindex_t bcpup, bstart, src_bstart;
18369 +       const unsigned char add_entry
18370 +               = au_ftest_wrdir(args->flags, ADD_ENTRY)
18371 +               | au_ftest_wrdir(args->flags, TMP_WHENTRY)
18372 +               | au_ftest_wrdir(args->flags, TMPFILE);
18373 +       struct super_block *sb;
18374 +       struct dentry *parent;
18375 +       struct au_sbinfo *sbinfo;
18376 +
18377 +       sb = dentry->d_sb;
18378 +       sbinfo = au_sbi(sb);
18379 +       parent = dget_parent(dentry);
18380 +       bstart = au_dbstart(dentry);
18381 +       bcpup = bstart;
18382 +       if (args->force_btgt < 0) {
18383 +               if (src_dentry) {
18384 +                       src_bstart = au_dbstart(src_dentry);
18385 +                       if (src_bstart < bstart)
18386 +                               bcpup = src_bstart;
18387 +               } else if (add_entry) {
18388 +                       flags = 0;
18389 +                       if (au_ftest_wrdir(args->flags, ISDIR))
18390 +                               au_fset_wbr(flags, DIR);
18391 +                       err = AuWbrCreate(sbinfo, dentry, flags);
18392 +                       bcpup = err;
18393 +               }
18394 +
18395 +               if (bcpup < 0 || au_test_ro(sb, bcpup, dentry->d_inode)) {
18396 +                       if (add_entry)
18397 +                               err = AuWbrCopyup(sbinfo, dentry);
18398 +                       else {
18399 +                               if (!IS_ROOT(dentry)) {
18400 +                                       di_read_lock_parent(parent, !AuLock_IR);
18401 +                                       err = AuWbrCopyup(sbinfo, dentry);
18402 +                                       di_read_unlock(parent, !AuLock_IR);
18403 +                               } else
18404 +                                       err = AuWbrCopyup(sbinfo, dentry);
18405 +                       }
18406 +                       bcpup = err;
18407 +                       if (unlikely(err < 0))
18408 +                               goto out;
18409 +               }
18410 +       } else {
18411 +               bcpup = args->force_btgt;
18412 +               AuDebugOn(au_test_ro(sb, bcpup, dentry->d_inode));
18413 +       }
18414 +
18415 +       AuDbg("bstart %d, bcpup %d\n", bstart, bcpup);
18416 +       err = bcpup;
18417 +       if (bcpup == bstart)
18418 +               goto out; /* success */
18419 +
18420 +       /* copyup the new parent into the branch we process */
18421 +       err = au_wr_dir_cpup(dentry, parent, add_entry, bcpup, bstart);
18422 +       if (err >= 0) {
18423 +               if (!dentry->d_inode) {
18424 +                       au_set_h_dptr(dentry, bstart, NULL);
18425 +                       au_set_dbstart(dentry, bcpup);
18426 +                       au_set_dbend(dentry, bcpup);
18427 +               }
18428 +               AuDebugOn(add_entry
18429 +                         && !au_ftest_wrdir(args->flags, TMPFILE)
18430 +                         && !au_h_dptr(dentry, bcpup));
18431 +       }
18432 +
18433 +out:
18434 +       dput(parent);
18435 +       return err;
18436 +}
18437 +
18438 +/* ---------------------------------------------------------------------- */
18439 +
18440 +void au_pin_hdir_unlock(struct au_pin *p)
18441 +{
18442 +       if (p->hdir)
18443 +               au_hn_imtx_unlock(p->hdir);
18444 +}
18445 +
18446 +int au_pin_hdir_lock(struct au_pin *p)
18447 +{
18448 +       int err;
18449 +
18450 +       err = 0;
18451 +       if (!p->hdir)
18452 +               goto out;
18453 +
18454 +       /* even if an error happens later, keep this lock */
18455 +       au_hn_imtx_lock_nested(p->hdir, p->lsc_hi);
18456 +
18457 +       err = -EBUSY;
18458 +       if (unlikely(p->hdir->hi_inode != p->h_parent->d_inode))
18459 +               goto out;
18460 +
18461 +       err = 0;
18462 +       if (p->h_dentry)
18463 +               err = au_h_verify(p->h_dentry, p->udba, p->hdir->hi_inode,
18464 +                                 p->h_parent, p->br);
18465 +
18466 +out:
18467 +       return err;
18468 +}
18469 +
18470 +int au_pin_hdir_relock(struct au_pin *p)
18471 +{
18472 +       int err, i;
18473 +       struct inode *h_i;
18474 +       struct dentry *h_d[] = {
18475 +               p->h_dentry,
18476 +               p->h_parent
18477 +       };
18478 +
18479 +       err = au_pin_hdir_lock(p);
18480 +       if (unlikely(err))
18481 +               goto out;
18482 +
18483 +       for (i = 0; !err && i < sizeof(h_d)/sizeof(*h_d); i++) {
18484 +               if (!h_d[i])
18485 +                       continue;
18486 +               h_i = h_d[i]->d_inode;
18487 +               if (h_i)
18488 +                       err = !h_i->i_nlink;
18489 +       }
18490 +
18491 +out:
18492 +       return err;
18493 +}
18494 +
18495 +void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task)
18496 +{
18497 +#if defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_SMP)
18498 +       p->hdir->hi_inode->i_mutex.owner = task;
18499 +#endif
18500 +}
18501 +
18502 +void au_pin_hdir_acquire_nest(struct au_pin *p)
18503 +{
18504 +       if (p->hdir) {
18505 +               mutex_acquire_nest(&p->hdir->hi_inode->i_mutex.dep_map,
18506 +                                  p->lsc_hi, 0, NULL, _RET_IP_);
18507 +               au_pin_hdir_set_owner(p, current);
18508 +       }
18509 +}
18510 +
18511 +void au_pin_hdir_release(struct au_pin *p)
18512 +{
18513 +       if (p->hdir) {
18514 +               au_pin_hdir_set_owner(p, p->task);
18515 +               mutex_release(&p->hdir->hi_inode->i_mutex.dep_map, 1, _RET_IP_);
18516 +       }
18517 +}
18518 +
18519 +struct dentry *au_pinned_h_parent(struct au_pin *pin)
18520 +{
18521 +       if (pin && pin->parent)
18522 +               return au_h_dptr(pin->parent, pin->bindex);
18523 +       return NULL;
18524 +}
18525 +
18526 +void au_unpin(struct au_pin *p)
18527 +{
18528 +       if (p->hdir)
18529 +               au_pin_hdir_unlock(p);
18530 +       if (p->h_mnt && au_ftest_pin(p->flags, MNT_WRITE))
18531 +               vfsub_mnt_drop_write(p->h_mnt);
18532 +       if (!p->hdir)
18533 +               return;
18534 +
18535 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
18536 +               di_read_unlock(p->parent, AuLock_IR);
18537 +       iput(p->hdir->hi_inode);
18538 +       dput(p->parent);
18539 +       p->parent = NULL;
18540 +       p->hdir = NULL;
18541 +       p->h_mnt = NULL;
18542 +       /* do not clear p->task */
18543 +}
18544 +
18545 +int au_do_pin(struct au_pin *p)
18546 +{
18547 +       int err;
18548 +       struct super_block *sb;
18549 +       struct inode *h_dir;
18550 +
18551 +       err = 0;
18552 +       sb = p->dentry->d_sb;
18553 +       p->br = au_sbr(sb, p->bindex);
18554 +       if (IS_ROOT(p->dentry)) {
18555 +               if (au_ftest_pin(p->flags, MNT_WRITE)) {
18556 +                       p->h_mnt = au_br_mnt(p->br);
18557 +                       err = vfsub_mnt_want_write(p->h_mnt);
18558 +                       if (unlikely(err)) {
18559 +                               au_fclr_pin(p->flags, MNT_WRITE);
18560 +                               goto out_err;
18561 +                       }
18562 +               }
18563 +               goto out;
18564 +       }
18565 +
18566 +       p->h_dentry = NULL;
18567 +       if (p->bindex <= au_dbend(p->dentry))
18568 +               p->h_dentry = au_h_dptr(p->dentry, p->bindex);
18569 +
18570 +       p->parent = dget_parent(p->dentry);
18571 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
18572 +               di_read_lock(p->parent, AuLock_IR, p->lsc_di);
18573 +
18574 +       h_dir = NULL;
18575 +       p->h_parent = au_h_dptr(p->parent, p->bindex);
18576 +       p->hdir = au_hi(p->parent->d_inode, p->bindex);
18577 +       if (p->hdir)
18578 +               h_dir = p->hdir->hi_inode;
18579 +
18580 +       /*
18581 +        * udba case, or
18582 +        * if DI_LOCKED is not set, then p->parent may be different
18583 +        * and h_parent can be NULL.
18584 +        */
18585 +       if (unlikely(!p->hdir || !h_dir || !p->h_parent)) {
18586 +               err = -EBUSY;
18587 +               if (!au_ftest_pin(p->flags, DI_LOCKED))
18588 +                       di_read_unlock(p->parent, AuLock_IR);
18589 +               dput(p->parent);
18590 +               p->parent = NULL;
18591 +               goto out_err;
18592 +       }
18593 +
18594 +       if (au_ftest_pin(p->flags, MNT_WRITE)) {
18595 +               p->h_mnt = au_br_mnt(p->br);
18596 +               err = vfsub_mnt_want_write(p->h_mnt);
18597 +               if (unlikely(err)) {
18598 +                       au_fclr_pin(p->flags, MNT_WRITE);
18599 +                       if (!au_ftest_pin(p->flags, DI_LOCKED))
18600 +                               di_read_unlock(p->parent, AuLock_IR);
18601 +                       dput(p->parent);
18602 +                       p->parent = NULL;
18603 +                       goto out_err;
18604 +               }
18605 +       }
18606 +
18607 +       au_igrab(h_dir);
18608 +       err = au_pin_hdir_lock(p);
18609 +       if (!err)
18610 +               goto out; /* success */
18611 +
18612 +       au_unpin(p);
18613 +
18614 +out_err:
18615 +       pr_err("err %d\n", err);
18616 +       err = au_busy_or_stale();
18617 +out:
18618 +       return err;
18619 +}
18620 +
18621 +void au_pin_init(struct au_pin *p, struct dentry *dentry,
18622 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
18623 +                unsigned int udba, unsigned char flags)
18624 +{
18625 +       p->dentry = dentry;
18626 +       p->udba = udba;
18627 +       p->lsc_di = lsc_di;
18628 +       p->lsc_hi = lsc_hi;
18629 +       p->flags = flags;
18630 +       p->bindex = bindex;
18631 +
18632 +       p->parent = NULL;
18633 +       p->hdir = NULL;
18634 +       p->h_mnt = NULL;
18635 +
18636 +       p->h_dentry = NULL;
18637 +       p->h_parent = NULL;
18638 +       p->br = NULL;
18639 +       p->task = current;
18640 +}
18641 +
18642 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
18643 +          unsigned int udba, unsigned char flags)
18644 +{
18645 +       au_pin_init(pin, dentry, bindex, AuLsc_DI_PARENT, AuLsc_I_PARENT2,
18646 +                   udba, flags);
18647 +       return au_do_pin(pin);
18648 +}
18649 +
18650 +/* ---------------------------------------------------------------------- */
18651 +
18652 +/*
18653 + * ->setattr() and ->getattr() are called in various cases.
18654 + * chmod, stat: dentry is revalidated.
18655 + * fchmod, fstat: file and dentry are not revalidated, additionally they may be
18656 + *               unhashed.
18657 + * for ->setattr(), ia->ia_file is passed from ftruncate only.
18658 + */
18659 +/* todo: consolidate with do_refresh() and simple_reval_dpath() */
18660 +int au_reval_for_attr(struct dentry *dentry, unsigned int sigen)
18661 +{
18662 +       int err;
18663 +       struct inode *inode;
18664 +       struct dentry *parent;
18665 +
18666 +       err = 0;
18667 +       inode = dentry->d_inode;
18668 +       if (au_digen_test(dentry, sigen)) {
18669 +               parent = dget_parent(dentry);
18670 +               di_read_lock_parent(parent, AuLock_IR);
18671 +               err = au_refresh_dentry(dentry, parent);
18672 +               di_read_unlock(parent, AuLock_IR);
18673 +               dput(parent);
18674 +       }
18675 +
18676 +       AuTraceErr(err);
18677 +       return err;
18678 +}
18679 +
18680 +int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
18681 +                    struct au_icpup_args *a)
18682 +{
18683 +       int err;
18684 +       loff_t sz;
18685 +       aufs_bindex_t bstart, ibstart;
18686 +       struct dentry *hi_wh, *parent;
18687 +       struct inode *inode;
18688 +       struct au_wr_dir_args wr_dir_args = {
18689 +               .force_btgt     = -1,
18690 +               .flags          = 0
18691 +       };
18692 +
18693 +       if (d_is_dir(dentry))
18694 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
18695 +       /* plink or hi_wh() case */
18696 +       bstart = au_dbstart(dentry);
18697 +       inode = dentry->d_inode;
18698 +       ibstart = au_ibstart(inode);
18699 +       if (bstart != ibstart && !au_test_ro(inode->i_sb, ibstart, inode))
18700 +               wr_dir_args.force_btgt = ibstart;
18701 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
18702 +       if (unlikely(err < 0))
18703 +               goto out;
18704 +       a->btgt = err;
18705 +       if (err != bstart)
18706 +               au_fset_icpup(a->flags, DID_CPUP);
18707 +
18708 +       err = 0;
18709 +       a->pin_flags = AuPin_MNT_WRITE;
18710 +       parent = NULL;
18711 +       if (!IS_ROOT(dentry)) {
18712 +               au_fset_pin(a->pin_flags, DI_LOCKED);
18713 +               parent = dget_parent(dentry);
18714 +               di_write_lock_parent(parent);
18715 +       }
18716 +
18717 +       err = au_pin(&a->pin, dentry, a->btgt, a->udba, a->pin_flags);
18718 +       if (unlikely(err))
18719 +               goto out_parent;
18720 +
18721 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
18722 +       a->h_inode = a->h_path.dentry->d_inode;
18723 +       sz = -1;
18724 +       if (ia && (ia->ia_valid & ATTR_SIZE)) {
18725 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
18726 +               if (ia->ia_size < i_size_read(a->h_inode))
18727 +                       sz = ia->ia_size;
18728 +               mutex_unlock(&a->h_inode->i_mutex);
18729 +       }
18730 +
18731 +       hi_wh = NULL;
18732 +       if (au_ftest_icpup(a->flags, DID_CPUP) && d_unlinked(dentry)) {
18733 +               hi_wh = au_hi_wh(inode, a->btgt);
18734 +               if (!hi_wh) {
18735 +                       struct au_cp_generic cpg = {
18736 +                               .dentry = dentry,
18737 +                               .bdst   = a->btgt,
18738 +                               .bsrc   = -1,
18739 +                               .len    = sz,
18740 +                               .pin    = &a->pin
18741 +                       };
18742 +                       err = au_sio_cpup_wh(&cpg, /*file*/NULL);
18743 +                       if (unlikely(err))
18744 +                               goto out_unlock;
18745 +                       hi_wh = au_hi_wh(inode, a->btgt);
18746 +                       /* todo: revalidate hi_wh? */
18747 +               }
18748 +       }
18749 +
18750 +       if (parent) {
18751 +               au_pin_set_parent_lflag(&a->pin, /*lflag*/0);
18752 +               di_downgrade_lock(parent, AuLock_IR);
18753 +               dput(parent);
18754 +               parent = NULL;
18755 +       }
18756 +       if (!au_ftest_icpup(a->flags, DID_CPUP))
18757 +               goto out; /* success */
18758 +
18759 +       if (!d_unhashed(dentry)) {
18760 +               struct au_cp_generic cpg = {
18761 +                       .dentry = dentry,
18762 +                       .bdst   = a->btgt,
18763 +                       .bsrc   = bstart,
18764 +                       .len    = sz,
18765 +                       .pin    = &a->pin,
18766 +                       .flags  = AuCpup_DTIME | AuCpup_HOPEN
18767 +               };
18768 +               err = au_sio_cpup_simple(&cpg);
18769 +               if (!err)
18770 +                       a->h_path.dentry = au_h_dptr(dentry, a->btgt);
18771 +       } else if (!hi_wh)
18772 +               a->h_path.dentry = au_h_dptr(dentry, a->btgt);
18773 +       else
18774 +               a->h_path.dentry = hi_wh; /* do not dget here */
18775 +
18776 +out_unlock:
18777 +       a->h_inode = a->h_path.dentry->d_inode;
18778 +       if (!err)
18779 +               goto out; /* success */
18780 +       au_unpin(&a->pin);
18781 +out_parent:
18782 +       if (parent) {
18783 +               di_write_unlock(parent);
18784 +               dput(parent);
18785 +       }
18786 +out:
18787 +       if (!err)
18788 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
18789 +       return err;
18790 +}
18791 +
18792 +static int aufs_setattr(struct dentry *dentry, struct iattr *ia)
18793 +{
18794 +       int err;
18795 +       struct inode *inode, *delegated;
18796 +       struct super_block *sb;
18797 +       struct file *file;
18798 +       struct au_icpup_args *a;
18799 +
18800 +       inode = dentry->d_inode;
18801 +       IMustLock(inode);
18802 +
18803 +       err = -ENOMEM;
18804 +       a = kzalloc(sizeof(*a), GFP_NOFS);
18805 +       if (unlikely(!a))
18806 +               goto out;
18807 +
18808 +       if (ia->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
18809 +               ia->ia_valid &= ~ATTR_MODE;
18810 +
18811 +       file = NULL;
18812 +       sb = dentry->d_sb;
18813 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18814 +       if (unlikely(err))
18815 +               goto out_kfree;
18816 +
18817 +       if (ia->ia_valid & ATTR_FILE) {
18818 +               /* currently ftruncate(2) only */
18819 +               AuDebugOn(!S_ISREG(inode->i_mode));
18820 +               file = ia->ia_file;
18821 +               err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
18822 +               if (unlikely(err))
18823 +                       goto out_si;
18824 +               ia->ia_file = au_hf_top(file);
18825 +               a->udba = AuOpt_UDBA_NONE;
18826 +       } else {
18827 +               /* fchmod() doesn't pass ia_file */
18828 +               a->udba = au_opt_udba(sb);
18829 +               di_write_lock_child(dentry);
18830 +               /* no d_unlinked(), to set UDBA_NONE for root */
18831 +               if (d_unhashed(dentry))
18832 +                       a->udba = AuOpt_UDBA_NONE;
18833 +               if (a->udba != AuOpt_UDBA_NONE) {
18834 +                       AuDebugOn(IS_ROOT(dentry));
18835 +                       err = au_reval_for_attr(dentry, au_sigen(sb));
18836 +                       if (unlikely(err))
18837 +                               goto out_dentry;
18838 +               }
18839 +       }
18840 +
18841 +       err = au_pin_and_icpup(dentry, ia, a);
18842 +       if (unlikely(err < 0))
18843 +               goto out_dentry;
18844 +       if (au_ftest_icpup(a->flags, DID_CPUP)) {
18845 +               ia->ia_file = NULL;
18846 +               ia->ia_valid &= ~ATTR_FILE;
18847 +       }
18848 +
18849 +       a->h_path.mnt = au_sbr_mnt(sb, a->btgt);
18850 +       if ((ia->ia_valid & (ATTR_MODE | ATTR_CTIME))
18851 +           == (ATTR_MODE | ATTR_CTIME)) {
18852 +               err = security_path_chmod(&a->h_path, ia->ia_mode);
18853 +               if (unlikely(err))
18854 +                       goto out_unlock;
18855 +       } else if ((ia->ia_valid & (ATTR_UID | ATTR_GID))
18856 +                  && (ia->ia_valid & ATTR_CTIME)) {
18857 +               err = security_path_chown(&a->h_path, ia->ia_uid, ia->ia_gid);
18858 +               if (unlikely(err))
18859 +                       goto out_unlock;
18860 +       }
18861 +
18862 +       if (ia->ia_valid & ATTR_SIZE) {
18863 +               struct file *f;
18864 +
18865 +               if (ia->ia_size < i_size_read(inode))
18866 +                       /* unmap only */
18867 +                       truncate_setsize(inode, ia->ia_size);
18868 +
18869 +               f = NULL;
18870 +               if (ia->ia_valid & ATTR_FILE)
18871 +                       f = ia->ia_file;
18872 +               mutex_unlock(&a->h_inode->i_mutex);
18873 +               err = vfsub_trunc(&a->h_path, ia->ia_size, ia->ia_valid, f);
18874 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
18875 +       } else {
18876 +               delegated = NULL;
18877 +               while (1) {
18878 +                       err = vfsub_notify_change(&a->h_path, ia, &delegated);
18879 +                       if (delegated) {
18880 +                               err = break_deleg_wait(&delegated);
18881 +                               if (!err)
18882 +                                       continue;
18883 +                       }
18884 +                       break;
18885 +               }
18886 +       }
18887 +       if (!err)
18888 +               au_cpup_attr_changeable(inode);
18889 +
18890 +out_unlock:
18891 +       mutex_unlock(&a->h_inode->i_mutex);
18892 +       au_unpin(&a->pin);
18893 +       if (unlikely(err))
18894 +               au_update_dbstart(dentry);
18895 +out_dentry:
18896 +       di_write_unlock(dentry);
18897 +       if (file) {
18898 +               fi_write_unlock(file);
18899 +               ia->ia_file = file;
18900 +               ia->ia_valid |= ATTR_FILE;
18901 +       }
18902 +out_si:
18903 +       si_read_unlock(sb);
18904 +out_kfree:
18905 +       kfree(a);
18906 +out:
18907 +       AuTraceErr(err);
18908 +       return err;
18909 +}
18910 +
18911 +#if IS_ENABLED(CONFIG_AUFS_XATTR) || IS_ENABLED(CONFIG_FS_POSIX_ACL)
18912 +static int au_h_path_to_set_attr(struct dentry *dentry,
18913 +                                struct au_icpup_args *a, struct path *h_path)
18914 +{
18915 +       int err;
18916 +       struct super_block *sb;
18917 +
18918 +       sb = dentry->d_sb;
18919 +       a->udba = au_opt_udba(sb);
18920 +       /* no d_unlinked(), to set UDBA_NONE for root */
18921 +       if (d_unhashed(dentry))
18922 +               a->udba = AuOpt_UDBA_NONE;
18923 +       if (a->udba != AuOpt_UDBA_NONE) {
18924 +               AuDebugOn(IS_ROOT(dentry));
18925 +               err = au_reval_for_attr(dentry, au_sigen(sb));
18926 +               if (unlikely(err))
18927 +                       goto out;
18928 +       }
18929 +       err = au_pin_and_icpup(dentry, /*ia*/NULL, a);
18930 +       if (unlikely(err < 0))
18931 +               goto out;
18932 +
18933 +       h_path->dentry = a->h_path.dentry;
18934 +       h_path->mnt = au_sbr_mnt(sb, a->btgt);
18935 +
18936 +out:
18937 +       return err;
18938 +}
18939 +
18940 +ssize_t au_srxattr(struct dentry *dentry, struct au_srxattr *arg)
18941 +{
18942 +       int err;
18943 +       struct path h_path;
18944 +       struct super_block *sb;
18945 +       struct au_icpup_args *a;
18946 +       struct inode *inode, *h_inode;
18947 +
18948 +       inode = dentry->d_inode;
18949 +       IMustLock(inode);
18950 +
18951 +       err = -ENOMEM;
18952 +       a = kzalloc(sizeof(*a), GFP_NOFS);
18953 +       if (unlikely(!a))
18954 +               goto out;
18955 +
18956 +       sb = dentry->d_sb;
18957 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
18958 +       if (unlikely(err))
18959 +               goto out_kfree;
18960 +
18961 +       h_path.dentry = NULL;   /* silence gcc */
18962 +       di_write_lock_child(dentry);
18963 +       err = au_h_path_to_set_attr(dentry, a, &h_path);
18964 +       if (unlikely(err))
18965 +               goto out_di;
18966 +
18967 +       mutex_unlock(&a->h_inode->i_mutex);
18968 +       switch (arg->type) {
18969 +       case AU_XATTR_SET:
18970 +               err = vfsub_setxattr(h_path.dentry,
18971 +                                    arg->u.set.name, arg->u.set.value,
18972 +                                    arg->u.set.size, arg->u.set.flags);
18973 +               break;
18974 +       case AU_XATTR_REMOVE:
18975 +               err = vfsub_removexattr(h_path.dentry, arg->u.remove.name);
18976 +               break;
18977 +       case AU_ACL_SET:
18978 +               err = -EOPNOTSUPP;
18979 +               h_inode = h_path.dentry->d_inode;
18980 +               if (h_inode->i_op->set_acl)
18981 +                       err = h_inode->i_op->set_acl(h_inode,
18982 +                                                    arg->u.acl_set.acl,
18983 +                                                    arg->u.acl_set.type);
18984 +               break;
18985 +       }
18986 +       if (!err)
18987 +               au_cpup_attr_timesizes(inode);
18988 +
18989 +       au_unpin(&a->pin);
18990 +       if (unlikely(err))
18991 +               au_update_dbstart(dentry);
18992 +
18993 +out_di:
18994 +       di_write_unlock(dentry);
18995 +       si_read_unlock(sb);
18996 +out_kfree:
18997 +       kfree(a);
18998 +out:
18999 +       AuTraceErr(err);
19000 +       return err;
19001 +}
19002 +#endif
19003 +
19004 +static void au_refresh_iattr(struct inode *inode, struct kstat *st,
19005 +                            unsigned int nlink)
19006 +{
19007 +       unsigned int n;
19008 +
19009 +       inode->i_mode = st->mode;
19010 +       /* don't i_[ug]id_write() here */
19011 +       inode->i_uid = st->uid;
19012 +       inode->i_gid = st->gid;
19013 +       inode->i_atime = st->atime;
19014 +       inode->i_mtime = st->mtime;
19015 +       inode->i_ctime = st->ctime;
19016 +
19017 +       au_cpup_attr_nlink(inode, /*force*/0);
19018 +       if (S_ISDIR(inode->i_mode)) {
19019 +               n = inode->i_nlink;
19020 +               n -= nlink;
19021 +               n += st->nlink;
19022 +               smp_mb(); /* for i_nlink */
19023 +               /* 0 can happen */
19024 +               set_nlink(inode, n);
19025 +       }
19026 +
19027 +       spin_lock(&inode->i_lock);
19028 +       inode->i_blocks = st->blocks;
19029 +       i_size_write(inode, st->size);
19030 +       spin_unlock(&inode->i_lock);
19031 +}
19032 +
19033 +/*
19034 + * common routine for aufs_getattr() and aufs_getxattr().
19035 + * returns zero or negative (an error).
19036 + * @dentry will be read-locked in success.
19037 + */
19038 +int au_h_path_getattr(struct dentry *dentry, int force, struct path *h_path)
19039 +{
19040 +       int err;
19041 +       unsigned int mnt_flags, sigen;
19042 +       unsigned char udba_none;
19043 +       aufs_bindex_t bindex;
19044 +       struct super_block *sb, *h_sb;
19045 +       struct inode *inode;
19046 +
19047 +       h_path->mnt = NULL;
19048 +       h_path->dentry = NULL;
19049 +
19050 +       err = 0;
19051 +       sb = dentry->d_sb;
19052 +       mnt_flags = au_mntflags(sb);
19053 +       udba_none = !!au_opt_test(mnt_flags, UDBA_NONE);
19054 +
19055 +       /* support fstat(2) */
19056 +       if (!d_unlinked(dentry) && !udba_none) {
19057 +               sigen = au_sigen(sb);
19058 +               err = au_digen_test(dentry, sigen);
19059 +               if (!err) {
19060 +                       di_read_lock_child(dentry, AuLock_IR);
19061 +                       err = au_dbrange_test(dentry);
19062 +                       if (unlikely(err)) {
19063 +                               di_read_unlock(dentry, AuLock_IR);
19064 +                               goto out;
19065 +                       }
19066 +               } else {
19067 +                       AuDebugOn(IS_ROOT(dentry));
19068 +                       di_write_lock_child(dentry);
19069 +                       err = au_dbrange_test(dentry);
19070 +                       if (!err)
19071 +                               err = au_reval_for_attr(dentry, sigen);
19072 +                       if (!err)
19073 +                               di_downgrade_lock(dentry, AuLock_IR);
19074 +                       else {
19075 +                               di_write_unlock(dentry);
19076 +                               goto out;
19077 +                       }
19078 +               }
19079 +       } else
19080 +               di_read_lock_child(dentry, AuLock_IR);
19081 +
19082 +       inode = dentry->d_inode;
19083 +       bindex = au_ibstart(inode);
19084 +       h_path->mnt = au_sbr_mnt(sb, bindex);
19085 +       h_sb = h_path->mnt->mnt_sb;
19086 +       if (!force
19087 +           && !au_test_fs_bad_iattr(h_sb)
19088 +           && udba_none)
19089 +               goto out; /* success */
19090 +
19091 +       if (au_dbstart(dentry) == bindex)
19092 +               h_path->dentry = au_h_dptr(dentry, bindex);
19093 +       else if (au_opt_test(mnt_flags, PLINK) && au_plink_test(inode)) {
19094 +               h_path->dentry = au_plink_lkup(inode, bindex);
19095 +               if (IS_ERR(h_path->dentry))
19096 +                       /* pretending success */
19097 +                       h_path->dentry = NULL;
19098 +               else
19099 +                       dput(h_path->dentry);
19100 +       }
19101 +
19102 +out:
19103 +       return err;
19104 +}
19105 +
19106 +static int aufs_getattr(struct vfsmount *mnt __maybe_unused,
19107 +                       struct dentry *dentry, struct kstat *st)
19108 +{
19109 +       int err;
19110 +       unsigned char positive;
19111 +       struct path h_path;
19112 +       struct inode *inode;
19113 +       struct super_block *sb;
19114 +
19115 +       inode = dentry->d_inode;
19116 +       sb = dentry->d_sb;
19117 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
19118 +       if (unlikely(err))
19119 +               goto out;
19120 +       err = au_h_path_getattr(dentry, /*force*/0, &h_path);
19121 +       if (unlikely(err))
19122 +               goto out_si;
19123 +       if (unlikely(!h_path.dentry))
19124 +               /* illegally overlapped or something */
19125 +               goto out_fill; /* pretending success */
19126 +
19127 +       positive = !!h_path.dentry->d_inode;
19128 +       if (positive)
19129 +               err = vfs_getattr(&h_path, st);
19130 +       if (!err) {
19131 +               if (positive)
19132 +                       au_refresh_iattr(inode, st,
19133 +                                        h_path.dentry->d_inode->i_nlink);
19134 +               goto out_fill; /* success */
19135 +       }
19136 +       AuTraceErr(err);
19137 +       goto out_di;
19138 +
19139 +out_fill:
19140 +       generic_fillattr(inode, st);
19141 +out_di:
19142 +       di_read_unlock(dentry, AuLock_IR);
19143 +out_si:
19144 +       si_read_unlock(sb);
19145 +out:
19146 +       AuTraceErr(err);
19147 +       return err;
19148 +}
19149 +
19150 +/* ---------------------------------------------------------------------- */
19151 +
19152 +static int h_readlink(struct dentry *dentry, int bindex, char __user *buf,
19153 +                     int bufsiz)
19154 +{
19155 +       int err;
19156 +       struct super_block *sb;
19157 +       struct dentry *h_dentry;
19158 +
19159 +       err = -EINVAL;
19160 +       h_dentry = au_h_dptr(dentry, bindex);
19161 +       if (unlikely(!h_dentry->d_inode->i_op->readlink))
19162 +               goto out;
19163 +
19164 +       err = security_inode_readlink(h_dentry);
19165 +       if (unlikely(err))
19166 +               goto out;
19167 +
19168 +       sb = dentry->d_sb;
19169 +       if (!au_test_ro(sb, bindex, dentry->d_inode)) {
19170 +               vfsub_touch_atime(au_sbr_mnt(sb, bindex), h_dentry);
19171 +               fsstack_copy_attr_atime(dentry->d_inode, h_dentry->d_inode);
19172 +       }
19173 +       err = h_dentry->d_inode->i_op->readlink(h_dentry, buf, bufsiz);
19174 +
19175 +out:
19176 +       return err;
19177 +}
19178 +
19179 +static int aufs_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
19180 +{
19181 +       int err;
19182 +
19183 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
19184 +       if (unlikely(err))
19185 +               goto out;
19186 +       err = au_d_hashed_positive(dentry);
19187 +       if (!err)
19188 +               err = h_readlink(dentry, au_dbstart(dentry), buf, bufsiz);
19189 +       aufs_read_unlock(dentry, AuLock_IR);
19190 +
19191 +out:
19192 +       return err;
19193 +}
19194 +
19195 +static void *aufs_follow_link(struct dentry *dentry, struct nameidata *nd)
19196 +{
19197 +       int err;
19198 +       mm_segment_t old_fs;
19199 +       union {
19200 +               char *k;
19201 +               char __user *u;
19202 +       } buf;
19203 +
19204 +       err = -ENOMEM;
19205 +       buf.k = (void *)__get_free_page(GFP_NOFS);
19206 +       if (unlikely(!buf.k))
19207 +               goto out;
19208 +
19209 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
19210 +       if (unlikely(err))
19211 +               goto out_name;
19212 +
19213 +       err = au_d_hashed_positive(dentry);
19214 +       if (!err) {
19215 +               old_fs = get_fs();
19216 +               set_fs(KERNEL_DS);
19217 +               err = h_readlink(dentry, au_dbstart(dentry), buf.u, PATH_MAX);
19218 +               set_fs(old_fs);
19219 +       }
19220 +       aufs_read_unlock(dentry, AuLock_IR);
19221 +
19222 +       if (err >= 0) {
19223 +               buf.k[err] = 0;
19224 +               /* will be freed by put_link */
19225 +               nd_set_link(nd, buf.k);
19226 +               return NULL; /* success */
19227 +       }
19228 +
19229 +out_name:
19230 +       free_page((unsigned long)buf.k);
19231 +out:
19232 +       AuTraceErr(err);
19233 +       return ERR_PTR(err);
19234 +}
19235 +
19236 +static void aufs_put_link(struct dentry *dentry __maybe_unused,
19237 +                         struct nameidata *nd, void *cookie __maybe_unused)
19238 +{
19239 +       char *p;
19240 +
19241 +       p = nd_get_link(nd);
19242 +       if (!IS_ERR_OR_NULL(p))
19243 +               free_page((unsigned long)p);
19244 +}
19245 +
19246 +/* ---------------------------------------------------------------------- */
19247 +
19248 +static int aufs_update_time(struct inode *inode, struct timespec *ts, int flags)
19249 +{
19250 +       int err;
19251 +       struct super_block *sb;
19252 +       struct inode *h_inode;
19253 +
19254 +       sb = inode->i_sb;
19255 +       /* mmap_sem might be acquired already, cf. aufs_mmap() */
19256 +       lockdep_off();
19257 +       si_read_lock(sb, AuLock_FLUSH);
19258 +       ii_write_lock_child(inode);
19259 +       lockdep_on();
19260 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
19261 +       err = vfsub_update_time(h_inode, ts, flags);
19262 +       lockdep_off();
19263 +       if (!err)
19264 +               au_cpup_attr_timesizes(inode);
19265 +       ii_write_unlock(inode);
19266 +       si_read_unlock(sb);
19267 +       lockdep_on();
19268 +
19269 +       if (!err && (flags & S_VERSION))
19270 +               inode_inc_iversion(inode);
19271 +
19272 +       return err;
19273 +}
19274 +
19275 +/* ---------------------------------------------------------------------- */
19276 +
19277 +struct inode_operations aufs_symlink_iop = {
19278 +       .permission     = aufs_permission,
19279 +#ifdef CONFIG_FS_POSIX_ACL
19280 +       .get_acl        = aufs_get_acl,
19281 +       .set_acl        = aufs_set_acl, /* unsupport for symlink? */
19282 +#endif
19283 +
19284 +       .setattr        = aufs_setattr,
19285 +       .getattr        = aufs_getattr,
19286 +
19287 +#ifdef CONFIG_AUFS_XATTR
19288 +       .setxattr       = aufs_setxattr,
19289 +       .getxattr       = aufs_getxattr,
19290 +       .listxattr      = aufs_listxattr,
19291 +       .removexattr    = aufs_removexattr,
19292 +#endif
19293 +
19294 +       .readlink       = aufs_readlink,
19295 +       .follow_link    = aufs_follow_link,
19296 +       .put_link       = aufs_put_link,
19297 +
19298 +       /* .update_time = aufs_update_time */
19299 +};
19300 +
19301 +struct inode_operations aufs_dir_iop = {
19302 +       .create         = aufs_create,
19303 +       .lookup         = aufs_lookup,
19304 +       .link           = aufs_link,
19305 +       .unlink         = aufs_unlink,
19306 +       .symlink        = aufs_symlink,
19307 +       .mkdir          = aufs_mkdir,
19308 +       .rmdir          = aufs_rmdir,
19309 +       .mknod          = aufs_mknod,
19310 +       .rename         = aufs_rename,
19311 +
19312 +       .permission     = aufs_permission,
19313 +#ifdef CONFIG_FS_POSIX_ACL
19314 +       .get_acl        = aufs_get_acl,
19315 +       .set_acl        = aufs_set_acl,
19316 +#endif
19317 +
19318 +       .setattr        = aufs_setattr,
19319 +       .getattr        = aufs_getattr,
19320 +
19321 +#ifdef CONFIG_AUFS_XATTR
19322 +       .setxattr       = aufs_setxattr,
19323 +       .getxattr       = aufs_getxattr,
19324 +       .listxattr      = aufs_listxattr,
19325 +       .removexattr    = aufs_removexattr,
19326 +#endif
19327 +
19328 +       .update_time    = aufs_update_time,
19329 +       /* no support for atomic_open() */
19330 +
19331 +       .tmpfile        = aufs_tmpfile
19332 +};
19333 +
19334 +struct inode_operations aufs_iop = {
19335 +       .permission     = aufs_permission,
19336 +#ifdef CONFIG_FS_POSIX_ACL
19337 +       .get_acl        = aufs_get_acl,
19338 +       .set_acl        = aufs_set_acl,
19339 +#endif
19340 +
19341 +       .setattr        = aufs_setattr,
19342 +       .getattr        = aufs_getattr,
19343 +
19344 +#ifdef CONFIG_AUFS_XATTR
19345 +       .setxattr       = aufs_setxattr,
19346 +       .getxattr       = aufs_getxattr,
19347 +       .listxattr      = aufs_listxattr,
19348 +       .removexattr    = aufs_removexattr,
19349 +#endif
19350 +
19351 +       .update_time    = aufs_update_time
19352 +};
19353 diff -urN /usr/share/empty/fs/aufs/i_op_del.c linux/fs/aufs/i_op_del.c
19354 --- /usr/share/empty/fs/aufs/i_op_del.c 1970-01-01 01:00:00.000000000 +0100
19355 +++ linux/fs/aufs/i_op_del.c    2015-03-27 21:56:35.466795000 +0100
19356 @@ -0,0 +1,507 @@
19357 +/*
19358 + * Copyright (C) 2005-2015 Junjiro R. Okajima
19359 + *
19360 + * This program, aufs is free software; you can redistribute it and/or modify
19361 + * it under the terms of the GNU General Public License as published by
19362 + * the Free Software Foundation; either version 2 of the License, or
19363 + * (at your option) any later version.
19364 + *
19365 + * This program is distributed in the hope that it will be useful,
19366 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19367 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19368 + * GNU General Public License for more details.
19369 + *
19370 + * You should have received a copy of the GNU General Public License
19371 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19372 + */
19373 +
19374 +/*
19375 + * inode operations (del entry)
19376 + */
19377 +
19378 +#include "aufs.h"
19379 +
19380 +/*
19381 + * decide if a new whiteout for @dentry is necessary or not.
19382 + * when it is necessary, prepare the parent dir for the upper branch whose
19383 + * branch index is @bcpup for creation. the actual creation of the whiteout will
19384 + * be done by caller.
19385 + * return value:
19386 + * 0: wh is unnecessary
19387 + * plus: wh is necessary
19388 + * minus: error
19389 + */
19390 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup)
19391 +{
19392 +       int need_wh, err;
19393 +       aufs_bindex_t bstart;
19394 +       struct super_block *sb;
19395 +
19396 +       sb = dentry->d_sb;
19397 +       bstart = au_dbstart(dentry);
19398 +       if (*bcpup < 0) {
19399 +               *bcpup = bstart;
19400 +               if (au_test_ro(sb, bstart, dentry->d_inode)) {
19401 +                       err = AuWbrCopyup(au_sbi(sb), dentry);
19402 +                       *bcpup = err;
19403 +                       if (unlikely(err < 0))
19404 +                               goto out;
19405 +               }
19406 +       } else
19407 +               AuDebugOn(bstart < *bcpup
19408 +                         || au_test_ro(sb, *bcpup, dentry->d_inode));
19409 +       AuDbg("bcpup %d, bstart %d\n", *bcpup, bstart);
19410 +
19411 +       if (*bcpup != bstart) {
19412 +               err = au_cpup_dirs(dentry, *bcpup);
19413 +               if (unlikely(err))
19414 +                       goto out;
19415 +               need_wh = 1;
19416 +       } else {
19417 +               struct au_dinfo *dinfo, *tmp;
19418 +
19419 +               need_wh = -ENOMEM;
19420 +               dinfo = au_di(dentry);
19421 +               tmp = au_di_alloc(sb, AuLsc_DI_TMP);
19422 +               if (tmp) {
19423 +                       au_di_cp(tmp, dinfo);
19424 +                       au_di_swap(tmp, dinfo);
19425 +                       /* returns the number of positive dentries */
19426 +                       need_wh = au_lkup_dentry(dentry, bstart + 1, /*type*/0);
19427 +                       au_di_swap(tmp, dinfo);
19428 +                       au_rw_write_unlock(&tmp->di_rwsem);
19429 +                       au_di_free(tmp);
19430 +               }
19431 +       }
19432 +       AuDbg("need_wh %d\n", need_wh);
19433 +       err = need_wh;
19434 +
19435 +out:
19436 +       return err;
19437 +}
19438 +
19439 +/*
19440 + * simple tests for the del-entry operations.
19441 + * following the checks in vfs, plus the parent-child relationship.
19442 + */
19443 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
19444 +              struct dentry *h_parent, int isdir)
19445 +{
19446 +       int err;
19447 +       umode_t h_mode;
19448 +       struct dentry *h_dentry, *h_latest;
19449 +       struct inode *h_inode;
19450 +
19451 +       h_dentry = au_h_dptr(dentry, bindex);
19452 +       h_inode = h_dentry->d_inode;
19453 +       if (dentry->d_inode) {
19454 +               err = -ENOENT;
19455 +               if (unlikely(!h_inode || !h_inode->i_nlink))
19456 +                       goto out;
19457 +
19458 +               h_mode = h_inode->i_mode;
19459 +               if (!isdir) {
19460 +                       err = -EISDIR;
19461 +                       if (unlikely(S_ISDIR(h_mode)))
19462 +                               goto out;
19463 +               } else if (unlikely(!S_ISDIR(h_mode))) {
19464 +                       err = -ENOTDIR;
19465 +                       goto out;
19466 +               }
19467 +       } else {
19468 +               /* rename(2) case */
19469 +               err = -EIO;
19470 +               if (unlikely(h_inode))
19471 +                       goto out;
19472 +       }
19473 +
19474 +       err = -ENOENT;
19475 +       /* expected parent dir is locked */
19476 +       if (unlikely(h_parent != h_dentry->d_parent))
19477 +               goto out;
19478 +       err = 0;
19479 +
19480 +       /*
19481 +        * rmdir a dir may break the consistency on some filesystem.
19482 +        * let's try heavy test.
19483 +        */
19484 +       err = -EACCES;
19485 +       if (unlikely(!au_opt_test(au_mntflags(dentry->d_sb), DIRPERM1)
19486 +                    && au_test_h_perm(h_parent->d_inode,
19487 +                                      MAY_EXEC | MAY_WRITE)))
19488 +               goto out;
19489 +
19490 +       h_latest = au_sio_lkup_one(&dentry->d_name, h_parent);
19491 +       err = -EIO;
19492 +       if (IS_ERR(h_latest))
19493 +               goto out;
19494 +       if (h_latest == h_dentry)
19495 +               err = 0;
19496 +       dput(h_latest);
19497 +
19498 +out:
19499 +       return err;
19500 +}
19501 +
19502 +/*
19503 + * decide the branch where we operate for @dentry. the branch index will be set
19504 + * @rbcpup. after diciding it, 'pin' it and store the timestamps of the parent
19505 + * dir for reverting.
19506 + * when a new whiteout is necessary, create it.
19507 + */
19508 +static struct dentry*
19509 +lock_hdir_create_wh(struct dentry *dentry, int isdir, aufs_bindex_t *rbcpup,
19510 +                   struct au_dtime *dt, struct au_pin *pin)
19511 +{
19512 +       struct dentry *wh_dentry;
19513 +       struct super_block *sb;
19514 +       struct path h_path;
19515 +       int err, need_wh;
19516 +       unsigned int udba;
19517 +       aufs_bindex_t bcpup;
19518 +
19519 +       need_wh = au_wr_dir_need_wh(dentry, isdir, rbcpup);
19520 +       wh_dentry = ERR_PTR(need_wh);
19521 +       if (unlikely(need_wh < 0))
19522 +               goto out;
19523 +
19524 +       sb = dentry->d_sb;
19525 +       udba = au_opt_udba(sb);
19526 +       bcpup = *rbcpup;
19527 +       err = au_pin(pin, dentry, bcpup, udba,
19528 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
19529 +       wh_dentry = ERR_PTR(err);
19530 +       if (unlikely(err))
19531 +               goto out;
19532 +
19533 +       h_path.dentry = au_pinned_h_parent(pin);
19534 +       if (udba != AuOpt_UDBA_NONE
19535 +           && au_dbstart(dentry) == bcpup) {
19536 +               err = au_may_del(dentry, bcpup, h_path.dentry, isdir);
19537 +               wh_dentry = ERR_PTR(err);
19538 +               if (unlikely(err))
19539 +                       goto out_unpin;
19540 +       }
19541 +
19542 +       h_path.mnt = au_sbr_mnt(sb, bcpup);
19543 +       au_dtime_store(dt, au_pinned_parent(pin), &h_path);
19544 +       wh_dentry = NULL;
19545 +       if (!need_wh)
19546 +               goto out; /* success, no need to create whiteout */
19547 +
19548 +       wh_dentry = au_wh_create(dentry, bcpup, h_path.dentry);
19549 +       if (IS_ERR(wh_dentry))
19550 +               goto out_unpin;
19551 +
19552 +       /* returns with the parent is locked and wh_dentry is dget-ed */
19553 +       goto out; /* success */
19554 +
19555 +out_unpin:
19556 +       au_unpin(pin);
19557 +out:
19558 +       return wh_dentry;
19559 +}
19560 +
19561 +/*
19562 + * when removing a dir, rename it to a unique temporary whiteout-ed name first
19563 + * in order to be revertible and save time for removing many child whiteouts
19564 + * under the dir.
19565 + * returns 1 when there are too many child whiteout and caller should remove
19566 + * them asynchronously. returns 0 when the number of children is enough small to
19567 + * remove now or the branch fs is a remote fs.
19568 + * otherwise return an error.
19569 + */
19570 +static int renwh_and_rmdir(struct dentry *dentry, aufs_bindex_t bindex,
19571 +                          struct au_nhash *whlist, struct inode *dir)
19572 +{
19573 +       int rmdir_later, err, dirwh;
19574 +       struct dentry *h_dentry;
19575 +       struct super_block *sb;
19576 +
19577 +       sb = dentry->d_sb;
19578 +       SiMustAnyLock(sb);
19579 +       h_dentry = au_h_dptr(dentry, bindex);
19580 +       err = au_whtmp_ren(h_dentry, au_sbr(sb, bindex));
19581 +       if (unlikely(err))
19582 +               goto out;
19583 +
19584 +       /* stop monitoring */
19585 +       au_hn_free(au_hi(dentry->d_inode, bindex));
19586 +
19587 +       if (!au_test_fs_remote(h_dentry->d_sb)) {
19588 +               dirwh = au_sbi(sb)->si_dirwh;
19589 +               rmdir_later = (dirwh <= 1);
19590 +               if (!rmdir_later)
19591 +                       rmdir_later = au_nhash_test_longer_wh(whlist, bindex,
19592 +                                                             dirwh);
19593 +               if (rmdir_later)
19594 +                       return rmdir_later;
19595 +       }
19596 +
19597 +       err = au_whtmp_rmdir(dir, bindex, h_dentry, whlist);
19598 +       if (unlikely(err)) {
19599 +               AuIOErr("rmdir %pd, b%d failed, %d. ignored\n",
19600 +                       h_dentry, bindex, err);
19601 +               err = 0;
19602 +       }
19603 +
19604 +out:
19605 +       AuTraceErr(err);
19606 +       return err;
19607 +}
19608 +
19609 +/*
19610 + * final procedure for deleting a entry.
19611 + * maintain dentry and iattr.
19612 + */
19613 +static void epilog(struct inode *dir, struct dentry *dentry,
19614 +                  aufs_bindex_t bindex)
19615 +{
19616 +       struct inode *inode;
19617 +
19618 +       inode = dentry->d_inode;
19619 +       d_drop(dentry);
19620 +       inode->i_ctime = dir->i_ctime;
19621 +
19622 +       if (au_ibstart(dir) == bindex)
19623 +               au_cpup_attr_timesizes(dir);
19624 +       dir->i_version++;
19625 +}
19626 +
19627 +/*
19628 + * when an error happened, remove the created whiteout and revert everything.
19629 + */
19630 +static int do_revert(int err, struct inode *dir, aufs_bindex_t bindex,
19631 +                    aufs_bindex_t bwh, struct dentry *wh_dentry,
19632 +                    struct dentry *dentry, struct au_dtime *dt)
19633 +{
19634 +       int rerr;
19635 +       struct path h_path = {
19636 +               .dentry = wh_dentry,
19637 +               .mnt    = au_sbr_mnt(dir->i_sb, bindex)
19638 +       };
19639 +
19640 +       rerr = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path, dentry);
19641 +       if (!rerr) {
19642 +               au_set_dbwh(dentry, bwh);
19643 +               au_dtime_revert(dt);
19644 +               return 0;
19645 +       }
19646 +
19647 +       AuIOErr("%pd reverting whiteout failed(%d, %d)\n", dentry, err, rerr);
19648 +       return -EIO;
19649 +}
19650 +
19651 +/* ---------------------------------------------------------------------- */
19652 +
19653 +int aufs_unlink(struct inode *dir, struct dentry *dentry)
19654 +{
19655 +       int err;
19656 +       aufs_bindex_t bwh, bindex, bstart;
19657 +       struct inode *inode, *h_dir, *delegated;
19658 +       struct dentry *parent, *wh_dentry;
19659 +       /* to reuduce stack size */
19660 +       struct {
19661 +               struct au_dtime dt;
19662 +               struct au_pin pin;
19663 +               struct path h_path;
19664 +       } *a;
19665 +
19666 +       IMustLock(dir);
19667 +
19668 +       err = -ENOMEM;
19669 +       a = kmalloc(sizeof(*a), GFP_NOFS);
19670 +       if (unlikely(!a))
19671 +               goto out;
19672 +
19673 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
19674 +       if (unlikely(err))
19675 +               goto out_free;
19676 +       err = au_d_hashed_positive(dentry);
19677 +       if (unlikely(err))
19678 +               goto out_unlock;
19679 +       inode = dentry->d_inode;
19680 +       IMustLock(inode);
19681 +       err = -EISDIR;
19682 +       if (unlikely(d_is_dir(dentry)))
19683 +               goto out_unlock; /* possible? */
19684 +
19685 +       bstart = au_dbstart(dentry);
19686 +       bwh = au_dbwh(dentry);
19687 +       bindex = -1;
19688 +       parent = dentry->d_parent; /* dir inode is locked */
19689 +       di_write_lock_parent(parent);
19690 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/0, &bindex, &a->dt,
19691 +                                       &a->pin);
19692 +       err = PTR_ERR(wh_dentry);
19693 +       if (IS_ERR(wh_dentry))
19694 +               goto out_parent;
19695 +
19696 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, bstart);
19697 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
19698 +       dget(a->h_path.dentry);
19699 +       if (bindex == bstart) {
19700 +               h_dir = au_pinned_h_dir(&a->pin);
19701 +               delegated = NULL;
19702 +               err = vfsub_unlink(h_dir, &a->h_path, &delegated, /*force*/0);
19703 +               if (unlikely(err == -EWOULDBLOCK)) {
19704 +                       pr_warn("cannot retry for NFSv4 delegation"
19705 +                               " for an internal unlink\n");
19706 +                       iput(delegated);
19707 +               }
19708 +       } else {
19709 +               /* dir inode is locked */
19710 +               h_dir = wh_dentry->d_parent->d_inode;
19711 +               IMustLock(h_dir);
19712 +               err = 0;
19713 +       }
19714 +
19715 +       if (!err) {
19716 +               vfsub_drop_nlink(inode);
19717 +               epilog(dir, dentry, bindex);
19718 +
19719 +               /* update target timestamps */
19720 +               if (bindex == bstart) {
19721 +                       vfsub_update_h_iattr(&a->h_path, /*did*/NULL);
19722 +                       /*ignore*/
19723 +                       inode->i_ctime = a->h_path.dentry->d_inode->i_ctime;
19724 +               } else
19725 +                       /* todo: this timestamp may be reverted later */
19726 +                       inode->i_ctime = h_dir->i_ctime;
19727 +               goto out_unpin; /* success */
19728 +       }
19729 +
19730 +       /* revert */
19731 +       if (wh_dentry) {
19732 +               int rerr;
19733 +
19734 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
19735 +                                &a->dt);
19736 +               if (rerr)
19737 +                       err = rerr;
19738 +       }
19739 +
19740 +out_unpin:
19741 +       au_unpin(&a->pin);
19742 +       dput(wh_dentry);
19743 +       dput(a->h_path.dentry);
19744 +out_parent:
19745 +       di_write_unlock(parent);
19746 +out_unlock:
19747 +       aufs_read_unlock(dentry, AuLock_DW);
19748 +out_free:
19749 +       kfree(a);
19750 +out:
19751 +       return err;
19752 +}
19753 +
19754 +int aufs_rmdir(struct inode *dir, struct dentry *dentry)
19755 +{
19756 +       int err, rmdir_later;
19757 +       aufs_bindex_t bwh, bindex, bstart;
19758 +       struct inode *inode;
19759 +       struct dentry *parent, *wh_dentry, *h_dentry;
19760 +       struct au_whtmp_rmdir *args;
19761 +       /* to reuduce stack size */
19762 +       struct {
19763 +               struct au_dtime dt;
19764 +               struct au_pin pin;
19765 +       } *a;
19766 +
19767 +       IMustLock(dir);
19768 +
19769 +       err = -ENOMEM;
19770 +       a = kmalloc(sizeof(*a), GFP_NOFS);
19771 +       if (unlikely(!a))
19772 +               goto out;
19773 +
19774 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
19775 +       if (unlikely(err))
19776 +               goto out_free;
19777 +       err = au_alive_dir(dentry);
19778 +       if (unlikely(err))
19779 +               goto out_unlock;
19780 +       inode = dentry->d_inode;
19781 +       IMustLock(inode);
19782 +       err = -ENOTDIR;
19783 +       if (unlikely(!d_is_dir(dentry)))
19784 +               goto out_unlock; /* possible? */
19785 +
19786 +       err = -ENOMEM;
19787 +       args = au_whtmp_rmdir_alloc(dir->i_sb, GFP_NOFS);
19788 +       if (unlikely(!args))
19789 +               goto out_unlock;
19790 +
19791 +       parent = dentry->d_parent; /* dir inode is locked */
19792 +       di_write_lock_parent(parent);
19793 +       err = au_test_empty(dentry, &args->whlist);
19794 +       if (unlikely(err))
19795 +               goto out_parent;
19796 +
19797 +       bstart = au_dbstart(dentry);
19798 +       bwh = au_dbwh(dentry);
19799 +       bindex = -1;
19800 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/1, &bindex, &a->dt,
19801 +                                       &a->pin);
19802 +       err = PTR_ERR(wh_dentry);
19803 +       if (IS_ERR(wh_dentry))
19804 +               goto out_parent;
19805 +
19806 +       h_dentry = au_h_dptr(dentry, bstart);
19807 +       dget(h_dentry);
19808 +       rmdir_later = 0;
19809 +       if (bindex == bstart) {
19810 +               err = renwh_and_rmdir(dentry, bstart, &args->whlist, dir);
19811 +               if (err > 0) {
19812 +                       rmdir_later = err;
19813 +                       err = 0;
19814 +               }
19815 +       } else {
19816 +               /* stop monitoring */
19817 +               au_hn_free(au_hi(inode, bstart));
19818 +
19819 +               /* dir inode is locked */
19820 +               IMustLock(wh_dentry->d_parent->d_inode);
19821 +               err = 0;
19822 +       }
19823 +
19824 +       if (!err) {
19825 +               vfsub_dead_dir(inode);
19826 +               au_set_dbdiropq(dentry, -1);
19827 +               epilog(dir, dentry, bindex);
19828 +
19829 +               if (rmdir_later) {
19830 +                       au_whtmp_kick_rmdir(dir, bstart, h_dentry, args);
19831 +                       args = NULL;
19832 +               }
19833 +
19834 +               goto out_unpin; /* success */
19835 +       }
19836 +
19837 +       /* revert */
19838 +       AuLabel(revert);
19839 +       if (wh_dentry) {
19840 +               int rerr;
19841 +
19842 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
19843 +                                &a->dt);
19844 +               if (rerr)
19845 +                       err = rerr;
19846 +       }
19847 +
19848 +out_unpin:
19849 +       au_unpin(&a->pin);
19850 +       dput(wh_dentry);
19851 +       dput(h_dentry);
19852 +out_parent:
19853 +       di_write_unlock(parent);
19854 +       if (args)
19855 +               au_whtmp_rmdir_free(args);
19856 +out_unlock:
19857 +       aufs_read_unlock(dentry, AuLock_DW);
19858 +out_free:
19859 +       kfree(a);
19860 +out:
19861 +       AuTraceErr(err);
19862 +       return err;
19863 +}
19864 diff -urN /usr/share/empty/fs/aufs/i_op_ren.c linux/fs/aufs/i_op_ren.c
19865 --- /usr/share/empty/fs/aufs/i_op_ren.c 1970-01-01 01:00:00.000000000 +0100
19866 +++ linux/fs/aufs/i_op_ren.c    2015-03-27 21:56:35.466795000 +0100
19867 @@ -0,0 +1,1035 @@
19868 +/*
19869 + * Copyright (C) 2005-2015 Junjiro R. Okajima
19870 + *
19871 + * This program, aufs is free software; you can redistribute it and/or modify
19872 + * it under the terms of the GNU General Public License as published by
19873 + * the Free Software Foundation; either version 2 of the License, or
19874 + * (at your option) any later version.
19875 + *
19876 + * This program is distributed in the hope that it will be useful,
19877 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19878 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19879 + * GNU General Public License for more details.
19880 + *
19881 + * You should have received a copy of the GNU General Public License
19882 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19883 + */
19884 +
19885 +/*
19886 + * inode operation (rename entry)
19887 + * todo: this is crazy monster
19888 + */
19889 +
19890 +#include "aufs.h"
19891 +
19892 +enum { AuSRC, AuDST, AuSrcDst };
19893 +enum { AuPARENT, AuCHILD, AuParentChild };
19894 +
19895 +#define AuRen_ISDIR    1
19896 +#define AuRen_ISSAMEDIR        (1 << 1)
19897 +#define AuRen_WHSRC    (1 << 2)
19898 +#define AuRen_WHDST    (1 << 3)
19899 +#define AuRen_MNT_WRITE        (1 << 4)
19900 +#define AuRen_DT_DSTDIR        (1 << 5)
19901 +#define AuRen_DIROPQ   (1 << 6)
19902 +#define AuRen_CPUP     (1 << 7)
19903 +#define au_ftest_ren(flags, name)      ((flags) & AuRen_##name)
19904 +#define au_fset_ren(flags, name) \
19905 +       do { (flags) |= AuRen_##name; } while (0)
19906 +#define au_fclr_ren(flags, name) \
19907 +       do { (flags) &= ~AuRen_##name; } while (0)
19908 +
19909 +struct au_ren_args {
19910 +       struct {
19911 +               struct dentry *dentry, *h_dentry, *parent, *h_parent,
19912 +                       *wh_dentry;
19913 +               struct inode *dir, *inode;
19914 +               struct au_hinode *hdir;
19915 +               struct au_dtime dt[AuParentChild];
19916 +               aufs_bindex_t bstart;
19917 +       } sd[AuSrcDst];
19918 +
19919 +#define src_dentry     sd[AuSRC].dentry
19920 +#define src_dir                sd[AuSRC].dir
19921 +#define src_inode      sd[AuSRC].inode
19922 +#define src_h_dentry   sd[AuSRC].h_dentry
19923 +#define src_parent     sd[AuSRC].parent
19924 +#define src_h_parent   sd[AuSRC].h_parent
19925 +#define src_wh_dentry  sd[AuSRC].wh_dentry
19926 +#define src_hdir       sd[AuSRC].hdir
19927 +#define src_h_dir      sd[AuSRC].hdir->hi_inode
19928 +#define src_dt         sd[AuSRC].dt
19929 +#define src_bstart     sd[AuSRC].bstart
19930 +
19931 +#define dst_dentry     sd[AuDST].dentry
19932 +#define dst_dir                sd[AuDST].dir
19933 +#define dst_inode      sd[AuDST].inode
19934 +#define dst_h_dentry   sd[AuDST].h_dentry
19935 +#define dst_parent     sd[AuDST].parent
19936 +#define dst_h_parent   sd[AuDST].h_parent
19937 +#define dst_wh_dentry  sd[AuDST].wh_dentry
19938 +#define dst_hdir       sd[AuDST].hdir
19939 +#define dst_h_dir      sd[AuDST].hdir->hi_inode
19940 +#define dst_dt         sd[AuDST].dt
19941 +#define dst_bstart     sd[AuDST].bstart
19942 +
19943 +       struct dentry *h_trap;
19944 +       struct au_branch *br;
19945 +       struct au_hinode *src_hinode;
19946 +       struct path h_path;
19947 +       struct au_nhash whlist;
19948 +       aufs_bindex_t btgt, src_bwh, src_bdiropq;
19949 +
19950 +       unsigned int flags;
19951 +
19952 +       struct au_whtmp_rmdir *thargs;
19953 +       struct dentry *h_dst;
19954 +};
19955 +
19956 +/* ---------------------------------------------------------------------- */
19957 +
19958 +/*
19959 + * functions for reverting.
19960 + * when an error happened in a single rename systemcall, we should revert
19961 + * everything as if nothing happend.
19962 + * we don't need to revert the copied-up/down the parent dir since they are
19963 + * harmless.
19964 + */
19965 +
19966 +#define RevertFailure(fmt, ...) do { \
19967 +       AuIOErr("revert failure: " fmt " (%d, %d)\n", \
19968 +               ##__VA_ARGS__, err, rerr); \
19969 +       err = -EIO; \
19970 +} while (0)
19971 +
19972 +static void au_ren_rev_diropq(int err, struct au_ren_args *a)
19973 +{
19974 +       int rerr;
19975 +
19976 +       au_hn_imtx_lock_nested(a->src_hinode, AuLsc_I_CHILD);
19977 +       rerr = au_diropq_remove(a->src_dentry, a->btgt);
19978 +       au_hn_imtx_unlock(a->src_hinode);
19979 +       au_set_dbdiropq(a->src_dentry, a->src_bdiropq);
19980 +       if (rerr)
19981 +               RevertFailure("remove diropq %pd", a->src_dentry);
19982 +}
19983 +
19984 +static void au_ren_rev_rename(int err, struct au_ren_args *a)
19985 +{
19986 +       int rerr;
19987 +       struct inode *delegated;
19988 +
19989 +       a->h_path.dentry = vfsub_lkup_one(&a->src_dentry->d_name,
19990 +                                         a->src_h_parent);
19991 +       rerr = PTR_ERR(a->h_path.dentry);
19992 +       if (IS_ERR(a->h_path.dentry)) {
19993 +               RevertFailure("lkup one %pd", a->src_dentry);
19994 +               return;
19995 +       }
19996 +
19997 +       delegated = NULL;
19998 +       rerr = vfsub_rename(a->dst_h_dir,
19999 +                           au_h_dptr(a->src_dentry, a->btgt),
20000 +                           a->src_h_dir, &a->h_path, &delegated);
20001 +       if (unlikely(rerr == -EWOULDBLOCK)) {
20002 +               pr_warn("cannot retry for NFSv4 delegation"
20003 +                       " for an internal rename\n");
20004 +               iput(delegated);
20005 +       }
20006 +       d_drop(a->h_path.dentry);
20007 +       dput(a->h_path.dentry);
20008 +       /* au_set_h_dptr(a->src_dentry, a->btgt, NULL); */
20009 +       if (rerr)
20010 +               RevertFailure("rename %pd", a->src_dentry);
20011 +}
20012 +
20013 +static void au_ren_rev_cpup(int err, struct au_ren_args *a)
20014 +{
20015 +       int rerr;
20016 +
20017 +       a->h_path.dentry = a->dst_h_dentry;
20018 +       /* no delegation since it is just created */
20019 +       rerr = vfsub_unlink(a->dst_h_dir, &a->h_path, /*delegated*/NULL,
20020 +                           /*force*/0);
20021 +       au_set_h_dptr(a->src_dentry, a->btgt, NULL);
20022 +       au_set_dbstart(a->src_dentry, a->src_bstart);
20023 +       if (rerr)
20024 +               RevertFailure("unlink %pd", a->dst_h_dentry);
20025 +}
20026 +
20027 +static void au_ren_rev_whtmp(int err, struct au_ren_args *a)
20028 +{
20029 +       int rerr;
20030 +       struct inode *delegated;
20031 +
20032 +       a->h_path.dentry = vfsub_lkup_one(&a->dst_dentry->d_name,
20033 +                                         a->dst_h_parent);
20034 +       rerr = PTR_ERR(a->h_path.dentry);
20035 +       if (IS_ERR(a->h_path.dentry)) {
20036 +               RevertFailure("lkup one %pd", a->dst_dentry);
20037 +               return;
20038 +       }
20039 +       if (a->h_path.dentry->d_inode) {
20040 +               d_drop(a->h_path.dentry);
20041 +               dput(a->h_path.dentry);
20042 +               return;
20043 +       }
20044 +
20045 +       delegated = NULL;
20046 +       rerr = vfsub_rename(a->dst_h_dir, a->h_dst, a->dst_h_dir, &a->h_path,
20047 +                           &delegated);
20048 +       if (unlikely(rerr == -EWOULDBLOCK)) {
20049 +               pr_warn("cannot retry for NFSv4 delegation"
20050 +                       " for an internal rename\n");
20051 +               iput(delegated);
20052 +       }
20053 +       d_drop(a->h_path.dentry);
20054 +       dput(a->h_path.dentry);
20055 +       if (!rerr)
20056 +               au_set_h_dptr(a->dst_dentry, a->btgt, dget(a->h_dst));
20057 +       else
20058 +               RevertFailure("rename %pd", a->h_dst);
20059 +}
20060 +
20061 +static void au_ren_rev_whsrc(int err, struct au_ren_args *a)
20062 +{
20063 +       int rerr;
20064 +
20065 +       a->h_path.dentry = a->src_wh_dentry;
20066 +       rerr = au_wh_unlink_dentry(a->src_h_dir, &a->h_path, a->src_dentry);
20067 +       au_set_dbwh(a->src_dentry, a->src_bwh);
20068 +       if (rerr)
20069 +               RevertFailure("unlink %pd", a->src_wh_dentry);
20070 +}
20071 +#undef RevertFailure
20072 +
20073 +/* ---------------------------------------------------------------------- */
20074 +
20075 +/*
20076 + * when we have to copyup the renaming entry, do it with the rename-target name
20077 + * in order to minimize the cost (the later actual rename is unnecessary).
20078 + * otherwise rename it on the target branch.
20079 + */
20080 +static int au_ren_or_cpup(struct au_ren_args *a)
20081 +{
20082 +       int err;
20083 +       struct dentry *d;
20084 +       struct inode *delegated;
20085 +
20086 +       d = a->src_dentry;
20087 +       if (au_dbstart(d) == a->btgt) {
20088 +               a->h_path.dentry = a->dst_h_dentry;
20089 +               if (au_ftest_ren(a->flags, DIROPQ)
20090 +                   && au_dbdiropq(d) == a->btgt)
20091 +                       au_fclr_ren(a->flags, DIROPQ);
20092 +               AuDebugOn(au_dbstart(d) != a->btgt);
20093 +               delegated = NULL;
20094 +               err = vfsub_rename(a->src_h_dir, au_h_dptr(d, a->btgt),
20095 +                                  a->dst_h_dir, &a->h_path, &delegated);
20096 +               if (unlikely(err == -EWOULDBLOCK)) {
20097 +                       pr_warn("cannot retry for NFSv4 delegation"
20098 +                               " for an internal rename\n");
20099 +                       iput(delegated);
20100 +               }
20101 +       } else
20102 +               BUG();
20103 +
20104 +       if (!err && a->h_dst)
20105 +               /* it will be set to dinfo later */
20106 +               dget(a->h_dst);
20107 +
20108 +       return err;
20109 +}
20110 +
20111 +/* cf. aufs_rmdir() */
20112 +static int au_ren_del_whtmp(struct au_ren_args *a)
20113 +{
20114 +       int err;
20115 +       struct inode *dir;
20116 +
20117 +       dir = a->dst_dir;
20118 +       SiMustAnyLock(dir->i_sb);
20119 +       if (!au_nhash_test_longer_wh(&a->whlist, a->btgt,
20120 +                                    au_sbi(dir->i_sb)->si_dirwh)
20121 +           || au_test_fs_remote(a->h_dst->d_sb)) {
20122 +               err = au_whtmp_rmdir(dir, a->btgt, a->h_dst, &a->whlist);
20123 +               if (unlikely(err))
20124 +                       pr_warn("failed removing whtmp dir %pd (%d), "
20125 +                               "ignored.\n", a->h_dst, err);
20126 +       } else {
20127 +               au_nhash_wh_free(&a->thargs->whlist);
20128 +               a->thargs->whlist = a->whlist;
20129 +               a->whlist.nh_num = 0;
20130 +               au_whtmp_kick_rmdir(dir, a->btgt, a->h_dst, a->thargs);
20131 +               dput(a->h_dst);
20132 +               a->thargs = NULL;
20133 +       }
20134 +
20135 +       return 0;
20136 +}
20137 +
20138 +/* make it 'opaque' dir. */
20139 +static int au_ren_diropq(struct au_ren_args *a)
20140 +{
20141 +       int err;
20142 +       struct dentry *diropq;
20143 +
20144 +       err = 0;
20145 +       a->src_bdiropq = au_dbdiropq(a->src_dentry);
20146 +       a->src_hinode = au_hi(a->src_inode, a->btgt);
20147 +       au_hn_imtx_lock_nested(a->src_hinode, AuLsc_I_CHILD);
20148 +       diropq = au_diropq_create(a->src_dentry, a->btgt);
20149 +       au_hn_imtx_unlock(a->src_hinode);
20150 +       if (IS_ERR(diropq))
20151 +               err = PTR_ERR(diropq);
20152 +       else
20153 +               dput(diropq);
20154 +
20155 +       return err;
20156 +}
20157 +
20158 +static int do_rename(struct au_ren_args *a)
20159 +{
20160 +       int err;
20161 +       struct dentry *d, *h_d;
20162 +
20163 +       /* prepare workqueue args for asynchronous rmdir */
20164 +       h_d = a->dst_h_dentry;
20165 +       if (au_ftest_ren(a->flags, ISDIR) && h_d->d_inode) {
20166 +               err = -ENOMEM;
20167 +               a->thargs = au_whtmp_rmdir_alloc(a->src_dentry->d_sb, GFP_NOFS);
20168 +               if (unlikely(!a->thargs))
20169 +                       goto out;
20170 +               a->h_dst = dget(h_d);
20171 +       }
20172 +
20173 +       /* create whiteout for src_dentry */
20174 +       if (au_ftest_ren(a->flags, WHSRC)) {
20175 +               a->src_bwh = au_dbwh(a->src_dentry);
20176 +               AuDebugOn(a->src_bwh >= 0);
20177 +               a->src_wh_dentry
20178 +                       = au_wh_create(a->src_dentry, a->btgt, a->src_h_parent);
20179 +               err = PTR_ERR(a->src_wh_dentry);
20180 +               if (IS_ERR(a->src_wh_dentry))
20181 +                       goto out_thargs;
20182 +       }
20183 +
20184 +       /* lookup whiteout for dentry */
20185 +       if (au_ftest_ren(a->flags, WHDST)) {
20186 +               h_d = au_wh_lkup(a->dst_h_parent, &a->dst_dentry->d_name,
20187 +                                a->br);
20188 +               err = PTR_ERR(h_d);
20189 +               if (IS_ERR(h_d))
20190 +                       goto out_whsrc;
20191 +               if (!h_d->d_inode)
20192 +                       dput(h_d);
20193 +               else
20194 +                       a->dst_wh_dentry = h_d;
20195 +       }
20196 +
20197 +       /* rename dentry to tmpwh */
20198 +       if (a->thargs) {
20199 +               err = au_whtmp_ren(a->dst_h_dentry, a->br);
20200 +               if (unlikely(err))
20201 +                       goto out_whdst;
20202 +
20203 +               d = a->dst_dentry;
20204 +               au_set_h_dptr(d, a->btgt, NULL);
20205 +               err = au_lkup_neg(d, a->btgt, /*wh*/0);
20206 +               if (unlikely(err))
20207 +                       goto out_whtmp;
20208 +               a->dst_h_dentry = au_h_dptr(d, a->btgt);
20209 +       }
20210 +
20211 +       BUG_ON(a->dst_h_dentry->d_inode && a->src_bstart != a->btgt);
20212 +
20213 +       /* rename by vfs_rename or cpup */
20214 +       d = a->dst_dentry;
20215 +       if (au_ftest_ren(a->flags, ISDIR)
20216 +           && (a->dst_wh_dentry
20217 +               || au_dbdiropq(d) == a->btgt
20218 +               /* hide the lower to keep xino */
20219 +               || a->btgt < au_dbend(d)
20220 +               || au_opt_test(au_mntflags(d->d_sb), ALWAYS_DIROPQ)))
20221 +               au_fset_ren(a->flags, DIROPQ);
20222 +       err = au_ren_or_cpup(a);
20223 +       if (unlikely(err))
20224 +               /* leave the copied-up one */
20225 +               goto out_whtmp;
20226 +
20227 +       /* make dir opaque */
20228 +       if (au_ftest_ren(a->flags, DIROPQ)) {
20229 +               err = au_ren_diropq(a);
20230 +               if (unlikely(err))
20231 +                       goto out_rename;
20232 +       }
20233 +
20234 +       /* update target timestamps */
20235 +       AuDebugOn(au_dbstart(a->src_dentry) != a->btgt);
20236 +       a->h_path.dentry = au_h_dptr(a->src_dentry, a->btgt);
20237 +       vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
20238 +       a->src_inode->i_ctime = a->h_path.dentry->d_inode->i_ctime;
20239 +
20240 +       /* remove whiteout for dentry */
20241 +       if (a->dst_wh_dentry) {
20242 +               a->h_path.dentry = a->dst_wh_dentry;
20243 +               err = au_wh_unlink_dentry(a->dst_h_dir, &a->h_path,
20244 +                                         a->dst_dentry);
20245 +               if (unlikely(err))
20246 +                       goto out_diropq;
20247 +       }
20248 +
20249 +       /* remove whtmp */
20250 +       if (a->thargs)
20251 +               au_ren_del_whtmp(a); /* ignore this error */
20252 +
20253 +       au_fhsm_wrote(a->src_dentry->d_sb, a->btgt, /*force*/0);
20254 +       err = 0;
20255 +       goto out_success;
20256 +
20257 +out_diropq:
20258 +       if (au_ftest_ren(a->flags, DIROPQ))
20259 +               au_ren_rev_diropq(err, a);
20260 +out_rename:
20261 +       if (!au_ftest_ren(a->flags, CPUP))
20262 +               au_ren_rev_rename(err, a);
20263 +       else
20264 +               au_ren_rev_cpup(err, a);
20265 +       dput(a->h_dst);
20266 +out_whtmp:
20267 +       if (a->thargs)
20268 +               au_ren_rev_whtmp(err, a);
20269 +out_whdst:
20270 +       dput(a->dst_wh_dentry);
20271 +       a->dst_wh_dentry = NULL;
20272 +out_whsrc:
20273 +       if (a->src_wh_dentry)
20274 +               au_ren_rev_whsrc(err, a);
20275 +out_success:
20276 +       dput(a->src_wh_dentry);
20277 +       dput(a->dst_wh_dentry);
20278 +out_thargs:
20279 +       if (a->thargs) {
20280 +               dput(a->h_dst);
20281 +               au_whtmp_rmdir_free(a->thargs);
20282 +               a->thargs = NULL;
20283 +       }
20284 +out:
20285 +       return err;
20286 +}
20287 +
20288 +/* ---------------------------------------------------------------------- */
20289 +
20290 +/*
20291 + * test if @dentry dir can be rename destination or not.
20292 + * success means, it is a logically empty dir.
20293 + */
20294 +static int may_rename_dstdir(struct dentry *dentry, struct au_nhash *whlist)
20295 +{
20296 +       return au_test_empty(dentry, whlist);
20297 +}
20298 +
20299 +/*
20300 + * test if @dentry dir can be rename source or not.
20301 + * if it can, return 0 and @children is filled.
20302 + * success means,
20303 + * - it is a logically empty dir.
20304 + * - or, it exists on writable branch and has no children including whiteouts
20305 + *       on the lower branch.
20306 + */
20307 +static int may_rename_srcdir(struct dentry *dentry, aufs_bindex_t btgt)
20308 +{
20309 +       int err;
20310 +       unsigned int rdhash;
20311 +       aufs_bindex_t bstart;
20312 +
20313 +       bstart = au_dbstart(dentry);
20314 +       if (bstart != btgt) {
20315 +               struct au_nhash whlist;
20316 +
20317 +               SiMustAnyLock(dentry->d_sb);
20318 +               rdhash = au_sbi(dentry->d_sb)->si_rdhash;
20319 +               if (!rdhash)
20320 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL,
20321 +                                                          dentry));
20322 +               err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
20323 +               if (unlikely(err))
20324 +                       goto out;
20325 +               err = au_test_empty(dentry, &whlist);
20326 +               au_nhash_wh_free(&whlist);
20327 +               goto out;
20328 +       }
20329 +
20330 +       if (bstart == au_dbtaildir(dentry))
20331 +               return 0; /* success */
20332 +
20333 +       err = au_test_empty_lower(dentry);
20334 +
20335 +out:
20336 +       if (err == -ENOTEMPTY) {
20337 +               AuWarn1("renaming dir who has child(ren) on multiple branches,"
20338 +                       " is not supported\n");
20339 +               err = -EXDEV;
20340 +       }
20341 +       return err;
20342 +}
20343 +
20344 +/* side effect: sets whlist and h_dentry */
20345 +static int au_ren_may_dir(struct au_ren_args *a)
20346 +{
20347 +       int err;
20348 +       unsigned int rdhash;
20349 +       struct dentry *d;
20350 +
20351 +       d = a->dst_dentry;
20352 +       SiMustAnyLock(d->d_sb);
20353 +
20354 +       err = 0;
20355 +       if (au_ftest_ren(a->flags, ISDIR) && a->dst_inode) {
20356 +               rdhash = au_sbi(d->d_sb)->si_rdhash;
20357 +               if (!rdhash)
20358 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, d));
20359 +               err = au_nhash_alloc(&a->whlist, rdhash, GFP_NOFS);
20360 +               if (unlikely(err))
20361 +                       goto out;
20362 +
20363 +               au_set_dbstart(d, a->dst_bstart);
20364 +               err = may_rename_dstdir(d, &a->whlist);
20365 +               au_set_dbstart(d, a->btgt);
20366 +       }
20367 +       a->dst_h_dentry = au_h_dptr(d, au_dbstart(d));
20368 +       if (unlikely(err))
20369 +               goto out;
20370 +
20371 +       d = a->src_dentry;
20372 +       a->src_h_dentry = au_h_dptr(d, au_dbstart(d));
20373 +       if (au_ftest_ren(a->flags, ISDIR)) {
20374 +               err = may_rename_srcdir(d, a->btgt);
20375 +               if (unlikely(err)) {
20376 +                       au_nhash_wh_free(&a->whlist);
20377 +                       a->whlist.nh_num = 0;
20378 +               }
20379 +       }
20380 +out:
20381 +       return err;
20382 +}
20383 +
20384 +/* ---------------------------------------------------------------------- */
20385 +
20386 +/*
20387 + * simple tests for rename.
20388 + * following the checks in vfs, plus the parent-child relationship.
20389 + */
20390 +static int au_may_ren(struct au_ren_args *a)
20391 +{
20392 +       int err, isdir;
20393 +       struct inode *h_inode;
20394 +
20395 +       if (a->src_bstart == a->btgt) {
20396 +               err = au_may_del(a->src_dentry, a->btgt, a->src_h_parent,
20397 +                                au_ftest_ren(a->flags, ISDIR));
20398 +               if (unlikely(err))
20399 +                       goto out;
20400 +               err = -EINVAL;
20401 +               if (unlikely(a->src_h_dentry == a->h_trap))
20402 +                       goto out;
20403 +       }
20404 +
20405 +       err = 0;
20406 +       if (a->dst_bstart != a->btgt)
20407 +               goto out;
20408 +
20409 +       err = -ENOTEMPTY;
20410 +       if (unlikely(a->dst_h_dentry == a->h_trap))
20411 +               goto out;
20412 +
20413 +       err = -EIO;
20414 +       h_inode = a->dst_h_dentry->d_inode;
20415 +       isdir = !!au_ftest_ren(a->flags, ISDIR);
20416 +       if (!a->dst_dentry->d_inode) {
20417 +               if (unlikely(h_inode))
20418 +                       goto out;
20419 +               err = au_may_add(a->dst_dentry, a->btgt, a->dst_h_parent,
20420 +                                isdir);
20421 +       } else {
20422 +               if (unlikely(!h_inode || !h_inode->i_nlink))
20423 +                       goto out;
20424 +               err = au_may_del(a->dst_dentry, a->btgt, a->dst_h_parent,
20425 +                                isdir);
20426 +               if (unlikely(err))
20427 +                       goto out;
20428 +       }
20429 +
20430 +out:
20431 +       if (unlikely(err == -ENOENT || err == -EEXIST))
20432 +               err = -EIO;
20433 +       AuTraceErr(err);
20434 +       return err;
20435 +}
20436 +
20437 +/* ---------------------------------------------------------------------- */
20438 +
20439 +/*
20440 + * locking order
20441 + * (VFS)
20442 + * - src_dir and dir by lock_rename()
20443 + * - inode if exitsts
20444 + * (aufs)
20445 + * - lock all
20446 + *   + src_dentry and dentry by aufs_read_and_write_lock2() which calls,
20447 + *     + si_read_lock
20448 + *     + di_write_lock2_child()
20449 + *       + di_write_lock_child()
20450 + *        + ii_write_lock_child()
20451 + *       + di_write_lock_child2()
20452 + *        + ii_write_lock_child2()
20453 + *     + src_parent and parent
20454 + *       + di_write_lock_parent()
20455 + *        + ii_write_lock_parent()
20456 + *       + di_write_lock_parent2()
20457 + *        + ii_write_lock_parent2()
20458 + *   + lower src_dir and dir by vfsub_lock_rename()
20459 + *   + verify the every relationships between child and parent. if any
20460 + *     of them failed, unlock all and return -EBUSY.
20461 + */
20462 +static void au_ren_unlock(struct au_ren_args *a)
20463 +{
20464 +       vfsub_unlock_rename(a->src_h_parent, a->src_hdir,
20465 +                           a->dst_h_parent, a->dst_hdir);
20466 +       if (au_ftest_ren(a->flags, MNT_WRITE))
20467 +               vfsub_mnt_drop_write(au_br_mnt(a->br));
20468 +}
20469 +
20470 +static int au_ren_lock(struct au_ren_args *a)
20471 +{
20472 +       int err;
20473 +       unsigned int udba;
20474 +
20475 +       err = 0;
20476 +       a->src_h_parent = au_h_dptr(a->src_parent, a->btgt);
20477 +       a->src_hdir = au_hi(a->src_dir, a->btgt);
20478 +       a->dst_h_parent = au_h_dptr(a->dst_parent, a->btgt);
20479 +       a->dst_hdir = au_hi(a->dst_dir, a->btgt);
20480 +
20481 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
20482 +       if (unlikely(err))
20483 +               goto out;
20484 +       au_fset_ren(a->flags, MNT_WRITE);
20485 +       a->h_trap = vfsub_lock_rename(a->src_h_parent, a->src_hdir,
20486 +                                     a->dst_h_parent, a->dst_hdir);
20487 +       udba = au_opt_udba(a->src_dentry->d_sb);
20488 +       if (unlikely(a->src_hdir->hi_inode != a->src_h_parent->d_inode
20489 +                    || a->dst_hdir->hi_inode != a->dst_h_parent->d_inode))
20490 +               err = au_busy_or_stale();
20491 +       if (!err && au_dbstart(a->src_dentry) == a->btgt)
20492 +               err = au_h_verify(a->src_h_dentry, udba,
20493 +                                 a->src_h_parent->d_inode, a->src_h_parent,
20494 +                                 a->br);
20495 +       if (!err && au_dbstart(a->dst_dentry) == a->btgt)
20496 +               err = au_h_verify(a->dst_h_dentry, udba,
20497 +                                 a->dst_h_parent->d_inode, a->dst_h_parent,
20498 +                                 a->br);
20499 +       if (!err)
20500 +               goto out; /* success */
20501 +
20502 +       err = au_busy_or_stale();
20503 +       au_ren_unlock(a);
20504 +
20505 +out:
20506 +       return err;
20507 +}
20508 +
20509 +/* ---------------------------------------------------------------------- */
20510 +
20511 +static void au_ren_refresh_dir(struct au_ren_args *a)
20512 +{
20513 +       struct inode *dir;
20514 +
20515 +       dir = a->dst_dir;
20516 +       dir->i_version++;
20517 +       if (au_ftest_ren(a->flags, ISDIR)) {
20518 +               /* is this updating defined in POSIX? */
20519 +               au_cpup_attr_timesizes(a->src_inode);
20520 +               au_cpup_attr_nlink(dir, /*force*/1);
20521 +       }
20522 +
20523 +       if (au_ibstart(dir) == a->btgt)
20524 +               au_cpup_attr_timesizes(dir);
20525 +
20526 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
20527 +               return;
20528 +
20529 +       dir = a->src_dir;
20530 +       dir->i_version++;
20531 +       if (au_ftest_ren(a->flags, ISDIR))
20532 +               au_cpup_attr_nlink(dir, /*force*/1);
20533 +       if (au_ibstart(dir) == a->btgt)
20534 +               au_cpup_attr_timesizes(dir);
20535 +}
20536 +
20537 +static void au_ren_refresh(struct au_ren_args *a)
20538 +{
20539 +       aufs_bindex_t bend, bindex;
20540 +       struct dentry *d, *h_d;
20541 +       struct inode *i, *h_i;
20542 +       struct super_block *sb;
20543 +
20544 +       d = a->dst_dentry;
20545 +       d_drop(d);
20546 +       if (a->h_dst)
20547 +               /* already dget-ed by au_ren_or_cpup() */
20548 +               au_set_h_dptr(d, a->btgt, a->h_dst);
20549 +
20550 +       i = a->dst_inode;
20551 +       if (i) {
20552 +               if (!au_ftest_ren(a->flags, ISDIR))
20553 +                       vfsub_drop_nlink(i);
20554 +               else {
20555 +                       vfsub_dead_dir(i);
20556 +                       au_cpup_attr_timesizes(i);
20557 +               }
20558 +               au_update_dbrange(d, /*do_put_zero*/1);
20559 +       } else {
20560 +               bend = a->btgt;
20561 +               for (bindex = au_dbstart(d); bindex < bend; bindex++)
20562 +                       au_set_h_dptr(d, bindex, NULL);
20563 +               bend = au_dbend(d);
20564 +               for (bindex = a->btgt + 1; bindex <= bend; bindex++)
20565 +                       au_set_h_dptr(d, bindex, NULL);
20566 +               au_update_dbrange(d, /*do_put_zero*/0);
20567 +       }
20568 +
20569 +       d = a->src_dentry;
20570 +       au_set_dbwh(d, -1);
20571 +       bend = au_dbend(d);
20572 +       for (bindex = a->btgt + 1; bindex <= bend; bindex++) {
20573 +               h_d = au_h_dptr(d, bindex);
20574 +               if (h_d)
20575 +                       au_set_h_dptr(d, bindex, NULL);
20576 +       }
20577 +       au_set_dbend(d, a->btgt);
20578 +
20579 +       sb = d->d_sb;
20580 +       i = a->src_inode;
20581 +       if (au_opt_test(au_mntflags(sb), PLINK) && au_plink_test(i))
20582 +               return; /* success */
20583 +
20584 +       bend = au_ibend(i);
20585 +       for (bindex = a->btgt + 1; bindex <= bend; bindex++) {
20586 +               h_i = au_h_iptr(i, bindex);
20587 +               if (h_i) {
20588 +                       au_xino_write(sb, bindex, h_i->i_ino, /*ino*/0);
20589 +                       /* ignore this error */
20590 +                       au_set_h_iptr(i, bindex, NULL, 0);
20591 +               }
20592 +       }
20593 +       au_set_ibend(i, a->btgt);
20594 +}
20595 +
20596 +/* ---------------------------------------------------------------------- */
20597 +
20598 +/* mainly for link(2) and rename(2) */
20599 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt)
20600 +{
20601 +       aufs_bindex_t bdiropq, bwh;
20602 +       struct dentry *parent;
20603 +       struct au_branch *br;
20604 +
20605 +       parent = dentry->d_parent;
20606 +       IMustLock(parent->d_inode); /* dir is locked */
20607 +
20608 +       bdiropq = au_dbdiropq(parent);
20609 +       bwh = au_dbwh(dentry);
20610 +       br = au_sbr(dentry->d_sb, btgt);
20611 +       if (au_br_rdonly(br)
20612 +           || (0 <= bdiropq && bdiropq < btgt)
20613 +           || (0 <= bwh && bwh < btgt))
20614 +               btgt = -1;
20615 +
20616 +       AuDbg("btgt %d\n", btgt);
20617 +       return btgt;
20618 +}
20619 +
20620 +/* sets src_bstart, dst_bstart and btgt */
20621 +static int au_ren_wbr(struct au_ren_args *a)
20622 +{
20623 +       int err;
20624 +       struct au_wr_dir_args wr_dir_args = {
20625 +               /* .force_btgt  = -1, */
20626 +               .flags          = AuWrDir_ADD_ENTRY
20627 +       };
20628 +
20629 +       a->src_bstart = au_dbstart(a->src_dentry);
20630 +       a->dst_bstart = au_dbstart(a->dst_dentry);
20631 +       if (au_ftest_ren(a->flags, ISDIR))
20632 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
20633 +       wr_dir_args.force_btgt = a->src_bstart;
20634 +       if (a->dst_inode && a->dst_bstart < a->src_bstart)
20635 +               wr_dir_args.force_btgt = a->dst_bstart;
20636 +       wr_dir_args.force_btgt = au_wbr(a->dst_dentry, wr_dir_args.force_btgt);
20637 +       err = au_wr_dir(a->dst_dentry, a->src_dentry, &wr_dir_args);
20638 +       a->btgt = err;
20639 +
20640 +       return err;
20641 +}
20642 +
20643 +static void au_ren_dt(struct au_ren_args *a)
20644 +{
20645 +       a->h_path.dentry = a->src_h_parent;
20646 +       au_dtime_store(a->src_dt + AuPARENT, a->src_parent, &a->h_path);
20647 +       if (!au_ftest_ren(a->flags, ISSAMEDIR)) {
20648 +               a->h_path.dentry = a->dst_h_parent;
20649 +               au_dtime_store(a->dst_dt + AuPARENT, a->dst_parent, &a->h_path);
20650 +       }
20651 +
20652 +       au_fclr_ren(a->flags, DT_DSTDIR);
20653 +       if (!au_ftest_ren(a->flags, ISDIR))
20654 +               return;
20655 +
20656 +       a->h_path.dentry = a->src_h_dentry;
20657 +       au_dtime_store(a->src_dt + AuCHILD, a->src_dentry, &a->h_path);
20658 +       if (a->dst_h_dentry->d_inode) {
20659 +               au_fset_ren(a->flags, DT_DSTDIR);
20660 +               a->h_path.dentry = a->dst_h_dentry;
20661 +               au_dtime_store(a->dst_dt + AuCHILD, a->dst_dentry, &a->h_path);
20662 +       }
20663 +}
20664 +
20665 +static void au_ren_rev_dt(int err, struct au_ren_args *a)
20666 +{
20667 +       struct dentry *h_d;
20668 +       struct mutex *h_mtx;
20669 +
20670 +       au_dtime_revert(a->src_dt + AuPARENT);
20671 +       if (!au_ftest_ren(a->flags, ISSAMEDIR))
20672 +               au_dtime_revert(a->dst_dt + AuPARENT);
20673 +
20674 +       if (au_ftest_ren(a->flags, ISDIR) && err != -EIO) {
20675 +               h_d = a->src_dt[AuCHILD].dt_h_path.dentry;
20676 +               h_mtx = &h_d->d_inode->i_mutex;
20677 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
20678 +               au_dtime_revert(a->src_dt + AuCHILD);
20679 +               mutex_unlock(h_mtx);
20680 +
20681 +               if (au_ftest_ren(a->flags, DT_DSTDIR)) {
20682 +                       h_d = a->dst_dt[AuCHILD].dt_h_path.dentry;
20683 +                       h_mtx = &h_d->d_inode->i_mutex;
20684 +                       mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
20685 +                       au_dtime_revert(a->dst_dt + AuCHILD);
20686 +                       mutex_unlock(h_mtx);
20687 +               }
20688 +       }
20689 +}
20690 +
20691 +/* ---------------------------------------------------------------------- */
20692 +
20693 +int aufs_rename(struct inode *_src_dir, struct dentry *_src_dentry,
20694 +               struct inode *_dst_dir, struct dentry *_dst_dentry)
20695 +{
20696 +       int err, flags;
20697 +       /* reduce stack space */
20698 +       struct au_ren_args *a;
20699 +
20700 +       AuDbg("%pd, %pd\n", _src_dentry, _dst_dentry);
20701 +       IMustLock(_src_dir);
20702 +       IMustLock(_dst_dir);
20703 +
20704 +       err = -ENOMEM;
20705 +       BUILD_BUG_ON(sizeof(*a) > PAGE_SIZE);
20706 +       a = kzalloc(sizeof(*a), GFP_NOFS);
20707 +       if (unlikely(!a))
20708 +               goto out;
20709 +
20710 +       a->src_dir = _src_dir;
20711 +       a->src_dentry = _src_dentry;
20712 +       a->src_inode = a->src_dentry->d_inode;
20713 +       a->src_parent = a->src_dentry->d_parent; /* dir inode is locked */
20714 +       a->dst_dir = _dst_dir;
20715 +       a->dst_dentry = _dst_dentry;
20716 +       a->dst_inode = a->dst_dentry->d_inode;
20717 +       a->dst_parent = a->dst_dentry->d_parent; /* dir inode is locked */
20718 +       if (a->dst_inode) {
20719 +               IMustLock(a->dst_inode);
20720 +               au_igrab(a->dst_inode);
20721 +       }
20722 +
20723 +       err = -ENOTDIR;
20724 +       flags = AuLock_FLUSH | AuLock_NOPLM | AuLock_GEN;
20725 +       if (d_is_dir(a->src_dentry)) {
20726 +               au_fset_ren(a->flags, ISDIR);
20727 +               if (unlikely(d_is_positive(a->dst_dentry)
20728 +                            && !d_is_dir(a->dst_dentry)))
20729 +                       goto out_free;
20730 +               err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry,
20731 +                                               AuLock_DIR | flags);
20732 +       } else
20733 +               err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry,
20734 +                                               flags);
20735 +       if (unlikely(err))
20736 +               goto out_free;
20737 +
20738 +       err = au_d_hashed_positive(a->src_dentry);
20739 +       if (unlikely(err))
20740 +               goto out_unlock;
20741 +       err = -ENOENT;
20742 +       if (a->dst_inode) {
20743 +               /*
20744 +                * If it is a dir, VFS unhash dst_dentry before this
20745 +                * function. It means we cannot rely upon d_unhashed().
20746 +                */
20747 +               if (unlikely(!a->dst_inode->i_nlink))
20748 +                       goto out_unlock;
20749 +               if (!S_ISDIR(a->dst_inode->i_mode)) {
20750 +                       err = au_d_hashed_positive(a->dst_dentry);
20751 +                       if (unlikely(err))
20752 +                               goto out_unlock;
20753 +               } else if (unlikely(IS_DEADDIR(a->dst_inode)))
20754 +                       goto out_unlock;
20755 +       } else if (unlikely(d_unhashed(a->dst_dentry)))
20756 +               goto out_unlock;
20757 +
20758 +       /*
20759 +        * is it possible?
20760 +        * yes, it happend (in linux-3.3-rcN) but I don't know why.
20761 +        * there may exist a problem somewhere else.
20762 +        */
20763 +       err = -EINVAL;
20764 +       if (unlikely(a->dst_parent->d_inode == a->src_dentry->d_inode))
20765 +               goto out_unlock;
20766 +
20767 +       au_fset_ren(a->flags, ISSAMEDIR); /* temporary */
20768 +       di_write_lock_parent(a->dst_parent);
20769 +
20770 +       /* which branch we process */
20771 +       err = au_ren_wbr(a);
20772 +       if (unlikely(err < 0))
20773 +               goto out_parent;
20774 +       a->br = au_sbr(a->dst_dentry->d_sb, a->btgt);
20775 +       a->h_path.mnt = au_br_mnt(a->br);
20776 +
20777 +       /* are they available to be renamed */
20778 +       err = au_ren_may_dir(a);
20779 +       if (unlikely(err))
20780 +               goto out_children;
20781 +
20782 +       /* prepare the writable parent dir on the same branch */
20783 +       if (a->dst_bstart == a->btgt) {
20784 +               au_fset_ren(a->flags, WHDST);
20785 +       } else {
20786 +               err = au_cpup_dirs(a->dst_dentry, a->btgt);
20787 +               if (unlikely(err))
20788 +                       goto out_children;
20789 +       }
20790 +
20791 +       if (a->src_dir != a->dst_dir) {
20792 +               /*
20793 +                * this temporary unlock is safe,
20794 +                * because both dir->i_mutex are locked.
20795 +                */
20796 +               di_write_unlock(a->dst_parent);
20797 +               di_write_lock_parent(a->src_parent);
20798 +               err = au_wr_dir_need_wh(a->src_dentry,
20799 +                                       au_ftest_ren(a->flags, ISDIR),
20800 +                                       &a->btgt);
20801 +               di_write_unlock(a->src_parent);
20802 +               di_write_lock2_parent(a->src_parent, a->dst_parent, /*isdir*/1);
20803 +               au_fclr_ren(a->flags, ISSAMEDIR);
20804 +       } else
20805 +               err = au_wr_dir_need_wh(a->src_dentry,
20806 +                                       au_ftest_ren(a->flags, ISDIR),
20807 +                                       &a->btgt);
20808 +       if (unlikely(err < 0))
20809 +               goto out_children;
20810 +       if (err)
20811 +               au_fset_ren(a->flags, WHSRC);
20812 +
20813 +       /* cpup src */
20814 +       if (a->src_bstart != a->btgt) {
20815 +               struct au_pin pin;
20816 +
20817 +               err = au_pin(&pin, a->src_dentry, a->btgt,
20818 +                            au_opt_udba(a->src_dentry->d_sb),
20819 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
20820 +               if (!err) {
20821 +                       struct au_cp_generic cpg = {
20822 +                               .dentry = a->src_dentry,
20823 +                               .bdst   = a->btgt,
20824 +                               .bsrc   = a->src_bstart,
20825 +                               .len    = -1,
20826 +                               .pin    = &pin,
20827 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
20828 +                       };
20829 +                       AuDebugOn(au_dbstart(a->src_dentry) != a->src_bstart);
20830 +                       err = au_sio_cpup_simple(&cpg);
20831 +                       au_unpin(&pin);
20832 +               }
20833 +               if (unlikely(err))
20834 +                       goto out_children;
20835 +               a->src_bstart = a->btgt;
20836 +               a->src_h_dentry = au_h_dptr(a->src_dentry, a->btgt);
20837 +               au_fset_ren(a->flags, WHSRC);
20838 +       }
20839 +
20840 +       /* lock them all */
20841 +       err = au_ren_lock(a);
20842 +       if (unlikely(err))
20843 +               /* leave the copied-up one */
20844 +               goto out_children;
20845 +
20846 +       if (!au_opt_test(au_mntflags(a->dst_dir->i_sb), UDBA_NONE))
20847 +               err = au_may_ren(a);
20848 +       else if (unlikely(a->dst_dentry->d_name.len > AUFS_MAX_NAMELEN))
20849 +               err = -ENAMETOOLONG;
20850 +       if (unlikely(err))
20851 +               goto out_hdir;
20852 +
20853 +       /* store timestamps to be revertible */
20854 +       au_ren_dt(a);
20855 +
20856 +       /* here we go */
20857 +       err = do_rename(a);
20858 +       if (unlikely(err))
20859 +               goto out_dt;
20860 +
20861 +       /* update dir attributes */
20862 +       au_ren_refresh_dir(a);
20863 +
20864 +       /* dput/iput all lower dentries */
20865 +       au_ren_refresh(a);
20866 +
20867 +       goto out_hdir; /* success */
20868 +
20869 +out_dt:
20870 +       au_ren_rev_dt(err, a);
20871 +out_hdir:
20872 +       au_ren_unlock(a);
20873 +out_children:
20874 +       au_nhash_wh_free(&a->whlist);
20875 +       if (err && a->dst_inode && a->dst_bstart != a->btgt) {
20876 +               AuDbg("bstart %d, btgt %d\n", a->dst_bstart, a->btgt);
20877 +               au_set_h_dptr(a->dst_dentry, a->btgt, NULL);
20878 +               au_set_dbstart(a->dst_dentry, a->dst_bstart);
20879 +       }
20880 +out_parent:
20881 +       if (!err)
20882 +               d_move(a->src_dentry, a->dst_dentry);
20883 +       else {
20884 +               au_update_dbstart(a->dst_dentry);
20885 +               if (!a->dst_inode)
20886 +                       d_drop(a->dst_dentry);
20887 +       }
20888 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
20889 +               di_write_unlock(a->dst_parent);
20890 +       else
20891 +               di_write_unlock2(a->src_parent, a->dst_parent);
20892 +out_unlock:
20893 +       aufs_read_and_write_unlock2(a->dst_dentry, a->src_dentry);
20894 +out_free:
20895 +       iput(a->dst_inode);
20896 +       if (a->thargs)
20897 +               au_whtmp_rmdir_free(a->thargs);
20898 +       kfree(a);
20899 +out:
20900 +       AuTraceErr(err);
20901 +       return err;
20902 +}
20903 diff -urN /usr/share/empty/fs/aufs/Kconfig linux/fs/aufs/Kconfig
20904 --- /usr/share/empty/fs/aufs/Kconfig    1970-01-01 01:00:00.000000000 +0100
20905 +++ linux/fs/aufs/Kconfig       2015-03-27 21:56:35.460128334 +0100
20906 @@ -0,0 +1,185 @@
20907 +config AUFS_FS
20908 +       tristate "Aufs (Advanced multi layered unification filesystem) support"
20909 +       help
20910 +       Aufs is a stackable unification filesystem such as Unionfs,
20911 +       which unifies several directories and provides a merged single
20912 +       directory.
20913 +       In the early days, aufs was entirely re-designed and
20914 +       re-implemented Unionfs Version 1.x series. Introducing many
20915 +       original ideas, approaches and improvements, it becomes totally
20916 +       different from Unionfs while keeping the basic features.
20917 +
20918 +if AUFS_FS
20919 +choice
20920 +       prompt "Maximum number of branches"
20921 +       default AUFS_BRANCH_MAX_127
20922 +       help
20923 +       Specifies the maximum number of branches (or member directories)
20924 +       in a single aufs. The larger value consumes more system
20925 +       resources and has a minor impact to performance.
20926 +config AUFS_BRANCH_MAX_127
20927 +       bool "127"
20928 +       help
20929 +       Specifies the maximum number of branches (or member directories)
20930 +       in a single aufs. The larger value consumes more system
20931 +       resources and has a minor impact to performance.
20932 +config AUFS_BRANCH_MAX_511
20933 +       bool "511"
20934 +       help
20935 +       Specifies the maximum number of branches (or member directories)
20936 +       in a single aufs. The larger value consumes more system
20937 +       resources and has a minor impact to performance.
20938 +config AUFS_BRANCH_MAX_1023
20939 +       bool "1023"
20940 +       help
20941 +       Specifies the maximum number of branches (or member directories)
20942 +       in a single aufs. The larger value consumes more system
20943 +       resources and has a minor impact to performance.
20944 +config AUFS_BRANCH_MAX_32767
20945 +       bool "32767"
20946 +       help
20947 +       Specifies the maximum number of branches (or member directories)
20948 +       in a single aufs. The larger value consumes more system
20949 +       resources and has a minor impact to performance.
20950 +endchoice
20951 +
20952 +config AUFS_SBILIST
20953 +       bool
20954 +       depends on AUFS_MAGIC_SYSRQ || PROC_FS
20955 +       default y
20956 +       help
20957 +       Automatic configuration for internal use.
20958 +       When aufs supports Magic SysRq or /proc, enabled automatically.
20959 +
20960 +config AUFS_HNOTIFY
20961 +       bool "Detect direct branch access (bypassing aufs)"
20962 +       help
20963 +       If you want to modify files on branches directly, eg. bypassing aufs,
20964 +       and want aufs to detect the changes of them fully, then enable this
20965 +       option and use 'udba=notify' mount option.
20966 +       Currently there is only one available configuration, "fsnotify".
20967 +       It will have a negative impact to the performance.
20968 +       See detail in aufs.5.
20969 +
20970 +choice
20971 +       prompt "method" if AUFS_HNOTIFY
20972 +       default AUFS_HFSNOTIFY
20973 +config AUFS_HFSNOTIFY
20974 +       bool "fsnotify"
20975 +       select FSNOTIFY
20976 +endchoice
20977 +
20978 +config AUFS_EXPORT
20979 +       bool "NFS-exportable aufs"
20980 +       depends on EXPORTFS
20981 +       help
20982 +       If you want to export your mounted aufs via NFS, then enable this
20983 +       option. There are several requirements for this configuration.
20984 +       See detail in aufs.5.
20985 +
20986 +config AUFS_INO_T_64
20987 +       bool
20988 +       depends on AUFS_EXPORT
20989 +       depends on 64BIT && !(ALPHA || S390)
20990 +       default y
20991 +       help
20992 +       Automatic configuration for internal use.
20993 +       /* typedef unsigned long/int __kernel_ino_t */
20994 +       /* alpha and s390x are int */
20995 +
20996 +config AUFS_XATTR
20997 +       bool "support for XATTR/EA (including Security Labels)"
20998 +       help
20999 +       If your branch fs supports XATTR/EA and you want to make them
21000 +       available in aufs too, then enable this opsion and specify the
21001 +       branch attributes for EA.
21002 +       See detail in aufs.5.
21003 +
21004 +config AUFS_FHSM
21005 +       bool "File-based Hierarchical Storage Management"
21006 +       help
21007 +       Hierarchical Storage Management (or HSM) is a well-known feature
21008 +       in the storage world. Aufs provides this feature as file-based.
21009 +       with multiple branches.
21010 +       These multiple branches are prioritized, ie. the topmost one
21011 +       should be the fastest drive and be used heavily.
21012 +
21013 +config AUFS_RDU
21014 +       bool "Readdir in userspace"
21015 +       help
21016 +       Aufs has two methods to provide a merged view for a directory,
21017 +       by a user-space library and by kernel-space natively. The latter
21018 +       is always enabled but sometimes large and slow.
21019 +       If you enable this option, install the library in aufs2-util
21020 +       package, and set some environment variables for your readdir(3),
21021 +       then the work will be handled in user-space which generally
21022 +       shows better performance in most cases.
21023 +       See detail in aufs.5.
21024 +
21025 +config AUFS_SHWH
21026 +       bool "Show whiteouts"
21027 +       help
21028 +       If you want to make the whiteouts in aufs visible, then enable
21029 +       this option and specify 'shwh' mount option. Although it may
21030 +       sounds like philosophy or something, but in technically it
21031 +       simply shows the name of whiteout with keeping its behaviour.
21032 +
21033 +config AUFS_BR_RAMFS
21034 +       bool "Ramfs (initramfs/rootfs) as an aufs branch"
21035 +       help
21036 +       If you want to use ramfs as an aufs branch fs, then enable this
21037 +       option. Generally tmpfs is recommended.
21038 +       Aufs prohibited them to be a branch fs by default, because
21039 +       initramfs becomes unusable after switch_root or something
21040 +       generally. If you sets initramfs as an aufs branch and boot your
21041 +       system by switch_root, you will meet a problem easily since the
21042 +       files in initramfs may be inaccessible.
21043 +       Unless you are going to use ramfs as an aufs branch fs without
21044 +       switch_root or something, leave it N.
21045 +
21046 +config AUFS_BR_FUSE
21047 +       bool "Fuse fs as an aufs branch"
21048 +       depends on FUSE_FS
21049 +       select AUFS_POLL
21050 +       help
21051 +       If you want to use fuse-based userspace filesystem as an aufs
21052 +       branch fs, then enable this option.
21053 +       It implements the internal poll(2) operation which is
21054 +       implemented by fuse only (curretnly).
21055 +
21056 +config AUFS_POLL
21057 +       bool
21058 +       help
21059 +       Automatic configuration for internal use.
21060 +
21061 +config AUFS_BR_HFSPLUS
21062 +       bool "Hfsplus as an aufs branch"
21063 +       depends on HFSPLUS_FS
21064 +       default y
21065 +       help
21066 +       If you want to use hfsplus fs as an aufs branch fs, then enable
21067 +       this option. This option introduces a small overhead at
21068 +       copying-up a file on hfsplus.
21069 +
21070 +config AUFS_BDEV_LOOP
21071 +       bool
21072 +       depends on BLK_DEV_LOOP
21073 +       default y
21074 +       help
21075 +       Automatic configuration for internal use.
21076 +       Convert =[ym] into =y.
21077 +
21078 +config AUFS_DEBUG
21079 +       bool "Debug aufs"
21080 +       help
21081 +       Enable this to compile aufs internal debug code.
21082 +       It will have a negative impact to the performance.
21083 +
21084 +config AUFS_MAGIC_SYSRQ
21085 +       bool
21086 +       depends on AUFS_DEBUG && MAGIC_SYSRQ
21087 +       default y
21088 +       help
21089 +       Automatic configuration for internal use.
21090 +       When aufs supports Magic SysRq, enabled automatically.
21091 +endif
21092 diff -urN /usr/share/empty/fs/aufs/loop.c linux/fs/aufs/loop.c
21093 --- /usr/share/empty/fs/aufs/loop.c     1970-01-01 01:00:00.000000000 +0100
21094 +++ linux/fs/aufs/loop.c        2015-03-27 21:56:35.466795000 +0100
21095 @@ -0,0 +1,145 @@
21096 +/*
21097 + * Copyright (C) 2005-2015 Junjiro R. Okajima
21098 + *
21099 + * This program, aufs is free software; you can redistribute it and/or modify
21100 + * it under the terms of the GNU General Public License as published by
21101 + * the Free Software Foundation; either version 2 of the License, or
21102 + * (at your option) any later version.
21103 + *
21104 + * This program is distributed in the hope that it will be useful,
21105 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21106 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21107 + * GNU General Public License for more details.
21108 + *
21109 + * You should have received a copy of the GNU General Public License
21110 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21111 + */
21112 +
21113 +/*
21114 + * support for loopback block device as a branch
21115 + */
21116 +
21117 +#include "aufs.h"
21118 +
21119 +/* added into drivers/block/loop.c */
21120 +static struct file *(*backing_file_func)(struct super_block *sb);
21121 +
21122 +/*
21123 + * test if two lower dentries have overlapping branches.
21124 + */
21125 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding)
21126 +{
21127 +       struct super_block *h_sb;
21128 +       struct file *backing_file;
21129 +
21130 +       if (unlikely(!backing_file_func)) {
21131 +               /* don't load "loop" module here */
21132 +               backing_file_func = symbol_get(loop_backing_file);
21133 +               if (unlikely(!backing_file_func))
21134 +                       /* "loop" module is not loaded */
21135 +                       return 0;
21136 +       }
21137 +
21138 +       h_sb = h_adding->d_sb;
21139 +       backing_file = backing_file_func(h_sb);
21140 +       if (!backing_file)
21141 +               return 0;
21142 +
21143 +       h_adding = backing_file->f_path.dentry;
21144 +       /*
21145 +        * h_adding can be local NFS.
21146 +        * in this case aufs cannot detect the loop.
21147 +        */
21148 +       if (unlikely(h_adding->d_sb == sb))
21149 +               return 1;
21150 +       return !!au_test_subdir(h_adding, sb->s_root);
21151 +}
21152 +
21153 +/* true if a kernel thread named 'loop[0-9].*' accesses a file */
21154 +int au_test_loopback_kthread(void)
21155 +{
21156 +       int ret;
21157 +       struct task_struct *tsk = current;
21158 +       char c, comm[sizeof(tsk->comm)];
21159 +
21160 +       ret = 0;
21161 +       if (tsk->flags & PF_KTHREAD) {
21162 +               get_task_comm(comm, tsk);
21163 +               c = comm[4];
21164 +               ret = ('0' <= c && c <= '9'
21165 +                      && !strncmp(comm, "loop", 4));
21166 +       }
21167 +
21168 +       return ret;
21169 +}
21170 +
21171 +/* ---------------------------------------------------------------------- */
21172 +
21173 +#define au_warn_loopback_step  16
21174 +static int au_warn_loopback_nelem = au_warn_loopback_step;
21175 +static unsigned long *au_warn_loopback_array;
21176 +
21177 +void au_warn_loopback(struct super_block *h_sb)
21178 +{
21179 +       int i, new_nelem;
21180 +       unsigned long *a, magic;
21181 +       static DEFINE_SPINLOCK(spin);
21182 +
21183 +       magic = h_sb->s_magic;
21184 +       spin_lock(&spin);
21185 +       a = au_warn_loopback_array;
21186 +       for (i = 0; i < au_warn_loopback_nelem && *a; i++)
21187 +               if (a[i] == magic) {
21188 +                       spin_unlock(&spin);
21189 +                       return;
21190 +               }
21191 +
21192 +       /* h_sb is new to us, print it */
21193 +       if (i < au_warn_loopback_nelem) {
21194 +               a[i] = magic;
21195 +               goto pr;
21196 +       }
21197 +
21198 +       /* expand the array */
21199 +       new_nelem = au_warn_loopback_nelem + au_warn_loopback_step;
21200 +       a = au_kzrealloc(au_warn_loopback_array,
21201 +                        au_warn_loopback_nelem * sizeof(unsigned long),
21202 +                        new_nelem * sizeof(unsigned long), GFP_ATOMIC);
21203 +       if (a) {
21204 +               au_warn_loopback_nelem = new_nelem;
21205 +               au_warn_loopback_array = a;
21206 +               a[i] = magic;
21207 +               goto pr;
21208 +       }
21209 +
21210 +       spin_unlock(&spin);
21211 +       AuWarn1("realloc failed, ignored\n");
21212 +       return;
21213 +
21214 +pr:
21215 +       spin_unlock(&spin);
21216 +       pr_warn("you may want to try another patch for loopback file "
21217 +               "on %s(0x%lx) branch\n", au_sbtype(h_sb), magic);
21218 +}
21219 +
21220 +int au_loopback_init(void)
21221 +{
21222 +       int err;
21223 +       struct super_block *sb __maybe_unused;
21224 +
21225 +       AuDebugOn(sizeof(sb->s_magic) != sizeof(unsigned long));
21226 +
21227 +       err = 0;
21228 +       au_warn_loopback_array = kcalloc(au_warn_loopback_step,
21229 +                                        sizeof(unsigned long), GFP_NOFS);
21230 +       if (unlikely(!au_warn_loopback_array))
21231 +               err = -ENOMEM;
21232 +
21233 +       return err;
21234 +}
21235 +
21236 +void au_loopback_fin(void)
21237 +{
21238 +       symbol_put(loop_backing_file);
21239 +       kfree(au_warn_loopback_array);
21240 +}
21241 diff -urN /usr/share/empty/fs/aufs/loop.h linux/fs/aufs/loop.h
21242 --- /usr/share/empty/fs/aufs/loop.h     1970-01-01 01:00:00.000000000 +0100
21243 +++ linux/fs/aufs/loop.h        2015-03-27 21:56:35.466795000 +0100
21244 @@ -0,0 +1,52 @@
21245 +/*
21246 + * Copyright (C) 2005-2015 Junjiro R. Okajima
21247 + *
21248 + * This program, aufs is free software; you can redistribute it and/or modify
21249 + * it under the terms of the GNU General Public License as published by
21250 + * the Free Software Foundation; either version 2 of the License, or
21251 + * (at your option) any later version.
21252 + *
21253 + * This program is distributed in the hope that it will be useful,
21254 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21255 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21256 + * GNU General Public License for more details.
21257 + *
21258 + * You should have received a copy of the GNU General Public License
21259 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21260 + */
21261 +
21262 +/*
21263 + * support for loopback mount as a branch
21264 + */
21265 +
21266 +#ifndef __AUFS_LOOP_H__
21267 +#define __AUFS_LOOP_H__
21268 +
21269 +#ifdef __KERNEL__
21270 +
21271 +struct dentry;
21272 +struct super_block;
21273 +
21274 +#ifdef CONFIG_AUFS_BDEV_LOOP
21275 +/* drivers/block/loop.c */
21276 +struct file *loop_backing_file(struct super_block *sb);
21277 +
21278 +/* loop.c */
21279 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding);
21280 +int au_test_loopback_kthread(void);
21281 +void au_warn_loopback(struct super_block *h_sb);
21282 +
21283 +int au_loopback_init(void);
21284 +void au_loopback_fin(void);
21285 +#else
21286 +AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
21287 +          struct dentry *h_adding)
21288 +AuStubInt0(au_test_loopback_kthread, void)
21289 +AuStubVoid(au_warn_loopback, struct super_block *h_sb)
21290 +
21291 +AuStubInt0(au_loopback_init, void)
21292 +AuStubVoid(au_loopback_fin, void)
21293 +#endif /* BLK_DEV_LOOP */
21294 +
21295 +#endif /* __KERNEL__ */
21296 +#endif /* __AUFS_LOOP_H__ */
21297 diff -urN /usr/share/empty/fs/aufs/magic.mk linux/fs/aufs/magic.mk
21298 --- /usr/share/empty/fs/aufs/magic.mk   1970-01-01 01:00:00.000000000 +0100
21299 +++ linux/fs/aufs/magic.mk      2015-03-27 21:56:35.466795000 +0100
21300 @@ -0,0 +1,54 @@
21301 +
21302 +# defined in ${srctree}/fs/fuse/inode.c
21303 +# tristate
21304 +ifdef CONFIG_FUSE_FS
21305 +ccflags-y += -DFUSE_SUPER_MAGIC=0x65735546
21306 +endif
21307 +
21308 +# defined in ${srctree}/fs/ocfs2/ocfs2_fs.h
21309 +# tristate
21310 +ifdef CONFIG_OCFS2_FS
21311 +ccflags-y += -DOCFS2_SUPER_MAGIC=0x7461636f
21312 +endif
21313 +
21314 +# defined in ${srctree}/fs/ocfs2/dlm/userdlm.h
21315 +# tristate
21316 +ifdef CONFIG_OCFS2_FS_O2CB
21317 +ccflags-y += -DDLMFS_MAGIC=0x76a9f425
21318 +endif
21319 +
21320 +# defined in ${srctree}/fs/cifs/cifsfs.c
21321 +# tristate
21322 +ifdef CONFIG_CIFS_FS
21323 +ccflags-y += -DCIFS_MAGIC_NUMBER=0xFF534D42
21324 +endif
21325 +
21326 +# defined in ${srctree}/fs/xfs/xfs_sb.h
21327 +# tristate
21328 +ifdef CONFIG_XFS_FS
21329 +ccflags-y += -DXFS_SB_MAGIC=0x58465342
21330 +endif
21331 +
21332 +# defined in ${srctree}/fs/configfs/mount.c
21333 +# tristate
21334 +ifdef CONFIG_CONFIGFS_FS
21335 +ccflags-y += -DCONFIGFS_MAGIC=0x62656570
21336 +endif
21337 +
21338 +# defined in ${srctree}/fs/9p/v9fs.h
21339 +# tristate
21340 +ifdef CONFIG_9P_FS
21341 +ccflags-y += -DV9FS_MAGIC=0x01021997
21342 +endif
21343 +
21344 +# defined in ${srctree}/fs/ubifs/ubifs.h
21345 +# tristate
21346 +ifdef CONFIG_UBIFS_FS
21347 +ccflags-y += -DUBIFS_SUPER_MAGIC=0x24051905
21348 +endif
21349 +
21350 +# defined in ${srctree}/fs/hfsplus/hfsplus_raw.h
21351 +# tristate
21352 +ifdef CONFIG_HFSPLUS_FS
21353 +ccflags-y += -DHFSPLUS_SUPER_MAGIC=0x482b
21354 +endif
21355 diff -urN /usr/share/empty/fs/aufs/Makefile linux/fs/aufs/Makefile
21356 --- /usr/share/empty/fs/aufs/Makefile   1970-01-01 01:00:00.000000000 +0100
21357 +++ linux/fs/aufs/Makefile      2015-03-27 21:56:35.460128334 +0100
21358 @@ -0,0 +1,44 @@
21359 +
21360 +include ${src}/magic.mk
21361 +ifeq (${CONFIG_AUFS_FS},m)
21362 +include ${src}/conf.mk
21363 +endif
21364 +-include ${src}/priv_def.mk
21365 +
21366 +# cf. include/linux/kernel.h
21367 +# enable pr_debug
21368 +ccflags-y += -DDEBUG
21369 +# sparse requires the full pathname
21370 +ifdef M
21371 +ccflags-y += -include ${M}/../../include/uapi/linux/aufs_type.h
21372 +else
21373 +ccflags-y += -include ${srctree}/include/uapi/linux/aufs_type.h
21374 +endif
21375 +
21376 +obj-$(CONFIG_AUFS_FS) += aufs.o
21377 +aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
21378 +       wkq.o vfsub.o dcsub.o \
21379 +       cpup.o whout.o wbr_policy.o \
21380 +       dinfo.o dentry.o \
21381 +       dynop.o \
21382 +       finfo.o file.o f_op.o \
21383 +       dir.o vdir.o \
21384 +       iinfo.o inode.o i_op.o i_op_add.o i_op_del.o i_op_ren.o \
21385 +       mvdown.o ioctl.o
21386 +
21387 +# all are boolean
21388 +aufs-$(CONFIG_PROC_FS) += procfs.o plink.o
21389 +aufs-$(CONFIG_SYSFS) += sysfs.o
21390 +aufs-$(CONFIG_DEBUG_FS) += dbgaufs.o
21391 +aufs-$(CONFIG_AUFS_BDEV_LOOP) += loop.o
21392 +aufs-$(CONFIG_AUFS_HNOTIFY) += hnotify.o
21393 +aufs-$(CONFIG_AUFS_HFSNOTIFY) += hfsnotify.o
21394 +aufs-$(CONFIG_AUFS_EXPORT) += export.o
21395 +aufs-$(CONFIG_AUFS_XATTR) += xattr.o
21396 +aufs-$(CONFIG_FS_POSIX_ACL) += posix_acl.o
21397 +aufs-$(CONFIG_AUFS_FHSM) += fhsm.o
21398 +aufs-$(CONFIG_AUFS_POLL) += poll.o
21399 +aufs-$(CONFIG_AUFS_RDU) += rdu.o
21400 +aufs-$(CONFIG_AUFS_BR_HFSPLUS) += hfsplus.o
21401 +aufs-$(CONFIG_AUFS_DEBUG) += debug.o
21402 +aufs-$(CONFIG_AUFS_MAGIC_SYSRQ) += sysrq.o
21403 diff -urN /usr/share/empty/fs/aufs/module.c linux/fs/aufs/module.c
21404 --- /usr/share/empty/fs/aufs/module.c   1970-01-01 01:00:00.000000000 +0100
21405 +++ linux/fs/aufs/module.c      2015-03-27 21:56:35.466795000 +0100
21406 @@ -0,0 +1,210 @@
21407 +/*
21408 + * Copyright (C) 2005-2015 Junjiro R. Okajima
21409 + *
21410 + * This program, aufs is free software; you can redistribute it and/or modify
21411 + * it under the terms of the GNU General Public License as published by
21412 + * the Free Software Foundation; either version 2 of the License, or
21413 + * (at your option) any later version.
21414 + *
21415 + * This program is distributed in the hope that it will be useful,
21416 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21417 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21418 + * GNU General Public License for more details.
21419 + *
21420 + * You should have received a copy of the GNU General Public License
21421 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21422 + */
21423 +
21424 +/*
21425 + * module global variables and operations
21426 + */
21427 +
21428 +#include <linux/module.h>
21429 +#include <linux/seq_file.h>
21430 +#include "aufs.h"
21431 +
21432 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp)
21433 +{
21434 +       if (new_sz <= nused)
21435 +               return p;
21436 +
21437 +       p = krealloc(p, new_sz, gfp);
21438 +       if (p)
21439 +               memset(p + nused, 0, new_sz - nused);
21440 +       return p;
21441 +}
21442 +
21443 +/* ---------------------------------------------------------------------- */
21444 +
21445 +/*
21446 + * aufs caches
21447 + */
21448 +struct kmem_cache *au_cachep[AuCache_Last];
21449 +static int __init au_cache_init(void)
21450 +{
21451 +       au_cachep[AuCache_DINFO] = AuCacheCtor(au_dinfo, au_di_init_once);
21452 +       if (au_cachep[AuCache_DINFO])
21453 +               /* SLAB_DESTROY_BY_RCU */
21454 +               au_cachep[AuCache_ICNTNR] = AuCacheCtor(au_icntnr,
21455 +                                                       au_icntnr_init_once);
21456 +       if (au_cachep[AuCache_ICNTNR])
21457 +               au_cachep[AuCache_FINFO] = AuCacheCtor(au_finfo,
21458 +                                                      au_fi_init_once);
21459 +       if (au_cachep[AuCache_FINFO])
21460 +               au_cachep[AuCache_VDIR] = AuCache(au_vdir);
21461 +       if (au_cachep[AuCache_VDIR])
21462 +               au_cachep[AuCache_DEHSTR] = AuCache(au_vdir_dehstr);
21463 +       if (au_cachep[AuCache_DEHSTR])
21464 +               return 0;
21465 +
21466 +       return -ENOMEM;
21467 +}
21468 +
21469 +static void au_cache_fin(void)
21470 +{
21471 +       int i;
21472 +
21473 +       /*
21474 +        * Make sure all delayed rcu free inodes are flushed before we
21475 +        * destroy cache.
21476 +        */
21477 +       rcu_barrier();
21478 +
21479 +       /* excluding AuCache_HNOTIFY */
21480 +       BUILD_BUG_ON(AuCache_HNOTIFY + 1 != AuCache_Last);
21481 +       for (i = 0; i < AuCache_HNOTIFY; i++)
21482 +               if (au_cachep[i]) {
21483 +                       kmem_cache_destroy(au_cachep[i]);
21484 +                       au_cachep[i] = NULL;
21485 +               }
21486 +}
21487 +
21488 +/* ---------------------------------------------------------------------- */
21489 +
21490 +int au_dir_roflags;
21491 +
21492 +#ifdef CONFIG_AUFS_SBILIST
21493 +/*
21494 + * iterate_supers_type() doesn't protect us from
21495 + * remounting (branch management)
21496 + */
21497 +struct au_splhead au_sbilist;
21498 +#endif
21499 +
21500 +struct lock_class_key au_lc_key[AuLcKey_Last];
21501 +
21502 +/*
21503 + * functions for module interface.
21504 + */
21505 +MODULE_LICENSE("GPL");
21506 +/* MODULE_LICENSE("GPL v2"); */
21507 +MODULE_AUTHOR("Junjiro R. Okajima <aufs-users@lists.sourceforge.net>");
21508 +MODULE_DESCRIPTION(AUFS_NAME
21509 +       " -- Advanced multi layered unification filesystem");
21510 +MODULE_VERSION(AUFS_VERSION);
21511 +MODULE_ALIAS_FS(AUFS_NAME);
21512 +
21513 +/* this module parameter has no meaning when SYSFS is disabled */
21514 +int sysaufs_brs = 1;
21515 +MODULE_PARM_DESC(brs, "use <sysfs>/fs/aufs/si_*/brN");
21516 +module_param_named(brs, sysaufs_brs, int, S_IRUGO);
21517 +
21518 +/* this module parameter has no meaning when USER_NS is disabled */
21519 +static bool au_userns;
21520 +MODULE_PARM_DESC(allow_userns, "allow unprivileged to mount under userns");
21521 +module_param_named(allow_userns, au_userns, bool, S_IRUGO);
21522 +
21523 +/* ---------------------------------------------------------------------- */
21524 +
21525 +static char au_esc_chars[0x20 + 3]; /* 0x01-0x20, backslash, del, and NULL */
21526 +
21527 +int au_seq_path(struct seq_file *seq, struct path *path)
21528 +{
21529 +       return seq_path(seq, path, au_esc_chars);
21530 +}
21531 +
21532 +/* ---------------------------------------------------------------------- */
21533 +
21534 +static int __init aufs_init(void)
21535 +{
21536 +       int err, i;
21537 +       char *p;
21538 +
21539 +       p = au_esc_chars;
21540 +       for (i = 1; i <= ' '; i++)
21541 +               *p++ = i;
21542 +       *p++ = '\\';
21543 +       *p++ = '\x7f';
21544 +       *p = 0;
21545 +
21546 +       au_dir_roflags = au_file_roflags(O_DIRECTORY | O_LARGEFILE);
21547 +
21548 +       au_sbilist_init();
21549 +       sysaufs_brs_init();
21550 +       au_debug_init();
21551 +       au_dy_init();
21552 +       err = sysaufs_init();
21553 +       if (unlikely(err))
21554 +               goto out;
21555 +       err = au_procfs_init();
21556 +       if (unlikely(err))
21557 +               goto out_sysaufs;
21558 +       err = au_wkq_init();
21559 +       if (unlikely(err))
21560 +               goto out_procfs;
21561 +       err = au_loopback_init();
21562 +       if (unlikely(err))
21563 +               goto out_wkq;
21564 +       err = au_hnotify_init();
21565 +       if (unlikely(err))
21566 +               goto out_loopback;
21567 +       err = au_sysrq_init();
21568 +       if (unlikely(err))
21569 +               goto out_hin;
21570 +       err = au_cache_init();
21571 +       if (unlikely(err))
21572 +               goto out_sysrq;
21573 +
21574 +       aufs_fs_type.fs_flags |= au_userns ? FS_USERNS_MOUNT : 0;
21575 +       err = register_filesystem(&aufs_fs_type);
21576 +       if (unlikely(err))
21577 +               goto out_cache;
21578 +
21579 +       /* since we define pr_fmt, call printk directly */
21580 +       printk(KERN_INFO AUFS_NAME " " AUFS_VERSION "\n");
21581 +       goto out; /* success */
21582 +
21583 +out_cache:
21584 +       au_cache_fin();
21585 +out_sysrq:
21586 +       au_sysrq_fin();
21587 +out_hin:
21588 +       au_hnotify_fin();
21589 +out_loopback:
21590 +       au_loopback_fin();
21591 +out_wkq:
21592 +       au_wkq_fin();
21593 +out_procfs:
21594 +       au_procfs_fin();
21595 +out_sysaufs:
21596 +       sysaufs_fin();
21597 +       au_dy_fin();
21598 +out:
21599 +       return err;
21600 +}
21601 +
21602 +static void __exit aufs_exit(void)
21603 +{
21604 +       unregister_filesystem(&aufs_fs_type);
21605 +       au_cache_fin();
21606 +       au_sysrq_fin();
21607 +       au_hnotify_fin();
21608 +       au_loopback_fin();
21609 +       au_wkq_fin();
21610 +       au_procfs_fin();
21611 +       sysaufs_fin();
21612 +       au_dy_fin();
21613 +}
21614 +
21615 +module_init(aufs_init);
21616 +module_exit(aufs_exit);
21617 diff -urN /usr/share/empty/fs/aufs/module.h linux/fs/aufs/module.h
21618 --- /usr/share/empty/fs/aufs/module.h   1970-01-01 01:00:00.000000000 +0100
21619 +++ linux/fs/aufs/module.h      2015-03-27 21:56:35.466795000 +0100
21620 @@ -0,0 +1,104 @@
21621 +/*
21622 + * Copyright (C) 2005-2015 Junjiro R. Okajima
21623 + *
21624 + * This program, aufs is free software; you can redistribute it and/or modify
21625 + * it under the terms of the GNU General Public License as published by
21626 + * the Free Software Foundation; either version 2 of the License, or
21627 + * (at your option) any later version.
21628 + *
21629 + * This program is distributed in the hope that it will be useful,
21630 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21631 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21632 + * GNU General Public License for more details.
21633 + *
21634 + * You should have received a copy of the GNU General Public License
21635 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21636 + */
21637 +
21638 +/*
21639 + * module initialization and module-global
21640 + */
21641 +
21642 +#ifndef __AUFS_MODULE_H__
21643 +#define __AUFS_MODULE_H__
21644 +
21645 +#ifdef __KERNEL__
21646 +
21647 +#include <linux/slab.h>
21648 +
21649 +struct path;
21650 +struct seq_file;
21651 +
21652 +/* module parameters */
21653 +extern int sysaufs_brs;
21654 +
21655 +/* ---------------------------------------------------------------------- */
21656 +
21657 +extern int au_dir_roflags;
21658 +
21659 +enum {
21660 +       AuLcNonDir_FIINFO,
21661 +       AuLcNonDir_DIINFO,
21662 +       AuLcNonDir_IIINFO,
21663 +
21664 +       AuLcDir_FIINFO,
21665 +       AuLcDir_DIINFO,
21666 +       AuLcDir_IIINFO,
21667 +
21668 +       AuLcSymlink_DIINFO,
21669 +       AuLcSymlink_IIINFO,
21670 +
21671 +       AuLcKey_Last
21672 +};
21673 +extern struct lock_class_key au_lc_key[AuLcKey_Last];
21674 +
21675 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp);
21676 +int au_seq_path(struct seq_file *seq, struct path *path);
21677 +
21678 +#ifdef CONFIG_PROC_FS
21679 +/* procfs.c */
21680 +int __init au_procfs_init(void);
21681 +void au_procfs_fin(void);
21682 +#else
21683 +AuStubInt0(au_procfs_init, void);
21684 +AuStubVoid(au_procfs_fin, void);
21685 +#endif
21686 +
21687 +/* ---------------------------------------------------------------------- */
21688 +
21689 +/* kmem cache */
21690 +enum {
21691 +       AuCache_DINFO,
21692 +       AuCache_ICNTNR,
21693 +       AuCache_FINFO,
21694 +       AuCache_VDIR,
21695 +       AuCache_DEHSTR,
21696 +       AuCache_HNOTIFY, /* must be last */
21697 +       AuCache_Last
21698 +};
21699 +
21700 +#define AuCacheFlags           (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD)
21701 +#define AuCache(type)          KMEM_CACHE(type, AuCacheFlags)
21702 +#define AuCacheCtor(type, ctor)        \
21703 +       kmem_cache_create(#type, sizeof(struct type), \
21704 +                         __alignof__(struct type), AuCacheFlags, ctor)
21705 +
21706 +extern struct kmem_cache *au_cachep[];
21707 +
21708 +#define AuCacheFuncs(name, index) \
21709 +static inline struct au_##name *au_cache_alloc_##name(void) \
21710 +{ return kmem_cache_alloc(au_cachep[AuCache_##index], GFP_NOFS); } \
21711 +static inline void au_cache_free_##name(struct au_##name *p) \
21712 +{ kmem_cache_free(au_cachep[AuCache_##index], p); }
21713 +
21714 +AuCacheFuncs(dinfo, DINFO);
21715 +AuCacheFuncs(icntnr, ICNTNR);
21716 +AuCacheFuncs(finfo, FINFO);
21717 +AuCacheFuncs(vdir, VDIR);
21718 +AuCacheFuncs(vdir_dehstr, DEHSTR);
21719 +#ifdef CONFIG_AUFS_HNOTIFY
21720 +AuCacheFuncs(hnotify, HNOTIFY);
21721 +#endif
21722 +
21723 +#endif /* __KERNEL__ */
21724 +#endif /* __AUFS_MODULE_H__ */
21725 diff -urN /usr/share/empty/fs/aufs/mvdown.c linux/fs/aufs/mvdown.c
21726 --- /usr/share/empty/fs/aufs/mvdown.c   1970-01-01 01:00:00.000000000 +0100
21727 +++ linux/fs/aufs/mvdown.c      2015-03-27 21:56:35.466795000 +0100
21728 @@ -0,0 +1,694 @@
21729 +/*
21730 + * Copyright (C) 2011-2015 Junjiro R. Okajima
21731 + *
21732 + * This program, aufs is free software; you can redistribute it and/or modify
21733 + * it under the terms of the GNU General Public License as published by
21734 + * the Free Software Foundation; either version 2 of the License, or
21735 + * (at your option) any later version.
21736 + *
21737 + * This program is distributed in the hope that it will be useful,
21738 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21739 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21740 + * GNU General Public License for more details.
21741 + *
21742 + * You should have received a copy of the GNU General Public License
21743 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21744 + */
21745 +
21746 +/*
21747 + * move-down, opposite of copy-up
21748 + */
21749 +
21750 +#include "aufs.h"
21751 +
21752 +struct au_mvd_args {
21753 +       struct {
21754 +               struct super_block *h_sb;
21755 +               struct dentry *h_parent;
21756 +               struct au_hinode *hdir;
21757 +               struct inode *h_dir, *h_inode;
21758 +               struct au_pin pin;
21759 +       } info[AUFS_MVDOWN_NARRAY];
21760 +
21761 +       struct aufs_mvdown mvdown;
21762 +       struct dentry *dentry, *parent;
21763 +       struct inode *inode, *dir;
21764 +       struct super_block *sb;
21765 +       aufs_bindex_t bopq, bwh, bfound;
21766 +       unsigned char rename_lock;
21767 +};
21768 +
21769 +#define mvd_errno              mvdown.au_errno
21770 +#define mvd_bsrc               mvdown.stbr[AUFS_MVDOWN_UPPER].bindex
21771 +#define mvd_src_brid           mvdown.stbr[AUFS_MVDOWN_UPPER].brid
21772 +#define mvd_bdst               mvdown.stbr[AUFS_MVDOWN_LOWER].bindex
21773 +#define mvd_dst_brid           mvdown.stbr[AUFS_MVDOWN_LOWER].brid
21774 +
21775 +#define mvd_h_src_sb           info[AUFS_MVDOWN_UPPER].h_sb
21776 +#define mvd_h_src_parent       info[AUFS_MVDOWN_UPPER].h_parent
21777 +#define mvd_hdir_src           info[AUFS_MVDOWN_UPPER].hdir
21778 +#define mvd_h_src_dir          info[AUFS_MVDOWN_UPPER].h_dir
21779 +#define mvd_h_src_inode                info[AUFS_MVDOWN_UPPER].h_inode
21780 +#define mvd_pin_src            info[AUFS_MVDOWN_UPPER].pin
21781 +
21782 +#define mvd_h_dst_sb           info[AUFS_MVDOWN_LOWER].h_sb
21783 +#define mvd_h_dst_parent       info[AUFS_MVDOWN_LOWER].h_parent
21784 +#define mvd_hdir_dst           info[AUFS_MVDOWN_LOWER].hdir
21785 +#define mvd_h_dst_dir          info[AUFS_MVDOWN_LOWER].h_dir
21786 +#define mvd_h_dst_inode                info[AUFS_MVDOWN_LOWER].h_inode
21787 +#define mvd_pin_dst            info[AUFS_MVDOWN_LOWER].pin
21788 +
21789 +#define AU_MVD_PR(flag, ...) do {                      \
21790 +               if (flag)                               \
21791 +                       pr_err(__VA_ARGS__);            \
21792 +       } while (0)
21793 +
21794 +static int find_lower_writable(struct au_mvd_args *a)
21795 +{
21796 +       struct super_block *sb;
21797 +       aufs_bindex_t bindex, bend;
21798 +       struct au_branch *br;
21799 +
21800 +       sb = a->sb;
21801 +       bindex = a->mvd_bsrc;
21802 +       bend = au_sbend(sb);
21803 +       if (a->mvdown.flags & AUFS_MVDOWN_FHSM_LOWER)
21804 +               for (bindex++; bindex <= bend; bindex++) {
21805 +                       br = au_sbr(sb, bindex);
21806 +                       if (au_br_fhsm(br->br_perm)
21807 +                           && (!(au_br_sb(br)->s_flags & MS_RDONLY)))
21808 +                               return bindex;
21809 +               }
21810 +       else if (!(a->mvdown.flags & AUFS_MVDOWN_ROLOWER))
21811 +               for (bindex++; bindex <= bend; bindex++) {
21812 +                       br = au_sbr(sb, bindex);
21813 +                       if (!au_br_rdonly(br))
21814 +                               return bindex;
21815 +               }
21816 +       else
21817 +               for (bindex++; bindex <= bend; bindex++) {
21818 +                       br = au_sbr(sb, bindex);
21819 +                       if (!(au_br_sb(br)->s_flags & MS_RDONLY)) {
21820 +                               if (au_br_rdonly(br))
21821 +                                       a->mvdown.flags
21822 +                                               |= AUFS_MVDOWN_ROLOWER_R;
21823 +                               return bindex;
21824 +                       }
21825 +               }
21826 +
21827 +       return -1;
21828 +}
21829 +
21830 +/* make the parent dir on bdst */
21831 +static int au_do_mkdir(const unsigned char dmsg, struct au_mvd_args *a)
21832 +{
21833 +       int err;
21834 +
21835 +       err = 0;
21836 +       a->mvd_hdir_src = au_hi(a->dir, a->mvd_bsrc);
21837 +       a->mvd_hdir_dst = au_hi(a->dir, a->mvd_bdst);
21838 +       a->mvd_h_src_parent = au_h_dptr(a->parent, a->mvd_bsrc);
21839 +       a->mvd_h_dst_parent = NULL;
21840 +       if (au_dbend(a->parent) >= a->mvd_bdst)
21841 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
21842 +       if (!a->mvd_h_dst_parent) {
21843 +               err = au_cpdown_dirs(a->dentry, a->mvd_bdst);
21844 +               if (unlikely(err)) {
21845 +                       AU_MVD_PR(dmsg, "cpdown_dirs failed\n");
21846 +                       goto out;
21847 +               }
21848 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
21849 +       }
21850 +
21851 +out:
21852 +       AuTraceErr(err);
21853 +       return err;
21854 +}
21855 +
21856 +/* lock them all */
21857 +static int au_do_lock(const unsigned char dmsg, struct au_mvd_args *a)
21858 +{
21859 +       int err;
21860 +       struct dentry *h_trap;
21861 +
21862 +       a->mvd_h_src_sb = au_sbr_sb(a->sb, a->mvd_bsrc);
21863 +       a->mvd_h_dst_sb = au_sbr_sb(a->sb, a->mvd_bdst);
21864 +       err = au_pin(&a->mvd_pin_dst, a->dentry, a->mvd_bdst,
21865 +                    au_opt_udba(a->sb),
21866 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
21867 +       AuTraceErr(err);
21868 +       if (unlikely(err)) {
21869 +               AU_MVD_PR(dmsg, "pin_dst failed\n");
21870 +               goto out;
21871 +       }
21872 +
21873 +       if (a->mvd_h_src_sb != a->mvd_h_dst_sb) {
21874 +               a->rename_lock = 0;
21875 +               au_pin_init(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
21876 +                           AuLsc_DI_PARENT, AuLsc_I_PARENT3,
21877 +                           au_opt_udba(a->sb),
21878 +                           AuPin_MNT_WRITE | AuPin_DI_LOCKED);
21879 +               err = au_do_pin(&a->mvd_pin_src);
21880 +               AuTraceErr(err);
21881 +               a->mvd_h_src_dir = a->mvd_h_src_parent->d_inode;
21882 +               if (unlikely(err)) {
21883 +                       AU_MVD_PR(dmsg, "pin_src failed\n");
21884 +                       goto out_dst;
21885 +               }
21886 +               goto out; /* success */
21887 +       }
21888 +
21889 +       a->rename_lock = 1;
21890 +       au_pin_hdir_unlock(&a->mvd_pin_dst);
21891 +       err = au_pin(&a->mvd_pin_src, a->dentry, a->mvd_bsrc,
21892 +                    au_opt_udba(a->sb),
21893 +                    AuPin_MNT_WRITE | AuPin_DI_LOCKED);
21894 +       AuTraceErr(err);
21895 +       a->mvd_h_src_dir = a->mvd_h_src_parent->d_inode;
21896 +       if (unlikely(err)) {
21897 +               AU_MVD_PR(dmsg, "pin_src failed\n");
21898 +               au_pin_hdir_lock(&a->mvd_pin_dst);
21899 +               goto out_dst;
21900 +       }
21901 +       au_pin_hdir_unlock(&a->mvd_pin_src);
21902 +       h_trap = vfsub_lock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
21903 +                                  a->mvd_h_dst_parent, a->mvd_hdir_dst);
21904 +       if (h_trap) {
21905 +               err = (h_trap != a->mvd_h_src_parent);
21906 +               if (err)
21907 +                       err = (h_trap != a->mvd_h_dst_parent);
21908 +       }
21909 +       BUG_ON(err); /* it should never happen */
21910 +       if (unlikely(a->mvd_h_src_dir != au_pinned_h_dir(&a->mvd_pin_src))) {
21911 +               err = -EBUSY;
21912 +               AuTraceErr(err);
21913 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
21914 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
21915 +               au_pin_hdir_lock(&a->mvd_pin_src);
21916 +               au_unpin(&a->mvd_pin_src);
21917 +               au_pin_hdir_lock(&a->mvd_pin_dst);
21918 +               goto out_dst;
21919 +       }
21920 +       goto out; /* success */
21921 +
21922 +out_dst:
21923 +       au_unpin(&a->mvd_pin_dst);
21924 +out:
21925 +       AuTraceErr(err);
21926 +       return err;
21927 +}
21928 +
21929 +static void au_do_unlock(const unsigned char dmsg, struct au_mvd_args *a)
21930 +{
21931 +       if (!a->rename_lock)
21932 +               au_unpin(&a->mvd_pin_src);
21933 +       else {
21934 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
21935 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
21936 +               au_pin_hdir_lock(&a->mvd_pin_src);
21937 +               au_unpin(&a->mvd_pin_src);
21938 +               au_pin_hdir_lock(&a->mvd_pin_dst);
21939 +       }
21940 +       au_unpin(&a->mvd_pin_dst);
21941 +}
21942 +
21943 +/* copy-down the file */
21944 +static int au_do_cpdown(const unsigned char dmsg, struct au_mvd_args *a)
21945 +{
21946 +       int err;
21947 +       struct au_cp_generic cpg = {
21948 +               .dentry = a->dentry,
21949 +               .bdst   = a->mvd_bdst,
21950 +               .bsrc   = a->mvd_bsrc,
21951 +               .len    = -1,
21952 +               .pin    = &a->mvd_pin_dst,
21953 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
21954 +       };
21955 +
21956 +       AuDbg("b%d, b%d\n", cpg.bsrc, cpg.bdst);
21957 +       if (a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
21958 +               au_fset_cpup(cpg.flags, OVERWRITE);
21959 +       if (a->mvdown.flags & AUFS_MVDOWN_ROLOWER)
21960 +               au_fset_cpup(cpg.flags, RWDST);
21961 +       err = au_sio_cpdown_simple(&cpg);
21962 +       if (unlikely(err))
21963 +               AU_MVD_PR(dmsg, "cpdown failed\n");
21964 +
21965 +       AuTraceErr(err);
21966 +       return err;
21967 +}
21968 +
21969 +/*
21970 + * unlink the whiteout on bdst if exist which may be created by UDBA while we
21971 + * were sleeping
21972 + */
21973 +static int au_do_unlink_wh(const unsigned char dmsg, struct au_mvd_args *a)
21974 +{
21975 +       int err;
21976 +       struct path h_path;
21977 +       struct au_branch *br;
21978 +       struct inode *delegated;
21979 +
21980 +       br = au_sbr(a->sb, a->mvd_bdst);
21981 +       h_path.dentry = au_wh_lkup(a->mvd_h_dst_parent, &a->dentry->d_name, br);
21982 +       err = PTR_ERR(h_path.dentry);
21983 +       if (IS_ERR(h_path.dentry)) {
21984 +               AU_MVD_PR(dmsg, "wh_lkup failed\n");
21985 +               goto out;
21986 +       }
21987 +
21988 +       err = 0;
21989 +       if (h_path.dentry->d_inode) {
21990 +               h_path.mnt = au_br_mnt(br);
21991 +               delegated = NULL;
21992 +               err = vfsub_unlink(a->mvd_h_dst_parent->d_inode, &h_path,
21993 +                                  &delegated, /*force*/0);
21994 +               if (unlikely(err == -EWOULDBLOCK)) {
21995 +                       pr_warn("cannot retry for NFSv4 delegation"
21996 +                               " for an internal unlink\n");
21997 +                       iput(delegated);
21998 +               }
21999 +               if (unlikely(err))
22000 +                       AU_MVD_PR(dmsg, "wh_unlink failed\n");
22001 +       }
22002 +       dput(h_path.dentry);
22003 +
22004 +out:
22005 +       AuTraceErr(err);
22006 +       return err;
22007 +}
22008 +
22009 +/*
22010 + * unlink the topmost h_dentry
22011 + */
22012 +static int au_do_unlink(const unsigned char dmsg, struct au_mvd_args *a)
22013 +{
22014 +       int err;
22015 +       struct path h_path;
22016 +       struct inode *delegated;
22017 +
22018 +       h_path.mnt = au_sbr_mnt(a->sb, a->mvd_bsrc);
22019 +       h_path.dentry = au_h_dptr(a->dentry, a->mvd_bsrc);
22020 +       delegated = NULL;
22021 +       err = vfsub_unlink(a->mvd_h_src_dir, &h_path, &delegated, /*force*/0);
22022 +       if (unlikely(err == -EWOULDBLOCK)) {
22023 +               pr_warn("cannot retry for NFSv4 delegation"
22024 +                       " for an internal unlink\n");
22025 +               iput(delegated);
22026 +       }
22027 +       if (unlikely(err))
22028 +               AU_MVD_PR(dmsg, "unlink failed\n");
22029 +
22030 +       AuTraceErr(err);
22031 +       return err;
22032 +}
22033 +
22034 +/* Since mvdown succeeded, we ignore an error of this function */
22035 +static void au_do_stfs(const unsigned char dmsg, struct au_mvd_args *a)
22036 +{
22037 +       int err;
22038 +       struct au_branch *br;
22039 +
22040 +       a->mvdown.flags |= AUFS_MVDOWN_STFS_FAILED;
22041 +       br = au_sbr(a->sb, a->mvd_bsrc);
22042 +       err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_UPPER].stfs);
22043 +       if (!err) {
22044 +               br = au_sbr(a->sb, a->mvd_bdst);
22045 +               a->mvdown.stbr[AUFS_MVDOWN_LOWER].brid = br->br_id;
22046 +               err = au_br_stfs(br, &a->mvdown.stbr[AUFS_MVDOWN_LOWER].stfs);
22047 +       }
22048 +       if (!err)
22049 +               a->mvdown.flags &= ~AUFS_MVDOWN_STFS_FAILED;
22050 +       else
22051 +               AU_MVD_PR(dmsg, "statfs failed (%d), ignored\n", err);
22052 +}
22053 +
22054 +/*
22055 + * copy-down the file and unlink the bsrc file.
22056 + * - unlink the bdst whout if exist
22057 + * - copy-down the file (with whtmp name and rename)
22058 + * - unlink the bsrc file
22059 + */
22060 +static int au_do_mvdown(const unsigned char dmsg, struct au_mvd_args *a)
22061 +{
22062 +       int err;
22063 +
22064 +       err = au_do_mkdir(dmsg, a);
22065 +       if (!err)
22066 +               err = au_do_lock(dmsg, a);
22067 +       if (unlikely(err))
22068 +               goto out;
22069 +
22070 +       /*
22071 +        * do not revert the activities we made on bdst since they should be
22072 +        * harmless in aufs.
22073 +        */
22074 +
22075 +       err = au_do_cpdown(dmsg, a);
22076 +       if (!err)
22077 +               err = au_do_unlink_wh(dmsg, a);
22078 +       if (!err && !(a->mvdown.flags & AUFS_MVDOWN_KUPPER))
22079 +               err = au_do_unlink(dmsg, a);
22080 +       if (unlikely(err))
22081 +               goto out_unlock;
22082 +
22083 +       AuDbg("%pd2, 0x%x, %d --> %d\n",
22084 +             a->dentry, a->mvdown.flags, a->mvd_bsrc, a->mvd_bdst);
22085 +       if (find_lower_writable(a) < 0)
22086 +               a->mvdown.flags |= AUFS_MVDOWN_BOTTOM;
22087 +
22088 +       if (a->mvdown.flags & AUFS_MVDOWN_STFS)
22089 +               au_do_stfs(dmsg, a);
22090 +
22091 +       /* maintain internal array */
22092 +       if (!(a->mvdown.flags & AUFS_MVDOWN_KUPPER)) {
22093 +               au_set_h_dptr(a->dentry, a->mvd_bsrc, NULL);
22094 +               au_set_dbstart(a->dentry, a->mvd_bdst);
22095 +               au_set_h_iptr(a->inode, a->mvd_bsrc, NULL, /*flags*/0);
22096 +               au_set_ibstart(a->inode, a->mvd_bdst);
22097 +       }
22098 +       if (au_dbend(a->dentry) < a->mvd_bdst)
22099 +               au_set_dbend(a->dentry, a->mvd_bdst);
22100 +       if (au_ibend(a->inode) < a->mvd_bdst)
22101 +               au_set_ibend(a->inode, a->mvd_bdst);
22102 +
22103 +out_unlock:
22104 +       au_do_unlock(dmsg, a);
22105 +out:
22106 +       AuTraceErr(err);
22107 +       return err;
22108 +}
22109 +
22110 +/* ---------------------------------------------------------------------- */
22111 +
22112 +/* make sure the file is idle */
22113 +static int au_mvd_args_busy(const unsigned char dmsg, struct au_mvd_args *a)
22114 +{
22115 +       int err, plinked;
22116 +
22117 +       err = 0;
22118 +       plinked = !!au_opt_test(au_mntflags(a->sb), PLINK);
22119 +       if (au_dbstart(a->dentry) == a->mvd_bsrc
22120 +           && au_dcount(a->dentry) == 1
22121 +           && atomic_read(&a->inode->i_count) == 1
22122 +           /* && a->mvd_h_src_inode->i_nlink == 1 */
22123 +           && (!plinked || !au_plink_test(a->inode))
22124 +           && a->inode->i_nlink == 1)
22125 +               goto out;
22126 +
22127 +       err = -EBUSY;
22128 +       AU_MVD_PR(dmsg,
22129 +                 "b%d, d{b%d, c%d?}, i{c%d?, l%u}, hi{l%u}, p{%d, %d}\n",
22130 +                 a->mvd_bsrc, au_dbstart(a->dentry), au_dcount(a->dentry),
22131 +                 atomic_read(&a->inode->i_count), a->inode->i_nlink,
22132 +                 a->mvd_h_src_inode->i_nlink,
22133 +                 plinked, plinked ? au_plink_test(a->inode) : 0);
22134 +
22135 +out:
22136 +       AuTraceErr(err);
22137 +       return err;
22138 +}
22139 +
22140 +/* make sure the parent dir is fine */
22141 +static int au_mvd_args_parent(const unsigned char dmsg,
22142 +                             struct au_mvd_args *a)
22143 +{
22144 +       int err;
22145 +       aufs_bindex_t bindex;
22146 +
22147 +       err = 0;
22148 +       if (unlikely(au_alive_dir(a->parent))) {
22149 +               err = -ENOENT;
22150 +               AU_MVD_PR(dmsg, "parent dir is dead\n");
22151 +               goto out;
22152 +       }
22153 +
22154 +       a->bopq = au_dbdiropq(a->parent);
22155 +       bindex = au_wbr_nonopq(a->dentry, a->mvd_bdst);
22156 +       AuDbg("b%d\n", bindex);
22157 +       if (unlikely((bindex >= 0 && bindex < a->mvd_bdst)
22158 +                    || (a->bopq != -1 && a->bopq < a->mvd_bdst))) {
22159 +               err = -EINVAL;
22160 +               a->mvd_errno = EAU_MVDOWN_OPAQUE;
22161 +               AU_MVD_PR(dmsg, "ancestor is opaque b%d, b%d\n",
22162 +                         a->bopq, a->mvd_bdst);
22163 +       }
22164 +
22165 +out:
22166 +       AuTraceErr(err);
22167 +       return err;
22168 +}
22169 +
22170 +static int au_mvd_args_intermediate(const unsigned char dmsg,
22171 +                                   struct au_mvd_args *a)
22172 +{
22173 +       int err;
22174 +       struct au_dinfo *dinfo, *tmp;
22175 +
22176 +       /* lookup the next lower positive entry */
22177 +       err = -ENOMEM;
22178 +       tmp = au_di_alloc(a->sb, AuLsc_DI_TMP);
22179 +       if (unlikely(!tmp))
22180 +               goto out;
22181 +
22182 +       a->bfound = -1;
22183 +       a->bwh = -1;
22184 +       dinfo = au_di(a->dentry);
22185 +       au_di_cp(tmp, dinfo);
22186 +       au_di_swap(tmp, dinfo);
22187 +
22188 +       /* returns the number of positive dentries */
22189 +       err = au_lkup_dentry(a->dentry, a->mvd_bsrc + 1, /*type*/0);
22190 +       if (!err)
22191 +               a->bwh = au_dbwh(a->dentry);
22192 +       else if (err > 0)
22193 +               a->bfound = au_dbstart(a->dentry);
22194 +
22195 +       au_di_swap(tmp, dinfo);
22196 +       au_rw_write_unlock(&tmp->di_rwsem);
22197 +       au_di_free(tmp);
22198 +       if (unlikely(err < 0))
22199 +               AU_MVD_PR(dmsg, "failed look-up lower\n");
22200 +
22201 +       /*
22202 +        * here, we have these cases.
22203 +        * bfound == -1
22204 +        *      no positive dentry under bsrc. there are more sub-cases.
22205 +        *      bwh < 0
22206 +        *              there no whiteout, we can safely move-down.
22207 +        *      bwh <= bsrc
22208 +        *              impossible
22209 +        *      bsrc < bwh && bwh < bdst
22210 +        *              there is a whiteout on RO branch. cannot proceed.
22211 +        *      bwh == bdst
22212 +        *              there is a whiteout on the RW target branch. it should
22213 +        *              be removed.
22214 +        *      bdst < bwh
22215 +        *              there is a whiteout somewhere unrelated branch.
22216 +        * -1 < bfound && bfound <= bsrc
22217 +        *      impossible.
22218 +        * bfound < bdst
22219 +        *      found, but it is on RO branch between bsrc and bdst. cannot
22220 +        *      proceed.
22221 +        * bfound == bdst
22222 +        *      found, replace it if AUFS_MVDOWN_FORCE is set. otherwise return
22223 +        *      error.
22224 +        * bdst < bfound
22225 +        *      found, after we create the file on bdst, it will be hidden.
22226 +        */
22227 +
22228 +       AuDebugOn(a->bfound == -1
22229 +                 && a->bwh != -1
22230 +                 && a->bwh <= a->mvd_bsrc);
22231 +       AuDebugOn(-1 < a->bfound
22232 +                 && a->bfound <= a->mvd_bsrc);
22233 +
22234 +       err = -EINVAL;
22235 +       if (a->bfound == -1
22236 +           && a->mvd_bsrc < a->bwh
22237 +           && a->bwh != -1
22238 +           && a->bwh < a->mvd_bdst) {
22239 +               a->mvd_errno = EAU_MVDOWN_WHITEOUT;
22240 +               AU_MVD_PR(dmsg, "bsrc %d, bdst %d, bfound %d, bwh %d\n",
22241 +                         a->mvd_bsrc, a->mvd_bdst, a->bfound, a->bwh);
22242 +               goto out;
22243 +       } else if (a->bfound != -1 && a->bfound < a->mvd_bdst) {
22244 +               a->mvd_errno = EAU_MVDOWN_UPPER;
22245 +               AU_MVD_PR(dmsg, "bdst %d, bfound %d\n",
22246 +                         a->mvd_bdst, a->bfound);
22247 +               goto out;
22248 +       }
22249 +
22250 +       err = 0; /* success */
22251 +
22252 +out:
22253 +       AuTraceErr(err);
22254 +       return err;
22255 +}
22256 +
22257 +static int au_mvd_args_exist(const unsigned char dmsg, struct au_mvd_args *a)
22258 +{
22259 +       int err;
22260 +
22261 +       err = 0;
22262 +       if (!(a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
22263 +           && a->bfound == a->mvd_bdst)
22264 +               err = -EEXIST;
22265 +       AuTraceErr(err);
22266 +       return err;
22267 +}
22268 +
22269 +static int au_mvd_args(const unsigned char dmsg, struct au_mvd_args *a)
22270 +{
22271 +       int err;
22272 +       struct au_branch *br;
22273 +
22274 +       err = -EISDIR;
22275 +       if (unlikely(S_ISDIR(a->inode->i_mode)))
22276 +               goto out;
22277 +
22278 +       err = -EINVAL;
22279 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_UPPER))
22280 +               a->mvd_bsrc = au_ibstart(a->inode);
22281 +       else {
22282 +               a->mvd_bsrc = au_br_index(a->sb, a->mvd_src_brid);
22283 +               if (unlikely(a->mvd_bsrc < 0
22284 +                            || (a->mvd_bsrc < au_dbstart(a->dentry)
22285 +                                || au_dbend(a->dentry) < a->mvd_bsrc
22286 +                                || !au_h_dptr(a->dentry, a->mvd_bsrc))
22287 +                            || (a->mvd_bsrc < au_ibstart(a->inode)
22288 +                                || au_ibend(a->inode) < a->mvd_bsrc
22289 +                                || !au_h_iptr(a->inode, a->mvd_bsrc)))) {
22290 +                       a->mvd_errno = EAU_MVDOWN_NOUPPER;
22291 +                       AU_MVD_PR(dmsg, "no upper\n");
22292 +                       goto out;
22293 +               }
22294 +       }
22295 +       if (unlikely(a->mvd_bsrc == au_sbend(a->sb))) {
22296 +               a->mvd_errno = EAU_MVDOWN_BOTTOM;
22297 +               AU_MVD_PR(dmsg, "on the bottom\n");
22298 +               goto out;
22299 +       }
22300 +       a->mvd_h_src_inode = au_h_iptr(a->inode, a->mvd_bsrc);
22301 +       br = au_sbr(a->sb, a->mvd_bsrc);
22302 +       err = au_br_rdonly(br);
22303 +       if (!(a->mvdown.flags & AUFS_MVDOWN_ROUPPER)) {
22304 +               if (unlikely(err))
22305 +                       goto out;
22306 +       } else if (!(vfsub_native_ro(a->mvd_h_src_inode)
22307 +                    || IS_APPEND(a->mvd_h_src_inode))) {
22308 +               if (err)
22309 +                       a->mvdown.flags |= AUFS_MVDOWN_ROUPPER_R;
22310 +               /* go on */
22311 +       } else
22312 +               goto out;
22313 +
22314 +       err = -EINVAL;
22315 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_LOWER)) {
22316 +               a->mvd_bdst = find_lower_writable(a);
22317 +               if (unlikely(a->mvd_bdst < 0)) {
22318 +                       a->mvd_errno = EAU_MVDOWN_BOTTOM;
22319 +                       AU_MVD_PR(dmsg, "no writable lower branch\n");
22320 +                       goto out;
22321 +               }
22322 +       } else {
22323 +               a->mvd_bdst = au_br_index(a->sb, a->mvd_dst_brid);
22324 +               if (unlikely(a->mvd_bdst < 0
22325 +                            || au_sbend(a->sb) < a->mvd_bdst)) {
22326 +                       a->mvd_errno = EAU_MVDOWN_NOLOWERBR;
22327 +                       AU_MVD_PR(dmsg, "no lower brid\n");
22328 +                       goto out;
22329 +               }
22330 +       }
22331 +
22332 +       err = au_mvd_args_busy(dmsg, a);
22333 +       if (!err)
22334 +               err = au_mvd_args_parent(dmsg, a);
22335 +       if (!err)
22336 +               err = au_mvd_args_intermediate(dmsg, a);
22337 +       if (!err)
22338 +               err = au_mvd_args_exist(dmsg, a);
22339 +       if (!err)
22340 +               AuDbg("b%d, b%d\n", a->mvd_bsrc, a->mvd_bdst);
22341 +
22342 +out:
22343 +       AuTraceErr(err);
22344 +       return err;
22345 +}
22346 +
22347 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *uarg)
22348 +{
22349 +       int err, e;
22350 +       unsigned char dmsg;
22351 +       struct au_mvd_args *args;
22352 +
22353 +       err = -EPERM;
22354 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
22355 +               goto out;
22356 +
22357 +       err = -ENOMEM;
22358 +       args = kmalloc(sizeof(*args), GFP_NOFS);
22359 +       if (unlikely(!args))
22360 +               goto out;
22361 +
22362 +       err = copy_from_user(&args->mvdown, uarg, sizeof(args->mvdown));
22363 +       if (!err)
22364 +               err = !access_ok(VERIFY_WRITE, uarg, sizeof(*uarg));
22365 +       if (unlikely(err)) {
22366 +               err = -EFAULT;
22367 +               AuTraceErr(err);
22368 +               goto out_free;
22369 +       }
22370 +       AuDbg("flags 0x%x\n", args->mvdown.flags);
22371 +       args->mvdown.flags &= ~(AUFS_MVDOWN_ROLOWER_R | AUFS_MVDOWN_ROUPPER_R);
22372 +       args->mvdown.au_errno = 0;
22373 +       args->dentry = dentry;
22374 +       args->inode = dentry->d_inode;
22375 +       args->sb = dentry->d_sb;
22376 +
22377 +       err = -ENOENT;
22378 +       dmsg = !!(args->mvdown.flags & AUFS_MVDOWN_DMSG);
22379 +       args->parent = dget_parent(dentry);
22380 +       args->dir = args->parent->d_inode;
22381 +       mutex_lock_nested(&args->dir->i_mutex, I_MUTEX_PARENT);
22382 +       dput(args->parent);
22383 +       if (unlikely(args->parent != dentry->d_parent)) {
22384 +               AU_MVD_PR(dmsg, "parent dir is moved\n");
22385 +               goto out_dir;
22386 +       }
22387 +
22388 +       mutex_lock_nested(&args->inode->i_mutex, I_MUTEX_CHILD);
22389 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH);
22390 +       if (unlikely(err))
22391 +               goto out_inode;
22392 +
22393 +       di_write_lock_parent(args->parent);
22394 +       err = au_mvd_args(dmsg, args);
22395 +       if (unlikely(err))
22396 +               goto out_parent;
22397 +
22398 +       err = au_do_mvdown(dmsg, args);
22399 +       if (unlikely(err))
22400 +               goto out_parent;
22401 +
22402 +       au_cpup_attr_timesizes(args->dir);
22403 +       au_cpup_attr_timesizes(args->inode);
22404 +       au_cpup_igen(args->inode, au_h_iptr(args->inode, args->mvd_bdst));
22405 +       /* au_digen_dec(dentry); */
22406 +
22407 +out_parent:
22408 +       di_write_unlock(args->parent);
22409 +       aufs_read_unlock(dentry, AuLock_DW);
22410 +out_inode:
22411 +       mutex_unlock(&args->inode->i_mutex);
22412 +out_dir:
22413 +       mutex_unlock(&args->dir->i_mutex);
22414 +out_free:
22415 +       e = copy_to_user(uarg, &args->mvdown, sizeof(args->mvdown));
22416 +       if (unlikely(e))
22417 +               err = -EFAULT;
22418 +       kfree(args);
22419 +out:
22420 +       AuTraceErr(err);
22421 +       return err;
22422 +}
22423 diff -urN /usr/share/empty/fs/aufs/opts.c linux/fs/aufs/opts.c
22424 --- /usr/share/empty/fs/aufs/opts.c     1970-01-01 01:00:00.000000000 +0100
22425 +++ linux/fs/aufs/opts.c        2015-03-27 21:56:35.466795000 +0100
22426 @@ -0,0 +1,1859 @@
22427 +/*
22428 + * Copyright (C) 2005-2015 Junjiro R. Okajima
22429 + *
22430 + * This program, aufs is free software; you can redistribute it and/or modify
22431 + * it under the terms of the GNU General Public License as published by
22432 + * the Free Software Foundation; either version 2 of the License, or
22433 + * (at your option) any later version.
22434 + *
22435 + * This program is distributed in the hope that it will be useful,
22436 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22437 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22438 + * GNU General Public License for more details.
22439 + *
22440 + * You should have received a copy of the GNU General Public License
22441 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22442 + */
22443 +
22444 +/*
22445 + * mount options/flags
22446 + */
22447 +
22448 +#include <linux/namei.h>
22449 +#include <linux/types.h> /* a distribution requires */
22450 +#include <linux/parser.h>
22451 +#include "aufs.h"
22452 +
22453 +/* ---------------------------------------------------------------------- */
22454 +
22455 +enum {
22456 +       Opt_br,
22457 +       Opt_add, Opt_del, Opt_mod, Opt_reorder, Opt_append, Opt_prepend,
22458 +       Opt_idel, Opt_imod, Opt_ireorder,
22459 +       Opt_dirwh, Opt_rdcache, Opt_rdblk, Opt_rdhash, Opt_rendir,
22460 +       Opt_rdblk_def, Opt_rdhash_def,
22461 +       Opt_xino, Opt_zxino, Opt_noxino,
22462 +       Opt_trunc_xino, Opt_trunc_xino_v, Opt_notrunc_xino,
22463 +       Opt_trunc_xino_path, Opt_itrunc_xino,
22464 +       Opt_trunc_xib, Opt_notrunc_xib,
22465 +       Opt_shwh, Opt_noshwh,
22466 +       Opt_plink, Opt_noplink, Opt_list_plink,
22467 +       Opt_udba,
22468 +       Opt_dio, Opt_nodio,
22469 +       /* Opt_lock, Opt_unlock, */
22470 +       Opt_cmd, Opt_cmd_args,
22471 +       Opt_diropq_a, Opt_diropq_w,
22472 +       Opt_warn_perm, Opt_nowarn_perm,
22473 +       Opt_wbr_copyup, Opt_wbr_create,
22474 +       Opt_fhsm_sec,
22475 +       Opt_refrof, Opt_norefrof,
22476 +       Opt_verbose, Opt_noverbose,
22477 +       Opt_sum, Opt_nosum, Opt_wsum,
22478 +       Opt_dirperm1, Opt_nodirperm1,
22479 +       Opt_acl, Opt_noacl,
22480 +       Opt_tail, Opt_ignore, Opt_ignore_silent, Opt_err
22481 +};
22482 +
22483 +static match_table_t options = {
22484 +       {Opt_br, "br=%s"},
22485 +       {Opt_br, "br:%s"},
22486 +
22487 +       {Opt_add, "add=%d:%s"},
22488 +       {Opt_add, "add:%d:%s"},
22489 +       {Opt_add, "ins=%d:%s"},
22490 +       {Opt_add, "ins:%d:%s"},
22491 +       {Opt_append, "append=%s"},
22492 +       {Opt_append, "append:%s"},
22493 +       {Opt_prepend, "prepend=%s"},
22494 +       {Opt_prepend, "prepend:%s"},
22495 +
22496 +       {Opt_del, "del=%s"},
22497 +       {Opt_del, "del:%s"},
22498 +       /* {Opt_idel, "idel:%d"}, */
22499 +       {Opt_mod, "mod=%s"},
22500 +       {Opt_mod, "mod:%s"},
22501 +       /* {Opt_imod, "imod:%d:%s"}, */
22502 +
22503 +       {Opt_dirwh, "dirwh=%d"},
22504 +
22505 +       {Opt_xino, "xino=%s"},
22506 +       {Opt_noxino, "noxino"},
22507 +       {Opt_trunc_xino, "trunc_xino"},
22508 +       {Opt_trunc_xino_v, "trunc_xino_v=%d:%d"},
22509 +       {Opt_notrunc_xino, "notrunc_xino"},
22510 +       {Opt_trunc_xino_path, "trunc_xino=%s"},
22511 +       {Opt_itrunc_xino, "itrunc_xino=%d"},
22512 +       /* {Opt_zxino, "zxino=%s"}, */
22513 +       {Opt_trunc_xib, "trunc_xib"},
22514 +       {Opt_notrunc_xib, "notrunc_xib"},
22515 +
22516 +#ifdef CONFIG_PROC_FS
22517 +       {Opt_plink, "plink"},
22518 +#else
22519 +       {Opt_ignore_silent, "plink"},
22520 +#endif
22521 +
22522 +       {Opt_noplink, "noplink"},
22523 +
22524 +#ifdef CONFIG_AUFS_DEBUG
22525 +       {Opt_list_plink, "list_plink"},
22526 +#endif
22527 +
22528 +       {Opt_udba, "udba=%s"},
22529 +
22530 +       {Opt_dio, "dio"},
22531 +       {Opt_nodio, "nodio"},
22532 +
22533 +#ifdef CONFIG_AUFS_FHSM
22534 +       {Opt_fhsm_sec, "fhsm_sec=%d"},
22535 +#else
22536 +       {Opt_ignore_silent, "fhsm_sec=%d"},
22537 +#endif
22538 +
22539 +       {Opt_diropq_a, "diropq=always"},
22540 +       {Opt_diropq_a, "diropq=a"},
22541 +       {Opt_diropq_w, "diropq=whiteouted"},
22542 +       {Opt_diropq_w, "diropq=w"},
22543 +
22544 +       {Opt_warn_perm, "warn_perm"},
22545 +       {Opt_nowarn_perm, "nowarn_perm"},
22546 +
22547 +       /* keep them temporary */
22548 +       {Opt_ignore_silent, "nodlgt"},
22549 +       {Opt_ignore_silent, "clean_plink"},
22550 +
22551 +#ifdef CONFIG_AUFS_SHWH
22552 +       {Opt_shwh, "shwh"},
22553 +#endif
22554 +       {Opt_noshwh, "noshwh"},
22555 +
22556 +       {Opt_dirperm1, "dirperm1"},
22557 +       {Opt_nodirperm1, "nodirperm1"},
22558 +
22559 +       {Opt_rendir, "rendir=%d"},
22560 +
22561 +       {Opt_refrof, "refrof"},
22562 +       {Opt_norefrof, "norefrof"},
22563 +
22564 +       {Opt_verbose, "verbose"},
22565 +       {Opt_verbose, "v"},
22566 +       {Opt_noverbose, "noverbose"},
22567 +       {Opt_noverbose, "quiet"},
22568 +       {Opt_noverbose, "q"},
22569 +       {Opt_noverbose, "silent"},
22570 +
22571 +       {Opt_sum, "sum"},
22572 +       {Opt_nosum, "nosum"},
22573 +       {Opt_wsum, "wsum"},
22574 +
22575 +       {Opt_rdcache, "rdcache=%d"},
22576 +       {Opt_rdblk, "rdblk=%d"},
22577 +       {Opt_rdblk_def, "rdblk=def"},
22578 +       {Opt_rdhash, "rdhash=%d"},
22579 +       {Opt_rdhash_def, "rdhash=def"},
22580 +
22581 +       {Opt_wbr_create, "create=%s"},
22582 +       {Opt_wbr_create, "create_policy=%s"},
22583 +       {Opt_wbr_copyup, "cpup=%s"},
22584 +       {Opt_wbr_copyup, "copyup=%s"},
22585 +       {Opt_wbr_copyup, "copyup_policy=%s"},
22586 +
22587 +       /* generic VFS flag */
22588 +#ifdef CONFIG_FS_POSIX_ACL
22589 +       {Opt_acl, "acl"},
22590 +       {Opt_noacl, "noacl"},
22591 +#else
22592 +       {Opt_ignore_silent, "acl"},
22593 +       {Opt_ignore_silent, "noacl"},
22594 +#endif
22595 +
22596 +       /* internal use for the scripts */
22597 +       {Opt_ignore_silent, "si=%s"},
22598 +
22599 +       {Opt_br, "dirs=%s"},
22600 +       {Opt_ignore, "debug=%d"},
22601 +       {Opt_ignore, "delete=whiteout"},
22602 +       {Opt_ignore, "delete=all"},
22603 +       {Opt_ignore, "imap=%s"},
22604 +
22605 +       /* temporary workaround, due to old mount(8)? */
22606 +       {Opt_ignore_silent, "relatime"},
22607 +
22608 +       {Opt_err, NULL}
22609 +};
22610 +
22611 +/* ---------------------------------------------------------------------- */
22612 +
22613 +static const char *au_parser_pattern(int val, match_table_t tbl)
22614 +{
22615 +       struct match_token *p;
22616 +
22617 +       p = tbl;
22618 +       while (p->pattern) {
22619 +               if (p->token == val)
22620 +                       return p->pattern;
22621 +               p++;
22622 +       }
22623 +       BUG();
22624 +       return "??";
22625 +}
22626 +
22627 +static const char *au_optstr(int *val, match_table_t tbl)
22628 +{
22629 +       struct match_token *p;
22630 +       int v;
22631 +
22632 +       v = *val;
22633 +       if (!v)
22634 +               goto out;
22635 +       p = tbl;
22636 +       while (p->pattern) {
22637 +               if (p->token
22638 +                   && (v & p->token) == p->token) {
22639 +                       *val &= ~p->token;
22640 +                       return p->pattern;
22641 +               }
22642 +               p++;
22643 +       }
22644 +
22645 +out:
22646 +       return NULL;
22647 +}
22648 +
22649 +/* ---------------------------------------------------------------------- */
22650 +
22651 +static match_table_t brperm = {
22652 +       {AuBrPerm_RO, AUFS_BRPERM_RO},
22653 +       {AuBrPerm_RR, AUFS_BRPERM_RR},
22654 +       {AuBrPerm_RW, AUFS_BRPERM_RW},
22655 +       {0, NULL}
22656 +};
22657 +
22658 +static match_table_t brattr = {
22659 +       /* general */
22660 +       {AuBrAttr_COO_REG, AUFS_BRATTR_COO_REG},
22661 +       {AuBrAttr_COO_ALL, AUFS_BRATTR_COO_ALL},
22662 +       /* 'unpin' attrib is meaningless since linux-3.18-rc1 */
22663 +       {AuBrAttr_UNPIN, AUFS_BRATTR_UNPIN},
22664 +#ifdef CONFIG_AUFS_FHSM
22665 +       {AuBrAttr_FHSM, AUFS_BRATTR_FHSM},
22666 +#endif
22667 +#ifdef CONFIG_AUFS_XATTR
22668 +       {AuBrAttr_ICEX, AUFS_BRATTR_ICEX},
22669 +       {AuBrAttr_ICEX_SEC, AUFS_BRATTR_ICEX_SEC},
22670 +       {AuBrAttr_ICEX_SYS, AUFS_BRATTR_ICEX_SYS},
22671 +       {AuBrAttr_ICEX_TR, AUFS_BRATTR_ICEX_TR},
22672 +       {AuBrAttr_ICEX_USR, AUFS_BRATTR_ICEX_USR},
22673 +       {AuBrAttr_ICEX_OTH, AUFS_BRATTR_ICEX_OTH},
22674 +#endif
22675 +
22676 +       /* ro/rr branch */
22677 +       {AuBrRAttr_WH, AUFS_BRRATTR_WH},
22678 +
22679 +       /* rw branch */
22680 +       {AuBrWAttr_MOO, AUFS_BRWATTR_MOO},
22681 +       {AuBrWAttr_NoLinkWH, AUFS_BRWATTR_NLWH},
22682 +
22683 +       {0, NULL}
22684 +};
22685 +
22686 +static int br_attr_val(char *str, match_table_t table, substring_t args[])
22687 +{
22688 +       int attr, v;
22689 +       char *p;
22690 +
22691 +       attr = 0;
22692 +       do {
22693 +               p = strchr(str, '+');
22694 +               if (p)
22695 +                       *p = 0;
22696 +               v = match_token(str, table, args);
22697 +               if (v) {
22698 +                       if (v & AuBrAttr_CMOO_Mask)
22699 +                               attr &= ~AuBrAttr_CMOO_Mask;
22700 +                       attr |= v;
22701 +               } else {
22702 +                       if (p)
22703 +                               *p = '+';
22704 +                       pr_warn("ignored branch attribute %s\n", str);
22705 +                       break;
22706 +               }
22707 +               if (p)
22708 +                       str = p + 1;
22709 +       } while (p);
22710 +
22711 +       return attr;
22712 +}
22713 +
22714 +static int au_do_optstr_br_attr(au_br_perm_str_t *str, int perm)
22715 +{
22716 +       int sz;
22717 +       const char *p;
22718 +       char *q;
22719 +
22720 +       q = str->a;
22721 +       *q = 0;
22722 +       p = au_optstr(&perm, brattr);
22723 +       if (p) {
22724 +               sz = strlen(p);
22725 +               memcpy(q, p, sz + 1);
22726 +               q += sz;
22727 +       } else
22728 +               goto out;
22729 +
22730 +       do {
22731 +               p = au_optstr(&perm, brattr);
22732 +               if (p) {
22733 +                       *q++ = '+';
22734 +                       sz = strlen(p);
22735 +                       memcpy(q, p, sz + 1);
22736 +                       q += sz;
22737 +               }
22738 +       } while (p);
22739 +
22740 +out:
22741 +       return q - str->a;
22742 +}
22743 +
22744 +static int noinline_for_stack br_perm_val(char *perm)
22745 +{
22746 +       int val, bad, sz;
22747 +       char *p;
22748 +       substring_t args[MAX_OPT_ARGS];
22749 +       au_br_perm_str_t attr;
22750 +
22751 +       p = strchr(perm, '+');
22752 +       if (p)
22753 +               *p = 0;
22754 +       val = match_token(perm, brperm, args);
22755 +       if (!val) {
22756 +               if (p)
22757 +                       *p = '+';
22758 +               pr_warn("ignored branch permission %s\n", perm);
22759 +               val = AuBrPerm_RO;
22760 +               goto out;
22761 +       }
22762 +       if (!p)
22763 +               goto out;
22764 +
22765 +       val |= br_attr_val(p + 1, brattr, args);
22766 +
22767 +       bad = 0;
22768 +       switch (val & AuBrPerm_Mask) {
22769 +       case AuBrPerm_RO:
22770 +       case AuBrPerm_RR:
22771 +               bad = val & AuBrWAttr_Mask;
22772 +               val &= ~AuBrWAttr_Mask;
22773 +               break;
22774 +       case AuBrPerm_RW:
22775 +               bad = val & AuBrRAttr_Mask;
22776 +               val &= ~AuBrRAttr_Mask;
22777 +               break;
22778 +       }
22779 +
22780 +       /*
22781 +        * 'unpin' attrib becomes meaningless since linux-3.18-rc1, but aufs
22782 +        * does not treat it as an error, just warning.
22783 +        * this is a tiny guard for the user operation.
22784 +        */
22785 +       if (val & AuBrAttr_UNPIN) {
22786 +               bad |= AuBrAttr_UNPIN;
22787 +               val &= ~AuBrAttr_UNPIN;
22788 +       }
22789 +
22790 +       if (unlikely(bad)) {
22791 +               sz = au_do_optstr_br_attr(&attr, bad);
22792 +               AuDebugOn(!sz);
22793 +               pr_warn("ignored branch attribute %s\n", attr.a);
22794 +       }
22795 +
22796 +out:
22797 +       return val;
22798 +}
22799 +
22800 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm)
22801 +{
22802 +       au_br_perm_str_t attr;
22803 +       const char *p;
22804 +       char *q;
22805 +       int sz;
22806 +
22807 +       q = str->a;
22808 +       p = au_optstr(&perm, brperm);
22809 +       AuDebugOn(!p || !*p);
22810 +       sz = strlen(p);
22811 +       memcpy(q, p, sz + 1);
22812 +       q += sz;
22813 +
22814 +       sz = au_do_optstr_br_attr(&attr, perm);
22815 +       if (sz) {
22816 +               *q++ = '+';
22817 +               memcpy(q, attr.a, sz + 1);
22818 +       }
22819 +
22820 +       AuDebugOn(strlen(str->a) >= sizeof(str->a));
22821 +}
22822 +
22823 +/* ---------------------------------------------------------------------- */
22824 +
22825 +static match_table_t udbalevel = {
22826 +       {AuOpt_UDBA_REVAL, "reval"},
22827 +       {AuOpt_UDBA_NONE, "none"},
22828 +#ifdef CONFIG_AUFS_HNOTIFY
22829 +       {AuOpt_UDBA_HNOTIFY, "notify"}, /* abstraction */
22830 +#ifdef CONFIG_AUFS_HFSNOTIFY
22831 +       {AuOpt_UDBA_HNOTIFY, "fsnotify"},
22832 +#endif
22833 +#endif
22834 +       {-1, NULL}
22835 +};
22836 +
22837 +static int noinline_for_stack udba_val(char *str)
22838 +{
22839 +       substring_t args[MAX_OPT_ARGS];
22840 +
22841 +       return match_token(str, udbalevel, args);
22842 +}
22843 +
22844 +const char *au_optstr_udba(int udba)
22845 +{
22846 +       return au_parser_pattern(udba, udbalevel);
22847 +}
22848 +
22849 +/* ---------------------------------------------------------------------- */
22850 +
22851 +static match_table_t au_wbr_create_policy = {
22852 +       {AuWbrCreate_TDP, "tdp"},
22853 +       {AuWbrCreate_TDP, "top-down-parent"},
22854 +       {AuWbrCreate_RR, "rr"},
22855 +       {AuWbrCreate_RR, "round-robin"},
22856 +       {AuWbrCreate_MFS, "mfs"},
22857 +       {AuWbrCreate_MFS, "most-free-space"},
22858 +       {AuWbrCreate_MFSV, "mfs:%d"},
22859 +       {AuWbrCreate_MFSV, "most-free-space:%d"},
22860 +
22861 +       {AuWbrCreate_MFSRR, "mfsrr:%d"},
22862 +       {AuWbrCreate_MFSRRV, "mfsrr:%d:%d"},
22863 +       {AuWbrCreate_PMFS, "pmfs"},
22864 +       {AuWbrCreate_PMFSV, "pmfs:%d"},
22865 +       {AuWbrCreate_PMFSRR, "pmfsrr:%d"},
22866 +       {AuWbrCreate_PMFSRRV, "pmfsrr:%d:%d"},
22867 +
22868 +       {-1, NULL}
22869 +};
22870 +
22871 +/*
22872 + * cf. linux/lib/parser.c and cmdline.c
22873 + * gave up calling memparse() since it uses simple_strtoull() instead of
22874 + * kstrto...().
22875 + */
22876 +static int noinline_for_stack
22877 +au_match_ull(substring_t *s, unsigned long long *result)
22878 +{
22879 +       int err;
22880 +       unsigned int len;
22881 +       char a[32];
22882 +
22883 +       err = -ERANGE;
22884 +       len = s->to - s->from;
22885 +       if (len + 1 <= sizeof(a)) {
22886 +               memcpy(a, s->from, len);
22887 +               a[len] = '\0';
22888 +               err = kstrtoull(a, 0, result);
22889 +       }
22890 +       return err;
22891 +}
22892 +
22893 +static int au_wbr_mfs_wmark(substring_t *arg, char *str,
22894 +                           struct au_opt_wbr_create *create)
22895 +{
22896 +       int err;
22897 +       unsigned long long ull;
22898 +
22899 +       err = 0;
22900 +       if (!au_match_ull(arg, &ull))
22901 +               create->mfsrr_watermark = ull;
22902 +       else {
22903 +               pr_err("bad integer in %s\n", str);
22904 +               err = -EINVAL;
22905 +       }
22906 +
22907 +       return err;
22908 +}
22909 +
22910 +static int au_wbr_mfs_sec(substring_t *arg, char *str,
22911 +                         struct au_opt_wbr_create *create)
22912 +{
22913 +       int n, err;
22914 +
22915 +       err = 0;
22916 +       if (!match_int(arg, &n) && 0 <= n && n <= AUFS_MFS_MAX_SEC)
22917 +               create->mfs_second = n;
22918 +       else {
22919 +               pr_err("bad integer in %s\n", str);
22920 +               err = -EINVAL;
22921 +       }
22922 +
22923 +       return err;
22924 +}
22925 +
22926 +static int noinline_for_stack
22927 +au_wbr_create_val(char *str, struct au_opt_wbr_create *create)
22928 +{
22929 +       int err, e;
22930 +       substring_t args[MAX_OPT_ARGS];
22931 +
22932 +       err = match_token(str, au_wbr_create_policy, args);
22933 +       create->wbr_create = err;
22934 +       switch (err) {
22935 +       case AuWbrCreate_MFSRRV:
22936 +       case AuWbrCreate_PMFSRRV:
22937 +               e = au_wbr_mfs_wmark(&args[0], str, create);
22938 +               if (!e)
22939 +                       e = au_wbr_mfs_sec(&args[1], str, create);
22940 +               if (unlikely(e))
22941 +                       err = e;
22942 +               break;
22943 +       case AuWbrCreate_MFSRR:
22944 +       case AuWbrCreate_PMFSRR:
22945 +               e = au_wbr_mfs_wmark(&args[0], str, create);
22946 +               if (unlikely(e)) {
22947 +                       err = e;
22948 +                       break;
22949 +               }
22950 +               /*FALLTHROUGH*/
22951 +       case AuWbrCreate_MFS:
22952 +       case AuWbrCreate_PMFS:
22953 +               create->mfs_second = AUFS_MFS_DEF_SEC;
22954 +               break;
22955 +       case AuWbrCreate_MFSV:
22956 +       case AuWbrCreate_PMFSV:
22957 +               e = au_wbr_mfs_sec(&args[0], str, create);
22958 +               if (unlikely(e))
22959 +                       err = e;
22960 +               break;
22961 +       }
22962 +
22963 +       return err;
22964 +}
22965 +
22966 +const char *au_optstr_wbr_create(int wbr_create)
22967 +{
22968 +       return au_parser_pattern(wbr_create, au_wbr_create_policy);
22969 +}
22970 +
22971 +static match_table_t au_wbr_copyup_policy = {
22972 +       {AuWbrCopyup_TDP, "tdp"},
22973 +       {AuWbrCopyup_TDP, "top-down-parent"},
22974 +       {AuWbrCopyup_BUP, "bup"},
22975 +       {AuWbrCopyup_BUP, "bottom-up-parent"},
22976 +       {AuWbrCopyup_BU, "bu"},
22977 +       {AuWbrCopyup_BU, "bottom-up"},
22978 +       {-1, NULL}
22979 +};
22980 +
22981 +static int noinline_for_stack au_wbr_copyup_val(char *str)
22982 +{
22983 +       substring_t args[MAX_OPT_ARGS];
22984 +
22985 +       return match_token(str, au_wbr_copyup_policy, args);
22986 +}
22987 +
22988 +const char *au_optstr_wbr_copyup(int wbr_copyup)
22989 +{
22990 +       return au_parser_pattern(wbr_copyup, au_wbr_copyup_policy);
22991 +}
22992 +
22993 +/* ---------------------------------------------------------------------- */
22994 +
22995 +static const int lkup_dirflags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
22996 +
22997 +static void dump_opts(struct au_opts *opts)
22998 +{
22999 +#ifdef CONFIG_AUFS_DEBUG
23000 +       /* reduce stack space */
23001 +       union {
23002 +               struct au_opt_add *add;
23003 +               struct au_opt_del *del;
23004 +               struct au_opt_mod *mod;
23005 +               struct au_opt_xino *xino;
23006 +               struct au_opt_xino_itrunc *xino_itrunc;
23007 +               struct au_opt_wbr_create *create;
23008 +       } u;
23009 +       struct au_opt *opt;
23010 +
23011 +       opt = opts->opt;
23012 +       while (opt->type != Opt_tail) {
23013 +               switch (opt->type) {
23014 +               case Opt_add:
23015 +                       u.add = &opt->add;
23016 +                       AuDbg("add {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_del:
23021 +               case Opt_idel:
23022 +                       u.del = &opt->del;
23023 +                       AuDbg("del {%s, %p}\n",
23024 +                             u.del->pathname, u.del->h_path.dentry);
23025 +                       break;
23026 +               case Opt_mod:
23027 +               case Opt_imod:
23028 +                       u.mod = &opt->mod;
23029 +                       AuDbg("mod {%s, 0x%x, %p}\n",
23030 +                                 u.mod->path, u.mod->perm, u.mod->h_root);
23031 +                       break;
23032 +               case Opt_append:
23033 +                       u.add = &opt->add;
23034 +                       AuDbg("append {b%d, %s, 0x%x, %p}\n",
23035 +                                 u.add->bindex, u.add->pathname, u.add->perm,
23036 +                                 u.add->path.dentry);
23037 +                       break;
23038 +               case Opt_prepend:
23039 +                       u.add = &opt->add;
23040 +                       AuDbg("prepend {b%d, %s, 0x%x, %p}\n",
23041 +                                 u.add->bindex, u.add->pathname, u.add->perm,
23042 +                                 u.add->path.dentry);
23043 +                       break;
23044 +               case Opt_dirwh:
23045 +                       AuDbg("dirwh %d\n", opt->dirwh);
23046 +                       break;
23047 +               case Opt_rdcache:
23048 +                       AuDbg("rdcache %d\n", opt->rdcache);
23049 +                       break;
23050 +               case Opt_rdblk:
23051 +                       AuDbg("rdblk %u\n", opt->rdblk);
23052 +                       break;
23053 +               case Opt_rdblk_def:
23054 +                       AuDbg("rdblk_def\n");
23055 +                       break;
23056 +               case Opt_rdhash:
23057 +                       AuDbg("rdhash %u\n", opt->rdhash);
23058 +                       break;
23059 +               case Opt_rdhash_def:
23060 +                       AuDbg("rdhash_def\n");
23061 +                       break;
23062 +               case Opt_xino:
23063 +                       u.xino = &opt->xino;
23064 +                       AuDbg("xino {%s %pD}\n", u.xino->path, u.xino->file);
23065 +                       break;
23066 +               case Opt_trunc_xino:
23067 +                       AuLabel(trunc_xino);
23068 +                       break;
23069 +               case Opt_notrunc_xino:
23070 +                       AuLabel(notrunc_xino);
23071 +                       break;
23072 +               case Opt_trunc_xino_path:
23073 +               case Opt_itrunc_xino:
23074 +                       u.xino_itrunc = &opt->xino_itrunc;
23075 +                       AuDbg("trunc_xino %d\n", u.xino_itrunc->bindex);
23076 +                       break;
23077 +
23078 +               case Opt_noxino:
23079 +                       AuLabel(noxino);
23080 +                       break;
23081 +               case Opt_trunc_xib:
23082 +                       AuLabel(trunc_xib);
23083 +                       break;
23084 +               case Opt_notrunc_xib:
23085 +                       AuLabel(notrunc_xib);
23086 +                       break;
23087 +               case Opt_shwh:
23088 +                       AuLabel(shwh);
23089 +                       break;
23090 +               case Opt_noshwh:
23091 +                       AuLabel(noshwh);
23092 +                       break;
23093 +               case Opt_dirperm1:
23094 +                       AuLabel(dirperm1);
23095 +                       break;
23096 +               case Opt_nodirperm1:
23097 +                       AuLabel(nodirperm1);
23098 +                       break;
23099 +               case Opt_plink:
23100 +                       AuLabel(plink);
23101 +                       break;
23102 +               case Opt_noplink:
23103 +                       AuLabel(noplink);
23104 +                       break;
23105 +               case Opt_list_plink:
23106 +                       AuLabel(list_plink);
23107 +                       break;
23108 +               case Opt_udba:
23109 +                       AuDbg("udba %d, %s\n",
23110 +                                 opt->udba, au_optstr_udba(opt->udba));
23111 +                       break;
23112 +               case Opt_dio:
23113 +                       AuLabel(dio);
23114 +                       break;
23115 +               case Opt_nodio:
23116 +                       AuLabel(nodio);
23117 +                       break;
23118 +               case Opt_diropq_a:
23119 +                       AuLabel(diropq_a);
23120 +                       break;
23121 +               case Opt_diropq_w:
23122 +                       AuLabel(diropq_w);
23123 +                       break;
23124 +               case Opt_warn_perm:
23125 +                       AuLabel(warn_perm);
23126 +                       break;
23127 +               case Opt_nowarn_perm:
23128 +                       AuLabel(nowarn_perm);
23129 +                       break;
23130 +               case Opt_refrof:
23131 +                       AuLabel(refrof);
23132 +                       break;
23133 +               case Opt_norefrof:
23134 +                       AuLabel(norefrof);
23135 +                       break;
23136 +               case Opt_verbose:
23137 +                       AuLabel(verbose);
23138 +                       break;
23139 +               case Opt_noverbose:
23140 +                       AuLabel(noverbose);
23141 +                       break;
23142 +               case Opt_sum:
23143 +                       AuLabel(sum);
23144 +                       break;
23145 +               case Opt_nosum:
23146 +                       AuLabel(nosum);
23147 +                       break;
23148 +               case Opt_wsum:
23149 +                       AuLabel(wsum);
23150 +                       break;
23151 +               case Opt_wbr_create:
23152 +                       u.create = &opt->wbr_create;
23153 +                       AuDbg("create %d, %s\n", u.create->wbr_create,
23154 +                                 au_optstr_wbr_create(u.create->wbr_create));
23155 +                       switch (u.create->wbr_create) {
23156 +                       case AuWbrCreate_MFSV:
23157 +                       case AuWbrCreate_PMFSV:
23158 +                               AuDbg("%d sec\n", u.create->mfs_second);
23159 +                               break;
23160 +                       case AuWbrCreate_MFSRR:
23161 +                               AuDbg("%llu watermark\n",
23162 +                                         u.create->mfsrr_watermark);
23163 +                               break;
23164 +                       case AuWbrCreate_MFSRRV:
23165 +                       case AuWbrCreate_PMFSRRV:
23166 +                               AuDbg("%llu watermark, %d sec\n",
23167 +                                         u.create->mfsrr_watermark,
23168 +                                         u.create->mfs_second);
23169 +                               break;
23170 +                       }
23171 +                       break;
23172 +               case Opt_wbr_copyup:
23173 +                       AuDbg("copyup %d, %s\n", opt->wbr_copyup,
23174 +                                 au_optstr_wbr_copyup(opt->wbr_copyup));
23175 +                       break;
23176 +               case Opt_fhsm_sec:
23177 +                       AuDbg("fhsm_sec %u\n", opt->fhsm_second);
23178 +                       break;
23179 +               case Opt_acl:
23180 +                       AuLabel(acl);
23181 +                       break;
23182 +               case Opt_noacl:
23183 +                       AuLabel(noacl);
23184 +                       break;
23185 +               default:
23186 +                       BUG();
23187 +               }
23188 +               opt++;
23189 +       }
23190 +#endif
23191 +}
23192 +
23193 +void au_opts_free(struct au_opts *opts)
23194 +{
23195 +       struct au_opt *opt;
23196 +
23197 +       opt = opts->opt;
23198 +       while (opt->type != Opt_tail) {
23199 +               switch (opt->type) {
23200 +               case Opt_add:
23201 +               case Opt_append:
23202 +               case Opt_prepend:
23203 +                       path_put(&opt->add.path);
23204 +                       break;
23205 +               case Opt_del:
23206 +               case Opt_idel:
23207 +                       path_put(&opt->del.h_path);
23208 +                       break;
23209 +               case Opt_mod:
23210 +               case Opt_imod:
23211 +                       dput(opt->mod.h_root);
23212 +                       break;
23213 +               case Opt_xino:
23214 +                       fput(opt->xino.file);
23215 +                       break;
23216 +               }
23217 +               opt++;
23218 +       }
23219 +}
23220 +
23221 +static int opt_add(struct au_opt *opt, char *opt_str, unsigned long sb_flags,
23222 +                  aufs_bindex_t bindex)
23223 +{
23224 +       int err;
23225 +       struct au_opt_add *add = &opt->add;
23226 +       char *p;
23227 +
23228 +       add->bindex = bindex;
23229 +       add->perm = AuBrPerm_RO;
23230 +       add->pathname = opt_str;
23231 +       p = strchr(opt_str, '=');
23232 +       if (p) {
23233 +               *p++ = 0;
23234 +               if (*p)
23235 +                       add->perm = br_perm_val(p);
23236 +       }
23237 +
23238 +       err = vfsub_kern_path(add->pathname, lkup_dirflags, &add->path);
23239 +       if (!err) {
23240 +               if (!p) {
23241 +                       add->perm = AuBrPerm_RO;
23242 +                       if (au_test_fs_rr(add->path.dentry->d_sb))
23243 +                               add->perm = AuBrPerm_RR;
23244 +                       else if (!bindex && !(sb_flags & MS_RDONLY))
23245 +                               add->perm = AuBrPerm_RW;
23246 +               }
23247 +               opt->type = Opt_add;
23248 +               goto out;
23249 +       }
23250 +       pr_err("lookup failed %s (%d)\n", add->pathname, err);
23251 +       err = -EINVAL;
23252 +
23253 +out:
23254 +       return err;
23255 +}
23256 +
23257 +static int au_opts_parse_del(struct au_opt_del *del, substring_t args[])
23258 +{
23259 +       int err;
23260 +
23261 +       del->pathname = args[0].from;
23262 +       AuDbg("del path %s\n", del->pathname);
23263 +
23264 +       err = vfsub_kern_path(del->pathname, lkup_dirflags, &del->h_path);
23265 +       if (unlikely(err))
23266 +               pr_err("lookup failed %s (%d)\n", del->pathname, err);
23267 +
23268 +       return err;
23269 +}
23270 +
23271 +#if 0 /* reserved for future use */
23272 +static int au_opts_parse_idel(struct super_block *sb, aufs_bindex_t bindex,
23273 +                             struct au_opt_del *del, substring_t args[])
23274 +{
23275 +       int err;
23276 +       struct dentry *root;
23277 +
23278 +       err = -EINVAL;
23279 +       root = sb->s_root;
23280 +       aufs_read_lock(root, AuLock_FLUSH);
23281 +       if (bindex < 0 || au_sbend(sb) < bindex) {
23282 +               pr_err("out of bounds, %d\n", bindex);
23283 +               goto out;
23284 +       }
23285 +
23286 +       err = 0;
23287 +       del->h_path.dentry = dget(au_h_dptr(root, bindex));
23288 +       del->h_path.mnt = mntget(au_sbr_mnt(sb, bindex));
23289 +
23290 +out:
23291 +       aufs_read_unlock(root, !AuLock_IR);
23292 +       return err;
23293 +}
23294 +#endif
23295 +
23296 +static int noinline_for_stack
23297 +au_opts_parse_mod(struct au_opt_mod *mod, substring_t args[])
23298 +{
23299 +       int err;
23300 +       struct path path;
23301 +       char *p;
23302 +
23303 +       err = -EINVAL;
23304 +       mod->path = args[0].from;
23305 +       p = strchr(mod->path, '=');
23306 +       if (unlikely(!p)) {
23307 +               pr_err("no permssion %s\n", args[0].from);
23308 +               goto out;
23309 +       }
23310 +
23311 +       *p++ = 0;
23312 +       err = vfsub_kern_path(mod->path, lkup_dirflags, &path);
23313 +       if (unlikely(err)) {
23314 +               pr_err("lookup failed %s (%d)\n", mod->path, err);
23315 +               goto out;
23316 +       }
23317 +
23318 +       mod->perm = br_perm_val(p);
23319 +       AuDbg("mod path %s, perm 0x%x, %s\n", mod->path, mod->perm, p);
23320 +       mod->h_root = dget(path.dentry);
23321 +       path_put(&path);
23322 +
23323 +out:
23324 +       return err;
23325 +}
23326 +
23327 +#if 0 /* reserved for future use */
23328 +static int au_opts_parse_imod(struct super_block *sb, aufs_bindex_t bindex,
23329 +                             struct au_opt_mod *mod, substring_t args[])
23330 +{
23331 +       int err;
23332 +       struct dentry *root;
23333 +
23334 +       err = -EINVAL;
23335 +       root = sb->s_root;
23336 +       aufs_read_lock(root, AuLock_FLUSH);
23337 +       if (bindex < 0 || au_sbend(sb) < bindex) {
23338 +               pr_err("out of bounds, %d\n", bindex);
23339 +               goto out;
23340 +       }
23341 +
23342 +       err = 0;
23343 +       mod->perm = br_perm_val(args[1].from);
23344 +       AuDbg("mod path %s, perm 0x%x, %s\n",
23345 +             mod->path, mod->perm, args[1].from);
23346 +       mod->h_root = dget(au_h_dptr(root, bindex));
23347 +
23348 +out:
23349 +       aufs_read_unlock(root, !AuLock_IR);
23350 +       return err;
23351 +}
23352 +#endif
23353 +
23354 +static int au_opts_parse_xino(struct super_block *sb, struct au_opt_xino *xino,
23355 +                             substring_t args[])
23356 +{
23357 +       int err;
23358 +       struct file *file;
23359 +
23360 +       file = au_xino_create(sb, args[0].from, /*silent*/0);
23361 +       err = PTR_ERR(file);
23362 +       if (IS_ERR(file))
23363 +               goto out;
23364 +
23365 +       err = -EINVAL;
23366 +       if (unlikely(file->f_path.dentry->d_sb == sb)) {
23367 +               fput(file);
23368 +               pr_err("%s must be outside\n", args[0].from);
23369 +               goto out;
23370 +       }
23371 +
23372 +       err = 0;
23373 +       xino->file = file;
23374 +       xino->path = args[0].from;
23375 +
23376 +out:
23377 +       return err;
23378 +}
23379 +
23380 +static int noinline_for_stack
23381 +au_opts_parse_xino_itrunc_path(struct super_block *sb,
23382 +                              struct au_opt_xino_itrunc *xino_itrunc,
23383 +                              substring_t args[])
23384 +{
23385 +       int err;
23386 +       aufs_bindex_t bend, bindex;
23387 +       struct path path;
23388 +       struct dentry *root;
23389 +
23390 +       err = vfsub_kern_path(args[0].from, lkup_dirflags, &path);
23391 +       if (unlikely(err)) {
23392 +               pr_err("lookup failed %s (%d)\n", args[0].from, err);
23393 +               goto out;
23394 +       }
23395 +
23396 +       xino_itrunc->bindex = -1;
23397 +       root = sb->s_root;
23398 +       aufs_read_lock(root, AuLock_FLUSH);
23399 +       bend = au_sbend(sb);
23400 +       for (bindex = 0; bindex <= bend; bindex++) {
23401 +               if (au_h_dptr(root, bindex) == path.dentry) {
23402 +                       xino_itrunc->bindex = bindex;
23403 +                       break;
23404 +               }
23405 +       }
23406 +       aufs_read_unlock(root, !AuLock_IR);
23407 +       path_put(&path);
23408 +
23409 +       if (unlikely(xino_itrunc->bindex < 0)) {
23410 +               pr_err("no such branch %s\n", args[0].from);
23411 +               err = -EINVAL;
23412 +       }
23413 +
23414 +out:
23415 +       return err;
23416 +}
23417 +
23418 +/* called without aufs lock */
23419 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
23420 +{
23421 +       int err, n, token;
23422 +       aufs_bindex_t bindex;
23423 +       unsigned char skipped;
23424 +       struct dentry *root;
23425 +       struct au_opt *opt, *opt_tail;
23426 +       char *opt_str;
23427 +       /* reduce the stack space */
23428 +       union {
23429 +               struct au_opt_xino_itrunc *xino_itrunc;
23430 +               struct au_opt_wbr_create *create;
23431 +       } u;
23432 +       struct {
23433 +               substring_t args[MAX_OPT_ARGS];
23434 +       } *a;
23435 +
23436 +       err = -ENOMEM;
23437 +       a = kmalloc(sizeof(*a), GFP_NOFS);
23438 +       if (unlikely(!a))
23439 +               goto out;
23440 +
23441 +       root = sb->s_root;
23442 +       err = 0;
23443 +       bindex = 0;
23444 +       opt = opts->opt;
23445 +       opt_tail = opt + opts->max_opt - 1;
23446 +       opt->type = Opt_tail;
23447 +       while (!err && (opt_str = strsep(&str, ",")) && *opt_str) {
23448 +               err = -EINVAL;
23449 +               skipped = 0;
23450 +               token = match_token(opt_str, options, a->args);
23451 +               switch (token) {
23452 +               case Opt_br:
23453 +                       err = 0;
23454 +                       while (!err && (opt_str = strsep(&a->args[0].from, ":"))
23455 +                              && *opt_str) {
23456 +                               err = opt_add(opt, opt_str, opts->sb_flags,
23457 +                                             bindex++);
23458 +                               if (unlikely(!err && ++opt > opt_tail)) {
23459 +                                       err = -E2BIG;
23460 +                                       break;
23461 +                               }
23462 +                               opt->type = Opt_tail;
23463 +                               skipped = 1;
23464 +                       }
23465 +                       break;
23466 +               case Opt_add:
23467 +                       if (unlikely(match_int(&a->args[0], &n))) {
23468 +                               pr_err("bad integer in %s\n", opt_str);
23469 +                               break;
23470 +                       }
23471 +                       bindex = n;
23472 +                       err = opt_add(opt, a->args[1].from, opts->sb_flags,
23473 +                                     bindex);
23474 +                       if (!err)
23475 +                               opt->type = token;
23476 +                       break;
23477 +               case Opt_append:
23478 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
23479 +                                     /*dummy bindex*/1);
23480 +                       if (!err)
23481 +                               opt->type = token;
23482 +                       break;
23483 +               case Opt_prepend:
23484 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
23485 +                                     /*bindex*/0);
23486 +                       if (!err)
23487 +                               opt->type = token;
23488 +                       break;
23489 +               case Opt_del:
23490 +                       err = au_opts_parse_del(&opt->del, a->args);
23491 +                       if (!err)
23492 +                               opt->type = token;
23493 +                       break;
23494 +#if 0 /* reserved for future use */
23495 +               case Opt_idel:
23496 +                       del->pathname = "(indexed)";
23497 +                       if (unlikely(match_int(&args[0], &n))) {
23498 +                               pr_err("bad integer in %s\n", opt_str);
23499 +                               break;
23500 +                       }
23501 +                       err = au_opts_parse_idel(sb, n, &opt->del, a->args);
23502 +                       if (!err)
23503 +                               opt->type = token;
23504 +                       break;
23505 +#endif
23506 +               case Opt_mod:
23507 +                       err = au_opts_parse_mod(&opt->mod, a->args);
23508 +                       if (!err)
23509 +                               opt->type = token;
23510 +                       break;
23511 +#ifdef IMOD /* reserved for future use */
23512 +               case Opt_imod:
23513 +                       u.mod->path = "(indexed)";
23514 +                       if (unlikely(match_int(&a->args[0], &n))) {
23515 +                               pr_err("bad integer in %s\n", opt_str);
23516 +                               break;
23517 +                       }
23518 +                       err = au_opts_parse_imod(sb, n, &opt->mod, a->args);
23519 +                       if (!err)
23520 +                               opt->type = token;
23521 +                       break;
23522 +#endif
23523 +               case Opt_xino:
23524 +                       err = au_opts_parse_xino(sb, &opt->xino, a->args);
23525 +                       if (!err)
23526 +                               opt->type = token;
23527 +                       break;
23528 +
23529 +               case Opt_trunc_xino_path:
23530 +                       err = au_opts_parse_xino_itrunc_path
23531 +                               (sb, &opt->xino_itrunc, a->args);
23532 +                       if (!err)
23533 +                               opt->type = token;
23534 +                       break;
23535 +
23536 +               case Opt_itrunc_xino:
23537 +                       u.xino_itrunc = &opt->xino_itrunc;
23538 +                       if (unlikely(match_int(&a->args[0], &n))) {
23539 +                               pr_err("bad integer in %s\n", opt_str);
23540 +                               break;
23541 +                       }
23542 +                       u.xino_itrunc->bindex = n;
23543 +                       aufs_read_lock(root, AuLock_FLUSH);
23544 +                       if (n < 0 || au_sbend(sb) < n) {
23545 +                               pr_err("out of bounds, %d\n", n);
23546 +                               aufs_read_unlock(root, !AuLock_IR);
23547 +                               break;
23548 +                       }
23549 +                       aufs_read_unlock(root, !AuLock_IR);
23550 +                       err = 0;
23551 +                       opt->type = token;
23552 +                       break;
23553 +
23554 +               case Opt_dirwh:
23555 +                       if (unlikely(match_int(&a->args[0], &opt->dirwh)))
23556 +                               break;
23557 +                       err = 0;
23558 +                       opt->type = token;
23559 +                       break;
23560 +
23561 +               case Opt_rdcache:
23562 +                       if (unlikely(match_int(&a->args[0], &n))) {
23563 +                               pr_err("bad integer in %s\n", opt_str);
23564 +                               break;
23565 +                       }
23566 +                       if (unlikely(n > AUFS_RDCACHE_MAX)) {
23567 +                               pr_err("rdcache must be smaller than %d\n",
23568 +                                      AUFS_RDCACHE_MAX);
23569 +                               break;
23570 +                       }
23571 +                       opt->rdcache = n;
23572 +                       err = 0;
23573 +                       opt->type = token;
23574 +                       break;
23575 +               case Opt_rdblk:
23576 +                       if (unlikely(match_int(&a->args[0], &n)
23577 +                                    || n < 0
23578 +                                    || n > KMALLOC_MAX_SIZE)) {
23579 +                               pr_err("bad integer in %s\n", opt_str);
23580 +                               break;
23581 +                       }
23582 +                       if (unlikely(n && n < NAME_MAX)) {
23583 +                               pr_err("rdblk must be larger than %d\n",
23584 +                                      NAME_MAX);
23585 +                               break;
23586 +                       }
23587 +                       opt->rdblk = n;
23588 +                       err = 0;
23589 +                       opt->type = token;
23590 +                       break;
23591 +               case Opt_rdhash:
23592 +                       if (unlikely(match_int(&a->args[0], &n)
23593 +                                    || n < 0
23594 +                                    || n * sizeof(struct hlist_head)
23595 +                                    > KMALLOC_MAX_SIZE)) {
23596 +                               pr_err("bad integer in %s\n", opt_str);
23597 +                               break;
23598 +                       }
23599 +                       opt->rdhash = n;
23600 +                       err = 0;
23601 +                       opt->type = token;
23602 +                       break;
23603 +
23604 +               case Opt_trunc_xino:
23605 +               case Opt_notrunc_xino:
23606 +               case Opt_noxino:
23607 +               case Opt_trunc_xib:
23608 +               case Opt_notrunc_xib:
23609 +               case Opt_shwh:
23610 +               case Opt_noshwh:
23611 +               case Opt_dirperm1:
23612 +               case Opt_nodirperm1:
23613 +               case Opt_plink:
23614 +               case Opt_noplink:
23615 +               case Opt_list_plink:
23616 +               case Opt_dio:
23617 +               case Opt_nodio:
23618 +               case Opt_diropq_a:
23619 +               case Opt_diropq_w:
23620 +               case Opt_warn_perm:
23621 +               case Opt_nowarn_perm:
23622 +               case Opt_refrof:
23623 +               case Opt_norefrof:
23624 +               case Opt_verbose:
23625 +               case Opt_noverbose:
23626 +               case Opt_sum:
23627 +               case Opt_nosum:
23628 +               case Opt_wsum:
23629 +               case Opt_rdblk_def:
23630 +               case Opt_rdhash_def:
23631 +               case Opt_acl:
23632 +               case Opt_noacl:
23633 +                       err = 0;
23634 +                       opt->type = token;
23635 +                       break;
23636 +
23637 +               case Opt_udba:
23638 +                       opt->udba = udba_val(a->args[0].from);
23639 +                       if (opt->udba >= 0) {
23640 +                               err = 0;
23641 +                               opt->type = token;
23642 +                       } else
23643 +                               pr_err("wrong value, %s\n", opt_str);
23644 +                       break;
23645 +
23646 +               case Opt_wbr_create:
23647 +                       u.create = &opt->wbr_create;
23648 +                       u.create->wbr_create
23649 +                               = au_wbr_create_val(a->args[0].from, u.create);
23650 +                       if (u.create->wbr_create >= 0) {
23651 +                               err = 0;
23652 +                               opt->type = token;
23653 +                       } else
23654 +                               pr_err("wrong value, %s\n", opt_str);
23655 +                       break;
23656 +               case Opt_wbr_copyup:
23657 +                       opt->wbr_copyup = au_wbr_copyup_val(a->args[0].from);
23658 +                       if (opt->wbr_copyup >= 0) {
23659 +                               err = 0;
23660 +                               opt->type = token;
23661 +                       } else
23662 +                               pr_err("wrong value, %s\n", opt_str);
23663 +                       break;
23664 +
23665 +               case Opt_fhsm_sec:
23666 +                       if (unlikely(match_int(&a->args[0], &n)
23667 +                                    || n < 0)) {
23668 +                               pr_err("bad integer in %s\n", opt_str);
23669 +                               break;
23670 +                       }
23671 +                       if (sysaufs_brs) {
23672 +                               opt->fhsm_second = n;
23673 +                               opt->type = token;
23674 +                       } else
23675 +                               pr_warn("ignored %s\n", opt_str);
23676 +                       err = 0;
23677 +                       break;
23678 +
23679 +               case Opt_ignore:
23680 +                       pr_warn("ignored %s\n", opt_str);
23681 +                       /*FALLTHROUGH*/
23682 +               case Opt_ignore_silent:
23683 +                       skipped = 1;
23684 +                       err = 0;
23685 +                       break;
23686 +               case Opt_err:
23687 +                       pr_err("unknown option %s\n", opt_str);
23688 +                       break;
23689 +               }
23690 +
23691 +               if (!err && !skipped) {
23692 +                       if (unlikely(++opt > opt_tail)) {
23693 +                               err = -E2BIG;
23694 +                               opt--;
23695 +                               opt->type = Opt_tail;
23696 +                               break;
23697 +                       }
23698 +                       opt->type = Opt_tail;
23699 +               }
23700 +       }
23701 +
23702 +       kfree(a);
23703 +       dump_opts(opts);
23704 +       if (unlikely(err))
23705 +               au_opts_free(opts);
23706 +
23707 +out:
23708 +       return err;
23709 +}
23710 +
23711 +static int au_opt_wbr_create(struct super_block *sb,
23712 +                            struct au_opt_wbr_create *create)
23713 +{
23714 +       int err;
23715 +       struct au_sbinfo *sbinfo;
23716 +
23717 +       SiMustWriteLock(sb);
23718 +
23719 +       err = 1; /* handled */
23720 +       sbinfo = au_sbi(sb);
23721 +       if (sbinfo->si_wbr_create_ops->fin) {
23722 +               err = sbinfo->si_wbr_create_ops->fin(sb);
23723 +               if (!err)
23724 +                       err = 1;
23725 +       }
23726 +
23727 +       sbinfo->si_wbr_create = create->wbr_create;
23728 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + create->wbr_create;
23729 +       switch (create->wbr_create) {
23730 +       case AuWbrCreate_MFSRRV:
23731 +       case AuWbrCreate_MFSRR:
23732 +       case AuWbrCreate_PMFSRR:
23733 +       case AuWbrCreate_PMFSRRV:
23734 +               sbinfo->si_wbr_mfs.mfsrr_watermark = create->mfsrr_watermark;
23735 +               /*FALLTHROUGH*/
23736 +       case AuWbrCreate_MFS:
23737 +       case AuWbrCreate_MFSV:
23738 +       case AuWbrCreate_PMFS:
23739 +       case AuWbrCreate_PMFSV:
23740 +               sbinfo->si_wbr_mfs.mfs_expire
23741 +                       = msecs_to_jiffies(create->mfs_second * MSEC_PER_SEC);
23742 +               break;
23743 +       }
23744 +
23745 +       if (sbinfo->si_wbr_create_ops->init)
23746 +               sbinfo->si_wbr_create_ops->init(sb); /* ignore */
23747 +
23748 +       return err;
23749 +}
23750 +
23751 +/*
23752 + * returns,
23753 + * plus: processed without an error
23754 + * zero: unprocessed
23755 + */
23756 +static int au_opt_simple(struct super_block *sb, struct au_opt *opt,
23757 +                        struct au_opts *opts)
23758 +{
23759 +       int err;
23760 +       struct au_sbinfo *sbinfo;
23761 +
23762 +       SiMustWriteLock(sb);
23763 +
23764 +       err = 1; /* handled */
23765 +       sbinfo = au_sbi(sb);
23766 +       switch (opt->type) {
23767 +       case Opt_udba:
23768 +               sbinfo->si_mntflags &= ~AuOptMask_UDBA;
23769 +               sbinfo->si_mntflags |= opt->udba;
23770 +               opts->given_udba |= opt->udba;
23771 +               break;
23772 +
23773 +       case Opt_plink:
23774 +               au_opt_set(sbinfo->si_mntflags, PLINK);
23775 +               break;
23776 +       case Opt_noplink:
23777 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
23778 +                       au_plink_put(sb, /*verbose*/1);
23779 +               au_opt_clr(sbinfo->si_mntflags, PLINK);
23780 +               break;
23781 +       case Opt_list_plink:
23782 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
23783 +                       au_plink_list(sb);
23784 +               break;
23785 +
23786 +       case Opt_dio:
23787 +               au_opt_set(sbinfo->si_mntflags, DIO);
23788 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
23789 +               break;
23790 +       case Opt_nodio:
23791 +               au_opt_clr(sbinfo->si_mntflags, DIO);
23792 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
23793 +               break;
23794 +
23795 +       case Opt_fhsm_sec:
23796 +               au_fhsm_set(sbinfo, opt->fhsm_second);
23797 +               break;
23798 +
23799 +       case Opt_diropq_a:
23800 +               au_opt_set(sbinfo->si_mntflags, ALWAYS_DIROPQ);
23801 +               break;
23802 +       case Opt_diropq_w:
23803 +               au_opt_clr(sbinfo->si_mntflags, ALWAYS_DIROPQ);
23804 +               break;
23805 +
23806 +       case Opt_warn_perm:
23807 +               au_opt_set(sbinfo->si_mntflags, WARN_PERM);
23808 +               break;
23809 +       case Opt_nowarn_perm:
23810 +               au_opt_clr(sbinfo->si_mntflags, WARN_PERM);
23811 +               break;
23812 +
23813 +       case Opt_refrof:
23814 +               au_opt_set(sbinfo->si_mntflags, REFROF);
23815 +               break;
23816 +       case Opt_norefrof:
23817 +               au_opt_clr(sbinfo->si_mntflags, REFROF);
23818 +               break;
23819 +
23820 +       case Opt_verbose:
23821 +               au_opt_set(sbinfo->si_mntflags, VERBOSE);
23822 +               break;
23823 +       case Opt_noverbose:
23824 +               au_opt_clr(sbinfo->si_mntflags, VERBOSE);
23825 +               break;
23826 +
23827 +       case Opt_sum:
23828 +               au_opt_set(sbinfo->si_mntflags, SUM);
23829 +               break;
23830 +       case Opt_wsum:
23831 +               au_opt_clr(sbinfo->si_mntflags, SUM);
23832 +               au_opt_set(sbinfo->si_mntflags, SUM_W);
23833 +       case Opt_nosum:
23834 +               au_opt_clr(sbinfo->si_mntflags, SUM);
23835 +               au_opt_clr(sbinfo->si_mntflags, SUM_W);
23836 +               break;
23837 +
23838 +       case Opt_wbr_create:
23839 +               err = au_opt_wbr_create(sb, &opt->wbr_create);
23840 +               break;
23841 +       case Opt_wbr_copyup:
23842 +               sbinfo->si_wbr_copyup = opt->wbr_copyup;
23843 +               sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + opt->wbr_copyup;
23844 +               break;
23845 +
23846 +       case Opt_dirwh:
23847 +               sbinfo->si_dirwh = opt->dirwh;
23848 +               break;
23849 +
23850 +       case Opt_rdcache:
23851 +               sbinfo->si_rdcache
23852 +                       = msecs_to_jiffies(opt->rdcache * MSEC_PER_SEC);
23853 +               break;
23854 +       case Opt_rdblk:
23855 +               sbinfo->si_rdblk = opt->rdblk;
23856 +               break;
23857 +       case Opt_rdblk_def:
23858 +               sbinfo->si_rdblk = AUFS_RDBLK_DEF;
23859 +               break;
23860 +       case Opt_rdhash:
23861 +               sbinfo->si_rdhash = opt->rdhash;
23862 +               break;
23863 +       case Opt_rdhash_def:
23864 +               sbinfo->si_rdhash = AUFS_RDHASH_DEF;
23865 +               break;
23866 +
23867 +       case Opt_shwh:
23868 +               au_opt_set(sbinfo->si_mntflags, SHWH);
23869 +               break;
23870 +       case Opt_noshwh:
23871 +               au_opt_clr(sbinfo->si_mntflags, SHWH);
23872 +               break;
23873 +
23874 +       case Opt_dirperm1:
23875 +               au_opt_set(sbinfo->si_mntflags, DIRPERM1);
23876 +               break;
23877 +       case Opt_nodirperm1:
23878 +               au_opt_clr(sbinfo->si_mntflags, DIRPERM1);
23879 +               break;
23880 +
23881 +       case Opt_trunc_xino:
23882 +               au_opt_set(sbinfo->si_mntflags, TRUNC_XINO);
23883 +               break;
23884 +       case Opt_notrunc_xino:
23885 +               au_opt_clr(sbinfo->si_mntflags, TRUNC_XINO);
23886 +               break;
23887 +
23888 +       case Opt_trunc_xino_path:
23889 +       case Opt_itrunc_xino:
23890 +               err = au_xino_trunc(sb, opt->xino_itrunc.bindex);
23891 +               if (!err)
23892 +                       err = 1;
23893 +               break;
23894 +
23895 +       case Opt_trunc_xib:
23896 +               au_fset_opts(opts->flags, TRUNC_XIB);
23897 +               break;
23898 +       case Opt_notrunc_xib:
23899 +               au_fclr_opts(opts->flags, TRUNC_XIB);
23900 +               break;
23901 +
23902 +       case Opt_acl:
23903 +               sb->s_flags |= MS_POSIXACL;
23904 +               break;
23905 +       case Opt_noacl:
23906 +               sb->s_flags &= ~MS_POSIXACL;
23907 +               break;
23908 +
23909 +       default:
23910 +               err = 0;
23911 +               break;
23912 +       }
23913 +
23914 +       return err;
23915 +}
23916 +
23917 +/*
23918 + * returns tri-state.
23919 + * plus: processed without an error
23920 + * zero: unprocessed
23921 + * minus: error
23922 + */
23923 +static int au_opt_br(struct super_block *sb, struct au_opt *opt,
23924 +                    struct au_opts *opts)
23925 +{
23926 +       int err, do_refresh;
23927 +
23928 +       err = 0;
23929 +       switch (opt->type) {
23930 +       case Opt_append:
23931 +               opt->add.bindex = au_sbend(sb) + 1;
23932 +               if (opt->add.bindex < 0)
23933 +                       opt->add.bindex = 0;
23934 +               goto add;
23935 +       case Opt_prepend:
23936 +               opt->add.bindex = 0;
23937 +       add: /* indented label */
23938 +       case Opt_add:
23939 +               err = au_br_add(sb, &opt->add,
23940 +                               au_ftest_opts(opts->flags, REMOUNT));
23941 +               if (!err) {
23942 +                       err = 1;
23943 +                       au_fset_opts(opts->flags, REFRESH);
23944 +               }
23945 +               break;
23946 +
23947 +       case Opt_del:
23948 +       case Opt_idel:
23949 +               err = au_br_del(sb, &opt->del,
23950 +                               au_ftest_opts(opts->flags, REMOUNT));
23951 +               if (!err) {
23952 +                       err = 1;
23953 +                       au_fset_opts(opts->flags, TRUNC_XIB);
23954 +                       au_fset_opts(opts->flags, REFRESH);
23955 +               }
23956 +               break;
23957 +
23958 +       case Opt_mod:
23959 +       case Opt_imod:
23960 +               err = au_br_mod(sb, &opt->mod,
23961 +                               au_ftest_opts(opts->flags, REMOUNT),
23962 +                               &do_refresh);
23963 +               if (!err) {
23964 +                       err = 1;
23965 +                       if (do_refresh)
23966 +                               au_fset_opts(opts->flags, REFRESH);
23967 +               }
23968 +               break;
23969 +       }
23970 +
23971 +       return err;
23972 +}
23973 +
23974 +static int au_opt_xino(struct super_block *sb, struct au_opt *opt,
23975 +                      struct au_opt_xino **opt_xino,
23976 +                      struct au_opts *opts)
23977 +{
23978 +       int err;
23979 +       aufs_bindex_t bend, bindex;
23980 +       struct dentry *root, *parent, *h_root;
23981 +
23982 +       err = 0;
23983 +       switch (opt->type) {
23984 +       case Opt_xino:
23985 +               err = au_xino_set(sb, &opt->xino,
23986 +                                 !!au_ftest_opts(opts->flags, REMOUNT));
23987 +               if (unlikely(err))
23988 +                       break;
23989 +
23990 +               *opt_xino = &opt->xino;
23991 +               au_xino_brid_set(sb, -1);
23992 +
23993 +               /* safe d_parent access */
23994 +               parent = opt->xino.file->f_path.dentry->d_parent;
23995 +               root = sb->s_root;
23996 +               bend = au_sbend(sb);
23997 +               for (bindex = 0; bindex <= bend; bindex++) {
23998 +                       h_root = au_h_dptr(root, bindex);
23999 +                       if (h_root == parent) {
24000 +                               au_xino_brid_set(sb, au_sbr_id(sb, bindex));
24001 +                               break;
24002 +                       }
24003 +               }
24004 +               break;
24005 +
24006 +       case Opt_noxino:
24007 +               au_xino_clr(sb);
24008 +               au_xino_brid_set(sb, -1);
24009 +               *opt_xino = (void *)-1;
24010 +               break;
24011 +       }
24012 +
24013 +       return err;
24014 +}
24015 +
24016 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
24017 +                  unsigned int pending)
24018 +{
24019 +       int err, fhsm;
24020 +       aufs_bindex_t bindex, bend;
24021 +       unsigned char do_plink, skip, do_free;
24022 +       struct au_branch *br;
24023 +       struct au_wbr *wbr;
24024 +       struct dentry *root;
24025 +       struct inode *dir, *h_dir;
24026 +       struct au_sbinfo *sbinfo;
24027 +       struct au_hinode *hdir;
24028 +
24029 +       SiMustAnyLock(sb);
24030 +
24031 +       sbinfo = au_sbi(sb);
24032 +       AuDebugOn(!(sbinfo->si_mntflags & AuOptMask_UDBA));
24033 +
24034 +       if (!(sb_flags & MS_RDONLY)) {
24035 +               if (unlikely(!au_br_writable(au_sbr_perm(sb, 0))))
24036 +                       pr_warn("first branch should be rw\n");
24037 +               if (unlikely(au_opt_test(sbinfo->si_mntflags, SHWH)))
24038 +                       pr_warn("shwh should be used with ro\n");
24039 +       }
24040 +
24041 +       if (au_opt_test((sbinfo->si_mntflags | pending), UDBA_HNOTIFY)
24042 +           && !au_opt_test(sbinfo->si_mntflags, XINO))
24043 +               pr_warn("udba=*notify requires xino\n");
24044 +
24045 +       if (au_opt_test(sbinfo->si_mntflags, DIRPERM1))
24046 +               pr_warn("dirperm1 breaks the protection"
24047 +                       " by the permission bits on the lower branch\n");
24048 +
24049 +       err = 0;
24050 +       fhsm = 0;
24051 +       root = sb->s_root;
24052 +       dir = root->d_inode;
24053 +       do_plink = !!au_opt_test(sbinfo->si_mntflags, PLINK);
24054 +       bend = au_sbend(sb);
24055 +       for (bindex = 0; !err && bindex <= bend; bindex++) {
24056 +               skip = 0;
24057 +               h_dir = au_h_iptr(dir, bindex);
24058 +               br = au_sbr(sb, bindex);
24059 +
24060 +               if ((br->br_perm & AuBrAttr_ICEX)
24061 +                   && !h_dir->i_op->listxattr)
24062 +                       br->br_perm &= ~AuBrAttr_ICEX;
24063 +#if 0
24064 +               if ((br->br_perm & AuBrAttr_ICEX_SEC)
24065 +                   && (au_br_sb(br)->s_flags & MS_NOSEC))
24066 +                       br->br_perm &= ~AuBrAttr_ICEX_SEC;
24067 +#endif
24068 +
24069 +               do_free = 0;
24070 +               wbr = br->br_wbr;
24071 +               if (wbr)
24072 +                       wbr_wh_read_lock(wbr);
24073 +
24074 +               if (!au_br_writable(br->br_perm)) {
24075 +                       do_free = !!wbr;
24076 +                       skip = (!wbr
24077 +                               || (!wbr->wbr_whbase
24078 +                                   && !wbr->wbr_plink
24079 +                                   && !wbr->wbr_orph));
24080 +               } else if (!au_br_wh_linkable(br->br_perm)) {
24081 +                       /* skip = (!br->br_whbase && !br->br_orph); */
24082 +                       skip = (!wbr || !wbr->wbr_whbase);
24083 +                       if (skip && wbr) {
24084 +                               if (do_plink)
24085 +                                       skip = !!wbr->wbr_plink;
24086 +                               else
24087 +                                       skip = !wbr->wbr_plink;
24088 +                       }
24089 +               } else {
24090 +                       /* skip = (br->br_whbase && br->br_ohph); */
24091 +                       skip = (wbr && wbr->wbr_whbase);
24092 +                       if (skip) {
24093 +                               if (do_plink)
24094 +                                       skip = !!wbr->wbr_plink;
24095 +                               else
24096 +                                       skip = !wbr->wbr_plink;
24097 +                       }
24098 +               }
24099 +               if (wbr)
24100 +                       wbr_wh_read_unlock(wbr);
24101 +
24102 +               if (au_br_fhsm(br->br_perm)) {
24103 +                       fhsm++;
24104 +                       AuDebugOn(!br->br_fhsm);
24105 +               }
24106 +
24107 +               if (skip)
24108 +                       continue;
24109 +
24110 +               hdir = au_hi(dir, bindex);
24111 +               au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
24112 +               if (wbr)
24113 +                       wbr_wh_write_lock(wbr);
24114 +               err = au_wh_init(br, sb);
24115 +               if (wbr)
24116 +                       wbr_wh_write_unlock(wbr);
24117 +               au_hn_imtx_unlock(hdir);
24118 +
24119 +               if (!err && do_free) {
24120 +                       kfree(wbr);
24121 +                       br->br_wbr = NULL;
24122 +               }
24123 +       }
24124 +
24125 +       if (fhsm >= 2) {
24126 +               au_fset_si(sbinfo, FHSM);
24127 +               for (bindex = bend; bindex >= 0; bindex--) {
24128 +                       br = au_sbr(sb, bindex);
24129 +                       if (au_br_fhsm(br->br_perm)) {
24130 +                               au_fhsm_set_bottom(sb, bindex);
24131 +                               break;
24132 +                       }
24133 +               }
24134 +       } else {
24135 +               au_fclr_si(sbinfo, FHSM);
24136 +               au_fhsm_set_bottom(sb, -1);
24137 +       }
24138 +
24139 +       return err;
24140 +}
24141 +
24142 +int au_opts_mount(struct super_block *sb, struct au_opts *opts)
24143 +{
24144 +       int err;
24145 +       unsigned int tmp;
24146 +       aufs_bindex_t bindex, bend;
24147 +       struct au_opt *opt;
24148 +       struct au_opt_xino *opt_xino, xino;
24149 +       struct au_sbinfo *sbinfo;
24150 +       struct au_branch *br;
24151 +       struct inode *dir;
24152 +
24153 +       SiMustWriteLock(sb);
24154 +
24155 +       err = 0;
24156 +       opt_xino = NULL;
24157 +       opt = opts->opt;
24158 +       while (err >= 0 && opt->type != Opt_tail)
24159 +               err = au_opt_simple(sb, opt++, opts);
24160 +       if (err > 0)
24161 +               err = 0;
24162 +       else if (unlikely(err < 0))
24163 +               goto out;
24164 +
24165 +       /* disable xino and udba temporary */
24166 +       sbinfo = au_sbi(sb);
24167 +       tmp = sbinfo->si_mntflags;
24168 +       au_opt_clr(sbinfo->si_mntflags, XINO);
24169 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_REVAL);
24170 +
24171 +       opt = opts->opt;
24172 +       while (err >= 0 && opt->type != Opt_tail)
24173 +               err = au_opt_br(sb, opt++, opts);
24174 +       if (err > 0)
24175 +               err = 0;
24176 +       else if (unlikely(err < 0))
24177 +               goto out;
24178 +
24179 +       bend = au_sbend(sb);
24180 +       if (unlikely(bend < 0)) {
24181 +               err = -EINVAL;
24182 +               pr_err("no branches\n");
24183 +               goto out;
24184 +       }
24185 +
24186 +       if (au_opt_test(tmp, XINO))
24187 +               au_opt_set(sbinfo->si_mntflags, XINO);
24188 +       opt = opts->opt;
24189 +       while (!err && opt->type != Opt_tail)
24190 +               err = au_opt_xino(sb, opt++, &opt_xino, opts);
24191 +       if (unlikely(err))
24192 +               goto out;
24193 +
24194 +       err = au_opts_verify(sb, sb->s_flags, tmp);
24195 +       if (unlikely(err))
24196 +               goto out;
24197 +
24198 +       /* restore xino */
24199 +       if (au_opt_test(tmp, XINO) && !opt_xino) {
24200 +               xino.file = au_xino_def(sb);
24201 +               err = PTR_ERR(xino.file);
24202 +               if (IS_ERR(xino.file))
24203 +                       goto out;
24204 +
24205 +               err = au_xino_set(sb, &xino, /*remount*/0);
24206 +               fput(xino.file);
24207 +               if (unlikely(err))
24208 +                       goto out;
24209 +       }
24210 +
24211 +       /* restore udba */
24212 +       tmp &= AuOptMask_UDBA;
24213 +       sbinfo->si_mntflags &= ~AuOptMask_UDBA;
24214 +       sbinfo->si_mntflags |= tmp;
24215 +       bend = au_sbend(sb);
24216 +       for (bindex = 0; bindex <= bend; bindex++) {
24217 +               br = au_sbr(sb, bindex);
24218 +               err = au_hnotify_reset_br(tmp, br, br->br_perm);
24219 +               if (unlikely(err))
24220 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
24221 +                               bindex, err);
24222 +               /* go on even if err */
24223 +       }
24224 +       if (au_opt_test(tmp, UDBA_HNOTIFY)) {
24225 +               dir = sb->s_root->d_inode;
24226 +               au_hn_reset(dir, au_hi_flags(dir, /*isdir*/1) & ~AuHi_XINO);
24227 +       }
24228 +
24229 +out:
24230 +       return err;
24231 +}
24232 +
24233 +int au_opts_remount(struct super_block *sb, struct au_opts *opts)
24234 +{
24235 +       int err, rerr;
24236 +       struct inode *dir;
24237 +       struct au_opt_xino *opt_xino;
24238 +       struct au_opt *opt;
24239 +       struct au_sbinfo *sbinfo;
24240 +
24241 +       SiMustWriteLock(sb);
24242 +
24243 +       dir = sb->s_root->d_inode;
24244 +       sbinfo = au_sbi(sb);
24245 +       err = 0;
24246 +       opt_xino = NULL;
24247 +       opt = opts->opt;
24248 +       while (err >= 0 && opt->type != Opt_tail) {
24249 +               err = au_opt_simple(sb, opt, opts);
24250 +               if (!err)
24251 +                       err = au_opt_br(sb, opt, opts);
24252 +               if (!err)
24253 +                       err = au_opt_xino(sb, opt, &opt_xino, opts);
24254 +               opt++;
24255 +       }
24256 +       if (err > 0)
24257 +               err = 0;
24258 +       AuTraceErr(err);
24259 +       /* go on even err */
24260 +
24261 +       rerr = au_opts_verify(sb, opts->sb_flags, /*pending*/0);
24262 +       if (unlikely(rerr && !err))
24263 +               err = rerr;
24264 +
24265 +       if (au_ftest_opts(opts->flags, TRUNC_XIB)) {
24266 +               rerr = au_xib_trunc(sb);
24267 +               if (unlikely(rerr && !err))
24268 +                       err = rerr;
24269 +       }
24270 +
24271 +       /* will be handled by the caller */
24272 +       if (!au_ftest_opts(opts->flags, REFRESH)
24273 +           && (opts->given_udba || au_opt_test(sbinfo->si_mntflags, XINO)))
24274 +               au_fset_opts(opts->flags, REFRESH);
24275 +
24276 +       AuDbg("status 0x%x\n", opts->flags);
24277 +       return err;
24278 +}
24279 +
24280 +/* ---------------------------------------------------------------------- */
24281 +
24282 +unsigned int au_opt_udba(struct super_block *sb)
24283 +{
24284 +       return au_mntflags(sb) & AuOptMask_UDBA;
24285 +}
24286 diff -urN /usr/share/empty/fs/aufs/opts.h linux/fs/aufs/opts.h
24287 --- /usr/share/empty/fs/aufs/opts.h     1970-01-01 01:00:00.000000000 +0100
24288 +++ linux/fs/aufs/opts.h        2015-03-27 21:56:35.466795000 +0100
24289 @@ -0,0 +1,213 @@
24290 +/*
24291 + * Copyright (C) 2005-2015 Junjiro R. Okajima
24292 + *
24293 + * This program, aufs is free software; you can redistribute it and/or modify
24294 + * it under the terms of the GNU General Public License as published by
24295 + * the Free Software Foundation; either version 2 of the License, or
24296 + * (at your option) any later version.
24297 + *
24298 + * This program is distributed in the hope that it will be useful,
24299 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24300 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24301 + * GNU General Public License for more details.
24302 + *
24303 + * You should have received a copy of the GNU General Public License
24304 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24305 + */
24306 +
24307 +/*
24308 + * mount options/flags
24309 + */
24310 +
24311 +#ifndef __AUFS_OPTS_H__
24312 +#define __AUFS_OPTS_H__
24313 +
24314 +#ifdef __KERNEL__
24315 +
24316 +#include <linux/path.h>
24317 +#include "branch.h"
24318 +
24319 +struct file;
24320 +struct super_block;
24321 +
24322 +/* ---------------------------------------------------------------------- */
24323 +
24324 +/* mount flags */
24325 +#define AuOpt_XINO             1               /* external inode number bitmap
24326 +                                                  and translation table */
24327 +#define AuOpt_TRUNC_XINO       (1 << 1)        /* truncate xino files */
24328 +#define AuOpt_UDBA_NONE                (1 << 2)        /* users direct branch access */
24329 +#define AuOpt_UDBA_REVAL       (1 << 3)
24330 +#define AuOpt_UDBA_HNOTIFY     (1 << 4)
24331 +#define AuOpt_SHWH             (1 << 5)        /* show whiteout */
24332 +#define AuOpt_PLINK            (1 << 6)        /* pseudo-link */
24333 +#define AuOpt_DIRPERM1         (1 << 7)        /* ignore the lower dir's perm
24334 +                                                  bits */
24335 +#define AuOpt_REFROF           (1 << 8)        /* unimplemented */
24336 +#define AuOpt_ALWAYS_DIROPQ    (1 << 9)        /* policy to creating diropq */
24337 +#define AuOpt_SUM              (1 << 10)       /* summation for statfs(2) */
24338 +#define AuOpt_SUM_W            (1 << 11)       /* unimplemented */
24339 +#define AuOpt_WARN_PERM                (1 << 12)       /* warn when add-branch */
24340 +#define AuOpt_VERBOSE          (1 << 13)       /* busy inode when del-branch */
24341 +#define AuOpt_DIO              (1 << 14)       /* direct io */
24342 +
24343 +#ifndef CONFIG_AUFS_HNOTIFY
24344 +#undef AuOpt_UDBA_HNOTIFY
24345 +#define AuOpt_UDBA_HNOTIFY     0
24346 +#endif
24347 +#ifndef CONFIG_AUFS_SHWH
24348 +#undef AuOpt_SHWH
24349 +#define AuOpt_SHWH             0
24350 +#endif
24351 +
24352 +#define AuOpt_Def      (AuOpt_XINO \
24353 +                        | AuOpt_UDBA_REVAL \
24354 +                        | AuOpt_PLINK \
24355 +                        /* | AuOpt_DIRPERM1 */ \
24356 +                        | AuOpt_WARN_PERM)
24357 +#define AuOptMask_UDBA (AuOpt_UDBA_NONE \
24358 +                        | AuOpt_UDBA_REVAL \
24359 +                        | AuOpt_UDBA_HNOTIFY)
24360 +
24361 +#define au_opt_test(flags, name)       (flags & AuOpt_##name)
24362 +#define au_opt_set(flags, name) do { \
24363 +       BUILD_BUG_ON(AuOpt_##name & AuOptMask_UDBA); \
24364 +       ((flags) |= AuOpt_##name); \
24365 +} while (0)
24366 +#define au_opt_set_udba(flags, name) do { \
24367 +       (flags) &= ~AuOptMask_UDBA; \
24368 +       ((flags) |= AuOpt_##name); \
24369 +} while (0)
24370 +#define au_opt_clr(flags, name) do { \
24371 +       ((flags) &= ~AuOpt_##name); \
24372 +} while (0)
24373 +
24374 +static inline unsigned int au_opts_plink(unsigned int mntflags)
24375 +{
24376 +#ifdef CONFIG_PROC_FS
24377 +       return mntflags;
24378 +#else
24379 +       return mntflags & ~AuOpt_PLINK;
24380 +#endif
24381 +}
24382 +
24383 +/* ---------------------------------------------------------------------- */
24384 +
24385 +/* policies to select one among multiple writable branches */
24386 +enum {
24387 +       AuWbrCreate_TDP,        /* top down parent */
24388 +       AuWbrCreate_RR,         /* round robin */
24389 +       AuWbrCreate_MFS,        /* most free space */
24390 +       AuWbrCreate_MFSV,       /* mfs with seconds */
24391 +       AuWbrCreate_MFSRR,      /* mfs then rr */
24392 +       AuWbrCreate_MFSRRV,     /* mfs then rr with seconds */
24393 +       AuWbrCreate_PMFS,       /* parent and mfs */
24394 +       AuWbrCreate_PMFSV,      /* parent and mfs with seconds */
24395 +       AuWbrCreate_PMFSRR,     /* parent, mfs and round-robin */
24396 +       AuWbrCreate_PMFSRRV,    /* plus seconds */
24397 +
24398 +       AuWbrCreate_Def = AuWbrCreate_TDP
24399 +};
24400 +
24401 +enum {
24402 +       AuWbrCopyup_TDP,        /* top down parent */
24403 +       AuWbrCopyup_BUP,        /* bottom up parent */
24404 +       AuWbrCopyup_BU,         /* bottom up */
24405 +
24406 +       AuWbrCopyup_Def = AuWbrCopyup_TDP
24407 +};
24408 +
24409 +/* ---------------------------------------------------------------------- */
24410 +
24411 +struct au_opt_add {
24412 +       aufs_bindex_t   bindex;
24413 +       char            *pathname;
24414 +       int             perm;
24415 +       struct path     path;
24416 +};
24417 +
24418 +struct au_opt_del {
24419 +       char            *pathname;
24420 +       struct path     h_path;
24421 +};
24422 +
24423 +struct au_opt_mod {
24424 +       char            *path;
24425 +       int             perm;
24426 +       struct dentry   *h_root;
24427 +};
24428 +
24429 +struct au_opt_xino {
24430 +       char            *path;
24431 +       struct file     *file;
24432 +};
24433 +
24434 +struct au_opt_xino_itrunc {
24435 +       aufs_bindex_t   bindex;
24436 +};
24437 +
24438 +struct au_opt_wbr_create {
24439 +       int                     wbr_create;
24440 +       int                     mfs_second;
24441 +       unsigned long long      mfsrr_watermark;
24442 +};
24443 +
24444 +struct au_opt {
24445 +       int type;
24446 +       union {
24447 +               struct au_opt_xino      xino;
24448 +               struct au_opt_xino_itrunc xino_itrunc;
24449 +               struct au_opt_add       add;
24450 +               struct au_opt_del       del;
24451 +               struct au_opt_mod       mod;
24452 +               int                     dirwh;
24453 +               int                     rdcache;
24454 +               unsigned int            rdblk;
24455 +               unsigned int            rdhash;
24456 +               int                     udba;
24457 +               struct au_opt_wbr_create wbr_create;
24458 +               int                     wbr_copyup;
24459 +               unsigned int            fhsm_second;
24460 +       };
24461 +};
24462 +
24463 +/* opts flags */
24464 +#define AuOpts_REMOUNT         1
24465 +#define AuOpts_REFRESH         (1 << 1)
24466 +#define AuOpts_TRUNC_XIB       (1 << 2)
24467 +#define AuOpts_REFRESH_DYAOP   (1 << 3)
24468 +#define au_ftest_opts(flags, name)     ((flags) & AuOpts_##name)
24469 +#define au_fset_opts(flags, name) \
24470 +       do { (flags) |= AuOpts_##name; } while (0)
24471 +#define au_fclr_opts(flags, name) \
24472 +       do { (flags) &= ~AuOpts_##name; } while (0)
24473 +
24474 +struct au_opts {
24475 +       struct au_opt   *opt;
24476 +       int             max_opt;
24477 +
24478 +       unsigned int    given_udba;
24479 +       unsigned int    flags;
24480 +       unsigned long   sb_flags;
24481 +};
24482 +
24483 +/* ---------------------------------------------------------------------- */
24484 +
24485 +void au_optstr_br_perm(au_br_perm_str_t *str, int perm);
24486 +const char *au_optstr_udba(int udba);
24487 +const char *au_optstr_wbr_copyup(int wbr_copyup);
24488 +const char *au_optstr_wbr_create(int wbr_create);
24489 +
24490 +void au_opts_free(struct au_opts *opts);
24491 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts);
24492 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
24493 +                  unsigned int pending);
24494 +int au_opts_mount(struct super_block *sb, struct au_opts *opts);
24495 +int au_opts_remount(struct super_block *sb, struct au_opts *opts);
24496 +
24497 +unsigned int au_opt_udba(struct super_block *sb);
24498 +
24499 +/* ---------------------------------------------------------------------- */
24500 +
24501 +#endif /* __KERNEL__ */
24502 +#endif /* __AUFS_OPTS_H__ */
24503 diff -urN /usr/share/empty/fs/aufs/plink.c linux/fs/aufs/plink.c
24504 --- /usr/share/empty/fs/aufs/plink.c    1970-01-01 01:00:00.000000000 +0100
24505 +++ linux/fs/aufs/plink.c       2015-03-27 21:56:35.466795000 +0100
24506 @@ -0,0 +1,532 @@
24507 +/*
24508 + * Copyright (C) 2005-2015 Junjiro R. Okajima
24509 + *
24510 + * This program, aufs is free software; you can redistribute it and/or modify
24511 + * it under the terms of the GNU General Public License as published by
24512 + * the Free Software Foundation; either version 2 of the License, or
24513 + * (at your option) any later version.
24514 + *
24515 + * This program is distributed in the hope that it will be useful,
24516 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24517 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24518 + * GNU General Public License for more details.
24519 + *
24520 + * You should have received a copy of the GNU General Public License
24521 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24522 + */
24523 +
24524 +/*
24525 + * pseudo-link
24526 + */
24527 +
24528 +#include "aufs.h"
24529 +
24530 +/*
24531 + * the pseudo-link maintenance mode.
24532 + * during a user process maintains the pseudo-links,
24533 + * prohibit adding a new plink and branch manipulation.
24534 + *
24535 + * Flags
24536 + * NOPLM:
24537 + *     For entry functions which will handle plink, and i_mutex is already held
24538 + *     in VFS.
24539 + *     They cannot wait and should return an error at once.
24540 + *     Callers has to check the error.
24541 + * NOPLMW:
24542 + *     For entry functions which will handle plink, but i_mutex is not held
24543 + *     in VFS.
24544 + *     They can wait the plink maintenance mode to finish.
24545 + *
24546 + * They behave like F_SETLK and F_SETLKW.
24547 + * If the caller never handle plink, then both flags are unnecessary.
24548 + */
24549 +
24550 +int au_plink_maint(struct super_block *sb, int flags)
24551 +{
24552 +       int err;
24553 +       pid_t pid, ppid;
24554 +       struct au_sbinfo *sbi;
24555 +
24556 +       SiMustAnyLock(sb);
24557 +
24558 +       err = 0;
24559 +       if (!au_opt_test(au_mntflags(sb), PLINK))
24560 +               goto out;
24561 +
24562 +       sbi = au_sbi(sb);
24563 +       pid = sbi->si_plink_maint_pid;
24564 +       if (!pid || pid == current->pid)
24565 +               goto out;
24566 +
24567 +       /* todo: it highly depends upon /sbin/mount.aufs */
24568 +       rcu_read_lock();
24569 +       ppid = task_pid_vnr(rcu_dereference(current->real_parent));
24570 +       rcu_read_unlock();
24571 +       if (pid == ppid)
24572 +               goto out;
24573 +
24574 +       if (au_ftest_lock(flags, NOPLMW)) {
24575 +               /* if there is no i_mutex lock in VFS, we don't need to wait */
24576 +               /* AuDebugOn(!lockdep_depth(current)); */
24577 +               while (sbi->si_plink_maint_pid) {
24578 +                       si_read_unlock(sb);
24579 +                       /* gave up wake_up_bit() */
24580 +                       wait_event(sbi->si_plink_wq, !sbi->si_plink_maint_pid);
24581 +
24582 +                       if (au_ftest_lock(flags, FLUSH))
24583 +                               au_nwt_flush(&sbi->si_nowait);
24584 +                       si_noflush_read_lock(sb);
24585 +               }
24586 +       } else if (au_ftest_lock(flags, NOPLM)) {
24587 +               AuDbg("ppid %d, pid %d\n", ppid, pid);
24588 +               err = -EAGAIN;
24589 +       }
24590 +
24591 +out:
24592 +       return err;
24593 +}
24594 +
24595 +void au_plink_maint_leave(struct au_sbinfo *sbinfo)
24596 +{
24597 +       spin_lock(&sbinfo->si_plink_maint_lock);
24598 +       sbinfo->si_plink_maint_pid = 0;
24599 +       spin_unlock(&sbinfo->si_plink_maint_lock);
24600 +       wake_up_all(&sbinfo->si_plink_wq);
24601 +}
24602 +
24603 +int au_plink_maint_enter(struct super_block *sb)
24604 +{
24605 +       int err;
24606 +       struct au_sbinfo *sbinfo;
24607 +
24608 +       err = 0;
24609 +       sbinfo = au_sbi(sb);
24610 +       /* make sure i am the only one in this fs */
24611 +       si_write_lock(sb, AuLock_FLUSH);
24612 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
24613 +               spin_lock(&sbinfo->si_plink_maint_lock);
24614 +               if (!sbinfo->si_plink_maint_pid)
24615 +                       sbinfo->si_plink_maint_pid = current->pid;
24616 +               else
24617 +                       err = -EBUSY;
24618 +               spin_unlock(&sbinfo->si_plink_maint_lock);
24619 +       }
24620 +       si_write_unlock(sb);
24621 +
24622 +       return err;
24623 +}
24624 +
24625 +/* ---------------------------------------------------------------------- */
24626 +
24627 +#ifdef CONFIG_AUFS_DEBUG
24628 +void au_plink_list(struct super_block *sb)
24629 +{
24630 +       int i;
24631 +       struct au_sbinfo *sbinfo;
24632 +       struct hlist_head *plink_hlist;
24633 +       struct pseudo_link *plink;
24634 +
24635 +       SiMustAnyLock(sb);
24636 +
24637 +       sbinfo = au_sbi(sb);
24638 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
24639 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
24640 +
24641 +       for (i = 0; i < AuPlink_NHASH; i++) {
24642 +               plink_hlist = &sbinfo->si_plink[i].head;
24643 +               rcu_read_lock();
24644 +               hlist_for_each_entry_rcu(plink, plink_hlist, hlist)
24645 +                       AuDbg("%lu\n", plink->inode->i_ino);
24646 +               rcu_read_unlock();
24647 +       }
24648 +}
24649 +#endif
24650 +
24651 +/* is the inode pseudo-linked? */
24652 +int au_plink_test(struct inode *inode)
24653 +{
24654 +       int found, i;
24655 +       struct au_sbinfo *sbinfo;
24656 +       struct hlist_head *plink_hlist;
24657 +       struct pseudo_link *plink;
24658 +
24659 +       sbinfo = au_sbi(inode->i_sb);
24660 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
24661 +       AuDebugOn(!au_opt_test(au_mntflags(inode->i_sb), PLINK));
24662 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
24663 +
24664 +       found = 0;
24665 +       i = au_plink_hash(inode->i_ino);
24666 +       plink_hlist = &sbinfo->si_plink[i].head;
24667 +       rcu_read_lock();
24668 +       hlist_for_each_entry_rcu(plink, plink_hlist, hlist)
24669 +               if (plink->inode == inode) {
24670 +                       found = 1;
24671 +                       break;
24672 +               }
24673 +       rcu_read_unlock();
24674 +       return found;
24675 +}
24676 +
24677 +/* ---------------------------------------------------------------------- */
24678 +
24679 +/*
24680 + * generate a name for plink.
24681 + * the file will be stored under AUFS_WH_PLINKDIR.
24682 + */
24683 +/* 20 is max digits length of ulong 64 */
24684 +#define PLINK_NAME_LEN ((20 + 1) * 2)
24685 +
24686 +static int plink_name(char *name, int len, struct inode *inode,
24687 +                     aufs_bindex_t bindex)
24688 +{
24689 +       int rlen;
24690 +       struct inode *h_inode;
24691 +
24692 +       h_inode = au_h_iptr(inode, bindex);
24693 +       rlen = snprintf(name, len, "%lu.%lu", inode->i_ino, h_inode->i_ino);
24694 +       return rlen;
24695 +}
24696 +
24697 +struct au_do_plink_lkup_args {
24698 +       struct dentry **errp;
24699 +       struct qstr *tgtname;
24700 +       struct dentry *h_parent;
24701 +       struct au_branch *br;
24702 +};
24703 +
24704 +static struct dentry *au_do_plink_lkup(struct qstr *tgtname,
24705 +                                      struct dentry *h_parent,
24706 +                                      struct au_branch *br)
24707 +{
24708 +       struct dentry *h_dentry;
24709 +       struct mutex *h_mtx;
24710 +
24711 +       h_mtx = &h_parent->d_inode->i_mutex;
24712 +       mutex_lock_nested(h_mtx, AuLsc_I_CHILD2);
24713 +       h_dentry = vfsub_lkup_one(tgtname, h_parent);
24714 +       mutex_unlock(h_mtx);
24715 +       return h_dentry;
24716 +}
24717 +
24718 +static void au_call_do_plink_lkup(void *args)
24719 +{
24720 +       struct au_do_plink_lkup_args *a = args;
24721 +       *a->errp = au_do_plink_lkup(a->tgtname, a->h_parent, a->br);
24722 +}
24723 +
24724 +/* lookup the plink-ed @inode under the branch at @bindex */
24725 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex)
24726 +{
24727 +       struct dentry *h_dentry, *h_parent;
24728 +       struct au_branch *br;
24729 +       struct inode *h_dir;
24730 +       int wkq_err;
24731 +       char a[PLINK_NAME_LEN];
24732 +       struct qstr tgtname = QSTR_INIT(a, 0);
24733 +
24734 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
24735 +
24736 +       br = au_sbr(inode->i_sb, bindex);
24737 +       h_parent = br->br_wbr->wbr_plink;
24738 +       h_dir = h_parent->d_inode;
24739 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
24740 +
24741 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
24742 +               struct au_do_plink_lkup_args args = {
24743 +                       .errp           = &h_dentry,
24744 +                       .tgtname        = &tgtname,
24745 +                       .h_parent       = h_parent,
24746 +                       .br             = br
24747 +               };
24748 +
24749 +               wkq_err = au_wkq_wait(au_call_do_plink_lkup, &args);
24750 +               if (unlikely(wkq_err))
24751 +                       h_dentry = ERR_PTR(wkq_err);
24752 +       } else
24753 +               h_dentry = au_do_plink_lkup(&tgtname, h_parent, br);
24754 +
24755 +       return h_dentry;
24756 +}
24757 +
24758 +/* create a pseudo-link */
24759 +static int do_whplink(struct qstr *tgt, struct dentry *h_parent,
24760 +                     struct dentry *h_dentry, struct au_branch *br)
24761 +{
24762 +       int err;
24763 +       struct path h_path = {
24764 +               .mnt = au_br_mnt(br)
24765 +       };
24766 +       struct inode *h_dir, *delegated;
24767 +
24768 +       h_dir = h_parent->d_inode;
24769 +       mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_CHILD2);
24770 +again:
24771 +       h_path.dentry = vfsub_lkup_one(tgt, h_parent);
24772 +       err = PTR_ERR(h_path.dentry);
24773 +       if (IS_ERR(h_path.dentry))
24774 +               goto out;
24775 +
24776 +       err = 0;
24777 +       /* wh.plink dir is not monitored */
24778 +       /* todo: is it really safe? */
24779 +       if (h_path.dentry->d_inode
24780 +           && h_path.dentry->d_inode != h_dentry->d_inode) {
24781 +               delegated = NULL;
24782 +               err = vfsub_unlink(h_dir, &h_path, &delegated, /*force*/0);
24783 +               if (unlikely(err == -EWOULDBLOCK)) {
24784 +                       pr_warn("cannot retry for NFSv4 delegation"
24785 +                               " for an internal unlink\n");
24786 +                       iput(delegated);
24787 +               }
24788 +               dput(h_path.dentry);
24789 +               h_path.dentry = NULL;
24790 +               if (!err)
24791 +                       goto again;
24792 +       }
24793 +       if (!err && !h_path.dentry->d_inode) {
24794 +               delegated = NULL;
24795 +               err = vfsub_link(h_dentry, h_dir, &h_path, &delegated);
24796 +               if (unlikely(err == -EWOULDBLOCK)) {
24797 +                       pr_warn("cannot retry for NFSv4 delegation"
24798 +                               " for an internal link\n");
24799 +                       iput(delegated);
24800 +               }
24801 +       }
24802 +       dput(h_path.dentry);
24803 +
24804 +out:
24805 +       mutex_unlock(&h_dir->i_mutex);
24806 +       return err;
24807 +}
24808 +
24809 +struct do_whplink_args {
24810 +       int *errp;
24811 +       struct qstr *tgt;
24812 +       struct dentry *h_parent;
24813 +       struct dentry *h_dentry;
24814 +       struct au_branch *br;
24815 +};
24816 +
24817 +static void call_do_whplink(void *args)
24818 +{
24819 +       struct do_whplink_args *a = args;
24820 +       *a->errp = do_whplink(a->tgt, a->h_parent, a->h_dentry, a->br);
24821 +}
24822 +
24823 +static int whplink(struct dentry *h_dentry, struct inode *inode,
24824 +                  aufs_bindex_t bindex, struct au_branch *br)
24825 +{
24826 +       int err, wkq_err;
24827 +       struct au_wbr *wbr;
24828 +       struct dentry *h_parent;
24829 +       struct inode *h_dir;
24830 +       char a[PLINK_NAME_LEN];
24831 +       struct qstr tgtname = QSTR_INIT(a, 0);
24832 +
24833 +       wbr = au_sbr(inode->i_sb, bindex)->br_wbr;
24834 +       h_parent = wbr->wbr_plink;
24835 +       h_dir = h_parent->d_inode;
24836 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
24837 +
24838 +       /* always superio. */
24839 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
24840 +               struct do_whplink_args args = {
24841 +                       .errp           = &err,
24842 +                       .tgt            = &tgtname,
24843 +                       .h_parent       = h_parent,
24844 +                       .h_dentry       = h_dentry,
24845 +                       .br             = br
24846 +               };
24847 +               wkq_err = au_wkq_wait(call_do_whplink, &args);
24848 +               if (unlikely(wkq_err))
24849 +                       err = wkq_err;
24850 +       } else
24851 +               err = do_whplink(&tgtname, h_parent, h_dentry, br);
24852 +
24853 +       return err;
24854 +}
24855 +
24856 +/* free a single plink */
24857 +static void do_put_plink(struct pseudo_link *plink, int do_del)
24858 +{
24859 +       if (do_del)
24860 +               hlist_del(&plink->hlist);
24861 +       iput(plink->inode);
24862 +       kfree(plink);
24863 +}
24864 +
24865 +static void do_put_plink_rcu(struct rcu_head *rcu)
24866 +{
24867 +       struct pseudo_link *plink;
24868 +
24869 +       plink = container_of(rcu, struct pseudo_link, rcu);
24870 +       iput(plink->inode);
24871 +       kfree(plink);
24872 +}
24873 +
24874 +/*
24875 + * create a new pseudo-link for @h_dentry on @bindex.
24876 + * the linked inode is held in aufs @inode.
24877 + */
24878 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
24879 +                    struct dentry *h_dentry)
24880 +{
24881 +       struct super_block *sb;
24882 +       struct au_sbinfo *sbinfo;
24883 +       struct hlist_head *plink_hlist;
24884 +       struct pseudo_link *plink, *tmp;
24885 +       struct au_sphlhead *sphl;
24886 +       int found, err, cnt, i;
24887 +
24888 +       sb = inode->i_sb;
24889 +       sbinfo = au_sbi(sb);
24890 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
24891 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
24892 +
24893 +       found = au_plink_test(inode);
24894 +       if (found)
24895 +               return;
24896 +
24897 +       i = au_plink_hash(inode->i_ino);
24898 +       sphl = sbinfo->si_plink + i;
24899 +       plink_hlist = &sphl->head;
24900 +       tmp = kmalloc(sizeof(*plink), GFP_NOFS);
24901 +       if (tmp)
24902 +               tmp->inode = au_igrab(inode);
24903 +       else {
24904 +               err = -ENOMEM;
24905 +               goto out;
24906 +       }
24907 +
24908 +       spin_lock(&sphl->spin);
24909 +       hlist_for_each_entry(plink, plink_hlist, hlist) {
24910 +               if (plink->inode == inode) {
24911 +                       found = 1;
24912 +                       break;
24913 +               }
24914 +       }
24915 +       if (!found)
24916 +               hlist_add_head_rcu(&tmp->hlist, plink_hlist);
24917 +       spin_unlock(&sphl->spin);
24918 +       if (!found) {
24919 +               cnt = au_sphl_count(sphl);
24920 +#define msg "unexpectedly unblanced or too many pseudo-links"
24921 +               if (cnt > AUFS_PLINK_WARN)
24922 +                       AuWarn1(msg ", %d\n", cnt);
24923 +#undef msg
24924 +               err = whplink(h_dentry, inode, bindex, au_sbr(sb, bindex));
24925 +       } else {
24926 +               do_put_plink(tmp, 0);
24927 +               return;
24928 +       }
24929 +
24930 +out:
24931 +       if (unlikely(err)) {
24932 +               pr_warn("err %d, damaged pseudo link.\n", err);
24933 +               if (tmp) {
24934 +                       au_sphl_del_rcu(&tmp->hlist, sphl);
24935 +                       call_rcu(&tmp->rcu, do_put_plink_rcu);
24936 +               }
24937 +       }
24938 +}
24939 +
24940 +/* free all plinks */
24941 +void au_plink_put(struct super_block *sb, int verbose)
24942 +{
24943 +       int i, warned;
24944 +       struct au_sbinfo *sbinfo;
24945 +       struct hlist_head *plink_hlist;
24946 +       struct hlist_node *tmp;
24947 +       struct pseudo_link *plink;
24948 +
24949 +       SiMustWriteLock(sb);
24950 +
24951 +       sbinfo = au_sbi(sb);
24952 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
24953 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
24954 +
24955 +       /* no spin_lock since sbinfo is write-locked */
24956 +       warned = 0;
24957 +       for (i = 0; i < AuPlink_NHASH; i++) {
24958 +               plink_hlist = &sbinfo->si_plink[i].head;
24959 +               if (!warned && verbose && !hlist_empty(plink_hlist)) {
24960 +                       pr_warn("pseudo-link is not flushed");
24961 +                       warned = 1;
24962 +               }
24963 +               hlist_for_each_entry_safe(plink, tmp, plink_hlist, hlist)
24964 +                       do_put_plink(plink, 0);
24965 +               INIT_HLIST_HEAD(plink_hlist);
24966 +       }
24967 +}
24968 +
24969 +void au_plink_clean(struct super_block *sb, int verbose)
24970 +{
24971 +       struct dentry *root;
24972 +
24973 +       root = sb->s_root;
24974 +       aufs_write_lock(root);
24975 +       if (au_opt_test(au_mntflags(sb), PLINK))
24976 +               au_plink_put(sb, verbose);
24977 +       aufs_write_unlock(root);
24978 +}
24979 +
24980 +static int au_plink_do_half_refresh(struct inode *inode, aufs_bindex_t br_id)
24981 +{
24982 +       int do_put;
24983 +       aufs_bindex_t bstart, bend, bindex;
24984 +
24985 +       do_put = 0;
24986 +       bstart = au_ibstart(inode);
24987 +       bend = au_ibend(inode);
24988 +       if (bstart >= 0) {
24989 +               for (bindex = bstart; bindex <= bend; bindex++) {
24990 +                       if (!au_h_iptr(inode, bindex)
24991 +                           || au_ii_br_id(inode, bindex) != br_id)
24992 +                               continue;
24993 +                       au_set_h_iptr(inode, bindex, NULL, 0);
24994 +                       do_put = 1;
24995 +                       break;
24996 +               }
24997 +               if (do_put)
24998 +                       for (bindex = bstart; bindex <= bend; bindex++)
24999 +                               if (au_h_iptr(inode, bindex)) {
25000 +                                       do_put = 0;
25001 +                                       break;
25002 +                               }
25003 +       } else
25004 +               do_put = 1;
25005 +
25006 +       return do_put;
25007 +}
25008 +
25009 +/* free the plinks on a branch specified by @br_id */
25010 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id)
25011 +{
25012 +       struct au_sbinfo *sbinfo;
25013 +       struct hlist_head *plink_hlist;
25014 +       struct hlist_node *tmp;
25015 +       struct pseudo_link *plink;
25016 +       struct inode *inode;
25017 +       int i, do_put;
25018 +
25019 +       SiMustWriteLock(sb);
25020 +
25021 +       sbinfo = au_sbi(sb);
25022 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
25023 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
25024 +
25025 +       /* no spin_lock since sbinfo is write-locked */
25026 +       for (i = 0; i < AuPlink_NHASH; i++) {
25027 +               plink_hlist = &sbinfo->si_plink[i].head;
25028 +               hlist_for_each_entry_safe(plink, tmp, plink_hlist, hlist) {
25029 +                       inode = au_igrab(plink->inode);
25030 +                       ii_write_lock_child(inode);
25031 +                       do_put = au_plink_do_half_refresh(inode, br_id);
25032 +                       if (do_put)
25033 +                               do_put_plink(plink, 1);
25034 +                       ii_write_unlock(inode);
25035 +                       iput(inode);
25036 +               }
25037 +       }
25038 +}
25039 diff -urN /usr/share/empty/fs/aufs/poll.c linux/fs/aufs/poll.c
25040 --- /usr/share/empty/fs/aufs/poll.c     1970-01-01 01:00:00.000000000 +0100
25041 +++ linux/fs/aufs/poll.c        2015-03-27 21:56:35.466795000 +0100
25042 @@ -0,0 +1,55 @@
25043 +/*
25044 + * Copyright (C) 2005-2015 Junjiro R. Okajima
25045 + *
25046 + * This program, aufs is free software; you can redistribute it and/or modify
25047 + * it under the terms of the GNU General Public License as published by
25048 + * the Free Software Foundation; either version 2 of the License, or
25049 + * (at your option) any later version.
25050 + *
25051 + * This program is distributed in the hope that it will be useful,
25052 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25053 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25054 + * GNU General Public License for more details.
25055 + *
25056 + * You should have received a copy of the GNU General Public License
25057 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25058 + */
25059 +
25060 +/*
25061 + * poll operation
25062 + * There is only one filesystem which implements ->poll operation, currently.
25063 + */
25064 +
25065 +#include "aufs.h"
25066 +
25067 +unsigned int aufs_poll(struct file *file, poll_table *wait)
25068 +{
25069 +       unsigned int mask;
25070 +       int err;
25071 +       struct file *h_file;
25072 +       struct dentry *dentry;
25073 +       struct super_block *sb;
25074 +
25075 +       /* We should pretend an error happened. */
25076 +       mask = POLLERR /* | POLLIN | POLLOUT */;
25077 +       dentry = file->f_path.dentry;
25078 +       sb = dentry->d_sb;
25079 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
25080 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
25081 +       if (unlikely(err))
25082 +               goto out;
25083 +
25084 +       /* it is not an error if h_file has no operation */
25085 +       mask = DEFAULT_POLLMASK;
25086 +       h_file = au_hf_top(file);
25087 +       if (h_file->f_op->poll)
25088 +               mask = h_file->f_op->poll(h_file, wait);
25089 +
25090 +       di_read_unlock(dentry, AuLock_IR);
25091 +       fi_read_unlock(file);
25092 +
25093 +out:
25094 +       si_read_unlock(sb);
25095 +       AuTraceErr((int)mask);
25096 +       return mask;
25097 +}
25098 diff -urN /usr/share/empty/fs/aufs/posix_acl.c linux/fs/aufs/posix_acl.c
25099 --- /usr/share/empty/fs/aufs/posix_acl.c        1970-01-01 01:00:00.000000000 +0100
25100 +++ linux/fs/aufs/posix_acl.c   2015-03-27 21:56:35.466795000 +0100
25101 @@ -0,0 +1,99 @@
25102 +/*
25103 + * Copyright (C) 2014-2015 Junjiro R. Okajima
25104 + *
25105 + * This program, aufs is free software; you can redistribute it and/or modify
25106 + * it under the terms of the GNU General Public License as published by
25107 + * the Free Software Foundation; either version 2 of the License, or
25108 + * (at your option) any later version.
25109 + *
25110 + * This program is distributed in the hope that it will be useful,
25111 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25112 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25113 + * GNU General Public License for more details.
25114 + *
25115 + * You should have received a copy of the GNU General Public License
25116 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25117 + */
25118 +
25119 +/*
25120 + * posix acl operations
25121 + */
25122 +
25123 +#include <linux/fs.h>
25124 +#include <linux/posix_acl.h>
25125 +#include "aufs.h"
25126 +
25127 +struct posix_acl *aufs_get_acl(struct inode *inode, int type)
25128 +{
25129 +       struct posix_acl *acl;
25130 +       int err;
25131 +       aufs_bindex_t bindex;
25132 +       struct inode *h_inode;
25133 +       struct super_block *sb;
25134 +
25135 +       acl = NULL;
25136 +       sb = inode->i_sb;
25137 +       si_read_lock(sb, AuLock_FLUSH);
25138 +       ii_read_lock_child(inode);
25139 +       if (!(sb->s_flags & MS_POSIXACL))
25140 +               goto out;
25141 +
25142 +       bindex = au_ibstart(inode);
25143 +       h_inode = au_h_iptr(inode, bindex);
25144 +       if (unlikely(!h_inode
25145 +                    || ((h_inode->i_mode & S_IFMT)
25146 +                        != (inode->i_mode & S_IFMT)))) {
25147 +               err = au_busy_or_stale();
25148 +               acl = ERR_PTR(err);
25149 +               goto out;
25150 +       }
25151 +
25152 +       /* always topmost only */
25153 +       acl = get_acl(h_inode, type);
25154 +
25155 +out:
25156 +       ii_read_unlock(inode);
25157 +       si_read_unlock(sb);
25158 +
25159 +       AuTraceErrPtr(acl);
25160 +       return acl;
25161 +}
25162 +
25163 +int aufs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
25164 +{
25165 +       int err;
25166 +       ssize_t ssz;
25167 +       struct dentry *dentry;
25168 +       struct au_srxattr arg = {
25169 +               .type = AU_ACL_SET,
25170 +               .u.acl_set = {
25171 +                       .acl    = acl,
25172 +                       .type   = type
25173 +               },
25174 +       };
25175 +
25176 +       mutex_lock(&inode->i_mutex);
25177 +       if (inode->i_ino == AUFS_ROOT_INO)
25178 +               dentry = dget(inode->i_sb->s_root);
25179 +       else {
25180 +               dentry = d_find_alias(inode);
25181 +               if (!dentry)
25182 +                       dentry = d_find_any_alias(inode);
25183 +               if (!dentry) {
25184 +                       pr_warn("cannot handle this inode, "
25185 +                               "please report to aufs-users ML\n");
25186 +                       err = -ENOENT;
25187 +                       goto out;
25188 +               }
25189 +       }
25190 +
25191 +       ssz = au_srxattr(dentry, &arg);
25192 +       dput(dentry);
25193 +       err = ssz;
25194 +       if (ssz >= 0)
25195 +               err = 0;
25196 +
25197 +out:
25198 +       mutex_unlock(&inode->i_mutex);
25199 +       return err;
25200 +}
25201 diff -urN /usr/share/empty/fs/aufs/procfs.c linux/fs/aufs/procfs.c
25202 --- /usr/share/empty/fs/aufs/procfs.c   1970-01-01 01:00:00.000000000 +0100
25203 +++ linux/fs/aufs/procfs.c      2015-03-27 21:56:35.466795000 +0100
25204 @@ -0,0 +1,169 @@
25205 +/*
25206 + * Copyright (C) 2010-2015 Junjiro R. Okajima
25207 + *
25208 + * This program, aufs is free software; you can redistribute it and/or modify
25209 + * it under the terms of the GNU General Public License as published by
25210 + * the Free Software Foundation; either version 2 of the License, or
25211 + * (at your option) any later version.
25212 + *
25213 + * This program is distributed in the hope that it will be useful,
25214 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25215 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25216 + * GNU General Public License for more details.
25217 + *
25218 + * You should have received a copy of the GNU General Public License
25219 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25220 + */
25221 +
25222 +/*
25223 + * procfs interfaces
25224 + */
25225 +
25226 +#include <linux/proc_fs.h>
25227 +#include "aufs.h"
25228 +
25229 +static int au_procfs_plm_release(struct inode *inode, struct file *file)
25230 +{
25231 +       struct au_sbinfo *sbinfo;
25232 +
25233 +       sbinfo = file->private_data;
25234 +       if (sbinfo) {
25235 +               au_plink_maint_leave(sbinfo);
25236 +               kobject_put(&sbinfo->si_kobj);
25237 +       }
25238 +
25239 +       return 0;
25240 +}
25241 +
25242 +static void au_procfs_plm_write_clean(struct file *file)
25243 +{
25244 +       struct au_sbinfo *sbinfo;
25245 +
25246 +       sbinfo = file->private_data;
25247 +       if (sbinfo)
25248 +               au_plink_clean(sbinfo->si_sb, /*verbose*/0);
25249 +}
25250 +
25251 +static int au_procfs_plm_write_si(struct file *file, unsigned long id)
25252 +{
25253 +       int err;
25254 +       struct super_block *sb;
25255 +       struct au_sbinfo *sbinfo;
25256 +
25257 +       err = -EBUSY;
25258 +       if (unlikely(file->private_data))
25259 +               goto out;
25260 +
25261 +       sb = NULL;
25262 +       /* don't use au_sbilist_lock() here */
25263 +       spin_lock(&au_sbilist.spin);
25264 +       list_for_each_entry(sbinfo, &au_sbilist.head, si_list)
25265 +               if (id == sysaufs_si_id(sbinfo)) {
25266 +                       kobject_get(&sbinfo->si_kobj);
25267 +                       sb = sbinfo->si_sb;
25268 +                       break;
25269 +               }
25270 +       spin_unlock(&au_sbilist.spin);
25271 +
25272 +       err = -EINVAL;
25273 +       if (unlikely(!sb))
25274 +               goto out;
25275 +
25276 +       err = au_plink_maint_enter(sb);
25277 +       if (!err)
25278 +               /* keep kobject_get() */
25279 +               file->private_data = sbinfo;
25280 +       else
25281 +               kobject_put(&sbinfo->si_kobj);
25282 +out:
25283 +       return err;
25284 +}
25285 +
25286 +/*
25287 + * Accept a valid "si=xxxx" only.
25288 + * Once it is accepted successfully, accept "clean" too.
25289 + */
25290 +static ssize_t au_procfs_plm_write(struct file *file, const char __user *ubuf,
25291 +                                  size_t count, loff_t *ppos)
25292 +{
25293 +       ssize_t err;
25294 +       unsigned long id;
25295 +       /* last newline is allowed */
25296 +       char buf[3 + sizeof(unsigned long) * 2 + 1];
25297 +
25298 +       err = -EACCES;
25299 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
25300 +               goto out;
25301 +
25302 +       err = -EINVAL;
25303 +       if (unlikely(count > sizeof(buf)))
25304 +               goto out;
25305 +
25306 +       err = copy_from_user(buf, ubuf, count);
25307 +       if (unlikely(err)) {
25308 +               err = -EFAULT;
25309 +               goto out;
25310 +       }
25311 +       buf[count] = 0;
25312 +
25313 +       err = -EINVAL;
25314 +       if (!strcmp("clean", buf)) {
25315 +               au_procfs_plm_write_clean(file);
25316 +               goto out_success;
25317 +       } else if (unlikely(strncmp("si=", buf, 3)))
25318 +               goto out;
25319 +
25320 +       err = kstrtoul(buf + 3, 16, &id);
25321 +       if (unlikely(err))
25322 +               goto out;
25323 +
25324 +       err = au_procfs_plm_write_si(file, id);
25325 +       if (unlikely(err))
25326 +               goto out;
25327 +
25328 +out_success:
25329 +       err = count; /* success */
25330 +out:
25331 +       return err;
25332 +}
25333 +
25334 +static const struct file_operations au_procfs_plm_fop = {
25335 +       .write          = au_procfs_plm_write,
25336 +       .release        = au_procfs_plm_release,
25337 +       .owner          = THIS_MODULE
25338 +};
25339 +
25340 +/* ---------------------------------------------------------------------- */
25341 +
25342 +static struct proc_dir_entry *au_procfs_dir;
25343 +
25344 +void au_procfs_fin(void)
25345 +{
25346 +       remove_proc_entry(AUFS_PLINK_MAINT_NAME, au_procfs_dir);
25347 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
25348 +}
25349 +
25350 +int __init au_procfs_init(void)
25351 +{
25352 +       int err;
25353 +       struct proc_dir_entry *entry;
25354 +
25355 +       err = -ENOMEM;
25356 +       au_procfs_dir = proc_mkdir(AUFS_PLINK_MAINT_DIR, NULL);
25357 +       if (unlikely(!au_procfs_dir))
25358 +               goto out;
25359 +
25360 +       entry = proc_create(AUFS_PLINK_MAINT_NAME, S_IFREG | S_IWUSR,
25361 +                           au_procfs_dir, &au_procfs_plm_fop);
25362 +       if (unlikely(!entry))
25363 +               goto out_dir;
25364 +
25365 +       err = 0;
25366 +       goto out; /* success */
25367 +
25368 +
25369 +out_dir:
25370 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
25371 +out:
25372 +       return err;
25373 +}
25374 diff -urN /usr/share/empty/fs/aufs/rdu.c linux/fs/aufs/rdu.c
25375 --- /usr/share/empty/fs/aufs/rdu.c      1970-01-01 01:00:00.000000000 +0100
25376 +++ linux/fs/aufs/rdu.c 2015-03-27 21:56:35.466795000 +0100
25377 @@ -0,0 +1,388 @@
25378 +/*
25379 + * Copyright (C) 2005-2015 Junjiro R. Okajima
25380 + *
25381 + * This program, aufs is free software; you can redistribute it and/or modify
25382 + * it under the terms of the GNU General Public License as published by
25383 + * the Free Software Foundation; either version 2 of the License, or
25384 + * (at your option) any later version.
25385 + *
25386 + * This program is distributed in the hope that it will be useful,
25387 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25388 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25389 + * GNU General Public License for more details.
25390 + *
25391 + * You should have received a copy of the GNU General Public License
25392 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25393 + */
25394 +
25395 +/*
25396 + * readdir in userspace.
25397 + */
25398 +
25399 +#include <linux/compat.h>
25400 +#include <linux/fs_stack.h>
25401 +#include <linux/security.h>
25402 +#include "aufs.h"
25403 +
25404 +/* bits for struct aufs_rdu.flags */
25405 +#define        AuRdu_CALLED    1
25406 +#define        AuRdu_CONT      (1 << 1)
25407 +#define        AuRdu_FULL      (1 << 2)
25408 +#define au_ftest_rdu(flags, name)      ((flags) & AuRdu_##name)
25409 +#define au_fset_rdu(flags, name) \
25410 +       do { (flags) |= AuRdu_##name; } while (0)
25411 +#define au_fclr_rdu(flags, name) \
25412 +       do { (flags) &= ~AuRdu_##name; } while (0)
25413 +
25414 +struct au_rdu_arg {
25415 +       struct dir_context              ctx;
25416 +       struct aufs_rdu                 *rdu;
25417 +       union au_rdu_ent_ul             ent;
25418 +       unsigned long                   end;
25419 +
25420 +       struct super_block              *sb;
25421 +       int                             err;
25422 +};
25423 +
25424 +static int au_rdu_fill(struct dir_context *ctx, const char *name, int nlen,
25425 +                      loff_t offset, u64 h_ino, unsigned int d_type)
25426 +{
25427 +       int err, len;
25428 +       struct au_rdu_arg *arg = container_of(ctx, struct au_rdu_arg, ctx);
25429 +       struct aufs_rdu *rdu = arg->rdu;
25430 +       struct au_rdu_ent ent;
25431 +
25432 +       err = 0;
25433 +       arg->err = 0;
25434 +       au_fset_rdu(rdu->cookie.flags, CALLED);
25435 +       len = au_rdu_len(nlen);
25436 +       if (arg->ent.ul + len  < arg->end) {
25437 +               ent.ino = h_ino;
25438 +               ent.bindex = rdu->cookie.bindex;
25439 +               ent.type = d_type;
25440 +               ent.nlen = nlen;
25441 +               if (unlikely(nlen > AUFS_MAX_NAMELEN))
25442 +                       ent.type = DT_UNKNOWN;
25443 +
25444 +               /* unnecessary to support mmap_sem since this is a dir */
25445 +               err = -EFAULT;
25446 +               if (copy_to_user(arg->ent.e, &ent, sizeof(ent)))
25447 +                       goto out;
25448 +               if (copy_to_user(arg->ent.e->name, name, nlen))
25449 +                       goto out;
25450 +               /* the terminating NULL */
25451 +               if (__put_user(0, arg->ent.e->name + nlen))
25452 +                       goto out;
25453 +               err = 0;
25454 +               /* AuDbg("%p, %.*s\n", arg->ent.p, nlen, name); */
25455 +               arg->ent.ul += len;
25456 +               rdu->rent++;
25457 +       } else {
25458 +               err = -EFAULT;
25459 +               au_fset_rdu(rdu->cookie.flags, FULL);
25460 +               rdu->full = 1;
25461 +               rdu->tail = arg->ent;
25462 +       }
25463 +
25464 +out:
25465 +       /* AuTraceErr(err); */
25466 +       return err;
25467 +}
25468 +
25469 +static int au_rdu_do(struct file *h_file, struct au_rdu_arg *arg)
25470 +{
25471 +       int err;
25472 +       loff_t offset;
25473 +       struct au_rdu_cookie *cookie = &arg->rdu->cookie;
25474 +
25475 +       /* we don't have to care (FMODE_32BITHASH | FMODE_64BITHASH) for ext4 */
25476 +       offset = vfsub_llseek(h_file, cookie->h_pos, SEEK_SET);
25477 +       err = offset;
25478 +       if (unlikely(offset != cookie->h_pos))
25479 +               goto out;
25480 +
25481 +       err = 0;
25482 +       do {
25483 +               arg->err = 0;
25484 +               au_fclr_rdu(cookie->flags, CALLED);
25485 +               /* smp_mb(); */
25486 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
25487 +               if (err >= 0)
25488 +                       err = arg->err;
25489 +       } while (!err
25490 +                && au_ftest_rdu(cookie->flags, CALLED)
25491 +                && !au_ftest_rdu(cookie->flags, FULL));
25492 +       cookie->h_pos = h_file->f_pos;
25493 +
25494 +out:
25495 +       AuTraceErr(err);
25496 +       return err;
25497 +}
25498 +
25499 +static int au_rdu(struct file *file, struct aufs_rdu *rdu)
25500 +{
25501 +       int err;
25502 +       aufs_bindex_t bend;
25503 +       struct au_rdu_arg arg = {
25504 +               .ctx = {
25505 +                       .actor = au_rdu_fill
25506 +               }
25507 +       };
25508 +       struct dentry *dentry;
25509 +       struct inode *inode;
25510 +       struct file *h_file;
25511 +       struct au_rdu_cookie *cookie = &rdu->cookie;
25512 +
25513 +       err = !access_ok(VERIFY_WRITE, rdu->ent.e, rdu->sz);
25514 +       if (unlikely(err)) {
25515 +               err = -EFAULT;
25516 +               AuTraceErr(err);
25517 +               goto out;
25518 +       }
25519 +       rdu->rent = 0;
25520 +       rdu->tail = rdu->ent;
25521 +       rdu->full = 0;
25522 +       arg.rdu = rdu;
25523 +       arg.ent = rdu->ent;
25524 +       arg.end = arg.ent.ul;
25525 +       arg.end += rdu->sz;
25526 +
25527 +       err = -ENOTDIR;
25528 +       if (unlikely(!file->f_op->iterate))
25529 +               goto out;
25530 +
25531 +       err = security_file_permission(file, MAY_READ);
25532 +       AuTraceErr(err);
25533 +       if (unlikely(err))
25534 +               goto out;
25535 +
25536 +       dentry = file->f_path.dentry;
25537 +       inode = dentry->d_inode;
25538 +#if 1
25539 +       mutex_lock(&inode->i_mutex);
25540 +#else
25541 +       err = mutex_lock_killable(&inode->i_mutex);
25542 +       AuTraceErr(err);
25543 +       if (unlikely(err))
25544 +               goto out;
25545 +#endif
25546 +
25547 +       arg.sb = inode->i_sb;
25548 +       err = si_read_lock(arg.sb, AuLock_FLUSH | AuLock_NOPLM);
25549 +       if (unlikely(err))
25550 +               goto out_mtx;
25551 +       err = au_alive_dir(dentry);
25552 +       if (unlikely(err))
25553 +               goto out_si;
25554 +       /* todo: reval? */
25555 +       fi_read_lock(file);
25556 +
25557 +       err = -EAGAIN;
25558 +       if (unlikely(au_ftest_rdu(cookie->flags, CONT)
25559 +                    && cookie->generation != au_figen(file)))
25560 +               goto out_unlock;
25561 +
25562 +       err = 0;
25563 +       if (!rdu->blk) {
25564 +               rdu->blk = au_sbi(arg.sb)->si_rdblk;
25565 +               if (!rdu->blk)
25566 +                       rdu->blk = au_dir_size(file, /*dentry*/NULL);
25567 +       }
25568 +       bend = au_fbstart(file);
25569 +       if (cookie->bindex < bend)
25570 +               cookie->bindex = bend;
25571 +       bend = au_fbend_dir(file);
25572 +       /* AuDbg("b%d, b%d\n", cookie->bindex, bend); */
25573 +       for (; !err && cookie->bindex <= bend;
25574 +            cookie->bindex++, cookie->h_pos = 0) {
25575 +               h_file = au_hf_dir(file, cookie->bindex);
25576 +               if (!h_file)
25577 +                       continue;
25578 +
25579 +               au_fclr_rdu(cookie->flags, FULL);
25580 +               err = au_rdu_do(h_file, &arg);
25581 +               AuTraceErr(err);
25582 +               if (unlikely(au_ftest_rdu(cookie->flags, FULL) || err))
25583 +                       break;
25584 +       }
25585 +       AuDbg("rent %llu\n", rdu->rent);
25586 +
25587 +       if (!err && !au_ftest_rdu(cookie->flags, CONT)) {
25588 +               rdu->shwh = !!au_opt_test(au_sbi(arg.sb)->si_mntflags, SHWH);
25589 +               au_fset_rdu(cookie->flags, CONT);
25590 +               cookie->generation = au_figen(file);
25591 +       }
25592 +
25593 +       ii_read_lock_child(inode);
25594 +       fsstack_copy_attr_atime(inode, au_h_iptr(inode, au_ibstart(inode)));
25595 +       ii_read_unlock(inode);
25596 +
25597 +out_unlock:
25598 +       fi_read_unlock(file);
25599 +out_si:
25600 +       si_read_unlock(arg.sb);
25601 +out_mtx:
25602 +       mutex_unlock(&inode->i_mutex);
25603 +out:
25604 +       AuTraceErr(err);
25605 +       return err;
25606 +}
25607 +
25608 +static int au_rdu_ino(struct file *file, struct aufs_rdu *rdu)
25609 +{
25610 +       int err;
25611 +       ino_t ino;
25612 +       unsigned long long nent;
25613 +       union au_rdu_ent_ul *u;
25614 +       struct au_rdu_ent ent;
25615 +       struct super_block *sb;
25616 +
25617 +       err = 0;
25618 +       nent = rdu->nent;
25619 +       u = &rdu->ent;
25620 +       sb = file->f_path.dentry->d_sb;
25621 +       si_read_lock(sb, AuLock_FLUSH);
25622 +       while (nent-- > 0) {
25623 +               /* unnecessary to support mmap_sem since this is a dir */
25624 +               err = copy_from_user(&ent, u->e, sizeof(ent));
25625 +               if (!err)
25626 +                       err = !access_ok(VERIFY_WRITE, &u->e->ino, sizeof(ino));
25627 +               if (unlikely(err)) {
25628 +                       err = -EFAULT;
25629 +                       AuTraceErr(err);
25630 +                       break;
25631 +               }
25632 +
25633 +               /* AuDbg("b%d, i%llu\n", ent.bindex, ent.ino); */
25634 +               if (!ent.wh)
25635 +                       err = au_ino(sb, ent.bindex, ent.ino, ent.type, &ino);
25636 +               else
25637 +                       err = au_wh_ino(sb, ent.bindex, ent.ino, ent.type,
25638 +                                       &ino);
25639 +               if (unlikely(err)) {
25640 +                       AuTraceErr(err);
25641 +                       break;
25642 +               }
25643 +
25644 +               err = __put_user(ino, &u->e->ino);
25645 +               if (unlikely(err)) {
25646 +                       err = -EFAULT;
25647 +                       AuTraceErr(err);
25648 +                       break;
25649 +               }
25650 +               u->ul += au_rdu_len(ent.nlen);
25651 +       }
25652 +       si_read_unlock(sb);
25653 +
25654 +       return err;
25655 +}
25656 +
25657 +/* ---------------------------------------------------------------------- */
25658 +
25659 +static int au_rdu_verify(struct aufs_rdu *rdu)
25660 +{
25661 +       AuDbg("rdu{%llu, %p, %u | %u | %llu, %u, %u | "
25662 +             "%llu, b%d, 0x%x, g%u}\n",
25663 +             rdu->sz, rdu->ent.e, rdu->verify[AufsCtlRduV_SZ],
25664 +             rdu->blk,
25665 +             rdu->rent, rdu->shwh, rdu->full,
25666 +             rdu->cookie.h_pos, rdu->cookie.bindex, rdu->cookie.flags,
25667 +             rdu->cookie.generation);
25668 +
25669 +       if (rdu->verify[AufsCtlRduV_SZ] == sizeof(*rdu))
25670 +               return 0;
25671 +
25672 +       AuDbg("%u:%u\n",
25673 +             rdu->verify[AufsCtlRduV_SZ], (unsigned int)sizeof(*rdu));
25674 +       return -EINVAL;
25675 +}
25676 +
25677 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
25678 +{
25679 +       long err, e;
25680 +       struct aufs_rdu rdu;
25681 +       void __user *p = (void __user *)arg;
25682 +
25683 +       err = copy_from_user(&rdu, p, sizeof(rdu));
25684 +       if (unlikely(err)) {
25685 +               err = -EFAULT;
25686 +               AuTraceErr(err);
25687 +               goto out;
25688 +       }
25689 +       err = au_rdu_verify(&rdu);
25690 +       if (unlikely(err))
25691 +               goto out;
25692 +
25693 +       switch (cmd) {
25694 +       case AUFS_CTL_RDU:
25695 +               err = au_rdu(file, &rdu);
25696 +               if (unlikely(err))
25697 +                       break;
25698 +
25699 +               e = copy_to_user(p, &rdu, sizeof(rdu));
25700 +               if (unlikely(e)) {
25701 +                       err = -EFAULT;
25702 +                       AuTraceErr(err);
25703 +               }
25704 +               break;
25705 +       case AUFS_CTL_RDU_INO:
25706 +               err = au_rdu_ino(file, &rdu);
25707 +               break;
25708 +
25709 +       default:
25710 +               /* err = -ENOTTY; */
25711 +               err = -EINVAL;
25712 +       }
25713 +
25714 +out:
25715 +       AuTraceErr(err);
25716 +       return err;
25717 +}
25718 +
25719 +#ifdef CONFIG_COMPAT
25720 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
25721 +{
25722 +       long err, e;
25723 +       struct aufs_rdu rdu;
25724 +       void __user *p = compat_ptr(arg);
25725 +
25726 +       /* todo: get_user()? */
25727 +       err = copy_from_user(&rdu, p, sizeof(rdu));
25728 +       if (unlikely(err)) {
25729 +               err = -EFAULT;
25730 +               AuTraceErr(err);
25731 +               goto out;
25732 +       }
25733 +       rdu.ent.e = compat_ptr(rdu.ent.ul);
25734 +       err = au_rdu_verify(&rdu);
25735 +       if (unlikely(err))
25736 +               goto out;
25737 +
25738 +       switch (cmd) {
25739 +       case AUFS_CTL_RDU:
25740 +               err = au_rdu(file, &rdu);
25741 +               if (unlikely(err))
25742 +                       break;
25743 +
25744 +               rdu.ent.ul = ptr_to_compat(rdu.ent.e);
25745 +               rdu.tail.ul = ptr_to_compat(rdu.tail.e);
25746 +               e = copy_to_user(p, &rdu, sizeof(rdu));
25747 +               if (unlikely(e)) {
25748 +                       err = -EFAULT;
25749 +                       AuTraceErr(err);
25750 +               }
25751 +               break;
25752 +       case AUFS_CTL_RDU_INO:
25753 +               err = au_rdu_ino(file, &rdu);
25754 +               break;
25755 +
25756 +       default:
25757 +               /* err = -ENOTTY; */
25758 +               err = -EINVAL;
25759 +       }
25760 +
25761 +out:
25762 +       AuTraceErr(err);
25763 +       return err;
25764 +}
25765 +#endif
25766 diff -urN /usr/share/empty/fs/aufs/rwsem.h linux/fs/aufs/rwsem.h
25767 --- /usr/share/empty/fs/aufs/rwsem.h    1970-01-01 01:00:00.000000000 +0100
25768 +++ linux/fs/aufs/rwsem.h       2015-03-27 21:56:35.466795000 +0100
25769 @@ -0,0 +1,191 @@
25770 +/*
25771 + * Copyright (C) 2005-2015 Junjiro R. Okajima
25772 + *
25773 + * This program, aufs is free software; you can redistribute it and/or modify
25774 + * it under the terms of the GNU General Public License as published by
25775 + * the Free Software Foundation; either version 2 of the License, or
25776 + * (at your option) any later version.
25777 + *
25778 + * This program is distributed in the hope that it will be useful,
25779 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25780 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25781 + * GNU General Public License for more details.
25782 + *
25783 + * You should have received a copy of the GNU General Public License
25784 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25785 + */
25786 +
25787 +/*
25788 + * simple read-write semaphore wrappers
25789 + */
25790 +
25791 +#ifndef __AUFS_RWSEM_H__
25792 +#define __AUFS_RWSEM_H__
25793 +
25794 +#ifdef __KERNEL__
25795 +
25796 +#include "debug.h"
25797 +
25798 +struct au_rwsem {
25799 +       struct rw_semaphore     rwsem;
25800 +#ifdef CONFIG_AUFS_DEBUG
25801 +       /* just for debugging, not almighty counter */
25802 +       atomic_t                rcnt, wcnt;
25803 +#endif
25804 +};
25805 +
25806 +#ifdef CONFIG_AUFS_DEBUG
25807 +#define AuDbgCntInit(rw) do { \
25808 +       atomic_set(&(rw)->rcnt, 0); \
25809 +       atomic_set(&(rw)->wcnt, 0); \
25810 +       smp_mb(); /* atomic set */ \
25811 +} while (0)
25812 +
25813 +#define AuDbgRcntInc(rw)       atomic_inc(&(rw)->rcnt)
25814 +#define AuDbgRcntDec(rw)       WARN_ON(atomic_dec_return(&(rw)->rcnt) < 0)
25815 +#define AuDbgWcntInc(rw)       atomic_inc(&(rw)->wcnt)
25816 +#define AuDbgWcntDec(rw)       WARN_ON(atomic_dec_return(&(rw)->wcnt) < 0)
25817 +#else
25818 +#define AuDbgCntInit(rw)       do {} while (0)
25819 +#define AuDbgRcntInc(rw)       do {} while (0)
25820 +#define AuDbgRcntDec(rw)       do {} while (0)
25821 +#define AuDbgWcntInc(rw)       do {} while (0)
25822 +#define AuDbgWcntDec(rw)       do {} while (0)
25823 +#endif /* CONFIG_AUFS_DEBUG */
25824 +
25825 +/* to debug easier, do not make them inlined functions */
25826 +#define AuRwMustNoWaiters(rw)  AuDebugOn(!list_empty(&(rw)->rwsem.wait_list))
25827 +/* rwsem_is_locked() is unusable */
25828 +#define AuRwMustReadLock(rw)   AuDebugOn(atomic_read(&(rw)->rcnt) <= 0)
25829 +#define AuRwMustWriteLock(rw)  AuDebugOn(atomic_read(&(rw)->wcnt) <= 0)
25830 +#define AuRwMustAnyLock(rw)    AuDebugOn(atomic_read(&(rw)->rcnt) <= 0 \
25831 +                                       && atomic_read(&(rw)->wcnt) <= 0)
25832 +#define AuRwDestroy(rw)                AuDebugOn(atomic_read(&(rw)->rcnt) \
25833 +                                       || atomic_read(&(rw)->wcnt))
25834 +
25835 +#define au_rw_class(rw, key)   lockdep_set_class(&(rw)->rwsem, key)
25836 +
25837 +static inline void au_rw_init(struct au_rwsem *rw)
25838 +{
25839 +       AuDbgCntInit(rw);
25840 +       init_rwsem(&rw->rwsem);
25841 +}
25842 +
25843 +static inline void au_rw_init_wlock(struct au_rwsem *rw)
25844 +{
25845 +       au_rw_init(rw);
25846 +       down_write(&rw->rwsem);
25847 +       AuDbgWcntInc(rw);
25848 +}
25849 +
25850 +static inline void au_rw_init_wlock_nested(struct au_rwsem *rw,
25851 +                                          unsigned int lsc)
25852 +{
25853 +       au_rw_init(rw);
25854 +       down_write_nested(&rw->rwsem, lsc);
25855 +       AuDbgWcntInc(rw);
25856 +}
25857 +
25858 +static inline void au_rw_read_lock(struct au_rwsem *rw)
25859 +{
25860 +       down_read(&rw->rwsem);
25861 +       AuDbgRcntInc(rw);
25862 +}
25863 +
25864 +static inline void au_rw_read_lock_nested(struct au_rwsem *rw, unsigned int lsc)
25865 +{
25866 +       down_read_nested(&rw->rwsem, lsc);
25867 +       AuDbgRcntInc(rw);
25868 +}
25869 +
25870 +static inline void au_rw_read_unlock(struct au_rwsem *rw)
25871 +{
25872 +       AuRwMustReadLock(rw);
25873 +       AuDbgRcntDec(rw);
25874 +       up_read(&rw->rwsem);
25875 +}
25876 +
25877 +static inline void au_rw_dgrade_lock(struct au_rwsem *rw)
25878 +{
25879 +       AuRwMustWriteLock(rw);
25880 +       AuDbgRcntInc(rw);
25881 +       AuDbgWcntDec(rw);
25882 +       downgrade_write(&rw->rwsem);
25883 +}
25884 +
25885 +static inline void au_rw_write_lock(struct au_rwsem *rw)
25886 +{
25887 +       down_write(&rw->rwsem);
25888 +       AuDbgWcntInc(rw);
25889 +}
25890 +
25891 +static inline void au_rw_write_lock_nested(struct au_rwsem *rw,
25892 +                                          unsigned int lsc)
25893 +{
25894 +       down_write_nested(&rw->rwsem, lsc);
25895 +       AuDbgWcntInc(rw);
25896 +}
25897 +
25898 +static inline void au_rw_write_unlock(struct au_rwsem *rw)
25899 +{
25900 +       AuRwMustWriteLock(rw);
25901 +       AuDbgWcntDec(rw);
25902 +       up_write(&rw->rwsem);
25903 +}
25904 +
25905 +/* why is not _nested version defined */
25906 +static inline int au_rw_read_trylock(struct au_rwsem *rw)
25907 +{
25908 +       int ret;
25909 +
25910 +       ret = down_read_trylock(&rw->rwsem);
25911 +       if (ret)
25912 +               AuDbgRcntInc(rw);
25913 +       return ret;
25914 +}
25915 +
25916 +static inline int au_rw_write_trylock(struct au_rwsem *rw)
25917 +{
25918 +       int ret;
25919 +
25920 +       ret = down_write_trylock(&rw->rwsem);
25921 +       if (ret)
25922 +               AuDbgWcntInc(rw);
25923 +       return ret;
25924 +}
25925 +
25926 +#undef AuDbgCntInit
25927 +#undef AuDbgRcntInc
25928 +#undef AuDbgRcntDec
25929 +#undef AuDbgWcntInc
25930 +#undef AuDbgWcntDec
25931 +
25932 +#define AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
25933 +static inline void prefix##_read_lock(param) \
25934 +{ au_rw_read_lock(rwsem); } \
25935 +static inline void prefix##_write_lock(param) \
25936 +{ au_rw_write_lock(rwsem); } \
25937 +static inline int prefix##_read_trylock(param) \
25938 +{ return au_rw_read_trylock(rwsem); } \
25939 +static inline int prefix##_write_trylock(param) \
25940 +{ return au_rw_write_trylock(rwsem); }
25941 +/* why is not _nested version defined */
25942 +/* static inline void prefix##_read_trylock_nested(param, lsc)
25943 +{ au_rw_read_trylock_nested(rwsem, lsc)); }
25944 +static inline void prefix##_write_trylock_nestd(param, lsc)
25945 +{ au_rw_write_trylock_nested(rwsem, lsc); } */
25946 +
25947 +#define AuSimpleUnlockRwsemFuncs(prefix, param, rwsem) \
25948 +static inline void prefix##_read_unlock(param) \
25949 +{ au_rw_read_unlock(rwsem); } \
25950 +static inline void prefix##_write_unlock(param) \
25951 +{ au_rw_write_unlock(rwsem); } \
25952 +static inline void prefix##_downgrade_lock(param) \
25953 +{ au_rw_dgrade_lock(rwsem); }
25954 +
25955 +#define AuSimpleRwsemFuncs(prefix, param, rwsem) \
25956 +       AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
25957 +       AuSimpleUnlockRwsemFuncs(prefix, param, rwsem)
25958 +
25959 +#endif /* __KERNEL__ */
25960 +#endif /* __AUFS_RWSEM_H__ */
25961 diff -urN /usr/share/empty/fs/aufs/sbinfo.c linux/fs/aufs/sbinfo.c
25962 --- /usr/share/empty/fs/aufs/sbinfo.c   1970-01-01 01:00:00.000000000 +0100
25963 +++ linux/fs/aufs/sbinfo.c      2015-03-27 21:56:35.466795000 +0100
25964 @@ -0,0 +1,353 @@
25965 +/*
25966 + * Copyright (C) 2005-2015 Junjiro R. Okajima
25967 + *
25968 + * This program, aufs is free software; you can redistribute it and/or modify
25969 + * it under the terms of the GNU General Public License as published by
25970 + * the Free Software Foundation; either version 2 of the License, or
25971 + * (at your option) any later version.
25972 + *
25973 + * This program is distributed in the hope that it will be useful,
25974 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25975 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25976 + * GNU General Public License for more details.
25977 + *
25978 + * You should have received a copy of the GNU General Public License
25979 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25980 + */
25981 +
25982 +/*
25983 + * superblock private data
25984 + */
25985 +
25986 +#include "aufs.h"
25987 +
25988 +/*
25989 + * they are necessary regardless sysfs is disabled.
25990 + */
25991 +void au_si_free(struct kobject *kobj)
25992 +{
25993 +       int i;
25994 +       struct au_sbinfo *sbinfo;
25995 +       char *locked __maybe_unused; /* debug only */
25996 +
25997 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
25998 +       for (i = 0; i < AuPlink_NHASH; i++)
25999 +               AuDebugOn(!hlist_empty(&sbinfo->si_plink[i].head));
26000 +       AuDebugOn(atomic_read(&sbinfo->si_nowait.nw_len));
26001 +
26002 +       au_rw_write_lock(&sbinfo->si_rwsem);
26003 +       au_br_free(sbinfo);
26004 +       au_rw_write_unlock(&sbinfo->si_rwsem);
26005 +
26006 +       AuDebugOn(radix_tree_gang_lookup
26007 +                 (&sbinfo->au_si_pid.tree, (void **)&locked,
26008 +                  /*first_index*/PID_MAX_DEFAULT - 1,
26009 +                  /*max_items*/sizeof(locked)/sizeof(*locked)));
26010 +
26011 +       kfree(sbinfo->si_branch);
26012 +       kfree(sbinfo->au_si_pid.bitmap);
26013 +       mutex_destroy(&sbinfo->si_xib_mtx);
26014 +       AuRwDestroy(&sbinfo->si_rwsem);
26015 +
26016 +       kfree(sbinfo);
26017 +}
26018 +
26019 +int au_si_alloc(struct super_block *sb)
26020 +{
26021 +       int err, i;
26022 +       struct au_sbinfo *sbinfo;
26023 +       static struct lock_class_key aufs_si;
26024 +
26025 +       err = -ENOMEM;
26026 +       sbinfo = kzalloc(sizeof(*sbinfo), GFP_NOFS);
26027 +       if (unlikely(!sbinfo))
26028 +               goto out;
26029 +
26030 +       BUILD_BUG_ON(sizeof(unsigned long) !=
26031 +                    sizeof(*sbinfo->au_si_pid.bitmap));
26032 +       sbinfo->au_si_pid.bitmap = kcalloc(BITS_TO_LONGS(PID_MAX_DEFAULT),
26033 +                                       sizeof(*sbinfo->au_si_pid.bitmap),
26034 +                                       GFP_NOFS);
26035 +       if (unlikely(!sbinfo->au_si_pid.bitmap))
26036 +               goto out_sbinfo;
26037 +
26038 +       /* will be reallocated separately */
26039 +       sbinfo->si_branch = kzalloc(sizeof(*sbinfo->si_branch), GFP_NOFS);
26040 +       if (unlikely(!sbinfo->si_branch))
26041 +               goto out_pidmap;
26042 +
26043 +       err = sysaufs_si_init(sbinfo);
26044 +       if (unlikely(err))
26045 +               goto out_br;
26046 +
26047 +       au_nwt_init(&sbinfo->si_nowait);
26048 +       au_rw_init_wlock(&sbinfo->si_rwsem);
26049 +       au_rw_class(&sbinfo->si_rwsem, &aufs_si);
26050 +       spin_lock_init(&sbinfo->au_si_pid.tree_lock);
26051 +       INIT_RADIX_TREE(&sbinfo->au_si_pid.tree, GFP_ATOMIC | __GFP_NOFAIL);
26052 +
26053 +       atomic_long_set(&sbinfo->si_ninodes, 0);
26054 +       atomic_long_set(&sbinfo->si_nfiles, 0);
26055 +
26056 +       sbinfo->si_bend = -1;
26057 +       sbinfo->si_last_br_id = AUFS_BRANCH_MAX / 2;
26058 +
26059 +       sbinfo->si_wbr_copyup = AuWbrCopyup_Def;
26060 +       sbinfo->si_wbr_create = AuWbrCreate_Def;
26061 +       sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + sbinfo->si_wbr_copyup;
26062 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + sbinfo->si_wbr_create;
26063 +
26064 +       au_fhsm_init(sbinfo);
26065 +
26066 +       sbinfo->si_mntflags = au_opts_plink(AuOpt_Def);
26067 +
26068 +       sbinfo->si_xino_jiffy = jiffies;
26069 +       sbinfo->si_xino_expire
26070 +               = msecs_to_jiffies(AUFS_XINO_DEF_SEC * MSEC_PER_SEC);
26071 +       mutex_init(&sbinfo->si_xib_mtx);
26072 +       sbinfo->si_xino_brid = -1;
26073 +       /* leave si_xib_last_pindex and si_xib_next_bit */
26074 +
26075 +       sbinfo->si_rdcache = msecs_to_jiffies(AUFS_RDCACHE_DEF * MSEC_PER_SEC);
26076 +       sbinfo->si_rdblk = AUFS_RDBLK_DEF;
26077 +       sbinfo->si_rdhash = AUFS_RDHASH_DEF;
26078 +       sbinfo->si_dirwh = AUFS_DIRWH_DEF;
26079 +
26080 +       for (i = 0; i < AuPlink_NHASH; i++)
26081 +               au_sphl_init(sbinfo->si_plink + i);
26082 +       init_waitqueue_head(&sbinfo->si_plink_wq);
26083 +       spin_lock_init(&sbinfo->si_plink_maint_lock);
26084 +
26085 +       au_sphl_init(&sbinfo->si_files);
26086 +
26087 +       /* leave other members for sysaufs and si_mnt. */
26088 +       sbinfo->si_sb = sb;
26089 +       sb->s_fs_info = sbinfo;
26090 +       si_pid_set(sb);
26091 +       au_debug_sbinfo_init(sbinfo);
26092 +       return 0; /* success */
26093 +
26094 +out_br:
26095 +       kfree(sbinfo->si_branch);
26096 +out_pidmap:
26097 +       kfree(sbinfo->au_si_pid.bitmap);
26098 +out_sbinfo:
26099 +       kfree(sbinfo);
26100 +out:
26101 +       return err;
26102 +}
26103 +
26104 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr)
26105 +{
26106 +       int err, sz;
26107 +       struct au_branch **brp;
26108 +
26109 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
26110 +
26111 +       err = -ENOMEM;
26112 +       sz = sizeof(*brp) * (sbinfo->si_bend + 1);
26113 +       if (unlikely(!sz))
26114 +               sz = sizeof(*brp);
26115 +       brp = au_kzrealloc(sbinfo->si_branch, sz, sizeof(*brp) * nbr, GFP_NOFS);
26116 +       if (brp) {
26117 +               sbinfo->si_branch = brp;
26118 +               err = 0;
26119 +       }
26120 +
26121 +       return err;
26122 +}
26123 +
26124 +/* ---------------------------------------------------------------------- */
26125 +
26126 +unsigned int au_sigen_inc(struct super_block *sb)
26127 +{
26128 +       unsigned int gen;
26129 +
26130 +       SiMustWriteLock(sb);
26131 +
26132 +       gen = ++au_sbi(sb)->si_generation;
26133 +       au_update_digen(sb->s_root);
26134 +       au_update_iigen(sb->s_root->d_inode, /*half*/0);
26135 +       sb->s_root->d_inode->i_version++;
26136 +       return gen;
26137 +}
26138 +
26139 +aufs_bindex_t au_new_br_id(struct super_block *sb)
26140 +{
26141 +       aufs_bindex_t br_id;
26142 +       int i;
26143 +       struct au_sbinfo *sbinfo;
26144 +
26145 +       SiMustWriteLock(sb);
26146 +
26147 +       sbinfo = au_sbi(sb);
26148 +       for (i = 0; i <= AUFS_BRANCH_MAX; i++) {
26149 +               br_id = ++sbinfo->si_last_br_id;
26150 +               AuDebugOn(br_id < 0);
26151 +               if (br_id && au_br_index(sb, br_id) < 0)
26152 +                       return br_id;
26153 +       }
26154 +
26155 +       return -1;
26156 +}
26157 +
26158 +/* ---------------------------------------------------------------------- */
26159 +
26160 +/* it is ok that new 'nwt' tasks are appended while we are sleeping */
26161 +int si_read_lock(struct super_block *sb, int flags)
26162 +{
26163 +       int err;
26164 +
26165 +       err = 0;
26166 +       if (au_ftest_lock(flags, FLUSH))
26167 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26168 +
26169 +       si_noflush_read_lock(sb);
26170 +       err = au_plink_maint(sb, flags);
26171 +       if (unlikely(err))
26172 +               si_read_unlock(sb);
26173 +
26174 +       return err;
26175 +}
26176 +
26177 +int si_write_lock(struct super_block *sb, int flags)
26178 +{
26179 +       int err;
26180 +
26181 +       if (au_ftest_lock(flags, FLUSH))
26182 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26183 +
26184 +       si_noflush_write_lock(sb);
26185 +       err = au_plink_maint(sb, flags);
26186 +       if (unlikely(err))
26187 +               si_write_unlock(sb);
26188 +
26189 +       return err;
26190 +}
26191 +
26192 +/* dentry and super_block lock. call at entry point */
26193 +int aufs_read_lock(struct dentry *dentry, int flags)
26194 +{
26195 +       int err;
26196 +       struct super_block *sb;
26197 +
26198 +       sb = dentry->d_sb;
26199 +       err = si_read_lock(sb, flags);
26200 +       if (unlikely(err))
26201 +               goto out;
26202 +
26203 +       if (au_ftest_lock(flags, DW))
26204 +               di_write_lock_child(dentry);
26205 +       else
26206 +               di_read_lock_child(dentry, flags);
26207 +
26208 +       if (au_ftest_lock(flags, GEN)) {
26209 +               err = au_digen_test(dentry, au_sigen(sb));
26210 +               AuDebugOn(!err && au_dbrange_test(dentry));
26211 +               if (unlikely(err))
26212 +                       aufs_read_unlock(dentry, flags);
26213 +       }
26214 +
26215 +out:
26216 +       return err;
26217 +}
26218 +
26219 +void aufs_read_unlock(struct dentry *dentry, int flags)
26220 +{
26221 +       if (au_ftest_lock(flags, DW))
26222 +               di_write_unlock(dentry);
26223 +       else
26224 +               di_read_unlock(dentry, flags);
26225 +       si_read_unlock(dentry->d_sb);
26226 +}
26227 +
26228 +void aufs_write_lock(struct dentry *dentry)
26229 +{
26230 +       si_write_lock(dentry->d_sb, AuLock_FLUSH | AuLock_NOPLMW);
26231 +       di_write_lock_child(dentry);
26232 +}
26233 +
26234 +void aufs_write_unlock(struct dentry *dentry)
26235 +{
26236 +       di_write_unlock(dentry);
26237 +       si_write_unlock(dentry->d_sb);
26238 +}
26239 +
26240 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags)
26241 +{
26242 +       int err;
26243 +       unsigned int sigen;
26244 +       struct super_block *sb;
26245 +
26246 +       sb = d1->d_sb;
26247 +       err = si_read_lock(sb, flags);
26248 +       if (unlikely(err))
26249 +               goto out;
26250 +
26251 +       di_write_lock2_child(d1, d2, au_ftest_lock(flags, DIR));
26252 +
26253 +       if (au_ftest_lock(flags, GEN)) {
26254 +               sigen = au_sigen(sb);
26255 +               err = au_digen_test(d1, sigen);
26256 +               AuDebugOn(!err && au_dbrange_test(d1));
26257 +               if (!err) {
26258 +                       err = au_digen_test(d2, sigen);
26259 +                       AuDebugOn(!err && au_dbrange_test(d2));
26260 +               }
26261 +               if (unlikely(err))
26262 +                       aufs_read_and_write_unlock2(d1, d2);
26263 +       }
26264 +
26265 +out:
26266 +       return err;
26267 +}
26268 +
26269 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2)
26270 +{
26271 +       di_write_unlock2(d1, d2);
26272 +       si_read_unlock(d1->d_sb);
26273 +}
26274 +
26275 +/* ---------------------------------------------------------------------- */
26276 +
26277 +int si_pid_test_slow(struct super_block *sb)
26278 +{
26279 +       void *p;
26280 +
26281 +       rcu_read_lock();
26282 +       p = radix_tree_lookup(&au_sbi(sb)->au_si_pid.tree, current->pid);
26283 +       rcu_read_unlock();
26284 +
26285 +       return (long)!!p;
26286 +}
26287 +
26288 +void si_pid_set_slow(struct super_block *sb)
26289 +{
26290 +       int err;
26291 +       struct au_sbinfo *sbinfo;
26292 +
26293 +       AuDebugOn(si_pid_test_slow(sb));
26294 +
26295 +       sbinfo = au_sbi(sb);
26296 +       err = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
26297 +       AuDebugOn(err);
26298 +       spin_lock(&sbinfo->au_si_pid.tree_lock);
26299 +       err = radix_tree_insert(&sbinfo->au_si_pid.tree, current->pid,
26300 +                               /*any valid ptr*/sb);
26301 +       spin_unlock(&sbinfo->au_si_pid.tree_lock);
26302 +       AuDebugOn(err);
26303 +       radix_tree_preload_end();
26304 +}
26305 +
26306 +void si_pid_clr_slow(struct super_block *sb)
26307 +{
26308 +       void *p;
26309 +       struct au_sbinfo *sbinfo;
26310 +
26311 +       AuDebugOn(!si_pid_test_slow(sb));
26312 +
26313 +       sbinfo = au_sbi(sb);
26314 +       spin_lock(&sbinfo->au_si_pid.tree_lock);
26315 +       p = radix_tree_delete(&sbinfo->au_si_pid.tree, current->pid);
26316 +       spin_unlock(&sbinfo->au_si_pid.tree_lock);
26317 +}
26318 diff -urN /usr/share/empty/fs/aufs/spl.h linux/fs/aufs/spl.h
26319 --- /usr/share/empty/fs/aufs/spl.h      1970-01-01 01:00:00.000000000 +0100
26320 +++ linux/fs/aufs/spl.h 2015-03-27 21:56:35.466795000 +0100
26321 @@ -0,0 +1,111 @@
26322 +/*
26323 + * Copyright (C) 2005-2015 Junjiro R. Okajima
26324 + *
26325 + * This program, aufs is free software; you can redistribute it and/or modify
26326 + * it under the terms of the GNU General Public License as published by
26327 + * the Free Software Foundation; either version 2 of the License, or
26328 + * (at your option) any later version.
26329 + *
26330 + * This program is distributed in the hope that it will be useful,
26331 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26332 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26333 + * GNU General Public License for more details.
26334 + *
26335 + * You should have received a copy of the GNU General Public License
26336 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26337 + */
26338 +
26339 +/*
26340 + * simple list protected by a spinlock
26341 + */
26342 +
26343 +#ifndef __AUFS_SPL_H__
26344 +#define __AUFS_SPL_H__
26345 +
26346 +#ifdef __KERNEL__
26347 +
26348 +struct au_splhead {
26349 +       spinlock_t              spin;
26350 +       struct list_head        head;
26351 +};
26352 +
26353 +static inline void au_spl_init(struct au_splhead *spl)
26354 +{
26355 +       spin_lock_init(&spl->spin);
26356 +       INIT_LIST_HEAD(&spl->head);
26357 +}
26358 +
26359 +static inline void au_spl_add(struct list_head *list, struct au_splhead *spl)
26360 +{
26361 +       spin_lock(&spl->spin);
26362 +       list_add(list, &spl->head);
26363 +       spin_unlock(&spl->spin);
26364 +}
26365 +
26366 +static inline void au_spl_del(struct list_head *list, struct au_splhead *spl)
26367 +{
26368 +       spin_lock(&spl->spin);
26369 +       list_del(list);
26370 +       spin_unlock(&spl->spin);
26371 +}
26372 +
26373 +static inline void au_spl_del_rcu(struct list_head *list,
26374 +                                 struct au_splhead *spl)
26375 +{
26376 +       spin_lock(&spl->spin);
26377 +       list_del_rcu(list);
26378 +       spin_unlock(&spl->spin);
26379 +}
26380 +
26381 +/* ---------------------------------------------------------------------- */
26382 +
26383 +struct au_sphlhead {
26384 +       spinlock_t              spin;
26385 +       struct hlist_head       head;
26386 +};
26387 +
26388 +static inline void au_sphl_init(struct au_sphlhead *sphl)
26389 +{
26390 +       spin_lock_init(&sphl->spin);
26391 +       INIT_HLIST_HEAD(&sphl->head);
26392 +}
26393 +
26394 +static inline void au_sphl_add(struct hlist_node *hlist,
26395 +                              struct au_sphlhead *sphl)
26396 +{
26397 +       spin_lock(&sphl->spin);
26398 +       hlist_add_head(hlist, &sphl->head);
26399 +       spin_unlock(&sphl->spin);
26400 +}
26401 +
26402 +static inline void au_sphl_del(struct hlist_node *hlist,
26403 +                              struct au_sphlhead *sphl)
26404 +{
26405 +       spin_lock(&sphl->spin);
26406 +       hlist_del(hlist);
26407 +       spin_unlock(&sphl->spin);
26408 +}
26409 +
26410 +static inline void au_sphl_del_rcu(struct hlist_node *hlist,
26411 +                                  struct au_sphlhead *sphl)
26412 +{
26413 +       spin_lock(&sphl->spin);
26414 +       hlist_del_rcu(hlist);
26415 +       spin_unlock(&sphl->spin);
26416 +}
26417 +
26418 +static inline unsigned long au_sphl_count(struct au_sphlhead *sphl)
26419 +{
26420 +       unsigned long cnt;
26421 +       struct hlist_node *pos;
26422 +
26423 +       cnt = 0;
26424 +       spin_lock(&sphl->spin);
26425 +       hlist_for_each(pos, &sphl->head)
26426 +               cnt++;
26427 +       spin_unlock(&sphl->spin);
26428 +       return cnt;
26429 +}
26430 +
26431 +#endif /* __KERNEL__ */
26432 +#endif /* __AUFS_SPL_H__ */
26433 diff -urN /usr/share/empty/fs/aufs/super.c linux/fs/aufs/super.c
26434 --- /usr/share/empty/fs/aufs/super.c    1970-01-01 01:00:00.000000000 +0100
26435 +++ linux/fs/aufs/super.c       2015-03-27 21:56:35.466795000 +0100
26436 @@ -0,0 +1,1009 @@
26437 +/*
26438 + * Copyright (C) 2005-2015 Junjiro R. Okajima
26439 + *
26440 + * This program, aufs is free software; you can redistribute it and/or modify
26441 + * it under the terms of the GNU General Public License as published by
26442 + * the Free Software Foundation; either version 2 of the License, or
26443 + * (at your option) any later version.
26444 + *
26445 + * This program is distributed in the hope that it will be useful,
26446 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26447 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26448 + * GNU General Public License for more details.
26449 + *
26450 + * You should have received a copy of the GNU General Public License
26451 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26452 + */
26453 +
26454 +/*
26455 + * mount and super_block operations
26456 + */
26457 +
26458 +#include <linux/mm.h>
26459 +#include <linux/module.h>
26460 +#include <linux/seq_file.h>
26461 +#include <linux/statfs.h>
26462 +#include <linux/vmalloc.h>
26463 +#include <linux/writeback.h>
26464 +#include "aufs.h"
26465 +
26466 +/*
26467 + * super_operations
26468 + */
26469 +static struct inode *aufs_alloc_inode(struct super_block *sb __maybe_unused)
26470 +{
26471 +       struct au_icntnr *c;
26472 +
26473 +       c = au_cache_alloc_icntnr();
26474 +       if (c) {
26475 +               au_icntnr_init(c);
26476 +               c->vfs_inode.i_version = 1; /* sigen(sb); */
26477 +               c->iinfo.ii_hinode = NULL;
26478 +               return &c->vfs_inode;
26479 +       }
26480 +       return NULL;
26481 +}
26482 +
26483 +static void aufs_destroy_inode_cb(struct rcu_head *head)
26484 +{
26485 +       struct inode *inode = container_of(head, struct inode, i_rcu);
26486 +
26487 +       INIT_HLIST_HEAD(&inode->i_dentry);
26488 +       au_cache_free_icntnr(container_of(inode, struct au_icntnr, vfs_inode));
26489 +}
26490 +
26491 +static void aufs_destroy_inode(struct inode *inode)
26492 +{
26493 +       au_iinfo_fin(inode);
26494 +       call_rcu(&inode->i_rcu, aufs_destroy_inode_cb);
26495 +}
26496 +
26497 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino)
26498 +{
26499 +       struct inode *inode;
26500 +       int err;
26501 +
26502 +       inode = iget_locked(sb, ino);
26503 +       if (unlikely(!inode)) {
26504 +               inode = ERR_PTR(-ENOMEM);
26505 +               goto out;
26506 +       }
26507 +       if (!(inode->i_state & I_NEW))
26508 +               goto out;
26509 +
26510 +       err = au_xigen_new(inode);
26511 +       if (!err)
26512 +               err = au_iinfo_init(inode);
26513 +       if (!err)
26514 +               inode->i_version++;
26515 +       else {
26516 +               iget_failed(inode);
26517 +               inode = ERR_PTR(err);
26518 +       }
26519 +
26520 +out:
26521 +       /* never return NULL */
26522 +       AuDebugOn(!inode);
26523 +       AuTraceErrPtr(inode);
26524 +       return inode;
26525 +}
26526 +
26527 +/* lock free root dinfo */
26528 +static int au_show_brs(struct seq_file *seq, struct super_block *sb)
26529 +{
26530 +       int err;
26531 +       aufs_bindex_t bindex, bend;
26532 +       struct path path;
26533 +       struct au_hdentry *hdp;
26534 +       struct au_branch *br;
26535 +       au_br_perm_str_t perm;
26536 +
26537 +       err = 0;
26538 +       bend = au_sbend(sb);
26539 +       hdp = au_di(sb->s_root)->di_hdentry;
26540 +       for (bindex = 0; !err && bindex <= bend; bindex++) {
26541 +               br = au_sbr(sb, bindex);
26542 +               path.mnt = au_br_mnt(br);
26543 +               path.dentry = hdp[bindex].hd_dentry;
26544 +               err = au_seq_path(seq, &path);
26545 +               if (err > 0) {
26546 +                       au_optstr_br_perm(&perm, br->br_perm);
26547 +                       err = seq_printf(seq, "=%s", perm.a);
26548 +                       if (err == -1)
26549 +                               err = -E2BIG;
26550 +               }
26551 +               if (!err && bindex != bend)
26552 +                       err = seq_putc(seq, ':');
26553 +       }
26554 +
26555 +       return err;
26556 +}
26557 +
26558 +static void au_show_wbr_create(struct seq_file *m, int v,
26559 +                              struct au_sbinfo *sbinfo)
26560 +{
26561 +       const char *pat;
26562 +
26563 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
26564 +
26565 +       seq_puts(m, ",create=");
26566 +       pat = au_optstr_wbr_create(v);
26567 +       switch (v) {
26568 +       case AuWbrCreate_TDP:
26569 +       case AuWbrCreate_RR:
26570 +       case AuWbrCreate_MFS:
26571 +       case AuWbrCreate_PMFS:
26572 +               seq_puts(m, pat);
26573 +               break;
26574 +       case AuWbrCreate_MFSV:
26575 +               seq_printf(m, /*pat*/"mfs:%lu",
26576 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26577 +                          / MSEC_PER_SEC);
26578 +               break;
26579 +       case AuWbrCreate_PMFSV:
26580 +               seq_printf(m, /*pat*/"pmfs:%lu",
26581 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26582 +                          / MSEC_PER_SEC);
26583 +               break;
26584 +       case AuWbrCreate_MFSRR:
26585 +               seq_printf(m, /*pat*/"mfsrr:%llu",
26586 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
26587 +               break;
26588 +       case AuWbrCreate_MFSRRV:
26589 +               seq_printf(m, /*pat*/"mfsrr:%llu:%lu",
26590 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
26591 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26592 +                          / MSEC_PER_SEC);
26593 +               break;
26594 +       case AuWbrCreate_PMFSRR:
26595 +               seq_printf(m, /*pat*/"pmfsrr:%llu",
26596 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
26597 +               break;
26598 +       case AuWbrCreate_PMFSRRV:
26599 +               seq_printf(m, /*pat*/"pmfsrr:%llu:%lu",
26600 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
26601 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
26602 +                          / MSEC_PER_SEC);
26603 +               break;
26604 +       }
26605 +}
26606 +
26607 +static int au_show_xino(struct seq_file *seq, struct super_block *sb)
26608 +{
26609 +#ifdef CONFIG_SYSFS
26610 +       return 0;
26611 +#else
26612 +       int err;
26613 +       const int len = sizeof(AUFS_XINO_FNAME) - 1;
26614 +       aufs_bindex_t bindex, brid;
26615 +       struct qstr *name;
26616 +       struct file *f;
26617 +       struct dentry *d, *h_root;
26618 +       struct au_hdentry *hdp;
26619 +
26620 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
26621 +
26622 +       err = 0;
26623 +       f = au_sbi(sb)->si_xib;
26624 +       if (!f)
26625 +               goto out;
26626 +
26627 +       /* stop printing the default xino path on the first writable branch */
26628 +       h_root = NULL;
26629 +       brid = au_xino_brid(sb);
26630 +       if (brid >= 0) {
26631 +               bindex = au_br_index(sb, brid);
26632 +               hdp = au_di(sb->s_root)->di_hdentry;
26633 +               h_root = hdp[0 + bindex].hd_dentry;
26634 +       }
26635 +       d = f->f_path.dentry;
26636 +       name = &d->d_name;
26637 +       /* safe ->d_parent because the file is unlinked */
26638 +       if (d->d_parent == h_root
26639 +           && name->len == len
26640 +           && !memcmp(name->name, AUFS_XINO_FNAME, len))
26641 +               goto out;
26642 +
26643 +       seq_puts(seq, ",xino=");
26644 +       err = au_xino_path(seq, f);
26645 +
26646 +out:
26647 +       return err;
26648 +#endif
26649 +}
26650 +
26651 +/* seq_file will re-call me in case of too long string */
26652 +static int aufs_show_options(struct seq_file *m, struct dentry *dentry)
26653 +{
26654 +       int err;
26655 +       unsigned int mnt_flags, v;
26656 +       struct super_block *sb;
26657 +       struct au_sbinfo *sbinfo;
26658 +
26659 +#define AuBool(name, str) do { \
26660 +       v = au_opt_test(mnt_flags, name); \
26661 +       if (v != au_opt_test(AuOpt_Def, name)) \
26662 +               seq_printf(m, ",%s" #str, v ? "" : "no"); \
26663 +} while (0)
26664 +
26665 +#define AuStr(name, str) do { \
26666 +       v = mnt_flags & AuOptMask_##name; \
26667 +       if (v != (AuOpt_Def & AuOptMask_##name)) \
26668 +               seq_printf(m, "," #str "=%s", au_optstr_##str(v)); \
26669 +} while (0)
26670 +
26671 +#define AuUInt(name, str, val) do { \
26672 +       if (val != AUFS_##name##_DEF) \
26673 +               seq_printf(m, "," #str "=%u", val); \
26674 +} while (0)
26675 +
26676 +       sb = dentry->d_sb;
26677 +       if (sb->s_flags & MS_POSIXACL)
26678 +               seq_puts(m, ",acl");
26679 +
26680 +       /* lock free root dinfo */
26681 +       si_noflush_read_lock(sb);
26682 +       sbinfo = au_sbi(sb);
26683 +       seq_printf(m, ",si=%lx", sysaufs_si_id(sbinfo));
26684 +
26685 +       mnt_flags = au_mntflags(sb);
26686 +       if (au_opt_test(mnt_flags, XINO)) {
26687 +               err = au_show_xino(m, sb);
26688 +               if (unlikely(err))
26689 +                       goto out;
26690 +       } else
26691 +               seq_puts(m, ",noxino");
26692 +
26693 +       AuBool(TRUNC_XINO, trunc_xino);
26694 +       AuStr(UDBA, udba);
26695 +       AuBool(SHWH, shwh);
26696 +       AuBool(PLINK, plink);
26697 +       AuBool(DIO, dio);
26698 +       AuBool(DIRPERM1, dirperm1);
26699 +       /* AuBool(REFROF, refrof); */
26700 +
26701 +       v = sbinfo->si_wbr_create;
26702 +       if (v != AuWbrCreate_Def)
26703 +               au_show_wbr_create(m, v, sbinfo);
26704 +
26705 +       v = sbinfo->si_wbr_copyup;
26706 +       if (v != AuWbrCopyup_Def)
26707 +               seq_printf(m, ",cpup=%s", au_optstr_wbr_copyup(v));
26708 +
26709 +       v = au_opt_test(mnt_flags, ALWAYS_DIROPQ);
26710 +       if (v != au_opt_test(AuOpt_Def, ALWAYS_DIROPQ))
26711 +               seq_printf(m, ",diropq=%c", v ? 'a' : 'w');
26712 +
26713 +       AuUInt(DIRWH, dirwh, sbinfo->si_dirwh);
26714 +
26715 +       v = jiffies_to_msecs(sbinfo->si_rdcache) / MSEC_PER_SEC;
26716 +       AuUInt(RDCACHE, rdcache, v);
26717 +
26718 +       AuUInt(RDBLK, rdblk, sbinfo->si_rdblk);
26719 +       AuUInt(RDHASH, rdhash, sbinfo->si_rdhash);
26720 +
26721 +       au_fhsm_show(m, sbinfo);
26722 +
26723 +       AuBool(SUM, sum);
26724 +       /* AuBool(SUM_W, wsum); */
26725 +       AuBool(WARN_PERM, warn_perm);
26726 +       AuBool(VERBOSE, verbose);
26727 +
26728 +out:
26729 +       /* be sure to print "br:" last */
26730 +       if (!sysaufs_brs) {
26731 +               seq_puts(m, ",br:");
26732 +               au_show_brs(m, sb);
26733 +       }
26734 +       si_read_unlock(sb);
26735 +       return 0;
26736 +
26737 +#undef AuBool
26738 +#undef AuStr
26739 +#undef AuUInt
26740 +}
26741 +
26742 +/* ---------------------------------------------------------------------- */
26743 +
26744 +/* sum mode which returns the summation for statfs(2) */
26745 +
26746 +static u64 au_add_till_max(u64 a, u64 b)
26747 +{
26748 +       u64 old;
26749 +
26750 +       old = a;
26751 +       a += b;
26752 +       if (old <= a)
26753 +               return a;
26754 +       return ULLONG_MAX;
26755 +}
26756 +
26757 +static u64 au_mul_till_max(u64 a, long mul)
26758 +{
26759 +       u64 old;
26760 +
26761 +       old = a;
26762 +       a *= mul;
26763 +       if (old <= a)
26764 +               return a;
26765 +       return ULLONG_MAX;
26766 +}
26767 +
26768 +static int au_statfs_sum(struct super_block *sb, struct kstatfs *buf)
26769 +{
26770 +       int err;
26771 +       long bsize, factor;
26772 +       u64 blocks, bfree, bavail, files, ffree;
26773 +       aufs_bindex_t bend, bindex, i;
26774 +       unsigned char shared;
26775 +       struct path h_path;
26776 +       struct super_block *h_sb;
26777 +
26778 +       err = 0;
26779 +       bsize = LONG_MAX;
26780 +       files = 0;
26781 +       ffree = 0;
26782 +       blocks = 0;
26783 +       bfree = 0;
26784 +       bavail = 0;
26785 +       bend = au_sbend(sb);
26786 +       for (bindex = 0; bindex <= bend; bindex++) {
26787 +               h_path.mnt = au_sbr_mnt(sb, bindex);
26788 +               h_sb = h_path.mnt->mnt_sb;
26789 +               shared = 0;
26790 +               for (i = 0; !shared && i < bindex; i++)
26791 +                       shared = (au_sbr_sb(sb, i) == h_sb);
26792 +               if (shared)
26793 +                       continue;
26794 +
26795 +               /* sb->s_root for NFS is unreliable */
26796 +               h_path.dentry = h_path.mnt->mnt_root;
26797 +               err = vfs_statfs(&h_path, buf);
26798 +               if (unlikely(err))
26799 +                       goto out;
26800 +
26801 +               if (bsize > buf->f_bsize) {
26802 +                       /*
26803 +                        * we will reduce bsize, so we have to expand blocks
26804 +                        * etc. to match them again
26805 +                        */
26806 +                       factor = (bsize / buf->f_bsize);
26807 +                       blocks = au_mul_till_max(blocks, factor);
26808 +                       bfree = au_mul_till_max(bfree, factor);
26809 +                       bavail = au_mul_till_max(bavail, factor);
26810 +                       bsize = buf->f_bsize;
26811 +               }
26812 +
26813 +               factor = (buf->f_bsize / bsize);
26814 +               blocks = au_add_till_max(blocks,
26815 +                               au_mul_till_max(buf->f_blocks, factor));
26816 +               bfree = au_add_till_max(bfree,
26817 +                               au_mul_till_max(buf->f_bfree, factor));
26818 +               bavail = au_add_till_max(bavail,
26819 +                               au_mul_till_max(buf->f_bavail, factor));
26820 +               files = au_add_till_max(files, buf->f_files);
26821 +               ffree = au_add_till_max(ffree, buf->f_ffree);
26822 +       }
26823 +
26824 +       buf->f_bsize = bsize;
26825 +       buf->f_blocks = blocks;
26826 +       buf->f_bfree = bfree;
26827 +       buf->f_bavail = bavail;
26828 +       buf->f_files = files;
26829 +       buf->f_ffree = ffree;
26830 +       buf->f_frsize = 0;
26831 +
26832 +out:
26833 +       return err;
26834 +}
26835 +
26836 +static int aufs_statfs(struct dentry *dentry, struct kstatfs *buf)
26837 +{
26838 +       int err;
26839 +       struct path h_path;
26840 +       struct super_block *sb;
26841 +
26842 +       /* lock free root dinfo */
26843 +       sb = dentry->d_sb;
26844 +       si_noflush_read_lock(sb);
26845 +       if (!au_opt_test(au_mntflags(sb), SUM)) {
26846 +               /* sb->s_root for NFS is unreliable */
26847 +               h_path.mnt = au_sbr_mnt(sb, 0);
26848 +               h_path.dentry = h_path.mnt->mnt_root;
26849 +               err = vfs_statfs(&h_path, buf);
26850 +       } else
26851 +               err = au_statfs_sum(sb, buf);
26852 +       si_read_unlock(sb);
26853 +
26854 +       if (!err) {
26855 +               buf->f_type = AUFS_SUPER_MAGIC;
26856 +               buf->f_namelen = AUFS_MAX_NAMELEN;
26857 +               memset(&buf->f_fsid, 0, sizeof(buf->f_fsid));
26858 +       }
26859 +       /* buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1; */
26860 +
26861 +       return err;
26862 +}
26863 +
26864 +/* ---------------------------------------------------------------------- */
26865 +
26866 +static int aufs_sync_fs(struct super_block *sb, int wait)
26867 +{
26868 +       int err, e;
26869 +       aufs_bindex_t bend, bindex;
26870 +       struct au_branch *br;
26871 +       struct super_block *h_sb;
26872 +
26873 +       err = 0;
26874 +       si_noflush_read_lock(sb);
26875 +       bend = au_sbend(sb);
26876 +       for (bindex = 0; bindex <= bend; bindex++) {
26877 +               br = au_sbr(sb, bindex);
26878 +               if (!au_br_writable(br->br_perm))
26879 +                       continue;
26880 +
26881 +               h_sb = au_sbr_sb(sb, bindex);
26882 +               if (h_sb->s_op->sync_fs) {
26883 +                       e = h_sb->s_op->sync_fs(h_sb, wait);
26884 +                       if (unlikely(e && !err))
26885 +                               err = e;
26886 +                       /* go on even if an error happens */
26887 +               }
26888 +       }
26889 +       si_read_unlock(sb);
26890 +
26891 +       return err;
26892 +}
26893 +
26894 +/* ---------------------------------------------------------------------- */
26895 +
26896 +/* final actions when unmounting a file system */
26897 +static void aufs_put_super(struct super_block *sb)
26898 +{
26899 +       struct au_sbinfo *sbinfo;
26900 +
26901 +       sbinfo = au_sbi(sb);
26902 +       if (!sbinfo)
26903 +               return;
26904 +
26905 +       dbgaufs_si_fin(sbinfo);
26906 +       kobject_put(&sbinfo->si_kobj);
26907 +}
26908 +
26909 +/* ---------------------------------------------------------------------- */
26910 +
26911 +void au_array_free(void *array)
26912 +{
26913 +       if (array) {
26914 +               if (!is_vmalloc_addr(array))
26915 +                       kfree(array);
26916 +               else
26917 +                       vfree(array);
26918 +       }
26919 +}
26920 +
26921 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb, void *arg)
26922 +{
26923 +       void *array;
26924 +       unsigned long long n, sz;
26925 +
26926 +       array = NULL;
26927 +       n = 0;
26928 +       if (!*hint)
26929 +               goto out;
26930 +
26931 +       if (*hint > ULLONG_MAX / sizeof(array)) {
26932 +               array = ERR_PTR(-EMFILE);
26933 +               pr_err("hint %llu\n", *hint);
26934 +               goto out;
26935 +       }
26936 +
26937 +       sz = sizeof(array) * *hint;
26938 +       array = kzalloc(sz, GFP_NOFS);
26939 +       if (unlikely(!array))
26940 +               array = vzalloc(sz);
26941 +       if (unlikely(!array)) {
26942 +               array = ERR_PTR(-ENOMEM);
26943 +               goto out;
26944 +       }
26945 +
26946 +       n = cb(array, *hint, arg);
26947 +       AuDebugOn(n > *hint);
26948 +
26949 +out:
26950 +       *hint = n;
26951 +       return array;
26952 +}
26953 +
26954 +static unsigned long long au_iarray_cb(void *a,
26955 +                                      unsigned long long max __maybe_unused,
26956 +                                      void *arg)
26957 +{
26958 +       unsigned long long n;
26959 +       struct inode **p, *inode;
26960 +       struct list_head *head;
26961 +
26962 +       n = 0;
26963 +       p = a;
26964 +       head = arg;
26965 +       spin_lock(&inode_sb_list_lock);
26966 +       list_for_each_entry(inode, head, i_sb_list) {
26967 +               if (!is_bad_inode(inode)
26968 +                   && au_ii(inode)->ii_bstart >= 0) {
26969 +                       spin_lock(&inode->i_lock);
26970 +                       if (atomic_read(&inode->i_count)) {
26971 +                               au_igrab(inode);
26972 +                               *p++ = inode;
26973 +                               n++;
26974 +                               AuDebugOn(n > max);
26975 +                       }
26976 +                       spin_unlock(&inode->i_lock);
26977 +               }
26978 +       }
26979 +       spin_unlock(&inode_sb_list_lock);
26980 +
26981 +       return n;
26982 +}
26983 +
26984 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max)
26985 +{
26986 +       *max = atomic_long_read(&au_sbi(sb)->si_ninodes);
26987 +       return au_array_alloc(max, au_iarray_cb, &sb->s_inodes);
26988 +}
26989 +
26990 +void au_iarray_free(struct inode **a, unsigned long long max)
26991 +{
26992 +       unsigned long long ull;
26993 +
26994 +       for (ull = 0; ull < max; ull++)
26995 +               iput(a[ull]);
26996 +       au_array_free(a);
26997 +}
26998 +
26999 +/* ---------------------------------------------------------------------- */
27000 +
27001 +/*
27002 + * refresh dentry and inode at remount time.
27003 + */
27004 +/* todo: consolidate with simple_reval_dpath() and au_reval_for_attr() */
27005 +static int au_do_refresh(struct dentry *dentry, unsigned int dir_flags,
27006 +                     struct dentry *parent)
27007 +{
27008 +       int err;
27009 +
27010 +       di_write_lock_child(dentry);
27011 +       di_read_lock_parent(parent, AuLock_IR);
27012 +       err = au_refresh_dentry(dentry, parent);
27013 +       if (!err && dir_flags)
27014 +               au_hn_reset(dentry->d_inode, dir_flags);
27015 +       di_read_unlock(parent, AuLock_IR);
27016 +       di_write_unlock(dentry);
27017 +
27018 +       return err;
27019 +}
27020 +
27021 +static int au_do_refresh_d(struct dentry *dentry, unsigned int sigen,
27022 +                          struct au_sbinfo *sbinfo,
27023 +                          const unsigned int dir_flags)
27024 +{
27025 +       int err;
27026 +       struct dentry *parent;
27027 +       struct inode *inode;
27028 +
27029 +       err = 0;
27030 +       parent = dget_parent(dentry);
27031 +       if (!au_digen_test(parent, sigen) && au_digen_test(dentry, sigen)) {
27032 +               inode = dentry->d_inode;
27033 +               if (inode) {
27034 +                       if (!S_ISDIR(inode->i_mode))
27035 +                               err = au_do_refresh(dentry, /*dir_flags*/0,
27036 +                                                parent);
27037 +                       else {
27038 +                               err = au_do_refresh(dentry, dir_flags, parent);
27039 +                               if (unlikely(err))
27040 +                                       au_fset_si(sbinfo, FAILED_REFRESH_DIR);
27041 +                       }
27042 +               } else
27043 +                       err = au_do_refresh(dentry, /*dir_flags*/0, parent);
27044 +               AuDbgDentry(dentry);
27045 +       }
27046 +       dput(parent);
27047 +
27048 +       AuTraceErr(err);
27049 +       return err;
27050 +}
27051 +
27052 +static int au_refresh_d(struct super_block *sb)
27053 +{
27054 +       int err, i, j, ndentry, e;
27055 +       unsigned int sigen;
27056 +       struct au_dcsub_pages dpages;
27057 +       struct au_dpage *dpage;
27058 +       struct dentry **dentries, *d;
27059 +       struct au_sbinfo *sbinfo;
27060 +       struct dentry *root = sb->s_root;
27061 +       const unsigned int dir_flags = au_hi_flags(root->d_inode, /*isdir*/1);
27062 +
27063 +       err = au_dpages_init(&dpages, GFP_NOFS);
27064 +       if (unlikely(err))
27065 +               goto out;
27066 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
27067 +       if (unlikely(err))
27068 +               goto out_dpages;
27069 +
27070 +       sigen = au_sigen(sb);
27071 +       sbinfo = au_sbi(sb);
27072 +       for (i = 0; i < dpages.ndpage; i++) {
27073 +               dpage = dpages.dpages + i;
27074 +               dentries = dpage->dentries;
27075 +               ndentry = dpage->ndentry;
27076 +               for (j = 0; j < ndentry; j++) {
27077 +                       d = dentries[j];
27078 +                       e = au_do_refresh_d(d, sigen, sbinfo, dir_flags);
27079 +                       if (unlikely(e && !err))
27080 +                               err = e;
27081 +                       /* go on even err */
27082 +               }
27083 +       }
27084 +
27085 +out_dpages:
27086 +       au_dpages_free(&dpages);
27087 +out:
27088 +       return err;
27089 +}
27090 +
27091 +static int au_refresh_i(struct super_block *sb)
27092 +{
27093 +       int err, e;
27094 +       unsigned int sigen;
27095 +       unsigned long long max, ull;
27096 +       struct inode *inode, **array;
27097 +
27098 +       array = au_iarray_alloc(sb, &max);
27099 +       err = PTR_ERR(array);
27100 +       if (IS_ERR(array))
27101 +               goto out;
27102 +
27103 +       err = 0;
27104 +       sigen = au_sigen(sb);
27105 +       for (ull = 0; ull < max; ull++) {
27106 +               inode = array[ull];
27107 +               if (unlikely(!inode))
27108 +                       break;
27109 +               if (au_iigen(inode, NULL) != sigen) {
27110 +                       ii_write_lock_child(inode);
27111 +                       e = au_refresh_hinode_self(inode);
27112 +                       ii_write_unlock(inode);
27113 +                       if (unlikely(e)) {
27114 +                               pr_err("error %d, i%lu\n", e, inode->i_ino);
27115 +                               if (!err)
27116 +                                       err = e;
27117 +                               /* go on even if err */
27118 +                       }
27119 +               }
27120 +       }
27121 +
27122 +       au_iarray_free(array, max);
27123 +
27124 +out:
27125 +       return err;
27126 +}
27127 +
27128 +static void au_remount_refresh(struct super_block *sb)
27129 +{
27130 +       int err, e;
27131 +       unsigned int udba;
27132 +       aufs_bindex_t bindex, bend;
27133 +       struct dentry *root;
27134 +       struct inode *inode;
27135 +       struct au_branch *br;
27136 +
27137 +       au_sigen_inc(sb);
27138 +       au_fclr_si(au_sbi(sb), FAILED_REFRESH_DIR);
27139 +
27140 +       root = sb->s_root;
27141 +       DiMustNoWaiters(root);
27142 +       inode = root->d_inode;
27143 +       IiMustNoWaiters(inode);
27144 +
27145 +       udba = au_opt_udba(sb);
27146 +       bend = au_sbend(sb);
27147 +       for (bindex = 0; bindex <= bend; bindex++) {
27148 +               br = au_sbr(sb, bindex);
27149 +               err = au_hnotify_reset_br(udba, br, br->br_perm);
27150 +               if (unlikely(err))
27151 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
27152 +                               bindex, err);
27153 +               /* go on even if err */
27154 +       }
27155 +       au_hn_reset(inode, au_hi_flags(inode, /*isdir*/1));
27156 +
27157 +       di_write_unlock(root);
27158 +       err = au_refresh_d(sb);
27159 +       e = au_refresh_i(sb);
27160 +       if (unlikely(e && !err))
27161 +               err = e;
27162 +       /* aufs_write_lock() calls ..._child() */
27163 +       di_write_lock_child(root);
27164 +
27165 +       au_cpup_attr_all(inode, /*force*/1);
27166 +
27167 +       if (unlikely(err))
27168 +               AuIOErr("refresh failed, ignored, %d\n", err);
27169 +}
27170 +
27171 +/* stop extra interpretation of errno in mount(8), and strange error messages */
27172 +static int cvt_err(int err)
27173 +{
27174 +       AuTraceErr(err);
27175 +
27176 +       switch (err) {
27177 +       case -ENOENT:
27178 +       case -ENOTDIR:
27179 +       case -EEXIST:
27180 +       case -EIO:
27181 +               err = -EINVAL;
27182 +       }
27183 +       return err;
27184 +}
27185 +
27186 +static int aufs_remount_fs(struct super_block *sb, int *flags, char *data)
27187 +{
27188 +       int err, do_dx;
27189 +       unsigned int mntflags;
27190 +       struct au_opts opts;
27191 +       struct dentry *root;
27192 +       struct inode *inode;
27193 +       struct au_sbinfo *sbinfo;
27194 +
27195 +       err = 0;
27196 +       root = sb->s_root;
27197 +       if (!data || !*data) {
27198 +               err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
27199 +               if (!err) {
27200 +                       di_write_lock_child(root);
27201 +                       err = au_opts_verify(sb, *flags, /*pending*/0);
27202 +                       aufs_write_unlock(root);
27203 +               }
27204 +               goto out;
27205 +       }
27206 +
27207 +       err = -ENOMEM;
27208 +       memset(&opts, 0, sizeof(opts));
27209 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
27210 +       if (unlikely(!opts.opt))
27211 +               goto out;
27212 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
27213 +       opts.flags = AuOpts_REMOUNT;
27214 +       opts.sb_flags = *flags;
27215 +
27216 +       /* parse it before aufs lock */
27217 +       err = au_opts_parse(sb, data, &opts);
27218 +       if (unlikely(err))
27219 +               goto out_opts;
27220 +
27221 +       sbinfo = au_sbi(sb);
27222 +       inode = root->d_inode;
27223 +       mutex_lock(&inode->i_mutex);
27224 +       err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
27225 +       if (unlikely(err))
27226 +               goto out_mtx;
27227 +       di_write_lock_child(root);
27228 +
27229 +       /* au_opts_remount() may return an error */
27230 +       err = au_opts_remount(sb, &opts);
27231 +       au_opts_free(&opts);
27232 +
27233 +       if (au_ftest_opts(opts.flags, REFRESH))
27234 +               au_remount_refresh(sb);
27235 +
27236 +       if (au_ftest_opts(opts.flags, REFRESH_DYAOP)) {
27237 +               mntflags = au_mntflags(sb);
27238 +               do_dx = !!au_opt_test(mntflags, DIO);
27239 +               au_dy_arefresh(do_dx);
27240 +       }
27241 +
27242 +       au_fhsm_wrote_all(sb, /*force*/1); /* ?? */
27243 +       aufs_write_unlock(root);
27244 +
27245 +out_mtx:
27246 +       mutex_unlock(&inode->i_mutex);
27247 +out_opts:
27248 +       free_page((unsigned long)opts.opt);
27249 +out:
27250 +       err = cvt_err(err);
27251 +       AuTraceErr(err);
27252 +       return err;
27253 +}
27254 +
27255 +static const struct super_operations aufs_sop = {
27256 +       .alloc_inode    = aufs_alloc_inode,
27257 +       .destroy_inode  = aufs_destroy_inode,
27258 +       /* always deleting, no clearing */
27259 +       .drop_inode     = generic_delete_inode,
27260 +       .show_options   = aufs_show_options,
27261 +       .statfs         = aufs_statfs,
27262 +       .put_super      = aufs_put_super,
27263 +       .sync_fs        = aufs_sync_fs,
27264 +       .remount_fs     = aufs_remount_fs
27265 +};
27266 +
27267 +/* ---------------------------------------------------------------------- */
27268 +
27269 +static int alloc_root(struct super_block *sb)
27270 +{
27271 +       int err;
27272 +       struct inode *inode;
27273 +       struct dentry *root;
27274 +
27275 +       err = -ENOMEM;
27276 +       inode = au_iget_locked(sb, AUFS_ROOT_INO);
27277 +       err = PTR_ERR(inode);
27278 +       if (IS_ERR(inode))
27279 +               goto out;
27280 +
27281 +       inode->i_op = &aufs_dir_iop;
27282 +       inode->i_fop = &aufs_dir_fop;
27283 +       inode->i_mode = S_IFDIR;
27284 +       set_nlink(inode, 2);
27285 +       unlock_new_inode(inode);
27286 +
27287 +       root = d_make_root(inode);
27288 +       if (unlikely(!root))
27289 +               goto out;
27290 +       err = PTR_ERR(root);
27291 +       if (IS_ERR(root))
27292 +               goto out;
27293 +
27294 +       err = au_di_init(root);
27295 +       if (!err) {
27296 +               sb->s_root = root;
27297 +               return 0; /* success */
27298 +       }
27299 +       dput(root);
27300 +
27301 +out:
27302 +       return err;
27303 +}
27304 +
27305 +static int aufs_fill_super(struct super_block *sb, void *raw_data,
27306 +                          int silent __maybe_unused)
27307 +{
27308 +       int err;
27309 +       struct au_opts opts;
27310 +       struct dentry *root;
27311 +       struct inode *inode;
27312 +       char *arg = raw_data;
27313 +
27314 +       if (unlikely(!arg || !*arg)) {
27315 +               err = -EINVAL;
27316 +               pr_err("no arg\n");
27317 +               goto out;
27318 +       }
27319 +
27320 +       err = -ENOMEM;
27321 +       memset(&opts, 0, sizeof(opts));
27322 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
27323 +       if (unlikely(!opts.opt))
27324 +               goto out;
27325 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
27326 +       opts.sb_flags = sb->s_flags;
27327 +
27328 +       err = au_si_alloc(sb);
27329 +       if (unlikely(err))
27330 +               goto out_opts;
27331 +
27332 +       /* all timestamps always follow the ones on the branch */
27333 +       sb->s_flags |= MS_NOATIME | MS_NODIRATIME;
27334 +       sb->s_op = &aufs_sop;
27335 +       sb->s_d_op = &aufs_dop;
27336 +       sb->s_magic = AUFS_SUPER_MAGIC;
27337 +       sb->s_maxbytes = 0;
27338 +       sb->s_stack_depth = 1;
27339 +       au_export_init(sb);
27340 +       /* au_xattr_init(sb); */
27341 +
27342 +       err = alloc_root(sb);
27343 +       if (unlikely(err)) {
27344 +               si_write_unlock(sb);
27345 +               goto out_info;
27346 +       }
27347 +       root = sb->s_root;
27348 +       inode = root->d_inode;
27349 +
27350 +       /*
27351 +        * actually we can parse options regardless aufs lock here.
27352 +        * but at remount time, parsing must be done before aufs lock.
27353 +        * so we follow the same rule.
27354 +        */
27355 +       ii_write_lock_parent(inode);
27356 +       aufs_write_unlock(root);
27357 +       err = au_opts_parse(sb, arg, &opts);
27358 +       if (unlikely(err))
27359 +               goto out_root;
27360 +
27361 +       /* lock vfs_inode first, then aufs. */
27362 +       mutex_lock(&inode->i_mutex);
27363 +       aufs_write_lock(root);
27364 +       err = au_opts_mount(sb, &opts);
27365 +       au_opts_free(&opts);
27366 +       aufs_write_unlock(root);
27367 +       mutex_unlock(&inode->i_mutex);
27368 +       if (!err)
27369 +               goto out_opts; /* success */
27370 +
27371 +out_root:
27372 +       dput(root);
27373 +       sb->s_root = NULL;
27374 +out_info:
27375 +       dbgaufs_si_fin(au_sbi(sb));
27376 +       kobject_put(&au_sbi(sb)->si_kobj);
27377 +       sb->s_fs_info = NULL;
27378 +out_opts:
27379 +       free_page((unsigned long)opts.opt);
27380 +out:
27381 +       AuTraceErr(err);
27382 +       err = cvt_err(err);
27383 +       AuTraceErr(err);
27384 +       return err;
27385 +}
27386 +
27387 +/* ---------------------------------------------------------------------- */
27388 +
27389 +static struct dentry *aufs_mount(struct file_system_type *fs_type, int flags,
27390 +                                const char *dev_name __maybe_unused,
27391 +                                void *raw_data)
27392 +{
27393 +       struct dentry *root;
27394 +       struct super_block *sb;
27395 +
27396 +       /* all timestamps always follow the ones on the branch */
27397 +       /* mnt->mnt_flags |= MNT_NOATIME | MNT_NODIRATIME; */
27398 +       root = mount_nodev(fs_type, flags, raw_data, aufs_fill_super);
27399 +       if (IS_ERR(root))
27400 +               goto out;
27401 +
27402 +       sb = root->d_sb;
27403 +       si_write_lock(sb, !AuLock_FLUSH);
27404 +       sysaufs_brs_add(sb, 0);
27405 +       si_write_unlock(sb);
27406 +       au_sbilist_add(sb);
27407 +
27408 +out:
27409 +       return root;
27410 +}
27411 +
27412 +static void aufs_kill_sb(struct super_block *sb)
27413 +{
27414 +       struct au_sbinfo *sbinfo;
27415 +
27416 +       sbinfo = au_sbi(sb);
27417 +       if (sbinfo) {
27418 +               au_sbilist_del(sb);
27419 +               aufs_write_lock(sb->s_root);
27420 +               au_fhsm_fin(sb);
27421 +               if (sbinfo->si_wbr_create_ops->fin)
27422 +                       sbinfo->si_wbr_create_ops->fin(sb);
27423 +               if (au_opt_test(sbinfo->si_mntflags, UDBA_HNOTIFY)) {
27424 +                       au_opt_set_udba(sbinfo->si_mntflags, UDBA_NONE);
27425 +                       au_remount_refresh(sb);
27426 +               }
27427 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
27428 +                       au_plink_put(sb, /*verbose*/1);
27429 +               au_xino_clr(sb);
27430 +               sbinfo->si_sb = NULL;
27431 +               aufs_write_unlock(sb->s_root);
27432 +               au_nwt_flush(&sbinfo->si_nowait);
27433 +       }
27434 +       kill_anon_super(sb);
27435 +}
27436 +
27437 +struct file_system_type aufs_fs_type = {
27438 +       .name           = AUFS_FSTYPE,
27439 +       /* a race between rename and others */
27440 +       .fs_flags       = FS_RENAME_DOES_D_MOVE,
27441 +       .mount          = aufs_mount,
27442 +       .kill_sb        = aufs_kill_sb,
27443 +       /* no need to __module_get() and module_put(). */
27444 +       .owner          = THIS_MODULE,
27445 +};
27446 diff -urN /usr/share/empty/fs/aufs/super.h linux/fs/aufs/super.h
27447 --- /usr/share/empty/fs/aufs/super.h    1970-01-01 01:00:00.000000000 +0100
27448 +++ linux/fs/aufs/super.h       2015-03-27 21:56:35.466795000 +0100
27449 @@ -0,0 +1,641 @@
27450 +/*
27451 + * Copyright (C) 2005-2015 Junjiro R. Okajima
27452 + *
27453 + * This program, aufs is free software; you can redistribute it and/or modify
27454 + * it under the terms of the GNU General Public License as published by
27455 + * the Free Software Foundation; either version 2 of the License, or
27456 + * (at your option) any later version.
27457 + *
27458 + * This program is distributed in the hope that it will be useful,
27459 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27460 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27461 + * GNU General Public License for more details.
27462 + *
27463 + * You should have received a copy of the GNU General Public License
27464 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27465 + */
27466 +
27467 +/*
27468 + * super_block operations
27469 + */
27470 +
27471 +#ifndef __AUFS_SUPER_H__
27472 +#define __AUFS_SUPER_H__
27473 +
27474 +#ifdef __KERNEL__
27475 +
27476 +#include <linux/fs.h>
27477 +#include "rwsem.h"
27478 +#include "spl.h"
27479 +#include "wkq.h"
27480 +
27481 +typedef ssize_t (*au_readf_t)(struct file *, char __user *, size_t, loff_t *);
27482 +typedef ssize_t (*au_writef_t)(struct file *, const char __user *, size_t,
27483 +                              loff_t *);
27484 +
27485 +/* policies to select one among multiple writable branches */
27486 +struct au_wbr_copyup_operations {
27487 +       int (*copyup)(struct dentry *dentry);
27488 +};
27489 +
27490 +#define AuWbr_DIR      1               /* target is a dir */
27491 +#define AuWbr_PARENT   (1 << 1)        /* always require a parent */
27492 +
27493 +#define au_ftest_wbr(flags, name)      ((flags) & AuWbr_##name)
27494 +#define au_fset_wbr(flags, name)       { (flags) |= AuWbr_##name; }
27495 +#define au_fclr_wbr(flags, name)       { (flags) &= ~AuWbr_##name; }
27496 +
27497 +struct au_wbr_create_operations {
27498 +       int (*create)(struct dentry *dentry, unsigned int flags);
27499 +       int (*init)(struct super_block *sb);
27500 +       int (*fin)(struct super_block *sb);
27501 +};
27502 +
27503 +struct au_wbr_mfs {
27504 +       struct mutex    mfs_lock; /* protect this structure */
27505 +       unsigned long   mfs_jiffy;
27506 +       unsigned long   mfs_expire;
27507 +       aufs_bindex_t   mfs_bindex;
27508 +
27509 +       unsigned long long      mfsrr_bytes;
27510 +       unsigned long long      mfsrr_watermark;
27511 +};
27512 +
27513 +struct pseudo_link {
27514 +       union {
27515 +               struct hlist_node hlist;
27516 +               struct rcu_head rcu;
27517 +       };
27518 +       struct inode *inode;
27519 +};
27520 +
27521 +#define AuPlink_NHASH 100
27522 +static inline int au_plink_hash(ino_t ino)
27523 +{
27524 +       return ino % AuPlink_NHASH;
27525 +}
27526 +
27527 +/* File-based Hierarchical Storage Management */
27528 +struct au_fhsm {
27529 +#ifdef CONFIG_AUFS_FHSM
27530 +       /* allow only one process who can receive the notification */
27531 +       spinlock_t              fhsm_spin;
27532 +       pid_t                   fhsm_pid;
27533 +       wait_queue_head_t       fhsm_wqh;
27534 +       atomic_t                fhsm_readable;
27535 +
27536 +       /* these are protected by si_rwsem */
27537 +       unsigned long           fhsm_expire;
27538 +       aufs_bindex_t           fhsm_bottom;
27539 +#endif
27540 +};
27541 +
27542 +struct au_branch;
27543 +struct au_sbinfo {
27544 +       /* nowait tasks in the system-wide workqueue */
27545 +       struct au_nowait_tasks  si_nowait;
27546 +
27547 +       /*
27548 +        * tried sb->s_umount, but failed due to the dependecy between i_mutex.
27549 +        * rwsem for au_sbinfo is necessary.
27550 +        */
27551 +       struct au_rwsem         si_rwsem;
27552 +
27553 +       /* prevent recursive locking in deleting inode */
27554 +       struct {
27555 +               unsigned long           *bitmap;
27556 +               spinlock_t              tree_lock;
27557 +               struct radix_tree_root  tree;
27558 +       } au_si_pid;
27559 +
27560 +       /*
27561 +        * dirty approach to protect sb->sb_inodes and ->s_files (gone) from
27562 +        * remount.
27563 +        */
27564 +       atomic_long_t           si_ninodes, si_nfiles;
27565 +
27566 +       /* branch management */
27567 +       unsigned int            si_generation;
27568 +
27569 +       /* see AuSi_ flags */
27570 +       unsigned char           au_si_status;
27571 +
27572 +       aufs_bindex_t           si_bend;
27573 +
27574 +       /* dirty trick to keep br_id plus */
27575 +       unsigned int            si_last_br_id :
27576 +                               sizeof(aufs_bindex_t) * BITS_PER_BYTE - 1;
27577 +       struct au_branch        **si_branch;
27578 +
27579 +       /* policy to select a writable branch */
27580 +       unsigned char           si_wbr_copyup;
27581 +       unsigned char           si_wbr_create;
27582 +       struct au_wbr_copyup_operations *si_wbr_copyup_ops;
27583 +       struct au_wbr_create_operations *si_wbr_create_ops;
27584 +
27585 +       /* round robin */
27586 +       atomic_t                si_wbr_rr_next;
27587 +
27588 +       /* most free space */
27589 +       struct au_wbr_mfs       si_wbr_mfs;
27590 +
27591 +       /* File-based Hierarchical Storage Management */
27592 +       struct au_fhsm          si_fhsm;
27593 +
27594 +       /* mount flags */
27595 +       /* include/asm-ia64/siginfo.h defines a macro named si_flags */
27596 +       unsigned int            si_mntflags;
27597 +
27598 +       /* external inode number (bitmap and translation table) */
27599 +       au_readf_t              si_xread;
27600 +       au_writef_t             si_xwrite;
27601 +       struct file             *si_xib;
27602 +       struct mutex            si_xib_mtx; /* protect xib members */
27603 +       unsigned long           *si_xib_buf;
27604 +       unsigned long           si_xib_last_pindex;
27605 +       int                     si_xib_next_bit;
27606 +       aufs_bindex_t           si_xino_brid;
27607 +       unsigned long           si_xino_jiffy;
27608 +       unsigned long           si_xino_expire;
27609 +       /* reserved for future use */
27610 +       /* unsigned long long   si_xib_limit; */        /* Max xib file size */
27611 +
27612 +#ifdef CONFIG_AUFS_EXPORT
27613 +       /* i_generation */
27614 +       struct file             *si_xigen;
27615 +       atomic_t                si_xigen_next;
27616 +#endif
27617 +
27618 +       /* vdir parameters */
27619 +       unsigned long           si_rdcache;     /* max cache time in jiffies */
27620 +       unsigned int            si_rdblk;       /* deblk size */
27621 +       unsigned int            si_rdhash;      /* hash size */
27622 +
27623 +       /*
27624 +        * If the number of whiteouts are larger than si_dirwh, leave all of
27625 +        * them after au_whtmp_ren to reduce the cost of rmdir(2).
27626 +        * future fsck.aufs or kernel thread will remove them later.
27627 +        * Otherwise, remove all whiteouts and the dir in rmdir(2).
27628 +        */
27629 +       unsigned int            si_dirwh;
27630 +
27631 +       /*
27632 +        * rename(2) a directory with all children.
27633 +        */
27634 +       /* reserved for future use */
27635 +       /* int                  si_rendir; */
27636 +
27637 +       /* pseudo_link list */
27638 +       struct au_sphlhead      si_plink[AuPlink_NHASH];
27639 +       wait_queue_head_t       si_plink_wq;
27640 +       spinlock_t              si_plink_maint_lock;
27641 +       pid_t                   si_plink_maint_pid;
27642 +
27643 +       /* file list */
27644 +       struct au_sphlhead      si_files;
27645 +
27646 +       /*
27647 +        * sysfs and lifetime management.
27648 +        * this is not a small structure and it may be a waste of memory in case
27649 +        * of sysfs is disabled, particulary when many aufs-es are mounted.
27650 +        * but using sysfs is majority.
27651 +        */
27652 +       struct kobject          si_kobj;
27653 +#ifdef CONFIG_DEBUG_FS
27654 +       struct dentry            *si_dbgaufs;
27655 +       struct dentry            *si_dbgaufs_plink;
27656 +       struct dentry            *si_dbgaufs_xib;
27657 +#ifdef CONFIG_AUFS_EXPORT
27658 +       struct dentry            *si_dbgaufs_xigen;
27659 +#endif
27660 +#endif
27661 +
27662 +#ifdef CONFIG_AUFS_SBILIST
27663 +       struct list_head        si_list;
27664 +#endif
27665 +
27666 +       /* dirty, necessary for unmounting, sysfs and sysrq */
27667 +       struct super_block      *si_sb;
27668 +};
27669 +
27670 +/* sbinfo status flags */
27671 +/*
27672 + * set true when refresh_dirs() failed at remount time.
27673 + * then try refreshing dirs at access time again.
27674 + * if it is false, refreshing dirs at access time is unnecesary
27675 + */
27676 +#define AuSi_FAILED_REFRESH_DIR        1
27677 +
27678 +#define AuSi_FHSM              (1 << 1)        /* fhsm is active now */
27679 +
27680 +#ifndef CONFIG_AUFS_FHSM
27681 +#undef AuSi_FHSM
27682 +#define AuSi_FHSM              0
27683 +#endif
27684 +
27685 +static inline unsigned char au_do_ftest_si(struct au_sbinfo *sbi,
27686 +                                          unsigned int flag)
27687 +{
27688 +       AuRwMustAnyLock(&sbi->si_rwsem);
27689 +       return sbi->au_si_status & flag;
27690 +}
27691 +#define au_ftest_si(sbinfo, name)      au_do_ftest_si(sbinfo, AuSi_##name)
27692 +#define au_fset_si(sbinfo, name) do { \
27693 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
27694 +       (sbinfo)->au_si_status |= AuSi_##name; \
27695 +} while (0)
27696 +#define au_fclr_si(sbinfo, name) do { \
27697 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
27698 +       (sbinfo)->au_si_status &= ~AuSi_##name; \
27699 +} while (0)
27700 +
27701 +/* ---------------------------------------------------------------------- */
27702 +
27703 +/* policy to select one among writable branches */
27704 +#define AuWbrCopyup(sbinfo, ...) \
27705 +       ((sbinfo)->si_wbr_copyup_ops->copyup(__VA_ARGS__))
27706 +#define AuWbrCreate(sbinfo, ...) \
27707 +       ((sbinfo)->si_wbr_create_ops->create(__VA_ARGS__))
27708 +
27709 +/* flags for si_read_lock()/aufs_read_lock()/di_read_lock() */
27710 +#define AuLock_DW              1               /* write-lock dentry */
27711 +#define AuLock_IR              (1 << 1)        /* read-lock inode */
27712 +#define AuLock_IW              (1 << 2)        /* write-lock inode */
27713 +#define AuLock_FLUSH           (1 << 3)        /* wait for 'nowait' tasks */
27714 +#define AuLock_DIR             (1 << 4)        /* target is a dir */
27715 +#define AuLock_NOPLM           (1 << 5)        /* return err in plm mode */
27716 +#define AuLock_NOPLMW          (1 << 6)        /* wait for plm mode ends */
27717 +#define AuLock_GEN             (1 << 7)        /* test digen/iigen */
27718 +#define au_ftest_lock(flags, name)     ((flags) & AuLock_##name)
27719 +#define au_fset_lock(flags, name) \
27720 +       do { (flags) |= AuLock_##name; } while (0)
27721 +#define au_fclr_lock(flags, name) \
27722 +       do { (flags) &= ~AuLock_##name; } while (0)
27723 +
27724 +/* ---------------------------------------------------------------------- */
27725 +
27726 +/* super.c */
27727 +extern struct file_system_type aufs_fs_type;
27728 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino);
27729 +typedef unsigned long long (*au_arraycb_t)(void *array, unsigned long long max,
27730 +                                          void *arg);
27731 +void au_array_free(void *array);
27732 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb, void *arg);
27733 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max);
27734 +void au_iarray_free(struct inode **a, unsigned long long max);
27735 +
27736 +/* sbinfo.c */
27737 +void au_si_free(struct kobject *kobj);
27738 +int au_si_alloc(struct super_block *sb);
27739 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr);
27740 +
27741 +unsigned int au_sigen_inc(struct super_block *sb);
27742 +aufs_bindex_t au_new_br_id(struct super_block *sb);
27743 +
27744 +int si_read_lock(struct super_block *sb, int flags);
27745 +int si_write_lock(struct super_block *sb, int flags);
27746 +int aufs_read_lock(struct dentry *dentry, int flags);
27747 +void aufs_read_unlock(struct dentry *dentry, int flags);
27748 +void aufs_write_lock(struct dentry *dentry);
27749 +void aufs_write_unlock(struct dentry *dentry);
27750 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags);
27751 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2);
27752 +
27753 +int si_pid_test_slow(struct super_block *sb);
27754 +void si_pid_set_slow(struct super_block *sb);
27755 +void si_pid_clr_slow(struct super_block *sb);
27756 +
27757 +/* wbr_policy.c */
27758 +extern struct au_wbr_copyup_operations au_wbr_copyup_ops[];
27759 +extern struct au_wbr_create_operations au_wbr_create_ops[];
27760 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst);
27761 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex);
27762 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t bstart);
27763 +
27764 +/* mvdown.c */
27765 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *arg);
27766 +
27767 +#ifdef CONFIG_AUFS_FHSM
27768 +/* fhsm.c */
27769 +
27770 +static inline pid_t au_fhsm_pid(struct au_fhsm *fhsm)
27771 +{
27772 +       pid_t pid;
27773 +
27774 +       spin_lock(&fhsm->fhsm_spin);
27775 +       pid = fhsm->fhsm_pid;
27776 +       spin_unlock(&fhsm->fhsm_spin);
27777 +
27778 +       return pid;
27779 +}
27780 +
27781 +void au_fhsm_wrote(struct super_block *sb, aufs_bindex_t bindex, int force);
27782 +void au_fhsm_wrote_all(struct super_block *sb, int force);
27783 +int au_fhsm_fd(struct super_block *sb, int oflags);
27784 +int au_fhsm_br_alloc(struct au_branch *br);
27785 +void au_fhsm_set_bottom(struct super_block *sb, aufs_bindex_t bindex);
27786 +void au_fhsm_fin(struct super_block *sb);
27787 +void au_fhsm_init(struct au_sbinfo *sbinfo);
27788 +void au_fhsm_set(struct au_sbinfo *sbinfo, unsigned int sec);
27789 +void au_fhsm_show(struct seq_file *seq, struct au_sbinfo *sbinfo);
27790 +#else
27791 +AuStubVoid(au_fhsm_wrote, struct super_block *sb, aufs_bindex_t bindex,
27792 +          int force)
27793 +AuStubVoid(au_fhsm_wrote_all, struct super_block *sb, int force)
27794 +AuStub(int, au_fhsm_fd, return -EOPNOTSUPP, struct super_block *sb, int oflags)
27795 +AuStub(pid_t, au_fhsm_pid, return 0, struct au_fhsm *fhsm)
27796 +AuStubInt0(au_fhsm_br_alloc, struct au_branch *br)
27797 +AuStubVoid(au_fhsm_set_bottom, struct super_block *sb, aufs_bindex_t bindex)
27798 +AuStubVoid(au_fhsm_fin, struct super_block *sb)
27799 +AuStubVoid(au_fhsm_init, struct au_sbinfo *sbinfo)
27800 +AuStubVoid(au_fhsm_set, struct au_sbinfo *sbinfo, unsigned int sec)
27801 +AuStubVoid(au_fhsm_show, struct seq_file *seq, struct au_sbinfo *sbinfo)
27802 +#endif
27803 +
27804 +/* ---------------------------------------------------------------------- */
27805 +
27806 +static inline struct au_sbinfo *au_sbi(struct super_block *sb)
27807 +{
27808 +       return sb->s_fs_info;
27809 +}
27810 +
27811 +/* ---------------------------------------------------------------------- */
27812 +
27813 +#ifdef CONFIG_AUFS_EXPORT
27814 +int au_test_nfsd(void);
27815 +void au_export_init(struct super_block *sb);
27816 +void au_xigen_inc(struct inode *inode);
27817 +int au_xigen_new(struct inode *inode);
27818 +int au_xigen_set(struct super_block *sb, struct file *base);
27819 +void au_xigen_clr(struct super_block *sb);
27820 +
27821 +static inline int au_busy_or_stale(void)
27822 +{
27823 +       if (!au_test_nfsd())
27824 +               return -EBUSY;
27825 +       return -ESTALE;
27826 +}
27827 +#else
27828 +AuStubInt0(au_test_nfsd, void)
27829 +AuStubVoid(au_export_init, struct super_block *sb)
27830 +AuStubVoid(au_xigen_inc, struct inode *inode)
27831 +AuStubInt0(au_xigen_new, struct inode *inode)
27832 +AuStubInt0(au_xigen_set, struct super_block *sb, struct file *base)
27833 +AuStubVoid(au_xigen_clr, struct super_block *sb)
27834 +AuStub(int, au_busy_or_stale, return -EBUSY, void)
27835 +#endif /* CONFIG_AUFS_EXPORT */
27836 +
27837 +/* ---------------------------------------------------------------------- */
27838 +
27839 +#ifdef CONFIG_AUFS_SBILIST
27840 +/* module.c */
27841 +extern struct au_splhead au_sbilist;
27842 +
27843 +static inline void au_sbilist_init(void)
27844 +{
27845 +       au_spl_init(&au_sbilist);
27846 +}
27847 +
27848 +static inline void au_sbilist_add(struct super_block *sb)
27849 +{
27850 +       au_spl_add(&au_sbi(sb)->si_list, &au_sbilist);
27851 +}
27852 +
27853 +static inline void au_sbilist_del(struct super_block *sb)
27854 +{
27855 +       au_spl_del(&au_sbi(sb)->si_list, &au_sbilist);
27856 +}
27857 +
27858 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
27859 +static inline void au_sbilist_lock(void)
27860 +{
27861 +       spin_lock(&au_sbilist.spin);
27862 +}
27863 +
27864 +static inline void au_sbilist_unlock(void)
27865 +{
27866 +       spin_unlock(&au_sbilist.spin);
27867 +}
27868 +#define AuGFP_SBILIST  GFP_ATOMIC
27869 +#else
27870 +AuStubVoid(au_sbilist_lock, void)
27871 +AuStubVoid(au_sbilist_unlock, void)
27872 +#define AuGFP_SBILIST  GFP_NOFS
27873 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
27874 +#else
27875 +AuStubVoid(au_sbilist_init, void)
27876 +AuStubVoid(au_sbilist_add, struct super_block *sb)
27877 +AuStubVoid(au_sbilist_del, struct super_block *sb)
27878 +AuStubVoid(au_sbilist_lock, void)
27879 +AuStubVoid(au_sbilist_unlock, void)
27880 +#define AuGFP_SBILIST  GFP_NOFS
27881 +#endif
27882 +
27883 +/* ---------------------------------------------------------------------- */
27884 +
27885 +static inline void dbgaufs_si_null(struct au_sbinfo *sbinfo)
27886 +{
27887 +       /*
27888 +        * This function is a dynamic '__init' function actually,
27889 +        * so the tiny check for si_rwsem is unnecessary.
27890 +        */
27891 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
27892 +#ifdef CONFIG_DEBUG_FS
27893 +       sbinfo->si_dbgaufs = NULL;
27894 +       sbinfo->si_dbgaufs_plink = NULL;
27895 +       sbinfo->si_dbgaufs_xib = NULL;
27896 +#ifdef CONFIG_AUFS_EXPORT
27897 +       sbinfo->si_dbgaufs_xigen = NULL;
27898 +#endif
27899 +#endif
27900 +}
27901 +
27902 +/* ---------------------------------------------------------------------- */
27903 +
27904 +static inline pid_t si_pid_bit(void)
27905 +{
27906 +       /* the origin of pid is 1, but the bitmap's is 0 */
27907 +       return current->pid - 1;
27908 +}
27909 +
27910 +static inline int si_pid_test(struct super_block *sb)
27911 +{
27912 +       pid_t bit;
27913 +
27914 +       bit = si_pid_bit();
27915 +       if (bit < PID_MAX_DEFAULT)
27916 +               return test_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
27917 +       return si_pid_test_slow(sb);
27918 +}
27919 +
27920 +static inline void si_pid_set(struct super_block *sb)
27921 +{
27922 +       pid_t bit;
27923 +
27924 +       bit = si_pid_bit();
27925 +       if (bit < PID_MAX_DEFAULT) {
27926 +               AuDebugOn(test_bit(bit, au_sbi(sb)->au_si_pid.bitmap));
27927 +               set_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
27928 +               /* smp_mb(); */
27929 +       } else
27930 +               si_pid_set_slow(sb);
27931 +}
27932 +
27933 +static inline void si_pid_clr(struct super_block *sb)
27934 +{
27935 +       pid_t bit;
27936 +
27937 +       bit = si_pid_bit();
27938 +       if (bit < PID_MAX_DEFAULT) {
27939 +               AuDebugOn(!test_bit(bit, au_sbi(sb)->au_si_pid.bitmap));
27940 +               clear_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
27941 +               /* smp_mb(); */
27942 +       } else
27943 +               si_pid_clr_slow(sb);
27944 +}
27945 +
27946 +/* ---------------------------------------------------------------------- */
27947 +
27948 +/* lock superblock. mainly for entry point functions */
27949 +/*
27950 + * __si_read_lock, __si_write_lock,
27951 + * __si_read_unlock, __si_write_unlock, __si_downgrade_lock
27952 + */
27953 +AuSimpleRwsemFuncs(__si, struct super_block *sb, &au_sbi(sb)->si_rwsem);
27954 +
27955 +#define SiMustNoWaiters(sb)    AuRwMustNoWaiters(&au_sbi(sb)->si_rwsem)
27956 +#define SiMustAnyLock(sb)      AuRwMustAnyLock(&au_sbi(sb)->si_rwsem)
27957 +#define SiMustWriteLock(sb)    AuRwMustWriteLock(&au_sbi(sb)->si_rwsem)
27958 +
27959 +static inline void si_noflush_read_lock(struct super_block *sb)
27960 +{
27961 +       __si_read_lock(sb);
27962 +       si_pid_set(sb);
27963 +}
27964 +
27965 +static inline int si_noflush_read_trylock(struct super_block *sb)
27966 +{
27967 +       int locked;
27968 +
27969 +       locked = __si_read_trylock(sb);
27970 +       if (locked)
27971 +               si_pid_set(sb);
27972 +       return locked;
27973 +}
27974 +
27975 +static inline void si_noflush_write_lock(struct super_block *sb)
27976 +{
27977 +       __si_write_lock(sb);
27978 +       si_pid_set(sb);
27979 +}
27980 +
27981 +static inline int si_noflush_write_trylock(struct super_block *sb)
27982 +{
27983 +       int locked;
27984 +
27985 +       locked = __si_write_trylock(sb);
27986 +       if (locked)
27987 +               si_pid_set(sb);
27988 +       return locked;
27989 +}
27990 +
27991 +#if 0 /* unused */
27992 +static inline int si_read_trylock(struct super_block *sb, int flags)
27993 +{
27994 +       if (au_ftest_lock(flags, FLUSH))
27995 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
27996 +       return si_noflush_read_trylock(sb);
27997 +}
27998 +#endif
27999 +
28000 +static inline void si_read_unlock(struct super_block *sb)
28001 +{
28002 +       si_pid_clr(sb);
28003 +       __si_read_unlock(sb);
28004 +}
28005 +
28006 +#if 0 /* unused */
28007 +static inline int si_write_trylock(struct super_block *sb, int flags)
28008 +{
28009 +       if (au_ftest_lock(flags, FLUSH))
28010 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
28011 +       return si_noflush_write_trylock(sb);
28012 +}
28013 +#endif
28014 +
28015 +static inline void si_write_unlock(struct super_block *sb)
28016 +{
28017 +       si_pid_clr(sb);
28018 +       __si_write_unlock(sb);
28019 +}
28020 +
28021 +#if 0 /* unused */
28022 +static inline void si_downgrade_lock(struct super_block *sb)
28023 +{
28024 +       __si_downgrade_lock(sb);
28025 +}
28026 +#endif
28027 +
28028 +/* ---------------------------------------------------------------------- */
28029 +
28030 +static inline aufs_bindex_t au_sbend(struct super_block *sb)
28031 +{
28032 +       SiMustAnyLock(sb);
28033 +       return au_sbi(sb)->si_bend;
28034 +}
28035 +
28036 +static inline unsigned int au_mntflags(struct super_block *sb)
28037 +{
28038 +       SiMustAnyLock(sb);
28039 +       return au_sbi(sb)->si_mntflags;
28040 +}
28041 +
28042 +static inline unsigned int au_sigen(struct super_block *sb)
28043 +{
28044 +       SiMustAnyLock(sb);
28045 +       return au_sbi(sb)->si_generation;
28046 +}
28047 +
28048 +static inline void au_ninodes_inc(struct super_block *sb)
28049 +{
28050 +       atomic_long_inc(&au_sbi(sb)->si_ninodes);
28051 +}
28052 +
28053 +static inline void au_ninodes_dec(struct super_block *sb)
28054 +{
28055 +       AuDebugOn(!atomic_long_read(&au_sbi(sb)->si_ninodes));
28056 +       atomic_long_dec(&au_sbi(sb)->si_ninodes);
28057 +}
28058 +
28059 +static inline void au_nfiles_inc(struct super_block *sb)
28060 +{
28061 +       atomic_long_inc(&au_sbi(sb)->si_nfiles);
28062 +}
28063 +
28064 +static inline void au_nfiles_dec(struct super_block *sb)
28065 +{
28066 +       AuDebugOn(!atomic_long_read(&au_sbi(sb)->si_nfiles));
28067 +       atomic_long_dec(&au_sbi(sb)->si_nfiles);
28068 +}
28069 +
28070 +static inline struct au_branch *au_sbr(struct super_block *sb,
28071 +                                      aufs_bindex_t bindex)
28072 +{
28073 +       SiMustAnyLock(sb);
28074 +       return au_sbi(sb)->si_branch[0 + bindex];
28075 +}
28076 +
28077 +static inline void au_xino_brid_set(struct super_block *sb, aufs_bindex_t brid)
28078 +{
28079 +       SiMustWriteLock(sb);
28080 +       au_sbi(sb)->si_xino_brid = brid;
28081 +}
28082 +
28083 +static inline aufs_bindex_t au_xino_brid(struct super_block *sb)
28084 +{
28085 +       SiMustAnyLock(sb);
28086 +       return au_sbi(sb)->si_xino_brid;
28087 +}
28088 +
28089 +#endif /* __KERNEL__ */
28090 +#endif /* __AUFS_SUPER_H__ */
28091 diff -urN /usr/share/empty/fs/aufs/sysaufs.c linux/fs/aufs/sysaufs.c
28092 --- /usr/share/empty/fs/aufs/sysaufs.c  1970-01-01 01:00:00.000000000 +0100
28093 +++ linux/fs/aufs/sysaufs.c     2015-03-27 21:56:35.466795000 +0100
28094 @@ -0,0 +1,104 @@
28095 +/*
28096 + * Copyright (C) 2005-2015 Junjiro R. Okajima
28097 + *
28098 + * This program, aufs is free software; you can redistribute it and/or modify
28099 + * it under the terms of the GNU General Public License as published by
28100 + * the Free Software Foundation; either version 2 of the License, or
28101 + * (at your option) any later version.
28102 + *
28103 + * This program is distributed in the hope that it will be useful,
28104 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28105 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28106 + * GNU General Public License for more details.
28107 + *
28108 + * You should have received a copy of the GNU General Public License
28109 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28110 + */
28111 +
28112 +/*
28113 + * sysfs interface and lifetime management
28114 + * they are necessary regardless sysfs is disabled.
28115 + */
28116 +
28117 +#include <linux/random.h>
28118 +#include "aufs.h"
28119 +
28120 +unsigned long sysaufs_si_mask;
28121 +struct kset *sysaufs_kset;
28122 +
28123 +#define AuSiAttr(_name) { \
28124 +       .attr   = { .name = __stringify(_name), .mode = 0444 }, \
28125 +       .show   = sysaufs_si_##_name,                           \
28126 +}
28127 +
28128 +static struct sysaufs_si_attr sysaufs_si_attr_xi_path = AuSiAttr(xi_path);
28129 +struct attribute *sysaufs_si_attrs[] = {
28130 +       &sysaufs_si_attr_xi_path.attr,
28131 +       NULL,
28132 +};
28133 +
28134 +static const struct sysfs_ops au_sbi_ops = {
28135 +       .show   = sysaufs_si_show
28136 +};
28137 +
28138 +static struct kobj_type au_sbi_ktype = {
28139 +       .release        = au_si_free,
28140 +       .sysfs_ops      = &au_sbi_ops,
28141 +       .default_attrs  = sysaufs_si_attrs
28142 +};
28143 +
28144 +/* ---------------------------------------------------------------------- */
28145 +
28146 +int sysaufs_si_init(struct au_sbinfo *sbinfo)
28147 +{
28148 +       int err;
28149 +
28150 +       sbinfo->si_kobj.kset = sysaufs_kset;
28151 +       /* cf. sysaufs_name() */
28152 +       err = kobject_init_and_add
28153 +               (&sbinfo->si_kobj, &au_sbi_ktype, /*&sysaufs_kset->kobj*/NULL,
28154 +                SysaufsSiNamePrefix "%lx", sysaufs_si_id(sbinfo));
28155 +
28156 +       dbgaufs_si_null(sbinfo);
28157 +       if (!err) {
28158 +               err = dbgaufs_si_init(sbinfo);
28159 +               if (unlikely(err))
28160 +                       kobject_put(&sbinfo->si_kobj);
28161 +       }
28162 +       return err;
28163 +}
28164 +
28165 +void sysaufs_fin(void)
28166 +{
28167 +       dbgaufs_fin();
28168 +       sysfs_remove_group(&sysaufs_kset->kobj, sysaufs_attr_group);
28169 +       kset_unregister(sysaufs_kset);
28170 +}
28171 +
28172 +int __init sysaufs_init(void)
28173 +{
28174 +       int err;
28175 +
28176 +       do {
28177 +               get_random_bytes(&sysaufs_si_mask, sizeof(sysaufs_si_mask));
28178 +       } while (!sysaufs_si_mask);
28179 +
28180 +       err = -EINVAL;
28181 +       sysaufs_kset = kset_create_and_add(AUFS_NAME, NULL, fs_kobj);
28182 +       if (unlikely(!sysaufs_kset))
28183 +               goto out;
28184 +       err = PTR_ERR(sysaufs_kset);
28185 +       if (IS_ERR(sysaufs_kset))
28186 +               goto out;
28187 +       err = sysfs_create_group(&sysaufs_kset->kobj, sysaufs_attr_group);
28188 +       if (unlikely(err)) {
28189 +               kset_unregister(sysaufs_kset);
28190 +               goto out;
28191 +       }
28192 +
28193 +       err = dbgaufs_init();
28194 +       if (unlikely(err))
28195 +               sysaufs_fin();
28196 +out:
28197 +       return err;
28198 +}
28199 diff -urN /usr/share/empty/fs/aufs/sysaufs.h linux/fs/aufs/sysaufs.h
28200 --- /usr/share/empty/fs/aufs/sysaufs.h  1970-01-01 01:00:00.000000000 +0100
28201 +++ linux/fs/aufs/sysaufs.h     2015-03-27 21:56:35.466795000 +0100
28202 @@ -0,0 +1,101 @@
28203 +/*
28204 + * Copyright (C) 2005-2015 Junjiro R. Okajima
28205 + *
28206 + * This program, aufs is free software; you can redistribute it and/or modify
28207 + * it under the terms of the GNU General Public License as published by
28208 + * the Free Software Foundation; either version 2 of the License, or
28209 + * (at your option) any later version.
28210 + *
28211 + * This program is distributed in the hope that it will be useful,
28212 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28213 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28214 + * GNU General Public License for more details.
28215 + *
28216 + * You should have received a copy of the GNU General Public License
28217 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28218 + */
28219 +
28220 +/*
28221 + * sysfs interface and mount lifetime management
28222 + */
28223 +
28224 +#ifndef __SYSAUFS_H__
28225 +#define __SYSAUFS_H__
28226 +
28227 +#ifdef __KERNEL__
28228 +
28229 +#include <linux/sysfs.h>
28230 +#include "module.h"
28231 +
28232 +struct super_block;
28233 +struct au_sbinfo;
28234 +
28235 +struct sysaufs_si_attr {
28236 +       struct attribute attr;
28237 +       int (*show)(struct seq_file *seq, struct super_block *sb);
28238 +};
28239 +
28240 +/* ---------------------------------------------------------------------- */
28241 +
28242 +/* sysaufs.c */
28243 +extern unsigned long sysaufs_si_mask;
28244 +extern struct kset *sysaufs_kset;
28245 +extern struct attribute *sysaufs_si_attrs[];
28246 +int sysaufs_si_init(struct au_sbinfo *sbinfo);
28247 +int __init sysaufs_init(void);
28248 +void sysaufs_fin(void);
28249 +
28250 +/* ---------------------------------------------------------------------- */
28251 +
28252 +/* some people doesn't like to show a pointer in kernel */
28253 +static inline unsigned long sysaufs_si_id(struct au_sbinfo *sbinfo)
28254 +{
28255 +       return sysaufs_si_mask ^ (unsigned long)sbinfo;
28256 +}
28257 +
28258 +#define SysaufsSiNamePrefix    "si_"
28259 +#define SysaufsSiNameLen       (sizeof(SysaufsSiNamePrefix) + 16)
28260 +static inline void sysaufs_name(struct au_sbinfo *sbinfo, char *name)
28261 +{
28262 +       snprintf(name, SysaufsSiNameLen, SysaufsSiNamePrefix "%lx",
28263 +                sysaufs_si_id(sbinfo));
28264 +}
28265 +
28266 +struct au_branch;
28267 +#ifdef CONFIG_SYSFS
28268 +/* sysfs.c */
28269 +extern struct attribute_group *sysaufs_attr_group;
28270 +
28271 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb);
28272 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
28273 +                        char *buf);
28274 +long au_brinfo_ioctl(struct file *file, unsigned long arg);
28275 +#ifdef CONFIG_COMPAT
28276 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg);
28277 +#endif
28278 +
28279 +void sysaufs_br_init(struct au_branch *br);
28280 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
28281 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
28282 +
28283 +#define sysaufs_brs_init()     do {} while (0)
28284 +
28285 +#else
28286 +#define sysaufs_attr_group     NULL
28287 +
28288 +AuStubInt0(sysaufs_si_xi_path, struct seq_file *seq, struct super_block *sb)
28289 +AuStub(ssize_t, sysaufs_si_show, return 0, struct kobject *kobj,
28290 +       struct attribute *attr, char *buf)
28291 +AuStubVoid(sysaufs_br_init, struct au_branch *br)
28292 +AuStubVoid(sysaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
28293 +AuStubVoid(sysaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
28294 +
28295 +static inline void sysaufs_brs_init(void)
28296 +{
28297 +       sysaufs_brs = 0;
28298 +}
28299 +
28300 +#endif /* CONFIG_SYSFS */
28301 +
28302 +#endif /* __KERNEL__ */
28303 +#endif /* __SYSAUFS_H__ */
28304 diff -urN /usr/share/empty/fs/aufs/sysfs.c linux/fs/aufs/sysfs.c
28305 --- /usr/share/empty/fs/aufs/sysfs.c    1970-01-01 01:00:00.000000000 +0100
28306 +++ linux/fs/aufs/sysfs.c       2015-03-27 21:56:35.466795000 +0100
28307 @@ -0,0 +1,372 @@
28308 +/*
28309 + * Copyright (C) 2005-2015 Junjiro R. Okajima
28310 + *
28311 + * This program, aufs is free software; you can redistribute it and/or modify
28312 + * it under the terms of the GNU General Public License as published by
28313 + * the Free Software Foundation; either version 2 of the License, or
28314 + * (at your option) any later version.
28315 + *
28316 + * This program is distributed in the hope that it will be useful,
28317 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28318 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28319 + * GNU General Public License for more details.
28320 + *
28321 + * You should have received a copy of the GNU General Public License
28322 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28323 + */
28324 +
28325 +/*
28326 + * sysfs interface
28327 + */
28328 +
28329 +#include <linux/compat.h>
28330 +#include <linux/seq_file.h>
28331 +#include "aufs.h"
28332 +
28333 +#ifdef CONFIG_AUFS_FS_MODULE
28334 +/* this entry violates the "one line per file" policy of sysfs */
28335 +static ssize_t config_show(struct kobject *kobj, struct kobj_attribute *attr,
28336 +                          char *buf)
28337 +{
28338 +       ssize_t err;
28339 +       static char *conf =
28340 +/* this file is generated at compiling */
28341 +#include "conf.str"
28342 +               ;
28343 +
28344 +       err = snprintf(buf, PAGE_SIZE, conf);
28345 +       if (unlikely(err >= PAGE_SIZE))
28346 +               err = -EFBIG;
28347 +       return err;
28348 +}
28349 +
28350 +static struct kobj_attribute au_config_attr = __ATTR_RO(config);
28351 +#endif
28352 +
28353 +static struct attribute *au_attr[] = {
28354 +#ifdef CONFIG_AUFS_FS_MODULE
28355 +       &au_config_attr.attr,
28356 +#endif
28357 +       NULL,   /* need to NULL terminate the list of attributes */
28358 +};
28359 +
28360 +static struct attribute_group sysaufs_attr_group_body = {
28361 +       .attrs = au_attr
28362 +};
28363 +
28364 +struct attribute_group *sysaufs_attr_group = &sysaufs_attr_group_body;
28365 +
28366 +/* ---------------------------------------------------------------------- */
28367 +
28368 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb)
28369 +{
28370 +       int err;
28371 +
28372 +       SiMustAnyLock(sb);
28373 +
28374 +       err = 0;
28375 +       if (au_opt_test(au_mntflags(sb), XINO)) {
28376 +               err = au_xino_path(seq, au_sbi(sb)->si_xib);
28377 +               seq_putc(seq, '\n');
28378 +       }
28379 +       return err;
28380 +}
28381 +
28382 +/*
28383 + * the lifetime of branch is independent from the entry under sysfs.
28384 + * sysfs handles the lifetime of the entry, and never call ->show() after it is
28385 + * unlinked.
28386 + */
28387 +static int sysaufs_si_br(struct seq_file *seq, struct super_block *sb,
28388 +                        aufs_bindex_t bindex, int idx)
28389 +{
28390 +       int err;
28391 +       struct path path;
28392 +       struct dentry *root;
28393 +       struct au_branch *br;
28394 +       au_br_perm_str_t perm;
28395 +
28396 +       AuDbg("b%d\n", bindex);
28397 +
28398 +       err = 0;
28399 +       root = sb->s_root;
28400 +       di_read_lock_parent(root, !AuLock_IR);
28401 +       br = au_sbr(sb, bindex);
28402 +
28403 +       switch (idx) {
28404 +       case AuBrSysfs_BR:
28405 +               path.mnt = au_br_mnt(br);
28406 +               path.dentry = au_h_dptr(root, bindex);
28407 +               au_seq_path(seq, &path);
28408 +               au_optstr_br_perm(&perm, br->br_perm);
28409 +               err = seq_printf(seq, "=%s\n", perm.a);
28410 +               break;
28411 +       case AuBrSysfs_BRID:
28412 +               err = seq_printf(seq, "%d\n", br->br_id);
28413 +               break;
28414 +       }
28415 +       di_read_unlock(root, !AuLock_IR);
28416 +       if (err == -1)
28417 +               err = -E2BIG;
28418 +
28419 +       return err;
28420 +}
28421 +
28422 +/* ---------------------------------------------------------------------- */
28423 +
28424 +static struct seq_file *au_seq(char *p, ssize_t len)
28425 +{
28426 +       struct seq_file *seq;
28427 +
28428 +       seq = kzalloc(sizeof(*seq), GFP_NOFS);
28429 +       if (seq) {
28430 +               /* mutex_init(&seq.lock); */
28431 +               seq->buf = p;
28432 +               seq->size = len;
28433 +               return seq; /* success */
28434 +       }
28435 +
28436 +       seq = ERR_PTR(-ENOMEM);
28437 +       return seq;
28438 +}
28439 +
28440 +#define SysaufsBr_PREFIX       "br"
28441 +#define SysaufsBrid_PREFIX     "brid"
28442 +
28443 +/* todo: file size may exceed PAGE_SIZE */
28444 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
28445 +                       char *buf)
28446 +{
28447 +       ssize_t err;
28448 +       int idx;
28449 +       long l;
28450 +       aufs_bindex_t bend;
28451 +       struct au_sbinfo *sbinfo;
28452 +       struct super_block *sb;
28453 +       struct seq_file *seq;
28454 +       char *name;
28455 +       struct attribute **cattr;
28456 +
28457 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
28458 +       sb = sbinfo->si_sb;
28459 +
28460 +       /*
28461 +        * prevent a race condition between sysfs and aufs.
28462 +        * for instance, sysfs_file_read() calls sysfs_get_active_two() which
28463 +        * prohibits maintaining the sysfs entries.
28464 +        * hew we acquire read lock after sysfs_get_active_two().
28465 +        * on the other hand, the remount process may maintain the sysfs/aufs
28466 +        * entries after acquiring write lock.
28467 +        * it can cause a deadlock.
28468 +        * simply we gave up processing read here.
28469 +        */
28470 +       err = -EBUSY;
28471 +       if (unlikely(!si_noflush_read_trylock(sb)))
28472 +               goto out;
28473 +
28474 +       seq = au_seq(buf, PAGE_SIZE);
28475 +       err = PTR_ERR(seq);
28476 +       if (IS_ERR(seq))
28477 +               goto out_unlock;
28478 +
28479 +       name = (void *)attr->name;
28480 +       cattr = sysaufs_si_attrs;
28481 +       while (*cattr) {
28482 +               if (!strcmp(name, (*cattr)->name)) {
28483 +                       err = container_of(*cattr, struct sysaufs_si_attr, attr)
28484 +                               ->show(seq, sb);
28485 +                       goto out_seq;
28486 +               }
28487 +               cattr++;
28488 +       }
28489 +
28490 +       if (!strncmp(name, SysaufsBrid_PREFIX,
28491 +                    sizeof(SysaufsBrid_PREFIX) - 1)) {
28492 +               idx = AuBrSysfs_BRID;
28493 +               name += sizeof(SysaufsBrid_PREFIX) - 1;
28494 +       } else if (!strncmp(name, SysaufsBr_PREFIX,
28495 +                           sizeof(SysaufsBr_PREFIX) - 1)) {
28496 +               idx = AuBrSysfs_BR;
28497 +               name += sizeof(SysaufsBr_PREFIX) - 1;
28498 +       } else
28499 +                 BUG();
28500 +
28501 +       err = kstrtol(name, 10, &l);
28502 +       if (!err) {
28503 +               bend = au_sbend(sb);
28504 +               if (l <= bend)
28505 +                       err = sysaufs_si_br(seq, sb, (aufs_bindex_t)l, idx);
28506 +               else
28507 +                       err = -ENOENT;
28508 +       }
28509 +
28510 +out_seq:
28511 +       if (!err) {
28512 +               err = seq->count;
28513 +               /* sysfs limit */
28514 +               if (unlikely(err == PAGE_SIZE))
28515 +                       err = -EFBIG;
28516 +       }
28517 +       kfree(seq);
28518 +out_unlock:
28519 +       si_read_unlock(sb);
28520 +out:
28521 +       return err;
28522 +}
28523 +
28524 +/* ---------------------------------------------------------------------- */
28525 +
28526 +static int au_brinfo(struct super_block *sb, union aufs_brinfo __user *arg)
28527 +{
28528 +       int err;
28529 +       int16_t brid;
28530 +       aufs_bindex_t bindex, bend;
28531 +       size_t sz;
28532 +       char *buf;
28533 +       struct seq_file *seq;
28534 +       struct au_branch *br;
28535 +
28536 +       si_read_lock(sb, AuLock_FLUSH);
28537 +       bend = au_sbend(sb);
28538 +       err = bend + 1;
28539 +       if (!arg)
28540 +               goto out;
28541 +
28542 +       err = -ENOMEM;
28543 +       buf = (void *)__get_free_page(GFP_NOFS);
28544 +       if (unlikely(!buf))
28545 +               goto out;
28546 +
28547 +       seq = au_seq(buf, PAGE_SIZE);
28548 +       err = PTR_ERR(seq);
28549 +       if (IS_ERR(seq))
28550 +               goto out_buf;
28551 +
28552 +       sz = sizeof(*arg) - offsetof(union aufs_brinfo, path);
28553 +       for (bindex = 0; bindex <= bend; bindex++, arg++) {
28554 +               err = !access_ok(VERIFY_WRITE, arg, sizeof(*arg));
28555 +               if (unlikely(err))
28556 +                       break;
28557 +
28558 +               br = au_sbr(sb, bindex);
28559 +               brid = br->br_id;
28560 +               BUILD_BUG_ON(sizeof(brid) != sizeof(arg->id));
28561 +               err = __put_user(brid, &arg->id);
28562 +               if (unlikely(err))
28563 +                       break;
28564 +
28565 +               BUILD_BUG_ON(sizeof(br->br_perm) != sizeof(arg->perm));
28566 +               err = __put_user(br->br_perm, &arg->perm);
28567 +               if (unlikely(err))
28568 +                       break;
28569 +
28570 +               au_seq_path(seq, &br->br_path);
28571 +               err = seq_putc(seq, '\0');
28572 +               if (!err && seq->count <= sz) {
28573 +                       err = copy_to_user(arg->path, seq->buf, seq->count);
28574 +                       seq->count = 0;
28575 +                       if (unlikely(err))
28576 +                               break;
28577 +               } else {
28578 +                       err = -E2BIG;
28579 +                       goto out_seq;
28580 +               }
28581 +       }
28582 +       if (unlikely(err))
28583 +               err = -EFAULT;
28584 +
28585 +out_seq:
28586 +       kfree(seq);
28587 +out_buf:
28588 +       free_page((unsigned long)buf);
28589 +out:
28590 +       si_read_unlock(sb);
28591 +       return err;
28592 +}
28593 +
28594 +long au_brinfo_ioctl(struct file *file, unsigned long arg)
28595 +{
28596 +       return au_brinfo(file->f_path.dentry->d_sb, (void __user *)arg);
28597 +}
28598 +
28599 +#ifdef CONFIG_COMPAT
28600 +long au_brinfo_compat_ioctl(struct file *file, unsigned long arg)
28601 +{
28602 +       return au_brinfo(file->f_path.dentry->d_sb, compat_ptr(arg));
28603 +}
28604 +#endif
28605 +
28606 +/* ---------------------------------------------------------------------- */
28607 +
28608 +void sysaufs_br_init(struct au_branch *br)
28609 +{
28610 +       int i;
28611 +       struct au_brsysfs *br_sysfs;
28612 +       struct attribute *attr;
28613 +
28614 +       br_sysfs = br->br_sysfs;
28615 +       for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
28616 +               attr = &br_sysfs->attr;
28617 +               sysfs_attr_init(attr);
28618 +               attr->name = br_sysfs->name;
28619 +               attr->mode = S_IRUGO;
28620 +               br_sysfs++;
28621 +       }
28622 +}
28623 +
28624 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
28625 +{
28626 +       struct au_branch *br;
28627 +       struct kobject *kobj;
28628 +       struct au_brsysfs *br_sysfs;
28629 +       int i;
28630 +       aufs_bindex_t bend;
28631 +
28632 +       dbgaufs_brs_del(sb, bindex);
28633 +
28634 +       if (!sysaufs_brs)
28635 +               return;
28636 +
28637 +       kobj = &au_sbi(sb)->si_kobj;
28638 +       bend = au_sbend(sb);
28639 +       for (; bindex <= bend; bindex++) {
28640 +               br = au_sbr(sb, bindex);
28641 +               br_sysfs = br->br_sysfs;
28642 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
28643 +                       sysfs_remove_file(kobj, &br_sysfs->attr);
28644 +                       br_sysfs++;
28645 +               }
28646 +       }
28647 +}
28648 +
28649 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
28650 +{
28651 +       int err, i;
28652 +       aufs_bindex_t bend;
28653 +       struct kobject *kobj;
28654 +       struct au_branch *br;
28655 +       struct au_brsysfs *br_sysfs;
28656 +
28657 +       dbgaufs_brs_add(sb, bindex);
28658 +
28659 +       if (!sysaufs_brs)
28660 +               return;
28661 +
28662 +       kobj = &au_sbi(sb)->si_kobj;
28663 +       bend = au_sbend(sb);
28664 +       for (; bindex <= bend; bindex++) {
28665 +               br = au_sbr(sb, bindex);
28666 +               br_sysfs = br->br_sysfs;
28667 +               snprintf(br_sysfs[AuBrSysfs_BR].name, sizeof(br_sysfs->name),
28668 +                        SysaufsBr_PREFIX "%d", bindex);
28669 +               snprintf(br_sysfs[AuBrSysfs_BRID].name, sizeof(br_sysfs->name),
28670 +                        SysaufsBrid_PREFIX "%d", bindex);
28671 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
28672 +                       err = sysfs_create_file(kobj, &br_sysfs->attr);
28673 +                       if (unlikely(err))
28674 +                               pr_warn("failed %s under sysfs(%d)\n",
28675 +                                       br_sysfs->name, err);
28676 +                       br_sysfs++;
28677 +               }
28678 +       }
28679 +}
28680 diff -urN /usr/share/empty/fs/aufs/sysrq.c linux/fs/aufs/sysrq.c
28681 --- /usr/share/empty/fs/aufs/sysrq.c    1970-01-01 01:00:00.000000000 +0100
28682 +++ linux/fs/aufs/sysrq.c       2015-03-27 21:56:35.466795000 +0100
28683 @@ -0,0 +1,157 @@
28684 +/*
28685 + * Copyright (C) 2005-2015 Junjiro R. Okajima
28686 + *
28687 + * This program, aufs is free software; you can redistribute it and/or modify
28688 + * it under the terms of the GNU General Public License as published by
28689 + * the Free Software Foundation; either version 2 of the License, or
28690 + * (at your option) any later version.
28691 + *
28692 + * This program is distributed in the hope that it will be useful,
28693 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28694 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28695 + * GNU General Public License for more details.
28696 + *
28697 + * You should have received a copy of the GNU General Public License
28698 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28699 + */
28700 +
28701 +/*
28702 + * magic sysrq hanlder
28703 + */
28704 +
28705 +/* #include <linux/sysrq.h> */
28706 +#include <linux/writeback.h>
28707 +#include "aufs.h"
28708 +
28709 +/* ---------------------------------------------------------------------- */
28710 +
28711 +static void sysrq_sb(struct super_block *sb)
28712 +{
28713 +       char *plevel;
28714 +       struct au_sbinfo *sbinfo;
28715 +       struct file *file;
28716 +       struct au_sphlhead *files;
28717 +       struct au_finfo *finfo;
28718 +
28719 +       plevel = au_plevel;
28720 +       au_plevel = KERN_WARNING;
28721 +
28722 +       /* since we define pr_fmt, call printk directly */
28723 +#define pr(str) printk(KERN_WARNING AUFS_NAME ": " str)
28724 +
28725 +       sbinfo = au_sbi(sb);
28726 +       printk(KERN_WARNING "si=%lx\n", sysaufs_si_id(sbinfo));
28727 +       pr("superblock\n");
28728 +       au_dpri_sb(sb);
28729 +
28730 +#if 0
28731 +       pr("root dentry\n");
28732 +       au_dpri_dentry(sb->s_root);
28733 +       pr("root inode\n");
28734 +       au_dpri_inode(sb->s_root->d_inode);
28735 +#endif
28736 +
28737 +#if 0
28738 +       do {
28739 +               int err, i, j, ndentry;
28740 +               struct au_dcsub_pages dpages;
28741 +               struct au_dpage *dpage;
28742 +
28743 +               err = au_dpages_init(&dpages, GFP_ATOMIC);
28744 +               if (unlikely(err))
28745 +                       break;
28746 +               err = au_dcsub_pages(&dpages, sb->s_root, NULL, NULL);
28747 +               if (!err)
28748 +                       for (i = 0; i < dpages.ndpage; i++) {
28749 +                               dpage = dpages.dpages + i;
28750 +                               ndentry = dpage->ndentry;
28751 +                               for (j = 0; j < ndentry; j++)
28752 +                                       au_dpri_dentry(dpage->dentries[j]);
28753 +                       }
28754 +               au_dpages_free(&dpages);
28755 +       } while (0);
28756 +#endif
28757 +
28758 +#if 1
28759 +       {
28760 +               struct inode *i;
28761 +
28762 +               pr("isolated inode\n");
28763 +               spin_lock(&inode_sb_list_lock);
28764 +               list_for_each_entry(i, &sb->s_inodes, i_sb_list) {
28765 +                       spin_lock(&i->i_lock);
28766 +                       if (1 || hlist_empty(&i->i_dentry))
28767 +                               au_dpri_inode(i);
28768 +                       spin_unlock(&i->i_lock);
28769 +               }
28770 +               spin_unlock(&inode_sb_list_lock);
28771 +       }
28772 +#endif
28773 +       pr("files\n");
28774 +       files = &au_sbi(sb)->si_files;
28775 +       spin_lock(&files->spin);
28776 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
28777 +               umode_t mode;
28778 +
28779 +               file = finfo->fi_file;
28780 +               mode = file_inode(file)->i_mode;
28781 +               if (!special_file(mode))
28782 +                       au_dpri_file(file);
28783 +       }
28784 +       spin_unlock(&files->spin);
28785 +       pr("done\n");
28786 +
28787 +#undef pr
28788 +       au_plevel = plevel;
28789 +}
28790 +
28791 +/* ---------------------------------------------------------------------- */
28792 +
28793 +/* module parameter */
28794 +static char *aufs_sysrq_key = "a";
28795 +module_param_named(sysrq, aufs_sysrq_key, charp, S_IRUGO);
28796 +MODULE_PARM_DESC(sysrq, "MagicSysRq key for " AUFS_NAME);
28797 +
28798 +static void au_sysrq(int key __maybe_unused)
28799 +{
28800 +       struct au_sbinfo *sbinfo;
28801 +
28802 +       lockdep_off();
28803 +       au_sbilist_lock();
28804 +       list_for_each_entry(sbinfo, &au_sbilist.head, si_list)
28805 +               sysrq_sb(sbinfo->si_sb);
28806 +       au_sbilist_unlock();
28807 +       lockdep_on();
28808 +}
28809 +
28810 +static struct sysrq_key_op au_sysrq_op = {
28811 +       .handler        = au_sysrq,
28812 +       .help_msg       = "Aufs",
28813 +       .action_msg     = "Aufs",
28814 +       .enable_mask    = SYSRQ_ENABLE_DUMP
28815 +};
28816 +
28817 +/* ---------------------------------------------------------------------- */
28818 +
28819 +int __init au_sysrq_init(void)
28820 +{
28821 +       int err;
28822 +       char key;
28823 +
28824 +       err = -1;
28825 +       key = *aufs_sysrq_key;
28826 +       if ('a' <= key && key <= 'z')
28827 +               err = register_sysrq_key(key, &au_sysrq_op);
28828 +       if (unlikely(err))
28829 +               pr_err("err %d, sysrq=%c\n", err, key);
28830 +       return err;
28831 +}
28832 +
28833 +void au_sysrq_fin(void)
28834 +{
28835 +       int err;
28836 +
28837 +       err = unregister_sysrq_key(*aufs_sysrq_key, &au_sysrq_op);
28838 +       if (unlikely(err))
28839 +               pr_err("err %d (ignored)\n", err);
28840 +}
28841 diff -urN /usr/share/empty/fs/aufs/vdir.c linux/fs/aufs/vdir.c
28842 --- /usr/share/empty/fs/aufs/vdir.c     1970-01-01 01:00:00.000000000 +0100
28843 +++ linux/fs/aufs/vdir.c        2015-03-27 21:56:35.466795000 +0100
28844 @@ -0,0 +1,889 @@
28845 +/*
28846 + * Copyright (C) 2005-2015 Junjiro R. Okajima
28847 + *
28848 + * This program, aufs is free software; you can redistribute it and/or modify
28849 + * it under the terms of the GNU General Public License as published by
28850 + * the Free Software Foundation; either version 2 of the License, or
28851 + * (at your option) any later version.
28852 + *
28853 + * This program is distributed in the hope that it will be useful,
28854 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28855 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28856 + * GNU General Public License for more details.
28857 + *
28858 + * You should have received a copy of the GNU General Public License
28859 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28860 + */
28861 +
28862 +/*
28863 + * virtual or vertical directory
28864 + */
28865 +
28866 +#include "aufs.h"
28867 +
28868 +static unsigned int calc_size(int nlen)
28869 +{
28870 +       return ALIGN(sizeof(struct au_vdir_de) + nlen, sizeof(ino_t));
28871 +}
28872 +
28873 +static int set_deblk_end(union au_vdir_deblk_p *p,
28874 +                        union au_vdir_deblk_p *deblk_end)
28875 +{
28876 +       if (calc_size(0) <= deblk_end->deblk - p->deblk) {
28877 +               p->de->de_str.len = 0;
28878 +               /* smp_mb(); */
28879 +               return 0;
28880 +       }
28881 +       return -1; /* error */
28882 +}
28883 +
28884 +/* returns true or false */
28885 +static int is_deblk_end(union au_vdir_deblk_p *p,
28886 +                       union au_vdir_deblk_p *deblk_end)
28887 +{
28888 +       if (calc_size(0) <= deblk_end->deblk - p->deblk)
28889 +               return !p->de->de_str.len;
28890 +       return 1;
28891 +}
28892 +
28893 +static unsigned char *last_deblk(struct au_vdir *vdir)
28894 +{
28895 +       return vdir->vd_deblk[vdir->vd_nblk - 1];
28896 +}
28897 +
28898 +/* ---------------------------------------------------------------------- */
28899 +
28900 +/* estimate the apropriate size for name hash table */
28901 +unsigned int au_rdhash_est(loff_t sz)
28902 +{
28903 +       unsigned int n;
28904 +
28905 +       n = UINT_MAX;
28906 +       sz >>= 10;
28907 +       if (sz < n)
28908 +               n = sz;
28909 +       if (sz < AUFS_RDHASH_DEF)
28910 +               n = AUFS_RDHASH_DEF;
28911 +       /* pr_info("n %u\n", n); */
28912 +       return n;
28913 +}
28914 +
28915 +/*
28916 + * the allocated memory has to be freed by
28917 + * au_nhash_wh_free() or au_nhash_de_free().
28918 + */
28919 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp)
28920 +{
28921 +       struct hlist_head *head;
28922 +       unsigned int u;
28923 +       size_t sz;
28924 +
28925 +       sz = sizeof(*nhash->nh_head) * num_hash;
28926 +       head = kmalloc(sz, gfp);
28927 +       if (head) {
28928 +               nhash->nh_num = num_hash;
28929 +               nhash->nh_head = head;
28930 +               for (u = 0; u < num_hash; u++)
28931 +                       INIT_HLIST_HEAD(head++);
28932 +               return 0; /* success */
28933 +       }
28934 +
28935 +       return -ENOMEM;
28936 +}
28937 +
28938 +static void nhash_count(struct hlist_head *head)
28939 +{
28940 +#if 0
28941 +       unsigned long n;
28942 +       struct hlist_node *pos;
28943 +
28944 +       n = 0;
28945 +       hlist_for_each(pos, head)
28946 +               n++;
28947 +       pr_info("%lu\n", n);
28948 +#endif
28949 +}
28950 +
28951 +static void au_nhash_wh_do_free(struct hlist_head *head)
28952 +{
28953 +       struct au_vdir_wh *pos;
28954 +       struct hlist_node *node;
28955 +
28956 +       hlist_for_each_entry_safe(pos, node, head, wh_hash)
28957 +               kfree(pos);
28958 +}
28959 +
28960 +static void au_nhash_de_do_free(struct hlist_head *head)
28961 +{
28962 +       struct au_vdir_dehstr *pos;
28963 +       struct hlist_node *node;
28964 +
28965 +       hlist_for_each_entry_safe(pos, node, head, hash)
28966 +               au_cache_free_vdir_dehstr(pos);
28967 +}
28968 +
28969 +static void au_nhash_do_free(struct au_nhash *nhash,
28970 +                            void (*free)(struct hlist_head *head))
28971 +{
28972 +       unsigned int n;
28973 +       struct hlist_head *head;
28974 +
28975 +       n = nhash->nh_num;
28976 +       if (!n)
28977 +               return;
28978 +
28979 +       head = nhash->nh_head;
28980 +       while (n-- > 0) {
28981 +               nhash_count(head);
28982 +               free(head++);
28983 +       }
28984 +       kfree(nhash->nh_head);
28985 +}
28986 +
28987 +void au_nhash_wh_free(struct au_nhash *whlist)
28988 +{
28989 +       au_nhash_do_free(whlist, au_nhash_wh_do_free);
28990 +}
28991 +
28992 +static void au_nhash_de_free(struct au_nhash *delist)
28993 +{
28994 +       au_nhash_do_free(delist, au_nhash_de_do_free);
28995 +}
28996 +
28997 +/* ---------------------------------------------------------------------- */
28998 +
28999 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
29000 +                           int limit)
29001 +{
29002 +       int num;
29003 +       unsigned int u, n;
29004 +       struct hlist_head *head;
29005 +       struct au_vdir_wh *pos;
29006 +
29007 +       num = 0;
29008 +       n = whlist->nh_num;
29009 +       head = whlist->nh_head;
29010 +       for (u = 0; u < n; u++, head++)
29011 +               hlist_for_each_entry(pos, head, wh_hash)
29012 +                       if (pos->wh_bindex == btgt && ++num > limit)
29013 +                               return 1;
29014 +       return 0;
29015 +}
29016 +
29017 +static struct hlist_head *au_name_hash(struct au_nhash *nhash,
29018 +                                      unsigned char *name,
29019 +                                      unsigned int len)
29020 +{
29021 +       unsigned int v;
29022 +       /* const unsigned int magic_bit = 12; */
29023 +
29024 +       AuDebugOn(!nhash->nh_num || !nhash->nh_head);
29025 +
29026 +       v = 0;
29027 +       while (len--)
29028 +               v += *name++;
29029 +       /* v = hash_long(v, magic_bit); */
29030 +       v %= nhash->nh_num;
29031 +       return nhash->nh_head + v;
29032 +}
29033 +
29034 +static int au_nhash_test_name(struct au_vdir_destr *str, const char *name,
29035 +                             int nlen)
29036 +{
29037 +       return str->len == nlen && !memcmp(str->name, name, nlen);
29038 +}
29039 +
29040 +/* returns found or not */
29041 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen)
29042 +{
29043 +       struct hlist_head *head;
29044 +       struct au_vdir_wh *pos;
29045 +       struct au_vdir_destr *str;
29046 +
29047 +       head = au_name_hash(whlist, name, nlen);
29048 +       hlist_for_each_entry(pos, head, wh_hash) {
29049 +               str = &pos->wh_str;
29050 +               AuDbg("%.*s\n", str->len, str->name);
29051 +               if (au_nhash_test_name(str, name, nlen))
29052 +                       return 1;
29053 +       }
29054 +       return 0;
29055 +}
29056 +
29057 +/* returns found(true) or not */
29058 +static int test_known(struct au_nhash *delist, char *name, int nlen)
29059 +{
29060 +       struct hlist_head *head;
29061 +       struct au_vdir_dehstr *pos;
29062 +       struct au_vdir_destr *str;
29063 +
29064 +       head = au_name_hash(delist, name, nlen);
29065 +       hlist_for_each_entry(pos, head, hash) {
29066 +               str = pos->str;
29067 +               AuDbg("%.*s\n", str->len, str->name);
29068 +               if (au_nhash_test_name(str, name, nlen))
29069 +                       return 1;
29070 +       }
29071 +       return 0;
29072 +}
29073 +
29074 +static void au_shwh_init_wh(struct au_vdir_wh *wh, ino_t ino,
29075 +                           unsigned char d_type)
29076 +{
29077 +#ifdef CONFIG_AUFS_SHWH
29078 +       wh->wh_ino = ino;
29079 +       wh->wh_type = d_type;
29080 +#endif
29081 +}
29082 +
29083 +/* ---------------------------------------------------------------------- */
29084 +
29085 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
29086 +                      unsigned int d_type, aufs_bindex_t bindex,
29087 +                      unsigned char shwh)
29088 +{
29089 +       int err;
29090 +       struct au_vdir_destr *str;
29091 +       struct au_vdir_wh *wh;
29092 +
29093 +       AuDbg("%.*s\n", nlen, name);
29094 +       AuDebugOn(!whlist->nh_num || !whlist->nh_head);
29095 +
29096 +       err = -ENOMEM;
29097 +       wh = kmalloc(sizeof(*wh) + nlen, GFP_NOFS);
29098 +       if (unlikely(!wh))
29099 +               goto out;
29100 +
29101 +       err = 0;
29102 +       wh->wh_bindex = bindex;
29103 +       if (shwh)
29104 +               au_shwh_init_wh(wh, ino, d_type);
29105 +       str = &wh->wh_str;
29106 +       str->len = nlen;
29107 +       memcpy(str->name, name, nlen);
29108 +       hlist_add_head(&wh->wh_hash, au_name_hash(whlist, name, nlen));
29109 +       /* smp_mb(); */
29110 +
29111 +out:
29112 +       return err;
29113 +}
29114 +
29115 +static int append_deblk(struct au_vdir *vdir)
29116 +{
29117 +       int err;
29118 +       unsigned long ul;
29119 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
29120 +       union au_vdir_deblk_p p, deblk_end;
29121 +       unsigned char **o;
29122 +
29123 +       err = -ENOMEM;
29124 +       o = krealloc(vdir->vd_deblk, sizeof(*o) * (vdir->vd_nblk + 1),
29125 +                    GFP_NOFS);
29126 +       if (unlikely(!o))
29127 +               goto out;
29128 +
29129 +       vdir->vd_deblk = o;
29130 +       p.deblk = kmalloc(deblk_sz, GFP_NOFS);
29131 +       if (p.deblk) {
29132 +               ul = vdir->vd_nblk++;
29133 +               vdir->vd_deblk[ul] = p.deblk;
29134 +               vdir->vd_last.ul = ul;
29135 +               vdir->vd_last.p.deblk = p.deblk;
29136 +               deblk_end.deblk = p.deblk + deblk_sz;
29137 +               err = set_deblk_end(&p, &deblk_end);
29138 +       }
29139 +
29140 +out:
29141 +       return err;
29142 +}
29143 +
29144 +static int append_de(struct au_vdir *vdir, char *name, int nlen, ino_t ino,
29145 +                    unsigned int d_type, struct au_nhash *delist)
29146 +{
29147 +       int err;
29148 +       unsigned int sz;
29149 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
29150 +       union au_vdir_deblk_p p, *room, deblk_end;
29151 +       struct au_vdir_dehstr *dehstr;
29152 +
29153 +       p.deblk = last_deblk(vdir);
29154 +       deblk_end.deblk = p.deblk + deblk_sz;
29155 +       room = &vdir->vd_last.p;
29156 +       AuDebugOn(room->deblk < p.deblk || deblk_end.deblk <= room->deblk
29157 +                 || !is_deblk_end(room, &deblk_end));
29158 +
29159 +       sz = calc_size(nlen);
29160 +       if (unlikely(sz > deblk_end.deblk - room->deblk)) {
29161 +               err = append_deblk(vdir);
29162 +               if (unlikely(err))
29163 +                       goto out;
29164 +
29165 +               p.deblk = last_deblk(vdir);
29166 +               deblk_end.deblk = p.deblk + deblk_sz;
29167 +               /* smp_mb(); */
29168 +               AuDebugOn(room->deblk != p.deblk);
29169 +       }
29170 +
29171 +       err = -ENOMEM;
29172 +       dehstr = au_cache_alloc_vdir_dehstr();
29173 +       if (unlikely(!dehstr))
29174 +               goto out;
29175 +
29176 +       dehstr->str = &room->de->de_str;
29177 +       hlist_add_head(&dehstr->hash, au_name_hash(delist, name, nlen));
29178 +       room->de->de_ino = ino;
29179 +       room->de->de_type = d_type;
29180 +       room->de->de_str.len = nlen;
29181 +       memcpy(room->de->de_str.name, name, nlen);
29182 +
29183 +       err = 0;
29184 +       room->deblk += sz;
29185 +       if (unlikely(set_deblk_end(room, &deblk_end)))
29186 +               err = append_deblk(vdir);
29187 +       /* smp_mb(); */
29188 +
29189 +out:
29190 +       return err;
29191 +}
29192 +
29193 +/* ---------------------------------------------------------------------- */
29194 +
29195 +void au_vdir_free(struct au_vdir *vdir)
29196 +{
29197 +       unsigned char **deblk;
29198 +
29199 +       deblk = vdir->vd_deblk;
29200 +       while (vdir->vd_nblk--)
29201 +               kfree(*deblk++);
29202 +       kfree(vdir->vd_deblk);
29203 +       au_cache_free_vdir(vdir);
29204 +}
29205 +
29206 +static struct au_vdir *alloc_vdir(struct file *file)
29207 +{
29208 +       struct au_vdir *vdir;
29209 +       struct super_block *sb;
29210 +       int err;
29211 +
29212 +       sb = file->f_path.dentry->d_sb;
29213 +       SiMustAnyLock(sb);
29214 +
29215 +       err = -ENOMEM;
29216 +       vdir = au_cache_alloc_vdir();
29217 +       if (unlikely(!vdir))
29218 +               goto out;
29219 +
29220 +       vdir->vd_deblk = kzalloc(sizeof(*vdir->vd_deblk), GFP_NOFS);
29221 +       if (unlikely(!vdir->vd_deblk))
29222 +               goto out_free;
29223 +
29224 +       vdir->vd_deblk_sz = au_sbi(sb)->si_rdblk;
29225 +       if (!vdir->vd_deblk_sz) {
29226 +               /* estimate the apropriate size for deblk */
29227 +               vdir->vd_deblk_sz = au_dir_size(file, /*dentry*/NULL);
29228 +               /* pr_info("vd_deblk_sz %u\n", vdir->vd_deblk_sz); */
29229 +       }
29230 +       vdir->vd_nblk = 0;
29231 +       vdir->vd_version = 0;
29232 +       vdir->vd_jiffy = 0;
29233 +       err = append_deblk(vdir);
29234 +       if (!err)
29235 +               return vdir; /* success */
29236 +
29237 +       kfree(vdir->vd_deblk);
29238 +
29239 +out_free:
29240 +       au_cache_free_vdir(vdir);
29241 +out:
29242 +       vdir = ERR_PTR(err);
29243 +       return vdir;
29244 +}
29245 +
29246 +static int reinit_vdir(struct au_vdir *vdir)
29247 +{
29248 +       int err;
29249 +       union au_vdir_deblk_p p, deblk_end;
29250 +
29251 +       while (vdir->vd_nblk > 1) {
29252 +               kfree(vdir->vd_deblk[vdir->vd_nblk - 1]);
29253 +               /* vdir->vd_deblk[vdir->vd_nblk - 1] = NULL; */
29254 +               vdir->vd_nblk--;
29255 +       }
29256 +       p.deblk = vdir->vd_deblk[0];
29257 +       deblk_end.deblk = p.deblk + vdir->vd_deblk_sz;
29258 +       err = set_deblk_end(&p, &deblk_end);
29259 +       /* keep vd_dblk_sz */
29260 +       vdir->vd_last.ul = 0;
29261 +       vdir->vd_last.p.deblk = vdir->vd_deblk[0];
29262 +       vdir->vd_version = 0;
29263 +       vdir->vd_jiffy = 0;
29264 +       /* smp_mb(); */
29265 +       return err;
29266 +}
29267 +
29268 +/* ---------------------------------------------------------------------- */
29269 +
29270 +#define AuFillVdir_CALLED      1
29271 +#define AuFillVdir_WHABLE      (1 << 1)
29272 +#define AuFillVdir_SHWH                (1 << 2)
29273 +#define au_ftest_fillvdir(flags, name) ((flags) & AuFillVdir_##name)
29274 +#define au_fset_fillvdir(flags, name) \
29275 +       do { (flags) |= AuFillVdir_##name; } while (0)
29276 +#define au_fclr_fillvdir(flags, name) \
29277 +       do { (flags) &= ~AuFillVdir_##name; } while (0)
29278 +
29279 +#ifndef CONFIG_AUFS_SHWH
29280 +#undef AuFillVdir_SHWH
29281 +#define AuFillVdir_SHWH                0
29282 +#endif
29283 +
29284 +struct fillvdir_arg {
29285 +       struct dir_context      ctx;
29286 +       struct file             *file;
29287 +       struct au_vdir          *vdir;
29288 +       struct au_nhash         delist;
29289 +       struct au_nhash         whlist;
29290 +       aufs_bindex_t           bindex;
29291 +       unsigned int            flags;
29292 +       int                     err;
29293 +};
29294 +
29295 +static int fillvdir(struct dir_context *ctx, const char *__name, int nlen,
29296 +                   loff_t offset __maybe_unused, u64 h_ino,
29297 +                   unsigned int d_type)
29298 +{
29299 +       struct fillvdir_arg *arg = container_of(ctx, struct fillvdir_arg, ctx);
29300 +       char *name = (void *)__name;
29301 +       struct super_block *sb;
29302 +       ino_t ino;
29303 +       const unsigned char shwh = !!au_ftest_fillvdir(arg->flags, SHWH);
29304 +
29305 +       arg->err = 0;
29306 +       sb = arg->file->f_path.dentry->d_sb;
29307 +       au_fset_fillvdir(arg->flags, CALLED);
29308 +       /* smp_mb(); */
29309 +       if (nlen <= AUFS_WH_PFX_LEN
29310 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
29311 +               if (test_known(&arg->delist, name, nlen)
29312 +                   || au_nhash_test_known_wh(&arg->whlist, name, nlen))
29313 +                       goto out; /* already exists or whiteouted */
29314 +
29315 +               sb = arg->file->f_path.dentry->d_sb;
29316 +               arg->err = au_ino(sb, arg->bindex, h_ino, d_type, &ino);
29317 +               if (!arg->err) {
29318 +                       if (unlikely(nlen > AUFS_MAX_NAMELEN))
29319 +                               d_type = DT_UNKNOWN;
29320 +                       arg->err = append_de(arg->vdir, name, nlen, ino,
29321 +                                            d_type, &arg->delist);
29322 +               }
29323 +       } else if (au_ftest_fillvdir(arg->flags, WHABLE)) {
29324 +               name += AUFS_WH_PFX_LEN;
29325 +               nlen -= AUFS_WH_PFX_LEN;
29326 +               if (au_nhash_test_known_wh(&arg->whlist, name, nlen))
29327 +                       goto out; /* already whiteouted */
29328 +
29329 +               if (shwh)
29330 +                       arg->err = au_wh_ino(sb, arg->bindex, h_ino, d_type,
29331 +                                            &ino);
29332 +               if (!arg->err) {
29333 +                       if (nlen <= AUFS_MAX_NAMELEN + AUFS_WH_PFX_LEN)
29334 +                               d_type = DT_UNKNOWN;
29335 +                       arg->err = au_nhash_append_wh
29336 +                               (&arg->whlist, name, nlen, ino, d_type,
29337 +                                arg->bindex, shwh);
29338 +               }
29339 +       }
29340 +
29341 +out:
29342 +       if (!arg->err)
29343 +               arg->vdir->vd_jiffy = jiffies;
29344 +       /* smp_mb(); */
29345 +       AuTraceErr(arg->err);
29346 +       return arg->err;
29347 +}
29348 +
29349 +static int au_handle_shwh(struct super_block *sb, struct au_vdir *vdir,
29350 +                         struct au_nhash *whlist, struct au_nhash *delist)
29351 +{
29352 +#ifdef CONFIG_AUFS_SHWH
29353 +       int err;
29354 +       unsigned int nh, u;
29355 +       struct hlist_head *head;
29356 +       struct au_vdir_wh *pos;
29357 +       struct hlist_node *n;
29358 +       char *p, *o;
29359 +       struct au_vdir_destr *destr;
29360 +
29361 +       AuDebugOn(!au_opt_test(au_mntflags(sb), SHWH));
29362 +
29363 +       err = -ENOMEM;
29364 +       o = p = (void *)__get_free_page(GFP_NOFS);
29365 +       if (unlikely(!p))
29366 +               goto out;
29367 +
29368 +       err = 0;
29369 +       nh = whlist->nh_num;
29370 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
29371 +       p += AUFS_WH_PFX_LEN;
29372 +       for (u = 0; u < nh; u++) {
29373 +               head = whlist->nh_head + u;
29374 +               hlist_for_each_entry_safe(pos, n, head, wh_hash) {
29375 +                       destr = &pos->wh_str;
29376 +                       memcpy(p, destr->name, destr->len);
29377 +                       err = append_de(vdir, o, destr->len + AUFS_WH_PFX_LEN,
29378 +                                       pos->wh_ino, pos->wh_type, delist);
29379 +                       if (unlikely(err))
29380 +                               break;
29381 +               }
29382 +       }
29383 +
29384 +       free_page((unsigned long)o);
29385 +
29386 +out:
29387 +       AuTraceErr(err);
29388 +       return err;
29389 +#else
29390 +       return 0;
29391 +#endif
29392 +}
29393 +
29394 +static int au_do_read_vdir(struct fillvdir_arg *arg)
29395 +{
29396 +       int err;
29397 +       unsigned int rdhash;
29398 +       loff_t offset;
29399 +       aufs_bindex_t bend, bindex, bstart;
29400 +       unsigned char shwh;
29401 +       struct file *hf, *file;
29402 +       struct super_block *sb;
29403 +
29404 +       file = arg->file;
29405 +       sb = file->f_path.dentry->d_sb;
29406 +       SiMustAnyLock(sb);
29407 +
29408 +       rdhash = au_sbi(sb)->si_rdhash;
29409 +       if (!rdhash)
29410 +               rdhash = au_rdhash_est(au_dir_size(file, /*dentry*/NULL));
29411 +       err = au_nhash_alloc(&arg->delist, rdhash, GFP_NOFS);
29412 +       if (unlikely(err))
29413 +               goto out;
29414 +       err = au_nhash_alloc(&arg->whlist, rdhash, GFP_NOFS);
29415 +       if (unlikely(err))
29416 +               goto out_delist;
29417 +
29418 +       err = 0;
29419 +       arg->flags = 0;
29420 +       shwh = 0;
29421 +       if (au_opt_test(au_mntflags(sb), SHWH)) {
29422 +               shwh = 1;
29423 +               au_fset_fillvdir(arg->flags, SHWH);
29424 +       }
29425 +       bstart = au_fbstart(file);
29426 +       bend = au_fbend_dir(file);
29427 +       for (bindex = bstart; !err && bindex <= bend; bindex++) {
29428 +               hf = au_hf_dir(file, bindex);
29429 +               if (!hf)
29430 +                       continue;
29431 +
29432 +               offset = vfsub_llseek(hf, 0, SEEK_SET);
29433 +               err = offset;
29434 +               if (unlikely(offset))
29435 +                       break;
29436 +
29437 +               arg->bindex = bindex;
29438 +               au_fclr_fillvdir(arg->flags, WHABLE);
29439 +               if (shwh
29440 +                   || (bindex != bend
29441 +                       && au_br_whable(au_sbr_perm(sb, bindex))))
29442 +                       au_fset_fillvdir(arg->flags, WHABLE);
29443 +               do {
29444 +                       arg->err = 0;
29445 +                       au_fclr_fillvdir(arg->flags, CALLED);
29446 +                       /* smp_mb(); */
29447 +                       err = vfsub_iterate_dir(hf, &arg->ctx);
29448 +                       if (err >= 0)
29449 +                               err = arg->err;
29450 +               } while (!err && au_ftest_fillvdir(arg->flags, CALLED));
29451 +
29452 +               /*
29453 +                * dir_relax() may be good for concurrency, but aufs should not
29454 +                * use it since it will cause a lockdep problem.
29455 +                */
29456 +       }
29457 +
29458 +       if (!err && shwh)
29459 +               err = au_handle_shwh(sb, arg->vdir, &arg->whlist, &arg->delist);
29460 +
29461 +       au_nhash_wh_free(&arg->whlist);
29462 +
29463 +out_delist:
29464 +       au_nhash_de_free(&arg->delist);
29465 +out:
29466 +       return err;
29467 +}
29468 +
29469 +static int read_vdir(struct file *file, int may_read)
29470 +{
29471 +       int err;
29472 +       unsigned long expire;
29473 +       unsigned char do_read;
29474 +       struct fillvdir_arg arg = {
29475 +               .ctx = {
29476 +                       .actor = fillvdir
29477 +               }
29478 +       };
29479 +       struct inode *inode;
29480 +       struct au_vdir *vdir, *allocated;
29481 +
29482 +       err = 0;
29483 +       inode = file_inode(file);
29484 +       IMustLock(inode);
29485 +       SiMustAnyLock(inode->i_sb);
29486 +
29487 +       allocated = NULL;
29488 +       do_read = 0;
29489 +       expire = au_sbi(inode->i_sb)->si_rdcache;
29490 +       vdir = au_ivdir(inode);
29491 +       if (!vdir) {
29492 +               do_read = 1;
29493 +               vdir = alloc_vdir(file);
29494 +               err = PTR_ERR(vdir);
29495 +               if (IS_ERR(vdir))
29496 +                       goto out;
29497 +               err = 0;
29498 +               allocated = vdir;
29499 +       } else if (may_read
29500 +                  && (inode->i_version != vdir->vd_version
29501 +                      || time_after(jiffies, vdir->vd_jiffy + expire))) {
29502 +               do_read = 1;
29503 +               err = reinit_vdir(vdir);
29504 +               if (unlikely(err))
29505 +                       goto out;
29506 +       }
29507 +
29508 +       if (!do_read)
29509 +               return 0; /* success */
29510 +
29511 +       arg.file = file;
29512 +       arg.vdir = vdir;
29513 +       err = au_do_read_vdir(&arg);
29514 +       if (!err) {
29515 +               /* file->f_pos = 0; */ /* todo: ctx->pos? */
29516 +               vdir->vd_version = inode->i_version;
29517 +               vdir->vd_last.ul = 0;
29518 +               vdir->vd_last.p.deblk = vdir->vd_deblk[0];
29519 +               if (allocated)
29520 +                       au_set_ivdir(inode, allocated);
29521 +       } else if (allocated)
29522 +               au_vdir_free(allocated);
29523 +
29524 +out:
29525 +       return err;
29526 +}
29527 +
29528 +static int copy_vdir(struct au_vdir *tgt, struct au_vdir *src)
29529 +{
29530 +       int err, rerr;
29531 +       unsigned long ul, n;
29532 +       const unsigned int deblk_sz = src->vd_deblk_sz;
29533 +
29534 +       AuDebugOn(tgt->vd_nblk != 1);
29535 +
29536 +       err = -ENOMEM;
29537 +       if (tgt->vd_nblk < src->vd_nblk) {
29538 +               unsigned char **p;
29539 +
29540 +               p = krealloc(tgt->vd_deblk, sizeof(*p) * src->vd_nblk,
29541 +                            GFP_NOFS);
29542 +               if (unlikely(!p))
29543 +                       goto out;
29544 +               tgt->vd_deblk = p;
29545 +       }
29546 +
29547 +       if (tgt->vd_deblk_sz != deblk_sz) {
29548 +               unsigned char *p;
29549 +
29550 +               tgt->vd_deblk_sz = deblk_sz;
29551 +               p = krealloc(tgt->vd_deblk[0], deblk_sz, GFP_NOFS);
29552 +               if (unlikely(!p))
29553 +                       goto out;
29554 +               tgt->vd_deblk[0] = p;
29555 +       }
29556 +       memcpy(tgt->vd_deblk[0], src->vd_deblk[0], deblk_sz);
29557 +       tgt->vd_version = src->vd_version;
29558 +       tgt->vd_jiffy = src->vd_jiffy;
29559 +
29560 +       n = src->vd_nblk;
29561 +       for (ul = 1; ul < n; ul++) {
29562 +               tgt->vd_deblk[ul] = kmemdup(src->vd_deblk[ul], deblk_sz,
29563 +                                           GFP_NOFS);
29564 +               if (unlikely(!tgt->vd_deblk[ul]))
29565 +                       goto out;
29566 +               tgt->vd_nblk++;
29567 +       }
29568 +       tgt->vd_nblk = n;
29569 +       tgt->vd_last.ul = tgt->vd_last.ul;
29570 +       tgt->vd_last.p.deblk = tgt->vd_deblk[tgt->vd_last.ul];
29571 +       tgt->vd_last.p.deblk += src->vd_last.p.deblk
29572 +               - src->vd_deblk[src->vd_last.ul];
29573 +       /* smp_mb(); */
29574 +       return 0; /* success */
29575 +
29576 +out:
29577 +       rerr = reinit_vdir(tgt);
29578 +       BUG_ON(rerr);
29579 +       return err;
29580 +}
29581 +
29582 +int au_vdir_init(struct file *file)
29583 +{
29584 +       int err;
29585 +       struct inode *inode;
29586 +       struct au_vdir *vdir_cache, *allocated;
29587 +
29588 +       /* test file->f_pos here instead of ctx->pos */
29589 +       err = read_vdir(file, !file->f_pos);
29590 +       if (unlikely(err))
29591 +               goto out;
29592 +
29593 +       allocated = NULL;
29594 +       vdir_cache = au_fvdir_cache(file);
29595 +       if (!vdir_cache) {
29596 +               vdir_cache = alloc_vdir(file);
29597 +               err = PTR_ERR(vdir_cache);
29598 +               if (IS_ERR(vdir_cache))
29599 +                       goto out;
29600 +               allocated = vdir_cache;
29601 +       } else if (!file->f_pos && vdir_cache->vd_version != file->f_version) {
29602 +               /* test file->f_pos here instead of ctx->pos */
29603 +               err = reinit_vdir(vdir_cache);
29604 +               if (unlikely(err))
29605 +                       goto out;
29606 +       } else
29607 +               return 0; /* success */
29608 +
29609 +       inode = file_inode(file);
29610 +       err = copy_vdir(vdir_cache, au_ivdir(inode));
29611 +       if (!err) {
29612 +               file->f_version = inode->i_version;
29613 +               if (allocated)
29614 +                       au_set_fvdir_cache(file, allocated);
29615 +       } else if (allocated)
29616 +               au_vdir_free(allocated);
29617 +
29618 +out:
29619 +       return err;
29620 +}
29621 +
29622 +static loff_t calc_offset(struct au_vdir *vdir)
29623 +{
29624 +       loff_t offset;
29625 +       union au_vdir_deblk_p p;
29626 +
29627 +       p.deblk = vdir->vd_deblk[vdir->vd_last.ul];
29628 +       offset = vdir->vd_last.p.deblk - p.deblk;
29629 +       offset += vdir->vd_deblk_sz * vdir->vd_last.ul;
29630 +       return offset;
29631 +}
29632 +
29633 +/* returns true or false */
29634 +static int seek_vdir(struct file *file, struct dir_context *ctx)
29635 +{
29636 +       int valid;
29637 +       unsigned int deblk_sz;
29638 +       unsigned long ul, n;
29639 +       loff_t offset;
29640 +       union au_vdir_deblk_p p, deblk_end;
29641 +       struct au_vdir *vdir_cache;
29642 +
29643 +       valid = 1;
29644 +       vdir_cache = au_fvdir_cache(file);
29645 +       offset = calc_offset(vdir_cache);
29646 +       AuDbg("offset %lld\n", offset);
29647 +       if (ctx->pos == offset)
29648 +               goto out;
29649 +
29650 +       vdir_cache->vd_last.ul = 0;
29651 +       vdir_cache->vd_last.p.deblk = vdir_cache->vd_deblk[0];
29652 +       if (!ctx->pos)
29653 +               goto out;
29654 +
29655 +       valid = 0;
29656 +       deblk_sz = vdir_cache->vd_deblk_sz;
29657 +       ul = div64_u64(ctx->pos, deblk_sz);
29658 +       AuDbg("ul %lu\n", ul);
29659 +       if (ul >= vdir_cache->vd_nblk)
29660 +               goto out;
29661 +
29662 +       n = vdir_cache->vd_nblk;
29663 +       for (; ul < n; ul++) {
29664 +               p.deblk = vdir_cache->vd_deblk[ul];
29665 +               deblk_end.deblk = p.deblk + deblk_sz;
29666 +               offset = ul;
29667 +               offset *= deblk_sz;
29668 +               while (!is_deblk_end(&p, &deblk_end) && offset < ctx->pos) {
29669 +                       unsigned int l;
29670 +
29671 +                       l = calc_size(p.de->de_str.len);
29672 +                       offset += l;
29673 +                       p.deblk += l;
29674 +               }
29675 +               if (!is_deblk_end(&p, &deblk_end)) {
29676 +                       valid = 1;
29677 +                       vdir_cache->vd_last.ul = ul;
29678 +                       vdir_cache->vd_last.p = p;
29679 +                       break;
29680 +               }
29681 +       }
29682 +
29683 +out:
29684 +       /* smp_mb(); */
29685 +       AuTraceErr(!valid);
29686 +       return valid;
29687 +}
29688 +
29689 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx)
29690 +{
29691 +       unsigned int l, deblk_sz;
29692 +       union au_vdir_deblk_p deblk_end;
29693 +       struct au_vdir *vdir_cache;
29694 +       struct au_vdir_de *de;
29695 +
29696 +       vdir_cache = au_fvdir_cache(file);
29697 +       if (!seek_vdir(file, ctx))
29698 +               return 0;
29699 +
29700 +       deblk_sz = vdir_cache->vd_deblk_sz;
29701 +       while (1) {
29702 +               deblk_end.deblk = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
29703 +               deblk_end.deblk += deblk_sz;
29704 +               while (!is_deblk_end(&vdir_cache->vd_last.p, &deblk_end)) {
29705 +                       de = vdir_cache->vd_last.p.de;
29706 +                       AuDbg("%.*s, off%lld, i%lu, dt%d\n",
29707 +                             de->de_str.len, de->de_str.name, ctx->pos,
29708 +                             (unsigned long)de->de_ino, de->de_type);
29709 +                       if (unlikely(!dir_emit(ctx, de->de_str.name,
29710 +                                              de->de_str.len, de->de_ino,
29711 +                                              de->de_type))) {
29712 +                               /* todo: ignore the error caused by udba? */
29713 +                               /* return err; */
29714 +                               return 0;
29715 +                       }
29716 +
29717 +                       l = calc_size(de->de_str.len);
29718 +                       vdir_cache->vd_last.p.deblk += l;
29719 +                       ctx->pos += l;
29720 +               }
29721 +               if (vdir_cache->vd_last.ul < vdir_cache->vd_nblk - 1) {
29722 +                       vdir_cache->vd_last.ul++;
29723 +                       vdir_cache->vd_last.p.deblk
29724 +                               = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
29725 +                       ctx->pos = deblk_sz * vdir_cache->vd_last.ul;
29726 +                       continue;
29727 +               }
29728 +               break;
29729 +       }
29730 +
29731 +       /* smp_mb(); */
29732 +       return 0;
29733 +}
29734 diff -urN /usr/share/empty/fs/aufs/vfsub.c linux/fs/aufs/vfsub.c
29735 --- /usr/share/empty/fs/aufs/vfsub.c    1970-01-01 01:00:00.000000000 +0100
29736 +++ linux/fs/aufs/vfsub.c       2015-03-27 21:56:35.470128334 +0100
29737 @@ -0,0 +1,796 @@
29738 +/*
29739 + * Copyright (C) 2005-2015 Junjiro R. Okajima
29740 + *
29741 + * This program, aufs is free software; you can redistribute it and/or modify
29742 + * it under the terms of the GNU General Public License as published by
29743 + * the Free Software Foundation; either version 2 of the License, or
29744 + * (at your option) any later version.
29745 + *
29746 + * This program is distributed in the hope that it will be useful,
29747 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29748 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29749 + * GNU General Public License for more details.
29750 + *
29751 + * You should have received a copy of the GNU General Public License
29752 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29753 + */
29754 +
29755 +/*
29756 + * sub-routines for VFS
29757 + */
29758 +
29759 +#include <linux/ima.h>
29760 +#include <linux/namei.h>
29761 +#include <linux/security.h>
29762 +#include <linux/splice.h>
29763 +#include "aufs.h"
29764 +
29765 +int vfsub_update_h_iattr(struct path *h_path, int *did)
29766 +{
29767 +       int err;
29768 +       struct kstat st;
29769 +       struct super_block *h_sb;
29770 +
29771 +       /* for remote fs, leave work for its getattr or d_revalidate */
29772 +       /* for bad i_attr fs, handle them in aufs_getattr() */
29773 +       /* still some fs may acquire i_mutex. we need to skip them */
29774 +       err = 0;
29775 +       if (!did)
29776 +               did = &err;
29777 +       h_sb = h_path->dentry->d_sb;
29778 +       *did = (!au_test_fs_remote(h_sb) && au_test_fs_refresh_iattr(h_sb));
29779 +       if (*did)
29780 +               err = vfs_getattr(h_path, &st);
29781 +
29782 +       return err;
29783 +}
29784 +
29785 +/* ---------------------------------------------------------------------- */
29786 +
29787 +struct file *vfsub_dentry_open(struct path *path, int flags)
29788 +{
29789 +       struct file *file;
29790 +
29791 +       file = dentry_open(path, flags /* | __FMODE_NONOTIFY */,
29792 +                          current_cred());
29793 +       if (!IS_ERR_OR_NULL(file)
29794 +           && (file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
29795 +               i_readcount_inc(path->dentry->d_inode);
29796 +
29797 +       return file;
29798 +}
29799 +
29800 +struct file *vfsub_filp_open(const char *path, int oflags, int mode)
29801 +{
29802 +       struct file *file;
29803 +
29804 +       lockdep_off();
29805 +       file = filp_open(path,
29806 +                        oflags /* | __FMODE_NONOTIFY */,
29807 +                        mode);
29808 +       lockdep_on();
29809 +       if (IS_ERR(file))
29810 +               goto out;
29811 +       vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
29812 +
29813 +out:
29814 +       return file;
29815 +}
29816 +
29817 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path)
29818 +{
29819 +       int err;
29820 +
29821 +       err = kern_path(name, flags, path);
29822 +       if (!err && path->dentry->d_inode)
29823 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
29824 +       return err;
29825 +}
29826 +
29827 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
29828 +                                   int len)
29829 +{
29830 +       struct path path = {
29831 +               .mnt = NULL
29832 +       };
29833 +
29834 +       /* VFS checks it too, but by WARN_ON_ONCE() */
29835 +       IMustLock(parent->d_inode);
29836 +
29837 +       path.dentry = lookup_one_len(name, parent, len);
29838 +       if (IS_ERR(path.dentry))
29839 +               goto out;
29840 +       if (path.dentry->d_inode)
29841 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
29842 +
29843 +out:
29844 +       AuTraceErrPtr(path.dentry);
29845 +       return path.dentry;
29846 +}
29847 +
29848 +void vfsub_call_lkup_one(void *args)
29849 +{
29850 +       struct vfsub_lkup_one_args *a = args;
29851 +       *a->errp = vfsub_lkup_one(a->name, a->parent);
29852 +}
29853 +
29854 +/* ---------------------------------------------------------------------- */
29855 +
29856 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
29857 +                                struct dentry *d2, struct au_hinode *hdir2)
29858 +{
29859 +       struct dentry *d;
29860 +
29861 +       lockdep_off();
29862 +       d = lock_rename(d1, d2);
29863 +       lockdep_on();
29864 +       au_hn_suspend(hdir1);
29865 +       if (hdir1 != hdir2)
29866 +               au_hn_suspend(hdir2);
29867 +
29868 +       return d;
29869 +}
29870 +
29871 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
29872 +                        struct dentry *d2, struct au_hinode *hdir2)
29873 +{
29874 +       au_hn_resume(hdir1);
29875 +       if (hdir1 != hdir2)
29876 +               au_hn_resume(hdir2);
29877 +       lockdep_off();
29878 +       unlock_rename(d1, d2);
29879 +       lockdep_on();
29880 +}
29881 +
29882 +/* ---------------------------------------------------------------------- */
29883 +
29884 +int vfsub_create(struct inode *dir, struct path *path, int mode, bool want_excl)
29885 +{
29886 +       int err;
29887 +       struct dentry *d;
29888 +
29889 +       IMustLock(dir);
29890 +
29891 +       d = path->dentry;
29892 +       path->dentry = d->d_parent;
29893 +       err = security_path_mknod(path, d, mode, 0);
29894 +       path->dentry = d;
29895 +       if (unlikely(err))
29896 +               goto out;
29897 +
29898 +       lockdep_off();
29899 +       err = vfs_create(dir, path->dentry, mode, want_excl);
29900 +       lockdep_on();
29901 +       if (!err) {
29902 +               struct path tmp = *path;
29903 +               int did;
29904 +
29905 +               vfsub_update_h_iattr(&tmp, &did);
29906 +               if (did) {
29907 +                       tmp.dentry = path->dentry->d_parent;
29908 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
29909 +               }
29910 +               /*ignore*/
29911 +       }
29912 +
29913 +out:
29914 +       return err;
29915 +}
29916 +
29917 +int vfsub_symlink(struct inode *dir, struct path *path, const char *symname)
29918 +{
29919 +       int err;
29920 +       struct dentry *d;
29921 +
29922 +       IMustLock(dir);
29923 +
29924 +       d = path->dentry;
29925 +       path->dentry = d->d_parent;
29926 +       err = security_path_symlink(path, d, symname);
29927 +       path->dentry = d;
29928 +       if (unlikely(err))
29929 +               goto out;
29930 +
29931 +       lockdep_off();
29932 +       err = vfs_symlink(dir, path->dentry, symname);
29933 +       lockdep_on();
29934 +       if (!err) {
29935 +               struct path tmp = *path;
29936 +               int did;
29937 +
29938 +               vfsub_update_h_iattr(&tmp, &did);
29939 +               if (did) {
29940 +                       tmp.dentry = path->dentry->d_parent;
29941 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
29942 +               }
29943 +               /*ignore*/
29944 +       }
29945 +
29946 +out:
29947 +       return err;
29948 +}
29949 +
29950 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev)
29951 +{
29952 +       int err;
29953 +       struct dentry *d;
29954 +
29955 +       IMustLock(dir);
29956 +
29957 +       d = path->dentry;
29958 +       path->dentry = d->d_parent;
29959 +       err = security_path_mknod(path, d, mode, new_encode_dev(dev));
29960 +       path->dentry = d;
29961 +       if (unlikely(err))
29962 +               goto out;
29963 +
29964 +       lockdep_off();
29965 +       err = vfs_mknod(dir, path->dentry, mode, dev);
29966 +       lockdep_on();
29967 +       if (!err) {
29968 +               struct path tmp = *path;
29969 +               int did;
29970 +
29971 +               vfsub_update_h_iattr(&tmp, &did);
29972 +               if (did) {
29973 +                       tmp.dentry = path->dentry->d_parent;
29974 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
29975 +               }
29976 +               /*ignore*/
29977 +       }
29978 +
29979 +out:
29980 +       return err;
29981 +}
29982 +
29983 +static int au_test_nlink(struct inode *inode)
29984 +{
29985 +       const unsigned int link_max = UINT_MAX >> 1; /* rough margin */
29986 +
29987 +       if (!au_test_fs_no_limit_nlink(inode->i_sb)
29988 +           || inode->i_nlink < link_max)
29989 +               return 0;
29990 +       return -EMLINK;
29991 +}
29992 +
29993 +int vfsub_link(struct dentry *src_dentry, struct inode *dir, struct path *path,
29994 +              struct inode **delegated_inode)
29995 +{
29996 +       int err;
29997 +       struct dentry *d;
29998 +
29999 +       IMustLock(dir);
30000 +
30001 +       err = au_test_nlink(src_dentry->d_inode);
30002 +       if (unlikely(err))
30003 +               return err;
30004 +
30005 +       /* we don't call may_linkat() */
30006 +       d = path->dentry;
30007 +       path->dentry = d->d_parent;
30008 +       err = security_path_link(src_dentry, path, d);
30009 +       path->dentry = d;
30010 +       if (unlikely(err))
30011 +               goto out;
30012 +
30013 +       lockdep_off();
30014 +       err = vfs_link(src_dentry, dir, path->dentry, delegated_inode);
30015 +       lockdep_on();
30016 +       if (!err) {
30017 +               struct path tmp = *path;
30018 +               int did;
30019 +
30020 +               /* fuse has different memory inode for the same inumber */
30021 +               vfsub_update_h_iattr(&tmp, &did);
30022 +               if (did) {
30023 +                       tmp.dentry = path->dentry->d_parent;
30024 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30025 +                       tmp.dentry = src_dentry;
30026 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30027 +               }
30028 +               /*ignore*/
30029 +       }
30030 +
30031 +out:
30032 +       return err;
30033 +}
30034 +
30035 +int vfsub_rename(struct inode *src_dir, struct dentry *src_dentry,
30036 +                struct inode *dir, struct path *path,
30037 +                struct inode **delegated_inode)
30038 +{
30039 +       int err;
30040 +       struct path tmp = {
30041 +               .mnt    = path->mnt
30042 +       };
30043 +       struct dentry *d;
30044 +
30045 +       IMustLock(dir);
30046 +       IMustLock(src_dir);
30047 +
30048 +       d = path->dentry;
30049 +       path->dentry = d->d_parent;
30050 +       tmp.dentry = src_dentry->d_parent;
30051 +       err = security_path_rename(&tmp, src_dentry, path, d, /*flags*/0);
30052 +       path->dentry = d;
30053 +       if (unlikely(err))
30054 +               goto out;
30055 +
30056 +       lockdep_off();
30057 +       err = vfs_rename(src_dir, src_dentry, dir, path->dentry,
30058 +                        delegated_inode, /*flags*/0);
30059 +       lockdep_on();
30060 +       if (!err) {
30061 +               int did;
30062 +
30063 +               tmp.dentry = d->d_parent;
30064 +               vfsub_update_h_iattr(&tmp, &did);
30065 +               if (did) {
30066 +                       tmp.dentry = src_dentry;
30067 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30068 +                       tmp.dentry = src_dentry->d_parent;
30069 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30070 +               }
30071 +               /*ignore*/
30072 +       }
30073 +
30074 +out:
30075 +       return err;
30076 +}
30077 +
30078 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode)
30079 +{
30080 +       int err;
30081 +       struct dentry *d;
30082 +
30083 +       IMustLock(dir);
30084 +
30085 +       d = path->dentry;
30086 +       path->dentry = d->d_parent;
30087 +       err = security_path_mkdir(path, d, mode);
30088 +       path->dentry = d;
30089 +       if (unlikely(err))
30090 +               goto out;
30091 +
30092 +       lockdep_off();
30093 +       err = vfs_mkdir(dir, path->dentry, mode);
30094 +       lockdep_on();
30095 +       if (!err) {
30096 +               struct path tmp = *path;
30097 +               int did;
30098 +
30099 +               vfsub_update_h_iattr(&tmp, &did);
30100 +               if (did) {
30101 +                       tmp.dentry = path->dentry->d_parent;
30102 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
30103 +               }
30104 +               /*ignore*/
30105 +       }
30106 +
30107 +out:
30108 +       return err;
30109 +}
30110 +
30111 +int vfsub_rmdir(struct inode *dir, struct path *path)
30112 +{
30113 +       int err;
30114 +       struct dentry *d;
30115 +
30116 +       IMustLock(dir);
30117 +
30118 +       d = path->dentry;
30119 +       path->dentry = d->d_parent;
30120 +       err = security_path_rmdir(path, d);
30121 +       path->dentry = d;
30122 +       if (unlikely(err))
30123 +               goto out;
30124 +
30125 +       lockdep_off();
30126 +       err = vfs_rmdir(dir, path->dentry);
30127 +       lockdep_on();
30128 +       if (!err) {
30129 +               struct path tmp = {
30130 +                       .dentry = path->dentry->d_parent,
30131 +                       .mnt    = path->mnt
30132 +               };
30133 +
30134 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
30135 +       }
30136 +
30137 +out:
30138 +       return err;
30139 +}
30140 +
30141 +/* ---------------------------------------------------------------------- */
30142 +
30143 +/* todo: support mmap_sem? */
30144 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
30145 +                    loff_t *ppos)
30146 +{
30147 +       ssize_t err;
30148 +
30149 +       lockdep_off();
30150 +       err = vfs_read(file, ubuf, count, ppos);
30151 +       lockdep_on();
30152 +       if (err >= 0)
30153 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30154 +       return err;
30155 +}
30156 +
30157 +/* todo: kernel_read()? */
30158 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
30159 +                    loff_t *ppos)
30160 +{
30161 +       ssize_t err;
30162 +       mm_segment_t oldfs;
30163 +       union {
30164 +               void *k;
30165 +               char __user *u;
30166 +       } buf;
30167 +
30168 +       buf.k = kbuf;
30169 +       oldfs = get_fs();
30170 +       set_fs(KERNEL_DS);
30171 +       err = vfsub_read_u(file, buf.u, count, ppos);
30172 +       set_fs(oldfs);
30173 +       return err;
30174 +}
30175 +
30176 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
30177 +                     loff_t *ppos)
30178 +{
30179 +       ssize_t err;
30180 +
30181 +       lockdep_off();
30182 +       err = vfs_write(file, ubuf, count, ppos);
30183 +       lockdep_on();
30184 +       if (err >= 0)
30185 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30186 +       return err;
30187 +}
30188 +
30189 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count, loff_t *ppos)
30190 +{
30191 +       ssize_t err;
30192 +       mm_segment_t oldfs;
30193 +       union {
30194 +               void *k;
30195 +               const char __user *u;
30196 +       } buf;
30197 +
30198 +       buf.k = kbuf;
30199 +       oldfs = get_fs();
30200 +       set_fs(KERNEL_DS);
30201 +       err = vfsub_write_u(file, buf.u, count, ppos);
30202 +       set_fs(oldfs);
30203 +       return err;
30204 +}
30205 +
30206 +int vfsub_flush(struct file *file, fl_owner_t id)
30207 +{
30208 +       int err;
30209 +
30210 +       err = 0;
30211 +       if (file->f_op->flush) {
30212 +               if (!au_test_nfs(file->f_path.dentry->d_sb))
30213 +                       err = file->f_op->flush(file, id);
30214 +               else {
30215 +                       lockdep_off();
30216 +                       err = file->f_op->flush(file, id);
30217 +                       lockdep_on();
30218 +               }
30219 +               if (!err)
30220 +                       vfsub_update_h_iattr(&file->f_path, /*did*/NULL);
30221 +               /*ignore*/
30222 +       }
30223 +       return err;
30224 +}
30225 +
30226 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx)
30227 +{
30228 +       int err;
30229 +
30230 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
30231 +
30232 +       lockdep_off();
30233 +       err = iterate_dir(file, ctx);
30234 +       lockdep_on();
30235 +       if (err >= 0)
30236 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
30237 +       return err;
30238 +}
30239 +
30240 +long vfsub_splice_to(struct file *in, loff_t *ppos,
30241 +                    struct pipe_inode_info *pipe, size_t len,
30242 +                    unsigned int flags)
30243 +{
30244 +       long err;
30245 +
30246 +       lockdep_off();
30247 +       err = do_splice_to(in, ppos, pipe, len, flags);
30248 +       lockdep_on();
30249 +       file_accessed(in);
30250 +       if (err >= 0)
30251 +               vfsub_update_h_iattr(&in->f_path, /*did*/NULL); /*ignore*/
30252 +       return err;
30253 +}
30254 +
30255 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
30256 +                      loff_t *ppos, size_t len, unsigned int flags)
30257 +{
30258 +       long err;
30259 +
30260 +       lockdep_off();
30261 +       err = do_splice_from(pipe, out, ppos, len, flags);
30262 +       lockdep_on();
30263 +       if (err >= 0)
30264 +               vfsub_update_h_iattr(&out->f_path, /*did*/NULL); /*ignore*/
30265 +       return err;
30266 +}
30267 +
30268 +int vfsub_fsync(struct file *file, struct path *path, int datasync)
30269 +{
30270 +       int err;
30271 +
30272 +       /* file can be NULL */
30273 +       lockdep_off();
30274 +       err = vfs_fsync(file, datasync);
30275 +       lockdep_on();
30276 +       if (!err) {
30277 +               if (!path) {
30278 +                       AuDebugOn(!file);
30279 +                       path = &file->f_path;
30280 +               }
30281 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
30282 +       }
30283 +       return err;
30284 +}
30285 +
30286 +/* cf. open.c:do_sys_truncate() and do_sys_ftruncate() */
30287 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
30288 +               struct file *h_file)
30289 +{
30290 +       int err;
30291 +       struct inode *h_inode;
30292 +       struct super_block *h_sb;
30293 +
30294 +       if (!h_file) {
30295 +               err = vfsub_truncate(h_path, length);
30296 +               goto out;
30297 +       }
30298 +
30299 +       h_inode = h_path->dentry->d_inode;
30300 +       h_sb = h_inode->i_sb;
30301 +       lockdep_off();
30302 +       sb_start_write(h_sb);
30303 +       lockdep_on();
30304 +       err = locks_verify_truncate(h_inode, h_file, length);
30305 +       if (!err)
30306 +               err = security_path_truncate(h_path);
30307 +       if (!err) {
30308 +               lockdep_off();
30309 +               err = do_truncate(h_path->dentry, length, attr, h_file);
30310 +               lockdep_on();
30311 +       }
30312 +       lockdep_off();
30313 +       sb_end_write(h_sb);
30314 +       lockdep_on();
30315 +
30316 +out:
30317 +       return err;
30318 +}
30319 +
30320 +/* ---------------------------------------------------------------------- */
30321 +
30322 +struct au_vfsub_mkdir_args {
30323 +       int *errp;
30324 +       struct inode *dir;
30325 +       struct path *path;
30326 +       int mode;
30327 +};
30328 +
30329 +static void au_call_vfsub_mkdir(void *args)
30330 +{
30331 +       struct au_vfsub_mkdir_args *a = args;
30332 +       *a->errp = vfsub_mkdir(a->dir, a->path, a->mode);
30333 +}
30334 +
30335 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode)
30336 +{
30337 +       int err, do_sio, wkq_err;
30338 +
30339 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
30340 +       if (!do_sio) {
30341 +               lockdep_off();
30342 +               err = vfsub_mkdir(dir, path, mode);
30343 +               lockdep_on();
30344 +       } else {
30345 +               struct au_vfsub_mkdir_args args = {
30346 +                       .errp   = &err,
30347 +                       .dir    = dir,
30348 +                       .path   = path,
30349 +                       .mode   = mode
30350 +               };
30351 +               wkq_err = au_wkq_wait(au_call_vfsub_mkdir, &args);
30352 +               if (unlikely(wkq_err))
30353 +                       err = wkq_err;
30354 +       }
30355 +
30356 +       return err;
30357 +}
30358 +
30359 +struct au_vfsub_rmdir_args {
30360 +       int *errp;
30361 +       struct inode *dir;
30362 +       struct path *path;
30363 +};
30364 +
30365 +static void au_call_vfsub_rmdir(void *args)
30366 +{
30367 +       struct au_vfsub_rmdir_args *a = args;
30368 +       *a->errp = vfsub_rmdir(a->dir, a->path);
30369 +}
30370 +
30371 +int vfsub_sio_rmdir(struct inode *dir, struct path *path)
30372 +{
30373 +       int err, do_sio, wkq_err;
30374 +
30375 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
30376 +       if (!do_sio) {
30377 +               lockdep_off();
30378 +               err = vfsub_rmdir(dir, path);
30379 +               lockdep_on();
30380 +       } else {
30381 +               struct au_vfsub_rmdir_args args = {
30382 +                       .errp   = &err,
30383 +                       .dir    = dir,
30384 +                       .path   = path
30385 +               };
30386 +               wkq_err = au_wkq_wait(au_call_vfsub_rmdir, &args);
30387 +               if (unlikely(wkq_err))
30388 +                       err = wkq_err;
30389 +       }
30390 +
30391 +       return err;
30392 +}
30393 +
30394 +/* ---------------------------------------------------------------------- */
30395 +
30396 +struct notify_change_args {
30397 +       int *errp;
30398 +       struct path *path;
30399 +       struct iattr *ia;
30400 +       struct inode **delegated_inode;
30401 +};
30402 +
30403 +static void call_notify_change(void *args)
30404 +{
30405 +       struct notify_change_args *a = args;
30406 +       struct inode *h_inode;
30407 +
30408 +       h_inode = a->path->dentry->d_inode;
30409 +       IMustLock(h_inode);
30410 +
30411 +       *a->errp = -EPERM;
30412 +       if (!IS_IMMUTABLE(h_inode) && !IS_APPEND(h_inode)) {
30413 +               lockdep_off();
30414 +               *a->errp = notify_change(a->path->dentry, a->ia,
30415 +                                        a->delegated_inode);
30416 +               lockdep_on();
30417 +               if (!*a->errp)
30418 +                       vfsub_update_h_iattr(a->path, /*did*/NULL); /*ignore*/
30419 +       }
30420 +       AuTraceErr(*a->errp);
30421 +}
30422 +
30423 +int vfsub_notify_change(struct path *path, struct iattr *ia,
30424 +                       struct inode **delegated_inode)
30425 +{
30426 +       int err;
30427 +       struct notify_change_args args = {
30428 +               .errp                   = &err,
30429 +               .path                   = path,
30430 +               .ia                     = ia,
30431 +               .delegated_inode        = delegated_inode
30432 +       };
30433 +
30434 +       call_notify_change(&args);
30435 +
30436 +       return err;
30437 +}
30438 +
30439 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
30440 +                           struct inode **delegated_inode)
30441 +{
30442 +       int err, wkq_err;
30443 +       struct notify_change_args args = {
30444 +               .errp                   = &err,
30445 +               .path                   = path,
30446 +               .ia                     = ia,
30447 +               .delegated_inode        = delegated_inode
30448 +       };
30449 +
30450 +       wkq_err = au_wkq_wait(call_notify_change, &args);
30451 +       if (unlikely(wkq_err))
30452 +               err = wkq_err;
30453 +
30454 +       return err;
30455 +}
30456 +
30457 +/* ---------------------------------------------------------------------- */
30458 +
30459 +struct unlink_args {
30460 +       int *errp;
30461 +       struct inode *dir;
30462 +       struct path *path;
30463 +       struct inode **delegated_inode;
30464 +};
30465 +
30466 +static void call_unlink(void *args)
30467 +{
30468 +       struct unlink_args *a = args;
30469 +       struct dentry *d = a->path->dentry;
30470 +       struct inode *h_inode;
30471 +       const int stop_sillyrename = (au_test_nfs(d->d_sb)
30472 +                                     && au_dcount(d) == 1);
30473 +
30474 +       IMustLock(a->dir);
30475 +
30476 +       a->path->dentry = d->d_parent;
30477 +       *a->errp = security_path_unlink(a->path, d);
30478 +       a->path->dentry = d;
30479 +       if (unlikely(*a->errp))
30480 +               return;
30481 +
30482 +       if (!stop_sillyrename)
30483 +               dget(d);
30484 +       h_inode = d->d_inode;
30485 +       if (h_inode)
30486 +               ihold(h_inode);
30487 +
30488 +       lockdep_off();
30489 +       *a->errp = vfs_unlink(a->dir, d, a->delegated_inode);
30490 +       lockdep_on();
30491 +       if (!*a->errp) {
30492 +               struct path tmp = {
30493 +                       .dentry = d->d_parent,
30494 +                       .mnt    = a->path->mnt
30495 +               };
30496 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
30497 +       }
30498 +
30499 +       if (!stop_sillyrename)
30500 +               dput(d);
30501 +       if (h_inode)
30502 +               iput(h_inode);
30503 +
30504 +       AuTraceErr(*a->errp);
30505 +}
30506 +
30507 +/*
30508 + * @dir: must be locked.
30509 + * @dentry: target dentry.
30510 + */
30511 +int vfsub_unlink(struct inode *dir, struct path *path,
30512 +                struct inode **delegated_inode, int force)
30513 +{
30514 +       int err;
30515 +       struct unlink_args args = {
30516 +               .errp                   = &err,
30517 +               .dir                    = dir,
30518 +               .path                   = path,
30519 +               .delegated_inode        = delegated_inode
30520 +       };
30521 +
30522 +       if (!force)
30523 +               call_unlink(&args);
30524 +       else {
30525 +               int wkq_err;
30526 +
30527 +               wkq_err = au_wkq_wait(call_unlink, &args);
30528 +               if (unlikely(wkq_err))
30529 +                       err = wkq_err;
30530 +       }
30531 +
30532 +       return err;
30533 +}
30534 diff -urN /usr/share/empty/fs/aufs/vfsub.h linux/fs/aufs/vfsub.h
30535 --- /usr/share/empty/fs/aufs/vfsub.h    1970-01-01 01:00:00.000000000 +0100
30536 +++ linux/fs/aufs/vfsub.h       2015-03-27 21:56:35.470128334 +0100
30537 @@ -0,0 +1,301 @@
30538 +/*
30539 + * Copyright (C) 2005-2015 Junjiro R. Okajima
30540 + *
30541 + * This program, aufs is free software; you can redistribute it and/or modify
30542 + * it under the terms of the GNU General Public License as published by
30543 + * the Free Software Foundation; either version 2 of the License, or
30544 + * (at your option) any later version.
30545 + *
30546 + * This program is distributed in the hope that it will be useful,
30547 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30548 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30549 + * GNU General Public License for more details.
30550 + *
30551 + * You should have received a copy of the GNU General Public License
30552 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30553 + */
30554 +
30555 +/*
30556 + * sub-routines for VFS
30557 + */
30558 +
30559 +#ifndef __AUFS_VFSUB_H__
30560 +#define __AUFS_VFSUB_H__
30561 +
30562 +#ifdef __KERNEL__
30563 +
30564 +#include <linux/fs.h>
30565 +#include <linux/lglock.h>
30566 +#include <linux/mount.h>
30567 +#include <linux/xattr.h>
30568 +#include "debug.h"
30569 +
30570 +/* copied from linux/fs/internal.h */
30571 +/* todo: BAD approach!! */
30572 +extern void __mnt_drop_write(struct vfsmount *);
30573 +extern spinlock_t inode_sb_list_lock;
30574 +
30575 +/* ---------------------------------------------------------------------- */
30576 +
30577 +/* lock subclass for lower inode */
30578 +/* default MAX_LOCKDEP_SUBCLASSES(8) is not enough */
30579 +/* reduce? gave up. */
30580 +enum {
30581 +       AuLsc_I_Begin = I_MUTEX_PARENT2, /* 5 */
30582 +       AuLsc_I_PARENT,         /* lower inode, parent first */
30583 +       AuLsc_I_PARENT2,        /* copyup dirs */
30584 +       AuLsc_I_PARENT3,        /* copyup wh */
30585 +       AuLsc_I_CHILD,
30586 +       AuLsc_I_CHILD2,
30587 +       AuLsc_I_End
30588 +};
30589 +
30590 +/* to debug easier, do not make them inlined functions */
30591 +#define MtxMustLock(mtx)       AuDebugOn(!mutex_is_locked(mtx))
30592 +#define IMustLock(i)           MtxMustLock(&(i)->i_mutex)
30593 +
30594 +/* ---------------------------------------------------------------------- */
30595 +
30596 +static inline void vfsub_drop_nlink(struct inode *inode)
30597 +{
30598 +       AuDebugOn(!inode->i_nlink);
30599 +       drop_nlink(inode);
30600 +}
30601 +
30602 +static inline void vfsub_dead_dir(struct inode *inode)
30603 +{
30604 +       AuDebugOn(!S_ISDIR(inode->i_mode));
30605 +       inode->i_flags |= S_DEAD;
30606 +       clear_nlink(inode);
30607 +}
30608 +
30609 +static inline int vfsub_native_ro(struct inode *inode)
30610 +{
30611 +       return (inode->i_sb->s_flags & MS_RDONLY)
30612 +               || IS_RDONLY(inode)
30613 +               /* || IS_APPEND(inode) */
30614 +               || IS_IMMUTABLE(inode);
30615 +}
30616 +
30617 +/* ---------------------------------------------------------------------- */
30618 +
30619 +int vfsub_update_h_iattr(struct path *h_path, int *did);
30620 +struct file *vfsub_dentry_open(struct path *path, int flags);
30621 +struct file *vfsub_filp_open(const char *path, int oflags, int mode);
30622 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path);
30623 +
30624 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
30625 +                                   int len);
30626 +
30627 +struct vfsub_lkup_one_args {
30628 +       struct dentry **errp;
30629 +       struct qstr *name;
30630 +       struct dentry *parent;
30631 +};
30632 +
30633 +static inline struct dentry *vfsub_lkup_one(struct qstr *name,
30634 +                                           struct dentry *parent)
30635 +{
30636 +       return vfsub_lookup_one_len(name->name, parent, name->len);
30637 +}
30638 +
30639 +void vfsub_call_lkup_one(void *args);
30640 +
30641 +/* ---------------------------------------------------------------------- */
30642 +
30643 +static inline int vfsub_mnt_want_write(struct vfsmount *mnt)
30644 +{
30645 +       int err;
30646 +
30647 +       lockdep_off();
30648 +       err = mnt_want_write(mnt);
30649 +       lockdep_on();
30650 +       return err;
30651 +}
30652 +
30653 +static inline void vfsub_mnt_drop_write(struct vfsmount *mnt)
30654 +{
30655 +       lockdep_off();
30656 +       mnt_drop_write(mnt);
30657 +       lockdep_on();
30658 +}
30659 +
30660 +static inline void vfsub_mnt_drop_write_file(struct file *file)
30661 +{
30662 +       lockdep_off();
30663 +       mnt_drop_write_file(file);
30664 +       lockdep_on();
30665 +}
30666 +
30667 +/* ---------------------------------------------------------------------- */
30668 +
30669 +struct au_hinode;
30670 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
30671 +                                struct dentry *d2, struct au_hinode *hdir2);
30672 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
30673 +                        struct dentry *d2, struct au_hinode *hdir2);
30674 +
30675 +int vfsub_create(struct inode *dir, struct path *path, int mode,
30676 +                bool want_excl);
30677 +int vfsub_symlink(struct inode *dir, struct path *path,
30678 +                 const char *symname);
30679 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev);
30680 +int vfsub_link(struct dentry *src_dentry, struct inode *dir,
30681 +              struct path *path, struct inode **delegated_inode);
30682 +int vfsub_rename(struct inode *src_hdir, struct dentry *src_dentry,
30683 +                struct inode *hdir, struct path *path,
30684 +                struct inode **delegated_inode);
30685 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode);
30686 +int vfsub_rmdir(struct inode *dir, struct path *path);
30687 +
30688 +/* ---------------------------------------------------------------------- */
30689 +
30690 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
30691 +                    loff_t *ppos);
30692 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
30693 +                       loff_t *ppos);
30694 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
30695 +                     loff_t *ppos);
30696 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count,
30697 +                     loff_t *ppos);
30698 +int vfsub_flush(struct file *file, fl_owner_t id);
30699 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx);
30700 +
30701 +static inline loff_t vfsub_f_size_read(struct file *file)
30702 +{
30703 +       return i_size_read(file_inode(file));
30704 +}
30705 +
30706 +static inline unsigned int vfsub_file_flags(struct file *file)
30707 +{
30708 +       unsigned int flags;
30709 +
30710 +       spin_lock(&file->f_lock);
30711 +       flags = file->f_flags;
30712 +       spin_unlock(&file->f_lock);
30713 +
30714 +       return flags;
30715 +}
30716 +
30717 +static inline void vfsub_file_accessed(struct file *h_file)
30718 +{
30719 +       file_accessed(h_file);
30720 +       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL); /*ignore*/
30721 +}
30722 +
30723 +static inline void vfsub_touch_atime(struct vfsmount *h_mnt,
30724 +                                    struct dentry *h_dentry)
30725 +{
30726 +       struct path h_path = {
30727 +               .dentry = h_dentry,
30728 +               .mnt    = h_mnt
30729 +       };
30730 +       touch_atime(&h_path);
30731 +       vfsub_update_h_iattr(&h_path, /*did*/NULL); /*ignore*/
30732 +}
30733 +
30734 +static inline int vfsub_update_time(struct inode *h_inode, struct timespec *ts,
30735 +                                   int flags)
30736 +{
30737 +       return update_time(h_inode, ts, flags);
30738 +       /* no vfsub_update_h_iattr() since we don't have struct path */
30739 +}
30740 +
30741 +long vfsub_splice_to(struct file *in, loff_t *ppos,
30742 +                    struct pipe_inode_info *pipe, size_t len,
30743 +                    unsigned int flags);
30744 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
30745 +                      loff_t *ppos, size_t len, unsigned int flags);
30746 +
30747 +static inline long vfsub_truncate(struct path *path, loff_t length)
30748 +{
30749 +       long err;
30750 +
30751 +       lockdep_off();
30752 +       err = vfs_truncate(path, length);
30753 +       lockdep_on();
30754 +       return err;
30755 +}
30756 +
30757 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
30758 +               struct file *h_file);
30759 +int vfsub_fsync(struct file *file, struct path *path, int datasync);
30760 +
30761 +/* ---------------------------------------------------------------------- */
30762 +
30763 +static inline loff_t vfsub_llseek(struct file *file, loff_t offset, int origin)
30764 +{
30765 +       loff_t err;
30766 +
30767 +       lockdep_off();
30768 +       err = vfs_llseek(file, offset, origin);
30769 +       lockdep_on();
30770 +       return err;
30771 +}
30772 +
30773 +/* ---------------------------------------------------------------------- */
30774 +
30775 +/* dirty workaround for strict type of fmode_t */
30776 +union vfsub_fmu {
30777 +       fmode_t fm;
30778 +       unsigned int ui;
30779 +};
30780 +
30781 +static inline unsigned int vfsub_fmode_to_uint(fmode_t fm)
30782 +{
30783 +       union vfsub_fmu u = {
30784 +               .fm = fm
30785 +       };
30786 +
30787 +       BUILD_BUG_ON(sizeof(u.fm) != sizeof(u.ui));
30788 +
30789 +       return u.ui;
30790 +}
30791 +
30792 +static inline fmode_t vfsub_uint_to_fmode(unsigned int ui)
30793 +{
30794 +       union vfsub_fmu u = {
30795 +               .ui = ui
30796 +       };
30797 +
30798 +       return u.fm;
30799 +}
30800 +
30801 +/* ---------------------------------------------------------------------- */
30802 +
30803 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode);
30804 +int vfsub_sio_rmdir(struct inode *dir, struct path *path);
30805 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
30806 +                           struct inode **delegated_inode);
30807 +int vfsub_notify_change(struct path *path, struct iattr *ia,
30808 +                       struct inode **delegated_inode);
30809 +int vfsub_unlink(struct inode *dir, struct path *path,
30810 +                struct inode **delegated_inode, int force);
30811 +
30812 +/* ---------------------------------------------------------------------- */
30813 +
30814 +static inline int vfsub_setxattr(struct dentry *dentry, const char *name,
30815 +                                const void *value, size_t size, int flags)
30816 +{
30817 +       int err;
30818 +
30819 +       lockdep_off();
30820 +       err = vfs_setxattr(dentry, name, value, size, flags);
30821 +       lockdep_on();
30822 +
30823 +       return err;
30824 +}
30825 +
30826 +static inline int vfsub_removexattr(struct dentry *dentry, const char *name)
30827 +{
30828 +       int err;
30829 +
30830 +       lockdep_off();
30831 +       err = vfs_removexattr(dentry, name);
30832 +       lockdep_on();
30833 +
30834 +       return err;
30835 +}
30836 +
30837 +#endif /* __KERNEL__ */
30838 +#endif /* __AUFS_VFSUB_H__ */
30839 diff -urN /usr/share/empty/fs/aufs/wbr_policy.c linux/fs/aufs/wbr_policy.c
30840 --- /usr/share/empty/fs/aufs/wbr_policy.c       1970-01-01 01:00:00.000000000 +0100
30841 +++ linux/fs/aufs/wbr_policy.c  2015-03-27 21:56:35.470128334 +0100
30842 @@ -0,0 +1,765 @@
30843 +/*
30844 + * Copyright (C) 2005-2015 Junjiro R. Okajima
30845 + *
30846 + * This program, aufs is free software; you can redistribute it and/or modify
30847 + * it under the terms of the GNU General Public License as published by
30848 + * the Free Software Foundation; either version 2 of the License, or
30849 + * (at your option) any later version.
30850 + *
30851 + * This program is distributed in the hope that it will be useful,
30852 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30853 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30854 + * GNU General Public License for more details.
30855 + *
30856 + * You should have received a copy of the GNU General Public License
30857 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30858 + */
30859 +
30860 +/*
30861 + * policies for selecting one among multiple writable branches
30862 + */
30863 +
30864 +#include <linux/statfs.h>
30865 +#include "aufs.h"
30866 +
30867 +/* subset of cpup_attr() */
30868 +static noinline_for_stack
30869 +int au_cpdown_attr(struct path *h_path, struct dentry *h_src)
30870 +{
30871 +       int err, sbits;
30872 +       struct iattr ia;
30873 +       struct inode *h_isrc;
30874 +
30875 +       h_isrc = h_src->d_inode;
30876 +       ia.ia_valid = ATTR_FORCE | ATTR_MODE | ATTR_UID | ATTR_GID;
30877 +       ia.ia_mode = h_isrc->i_mode;
30878 +       ia.ia_uid = h_isrc->i_uid;
30879 +       ia.ia_gid = h_isrc->i_gid;
30880 +       sbits = !!(ia.ia_mode & (S_ISUID | S_ISGID));
30881 +       au_cpup_attr_flags(h_path->dentry->d_inode, h_isrc->i_flags);
30882 +       /* no delegation since it is just created */
30883 +       err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
30884 +
30885 +       /* is this nfs only? */
30886 +       if (!err && sbits && au_test_nfs(h_path->dentry->d_sb)) {
30887 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
30888 +               ia.ia_mode = h_isrc->i_mode;
30889 +               err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
30890 +       }
30891 +
30892 +       return err;
30893 +}
30894 +
30895 +#define AuCpdown_PARENT_OPQ    1
30896 +#define AuCpdown_WHED          (1 << 1)
30897 +#define AuCpdown_MADE_DIR      (1 << 2)
30898 +#define AuCpdown_DIROPQ                (1 << 3)
30899 +#define au_ftest_cpdown(flags, name)   ((flags) & AuCpdown_##name)
30900 +#define au_fset_cpdown(flags, name) \
30901 +       do { (flags) |= AuCpdown_##name; } while (0)
30902 +#define au_fclr_cpdown(flags, name) \
30903 +       do { (flags) &= ~AuCpdown_##name; } while (0)
30904 +
30905 +static int au_cpdown_dir_opq(struct dentry *dentry, aufs_bindex_t bdst,
30906 +                            unsigned int *flags)
30907 +{
30908 +       int err;
30909 +       struct dentry *opq_dentry;
30910 +
30911 +       opq_dentry = au_diropq_create(dentry, bdst);
30912 +       err = PTR_ERR(opq_dentry);
30913 +       if (IS_ERR(opq_dentry))
30914 +               goto out;
30915 +       dput(opq_dentry);
30916 +       au_fset_cpdown(*flags, DIROPQ);
30917 +
30918 +out:
30919 +       return err;
30920 +}
30921 +
30922 +static int au_cpdown_dir_wh(struct dentry *dentry, struct dentry *h_parent,
30923 +                           struct inode *dir, aufs_bindex_t bdst)
30924 +{
30925 +       int err;
30926 +       struct path h_path;
30927 +       struct au_branch *br;
30928 +
30929 +       br = au_sbr(dentry->d_sb, bdst);
30930 +       h_path.dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
30931 +       err = PTR_ERR(h_path.dentry);
30932 +       if (IS_ERR(h_path.dentry))
30933 +               goto out;
30934 +
30935 +       err = 0;
30936 +       if (h_path.dentry->d_inode) {
30937 +               h_path.mnt = au_br_mnt(br);
30938 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bdst), &h_path,
30939 +                                         dentry);
30940 +       }
30941 +       dput(h_path.dentry);
30942 +
30943 +out:
30944 +       return err;
30945 +}
30946 +
30947 +static int au_cpdown_dir(struct dentry *dentry, aufs_bindex_t bdst,
30948 +                        struct au_pin *pin,
30949 +                        struct dentry *h_parent, void *arg)
30950 +{
30951 +       int err, rerr;
30952 +       aufs_bindex_t bopq, bstart;
30953 +       struct path h_path;
30954 +       struct dentry *parent;
30955 +       struct inode *h_dir, *h_inode, *inode, *dir;
30956 +       unsigned int *flags = arg;
30957 +
30958 +       bstart = au_dbstart(dentry);
30959 +       /* dentry is di-locked */
30960 +       parent = dget_parent(dentry);
30961 +       dir = parent->d_inode;
30962 +       h_dir = h_parent->d_inode;
30963 +       AuDebugOn(h_dir != au_h_iptr(dir, bdst));
30964 +       IMustLock(h_dir);
30965 +
30966 +       err = au_lkup_neg(dentry, bdst, /*wh*/0);
30967 +       if (unlikely(err < 0))
30968 +               goto out;
30969 +       h_path.dentry = au_h_dptr(dentry, bdst);
30970 +       h_path.mnt = au_sbr_mnt(dentry->d_sb, bdst);
30971 +       err = vfsub_sio_mkdir(au_h_iptr(dir, bdst), &h_path,
30972 +                             S_IRWXU | S_IRUGO | S_IXUGO);
30973 +       if (unlikely(err))
30974 +               goto out_put;
30975 +       au_fset_cpdown(*flags, MADE_DIR);
30976 +
30977 +       bopq = au_dbdiropq(dentry);
30978 +       au_fclr_cpdown(*flags, WHED);
30979 +       au_fclr_cpdown(*flags, DIROPQ);
30980 +       if (au_dbwh(dentry) == bdst)
30981 +               au_fset_cpdown(*flags, WHED);
30982 +       if (!au_ftest_cpdown(*flags, PARENT_OPQ) && bopq <= bdst)
30983 +               au_fset_cpdown(*flags, PARENT_OPQ);
30984 +       h_inode = h_path.dentry->d_inode;
30985 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
30986 +       if (au_ftest_cpdown(*flags, WHED)) {
30987 +               err = au_cpdown_dir_opq(dentry, bdst, flags);
30988 +               if (unlikely(err)) {
30989 +                       mutex_unlock(&h_inode->i_mutex);
30990 +                       goto out_dir;
30991 +               }
30992 +       }
30993 +
30994 +       err = au_cpdown_attr(&h_path, au_h_dptr(dentry, bstart));
30995 +       mutex_unlock(&h_inode->i_mutex);
30996 +       if (unlikely(err))
30997 +               goto out_opq;
30998 +
30999 +       if (au_ftest_cpdown(*flags, WHED)) {
31000 +               err = au_cpdown_dir_wh(dentry, h_parent, dir, bdst);
31001 +               if (unlikely(err))
31002 +                       goto out_opq;
31003 +       }
31004 +
31005 +       inode = dentry->d_inode;
31006 +       if (au_ibend(inode) < bdst)
31007 +               au_set_ibend(inode, bdst);
31008 +       au_set_h_iptr(inode, bdst, au_igrab(h_inode),
31009 +                     au_hi_flags(inode, /*isdir*/1));
31010 +       au_fhsm_wrote(dentry->d_sb, bdst, /*force*/0);
31011 +       goto out; /* success */
31012 +
31013 +       /* revert */
31014 +out_opq:
31015 +       if (au_ftest_cpdown(*flags, DIROPQ)) {
31016 +               mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
31017 +               rerr = au_diropq_remove(dentry, bdst);
31018 +               mutex_unlock(&h_inode->i_mutex);
31019 +               if (unlikely(rerr)) {
31020 +                       AuIOErr("failed removing diropq for %pd b%d (%d)\n",
31021 +                               dentry, bdst, rerr);
31022 +                       err = -EIO;
31023 +                       goto out;
31024 +               }
31025 +       }
31026 +out_dir:
31027 +       if (au_ftest_cpdown(*flags, MADE_DIR)) {
31028 +               rerr = vfsub_sio_rmdir(au_h_iptr(dir, bdst), &h_path);
31029 +               if (unlikely(rerr)) {
31030 +                       AuIOErr("failed removing %pd b%d (%d)\n",
31031 +                               dentry, bdst, rerr);
31032 +                       err = -EIO;
31033 +               }
31034 +       }
31035 +out_put:
31036 +       au_set_h_dptr(dentry, bdst, NULL);
31037 +       if (au_dbend(dentry) == bdst)
31038 +               au_update_dbend(dentry);
31039 +out:
31040 +       dput(parent);
31041 +       return err;
31042 +}
31043 +
31044 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst)
31045 +{
31046 +       int err;
31047 +       unsigned int flags;
31048 +
31049 +       flags = 0;
31050 +       err = au_cp_dirs(dentry, bdst, au_cpdown_dir, &flags);
31051 +
31052 +       return err;
31053 +}
31054 +
31055 +/* ---------------------------------------------------------------------- */
31056 +
31057 +/* policies for create */
31058 +
31059 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex)
31060 +{
31061 +       int err, i, j, ndentry;
31062 +       aufs_bindex_t bopq;
31063 +       struct au_dcsub_pages dpages;
31064 +       struct au_dpage *dpage;
31065 +       struct dentry **dentries, *parent, *d;
31066 +
31067 +       err = au_dpages_init(&dpages, GFP_NOFS);
31068 +       if (unlikely(err))
31069 +               goto out;
31070 +       parent = dget_parent(dentry);
31071 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/0);
31072 +       if (unlikely(err))
31073 +               goto out_free;
31074 +
31075 +       err = bindex;
31076 +       for (i = 0; i < dpages.ndpage; i++) {
31077 +               dpage = dpages.dpages + i;
31078 +               dentries = dpage->dentries;
31079 +               ndentry = dpage->ndentry;
31080 +               for (j = 0; j < ndentry; j++) {
31081 +                       d = dentries[j];
31082 +                       di_read_lock_parent2(d, !AuLock_IR);
31083 +                       bopq = au_dbdiropq(d);
31084 +                       di_read_unlock(d, !AuLock_IR);
31085 +                       if (bopq >= 0 && bopq < err)
31086 +                               err = bopq;
31087 +               }
31088 +       }
31089 +
31090 +out_free:
31091 +       dput(parent);
31092 +       au_dpages_free(&dpages);
31093 +out:
31094 +       return err;
31095 +}
31096 +
31097 +static int au_wbr_bu(struct super_block *sb, aufs_bindex_t bindex)
31098 +{
31099 +       for (; bindex >= 0; bindex--)
31100 +               if (!au_br_rdonly(au_sbr(sb, bindex)))
31101 +                       return bindex;
31102 +       return -EROFS;
31103 +}
31104 +
31105 +/* top down parent */
31106 +static int au_wbr_create_tdp(struct dentry *dentry,
31107 +                            unsigned int flags __maybe_unused)
31108 +{
31109 +       int err;
31110 +       aufs_bindex_t bstart, bindex;
31111 +       struct super_block *sb;
31112 +       struct dentry *parent, *h_parent;
31113 +
31114 +       sb = dentry->d_sb;
31115 +       bstart = au_dbstart(dentry);
31116 +       err = bstart;
31117 +       if (!au_br_rdonly(au_sbr(sb, bstart)))
31118 +               goto out;
31119 +
31120 +       err = -EROFS;
31121 +       parent = dget_parent(dentry);
31122 +       for (bindex = au_dbstart(parent); bindex < bstart; bindex++) {
31123 +               h_parent = au_h_dptr(parent, bindex);
31124 +               if (!h_parent || !h_parent->d_inode)
31125 +                       continue;
31126 +
31127 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
31128 +                       err = bindex;
31129 +                       break;
31130 +               }
31131 +       }
31132 +       dput(parent);
31133 +
31134 +       /* bottom up here */
31135 +       if (unlikely(err < 0)) {
31136 +               err = au_wbr_bu(sb, bstart - 1);
31137 +               if (err >= 0)
31138 +                       err = au_wbr_nonopq(dentry, err);
31139 +       }
31140 +
31141 +out:
31142 +       AuDbg("b%d\n", err);
31143 +       return err;
31144 +}
31145 +
31146 +/* ---------------------------------------------------------------------- */
31147 +
31148 +/* an exception for the policy other than tdp */
31149 +static int au_wbr_create_exp(struct dentry *dentry)
31150 +{
31151 +       int err;
31152 +       aufs_bindex_t bwh, bdiropq;
31153 +       struct dentry *parent;
31154 +
31155 +       err = -1;
31156 +       bwh = au_dbwh(dentry);
31157 +       parent = dget_parent(dentry);
31158 +       bdiropq = au_dbdiropq(parent);
31159 +       if (bwh >= 0) {
31160 +               if (bdiropq >= 0)
31161 +                       err = min(bdiropq, bwh);
31162 +               else
31163 +                       err = bwh;
31164 +               AuDbg("%d\n", err);
31165 +       } else if (bdiropq >= 0) {
31166 +               err = bdiropq;
31167 +               AuDbg("%d\n", err);
31168 +       }
31169 +       dput(parent);
31170 +
31171 +       if (err >= 0)
31172 +               err = au_wbr_nonopq(dentry, err);
31173 +
31174 +       if (err >= 0 && au_br_rdonly(au_sbr(dentry->d_sb, err)))
31175 +               err = -1;
31176 +
31177 +       AuDbg("%d\n", err);
31178 +       return err;
31179 +}
31180 +
31181 +/* ---------------------------------------------------------------------- */
31182 +
31183 +/* round robin */
31184 +static int au_wbr_create_init_rr(struct super_block *sb)
31185 +{
31186 +       int err;
31187 +
31188 +       err = au_wbr_bu(sb, au_sbend(sb));
31189 +       atomic_set(&au_sbi(sb)->si_wbr_rr_next, -err); /* less important */
31190 +       /* smp_mb(); */
31191 +
31192 +       AuDbg("b%d\n", err);
31193 +       return err;
31194 +}
31195 +
31196 +static int au_wbr_create_rr(struct dentry *dentry, unsigned int flags)
31197 +{
31198 +       int err, nbr;
31199 +       unsigned int u;
31200 +       aufs_bindex_t bindex, bend;
31201 +       struct super_block *sb;
31202 +       atomic_t *next;
31203 +
31204 +       err = au_wbr_create_exp(dentry);
31205 +       if (err >= 0)
31206 +               goto out;
31207 +
31208 +       sb = dentry->d_sb;
31209 +       next = &au_sbi(sb)->si_wbr_rr_next;
31210 +       bend = au_sbend(sb);
31211 +       nbr = bend + 1;
31212 +       for (bindex = 0; bindex <= bend; bindex++) {
31213 +               if (!au_ftest_wbr(flags, DIR)) {
31214 +                       err = atomic_dec_return(next) + 1;
31215 +                       /* modulo for 0 is meaningless */
31216 +                       if (unlikely(!err))
31217 +                               err = atomic_dec_return(next) + 1;
31218 +               } else
31219 +                       err = atomic_read(next);
31220 +               AuDbg("%d\n", err);
31221 +               u = err;
31222 +               err = u % nbr;
31223 +               AuDbg("%d\n", err);
31224 +               if (!au_br_rdonly(au_sbr(sb, err)))
31225 +                       break;
31226 +               err = -EROFS;
31227 +       }
31228 +
31229 +       if (err >= 0)
31230 +               err = au_wbr_nonopq(dentry, err);
31231 +
31232 +out:
31233 +       AuDbg("%d\n", err);
31234 +       return err;
31235 +}
31236 +
31237 +/* ---------------------------------------------------------------------- */
31238 +
31239 +/* most free space */
31240 +static void au_mfs(struct dentry *dentry, struct dentry *parent)
31241 +{
31242 +       struct super_block *sb;
31243 +       struct au_branch *br;
31244 +       struct au_wbr_mfs *mfs;
31245 +       struct dentry *h_parent;
31246 +       aufs_bindex_t bindex, bend;
31247 +       int err;
31248 +       unsigned long long b, bavail;
31249 +       struct path h_path;
31250 +       /* reduce the stack usage */
31251 +       struct kstatfs *st;
31252 +
31253 +       st = kmalloc(sizeof(*st), GFP_NOFS);
31254 +       if (unlikely(!st)) {
31255 +               AuWarn1("failed updating mfs(%d), ignored\n", -ENOMEM);
31256 +               return;
31257 +       }
31258 +
31259 +       bavail = 0;
31260 +       sb = dentry->d_sb;
31261 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31262 +       MtxMustLock(&mfs->mfs_lock);
31263 +       mfs->mfs_bindex = -EROFS;
31264 +       mfs->mfsrr_bytes = 0;
31265 +       if (!parent) {
31266 +               bindex = 0;
31267 +               bend = au_sbend(sb);
31268 +       } else {
31269 +               bindex = au_dbstart(parent);
31270 +               bend = au_dbtaildir(parent);
31271 +       }
31272 +
31273 +       for (; bindex <= bend; bindex++) {
31274 +               if (parent) {
31275 +                       h_parent = au_h_dptr(parent, bindex);
31276 +                       if (!h_parent || !h_parent->d_inode)
31277 +                               continue;
31278 +               }
31279 +               br = au_sbr(sb, bindex);
31280 +               if (au_br_rdonly(br))
31281 +                       continue;
31282 +
31283 +               /* sb->s_root for NFS is unreliable */
31284 +               h_path.mnt = au_br_mnt(br);
31285 +               h_path.dentry = h_path.mnt->mnt_root;
31286 +               err = vfs_statfs(&h_path, st);
31287 +               if (unlikely(err)) {
31288 +                       AuWarn1("failed statfs, b%d, %d\n", bindex, err);
31289 +                       continue;
31290 +               }
31291 +
31292 +               /* when the available size is equal, select the lower one */
31293 +               BUILD_BUG_ON(sizeof(b) < sizeof(st->f_bavail)
31294 +                            || sizeof(b) < sizeof(st->f_bsize));
31295 +               b = st->f_bavail * st->f_bsize;
31296 +               br->br_wbr->wbr_bytes = b;
31297 +               if (b >= bavail) {
31298 +                       bavail = b;
31299 +                       mfs->mfs_bindex = bindex;
31300 +                       mfs->mfs_jiffy = jiffies;
31301 +               }
31302 +       }
31303 +
31304 +       mfs->mfsrr_bytes = bavail;
31305 +       AuDbg("b%d\n", mfs->mfs_bindex);
31306 +       kfree(st);
31307 +}
31308 +
31309 +static int au_wbr_create_mfs(struct dentry *dentry, unsigned int flags)
31310 +{
31311 +       int err;
31312 +       struct dentry *parent;
31313 +       struct super_block *sb;
31314 +       struct au_wbr_mfs *mfs;
31315 +
31316 +       err = au_wbr_create_exp(dentry);
31317 +       if (err >= 0)
31318 +               goto out;
31319 +
31320 +       sb = dentry->d_sb;
31321 +       parent = NULL;
31322 +       if (au_ftest_wbr(flags, PARENT))
31323 +               parent = dget_parent(dentry);
31324 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31325 +       mutex_lock(&mfs->mfs_lock);
31326 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
31327 +           || mfs->mfs_bindex < 0
31328 +           || au_br_rdonly(au_sbr(sb, mfs->mfs_bindex)))
31329 +               au_mfs(dentry, parent);
31330 +       mutex_unlock(&mfs->mfs_lock);
31331 +       err = mfs->mfs_bindex;
31332 +       dput(parent);
31333 +
31334 +       if (err >= 0)
31335 +               err = au_wbr_nonopq(dentry, err);
31336 +
31337 +out:
31338 +       AuDbg("b%d\n", err);
31339 +       return err;
31340 +}
31341 +
31342 +static int au_wbr_create_init_mfs(struct super_block *sb)
31343 +{
31344 +       struct au_wbr_mfs *mfs;
31345 +
31346 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31347 +       mutex_init(&mfs->mfs_lock);
31348 +       mfs->mfs_jiffy = 0;
31349 +       mfs->mfs_bindex = -EROFS;
31350 +
31351 +       return 0;
31352 +}
31353 +
31354 +static int au_wbr_create_fin_mfs(struct super_block *sb __maybe_unused)
31355 +{
31356 +       mutex_destroy(&au_sbi(sb)->si_wbr_mfs.mfs_lock);
31357 +       return 0;
31358 +}
31359 +
31360 +/* ---------------------------------------------------------------------- */
31361 +
31362 +/* most free space and then round robin */
31363 +static int au_wbr_create_mfsrr(struct dentry *dentry, unsigned int flags)
31364 +{
31365 +       int err;
31366 +       struct au_wbr_mfs *mfs;
31367 +
31368 +       err = au_wbr_create_mfs(dentry, flags);
31369 +       if (err >= 0) {
31370 +               mfs = &au_sbi(dentry->d_sb)->si_wbr_mfs;
31371 +               mutex_lock(&mfs->mfs_lock);
31372 +               if (mfs->mfsrr_bytes < mfs->mfsrr_watermark)
31373 +                       err = au_wbr_create_rr(dentry, flags);
31374 +               mutex_unlock(&mfs->mfs_lock);
31375 +       }
31376 +
31377 +       AuDbg("b%d\n", err);
31378 +       return err;
31379 +}
31380 +
31381 +static int au_wbr_create_init_mfsrr(struct super_block *sb)
31382 +{
31383 +       int err;
31384 +
31385 +       au_wbr_create_init_mfs(sb); /* ignore */
31386 +       err = au_wbr_create_init_rr(sb);
31387 +
31388 +       return err;
31389 +}
31390 +
31391 +/* ---------------------------------------------------------------------- */
31392 +
31393 +/* top down parent and most free space */
31394 +static int au_wbr_create_pmfs(struct dentry *dentry, unsigned int flags)
31395 +{
31396 +       int err, e2;
31397 +       unsigned long long b;
31398 +       aufs_bindex_t bindex, bstart, bend;
31399 +       struct super_block *sb;
31400 +       struct dentry *parent, *h_parent;
31401 +       struct au_branch *br;
31402 +
31403 +       err = au_wbr_create_tdp(dentry, flags);
31404 +       if (unlikely(err < 0))
31405 +               goto out;
31406 +       parent = dget_parent(dentry);
31407 +       bstart = au_dbstart(parent);
31408 +       bend = au_dbtaildir(parent);
31409 +       if (bstart == bend)
31410 +               goto out_parent; /* success */
31411 +
31412 +       e2 = au_wbr_create_mfs(dentry, flags);
31413 +       if (e2 < 0)
31414 +               goto out_parent; /* success */
31415 +
31416 +       /* when the available size is equal, select upper one */
31417 +       sb = dentry->d_sb;
31418 +       br = au_sbr(sb, err);
31419 +       b = br->br_wbr->wbr_bytes;
31420 +       AuDbg("b%d, %llu\n", err, b);
31421 +
31422 +       for (bindex = bstart; bindex <= bend; bindex++) {
31423 +               h_parent = au_h_dptr(parent, bindex);
31424 +               if (!h_parent || !h_parent->d_inode)
31425 +                       continue;
31426 +
31427 +               br = au_sbr(sb, bindex);
31428 +               if (!au_br_rdonly(br) && br->br_wbr->wbr_bytes > b) {
31429 +                       b = br->br_wbr->wbr_bytes;
31430 +                       err = bindex;
31431 +                       AuDbg("b%d, %llu\n", err, b);
31432 +               }
31433 +       }
31434 +
31435 +       if (err >= 0)
31436 +               err = au_wbr_nonopq(dentry, err);
31437 +
31438 +out_parent:
31439 +       dput(parent);
31440 +out:
31441 +       AuDbg("b%d\n", err);
31442 +       return err;
31443 +}
31444 +
31445 +/* ---------------------------------------------------------------------- */
31446 +
31447 +/*
31448 + * - top down parent
31449 + * - most free space with parent
31450 + * - most free space round-robin regardless parent
31451 + */
31452 +static int au_wbr_create_pmfsrr(struct dentry *dentry, unsigned int flags)
31453 +{
31454 +       int err;
31455 +       unsigned long long watermark;
31456 +       struct super_block *sb;
31457 +       struct au_branch *br;
31458 +       struct au_wbr_mfs *mfs;
31459 +
31460 +       err = au_wbr_create_pmfs(dentry, flags | AuWbr_PARENT);
31461 +       if (unlikely(err < 0))
31462 +               goto out;
31463 +
31464 +       sb = dentry->d_sb;
31465 +       br = au_sbr(sb, err);
31466 +       mfs = &au_sbi(sb)->si_wbr_mfs;
31467 +       mutex_lock(&mfs->mfs_lock);
31468 +       watermark = mfs->mfsrr_watermark;
31469 +       mutex_unlock(&mfs->mfs_lock);
31470 +       if (br->br_wbr->wbr_bytes < watermark)
31471 +               /* regardless the parent dir */
31472 +               err = au_wbr_create_mfsrr(dentry, flags);
31473 +
31474 +out:
31475 +       AuDbg("b%d\n", err);
31476 +       return err;
31477 +}
31478 +
31479 +/* ---------------------------------------------------------------------- */
31480 +
31481 +/* policies for copyup */
31482 +
31483 +/* top down parent */
31484 +static int au_wbr_copyup_tdp(struct dentry *dentry)
31485 +{
31486 +       return au_wbr_create_tdp(dentry, /*flags, anything is ok*/0);
31487 +}
31488 +
31489 +/* bottom up parent */
31490 +static int au_wbr_copyup_bup(struct dentry *dentry)
31491 +{
31492 +       int err;
31493 +       aufs_bindex_t bindex, bstart;
31494 +       struct dentry *parent, *h_parent;
31495 +       struct super_block *sb;
31496 +
31497 +       err = -EROFS;
31498 +       sb = dentry->d_sb;
31499 +       parent = dget_parent(dentry);
31500 +       bstart = au_dbstart(parent);
31501 +       for (bindex = au_dbstart(dentry); bindex >= bstart; bindex--) {
31502 +               h_parent = au_h_dptr(parent, bindex);
31503 +               if (!h_parent || !h_parent->d_inode)
31504 +                       continue;
31505 +
31506 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
31507 +                       err = bindex;
31508 +                       break;
31509 +               }
31510 +       }
31511 +       dput(parent);
31512 +
31513 +       /* bottom up here */
31514 +       if (unlikely(err < 0))
31515 +               err = au_wbr_bu(sb, bstart - 1);
31516 +
31517 +       AuDbg("b%d\n", err);
31518 +       return err;
31519 +}
31520 +
31521 +/* bottom up */
31522 +int au_wbr_do_copyup_bu(struct dentry *dentry, aufs_bindex_t bstart)
31523 +{
31524 +       int err;
31525 +
31526 +       err = au_wbr_bu(dentry->d_sb, bstart);
31527 +       AuDbg("b%d\n", err);
31528 +       if (err > bstart)
31529 +               err = au_wbr_nonopq(dentry, err);
31530 +
31531 +       AuDbg("b%d\n", err);
31532 +       return err;
31533 +}
31534 +
31535 +static int au_wbr_copyup_bu(struct dentry *dentry)
31536 +{
31537 +       int err;
31538 +       aufs_bindex_t bstart;
31539 +
31540 +       bstart = au_dbstart(dentry);
31541 +       err = au_wbr_do_copyup_bu(dentry, bstart);
31542 +       return err;
31543 +}
31544 +
31545 +/* ---------------------------------------------------------------------- */
31546 +
31547 +struct au_wbr_copyup_operations au_wbr_copyup_ops[] = {
31548 +       [AuWbrCopyup_TDP] = {
31549 +               .copyup = au_wbr_copyup_tdp
31550 +       },
31551 +       [AuWbrCopyup_BUP] = {
31552 +               .copyup = au_wbr_copyup_bup
31553 +       },
31554 +       [AuWbrCopyup_BU] = {
31555 +               .copyup = au_wbr_copyup_bu
31556 +       }
31557 +};
31558 +
31559 +struct au_wbr_create_operations au_wbr_create_ops[] = {
31560 +       [AuWbrCreate_TDP] = {
31561 +               .create = au_wbr_create_tdp
31562 +       },
31563 +       [AuWbrCreate_RR] = {
31564 +               .create = au_wbr_create_rr,
31565 +               .init   = au_wbr_create_init_rr
31566 +       },
31567 +       [AuWbrCreate_MFS] = {
31568 +               .create = au_wbr_create_mfs,
31569 +               .init   = au_wbr_create_init_mfs,
31570 +               .fin    = au_wbr_create_fin_mfs
31571 +       },
31572 +       [AuWbrCreate_MFSV] = {
31573 +               .create = au_wbr_create_mfs,
31574 +               .init   = au_wbr_create_init_mfs,
31575 +               .fin    = au_wbr_create_fin_mfs
31576 +       },
31577 +       [AuWbrCreate_MFSRR] = {
31578 +               .create = au_wbr_create_mfsrr,
31579 +               .init   = au_wbr_create_init_mfsrr,
31580 +               .fin    = au_wbr_create_fin_mfs
31581 +       },
31582 +       [AuWbrCreate_MFSRRV] = {
31583 +               .create = au_wbr_create_mfsrr,
31584 +               .init   = au_wbr_create_init_mfsrr,
31585 +               .fin    = au_wbr_create_fin_mfs
31586 +       },
31587 +       [AuWbrCreate_PMFS] = {
31588 +               .create = au_wbr_create_pmfs,
31589 +               .init   = au_wbr_create_init_mfs,
31590 +               .fin    = au_wbr_create_fin_mfs
31591 +       },
31592 +       [AuWbrCreate_PMFSV] = {
31593 +               .create = au_wbr_create_pmfs,
31594 +               .init   = au_wbr_create_init_mfs,
31595 +               .fin    = au_wbr_create_fin_mfs
31596 +       },
31597 +       [AuWbrCreate_PMFSRR] = {
31598 +               .create = au_wbr_create_pmfsrr,
31599 +               .init   = au_wbr_create_init_mfsrr,
31600 +               .fin    = au_wbr_create_fin_mfs
31601 +       },
31602 +       [AuWbrCreate_PMFSRRV] = {
31603 +               .create = au_wbr_create_pmfsrr,
31604 +               .init   = au_wbr_create_init_mfsrr,
31605 +               .fin    = au_wbr_create_fin_mfs
31606 +       }
31607 +};
31608 diff -urN /usr/share/empty/fs/aufs/whout.c linux/fs/aufs/whout.c
31609 --- /usr/share/empty/fs/aufs/whout.c    1970-01-01 01:00:00.000000000 +0100
31610 +++ linux/fs/aufs/whout.c       2015-03-27 21:56:35.470128334 +0100
31611 @@ -0,0 +1,1064 @@
31612 +/*
31613 + * Copyright (C) 2005-2015 Junjiro R. Okajima
31614 + *
31615 + * This program, aufs is free software; you can redistribute it and/or modify
31616 + * it under the terms of the GNU General Public License as published by
31617 + * the Free Software Foundation; either version 2 of the License, or
31618 + * (at your option) any later version.
31619 + *
31620 + * This program is distributed in the hope that it will be useful,
31621 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31622 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31623 + * GNU General Public License for more details.
31624 + *
31625 + * You should have received a copy of the GNU General Public License
31626 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31627 + */
31628 +
31629 +/*
31630 + * whiteout for logical deletion and opaque directory
31631 + */
31632 +
31633 +#include "aufs.h"
31634 +
31635 +#define WH_MASK                        S_IRUGO
31636 +
31637 +/*
31638 + * If a directory contains this file, then it is opaque.  We start with the
31639 + * .wh. flag so that it is blocked by lookup.
31640 + */
31641 +static struct qstr diropq_name = QSTR_INIT(AUFS_WH_DIROPQ,
31642 +                                          sizeof(AUFS_WH_DIROPQ) - 1);
31643 +
31644 +/*
31645 + * generate whiteout name, which is NOT terminated by NULL.
31646 + * @name: original d_name.name
31647 + * @len: original d_name.len
31648 + * @wh: whiteout qstr
31649 + * returns zero when succeeds, otherwise error.
31650 + * succeeded value as wh->name should be freed by kfree().
31651 + */
31652 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name)
31653 +{
31654 +       char *p;
31655 +
31656 +       if (unlikely(name->len > PATH_MAX - AUFS_WH_PFX_LEN))
31657 +               return -ENAMETOOLONG;
31658 +
31659 +       wh->len = name->len + AUFS_WH_PFX_LEN;
31660 +       p = kmalloc(wh->len, GFP_NOFS);
31661 +       wh->name = p;
31662 +       if (p) {
31663 +               memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
31664 +               memcpy(p + AUFS_WH_PFX_LEN, name->name, name->len);
31665 +               /* smp_mb(); */
31666 +               return 0;
31667 +       }
31668 +       return -ENOMEM;
31669 +}
31670 +
31671 +/* ---------------------------------------------------------------------- */
31672 +
31673 +/*
31674 + * test if the @wh_name exists under @h_parent.
31675 + * @try_sio specifies the necessary of super-io.
31676 + */
31677 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio)
31678 +{
31679 +       int err;
31680 +       struct dentry *wh_dentry;
31681 +
31682 +       if (!try_sio)
31683 +               wh_dentry = vfsub_lkup_one(wh_name, h_parent);
31684 +       else
31685 +               wh_dentry = au_sio_lkup_one(wh_name, h_parent);
31686 +       err = PTR_ERR(wh_dentry);
31687 +       if (IS_ERR(wh_dentry)) {
31688 +               if (err == -ENAMETOOLONG)
31689 +                       err = 0;
31690 +               goto out;
31691 +       }
31692 +
31693 +       err = 0;
31694 +       if (!wh_dentry->d_inode)
31695 +               goto out_wh; /* success */
31696 +
31697 +       err = 1;
31698 +       if (S_ISREG(wh_dentry->d_inode->i_mode))
31699 +               goto out_wh; /* success */
31700 +
31701 +       err = -EIO;
31702 +       AuIOErr("%pd Invalid whiteout entry type 0%o.\n",
31703 +               wh_dentry, wh_dentry->d_inode->i_mode);
31704 +
31705 +out_wh:
31706 +       dput(wh_dentry);
31707 +out:
31708 +       return err;
31709 +}
31710 +
31711 +/*
31712 + * test if the @h_dentry sets opaque or not.
31713 + */
31714 +int au_diropq_test(struct dentry *h_dentry)
31715 +{
31716 +       int err;
31717 +       struct inode *h_dir;
31718 +
31719 +       h_dir = h_dentry->d_inode;
31720 +       err = au_wh_test(h_dentry, &diropq_name,
31721 +                        au_test_h_perm_sio(h_dir, MAY_EXEC));
31722 +       return err;
31723 +}
31724 +
31725 +/*
31726 + * returns a negative dentry whose name is unique and temporary.
31727 + */
31728 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
31729 +                            struct qstr *prefix)
31730 +{
31731 +       struct dentry *dentry;
31732 +       int i;
31733 +       char defname[NAME_MAX - AUFS_MAX_NAMELEN + DNAME_INLINE_LEN + 1],
31734 +               *name, *p;
31735 +       /* strict atomic_t is unnecessary here */
31736 +       static unsigned short cnt;
31737 +       struct qstr qs;
31738 +
31739 +       BUILD_BUG_ON(sizeof(cnt) * 2 > AUFS_WH_TMP_LEN);
31740 +
31741 +       name = defname;
31742 +       qs.len = sizeof(defname) - DNAME_INLINE_LEN + prefix->len - 1;
31743 +       if (unlikely(prefix->len > DNAME_INLINE_LEN)) {
31744 +               dentry = ERR_PTR(-ENAMETOOLONG);
31745 +               if (unlikely(qs.len > NAME_MAX))
31746 +                       goto out;
31747 +               dentry = ERR_PTR(-ENOMEM);
31748 +               name = kmalloc(qs.len + 1, GFP_NOFS);
31749 +               if (unlikely(!name))
31750 +                       goto out;
31751 +       }
31752 +
31753 +       /* doubly whiteout-ed */
31754 +       memcpy(name, AUFS_WH_PFX AUFS_WH_PFX, AUFS_WH_PFX_LEN * 2);
31755 +       p = name + AUFS_WH_PFX_LEN * 2;
31756 +       memcpy(p, prefix->name, prefix->len);
31757 +       p += prefix->len;
31758 +       *p++ = '.';
31759 +       AuDebugOn(name + qs.len + 1 - p <= AUFS_WH_TMP_LEN);
31760 +
31761 +       qs.name = name;
31762 +       for (i = 0; i < 3; i++) {
31763 +               sprintf(p, "%.*x", AUFS_WH_TMP_LEN, cnt++);
31764 +               dentry = au_sio_lkup_one(&qs, h_parent);
31765 +               if (IS_ERR(dentry) || !dentry->d_inode)
31766 +                       goto out_name;
31767 +               dput(dentry);
31768 +       }
31769 +       /* pr_warn("could not get random name\n"); */
31770 +       dentry = ERR_PTR(-EEXIST);
31771 +       AuDbg("%.*s\n", AuLNPair(&qs));
31772 +       BUG();
31773 +
31774 +out_name:
31775 +       if (name != defname)
31776 +               kfree(name);
31777 +out:
31778 +       AuTraceErrPtr(dentry);
31779 +       return dentry;
31780 +}
31781 +
31782 +/*
31783 + * rename the @h_dentry on @br to the whiteouted temporary name.
31784 + */
31785 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br)
31786 +{
31787 +       int err;
31788 +       struct path h_path = {
31789 +               .mnt = au_br_mnt(br)
31790 +       };
31791 +       struct inode *h_dir, *delegated;
31792 +       struct dentry *h_parent;
31793 +
31794 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
31795 +       h_dir = h_parent->d_inode;
31796 +       IMustLock(h_dir);
31797 +
31798 +       h_path.dentry = au_whtmp_lkup(h_parent, br, &h_dentry->d_name);
31799 +       err = PTR_ERR(h_path.dentry);
31800 +       if (IS_ERR(h_path.dentry))
31801 +               goto out;
31802 +
31803 +       /* under the same dir, no need to lock_rename() */
31804 +       delegated = NULL;
31805 +       err = vfsub_rename(h_dir, h_dentry, h_dir, &h_path, &delegated);
31806 +       AuTraceErr(err);
31807 +       if (unlikely(err == -EWOULDBLOCK)) {
31808 +               pr_warn("cannot retry for NFSv4 delegation"
31809 +                       " for an internal rename\n");
31810 +               iput(delegated);
31811 +       }
31812 +       dput(h_path.dentry);
31813 +
31814 +out:
31815 +       AuTraceErr(err);
31816 +       return err;
31817 +}
31818 +
31819 +/* ---------------------------------------------------------------------- */
31820 +/*
31821 + * functions for removing a whiteout
31822 + */
31823 +
31824 +static int do_unlink_wh(struct inode *h_dir, struct path *h_path)
31825 +{
31826 +       int err, force;
31827 +       struct inode *delegated;
31828 +
31829 +       /*
31830 +        * forces superio when the dir has a sticky bit.
31831 +        * this may be a violation of unix fs semantics.
31832 +        */
31833 +       force = (h_dir->i_mode & S_ISVTX)
31834 +               && !uid_eq(current_fsuid(), h_path->dentry->d_inode->i_uid);
31835 +       delegated = NULL;
31836 +       err = vfsub_unlink(h_dir, h_path, &delegated, force);
31837 +       if (unlikely(err == -EWOULDBLOCK)) {
31838 +               pr_warn("cannot retry for NFSv4 delegation"
31839 +                       " for an internal unlink\n");
31840 +               iput(delegated);
31841 +       }
31842 +       return err;
31843 +}
31844 +
31845 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
31846 +                       struct dentry *dentry)
31847 +{
31848 +       int err;
31849 +
31850 +       err = do_unlink_wh(h_dir, h_path);
31851 +       if (!err && dentry)
31852 +               au_set_dbwh(dentry, -1);
31853 +
31854 +       return err;
31855 +}
31856 +
31857 +static int unlink_wh_name(struct dentry *h_parent, struct qstr *wh,
31858 +                         struct au_branch *br)
31859 +{
31860 +       int err;
31861 +       struct path h_path = {
31862 +               .mnt = au_br_mnt(br)
31863 +       };
31864 +
31865 +       err = 0;
31866 +       h_path.dentry = vfsub_lkup_one(wh, h_parent);
31867 +       if (IS_ERR(h_path.dentry))
31868 +               err = PTR_ERR(h_path.dentry);
31869 +       else {
31870 +               if (h_path.dentry->d_inode
31871 +                   && S_ISREG(h_path.dentry->d_inode->i_mode))
31872 +                       err = do_unlink_wh(h_parent->d_inode, &h_path);
31873 +               dput(h_path.dentry);
31874 +       }
31875 +
31876 +       return err;
31877 +}
31878 +
31879 +/* ---------------------------------------------------------------------- */
31880 +/*
31881 + * initialize/clean whiteout for a branch
31882 + */
31883 +
31884 +static void au_wh_clean(struct inode *h_dir, struct path *whpath,
31885 +                       const int isdir)
31886 +{
31887 +       int err;
31888 +       struct inode *delegated;
31889 +
31890 +       if (!whpath->dentry->d_inode)
31891 +               return;
31892 +
31893 +       if (isdir)
31894 +               err = vfsub_rmdir(h_dir, whpath);
31895 +       else {
31896 +               delegated = NULL;
31897 +               err = vfsub_unlink(h_dir, whpath, &delegated, /*force*/0);
31898 +               if (unlikely(err == -EWOULDBLOCK)) {
31899 +                       pr_warn("cannot retry for NFSv4 delegation"
31900 +                               " for an internal unlink\n");
31901 +                       iput(delegated);
31902 +               }
31903 +       }
31904 +       if (unlikely(err))
31905 +               pr_warn("failed removing %pd (%d), ignored.\n",
31906 +                       whpath->dentry, err);
31907 +}
31908 +
31909 +static int test_linkable(struct dentry *h_root)
31910 +{
31911 +       struct inode *h_dir = h_root->d_inode;
31912 +
31913 +       if (h_dir->i_op->link)
31914 +               return 0;
31915 +
31916 +       pr_err("%pd (%s) doesn't support link(2), use noplink and rw+nolwh\n",
31917 +              h_root, au_sbtype(h_root->d_sb));
31918 +       return -ENOSYS;
31919 +}
31920 +
31921 +/* todo: should this mkdir be done in /sbin/mount.aufs helper? */
31922 +static int au_whdir(struct inode *h_dir, struct path *path)
31923 +{
31924 +       int err;
31925 +
31926 +       err = -EEXIST;
31927 +       if (!path->dentry->d_inode) {
31928 +               int mode = S_IRWXU;
31929 +
31930 +               if (au_test_nfs(path->dentry->d_sb))
31931 +                       mode |= S_IXUGO;
31932 +               err = vfsub_mkdir(h_dir, path, mode);
31933 +       } else if (d_is_dir(path->dentry))
31934 +               err = 0;
31935 +       else
31936 +               pr_err("unknown %pd exists\n", path->dentry);
31937 +
31938 +       return err;
31939 +}
31940 +
31941 +struct au_wh_base {
31942 +       const struct qstr *name;
31943 +       struct dentry *dentry;
31944 +};
31945 +
31946 +static void au_wh_init_ro(struct inode *h_dir, struct au_wh_base base[],
31947 +                         struct path *h_path)
31948 +{
31949 +       h_path->dentry = base[AuBrWh_BASE].dentry;
31950 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
31951 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
31952 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
31953 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
31954 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
31955 +}
31956 +
31957 +/*
31958 + * returns tri-state,
31959 + * minus: error, caller should print the message
31960 + * zero: succuess
31961 + * plus: error, caller should NOT print the message
31962 + */
31963 +static int au_wh_init_rw_nolink(struct dentry *h_root, struct au_wbr *wbr,
31964 +                               int do_plink, struct au_wh_base base[],
31965 +                               struct path *h_path)
31966 +{
31967 +       int err;
31968 +       struct inode *h_dir;
31969 +
31970 +       h_dir = h_root->d_inode;
31971 +       h_path->dentry = base[AuBrWh_BASE].dentry;
31972 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
31973 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
31974 +       if (do_plink) {
31975 +               err = test_linkable(h_root);
31976 +               if (unlikely(err)) {
31977 +                       err = 1;
31978 +                       goto out;
31979 +               }
31980 +
31981 +               err = au_whdir(h_dir, h_path);
31982 +               if (unlikely(err))
31983 +                       goto out;
31984 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
31985 +       } else
31986 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
31987 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
31988 +       err = au_whdir(h_dir, h_path);
31989 +       if (unlikely(err))
31990 +               goto out;
31991 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
31992 +
31993 +out:
31994 +       return err;
31995 +}
31996 +
31997 +/*
31998 + * for the moment, aufs supports the branch filesystem which does not support
31999 + * link(2). testing on FAT which does not support i_op->setattr() fully either,
32000 + * copyup failed. finally, such filesystem will not be used as the writable
32001 + * branch.
32002 + *
32003 + * returns tri-state, see above.
32004 + */
32005 +static int au_wh_init_rw(struct dentry *h_root, struct au_wbr *wbr,
32006 +                        int do_plink, struct au_wh_base base[],
32007 +                        struct path *h_path)
32008 +{
32009 +       int err;
32010 +       struct inode *h_dir;
32011 +
32012 +       WbrWhMustWriteLock(wbr);
32013 +
32014 +       err = test_linkable(h_root);
32015 +       if (unlikely(err)) {
32016 +               err = 1;
32017 +               goto out;
32018 +       }
32019 +
32020 +       /*
32021 +        * todo: should this create be done in /sbin/mount.aufs helper?
32022 +        */
32023 +       err = -EEXIST;
32024 +       h_dir = h_root->d_inode;
32025 +       if (!base[AuBrWh_BASE].dentry->d_inode) {
32026 +               h_path->dentry = base[AuBrWh_BASE].dentry;
32027 +               err = vfsub_create(h_dir, h_path, WH_MASK, /*want_excl*/true);
32028 +       } else if (S_ISREG(base[AuBrWh_BASE].dentry->d_inode->i_mode))
32029 +               err = 0;
32030 +       else
32031 +               pr_err("unknown %pd2 exists\n", base[AuBrWh_BASE].dentry);
32032 +       if (unlikely(err))
32033 +               goto out;
32034 +
32035 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
32036 +       if (do_plink) {
32037 +               err = au_whdir(h_dir, h_path);
32038 +               if (unlikely(err))
32039 +                       goto out;
32040 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
32041 +       } else
32042 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
32043 +       wbr->wbr_whbase = dget(base[AuBrWh_BASE].dentry);
32044 +
32045 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
32046 +       err = au_whdir(h_dir, h_path);
32047 +       if (unlikely(err))
32048 +               goto out;
32049 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
32050 +
32051 +out:
32052 +       return err;
32053 +}
32054 +
32055 +/*
32056 + * initialize the whiteout base file/dir for @br.
32057 + */
32058 +int au_wh_init(struct au_branch *br, struct super_block *sb)
32059 +{
32060 +       int err, i;
32061 +       const unsigned char do_plink
32062 +               = !!au_opt_test(au_mntflags(sb), PLINK);
32063 +       struct inode *h_dir;
32064 +       struct path path = br->br_path;
32065 +       struct dentry *h_root = path.dentry;
32066 +       struct au_wbr *wbr = br->br_wbr;
32067 +       static const struct qstr base_name[] = {
32068 +               [AuBrWh_BASE] = QSTR_INIT(AUFS_BASE_NAME,
32069 +                                         sizeof(AUFS_BASE_NAME) - 1),
32070 +               [AuBrWh_PLINK] = QSTR_INIT(AUFS_PLINKDIR_NAME,
32071 +                                          sizeof(AUFS_PLINKDIR_NAME) - 1),
32072 +               [AuBrWh_ORPH] = QSTR_INIT(AUFS_ORPHDIR_NAME,
32073 +                                         sizeof(AUFS_ORPHDIR_NAME) - 1)
32074 +       };
32075 +       struct au_wh_base base[] = {
32076 +               [AuBrWh_BASE] = {
32077 +                       .name   = base_name + AuBrWh_BASE,
32078 +                       .dentry = NULL
32079 +               },
32080 +               [AuBrWh_PLINK] = {
32081 +                       .name   = base_name + AuBrWh_PLINK,
32082 +                       .dentry = NULL
32083 +               },
32084 +               [AuBrWh_ORPH] = {
32085 +                       .name   = base_name + AuBrWh_ORPH,
32086 +                       .dentry = NULL
32087 +               }
32088 +       };
32089 +
32090 +       if (wbr)
32091 +               WbrWhMustWriteLock(wbr);
32092 +
32093 +       for (i = 0; i < AuBrWh_Last; i++) {
32094 +               /* doubly whiteouted */
32095 +               struct dentry *d;
32096 +
32097 +               d = au_wh_lkup(h_root, (void *)base[i].name, br);
32098 +               err = PTR_ERR(d);
32099 +               if (IS_ERR(d))
32100 +                       goto out;
32101 +
32102 +               base[i].dentry = d;
32103 +               AuDebugOn(wbr
32104 +                         && wbr->wbr_wh[i]
32105 +                         && wbr->wbr_wh[i] != base[i].dentry);
32106 +       }
32107 +
32108 +       if (wbr)
32109 +               for (i = 0; i < AuBrWh_Last; i++) {
32110 +                       dput(wbr->wbr_wh[i]);
32111 +                       wbr->wbr_wh[i] = NULL;
32112 +               }
32113 +
32114 +       err = 0;
32115 +       if (!au_br_writable(br->br_perm)) {
32116 +               h_dir = h_root->d_inode;
32117 +               au_wh_init_ro(h_dir, base, &path);
32118 +       } else if (!au_br_wh_linkable(br->br_perm)) {
32119 +               err = au_wh_init_rw_nolink(h_root, wbr, do_plink, base, &path);
32120 +               if (err > 0)
32121 +                       goto out;
32122 +               else if (err)
32123 +                       goto out_err;
32124 +       } else {
32125 +               err = au_wh_init_rw(h_root, wbr, do_plink, base, &path);
32126 +               if (err > 0)
32127 +                       goto out;
32128 +               else if (err)
32129 +                       goto out_err;
32130 +       }
32131 +       goto out; /* success */
32132 +
32133 +out_err:
32134 +       pr_err("an error(%d) on the writable branch %pd(%s)\n",
32135 +              err, h_root, au_sbtype(h_root->d_sb));
32136 +out:
32137 +       for (i = 0; i < AuBrWh_Last; i++)
32138 +               dput(base[i].dentry);
32139 +       return err;
32140 +}
32141 +
32142 +/* ---------------------------------------------------------------------- */
32143 +/*
32144 + * whiteouts are all hard-linked usually.
32145 + * when its link count reaches a ceiling, we create a new whiteout base
32146 + * asynchronously.
32147 + */
32148 +
32149 +struct reinit_br_wh {
32150 +       struct super_block *sb;
32151 +       struct au_branch *br;
32152 +};
32153 +
32154 +static void reinit_br_wh(void *arg)
32155 +{
32156 +       int err;
32157 +       aufs_bindex_t bindex;
32158 +       struct path h_path;
32159 +       struct reinit_br_wh *a = arg;
32160 +       struct au_wbr *wbr;
32161 +       struct inode *dir, *delegated;
32162 +       struct dentry *h_root;
32163 +       struct au_hinode *hdir;
32164 +
32165 +       err = 0;
32166 +       wbr = a->br->br_wbr;
32167 +       /* big aufs lock */
32168 +       si_noflush_write_lock(a->sb);
32169 +       if (!au_br_writable(a->br->br_perm))
32170 +               goto out;
32171 +       bindex = au_br_index(a->sb, a->br->br_id);
32172 +       if (unlikely(bindex < 0))
32173 +               goto out;
32174 +
32175 +       di_read_lock_parent(a->sb->s_root, AuLock_IR);
32176 +       dir = a->sb->s_root->d_inode;
32177 +       hdir = au_hi(dir, bindex);
32178 +       h_root = au_h_dptr(a->sb->s_root, bindex);
32179 +       AuDebugOn(h_root != au_br_dentry(a->br));
32180 +
32181 +       au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
32182 +       wbr_wh_write_lock(wbr);
32183 +       err = au_h_verify(wbr->wbr_whbase, au_opt_udba(a->sb), hdir->hi_inode,
32184 +                         h_root, a->br);
32185 +       if (!err) {
32186 +               h_path.dentry = wbr->wbr_whbase;
32187 +               h_path.mnt = au_br_mnt(a->br);
32188 +               delegated = NULL;
32189 +               err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated,
32190 +                                  /*force*/0);
32191 +               if (unlikely(err == -EWOULDBLOCK)) {
32192 +                       pr_warn("cannot retry for NFSv4 delegation"
32193 +                               " for an internal unlink\n");
32194 +                       iput(delegated);
32195 +               }
32196 +       } else {
32197 +               pr_warn("%pd is moved, ignored\n", wbr->wbr_whbase);
32198 +               err = 0;
32199 +       }
32200 +       dput(wbr->wbr_whbase);
32201 +       wbr->wbr_whbase = NULL;
32202 +       if (!err)
32203 +               err = au_wh_init(a->br, a->sb);
32204 +       wbr_wh_write_unlock(wbr);
32205 +       au_hn_imtx_unlock(hdir);
32206 +       di_read_unlock(a->sb->s_root, AuLock_IR);
32207 +       if (!err)
32208 +               au_fhsm_wrote(a->sb, bindex, /*force*/0);
32209 +
32210 +out:
32211 +       if (wbr)
32212 +               atomic_dec(&wbr->wbr_wh_running);
32213 +       atomic_dec(&a->br->br_count);
32214 +       si_write_unlock(a->sb);
32215 +       au_nwt_done(&au_sbi(a->sb)->si_nowait);
32216 +       kfree(arg);
32217 +       if (unlikely(err))
32218 +               AuIOErr("err %d\n", err);
32219 +}
32220 +
32221 +static void kick_reinit_br_wh(struct super_block *sb, struct au_branch *br)
32222 +{
32223 +       int do_dec, wkq_err;
32224 +       struct reinit_br_wh *arg;
32225 +
32226 +       do_dec = 1;
32227 +       if (atomic_inc_return(&br->br_wbr->wbr_wh_running) != 1)
32228 +               goto out;
32229 +
32230 +       /* ignore ENOMEM */
32231 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
32232 +       if (arg) {
32233 +               /*
32234 +                * dec(wh_running), kfree(arg) and dec(br_count)
32235 +                * in reinit function
32236 +                */
32237 +               arg->sb = sb;
32238 +               arg->br = br;
32239 +               atomic_inc(&br->br_count);
32240 +               wkq_err = au_wkq_nowait(reinit_br_wh, arg, sb, /*flags*/0);
32241 +               if (unlikely(wkq_err)) {
32242 +                       atomic_dec(&br->br_wbr->wbr_wh_running);
32243 +                       atomic_dec(&br->br_count);
32244 +                       kfree(arg);
32245 +               }
32246 +               do_dec = 0;
32247 +       }
32248 +
32249 +out:
32250 +       if (do_dec)
32251 +               atomic_dec(&br->br_wbr->wbr_wh_running);
32252 +}
32253 +
32254 +/* ---------------------------------------------------------------------- */
32255 +
32256 +/*
32257 + * create the whiteout @wh.
32258 + */
32259 +static int link_or_create_wh(struct super_block *sb, aufs_bindex_t bindex,
32260 +                            struct dentry *wh)
32261 +{
32262 +       int err;
32263 +       struct path h_path = {
32264 +               .dentry = wh
32265 +       };
32266 +       struct au_branch *br;
32267 +       struct au_wbr *wbr;
32268 +       struct dentry *h_parent;
32269 +       struct inode *h_dir, *delegated;
32270 +
32271 +       h_parent = wh->d_parent; /* dir inode is locked */
32272 +       h_dir = h_parent->d_inode;
32273 +       IMustLock(h_dir);
32274 +
32275 +       br = au_sbr(sb, bindex);
32276 +       h_path.mnt = au_br_mnt(br);
32277 +       wbr = br->br_wbr;
32278 +       wbr_wh_read_lock(wbr);
32279 +       if (wbr->wbr_whbase) {
32280 +               delegated = NULL;
32281 +               err = vfsub_link(wbr->wbr_whbase, h_dir, &h_path, &delegated);
32282 +               if (unlikely(err == -EWOULDBLOCK)) {
32283 +                       pr_warn("cannot retry for NFSv4 delegation"
32284 +                               " for an internal link\n");
32285 +                       iput(delegated);
32286 +               }
32287 +               if (!err || err != -EMLINK)
32288 +                       goto out;
32289 +
32290 +               /* link count full. re-initialize br_whbase. */
32291 +               kick_reinit_br_wh(sb, br);
32292 +       }
32293 +
32294 +       /* return this error in this context */
32295 +       err = vfsub_create(h_dir, &h_path, WH_MASK, /*want_excl*/true);
32296 +       if (!err)
32297 +               au_fhsm_wrote(sb, bindex, /*force*/0);
32298 +
32299 +out:
32300 +       wbr_wh_read_unlock(wbr);
32301 +       return err;
32302 +}
32303 +
32304 +/* ---------------------------------------------------------------------- */
32305 +
32306 +/*
32307 + * create or remove the diropq.
32308 + */
32309 +static struct dentry *do_diropq(struct dentry *dentry, aufs_bindex_t bindex,
32310 +                               unsigned int flags)
32311 +{
32312 +       struct dentry *opq_dentry, *h_dentry;
32313 +       struct super_block *sb;
32314 +       struct au_branch *br;
32315 +       int err;
32316 +
32317 +       sb = dentry->d_sb;
32318 +       br = au_sbr(sb, bindex);
32319 +       h_dentry = au_h_dptr(dentry, bindex);
32320 +       opq_dentry = vfsub_lkup_one(&diropq_name, h_dentry);
32321 +       if (IS_ERR(opq_dentry))
32322 +               goto out;
32323 +
32324 +       if (au_ftest_diropq(flags, CREATE)) {
32325 +               err = link_or_create_wh(sb, bindex, opq_dentry);
32326 +               if (!err) {
32327 +                       au_set_dbdiropq(dentry, bindex);
32328 +                       goto out; /* success */
32329 +               }
32330 +       } else {
32331 +               struct path tmp = {
32332 +                       .dentry = opq_dentry,
32333 +                       .mnt    = au_br_mnt(br)
32334 +               };
32335 +               err = do_unlink_wh(au_h_iptr(dentry->d_inode, bindex), &tmp);
32336 +               if (!err)
32337 +                       au_set_dbdiropq(dentry, -1);
32338 +       }
32339 +       dput(opq_dentry);
32340 +       opq_dentry = ERR_PTR(err);
32341 +
32342 +out:
32343 +       return opq_dentry;
32344 +}
32345 +
32346 +struct do_diropq_args {
32347 +       struct dentry **errp;
32348 +       struct dentry *dentry;
32349 +       aufs_bindex_t bindex;
32350 +       unsigned int flags;
32351 +};
32352 +
32353 +static void call_do_diropq(void *args)
32354 +{
32355 +       struct do_diropq_args *a = args;
32356 +       *a->errp = do_diropq(a->dentry, a->bindex, a->flags);
32357 +}
32358 +
32359 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
32360 +                            unsigned int flags)
32361 +{
32362 +       struct dentry *diropq, *h_dentry;
32363 +
32364 +       h_dentry = au_h_dptr(dentry, bindex);
32365 +       if (!au_test_h_perm_sio(h_dentry->d_inode, MAY_EXEC | MAY_WRITE))
32366 +               diropq = do_diropq(dentry, bindex, flags);
32367 +       else {
32368 +               int wkq_err;
32369 +               struct do_diropq_args args = {
32370 +                       .errp           = &diropq,
32371 +                       .dentry         = dentry,
32372 +                       .bindex         = bindex,
32373 +                       .flags          = flags
32374 +               };
32375 +
32376 +               wkq_err = au_wkq_wait(call_do_diropq, &args);
32377 +               if (unlikely(wkq_err))
32378 +                       diropq = ERR_PTR(wkq_err);
32379 +       }
32380 +
32381 +       return diropq;
32382 +}
32383 +
32384 +/* ---------------------------------------------------------------------- */
32385 +
32386 +/*
32387 + * lookup whiteout dentry.
32388 + * @h_parent: lower parent dentry which must exist and be locked
32389 + * @base_name: name of dentry which will be whiteouted
32390 + * returns dentry for whiteout.
32391 + */
32392 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
32393 +                         struct au_branch *br)
32394 +{
32395 +       int err;
32396 +       struct qstr wh_name;
32397 +       struct dentry *wh_dentry;
32398 +
32399 +       err = au_wh_name_alloc(&wh_name, base_name);
32400 +       wh_dentry = ERR_PTR(err);
32401 +       if (!err) {
32402 +               wh_dentry = vfsub_lkup_one(&wh_name, h_parent);
32403 +               kfree(wh_name.name);
32404 +       }
32405 +       return wh_dentry;
32406 +}
32407 +
32408 +/*
32409 + * link/create a whiteout for @dentry on @bindex.
32410 + */
32411 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
32412 +                           struct dentry *h_parent)
32413 +{
32414 +       struct dentry *wh_dentry;
32415 +       struct super_block *sb;
32416 +       int err;
32417 +
32418 +       sb = dentry->d_sb;
32419 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, au_sbr(sb, bindex));
32420 +       if (!IS_ERR(wh_dentry) && !wh_dentry->d_inode) {
32421 +               err = link_or_create_wh(sb, bindex, wh_dentry);
32422 +               if (!err) {
32423 +                       au_set_dbwh(dentry, bindex);
32424 +                       au_fhsm_wrote(sb, bindex, /*force*/0);
32425 +               } else {
32426 +                       dput(wh_dentry);
32427 +                       wh_dentry = ERR_PTR(err);
32428 +               }
32429 +       }
32430 +
32431 +       return wh_dentry;
32432 +}
32433 +
32434 +/* ---------------------------------------------------------------------- */
32435 +
32436 +/* Delete all whiteouts in this directory on branch bindex. */
32437 +static int del_wh_children(struct dentry *h_dentry, struct au_nhash *whlist,
32438 +                          aufs_bindex_t bindex, struct au_branch *br)
32439 +{
32440 +       int err;
32441 +       unsigned long ul, n;
32442 +       struct qstr wh_name;
32443 +       char *p;
32444 +       struct hlist_head *head;
32445 +       struct au_vdir_wh *pos;
32446 +       struct au_vdir_destr *str;
32447 +
32448 +       err = -ENOMEM;
32449 +       p = (void *)__get_free_page(GFP_NOFS);
32450 +       wh_name.name = p;
32451 +       if (unlikely(!wh_name.name))
32452 +               goto out;
32453 +
32454 +       err = 0;
32455 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
32456 +       p += AUFS_WH_PFX_LEN;
32457 +       n = whlist->nh_num;
32458 +       head = whlist->nh_head;
32459 +       for (ul = 0; !err && ul < n; ul++, head++) {
32460 +               hlist_for_each_entry(pos, head, wh_hash) {
32461 +                       if (pos->wh_bindex != bindex)
32462 +                               continue;
32463 +
32464 +                       str = &pos->wh_str;
32465 +                       if (str->len + AUFS_WH_PFX_LEN <= PATH_MAX) {
32466 +                               memcpy(p, str->name, str->len);
32467 +                               wh_name.len = AUFS_WH_PFX_LEN + str->len;
32468 +                               err = unlink_wh_name(h_dentry, &wh_name, br);
32469 +                               if (!err)
32470 +                                       continue;
32471 +                               break;
32472 +                       }
32473 +                       AuIOErr("whiteout name too long %.*s\n",
32474 +                               str->len, str->name);
32475 +                       err = -EIO;
32476 +                       break;
32477 +               }
32478 +       }
32479 +       free_page((unsigned long)wh_name.name);
32480 +
32481 +out:
32482 +       return err;
32483 +}
32484 +
32485 +struct del_wh_children_args {
32486 +       int *errp;
32487 +       struct dentry *h_dentry;
32488 +       struct au_nhash *whlist;
32489 +       aufs_bindex_t bindex;
32490 +       struct au_branch *br;
32491 +};
32492 +
32493 +static void call_del_wh_children(void *args)
32494 +{
32495 +       struct del_wh_children_args *a = args;
32496 +       *a->errp = del_wh_children(a->h_dentry, a->whlist, a->bindex, a->br);
32497 +}
32498 +
32499 +/* ---------------------------------------------------------------------- */
32500 +
32501 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp)
32502 +{
32503 +       struct au_whtmp_rmdir *whtmp;
32504 +       int err;
32505 +       unsigned int rdhash;
32506 +
32507 +       SiMustAnyLock(sb);
32508 +
32509 +       whtmp = kmalloc(sizeof(*whtmp), gfp);
32510 +       if (unlikely(!whtmp)) {
32511 +               whtmp = ERR_PTR(-ENOMEM);
32512 +               goto out;
32513 +       }
32514 +
32515 +       whtmp->dir = NULL;
32516 +       whtmp->br = NULL;
32517 +       whtmp->wh_dentry = NULL;
32518 +       /* no estimation for dir size */
32519 +       rdhash = au_sbi(sb)->si_rdhash;
32520 +       if (!rdhash)
32521 +               rdhash = AUFS_RDHASH_DEF;
32522 +       err = au_nhash_alloc(&whtmp->whlist, rdhash, gfp);
32523 +       if (unlikely(err)) {
32524 +               kfree(whtmp);
32525 +               whtmp = ERR_PTR(err);
32526 +       }
32527 +
32528 +out:
32529 +       return whtmp;
32530 +}
32531 +
32532 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp)
32533 +{
32534 +       if (whtmp->br)
32535 +               atomic_dec(&whtmp->br->br_count);
32536 +       dput(whtmp->wh_dentry);
32537 +       iput(whtmp->dir);
32538 +       au_nhash_wh_free(&whtmp->whlist);
32539 +       kfree(whtmp);
32540 +}
32541 +
32542 +/*
32543 + * rmdir the whiteouted temporary named dir @h_dentry.
32544 + * @whlist: whiteouted children.
32545 + */
32546 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
32547 +                  struct dentry *wh_dentry, struct au_nhash *whlist)
32548 +{
32549 +       int err;
32550 +       unsigned int h_nlink;
32551 +       struct path h_tmp;
32552 +       struct inode *wh_inode, *h_dir;
32553 +       struct au_branch *br;
32554 +
32555 +       h_dir = wh_dentry->d_parent->d_inode; /* dir inode is locked */
32556 +       IMustLock(h_dir);
32557 +
32558 +       br = au_sbr(dir->i_sb, bindex);
32559 +       wh_inode = wh_dentry->d_inode;
32560 +       mutex_lock_nested(&wh_inode->i_mutex, AuLsc_I_CHILD);
32561 +
32562 +       /*
32563 +        * someone else might change some whiteouts while we were sleeping.
32564 +        * it means this whlist may have an obsoleted entry.
32565 +        */
32566 +       if (!au_test_h_perm_sio(wh_inode, MAY_EXEC | MAY_WRITE))
32567 +               err = del_wh_children(wh_dentry, whlist, bindex, br);
32568 +       else {
32569 +               int wkq_err;
32570 +               struct del_wh_children_args args = {
32571 +                       .errp           = &err,
32572 +                       .h_dentry       = wh_dentry,
32573 +                       .whlist         = whlist,
32574 +                       .bindex         = bindex,
32575 +                       .br             = br
32576 +               };
32577 +
32578 +               wkq_err = au_wkq_wait(call_del_wh_children, &args);
32579 +               if (unlikely(wkq_err))
32580 +                       err = wkq_err;
32581 +       }
32582 +       mutex_unlock(&wh_inode->i_mutex);
32583 +
32584 +       if (!err) {
32585 +               h_tmp.dentry = wh_dentry;
32586 +               h_tmp.mnt = au_br_mnt(br);
32587 +               h_nlink = h_dir->i_nlink;
32588 +               err = vfsub_rmdir(h_dir, &h_tmp);
32589 +               /* some fs doesn't change the parent nlink in some cases */
32590 +               h_nlink -= h_dir->i_nlink;
32591 +       }
32592 +
32593 +       if (!err) {
32594 +               if (au_ibstart(dir) == bindex) {
32595 +                       /* todo: dir->i_mutex is necessary */
32596 +                       au_cpup_attr_timesizes(dir);
32597 +                       if (h_nlink)
32598 +                               vfsub_drop_nlink(dir);
32599 +               }
32600 +               return 0; /* success */
32601 +       }
32602 +
32603 +       pr_warn("failed removing %pd(%d), ignored\n", wh_dentry, err);
32604 +       return err;
32605 +}
32606 +
32607 +static void call_rmdir_whtmp(void *args)
32608 +{
32609 +       int err;
32610 +       aufs_bindex_t bindex;
32611 +       struct au_whtmp_rmdir *a = args;
32612 +       struct super_block *sb;
32613 +       struct dentry *h_parent;
32614 +       struct inode *h_dir;
32615 +       struct au_hinode *hdir;
32616 +
32617 +       /* rmdir by nfsd may cause deadlock with this i_mutex */
32618 +       /* mutex_lock(&a->dir->i_mutex); */
32619 +       err = -EROFS;
32620 +       sb = a->dir->i_sb;
32621 +       si_read_lock(sb, !AuLock_FLUSH);
32622 +       if (!au_br_writable(a->br->br_perm))
32623 +               goto out;
32624 +       bindex = au_br_index(sb, a->br->br_id);
32625 +       if (unlikely(bindex < 0))
32626 +               goto out;
32627 +
32628 +       err = -EIO;
32629 +       ii_write_lock_parent(a->dir);
32630 +       h_parent = dget_parent(a->wh_dentry);
32631 +       h_dir = h_parent->d_inode;
32632 +       hdir = au_hi(a->dir, bindex);
32633 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
32634 +       if (unlikely(err))
32635 +               goto out_mnt;
32636 +       au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
32637 +       err = au_h_verify(a->wh_dentry, au_opt_udba(sb), h_dir, h_parent,
32638 +                         a->br);
32639 +       if (!err)
32640 +               err = au_whtmp_rmdir(a->dir, bindex, a->wh_dentry, &a->whlist);
32641 +       au_hn_imtx_unlock(hdir);
32642 +       vfsub_mnt_drop_write(au_br_mnt(a->br));
32643 +
32644 +out_mnt:
32645 +       dput(h_parent);
32646 +       ii_write_unlock(a->dir);
32647 +out:
32648 +       /* mutex_unlock(&a->dir->i_mutex); */
32649 +       au_whtmp_rmdir_free(a);
32650 +       si_read_unlock(sb);
32651 +       au_nwt_done(&au_sbi(sb)->si_nowait);
32652 +       if (unlikely(err))
32653 +               AuIOErr("err %d\n", err);
32654 +}
32655 +
32656 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
32657 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args)
32658 +{
32659 +       int wkq_err;
32660 +       struct super_block *sb;
32661 +
32662 +       IMustLock(dir);
32663 +
32664 +       /* all post-process will be done in do_rmdir_whtmp(). */
32665 +       sb = dir->i_sb;
32666 +       args->dir = au_igrab(dir);
32667 +       args->br = au_sbr(sb, bindex);
32668 +       atomic_inc(&args->br->br_count);
32669 +       args->wh_dentry = dget(wh_dentry);
32670 +       wkq_err = au_wkq_nowait(call_rmdir_whtmp, args, sb, /*flags*/0);
32671 +       if (unlikely(wkq_err)) {
32672 +               pr_warn("rmdir error %pd (%d), ignored\n", wh_dentry, wkq_err);
32673 +               au_whtmp_rmdir_free(args);
32674 +       }
32675 +}
32676 diff -urN /usr/share/empty/fs/aufs/whout.h linux/fs/aufs/whout.h
32677 --- /usr/share/empty/fs/aufs/whout.h    1970-01-01 01:00:00.000000000 +0100
32678 +++ linux/fs/aufs/whout.h       2015-03-27 21:56:35.470128334 +0100
32679 @@ -0,0 +1,85 @@
32680 +/*
32681 + * Copyright (C) 2005-2015 Junjiro R. Okajima
32682 + *
32683 + * This program, aufs is free software; you can redistribute it and/or modify
32684 + * it under the terms of the GNU General Public License as published by
32685 + * the Free Software Foundation; either version 2 of the License, or
32686 + * (at your option) any later version.
32687 + *
32688 + * This program is distributed in the hope that it will be useful,
32689 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32690 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32691 + * GNU General Public License for more details.
32692 + *
32693 + * You should have received a copy of the GNU General Public License
32694 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32695 + */
32696 +
32697 +/*
32698 + * whiteout for logical deletion and opaque directory
32699 + */
32700 +
32701 +#ifndef __AUFS_WHOUT_H__
32702 +#define __AUFS_WHOUT_H__
32703 +
32704 +#ifdef __KERNEL__
32705 +
32706 +#include "dir.h"
32707 +
32708 +/* whout.c */
32709 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name);
32710 +struct au_branch;
32711 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name, int try_sio);
32712 +int au_diropq_test(struct dentry *h_dentry);
32713 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
32714 +                            struct qstr *prefix);
32715 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br);
32716 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
32717 +                       struct dentry *dentry);
32718 +int au_wh_init(struct au_branch *br, struct super_block *sb);
32719 +
32720 +/* diropq flags */
32721 +#define AuDiropq_CREATE        1
32722 +#define au_ftest_diropq(flags, name)   ((flags) & AuDiropq_##name)
32723 +#define au_fset_diropq(flags, name) \
32724 +       do { (flags) |= AuDiropq_##name; } while (0)
32725 +#define au_fclr_diropq(flags, name) \
32726 +       do { (flags) &= ~AuDiropq_##name; } while (0)
32727 +
32728 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
32729 +                            unsigned int flags);
32730 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
32731 +                         struct au_branch *br);
32732 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
32733 +                           struct dentry *h_parent);
32734 +
32735 +/* real rmdir for the whiteout-ed dir */
32736 +struct au_whtmp_rmdir {
32737 +       struct inode *dir;
32738 +       struct au_branch *br;
32739 +       struct dentry *wh_dentry;
32740 +       struct au_nhash whlist;
32741 +};
32742 +
32743 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp);
32744 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp);
32745 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
32746 +                  struct dentry *wh_dentry, struct au_nhash *whlist);
32747 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
32748 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args);
32749 +
32750 +/* ---------------------------------------------------------------------- */
32751 +
32752 +static inline struct dentry *au_diropq_create(struct dentry *dentry,
32753 +                                             aufs_bindex_t bindex)
32754 +{
32755 +       return au_diropq_sio(dentry, bindex, AuDiropq_CREATE);
32756 +}
32757 +
32758 +static inline int au_diropq_remove(struct dentry *dentry, aufs_bindex_t bindex)
32759 +{
32760 +       return PTR_ERR(au_diropq_sio(dentry, bindex, !AuDiropq_CREATE));
32761 +}
32762 +
32763 +#endif /* __KERNEL__ */
32764 +#endif /* __AUFS_WHOUT_H__ */
32765 diff -urN /usr/share/empty/fs/aufs/wkq.c linux/fs/aufs/wkq.c
32766 --- /usr/share/empty/fs/aufs/wkq.c      1970-01-01 01:00:00.000000000 +0100
32767 +++ linux/fs/aufs/wkq.c 2015-03-27 21:56:35.470128334 +0100
32768 @@ -0,0 +1,213 @@
32769 +/*
32770 + * Copyright (C) 2005-2015 Junjiro R. Okajima
32771 + *
32772 + * This program, aufs is free software; you can redistribute it and/or modify
32773 + * it under the terms of the GNU General Public License as published by
32774 + * the Free Software Foundation; either version 2 of the License, or
32775 + * (at your option) any later version.
32776 + *
32777 + * This program is distributed in the hope that it will be useful,
32778 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32779 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32780 + * GNU General Public License for more details.
32781 + *
32782 + * You should have received a copy of the GNU General Public License
32783 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32784 + */
32785 +
32786 +/*
32787 + * workqueue for asynchronous/super-io operations
32788 + * todo: try new dredential scheme
32789 + */
32790 +
32791 +#include <linux/module.h>
32792 +#include "aufs.h"
32793 +
32794 +/* internal workqueue named AUFS_WKQ_NAME */
32795 +
32796 +static struct workqueue_struct *au_wkq;
32797 +
32798 +struct au_wkinfo {
32799 +       struct work_struct wk;
32800 +       struct kobject *kobj;
32801 +
32802 +       unsigned int flags; /* see wkq.h */
32803 +
32804 +       au_wkq_func_t func;
32805 +       void *args;
32806 +
32807 +       struct completion *comp;
32808 +};
32809 +
32810 +/* ---------------------------------------------------------------------- */
32811 +
32812 +static void wkq_func(struct work_struct *wk)
32813 +{
32814 +       struct au_wkinfo *wkinfo = container_of(wk, struct au_wkinfo, wk);
32815 +
32816 +       AuDebugOn(!uid_eq(current_fsuid(), GLOBAL_ROOT_UID));
32817 +       AuDebugOn(rlimit(RLIMIT_FSIZE) != RLIM_INFINITY);
32818 +
32819 +       wkinfo->func(wkinfo->args);
32820 +       if (au_ftest_wkq(wkinfo->flags, WAIT))
32821 +               complete(wkinfo->comp);
32822 +       else {
32823 +               kobject_put(wkinfo->kobj);
32824 +               module_put(THIS_MODULE); /* todo: ?? */
32825 +               kfree(wkinfo);
32826 +       }
32827 +}
32828 +
32829 +/*
32830 + * Since struct completion is large, try allocating it dynamically.
32831 + */
32832 +#if 1 /* defined(CONFIG_4KSTACKS) || defined(AuTest4KSTACKS) */
32833 +#define AuWkqCompDeclare(name) struct completion *comp = NULL
32834 +
32835 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
32836 +{
32837 +       *comp = kmalloc(sizeof(**comp), GFP_NOFS);
32838 +       if (*comp) {
32839 +               init_completion(*comp);
32840 +               wkinfo->comp = *comp;
32841 +               return 0;
32842 +       }
32843 +       return -ENOMEM;
32844 +}
32845 +
32846 +static void au_wkq_comp_free(struct completion *comp)
32847 +{
32848 +       kfree(comp);
32849 +}
32850 +
32851 +#else
32852 +
32853 +/* no braces */
32854 +#define AuWkqCompDeclare(name) \
32855 +       DECLARE_COMPLETION_ONSTACK(_ ## name); \
32856 +       struct completion *comp = &_ ## name
32857 +
32858 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
32859 +{
32860 +       wkinfo->comp = *comp;
32861 +       return 0;
32862 +}
32863 +
32864 +static void au_wkq_comp_free(struct completion *comp __maybe_unused)
32865 +{
32866 +       /* empty */
32867 +}
32868 +#endif /* 4KSTACKS */
32869 +
32870 +static void au_wkq_run(struct au_wkinfo *wkinfo)
32871 +{
32872 +       if (au_ftest_wkq(wkinfo->flags, NEST)) {
32873 +               if (au_wkq_test()) {
32874 +                       AuWarn1("wkq from wkq, unless silly-rename on NFS,"
32875 +                               " due to a dead dir by UDBA?\n");
32876 +                       AuDebugOn(au_ftest_wkq(wkinfo->flags, WAIT));
32877 +               }
32878 +       } else
32879 +               au_dbg_verify_kthread();
32880 +
32881 +       if (au_ftest_wkq(wkinfo->flags, WAIT)) {
32882 +               INIT_WORK_ONSTACK(&wkinfo->wk, wkq_func);
32883 +               queue_work(au_wkq, &wkinfo->wk);
32884 +       } else {
32885 +               INIT_WORK(&wkinfo->wk, wkq_func);
32886 +               schedule_work(&wkinfo->wk);
32887 +       }
32888 +}
32889 +
32890 +/*
32891 + * Be careful. It is easy to make deadlock happen.
32892 + * processA: lock, wkq and wait
32893 + * processB: wkq and wait, lock in wkq
32894 + * --> deadlock
32895 + */
32896 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args)
32897 +{
32898 +       int err;
32899 +       AuWkqCompDeclare(comp);
32900 +       struct au_wkinfo wkinfo = {
32901 +               .flags  = flags,
32902 +               .func   = func,
32903 +               .args   = args
32904 +       };
32905 +
32906 +       err = au_wkq_comp_alloc(&wkinfo, &comp);
32907 +       if (!err) {
32908 +               au_wkq_run(&wkinfo);
32909 +               /* no timeout, no interrupt */
32910 +               wait_for_completion(wkinfo.comp);
32911 +               au_wkq_comp_free(comp);
32912 +               destroy_work_on_stack(&wkinfo.wk);
32913 +       }
32914 +
32915 +       return err;
32916 +
32917 +}
32918 +
32919 +/*
32920 + * Note: dget/dput() in func for aufs dentries are not supported. It will be a
32921 + * problem in a concurrent umounting.
32922 + */
32923 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
32924 +                 unsigned int flags)
32925 +{
32926 +       int err;
32927 +       struct au_wkinfo *wkinfo;
32928 +
32929 +       atomic_inc(&au_sbi(sb)->si_nowait.nw_len);
32930 +
32931 +       /*
32932 +        * wkq_func() must free this wkinfo.
32933 +        * it highly depends upon the implementation of workqueue.
32934 +        */
32935 +       err = 0;
32936 +       wkinfo = kmalloc(sizeof(*wkinfo), GFP_NOFS);
32937 +       if (wkinfo) {
32938 +               wkinfo->kobj = &au_sbi(sb)->si_kobj;
32939 +               wkinfo->flags = flags & ~AuWkq_WAIT;
32940 +               wkinfo->func = func;
32941 +               wkinfo->args = args;
32942 +               wkinfo->comp = NULL;
32943 +               kobject_get(wkinfo->kobj);
32944 +               __module_get(THIS_MODULE); /* todo: ?? */
32945 +
32946 +               au_wkq_run(wkinfo);
32947 +       } else {
32948 +               err = -ENOMEM;
32949 +               au_nwt_done(&au_sbi(sb)->si_nowait);
32950 +       }
32951 +
32952 +       return err;
32953 +}
32954 +
32955 +/* ---------------------------------------------------------------------- */
32956 +
32957 +void au_nwt_init(struct au_nowait_tasks *nwt)
32958 +{
32959 +       atomic_set(&nwt->nw_len, 0);
32960 +       /* smp_mb(); */ /* atomic_set */
32961 +       init_waitqueue_head(&nwt->nw_wq);
32962 +}
32963 +
32964 +void au_wkq_fin(void)
32965 +{
32966 +       destroy_workqueue(au_wkq);
32967 +}
32968 +
32969 +int __init au_wkq_init(void)
32970 +{
32971 +       int err;
32972 +
32973 +       err = 0;
32974 +       au_wkq = alloc_workqueue(AUFS_WKQ_NAME, 0, WQ_DFL_ACTIVE);
32975 +       if (IS_ERR(au_wkq))
32976 +               err = PTR_ERR(au_wkq);
32977 +       else if (!au_wkq)
32978 +               err = -ENOMEM;
32979 +
32980 +       return err;
32981 +}
32982 diff -urN /usr/share/empty/fs/aufs/wkq.h linux/fs/aufs/wkq.h
32983 --- /usr/share/empty/fs/aufs/wkq.h      1970-01-01 01:00:00.000000000 +0100
32984 +++ linux/fs/aufs/wkq.h 2015-03-27 21:56:35.470128334 +0100
32985 @@ -0,0 +1,91 @@
32986 +/*
32987 + * Copyright (C) 2005-2015 Junjiro R. Okajima
32988 + *
32989 + * This program, aufs is free software; you can redistribute it and/or modify
32990 + * it under the terms of the GNU General Public License as published by
32991 + * the Free Software Foundation; either version 2 of the License, or
32992 + * (at your option) any later version.
32993 + *
32994 + * This program is distributed in the hope that it will be useful,
32995 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32996 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32997 + * GNU General Public License for more details.
32998 + *
32999 + * You should have received a copy of the GNU General Public License
33000 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33001 + */
33002 +
33003 +/*
33004 + * workqueue for asynchronous/super-io operations
33005 + * todo: try new credentials management scheme
33006 + */
33007 +
33008 +#ifndef __AUFS_WKQ_H__
33009 +#define __AUFS_WKQ_H__
33010 +
33011 +#ifdef __KERNEL__
33012 +
33013 +struct super_block;
33014 +
33015 +/* ---------------------------------------------------------------------- */
33016 +
33017 +/*
33018 + * in the next operation, wait for the 'nowait' tasks in system-wide workqueue
33019 + */
33020 +struct au_nowait_tasks {
33021 +       atomic_t                nw_len;
33022 +       wait_queue_head_t       nw_wq;
33023 +};
33024 +
33025 +/* ---------------------------------------------------------------------- */
33026 +
33027 +typedef void (*au_wkq_func_t)(void *args);
33028 +
33029 +/* wkq flags */
33030 +#define AuWkq_WAIT     1
33031 +#define AuWkq_NEST     (1 << 1)
33032 +#define au_ftest_wkq(flags, name)      ((flags) & AuWkq_##name)
33033 +#define au_fset_wkq(flags, name) \
33034 +       do { (flags) |= AuWkq_##name; } while (0)
33035 +#define au_fclr_wkq(flags, name) \
33036 +       do { (flags) &= ~AuWkq_##name; } while (0)
33037 +
33038 +#ifndef CONFIG_AUFS_HNOTIFY
33039 +#undef AuWkq_NEST
33040 +#define AuWkq_NEST     0
33041 +#endif
33042 +
33043 +/* wkq.c */
33044 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args);
33045 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
33046 +                 unsigned int flags);
33047 +void au_nwt_init(struct au_nowait_tasks *nwt);
33048 +int __init au_wkq_init(void);
33049 +void au_wkq_fin(void);
33050 +
33051 +/* ---------------------------------------------------------------------- */
33052 +
33053 +static inline int au_wkq_test(void)
33054 +{
33055 +       return current->flags & PF_WQ_WORKER;
33056 +}
33057 +
33058 +static inline int au_wkq_wait(au_wkq_func_t func, void *args)
33059 +{
33060 +       return au_wkq_do_wait(AuWkq_WAIT, func, args);
33061 +}
33062 +
33063 +static inline void au_nwt_done(struct au_nowait_tasks *nwt)
33064 +{
33065 +       if (atomic_dec_and_test(&nwt->nw_len))
33066 +               wake_up_all(&nwt->nw_wq);
33067 +}
33068 +
33069 +static inline int au_nwt_flush(struct au_nowait_tasks *nwt)
33070 +{
33071 +       wait_event(nwt->nw_wq, !atomic_read(&nwt->nw_len));
33072 +       return 0;
33073 +}
33074 +
33075 +#endif /* __KERNEL__ */
33076 +#endif /* __AUFS_WKQ_H__ */
33077 diff -urN /usr/share/empty/fs/aufs/xattr.c linux/fs/aufs/xattr.c
33078 --- /usr/share/empty/fs/aufs/xattr.c    1970-01-01 01:00:00.000000000 +0100
33079 +++ linux/fs/aufs/xattr.c       2015-03-27 21:56:35.470128334 +0100
33080 @@ -0,0 +1,334 @@
33081 +/*
33082 + * Copyright (C) 2014-2015 Junjiro R. Okajima
33083 + *
33084 + * This program, aufs is free software; you can redistribute it and/or modify
33085 + * it under the terms of the GNU General Public License as published by
33086 + * the Free Software Foundation; either version 2 of the License, or
33087 + * (at your option) any later version.
33088 + *
33089 + * This program is distributed in the hope that it will be useful,
33090 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33091 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33092 + * GNU General Public License for more details.
33093 + *
33094 + * You should have received a copy of the GNU General Public License
33095 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33096 + */
33097 +
33098 +/*
33099 + * handling xattr functions
33100 + */
33101 +
33102 +#include <linux/xattr.h>
33103 +#include "aufs.h"
33104 +
33105 +static int au_xattr_ignore(int err, char *name, unsigned int ignore_flags)
33106 +{
33107 +       if (!ignore_flags)
33108 +               goto out;
33109 +       switch (err) {
33110 +       case -ENOMEM:
33111 +       case -EDQUOT:
33112 +               goto out;
33113 +       }
33114 +
33115 +       if ((ignore_flags & AuBrAttr_ICEX) == AuBrAttr_ICEX) {
33116 +               err = 0;
33117 +               goto out;
33118 +       }
33119 +
33120 +#define cmp(brattr, prefix) do {                                       \
33121 +               if (!strncmp(name, XATTR_##prefix##_PREFIX,             \
33122 +                            XATTR_##prefix##_PREFIX_LEN)) {            \
33123 +                       if (ignore_flags & AuBrAttr_ICEX_##brattr)      \
33124 +                               err = 0;                                \
33125 +                       goto out;                                       \
33126 +               }                                                       \
33127 +       } while (0)
33128 +
33129 +       cmp(SEC, SECURITY);
33130 +       cmp(SYS, SYSTEM);
33131 +       cmp(TR, TRUSTED);
33132 +       cmp(USR, USER);
33133 +#undef cmp
33134 +
33135 +       if (ignore_flags & AuBrAttr_ICEX_OTH)
33136 +               err = 0;
33137 +
33138 +out:
33139 +       return err;
33140 +}
33141 +
33142 +static const int au_xattr_out_of_list = AuBrAttr_ICEX_OTH << 1;
33143 +
33144 +static int au_do_cpup_xattr(struct dentry *h_dst, struct dentry *h_src,
33145 +                           char *name, char **buf, unsigned int ignore_flags)
33146 +{
33147 +       int err;
33148 +       ssize_t ssz;
33149 +       struct inode *h_idst;
33150 +
33151 +       ssz = vfs_getxattr_alloc(h_src, name, buf, 0, GFP_NOFS);
33152 +       err = ssz;
33153 +       if (unlikely(err <= 0)) {
33154 +               AuTraceErr(err);
33155 +               if (err == -ENODATA
33156 +                   || (err == -EOPNOTSUPP
33157 +                       && (ignore_flags & au_xattr_out_of_list)))
33158 +                       err = 0;
33159 +               goto out;
33160 +       }
33161 +
33162 +       /* unlock it temporary */
33163 +       h_idst = h_dst->d_inode;
33164 +       mutex_unlock(&h_idst->i_mutex);
33165 +       err = vfsub_setxattr(h_dst, name, *buf, ssz, /*flags*/0);
33166 +       mutex_lock_nested(&h_idst->i_mutex, AuLsc_I_CHILD2);
33167 +       if (unlikely(err)) {
33168 +               AuDbg("%s, err %d\n", name, err);
33169 +               err = au_xattr_ignore(err, name, ignore_flags);
33170 +       }
33171 +
33172 +out:
33173 +       return err;
33174 +}
33175 +
33176 +int au_cpup_xattr(struct dentry *h_dst, struct dentry *h_src, int ignore_flags)
33177 +{
33178 +       int err, unlocked, acl_access, acl_default;
33179 +       ssize_t ssz;
33180 +       struct inode *h_isrc, *h_idst;
33181 +       char *value, *p, *o, *e;
33182 +
33183 +       /* try stopping to update the source inode while we are referencing */
33184 +       /* there should not be the parent-child relation ship between them */
33185 +       h_isrc = h_src->d_inode;
33186 +       h_idst = h_dst->d_inode;
33187 +       mutex_unlock(&h_idst->i_mutex);
33188 +       mutex_lock_nested(&h_isrc->i_mutex, AuLsc_I_CHILD);
33189 +       mutex_lock_nested(&h_idst->i_mutex, AuLsc_I_CHILD2);
33190 +       unlocked = 0;
33191 +
33192 +       /* some filesystems don't list POSIX ACL, for example tmpfs */
33193 +       ssz = vfs_listxattr(h_src, NULL, 0);
33194 +       err = ssz;
33195 +       if (unlikely(err < 0)) {
33196 +               AuTraceErr(err);
33197 +               if (err == -ENODATA
33198 +                   || err == -EOPNOTSUPP)
33199 +                       err = 0;        /* ignore */
33200 +               goto out;
33201 +       }
33202 +
33203 +       err = 0;
33204 +       p = NULL;
33205 +       o = NULL;
33206 +       if (ssz) {
33207 +               err = -ENOMEM;
33208 +               p = kmalloc(ssz, GFP_NOFS);
33209 +               o = p;
33210 +               if (unlikely(!p))
33211 +                       goto out;
33212 +               err = vfs_listxattr(h_src, p, ssz);
33213 +       }
33214 +       mutex_unlock(&h_isrc->i_mutex);
33215 +       unlocked = 1;
33216 +       AuDbg("err %d, ssz %zd\n", err, ssz);
33217 +       if (unlikely(err < 0))
33218 +               goto out_free;
33219 +
33220 +       err = 0;
33221 +       e = p + ssz;
33222 +       value = NULL;
33223 +       acl_access = 0;
33224 +       acl_default = 0;
33225 +       while (!err && p < e) {
33226 +               acl_access |= !strncmp(p, XATTR_NAME_POSIX_ACL_ACCESS,
33227 +                                      sizeof(XATTR_NAME_POSIX_ACL_ACCESS) - 1);
33228 +               acl_default |= !strncmp(p, XATTR_NAME_POSIX_ACL_DEFAULT,
33229 +                                       sizeof(XATTR_NAME_POSIX_ACL_DEFAULT)
33230 +                                       - 1);
33231 +               err = au_do_cpup_xattr(h_dst, h_src, p, &value, ignore_flags);
33232 +               p += strlen(p) + 1;
33233 +       }
33234 +       AuTraceErr(err);
33235 +       ignore_flags |= au_xattr_out_of_list;
33236 +       if (!err && !acl_access) {
33237 +               err = au_do_cpup_xattr(h_dst, h_src,
33238 +                                      XATTR_NAME_POSIX_ACL_ACCESS, &value,
33239 +                                      ignore_flags);
33240 +               AuTraceErr(err);
33241 +       }
33242 +       if (!err && !acl_default) {
33243 +               err = au_do_cpup_xattr(h_dst, h_src,
33244 +                                      XATTR_NAME_POSIX_ACL_DEFAULT, &value,
33245 +                                      ignore_flags);
33246 +               AuTraceErr(err);
33247 +       }
33248 +
33249 +       kfree(value);
33250 +
33251 +out_free:
33252 +       kfree(o);
33253 +out:
33254 +       if (!unlocked)
33255 +               mutex_unlock(&h_isrc->i_mutex);
33256 +       AuTraceErr(err);
33257 +       return err;
33258 +}
33259 +
33260 +/* ---------------------------------------------------------------------- */
33261 +
33262 +enum {
33263 +       AU_XATTR_LIST,
33264 +       AU_XATTR_GET
33265 +};
33266 +
33267 +struct au_lgxattr {
33268 +       int type;
33269 +       union {
33270 +               struct {
33271 +                       char    *list;
33272 +                       size_t  size;
33273 +               } list;
33274 +               struct {
33275 +                       const char      *name;
33276 +                       void            *value;
33277 +                       size_t          size;
33278 +               } get;
33279 +       } u;
33280 +};
33281 +
33282 +static ssize_t au_lgxattr(struct dentry *dentry, struct au_lgxattr *arg)
33283 +{
33284 +       ssize_t err;
33285 +       struct path h_path;
33286 +       struct super_block *sb;
33287 +
33288 +       sb = dentry->d_sb;
33289 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
33290 +       if (unlikely(err))
33291 +               goto out;
33292 +       err = au_h_path_getattr(dentry, /*force*/1, &h_path);
33293 +       if (unlikely(err))
33294 +               goto out_si;
33295 +       if (unlikely(!h_path.dentry))
33296 +               /* illegally overlapped or something */
33297 +               goto out_di; /* pretending success */
33298 +
33299 +       /* always topmost entry only */
33300 +       switch (arg->type) {
33301 +       case AU_XATTR_LIST:
33302 +               err = vfs_listxattr(h_path.dentry,
33303 +                                   arg->u.list.list, arg->u.list.size);
33304 +               break;
33305 +       case AU_XATTR_GET:
33306 +               err = vfs_getxattr(h_path.dentry,
33307 +                                  arg->u.get.name, arg->u.get.value,
33308 +                                  arg->u.get.size);
33309 +               break;
33310 +       }
33311 +
33312 +out_di:
33313 +       di_read_unlock(dentry, AuLock_IR);
33314 +out_si:
33315 +       si_read_unlock(sb);
33316 +out:
33317 +       AuTraceErr(err);
33318 +       return err;
33319 +}
33320 +
33321 +ssize_t aufs_listxattr(struct dentry *dentry, char *list, size_t size)
33322 +{
33323 +       struct au_lgxattr arg = {
33324 +               .type = AU_XATTR_LIST,
33325 +               .u.list = {
33326 +                       .list   = list,
33327 +                       .size   = size
33328 +               },
33329 +       };
33330 +
33331 +       return au_lgxattr(dentry, &arg);
33332 +}
33333 +
33334 +ssize_t aufs_getxattr(struct dentry *dentry, const char *name, void *value,
33335 +                     size_t size)
33336 +{
33337 +       struct au_lgxattr arg = {
33338 +               .type = AU_XATTR_GET,
33339 +               .u.get = {
33340 +                       .name   = name,
33341 +                       .value  = value,
33342 +                       .size   = size
33343 +               },
33344 +       };
33345 +
33346 +       return au_lgxattr(dentry, &arg);
33347 +}
33348 +
33349 +int aufs_setxattr(struct dentry *dentry, const char *name, const void *value,
33350 +                 size_t size, int flags)
33351 +{
33352 +       struct au_srxattr arg = {
33353 +               .type = AU_XATTR_SET,
33354 +               .u.set = {
33355 +                       .name   = name,
33356 +                       .value  = value,
33357 +                       .size   = size,
33358 +                       .flags  = flags
33359 +               },
33360 +       };
33361 +
33362 +       return au_srxattr(dentry, &arg);
33363 +}
33364 +
33365 +int aufs_removexattr(struct dentry *dentry, const char *name)
33366 +{
33367 +       struct au_srxattr arg = {
33368 +               .type = AU_XATTR_REMOVE,
33369 +               .u.remove = {
33370 +                       .name   = name
33371 +               },
33372 +       };
33373 +
33374 +       return au_srxattr(dentry, &arg);
33375 +}
33376 +
33377 +/* ---------------------------------------------------------------------- */
33378 +
33379 +#if 0
33380 +static size_t au_xattr_list(struct dentry *dentry, char *list, size_t list_size,
33381 +                           const char *name, size_t name_len, int type)
33382 +{
33383 +       return aufs_listxattr(dentry, list, list_size);
33384 +}
33385 +
33386 +static int au_xattr_get(struct dentry *dentry, const char *name, void *buffer,
33387 +                       size_t size, int type)
33388 +{
33389 +       return aufs_getxattr(dentry, name, buffer, size);
33390 +}
33391 +
33392 +static int au_xattr_set(struct dentry *dentry, const char *name,
33393 +                       const void *value, size_t size, int flags, int type)
33394 +{
33395 +       return aufs_setxattr(dentry, name, value, size, flags);
33396 +}
33397 +
33398 +static const struct xattr_handler au_xattr_handler = {
33399 +       /* no prefix, no flags */
33400 +       .list   = au_xattr_list,
33401 +       .get    = au_xattr_get,
33402 +       .set    = au_xattr_set
33403 +       /* why no remove? */
33404 +};
33405 +
33406 +static const struct xattr_handler *au_xattr_handlers[] = {
33407 +       &au_xattr_handler
33408 +};
33409 +
33410 +void au_xattr_init(struct super_block *sb)
33411 +{
33412 +       /* sb->s_xattr = au_xattr_handlers; */
33413 +}
33414 +#endif
33415 diff -urN /usr/share/empty/fs/aufs/xino.c linux/fs/aufs/xino.c
33416 --- /usr/share/empty/fs/aufs/xino.c     1970-01-01 01:00:00.000000000 +0100
33417 +++ linux/fs/aufs/xino.c        2015-03-27 21:56:35.470128334 +0100
33418 @@ -0,0 +1,1318 @@
33419 +/*
33420 + * Copyright (C) 2005-2015 Junjiro R. Okajima
33421 + *
33422 + * This program, aufs is free software; you can redistribute it and/or modify
33423 + * it under the terms of the GNU General Public License as published by
33424 + * the Free Software Foundation; either version 2 of the License, or
33425 + * (at your option) any later version.
33426 + *
33427 + * This program is distributed in the hope that it will be useful,
33428 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
33429 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
33430 + * GNU General Public License for more details.
33431 + *
33432 + * You should have received a copy of the GNU General Public License
33433 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
33434 + */
33435 +
33436 +/*
33437 + * external inode number translation table and bitmap
33438 + */
33439 +
33440 +#include <linux/seq_file.h>
33441 +#include <linux/statfs.h>
33442 +#include "aufs.h"
33443 +
33444 +/* todo: unnecessary to support mmap_sem since kernel-space? */
33445 +ssize_t xino_fread(au_readf_t func, struct file *file, void *kbuf, size_t size,
33446 +                  loff_t *pos)
33447 +{
33448 +       ssize_t err;
33449 +       mm_segment_t oldfs;
33450 +       union {
33451 +               void *k;
33452 +               char __user *u;
33453 +       } buf;
33454 +
33455 +       buf.k = kbuf;
33456 +       oldfs = get_fs();
33457 +       set_fs(KERNEL_DS);
33458 +       do {
33459 +               /* todo: signal_pending? */
33460 +               err = func(file, buf.u, size, pos);
33461 +       } while (err == -EAGAIN || err == -EINTR);
33462 +       set_fs(oldfs);
33463 +
33464 +#if 0 /* reserved for future use */
33465 +       if (err > 0)
33466 +               fsnotify_access(file->f_path.dentry);
33467 +#endif
33468 +
33469 +       return err;
33470 +}
33471 +
33472 +/* ---------------------------------------------------------------------- */
33473 +
33474 +static ssize_t do_xino_fwrite(au_writef_t func, struct file *file, void *kbuf,
33475 +                             size_t size, loff_t *pos)
33476 +{
33477 +       ssize_t err;
33478 +       mm_segment_t oldfs;
33479 +       union {
33480 +               void *k;
33481 +               const char __user *u;
33482 +       } buf;
33483 +
33484 +       buf.k = kbuf;
33485 +       oldfs = get_fs();
33486 +       set_fs(KERNEL_DS);
33487 +       do {
33488 +               /* todo: signal_pending? */
33489 +               err = func(file, buf.u, size, pos);
33490 +       } while (err == -EAGAIN || err == -EINTR);
33491 +       set_fs(oldfs);
33492 +
33493 +#if 0 /* reserved for future use */
33494 +       if (err > 0)
33495 +               fsnotify_modify(file->f_path.dentry);
33496 +#endif
33497 +
33498 +       return err;
33499 +}
33500 +
33501 +struct do_xino_fwrite_args {
33502 +       ssize_t *errp;
33503 +       au_writef_t func;
33504 +       struct file *file;
33505 +       void *buf;
33506 +       size_t size;
33507 +       loff_t *pos;
33508 +};
33509 +
33510 +static void call_do_xino_fwrite(void *args)
33511 +{
33512 +       struct do_xino_fwrite_args *a = args;
33513 +       *a->errp = do_xino_fwrite(a->func, a->file, a->buf, a->size, a->pos);
33514 +}
33515 +
33516 +ssize_t xino_fwrite(au_writef_t func, struct file *file, void *buf, size_t size,
33517 +                   loff_t *pos)
33518 +{
33519 +       ssize_t err;
33520 +
33521 +       /* todo: signal block and no wkq? */
33522 +       if (rlimit(RLIMIT_FSIZE) == RLIM_INFINITY) {
33523 +               lockdep_off();
33524 +               err = do_xino_fwrite(func, file, buf, size, pos);
33525 +               lockdep_on();
33526 +       } else {
33527 +               /*
33528 +                * it breaks RLIMIT_FSIZE and normal user's limit,
33529 +                * users should care about quota and real 'filesystem full.'
33530 +                */
33531 +               int wkq_err;
33532 +               struct do_xino_fwrite_args args = {
33533 +                       .errp   = &err,
33534 +                       .func   = func,
33535 +                       .file   = file,
33536 +                       .buf    = buf,
33537 +                       .size   = size,
33538 +                       .pos    = pos
33539 +               };
33540 +
33541 +               wkq_err = au_wkq_wait(call_do_xino_fwrite, &args);
33542 +               if (unlikely(wkq_err))
33543 +                       err = wkq_err;
33544 +       }
33545 +
33546 +       return err;
33547 +}
33548 +
33549 +/* ---------------------------------------------------------------------- */
33550 +
33551 +/*
33552 + * create a new xinofile at the same place/path as @base_file.
33553 + */
33554 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src)
33555 +{
33556 +       struct file *file;
33557 +       struct dentry *base, *parent;
33558 +       struct inode *dir, *delegated;
33559 +       struct qstr *name;
33560 +       struct path path;
33561 +       int err;
33562 +
33563 +       base = base_file->f_path.dentry;
33564 +       parent = base->d_parent; /* dir inode is locked */
33565 +       dir = parent->d_inode;
33566 +       IMustLock(dir);
33567 +
33568 +       file = ERR_PTR(-EINVAL);
33569 +       name = &base->d_name;
33570 +       path.dentry = vfsub_lookup_one_len(name->name, parent, name->len);
33571 +       if (IS_ERR(path.dentry)) {
33572 +               file = (void *)path.dentry;
33573 +               pr_err("%pd lookup err %ld\n",
33574 +                      base, PTR_ERR(path.dentry));
33575 +               goto out;
33576 +       }
33577 +
33578 +       /* no need to mnt_want_write() since we call dentry_open() later */
33579 +       err = vfs_create(dir, path.dentry, S_IRUGO | S_IWUGO, NULL);
33580 +       if (unlikely(err)) {
33581 +               file = ERR_PTR(err);
33582 +               pr_err("%pd create err %d\n", base, err);
33583 +               goto out_dput;
33584 +       }
33585 +
33586 +       path.mnt = base_file->f_path.mnt;
33587 +       file = vfsub_dentry_open(&path,
33588 +                                O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
33589 +                                /* | __FMODE_NONOTIFY */);
33590 +       if (IS_ERR(file)) {
33591 +               pr_err("%pd open err %ld\n", base, PTR_ERR(file));
33592 +               goto out_dput;
33593 +       }
33594 +
33595 +       delegated = NULL;
33596 +       err = vfsub_unlink(dir, &file->f_path, &delegated, /*force*/0);
33597 +       if (unlikely(err == -EWOULDBLOCK)) {
33598 +               pr_warn("cannot retry for NFSv4 delegation"
33599 +                       " for an internal unlink\n");
33600 +               iput(delegated);
33601 +       }
33602 +       if (unlikely(err)) {
33603 +               pr_err("%pd unlink err %d\n", base, err);
33604 +               goto out_fput;
33605 +       }
33606 +
33607 +       if (copy_src) {
33608 +               /* no one can touch copy_src xino */
33609 +               err = au_copy_file(file, copy_src, vfsub_f_size_read(copy_src));
33610 +               if (unlikely(err)) {
33611 +                       pr_err("%pd copy err %d\n", base, err);
33612 +                       goto out_fput;
33613 +               }
33614 +       }
33615 +       goto out_dput; /* success */
33616 +
33617 +out_fput:
33618 +       fput(file);
33619 +       file = ERR_PTR(err);
33620 +out_dput:
33621 +       dput(path.dentry);
33622 +out:
33623 +       return file;
33624 +}
33625 +
33626 +struct au_xino_lock_dir {
33627 +       struct au_hinode *hdir;
33628 +       struct dentry *parent;
33629 +       struct mutex *mtx;
33630 +};
33631 +
33632 +static void au_xino_lock_dir(struct super_block *sb, struct file *xino,
33633 +                            struct au_xino_lock_dir *ldir)
33634 +{
33635 +       aufs_bindex_t brid, bindex;
33636 +
33637 +       ldir->hdir = NULL;
33638 +       bindex = -1;
33639 +       brid = au_xino_brid(sb);
33640 +       if (brid >= 0)
33641 +               bindex = au_br_index(sb, brid);
33642 +       if (bindex >= 0) {
33643 +               ldir->hdir = au_hi(sb->s_root->d_inode, bindex);
33644 +               au_hn_imtx_lock_nested(ldir->hdir, AuLsc_I_PARENT);
33645 +       } else {
33646 +               ldir->parent = dget_parent(xino->f_path.dentry);
33647 +               ldir->mtx = &ldir->parent->d_inode->i_mutex;
33648 +               mutex_lock_nested(ldir->mtx, AuLsc_I_PARENT);
33649 +       }
33650 +}
33651 +
33652 +static void au_xino_unlock_dir(struct au_xino_lock_dir *ldir)
33653 +{
33654 +       if (ldir->hdir)
33655 +               au_hn_imtx_unlock(ldir->hdir);
33656 +       else {
33657 +               mutex_unlock(ldir->mtx);
33658 +               dput(ldir->parent);
33659 +       }
33660 +}
33661 +
33662 +/* ---------------------------------------------------------------------- */
33663 +
33664 +/* trucate xino files asynchronously */
33665 +
33666 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex)
33667 +{
33668 +       int err;
33669 +       unsigned long jiffy;
33670 +       blkcnt_t blocks;
33671 +       aufs_bindex_t bi, bend;
33672 +       struct kstatfs *st;
33673 +       struct au_branch *br;
33674 +       struct file *new_xino, *file;
33675 +       struct super_block *h_sb;
33676 +       struct au_xino_lock_dir ldir;
33677 +
33678 +       err = -ENOMEM;
33679 +       st = kzalloc(sizeof(*st), GFP_NOFS);
33680 +       if (unlikely(!st))
33681 +               goto out;
33682 +
33683 +       err = -EINVAL;
33684 +       bend = au_sbend(sb);
33685 +       if (unlikely(bindex < 0 || bend < bindex))
33686 +               goto out_st;
33687 +       br = au_sbr(sb, bindex);
33688 +       file = br->br_xino.xi_file;
33689 +       if (!file)
33690 +               goto out_st;
33691 +
33692 +       err = vfs_statfs(&file->f_path, st);
33693 +       if (unlikely(err))
33694 +               AuErr1("statfs err %d, ignored\n", err);
33695 +       jiffy = jiffies;
33696 +       blocks = file_inode(file)->i_blocks;
33697 +       pr_info("begin truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
33698 +               bindex, (u64)blocks, st->f_bfree, st->f_blocks);
33699 +
33700 +       au_xino_lock_dir(sb, file, &ldir);
33701 +       /* mnt_want_write() is unnecessary here */
33702 +       new_xino = au_xino_create2(file, file);
33703 +       au_xino_unlock_dir(&ldir);
33704 +       err = PTR_ERR(new_xino);
33705 +       if (IS_ERR(new_xino)) {
33706 +               pr_err("err %d, ignored\n", err);
33707 +               goto out_st;
33708 +       }
33709 +       err = 0;
33710 +       fput(file);
33711 +       br->br_xino.xi_file = new_xino;
33712 +
33713 +       h_sb = au_br_sb(br);
33714 +       for (bi = 0; bi <= bend; bi++) {
33715 +               if (unlikely(bi == bindex))
33716 +                       continue;
33717 +               br = au_sbr(sb, bi);
33718 +               if (au_br_sb(br) != h_sb)
33719 +                       continue;
33720 +
33721 +               fput(br->br_xino.xi_file);
33722 +               br->br_xino.xi_file = new_xino;
33723 +               get_file(new_xino);
33724 +       }
33725 +
33726 +       err = vfs_statfs(&new_xino->f_path, st);
33727 +       if (!err) {
33728 +               pr_info("end truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
33729 +                       bindex, (u64)file_inode(new_xino)->i_blocks,
33730 +                       st->f_bfree, st->f_blocks);
33731 +               if (file_inode(new_xino)->i_blocks < blocks)
33732 +                       au_sbi(sb)->si_xino_jiffy = jiffy;
33733 +       } else
33734 +               AuErr1("statfs err %d, ignored\n", err);
33735 +
33736 +out_st:
33737 +       kfree(st);
33738 +out:
33739 +       return err;
33740 +}
33741 +
33742 +struct xino_do_trunc_args {
33743 +       struct super_block *sb;
33744 +       struct au_branch *br;
33745 +};
33746 +
33747 +static void xino_do_trunc(void *_args)
33748 +{
33749 +       struct xino_do_trunc_args *args = _args;
33750 +       struct super_block *sb;
33751 +       struct au_branch *br;
33752 +       struct inode *dir;
33753 +       int err;
33754 +       aufs_bindex_t bindex;
33755 +
33756 +       err = 0;
33757 +       sb = args->sb;
33758 +       dir = sb->s_root->d_inode;
33759 +       br = args->br;
33760 +
33761 +       si_noflush_write_lock(sb);
33762 +       ii_read_lock_parent(dir);
33763 +       bindex = au_br_index(sb, br->br_id);
33764 +       err = au_xino_trunc(sb, bindex);
33765 +       ii_read_unlock(dir);
33766 +       if (unlikely(err))
33767 +               pr_warn("err b%d, (%d)\n", bindex, err);
33768 +       atomic_dec(&br->br_xino_running);
33769 +       atomic_dec(&br->br_count);
33770 +       si_write_unlock(sb);
33771 +       au_nwt_done(&au_sbi(sb)->si_nowait);
33772 +       kfree(args);
33773 +}
33774 +
33775 +static int xino_trunc_test(struct super_block *sb, struct au_branch *br)
33776 +{
33777 +       int err;
33778 +       struct kstatfs st;
33779 +       struct au_sbinfo *sbinfo;
33780 +
33781 +       /* todo: si_xino_expire and the ratio should be customizable */
33782 +       sbinfo = au_sbi(sb);
33783 +       if (time_before(jiffies,
33784 +                       sbinfo->si_xino_jiffy + sbinfo->si_xino_expire))
33785 +               return 0;
33786 +
33787 +       /* truncation border */
33788 +       err = vfs_statfs(&br->br_xino.xi_file->f_path, &st);
33789 +       if (unlikely(err)) {
33790 +               AuErr1("statfs err %d, ignored\n", err);
33791 +               return 0;
33792 +       }
33793 +       if (div64_u64(st.f_bfree * 100, st.f_blocks) >= AUFS_XINO_DEF_TRUNC)
33794 +               return 0;
33795 +
33796 +       return 1;
33797 +}
33798 +
33799 +static void xino_try_trunc(struct super_block *sb, struct au_branch *br)
33800 +{
33801 +       struct xino_do_trunc_args *args;
33802 +       int wkq_err;
33803 +
33804 +       if (!xino_trunc_test(sb, br))
33805 +               return;
33806 +
33807 +       if (atomic_inc_return(&br->br_xino_running) > 1)
33808 +               goto out;
33809 +
33810 +       /* lock and kfree() will be called in trunc_xino() */
33811 +       args = kmalloc(sizeof(*args), GFP_NOFS);
33812 +       if (unlikely(!args)) {
33813 +               AuErr1("no memory\n");
33814 +               goto out_args;
33815 +       }
33816 +
33817 +       atomic_inc(&br->br_count);
33818 +       args->sb = sb;
33819 +       args->br = br;
33820 +       wkq_err = au_wkq_nowait(xino_do_trunc, args, sb, /*flags*/0);
33821 +       if (!wkq_err)
33822 +               return; /* success */
33823 +
33824 +       pr_err("wkq %d\n", wkq_err);
33825 +       atomic_dec(&br->br_count);
33826 +
33827 +out_args:
33828 +       kfree(args);
33829 +out:
33830 +       atomic_dec(&br->br_xino_running);
33831 +}
33832 +
33833 +/* ---------------------------------------------------------------------- */
33834 +
33835 +static int au_xino_do_write(au_writef_t write, struct file *file,
33836 +                           ino_t h_ino, ino_t ino)
33837 +{
33838 +       loff_t pos;
33839 +       ssize_t sz;
33840 +
33841 +       pos = h_ino;
33842 +       if (unlikely(au_loff_max / sizeof(ino) - 1 < pos)) {
33843 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
33844 +               return -EFBIG;
33845 +       }
33846 +       pos *= sizeof(ino);
33847 +       sz = xino_fwrite(write, file, &ino, sizeof(ino), &pos);
33848 +       if (sz == sizeof(ino))
33849 +               return 0; /* success */
33850 +
33851 +       AuIOErr("write failed (%zd)\n", sz);
33852 +       return -EIO;
33853 +}
33854 +
33855 +/*
33856 + * write @ino to the xinofile for the specified branch{@sb, @bindex}
33857 + * at the position of @h_ino.
33858 + * even if @ino is zero, it is written to the xinofile and means no entry.
33859 + * if the size of the xino file on a specific filesystem exceeds the watermark,
33860 + * try truncating it.
33861 + */
33862 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
33863 +                 ino_t ino)
33864 +{
33865 +       int err;
33866 +       unsigned int mnt_flags;
33867 +       struct au_branch *br;
33868 +
33869 +       BUILD_BUG_ON(sizeof(long long) != sizeof(au_loff_max)
33870 +                    || ((loff_t)-1) > 0);
33871 +       SiMustAnyLock(sb);
33872 +
33873 +       mnt_flags = au_mntflags(sb);
33874 +       if (!au_opt_test(mnt_flags, XINO))
33875 +               return 0;
33876 +
33877 +       br = au_sbr(sb, bindex);
33878 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
33879 +                              h_ino, ino);
33880 +       if (!err) {
33881 +               if (au_opt_test(mnt_flags, TRUNC_XINO)
33882 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
33883 +                       xino_try_trunc(sb, br);
33884 +               return 0; /* success */
33885 +       }
33886 +
33887 +       AuIOErr("write failed (%d)\n", err);
33888 +       return -EIO;
33889 +}
33890 +
33891 +/* ---------------------------------------------------------------------- */
33892 +
33893 +/* aufs inode number bitmap */
33894 +
33895 +static const int page_bits = (int)PAGE_SIZE * BITS_PER_BYTE;
33896 +static ino_t xib_calc_ino(unsigned long pindex, int bit)
33897 +{
33898 +       ino_t ino;
33899 +
33900 +       AuDebugOn(bit < 0 || page_bits <= bit);
33901 +       ino = AUFS_FIRST_INO + pindex * page_bits + bit;
33902 +       return ino;
33903 +}
33904 +
33905 +static void xib_calc_bit(ino_t ino, unsigned long *pindex, int *bit)
33906 +{
33907 +       AuDebugOn(ino < AUFS_FIRST_INO);
33908 +       ino -= AUFS_FIRST_INO;
33909 +       *pindex = ino / page_bits;
33910 +       *bit = ino % page_bits;
33911 +}
33912 +
33913 +static int xib_pindex(struct super_block *sb, unsigned long pindex)
33914 +{
33915 +       int err;
33916 +       loff_t pos;
33917 +       ssize_t sz;
33918 +       struct au_sbinfo *sbinfo;
33919 +       struct file *xib;
33920 +       unsigned long *p;
33921 +
33922 +       sbinfo = au_sbi(sb);
33923 +       MtxMustLock(&sbinfo->si_xib_mtx);
33924 +       AuDebugOn(pindex > ULONG_MAX / PAGE_SIZE
33925 +                 || !au_opt_test(sbinfo->si_mntflags, XINO));
33926 +
33927 +       if (pindex == sbinfo->si_xib_last_pindex)
33928 +               return 0;
33929 +
33930 +       xib = sbinfo->si_xib;
33931 +       p = sbinfo->si_xib_buf;
33932 +       pos = sbinfo->si_xib_last_pindex;
33933 +       pos *= PAGE_SIZE;
33934 +       sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
33935 +       if (unlikely(sz != PAGE_SIZE))
33936 +               goto out;
33937 +
33938 +       pos = pindex;
33939 +       pos *= PAGE_SIZE;
33940 +       if (vfsub_f_size_read(xib) >= pos + PAGE_SIZE)
33941 +               sz = xino_fread(sbinfo->si_xread, xib, p, PAGE_SIZE, &pos);
33942 +       else {
33943 +               memset(p, 0, PAGE_SIZE);
33944 +               sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
33945 +       }
33946 +       if (sz == PAGE_SIZE) {
33947 +               sbinfo->si_xib_last_pindex = pindex;
33948 +               return 0; /* success */
33949 +       }
33950 +
33951 +out:
33952 +       AuIOErr1("write failed (%zd)\n", sz);
33953 +       err = sz;
33954 +       if (sz >= 0)
33955 +               err = -EIO;
33956 +       return err;
33957 +}
33958 +
33959 +/* ---------------------------------------------------------------------- */
33960 +
33961 +static void au_xib_clear_bit(struct inode *inode)
33962 +{
33963 +       int err, bit;
33964 +       unsigned long pindex;
33965 +       struct super_block *sb;
33966 +       struct au_sbinfo *sbinfo;
33967 +
33968 +       AuDebugOn(inode->i_nlink);
33969 +
33970 +       sb = inode->i_sb;
33971 +       xib_calc_bit(inode->i_ino, &pindex, &bit);
33972 +       AuDebugOn(page_bits <= bit);
33973 +       sbinfo = au_sbi(sb);
33974 +       mutex_lock(&sbinfo->si_xib_mtx);
33975 +       err = xib_pindex(sb, pindex);
33976 +       if (!err) {
33977 +               clear_bit(bit, sbinfo->si_xib_buf);
33978 +               sbinfo->si_xib_next_bit = bit;
33979 +       }
33980 +       mutex_unlock(&sbinfo->si_xib_mtx);
33981 +}
33982 +
33983 +/* for s_op->delete_inode() */
33984 +void au_xino_delete_inode(struct inode *inode, const int unlinked)
33985 +{
33986 +       int err;
33987 +       unsigned int mnt_flags;
33988 +       aufs_bindex_t bindex, bend, bi;
33989 +       unsigned char try_trunc;
33990 +       struct au_iinfo *iinfo;
33991 +       struct super_block *sb;
33992 +       struct au_hinode *hi;
33993 +       struct inode *h_inode;
33994 +       struct au_branch *br;
33995 +       au_writef_t xwrite;
33996 +
33997 +       sb = inode->i_sb;
33998 +       mnt_flags = au_mntflags(sb);
33999 +       if (!au_opt_test(mnt_flags, XINO)
34000 +           || inode->i_ino == AUFS_ROOT_INO)
34001 +               return;
34002 +
34003 +       if (unlinked) {
34004 +               au_xigen_inc(inode);
34005 +               au_xib_clear_bit(inode);
34006 +       }
34007 +
34008 +       iinfo = au_ii(inode);
34009 +       if (!iinfo)
34010 +               return;
34011 +
34012 +       bindex = iinfo->ii_bstart;
34013 +       if (bindex < 0)
34014 +               return;
34015 +
34016 +       xwrite = au_sbi(sb)->si_xwrite;
34017 +       try_trunc = !!au_opt_test(mnt_flags, TRUNC_XINO);
34018 +       hi = iinfo->ii_hinode + bindex;
34019 +       bend = iinfo->ii_bend;
34020 +       for (; bindex <= bend; bindex++, hi++) {
34021 +               h_inode = hi->hi_inode;
34022 +               if (!h_inode
34023 +                   || (!unlinked && h_inode->i_nlink))
34024 +                       continue;
34025 +
34026 +               /* inode may not be revalidated */
34027 +               bi = au_br_index(sb, hi->hi_id);
34028 +               if (bi < 0)
34029 +                       continue;
34030 +
34031 +               br = au_sbr(sb, bi);
34032 +               err = au_xino_do_write(xwrite, br->br_xino.xi_file,
34033 +                                      h_inode->i_ino, /*ino*/0);
34034 +               if (!err && try_trunc
34035 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
34036 +                       xino_try_trunc(sb, br);
34037 +       }
34038 +}
34039 +
34040 +/* get an unused inode number from bitmap */
34041 +ino_t au_xino_new_ino(struct super_block *sb)
34042 +{
34043 +       ino_t ino;
34044 +       unsigned long *p, pindex, ul, pend;
34045 +       struct au_sbinfo *sbinfo;
34046 +       struct file *file;
34047 +       int free_bit, err;
34048 +
34049 +       if (!au_opt_test(au_mntflags(sb), XINO))
34050 +               return iunique(sb, AUFS_FIRST_INO);
34051 +
34052 +       sbinfo = au_sbi(sb);
34053 +       mutex_lock(&sbinfo->si_xib_mtx);
34054 +       p = sbinfo->si_xib_buf;
34055 +       free_bit = sbinfo->si_xib_next_bit;
34056 +       if (free_bit < page_bits && !test_bit(free_bit, p))
34057 +               goto out; /* success */
34058 +       free_bit = find_first_zero_bit(p, page_bits);
34059 +       if (free_bit < page_bits)
34060 +               goto out; /* success */
34061 +
34062 +       pindex = sbinfo->si_xib_last_pindex;
34063 +       for (ul = pindex - 1; ul < ULONG_MAX; ul--) {
34064 +               err = xib_pindex(sb, ul);
34065 +               if (unlikely(err))
34066 +                       goto out_err;
34067 +               free_bit = find_first_zero_bit(p, page_bits);
34068 +               if (free_bit < page_bits)
34069 +                       goto out; /* success */
34070 +       }
34071 +
34072 +       file = sbinfo->si_xib;
34073 +       pend = vfsub_f_size_read(file) / PAGE_SIZE;
34074 +       for (ul = pindex + 1; ul <= pend; ul++) {
34075 +               err = xib_pindex(sb, ul);
34076 +               if (unlikely(err))
34077 +                       goto out_err;
34078 +               free_bit = find_first_zero_bit(p, page_bits);
34079 +               if (free_bit < page_bits)
34080 +                       goto out; /* success */
34081 +       }
34082 +       BUG();
34083 +
34084 +out:
34085 +       set_bit(free_bit, p);
34086 +       sbinfo->si_xib_next_bit = free_bit + 1;
34087 +       pindex = sbinfo->si_xib_last_pindex;
34088 +       mutex_unlock(&sbinfo->si_xib_mtx);
34089 +       ino = xib_calc_ino(pindex, free_bit);
34090 +       AuDbg("i%lu\n", (unsigned long)ino);
34091 +       return ino;
34092 +out_err:
34093 +       mutex_unlock(&sbinfo->si_xib_mtx);
34094 +       AuDbg("i0\n");
34095 +       return 0;
34096 +}
34097 +
34098 +/*
34099 + * read @ino from xinofile for the specified branch{@sb, @bindex}
34100 + * at the position of @h_ino.
34101 + * if @ino does not exist and @do_new is true, get new one.
34102 + */
34103 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
34104 +                ino_t *ino)
34105 +{
34106 +       int err;
34107 +       ssize_t sz;
34108 +       loff_t pos;
34109 +       struct file *file;
34110 +       struct au_sbinfo *sbinfo;
34111 +
34112 +       *ino = 0;
34113 +       if (!au_opt_test(au_mntflags(sb), XINO))
34114 +               return 0; /* no xino */
34115 +
34116 +       err = 0;
34117 +       sbinfo = au_sbi(sb);
34118 +       pos = h_ino;
34119 +       if (unlikely(au_loff_max / sizeof(*ino) - 1 < pos)) {
34120 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
34121 +               return -EFBIG;
34122 +       }
34123 +       pos *= sizeof(*ino);
34124 +
34125 +       file = au_sbr(sb, bindex)->br_xino.xi_file;
34126 +       if (vfsub_f_size_read(file) < pos + sizeof(*ino))
34127 +               return 0; /* no ino */
34128 +
34129 +       sz = xino_fread(sbinfo->si_xread, file, ino, sizeof(*ino), &pos);
34130 +       if (sz == sizeof(*ino))
34131 +               return 0; /* success */
34132 +
34133 +       err = sz;
34134 +       if (unlikely(sz >= 0)) {
34135 +               err = -EIO;
34136 +               AuIOErr("xino read error (%zd)\n", sz);
34137 +       }
34138 +
34139 +       return err;
34140 +}
34141 +
34142 +/* ---------------------------------------------------------------------- */
34143 +
34144 +/* create and set a new xino file */
34145 +
34146 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent)
34147 +{
34148 +       struct file *file;
34149 +       struct dentry *h_parent, *d;
34150 +       struct inode *h_dir;
34151 +       int err;
34152 +
34153 +       /*
34154 +        * at mount-time, and the xino file is the default path,
34155 +        * hnotify is disabled so we have no notify events to ignore.
34156 +        * when a user specified the xino, we cannot get au_hdir to be ignored.
34157 +        */
34158 +       file = vfsub_filp_open(fname, O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
34159 +                              /* | __FMODE_NONOTIFY */,
34160 +                              S_IRUGO | S_IWUGO);
34161 +       if (IS_ERR(file)) {
34162 +               if (!silent)
34163 +                       pr_err("open %s(%ld)\n", fname, PTR_ERR(file));
34164 +               return file;
34165 +       }
34166 +
34167 +       /* keep file count */
34168 +       h_parent = dget_parent(file->f_path.dentry);
34169 +       h_dir = h_parent->d_inode;
34170 +       mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
34171 +       /* mnt_want_write() is unnecessary here */
34172 +       /* no delegation since it is just created */
34173 +       err = vfsub_unlink(h_dir, &file->f_path, /*delegated*/NULL, /*force*/0);
34174 +       mutex_unlock(&h_dir->i_mutex);
34175 +       dput(h_parent);
34176 +       if (unlikely(err)) {
34177 +               if (!silent)
34178 +                       pr_err("unlink %s(%d)\n", fname, err);
34179 +               goto out;
34180 +       }
34181 +
34182 +       err = -EINVAL;
34183 +       d = file->f_path.dentry;
34184 +       if (unlikely(sb == d->d_sb)) {
34185 +               if (!silent)
34186 +                       pr_err("%s must be outside\n", fname);
34187 +               goto out;
34188 +       }
34189 +       if (unlikely(au_test_fs_bad_xino(d->d_sb))) {
34190 +               if (!silent)
34191 +                       pr_err("xino doesn't support %s(%s)\n",
34192 +                              fname, au_sbtype(d->d_sb));
34193 +               goto out;
34194 +       }
34195 +       return file; /* success */
34196 +
34197 +out:
34198 +       fput(file);
34199 +       file = ERR_PTR(err);
34200 +       return file;
34201 +}
34202 +
34203 +/*
34204 + * find another branch who is on the same filesystem of the specified
34205 + * branch{@btgt}. search until @bend.
34206 + */
34207 +static int is_sb_shared(struct super_block *sb, aufs_bindex_t btgt,
34208 +                       aufs_bindex_t bend)
34209 +{
34210 +       aufs_bindex_t bindex;
34211 +       struct super_block *tgt_sb = au_sbr_sb(sb, btgt);
34212 +
34213 +       for (bindex = 0; bindex < btgt; bindex++)
34214 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
34215 +                       return bindex;
34216 +       for (bindex++; bindex <= bend; bindex++)
34217 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
34218 +                       return bindex;
34219 +       return -1;
34220 +}
34221 +
34222 +/* ---------------------------------------------------------------------- */
34223 +
34224 +/*
34225 + * initialize the xinofile for the specified branch @br
34226 + * at the place/path where @base_file indicates.
34227 + * test whether another branch is on the same filesystem or not,
34228 + * if @do_test is true.
34229 + */
34230 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t h_ino,
34231 +              struct file *base_file, int do_test)
34232 +{
34233 +       int err;
34234 +       ino_t ino;
34235 +       aufs_bindex_t bend, bindex;
34236 +       struct au_branch *shared_br, *b;
34237 +       struct file *file;
34238 +       struct super_block *tgt_sb;
34239 +
34240 +       shared_br = NULL;
34241 +       bend = au_sbend(sb);
34242 +       if (do_test) {
34243 +               tgt_sb = au_br_sb(br);
34244 +               for (bindex = 0; bindex <= bend; bindex++) {
34245 +                       b = au_sbr(sb, bindex);
34246 +                       if (tgt_sb == au_br_sb(b)) {
34247 +                               shared_br = b;
34248 +                               break;
34249 +                       }
34250 +               }
34251 +       }
34252 +
34253 +       if (!shared_br || !shared_br->br_xino.xi_file) {
34254 +               struct au_xino_lock_dir ldir;
34255 +
34256 +               au_xino_lock_dir(sb, base_file, &ldir);
34257 +               /* mnt_want_write() is unnecessary here */
34258 +               file = au_xino_create2(base_file, NULL);
34259 +               au_xino_unlock_dir(&ldir);
34260 +               err = PTR_ERR(file);
34261 +               if (IS_ERR(file))
34262 +                       goto out;
34263 +               br->br_xino.xi_file = file;
34264 +       } else {
34265 +               br->br_xino.xi_file = shared_br->br_xino.xi_file;
34266 +               get_file(br->br_xino.xi_file);
34267 +       }
34268 +
34269 +       ino = AUFS_ROOT_INO;
34270 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
34271 +                              h_ino, ino);
34272 +       if (unlikely(err)) {
34273 +               fput(br->br_xino.xi_file);
34274 +               br->br_xino.xi_file = NULL;
34275 +       }
34276 +
34277 +out:
34278 +       return err;
34279 +}
34280 +
34281 +/* ---------------------------------------------------------------------- */
34282 +
34283 +/* trucate a xino bitmap file */
34284 +
34285 +/* todo: slow */
34286 +static int do_xib_restore(struct super_block *sb, struct file *file, void *page)
34287 +{
34288 +       int err, bit;
34289 +       ssize_t sz;
34290 +       unsigned long pindex;
34291 +       loff_t pos, pend;
34292 +       struct au_sbinfo *sbinfo;
34293 +       au_readf_t func;
34294 +       ino_t *ino;
34295 +       unsigned long *p;
34296 +
34297 +       err = 0;
34298 +       sbinfo = au_sbi(sb);
34299 +       MtxMustLock(&sbinfo->si_xib_mtx);
34300 +       p = sbinfo->si_xib_buf;
34301 +       func = sbinfo->si_xread;
34302 +       pend = vfsub_f_size_read(file);
34303 +       pos = 0;
34304 +       while (pos < pend) {
34305 +               sz = xino_fread(func, file, page, PAGE_SIZE, &pos);
34306 +               err = sz;
34307 +               if (unlikely(sz <= 0))
34308 +                       goto out;
34309 +
34310 +               err = 0;
34311 +               for (ino = page; sz > 0; ino++, sz -= sizeof(ino)) {
34312 +                       if (unlikely(*ino < AUFS_FIRST_INO))
34313 +                               continue;
34314 +
34315 +                       xib_calc_bit(*ino, &pindex, &bit);
34316 +                       AuDebugOn(page_bits <= bit);
34317 +                       err = xib_pindex(sb, pindex);
34318 +                       if (!err)
34319 +                               set_bit(bit, p);
34320 +                       else
34321 +                               goto out;
34322 +               }
34323 +       }
34324 +
34325 +out:
34326 +       return err;
34327 +}
34328 +
34329 +static int xib_restore(struct super_block *sb)
34330 +{
34331 +       int err;
34332 +       aufs_bindex_t bindex, bend;
34333 +       void *page;
34334 +
34335 +       err = -ENOMEM;
34336 +       page = (void *)__get_free_page(GFP_NOFS);
34337 +       if (unlikely(!page))
34338 +               goto out;
34339 +
34340 +       err = 0;
34341 +       bend = au_sbend(sb);
34342 +       for (bindex = 0; !err && bindex <= bend; bindex++)
34343 +               if (!bindex || is_sb_shared(sb, bindex, bindex - 1) < 0)
34344 +                       err = do_xib_restore
34345 +                               (sb, au_sbr(sb, bindex)->br_xino.xi_file, page);
34346 +               else
34347 +                       AuDbg("b%d\n", bindex);
34348 +       free_page((unsigned long)page);
34349 +
34350 +out:
34351 +       return err;
34352 +}
34353 +
34354 +int au_xib_trunc(struct super_block *sb)
34355 +{
34356 +       int err;
34357 +       ssize_t sz;
34358 +       loff_t pos;
34359 +       struct au_xino_lock_dir ldir;
34360 +       struct au_sbinfo *sbinfo;
34361 +       unsigned long *p;
34362 +       struct file *file;
34363 +
34364 +       SiMustWriteLock(sb);
34365 +
34366 +       err = 0;
34367 +       sbinfo = au_sbi(sb);
34368 +       if (!au_opt_test(sbinfo->si_mntflags, XINO))
34369 +               goto out;
34370 +
34371 +       file = sbinfo->si_xib;
34372 +       if (vfsub_f_size_read(file) <= PAGE_SIZE)
34373 +               goto out;
34374 +
34375 +       au_xino_lock_dir(sb, file, &ldir);
34376 +       /* mnt_want_write() is unnecessary here */
34377 +       file = au_xino_create2(sbinfo->si_xib, NULL);
34378 +       au_xino_unlock_dir(&ldir);
34379 +       err = PTR_ERR(file);
34380 +       if (IS_ERR(file))
34381 +               goto out;
34382 +       fput(sbinfo->si_xib);
34383 +       sbinfo->si_xib = file;
34384 +
34385 +       p = sbinfo->si_xib_buf;
34386 +       memset(p, 0, PAGE_SIZE);
34387 +       pos = 0;
34388 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xib, p, PAGE_SIZE, &pos);
34389 +       if (unlikely(sz != PAGE_SIZE)) {
34390 +               err = sz;
34391 +               AuIOErr("err %d\n", err);
34392 +               if (sz >= 0)
34393 +                       err = -EIO;
34394 +               goto out;
34395 +       }
34396 +
34397 +       mutex_lock(&sbinfo->si_xib_mtx);
34398 +       /* mnt_want_write() is unnecessary here */
34399 +       err = xib_restore(sb);
34400 +       mutex_unlock(&sbinfo->si_xib_mtx);
34401 +
34402 +out:
34403 +       return err;
34404 +}
34405 +
34406 +/* ---------------------------------------------------------------------- */
34407 +
34408 +/*
34409 + * xino mount option handlers
34410 + */
34411 +static au_readf_t find_readf(struct file *h_file)
34412 +{
34413 +       const struct file_operations *fop = h_file->f_op;
34414 +
34415 +       if (fop->read)
34416 +               return fop->read;
34417 +       if (fop->aio_read)
34418 +               return do_sync_read;
34419 +       if (fop->read_iter)
34420 +               return new_sync_read;
34421 +       return ERR_PTR(-ENOSYS);
34422 +}
34423 +
34424 +static au_writef_t find_writef(struct file *h_file)
34425 +{
34426 +       const struct file_operations *fop = h_file->f_op;
34427 +
34428 +       if (fop->write)
34429 +               return fop->write;
34430 +       if (fop->aio_write)
34431 +               return do_sync_write;
34432 +       if (fop->write_iter)
34433 +               return new_sync_write;
34434 +       return ERR_PTR(-ENOSYS);
34435 +}
34436 +
34437 +/* xino bitmap */
34438 +static void xino_clear_xib(struct super_block *sb)
34439 +{
34440 +       struct au_sbinfo *sbinfo;
34441 +
34442 +       SiMustWriteLock(sb);
34443 +
34444 +       sbinfo = au_sbi(sb);
34445 +       sbinfo->si_xread = NULL;
34446 +       sbinfo->si_xwrite = NULL;
34447 +       if (sbinfo->si_xib)
34448 +               fput(sbinfo->si_xib);
34449 +       sbinfo->si_xib = NULL;
34450 +       free_page((unsigned long)sbinfo->si_xib_buf);
34451 +       sbinfo->si_xib_buf = NULL;
34452 +}
34453 +
34454 +static int au_xino_set_xib(struct super_block *sb, struct file *base)
34455 +{
34456 +       int err;
34457 +       loff_t pos;
34458 +       struct au_sbinfo *sbinfo;
34459 +       struct file *file;
34460 +
34461 +       SiMustWriteLock(sb);
34462 +
34463 +       sbinfo = au_sbi(sb);
34464 +       file = au_xino_create2(base, sbinfo->si_xib);
34465 +       err = PTR_ERR(file);
34466 +       if (IS_ERR(file))
34467 +               goto out;
34468 +       if (sbinfo->si_xib)
34469 +               fput(sbinfo->si_xib);
34470 +       sbinfo->si_xib = file;
34471 +       sbinfo->si_xread = find_readf(file);
34472 +       sbinfo->si_xwrite = find_writef(file);
34473 +
34474 +       err = -ENOMEM;
34475 +       if (!sbinfo->si_xib_buf)
34476 +               sbinfo->si_xib_buf = (void *)get_zeroed_page(GFP_NOFS);
34477 +       if (unlikely(!sbinfo->si_xib_buf))
34478 +               goto out_unset;
34479 +
34480 +       sbinfo->si_xib_last_pindex = 0;
34481 +       sbinfo->si_xib_next_bit = 0;
34482 +       if (vfsub_f_size_read(file) < PAGE_SIZE) {
34483 +               pos = 0;
34484 +               err = xino_fwrite(sbinfo->si_xwrite, file, sbinfo->si_xib_buf,
34485 +                                 PAGE_SIZE, &pos);
34486 +               if (unlikely(err != PAGE_SIZE))
34487 +                       goto out_free;
34488 +       }
34489 +       err = 0;
34490 +       goto out; /* success */
34491 +
34492 +out_free:
34493 +       free_page((unsigned long)sbinfo->si_xib_buf);
34494 +       sbinfo->si_xib_buf = NULL;
34495 +       if (err >= 0)
34496 +               err = -EIO;
34497 +out_unset:
34498 +       fput(sbinfo->si_xib);
34499 +       sbinfo->si_xib = NULL;
34500 +       sbinfo->si_xread = NULL;
34501 +       sbinfo->si_xwrite = NULL;
34502 +out:
34503 +       return err;
34504 +}
34505 +
34506 +/* xino for each branch */
34507 +static void xino_clear_br(struct super_block *sb)
34508 +{
34509 +       aufs_bindex_t bindex, bend;
34510 +       struct au_branch *br;
34511 +
34512 +       bend = au_sbend(sb);
34513 +       for (bindex = 0; bindex <= bend; bindex++) {
34514 +               br = au_sbr(sb, bindex);
34515 +               if (!br || !br->br_xino.xi_file)
34516 +                       continue;
34517 +
34518 +               fput(br->br_xino.xi_file);
34519 +               br->br_xino.xi_file = NULL;
34520 +       }
34521 +}
34522 +
34523 +static int au_xino_set_br(struct super_block *sb, struct file *base)
34524 +{
34525 +       int err;
34526 +       ino_t ino;
34527 +       aufs_bindex_t bindex, bend, bshared;
34528 +       struct {
34529 +               struct file *old, *new;
34530 +       } *fpair, *p;
34531 +       struct au_branch *br;
34532 +       struct inode *inode;
34533 +       au_writef_t writef;
34534 +
34535 +       SiMustWriteLock(sb);
34536 +
34537 +       err = -ENOMEM;
34538 +       bend = au_sbend(sb);
34539 +       fpair = kcalloc(bend + 1, sizeof(*fpair), GFP_NOFS);
34540 +       if (unlikely(!fpair))
34541 +               goto out;
34542 +
34543 +       inode = sb->s_root->d_inode;
34544 +       ino = AUFS_ROOT_INO;
34545 +       writef = au_sbi(sb)->si_xwrite;
34546 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++) {
34547 +               br = au_sbr(sb, bindex);
34548 +               bshared = is_sb_shared(sb, bindex, bindex - 1);
34549 +               if (bshared >= 0) {
34550 +                       /* shared xino */
34551 +                       *p = fpair[bshared];
34552 +                       get_file(p->new);
34553 +               }
34554 +
34555 +               if (!p->new) {
34556 +                       /* new xino */
34557 +                       p->old = br->br_xino.xi_file;
34558 +                       p->new = au_xino_create2(base, br->br_xino.xi_file);
34559 +                       err = PTR_ERR(p->new);
34560 +                       if (IS_ERR(p->new)) {
34561 +                               p->new = NULL;
34562 +                               goto out_pair;
34563 +                       }
34564 +               }
34565 +
34566 +               err = au_xino_do_write(writef, p->new,
34567 +                                      au_h_iptr(inode, bindex)->i_ino, ino);
34568 +               if (unlikely(err))
34569 +                       goto out_pair;
34570 +       }
34571 +
34572 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++) {
34573 +               br = au_sbr(sb, bindex);
34574 +               if (br->br_xino.xi_file)
34575 +                       fput(br->br_xino.xi_file);
34576 +               get_file(p->new);
34577 +               br->br_xino.xi_file = p->new;
34578 +       }
34579 +
34580 +out_pair:
34581 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++)
34582 +               if (p->new)
34583 +                       fput(p->new);
34584 +               else
34585 +                       break;
34586 +       kfree(fpair);
34587 +out:
34588 +       return err;
34589 +}
34590 +
34591 +void au_xino_clr(struct super_block *sb)
34592 +{
34593 +       struct au_sbinfo *sbinfo;
34594 +
34595 +       au_xigen_clr(sb);
34596 +       xino_clear_xib(sb);
34597 +       xino_clear_br(sb);
34598 +       sbinfo = au_sbi(sb);
34599 +       /* lvalue, do not call au_mntflags() */
34600 +       au_opt_clr(sbinfo->si_mntflags, XINO);
34601 +}
34602 +
34603 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount)
34604 +{
34605 +       int err, skip;
34606 +       struct dentry *parent, *cur_parent;
34607 +       struct qstr *dname, *cur_name;
34608 +       struct file *cur_xino;
34609 +       struct inode *dir;
34610 +       struct au_sbinfo *sbinfo;
34611 +
34612 +       SiMustWriteLock(sb);
34613 +
34614 +       err = 0;
34615 +       sbinfo = au_sbi(sb);
34616 +       parent = dget_parent(xino->file->f_path.dentry);
34617 +       if (remount) {
34618 +               skip = 0;
34619 +               dname = &xino->file->f_path.dentry->d_name;
34620 +               cur_xino = sbinfo->si_xib;
34621 +               if (cur_xino) {
34622 +                       cur_parent = dget_parent(cur_xino->f_path.dentry);
34623 +                       cur_name = &cur_xino->f_path.dentry->d_name;
34624 +                       skip = (cur_parent == parent
34625 +                               && au_qstreq(dname, cur_name));
34626 +                       dput(cur_parent);
34627 +               }
34628 +               if (skip)
34629 +                       goto out;
34630 +       }
34631 +
34632 +       au_opt_set(sbinfo->si_mntflags, XINO);
34633 +       dir = parent->d_inode;
34634 +       mutex_lock_nested(&dir->i_mutex, AuLsc_I_PARENT);
34635 +       /* mnt_want_write() is unnecessary here */
34636 +       err = au_xino_set_xib(sb, xino->file);
34637 +       if (!err)
34638 +               err = au_xigen_set(sb, xino->file);
34639 +       if (!err)
34640 +               err = au_xino_set_br(sb, xino->file);
34641 +       mutex_unlock(&dir->i_mutex);
34642 +       if (!err)
34643 +               goto out; /* success */
34644 +
34645 +       /* reset all */
34646 +       AuIOErr("failed creating xino(%d).\n", err);
34647 +       au_xigen_clr(sb);
34648 +       xino_clear_xib(sb);
34649 +
34650 +out:
34651 +       dput(parent);
34652 +       return err;
34653 +}
34654 +
34655 +/* ---------------------------------------------------------------------- */
34656 +
34657 +/*
34658 + * create a xinofile at the default place/path.
34659 + */
34660 +struct file *au_xino_def(struct super_block *sb)
34661 +{
34662 +       struct file *file;
34663 +       char *page, *p;
34664 +       struct au_branch *br;
34665 +       struct super_block *h_sb;
34666 +       struct path path;
34667 +       aufs_bindex_t bend, bindex, bwr;
34668 +
34669 +       br = NULL;
34670 +       bend = au_sbend(sb);
34671 +       bwr = -1;
34672 +       for (bindex = 0; bindex <= bend; bindex++) {
34673 +               br = au_sbr(sb, bindex);
34674 +               if (au_br_writable(br->br_perm)
34675 +                   && !au_test_fs_bad_xino(au_br_sb(br))) {
34676 +                       bwr = bindex;
34677 +                       break;
34678 +               }
34679 +       }
34680 +
34681 +       if (bwr >= 0) {
34682 +               file = ERR_PTR(-ENOMEM);
34683 +               page = (void *)__get_free_page(GFP_NOFS);
34684 +               if (unlikely(!page))
34685 +                       goto out;
34686 +               path.mnt = au_br_mnt(br);
34687 +               path.dentry = au_h_dptr(sb->s_root, bwr);
34688 +               p = d_path(&path, page, PATH_MAX - sizeof(AUFS_XINO_FNAME));
34689 +               file = (void *)p;
34690 +               if (!IS_ERR(p)) {
34691 +                       strcat(p, "/" AUFS_XINO_FNAME);
34692 +                       AuDbg("%s\n", p);
34693 +                       file = au_xino_create(sb, p, /*silent*/0);
34694 +                       if (!IS_ERR(file))
34695 +                               au_xino_brid_set(sb, br->br_id);
34696 +               }
34697 +               free_page((unsigned long)page);
34698 +       } else {
34699 +               file = au_xino_create(sb, AUFS_XINO_DEFPATH, /*silent*/0);
34700 +               if (IS_ERR(file))
34701 +                       goto out;
34702 +               h_sb = file->f_path.dentry->d_sb;
34703 +               if (unlikely(au_test_fs_bad_xino(h_sb))) {
34704 +                       pr_err("xino doesn't support %s(%s)\n",
34705 +                              AUFS_XINO_DEFPATH, au_sbtype(h_sb));
34706 +                       fput(file);
34707 +                       file = ERR_PTR(-EINVAL);
34708 +               }
34709 +               if (!IS_ERR(file))
34710 +                       au_xino_brid_set(sb, -1);
34711 +       }
34712 +
34713 +out:
34714 +       return file;
34715 +}
34716 +
34717 +/* ---------------------------------------------------------------------- */
34718 +
34719 +int au_xino_path(struct seq_file *seq, struct file *file)
34720 +{
34721 +       int err;
34722 +
34723 +       err = au_seq_path(seq, &file->f_path);
34724 +       if (unlikely(err < 0))
34725 +               goto out;
34726 +
34727 +       err = 0;
34728 +#define Deleted "\\040(deleted)"
34729 +       seq->count -= sizeof(Deleted) - 1;
34730 +       AuDebugOn(memcmp(seq->buf + seq->count, Deleted,
34731 +                        sizeof(Deleted) - 1));
34732 +#undef Deleted
34733 +
34734 +out:
34735 +       return err;
34736 +}
34737 diff -urN /usr/share/empty/include/uapi/linux/aufs_type.h linux/include/uapi/linux/aufs_type.h
34738 --- /usr/share/empty/include/uapi/linux/aufs_type.h     1970-01-01 01:00:00.000000000 +0100
34739 +++ linux/include/uapi/linux/aufs_type.h        2015-03-27 21:56:35.470128334 +0100
34740 @@ -0,0 +1,419 @@
34741 +/*
34742 + * Copyright (C) 2005-2015 Junjiro R. Okajima
34743 + *
34744 + * This program, aufs is free software; you can redistribute it and/or modify
34745 + * it under the terms of the GNU General Public License as published by
34746 + * the Free Software Foundation; either version 2 of the License, or
34747 + * (at your option) any later version.
34748 + *
34749 + * This program is distributed in the hope that it will be useful,
34750 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
34751 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
34752 + * GNU General Public License for more details.
34753 + *
34754 + * You should have received a copy of the GNU General Public License
34755 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
34756 + */
34757 +
34758 +#ifndef __AUFS_TYPE_H__
34759 +#define __AUFS_TYPE_H__
34760 +
34761 +#define AUFS_NAME      "aufs"
34762 +
34763 +#ifdef __KERNEL__
34764 +/*
34765 + * define it before including all other headers.
34766 + * sched.h may use pr_* macros before defining "current", so define the
34767 + * no-current version first, and re-define later.
34768 + */
34769 +#define pr_fmt(fmt)    AUFS_NAME " %s:%d: " fmt, __func__, __LINE__
34770 +#include <linux/sched.h>
34771 +#undef pr_fmt
34772 +#define pr_fmt(fmt) \
34773 +               AUFS_NAME " %s:%d:%.*s[%d]: " fmt, __func__, __LINE__, \
34774 +               (int)sizeof(current->comm), current->comm, current->pid
34775 +#else
34776 +#include <stdint.h>
34777 +#include <sys/types.h>
34778 +#endif /* __KERNEL__ */
34779 +
34780 +#include <linux/limits.h>
34781 +
34782 +#define AUFS_VERSION   "3.19-20150323"
34783 +
34784 +/* todo? move this to linux-2.6.19/include/magic.h */
34785 +#define AUFS_SUPER_MAGIC       ('a' << 24 | 'u' << 16 | 'f' << 8 | 's')
34786 +
34787 +/* ---------------------------------------------------------------------- */
34788 +
34789 +#ifdef CONFIG_AUFS_BRANCH_MAX_127
34790 +typedef int8_t aufs_bindex_t;
34791 +#define AUFS_BRANCH_MAX 127
34792 +#else
34793 +typedef int16_t aufs_bindex_t;
34794 +#ifdef CONFIG_AUFS_BRANCH_MAX_511
34795 +#define AUFS_BRANCH_MAX 511
34796 +#elif defined(CONFIG_AUFS_BRANCH_MAX_1023)
34797 +#define AUFS_BRANCH_MAX 1023
34798 +#elif defined(CONFIG_AUFS_BRANCH_MAX_32767)
34799 +#define AUFS_BRANCH_MAX 32767
34800 +#endif
34801 +#endif
34802 +
34803 +#ifdef __KERNEL__
34804 +#ifndef AUFS_BRANCH_MAX
34805 +#error unknown CONFIG_AUFS_BRANCH_MAX value
34806 +#endif
34807 +#endif /* __KERNEL__ */
34808 +
34809 +/* ---------------------------------------------------------------------- */
34810 +
34811 +#define AUFS_FSTYPE            AUFS_NAME
34812 +
34813 +#define AUFS_ROOT_INO          2
34814 +#define AUFS_FIRST_INO         11
34815 +
34816 +#define AUFS_WH_PFX            ".wh."
34817 +#define AUFS_WH_PFX_LEN                ((int)sizeof(AUFS_WH_PFX) - 1)
34818 +#define AUFS_WH_TMP_LEN                4
34819 +/* a limit for rmdir/rename a dir and copyup */
34820 +#define AUFS_MAX_NAMELEN       (NAME_MAX \
34821 +                               - AUFS_WH_PFX_LEN * 2   /* doubly whiteouted */\
34822 +                               - 1                     /* dot */\
34823 +                               - AUFS_WH_TMP_LEN)      /* hex */
34824 +#define AUFS_XINO_FNAME                "." AUFS_NAME ".xino"
34825 +#define AUFS_XINO_DEFPATH      "/tmp/" AUFS_XINO_FNAME
34826 +#define AUFS_XINO_DEF_SEC      30 /* seconds */
34827 +#define AUFS_XINO_DEF_TRUNC    45 /* percentage */
34828 +#define AUFS_DIRWH_DEF         3
34829 +#define AUFS_RDCACHE_DEF       10 /* seconds */
34830 +#define AUFS_RDCACHE_MAX       3600 /* seconds */
34831 +#define AUFS_RDBLK_DEF         512 /* bytes */
34832 +#define AUFS_RDHASH_DEF                32
34833 +#define AUFS_WKQ_NAME          AUFS_NAME "d"
34834 +#define AUFS_MFS_DEF_SEC       30 /* seconds */
34835 +#define AUFS_MFS_MAX_SEC       3600 /* seconds */
34836 +#define AUFS_FHSM_CACHE_DEF_SEC        30 /* seconds */
34837 +#define AUFS_PLINK_WARN                50 /* number of plinks in a single bucket */
34838 +
34839 +/* pseudo-link maintenace under /proc */
34840 +#define AUFS_PLINK_MAINT_NAME  "plink_maint"
34841 +#define AUFS_PLINK_MAINT_DIR   "fs/" AUFS_NAME
34842 +#define AUFS_PLINK_MAINT_PATH  AUFS_PLINK_MAINT_DIR "/" AUFS_PLINK_MAINT_NAME
34843 +
34844 +#define AUFS_DIROPQ_NAME       AUFS_WH_PFX ".opq" /* whiteouted doubly */
34845 +#define AUFS_WH_DIROPQ         AUFS_WH_PFX AUFS_DIROPQ_NAME
34846 +
34847 +#define AUFS_BASE_NAME         AUFS_WH_PFX AUFS_NAME
34848 +#define AUFS_PLINKDIR_NAME     AUFS_WH_PFX "plnk"
34849 +#define AUFS_ORPHDIR_NAME      AUFS_WH_PFX "orph"
34850 +
34851 +/* doubly whiteouted */
34852 +#define AUFS_WH_BASE           AUFS_WH_PFX AUFS_BASE_NAME
34853 +#define AUFS_WH_PLINKDIR       AUFS_WH_PFX AUFS_PLINKDIR_NAME
34854 +#define AUFS_WH_ORPHDIR                AUFS_WH_PFX AUFS_ORPHDIR_NAME
34855 +
34856 +/* branch permissions and attributes */
34857 +#define AUFS_BRPERM_RW         "rw"
34858 +#define AUFS_BRPERM_RO         "ro"
34859 +#define AUFS_BRPERM_RR         "rr"
34860 +#define AUFS_BRATTR_COO_REG    "coo_reg"
34861 +#define AUFS_BRATTR_COO_ALL    "coo_all"
34862 +#define AUFS_BRATTR_FHSM       "fhsm"
34863 +#define AUFS_BRATTR_UNPIN      "unpin"
34864 +#define AUFS_BRATTR_ICEX       "icex"
34865 +#define AUFS_BRATTR_ICEX_SEC   "icexsec"
34866 +#define AUFS_BRATTR_ICEX_SYS   "icexsys"
34867 +#define AUFS_BRATTR_ICEX_TR    "icextr"
34868 +#define AUFS_BRATTR_ICEX_USR   "icexusr"
34869 +#define AUFS_BRATTR_ICEX_OTH   "icexoth"
34870 +#define AUFS_BRRATTR_WH                "wh"
34871 +#define AUFS_BRWATTR_NLWH      "nolwh"
34872 +#define AUFS_BRWATTR_MOO       "moo"
34873 +
34874 +#define AuBrPerm_RW            1               /* writable, hardlinkable wh */
34875 +#define AuBrPerm_RO            (1 << 1)        /* readonly */
34876 +#define AuBrPerm_RR            (1 << 2)        /* natively readonly */
34877 +#define AuBrPerm_Mask          (AuBrPerm_RW | AuBrPerm_RO | AuBrPerm_RR)
34878 +
34879 +#define AuBrAttr_COO_REG       (1 << 3)        /* copy-up on open */
34880 +#define AuBrAttr_COO_ALL       (1 << 4)
34881 +#define AuBrAttr_COO_Mask      (AuBrAttr_COO_REG | AuBrAttr_COO_ALL)
34882 +
34883 +#define AuBrAttr_FHSM          (1 << 5)        /* file-based hsm */
34884 +#define AuBrAttr_UNPIN         (1 << 6)        /* rename-able top dir of
34885 +                                                  branch. meaningless since
34886 +                                                  linux-3.18-rc1 */
34887 +
34888 +/* ignore error in copying XATTR */
34889 +#define AuBrAttr_ICEX_SEC      (1 << 7)
34890 +#define AuBrAttr_ICEX_SYS      (1 << 8)
34891 +#define AuBrAttr_ICEX_TR       (1 << 9)
34892 +#define AuBrAttr_ICEX_USR      (1 << 10)
34893 +#define AuBrAttr_ICEX_OTH      (1 << 11)
34894 +#define AuBrAttr_ICEX          (AuBrAttr_ICEX_SEC      \
34895 +                                | AuBrAttr_ICEX_SYS    \
34896 +                                | AuBrAttr_ICEX_TR     \
34897 +                                | AuBrAttr_ICEX_USR    \
34898 +                                | AuBrAttr_ICEX_OTH)
34899 +
34900 +#define AuBrRAttr_WH           (1 << 12)       /* whiteout-able */
34901 +#define AuBrRAttr_Mask         AuBrRAttr_WH
34902 +
34903 +#define AuBrWAttr_NoLinkWH     (1 << 13)       /* un-hardlinkable whiteouts */
34904 +#define AuBrWAttr_MOO          (1 << 14)       /* move-up on open */
34905 +#define AuBrWAttr_Mask         (AuBrWAttr_NoLinkWH | AuBrWAttr_MOO)
34906 +
34907 +#define AuBrAttr_CMOO_Mask     (AuBrAttr_COO_Mask | AuBrWAttr_MOO)
34908 +
34909 +/* #warning test userspace */
34910 +#ifdef __KERNEL__
34911 +#ifndef CONFIG_AUFS_FHSM
34912 +#undef AuBrAttr_FHSM
34913 +#define AuBrAttr_FHSM          0
34914 +#endif
34915 +#ifndef CONFIG_AUFS_XATTR
34916 +#undef AuBrAttr_ICEX
34917 +#define AuBrAttr_ICEX          0
34918 +#undef AuBrAttr_ICEX_SEC
34919 +#define AuBrAttr_ICEX_SEC      0
34920 +#undef AuBrAttr_ICEX_SYS
34921 +#define AuBrAttr_ICEX_SYS      0
34922 +#undef AuBrAttr_ICEX_TR
34923 +#define AuBrAttr_ICEX_TR       0
34924 +#undef AuBrAttr_ICEX_USR
34925 +#define AuBrAttr_ICEX_USR      0
34926 +#undef AuBrAttr_ICEX_OTH
34927 +#define AuBrAttr_ICEX_OTH      0
34928 +#endif
34929 +#endif
34930 +
34931 +/* the longest combination */
34932 +/* AUFS_BRATTR_ICEX and AUFS_BRATTR_ICEX_TR don't affect here */
34933 +#define AuBrPermStrSz  sizeof(AUFS_BRPERM_RW                   \
34934 +                              "+" AUFS_BRATTR_COO_REG          \
34935 +                              "+" AUFS_BRATTR_FHSM             \
34936 +                              "+" AUFS_BRATTR_UNPIN            \
34937 +                              "+" AUFS_BRATTR_ICEX_SEC \
34938 +                              "+" AUFS_BRATTR_ICEX_SYS \
34939 +                              "+" AUFS_BRATTR_ICEX_USR \
34940 +                              "+" AUFS_BRATTR_ICEX_OTH \
34941 +                              "+" AUFS_BRWATTR_NLWH)
34942 +
34943 +typedef struct {
34944 +       char a[AuBrPermStrSz];
34945 +} au_br_perm_str_t;
34946 +
34947 +static inline int au_br_writable(int brperm)
34948 +{
34949 +       return brperm & AuBrPerm_RW;
34950 +}
34951 +
34952 +static inline int au_br_whable(int brperm)
34953 +{
34954 +       return brperm & (AuBrPerm_RW | AuBrRAttr_WH);
34955 +}
34956 +
34957 +static inline int au_br_wh_linkable(int brperm)
34958 +{
34959 +       return !(brperm & AuBrWAttr_NoLinkWH);
34960 +}
34961 +
34962 +static inline int au_br_cmoo(int brperm)
34963 +{
34964 +       return brperm & AuBrAttr_CMOO_Mask;
34965 +}
34966 +
34967 +static inline int au_br_fhsm(int brperm)
34968 +{
34969 +       return brperm & AuBrAttr_FHSM;
34970 +}
34971 +
34972 +/* ---------------------------------------------------------------------- */
34973 +
34974 +/* ioctl */
34975 +enum {
34976 +       /* readdir in userspace */
34977 +       AuCtl_RDU,
34978 +       AuCtl_RDU_INO,
34979 +
34980 +       AuCtl_WBR_FD,   /* pathconf wrapper */
34981 +       AuCtl_IBUSY,    /* busy inode */
34982 +       AuCtl_MVDOWN,   /* move-down */
34983 +       AuCtl_BR,       /* info about branches */
34984 +       AuCtl_FHSM_FD   /* connection for fhsm */
34985 +};
34986 +
34987 +/* borrowed from linux/include/linux/kernel.h */
34988 +#ifndef ALIGN
34989 +#define ALIGN(x, a)            __ALIGN_MASK(x, (typeof(x))(a)-1)
34990 +#define __ALIGN_MASK(x, mask)  (((x)+(mask))&~(mask))
34991 +#endif
34992 +
34993 +/* borrowed from linux/include/linux/compiler-gcc3.h */
34994 +#ifndef __aligned
34995 +#define __aligned(x)                   __attribute__((aligned(x)))
34996 +#endif
34997 +
34998 +#ifdef __KERNEL__
34999 +#ifndef __packed
35000 +#define __packed                       __attribute__((packed))
35001 +#endif
35002 +#endif
35003 +
35004 +struct au_rdu_cookie {
35005 +       uint64_t        h_pos;
35006 +       int16_t         bindex;
35007 +       uint8_t         flags;
35008 +       uint8_t         pad;
35009 +       uint32_t        generation;
35010 +} __aligned(8);
35011 +
35012 +struct au_rdu_ent {
35013 +       uint64_t        ino;
35014 +       int16_t         bindex;
35015 +       uint8_t         type;
35016 +       uint8_t         nlen;
35017 +       uint8_t         wh;
35018 +       char            name[0];
35019 +} __aligned(8);
35020 +
35021 +static inline int au_rdu_len(int nlen)
35022 +{
35023 +       /* include the terminating NULL */
35024 +       return ALIGN(sizeof(struct au_rdu_ent) + nlen + 1,
35025 +                    sizeof(uint64_t));
35026 +}
35027 +
35028 +union au_rdu_ent_ul {
35029 +       struct au_rdu_ent __user        *e;
35030 +       uint64_t                        ul;
35031 +};
35032 +
35033 +enum {
35034 +       AufsCtlRduV_SZ,
35035 +       AufsCtlRduV_End
35036 +};
35037 +
35038 +struct aufs_rdu {
35039 +       /* input */
35040 +       union {
35041 +               uint64_t        sz;     /* AuCtl_RDU */
35042 +               uint64_t        nent;   /* AuCtl_RDU_INO */
35043 +       };
35044 +       union au_rdu_ent_ul     ent;
35045 +       uint16_t                verify[AufsCtlRduV_End];
35046 +
35047 +       /* input/output */
35048 +       uint32_t                blk;
35049 +
35050 +       /* output */
35051 +       union au_rdu_ent_ul     tail;
35052 +       /* number of entries which were added in a single call */
35053 +       uint64_t                rent;
35054 +       uint8_t                 full;
35055 +       uint8_t                 shwh;
35056 +
35057 +       struct au_rdu_cookie    cookie;
35058 +} __aligned(8);
35059 +
35060 +/* ---------------------------------------------------------------------- */
35061 +
35062 +struct aufs_wbr_fd {
35063 +       uint32_t        oflags;
35064 +       int16_t         brid;
35065 +} __aligned(8);
35066 +
35067 +/* ---------------------------------------------------------------------- */
35068 +
35069 +struct aufs_ibusy {
35070 +       uint64_t        ino, h_ino;
35071 +       int16_t         bindex;
35072 +} __aligned(8);
35073 +
35074 +/* ---------------------------------------------------------------------- */
35075 +
35076 +/* error code for move-down */
35077 +/* the actual message strings are implemented in aufs-util.git */
35078 +enum {
35079 +       EAU_MVDOWN_OPAQUE = 1,
35080 +       EAU_MVDOWN_WHITEOUT,
35081 +       EAU_MVDOWN_UPPER,
35082 +       EAU_MVDOWN_BOTTOM,
35083 +       EAU_MVDOWN_NOUPPER,
35084 +       EAU_MVDOWN_NOLOWERBR,
35085 +       EAU_Last
35086 +};
35087 +
35088 +/* flags for move-down */
35089 +#define AUFS_MVDOWN_DMSG       1
35090 +#define AUFS_MVDOWN_OWLOWER    (1 << 1)        /* overwrite lower */
35091 +#define AUFS_MVDOWN_KUPPER     (1 << 2)        /* keep upper */
35092 +#define AUFS_MVDOWN_ROLOWER    (1 << 3)        /* do even if lower is RO */
35093 +#define AUFS_MVDOWN_ROLOWER_R  (1 << 4)        /* did on lower RO */
35094 +#define AUFS_MVDOWN_ROUPPER    (1 << 5)        /* do even if upper is RO */
35095 +#define AUFS_MVDOWN_ROUPPER_R  (1 << 6)        /* did on upper RO */
35096 +#define AUFS_MVDOWN_BRID_UPPER (1 << 7)        /* upper brid */
35097 +#define AUFS_MVDOWN_BRID_LOWER (1 << 8)        /* lower brid */
35098 +#define AUFS_MVDOWN_FHSM_LOWER (1 << 9)        /* find fhsm attr for lower */
35099 +#define AUFS_MVDOWN_STFS       (1 << 10)       /* req. stfs */
35100 +#define AUFS_MVDOWN_STFS_FAILED        (1 << 11)       /* output: stfs is unusable */
35101 +#define AUFS_MVDOWN_BOTTOM     (1 << 12)       /* output: no more lowers */
35102 +
35103 +/* index for move-down */
35104 +enum {
35105 +       AUFS_MVDOWN_UPPER,
35106 +       AUFS_MVDOWN_LOWER,
35107 +       AUFS_MVDOWN_NARRAY
35108 +};
35109 +
35110 +/*
35111 + * additional info of move-down
35112 + * number of free blocks and inodes.
35113 + * subset of struct kstatfs, but smaller and always 64bit.
35114 + */
35115 +struct aufs_stfs {
35116 +       uint64_t        f_blocks;
35117 +       uint64_t        f_bavail;
35118 +       uint64_t        f_files;
35119 +       uint64_t        f_ffree;
35120 +};
35121 +
35122 +struct aufs_stbr {
35123 +       int16_t                 brid;   /* optional input */
35124 +       int16_t                 bindex; /* output */
35125 +       struct aufs_stfs        stfs;   /* output when AUFS_MVDOWN_STFS set */
35126 +} __aligned(8);
35127 +
35128 +struct aufs_mvdown {
35129 +       uint32_t                flags;                  /* input/output */
35130 +       struct aufs_stbr        stbr[AUFS_MVDOWN_NARRAY]; /* input/output */
35131 +       int8_t                  au_errno;               /* output */
35132 +} __aligned(8);
35133 +
35134 +/* ---------------------------------------------------------------------- */
35135 +
35136 +union aufs_brinfo {
35137 +       /* PATH_MAX may differ between kernel-space and user-space */
35138 +       char    _spacer[4096];
35139 +       struct {
35140 +               int16_t id;
35141 +               int     perm;
35142 +               char    path[0];
35143 +       };
35144 +} __aligned(8);
35145 +
35146 +/* ---------------------------------------------------------------------- */
35147 +
35148 +#define AuCtlType              'A'
35149 +#define AUFS_CTL_RDU           _IOWR(AuCtlType, AuCtl_RDU, struct aufs_rdu)
35150 +#define AUFS_CTL_RDU_INO       _IOWR(AuCtlType, AuCtl_RDU_INO, struct aufs_rdu)
35151 +#define AUFS_CTL_WBR_FD                _IOW(AuCtlType, AuCtl_WBR_FD, \
35152 +                                    struct aufs_wbr_fd)
35153 +#define AUFS_CTL_IBUSY         _IOWR(AuCtlType, AuCtl_IBUSY, struct aufs_ibusy)
35154 +#define AUFS_CTL_MVDOWN                _IOWR(AuCtlType, AuCtl_MVDOWN, \
35155 +                                     struct aufs_mvdown)
35156 +#define AUFS_CTL_BRINFO                _IOW(AuCtlType, AuCtl_BR, union aufs_brinfo)
35157 +#define AUFS_CTL_FHSM_FD       _IOW(AuCtlType, AuCtl_FHSM_FD, int)
35158 +
35159 +#endif /* __AUFS_TYPE_H__ */
35160 aufs3.19 loopback patch
35161
35162 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
35163 index 30efd68..2a9b789 100644
35164 --- a/drivers/block/loop.c
35165 +++ b/drivers/block/loop.c
35166 @@ -514,7 +514,7 @@ out:
35167  }
35168  
35169  struct switch_request {
35170 -       struct file *file;
35171 +       struct file *file, *virt_file;
35172         struct completion wait;
35173  };
35174  
35175 @@ -576,7 +576,8 @@ static int loop_thread(void *data)
35176   * First it needs to flush existing IO, it does this by sending a magic
35177   * BIO down the pipe. The completion of this BIO does the actual switch.
35178   */
35179 -static int loop_switch(struct loop_device *lo, struct file *file)
35180 +static int loop_switch(struct loop_device *lo, struct file *file,
35181 +                      struct file *virt_file)
35182  {
35183         struct switch_request w;
35184         struct bio *bio = bio_alloc(GFP_KERNEL, 0);
35185 @@ -584,6 +585,7 @@ static int loop_switch(struct loop_device *lo, struct file *file)
35186                 return -ENOMEM;
35187         init_completion(&w.wait);
35188         w.file = file;
35189 +       w.virt_file = virt_file;
35190         bio->bi_private = &w;
35191         bio->bi_bdev = NULL;
35192         loop_make_request(lo->lo_queue, bio);
35193 @@ -600,7 +602,7 @@ static int loop_flush(struct loop_device *lo)
35194         if (!lo->lo_thread)
35195                 return 0;
35196  
35197 -       return loop_switch(lo, NULL);
35198 +       return loop_switch(lo, NULL, NULL);
35199  }
35200  
35201  /*
35202 @@ -619,6 +621,7 @@ static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
35203         mapping = file->f_mapping;
35204         mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
35205         lo->lo_backing_file = file;
35206 +       lo->lo_backing_virt_file = p->virt_file;
35207         lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
35208                 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
35209         lo->old_gfp_mask = mapping_gfp_mask(mapping);
35210 @@ -627,6 +630,13 @@ out:
35211         complete(&p->wait);
35212  }
35213  
35214 +static struct file *loop_real_file(struct file *file)
35215 +{
35216 +       struct file *f = NULL;
35217 +       if (file->f_path.dentry->d_sb->s_op->real_loop)
35218 +               f = file->f_path.dentry->d_sb->s_op->real_loop(file);
35219 +       return f;
35220 +}
35221  
35222  /*
35223   * loop_change_fd switched the backing store of a loopback device to
35224 @@ -640,6 +650,7 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
35225                           unsigned int arg)
35226  {
35227         struct file     *file, *old_file;
35228 +       struct file     *f, *virt_file = NULL, *old_virt_file;
35229         struct inode    *inode;
35230         int             error;
35231  
35232 @@ -656,9 +667,16 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
35233         file = fget(arg);
35234         if (!file)
35235                 goto out;
35236 +       f = loop_real_file(file);
35237 +       if (f) {
35238 +               virt_file = file;
35239 +               file = f;
35240 +               get_file(file);
35241 +       }
35242  
35243         inode = file->f_mapping->host;
35244         old_file = lo->lo_backing_file;
35245 +       old_virt_file = lo->lo_backing_virt_file;
35246  
35247         error = -EINVAL;
35248  
35249 @@ -670,17 +688,21 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
35250                 goto out_putf;
35251  
35252         /* and ... switch */
35253 -       error = loop_switch(lo, file);
35254 +       error = loop_switch(lo, file, virt_file);
35255         if (error)
35256                 goto out_putf;
35257  
35258         fput(old_file);
35259 +       if (old_virt_file)
35260 +               fput(old_virt_file);
35261         if (lo->lo_flags & LO_FLAGS_PARTSCAN)
35262                 ioctl_by_bdev(bdev, BLKRRPART, 0);
35263         return 0;
35264  
35265   out_putf:
35266         fput(file);
35267 +       if (virt_file)
35268 +               fput(virt_file);
35269   out:
35270         return error;
35271  }
35272 @@ -841,7 +863,7 @@ static void loop_config_discard(struct loop_device *lo)
35273  static int loop_set_fd(struct loop_device *lo, fmode_t mode,
35274                        struct block_device *bdev, unsigned int arg)
35275  {
35276 -       struct file     *file, *f;
35277 +       struct file     *file, *f, *virt_file = NULL;
35278         struct inode    *inode;
35279         struct address_space *mapping;
35280         unsigned lo_blocksize;
35281 @@ -856,6 +878,12 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
35282         file = fget(arg);
35283         if (!file)
35284                 goto out;
35285 +       f = loop_real_file(file);
35286 +       if (f) {
35287 +               virt_file = file;
35288 +               file = f;
35289 +               get_file(file);
35290 +       }
35291  
35292         error = -EBUSY;
35293         if (lo->lo_state != Lo_unbound)
35294 @@ -904,6 +932,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
35295         lo->lo_device = bdev;
35296         lo->lo_flags = lo_flags;
35297         lo->lo_backing_file = file;
35298 +       lo->lo_backing_virt_file = virt_file;
35299         lo->transfer = transfer_none;
35300         lo->ioctl = NULL;
35301         lo->lo_sizelimit = 0;
35302 @@ -948,6 +977,7 @@ out_clr:
35303         lo->lo_thread = NULL;
35304         lo->lo_device = NULL;
35305         lo->lo_backing_file = NULL;
35306 +       lo->lo_backing_virt_file = NULL;
35307         lo->lo_flags = 0;
35308         set_capacity(lo->lo_disk, 0);
35309         invalidate_bdev(bdev);
35310 @@ -957,6 +987,8 @@ out_clr:
35311         lo->lo_state = Lo_unbound;
35312   out_putf:
35313         fput(file);
35314 +       if (virt_file)
35315 +               fput(virt_file);
35316   out:
35317         /* This is safe: open() is still holding a reference. */
35318         module_put(THIS_MODULE);
35319 @@ -1003,6 +1035,7 @@ loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
35320  static int loop_clr_fd(struct loop_device *lo)
35321  {
35322         struct file *filp = lo->lo_backing_file;
35323 +       struct file *virt_filp = lo->lo_backing_virt_file;
35324         gfp_t gfp = lo->old_gfp_mask;
35325         struct block_device *bdev = lo->lo_device;
35326  
35327 @@ -1036,6 +1069,7 @@ static int loop_clr_fd(struct loop_device *lo)
35328  
35329         spin_lock_irq(&lo->lo_lock);
35330         lo->lo_backing_file = NULL;
35331 +       lo->lo_backing_virt_file = NULL;
35332         spin_unlock_irq(&lo->lo_lock);
35333  
35334         loop_release_xfer(lo);
35335 @@ -1078,6 +1112,8 @@ static int loop_clr_fd(struct loop_device *lo)
35336          * bd_mutex which is usually taken before lo_ctl_mutex.
35337          */
35338         fput(filp);
35339 +       if (virt_filp)
35340 +               fput(virt_filp);
35341         return 0;
35342  }
35343  
35344 diff --git a/drivers/block/loop.h b/drivers/block/loop.h
35345 index 90df5d6..cb91822 100644
35346 --- a/drivers/block/loop.h
35347 +++ b/drivers/block/loop.h
35348 @@ -44,7 +44,7 @@ struct loop_device {
35349         int             (*ioctl)(struct loop_device *, int cmd, 
35350                                  unsigned long arg); 
35351  
35352 -       struct file *   lo_backing_file;
35353 +       struct file *   lo_backing_file, *lo_backing_virt_file;
35354         struct block_device *lo_device;
35355         unsigned        lo_blocksize;
35356         void            *key_data; 
35357 diff --git a/fs/aufs/f_op.c b/fs/aufs/f_op.c
35358 index 41c4c4a..6a68365 100644
35359 --- a/fs/aufs/f_op.c
35360 +++ b/fs/aufs/f_op.c
35361 @@ -368,7 +368,7 @@ static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
35362         err = -EINVAL;
35363         h_file = au_hf_top(file);
35364         get_file(h_file);
35365 -       if (au_test_loopback_kthread()) {
35366 +       if (0 && au_test_loopback_kthread()) {
35367                 au_warn_loopback(h_file->f_path.dentry->d_sb);
35368                 if (file->f_mapping != h_file->f_mapping) {
35369                         file->f_mapping = h_file->f_mapping;
35370 diff --git a/fs/aufs/loop.c b/fs/aufs/loop.c
35371 index 69f7e96..7941063 100644
35372 --- a/fs/aufs/loop.c
35373 +++ b/fs/aufs/loop.c
35374 @@ -130,3 +130,19 @@ void au_loopback_fin(void)
35375         symbol_put(loop_backing_file);
35376         kfree(au_warn_loopback_array);
35377  }
35378 +
35379 +/* ---------------------------------------------------------------------- */
35380 +
35381 +/* support the loopback block device insude aufs */
35382 +
35383 +struct file *aufs_real_loop(struct file *file)
35384 +{
35385 +       struct file *f;
35386 +
35387 +       BUG_ON(!au_test_aufs(file->f_path.dentry->d_sb));
35388 +       fi_read_lock(file);
35389 +       f = au_hf_top(file);
35390 +       fi_read_unlock(file);
35391 +       AuDebugOn(!f);
35392 +       return f;
35393 +}
35394 diff --git a/fs/aufs/loop.h b/fs/aufs/loop.h
35395 index 6d9864d..3322557 100644
35396 --- a/fs/aufs/loop.h
35397 +++ b/fs/aufs/loop.h
35398 @@ -25,7 +25,11 @@ void au_warn_loopback(struct super_block *h_sb);
35399  
35400  int au_loopback_init(void);
35401  void au_loopback_fin(void);
35402 +
35403 +struct file *aufs_real_loop(struct file *file);
35404  #else
35405 +AuStub(struct file *, loop_backing_file, return NULL)
35406 +
35407  AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
35408            struct dentry *h_adding)
35409  AuStubInt0(au_test_loopback_kthread, void)
35410 @@ -33,6 +37,8 @@ AuStubVoid(au_warn_loopback, struct super_block *h_sb)
35411  
35412  AuStubInt0(au_loopback_init, void)
35413  AuStubVoid(au_loopback_fin, void)
35414 +
35415 +AuStub(struct file *, aufs_real_loop, return NULL, struct file *file)
35416  #endif /* BLK_DEV_LOOP */
35417  
35418  #endif /* __KERNEL__ */
35419 diff --git a/fs/aufs/super.c b/fs/aufs/super.c
35420 index 9acda0e..bb80e3b 100644
35421 --- a/fs/aufs/super.c
35422 +++ b/fs/aufs/super.c
35423 @@ -812,7 +812,10 @@ static const struct super_operations aufs_sop = {
35424         .statfs         = aufs_statfs,
35425         .put_super      = aufs_put_super,
35426         .sync_fs        = aufs_sync_fs,
35427 -       .remount_fs     = aufs_remount_fs
35428 +       .remount_fs     = aufs_remount_fs,
35429 +#ifdef CONFIG_AUFS_BDEV_LOOP
35430 +       .real_loop      = aufs_real_loop
35431 +#endif
35432  };
35433  
35434  /* ---------------------------------------------------------------------- */
35435 diff --git a/include/linux/fs.h b/include/linux/fs.h
35436 index f634198..bfecf84 100644
35437 --- a/include/linux/fs.h
35438 +++ b/include/linux/fs.h
35439 @@ -1620,6 +1620,10 @@ struct super_operations {
35440         int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
35441         long (*nr_cached_objects)(struct super_block *, int);
35442         long (*free_cached_objects)(struct super_block *, long, int);
35443 +#if defined(CONFIG_BLK_DEV_LOOP) ||  defined(CONFIG_BLK_DEV_LOOP_MODULE)
35444 +       /* and aufs */
35445 +       struct file *(*real_loop)(struct file *);
35446 +#endif
35447  };
35448  
35449  /*
This page took 4.134984 seconds and 3 git commands to generate.