]> git.pld-linux.org Git - packages/kernel.git/blob - kernel-aufs3.patch
- 3.14.55
[packages/kernel.git] / kernel-aufs3.patch
1 aufs3.14 kbuild patch
2
3 diff --git a/fs/Kconfig b/fs/Kconfig
4 index 7385e54..d5c769c 100644
5 --- a/fs/Kconfig
6 +++ b/fs/Kconfig
7 @@ -208,6 +208,7 @@ 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 47ac07b..0c6a294 100644
17 --- a/fs/Makefile
18 +++ b/fs/Makefile
19 @@ -125,3 +125,4 @@ 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 3ce25b5..9faebdc 100644
26 --- a/include/uapi/linux/Kbuild
27 +++ b/include/uapi/linux/Kbuild
28 @@ -56,6 +56,7 @@ header-y += atmppp.h
29  header-y += atmsap.h
30  header-y += atmsvc.h
31  header-y += audit.h
32 +header-y += aufs_type.h
33  header-y += auto_fs.h
34  header-y += auto_fs4.h
35  header-y += auxvec.h
36 aufs3.14 base patch
37
38 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
39 index 66e8c3b..ec278ac 100644
40 --- a/drivers/block/loop.c
41 +++ b/drivers/block/loop.c
42 @@ -692,6 +692,24 @@ static inline int is_loop_device(struct file *file)
43         return i && S_ISBLK(i->i_mode) && MAJOR(i->i_rdev) == LOOP_MAJOR;
44  }
45  
46 +/*
47 + * for AUFS
48 + * no get/put for file.
49 + */
50 +struct file *loop_backing_file(struct super_block *sb)
51 +{
52 +       struct file *ret;
53 +       struct loop_device *l;
54 +
55 +       ret = NULL;
56 +       if (MAJOR(sb->s_dev) == LOOP_MAJOR) {
57 +               l = sb->s_bdev->bd_disk->private_data;
58 +               ret = l->lo_backing_file;
59 +       }
60 +       return ret;
61 +}
62 +EXPORT_SYMBOL(loop_backing_file);
63 +
64  /* loop sysfs attributes */
65  
66  static ssize_t loop_attr_show(struct device *dev, char *page,
67 diff --git a/fs/inode.c b/fs/inode.c
68 index 4bcdad3..bc83168 100644
69 --- a/fs/inode.c
70 +++ b/fs/inode.c
71 @@ -1497,7 +1497,7 @@ static int relatime_need_update(struct vfsmount *mnt, struct inode *inode,
72   * This does the actual work of updating an inodes time or version.  Must have
73   * had called mnt_want_write() before calling this.
74   */
75 -static int update_time(struct inode *inode, struct timespec *time, int flags)
76 +int update_time(struct inode *inode, struct timespec *time, int flags)
77  {
78         if (inode->i_op->update_time)
79                 return inode->i_op->update_time(inode, time, flags);
80 diff --git a/fs/splice.c b/fs/splice.c
81 index 12028fa..f26cbaf 100644
82 --- a/fs/splice.c
83 +++ b/fs/splice.c
84 @@ -1111,8 +1111,8 @@ EXPORT_SYMBOL(generic_splice_sendpage);
85  /*
86   * Attempt to initiate a splice from pipe to file.
87   */
88 -static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
89 -                          loff_t *ppos, size_t len, unsigned int flags)
90 +long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
91 +                   loff_t *ppos, size_t len, unsigned int flags)
92  {
93         ssize_t (*splice_write)(struct pipe_inode_info *, struct file *,
94                                 loff_t *, size_t, unsigned int);
95 @@ -1128,9 +1128,9 @@ static long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
96  /*
97   * Attempt to initiate a splice from a file to a pipe.
98   */
99 -static long do_splice_to(struct file *in, loff_t *ppos,
100 -                        struct pipe_inode_info *pipe, size_t len,
101 -                        unsigned int flags)
102 +long do_splice_to(struct file *in, loff_t *ppos,
103 +                 struct pipe_inode_info *pipe, size_t len,
104 +                 unsigned int flags)
105  {
106         ssize_t (*splice_read)(struct file *, loff_t *,
107                                struct pipe_inode_info *, size_t, unsigned int);
108 diff --git a/include/linux/fs.h b/include/linux/fs.h
109 index 23b2a35..f3f635c 100644
110 --- a/include/linux/fs.h
111 +++ b/include/linux/fs.h
112 @@ -2669,6 +2669,7 @@ extern int inode_change_ok(const struct inode *, struct iattr *);
113  extern int inode_newsize_ok(const struct inode *, loff_t offset);
114  extern void setattr_copy(struct inode *inode, const struct iattr *attr);
115  
116 +extern int update_time(struct inode *, struct timespec *, int);
117  extern int file_update_time(struct file *file);
118  
119  extern int generic_show_options(struct seq_file *m, struct dentry *root);
120 diff --git a/include/linux/splice.h b/include/linux/splice.h
121 index 0e43906..304169e 100644
122 --- a/include/linux/splice.h
123 +++ b/include/linux/splice.h
124 @@ -93,4 +93,10 @@ extern void splice_shrink_spd(struct splice_pipe_desc *);
125  extern void spd_release_page(struct splice_pipe_desc *, unsigned int);
126  
127  extern const struct pipe_buf_operations page_cache_pipe_buf_ops;
128 +
129 +extern long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
130 +                          loff_t *ppos, size_t len, unsigned int flags);
131 +extern long do_splice_to(struct file *in, loff_t *ppos,
132 +                        struct pipe_inode_info *pipe, size_t len,
133 +                        unsigned int flags);
134  #endif
135 aufs3.14 mmap patch
136
137 diff --git a/fs/buffer.c b/fs/buffer.c
138 index 27265a8..75427a6 100644
139 --- a/fs/buffer.c
140 +++ b/fs/buffer.c
141 @@ -2448,7 +2448,7 @@ int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
142          * Update file times before taking page lock. We may end up failing the
143          * fault so this update may be superfluous but who really cares...
144          */
145 -       file_update_time(vma->vm_file);
146 +       vma_file_update_time(vma);
147  
148         ret = __block_page_mkwrite(vma, vmf, get_block);
149         sb_end_pagefault(sb);
150 diff --git a/fs/proc/nommu.c b/fs/proc/nommu.c
151 index d4a3574..e44a744 100644
152 --- a/fs/proc/nommu.c
153 +++ b/fs/proc/nommu.c
154 @@ -45,7 +45,9 @@ static int nommu_region_show(struct seq_file *m, struct vm_region *region)
155         file = region->vm_file;
156  
157         if (file) {
158 -               struct inode *inode = file_inode(region->vm_file);
159 +               struct inode *inode;
160 +               file = vmr_pr_or_file(region);
161 +               inode = file_inode(file);
162                 dev = inode->i_sb->s_dev;
163                 ino = inode->i_ino;
164         }
165 diff --git a/fs/proc/task_mmu.c b/fs/proc/task_mmu.c
166 index fb52b54..1aca72e 100644
167 --- a/fs/proc/task_mmu.c
168 +++ b/fs/proc/task_mmu.c
169 @@ -264,7 +264,9 @@ show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid)
170         const char *name = NULL;
171  
172         if (file) {
173 -               struct inode *inode = file_inode(vma->vm_file);
174 +               struct inode *inode;
175 +               file = vma_pr_or_file(vma);
176 +               inode = file_inode(file);
177                 dev = inode->i_sb->s_dev;
178                 ino = inode->i_ino;
179                 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT;
180 @@ -1407,6 +1409,7 @@ static int show_numa_map(struct seq_file *m, void *v, int is_pid)
181         seq_printf(m, "%08lx %s", vma->vm_start, buffer);
182  
183         if (file) {
184 +               file = vma_pr_or_file(vma);
185                 seq_printf(m, " file=");
186                 seq_path(m, &file->f_path, "\n\t= ");
187         } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) {
188 diff --git a/fs/proc/task_nommu.c b/fs/proc/task_nommu.c
189 index 678455d..ad0ce45 100644
190 --- a/fs/proc/task_nommu.c
191 +++ b/fs/proc/task_nommu.c
192 @@ -141,7 +141,9 @@ static int nommu_vma_show(struct seq_file *m, struct vm_area_struct *vma,
193         file = vma->vm_file;
194  
195         if (file) {
196 -               struct inode *inode = file_inode(vma->vm_file);
197 +               struct inode *inode;
198 +               file = vma_pr_or_file(file);
199 +               inode = file_inode(file);
200                 dev = inode->i_sb->s_dev;
201                 ino = inode->i_ino;
202                 pgoff = (loff_t)vma->vm_pgoff << PAGE_SHIFT;
203 diff --git a/include/linux/mm.h b/include/linux/mm.h
204 index c1b7414..02036e0 100644
205 --- a/include/linux/mm.h
206 +++ b/include/linux/mm.h
207 @@ -18,6 +18,9 @@
208  #include <linux/pfn.h>
209  #include <linux/bit_spinlock.h>
210  #include <linux/shrinker.h>
211 +#include <linux/dcache.h>
212 +#include <linux/file.h>
213 +#include <linux/fs.h>
214  
215  struct mempolicy;
216  struct anon_vma;
217 @@ -1152,6 +1155,87 @@ static inline int fixup_user_fault(struct task_struct *tsk,
218  }
219  #endif
220  
221 +/*
222 + * Mainly for aufs which mmap(2) diffrent file and wants to print different path
223 + * in /proc/PID/maps.
224 + */
225 +/* #define AUFS_DEBUG_MMAP */
226 +static inline void aufs_trace(struct file *f, struct file *pr,
227 +                             const char func[], int line, const char func2[])
228 +{
229 +#ifdef AUFS_DEBUG_MMAP
230 +       if (pr)
231 +               pr_info("%s:%d: %s, %p\n", func, line, func2,
232 +                       f ? (char *)f->f_dentry->d_name.name : "(null)");
233 +#endif
234 +}
235 +
236 +static inline struct file *vmr_do_pr_or_file(struct vm_region *region,
237 +                                            const char func[], int line)
238 +{
239 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
240 +       aufs_trace(f, pr, func, line, __func__);
241 +       return (f && pr) ? pr : f;
242 +}
243 +
244 +static inline void vmr_do_fput(struct vm_region *region,
245 +                              const char func[], int line)
246 +{
247 +       struct file *f = region->vm_file, *pr = region->vm_prfile;
248 +       aufs_trace(f, pr, func, line, __func__);
249 +       fput(f);
250 +       if (f && pr)
251 +               fput(pr);
252 +}
253 +
254 +static inline void vma_do_file_update_time(struct vm_area_struct *vma,
255 +                                          const char func[], int line)
256 +{
257 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
258 +       aufs_trace(f, pr, func, line, __func__);
259 +       file_update_time(f);
260 +       if (f && pr)
261 +               file_update_time(pr);
262 +}
263 +
264 +static inline struct file *vma_do_pr_or_file(struct vm_area_struct *vma,
265 +                                            const char func[], int line)
266 +{
267 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
268 +       aufs_trace(f, pr, func, line, __func__);
269 +       return (f && pr) ? pr : f;
270 +}
271 +
272 +static inline void vma_do_get_file(struct vm_area_struct *vma,
273 +                                  const char func[], int line)
274 +{
275 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
276 +       aufs_trace(f, pr, func, line, __func__);
277 +       get_file(f);
278 +       if (f && pr)
279 +               get_file(pr);
280 +}
281 +
282 +static inline void vma_do_fput(struct vm_area_struct *vma,
283 +                              const char func[], int line)
284 +{
285 +       struct file *f = vma->vm_file, *pr = vma->vm_prfile;
286 +       aufs_trace(f, pr, func, line, __func__);
287 +       fput(f);
288 +       if (f && pr)
289 +               fput(pr);
290 +}
291 +
292 +#define vmr_pr_or_file(region)         vmr_do_pr_or_file(region, __func__, \
293 +                                                         __LINE__)
294 +#define vmr_fput(region)               vmr_do_fput(region, __func__, __LINE__)
295 +#define vma_file_update_time(vma)      vma_do_file_update_time(vma, __func__, \
296 +                                                               __LINE__)
297 +#define vma_pr_or_file(vma)            vma_do_pr_or_file(vma, __func__, \
298 +                                                         __LINE__)
299 +#define vma_get_file(vma)              vma_do_get_file(vma, __func__, __LINE__)
300 +#define vma_fput(vma)                  vma_do_fput(vma, __func__, __LINE__)
301 +
302  extern int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write);
303  extern int access_remote_vm(struct mm_struct *mm, unsigned long addr,
304                 void *buf, int len, int write);
305 diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
306 index 290901a..c21588b 100644
307 --- a/include/linux/mm_types.h
308 +++ b/include/linux/mm_types.h
309 @@ -231,6 +231,7 @@ struct vm_region {
310         unsigned long   vm_top;         /* region allocated to here */
311         unsigned long   vm_pgoff;       /* the offset in vm_file corresponding to vm_start */
312         struct file     *vm_file;       /* the backing file or NULL */
313 +       struct file     *vm_prfile;     /* the virtual backing file or NULL */
314  
315         int             vm_usage;       /* region usage count (access under nommu_region_sem) */
316         bool            vm_icache_flushed : 1; /* true if the icache has been flushed for
317 @@ -299,6 +300,7 @@ struct vm_area_struct {
318         unsigned long vm_pgoff;         /* Offset (within vm_file) in PAGE_SIZE
319                                            units, *not* PAGE_CACHE_SIZE */
320         struct file * vm_file;          /* File we map to (can be NULL). */
321 +       struct file *vm_prfile;         /* shadow of vm_file */
322         void * vm_private_data;         /* was vm_pte (shared mem) */
323  
324  #ifndef CONFIG_MMU
325 diff --git a/kernel/fork.c b/kernel/fork.c
326 index a17621c..40d9f6a 100644
327 --- a/kernel/fork.c
328 +++ b/kernel/fork.c
329 @@ -412,7 +412,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
330                         struct inode *inode = file_inode(file);
331                         struct address_space *mapping = file->f_mapping;
332  
333 -                       get_file(file);
334 +                       vma_get_file(tmp);
335                         if (tmp->vm_flags & VM_DENYWRITE)
336                                 atomic_dec(&inode->i_writecount);
337                         mutex_lock(&mapping->i_mmap_mutex);
338 diff --git a/mm/filemap.c b/mm/filemap.c
339 index 7a13f6a..f1805df 100644
340 --- a/mm/filemap.c
341 +++ b/mm/filemap.c
342 @@ -1733,7 +1733,7 @@ int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
343         int ret = VM_FAULT_LOCKED;
344  
345         sb_start_pagefault(inode->i_sb);
346 -       file_update_time(vma->vm_file);
347 +       vma_file_update_time(vma);
348         lock_page(page);
349         if (page->mapping != inode->i_mapping) {
350                 unlock_page(page);
351 diff --git a/mm/fremap.c b/mm/fremap.c
352 index 34feba6..d857364 100644
353 --- a/mm/fremap.c
354 +++ b/mm/fremap.c
355 @@ -227,7 +227,9 @@ get_write_lock:
356                         /* mmap_region may free vma; grab the info now */
357                         vm_flags = vma->vm_flags;
358  
359 +                       vma_get_file(vma);
360                         addr = mmap_region(file, start, size, vm_flags, pgoff);
361 +                       vma_fput(vma);
362                         fput(file);
363                         if (IS_ERR_VALUE(addr)) {
364                                 err = addr;
365 diff --git a/mm/madvise.c b/mm/madvise.c
366 index 539eeb9..5e700b1 100644
367 --- a/mm/madvise.c
368 +++ b/mm/madvise.c
369 @@ -327,12 +327,12 @@ static long madvise_remove(struct vm_area_struct *vma,
370          * vma's reference to the file) can go away as soon as we drop
371          * mmap_sem.
372          */
373 -       get_file(f);
374 +       vma_get_file(vma);
375         up_read(&current->mm->mmap_sem);
376         error = do_fallocate(f,
377                                 FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
378                                 offset, end - start);
379 -       fput(f);
380 +       vma_fput(vma);
381         down_read(&current->mm->mmap_sem);
382         return error;
383  }
384 diff --git a/mm/memory.c b/mm/memory.c
385 index 22dfa61..81813d9 100644
386 --- a/mm/memory.c
387 +++ b/mm/memory.c
388 @@ -2755,7 +2755,7 @@ reuse:
389                         set_page_dirty_balance(dirty_page, page_mkwrite);
390                         /* file_update_time outside page_lock */
391                         if (vma->vm_file)
392 -                               file_update_time(vma->vm_file);
393 +                               vma_file_update_time(vma);
394                 }
395                 put_page(dirty_page);
396                 if (page_mkwrite) {
397 @@ -3467,7 +3467,7 @@ static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
398  
399                 /* file_update_time outside page_lock */
400                 if (vma->vm_file && !page_mkwrite)
401 -                       file_update_time(vma->vm_file);
402 +                       vma_file_update_time(vma);
403         } else {
404                 unlock_page(vmf.page);
405                 if (anon)
406 diff --git a/mm/mmap.c b/mm/mmap.c
407 index 20ff0c3..f743033 100644
408 --- a/mm/mmap.c
409 +++ b/mm/mmap.c
410 @@ -249,7 +249,7 @@ static struct vm_area_struct *remove_vma(struct vm_area_struct *vma)
411         if (vma->vm_ops && vma->vm_ops->close)
412                 vma->vm_ops->close(vma);
413         if (vma->vm_file)
414 -               fput(vma->vm_file);
415 +               vma_fput(vma);
416         mpol_put(vma_policy(vma));
417         kmem_cache_free(vm_area_cachep, vma);
418         return next;
419 @@ -859,7 +859,7 @@ again:                      remove_next = 1 + (end > next->vm_end);
420         if (remove_next) {
421                 if (file) {
422                         uprobe_munmap(next, next->vm_start, next->vm_end);
423 -                       fput(file);
424 +                       vma_fput(vma);
425                 }
426                 if (next->anon_vma)
427                         anon_vma_merge(vma, next);
428 @@ -1639,8 +1639,8 @@ out:
429  unmap_and_free_vma:
430         if (vm_flags & VM_DENYWRITE)
431                 allow_write_access(file);
432 +       vma_fput(vma);
433         vma->vm_file = NULL;
434 -       fput(file);
435  
436         /* Undo any partial mapping done by a device driver. */
437         unmap_region(mm, vma, prev, vma->vm_start, vma->vm_end);
438 @@ -2429,7 +2429,7 @@ static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
439                 goto out_free_mpol;
440  
441         if (new->vm_file)
442 -               get_file(new->vm_file);
443 +               vma_get_file(new);
444  
445         if (new->vm_ops && new->vm_ops->open)
446                 new->vm_ops->open(new);
447 @@ -2448,7 +2448,7 @@ static int __split_vma(struct mm_struct * mm, struct vm_area_struct * vma,
448         if (new->vm_ops && new->vm_ops->close)
449                 new->vm_ops->close(new);
450         if (new->vm_file)
451 -               fput(new->vm_file);
452 +               vma_fput(new);
453         unlink_anon_vmas(new);
454   out_free_mpol:
455         mpol_put(vma_policy(new));
456 @@ -2837,7 +2837,7 @@ struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
457                         if (anon_vma_clone(new_vma, vma))
458                                 goto out_free_mempol;
459                         if (new_vma->vm_file)
460 -                               get_file(new_vma->vm_file);
461 +                               vma_get_file(new_vma);
462                         if (new_vma->vm_ops && new_vma->vm_ops->open)
463                                 new_vma->vm_ops->open(new_vma);
464                         vma_link(mm, new_vma, prev, rb_link, rb_parent);
465 diff --git a/mm/msync.c b/mm/msync.c
466 index 632df45..02d770e 100644
467 --- a/mm/msync.c
468 +++ b/mm/msync.c
469 @@ -80,10 +80,10 @@ SYSCALL_DEFINE3(msync, unsigned long, start, size_t, len, int, flags)
470                 start = vma->vm_end;
471                 if ((flags & MS_SYNC) && file &&
472                                 (vma->vm_flags & VM_SHARED)) {
473 -                       get_file(file);
474 +                       vma_get_file(vma);
475                         up_read(&mm->mmap_sem);
476                         error = vfs_fsync(file, 0);
477 -                       fput(file);
478 +                       vma_fput(vma);
479                         if (error || start >= end)
480                                 goto out;
481                         down_read(&mm->mmap_sem);
482 diff --git a/mm/nommu.c b/mm/nommu.c
483 index 8740213..ea7e336 100644
484 --- a/mm/nommu.c
485 +++ b/mm/nommu.c
486 @@ -653,7 +653,7 @@ static void __put_nommu_region(struct vm_region *region)
487                 up_write(&nommu_region_sem);
488  
489                 if (region->vm_file)
490 -                       fput(region->vm_file);
491 +                       vmr_fput(region);
492  
493                 /* IO memory and memory shared directly out of the pagecache
494                  * from ramfs/tmpfs mustn't be released here */
495 @@ -811,7 +811,7 @@ static void delete_vma(struct mm_struct *mm, struct vm_area_struct *vma)
496         if (vma->vm_ops && vma->vm_ops->close)
497                 vma->vm_ops->close(vma);
498         if (vma->vm_file)
499 -               fput(vma->vm_file);
500 +               vma_fput(vma);
501         put_nommu_region(vma->vm_region);
502         kmem_cache_free(vm_area_cachep, vma);
503  }
504 @@ -1377,7 +1377,7 @@ unsigned long do_mmap_pgoff(struct file *file,
505                                         goto error_just_free;
506                                 }
507                         }
508 -                       fput(region->vm_file);
509 +                       vmr_fput(region);
510                         kmem_cache_free(vm_region_jar, region);
511                         region = pregion;
512                         result = start;
513 @@ -1453,10 +1453,10 @@ error_just_free:
514         up_write(&nommu_region_sem);
515  error:
516         if (region->vm_file)
517 -               fput(region->vm_file);
518 +               vmr_fput(region);
519         kmem_cache_free(vm_region_jar, region);
520         if (vma->vm_file)
521 -               fput(vma->vm_file);
522 +               vma_fput(vma);
523         kmem_cache_free(vm_area_cachep, vma);
524         kleave(" = %d", ret);
525         return ret;
526 aufs3.14 standalone patch
527
528 diff --git a/fs/inode.c b/fs/inode.c
529 index bc83168..6dd1207 100644
530 --- a/fs/inode.c
531 +++ b/fs/inode.c
532 @@ -57,6 +57,7 @@ static struct hlist_head *inode_hashtable __read_mostly;
533  static __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_hash_lock);
534  
535  __cacheline_aligned_in_smp DEFINE_SPINLOCK(inode_sb_list_lock);
536 +EXPORT_SYMBOL(inode_sb_list_lock);
537  
538  /*
539   * Empty aops. Can be used for the cases where the user does not
540 @@ -1513,6 +1514,7 @@ int update_time(struct inode *inode, struct timespec *time, int flags)
541         mark_inode_dirty_sync(inode);
542         return 0;
543  }
544 +EXPORT_SYMBOL(update_time);
545  
546  /**
547   *     touch_atime     -       update the access time
548 diff --git a/fs/namespace.c b/fs/namespace.c
549 index 2ffc5a2..785a51f 100644
550 --- a/fs/namespace.c
551 +++ b/fs/namespace.c
552 @@ -455,6 +455,7 @@ void __mnt_drop_write(struct vfsmount *mnt)
553         mnt_dec_writers(real_mount(mnt));
554         preempt_enable();
555  }
556 +EXPORT_SYMBOL_GPL(__mnt_drop_write);
557  
558  /**
559   * mnt_drop_write - give up write access to a mount
560 @@ -1555,6 +1556,7 @@ int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
561         }
562         return 0;
563  }
564 +EXPORT_SYMBOL(iterate_mounts);
565  
566  static void cleanup_group_ids(struct mount *mnt, struct mount *end)
567  {
568 diff --git a/fs/notify/group.c b/fs/notify/group.c
569 index ad19959..adf290d 100644
570 --- a/fs/notify/group.c
571 +++ b/fs/notify/group.c
572 @@ -22,6 +22,7 @@
573  #include <linux/srcu.h>
574  #include <linux/rculist.h>
575  #include <linux/wait.h>
576 +#include <linux/module.h>
577  
578  #include <linux/fsnotify_backend.h>
579  #include "fsnotify.h"
580 @@ -72,6 +73,7 @@ void fsnotify_get_group(struct fsnotify_group *group)
581  {
582         atomic_inc(&group->refcnt);
583  }
584 +EXPORT_SYMBOL(fsnotify_get_group);
585  
586  /*
587   * Drop a reference to a group.  Free it if it's through.
588 @@ -81,6 +83,7 @@ void fsnotify_put_group(struct fsnotify_group *group)
589         if (atomic_dec_and_test(&group->refcnt))
590                 fsnotify_final_destroy_group(group);
591  }
592 +EXPORT_SYMBOL(fsnotify_put_group);
593  
594  /*
595   * Create a new fsnotify_group and hold a reference for the group returned.
596 @@ -109,6 +112,7 @@ struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops)
597  
598         return group;
599  }
600 +EXPORT_SYMBOL(fsnotify_alloc_group);
601  
602  int fsnotify_fasync(int fd, struct file *file, int on)
603  {
604 diff --git a/fs/notify/mark.c b/fs/notify/mark.c
605 index 923fe4a..176b435 100644
606 --- a/fs/notify/mark.c
607 +++ b/fs/notify/mark.c
608 @@ -109,6 +109,7 @@ void fsnotify_put_mark(struct fsnotify_mark *mark)
609                 mark->free_mark(mark);
610         }
611  }
612 +EXPORT_SYMBOL(fsnotify_put_mark);
613  
614  /*
615   * Any time a mark is getting freed we end up here.
616 @@ -191,6 +192,7 @@ void fsnotify_destroy_mark(struct fsnotify_mark *mark,
617         fsnotify_destroy_mark_locked(mark, group);
618         mutex_unlock(&group->mark_mutex);
619  }
620 +EXPORT_SYMBOL(fsnotify_destroy_mark);
621  
622  void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask)
623  {
624 @@ -275,6 +277,7 @@ err:
625  
626         return ret;
627  }
628 +EXPORT_SYMBOL(fsnotify_add_mark);
629  
630  int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
631                       struct inode *inode, struct vfsmount *mnt, int allow_dups)
632 @@ -336,6 +339,7 @@ void fsnotify_init_mark(struct fsnotify_mark *mark,
633         atomic_set(&mark->refcnt, 1);
634         mark->free_mark = free_mark;
635  }
636 +EXPORT_SYMBOL(fsnotify_init_mark);
637  
638  static int fsnotify_mark_destroy(void *ignored)
639  {
640 diff --git a/fs/open.c b/fs/open.c
641 index b9ed8b2..3ea66972 100644
642 --- a/fs/open.c
643 +++ b/fs/open.c
644 @@ -62,6 +62,7 @@ int do_truncate(struct dentry *dentry, loff_t length, unsigned int time_attrs,
645         mutex_unlock(&dentry->d_inode->i_mutex);
646         return ret;
647  }
648 +EXPORT_SYMBOL(do_truncate);
649  
650  long vfs_truncate(struct path *path, loff_t length)
651  {
652 diff --git a/fs/splice.c b/fs/splice.c
653 index f26cbaf..ac02366 100644
654 --- a/fs/splice.c
655 +++ b/fs/splice.c
656 @@ -1124,6 +1124,7 @@ long do_splice_from(struct pipe_inode_info *pipe, struct file *out,
657  
658         return splice_write(pipe, out, ppos, len, flags);
659  }
660 +EXPORT_SYMBOL(do_splice_from);
661  
662  /*
663   * Attempt to initiate a splice from a file to a pipe.
664 @@ -1150,6 +1151,7 @@ long do_splice_to(struct file *in, loff_t *ppos,
665  
666         return splice_read(in, ppos, pipe, len, flags);
667  }
668 +EXPORT_SYMBOL(do_splice_to);
669  
670  /**
671   * splice_direct_to_actor - splices data directly between two non-pipes
672 diff --git a/security/commoncap.c b/security/commoncap.c
673 index b9d613e..ba3b618 100644
674 --- a/security/commoncap.c
675 +++ b/security/commoncap.c
676 @@ -988,9 +988,11 @@ int cap_mmap_addr(unsigned long addr)
677         }
678         return ret;
679  }
680 +EXPORT_SYMBOL(cap_mmap_addr);
681  
682  int cap_mmap_file(struct file *file, unsigned long reqprot,
683                   unsigned long prot, unsigned long flags)
684  {
685         return 0;
686  }
687 +EXPORT_SYMBOL(cap_mmap_file);
688 diff --git a/security/device_cgroup.c b/security/device_cgroup.c
689 index d3b6d2c..5076912 100644
690 --- a/security/device_cgroup.c
691 +++ b/security/device_cgroup.c
692 @@ -7,6 +7,7 @@
693  #include <linux/device_cgroup.h>
694  #include <linux/cgroup.h>
695  #include <linux/ctype.h>
696 +#include <linux/export.h>
697  #include <linux/list.h>
698  #include <linux/uaccess.h>
699  #include <linux/seq_file.h>
700 @@ -744,6 +745,7 @@ int __devcgroup_inode_permission(struct inode *inode, int mask)
701         return __devcgroup_check_permission(type, imajor(inode), iminor(inode),
702                         access);
703  }
704 +EXPORT_SYMBOL(__devcgroup_inode_permission);
705  
706  int devcgroup_inode_mknod(int mode, dev_t dev)
707  {
708 diff --git a/security/security.c b/security/security.c
709 index 919cad9..f9e9e17 100644
710 --- a/security/security.c
711 +++ b/security/security.c
712 @@ -407,6 +407,7 @@ int security_path_rmdir(struct path *dir, struct dentry *dentry)
713                 return 0;
714         return security_ops->path_rmdir(dir, dentry);
715  }
716 +EXPORT_SYMBOL(security_path_rmdir);
717  
718  int security_path_unlink(struct path *dir, struct dentry *dentry)
719  {
720 @@ -423,6 +424,7 @@ int security_path_symlink(struct path *dir, struct dentry *dentry,
721                 return 0;
722         return security_ops->path_symlink(dir, dentry, old_name);
723  }
724 +EXPORT_SYMBOL(security_path_symlink);
725  
726  int security_path_link(struct dentry *old_dentry, struct path *new_dir,
727                        struct dentry *new_dentry)
728 @@ -431,6 +433,7 @@ int security_path_link(struct dentry *old_dentry, struct path *new_dir,
729                 return 0;
730         return security_ops->path_link(old_dentry, new_dir, new_dentry);
731  }
732 +EXPORT_SYMBOL(security_path_link);
733  
734  int security_path_rename(struct path *old_dir, struct dentry *old_dentry,
735                          struct path *new_dir, struct dentry *new_dentry)
736 @@ -449,6 +452,7 @@ int security_path_truncate(struct path *path)
737                 return 0;
738         return security_ops->path_truncate(path);
739  }
740 +EXPORT_SYMBOL(security_path_truncate);
741  
742  int security_path_chmod(struct path *path, umode_t mode)
743  {
744 @@ -456,6 +460,7 @@ int security_path_chmod(struct path *path, umode_t mode)
745                 return 0;
746         return security_ops->path_chmod(path, mode);
747  }
748 +EXPORT_SYMBOL(security_path_chmod);
749  
750  int security_path_chown(struct path *path, kuid_t uid, kgid_t gid)
751  {
752 @@ -463,6 +468,7 @@ int security_path_chown(struct path *path, kuid_t uid, kgid_t gid)
753                 return 0;
754         return security_ops->path_chown(path, uid, gid);
755  }
756 +EXPORT_SYMBOL(security_path_chown);
757  
758  int security_path_chroot(struct path *path)
759  {
760 @@ -539,6 +545,7 @@ int security_inode_readlink(struct dentry *dentry)
761                 return 0;
762         return security_ops->inode_readlink(dentry);
763  }
764 +EXPORT_SYMBOL(security_inode_readlink);
765  
766  int security_inode_follow_link(struct dentry *dentry, struct nameidata *nd)
767  {
768 @@ -553,6 +560,7 @@ int security_inode_permission(struct inode *inode, int mask)
769                 return 0;
770         return security_ops->inode_permission(inode, mask);
771  }
772 +EXPORT_SYMBOL(security_inode_permission);
773  
774  int security_inode_setattr(struct dentry *dentry, struct iattr *attr)
775  {
776 @@ -675,6 +683,7 @@ int security_file_permission(struct file *file, int mask)
777  
778         return fsnotify_perm(file, mask);
779  }
780 +EXPORT_SYMBOL(security_file_permission);
781  
782  int security_file_alloc(struct file *file)
783  {
784 @@ -735,6 +744,7 @@ int security_mmap_file(struct file *file, unsigned long prot,
785                 return ret;
786         return ima_file_mmap(file, prot);
787  }
788 +EXPORT_SYMBOL(security_mmap_file);
789  
790  int security_mmap_addr(unsigned long addr)
791  {
792 diff -urN /usr/share/empty/Documentation/ABI/testing/debugfs-aufs linux/Documentation/ABI/testing/debugfs-aufs
793 --- /usr/share/empty/Documentation/ABI/testing/debugfs-aufs     1970-01-01 01:00:00.000000000 +0100
794 +++ linux/Documentation/ABI/testing/debugfs-aufs        2014-04-24 22:11:10.471931783 +0200
795 @@ -0,0 +1,50 @@
796 +What:          /debug/aufs/si_<id>/
797 +Date:          March 2009
798 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
799 +Description:
800 +               Under /debug/aufs, a directory named si_<id> is created
801 +               per aufs mount, where <id> is a unique id generated
802 +               internally.
803 +
804 +What:          /debug/aufs/si_<id>/plink
805 +Date:          Apr 2013
806 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
807 +Description:
808 +               It has three lines and shows the information about the
809 +               pseudo-link. The first line is a single number
810 +               representing a number of buckets. The second line is a
811 +               number of pseudo-links per buckets (separated by a
812 +               blank). The last line is a single number representing a
813 +               total number of psedo-links.
814 +               When the aufs mount option 'noplink' is specified, it
815 +               will show "1\n0\n0\n".
816 +
817 +What:          /debug/aufs/si_<id>/xib
818 +Date:          March 2009
819 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
820 +Description:
821 +               It shows the consumed blocks by xib (External Inode Number
822 +               Bitmap), its block size and file size.
823 +               When the aufs mount option 'noxino' is specified, it
824 +               will be empty. About XINO files, see the aufs manual.
825 +
826 +What:          /debug/aufs/si_<id>/xino0, xino1 ... xinoN
827 +Date:          March 2009
828 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
829 +Description:
830 +               It shows the consumed blocks by xino (External Inode Number
831 +               Translation Table), its link count, block size and file
832 +               size.
833 +               When the aufs mount option 'noxino' is specified, it
834 +               will be empty. About XINO files, see the aufs manual.
835 +
836 +What:          /debug/aufs/si_<id>/xigen
837 +Date:          March 2009
838 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
839 +Description:
840 +               It shows the consumed blocks by xigen (External Inode
841 +               Generation Table), its block size and file size.
842 +               If CONFIG_AUFS_EXPORT is disabled, this entry will not
843 +               be created.
844 +               When the aufs mount option 'noxino' is specified, it
845 +               will be empty. About XINO files, see the aufs manual.
846 diff -urN /usr/share/empty/Documentation/ABI/testing/sysfs-aufs linux/Documentation/ABI/testing/sysfs-aufs
847 --- /usr/share/empty/Documentation/ABI/testing/sysfs-aufs       1970-01-01 01:00:00.000000000 +0100
848 +++ linux/Documentation/ABI/testing/sysfs-aufs  2014-04-24 22:11:10.471931783 +0200
849 @@ -0,0 +1,31 @@
850 +What:          /sys/fs/aufs/si_<id>/
851 +Date:          March 2009
852 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
853 +Description:
854 +               Under /sys/fs/aufs, a directory named si_<id> is created
855 +               per aufs mount, where <id> is a unique id generated
856 +               internally.
857 +
858 +What:          /sys/fs/aufs/si_<id>/br0, br1 ... brN
859 +Date:          March 2009
860 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
861 +Description:
862 +               It shows the abolute path of a member directory (which
863 +               is called branch) in aufs, and its permission.
864 +
865 +What:          /sys/fs/aufs/si_<id>/brid0, brid1 ... bridN
866 +Date:          July 2013
867 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
868 +Description:
869 +               It shows the id of a member directory (which is called
870 +               branch) in aufs.
871 +
872 +What:          /sys/fs/aufs/si_<id>/xi_path
873 +Date:          March 2009
874 +Contact:       J. R. Okajima <hooanon05g@gmail.com>
875 +Description:
876 +               It shows the abolute path of XINO (External Inode Number
877 +               Bitmap, Translation Table and Generation Table) file
878 +               even if it is the default path.
879 +               When the aufs mount option 'noxino' is specified, it
880 +               will be empty. About XINO files, see the aufs manual.
881 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt linux/Documentation/filesystems/aufs/design/01intro.txt
882 --- /usr/share/empty/Documentation/filesystems/aufs/design/01intro.txt  1970-01-01 01:00:00.000000000 +0100
883 +++ linux/Documentation/filesystems/aufs/design/01intro.txt     2014-04-24 22:11:10.471931783 +0200
884 @@ -0,0 +1,161 @@
885 +
886 +# Copyright (C) 2005-2014 Junjiro R. Okajima
887 +# 
888 +# This program is free software; you can redistribute it and/or modify
889 +# it under the terms of the GNU General Public License as published by
890 +# the Free Software Foundation; either version 2 of the License, or
891 +# (at your option) any later version.
892 +# 
893 +# This program is distributed in the hope that it will be useful,
894 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
895 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
896 +# GNU General Public License for more details.
897 +# 
898 +# You should have received a copy of the GNU General Public License
899 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
900 +
901 +Introduction
902 +----------------------------------------
903 +
904 +aufs [ei ju: ef es] | [a u f s]
905 +1. abbrev. for "advanced multi-layered unification filesystem".
906 +2. abbrev. for "another unionfs".
907 +3. abbrev. for "auf das" in German which means "on the" in English.
908 +   Ex. "Butter aufs Brot"(G) means "butter onto bread"(E).
909 +   But "Filesystem aufs Filesystem" is hard to understand.
910 +
911 +AUFS is a filesystem with features:
912 +- multi layered stackable unification filesystem, the member directory
913 +  is called as a branch.
914 +- branch permission and attribute, 'readonly', 'real-readonly',
915 +  'readwrite', 'whiteout-able', 'link-able whiteout' and their
916 +  combination.
917 +- internal "file copy-on-write".
918 +- logical deletion, whiteout.
919 +- dynamic branch manipulation, adding, deleting and changing permission.
920 +- allow bypassing aufs, user's direct branch access.
921 +- external inode number translation table and bitmap which maintains the
922 +  persistent aufs inode number.
923 +- seekable directory, including NFS readdir.
924 +- file mapping, mmap and sharing pages.
925 +- pseudo-link, hardlink over branches.
926 +- loopback mounted filesystem as a branch.
927 +- several policies to select one among multiple writable branches.
928 +- revert a single systemcall when an error occurs in aufs.
929 +- and more...
930 +
931 +
932 +Multi Layered Stackable Unification Filesystem
933 +----------------------------------------------------------------------
934 +Most people already knows what it is.
935 +It is a filesystem which unifies several directories and provides a
936 +merged single directory. When users access a file, the access will be
937 +passed/re-directed/converted (sorry, I am not sure which English word is
938 +correct) to the real file on the member filesystem. The member
939 +filesystem is called 'lower filesystem' or 'branch' and has a mode
940 +'readonly' and 'readwrite.' And the deletion for a file on the lower
941 +readonly branch is handled by creating 'whiteout' on the upper writable
942 +branch.
943 +
944 +On LKML, there have been discussions about UnionMount (Jan Blunck,
945 +Bharata B Rao and Valerie Aurora) and Unionfs (Erez Zadok). They took
946 +different approaches to implement the merged-view.
947 +The former tries putting it into VFS, and the latter implements as a
948 +separate filesystem.
949 +(If I misunderstand about these implementations, please let me know and
950 +I shall correct it. Because it is a long time ago when I read their
951 +source files last time).
952 +
953 +UnionMount's approach will be able to small, but may be hard to share
954 +branches between several UnionMount since the whiteout in it is
955 +implemented in the inode on branch filesystem and always
956 +shared. According to Bharata's post, readdir does not seems to be
957 +finished yet.
958 +There are several missing features known in this implementations such as
959 +- for users, the inode number may change silently. eg. copy-up.
960 +- link(2) may break by copy-up.
961 +- read(2) may get an obsoleted filedata (fstat(2) too).
962 +- fcntl(F_SETLK) may be broken by copy-up.
963 +- unnecessary copy-up may happen, for example mmap(MAP_PRIVATE) after
964 +  open(O_RDWR).
965 +
966 +Unionfs has a longer history. When I started implementing a stacking filesystem
967 +(Aug 2005), it already existed. It has virtual super_block, inode,
968 +dentry and file objects and they have an array pointing lower same kind
969 +objects. After contributing many patches for Unionfs, I re-started my
970 +project AUFS (Jun 2006).
971 +
972 +In AUFS, the structure of filesystem resembles to Unionfs, but I
973 +implemented my own ideas, approaches and enhancements and it became
974 +totally different one.
975 +
976 +Comparing DM snapshot and fs based implementation
977 +- the number of bytes to be copied between devices is much smaller.
978 +- the type of filesystem must be one and only.
979 +- the fs must be writable, no readonly fs, even for the lower original
980 +  device. so the compression fs will not be usable. but if we use
981 +  loopback mount, we may address this issue.
982 +  for instance,
983 +       mount /cdrom/squashfs.img /sq
984 +       losetup /sq/ext2.img
985 +       losetup /somewhere/cow
986 +       dmsetup "snapshot /dev/loop0 /dev/loop1 ..."
987 +- it will be difficult (or needs more operations) to extract the
988 +  difference between the original device and COW.
989 +- DM snapshot-merge may help a lot when users try merging. in the
990 +  fs-layer union, users will use rsync(1).
991 +
992 +
993 +Several characters/aspects of aufs
994 +----------------------------------------------------------------------
995 +
996 +Aufs has several characters or aspects.
997 +1. a filesystem, callee of VFS helper
998 +2. sub-VFS, caller of VFS helper for branches
999 +3. a virtual filesystem which maintains persistent inode number
1000 +4. reader/writer of files on branches such like an application
1001 +
1002 +1. Callee of VFS Helper
1003 +As an ordinary linux filesystem, aufs is a callee of VFS. For instance,
1004 +unlink(2) from an application reaches sys_unlink() kernel function and
1005 +then vfs_unlink() is called. vfs_unlink() is one of VFS helper and it
1006 +calls filesystem specific unlink operation. Actually aufs implements the
1007 +unlink operation but it behaves like a redirector.
1008 +
1009 +2. Caller of VFS Helper for Branches
1010 +aufs_unlink() passes the unlink request to the branch filesystem as if
1011 +it were called from VFS. So the called unlink operation of the branch
1012 +filesystem acts as usual. As a caller of VFS helper, aufs should handle
1013 +every necessary pre/post operation for the branch filesystem.
1014 +- acquire the lock for the parent dir on a branch
1015 +- lookup in a branch
1016 +- revalidate dentry on a branch
1017 +- mnt_want_write() for a branch
1018 +- vfs_unlink() for a branch
1019 +- mnt_drop_write() for a branch
1020 +- release the lock on a branch
1021 +
1022 +3. Persistent Inode Number
1023 +One of the most important issue for a filesystem is to maintain inode
1024 +numbers. This is particularly important to support exporting a
1025 +filesystem via NFS. Aufs is a virtual filesystem which doesn't have a
1026 +backend block device for its own. But some storage is necessary to
1027 +maintain inode number. It may be a large space and may not suit to keep
1028 +in memory. Aufs rents some space from its first writable branch
1029 +filesystem (by default) and creates file(s) on it. These files are
1030 +created by aufs internally and removed soon (currently) keeping opened.
1031 +Note: Because these files are removed, they are totally gone after
1032 +      unmounting aufs. It means the inode numbers are not persistent
1033 +      across unmount or reboot. I have a plan to make them really
1034 +      persistent which will be important for aufs on NFS server.
1035 +
1036 +4. Read/Write Files Internally (copy-on-write)
1037 +Because a branch can be readonly, when you write a file on it, aufs will
1038 +"copy-up" it to the upper writable branch internally. And then write the
1039 +originally requested thing to the file. Generally kernel doesn't
1040 +open/read/write file actively. In aufs, even a single write may cause a
1041 +internal "file copy". This behaviour is very similar to cp(1) command.
1042 +
1043 +Some people may think it is better to pass such work to user space
1044 +helper, instead of doing in kernel space. Actually I am still thinking
1045 +about it. But currently I have implemented it in kernel space.
1046 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt linux/Documentation/filesystems/aufs/design/02struct.txt
1047 --- /usr/share/empty/Documentation/filesystems/aufs/design/02struct.txt 1970-01-01 01:00:00.000000000 +0100
1048 +++ linux/Documentation/filesystems/aufs/design/02struct.txt    2014-04-24 22:11:10.471931783 +0200
1049 @@ -0,0 +1,242 @@
1050 +
1051 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1052 +# 
1053 +# This program is free software; you can redistribute it and/or modify
1054 +# it under the terms of the GNU General Public License as published by
1055 +# the Free Software Foundation; either version 2 of the License, or
1056 +# (at your option) any later version.
1057 +# 
1058 +# This program is distributed in the hope that it will be useful,
1059 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1060 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1061 +# GNU General Public License for more details.
1062 +# 
1063 +# You should have received a copy of the GNU General Public License
1064 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1065 +
1066 +Basic Aufs Internal Structure
1067 +
1068 +Superblock/Inode/Dentry/File Objects
1069 +----------------------------------------------------------------------
1070 +As like an ordinary filesystem, aufs has its own
1071 +superblock/inode/dentry/file objects. All these objects have a
1072 +dynamically allocated array and store the same kind of pointers to the
1073 +lower filesystem, branch.
1074 +For example, when you build a union with one readwrite branch and one
1075 +readonly, mounted /au, /rw and /ro respectively.
1076 +- /au = /rw + /ro
1077 +- /ro/fileA exists but /rw/fileA
1078 +
1079 +Aufs lookup operation finds /ro/fileA and gets dentry for that. These
1080 +pointers are stored in a aufs dentry. The array in aufs dentry will be,
1081 +- [0] = NULL
1082 +- [1] = /ro/fileA
1083 +
1084 +This style of an array is essentially same to the aufs
1085 +superblock/inode/dentry/file objects.
1086 +
1087 +Because aufs supports manipulating branches, ie. add/delete/change
1088 +dynamically, these objects has its own generation. When branches are
1089 +changed, the generation in aufs superblock is incremented. And a
1090 +generation in other object are compared when it is accessed.
1091 +When a generation in other objects are obsoleted, aufs refreshes the
1092 +internal array.
1093 +
1094 +
1095 +Superblock
1096 +----------------------------------------------------------------------
1097 +Additionally aufs superblock has some data for policies to select one
1098 +among multiple writable branches, XIB files, pseudo-links and kobject.
1099 +See below in detail.
1100 +About the policies which supports copy-down a directory, see policy.txt
1101 +too.
1102 +
1103 +
1104 +Branch and XINO(External Inode Number Translation Table)
1105 +----------------------------------------------------------------------
1106 +Every branch has its own xino (external inode number translation table)
1107 +file. The xino file is created and unlinked by aufs internally. When two
1108 +members of a union exist on the same filesystem, they share the single
1109 +xino file.
1110 +The struct of a xino file is simple, just a sequence of aufs inode
1111 +numbers which is indexed by the lower inode number.
1112 +In the above sample, assume the inode number of /ro/fileA is i111 and
1113 +aufs assigns the inode number i999 for fileA. Then aufs writes 999 as
1114 +4(8) bytes at 111 * 4(8) bytes offset in the xino file.
1115 +
1116 +When the inode numbers are not contiguous, the xino file will be sparse
1117 +which has a hole in it and doesn't consume as much disk space as it
1118 +might appear. If your branch filesystem consumes disk space for such
1119 +holes, then you should specify 'xino=' option at mounting aufs.
1120 +
1121 +Also a writable branch has three kinds of "whiteout bases". All these
1122 +are existed when the branch is joined to aufs and the names are
1123 +whiteout-ed doubly, so that users will never see their names in aufs
1124 +hierarchy.
1125 +1. a regular file which will be linked to all whiteouts.
1126 +2. a directory to store a pseudo-link.
1127 +3. a directory to store an "orphan-ed" file temporary.
1128 +
1129 +1. Whiteout Base
1130 +   When you remove a file on a readonly branch, aufs handles it as a
1131 +   logical deletion and creates a whiteout on the upper writable branch
1132 +   as a hardlink of this file in order not to consume inode on the
1133 +   writable branch.
1134 +2. Pseudo-link Dir
1135 +   See below, Pseudo-link.
1136 +3. Step-Parent Dir
1137 +   When "fileC" exists on the lower readonly branch only and it is
1138 +   opened and removed with its parent dir, and then user writes
1139 +   something into it, then aufs copies-up fileC to this
1140 +   directory. Because there is no other dir to store fileC. After
1141 +   creating a file under this dir, the file is unlinked.
1142 +
1143 +Because aufs supports manipulating branches, ie. add/delete/change
1144 +dynamically, a branch has its own id. When the branch order changes, aufs
1145 +finds the new index by searching the branch id.
1146 +
1147 +
1148 +Pseudo-link
1149 +----------------------------------------------------------------------
1150 +Assume "fileA" exists on the lower readonly branch only and it is
1151 +hardlinked to "fileB" on the branch. When you write something to fileA,
1152 +aufs copies-up it to the upper writable branch. Additionally aufs
1153 +creates a hardlink under the Pseudo-link Directory of the writable
1154 +branch. The inode of a pseudo-link is kept in aufs super_block as a
1155 +simple list. If fileB is read after unlinking fileA, aufs returns
1156 +filedata from the pseudo-link instead of the lower readonly
1157 +branch. Because the pseudo-link is based upon the inode, to keep the
1158 +inode number by xino (see above) is important.
1159 +
1160 +All the hardlinks under the Pseudo-link Directory of the writable branch
1161 +should be restored in a proper location later. Aufs provides a utility
1162 +to do this. The userspace helpers executed at remounting and unmounting
1163 +aufs by default.
1164 +During this utility is running, it puts aufs into the pseudo-link
1165 +maintenance mode. In this mode, only the process which began the
1166 +maintenance mode (and its child processes) is allowed to operate in
1167 +aufs. Some other processes which are not related to the pseudo-link will
1168 +be allowed to run too, but the rest have to return an error or wait
1169 +until the maintenance mode ends. If a process already acquires an inode
1170 +mutex (in VFS), it has to return an error.
1171 +
1172 +
1173 +XIB(external inode number bitmap)
1174 +----------------------------------------------------------------------
1175 +Addition to the xino file per a branch, aufs has an external inode number
1176 +bitmap in a superblock object. It is also a file such like a xino file.
1177 +It is a simple bitmap to mark whether the aufs inode number is in-use or
1178 +not.
1179 +To reduce the file I/O, aufs prepares a single memory page to cache xib.
1180 +
1181 +Aufs implements a feature to truncate/refresh both of xino and xib to
1182 +reduce the number of consumed disk blocks for these files.
1183 +
1184 +
1185 +Virtual or Vertical Dir, and Readdir in Userspace
1186 +----------------------------------------------------------------------
1187 +In order to support multiple layers (branches), aufs readdir operation
1188 +constructs a virtual dir block on memory. For readdir, aufs calls
1189 +vfs_readdir() internally for each dir on branches, merges their entries
1190 +with eliminating the whiteout-ed ones, and sets it to file (dir)
1191 +object. So the file object has its entry list until it is closed. The
1192 +entry list will be updated when the file position is zero and becomes
1193 +old. This decision is made in aufs automatically.
1194 +
1195 +The dynamically allocated memory block for the name of entries has a
1196 +unit of 512 bytes (by default) and stores the names contiguously (no
1197 +padding). Another block for each entry is handled by kmem_cache too.
1198 +During building dir blocks, aufs creates hash list and judging whether
1199 +the entry is whiteouted by its upper branch or already listed.
1200 +The merged result is cached in the corresponding inode object and
1201 +maintained by a customizable life-time option.
1202 +
1203 +Some people may call it can be a security hole or invite DoS attack
1204 +since the opened and once readdir-ed dir (file object) holds its entry
1205 +list and becomes a pressure for system memory. But I'd say it is similar
1206 +to files under /proc or /sys. The virtual files in them also holds a
1207 +memory page (generally) while they are opened. When an idea to reduce
1208 +memory for them is introduced, it will be applied to aufs too.
1209 +For those who really hate this situation, I've developed readdir(3)
1210 +library which operates this merging in userspace. You just need to set
1211 +LD_PRELOAD environment variable, and aufs will not consume no memory in
1212 +kernel space for readdir(3).
1213 +
1214 +
1215 +Workqueue
1216 +----------------------------------------------------------------------
1217 +Aufs sometimes requires privilege access to a branch. For instance,
1218 +in copy-up/down operation. When a user process is going to make changes
1219 +to a file which exists in the lower readonly branch only, and the mode
1220 +of one of ancestor directories may not be writable by a user
1221 +process. Here aufs copy-up the file with its ancestors and they may
1222 +require privilege to set its owner/group/mode/etc.
1223 +This is a typical case of a application character of aufs (see
1224 +Introduction).
1225 +
1226 +Aufs uses workqueue synchronously for this case. It creates its own
1227 +workqueue. The workqueue is a kernel thread and has privilege. Aufs
1228 +passes the request to call mkdir or write (for example), and wait for
1229 +its completion. This approach solves a problem of a signal handler
1230 +simply.
1231 +If aufs didn't adopt the workqueue and changed the privilege of the
1232 +process, and if the mkdir/write call arises SIGXFSZ or other signal,
1233 +then the user process might gain a privilege or the generated core file
1234 +was owned by a superuser.
1235 +
1236 +Also aufs uses the system global workqueue ("events" kernel thread) too
1237 +for asynchronous tasks, such like handling inotify/fsnotify, re-creating a
1238 +whiteout base and etc. This is unrelated to a privilege.
1239 +Most of aufs operation tries acquiring a rw_semaphore for aufs
1240 +superblock at the beginning, at the same time waits for the completion
1241 +of all queued asynchronous tasks.
1242 +
1243 +
1244 +Whiteout
1245 +----------------------------------------------------------------------
1246 +The whiteout in aufs is very similar to Unionfs's. That is represented
1247 +by its filename. UnionMount takes an approach of a file mode, but I am
1248 +afraid several utilities (find(1) or something) will have to support it.
1249 +
1250 +Basically the whiteout represents "logical deletion" which stops aufs to
1251 +lookup further, but also it represents "dir is opaque" which also stop
1252 +lookup.
1253 +
1254 +In aufs, rmdir(2) and rename(2) for dir uses whiteout alternatively.
1255 +In order to make several functions in a single systemcall to be
1256 +revertible, aufs adopts an approach to rename a directory to a temporary
1257 +unique whiteouted name.
1258 +For example, in rename(2) dir where the target dir already existed, aufs
1259 +renames the target dir to a temporary unique whiteouted name before the
1260 +actual rename on a branch and then handles other actions (make it opaque,
1261 +update the attributes, etc). If an error happens in these actions, aufs
1262 +simply renames the whiteouted name back and returns an error. If all are
1263 +succeeded, aufs registers a function to remove the whiteouted unique
1264 +temporary name completely and asynchronously to the system global
1265 +workqueue.
1266 +
1267 +
1268 +Copy-up
1269 +----------------------------------------------------------------------
1270 +It is a well-known feature or concept.
1271 +When user modifies a file on a readonly branch, aufs operate "copy-up"
1272 +internally and makes change to the new file on the upper writable branch.
1273 +When the trigger systemcall does not update the timestamps of the parent
1274 +dir, aufs reverts it after copy-up.
1275 +
1276 +
1277 +Move-down (aufs3.9 and later)
1278 +----------------------------------------------------------------------
1279 +"Copy-up" is one of the essential feature in aufs. It copies a file from
1280 +the lower readonly branch to the upper writable branch when a user
1281 +changes something about the file.
1282 +"Move-down" is an opposite action of copy-up. Basically this action is
1283 +ran manually instead of automatically and internally.
1284 +
1285 +Sometimes users want to move-down a file from the upper writable branch
1286 +to the lower readonly or writable branch. For instance,
1287 +- the free space of the upper writable branch is going to run out.
1288 +- create a new intermediate branch between the upper and lower branch.
1289 +- etc.
1290 +
1291 +For this purpose, use "aumvdown" command in aufs-util.git.
1292 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt linux/Documentation/filesystems/aufs/design/03lookup.txt
1293 --- /usr/share/empty/Documentation/filesystems/aufs/design/03lookup.txt 1970-01-01 01:00:00.000000000 +0100
1294 +++ linux/Documentation/filesystems/aufs/design/03lookup.txt    2014-04-24 22:11:10.471931783 +0200
1295 @@ -0,0 +1,105 @@
1296 +
1297 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1298 +# 
1299 +# This program is free software; you can redistribute it and/or modify
1300 +# it under the terms of the GNU General Public License as published by
1301 +# the Free Software Foundation; either version 2 of the License, or
1302 +# (at your option) any later version.
1303 +# 
1304 +# This program is distributed in the hope that it will be useful,
1305 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1306 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1307 +# GNU General Public License for more details.
1308 +# 
1309 +# You should have received a copy of the GNU General Public License
1310 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1311 +
1312 +Lookup in a Branch
1313 +----------------------------------------------------------------------
1314 +Since aufs has a character of sub-VFS (see Introduction), it operates
1315 +lookup for branches as VFS does. It may be a heavy work. Generally
1316 +speaking struct nameidata is a bigger structure and includes many
1317 +information. But almost all lookup operation in aufs is the simplest
1318 +case, ie. lookup only an entry directly connected to its parent. Digging
1319 +down the directory hierarchy is unnecessary.
1320 +
1321 +VFS has a function lookup_one_len() for that use, but it is not usable
1322 +for a branch filesystem which requires struct nameidata. So aufs
1323 +implements a simple lookup wrapper function. When a branch filesystem
1324 +allows NULL as nameidata, it calls lookup_one_len(). Otherwise it builds
1325 +a simplest nameidata and calls lookup_hash().
1326 +Here aufs applies "a principle in NFSD", ie. if the filesystem supports
1327 +NFS-export, then it has to support NULL as a nameidata parameter for
1328 +->create(), ->lookup() and ->d_revalidate(). So the lookup wrapper in
1329 +aufs tests if ->s_export_op in the branch is NULL or not.
1330 +
1331 +When a branch is a remote filesystem, aufs basically trusts its
1332 +->d_revalidate(), also aufs forces the hardest revalidate tests for
1333 +them.
1334 +For d_revalidate, aufs implements three levels of revalidate tests. See
1335 +"Revalidate Dentry and UDBA" in detail.
1336 +
1337 +
1338 +Loopback Mount
1339 +----------------------------------------------------------------------
1340 +Basically aufs supports any type of filesystem and block device for a
1341 +branch (actually there are some exceptions). But it is prohibited to add
1342 +a loopback mounted one whose backend file exists in a filesystem which is
1343 +already added to aufs. The reason is to protect aufs from a recursive
1344 +lookup. If it was allowed, the aufs lookup operation might re-enter a
1345 +lookup for the loopback mounted branch in the same context, and will
1346 +cause a deadlock.
1347 +
1348 +
1349 +Revalidate Dentry and UDBA (User's Direct Branch Access)
1350 +----------------------------------------------------------------------
1351 +Generally VFS helpers re-validate a dentry as a part of lookup.
1352 +0. digging down the directory hierarchy.
1353 +1. lock the parent dir by its i_mutex.
1354 +2. lookup the final (child) entry.
1355 +3. revalidate it.
1356 +4. call the actual operation (create, unlink, etc.)
1357 +5. unlock the parent dir
1358 +
1359 +If the filesystem implements its ->d_revalidate() (step 3), then it is
1360 +called. Actually aufs implements it and checks the dentry on a branch is
1361 +still valid.
1362 +But it is not enough. Because aufs has to release the lock for the
1363 +parent dir on a branch at the end of ->lookup() (step 2) and
1364 +->d_revalidate() (step 3) while the i_mutex of the aufs dir is still
1365 +held by VFS.
1366 +If the file on a branch is changed directly, eg. bypassing aufs, after
1367 +aufs released the lock, then the subsequent operation may cause
1368 +something unpleasant result.
1369 +
1370 +This situation is a result of VFS architecture, ->lookup() and
1371 +->d_revalidate() is separated. But I never say it is wrong. It is a good
1372 +design from VFS's point of view. It is just not suitable for sub-VFS
1373 +character in aufs.
1374 +
1375 +Aufs supports such case by three level of revalidation which is
1376 +selectable by user.
1377 +1. Simple Revalidate
1378 +   Addition to the native flow in VFS's, confirm the child-parent
1379 +   relationship on the branch just after locking the parent dir on the
1380 +   branch in the "actual operation" (step 4). When this validation
1381 +   fails, aufs returns EBUSY. ->d_revalidate() (step 3) in aufs still
1382 +   checks the validation of the dentry on branches.
1383 +2. Monitor Changes Internally by Inotify/Fsnotify
1384 +   Addition to above, in the "actual operation" (step 4) aufs re-lookup
1385 +   the dentry on the branch, and returns EBUSY if it finds different
1386 +   dentry.
1387 +   Additionally, aufs sets the inotify/fsnotify watch for every dir on branches
1388 +   during it is in cache. When the event is notified, aufs registers a
1389 +   function to kernel 'events' thread by schedule_work(). And the
1390 +   function sets some special status to the cached aufs dentry and inode
1391 +   private data. If they are not cached, then aufs has nothing to
1392 +   do. When the same file is accessed through aufs (step 0-3) later,
1393 +   aufs will detect the status and refresh all necessary data.
1394 +   In this mode, aufs has to ignore the event which is fired by aufs
1395 +   itself.
1396 +3. No Extra Validation
1397 +   This is the simplest test and doesn't add any additional revalidation
1398 +   test, and skip therevalidatin in step 4. It is useful and improves
1399 +   aufs performance when system surely hide the aufs branches from user,
1400 +   by over-mounting something (or another method).
1401 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt linux/Documentation/filesystems/aufs/design/04branch.txt
1402 --- /usr/share/empty/Documentation/filesystems/aufs/design/04branch.txt 1970-01-01 01:00:00.000000000 +0100
1403 +++ linux/Documentation/filesystems/aufs/design/04branch.txt    2014-04-24 22:11:10.471931783 +0200
1404 @@ -0,0 +1,75 @@
1405 +
1406 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1407 +# 
1408 +# This program is free software; you can redistribute it and/or modify
1409 +# it under the terms of the GNU General Public License as published by
1410 +# the Free Software Foundation; either version 2 of the License, or
1411 +# (at your option) any later version.
1412 +# 
1413 +# This program is distributed in the hope that it will be useful,
1414 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1415 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1416 +# GNU General Public License for more details.
1417 +# 
1418 +# You should have received a copy of the GNU General Public License
1419 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1420 +
1421 +Branch Manipulation
1422 +
1423 +Since aufs supports dynamic branch manipulation, ie. add/remove a branch
1424 +and changing its permission/attribute, there are a lot of works to do.
1425 +
1426 +
1427 +Add a Branch
1428 +----------------------------------------------------------------------
1429 +o Confirm the adding dir exists outside of aufs, including loopback
1430 +  mount.
1431 +- and other various attributes...
1432 +o Initialize the xino file and whiteout bases if necessary.
1433 +  See struct.txt.
1434 +
1435 +o Check the owner/group/mode of the directory
1436 +  When the owner/group/mode of the adding directory differs from the
1437 +  existing branch, aufs issues a warning because it may impose a
1438 +  security risk.
1439 +  For example, when a upper writable branch has a world writable empty
1440 +  top directory, a malicious user can create any files on the writable
1441 +  branch directly, like copy-up and modify manually. If something like
1442 +  /etc/{passwd,shadow} exists on the lower readonly branch but the upper
1443 +  writable branch, and the writable branch is world-writable, then a
1444 +  malicious guy may create /etc/passwd on the writable branch directly
1445 +  and the infected file will be valid in aufs.
1446 +  I am afraid it can be a security issue, but nothing to do except
1447 +  producing a warning.
1448 +
1449 +
1450 +Delete a Branch
1451 +----------------------------------------------------------------------
1452 +o Confirm the deleting branch is not busy
1453 +  To be general, there is one merit to adopt "remount" interface to
1454 +  manipulate branches. It is to discard caches. At deleting a branch,
1455 +  aufs checks the still cached (and connected) dentries and inodes. If
1456 +  there are any, then they are all in-use. An inode without its
1457 +  corresponding dentry can be alive alone (for example, inotify/fsnotify case).
1458 +
1459 +  For the cached one, aufs checks whether the same named entry exists on
1460 +  other branches.
1461 +  If the cached one is a directory, because aufs provides a merged view
1462 +  to users, as long as one dir is left on any branch aufs can show the
1463 +  dir to users. In this case, the branch can be removed from aufs.
1464 +  Otherwise aufs rejects deleting the branch.
1465 +
1466 +  If any file on the deleting branch is opened by aufs, then aufs
1467 +  rejects deleting.
1468 +
1469 +
1470 +Modify the Permission of a Branch
1471 +----------------------------------------------------------------------
1472 +o Re-initialize or remove the xino file and whiteout bases if necessary.
1473 +  See struct.txt.
1474 +
1475 +o rw --> ro: Confirm the modifying branch is not busy
1476 +  Aufs rejects the request if any of these conditions are true.
1477 +  - a file on the branch is mmap-ed.
1478 +  - a regular file on the branch is opened for write and there is no
1479 +    same named entry on the upper branch.
1480 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt linux/Documentation/filesystems/aufs/design/05wbr_policy.txt
1481 --- /usr/share/empty/Documentation/filesystems/aufs/design/05wbr_policy.txt     1970-01-01 01:00:00.000000000 +0100
1482 +++ linux/Documentation/filesystems/aufs/design/05wbr_policy.txt        2014-04-24 22:11:10.471931783 +0200
1483 @@ -0,0 +1,64 @@
1484 +
1485 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1486 +# 
1487 +# This program is free software; you can redistribute it and/or modify
1488 +# it under the terms of the GNU General Public License as published by
1489 +# the Free Software Foundation; either version 2 of the License, or
1490 +# (at your option) any later version.
1491 +# 
1492 +# This program is distributed in the hope that it will be useful,
1493 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1494 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1495 +# GNU General Public License for more details.
1496 +# 
1497 +# You should have received a copy of the GNU General Public License
1498 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1499 +
1500 +Policies to Select One among Multiple Writable Branches
1501 +----------------------------------------------------------------------
1502 +When the number of writable branch is more than one, aufs has to decide
1503 +the target branch for file creation or copy-up. By default, the highest
1504 +writable branch which has the parent (or ancestor) dir of the target
1505 +file is chosen (top-down-parent policy).
1506 +By user's request, aufs implements some other policies to select the
1507 +writable branch, for file creation two policies, round-robin and
1508 +most-free-space policies. For copy-up three policies, top-down-parent,
1509 +bottom-up-parent and bottom-up policies.
1510 +
1511 +As expected, the round-robin policy selects the branch in circular. When
1512 +you have two writable branches and creates 10 new files, 5 files will be
1513 +created for each branch. mkdir(2) systemcall is an exception. When you
1514 +create 10 new directories, all will be created on the same branch.
1515 +And the most-free-space policy selects the one which has most free
1516 +space among the writable branches. The amount of free space will be
1517 +checked by aufs internally, and users can specify its time interval.
1518 +
1519 +The policies for copy-up is more simple,
1520 +top-down-parent is equivalent to the same named on in create policy,
1521 +bottom-up-parent selects the writable branch where the parent dir
1522 +exists and the nearest upper one from the copyup-source,
1523 +bottom-up selects the nearest upper writable branch from the
1524 +copyup-source, regardless the existence of the parent dir.
1525 +
1526 +There are some rules or exceptions to apply these policies.
1527 +- If there is a readonly branch above the policy-selected branch and
1528 +  the parent dir is marked as opaque (a variation of whiteout), or the
1529 +  target (creating) file is whiteout-ed on the upper readonly branch,
1530 +  then the result of the policy is ignored and the target file will be
1531 +  created on the nearest upper writable branch than the readonly branch.
1532 +- If there is a writable branch above the policy-selected branch and
1533 +  the parent dir is marked as opaque or the target file is whiteouted
1534 +  on the branch, then the result of the policy is ignored and the target
1535 +  file will be created on the highest one among the upper writable
1536 +  branches who has diropq or whiteout. In case of whiteout, aufs removes
1537 +  it as usual.
1538 +- link(2) and rename(2) systemcalls are exceptions in every policy.
1539 +  They try selecting the branch where the source exists as possible
1540 +  since copyup a large file will take long time. If it can't be,
1541 +  ie. the branch where the source exists is readonly, then they will
1542 +  follow the copyup policy.
1543 +- There is an exception for rename(2) when the target exists.
1544 +  If the rename target exists, aufs compares the index of the branches
1545 +  where the source and the target exists and selects the higher
1546 +  one. If the selected branch is readonly, then aufs follows the
1547 +  copyup policy.
1548 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt linux/Documentation/filesystems/aufs/design/06mmap.txt
1549 --- /usr/share/empty/Documentation/filesystems/aufs/design/06mmap.txt   1970-01-01 01:00:00.000000000 +0100
1550 +++ linux/Documentation/filesystems/aufs/design/06mmap.txt      2014-04-24 22:11:10.471931783 +0200
1551 @@ -0,0 +1,46 @@
1552 +
1553 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1554 +# 
1555 +# This program is free software; you can redistribute it and/or modify
1556 +# it under the terms of the GNU General Public License as published by
1557 +# the Free Software Foundation; either version 2 of the License, or
1558 +# (at your option) any later version.
1559 +# 
1560 +# This program is distributed in the hope that it will be useful,
1561 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1562 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1563 +# GNU General Public License for more details.
1564 +# 
1565 +# You should have received a copy of the GNU General Public License
1566 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1567 +
1568 +mmap(2) -- File Memory Mapping
1569 +----------------------------------------------------------------------
1570 +In aufs, the file-mapped pages are handled by a branch fs directly, no
1571 +interaction with aufs. It means aufs_mmap() calls the branch fs's
1572 +->mmap().
1573 +This approach is simple and good, but there is one problem.
1574 +Under /proc, several entries show the mmap-ped files by its path (with
1575 +device and inode number), and the printed path will be the path on the
1576 +branch fs's instead of virtual aufs's.
1577 +This is not a problem in most cases, but some utilities lsof(1) (and its
1578 +user) may expect the path on aufs.
1579 +
1580 +To address this issue, aufs adds a new member called vm_prfile in struct
1581 +vm_area_struct (and struct vm_region). The original vm_file points to
1582 +the file on the branch fs in order to handle everything correctly as
1583 +usual. The new vm_prfile points to a virtual file in aufs, and the
1584 +show-functions in procfs refers to vm_prfile if it is set.
1585 +Also we need to maintain several other places where touching vm_file
1586 +such like
1587 +- fork()/clone() copies vma and the reference count of vm_file is
1588 +  incremented.
1589 +- merging vma maintains the ref count too.
1590 +
1591 +This is not a good approach. It just faking the printed path. But it
1592 +leaves all behaviour around f_mapping unchanged. This is surely an
1593 +advantage.
1594 +Actually aufs had adopted another complicated approach which calls
1595 +generic_file_mmap() and handles struct vm_operations_struct. In this
1596 +approach, aufs met a hard problem and I could not solve it without
1597 +switching the approach.
1598 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt linux/Documentation/filesystems/aufs/design/07export.txt
1599 --- /usr/share/empty/Documentation/filesystems/aufs/design/07export.txt 1970-01-01 01:00:00.000000000 +0100
1600 +++ linux/Documentation/filesystems/aufs/design/07export.txt    2014-04-24 22:11:10.471931783 +0200
1601 @@ -0,0 +1,58 @@
1602 +
1603 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1604 +# 
1605 +# This program is free software; you can redistribute it and/or modify
1606 +# it under the terms of the GNU General Public License as published by
1607 +# the Free Software Foundation; either version 2 of the License, or
1608 +# (at your option) any later version.
1609 +# 
1610 +# This program is distributed in the hope that it will be useful,
1611 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1612 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1613 +# GNU General Public License for more details.
1614 +# 
1615 +# You should have received a copy of the GNU General Public License
1616 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1617 +
1618 +Export Aufs via NFS
1619 +----------------------------------------------------------------------
1620 +Here is an approach.
1621 +- like xino/xib, add a new file 'xigen' which stores aufs inode
1622 +  generation.
1623 +- iget_locked(): initialize aufs inode generation for a new inode, and
1624 +  store it in xigen file.
1625 +- destroy_inode(): increment aufs inode generation and store it in xigen
1626 +  file. it is necessary even if it is not unlinked, because any data of
1627 +  inode may be changed by UDBA.
1628 +- encode_fh(): for a root dir, simply return FILEID_ROOT. otherwise
1629 +  build file handle by
1630 +  + branch id (4 bytes)
1631 +  + superblock generation (4 bytes)
1632 +  + inode number (4 or 8 bytes)
1633 +  + parent dir inode number (4 or 8 bytes)
1634 +  + inode generation (4 bytes))
1635 +  + return value of exportfs_encode_fh() for the parent on a branch (4
1636 +    bytes)
1637 +  + file handle for a branch (by exportfs_encode_fh())
1638 +- fh_to_dentry():
1639 +  + find the index of a branch from its id in handle, and check it is
1640 +    still exist in aufs.
1641 +  + 1st level: get the inode number from handle and search it in cache.
1642 +  + 2nd level: if not found, get the parent inode number from handle and
1643 +    search it in cache. and then open the parent dir, find the matching
1644 +    inode number by vfs_readdir() and get its name, and call
1645 +    lookup_one_len() for the target dentry.
1646 +  + 3rd level: if the parent dir is not cached, call
1647 +    exportfs_decode_fh() for a branch and get the parent on a branch,
1648 +    build a pathname of it, convert it a pathname in aufs, call
1649 +    path_lookup(). now aufs gets a parent dir dentry, then handle it as
1650 +    the 2nd level.
1651 +  + to open the dir, aufs needs struct vfsmount. aufs keeps vfsmount
1652 +    for every branch, but not itself. to get this, (currently) aufs
1653 +    searches in current->nsproxy->mnt_ns list. it may not be a good
1654 +    idea, but I didn't get other approach.
1655 +  + test the generation of the gotten inode.
1656 +- every inode operation: they may get EBUSY due to UDBA. in this case,
1657 +  convert it into ESTALE for NFSD.
1658 +- readdir(): call lockdep_on/off() because filldir in NFSD calls
1659 +  lookup_one_len(), vfs_getattr(), encode_fh() and others.
1660 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt linux/Documentation/filesystems/aufs/design/08shwh.txt
1661 --- /usr/share/empty/Documentation/filesystems/aufs/design/08shwh.txt   1970-01-01 01:00:00.000000000 +0100
1662 +++ linux/Documentation/filesystems/aufs/design/08shwh.txt      2014-04-24 22:11:10.471931783 +0200
1663 @@ -0,0 +1,52 @@
1664 +
1665 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1666 +# 
1667 +# This program is free software; you can redistribute it and/or modify
1668 +# it under the terms of the GNU General Public License as published by
1669 +# the Free Software Foundation; either version 2 of the License, or
1670 +# (at your option) any later version.
1671 +# 
1672 +# This program is distributed in the hope that it will be useful,
1673 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1674 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1675 +# GNU General Public License for more details.
1676 +# 
1677 +# You should have received a copy of the GNU General Public License
1678 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1679 +
1680 +Show Whiteout Mode (shwh)
1681 +----------------------------------------------------------------------
1682 +Generally aufs hides the name of whiteouts. But in some cases, to show
1683 +them is very useful for users. For instance, creating a new middle layer
1684 +(branch) by merging existing layers.
1685 +
1686 +(borrowing aufs1 HOW-TO from a user, Michael Towers)
1687 +When you have three branches,
1688 +- Bottom: 'system', squashfs (underlying base system), read-only
1689 +- Middle: 'mods', squashfs, read-only
1690 +- Top: 'overlay', ram (tmpfs), read-write
1691 +
1692 +The top layer is loaded at boot time and saved at shutdown, to preserve
1693 +the changes made to the system during the session.
1694 +When larger changes have been made, or smaller changes have accumulated,
1695 +the size of the saved top layer data grows. At this point, it would be
1696 +nice to be able to merge the two overlay branches ('mods' and 'overlay')
1697 +and rewrite the 'mods' squashfs, clearing the top layer and thus
1698 +restoring save and load speed.
1699 +
1700 +This merging is simplified by the use of another aufs mount, of just the
1701 +two overlay branches using the 'shwh' option.
1702 +# mount -t aufs -o ro,shwh,br:/livesys/overlay=ro+wh:/livesys/mods=rr+wh \
1703 +       aufs /livesys/merge_union
1704 +
1705 +A merged view of these two branches is then available at
1706 +/livesys/merge_union, and the new feature is that the whiteouts are
1707 +visible!
1708 +Note that in 'shwh' mode the aufs mount must be 'ro', which will disable
1709 +writing to all branches. Also the default mode for all branches is 'ro'.
1710 +It is now possible to save the combined contents of the two overlay
1711 +branches to a new squashfs, e.g.:
1712 +# mksquashfs /livesys/merge_union /path/to/newmods.squash
1713 +
1714 +This new squashfs archive can be stored on the boot device and the
1715 +initramfs will use it to replace the old one at the next boot.
1716 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt linux/Documentation/filesystems/aufs/design/10dynop.txt
1717 --- /usr/share/empty/Documentation/filesystems/aufs/design/10dynop.txt  1970-01-01 01:00:00.000000000 +0100
1718 +++ linux/Documentation/filesystems/aufs/design/10dynop.txt     2014-04-24 22:11:10.471931783 +0200
1719 @@ -0,0 +1,46 @@
1720 +
1721 +# Copyright (C) 2010-2014 Junjiro R. Okajima
1722 +#
1723 +# This program is free software; you can redistribute it and/or modify
1724 +# it under the terms of the GNU General Public License as published by
1725 +# the Free Software Foundation; either version 2 of the License, or
1726 +# (at your option) any later version.
1727 +#
1728 +# This program is distributed in the hope that it will be useful,
1729 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1730 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1731 +# GNU General Public License for more details.
1732 +#
1733 +# You should have received a copy of the GNU General Public License
1734 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1735 +
1736 +Dynamically customizable FS operations
1737 +----------------------------------------------------------------------
1738 +Generally FS operations (struct inode_operations, struct
1739 +address_space_operations, struct file_operations, etc.) are defined as
1740 +"static const", but it never means that FS have only one set of
1741 +operation. Some FS have multiple sets of them. For instance, ext2 has
1742 +three sets, one for XIP, for NOBH, and for normal.
1743 +Since aufs overrides and redirects these operations, sometimes aufs has
1744 +to change its behaviour according to the branch FS type. More imporantly
1745 +VFS acts differently if a function (member in the struct) is set or
1746 +not. It means aufs should have several sets of operations and select one
1747 +among them according to the branch FS definition.
1748 +
1749 +In order to solve this problem and not to affect the behavour of VFS,
1750 +aufs defines these operations dynamically. For instance, aufs defines
1751 +aio_read function for struct file_operations, but it may not be set to
1752 +the file_operations. When the branch FS doesn't have it, aufs doesn't
1753 +set it to its file_operations while the function definition itself is
1754 +still alive. So the behaviour of io_submit(2) will not change, and it
1755 +will return an error when aio_read is not defined.
1756 +
1757 +The lifetime of these dynamically generated operation object is
1758 +maintained by aufs branch object. When the branch is removed from aufs,
1759 +the reference counter of the object is decremented. When it reaches
1760 +zero, the dynamically generated operation object will be freed.
1761 +
1762 +This approach is designed to support AIO (io_submit), Direcit I/O and
1763 +XIP mainly.
1764 +Currently this approach is applied to file_operations and
1765 +vm_operations_struct for regular files only.
1766 diff -urN /usr/share/empty/Documentation/filesystems/aufs/design/99plan.txt linux/Documentation/filesystems/aufs/design/99plan.txt
1767 --- /usr/share/empty/Documentation/filesystems/aufs/design/99plan.txt   1970-01-01 01:00:00.000000000 +0100
1768 +++ linux/Documentation/filesystems/aufs/design/99plan.txt      2014-04-24 22:11:10.471931783 +0200
1769 @@ -0,0 +1,95 @@
1770 +
1771 +# Copyright (C) 2005-2014 Junjiro R. Okajima
1772 +# 
1773 +# This program is free software; you can redistribute it and/or modify
1774 +# it under the terms of the GNU General Public License as published by
1775 +# the Free Software Foundation; either version 2 of the License, or
1776 +# (at your option) any later version.
1777 +# 
1778 +# This program is distributed in the hope that it will be useful,
1779 +# but WITHOUT ANY WARRANTY; without even the implied warranty of
1780 +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
1781 +# GNU General Public License for more details.
1782 +# 
1783 +# You should have received a copy of the GNU General Public License
1784 +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
1785 +
1786 +Plan
1787 +
1788 +Restoring some features which was implemented in aufs1.
1789 +They were dropped in aufs2 in order to make source files simpler and
1790 +easier to be reviewed.
1791 +
1792 +
1793 +Test Only the Highest One for the Directory Permission (dirperm1 option)
1794 +----------------------------------------------------------------------
1795 +Let's try case study.
1796 +- aufs has two branches, upper readwrite and lower readonly.
1797 +  /au = /rw + /ro
1798 +- "dirA" exists under /ro, but /rw. and its mode is 0700.
1799 +- user invoked "chmod a+rx /au/dirA"
1800 +- then "dirA" becomes world readable?
1801 +
1802 +In this case, /ro/dirA is still 0700 since it exists in readonly branch,
1803 +or it may be a natively readonly filesystem. If aufs respects the lower
1804 +branch, it should not respond readdir request from other users. But user
1805 +allowed it by chmod. Should really aufs rejects showing the entries
1806 +under /ro/dirA?
1807 +
1808 +To be honest, I don't have a best solution for this case. So I
1809 +implemented 'dirperm1' and 'nodirperm1' option in aufs1, and leave it to
1810 +users.
1811 +When dirperm1 is specified, aufs checks only the highest one for the
1812 +directory permission, and shows the entries. Otherwise, as usual, checks
1813 +every dir existing on all branches and rejects the request.
1814 +
1815 +As a side effect, dirperm1 option improves the performance of aufs
1816 +because the number of permission check is reduced.
1817 +
1818 +
1819 +Being Another Aufs's Readonly Branch (robr)
1820 +----------------------------------------------------------------------
1821 +Aufs1 allows aufs to be another aufs's readonly branch.
1822 +This feature was developed by a user's request. But it may not be used
1823 +currecnly.
1824 +
1825 +
1826 +Copy-up on Open (coo=)
1827 +----------------------------------------------------------------------
1828 +By default the internal copy-up is executed when it is really necessary.
1829 +It is not done when a file is opened for writing, but when write(2) is
1830 +done. Users who have many (over 100) branches want to know and analyse
1831 +when and what file is copied-up. To insert a new upper branch which
1832 +contains such files only may improve the performance of aufs.
1833 +
1834 +Aufs1 implemented "coo=none | leaf | all" option.
1835 +
1836 +
1837 +Refresh the Opened File (refrof)
1838 +----------------------------------------------------------------------
1839 +This option is implemented in aufs1 but incomplete.
1840 +
1841 +When user reads from a file, he expects to get its latest filedata
1842 +generally. If the file is removed and a new same named file is created,
1843 +the content he gets is unchanged, ie. the unlinked filedata.
1844 +
1845 +Let's try case study again.
1846 +- aufs has two branches.
1847 +  /au = /rw + /ro
1848 +- "fileA" exists under /ro, but /rw.
1849 +- user opened "/au/fileA".
1850 +- he or someone else inserts a branch (/new) between /rw and /ro.
1851 +  /au = /rw + /new + /ro
1852 +- the new branch has "fileA".
1853 +- user reads from the opened "fileA"
1854 +- which filedata should aufs return, from /ro or /new?
1855 +
1856 +Some people says it has to be "from /ro" and it is a semantics of Unix.
1857 +The others say it should be "from /new" because the file is not removed
1858 +and it is equivalent to the case of someone else modifies the file.
1859 +
1860 +Here again I don't have a best and final answer. I got an idea to
1861 +implement 'refrof' and 'norefrof' option. When 'refrof' (REFResh the
1862 +Opened File) is specified (by default), aufs returns the filedata from
1863 +/new.
1864 +Otherwise from /new.
1865 diff -urN /usr/share/empty/Documentation/filesystems/aufs/README linux/Documentation/filesystems/aufs/README
1866 --- /usr/share/empty/Documentation/filesystems/aufs/README      1970-01-01 01:00:00.000000000 +0100
1867 +++ linux/Documentation/filesystems/aufs/README 2014-04-24 22:11:10.471931783 +0200
1868 @@ -0,0 +1,344 @@
1869 +
1870 +Aufs3 -- advanced multi layered unification filesystem version 3.x
1871 +http://aufs.sf.net
1872 +Junjiro R. Okajima
1873 +
1874 +
1875 +0. Introduction
1876 +----------------------------------------
1877 +In the early days, aufs was entirely re-designed and re-implemented
1878 +Unionfs Version 1.x series. After many original ideas, approaches,
1879 +improvements and implementations, it becomes totally different from
1880 +Unionfs while keeping the basic features.
1881 +Recently, Unionfs Version 2.x series begin taking some of the same
1882 +approaches to aufs1's.
1883 +Unionfs is being developed by Professor Erez Zadok at Stony Brook
1884 +University and his team.
1885 +
1886 +Aufs3 supports linux-3.0 and later.
1887 +If you want older kernel version support, try aufs2-2.6.git or
1888 +aufs2-standalone.git repository, aufs1 from CVS on SourceForge.
1889 +
1890 +Note: it becomes clear that "Aufs was rejected. Let's give it up."
1891 +According to Christoph Hellwig, linux rejects all union-type filesystems
1892 +but UnionMount.
1893 +<http://marc.info/?l=linux-kernel&m=123938533724484&w=2>
1894 +
1895 +
1896 +1. Features
1897 +----------------------------------------
1898 +- unite several directories into a single virtual filesystem. The member
1899 +  directory is called as a branch.
1900 +- you can specify the permission flags to the branch, which are 'readonly',
1901 +  'readwrite' and 'whiteout-able.'
1902 +- by upper writable branch, internal copyup and whiteout, files/dirs on
1903 +  readonly branch are modifiable logically.
1904 +- dynamic branch manipulation, add, del.
1905 +- etc...
1906 +
1907 +Also there are many enhancements in aufs1, such as:
1908 +- readdir(3) in userspace.
1909 +- keep inode number by external inode number table
1910 +- keep the timestamps of file/dir in internal copyup operation
1911 +- seekable directory, supporting NFS readdir.
1912 +- whiteout is hardlinked in order to reduce the consumption of inodes
1913 +  on branch
1914 +- do not copyup, nor create a whiteout when it is unnecessary
1915 +- revert a single systemcall when an error occurs in aufs
1916 +- remount interface instead of ioctl
1917 +- maintain /etc/mtab by an external command, /sbin/mount.aufs.
1918 +- loopback mounted filesystem as a branch
1919 +- kernel thread for removing the dir who has a plenty of whiteouts
1920 +- support copyup sparse file (a file which has a 'hole' in it)
1921 +- default permission flags for branches
1922 +- selectable permission flags for ro branch, whether whiteout can
1923 +  exist or not
1924 +- export via NFS.
1925 +- support <sysfs>/fs/aufs and <debugfs>/aufs.
1926 +- support multiple writable branches, some policies to select one
1927 +  among multiple writable branches.
1928 +- a new semantics for link(2) and rename(2) to support multiple
1929 +  writable branches.
1930 +- no glibc changes are required.
1931 +- pseudo hardlink (hardlink over branches)
1932 +- allow a direct access manually to a file on branch, e.g. bypassing aufs.
1933 +  including NFS or remote filesystem branch.
1934 +- userspace wrapper for pathconf(3)/fpathconf(3) with _PC_LINK_MAX.
1935 +- and more...
1936 +
1937 +Currently these features are dropped temporary from aufs3.
1938 +See design/08plan.txt in detail.
1939 +- test only the highest one for the directory permission (dirperm1)
1940 +- copyup on open (coo=)
1941 +- nested mount, i.e. aufs as readonly no-whiteout branch of another aufs
1942 +  (robr)
1943 +- statistics of aufs thread (/sys/fs/aufs/stat)
1944 +- delegation mode (dlgt)
1945 +  a delegation of the internal branch access to support task I/O
1946 +  accounting, which also supports Linux Security Modules (LSM) mainly
1947 +  for Suse AppArmor.
1948 +- intent.open/create (file open in a single lookup)
1949 +
1950 +Features or just an idea in the future (see also design/*.txt),
1951 +- reorder the branch index without del/re-add.
1952 +- permanent xino files for NFSD
1953 +- an option for refreshing the opened files after add/del branches
1954 +- 'move' policy for copy-up between two writable branches, after
1955 +  checking free space.
1956 +- light version, without branch manipulation. (unnecessary?)
1957 +- copyup in userspace
1958 +- inotify in userspace
1959 +- readv/writev
1960 +- xattr, acl
1961 +
1962 +
1963 +2. Download
1964 +----------------------------------------
1965 +There were three GIT trees for aufs3, aufs3-linux.git,
1966 +aufs3-standalone.git, and aufs-util.git. Note that there is no "3" in
1967 +"aufs-util.git."
1968 +While the aufs-util is always necessary, you need either of aufs3-linux
1969 +or aufs3-standalone.
1970 +
1971 +The aufs3-linux tree includes the whole linux mainline GIT tree,
1972 +git://git.kernel.org/.../torvalds/linux.git.
1973 +And you cannot select CONFIG_AUFS_FS=m for this version, eg. you cannot
1974 +build aufs3 as an external kernel module.
1975 +
1976 +On the other hand, the aufs3-standalone tree has only aufs source files
1977 +and necessary patches, and you can select CONFIG_AUFS_FS=m.
1978 +
1979 +You will find GIT branches whose name is in form of "aufs3.x" where "x"
1980 +represents the linux kernel version, "linux-3.x". For instance,
1981 +"aufs3.0" is for linux-3.0. For latest "linux-3.x-rcN", use
1982 +"aufs3.x-rcN" branch.
1983 +
1984 +o aufs3-linux tree
1985 +$ git clone --reference /your/linux/git/tree \
1986 +       git://git.code.sf.net/p/aufs/aufs3-linux aufs-aufs3-linux \
1987 +       aufs3-linux.git
1988 +- if you don't have linux GIT tree, then remove "--reference ..."
1989 +$ cd aufs3-linux.git
1990 +$ git checkout origin/aufs3.0
1991 +
1992 +o aufs3-standalone tree
1993 +$ git clone git://git.code.sf.net/p/aufs/aufs3-standalone \
1994 +       aufs3-standalone.git
1995 +$ cd aufs3-standalone.git
1996 +$ git checkout origin/aufs3.0
1997 +
1998 +o aufs-util tree
1999 +$ git clone git://git.code.sf.net/p/aufs/aufs-util \
2000 +       aufs-util.git
2001 +$ cd aufs-util.git
2002 +$ git checkout origin/aufs3.0
2003 +
2004 +Note: The 3.x-rcN branch is to be used with `rc' kernel versions ONLY.
2005 +The minor version number, 'x' in '3.x', of aufs may not always
2006 +follow the minor version number of the kernel.
2007 +Because changes in the kernel that cause the use of a new
2008 +minor version number do not always require changes to aufs-util.
2009 +
2010 +Since aufs-util has its own minor version number, you may not be
2011 +able to find a GIT branch in aufs-util for your kernel's
2012 +exact minor version number.
2013 +In this case, you should git-checkout the branch for the
2014 +nearest lower number.
2015 +
2016 +For (an unreleased) example:
2017 +If you are using "linux-3.10" and the "aufs3.10" branch
2018 +does not exist in aufs-util repository, then "aufs3.9", "aufs3.8"
2019 +or something numerically smaller is the branch for your kernel.
2020 +
2021 +Also you can view all branches by
2022 +       $ git branch -a
2023 +
2024 +
2025 +3. Configuration and Compilation
2026 +----------------------------------------
2027 +Make sure you have git-checkout'ed the correct branch.
2028 +
2029 +For aufs3-linux tree,
2030 +- enable CONFIG_AUFS_FS.
2031 +- set other aufs configurations if necessary.
2032 +
2033 +For aufs3-standalone tree,
2034 +There are several ways to build.
2035 +
2036 +1.
2037 +- apply ./aufs3-kbuild.patch to your kernel source files.
2038 +- apply ./aufs3-base.patch too.
2039 +- apply ./aufs3-mmap.patch too.
2040 +- apply ./aufs3-standalone.patch too, if you have a plan to set
2041 +  CONFIG_AUFS_FS=m. otherwise you don't need ./aufs3-standalone.patch.
2042 +- copy ./{Documentation,fs,include/uapi/linux/aufs_type.h} files to your
2043 +  kernel source tree. Never copy $PWD/include/uapi/linux/Kbuild.
2044 +- enable CONFIG_AUFS_FS, you can select either
2045 +  =m or =y.
2046 +- and build your kernel as usual.
2047 +- install the built kernel.
2048 +  Note: Since linux-3.9, every filesystem module requires an alias
2049 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2050 +  modules.aliases file if you set CONFIG_AUFS_FS=m.
2051 +- install the header files too by "make headers_install" to the
2052 +  directory where you specify. By default, it is $PWD/usr.
2053 +  "make help" shows a brief note for headers_install.
2054 +- and reboot your system.
2055 +
2056 +2.
2057 +- module only (CONFIG_AUFS_FS=m).
2058 +- apply ./aufs3-base.patch to your kernel source files.
2059 +- apply ./aufs3-mmap.patch too.
2060 +- apply ./aufs3-standalone.patch too.
2061 +- build your kernel, don't forget "make headers_install", and reboot.
2062 +- edit ./config.mk and set other aufs configurations if necessary.
2063 +  Note: You should read $PWD/fs/aufs/Kconfig carefully which describes
2064 +  every aufs configurations.
2065 +- build the module by simple "make".
2066 +  Note: Since linux-3.9, every filesystem module requires an alias
2067 +  "fs-<fsname>". You should make sure that "fs-aufs" is listed in your
2068 +  modules.aliases file.
2069 +- you can specify ${KDIR} make variable which points to your kernel
2070 +  source tree.
2071 +- install the files
2072 +  + run "make install" to install the aufs module, or copy the built
2073 +    $PWD/aufs.ko to /lib/modules/... and run depmod -a (or reboot simply).
2074 +  + run "make install_headers" (instead of headers_install) to install
2075 +    the modified aufs header file (you can specify DESTDIR which is
2076 +    available in aufs standalone version's Makefile only), or copy
2077 +    $PWD/usr/include/linux/aufs_type.h to /usr/include/linux or wherever
2078 +    you like manually. By default, the target directory is $PWD/usr.
2079 +- no need to apply aufs3-kbuild.patch, nor copying source files to your
2080 +  kernel source tree.
2081 +
2082 +Note: The header file aufs_type.h is necessary to build aufs-util
2083 +      as well as "make headers_install" in the kernel source tree.
2084 +      headers_install is subject to be forgotten, but it is essentially
2085 +      necessary, not only for building aufs-util.
2086 +      You may not meet problems without headers_install in some older
2087 +      version though.
2088 +
2089 +And then,
2090 +- read README in aufs-util, build and install it
2091 +- note that your distribution may contain an obsoleted version of
2092 +  aufs_type.h in /usr/include/linux or something. When you build aufs
2093 +  utilities, make sure that your compiler refers the correct aufs header
2094 +  file which is built by "make headers_install."
2095 +- if you want to use readdir(3) in userspace or pathconf(3) wrapper,
2096 +  then run "make install_ulib" too. And refer to the aufs manual in
2097 +  detail.
2098 +
2099 +
2100 +4. Usage
2101 +----------------------------------------
2102 +At first, make sure aufs-util are installed, and please read the aufs
2103 +manual, aufs.5 in aufs-util.git tree.
2104 +$ man -l aufs.5
2105 +
2106 +And then,
2107 +$ mkdir /tmp/rw /tmp/aufs
2108 +# mount -t aufs -o br=/tmp/rw:${HOME} none /tmp/aufs
2109 +
2110 +Here is another example. The result is equivalent.
2111 +# mount -t aufs -o br=/tmp/rw=rw:${HOME}=ro none /tmp/aufs
2112 +  Or
2113 +# mount -t aufs -o br:/tmp/rw none /tmp/aufs
2114 +# mount -o remount,append:${HOME} /tmp/aufs
2115 +
2116 +Then, you can see whole tree of your home dir through /tmp/aufs. If
2117 +you modify a file under /tmp/aufs, the one on your home directory is
2118 +not affected, instead the same named file will be newly created under
2119 +/tmp/rw. And all of your modification to a file will be applied to
2120 +the one under /tmp/rw. This is called the file based Copy on Write
2121 +(COW) method.
2122 +Aufs mount options are described in aufs.5.
2123 +If you run chroot or something and make your aufs as a root directory,
2124 +then you need to customize the shutdown script. See the aufs manual in
2125 +detail.
2126 +
2127 +Additionally, there are some sample usages of aufs which are a
2128 +diskless system with network booting, and LiveCD over NFS.
2129 +See sample dir in CVS tree on SourceForge.
2130 +
2131 +
2132 +5. Contact
2133 +----------------------------------------
2134 +When you have any problems or strange behaviour in aufs, please let me
2135 +know with:
2136 +- /proc/mounts (instead of the output of mount(8))
2137 +- /sys/module/aufs/*
2138 +- /sys/fs/aufs/* (if you have them)
2139 +- /debug/aufs/* (if you have them)
2140 +- linux kernel version
2141 +  if your kernel is not plain, for example modified by distributor,
2142 +  the url where i can download its source is necessary too.
2143 +- aufs version which was printed at loading the module or booting the
2144 +  system, instead of the date you downloaded.
2145 +- configuration (define/undefine CONFIG_AUFS_xxx)
2146 +- kernel configuration or /proc/config.gz (if you have it)
2147 +- behaviour which you think to be incorrect
2148 +- actual operation, reproducible one is better
2149 +- mailto: aufs-users at lists.sourceforge.net
2150 +
2151 +Usually, I don't watch the Public Areas(Bugs, Support Requests, Patches,
2152 +and Feature Requests) on SourceForge. Please join and write to
2153 +aufs-users ML.
2154 +
2155 +
2156 +6. Acknowledgements
2157 +----------------------------------------
2158 +Thanks to everyone who have tried and are using aufs, whoever
2159 +have reported a bug or any feedback.
2160 +
2161 +Especially donators:
2162 +Tomas Matejicek(slax.org) made a donation (much more than once).
2163 +       Since Apr 2010, Tomas M (the author of Slax and Linux Live
2164 +       scripts) is making "doubling" donations.
2165 +       Unfortunately I cannot list all of the donators, but I really
2166 +       appreciate.
2167 +       It ends Aug 2010, but the ordinary donation URL is still available.
2168 +       <http://sourceforge.net/donate/index.php?group_id=167503>
2169 +Dai Itasaka made a donation (2007/8).
2170 +Chuck Smith made a donation (2008/4, 10 and 12).
2171 +Henk Schoneveld made a donation (2008/9).
2172 +Chih-Wei Huang, ASUS, CTC donated Eee PC 4G (2008/10).
2173 +Francois Dupoux made a donation (2008/11).
2174 +Bruno Cesar Ribas and Luis Carlos Erpen de Bona, C3SL serves public
2175 +       aufs2 GIT tree (2009/2).
2176 +William Grant made a donation (2009/3).
2177 +Patrick Lane made a donation (2009/4).
2178 +The Mail Archive (mail-archive.com) made donations (2009/5).
2179 +Nippy Networks (Ed Wildgoose) made a donation (2009/7).
2180 +New Dream Network, LLC (www.dreamhost.com) made a donation (2009/11).
2181 +Pavel Pronskiy made a donation (2011/2).
2182 +Iridium and Inmarsat satellite phone retailer (www.mailasail.com), Nippy
2183 +       Networks (Ed Wildgoose) made a donation for hardware (2011/3).
2184 +Max Lekomcev (DOM-TV project) made a donation (2011/7, 12, 2012/3, 6 and
2185 +11).
2186 +Sam Liddicott made a donation (2011/9).
2187 +Era Scarecrow made a donation (2013/4).
2188 +Bor Ratajc made a donation (2013/4).
2189 +Alessandro Gorreta made a donation (2013/4).
2190 +POIRETTE Marc made a donation (2013/4).
2191 +Alessandro Gorreta made a donation (2013/4).
2192 +lauri kasvandik made a donation (2013/5).
2193 +"pemasu from Finland" made a donation (2013/7).
2194 +The Parted Magic Project made a donation (2013/9 and 11).
2195 +Pavel Barta made a donation (2013/10).
2196 +
2197 +Thank you very much.
2198 +Donations are always, including future donations, very important and
2199 +helpful for me to keep on developing aufs.
2200 +
2201 +
2202 +7.
2203 +----------------------------------------
2204 +If you are an experienced user, no explanation is needed. Aufs is
2205 +just a linux filesystem.
2206 +
2207 +
2208 +Enjoy!
2209 +
2210 +# Local variables: ;
2211 +# mode: text;
2212 +# End: ;
2213 diff -urN /usr/share/empty/fs/aufs/aufs.h linux/fs/aufs/aufs.h
2214 --- /usr/share/empty/fs/aufs/aufs.h     1970-01-01 01:00:00.000000000 +0100
2215 +++ linux/fs/aufs/aufs.h        2014-04-24 22:11:10.848602379 +0200
2216 @@ -0,0 +1,59 @@
2217 +/*
2218 + * Copyright (C) 2005-2014 Junjiro R. Okajima
2219 + *
2220 + * This program, aufs is free software; you can redistribute it and/or modify
2221 + * it under the terms of the GNU General Public License as published by
2222 + * the Free Software Foundation; either version 2 of the License, or
2223 + * (at your option) any later version.
2224 + *
2225 + * This program is distributed in the hope that it will be useful,
2226 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2227 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2228 + * GNU General Public License for more details.
2229 + *
2230 + * You should have received a copy of the GNU General Public License
2231 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2232 + */
2233 +
2234 +/*
2235 + * all header files
2236 + */
2237 +
2238 +#ifndef __AUFS_H__
2239 +#define __AUFS_H__
2240 +
2241 +#ifdef __KERNEL__
2242 +
2243 +#define AuStub(type, name, body, ...) \
2244 +       static inline type name(__VA_ARGS__) { body; }
2245 +
2246 +#define AuStubVoid(name, ...) \
2247 +       AuStub(void, name, , __VA_ARGS__)
2248 +#define AuStubInt0(name, ...) \
2249 +       AuStub(int, name, return 0, __VA_ARGS__)
2250 +
2251 +#include "debug.h"
2252 +
2253 +#include "branch.h"
2254 +#include "cpup.h"
2255 +#include "dcsub.h"
2256 +#include "dbgaufs.h"
2257 +#include "dentry.h"
2258 +#include "dir.h"
2259 +#include "dynop.h"
2260 +#include "file.h"
2261 +#include "fstype.h"
2262 +#include "inode.h"
2263 +#include "loop.h"
2264 +#include "module.h"
2265 +#include "opts.h"
2266 +#include "rwsem.h"
2267 +#include "spl.h"
2268 +#include "super.h"
2269 +#include "sysaufs.h"
2270 +#include "vfsub.h"
2271 +#include "whout.h"
2272 +#include "wkq.h"
2273 +
2274 +#endif /* __KERNEL__ */
2275 +#endif /* __AUFS_H__ */
2276 diff -urN /usr/share/empty/fs/aufs/branch.c linux/fs/aufs/branch.c
2277 --- /usr/share/empty/fs/aufs/branch.c   1970-01-01 01:00:00.000000000 +0100
2278 +++ linux/fs/aufs/branch.c      2014-04-24 22:11:10.848602379 +0200
2279 @@ -0,0 +1,1219 @@
2280 +/*
2281 + * Copyright (C) 2005-2014 Junjiro R. Okajima
2282 + *
2283 + * This program, aufs is free software; you can redistribute it and/or modify
2284 + * it under the terms of the GNU General Public License as published by
2285 + * the Free Software Foundation; either version 2 of the License, or
2286 + * (at your option) any later version.
2287 + *
2288 + * This program is distributed in the hope that it will be useful,
2289 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
2290 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
2291 + * GNU General Public License for more details.
2292 + *
2293 + * You should have received a copy of the GNU General Public License
2294 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
2295 + */
2296 +
2297 +/*
2298 + * branch management
2299 + */
2300 +
2301 +#include <linux/compat.h>
2302 +#include <linux/statfs.h>
2303 +#include "aufs.h"
2304 +
2305 +/*
2306 + * free a single branch
2307 + */
2308 +
2309 +/* prohibit rmdir to the root of the branch */
2310 +/* todo: another new flag? */
2311 +static void au_br_dflags_force(struct au_branch *br)
2312 +{
2313 +       struct dentry *h_dentry;
2314 +
2315 +       h_dentry = au_br_dentry(br);
2316 +       spin_lock(&h_dentry->d_lock);
2317 +       br->br_dflags = h_dentry->d_flags & DCACHE_MOUNTED;
2318 +       h_dentry->d_flags |= DCACHE_MOUNTED;
2319 +       spin_unlock(&h_dentry->d_lock);
2320 +}
2321 +
2322 +/* restore its d_flags */
2323 +static void au_br_dflags_restore(struct au_branch *br)
2324 +{
2325 +       struct dentry *h_dentry;
2326 +
2327 +       if (br->br_dflags)
2328 +               return;
2329 +
2330 +       h_dentry = au_br_dentry(br);
2331 +       spin_lock(&h_dentry->d_lock);
2332 +       h_dentry->d_flags &= ~DCACHE_MOUNTED;
2333 +       spin_unlock(&h_dentry->d_lock);
2334 +}
2335 +
2336 +static void au_br_do_free(struct au_branch *br)
2337 +{
2338 +       int i;
2339 +       struct au_wbr *wbr;
2340 +       struct au_dykey **key;
2341 +
2342 +       au_hnotify_fin_br(br);
2343 +
2344 +       if (br->br_xino.xi_file)
2345 +               fput(br->br_xino.xi_file);
2346 +       mutex_destroy(&br->br_xino.xi_nondir_mtx);
2347 +
2348 +       AuDebugOn(atomic_read(&br->br_count));
2349 +
2350 +       wbr = br->br_wbr;
2351 +       if (wbr) {
2352 +               for (i = 0; i < AuBrWh_Last; i++)
2353 +                       dput(wbr->wbr_wh[i]);
2354 +               AuDebugOn(atomic_read(&wbr->wbr_wh_running));
2355 +               AuRwDestroy(&wbr->wbr_wh_rwsem);
2356 +       }
2357 +
2358 +       key = br->br_dykey;
2359 +       for (i = 0; i < AuBrDynOp; i++, key++)
2360 +               if (*key)
2361 +                       au_dy_put(*key);
2362 +               else
2363 +                       break;
2364 +
2365 +       au_br_dflags_restore(br);
2366 +
2367 +       /* recursive lock, s_umount of branch's */
2368 +       lockdep_off();
2369 +       path_put(&br->br_path);
2370 +       lockdep_on();
2371 +       kfree(wbr);
2372 +       kfree(br);
2373 +}
2374 +
2375 +/*
2376 + * frees all branches
2377 + */
2378 +void au_br_free(struct au_sbinfo *sbinfo)
2379 +{
2380 +       aufs_bindex_t bmax;
2381 +       struct au_branch **br;
2382 +
2383 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
2384 +
2385 +       bmax = sbinfo->si_bend + 1;
2386 +       br = sbinfo->si_branch;
2387 +       while (bmax--)
2388 +               au_br_do_free(*br++);
2389 +}
2390 +
2391 +/*
2392 + * find the index of a branch which is specified by @br_id.
2393 + */
2394 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id)
2395 +{
2396 +       aufs_bindex_t bindex, bend;
2397 +
2398 +       bend = au_sbend(sb);
2399 +       for (bindex = 0; bindex <= bend; bindex++)
2400 +               if (au_sbr_id(sb, bindex) == br_id)
2401 +                       return bindex;
2402 +       return -1;
2403 +}
2404 +
2405 +/* ---------------------------------------------------------------------- */
2406 +
2407 +/*
2408 + * add a branch
2409 + */
2410 +
2411 +static int test_overlap(struct super_block *sb, struct dentry *h_adding,
2412 +                       struct dentry *h_root)
2413 +{
2414 +       if (unlikely(h_adding == h_root
2415 +                    || au_test_loopback_overlap(sb, h_adding)))
2416 +               return 1;
2417 +       if (h_adding->d_sb != h_root->d_sb)
2418 +               return 0;
2419 +       return au_test_subdir(h_adding, h_root)
2420 +               || au_test_subdir(h_root, h_adding);
2421 +}
2422 +
2423 +/*
2424 + * returns a newly allocated branch. @new_nbranch is a number of branches
2425 + * after adding a branch.
2426 + */
2427 +static struct au_branch *au_br_alloc(struct super_block *sb, int new_nbranch,
2428 +                                    int perm)
2429 +{
2430 +       struct au_branch *add_branch;
2431 +       struct dentry *root;
2432 +       int err;
2433 +
2434 +       err = -ENOMEM;
2435 +       root = sb->s_root;
2436 +       add_branch = kmalloc(sizeof(*add_branch), GFP_NOFS);
2437 +       if (unlikely(!add_branch))
2438 +               goto out;
2439 +
2440 +       err = au_hnotify_init_br(add_branch, perm);
2441 +       if (unlikely(err))
2442 +               goto out_br;
2443 +
2444 +       add_branch->br_wbr = NULL;
2445 +       if (au_br_writable(perm)) {
2446 +               /* may be freed separately at changing the branch permission */
2447 +               add_branch->br_wbr = kmalloc(sizeof(*add_branch->br_wbr),
2448 +                                            GFP_NOFS);
2449 +               if (unlikely(!add_branch->br_wbr))
2450 +                       goto out_hnotify;
2451 +       }
2452 +
2453 +       err = au_sbr_realloc(au_sbi(sb), new_nbranch);
2454 +       if (!err)
2455 +               err = au_di_realloc(au_di(root), new_nbranch);
2456 +       if (!err)
2457 +               err = au_ii_realloc(au_ii(root->d_inode), new_nbranch);
2458 +       if (!err)
2459 +               return add_branch; /* success */
2460 +
2461 +       kfree(add_branch->br_wbr);
2462 +
2463 +out_hnotify:
2464 +       au_hnotify_fin_br(add_branch);
2465 +out_br:
2466 +       kfree(add_branch);
2467 +out:
2468 +       return ERR_PTR(err);
2469 +}
2470 +
2471 +/*
2472 + * test if the branch permission is legal or not.
2473 + */
2474 +static int test_br(struct inode *inode, int brperm, char *path)
2475 +{
2476 +       int err;
2477 +
2478 +       err = (au_br_writable(brperm) && IS_RDONLY(inode));
2479 +       if (!err)
2480 +               goto out;
2481 +
2482 +       err = -EINVAL;
2483 +       pr_err("write permission for readonly mount or inode, %s\n", path);
2484 +
2485 +out:
2486 +       return err;
2487 +}
2488 +
2489 +/*
2490 + * returns:
2491 + * 0: success, the caller will add it
2492 + * plus: success, it is already unified, the caller should ignore it
2493 + * minus: error
2494 + */
2495 +static int test_add(struct super_block *sb, struct au_opt_add *add, int remount)
2496 +{
2497 +       int err;
2498 +       aufs_bindex_t bend, bindex;
2499 +       struct dentry *root;
2500 +       struct inode *inode, *h_inode;
2501 +
2502 +       root = sb->s_root;
2503 +       bend = au_sbend(sb);
2504 +       if (unlikely(bend >= 0
2505 +                    && au_find_dbindex(root, add->path.dentry) >= 0)) {
2506 +               err = 1;
2507 +               if (!remount) {
2508 +                       err = -EINVAL;
2509 +                       pr_err("%s duplicated\n", add->pathname);
2510 +               }
2511 +               goto out;
2512 +       }
2513 +
2514 +       err = -ENOSPC; /* -E2BIG; */
2515 +       if (unlikely(AUFS_BRANCH_MAX <= add->bindex
2516 +                    || AUFS_BRANCH_MAX - 1 <= bend)) {
2517 +               pr_err("number of branches exceeded %s\n", add->pathname);
2518 +               goto out;
2519 +       }
2520 +
2521 +       err = -EDOM;
2522 +       if (unlikely(add->bindex < 0 || bend + 1 < add->bindex)) {
2523 +               pr_err("bad index %d\n", add->bindex);
2524 +               goto out;
2525 +       }
2526 +
2527 +       inode = add->path.dentry->d_inode;
2528 +       err = -ENOENT;
2529 +       if (unlikely(!inode->i_nlink)) {
2530 +               pr_err("no existence %s\n", add->pathname);
2531 +               goto out;
2532 +       }
2533 +
2534 +       err = -EINVAL;
2535 +       if (unlikely(inode->i_sb == sb)) {
2536 +               pr_err("%s must be outside\n", add->pathname);
2537 +               goto out;
2538 +       }
2539 +
2540 +       if (unlikely(au_test_fs_unsuppoted(inode->i_sb))) {
2541 +               pr_err("unsupported filesystem, %s (%s)\n",
2542 +                      add->pathname, au_sbtype(inode->i_sb));
2543 +               goto out;
2544 +       }
2545 +
2546 +       err = test_br(add->path.dentry->d_inode, add->perm, add->pathname);
2547 +       if (unlikely(err))
2548 +               goto out;
2549 +
2550 +       if (bend < 0)
2551 +               return 0; /* success */
2552 +
2553 +       err = -EINVAL;
2554 +       for (bindex = 0; bindex <= bend; bindex++)
2555 +               if (unlikely(test_overlap(sb, add->path.dentry,
2556 +                                         au_h_dptr(root, bindex)))) {
2557 +                       pr_err("%s is overlapped\n", add->pathname);
2558 +                       goto out;
2559 +               }
2560 +
2561 +       err = 0;
2562 +       if (au_opt_test(au_mntflags(sb), WARN_PERM)) {
2563 +               h_inode = au_h_dptr(root, 0)->d_inode;
2564 +               if ((h_inode->i_mode & S_IALLUGO) != (inode->i_mode & S_IALLUGO)
2565 +                   || !uid_eq(h_inode->i_uid, inode->i_uid)
2566 +                   || !gid_eq(h_inode->i_gid, inode->i_gid))
2567 +                       pr_warn("uid/gid/perm %s %u/%u/0%o, %u/%u/0%o\n",
2568 +                               add->pathname,
2569 +                               i_uid_read(inode), i_gid_read(inode),
2570 +                               (inode->i_mode & S_IALLUGO),
2571 +                               i_uid_read(h_inode), i_gid_read(h_inode),
2572 +                               (h_inode->i_mode & S_IALLUGO));
2573 +       }
2574 +
2575 +out:
2576 +       return err;
2577 +}
2578 +
2579 +/*
2580 + * initialize or clean the whiteouts for an adding branch
2581 + */
2582 +static int au_br_init_wh(struct super_block *sb, struct au_branch *br,
2583 +                        int new_perm)
2584 +{
2585 +       int err, old_perm;
2586 +       aufs_bindex_t bindex;
2587 +       struct mutex *h_mtx;
2588 +       struct au_wbr *wbr;
2589 +       struct au_hinode *hdir;
2590 +
2591 +       err = vfsub_mnt_want_write(au_br_mnt(br));
2592 +       if (unlikely(err))
2593 +               goto out;
2594 +
2595 +       wbr = br->br_wbr;
2596 +       old_perm = br->br_perm;
2597 +       br->br_perm = new_perm;
2598 +       hdir = NULL;
2599 +       h_mtx = NULL;
2600 +       bindex = au_br_index(sb, br->br_id);
2601 +       if (0 <= bindex) {
2602 +               hdir = au_hi(sb->s_root->d_inode, bindex);
2603 +               au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
2604 +       } else {
2605 +               h_mtx = &au_br_dentry(br)->d_inode->i_mutex;
2606 +               mutex_lock_nested(h_mtx, AuLsc_I_PARENT);
2607 +       }
2608 +       if (!wbr)
2609 +               err = au_wh_init(br, sb);
2610 +       else {
2611 +               wbr_wh_write_lock(wbr);
2612 +               err = au_wh_init(br, sb);
2613 +               wbr_wh_write_unlock(wbr);
2614 +       }
2615 +       if (hdir)
2616 +               au_hn_imtx_unlock(hdir);
2617 +       else
2618 +               mutex_unlock(h_mtx);
2619 +       vfsub_mnt_drop_write(au_br_mnt(br));
2620 +       br->br_perm = old_perm;
2621 +
2622 +       if (!err && wbr && !au_br_writable(new_perm)) {
2623 +               kfree(wbr);
2624 +               br->br_wbr = NULL;
2625 +       }
2626 +
2627 +out:
2628 +       return err;
2629 +}
2630 +
2631 +static int au_wbr_init(struct au_branch *br, struct super_block *sb,
2632 +                      int perm)
2633 +{
2634 +       int err;
2635 +       struct kstatfs kst;
2636 +       struct au_wbr *wbr;
2637 +
2638 +       wbr = br->br_wbr;
2639 +       au_rw_init(&wbr->wbr_wh_rwsem);
2640 +       memset(wbr->wbr_wh, 0, sizeof(wbr->wbr_wh));
2641 +       atomic_set(&wbr->wbr_wh_running, 0);
2642 +       wbr->wbr_bytes = 0;
2643 +
2644 +       /*
2645 +        * a limit for rmdir/rename a dir
2646 +        * cf. AUFS_MAX_NAMELEN in include/uapi/linux/aufs_type.h
2647 +        */
2648 +       err = vfs_statfs(&br->br_path, &kst);
2649 +       if (unlikely(err))
2650 +               goto out;
2651 +       err = -EINVAL;
2652 +       if (kst.f_namelen >= NAME_MAX)
2653 +               err = au_br_init_wh(sb, br, perm);
2654 +       else
2655 +               pr_err("%pd(%s), unsupported namelen %ld\n",
2656 +                      au_br_dentry(br),
2657 +                      au_sbtype(au_br_dentry(br)->d_sb), kst.f_namelen);
2658 +
2659 +out:
2660 +       return err;
2661 +}
2662 +
2663 +/* intialize a new branch */
2664 +static int au_br_init(struct au_branch *br, struct super_block *sb,
2665 +                     struct au_opt_add *add)
2666 +{
2667 +       int err;
2668 +
2669 +       err = 0;
2670 +       memset(&br->br_xino, 0, sizeof(br->br_xino));
2671 +       mutex_init(&br->br_xino.xi_nondir_mtx);
2672 +       br->br_perm = add->perm;
2673 +       BUILD_BUG_ON(sizeof(br->br_dflags)
2674 +                    != sizeof(br->br_path.dentry->d_flags));
2675 +       br->br_dflags = DCACHE_MOUNTED;
2676 +       br->br_path = add->path; /* set first, path_get() later */
2677 +       spin_lock_init(&br->br_dykey_lock);
2678 +       memset(br->br_dykey, 0, sizeof(br->br_dykey));
2679 +       atomic_set(&br->br_count, 0);
2680 +       atomic_set(&br->br_xino_running, 0);
2681 +       br->br_id = au_new_br_id(sb);
2682 +       AuDebugOn(br->br_id < 0);
2683 +
2684 +       if (au_br_writable(add->perm)) {
2685 +               err = au_wbr_init(br, sb, add->perm);
2686 +               if (unlikely(err))
2687 +                       goto out_err;
2688 +       }
2689 +
2690 +       if (au_opt_test(au_mntflags(sb), XINO)) {
2691 +               err = au_xino_br(sb, br, add->path.dentry->d_inode->i_ino,
2692 +                                au_sbr(sb, 0)->br_xino.xi_file, /*do_test*/1);
2693 +               if (unlikely(err)) {
2694 +                       AuDebugOn(br->br_xino.xi_file);
2695 +                       goto out_err;
2696 +               }
2697 +       }
2698 +
2699 +       sysaufs_br_init(br);
2700 +       path_get(&br->br_path);
2701 +       goto out; /* success */
2702 +
2703 +out_err:
2704 +       memset(&br->br_path, 0, sizeof(br->br_path));
2705 +out:
2706 +       return err;
2707 +}
2708 +
2709 +static void au_br_do_add_brp(struct au_sbinfo *sbinfo, aufs_bindex_t bindex,
2710 +                            struct au_branch *br, aufs_bindex_t bend,
2711 +                            aufs_bindex_t amount)
2712 +{
2713 +       struct au_branch **brp;
2714 +
2715 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
2716 +
2717 +       brp = sbinfo->si_branch + bindex;
2718 +       memmove(brp + 1, brp, sizeof(*brp) * amount);
2719 +       *brp = br;
2720 +       sbinfo->si_bend++;
2721 +       if (unlikely(bend < 0))
2722 +               sbinfo->si_bend = 0;
2723 +}
2724 +
2725 +static void au_br_do_add_hdp(struct au_dinfo *dinfo, aufs_bindex_t bindex,
2726 +                            aufs_bindex_t bend, aufs_bindex_t amount)
2727 +{
2728 +       struct au_hdentry *hdp;
2729 +
2730 +       AuRwMustWriteLock(&dinfo->di_rwsem);
2731 +
2732 +       hdp = dinfo->di_hdentry + bindex;
2733 +       memmove(hdp + 1, hdp, sizeof(*hdp) * amount);
2734 +       au_h_dentry_init(hdp);
2735 +       dinfo->di_bend++;
2736 +       if (unlikely(bend < 0))
2737 +               dinfo->di_bstart = 0;
2738 +}
2739 +
2740 +static void au_br_do_add_hip(struct au_iinfo *iinfo, aufs_bindex_t bindex,
2741 +                            aufs_bindex_t bend, aufs_bindex_t amount)
2742 +{
2743 +       struct au_hinode *hip;
2744 +
2745 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
2746 +
2747 +       hip = iinfo->ii_hinode + bindex;
2748 +       memmove(hip + 1, hip, sizeof(*hip) * amount);
2749 +       hip->hi_inode = NULL;
2750 +       au_hn_init(hip);
2751 +       iinfo->ii_bend++;
2752 +       if (unlikely(bend < 0))
2753 +               iinfo->ii_bstart = 0;
2754 +}
2755 +
2756 +static void au_br_do_add(struct super_block *sb, struct au_branch *br,
2757 +                        aufs_bindex_t bindex)
2758 +{
2759 +       struct dentry *root, *h_dentry;
2760 +       struct inode *root_inode;
2761 +       aufs_bindex_t bend, amount;
2762 +
2763 +       au_br_dflags_force(br);
2764 +
2765 +       root = sb->s_root;
2766 +       root_inode = root->d_inode;
2767 +       bend = au_sbend(sb);
2768 +       amount = bend + 1 - bindex;
2769 +       h_dentry = au_br_dentry(br);
2770 +       au_sbilist_lock();
2771 +       au_br_do_add_brp(au_sbi(sb), bindex, br, bend, amount);
2772 +       au_br_do_add_hdp(au_di(root), bindex, bend, amount);
2773 +       au_br_do_add_hip(au_ii(root_inode), bindex, bend, amount);
2774 +       au_set_h_dptr(root, bindex, dget(h_dentry));
2775 +       au_set_h_iptr(root_inode, bindex, au_igrab(h_dentry->d_inode),
2776 +                     /*flags*/0);
2777 +       au_sbilist_unlock();
2778 +}
2779 +
2780 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount)
2781 +{
2782 +       int err;
2783 +       aufs_bindex_t bend, add_bindex;
2784 +       struct dentry *root, *h_dentry;
2785 +       struct inode *root_inode;
2786 +       struct au_branch *add_branch;
2787 +
2788 +       root = sb->s_root;
2789 +       root_inode = root->d_inode;
2790 +       IMustLock(root_inode);
2791 +       err = test_add(sb, add, remount);
2792 +       if (unlikely(err < 0))
2793 +               goto out;
2794 +       if (err) {
2795 +               err = 0;
2796 +               goto out; /* success */
2797 +       }
2798 +
2799 +       bend = au_sbend(sb);
2800 +       add_branch = au_br_alloc(sb, bend + 2, add->perm);
2801 +       err = PTR_ERR(add_branch);
2802 +       if (IS_ERR(add_branch))
2803 +               goto out;
2804 +
2805 +       err = au_br_init(add_branch, sb, add);
2806 +       if (unlikely(err)) {
2807 +               au_br_do_free(add_branch);
2808 +               goto out;
2809 +       }
2810 +
2811 +       add_bindex = add->bindex;
2812 +       if (!remount)
2813 +               au_br_do_add(sb, add_branch, add_bindex);
2814 +       else {
2815 +               sysaufs_brs_del(sb, add_bindex);
2816 +               au_br_do_add(sb, add_branch, add_bindex);
2817 +               sysaufs_brs_add(sb, add_bindex);
2818 +       }
2819 +
2820 +       h_dentry = add->path.dentry;
2821 +       if (!add_bindex) {
2822 +               au_cpup_attr_all(root_inode, /*force*/1);
2823 +               sb->s_maxbytes = h_dentry->d_sb->s_maxbytes;
2824 +       } else
2825 +               au_add_nlink(root_inode, h_dentry->d_inode);
2826 +
2827 +       /*
2828 +        * this test/set prevents aufs from handling unnecesary notify events
2829 +        * of xino files, in case of re-adding a writable branch which was
2830 +        * once detached from aufs.
2831 +        */
2832 +       if (au_xino_brid(sb) < 0
2833 +           && au_br_writable(add_branch->br_perm)
2834 +           && !au_test_fs_bad_xino(h_dentry->d_sb)
2835 +           && add_branch->br_xino.xi_file
2836 +           && add_branch->br_xino.xi_file->f_dentry->d_parent == h_dentry)
2837 +               au_xino_brid_set(sb, add_branch->br_id);
2838 +
2839 +out:
2840 +       return err;
2841 +}
2842 +
2843 +/* ---------------------------------------------------------------------- */
2844 +
2845 +/*
2846 + * delete a branch
2847 + */
2848 +
2849 +/* to show the line number, do not make it inlined function */
2850 +#define AuVerbose(do_info, fmt, ...) do { \
2851 +       if (do_info) \
2852 +               pr_info(fmt, ##__VA_ARGS__); \
2853 +} while (0)
2854 +
2855 +static int au_test_ibusy(struct inode *inode, aufs_bindex_t bstart,
2856 +                        aufs_bindex_t bend)
2857 +{
2858 +       return (inode && !S_ISDIR(inode->i_mode)) || bstart == bend;
2859 +}
2860 +
2861 +static int au_test_dbusy(struct dentry *dentry, aufs_bindex_t bstart,
2862 +                        aufs_bindex_t bend)
2863 +{
2864 +       return au_test_ibusy(dentry->d_inode, bstart, bend);
2865 +}
2866 +
2867 +/*
2868 + * test if the branch is deletable or not.
2869 + */
2870 +static int test_dentry_busy(struct dentry *root, aufs_bindex_t bindex,
2871 +                           unsigned int sigen, const unsigned int verbose)
2872 +{
2873 +       int err, i, j, ndentry;
2874 +       aufs_bindex_t bstart, bend;
2875 +       struct au_dcsub_pages dpages;
2876 +       struct au_dpage *dpage;
2877 +       struct dentry *d;
2878 +
2879 +       err = au_dpages_init(&dpages, GFP_NOFS);
2880 +       if (unlikely(err))
2881 +               goto out;
2882 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
2883 +       if (unlikely(err))
2884 +               goto out_dpages;
2885 +
2886 +       for (i = 0; !err && i < dpages.ndpage; i++) {
2887 +               dpage = dpages.dpages + i;
2888 +               ndentry = dpage->ndentry;
2889 +               for (j = 0; !err && j < ndentry; j++) {
2890 +                       d = dpage->dentries[j];
2891 +                       AuDebugOn(!d_count(d));
2892 +                       if (!au_digen_test(d, sigen)) {
2893 +                               di_read_lock_child(d, AuLock_IR);
2894 +                               if (unlikely(au_dbrange_test(d))) {
2895 +                                       di_read_unlock(d, AuLock_IR);
2896 +                                       continue;
2897 +                               }
2898 +                       } else {
2899 +                               di_write_lock_child(d);
2900 +                               if (unlikely(au_dbrange_test(d))) {
2901 +                                       di_write_unlock(d);
2902 +                                       continue;
2903 +                               }
2904 +                               err = au_reval_dpath(d, sigen);
2905 +                               if (!err)
2906 +                                       di_downgrade_lock(d, AuLock_IR);
2907 +                               else {
2908 +                                       di_write_unlock(d);
2909 +                                       break;
2910 +                               }
2911 +                       }
2912 +
2913 +                       /* AuDbgDentry(d); */
2914 +                       bstart = au_dbstart(d);
2915 +                       bend = au_dbend(d);
2916 +                       if (bstart <= bindex
2917 +                           && bindex <= bend
2918 +                           && au_h_dptr(d, bindex)
2919 +                           && au_test_dbusy(d, bstart, bend)) {
2920 +                               err = -EBUSY;
2921 +                               AuVerbose(verbose, "busy %pd\n", d);
2922 +                               AuDbgDentry(d);
2923 +                       }
2924 +                       di_read_unlock(d, AuLock_IR);
2925 +               }
2926 +       }
2927 +
2928 +out_dpages:
2929 +       au_dpages_free(&dpages);
2930 +out:
2931 +       return err;
2932 +}
2933 +
2934 +static int test_inode_busy(struct super_block *sb, aufs_bindex_t bindex,
2935 +                          unsigned int sigen, const unsigned int verbose)
2936 +{
2937 +       int err;
2938 +       unsigned long long max, ull;
2939 +       struct inode *i, **array;
2940 +       aufs_bindex_t bstart, bend;
2941 +
2942 +       array = au_iarray_alloc(sb, &max);
2943 +       err = PTR_ERR(array);
2944 +       if (IS_ERR(array))
2945 +               goto out;
2946 +
2947 +       err = 0;
2948 +       AuDbg("b%d\n", bindex);
2949 +       for (ull = 0; !err && ull < max; ull++) {
2950 +               i = array[ull];
2951 +               if (i->i_ino == AUFS_ROOT_INO)
2952 +                       continue;
2953 +
2954 +               /* AuDbgInode(i); */
2955 +               if (au_iigen(i, NULL) == sigen)
2956 +                       ii_read_lock_child(i);
2957 +               else {
2958 +                       ii_write_lock_child(i);
2959 +                       err = au_refresh_hinode_self(i);
2960 +                       au_iigen_dec(i);
2961 +                       if (!err)
2962 +                               ii_downgrade_lock(i);
2963 +                       else {
2964 +                               ii_write_unlock(i);
2965 +                               break;
2966 +                       }
2967 +               }
2968 +
2969 +               bstart = au_ibstart(i);
2970 +               bend = au_ibend(i);
2971 +               if (bstart <= bindex
2972 +                   && bindex <= bend
2973 +                   && au_h_iptr(i, bindex)
2974 +                   && au_test_ibusy(i, bstart, bend)) {
2975 +                       err = -EBUSY;
2976 +                       AuVerbose(verbose, "busy i%lu\n", i->i_ino);
2977 +                       AuDbgInode(i);
2978 +               }
2979 +               ii_read_unlock(i);
2980 +       }
2981 +       au_iarray_free(array, max);
2982 +
2983 +out:
2984 +       return err;
2985 +}
2986 +
2987 +static int test_children_busy(struct dentry *root, aufs_bindex_t bindex,
2988 +                             const unsigned int verbose)
2989 +{
2990 +       int err;
2991 +       unsigned int sigen;
2992 +
2993 +       sigen = au_sigen(root->d_sb);
2994 +       DiMustNoWaiters(root);
2995 +       IiMustNoWaiters(root->d_inode);
2996 +       di_write_unlock(root);
2997 +       err = test_dentry_busy(root, bindex, sigen, verbose);
2998 +       if (!err)
2999 +               err = test_inode_busy(root->d_sb, bindex, sigen, verbose);
3000 +       di_write_lock_child(root); /* aufs_write_lock() calls ..._child() */
3001 +
3002 +       return err;
3003 +}
3004 +
3005 +static void au_br_do_del_brp(struct au_sbinfo *sbinfo,
3006 +                            const aufs_bindex_t bindex,
3007 +                            const aufs_bindex_t bend)
3008 +{
3009 +       struct au_branch **brp, **p;
3010 +
3011 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
3012 +
3013 +       brp = sbinfo->si_branch + bindex;
3014 +       if (bindex < bend)
3015 +               memmove(brp, brp + 1, sizeof(*brp) * (bend - bindex));
3016 +       sbinfo->si_branch[0 + bend] = NULL;
3017 +       sbinfo->si_bend--;
3018 +
3019 +       p = krealloc(sbinfo->si_branch, sizeof(*p) * bend, AuGFP_SBILIST);
3020 +       if (p)
3021 +               sbinfo->si_branch = p;
3022 +       /* harmless error */
3023 +}
3024 +
3025 +static void au_br_do_del_hdp(struct au_dinfo *dinfo, const aufs_bindex_t bindex,
3026 +                            const aufs_bindex_t bend)
3027 +{
3028 +       struct au_hdentry *hdp, *p;
3029 +
3030 +       AuRwMustWriteLock(&dinfo->di_rwsem);
3031 +
3032 +       hdp = dinfo->di_hdentry;
3033 +       if (bindex < bend)
3034 +               memmove(hdp + bindex, hdp + bindex + 1,
3035 +                       sizeof(*hdp) * (bend - bindex));
3036 +       hdp[0 + bend].hd_dentry = NULL;
3037 +       dinfo->di_bend--;
3038 +
3039 +       p = krealloc(hdp, sizeof(*p) * bend, AuGFP_SBILIST);
3040 +       if (p)
3041 +               dinfo->di_hdentry = p;
3042 +       /* harmless error */
3043 +}
3044 +
3045 +static void au_br_do_del_hip(struct au_iinfo *iinfo, const aufs_bindex_t bindex,
3046 +                            const aufs_bindex_t bend)
3047 +{
3048 +       struct au_hinode *hip, *p;
3049 +
3050 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
3051 +
3052 +       hip = iinfo->ii_hinode + bindex;
3053 +       if (bindex < bend)
3054 +               memmove(hip, hip + 1, sizeof(*hip) * (bend - bindex));
3055 +       iinfo->ii_hinode[0 + bend].hi_inode = NULL;
3056 +       au_hn_init(iinfo->ii_hinode + bend);
3057 +       iinfo->ii_bend--;
3058 +
3059 +       p = krealloc(iinfo->ii_hinode, sizeof(*p) * bend, AuGFP_SBILIST);
3060 +       if (p)
3061 +               iinfo->ii_hinode = p;
3062 +       /* harmless error */
3063 +}
3064 +
3065 +static void au_br_do_del(struct super_block *sb, aufs_bindex_t bindex,
3066 +                        struct au_branch *br)
3067 +{
3068 +       aufs_bindex_t bend;
3069 +       struct au_sbinfo *sbinfo;
3070 +       struct dentry *root, *h_root;
3071 +       struct inode *inode, *h_inode;
3072 +       struct au_hinode *hinode;
3073 +
3074 +       SiMustWriteLock(sb);
3075 +
3076 +       root = sb->s_root;
3077 +       inode = root->d_inode;
3078 +       sbinfo = au_sbi(sb);
3079 +       bend = sbinfo->si_bend;
3080 +
3081 +       h_root = au_h_dptr(root, bindex);
3082 +       hinode = au_hi(inode, bindex);
3083 +       h_inode = au_igrab(hinode->hi_inode);
3084 +       au_hiput(hinode);
3085 +
3086 +       au_sbilist_lock();
3087 +       au_br_do_del_brp(sbinfo, bindex, bend);
3088 +       au_br_do_del_hdp(au_di(root), bindex, bend);
3089 +       au_br_do_del_hip(au_ii(inode), bindex, bend);
3090 +       au_sbilist_unlock();
3091 +
3092 +       dput(h_root);
3093 +       iput(h_inode);
3094 +       au_br_do_free(br);
3095 +}
3096 +
3097 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount)
3098 +{
3099 +       int err, rerr, i;
3100 +       unsigned int mnt_flags;
3101 +       aufs_bindex_t bindex, bend, br_id;
3102 +       unsigned char do_wh, verbose;
3103 +       struct au_branch *br;
3104 +       struct au_wbr *wbr;
3105 +
3106 +       err = 0;
3107 +       bindex = au_find_dbindex(sb->s_root, del->h_path.dentry);
3108 +       if (bindex < 0) {
3109 +               if (remount)
3110 +                       goto out; /* success */
3111 +               err = -ENOENT;
3112 +               pr_err("%s no such branch\n", del->pathname);
3113 +               goto out;
3114 +       }
3115 +       AuDbg("bindex b%d\n", bindex);
3116 +
3117 +       err = -EBUSY;
3118 +       mnt_flags = au_mntflags(sb);
3119 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3120 +       bend = au_sbend(sb);
3121 +       if (unlikely(!bend)) {
3122 +               AuVerbose(verbose, "no more branches left\n");
3123 +               goto out;
3124 +       }
3125 +       br = au_sbr(sb, bindex);
3126 +       AuDebugOn(!path_equal(&br->br_path, &del->h_path));
3127 +       i = atomic_read(&br->br_count);
3128 +       if (unlikely(i)) {
3129 +               AuVerbose(verbose, "%d file(s) opened\n", i);
3130 +               goto out;
3131 +       }
3132 +
3133 +       wbr = br->br_wbr;
3134 +       do_wh = wbr && (wbr->wbr_whbase || wbr->wbr_plink || wbr->wbr_orph);
3135 +       if (do_wh) {
3136 +               /* instead of WbrWhMustWriteLock(wbr) */
3137 +               SiMustWriteLock(sb);
3138 +               for (i = 0; i < AuBrWh_Last; i++) {
3139 +                       dput(wbr->wbr_wh[i]);
3140 +                       wbr->wbr_wh[i] = NULL;
3141 +               }
3142 +       }
3143 +
3144 +       err = test_children_busy(sb->s_root, bindex, verbose);
3145 +       if (unlikely(err)) {
3146 +               if (do_wh)
3147 +                       goto out_wh;
3148 +               goto out;
3149 +       }
3150 +
3151 +       err = 0;
3152 +       br_id = br->br_id;
3153 +       if (!remount)
3154 +               au_br_do_del(sb, bindex, br);
3155 +       else {
3156 +               sysaufs_brs_del(sb, bindex);
3157 +               au_br_do_del(sb, bindex, br);
3158 +               sysaufs_brs_add(sb, bindex);
3159 +       }
3160 +
3161 +       if (!bindex) {
3162 +               au_cpup_attr_all(sb->s_root->d_inode, /*force*/1);
3163 +               sb->s_maxbytes = au_sbr_sb(sb, 0)->s_maxbytes;
3164 +       } else
3165 +               au_sub_nlink(sb->s_root->d_inode, del->h_path.dentry->d_inode);
3166 +       if (au_opt_test(mnt_flags, PLINK))
3167 +               au_plink_half_refresh(sb, br_id);
3168 +
3169 +       if (au_xino_brid(sb) == br_id)
3170 +               au_xino_brid_set(sb, -1);
3171 +       goto out; /* success */
3172 +
3173 +out_wh:
3174 +       /* revert */
3175 +       rerr = au_br_init_wh(sb, br, br->br_perm);
3176 +       if (rerr)
3177 +               pr_warn("failed re-creating base whiteout, %s. (%d)\n",
3178 +                       del->pathname, rerr);
3179 +out:
3180 +       return err;
3181 +}
3182 +
3183 +/* ---------------------------------------------------------------------- */
3184 +
3185 +static int au_ibusy(struct super_block *sb, struct aufs_ibusy __user *arg)
3186 +{
3187 +       int err;
3188 +       aufs_bindex_t bstart, bend;
3189 +       struct aufs_ibusy ibusy;
3190 +       struct inode *inode, *h_inode;
3191 +
3192 +       err = -EPERM;
3193 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
3194 +               goto out;
3195 +
3196 +       err = copy_from_user(&ibusy, arg, sizeof(ibusy));
3197 +       if (!err)
3198 +               err = !access_ok(VERIFY_WRITE, &arg->h_ino, sizeof(arg->h_ino));
3199 +       if (unlikely(err)) {
3200 +               err = -EFAULT;
3201 +               AuTraceErr(err);
3202 +               goto out;
3203 +       }
3204 +
3205 +       err = -EINVAL;
3206 +       si_read_lock(sb, AuLock_FLUSH);
3207 +       if (unlikely(ibusy.bindex < 0 || ibusy.bindex > au_sbend(sb)))
3208 +               goto out_unlock;
3209 +
3210 +       err = 0;
3211 +       ibusy.h_ino = 0; /* invalid */
3212 +       inode = ilookup(sb, ibusy.ino);
3213 +       if (!inode
3214 +           || inode->i_ino == AUFS_ROOT_INO
3215 +           || is_bad_inode(inode))
3216 +               goto out_unlock;
3217 +
3218 +       ii_read_lock_child(inode);
3219 +       bstart = au_ibstart(inode);
3220 +       bend = au_ibend(inode);
3221 +       if (bstart <= ibusy.bindex && ibusy.bindex <= bend) {
3222 +               h_inode = au_h_iptr(inode, ibusy.bindex);
3223 +               if (h_inode && au_test_ibusy(inode, bstart, bend))
3224 +                       ibusy.h_ino = h_inode->i_ino;
3225 +       }
3226 +       ii_read_unlock(inode);
3227 +       iput(inode);
3228 +
3229 +out_unlock:
3230 +       si_read_unlock(sb);
3231 +       if (!err) {
3232 +               err = __put_user(ibusy.h_ino, &arg->h_ino);
3233 +               if (unlikely(err)) {
3234 +                       err = -EFAULT;
3235 +                       AuTraceErr(err);
3236 +               }
3237 +       }
3238 +out:
3239 +       return err;
3240 +}
3241 +
3242 +long au_ibusy_ioctl(struct file *file, unsigned long arg)
3243 +{
3244 +       return au_ibusy(file->f_dentry->d_sb, (void __user *)arg);
3245 +}
3246 +
3247 +#ifdef CONFIG_COMPAT
3248 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg)
3249 +{
3250 +       return au_ibusy(file->f_dentry->d_sb, compat_ptr(arg));
3251 +}
3252 +#endif
3253 +
3254 +/* ---------------------------------------------------------------------- */
3255 +
3256 +/*
3257 + * change a branch permission
3258 + */
3259 +
3260 +static void au_warn_ima(void)
3261 +{
3262 +#ifdef CONFIG_IMA
3263 +       /* since it doesn't support mark_files_ro() */
3264 +       AuWarn1("RW -> RO makes IMA to produce wrong message\n");
3265 +#endif
3266 +}
3267 +
3268 +static int do_need_sigen_inc(int a, int b)
3269 +{
3270 +       return au_br_whable(a) && !au_br_whable(b);
3271 +}
3272 +
3273 +static int need_sigen_inc(int old, int new)
3274 +{
3275 +       return do_need_sigen_inc(old, new)
3276 +               || do_need_sigen_inc(new, old);
3277 +}
3278 +
3279 +static unsigned long long au_farray_cb(void *a,
3280 +                                      unsigned long long max __maybe_unused,
3281 +                                      void *arg)
3282 +{
3283 +       unsigned long long n;
3284 +       struct file **p, *f;
3285 +       struct au_sphlhead *files;
3286 +       struct au_finfo *finfo;
3287 +       struct super_block *sb = arg;
3288 +
3289 +       n = 0;
3290 +       p = a;
3291 +       files = &au_sbi(sb)->si_files;
3292 +       spin_lock(&files->spin);
3293 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
3294 +               f = finfo->fi_file;
3295 +               if (file_count(f)
3296 +                   && !special_file(file_inode(f)->i_mode)) {
3297 +                       get_file(f);
3298 +                       *p++ = f;
3299 +                       n++;
3300 +                       AuDebugOn(n > max);
3301 +               }
3302 +       }
3303 +       spin_unlock(&files->spin);
3304 +
3305 +       return n;
3306 +}
3307 +
3308 +static struct file **au_farray_alloc(struct super_block *sb,
3309 +                                    unsigned long long *max)
3310 +{
3311 +       *max = atomic_long_read(&au_sbi(sb)->si_nfiles);
3312 +       return au_array_alloc(max, au_farray_cb, sb);
3313 +}
3314 +
3315 +static void au_farray_free(struct file **a, unsigned long long max)
3316 +{
3317 +       unsigned long long ull;
3318 +
3319 +       for (ull = 0; ull < max; ull++)
3320 +               if (a[ull])
3321 +                       fput(a[ull]);
3322 +       au_array_free(a);
3323 +}
3324 +
3325 +static int au_br_mod_files_ro(struct super_block *sb, aufs_bindex_t bindex)
3326 +{
3327 +       int err, do_warn;
3328 +       unsigned int mnt_flags;
3329 +       unsigned long long ull, max;
3330 +       aufs_bindex_t br_id;
3331 +       unsigned char verbose;
3332 +       struct file *file, *hf, **array;
3333 +       struct inode *inode;
3334 +       struct au_hfile *hfile;
3335 +
3336 +       mnt_flags = au_mntflags(sb);
3337 +       verbose = !!au_opt_test(mnt_flags, VERBOSE);
3338 +
3339 +       array = au_farray_alloc(sb, &max);
3340 +       err = PTR_ERR(array);
3341 +       if (IS_ERR(array))
3342 +               goto out;
3343 +
3344 +       do_warn = 0;
3345 +       br_id = au_sbr_id(sb, bindex);
3346 +       for (ull = 0; ull < max; ull++) {
3347 +               file = array[ull];
3348 +
3349 +               /* AuDbg("%pD\n", file); */
3350 +               fi_read_lock(file);
3351 +               if (unlikely(au_test_mmapped(file))) {
3352 +                       err = -EBUSY;
3353 +                       AuVerbose(verbose, "mmapped %pD\n", file);
3354 +                       AuDbgFile(file);
3355 +                       FiMustNoWaiters(file);
3356 +                       fi_read_unlock(file);
3357 +                       goto out_array;
3358 +               }
3359 +
3360 +               inode = file_inode(file);
3361 +               hfile = &au_fi(file)->fi_htop;
3362 +               hf = hfile->hf_file;
3363 +               if (!S_ISREG(inode->i_mode)
3364 +                   || !(file->f_mode & FMODE_WRITE)
3365 +                   || hfile->hf_br->br_id != br_id
3366 +                   || !(hf->f_mode & FMODE_WRITE))
3367 +                       array[ull] = NULL;
3368 +               else {
3369 +                       do_warn = 1;
3370 +                       get_file(file);
3371 +               }
3372 +
3373 +               FiMustNoWaiters(file);
3374 +               fi_read_unlock(file);
3375 +               fput(file);
3376 +       }
3377 +
3378 +       err = 0;
3379 +       if (do_warn)
3380 +               au_warn_ima();
3381 +
3382 +       for (ull = 0; ull < max; ull++) {
3383 +               file = array[ull];
3384 +               if (!file)
3385 +                       continue;
3386 +
3387 +               /* todo: already flushed? */
3388 +               /*
3389 +                * fs/super.c:mark_files_ro() is gone, but aufs keeps its
3390 +                * approach which resets f_mode and calls mnt_drop_write() and
3391 +                * file_release_write() for each file, because the branch
3392 +                * attribute in aufs world is totally different from the native
3393 +                * fs rw/ro mode.
3394 +               */
3395 +               /* fi_read_lock(file); */
3396 +               hfile = &au_fi(file)->fi_htop;
3397 +               hf = hfile->hf_file;
3398 +               /* fi_read_unlock(file); */
3399 +               spin_lock(&hf->f_lock);
3400 +               hf->f_mode &= ~FMODE_WRITE;
3401 +               spin_unlock(&hf->f_lock);
3402 +               if (!file_check_writeable(hf)) {
3403 +                       __mnt_drop_write(hf->f_path.mnt);
3404 +                       file_release_write(hf);
3405 +               }
3406 +       }
3407 +
3408 +out_array:
3409 +       au_farray_free(array, max);
3410 +out:
3411 +       AuTraceErr(err);
3412 +       return err;
3413 +}
3414 +
3415 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
3416 +             int *do_refresh)
3417 +{
3418 +       int err, rerr;
3419 +       aufs_bindex_t bindex;
3420 +       struct dentry *root;
3421 +       struct au_branch *br;
3422 +
3423 +       root = sb->s_root;
3424 +       bindex = au_find_dbindex(root, mod->h_root);
3425 +       if (bindex < 0) {
3426 +               if (remount)
3427 +                       return 0; /* success */
3428 +               err = -ENOENT;
3429 +               pr_err("%s no such branch\n", mod->path);
3430 +               goto out;
3431 +       }
3432 +       AuDbg("bindex b%d\n", bindex);
3433 +
3434 +       err = test_br(mod->h_root->d_inode, mod->perm, mod->path);
3435 +       if (unlikely(err))
3436 +               goto out;
3437 +
3438 +       br = au_sbr(sb, bindex);
3439 +       AuDebugOn(mod->h_root != au_br_dentry(br));
3440 +       if (br->br_perm == mod->perm)
3441 +               return 0; /* success */
3442 +
3443 +       if (au_br_writable(br->br_perm)) {
3444 +               /* remove whiteout base */
3445 +               err = au_br_init_wh(sb, br, mod->perm);
3446 +               if (unlikely(err))
3447 +                       goto out;
3448 +
3449 +               if (!au_br_writable(mod->perm)) {
3450 +                       /* rw --> ro, file might be mmapped */
3451 +                       DiMustNoWaiters(root);
3452 +                       IiMustNoWaiters(root->d_inode);
3453 +                       di_write_unlock(root);
3454 +                       err = au_br_mod_files_ro(sb, bindex);
3455 +                       /* aufs_write_lock() calls ..._child() */
3456 +                       di_write_lock_child(root);
3457 +
3458 +                       if (unlikely(err)) {
3459 +                               rerr = -ENOMEM;
3460 +                               br->br_wbr = kmalloc(sizeof(*br->br_wbr),
3461 +                                                    GFP_NOFS);
3462 +                               if (br->br_wbr)
3463 +                                       rerr = au_wbr_init(br, sb, br->br_perm);
3464 +                               if (unlikely(rerr)) {
3465 +                                       AuIOErr("nested error %d (%d)\n",
3466 +                                               rerr, err);
3467 +                                       br->br_perm = mod->perm;
3468 +                               }
3469 +                       }
3470 +               }
3471 +       } else if (au_br_writable(mod->perm)) {
3472 +               /* ro --> rw */
3473 +               err = -ENOMEM;
3474 +               br->br_wbr = kmalloc(sizeof(*br->br_wbr), GFP_NOFS);
3475 +               if (br->br_wbr) {
3476 +                       err = au_wbr_init(br, sb, mod->perm);
3477 +                       if (unlikely(err)) {
3478 +                               kfree(br->br_wbr);
3479 +                               br->br_wbr = NULL;
3480 +                       }
3481 +               }
3482 +       }
3483 +
3484 +       if (!err) {
3485 +               if ((br->br_perm & AuBrAttr_UNPIN)
3486 +                   && !(mod->perm & AuBrAttr_UNPIN))
3487 +                       au_br_dflags_force(br);
3488 +               else if (!(br->br_perm & AuBrAttr_UNPIN)
3489 +                        && (mod->perm & AuBrAttr_UNPIN))
3490 +                       au_br_dflags_restore(br);
3491 +               *do_refresh |= need_sigen_inc(br->br_perm, mod->perm);
3492 +               br->br_perm = mod->perm;
3493 +       }
3494 +
3495 +out:
3496 +       AuTraceErr(err);
3497 +       return err;
3498 +}
3499 diff -urN /usr/share/empty/fs/aufs/branch.h linux/fs/aufs/branch.h
3500 --- /usr/share/empty/fs/aufs/branch.h   1970-01-01 01:00:00.000000000 +0100
3501 +++ linux/fs/aufs/branch.h      2014-04-24 22:11:10.848602379 +0200
3502 @@ -0,0 +1,264 @@
3503 +/*
3504 + * Copyright (C) 2005-2014 Junjiro R. Okajima
3505 + *
3506 + * This program, aufs is free software; you can redistribute it and/or modify
3507 + * it under the terms of the GNU General Public License as published by
3508 + * the Free Software Foundation; either version 2 of the License, or
3509 + * (at your option) any later version.
3510 + *
3511 + * This program is distributed in the hope that it will be useful,
3512 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
3513 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
3514 + * GNU General Public License for more details.
3515 + *
3516 + * You should have received a copy of the GNU General Public License
3517 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
3518 + */
3519 +
3520 +/*
3521 + * branch filesystems and xino for them
3522 + */
3523 +
3524 +#ifndef __AUFS_BRANCH_H__
3525 +#define __AUFS_BRANCH_H__
3526 +
3527 +#ifdef __KERNEL__
3528 +
3529 +#include <linux/mount.h>
3530 +#include "dynop.h"
3531 +#include "rwsem.h"
3532 +#include "super.h"
3533 +
3534 +/* ---------------------------------------------------------------------- */
3535 +
3536 +/* a xino file */
3537 +struct au_xino_file {
3538 +       struct file             *xi_file;
3539 +       struct mutex            xi_nondir_mtx;
3540 +
3541 +       /* todo: make xino files an array to support huge inode number */
3542 +
3543 +#ifdef CONFIG_DEBUG_FS
3544 +       struct dentry            *xi_dbgaufs;
3545 +#endif
3546 +};
3547 +
3548 +/* members for writable branch only */
3549 +enum {AuBrWh_BASE, AuBrWh_PLINK, AuBrWh_ORPH, AuBrWh_Last};
3550 +struct au_wbr {
3551 +       struct au_rwsem         wbr_wh_rwsem;
3552 +       struct dentry           *wbr_wh[AuBrWh_Last];
3553 +       atomic_t                wbr_wh_running;
3554 +#define wbr_whbase             wbr_wh[AuBrWh_BASE]     /* whiteout base */
3555 +#define wbr_plink              wbr_wh[AuBrWh_PLINK]    /* pseudo-link dir */
3556 +#define wbr_orph               wbr_wh[AuBrWh_ORPH]     /* dir for orphans */
3557 +
3558 +       /* mfs mode */
3559 +       unsigned long long      wbr_bytes;
3560 +};
3561 +
3562 +/* ext2 has 3 types of operations at least, ext3 has 4 */
3563 +#define AuBrDynOp (AuDyLast * 4)
3564 +
3565 +#ifdef CONFIG_AUFS_HFSNOTIFY
3566 +/* support for asynchronous destruction */
3567 +struct au_br_hfsnotify {
3568 +       struct fsnotify_group   *hfsn_group;
3569 +};
3570 +#endif
3571 +
3572 +/* sysfs entries */
3573 +struct au_brsysfs {
3574 +       char                    name[16];
3575 +       struct attribute        attr;
3576 +};
3577 +
3578 +enum {
3579 +       AuBrSysfs_BR,
3580 +       AuBrSysfs_BRID,
3581 +       AuBrSysfs_Last
3582 +};
3583 +
3584 +/* protected by superblock rwsem */
3585 +struct au_branch {
3586 +       struct au_xino_file     br_xino;
3587 +
3588 +       aufs_bindex_t           br_id;
3589 +
3590 +       int                     br_perm;
3591 +       unsigned int            br_dflags;
3592 +       struct path             br_path;
3593 +       spinlock_t              br_dykey_lock;
3594 +       struct au_dykey         *br_dykey[AuBrDynOp];
3595 +       atomic_t                br_count;
3596 +
3597 +       struct au_wbr           *br_wbr;
3598 +
3599 +       /* xino truncation */
3600 +       atomic_t                br_xino_running;
3601 +
3602 +#ifdef CONFIG_AUFS_HFSNOTIFY
3603 +       struct au_br_hfsnotify  *br_hfsn;
3604 +#endif
3605 +
3606 +#ifdef CONFIG_SYSFS
3607 +       /* entries under sysfs per mount-point */
3608 +       struct au_brsysfs       br_sysfs[AuBrSysfs_Last];
3609 +#endif
3610 +};
3611 +
3612 +/* ---------------------------------------------------------------------- */
3613 +
3614 +static inline struct vfsmount *au_br_mnt(struct au_branch *br)
3615 +{
3616 +       return br->br_path.mnt;
3617 +}
3618 +
3619 +static inline struct dentry *au_br_dentry(struct au_branch *br)
3620 +{
3621 +       return br->br_path.dentry;
3622 +}
3623 +
3624 +static inline struct super_block *au_br_sb(struct au_branch *br)
3625 +{
3626 +       return au_br_mnt(br)->mnt_sb;
3627 +}
3628 +
3629 +/* branch permissions and attributes */
3630 +#define AuBrPerm_RW            1               /* writable, hardlinkable wh */
3631 +#define AuBrPerm_RO            (1 << 1)        /* readonly */
3632 +#define AuBrPerm_RR            (1 << 2)        /* natively readonly */
3633 +#define AuBrPerm_Mask          (AuBrPerm_RW | AuBrPerm_RO | AuBrPerm_RR)
3634 +
3635 +#define AuBrRAttr_WH           (1 << 3)        /* whiteout-able */
3636 +
3637 +#define AuBrWAttr_NoLinkWH     (1 << 4)        /* un-hardlinkable whiteouts */
3638 +
3639 +#define AuBrAttr_UNPIN         (1 << 5)        /* rename-able top dir of
3640 +                                                  branch */
3641 +
3642 +static inline int au_br_writable(int brperm)
3643 +{
3644 +       return brperm & AuBrPerm_RW;
3645 +}
3646 +
3647 +static inline int au_br_whable(int brperm)
3648 +{
3649 +       return brperm & (AuBrPerm_RW | AuBrRAttr_WH);
3650 +}
3651 +
3652 +static inline int au_br_wh_linkable(int brperm)
3653 +{
3654 +       return !(brperm & AuBrWAttr_NoLinkWH);
3655 +}
3656 +
3657 +static inline int au_br_rdonly(struct au_branch *br)
3658 +{
3659 +       return ((au_br_sb(br)->s_flags & MS_RDONLY)
3660 +               || !au_br_writable(br->br_perm))
3661 +               ? -EROFS : 0;
3662 +}
3663 +
3664 +static inline int au_br_hnotifyable(int brperm __maybe_unused)
3665 +{
3666 +#ifdef CONFIG_AUFS_HNOTIFY
3667 +       return !(brperm & AuBrPerm_RR);
3668 +#else
3669 +       return 0;
3670 +#endif
3671 +}
3672 +
3673 +/* ---------------------------------------------------------------------- */
3674 +
3675 +/* branch.c */
3676 +struct au_sbinfo;
3677 +void au_br_free(struct au_sbinfo *sinfo);
3678 +int au_br_index(struct super_block *sb, aufs_bindex_t br_id);
3679 +struct au_opt_add;
3680 +int au_br_add(struct super_block *sb, struct au_opt_add *add, int remount);
3681 +struct au_opt_del;
3682 +int au_br_del(struct super_block *sb, struct au_opt_del *del, int remount);
3683 +long au_ibusy_ioctl(struct file *file, unsigned long arg);
3684 +#ifdef CONFIG_COMPAT
3685 +long au_ibusy_compat_ioctl(struct file *file, unsigned long arg);
3686 +#endif
3687 +struct au_opt_mod;
3688 +int au_br_mod(struct super_block *sb, struct au_opt_mod *mod, int remount,
3689 +             int *do_refresh);
3690 +
3691 +/* xino.c */
3692 +static const loff_t au_loff_max = LLONG_MAX;
3693 +
3694 +int au_xib_trunc(struct super_block *sb);
3695 +ssize_t xino_fread(au_readf_t func, struct file *file, void *buf, size_t size,
3696 +                  loff_t *pos);
3697 +ssize_t xino_fwrite(au_writef_t func, struct file *file, void *buf, size_t size,
3698 +                   loff_t *pos);
3699 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src);
3700 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent);
3701 +ino_t au_xino_new_ino(struct super_block *sb);
3702 +void au_xino_delete_inode(struct inode *inode, const int unlinked);
3703 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
3704 +                 ino_t ino);
3705 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
3706 +                ino_t *ino);
3707 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t hino,
3708 +              struct file *base_file, int do_test);
3709 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex);
3710 +
3711 +struct au_opt_xino;
3712 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount);
3713 +void au_xino_clr(struct super_block *sb);
3714 +struct file *au_xino_def(struct super_block *sb);
3715 +int au_xino_path(struct seq_file *seq, struct file *file);
3716 +
3717 +/* ---------------------------------------------------------------------- */
3718 +
3719 +/* Superblock to branch */
3720 +static inline
3721 +aufs_bindex_t au_sbr_id(struct super_block *sb, aufs_bindex_t bindex)
3722 +{
3723 +       return au_sbr(sb, bindex)->br_id;
3724 +}
3725 +
3726 +static inline
3727 +struct vfsmount *au_sbr_mnt(struct super_block *sb, aufs_bindex_t bindex)
3728 +{
3729 +       return au_br_mnt(au_sbr(sb, bindex));
3730 +}
3731 +
3732 +static inline
3733 +struct super_block *au_sbr_sb(struct super_block *sb, aufs_bindex_t bindex)
3734 +{
3735 +       return au_br_sb(au_sbr(sb, bindex));
3736 +}
3737 +
3738 +static inline void au_sbr_put(struct super_block *sb, aufs_bindex_t bindex)
3739 +{
3740 +       atomic_dec(&au_sbr(sb, bindex)->br_count);
3741 +}
3742 +
3743 +static inline int au_sbr_perm(struct super_block *sb, aufs_bindex_t bindex)
3744 +{
3745 +       return au_sbr(sb, bindex)->br_perm;
3746 +}
3747 +
3748 +static inline int au_sbr_whable(struct super_block *sb, aufs_bindex_t bindex)
3749 +{
3750 +       return au_br_whable(au_sbr_perm(sb, bindex));
3751 +}
3752 +
3753 +/* ---------------------------------------------------------------------- */
3754 +
3755 +/*
3756 + * wbr_wh_read_lock, wbr_wh_write_lock
3757 + * wbr_wh_read_unlock, wbr_wh_write_unlock, wbr_wh_downgrade_lock
3758 + */
3759 +AuSimpleRwsemFuncs(wbr_wh, struct au_wbr *wbr, &wbr->wbr_wh_rwsem);
3760 +
3761 +#define WbrWhMustNoWaiters(wbr)        AuRwMustNoWaiters(&wbr->wbr_wh_rwsem)
3762 +#define WbrWhMustAnyLock(wbr)  AuRwMustAnyLock(&wbr->wbr_wh_rwsem)
3763 +#define WbrWhMustWriteLock(wbr)        AuRwMustWriteLock(&wbr->wbr_wh_rwsem)
3764 +
3765 +#endif /* __KERNEL__ */
3766 +#endif /* __AUFS_BRANCH_H__ */
3767 diff -urN /usr/share/empty/fs/aufs/conf.mk linux/fs/aufs/conf.mk
3768 --- /usr/share/empty/fs/aufs/conf.mk    1970-01-01 01:00:00.000000000 +0100
3769 +++ linux/fs/aufs/conf.mk       2014-04-24 22:11:10.848602379 +0200
3770 @@ -0,0 +1,37 @@
3771 +
3772 +AuConfStr = CONFIG_AUFS_FS=${CONFIG_AUFS_FS}
3773 +
3774 +define AuConf
3775 +ifdef ${1}
3776 +AuConfStr += ${1}=${${1}}
3777 +endif
3778 +endef
3779 +
3780 +AuConfAll = BRANCH_MAX_127 BRANCH_MAX_511 BRANCH_MAX_1023 BRANCH_MAX_32767 \
3781 +       SBILIST \
3782 +       HNOTIFY HFSNOTIFY \
3783 +       EXPORT INO_T_64 \
3784 +       RDU \
3785 +       SP_IATTR \
3786 +       SHWH \
3787 +       BR_RAMFS \
3788 +       BR_FUSE POLL \
3789 +       BR_HFSPLUS \
3790 +       BDEV_LOOP \
3791 +       DEBUG MAGIC_SYSRQ
3792 +$(foreach i, ${AuConfAll}, \
3793 +       $(eval $(call AuConf,CONFIG_AUFS_${i})))
3794 +
3795 +AuConfName = ${obj}/conf.str
3796 +${AuConfName}.tmp: FORCE
3797 +       @echo ${AuConfStr} | tr ' ' '\n' | sed -e 's/^/"/' -e 's/$$/\\n"/' > $@
3798 +${AuConfName}: ${AuConfName}.tmp
3799 +       @diff -q $< $@ > /dev/null 2>&1 || { \
3800 +       echo '  GEN    ' $@; \
3801 +       cp -p $< $@; \
3802 +       }
3803 +FORCE:
3804 +clean-files += ${AuConfName} ${AuConfName}.tmp
3805 +${obj}/sysfs.o: ${AuConfName}
3806 +
3807 +-include ${srctree}/${src}/conf_priv.mk
3808 diff -urN /usr/share/empty/fs/aufs/cpup.c linux/fs/aufs/cpup.c
3809 --- /usr/share/empty/fs/aufs/cpup.c     1970-01-01 01:00:00.000000000 +0100
3810 +++ linux/fs/aufs/cpup.c        2014-04-24 22:11:10.848602379 +0200
3811 @@ -0,0 +1,1277 @@
3812 +/*
3813 + * Copyright (C) 2005-2014 Junjiro R. Okajima
3814 + *
3815 + * This program, aufs is free software; you can redistribute it and/or modify
3816 + * it under the terms of the GNU General Public License as published by
3817 + * the Free Software Foundation; either version 2 of the License, or
3818 + * (at your option) any later version.
3819 + *
3820 + * This program is distributed in the hope that it will be useful,
3821 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
3822 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
3823 + * GNU General Public License for more details.
3824 + *
3825 + * You should have received a copy of the GNU General Public License
3826 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
3827 + */
3828 +
3829 +/*
3830 + * copy-up functions, see wbr_policy.c for copy-down
3831 + */
3832 +
3833 +#include <linux/fs_stack.h>
3834 +#include <linux/mm.h>
3835 +#include "aufs.h"
3836 +
3837 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags)
3838 +{
3839 +       const unsigned int mask = S_DEAD | S_SWAPFILE | S_PRIVATE
3840 +               | S_NOATIME | S_NOCMTIME | S_AUTOMOUNT;
3841 +
3842 +       BUILD_BUG_ON(sizeof(iflags) != sizeof(dst->i_flags));
3843 +
3844 +       dst->i_flags |= iflags & ~mask;
3845 +       if (au_test_fs_notime(dst->i_sb))
3846 +               dst->i_flags |= S_NOATIME | S_NOCMTIME;
3847 +}
3848 +
3849 +void au_cpup_attr_timesizes(struct inode *inode)
3850 +{
3851 +       struct inode *h_inode;
3852 +
3853 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
3854 +       fsstack_copy_attr_times(inode, h_inode);
3855 +       fsstack_copy_inode_size(inode, h_inode);
3856 +}
3857 +
3858 +void au_cpup_attr_nlink(struct inode *inode, int force)
3859 +{
3860 +       struct inode *h_inode;
3861 +       struct super_block *sb;
3862 +       aufs_bindex_t bindex, bend;
3863 +
3864 +       sb = inode->i_sb;
3865 +       bindex = au_ibstart(inode);
3866 +       h_inode = au_h_iptr(inode, bindex);
3867 +       if (!force
3868 +           && !S_ISDIR(h_inode->i_mode)
3869 +           && au_opt_test(au_mntflags(sb), PLINK)
3870 +           && au_plink_test(inode))
3871 +               return;
3872 +
3873 +       /*
3874 +        * 0 can happen in revalidating.
3875 +        * h_inode->i_mutex is not held, but it is harmless since once i_nlink
3876 +        * reaches 0, it will never become positive.
3877 +        */
3878 +       set_nlink(inode, h_inode->i_nlink);
3879 +
3880 +       /*
3881 +        * fewer nlink makes find(1) noisy, but larger nlink doesn't.
3882 +        * it may includes whplink directory.
3883 +        */
3884 +       if (S_ISDIR(h_inode->i_mode)) {
3885 +               bend = au_ibend(inode);
3886 +               for (bindex++; bindex <= bend; bindex++) {
3887 +                       h_inode = au_h_iptr(inode, bindex);
3888 +                       if (h_inode)
3889 +                               au_add_nlink(inode, h_inode);
3890 +               }
3891 +       }
3892 +}
3893 +
3894 +void au_cpup_attr_changeable(struct inode *inode)
3895 +{
3896 +       struct inode *h_inode;
3897 +
3898 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
3899 +       inode->i_mode = h_inode->i_mode;
3900 +       inode->i_uid = h_inode->i_uid;
3901 +       inode->i_gid = h_inode->i_gid;
3902 +       au_cpup_attr_timesizes(inode);
3903 +       au_cpup_attr_flags(inode, h_inode->i_flags);
3904 +}
3905 +
3906 +void au_cpup_igen(struct inode *inode, struct inode *h_inode)
3907 +{
3908 +       struct au_iinfo *iinfo = au_ii(inode);
3909 +
3910 +       IiMustWriteLock(inode);
3911 +
3912 +       iinfo->ii_higen = h_inode->i_generation;
3913 +       iinfo->ii_hsb1 = h_inode->i_sb;
3914 +}
3915 +
3916 +void au_cpup_attr_all(struct inode *inode, int force)
3917 +{
3918 +       struct inode *h_inode;
3919 +
3920 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
3921 +       au_cpup_attr_changeable(inode);
3922 +       if (inode->i_nlink > 0)
3923 +               au_cpup_attr_nlink(inode, force);
3924 +       inode->i_rdev = h_inode->i_rdev;
3925 +       inode->i_blkbits = h_inode->i_blkbits;
3926 +       au_cpup_igen(inode, h_inode);
3927 +}
3928 +
3929 +/* ---------------------------------------------------------------------- */
3930 +
3931 +/* Note: dt_dentry and dt_h_dentry are not dget/dput-ed */
3932 +
3933 +/* keep the timestamps of the parent dir when cpup */
3934 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
3935 +                   struct path *h_path)
3936 +{
3937 +       struct inode *h_inode;
3938 +
3939 +       dt->dt_dentry = dentry;
3940 +       dt->dt_h_path = *h_path;
3941 +       h_inode = h_path->dentry->d_inode;
3942 +       dt->dt_atime = h_inode->i_atime;
3943 +       dt->dt_mtime = h_inode->i_mtime;
3944 +       /* smp_mb(); */
3945 +}
3946 +
3947 +void au_dtime_revert(struct au_dtime *dt)
3948 +{
3949 +       struct iattr attr;
3950 +       int err;
3951 +
3952 +       attr.ia_atime = dt->dt_atime;
3953 +       attr.ia_mtime = dt->dt_mtime;
3954 +       attr.ia_valid = ATTR_FORCE | ATTR_MTIME | ATTR_MTIME_SET
3955 +               | ATTR_ATIME | ATTR_ATIME_SET;
3956 +
3957 +       /* no delegation since this is a directory */
3958 +       err = vfsub_notify_change(&dt->dt_h_path, &attr, /*delegated*/NULL);
3959 +       if (unlikely(err))
3960 +               pr_warn("restoring timestamps failed(%d). ignored\n", err);
3961 +}
3962 +
3963 +/* ---------------------------------------------------------------------- */
3964 +
3965 +/* internal use only */
3966 +struct au_cpup_reg_attr {
3967 +       int             valid;
3968 +       struct kstat    st;
3969 +       unsigned int    iflags; /* inode->i_flags */
3970 +};
3971 +
3972 +static noinline_for_stack
3973 +int cpup_iattr(struct dentry *dst, aufs_bindex_t bindex, struct dentry *h_src,
3974 +              struct au_cpup_reg_attr *h_src_attr)
3975 +{
3976 +       int err, sbits;
3977 +       struct iattr ia;
3978 +       struct path h_path;
3979 +       struct inode *h_isrc, *h_idst;
3980 +       struct kstat *h_st;
3981 +
3982 +       h_path.dentry = au_h_dptr(dst, bindex);
3983 +       h_idst = h_path.dentry->d_inode;
3984 +       h_path.mnt = au_sbr_mnt(dst->d_sb, bindex);
3985 +       h_isrc = h_src->d_inode;
3986 +       ia.ia_valid = ATTR_FORCE | ATTR_UID | ATTR_GID
3987 +               | ATTR_ATIME | ATTR_MTIME
3988 +               | ATTR_ATIME_SET | ATTR_MTIME_SET;
3989 +       if (h_src_attr && h_src_attr->valid) {
3990 +               h_st = &h_src_attr->st;
3991 +               ia.ia_uid = h_st->uid;
3992 +               ia.ia_gid = h_st->gid;
3993 +               ia.ia_atime = h_st->atime;
3994 +               ia.ia_mtime = h_st->mtime;
3995 +               if (h_idst->i_mode != h_st->mode
3996 +                   && !S_ISLNK(h_idst->i_mode)) {
3997 +                       ia.ia_valid |= ATTR_MODE;
3998 +                       ia.ia_mode = h_st->mode;
3999 +               }
4000 +               sbits = !!(h_st->mode & (S_ISUID | S_ISGID));
4001 +               au_cpup_attr_flags(h_idst, h_src_attr->iflags);
4002 +       } else {
4003 +               ia.ia_uid = h_isrc->i_uid;
4004 +               ia.ia_gid = h_isrc->i_gid;
4005 +               ia.ia_atime = h_isrc->i_atime;
4006 +               ia.ia_mtime = h_isrc->i_mtime;
4007 +               if (h_idst->i_mode != h_isrc->i_mode
4008 +                   && !S_ISLNK(h_idst->i_mode)) {
4009 +                       ia.ia_valid |= ATTR_MODE;
4010 +                       ia.ia_mode = h_isrc->i_mode;
4011 +               }
4012 +               sbits = !!(h_isrc->i_mode & (S_ISUID | S_ISGID));
4013 +               au_cpup_attr_flags(h_idst, h_isrc->i_flags);
4014 +       }
4015 +       /* no delegation since it is just created */
4016 +       err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4017 +
4018 +       /* is this nfs only? */
4019 +       if (!err && sbits && au_test_nfs(h_path.dentry->d_sb)) {
4020 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
4021 +               ia.ia_mode = h_isrc->i_mode;
4022 +               err = vfsub_notify_change(&h_path, &ia, /*delegated*/NULL);
4023 +       }
4024 +
4025 +       return err;
4026 +}
4027 +
4028 +/* ---------------------------------------------------------------------- */
4029 +
4030 +static int au_do_copy_file(struct file *dst, struct file *src, loff_t len,
4031 +                          char *buf, unsigned long blksize)
4032 +{
4033 +       int err;
4034 +       size_t sz, rbytes, wbytes;
4035 +       unsigned char all_zero;
4036 +       char *p, *zp;
4037 +       struct mutex *h_mtx;
4038 +       /* reduce stack usage */
4039 +       struct iattr *ia;
4040 +
4041 +       zp = page_address(ZERO_PAGE(0));
4042 +       if (unlikely(!zp))
4043 +               return -ENOMEM; /* possible? */
4044 +
4045 +       err = 0;
4046 +       all_zero = 0;
4047 +       while (len) {
4048 +               AuDbg("len %lld\n", len);
4049 +               sz = blksize;
4050 +               if (len < blksize)
4051 +                       sz = len;
4052 +
4053 +               rbytes = 0;
4054 +               /* todo: signal_pending? */
4055 +               while (!rbytes || err == -EAGAIN || err == -EINTR) {
4056 +                       rbytes = vfsub_read_k(src, buf, sz, &src->f_pos);
4057 +                       err = rbytes;
4058 +               }
4059 +               if (unlikely(err < 0))
4060 +                       break;
4061 +
4062 +               all_zero = 0;
4063 +               if (len >= rbytes && rbytes == blksize)
4064 +                       all_zero = !memcmp(buf, zp, rbytes);
4065 +               if (!all_zero) {
4066 +                       wbytes = rbytes;
4067 +                       p = buf;
4068 +                       while (wbytes) {
4069 +                               size_t b;
4070 +
4071 +                               b = vfsub_write_k(dst, p, wbytes, &dst->f_pos);
4072 +                               err = b;
4073 +                               /* todo: signal_pending? */
4074 +                               if (unlikely(err == -EAGAIN || err == -EINTR))
4075 +                                       continue;
4076 +                               if (unlikely(err < 0))
4077 +                                       break;
4078 +                               wbytes -= b;
4079 +                               p += b;
4080 +                       }
4081 +                       if (unlikely(err < 0))
4082 +                               break;
4083 +               } else {
4084 +                       loff_t res;
4085 +
4086 +                       AuLabel(hole);
4087 +                       res = vfsub_llseek(dst, rbytes, SEEK_CUR);
4088 +                       err = res;
4089 +                       if (unlikely(res < 0))
4090 +                               break;
4091 +               }
4092 +               len -= rbytes;
4093 +               err = 0;
4094 +       }
4095 +
4096 +       /* the last block may be a hole */
4097 +       if (!err && all_zero) {
4098 +               AuLabel(last hole);
4099 +
4100 +               err = 1;
4101 +               if (au_test_nfs(dst->f_dentry->d_sb)) {
4102 +                       /* nfs requires this step to make last hole */
4103 +                       /* is this only nfs? */
4104 +                       do {
4105 +                               /* todo: signal_pending? */
4106 +                               err = vfsub_write_k(dst, "\0", 1, &dst->f_pos);
4107 +                       } while (err == -EAGAIN || err == -EINTR);
4108 +                       if (err == 1)
4109 +                               dst->f_pos--;
4110 +               }
4111 +
4112 +               if (err == 1) {
4113 +                       ia = (void *)buf;
4114 +                       ia->ia_size = dst->f_pos;
4115 +                       ia->ia_valid = ATTR_SIZE | ATTR_FILE;
4116 +                       ia->ia_file = dst;
4117 +                       h_mtx = &file_inode(dst)->i_mutex;
4118 +                       mutex_lock_nested(h_mtx, AuLsc_I_CHILD2);
4119 +                       /* no delegation since it is just created */
4120 +                       err = vfsub_notify_change(&dst->f_path, ia,
4121 +                                                 /*delegated*/NULL);
4122 +                       mutex_unlock(h_mtx);
4123 +               }
4124 +       }
4125 +
4126 +       return err;
4127 +}
4128 +
4129 +int au_copy_file(struct file *dst, struct file *src, loff_t len)
4130 +{
4131 +       int err;
4132 +       unsigned long blksize;
4133 +       unsigned char do_kfree;
4134 +       char *buf;
4135 +
4136 +       err = -ENOMEM;
4137 +       blksize = dst->f_dentry->d_sb->s_blocksize;
4138 +       if (!blksize || PAGE_SIZE < blksize)
4139 +               blksize = PAGE_SIZE;
4140 +       AuDbg("blksize %lu\n", blksize);
4141 +       do_kfree = (blksize != PAGE_SIZE && blksize >= sizeof(struct iattr *));
4142 +       if (do_kfree)
4143 +               buf = kmalloc(blksize, GFP_NOFS);
4144 +       else
4145 +               buf = (void *)__get_free_page(GFP_NOFS);
4146 +       if (unlikely(!buf))
4147 +               goto out;
4148 +
4149 +       if (len > (1 << 22))
4150 +               AuDbg("copying a large file %lld\n", (long long)len);
4151 +
4152 +       src->f_pos = 0;
4153 +       dst->f_pos = 0;
4154 +       err = au_do_copy_file(dst, src, len, buf, blksize);
4155 +       if (do_kfree)
4156 +               kfree(buf);
4157 +       else
4158 +               free_page((unsigned long)buf);
4159 +
4160 +out:
4161 +       return err;
4162 +}
4163 +
4164 +/*
4165 + * to support a sparse file which is opened with O_APPEND,
4166 + * we need to close the file.
4167 + */
4168 +static int au_cp_regular(struct au_cp_generic *cpg)
4169 +{
4170 +       int err, i;
4171 +       enum { SRC, DST };
4172 +       struct {
4173 +               aufs_bindex_t bindex;
4174 +               unsigned int flags;
4175 +               struct dentry *dentry;
4176 +               int force_wr;
4177 +               struct file *file;
4178 +               void *label;
4179 +       } *f, file[] = {
4180 +               {
4181 +                       .bindex = cpg->bsrc,
4182 +                       .flags = O_RDONLY | O_NOATIME | O_LARGEFILE,
4183 +                       .label = &&out
4184 +               },
4185 +               {
4186 +                       .bindex = cpg->bdst,
4187 +                       .flags = O_WRONLY | O_NOATIME | O_LARGEFILE,
4188 +                       .force_wr = !!au_ftest_cpup(cpg->flags, RWDST),
4189 +                       .label = &&out_src
4190 +               }
4191 +       };
4192 +       struct super_block *sb;
4193 +
4194 +       /* bsrc branch can be ro/rw. */
4195 +       sb = cpg->dentry->d_sb;
4196 +       f = file;
4197 +       for (i = 0; i < 2; i++, f++) {
4198 +               f->dentry = au_h_dptr(cpg->dentry, f->bindex);
4199 +               f->file = au_h_open(cpg->dentry, f->bindex, f->flags,
4200 +                                   /*file*/NULL, f->force_wr);
4201 +               err = PTR_ERR(f->file);
4202 +               if (IS_ERR(f->file))
4203 +                       goto *f->label;
4204 +       }
4205 +
4206 +       /* try stopping to update while we copyup */
4207 +       IMustLock(file[SRC].dentry->d_inode);
4208 +       err = au_copy_file(file[DST].file, file[SRC].file, cpg->len);
4209 +
4210 +       fput(file[DST].file);
4211 +       au_sbr_put(sb, file[DST].bindex);
4212 +
4213 +out_src:
4214 +       fput(file[SRC].file);
4215 +       au_sbr_put(sb, file[SRC].bindex);
4216 +out:
4217 +       return err;
4218 +}
4219 +
4220 +static int au_do_cpup_regular(struct au_cp_generic *cpg,
4221 +                             struct au_cpup_reg_attr *h_src_attr)
4222 +{
4223 +       int err, rerr;
4224 +       loff_t l;
4225 +       struct path h_path;
4226 +       struct inode *h_src_inode;
4227 +
4228 +       err = 0;
4229 +       h_src_inode = au_h_iptr(cpg->dentry->d_inode, cpg->bsrc);
4230 +       l = i_size_read(h_src_inode);
4231 +       if (cpg->len == -1 || l < cpg->len)
4232 +               cpg->len = l;
4233 +       if (cpg->len) {
4234 +               /* try stopping to update while we are referencing */
4235 +               mutex_lock_nested(&h_src_inode->i_mutex, AuLsc_I_CHILD);
4236 +               au_pin_hdir_unlock(cpg->pin);
4237 +
4238 +               h_path.dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
4239 +               h_path.mnt = au_sbr_mnt(cpg->dentry->d_sb, cpg->bsrc);
4240 +               h_src_attr->iflags = h_src_inode->i_flags;
4241 +               err = vfs_getattr(&h_path, &h_src_attr->st);
4242 +               if (unlikely(err)) {
4243 +                       mutex_unlock(&h_src_inode->i_mutex);
4244 +                       goto out;
4245 +               }
4246 +               h_src_attr->valid = 1;
4247 +               err = au_cp_regular(cpg);
4248 +               mutex_unlock(&h_src_inode->i_mutex);
4249 +               rerr = au_pin_hdir_relock(cpg->pin);
4250 +               if (!err && rerr)
4251 +                       err = rerr;
4252 +       }
4253 +
4254 +out:
4255 +       return err;
4256 +}
4257 +
4258 +static int au_do_cpup_symlink(struct path *h_path, struct dentry *h_src,
4259 +                             struct inode *h_dir)
4260 +{
4261 +       int err, symlen;
4262 +       mm_segment_t old_fs;
4263 +       union {
4264 +               char *k;
4265 +               char __user *u;
4266 +       } sym;
4267 +
4268 +       err = -ENOSYS;
4269 +       if (unlikely(!h_src->d_inode->i_op->readlink))
4270 +               goto out;
4271 +
4272 +       err = -ENOMEM;
4273 +       sym.k = (void *)__get_free_page(GFP_NOFS);
4274 +       if (unlikely(!sym.k))
4275 +               goto out;
4276 +
4277 +       /* unnecessary to support mmap_sem since symlink is not mmap-able */
4278 +       old_fs = get_fs();
4279 +       set_fs(KERNEL_DS);
4280 +       symlen = h_src->d_inode->i_op->readlink(h_src, sym.u, PATH_MAX);
4281 +       err = symlen;
4282 +       set_fs(old_fs);
4283 +
4284 +       if (symlen > 0) {
4285 +               sym.k[symlen] = 0;
4286 +               err = vfsub_symlink(h_dir, h_path, sym.k);
4287 +       }
4288 +       free_page((unsigned long)sym.k);
4289 +
4290 +out:
4291 +       return err;
4292 +}
4293 +
4294 +static noinline_for_stack
4295 +int cpup_entry(struct au_cp_generic *cpg, struct dentry *dst_parent,
4296 +              struct au_cpup_reg_attr *h_src_attr)
4297 +{
4298 +       int err;
4299 +       umode_t mode;
4300 +       unsigned int mnt_flags;
4301 +       unsigned char isdir;
4302 +       const unsigned char do_dt = !!au_ftest_cpup(cpg->flags, DTIME);
4303 +       struct au_dtime dt;
4304 +       struct path h_path;
4305 +       struct dentry *h_src, *h_dst, *h_parent;
4306 +       struct inode *h_inode, *h_dir;
4307 +       struct super_block *sb;
4308 +
4309 +       /* bsrc branch can be ro/rw. */
4310 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
4311 +       h_inode = h_src->d_inode;
4312 +       AuDebugOn(h_inode != au_h_iptr(cpg->dentry->d_inode, cpg->bsrc));
4313 +
4314 +       /* try stopping to be referenced while we are creating */
4315 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
4316 +       if (au_ftest_cpup(cpg->flags, RENAME))
4317 +               AuDebugOn(strncmp(h_dst->d_name.name, AUFS_WH_PFX,
4318 +                                 AUFS_WH_PFX_LEN));
4319 +       h_parent = h_dst->d_parent; /* dir inode is locked */
4320 +       h_dir = h_parent->d_inode;
4321 +       IMustLock(h_dir);
4322 +       AuDebugOn(h_parent != h_dst->d_parent);
4323 +
4324 +       sb = cpg->dentry->d_sb;
4325 +       h_path.mnt = au_sbr_mnt(sb, cpg->bdst);
4326 +       if (do_dt) {
4327 +               h_path.dentry = h_parent;
4328 +               au_dtime_store(&dt, dst_parent, &h_path);
4329 +       }
4330 +       h_path.dentry = h_dst;
4331 +
4332 +       isdir = 0;
4333 +       mode = h_inode->i_mode;
4334 +       switch (mode & S_IFMT) {
4335 +       case S_IFREG:
4336 +               err = vfsub_create(h_dir, &h_path, mode | S_IWUSR,
4337 +                                  /*want_excl*/true);
4338 +               if (!err)
4339 +                       err = au_do_cpup_regular(cpg, h_src_attr);
4340 +               break;
4341 +       case S_IFDIR:
4342 +               isdir = 1;
4343 +               err = vfsub_mkdir(h_dir, &h_path, mode);
4344 +               if (!err) {
4345 +                       /*
4346 +                        * strange behaviour from the users view,
4347 +                        * particularry setattr case
4348 +                        */
4349 +                       if (au_ibstart(dst_parent->d_inode) == cpg->bdst)
4350 +                               au_cpup_attr_nlink(dst_parent->d_inode,
4351 +                                                  /*force*/1);
4352 +                       au_cpup_attr_nlink(cpg->dentry->d_inode, /*force*/1);
4353 +               }
4354 +               break;
4355 +       case S_IFLNK:
4356 +               err = au_do_cpup_symlink(&h_path, h_src, h_dir);
4357 +               break;
4358 +       case S_IFCHR:
4359 +       case S_IFBLK:
4360 +               AuDebugOn(!capable(CAP_MKNOD));
4361 +               /*FALLTHROUGH*/
4362 +       case S_IFIFO:
4363 +       case S_IFSOCK:
4364 +               err = vfsub_mknod(h_dir, &h_path, mode, h_inode->i_rdev);
4365 +               break;
4366 +       default:
4367 +               AuIOErr("Unknown inode type 0%o\n", mode);
4368 +               err = -EIO;
4369 +       }
4370 +
4371 +       mnt_flags = au_mntflags(sb);
4372 +       if (!au_opt_test(mnt_flags, UDBA_NONE)
4373 +           && !isdir
4374 +           && au_opt_test(mnt_flags, XINO)
4375 +           && h_inode->i_nlink == 1
4376 +           /* todo: unnecessary? */
4377 +           /* && cpg->dentry->d_inode->i_nlink == 1 */
4378 +           && cpg->bdst < cpg->bsrc
4379 +           && !au_ftest_cpup(cpg->flags, KEEPLINO))
4380 +               au_xino_write(sb, cpg->bsrc, h_inode->i_ino, /*ino*/0);
4381 +               /* ignore this error */
4382 +
4383 +       if (do_dt)
4384 +               au_dtime_revert(&dt);
4385 +       return err;
4386 +}
4387 +
4388 +static int au_do_ren_after_cpup(struct au_cp_generic *cpg, struct path *h_path)
4389 +{
4390 +       int err;
4391 +       struct dentry *dentry, *h_dentry, *h_parent, *parent;
4392 +       struct inode *h_dir;
4393 +       aufs_bindex_t bdst;
4394 +
4395 +       dentry = cpg->dentry;
4396 +       bdst = cpg->bdst;
4397 +       h_dentry = au_h_dptr(dentry, bdst);
4398 +       if (!au_ftest_cpup(cpg->flags, OVERWRITE)) {
4399 +               dget(h_dentry);
4400 +               au_set_h_dptr(dentry, bdst, NULL);
4401 +               err = au_lkup_neg(dentry, bdst, /*wh*/0);
4402 +               if (!err)
4403 +                       h_path->dentry = dget(au_h_dptr(dentry, bdst));
4404 +               au_set_h_dptr(dentry, bdst, h_dentry);
4405 +       } else {
4406 +               err = 0;
4407 +               parent = dget_parent(dentry);
4408 +               h_parent = au_h_dptr(parent, bdst);
4409 +               dput(parent);
4410 +               h_path->dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
4411 +               if (IS_ERR(h_path->dentry))
4412 +                       err = PTR_ERR(h_path->dentry);
4413 +       }
4414 +       if (unlikely(err))
4415 +               goto out;
4416 +
4417 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
4418 +       h_dir = h_parent->d_inode;
4419 +       IMustLock(h_dir);
4420 +       AuDbg("%pd %pd\n", h_dentry, h_path->dentry);
4421 +       /* no delegation since it is just created */
4422 +       err = vfsub_rename(h_dir, h_dentry, h_dir, h_path, /*delegated*/NULL);
4423 +       dput(h_path->dentry);
4424 +
4425 +out:
4426 +       return err;
4427 +}
4428 +
4429 +/*
4430 + * copyup the @dentry from @bsrc to @bdst.
4431 + * the caller must set the both of lower dentries.
4432 + * @len is for truncating when it is -1 copyup the entire file.
4433 + * in link/rename cases, @dst_parent may be different from the real one.
4434 + * basic->bsrc can be larger than basic->bdst.
4435 + */
4436 +static int au_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
4437 +{
4438 +       int err, rerr;
4439 +       aufs_bindex_t old_ibstart;
4440 +       unsigned char isdir, plink;
4441 +       struct dentry *h_src, *h_dst, *h_parent;
4442 +       struct inode *dst_inode, *h_dir, *inode, *delegated;
4443 +       struct super_block *sb;
4444 +       struct au_branch *br;
4445 +       /* to reuduce stack size */
4446 +       struct {
4447 +               struct au_dtime dt;
4448 +               struct path h_path;
4449 +               struct au_cpup_reg_attr h_src_attr;
4450 +       } *a;
4451 +
4452 +       err = -ENOMEM;
4453 +       a = kmalloc(sizeof(*a), GFP_NOFS);
4454 +       if (unlikely(!a))
4455 +               goto out;
4456 +       a->h_src_attr.valid = 0;
4457 +
4458 +       sb = cpg->dentry->d_sb;
4459 +       br = au_sbr(sb, cpg->bdst);
4460 +       a->h_path.mnt = au_br_mnt(br);
4461 +       h_dst = au_h_dptr(cpg->dentry, cpg->bdst);
4462 +       h_parent = h_dst->d_parent; /* dir inode is locked */
4463 +       h_dir = h_parent->d_inode;
4464 +       IMustLock(h_dir);
4465 +
4466 +       h_src = au_h_dptr(cpg->dentry, cpg->bsrc);
4467 +       inode = cpg->dentry->d_inode;
4468 +
4469 +       if (!dst_parent)
4470 +               dst_parent = dget_parent(cpg->dentry);
4471 +       else
4472 +               dget(dst_parent);
4473 +
4474 +       plink = !!au_opt_test(au_mntflags(sb), PLINK);
4475 +       dst_inode = au_h_iptr(inode, cpg->bdst);
4476 +       if (dst_inode) {
4477 +               if (unlikely(!plink)) {
4478 +                       err = -EIO;
4479 +                       AuIOErr("hi%lu(i%lu) exists on b%d "
4480 +                               "but plink is disabled\n",
4481 +                               dst_inode->i_ino, inode->i_ino, cpg->bdst);
4482 +                       goto out_parent;
4483 +               }
4484 +
4485 +               if (dst_inode->i_nlink) {
4486 +                       const int do_dt = au_ftest_cpup(cpg->flags, DTIME);
4487 +
4488 +                       h_src = au_plink_lkup(inode, cpg->bdst);
4489 +                       err = PTR_ERR(h_src);
4490 +                       if (IS_ERR(h_src))
4491 +                               goto out_parent;
4492 +                       if (unlikely(!h_src->d_inode)) {
4493 +                               err = -EIO;
4494 +                               AuIOErr("i%lu exists on a upper branch "
4495 +                                       "but not pseudo-linked\n",
4496 +                                       inode->i_ino);
4497 +                               dput(h_src);
4498 +                               goto out_parent;
4499 +                       }
4500 +
4501 +                       if (do_dt) {
4502 +                               a->h_path.dentry = h_parent;
4503 +                               au_dtime_store(&a->dt, dst_parent, &a->h_path);
4504 +                       }
4505 +
4506 +                       a->h_path.dentry = h_dst;
4507 +                       delegated = NULL;
4508 +                       err = vfsub_link(h_src, h_dir, &a->h_path, &delegated);
4509 +                       if (!err && au_ftest_cpup(cpg->flags, RENAME))
4510 +                               err = au_do_ren_after_cpup(cpg, &a->h_path);
4511 +                       if (do_dt)
4512 +                               au_dtime_revert(&a->dt);
4513 +                       if (unlikely(err == -EWOULDBLOCK)) {
4514 +                               pr_warn("cannot retry for NFSv4 delegation"
4515 +                                       " for an internal link\n");
4516 +                               iput(delegated);
4517 +                       }
4518 +                       dput(h_src);
4519 +                       goto out_parent;
4520 +               } else
4521 +                       /* todo: cpup_wh_file? */
4522 +                       /* udba work */
4523 +                       au_update_ibrange(inode, /*do_put_zero*/1);
4524 +       }
4525 +
4526 +       isdir = S_ISDIR(inode->i_mode);
4527 +       old_ibstart = au_ibstart(inode);
4528 +       err = cpup_entry(cpg, dst_parent, &a->h_src_attr);
4529 +       if (unlikely(err))
4530 +               goto out_rev;
4531 +       dst_inode = h_dst->d_inode;
4532 +       mutex_lock_nested(&dst_inode->i_mutex, AuLsc_I_CHILD2);
4533 +       /* todo: necessary? */
4534 +       /* au_pin_hdir_unlock(cpg->pin); */
4535 +
4536 +       err = cpup_iattr(cpg->dentry, cpg->bdst, h_src, &a->h_src_attr);
4537 +       if (unlikely(err)) {
4538 +               /* todo: necessary? */
4539 +               /* au_pin_hdir_relock(cpg->pin); */ /* ignore an error */
4540 +               mutex_unlock(&dst_inode->i_mutex);
4541 +               goto out_rev;
4542 +       }
4543 +
4544 +       if (cpg->bdst < old_ibstart) {
4545 +               if (S_ISREG(inode->i_mode)) {
4546 +                       err = au_dy_iaop(inode, cpg->bdst, dst_inode);
4547 +                       if (unlikely(err)) {
4548 +                               /* ignore an error */
4549 +                               /* au_pin_hdir_relock(cpg->pin); */
4550 +                               mutex_unlock(&dst_inode->i_mutex);
4551 +                               goto out_rev;
4552 +                       }
4553 +               }
4554 +               au_set_ibstart(inode, cpg->bdst);
4555 +       } else
4556 +               au_set_ibend(inode, cpg->bdst);
4557 +       au_set_h_iptr(inode, cpg->bdst, au_igrab(dst_inode),
4558 +                     au_hi_flags(inode, isdir));
4559 +
4560 +       /* todo: necessary? */
4561 +       /* err = au_pin_hdir_relock(cpg->pin); */
4562 +       mutex_unlock(&dst_inode->i_mutex);
4563 +       if (unlikely(err))
4564 +               goto out_rev;
4565 +
4566 +       if (!isdir
4567 +           && h_src->d_inode->i_nlink > 1
4568 +           && plink)
4569 +               au_plink_append(inode, cpg->bdst, h_dst);
4570 +
4571 +       if (au_ftest_cpup(cpg->flags, RENAME)) {
4572 +               a->h_path.dentry = h_dst;
4573 +               err = au_do_ren_after_cpup(cpg, &a->h_path);
4574 +       }
4575 +       if (!err)
4576 +               goto out_parent; /* success */
4577 +
4578 +       /* revert */
4579 +out_rev:
4580 +       a->h_path.dentry = h_parent;
4581 +       au_dtime_store(&a->dt, dst_parent, &a->h_path);
4582 +       a->h_path.dentry = h_dst;
4583 +       rerr = 0;
4584 +       if (h_dst->d_inode) {
4585 +               if (!isdir) {
4586 +                       /* no delegation since it is just created */
4587 +                       rerr = vfsub_unlink(h_dir, &a->h_path,
4588 +                                           /*delegated*/NULL, /*force*/0);
4589 +               } else
4590 +                       rerr = vfsub_rmdir(h_dir, &a->h_path);
4591 +       }
4592 +       au_dtime_revert(&a->dt);
4593 +       if (rerr) {
4594 +               AuIOErr("failed removing broken entry(%d, %d)\n", err, rerr);
4595 +               err = -EIO;
4596 +       }
4597 +out_parent:
4598 +       dput(dst_parent);
4599 +       kfree(a);
4600 +out:
4601 +       return err;
4602 +}
4603 +
4604 +#if 0 /* unused */
4605 +struct au_cpup_single_args {
4606 +       int *errp;
4607 +       struct au_cp_generic *cpg;
4608 +       struct dentry *dst_parent;
4609 +};
4610 +
4611 +static void au_call_cpup_single(void *args)
4612 +{
4613 +       struct au_cpup_single_args *a = args;
4614 +
4615 +       au_pin_hdir_acquire_nest(a->cpg->pin);
4616 +       *a->errp = au_cpup_single(a->cpg, a->dst_parent);
4617 +       au_pin_hdir_release(a->cpg->pin);
4618 +}
4619 +#endif
4620 +
4621 +/*
4622 + * prevent SIGXFSZ in copy-up.
4623 + * testing CAP_MKNOD is for generic fs,
4624 + * but CAP_FSETID is for xfs only, currently.
4625 + */
4626 +static int au_cpup_sio_test(struct au_pin *pin, umode_t mode)
4627 +{
4628 +       int do_sio;
4629 +       struct super_block *sb;
4630 +       struct inode *h_dir;
4631 +
4632 +       do_sio = 0;
4633 +       sb = au_pinned_parent(pin)->d_sb;
4634 +       if (!au_wkq_test()
4635 +           && (!au_sbi(sb)->si_plink_maint_pid
4636 +               || au_plink_maint(sb, AuLock_NOPLM))) {
4637 +               switch (mode & S_IFMT) {
4638 +               case S_IFREG:
4639 +                       /* no condition about RLIMIT_FSIZE and the file size */
4640 +                       do_sio = 1;
4641 +                       break;
4642 +               case S_IFCHR:
4643 +               case S_IFBLK:
4644 +                       do_sio = !capable(CAP_MKNOD);
4645 +                       break;
4646 +               }
4647 +               if (!do_sio)
4648 +                       do_sio = ((mode & (S_ISUID | S_ISGID))
4649 +                                 && !capable(CAP_FSETID));
4650 +               /* this workaround may be removed in the future */
4651 +               if (!do_sio) {
4652 +                       h_dir = au_pinned_h_dir(pin);
4653 +                       do_sio = h_dir->i_mode & S_ISVTX;
4654 +               }
4655 +       }
4656 +
4657 +       return do_sio;
4658 +}
4659 +
4660 +#if 0 /* unused */
4661 +int au_sio_cpup_single(struct au_cp_generic *cpg, struct dentry *dst_parent)
4662 +{
4663 +       int err, wkq_err;
4664 +       struct dentry *h_dentry;
4665 +
4666 +       h_dentry = au_h_dptr(cpg->dentry, cpg->bsrc);
4667 +       if (!au_cpup_sio_test(pin, h_dentry->d_inode->i_mode))
4668 +               err = au_cpup_single(cpg, dst_parent);
4669 +       else {
4670 +               struct au_cpup_single_args args = {
4671 +                       .errp           = &err,
4672 +                       .cpg            = cpg,
4673 +                       .dst_parent     = dst_parent
4674 +               };
4675 +               wkq_err = au_wkq_wait(au_call_cpup_single, &args);
4676 +               if (unlikely(wkq_err))
4677 +                       err = wkq_err;
4678 +       }
4679 +
4680 +       return err;
4681 +}
4682 +#endif
4683 +
4684 +/*
4685 + * copyup the @dentry from the first active lower branch to @bdst,
4686 + * using au_cpup_single().
4687 + */
4688 +static int au_cpup_simple(struct au_cp_generic *cpg)
4689 +{
4690 +       int err;
4691 +       unsigned int flags_orig;
4692 +       struct dentry *dentry;
4693 +
4694 +       AuDebugOn(cpg->bsrc < 0);
4695 +
4696 +       dentry = cpg->dentry;
4697 +       DiMustWriteLock(dentry);
4698 +
4699 +       err = au_lkup_neg(dentry, cpg->bdst, /*wh*/1);
4700 +       if (!err) {
4701 +               flags_orig = cpg->flags;
4702 +               au_fset_cpup(cpg->flags, RENAME);
4703 +               err = au_cpup_single(cpg, NULL);
4704 +               cpg->flags = flags_orig;
4705 +               if (!err)
4706 +                       return 0; /* success */
4707 +
4708 +               /* revert */
4709 +               au_set_h_dptr(dentry, cpg->bdst, NULL);
4710 +               au_set_dbstart(dentry, cpg->bsrc);
4711 +       }
4712 +
4713 +       return err;
4714 +}
4715 +
4716 +struct au_cpup_simple_args {
4717 +       int *errp;
4718 +       struct au_cp_generic *cpg;
4719 +};
4720 +
4721 +static void au_call_cpup_simple(void *args)
4722 +{
4723 +       struct au_cpup_simple_args *a = args;
4724 +
4725 +       au_pin_hdir_acquire_nest(a->cpg->pin);
4726 +       *a->errp = au_cpup_simple(a->cpg);
4727 +       au_pin_hdir_release(a->cpg->pin);
4728 +}
4729 +
4730 +static int au_do_sio_cpup_simple(struct au_cp_generic *cpg)
4731 +{
4732 +       int err, wkq_err;
4733 +       struct dentry *dentry, *parent;
4734 +       struct file *h_file;
4735 +       struct inode *h_dir;
4736 +
4737 +       dentry = cpg->dentry;
4738 +       h_file = NULL;
4739 +       if (au_ftest_cpup(cpg->flags, HOPEN)) {
4740 +               AuDebugOn(cpg->bsrc < 0);
4741 +               h_file = au_h_open_pre(dentry, cpg->bsrc, /*force_wr*/0);
4742 +               err = PTR_ERR(h_file);
4743 +               if (IS_ERR(h_file))
4744 +                       goto out;
4745 +       }
4746 +
4747 +       parent = dget_parent(dentry);
4748 +       h_dir = au_h_iptr(parent->d_inode, cpg->bdst);
4749 +       if (!au_test_h_perm_sio(h_dir, MAY_EXEC | MAY_WRITE)
4750 +           && !au_cpup_sio_test(cpg->pin, dentry->d_inode->i_mode))
4751 +               err = au_cpup_simple(cpg);
4752 +       else {
4753 +               struct au_cpup_simple_args args = {
4754 +                       .errp           = &err,
4755 +                       .cpg            = cpg
4756 +               };
4757 +               wkq_err = au_wkq_wait(au_call_cpup_simple, &args);
4758 +               if (unlikely(wkq_err))
4759 +                       err = wkq_err;
4760 +       }
4761 +
4762 +       dput(parent);
4763 +       if (h_file)
4764 +               au_h_open_post(dentry, cpg->bsrc, h_file);
4765 +
4766 +out:
4767 +       return err;
4768 +}
4769 +
4770 +int au_sio_cpup_simple(struct au_cp_generic *cpg)
4771 +{
4772 +       aufs_bindex_t bsrc, bend;
4773 +       struct dentry *dentry, *h_dentry;
4774 +
4775 +       if (cpg->bsrc < 0) {
4776 +               dentry = cpg->dentry;
4777 +               bend = au_dbend(dentry);
4778 +               for (bsrc = cpg->bdst + 1; bsrc <= bend; bsrc++) {
4779 +                       h_dentry = au_h_dptr(dentry, bsrc);
4780 +                       if (h_dentry) {
4781 +                               AuDebugOn(!h_dentry->d_inode);
4782 +                               break;
4783 +                       }
4784 +               }
4785 +               AuDebugOn(bsrc > bend);
4786 +               cpg->bsrc = bsrc;
4787 +       }
4788 +       AuDebugOn(cpg->bsrc <= cpg->bdst);
4789 +       return au_do_sio_cpup_simple(cpg);
4790 +}
4791 +
4792 +int au_sio_cpdown_simple(struct au_cp_generic *cpg)
4793 +{
4794 +       AuDebugOn(cpg->bdst <= cpg->bsrc);
4795 +       return au_do_sio_cpup_simple(cpg);
4796 +}
4797 +
4798 +/* ---------------------------------------------------------------------- */
4799 +
4800 +/*
4801 + * copyup the deleted file for writing.
4802 + */
4803 +static int au_do_cpup_wh(struct au_cp_generic *cpg, struct dentry *wh_dentry,
4804 +                        struct file *file)
4805 +{
4806 +       int err;
4807 +       unsigned int flags_orig;
4808 +       aufs_bindex_t bsrc_orig;
4809 +       struct dentry *h_d_dst, *h_d_start;
4810 +       struct au_dinfo *dinfo;
4811 +       struct au_hdentry *hdp;
4812 +
4813 +       dinfo = au_di(cpg->dentry);
4814 +       AuRwMustWriteLock(&dinfo->di_rwsem);
4815 +
4816 +       bsrc_orig = cpg->bsrc;
4817 +       cpg->bsrc = dinfo->di_bstart;
4818 +       hdp = dinfo->di_hdentry;
4819 +       h_d_dst = hdp[0 + cpg->bdst].hd_dentry;
4820 +       dinfo->di_bstart = cpg->bdst;
4821 +       hdp[0 + cpg->bdst].hd_dentry = wh_dentry;
4822 +       h_d_start = NULL;
4823 +       if (file) {
4824 +               h_d_start = hdp[0 + cpg->bsrc].hd_dentry;
4825 +               hdp[0 + cpg->bsrc].hd_dentry = au_hf_top(file)->f_dentry;
4826 +       }
4827 +       flags_orig = cpg->flags;
4828 +       cpg->flags = !AuCpup_DTIME;
4829 +       err = au_cpup_single(cpg, /*h_parent*/NULL);
4830 +       cpg->flags = flags_orig;
4831 +       if (file) {
4832 +               if (!err)
4833 +                       err = au_reopen_nondir(file);
4834 +               hdp[0 + cpg->bsrc].hd_dentry = h_d_start;
4835 +       }
4836 +       hdp[0 + cpg->bdst].hd_dentry = h_d_dst;
4837 +       dinfo->di_bstart = cpg->bsrc;
4838 +       cpg->bsrc = bsrc_orig;
4839 +
4840 +       return err;
4841 +}
4842 +
4843 +static int au_cpup_wh(struct au_cp_generic *cpg, struct file *file)
4844 +{
4845 +       int err;
4846 +       aufs_bindex_t bdst;
4847 +       struct au_dtime dt;
4848 +       struct dentry *dentry, *parent, *h_parent, *wh_dentry;
4849 +       struct au_branch *br;
4850 +       struct path h_path;
4851 +
4852 +       dentry = cpg->dentry;
4853 +       bdst = cpg->bdst;
4854 +       br = au_sbr(dentry->d_sb, bdst);
4855 +       parent = dget_parent(dentry);
4856 +       h_parent = au_h_dptr(parent, bdst);
4857 +       wh_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
4858 +       err = PTR_ERR(wh_dentry);
4859 +       if (IS_ERR(wh_dentry))
4860 +               goto out;
4861 +
4862 +       h_path.dentry = h_parent;
4863 +       h_path.mnt = au_br_mnt(br);
4864 +       au_dtime_store(&dt, parent, &h_path);
4865 +       err = au_do_cpup_wh(cpg, wh_dentry, file);
4866 +       if (unlikely(err))
4867 +               goto out_wh;
4868 +
4869 +       dget(wh_dentry);
4870 +       h_path.dentry = wh_dentry;
4871 +       if (!S_ISDIR(wh_dentry->d_inode->i_mode)) {
4872 +               /* no delegation since it is just created */
4873 +               err = vfsub_unlink(h_parent->d_inode, &h_path,
4874 +                                  /*delegated*/NULL, /*force*/0);
4875 +       } else
4876 +               err = vfsub_rmdir(h_parent->d_inode, &h_path);
4877 +       if (unlikely(err)) {
4878 +               AuIOErr("failed remove copied-up tmp file %pd(%d)\n",
4879 +                       wh_dentry, err);
4880 +               err = -EIO;
4881 +       }
4882 +       au_dtime_revert(&dt);
4883 +       au_set_hi_wh(dentry->d_inode, bdst, wh_dentry);
4884 +
4885 +out_wh:
4886 +       dput(wh_dentry);
4887 +out:
4888 +       dput(parent);
4889 +       return err;
4890 +}
4891 +
4892 +struct au_cpup_wh_args {
4893 +       int *errp;
4894 +       struct au_cp_generic *cpg;
4895 +       struct file *file;
4896 +};
4897 +
4898 +static void au_call_cpup_wh(void *args)
4899 +{
4900 +       struct au_cpup_wh_args *a = args;
4901 +
4902 +       au_pin_hdir_acquire_nest(a->cpg->pin);
4903 +       *a->errp = au_cpup_wh(a->cpg, a->file);
4904 +       au_pin_hdir_release(a->cpg->pin);
4905 +}
4906 +
4907 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file)
4908 +{
4909 +       int err, wkq_err;
4910 +       aufs_bindex_t bdst;
4911 +       struct dentry *dentry, *parent, *h_orph, *h_parent, *h_dentry;
4912 +       struct inode *dir, *h_dir, *h_tmpdir;
4913 +       struct au_wbr *wbr;
4914 +       struct au_pin wh_pin, *pin_orig;
4915 +
4916 +       dentry = cpg->dentry;
4917 +       bdst = cpg->bdst;
4918 +       parent = dget_parent(dentry);
4919 +       dir = parent->d_inode;
4920 +       h_orph = NULL;
4921 +       h_parent = NULL;
4922 +       h_dir = au_igrab(au_h_iptr(dir, bdst));
4923 +       h_tmpdir = h_dir;
4924 +       pin_orig = NULL;
4925 +       if (!h_dir->i_nlink) {
4926 +               wbr = au_sbr(dentry->d_sb, bdst)->br_wbr;
4927 +               h_orph = wbr->wbr_orph;
4928 +
4929 +               h_parent = dget(au_h_dptr(parent, bdst));
4930 +               au_set_h_dptr(parent, bdst, dget(h_orph));
4931 +               h_tmpdir = h_orph->d_inode;
4932 +               au_set_h_iptr(dir, bdst, au_igrab(h_tmpdir), /*flags*/0);
4933 +
4934 +               if (file)
4935 +                       h_dentry = au_hf_top(file)->f_dentry;
4936 +               else
4937 +                       h_dentry = au_h_dptr(dentry, au_dbstart(dentry));
4938 +               mutex_lock_nested(&h_tmpdir->i_mutex, AuLsc_I_PARENT3);
4939 +               /* todo: au_h_open_pre()? */
4940 +
4941 +               pin_orig = cpg->pin;
4942 +               au_pin_init(&wh_pin, dentry, bdst, AuLsc_DI_PARENT,
4943 +                           AuLsc_I_PARENT3, cpg->pin->udba, AuPin_DI_LOCKED);
4944 +               cpg->pin = &wh_pin;
4945 +       }
4946 +
4947 +       if (!au_test_h_perm_sio(h_tmpdir, MAY_EXEC | MAY_WRITE)
4948 +           && !au_cpup_sio_test(cpg->pin, dentry->d_inode->i_mode))
4949 +               err = au_cpup_wh(cpg, file);
4950 +       else {
4951 +               struct au_cpup_wh_args args = {
4952 +                       .errp   = &err,
4953 +                       .cpg    = cpg,
4954 +                       .file   = file
4955 +               };
4956 +               wkq_err = au_wkq_wait(au_call_cpup_wh, &args);
4957 +               if (unlikely(wkq_err))
4958 +                       err = wkq_err;
4959 +       }
4960 +
4961 +       if (h_orph) {
4962 +               mutex_unlock(&h_tmpdir->i_mutex);
4963 +               /* todo: au_h_open_post()? */
4964 +               au_set_h_iptr(dir, bdst, au_igrab(h_dir), /*flags*/0);
4965 +               au_set_h_dptr(parent, bdst, h_parent);
4966 +               AuDebugOn(!pin_orig);
4967 +               cpg->pin = pin_orig;
4968 +       }
4969 +       iput(h_dir);
4970 +       dput(parent);
4971 +
4972 +       return err;
4973 +}
4974 +
4975 +/* ---------------------------------------------------------------------- */
4976 +
4977 +/*
4978 + * generic routine for both of copy-up and copy-down.
4979 + */
4980 +/* cf. revalidate function in file.c */
4981 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
4982 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
4983 +                        struct au_pin *pin,
4984 +                        struct dentry *h_parent, void *arg),
4985 +              void *arg)
4986 +{
4987 +       int err;
4988 +       struct au_pin pin;
4989 +       struct dentry *d, *parent, *h_parent, *real_parent;
4990 +
4991 +       err = 0;
4992 +       parent = dget_parent(dentry);
4993 +       if (IS_ROOT(parent))
4994 +               goto out;
4995 +
4996 +       au_pin_init(&pin, dentry, bdst, AuLsc_DI_PARENT2, AuLsc_I_PARENT2,
4997 +                   au_opt_udba(dentry->d_sb), AuPin_MNT_WRITE);
4998 +
4999 +       /* do not use au_dpage */
5000 +       real_parent = parent;
5001 +       while (1) {
5002 +               dput(parent);
5003 +               parent = dget_parent(dentry);
5004 +               h_parent = au_h_dptr(parent, bdst);
5005 +               if (h_parent)
5006 +                       goto out; /* success */
5007 +
5008 +               /* find top dir which is necessary to cpup */
5009 +               do {
5010 +                       d = parent;
5011 +                       dput(parent);
5012 +                       parent = dget_parent(d);
5013 +                       di_read_lock_parent3(parent, !AuLock_IR);
5014 +                       h_parent = au_h_dptr(parent, bdst);
5015 +                       di_read_unlock(parent, !AuLock_IR);
5016 +               } while (!h_parent);
5017 +
5018 +               if (d != real_parent)
5019 +                       di_write_lock_child3(d);
5020 +
5021 +               /* somebody else might create while we were sleeping */
5022 +               if (!au_h_dptr(d, bdst) || !au_h_dptr(d, bdst)->d_inode) {
5023 +                       if (au_h_dptr(d, bdst))
5024 +                               au_update_dbstart(d);
5025 +
5026 +                       au_pin_set_dentry(&pin, d);
5027 +                       err = au_do_pin(&pin);
5028 +                       if (!err) {
5029 +                               err = cp(d, bdst, &pin, h_parent, arg);
5030 +                               au_unpin(&pin);
5031 +                       }
5032 +               }
5033 +
5034 +               if (d != real_parent)
5035 +                       di_write_unlock(d);
5036 +               if (unlikely(err))
5037 +                       break;
5038 +       }
5039 +
5040 +out:
5041 +       dput(parent);
5042 +       return err;
5043 +}
5044 +
5045 +static int au_cpup_dir(struct dentry *dentry, aufs_bindex_t bdst,
5046 +                      struct au_pin *pin,
5047 +                      struct dentry *h_parent __maybe_unused ,
5048 +                      void *arg __maybe_unused)
5049 +{
5050 +       struct au_cp_generic cpg = {
5051 +               .dentry = dentry,
5052 +               .bdst   = bdst,
5053 +               .bsrc   = -1,
5054 +               .len    = 0,
5055 +               .pin    = pin,
5056 +               .flags  = AuCpup_DTIME
5057 +       };
5058 +       return au_sio_cpup_simple(&cpg);
5059 +}
5060 +
5061 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
5062 +{
5063 +       return au_cp_dirs(dentry, bdst, au_cpup_dir, NULL);
5064 +}
5065 +
5066 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst)
5067 +{
5068 +       int err;
5069 +       struct dentry *parent;
5070 +       struct inode *dir;
5071 +
5072 +       parent = dget_parent(dentry);
5073 +       dir = parent->d_inode;
5074 +       err = 0;
5075 +       if (au_h_iptr(dir, bdst))
5076 +               goto out;
5077 +
5078 +       di_read_unlock(parent, AuLock_IR);
5079 +       di_write_lock_parent(parent);
5080 +       /* someone else might change our inode while we were sleeping */
5081 +       if (!au_h_iptr(dir, bdst))
5082 +               err = au_cpup_dirs(dentry, bdst);
5083 +       di_downgrade_lock(parent, AuLock_IR);
5084 +
5085 +out:
5086 +       dput(parent);
5087 +       return err;
5088 +}
5089 diff -urN /usr/share/empty/fs/aufs/cpup.h linux/fs/aufs/cpup.h
5090 --- /usr/share/empty/fs/aufs/cpup.h     1970-01-01 01:00:00.000000000 +0100
5091 +++ linux/fs/aufs/cpup.h        2014-04-24 22:11:10.848602379 +0200
5092 @@ -0,0 +1,94 @@
5093 +/*
5094 + * Copyright (C) 2005-2014 Junjiro R. Okajima
5095 + *
5096 + * This program, aufs is free software; you can redistribute it and/or modify
5097 + * it under the terms of the GNU General Public License as published by
5098 + * the Free Software Foundation; either version 2 of the License, or
5099 + * (at your option) any later version.
5100 + *
5101 + * This program is distributed in the hope that it will be useful,
5102 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5103 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5104 + * GNU General Public License for more details.
5105 + *
5106 + * You should have received a copy of the GNU General Public License
5107 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5108 + */
5109 +
5110 +/*
5111 + * copy-up/down functions
5112 + */
5113 +
5114 +#ifndef __AUFS_CPUP_H__
5115 +#define __AUFS_CPUP_H__
5116 +
5117 +#ifdef __KERNEL__
5118 +
5119 +#include <linux/path.h>
5120 +
5121 +struct inode;
5122 +struct file;
5123 +struct au_pin;
5124 +
5125 +void au_cpup_attr_flags(struct inode *dst, unsigned int iflags);
5126 +void au_cpup_attr_timesizes(struct inode *inode);
5127 +void au_cpup_attr_nlink(struct inode *inode, int force);
5128 +void au_cpup_attr_changeable(struct inode *inode);
5129 +void au_cpup_igen(struct inode *inode, struct inode *h_inode);
5130 +void au_cpup_attr_all(struct inode *inode, int force);
5131 +
5132 +/* ---------------------------------------------------------------------- */
5133 +
5134 +struct au_cp_generic {
5135 +       struct dentry   *dentry;
5136 +       aufs_bindex_t   bdst, bsrc;
5137 +       loff_t          len;
5138 +       struct au_pin   *pin;
5139 +       unsigned int    flags;
5140 +};
5141 +
5142 +/* cpup flags */
5143 +#define AuCpup_DTIME           1               /* do dtime_store/revert */
5144 +#define AuCpup_KEEPLINO                (1 << 1)        /* do not clear the lower xino,
5145 +                                                  for link(2) */
5146 +#define AuCpup_RENAME          (1 << 2)        /* rename after cpup */
5147 +#define AuCpup_HOPEN           (1 << 3)        /* call h_open_pre/post() in
5148 +                                                  cpup */
5149 +#define AuCpup_OVERWRITE       (1 << 4)        /* allow overwriting the
5150 +                                                  existing entry */
5151 +#define AuCpup_RWDST           (1 << 5)        /* force write target even if
5152 +                                                  the branch is marked as RO */
5153 +
5154 +#define au_ftest_cpup(flags, name)     ((flags) & AuCpup_##name)
5155 +#define au_fset_cpup(flags, name) \
5156 +       do { (flags) |= AuCpup_##name; } while (0)
5157 +#define au_fclr_cpup(flags, name) \
5158 +       do { (flags) &= ~AuCpup_##name; } while (0)
5159 +
5160 +int au_copy_file(struct file *dst, struct file *src, loff_t len);
5161 +int au_sio_cpup_simple(struct au_cp_generic *cpg);
5162 +int au_sio_cpdown_simple(struct au_cp_generic *cpg);
5163 +int au_sio_cpup_wh(struct au_cp_generic *cpg, struct file *file);
5164 +
5165 +int au_cp_dirs(struct dentry *dentry, aufs_bindex_t bdst,
5166 +              int (*cp)(struct dentry *dentry, aufs_bindex_t bdst,
5167 +                        struct au_pin *pin,
5168 +                        struct dentry *h_parent, void *arg),
5169 +              void *arg);
5170 +int au_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
5171 +int au_test_and_cpup_dirs(struct dentry *dentry, aufs_bindex_t bdst);
5172 +
5173 +/* ---------------------------------------------------------------------- */
5174 +
5175 +/* keep timestamps when copyup */
5176 +struct au_dtime {
5177 +       struct dentry *dt_dentry;
5178 +       struct path dt_h_path;
5179 +       struct timespec dt_atime, dt_mtime;
5180 +};
5181 +void au_dtime_store(struct au_dtime *dt, struct dentry *dentry,
5182 +                   struct path *h_path);
5183 +void au_dtime_revert(struct au_dtime *dt);
5184 +
5185 +#endif /* __KERNEL__ */
5186 +#endif /* __AUFS_CPUP_H__ */
5187 diff -urN /usr/share/empty/fs/aufs/dbgaufs.c linux/fs/aufs/dbgaufs.c
5188 --- /usr/share/empty/fs/aufs/dbgaufs.c  1970-01-01 01:00:00.000000000 +0100
5189 +++ linux/fs/aufs/dbgaufs.c     2014-04-24 22:11:10.848602379 +0200
5190 @@ -0,0 +1,432 @@
5191 +/*
5192 + * Copyright (C) 2005-2014 Junjiro R. Okajima
5193 + *
5194 + * This program, aufs is free software; you can redistribute it and/or modify
5195 + * it under the terms of the GNU General Public License as published by
5196 + * the Free Software Foundation; either version 2 of the License, or
5197 + * (at your option) any later version.
5198 + *
5199 + * This program is distributed in the hope that it will be useful,
5200 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5201 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5202 + * GNU General Public License for more details.
5203 + *
5204 + * You should have received a copy of the GNU General Public License
5205 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5206 + */
5207 +
5208 +/*
5209 + * debugfs interface
5210 + */
5211 +
5212 +#include <linux/debugfs.h>
5213 +#include "aufs.h"
5214 +
5215 +#ifndef CONFIG_SYSFS
5216 +#error DEBUG_FS depends upon SYSFS
5217 +#endif
5218 +
5219 +static struct dentry *dbgaufs;
5220 +static const mode_t dbgaufs_mode = S_IRUSR | S_IRGRP | S_IROTH;
5221 +
5222 +/* 20 is max digits length of ulong 64 */
5223 +struct dbgaufs_arg {
5224 +       int n;
5225 +       char a[20 * 4];
5226 +};
5227 +
5228 +/*
5229 + * common function for all XINO files
5230 + */
5231 +static int dbgaufs_xi_release(struct inode *inode __maybe_unused,
5232 +                             struct file *file)
5233 +{
5234 +       kfree(file->private_data);
5235 +       return 0;
5236 +}
5237 +
5238 +static int dbgaufs_xi_open(struct file *xf, struct file *file, int do_fcnt)
5239 +{
5240 +       int err;
5241 +       struct kstat st;
5242 +       struct dbgaufs_arg *p;
5243 +
5244 +       err = -ENOMEM;
5245 +       p = kmalloc(sizeof(*p), GFP_NOFS);
5246 +       if (unlikely(!p))
5247 +               goto out;
5248 +
5249 +       err = 0;
5250 +       p->n = 0;
5251 +       file->private_data = p;
5252 +       if (!xf)
5253 +               goto out;
5254 +
5255 +       err = vfs_getattr(&xf->f_path, &st);
5256 +       if (!err) {
5257 +               if (do_fcnt)
5258 +                       p->n = snprintf
5259 +                               (p->a, sizeof(p->a), "%ld, %llux%lu %lld\n",
5260 +                                (long)file_count(xf), st.blocks, st.blksize,
5261 +                                (long long)st.size);
5262 +               else
5263 +                       p->n = snprintf(p->a, sizeof(p->a), "%llux%lu %lld\n",
5264 +                                       st.blocks, st.blksize,
5265 +                                       (long long)st.size);
5266 +               AuDebugOn(p->n >= sizeof(p->a));
5267 +       } else {
5268 +               p->n = snprintf(p->a, sizeof(p->a), "err %d\n", err);
5269 +               err = 0;
5270 +       }
5271 +
5272 +out:
5273 +       return err;
5274 +
5275 +}
5276 +
5277 +static ssize_t dbgaufs_xi_read(struct file *file, char __user *buf,
5278 +                              size_t count, loff_t *ppos)
5279 +{
5280 +       struct dbgaufs_arg *p;
5281 +
5282 +       p = file->private_data;
5283 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
5284 +}
5285 +
5286 +/* ---------------------------------------------------------------------- */
5287 +
5288 +struct dbgaufs_plink_arg {
5289 +       int n;
5290 +       char a[];
5291 +};
5292 +
5293 +static int dbgaufs_plink_release(struct inode *inode __maybe_unused,
5294 +                                struct file *file)
5295 +{
5296 +       free_page((unsigned long)file->private_data);
5297 +       return 0;
5298 +}
5299 +
5300 +static int dbgaufs_plink_open(struct inode *inode, struct file *file)
5301 +{
5302 +       int err, i, limit;
5303 +       unsigned long n, sum;
5304 +       struct dbgaufs_plink_arg *p;
5305 +       struct au_sbinfo *sbinfo;
5306 +       struct super_block *sb;
5307 +       struct au_sphlhead *sphl;
5308 +
5309 +       err = -ENOMEM;
5310 +       p = (void *)get_zeroed_page(GFP_NOFS);
5311 +       if (unlikely(!p))
5312 +               goto out;
5313 +
5314 +       err = -EFBIG;
5315 +       sbinfo = inode->i_private;
5316 +       sb = sbinfo->si_sb;
5317 +       si_noflush_read_lock(sb);
5318 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
5319 +               limit = PAGE_SIZE - sizeof(p->n);
5320 +
5321 +               /* the number of buckets */
5322 +               n = snprintf(p->a + p->n, limit, "%d\n", AuPlink_NHASH);
5323 +               p->n += n;
5324 +               limit -= n;
5325 +
5326 +               sum = 0;
5327 +               for (i = 0, sphl = sbinfo->si_plink;
5328 +                    i < AuPlink_NHASH;
5329 +                    i++, sphl++) {
5330 +                       n = au_sphl_count(sphl);
5331 +                       sum += n;
5332 +
5333 +                       n = snprintf(p->a + p->n, limit, "%lu ", n);
5334 +                       p->n += n;
5335 +                       limit -= n;
5336 +                       if (unlikely(limit <= 0))
5337 +                               goto out_free;
5338 +               }
5339 +               p->a[p->n - 1] = '\n';
5340 +
5341 +               /* the sum of plinks */
5342 +               n = snprintf(p->a + p->n, limit, "%lu\n", sum);
5343 +               p->n += n;
5344 +               limit -= n;
5345 +               if (unlikely(limit <= 0))
5346 +                       goto out_free;
5347 +       } else {
5348 +#define str "1\n0\n0\n"
5349 +               p->n = sizeof(str) - 1;
5350 +               strcpy(p->a, str);
5351 +#undef str
5352 +       }
5353 +       si_read_unlock(sb);
5354 +
5355 +       err = 0;
5356 +       file->private_data = p;
5357 +       goto out; /* success */
5358 +
5359 +out_free:
5360 +       free_page((unsigned long)p);
5361 +out:
5362 +       return err;
5363 +}
5364 +
5365 +static ssize_t dbgaufs_plink_read(struct file *file, char __user *buf,
5366 +                                 size_t count, loff_t *ppos)
5367 +{
5368 +       struct dbgaufs_plink_arg *p;
5369 +
5370 +       p = file->private_data;
5371 +       return simple_read_from_buffer(buf, count, ppos, p->a, p->n);
5372 +}
5373 +
5374 +static const struct file_operations dbgaufs_plink_fop = {
5375 +       .owner          = THIS_MODULE,
5376 +       .open           = dbgaufs_plink_open,
5377 +       .release        = dbgaufs_plink_release,
5378 +       .read           = dbgaufs_plink_read
5379 +};
5380 +
5381 +/* ---------------------------------------------------------------------- */
5382 +
5383 +static int dbgaufs_xib_open(struct inode *inode, struct file *file)
5384 +{
5385 +       int err;
5386 +       struct au_sbinfo *sbinfo;
5387 +       struct super_block *sb;
5388 +
5389 +       sbinfo = inode->i_private;
5390 +       sb = sbinfo->si_sb;
5391 +       si_noflush_read_lock(sb);
5392 +       err = dbgaufs_xi_open(sbinfo->si_xib, file, /*do_fcnt*/0);
5393 +       si_read_unlock(sb);
5394 +       return err;
5395 +}
5396 +
5397 +static const struct file_operations dbgaufs_xib_fop = {
5398 +       .owner          = THIS_MODULE,
5399 +       .open           = dbgaufs_xib_open,
5400 +       .release        = dbgaufs_xi_release,
5401 +       .read           = dbgaufs_xi_read
5402 +};
5403 +
5404 +/* ---------------------------------------------------------------------- */
5405 +
5406 +#define DbgaufsXi_PREFIX "xi"
5407 +
5408 +static int dbgaufs_xino_open(struct inode *inode, struct file *file)
5409 +{
5410 +       int err;
5411 +       long l;
5412 +       struct au_sbinfo *sbinfo;
5413 +       struct super_block *sb;
5414 +       struct file *xf;
5415 +       struct qstr *name;
5416 +
5417 +       err = -ENOENT;
5418 +       xf = NULL;
5419 +       name = &file->f_dentry->d_name;
5420 +       if (unlikely(name->len < sizeof(DbgaufsXi_PREFIX)
5421 +                    || memcmp(name->name, DbgaufsXi_PREFIX,
5422 +                              sizeof(DbgaufsXi_PREFIX) - 1)))
5423 +               goto out;
5424 +       err = kstrtol(name->name + sizeof(DbgaufsXi_PREFIX) - 1, 10, &l);
5425 +       if (unlikely(err))
5426 +               goto out;
5427 +
5428 +       sbinfo = inode->i_private;
5429 +       sb = sbinfo->si_sb;
5430 +       si_noflush_read_lock(sb);
5431 +       if (l <= au_sbend(sb)) {
5432 +               xf = au_sbr(sb, (aufs_bindex_t)l)->br_xino.xi_file;
5433 +               err = dbgaufs_xi_open(xf, file, /*do_fcnt*/1);
5434 +       } else
5435 +               err = -ENOENT;
5436 +       si_read_unlock(sb);
5437 +
5438 +out:
5439 +       return err;
5440 +}
5441 +
5442 +static const struct file_operations dbgaufs_xino_fop = {
5443 +       .owner          = THIS_MODULE,
5444 +       .open           = dbgaufs_xino_open,
5445 +       .release        = dbgaufs_xi_release,
5446 +       .read           = dbgaufs_xi_read
5447 +};
5448 +
5449 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
5450 +{
5451 +       aufs_bindex_t bend;
5452 +       struct au_branch *br;
5453 +       struct au_xino_file *xi;
5454 +
5455 +       if (!au_sbi(sb)->si_dbgaufs)
5456 +               return;
5457 +
5458 +       bend = au_sbend(sb);
5459 +       for (; bindex <= bend; bindex++) {
5460 +               br = au_sbr(sb, bindex);
5461 +               xi = &br->br_xino;
5462 +               debugfs_remove(xi->xi_dbgaufs);
5463 +               xi->xi_dbgaufs = NULL;
5464 +       }
5465 +}
5466 +
5467 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
5468 +{
5469 +       struct au_sbinfo *sbinfo;
5470 +       struct dentry *parent;
5471 +       struct au_branch *br;
5472 +       struct au_xino_file *xi;
5473 +       aufs_bindex_t bend;
5474 +       char name[sizeof(DbgaufsXi_PREFIX) + 5]; /* "xi" bindex NULL */
5475 +
5476 +       sbinfo = au_sbi(sb);
5477 +       parent = sbinfo->si_dbgaufs;
5478 +       if (!parent)
5479 +               return;
5480 +
5481 +       bend = au_sbend(sb);
5482 +       for (; bindex <= bend; bindex++) {
5483 +               snprintf(name, sizeof(name), DbgaufsXi_PREFIX "%d", bindex);
5484 +               br = au_sbr(sb, bindex);
5485 +               xi = &br->br_xino;
5486 +               AuDebugOn(xi->xi_dbgaufs);
5487 +               xi->xi_dbgaufs = debugfs_create_file(name, dbgaufs_mode, parent,
5488 +                                                    sbinfo, &dbgaufs_xino_fop);
5489 +               /* ignore an error */
5490 +               if (unlikely(!xi->xi_dbgaufs))
5491 +                       AuWarn1("failed %s under debugfs\n", name);
5492 +       }
5493 +}
5494 +
5495 +/* ---------------------------------------------------------------------- */
5496 +
5497 +#ifdef CONFIG_AUFS_EXPORT
5498 +static int dbgaufs_xigen_open(struct inode *inode, struct file *file)
5499 +{
5500 +       int err;
5501 +       struct au_sbinfo *sbinfo;
5502 +       struct super_block *sb;
5503 +
5504 +       sbinfo = inode->i_private;
5505 +       sb = sbinfo->si_sb;
5506 +       si_noflush_read_lock(sb);
5507 +       err = dbgaufs_xi_open(sbinfo->si_xigen, file, /*do_fcnt*/0);
5508 +       si_read_unlock(sb);
5509 +       return err;
5510 +}
5511 +
5512 +static const struct file_operations dbgaufs_xigen_fop = {
5513 +       .owner          = THIS_MODULE,
5514 +       .open           = dbgaufs_xigen_open,
5515 +       .release        = dbgaufs_xi_release,
5516 +       .read           = dbgaufs_xi_read
5517 +};
5518 +
5519 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
5520 +{
5521 +       int err;
5522 +
5523 +       /*
5524 +        * This function is a dynamic '__init' fucntion actually,
5525 +        * so the tiny check for si_rwsem is unnecessary.
5526 +        */
5527 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
5528 +
5529 +       err = -EIO;
5530 +       sbinfo->si_dbgaufs_xigen = debugfs_create_file
5531 +               ("xigen", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
5532 +                &dbgaufs_xigen_fop);
5533 +       if (sbinfo->si_dbgaufs_xigen)
5534 +               err = 0;
5535 +
5536 +       return err;
5537 +}
5538 +#else
5539 +static int dbgaufs_xigen_init(struct au_sbinfo *sbinfo)
5540 +{
5541 +       return 0;
5542 +}
5543 +#endif /* CONFIG_AUFS_EXPORT */
5544 +
5545 +/* ---------------------------------------------------------------------- */
5546 +
5547 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo)
5548 +{
5549 +       /*
5550 +        * This function is a dynamic '__init' fucntion actually,
5551 +        * so the tiny check for si_rwsem is unnecessary.
5552 +        */
5553 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
5554 +
5555 +       debugfs_remove_recursive(sbinfo->si_dbgaufs);
5556 +       sbinfo->si_dbgaufs = NULL;
5557 +       kobject_put(&sbinfo->si_kobj);
5558 +}
5559 +
5560 +int dbgaufs_si_init(struct au_sbinfo *sbinfo)
5561 +{
5562 +       int err;
5563 +       char name[SysaufsSiNameLen];
5564 +
5565 +       /*
5566 +        * This function is a dynamic '__init' fucntion actually,
5567 +        * so the tiny check for si_rwsem is unnecessary.
5568 +        */
5569 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
5570 +
5571 +       err = -ENOENT;
5572 +       if (!dbgaufs) {
5573 +               AuErr1("/debug/aufs is uninitialized\n");
5574 +               goto out;
5575 +       }
5576 +
5577 +       err = -EIO;
5578 +       sysaufs_name(sbinfo, name);
5579 +       sbinfo->si_dbgaufs = debugfs_create_dir(name, dbgaufs);
5580 +       if (unlikely(!sbinfo->si_dbgaufs))
5581 +               goto out;
5582 +       kobject_get(&sbinfo->si_kobj);
5583 +
5584 +       sbinfo->si_dbgaufs_xib = debugfs_create_file
5585 +               ("xib", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
5586 +                &dbgaufs_xib_fop);
5587 +       if (unlikely(!sbinfo->si_dbgaufs_xib))
5588 +               goto out_dir;
5589 +
5590 +       sbinfo->si_dbgaufs_plink = debugfs_create_file
5591 +               ("plink", dbgaufs_mode, sbinfo->si_dbgaufs, sbinfo,
5592 +                &dbgaufs_plink_fop);
5593 +       if (unlikely(!sbinfo->si_dbgaufs_plink))
5594 +               goto out_dir;
5595 +
5596 +       err = dbgaufs_xigen_init(sbinfo);
5597 +       if (!err)
5598 +               goto out; /* success */
5599 +
5600 +out_dir:
5601 +       dbgaufs_si_fin(sbinfo);
5602 +out:
5603 +       return err;
5604 +}
5605 +
5606 +/* ---------------------------------------------------------------------- */
5607 +
5608 +void dbgaufs_fin(void)
5609 +{
5610 +       debugfs_remove(dbgaufs);
5611 +}
5612 +
5613 +int __init dbgaufs_init(void)
5614 +{
5615 +       int err;
5616 +
5617 +       err = -EIO;
5618 +       dbgaufs = debugfs_create_dir(AUFS_NAME, NULL);
5619 +       if (dbgaufs)
5620 +               err = 0;
5621 +       return err;
5622 +}
5623 diff -urN /usr/share/empty/fs/aufs/dbgaufs.h linux/fs/aufs/dbgaufs.h
5624 --- /usr/share/empty/fs/aufs/dbgaufs.h  1970-01-01 01:00:00.000000000 +0100
5625 +++ linux/fs/aufs/dbgaufs.h     2014-04-24 22:11:10.848602379 +0200
5626 @@ -0,0 +1,48 @@
5627 +/*
5628 + * Copyright (C) 2005-2014 Junjiro R. Okajima
5629 + *
5630 + * This program, aufs is free software; you can redistribute it and/or modify
5631 + * it under the terms of the GNU General Public License as published by
5632 + * the Free Software Foundation; either version 2 of the License, or
5633 + * (at your option) any later version.
5634 + *
5635 + * This program is distributed in the hope that it will be useful,
5636 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5637 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5638 + * GNU General Public License for more details.
5639 + *
5640 + * You should have received a copy of the GNU General Public License
5641 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5642 + */
5643 +
5644 +/*
5645 + * debugfs interface
5646 + */
5647 +
5648 +#ifndef __DBGAUFS_H__
5649 +#define __DBGAUFS_H__
5650 +
5651 +#ifdef __KERNEL__
5652 +
5653 +struct super_block;
5654 +struct au_sbinfo;
5655 +
5656 +#ifdef CONFIG_DEBUG_FS
5657 +/* dbgaufs.c */
5658 +void dbgaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
5659 +void dbgaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
5660 +void dbgaufs_si_fin(struct au_sbinfo *sbinfo);
5661 +int dbgaufs_si_init(struct au_sbinfo *sbinfo);
5662 +void dbgaufs_fin(void);
5663 +int __init dbgaufs_init(void);
5664 +#else
5665 +AuStubVoid(dbgaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
5666 +AuStubVoid(dbgaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
5667 +AuStubVoid(dbgaufs_si_fin, struct au_sbinfo *sbinfo)
5668 +AuStubInt0(dbgaufs_si_init, struct au_sbinfo *sbinfo)
5669 +AuStubVoid(dbgaufs_fin, void)
5670 +AuStubInt0(__init dbgaufs_init, void)
5671 +#endif /* CONFIG_DEBUG_FS */
5672 +
5673 +#endif /* __KERNEL__ */
5674 +#endif /* __DBGAUFS_H__ */
5675 diff -urN /usr/share/empty/fs/aufs/dcsub.c linux/fs/aufs/dcsub.c
5676 --- /usr/share/empty/fs/aufs/dcsub.c    1970-01-01 01:00:00.000000000 +0100
5677 +++ linux/fs/aufs/dcsub.c       2014-04-24 22:11:10.848602379 +0200
5678 @@ -0,0 +1,243 @@
5679 +/*
5680 + * Copyright (C) 2005-2014 Junjiro R. Okajima
5681 + *
5682 + * This program, aufs is free software; you can redistribute it and/or modify
5683 + * it under the terms of the GNU General Public License as published by
5684 + * the Free Software Foundation; either version 2 of the License, or
5685 + * (at your option) any later version.
5686 + *
5687 + * This program is distributed in the hope that it will be useful,
5688 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5689 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5690 + * GNU General Public License for more details.
5691 + *
5692 + * You should have received a copy of the GNU General Public License
5693 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5694 + */
5695 +
5696 +/*
5697 + * sub-routines for dentry cache
5698 + */
5699 +
5700 +#include "aufs.h"
5701 +
5702 +static void au_dpage_free(struct au_dpage *dpage)
5703 +{
5704 +       int i;
5705 +       struct dentry **p;
5706 +
5707 +       p = dpage->dentries;
5708 +       for (i = 0; i < dpage->ndentry; i++)
5709 +               dput(*p++);
5710 +       free_page((unsigned long)dpage->dentries);
5711 +}
5712 +
5713 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp)
5714 +{
5715 +       int err;
5716 +       void *p;
5717 +
5718 +       err = -ENOMEM;
5719 +       dpages->dpages = kmalloc(sizeof(*dpages->dpages), gfp);
5720 +       if (unlikely(!dpages->dpages))
5721 +               goto out;
5722 +
5723 +       p = (void *)__get_free_page(gfp);
5724 +       if (unlikely(!p))
5725 +               goto out_dpages;
5726 +
5727 +       dpages->dpages[0].ndentry = 0;
5728 +       dpages->dpages[0].dentries = p;
5729 +       dpages->ndpage = 1;
5730 +       return 0; /* success */
5731 +
5732 +out_dpages:
5733 +       kfree(dpages->dpages);
5734 +out:
5735 +       return err;
5736 +}
5737 +
5738 +void au_dpages_free(struct au_dcsub_pages *dpages)
5739 +{
5740 +       int i;
5741 +       struct au_dpage *p;
5742 +
5743 +       p = dpages->dpages;
5744 +       for (i = 0; i < dpages->ndpage; i++)
5745 +               au_dpage_free(p++);
5746 +       kfree(dpages->dpages);
5747 +}
5748 +
5749 +static int au_dpages_append(struct au_dcsub_pages *dpages,
5750 +                           struct dentry *dentry, gfp_t gfp)
5751 +{
5752 +       int err, sz;
5753 +       struct au_dpage *dpage;
5754 +       void *p;
5755 +
5756 +       dpage = dpages->dpages + dpages->ndpage - 1;
5757 +       sz = PAGE_SIZE / sizeof(dentry);
5758 +       if (unlikely(dpage->ndentry >= sz)) {
5759 +               AuLabel(new dpage);
5760 +               err = -ENOMEM;
5761 +               sz = dpages->ndpage * sizeof(*dpages->dpages);
5762 +               p = au_kzrealloc(dpages->dpages, sz,
5763 +                                sz + sizeof(*dpages->dpages), gfp);
5764 +               if (unlikely(!p))
5765 +                       goto out;
5766 +
5767 +               dpages->dpages = p;
5768 +               dpage = dpages->dpages + dpages->ndpage;
5769 +               p = (void *)__get_free_page(gfp);
5770 +               if (unlikely(!p))
5771 +                       goto out;
5772 +
5773 +               dpage->ndentry = 0;
5774 +               dpage->dentries = p;
5775 +               dpages->ndpage++;
5776 +       }
5777 +
5778 +       AuDebugOn(!d_count(dentry));
5779 +       dpage->dentries[dpage->ndentry++] = dget_dlock(dentry);
5780 +       return 0; /* success */
5781 +
5782 +out:
5783 +       return err;
5784 +}
5785 +
5786 +/* try d_walk() in linux/fs/dcache.c */
5787 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
5788 +                  au_dpages_test test, void *arg)
5789 +{
5790 +       int err;
5791 +       struct dentry *this_parent;
5792 +       struct list_head *next;
5793 +       struct super_block *sb = root->d_sb;
5794 +
5795 +       err = 0;
5796 +       write_seqlock(&rename_lock);
5797 +       this_parent = root;
5798 +       spin_lock(&this_parent->d_lock);
5799 +repeat:
5800 +       next = this_parent->d_subdirs.next;
5801 +resume:
5802 +       if (this_parent->d_sb == sb
5803 +           && !IS_ROOT(this_parent)
5804 +           && au_di(this_parent)
5805 +           && d_count(this_parent)
5806 +           && (!test || test(this_parent, arg))) {
5807 +               err = au_dpages_append(dpages, this_parent, GFP_ATOMIC);
5808 +               if (unlikely(err))
5809 +                       goto out;
5810 +       }
5811 +
5812 +       while (next != &this_parent->d_subdirs) {
5813 +               struct list_head *tmp = next;
5814 +               struct dentry *dentry = list_entry(tmp, struct dentry,
5815 +                                                  d_child);
5816 +
5817 +               next = tmp->next;
5818 +               spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
5819 +               if (d_count(dentry)) {
5820 +                       if (!list_empty(&dentry->d_subdirs)) {
5821 +                               spin_unlock(&this_parent->d_lock);
5822 +                               spin_release(&dentry->d_lock.dep_map, 1,
5823 +                                            _RET_IP_);
5824 +                               this_parent = dentry;
5825 +                               spin_acquire(&this_parent->d_lock.dep_map, 0, 1,
5826 +                                            _RET_IP_);
5827 +                               goto repeat;
5828 +                       }
5829 +                       if (dentry->d_sb == sb
5830 +                           && au_di(dentry)
5831 +                           && (!test || test(dentry, arg)))
5832 +                               err = au_dpages_append(dpages, dentry,
5833 +                                                      GFP_ATOMIC);
5834 +               }
5835 +               spin_unlock(&dentry->d_lock);
5836 +               if (unlikely(err))
5837 +                       goto out;
5838 +       }
5839 +
5840 +       if (this_parent != root) {
5841 +               struct dentry *tmp;
5842 +               struct dentry *child;
5843 +
5844 +               tmp = this_parent->d_parent;
5845 +               rcu_read_lock();
5846 +               spin_unlock(&this_parent->d_lock);
5847 +               child = this_parent;
5848 +               this_parent = tmp;
5849 +               spin_lock(&this_parent->d_lock);
5850 +               rcu_read_unlock();
5851 +               next = child->d_child.next;
5852 +               goto resume;
5853 +       }
5854 +
5855 +out:
5856 +       spin_unlock(&this_parent->d_lock);
5857 +       write_sequnlock(&rename_lock);
5858 +       return err;
5859 +}
5860 +
5861 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
5862 +                      int do_include, au_dpages_test test, void *arg)
5863 +{
5864 +       int err;
5865 +
5866 +       err = 0;
5867 +       write_seqlock(&rename_lock);
5868 +       spin_lock(&dentry->d_lock);
5869 +       if (do_include
5870 +           && d_count(dentry)
5871 +           && (!test || test(dentry, arg)))
5872 +               err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
5873 +       spin_unlock(&dentry->d_lock);
5874 +       if (unlikely(err))
5875 +               goto out;
5876 +
5877 +       /*
5878 +        * RCU for vfsmount is unnecessary since this is a traverse in a single
5879 +        * mount
5880 +        */
5881 +       while (!IS_ROOT(dentry)) {
5882 +               dentry = dentry->d_parent; /* rename_lock is locked */
5883 +               spin_lock(&dentry->d_lock);
5884 +               if (d_count(dentry)
5885 +                   && (!test || test(dentry, arg)))
5886 +                       err = au_dpages_append(dpages, dentry, GFP_ATOMIC);
5887 +               spin_unlock(&dentry->d_lock);
5888 +               if (unlikely(err))
5889 +                       break;
5890 +       }
5891 +
5892 +out:
5893 +       write_sequnlock(&rename_lock);
5894 +       return err;
5895 +}
5896 +
5897 +static inline int au_dcsub_dpages_aufs(struct dentry *dentry, void *arg)
5898 +{
5899 +       return au_di(dentry) && dentry->d_sb == arg;
5900 +}
5901 +
5902 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
5903 +                           struct dentry *dentry, int do_include)
5904 +{
5905 +       return au_dcsub_pages_rev(dpages, dentry, do_include,
5906 +                                 au_dcsub_dpages_aufs, dentry->d_sb);
5907 +}
5908 +
5909 +int au_test_subdir(struct dentry *d1, struct dentry *d2)
5910 +{
5911 +       struct path path[2] = {
5912 +               {
5913 +                       .dentry = d1
5914 +               },
5915 +               {
5916 +                       .dentry = d2
5917 +               }
5918 +       };
5919 +
5920 +       return path_is_under(path + 0, path + 1);
5921 +}
5922 diff -urN /usr/share/empty/fs/aufs/dcsub.h linux/fs/aufs/dcsub.h
5923 --- /usr/share/empty/fs/aufs/dcsub.h    1970-01-01 01:00:00.000000000 +0100
5924 +++ linux/fs/aufs/dcsub.h       2014-04-24 22:11:10.848602379 +0200
5925 @@ -0,0 +1,98 @@
5926 +/*
5927 + * Copyright (C) 2005-2014 Junjiro R. Okajima
5928 + *
5929 + * This program, aufs is free software; you can redistribute it and/or modify
5930 + * it under the terms of the GNU General Public License as published by
5931 + * the Free Software Foundation; either version 2 of the License, or
5932 + * (at your option) any later version.
5933 + *
5934 + * This program is distributed in the hope that it will be useful,
5935 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
5936 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
5937 + * GNU General Public License for more details.
5938 + *
5939 + * You should have received a copy of the GNU General Public License
5940 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
5941 + */
5942 +
5943 +/*
5944 + * sub-routines for dentry cache
5945 + */
5946 +
5947 +#ifndef __AUFS_DCSUB_H__
5948 +#define __AUFS_DCSUB_H__
5949 +
5950 +#ifdef __KERNEL__
5951 +
5952 +#include <linux/dcache.h>
5953 +#include <linux/fs.h>
5954 +
5955 +struct dentry;
5956 +
5957 +struct au_dpage {
5958 +       int ndentry;
5959 +       struct dentry **dentries;
5960 +};
5961 +
5962 +struct au_dcsub_pages {
5963 +       int ndpage;
5964 +       struct au_dpage *dpages;
5965 +};
5966 +
5967 +/* ---------------------------------------------------------------------- */
5968 +
5969 +/* dcsub.c */
5970 +int au_dpages_init(struct au_dcsub_pages *dpages, gfp_t gfp);
5971 +void au_dpages_free(struct au_dcsub_pages *dpages);
5972 +typedef int (*au_dpages_test)(struct dentry *dentry, void *arg);
5973 +int au_dcsub_pages(struct au_dcsub_pages *dpages, struct dentry *root,
5974 +                  au_dpages_test test, void *arg);
5975 +int au_dcsub_pages_rev(struct au_dcsub_pages *dpages, struct dentry *dentry,
5976 +                      int do_include, au_dpages_test test, void *arg);
5977 +int au_dcsub_pages_rev_aufs(struct au_dcsub_pages *dpages,
5978 +                           struct dentry *dentry, int do_include);
5979 +int au_test_subdir(struct dentry *d1, struct dentry *d2);
5980 +
5981 +/* ---------------------------------------------------------------------- */
5982 +
5983 +/*
5984 + * todo: in linux-3.13, several similar (but faster) helpers are added to
5985 + * include/linux/dcache.h. Try them (in the future).
5986 + */
5987 +
5988 +static inline int au_d_hashed_positive(struct dentry *d)
5989 +{
5990 +       int err;
5991 +       struct inode *inode = d->d_inode;
5992 +       err = 0;
5993 +       if (unlikely(d_unhashed(d) || !inode || !inode->i_nlink))
5994 +               err = -ENOENT;
5995 +       return err;
5996 +}
5997 +
5998 +static inline int au_d_alive(struct dentry *d)
5999 +{
6000 +       int err;
6001 +       struct inode *inode;
6002 +       err = 0;
6003 +       if (!IS_ROOT(d))
6004 +               err = au_d_hashed_positive(d);
6005 +       else {
6006 +               inode = d->d_inode;
6007 +               if (unlikely(d_unlinked(d) || !inode || !inode->i_nlink))
6008 +                       err = -ENOENT;
6009 +       }
6010 +       return err;
6011 +}
6012 +
6013 +static inline int au_alive_dir(struct dentry *d)
6014 +{
6015 +       int err;
6016 +       err = au_d_alive(d);
6017 +       if (unlikely(err || IS_DEADDIR(d->d_inode)))
6018 +               err = -ENOENT;
6019 +       return err;
6020 +}
6021 +
6022 +#endif /* __KERNEL__ */
6023 +#endif /* __AUFS_DCSUB_H__ */
6024 diff -urN /usr/share/empty/fs/aufs/debug.c linux/fs/aufs/debug.c
6025 --- /usr/share/empty/fs/aufs/debug.c    1970-01-01 01:00:00.000000000 +0100
6026 +++ linux/fs/aufs/debug.c       2014-04-24 22:11:10.848602379 +0200
6027 @@ -0,0 +1,517 @@
6028 +/*
6029 + * Copyright (C) 2005-2014 Junjiro R. Okajima
6030 + *
6031 + * This program, aufs is free software; you can redistribute it and/or modify
6032 + * it under the terms of the GNU General Public License as published by
6033 + * the Free Software Foundation; either version 2 of the License, or
6034 + * (at your option) any later version.
6035 + *
6036 + * This program is distributed in the hope that it will be useful,
6037 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6038 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6039 + * GNU General Public License for more details.
6040 + *
6041 + * You should have received a copy of the GNU General Public License
6042 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6043 + */
6044 +
6045 +/*
6046 + * debug print functions
6047 + */
6048 +
6049 +#include <linux/vt_kern.h>
6050 +#include "aufs.h"
6051 +
6052 +/* Returns 0, or -errno.  arg is in kp->arg. */
6053 +static int param_atomic_t_set(const char *val, const struct kernel_param *kp)
6054 +{
6055 +       int err, n;
6056 +
6057 +       err = kstrtoint(val, 0, &n);
6058 +       if (!err) {
6059 +               if (n > 0)
6060 +                       au_debug_on();
6061 +               else
6062 +                       au_debug_off();
6063 +       }
6064 +       return err;
6065 +}
6066 +
6067 +/* Returns length written or -errno.  Buffer is 4k (ie. be short!) */
6068 +static int param_atomic_t_get(char *buffer, const struct kernel_param *kp)
6069 +{
6070 +       atomic_t *a;
6071 +
6072 +       a = kp->arg;
6073 +       return sprintf(buffer, "%d", atomic_read(a));
6074 +}
6075 +
6076 +static struct kernel_param_ops param_ops_atomic_t = {
6077 +       .set = param_atomic_t_set,
6078 +       .get = param_atomic_t_get
6079 +       /* void (*free)(void *arg) */
6080 +};
6081 +
6082 +atomic_t aufs_debug = ATOMIC_INIT(0);
6083 +MODULE_PARM_DESC(debug, "debug print");
6084 +module_param_named(debug, aufs_debug, atomic_t, S_IRUGO | S_IWUSR | S_IWGRP);
6085 +
6086 +char *au_plevel = KERN_DEBUG;
6087 +#define dpri(fmt, ...) do {                                    \
6088 +       if ((au_plevel                                          \
6089 +            && strcmp(au_plevel, KERN_DEBUG))                  \
6090 +           || au_debug_test())                                 \
6091 +               printk("%s" fmt, au_plevel, ##__VA_ARGS__);     \
6092 +} while (0)
6093 +
6094 +/* ---------------------------------------------------------------------- */
6095 +
6096 +void au_dpri_whlist(struct au_nhash *whlist)
6097 +{
6098 +       unsigned long ul, n;
6099 +       struct hlist_head *head;
6100 +       struct au_vdir_wh *pos;
6101 +
6102 +       n = whlist->nh_num;
6103 +       head = whlist->nh_head;
6104 +       for (ul = 0; ul < n; ul++) {
6105 +               hlist_for_each_entry(pos, head, wh_hash)
6106 +                       dpri("b%d, %.*s, %d\n",
6107 +                            pos->wh_bindex,
6108 +                            pos->wh_str.len, pos->wh_str.name,
6109 +                            pos->wh_str.len);
6110 +               head++;
6111 +       }
6112 +}
6113 +
6114 +void au_dpri_vdir(struct au_vdir *vdir)
6115 +{
6116 +       unsigned long ul;
6117 +       union au_vdir_deblk_p p;
6118 +       unsigned char *o;
6119 +
6120 +       if (!vdir || IS_ERR(vdir)) {
6121 +               dpri("err %ld\n", PTR_ERR(vdir));
6122 +               return;
6123 +       }
6124 +
6125 +       dpri("deblk %u, nblk %lu, deblk %p, last{%lu, %p}, ver %lu\n",
6126 +            vdir->vd_deblk_sz, vdir->vd_nblk, vdir->vd_deblk,
6127 +            vdir->vd_last.ul, vdir->vd_last.p.deblk, vdir->vd_version);
6128 +       for (ul = 0; ul < vdir->vd_nblk; ul++) {
6129 +               p.deblk = vdir->vd_deblk[ul];
6130 +               o = p.deblk;
6131 +               dpri("[%lu]: %p\n", ul, o);
6132 +       }
6133 +}
6134 +
6135 +static int do_pri_inode(aufs_bindex_t bindex, struct inode *inode, int hn,
6136 +                       struct dentry *wh)
6137 +{
6138 +       char *n = NULL;
6139 +       int l = 0;
6140 +
6141 +       if (!inode || IS_ERR(inode)) {
6142 +               dpri("i%d: err %ld\n", bindex, PTR_ERR(inode));
6143 +               return -1;
6144 +       }
6145 +
6146 +       /* the type of i_blocks depends upon CONFIG_LBDAF */
6147 +       BUILD_BUG_ON(sizeof(inode->i_blocks) != sizeof(unsigned long)
6148 +                    && sizeof(inode->i_blocks) != sizeof(u64));
6149 +       if (wh) {
6150 +               n = (void *)wh->d_name.name;
6151 +               l = wh->d_name.len;
6152 +       }
6153 +
6154 +       dpri("i%d: %p, i%lu, %s, cnt %d, nl %u, 0%o, sz %llu, blk %llu,"
6155 +            " hn %d, ct %lld, np %lu, st 0x%lx, f 0x%x, v %llu, g %x%s%.*s\n",
6156 +            bindex, inode,
6157 +            inode->i_ino, inode->i_sb ? au_sbtype(inode->i_sb) : "??",
6158 +            atomic_read(&inode->i_count), inode->i_nlink, inode->i_mode,
6159 +            i_size_read(inode), (unsigned long long)inode->i_blocks,
6160 +            hn, (long long)timespec_to_ns(&inode->i_ctime) & 0x0ffff,
6161 +            inode->i_mapping ? inode->i_mapping->nrpages : 0,
6162 +            inode->i_state, inode->i_flags, inode->i_version,
6163 +            inode->i_generation,
6164 +            l ? ", wh " : "", l, n);
6165 +       return 0;
6166 +}
6167 +
6168 +void au_dpri_inode(struct inode *inode)
6169 +{
6170 +       struct au_iinfo *iinfo;
6171 +       aufs_bindex_t bindex;
6172 +       int err, hn;
6173 +
6174 +       err = do_pri_inode(-1, inode, -1, NULL);
6175 +       if (err || !au_test_aufs(inode->i_sb))
6176 +               return;
6177 +
6178 +       iinfo = au_ii(inode);
6179 +       if (!iinfo)
6180 +               return;
6181 +       dpri("i-1: bstart %d, bend %d, gen %d\n",
6182 +            iinfo->ii_bstart, iinfo->ii_bend, au_iigen(inode, NULL));
6183 +       if (iinfo->ii_bstart < 0)
6184 +               return;
6185 +       hn = 0;
6186 +       for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend; bindex++) {
6187 +               hn = !!au_hn(iinfo->ii_hinode + bindex);
6188 +               do_pri_inode(bindex, iinfo->ii_hinode[0 + bindex].hi_inode, hn,
6189 +                            iinfo->ii_hinode[0 + bindex].hi_whdentry);
6190 +       }
6191 +}
6192 +
6193 +void au_dpri_dalias(struct inode *inode)
6194 +{
6195 +       struct dentry *d;
6196 +
6197 +       spin_lock(&inode->i_lock);
6198 +       hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias)
6199 +               au_dpri_dentry(d);
6200 +       spin_unlock(&inode->i_lock);
6201 +}
6202 +
6203 +static int do_pri_dentry(aufs_bindex_t bindex, struct dentry *dentry)
6204 +{
6205 +       struct dentry *wh = NULL;
6206 +       int hn;
6207 +
6208 +       if (!dentry || IS_ERR(dentry)) {
6209 +               dpri("d%d: err %ld\n", bindex, PTR_ERR(dentry));
6210 +               return -1;
6211 +       }
6212 +       /* do not call dget_parent() here */
6213 +       /* note: access d_xxx without d_lock */
6214 +       dpri("d%d: %p, %pd2?, %s, cnt %d, flags 0x%x, %shashed\n",
6215 +            bindex, dentry, dentry,
6216 +            dentry->d_sb ? au_sbtype(dentry->d_sb) : "??",
6217 +            d_count(dentry), dentry->d_flags,
6218 +            d_unhashed(dentry) ? "un" : "");
6219 +       hn = -1;
6220 +       if (bindex >= 0 && dentry->d_inode && au_test_aufs(dentry->d_sb)) {
6221 +               struct au_iinfo *iinfo = au_ii(dentry->d_inode);
6222 +               if (iinfo) {
6223 +                       hn = !!au_hn(iinfo->ii_hinode + bindex);
6224 +                       wh = iinfo->ii_hinode[0 + bindex].hi_whdentry;
6225 +               }
6226 +       }
6227 +       do_pri_inode(bindex, dentry->d_inode, hn, wh);
6228 +       return 0;
6229 +}
6230 +
6231 +void au_dpri_dentry(struct dentry *dentry)
6232 +{
6233 +       struct au_dinfo *dinfo;
6234 +       aufs_bindex_t bindex;
6235 +       int err;
6236 +       struct au_hdentry *hdp;
6237 +
6238 +       err = do_pri_dentry(-1, dentry);
6239 +       if (err || !au_test_aufs(dentry->d_sb))
6240 +               return;
6241 +
6242 +       dinfo = au_di(dentry);
6243 +       if (!dinfo)
6244 +               return;
6245 +       dpri("d-1: bstart %d, bend %d, bwh %d, bdiropq %d, gen %d\n",
6246 +            dinfo->di_bstart, dinfo->di_bend,
6247 +            dinfo->di_bwh, dinfo->di_bdiropq, au_digen(dentry));
6248 +       if (dinfo->di_bstart < 0)
6249 +               return;
6250 +       hdp = dinfo->di_hdentry;
6251 +       for (bindex = dinfo->di_bstart; bindex <= dinfo->di_bend; bindex++)
6252 +               do_pri_dentry(bindex, hdp[0 + bindex].hd_dentry);
6253 +}
6254 +
6255 +static int do_pri_file(aufs_bindex_t bindex, struct file *file)
6256 +{
6257 +       char a[32];
6258 +
6259 +       if (!file || IS_ERR(file)) {
6260 +               dpri("f%d: err %ld\n", bindex, PTR_ERR(file));
6261 +               return -1;
6262 +       }
6263 +       a[0] = 0;
6264 +       if (bindex < 0
6265 +           && file->f_dentry
6266 +           && au_test_aufs(file->f_dentry->d_sb)
6267 +           && au_fi(file))
6268 +               snprintf(a, sizeof(a), ", gen %d, mmapped %d",
6269 +                        au_figen(file), atomic_read(&au_fi(file)->fi_mmapped));
6270 +       dpri("f%d: mode 0x%x, flags 0%o, cnt %ld, v %llu, pos %llu%s\n",
6271 +            bindex, file->f_mode, file->f_flags, (long)file_count(file),
6272 +            file->f_version, file->f_pos, a);
6273 +       if (file->f_dentry)
6274 +               do_pri_dentry(bindex, file->f_dentry);
6275 +       return 0;
6276 +}
6277 +
6278 +void au_dpri_file(struct file *file)
6279 +{
6280 +       struct au_finfo *finfo;
6281 +       struct au_fidir *fidir;
6282 +       struct au_hfile *hfile;
6283 +       aufs_bindex_t bindex;
6284 +       int err;
6285 +
6286 +       err = do_pri_file(-1, file);
6287 +       if (err || !file->f_dentry || !au_test_aufs(file->f_dentry->d_sb))
6288 +               return;
6289 +
6290 +       finfo = au_fi(file);
6291 +       if (!finfo)
6292 +               return;
6293 +       if (finfo->fi_btop < 0)
6294 +               return;
6295 +       fidir = finfo->fi_hdir;
6296 +       if (!fidir)
6297 +               do_pri_file(finfo->fi_btop, finfo->fi_htop.hf_file);
6298 +       else
6299 +               for (bindex = finfo->fi_btop;
6300 +                    bindex >= 0 && bindex <= fidir->fd_bbot;
6301 +                    bindex++) {
6302 +                       hfile = fidir->fd_hfile + bindex;
6303 +                       do_pri_file(bindex, hfile ? hfile->hf_file : NULL);
6304 +               }
6305 +}
6306 +
6307 +static int do_pri_br(aufs_bindex_t bindex, struct au_branch *br)
6308 +{
6309 +       struct vfsmount *mnt;
6310 +       struct super_block *sb;
6311 +
6312 +       if (!br || IS_ERR(br))
6313 +               goto out;
6314 +       mnt = au_br_mnt(br);
6315 +       if (!mnt || IS_ERR(mnt))
6316 +               goto out;
6317 +       sb = mnt->mnt_sb;
6318 +       if (!sb || IS_ERR(sb))
6319 +               goto out;
6320 +
6321 +       dpri("s%d: {perm 0x%x, id %d, cnt %d, wbr %p}, "
6322 +            "%s, dev 0x%02x%02x, flags 0x%lx, cnt %d, active %d, "
6323 +            "xino %d\n",
6324 +            bindex, br->br_perm, br->br_id, atomic_read(&br->br_count),
6325 +            br->br_wbr, au_sbtype(sb), MAJOR(sb->s_dev), MINOR(sb->s_dev),
6326 +            sb->s_flags, sb->s_count,
6327 +            atomic_read(&sb->s_active), !!br->br_xino.xi_file);
6328 +       return 0;
6329 +
6330 +out:
6331 +       dpri("s%d: err %ld\n", bindex, PTR_ERR(br));
6332 +       return -1;
6333 +}
6334 +
6335 +void au_dpri_sb(struct super_block *sb)
6336 +{
6337 +       struct au_sbinfo *sbinfo;
6338 +       aufs_bindex_t bindex;
6339 +       int err;
6340 +       /* to reuduce stack size */
6341 +       struct {
6342 +               struct vfsmount mnt;
6343 +               struct au_branch fake;
6344 +       } *a;
6345 +
6346 +       /* this function can be called from magic sysrq */
6347 +       a = kzalloc(sizeof(*a), GFP_ATOMIC);
6348 +       if (unlikely(!a)) {
6349 +               dpri("no memory\n");
6350 +               return;
6351 +       }
6352 +
6353 +       a->mnt.mnt_sb = sb;
6354 +       a->fake.br_perm = 0;
6355 +       a->fake.br_path.mnt = &a->mnt;
6356 +       a->fake.br_xino.xi_file = NULL;
6357 +       atomic_set(&a->fake.br_count, 0);
6358 +       smp_mb(); /* atomic_set */
6359 +       err = do_pri_br(-1, &a->fake);
6360 +       kfree(a);
6361 +       dpri("dev 0x%x\n", sb->s_dev);
6362 +       if (err || !au_test_aufs(sb))
6363 +               return;
6364 +
6365 +       sbinfo = au_sbi(sb);
6366 +       if (!sbinfo)
6367 +               return;
6368 +       dpri("nw %d, gen %u, kobj %d\n",
6369 +            atomic_read(&sbinfo->si_nowait.nw_len), sbinfo->si_generation,
6370 +            atomic_read(&sbinfo->si_kobj.kref.refcount));
6371 +       for (bindex = 0; bindex <= sbinfo->si_bend; bindex++)
6372 +               do_pri_br(bindex, sbinfo->si_branch[0 + bindex]);
6373 +}
6374 +
6375 +/* ---------------------------------------------------------------------- */
6376 +
6377 +void au_dbg_sleep_jiffy(int jiffy)
6378 +{
6379 +       while (jiffy)
6380 +               jiffy = schedule_timeout_uninterruptible(jiffy);
6381 +}
6382 +
6383 +void au_dbg_iattr(struct iattr *ia)
6384 +{
6385 +#define AuBit(name)                                    \
6386 +       do {                                            \
6387 +               if (ia->ia_valid & ATTR_ ## name)       \
6388 +                       dpri(#name "\n");               \
6389 +       } while (0)
6390 +       AuBit(MODE);
6391 +       AuBit(UID);
6392 +       AuBit(GID);
6393 +       AuBit(SIZE);
6394 +       AuBit(ATIME);
6395 +       AuBit(MTIME);
6396 +       AuBit(CTIME);
6397 +       AuBit(ATIME_SET);
6398 +       AuBit(MTIME_SET);
6399 +       AuBit(FORCE);
6400 +       AuBit(ATTR_FLAG);
6401 +       AuBit(KILL_SUID);
6402 +       AuBit(KILL_SGID);
6403 +       AuBit(FILE);
6404 +       AuBit(KILL_PRIV);
6405 +       AuBit(OPEN);
6406 +       AuBit(TIMES_SET);
6407 +#undef AuBit
6408 +       dpri("ia_file %p\n", ia->ia_file);
6409 +}
6410 +
6411 +/* ---------------------------------------------------------------------- */
6412 +
6413 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line)
6414 +{
6415 +       struct inode *h_inode, *inode = dentry->d_inode;
6416 +       struct dentry *h_dentry;
6417 +       aufs_bindex_t bindex, bend, bi;
6418 +
6419 +       if (!inode /* || au_di(dentry)->di_lsc == AuLsc_DI_TMP */)
6420 +               return;
6421 +
6422 +       bend = au_dbend(dentry);
6423 +       bi = au_ibend(inode);
6424 +       if (bi < bend)
6425 +               bend = bi;
6426 +       bindex = au_dbstart(dentry);
6427 +       bi = au_ibstart(inode);
6428 +       if (bi > bindex)
6429 +               bindex = bi;
6430 +
6431 +       for (; bindex <= bend; bindex++) {
6432 +               h_dentry = au_h_dptr(dentry, bindex);
6433 +               if (!h_dentry)
6434 +                       continue;
6435 +               h_inode = au_h_iptr(inode, bindex);
6436 +               if (unlikely(h_inode != h_dentry->d_inode)) {
6437 +                       au_debug_on();
6438 +                       AuDbg("b%d, %s:%d\n", bindex, func, line);
6439 +                       AuDbgDentry(dentry);
6440 +                       AuDbgInode(inode);
6441 +                       au_debug_off();
6442 +                       BUG();
6443 +               }
6444 +       }
6445 +}
6446 +
6447 +void au_dbg_verify_dir_parent(struct dentry *dentry, unsigned int sigen)
6448 +{
6449 +       struct dentry *parent;
6450 +
6451 +       parent = dget_parent(dentry);
6452 +       AuDebugOn(!S_ISDIR(dentry->d_inode->i_mode));
6453 +       AuDebugOn(IS_ROOT(dentry));
6454 +       AuDebugOn(au_digen_test(parent, sigen));
6455 +       dput(parent);
6456 +}
6457 +
6458 +void au_dbg_verify_nondir_parent(struct dentry *dentry, unsigned int sigen)
6459 +{
6460 +       struct dentry *parent;
6461 +       struct inode *inode;
6462 +
6463 +       parent = dget_parent(dentry);
6464 +       inode = dentry->d_inode;
6465 +       AuDebugOn(inode && S_ISDIR(dentry->d_inode->i_mode));
6466 +       AuDebugOn(au_digen_test(parent, sigen));
6467 +       dput(parent);
6468 +}
6469 +
6470 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen)
6471 +{
6472 +       int err, i, j;
6473 +       struct au_dcsub_pages dpages;
6474 +       struct au_dpage *dpage;
6475 +       struct dentry **dentries;
6476 +
6477 +       err = au_dpages_init(&dpages, GFP_NOFS);
6478 +       AuDebugOn(err);
6479 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/1);
6480 +       AuDebugOn(err);
6481 +       for (i = dpages.ndpage - 1; !err && i >= 0; i--) {
6482 +               dpage = dpages.dpages + i;
6483 +               dentries = dpage->dentries;
6484 +               for (j = dpage->ndentry - 1; !err && j >= 0; j--)
6485 +                       AuDebugOn(au_digen_test(dentries[j], sigen));
6486 +       }
6487 +       au_dpages_free(&dpages);
6488 +}
6489 +
6490 +void au_dbg_verify_kthread(void)
6491 +{
6492 +       if (au_wkq_test()) {
6493 +               au_dbg_blocked();
6494 +               /*
6495 +                * It may be recursive, but udba=notify between two aufs mounts,
6496 +                * where a single ro branch is shared, is not a problem.
6497 +                */
6498 +               /* WARN_ON(1); */
6499 +       }
6500 +}
6501 +
6502 +/* ---------------------------------------------------------------------- */
6503 +
6504 +void au_debug_sbinfo_init(struct au_sbinfo *sbinfo __maybe_unused)
6505 +{
6506 +#ifdef AuForceNoPlink
6507 +       au_opt_clr(sbinfo->si_mntflags, PLINK);
6508 +#endif
6509 +#ifdef AuForceNoXino
6510 +       au_opt_clr(sbinfo->si_mntflags, XINO);
6511 +#endif
6512 +#ifdef AuForceNoRefrof
6513 +       au_opt_clr(sbinfo->si_mntflags, REFROF);
6514 +#endif
6515 +#ifdef AuForceHnotify
6516 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_HNOTIFY);
6517 +#endif
6518 +#ifdef AuForceRd0
6519 +       sbinfo->si_rdblk = 0;
6520 +       sbinfo->si_rdhash = 0;
6521 +#endif
6522 +}
6523 +
6524 +int __init au_debug_init(void)
6525 +{
6526 +       aufs_bindex_t bindex;
6527 +       struct au_vdir_destr destr;
6528 +
6529 +       bindex = -1;
6530 +       AuDebugOn(bindex >= 0);
6531 +
6532 +       destr.len = -1;
6533 +       AuDebugOn(destr.len < NAME_MAX);
6534 +
6535 +#ifdef CONFIG_4KSTACKS
6536 +       pr_warn("CONFIG_4KSTACKS is defined.\n");
6537 +#endif
6538 +
6539 +#ifdef AuForceNoBrs
6540 +       sysaufs_brs = 0;
6541 +#endif
6542 +
6543 +       return 0;
6544 +}
6545 diff -urN /usr/share/empty/fs/aufs/debug.h linux/fs/aufs/debug.h
6546 --- /usr/share/empty/fs/aufs/debug.h    1970-01-01 01:00:00.000000000 +0100
6547 +++ linux/fs/aufs/debug.h       2014-04-24 22:11:10.848602379 +0200
6548 @@ -0,0 +1,247 @@
6549 +/*
6550 + * Copyright (C) 2005-2014 Junjiro R. Okajima
6551 + *
6552 + * This program, aufs is free software; you can redistribute it and/or modify
6553 + * it under the terms of the GNU General Public License as published by
6554 + * the Free Software Foundation; either version 2 of the License, or
6555 + * (at your option) any later version.
6556 + *
6557 + * This program is distributed in the hope that it will be useful,
6558 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6559 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6560 + * GNU General Public License for more details.
6561 + *
6562 + * You should have received a copy of the GNU General Public License
6563 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6564 + */
6565 +
6566 +/*
6567 + * debug print functions
6568 + */
6569 +
6570 +#ifndef __AUFS_DEBUG_H__
6571 +#define __AUFS_DEBUG_H__
6572 +
6573 +#ifdef __KERNEL__
6574 +
6575 +#include <linux/atomic.h>
6576 +#include <linux/module.h>
6577 +#include <linux/kallsyms.h>
6578 +#include <linux/sysrq.h>
6579 +
6580 +#ifdef CONFIG_AUFS_DEBUG
6581 +#define AuDebugOn(a)           BUG_ON(a)
6582 +
6583 +/* module parameter */
6584 +extern atomic_t aufs_debug;
6585 +static inline void au_debug_on(void)
6586 +{
6587 +       atomic_inc(&aufs_debug);
6588 +}
6589 +static inline void au_debug_off(void)
6590 +{
6591 +       atomic_dec_if_positive(&aufs_debug);
6592 +}
6593 +
6594 +static inline int au_debug_test(void)
6595 +{
6596 +       return atomic_read(&aufs_debug) > 0;
6597 +}
6598 +#else
6599 +#define AuDebugOn(a)           do {} while (0)
6600 +AuStubVoid(au_debug_on, void)
6601 +AuStubVoid(au_debug_off, void)
6602 +AuStubInt0(au_debug_test, void)
6603 +#endif /* CONFIG_AUFS_DEBUG */
6604 +
6605 +#define param_check_atomic_t(name, p) __param_check(name, p, atomic_t)
6606 +
6607 +/* ---------------------------------------------------------------------- */
6608 +
6609 +/* debug print */
6610 +
6611 +#define AuDbg(fmt, ...) do { \
6612 +       if (au_debug_test()) \
6613 +               pr_debug("DEBUG: " fmt, ##__VA_ARGS__); \
6614 +} while (0)
6615 +#define AuLabel(l)             AuDbg(#l "\n")
6616 +#define AuIOErr(fmt, ...)      pr_err("I/O Error, " fmt, ##__VA_ARGS__)
6617 +#define AuWarn1(fmt, ...) do { \
6618 +       static unsigned char _c; \
6619 +       if (!_c++) \
6620 +               pr_warn(fmt, ##__VA_ARGS__); \
6621 +} while (0)
6622 +
6623 +#define AuErr1(fmt, ...) do { \
6624 +       static unsigned char _c; \
6625 +       if (!_c++) \
6626 +               pr_err(fmt, ##__VA_ARGS__); \
6627 +} while (0)
6628 +
6629 +#define AuIOErr1(fmt, ...) do { \
6630 +       static unsigned char _c; \
6631 +       if (!_c++) \
6632 +               AuIOErr(fmt, ##__VA_ARGS__); \
6633 +} while (0)
6634 +
6635 +#define AuUnsupportMsg "This operation is not supported." \
6636 +                       " Please report this application to aufs-users ML."
6637 +#define AuUnsupport(fmt, ...) do { \
6638 +       pr_err(AuUnsupportMsg "\n" fmt, ##__VA_ARGS__); \
6639 +       dump_stack(); \
6640 +} while (0)
6641 +
6642 +#define AuTraceErr(e) do { \
6643 +       if (unlikely((e) < 0)) \
6644 +               AuDbg("err %d\n", (int)(e)); \
6645 +} while (0)
6646 +
6647 +#define AuTraceErrPtr(p) do { \
6648 +       if (IS_ERR(p)) \
6649 +               AuDbg("err %ld\n", PTR_ERR(p)); \
6650 +} while (0)
6651 +
6652 +/* dirty macros for debug print, use with "%.*s" and caution */
6653 +#define AuLNPair(qstr)         (qstr)->len, (qstr)->name
6654 +
6655 +/* ---------------------------------------------------------------------- */
6656 +
6657 +struct au_sbinfo;
6658 +struct au_finfo;
6659 +struct dentry;
6660 +#ifdef CONFIG_AUFS_DEBUG
6661 +extern char *au_plevel;
6662 +struct au_nhash;
6663 +void au_dpri_whlist(struct au_nhash *whlist);
6664 +struct au_vdir;
6665 +void au_dpri_vdir(struct au_vdir *vdir);
6666 +struct inode;
6667 +void au_dpri_inode(struct inode *inode);
6668 +void au_dpri_dalias(struct inode *inode);
6669 +void au_dpri_dentry(struct dentry *dentry);
6670 +struct file;
6671 +void au_dpri_file(struct file *filp);
6672 +struct super_block;
6673 +void au_dpri_sb(struct super_block *sb);
6674 +
6675 +void au_dbg_sleep_jiffy(int jiffy);
6676 +struct iattr;
6677 +void au_dbg_iattr(struct iattr *ia);
6678 +
6679 +#define au_dbg_verify_dinode(d) __au_dbg_verify_dinode(d, __func__, __LINE__)
6680 +void __au_dbg_verify_dinode(struct dentry *dentry, const char *func, int line);
6681 +void au_dbg_verify_dir_parent(struct dentry *dentry, unsigned int sigen);
6682 +void au_dbg_verify_nondir_parent(struct dentry *dentry, unsigned int sigen);
6683 +void au_dbg_verify_gen(struct dentry *parent, unsigned int sigen);
6684 +void au_dbg_verify_kthread(void);
6685 +
6686 +int __init au_debug_init(void);
6687 +void au_debug_sbinfo_init(struct au_sbinfo *sbinfo);
6688 +#define AuDbgWhlist(w) do { \
6689 +       AuDbg(#w "\n"); \
6690 +       au_dpri_whlist(w); \
6691 +} while (0)
6692 +
6693 +#define AuDbgVdir(v) do { \
6694 +       AuDbg(#v "\n"); \
6695 +       au_dpri_vdir(v); \
6696 +} while (0)
6697 +
6698 +#define AuDbgInode(i) do { \
6699 +       AuDbg(#i "\n"); \
6700 +       au_dpri_inode(i); \
6701 +} while (0)
6702 +
6703 +#define AuDbgDAlias(i) do { \
6704 +       AuDbg(#i "\n"); \
6705 +       au_dpri_dalias(i); \
6706 +} while (0)
6707 +
6708 +#define AuDbgDentry(d) do { \
6709 +       AuDbg(#d "\n"); \
6710 +       au_dpri_dentry(d); \
6711 +} while (0)
6712 +
6713 +#define AuDbgFile(f) do { \
6714 +       AuDbg(#f "\n"); \
6715 +       au_dpri_file(f); \
6716 +} while (0)
6717 +
6718 +#define AuDbgSb(sb) do { \
6719 +       AuDbg(#sb "\n"); \
6720 +       au_dpri_sb(sb); \
6721 +} while (0)
6722 +
6723 +#define AuDbgSleep(sec) do { \
6724 +       AuDbg("sleep %d sec\n", sec); \
6725 +       ssleep(sec); \
6726 +} while (0)
6727 +
6728 +#define AuDbgSleepJiffy(jiffy) do { \
6729 +       AuDbg("sleep %d jiffies\n", jiffy); \
6730 +       au_dbg_sleep_jiffy(jiffy); \
6731 +} while (0)
6732 +
6733 +#define AuDbgIAttr(ia) do { \
6734 +       AuDbg("ia_valid 0x%x\n", (ia)->ia_valid); \
6735 +       au_dbg_iattr(ia); \
6736 +} while (0)
6737 +
6738 +#define AuDbgSym(addr) do {                            \
6739 +       char sym[KSYM_SYMBOL_LEN];                      \
6740 +       sprint_symbol(sym, (unsigned long)addr);        \
6741 +       AuDbg("%s\n", sym);                             \
6742 +} while (0)
6743 +
6744 +#define AuInfoSym(addr) do {                           \
6745 +       char sym[KSYM_SYMBOL_LEN];                      \
6746 +       sprint_symbol(sym, (unsigned long)addr);        \
6747 +       AuInfo("%s\n", sym);                            \
6748 +} while (0)
6749 +#else
6750 +AuStubVoid(au_dbg_verify_dinode, struct dentry *dentry)
6751 +AuStubVoid(au_dbg_verify_dir_parent, struct dentry *dentry, unsigned int sigen)
6752 +AuStubVoid(au_dbg_verify_nondir_parent, struct dentry *dentry,
6753 +          unsigned int sigen)
6754 +AuStubVoid(au_dbg_verify_gen, struct dentry *parent, unsigned int sigen)
6755 +AuStubVoid(au_dbg_verify_kthread, void)
6756 +AuStubInt0(__init au_debug_init, void)
6757 +AuStubVoid(au_debug_sbinfo_init, struct au_sbinfo *sbinfo)
6758 +
6759 +#define AuDbgWhlist(w)         do {} while (0)
6760 +#define AuDbgVdir(v)           do {} while (0)
6761 +#define AuDbgInode(i)          do {} while (0)
6762 +#define AuDbgDAlias(i)         do {} while (0)
6763 +#define AuDbgDentry(d)         do {} while (0)
6764 +#define AuDbgFile(f)           do {} while (0)
6765 +#define AuDbgSb(sb)            do {} while (0)
6766 +#define AuDbgSleep(sec)                do {} while (0)
6767 +#define AuDbgSleepJiffy(jiffy) do {} while (0)
6768 +#define AuDbgIAttr(ia)         do {} while (0)
6769 +#define AuDbgSym(addr)         do {} while (0)
6770 +#define AuInfoSym(addr)                do {} while (0)
6771 +#endif /* CONFIG_AUFS_DEBUG */
6772 +
6773 +/* ---------------------------------------------------------------------- */
6774 +
6775 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
6776 +int __init au_sysrq_init(void);
6777 +void au_sysrq_fin(void);
6778 +
6779 +#ifdef CONFIG_HW_CONSOLE
6780 +#define au_dbg_blocked() do { \
6781 +       WARN_ON(1); \
6782 +       handle_sysrq('w'); \
6783 +} while (0)
6784 +#else
6785 +AuStubVoid(au_dbg_blocked, void)
6786 +#endif
6787 +
6788 +#else
6789 +AuStubInt0(__init au_sysrq_init, void)
6790 +AuStubVoid(au_sysrq_fin, void)
6791 +AuStubVoid(au_dbg_blocked, void)
6792 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
6793 +
6794 +#endif /* __KERNEL__ */
6795 +#endif /* __AUFS_DEBUG_H__ */
6796 diff -urN /usr/share/empty/fs/aufs/dentry.c linux/fs/aufs/dentry.c
6797 --- /usr/share/empty/fs/aufs/dentry.c   1970-01-01 01:00:00.000000000 +0100
6798 +++ linux/fs/aufs/dentry.c      2014-04-24 22:11:10.848602379 +0200
6799 @@ -0,0 +1,1081 @@
6800 +/*
6801 + * Copyright (C) 2005-2014 Junjiro R. Okajima
6802 + *
6803 + * This program, aufs is free software; you can redistribute it and/or modify
6804 + * it under the terms of the GNU General Public License as published by
6805 + * the Free Software Foundation; either version 2 of the License, or
6806 + * (at your option) any later version.
6807 + *
6808 + * This program is distributed in the hope that it will be useful,
6809 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
6810 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
6811 + * GNU General Public License for more details.
6812 + *
6813 + * You should have received a copy of the GNU General Public License
6814 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
6815 + */
6816 +
6817 +/*
6818 + * lookup and dentry operations
6819 + */
6820 +
6821 +#include <linux/namei.h>
6822 +#include "aufs.h"
6823 +
6824 +#define AuLkup_ALLOW_NEG       1
6825 +#define au_ftest_lkup(flags, name)     ((flags) & AuLkup_##name)
6826 +#define au_fset_lkup(flags, name) \
6827 +       do { (flags) |= AuLkup_##name; } while (0)
6828 +#define au_fclr_lkup(flags, name) \
6829 +       do { (flags) &= ~AuLkup_##name; } while (0)
6830 +
6831 +struct au_do_lookup_args {
6832 +       unsigned int            flags;
6833 +       mode_t                  type;
6834 +};
6835 +
6836 +/*
6837 + * returns positive/negative dentry, NULL or an error.
6838 + * NULL means whiteout-ed or not-found.
6839 + */
6840 +static struct dentry*
6841 +au_do_lookup(struct dentry *h_parent, struct dentry *dentry,
6842 +            aufs_bindex_t bindex, struct qstr *wh_name,
6843 +            struct au_do_lookup_args *args)
6844 +{
6845 +       struct dentry *h_dentry;
6846 +       struct inode *h_inode, *inode;
6847 +       struct au_branch *br;
6848 +       int wh_found, opq;
6849 +       unsigned char wh_able;
6850 +       const unsigned char allow_neg = !!au_ftest_lkup(args->flags, ALLOW_NEG);
6851 +
6852 +       wh_found = 0;
6853 +       br = au_sbr(dentry->d_sb, bindex);
6854 +       wh_able = !!au_br_whable(br->br_perm);
6855 +       if (wh_able)
6856 +               wh_found = au_wh_test(h_parent, wh_name, br, /*try_sio*/0);
6857 +       h_dentry = ERR_PTR(wh_found);
6858 +       if (!wh_found)
6859 +               goto real_lookup;
6860 +       if (unlikely(wh_found < 0))
6861 +               goto out;
6862 +
6863 +       /* We found a whiteout */
6864 +       /* au_set_dbend(dentry, bindex); */
6865 +       au_set_dbwh(dentry, bindex);
6866 +       if (!allow_neg)
6867 +               return NULL; /* success */
6868 +
6869 +real_lookup:
6870 +       h_dentry = vfsub_lkup_one(&dentry->d_name, h_parent);
6871 +       if (IS_ERR(h_dentry))
6872 +               goto out;
6873 +
6874 +       h_inode = h_dentry->d_inode;
6875 +       if (!h_inode) {
6876 +               if (!allow_neg)
6877 +                       goto out_neg;
6878 +       } else if (wh_found
6879 +                  || (args->type && args->type != (h_inode->i_mode & S_IFMT)))
6880 +               goto out_neg;
6881 +
6882 +       if (au_dbend(dentry) <= bindex)
6883 +               au_set_dbend(dentry, bindex);
6884 +       if (au_dbstart(dentry) < 0 || bindex < au_dbstart(dentry))
6885 +               au_set_dbstart(dentry, bindex);
6886 +       au_set_h_dptr(dentry, bindex, h_dentry);
6887 +
6888 +       inode = dentry->d_inode;
6889 +       if (!h_inode || !S_ISDIR(h_inode->i_mode) || !wh_able
6890 +           || (inode && !S_ISDIR(inode->i_mode)))
6891 +               goto out; /* success */
6892 +
6893 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
6894 +       opq = au_diropq_test(h_dentry, br);
6895 +       mutex_unlock(&h_inode->i_mutex);
6896 +       if (opq > 0)
6897 +               au_set_dbdiropq(dentry, bindex);
6898 +       else if (unlikely(opq < 0)) {
6899 +               au_set_h_dptr(dentry, bindex, NULL);
6900 +               h_dentry = ERR_PTR(opq);
6901 +       }
6902 +       goto out;
6903 +
6904 +out_neg:
6905 +       dput(h_dentry);
6906 +       h_dentry = NULL;
6907 +out:
6908 +       return h_dentry;
6909 +}
6910 +
6911 +static int au_test_shwh(struct super_block *sb, const struct qstr *name)
6912 +{
6913 +       if (unlikely(!au_opt_test(au_mntflags(sb), SHWH)
6914 +                    && !strncmp(name->name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)))
6915 +               return -EPERM;
6916 +       return 0;
6917 +}
6918 +
6919 +/*
6920 + * returns the number of lower positive dentries,
6921 + * otherwise an error.
6922 + * can be called at unlinking with @type is zero.
6923 + */
6924 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t bstart, mode_t type)
6925 +{
6926 +       int npositive, err;
6927 +       aufs_bindex_t bindex, btail, bdiropq;
6928 +       unsigned char isdir;
6929 +       struct qstr whname;
6930 +       struct au_do_lookup_args args = {
6931 +               .flags          = 0,
6932 +               .type           = type
6933 +       };
6934 +       const struct qstr *name = &dentry->d_name;
6935 +       struct dentry *parent;
6936 +       struct inode *inode;
6937 +
6938 +       err = au_test_shwh(dentry->d_sb, name);
6939 +       if (unlikely(err))
6940 +               goto out;
6941 +
6942 +       err = au_wh_name_alloc(&whname, name);
6943 +       if (unlikely(err))
6944 +               goto out;
6945 +
6946 +       inode = dentry->d_inode;
6947 +       isdir = !!(inode && S_ISDIR(inode->i_mode));
6948 +       if (!type)
6949 +               au_fset_lkup(args.flags, ALLOW_NEG);
6950 +
6951 +       npositive = 0;
6952 +       parent = dget_parent(dentry);
6953 +       btail = au_dbtaildir(parent);
6954 +       for (bindex = bstart; bindex <= btail; bindex++) {
6955 +               struct dentry *h_parent, *h_dentry;
6956 +               struct inode *h_inode, *h_dir;
6957 +
6958 +               h_dentry = au_h_dptr(dentry, bindex);
6959 +               if (h_dentry) {
6960 +                       if (h_dentry->d_inode)
6961 +                               npositive++;
6962 +                       if (type != S_IFDIR)
6963 +                               break;
6964 +                       continue;
6965 +               }
6966 +               h_parent = au_h_dptr(parent, bindex);
6967 +               if (!h_parent)
6968 +                       continue;
6969 +               h_dir = h_parent->d_inode;
6970 +               if (!h_dir || !S_ISDIR(h_dir->i_mode))
6971 +                       continue;
6972 +
6973 +               mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
6974 +               h_dentry = au_do_lookup(h_parent, dentry, bindex, &whname,
6975 +                                       &args);
6976 +               mutex_unlock(&h_dir->i_mutex);
6977 +               err = PTR_ERR(h_dentry);
6978 +               if (IS_ERR(h_dentry))
6979 +                       goto out_parent;
6980 +               au_fclr_lkup(args.flags, ALLOW_NEG);
6981 +
6982 +               if (au_dbwh(dentry) >= 0)
6983 +                       break;
6984 +               if (!h_dentry)
6985 +                       continue;
6986 +               h_inode = h_dentry->d_inode;
6987 +               if (!h_inode)
6988 +                       continue;
6989 +               npositive++;
6990 +               if (!args.type)
6991 +                       args.type = h_inode->i_mode & S_IFMT;
6992 +               if (args.type != S_IFDIR)
6993 +                       break;
6994 +               else if (isdir) {
6995 +                       /* the type of lower may be different */
6996 +                       bdiropq = au_dbdiropq(dentry);
6997 +                       if (bdiropq >= 0 && bdiropq <= bindex)
6998 +                               break;
6999 +               }
7000 +       }
7001 +
7002 +       if (npositive) {
7003 +               AuLabel(positive);
7004 +               au_update_dbstart(dentry);
7005 +       }
7006 +       err = npositive;
7007 +       if (unlikely(!au_opt_test(au_mntflags(dentry->d_sb), UDBA_NONE)
7008 +                    && au_dbstart(dentry) < 0)) {
7009 +               err = -EIO;
7010 +               AuIOErr("both of real entry and whiteout found, %pd, err %d\n",
7011 +                       dentry, err);
7012 +       }
7013 +
7014 +out_parent:
7015 +       dput(parent);
7016 +       kfree(whname.name);
7017 +out:
7018 +       return err;
7019 +}
7020 +
7021 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent,
7022 +                              struct au_branch *br)
7023 +{
7024 +       struct dentry *dentry;
7025 +       int wkq_err;
7026 +
7027 +       if (!au_test_h_perm_sio(parent->d_inode, MAY_EXEC))
7028 +               dentry = vfsub_lkup_one(name, parent);
7029 +       else {
7030 +               struct vfsub_lkup_one_args args = {
7031 +                       .errp   = &dentry,
7032 +                       .name   = name,
7033 +                       .parent = parent
7034 +               };
7035 +
7036 +               wkq_err = au_wkq_wait(vfsub_call_lkup_one, &args);
7037 +               if (unlikely(wkq_err))
7038 +                       dentry = ERR_PTR(wkq_err);
7039 +       }
7040 +
7041 +       return dentry;
7042 +}
7043 +
7044 +/*
7045 + * lookup @dentry on @bindex which should be negative.
7046 + */
7047 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh)
7048 +{
7049 +       int err;
7050 +       struct dentry *parent, *h_parent, *h_dentry;
7051 +       struct au_branch *br;
7052 +
7053 +       parent = dget_parent(dentry);
7054 +       h_parent = au_h_dptr(parent, bindex);
7055 +       br = au_sbr(dentry->d_sb, bindex);
7056 +       if (wh)
7057 +               h_dentry = au_whtmp_lkup(h_parent, br, &dentry->d_name);
7058 +       else
7059 +               h_dentry = au_sio_lkup_one(&dentry->d_name, h_parent, br);
7060 +       err = PTR_ERR(h_dentry);
7061 +       if (IS_ERR(h_dentry))
7062 +               goto out;
7063 +       if (unlikely(h_dentry->d_inode)) {
7064 +               err = -EIO;
7065 +               AuIOErr("%pd should be negative on b%d.\n", h_dentry, bindex);
7066 +               dput(h_dentry);
7067 +               goto out;
7068 +       }
7069 +
7070 +       err = 0;
7071 +       if (bindex < au_dbstart(dentry))
7072 +               au_set_dbstart(dentry, bindex);
7073 +       if (au_dbend(dentry) < bindex)
7074 +               au_set_dbend(dentry, bindex);
7075 +       au_set_h_dptr(dentry, bindex, h_dentry);
7076 +
7077 +out:
7078 +       dput(parent);
7079 +       return err;
7080 +}
7081 +
7082 +/* ---------------------------------------------------------------------- */
7083 +
7084 +/* subset of struct inode */
7085 +struct au_iattr {
7086 +       unsigned long           i_ino;
7087 +       /* unsigned int         i_nlink; */
7088 +       kuid_t                  i_uid;
7089 +       kgid_t                  i_gid;
7090 +       u64                     i_version;
7091 +/*
7092 +       loff_t                  i_size;
7093 +       blkcnt_t                i_blocks;
7094 +*/
7095 +       umode_t                 i_mode;
7096 +};
7097 +
7098 +static void au_iattr_save(struct au_iattr *ia, struct inode *h_inode)
7099 +{
7100 +       ia->i_ino = h_inode->i_ino;
7101 +       /* ia->i_nlink = h_inode->i_nlink; */
7102 +       ia->i_uid = h_inode->i_uid;
7103 +       ia->i_gid = h_inode->i_gid;
7104 +       ia->i_version = h_inode->i_version;
7105 +/*
7106 +       ia->i_size = h_inode->i_size;
7107 +       ia->i_blocks = h_inode->i_blocks;
7108 +*/
7109 +       ia->i_mode = (h_inode->i_mode & S_IFMT);
7110 +}
7111 +
7112 +static int au_iattr_test(struct au_iattr *ia, struct inode *h_inode)
7113 +{
7114 +       return ia->i_ino != h_inode->i_ino
7115 +               /* || ia->i_nlink != h_inode->i_nlink */
7116 +               || !uid_eq(ia->i_uid, h_inode->i_uid)
7117 +               || !gid_eq(ia->i_gid, h_inode->i_gid)
7118 +               || ia->i_version != h_inode->i_version
7119 +/*
7120 +               || ia->i_size != h_inode->i_size
7121 +               || ia->i_blocks != h_inode->i_blocks
7122 +*/
7123 +               || ia->i_mode != (h_inode->i_mode & S_IFMT);
7124 +}
7125 +
7126 +static int au_h_verify_dentry(struct dentry *h_dentry, struct dentry *h_parent,
7127 +                             struct au_branch *br)
7128 +{
7129 +       int err;
7130 +       struct au_iattr ia;
7131 +       struct inode *h_inode;
7132 +       struct dentry *h_d;
7133 +       struct super_block *h_sb;
7134 +
7135 +       err = 0;
7136 +       memset(&ia, -1, sizeof(ia));
7137 +       h_sb = h_dentry->d_sb;
7138 +       h_inode = h_dentry->d_inode;
7139 +       if (h_inode)
7140 +               au_iattr_save(&ia, h_inode);
7141 +       else if (au_test_nfs(h_sb) || au_test_fuse(h_sb))
7142 +               /* nfs d_revalidate may return 0 for negative dentry */
7143 +               /* fuse d_revalidate always return 0 for negative dentry */
7144 +               goto out;
7145 +
7146 +       /* main purpose is namei.c:cached_lookup() and d_revalidate */
7147 +       h_d = vfsub_lkup_one(&h_dentry->d_name, h_parent);
7148 +       err = PTR_ERR(h_d);
7149 +       if (IS_ERR(h_d))
7150 +               goto out;
7151 +
7152 +       err = 0;
7153 +       if (unlikely(h_d != h_dentry
7154 +                    || h_d->d_inode != h_inode
7155 +                    || (h_inode && au_iattr_test(&ia, h_inode))))
7156 +               err = au_busy_or_stale();
7157 +       dput(h_d);
7158 +
7159 +out:
7160 +       AuTraceErr(err);
7161 +       return err;
7162 +}
7163 +
7164 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
7165 +               struct dentry *h_parent, struct au_branch *br)
7166 +{
7167 +       int err;
7168 +
7169 +       err = 0;
7170 +       if (udba == AuOpt_UDBA_REVAL
7171 +           && !au_test_fs_remote(h_dentry->d_sb)) {
7172 +               IMustLock(h_dir);
7173 +               err = (h_dentry->d_parent->d_inode != h_dir);
7174 +       } else if (udba != AuOpt_UDBA_NONE)
7175 +               err = au_h_verify_dentry(h_dentry, h_parent, br);
7176 +
7177 +       return err;
7178 +}
7179 +
7180 +/* ---------------------------------------------------------------------- */
7181 +
7182 +static int au_do_refresh_hdentry(struct dentry *dentry, struct dentry *parent)
7183 +{
7184 +       int err;
7185 +       aufs_bindex_t new_bindex, bindex, bend, bwh, bdiropq;
7186 +       struct au_hdentry tmp, *p, *q;
7187 +       struct au_dinfo *dinfo;
7188 +       struct super_block *sb;
7189 +
7190 +       DiMustWriteLock(dentry);
7191 +
7192 +       sb = dentry->d_sb;
7193 +       dinfo = au_di(dentry);
7194 +       bend = dinfo->di_bend;
7195 +       bwh = dinfo->di_bwh;
7196 +       bdiropq = dinfo->di_bdiropq;
7197 +       p = dinfo->di_hdentry + dinfo->di_bstart;
7198 +       for (bindex = dinfo->di_bstart; bindex <= bend; bindex++, p++) {
7199 +               if (!p->hd_dentry)
7200 +                       continue;
7201 +
7202 +               new_bindex = au_br_index(sb, p->hd_id);
7203 +               if (new_bindex == bindex)
7204 +                       continue;
7205 +
7206 +               if (dinfo->di_bwh == bindex)
7207 +                       bwh = new_bindex;
7208 +               if (dinfo->di_bdiropq == bindex)
7209 +                       bdiropq = new_bindex;
7210 +               if (new_bindex < 0) {
7211 +                       au_hdput(p);
7212 +                       p->hd_dentry = NULL;
7213 +                       continue;
7214 +               }
7215 +
7216 +               /* swap two lower dentries, and loop again */
7217 +               q = dinfo->di_hdentry + new_bindex;
7218 +               tmp = *q;
7219 +               *q = *p;
7220 +               *p = tmp;
7221 +               if (tmp.hd_dentry) {
7222 +                       bindex--;
7223 +                       p--;
7224 +               }
7225 +       }
7226 +
7227 +       dinfo->di_bwh = -1;
7228 +       if (bwh >= 0 && bwh <= au_sbend(sb) && au_sbr_whable(sb, bwh))
7229 +               dinfo->di_bwh = bwh;
7230 +
7231 +       dinfo->di_bdiropq = -1;
7232 +       if (bdiropq >= 0
7233 +           && bdiropq <= au_sbend(sb)
7234 +           && au_sbr_whable(sb, bdiropq))
7235 +               dinfo->di_bdiropq = bdiropq;
7236 +
7237 +       err = -EIO;
7238 +       dinfo->di_bstart = -1;
7239 +       dinfo->di_bend = -1;
7240 +       bend = au_dbend(parent);
7241 +       p = dinfo->di_hdentry;
7242 +       for (bindex = 0; bindex <= bend; bindex++, p++)
7243 +               if (p->hd_dentry) {
7244 +                       dinfo->di_bstart = bindex;
7245 +                       break;
7246 +               }
7247 +
7248 +       if (dinfo->di_bstart >= 0) {
7249 +               p = dinfo->di_hdentry + bend;
7250 +               for (bindex = bend; bindex >= 0; bindex--, p--)
7251 +                       if (p->hd_dentry) {
7252 +                               dinfo->di_bend = bindex;
7253 +                               err = 0;
7254 +                               break;
7255 +                       }
7256 +       }
7257 +
7258 +       return err;
7259 +}
7260 +
7261 +static void au_do_hide(struct dentry *dentry)
7262 +{
7263 +       struct inode *inode;
7264 +
7265 +       inode = dentry->d_inode;
7266 +       if (inode) {
7267 +               if (!S_ISDIR(inode->i_mode)) {
7268 +                       if (inode->i_nlink && !d_unhashed(dentry))
7269 +                               drop_nlink(inode);
7270 +               } else {
7271 +                       clear_nlink(inode);
7272 +                       /* stop next lookup */
7273 +                       inode->i_flags |= S_DEAD;
7274 +               }
7275 +               smp_mb(); /* necessary? */
7276 +       }
7277 +       d_drop(dentry);
7278 +}
7279 +
7280 +static int au_hide_children(struct dentry *parent)
7281 +{
7282 +       int err, i, j, ndentry;
7283 +       struct au_dcsub_pages dpages;
7284 +       struct au_dpage *dpage;
7285 +       struct dentry *dentry;
7286 +
7287 +       err = au_dpages_init(&dpages, GFP_NOFS);
7288 +       if (unlikely(err))
7289 +               goto out;
7290 +       err = au_dcsub_pages(&dpages, parent, NULL, NULL);
7291 +       if (unlikely(err))
7292 +               goto out_dpages;
7293 +
7294 +       /* in reverse order */
7295 +       for (i = dpages.ndpage - 1; i >= 0; i--) {
7296 +               dpage = dpages.dpages + i;
7297 +               ndentry = dpage->ndentry;
7298 +               for (j = ndentry - 1; j >= 0; j--) {
7299 +                       dentry = dpage->dentries[j];
7300 +                       if (dentry != parent)
7301 +                               au_do_hide(dentry);
7302 +               }
7303 +       }
7304 +
7305 +out_dpages:
7306 +       au_dpages_free(&dpages);
7307 +out:
7308 +       return err;
7309 +}
7310 +
7311 +static void au_hide(struct dentry *dentry)
7312 +{
7313 +       int err;
7314 +       struct inode *inode;
7315 +
7316 +       AuDbgDentry(dentry);
7317 +       inode = dentry->d_inode;
7318 +       if (inode && S_ISDIR(inode->i_mode)) {
7319 +               /* shrink_dcache_parent(dentry); */
7320 +               err = au_hide_children(dentry);
7321 +               if (unlikely(err))
7322 +                       AuIOErr("%pd, failed hiding children, ignored %d\n",
7323 +                               dentry, err);
7324 +       }
7325 +       au_do_hide(dentry);
7326 +}
7327 +
7328 +/*
7329 + * By adding a dirty branch, a cached dentry may be affected in various ways.
7330 + *
7331 + * a dirty branch is added
7332 + * - on the top of layers
7333 + * - in the middle of layers
7334 + * - to the bottom of layers
7335 + *
7336 + * on the added branch there exists
7337 + * - a whiteout
7338 + * - a diropq
7339 + * - a same named entry
7340 + *   + exist
7341 + *     * negative --> positive
7342 + *     * positive --> positive
7343 + *      - type is unchanged
7344 + *      - type is changed
7345 + *   + doesn't exist
7346 + *     * negative --> negative
7347 + *     * positive --> negative (rejected by au_br_del() for non-dir case)
7348 + * - none
7349 + */
7350 +static int au_refresh_by_dinfo(struct dentry *dentry, struct au_dinfo *dinfo,
7351 +                              struct au_dinfo *tmp)
7352 +{
7353 +       int err;
7354 +       aufs_bindex_t bindex, bend;
7355 +       struct {
7356 +               struct dentry *dentry;
7357 +               struct inode *inode;
7358 +               mode_t mode;
7359 +       } orig_h, tmp_h;
7360 +       struct au_hdentry *hd;
7361 +       struct inode *inode, *h_inode;
7362 +       struct dentry *h_dentry;
7363 +
7364 +       err = 0;
7365 +       AuDebugOn(dinfo->di_bstart < 0);
7366 +       orig_h.dentry = dinfo->di_hdentry[dinfo->di_bstart].hd_dentry;
7367 +       orig_h.inode = orig_h.dentry->d_inode;
7368 +       orig_h.mode = 0;
7369 +       if (orig_h.inode)
7370 +               orig_h.mode = orig_h.inode->i_mode & S_IFMT;
7371 +       memset(&tmp_h, 0, sizeof(tmp_h));
7372 +       if (tmp->di_bstart >= 0) {
7373 +               tmp_h.dentry = tmp->di_hdentry[tmp->di_bstart].hd_dentry;
7374 +               tmp_h.inode = tmp_h.dentry->d_inode;
7375 +               if (tmp_h.inode)
7376 +                       tmp_h.mode = tmp_h.inode->i_mode & S_IFMT;
7377 +       }
7378 +
7379 +       inode = dentry->d_inode;
7380 +       if (!orig_h.inode) {
7381 +               AuDbg("nagative originally\n");
7382 +               if (inode) {
7383 +                       au_hide(dentry);
7384 +                       goto out;
7385 +               }
7386 +               AuDebugOn(inode);
7387 +               AuDebugOn(dinfo->di_bstart != dinfo->di_bend);
7388 +               AuDebugOn(dinfo->di_bdiropq != -1);
7389 +
7390 +               if (!tmp_h.inode) {
7391 +                       AuDbg("negative --> negative\n");
7392 +                       /* should have only one negative lower */
7393 +                       if (tmp->di_bstart >= 0
7394 +                           && tmp->di_bstart < dinfo->di_bstart) {
7395 +                               AuDebugOn(tmp->di_bstart != tmp->di_bend);
7396 +                               AuDebugOn(dinfo->di_bstart != dinfo->di_bend);
7397 +                               au_set_h_dptr(dentry, dinfo->di_bstart, NULL);
7398 +                               au_di_cp(dinfo, tmp);
7399 +                               hd = tmp->di_hdentry + tmp->di_bstart;
7400 +                               au_set_h_dptr(dentry, tmp->di_bstart,
7401 +                                             dget(hd->hd_dentry));
7402 +                       }
7403 +                       au_dbg_verify_dinode(dentry);
7404 +               } else {
7405 +                       AuDbg("negative --> positive\n");
7406 +                       /*
7407 +                        * similar to the behaviour of creating with bypassing
7408 +                        * aufs.
7409 +                        * unhash it in order to force an error in the
7410 +                        * succeeding create operation.
7411 +                        * we should not set S_DEAD here.
7412 +                        */
7413 +                       d_drop(dentry);
7414 +                       /* au_di_swap(tmp, dinfo); */
7415 +                       au_dbg_verify_dinode(dentry);
7416 +               }
7417 +       } else {
7418 +               AuDbg("positive originally\n");
7419 +               /* inode may be NULL */
7420 +               AuDebugOn(inode && (inode->i_mode & S_IFMT) != orig_h.mode);
7421 +               if (!tmp_h.inode) {
7422 +                       AuDbg("positive --> negative\n");
7423 +                       /* or bypassing aufs */
7424 +                       au_hide(dentry);
7425 +                       if (tmp->di_bwh >= 0 && tmp->di_bwh <= dinfo->di_bstart)
7426 +                               dinfo->di_bwh = tmp->di_bwh;
7427 +                       if (inode)
7428 +                               err = au_refresh_hinode_self(inode);
7429 +                       au_dbg_verify_dinode(dentry);
7430 +               } else if (orig_h.mode == tmp_h.mode) {
7431 +                       AuDbg("positive --> positive, same type\n");
7432 +                       if (!S_ISDIR(orig_h.mode)
7433 +                           && dinfo->di_bstart > tmp->di_bstart) {
7434 +                               /*
7435 +                                * similar to the behaviour of removing and
7436 +                                * creating.
7437 +                                */
7438 +                               au_hide(dentry);
7439 +                               if (inode)
7440 +                                       err = au_refresh_hinode_self(inode);
7441 +                               au_dbg_verify_dinode(dentry);
7442 +                       } else {
7443 +                               /* fill empty slots */
7444 +                               if (dinfo->di_bstart > tmp->di_bstart)
7445 +                                       dinfo->di_bstart = tmp->di_bstart;
7446 +                               if (dinfo->di_bend < tmp->di_bend)
7447 +                                       dinfo->di_bend = tmp->di_bend;
7448 +                               dinfo->di_bwh = tmp->di_bwh;
7449 +                               dinfo->di_bdiropq = tmp->di_bdiropq;
7450 +                               hd = tmp->di_hdentry;
7451 +                               bend = dinfo->di_bend;
7452 +                               for (bindex = tmp->di_bstart; bindex <= bend;
7453 +                                    bindex++) {
7454 +                                       if (au_h_dptr(dentry, bindex))
7455 +                                               continue;
7456 +                                       h_dentry = hd[bindex].hd_dentry;
7457 +                                       if (!h_dentry)
7458 +                                               continue;
7459 +                                       h_inode = h_dentry->d_inode;
7460 +                                       AuDebugOn(!h_inode);
7461 +                                       AuDebugOn(orig_h.mode
7462 +                                                 != (h_inode->i_mode
7463 +                                                     & S_IFMT));
7464 +                                       au_set_h_dptr(dentry, bindex,
7465 +                                                     dget(h_dentry));
7466 +                               }
7467 +                               err = au_refresh_hinode(inode, dentry);
7468 +                               au_dbg_verify_dinode(dentry);
7469 +                       }
7470 +               } else {
7471 +                       AuDbg("positive --> positive, different type\n");
7472 +                       /* similar to the behaviour of removing and creating */
7473 +                       au_hide(dentry);
7474 +                       if (inode)
7475 +                               err = au_refresh_hinode_self(inode);
7476 +                       au_dbg_verify_dinode(dentry);
7477 +               }
7478 +       }
7479 +
7480 +out:
7481 +       return err;
7482 +}
7483 +
7484 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent)
7485 +{
7486 +       int err, ebrange;
7487 +       unsigned int sigen;
7488 +       struct au_dinfo *dinfo, *tmp;
7489 +       struct super_block *sb;
7490 +       struct inode *inode;
7491 +
7492 +       DiMustWriteLock(dentry);
7493 +       AuDebugOn(IS_ROOT(dentry));
7494 +       AuDebugOn(!parent->d_inode);
7495 +
7496 +       sb = dentry->d_sb;
7497 +       inode = dentry->d_inode;
7498 +       sigen = au_sigen(sb);
7499 +       err = au_digen_test(parent, sigen);
7500 +       if (unlikely(err))
7501 +               goto out;
7502 +
7503 +       dinfo = au_di(dentry);
7504 +       err = au_di_realloc(dinfo, au_sbend(sb) + 1);
7505 +       if (unlikely(err))
7506 +               goto out;
7507 +       ebrange = au_dbrange_test(dentry);
7508 +       if (!ebrange)
7509 +               ebrange = au_do_refresh_hdentry(dentry, parent);
7510 +
7511 +       if (d_unhashed(dentry) || ebrange) {
7512 +               AuDebugOn(au_dbstart(dentry) < 0 && au_dbend(dentry) >= 0);
7513 +               if (inode)
7514 +                       err = au_refresh_hinode_self(inode);
7515 +               au_dbg_verify_dinode(dentry);
7516 +               if (!err)
7517 +                       goto out_dgen; /* success */
7518 +               goto out;
7519 +       }
7520 +
7521 +       /* temporary dinfo */
7522 +       AuDbgDentry(dentry);
7523 +       err = -ENOMEM;
7524 +       tmp = au_di_alloc(sb, AuLsc_DI_TMP);
7525 +       if (unlikely(!tmp))
7526 +               goto out;
7527 +       au_di_swap(tmp, dinfo);
7528 +       /* returns the number of positive dentries */
7529 +       /*
7530 +        * if current working dir is removed, it returns an error.
7531 +        * but the dentry is legal.
7532 +        */
7533 +       err = au_lkup_dentry(dentry, /*bstart*/0, /*type*/0);
7534 +       AuDbgDentry(dentry);
7535 +       au_di_swap(tmp, dinfo);
7536 +       if (err == -ENOENT)
7537 +               err = 0;
7538 +       if (err >= 0) {
7539 +               /* compare/refresh by dinfo */
7540 +               AuDbgDentry(dentry);
7541 +               err = au_refresh_by_dinfo(dentry, dinfo, tmp);
7542 +               au_dbg_verify_dinode(dentry);
7543 +               AuTraceErr(err);
7544 +       }
7545 +       au_rw_write_unlock(&tmp->di_rwsem);
7546 +       au_di_free(tmp);
7547 +       if (unlikely(err))
7548 +               goto out;
7549 +
7550 +out_dgen:
7551 +       au_update_digen(dentry);
7552 +out:
7553 +       if (unlikely(err && !(dentry->d_flags & DCACHE_NFSFS_RENAMED))) {
7554 +               AuIOErr("failed refreshing %pd, %d\n", dentry, err);
7555 +               AuDbgDentry(dentry);
7556 +       }
7557 +       AuTraceErr(err);
7558 +       return err;
7559 +}
7560 +
7561 +static int au_do_h_d_reval(struct dentry *h_dentry, unsigned int flags,
7562 +                          struct dentry *dentry, aufs_bindex_t bindex)
7563 +{
7564 +       int err, valid;
7565 +
7566 +       err = 0;
7567 +       if (!(h_dentry->d_flags & DCACHE_OP_REVALIDATE))
7568 +               goto out;
7569 +
7570 +       AuDbg("b%d\n", bindex);
7571 +       /*
7572 +        * gave up supporting LOOKUP_CREATE/OPEN for lower fs,
7573 +        * due to whiteout and branch permission.
7574 +        */
7575 +       flags &= ~(/*LOOKUP_PARENT |*/ LOOKUP_OPEN | LOOKUP_CREATE
7576 +                  | LOOKUP_FOLLOW | LOOKUP_EXCL);
7577 +       /* it may return tri-state */
7578 +       valid = h_dentry->d_op->d_revalidate(h_dentry, flags);
7579 +
7580 +       if (unlikely(valid < 0))
7581 +               err = valid;
7582 +       else if (!valid)
7583 +               err = -EINVAL;
7584 +
7585 +out:
7586 +       AuTraceErr(err);
7587 +       return err;
7588 +}
7589 +
7590 +/* todo: remove this */
7591 +static int h_d_revalidate(struct dentry *dentry, struct inode *inode,
7592 +                         unsigned int flags, int do_udba)
7593 +{
7594 +       int err;
7595 +       umode_t mode, h_mode;
7596 +       aufs_bindex_t bindex, btail, bstart, ibs, ibe;
7597 +       unsigned char plus, unhashed, is_root, h_plus, h_nfs;
7598 +       struct inode *h_inode, *h_cached_inode;
7599 +       struct dentry *h_dentry;
7600 +       struct qstr *name, *h_name;
7601 +
7602 +       err = 0;
7603 +       plus = 0;
7604 +       mode = 0;
7605 +       ibs = -1;
7606 +       ibe = -1;
7607 +       unhashed = !!d_unhashed(dentry);
7608 +       is_root = !!IS_ROOT(dentry);
7609 +       name = &dentry->d_name;
7610 +
7611 +       /*
7612 +        * Theoretically, REVAL test should be unnecessary in case of
7613 +        * {FS,I}NOTIFY.
7614 +        * But {fs,i}notify doesn't fire some necessary events,
7615 +        *      IN_ATTRIB for atime/nlink/pageio
7616 +        *      IN_DELETE for NFS dentry
7617 +        * Let's do REVAL test too.
7618 +        */
7619 +       if (do_udba && inode) {
7620 +               mode = (inode->i_mode & S_IFMT);
7621 +               plus = (inode->i_nlink > 0);
7622 +               ibs = au_ibstart(inode);
7623 +               ibe = au_ibend(inode);
7624 +       }
7625 +
7626 +       bstart = au_dbstart(dentry);
7627 +       btail = bstart;
7628 +       if (inode && S_ISDIR(inode->i_mode))
7629 +               btail = au_dbtaildir(dentry);
7630 +       for (bindex = bstart; bindex <= btail; bindex++) {
7631 +               h_dentry = au_h_dptr(dentry, bindex);
7632 +               if (!h_dentry)
7633 +                       continue;
7634 +
7635 +               AuDbg("b%d, %pd\n", bindex, h_dentry);
7636 +               h_nfs = !!au_test_nfs(h_dentry->d_sb);
7637 +               spin_lock(&h_dentry->d_lock);
7638 +               h_name = &h_dentry->d_name;
7639 +               if (unlikely(do_udba
7640 +                            && !is_root
7641 +                            && ((!h_nfs
7642 +                                 && (unhashed != !!d_unhashed(h_dentry)
7643 +                                     || name->len != h_name->len
7644 +                                     || memcmp(name->name, h_name->name,
7645 +                                               name->len)))
7646 +                                || (h_nfs
7647 +                                    && !(flags & LOOKUP_OPEN)
7648 +                                    && (h_dentry->d_flags
7649 +                                        & DCACHE_NFSFS_RENAMED)))
7650 +                           )) {
7651 +                       AuDbg("unhash 0x%x 0x%x, %pd %pd\n",
7652 +                             unhashed, d_unhashed(h_dentry),
7653 +                             dentry, h_dentry);
7654 +                       spin_unlock(&h_dentry->d_lock);
7655 +                       goto err;
7656 +               }
7657 +               spin_unlock(&h_dentry->d_lock);
7658 +
7659 +               err = au_do_h_d_reval(h_dentry, flags, dentry, bindex);
7660 +               if (unlikely(err))
7661 +                       /* do not goto err, to keep the errno */
7662 +                       break;
7663 +
7664 +               /* todo: plink too? */
7665 +               if (!do_udba)
7666 +                       continue;
7667 +
7668 +               /* UDBA tests */
7669 +               h_inode = h_dentry->d_inode;
7670 +               if (unlikely(!!inode != !!h_inode))
7671 +                       goto err;
7672 +
7673 +               h_plus = plus;
7674 +               h_mode = mode;
7675 +               h_cached_inode = h_inode;
7676 +               if (h_inode) {
7677 +                       h_mode = (h_inode->i_mode & S_IFMT);
7678 +                       h_plus = (h_inode->i_nlink > 0);
7679 +               }
7680 +               if (inode && ibs <= bindex && bindex <= ibe)
7681 +                       h_cached_inode = au_h_iptr(inode, bindex);
7682 +
7683 +               if (!h_nfs) {
7684 +                       if (unlikely(plus != h_plus))
7685 +                               goto err;
7686 +               } else {
7687 +                       if (unlikely(!(h_dentry->d_flags & DCACHE_NFSFS_RENAMED)
7688 +                                    && !is_root
7689 +                                    && !IS_ROOT(h_dentry)
7690 +                                    && unhashed != d_unhashed(h_dentry)))
7691 +                               goto err;
7692 +               }
7693 +               if (unlikely(mode != h_mode
7694 +                            || h_cached_inode != h_inode))
7695 +                       goto err;
7696 +               continue;
7697 +
7698 +err:
7699 +               err = -EINVAL;
7700 +               break;
7701 +       }
7702 +
7703 +       AuTraceErr(err);
7704 +       return err;
7705 +}
7706 +
7707 +/* todo: consolidate with do_refresh() and au_reval_for_attr() */
7708 +static int simple_reval_dpath(struct dentry *dentry, unsigned int sigen)
7709 +{
7710 +       int err;
7711 +       struct dentry *parent;
7712 +
7713 +       if (!au_digen_test(dentry, sigen))
7714 +               return 0;
7715 +
7716 +       parent = dget_parent(dentry);
7717 +       di_read_lock_parent(parent, AuLock_IR);
7718 +       AuDebugOn(au_digen_test(parent, sigen));
7719 +       au_dbg_verify_gen(parent, sigen);
7720 +       err = au_refresh_dentry(dentry, parent);
7721 +       di_read_unlock(parent, AuLock_IR);
7722 +       dput(parent);
7723 +       AuTraceErr(err);
7724 +       return err;
7725 +}
7726 +
7727 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen)
7728 +{
7729 +       int err;
7730 +       struct dentry *d, *parent;
7731 +       struct inode *inode;
7732 +
7733 +       if (!au_ftest_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR))
7734 +               return simple_reval_dpath(dentry, sigen);
7735 +
7736 +       /* slow loop, keep it simple and stupid */
7737 +       /* cf: au_cpup_dirs() */
7738 +       err = 0;
7739 +       parent = NULL;
7740 +       while (au_digen_test(dentry, sigen)) {
7741 +               d = dentry;
7742 +               while (1) {
7743 +                       dput(parent);
7744 +                       parent = dget_parent(d);
7745 +                       if (!au_digen_test(parent, sigen))
7746 +                               break;
7747 +                       d = parent;
7748 +               }
7749 +
7750 +               inode = d->d_inode;
7751 +               if (d != dentry)
7752 +                       di_write_lock_child2(d);
7753 +
7754 +               /* someone might update our dentry while we were sleeping */
7755 +               if (au_digen_test(d, sigen)) {
7756 +                       /*
7757 +                        * todo: consolidate with simple_reval_dpath(),
7758 +                        * do_refresh() and au_reval_for_attr().
7759 +                        */
7760 +                       di_read_lock_parent(parent, AuLock_IR);
7761 +                       err = au_refresh_dentry(d, parent);
7762 +                       di_read_unlock(parent, AuLock_IR);
7763 +               }
7764 +
7765 +               if (d != dentry)
7766 +                       di_write_unlock(d);
7767 +               dput(parent);
7768 +               if (unlikely(err))
7769 +                       break;
7770 +       }
7771 +
7772 +       return err;
7773 +}
7774 +
7775 +/*
7776 + * if valid returns 1, otherwise 0.
7777 + */
7778 +static int aufs_d_revalidate(struct dentry *dentry, unsigned int flags)
7779 +{
7780 +       int valid, err;
7781 +       unsigned int sigen;
7782 +       unsigned char do_udba;
7783 +       struct super_block *sb;
7784 +       struct inode *inode;
7785 +
7786 +       /* todo: support rcu-walk? */
7787 +       if (flags & LOOKUP_RCU)
7788 +               return -ECHILD;
7789 +
7790 +       valid = 0;
7791 +       if (unlikely(!au_di(dentry)))
7792 +               goto out;
7793 +
7794 +       inode = dentry->d_inode;
7795 +       if (inode && is_bad_inode(inode))
7796 +               goto out;
7797 +
7798 +       valid = 1;
7799 +       sb = dentry->d_sb;
7800 +       /*
7801 +        * todo: very ugly
7802 +        * i_mutex of parent dir may be held,
7803 +        * but we should not return 'invalid' due to busy.
7804 +        */
7805 +       err = aufs_read_lock(dentry, AuLock_FLUSH | AuLock_DW | AuLock_NOPLM);
7806 +       if (unlikely(err)) {
7807 +               valid = err;
7808 +               AuTraceErr(err);
7809 +               goto out;
7810 +       }
7811 +       if (unlikely(au_dbrange_test(dentry))) {
7812 +               err = -EINVAL;
7813 +               AuTraceErr(err);
7814 +               goto out_dgrade;
7815 +       }
7816 +
7817 +       sigen = au_sigen(sb);
7818 +       if (au_digen_test(dentry, sigen)) {
7819 +               AuDebugOn(IS_ROOT(dentry));
7820 +               err = au_reval_dpath(dentry, sigen);
7821 +               if (unlikely(err)) {
7822 +                       AuTraceErr(err);
7823 +                       goto out_dgrade;
7824 +               }
7825 +       }
7826 +       di_downgrade_lock(dentry, AuLock_IR);
7827 +
7828 +       err = -EINVAL;
7829 +       if (!(flags & LOOKUP_OPEN)
7830 +           && inode
7831 +           && (IS_DEADDIR(inode) || !inode->i_nlink))
7832 +               goto out_inval;
7833 +
7834 +       do_udba = !au_opt_test(au_mntflags(sb), UDBA_NONE);
7835 +       if (do_udba && inode) {
7836 +               aufs_bindex_t bstart = au_ibstart(inode);
7837 +               struct inode *h_inode;
7838 +
7839 +               if (bstart >= 0) {
7840 +                       h_inode = au_h_iptr(inode, bstart);
7841 +                       if (h_inode && au_test_higen(inode, h_inode))
7842 +                               goto out_inval;
7843 +               }
7844 +       }
7845 +
7846 +       err = h_d_revalidate(dentry, inode, flags, do_udba);
7847 +       if (unlikely(!err && do_udba && au_dbstart(dentry) < 0)) {
7848 +               err = -EIO;
7849 +               AuDbg("both of real entry and whiteout found, %p, err %d\n",
7850 +                     dentry, err);
7851 +       }
7852 +       goto out_inval;
7853 +
7854 +out_dgrade:
7855 +       di_downgrade_lock(dentry, AuLock_IR);
7856 +out_inval:
7857 +       aufs_read_unlock(dentry, AuLock_IR);
7858 +       AuTraceErr(err);
7859 +       valid = !err;
7860 +out:
7861 +       if (!valid) {
7862 +               AuDbg("%pd invalid, %d\n", dentry, valid);
7863 +               d_drop(dentry);
7864 +       }
7865 +       return valid;
7866 +}
7867 +
7868 +static void aufs_d_release(struct dentry *dentry)
7869 +{
7870 +       if (au_di(dentry)) {
7871 +               au_di_fin(dentry);
7872 +               au_hn_di_reinit(dentry);
7873 +       }
7874 +}
7875 +
7876 +const struct dentry_operations aufs_dop = {
7877 +       .d_revalidate           = aufs_d_revalidate,
7878 +       .d_weak_revalidate      = aufs_d_revalidate,
7879 +       .d_release              = aufs_d_release
7880 +};
7881 diff -urN /usr/share/empty/fs/aufs/dentry.h linux/fs/aufs/dentry.h
7882 --- /usr/share/empty/fs/aufs/dentry.h   1970-01-01 01:00:00.000000000 +0100
7883 +++ linux/fs/aufs/dentry.h      2014-04-24 22:11:10.848602379 +0200
7884 @@ -0,0 +1,233 @@
7885 +/*
7886 + * Copyright (C) 2005-2014 Junjiro R. Okajima
7887 + *
7888 + * This program, aufs is free software; you can redistribute it and/or modify
7889 + * it under the terms of the GNU General Public License as published by
7890 + * the Free Software Foundation; either version 2 of the License, or
7891 + * (at your option) any later version.
7892 + *
7893 + * This program is distributed in the hope that it will be useful,
7894 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
7895 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
7896 + * GNU General Public License for more details.
7897 + *
7898 + * You should have received a copy of the GNU General Public License
7899 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
7900 + */
7901 +
7902 +/*
7903 + * lookup and dentry operations
7904 + */
7905 +
7906 +#ifndef __AUFS_DENTRY_H__
7907 +#define __AUFS_DENTRY_H__
7908 +
7909 +#ifdef __KERNEL__
7910 +
7911 +#include <linux/dcache.h>
7912 +#include "rwsem.h"
7913 +
7914 +struct au_hdentry {
7915 +       struct dentry           *hd_dentry;
7916 +       aufs_bindex_t           hd_id;
7917 +};
7918 +
7919 +struct au_dinfo {
7920 +       atomic_t                di_generation;
7921 +
7922 +       struct au_rwsem         di_rwsem;
7923 +       aufs_bindex_t           di_bstart, di_bend, di_bwh, di_bdiropq;
7924 +       struct au_hdentry       *di_hdentry;
7925 +} ____cacheline_aligned_in_smp;
7926 +
7927 +/* ---------------------------------------------------------------------- */
7928 +
7929 +/* dentry.c */
7930 +extern const struct dentry_operations aufs_dop;
7931 +struct au_branch;
7932 +struct dentry *au_sio_lkup_one(struct qstr *name, struct dentry *parent,
7933 +                              struct au_branch *br);
7934 +int au_h_verify(struct dentry *h_dentry, unsigned int udba, struct inode *h_dir,
7935 +               struct dentry *h_parent, struct au_branch *br);
7936 +
7937 +int au_lkup_dentry(struct dentry *dentry, aufs_bindex_t bstart, mode_t type);
7938 +int au_lkup_neg(struct dentry *dentry, aufs_bindex_t bindex, int wh);
7939 +int au_refresh_dentry(struct dentry *dentry, struct dentry *parent);
7940 +int au_reval_dpath(struct dentry *dentry, unsigned int sigen);
7941 +
7942 +/* dinfo.c */
7943 +void au_di_init_once(void *_di);
7944 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc);
7945 +void au_di_free(struct au_dinfo *dinfo);
7946 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b);
7947 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src);
7948 +int au_di_init(struct dentry *dentry);
7949 +void au_di_fin(struct dentry *dentry);
7950 +int au_di_realloc(struct au_dinfo *dinfo, int nbr);
7951 +
7952 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc);
7953 +void di_read_unlock(struct dentry *d, int flags);
7954 +void di_downgrade_lock(struct dentry *d, int flags);
7955 +void di_write_lock(struct dentry *d, unsigned int lsc);
7956 +void di_write_unlock(struct dentry *d);
7957 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir);
7958 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir);
7959 +void di_write_unlock2(struct dentry *d1, struct dentry *d2);
7960 +
7961 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex);
7962 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex);
7963 +aufs_bindex_t au_dbtail(struct dentry *dentry);
7964 +aufs_bindex_t au_dbtaildir(struct dentry *dentry);
7965 +
7966 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
7967 +                  struct dentry *h_dentry);
7968 +int au_digen_test(struct dentry *dentry, unsigned int sigen);
7969 +int au_dbrange_test(struct dentry *dentry);
7970 +void au_update_digen(struct dentry *dentry);
7971 +void au_update_dbrange(struct dentry *dentry, int do_put_zero);
7972 +void au_update_dbstart(struct dentry *dentry);
7973 +void au_update_dbend(struct dentry *dentry);
7974 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry);
7975 +
7976 +/* ---------------------------------------------------------------------- */
7977 +
7978 +static inline struct au_dinfo *au_di(struct dentry *dentry)
7979 +{
7980 +       return dentry->d_fsdata;
7981 +}
7982 +
7983 +/* ---------------------------------------------------------------------- */
7984 +
7985 +/* lock subclass for dinfo */
7986 +enum {
7987 +       AuLsc_DI_CHILD,         /* child first */
7988 +       AuLsc_DI_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
7989 +       AuLsc_DI_CHILD3,        /* copyup dirs */
7990 +       AuLsc_DI_PARENT,
7991 +       AuLsc_DI_PARENT2,
7992 +       AuLsc_DI_PARENT3,
7993 +       AuLsc_DI_TMP            /* temp for replacing dinfo */
7994 +};
7995 +
7996 +/*
7997 + * di_read_lock_child, di_write_lock_child,
7998 + * di_read_lock_child2, di_write_lock_child2,
7999 + * di_read_lock_child3, di_write_lock_child3,
8000 + * di_read_lock_parent, di_write_lock_parent,
8001 + * di_read_lock_parent2, di_write_lock_parent2,
8002 + * di_read_lock_parent3, di_write_lock_parent3,
8003 + */
8004 +#define AuReadLockFunc(name, lsc) \
8005 +static inline void di_read_lock_##name(struct dentry *d, int flags) \
8006 +{ di_read_lock(d, flags, AuLsc_DI_##lsc); }
8007 +
8008 +#define AuWriteLockFunc(name, lsc) \
8009 +static inline void di_write_lock_##name(struct dentry *d) \
8010 +{ di_write_lock(d, AuLsc_DI_##lsc); }
8011 +
8012 +#define AuRWLockFuncs(name, lsc) \
8013 +       AuReadLockFunc(name, lsc) \
8014 +       AuWriteLockFunc(name, lsc)
8015 +
8016 +AuRWLockFuncs(child, CHILD);
8017 +AuRWLockFuncs(child2, CHILD2);
8018 +AuRWLockFuncs(child3, CHILD3);
8019 +AuRWLockFuncs(parent, PARENT);
8020 +AuRWLockFuncs(parent2, PARENT2);
8021 +AuRWLockFuncs(parent3, PARENT3);
8022 +
8023 +#undef AuReadLockFunc
8024 +#undef AuWriteLockFunc
8025 +#undef AuRWLockFuncs
8026 +
8027 +#define DiMustNoWaiters(d)     AuRwMustNoWaiters(&au_di(d)->di_rwsem)
8028 +#define DiMustAnyLock(d)       AuRwMustAnyLock(&au_di(d)->di_rwsem)
8029 +#define DiMustWriteLock(d)     AuRwMustWriteLock(&au_di(d)->di_rwsem)
8030 +
8031 +/* ---------------------------------------------------------------------- */
8032 +
8033 +/* todo: memory barrier? */
8034 +static inline unsigned int au_digen(struct dentry *d)
8035 +{
8036 +       return atomic_read(&au_di(d)->di_generation);
8037 +}
8038 +
8039 +static inline void au_h_dentry_init(struct au_hdentry *hdentry)
8040 +{
8041 +       hdentry->hd_dentry = NULL;
8042 +}
8043 +
8044 +static inline void au_hdput(struct au_hdentry *hd)
8045 +{
8046 +       if (hd)
8047 +               dput(hd->hd_dentry);
8048 +}
8049 +
8050 +static inline aufs_bindex_t au_dbstart(struct dentry *dentry)
8051 +{
8052 +       DiMustAnyLock(dentry);
8053 +       return au_di(dentry)->di_bstart;
8054 +}
8055 +
8056 +static inline aufs_bindex_t au_dbend(struct dentry *dentry)
8057 +{
8058 +       DiMustAnyLock(dentry);
8059 +       return au_di(dentry)->di_bend;
8060 +}
8061 +
8062 +static inline aufs_bindex_t au_dbwh(struct dentry *dentry)
8063 +{
8064 +       DiMustAnyLock(dentry);
8065 +       return au_di(dentry)->di_bwh;
8066 +}
8067 +
8068 +static inline aufs_bindex_t au_dbdiropq(struct dentry *dentry)
8069 +{
8070 +       DiMustAnyLock(dentry);
8071 +       return au_di(dentry)->di_bdiropq;
8072 +}
8073 +
8074 +/* todo: hard/soft set? */
8075 +static inline void au_set_dbstart(struct dentry *dentry, aufs_bindex_t bindex)
8076 +{
8077 +       DiMustWriteLock(dentry);
8078 +       au_di(dentry)->di_bstart = bindex;
8079 +}
8080 +
8081 +static inline void au_set_dbend(struct dentry *dentry, aufs_bindex_t bindex)
8082 +{
8083 +       DiMustWriteLock(dentry);
8084 +       au_di(dentry)->di_bend = bindex;
8085 +}
8086 +
8087 +static inline void au_set_dbwh(struct dentry *dentry, aufs_bindex_t bindex)
8088 +{
8089 +       DiMustWriteLock(dentry);
8090 +       /* dbwh can be outside of bstart - bend range */
8091 +       au_di(dentry)->di_bwh = bindex;
8092 +}
8093 +
8094 +static inline void au_set_dbdiropq(struct dentry *dentry, aufs_bindex_t bindex)
8095 +{
8096 +       DiMustWriteLock(dentry);
8097 +       au_di(dentry)->di_bdiropq = bindex;
8098 +}
8099 +
8100 +/* ---------------------------------------------------------------------- */
8101 +
8102 +#ifdef CONFIG_AUFS_HNOTIFY
8103 +static inline void au_digen_dec(struct dentry *d)
8104 +{
8105 +       atomic_dec(&au_di(d)->di_generation);
8106 +}
8107 +
8108 +static inline void au_hn_di_reinit(struct dentry *dentry)
8109 +{
8110 +       dentry->d_fsdata = NULL;
8111 +}
8112 +#else
8113 +AuStubVoid(au_hn_di_reinit, struct dentry *dentry __maybe_unused)
8114 +#endif /* CONFIG_AUFS_HNOTIFY */
8115 +
8116 +#endif /* __KERNEL__ */
8117 +#endif /* __AUFS_DENTRY_H__ */
8118 diff -urN /usr/share/empty/fs/aufs/dinfo.c linux/fs/aufs/dinfo.c
8119 --- /usr/share/empty/fs/aufs/dinfo.c    1970-01-01 01:00:00.000000000 +0100
8120 +++ linux/fs/aufs/dinfo.c       2014-04-24 22:11:10.848602379 +0200
8121 @@ -0,0 +1,542 @@
8122 +/*
8123 + * Copyright (C) 2005-2014 Junjiro R. Okajima
8124 + *
8125 + * This program, aufs is free software; you can redistribute it and/or modify
8126 + * it under the terms of the GNU General Public License as published by
8127 + * the Free Software Foundation; either version 2 of the License, or
8128 + * (at your option) any later version.
8129 + *
8130 + * This program is distributed in the hope that it will be useful,
8131 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8132 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8133 + * GNU General Public License for more details.
8134 + *
8135 + * You should have received a copy of the GNU General Public License
8136 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8137 + */
8138 +
8139 +/*
8140 + * dentry private data
8141 + */
8142 +
8143 +#include "aufs.h"
8144 +
8145 +void au_di_init_once(void *_dinfo)
8146 +{
8147 +       struct au_dinfo *dinfo = _dinfo;
8148 +       static struct lock_class_key aufs_di;
8149 +
8150 +       au_rw_init(&dinfo->di_rwsem);
8151 +       au_rw_class(&dinfo->di_rwsem, &aufs_di);
8152 +}
8153 +
8154 +struct au_dinfo *au_di_alloc(struct super_block *sb, unsigned int lsc)
8155 +{
8156 +       struct au_dinfo *dinfo;
8157 +       int nbr, i;
8158 +
8159 +       dinfo = au_cache_alloc_dinfo();
8160 +       if (unlikely(!dinfo))
8161 +               goto out;
8162 +
8163 +       nbr = au_sbend(sb) + 1;
8164 +       if (nbr <= 0)
8165 +               nbr = 1;
8166 +       dinfo->di_hdentry = kcalloc(nbr, sizeof(*dinfo->di_hdentry), GFP_NOFS);
8167 +       if (dinfo->di_hdentry) {
8168 +               au_rw_write_lock_nested(&dinfo->di_rwsem, lsc);
8169 +               dinfo->di_bstart = -1;
8170 +               dinfo->di_bend = -1;
8171 +               dinfo->di_bwh = -1;
8172 +               dinfo->di_bdiropq = -1;
8173 +               for (i = 0; i < nbr; i++)
8174 +                       dinfo->di_hdentry[i].hd_id = -1;
8175 +               goto out;
8176 +       }
8177 +
8178 +       au_cache_free_dinfo(dinfo);
8179 +       dinfo = NULL;
8180 +
8181 +out:
8182 +       return dinfo;
8183 +}
8184 +
8185 +void au_di_free(struct au_dinfo *dinfo)
8186 +{
8187 +       struct au_hdentry *p;
8188 +       aufs_bindex_t bend, bindex;
8189 +
8190 +       /* dentry may not be revalidated */
8191 +       bindex = dinfo->di_bstart;
8192 +       if (bindex >= 0) {
8193 +               bend = dinfo->di_bend;
8194 +               p = dinfo->di_hdentry + bindex;
8195 +               while (bindex++ <= bend)
8196 +                       au_hdput(p++);
8197 +       }
8198 +       kfree(dinfo->di_hdentry);
8199 +       au_cache_free_dinfo(dinfo);
8200 +}
8201 +
8202 +void au_di_swap(struct au_dinfo *a, struct au_dinfo *b)
8203 +{
8204 +       struct au_hdentry *p;
8205 +       aufs_bindex_t bi;
8206 +
8207 +       AuRwMustWriteLock(&a->di_rwsem);
8208 +       AuRwMustWriteLock(&b->di_rwsem);
8209 +
8210 +#define DiSwap(v, name)                                \
8211 +       do {                                    \
8212 +               v = a->di_##name;               \
8213 +               a->di_##name = b->di_##name;    \
8214 +               b->di_##name = v;               \
8215 +       } while (0)
8216 +
8217 +       DiSwap(p, hdentry);
8218 +       DiSwap(bi, bstart);
8219 +       DiSwap(bi, bend);
8220 +       DiSwap(bi, bwh);
8221 +       DiSwap(bi, bdiropq);
8222 +       /* smp_mb(); */
8223 +
8224 +#undef DiSwap
8225 +}
8226 +
8227 +void au_di_cp(struct au_dinfo *dst, struct au_dinfo *src)
8228 +{
8229 +       AuRwMustWriteLock(&dst->di_rwsem);
8230 +       AuRwMustWriteLock(&src->di_rwsem);
8231 +
8232 +       dst->di_bstart = src->di_bstart;
8233 +       dst->di_bend = src->di_bend;
8234 +       dst->di_bwh = src->di_bwh;
8235 +       dst->di_bdiropq = src->di_bdiropq;
8236 +       /* smp_mb(); */
8237 +}
8238 +
8239 +int au_di_init(struct dentry *dentry)
8240 +{
8241 +       int err;
8242 +       struct super_block *sb;
8243 +       struct au_dinfo *dinfo;
8244 +
8245 +       err = 0;
8246 +       sb = dentry->d_sb;
8247 +       dinfo = au_di_alloc(sb, AuLsc_DI_CHILD);
8248 +       if (dinfo) {
8249 +               atomic_set(&dinfo->di_generation, au_sigen(sb));
8250 +               /* smp_mb(); */ /* atomic_set */
8251 +               dentry->d_fsdata = dinfo;
8252 +       } else
8253 +               err = -ENOMEM;
8254 +
8255 +       return err;
8256 +}
8257 +
8258 +void au_di_fin(struct dentry *dentry)
8259 +{
8260 +       struct au_dinfo *dinfo;
8261 +
8262 +       dinfo = au_di(dentry);
8263 +       AuRwDestroy(&dinfo->di_rwsem);
8264 +       au_di_free(dinfo);
8265 +}
8266 +
8267 +int au_di_realloc(struct au_dinfo *dinfo, int nbr)
8268 +{
8269 +       int err, sz;
8270 +       struct au_hdentry *hdp;
8271 +
8272 +       AuRwMustWriteLock(&dinfo->di_rwsem);
8273 +
8274 +       err = -ENOMEM;
8275 +       sz = sizeof(*hdp) * (dinfo->di_bend + 1);
8276 +       if (!sz)
8277 +               sz = sizeof(*hdp);
8278 +       hdp = au_kzrealloc(dinfo->di_hdentry, sz, sizeof(*hdp) * nbr, GFP_NOFS);
8279 +       if (hdp) {
8280 +               dinfo->di_hdentry = hdp;
8281 +               err = 0;
8282 +       }
8283 +
8284 +       return err;
8285 +}
8286 +
8287 +/* ---------------------------------------------------------------------- */
8288 +
8289 +static void do_ii_write_lock(struct inode *inode, unsigned int lsc)
8290 +{
8291 +       switch (lsc) {
8292 +       case AuLsc_DI_CHILD:
8293 +               ii_write_lock_child(inode);
8294 +               break;
8295 +       case AuLsc_DI_CHILD2:
8296 +               ii_write_lock_child2(inode);
8297 +               break;
8298 +       case AuLsc_DI_CHILD3:
8299 +               ii_write_lock_child3(inode);
8300 +               break;
8301 +       case AuLsc_DI_PARENT:
8302 +               ii_write_lock_parent(inode);
8303 +               break;
8304 +       case AuLsc_DI_PARENT2:
8305 +               ii_write_lock_parent2(inode);
8306 +               break;
8307 +       case AuLsc_DI_PARENT3:
8308 +               ii_write_lock_parent3(inode);
8309 +               break;
8310 +       default:
8311 +               BUG();
8312 +       }
8313 +}
8314 +
8315 +static void do_ii_read_lock(struct inode *inode, unsigned int lsc)
8316 +{
8317 +       switch (lsc) {
8318 +       case AuLsc_DI_CHILD:
8319 +               ii_read_lock_child(inode);
8320 +               break;
8321 +       case AuLsc_DI_CHILD2:
8322 +               ii_read_lock_child2(inode);
8323 +               break;
8324 +       case AuLsc_DI_CHILD3:
8325 +               ii_read_lock_child3(inode);
8326 +               break;
8327 +       case AuLsc_DI_PARENT:
8328 +               ii_read_lock_parent(inode);
8329 +               break;
8330 +       case AuLsc_DI_PARENT2:
8331 +               ii_read_lock_parent2(inode);
8332 +               break;
8333 +       case AuLsc_DI_PARENT3:
8334 +               ii_read_lock_parent3(inode);
8335 +               break;
8336 +       default:
8337 +               BUG();
8338 +       }
8339 +}
8340 +
8341 +void di_read_lock(struct dentry *d, int flags, unsigned int lsc)
8342 +{
8343 +       au_rw_read_lock_nested(&au_di(d)->di_rwsem, lsc);
8344 +       if (d->d_inode) {
8345 +               if (au_ftest_lock(flags, IW))
8346 +                       do_ii_write_lock(d->d_inode, lsc);
8347 +               else if (au_ftest_lock(flags, IR))
8348 +                       do_ii_read_lock(d->d_inode, lsc);
8349 +       }
8350 +}
8351 +
8352 +void di_read_unlock(struct dentry *d, int flags)
8353 +{
8354 +       if (d->d_inode) {
8355 +               if (au_ftest_lock(flags, IW)) {
8356 +                       au_dbg_verify_dinode(d);
8357 +                       ii_write_unlock(d->d_inode);
8358 +               } else if (au_ftest_lock(flags, IR)) {
8359 +                       au_dbg_verify_dinode(d);
8360 +                       ii_read_unlock(d->d_inode);
8361 +               }
8362 +       }
8363 +       au_rw_read_unlock(&au_di(d)->di_rwsem);
8364 +}
8365 +
8366 +void di_downgrade_lock(struct dentry *d, int flags)
8367 +{
8368 +       if (d->d_inode && au_ftest_lock(flags, IR))
8369 +               ii_downgrade_lock(d->d_inode);
8370 +       au_rw_dgrade_lock(&au_di(d)->di_rwsem);
8371 +}
8372 +
8373 +void di_write_lock(struct dentry *d, unsigned int lsc)
8374 +{
8375 +       au_rw_write_lock_nested(&au_di(d)->di_rwsem, lsc);
8376 +       if (d->d_inode)
8377 +               do_ii_write_lock(d->d_inode, lsc);
8378 +}
8379 +
8380 +void di_write_unlock(struct dentry *d)
8381 +{
8382 +       au_dbg_verify_dinode(d);
8383 +       if (d->d_inode)
8384 +               ii_write_unlock(d->d_inode);
8385 +       au_rw_write_unlock(&au_di(d)->di_rwsem);
8386 +}
8387 +
8388 +void di_write_lock2_child(struct dentry *d1, struct dentry *d2, int isdir)
8389 +{
8390 +       AuDebugOn(d1 == d2
8391 +                 || d1->d_inode == d2->d_inode
8392 +                 || d1->d_sb != d2->d_sb);
8393 +
8394 +       if (isdir && au_test_subdir(d1, d2)) {
8395 +               di_write_lock_child(d1);
8396 +               di_write_lock_child2(d2);
8397 +       } else {
8398 +               /* there should be no races */
8399 +               di_write_lock_child(d2);
8400 +               di_write_lock_child2(d1);
8401 +       }
8402 +}
8403 +
8404 +void di_write_lock2_parent(struct dentry *d1, struct dentry *d2, int isdir)
8405 +{
8406 +       AuDebugOn(d1 == d2
8407 +                 || d1->d_inode == d2->d_inode
8408 +                 || d1->d_sb != d2->d_sb);
8409 +
8410 +       if (isdir && au_test_subdir(d1, d2)) {
8411 +               di_write_lock_parent(d1);
8412 +               di_write_lock_parent2(d2);
8413 +       } else {
8414 +               /* there should be no races */
8415 +               di_write_lock_parent(d2);
8416 +               di_write_lock_parent2(d1);
8417 +       }
8418 +}
8419 +
8420 +void di_write_unlock2(struct dentry *d1, struct dentry *d2)
8421 +{
8422 +       di_write_unlock(d1);
8423 +       if (d1->d_inode == d2->d_inode)
8424 +               au_rw_write_unlock(&au_di(d2)->di_rwsem);
8425 +       else
8426 +               di_write_unlock(d2);
8427 +}
8428 +
8429 +/* ---------------------------------------------------------------------- */
8430 +
8431 +struct dentry *au_h_dptr(struct dentry *dentry, aufs_bindex_t bindex)
8432 +{
8433 +       struct dentry *d;
8434 +
8435 +       DiMustAnyLock(dentry);
8436 +
8437 +       if (au_dbstart(dentry) < 0 || bindex < au_dbstart(dentry))
8438 +               return NULL;
8439 +       AuDebugOn(bindex < 0);
8440 +       d = au_di(dentry)->di_hdentry[0 + bindex].hd_dentry;
8441 +       AuDebugOn(d && d_count(d) <= 0);
8442 +       return d;
8443 +}
8444 +
8445 +/*
8446 + * extended version of au_h_dptr().
8447 + * returns a hashed and positive h_dentry in bindex, NULL, or error.
8448 + */
8449 +struct dentry *au_h_d_alias(struct dentry *dentry, aufs_bindex_t bindex)
8450 +{
8451 +       struct dentry *h_dentry;
8452 +       struct inode *inode, *h_inode;
8453 +
8454 +       inode = dentry->d_inode;
8455 +       AuDebugOn(!inode);
8456 +
8457 +       h_dentry = NULL;
8458 +       if (au_dbstart(dentry) <= bindex
8459 +           && bindex <= au_dbend(dentry))
8460 +               h_dentry = au_h_dptr(dentry, bindex);
8461 +       if (h_dentry && !au_d_hashed_positive(h_dentry)) {
8462 +               dget(h_dentry);
8463 +               goto out; /* success */
8464 +       }
8465 +
8466 +       AuDebugOn(bindex < au_ibstart(inode));
8467 +       AuDebugOn(au_ibend(inode) < bindex);
8468 +       h_inode = au_h_iptr(inode, bindex);
8469 +       h_dentry = d_find_alias(h_inode);
8470 +       if (h_dentry) {
8471 +               if (!IS_ERR(h_dentry)) {
8472 +                       if (!au_d_hashed_positive(h_dentry))
8473 +                               goto out; /* success */
8474 +                       dput(h_dentry);
8475 +               } else
8476 +                       goto out;
8477 +       }
8478 +
8479 +       if (au_opt_test(au_mntflags(dentry->d_sb), PLINK)) {
8480 +               h_dentry = au_plink_lkup(inode, bindex);
8481 +               AuDebugOn(!h_dentry);
8482 +               if (!IS_ERR(h_dentry)) {
8483 +                       if (!au_d_hashed_positive(h_dentry))
8484 +                               goto out; /* success */
8485 +                       dput(h_dentry);
8486 +                       h_dentry = NULL;
8487 +               }
8488 +       }
8489 +
8490 +out:
8491 +       AuDbgDentry(h_dentry);
8492 +       return h_dentry;
8493 +}
8494 +
8495 +aufs_bindex_t au_dbtail(struct dentry *dentry)
8496 +{
8497 +       aufs_bindex_t bend, bwh;
8498 +
8499 +       bend = au_dbend(dentry);
8500 +       if (0 <= bend) {
8501 +               bwh = au_dbwh(dentry);
8502 +               if (!bwh)
8503 +                       return bwh;
8504 +               if (0 < bwh && bwh < bend)
8505 +                       return bwh - 1;
8506 +       }
8507 +       return bend;
8508 +}
8509 +
8510 +aufs_bindex_t au_dbtaildir(struct dentry *dentry)
8511 +{
8512 +       aufs_bindex_t bend, bopq;
8513 +
8514 +       bend = au_dbtail(dentry);
8515 +       if (0 <= bend) {
8516 +               bopq = au_dbdiropq(dentry);
8517 +               if (0 <= bopq && bopq < bend)
8518 +                       bend = bopq;
8519 +       }
8520 +       return bend;
8521 +}
8522 +
8523 +/* ---------------------------------------------------------------------- */
8524 +
8525 +void au_set_h_dptr(struct dentry *dentry, aufs_bindex_t bindex,
8526 +                  struct dentry *h_dentry)
8527 +{
8528 +       struct au_hdentry *hd = au_di(dentry)->di_hdentry + bindex;
8529 +       struct au_branch *br;
8530 +
8531 +       DiMustWriteLock(dentry);
8532 +
8533 +       au_hdput(hd);
8534 +       hd->hd_dentry = h_dentry;
8535 +       if (h_dentry) {
8536 +               br = au_sbr(dentry->d_sb, bindex);
8537 +               hd->hd_id = br->br_id;
8538 +       }
8539 +}
8540 +
8541 +int au_dbrange_test(struct dentry *dentry)
8542 +{
8543 +       int err;
8544 +       aufs_bindex_t bstart, bend;
8545 +
8546 +       err = 0;
8547 +       bstart = au_dbstart(dentry);
8548 +       bend = au_dbend(dentry);
8549 +       if (bstart >= 0)
8550 +               AuDebugOn(bend < 0 && bstart > bend);
8551 +       else {
8552 +               err = -EIO;
8553 +               AuDebugOn(bend >= 0);
8554 +       }
8555 +
8556 +       return err;
8557 +}
8558 +
8559 +int au_digen_test(struct dentry *dentry, unsigned int sigen)
8560 +{
8561 +       int err;
8562 +
8563 +       err = 0;
8564 +       if (unlikely(au_digen(dentry) != sigen
8565 +                    || au_iigen_test(dentry->d_inode, sigen)))
8566 +               err = -EIO;
8567 +
8568 +       return err;
8569 +}
8570 +
8571 +void au_update_digen(struct dentry *dentry)
8572 +{
8573 +       atomic_set(&au_di(dentry)->di_generation, au_sigen(dentry->d_sb));
8574 +       /* smp_mb(); */ /* atomic_set */
8575 +}
8576 +
8577 +void au_update_dbrange(struct dentry *dentry, int do_put_zero)
8578 +{
8579 +       struct au_dinfo *dinfo;
8580 +       struct dentry *h_d;
8581 +       struct au_hdentry *hdp;
8582 +
8583 +       DiMustWriteLock(dentry);
8584 +
8585 +       dinfo = au_di(dentry);
8586 +       if (!dinfo || dinfo->di_bstart < 0)
8587 +               return;
8588 +
8589 +       hdp = dinfo->di_hdentry;
8590 +       if (do_put_zero) {
8591 +               aufs_bindex_t bindex, bend;
8592 +
8593 +               bend = dinfo->di_bend;
8594 +               for (bindex = dinfo->di_bstart; bindex <= bend; bindex++) {
8595 +                       h_d = hdp[0 + bindex].hd_dentry;
8596 +                       if (h_d && !h_d->d_inode)
8597 +                               au_set_h_dptr(dentry, bindex, NULL);
8598 +               }
8599 +       }
8600 +
8601 +       dinfo->di_bstart = -1;
8602 +       while (++dinfo->di_bstart <= dinfo->di_bend)
8603 +               if (hdp[0 + dinfo->di_bstart].hd_dentry)
8604 +                       break;
8605 +       if (dinfo->di_bstart > dinfo->di_bend) {
8606 +               dinfo->di_bstart = -1;
8607 +               dinfo->di_bend = -1;
8608 +               return;
8609 +       }
8610 +
8611 +       dinfo->di_bend++;
8612 +       while (0 <= --dinfo->di_bend)
8613 +               if (hdp[0 + dinfo->di_bend].hd_dentry)
8614 +                       break;
8615 +       AuDebugOn(dinfo->di_bstart > dinfo->di_bend || dinfo->di_bend < 0);
8616 +}
8617 +
8618 +void au_update_dbstart(struct dentry *dentry)
8619 +{
8620 +       aufs_bindex_t bindex, bend;
8621 +       struct dentry *h_dentry;
8622 +
8623 +       bend = au_dbend(dentry);
8624 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++) {
8625 +               h_dentry = au_h_dptr(dentry, bindex);
8626 +               if (!h_dentry)
8627 +                       continue;
8628 +               if (h_dentry->d_inode) {
8629 +                       au_set_dbstart(dentry, bindex);
8630 +                       return;
8631 +               }
8632 +               au_set_h_dptr(dentry, bindex, NULL);
8633 +       }
8634 +}
8635 +
8636 +void au_update_dbend(struct dentry *dentry)
8637 +{
8638 +       aufs_bindex_t bindex, bstart;
8639 +       struct dentry *h_dentry;
8640 +
8641 +       bstart = au_dbstart(dentry);
8642 +       for (bindex = au_dbend(dentry); bindex >= bstart; bindex--) {
8643 +               h_dentry = au_h_dptr(dentry, bindex);
8644 +               if (!h_dentry)
8645 +                       continue;
8646 +               if (h_dentry->d_inode) {
8647 +                       au_set_dbend(dentry, bindex);
8648 +                       return;
8649 +               }
8650 +               au_set_h_dptr(dentry, bindex, NULL);
8651 +       }
8652 +}
8653 +
8654 +int au_find_dbindex(struct dentry *dentry, struct dentry *h_dentry)
8655 +{
8656 +       aufs_bindex_t bindex, bend;
8657 +
8658 +       bend = au_dbend(dentry);
8659 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++)
8660 +               if (au_h_dptr(dentry, bindex) == h_dentry)
8661 +                       return bindex;
8662 +       return -1;
8663 +}
8664 diff -urN /usr/share/empty/fs/aufs/dir.c linux/fs/aufs/dir.c
8665 --- /usr/share/empty/fs/aufs/dir.c      1970-01-01 01:00:00.000000000 +0100
8666 +++ linux/fs/aufs/dir.c 2014-04-24 22:11:10.848602379 +0200
8667 @@ -0,0 +1,639 @@
8668 +/*
8669 + * Copyright (C) 2005-2014 Junjiro R. Okajima
8670 + *
8671 + * This program, aufs is free software; you can redistribute it and/or modify
8672 + * it under the terms of the GNU General Public License as published by
8673 + * the Free Software Foundation; either version 2 of the License, or
8674 + * (at your option) any later version.
8675 + *
8676 + * This program is distributed in the hope that it will be useful,
8677 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
8678 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
8679 + * GNU General Public License for more details.
8680 + *
8681 + * You should have received a copy of the GNU General Public License
8682 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
8683 + */
8684 +
8685 +/*
8686 + * directory operations
8687 + */
8688 +
8689 +#include <linux/fs_stack.h>
8690 +#include "aufs.h"
8691 +
8692 +void au_add_nlink(struct inode *dir, struct inode *h_dir)
8693 +{
8694 +       unsigned int nlink;
8695 +
8696 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
8697 +
8698 +       nlink = dir->i_nlink;
8699 +       nlink += h_dir->i_nlink - 2;
8700 +       if (h_dir->i_nlink < 2)
8701 +               nlink += 2;
8702 +       smp_mb(); /* for i_nlink */
8703 +       /* 0 can happen in revaliding */
8704 +       set_nlink(dir, nlink);
8705 +}
8706 +
8707 +void au_sub_nlink(struct inode *dir, struct inode *h_dir)
8708 +{
8709 +       unsigned int nlink;
8710 +
8711 +       AuDebugOn(!S_ISDIR(dir->i_mode) || !S_ISDIR(h_dir->i_mode));
8712 +
8713 +       nlink = dir->i_nlink;
8714 +       nlink -= h_dir->i_nlink - 2;
8715 +       if (h_dir->i_nlink < 2)
8716 +               nlink -= 2;
8717 +       smp_mb(); /* for i_nlink */
8718 +       /* nlink == 0 means the branch-fs is broken */
8719 +       set_nlink(dir, nlink);
8720 +}
8721 +
8722 +loff_t au_dir_size(struct file *file, struct dentry *dentry)
8723 +{
8724 +       loff_t sz;
8725 +       aufs_bindex_t bindex, bend;
8726 +       struct file *h_file;
8727 +       struct dentry *h_dentry;
8728 +
8729 +       sz = 0;
8730 +       if (file) {
8731 +               AuDebugOn(!file_inode(file));
8732 +               AuDebugOn(!S_ISDIR(file_inode(file)->i_mode));
8733 +
8734 +               bend = au_fbend_dir(file);
8735 +               for (bindex = au_fbstart(file);
8736 +                    bindex <= bend && sz < KMALLOC_MAX_SIZE;
8737 +                    bindex++) {
8738 +                       h_file = au_hf_dir(file, bindex);
8739 +                       if (h_file && file_inode(h_file))
8740 +                               sz += vfsub_f_size_read(h_file);
8741 +               }
8742 +       } else {
8743 +               AuDebugOn(!dentry);
8744 +               AuDebugOn(!dentry->d_inode);
8745 +               AuDebugOn(!S_ISDIR(dentry->d_inode->i_mode));
8746 +
8747 +               bend = au_dbtaildir(dentry);
8748 +               for (bindex = au_dbstart(dentry);
8749 +                    bindex <= bend && sz < KMALLOC_MAX_SIZE;
8750 +                    bindex++) {
8751 +                       h_dentry = au_h_dptr(dentry, bindex);
8752 +                       if (h_dentry && h_dentry->d_inode)
8753 +                               sz += i_size_read(h_dentry->d_inode);
8754 +               }
8755 +       }
8756 +       if (sz < KMALLOC_MAX_SIZE)
8757 +               sz = roundup_pow_of_two(sz);
8758 +       if (sz > KMALLOC_MAX_SIZE)
8759 +               sz = KMALLOC_MAX_SIZE;
8760 +       else if (sz < NAME_MAX) {
8761 +               BUILD_BUG_ON(AUFS_RDBLK_DEF < NAME_MAX);
8762 +               sz = AUFS_RDBLK_DEF;
8763 +       }
8764 +       return sz;
8765 +}
8766 +
8767 +/* ---------------------------------------------------------------------- */
8768 +
8769 +static int reopen_dir(struct file *file)
8770 +{
8771 +       int err;
8772 +       unsigned int flags;
8773 +       aufs_bindex_t bindex, btail, bstart;
8774 +       struct dentry *dentry, *h_dentry;
8775 +       struct file *h_file;
8776 +
8777 +       /* open all lower dirs */
8778 +       dentry = file->f_dentry;
8779 +       bstart = au_dbstart(dentry);
8780 +       for (bindex = au_fbstart(file); bindex < bstart; bindex++)
8781 +               au_set_h_fptr(file, bindex, NULL);
8782 +       au_set_fbstart(file, bstart);
8783 +
8784 +       btail = au_dbtaildir(dentry);
8785 +       for (bindex = au_fbend_dir(file); btail < bindex; bindex--)
8786 +               au_set_h_fptr(file, bindex, NULL);
8787 +       au_set_fbend_dir(file, btail);
8788 +
8789 +       flags = vfsub_file_flags(file);
8790 +       for (bindex = bstart; bindex <= btail; bindex++) {
8791 +               h_dentry = au_h_dptr(dentry, bindex);
8792 +               if (!h_dentry)
8793 +                       continue;
8794 +               h_file = au_hf_dir(file, bindex);
8795 +               if (h_file)
8796 +                       continue;
8797 +
8798 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
8799 +               err = PTR_ERR(h_file);
8800 +               if (IS_ERR(h_file))
8801 +                       goto out; /* close all? */
8802 +               au_set_h_fptr(file, bindex, h_file);
8803 +       }
8804 +       au_update_figen(file);
8805 +       /* todo: necessary? */
8806 +       /* file->f_ra = h_file->f_ra; */
8807 +       err = 0;
8808 +
8809 +out:
8810 +       return err;
8811 +}
8812 +
8813 +static int do_open_dir(struct file *file, int flags)
8814 +{
8815 +       int err;
8816 +       aufs_bindex_t bindex, btail;
8817 +       struct dentry *dentry, *h_dentry;
8818 +       struct file *h_file;
8819 +
8820 +       FiMustWriteLock(file);
8821 +
8822 +       err = 0;
8823 +       dentry = file->f_dentry;
8824 +       file->f_version = dentry->d_inode->i_version;
8825 +       bindex = au_dbstart(dentry);
8826 +       au_set_fbstart(file, bindex);
8827 +       btail = au_dbtaildir(dentry);
8828 +       au_set_fbend_dir(file, btail);
8829 +       for (; !err && bindex <= btail; bindex++) {
8830 +               h_dentry = au_h_dptr(dentry, bindex);
8831 +               if (!h_dentry)
8832 +                       continue;
8833 +
8834 +               h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
8835 +               if (IS_ERR(h_file)) {
8836 +                       err = PTR_ERR(h_file);
8837 +                       break;
8838 +               }
8839 +               au_set_h_fptr(file, bindex, h_file);
8840 +       }
8841 +       au_update_figen(file);
8842 +       /* todo: necessary? */
8843 +       /* file->f_ra = h_file->f_ra; */
8844 +       if (!err)
8845 +               return 0; /* success */
8846 +
8847 +       /* close all */
8848 +       for (bindex = au_fbstart(file); bindex <= btail; bindex++)
8849 +               au_set_h_fptr(file, bindex, NULL);
8850 +       au_set_fbstart(file, -1);
8851 +       au_set_fbend_dir(file, -1);
8852 +
8853 +       return err;
8854 +}
8855 +
8856 +static int aufs_open_dir(struct inode *inode __maybe_unused,
8857 +                        struct file *file)
8858 +{
8859 +       int err;
8860 +       struct super_block *sb;
8861 +       struct au_fidir *fidir;
8862 +
8863 +       err = -ENOMEM;
8864 +       sb = file->f_dentry->d_sb;
8865 +       si_read_lock(sb, AuLock_FLUSH);
8866 +       fidir = au_fidir_alloc(sb);
8867 +       if (fidir) {
8868 +               err = au_do_open(file, do_open_dir, fidir);
8869 +               if (unlikely(err))
8870 +                       kfree(fidir);
8871 +       }
8872 +       si_read_unlock(sb);
8873 +       return err;
8874 +}
8875 +
8876 +static int aufs_release_dir(struct inode *inode __maybe_unused,
8877 +                           struct file *file)
8878 +{
8879 +       struct au_vdir *vdir_cache;
8880 +       struct au_finfo *finfo;
8881 +       struct au_fidir *fidir;
8882 +       aufs_bindex_t bindex, bend;
8883 +
8884 +       finfo = au_fi(file);
8885 +       fidir = finfo->fi_hdir;
8886 +       if (fidir) {
8887 +               vdir_cache = fidir->fd_vdir_cache; /* lock-free */
8888 +               if (vdir_cache)
8889 +                       au_vdir_free(vdir_cache);
8890 +
8891 +               bindex = finfo->fi_btop;
8892 +               if (bindex >= 0) {
8893 +                       /*
8894 +                        * calls fput() instead of filp_close(),
8895 +                        * since no dnotify or lock for the lower file.
8896 +                        */
8897 +                       bend = fidir->fd_bbot;
8898 +                       for (; bindex <= bend; bindex++)
8899 +                               au_set_h_fptr(file, bindex, NULL);
8900 +               }
8901 +               kfree(fidir);
8902 +               finfo->fi_hdir = NULL;
8903 +       }
8904 +       au_finfo_fin(file);
8905 +       return 0;
8906 +}
8907 +
8908 +/* ---------------------------------------------------------------------- */
8909 +
8910 +static int au_do_flush_dir(struct file *file, fl_owner_t id)
8911 +{
8912 +       int err;
8913 +       aufs_bindex_t bindex, bend;
8914 +       struct file *h_file;
8915 +
8916 +       err = 0;
8917 +       bend = au_fbend_dir(file);
8918 +       for (bindex = au_fbstart(file); !err && bindex <= bend; bindex++) {
8919 +               h_file = au_hf_dir(file, bindex);
8920 +               if (h_file)
8921 +                       err = vfsub_flush(h_file, id);
8922 +       }
8923 +       return err;
8924 +}
8925 +
8926 +static int aufs_flush_dir(struct file *file, fl_owner_t id)
8927 +{
8928 +       return au_do_flush(file, id, au_do_flush_dir);
8929 +}
8930 +
8931 +/* ---------------------------------------------------------------------- */
8932 +
8933 +static int au_do_fsync_dir_no_file(struct dentry *dentry, int datasync)
8934 +{
8935 +       int err;
8936 +       aufs_bindex_t bend, bindex;
8937 +       struct inode *inode;
8938 +       struct super_block *sb;
8939 +
8940 +       err = 0;
8941 +       sb = dentry->d_sb;
8942 +       inode = dentry->d_inode;
8943 +       IMustLock(inode);
8944 +       bend = au_dbend(dentry);
8945 +       for (bindex = au_dbstart(dentry); !err && bindex <= bend; bindex++) {
8946 +               struct path h_path;
8947 +
8948 +               if (au_test_ro(sb, bindex, inode))
8949 +                       continue;
8950 +               h_path.dentry = au_h_dptr(dentry, bindex);
8951 +               if (!h_path.dentry)
8952 +                       continue;
8953 +
8954 +               h_path.mnt = au_sbr_mnt(sb, bindex);
8955 +               err = vfsub_fsync(NULL, &h_path, datasync);
8956 +       }
8957 +
8958 +       return err;
8959 +}
8960 +
8961 +static int au_do_fsync_dir(struct file *file, int datasync)
8962 +{
8963 +       int err;
8964 +       aufs_bindex_t bend, bindex;
8965 +       struct file *h_file;
8966 +       struct super_block *sb;
8967 +       struct inode *inode;
8968 +
8969 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
8970 +       if (unlikely(err))
8971 +               goto out;
8972 +
8973 +       sb = file->f_dentry->d_sb;
8974 +       inode = file_inode(file);
8975 +       bend = au_fbend_dir(file);
8976 +       for (bindex = au_fbstart(file); !err && bindex <= bend; bindex++) {
8977 +               h_file = au_hf_dir(file, bindex);
8978 +               if (!h_file || au_test_ro(sb, bindex, inode))
8979 +                       continue;
8980 +
8981 +               err = vfsub_fsync(h_file, &h_file->f_path, datasync);
8982 +       }
8983 +
8984 +out:
8985 +       return err;
8986 +}
8987 +
8988 +/*
8989 + * @file may be NULL
8990 + */
8991 +static int aufs_fsync_dir(struct file *file, loff_t start, loff_t end,
8992 +                         int datasync)
8993 +{
8994 +       int err;
8995 +       struct dentry *dentry;
8996 +       struct super_block *sb;
8997 +       struct mutex *mtx;
8998 +
8999 +       err = 0;
9000 +       dentry = file->f_dentry;
9001 +       mtx = &dentry->d_inode->i_mutex;
9002 +       mutex_lock(mtx);
9003 +       sb = dentry->d_sb;
9004 +       si_noflush_read_lock(sb);
9005 +       if (file)
9006 +               err = au_do_fsync_dir(file, datasync);
9007 +       else {
9008 +               di_write_lock_child(dentry);
9009 +               err = au_do_fsync_dir_no_file(dentry, datasync);
9010 +       }
9011 +       au_cpup_attr_timesizes(dentry->d_inode);
9012 +       di_write_unlock(dentry);
9013 +       if (file)
9014 +               fi_write_unlock(file);
9015 +
9016 +       si_read_unlock(sb);
9017 +       mutex_unlock(mtx);
9018 +       return err;
9019 +}
9020 +
9021 +/* ---------------------------------------------------------------------- */
9022 +
9023 +static int aufs_iterate(struct file *file, struct dir_context *ctx)
9024 +{
9025 +       int err;
9026 +       struct dentry *dentry;
9027 +       struct inode *inode, *h_inode;
9028 +       struct super_block *sb;
9029 +
9030 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
9031 +
9032 +       dentry = file->f_dentry;
9033 +       inode = dentry->d_inode;
9034 +       IMustLock(inode);
9035 +
9036 +       sb = dentry->d_sb;
9037 +       si_read_lock(sb, AuLock_FLUSH);
9038 +       err = au_reval_and_lock_fdi(file, reopen_dir, /*wlock*/1);
9039 +       if (unlikely(err))
9040 +               goto out;
9041 +       err = au_alive_dir(dentry);
9042 +       if (!err)
9043 +               err = au_vdir_init(file);
9044 +       di_downgrade_lock(dentry, AuLock_IR);
9045 +       if (unlikely(err))
9046 +               goto out_unlock;
9047 +
9048 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
9049 +       if (!au_test_nfsd()) {
9050 +               err = au_vdir_fill_de(file, ctx);
9051 +               fsstack_copy_attr_atime(inode, h_inode);
9052 +       } else {
9053 +               /*
9054 +                * nfsd filldir may call lookup_one_len(), vfs_getattr(),
9055 +                * encode_fh() and others.
9056 +                */
9057 +               atomic_inc(&h_inode->i_count);
9058 +               di_read_unlock(dentry, AuLock_IR);
9059 +               si_read_unlock(sb);
9060 +               err = au_vdir_fill_de(file, ctx);
9061 +               fsstack_copy_attr_atime(inode, h_inode);
9062 +               fi_write_unlock(file);
9063 +               iput(h_inode);
9064 +
9065 +               AuTraceErr(err);
9066 +               return err;
9067 +       }
9068 +
9069 +out_unlock:
9070 +       di_read_unlock(dentry, AuLock_IR);
9071 +       fi_write_unlock(file);
9072 +out:
9073 +       si_read_unlock(sb);
9074 +       return err;
9075 +}
9076 +
9077 +/* ---------------------------------------------------------------------- */
9078 +
9079 +#define AuTestEmpty_WHONLY     1
9080 +#define AuTestEmpty_CALLED     (1 << 1)
9081 +#define AuTestEmpty_SHWH       (1 << 2)
9082 +#define au_ftest_testempty(flags, name)        ((flags) & AuTestEmpty_##name)
9083 +#define au_fset_testempty(flags, name) \
9084 +       do { (flags) |= AuTestEmpty_##name; } while (0)
9085 +#define au_fclr_testempty(flags, name) \
9086 +       do { (flags) &= ~AuTestEmpty_##name; } while (0)
9087 +
9088 +#ifndef CONFIG_AUFS_SHWH
9089 +#undef AuTestEmpty_SHWH
9090 +#define AuTestEmpty_SHWH       0
9091 +#endif
9092 +
9093 +struct test_empty_arg {
9094 +       struct dir_context ctx;
9095 +       struct au_nhash *whlist;
9096 +       unsigned int flags;
9097 +       int err;
9098 +       aufs_bindex_t bindex;
9099 +};
9100 +
9101 +static int test_empty_cb(struct dir_context *ctx, const char *__name,
9102 +                        int namelen, loff_t offset __maybe_unused, u64 ino,
9103 +                        unsigned int d_type)
9104 +{
9105 +       struct test_empty_arg *arg = container_of(ctx, struct test_empty_arg,
9106 +                                                 ctx);
9107 +       char *name = (void *)__name;
9108 +
9109 +       arg->err = 0;
9110 +       au_fset_testempty(arg->flags, CALLED);
9111 +       /* smp_mb(); */
9112 +       if (name[0] == '.'
9113 +           && (namelen == 1 || (name[1] == '.' && namelen == 2)))
9114 +               goto out; /* success */
9115 +
9116 +       if (namelen <= AUFS_WH_PFX_LEN
9117 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
9118 +               if (au_ftest_testempty(arg->flags, WHONLY)
9119 +                   && !au_nhash_test_known_wh(arg->whlist, name, namelen))
9120 +                       arg->err = -ENOTEMPTY;
9121 +               goto out;
9122 +       }
9123 +
9124 +       name += AUFS_WH_PFX_LEN;
9125 +       namelen -= AUFS_WH_PFX_LEN;
9126 +       if (!au_nhash_test_known_wh(arg->whlist, name, namelen))
9127 +               arg->err = au_nhash_append_wh
9128 +                       (arg->whlist, name, namelen, ino, d_type, arg->bindex,
9129 +                        au_ftest_testempty(arg->flags, SHWH));
9130 +
9131 +out:
9132 +       /* smp_mb(); */
9133 +       AuTraceErr(arg->err);
9134 +       return arg->err;
9135 +}
9136 +
9137 +static int do_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
9138 +{
9139 +       int err;
9140 +       struct file *h_file;
9141 +
9142 +       h_file = au_h_open(dentry, arg->bindex,
9143 +                          O_RDONLY | O_NONBLOCK | O_DIRECTORY | O_LARGEFILE,
9144 +                          /*file*/NULL, /*force_wr*/0);
9145 +       err = PTR_ERR(h_file);
9146 +       if (IS_ERR(h_file))
9147 +               goto out;
9148 +
9149 +       err = 0;
9150 +       if (!au_opt_test(au_mntflags(dentry->d_sb), UDBA_NONE)
9151 +           && !file_inode(h_file)->i_nlink)
9152 +               goto out_put;
9153 +
9154 +       do {
9155 +               arg->err = 0;
9156 +               au_fclr_testempty(arg->flags, CALLED);
9157 +               /* smp_mb(); */
9158 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
9159 +               if (err >= 0)
9160 +                       err = arg->err;
9161 +       } while (!err && au_ftest_testempty(arg->flags, CALLED));
9162 +
9163 +out_put:
9164 +       fput(h_file);
9165 +       au_sbr_put(dentry->d_sb, arg->bindex);
9166 +out:
9167 +       return err;
9168 +}
9169 +
9170 +struct do_test_empty_args {
9171 +       int *errp;
9172 +       struct dentry *dentry;
9173 +       struct test_empty_arg *arg;
9174 +};
9175 +
9176 +static void call_do_test_empty(void *args)
9177 +{
9178 +       struct do_test_empty_args *a = args;
9179 +       *a->errp = do_test_empty(a->dentry, a->arg);
9180 +}
9181 +
9182 +static int sio_test_empty(struct dentry *dentry, struct test_empty_arg *arg)
9183 +{
9184 +       int err, wkq_err;
9185 +       struct dentry *h_dentry;
9186 +       struct inode *h_inode;
9187 +
9188 +       h_dentry = au_h_dptr(dentry, arg->bindex);
9189 +       h_inode = h_dentry->d_inode;
9190 +       /* todo: i_mode changes anytime? */
9191 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
9192 +       err = au_test_h_perm_sio(h_inode, MAY_EXEC | MAY_READ);
9193 +       mutex_unlock(&h_inode->i_mutex);
9194 +       if (!err)
9195 +               err = do_test_empty(dentry, arg);
9196 +       else {
9197 +               struct do_test_empty_args args = {
9198 +                       .errp   = &err,
9199 +                       .dentry = dentry,
9200 +                       .arg    = arg
9201 +               };
9202 +               unsigned int flags = arg->flags;
9203 +
9204 +               wkq_err = au_wkq_wait(call_do_test_empty, &args);
9205 +               if (unlikely(wkq_err))
9206 +                       err = wkq_err;
9207 +               arg->flags = flags;
9208 +       }
9209 +
9210 +       return err;
9211 +}
9212 +
9213 +int au_test_empty_lower(struct dentry *dentry)
9214 +{
9215 +       int err;
9216 +       unsigned int rdhash;
9217 +       aufs_bindex_t bindex, bstart, btail;
9218 +       struct au_nhash whlist;
9219 +       struct test_empty_arg arg = {
9220 +               .ctx = {
9221 +                       .actor = au_diractor(test_empty_cb)
9222 +               }
9223 +       };
9224 +
9225 +       SiMustAnyLock(dentry->d_sb);
9226 +
9227 +       rdhash = au_sbi(dentry->d_sb)->si_rdhash;
9228 +       if (!rdhash)
9229 +               rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, dentry));
9230 +       err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
9231 +       if (unlikely(err))
9232 +               goto out;
9233 +
9234 +       arg.flags = 0;
9235 +       arg.whlist = &whlist;
9236 +       bstart = au_dbstart(dentry);
9237 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
9238 +               au_fset_testempty(arg.flags, SHWH);
9239 +       arg.bindex = bstart;
9240 +       err = do_test_empty(dentry, &arg);
9241 +       if (unlikely(err))
9242 +               goto out_whlist;
9243 +
9244 +       au_fset_testempty(arg.flags, WHONLY);
9245 +       btail = au_dbtaildir(dentry);
9246 +       for (bindex = bstart + 1; !err && bindex <= btail; bindex++) {
9247 +               struct dentry *h_dentry;
9248 +
9249 +               h_dentry = au_h_dptr(dentry, bindex);
9250 +               if (h_dentry && h_dentry->d_inode) {
9251 +                       arg.bindex = bindex;
9252 +                       err = do_test_empty(dentry, &arg);
9253 +               }
9254 +       }
9255 +
9256 +out_whlist:
9257 +       au_nhash_wh_free(&whlist);
9258 +out:
9259 +       return err;
9260 +}
9261 +
9262 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist)
9263 +{
9264 +       int err;
9265 +       struct test_empty_arg arg = {
9266 +               .ctx = {
9267 +                       .actor = au_diractor(test_empty_cb)
9268 +               }
9269 +       };
9270 +       aufs_bindex_t bindex, btail;
9271 +
9272 +       err = 0;
9273 +       arg.whlist = whlist;
9274 +       arg.flags = AuTestEmpty_WHONLY;
9275 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH))
9276 +               au_fset_testempty(arg.flags, SHWH);
9277 +       btail = au_dbtaildir(dentry);
9278 +       for (bindex = au_dbstart(dentry); !err && bindex <= btail; bindex++) {
9279 +               struct dentry *h_dentry;
9280 +
9281 +               h_dentry = au_h_dptr(dentry, bindex);
9282 +               if (h_dentry && h_dentry->d_inode) {
9283 +                       arg.bindex = bindex;
9284 +                       err = sio_test_empty(dentry, &arg);
9285 +               }
9286 +       }
9287 +
9288 +       return err;
9289 +}
9290 +
9291 +/* ---------------------------------------------------------------------- */
9292 +
9293 +const struct file_operations aufs_dir_fop = {
9294 +       .owner          = THIS_MODULE,
9295 +       .llseek         = default_llseek,
9296 +       .read           = generic_read_dir,
9297 +       .iterate        = aufs_iterate,
9298 +       .unlocked_ioctl = aufs_ioctl_dir,
9299 +#ifdef CONFIG_COMPAT
9300 +       .compat_ioctl   = aufs_compat_ioctl_dir,
9301 +#endif
9302 +       .open           = aufs_open_dir,
9303 +       .release        = aufs_release_dir,
9304 +       .flush          = aufs_flush_dir,
9305 +       .fsync          = aufs_fsync_dir
9306 +};
9307 diff -urN /usr/share/empty/fs/aufs/dir.h linux/fs/aufs/dir.h
9308 --- /usr/share/empty/fs/aufs/dir.h      1970-01-01 01:00:00.000000000 +0100
9309 +++ linux/fs/aufs/dir.h 2014-04-24 22:11:10.851935747 +0200
9310 @@ -0,0 +1,136 @@
9311 +/*
9312 + * Copyright (C) 2005-2014 Junjiro R. Okajima
9313 + *
9314 + * This program, aufs is free software; you can redistribute it and/or modify
9315 + * it under the terms of the GNU General Public License as published by
9316 + * the Free Software Foundation; either version 2 of the License, or
9317 + * (at your option) any later version.
9318 + *
9319 + * This program is distributed in the hope that it will be useful,
9320 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9321 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9322 + * GNU General Public License for more details.
9323 + *
9324 + * You should have received a copy of the GNU General Public License
9325 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9326 + */
9327 +
9328 +/*
9329 + * directory operations
9330 + */
9331 +
9332 +#ifndef __AUFS_DIR_H__
9333 +#define __AUFS_DIR_H__
9334 +
9335 +#ifdef __KERNEL__
9336 +
9337 +#include <linux/fs.h>
9338 +
9339 +/* ---------------------------------------------------------------------- */
9340 +
9341 +/* need to be faster and smaller */
9342 +
9343 +struct au_nhash {
9344 +       unsigned int            nh_num;
9345 +       struct hlist_head       *nh_head;
9346 +};
9347 +
9348 +struct au_vdir_destr {
9349 +       unsigned char   len;
9350 +       unsigned char   name[0];
9351 +} __packed;
9352 +
9353 +struct au_vdir_dehstr {
9354 +       struct hlist_node       hash;
9355 +       struct au_vdir_destr    *str;
9356 +} ____cacheline_aligned_in_smp;
9357 +
9358 +struct au_vdir_de {
9359 +       ino_t                   de_ino;
9360 +       unsigned char           de_type;
9361 +       /* caution: packed */
9362 +       struct au_vdir_destr    de_str;
9363 +} __packed;
9364 +
9365 +struct au_vdir_wh {
9366 +       struct hlist_node       wh_hash;
9367 +#ifdef CONFIG_AUFS_SHWH
9368 +       ino_t                   wh_ino;
9369 +       aufs_bindex_t           wh_bindex;
9370 +       unsigned char           wh_type;
9371 +#else
9372 +       aufs_bindex_t           wh_bindex;
9373 +#endif
9374 +       /* caution: packed */
9375 +       struct au_vdir_destr    wh_str;
9376 +} __packed;
9377 +
9378 +union au_vdir_deblk_p {
9379 +       unsigned char           *deblk;
9380 +       struct au_vdir_de       *de;
9381 +};
9382 +
9383 +struct au_vdir {
9384 +       unsigned char   **vd_deblk;
9385 +       unsigned long   vd_nblk;
9386 +       struct {
9387 +               unsigned long           ul;
9388 +               union au_vdir_deblk_p   p;
9389 +       } vd_last;
9390 +
9391 +       unsigned long   vd_version;
9392 +       unsigned int    vd_deblk_sz;
9393 +       unsigned long   vd_jiffy;
9394 +} ____cacheline_aligned_in_smp;
9395 +
9396 +/* ---------------------------------------------------------------------- */
9397 +
9398 +/* dir.c */
9399 +extern const struct file_operations aufs_dir_fop;
9400 +void au_add_nlink(struct inode *dir, struct inode *h_dir);
9401 +void au_sub_nlink(struct inode *dir, struct inode *h_dir);
9402 +loff_t au_dir_size(struct file *file, struct dentry *dentry);
9403 +int au_test_empty_lower(struct dentry *dentry);
9404 +int au_test_empty(struct dentry *dentry, struct au_nhash *whlist);
9405 +
9406 +/* vdir.c */
9407 +unsigned int au_rdhash_est(loff_t sz);
9408 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp);
9409 +void au_nhash_wh_free(struct au_nhash *whlist);
9410 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
9411 +                           int limit);
9412 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen);
9413 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
9414 +                      unsigned int d_type, aufs_bindex_t bindex,
9415 +                      unsigned char shwh);
9416 +void au_vdir_free(struct au_vdir *vdir);
9417 +int au_vdir_init(struct file *file);
9418 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx);
9419 +
9420 +/* ioctl.c */
9421 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg);
9422 +
9423 +#ifdef CONFIG_AUFS_RDU
9424 +/* rdu.c */
9425 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
9426 +#ifdef CONFIG_COMPAT
9427 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd,
9428 +                        unsigned long arg);
9429 +#endif
9430 +#else
9431 +static inline long au_rdu_ioctl(struct file *file, unsigned int cmd,
9432 +                               unsigned long arg)
9433 +{
9434 +       return -EINVAL;
9435 +}
9436 +#ifdef CONFIG_COMPAT
9437 +static inline long au_rdu_compat_ioctl(struct file *file, unsigned int cmd,
9438 +                                      unsigned long arg)
9439 +{
9440 +       return -EINVAL;
9441 +}
9442 +#endif
9443 +#endif
9444 +
9445 +#endif /* __KERNEL__ */
9446 +#endif /* __AUFS_DIR_H__ */
9447 diff -urN /usr/share/empty/fs/aufs/dynop.c linux/fs/aufs/dynop.c
9448 --- /usr/share/empty/fs/aufs/dynop.c    1970-01-01 01:00:00.000000000 +0100
9449 +++ linux/fs/aufs/dynop.c       2014-04-24 22:11:10.851935747 +0200
9450 @@ -0,0 +1,379 @@
9451 +/*
9452 + * Copyright (C) 2010-2014 Junjiro R. Okajima
9453 + *
9454 + * This program, aufs is free software; you can redistribute it and/or modify
9455 + * it under the terms of the GNU General Public License as published by
9456 + * the Free Software Foundation; either version 2 of the License, or
9457 + * (at your option) any later version.
9458 + *
9459 + * This program is distributed in the hope that it will be useful,
9460 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9461 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9462 + * GNU General Public License for more details.
9463 + *
9464 + * You should have received a copy of the GNU General Public License
9465 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9466 + */
9467 +
9468 +/*
9469 + * dynamically customizable operations for regular files
9470 + */
9471 +
9472 +#include "aufs.h"
9473 +
9474 +#define DyPrSym(key)   AuDbgSym(key->dk_op.dy_hop)
9475 +
9476 +/*
9477 + * How large will these lists be?
9478 + * Usually just a few elements, 20-30 at most for each, I guess.
9479 + */
9480 +static struct au_splhead dynop[AuDyLast];
9481 +
9482 +static struct au_dykey *dy_gfind_get(struct au_splhead *spl, const void *h_op)
9483 +{
9484 +       struct au_dykey *key, *tmp;
9485 +       struct list_head *head;
9486 +
9487 +       key = NULL;
9488 +       head = &spl->head;
9489 +       rcu_read_lock();
9490 +       list_for_each_entry_rcu(tmp, head, dk_list)
9491 +               if (tmp->dk_op.dy_hop == h_op) {
9492 +                       key = tmp;
9493 +                       kref_get(&key->dk_kref);
9494 +                       break;
9495 +               }
9496 +       rcu_read_unlock();
9497 +
9498 +       return key;
9499 +}
9500 +
9501 +static struct au_dykey *dy_bradd(struct au_branch *br, struct au_dykey *key)
9502 +{
9503 +       struct au_dykey **k, *found;
9504 +       const void *h_op = key->dk_op.dy_hop;
9505 +       int i;
9506 +
9507 +       found = NULL;
9508 +       k = br->br_dykey;
9509 +       for (i = 0; i < AuBrDynOp; i++)
9510 +               if (k[i]) {
9511 +                       if (k[i]->dk_op.dy_hop == h_op) {
9512 +                               found = k[i];
9513 +                               break;
9514 +                       }
9515 +               } else
9516 +                       break;
9517 +       if (!found) {
9518 +               spin_lock(&br->br_dykey_lock);
9519 +               for (; i < AuBrDynOp; i++)
9520 +                       if (k[i]) {
9521 +                               if (k[i]->dk_op.dy_hop == h_op) {
9522 +                                       found = k[i];
9523 +                                       break;
9524 +                               }
9525 +                       } else {
9526 +                               k[i] = key;
9527 +                               break;
9528 +                       }
9529 +               spin_unlock(&br->br_dykey_lock);
9530 +               BUG_ON(i == AuBrDynOp); /* expand the array */
9531 +       }
9532 +
9533 +       return found;
9534 +}
9535 +
9536 +/* kref_get() if @key is already added */
9537 +static struct au_dykey *dy_gadd(struct au_splhead *spl, struct au_dykey *key)
9538 +{
9539 +       struct au_dykey *tmp, *found;
9540 +       struct list_head *head;
9541 +       const void *h_op = key->dk_op.dy_hop;
9542 +
9543 +       found = NULL;
9544 +       head = &spl->head;
9545 +       spin_lock(&spl->spin);
9546 +       list_for_each_entry(tmp, head, dk_list)
9547 +               if (tmp->dk_op.dy_hop == h_op) {
9548 +                       kref_get(&tmp->dk_kref);
9549 +                       found = tmp;
9550 +                       break;
9551 +               }
9552 +       if (!found)
9553 +               list_add_rcu(&key->dk_list, head);
9554 +       spin_unlock(&spl->spin);
9555 +
9556 +       if (!found)
9557 +               DyPrSym(key);
9558 +       return found;
9559 +}
9560 +
9561 +static void dy_free_rcu(struct rcu_head *rcu)
9562 +{
9563 +       struct au_dykey *key;
9564 +
9565 +       key = container_of(rcu, struct au_dykey, dk_rcu);
9566 +       DyPrSym(key);
9567 +       kfree(key);
9568 +}
9569 +
9570 +static void dy_free(struct kref *kref)
9571 +{
9572 +       struct au_dykey *key;
9573 +       struct au_splhead *spl;
9574 +
9575 +       key = container_of(kref, struct au_dykey, dk_kref);
9576 +       spl = dynop + key->dk_op.dy_type;
9577 +       au_spl_del_rcu(&key->dk_list, spl);
9578 +       call_rcu(&key->dk_rcu, dy_free_rcu);
9579 +}
9580 +
9581 +void au_dy_put(struct au_dykey *key)
9582 +{
9583 +       kref_put(&key->dk_kref, dy_free);
9584 +}
9585 +
9586 +/* ---------------------------------------------------------------------- */
9587 +
9588 +#define DyDbgSize(cnt, op)     AuDebugOn(cnt != sizeof(op)/sizeof(void *))
9589 +
9590 +#ifdef CONFIG_AUFS_DEBUG
9591 +#define DyDbgDeclare(cnt)      unsigned int cnt = 0
9592 +#define DyDbgInc(cnt)          do { cnt++; } while (0)
9593 +#else
9594 +#define DyDbgDeclare(cnt)      do {} while (0)
9595 +#define DyDbgInc(cnt)          do {} while (0)
9596 +#endif
9597 +
9598 +#define DySet(func, dst, src, h_op, h_sb) do {                         \
9599 +       DyDbgInc(cnt);                                                  \
9600 +       if (h_op->func) {                                               \
9601 +               if (src.func)                                           \
9602 +                       dst.func = src.func;                            \
9603 +               else                                                    \
9604 +                       AuDbg("%s %s\n", au_sbtype(h_sb), #func);       \
9605 +       }                                                               \
9606 +} while (0)
9607 +
9608 +#define DySetForce(func, dst, src) do {                \
9609 +       AuDebugOn(!src.func);                   \
9610 +       DyDbgInc(cnt);                          \
9611 +       dst.func = src.func;                    \
9612 +} while (0)
9613 +
9614 +#define DySetAop(func) \
9615 +       DySet(func, dyaop->da_op, aufs_aop, h_aop, h_sb)
9616 +#define DySetAopForce(func) \
9617 +       DySetForce(func, dyaop->da_op, aufs_aop)
9618 +
9619 +static void dy_aop(struct au_dykey *key, const void *h_op,
9620 +                  struct super_block *h_sb __maybe_unused)
9621 +{
9622 +       struct au_dyaop *dyaop = (void *)key;
9623 +       const struct address_space_operations *h_aop = h_op;
9624 +       DyDbgDeclare(cnt);
9625 +
9626 +       AuDbg("%s\n", au_sbtype(h_sb));
9627 +
9628 +       DySetAop(writepage);
9629 +       DySetAopForce(readpage);        /* force */
9630 +       DySetAop(writepages);
9631 +       DySetAop(set_page_dirty);
9632 +       DySetAop(readpages);
9633 +       DySetAop(write_begin);
9634 +       DySetAop(write_end);
9635 +       DySetAop(bmap);
9636 +       DySetAop(invalidatepage);
9637 +       DySetAop(releasepage);
9638 +       DySetAop(freepage);
9639 +       /* these two will be changed according to an aufs mount option */
9640 +       DySetAop(direct_IO);
9641 +       DySetAop(get_xip_mem);
9642 +       DySetAop(migratepage);
9643 +       DySetAop(launder_page);
9644 +       DySetAop(is_partially_uptodate);
9645 +       DySetAop(is_dirty_writeback);
9646 +       DySetAop(error_remove_page);
9647 +       DySetAop(swap_activate);
9648 +       DySetAop(swap_deactivate);
9649 +
9650 +       DyDbgSize(cnt, *h_aop);
9651 +       dyaop->da_get_xip_mem = h_aop->get_xip_mem;
9652 +}
9653 +
9654 +/* ---------------------------------------------------------------------- */
9655 +
9656 +static void dy_bug(struct kref *kref)
9657 +{
9658 +       BUG();
9659 +}
9660 +
9661 +static struct au_dykey *dy_get(struct au_dynop *op, struct au_branch *br)
9662 +{
9663 +       struct au_dykey *key, *old;
9664 +       struct au_splhead *spl;
9665 +       struct op {
9666 +               unsigned int sz;
9667 +               void (*set)(struct au_dykey *key, const void *h_op,
9668 +                           struct super_block *h_sb __maybe_unused);
9669 +       };
9670 +       static const struct op a[] = {
9671 +               [AuDy_AOP] = {
9672 +                       .sz     = sizeof(struct au_dyaop),
9673 +                       .set    = dy_aop
9674 +               }
9675 +       };
9676 +       const struct op *p;
9677 +
9678 +       spl = dynop + op->dy_type;
9679 +       key = dy_gfind_get(spl, op->dy_hop);
9680 +       if (key)
9681 +               goto out_add; /* success */
9682 +
9683 +       p = a + op->dy_type;
9684 +       key = kzalloc(p->sz, GFP_NOFS);
9685 +       if (unlikely(!key)) {
9686 +               key = ERR_PTR(-ENOMEM);
9687 +               goto out;
9688 +       }
9689 +
9690 +       key->dk_op.dy_hop = op->dy_hop;
9691 +       kref_init(&key->dk_kref);
9692 +       p->set(key, op->dy_hop, au_br_sb(br));
9693 +       old = dy_gadd(spl, key);
9694 +       if (old) {
9695 +               kfree(key);
9696 +               key = old;
9697 +       }
9698 +
9699 +out_add:
9700 +       old = dy_bradd(br, key);
9701 +       if (old)
9702 +               /* its ref-count should never be zero here */
9703 +               kref_put(&key->dk_kref, dy_bug);
9704 +out:
9705 +       return key;
9706 +}
9707 +
9708 +/* ---------------------------------------------------------------------- */
9709 +/*
9710 + * Aufs prohibits O_DIRECT by defaut even if the branch supports it.
9711 + * This behaviour is neccessary to return an error from open(O_DIRECT) instead
9712 + * of the succeeding I/O. The dio mount option enables O_DIRECT and makes
9713 + * open(O_DIRECT) always succeed, but the succeeding I/O may return an error.
9714 + * See the aufs manual in detail.
9715 + *
9716 + * To keep this behaviour, aufs has to set NULL to ->get_xip_mem too, and the
9717 + * performance of fadvise() and madvise() may be affected.
9718 + */
9719 +static void dy_adx(struct au_dyaop *dyaop, int do_dx)
9720 +{
9721 +       if (!do_dx) {
9722 +               dyaop->da_op.direct_IO = NULL;
9723 +               dyaop->da_op.get_xip_mem = NULL;
9724 +       } else {
9725 +               dyaop->da_op.direct_IO = aufs_aop.direct_IO;
9726 +               dyaop->da_op.get_xip_mem = aufs_aop.get_xip_mem;
9727 +               if (!dyaop->da_get_xip_mem)
9728 +                       dyaop->da_op.get_xip_mem = NULL;
9729 +       }
9730 +}
9731 +
9732 +static struct au_dyaop *dy_aget(struct au_branch *br,
9733 +                               const struct address_space_operations *h_aop,
9734 +                               int do_dx)
9735 +{
9736 +       struct au_dyaop *dyaop;
9737 +       struct au_dynop op;
9738 +
9739 +       op.dy_type = AuDy_AOP;
9740 +       op.dy_haop = h_aop;
9741 +       dyaop = (void *)dy_get(&op, br);
9742 +       if (IS_ERR(dyaop))
9743 +               goto out;
9744 +       dy_adx(dyaop, do_dx);
9745 +
9746 +out:
9747 +       return dyaop;
9748 +}
9749 +
9750 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
9751 +               struct inode *h_inode)
9752 +{
9753 +       int err, do_dx;
9754 +       struct super_block *sb;
9755 +       struct au_branch *br;
9756 +       struct au_dyaop *dyaop;
9757 +
9758 +       AuDebugOn(!S_ISREG(h_inode->i_mode));
9759 +       IiMustWriteLock(inode);
9760 +
9761 +       sb = inode->i_sb;
9762 +       br = au_sbr(sb, bindex);
9763 +       do_dx = !!au_opt_test(au_mntflags(sb), DIO);
9764 +       dyaop = dy_aget(br, h_inode->i_mapping->a_ops, do_dx);
9765 +       err = PTR_ERR(dyaop);
9766 +       if (IS_ERR(dyaop))
9767 +               /* unnecessary to call dy_fput() */
9768 +               goto out;
9769 +
9770 +       err = 0;
9771 +       inode->i_mapping->a_ops = &dyaop->da_op;
9772 +
9773 +out:
9774 +       return err;
9775 +}
9776 +
9777 +/*
9778 + * Is it safe to replace a_ops during the inode/file is in operation?
9779 + * Yes, I hope so.
9780 + */
9781 +int au_dy_irefresh(struct inode *inode)
9782 +{
9783 +       int err;
9784 +       aufs_bindex_t bstart;
9785 +       struct inode *h_inode;
9786 +
9787 +       err = 0;
9788 +       if (S_ISREG(inode->i_mode)) {
9789 +               bstart = au_ibstart(inode);
9790 +               h_inode = au_h_iptr(inode, bstart);
9791 +               err = au_dy_iaop(inode, bstart, h_inode);
9792 +       }
9793 +       return err;
9794 +}
9795 +
9796 +void au_dy_arefresh(int do_dx)
9797 +{
9798 +       struct au_splhead *spl;
9799 +       struct list_head *head;
9800 +       struct au_dykey *key;
9801 +
9802 +       spl = dynop + AuDy_AOP;
9803 +       head = &spl->head;
9804 +       spin_lock(&spl->spin);
9805 +       list_for_each_entry(key, head, dk_list)
9806 +               dy_adx((void *)key, do_dx);
9807 +       spin_unlock(&spl->spin);
9808 +}
9809 +
9810 +/* ---------------------------------------------------------------------- */
9811 +
9812 +void __init au_dy_init(void)
9813 +{
9814 +       int i;
9815 +
9816 +       /* make sure that 'struct au_dykey *' can be any type */
9817 +       BUILD_BUG_ON(offsetof(struct au_dyaop, da_key));
9818 +
9819 +       for (i = 0; i < AuDyLast; i++)
9820 +               au_spl_init(dynop + i);
9821 +}
9822 +
9823 +void au_dy_fin(void)
9824 +{
9825 +       int i;
9826 +
9827 +       for (i = 0; i < AuDyLast; i++)
9828 +               WARN_ON(!list_empty(&dynop[i].head));
9829 +}
9830 diff -urN /usr/share/empty/fs/aufs/dynop.h linux/fs/aufs/dynop.h
9831 --- /usr/share/empty/fs/aufs/dynop.h    1970-01-01 01:00:00.000000000 +0100
9832 +++ linux/fs/aufs/dynop.h       2014-04-24 22:11:10.851935747 +0200
9833 @@ -0,0 +1,75 @@
9834 +/*
9835 + * Copyright (C) 2010-2014 Junjiro R. Okajima
9836 + *
9837 + * This program, aufs is free software; you can redistribute it and/or modify
9838 + * it under the terms of the GNU General Public License as published by
9839 + * the Free Software Foundation; either version 2 of the License, or
9840 + * (at your option) any later version.
9841 + *
9842 + * This program is distributed in the hope that it will be useful,
9843 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9844 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9845 + * GNU General Public License for more details.
9846 + *
9847 + * You should have received a copy of the GNU General Public License
9848 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9849 + */
9850 +
9851 +/*
9852 + * dynamically customizable operations (for regular files only)
9853 + */
9854 +
9855 +#ifndef __AUFS_DYNOP_H__
9856 +#define __AUFS_DYNOP_H__
9857 +
9858 +#ifdef __KERNEL__
9859 +
9860 +#include "inode.h"
9861 +
9862 +enum {AuDy_AOP, AuDyLast};
9863 +
9864 +struct au_dynop {
9865 +       int                                             dy_type;
9866 +       union {
9867 +               const void                              *dy_hop;
9868 +               const struct address_space_operations   *dy_haop;
9869 +       };
9870 +};
9871 +
9872 +struct au_dykey {
9873 +       union {
9874 +               struct list_head        dk_list;
9875 +               struct rcu_head         dk_rcu;
9876 +       };
9877 +       struct au_dynop         dk_op;
9878 +
9879 +       /*
9880 +        * during I am in the branch local array, kref is gotten. when the
9881 +        * branch is removed, kref is put.
9882 +        */
9883 +       struct kref             dk_kref;
9884 +};
9885 +
9886 +/* stop unioning since their sizes are very different from each other */
9887 +struct au_dyaop {
9888 +       struct au_dykey                 da_key;
9889 +       struct address_space_operations da_op; /* not const */
9890 +       int (*da_get_xip_mem)(struct address_space *, pgoff_t, int,
9891 +                             void **, unsigned long *);
9892 +};
9893 +
9894 +/* ---------------------------------------------------------------------- */
9895 +
9896 +/* dynop.c */
9897 +struct au_branch;
9898 +void au_dy_put(struct au_dykey *key);
9899 +int au_dy_iaop(struct inode *inode, aufs_bindex_t bindex,
9900 +               struct inode *h_inode);
9901 +int au_dy_irefresh(struct inode *inode);
9902 +void au_dy_arefresh(int do_dio);
9903 +
9904 +void __init au_dy_init(void);
9905 +void au_dy_fin(void);
9906 +
9907 +#endif /* __KERNEL__ */
9908 +#endif /* __AUFS_DYNOP_H__ */
9909 diff -urN /usr/share/empty/fs/aufs/export.c linux/fs/aufs/export.c
9910 --- /usr/share/empty/fs/aufs/export.c   1970-01-01 01:00:00.000000000 +0100
9911 +++ linux/fs/aufs/export.c      2014-04-24 22:11:10.851935747 +0200
9912 @@ -0,0 +1,831 @@
9913 +/*
9914 + * Copyright (C) 2005-2014 Junjiro R. Okajima
9915 + *
9916 + * This program, aufs is free software; you can redistribute it and/or modify
9917 + * it under the terms of the GNU General Public License as published by
9918 + * the Free Software Foundation; either version 2 of the License, or
9919 + * (at your option) any later version.
9920 + *
9921 + * This program is distributed in the hope that it will be useful,
9922 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
9923 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
9924 + * GNU General Public License for more details.
9925 + *
9926 + * You should have received a copy of the GNU General Public License
9927 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
9928 + */
9929 +
9930 +/*
9931 + * export via nfs
9932 + */
9933 +
9934 +#include <linux/exportfs.h>
9935 +#include <linux/fs_struct.h>
9936 +#include <linux/namei.h>
9937 +#include <linux/nsproxy.h>
9938 +#include <linux/random.h>
9939 +#include <linux/writeback.h>
9940 +#include "../fs/mount.h"
9941 +#include "aufs.h"
9942 +
9943 +union conv {
9944 +#ifdef CONFIG_AUFS_INO_T_64
9945 +       __u32 a[2];
9946 +#else
9947 +       __u32 a[1];
9948 +#endif
9949 +       ino_t ino;
9950 +};
9951 +
9952 +static ino_t decode_ino(__u32 *a)
9953 +{
9954 +       union conv u;
9955 +
9956 +       BUILD_BUG_ON(sizeof(u.ino) != sizeof(u.a));
9957 +       u.a[0] = a[0];
9958 +#ifdef CONFIG_AUFS_INO_T_64
9959 +       u.a[1] = a[1];
9960 +#endif
9961 +       return u.ino;
9962 +}
9963 +
9964 +static void encode_ino(__u32 *a, ino_t ino)
9965 +{
9966 +       union conv u;
9967 +
9968 +       u.ino = ino;
9969 +       a[0] = u.a[0];
9970 +#ifdef CONFIG_AUFS_INO_T_64
9971 +       a[1] = u.a[1];
9972 +#endif
9973 +}
9974 +
9975 +/* NFS file handle */
9976 +enum {
9977 +       Fh_br_id,
9978 +       Fh_sigen,
9979 +#ifdef CONFIG_AUFS_INO_T_64
9980 +       /* support 64bit inode number */
9981 +       Fh_ino1,
9982 +       Fh_ino2,
9983 +       Fh_dir_ino1,
9984 +       Fh_dir_ino2,
9985 +#else
9986 +       Fh_ino1,
9987 +       Fh_dir_ino1,
9988 +#endif
9989 +       Fh_igen,
9990 +       Fh_h_type,
9991 +       Fh_tail,
9992 +
9993 +       Fh_ino = Fh_ino1,
9994 +       Fh_dir_ino = Fh_dir_ino1
9995 +};
9996 +
9997 +static int au_test_anon(struct dentry *dentry)
9998 +{
9999 +       /* note: read d_flags without d_lock */
10000 +       return !!(dentry->d_flags & DCACHE_DISCONNECTED);
10001 +}
10002 +
10003 +int au_test_nfsd(void)
10004 +{
10005 +       int ret;
10006 +       struct task_struct *tsk = current;
10007 +       char comm[sizeof(tsk->comm)];
10008 +
10009 +       ret = 0;
10010 +       if (tsk->flags & PF_KTHREAD) {
10011 +               get_task_comm(comm, tsk);
10012 +               ret = !strcmp(comm, "nfsd");
10013 +       }
10014 +
10015 +       return ret;
10016 +}
10017 +
10018 +/* ---------------------------------------------------------------------- */
10019 +/* inode generation external table */
10020 +
10021 +void au_xigen_inc(struct inode *inode)
10022 +{
10023 +       loff_t pos;
10024 +       ssize_t sz;
10025 +       __u32 igen;
10026 +       struct super_block *sb;
10027 +       struct au_sbinfo *sbinfo;
10028 +
10029 +       sb = inode->i_sb;
10030 +       AuDebugOn(!au_opt_test(au_mntflags(sb), XINO));
10031 +
10032 +       sbinfo = au_sbi(sb);
10033 +       pos = inode->i_ino;
10034 +       pos *= sizeof(igen);
10035 +       igen = inode->i_generation + 1;
10036 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xigen, &igen,
10037 +                        sizeof(igen), &pos);
10038 +       if (sz == sizeof(igen))
10039 +               return; /* success */
10040 +
10041 +       if (unlikely(sz >= 0))
10042 +               AuIOErr("xigen error (%zd)\n", sz);
10043 +}
10044 +
10045 +int au_xigen_new(struct inode *inode)
10046 +{
10047 +       int err;
10048 +       loff_t pos;
10049 +       ssize_t sz;
10050 +       struct super_block *sb;
10051 +       struct au_sbinfo *sbinfo;
10052 +       struct file *file;
10053 +
10054 +       err = 0;
10055 +       /* todo: dirty, at mount time */
10056 +       if (inode->i_ino == AUFS_ROOT_INO)
10057 +               goto out;
10058 +       sb = inode->i_sb;
10059 +       SiMustAnyLock(sb);
10060 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
10061 +               goto out;
10062 +
10063 +       err = -EFBIG;
10064 +       pos = inode->i_ino;
10065 +       if (unlikely(au_loff_max / sizeof(inode->i_generation) - 1 < pos)) {
10066 +               AuIOErr1("too large i%lld\n", pos);
10067 +               goto out;
10068 +       }
10069 +       pos *= sizeof(inode->i_generation);
10070 +
10071 +       err = 0;
10072 +       sbinfo = au_sbi(sb);
10073 +       file = sbinfo->si_xigen;
10074 +       BUG_ON(!file);
10075 +
10076 +       if (vfsub_f_size_read(file)
10077 +           < pos + sizeof(inode->i_generation)) {
10078 +               inode->i_generation = atomic_inc_return(&sbinfo->si_xigen_next);
10079 +               sz = xino_fwrite(sbinfo->si_xwrite, file, &inode->i_generation,
10080 +                                sizeof(inode->i_generation), &pos);
10081 +       } else
10082 +               sz = xino_fread(sbinfo->si_xread, file, &inode->i_generation,
10083 +                               sizeof(inode->i_generation), &pos);
10084 +       if (sz == sizeof(inode->i_generation))
10085 +               goto out; /* success */
10086 +
10087 +       err = sz;
10088 +       if (unlikely(sz >= 0)) {
10089 +               err = -EIO;
10090 +               AuIOErr("xigen error (%zd)\n", sz);
10091 +       }
10092 +
10093 +out:
10094 +       return err;
10095 +}
10096 +
10097 +int au_xigen_set(struct super_block *sb, struct file *base)
10098 +{
10099 +       int err;
10100 +       struct au_sbinfo *sbinfo;
10101 +       struct file *file;
10102 +
10103 +       SiMustWriteLock(sb);
10104 +
10105 +       sbinfo = au_sbi(sb);
10106 +       file = au_xino_create2(base, sbinfo->si_xigen);
10107 +       err = PTR_ERR(file);
10108 +       if (IS_ERR(file))
10109 +               goto out;
10110 +       err = 0;
10111 +       if (sbinfo->si_xigen)
10112 +               fput(sbinfo->si_xigen);
10113 +       sbinfo->si_xigen = file;
10114 +
10115 +out:
10116 +       return err;
10117 +}
10118 +
10119 +void au_xigen_clr(struct super_block *sb)
10120 +{
10121 +       struct au_sbinfo *sbinfo;
10122 +
10123 +       SiMustWriteLock(sb);
10124 +
10125 +       sbinfo = au_sbi(sb);
10126 +       if (sbinfo->si_xigen) {
10127 +               fput(sbinfo->si_xigen);
10128 +               sbinfo->si_xigen = NULL;
10129 +       }
10130 +}
10131 +
10132 +/* ---------------------------------------------------------------------- */
10133 +
10134 +static struct dentry *decode_by_ino(struct super_block *sb, ino_t ino,
10135 +                                   ino_t dir_ino)
10136 +{
10137 +       struct dentry *dentry, *d;
10138 +       struct inode *inode;
10139 +       unsigned int sigen;
10140 +
10141 +       dentry = NULL;
10142 +       inode = ilookup(sb, ino);
10143 +       if (!inode)
10144 +               goto out;
10145 +
10146 +       dentry = ERR_PTR(-ESTALE);
10147 +       sigen = au_sigen(sb);
10148 +       if (unlikely(is_bad_inode(inode)
10149 +                    || IS_DEADDIR(inode)
10150 +                    || sigen != au_iigen(inode, NULL)))
10151 +               goto out_iput;
10152 +
10153 +       dentry = NULL;
10154 +       if (!dir_ino || S_ISDIR(inode->i_mode))
10155 +               dentry = d_find_alias(inode);
10156 +       else {
10157 +               spin_lock(&inode->i_lock);
10158 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
10159 +                       spin_lock(&d->d_lock);
10160 +                       if (!au_test_anon(d)
10161 +                           && d->d_parent->d_inode->i_ino == dir_ino) {
10162 +                               dentry = dget_dlock(d);
10163 +                               spin_unlock(&d->d_lock);
10164 +                               break;
10165 +                       }
10166 +                       spin_unlock(&d->d_lock);
10167 +               }
10168 +               spin_unlock(&inode->i_lock);
10169 +       }
10170 +       if (unlikely(dentry && au_digen_test(dentry, sigen))) {
10171 +               /* need to refresh */
10172 +               dput(dentry);
10173 +               dentry = NULL;
10174 +       }
10175 +
10176 +out_iput:
10177 +       iput(inode);
10178 +out:
10179 +       AuTraceErrPtr(dentry);
10180 +       return dentry;
10181 +}
10182 +
10183 +/* ---------------------------------------------------------------------- */
10184 +
10185 +/* todo: dirty? */
10186 +/* if exportfs_decode_fh() passed vfsmount*, we could be happy */
10187 +
10188 +struct au_compare_mnt_args {
10189 +       /* input */
10190 +       struct super_block *sb;
10191 +
10192 +       /* output */
10193 +       struct vfsmount *mnt;
10194 +};
10195 +
10196 +static int au_compare_mnt(struct vfsmount *mnt, void *arg)
10197 +{
10198 +       struct au_compare_mnt_args *a = arg;
10199 +
10200 +       if (mnt->mnt_sb != a->sb)
10201 +               return 0;
10202 +       a->mnt = mntget(mnt);
10203 +       return 1;
10204 +}
10205 +
10206 +static struct vfsmount *au_mnt_get(struct super_block *sb)
10207 +{
10208 +       int err;
10209 +       struct path root;
10210 +       struct au_compare_mnt_args args = {
10211 +               .sb = sb
10212 +       };
10213 +
10214 +       get_fs_root(current->fs, &root);
10215 +       rcu_read_lock();
10216 +       err = iterate_mounts(au_compare_mnt, &args, root.mnt);
10217 +       rcu_read_unlock();
10218 +       path_put(&root);
10219 +       AuDebugOn(!err);
10220 +       AuDebugOn(!args.mnt);
10221 +       return args.mnt;
10222 +}
10223 +
10224 +struct au_nfsd_si_lock {
10225 +       unsigned int sigen;
10226 +       aufs_bindex_t bindex, br_id;
10227 +       unsigned char force_lock;
10228 +};
10229 +
10230 +static int si_nfsd_read_lock(struct super_block *sb,
10231 +                            struct au_nfsd_si_lock *nsi_lock)
10232 +{
10233 +       int err;
10234 +       aufs_bindex_t bindex;
10235 +
10236 +       si_read_lock(sb, AuLock_FLUSH);
10237 +
10238 +       /* branch id may be wrapped around */
10239 +       err = 0;
10240 +       bindex = au_br_index(sb, nsi_lock->br_id);
10241 +       if (bindex >= 0 && nsi_lock->sigen + AUFS_BRANCH_MAX > au_sigen(sb))
10242 +               goto out; /* success */
10243 +
10244 +       err = -ESTALE;
10245 +       bindex = -1;
10246 +       if (!nsi_lock->force_lock)
10247 +               si_read_unlock(sb);
10248 +
10249 +out:
10250 +       nsi_lock->bindex = bindex;
10251 +       return err;
10252 +}
10253 +
10254 +struct find_name_by_ino {
10255 +       struct dir_context ctx;
10256 +       int called, found;
10257 +       ino_t ino;
10258 +       char *name;
10259 +       int namelen;
10260 +};
10261 +
10262 +static int
10263 +find_name_by_ino(struct dir_context *ctx, const char *name, int namelen,
10264 +                loff_t offset, u64 ino, unsigned int d_type)
10265 +{
10266 +       struct find_name_by_ino *a = container_of(ctx, struct find_name_by_ino,
10267 +                                                 ctx);
10268 +
10269 +       a->called++;
10270 +       if (a->ino != ino)
10271 +               return 0;
10272 +
10273 +       memcpy(a->name, name, namelen);
10274 +       a->namelen = namelen;
10275 +       a->found = 1;
10276 +       return 1;
10277 +}
10278 +
10279 +static struct dentry *au_lkup_by_ino(struct path *path, ino_t ino,
10280 +                                    struct au_nfsd_si_lock *nsi_lock)
10281 +{
10282 +       struct dentry *dentry, *parent;
10283 +       struct file *file;
10284 +       struct inode *dir;
10285 +       struct find_name_by_ino arg = {
10286 +               .ctx = {
10287 +                       .actor = au_diractor(find_name_by_ino)
10288 +               }
10289 +       };
10290 +       int err;
10291 +
10292 +       parent = path->dentry;
10293 +       if (nsi_lock)
10294 +               si_read_unlock(parent->d_sb);
10295 +       file = vfsub_dentry_open(path, au_dir_roflags);
10296 +       dentry = (void *)file;
10297 +       if (IS_ERR(file))
10298 +               goto out;
10299 +
10300 +       dentry = ERR_PTR(-ENOMEM);
10301 +       arg.name = (void *)__get_free_page(GFP_NOFS);
10302 +       if (unlikely(!arg.name))
10303 +               goto out_file;
10304 +       arg.ino = ino;
10305 +       arg.found = 0;
10306 +       do {
10307 +               arg.called = 0;
10308 +               /* smp_mb(); */
10309 +               err = vfsub_iterate_dir(file, &arg.ctx);
10310 +       } while (!err && !arg.found && arg.called);
10311 +       dentry = ERR_PTR(err);
10312 +       if (unlikely(err))
10313 +               goto out_name;
10314 +       /* instead of ENOENT */
10315 +       dentry = ERR_PTR(-ESTALE);
10316 +       if (!arg.found)
10317 +               goto out_name;
10318 +
10319 +       /* do not call vfsub_lkup_one() */
10320 +       dir = parent->d_inode;
10321 +       mutex_lock(&dir->i_mutex);
10322 +       dentry = vfsub_lookup_one_len(arg.name, parent, arg.namelen);
10323 +       mutex_unlock(&dir->i_mutex);
10324 +       AuTraceErrPtr(dentry);
10325 +       if (IS_ERR(dentry))
10326 +               goto out_name;
10327 +       AuDebugOn(au_test_anon(dentry));
10328 +       if (unlikely(!dentry->d_inode)) {
10329 +               dput(dentry);
10330 +               dentry = ERR_PTR(-ENOENT);
10331 +       }
10332 +
10333 +out_name:
10334 +       free_page((unsigned long)arg.name);
10335 +out_file:
10336 +       fput(file);
10337 +out:
10338 +       if (unlikely(nsi_lock
10339 +                    && si_nfsd_read_lock(parent->d_sb, nsi_lock) < 0))
10340 +               if (!IS_ERR(dentry)) {
10341 +                       dput(dentry);
10342 +                       dentry = ERR_PTR(-ESTALE);
10343 +               }
10344 +       AuTraceErrPtr(dentry);
10345 +       return dentry;
10346 +}
10347 +
10348 +static struct dentry *decode_by_dir_ino(struct super_block *sb, ino_t ino,
10349 +                                       ino_t dir_ino,
10350 +                                       struct au_nfsd_si_lock *nsi_lock)
10351 +{
10352 +       struct dentry *dentry;
10353 +       struct path path;
10354 +
10355 +       if (dir_ino != AUFS_ROOT_INO) {
10356 +               path.dentry = decode_by_ino(sb, dir_ino, 0);
10357 +               dentry = path.dentry;
10358 +               if (!path.dentry || IS_ERR(path.dentry))
10359 +                       goto out;
10360 +               AuDebugOn(au_test_anon(path.dentry));
10361 +       } else
10362 +               path.dentry = dget(sb->s_root);
10363 +
10364 +       path.mnt = au_mnt_get(sb);
10365 +       dentry = au_lkup_by_ino(&path, ino, nsi_lock);
10366 +       path_put(&path);
10367 +
10368 +out:
10369 +       AuTraceErrPtr(dentry);
10370 +       return dentry;
10371 +}
10372 +
10373 +/* ---------------------------------------------------------------------- */
10374 +
10375 +static int h_acceptable(void *expv, struct dentry *dentry)
10376 +{
10377 +       return 1;
10378 +}
10379 +
10380 +static char *au_build_path(struct dentry *h_parent, struct path *h_rootpath,
10381 +                          char *buf, int len, struct super_block *sb)
10382 +{
10383 +       char *p;
10384 +       int n;
10385 +       struct path path;
10386 +
10387 +       p = d_path(h_rootpath, buf, len);
10388 +       if (IS_ERR(p))
10389 +               goto out;
10390 +       n = strlen(p);
10391 +
10392 +       path.mnt = h_rootpath->mnt;
10393 +       path.dentry = h_parent;
10394 +       p = d_path(&path, buf, len);
10395 +       if (IS_ERR(p))
10396 +               goto out;
10397 +       if (n != 1)
10398 +               p += n;
10399 +
10400 +       path.mnt = au_mnt_get(sb);
10401 +       path.dentry = sb->s_root;
10402 +       p = d_path(&path, buf, len - strlen(p));
10403 +       mntput(path.mnt);
10404 +       if (IS_ERR(p))
10405 +               goto out;
10406 +       if (n != 1)
10407 +               p[strlen(p)] = '/';
10408 +
10409 +out:
10410 +       AuTraceErrPtr(p);
10411 +       return p;
10412 +}
10413 +
10414 +static
10415 +struct dentry *decode_by_path(struct super_block *sb, ino_t ino, __u32 *fh,
10416 +                             int fh_len, struct au_nfsd_si_lock *nsi_lock)
10417 +{
10418 +       struct dentry *dentry, *h_parent, *root;
10419 +       struct super_block *h_sb;
10420 +       char *pathname, *p;
10421 +       struct vfsmount *h_mnt;
10422 +       struct au_branch *br;
10423 +       int err;
10424 +       struct path path;
10425 +
10426 +       br = au_sbr(sb, nsi_lock->bindex);
10427 +       h_mnt = au_br_mnt(br);
10428 +       h_sb = h_mnt->mnt_sb;
10429 +       /* todo: call lower fh_to_dentry()? fh_to_parent()? */
10430 +       h_parent = exportfs_decode_fh(h_mnt, (void *)(fh + Fh_tail),
10431 +                                     fh_len - Fh_tail, fh[Fh_h_type],
10432 +                                     h_acceptable, /*context*/NULL);
10433 +       dentry = h_parent;
10434 +       if (unlikely(!h_parent || IS_ERR(h_parent))) {
10435 +               AuWarn1("%s decode_fh failed, %ld\n",
10436 +                       au_sbtype(h_sb), PTR_ERR(h_parent));
10437 +               goto out;
10438 +       }
10439 +       dentry = NULL;
10440 +       if (unlikely(au_test_anon(h_parent))) {
10441 +               AuWarn1("%s decode_fh returned a disconnected dentry\n",
10442 +                       au_sbtype(h_sb));
10443 +               goto out_h_parent;
10444 +       }
10445 +
10446 +       dentry = ERR_PTR(-ENOMEM);
10447 +       pathname = (void *)__get_free_page(GFP_NOFS);
10448 +       if (unlikely(!pathname))
10449 +               goto out_h_parent;
10450 +
10451 +       root = sb->s_root;
10452 +       path.mnt = h_mnt;
10453 +       di_read_lock_parent(root, !AuLock_IR);
10454 +       path.dentry = au_h_dptr(root, nsi_lock->bindex);
10455 +       di_read_unlock(root, !AuLock_IR);
10456 +       p = au_build_path(h_parent, &path, pathname, PAGE_SIZE, sb);
10457 +       dentry = (void *)p;
10458 +       if (IS_ERR(p))
10459 +               goto out_pathname;
10460 +
10461 +       si_read_unlock(sb);
10462 +       err = vfsub_kern_path(p, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &path);
10463 +       dentry = ERR_PTR(err);
10464 +       if (unlikely(err))
10465 +               goto out_relock;
10466 +
10467 +       dentry = ERR_PTR(-ENOENT);
10468 +       AuDebugOn(au_test_anon(path.dentry));
10469 +       if (unlikely(!path.dentry->d_inode))
10470 +               goto out_path;
10471 +
10472 +       if (ino != path.dentry->d_inode->i_ino)
10473 +               dentry = au_lkup_by_ino(&path, ino, /*nsi_lock*/NULL);
10474 +       else
10475 +               dentry = dget(path.dentry);
10476 +
10477 +out_path:
10478 +       path_put(&path);
10479 +out_relock:
10480 +       if (unlikely(si_nfsd_read_lock(sb, nsi_lock) < 0))
10481 +               if (!IS_ERR(dentry)) {
10482 +                       dput(dentry);
10483 +                       dentry = ERR_PTR(-ESTALE);
10484 +               }
10485 +out_pathname:
10486 +       free_page((unsigned long)pathname);
10487 +out_h_parent:
10488 +       dput(h_parent);
10489 +out:
10490 +       AuTraceErrPtr(dentry);
10491 +       return dentry;
10492 +}
10493 +
10494 +/* ---------------------------------------------------------------------- */
10495 +
10496 +static struct dentry *
10497 +aufs_fh_to_dentry(struct super_block *sb, struct fid *fid, int fh_len,
10498 +                 int fh_type)
10499 +{
10500 +       struct dentry *dentry;
10501 +       __u32 *fh = fid->raw;
10502 +       struct au_branch *br;
10503 +       ino_t ino, dir_ino;
10504 +       struct au_nfsd_si_lock nsi_lock = {
10505 +               .force_lock     = 0
10506 +       };
10507 +
10508 +       dentry = ERR_PTR(-ESTALE);
10509 +       /* it should never happen, but the file handle is unreliable */
10510 +       if (unlikely(fh_len < Fh_tail))
10511 +               goto out;
10512 +       nsi_lock.sigen = fh[Fh_sigen];
10513 +       nsi_lock.br_id = fh[Fh_br_id];
10514 +
10515 +       /* branch id may be wrapped around */
10516 +       br = NULL;
10517 +       if (unlikely(si_nfsd_read_lock(sb, &nsi_lock)))
10518 +               goto out;
10519 +       nsi_lock.force_lock = 1;
10520 +
10521 +       /* is this inode still cached? */
10522 +       ino = decode_ino(fh + Fh_ino);
10523 +       /* it should never happen */
10524 +       if (unlikely(ino == AUFS_ROOT_INO))
10525 +               goto out;
10526 +
10527 +       dir_ino = decode_ino(fh + Fh_dir_ino);
10528 +       dentry = decode_by_ino(sb, ino, dir_ino);
10529 +       if (IS_ERR(dentry))
10530 +               goto out_unlock;
10531 +       if (dentry)
10532 +               goto accept;
10533 +
10534 +       /* is the parent dir cached? */
10535 +       br = au_sbr(sb, nsi_lock.bindex);
10536 +       atomic_inc(&br->br_count);
10537 +       dentry = decode_by_dir_ino(sb, ino, dir_ino, &nsi_lock);
10538 +       if (IS_ERR(dentry))
10539 +               goto out_unlock;
10540 +       if (dentry)
10541 +               goto accept;
10542 +
10543 +       /* lookup path */
10544 +       dentry = decode_by_path(sb, ino, fh, fh_len, &nsi_lock);
10545 +       if (IS_ERR(dentry))
10546 +               goto out_unlock;
10547 +       if (unlikely(!dentry))
10548 +               /* todo?: make it ESTALE */
10549 +               goto out_unlock;
10550 +
10551 +accept:
10552 +       if (!au_digen_test(dentry, au_sigen(sb))
10553 +           && dentry->d_inode->i_generation == fh[Fh_igen])
10554 +               goto out_unlock; /* success */
10555 +
10556 +       dput(dentry);
10557 +       dentry = ERR_PTR(-ESTALE);
10558 +out_unlock:
10559 +       if (br)
10560 +               atomic_dec(&br->br_count);
10561 +       si_read_unlock(sb);
10562 +out:
10563 +       AuTraceErrPtr(dentry);
10564 +       return dentry;
10565 +}
10566 +
10567 +#if 0 /* reserved for future use */
10568 +/* support subtreecheck option */
10569 +static struct dentry *aufs_fh_to_parent(struct super_block *sb, struct fid *fid,
10570 +                                       int fh_len, int fh_type)
10571 +{
10572 +       struct dentry *parent;
10573 +       __u32 *fh = fid->raw;
10574 +       ino_t dir_ino;
10575 +
10576 +       dir_ino = decode_ino(fh + Fh_dir_ino);
10577 +       parent = decode_by_ino(sb, dir_ino, 0);
10578 +       if (IS_ERR(parent))
10579 +               goto out;
10580 +       if (!parent)
10581 +               parent = decode_by_path(sb, au_br_index(sb, fh[Fh_br_id]),
10582 +                                       dir_ino, fh, fh_len);
10583 +
10584 +out:
10585 +       AuTraceErrPtr(parent);
10586 +       return parent;
10587 +}
10588 +#endif
10589 +
10590 +/* ---------------------------------------------------------------------- */
10591 +
10592 +static int aufs_encode_fh(struct inode *inode, __u32 *fh, int *max_len,
10593 +                         struct inode *dir)
10594 +{
10595 +       int err;
10596 +       aufs_bindex_t bindex;
10597 +       struct super_block *sb, *h_sb;
10598 +       struct dentry *dentry, *parent, *h_parent;
10599 +       struct inode *h_dir;
10600 +       struct au_branch *br;
10601 +
10602 +       err = -ENOSPC;
10603 +       if (unlikely(*max_len <= Fh_tail)) {
10604 +               AuWarn1("NFSv2 client (max_len %d)?\n", *max_len);
10605 +               goto out;
10606 +       }
10607 +
10608 +       err = FILEID_ROOT;
10609 +       if (inode->i_ino == AUFS_ROOT_INO) {
10610 +               AuDebugOn(inode->i_ino != AUFS_ROOT_INO);
10611 +               goto out;
10612 +       }
10613 +
10614 +       h_parent = NULL;
10615 +       sb = inode->i_sb;
10616 +       err = si_read_lock(sb, AuLock_FLUSH);
10617 +       if (unlikely(err))
10618 +               goto out;
10619 +
10620 +#ifdef CONFIG_AUFS_DEBUG
10621 +       if (unlikely(!au_opt_test(au_mntflags(sb), XINO)))
10622 +               AuWarn1("NFS-exporting requires xino\n");
10623 +#endif
10624 +       err = -EIO;
10625 +       parent = NULL;
10626 +       ii_read_lock_child(inode);
10627 +       bindex = au_ibstart(inode);
10628 +       if (!dir) {
10629 +               dentry = d_find_alias(inode);
10630 +               if (unlikely(!dentry))
10631 +                       goto out_unlock;
10632 +               AuDebugOn(au_test_anon(dentry));
10633 +               parent = dget_parent(dentry);
10634 +               dput(dentry);
10635 +               if (unlikely(!parent))
10636 +                       goto out_unlock;
10637 +               dir = parent->d_inode;
10638 +       }
10639 +
10640 +       ii_read_lock_parent(dir);
10641 +       h_dir = au_h_iptr(dir, bindex);
10642 +       ii_read_unlock(dir);
10643 +       if (unlikely(!h_dir))
10644 +               goto out_parent;
10645 +       h_parent = d_find_alias(h_dir);
10646 +       if (unlikely(!h_parent))
10647 +               goto out_hparent;
10648 +
10649 +       err = -EPERM;
10650 +       br = au_sbr(sb, bindex);
10651 +       h_sb = au_br_sb(br);
10652 +       if (unlikely(!h_sb->s_export_op)) {
10653 +               AuErr1("%s branch is not exportable\n", au_sbtype(h_sb));
10654 +               goto out_hparent;
10655 +       }
10656 +
10657 +       fh[Fh_br_id] = br->br_id;
10658 +       fh[Fh_sigen] = au_sigen(sb);
10659 +       encode_ino(fh + Fh_ino, inode->i_ino);
10660 +       encode_ino(fh + Fh_dir_ino, dir->i_ino);
10661 +       fh[Fh_igen] = inode->i_generation;
10662 +
10663 +       *max_len -= Fh_tail;
10664 +       fh[Fh_h_type] = exportfs_encode_fh(h_parent, (void *)(fh + Fh_tail),
10665 +                                          max_len,
10666 +                                          /*connectable or subtreecheck*/0);
10667 +       err = fh[Fh_h_type];
10668 +       *max_len += Fh_tail;
10669 +       /* todo: macros? */
10670 +       if (err != FILEID_INVALID)
10671 +               err = 99;
10672 +       else
10673 +               AuWarn1("%s encode_fh failed\n", au_sbtype(h_sb));
10674 +
10675 +out_hparent:
10676 +       dput(h_parent);
10677 +out_parent:
10678 +       dput(parent);
10679 +out_unlock:
10680 +       ii_read_unlock(inode);
10681 +       si_read_unlock(sb);
10682 +out:
10683 +       if (unlikely(err < 0))
10684 +               err = FILEID_INVALID;
10685 +       return err;
10686 +}
10687 +
10688 +/* ---------------------------------------------------------------------- */
10689 +
10690 +static int aufs_commit_metadata(struct inode *inode)
10691 +{
10692 +       int err;
10693 +       aufs_bindex_t bindex;
10694 +       struct super_block *sb;
10695 +       struct inode *h_inode;
10696 +       int (*f)(struct inode *inode);
10697 +
10698 +       sb = inode->i_sb;
10699 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
10700 +       ii_write_lock_child(inode);
10701 +       bindex = au_ibstart(inode);
10702 +       AuDebugOn(bindex < 0);
10703 +       h_inode = au_h_iptr(inode, bindex);
10704 +
10705 +       f = h_inode->i_sb->s_export_op->commit_metadata;
10706 +       if (f)
10707 +               err = f(h_inode);
10708 +       else {
10709 +               struct writeback_control wbc = {
10710 +                       .sync_mode      = WB_SYNC_ALL,
10711 +                       .nr_to_write    = 0 /* metadata only */
10712 +               };
10713 +
10714 +               err = sync_inode(h_inode, &wbc);
10715 +       }
10716 +
10717 +       au_cpup_attr_timesizes(inode);
10718 +       ii_write_unlock(inode);
10719 +       si_read_unlock(sb);
10720 +       return err;
10721 +}
10722 +
10723 +/* ---------------------------------------------------------------------- */
10724 +
10725 +static struct export_operations aufs_export_op = {
10726 +       .fh_to_dentry           = aufs_fh_to_dentry,
10727 +       /* .fh_to_parent        = aufs_fh_to_parent, */
10728 +       .encode_fh              = aufs_encode_fh,
10729 +       .commit_metadata        = aufs_commit_metadata
10730 +};
10731 +
10732 +void au_export_init(struct super_block *sb)
10733 +{
10734 +       struct au_sbinfo *sbinfo;
10735 +       __u32 u;
10736 +
10737 +       sb->s_export_op = &aufs_export_op;
10738 +       sbinfo = au_sbi(sb);
10739 +       sbinfo->si_xigen = NULL;
10740 +       get_random_bytes(&u, sizeof(u));
10741 +       BUILD_BUG_ON(sizeof(u) != sizeof(int));
10742 +       atomic_set(&sbinfo->si_xigen_next, u);
10743 +}
10744 diff -urN /usr/share/empty/fs/aufs/file.c linux/fs/aufs/file.c
10745 --- /usr/share/empty/fs/aufs/file.c     1970-01-01 01:00:00.000000000 +0100
10746 +++ linux/fs/aufs/file.c        2014-04-24 22:11:10.851935747 +0200
10747 @@ -0,0 +1,724 @@
10748 +/*
10749 + * Copyright (C) 2005-2014 Junjiro R. Okajima
10750 + *
10751 + * This program, aufs is free software; you can redistribute it and/or modify
10752 + * it under the terms of the GNU General Public License as published by
10753 + * the Free Software Foundation; either version 2 of the License, or
10754 + * (at your option) any later version.
10755 + *
10756 + * This program is distributed in the hope that it will be useful,
10757 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
10758 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
10759 + * GNU General Public License for more details.
10760 + *
10761 + * You should have received a copy of the GNU General Public License
10762 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
10763 + */
10764 +
10765 +/*
10766 + * handling file/dir, and address_space operation
10767 + */
10768 +
10769 +#ifdef CONFIG_AUFS_DEBUG
10770 +#include <linux/migrate.h>
10771 +#endif
10772 +#include <linux/pagemap.h>
10773 +#include "aufs.h"
10774 +
10775 +/* drop flags for writing */
10776 +unsigned int au_file_roflags(unsigned int flags)
10777 +{
10778 +       flags &= ~(O_WRONLY | O_RDWR | O_APPEND | O_CREAT | O_TRUNC);
10779 +       flags |= O_RDONLY | O_NOATIME;
10780 +       return flags;
10781 +}
10782 +
10783 +/* common functions to regular file and dir */
10784 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
10785 +                      struct file *file, int force_wr)
10786 +{
10787 +       struct file *h_file;
10788 +       struct dentry *h_dentry;
10789 +       struct inode *h_inode;
10790 +       struct super_block *sb;
10791 +       struct au_branch *br;
10792 +       struct path h_path;
10793 +       int err, exec_flag;
10794 +
10795 +       /* a race condition can happen between open and unlink/rmdir */
10796 +       h_file = ERR_PTR(-ENOENT);
10797 +       h_dentry = au_h_dptr(dentry, bindex);
10798 +       if (au_test_nfsd() && !h_dentry)
10799 +               goto out;
10800 +       h_inode = h_dentry->d_inode;
10801 +       if (au_test_nfsd() && !h_inode)
10802 +               goto out;
10803 +       spin_lock(&h_dentry->d_lock);
10804 +       err = (!d_unhashed(dentry) && d_unlinked(h_dentry))
10805 +               || !h_inode
10806 +               /* || !dentry->d_inode->i_nlink */
10807 +               ;
10808 +       spin_unlock(&h_dentry->d_lock);
10809 +       if (unlikely(err))
10810 +               goto out;
10811 +
10812 +       sb = dentry->d_sb;
10813 +       br = au_sbr(sb, bindex);
10814 +       h_file = ERR_PTR(-EACCES);
10815 +       exec_flag = flags & __FMODE_EXEC;
10816 +       if (exec_flag && (au_br_mnt(br)->mnt_flags & MNT_NOEXEC))
10817 +               goto out;
10818 +
10819 +       /* drop flags for writing */
10820 +       if (au_test_ro(sb, bindex, dentry->d_inode)) {
10821 +               if (force_wr && !(flags & O_WRONLY))
10822 +                       force_wr = 0;
10823 +               flags = au_file_roflags(flags);
10824 +               if (force_wr) {
10825 +                       h_file = ERR_PTR(-EROFS);
10826 +                       flags = au_file_roflags(flags);
10827 +                       if (unlikely(vfsub_native_ro(h_inode)
10828 +                                    || IS_APPEND(h_inode)))
10829 +                               goto out;
10830 +                       flags &= ~O_ACCMODE;
10831 +                       flags |= O_WRONLY;
10832 +               }
10833 +       }
10834 +       flags &= ~O_CREAT;
10835 +       atomic_inc(&br->br_count);
10836 +       h_path.dentry = h_dentry;
10837 +       h_path.mnt = au_br_mnt(br);
10838 +       if (!au_special_file(h_inode->i_mode))
10839 +               h_file = vfsub_dentry_open(&h_path, flags);
10840 +       else {
10841 +               /* this block depends upon the configuration */
10842 +               di_read_unlock(dentry, AuLock_IR);
10843 +               fi_write_unlock(file);
10844 +               si_read_unlock(sb);
10845 +               h_file = vfsub_dentry_open(&h_path, flags);
10846 +               si_noflush_read_lock(sb);
10847 +               fi_write_lock(file);
10848 +               di_read_lock_child(dentry, AuLock_IR);
10849 +       }
10850 +       if (IS_ERR(h_file))
10851 +               goto out_br;
10852 +
10853 +       if (exec_flag) {
10854 +               err = deny_write_access(h_file);
10855 +               if (unlikely(err)) {
10856 +                       fput(h_file);
10857 +                       h_file = ERR_PTR(err);
10858 +                       goto out_br;
10859 +               }
10860 +       }
10861 +       fsnotify_open(h_file);
10862 +       goto out; /* success */
10863 +
10864 +out_br:
10865 +       atomic_dec(&br->br_count);
10866 +out:
10867 +       return h_file;
10868 +}
10869 +
10870 +int au_do_open(struct file *file, int (*open)(struct file *file, int flags),
10871 +              struct au_fidir *fidir)
10872 +{
10873 +       int err;
10874 +       struct dentry *dentry;
10875 +
10876 +       err = au_finfo_init(file, fidir);
10877 +       if (unlikely(err))
10878 +               goto out;
10879 +
10880 +       dentry = file->f_dentry;
10881 +       di_read_lock_child(dentry, AuLock_IR);
10882 +       err = open(file, vfsub_file_flags(file));
10883 +       di_read_unlock(dentry, AuLock_IR);
10884 +
10885 +       fi_write_unlock(file);
10886 +       if (unlikely(err)) {
10887 +               au_fi(file)->fi_hdir = NULL;
10888 +               au_finfo_fin(file);
10889 +       }
10890 +
10891 +out:
10892 +       return err;
10893 +}
10894 +
10895 +int au_reopen_nondir(struct file *file)
10896 +{
10897 +       int err;
10898 +       aufs_bindex_t bstart;
10899 +       struct dentry *dentry;
10900 +       struct file *h_file, *h_file_tmp;
10901 +
10902 +       dentry = file->f_dentry;
10903 +       AuDebugOn(au_special_file(dentry->d_inode->i_mode));
10904 +       bstart = au_dbstart(dentry);
10905 +       h_file_tmp = NULL;
10906 +       if (au_fbstart(file) == bstart) {
10907 +               h_file = au_hf_top(file);
10908 +               if (file->f_mode == h_file->f_mode)
10909 +                       return 0; /* success */
10910 +               h_file_tmp = h_file;
10911 +               get_file(h_file_tmp);
10912 +               au_set_h_fptr(file, bstart, NULL);
10913 +       }
10914 +       AuDebugOn(au_fi(file)->fi_hdir);
10915 +       /*
10916 +        * it can happen
10917 +        * file exists on both of rw and ro
10918 +        * open --> dbstart and fbstart are both 0
10919 +        * prepend a branch as rw, "rw" become ro
10920 +        * remove rw/file
10921 +        * delete the top branch, "rw" becomes rw again
10922 +        *      --> dbstart is 1, fbstart is still 0
10923 +        * write --> fbstart is 0 but dbstart is 1
10924 +        */
10925 +       /* AuDebugOn(au_fbstart(file) < bstart); */
10926 +
10927 +       h_file = au_h_open(dentry, bstart, vfsub_file_flags(file) & ~O_TRUNC,
10928 +                          file, /*force_wr*/0);
10929 +       err = PTR_ERR(h_file);
10930 +       if (IS_ERR(h_file)) {
10931 +               if (h_file_tmp) {
10932 +                       atomic_inc(&au_sbr(dentry->d_sb, bstart)->br_count);
10933 +                       au_set_h_fptr(file, bstart, h_file_tmp);
10934 +                       h_file_tmp = NULL;
10935 +               }
10936 +               goto out; /* todo: close all? */
10937 +       }
10938 +
10939 +       err = 0;
10940 +       au_set_fbstart(file, bstart);
10941 +       au_set_h_fptr(file, bstart, h_file);
10942 +       au_update_figen(file);
10943 +       /* todo: necessary? */
10944 +       /* file->f_ra = h_file->f_ra; */
10945 +
10946 +out:
10947 +       if (h_file_tmp)
10948 +               fput(h_file_tmp);
10949 +       return err;
10950 +}
10951 +
10952 +/* ---------------------------------------------------------------------- */
10953 +
10954 +static int au_reopen_wh(struct file *file, aufs_bindex_t btgt,
10955 +                       struct dentry *hi_wh)
10956 +{
10957 +       int err;
10958 +       aufs_bindex_t bstart;
10959 +       struct au_dinfo *dinfo;
10960 +       struct dentry *h_dentry;
10961 +       struct au_hdentry *hdp;
10962 +
10963 +       dinfo = au_di(file->f_dentry);
10964 +       AuRwMustWriteLock(&dinfo->di_rwsem);
10965 +
10966 +       bstart = dinfo->di_bstart;
10967 +       dinfo->di_bstart = btgt;
10968 +       hdp = dinfo->di_hdentry;
10969 +       h_dentry = hdp[0 + btgt].hd_dentry;
10970 +       hdp[0 + btgt].hd_dentry = hi_wh;
10971 +       err = au_reopen_nondir(file);
10972 +       hdp[0 + btgt].hd_dentry = h_dentry;
10973 +       dinfo->di_bstart = bstart;
10974 +
10975 +       return err;
10976 +}
10977 +
10978 +static int au_ready_to_write_wh(struct file *file, loff_t len,
10979 +                               aufs_bindex_t bcpup, struct au_pin *pin)
10980 +{
10981 +       int err;
10982 +       struct inode *inode, *h_inode;
10983 +       struct dentry *h_dentry, *hi_wh;
10984 +       struct au_cp_generic cpg = {
10985 +               .dentry = file->f_dentry,
10986 +               .bdst   = bcpup,
10987 +               .bsrc   = -1,
10988 +               .len    = len,
10989 +               .pin    = pin
10990 +       };
10991 +
10992 +       au_update_dbstart(cpg.dentry);
10993 +       inode = cpg.dentry->d_inode;
10994 +       h_inode = NULL;
10995 +       if (au_dbstart(cpg.dentry) <= bcpup
10996 +           && au_dbend(cpg.dentry) >= bcpup) {
10997 +               h_dentry = au_h_dptr(cpg.dentry, bcpup);
10998 +               if (h_dentry)
10999 +                       h_inode = h_dentry->d_inode;
11000 +       }
11001 +       hi_wh = au_hi_wh(inode, bcpup);
11002 +       if (!hi_wh && !h_inode)
11003 +               err = au_sio_cpup_wh(&cpg, file);
11004 +       else
11005 +               /* already copied-up after unlink */
11006 +               err = au_reopen_wh(file, bcpup, hi_wh);
11007 +
11008 +       if (!err
11009 +           && inode->i_nlink > 1
11010 +           && au_opt_test(au_mntflags(cpg.dentry->d_sb), PLINK))
11011 +               au_plink_append(inode, bcpup, au_h_dptr(cpg.dentry, bcpup));
11012 +
11013 +       return err;
11014 +}
11015 +
11016 +/*
11017 + * prepare the @file for writing.
11018 + */
11019 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin)
11020 +{
11021 +       int err;
11022 +       aufs_bindex_t dbstart;
11023 +       struct dentry *parent, *h_dentry;
11024 +       struct inode *inode;
11025 +       struct super_block *sb;
11026 +       struct file *h_file;
11027 +       struct au_cp_generic cpg = {
11028 +               .dentry = file->f_dentry,
11029 +               .bdst   = -1,
11030 +               .bsrc   = -1,
11031 +               .len    = len,
11032 +               .pin    = pin,
11033 +               .flags  = AuCpup_DTIME
11034 +       };
11035 +
11036 +       sb = cpg.dentry->d_sb;
11037 +       inode = cpg.dentry->d_inode;
11038 +       AuDebugOn(au_special_file(inode->i_mode));
11039 +       cpg.bsrc = au_fbstart(file);
11040 +       err = au_test_ro(sb, cpg.bsrc, inode);
11041 +       if (!err && (au_hf_top(file)->f_mode & FMODE_WRITE)) {
11042 +               err = au_pin(pin, cpg.dentry, cpg.bsrc, AuOpt_UDBA_NONE,
11043 +                            /*flags*/0);
11044 +               goto out;
11045 +       }
11046 +
11047 +       /* need to cpup or reopen */
11048 +       parent = dget_parent(cpg.dentry);
11049 +       di_write_lock_parent(parent);
11050 +       err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
11051 +       cpg.bdst = err;
11052 +       if (unlikely(err < 0))
11053 +               goto out_dgrade;
11054 +       err = 0;
11055 +
11056 +       if (!d_unhashed(cpg.dentry) && !au_h_dptr(parent, cpg.bdst)) {
11057 +               err = au_cpup_dirs(cpg.dentry, cpg.bdst);
11058 +               if (unlikely(err))
11059 +                       goto out_dgrade;
11060 +       }
11061 +
11062 +       err = au_pin(pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
11063 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
11064 +       if (unlikely(err))
11065 +               goto out_dgrade;
11066 +
11067 +       h_dentry = au_hf_top(file)->f_dentry;
11068 +       dbstart = au_dbstart(cpg.dentry);
11069 +       if (dbstart <= cpg.bdst) {
11070 +               h_dentry = au_h_dptr(cpg.dentry, cpg.bdst);
11071 +               AuDebugOn(!h_dentry);
11072 +               cpg.bsrc = cpg.bdst;
11073 +       }
11074 +
11075 +       if (dbstart <= cpg.bdst         /* just reopen */
11076 +           || !d_unhashed(cpg.dentry)  /* copyup and reopen */
11077 +               ) {
11078 +               h_file = au_h_open_pre(cpg.dentry, cpg.bsrc, /*force_wr*/0);
11079 +               if (IS_ERR(h_file))
11080 +                       err = PTR_ERR(h_file);
11081 +               else {
11082 +                       di_downgrade_lock(parent, AuLock_IR);
11083 +                       if (dbstart > cpg.bdst)
11084 +                               err = au_sio_cpup_simple(&cpg);
11085 +                       if (!err)
11086 +                               err = au_reopen_nondir(file);
11087 +                       au_h_open_post(cpg.dentry, cpg.bsrc, h_file);
11088 +               }
11089 +       } else {                        /* copyup as wh and reopen */
11090 +               /*
11091 +                * since writable hfsplus branch is not supported,
11092 +                * h_open_pre/post() are unnecessary.
11093 +                */
11094 +               err = au_ready_to_write_wh(file, len, cpg.bdst, pin);
11095 +               di_downgrade_lock(parent, AuLock_IR);
11096 +       }
11097 +
11098 +       if (!err) {
11099 +               au_pin_set_parent_lflag(pin, /*lflag*/0);
11100 +               goto out_dput; /* success */
11101 +       }
11102 +       au_unpin(pin);
11103 +       goto out_unlock;
11104 +
11105 +out_dgrade:
11106 +       di_downgrade_lock(parent, AuLock_IR);
11107 +out_unlock:
11108 +       di_read_unlock(parent, AuLock_IR);
11109 +out_dput:
11110 +       dput(parent);
11111 +out:
11112 +       return err;
11113 +}
11114 +
11115 +/* ---------------------------------------------------------------------- */
11116 +
11117 +int au_do_flush(struct file *file, fl_owner_t id,
11118 +               int (*flush)(struct file *file, fl_owner_t id))
11119 +{
11120 +       int err;
11121 +       struct super_block *sb;
11122 +       struct inode *inode;
11123 +
11124 +       inode = file_inode(file);
11125 +       sb = inode->i_sb;
11126 +       si_noflush_read_lock(sb);
11127 +       fi_read_lock(file);
11128 +       ii_read_lock_child(inode);
11129 +
11130 +       err = flush(file, id);
11131 +       au_cpup_attr_timesizes(inode);
11132 +
11133 +       ii_read_unlock(inode);
11134 +       fi_read_unlock(file);
11135 +       si_read_unlock(sb);
11136 +       return err;
11137 +}
11138 +
11139 +/* ---------------------------------------------------------------------- */
11140 +
11141 +static int au_file_refresh_by_inode(struct file *file, int *need_reopen)
11142 +{
11143 +       int err;
11144 +       struct au_pin pin;
11145 +       struct au_finfo *finfo;
11146 +       struct dentry *parent, *hi_wh;
11147 +       struct inode *inode;
11148 +       struct super_block *sb;
11149 +       struct au_cp_generic cpg = {
11150 +               .dentry = file->f_dentry,
11151 +               .bdst   = -1,
11152 +               .bsrc   = -1,
11153 +               .len    = -1,
11154 +               .pin    = &pin,
11155 +               .flags  = AuCpup_DTIME
11156 +       };
11157 +
11158 +       FiMustWriteLock(file);
11159 +
11160 +       err = 0;
11161 +       finfo = au_fi(file);
11162 +       sb = cpg.dentry->d_sb;
11163 +       inode = cpg.dentry->d_inode;
11164 +       cpg.bdst = au_ibstart(inode);
11165 +       if (cpg.bdst == finfo->fi_btop || IS_ROOT(cpg.dentry))
11166 +               goto out;
11167 +
11168 +       parent = dget_parent(cpg.dentry);
11169 +       if (au_test_ro(sb, cpg.bdst, inode)) {
11170 +               di_read_lock_parent(parent, !AuLock_IR);
11171 +               err = AuWbrCopyup(au_sbi(sb), cpg.dentry);
11172 +               cpg.bdst = err;
11173 +               di_read_unlock(parent, !AuLock_IR);
11174 +               if (unlikely(err < 0))
11175 +                       goto out_parent;
11176 +               err = 0;
11177 +       }
11178 +
11179 +       di_read_lock_parent(parent, AuLock_IR);
11180 +       hi_wh = au_hi_wh(inode, cpg.bdst);
11181 +       if (!S_ISDIR(inode->i_mode)
11182 +           && au_opt_test(au_mntflags(sb), PLINK)
11183 +           && au_plink_test(inode)
11184 +           && !d_unhashed(cpg.dentry)
11185 +           && cpg.bdst < au_dbstart(cpg.dentry)) {
11186 +               err = au_test_and_cpup_dirs(cpg.dentry, cpg.bdst);
11187 +               if (unlikely(err))
11188 +                       goto out_unlock;
11189 +
11190 +               /* always superio. */
11191 +               err = au_pin(&pin, cpg.dentry, cpg.bdst, AuOpt_UDBA_NONE,
11192 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
11193 +               if (!err) {
11194 +                       err = au_sio_cpup_simple(&cpg);
11195 +                       au_unpin(&pin);
11196 +               }
11197 +       } else if (hi_wh) {
11198 +               /* already copied-up after unlink */
11199 +               err = au_reopen_wh(file, cpg.bdst, hi_wh);
11200 +               *need_reopen = 0;
11201 +       }
11202 +
11203 +out_unlock:
11204 +       di_read_unlock(parent, AuLock_IR);
11205 +out_parent:
11206 +       dput(parent);
11207 +out:
11208 +       return err;
11209 +}
11210 +
11211 +static void au_do_refresh_dir(struct file *file)
11212 +{
11213 +       aufs_bindex_t bindex, bend, new_bindex, brid;
11214 +       struct au_hfile *p, tmp, *q;
11215 +       struct au_finfo *finfo;
11216 +       struct super_block *sb;
11217 +       struct au_fidir *fidir;
11218 +
11219 +       FiMustWriteLock(file);
11220 +
11221 +       sb = file->f_dentry->d_sb;
11222 +       finfo = au_fi(file);
11223 +       fidir = finfo->fi_hdir;
11224 +       AuDebugOn(!fidir);
11225 +       p = fidir->fd_hfile + finfo->fi_btop;
11226 +       brid = p->hf_br->br_id;
11227 +       bend = fidir->fd_bbot;
11228 +       for (bindex = finfo->fi_btop; bindex <= bend; bindex++, p++) {
11229 +               if (!p->hf_file)
11230 +                       continue;
11231 +
11232 +               new_bindex = au_br_index(sb, p->hf_br->br_id);
11233 +               if (new_bindex == bindex)
11234 +                       continue;
11235 +               if (new_bindex < 0) {
11236 +                       au_set_h_fptr(file, bindex, NULL);
11237 +                       continue;
11238 +               }
11239 +
11240 +               /* swap two lower inode, and loop again */
11241 +               q = fidir->fd_hfile + new_bindex;
11242 +               tmp = *q;
11243 +               *q = *p;
11244 +               *p = tmp;
11245 +               if (tmp.hf_file) {
11246 +                       bindex--;
11247 +                       p--;
11248 +               }
11249 +       }
11250 +
11251 +       p = fidir->fd_hfile;
11252 +       if (!au_test_mmapped(file) && !d_unlinked(file->f_dentry)) {
11253 +               bend = au_sbend(sb);
11254 +               for (finfo->fi_btop = 0; finfo->fi_btop <= bend;
11255 +                    finfo->fi_btop++, p++)
11256 +                       if (p->hf_file) {
11257 +                               if (file_inode(p->hf_file))
11258 +                                       break;
11259 +                               else
11260 +                                       au_hfput(p, file);
11261 +                       }
11262 +       } else {
11263 +               bend = au_br_index(sb, brid);
11264 +               for (finfo->fi_btop = 0; finfo->fi_btop < bend;
11265 +                    finfo->fi_btop++, p++)
11266 +                       if (p->hf_file)
11267 +                               au_hfput(p, file);
11268 +               bend = au_sbend(sb);
11269 +       }
11270 +
11271 +       p = fidir->fd_hfile + bend;
11272 +       for (fidir->fd_bbot = bend; fidir->fd_bbot >= finfo->fi_btop;
11273 +            fidir->fd_bbot--, p--)
11274 +               if (p->hf_file) {
11275 +                       if (file_inode(p->hf_file))
11276 +                               break;
11277 +                       else
11278 +                               au_hfput(p, file);
11279 +               }
11280 +       AuDebugOn(fidir->fd_bbot < finfo->fi_btop);
11281 +}
11282 +
11283 +/*
11284 + * after branch manipulating, refresh the file.
11285 + */
11286 +static int refresh_file(struct file *file, int (*reopen)(struct file *file))
11287 +{
11288 +       int err, need_reopen;
11289 +       aufs_bindex_t bend, bindex;
11290 +       struct dentry *dentry;
11291 +       struct au_finfo *finfo;
11292 +       struct au_hfile *hfile;
11293 +
11294 +       dentry = file->f_dentry;
11295 +       finfo = au_fi(file);
11296 +       if (!finfo->fi_hdir) {
11297 +               hfile = &finfo->fi_htop;
11298 +               AuDebugOn(!hfile->hf_file);
11299 +               bindex = au_br_index(dentry->d_sb, hfile->hf_br->br_id);
11300 +               AuDebugOn(bindex < 0);
11301 +               if (bindex != finfo->fi_btop)
11302 +                       au_set_fbstart(file, bindex);
11303 +       } else {
11304 +               err = au_fidir_realloc(finfo, au_sbend(dentry->d_sb) + 1);
11305 +               if (unlikely(err))
11306 +                       goto out;
11307 +               au_do_refresh_dir(file);
11308 +       }
11309 +
11310 +       err = 0;
11311 +       need_reopen = 1;
11312 +       if (!au_test_mmapped(file))
11313 +               err = au_file_refresh_by_inode(file, &need_reopen);
11314 +       if (!err && need_reopen && !d_unlinked(dentry))
11315 +               err = reopen(file);
11316 +       if (!err) {
11317 +               au_update_figen(file);
11318 +               goto out; /* success */
11319 +       }
11320 +
11321 +       /* error, close all lower files */
11322 +       if (finfo->fi_hdir) {
11323 +               bend = au_fbend_dir(file);
11324 +               for (bindex = au_fbstart(file); bindex <= bend; bindex++)
11325 +                       au_set_h_fptr(file, bindex, NULL);
11326 +       }
11327 +
11328 +out:
11329 +       return err;
11330 +}
11331 +
11332 +/* common function to regular file and dir */
11333 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
11334 +                         int wlock)
11335 +{
11336 +       int err;
11337 +       unsigned int sigen, figen;
11338 +       aufs_bindex_t bstart;
11339 +       unsigned char pseudo_link;
11340 +       struct dentry *dentry;
11341 +       struct inode *inode;
11342 +
11343 +       err = 0;
11344 +       dentry = file->f_dentry;
11345 +       inode = dentry->d_inode;
11346 +       AuDebugOn(au_special_file(inode->i_mode));
11347 +       sigen = au_sigen(dentry->d_sb);
11348 +       fi_write_lock(file);
11349 +       figen = au_figen(file);
11350 +       di_write_lock_child(dentry);
11351 +       bstart = au_dbstart(dentry);
11352 +       pseudo_link = (bstart != au_ibstart(inode));
11353 +       if (sigen == figen && !pseudo_link && au_fbstart(file) == bstart) {
11354 +               if (!wlock) {
11355 +                       di_downgrade_lock(dentry, AuLock_IR);
11356 +                       fi_downgrade_lock(file);
11357 +               }
11358 +               goto out; /* success */
11359 +       }
11360 +
11361 +       AuDbg("sigen %d, figen %d\n", sigen, figen);
11362 +       if (au_digen_test(dentry, sigen)) {
11363 +               err = au_reval_dpath(dentry, sigen);
11364 +               AuDebugOn(!err && au_digen_test(dentry, sigen));
11365 +       }
11366 +
11367 +       if (!err)
11368 +               err = refresh_file(file, reopen);
11369 +       if (!err) {
11370 +               if (!wlock) {
11371 +                       di_downgrade_lock(dentry, AuLock_IR);
11372 +                       fi_downgrade_lock(file);
11373 +               }
11374 +       } else {
11375 +               di_write_unlock(dentry);
11376 +               fi_write_unlock(file);
11377 +       }
11378 +
11379 +out:
11380 +       return err;
11381 +}
11382 +
11383 +/* ---------------------------------------------------------------------- */
11384 +
11385 +/* cf. aufs_nopage() */
11386 +/* for madvise(2) */
11387 +static int aufs_readpage(struct file *file __maybe_unused, struct page *page)
11388 +{
11389 +       unlock_page(page);
11390 +       return 0;
11391 +}
11392 +
11393 +/* it will never be called, but necessary to support O_DIRECT */
11394 +static ssize_t aufs_direct_IO(int rw, struct kiocb *iocb,
11395 +                             const struct iovec *iov, loff_t offset,
11396 +                             unsigned long nr_segs)
11397 +{ BUG(); return 0; }
11398 +
11399 +/*
11400 + * it will never be called, but madvise and fadvise behaves differently
11401 + * when get_xip_mem is defined
11402 + */
11403 +static int aufs_get_xip_mem(struct address_space *mapping, pgoff_t pgoff,
11404 +                           int create, void **kmem, unsigned long *pfn)
11405 +{ BUG(); return 0; }
11406 +
11407 +/* they will never be called. */
11408 +#ifdef CONFIG_AUFS_DEBUG
11409 +static int aufs_write_begin(struct file *file, struct address_space *mapping,
11410 +                           loff_t pos, unsigned len, unsigned flags,
11411 +                           struct page **pagep, void **fsdata)
11412 +{ AuUnsupport(); return 0; }
11413 +static int aufs_write_end(struct file *file, struct address_space *mapping,
11414 +                         loff_t pos, unsigned len, unsigned copied,
11415 +                         struct page *page, void *fsdata)
11416 +{ AuUnsupport(); return 0; }
11417 +static int aufs_writepage(struct page *page, struct writeback_control *wbc)
11418 +{ AuUnsupport(); return 0; }
11419 +
11420 +static int aufs_set_page_dirty(struct page *page)
11421 +{ AuUnsupport(); return 0; }
11422 +static void aufs_invalidatepage(struct page *page, unsigned int offset,
11423 +                               unsigned int length)
11424 +{ AuUnsupport(); }
11425 +static int aufs_releasepage(struct page *page, gfp_t gfp)
11426 +{ AuUnsupport(); return 0; }
11427 +static int aufs_migratepage(struct address_space *mapping, struct page *newpage,
11428 +                           struct page *page, enum migrate_mode mode)
11429 +{ AuUnsupport(); return 0; }
11430 +static int aufs_launder_page(struct page *page)
11431 +{ AuUnsupport(); return 0; }
11432 +static int aufs_is_partially_uptodate(struct page *page,
11433 +                                     read_descriptor_t *desc,
11434 +                                     unsigned long from)
11435 +{ AuUnsupport(); return 0; }
11436 +static void aufs_is_dirty_writeback(struct page *page, bool *dirty,
11437 +                                   bool *writeback)
11438 +{ AuUnsupport(); }
11439 +static int aufs_error_remove_page(struct address_space *mapping,
11440 +                                 struct page *page)
11441 +{ AuUnsupport(); return 0; }
11442 +static int aufs_swap_activate(struct swap_info_struct *sis, struct file *file,
11443 +                             sector_t *span)
11444 +{ AuUnsupport(); return 0; }
11445 +static void aufs_swap_deactivate(struct file *file)
11446 +{ AuUnsupport(); }
11447 +#endif /* CONFIG_AUFS_DEBUG */
11448 +
11449 +const struct address_space_operations aufs_aop = {
11450 +       .readpage               = aufs_readpage,
11451 +       .direct_IO              = aufs_direct_IO,
11452 +       .get_xip_mem            = aufs_get_xip_mem,
11453 +#ifdef CONFIG_AUFS_DEBUG
11454 +       .writepage              = aufs_writepage,
11455 +       /* no writepages, because of writepage */
11456 +       .set_page_dirty         = aufs_set_page_dirty,
11457 +       /* no readpages, because of readpage */
11458 +       .write_begin            = aufs_write_begin,
11459 +       .write_end              = aufs_write_end,
11460 +       /* no bmap, no block device */
11461 +       .invalidatepage         = aufs_invalidatepage,
11462 +       .releasepage            = aufs_releasepage,
11463 +       .migratepage            = aufs_migratepage,
11464 +       .launder_page           = aufs_launder_page,
11465 +       .is_partially_uptodate  = aufs_is_partially_uptodate,
11466 +       .is_dirty_writeback     = aufs_is_dirty_writeback,
11467 +       .error_remove_page      = aufs_error_remove_page,
11468 +       .swap_activate          = aufs_swap_activate,
11469 +       .swap_deactivate        = aufs_swap_deactivate
11470 +#endif /* CONFIG_AUFS_DEBUG */
11471 +};
11472 diff -urN /usr/share/empty/fs/aufs/file.h linux/fs/aufs/file.h
11473 --- /usr/share/empty/fs/aufs/file.h     1970-01-01 01:00:00.000000000 +0100
11474 +++ linux/fs/aufs/file.h        2014-04-24 22:11:10.851935747 +0200
11475 @@ -0,0 +1,312 @@
11476 +/*
11477 + * Copyright (C) 2005-2014 Junjiro R. Okajima
11478 + *
11479 + * This program, aufs is free software; you can redistribute it and/or modify
11480 + * it under the terms of the GNU General Public License as published by
11481 + * the Free Software Foundation; either version 2 of the License, or
11482 + * (at your option) any later version.
11483 + *
11484 + * This program is distributed in the hope that it will be useful,
11485 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11486 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11487 + * GNU General Public License for more details.
11488 + *
11489 + * You should have received a copy of the GNU General Public License
11490 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
11491 + */
11492 +
11493 +/*
11494 + * file operations
11495 + */
11496 +
11497 +#ifndef __AUFS_FILE_H__
11498 +#define __AUFS_FILE_H__
11499 +
11500 +#ifdef __KERNEL__
11501 +
11502 +#include <linux/file.h>
11503 +#include <linux/fs.h>
11504 +#include <linux/poll.h>
11505 +#include "rwsem.h"
11506 +
11507 +struct au_branch;
11508 +struct au_hfile {
11509 +       struct file             *hf_file;
11510 +       struct au_branch        *hf_br;
11511 +};
11512 +
11513 +struct au_vdir;
11514 +struct au_fidir {
11515 +       aufs_bindex_t           fd_bbot;
11516 +       aufs_bindex_t           fd_nent;
11517 +       struct au_vdir          *fd_vdir_cache;
11518 +       struct au_hfile         fd_hfile[];
11519 +};
11520 +
11521 +static inline int au_fidir_sz(int nent)
11522 +{
11523 +       AuDebugOn(nent < 0);
11524 +       return sizeof(struct au_fidir) + sizeof(struct au_hfile) * nent;
11525 +}
11526 +
11527 +struct au_finfo {
11528 +       atomic_t                fi_generation;
11529 +
11530 +       struct au_rwsem         fi_rwsem;
11531 +       aufs_bindex_t           fi_btop;
11532 +
11533 +       /* do not union them */
11534 +       struct {                                /* for non-dir */
11535 +               struct au_hfile                 fi_htop;
11536 +               atomic_t                        fi_mmapped;
11537 +       };
11538 +       struct au_fidir         *fi_hdir;       /* for dir only */
11539 +
11540 +       struct hlist_node       fi_hlist;
11541 +       struct file             *fi_file;       /* very ugly */
11542 +} ____cacheline_aligned_in_smp;
11543 +
11544 +/* ---------------------------------------------------------------------- */
11545 +
11546 +/* file.c */
11547 +extern const struct address_space_operations aufs_aop;
11548 +unsigned int au_file_roflags(unsigned int flags);
11549 +struct file *au_h_open(struct dentry *dentry, aufs_bindex_t bindex, int flags,
11550 +                      struct file *file, int force_wr);
11551 +int au_do_open(struct file *file, int (*open)(struct file *file, int flags),
11552 +              struct au_fidir *fidir);
11553 +int au_reopen_nondir(struct file *file);
11554 +struct au_pin;
11555 +int au_ready_to_write(struct file *file, loff_t len, struct au_pin *pin);
11556 +int au_reval_and_lock_fdi(struct file *file, int (*reopen)(struct file *file),
11557 +                         int wlock);
11558 +int au_do_flush(struct file *file, fl_owner_t id,
11559 +               int (*flush)(struct file *file, fl_owner_t id));
11560 +
11561 +/* poll.c */
11562 +#ifdef CONFIG_AUFS_POLL
11563 +unsigned int aufs_poll(struct file *file, poll_table *wait);
11564 +#endif
11565 +
11566 +#ifdef CONFIG_AUFS_BR_HFSPLUS
11567 +/* hfsplus.c */
11568 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
11569 +                          int force_wr);
11570 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
11571 +                   struct file *h_file);
11572 +#else
11573 +static inline
11574 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
11575 +                          int force_wr)
11576 +{
11577 +       return NULL;
11578 +}
11579 +
11580 +AuStubVoid(au_h_open_post, struct dentry *dentry, aufs_bindex_t bindex,
11581 +          struct file *h_file);
11582 +#endif
11583 +
11584 +/* f_op.c */
11585 +extern const struct file_operations aufs_file_fop;
11586 +int au_do_open_nondir(struct file *file, int flags);
11587 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file);
11588 +
11589 +#ifdef CONFIG_AUFS_SP_IATTR
11590 +/* f_op_sp.c */
11591 +struct au_finfo *au_fi_sp(struct file *file);
11592 +int au_special_file(umode_t mode);
11593 +void au_init_special_fop(struct inode *inode, umode_t mode, dev_t rdev);
11594 +#else
11595 +static inline struct au_finfo *au_fi_sp(struct file *file)
11596 +{
11597 +       return NULL;
11598 +}
11599 +AuStubInt0(au_special_file, umode_t mode)
11600 +static inline void au_init_special_fop(struct inode *inode, umode_t mode,
11601 +                                      dev_t rdev)
11602 +{
11603 +       init_special_inode(inode, mode, rdev);
11604 +}
11605 +#endif
11606 +
11607 +/* finfo.c */
11608 +void au_hfput(struct au_hfile *hf, struct file *file);
11609 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex,
11610 +                  struct file *h_file);
11611 +
11612 +void au_update_figen(struct file *file);
11613 +struct au_fidir *au_fidir_alloc(struct super_block *sb);
11614 +int au_fidir_realloc(struct au_finfo *finfo, int nbr);
11615 +
11616 +void au_fi_init_once(void *_fi);
11617 +void au_finfo_fin(struct file *file);
11618 +int au_finfo_init(struct file *file, struct au_fidir *fidir);
11619 +
11620 +/* ioctl.c */
11621 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg);
11622 +#ifdef CONFIG_COMPAT
11623 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
11624 +                          unsigned long arg);
11625 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
11626 +                             unsigned long arg);
11627 +#endif
11628 +
11629 +/* ---------------------------------------------------------------------- */
11630 +
11631 +static inline struct au_finfo *au_fi(struct file *file)
11632 +{
11633 +       struct au_finfo *finfo;
11634 +
11635 +       finfo = au_fi_sp(file);
11636 +       if (!finfo)
11637 +               finfo = file->private_data;
11638 +       return finfo;
11639 +}
11640 +
11641 +/* ---------------------------------------------------------------------- */
11642 +
11643 +/*
11644 + * fi_read_lock, fi_write_lock,
11645 + * fi_read_unlock, fi_write_unlock, fi_downgrade_lock
11646 + */
11647 +AuSimpleRwsemFuncs(fi, struct file *f, &au_fi(f)->fi_rwsem);
11648 +
11649 +#define FiMustNoWaiters(f)     AuRwMustNoWaiters(&au_fi(f)->fi_rwsem)
11650 +#define FiMustAnyLock(f)       AuRwMustAnyLock(&au_fi(f)->fi_rwsem)
11651 +#define FiMustWriteLock(f)     AuRwMustWriteLock(&au_fi(f)->fi_rwsem)
11652 +
11653 +/* ---------------------------------------------------------------------- */
11654 +
11655 +/* todo: hard/soft set? */
11656 +static inline aufs_bindex_t au_fbstart(struct file *file)
11657 +{
11658 +       FiMustAnyLock(file);
11659 +       return au_fi(file)->fi_btop;
11660 +}
11661 +
11662 +static inline aufs_bindex_t au_fbend_dir(struct file *file)
11663 +{
11664 +       FiMustAnyLock(file);
11665 +       AuDebugOn(!au_fi(file)->fi_hdir);
11666 +       return au_fi(file)->fi_hdir->fd_bbot;
11667 +}
11668 +
11669 +static inline struct au_vdir *au_fvdir_cache(struct file *file)
11670 +{
11671 +       FiMustAnyLock(file);
11672 +       AuDebugOn(!au_fi(file)->fi_hdir);
11673 +       return au_fi(file)->fi_hdir->fd_vdir_cache;
11674 +}
11675 +
11676 +static inline void au_set_fbstart(struct file *file, aufs_bindex_t bindex)
11677 +{
11678 +       FiMustWriteLock(file);
11679 +       au_fi(file)->fi_btop = bindex;
11680 +}
11681 +
11682 +static inline void au_set_fbend_dir(struct file *file, aufs_bindex_t bindex)
11683 +{
11684 +       FiMustWriteLock(file);
11685 +       AuDebugOn(!au_fi(file)->fi_hdir);
11686 +       au_fi(file)->fi_hdir->fd_bbot = bindex;
11687 +}
11688 +
11689 +static inline void au_set_fvdir_cache(struct file *file,
11690 +                                     struct au_vdir *vdir_cache)
11691 +{
11692 +       FiMustWriteLock(file);
11693 +       AuDebugOn(!au_fi(file)->fi_hdir);
11694 +       au_fi(file)->fi_hdir->fd_vdir_cache = vdir_cache;
11695 +}
11696 +
11697 +static inline struct file *au_hf_top(struct file *file)
11698 +{
11699 +       FiMustAnyLock(file);
11700 +       AuDebugOn(au_fi(file)->fi_hdir);
11701 +       return au_fi(file)->fi_htop.hf_file;
11702 +}
11703 +
11704 +static inline struct file *au_hf_dir(struct file *file, aufs_bindex_t bindex)
11705 +{
11706 +       FiMustAnyLock(file);
11707 +       AuDebugOn(!au_fi(file)->fi_hdir);
11708 +       return au_fi(file)->fi_hdir->fd_hfile[0 + bindex].hf_file;
11709 +}
11710 +
11711 +/* todo: memory barrier? */
11712 +static inline unsigned int au_figen(struct file *f)
11713 +{
11714 +       return atomic_read(&au_fi(f)->fi_generation);
11715 +}
11716 +
11717 +static inline void au_set_mmapped(struct file *f)
11718 +{
11719 +       if (atomic_inc_return(&au_fi(f)->fi_mmapped))
11720 +               return;
11721 +       pr_warn("fi_mmapped wrapped around\n");
11722 +       while (!atomic_inc_return(&au_fi(f)->fi_mmapped))
11723 +               ;
11724 +}
11725 +
11726 +static inline void au_unset_mmapped(struct file *f)
11727 +{
11728 +       atomic_dec(&au_fi(f)->fi_mmapped);
11729 +}
11730 +
11731 +static inline int au_test_mmapped(struct file *f)
11732 +{
11733 +       return atomic_read(&au_fi(f)->fi_mmapped);
11734 +}
11735 +
11736 +/* customize vma->vm_file */
11737 +
11738 +static inline void au_do_vm_file_reset(struct vm_area_struct *vma,
11739 +                                      struct file *file)
11740 +{
11741 +       struct file *f;
11742 +
11743 +       f = vma->vm_file;
11744 +       get_file(file);
11745 +       vma->vm_file = file;
11746 +       fput(f);
11747 +}
11748 +
11749 +#ifdef CONFIG_MMU
11750 +#define AuDbgVmRegion(file, vma) do {} while (0)
11751 +
11752 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
11753 +                                   struct file *file)
11754 +{
11755 +       au_do_vm_file_reset(vma, file);
11756 +}
11757 +#else
11758 +#define AuDbgVmRegion(file, vma) \
11759 +       AuDebugOn((vma)->vm_region && (vma)->vm_region->vm_file != (file))
11760 +
11761 +static inline void au_vm_file_reset(struct vm_area_struct *vma,
11762 +                                   struct file *file)
11763 +{
11764 +       struct file *f;
11765 +
11766 +       au_do_vm_file_reset(vma, file);
11767 +       f = vma->vm_region->vm_file;
11768 +       get_file(file);
11769 +       vma->vm_region->vm_file = file;
11770 +       fput(f);
11771 +}
11772 +#endif /* CONFIG_MMU */
11773 +
11774 +/* handle vma->vm_prfile */
11775 +static inline void au_vm_prfile_set(struct vm_area_struct *vma,
11776 +                                   struct file *file)
11777 +{
11778 +       get_file(file);
11779 +       vma->vm_prfile = file;
11780 +#ifndef CONFIG_MMU
11781 +       get_file(file);
11782 +       vma->vm_region->vm_prfile = file;
11783 +#endif
11784 +}
11785 +
11786 +#endif /* __KERNEL__ */
11787 +#endif /* __AUFS_FILE_H__ */
11788 diff -urN /usr/share/empty/fs/aufs/finfo.c linux/fs/aufs/finfo.c
11789 --- /usr/share/empty/fs/aufs/finfo.c    1970-01-01 01:00:00.000000000 +0100
11790 +++ linux/fs/aufs/finfo.c       2014-04-24 22:11:10.851935747 +0200
11791 @@ -0,0 +1,156 @@
11792 +/*
11793 + * Copyright (C) 2005-2014 Junjiro R. Okajima
11794 + *
11795 + * This program, aufs is free software; you can redistribute it and/or modify
11796 + * it under the terms of the GNU General Public License as published by
11797 + * the Free Software Foundation; either version 2 of the License, or
11798 + * (at your option) any later version.
11799 + *
11800 + * This program is distributed in the hope that it will be useful,
11801 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11802 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11803 + * GNU General Public License for more details.
11804 + *
11805 + * You should have received a copy of the GNU General Public License
11806 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
11807 + */
11808 +
11809 +/*
11810 + * file private data
11811 + */
11812 +
11813 +#include "aufs.h"
11814 +
11815 +void au_hfput(struct au_hfile *hf, struct file *file)
11816 +{
11817 +       /* todo: direct access f_flags */
11818 +       if (vfsub_file_flags(file) & __FMODE_EXEC)
11819 +               allow_write_access(hf->hf_file);
11820 +       fput(hf->hf_file);
11821 +       hf->hf_file = NULL;
11822 +       atomic_dec(&hf->hf_br->br_count);
11823 +       hf->hf_br = NULL;
11824 +}
11825 +
11826 +void au_set_h_fptr(struct file *file, aufs_bindex_t bindex, struct file *val)
11827 +{
11828 +       struct au_finfo *finfo = au_fi(file);
11829 +       struct au_hfile *hf;
11830 +       struct au_fidir *fidir;
11831 +
11832 +       fidir = finfo->fi_hdir;
11833 +       if (!fidir) {
11834 +               AuDebugOn(finfo->fi_btop != bindex);
11835 +               hf = &finfo->fi_htop;
11836 +       } else
11837 +               hf = fidir->fd_hfile + bindex;
11838 +
11839 +       if (hf && hf->hf_file)
11840 +               au_hfput(hf, file);
11841 +       if (val) {
11842 +               FiMustWriteLock(file);
11843 +               hf->hf_file = val;
11844 +               hf->hf_br = au_sbr(file->f_dentry->d_sb, bindex);
11845 +       }
11846 +}
11847 +
11848 +void au_update_figen(struct file *file)
11849 +{
11850 +       atomic_set(&au_fi(file)->fi_generation, au_digen(file->f_dentry));
11851 +       /* smp_mb(); */ /* atomic_set */
11852 +}
11853 +
11854 +/* ---------------------------------------------------------------------- */
11855 +
11856 +struct au_fidir *au_fidir_alloc(struct super_block *sb)
11857 +{
11858 +       struct au_fidir *fidir;
11859 +       int nbr;
11860 +
11861 +       nbr = au_sbend(sb) + 1;
11862 +       if (nbr < 2)
11863 +               nbr = 2; /* initial allocate for 2 branches */
11864 +       fidir = kzalloc(au_fidir_sz(nbr), GFP_NOFS);
11865 +       if (fidir) {
11866 +               fidir->fd_bbot = -1;
11867 +               fidir->fd_nent = nbr;
11868 +               fidir->fd_vdir_cache = NULL;
11869 +       }
11870 +
11871 +       return fidir;
11872 +}
11873 +
11874 +int au_fidir_realloc(struct au_finfo *finfo, int nbr)
11875 +{
11876 +       int err;
11877 +       struct au_fidir *fidir, *p;
11878 +
11879 +       AuRwMustWriteLock(&finfo->fi_rwsem);
11880 +       fidir = finfo->fi_hdir;
11881 +       AuDebugOn(!fidir);
11882 +
11883 +       err = -ENOMEM;
11884 +       p = au_kzrealloc(fidir, au_fidir_sz(fidir->fd_nent), au_fidir_sz(nbr),
11885 +                        GFP_NOFS);
11886 +       if (p) {
11887 +               p->fd_nent = nbr;
11888 +               finfo->fi_hdir = p;
11889 +               err = 0;
11890 +       }
11891 +
11892 +       return err;
11893 +}
11894 +
11895 +/* ---------------------------------------------------------------------- */
11896 +
11897 +void au_finfo_fin(struct file *file)
11898 +{
11899 +       struct au_finfo *finfo;
11900 +
11901 +       au_nfiles_dec(file->f_dentry->d_sb);
11902 +
11903 +       finfo = au_fi(file);
11904 +       AuDebugOn(finfo->fi_hdir);
11905 +       AuRwDestroy(&finfo->fi_rwsem);
11906 +       au_cache_free_finfo(finfo);
11907 +}
11908 +
11909 +void au_fi_init_once(void *_finfo)
11910 +{
11911 +       struct au_finfo *finfo = _finfo;
11912 +       static struct lock_class_key aufs_fi;
11913 +
11914 +       au_rw_init(&finfo->fi_rwsem);
11915 +       au_rw_class(&finfo->fi_rwsem, &aufs_fi);
11916 +}
11917 +
11918 +int au_finfo_init(struct file *file, struct au_fidir *fidir)
11919 +{
11920 +       int err;
11921 +       struct au_finfo *finfo;
11922 +       struct dentry *dentry;
11923 +
11924 +       err = -ENOMEM;
11925 +       dentry = file->f_dentry;
11926 +       finfo = au_cache_alloc_finfo();
11927 +       if (unlikely(!finfo))
11928 +               goto out;
11929 +
11930 +       err = 0;
11931 +       au_nfiles_inc(dentry->d_sb);
11932 +       /* verbose coding for lock class name */
11933 +       if (!fidir)
11934 +               au_rw_class(&finfo->fi_rwsem, au_lc_key + AuLcNonDir_FIINFO);
11935 +       else
11936 +               au_rw_class(&finfo->fi_rwsem, au_lc_key + AuLcDir_FIINFO);
11937 +       au_rw_write_lock(&finfo->fi_rwsem);
11938 +       finfo->fi_btop = -1;
11939 +       finfo->fi_hdir = fidir;
11940 +       atomic_set(&finfo->fi_generation, au_digen(dentry));
11941 +       /* smp_mb(); */ /* atomic_set */
11942 +
11943 +       file->private_data = finfo;
11944 +
11945 +out:
11946 +       return err;
11947 +}
11948 diff -urN /usr/share/empty/fs/aufs/f_op.c linux/fs/aufs/f_op.c
11949 --- /usr/share/empty/fs/aufs/f_op.c     1970-01-01 01:00:00.000000000 +0100
11950 +++ linux/fs/aufs/f_op.c        2014-04-24 22:11:10.851935747 +0200
11951 @@ -0,0 +1,718 @@
11952 +/*
11953 + * Copyright (C) 2005-2014 Junjiro R. Okajima
11954 + *
11955 + * This program, aufs is free software; you can redistribute it and/or modify
11956 + * it under the terms of the GNU General Public License as published by
11957 + * the Free Software Foundation; either version 2 of the License, or
11958 + * (at your option) any later version.
11959 + *
11960 + * This program is distributed in the hope that it will be useful,
11961 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
11962 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11963 + * GNU General Public License for more details.
11964 + *
11965 + * You should have received a copy of the GNU General Public License
11966 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
11967 + */
11968 +
11969 +/*
11970 + * file and vm operations
11971 + */
11972 +
11973 +#include <linux/aio.h>
11974 +#include <linux/fs_stack.h>
11975 +#include <linux/mman.h>
11976 +#include <linux/security.h>
11977 +#include "aufs.h"
11978 +
11979 +int au_do_open_nondir(struct file *file, int flags)
11980 +{
11981 +       int err;
11982 +       aufs_bindex_t bindex;
11983 +       struct file *h_file;
11984 +       struct dentry *dentry;
11985 +       struct au_finfo *finfo;
11986 +
11987 +       FiMustWriteLock(file);
11988 +
11989 +       err = 0;
11990 +       dentry = file->f_dentry;
11991 +       finfo = au_fi(file);
11992 +       memset(&finfo->fi_htop, 0, sizeof(finfo->fi_htop));
11993 +       atomic_set(&finfo->fi_mmapped, 0);
11994 +       bindex = au_dbstart(dentry);
11995 +       h_file = au_h_open(dentry, bindex, flags, file, /*force_wr*/0);
11996 +       if (IS_ERR(h_file))
11997 +               err = PTR_ERR(h_file);
11998 +       else {
11999 +               au_set_fbstart(file, bindex);
12000 +               au_set_h_fptr(file, bindex, h_file);
12001 +               au_update_figen(file);
12002 +               finfo->fi_file = file;
12003 +               au_sphl_add(&finfo->fi_hlist, &au_sbi(dentry->d_sb)->si_files);
12004 +               /* todo: necessary? */
12005 +               /* file->f_ra = h_file->f_ra; */
12006 +       }
12007 +
12008 +       return err;
12009 +}
12010 +
12011 +static int aufs_open_nondir(struct inode *inode __maybe_unused,
12012 +                           struct file *file)
12013 +{
12014 +       int err;
12015 +       struct super_block *sb;
12016 +
12017 +       AuDbg("%pD, f_flags 0x%x, f_mode 0x%x\n",
12018 +             file, vfsub_file_flags(file), file->f_mode);
12019 +
12020 +       sb = file->f_dentry->d_sb;
12021 +       si_read_lock(sb, AuLock_FLUSH);
12022 +       err = au_do_open(file, au_do_open_nondir, /*fidir*/NULL);
12023 +       si_read_unlock(sb);
12024 +       return err;
12025 +}
12026 +
12027 +int aufs_release_nondir(struct inode *inode __maybe_unused, struct file *file)
12028 +{
12029 +       struct au_finfo *finfo;
12030 +       aufs_bindex_t bindex;
12031 +
12032 +       finfo = au_fi(file);
12033 +       au_sphl_del(&finfo->fi_hlist, &au_sbi(file->f_dentry->d_sb)->si_files);
12034 +       bindex = finfo->fi_btop;
12035 +       if (bindex >= 0)
12036 +               au_set_h_fptr(file, bindex, NULL);
12037 +
12038 +       au_finfo_fin(file);
12039 +       return 0;
12040 +}
12041 +
12042 +/* ---------------------------------------------------------------------- */
12043 +
12044 +static int au_do_flush_nondir(struct file *file, fl_owner_t id)
12045 +{
12046 +       int err;
12047 +       struct file *h_file;
12048 +
12049 +       err = 0;
12050 +       h_file = au_hf_top(file);
12051 +       if (h_file)
12052 +               err = vfsub_flush(h_file, id);
12053 +       return err;
12054 +}
12055 +
12056 +static int aufs_flush_nondir(struct file *file, fl_owner_t id)
12057 +{
12058 +       return au_do_flush(file, id, au_do_flush_nondir);
12059 +}
12060 +
12061 +/* ---------------------------------------------------------------------- */
12062 +/*
12063 + * read and write functions acquire [fdi]_rwsem once, but release before
12064 + * mmap_sem. This is because to stop a race condition between mmap(2).
12065 + * Releasing these aufs-rwsem should be safe, no branch-mamagement (by keeping
12066 + * si_rwsem), no harmful copy-up should happen. Actually copy-up may happen in
12067 + * read functions after [fdi]_rwsem are released, but it should be harmless.
12068 + */
12069 +
12070 +static ssize_t aufs_read(struct file *file, char __user *buf, size_t count,
12071 +                        loff_t *ppos)
12072 +{
12073 +       ssize_t err;
12074 +       struct dentry *dentry;
12075 +       struct file *h_file;
12076 +       struct super_block *sb;
12077 +
12078 +       dentry = file->f_dentry;
12079 +       sb = dentry->d_sb;
12080 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
12081 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
12082 +       if (unlikely(err))
12083 +               goto out;
12084 +
12085 +       h_file = au_hf_top(file);
12086 +       get_file(h_file);
12087 +       di_read_unlock(dentry, AuLock_IR);
12088 +       fi_read_unlock(file);
12089 +
12090 +       /* filedata may be obsoleted by concurrent copyup, but no problem */
12091 +       err = vfsub_read_u(h_file, buf, count, ppos);
12092 +       /* todo: necessary? */
12093 +       /* file->f_ra = h_file->f_ra; */
12094 +       /* update without lock, I don't think it a problem */
12095 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
12096 +       fput(h_file);
12097 +
12098 +out:
12099 +       si_read_unlock(sb);
12100 +       return err;
12101 +}
12102 +
12103 +/*
12104 + * todo: very ugly
12105 + * it locks both of i_mutex and si_rwsem for read in safe.
12106 + * if the plink maintenance mode continues forever (that is the problem),
12107 + * may loop forever.
12108 + */
12109 +static void au_mtx_and_read_lock(struct inode *inode)
12110 +{
12111 +       int err;
12112 +       struct super_block *sb = inode->i_sb;
12113 +
12114 +       while (1) {
12115 +               mutex_lock(&inode->i_mutex);
12116 +               err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
12117 +               if (!err)
12118 +                       break;
12119 +               mutex_unlock(&inode->i_mutex);
12120 +               si_read_lock(sb, AuLock_NOPLMW);
12121 +               si_read_unlock(sb);
12122 +       }
12123 +}
12124 +
12125 +static ssize_t aufs_write(struct file *file, const char __user *ubuf,
12126 +                         size_t count, loff_t *ppos)
12127 +{
12128 +       ssize_t err;
12129 +       struct au_pin pin;
12130 +       struct dentry *dentry;
12131 +       struct super_block *sb;
12132 +       struct inode *inode;
12133 +       struct file *h_file;
12134 +       char __user *buf = (char __user *)ubuf;
12135 +
12136 +       dentry = file->f_dentry;
12137 +       sb = dentry->d_sb;
12138 +       inode = dentry->d_inode;
12139 +       au_mtx_and_read_lock(inode);
12140 +
12141 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
12142 +       if (unlikely(err))
12143 +               goto out;
12144 +
12145 +       err = au_ready_to_write(file, -1, &pin);
12146 +       di_downgrade_lock(dentry, AuLock_IR);
12147 +       if (unlikely(err)) {
12148 +               di_read_unlock(dentry, AuLock_IR);
12149 +               fi_write_unlock(file);
12150 +               goto out;
12151 +       }
12152 +
12153 +       h_file = au_hf_top(file);
12154 +       get_file(h_file);
12155 +       au_unpin(&pin);
12156 +       di_read_unlock(dentry, AuLock_IR);
12157 +       fi_write_unlock(file);
12158 +
12159 +       err = vfsub_write_u(h_file, buf, count, ppos);
12160 +       ii_write_lock_child(inode);
12161 +       au_cpup_attr_timesizes(inode);
12162 +       inode->i_mode = file_inode(h_file)->i_mode;
12163 +       ii_write_unlock(inode);
12164 +       fput(h_file);
12165 +
12166 +out:
12167 +       si_read_unlock(sb);
12168 +       mutex_unlock(&inode->i_mutex);
12169 +       return err;
12170 +}
12171 +
12172 +static ssize_t au_do_aio(struct file *h_file, int rw, struct kiocb *kio,
12173 +                        const struct iovec *iov, unsigned long nv, loff_t pos)
12174 +{
12175 +       ssize_t err;
12176 +       struct file *file;
12177 +       ssize_t (*func)(struct kiocb *, const struct iovec *, unsigned long,
12178 +                       loff_t);
12179 +
12180 +       err = security_file_permission(h_file, rw);
12181 +       if (unlikely(err))
12182 +               goto out;
12183 +
12184 +       err = -ENOSYS;
12185 +       func = NULL;
12186 +       if (rw == MAY_READ)
12187 +               func = h_file->f_op->aio_read;
12188 +       else if (rw == MAY_WRITE)
12189 +               func = h_file->f_op->aio_write;
12190 +       if (func) {
12191 +               file = kio->ki_filp;
12192 +               kio->ki_filp = h_file;
12193 +               lockdep_off();
12194 +               err = func(kio, iov, nv, pos);
12195 +               lockdep_on();
12196 +               kio->ki_filp = file;
12197 +       } else
12198 +               /* currently there is no such fs */
12199 +               WARN_ON_ONCE(1);
12200 +
12201 +out:
12202 +       return err;
12203 +}
12204 +
12205 +static ssize_t aufs_aio_read(struct kiocb *kio, const struct iovec *iov,
12206 +                            unsigned long nv, loff_t pos)
12207 +{
12208 +       ssize_t err;
12209 +       struct file *file, *h_file;
12210 +       struct dentry *dentry;
12211 +       struct super_block *sb;
12212 +
12213 +       file = kio->ki_filp;
12214 +       dentry = file->f_dentry;
12215 +       sb = dentry->d_sb;
12216 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
12217 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
12218 +       if (unlikely(err))
12219 +               goto out;
12220 +
12221 +       h_file = au_hf_top(file);
12222 +       get_file(h_file);
12223 +       di_read_unlock(dentry, AuLock_IR);
12224 +       fi_read_unlock(file);
12225 +
12226 +       err = au_do_aio(h_file, MAY_READ, kio, iov, nv, pos);
12227 +       /* todo: necessary? */
12228 +       /* file->f_ra = h_file->f_ra; */
12229 +       /* update without lock, I don't think it a problem */
12230 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
12231 +       fput(h_file);
12232 +
12233 +out:
12234 +       si_read_unlock(sb);
12235 +       return err;
12236 +}
12237 +
12238 +static ssize_t aufs_aio_write(struct kiocb *kio, const struct iovec *iov,
12239 +                             unsigned long nv, loff_t pos)
12240 +{
12241 +       ssize_t err;
12242 +       struct au_pin pin;
12243 +       struct dentry *dentry;
12244 +       struct inode *inode;
12245 +       struct file *file, *h_file;
12246 +       struct super_block *sb;
12247 +
12248 +       file = kio->ki_filp;
12249 +       dentry = file->f_dentry;
12250 +       sb = dentry->d_sb;
12251 +       inode = dentry->d_inode;
12252 +       au_mtx_and_read_lock(inode);
12253 +
12254 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
12255 +       if (unlikely(err))
12256 +               goto out;
12257 +
12258 +       err = au_ready_to_write(file, -1, &pin);
12259 +       di_downgrade_lock(dentry, AuLock_IR);
12260 +       if (unlikely(err)) {
12261 +               di_read_unlock(dentry, AuLock_IR);
12262 +               fi_write_unlock(file);
12263 +               goto out;
12264 +       }
12265 +
12266 +       h_file = au_hf_top(file);
12267 +       get_file(h_file);
12268 +       au_unpin(&pin);
12269 +       di_read_unlock(dentry, AuLock_IR);
12270 +       fi_write_unlock(file);
12271 +
12272 +       err = au_do_aio(h_file, MAY_WRITE, kio, iov, nv, pos);
12273 +       ii_write_lock_child(inode);
12274 +       au_cpup_attr_timesizes(inode);
12275 +       inode->i_mode = file_inode(h_file)->i_mode;
12276 +       ii_write_unlock(inode);
12277 +       fput(h_file);
12278 +
12279 +out:
12280 +       si_read_unlock(sb);
12281 +       mutex_unlock(&inode->i_mutex);
12282 +       return err;
12283 +}
12284 +
12285 +static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
12286 +                               struct pipe_inode_info *pipe, size_t len,
12287 +                               unsigned int flags)
12288 +{
12289 +       ssize_t err;
12290 +       struct file *h_file;
12291 +       struct dentry *dentry;
12292 +       struct super_block *sb;
12293 +
12294 +       dentry = file->f_dentry;
12295 +       sb = dentry->d_sb;
12296 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
12297 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
12298 +       if (unlikely(err))
12299 +               goto out;
12300 +
12301 +       err = -EINVAL;
12302 +       h_file = au_hf_top(file);
12303 +       get_file(h_file);
12304 +       if (au_test_loopback_kthread()) {
12305 +               au_warn_loopback(h_file->f_dentry->d_sb);
12306 +               if (file->f_mapping != h_file->f_mapping) {
12307 +                       file->f_mapping = h_file->f_mapping;
12308 +                       smp_mb(); /* unnecessary? */
12309 +               }
12310 +       }
12311 +       di_read_unlock(dentry, AuLock_IR);
12312 +       fi_read_unlock(file);
12313 +
12314 +       err = vfsub_splice_to(h_file, ppos, pipe, len, flags);
12315 +       /* todo: necessasry? */
12316 +       /* file->f_ra = h_file->f_ra; */
12317 +       /* update without lock, I don't think it a problem */
12318 +       fsstack_copy_attr_atime(dentry->d_inode, file_inode(h_file));
12319 +       fput(h_file);
12320 +
12321 +out:
12322 +       si_read_unlock(sb);
12323 +       return err;
12324 +}
12325 +
12326 +static ssize_t
12327 +aufs_splice_write(struct pipe_inode_info *pipe, struct file *file, loff_t *ppos,
12328 +                 size_t len, unsigned int flags)
12329 +{
12330 +       ssize_t err;
12331 +       struct au_pin pin;
12332 +       struct dentry *dentry;
12333 +       struct inode *inode;
12334 +       struct file *h_file;
12335 +       struct super_block *sb;
12336 +
12337 +       dentry = file->f_dentry;
12338 +       sb = dentry->d_sb;
12339 +       inode = dentry->d_inode;
12340 +       au_mtx_and_read_lock(inode);
12341 +
12342 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
12343 +       if (unlikely(err))
12344 +               goto out;
12345 +
12346 +       err = au_ready_to_write(file, -1, &pin);
12347 +       di_downgrade_lock(dentry, AuLock_IR);
12348 +       if (unlikely(err)) {
12349 +               di_read_unlock(dentry, AuLock_IR);
12350 +               fi_write_unlock(file);
12351 +               goto out;
12352 +       }
12353 +
12354 +       h_file = au_hf_top(file);
12355 +       get_file(h_file);
12356 +       au_unpin(&pin);
12357 +       di_read_unlock(dentry, AuLock_IR);
12358 +       fi_write_unlock(file);
12359 +
12360 +       err = vfsub_splice_from(pipe, h_file, ppos, len, flags);
12361 +       ii_write_lock_child(inode);
12362 +       au_cpup_attr_timesizes(inode);
12363 +       inode->i_mode = file_inode(h_file)->i_mode;
12364 +       ii_write_unlock(inode);
12365 +       fput(h_file);
12366 +
12367 +out:
12368 +       si_read_unlock(sb);
12369 +       mutex_unlock(&inode->i_mutex);
12370 +       return err;
12371 +}
12372 +
12373 +/* ---------------------------------------------------------------------- */
12374 +
12375 +/*
12376 + * The locking order around current->mmap_sem.
12377 + * - in most and regular cases
12378 + *   file I/O syscall -- aufs_read() or something
12379 + *     -- si_rwsem for read -- mmap_sem
12380 + *     (Note that [fdi]i_rwsem are released before mmap_sem).
12381 + * - in mmap case
12382 + *   mmap(2) -- mmap_sem -- aufs_mmap() -- si_rwsem for read -- [fdi]i_rwsem
12383 + * This AB-BA order is definitly bad, but is not a problem since "si_rwsem for
12384 + * read" allows muliple processes to acquire it and [fdi]i_rwsem are not held in
12385 + * file I/O. Aufs needs to stop lockdep in aufs_mmap() though.
12386 + * It means that when aufs acquires si_rwsem for write, the process should never
12387 + * acquire mmap_sem.
12388 + *
12389 + * Actually aufs_iterate() holds [fdi]i_rwsem before mmap_sem, but this is not a
12390 + * problem either since any directory is not able to be mmap-ed.
12391 + * The similar scenario is applied to aufs_readlink() too.
12392 + */
12393 +
12394 +/* cf. linux/include/linux/mman.h: calc_vm_prot_bits() */
12395 +#define AuConv_VM_PROT(f, b)   _calc_vm_trans(f, VM_##b, PROT_##b)
12396 +
12397 +static unsigned long au_arch_prot_conv(unsigned long flags)
12398 +{
12399 +       /* currently ppc64 only */
12400 +#ifdef CONFIG_PPC64
12401 +       /* cf. linux/arch/powerpc/include/asm/mman.h */
12402 +       AuDebugOn(arch_calc_vm_prot_bits(-1) != VM_SAO);
12403 +       return AuConv_VM_PROT(flags, SAO);
12404 +#else
12405 +       AuDebugOn(arch_calc_vm_prot_bits(-1));
12406 +       return 0;
12407 +#endif
12408 +}
12409 +
12410 +static unsigned long au_prot_conv(unsigned long flags)
12411 +{
12412 +       return AuConv_VM_PROT(flags, READ)
12413 +               | AuConv_VM_PROT(flags, WRITE)
12414 +               | AuConv_VM_PROT(flags, EXEC)
12415 +               | au_arch_prot_conv(flags);
12416 +}
12417 +
12418 +/* cf. linux/include/linux/mman.h: calc_vm_flag_bits() */
12419 +#define AuConv_VM_MAP(f, b)    _calc_vm_trans(f, VM_##b, MAP_##b)
12420 +
12421 +static unsigned long au_flag_conv(unsigned long flags)
12422 +{
12423 +       return AuConv_VM_MAP(flags, GROWSDOWN)
12424 +               | AuConv_VM_MAP(flags, DENYWRITE)
12425 +               | AuConv_VM_MAP(flags, LOCKED);
12426 +}
12427 +
12428 +static int aufs_mmap(struct file *file, struct vm_area_struct *vma)
12429 +{
12430 +       int err;
12431 +       aufs_bindex_t bstart;
12432 +       const unsigned char wlock
12433 +               = (file->f_mode & FMODE_WRITE) && (vma->vm_flags & VM_SHARED);
12434 +       struct dentry *dentry;
12435 +       struct super_block *sb;
12436 +       struct file *h_file;
12437 +       struct au_branch *br;
12438 +       struct au_pin pin;
12439 +
12440 +       AuDbgVmRegion(file, vma);
12441 +
12442 +       dentry = file->f_dentry;
12443 +       sb = dentry->d_sb;
12444 +       lockdep_off();
12445 +       si_read_lock(sb, AuLock_NOPLMW);
12446 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
12447 +       if (unlikely(err))
12448 +               goto out;
12449 +
12450 +       if (wlock) {
12451 +               err = au_ready_to_write(file, -1, &pin);
12452 +               di_write_unlock(dentry);
12453 +               if (unlikely(err)) {
12454 +                       fi_write_unlock(file);
12455 +                       goto out;
12456 +               }
12457 +               au_unpin(&pin);
12458 +       } else
12459 +               di_write_unlock(dentry);
12460 +
12461 +       bstart = au_fbstart(file);
12462 +       br = au_sbr(sb, bstart);
12463 +       h_file = au_hf_top(file);
12464 +       get_file(h_file);
12465 +       au_set_mmapped(file);
12466 +       fi_write_unlock(file);
12467 +       lockdep_on();
12468 +
12469 +       au_vm_file_reset(vma, h_file);
12470 +       err = security_mmap_file(h_file, au_prot_conv(vma->vm_flags),
12471 +                                au_flag_conv(vma->vm_flags));
12472 +       if (!err)
12473 +               err = h_file->f_op->mmap(h_file, vma);
12474 +       if (unlikely(err))
12475 +               goto out_reset;
12476 +
12477 +       au_vm_prfile_set(vma, file);
12478 +       /* update without lock, I don't think it a problem */
12479 +       fsstack_copy_attr_atime(file_inode(file), file_inode(h_file));
12480 +       goto out_fput; /* success */
12481 +
12482 +out_reset:
12483 +       au_unset_mmapped(file);
12484 +       au_vm_file_reset(vma, file);
12485 +out_fput:
12486 +       fput(h_file);
12487 +       lockdep_off();
12488 +out:
12489 +       si_read_unlock(sb);
12490 +       lockdep_on();
12491 +       AuTraceErr(err);
12492 +       return err;
12493 +}
12494 +
12495 +/* ---------------------------------------------------------------------- */
12496 +
12497 +static int aufs_fsync_nondir(struct file *file, loff_t start, loff_t end,
12498 +                            int datasync)
12499 +{
12500 +       int err;
12501 +       struct au_pin pin;
12502 +       struct dentry *dentry;
12503 +       struct inode *inode;
12504 +       struct file *h_file;
12505 +       struct super_block *sb;
12506 +
12507 +       dentry = file->f_dentry;
12508 +       inode = dentry->d_inode;
12509 +       sb = dentry->d_sb;
12510 +       mutex_lock(&inode->i_mutex);
12511 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
12512 +       if (unlikely(err))
12513 +               goto out;
12514 +
12515 +       err = 0; /* -EBADF; */ /* posix? */
12516 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
12517 +               goto out_si;
12518 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
12519 +       if (unlikely(err))
12520 +               goto out_si;
12521 +
12522 +       err = au_ready_to_write(file, -1, &pin);
12523 +       di_downgrade_lock(dentry, AuLock_IR);
12524 +       if (unlikely(err))
12525 +               goto out_unlock;
12526 +       au_unpin(&pin);
12527 +
12528 +       err = -EINVAL;
12529 +       h_file = au_hf_top(file);
12530 +       err = vfsub_fsync(h_file, &h_file->f_path, datasync);
12531 +       au_cpup_attr_timesizes(inode);
12532 +
12533 +out_unlock:
12534 +       di_read_unlock(dentry, AuLock_IR);
12535 +       fi_write_unlock(file);
12536 +out_si:
12537 +       si_read_unlock(sb);
12538 +out:
12539 +       mutex_unlock(&inode->i_mutex);
12540 +       return err;
12541 +}
12542 +
12543 +/* no one supports this operation, currently */
12544 +#if 0
12545 +static int aufs_aio_fsync_nondir(struct kiocb *kio, int datasync)
12546 +{
12547 +       int err;
12548 +       struct au_pin pin;
12549 +       struct dentry *dentry;
12550 +       struct inode *inode;
12551 +       struct file *file, *h_file;
12552 +
12553 +       file = kio->ki_filp;
12554 +       dentry = file->f_dentry;
12555 +       inode = dentry->d_inode;
12556 +       au_mtx_and_read_lock(inode);
12557 +
12558 +       err = 0; /* -EBADF; */ /* posix? */
12559 +       if (unlikely(!(file->f_mode & FMODE_WRITE)))
12560 +               goto out;
12561 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
12562 +       if (unlikely(err))
12563 +               goto out;
12564 +
12565 +       err = au_ready_to_write(file, -1, &pin);
12566 +       di_downgrade_lock(dentry, AuLock_IR);
12567 +       if (unlikely(err))
12568 +               goto out_unlock;
12569 +       au_unpin(&pin);
12570 +
12571 +       err = -ENOSYS;
12572 +       h_file = au_hf_top(file);
12573 +       if (h_file->f_op->aio_fsync) {
12574 +               struct mutex *h_mtx;
12575 +
12576 +               h_mtx = &file_inode(h_file)->i_mutex;
12577 +               if (!is_sync_kiocb(kio)) {
12578 +                       get_file(h_file);
12579 +                       fput(file);
12580 +               }
12581 +               kio->ki_filp = h_file;
12582 +               err = h_file->f_op->aio_fsync(kio, datasync);
12583 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
12584 +               if (!err)
12585 +                       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL);
12586 +               /*ignore*/
12587 +               au_cpup_attr_timesizes(inode);
12588 +               mutex_unlock(h_mtx);
12589 +       }
12590 +
12591 +out_unlock:
12592 +       di_read_unlock(dentry, AuLock_IR);
12593 +       fi_write_unlock(file);
12594 +out:
12595 +       si_read_unlock(inode->sb);
12596 +       mutex_unlock(&inode->i_mutex);
12597 +       return err;
12598 +}
12599 +#endif
12600 +
12601 +static int aufs_fasync(int fd, struct file *file, int flag)
12602 +{
12603 +       int err;
12604 +       struct file *h_file;
12605 +       struct dentry *dentry;
12606 +       struct super_block *sb;
12607 +
12608 +       dentry = file->f_dentry;
12609 +       sb = dentry->d_sb;
12610 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
12611 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
12612 +       if (unlikely(err))
12613 +               goto out;
12614 +
12615 +       h_file = au_hf_top(file);
12616 +       if (h_file->f_op->fasync)
12617 +               err = h_file->f_op->fasync(fd, h_file, flag);
12618 +
12619 +       di_read_unlock(dentry, AuLock_IR);
12620 +       fi_read_unlock(file);
12621 +
12622 +out:
12623 +       si_read_unlock(sb);
12624 +       return err;
12625 +}
12626 +
12627 +/* ---------------------------------------------------------------------- */
12628 +
12629 +/* no one supports this operation, currently */
12630 +#if 0
12631 +static ssize_t aufs_sendpage(struct file *file, struct page *page, int offset,
12632 +                            size_t len, loff_t *pos , int more)
12633 +{
12634 +}
12635 +#endif
12636 +
12637 +/* ---------------------------------------------------------------------- */
12638 +
12639 +const struct file_operations aufs_file_fop = {
12640 +       .owner          = THIS_MODULE,
12641 +
12642 +       .llseek         = default_llseek,
12643 +
12644 +       .read           = aufs_read,
12645 +       .write          = aufs_write,
12646 +       .aio_read       = aufs_aio_read,
12647 +       .aio_write      = aufs_aio_write,
12648 +#ifdef CONFIG_AUFS_POLL
12649 +       .poll           = aufs_poll,
12650 +#endif
12651 +       .unlocked_ioctl = aufs_ioctl_nondir,
12652 +#ifdef CONFIG_COMPAT
12653 +       .compat_ioctl   = aufs_compat_ioctl_nondir,
12654 +#endif
12655 +       .mmap           = aufs_mmap,
12656 +       .open           = aufs_open_nondir,
12657 +       .flush          = aufs_flush_nondir,
12658 +       .release        = aufs_release_nondir,
12659 +       .fsync          = aufs_fsync_nondir,
12660 +       /* .aio_fsync   = aufs_aio_fsync_nondir, */
12661 +       .fasync         = aufs_fasync,
12662 +       /* .sendpage    = aufs_sendpage, */
12663 +       .splice_write   = aufs_splice_write,
12664 +       .splice_read    = aufs_splice_read,
12665 +#if 0
12666 +       .aio_splice_write = aufs_aio_splice_write,
12667 +       .aio_splice_read  = aufs_aio_splice_read
12668 +#endif
12669 +};
12670 diff -urN /usr/share/empty/fs/aufs/f_op_sp.c linux/fs/aufs/f_op_sp.c
12671 --- /usr/share/empty/fs/aufs/f_op_sp.c  1970-01-01 01:00:00.000000000 +0100
12672 +++ linux/fs/aufs/f_op_sp.c     2014-04-24 22:11:10.851935747 +0200
12673 @@ -0,0 +1,382 @@
12674 +/*
12675 + * Copyright (C) 2005-2014 Junjiro R. Okajima
12676 + *
12677 + * This program, aufs is free software; you can redistribute it and/or modify
12678 + * it under the terms of the GNU General Public License as published by
12679 + * the Free Software Foundation; either version 2 of the License, or
12680 + * (at your option) any later version.
12681 + *
12682 + * This program is distributed in the hope that it will be useful,
12683 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
12684 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12685 + * GNU General Public License for more details.
12686 + *
12687 + * You should have received a copy of the GNU General Public License
12688 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
12689 + */
12690 +
12691 +/*
12692 + * file operations for special files.
12693 + * while they exist in aufs virtually,
12694 + * their file I/O is handled out of aufs.
12695 + */
12696 +
12697 +#include <linux/aio.h>
12698 +#include "aufs.h"
12699 +
12700 +/*
12701 + * I don't think the size of this list grows much.
12702 + * so here is a very simple list implemented in order to find finfo matching a
12703 + * given file.
12704 + */
12705 +static struct au_sphlhead au_finfo_sp = {
12706 +       .spin   = __SPIN_LOCK_INITIALIZER(au_finfo_sp.spin),
12707 +       .head   = HLIST_HEAD_INIT
12708 +};
12709 +
12710 +struct au_finfo_sp {
12711 +       struct hlist_node       hlist;
12712 +       struct file             *file;
12713 +       struct au_finfo         *finfo;
12714 +};
12715 +
12716 +struct au_finfo *au_fi_sp(struct file *file)
12717 +{
12718 +       struct au_finfo *finfo;
12719 +       struct au_finfo_sp *sp;
12720 +
12721 +       finfo = NULL;
12722 +       spin_lock(&au_finfo_sp.spin);
12723 +       hlist_for_each_entry(sp, &au_finfo_sp.head, hlist) {
12724 +               if (sp->file != file)
12725 +                       continue;
12726 +               finfo = sp->finfo;
12727 +               break;
12728 +       }
12729 +       spin_unlock(&au_finfo_sp.spin);
12730 +
12731 +       return finfo;
12732 +}
12733 +
12734 +static int au_fi_sp_add(struct file *file)
12735 +{
12736 +       int err;
12737 +       struct au_finfo_sp *sp;
12738 +
12739 +       err = -ENOMEM;
12740 +       sp = kmalloc(sizeof(*sp), GFP_NOFS);
12741 +       if (sp) {
12742 +               err = 0;
12743 +               sp->file = file;
12744 +               sp->finfo = file->private_data;
12745 +               spin_lock(&au_finfo_sp.spin);
12746 +               hlist_add_head(&sp->hlist, &au_finfo_sp.head);
12747 +               spin_unlock(&au_finfo_sp.spin);
12748 +       }
12749 +       return err;
12750 +}
12751 +
12752 +static void au_fi_sp_del(struct file *file)
12753 +{
12754 +       struct au_finfo_sp *sp, *do_free;
12755 +
12756 +       do_free = NULL;
12757 +       spin_lock(&au_finfo_sp.spin);
12758 +       hlist_for_each_entry(sp, &au_finfo_sp.head, hlist) {
12759 +               if (sp->file != file)
12760 +                       continue;
12761 +               hlist_del(&sp->hlist);
12762 +               do_free = sp;
12763 +               break;
12764 +       }
12765 +       spin_unlock(&au_finfo_sp.spin);
12766 +       kfree(do_free);
12767 +}
12768 +
12769 +/* ---------------------------------------------------------------------- */
12770 +
12771 +static ssize_t aufs_aio_read_sp(struct kiocb *kio, const struct iovec *iov,
12772 +                               unsigned long nv, loff_t pos)
12773 +{
12774 +       ssize_t err;
12775 +       aufs_bindex_t bstart;
12776 +       unsigned char wbr;
12777 +       struct file *file, *h_file;
12778 +       struct super_block *sb;
12779 +
12780 +       file = kio->ki_filp;
12781 +       sb = file->f_dentry->d_sb;
12782 +       si_read_lock(sb, AuLock_FLUSH);
12783 +       fi_read_lock(file);
12784 +       bstart = au_fbstart(file);
12785 +       h_file = au_hf_top(file);
12786 +       fi_read_unlock(file);
12787 +       wbr = !!au_br_writable(au_sbr(sb, bstart)->br_perm);
12788 +       si_read_unlock(sb);
12789 +
12790 +       /* do not change the file in kio */
12791 +       AuDebugOn(!h_file->f_op || !h_file->f_op->aio_read);
12792 +       err = h_file->f_op->aio_read(kio, iov, nv, pos);
12793 +       if (err > 0 && wbr)
12794 +               file_accessed(h_file);
12795 +
12796 +       return err;
12797 +}
12798 +
12799 +static ssize_t aufs_aio_write_sp(struct kiocb *kio, const struct iovec *iov,
12800 +                                unsigned long nv, loff_t pos)
12801 +{
12802 +       ssize_t err;
12803 +       aufs_bindex_t bstart;
12804 +       unsigned char wbr;
12805 +       struct super_block *sb;
12806 +       struct file *file, *h_file;
12807 +
12808 +       file = kio->ki_filp;
12809 +       sb = file->f_dentry->d_sb;
12810 +       si_read_lock(sb, AuLock_FLUSH);
12811 +       fi_read_lock(file);
12812 +       bstart = au_fbstart(file);
12813 +       h_file = au_hf_top(file);
12814 +       fi_read_unlock(file);
12815 +       wbr = !!au_br_writable(au_sbr(sb, bstart)->br_perm);
12816 +       si_read_unlock(sb);
12817 +
12818 +       /* do not change the file in kio */
12819 +       AuDebugOn(!h_file->f_op || !h_file->f_op->aio_write);
12820 +       err = h_file->f_op->aio_write(kio, iov, nv, pos);
12821 +       return err;
12822 +}
12823 +
12824 +/* ---------------------------------------------------------------------- */
12825 +
12826 +static int aufs_release_sp(struct inode *inode, struct file *file)
12827 +{
12828 +       int err;
12829 +       struct file *h_file;
12830 +
12831 +       fi_read_lock(file);
12832 +       h_file = au_hf_top(file);
12833 +       fi_read_unlock(file);
12834 +       /* close this fifo in aufs */
12835 +       err = h_file->f_op->release(inode, file); /* ignore */
12836 +       aufs_release_nondir(inode, file); /* ignore */
12837 +       au_fi_sp_del(file);
12838 +       return err;
12839 +}
12840 +
12841 +/* ---------------------------------------------------------------------- */
12842 +
12843 +/* currently, support only FIFO */
12844 +enum {
12845 +       AuSp_FIFO, AuSp_FIFO_R, AuSp_FIFO_W, AuSp_FIFO_RW,
12846 +       /* AuSp_SOCK, AuSp_CHR, AuSp_BLK, */
12847 +       AuSp_Last
12848 +};
12849 +static int aufs_open_sp(struct inode *inode, struct file *file);
12850 +static struct au_sp_fop {
12851 +       int                     done;
12852 +       struct file_operations  fop;    /* not 'const' */
12853 +       spinlock_t              spin;
12854 +} au_sp_fop[AuSp_Last] = {
12855 +       [AuSp_FIFO] = {
12856 +               .fop    = {
12857 +                       .owner  = THIS_MODULE,
12858 +                       .open   = aufs_open_sp
12859 +               }
12860 +       }
12861 +};
12862 +
12863 +static void au_init_fop_sp(struct file *file)
12864 +{
12865 +       struct au_sp_fop *p;
12866 +       int i;
12867 +       struct file *h_file;
12868 +
12869 +       p = au_sp_fop;
12870 +       if (unlikely(!p->done)) {
12871 +               /* initialize first time only */
12872 +               static DEFINE_SPINLOCK(spin);
12873 +
12874 +               spin_lock(&spin);
12875 +               if (!p->done) {
12876 +                       BUILD_BUG_ON(sizeof(au_sp_fop)/sizeof(*au_sp_fop)
12877 +                                    != AuSp_Last);
12878 +                       for (i = 0; i < AuSp_Last; i++)
12879 +                               spin_lock_init(&p[i].spin);
12880 +                       p->done = 1;
12881 +               }
12882 +               spin_unlock(&spin);
12883 +       }
12884 +
12885 +       switch (file->f_mode & (FMODE_READ | FMODE_WRITE)) {
12886 +       case FMODE_READ:
12887 +               i = AuSp_FIFO_R;
12888 +               break;
12889 +       case FMODE_WRITE:
12890 +               i = AuSp_FIFO_W;
12891 +               break;
12892 +       case FMODE_READ | FMODE_WRITE:
12893 +               i = AuSp_FIFO_RW;
12894 +               break;
12895 +       default:
12896 +               BUG();
12897 +       }
12898 +
12899 +       p += i;
12900 +       if (unlikely(!p->done)) {
12901 +               /* initialize first time only */
12902 +               h_file = au_hf_top(file);
12903 +               spin_lock(&p->spin);
12904 +               if (!p->done) {
12905 +                       p->fop = *h_file->f_op;
12906 +                       p->fop.owner = THIS_MODULE;
12907 +                       if (p->fop.aio_read)
12908 +                               p->fop.aio_read = aufs_aio_read_sp;
12909 +                       if (p->fop.aio_write)
12910 +                               p->fop.aio_write = aufs_aio_write_sp;
12911 +                       p->fop.release = aufs_release_sp;
12912 +                       p->done = 1;
12913 +               }
12914 +               spin_unlock(&p->spin);
12915 +       }
12916 +       file->f_op = &p->fop;
12917 +}
12918 +
12919 +static int au_cpup_sp(struct dentry *dentry)
12920 +{
12921 +       int err;
12922 +       struct au_pin pin;
12923 +       struct au_wr_dir_args wr_dir_args = {
12924 +               .force_btgt     = -1,
12925 +               .flags          = 0
12926 +       };
12927 +       struct au_cp_generic cpg = {
12928 +               .dentry = dentry,
12929 +               .bdst   = -1,
12930 +               .bsrc   = -1,
12931 +               .len    = -1,
12932 +               .pin    = &pin,
12933 +               .flags  = AuCpup_DTIME
12934 +       };
12935 +
12936 +       AuDbg("%pd\n", dentry);
12937 +
12938 +       di_read_unlock(dentry, AuLock_IR);
12939 +       di_write_lock_child(dentry);
12940 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
12941 +       if (unlikely(err < 0))
12942 +               goto out;
12943 +       cpg.bdst = err;
12944 +       err = 0;
12945 +       if (cpg.bdst == au_dbstart(dentry))
12946 +               goto out; /* success */
12947 +
12948 +       err = au_pin(&pin, dentry, cpg.bdst, au_opt_udba(dentry->d_sb),
12949 +                    AuPin_MNT_WRITE);
12950 +       if (!err) {
12951 +               err = au_sio_cpup_simple(&cpg);
12952 +               au_unpin(&pin);
12953 +       }
12954 +
12955 +out:
12956 +       di_downgrade_lock(dentry, AuLock_IR);
12957 +       return err;
12958 +}
12959 +
12960 +static int au_do_open_sp(struct file *file, int flags)
12961 +{
12962 +       int err;
12963 +       struct dentry *dentry;
12964 +       struct super_block *sb;
12965 +       struct file *h_file;
12966 +       struct inode *h_inode;
12967 +
12968 +       err = au_fi_sp_add(file);
12969 +       if (unlikely(err))
12970 +               goto out;
12971 +
12972 +       dentry = file->f_dentry;
12973 +       AuDbg("%pd\n", dentry);
12974 +
12975 +       /*
12976 +        * try copying-up.
12977 +        * operate on the ro branch is not an error.
12978 +        */
12979 +       au_cpup_sp(dentry); /* ignore */
12980 +
12981 +       /* prepare h_file */
12982 +       err = au_do_open_nondir(file, vfsub_file_flags(file));
12983 +       if (unlikely(err))
12984 +               goto out_del;
12985 +
12986 +       sb = dentry->d_sb;
12987 +       h_file = au_hf_top(file);
12988 +       h_inode = file_inode(h_file);
12989 +       di_read_unlock(dentry, AuLock_IR);
12990 +       fi_write_unlock(file);
12991 +       si_read_unlock(sb);
12992 +       /* open this fifo in aufs */
12993 +       err = h_inode->i_fop->open(file_inode(file), file);
12994 +       si_noflush_read_lock(sb);
12995 +       fi_write_lock(file);
12996 +       di_read_lock_child(dentry, AuLock_IR);
12997 +       if (!err) {
12998 +               au_init_fop_sp(file);
12999 +               goto out; /* success */
13000 +       }
13001 +
13002 +out_del:
13003 +       au_fi_sp_del(file);
13004 +out:
13005 +       return err;
13006 +}
13007 +
13008 +static int aufs_open_sp(struct inode *inode, struct file *file)
13009 +{
13010 +       int err;
13011 +       struct super_block *sb;
13012 +
13013 +       sb = file->f_dentry->d_sb;
13014 +       si_read_lock(sb, AuLock_FLUSH);
13015 +       err = au_do_open(file, au_do_open_sp, /*fidir*/NULL);
13016 +       si_read_unlock(sb);
13017 +       return err;
13018 +}
13019 +
13020 +/* ---------------------------------------------------------------------- */
13021 +
13022 +void au_init_special_fop(struct inode *inode, umode_t mode, dev_t rdev)
13023 +{
13024 +       init_special_inode(inode, mode, rdev);
13025 +
13026 +       switch (mode & S_IFMT) {
13027 +       case S_IFIFO:
13028 +               inode->i_fop = &au_sp_fop[AuSp_FIFO].fop;
13029 +               /*FALLTHROUGH*/
13030 +       case S_IFCHR:
13031 +       case S_IFBLK:
13032 +       case S_IFSOCK:
13033 +               break;
13034 +       default:
13035 +               AuDebugOn(1);
13036 +       }
13037 +}
13038 +
13039 +int au_special_file(umode_t mode)
13040 +{
13041 +       int ret;
13042 +
13043 +       ret = 0;
13044 +       switch (mode & S_IFMT) {
13045 +       case S_IFIFO:
13046 +#if 0
13047 +       case S_IFCHR:
13048 +       case S_IFBLK:
13049 +       case S_IFSOCK:
13050 +#endif
13051 +               ret = 1;
13052 +       }
13053 +
13054 +       return ret;
13055 +}
13056 diff -urN /usr/share/empty/fs/aufs/fstype.h linux/fs/aufs/fstype.h
13057 --- /usr/share/empty/fs/aufs/fstype.h   1970-01-01 01:00:00.000000000 +0100
13058 +++ linux/fs/aufs/fstype.h      2014-04-24 22:11:10.851935747 +0200
13059 @@ -0,0 +1,469 @@
13060 +/*
13061 + * Copyright (C) 2005-2014 Junjiro R. Okajima
13062 + *
13063 + * This program, aufs is free software; you can redistribute it and/or modify
13064 + * it under the terms of the GNU General Public License as published by
13065 + * the Free Software Foundation; either version 2 of the License, or
13066 + * (at your option) any later version.
13067 + *
13068 + * This program is distributed in the hope that it will be useful,
13069 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13070 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13071 + * GNU General Public License for more details.
13072 + *
13073 + * You should have received a copy of the GNU General Public License
13074 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13075 + */
13076 +
13077 +/*
13078 + * judging filesystem type
13079 + */
13080 +
13081 +#ifndef __AUFS_FSTYPE_H__
13082 +#define __AUFS_FSTYPE_H__
13083 +
13084 +#ifdef __KERNEL__
13085 +
13086 +#include <linux/fs.h>
13087 +#include <linux/magic.h>
13088 +#include <linux/romfs_fs.h>
13089 +
13090 +static inline int au_test_aufs(struct super_block *sb)
13091 +{
13092 +       return sb->s_magic == AUFS_SUPER_MAGIC;
13093 +}
13094 +
13095 +static inline const char *au_sbtype(struct super_block *sb)
13096 +{
13097 +       return sb->s_type->name;
13098 +}
13099 +
13100 +static inline int au_test_iso9660(struct super_block *sb __maybe_unused)
13101 +{
13102 +#if defined(CONFIG_ROMFS_FS) || defined(CONFIG_ROMFS_FS_MODULE)
13103 +       return sb->s_magic == ROMFS_MAGIC;
13104 +#else
13105 +       return 0;
13106 +#endif
13107 +}
13108 +
13109 +static inline int au_test_romfs(struct super_block *sb __maybe_unused)
13110 +{
13111 +#if defined(CONFIG_ISO9660_FS) || defined(CONFIG_ISO9660_FS_MODULE)
13112 +       return sb->s_magic == ISOFS_SUPER_MAGIC;
13113 +#else
13114 +       return 0;
13115 +#endif
13116 +}
13117 +
13118 +static inline int au_test_cramfs(struct super_block *sb __maybe_unused)
13119 +{
13120 +#if defined(CONFIG_CRAMFS) || defined(CONFIG_CRAMFS_MODULE)
13121 +       return sb->s_magic == CRAMFS_MAGIC;
13122 +#endif
13123 +       return 0;
13124 +}
13125 +
13126 +static inline int au_test_nfs(struct super_block *sb __maybe_unused)
13127 +{
13128 +#if defined(CONFIG_NFS_FS) || defined(CONFIG_NFS_FS_MODULE)
13129 +       return sb->s_magic == NFS_SUPER_MAGIC;
13130 +#else
13131 +       return 0;
13132 +#endif
13133 +}
13134 +
13135 +static inline int au_test_fuse(struct super_block *sb __maybe_unused)
13136 +{
13137 +#if defined(CONFIG_FUSE_FS) || defined(CONFIG_FUSE_FS_MODULE)
13138 +       return sb->s_magic == FUSE_SUPER_MAGIC;
13139 +#else
13140 +       return 0;
13141 +#endif
13142 +}
13143 +
13144 +static inline int au_test_xfs(struct super_block *sb __maybe_unused)
13145 +{
13146 +#if defined(CONFIG_XFS_FS) || defined(CONFIG_XFS_FS_MODULE)
13147 +       return sb->s_magic == XFS_SB_MAGIC;
13148 +#else
13149 +       return 0;
13150 +#endif
13151 +}
13152 +
13153 +static inline int au_test_tmpfs(struct super_block *sb __maybe_unused)
13154 +{
13155 +#ifdef CONFIG_TMPFS
13156 +       return sb->s_magic == TMPFS_MAGIC;
13157 +#else
13158 +       return 0;
13159 +#endif
13160 +}
13161 +
13162 +static inline int au_test_ecryptfs(struct super_block *sb __maybe_unused)
13163 +{
13164 +#if defined(CONFIG_ECRYPT_FS) || defined(CONFIG_ECRYPT_FS_MODULE)
13165 +       return !strcmp(au_sbtype(sb), "ecryptfs");
13166 +#else
13167 +       return 0;
13168 +#endif
13169 +}
13170 +
13171 +static inline int au_test_ocfs2(struct super_block *sb __maybe_unused)
13172 +{
13173 +#if defined(CONFIG_OCFS2_FS) || defined(CONFIG_OCFS2_FS_MODULE)
13174 +       return sb->s_magic == OCFS2_SUPER_MAGIC;
13175 +#else
13176 +       return 0;
13177 +#endif
13178 +}
13179 +
13180 +static inline int au_test_ocfs2_dlmfs(struct super_block *sb __maybe_unused)
13181 +{
13182 +#if defined(CONFIG_OCFS2_FS_O2CB) || defined(CONFIG_OCFS2_FS_O2CB_MODULE)
13183 +       return sb->s_magic == DLMFS_MAGIC;
13184 +#else
13185 +       return 0;
13186 +#endif
13187 +}
13188 +
13189 +static inline int au_test_coda(struct super_block *sb __maybe_unused)
13190 +{
13191 +#if defined(CONFIG_CODA_FS) || defined(CONFIG_CODA_FS_MODULE)
13192 +       return sb->s_magic == CODA_SUPER_MAGIC;
13193 +#else
13194 +       return 0;
13195 +#endif
13196 +}
13197 +
13198 +static inline int au_test_v9fs(struct super_block *sb __maybe_unused)
13199 +{
13200 +#if defined(CONFIG_9P_FS) || defined(CONFIG_9P_FS_MODULE)
13201 +       return sb->s_magic == V9FS_MAGIC;
13202 +#else
13203 +       return 0;
13204 +#endif
13205 +}
13206 +
13207 +static inline int au_test_ext4(struct super_block *sb __maybe_unused)
13208 +{
13209 +#if defined(CONFIG_EXT4_FS) || defined(CONFIG_EXT4_FS_MODULE)
13210 +       return sb->s_magic == EXT4_SUPER_MAGIC;
13211 +#else
13212 +       return 0;
13213 +#endif
13214 +}
13215 +
13216 +static inline int au_test_sysv(struct super_block *sb __maybe_unused)
13217 +{
13218 +#if defined(CONFIG_SYSV_FS) || defined(CONFIG_SYSV_FS_MODULE)
13219 +       return !strcmp(au_sbtype(sb), "sysv");
13220 +#else
13221 +       return 0;
13222 +#endif
13223 +}
13224 +
13225 +static inline int au_test_ramfs(struct super_block *sb)
13226 +{
13227 +       return sb->s_magic == RAMFS_MAGIC;
13228 +}
13229 +
13230 +static inline int au_test_ubifs(struct super_block *sb __maybe_unused)
13231 +{
13232 +#if defined(CONFIG_UBIFS_FS) || defined(CONFIG_UBIFS_FS_MODULE)
13233 +       return sb->s_magic == UBIFS_SUPER_MAGIC;
13234 +#else
13235 +       return 0;
13236 +#endif
13237 +}
13238 +
13239 +static inline int au_test_procfs(struct super_block *sb __maybe_unused)
13240 +{
13241 +#ifdef CONFIG_PROC_FS
13242 +       return sb->s_magic == PROC_SUPER_MAGIC;
13243 +#else
13244 +       return 0;
13245 +#endif
13246 +}
13247 +
13248 +static inline int au_test_sysfs(struct super_block *sb __maybe_unused)
13249 +{
13250 +#ifdef CONFIG_SYSFS
13251 +       return sb->s_magic == SYSFS_MAGIC;
13252 +#else
13253 +       return 0;
13254 +#endif
13255 +}
13256 +
13257 +static inline int au_test_configfs(struct super_block *sb __maybe_unused)
13258 +{
13259 +#if defined(CONFIG_CONFIGFS_FS) || defined(CONFIG_CONFIGFS_FS_MODULE)
13260 +       return sb->s_magic == CONFIGFS_MAGIC;
13261 +#else
13262 +       return 0;
13263 +#endif
13264 +}
13265 +
13266 +static inline int au_test_minix(struct super_block *sb __maybe_unused)
13267 +{
13268 +#if defined(CONFIG_MINIX_FS) || defined(CONFIG_MINIX_FS_MODULE)
13269 +       return sb->s_magic == MINIX3_SUPER_MAGIC
13270 +               || sb->s_magic == MINIX2_SUPER_MAGIC
13271 +               || sb->s_magic == MINIX2_SUPER_MAGIC2
13272 +               || sb->s_magic == MINIX_SUPER_MAGIC
13273 +               || sb->s_magic == MINIX_SUPER_MAGIC2;
13274 +#else
13275 +       return 0;
13276 +#endif
13277 +}
13278 +
13279 +static inline int au_test_cifs(struct super_block *sb __maybe_unused)
13280 +{
13281 +#if defined(CONFIG_CIFS_FS) || defined(CONFIGCIFS_FS_MODULE)
13282 +       return sb->s_magic == CIFS_MAGIC_NUMBER;
13283 +#else
13284 +       return 0;
13285 +#endif
13286 +}
13287 +
13288 +static inline int au_test_fat(struct super_block *sb __maybe_unused)
13289 +{
13290 +#if defined(CONFIG_FAT_FS) || defined(CONFIG_FAT_FS_MODULE)
13291 +       return sb->s_magic == MSDOS_SUPER_MAGIC;
13292 +#else
13293 +       return 0;
13294 +#endif
13295 +}
13296 +
13297 +static inline int au_test_msdos(struct super_block *sb)
13298 +{
13299 +       return au_test_fat(sb);
13300 +}
13301 +
13302 +static inline int au_test_vfat(struct super_block *sb)
13303 +{
13304 +       return au_test_fat(sb);
13305 +}
13306 +
13307 +static inline int au_test_securityfs(struct super_block *sb __maybe_unused)
13308 +{
13309 +#ifdef CONFIG_SECURITYFS
13310 +       return sb->s_magic == SECURITYFS_MAGIC;
13311 +#else
13312 +       return 0;
13313 +#endif
13314 +}
13315 +
13316 +static inline int au_test_squashfs(struct super_block *sb __maybe_unused)
13317 +{
13318 +#if defined(CONFIG_SQUASHFS) || defined(CONFIG_SQUASHFS_MODULE)
13319 +       return sb->s_magic == SQUASHFS_MAGIC;
13320 +#else
13321 +       return 0;
13322 +#endif
13323 +}
13324 +
13325 +static inline int au_test_btrfs(struct super_block *sb __maybe_unused)
13326 +{
13327 +#if defined(CONFIG_BTRFS_FS) || defined(CONFIG_BTRFS_FS_MODULE)
13328 +       return sb->s_magic == BTRFS_SUPER_MAGIC;
13329 +#else
13330 +       return 0;
13331 +#endif
13332 +}
13333 +
13334 +static inline int au_test_xenfs(struct super_block *sb __maybe_unused)
13335 +{
13336 +#if defined(CONFIG_XENFS) || defined(CONFIG_XENFS_MODULE)
13337 +       return sb->s_magic == XENFS_SUPER_MAGIC;
13338 +#else
13339 +       return 0;
13340 +#endif
13341 +}
13342 +
13343 +static inline int au_test_debugfs(struct super_block *sb __maybe_unused)
13344 +{
13345 +#ifdef CONFIG_DEBUG_FS
13346 +       return sb->s_magic == DEBUGFS_MAGIC;
13347 +#else
13348 +       return 0;
13349 +#endif
13350 +}
13351 +
13352 +static inline int au_test_nilfs(struct super_block *sb __maybe_unused)
13353 +{
13354 +#if defined(CONFIG_NILFS) || defined(CONFIG_NILFS_MODULE)
13355 +       return sb->s_magic == NILFS_SUPER_MAGIC;
13356 +#else
13357 +       return 0;
13358 +#endif
13359 +}
13360 +
13361 +static inline int au_test_hfsplus(struct super_block *sb __maybe_unused)
13362 +{
13363 +#if defined(CONFIG_HFSPLUS_FS) || defined(CONFIG_HFSPLUS_FS_MODULE)
13364 +       return sb->s_magic == HFSPLUS_SUPER_MAGIC;
13365 +#else
13366 +       return 0;
13367 +#endif
13368 +}
13369 +
13370 +/* ---------------------------------------------------------------------- */
13371 +/*
13372 + * they can't be an aufs branch.
13373 + */
13374 +static inline int au_test_fs_unsuppoted(struct super_block *sb)
13375 +{
13376 +       return
13377 +#ifndef CONFIG_AUFS_BR_RAMFS
13378 +               au_test_ramfs(sb) ||
13379 +#endif
13380 +               au_test_procfs(sb)
13381 +               || au_test_sysfs(sb)
13382 +               || au_test_configfs(sb)
13383 +               || au_test_debugfs(sb)
13384 +               || au_test_securityfs(sb)
13385 +               || au_test_xenfs(sb)
13386 +               || au_test_ecryptfs(sb)
13387 +               /* || !strcmp(au_sbtype(sb), "unionfs") */
13388 +               || au_test_aufs(sb); /* will be supported in next version */
13389 +}
13390 +
13391 +static inline int au_test_fs_remote(struct super_block *sb)
13392 +{
13393 +       return !au_test_tmpfs(sb)
13394 +#ifdef CONFIG_AUFS_BR_RAMFS
13395 +               && !au_test_ramfs(sb)
13396 +#endif
13397 +               && !(sb->s_type->fs_flags & FS_REQUIRES_DEV);
13398 +}
13399 +
13400 +/* ---------------------------------------------------------------------- */
13401 +
13402 +/*
13403 + * Note: these functions (below) are created after reading ->getattr() in all
13404 + * filesystems under linux/fs. it means we have to do so in every update...
13405 + */
13406 +
13407 +/*
13408 + * some filesystems require getattr to refresh the inode attributes before
13409 + * referencing.
13410 + * in most cases, we can rely on the inode attribute in NFS (or every remote fs)
13411 + * and leave the work for d_revalidate()
13412 + */
13413 +static inline int au_test_fs_refresh_iattr(struct super_block *sb)
13414 +{
13415 +       return au_test_nfs(sb)
13416 +               || au_test_fuse(sb)
13417 +               /* || au_test_ocfs2(sb) */      /* untested */
13418 +               /* || au_test_btrfs(sb) */      /* untested */
13419 +               /* || au_test_coda(sb) */       /* untested */
13420 +               /* || au_test_v9fs(sb) */       /* untested */
13421 +               ;
13422 +}
13423 +
13424 +/*
13425 + * filesystems which don't maintain i_size or i_blocks.
13426 + */
13427 +static inline int au_test_fs_bad_iattr_size(struct super_block *sb)
13428 +{
13429 +       return au_test_xfs(sb)
13430 +               || au_test_btrfs(sb)
13431 +               || au_test_ubifs(sb)
13432 +               || au_test_hfsplus(sb)  /* maintained, but incorrect */
13433 +               /* || au_test_ext4(sb) */       /* untested */
13434 +               /* || au_test_ocfs2(sb) */      /* untested */
13435 +               /* || au_test_ocfs2_dlmfs(sb) */ /* untested */
13436 +               /* || au_test_sysv(sb) */       /* untested */
13437 +               /* || au_test_minix(sb) */      /* untested */
13438 +               ;
13439 +}
13440 +
13441 +/*
13442 + * filesystems which don't store the correct value in some of their inode
13443 + * attributes.
13444 + */
13445 +static inline int au_test_fs_bad_iattr(struct super_block *sb)
13446 +{
13447 +       return au_test_fs_bad_iattr_size(sb)
13448 +               /* || au_test_cifs(sb) */       /* untested */
13449 +               || au_test_fat(sb)
13450 +               || au_test_msdos(sb)
13451 +               || au_test_vfat(sb);
13452 +}
13453 +
13454 +/* they don't check i_nlink in link(2) */
13455 +static inline int au_test_fs_no_limit_nlink(struct super_block *sb)
13456 +{
13457 +       return au_test_tmpfs(sb)
13458 +#ifdef CONFIG_AUFS_BR_RAMFS
13459 +               || au_test_ramfs(sb)
13460 +#endif
13461 +               || au_test_ubifs(sb)
13462 +               || au_test_hfsplus(sb);
13463 +}
13464 +
13465 +/*
13466 + * filesystems which sets S_NOATIME and S_NOCMTIME.
13467 + */
13468 +static inline int au_test_fs_notime(struct super_block *sb)
13469 +{
13470 +       return au_test_nfs(sb)
13471 +               || au_test_fuse(sb)
13472 +               || au_test_ubifs(sb)
13473 +               /* || au_test_cifs(sb) */       /* untested */
13474 +               ;
13475 +}
13476 +
13477 +/*
13478 + * filesystems which requires replacing i_mapping.
13479 + */
13480 +static inline int au_test_fs_bad_mapping(struct super_block *sb)
13481 +{
13482 +       return au_test_fuse(sb)
13483 +               || au_test_ubifs(sb);
13484 +}
13485 +
13486 +/* temporary support for i#1 in cramfs */
13487 +static inline int au_test_fs_unique_ino(struct inode *inode)
13488 +{
13489 +       if (au_test_cramfs(inode->i_sb))
13490 +               return inode->i_ino != 1;
13491 +       return 1;
13492 +}
13493 +
13494 +/* ---------------------------------------------------------------------- */
13495 +
13496 +/*
13497 + * the filesystem where the xino files placed must support i/o after unlink and
13498 + * maintain i_size and i_blocks.
13499 + */
13500 +static inline int au_test_fs_bad_xino(struct super_block *sb)
13501 +{
13502 +       return au_test_fs_remote(sb)
13503 +               || au_test_fs_bad_iattr_size(sb)
13504 +               /* don't want unnecessary work for xino */
13505 +               || au_test_aufs(sb)
13506 +               || au_test_ecryptfs(sb)
13507 +               || au_test_nilfs(sb);
13508 +}
13509 +
13510 +static inline int au_test_fs_trunc_xino(struct super_block *sb)
13511 +{
13512 +       return au_test_tmpfs(sb)
13513 +               || au_test_ramfs(sb);
13514 +}
13515 +
13516 +/*
13517 + * test if the @sb is real-readonly.
13518 + */
13519 +static inline int au_test_fs_rr(struct super_block *sb)
13520 +{
13521 +       return au_test_squashfs(sb)
13522 +               || au_test_iso9660(sb)
13523 +               || au_test_cramfs(sb)
13524 +               || au_test_romfs(sb);
13525 +}
13526 +
13527 +#endif /* __KERNEL__ */
13528 +#endif /* __AUFS_FSTYPE_H__ */
13529 diff -urN /usr/share/empty/fs/aufs/hfsnotify.c linux/fs/aufs/hfsnotify.c
13530 --- /usr/share/empty/fs/aufs/hfsnotify.c        1970-01-01 01:00:00.000000000 +0100
13531 +++ linux/fs/aufs/hfsnotify.c   2014-04-24 22:11:10.851935747 +0200
13532 @@ -0,0 +1,281 @@
13533 +/*
13534 + * Copyright (C) 2005-2014 Junjiro R. Okajima
13535 + *
13536 + * This program, aufs is free software; you can redistribute it and/or modify
13537 + * it under the terms of the GNU General Public License as published by
13538 + * the Free Software Foundation; either version 2 of the License, or
13539 + * (at your option) any later version.
13540 + *
13541 + * This program is distributed in the hope that it will be useful,
13542 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13543 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13544 + * GNU General Public License for more details.
13545 + *
13546 + * You should have received a copy of the GNU General Public License
13547 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13548 + */
13549 +
13550 +/*
13551 + * fsnotify for the lower directories
13552 + */
13553 +
13554 +#include "aufs.h"
13555 +
13556 +/* FS_IN_IGNORED is unnecessary */
13557 +static const __u32 AuHfsnMask = (FS_MOVED_TO | FS_MOVED_FROM | FS_DELETE
13558 +                                | FS_CREATE | FS_EVENT_ON_CHILD);
13559 +static DECLARE_WAIT_QUEUE_HEAD(au_hfsn_wq);
13560 +static __cacheline_aligned_in_smp atomic64_t au_hfsn_ifree = ATOMIC64_INIT(0);
13561 +
13562 +static void au_hfsn_free_mark(struct fsnotify_mark *mark)
13563 +{
13564 +       struct au_hnotify *hn = container_of(mark, struct au_hnotify,
13565 +                                            hn_mark);
13566 +       AuDbg("here\n");
13567 +       au_cache_free_hnotify(hn);
13568 +       smp_mb__before_atomic_dec();
13569 +       if (atomic64_dec_and_test(&au_hfsn_ifree))
13570 +               wake_up(&au_hfsn_wq);
13571 +}
13572 +
13573 +static int au_hfsn_alloc(struct au_hinode *hinode)
13574 +{
13575 +       int err;
13576 +       struct au_hnotify *hn;
13577 +       struct super_block *sb;
13578 +       struct au_branch *br;
13579 +       struct fsnotify_mark *mark;
13580 +       aufs_bindex_t bindex;
13581 +
13582 +       hn = hinode->hi_notify;
13583 +       sb = hn->hn_aufs_inode->i_sb;
13584 +       bindex = au_br_index(sb, hinode->hi_id);
13585 +       br = au_sbr(sb, bindex);
13586 +       AuDebugOn(!br->br_hfsn);
13587 +
13588 +       mark = &hn->hn_mark;
13589 +       fsnotify_init_mark(mark, au_hfsn_free_mark);
13590 +       mark->mask = AuHfsnMask;
13591 +       /*
13592 +        * by udba rename or rmdir, aufs assign a new inode to the known
13593 +        * h_inode, so specify 1 to allow dups.
13594 +        */
13595 +       err = fsnotify_add_mark(mark, br->br_hfsn->hfsn_group, hinode->hi_inode,
13596 +                                /*mnt*/NULL, /*allow_dups*/1);
13597 +       /* even if err */
13598 +       fsnotify_put_mark(mark);
13599 +
13600 +       return err;
13601 +}
13602 +
13603 +static int au_hfsn_free(struct au_hinode *hinode, struct au_hnotify *hn)
13604 +{
13605 +       struct fsnotify_mark *mark;
13606 +       unsigned long long ull;
13607 +       struct fsnotify_group *group;
13608 +
13609 +       ull = atomic64_inc_return(&au_hfsn_ifree);
13610 +       BUG_ON(!ull);
13611 +
13612 +       mark = &hn->hn_mark;
13613 +       spin_lock(&mark->lock);
13614 +       group = mark->group;
13615 +       fsnotify_get_group(group);
13616 +       spin_unlock(&mark->lock);
13617 +       fsnotify_destroy_mark(mark, group);
13618 +       fsnotify_put_group(group);
13619 +
13620 +       /* free hn by myself */
13621 +       return 0;
13622 +}
13623 +
13624 +/* ---------------------------------------------------------------------- */
13625 +
13626 +static void au_hfsn_ctl(struct au_hinode *hinode, int do_set)
13627 +{
13628 +       struct fsnotify_mark *mark;
13629 +
13630 +       mark = &hinode->hi_notify->hn_mark;
13631 +       spin_lock(&mark->lock);
13632 +       if (do_set) {
13633 +               AuDebugOn(mark->mask & AuHfsnMask);
13634 +               mark->mask |= AuHfsnMask;
13635 +       } else {
13636 +               AuDebugOn(!(mark->mask & AuHfsnMask));
13637 +               mark->mask &= ~AuHfsnMask;
13638 +       }
13639 +       spin_unlock(&mark->lock);
13640 +       /* fsnotify_recalc_inode_mask(hinode->hi_inode); */
13641 +}
13642 +
13643 +/* ---------------------------------------------------------------------- */
13644 +
13645 +/* #define AuDbgHnotify */
13646 +#ifdef AuDbgHnotify
13647 +static char *au_hfsn_name(u32 mask)
13648 +{
13649 +#ifdef CONFIG_AUFS_DEBUG
13650 +#define test_ret(flag)                         \
13651 +       do {                                    \
13652 +               if (mask & flag)                \
13653 +                       return #flag;           \
13654 +       } while (0)
13655 +       test_ret(FS_ACCESS);
13656 +       test_ret(FS_MODIFY);
13657 +       test_ret(FS_ATTRIB);
13658 +       test_ret(FS_CLOSE_WRITE);
13659 +       test_ret(FS_CLOSE_NOWRITE);
13660 +       test_ret(FS_OPEN);
13661 +       test_ret(FS_MOVED_FROM);
13662 +       test_ret(FS_MOVED_TO);
13663 +       test_ret(FS_CREATE);
13664 +       test_ret(FS_DELETE);
13665 +       test_ret(FS_DELETE_SELF);
13666 +       test_ret(FS_MOVE_SELF);
13667 +       test_ret(FS_UNMOUNT);
13668 +       test_ret(FS_Q_OVERFLOW);
13669 +       test_ret(FS_IN_IGNORED);
13670 +       test_ret(FS_IN_ISDIR);
13671 +       test_ret(FS_IN_ONESHOT);
13672 +       test_ret(FS_EVENT_ON_CHILD);
13673 +       return "";
13674 +#undef test_ret
13675 +#else
13676 +       return "??";
13677 +#endif
13678 +}
13679 +#endif
13680 +
13681 +/* ---------------------------------------------------------------------- */
13682 +
13683 +static void au_hfsn_free_group(struct fsnotify_group *group)
13684 +{
13685 +       struct au_br_hfsnotify *hfsn = group->private;
13686 +
13687 +       AuDbg("here\n");
13688 +       kfree(hfsn);
13689 +}
13690 +
13691 +static int au_hfsn_handle_event(struct fsnotify_group *group,
13692 +                               struct inode *inode,
13693 +                               struct fsnotify_mark *inode_mark,
13694 +                               struct fsnotify_mark *vfsmount_mark,
13695 +                               u32 mask, void *data, int data_type,
13696 +                               const unsigned char *file_name, u32 cookie)
13697 +{
13698 +       int err;
13699 +       struct au_hnotify *hnotify;
13700 +       struct inode *h_dir, *h_inode;
13701 +       struct qstr h_child_qstr = QSTR_INIT(file_name, strlen(file_name));
13702 +
13703 +       AuDebugOn(data_type != FSNOTIFY_EVENT_INODE);
13704 +
13705 +       err = 0;
13706 +       /* if FS_UNMOUNT happens, there must be another bug */
13707 +       AuDebugOn(mask & FS_UNMOUNT);
13708 +       if (mask & (FS_IN_IGNORED | FS_UNMOUNT))
13709 +               goto out;
13710 +
13711 +       h_dir = inode;
13712 +       h_inode = NULL;
13713 +#ifdef AuDbgHnotify
13714 +       au_debug_on();
13715 +       if (1 || h_child_qstr.len != sizeof(AUFS_XINO_FNAME) - 1
13716 +           || strncmp(h_child_qstr.name, AUFS_XINO_FNAME, h_child_qstr.len)) {
13717 +               AuDbg("i%lu, mask 0x%x %s, hcname %.*s, hi%lu\n",
13718 +                     h_dir->i_ino, mask, au_hfsn_name(mask),
13719 +                     AuLNPair(&h_child_qstr), h_inode ? h_inode->i_ino : 0);
13720 +               /* WARN_ON(1); */
13721 +       }
13722 +       au_debug_off();
13723 +#endif
13724 +
13725 +       AuDebugOn(!inode_mark);
13726 +       hnotify = container_of(inode_mark, struct au_hnotify, hn_mark);
13727 +       err = au_hnotify(h_dir, hnotify, mask, &h_child_qstr, h_inode);
13728 +
13729 +out:
13730 +       return err;
13731 +}
13732 +
13733 +static struct fsnotify_ops au_hfsn_ops = {
13734 +       .handle_event           = au_hfsn_handle_event,
13735 +       .free_group_priv        = au_hfsn_free_group
13736 +};
13737 +
13738 +/* ---------------------------------------------------------------------- */
13739 +
13740 +static void au_hfsn_fin_br(struct au_branch *br)
13741 +{
13742 +       struct au_br_hfsnotify *hfsn;
13743 +
13744 +       hfsn = br->br_hfsn;
13745 +       if (hfsn)
13746 +               fsnotify_put_group(hfsn->hfsn_group);
13747 +}
13748 +
13749 +static int au_hfsn_init_br(struct au_branch *br, int perm)
13750 +{
13751 +       int err;
13752 +       struct fsnotify_group *group;
13753 +       struct au_br_hfsnotify *hfsn;
13754 +
13755 +       err = 0;
13756 +       br->br_hfsn = NULL;
13757 +       if (!au_br_hnotifyable(perm))
13758 +               goto out;
13759 +
13760 +       err = -ENOMEM;
13761 +       hfsn = kmalloc(sizeof(*hfsn), GFP_NOFS);
13762 +       if (unlikely(!hfsn))
13763 +               goto out;
13764 +
13765 +       err = 0;
13766 +       group = fsnotify_alloc_group(&au_hfsn_ops);
13767 +       if (IS_ERR(group)) {
13768 +               err = PTR_ERR(group);
13769 +               pr_err("fsnotify_alloc_group() failed, %d\n", err);
13770 +               goto out_hfsn;
13771 +       }
13772 +
13773 +       group->private = hfsn;
13774 +       hfsn->hfsn_group = group;
13775 +       br->br_hfsn = hfsn;
13776 +       goto out; /* success */
13777 +
13778 +out_hfsn:
13779 +       kfree(hfsn);
13780 +out:
13781 +       return err;
13782 +}
13783 +
13784 +static int au_hfsn_reset_br(unsigned int udba, struct au_branch *br, int perm)
13785 +{
13786 +       int err;
13787 +
13788 +       err = 0;
13789 +       if (!br->br_hfsn)
13790 +               err = au_hfsn_init_br(br, perm);
13791 +
13792 +       return err;
13793 +}
13794 +
13795 +/* ---------------------------------------------------------------------- */
13796 +
13797 +static void au_hfsn_fin(void)
13798 +{
13799 +       AuDbg("au_hfsn_ifree %lld\n", (long long)atomic64_read(&au_hfsn_ifree));
13800 +       wait_event(au_hfsn_wq, !atomic64_read(&au_hfsn_ifree));
13801 +}
13802 +
13803 +const struct au_hnotify_op au_hnotify_op = {
13804 +       .ctl            = au_hfsn_ctl,
13805 +       .alloc          = au_hfsn_alloc,
13806 +       .free           = au_hfsn_free,
13807 +
13808 +       .fin            = au_hfsn_fin,
13809 +
13810 +       .reset_br       = au_hfsn_reset_br,
13811 +       .fin_br         = au_hfsn_fin_br,
13812 +       .init_br        = au_hfsn_init_br
13813 +};
13814 diff -urN /usr/share/empty/fs/aufs/hfsplus.c linux/fs/aufs/hfsplus.c
13815 --- /usr/share/empty/fs/aufs/hfsplus.c  1970-01-01 01:00:00.000000000 +0100
13816 +++ linux/fs/aufs/hfsplus.c     2014-04-24 22:11:10.851935747 +0200
13817 @@ -0,0 +1,56 @@
13818 +/*
13819 + * Copyright (C) 2010-2014 Junjiro R. Okajima
13820 + *
13821 + * This program, aufs is free software; you can redistribute it and/or modify
13822 + * it under the terms of the GNU General Public License as published by
13823 + * the Free Software Foundation; either version 2 of the License, or
13824 + * (at your option) any later version.
13825 + *
13826 + * This program is distributed in the hope that it will be useful,
13827 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13828 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13829 + * GNU General Public License for more details.
13830 + *
13831 + * You should have received a copy of the GNU General Public License
13832 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13833 + */
13834 +
13835 +/*
13836 + * special support for filesystems which aqucires an inode mutex
13837 + * at final closing a file, eg, hfsplus.
13838 + *
13839 + * This trick is very simple and stupid, just to open the file before really
13840 + * neceeary open to tell hfsplus that this is not the final closing.
13841 + * The caller should call au_h_open_pre() after acquiring the inode mutex,
13842 + * and au_h_open_post() after releasing it.
13843 + */
13844 +
13845 +#include "aufs.h"
13846 +
13847 +struct file *au_h_open_pre(struct dentry *dentry, aufs_bindex_t bindex,
13848 +                          int force_wr)
13849 +{
13850 +       struct file *h_file;
13851 +       struct dentry *h_dentry;
13852 +
13853 +       h_dentry = au_h_dptr(dentry, bindex);
13854 +       AuDebugOn(!h_dentry);
13855 +       AuDebugOn(!h_dentry->d_inode);
13856 +
13857 +       h_file = NULL;
13858 +       if (au_test_hfsplus(h_dentry->d_sb)
13859 +           && S_ISREG(h_dentry->d_inode->i_mode))
13860 +               h_file = au_h_open(dentry, bindex,
13861 +                                  O_RDONLY | O_NOATIME | O_LARGEFILE,
13862 +                                  /*file*/NULL, force_wr);
13863 +       return h_file;
13864 +}
13865 +
13866 +void au_h_open_post(struct dentry *dentry, aufs_bindex_t bindex,
13867 +                   struct file *h_file)
13868 +{
13869 +       if (h_file) {
13870 +               fput(h_file);
13871 +               au_sbr_put(dentry->d_sb, bindex);
13872 +       }
13873 +}
13874 diff -urN /usr/share/empty/fs/aufs/hnotify.c linux/fs/aufs/hnotify.c
13875 --- /usr/share/empty/fs/aufs/hnotify.c  1970-01-01 01:00:00.000000000 +0100
13876 +++ linux/fs/aufs/hnotify.c     2014-04-24 22:11:10.851935747 +0200
13877 @@ -0,0 +1,710 @@
13878 +/*
13879 + * Copyright (C) 2005-2014 Junjiro R. Okajima
13880 + *
13881 + * This program, aufs is free software; you can redistribute it and/or modify
13882 + * it under the terms of the GNU General Public License as published by
13883 + * the Free Software Foundation; either version 2 of the License, or
13884 + * (at your option) any later version.
13885 + *
13886 + * This program is distributed in the hope that it will be useful,
13887 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
13888 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13889 + * GNU General Public License for more details.
13890 + *
13891 + * You should have received a copy of the GNU General Public License
13892 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
13893 + */
13894 +
13895 +/*
13896 + * abstraction to notify the direct changes on lower directories
13897 + */
13898 +
13899 +#include "aufs.h"
13900 +
13901 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode)
13902 +{
13903 +       int err;
13904 +       struct au_hnotify *hn;
13905 +
13906 +       err = -ENOMEM;
13907 +       hn = au_cache_alloc_hnotify();
13908 +       if (hn) {
13909 +               hn->hn_aufs_inode = inode;
13910 +               hinode->hi_notify = hn;
13911 +               err = au_hnotify_op.alloc(hinode);
13912 +               AuTraceErr(err);
13913 +               if (unlikely(err)) {
13914 +                       hinode->hi_notify = NULL;
13915 +                       au_cache_free_hnotify(hn);
13916 +                       /*
13917 +                        * The upper dir was removed by udba, but the same named
13918 +                        * dir left. In this case, aufs assignes a new inode
13919 +                        * number and set the monitor again.
13920 +                        * For the lower dir, the old monitnor is still left.
13921 +                        */
13922 +                       if (err == -EEXIST)
13923 +                               err = 0;
13924 +               }
13925 +       }
13926 +
13927 +       AuTraceErr(err);
13928 +       return err;
13929 +}
13930 +
13931 +void au_hn_free(struct au_hinode *hinode)
13932 +{
13933 +       struct au_hnotify *hn;
13934 +
13935 +       hn = hinode->hi_notify;
13936 +       if (hn) {
13937 +               hinode->hi_notify = NULL;
13938 +               if (au_hnotify_op.free(hinode, hn))
13939 +                       au_cache_free_hnotify(hn);
13940 +       }
13941 +}
13942 +
13943 +/* ---------------------------------------------------------------------- */
13944 +
13945 +void au_hn_ctl(struct au_hinode *hinode, int do_set)
13946 +{
13947 +       if (hinode->hi_notify)
13948 +               au_hnotify_op.ctl(hinode, do_set);
13949 +}
13950 +
13951 +void au_hn_reset(struct inode *inode, unsigned int flags)
13952 +{
13953 +       aufs_bindex_t bindex, bend;
13954 +       struct inode *hi;
13955 +       struct dentry *iwhdentry;
13956 +
13957 +       bend = au_ibend(inode);
13958 +       for (bindex = au_ibstart(inode); bindex <= bend; bindex++) {
13959 +               hi = au_h_iptr(inode, bindex);
13960 +               if (!hi)
13961 +                       continue;
13962 +
13963 +               /* mutex_lock_nested(&hi->i_mutex, AuLsc_I_CHILD); */
13964 +               iwhdentry = au_hi_wh(inode, bindex);
13965 +               if (iwhdentry)
13966 +                       dget(iwhdentry);
13967 +               au_igrab(hi);
13968 +               au_set_h_iptr(inode, bindex, NULL, 0);
13969 +               au_set_h_iptr(inode, bindex, au_igrab(hi),
13970 +                             flags & ~AuHi_XINO);
13971 +               iput(hi);
13972 +               dput(iwhdentry);
13973 +               /* mutex_unlock(&hi->i_mutex); */
13974 +       }
13975 +}
13976 +
13977 +/* ---------------------------------------------------------------------- */
13978 +
13979 +static int hn_xino(struct inode *inode, struct inode *h_inode)
13980 +{
13981 +       int err;
13982 +       aufs_bindex_t bindex, bend, bfound, bstart;
13983 +       struct inode *h_i;
13984 +
13985 +       err = 0;
13986 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
13987 +               pr_warn("branch root dir was changed\n");
13988 +               goto out;
13989 +       }
13990 +
13991 +       bfound = -1;
13992 +       bend = au_ibend(inode);
13993 +       bstart = au_ibstart(inode);
13994 +#if 0 /* reserved for future use */
13995 +       if (bindex == bend) {
13996 +               /* keep this ino in rename case */
13997 +               goto out;
13998 +       }
13999 +#endif
14000 +       for (bindex = bstart; bindex <= bend; bindex++)
14001 +               if (au_h_iptr(inode, bindex) == h_inode) {
14002 +                       bfound = bindex;
14003 +                       break;
14004 +               }
14005 +       if (bfound < 0)
14006 +               goto out;
14007 +
14008 +       for (bindex = bstart; bindex <= bend; bindex++) {
14009 +               h_i = au_h_iptr(inode, bindex);
14010 +               if (!h_i)
14011 +                       continue;
14012 +
14013 +               err = au_xino_write(inode->i_sb, bindex, h_i->i_ino, /*ino*/0);
14014 +               /* ignore this error */
14015 +               /* bad action? */
14016 +       }
14017 +
14018 +       /* children inode number will be broken */
14019 +
14020 +out:
14021 +       AuTraceErr(err);
14022 +       return err;
14023 +}
14024 +
14025 +static int hn_gen_tree(struct dentry *dentry)
14026 +{
14027 +       int err, i, j, ndentry;
14028 +       struct au_dcsub_pages dpages;
14029 +       struct au_dpage *dpage;
14030 +       struct dentry **dentries;
14031 +
14032 +       err = au_dpages_init(&dpages, GFP_NOFS);
14033 +       if (unlikely(err))
14034 +               goto out;
14035 +       err = au_dcsub_pages(&dpages, dentry, NULL, NULL);
14036 +       if (unlikely(err))
14037 +               goto out_dpages;
14038 +
14039 +       for (i = 0; i < dpages.ndpage; i++) {
14040 +               dpage = dpages.dpages + i;
14041 +               dentries = dpage->dentries;
14042 +               ndentry = dpage->ndentry;
14043 +               for (j = 0; j < ndentry; j++) {
14044 +                       struct dentry *d;
14045 +
14046 +                       d = dentries[j];
14047 +                       if (IS_ROOT(d))
14048 +                               continue;
14049 +
14050 +                       au_digen_dec(d);
14051 +                       if (d->d_inode)
14052 +                               /* todo: reset children xino?
14053 +                                  cached children only? */
14054 +                               au_iigen_dec(d->d_inode);
14055 +               }
14056 +       }
14057 +
14058 +out_dpages:
14059 +       au_dpages_free(&dpages);
14060 +
14061 +#if 0
14062 +       /* discard children */
14063 +       dentry_unhash(dentry);
14064 +       dput(dentry);
14065 +#endif
14066 +out:
14067 +       return err;
14068 +}
14069 +
14070 +/*
14071 + * return 0 if processed.
14072 + */
14073 +static int hn_gen_by_inode(char *name, unsigned int nlen, struct inode *inode,
14074 +                          const unsigned int isdir)
14075 +{
14076 +       int err;
14077 +       struct dentry *d;
14078 +       struct qstr *dname;
14079 +
14080 +       err = 1;
14081 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
14082 +               pr_warn("branch root dir was changed\n");
14083 +               err = 0;
14084 +               goto out;
14085 +       }
14086 +
14087 +       if (!isdir) {
14088 +               AuDebugOn(!name);
14089 +               au_iigen_dec(inode);
14090 +               spin_lock(&inode->i_lock);
14091 +               hlist_for_each_entry(d, &inode->i_dentry, d_u.d_alias) {
14092 +                       spin_lock(&d->d_lock);
14093 +                       dname = &d->d_name;
14094 +                       if (dname->len != nlen
14095 +                           && memcmp(dname->name, name, nlen)) {
14096 +                               spin_unlock(&d->d_lock);
14097 +                               continue;
14098 +                       }
14099 +                       err = 0;
14100 +                       au_digen_dec(d);
14101 +                       spin_unlock(&d->d_lock);
14102 +                       break;
14103 +               }
14104 +               spin_unlock(&inode->i_lock);
14105 +       } else {
14106 +               au_fset_si(au_sbi(inode->i_sb), FAILED_REFRESH_DIR);
14107 +               d = d_find_alias(inode);
14108 +               if (!d) {
14109 +                       au_iigen_dec(inode);
14110 +                       goto out;
14111 +               }
14112 +
14113 +               spin_lock(&d->d_lock);
14114 +               dname = &d->d_name;
14115 +               if (dname->len == nlen && !memcmp(dname->name, name, nlen)) {
14116 +                       spin_unlock(&d->d_lock);
14117 +                       err = hn_gen_tree(d);
14118 +                       spin_lock(&d->d_lock);
14119 +               }
14120 +               spin_unlock(&d->d_lock);
14121 +               dput(d);
14122 +       }
14123 +
14124 +out:
14125 +       AuTraceErr(err);
14126 +       return err;
14127 +}
14128 +
14129 +static int hn_gen_by_name(struct dentry *dentry, const unsigned int isdir)
14130 +{
14131 +       int err;
14132 +       struct inode *inode;
14133 +
14134 +       inode = dentry->d_inode;
14135 +       if (IS_ROOT(dentry)
14136 +           /* || (inode && inode->i_ino == AUFS_ROOT_INO) */
14137 +               ) {
14138 +               pr_warn("branch root dir was changed\n");
14139 +               return 0;
14140 +       }
14141 +
14142 +       err = 0;
14143 +       if (!isdir) {
14144 +               au_digen_dec(dentry);
14145 +               if (inode)
14146 +                       au_iigen_dec(inode);
14147 +       } else {
14148 +               au_fset_si(au_sbi(dentry->d_sb), FAILED_REFRESH_DIR);
14149 +               if (inode)
14150 +                       err = hn_gen_tree(dentry);
14151 +       }
14152 +
14153 +       AuTraceErr(err);
14154 +       return err;
14155 +}
14156 +
14157 +/* ---------------------------------------------------------------------- */
14158 +
14159 +/* hnotify job flags */
14160 +#define AuHnJob_XINO0          1
14161 +#define AuHnJob_GEN            (1 << 1)
14162 +#define AuHnJob_DIRENT         (1 << 2)
14163 +#define AuHnJob_ISDIR          (1 << 3)
14164 +#define AuHnJob_TRYXINO0       (1 << 4)
14165 +#define AuHnJob_MNTPNT         (1 << 5)
14166 +#define au_ftest_hnjob(flags, name)    ((flags) & AuHnJob_##name)
14167 +#define au_fset_hnjob(flags, name) \
14168 +       do { (flags) |= AuHnJob_##name; } while (0)
14169 +#define au_fclr_hnjob(flags, name) \
14170 +       do { (flags) &= ~AuHnJob_##name; } while (0)
14171 +
14172 +enum {
14173 +       AuHn_CHILD,
14174 +       AuHn_PARENT,
14175 +       AuHnLast
14176 +};
14177 +
14178 +struct au_hnotify_args {
14179 +       struct inode *h_dir, *dir, *h_child_inode;
14180 +       u32 mask;
14181 +       unsigned int flags[AuHnLast];
14182 +       unsigned int h_child_nlen;
14183 +       char h_child_name[];
14184 +};
14185 +
14186 +struct hn_job_args {
14187 +       unsigned int flags;
14188 +       struct inode *inode, *h_inode, *dir, *h_dir;
14189 +       struct dentry *dentry;
14190 +       char *h_name;
14191 +       int h_nlen;
14192 +};
14193 +
14194 +static int hn_job(struct hn_job_args *a)
14195 +{
14196 +       const unsigned int isdir = au_ftest_hnjob(a->flags, ISDIR);
14197 +
14198 +       /* reset xino */
14199 +       if (au_ftest_hnjob(a->flags, XINO0) && a->inode)
14200 +               hn_xino(a->inode, a->h_inode); /* ignore this error */
14201 +
14202 +       if (au_ftest_hnjob(a->flags, TRYXINO0)
14203 +           && a->inode
14204 +           && a->h_inode) {
14205 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
14206 +               if (!a->h_inode->i_nlink)
14207 +                       hn_xino(a->inode, a->h_inode); /* ignore this error */
14208 +               mutex_unlock(&a->h_inode->i_mutex);
14209 +       }
14210 +
14211 +       /* make the generation obsolete */
14212 +       if (au_ftest_hnjob(a->flags, GEN)) {
14213 +               int err = -1;
14214 +               if (a->inode)
14215 +                       err = hn_gen_by_inode(a->h_name, a->h_nlen, a->inode,
14216 +                                             isdir);
14217 +               if (err && a->dentry)
14218 +                       hn_gen_by_name(a->dentry, isdir);
14219 +               /* ignore this error */
14220 +       }
14221 +
14222 +       /* make dir entries obsolete */
14223 +       if (au_ftest_hnjob(a->flags, DIRENT) && a->inode) {
14224 +               struct au_vdir *vdir;
14225 +
14226 +               vdir = au_ivdir(a->inode);
14227 +               if (vdir)
14228 +                       vdir->vd_jiffy = 0;
14229 +               /* IMustLock(a->inode); */
14230 +               /* a->inode->i_version++; */
14231 +       }
14232 +
14233 +       /* can do nothing but warn */
14234 +       if (au_ftest_hnjob(a->flags, MNTPNT)
14235 +           && a->dentry
14236 +           && d_mountpoint(a->dentry))
14237 +               pr_warn("mount-point %pd is removed or renamed\n", a->dentry);
14238 +
14239 +       return 0;
14240 +}
14241 +
14242 +/* ---------------------------------------------------------------------- */
14243 +
14244 +static struct dentry *lookup_wlock_by_name(char *name, unsigned int nlen,
14245 +                                          struct inode *dir)
14246 +{
14247 +       struct dentry *dentry, *d, *parent;
14248 +       struct qstr *dname;
14249 +
14250 +       parent = d_find_alias(dir);
14251 +       if (!parent)
14252 +               return NULL;
14253 +
14254 +       dentry = NULL;
14255 +       spin_lock(&parent->d_lock);
14256 +       list_for_each_entry(d, &parent->d_subdirs, d_child) {
14257 +               /* AuDbg("%pd\n", d); */
14258 +               spin_lock_nested(&d->d_lock, DENTRY_D_LOCK_NESTED);
14259 +               dname = &d->d_name;
14260 +               if (dname->len != nlen || memcmp(dname->name, name, nlen))
14261 +                       goto cont_unlock;
14262 +               if (au_di(d))
14263 +                       au_digen_dec(d);
14264 +               else
14265 +                       goto cont_unlock;
14266 +               if (d_count(d)) {
14267 +                       dentry = dget_dlock(d);
14268 +                       spin_unlock(&d->d_lock);
14269 +                       break;
14270 +               }
14271 +
14272 +cont_unlock:
14273 +               spin_unlock(&d->d_lock);
14274 +       }
14275 +       spin_unlock(&parent->d_lock);
14276 +       dput(parent);
14277 +
14278 +       if (dentry)
14279 +               di_write_lock_child(dentry);
14280 +
14281 +       return dentry;
14282 +}
14283 +
14284 +static struct inode *lookup_wlock_by_ino(struct super_block *sb,
14285 +                                        aufs_bindex_t bindex, ino_t h_ino)
14286 +{
14287 +       struct inode *inode;
14288 +       ino_t ino;
14289 +       int err;
14290 +
14291 +       inode = NULL;
14292 +       err = au_xino_read(sb, bindex, h_ino, &ino);
14293 +       if (!err && ino)
14294 +               inode = ilookup(sb, ino);
14295 +       if (!inode)
14296 +               goto out;
14297 +
14298 +       if (unlikely(inode->i_ino == AUFS_ROOT_INO)) {
14299 +               pr_warn("wrong root branch\n");
14300 +               iput(inode);
14301 +               inode = NULL;
14302 +               goto out;
14303 +       }
14304 +
14305 +       ii_write_lock_child(inode);
14306 +
14307 +out:
14308 +       return inode;
14309 +}
14310 +
14311 +static void au_hn_bh(void *_args)
14312 +{
14313 +       struct au_hnotify_args *a = _args;
14314 +       struct super_block *sb;
14315 +       aufs_bindex_t bindex, bend, bfound;
14316 +       unsigned char xino, try_iput;
14317 +       int err;
14318 +       struct inode *inode;
14319 +       ino_t h_ino;
14320 +       struct hn_job_args args;
14321 +       struct dentry *dentry;
14322 +       struct au_sbinfo *sbinfo;
14323 +
14324 +       AuDebugOn(!_args);
14325 +       AuDebugOn(!a->h_dir);
14326 +       AuDebugOn(!a->dir);
14327 +       AuDebugOn(!a->mask);
14328 +       AuDbg("mask 0x%x, i%lu, hi%lu, hci%lu\n",
14329 +             a->mask, a->dir->i_ino, a->h_dir->i_ino,
14330 +             a->h_child_inode ? a->h_child_inode->i_ino : 0);
14331 +
14332 +       inode = NULL;
14333 +       dentry = NULL;
14334 +       /*
14335 +        * do not lock a->dir->i_mutex here
14336 +        * because of d_revalidate() may cause a deadlock.
14337 +        */
14338 +       sb = a->dir->i_sb;
14339 +       AuDebugOn(!sb);
14340 +       sbinfo = au_sbi(sb);
14341 +       AuDebugOn(!sbinfo);
14342 +       si_write_lock(sb, AuLock_NOPLMW);
14343 +
14344 +       ii_read_lock_parent(a->dir);
14345 +       bfound = -1;
14346 +       bend = au_ibend(a->dir);
14347 +       for (bindex = au_ibstart(a->dir); bindex <= bend; bindex++)
14348 +               if (au_h_iptr(a->dir, bindex) == a->h_dir) {
14349 +                       bfound = bindex;
14350 +                       break;
14351 +               }
14352 +       ii_read_unlock(a->dir);
14353 +       if (unlikely(bfound < 0))
14354 +               goto out;
14355 +
14356 +       xino = !!au_opt_test(au_mntflags(sb), XINO);
14357 +       h_ino = 0;
14358 +       if (a->h_child_inode)
14359 +               h_ino = a->h_child_inode->i_ino;
14360 +
14361 +       if (a->h_child_nlen
14362 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], GEN)
14363 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], MNTPNT)))
14364 +               dentry = lookup_wlock_by_name(a->h_child_name, a->h_child_nlen,
14365 +                                             a->dir);
14366 +       try_iput = 0;
14367 +       if (dentry)
14368 +               inode = dentry->d_inode;
14369 +       if (xino && !inode && h_ino
14370 +           && (au_ftest_hnjob(a->flags[AuHn_CHILD], XINO0)
14371 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], TRYXINO0)
14372 +               || au_ftest_hnjob(a->flags[AuHn_CHILD], GEN))) {
14373 +               inode = lookup_wlock_by_ino(sb, bfound, h_ino);
14374 +               try_iput = 1;
14375 +           }
14376 +
14377 +       args.flags = a->flags[AuHn_CHILD];
14378 +       args.dentry = dentry;
14379 +       args.inode = inode;
14380 +       args.h_inode = a->h_child_inode;
14381 +       args.dir = a->dir;
14382 +       args.h_dir = a->h_dir;
14383 +       args.h_name = a->h_child_name;
14384 +       args.h_nlen = a->h_child_nlen;
14385 +       err = hn_job(&args);
14386 +       if (dentry) {
14387 +               if (au_di(dentry))
14388 +                       di_write_unlock(dentry);
14389 +               dput(dentry);
14390 +       }
14391 +       if (inode && try_iput) {
14392 +               ii_write_unlock(inode);
14393 +               iput(inode);
14394 +       }
14395 +
14396 +       ii_write_lock_parent(a->dir);
14397 +       args.flags = a->flags[AuHn_PARENT];
14398 +       args.dentry = NULL;
14399 +       args.inode = a->dir;
14400 +       args.h_inode = a->h_dir;
14401 +       args.dir = NULL;
14402 +       args.h_dir = NULL;
14403 +       args.h_name = NULL;
14404 +       args.h_nlen = 0;
14405 +       err = hn_job(&args);
14406 +       ii_write_unlock(a->dir);
14407 +
14408 +out:
14409 +       iput(a->h_child_inode);
14410 +       iput(a->h_dir);
14411 +       iput(a->dir);
14412 +       si_write_unlock(sb);
14413 +       au_nwt_done(&sbinfo->si_nowait);
14414 +       kfree(a);
14415 +}
14416 +
14417 +/* ---------------------------------------------------------------------- */
14418 +
14419 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
14420 +              struct qstr *h_child_qstr, struct inode *h_child_inode)
14421 +{
14422 +       int err, len;
14423 +       unsigned int flags[AuHnLast], f;
14424 +       unsigned char isdir, isroot, wh;
14425 +       struct inode *dir;
14426 +       struct au_hnotify_args *args;
14427 +       char *p, *h_child_name;
14428 +
14429 +       err = 0;
14430 +       AuDebugOn(!hnotify || !hnotify->hn_aufs_inode);
14431 +       dir = igrab(hnotify->hn_aufs_inode);
14432 +       if (!dir)
14433 +               goto out;
14434 +
14435 +       isroot = (dir->i_ino == AUFS_ROOT_INO);
14436 +       wh = 0;
14437 +       h_child_name = (void *)h_child_qstr->name;
14438 +       len = h_child_qstr->len;
14439 +       if (h_child_name) {
14440 +               if (len > AUFS_WH_PFX_LEN
14441 +                   && !memcmp(h_child_name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
14442 +                       h_child_name += AUFS_WH_PFX_LEN;
14443 +                       len -= AUFS_WH_PFX_LEN;
14444 +                       wh = 1;
14445 +               }
14446 +       }
14447 +
14448 +       isdir = 0;
14449 +       if (h_child_inode)
14450 +               isdir = !!S_ISDIR(h_child_inode->i_mode);
14451 +       flags[AuHn_PARENT] = AuHnJob_ISDIR;
14452 +       flags[AuHn_CHILD] = 0;
14453 +       if (isdir)
14454 +               flags[AuHn_CHILD] = AuHnJob_ISDIR;
14455 +       au_fset_hnjob(flags[AuHn_PARENT], DIRENT);
14456 +       au_fset_hnjob(flags[AuHn_CHILD], GEN);
14457 +       switch (mask & FS_EVENTS_POSS_ON_CHILD) {
14458 +       case FS_MOVED_FROM:
14459 +       case FS_MOVED_TO:
14460 +               au_fset_hnjob(flags[AuHn_CHILD], XINO0);
14461 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
14462 +               /*FALLTHROUGH*/
14463 +       case FS_CREATE:
14464 +               AuDebugOn(!h_child_name);
14465 +               break;
14466 +
14467 +       case FS_DELETE:
14468 +               /*
14469 +                * aufs never be able to get this child inode.
14470 +                * revalidation should be in d_revalidate()
14471 +                * by checking i_nlink, i_generation or d_unhashed().
14472 +                */
14473 +               AuDebugOn(!h_child_name);
14474 +               au_fset_hnjob(flags[AuHn_CHILD], TRYXINO0);
14475 +               au_fset_hnjob(flags[AuHn_CHILD], MNTPNT);
14476 +               break;
14477 +
14478 +       default:
14479 +               AuDebugOn(1);
14480 +       }
14481 +
14482 +       if (wh)
14483 +               h_child_inode = NULL;
14484 +
14485 +       err = -ENOMEM;
14486 +       /* iput() and kfree() will be called in au_hnotify() */
14487 +       args = kmalloc(sizeof(*args) + len + 1, GFP_NOFS);
14488 +       if (unlikely(!args)) {
14489 +               AuErr1("no memory\n");
14490 +               iput(dir);
14491 +               goto out;
14492 +       }
14493 +       args->flags[AuHn_PARENT] = flags[AuHn_PARENT];
14494 +       args->flags[AuHn_CHILD] = flags[AuHn_CHILD];
14495 +       args->mask = mask;
14496 +       args->dir = dir;
14497 +       args->h_dir = igrab(h_dir);
14498 +       if (h_child_inode)
14499 +               h_child_inode = igrab(h_child_inode); /* can be NULL */
14500 +       args->h_child_inode = h_child_inode;
14501 +       args->h_child_nlen = len;
14502 +       if (len) {
14503 +               p = (void *)args;
14504 +               p += sizeof(*args);
14505 +               memcpy(p, h_child_name, len);
14506 +               p[len] = 0;
14507 +       }
14508 +
14509 +       f = 0;
14510 +       if (!dir->i_nlink)
14511 +               f = AuWkq_NEST;
14512 +       err = au_wkq_nowait(au_hn_bh, args, dir->i_sb, f);
14513 +       if (unlikely(err)) {
14514 +               pr_err("wkq %d\n", err);
14515 +               iput(args->h_child_inode);
14516 +               iput(args->h_dir);
14517 +               iput(args->dir);
14518 +               kfree(args);
14519 +       }
14520 +
14521 +out:
14522 +       return err;
14523 +}
14524 +
14525 +/* ---------------------------------------------------------------------- */
14526 +
14527 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm)
14528 +{
14529 +       int err;
14530 +
14531 +       AuDebugOn(!(udba & AuOptMask_UDBA));
14532 +
14533 +       err = 0;
14534 +       if (au_hnotify_op.reset_br)
14535 +               err = au_hnotify_op.reset_br(udba, br, perm);
14536 +
14537 +       return err;
14538 +}
14539 +
14540 +int au_hnotify_init_br(struct au_branch *br, int perm)
14541 +{
14542 +       int err;
14543 +
14544 +       err = 0;
14545 +       if (au_hnotify_op.init_br)
14546 +               err = au_hnotify_op.init_br(br, perm);
14547 +
14548 +       return err;
14549 +}
14550 +
14551 +void au_hnotify_fin_br(struct au_branch *br)
14552 +{
14553 +       if (au_hnotify_op.fin_br)
14554 +               au_hnotify_op.fin_br(br);
14555 +}
14556 +
14557 +static void au_hn_destroy_cache(void)
14558 +{
14559 +       kmem_cache_destroy(au_cachep[AuCache_HNOTIFY]);
14560 +       au_cachep[AuCache_HNOTIFY] = NULL;
14561 +}
14562 +
14563 +int __init au_hnotify_init(void)
14564 +{
14565 +       int err;
14566 +
14567 +       err = -ENOMEM;
14568 +       au_cachep[AuCache_HNOTIFY] = AuCache(au_hnotify);
14569 +       if (au_cachep[AuCache_HNOTIFY]) {
14570 +               err = 0;
14571 +               if (au_hnotify_op.init)
14572 +                       err = au_hnotify_op.init();
14573 +               if (unlikely(err))
14574 +                       au_hn_destroy_cache();
14575 +       }
14576 +       AuTraceErr(err);
14577 +       return err;
14578 +}
14579 +
14580 +void au_hnotify_fin(void)
14581 +{
14582 +       if (au_hnotify_op.fin)
14583 +               au_hnotify_op.fin();
14584 +       /* cf. au_cache_fin() */
14585 +       if (au_cachep[AuCache_HNOTIFY])
14586 +               au_hn_destroy_cache();
14587 +}
14588 diff -urN /usr/share/empty/fs/aufs/iinfo.c linux/fs/aufs/iinfo.c
14589 --- /usr/share/empty/fs/aufs/iinfo.c    1970-01-01 01:00:00.000000000 +0100
14590 +++ linux/fs/aufs/iinfo.c       2014-04-24 22:11:10.855269116 +0200
14591 @@ -0,0 +1,275 @@
14592 +/*
14593 + * Copyright (C) 2005-2014 Junjiro R. Okajima
14594 + *
14595 + * This program, aufs is free software; you can redistribute it and/or modify
14596 + * it under the terms of the GNU General Public License as published by
14597 + * the Free Software Foundation; either version 2 of the License, or
14598 + * (at your option) any later version.
14599 + *
14600 + * This program is distributed in the hope that it will be useful,
14601 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14602 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14603 + * GNU General Public License for more details.
14604 + *
14605 + * You should have received a copy of the GNU General Public License
14606 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14607 + */
14608 +
14609 +/*
14610 + * inode private data
14611 + */
14612 +
14613 +#include "aufs.h"
14614 +
14615 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex)
14616 +{
14617 +       struct inode *h_inode;
14618 +
14619 +       IiMustAnyLock(inode);
14620 +
14621 +       h_inode = au_ii(inode)->ii_hinode[0 + bindex].hi_inode;
14622 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
14623 +       return h_inode;
14624 +}
14625 +
14626 +/* todo: hard/soft set? */
14627 +void au_hiput(struct au_hinode *hinode)
14628 +{
14629 +       au_hn_free(hinode);
14630 +       dput(hinode->hi_whdentry);
14631 +       iput(hinode->hi_inode);
14632 +}
14633 +
14634 +unsigned int au_hi_flags(struct inode *inode, int isdir)
14635 +{
14636 +       unsigned int flags;
14637 +       const unsigned int mnt_flags = au_mntflags(inode->i_sb);
14638 +
14639 +       flags = 0;
14640 +       if (au_opt_test(mnt_flags, XINO))
14641 +               au_fset_hi(flags, XINO);
14642 +       if (isdir && au_opt_test(mnt_flags, UDBA_HNOTIFY))
14643 +               au_fset_hi(flags, HNOTIFY);
14644 +       return flags;
14645 +}
14646 +
14647 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
14648 +                  struct inode *h_inode, unsigned int flags)
14649 +{
14650 +       struct au_hinode *hinode;
14651 +       struct inode *hi;
14652 +       struct au_iinfo *iinfo = au_ii(inode);
14653 +
14654 +       IiMustWriteLock(inode);
14655 +
14656 +       hinode = iinfo->ii_hinode + bindex;
14657 +       hi = hinode->hi_inode;
14658 +       AuDebugOn(h_inode && atomic_read(&h_inode->i_count) <= 0);
14659 +
14660 +       if (hi)
14661 +               au_hiput(hinode);
14662 +       hinode->hi_inode = h_inode;
14663 +       if (h_inode) {
14664 +               int err;
14665 +               struct super_block *sb = inode->i_sb;
14666 +               struct au_branch *br;
14667 +
14668 +               AuDebugOn(inode->i_mode
14669 +                         && (h_inode->i_mode & S_IFMT)
14670 +                         != (inode->i_mode & S_IFMT));
14671 +               if (bindex == iinfo->ii_bstart)
14672 +                       au_cpup_igen(inode, h_inode);
14673 +               br = au_sbr(sb, bindex);
14674 +               hinode->hi_id = br->br_id;
14675 +               if (au_ftest_hi(flags, XINO)) {
14676 +                       err = au_xino_write(sb, bindex, h_inode->i_ino,
14677 +                                           inode->i_ino);
14678 +                       if (unlikely(err))
14679 +                               AuIOErr1("failed au_xino_write() %d\n", err);
14680 +               }
14681 +
14682 +               if (au_ftest_hi(flags, HNOTIFY)
14683 +                   && au_br_hnotifyable(br->br_perm)) {
14684 +                       err = au_hn_alloc(hinode, inode);
14685 +                       if (unlikely(err))
14686 +                               AuIOErr1("au_hn_alloc() %d\n", err);
14687 +               }
14688 +       }
14689 +}
14690 +
14691 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
14692 +                 struct dentry *h_wh)
14693 +{
14694 +       struct au_hinode *hinode;
14695 +
14696 +       IiMustWriteLock(inode);
14697 +
14698 +       hinode = au_ii(inode)->ii_hinode + bindex;
14699 +       AuDebugOn(hinode->hi_whdentry);
14700 +       hinode->hi_whdentry = h_wh;
14701 +}
14702 +
14703 +void au_update_iigen(struct inode *inode, int half)
14704 +{
14705 +       struct au_iinfo *iinfo;
14706 +       struct au_iigen *iigen;
14707 +       unsigned int sigen;
14708 +
14709 +       sigen = au_sigen(inode->i_sb);
14710 +       iinfo = au_ii(inode);
14711 +       iigen = &iinfo->ii_generation;
14712 +       spin_lock(&iinfo->ii_genspin);
14713 +       iigen->ig_generation = sigen;
14714 +       if (half)
14715 +               au_ig_fset(iigen->ig_flags, HALF_REFRESHED);
14716 +       else
14717 +               au_ig_fclr(iigen->ig_flags, HALF_REFRESHED);
14718 +       spin_unlock(&iinfo->ii_genspin);
14719 +}
14720 +
14721 +/* it may be called at remount time, too */
14722 +void au_update_ibrange(struct inode *inode, int do_put_zero)
14723 +{
14724 +       struct au_iinfo *iinfo;
14725 +       aufs_bindex_t bindex, bend;
14726 +
14727 +       iinfo = au_ii(inode);
14728 +       if (!iinfo)
14729 +               return;
14730 +
14731 +       IiMustWriteLock(inode);
14732 +
14733 +       if (do_put_zero && iinfo->ii_bstart >= 0) {
14734 +               for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend;
14735 +                    bindex++) {
14736 +                       struct inode *h_i;
14737 +
14738 +                       h_i = iinfo->ii_hinode[0 + bindex].hi_inode;
14739 +                       if (h_i && !h_i->i_nlink)
14740 +                               au_set_h_iptr(inode, bindex, NULL, 0);
14741 +               }
14742 +       }
14743 +
14744 +       iinfo->ii_bstart = -1;
14745 +       iinfo->ii_bend = -1;
14746 +       bend = au_sbend(inode->i_sb);
14747 +       for (bindex = 0; bindex <= bend; bindex++)
14748 +               if (iinfo->ii_hinode[0 + bindex].hi_inode) {
14749 +                       iinfo->ii_bstart = bindex;
14750 +                       break;
14751 +               }
14752 +       if (iinfo->ii_bstart >= 0)
14753 +               for (bindex = bend; bindex >= iinfo->ii_bstart; bindex--)
14754 +                       if (iinfo->ii_hinode[0 + bindex].hi_inode) {
14755 +                               iinfo->ii_bend = bindex;
14756 +                               break;
14757 +                       }
14758 +       AuDebugOn(iinfo->ii_bstart > iinfo->ii_bend);
14759 +}
14760 +
14761 +/* ---------------------------------------------------------------------- */
14762 +
14763 +void au_icntnr_init_once(void *_c)
14764 +{
14765 +       struct au_icntnr *c = _c;
14766 +       struct au_iinfo *iinfo = &c->iinfo;
14767 +       static struct lock_class_key aufs_ii;
14768 +
14769 +       spin_lock_init(&iinfo->ii_genspin);
14770 +       au_rw_init(&iinfo->ii_rwsem);
14771 +       au_rw_class(&iinfo->ii_rwsem, &aufs_ii);
14772 +       inode_init_once(&c->vfs_inode);
14773 +}
14774 +
14775 +int au_iinfo_init(struct inode *inode)
14776 +{
14777 +       struct au_iinfo *iinfo;
14778 +       struct super_block *sb;
14779 +       int nbr, i;
14780 +
14781 +       sb = inode->i_sb;
14782 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
14783 +       nbr = au_sbend(sb) + 1;
14784 +       if (unlikely(nbr <= 0))
14785 +               nbr = 1;
14786 +       iinfo->ii_hinode = kcalloc(nbr, sizeof(*iinfo->ii_hinode), GFP_NOFS);
14787 +       if (iinfo->ii_hinode) {
14788 +               au_ninodes_inc(sb);
14789 +               for (i = 0; i < nbr; i++)
14790 +                       iinfo->ii_hinode[i].hi_id = -1;
14791 +
14792 +               iinfo->ii_generation.ig_generation = au_sigen(sb);
14793 +               iinfo->ii_bstart = -1;
14794 +               iinfo->ii_bend = -1;
14795 +               iinfo->ii_vdir = NULL;
14796 +               return 0;
14797 +       }
14798 +       return -ENOMEM;
14799 +}
14800 +
14801 +int au_ii_realloc(struct au_iinfo *iinfo, int nbr)
14802 +{
14803 +       int err, sz;
14804 +       struct au_hinode *hip;
14805 +
14806 +       AuRwMustWriteLock(&iinfo->ii_rwsem);
14807 +
14808 +       err = -ENOMEM;
14809 +       sz = sizeof(*hip) * (iinfo->ii_bend + 1);
14810 +       if (!sz)
14811 +               sz = sizeof(*hip);
14812 +       hip = au_kzrealloc(iinfo->ii_hinode, sz, sizeof(*hip) * nbr, GFP_NOFS);
14813 +       if (hip) {
14814 +               iinfo->ii_hinode = hip;
14815 +               err = 0;
14816 +       }
14817 +
14818 +       return err;
14819 +}
14820 +
14821 +void au_iinfo_fin(struct inode *inode)
14822 +{
14823 +       struct au_iinfo *iinfo;
14824 +       struct au_hinode *hi;
14825 +       struct super_block *sb;
14826 +       aufs_bindex_t bindex, bend;
14827 +       const unsigned char unlinked = !inode->i_nlink;
14828 +
14829 +       iinfo = au_ii(inode);
14830 +       /* bad_inode case */
14831 +       if (!iinfo)
14832 +               return;
14833 +
14834 +       sb = inode->i_sb;
14835 +       au_ninodes_dec(sb);
14836 +       if (si_pid_test(sb))
14837 +               au_xino_delete_inode(inode, unlinked);
14838 +       else {
14839 +               /*
14840 +                * it is safe to hide the dependency between sbinfo and
14841 +                * sb->s_umount.
14842 +                */
14843 +               lockdep_off();
14844 +               si_noflush_read_lock(sb);
14845 +               au_xino_delete_inode(inode, unlinked);
14846 +               si_read_unlock(sb);
14847 +               lockdep_on();
14848 +       }
14849 +
14850 +       if (iinfo->ii_vdir)
14851 +               au_vdir_free(iinfo->ii_vdir);
14852 +
14853 +       bindex = iinfo->ii_bstart;
14854 +       if (bindex >= 0) {
14855 +               hi = iinfo->ii_hinode + bindex;
14856 +               bend = iinfo->ii_bend;
14857 +               while (bindex++ <= bend) {
14858 +                       if (hi->hi_inode)
14859 +                               au_hiput(hi);
14860 +                       hi++;
14861 +               }
14862 +       }
14863 +       kfree(iinfo->ii_hinode);
14864 +       iinfo->ii_hinode = NULL;
14865 +       AuRwDestroy(&iinfo->ii_rwsem);
14866 +}
14867 diff -urN /usr/share/empty/fs/aufs/inode.c linux/fs/aufs/inode.c
14868 --- /usr/share/empty/fs/aufs/inode.c    1970-01-01 01:00:00.000000000 +0100
14869 +++ linux/fs/aufs/inode.c       2014-04-24 22:11:10.855269116 +0200
14870 @@ -0,0 +1,491 @@
14871 +/*
14872 + * Copyright (C) 2005-2014 Junjiro R. Okajima
14873 + *
14874 + * This program, aufs is free software; you can redistribute it and/or modify
14875 + * it under the terms of the GNU General Public License as published by
14876 + * the Free Software Foundation; either version 2 of the License, or
14877 + * (at your option) any later version.
14878 + *
14879 + * This program is distributed in the hope that it will be useful,
14880 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
14881 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
14882 + * GNU General Public License for more details.
14883 + *
14884 + * You should have received a copy of the GNU General Public License
14885 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
14886 + */
14887 +
14888 +/*
14889 + * inode functions
14890 + */
14891 +
14892 +#include "aufs.h"
14893 +
14894 +struct inode *au_igrab(struct inode *inode)
14895 +{
14896 +       if (inode) {
14897 +               AuDebugOn(!atomic_read(&inode->i_count));
14898 +               ihold(inode);
14899 +       }
14900 +       return inode;
14901 +}
14902 +
14903 +static void au_refresh_hinode_attr(struct inode *inode, int do_version)
14904 +{
14905 +       au_cpup_attr_all(inode, /*force*/0);
14906 +       au_update_iigen(inode, /*half*/1);
14907 +       if (do_version)
14908 +               inode->i_version++;
14909 +}
14910 +
14911 +static int au_ii_refresh(struct inode *inode, int *update)
14912 +{
14913 +       int err, e;
14914 +       umode_t type;
14915 +       aufs_bindex_t bindex, new_bindex;
14916 +       struct super_block *sb;
14917 +       struct au_iinfo *iinfo;
14918 +       struct au_hinode *p, *q, tmp;
14919 +
14920 +       IiMustWriteLock(inode);
14921 +
14922 +       *update = 0;
14923 +       sb = inode->i_sb;
14924 +       type = inode->i_mode & S_IFMT;
14925 +       iinfo = au_ii(inode);
14926 +       err = au_ii_realloc(iinfo, au_sbend(sb) + 1);
14927 +       if (unlikely(err))
14928 +               goto out;
14929 +
14930 +       AuDebugOn(iinfo->ii_bstart < 0);
14931 +       p = iinfo->ii_hinode + iinfo->ii_bstart;
14932 +       for (bindex = iinfo->ii_bstart; bindex <= iinfo->ii_bend;
14933 +            bindex++, p++) {
14934 +               if (!p->hi_inode)
14935 +                       continue;
14936 +
14937 +               AuDebugOn(type != (p->hi_inode->i_mode & S_IFMT));
14938 +               new_bindex = au_br_index(sb, p->hi_id);
14939 +               if (new_bindex == bindex)
14940 +                       continue;
14941 +
14942 +               if (new_bindex < 0) {
14943 +                       *update = 1;
14944 +                       au_hiput(p);
14945 +                       p->hi_inode = NULL;
14946 +                       continue;
14947 +               }
14948 +
14949 +               if (new_bindex < iinfo->ii_bstart)
14950 +                       iinfo->ii_bstart = new_bindex;
14951 +               if (iinfo->ii_bend < new_bindex)
14952 +                       iinfo->ii_bend = new_bindex;
14953 +               /* swap two lower inode, and loop again */
14954 +               q = iinfo->ii_hinode + new_bindex;
14955 +               tmp = *q;
14956 +               *q = *p;
14957 +               *p = tmp;
14958 +               if (tmp.hi_inode) {
14959 +                       bindex--;
14960 +                       p--;
14961 +               }
14962 +       }
14963 +       au_update_ibrange(inode, /*do_put_zero*/0);
14964 +       e = au_dy_irefresh(inode);
14965 +       if (unlikely(e && !err))
14966 +               err = e;
14967 +
14968 +out:
14969 +       AuTraceErr(err);
14970 +       return err;
14971 +}
14972 +
14973 +int au_refresh_hinode_self(struct inode *inode)
14974 +{
14975 +       int err, update;
14976 +
14977 +       err = au_ii_refresh(inode, &update);
14978 +       if (!err)
14979 +               au_refresh_hinode_attr(inode, update && S_ISDIR(inode->i_mode));
14980 +
14981 +       AuTraceErr(err);
14982 +       return err;
14983 +}
14984 +
14985 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry)
14986 +{
14987 +       int err, e, update;
14988 +       unsigned int flags;
14989 +       umode_t mode;
14990 +       aufs_bindex_t bindex, bend;
14991 +       unsigned char isdir;
14992 +       struct au_hinode *p;
14993 +       struct au_iinfo *iinfo;
14994 +
14995 +       err = au_ii_refresh(inode, &update);
14996 +       if (unlikely(err))
14997 +               goto out;
14998 +
14999 +       update = 0;
15000 +       iinfo = au_ii(inode);
15001 +       p = iinfo->ii_hinode + iinfo->ii_bstart;
15002 +       mode = (inode->i_mode & S_IFMT);
15003 +       isdir = S_ISDIR(mode);
15004 +       flags = au_hi_flags(inode, isdir);
15005 +       bend = au_dbend(dentry);
15006 +       for (bindex = au_dbstart(dentry); bindex <= bend; bindex++) {
15007 +               struct inode *h_i;
15008 +               struct dentry *h_d;
15009 +
15010 +               h_d = au_h_dptr(dentry, bindex);
15011 +               if (!h_d || !h_d->d_inode)
15012 +                       continue;
15013 +
15014 +               AuDebugOn(mode != (h_d->d_inode->i_mode & S_IFMT));
15015 +               if (iinfo->ii_bstart <= bindex && bindex <= iinfo->ii_bend) {
15016 +                       h_i = au_h_iptr(inode, bindex);
15017 +                       if (h_i) {
15018 +                               if (h_i == h_d->d_inode)
15019 +                                       continue;
15020 +                               err = -EIO;
15021 +                               break;
15022 +                       }
15023 +               }
15024 +               if (bindex < iinfo->ii_bstart)
15025 +                       iinfo->ii_bstart = bindex;
15026 +               if (iinfo->ii_bend < bindex)
15027 +                       iinfo->ii_bend = bindex;
15028 +               au_set_h_iptr(inode, bindex, au_igrab(h_d->d_inode), flags);
15029 +               update = 1;
15030 +       }
15031 +       au_update_ibrange(inode, /*do_put_zero*/0);
15032 +       e = au_dy_irefresh(inode);
15033 +       if (unlikely(e && !err))
15034 +               err = e;
15035 +       if (!err)
15036 +               au_refresh_hinode_attr(inode, update && isdir);
15037 +
15038 +out:
15039 +       AuTraceErr(err);
15040 +       return err;
15041 +}
15042 +
15043 +static int set_inode(struct inode *inode, struct dentry *dentry)
15044 +{
15045 +       int err;
15046 +       unsigned int flags;
15047 +       umode_t mode;
15048 +       aufs_bindex_t bindex, bstart, btail;
15049 +       unsigned char isdir;
15050 +       struct dentry *h_dentry;
15051 +       struct inode *h_inode;
15052 +       struct au_iinfo *iinfo;
15053 +
15054 +       IiMustWriteLock(inode);
15055 +
15056 +       err = 0;
15057 +       isdir = 0;
15058 +       bstart = au_dbstart(dentry);
15059 +       h_inode = au_h_dptr(dentry, bstart)->d_inode;
15060 +       mode = h_inode->i_mode;
15061 +       switch (mode & S_IFMT) {
15062 +       case S_IFREG:
15063 +               btail = au_dbtail(dentry);
15064 +               inode->i_op = &aufs_iop;
15065 +               inode->i_fop = &aufs_file_fop;
15066 +               err = au_dy_iaop(inode, bstart, h_inode);
15067 +               if (unlikely(err))
15068 +                       goto out;
15069 +               break;
15070 +       case S_IFDIR:
15071 +               isdir = 1;
15072 +               btail = au_dbtaildir(dentry);
15073 +               inode->i_op = &aufs_dir_iop;
15074 +               inode->i_fop = &aufs_dir_fop;
15075 +               break;
15076 +       case S_IFLNK:
15077 +               btail = au_dbtail(dentry);
15078 +               inode->i_op = &aufs_symlink_iop;
15079 +               break;
15080 +       case S_IFBLK:
15081 +       case S_IFCHR:
15082 +       case S_IFIFO:
15083 +       case S_IFSOCK:
15084 +               btail = au_dbtail(dentry);
15085 +               inode->i_op = &aufs_iop;
15086 +               au_init_special_fop(inode, mode, h_inode->i_rdev);
15087 +               break;
15088 +       default:
15089 +               AuIOErr("Unknown file type 0%o\n", mode);
15090 +               err = -EIO;
15091 +               goto out;
15092 +       }
15093 +
15094 +       /* do not set hnotify for whiteouted dirs (SHWH mode) */
15095 +       flags = au_hi_flags(inode, isdir);
15096 +       if (au_opt_test(au_mntflags(dentry->d_sb), SHWH)
15097 +           && au_ftest_hi(flags, HNOTIFY)
15098 +           && dentry->d_name.len > AUFS_WH_PFX_LEN
15099 +           && !memcmp(dentry->d_name.name, AUFS_WH_PFX, AUFS_WH_PFX_LEN))
15100 +               au_fclr_hi(flags, HNOTIFY);
15101 +       iinfo = au_ii(inode);
15102 +       iinfo->ii_bstart = bstart;
15103 +       iinfo->ii_bend = btail;
15104 +       for (bindex = bstart; bindex <= btail; bindex++) {
15105 +               h_dentry = au_h_dptr(dentry, bindex);
15106 +               if (h_dentry)
15107 +                       au_set_h_iptr(inode, bindex,
15108 +                                     au_igrab(h_dentry->d_inode), flags);
15109 +       }
15110 +       au_cpup_attr_all(inode, /*force*/1);
15111 +
15112 +out:
15113 +       return err;
15114 +}
15115 +
15116 +/*
15117 + * successful returns with iinfo write_locked
15118 + * minus: errno
15119 + * zero: success, matched
15120 + * plus: no error, but unmatched
15121 + */
15122 +static int reval_inode(struct inode *inode, struct dentry *dentry)
15123 +{
15124 +       int err;
15125 +       unsigned int gen;
15126 +       struct au_iigen iigen;
15127 +       aufs_bindex_t bindex, bend;
15128 +       struct inode *h_inode, *h_dinode;
15129 +
15130 +       /*
15131 +        * before this function, if aufs got any iinfo lock, it must be only
15132 +        * one, the parent dir.
15133 +        * it can happen by UDBA and the obsoleted inode number.
15134 +        */
15135 +       err = -EIO;
15136 +       if (unlikely(inode->i_ino == parent_ino(dentry)))
15137 +               goto out;
15138 +
15139 +       err = 1;
15140 +       ii_write_lock_new_child(inode);
15141 +       h_dinode = au_h_dptr(dentry, au_dbstart(dentry))->d_inode;
15142 +       bend = au_ibend(inode);
15143 +       for (bindex = au_ibstart(inode); bindex <= bend; bindex++) {
15144 +               h_inode = au_h_iptr(inode, bindex);
15145 +               if (!h_inode || h_inode != h_dinode)
15146 +                       continue;
15147 +
15148 +               err = 0;
15149 +               gen = au_iigen(inode, &iigen);
15150 +               if (gen == au_digen(dentry)
15151 +                   && !au_ig_ftest(iigen.ig_flags, HALF_REFRESHED))
15152 +                       break;
15153 +
15154 +               /* fully refresh inode using dentry */
15155 +               err = au_refresh_hinode(inode, dentry);
15156 +               if (!err)
15157 +                       au_update_iigen(inode, /*half*/0);
15158 +               break;
15159 +       }
15160 +
15161 +       if (unlikely(err))
15162 +               ii_write_unlock(inode);
15163 +out:
15164 +       return err;
15165 +}
15166 +
15167 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
15168 +          unsigned int d_type, ino_t *ino)
15169 +{
15170 +       int err;
15171 +       struct mutex *mtx;
15172 +
15173 +       /* prevent hardlinked inode number from race condition */
15174 +       mtx = NULL;
15175 +       if (d_type != DT_DIR) {
15176 +               mtx = &au_sbr(sb, bindex)->br_xino.xi_nondir_mtx;
15177 +               mutex_lock(mtx);
15178 +       }
15179 +       err = au_xino_read(sb, bindex, h_ino, ino);
15180 +       if (unlikely(err))
15181 +               goto out;
15182 +
15183 +       if (!*ino) {
15184 +               err = -EIO;
15185 +               *ino = au_xino_new_ino(sb);
15186 +               if (unlikely(!*ino))
15187 +                       goto out;
15188 +               err = au_xino_write(sb, bindex, h_ino, *ino);
15189 +               if (unlikely(err))
15190 +                       goto out;
15191 +       }
15192 +
15193 +out:
15194 +       if (mtx)
15195 +               mutex_unlock(mtx);
15196 +       return err;
15197 +}
15198 +
15199 +/* successful returns with iinfo write_locked */
15200 +/* todo: return with unlocked? */
15201 +struct inode *au_new_inode(struct dentry *dentry, int must_new)
15202 +{
15203 +       struct inode *inode, *h_inode;
15204 +       struct dentry *h_dentry;
15205 +       struct super_block *sb;
15206 +       struct mutex *mtx;
15207 +       ino_t h_ino, ino;
15208 +       int err;
15209 +       aufs_bindex_t bstart;
15210 +
15211 +       sb = dentry->d_sb;
15212 +       bstart = au_dbstart(dentry);
15213 +       h_dentry = au_h_dptr(dentry, bstart);
15214 +       h_inode = h_dentry->d_inode;
15215 +       h_ino = h_inode->i_ino;
15216 +
15217 +       /*
15218 +        * stop 'race'-ing between hardlinks under different
15219 +        * parents.
15220 +        */
15221 +       mtx = NULL;
15222 +       if (!S_ISDIR(h_inode->i_mode))
15223 +               mtx = &au_sbr(sb, bstart)->br_xino.xi_nondir_mtx;
15224 +
15225 +new_ino:
15226 +       if (mtx)
15227 +               mutex_lock(mtx);
15228 +       err = au_xino_read(sb, bstart, h_ino, &ino);
15229 +       inode = ERR_PTR(err);
15230 +       if (unlikely(err))
15231 +               goto out;
15232 +
15233 +       if (!ino) {
15234 +               ino = au_xino_new_ino(sb);
15235 +               if (unlikely(!ino)) {
15236 +                       inode = ERR_PTR(-EIO);
15237 +                       goto out;
15238 +               }
15239 +       }
15240 +
15241 +       AuDbg("i%lu\n", (unsigned long)ino);
15242 +       inode = au_iget_locked(sb, ino);
15243 +       err = PTR_ERR(inode);
15244 +       if (IS_ERR(inode))
15245 +               goto out;
15246 +
15247 +       AuDbg("%lx, new %d\n", inode->i_state, !!(inode->i_state & I_NEW));
15248 +       if (inode->i_state & I_NEW) {
15249 +               /* verbose coding for lock class name */
15250 +               if (unlikely(S_ISLNK(h_inode->i_mode)))
15251 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
15252 +                                   au_lc_key + AuLcSymlink_IIINFO);
15253 +               else if (unlikely(S_ISDIR(h_inode->i_mode)))
15254 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
15255 +                                   au_lc_key + AuLcDir_IIINFO);
15256 +               else /* likely */
15257 +                       au_rw_class(&au_ii(inode)->ii_rwsem,
15258 +                                   au_lc_key + AuLcNonDir_IIINFO);
15259 +
15260 +               ii_write_lock_new_child(inode);
15261 +               err = set_inode(inode, dentry);
15262 +               if (!err) {
15263 +                       unlock_new_inode(inode);
15264 +                       goto out; /* success */
15265 +               }
15266 +
15267 +               /*
15268 +                * iget_failed() calls iput(), but we need to call
15269 +                * ii_write_unlock() after iget_failed(). so dirty hack for
15270 +                * i_count.
15271 +                */
15272 +               atomic_inc(&inode->i_count);
15273 +               iget_failed(inode);
15274 +               ii_write_unlock(inode);
15275 +               au_xino_write(sb, bstart, h_ino, /*ino*/0);
15276 +               /* ignore this error */
15277 +               goto out_iput;
15278 +       } else if (!must_new && !IS_DEADDIR(inode) && inode->i_nlink) {
15279 +               /*
15280 +                * horrible race condition between lookup, readdir and copyup
15281 +                * (or something).
15282 +                */
15283 +               if (mtx)
15284 +                       mutex_unlock(mtx);
15285 +               err = reval_inode(inode, dentry);
15286 +               if (unlikely(err < 0)) {
15287 +                       mtx = NULL;
15288 +                       goto out_iput;
15289 +               }
15290 +
15291 +               if (!err) {
15292 +                       mtx = NULL;
15293 +                       goto out; /* success */
15294 +               } else if (mtx)
15295 +                       mutex_lock(mtx);
15296 +       }
15297 +
15298 +       if (unlikely(au_test_fs_unique_ino(h_dentry->d_inode)))
15299 +               AuWarn1("Warning: Un-notified UDBA or repeatedly renamed dir,"
15300 +                       " b%d, %s, %pd, hi%lu, i%lu.\n",
15301 +                       bstart, au_sbtype(h_dentry->d_sb), dentry,
15302 +                       (unsigned long)h_ino, (unsigned long)ino);
15303 +       ino = 0;
15304 +       err = au_xino_write(sb, bstart, h_ino, /*ino*/0);
15305 +       if (!err) {
15306 +               iput(inode);
15307 +               if (mtx)
15308 +                       mutex_unlock(mtx);
15309 +               goto new_ino;
15310 +       }
15311 +
15312 +out_iput:
15313 +       iput(inode);
15314 +       inode = ERR_PTR(err);
15315 +out:
15316 +       if (mtx)
15317 +               mutex_unlock(mtx);
15318 +       return inode;
15319 +}
15320 +
15321 +/* ---------------------------------------------------------------------- */
15322 +
15323 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
15324 +              struct inode *inode)
15325 +{
15326 +       int err;
15327 +
15328 +       err = au_br_rdonly(au_sbr(sb, bindex));
15329 +
15330 +       /* pseudo-link after flushed may happen out of bounds */
15331 +       if (!err
15332 +           && inode
15333 +           && au_ibstart(inode) <= bindex
15334 +           && bindex <= au_ibend(inode)) {
15335 +               /*
15336 +                * permission check is unnecessary since vfsub routine
15337 +                * will be called later
15338 +                */
15339 +               struct inode *hi = au_h_iptr(inode, bindex);
15340 +               if (hi)
15341 +                       err = IS_IMMUTABLE(hi) ? -EROFS : 0;
15342 +       }
15343 +
15344 +       return err;
15345 +}
15346 +
15347 +int au_test_h_perm(struct inode *h_inode, int mask)
15348 +{
15349 +       if (uid_eq(current_fsuid(), GLOBAL_ROOT_UID))
15350 +               return 0;
15351 +       return inode_permission(h_inode, mask);
15352 +}
15353 +
15354 +int au_test_h_perm_sio(struct inode *h_inode, int mask)
15355 +{
15356 +       if (au_test_nfs(h_inode->i_sb)
15357 +           && (mask & MAY_WRITE)
15358 +           && S_ISDIR(h_inode->i_mode))
15359 +               mask |= MAY_READ; /* force permission check */
15360 +       return au_test_h_perm(h_inode, mask);
15361 +}
15362 diff -urN /usr/share/empty/fs/aufs/inode.h linux/fs/aufs/inode.h
15363 --- /usr/share/empty/fs/aufs/inode.h    1970-01-01 01:00:00.000000000 +0100
15364 +++ linux/fs/aufs/inode.h       2014-04-24 22:11:10.855269116 +0200
15365 @@ -0,0 +1,599 @@
15366 +/*
15367 + * Copyright (C) 2005-2014 Junjiro R. Okajima
15368 + *
15369 + * This program, aufs is free software; you can redistribute it and/or modify
15370 + * it under the terms of the GNU General Public License as published by
15371 + * the Free Software Foundation; either version 2 of the License, or
15372 + * (at your option) any later version.
15373 + *
15374 + * This program is distributed in the hope that it will be useful,
15375 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15376 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15377 + * GNU General Public License for more details.
15378 + *
15379 + * You should have received a copy of the GNU General Public License
15380 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15381 + */
15382 +
15383 +/*
15384 + * inode operations
15385 + */
15386 +
15387 +#ifndef __AUFS_INODE_H__
15388 +#define __AUFS_INODE_H__
15389 +
15390 +#ifdef __KERNEL__
15391 +
15392 +#include <linux/fsnotify.h>
15393 +#include "rwsem.h"
15394 +
15395 +struct vfsmount;
15396 +
15397 +struct au_hnotify {
15398 +#ifdef CONFIG_AUFS_HNOTIFY
15399 +#ifdef CONFIG_AUFS_HFSNOTIFY
15400 +       /* never use fsnotify_add_vfsmount_mark() */
15401 +       struct fsnotify_mark            hn_mark;
15402 +#endif
15403 +       struct inode                    *hn_aufs_inode; /* no get/put */
15404 +#endif
15405 +} ____cacheline_aligned_in_smp;
15406 +
15407 +struct au_hinode {
15408 +       struct inode            *hi_inode;
15409 +       aufs_bindex_t           hi_id;
15410 +#ifdef CONFIG_AUFS_HNOTIFY
15411 +       struct au_hnotify       *hi_notify;
15412 +#endif
15413 +
15414 +       /* reference to the copied-up whiteout with get/put */
15415 +       struct dentry           *hi_whdentry;
15416 +};
15417 +
15418 +/* ig_flags */
15419 +#define AuIG_HALF_REFRESHED            1
15420 +#define au_ig_ftest(flags, name)       ((flags) & AuIG_##name)
15421 +#define au_ig_fset(flags, name) \
15422 +       do { (flags) |= AuIG_##name; } while (0)
15423 +#define au_ig_fclr(flags, name) \
15424 +       do { (flags) &= ~AuIG_##name; } while (0)
15425 +
15426 +struct au_iigen {
15427 +       __u32           ig_generation, ig_flags;
15428 +};
15429 +
15430 +struct au_vdir;
15431 +struct au_iinfo {
15432 +       spinlock_t              ii_genspin;
15433 +       struct au_iigen         ii_generation;
15434 +       struct super_block      *ii_hsb1;       /* no get/put */
15435 +
15436 +       struct au_rwsem         ii_rwsem;
15437 +       aufs_bindex_t           ii_bstart, ii_bend;
15438 +       __u32                   ii_higen;
15439 +       struct au_hinode        *ii_hinode;
15440 +       struct au_vdir          *ii_vdir;
15441 +};
15442 +
15443 +struct au_icntnr {
15444 +       struct au_iinfo iinfo;
15445 +       struct inode vfs_inode;
15446 +} ____cacheline_aligned_in_smp;
15447 +
15448 +/* au_pin flags */
15449 +#define AuPin_DI_LOCKED                1
15450 +#define AuPin_MNT_WRITE                (1 << 1)
15451 +#define au_ftest_pin(flags, name)      ((flags) & AuPin_##name)
15452 +#define au_fset_pin(flags, name) \
15453 +       do { (flags) |= AuPin_##name; } while (0)
15454 +#define au_fclr_pin(flags, name) \
15455 +       do { (flags) &= ~AuPin_##name; } while (0)
15456 +
15457 +struct au_pin {
15458 +       /* input */
15459 +       struct dentry *dentry;
15460 +       unsigned int udba;
15461 +       unsigned char lsc_di, lsc_hi, flags;
15462 +       aufs_bindex_t bindex;
15463 +
15464 +       /* output */
15465 +       struct dentry *parent;
15466 +       struct au_hinode *hdir;
15467 +       struct vfsmount *h_mnt;
15468 +
15469 +       /* temporary unlock/relock for copyup */
15470 +       struct dentry *h_dentry, *h_parent;
15471 +       struct au_branch *br;
15472 +       struct task_struct *task;
15473 +};
15474 +
15475 +void au_pin_hdir_unlock(struct au_pin *p);
15476 +int au_pin_hdir_relock(struct au_pin *p);
15477 +void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task);
15478 +void au_pin_hdir_acquire_nest(struct au_pin *p);
15479 +void au_pin_hdir_release(struct au_pin *p);
15480 +
15481 +/* ---------------------------------------------------------------------- */
15482 +
15483 +static inline struct au_iinfo *au_ii(struct inode *inode)
15484 +{
15485 +       struct au_iinfo *iinfo;
15486 +
15487 +       iinfo = &(container_of(inode, struct au_icntnr, vfs_inode)->iinfo);
15488 +       if (iinfo->ii_hinode)
15489 +               return iinfo;
15490 +       return NULL; /* debugging bad_inode case */
15491 +}
15492 +
15493 +/* ---------------------------------------------------------------------- */
15494 +
15495 +/* inode.c */
15496 +struct inode *au_igrab(struct inode *inode);
15497 +int au_refresh_hinode_self(struct inode *inode);
15498 +int au_refresh_hinode(struct inode *inode, struct dentry *dentry);
15499 +int au_ino(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
15500 +          unsigned int d_type, ino_t *ino);
15501 +struct inode *au_new_inode(struct dentry *dentry, int must_new);
15502 +int au_test_ro(struct super_block *sb, aufs_bindex_t bindex,
15503 +              struct inode *inode);
15504 +int au_test_h_perm(struct inode *h_inode, int mask);
15505 +int au_test_h_perm_sio(struct inode *h_inode, int mask);
15506 +
15507 +static inline int au_wh_ino(struct super_block *sb, aufs_bindex_t bindex,
15508 +                           ino_t h_ino, unsigned int d_type, ino_t *ino)
15509 +{
15510 +#ifdef CONFIG_AUFS_SHWH
15511 +       return au_ino(sb, bindex, h_ino, d_type, ino);
15512 +#else
15513 +       return 0;
15514 +#endif
15515 +}
15516 +
15517 +/* i_op.c */
15518 +extern struct inode_operations aufs_iop, aufs_symlink_iop, aufs_dir_iop;
15519 +
15520 +/* au_wr_dir flags */
15521 +#define AuWrDir_ADD_ENTRY      1
15522 +#define AuWrDir_TMP_WHENTRY    (1 << 1)
15523 +#define AuWrDir_ISDIR          (1 << 2)
15524 +#define au_ftest_wrdir(flags, name)    ((flags) & AuWrDir_##name)
15525 +#define au_fset_wrdir(flags, name) \
15526 +       do { (flags) |= AuWrDir_##name; } while (0)
15527 +#define au_fclr_wrdir(flags, name) \
15528 +       do { (flags) &= ~AuWrDir_##name; } while (0)
15529 +
15530 +struct au_wr_dir_args {
15531 +       aufs_bindex_t force_btgt;
15532 +       unsigned char flags;
15533 +};
15534 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
15535 +             struct au_wr_dir_args *args);
15536 +
15537 +struct dentry *au_pinned_h_parent(struct au_pin *pin);
15538 +void au_pin_init(struct au_pin *pin, struct dentry *dentry,
15539 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
15540 +                unsigned int udba, unsigned char flags);
15541 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
15542 +          unsigned int udba, unsigned char flags) __must_check;
15543 +int au_do_pin(struct au_pin *pin) __must_check;
15544 +void au_unpin(struct au_pin *pin);
15545 +
15546 +/* i_op_add.c */
15547 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
15548 +              struct dentry *h_parent, int isdir);
15549 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
15550 +              dev_t dev);
15551 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname);
15552 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
15553 +               bool want_excl);
15554 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
15555 +             struct dentry *dentry);
15556 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
15557 +
15558 +/* i_op_del.c */
15559 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup);
15560 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
15561 +              struct dentry *h_parent, int isdir);
15562 +int aufs_unlink(struct inode *dir, struct dentry *dentry);
15563 +int aufs_rmdir(struct inode *dir, struct dentry *dentry);
15564 +
15565 +/* i_op_ren.c */
15566 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt);
15567 +int aufs_rename(struct inode *src_dir, struct dentry *src_dentry,
15568 +               struct inode *dir, struct dentry *dentry);
15569 +
15570 +/* iinfo.c */
15571 +struct inode *au_h_iptr(struct inode *inode, aufs_bindex_t bindex);
15572 +void au_hiput(struct au_hinode *hinode);
15573 +void au_set_hi_wh(struct inode *inode, aufs_bindex_t bindex,
15574 +                 struct dentry *h_wh);
15575 +unsigned int au_hi_flags(struct inode *inode, int isdir);
15576 +
15577 +/* hinode flags */
15578 +#define AuHi_XINO      1
15579 +#define AuHi_HNOTIFY   (1 << 1)
15580 +#define au_ftest_hi(flags, name)       ((flags) & AuHi_##name)
15581 +#define au_fset_hi(flags, name) \
15582 +       do { (flags) |= AuHi_##name; } while (0)
15583 +#define au_fclr_hi(flags, name) \
15584 +       do { (flags) &= ~AuHi_##name; } while (0)
15585 +
15586 +#ifndef CONFIG_AUFS_HNOTIFY
15587 +#undef AuHi_HNOTIFY
15588 +#define AuHi_HNOTIFY   0
15589 +#endif
15590 +
15591 +void au_set_h_iptr(struct inode *inode, aufs_bindex_t bindex,
15592 +                  struct inode *h_inode, unsigned int flags);
15593 +
15594 +void au_update_iigen(struct inode *inode, int half);
15595 +void au_update_ibrange(struct inode *inode, int do_put_zero);
15596 +
15597 +void au_icntnr_init_once(void *_c);
15598 +int au_iinfo_init(struct inode *inode);
15599 +void au_iinfo_fin(struct inode *inode);
15600 +int au_ii_realloc(struct au_iinfo *iinfo, int nbr);
15601 +
15602 +#ifdef CONFIG_PROC_FS
15603 +/* plink.c */
15604 +int au_plink_maint(struct super_block *sb, int flags);
15605 +void au_plink_maint_leave(struct au_sbinfo *sbinfo);
15606 +int au_plink_maint_enter(struct super_block *sb);
15607 +#ifdef CONFIG_AUFS_DEBUG
15608 +void au_plink_list(struct super_block *sb);
15609 +#else
15610 +AuStubVoid(au_plink_list, struct super_block *sb)
15611 +#endif
15612 +int au_plink_test(struct inode *inode);
15613 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex);
15614 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
15615 +                    struct dentry *h_dentry);
15616 +void au_plink_put(struct super_block *sb, int verbose);
15617 +void au_plink_clean(struct super_block *sb, int verbose);
15618 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id);
15619 +#else
15620 +AuStubInt0(au_plink_maint, struct super_block *sb, int flags);
15621 +AuStubVoid(au_plink_maint_leave, struct au_sbinfo *sbinfo);
15622 +AuStubInt0(au_plink_maint_enter, struct super_block *sb);
15623 +AuStubVoid(au_plink_list, struct super_block *sb);
15624 +AuStubInt0(au_plink_test, struct inode *inode);
15625 +AuStub(struct dentry *, au_plink_lkup, return NULL,
15626 +       struct inode *inode, aufs_bindex_t bindex);
15627 +AuStubVoid(au_plink_append, struct inode *inode, aufs_bindex_t bindex,
15628 +          struct dentry *h_dentry);
15629 +AuStubVoid(au_plink_put, struct super_block *sb, int verbose);
15630 +AuStubVoid(au_plink_clean, struct super_block *sb, int verbose);
15631 +AuStubVoid(au_plink_half_refresh, struct super_block *sb, aufs_bindex_t br_id);
15632 +#endif /* CONFIG_PROC_FS */
15633 +
15634 +/* ---------------------------------------------------------------------- */
15635 +
15636 +/* lock subclass for iinfo */
15637 +enum {
15638 +       AuLsc_II_CHILD,         /* child first */
15639 +       AuLsc_II_CHILD2,        /* rename(2), link(2), and cpup at hnotify */
15640 +       AuLsc_II_CHILD3,        /* copyup dirs */
15641 +       AuLsc_II_PARENT,        /* see AuLsc_I_PARENT in vfsub.h */
15642 +       AuLsc_II_PARENT2,
15643 +       AuLsc_II_PARENT3,       /* copyup dirs */
15644 +       AuLsc_II_NEW_CHILD
15645 +};
15646 +
15647 +/*
15648 + * ii_read_lock_child, ii_write_lock_child,
15649 + * ii_read_lock_child2, ii_write_lock_child2,
15650 + * ii_read_lock_child3, ii_write_lock_child3,
15651 + * ii_read_lock_parent, ii_write_lock_parent,
15652 + * ii_read_lock_parent2, ii_write_lock_parent2,
15653 + * ii_read_lock_parent3, ii_write_lock_parent3,
15654 + * ii_read_lock_new_child, ii_write_lock_new_child,
15655 + */
15656 +#define AuReadLockFunc(name, lsc) \
15657 +static inline void ii_read_lock_##name(struct inode *i) \
15658 +{ \
15659 +       au_rw_read_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
15660 +}
15661 +
15662 +#define AuWriteLockFunc(name, lsc) \
15663 +static inline void ii_write_lock_##name(struct inode *i) \
15664 +{ \
15665 +       au_rw_write_lock_nested(&au_ii(i)->ii_rwsem, AuLsc_II_##lsc); \
15666 +}
15667 +
15668 +#define AuRWLockFuncs(name, lsc) \
15669 +       AuReadLockFunc(name, lsc) \
15670 +       AuWriteLockFunc(name, lsc)
15671 +
15672 +AuRWLockFuncs(child, CHILD);
15673 +AuRWLockFuncs(child2, CHILD2);
15674 +AuRWLockFuncs(child3, CHILD3);
15675 +AuRWLockFuncs(parent, PARENT);
15676 +AuRWLockFuncs(parent2, PARENT2);
15677 +AuRWLockFuncs(parent3, PARENT3);
15678 +AuRWLockFuncs(new_child, NEW_CHILD);
15679 +
15680 +#undef AuReadLockFunc
15681 +#undef AuWriteLockFunc
15682 +#undef AuRWLockFuncs
15683 +
15684 +/*
15685 + * ii_read_unlock, ii_write_unlock, ii_downgrade_lock
15686 + */
15687 +AuSimpleUnlockRwsemFuncs(ii, struct inode *i, &au_ii(i)->ii_rwsem);
15688 +
15689 +#define IiMustNoWaiters(i)     AuRwMustNoWaiters(&au_ii(i)->ii_rwsem)
15690 +#define IiMustAnyLock(i)       AuRwMustAnyLock(&au_ii(i)->ii_rwsem)
15691 +#define IiMustWriteLock(i)     AuRwMustWriteLock(&au_ii(i)->ii_rwsem)
15692 +
15693 +/* ---------------------------------------------------------------------- */
15694 +
15695 +static inline void au_icntnr_init(struct au_icntnr *c)
15696 +{
15697 +#ifdef CONFIG_AUFS_DEBUG
15698 +       c->vfs_inode.i_mode = 0;
15699 +#endif
15700 +}
15701 +
15702 +static inline unsigned int au_iigen(struct inode *inode, struct au_iigen *iigen)
15703 +{
15704 +       unsigned int gen;
15705 +       struct au_iinfo *iinfo;
15706 +
15707 +       iinfo = au_ii(inode);
15708 +       spin_lock(&iinfo->ii_genspin);
15709 +       if (iigen)
15710 +               *iigen = iinfo->ii_generation;
15711 +       gen = iinfo->ii_generation.ig_generation;
15712 +       spin_unlock(&iinfo->ii_genspin);
15713 +
15714 +       return gen;
15715 +}
15716 +
15717 +/* tiny test for inode number */
15718 +/* tmpfs generation is too rough */
15719 +static inline int au_test_higen(struct inode *inode, struct inode *h_inode)
15720 +{
15721 +       struct au_iinfo *iinfo;
15722 +
15723 +       iinfo = au_ii(inode);
15724 +       AuRwMustAnyLock(&iinfo->ii_rwsem);
15725 +       return !(iinfo->ii_hsb1 == h_inode->i_sb
15726 +                && iinfo->ii_higen == h_inode->i_generation);
15727 +}
15728 +
15729 +static inline void au_iigen_dec(struct inode *inode)
15730 +{
15731 +       struct au_iinfo *iinfo;
15732 +
15733 +       iinfo = au_ii(inode);
15734 +       spin_lock(&iinfo->ii_genspin);
15735 +       iinfo->ii_generation.ig_generation--;
15736 +       spin_unlock(&iinfo->ii_genspin);
15737 +}
15738 +
15739 +static inline int au_iigen_test(struct inode *inode, unsigned int sigen)
15740 +{
15741 +       int err;
15742 +
15743 +       err = 0;
15744 +       if (unlikely(inode && au_iigen(inode, NULL) != sigen))
15745 +               err = -EIO;
15746 +
15747 +       return err;
15748 +}
15749 +
15750 +/* ---------------------------------------------------------------------- */
15751 +
15752 +static inline aufs_bindex_t au_ii_br_id(struct inode *inode,
15753 +                                       aufs_bindex_t bindex)
15754 +{
15755 +       IiMustAnyLock(inode);
15756 +       return au_ii(inode)->ii_hinode[0 + bindex].hi_id;
15757 +}
15758 +
15759 +static inline aufs_bindex_t au_ibstart(struct inode *inode)
15760 +{
15761 +       IiMustAnyLock(inode);
15762 +       return au_ii(inode)->ii_bstart;
15763 +}
15764 +
15765 +static inline aufs_bindex_t au_ibend(struct inode *inode)
15766 +{
15767 +       IiMustAnyLock(inode);
15768 +       return au_ii(inode)->ii_bend;
15769 +}
15770 +
15771 +static inline struct au_vdir *au_ivdir(struct inode *inode)
15772 +{
15773 +       IiMustAnyLock(inode);
15774 +       return au_ii(inode)->ii_vdir;
15775 +}
15776 +
15777 +static inline struct dentry *au_hi_wh(struct inode *inode, aufs_bindex_t bindex)
15778 +{
15779 +       IiMustAnyLock(inode);
15780 +       return au_ii(inode)->ii_hinode[0 + bindex].hi_whdentry;
15781 +}
15782 +
15783 +static inline void au_set_ibstart(struct inode *inode, aufs_bindex_t bindex)
15784 +{
15785 +       IiMustWriteLock(inode);
15786 +       au_ii(inode)->ii_bstart = bindex;
15787 +}
15788 +
15789 +static inline void au_set_ibend(struct inode *inode, aufs_bindex_t bindex)
15790 +{
15791 +       IiMustWriteLock(inode);
15792 +       au_ii(inode)->ii_bend = bindex;
15793 +}
15794 +
15795 +static inline void au_set_ivdir(struct inode *inode, struct au_vdir *vdir)
15796 +{
15797 +       IiMustWriteLock(inode);
15798 +       au_ii(inode)->ii_vdir = vdir;
15799 +}
15800 +
15801 +static inline struct au_hinode *au_hi(struct inode *inode, aufs_bindex_t bindex)
15802 +{
15803 +       IiMustAnyLock(inode);
15804 +       return au_ii(inode)->ii_hinode + bindex;
15805 +}
15806 +
15807 +/* ---------------------------------------------------------------------- */
15808 +
15809 +static inline struct dentry *au_pinned_parent(struct au_pin *pin)
15810 +{
15811 +       if (pin)
15812 +               return pin->parent;
15813 +       return NULL;
15814 +}
15815 +
15816 +static inline struct inode *au_pinned_h_dir(struct au_pin *pin)
15817 +{
15818 +       if (pin && pin->hdir)
15819 +               return pin->hdir->hi_inode;
15820 +       return NULL;
15821 +}
15822 +
15823 +static inline struct au_hinode *au_pinned_hdir(struct au_pin *pin)
15824 +{
15825 +       if (pin)
15826 +               return pin->hdir;
15827 +       return NULL;
15828 +}
15829 +
15830 +static inline void au_pin_set_dentry(struct au_pin *pin, struct dentry *dentry)
15831 +{
15832 +       if (pin)
15833 +               pin->dentry = dentry;
15834 +}
15835 +
15836 +static inline void au_pin_set_parent_lflag(struct au_pin *pin,
15837 +                                          unsigned char lflag)
15838 +{
15839 +       if (pin) {
15840 +               if (lflag)
15841 +                       au_fset_pin(pin->flags, DI_LOCKED);
15842 +               else
15843 +                       au_fclr_pin(pin->flags, DI_LOCKED);
15844 +       }
15845 +}
15846 +
15847 +static inline void au_pin_set_parent(struct au_pin *pin, struct dentry *parent)
15848 +{
15849 +       if (pin) {
15850 +               dput(pin->parent);
15851 +               pin->parent = dget(parent);
15852 +       }
15853 +}
15854 +
15855 +/* ---------------------------------------------------------------------- */
15856 +
15857 +struct au_branch;
15858 +#ifdef CONFIG_AUFS_HNOTIFY
15859 +struct au_hnotify_op {
15860 +       void (*ctl)(struct au_hinode *hinode, int do_set);
15861 +       int (*alloc)(struct au_hinode *hinode);
15862 +
15863 +       /*
15864 +        * if it returns true, the the caller should free hinode->hi_notify,
15865 +        * otherwise ->free() frees it.
15866 +        */
15867 +       int (*free)(struct au_hinode *hinode,
15868 +                   struct au_hnotify *hn) __must_check;
15869 +
15870 +       void (*fin)(void);
15871 +       int (*init)(void);
15872 +
15873 +       int (*reset_br)(unsigned int udba, struct au_branch *br, int perm);
15874 +       void (*fin_br)(struct au_branch *br);
15875 +       int (*init_br)(struct au_branch *br, int perm);
15876 +};
15877 +
15878 +/* hnotify.c */
15879 +int au_hn_alloc(struct au_hinode *hinode, struct inode *inode);
15880 +void au_hn_free(struct au_hinode *hinode);
15881 +void au_hn_ctl(struct au_hinode *hinode, int do_set);
15882 +void au_hn_reset(struct inode *inode, unsigned int flags);
15883 +int au_hnotify(struct inode *h_dir, struct au_hnotify *hnotify, u32 mask,
15884 +              struct qstr *h_child_qstr, struct inode *h_child_inode);
15885 +int au_hnotify_reset_br(unsigned int udba, struct au_branch *br, int perm);
15886 +int au_hnotify_init_br(struct au_branch *br, int perm);
15887 +void au_hnotify_fin_br(struct au_branch *br);
15888 +int __init au_hnotify_init(void);
15889 +void au_hnotify_fin(void);
15890 +
15891 +/* hfsnotify.c */
15892 +extern const struct au_hnotify_op au_hnotify_op;
15893 +
15894 +static inline
15895 +void au_hn_init(struct au_hinode *hinode)
15896 +{
15897 +       hinode->hi_notify = NULL;
15898 +}
15899 +
15900 +static inline struct au_hnotify *au_hn(struct au_hinode *hinode)
15901 +{
15902 +       return hinode->hi_notify;
15903 +}
15904 +
15905 +#else
15906 +static inline
15907 +int au_hn_alloc(struct au_hinode *hinode __maybe_unused,
15908 +               struct inode *inode __maybe_unused)
15909 +{
15910 +       return -EOPNOTSUPP;
15911 +}
15912 +
15913 +static inline struct au_hnotify *au_hn(struct au_hinode *hinode)
15914 +{
15915 +       return NULL;
15916 +}
15917 +
15918 +AuStubVoid(au_hn_free, struct au_hinode *hinode __maybe_unused)
15919 +AuStubVoid(au_hn_ctl, struct au_hinode *hinode __maybe_unused,
15920 +          int do_set __maybe_unused)
15921 +AuStubVoid(au_hn_reset, struct inode *inode __maybe_unused,
15922 +          unsigned int flags __maybe_unused)
15923 +AuStubInt0(au_hnotify_reset_br, unsigned int udba __maybe_unused,
15924 +          struct au_branch *br __maybe_unused,
15925 +          int perm __maybe_unused)
15926 +AuStubInt0(au_hnotify_init_br, struct au_branch *br __maybe_unused,
15927 +          int perm __maybe_unused)
15928 +AuStubVoid(au_hnotify_fin_br, struct au_branch *br __maybe_unused)
15929 +AuStubInt0(__init au_hnotify_init, void)
15930 +AuStubVoid(au_hnotify_fin, void)
15931 +AuStubVoid(au_hn_init, struct au_hinode *hinode __maybe_unused)
15932 +#endif /* CONFIG_AUFS_HNOTIFY */
15933 +
15934 +static inline void au_hn_suspend(struct au_hinode *hdir)
15935 +{
15936 +       au_hn_ctl(hdir, /*do_set*/0);
15937 +}
15938 +
15939 +static inline void au_hn_resume(struct au_hinode *hdir)
15940 +{
15941 +       au_hn_ctl(hdir, /*do_set*/1);
15942 +}
15943 +
15944 +static inline void au_hn_imtx_lock(struct au_hinode *hdir)
15945 +{
15946 +       mutex_lock(&hdir->hi_inode->i_mutex);
15947 +       au_hn_suspend(hdir);
15948 +}
15949 +
15950 +static inline void au_hn_imtx_lock_nested(struct au_hinode *hdir,
15951 +                                         unsigned int sc __maybe_unused)
15952 +{
15953 +       mutex_lock_nested(&hdir->hi_inode->i_mutex, sc);
15954 +       au_hn_suspend(hdir);
15955 +}
15956 +
15957 +static inline void au_hn_imtx_unlock(struct au_hinode *hdir)
15958 +{
15959 +       au_hn_resume(hdir);
15960 +       mutex_unlock(&hdir->hi_inode->i_mutex);
15961 +}
15962 +
15963 +#endif /* __KERNEL__ */
15964 +#endif /* __AUFS_INODE_H__ */
15965 diff -urN /usr/share/empty/fs/aufs/ioctl.c linux/fs/aufs/ioctl.c
15966 --- /usr/share/empty/fs/aufs/ioctl.c    1970-01-01 01:00:00.000000000 +0100
15967 +++ linux/fs/aufs/ioctl.c       2014-04-24 22:11:10.855269116 +0200
15968 @@ -0,0 +1,201 @@
15969 +/*
15970 + * Copyright (C) 2005-2014 Junjiro R. Okajima
15971 + *
15972 + * This program, aufs is free software; you can redistribute it and/or modify
15973 + * it under the terms of the GNU General Public License as published by
15974 + * the Free Software Foundation; either version 2 of the License, or
15975 + * (at your option) any later version.
15976 + *
15977 + * This program is distributed in the hope that it will be useful,
15978 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
15979 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15980 + * GNU General Public License for more details.
15981 + *
15982 + * You should have received a copy of the GNU General Public License
15983 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
15984 + */
15985 +
15986 +/*
15987 + * ioctl
15988 + * plink-management and readdir in userspace.
15989 + * assist the pathconf(3) wrapper library.
15990 + * move-down
15991 + */
15992 +
15993 +#include <linux/compat.h>
15994 +#include <linux/file.h>
15995 +#include "aufs.h"
15996 +
15997 +static int au_wbr_fd(struct path *path, struct aufs_wbr_fd __user *arg)
15998 +{
15999 +       int err, fd;
16000 +       aufs_bindex_t wbi, bindex, bend;
16001 +       struct file *h_file;
16002 +       struct super_block *sb;
16003 +       struct dentry *root;
16004 +       struct au_branch *br;
16005 +       struct aufs_wbr_fd wbrfd = {
16006 +               .oflags = au_dir_roflags,
16007 +               .brid   = -1
16008 +       };
16009 +       const int valid = O_RDONLY | O_NONBLOCK | O_LARGEFILE | O_DIRECTORY
16010 +               | O_NOATIME | O_CLOEXEC;
16011 +
16012 +       AuDebugOn(wbrfd.oflags & ~valid);
16013 +
16014 +       if (arg) {
16015 +               err = copy_from_user(&wbrfd, arg, sizeof(wbrfd));
16016 +               if (unlikely(err)) {
16017 +                       err = -EFAULT;
16018 +                       goto out;
16019 +               }
16020 +
16021 +               err = -EINVAL;
16022 +               AuDbg("wbrfd{0%o, %d}\n", wbrfd.oflags, wbrfd.brid);
16023 +               wbrfd.oflags |= au_dir_roflags;
16024 +               AuDbg("0%o\n", wbrfd.oflags);
16025 +               if (unlikely(wbrfd.oflags & ~valid))
16026 +                       goto out;
16027 +       }
16028 +
16029 +       fd = get_unused_fd();
16030 +       err = fd;
16031 +       if (unlikely(fd < 0))
16032 +               goto out;
16033 +
16034 +       h_file = ERR_PTR(-EINVAL);
16035 +       wbi = 0;
16036 +       br = NULL;
16037 +       sb = path->dentry->d_sb;
16038 +       root = sb->s_root;
16039 +       aufs_read_lock(root, AuLock_IR);
16040 +       bend = au_sbend(sb);
16041 +       if (wbrfd.brid >= 0) {
16042 +               wbi = au_br_index(sb, wbrfd.brid);
16043 +               if (unlikely(wbi < 0 || wbi > bend))
16044 +                       goto out_unlock;
16045 +       }
16046 +
16047 +       h_file = ERR_PTR(-ENOENT);
16048 +       br = au_sbr(sb, wbi);
16049 +       if (!au_br_writable(br->br_perm)) {
16050 +               if (arg)
16051 +                       goto out_unlock;
16052 +
16053 +               bindex = wbi + 1;
16054 +               wbi = -1;
16055 +               for (; bindex <= bend; bindex++) {
16056 +                       br = au_sbr(sb, bindex);
16057 +                       if (au_br_writable(br->br_perm)) {
16058 +                               wbi = bindex;
16059 +                               br = au_sbr(sb, wbi);
16060 +                               break;
16061 +                       }
16062 +               }
16063 +       }
16064 +       AuDbg("wbi %d\n", wbi);
16065 +       if (wbi >= 0)
16066 +               h_file = au_h_open(root, wbi, wbrfd.oflags, NULL,
16067 +                                  /*force_wr*/0);
16068 +
16069 +out_unlock:
16070 +       aufs_read_unlock(root, AuLock_IR);
16071 +       err = PTR_ERR(h_file);
16072 +       if (IS_ERR(h_file))
16073 +               goto out_fd;
16074 +
16075 +       atomic_dec(&br->br_count); /* cf. au_h_open() */
16076 +       fd_install(fd, h_file);
16077 +       err = fd;
16078 +       goto out; /* success */
16079 +
16080 +out_fd:
16081 +       put_unused_fd(fd);
16082 +out:
16083 +       AuTraceErr(err);
16084 +       return err;
16085 +}
16086 +
16087 +/* ---------------------------------------------------------------------- */
16088 +
16089 +long aufs_ioctl_dir(struct file *file, unsigned int cmd, unsigned long arg)
16090 +{
16091 +       long err;
16092 +
16093 +       switch (cmd) {
16094 +       case AUFS_CTL_RDU:
16095 +       case AUFS_CTL_RDU_INO:
16096 +               err = au_rdu_ioctl(file, cmd, arg);
16097 +               break;
16098 +
16099 +       case AUFS_CTL_WBR_FD:
16100 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
16101 +               break;
16102 +
16103 +       case AUFS_CTL_IBUSY:
16104 +               err = au_ibusy_ioctl(file, arg);
16105 +               break;
16106 +
16107 +       default:
16108 +               /* do not call the lower */
16109 +               AuDbg("0x%x\n", cmd);
16110 +               err = -ENOTTY;
16111 +       }
16112 +
16113 +       AuTraceErr(err);
16114 +       return err;
16115 +}
16116 +
16117 +long aufs_ioctl_nondir(struct file *file, unsigned int cmd, unsigned long arg)
16118 +{
16119 +       long err;
16120 +
16121 +       switch (cmd) {
16122 +       case AUFS_CTL_MVDOWN:
16123 +               err = au_mvdown(file->f_dentry, (void __user *)arg);
16124 +               break;
16125 +
16126 +       case AUFS_CTL_WBR_FD:
16127 +               err = au_wbr_fd(&file->f_path, (void __user *)arg);
16128 +               break;
16129 +
16130 +       default:
16131 +               /* do not call the lower */
16132 +               AuDbg("0x%x\n", cmd);
16133 +               err = -ENOTTY;
16134 +       }
16135 +
16136 +       AuTraceErr(err);
16137 +       return err;
16138 +}
16139 +
16140 +#ifdef CONFIG_COMPAT
16141 +long aufs_compat_ioctl_dir(struct file *file, unsigned int cmd,
16142 +                          unsigned long arg)
16143 +{
16144 +       long err;
16145 +
16146 +       switch (cmd) {
16147 +       case AUFS_CTL_RDU:
16148 +       case AUFS_CTL_RDU_INO:
16149 +               err = au_rdu_compat_ioctl(file, cmd, arg);
16150 +               break;
16151 +
16152 +       case AUFS_CTL_IBUSY:
16153 +               err = au_ibusy_compat_ioctl(file, arg);
16154 +               break;
16155 +
16156 +       default:
16157 +               err = aufs_ioctl_dir(file, cmd, arg);
16158 +       }
16159 +
16160 +       AuTraceErr(err);
16161 +       return err;
16162 +}
16163 +
16164 +long aufs_compat_ioctl_nondir(struct file *file, unsigned int cmd,
16165 +                             unsigned long arg)
16166 +{
16167 +       return aufs_ioctl_nondir(file, cmd, (unsigned long)compat_ptr(arg));
16168 +}
16169 +#endif
16170 diff -urN /usr/share/empty/fs/aufs/i_op_add.c linux/fs/aufs/i_op_add.c
16171 --- /usr/share/empty/fs/aufs/i_op_add.c 1970-01-01 01:00:00.000000000 +0100
16172 +++ linux/fs/aufs/i_op_add.c    2014-04-24 22:11:10.851935747 +0200
16173 @@ -0,0 +1,762 @@
16174 +/*
16175 + * Copyright (C) 2005-2014 Junjiro R. Okajima
16176 + *
16177 + * This program, aufs is free software; you can redistribute it and/or modify
16178 + * it under the terms of the GNU General Public License as published by
16179 + * the Free Software Foundation; either version 2 of the License, or
16180 + * (at your option) any later version.
16181 + *
16182 + * This program is distributed in the hope that it will be useful,
16183 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16184 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16185 + * GNU General Public License for more details.
16186 + *
16187 + * You should have received a copy of the GNU General Public License
16188 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16189 + */
16190 +
16191 +/*
16192 + * inode operations (add entry)
16193 + */
16194 +
16195 +#include "aufs.h"
16196 +
16197 +/*
16198 + * final procedure of adding a new entry, except link(2).
16199 + * remove whiteout, instantiate, copyup the parent dir's times and size
16200 + * and update version.
16201 + * if it failed, re-create the removed whiteout.
16202 + */
16203 +static int epilog(struct inode *dir, aufs_bindex_t bindex,
16204 +                 struct dentry *wh_dentry, struct dentry *dentry)
16205 +{
16206 +       int err, rerr;
16207 +       aufs_bindex_t bwh;
16208 +       struct path h_path;
16209 +       struct inode *inode, *h_dir;
16210 +       struct dentry *wh;
16211 +
16212 +       bwh = -1;
16213 +       if (wh_dentry) {
16214 +               h_dir = wh_dentry->d_parent->d_inode; /* dir inode is locked */
16215 +               IMustLock(h_dir);
16216 +               AuDebugOn(au_h_iptr(dir, bindex) != h_dir);
16217 +               bwh = au_dbwh(dentry);
16218 +               h_path.dentry = wh_dentry;
16219 +               h_path.mnt = au_sbr_mnt(dir->i_sb, bindex);
16220 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path,
16221 +                                         dentry);
16222 +               if (unlikely(err))
16223 +                       goto out;
16224 +       }
16225 +
16226 +       inode = au_new_inode(dentry, /*must_new*/1);
16227 +       if (!IS_ERR(inode)) {
16228 +               d_instantiate(dentry, inode);
16229 +               dir = dentry->d_parent->d_inode; /* dir inode is locked */
16230 +               IMustLock(dir);
16231 +               if (au_ibstart(dir) == au_dbstart(dentry))
16232 +                       au_cpup_attr_timesizes(dir);
16233 +               dir->i_version++;
16234 +               return 0; /* success */
16235 +       }
16236 +
16237 +       err = PTR_ERR(inode);
16238 +       if (!wh_dentry)
16239 +               goto out;
16240 +
16241 +       /* revert */
16242 +       /* dir inode is locked */
16243 +       wh = au_wh_create(dentry, bwh, wh_dentry->d_parent);
16244 +       rerr = PTR_ERR(wh);
16245 +       if (IS_ERR(wh)) {
16246 +               AuIOErr("%pd reverting whiteout failed(%d, %d)\n",
16247 +                       dentry, err, rerr);
16248 +               err = -EIO;
16249 +       } else
16250 +               dput(wh);
16251 +
16252 +out:
16253 +       return err;
16254 +}
16255 +
16256 +static int au_d_may_add(struct dentry *dentry)
16257 +{
16258 +       int err;
16259 +
16260 +       err = 0;
16261 +       if (unlikely(d_unhashed(dentry)))
16262 +               err = -ENOENT;
16263 +       if (unlikely(dentry->d_inode))
16264 +               err = -EEXIST;
16265 +       return err;
16266 +}
16267 +
16268 +/*
16269 + * simple tests for the adding inode operations.
16270 + * following the checks in vfs, plus the parent-child relationship.
16271 + */
16272 +int au_may_add(struct dentry *dentry, aufs_bindex_t bindex,
16273 +              struct dentry *h_parent, int isdir)
16274 +{
16275 +       int err;
16276 +       umode_t h_mode;
16277 +       struct dentry *h_dentry;
16278 +       struct inode *h_inode;
16279 +
16280 +       err = -ENAMETOOLONG;
16281 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
16282 +               goto out;
16283 +
16284 +       h_dentry = au_h_dptr(dentry, bindex);
16285 +       h_inode = h_dentry->d_inode;
16286 +       if (!dentry->d_inode) {
16287 +               err = -EEXIST;
16288 +               if (unlikely(h_inode))
16289 +                       goto out;
16290 +       } else {
16291 +               /* rename(2) case */
16292 +               err = -EIO;
16293 +               if (unlikely(!h_inode || !h_inode->i_nlink))
16294 +                       goto out;
16295 +
16296 +               h_mode = h_inode->i_mode;
16297 +               if (!isdir) {
16298 +                       err = -EISDIR;
16299 +                       if (unlikely(S_ISDIR(h_mode)))
16300 +                               goto out;
16301 +               } else if (unlikely(!S_ISDIR(h_mode))) {
16302 +                       err = -ENOTDIR;
16303 +                       goto out;
16304 +               }
16305 +       }
16306 +
16307 +       err = 0;
16308 +       /* expected parent dir is locked */
16309 +       if (unlikely(h_parent != h_dentry->d_parent))
16310 +               err = -EIO;
16311 +
16312 +out:
16313 +       AuTraceErr(err);
16314 +       return err;
16315 +}
16316 +
16317 +/*
16318 + * initial procedure of adding a new entry.
16319 + * prepare writable branch and the parent dir, lock it,
16320 + * and lookup whiteout for the new entry.
16321 + */
16322 +static struct dentry*
16323 +lock_hdir_lkup_wh(struct dentry *dentry, struct au_dtime *dt,
16324 +                 struct dentry *src_dentry, struct au_pin *pin,
16325 +                 struct au_wr_dir_args *wr_dir_args)
16326 +{
16327 +       struct dentry *wh_dentry, *h_parent;
16328 +       struct super_block *sb;
16329 +       struct au_branch *br;
16330 +       int err;
16331 +       unsigned int udba;
16332 +       aufs_bindex_t bcpup;
16333 +
16334 +       AuDbg("%pd\n", dentry);
16335 +
16336 +       err = au_wr_dir(dentry, src_dentry, wr_dir_args);
16337 +       bcpup = err;
16338 +       wh_dentry = ERR_PTR(err);
16339 +       if (unlikely(err < 0))
16340 +               goto out;
16341 +
16342 +       sb = dentry->d_sb;
16343 +       udba = au_opt_udba(sb);
16344 +       err = au_pin(pin, dentry, bcpup, udba,
16345 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
16346 +       wh_dentry = ERR_PTR(err);
16347 +       if (unlikely(err))
16348 +               goto out;
16349 +
16350 +       h_parent = au_pinned_h_parent(pin);
16351 +       if (udba != AuOpt_UDBA_NONE
16352 +           && au_dbstart(dentry) == bcpup)
16353 +               err = au_may_add(dentry, bcpup, h_parent,
16354 +                                au_ftest_wrdir(wr_dir_args->flags, ISDIR));
16355 +       else if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
16356 +               err = -ENAMETOOLONG;
16357 +       wh_dentry = ERR_PTR(err);
16358 +       if (unlikely(err))
16359 +               goto out_unpin;
16360 +
16361 +       br = au_sbr(sb, bcpup);
16362 +       if (dt) {
16363 +               struct path tmp = {
16364 +                       .dentry = h_parent,
16365 +                       .mnt    = au_br_mnt(br)
16366 +               };
16367 +               au_dtime_store(dt, au_pinned_parent(pin), &tmp);
16368 +       }
16369 +
16370 +       wh_dentry = NULL;
16371 +       if (bcpup != au_dbwh(dentry))
16372 +               goto out; /* success */
16373 +
16374 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
16375 +
16376 +out_unpin:
16377 +       if (IS_ERR(wh_dentry))
16378 +               au_unpin(pin);
16379 +out:
16380 +       return wh_dentry;
16381 +}
16382 +
16383 +/* ---------------------------------------------------------------------- */
16384 +
16385 +enum { Mknod, Symlink, Creat };
16386 +struct simple_arg {
16387 +       int type;
16388 +       union {
16389 +               struct {
16390 +                       umode_t mode;
16391 +                       bool want_excl;
16392 +               } c;
16393 +               struct {
16394 +                       const char *symname;
16395 +               } s;
16396 +               struct {
16397 +                       umode_t mode;
16398 +                       dev_t dev;
16399 +               } m;
16400 +       } u;
16401 +};
16402 +
16403 +static int add_simple(struct inode *dir, struct dentry *dentry,
16404 +                     struct simple_arg *arg)
16405 +{
16406 +       int err;
16407 +       aufs_bindex_t bstart;
16408 +       unsigned char created;
16409 +       struct dentry *wh_dentry, *parent;
16410 +       struct inode *h_dir;
16411 +       /* to reuduce stack size */
16412 +       struct {
16413 +               struct au_dtime dt;
16414 +               struct au_pin pin;
16415 +               struct path h_path;
16416 +               struct au_wr_dir_args wr_dir_args;
16417 +       } *a;
16418 +
16419 +       AuDbg("%pd\n", dentry);
16420 +       IMustLock(dir);
16421 +
16422 +       err = -ENOMEM;
16423 +       a = kmalloc(sizeof(*a), GFP_NOFS);
16424 +       if (unlikely(!a))
16425 +               goto out;
16426 +       a->wr_dir_args.force_btgt = -1;
16427 +       a->wr_dir_args.flags = AuWrDir_ADD_ENTRY;
16428 +
16429 +       parent = dentry->d_parent; /* dir inode is locked */
16430 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
16431 +       if (unlikely(err))
16432 +               goto out_free;
16433 +       err = au_d_may_add(dentry);
16434 +       if (unlikely(err))
16435 +               goto out_unlock;
16436 +       di_write_lock_parent(parent);
16437 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
16438 +                                     &a->pin, &a->wr_dir_args);
16439 +       err = PTR_ERR(wh_dentry);
16440 +       if (IS_ERR(wh_dentry))
16441 +               goto out_parent;
16442 +
16443 +       bstart = au_dbstart(dentry);
16444 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
16445 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, bstart);
16446 +       h_dir = au_pinned_h_dir(&a->pin);
16447 +       switch (arg->type) {
16448 +       case Creat:
16449 +               err = vfsub_create(h_dir, &a->h_path, arg->u.c.mode,
16450 +                                  arg->u.c.want_excl);
16451 +               break;
16452 +       case Symlink:
16453 +               err = vfsub_symlink(h_dir, &a->h_path, arg->u.s.symname);
16454 +               break;
16455 +       case Mknod:
16456 +               err = vfsub_mknod(h_dir, &a->h_path, arg->u.m.mode,
16457 +                                 arg->u.m.dev);
16458 +               break;
16459 +       default:
16460 +               BUG();
16461 +       }
16462 +       created = !err;
16463 +       if (!err)
16464 +               err = epilog(dir, bstart, wh_dentry, dentry);
16465 +
16466 +       /* revert */
16467 +       if (unlikely(created && err && a->h_path.dentry->d_inode)) {
16468 +               int rerr;
16469 +               /* no delegation since it is just created */
16470 +               rerr = vfsub_unlink(h_dir, &a->h_path, /*delegated*/NULL,
16471 +                                   /*force*/0);
16472 +               if (rerr) {
16473 +                       AuIOErr("%pd revert failure(%d, %d)\n",
16474 +                               dentry, err, rerr);
16475 +                       err = -EIO;
16476 +               }
16477 +               au_dtime_revert(&a->dt);
16478 +       }
16479 +
16480 +       au_unpin(&a->pin);
16481 +       dput(wh_dentry);
16482 +
16483 +out_parent:
16484 +       di_write_unlock(parent);
16485 +out_unlock:
16486 +       if (unlikely(err)) {
16487 +               au_update_dbstart(dentry);
16488 +               d_drop(dentry);
16489 +       }
16490 +       aufs_read_unlock(dentry, AuLock_DW);
16491 +out_free:
16492 +       kfree(a);
16493 +out:
16494 +       return err;
16495 +}
16496 +
16497 +int aufs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
16498 +              dev_t dev)
16499 +{
16500 +       struct simple_arg arg = {
16501 +               .type = Mknod,
16502 +               .u.m = {
16503 +                       .mode   = mode,
16504 +                       .dev    = dev
16505 +               }
16506 +       };
16507 +       return add_simple(dir, dentry, &arg);
16508 +}
16509 +
16510 +int aufs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
16511 +{
16512 +       struct simple_arg arg = {
16513 +               .type = Symlink,
16514 +               .u.s.symname = symname
16515 +       };
16516 +       return add_simple(dir, dentry, &arg);
16517 +}
16518 +
16519 +int aufs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
16520 +               bool want_excl)
16521 +{
16522 +       struct simple_arg arg = {
16523 +               .type = Creat,
16524 +               .u.c = {
16525 +                       .mode           = mode,
16526 +                       .want_excl      = want_excl
16527 +               }
16528 +       };
16529 +       return add_simple(dir, dentry, &arg);
16530 +}
16531 +
16532 +/* ---------------------------------------------------------------------- */
16533 +
16534 +struct au_link_args {
16535 +       aufs_bindex_t bdst, bsrc;
16536 +       struct au_pin pin;
16537 +       struct path h_path;
16538 +       struct dentry *src_parent, *parent;
16539 +};
16540 +
16541 +static int au_cpup_before_link(struct dentry *src_dentry,
16542 +                              struct au_link_args *a)
16543 +{
16544 +       int err;
16545 +       struct dentry *h_src_dentry;
16546 +       struct au_cp_generic cpg = {
16547 +               .dentry = src_dentry,
16548 +               .bdst   = a->bdst,
16549 +               .bsrc   = a->bsrc,
16550 +               .len    = -1,
16551 +               .pin    = &a->pin,
16552 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN /* | AuCpup_KEEPLINO */
16553 +       };
16554 +
16555 +       di_read_lock_parent(a->src_parent, AuLock_IR);
16556 +       err = au_test_and_cpup_dirs(src_dentry, a->bdst);
16557 +       if (unlikely(err))
16558 +               goto out;
16559 +
16560 +       h_src_dentry = au_h_dptr(src_dentry, a->bsrc);
16561 +       err = au_pin(&a->pin, src_dentry, a->bdst,
16562 +                    au_opt_udba(src_dentry->d_sb),
16563 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
16564 +       if (unlikely(err))
16565 +               goto out;
16566 +
16567 +       err = au_sio_cpup_simple(&cpg);
16568 +       au_unpin(&a->pin);
16569 +
16570 +out:
16571 +       di_read_unlock(a->src_parent, AuLock_IR);
16572 +       return err;
16573 +}
16574 +
16575 +static int au_cpup_or_link(struct dentry *src_dentry, struct dentry *dentry,
16576 +                          struct au_link_args *a)
16577 +{
16578 +       int err;
16579 +       unsigned char plink;
16580 +       aufs_bindex_t bend;
16581 +       struct dentry *h_src_dentry;
16582 +       struct inode *h_inode, *inode, *delegated;
16583 +       struct super_block *sb;
16584 +       struct file *h_file;
16585 +
16586 +       plink = 0;
16587 +       h_inode = NULL;
16588 +       sb = src_dentry->d_sb;
16589 +       inode = src_dentry->d_inode;
16590 +       if (au_ibstart(inode) <= a->bdst)
16591 +               h_inode = au_h_iptr(inode, a->bdst);
16592 +       if (!h_inode || !h_inode->i_nlink) {
16593 +               /* copyup src_dentry as the name of dentry. */
16594 +               bend = au_dbend(dentry);
16595 +               if (bend < a->bsrc)
16596 +                       au_set_dbend(dentry, a->bsrc);
16597 +               au_set_h_dptr(dentry, a->bsrc,
16598 +                             dget(au_h_dptr(src_dentry, a->bsrc)));
16599 +               dget(a->h_path.dentry);
16600 +               au_set_h_dptr(dentry, a->bdst, NULL);
16601 +               dentry->d_inode = src_dentry->d_inode; /* tmp */
16602 +               h_file = au_h_open_pre(dentry, a->bsrc, /*force_wr*/0);
16603 +               if (IS_ERR(h_file))
16604 +                       err = PTR_ERR(h_file);
16605 +               else {
16606 +                       struct au_cp_generic cpg = {
16607 +                               .dentry = dentry,
16608 +                               .bdst   = a->bdst,
16609 +                               .bsrc   = -1,
16610 +                               .len    = -1,
16611 +                               .pin    = &a->pin,
16612 +                               .flags  = AuCpup_KEEPLINO
16613 +                       };
16614 +                       err = au_sio_cpup_simple(&cpg);
16615 +                       au_h_open_post(dentry, a->bsrc, h_file);
16616 +                       if (!err) {
16617 +                               dput(a->h_path.dentry);
16618 +                               a->h_path.dentry = au_h_dptr(dentry, a->bdst);
16619 +                       } else
16620 +                               au_set_h_dptr(dentry, a->bdst,
16621 +                                             a->h_path.dentry);
16622 +               }
16623 +               dentry->d_inode = NULL; /* restore */
16624 +               au_set_h_dptr(dentry, a->bsrc, NULL);
16625 +               au_set_dbend(dentry, bend);
16626 +       } else {
16627 +               /* the inode of src_dentry already exists on a.bdst branch */
16628 +               h_src_dentry = d_find_alias(h_inode);
16629 +               if (!h_src_dentry && au_plink_test(inode)) {
16630 +                       plink = 1;
16631 +                       h_src_dentry = au_plink_lkup(inode, a->bdst);
16632 +                       err = PTR_ERR(h_src_dentry);
16633 +                       if (IS_ERR(h_src_dentry))
16634 +                               goto out;
16635 +
16636 +                       if (unlikely(!h_src_dentry->d_inode)) {
16637 +                               dput(h_src_dentry);
16638 +                               h_src_dentry = NULL;
16639 +                       }
16640 +
16641 +               }
16642 +               if (h_src_dentry) {
16643 +                       delegated = NULL;
16644 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
16645 +                                        &a->h_path, &delegated);
16646 +                       if (unlikely(err == -EWOULDBLOCK)) {
16647 +                               pr_warn("cannot retry for NFSv4 delegation"
16648 +                                       " for an internal link\n");
16649 +                               iput(delegated);
16650 +                       }
16651 +                       dput(h_src_dentry);
16652 +               } else {
16653 +                       AuIOErr("no dentry found for hi%lu on b%d\n",
16654 +                               h_inode->i_ino, a->bdst);
16655 +                       err = -EIO;
16656 +               }
16657 +       }
16658 +
16659 +       if (!err && !plink)
16660 +               au_plink_append(inode, a->bdst, a->h_path.dentry);
16661 +
16662 +out:
16663 +       AuTraceErr(err);
16664 +       return err;
16665 +}
16666 +
16667 +int aufs_link(struct dentry *src_dentry, struct inode *dir,
16668 +             struct dentry *dentry)
16669 +{
16670 +       int err, rerr;
16671 +       struct au_dtime dt;
16672 +       struct au_link_args *a;
16673 +       struct dentry *wh_dentry, *h_src_dentry;
16674 +       struct inode *inode, *delegated;
16675 +       struct super_block *sb;
16676 +       struct au_wr_dir_args wr_dir_args = {
16677 +               /* .force_btgt  = -1, */
16678 +               .flags          = AuWrDir_ADD_ENTRY
16679 +       };
16680 +
16681 +       IMustLock(dir);
16682 +       inode = src_dentry->d_inode;
16683 +       IMustLock(inode);
16684 +
16685 +       err = -ENOMEM;
16686 +       a = kzalloc(sizeof(*a), GFP_NOFS);
16687 +       if (unlikely(!a))
16688 +               goto out;
16689 +
16690 +       a->parent = dentry->d_parent; /* dir inode is locked */
16691 +       err = aufs_read_and_write_lock2(dentry, src_dentry,
16692 +                                       AuLock_NOPLM | AuLock_GEN);
16693 +       if (unlikely(err))
16694 +               goto out_kfree;
16695 +       err = au_d_hashed_positive(src_dentry);
16696 +       if (unlikely(err))
16697 +               goto out_unlock;
16698 +       err = au_d_may_add(dentry);
16699 +       if (unlikely(err))
16700 +               goto out_unlock;
16701 +
16702 +       a->src_parent = dget_parent(src_dentry);
16703 +       wr_dir_args.force_btgt = au_ibstart(inode);
16704 +
16705 +       di_write_lock_parent(a->parent);
16706 +       wr_dir_args.force_btgt = au_wbr(dentry, wr_dir_args.force_btgt);
16707 +       wh_dentry = lock_hdir_lkup_wh(dentry, &dt, src_dentry, &a->pin,
16708 +                                     &wr_dir_args);
16709 +       err = PTR_ERR(wh_dentry);
16710 +       if (IS_ERR(wh_dentry))
16711 +               goto out_parent;
16712 +
16713 +       err = 0;
16714 +       sb = dentry->d_sb;
16715 +       a->bdst = au_dbstart(dentry);
16716 +       a->h_path.dentry = au_h_dptr(dentry, a->bdst);
16717 +       a->h_path.mnt = au_sbr_mnt(sb, a->bdst);
16718 +       a->bsrc = au_ibstart(inode);
16719 +       h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
16720 +       if (!h_src_dentry) {
16721 +               a->bsrc = au_dbstart(src_dentry);
16722 +               h_src_dentry = au_h_d_alias(src_dentry, a->bsrc);
16723 +               AuDebugOn(!h_src_dentry);
16724 +       } else if (IS_ERR(h_src_dentry))
16725 +               goto out_parent;
16726 +
16727 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
16728 +               if (a->bdst < a->bsrc
16729 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */)
16730 +                       err = au_cpup_or_link(src_dentry, dentry, a);
16731 +               else {
16732 +                       delegated = NULL;
16733 +                       err = vfsub_link(h_src_dentry, au_pinned_h_dir(&a->pin),
16734 +                                        &a->h_path, &delegated);
16735 +                       if (unlikely(err == -EWOULDBLOCK)) {
16736 +                               pr_warn("cannot retry for NFSv4 delegation"
16737 +                                       " for an internal link\n");
16738 +                               iput(delegated);
16739 +                       }
16740 +               }
16741 +               dput(h_src_dentry);
16742 +       } else {
16743 +               /*
16744 +                * copyup src_dentry to the branch we process,
16745 +                * and then link(2) to it.
16746 +                */
16747 +               dput(h_src_dentry);
16748 +               if (a->bdst < a->bsrc
16749 +                   /* && h_src_dentry->d_sb != a->h_path.dentry->d_sb */) {
16750 +                       au_unpin(&a->pin);
16751 +                       di_write_unlock(a->parent);
16752 +                       err = au_cpup_before_link(src_dentry, a);
16753 +                       di_write_lock_parent(a->parent);
16754 +                       if (!err)
16755 +                               err = au_pin(&a->pin, dentry, a->bdst,
16756 +                                            au_opt_udba(sb),
16757 +                                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
16758 +                       if (unlikely(err))
16759 +                               goto out_wh;
16760 +               }
16761 +               if (!err) {
16762 +                       h_src_dentry = au_h_dptr(src_dentry, a->bdst);
16763 +                       err = -ENOENT;
16764 +                       if (h_src_dentry && h_src_dentry->d_inode) {
16765 +                               delegated = NULL;
16766 +                               err = vfsub_link(h_src_dentry,
16767 +                                                au_pinned_h_dir(&a->pin),
16768 +                                                &a->h_path, &delegated);
16769 +                               if (unlikely(err == -EWOULDBLOCK)) {
16770 +                                       pr_warn("cannot retry"
16771 +                                               " for NFSv4 delegation"
16772 +                                               " for an internal link\n");
16773 +                                       iput(delegated);
16774 +                               }
16775 +                       }
16776 +               }
16777 +       }
16778 +       if (unlikely(err))
16779 +               goto out_unpin;
16780 +
16781 +       if (wh_dentry) {
16782 +               a->h_path.dentry = wh_dentry;
16783 +               err = au_wh_unlink_dentry(au_pinned_h_dir(&a->pin), &a->h_path,
16784 +                                         dentry);
16785 +               if (unlikely(err))
16786 +                       goto out_revert;
16787 +       }
16788 +
16789 +       dir->i_version++;
16790 +       if (au_ibstart(dir) == au_dbstart(dentry))
16791 +               au_cpup_attr_timesizes(dir);
16792 +       inc_nlink(inode);
16793 +       inode->i_ctime = dir->i_ctime;
16794 +       d_instantiate(dentry, au_igrab(inode));
16795 +       if (d_unhashed(a->h_path.dentry))
16796 +               /* some filesystem calls d_drop() */
16797 +               d_drop(dentry);
16798 +       goto out_unpin; /* success */
16799 +
16800 +out_revert:
16801 +       /* no delegation since it is just created */
16802 +       rerr = vfsub_unlink(au_pinned_h_dir(&a->pin), &a->h_path,
16803 +                           /*delegated*/NULL, /*force*/0);
16804 +       if (unlikely(rerr)) {
16805 +               AuIOErr("%pd reverting failed(%d, %d)\n", dentry, err, rerr);
16806 +               err = -EIO;
16807 +       }
16808 +       au_dtime_revert(&dt);
16809 +out_unpin:
16810 +       au_unpin(&a->pin);
16811 +out_wh:
16812 +       dput(wh_dentry);
16813 +out_parent:
16814 +       di_write_unlock(a->parent);
16815 +       dput(a->src_parent);
16816 +out_unlock:
16817 +       if (unlikely(err)) {
16818 +               au_update_dbstart(dentry);
16819 +               d_drop(dentry);
16820 +       }
16821 +       aufs_read_and_write_unlock2(dentry, src_dentry);
16822 +out_kfree:
16823 +       kfree(a);
16824 +out:
16825 +       AuTraceErr(err);
16826 +       return err;
16827 +}
16828 +
16829 +int aufs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
16830 +{
16831 +       int err, rerr;
16832 +       aufs_bindex_t bindex;
16833 +       unsigned char diropq;
16834 +       struct path h_path;
16835 +       struct dentry *wh_dentry, *parent, *opq_dentry;
16836 +       struct mutex *h_mtx;
16837 +       struct super_block *sb;
16838 +       struct {
16839 +               struct au_pin pin;
16840 +               struct au_dtime dt;
16841 +       } *a; /* reduce the stack usage */
16842 +       struct au_wr_dir_args wr_dir_args = {
16843 +               .force_btgt     = -1,
16844 +               .flags          = AuWrDir_ADD_ENTRY | AuWrDir_ISDIR
16845 +       };
16846 +
16847 +       IMustLock(dir);
16848 +
16849 +       err = -ENOMEM;
16850 +       a = kmalloc(sizeof(*a), GFP_NOFS);
16851 +       if (unlikely(!a))
16852 +               goto out;
16853 +
16854 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
16855 +       if (unlikely(err))
16856 +               goto out_free;
16857 +       err = au_d_may_add(dentry);
16858 +       if (unlikely(err))
16859 +               goto out_unlock;
16860 +
16861 +       parent = dentry->d_parent; /* dir inode is locked */
16862 +       di_write_lock_parent(parent);
16863 +       wh_dentry = lock_hdir_lkup_wh(dentry, &a->dt, /*src_dentry*/NULL,
16864 +                                     &a->pin, &wr_dir_args);
16865 +       err = PTR_ERR(wh_dentry);
16866 +       if (IS_ERR(wh_dentry))
16867 +               goto out_parent;
16868 +
16869 +       sb = dentry->d_sb;
16870 +       bindex = au_dbstart(dentry);
16871 +       h_path.dentry = au_h_dptr(dentry, bindex);
16872 +       h_path.mnt = au_sbr_mnt(sb, bindex);
16873 +       err = vfsub_mkdir(au_pinned_h_dir(&a->pin), &h_path, mode);
16874 +       if (unlikely(err))
16875 +               goto out_unpin;
16876 +
16877 +       /* make the dir opaque */
16878 +       diropq = 0;
16879 +       h_mtx = &h_path.dentry->d_inode->i_mutex;
16880 +       if (wh_dentry
16881 +           || au_opt_test(au_mntflags(sb), ALWAYS_DIROPQ)) {
16882 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
16883 +               opq_dentry = au_diropq_create(dentry, bindex);
16884 +               mutex_unlock(h_mtx);
16885 +               err = PTR_ERR(opq_dentry);
16886 +               if (IS_ERR(opq_dentry))
16887 +                       goto out_dir;
16888 +               dput(opq_dentry);
16889 +               diropq = 1;
16890 +       }
16891 +
16892 +       err = epilog(dir, bindex, wh_dentry, dentry);
16893 +       if (!err) {
16894 +               inc_nlink(dir);
16895 +               goto out_unpin; /* success */
16896 +       }
16897 +
16898 +       /* revert */
16899 +       if (diropq) {
16900 +               AuLabel(revert opq);
16901 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
16902 +               rerr = au_diropq_remove(dentry, bindex);
16903 +               mutex_unlock(h_mtx);
16904 +               if (rerr) {
16905 +                       AuIOErr("%pd reverting diropq failed(%d, %d)\n",
16906 +                               dentry, err, rerr);
16907 +                       err = -EIO;
16908 +               }
16909 +       }
16910 +
16911 +out_dir:
16912 +       AuLabel(revert dir);
16913 +       rerr = vfsub_rmdir(au_pinned_h_dir(&a->pin), &h_path);
16914 +       if (rerr) {
16915 +               AuIOErr("%pd reverting dir failed(%d, %d)\n",
16916 +                       dentry, err, rerr);
16917 +               err = -EIO;
16918 +       }
16919 +       au_dtime_revert(&a->dt);
16920 +out_unpin:
16921 +       au_unpin(&a->pin);
16922 +       dput(wh_dentry);
16923 +out_parent:
16924 +       di_write_unlock(parent);
16925 +out_unlock:
16926 +       if (unlikely(err)) {
16927 +               au_update_dbstart(dentry);
16928 +               d_drop(dentry);
16929 +       }
16930 +       aufs_read_unlock(dentry, AuLock_DW);
16931 +out_free:
16932 +       kfree(a);
16933 +out:
16934 +       return err;
16935 +}
16936 diff -urN /usr/share/empty/fs/aufs/i_op.c linux/fs/aufs/i_op.c
16937 --- /usr/share/empty/fs/aufs/i_op.c     1970-01-01 01:00:00.000000000 +0100
16938 +++ linux/fs/aufs/i_op.c        2014-04-24 22:11:10.851935747 +0200
16939 @@ -0,0 +1,1127 @@
16940 +/*
16941 + * Copyright (C) 2005-2014 Junjiro R. Okajima
16942 + *
16943 + * This program, aufs is free software; you can redistribute it and/or modify
16944 + * it under the terms of the GNU General Public License as published by
16945 + * the Free Software Foundation; either version 2 of the License, or
16946 + * (at your option) any later version.
16947 + *
16948 + * This program is distributed in the hope that it will be useful,
16949 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
16950 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16951 + * GNU General Public License for more details.
16952 + *
16953 + * You should have received a copy of the GNU General Public License
16954 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
16955 + */
16956 +
16957 +/*
16958 + * inode operations (except add/del/rename)
16959 + */
16960 +
16961 +#include <linux/device_cgroup.h>
16962 +#include <linux/fs_stack.h>
16963 +#include <linux/mm.h>
16964 +#include <linux/namei.h>
16965 +#include <linux/security.h>
16966 +#include "aufs.h"
16967 +
16968 +static int h_permission(struct inode *h_inode, int mask,
16969 +                       struct vfsmount *h_mnt, int brperm)
16970 +{
16971 +       int err;
16972 +       const unsigned char write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
16973 +
16974 +       err = -EACCES;
16975 +       if ((write_mask && IS_IMMUTABLE(h_inode))
16976 +           || ((mask & MAY_EXEC)
16977 +               && S_ISREG(h_inode->i_mode)
16978 +               && ((h_mnt->mnt_flags & MNT_NOEXEC)
16979 +                   || !(h_inode->i_mode & S_IXUGO))))
16980 +               goto out;
16981 +
16982 +       /*
16983 +        * - skip the lower fs test in the case of write to ro branch.
16984 +        * - nfs dir permission write check is optimized, but a policy for
16985 +        *   link/rename requires a real check.
16986 +        */
16987 +       if ((write_mask && !au_br_writable(brperm))
16988 +           || (au_test_nfs(h_inode->i_sb) && S_ISDIR(h_inode->i_mode)
16989 +               && write_mask && !(mask & MAY_READ))
16990 +           || !h_inode->i_op->permission) {
16991 +               /* AuLabel(generic_permission); */
16992 +               err = generic_permission(h_inode, mask);
16993 +       } else {
16994 +               /* AuLabel(h_inode->permission); */
16995 +               err = h_inode->i_op->permission(h_inode, mask);
16996 +               AuTraceErr(err);
16997 +       }
16998 +
16999 +       if (!err)
17000 +               err = devcgroup_inode_permission(h_inode, mask);
17001 +       if (!err)
17002 +               err = security_inode_permission(h_inode, mask);
17003 +
17004 +#if 0
17005 +       if (!err) {
17006 +               /* todo: do we need to call ima_path_check()? */
17007 +               struct path h_path = {
17008 +                       .dentry =
17009 +                       .mnt    = h_mnt
17010 +               };
17011 +               err = ima_path_check(&h_path,
17012 +                                    mask & (MAY_READ | MAY_WRITE | MAY_EXEC),
17013 +                                    IMA_COUNT_LEAVE);
17014 +       }
17015 +#endif
17016 +
17017 +out:
17018 +       return err;
17019 +}
17020 +
17021 +static int aufs_permission(struct inode *inode, int mask)
17022 +{
17023 +       int err;
17024 +       aufs_bindex_t bindex, bend;
17025 +       const unsigned char isdir = !!S_ISDIR(inode->i_mode),
17026 +               write_mask = !!(mask & (MAY_WRITE | MAY_APPEND));
17027 +       struct inode *h_inode;
17028 +       struct super_block *sb;
17029 +       struct au_branch *br;
17030 +
17031 +       /* todo: support rcu-walk? */
17032 +       if (mask & MAY_NOT_BLOCK)
17033 +               return -ECHILD;
17034 +
17035 +       sb = inode->i_sb;
17036 +       si_read_lock(sb, AuLock_FLUSH);
17037 +       ii_read_lock_child(inode);
17038 +#if 0
17039 +       err = au_iigen_test(inode, au_sigen(sb));
17040 +       if (unlikely(err))
17041 +               goto out;
17042 +#endif
17043 +
17044 +       if (!isdir || write_mask) {
17045 +               err = au_busy_or_stale();
17046 +               h_inode = au_h_iptr(inode, au_ibstart(inode));
17047 +               if (unlikely(!h_inode
17048 +                            || (h_inode->i_mode & S_IFMT)
17049 +                            != (inode->i_mode & S_IFMT)))
17050 +                       goto out;
17051 +
17052 +               err = 0;
17053 +               bindex = au_ibstart(inode);
17054 +               br = au_sbr(sb, bindex);
17055 +               err = h_permission(h_inode, mask, au_br_mnt(br), br->br_perm);
17056 +               if (write_mask
17057 +                   && !err
17058 +                   && !special_file(h_inode->i_mode)) {
17059 +                       /* test whether the upper writable branch exists */
17060 +                       err = -EROFS;
17061 +                       for (; bindex >= 0; bindex--)
17062 +                               if (!au_br_rdonly(au_sbr(sb, bindex))) {
17063 +                                       err = 0;
17064 +                                       break;
17065 +                               }
17066 +               }
17067 +               goto out;
17068 +       }
17069 +
17070 +       /* non-write to dir */
17071 +       err = 0;
17072 +       bend = au_ibend(inode);
17073 +       for (bindex = au_ibstart(inode); !err && bindex <= bend; bindex++) {
17074 +               h_inode = au_h_iptr(inode, bindex);
17075 +               if (h_inode) {
17076 +                       err = au_busy_or_stale();
17077 +                       if (unlikely(!S_ISDIR(h_inode->i_mode)))
17078 +                               break;
17079 +
17080 +                       br = au_sbr(sb, bindex);
17081 +                       err = h_permission(h_inode, mask, au_br_mnt(br),
17082 +                                          br->br_perm);
17083 +               }
17084 +       }
17085 +
17086 +out:
17087 +       ii_read_unlock(inode);
17088 +       si_read_unlock(sb);
17089 +       return err;
17090 +}
17091 +
17092 +/* ---------------------------------------------------------------------- */
17093 +
17094 +static struct dentry *aufs_lookup(struct inode *dir, struct dentry *dentry,
17095 +                                 unsigned int flags)
17096 +{
17097 +       struct dentry *ret, *parent;
17098 +       struct inode *inode;
17099 +       struct super_block *sb;
17100 +       int err, npositive;
17101 +
17102 +       IMustLock(dir);
17103 +
17104 +       /* todo: support rcu-walk? */
17105 +       ret = ERR_PTR(-ECHILD);
17106 +       if (flags & LOOKUP_RCU)
17107 +               goto out;
17108 +
17109 +       ret = ERR_PTR(-ENAMETOOLONG);
17110 +       if (unlikely(dentry->d_name.len > AUFS_MAX_NAMELEN))
17111 +               goto out;
17112 +
17113 +       sb = dir->i_sb;
17114 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
17115 +       ret = ERR_PTR(err);
17116 +       if (unlikely(err))
17117 +               goto out;
17118 +
17119 +       err = au_di_init(dentry);
17120 +       ret = ERR_PTR(err);
17121 +       if (unlikely(err))
17122 +               goto out_si;
17123 +
17124 +       inode = NULL;
17125 +       npositive = 0; /* suppress a warning */
17126 +       parent = dentry->d_parent; /* dir inode is locked */
17127 +       di_read_lock_parent(parent, AuLock_IR);
17128 +       err = au_alive_dir(parent);
17129 +       if (!err)
17130 +               err = au_digen_test(parent, au_sigen(sb));
17131 +       if (!err) {
17132 +               npositive = au_lkup_dentry(dentry, au_dbstart(parent),
17133 +                                          /*type*/0);
17134 +               err = npositive;
17135 +       }
17136 +       di_read_unlock(parent, AuLock_IR);
17137 +       ret = ERR_PTR(err);
17138 +       if (unlikely(err < 0))
17139 +               goto out_unlock;
17140 +
17141 +       if (npositive) {
17142 +               inode = au_new_inode(dentry, /*must_new*/0);
17143 +               ret = (void *)inode;
17144 +       }
17145 +       if (IS_ERR(inode)) {
17146 +               inode = NULL;
17147 +               goto out_unlock;
17148 +       }
17149 +
17150 +       ret = d_splice_alias(inode, dentry);
17151 +#if 0
17152 +       if (unlikely(d_need_lookup(dentry))) {
17153 +               spin_lock(&dentry->d_lock);
17154 +               dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
17155 +               spin_unlock(&dentry->d_lock);
17156 +       } else
17157 +#endif
17158 +       if (unlikely(IS_ERR(ret) && inode)) {
17159 +               ii_write_unlock(inode);
17160 +               iput(inode);
17161 +               inode = NULL;
17162 +       }
17163 +
17164 +out_unlock:
17165 +       di_write_unlock(dentry);
17166 +       if (inode) {
17167 +               /* verbose coding for lock class name */
17168 +               if (unlikely(S_ISLNK(inode->i_mode)))
17169 +                       au_rw_class(&au_di(dentry)->di_rwsem,
17170 +                                   au_lc_key + AuLcSymlink_DIINFO);
17171 +               else if (unlikely(S_ISDIR(inode->i_mode)))
17172 +                       au_rw_class(&au_di(dentry)->di_rwsem,
17173 +                                   au_lc_key + AuLcDir_DIINFO);
17174 +               else /* likely */
17175 +                       au_rw_class(&au_di(dentry)->di_rwsem,
17176 +                                   au_lc_key + AuLcNonDir_DIINFO);
17177 +       }
17178 +out_si:
17179 +       si_read_unlock(sb);
17180 +out:
17181 +       return ret;
17182 +}
17183 +
17184 +/* ---------------------------------------------------------------------- */
17185 +
17186 +static int au_wr_dir_cpup(struct dentry *dentry, struct dentry *parent,
17187 +                         const unsigned char add_entry, aufs_bindex_t bcpup,
17188 +                         aufs_bindex_t bstart)
17189 +{
17190 +       int err;
17191 +       struct dentry *h_parent;
17192 +       struct inode *h_dir;
17193 +
17194 +       if (add_entry)
17195 +               IMustLock(parent->d_inode);
17196 +       else
17197 +               di_write_lock_parent(parent);
17198 +
17199 +       err = 0;
17200 +       if (!au_h_dptr(parent, bcpup)) {
17201 +               if (bstart > bcpup)
17202 +                       err = au_cpup_dirs(dentry, bcpup);
17203 +               else if (bstart < bcpup)
17204 +                       err = au_cpdown_dirs(dentry, bcpup);
17205 +               else
17206 +                       BUG();
17207 +       }
17208 +       if (!err && add_entry) {
17209 +               h_parent = au_h_dptr(parent, bcpup);
17210 +               h_dir = h_parent->d_inode;
17211 +               mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
17212 +               err = au_lkup_neg(dentry, bcpup,
17213 +                                 au_ftest_wrdir(add_entry, TMP_WHENTRY));
17214 +               /* todo: no unlock here */
17215 +               mutex_unlock(&h_dir->i_mutex);
17216 +
17217 +               AuDbg("bcpup %d\n", bcpup);
17218 +               if (!err) {
17219 +                       if (!dentry->d_inode)
17220 +                               au_set_h_dptr(dentry, bstart, NULL);
17221 +                       au_update_dbrange(dentry, /*do_put_zero*/0);
17222 +               }
17223 +       }
17224 +
17225 +       if (!add_entry)
17226 +               di_write_unlock(parent);
17227 +       if (!err)
17228 +               err = bcpup; /* success */
17229 +
17230 +       AuTraceErr(err);
17231 +       return err;
17232 +}
17233 +
17234 +/*
17235 + * decide the branch and the parent dir where we will create a new entry.
17236 + * returns new bindex or an error.
17237 + * copyup the parent dir if needed.
17238 + */
17239 +int au_wr_dir(struct dentry *dentry, struct dentry *src_dentry,
17240 +             struct au_wr_dir_args *args)
17241 +{
17242 +       int err;
17243 +       unsigned int flags;
17244 +       aufs_bindex_t bcpup, bstart, src_bstart;
17245 +       const unsigned char add_entry
17246 +               = au_ftest_wrdir(args->flags, ADD_ENTRY)
17247 +               | au_ftest_wrdir(args->flags, TMP_WHENTRY);
17248 +       struct super_block *sb;
17249 +       struct dentry *parent;
17250 +       struct au_sbinfo *sbinfo;
17251 +
17252 +       sb = dentry->d_sb;
17253 +       sbinfo = au_sbi(sb);
17254 +       parent = dget_parent(dentry);
17255 +       bstart = au_dbstart(dentry);
17256 +       bcpup = bstart;
17257 +       if (args->force_btgt < 0) {
17258 +               if (src_dentry) {
17259 +                       src_bstart = au_dbstart(src_dentry);
17260 +                       if (src_bstart < bstart)
17261 +                               bcpup = src_bstart;
17262 +               } else if (add_entry) {
17263 +                       flags = 0;
17264 +                       if (au_ftest_wrdir(args->flags, ISDIR))
17265 +                               au_fset_wbr(flags, DIR);
17266 +                       err = AuWbrCreate(sbinfo, dentry, flags);
17267 +                       bcpup = err;
17268 +               }
17269 +
17270 +               if (bcpup < 0 || au_test_ro(sb, bcpup, dentry->d_inode)) {
17271 +                       if (add_entry)
17272 +                               err = AuWbrCopyup(sbinfo, dentry);
17273 +                       else {
17274 +                               if (!IS_ROOT(dentry)) {
17275 +                                       di_read_lock_parent(parent, !AuLock_IR);
17276 +                                       err = AuWbrCopyup(sbinfo, dentry);
17277 +                                       di_read_unlock(parent, !AuLock_IR);
17278 +                               } else
17279 +                                       err = AuWbrCopyup(sbinfo, dentry);
17280 +                       }
17281 +                       bcpup = err;
17282 +                       if (unlikely(err < 0))
17283 +                               goto out;
17284 +               }
17285 +       } else {
17286 +               bcpup = args->force_btgt;
17287 +               AuDebugOn(au_test_ro(sb, bcpup, dentry->d_inode));
17288 +       }
17289 +
17290 +       AuDbg("bstart %d, bcpup %d\n", bstart, bcpup);
17291 +       err = bcpup;
17292 +       if (bcpup == bstart)
17293 +               goto out; /* success */
17294 +
17295 +       /* copyup the new parent into the branch we process */
17296 +       err = au_wr_dir_cpup(dentry, parent, add_entry, bcpup, bstart);
17297 +       if (err >= 0) {
17298 +               if (!dentry->d_inode) {
17299 +                       au_set_h_dptr(dentry, bstart, NULL);
17300 +                       au_set_dbstart(dentry, bcpup);
17301 +                       au_set_dbend(dentry, bcpup);
17302 +               }
17303 +               AuDebugOn(add_entry && !au_h_dptr(dentry, bcpup));
17304 +       }
17305 +
17306 +out:
17307 +       dput(parent);
17308 +       return err;
17309 +}
17310 +
17311 +/* ---------------------------------------------------------------------- */
17312 +
17313 +void au_pin_hdir_unlock(struct au_pin *p)
17314 +{
17315 +       if (p->hdir)
17316 +               au_hn_imtx_unlock(p->hdir);
17317 +}
17318 +
17319 +static int au_pin_hdir_lock(struct au_pin *p)
17320 +{
17321 +       int err;
17322 +
17323 +       err = 0;
17324 +       if (!p->hdir)
17325 +               goto out;
17326 +
17327 +       /* even if an error happens later, keep this lock */
17328 +       au_hn_imtx_lock_nested(p->hdir, p->lsc_hi);
17329 +
17330 +       err = -EBUSY;
17331 +       if (unlikely(p->hdir->hi_inode != p->h_parent->d_inode))
17332 +               goto out;
17333 +
17334 +       err = 0;
17335 +       if (p->h_dentry)
17336 +               err = au_h_verify(p->h_dentry, p->udba, p->hdir->hi_inode,
17337 +                                 p->h_parent, p->br);
17338 +
17339 +out:
17340 +       return err;
17341 +}
17342 +
17343 +int au_pin_hdir_relock(struct au_pin *p)
17344 +{
17345 +       int err, i;
17346 +       struct inode *h_i;
17347 +       struct dentry *h_d[] = {
17348 +               p->h_dentry,
17349 +               p->h_parent
17350 +       };
17351 +
17352 +       err = au_pin_hdir_lock(p);
17353 +       if (unlikely(err))
17354 +               goto out;
17355 +
17356 +       for (i = 0; !err && i < sizeof(h_d)/sizeof(*h_d); i++) {
17357 +               if (!h_d[i])
17358 +                       continue;
17359 +               h_i = h_d[i]->d_inode;
17360 +               if (h_i)
17361 +                       err = !h_i->i_nlink;
17362 +       }
17363 +
17364 +out:
17365 +       return err;
17366 +}
17367 +
17368 +void au_pin_hdir_set_owner(struct au_pin *p, struct task_struct *task)
17369 +{
17370 +#if defined(CONFIG_DEBUG_MUTEXES) || defined(CONFIG_SMP)
17371 +       p->hdir->hi_inode->i_mutex.owner = task;
17372 +#endif
17373 +}
17374 +
17375 +void au_pin_hdir_acquire_nest(struct au_pin *p)
17376 +{
17377 +       if (p->hdir) {
17378 +               mutex_acquire_nest(&p->hdir->hi_inode->i_mutex.dep_map,
17379 +                                  p->lsc_hi, 0, NULL, _RET_IP_);
17380 +               au_pin_hdir_set_owner(p, current);
17381 +       }
17382 +}
17383 +
17384 +void au_pin_hdir_release(struct au_pin *p)
17385 +{
17386 +       if (p->hdir) {
17387 +               au_pin_hdir_set_owner(p, p->task);
17388 +               mutex_release(&p->hdir->hi_inode->i_mutex.dep_map, 1, _RET_IP_);
17389 +       }
17390 +}
17391 +
17392 +struct dentry *au_pinned_h_parent(struct au_pin *pin)
17393 +{
17394 +       if (pin && pin->parent)
17395 +               return au_h_dptr(pin->parent, pin->bindex);
17396 +       return NULL;
17397 +}
17398 +
17399 +void au_unpin(struct au_pin *p)
17400 +{
17401 +       if (p->hdir)
17402 +               au_pin_hdir_unlock(p);
17403 +       if (p->h_mnt && au_ftest_pin(p->flags, MNT_WRITE))
17404 +               vfsub_mnt_drop_write(p->h_mnt);
17405 +       if (!p->hdir)
17406 +               return;
17407 +
17408 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
17409 +               di_read_unlock(p->parent, AuLock_IR);
17410 +       iput(p->hdir->hi_inode);
17411 +       dput(p->parent);
17412 +       p->parent = NULL;
17413 +       p->hdir = NULL;
17414 +       p->h_mnt = NULL;
17415 +       /* do not clear p->task */
17416 +}
17417 +
17418 +int au_do_pin(struct au_pin *p)
17419 +{
17420 +       int err;
17421 +       struct super_block *sb;
17422 +       struct inode *h_dir;
17423 +
17424 +       err = 0;
17425 +       sb = p->dentry->d_sb;
17426 +       p->br = au_sbr(sb, p->bindex);
17427 +       if (IS_ROOT(p->dentry)) {
17428 +               if (au_ftest_pin(p->flags, MNT_WRITE)) {
17429 +                       p->h_mnt = au_br_mnt(p->br);
17430 +                       err = vfsub_mnt_want_write(p->h_mnt);
17431 +                       if (unlikely(err)) {
17432 +                               au_fclr_pin(p->flags, MNT_WRITE);
17433 +                               goto out_err;
17434 +                       }
17435 +               }
17436 +               goto out;
17437 +       }
17438 +
17439 +       p->h_dentry = NULL;
17440 +       if (p->bindex <= au_dbend(p->dentry))
17441 +               p->h_dentry = au_h_dptr(p->dentry, p->bindex);
17442 +
17443 +       p->parent = dget_parent(p->dentry);
17444 +       if (!au_ftest_pin(p->flags, DI_LOCKED))
17445 +               di_read_lock(p->parent, AuLock_IR, p->lsc_di);
17446 +
17447 +       h_dir = NULL;
17448 +       p->h_parent = au_h_dptr(p->parent, p->bindex);
17449 +       p->hdir = au_hi(p->parent->d_inode, p->bindex);
17450 +       if (p->hdir)
17451 +               h_dir = p->hdir->hi_inode;
17452 +
17453 +       /*
17454 +        * udba case, or
17455 +        * if DI_LOCKED is not set, then p->parent may be different
17456 +        * and h_parent can be NULL.
17457 +        */
17458 +       if (unlikely(!p->hdir || !h_dir || !p->h_parent)) {
17459 +               err = -EBUSY;
17460 +               if (!au_ftest_pin(p->flags, DI_LOCKED))
17461 +                       di_read_unlock(p->parent, AuLock_IR);
17462 +               dput(p->parent);
17463 +               p->parent = NULL;
17464 +               goto out_err;
17465 +       }
17466 +
17467 +       if (au_ftest_pin(p->flags, MNT_WRITE)) {
17468 +               p->h_mnt = au_br_mnt(p->br);
17469 +               err = vfsub_mnt_want_write(p->h_mnt);
17470 +               if (unlikely(err)) {
17471 +                       au_fclr_pin(p->flags, MNT_WRITE);
17472 +                       if (!au_ftest_pin(p->flags, DI_LOCKED))
17473 +                               di_read_unlock(p->parent, AuLock_IR);
17474 +                       dput(p->parent);
17475 +                       p->parent = NULL;
17476 +                       goto out_err;
17477 +               }
17478 +       }
17479 +
17480 +       au_igrab(h_dir);
17481 +       err = au_pin_hdir_lock(p);
17482 +       if (!err)
17483 +               goto out; /* success */
17484 +
17485 +out_err:
17486 +       pr_err("err %d\n", err);
17487 +       err = au_busy_or_stale();
17488 +out:
17489 +       return err;
17490 +}
17491 +
17492 +void au_pin_init(struct au_pin *p, struct dentry *dentry,
17493 +                aufs_bindex_t bindex, int lsc_di, int lsc_hi,
17494 +                unsigned int udba, unsigned char flags)
17495 +{
17496 +       p->dentry = dentry;
17497 +       p->udba = udba;
17498 +       p->lsc_di = lsc_di;
17499 +       p->lsc_hi = lsc_hi;
17500 +       p->flags = flags;
17501 +       p->bindex = bindex;
17502 +
17503 +       p->parent = NULL;
17504 +       p->hdir = NULL;
17505 +       p->h_mnt = NULL;
17506 +
17507 +       p->h_dentry = NULL;
17508 +       p->h_parent = NULL;
17509 +       p->br = NULL;
17510 +       p->task = current;
17511 +}
17512 +
17513 +int au_pin(struct au_pin *pin, struct dentry *dentry, aufs_bindex_t bindex,
17514 +          unsigned int udba, unsigned char flags)
17515 +{
17516 +       au_pin_init(pin, dentry, bindex, AuLsc_DI_PARENT, AuLsc_I_PARENT2,
17517 +                   udba, flags);
17518 +       return au_do_pin(pin);
17519 +}
17520 +
17521 +/* ---------------------------------------------------------------------- */
17522 +
17523 +/*
17524 + * ->setattr() and ->getattr() are called in various cases.
17525 + * chmod, stat: dentry is revalidated.
17526 + * fchmod, fstat: file and dentry are not revalidated, additionally they may be
17527 + *               unhashed.
17528 + * for ->setattr(), ia->ia_file is passed from ftruncate only.
17529 + */
17530 +/* todo: consolidate with do_refresh() and simple_reval_dpath() */
17531 +static int au_reval_for_attr(struct dentry *dentry, unsigned int sigen)
17532 +{
17533 +       int err;
17534 +       struct inode *inode;
17535 +       struct dentry *parent;
17536 +
17537 +       err = 0;
17538 +       inode = dentry->d_inode;
17539 +       if (au_digen_test(dentry, sigen)) {
17540 +               parent = dget_parent(dentry);
17541 +               di_read_lock_parent(parent, AuLock_IR);
17542 +               err = au_refresh_dentry(dentry, parent);
17543 +               di_read_unlock(parent, AuLock_IR);
17544 +               dput(parent);
17545 +       }
17546 +
17547 +       AuTraceErr(err);
17548 +       return err;
17549 +}
17550 +
17551 +#define AuIcpup_DID_CPUP       1
17552 +#define au_ftest_icpup(flags, name)    ((flags) & AuIcpup_##name)
17553 +#define au_fset_icpup(flags, name) \
17554 +       do { (flags) |= AuIcpup_##name; } while (0)
17555 +#define au_fclr_icpup(flags, name) \
17556 +       do { (flags) &= ~AuIcpup_##name; } while (0)
17557 +
17558 +struct au_icpup_args {
17559 +       unsigned char flags;
17560 +       unsigned char pin_flags;
17561 +       aufs_bindex_t btgt;
17562 +       unsigned int udba;
17563 +       struct au_pin pin;
17564 +       struct path h_path;
17565 +       struct inode *h_inode;
17566 +};
17567 +
17568 +static int au_pin_and_icpup(struct dentry *dentry, struct iattr *ia,
17569 +                           struct au_icpup_args *a)
17570 +{
17571 +       int err;
17572 +       loff_t sz;
17573 +       aufs_bindex_t bstart, ibstart;
17574 +       struct dentry *hi_wh, *parent;
17575 +       struct inode *inode;
17576 +       struct au_wr_dir_args wr_dir_args = {
17577 +               .force_btgt     = -1,
17578 +               .flags          = 0
17579 +       };
17580 +
17581 +       bstart = au_dbstart(dentry);
17582 +       inode = dentry->d_inode;
17583 +       if (S_ISDIR(inode->i_mode))
17584 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
17585 +       /* plink or hi_wh() case */
17586 +       ibstart = au_ibstart(inode);
17587 +       if (bstart != ibstart && !au_test_ro(inode->i_sb, ibstart, inode))
17588 +               wr_dir_args.force_btgt = ibstart;
17589 +       err = au_wr_dir(dentry, /*src_dentry*/NULL, &wr_dir_args);
17590 +       if (unlikely(err < 0))
17591 +               goto out;
17592 +       a->btgt = err;
17593 +       if (err != bstart)
17594 +               au_fset_icpup(a->flags, DID_CPUP);
17595 +
17596 +       err = 0;
17597 +       a->pin_flags = AuPin_MNT_WRITE;
17598 +       parent = NULL;
17599 +       if (!IS_ROOT(dentry)) {
17600 +               au_fset_pin(a->pin_flags, DI_LOCKED);
17601 +               parent = dget_parent(dentry);
17602 +               di_write_lock_parent(parent);
17603 +       }
17604 +
17605 +       err = au_pin(&a->pin, dentry, a->btgt, a->udba, a->pin_flags);
17606 +       if (unlikely(err))
17607 +               goto out_parent;
17608 +
17609 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
17610 +       a->h_inode = a->h_path.dentry->d_inode;
17611 +       mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
17612 +       sz = -1;
17613 +       if ((ia->ia_valid & ATTR_SIZE) && ia->ia_size < i_size_read(a->h_inode))
17614 +               sz = ia->ia_size;
17615 +       mutex_unlock(&a->h_inode->i_mutex);
17616 +
17617 +       hi_wh = NULL;
17618 +       if (au_ftest_icpup(a->flags, DID_CPUP) && d_unlinked(dentry)) {
17619 +               hi_wh = au_hi_wh(inode, a->btgt);
17620 +               if (!hi_wh) {
17621 +                       struct au_cp_generic cpg = {
17622 +                               .dentry = dentry,
17623 +                               .bdst   = a->btgt,
17624 +                               .bsrc   = -1,
17625 +                               .len    = sz,
17626 +                               .pin    = &a->pin
17627 +                       };
17628 +                       err = au_sio_cpup_wh(&cpg, /*file*/NULL);
17629 +                       if (unlikely(err))
17630 +                               goto out_unlock;
17631 +                       hi_wh = au_hi_wh(inode, a->btgt);
17632 +                       /* todo: revalidate hi_wh? */
17633 +               }
17634 +       }
17635 +
17636 +       if (parent) {
17637 +               au_pin_set_parent_lflag(&a->pin, /*lflag*/0);
17638 +               di_downgrade_lock(parent, AuLock_IR);
17639 +               dput(parent);
17640 +               parent = NULL;
17641 +       }
17642 +       if (!au_ftest_icpup(a->flags, DID_CPUP))
17643 +               goto out; /* success */
17644 +
17645 +       if (!d_unhashed(dentry)) {
17646 +               struct au_cp_generic cpg = {
17647 +                       .dentry = dentry,
17648 +                       .bdst   = a->btgt,
17649 +                       .bsrc   = bstart,
17650 +                       .len    = sz,
17651 +                       .pin    = &a->pin,
17652 +                       .flags  = AuCpup_DTIME | AuCpup_HOPEN
17653 +               };
17654 +               err = au_sio_cpup_simple(&cpg);
17655 +               if (!err)
17656 +                       a->h_path.dentry = au_h_dptr(dentry, a->btgt);
17657 +       } else if (!hi_wh)
17658 +               a->h_path.dentry = au_h_dptr(dentry, a->btgt);
17659 +       else
17660 +               a->h_path.dentry = hi_wh; /* do not dget here */
17661 +
17662 +out_unlock:
17663 +       a->h_inode = a->h_path.dentry->d_inode;
17664 +       if (!err)
17665 +               goto out; /* success */
17666 +       au_unpin(&a->pin);
17667 +out_parent:
17668 +       if (parent) {
17669 +               di_write_unlock(parent);
17670 +               dput(parent);
17671 +       }
17672 +out:
17673 +       if (!err)
17674 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
17675 +       return err;
17676 +}
17677 +
17678 +static int aufs_setattr(struct dentry *dentry, struct iattr *ia)
17679 +{
17680 +       int err;
17681 +       struct inode *inode, *delegated;
17682 +       struct super_block *sb;
17683 +       struct file *file;
17684 +       struct au_icpup_args *a;
17685 +
17686 +       inode = dentry->d_inode;
17687 +       IMustLock(inode);
17688 +
17689 +       err = -ENOMEM;
17690 +       a = kzalloc(sizeof(*a), GFP_NOFS);
17691 +       if (unlikely(!a))
17692 +               goto out;
17693 +
17694 +       if (ia->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
17695 +               ia->ia_valid &= ~ATTR_MODE;
17696 +
17697 +       file = NULL;
17698 +       sb = dentry->d_sb;
17699 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
17700 +       if (unlikely(err))
17701 +               goto out_kfree;
17702 +
17703 +       if (ia->ia_valid & ATTR_FILE) {
17704 +               /* currently ftruncate(2) only */
17705 +               AuDebugOn(!S_ISREG(inode->i_mode));
17706 +               file = ia->ia_file;
17707 +               err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/1);
17708 +               if (unlikely(err))
17709 +                       goto out_si;
17710 +               ia->ia_file = au_hf_top(file);
17711 +               a->udba = AuOpt_UDBA_NONE;
17712 +       } else {
17713 +               /* fchmod() doesn't pass ia_file */
17714 +               a->udba = au_opt_udba(sb);
17715 +               di_write_lock_child(dentry);
17716 +               /* no d_unlinked(), to set UDBA_NONE for root */
17717 +               if (d_unhashed(dentry))
17718 +                       a->udba = AuOpt_UDBA_NONE;
17719 +               if (a->udba != AuOpt_UDBA_NONE) {
17720 +                       AuDebugOn(IS_ROOT(dentry));
17721 +                       err = au_reval_for_attr(dentry, au_sigen(sb));
17722 +                       if (unlikely(err))
17723 +                               goto out_dentry;
17724 +               }
17725 +       }
17726 +
17727 +       err = au_pin_and_icpup(dentry, ia, a);
17728 +       if (unlikely(err < 0))
17729 +               goto out_dentry;
17730 +       if (au_ftest_icpup(a->flags, DID_CPUP)) {
17731 +               ia->ia_file = NULL;
17732 +               ia->ia_valid &= ~ATTR_FILE;
17733 +       }
17734 +
17735 +       a->h_path.mnt = au_sbr_mnt(sb, a->btgt);
17736 +       if ((ia->ia_valid & (ATTR_MODE | ATTR_CTIME))
17737 +           == (ATTR_MODE | ATTR_CTIME)) {
17738 +               err = security_path_chmod(&a->h_path, ia->ia_mode);
17739 +               if (unlikely(err))
17740 +                       goto out_unlock;
17741 +       } else if ((ia->ia_valid & (ATTR_UID | ATTR_GID))
17742 +                  && (ia->ia_valid & ATTR_CTIME)) {
17743 +               err = security_path_chown(&a->h_path, ia->ia_uid, ia->ia_gid);
17744 +               if (unlikely(err))
17745 +                       goto out_unlock;
17746 +       }
17747 +
17748 +       if (ia->ia_valid & ATTR_SIZE) {
17749 +               struct file *f;
17750 +
17751 +               if (ia->ia_size < i_size_read(inode))
17752 +                       /* unmap only */
17753 +                       truncate_setsize(inode, ia->ia_size);
17754 +
17755 +               f = NULL;
17756 +               if (ia->ia_valid & ATTR_FILE)
17757 +                       f = ia->ia_file;
17758 +               mutex_unlock(&a->h_inode->i_mutex);
17759 +               err = vfsub_trunc(&a->h_path, ia->ia_size, ia->ia_valid, f);
17760 +               mutex_lock_nested(&a->h_inode->i_mutex, AuLsc_I_CHILD);
17761 +       } else {
17762 +               delegated = NULL;
17763 +               while (1) {
17764 +                       err = vfsub_notify_change(&a->h_path, ia, &delegated);
17765 +                       if (delegated) {
17766 +                               err = break_deleg_wait(&delegated);
17767 +                               if (!err)
17768 +                                       continue;
17769 +                       }
17770 +                       break;
17771 +               }
17772 +       }
17773 +       if (!err)
17774 +               au_cpup_attr_changeable(inode);
17775 +
17776 +out_unlock:
17777 +       mutex_unlock(&a->h_inode->i_mutex);
17778 +       au_unpin(&a->pin);
17779 +       if (unlikely(err))
17780 +               au_update_dbstart(dentry);
17781 +out_dentry:
17782 +       di_write_unlock(dentry);
17783 +       if (file) {
17784 +               fi_write_unlock(file);
17785 +               ia->ia_file = file;
17786 +               ia->ia_valid |= ATTR_FILE;
17787 +       }
17788 +out_si:
17789 +       si_read_unlock(sb);
17790 +out_kfree:
17791 +       kfree(a);
17792 +out:
17793 +       AuTraceErr(err);
17794 +       return err;
17795 +}
17796 +
17797 +static void au_refresh_iattr(struct inode *inode, struct kstat *st,
17798 +                            unsigned int nlink)
17799 +{
17800 +       unsigned int n;
17801 +
17802 +       inode->i_mode = st->mode;
17803 +       /* don't i_[ug]id_write() here */
17804 +       inode->i_uid = st->uid;
17805 +       inode->i_gid = st->gid;
17806 +       inode->i_atime = st->atime;
17807 +       inode->i_mtime = st->mtime;
17808 +       inode->i_ctime = st->ctime;
17809 +
17810 +       au_cpup_attr_nlink(inode, /*force*/0);
17811 +       if (S_ISDIR(inode->i_mode)) {
17812 +               n = inode->i_nlink;
17813 +               n -= nlink;
17814 +               n += st->nlink;
17815 +               smp_mb(); /* for i_nlink */
17816 +               /* 0 can happen */
17817 +               set_nlink(inode, n);
17818 +       }
17819 +
17820 +       spin_lock(&inode->i_lock);
17821 +       inode->i_blocks = st->blocks;
17822 +       i_size_write(inode, st->size);
17823 +       spin_unlock(&inode->i_lock);
17824 +}
17825 +
17826 +static int aufs_getattr(struct vfsmount *mnt __maybe_unused,
17827 +                       struct dentry *dentry, struct kstat *st)
17828 +{
17829 +       int err;
17830 +       unsigned int mnt_flags;
17831 +       aufs_bindex_t bindex;
17832 +       unsigned char udba_none, positive;
17833 +       struct super_block *sb, *h_sb;
17834 +       struct inode *inode;
17835 +       struct path h_path;
17836 +
17837 +       sb = dentry->d_sb;
17838 +       inode = dentry->d_inode;
17839 +       err = si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
17840 +       if (unlikely(err))
17841 +               goto out;
17842 +       mnt_flags = au_mntflags(sb);
17843 +       udba_none = !!au_opt_test(mnt_flags, UDBA_NONE);
17844 +
17845 +       /* support fstat(2) */
17846 +       if (!d_unlinked(dentry) && !udba_none) {
17847 +               unsigned int sigen = au_sigen(sb);
17848 +               err = au_digen_test(dentry, sigen);
17849 +               if (!err) {
17850 +                       di_read_lock_child(dentry, AuLock_IR);
17851 +                       err = au_dbrange_test(dentry);
17852 +                       if (unlikely(err))
17853 +                               goto out_unlock;
17854 +               } else {
17855 +                       AuDebugOn(IS_ROOT(dentry));
17856 +                       di_write_lock_child(dentry);
17857 +                       err = au_dbrange_test(dentry);
17858 +                       if (!err)
17859 +                               err = au_reval_for_attr(dentry, sigen);
17860 +                       di_downgrade_lock(dentry, AuLock_IR);
17861 +                       if (unlikely(err))
17862 +                               goto out_unlock;
17863 +               }
17864 +       } else
17865 +               di_read_lock_child(dentry, AuLock_IR);
17866 +
17867 +       bindex = au_ibstart(inode);
17868 +       h_path.mnt = au_sbr_mnt(sb, bindex);
17869 +       h_sb = h_path.mnt->mnt_sb;
17870 +       if (!au_test_fs_bad_iattr(h_sb) && udba_none)
17871 +               goto out_fill; /* success */
17872 +
17873 +       h_path.dentry = NULL;
17874 +       if (au_dbstart(dentry) == bindex)
17875 +               h_path.dentry = dget(au_h_dptr(dentry, bindex));
17876 +       else if (au_opt_test(mnt_flags, PLINK) && au_plink_test(inode)) {
17877 +               h_path.dentry = au_plink_lkup(inode, bindex);
17878 +               if (IS_ERR(h_path.dentry))
17879 +                       goto out_fill; /* pretending success */
17880 +       }
17881 +       /* illegally overlapped or something */
17882 +       if (unlikely(!h_path.dentry))
17883 +               goto out_fill; /* pretending success */
17884 +
17885 +       positive = !!h_path.dentry->d_inode;
17886 +       if (positive)
17887 +               err = vfs_getattr(&h_path, st);
17888 +       dput(h_path.dentry);
17889 +       if (!err) {
17890 +               if (positive)
17891 +                       au_refresh_iattr(inode, st,
17892 +                                        h_path.dentry->d_inode->i_nlink);
17893 +               goto out_fill; /* success */
17894 +       }
17895 +       AuTraceErr(err);
17896 +       goto out_unlock;
17897 +
17898 +out_fill:
17899 +       generic_fillattr(inode, st);
17900 +out_unlock:
17901 +       di_read_unlock(dentry, AuLock_IR);
17902 +       si_read_unlock(sb);
17903 +out:
17904 +       AuTraceErr(err);
17905 +       return err;
17906 +}
17907 +
17908 +/* ---------------------------------------------------------------------- */
17909 +
17910 +static int h_readlink(struct dentry *dentry, int bindex, char __user *buf,
17911 +                     int bufsiz)
17912 +{
17913 +       int err;
17914 +       struct super_block *sb;
17915 +       struct dentry *h_dentry;
17916 +
17917 +       err = -EINVAL;
17918 +       h_dentry = au_h_dptr(dentry, bindex);
17919 +       if (unlikely(!h_dentry->d_inode->i_op->readlink))
17920 +               goto out;
17921 +
17922 +       err = security_inode_readlink(h_dentry);
17923 +       if (unlikely(err))
17924 +               goto out;
17925 +
17926 +       sb = dentry->d_sb;
17927 +       if (!au_test_ro(sb, bindex, dentry->d_inode)) {
17928 +               vfsub_touch_atime(au_sbr_mnt(sb, bindex), h_dentry);
17929 +               fsstack_copy_attr_atime(dentry->d_inode, h_dentry->d_inode);
17930 +       }
17931 +       err = h_dentry->d_inode->i_op->readlink(h_dentry, buf, bufsiz);
17932 +
17933 +out:
17934 +       return err;
17935 +}
17936 +
17937 +static int aufs_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
17938 +{
17939 +       int err;
17940 +
17941 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
17942 +       if (unlikely(err))
17943 +               goto out;
17944 +       err = au_d_hashed_positive(dentry);
17945 +       if (!err)
17946 +               err = h_readlink(dentry, au_dbstart(dentry), buf, bufsiz);
17947 +       aufs_read_unlock(dentry, AuLock_IR);
17948 +
17949 +out:
17950 +       return err;
17951 +}
17952 +
17953 +static void *aufs_follow_link(struct dentry *dentry, struct nameidata *nd)
17954 +{
17955 +       int err;
17956 +       mm_segment_t old_fs;
17957 +       union {
17958 +               char *k;
17959 +               char __user *u;
17960 +       } buf;
17961 +
17962 +       err = -ENOMEM;
17963 +       buf.k = (void *)__get_free_page(GFP_NOFS);
17964 +       if (unlikely(!buf.k))
17965 +               goto out;
17966 +
17967 +       err = aufs_read_lock(dentry, AuLock_IR | AuLock_GEN);
17968 +       if (unlikely(err))
17969 +               goto out_name;
17970 +
17971 +       err = au_d_hashed_positive(dentry);
17972 +       if (!err) {
17973 +               old_fs = get_fs();
17974 +               set_fs(KERNEL_DS);
17975 +               err = h_readlink(dentry, au_dbstart(dentry), buf.u, PATH_MAX);
17976 +               set_fs(old_fs);
17977 +       }
17978 +       aufs_read_unlock(dentry, AuLock_IR);
17979 +
17980 +       if (err >= 0) {
17981 +               buf.k[err] = 0;
17982 +               /* will be freed by put_link */
17983 +               nd_set_link(nd, buf.k);
17984 +               return NULL; /* success */
17985 +       }
17986 +
17987 +out_name:
17988 +       free_page((unsigned long)buf.k);
17989 +out:
17990 +       AuTraceErr(err);
17991 +       return ERR_PTR(err);
17992 +}
17993 +
17994 +static void aufs_put_link(struct dentry *dentry __maybe_unused,
17995 +                         struct nameidata *nd, void *cookie __maybe_unused)
17996 +{
17997 +       char *p;
17998 +
17999 +       p = nd_get_link(nd);
18000 +       if (!IS_ERR_OR_NULL(p))
18001 +               free_page((unsigned long)p);
18002 +}
18003 +
18004 +/* ---------------------------------------------------------------------- */
18005 +
18006 +static int aufs_update_time(struct inode *inode, struct timespec *ts, int flags)
18007 +{
18008 +       int err;
18009 +       struct super_block *sb;
18010 +       struct inode *h_inode;
18011 +
18012 +       sb = inode->i_sb;
18013 +       /* mmap_sem might be acquired already, cf. aufs_mmap() */
18014 +       lockdep_off();
18015 +       si_read_lock(sb, AuLock_FLUSH);
18016 +       ii_write_lock_child(inode);
18017 +       lockdep_on();
18018 +       h_inode = au_h_iptr(inode, au_ibstart(inode));
18019 +       err = vfsub_update_time(h_inode, ts, flags);
18020 +       lockdep_off();
18021 +       ii_write_unlock(inode);
18022 +       si_read_unlock(sb);
18023 +       lockdep_on();
18024 +       return err;
18025 +}
18026 +
18027 +/* ---------------------------------------------------------------------- */
18028 +
18029 +struct inode_operations aufs_symlink_iop = {
18030 +       .permission     = aufs_permission,
18031 +       .setattr        = aufs_setattr,
18032 +       .getattr        = aufs_getattr,
18033 +
18034 +       .readlink       = aufs_readlink,
18035 +       .follow_link    = aufs_follow_link,
18036 +       .put_link       = aufs_put_link,
18037 +
18038 +       /* .update_time = aufs_update_time */
18039 +};
18040 +
18041 +struct inode_operations aufs_dir_iop = {
18042 +       .create         = aufs_create,
18043 +       .lookup         = aufs_lookup,
18044 +       .link           = aufs_link,
18045 +       .unlink         = aufs_unlink,
18046 +       .symlink        = aufs_symlink,
18047 +       .mkdir          = aufs_mkdir,
18048 +       .rmdir          = aufs_rmdir,
18049 +       .mknod          = aufs_mknod,
18050 +       .rename         = aufs_rename,
18051 +
18052 +       .permission     = aufs_permission,
18053 +       .setattr        = aufs_setattr,
18054 +       .getattr        = aufs_getattr,
18055 +
18056 +       .update_time    = aufs_update_time
18057 +       /* no support for atomic_open() */
18058 +};
18059 +
18060 +struct inode_operations aufs_iop = {
18061 +       .permission     = aufs_permission,
18062 +       .setattr        = aufs_setattr,
18063 +       .getattr        = aufs_getattr,
18064 +
18065 +       .update_time    = aufs_update_time
18066 +};
18067 diff -urN /usr/share/empty/fs/aufs/i_op_del.c linux/fs/aufs/i_op_del.c
18068 --- /usr/share/empty/fs/aufs/i_op_del.c 1970-01-01 01:00:00.000000000 +0100
18069 +++ linux/fs/aufs/i_op_del.c    2014-04-24 22:11:10.855269116 +0200
18070 @@ -0,0 +1,506 @@
18071 +/*
18072 + * Copyright (C) 2005-2014 Junjiro R. Okajima
18073 + *
18074 + * This program, aufs is free software; you can redistribute it and/or modify
18075 + * it under the terms of the GNU General Public License as published by
18076 + * the Free Software Foundation; either version 2 of the License, or
18077 + * (at your option) any later version.
18078 + *
18079 + * This program is distributed in the hope that it will be useful,
18080 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18081 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18082 + * GNU General Public License for more details.
18083 + *
18084 + * You should have received a copy of the GNU General Public License
18085 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18086 + */
18087 +
18088 +/*
18089 + * inode operations (del entry)
18090 + */
18091 +
18092 +#include "aufs.h"
18093 +
18094 +/*
18095 + * decide if a new whiteout for @dentry is necessary or not.
18096 + * when it is necessary, prepare the parent dir for the upper branch whose
18097 + * branch index is @bcpup for creation. the actual creation of the whiteout will
18098 + * be done by caller.
18099 + * return value:
18100 + * 0: wh is unnecessary
18101 + * plus: wh is necessary
18102 + * minus: error
18103 + */
18104 +int au_wr_dir_need_wh(struct dentry *dentry, int isdir, aufs_bindex_t *bcpup)
18105 +{
18106 +       int need_wh, err;
18107 +       aufs_bindex_t bstart;
18108 +       struct super_block *sb;
18109 +
18110 +       sb = dentry->d_sb;
18111 +       bstart = au_dbstart(dentry);
18112 +       if (*bcpup < 0) {
18113 +               *bcpup = bstart;
18114 +               if (au_test_ro(sb, bstart, dentry->d_inode)) {
18115 +                       err = AuWbrCopyup(au_sbi(sb), dentry);
18116 +                       *bcpup = err;
18117 +                       if (unlikely(err < 0))
18118 +                               goto out;
18119 +               }
18120 +       } else
18121 +               AuDebugOn(bstart < *bcpup
18122 +                         || au_test_ro(sb, *bcpup, dentry->d_inode));
18123 +       AuDbg("bcpup %d, bstart %d\n", *bcpup, bstart);
18124 +
18125 +       if (*bcpup != bstart) {
18126 +               err = au_cpup_dirs(dentry, *bcpup);
18127 +               if (unlikely(err))
18128 +                       goto out;
18129 +               need_wh = 1;
18130 +       } else {
18131 +               struct au_dinfo *dinfo, *tmp;
18132 +
18133 +               need_wh = -ENOMEM;
18134 +               dinfo = au_di(dentry);
18135 +               tmp = au_di_alloc(sb, AuLsc_DI_TMP);
18136 +               if (tmp) {
18137 +                       au_di_cp(tmp, dinfo);
18138 +                       au_di_swap(tmp, dinfo);
18139 +                       /* returns the number of positive dentries */
18140 +                       need_wh = au_lkup_dentry(dentry, bstart + 1, /*type*/0);
18141 +                       au_di_swap(tmp, dinfo);
18142 +                       au_rw_write_unlock(&tmp->di_rwsem);
18143 +                       au_di_free(tmp);
18144 +               }
18145 +       }
18146 +       AuDbg("need_wh %d\n", need_wh);
18147 +       err = need_wh;
18148 +
18149 +out:
18150 +       return err;
18151 +}
18152 +
18153 +/*
18154 + * simple tests for the del-entry operations.
18155 + * following the checks in vfs, plus the parent-child relationship.
18156 + */
18157 +int au_may_del(struct dentry *dentry, aufs_bindex_t bindex,
18158 +              struct dentry *h_parent, int isdir)
18159 +{
18160 +       int err;
18161 +       umode_t h_mode;
18162 +       struct dentry *h_dentry, *h_latest;
18163 +       struct inode *h_inode;
18164 +
18165 +       h_dentry = au_h_dptr(dentry, bindex);
18166 +       h_inode = h_dentry->d_inode;
18167 +       if (dentry->d_inode) {
18168 +               err = -ENOENT;
18169 +               if (unlikely(!h_inode || !h_inode->i_nlink))
18170 +                       goto out;
18171 +
18172 +               h_mode = h_inode->i_mode;
18173 +               if (!isdir) {
18174 +                       err = -EISDIR;
18175 +                       if (unlikely(S_ISDIR(h_mode)))
18176 +                               goto out;
18177 +               } else if (unlikely(!S_ISDIR(h_mode))) {
18178 +                       err = -ENOTDIR;
18179 +                       goto out;
18180 +               }
18181 +       } else {
18182 +               /* rename(2) case */
18183 +               err = -EIO;
18184 +               if (unlikely(h_inode))
18185 +                       goto out;
18186 +       }
18187 +
18188 +       err = -ENOENT;
18189 +       /* expected parent dir is locked */
18190 +       if (unlikely(h_parent != h_dentry->d_parent))
18191 +               goto out;
18192 +       err = 0;
18193 +
18194 +       /*
18195 +        * rmdir a dir may break the consistency on some filesystem.
18196 +        * let's try heavy test.
18197 +        */
18198 +       err = -EACCES;
18199 +       if (unlikely(au_test_h_perm(h_parent->d_inode, MAY_EXEC | MAY_WRITE)))
18200 +               goto out;
18201 +
18202 +       h_latest = au_sio_lkup_one(&dentry->d_name, h_parent,
18203 +                                  au_sbr(dentry->d_sb, bindex));
18204 +       err = -EIO;
18205 +       if (IS_ERR(h_latest))
18206 +               goto out;
18207 +       if (h_latest == h_dentry)
18208 +               err = 0;
18209 +       dput(h_latest);
18210 +
18211 +out:
18212 +       return err;
18213 +}
18214 +
18215 +/*
18216 + * decide the branch where we operate for @dentry. the branch index will be set
18217 + * @rbcpup. after diciding it, 'pin' it and store the timestamps of the parent
18218 + * dir for reverting.
18219 + * when a new whiteout is necessary, create it.
18220 + */
18221 +static struct dentry*
18222 +lock_hdir_create_wh(struct dentry *dentry, int isdir, aufs_bindex_t *rbcpup,
18223 +                   struct au_dtime *dt, struct au_pin *pin)
18224 +{
18225 +       struct dentry *wh_dentry;
18226 +       struct super_block *sb;
18227 +       struct path h_path;
18228 +       int err, need_wh;
18229 +       unsigned int udba;
18230 +       aufs_bindex_t bcpup;
18231 +
18232 +       need_wh = au_wr_dir_need_wh(dentry, isdir, rbcpup);
18233 +       wh_dentry = ERR_PTR(need_wh);
18234 +       if (unlikely(need_wh < 0))
18235 +               goto out;
18236 +
18237 +       sb = dentry->d_sb;
18238 +       udba = au_opt_udba(sb);
18239 +       bcpup = *rbcpup;
18240 +       err = au_pin(pin, dentry, bcpup, udba,
18241 +                    AuPin_DI_LOCKED | AuPin_MNT_WRITE);
18242 +       wh_dentry = ERR_PTR(err);
18243 +       if (unlikely(err))
18244 +               goto out;
18245 +
18246 +       h_path.dentry = au_pinned_h_parent(pin);
18247 +       if (udba != AuOpt_UDBA_NONE
18248 +           && au_dbstart(dentry) == bcpup) {
18249 +               err = au_may_del(dentry, bcpup, h_path.dentry, isdir);
18250 +               wh_dentry = ERR_PTR(err);
18251 +               if (unlikely(err))
18252 +                       goto out_unpin;
18253 +       }
18254 +
18255 +       h_path.mnt = au_sbr_mnt(sb, bcpup);
18256 +       au_dtime_store(dt, au_pinned_parent(pin), &h_path);
18257 +       wh_dentry = NULL;
18258 +       if (!need_wh)
18259 +               goto out; /* success, no need to create whiteout */
18260 +
18261 +       wh_dentry = au_wh_create(dentry, bcpup, h_path.dentry);
18262 +       if (IS_ERR(wh_dentry))
18263 +               goto out_unpin;
18264 +
18265 +       /* returns with the parent is locked and wh_dentry is dget-ed */
18266 +       goto out; /* success */
18267 +
18268 +out_unpin:
18269 +       au_unpin(pin);
18270 +out:
18271 +       return wh_dentry;
18272 +}
18273 +
18274 +/*
18275 + * when removing a dir, rename it to a unique temporary whiteout-ed name first
18276 + * in order to be revertible and save time for removing many child whiteouts
18277 + * under the dir.
18278 + * returns 1 when there are too many child whiteout and caller should remove
18279 + * them asynchronously. returns 0 when the number of children is enough small to
18280 + * remove now or the branch fs is a remote fs.
18281 + * otherwise return an error.
18282 + */
18283 +static int renwh_and_rmdir(struct dentry *dentry, aufs_bindex_t bindex,
18284 +                          struct au_nhash *whlist, struct inode *dir)
18285 +{
18286 +       int rmdir_later, err, dirwh;
18287 +       struct dentry *h_dentry;
18288 +       struct super_block *sb;
18289 +
18290 +       sb = dentry->d_sb;
18291 +       SiMustAnyLock(sb);
18292 +       h_dentry = au_h_dptr(dentry, bindex);
18293 +       err = au_whtmp_ren(h_dentry, au_sbr(sb, bindex));
18294 +       if (unlikely(err))
18295 +               goto out;
18296 +
18297 +       /* stop monitoring */
18298 +       au_hn_free(au_hi(dentry->d_inode, bindex));
18299 +
18300 +       if (!au_test_fs_remote(h_dentry->d_sb)) {
18301 +               dirwh = au_sbi(sb)->si_dirwh;
18302 +               rmdir_later = (dirwh <= 1);
18303 +               if (!rmdir_later)
18304 +                       rmdir_later = au_nhash_test_longer_wh(whlist, bindex,
18305 +                                                             dirwh);
18306 +               if (rmdir_later)
18307 +                       return rmdir_later;
18308 +       }
18309 +
18310 +       err = au_whtmp_rmdir(dir, bindex, h_dentry, whlist);
18311 +       if (unlikely(err)) {
18312 +               AuIOErr("rmdir %pd, b%d failed, %d. ignored\n",
18313 +                       h_dentry, bindex, err);
18314 +               err = 0;
18315 +       }
18316 +
18317 +out:
18318 +       AuTraceErr(err);
18319 +       return err;
18320 +}
18321 +
18322 +/*
18323 + * final procedure for deleting a entry.
18324 + * maintain dentry and iattr.
18325 + */
18326 +static void epilog(struct inode *dir, struct dentry *dentry,
18327 +                  aufs_bindex_t bindex)
18328 +{
18329 +       struct inode *inode;
18330 +
18331 +       inode = dentry->d_inode;
18332 +       d_drop(dentry);
18333 +       inode->i_ctime = dir->i_ctime;
18334 +
18335 +       if (au_ibstart(dir) == bindex)
18336 +               au_cpup_attr_timesizes(dir);
18337 +       dir->i_version++;
18338 +}
18339 +
18340 +/*
18341 + * when an error happened, remove the created whiteout and revert everything.
18342 + */
18343 +static int do_revert(int err, struct inode *dir, aufs_bindex_t bindex,
18344 +                    aufs_bindex_t bwh, struct dentry *wh_dentry,
18345 +                    struct dentry *dentry, struct au_dtime *dt)
18346 +{
18347 +       int rerr;
18348 +       struct path h_path = {
18349 +               .dentry = wh_dentry,
18350 +               .mnt    = au_sbr_mnt(dir->i_sb, bindex)
18351 +       };
18352 +
18353 +       rerr = au_wh_unlink_dentry(au_h_iptr(dir, bindex), &h_path, dentry);
18354 +       if (!rerr) {
18355 +               au_set_dbwh(dentry, bwh);
18356 +               au_dtime_revert(dt);
18357 +               return 0;
18358 +       }
18359 +
18360 +       AuIOErr("%pd reverting whiteout failed(%d, %d)\n", dentry, err, rerr);
18361 +       return -EIO;
18362 +}
18363 +
18364 +/* ---------------------------------------------------------------------- */
18365 +
18366 +int aufs_unlink(struct inode *dir, struct dentry *dentry)
18367 +{
18368 +       int err;
18369 +       aufs_bindex_t bwh, bindex, bstart;
18370 +       struct inode *inode, *h_dir, *delegated;
18371 +       struct dentry *parent, *wh_dentry;
18372 +       /* to reuduce stack size */
18373 +       struct {
18374 +               struct au_dtime dt;
18375 +               struct au_pin pin;
18376 +               struct path h_path;
18377 +       } *a;
18378 +
18379 +       IMustLock(dir);
18380 +
18381 +       err = -ENOMEM;
18382 +       a = kmalloc(sizeof(*a), GFP_NOFS);
18383 +       if (unlikely(!a))
18384 +               goto out;
18385 +
18386 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_GEN);
18387 +       if (unlikely(err))
18388 +               goto out_free;
18389 +       err = au_d_hashed_positive(dentry);
18390 +       if (unlikely(err))
18391 +               goto out_unlock;
18392 +       inode = dentry->d_inode;
18393 +       IMustLock(inode);
18394 +       err = -EISDIR;
18395 +       if (unlikely(S_ISDIR(inode->i_mode)))
18396 +               goto out_unlock; /* possible? */
18397 +
18398 +       bstart = au_dbstart(dentry);
18399 +       bwh = au_dbwh(dentry);
18400 +       bindex = -1;
18401 +       parent = dentry->d_parent; /* dir inode is locked */
18402 +       di_write_lock_parent(parent);
18403 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/0, &bindex, &a->dt,
18404 +                                       &a->pin);
18405 +       err = PTR_ERR(wh_dentry);
18406 +       if (IS_ERR(wh_dentry))
18407 +               goto out_parent;
18408 +
18409 +       a->h_path.mnt = au_sbr_mnt(dentry->d_sb, bstart);
18410 +       a->h_path.dentry = au_h_dptr(dentry, bstart);
18411 +       dget(a->h_path.dentry);
18412 +       if (bindex == bstart) {
18413 +               h_dir = au_pinned_h_dir(&a->pin);
18414 +               delegated = NULL;
18415 +               err = vfsub_unlink(h_dir, &a->h_path, &delegated, /*force*/0);
18416 +               if (unlikely(err == -EWOULDBLOCK)) {
18417 +                       pr_warn("cannot retry for NFSv4 delegation"
18418 +                               " for an internal unlink\n");
18419 +                       iput(delegated);
18420 +               }
18421 +       } else {
18422 +               /* dir inode is locked */
18423 +               h_dir = wh_dentry->d_parent->d_inode;
18424 +               IMustLock(h_dir);
18425 +               err = 0;
18426 +       }
18427 +
18428 +       if (!err) {
18429 +               vfsub_drop_nlink(inode);
18430 +               epilog(dir, dentry, bindex);
18431 +
18432 +               /* update target timestamps */
18433 +               if (bindex == bstart) {
18434 +                       vfsub_update_h_iattr(&a->h_path, /*did*/NULL);
18435 +                       /*ignore*/
18436 +                       inode->i_ctime = a->h_path.dentry->d_inode->i_ctime;
18437 +               } else
18438 +                       /* todo: this timestamp may be reverted later */
18439 +                       inode->i_ctime = h_dir->i_ctime;
18440 +               goto out_unpin; /* success */
18441 +       }
18442 +
18443 +       /* revert */
18444 +       if (wh_dentry) {
18445 +               int rerr;
18446 +
18447 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
18448 +                                &a->dt);
18449 +               if (rerr)
18450 +                       err = rerr;
18451 +       }
18452 +
18453 +out_unpin:
18454 +       au_unpin(&a->pin);
18455 +       dput(wh_dentry);
18456 +       dput(a->h_path.dentry);
18457 +out_parent:
18458 +       di_write_unlock(parent);
18459 +out_unlock:
18460 +       aufs_read_unlock(dentry, AuLock_DW);
18461 +out_free:
18462 +       kfree(a);
18463 +out:
18464 +       return err;
18465 +}
18466 +
18467 +int aufs_rmdir(struct inode *dir, struct dentry *dentry)
18468 +{
18469 +       int err, rmdir_later;
18470 +       aufs_bindex_t bwh, bindex, bstart;
18471 +       struct inode *inode;
18472 +       struct dentry *parent, *wh_dentry, *h_dentry;
18473 +       struct au_whtmp_rmdir *args;
18474 +       /* to reuduce stack size */
18475 +       struct {
18476 +               struct au_dtime dt;
18477 +               struct au_pin pin;
18478 +       } *a;
18479 +
18480 +       IMustLock(dir);
18481 +
18482 +       err = -ENOMEM;
18483 +       a = kmalloc(sizeof(*a), GFP_NOFS);
18484 +       if (unlikely(!a))
18485 +               goto out;
18486 +
18487 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH | AuLock_GEN);
18488 +       if (unlikely(err))
18489 +               goto out_free;
18490 +       err = au_alive_dir(dentry);
18491 +       if (unlikely(err))
18492 +               goto out_unlock;
18493 +       inode = dentry->d_inode;
18494 +       IMustLock(inode);
18495 +       err = -ENOTDIR;
18496 +       if (unlikely(!S_ISDIR(inode->i_mode)))
18497 +               goto out_unlock; /* possible? */
18498 +
18499 +       err = -ENOMEM;
18500 +       args = au_whtmp_rmdir_alloc(dir->i_sb, GFP_NOFS);
18501 +       if (unlikely(!args))
18502 +               goto out_unlock;
18503 +
18504 +       parent = dentry->d_parent; /* dir inode is locked */
18505 +       di_write_lock_parent(parent);
18506 +       err = au_test_empty(dentry, &args->whlist);
18507 +       if (unlikely(err))
18508 +               goto out_parent;
18509 +
18510 +       bstart = au_dbstart(dentry);
18511 +       bwh = au_dbwh(dentry);
18512 +       bindex = -1;
18513 +       wh_dentry = lock_hdir_create_wh(dentry, /*isdir*/1, &bindex, &a->dt,
18514 +                                       &a->pin);
18515 +       err = PTR_ERR(wh_dentry);
18516 +       if (IS_ERR(wh_dentry))
18517 +               goto out_parent;
18518 +
18519 +       h_dentry = au_h_dptr(dentry, bstart);
18520 +       dget(h_dentry);
18521 +       rmdir_later = 0;
18522 +       if (bindex == bstart) {
18523 +               err = renwh_and_rmdir(dentry, bstart, &args->whlist, dir);
18524 +               if (err > 0) {
18525 +                       rmdir_later = err;
18526 +                       err = 0;
18527 +               }
18528 +       } else {
18529 +               /* stop monitoring */
18530 +               au_hn_free(au_hi(inode, bstart));
18531 +
18532 +               /* dir inode is locked */
18533 +               IMustLock(wh_dentry->d_parent->d_inode);
18534 +               err = 0;
18535 +       }
18536 +
18537 +       if (!err) {
18538 +               vfsub_dead_dir(inode);
18539 +               au_set_dbdiropq(dentry, -1);
18540 +               epilog(dir, dentry, bindex);
18541 +
18542 +               if (rmdir_later) {
18543 +                       au_whtmp_kick_rmdir(dir, bstart, h_dentry, args);
18544 +                       args = NULL;
18545 +               }
18546 +
18547 +               goto out_unpin; /* success */
18548 +       }
18549 +
18550 +       /* revert */
18551 +       AuLabel(revert);
18552 +       if (wh_dentry) {
18553 +               int rerr;
18554 +
18555 +               rerr = do_revert(err, dir, bindex, bwh, wh_dentry, dentry,
18556 +                                &a->dt);
18557 +               if (rerr)
18558 +                       err = rerr;
18559 +       }
18560 +
18561 +out_unpin:
18562 +       au_unpin(&a->pin);
18563 +       dput(wh_dentry);
18564 +       dput(h_dentry);
18565 +out_parent:
18566 +       di_write_unlock(parent);
18567 +       if (args)
18568 +               au_whtmp_rmdir_free(args);
18569 +out_unlock:
18570 +       aufs_read_unlock(dentry, AuLock_DW);
18571 +out_free:
18572 +       kfree(a);
18573 +out:
18574 +       AuTraceErr(err);
18575 +       return err;
18576 +}
18577 diff -urN /usr/share/empty/fs/aufs/i_op_ren.c linux/fs/aufs/i_op_ren.c
18578 --- /usr/share/empty/fs/aufs/i_op_ren.c 1970-01-01 01:00:00.000000000 +0100
18579 +++ linux/fs/aufs/i_op_ren.c    2014-04-24 22:11:10.855269116 +0200
18580 @@ -0,0 +1,1032 @@
18581 +/*
18582 + * Copyright (C) 2005-2014 Junjiro R. Okajima
18583 + *
18584 + * This program, aufs is free software; you can redistribute it and/or modify
18585 + * it under the terms of the GNU General Public License as published by
18586 + * the Free Software Foundation; either version 2 of the License, or
18587 + * (at your option) any later version.
18588 + *
18589 + * This program is distributed in the hope that it will be useful,
18590 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
18591 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18592 + * GNU General Public License for more details.
18593 + *
18594 + * You should have received a copy of the GNU General Public License
18595 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
18596 + */
18597 +
18598 +/*
18599 + * inode operation (rename entry)
18600 + * todo: this is crazy monster
18601 + */
18602 +
18603 +#include "aufs.h"
18604 +
18605 +enum { AuSRC, AuDST, AuSrcDst };
18606 +enum { AuPARENT, AuCHILD, AuParentChild };
18607 +
18608 +#define AuRen_ISDIR    1
18609 +#define AuRen_ISSAMEDIR        (1 << 1)
18610 +#define AuRen_WHSRC    (1 << 2)
18611 +#define AuRen_WHDST    (1 << 3)
18612 +#define AuRen_MNT_WRITE        (1 << 4)
18613 +#define AuRen_DT_DSTDIR        (1 << 5)
18614 +#define AuRen_DIROPQ   (1 << 6)
18615 +#define AuRen_CPUP     (1 << 7)
18616 +#define au_ftest_ren(flags, name)      ((flags) & AuRen_##name)
18617 +#define au_fset_ren(flags, name) \
18618 +       do { (flags) |= AuRen_##name; } while (0)
18619 +#define au_fclr_ren(flags, name) \
18620 +       do { (flags) &= ~AuRen_##name; } while (0)
18621 +
18622 +struct au_ren_args {
18623 +       struct {
18624 +               struct dentry *dentry, *h_dentry, *parent, *h_parent,
18625 +                       *wh_dentry;
18626 +               struct inode *dir, *inode;
18627 +               struct au_hinode *hdir;
18628 +               struct au_dtime dt[AuParentChild];
18629 +               aufs_bindex_t bstart;
18630 +       } sd[AuSrcDst];
18631 +
18632 +#define src_dentry     sd[AuSRC].dentry
18633 +#define src_dir                sd[AuSRC].dir
18634 +#define src_inode      sd[AuSRC].inode
18635 +#define src_h_dentry   sd[AuSRC].h_dentry
18636 +#define src_parent     sd[AuSRC].parent
18637 +#define src_h_parent   sd[AuSRC].h_parent
18638 +#define src_wh_dentry  sd[AuSRC].wh_dentry
18639 +#define src_hdir       sd[AuSRC].hdir
18640 +#define src_h_dir      sd[AuSRC].hdir->hi_inode
18641 +#define src_dt         sd[AuSRC].dt
18642 +#define src_bstart     sd[AuSRC].bstart
18643 +
18644 +#define dst_dentry     sd[AuDST].dentry
18645 +#define dst_dir                sd[AuDST].dir
18646 +#define dst_inode      sd[AuDST].inode
18647 +#define dst_h_dentry   sd[AuDST].h_dentry
18648 +#define dst_parent     sd[AuDST].parent
18649 +#define dst_h_parent   sd[AuDST].h_parent
18650 +#define dst_wh_dentry  sd[AuDST].wh_dentry
18651 +#define dst_hdir       sd[AuDST].hdir
18652 +#define dst_h_dir      sd[AuDST].hdir->hi_inode
18653 +#define dst_dt         sd[AuDST].dt
18654 +#define dst_bstart     sd[AuDST].bstart
18655 +
18656 +       struct dentry *h_trap;
18657 +       struct au_branch *br;
18658 +       struct au_hinode *src_hinode;
18659 +       struct path h_path;
18660 +       struct au_nhash whlist;
18661 +       aufs_bindex_t btgt, src_bwh, src_bdiropq;
18662 +
18663 +       unsigned int flags;
18664 +
18665 +       struct au_whtmp_rmdir *thargs;
18666 +       struct dentry *h_dst;
18667 +};
18668 +
18669 +/* ---------------------------------------------------------------------- */
18670 +
18671 +/*
18672 + * functions for reverting.
18673 + * when an error happened in a single rename systemcall, we should revert
18674 + * everything as if nothing happend.
18675 + * we don't need to revert the copied-up/down the parent dir since they are
18676 + * harmless.
18677 + */
18678 +
18679 +#define RevertFailure(fmt, ...) do { \
18680 +       AuIOErr("revert failure: " fmt " (%d, %d)\n", \
18681 +               ##__VA_ARGS__, err, rerr); \
18682 +       err = -EIO; \
18683 +} while (0)
18684 +
18685 +static void au_ren_rev_diropq(int err, struct au_ren_args *a)
18686 +{
18687 +       int rerr;
18688 +
18689 +       au_hn_imtx_lock_nested(a->src_hinode, AuLsc_I_CHILD);
18690 +       rerr = au_diropq_remove(a->src_dentry, a->btgt);
18691 +       au_hn_imtx_unlock(a->src_hinode);
18692 +       au_set_dbdiropq(a->src_dentry, a->src_bdiropq);
18693 +       if (rerr)
18694 +               RevertFailure("remove diropq %pd", a->src_dentry);
18695 +}
18696 +
18697 +static void au_ren_rev_rename(int err, struct au_ren_args *a)
18698 +{
18699 +       int rerr;
18700 +       struct inode *delegated;
18701 +
18702 +       a->h_path.dentry = vfsub_lkup_one(&a->src_dentry->d_name,
18703 +                                         a->src_h_parent);
18704 +       rerr = PTR_ERR(a->h_path.dentry);
18705 +       if (IS_ERR(a->h_path.dentry)) {
18706 +               RevertFailure("lkup one %pd", a->src_dentry);
18707 +               return;
18708 +       }
18709 +
18710 +       delegated = NULL;
18711 +       rerr = vfsub_rename(a->dst_h_dir,
18712 +                           au_h_dptr(a->src_dentry, a->btgt),
18713 +                           a->src_h_dir, &a->h_path, &delegated);
18714 +       if (unlikely(rerr == -EWOULDBLOCK)) {
18715 +               pr_warn("cannot retry for NFSv4 delegation"
18716 +                       " for an internal rename\n");
18717 +               iput(delegated);
18718 +       }
18719 +       d_drop(a->h_path.dentry);
18720 +       dput(a->h_path.dentry);
18721 +       /* au_set_h_dptr(a->src_dentry, a->btgt, NULL); */
18722 +       if (rerr)
18723 +               RevertFailure("rename %pd", a->src_dentry);
18724 +}
18725 +
18726 +static void au_ren_rev_cpup(int err, struct au_ren_args *a)
18727 +{
18728 +       int rerr;
18729 +
18730 +       a->h_path.dentry = a->dst_h_dentry;
18731 +       /* no delegation since it is just created */
18732 +       rerr = vfsub_unlink(a->dst_h_dir, &a->h_path, /*delegated*/NULL,
18733 +                           /*force*/0);
18734 +       au_set_h_dptr(a->src_dentry, a->btgt, NULL);
18735 +       au_set_dbstart(a->src_dentry, a->src_bstart);
18736 +       if (rerr)
18737 +               RevertFailure("unlink %pd", a->dst_h_dentry);
18738 +}
18739 +
18740 +static void au_ren_rev_whtmp(int err, struct au_ren_args *a)
18741 +{
18742 +       int rerr;
18743 +       struct inode *delegated;
18744 +
18745 +       a->h_path.dentry = vfsub_lkup_one(&a->dst_dentry->d_name,
18746 +                                         a->dst_h_parent);
18747 +       rerr = PTR_ERR(a->h_path.dentry);
18748 +       if (IS_ERR(a->h_path.dentry)) {
18749 +               RevertFailure("lkup one %pd", a->dst_dentry);
18750 +               return;
18751 +       }
18752 +       if (a->h_path.dentry->d_inode) {
18753 +               d_drop(a->h_path.dentry);
18754 +               dput(a->h_path.dentry);
18755 +               return;
18756 +       }
18757 +
18758 +       delegated = NULL;
18759 +       rerr = vfsub_rename(a->dst_h_dir, a->h_dst, a->dst_h_dir, &a->h_path,
18760 +                           &delegated);
18761 +       if (unlikely(rerr == -EWOULDBLOCK)) {
18762 +               pr_warn("cannot retry for NFSv4 delegation"
18763 +                       " for an internal rename\n");
18764 +               iput(delegated);
18765 +       }
18766 +       d_drop(a->h_path.dentry);
18767 +       dput(a->h_path.dentry);
18768 +       if (!rerr)
18769 +               au_set_h_dptr(a->dst_dentry, a->btgt, dget(a->h_dst));
18770 +       else
18771 +               RevertFailure("rename %pd", a->h_dst);
18772 +}
18773 +
18774 +static void au_ren_rev_whsrc(int err, struct au_ren_args *a)
18775 +{
18776 +       int rerr;
18777 +
18778 +       a->h_path.dentry = a->src_wh_dentry;
18779 +       rerr = au_wh_unlink_dentry(a->src_h_dir, &a->h_path, a->src_dentry);
18780 +       au_set_dbwh(a->src_dentry, a->src_bwh);
18781 +       if (rerr)
18782 +               RevertFailure("unlink %pd", a->src_wh_dentry);
18783 +}
18784 +#undef RevertFailure
18785 +
18786 +/* ---------------------------------------------------------------------- */
18787 +
18788 +/*
18789 + * when we have to copyup the renaming entry, do it with the rename-target name
18790 + * in order to minimize the cost (the later actual rename is unnecessary).
18791 + * otherwise rename it on the target branch.
18792 + */
18793 +static int au_ren_or_cpup(struct au_ren_args *a)
18794 +{
18795 +       int err;
18796 +       struct dentry *d;
18797 +       struct inode *delegated;
18798 +
18799 +       d = a->src_dentry;
18800 +       if (au_dbstart(d) == a->btgt) {
18801 +               a->h_path.dentry = a->dst_h_dentry;
18802 +               if (au_ftest_ren(a->flags, DIROPQ)
18803 +                   && au_dbdiropq(d) == a->btgt)
18804 +                       au_fclr_ren(a->flags, DIROPQ);
18805 +               AuDebugOn(au_dbstart(d) != a->btgt);
18806 +               delegated = NULL;
18807 +               err = vfsub_rename(a->src_h_dir, au_h_dptr(d, a->btgt),
18808 +                                  a->dst_h_dir, &a->h_path, &delegated);
18809 +               if (unlikely(err == -EWOULDBLOCK)) {
18810 +                       pr_warn("cannot retry for NFSv4 delegation"
18811 +                               " for an internal rename\n");
18812 +                       iput(delegated);
18813 +               }
18814 +       } else
18815 +               BUG();
18816 +
18817 +       if (!err && a->h_dst)
18818 +               /* it will be set to dinfo later */
18819 +               dget(a->h_dst);
18820 +
18821 +       return err;
18822 +}
18823 +
18824 +/* cf. aufs_rmdir() */
18825 +static int au_ren_del_whtmp(struct au_ren_args *a)
18826 +{
18827 +       int err;
18828 +       struct inode *dir;
18829 +
18830 +       dir = a->dst_dir;
18831 +       SiMustAnyLock(dir->i_sb);
18832 +       if (!au_nhash_test_longer_wh(&a->whlist, a->btgt,
18833 +                                    au_sbi(dir->i_sb)->si_dirwh)
18834 +           || au_test_fs_remote(a->h_dst->d_sb)) {
18835 +               err = au_whtmp_rmdir(dir, a->btgt, a->h_dst, &a->whlist);
18836 +               if (unlikely(err))
18837 +                       pr_warn("failed removing whtmp dir %pd (%d), "
18838 +                               "ignored.\n", a->h_dst, err);
18839 +       } else {
18840 +               au_nhash_wh_free(&a->thargs->whlist);
18841 +               a->thargs->whlist = a->whlist;
18842 +               a->whlist.nh_num = 0;
18843 +               au_whtmp_kick_rmdir(dir, a->btgt, a->h_dst, a->thargs);
18844 +               dput(a->h_dst);
18845 +               a->thargs = NULL;
18846 +       }
18847 +
18848 +       return 0;
18849 +}
18850 +
18851 +/* make it 'opaque' dir. */
18852 +static int au_ren_diropq(struct au_ren_args *a)
18853 +{
18854 +       int err;
18855 +       struct dentry *diropq;
18856 +
18857 +       err = 0;
18858 +       a->src_bdiropq = au_dbdiropq(a->src_dentry);
18859 +       a->src_hinode = au_hi(a->src_inode, a->btgt);
18860 +       au_hn_imtx_lock_nested(a->src_hinode, AuLsc_I_CHILD);
18861 +       diropq = au_diropq_create(a->src_dentry, a->btgt);
18862 +       au_hn_imtx_unlock(a->src_hinode);
18863 +       if (IS_ERR(diropq))
18864 +               err = PTR_ERR(diropq);
18865 +       dput(diropq);
18866 +
18867 +       return err;
18868 +}
18869 +
18870 +static int do_rename(struct au_ren_args *a)
18871 +{
18872 +       int err;
18873 +       struct dentry *d, *h_d;
18874 +
18875 +       /* prepare workqueue args for asynchronous rmdir */
18876 +       h_d = a->dst_h_dentry;
18877 +       if (au_ftest_ren(a->flags, ISDIR) && h_d->d_inode) {
18878 +               err = -ENOMEM;
18879 +               a->thargs = au_whtmp_rmdir_alloc(a->src_dentry->d_sb, GFP_NOFS);
18880 +               if (unlikely(!a->thargs))
18881 +                       goto out;
18882 +               a->h_dst = dget(h_d);
18883 +       }
18884 +
18885 +       /* create whiteout for src_dentry */
18886 +       if (au_ftest_ren(a->flags, WHSRC)) {
18887 +               a->src_bwh = au_dbwh(a->src_dentry);
18888 +               AuDebugOn(a->src_bwh >= 0);
18889 +               a->src_wh_dentry
18890 +                       = au_wh_create(a->src_dentry, a->btgt, a->src_h_parent);
18891 +               err = PTR_ERR(a->src_wh_dentry);
18892 +               if (IS_ERR(a->src_wh_dentry))
18893 +                       goto out_thargs;
18894 +       }
18895 +
18896 +       /* lookup whiteout for dentry */
18897 +       if (au_ftest_ren(a->flags, WHDST)) {
18898 +               h_d = au_wh_lkup(a->dst_h_parent, &a->dst_dentry->d_name,
18899 +                                a->br);
18900 +               err = PTR_ERR(h_d);
18901 +               if (IS_ERR(h_d))
18902 +                       goto out_whsrc;
18903 +               if (!h_d->d_inode)
18904 +                       dput(h_d);
18905 +               else
18906 +                       a->dst_wh_dentry = h_d;
18907 +       }
18908 +
18909 +       /* rename dentry to tmpwh */
18910 +       if (a->thargs) {
18911 +               err = au_whtmp_ren(a->dst_h_dentry, a->br);
18912 +               if (unlikely(err))
18913 +                       goto out_whdst;
18914 +
18915 +               d = a->dst_dentry;
18916 +               au_set_h_dptr(d, a->btgt, NULL);
18917 +               err = au_lkup_neg(d, a->btgt, /*wh*/0);
18918 +               if (unlikely(err))
18919 +                       goto out_whtmp;
18920 +               a->dst_h_dentry = au_h_dptr(d, a->btgt);
18921 +       }
18922 +
18923 +       BUG_ON(a->dst_h_dentry->d_inode && a->src_bstart != a->btgt);
18924 +
18925 +       /* rename by vfs_rename or cpup */
18926 +       d = a->dst_dentry;
18927 +       if (au_ftest_ren(a->flags, ISDIR)
18928 +           && (a->dst_wh_dentry
18929 +               || au_dbdiropq(d) == a->btgt
18930 +               /* hide the lower to keep xino */
18931 +               || a->btgt < au_dbend(d)
18932 +               || au_opt_test(au_mntflags(d->d_sb), ALWAYS_DIROPQ)))
18933 +               au_fset_ren(a->flags, DIROPQ);
18934 +       err = au_ren_or_cpup(a);
18935 +       if (unlikely(err))
18936 +               /* leave the copied-up one */
18937 +               goto out_whtmp;
18938 +
18939 +       /* make dir opaque */
18940 +       if (au_ftest_ren(a->flags, DIROPQ)) {
18941 +               err = au_ren_diropq(a);
18942 +               if (unlikely(err))
18943 +                       goto out_rename;
18944 +       }
18945 +
18946 +       /* update target timestamps */
18947 +       AuDebugOn(au_dbstart(a->src_dentry) != a->btgt);
18948 +       a->h_path.dentry = au_h_dptr(a->src_dentry, a->btgt);
18949 +       vfsub_update_h_iattr(&a->h_path, /*did*/NULL); /*ignore*/
18950 +       a->src_inode->i_ctime = a->h_path.dentry->d_inode->i_ctime;
18951 +
18952 +       /* remove whiteout for dentry */
18953 +       if (a->dst_wh_dentry) {
18954 +               a->h_path.dentry = a->dst_wh_dentry;
18955 +               err = au_wh_unlink_dentry(a->dst_h_dir, &a->h_path,
18956 +                                         a->dst_dentry);
18957 +               if (unlikely(err))
18958 +                       goto out_diropq;
18959 +       }
18960 +
18961 +       /* remove whtmp */
18962 +       if (a->thargs)
18963 +               au_ren_del_whtmp(a); /* ignore this error */
18964 +
18965 +       err = 0;
18966 +       goto out_success;
18967 +
18968 +out_diropq:
18969 +       if (au_ftest_ren(a->flags, DIROPQ))
18970 +               au_ren_rev_diropq(err, a);
18971 +out_rename:
18972 +       if (!au_ftest_ren(a->flags, CPUP))
18973 +               au_ren_rev_rename(err, a);
18974 +       else
18975 +               au_ren_rev_cpup(err, a);
18976 +       dput(a->h_dst);
18977 +out_whtmp:
18978 +       if (a->thargs)
18979 +               au_ren_rev_whtmp(err, a);
18980 +out_whdst:
18981 +       dput(a->dst_wh_dentry);
18982 +       a->dst_wh_dentry = NULL;
18983 +out_whsrc:
18984 +       if (a->src_wh_dentry)
18985 +               au_ren_rev_whsrc(err, a);
18986 +out_success:
18987 +       dput(a->src_wh_dentry);
18988 +       dput(a->dst_wh_dentry);
18989 +out_thargs:
18990 +       if (a->thargs) {
18991 +               dput(a->h_dst);
18992 +               au_whtmp_rmdir_free(a->thargs);
18993 +               a->thargs = NULL;
18994 +       }
18995 +out:
18996 +       return err;
18997 +}
18998 +
18999 +/* ---------------------------------------------------------------------- */
19000 +
19001 +/*
19002 + * test if @dentry dir can be rename destination or not.
19003 + * success means, it is a logically empty dir.
19004 + */
19005 +static int may_rename_dstdir(struct dentry *dentry, struct au_nhash *whlist)
19006 +{
19007 +       return au_test_empty(dentry, whlist);
19008 +}
19009 +
19010 +/*
19011 + * test if @dentry dir can be rename source or not.
19012 + * if it can, return 0 and @children is filled.
19013 + * success means,
19014 + * - it is a logically empty dir.
19015 + * - or, it exists on writable branch and has no children including whiteouts
19016 + *       on the lower branch.
19017 + */
19018 +static int may_rename_srcdir(struct dentry *dentry, aufs_bindex_t btgt)
19019 +{
19020 +       int err;
19021 +       unsigned int rdhash;
19022 +       aufs_bindex_t bstart;
19023 +
19024 +       bstart = au_dbstart(dentry);
19025 +       if (bstart != btgt) {
19026 +               struct au_nhash whlist;
19027 +
19028 +               SiMustAnyLock(dentry->d_sb);
19029 +               rdhash = au_sbi(dentry->d_sb)->si_rdhash;
19030 +               if (!rdhash)
19031 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL,
19032 +                                                          dentry));
19033 +               err = au_nhash_alloc(&whlist, rdhash, GFP_NOFS);
19034 +               if (unlikely(err))
19035 +                       goto out;
19036 +               err = au_test_empty(dentry, &whlist);
19037 +               au_nhash_wh_free(&whlist);
19038 +               goto out;
19039 +       }
19040 +
19041 +       if (bstart == au_dbtaildir(dentry))
19042 +               return 0; /* success */
19043 +
19044 +       err = au_test_empty_lower(dentry);
19045 +
19046 +out:
19047 +       if (err == -ENOTEMPTY) {
19048 +               AuWarn1("renaming dir who has child(ren) on multiple branches,"
19049 +                       " is not supported\n");
19050 +               err = -EXDEV;
19051 +       }
19052 +       return err;
19053 +}
19054 +
19055 +/* side effect: sets whlist and h_dentry */
19056 +static int au_ren_may_dir(struct au_ren_args *a)
19057 +{
19058 +       int err;
19059 +       unsigned int rdhash;
19060 +       struct dentry *d;
19061 +
19062 +       d = a->dst_dentry;
19063 +       SiMustAnyLock(d->d_sb);
19064 +
19065 +       err = 0;
19066 +       if (au_ftest_ren(a->flags, ISDIR) && a->dst_inode) {
19067 +               rdhash = au_sbi(d->d_sb)->si_rdhash;
19068 +               if (!rdhash)
19069 +                       rdhash = au_rdhash_est(au_dir_size(/*file*/NULL, d));
19070 +               err = au_nhash_alloc(&a->whlist, rdhash, GFP_NOFS);
19071 +               if (unlikely(err))
19072 +                       goto out;
19073 +
19074 +               au_set_dbstart(d, a->dst_bstart);
19075 +               err = may_rename_dstdir(d, &a->whlist);
19076 +               au_set_dbstart(d, a->btgt);
19077 +       }
19078 +       a->dst_h_dentry = au_h_dptr(d, au_dbstart(d));
19079 +       if (unlikely(err))
19080 +               goto out;
19081 +
19082 +       d = a->src_dentry;
19083 +       a->src_h_dentry = au_h_dptr(d, au_dbstart(d));
19084 +       if (au_ftest_ren(a->flags, ISDIR)) {
19085 +               err = may_rename_srcdir(d, a->btgt);
19086 +               if (unlikely(err)) {
19087 +                       au_nhash_wh_free(&a->whlist);
19088 +                       a->whlist.nh_num = 0;
19089 +               }
19090 +       }
19091 +out:
19092 +       return err;
19093 +}
19094 +
19095 +/* ---------------------------------------------------------------------- */
19096 +
19097 +/*
19098 + * simple tests for rename.
19099 + * following the checks in vfs, plus the parent-child relationship.
19100 + */
19101 +static int au_may_ren(struct au_ren_args *a)
19102 +{
19103 +       int err, isdir;
19104 +       struct inode *h_inode;
19105 +
19106 +       if (a->src_bstart == a->btgt) {
19107 +               err = au_may_del(a->src_dentry, a->btgt, a->src_h_parent,
19108 +                                au_ftest_ren(a->flags, ISDIR));
19109 +               if (unlikely(err))
19110 +                       goto out;
19111 +               err = -EINVAL;
19112 +               if (unlikely(a->src_h_dentry == a->h_trap))
19113 +                       goto out;
19114 +       }
19115 +
19116 +       err = 0;
19117 +       if (a->dst_bstart != a->btgt)
19118 +               goto out;
19119 +
19120 +       err = -ENOTEMPTY;
19121 +       if (unlikely(a->dst_h_dentry == a->h_trap))
19122 +               goto out;
19123 +
19124 +       err = -EIO;
19125 +       h_inode = a->dst_h_dentry->d_inode;
19126 +       isdir = !!au_ftest_ren(a->flags, ISDIR);
19127 +       if (!a->dst_dentry->d_inode) {
19128 +               if (unlikely(h_inode))
19129 +                       goto out;
19130 +               err = au_may_add(a->dst_dentry, a->btgt, a->dst_h_parent,
19131 +                                isdir);
19132 +       } else {
19133 +               if (unlikely(!h_inode || !h_inode->i_nlink))
19134 +                       goto out;
19135 +               err = au_may_del(a->dst_dentry, a->btgt, a->dst_h_parent,
19136 +                                isdir);
19137 +               if (unlikely(err))
19138 +                       goto out;
19139 +       }
19140 +
19141 +out:
19142 +       if (unlikely(err == -ENOENT || err == -EEXIST))
19143 +               err = -EIO;
19144 +       AuTraceErr(err);
19145 +       return err;
19146 +}
19147 +
19148 +/* ---------------------------------------------------------------------- */
19149 +
19150 +/*
19151 + * locking order
19152 + * (VFS)
19153 + * - src_dir and dir by lock_rename()
19154 + * - inode if exitsts
19155 + * (aufs)
19156 + * - lock all
19157 + *   + src_dentry and dentry by aufs_read_and_write_lock2() which calls,
19158 + *     + si_read_lock
19159 + *     + di_write_lock2_child()
19160 + *       + di_write_lock_child()
19161 + *        + ii_write_lock_child()
19162 + *       + di_write_lock_child2()
19163 + *        + ii_write_lock_child2()
19164 + *     + src_parent and parent
19165 + *       + di_write_lock_parent()
19166 + *        + ii_write_lock_parent()
19167 + *       + di_write_lock_parent2()
19168 + *        + ii_write_lock_parent2()
19169 + *   + lower src_dir and dir by vfsub_lock_rename()
19170 + *   + verify the every relationships between child and parent. if any
19171 + *     of them failed, unlock all and return -EBUSY.
19172 + */
19173 +static void au_ren_unlock(struct au_ren_args *a)
19174 +{
19175 +       vfsub_unlock_rename(a->src_h_parent, a->src_hdir,
19176 +                           a->dst_h_parent, a->dst_hdir);
19177 +       if (au_ftest_ren(a->flags, MNT_WRITE))
19178 +               vfsub_mnt_drop_write(au_br_mnt(a->br));
19179 +}
19180 +
19181 +static int au_ren_lock(struct au_ren_args *a)
19182 +{
19183 +       int err;
19184 +       unsigned int udba;
19185 +
19186 +       err = 0;
19187 +       a->src_h_parent = au_h_dptr(a->src_parent, a->btgt);
19188 +       a->src_hdir = au_hi(a->src_dir, a->btgt);
19189 +       a->dst_h_parent = au_h_dptr(a->dst_parent, a->btgt);
19190 +       a->dst_hdir = au_hi(a->dst_dir, a->btgt);
19191 +
19192 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
19193 +       if (unlikely(err))
19194 +               goto out;
19195 +       au_fset_ren(a->flags, MNT_WRITE);
19196 +       a->h_trap = vfsub_lock_rename(a->src_h_parent, a->src_hdir,
19197 +                                     a->dst_h_parent, a->dst_hdir);
19198 +       udba = au_opt_udba(a->src_dentry->d_sb);
19199 +       if (unlikely(a->src_hdir->hi_inode != a->src_h_parent->d_inode
19200 +                    || a->dst_hdir->hi_inode != a->dst_h_parent->d_inode))
19201 +               err = au_busy_or_stale();
19202 +       if (!err && au_dbstart(a->src_dentry) == a->btgt)
19203 +               err = au_h_verify(a->src_h_dentry, udba,
19204 +                                 a->src_h_parent->d_inode, a->src_h_parent,
19205 +                                 a->br);
19206 +       if (!err && au_dbstart(a->dst_dentry) == a->btgt)
19207 +               err = au_h_verify(a->dst_h_dentry, udba,
19208 +                                 a->dst_h_parent->d_inode, a->dst_h_parent,
19209 +                                 a->br);
19210 +       if (!err)
19211 +               goto out; /* success */
19212 +
19213 +       err = au_busy_or_stale();
19214 +       au_ren_unlock(a);
19215 +
19216 +out:
19217 +       return err;
19218 +}
19219 +
19220 +/* ---------------------------------------------------------------------- */
19221 +
19222 +static void au_ren_refresh_dir(struct au_ren_args *a)
19223 +{
19224 +       struct inode *dir;
19225 +
19226 +       dir = a->dst_dir;
19227 +       dir->i_version++;
19228 +       if (au_ftest_ren(a->flags, ISDIR)) {
19229 +               /* is this updating defined in POSIX? */
19230 +               au_cpup_attr_timesizes(a->src_inode);
19231 +               au_cpup_attr_nlink(dir, /*force*/1);
19232 +       }
19233 +
19234 +       if (au_ibstart(dir) == a->btgt)
19235 +               au_cpup_attr_timesizes(dir);
19236 +
19237 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
19238 +               return;
19239 +
19240 +       dir = a->src_dir;
19241 +       dir->i_version++;
19242 +       if (au_ftest_ren(a->flags, ISDIR))
19243 +               au_cpup_attr_nlink(dir, /*force*/1);
19244 +       if (au_ibstart(dir) == a->btgt)
19245 +               au_cpup_attr_timesizes(dir);
19246 +}
19247 +
19248 +static void au_ren_refresh(struct au_ren_args *a)
19249 +{
19250 +       aufs_bindex_t bend, bindex;
19251 +       struct dentry *d, *h_d;
19252 +       struct inode *i, *h_i;
19253 +       struct super_block *sb;
19254 +
19255 +       d = a->dst_dentry;
19256 +       d_drop(d);
19257 +       if (a->h_dst)
19258 +               /* already dget-ed by au_ren_or_cpup() */
19259 +               au_set_h_dptr(d, a->btgt, a->h_dst);
19260 +
19261 +       i = a->dst_inode;
19262 +       if (i) {
19263 +               if (!au_ftest_ren(a->flags, ISDIR))
19264 +                       vfsub_drop_nlink(i);
19265 +               else {
19266 +                       vfsub_dead_dir(i);
19267 +                       au_cpup_attr_timesizes(i);
19268 +               }
19269 +               au_update_dbrange(d, /*do_put_zero*/1);
19270 +       } else {
19271 +               bend = a->btgt;
19272 +               for (bindex = au_dbstart(d); bindex < bend; bindex++)
19273 +                       au_set_h_dptr(d, bindex, NULL);
19274 +               bend = au_dbend(d);
19275 +               for (bindex = a->btgt + 1; bindex <= bend; bindex++)
19276 +                       au_set_h_dptr(d, bindex, NULL);
19277 +               au_update_dbrange(d, /*do_put_zero*/0);
19278 +       }
19279 +
19280 +       d = a->src_dentry;
19281 +       au_set_dbwh(d, -1);
19282 +       bend = au_dbend(d);
19283 +       for (bindex = a->btgt + 1; bindex <= bend; bindex++) {
19284 +               h_d = au_h_dptr(d, bindex);
19285 +               if (h_d)
19286 +                       au_set_h_dptr(d, bindex, NULL);
19287 +       }
19288 +       au_set_dbend(d, a->btgt);
19289 +
19290 +       sb = d->d_sb;
19291 +       i = a->src_inode;
19292 +       if (au_opt_test(au_mntflags(sb), PLINK) && au_plink_test(i))
19293 +               return; /* success */
19294 +
19295 +       bend = au_ibend(i);
19296 +       for (bindex = a->btgt + 1; bindex <= bend; bindex++) {
19297 +               h_i = au_h_iptr(i, bindex);
19298 +               if (h_i) {
19299 +                       au_xino_write(sb, bindex, h_i->i_ino, /*ino*/0);
19300 +                       /* ignore this error */
19301 +                       au_set_h_iptr(i, bindex, NULL, 0);
19302 +               }
19303 +       }
19304 +       au_set_ibend(i, a->btgt);
19305 +}
19306 +
19307 +/* ---------------------------------------------------------------------- */
19308 +
19309 +/* mainly for link(2) and rename(2) */
19310 +int au_wbr(struct dentry *dentry, aufs_bindex_t btgt)
19311 +{
19312 +       aufs_bindex_t bdiropq, bwh;
19313 +       struct dentry *parent;
19314 +       struct au_branch *br;
19315 +
19316 +       parent = dentry->d_parent;
19317 +       IMustLock(parent->d_inode); /* dir is locked */
19318 +
19319 +       bdiropq = au_dbdiropq(parent);
19320 +       bwh = au_dbwh(dentry);
19321 +       br = au_sbr(dentry->d_sb, btgt);
19322 +       if (au_br_rdonly(br)
19323 +           || (0 <= bdiropq && bdiropq < btgt)
19324 +           || (0 <= bwh && bwh < btgt))
19325 +               btgt = -1;
19326 +
19327 +       AuDbg("btgt %d\n", btgt);
19328 +       return btgt;
19329 +}
19330 +
19331 +/* sets src_bstart, dst_bstart and btgt */
19332 +static int au_ren_wbr(struct au_ren_args *a)
19333 +{
19334 +       int err;
19335 +       struct au_wr_dir_args wr_dir_args = {
19336 +               /* .force_btgt  = -1, */
19337 +               .flags          = AuWrDir_ADD_ENTRY
19338 +       };
19339 +
19340 +       a->src_bstart = au_dbstart(a->src_dentry);
19341 +       a->dst_bstart = au_dbstart(a->dst_dentry);
19342 +       if (au_ftest_ren(a->flags, ISDIR))
19343 +               au_fset_wrdir(wr_dir_args.flags, ISDIR);
19344 +       wr_dir_args.force_btgt = a->src_bstart;
19345 +       if (a->dst_inode && a->dst_bstart < a->src_bstart)
19346 +               wr_dir_args.force_btgt = a->dst_bstart;
19347 +       wr_dir_args.force_btgt = au_wbr(a->dst_dentry, wr_dir_args.force_btgt);
19348 +       err = au_wr_dir(a->dst_dentry, a->src_dentry, &wr_dir_args);
19349 +       a->btgt = err;
19350 +
19351 +       return err;
19352 +}
19353 +
19354 +static void au_ren_dt(struct au_ren_args *a)
19355 +{
19356 +       a->h_path.dentry = a->src_h_parent;
19357 +       au_dtime_store(a->src_dt + AuPARENT, a->src_parent, &a->h_path);
19358 +       if (!au_ftest_ren(a->flags, ISSAMEDIR)) {
19359 +               a->h_path.dentry = a->dst_h_parent;
19360 +               au_dtime_store(a->dst_dt + AuPARENT, a->dst_parent, &a->h_path);
19361 +       }
19362 +
19363 +       au_fclr_ren(a->flags, DT_DSTDIR);
19364 +       if (!au_ftest_ren(a->flags, ISDIR))
19365 +               return;
19366 +
19367 +       a->h_path.dentry = a->src_h_dentry;
19368 +       au_dtime_store(a->src_dt + AuCHILD, a->src_dentry, &a->h_path);
19369 +       if (a->dst_h_dentry->d_inode) {
19370 +               au_fset_ren(a->flags, DT_DSTDIR);
19371 +               a->h_path.dentry = a->dst_h_dentry;
19372 +               au_dtime_store(a->dst_dt + AuCHILD, a->dst_dentry, &a->h_path);
19373 +       }
19374 +}
19375 +
19376 +static void au_ren_rev_dt(int err, struct au_ren_args *a)
19377 +{
19378 +       struct dentry *h_d;
19379 +       struct mutex *h_mtx;
19380 +
19381 +       au_dtime_revert(a->src_dt + AuPARENT);
19382 +       if (!au_ftest_ren(a->flags, ISSAMEDIR))
19383 +               au_dtime_revert(a->dst_dt + AuPARENT);
19384 +
19385 +       if (au_ftest_ren(a->flags, ISDIR) && err != -EIO) {
19386 +               h_d = a->src_dt[AuCHILD].dt_h_path.dentry;
19387 +               h_mtx = &h_d->d_inode->i_mutex;
19388 +               mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
19389 +               au_dtime_revert(a->src_dt + AuCHILD);
19390 +               mutex_unlock(h_mtx);
19391 +
19392 +               if (au_ftest_ren(a->flags, DT_DSTDIR)) {
19393 +                       h_d = a->dst_dt[AuCHILD].dt_h_path.dentry;
19394 +                       h_mtx = &h_d->d_inode->i_mutex;
19395 +                       mutex_lock_nested(h_mtx, AuLsc_I_CHILD);
19396 +                       au_dtime_revert(a->dst_dt + AuCHILD);
19397 +                       mutex_unlock(h_mtx);
19398 +               }
19399 +       }
19400 +}
19401 +
19402 +/* ---------------------------------------------------------------------- */
19403 +
19404 +int aufs_rename(struct inode *_src_dir, struct dentry *_src_dentry,
19405 +               struct inode *_dst_dir, struct dentry *_dst_dentry)
19406 +{
19407 +       int err, flags;
19408 +       /* reduce stack space */
19409 +       struct au_ren_args *a;
19410 +
19411 +       AuDbg("%pd, %pd\n", _src_dentry, _dst_dentry);
19412 +       IMustLock(_src_dir);
19413 +       IMustLock(_dst_dir);
19414 +
19415 +       err = -ENOMEM;
19416 +       BUILD_BUG_ON(sizeof(*a) > PAGE_SIZE);
19417 +       a = kzalloc(sizeof(*a), GFP_NOFS);
19418 +       if (unlikely(!a))
19419 +               goto out;
19420 +
19421 +       a->src_dir = _src_dir;
19422 +       a->src_dentry = _src_dentry;
19423 +       a->src_inode = a->src_dentry->d_inode;
19424 +       a->src_parent = a->src_dentry->d_parent; /* dir inode is locked */
19425 +       a->dst_dir = _dst_dir;
19426 +       a->dst_dentry = _dst_dentry;
19427 +       a->dst_inode = a->dst_dentry->d_inode;
19428 +       a->dst_parent = a->dst_dentry->d_parent; /* dir inode is locked */
19429 +       if (a->dst_inode) {
19430 +               IMustLock(a->dst_inode);
19431 +               au_igrab(a->dst_inode);
19432 +       }
19433 +
19434 +       err = -ENOTDIR;
19435 +       flags = AuLock_FLUSH | AuLock_NOPLM | AuLock_GEN;
19436 +       if (S_ISDIR(a->src_inode->i_mode)) {
19437 +               au_fset_ren(a->flags, ISDIR);
19438 +               if (unlikely(a->dst_inode && !S_ISDIR(a->dst_inode->i_mode)))
19439 +                       goto out_free;
19440 +               err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry,
19441 +                                               AuLock_DIR | flags);
19442 +       } else
19443 +               err = aufs_read_and_write_lock2(a->dst_dentry, a->src_dentry,
19444 +                                               flags);
19445 +       if (unlikely(err))
19446 +               goto out_free;
19447 +
19448 +       err = au_d_hashed_positive(a->src_dentry);
19449 +       if (unlikely(err))
19450 +               goto out_unlock;
19451 +       err = -ENOENT;
19452 +       if (a->dst_inode) {
19453 +               /*
19454 +                * If it is a dir, VFS unhash dst_dentry before this
19455 +                * function. It means we cannot rely upon d_unhashed().
19456 +                */
19457 +               if (unlikely(!a->dst_inode->i_nlink))
19458 +                       goto out_unlock;
19459 +               if (!S_ISDIR(a->dst_inode->i_mode)) {
19460 +                       err = au_d_hashed_positive(a->dst_dentry);
19461 +                       if (unlikely(err))
19462 +                               goto out_unlock;
19463 +               } else if (unlikely(IS_DEADDIR(a->dst_inode)))
19464 +                       goto out_unlock;
19465 +       } else if (unlikely(d_unhashed(a->dst_dentry)))
19466 +               goto out_unlock;
19467 +
19468 +       /*
19469 +        * is it possible?
19470 +        * yes, it happend (in linux-3.3-rcN) but I don't know why.
19471 +        * there may exist a problem somewhere else.
19472 +        */
19473 +       err = -EINVAL;
19474 +       if (unlikely(a->dst_parent->d_inode == a->src_dentry->d_inode))
19475 +               goto out_unlock;
19476 +
19477 +       au_fset_ren(a->flags, ISSAMEDIR); /* temporary */
19478 +       di_write_lock_parent(a->dst_parent);
19479 +
19480 +       /* which branch we process */
19481 +       err = au_ren_wbr(a);
19482 +       if (unlikely(err < 0))
19483 +               goto out_parent;
19484 +       a->br = au_sbr(a->dst_dentry->d_sb, a->btgt);
19485 +       a->h_path.mnt = au_br_mnt(a->br);
19486 +
19487 +       /* are they available to be renamed */
19488 +       err = au_ren_may_dir(a);
19489 +       if (unlikely(err))
19490 +               goto out_children;
19491 +
19492 +       /* prepare the writable parent dir on the same branch */
19493 +       if (a->dst_bstart == a->btgt) {
19494 +               au_fset_ren(a->flags, WHDST);
19495 +       } else {
19496 +               err = au_cpup_dirs(a->dst_dentry, a->btgt);
19497 +               if (unlikely(err))
19498 +                       goto out_children;
19499 +       }
19500 +
19501 +       if (a->src_dir != a->dst_dir) {
19502 +               /*
19503 +                * this temporary unlock is safe,
19504 +                * because both dir->i_mutex are locked.
19505 +                */
19506 +               di_write_unlock(a->dst_parent);
19507 +               di_write_lock_parent(a->src_parent);
19508 +               err = au_wr_dir_need_wh(a->src_dentry,
19509 +                                       au_ftest_ren(a->flags, ISDIR),
19510 +                                       &a->btgt);
19511 +               di_write_unlock(a->src_parent);
19512 +               di_write_lock2_parent(a->src_parent, a->dst_parent, /*isdir*/1);
19513 +               au_fclr_ren(a->flags, ISSAMEDIR);
19514 +       } else
19515 +               err = au_wr_dir_need_wh(a->src_dentry,
19516 +                                       au_ftest_ren(a->flags, ISDIR),
19517 +                                       &a->btgt);
19518 +       if (unlikely(err < 0))
19519 +               goto out_children;
19520 +       if (err)
19521 +               au_fset_ren(a->flags, WHSRC);
19522 +
19523 +       /* cpup src */
19524 +       if (a->src_bstart != a->btgt) {
19525 +               struct au_pin pin;
19526 +
19527 +               err = au_pin(&pin, a->src_dentry, a->btgt,
19528 +                            au_opt_udba(a->src_dentry->d_sb),
19529 +                            AuPin_DI_LOCKED | AuPin_MNT_WRITE);
19530 +               if (!err) {
19531 +                       struct au_cp_generic cpg = {
19532 +                               .dentry = a->src_dentry,
19533 +                               .bdst   = a->btgt,
19534 +                               .bsrc   = a->src_bstart,
19535 +                               .len    = -1,
19536 +                               .pin    = &pin,
19537 +                               .flags  = AuCpup_DTIME | AuCpup_HOPEN
19538 +                       };
19539 +                       AuDebugOn(au_dbstart(a->src_dentry) != a->src_bstart);
19540 +                       err = au_sio_cpup_simple(&cpg);
19541 +                       au_unpin(&pin);
19542 +               }
19543 +               if (unlikely(err))
19544 +                       goto out_children;
19545 +               a->src_bstart = a->btgt;
19546 +               a->src_h_dentry = au_h_dptr(a->src_dentry, a->btgt);
19547 +               au_fset_ren(a->flags, WHSRC);
19548 +       }
19549 +
19550 +       /* lock them all */
19551 +       err = au_ren_lock(a);
19552 +       if (unlikely(err))
19553 +               /* leave the copied-up one */
19554 +               goto out_children;
19555 +
19556 +       if (!au_opt_test(au_mntflags(a->dst_dir->i_sb), UDBA_NONE))
19557 +               err = au_may_ren(a);
19558 +       else if (unlikely(a->dst_dentry->d_name.len > AUFS_MAX_NAMELEN))
19559 +               err = -ENAMETOOLONG;
19560 +       if (unlikely(err))
19561 +               goto out_hdir;
19562 +
19563 +       /* store timestamps to be revertible */
19564 +       au_ren_dt(a);
19565 +
19566 +       /* here we go */
19567 +       err = do_rename(a);
19568 +       if (unlikely(err))
19569 +               goto out_dt;
19570 +
19571 +       /* update dir attributes */
19572 +       au_ren_refresh_dir(a);
19573 +
19574 +       /* dput/iput all lower dentries */
19575 +       au_ren_refresh(a);
19576 +
19577 +       goto out_hdir; /* success */
19578 +
19579 +out_dt:
19580 +       au_ren_rev_dt(err, a);
19581 +out_hdir:
19582 +       au_ren_unlock(a);
19583 +out_children:
19584 +       au_nhash_wh_free(&a->whlist);
19585 +       if (err && a->dst_inode && a->dst_bstart != a->btgt) {
19586 +               AuDbg("bstart %d, btgt %d\n", a->dst_bstart, a->btgt);
19587 +               au_set_h_dptr(a->dst_dentry, a->btgt, NULL);
19588 +               au_set_dbstart(a->dst_dentry, a->dst_bstart);
19589 +       }
19590 +out_parent:
19591 +       if (!err)
19592 +               d_move(a->src_dentry, a->dst_dentry);
19593 +       else {
19594 +               au_update_dbstart(a->dst_dentry);
19595 +               if (!a->dst_inode)
19596 +                       d_drop(a->dst_dentry);
19597 +       }
19598 +       if (au_ftest_ren(a->flags, ISSAMEDIR))
19599 +               di_write_unlock(a->dst_parent);
19600 +       else
19601 +               di_write_unlock2(a->src_parent, a->dst_parent);
19602 +out_unlock:
19603 +       aufs_read_and_write_unlock2(a->dst_dentry, a->src_dentry);
19604 +out_free:
19605 +       iput(a->dst_inode);
19606 +       if (a->thargs)
19607 +               au_whtmp_rmdir_free(a->thargs);
19608 +       kfree(a);
19609 +out:
19610 +       AuTraceErr(err);
19611 +       return err;
19612 +}
19613 diff -urN /usr/share/empty/fs/aufs/Kconfig linux/fs/aufs/Kconfig
19614 --- /usr/share/empty/fs/aufs/Kconfig    1970-01-01 01:00:00.000000000 +0100
19615 +++ linux/fs/aufs/Kconfig       2014-04-24 22:11:10.835268907 +0200
19616 @@ -0,0 +1,179 @@
19617 +config AUFS_FS
19618 +       tristate "Aufs (Advanced multi layered unification filesystem) support"
19619 +       help
19620 +       Aufs is a stackable unification filesystem such as Unionfs,
19621 +       which unifies several directories and provides a merged single
19622 +       directory.
19623 +       In the early days, aufs was entirely re-designed and
19624 +       re-implemented Unionfs Version 1.x series. Introducing many
19625 +       original ideas, approaches and improvements, it becomes totally
19626 +       different from Unionfs while keeping the basic features.
19627 +
19628 +if AUFS_FS
19629 +choice
19630 +       prompt "Maximum number of branches"
19631 +       default AUFS_BRANCH_MAX_127
19632 +       help
19633 +       Specifies the maximum number of branches (or member directories)
19634 +       in a single aufs. The larger value consumes more system
19635 +       resources and has a minor impact to performance.
19636 +config AUFS_BRANCH_MAX_127
19637 +       bool "127"
19638 +       help
19639 +       Specifies the maximum number of branches (or member directories)
19640 +       in a single aufs. The larger value consumes more system
19641 +       resources and has a minor impact to performance.
19642 +config AUFS_BRANCH_MAX_511
19643 +       bool "511"
19644 +       help
19645 +       Specifies the maximum number of branches (or member directories)
19646 +       in a single aufs. The larger value consumes more system
19647 +       resources and has a minor impact to performance.
19648 +config AUFS_BRANCH_MAX_1023
19649 +       bool "1023"
19650 +       help
19651 +       Specifies the maximum number of branches (or member directories)
19652 +       in a single aufs. The larger value consumes more system
19653 +       resources and has a minor impact to performance.
19654 +config AUFS_BRANCH_MAX_32767
19655 +       bool "32767"
19656 +       help
19657 +       Specifies the maximum number of branches (or member directories)
19658 +       in a single aufs. The larger value consumes more system
19659 +       resources and has a minor impact to performance.
19660 +endchoice
19661 +
19662 +config AUFS_SBILIST
19663 +       bool
19664 +       depends on AUFS_MAGIC_SYSRQ || PROC_FS
19665 +       default y
19666 +       help
19667 +       Automatic configuration for internal use.
19668 +       When aufs supports Magic SysRq or /proc, enabled automatically.
19669 +
19670 +config AUFS_HNOTIFY
19671 +       bool "Detect direct branch access (bypassing aufs)"
19672 +       help
19673 +       If you want to modify files on branches directly, eg. bypassing aufs,
19674 +       and want aufs to detect the changes of them fully, then enable this
19675 +       option and use 'udba=notify' mount option.
19676 +       Currently there is only one available configuration, "fsnotify".
19677 +       It will have a negative impact to the performance.
19678 +       See detail in aufs.5.
19679 +
19680 +choice
19681 +       prompt "method" if AUFS_HNOTIFY
19682 +       default AUFS_HFSNOTIFY
19683 +config AUFS_HFSNOTIFY
19684 +       bool "fsnotify"
19685 +       select FSNOTIFY
19686 +endchoice
19687 +
19688 +config AUFS_EXPORT
19689 +       bool "NFS-exportable aufs"
19690 +       depends on EXPORTFS
19691 +       help
19692 +       If you want to export your mounted aufs via NFS, then enable this
19693 +       option. There are several requirements for this configuration.
19694 +       See detail in aufs.5.
19695 +
19696 +config AUFS_INO_T_64
19697 +       bool
19698 +       depends on AUFS_EXPORT
19699 +       depends on 64BIT && !(ALPHA || S390)
19700 +       default y
19701 +       help
19702 +       Automatic configuration for internal use.
19703 +       /* typedef unsigned long/int __kernel_ino_t */
19704 +       /* alpha and s390x are int */
19705 +
19706 +config AUFS_RDU
19707 +       bool "Readdir in userspace"
19708 +       help
19709 +       Aufs has two methods to provide a merged view for a directory,
19710 +       by a user-space library and by kernel-space natively. The latter
19711 +       is always enabled but sometimes large and slow.
19712 +       If you enable this option, install the library in aufs2-util
19713 +       package, and set some environment variables for your readdir(3),
19714 +       then the work will be handled in user-space which generally
19715 +       shows better performance in most cases.
19716 +       See detail in aufs.5.
19717 +
19718 +config AUFS_SP_IATTR
19719 +       bool "Respect the attributes (mtime/ctime mainly) of special files"
19720 +       help
19721 +       When you write something to a special file, some attributes of it
19722 +       (mtime/ctime mainly) may be updated. Generally such updates are
19723 +       less important (actually some device drivers and NFS ignore
19724 +       it). But some applications (such like test program) requires
19725 +       such updates. If you need these updates, then enable this
19726 +       configuration which introduces some overhead.
19727 +       Currently this configuration handles FIFO only.
19728 +
19729 +config AUFS_SHWH
19730 +       bool "Show whiteouts"
19731 +       help
19732 +       If you want to make the whiteouts in aufs visible, then enable
19733 +       this option and specify 'shwh' mount option. Although it may
19734 +       sounds like philosophy or something, but in technically it
19735 +       simply shows the name of whiteout with keeping its behaviour.
19736 +
19737 +config AUFS_BR_RAMFS
19738 +       bool "Ramfs (initramfs/rootfs) as an aufs branch"
19739 +       help
19740 +       If you want to use ramfs as an aufs branch fs, then enable this
19741 +       option. Generally tmpfs is recommended.
19742 +       Aufs prohibited them to be a branch fs by default, because
19743 +       initramfs becomes unusable after switch_root or something
19744 +       generally. If you sets initramfs as an aufs branch and boot your
19745 +       system by switch_root, you will meet a problem easily since the
19746 +       files in initramfs may be inaccessible.
19747 +       Unless you are going to use ramfs as an aufs branch fs without
19748 +       switch_root or something, leave it N.
19749 +
19750 +config AUFS_BR_FUSE
19751 +       bool "Fuse fs as an aufs branch"
19752 +       depends on FUSE_FS
19753 +       select AUFS_POLL
19754 +       help
19755 +       If you want to use fuse-based userspace filesystem as an aufs
19756 +       branch fs, then enable this option.
19757 +       It implements the internal poll(2) operation which is
19758 +       implemented by fuse only (curretnly).
19759 +
19760 +config AUFS_POLL
19761 +       bool
19762 +       help
19763 +       Automatic configuration for internal use.
19764 +
19765 +config AUFS_BR_HFSPLUS
19766 +       bool "Hfsplus as an aufs branch"
19767 +       depends on HFSPLUS_FS
19768 +       default y
19769 +       help
19770 +       If you want to use hfsplus fs as an aufs branch fs, then enable
19771 +       this option. This option introduces a small overhead at
19772 +       copying-up a file on hfsplus.
19773 +
19774 +config AUFS_BDEV_LOOP
19775 +       bool
19776 +       depends on BLK_DEV_LOOP
19777 +       default y
19778 +       help
19779 +       Automatic configuration for internal use.
19780 +       Convert =[ym] into =y.
19781 +
19782 +config AUFS_DEBUG
19783 +       bool "Debug aufs"
19784 +       help
19785 +       Enable this to compile aufs internal debug code.
19786 +       It will have a negative impact to the performance.
19787 +
19788 +config AUFS_MAGIC_SYSRQ
19789 +       bool
19790 +       depends on AUFS_DEBUG && MAGIC_SYSRQ
19791 +       default y
19792 +       help
19793 +       Automatic configuration for internal use.
19794 +       When aufs supports Magic SysRq, enabled automatically.
19795 +endif
19796 diff -urN /usr/share/empty/fs/aufs/loop.c linux/fs/aufs/loop.c
19797 --- /usr/share/empty/fs/aufs/loop.c     1970-01-01 01:00:00.000000000 +0100
19798 +++ linux/fs/aufs/loop.c        2014-04-24 22:11:10.855269116 +0200
19799 @@ -0,0 +1,145 @@
19800 +/*
19801 + * Copyright (C) 2005-2014 Junjiro R. Okajima
19802 + *
19803 + * This program, aufs is free software; you can redistribute it and/or modify
19804 + * it under the terms of the GNU General Public License as published by
19805 + * the Free Software Foundation; either version 2 of the License, or
19806 + * (at your option) any later version.
19807 + *
19808 + * This program is distributed in the hope that it will be useful,
19809 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19810 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19811 + * GNU General Public License for more details.
19812 + *
19813 + * You should have received a copy of the GNU General Public License
19814 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19815 + */
19816 +
19817 +/*
19818 + * support for loopback block device as a branch
19819 + */
19820 +
19821 +#include "aufs.h"
19822 +
19823 +/* added into drivers/block/loop.c */
19824 +static struct file *(*backing_file_func)(struct super_block *sb);
19825 +
19826 +/*
19827 + * test if two lower dentries have overlapping branches.
19828 + */
19829 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding)
19830 +{
19831 +       struct super_block *h_sb;
19832 +       struct file *backing_file;
19833 +
19834 +       if (unlikely(!backing_file_func)) {
19835 +               /* don't load "loop" module here */
19836 +               backing_file_func = symbol_get(loop_backing_file);
19837 +               if (unlikely(!backing_file_func))
19838 +                       /* "loop" module is not loaded */
19839 +                       return 0;
19840 +       }
19841 +
19842 +       h_sb = h_adding->d_sb;
19843 +       backing_file = backing_file_func(h_sb);
19844 +       if (!backing_file)
19845 +               return 0;
19846 +
19847 +       h_adding = backing_file->f_dentry;
19848 +       /*
19849 +        * h_adding can be local NFS.
19850 +        * in this case aufs cannot detect the loop.
19851 +        */
19852 +       if (unlikely(h_adding->d_sb == sb))
19853 +               return 1;
19854 +       return !!au_test_subdir(h_adding, sb->s_root);
19855 +}
19856 +
19857 +/* true if a kernel thread named 'loop[0-9].*' accesses a file */
19858 +int au_test_loopback_kthread(void)
19859 +{
19860 +       int ret;
19861 +       struct task_struct *tsk = current;
19862 +       char c, comm[sizeof(tsk->comm)];
19863 +
19864 +       ret = 0;
19865 +       if (tsk->flags & PF_KTHREAD) {
19866 +               get_task_comm(comm, tsk);
19867 +               c = comm[4];
19868 +               ret = ('0' <= c && c <= '9'
19869 +                      && !strncmp(comm, "loop", 4));
19870 +       }
19871 +
19872 +       return ret;
19873 +}
19874 +
19875 +/* ---------------------------------------------------------------------- */
19876 +
19877 +#define au_warn_loopback_step  16
19878 +static int au_warn_loopback_nelem = au_warn_loopback_step;
19879 +static unsigned long *au_warn_loopback_array;
19880 +
19881 +void au_warn_loopback(struct super_block *h_sb)
19882 +{
19883 +       int i, new_nelem;
19884 +       unsigned long *a, magic;
19885 +       static DEFINE_SPINLOCK(spin);
19886 +
19887 +       magic = h_sb->s_magic;
19888 +       spin_lock(&spin);
19889 +       a = au_warn_loopback_array;
19890 +       for (i = 0; i < au_warn_loopback_nelem && *a; i++)
19891 +               if (a[i] == magic) {
19892 +                       spin_unlock(&spin);
19893 +                       return;
19894 +               }
19895 +
19896 +       /* h_sb is new to us, print it */
19897 +       if (i < au_warn_loopback_nelem) {
19898 +               a[i] = magic;
19899 +               goto pr;
19900 +       }
19901 +
19902 +       /* expand the array */
19903 +       new_nelem = au_warn_loopback_nelem + au_warn_loopback_step;
19904 +       a = au_kzrealloc(au_warn_loopback_array,
19905 +                        au_warn_loopback_nelem * sizeof(unsigned long),
19906 +                        new_nelem * sizeof(unsigned long), GFP_ATOMIC);
19907 +       if (a) {
19908 +               au_warn_loopback_nelem = new_nelem;
19909 +               au_warn_loopback_array = a;
19910 +               a[i] = magic;
19911 +               goto pr;
19912 +       }
19913 +
19914 +       spin_unlock(&spin);
19915 +       AuWarn1("realloc failed, ignored\n");
19916 +       return;
19917 +
19918 +pr:
19919 +       spin_unlock(&spin);
19920 +       pr_warn("you may want to try another patch for loopback file "
19921 +               "on %s(0x%lx) branch\n", au_sbtype(h_sb), magic);
19922 +}
19923 +
19924 +int au_loopback_init(void)
19925 +{
19926 +       int err;
19927 +       struct super_block *sb __maybe_unused;
19928 +
19929 +       AuDebugOn(sizeof(sb->s_magic) != sizeof(unsigned long));
19930 +
19931 +       err = 0;
19932 +       au_warn_loopback_array = kcalloc(au_warn_loopback_step,
19933 +                                        sizeof(unsigned long), GFP_NOFS);
19934 +       if (unlikely(!au_warn_loopback_array))
19935 +               err = -ENOMEM;
19936 +
19937 +       return err;
19938 +}
19939 +
19940 +void au_loopback_fin(void)
19941 +{
19942 +       symbol_put(loop_backing_file);
19943 +       kfree(au_warn_loopback_array);
19944 +}
19945 diff -urN /usr/share/empty/fs/aufs/loop.h linux/fs/aufs/loop.h
19946 --- /usr/share/empty/fs/aufs/loop.h     1970-01-01 01:00:00.000000000 +0100
19947 +++ linux/fs/aufs/loop.h        2014-04-24 22:11:10.855269116 +0200
19948 @@ -0,0 +1,52 @@
19949 +/*
19950 + * Copyright (C) 2005-2014 Junjiro R. Okajima
19951 + *
19952 + * This program, aufs is free software; you can redistribute it and/or modify
19953 + * it under the terms of the GNU General Public License as published by
19954 + * the Free Software Foundation; either version 2 of the License, or
19955 + * (at your option) any later version.
19956 + *
19957 + * This program is distributed in the hope that it will be useful,
19958 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
19959 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19960 + * GNU General Public License for more details.
19961 + *
19962 + * You should have received a copy of the GNU General Public License
19963 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19964 + */
19965 +
19966 +/*
19967 + * support for loopback mount as a branch
19968 + */
19969 +
19970 +#ifndef __AUFS_LOOP_H__
19971 +#define __AUFS_LOOP_H__
19972 +
19973 +#ifdef __KERNEL__
19974 +
19975 +struct dentry;
19976 +struct super_block;
19977 +
19978 +#ifdef CONFIG_AUFS_BDEV_LOOP
19979 +/* drivers/block/loop.c */
19980 +struct file *loop_backing_file(struct super_block *sb);
19981 +
19982 +/* loop.c */
19983 +int au_test_loopback_overlap(struct super_block *sb, struct dentry *h_adding);
19984 +int au_test_loopback_kthread(void);
19985 +void au_warn_loopback(struct super_block *h_sb);
19986 +
19987 +int au_loopback_init(void);
19988 +void au_loopback_fin(void);
19989 +#else
19990 +AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
19991 +          struct dentry *h_adding)
19992 +AuStubInt0(au_test_loopback_kthread, void)
19993 +AuStubVoid(au_warn_loopback, struct super_block *h_sb)
19994 +
19995 +AuStubInt0(au_loopback_init, void)
19996 +AuStubVoid(au_loopback_fin, void)
19997 +#endif /* BLK_DEV_LOOP */
19998 +
19999 +#endif /* __KERNEL__ */
20000 +#endif /* __AUFS_LOOP_H__ */
20001 diff -urN /usr/share/empty/fs/aufs/magic.mk linux/fs/aufs/magic.mk
20002 --- /usr/share/empty/fs/aufs/magic.mk   1970-01-01 01:00:00.000000000 +0100
20003 +++ linux/fs/aufs/magic.mk      2012-10-05 20:40:23.980965955 +0200
20004 @@ -0,0 +1,54 @@
20005 +
20006 +# defined in ${srctree}/fs/fuse/inode.c
20007 +# tristate
20008 +ifdef CONFIG_FUSE_FS
20009 +ccflags-y += -DFUSE_SUPER_MAGIC=0x65735546
20010 +endif
20011 +
20012 +# defined in ${srctree}/fs/ocfs2/ocfs2_fs.h
20013 +# tristate
20014 +ifdef CONFIG_OCFS2_FS
20015 +ccflags-y += -DOCFS2_SUPER_MAGIC=0x7461636f
20016 +endif
20017 +
20018 +# defined in ${srctree}/fs/ocfs2/dlm/userdlm.h
20019 +# tristate
20020 +ifdef CONFIG_OCFS2_FS_O2CB
20021 +ccflags-y += -DDLMFS_MAGIC=0x76a9f425
20022 +endif
20023 +
20024 +# defined in ${srctree}/fs/cifs/cifsfs.c
20025 +# tristate
20026 +ifdef CONFIG_CIFS_FS
20027 +ccflags-y += -DCIFS_MAGIC_NUMBER=0xFF534D42
20028 +endif
20029 +
20030 +# defined in ${srctree}/fs/xfs/xfs_sb.h
20031 +# tristate
20032 +ifdef CONFIG_XFS_FS
20033 +ccflags-y += -DXFS_SB_MAGIC=0x58465342
20034 +endif
20035 +
20036 +# defined in ${srctree}/fs/configfs/mount.c
20037 +# tristate
20038 +ifdef CONFIG_CONFIGFS_FS
20039 +ccflags-y += -DCONFIGFS_MAGIC=0x62656570
20040 +endif
20041 +
20042 +# defined in ${srctree}/fs/9p/v9fs.h
20043 +# tristate
20044 +ifdef CONFIG_9P_FS
20045 +ccflags-y += -DV9FS_MAGIC=0x01021997
20046 +endif
20047 +
20048 +# defined in ${srctree}/fs/ubifs/ubifs.h
20049 +# tristate
20050 +ifdef CONFIG_UBIFS_FS
20051 +ccflags-y += -DUBIFS_SUPER_MAGIC=0x24051905
20052 +endif
20053 +
20054 +# defined in ${srctree}/fs/hfsplus/hfsplus_raw.h
20055 +# tristate
20056 +ifdef CONFIG_HFSPLUS_FS
20057 +ccflags-y += -DHFSPLUS_SUPER_MAGIC=0x482b
20058 +endif
20059 diff -urN /usr/share/empty/fs/aufs/Makefile linux/fs/aufs/Makefile
20060 --- /usr/share/empty/fs/aufs/Makefile   1970-01-01 01:00:00.000000000 +0100
20061 +++ linux/fs/aufs/Makefile      2014-04-24 22:11:10.845269011 +0200
20062 @@ -0,0 +1,42 @@
20063 +
20064 +include ${src}/magic.mk
20065 +ifeq (${CONFIG_AUFS_FS},m)
20066 +include ${src}/conf.mk
20067 +endif
20068 +-include ${src}/priv_def.mk
20069 +
20070 +# cf. include/linux/kernel.h
20071 +# enable pr_debug
20072 +ccflags-y += -DDEBUG
20073 +# sparse requires the full pathname
20074 +ifdef M
20075 +ccflags-y += -include ${M}/../../include/uapi/linux/aufs_type.h
20076 +else
20077 +ccflags-y += -include ${srctree}/include/uapi/linux/aufs_type.h
20078 +endif
20079 +
20080 +obj-$(CONFIG_AUFS_FS) += aufs.o
20081 +aufs-y := module.o sbinfo.o super.o branch.o xino.o sysaufs.o opts.o \
20082 +       wkq.o vfsub.o dcsub.o \
20083 +       cpup.o whout.o wbr_policy.o \
20084 +       dinfo.o dentry.o \
20085 +       dynop.o \
20086 +       finfo.o file.o f_op.o \
20087 +       dir.o vdir.o \
20088 +       iinfo.o inode.o i_op.o i_op_add.o i_op_del.o i_op_ren.o \
20089 +       mvdown.o ioctl.o
20090 +
20091 +# all are boolean
20092 +aufs-$(CONFIG_PROC_FS) += procfs.o plink.o
20093 +aufs-$(CONFIG_SYSFS) += sysfs.o
20094 +aufs-$(CONFIG_DEBUG_FS) += dbgaufs.o
20095 +aufs-$(CONFIG_AUFS_BDEV_LOOP) += loop.o
20096 +aufs-$(CONFIG_AUFS_HNOTIFY) += hnotify.o
20097 +aufs-$(CONFIG_AUFS_HFSNOTIFY) += hfsnotify.o
20098 +aufs-$(CONFIG_AUFS_EXPORT) += export.o
20099 +aufs-$(CONFIG_AUFS_POLL) += poll.o
20100 +aufs-$(CONFIG_AUFS_RDU) += rdu.o
20101 +aufs-$(CONFIG_AUFS_SP_IATTR) += f_op_sp.o
20102 +aufs-$(CONFIG_AUFS_BR_HFSPLUS) += hfsplus.o
20103 +aufs-$(CONFIG_AUFS_DEBUG) += debug.o
20104 +aufs-$(CONFIG_AUFS_MAGIC_SYSRQ) += sysrq.o
20105 diff -urN /usr/share/empty/fs/aufs/module.c linux/fs/aufs/module.c
20106 --- /usr/share/empty/fs/aufs/module.c   1970-01-01 01:00:00.000000000 +0100
20107 +++ linux/fs/aufs/module.c      2014-04-24 22:11:10.855269116 +0200
20108 @@ -0,0 +1,202 @@
20109 +/*
20110 + * Copyright (C) 2005-2014 Junjiro R. Okajima
20111 + *
20112 + * This program, aufs is free software; you can redistribute it and/or modify
20113 + * it under the terms of the GNU General Public License as published by
20114 + * the Free Software Foundation; either version 2 of the License, or
20115 + * (at your option) any later version.
20116 + *
20117 + * This program is distributed in the hope that it will be useful,
20118 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
20119 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20120 + * GNU General Public License for more details.
20121 + *
20122 + * You should have received a copy of the GNU General Public License
20123 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20124 + */
20125 +
20126 +/*
20127 + * module global variables and operations
20128 + */
20129 +
20130 +#include <linux/module.h>
20131 +#include <linux/seq_file.h>
20132 +#include "aufs.h"
20133 +
20134 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp)
20135 +{
20136 +       if (new_sz <= nused)
20137 +               return p;
20138 +
20139 +       p = krealloc(p, new_sz, gfp);
20140 +       if (p)
20141 +               memset(p + nused, 0, new_sz - nused);
20142 +       return p;
20143 +}
20144 +
20145 +/* ---------------------------------------------------------------------- */
20146 +
20147 +/*
20148 + * aufs caches
20149 + */
20150 +struct kmem_cache *au_cachep[AuCache_Last];
20151 +static int __init au_cache_init(void)
20152 +{
20153 +       au_cachep[AuCache_DINFO] = AuCacheCtor(au_dinfo, au_di_init_once);
20154 +       if (au_cachep[AuCache_DINFO])
20155 +               /* SLAB_DESTROY_BY_RCU */
20156 +               au_cachep[AuCache_ICNTNR] = AuCacheCtor(au_icntnr,
20157 +                                                       au_icntnr_init_once);
20158 +       if (au_cachep[AuCache_ICNTNR])
20159 +               au_cachep[AuCache_FINFO] = AuCacheCtor(au_finfo,
20160 +                                                      au_fi_init_once);
20161 +       if (au_cachep[AuCache_FINFO])
20162 +               au_cachep[AuCache_VDIR] = AuCache(au_vdir);
20163 +       if (au_cachep[AuCache_VDIR])
20164 +               au_cachep[AuCache_DEHSTR] = AuCache(au_vdir_dehstr);
20165 +       if (au_cachep[AuCache_DEHSTR])
20166 +               return 0;
20167 +
20168 +       return -ENOMEM;
20169 +}
20170 +
20171 +static void au_cache_fin(void)
20172 +{
20173 +       int i;
20174 +
20175 +       /*
20176 +        * Make sure all delayed rcu free inodes are flushed before we
20177 +        * destroy cache.
20178 +        */
20179 +       rcu_barrier();
20180 +
20181 +       /* excluding AuCache_HNOTIFY */
20182 +       BUILD_BUG_ON(AuCache_HNOTIFY + 1 != AuCache_Last);
20183 +       for (i = 0; i < AuCache_HNOTIFY; i++)
20184 +               if (au_cachep[i]) {
20185 +                       kmem_cache_destroy(au_cachep[i]);
20186 +                       au_cachep[i] = NULL;
20187 +               }
20188 +}
20189 +
20190 +/* ---------------------------------------------------------------------- */
20191 +
20192 +int au_dir_roflags;
20193 +
20194 +#ifdef CONFIG_AUFS_SBILIST
20195 +/*
20196 + * iterate_supers_type() doesn't protect us from
20197 + * remounting (branch management)
20198 + */
20199 +struct au_splhead au_sbilist;
20200 +#endif
20201 +
20202 +struct lock_class_key au_lc_key[AuLcKey_Last];
20203 +
20204 +/*
20205 + * functions for module interface.
20206 + */
20207 +MODULE_LICENSE("GPL");
20208 +/* MODULE_LICENSE("GPL v2"); */
20209 +MODULE_AUTHOR("Junjiro R. Okajima <aufs-users@lists.sourceforge.net>");
20210 +MODULE_DESCRIPTION(AUFS_NAME
20211 +       " -- Advanced multi layered unification filesystem");
20212 +MODULE_VERSION(AUFS_VERSION);
20213 +MODULE_ALIAS_FS(AUFS_NAME);
20214 +
20215 +/* this module parameter has no meaning when SYSFS is disabled */
20216 +int sysaufs_brs = 1;
20217 +MODULE_PARM_DESC(brs, "use <sysfs>/fs/aufs/si_*/brN");
20218 +module_param_named(brs, sysaufs_brs, int, S_IRUGO);
20219 +
20220 +/* ---------------------------------------------------------------------- */
20221 +
20222 +static char au_esc_chars[0x20 + 3]; /* 0x01-0x20, backslash, del, and NULL */
20223 +
20224 +int au_seq_path(struct seq_file *seq, struct path *path)
20225 +{
20226 +       return seq_path(seq, path, au_esc_chars);
20227 +}
20228 +
20229 +/* ---------------------------------------------------------------------- */
20230 +
20231 +static int __init aufs_init(void)
20232 +{
20233 +       int err, i;
20234 +       char *p;
20235 +
20236 +       p = au_esc_chars;
20237 +       for (i = 1; i <= ' '; i++)
20238 +               *p++ = i;
20239 +       *p++ = '\\';
20240 +       *p++ = '\x7f';
20241 +       *p = 0;
20242 +
20243 +       au_dir_roflags = au_file_roflags(O_DIRECTORY | O_LARGEFILE);
20244 +
20245 +       au_sbilist_init();
20246 +       sysaufs_brs_init();
20247 +       au_debug_init();
20248 +       au_dy_init();
20249 +       err = sysaufs_init();
20250 +       if (unlikely(err))
20251 +               goto out;
20252 +       err = au_procfs_init();
20253 +       if (unlikely(err))
20254 +               goto out_sysaufs;
20255 +       err = au_wkq_init();
20256 +       if (unlikely(err))
20257 +               goto out_procfs;
20258 +       err = au_loopback_init();
20259 +       if (unlikely(err))
20260 +               goto out_wkq;
20261 +       err = au_hnotify_init();
20262 +       if (unlikely(err))
20263 +               goto out_loopback;
20264 +       err = au_sysrq_init();
20265 +       if (unlikely(err))
20266 +               goto out_hin;
20267 +       err = au_cache_init();
20268 +       if (unlikely(err))
20269 +               goto out_sysrq;
20270 +       err = register_filesystem(&aufs_fs_type);
20271 +       if (unlikely(err))
20272 +               goto out_cache;
20273 +       /* since we define pr_fmt, call printk directly */
20274 +       printk(KERN_INFO AUFS_NAME " " AUFS_VERSION "\n");
20275 +       goto out; /* success */
20276 +
20277 +out_cache:
20278 +       au_cache_fin();
20279 +out_sysrq:
20280 +       au_sysrq_fin();
20281 +out_hin:
20282 +       au_hnotify_fin();
20283 +out_loopback:
20284 +       au_loopback_fin();
20285 +out_wkq:
20286 +       au_wkq_fin();
20287 +out_procfs:
20288 +       au_procfs_fin();
20289 +out_sysaufs:
20290 +       sysaufs_fin();
20291 +       au_dy_fin();
20292 +out:
20293 +       return err;
20294 +}
20295 +
20296 +static void __exit aufs_exit(void)
20297 +{
20298 +       unregister_filesystem(&aufs_fs_type);
20299 +       au_cache_fin();
20300 +       au_sysrq_fin();
20301 +       au_hnotify_fin();
20302 +       au_loopback_fin();
20303 +       au_wkq_fin();
20304 +       au_procfs_fin();
20305 +       sysaufs_fin();
20306 +       au_dy_fin();
20307 +}
20308 +
20309 +module_init(aufs_init);
20310 +module_exit(aufs_exit);
20311 diff -urN /usr/share/empty/fs/aufs/module.h linux/fs/aufs/module.h
20312 --- /usr/share/empty/fs/aufs/module.h   1970-01-01 01:00:00.000000000 +0100
20313 +++ linux/fs/aufs/module.h      2014-04-24 22:11:10.855269116 +0200
20314 @@ -0,0 +1,104 @@
20315 +/*
20316 + * Copyright (C) 2005-2014 Junjiro R. Okajima
20317 + *
20318 + * This program, aufs is free software; you can redistribute it and/or modify
20319 + * it under the terms of the GNU General Public License as published by
20320 + * the Free Software Foundation; either version 2 of the License, or
20321 + * (at your option) any later version.
20322 + *
20323 + * This program is distributed in the hope that it will be useful,
20324 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
20325 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20326 + * GNU General Public License for more details.
20327 + *
20328 + * You should have received a copy of the GNU General Public License
20329 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20330 + */
20331 +
20332 +/*
20333 + * module initialization and module-global
20334 + */
20335 +
20336 +#ifndef __AUFS_MODULE_H__
20337 +#define __AUFS_MODULE_H__
20338 +
20339 +#ifdef __KERNEL__
20340 +
20341 +#include <linux/slab.h>
20342 +
20343 +struct path;
20344 +struct seq_file;
20345 +
20346 +/* module parameters */
20347 +extern int sysaufs_brs;
20348 +
20349 +/* ---------------------------------------------------------------------- */
20350 +
20351 +extern int au_dir_roflags;
20352 +
20353 +enum {
20354 +       AuLcNonDir_FIINFO,
20355 +       AuLcNonDir_DIINFO,
20356 +       AuLcNonDir_IIINFO,
20357 +
20358 +       AuLcDir_FIINFO,
20359 +       AuLcDir_DIINFO,
20360 +       AuLcDir_IIINFO,
20361 +
20362 +       AuLcSymlink_DIINFO,
20363 +       AuLcSymlink_IIINFO,
20364 +
20365 +       AuLcKey_Last
20366 +};
20367 +extern struct lock_class_key au_lc_key[AuLcKey_Last];
20368 +
20369 +void *au_kzrealloc(void *p, unsigned int nused, unsigned int new_sz, gfp_t gfp);
20370 +int au_seq_path(struct seq_file *seq, struct path *path);
20371 +
20372 +#ifdef CONFIG_PROC_FS
20373 +/* procfs.c */
20374 +int __init au_procfs_init(void);
20375 +void au_procfs_fin(void);
20376 +#else
20377 +AuStubInt0(au_procfs_init, void);
20378 +AuStubVoid(au_procfs_fin, void);
20379 +#endif
20380 +
20381 +/* ---------------------------------------------------------------------- */
20382 +
20383 +/* kmem cache */
20384 +enum {
20385 +       AuCache_DINFO,
20386 +       AuCache_ICNTNR,
20387 +       AuCache_FINFO,
20388 +       AuCache_VDIR,
20389 +       AuCache_DEHSTR,
20390 +       AuCache_HNOTIFY, /* must be last */
20391 +       AuCache_Last
20392 +};
20393 +
20394 +#define AuCacheFlags           (SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD)
20395 +#define AuCache(type)          KMEM_CACHE(type, AuCacheFlags)
20396 +#define AuCacheCtor(type, ctor)        \
20397 +       kmem_cache_create(#type, sizeof(struct type), \
20398 +                         __alignof__(struct type), AuCacheFlags, ctor)
20399 +
20400 +extern struct kmem_cache *au_cachep[];
20401 +
20402 +#define AuCacheFuncs(name, index) \
20403 +static inline struct au_##name *au_cache_alloc_##name(void) \
20404 +{ return kmem_cache_alloc(au_cachep[AuCache_##index], GFP_NOFS); } \
20405 +static inline void au_cache_free_##name(struct au_##name *p) \
20406 +{ kmem_cache_free(au_cachep[AuCache_##index], p); }
20407 +
20408 +AuCacheFuncs(dinfo, DINFO);
20409 +AuCacheFuncs(icntnr, ICNTNR);
20410 +AuCacheFuncs(finfo, FINFO);
20411 +AuCacheFuncs(vdir, VDIR);
20412 +AuCacheFuncs(vdir_dehstr, DEHSTR);
20413 +#ifdef CONFIG_AUFS_HNOTIFY
20414 +AuCacheFuncs(hnotify, HNOTIFY);
20415 +#endif
20416 +
20417 +#endif /* __KERNEL__ */
20418 +#endif /* __AUFS_MODULE_H__ */
20419 diff -urN /usr/share/empty/fs/aufs/mvdown.c linux/fs/aufs/mvdown.c
20420 --- /usr/share/empty/fs/aufs/mvdown.c   1970-01-01 01:00:00.000000000 +0100
20421 +++ linux/fs/aufs/mvdown.c      2014-04-24 22:11:10.855269116 +0200
20422 @@ -0,0 +1,627 @@
20423 +/*
20424 + * Copyright (C) 2011-2014 Junjiro R. Okajima
20425 + *
20426 + * This program, aufs is free software; you can redistribute it and/or modify
20427 + * it under the terms of the GNU General Public License as published by
20428 + * the Free Software Foundation; either version 2 of the License, or
20429 + * (at your option) any later version.
20430 + *
20431 + * This program is distributed in the hope that it will be useful,
20432 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
20433 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
20434 + * GNU General Public License for more details.
20435 + *
20436 + * You should have received a copy of the GNU General Public License
20437 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20438 + */
20439 +
20440 +/*
20441 + * move-down, opposite of copy-up
20442 + */
20443 +
20444 +#include "aufs.h"
20445 +
20446 +struct au_mvd_args {
20447 +       struct {
20448 +               struct super_block *h_sb;
20449 +               struct dentry *h_parent;
20450 +               struct au_hinode *hdir;
20451 +               struct inode *h_dir, *h_inode;
20452 +       } info[AUFS_MVDOWN_NARRAY];
20453 +
20454 +       struct aufs_mvdown mvdown;
20455 +       struct dentry *dentry, *parent;
20456 +       struct inode *inode, *dir;
20457 +       struct super_block *sb;
20458 +       aufs_bindex_t bopq, bwh, bfound;
20459 +       unsigned char rename_lock;
20460 +       struct au_pin pin;
20461 +};
20462 +
20463 +#define mvd_errno              mvdown.au_errno
20464 +#define mvd_bsrc               mvdown.a[AUFS_MVDOWN_UPPER].bindex
20465 +#define mvd_src_brid           mvdown.a[AUFS_MVDOWN_UPPER].brid
20466 +#define mvd_bdst               mvdown.a[AUFS_MVDOWN_LOWER].bindex
20467 +#define mvd_dst_brid           mvdown.a[AUFS_MVDOWN_LOWER].brid
20468 +
20469 +#define mvd_h_src_sb           info[AUFS_MVDOWN_UPPER].h_sb
20470 +#define mvd_h_src_parent       info[AUFS_MVDOWN_UPPER].h_parent
20471 +#define mvd_hdir_src           info[AUFS_MVDOWN_UPPER].hdir
20472 +#define mvd_h_src_dir          info[AUFS_MVDOWN_UPPER].h_dir
20473 +#define mvd_h_src_inode                info[AUFS_MVDOWN_UPPER].h_inode
20474 +
20475 +#define mvd_h_dst_sb           info[AUFS_MVDOWN_LOWER].h_sb
20476 +#define mvd_h_dst_parent       info[AUFS_MVDOWN_LOWER].h_parent
20477 +#define mvd_hdir_dst           info[AUFS_MVDOWN_LOWER].hdir
20478 +#define mvd_h_dst_dir          info[AUFS_MVDOWN_LOWER].h_dir
20479 +#define mvd_h_dst_inode                info[AUFS_MVDOWN_LOWER].h_inode
20480 +
20481 +#define AU_MVD_PR(flag, ...) do {                      \
20482 +               if (flag)                               \
20483 +                       pr_err(__VA_ARGS__);            \
20484 +       } while (0)
20485 +
20486 +/* make the parent dir on bdst */
20487 +static int au_do_mkdir(const unsigned char dmsg, struct au_mvd_args *a)
20488 +{
20489 +       int err;
20490 +
20491 +       err = 0;
20492 +       a->mvd_hdir_src = au_hi(a->dir, a->mvd_bsrc);
20493 +       a->mvd_hdir_dst = au_hi(a->dir, a->mvd_bdst);
20494 +       a->mvd_h_src_parent = au_h_dptr(a->parent, a->mvd_bsrc);
20495 +       a->mvd_h_dst_parent = NULL;
20496 +       if (au_dbend(a->parent) >= a->mvd_bdst)
20497 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
20498 +       if (!a->mvd_h_dst_parent) {
20499 +               err = au_cpdown_dirs(a->dentry, a->mvd_bdst);
20500 +               if (unlikely(err)) {
20501 +                       AU_MVD_PR(dmsg, "cpdown_dirs failed\n");
20502 +                       goto out;
20503 +               }
20504 +               a->mvd_h_dst_parent = au_h_dptr(a->parent, a->mvd_bdst);
20505 +       }
20506 +
20507 +out:
20508 +       AuTraceErr(err);
20509 +       return err;
20510 +}
20511 +
20512 +/* lock them all */
20513 +static int au_do_lock(const unsigned char dmsg, struct au_mvd_args *a)
20514 +{
20515 +       int err;
20516 +       struct dentry *h_trap;
20517 +
20518 +       a->mvd_h_src_sb = au_sbr_sb(a->sb, a->mvd_bsrc);
20519 +       a->mvd_h_dst_sb = au_sbr_sb(a->sb, a->mvd_bdst);
20520 +       if (a->mvd_h_src_sb != a->mvd_h_dst_sb) {
20521 +               a->rename_lock = 0;
20522 +               err = au_pin(&a->pin, a->dentry, a->mvd_bdst,
20523 +                            au_opt_udba(a->sb),
20524 +                            AuPin_MNT_WRITE | AuPin_DI_LOCKED);
20525 +               if (!err) {
20526 +                       a->mvd_h_src_dir = a->mvd_h_src_parent->d_inode;
20527 +                       mutex_lock_nested(&a->mvd_h_src_dir->i_mutex,
20528 +                                         AuLsc_I_PARENT3);
20529 +               } else
20530 +                       AU_MVD_PR(dmsg, "pin failed\n");
20531 +               goto out;
20532 +       }
20533 +
20534 +       err = 0;
20535 +       a->rename_lock = 1;
20536 +       h_trap = vfsub_lock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
20537 +                                  a->mvd_h_dst_parent, a->mvd_hdir_dst);
20538 +       if (h_trap) {
20539 +               err = (h_trap != a->mvd_h_src_parent);
20540 +               if (err)
20541 +                       err = (h_trap != a->mvd_h_dst_parent);
20542 +       }
20543 +       BUG_ON(err); /* it should never happen */
20544 +
20545 +out:
20546 +       AuTraceErr(err);
20547 +       return err;
20548 +}
20549 +
20550 +static void au_do_unlock(const unsigned char dmsg, struct au_mvd_args *a)
20551 +{
20552 +       if (!a->rename_lock) {
20553 +               mutex_unlock(&a->mvd_h_src_dir->i_mutex);
20554 +               au_unpin(&a->pin);
20555 +       } else
20556 +               vfsub_unlock_rename(a->mvd_h_src_parent, a->mvd_hdir_src,
20557 +                                   a->mvd_h_dst_parent, a->mvd_hdir_dst);
20558 +}
20559 +
20560 +/* copy-down the file */
20561 +static int au_do_cpdown(const unsigned char dmsg, struct au_mvd_args *a)
20562 +{
20563 +       int err;
20564 +       struct au_cp_generic cpg = {
20565 +               .dentry = a->dentry,
20566 +               .bdst   = a->mvd_bdst,
20567 +               .bsrc   = a->mvd_bsrc,
20568 +               .len    = -1,
20569 +               .pin    = &a->pin,
20570 +               .flags  = AuCpup_DTIME | AuCpup_HOPEN
20571 +       };
20572 +
20573 +       AuDbg("b%d, b%d\n", cpg.bsrc, cpg.bdst);
20574 +       if (a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
20575 +               au_fset_cpup(cpg.flags, OVERWRITE);
20576 +       if (a->mvdown.flags & AUFS_MVDOWN_ROLOWER)
20577 +               au_fset_cpup(cpg.flags, RWDST);
20578 +       err = au_sio_cpdown_simple(&cpg);
20579 +       if (unlikely(err))
20580 +               AU_MVD_PR(dmsg, "cpdown failed\n");
20581 +
20582 +       AuTraceErr(err);
20583 +       return err;
20584 +}
20585 +
20586 +/*
20587 + * unlink the whiteout on bdst if exist which may be created by UDBA while we
20588 + * were sleeping
20589 + */
20590 +static int au_do_unlink_wh(const unsigned char dmsg, struct au_mvd_args *a)
20591 +{
20592 +       int err;
20593 +       struct path h_path;
20594 +       struct au_branch *br;
20595 +       struct inode *delegated;
20596 +
20597 +       br = au_sbr(a->sb, a->mvd_bdst);
20598 +       h_path.dentry = au_wh_lkup(a->mvd_h_dst_parent, &a->dentry->d_name, br);
20599 +       err = PTR_ERR(h_path.dentry);
20600 +       if (IS_ERR(h_path.dentry)) {
20601 +               AU_MVD_PR(dmsg, "wh_lkup failed\n");
20602 +               goto out;
20603 +       }
20604 +
20605 +       err = 0;
20606 +       if (h_path.dentry->d_inode) {
20607 +               h_path.mnt = au_br_mnt(br);
20608 +               delegated = NULL;
20609 +               err = vfsub_unlink(a->mvd_h_dst_parent->d_inode, &h_path,
20610 +                                  &delegated, /*force*/0);
20611 +               if (unlikely(err == -EWOULDBLOCK)) {
20612 +                       pr_warn("cannot retry for NFSv4 delegation"
20613 +                               " for an internal unlink\n");
20614 +                       iput(delegated);
20615 +               }
20616 +               if (unlikely(err))
20617 +                       AU_MVD_PR(dmsg, "wh_unlink failed\n");
20618 +       }
20619 +       dput(h_path.dentry);
20620 +
20621 +out:
20622 +       AuTraceErr(err);
20623 +       return err;
20624 +}
20625 +
20626 +/*
20627 + * unlink the topmost h_dentry
20628 + * Note: the target file MAY be modified by UDBA between this mutex_unlock() and
20629 + *     mutex_lock() in vfs_unlink(). in this case, such changes may be lost.
20630 + */
20631 +static int au_do_unlink(const unsigned char dmsg, struct au_mvd_args *a)
20632 +{
20633 +       int err;
20634 +       struct path h_path;
20635 +       struct inode *delegated;
20636 +
20637 +       h_path.mnt = au_sbr_mnt(a->sb, a->mvd_bsrc);
20638 +       h_path.dentry = au_h_dptr(a->dentry, a->mvd_bsrc);
20639 +       delegated = NULL;
20640 +       err = vfsub_unlink(a->mvd_h_src_dir, &h_path, &delegated, /*force*/0);
20641 +       if (unlikely(err == -EWOULDBLOCK)) {
20642 +               pr_warn("cannot retry for NFSv4 delegation"
20643 +                       " for an internal unlink\n");
20644 +               iput(delegated);
20645 +       }
20646 +       if (unlikely(err))
20647 +               AU_MVD_PR(dmsg, "unlink failed\n");
20648 +
20649 +       AuTraceErr(err);
20650 +       return err;
20651 +}
20652 +
20653 +/*
20654 + * copy-down the file and unlink the bsrc file.
20655 + * - unlink the bdst whout if exist
20656 + * - copy-down the file (with whtmp name and rename)
20657 + * - unlink the bsrc file
20658 + */
20659 +static int au_do_mvdown(const unsigned char dmsg, struct au_mvd_args *a)
20660 +{
20661 +       int err;
20662 +
20663 +       err = au_do_mkdir(dmsg, a);
20664 +       if (!err)
20665 +               err = au_do_lock(dmsg, a);
20666 +       if (unlikely(err))
20667 +               goto out;
20668 +
20669 +       /*
20670 +        * do not revert the activities we made on bdst since they should be
20671 +        * harmless in aufs.
20672 +        */
20673 +
20674 +       err = au_do_cpdown(dmsg, a);
20675 +       if (!err)
20676 +               err = au_do_unlink_wh(dmsg, a);
20677 +       if (!err && !(a->mvdown.flags & AUFS_MVDOWN_KUPPER))
20678 +               err = au_do_unlink(dmsg, a);
20679 +       if (unlikely(err))
20680 +               goto out_unlock;
20681 +
20682 +       /* maintain internal array */
20683 +       if (!(a->mvdown.flags & AUFS_MVDOWN_KUPPER)) {
20684 +               au_set_h_dptr(a->dentry, a->mvd_bsrc, NULL);
20685 +               au_set_dbstart(a->dentry, a->mvd_bdst);
20686 +               au_set_h_iptr(a->inode, a->mvd_bsrc, NULL, /*flags*/0);
20687 +               au_set_ibstart(a->inode, a->mvd_bdst);
20688 +       }
20689 +       if (au_dbend(a->dentry) < a->mvd_bdst)
20690 +               au_set_dbend(a->dentry, a->mvd_bdst);
20691 +       if (au_ibend(a->inode) < a->mvd_bdst)
20692 +               au_set_ibend(a->inode, a->mvd_bdst);
20693 +
20694 +out_unlock:
20695 +       au_do_unlock(dmsg, a);
20696 +out:
20697 +       AuTraceErr(err);
20698 +       return err;
20699 +}
20700 +
20701 +/* ---------------------------------------------------------------------- */
20702 +
20703 +static int find_lower_writable(struct au_mvd_args *a)
20704 +{
20705 +       struct super_block *sb;
20706 +       aufs_bindex_t bindex, bend;
20707 +       struct au_branch *br;
20708 +
20709 +       sb = a->sb;
20710 +       bindex = a->mvd_bsrc;
20711 +       bend = au_sbend(sb);
20712 +       if (!(a->mvdown.flags & AUFS_MVDOWN_ROLOWER)) {
20713 +               for (bindex++; bindex <= bend; bindex++) {
20714 +                       br = au_sbr(sb, bindex);
20715 +                       if (!au_br_rdonly(br))
20716 +                               return bindex;
20717 +               }
20718 +       } else {
20719 +               for (bindex++; bindex <= bend; bindex++) {
20720 +                       br = au_sbr(sb, bindex);
20721 +                       if (!(au_br_sb(br)->s_flags & MS_RDONLY)) {
20722 +                               if (au_br_rdonly(br))
20723 +                                       a->mvdown.flags
20724 +                                               |= AUFS_MVDOWN_ROLOWER_R;
20725 +                               return bindex;
20726 +                       }
20727 +               }
20728 +       }
20729 +
20730 +       return -1;
20731 +}
20732 +
20733 +/* make sure the file is idle */
20734 +static int au_mvd_args_busy(const unsigned char dmsg, struct au_mvd_args *a)
20735 +{
20736 +       int err, plinked;
20737 +
20738 +       err = 0;
20739 +       plinked = !!au_opt_test(au_mntflags(a->sb), PLINK);
20740 +       if (au_dbstart(a->dentry) == a->mvd_bsrc
20741 +           && d_count(a->dentry) == 1
20742 +           && atomic_read(&a->inode->i_count) == 1
20743 +           /* && a->mvd_h_src_inode->i_nlink == 1 */
20744 +           && (!plinked || !au_plink_test(a->inode))
20745 +           && a->inode->i_nlink == 1)
20746 +               goto out;
20747 +
20748 +       err = -EBUSY;
20749 +       AU_MVD_PR(dmsg,
20750 +                 "b%d, d{b%d, c%u?}, i{c%d?, l%u}, hi{l%u}, p{%d, %d}\n",
20751 +                 a->mvd_bsrc, au_dbstart(a->dentry), d_count(a->dentry),
20752 +                 atomic_read(&a->inode->i_count), a->inode->i_nlink,
20753 +                 a->mvd_h_src_inode->i_nlink,
20754 +                 plinked, plinked ? au_plink_test(a->inode) : 0);
20755 +
20756 +out:
20757 +       AuTraceErr(err);
20758 +       return err;
20759 +}
20760 +
20761 +/* make sure the parent dir is fine */
20762 +static int au_mvd_args_parent(const unsigned char dmsg,
20763 +                             struct au_mvd_args *a)
20764 +{
20765 +       int err;
20766 +       aufs_bindex_t bindex;
20767 +
20768 +       err = 0;
20769 +       if (unlikely(au_alive_dir(a->parent))) {
20770 +               err = -ENOENT;
20771 +               AU_MVD_PR(dmsg, "parent dir is dead\n");
20772 +               goto out;
20773 +       }
20774 +
20775 +       a->bopq = au_dbdiropq(a->parent);
20776 +       bindex = au_wbr_nonopq(a->dentry, a->mvd_bdst);
20777 +       AuDbg("b%d\n", bindex);
20778 +       if (unlikely((bindex >= 0 && bindex < a->mvd_bdst)
20779 +                    || (a->bopq != -1 && a->bopq < a->mvd_bdst))) {
20780 +               err = -EINVAL;
20781 +               a->mvd_errno = EAU_MVDOWN_OPAQUE;
20782 +               AU_MVD_PR(dmsg, "ancestor is opaque b%d, b%d\n",
20783 +                         a->bopq, a->mvd_bdst);
20784 +       }
20785 +
20786 +out:
20787 +       AuTraceErr(err);
20788 +       return err;
20789 +}
20790 +
20791 +static int au_mvd_args_intermediate(const unsigned char dmsg,
20792 +                                   struct au_mvd_args *a)
20793 +{
20794 +       int err;
20795 +       struct au_dinfo *dinfo, *tmp;
20796 +
20797 +       /* lookup the next lower positive entry */
20798 +       err = -ENOMEM;
20799 +       tmp = au_di_alloc(a->sb, AuLsc_DI_TMP);
20800 +       if (unlikely(!tmp))
20801 +               goto out;
20802 +
20803 +       a->bfound = -1;
20804 +       a->bwh = -1;
20805 +       dinfo = au_di(a->dentry);
20806 +       au_di_cp(tmp, dinfo);
20807 +       au_di_swap(tmp, dinfo);
20808 +
20809 +       /* returns the number of positive dentries */
20810 +       err = au_lkup_dentry(a->dentry, a->mvd_bsrc + 1, /*type*/0);
20811 +       if (!err)
20812 +               a->bwh = au_dbwh(a->dentry);
20813 +       else if (err > 0)
20814 +               a->bfound = au_dbstart(a->dentry);
20815 +
20816 +       au_di_swap(tmp, dinfo);
20817 +       au_rw_write_unlock(&tmp->di_rwsem);
20818 +       au_di_free(tmp);
20819 +       if (unlikely(err < 0))
20820 +               AU_MVD_PR(dmsg, "failed look-up lower\n");
20821 +
20822 +       /*
20823 +        * here, we have these cases.
20824 +        * bfound == -1
20825 +        *      no positive dentry under bsrc. there are more sub-cases.
20826 +        *      bwh < 0
20827 +        *              there no whiteout, we can safely move-down.
20828 +        *      bwh <= bsrc
20829 +        *              impossible
20830 +        *      bsrc < bwh && bwh < bdst
20831 +        *              there is a whiteout on RO branch. cannot proceed.
20832 +        *      bwh == bdst
20833 +        *              there is a whiteout on the RW target branch. it should
20834 +        *              be removed.
20835 +        *      bdst < bwh
20836 +        *              there is a whiteout somewhere unrelated branch.
20837 +        * -1 < bfound && bfound <= bsrc
20838 +        *      impossible.
20839 +        * bfound < bdst
20840 +        *      found, but it is on RO branch between bsrc and bdst. cannot
20841 +        *      proceed.
20842 +        * bfound == bdst
20843 +        *      found, replace it if AUFS_MVDOWN_FORCE is set. otherwise return
20844 +        *      error.
20845 +        * bdst < bfound
20846 +        *      found, after we create the file on bdst, it will be hidden.
20847 +        */
20848 +
20849 +       AuDebugOn(a->bfound == -1
20850 +                 && a->bwh != -1
20851 +                 && a->bwh <= a->mvd_bsrc);
20852 +       AuDebugOn(-1 < a->bfound
20853 +                 && a->bfound <= a->mvd_bsrc);
20854 +
20855 +       err = -EINVAL;
20856 +       if (a->bfound == -1
20857 +           && a->mvd_bsrc < a->bwh
20858 +           && a->bwh != -1
20859 +           && a->bwh < a->mvd_bdst) {
20860 +               a->mvd_errno = EAU_MVDOWN_WHITEOUT;
20861 +               AU_MVD_PR(dmsg, "bsrc %d, bdst %d, bfound %d, bwh %d\n",
20862 +                         a->mvd_bsrc, a->mvd_bdst, a->bfound, a->bwh);
20863 +               goto out;
20864 +       } else if (a->bfound != -1 && a->bfound < a->mvd_bdst) {
20865 +               a->mvd_errno = EAU_MVDOWN_UPPER;
20866 +               AU_MVD_PR(dmsg, "bdst %d, bfound %d\n",
20867 +                         a->mvd_bdst, a->bfound);
20868 +               goto out;
20869 +       }
20870 +
20871 +       err = 0; /* success */
20872 +
20873 +out:
20874 +       AuTraceErr(err);
20875 +       return err;
20876 +}
20877 +
20878 +static int au_mvd_args_exist(const unsigned char dmsg, struct au_mvd_args *a)
20879 +{
20880 +       int err;
20881 +
20882 +       err = 0;
20883 +       if (!(a->mvdown.flags & AUFS_MVDOWN_OWLOWER)
20884 +           && a->bfound == a->mvd_bdst)
20885 +               err = -EEXIST;
20886 +       AuTraceErr(err);
20887 +       return err;
20888 +}
20889 +
20890 +static int au_mvd_args(const unsigned char dmsg, struct au_mvd_args *a)
20891 +{
20892 +       int err;
20893 +       struct au_branch *br;
20894 +
20895 +       err = -EISDIR;
20896 +       if (unlikely(S_ISDIR(a->inode->i_mode)))
20897 +               goto out;
20898 +
20899 +       err = -EINVAL;
20900 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_UPPER))
20901 +               a->mvd_bsrc = au_ibstart(a->inode);
20902 +       else {
20903 +               a->mvd_bsrc = au_br_index(a->sb, a->mvd_src_brid);
20904 +               if (unlikely(a->mvd_bsrc < 0
20905 +                            || (a->mvd_bsrc < au_dbstart(a->dentry)
20906 +                                || au_dbend(a->dentry) < a->mvd_bsrc
20907 +                                || !au_h_dptr(a->dentry, a->mvd_bsrc))
20908 +                            || (a->mvd_bsrc < au_ibstart(a->inode)
20909 +                                || au_ibend(a->inode) < a->mvd_bsrc
20910 +                                || !au_h_iptr(a->inode, a->mvd_bsrc)))) {
20911 +                       a->mvd_errno = EAU_MVDOWN_NOUPPER;
20912 +                       AU_MVD_PR(dmsg, "no upper\n");
20913 +                       goto out;
20914 +               }
20915 +       }
20916 +       if (unlikely(a->mvd_bsrc == au_sbend(a->sb))) {
20917 +               a->mvd_errno = EAU_MVDOWN_BOTTOM;
20918 +               AU_MVD_PR(dmsg, "on the bottom\n");
20919 +               goto out;
20920 +       }
20921 +       a->mvd_h_src_inode = au_h_iptr(a->inode, a->mvd_bsrc);
20922 +       br = au_sbr(a->sb, a->mvd_bsrc);
20923 +       err = au_br_rdonly(br);
20924 +       if (!(a->mvdown.flags & AUFS_MVDOWN_ROUPPER)) {
20925 +               if (unlikely(err))
20926 +                       goto out;
20927 +       } else if (!(vfsub_native_ro(a->mvd_h_src_inode)
20928 +                    || IS_APPEND(a->mvd_h_src_inode))) {
20929 +               if (err)
20930 +                       a->mvdown.flags |= AUFS_MVDOWN_ROUPPER_R;
20931 +               /* go on */
20932 +       } else
20933 +               goto out;
20934 +
20935 +       err = -EINVAL;
20936 +       if (!(a->mvdown.flags & AUFS_MVDOWN_BRID_LOWER)) {
20937 +               a->mvd_bdst = find_lower_writable(a);
20938 +               if (unlikely(a->mvd_bdst < 0)) {
20939 +                       a->mvd_errno = EAU_MVDOWN_BOTTOM;
20940 +                       AU_MVD_PR(dmsg, "no writable lower branch\n");
20941 +                       goto out;
20942 +               }
20943 +       } else {
20944 +               a->mvd_bdst = au_br_index(a->sb, a->mvd_dst_brid);
20945 +               if (unlikely(a->mvd_bdst < 0
20946 +                            || au_sbend(a->sb) < a->mvd_bdst)) {
20947 +                       a->mvd_errno = EAU_MVDOWN_NOLOWERBR;
20948 +                       AU_MVD_PR(dmsg, "no lower brid\n");
20949 +                       goto out;
20950 +               }
20951 +       }
20952 +
20953 +       err = au_mvd_args_busy(dmsg, a);
20954 +       if (!err)
20955 +               err = au_mvd_args_parent(dmsg, a);
20956 +       if (!err)
20957 +               err = au_mvd_args_intermediate(dmsg, a);
20958 +       if (!err)
20959 +               err = au_mvd_args_exist(dmsg, a);
20960 +       if (!err)
20961 +               AuDbg("b%d, b%d\n", a->mvd_bsrc, a->mvd_bdst);
20962 +
20963 +out:
20964 +       AuTraceErr(err);
20965 +       return err;
20966 +}
20967 +
20968 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *uarg)
20969 +{
20970 +       int err, e;
20971 +       unsigned char dmsg;
20972 +       struct au_mvd_args *args;
20973 +
20974 +       err = -EPERM;
20975 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
20976 +               goto out;
20977 +
20978 +       WARN_ONCE(1, "move-down is still testing...\n");
20979 +
20980 +       err = -ENOMEM;
20981 +       args = kmalloc(sizeof(*args), GFP_NOFS);
20982 +       if (unlikely(!args))
20983 +               goto out;
20984 +
20985 +       err = copy_from_user(&args->mvdown, uarg, sizeof(args->mvdown));
20986 +       if (!err)
20987 +               err = !access_ok(VERIFY_WRITE, uarg, sizeof(*uarg));
20988 +       if (unlikely(err)) {
20989 +               err = -EFAULT;
20990 +               AuTraceErr(err);
20991 +               goto out_free;
20992 +       }
20993 +       AuDbg("flags 0x%x\n", args->mvdown.flags);
20994 +       args->mvdown.flags &= ~(AUFS_MVDOWN_ROLOWER_R | AUFS_MVDOWN_ROUPPER_R);
20995 +       args->mvdown.au_errno = 0;
20996 +       args->dentry = dentry;
20997 +       args->inode = dentry->d_inode;
20998 +       args->sb = dentry->d_sb;
20999 +
21000 +       err = -ENOENT;
21001 +       dmsg = !!(args->mvdown.flags & AUFS_MVDOWN_DMSG);
21002 +       args->parent = dget_parent(dentry);
21003 +       args->dir = args->parent->d_inode;
21004 +       mutex_lock_nested(&args->dir->i_mutex, I_MUTEX_PARENT);
21005 +       dput(args->parent);
21006 +       if (unlikely(args->parent != dentry->d_parent)) {
21007 +               AU_MVD_PR(dmsg, "parent dir is moved\n");
21008 +               goto out_dir;
21009 +       }
21010 +
21011 +       mutex_lock_nested(&args->inode->i_mutex, I_MUTEX_CHILD);
21012 +       err = aufs_read_lock(dentry, AuLock_DW | AuLock_FLUSH);
21013 +       if (unlikely(err))
21014 +               goto out_inode;
21015 +
21016 +       di_write_lock_parent(args->parent);
21017 +       err = au_mvd_args(dmsg, args);
21018 +       if (unlikely(err))
21019 +               goto out_parent;
21020 +
21021 +       AuDbgDentry(dentry);
21022 +       AuDbgInode(args->inode);
21023 +       err = au_do_mvdown(dmsg, args);
21024 +       if (unlikely(err))
21025 +               goto out_parent;
21026 +       AuDbgDentry(dentry);
21027 +       AuDbgInode(args->inode);
21028 +
21029 +       au_cpup_attr_timesizes(args->dir);
21030 +       au_cpup_attr_timesizes(args->inode);
21031 +       au_cpup_igen(args->inode, au_h_iptr(args->inode, args->mvd_bdst));
21032 +       /* au_digen_dec(dentry); */
21033 +
21034 +out_parent:
21035 +       di_write_unlock(args->parent);
21036 +       aufs_read_unlock(dentry, AuLock_DW);
21037 +out_inode:
21038 +       mutex_unlock(&args->inode->i_mutex);
21039 +out_dir:
21040 +       mutex_unlock(&args->dir->i_mutex);
21041 +out_free:
21042 +       e = copy_to_user(uarg, &args->mvdown, sizeof(args->mvdown));
21043 +       if (unlikely(e))
21044 +               err = -EFAULT;
21045 +       kfree(args);
21046 +out:
21047 +       AuTraceErr(err);
21048 +       return err;
21049 +}
21050 diff -urN /usr/share/empty/fs/aufs/opts.c linux/fs/aufs/opts.c
21051 --- /usr/share/empty/fs/aufs/opts.c     1970-01-01 01:00:00.000000000 +0100
21052 +++ linux/fs/aufs/opts.c        2014-04-24 22:11:10.855269116 +0200
21053 @@ -0,0 +1,1701 @@
21054 +/*
21055 + * Copyright (C) 2005-2014 Junjiro R. Okajima
21056 + *
21057 + * This program, aufs is free software; you can redistribute it and/or modify
21058 + * it under the terms of the GNU General Public License as published by
21059 + * the Free Software Foundation; either version 2 of the License, or
21060 + * (at your option) any later version.
21061 + *
21062 + * This program is distributed in the hope that it will be useful,
21063 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
21064 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
21065 + * GNU General Public License for more details.
21066 + *
21067 + * You should have received a copy of the GNU General Public License
21068 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
21069 + */
21070 +
21071 +/*
21072 + * mount options/flags
21073 + */
21074 +
21075 +#include <linux/namei.h>
21076 +#include <linux/types.h> /* a distribution requires */
21077 +#include <linux/parser.h>
21078 +#include "aufs.h"
21079 +
21080 +/* ---------------------------------------------------------------------- */
21081 +
21082 +enum {
21083 +       Opt_br,
21084 +       Opt_add, Opt_del, Opt_mod, Opt_reorder, Opt_append, Opt_prepend,
21085 +       Opt_idel, Opt_imod, Opt_ireorder,
21086 +       Opt_dirwh, Opt_rdcache, Opt_rdblk, Opt_rdhash, Opt_rendir,
21087 +       Opt_rdblk_def, Opt_rdhash_def,
21088 +       Opt_xino, Opt_zxino, Opt_noxino,
21089 +       Opt_trunc_xino, Opt_trunc_xino_v, Opt_notrunc_xino,
21090 +       Opt_trunc_xino_path, Opt_itrunc_xino,
21091 +       Opt_trunc_xib, Opt_notrunc_xib,
21092 +       Opt_shwh, Opt_noshwh,
21093 +       Opt_plink, Opt_noplink, Opt_list_plink,
21094 +       Opt_udba,
21095 +       Opt_dio, Opt_nodio,
21096 +       /* Opt_lock, Opt_unlock, */
21097 +       Opt_cmd, Opt_cmd_args,
21098 +       Opt_diropq_a, Opt_diropq_w,
21099 +       Opt_warn_perm, Opt_nowarn_perm,
21100 +       Opt_wbr_copyup, Opt_wbr_create,
21101 +       Opt_refrof, Opt_norefrof,
21102 +       Opt_verbose, Opt_noverbose,
21103 +       Opt_sum, Opt_nosum, Opt_wsum,
21104 +       Opt_tail, Opt_ignore, Opt_ignore_silent, Opt_err
21105 +};
21106 +
21107 +static match_table_t options = {
21108 +       {Opt_br, "br=%s"},
21109 +       {Opt_br, "br:%s"},
21110 +
21111 +       {Opt_add, "add=%d:%s"},
21112 +       {Opt_add, "add:%d:%s"},
21113 +       {Opt_add, "ins=%d:%s"},
21114 +       {Opt_add, "ins:%d:%s"},
21115 +       {Opt_append, "append=%s"},
21116 +       {Opt_append, "append:%s"},
21117 +       {Opt_prepend, "prepend=%s"},
21118 +       {Opt_prepend, "prepend:%s"},
21119 +
21120 +       {Opt_del, "del=%s"},
21121 +       {Opt_del, "del:%s"},
21122 +       /* {Opt_idel, "idel:%d"}, */
21123 +       {Opt_mod, "mod=%s"},
21124 +       {Opt_mod, "mod:%s"},
21125 +       /* {Opt_imod, "imod:%d:%s"}, */
21126 +
21127 +       {Opt_dirwh, "dirwh=%d"},
21128 +
21129 +       {Opt_xino, "xino=%s"},
21130 +       {Opt_noxino, "noxino"},
21131 +       {Opt_trunc_xino, "trunc_xino"},
21132 +       {Opt_trunc_xino_v, "trunc_xino_v=%d:%d"},
21133 +       {Opt_notrunc_xino, "notrunc_xino"},
21134 +       {Opt_trunc_xino_path, "trunc_xino=%s"},
21135 +       {Opt_itrunc_xino, "itrunc_xino=%d"},
21136 +       /* {Opt_zxino, "zxino=%s"}, */
21137 +       {Opt_trunc_xib, "trunc_xib"},
21138 +       {Opt_notrunc_xib, "notrunc_xib"},
21139 +
21140 +#ifdef CONFIG_PROC_FS
21141 +       {Opt_plink, "plink"},
21142 +#else
21143 +       {Opt_ignore_silent, "plink"},
21144 +#endif
21145 +
21146 +       {Opt_noplink, "noplink"},
21147 +
21148 +#ifdef CONFIG_AUFS_DEBUG
21149 +       {Opt_list_plink, "list_plink"},
21150 +#endif
21151 +
21152 +       {Opt_udba, "udba=%s"},
21153 +
21154 +       {Opt_dio, "dio"},
21155 +       {Opt_nodio, "nodio"},
21156 +
21157 +       {Opt_diropq_a, "diropq=always"},
21158 +       {Opt_diropq_a, "diropq=a"},
21159 +       {Opt_diropq_w, "diropq=whiteouted"},
21160 +       {Opt_diropq_w, "diropq=w"},
21161 +
21162 +       {Opt_warn_perm, "warn_perm"},
21163 +       {Opt_nowarn_perm, "nowarn_perm"},
21164 +
21165 +       /* keep them temporary */
21166 +       {Opt_ignore_silent, "coo=%s"},
21167 +       {Opt_ignore_silent, "nodlgt"},
21168 +       {Opt_ignore_silent, "nodirperm1"},
21169 +       {Opt_ignore_silent, "clean_plink"},
21170 +
21171 +#ifdef CONFIG_AUFS_SHWH
21172 +       {Opt_shwh, "shwh"},
21173 +#endif
21174 +       {Opt_noshwh, "noshwh"},
21175 +
21176 +       {Opt_rendir, "rendir=%d"},
21177 +
21178 +       {Opt_refrof, "refrof"},
21179 +       {Opt_norefrof, "norefrof"},
21180 +
21181 +       {Opt_verbose, "verbose"},
21182 +       {Opt_verbose, "v"},
21183 +       {Opt_noverbose, "noverbose"},
21184 +       {Opt_noverbose, "quiet"},
21185 +       {Opt_noverbose, "q"},
21186 +       {Opt_noverbose, "silent"},
21187 +
21188 +       {Opt_sum, "sum"},
21189 +       {Opt_nosum, "nosum"},
21190 +       {Opt_wsum, "wsum"},
21191 +
21192 +       {Opt_rdcache, "rdcache=%d"},
21193 +       {Opt_rdblk, "rdblk=%d"},
21194 +       {Opt_rdblk_def, "rdblk=def"},
21195 +       {Opt_rdhash, "rdhash=%d"},
21196 +       {Opt_rdhash_def, "rdhash=def"},
21197 +
21198 +       {Opt_wbr_create, "create=%s"},
21199 +       {Opt_wbr_create, "create_policy=%s"},
21200 +       {Opt_wbr_copyup, "cpup=%s"},
21201 +       {Opt_wbr_copyup, "copyup=%s"},
21202 +       {Opt_wbr_copyup, "copyup_policy=%s"},
21203 +
21204 +       /* internal use for the scripts */
21205 +       {Opt_ignore_silent, "si=%s"},
21206 +
21207 +       {Opt_br, "dirs=%s"},
21208 +       {Opt_ignore, "debug=%d"},
21209 +       {Opt_ignore, "delete=whiteout"},
21210 +       {Opt_ignore, "delete=all"},
21211 +       {Opt_ignore, "imap=%s"},
21212 +
21213 +       /* temporary workaround, due to old mount(8)? */
21214 +       {Opt_ignore_silent, "relatime"},
21215 +
21216 +       {Opt_err, NULL}
21217 +};
21218 +
21219 +/* ---------------------------------------------------------------------- */
21220 +
21221 +static const char *au_parser_pattern(int val, struct match_token *token)
21222 +{
21223 +       while (token->pattern) {
21224 +               if (token->token == val)
21225 +                       return token->pattern;
21226 +               token++;
21227 +       }
21228 +       BUG();
21229 +       return "??";
21230 +}
21231 +
21232 +/* ---------------------------------------------------------------------- */
21233 +
21234 +static match_table_t brperm = {
21235 +       {AuBrPerm_RO, AUFS_BRPERM_RO},
21236 +       {AuBrPerm_RR, AUFS_BRPERM_RR},
21237 +       {AuBrPerm_RW, AUFS_BRPERM_RW},
21238 +       {0, NULL}
21239 +};
21240 +
21241 +static match_table_t brattr = {
21242 +       {AuBrAttr_UNPIN, AUFS_BRATTR_UNPIN},
21243 +       {AuBrRAttr_WH, AUFS_BRRATTR_WH},
21244 +       {AuBrWAttr_NoLinkWH, AUFS_BRWATTR_NLWH},
21245 +       {0, NULL}
21246 +};
21247 +
21248 +#define AuBrStr_LONGEST        AUFS_BRPERM_RW \
21249 +       "+" AUFS_BRATTR_UNPIN \
21250 +       "+" AUFS_BRWATTR_NLWH
21251 +
21252 +static int br_attr_val(char *str, match_table_t table, substring_t args[])
21253 +{
21254 +       int attr, v;
21255 +       char *p;
21256 +
21257 +       attr = 0;
21258 +       do {
21259 +               p = strchr(str, '+');
21260 +               if (p)
21261 +                       *p = 0;
21262 +               v = match_token(str, table, args);
21263 +               if (v)
21264 +                       attr |= v;
21265 +               else {
21266 +                       if (p)
21267 +                               *p = '+';
21268 +                       pr_warn("ignored branch attribute %s\n", str);
21269 +                       break;
21270 +               }
21271 +               if (p)
21272 +                       str = p + 1;
21273 +       } while (p);
21274 +
21275 +       return attr;
21276 +}
21277 +
21278 +static int noinline_for_stack br_perm_val(char *perm)
21279 +{
21280 +       int val;
21281 +       char *p, *q;
21282 +       substring_t args[MAX_OPT_ARGS];
21283 +
21284 +       p = strchr(perm, '+');
21285 +       if (p)
21286 +               *p = 0;
21287 +       val = match_token(perm, brperm, args);
21288 +       if (!val) {
21289 +               if (p)
21290 +                       *p = '+';
21291 +               pr_warn("ignored branch permission %s\n", perm);
21292 +               val = AuBrPerm_RO;
21293 +               goto out;
21294 +       }
21295 +       if (!p)
21296 +               goto out;
21297 +
21298 +       p++;
21299 +       while (1) {
21300 +               q = strchr(p, '+');
21301 +               if (q)
21302 +                       *q = 0;
21303 +               val |= br_attr_val(p, brattr, args);
21304 +               if (q) {
21305 +                       *q = '+';
21306 +                       p = q + 1;
21307 +               } else
21308 +                       break;
21309 +       }
21310 +       switch (val & AuBrPerm_Mask) {
21311 +       case AuBrPerm_RO:
21312 +       case AuBrPerm_RR:
21313 +               if (unlikely(val & AuBrWAttr_NoLinkWH)) {
21314 +                       pr_warn("ignored branch attribute %s\n",
21315 +                               AUFS_BRWATTR_NLWH);
21316 +                       val &= ~AuBrWAttr_NoLinkWH;
21317 +               }
21318 +               break;
21319 +       case AuBrPerm_RW:
21320 +               if (unlikely(val & AuBrRAttr_WH)) {
21321 +                       pr_warn("ignored branch attribute %s\n",
21322 +                               AUFS_BRRATTR_WH);
21323 +                       val &= ~AuBrRAttr_WH;
21324 +               }
21325 +               break;
21326 +       }
21327 +
21328 +out:
21329 +       return val;
21330 +}
21331 +
21332 +/* Caller should free the return value */
21333 +char *au_optstr_br_perm(int brperm)
21334 +{
21335 +       char *p, a[sizeof(AuBrStr_LONGEST)];
21336 +       int sz;
21337 +
21338 +#define SetPerm(str) do {                      \
21339 +               sz = sizeof(str);               \
21340 +               memcpy(a, str, sz);             \
21341 +               p = a + sz - 1;                 \
21342 +       } while (0)
21343 +
21344 +#define AppendAttr(flag, str) do {                     \
21345 +               if (brperm & flag) {            \
21346 +                       sz = sizeof(str);       \
21347 +                       *p++ = '+';             \
21348 +                       memcpy(p, str, sz);     \
21349 +                       p += sz - 1;            \
21350 +               }                               \
21351 +       } while (0)
21352 +
21353 +       switch (brperm & AuBrPerm_Mask) {
21354 +       case AuBrPerm_RO:
21355 +               SetPerm(AUFS_BRPERM_RO);
21356 +               break;
21357 +       case AuBrPerm_RR:
21358 +               SetPerm(AUFS_BRPERM_RR);
21359 +               break;
21360 +       case AuBrPerm_RW:
21361 +               SetPerm(AUFS_BRPERM_RW);
21362 +               break;
21363 +       default:
21364 +               AuDebugOn(1);
21365 +       }
21366 +
21367 +       AppendAttr(AuBrAttr_UNPIN, AUFS_BRATTR_UNPIN);
21368 +       AppendAttr(AuBrRAttr_WH, AUFS_BRRATTR_WH);
21369 +       AppendAttr(AuBrWAttr_NoLinkWH, AUFS_BRWATTR_NLWH);
21370 +
21371 +       AuDebugOn(strlen(a) >= sizeof(a));
21372 +       return kstrdup(a, GFP_NOFS);
21373 +#undef SetPerm
21374 +#undef AppendAttr
21375 +}
21376 +
21377 +/* ---------------------------------------------------------------------- */
21378 +
21379 +static match_table_t udbalevel = {
21380 +       {AuOpt_UDBA_REVAL, "reval"},
21381 +       {AuOpt_UDBA_NONE, "none"},
21382 +#ifdef CONFIG_AUFS_HNOTIFY
21383 +       {AuOpt_UDBA_HNOTIFY, "notify"}, /* abstraction */
21384 +#ifdef CONFIG_AUFS_HFSNOTIFY
21385 +       {AuOpt_UDBA_HNOTIFY, "fsnotify"},
21386 +#endif
21387 +#endif
21388 +       {-1, NULL}
21389 +};
21390 +
21391 +static int noinline_for_stack udba_val(char *str)
21392 +{
21393 +       substring_t args[MAX_OPT_ARGS];
21394 +
21395 +       return match_token(str, udbalevel, args);
21396 +}
21397 +
21398 +const char *au_optstr_udba(int udba)
21399 +{
21400 +       return au_parser_pattern(udba, (void *)udbalevel);
21401 +}
21402 +
21403 +/* ---------------------------------------------------------------------- */
21404 +
21405 +static match_table_t au_wbr_create_policy = {
21406 +       {AuWbrCreate_TDP, "tdp"},
21407 +       {AuWbrCreate_TDP, "top-down-parent"},
21408 +       {AuWbrCreate_RR, "rr"},
21409 +       {AuWbrCreate_RR, "round-robin"},
21410 +       {AuWbrCreate_MFS, "mfs"},
21411 +       {AuWbrCreate_MFS, "most-free-space"},
21412 +       {AuWbrCreate_MFSV, "mfs:%d"},
21413 +       {AuWbrCreate_MFSV, "most-free-space:%d"},
21414 +
21415 +       {AuWbrCreate_MFSRR, "mfsrr:%d"},
21416 +       {AuWbrCreate_MFSRRV, "mfsrr:%d:%d"},
21417 +       {AuWbrCreate_PMFS, "pmfs"},
21418 +       {AuWbrCreate_PMFSV, "pmfs:%d"},
21419 +       {AuWbrCreate_PMFSRR, "pmfsrr:%d"},
21420 +       {AuWbrCreate_PMFSRRV, "pmfsrr:%d:%d"},
21421 +
21422 +       {-1, NULL}
21423 +};
21424 +
21425 +/*
21426 + * cf. linux/lib/parser.c and cmdline.c
21427 + * gave up calling memparse() since it uses simple_strtoull() instead of
21428 + * kstrto...().
21429 + */
21430 +static int noinline_for_stack
21431 +au_match_ull(substring_t *s, unsigned long long *result)
21432 +{
21433 +       int err;
21434 +       unsigned int len;
21435 +       char a[32];
21436 +
21437 +       err = -ERANGE;
21438 +       len = s->to - s->from;
21439 +       if (len + 1 <= sizeof(a)) {
21440 +               memcpy(a, s->from, len);
21441 +               a[len] = '\0';
21442 +               err = kstrtoull(a, 0, result);
21443 +       }
21444 +       return err;
21445 +}
21446 +
21447 +static int au_wbr_mfs_wmark(substring_t *arg, char *str,
21448 +                           struct au_opt_wbr_create *create)
21449 +{
21450 +       int err;
21451 +       unsigned long long ull;
21452 +
21453 +       err = 0;
21454 +       if (!au_match_ull(arg, &ull))
21455 +               create->mfsrr_watermark = ull;
21456 +       else {
21457 +               pr_err("bad integer in %s\n", str);
21458 +               err = -EINVAL;
21459 +       }
21460 +
21461 +       return err;
21462 +}
21463 +
21464 +static int au_wbr_mfs_sec(substring_t *arg, char *str,
21465 +                         struct au_opt_wbr_create *create)
21466 +{
21467 +       int n, err;
21468 +
21469 +       err = 0;
21470 +       if (!match_int(arg, &n) && 0 <= n && n <= AUFS_MFS_MAX_SEC)
21471 +               create->mfs_second = n;
21472 +       else {
21473 +               pr_err("bad integer in %s\n", str);
21474 +               err = -EINVAL;
21475 +       }
21476 +
21477 +       return err;
21478 +}
21479 +
21480 +static int noinline_for_stack
21481 +au_wbr_create_val(char *str, struct au_opt_wbr_create *create)
21482 +{
21483 +       int err, e;
21484 +       substring_t args[MAX_OPT_ARGS];
21485 +
21486 +       err = match_token(str, au_wbr_create_policy, args);
21487 +       create->wbr_create = err;
21488 +       switch (err) {
21489 +       case AuWbrCreate_MFSRRV:
21490 +       case AuWbrCreate_PMFSRRV:
21491 +               e = au_wbr_mfs_wmark(&args[0], str, create);
21492 +               if (!e)
21493 +                       e = au_wbr_mfs_sec(&args[1], str, create);
21494 +               if (unlikely(e))
21495 +                       err = e;
21496 +               break;
21497 +       case AuWbrCreate_MFSRR:
21498 +       case AuWbrCreate_PMFSRR:
21499 +               e = au_wbr_mfs_wmark(&args[0], str, create);
21500 +               if (unlikely(e)) {
21501 +                       err = e;
21502 +                       break;
21503 +               }
21504 +               /*FALLTHROUGH*/
21505 +       case AuWbrCreate_MFS:
21506 +       case AuWbrCreate_PMFS:
21507 +               create->mfs_second = AUFS_MFS_DEF_SEC;
21508 +               break;
21509 +       case AuWbrCreate_MFSV:
21510 +       case AuWbrCreate_PMFSV:
21511 +               e = au_wbr_mfs_sec(&args[0], str, create);
21512 +               if (unlikely(e))
21513 +                       err = e;
21514 +               break;
21515 +       }
21516 +
21517 +       return err;
21518 +}
21519 +
21520 +const char *au_optstr_wbr_create(int wbr_create)
21521 +{
21522 +       return au_parser_pattern(wbr_create, (void *)au_wbr_create_policy);
21523 +}
21524 +
21525 +static match_table_t au_wbr_copyup_policy = {
21526 +       {AuWbrCopyup_TDP, "tdp"},
21527 +       {AuWbrCopyup_TDP, "top-down-parent"},
21528 +       {AuWbrCopyup_BUP, "bup"},
21529 +       {AuWbrCopyup_BUP, "bottom-up-parent"},
21530 +       {AuWbrCopyup_BU, "bu"},
21531 +       {AuWbrCopyup_BU, "bottom-up"},
21532 +       {-1, NULL}
21533 +};
21534 +
21535 +static int noinline_for_stack au_wbr_copyup_val(char *str)
21536 +{
21537 +       substring_t args[MAX_OPT_ARGS];
21538 +
21539 +       return match_token(str, au_wbr_copyup_policy, args);
21540 +}
21541 +
21542 +const char *au_optstr_wbr_copyup(int wbr_copyup)
21543 +{
21544 +       return au_parser_pattern(wbr_copyup, (void *)au_wbr_copyup_policy);
21545 +}
21546 +
21547 +/* ---------------------------------------------------------------------- */
21548 +
21549 +static const int lkup_dirflags = LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
21550 +
21551 +static void dump_opts(struct au_opts *opts)
21552 +{
21553 +#ifdef CONFIG_AUFS_DEBUG
21554 +       /* reduce stack space */
21555 +       union {
21556 +               struct au_opt_add *add;
21557 +               struct au_opt_del *del;
21558 +               struct au_opt_mod *mod;
21559 +               struct au_opt_xino *xino;
21560 +               struct au_opt_xino_itrunc *xino_itrunc;
21561 +               struct au_opt_wbr_create *create;
21562 +       } u;
21563 +       struct au_opt *opt;
21564 +
21565 +       opt = opts->opt;
21566 +       while (opt->type != Opt_tail) {
21567 +               switch (opt->type) {
21568 +               case Opt_add:
21569 +                       u.add = &opt->add;
21570 +                       AuDbg("add {b%d, %s, 0x%x, %p}\n",
21571 +                                 u.add->bindex, u.add->pathname, u.add->perm,
21572 +                                 u.add->path.dentry);
21573 +                       break;
21574 +               case Opt_del:
21575 +               case Opt_idel:
21576 +                       u.del = &opt->del;
21577 +                       AuDbg("del {%s, %p}\n",
21578 +                             u.del->pathname, u.del->h_path.dentry);
21579 +                       break;
21580 +               case Opt_mod:
21581 +               case Opt_imod:
21582 +                       u.mod = &opt->mod;
21583 +                       AuDbg("mod {%s, 0x%x, %p}\n",
21584 +                                 u.mod->path, u.mod->perm, u.mod->h_root);
21585 +                       break;
21586 +               case Opt_append:
21587 +                       u.add = &opt->add;
21588 +                       AuDbg("append {b%d, %s, 0x%x, %p}\n",
21589 +                                 u.add->bindex, u.add->pathname, u.add->perm,
21590 +                                 u.add->path.dentry);
21591 +                       break;
21592 +               case Opt_prepend:
21593 +                       u.add = &opt->add;
21594 +                       AuDbg("prepend {b%d, %s, 0x%x, %p}\n",
21595 +                                 u.add->bindex, u.add->pathname, u.add->perm,
21596 +                                 u.add->path.dentry);
21597 +                       break;
21598 +               case Opt_dirwh:
21599 +                       AuDbg("dirwh %d\n", opt->dirwh);
21600 +                       break;
21601 +               case Opt_rdcache:
21602 +                       AuDbg("rdcache %d\n", opt->rdcache);
21603 +                       break;
21604 +               case Opt_rdblk:
21605 +                       AuDbg("rdblk %u\n", opt->rdblk);
21606 +                       break;
21607 +               case Opt_rdblk_def:
21608 +                       AuDbg("rdblk_def\n");
21609 +                       break;
21610 +               case Opt_rdhash:
21611 +                       AuDbg("rdhash %u\n", opt->rdhash);
21612 +                       break;
21613 +               case Opt_rdhash_def:
21614 +                       AuDbg("rdhash_def\n");
21615 +                       break;
21616 +               case Opt_xino:
21617 +                       u.xino = &opt->xino;
21618 +                       AuDbg("xino {%s %pD}\n", u.xino->path, u.xino->file);
21619 +                       break;
21620 +               case Opt_trunc_xino:
21621 +                       AuLabel(trunc_xino);
21622 +                       break;
21623 +               case Opt_notrunc_xino:
21624 +                       AuLabel(notrunc_xino);
21625 +                       break;
21626 +               case Opt_trunc_xino_path:
21627 +               case Opt_itrunc_xino:
21628 +                       u.xino_itrunc = &opt->xino_itrunc;
21629 +                       AuDbg("trunc_xino %d\n", u.xino_itrunc->bindex);
21630 +                       break;
21631 +
21632 +               case Opt_noxino:
21633 +                       AuLabel(noxino);
21634 +                       break;
21635 +               case Opt_trunc_xib:
21636 +                       AuLabel(trunc_xib);
21637 +                       break;
21638 +               case Opt_notrunc_xib:
21639 +                       AuLabel(notrunc_xib);
21640 +                       break;
21641 +               case Opt_shwh:
21642 +                       AuLabel(shwh);
21643 +                       break;
21644 +               case Opt_noshwh:
21645 +                       AuLabel(noshwh);
21646 +                       break;
21647 +               case Opt_plink:
21648 +                       AuLabel(plink);
21649 +                       break;
21650 +               case Opt_noplink:
21651 +                       AuLabel(noplink);
21652 +                       break;
21653 +               case Opt_list_plink:
21654 +                       AuLabel(list_plink);
21655 +                       break;
21656 +               case Opt_udba:
21657 +                       AuDbg("udba %d, %s\n",
21658 +                                 opt->udba, au_optstr_udba(opt->udba));
21659 +                       break;
21660 +               case Opt_dio:
21661 +                       AuLabel(dio);
21662 +                       break;
21663 +               case Opt_nodio:
21664 +                       AuLabel(nodio);
21665 +                       break;
21666 +               case Opt_diropq_a:
21667 +                       AuLabel(diropq_a);
21668 +                       break;
21669 +               case Opt_diropq_w:
21670 +                       AuLabel(diropq_w);
21671 +                       break;
21672 +               case Opt_warn_perm:
21673 +                       AuLabel(warn_perm);
21674 +                       break;
21675 +               case Opt_nowarn_perm:
21676 +                       AuLabel(nowarn_perm);
21677 +                       break;
21678 +               case Opt_refrof:
21679 +                       AuLabel(refrof);
21680 +                       break;
21681 +               case Opt_norefrof:
21682 +                       AuLabel(norefrof);
21683 +                       break;
21684 +               case Opt_verbose:
21685 +                       AuLabel(verbose);
21686 +                       break;
21687 +               case Opt_noverbose:
21688 +                       AuLabel(noverbose);
21689 +                       break;
21690 +               case Opt_sum:
21691 +                       AuLabel(sum);
21692 +                       break;
21693 +               case Opt_nosum:
21694 +                       AuLabel(nosum);
21695 +                       break;
21696 +               case Opt_wsum:
21697 +                       AuLabel(wsum);
21698 +                       break;
21699 +               case Opt_wbr_create:
21700 +                       u.create = &opt->wbr_create;
21701 +                       AuDbg("create %d, %s\n", u.create->wbr_create,
21702 +                                 au_optstr_wbr_create(u.create->wbr_create));
21703 +                       switch (u.create->wbr_create) {
21704 +                       case AuWbrCreate_MFSV:
21705 +                       case AuWbrCreate_PMFSV:
21706 +                               AuDbg("%d sec\n", u.create->mfs_second);
21707 +                               break;
21708 +                       case AuWbrCreate_MFSRR:
21709 +                               AuDbg("%llu watermark\n",
21710 +                                         u.create->mfsrr_watermark);
21711 +                               break;
21712 +                       case AuWbrCreate_MFSRRV:
21713 +                       case AuWbrCreate_PMFSRRV:
21714 +                               AuDbg("%llu watermark, %d sec\n",
21715 +                                         u.create->mfsrr_watermark,
21716 +                                         u.create->mfs_second);
21717 +                               break;
21718 +                       }
21719 +                       break;
21720 +               case Opt_wbr_copyup:
21721 +                       AuDbg("copyup %d, %s\n", opt->wbr_copyup,
21722 +                                 au_optstr_wbr_copyup(opt->wbr_copyup));
21723 +                       break;
21724 +               default:
21725 +                       BUG();
21726 +               }
21727 +               opt++;
21728 +       }
21729 +#endif
21730 +}
21731 +
21732 +void au_opts_free(struct au_opts *opts)
21733 +{
21734 +       struct au_opt *opt;
21735 +
21736 +       opt = opts->opt;
21737 +       while (opt->type != Opt_tail) {
21738 +               switch (opt->type) {
21739 +               case Opt_add:
21740 +               case Opt_append:
21741 +               case Opt_prepend:
21742 +                       path_put(&opt->add.path);
21743 +                       break;
21744 +               case Opt_del:
21745 +               case Opt_idel:
21746 +                       path_put(&opt->del.h_path);
21747 +                       break;
21748 +               case Opt_mod:
21749 +               case Opt_imod:
21750 +                       dput(opt->mod.h_root);
21751 +                       break;
21752 +               case Opt_xino:
21753 +                       fput(opt->xino.file);
21754 +                       break;
21755 +               }
21756 +               opt++;
21757 +       }
21758 +}
21759 +
21760 +static int opt_add(struct au_opt *opt, char *opt_str, unsigned long sb_flags,
21761 +                  aufs_bindex_t bindex)
21762 +{
21763 +       int err;
21764 +       struct au_opt_add *add = &opt->add;
21765 +       char *p;
21766 +
21767 +       add->bindex = bindex;
21768 +       add->perm = AuBrPerm_RO;
21769 +       add->pathname = opt_str;
21770 +       p = strchr(opt_str, '=');
21771 +       if (p) {
21772 +               *p++ = 0;
21773 +               if (*p)
21774 +                       add->perm = br_perm_val(p);
21775 +       }
21776 +
21777 +       err = vfsub_kern_path(add->pathname, lkup_dirflags, &add->path);
21778 +       if (!err) {
21779 +               if (!p) {
21780 +                       add->perm = AuBrPerm_RO;
21781 +                       if (au_test_fs_rr(add->path.dentry->d_sb))
21782 +                               add->perm = AuBrPerm_RR;
21783 +                       else if (!bindex && !(sb_flags & MS_RDONLY))
21784 +                               add->perm = AuBrPerm_RW;
21785 +               }
21786 +               opt->type = Opt_add;
21787 +               goto out;
21788 +       }
21789 +       pr_err("lookup failed %s (%d)\n", add->pathname, err);
21790 +       err = -EINVAL;
21791 +
21792 +out:
21793 +       return err;
21794 +}
21795 +
21796 +static int au_opts_parse_del(struct au_opt_del *del, substring_t args[])
21797 +{
21798 +       int err;
21799 +
21800 +       del->pathname = args[0].from;
21801 +       AuDbg("del path %s\n", del->pathname);
21802 +
21803 +       err = vfsub_kern_path(del->pathname, lkup_dirflags, &del->h_path);
21804 +       if (unlikely(err))
21805 +               pr_err("lookup failed %s (%d)\n", del->pathname, err);
21806 +
21807 +       return err;
21808 +}
21809 +
21810 +#if 0 /* reserved for future use */
21811 +static int au_opts_parse_idel(struct super_block *sb, aufs_bindex_t bindex,
21812 +                             struct au_opt_del *del, substring_t args[])
21813 +{
21814 +       int err;
21815 +       struct dentry *root;
21816 +
21817 +       err = -EINVAL;
21818 +       root = sb->s_root;
21819 +       aufs_read_lock(root, AuLock_FLUSH);
21820 +       if (bindex < 0 || au_sbend(sb) < bindex) {
21821 +               pr_err("out of bounds, %d\n", bindex);
21822 +               goto out;
21823 +       }
21824 +
21825 +       err = 0;
21826 +       del->h_path.dentry = dget(au_h_dptr(root, bindex));
21827 +       del->h_path.mnt = mntget(au_sbr_mnt(sb, bindex));
21828 +
21829 +out:
21830 +       aufs_read_unlock(root, !AuLock_IR);
21831 +       return err;
21832 +}
21833 +#endif
21834 +
21835 +static int noinline_for_stack
21836 +au_opts_parse_mod(struct au_opt_mod *mod, substring_t args[])
21837 +{
21838 +       int err;
21839 +       struct path path;
21840 +       char *p;
21841 +
21842 +       err = -EINVAL;
21843 +       mod->path = args[0].from;
21844 +       p = strchr(mod->path, '=');
21845 +       if (unlikely(!p)) {
21846 +               pr_err("no permssion %s\n", args[0].from);
21847 +               goto out;
21848 +       }
21849 +
21850 +       *p++ = 0;
21851 +       err = vfsub_kern_path(mod->path, lkup_dirflags, &path);
21852 +       if (unlikely(err)) {
21853 +               pr_err("lookup failed %s (%d)\n", mod->path, err);
21854 +               goto out;
21855 +       }
21856 +
21857 +       mod->perm = br_perm_val(p);
21858 +       AuDbg("mod path %s, perm 0x%x, %s\n", mod->path, mod->perm, p);
21859 +       mod->h_root = dget(path.dentry);
21860 +       path_put(&path);
21861 +
21862 +out:
21863 +       return err;
21864 +}
21865 +
21866 +#if 0 /* reserved for future use */
21867 +static int au_opts_parse_imod(struct super_block *sb, aufs_bindex_t bindex,
21868 +                             struct au_opt_mod *mod, substring_t args[])
21869 +{
21870 +       int err;
21871 +       struct dentry *root;
21872 +
21873 +       err = -EINVAL;
21874 +       root = sb->s_root;
21875 +       aufs_read_lock(root, AuLock_FLUSH);
21876 +       if (bindex < 0 || au_sbend(sb) < bindex) {
21877 +               pr_err("out of bounds, %d\n", bindex);
21878 +               goto out;
21879 +       }
21880 +
21881 +       err = 0;
21882 +       mod->perm = br_perm_val(args[1].from);
21883 +       AuDbg("mod path %s, perm 0x%x, %s\n",
21884 +             mod->path, mod->perm, args[1].from);
21885 +       mod->h_root = dget(au_h_dptr(root, bindex));
21886 +
21887 +out:
21888 +       aufs_read_unlock(root, !AuLock_IR);
21889 +       return err;
21890 +}
21891 +#endif
21892 +
21893 +static int au_opts_parse_xino(struct super_block *sb, struct au_opt_xino *xino,
21894 +                             substring_t args[])
21895 +{
21896 +       int err;
21897 +       struct file *file;
21898 +
21899 +       file = au_xino_create(sb, args[0].from, /*silent*/0);
21900 +       err = PTR_ERR(file);
21901 +       if (IS_ERR(file))
21902 +               goto out;
21903 +
21904 +       err = -EINVAL;
21905 +       if (unlikely(file->f_dentry->d_sb == sb)) {
21906 +               fput(file);
21907 +               pr_err("%s must be outside\n", args[0].from);
21908 +               goto out;
21909 +       }
21910 +
21911 +       err = 0;
21912 +       xino->file = file;
21913 +       xino->path = args[0].from;
21914 +
21915 +out:
21916 +       return err;
21917 +}
21918 +
21919 +static int noinline_for_stack
21920 +au_opts_parse_xino_itrunc_path(struct super_block *sb,
21921 +                              struct au_opt_xino_itrunc *xino_itrunc,
21922 +                              substring_t args[])
21923 +{
21924 +       int err;
21925 +       aufs_bindex_t bend, bindex;
21926 +       struct path path;
21927 +       struct dentry *root;
21928 +
21929 +       err = vfsub_kern_path(args[0].from, lkup_dirflags, &path);
21930 +       if (unlikely(err)) {
21931 +               pr_err("lookup failed %s (%d)\n", args[0].from, err);
21932 +               goto out;
21933 +       }
21934 +
21935 +       xino_itrunc->bindex = -1;
21936 +       root = sb->s_root;
21937 +       aufs_read_lock(root, AuLock_FLUSH);
21938 +       bend = au_sbend(sb);
21939 +       for (bindex = 0; bindex <= bend; bindex++) {
21940 +               if (au_h_dptr(root, bindex) == path.dentry) {
21941 +                       xino_itrunc->bindex = bindex;
21942 +                       break;
21943 +               }
21944 +       }
21945 +       aufs_read_unlock(root, !AuLock_IR);
21946 +       path_put(&path);
21947 +
21948 +       if (unlikely(xino_itrunc->bindex < 0)) {
21949 +               pr_err("no such branch %s\n", args[0].from);
21950 +               err = -EINVAL;
21951 +       }
21952 +
21953 +out:
21954 +       return err;
21955 +}
21956 +
21957 +/* called without aufs lock */
21958 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts)
21959 +{
21960 +       int err, n, token;
21961 +       aufs_bindex_t bindex;
21962 +       unsigned char skipped;
21963 +       struct dentry *root;
21964 +       struct au_opt *opt, *opt_tail;
21965 +       char *opt_str;
21966 +       /* reduce the stack space */
21967 +       union {
21968 +               struct au_opt_xino_itrunc *xino_itrunc;
21969 +               struct au_opt_wbr_create *create;
21970 +       } u;
21971 +       struct {
21972 +               substring_t args[MAX_OPT_ARGS];
21973 +       } *a;
21974 +
21975 +       err = -ENOMEM;
21976 +       a = kmalloc(sizeof(*a), GFP_NOFS);
21977 +       if (unlikely(!a))
21978 +               goto out;
21979 +
21980 +       root = sb->s_root;
21981 +       err = 0;
21982 +       bindex = 0;
21983 +       opt = opts->opt;
21984 +       opt_tail = opt + opts->max_opt - 1;
21985 +       opt->type = Opt_tail;
21986 +       while (!err && (opt_str = strsep(&str, ",")) && *opt_str) {
21987 +               err = -EINVAL;
21988 +               skipped = 0;
21989 +               token = match_token(opt_str, options, a->args);
21990 +               switch (token) {
21991 +               case Opt_br:
21992 +                       err = 0;
21993 +                       while (!err && (opt_str = strsep(&a->args[0].from, ":"))
21994 +                              && *opt_str) {
21995 +                               err = opt_add(opt, opt_str, opts->sb_flags,
21996 +                                             bindex++);
21997 +                               if (unlikely(!err && ++opt > opt_tail)) {
21998 +                                       err = -E2BIG;
21999 +                                       break;
22000 +                               }
22001 +                               opt->type = Opt_tail;
22002 +                               skipped = 1;
22003 +                       }
22004 +                       break;
22005 +               case Opt_add:
22006 +                       if (unlikely(match_int(&a->args[0], &n))) {
22007 +                               pr_err("bad integer in %s\n", opt_str);
22008 +                               break;
22009 +                       }
22010 +                       bindex = n;
22011 +                       err = opt_add(opt, a->args[1].from, opts->sb_flags,
22012 +                                     bindex);
22013 +                       if (!err)
22014 +                               opt->type = token;
22015 +                       break;
22016 +               case Opt_append:
22017 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
22018 +                                     /*dummy bindex*/1);
22019 +                       if (!err)
22020 +                               opt->type = token;
22021 +                       break;
22022 +               case Opt_prepend:
22023 +                       err = opt_add(opt, a->args[0].from, opts->sb_flags,
22024 +                                     /*bindex*/0);
22025 +                       if (!err)
22026 +                               opt->type = token;
22027 +                       break;
22028 +               case Opt_del:
22029 +                       err = au_opts_parse_del(&opt->del, a->args);
22030 +                       if (!err)
22031 +                               opt->type = token;
22032 +                       break;
22033 +#if 0 /* reserved for future use */
22034 +               case Opt_idel:
22035 +                       del->pathname = "(indexed)";
22036 +                       if (unlikely(match_int(&args[0], &n))) {
22037 +                               pr_err("bad integer in %s\n", opt_str);
22038 +                               break;
22039 +                       }
22040 +                       err = au_opts_parse_idel(sb, n, &opt->del, a->args);
22041 +                       if (!err)
22042 +                               opt->type = token;
22043 +                       break;
22044 +#endif
22045 +               case Opt_mod:
22046 +                       err = au_opts_parse_mod(&opt->mod, a->args);
22047 +                       if (!err)
22048 +                               opt->type = token;
22049 +                       break;
22050 +#ifdef IMOD /* reserved for future use */
22051 +               case Opt_imod:
22052 +                       u.mod->path = "(indexed)";
22053 +                       if (unlikely(match_int(&a->args[0], &n))) {
22054 +                               pr_err("bad integer in %s\n", opt_str);
22055 +                               break;
22056 +                       }
22057 +                       err = au_opts_parse_imod(sb, n, &opt->mod, a->args);
22058 +                       if (!err)
22059 +                               opt->type = token;
22060 +                       break;
22061 +#endif
22062 +               case Opt_xino:
22063 +                       err = au_opts_parse_xino(sb, &opt->xino, a->args);
22064 +                       if (!err)
22065 +                               opt->type = token;
22066 +                       break;
22067 +
22068 +               case Opt_trunc_xino_path:
22069 +                       err = au_opts_parse_xino_itrunc_path
22070 +                               (sb, &opt->xino_itrunc, a->args);
22071 +                       if (!err)
22072 +                               opt->type = token;
22073 +                       break;
22074 +
22075 +               case Opt_itrunc_xino:
22076 +                       u.xino_itrunc = &opt->xino_itrunc;
22077 +                       if (unlikely(match_int(&a->args[0], &n))) {
22078 +                               pr_err("bad integer in %s\n", opt_str);
22079 +                               break;
22080 +                       }
22081 +                       u.xino_itrunc->bindex = n;
22082 +                       aufs_read_lock(root, AuLock_FLUSH);
22083 +                       if (n < 0 || au_sbend(sb) < n) {
22084 +                               pr_err("out of bounds, %d\n", n);
22085 +                               aufs_read_unlock(root, !AuLock_IR);
22086 +                               break;
22087 +                       }
22088 +                       aufs_read_unlock(root, !AuLock_IR);
22089 +                       err = 0;
22090 +                       opt->type = token;
22091 +                       break;
22092 +
22093 +               case Opt_dirwh:
22094 +                       if (unlikely(match_int(&a->args[0], &opt->dirwh)))
22095 +                               break;
22096 +                       err = 0;
22097 +                       opt->type = token;
22098 +                       break;
22099 +
22100 +               case Opt_rdcache:
22101 +                       if (unlikely(match_int(&a->args[0], &n))) {
22102 +                               pr_err("bad integer in %s\n", opt_str);
22103 +                               break;
22104 +                       }
22105 +                       if (unlikely(n > AUFS_RDCACHE_MAX)) {
22106 +                               pr_err("rdcache must be smaller than %d\n",
22107 +                                      AUFS_RDCACHE_MAX);
22108 +                               break;
22109 +                       }
22110 +                       opt->rdcache = n;
22111 +                       err = 0;
22112 +                       opt->type = token;
22113 +                       break;
22114 +               case Opt_rdblk:
22115 +                       if (unlikely(match_int(&a->args[0], &n)
22116 +                                    || n < 0
22117 +                                    || n > KMALLOC_MAX_SIZE)) {
22118 +                               pr_err("bad integer in %s\n", opt_str);
22119 +                               break;
22120 +                       }
22121 +                       if (unlikely(n && n < NAME_MAX)) {
22122 +                               pr_err("rdblk must be larger than %d\n",
22123 +                                      NAME_MAX);
22124 +                               break;
22125 +                       }
22126 +                       opt->rdblk = n;
22127 +                       err = 0;
22128 +                       opt->type = token;
22129 +                       break;
22130 +               case Opt_rdhash:
22131 +                       if (unlikely(match_int(&a->args[0], &n)
22132 +                                    || n < 0
22133 +                                    || n * sizeof(struct hlist_head)
22134 +                                    > KMALLOC_MAX_SIZE)) {
22135 +                               pr_err("bad integer in %s\n", opt_str);
22136 +                               break;
22137 +                       }
22138 +                       opt->rdhash = n;
22139 +                       err = 0;
22140 +                       opt->type = token;
22141 +                       break;
22142 +
22143 +               case Opt_trunc_xino:
22144 +               case Opt_notrunc_xino:
22145 +               case Opt_noxino:
22146 +               case Opt_trunc_xib:
22147 +               case Opt_notrunc_xib:
22148 +               case Opt_shwh:
22149 +               case Opt_noshwh:
22150 +               case Opt_plink:
22151 +               case Opt_noplink:
22152 +               case Opt_list_plink:
22153 +               case Opt_dio:
22154 +               case Opt_nodio:
22155 +               case Opt_diropq_a:
22156 +               case Opt_diropq_w:
22157 +               case Opt_warn_perm:
22158 +               case Opt_nowarn_perm:
22159 +               case Opt_refrof:
22160 +               case Opt_norefrof:
22161 +               case Opt_verbose:
22162 +               case Opt_noverbose:
22163 +               case Opt_sum:
22164 +               case Opt_nosum:
22165 +               case Opt_wsum:
22166 +               case Opt_rdblk_def:
22167 +               case Opt_rdhash_def:
22168 +                       err = 0;
22169 +                       opt->type = token;
22170 +                       break;
22171 +
22172 +               case Opt_udba:
22173 +                       opt->udba = udba_val(a->args[0].from);
22174 +                       if (opt->udba >= 0) {
22175 +                               err = 0;
22176 +                               opt->type = token;
22177 +                       } else
22178 +                               pr_err("wrong value, %s\n", opt_str);
22179 +                       break;
22180 +
22181 +               case Opt_wbr_create:
22182 +                       u.create = &opt->wbr_create;
22183 +                       u.create->wbr_create
22184 +                               = au_wbr_create_val(a->args[0].from, u.create);
22185 +                       if (u.create->wbr_create >= 0) {
22186 +                               err = 0;
22187 +                               opt->type = token;
22188 +                       } else
22189 +                               pr_err("wrong value, %s\n", opt_str);
22190 +                       break;
22191 +               case Opt_wbr_copyup:
22192 +                       opt->wbr_copyup = au_wbr_copyup_val(a->args[0].from);
22193 +                       if (opt->wbr_copyup >= 0) {
22194 +                               err = 0;
22195 +                               opt->type = token;
22196 +                       } else
22197 +                               pr_err("wrong value, %s\n", opt_str);
22198 +                       break;
22199 +
22200 +               case Opt_ignore:
22201 +                       pr_warn("ignored %s\n", opt_str);
22202 +                       /*FALLTHROUGH*/
22203 +               case Opt_ignore_silent:
22204 +                       skipped = 1;
22205 +                       err = 0;
22206 +                       break;
22207 +               case Opt_err:
22208 +                       pr_err("unknown option %s\n", opt_str);
22209 +                       break;
22210 +               }
22211 +
22212 +               if (!err && !skipped) {
22213 +                       if (unlikely(++opt > opt_tail)) {
22214 +                               err = -E2BIG;
22215 +                               opt--;
22216 +                               opt->type = Opt_tail;
22217 +                               break;
22218 +                       }
22219 +                       opt->type = Opt_tail;
22220 +               }
22221 +       }
22222 +
22223 +       kfree(a);
22224 +       dump_opts(opts);
22225 +       if (unlikely(err))
22226 +               au_opts_free(opts);
22227 +
22228 +out:
22229 +       return err;
22230 +}
22231 +
22232 +static int au_opt_wbr_create(struct super_block *sb,
22233 +                            struct au_opt_wbr_create *create)
22234 +{
22235 +       int err;
22236 +       struct au_sbinfo *sbinfo;
22237 +
22238 +       SiMustWriteLock(sb);
22239 +
22240 +       err = 1; /* handled */
22241 +       sbinfo = au_sbi(sb);
22242 +       if (sbinfo->si_wbr_create_ops->fin) {
22243 +               err = sbinfo->si_wbr_create_ops->fin(sb);
22244 +               if (!err)
22245 +                       err = 1;
22246 +       }
22247 +
22248 +       sbinfo->si_wbr_create = create->wbr_create;
22249 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + create->wbr_create;
22250 +       switch (create->wbr_create) {
22251 +       case AuWbrCreate_MFSRRV:
22252 +       case AuWbrCreate_MFSRR:
22253 +       case AuWbrCreate_PMFSRR:
22254 +       case AuWbrCreate_PMFSRRV:
22255 +               sbinfo->si_wbr_mfs.mfsrr_watermark = create->mfsrr_watermark;
22256 +               /*FALLTHROUGH*/
22257 +       case AuWbrCreate_MFS:
22258 +       case AuWbrCreate_MFSV:
22259 +       case AuWbrCreate_PMFS:
22260 +       case AuWbrCreate_PMFSV:
22261 +               sbinfo->si_wbr_mfs.mfs_expire
22262 +                       = msecs_to_jiffies(create->mfs_second * MSEC_PER_SEC);
22263 +               break;
22264 +       }
22265 +
22266 +       if (sbinfo->si_wbr_create_ops->init)
22267 +               sbinfo->si_wbr_create_ops->init(sb); /* ignore */
22268 +
22269 +       return err;
22270 +}
22271 +
22272 +/*
22273 + * returns,
22274 + * plus: processed without an error
22275 + * zero: unprocessed
22276 + */
22277 +static int au_opt_simple(struct super_block *sb, struct au_opt *opt,
22278 +                        struct au_opts *opts)
22279 +{
22280 +       int err;
22281 +       struct au_sbinfo *sbinfo;
22282 +
22283 +       SiMustWriteLock(sb);
22284 +
22285 +       err = 1; /* handled */
22286 +       sbinfo = au_sbi(sb);
22287 +       switch (opt->type) {
22288 +       case Opt_udba:
22289 +               sbinfo->si_mntflags &= ~AuOptMask_UDBA;
22290 +               sbinfo->si_mntflags |= opt->udba;
22291 +               opts->given_udba |= opt->udba;
22292 +               break;
22293 +
22294 +       case Opt_plink:
22295 +               au_opt_set(sbinfo->si_mntflags, PLINK);
22296 +               break;
22297 +       case Opt_noplink:
22298 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
22299 +                       au_plink_put(sb, /*verbose*/1);
22300 +               au_opt_clr(sbinfo->si_mntflags, PLINK);
22301 +               break;
22302 +       case Opt_list_plink:
22303 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
22304 +                       au_plink_list(sb);
22305 +               break;
22306 +
22307 +       case Opt_dio:
22308 +               au_opt_set(sbinfo->si_mntflags, DIO);
22309 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
22310 +               break;
22311 +       case Opt_nodio:
22312 +               au_opt_clr(sbinfo->si_mntflags, DIO);
22313 +               au_fset_opts(opts->flags, REFRESH_DYAOP);
22314 +               break;
22315 +
22316 +       case Opt_diropq_a:
22317 +               au_opt_set(sbinfo->si_mntflags, ALWAYS_DIROPQ);
22318 +               break;
22319 +       case Opt_diropq_w:
22320 +               au_opt_clr(sbinfo->si_mntflags, ALWAYS_DIROPQ);
22321 +               break;
22322 +
22323 +       case Opt_warn_perm:
22324 +               au_opt_set(sbinfo->si_mntflags, WARN_PERM);
22325 +               break;
22326 +       case Opt_nowarn_perm:
22327 +               au_opt_clr(sbinfo->si_mntflags, WARN_PERM);
22328 +               break;
22329 +
22330 +       case Opt_refrof:
22331 +               au_opt_set(sbinfo->si_mntflags, REFROF);
22332 +               break;
22333 +       case Opt_norefrof:
22334 +               au_opt_clr(sbinfo->si_mntflags, REFROF);
22335 +               break;
22336 +
22337 +       case Opt_verbose:
22338 +               au_opt_set(sbinfo->si_mntflags, VERBOSE);
22339 +               break;
22340 +       case Opt_noverbose:
22341 +               au_opt_clr(sbinfo->si_mntflags, VERBOSE);
22342 +               break;
22343 +
22344 +       case Opt_sum:
22345 +               au_opt_set(sbinfo->si_mntflags, SUM);
22346 +               break;
22347 +       case Opt_wsum:
22348 +               au_opt_clr(sbinfo->si_mntflags, SUM);
22349 +               au_opt_set(sbinfo->si_mntflags, SUM_W);
22350 +       case Opt_nosum:
22351 +               au_opt_clr(sbinfo->si_mntflags, SUM);
22352 +               au_opt_clr(sbinfo->si_mntflags, SUM_W);
22353 +               break;
22354 +
22355 +       case Opt_wbr_create:
22356 +               err = au_opt_wbr_create(sb, &opt->wbr_create);
22357 +               break;
22358 +       case Opt_wbr_copyup:
22359 +               sbinfo->si_wbr_copyup = opt->wbr_copyup;
22360 +               sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + opt->wbr_copyup;
22361 +               break;
22362 +
22363 +       case Opt_dirwh:
22364 +               sbinfo->si_dirwh = opt->dirwh;
22365 +               break;
22366 +
22367 +       case Opt_rdcache:
22368 +               sbinfo->si_rdcache
22369 +                       = msecs_to_jiffies(opt->rdcache * MSEC_PER_SEC);
22370 +               break;
22371 +       case Opt_rdblk:
22372 +               sbinfo->si_rdblk = opt->rdblk;
22373 +               break;
22374 +       case Opt_rdblk_def:
22375 +               sbinfo->si_rdblk = AUFS_RDBLK_DEF;
22376 +               break;
22377 +       case Opt_rdhash:
22378 +               sbinfo->si_rdhash = opt->rdhash;
22379 +               break;
22380 +       case Opt_rdhash_def:
22381 +               sbinfo->si_rdhash = AUFS_RDHASH_DEF;
22382 +               break;
22383 +
22384 +       case Opt_shwh:
22385 +               au_opt_set(sbinfo->si_mntflags, SHWH);
22386 +               break;
22387 +       case Opt_noshwh:
22388 +               au_opt_clr(sbinfo->si_mntflags, SHWH);
22389 +               break;
22390 +
22391 +       case Opt_trunc_xino:
22392 +               au_opt_set(sbinfo->si_mntflags, TRUNC_XINO);
22393 +               break;
22394 +       case Opt_notrunc_xino:
22395 +               au_opt_clr(sbinfo->si_mntflags, TRUNC_XINO);
22396 +               break;
22397 +
22398 +       case Opt_trunc_xino_path:
22399 +       case Opt_itrunc_xino:
22400 +               err = au_xino_trunc(sb, opt->xino_itrunc.bindex);
22401 +               if (!err)
22402 +                       err = 1;
22403 +               break;
22404 +
22405 +       case Opt_trunc_xib:
22406 +               au_fset_opts(opts->flags, TRUNC_XIB);
22407 +               break;
22408 +       case Opt_notrunc_xib:
22409 +               au_fclr_opts(opts->flags, TRUNC_XIB);
22410 +               break;
22411 +
22412 +       default:
22413 +               err = 0;
22414 +               break;
22415 +       }
22416 +
22417 +       return err;
22418 +}
22419 +
22420 +/*
22421 + * returns tri-state.
22422 + * plus: processed without an error
22423 + * zero: unprocessed
22424 + * minus: error
22425 + */
22426 +static int au_opt_br(struct super_block *sb, struct au_opt *opt,
22427 +                    struct au_opts *opts)
22428 +{
22429 +       int err, do_refresh;
22430 +
22431 +       err = 0;
22432 +       switch (opt->type) {
22433 +       case Opt_append:
22434 +               opt->add.bindex = au_sbend(sb) + 1;
22435 +               if (opt->add.bindex < 0)
22436 +                       opt->add.bindex = 0;
22437 +               goto add;
22438 +       case Opt_prepend:
22439 +               opt->add.bindex = 0;
22440 +       add: /* indented label */
22441 +       case Opt_add:
22442 +               err = au_br_add(sb, &opt->add,
22443 +                               au_ftest_opts(opts->flags, REMOUNT));
22444 +               if (!err) {
22445 +                       err = 1;
22446 +                       au_fset_opts(opts->flags, REFRESH);
22447 +               }
22448 +               break;
22449 +
22450 +       case Opt_del:
22451 +       case Opt_idel:
22452 +               err = au_br_del(sb, &opt->del,
22453 +                               au_ftest_opts(opts->flags, REMOUNT));
22454 +               if (!err) {
22455 +                       err = 1;
22456 +                       au_fset_opts(opts->flags, TRUNC_XIB);
22457 +                       au_fset_opts(opts->flags, REFRESH);
22458 +               }
22459 +               break;
22460 +
22461 +       case Opt_mod:
22462 +       case Opt_imod:
22463 +               err = au_br_mod(sb, &opt->mod,
22464 +                               au_ftest_opts(opts->flags, REMOUNT),
22465 +                               &do_refresh);
22466 +               if (!err) {
22467 +                       err = 1;
22468 +                       if (do_refresh)
22469 +                               au_fset_opts(opts->flags, REFRESH);
22470 +               }
22471 +               break;
22472 +       }
22473 +
22474 +       return err;
22475 +}
22476 +
22477 +static int au_opt_xino(struct super_block *sb, struct au_opt *opt,
22478 +                      struct au_opt_xino **opt_xino,
22479 +                      struct au_opts *opts)
22480 +{
22481 +       int err;
22482 +       aufs_bindex_t bend, bindex;
22483 +       struct dentry *root, *parent, *h_root;
22484 +
22485 +       err = 0;
22486 +       switch (opt->type) {
22487 +       case Opt_xino:
22488 +               err = au_xino_set(sb, &opt->xino,
22489 +                                 !!au_ftest_opts(opts->flags, REMOUNT));
22490 +               if (unlikely(err))
22491 +                       break;
22492 +
22493 +               *opt_xino = &opt->xino;
22494 +               au_xino_brid_set(sb, -1);
22495 +
22496 +               /* safe d_parent access */
22497 +               parent = opt->xino.file->f_dentry->d_parent;
22498 +               root = sb->s_root;
22499 +               bend = au_sbend(sb);
22500 +               for (bindex = 0; bindex <= bend; bindex++) {
22501 +                       h_root = au_h_dptr(root, bindex);
22502 +                       if (h_root == parent) {
22503 +                               au_xino_brid_set(sb, au_sbr_id(sb, bindex));
22504 +                               break;
22505 +                       }
22506 +               }
22507 +               break;
22508 +
22509 +       case Opt_noxino:
22510 +               au_xino_clr(sb);
22511 +               au_xino_brid_set(sb, -1);
22512 +               *opt_xino = (void *)-1;
22513 +               break;
22514 +       }
22515 +
22516 +       return err;
22517 +}
22518 +
22519 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
22520 +                  unsigned int pending)
22521 +{
22522 +       int err;
22523 +       aufs_bindex_t bindex, bend;
22524 +       unsigned char do_plink, skip, do_free;
22525 +       struct au_branch *br;
22526 +       struct au_wbr *wbr;
22527 +       struct dentry *root;
22528 +       struct inode *dir, *h_dir;
22529 +       struct au_sbinfo *sbinfo;
22530 +       struct au_hinode *hdir;
22531 +
22532 +       SiMustAnyLock(sb);
22533 +
22534 +       sbinfo = au_sbi(sb);
22535 +       AuDebugOn(!(sbinfo->si_mntflags & AuOptMask_UDBA));
22536 +
22537 +       if (!(sb_flags & MS_RDONLY)) {
22538 +               if (unlikely(!au_br_writable(au_sbr_perm(sb, 0))))
22539 +                       pr_warn("first branch should be rw\n");
22540 +               if (unlikely(au_opt_test(sbinfo->si_mntflags, SHWH)))
22541 +                       pr_warn("shwh should be used with ro\n");
22542 +       }
22543 +
22544 +       if (au_opt_test((sbinfo->si_mntflags | pending), UDBA_HNOTIFY)
22545 +           && !au_opt_test(sbinfo->si_mntflags, XINO))
22546 +               pr_warn("udba=*notify requires xino\n");
22547 +
22548 +       err = 0;
22549 +       root = sb->s_root;
22550 +       dir = root->d_inode;
22551 +       do_plink = !!au_opt_test(sbinfo->si_mntflags, PLINK);
22552 +       bend = au_sbend(sb);
22553 +       for (bindex = 0; !err && bindex <= bend; bindex++) {
22554 +               skip = 0;
22555 +               h_dir = au_h_iptr(dir, bindex);
22556 +               br = au_sbr(sb, bindex);
22557 +               do_free = 0;
22558 +
22559 +               wbr = br->br_wbr;
22560 +               if (wbr)
22561 +                       wbr_wh_read_lock(wbr);
22562 +
22563 +               if (!au_br_writable(br->br_perm)) {
22564 +                       do_free = !!wbr;
22565 +                       skip = (!wbr
22566 +                               || (!wbr->wbr_whbase
22567 +                                   && !wbr->wbr_plink
22568 +                                   && !wbr->wbr_orph));
22569 +               } else if (!au_br_wh_linkable(br->br_perm)) {
22570 +                       /* skip = (!br->br_whbase && !br->br_orph); */
22571 +                       skip = (!wbr || !wbr->wbr_whbase);
22572 +                       if (skip && wbr) {
22573 +                               if (do_plink)
22574 +                                       skip = !!wbr->wbr_plink;
22575 +                               else
22576 +                                       skip = !wbr->wbr_plink;
22577 +                       }
22578 +               } else {
22579 +                       /* skip = (br->br_whbase && br->br_ohph); */
22580 +                       skip = (wbr && wbr->wbr_whbase);
22581 +                       if (skip) {
22582 +                               if (do_plink)
22583 +                                       skip = !!wbr->wbr_plink;
22584 +                               else
22585 +                                       skip = !wbr->wbr_plink;
22586 +                       }
22587 +               }
22588 +               if (wbr)
22589 +                       wbr_wh_read_unlock(wbr);
22590 +
22591 +               if (skip)
22592 +                       continue;
22593 +
22594 +               hdir = au_hi(dir, bindex);
22595 +               au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
22596 +               if (wbr)
22597 +                       wbr_wh_write_lock(wbr);
22598 +               err = au_wh_init(br, sb);
22599 +               if (wbr)
22600 +                       wbr_wh_write_unlock(wbr);
22601 +               au_hn_imtx_unlock(hdir);
22602 +
22603 +               if (!err && do_free) {
22604 +                       kfree(wbr);
22605 +                       br->br_wbr = NULL;
22606 +               }
22607 +       }
22608 +
22609 +       return err;
22610 +}
22611 +
22612 +int au_opts_mount(struct super_block *sb, struct au_opts *opts)
22613 +{
22614 +       int err;
22615 +       unsigned int tmp;
22616 +       aufs_bindex_t bindex, bend;
22617 +       struct au_opt *opt;
22618 +       struct au_opt_xino *opt_xino, xino;
22619 +       struct au_sbinfo *sbinfo;
22620 +       struct au_branch *br;
22621 +
22622 +       SiMustWriteLock(sb);
22623 +
22624 +       err = 0;
22625 +       opt_xino = NULL;
22626 +       opt = opts->opt;
22627 +       while (err >= 0 && opt->type != Opt_tail)
22628 +               err = au_opt_simple(sb, opt++, opts);
22629 +       if (err > 0)
22630 +               err = 0;
22631 +       else if (unlikely(err < 0))
22632 +               goto out;
22633 +
22634 +       /* disable xino and udba temporary */
22635 +       sbinfo = au_sbi(sb);
22636 +       tmp = sbinfo->si_mntflags;
22637 +       au_opt_clr(sbinfo->si_mntflags, XINO);
22638 +       au_opt_set_udba(sbinfo->si_mntflags, UDBA_REVAL);
22639 +
22640 +       opt = opts->opt;
22641 +       while (err >= 0 && opt->type != Opt_tail)
22642 +               err = au_opt_br(sb, opt++, opts);
22643 +       if (err > 0)
22644 +               err = 0;
22645 +       else if (unlikely(err < 0))
22646 +               goto out;
22647 +
22648 +       bend = au_sbend(sb);
22649 +       if (unlikely(bend < 0)) {
22650 +               err = -EINVAL;
22651 +               pr_err("no branches\n");
22652 +               goto out;
22653 +       }
22654 +
22655 +       if (au_opt_test(tmp, XINO))
22656 +               au_opt_set(sbinfo->si_mntflags, XINO);
22657 +       opt = opts->opt;
22658 +       while (!err && opt->type != Opt_tail)
22659 +               err = au_opt_xino(sb, opt++, &opt_xino, opts);
22660 +       if (unlikely(err))
22661 +               goto out;
22662 +
22663 +       err = au_opts_verify(sb, sb->s_flags, tmp);
22664 +       if (unlikely(err))
22665 +               goto out;
22666 +
22667 +       /* restore xino */
22668 +       if (au_opt_test(tmp, XINO) && !opt_xino) {
22669 +               xino.file = au_xino_def(sb);
22670 +               err = PTR_ERR(xino.file);
22671 +               if (IS_ERR(xino.file))
22672 +                       goto out;
22673 +
22674 +               err = au_xino_set(sb, &xino, /*remount*/0);
22675 +               fput(xino.file);
22676 +               if (unlikely(err))
22677 +                       goto out;
22678 +       }
22679 +
22680 +       /* restore udba */
22681 +       tmp &= AuOptMask_UDBA;
22682 +       sbinfo->si_mntflags &= ~AuOptMask_UDBA;
22683 +       sbinfo->si_mntflags |= tmp;
22684 +       bend = au_sbend(sb);
22685 +       for (bindex = 0; bindex <= bend; bindex++) {
22686 +               br = au_sbr(sb, bindex);
22687 +               err = au_hnotify_reset_br(tmp, br, br->br_perm);
22688 +               if (unlikely(err))
22689 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
22690 +                               bindex, err);
22691 +               /* go on even if err */
22692 +       }
22693 +       if (au_opt_test(tmp, UDBA_HNOTIFY)) {
22694 +               struct inode *dir = sb->s_root->d_inode;
22695 +               au_hn_reset(dir, au_hi_flags(dir, /*isdir*/1) & ~AuHi_XINO);
22696 +       }
22697 +
22698 +out:
22699 +       return err;
22700 +}
22701 +
22702 +int au_opts_remount(struct super_block *sb, struct au_opts *opts)
22703 +{
22704 +       int err, rerr;
22705 +       struct inode *dir;
22706 +       struct au_opt_xino *opt_xino;
22707 +       struct au_opt *opt;
22708 +       struct au_sbinfo *sbinfo;
22709 +
22710 +       SiMustWriteLock(sb);
22711 +
22712 +       dir = sb->s_root->d_inode;
22713 +       sbinfo = au_sbi(sb);
22714 +       err = 0;
22715 +       opt_xino = NULL;
22716 +       opt = opts->opt;
22717 +       while (err >= 0 && opt->type != Opt_tail) {
22718 +               err = au_opt_simple(sb, opt, opts);
22719 +               if (!err)
22720 +                       err = au_opt_br(sb, opt, opts);
22721 +               if (!err)
22722 +                       err = au_opt_xino(sb, opt, &opt_xino, opts);
22723 +               opt++;
22724 +       }
22725 +       if (err > 0)
22726 +               err = 0;
22727 +       AuTraceErr(err);
22728 +       /* go on even err */
22729 +
22730 +       rerr = au_opts_verify(sb, opts->sb_flags, /*pending*/0);
22731 +       if (unlikely(rerr && !err))
22732 +               err = rerr;
22733 +
22734 +       if (au_ftest_opts(opts->flags, TRUNC_XIB)) {
22735 +               rerr = au_xib_trunc(sb);
22736 +               if (unlikely(rerr && !err))
22737 +                       err = rerr;
22738 +       }
22739 +
22740 +       /* will be handled by the caller */
22741 +       if (!au_ftest_opts(opts->flags, REFRESH)
22742 +           && (opts->given_udba || au_opt_test(sbinfo->si_mntflags, XINO)))
22743 +               au_fset_opts(opts->flags, REFRESH);
22744 +
22745 +       AuDbg("status 0x%x\n", opts->flags);
22746 +       return err;
22747 +}
22748 +
22749 +/* ---------------------------------------------------------------------- */
22750 +
22751 +unsigned int au_opt_udba(struct super_block *sb)
22752 +{
22753 +       return au_mntflags(sb) & AuOptMask_UDBA;
22754 +}
22755 diff -urN /usr/share/empty/fs/aufs/opts.h linux/fs/aufs/opts.h
22756 --- /usr/share/empty/fs/aufs/opts.h     1970-01-01 01:00:00.000000000 +0100
22757 +++ linux/fs/aufs/opts.h        2014-04-24 22:11:10.855269116 +0200
22758 @@ -0,0 +1,210 @@
22759 +/*
22760 + * Copyright (C) 2005-2014 Junjiro R. Okajima
22761 + *
22762 + * This program, aufs is free software; you can redistribute it and/or modify
22763 + * it under the terms of the GNU General Public License as published by
22764 + * the Free Software Foundation; either version 2 of the License, or
22765 + * (at your option) any later version.
22766 + *
22767 + * This program is distributed in the hope that it will be useful,
22768 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22769 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22770 + * GNU General Public License for more details.
22771 + *
22772 + * You should have received a copy of the GNU General Public License
22773 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22774 + */
22775 +
22776 +/*
22777 + * mount options/flags
22778 + */
22779 +
22780 +#ifndef __AUFS_OPTS_H__
22781 +#define __AUFS_OPTS_H__
22782 +
22783 +#ifdef __KERNEL__
22784 +
22785 +#include <linux/path.h>
22786 +
22787 +struct file;
22788 +struct super_block;
22789 +
22790 +/* ---------------------------------------------------------------------- */
22791 +
22792 +/* mount flags */
22793 +#define AuOpt_XINO             1               /* external inode number bitmap
22794 +                                                  and translation table */
22795 +#define AuOpt_TRUNC_XINO       (1 << 1)        /* truncate xino files */
22796 +#define AuOpt_UDBA_NONE                (1 << 2)        /* users direct branch access */
22797 +#define AuOpt_UDBA_REVAL       (1 << 3)
22798 +#define AuOpt_UDBA_HNOTIFY     (1 << 4)
22799 +#define AuOpt_SHWH             (1 << 5)        /* show whiteout */
22800 +#define AuOpt_PLINK            (1 << 6)        /* pseudo-link */
22801 +#define AuOpt_DIRPERM1         (1 << 7)        /* unimplemented */
22802 +#define AuOpt_REFROF           (1 << 8)        /* unimplemented */
22803 +#define AuOpt_ALWAYS_DIROPQ    (1 << 9)        /* policy to creating diropq */
22804 +#define AuOpt_SUM              (1 << 10)       /* summation for statfs(2) */
22805 +#define AuOpt_SUM_W            (1 << 11)       /* unimplemented */
22806 +#define AuOpt_WARN_PERM                (1 << 12)       /* warn when add-branch */
22807 +#define AuOpt_VERBOSE          (1 << 13)       /* busy inode when del-branch */
22808 +#define AuOpt_DIO              (1 << 14)       /* direct io */
22809 +
22810 +#ifndef CONFIG_AUFS_HNOTIFY
22811 +#undef AuOpt_UDBA_HNOTIFY
22812 +#define AuOpt_UDBA_HNOTIFY     0
22813 +#endif
22814 +#ifndef CONFIG_AUFS_SHWH
22815 +#undef AuOpt_SHWH
22816 +#define AuOpt_SHWH             0
22817 +#endif
22818 +
22819 +#define AuOpt_Def      (AuOpt_XINO \
22820 +                        | AuOpt_UDBA_REVAL \
22821 +                        | AuOpt_PLINK \
22822 +                        /* | AuOpt_DIRPERM1 */ \
22823 +                        | AuOpt_WARN_PERM)
22824 +#define AuOptMask_UDBA (AuOpt_UDBA_NONE \
22825 +                        | AuOpt_UDBA_REVAL \
22826 +                        | AuOpt_UDBA_HNOTIFY)
22827 +
22828 +#define au_opt_test(flags, name)       (flags & AuOpt_##name)
22829 +#define au_opt_set(flags, name) do { \
22830 +       BUILD_BUG_ON(AuOpt_##name & AuOptMask_UDBA); \
22831 +       ((flags) |= AuOpt_##name); \
22832 +} while (0)
22833 +#define au_opt_set_udba(flags, name) do { \
22834 +       (flags) &= ~AuOptMask_UDBA; \
22835 +       ((flags) |= AuOpt_##name); \
22836 +} while (0)
22837 +#define au_opt_clr(flags, name) do { \
22838 +       ((flags) &= ~AuOpt_##name); \
22839 +} while (0)
22840 +
22841 +static inline unsigned int au_opts_plink(unsigned int mntflags)
22842 +{
22843 +#ifdef CONFIG_PROC_FS
22844 +       return mntflags;
22845 +#else
22846 +       return mntflags & ~AuOpt_PLINK;
22847 +#endif
22848 +}
22849 +
22850 +/* ---------------------------------------------------------------------- */
22851 +
22852 +/* policies to select one among multiple writable branches */
22853 +enum {
22854 +       AuWbrCreate_TDP,        /* top down parent */
22855 +       AuWbrCreate_RR,         /* round robin */
22856 +       AuWbrCreate_MFS,        /* most free space */
22857 +       AuWbrCreate_MFSV,       /* mfs with seconds */
22858 +       AuWbrCreate_MFSRR,      /* mfs then rr */
22859 +       AuWbrCreate_MFSRRV,     /* mfs then rr with seconds */
22860 +       AuWbrCreate_PMFS,       /* parent and mfs */
22861 +       AuWbrCreate_PMFSV,      /* parent and mfs with seconds */
22862 +       AuWbrCreate_PMFSRR,     /* parent, mfs and round-robin */
22863 +       AuWbrCreate_PMFSRRV,    /* plus seconds */
22864 +
22865 +       AuWbrCreate_Def = AuWbrCreate_TDP
22866 +};
22867 +
22868 +enum {
22869 +       AuWbrCopyup_TDP,        /* top down parent */
22870 +       AuWbrCopyup_BUP,        /* bottom up parent */
22871 +       AuWbrCopyup_BU,         /* bottom up */
22872 +
22873 +       AuWbrCopyup_Def = AuWbrCopyup_TDP
22874 +};
22875 +
22876 +/* ---------------------------------------------------------------------- */
22877 +
22878 +struct au_opt_add {
22879 +       aufs_bindex_t   bindex;
22880 +       char            *pathname;
22881 +       int             perm;
22882 +       struct path     path;
22883 +};
22884 +
22885 +struct au_opt_del {
22886 +       char            *pathname;
22887 +       struct path     h_path;
22888 +};
22889 +
22890 +struct au_opt_mod {
22891 +       char            *path;
22892 +       int             perm;
22893 +       struct dentry   *h_root;
22894 +};
22895 +
22896 +struct au_opt_xino {
22897 +       char            *path;
22898 +       struct file     *file;
22899 +};
22900 +
22901 +struct au_opt_xino_itrunc {
22902 +       aufs_bindex_t   bindex;
22903 +};
22904 +
22905 +struct au_opt_wbr_create {
22906 +       int                     wbr_create;
22907 +       int                     mfs_second;
22908 +       unsigned long long      mfsrr_watermark;
22909 +};
22910 +
22911 +struct au_opt {
22912 +       int type;
22913 +       union {
22914 +               struct au_opt_xino      xino;
22915 +               struct au_opt_xino_itrunc xino_itrunc;
22916 +               struct au_opt_add       add;
22917 +               struct au_opt_del       del;
22918 +               struct au_opt_mod       mod;
22919 +               int                     dirwh;
22920 +               int                     rdcache;
22921 +               unsigned int            rdblk;
22922 +               unsigned int            rdhash;
22923 +               int                     udba;
22924 +               struct au_opt_wbr_create wbr_create;
22925 +               int                     wbr_copyup;
22926 +       };
22927 +};
22928 +
22929 +/* opts flags */
22930 +#define AuOpts_REMOUNT         1
22931 +#define AuOpts_REFRESH         (1 << 1)
22932 +#define AuOpts_TRUNC_XIB       (1 << 2)
22933 +#define AuOpts_REFRESH_DYAOP   (1 << 3)
22934 +#define au_ftest_opts(flags, name)     ((flags) & AuOpts_##name)
22935 +#define au_fset_opts(flags, name) \
22936 +       do { (flags) |= AuOpts_##name; } while (0)
22937 +#define au_fclr_opts(flags, name) \
22938 +       do { (flags) &= ~AuOpts_##name; } while (0)
22939 +
22940 +struct au_opts {
22941 +       struct au_opt   *opt;
22942 +       int             max_opt;
22943 +
22944 +       unsigned int    given_udba;
22945 +       unsigned int    flags;
22946 +       unsigned long   sb_flags;
22947 +};
22948 +
22949 +/* ---------------------------------------------------------------------- */
22950 +
22951 +char *au_optstr_br_perm(int brperm);
22952 +const char *au_optstr_udba(int udba);
22953 +const char *au_optstr_wbr_copyup(int wbr_copyup);
22954 +const char *au_optstr_wbr_create(int wbr_create);
22955 +
22956 +void au_opts_free(struct au_opts *opts);
22957 +int au_opts_parse(struct super_block *sb, char *str, struct au_opts *opts);
22958 +int au_opts_verify(struct super_block *sb, unsigned long sb_flags,
22959 +                  unsigned int pending);
22960 +int au_opts_mount(struct super_block *sb, struct au_opts *opts);
22961 +int au_opts_remount(struct super_block *sb, struct au_opts *opts);
22962 +
22963 +unsigned int au_opt_udba(struct super_block *sb);
22964 +
22965 +/* ---------------------------------------------------------------------- */
22966 +
22967 +#endif /* __KERNEL__ */
22968 +#endif /* __AUFS_OPTS_H__ */
22969 diff -urN /usr/share/empty/fs/aufs/plink.c linux/fs/aufs/plink.c
22970 --- /usr/share/empty/fs/aufs/plink.c    1970-01-01 01:00:00.000000000 +0100
22971 +++ linux/fs/aufs/plink.c       2014-04-24 22:11:10.855269116 +0200
22972 @@ -0,0 +1,532 @@
22973 +/*
22974 + * Copyright (C) 2005-2014 Junjiro R. Okajima
22975 + *
22976 + * This program, aufs is free software; you can redistribute it and/or modify
22977 + * it under the terms of the GNU General Public License as published by
22978 + * the Free Software Foundation; either version 2 of the License, or
22979 + * (at your option) any later version.
22980 + *
22981 + * This program is distributed in the hope that it will be useful,
22982 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
22983 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
22984 + * GNU General Public License for more details.
22985 + *
22986 + * You should have received a copy of the GNU General Public License
22987 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
22988 + */
22989 +
22990 +/*
22991 + * pseudo-link
22992 + */
22993 +
22994 +#include "aufs.h"
22995 +
22996 +/*
22997 + * the pseudo-link maintenance mode.
22998 + * during a user process maintains the pseudo-links,
22999 + * prohibit adding a new plink and branch manipulation.
23000 + *
23001 + * Flags
23002 + * NOPLM:
23003 + *     For entry functions which will handle plink, and i_mutex is already held
23004 + *     in VFS.
23005 + *     They cannot wait and should return an error at once.
23006 + *     Callers has to check the error.
23007 + * NOPLMW:
23008 + *     For entry functions which will handle plink, but i_mutex is not held
23009 + *     in VFS.
23010 + *     They can wait the plink maintenance mode to finish.
23011 + *
23012 + * They behave like F_SETLK and F_SETLKW.
23013 + * If the caller never handle plink, then both flags are unnecessary.
23014 + */
23015 +
23016 +int au_plink_maint(struct super_block *sb, int flags)
23017 +{
23018 +       int err;
23019 +       pid_t pid, ppid;
23020 +       struct au_sbinfo *sbi;
23021 +
23022 +       SiMustAnyLock(sb);
23023 +
23024 +       err = 0;
23025 +       if (!au_opt_test(au_mntflags(sb), PLINK))
23026 +               goto out;
23027 +
23028 +       sbi = au_sbi(sb);
23029 +       pid = sbi->si_plink_maint_pid;
23030 +       if (!pid || pid == current->pid)
23031 +               goto out;
23032 +
23033 +       /* todo: it highly depends upon /sbin/mount.aufs */
23034 +       rcu_read_lock();
23035 +       ppid = task_pid_vnr(rcu_dereference(current->real_parent));
23036 +       rcu_read_unlock();
23037 +       if (pid == ppid)
23038 +               goto out;
23039 +
23040 +       if (au_ftest_lock(flags, NOPLMW)) {
23041 +               /* if there is no i_mutex lock in VFS, we don't need to wait */
23042 +               /* AuDebugOn(!lockdep_depth(current)); */
23043 +               while (sbi->si_plink_maint_pid) {
23044 +                       si_read_unlock(sb);
23045 +                       /* gave up wake_up_bit() */
23046 +                       wait_event(sbi->si_plink_wq, !sbi->si_plink_maint_pid);
23047 +
23048 +                       if (au_ftest_lock(flags, FLUSH))
23049 +                               au_nwt_flush(&sbi->si_nowait);
23050 +                       si_noflush_read_lock(sb);
23051 +               }
23052 +       } else if (au_ftest_lock(flags, NOPLM)) {
23053 +               AuDbg("ppid %d, pid %d\n", ppid, pid);
23054 +               err = -EAGAIN;
23055 +       }
23056 +
23057 +out:
23058 +       return err;
23059 +}
23060 +
23061 +void au_plink_maint_leave(struct au_sbinfo *sbinfo)
23062 +{
23063 +       spin_lock(&sbinfo->si_plink_maint_lock);
23064 +       sbinfo->si_plink_maint_pid = 0;
23065 +       spin_unlock(&sbinfo->si_plink_maint_lock);
23066 +       wake_up_all(&sbinfo->si_plink_wq);
23067 +}
23068 +
23069 +int au_plink_maint_enter(struct super_block *sb)
23070 +{
23071 +       int err;
23072 +       struct au_sbinfo *sbinfo;
23073 +
23074 +       err = 0;
23075 +       sbinfo = au_sbi(sb);
23076 +       /* make sure i am the only one in this fs */
23077 +       si_write_lock(sb, AuLock_FLUSH);
23078 +       if (au_opt_test(au_mntflags(sb), PLINK)) {
23079 +               spin_lock(&sbinfo->si_plink_maint_lock);
23080 +               if (!sbinfo->si_plink_maint_pid)
23081 +                       sbinfo->si_plink_maint_pid = current->pid;
23082 +               else
23083 +                       err = -EBUSY;
23084 +               spin_unlock(&sbinfo->si_plink_maint_lock);
23085 +       }
23086 +       si_write_unlock(sb);
23087 +
23088 +       return err;
23089 +}
23090 +
23091 +/* ---------------------------------------------------------------------- */
23092 +
23093 +#ifdef CONFIG_AUFS_DEBUG
23094 +void au_plink_list(struct super_block *sb)
23095 +{
23096 +       int i;
23097 +       struct au_sbinfo *sbinfo;
23098 +       struct hlist_head *plink_hlist;
23099 +       struct pseudo_link *plink;
23100 +
23101 +       SiMustAnyLock(sb);
23102 +
23103 +       sbinfo = au_sbi(sb);
23104 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
23105 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
23106 +
23107 +       for (i = 0; i < AuPlink_NHASH; i++) {
23108 +               plink_hlist = &sbinfo->si_plink[i].head;
23109 +               rcu_read_lock();
23110 +               hlist_for_each_entry_rcu(plink, plink_hlist, hlist)
23111 +                       AuDbg("%lu\n", plink->inode->i_ino);
23112 +               rcu_read_unlock();
23113 +       }
23114 +}
23115 +#endif
23116 +
23117 +/* is the inode pseudo-linked? */
23118 +int au_plink_test(struct inode *inode)
23119 +{
23120 +       int found, i;
23121 +       struct au_sbinfo *sbinfo;
23122 +       struct hlist_head *plink_hlist;
23123 +       struct pseudo_link *plink;
23124 +
23125 +       sbinfo = au_sbi(inode->i_sb);
23126 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
23127 +       AuDebugOn(!au_opt_test(au_mntflags(inode->i_sb), PLINK));
23128 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
23129 +
23130 +       found = 0;
23131 +       i = au_plink_hash(inode->i_ino);
23132 +       plink_hlist = &sbinfo->si_plink[i].head;
23133 +       rcu_read_lock();
23134 +       hlist_for_each_entry_rcu(plink, plink_hlist, hlist)
23135 +               if (plink->inode == inode) {
23136 +                       found = 1;
23137 +                       break;
23138 +               }
23139 +       rcu_read_unlock();
23140 +       return found;
23141 +}
23142 +
23143 +/* ---------------------------------------------------------------------- */
23144 +
23145 +/*
23146 + * generate a name for plink.
23147 + * the file will be stored under AUFS_WH_PLINKDIR.
23148 + */
23149 +/* 20 is max digits length of ulong 64 */
23150 +#define PLINK_NAME_LEN ((20 + 1) * 2)
23151 +
23152 +static int plink_name(char *name, int len, struct inode *inode,
23153 +                     aufs_bindex_t bindex)
23154 +{
23155 +       int rlen;
23156 +       struct inode *h_inode;
23157 +
23158 +       h_inode = au_h_iptr(inode, bindex);
23159 +       rlen = snprintf(name, len, "%lu.%lu", inode->i_ino, h_inode->i_ino);
23160 +       return rlen;
23161 +}
23162 +
23163 +struct au_do_plink_lkup_args {
23164 +       struct dentry **errp;
23165 +       struct qstr *tgtname;
23166 +       struct dentry *h_parent;
23167 +       struct au_branch *br;
23168 +};
23169 +
23170 +static struct dentry *au_do_plink_lkup(struct qstr *tgtname,
23171 +                                      struct dentry *h_parent,
23172 +                                      struct au_branch *br)
23173 +{
23174 +       struct dentry *h_dentry;
23175 +       struct mutex *h_mtx;
23176 +
23177 +       h_mtx = &h_parent->d_inode->i_mutex;
23178 +       mutex_lock_nested(h_mtx, AuLsc_I_CHILD2);
23179 +       h_dentry = vfsub_lkup_one(tgtname, h_parent);
23180 +       mutex_unlock(h_mtx);
23181 +       return h_dentry;
23182 +}
23183 +
23184 +static void au_call_do_plink_lkup(void *args)
23185 +{
23186 +       struct au_do_plink_lkup_args *a = args;
23187 +       *a->errp = au_do_plink_lkup(a->tgtname, a->h_parent, a->br);
23188 +}
23189 +
23190 +/* lookup the plink-ed @inode under the branch at @bindex */
23191 +struct dentry *au_plink_lkup(struct inode *inode, aufs_bindex_t bindex)
23192 +{
23193 +       struct dentry *h_dentry, *h_parent;
23194 +       struct au_branch *br;
23195 +       struct inode *h_dir;
23196 +       int wkq_err;
23197 +       char a[PLINK_NAME_LEN];
23198 +       struct qstr tgtname = QSTR_INIT(a, 0);
23199 +
23200 +       AuDebugOn(au_plink_maint(inode->i_sb, AuLock_NOPLM));
23201 +
23202 +       br = au_sbr(inode->i_sb, bindex);
23203 +       h_parent = br->br_wbr->wbr_plink;
23204 +       h_dir = h_parent->d_inode;
23205 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
23206 +
23207 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
23208 +               struct au_do_plink_lkup_args args = {
23209 +                       .errp           = &h_dentry,
23210 +                       .tgtname        = &tgtname,
23211 +                       .h_parent       = h_parent,
23212 +                       .br             = br
23213 +               };
23214 +
23215 +               wkq_err = au_wkq_wait(au_call_do_plink_lkup, &args);
23216 +               if (unlikely(wkq_err))
23217 +                       h_dentry = ERR_PTR(wkq_err);
23218 +       } else
23219 +               h_dentry = au_do_plink_lkup(&tgtname, h_parent, br);
23220 +
23221 +       return h_dentry;
23222 +}
23223 +
23224 +/* create a pseudo-link */
23225 +static int do_whplink(struct qstr *tgt, struct dentry *h_parent,
23226 +                     struct dentry *h_dentry, struct au_branch *br)
23227 +{
23228 +       int err;
23229 +       struct path h_path = {
23230 +               .mnt = au_br_mnt(br)
23231 +       };
23232 +       struct inode *h_dir, *delegated;
23233 +
23234 +       h_dir = h_parent->d_inode;
23235 +       mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_CHILD2);
23236 +again:
23237 +       h_path.dentry = vfsub_lkup_one(tgt, h_parent);
23238 +       err = PTR_ERR(h_path.dentry);
23239 +       if (IS_ERR(h_path.dentry))
23240 +               goto out;
23241 +
23242 +       err = 0;
23243 +       /* wh.plink dir is not monitored */
23244 +       /* todo: is it really safe? */
23245 +       if (h_path.dentry->d_inode
23246 +           && h_path.dentry->d_inode != h_dentry->d_inode) {
23247 +               delegated = NULL;
23248 +               err = vfsub_unlink(h_dir, &h_path, &delegated, /*force*/0);
23249 +               if (unlikely(err == -EWOULDBLOCK)) {
23250 +                       pr_warn("cannot retry for NFSv4 delegation"
23251 +                               " for an internal unlink\n");
23252 +                       iput(delegated);
23253 +               }
23254 +               dput(h_path.dentry);
23255 +               h_path.dentry = NULL;
23256 +               if (!err)
23257 +                       goto again;
23258 +       }
23259 +       if (!err && !h_path.dentry->d_inode) {
23260 +               delegated = NULL;
23261 +               err = vfsub_link(h_dentry, h_dir, &h_path, &delegated);
23262 +               if (unlikely(err == -EWOULDBLOCK)) {
23263 +                       pr_warn("cannot retry for NFSv4 delegation"
23264 +                               " for an internal link\n");
23265 +                       iput(delegated);
23266 +               }
23267 +       }
23268 +       dput(h_path.dentry);
23269 +
23270 +out:
23271 +       mutex_unlock(&h_dir->i_mutex);
23272 +       return err;
23273 +}
23274 +
23275 +struct do_whplink_args {
23276 +       int *errp;
23277 +       struct qstr *tgt;
23278 +       struct dentry *h_parent;
23279 +       struct dentry *h_dentry;
23280 +       struct au_branch *br;
23281 +};
23282 +
23283 +static void call_do_whplink(void *args)
23284 +{
23285 +       struct do_whplink_args *a = args;
23286 +       *a->errp = do_whplink(a->tgt, a->h_parent, a->h_dentry, a->br);
23287 +}
23288 +
23289 +static int whplink(struct dentry *h_dentry, struct inode *inode,
23290 +                  aufs_bindex_t bindex, struct au_branch *br)
23291 +{
23292 +       int err, wkq_err;
23293 +       struct au_wbr *wbr;
23294 +       struct dentry *h_parent;
23295 +       struct inode *h_dir;
23296 +       char a[PLINK_NAME_LEN];
23297 +       struct qstr tgtname = QSTR_INIT(a, 0);
23298 +
23299 +       wbr = au_sbr(inode->i_sb, bindex)->br_wbr;
23300 +       h_parent = wbr->wbr_plink;
23301 +       h_dir = h_parent->d_inode;
23302 +       tgtname.len = plink_name(a, sizeof(a), inode, bindex);
23303 +
23304 +       /* always superio. */
23305 +       if (!uid_eq(current_fsuid(), GLOBAL_ROOT_UID)) {
23306 +               struct do_whplink_args args = {
23307 +                       .errp           = &err,
23308 +                       .tgt            = &tgtname,
23309 +                       .h_parent       = h_parent,
23310 +                       .h_dentry       = h_dentry,
23311 +                       .br             = br
23312 +               };
23313 +               wkq_err = au_wkq_wait(call_do_whplink, &args);
23314 +               if (unlikely(wkq_err))
23315 +                       err = wkq_err;
23316 +       } else
23317 +               err = do_whplink(&tgtname, h_parent, h_dentry, br);
23318 +
23319 +       return err;
23320 +}
23321 +
23322 +/* free a single plink */
23323 +static void do_put_plink(struct pseudo_link *plink, int do_del)
23324 +{
23325 +       if (do_del)
23326 +               hlist_del(&plink->hlist);
23327 +       iput(plink->inode);
23328 +       kfree(plink);
23329 +}
23330 +
23331 +static void do_put_plink_rcu(struct rcu_head *rcu)
23332 +{
23333 +       struct pseudo_link *plink;
23334 +
23335 +       plink = container_of(rcu, struct pseudo_link, rcu);
23336 +       iput(plink->inode);
23337 +       kfree(plink);
23338 +}
23339 +
23340 +/*
23341 + * create a new pseudo-link for @h_dentry on @bindex.
23342 + * the linked inode is held in aufs @inode.
23343 + */
23344 +void au_plink_append(struct inode *inode, aufs_bindex_t bindex,
23345 +                    struct dentry *h_dentry)
23346 +{
23347 +       struct super_block *sb;
23348 +       struct au_sbinfo *sbinfo;
23349 +       struct hlist_head *plink_hlist;
23350 +       struct pseudo_link *plink, *tmp;
23351 +       struct au_sphlhead *sphl;
23352 +       int found, err, cnt, i;
23353 +
23354 +       sb = inode->i_sb;
23355 +       sbinfo = au_sbi(sb);
23356 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
23357 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
23358 +
23359 +       found = au_plink_test(inode);
23360 +       if (found)
23361 +               return;
23362 +
23363 +       i = au_plink_hash(inode->i_ino);
23364 +       sphl = sbinfo->si_plink + i;
23365 +       plink_hlist = &sphl->head;
23366 +       tmp = kmalloc(sizeof(*plink), GFP_NOFS);
23367 +       if (tmp)
23368 +               tmp->inode = au_igrab(inode);
23369 +       else {
23370 +               err = -ENOMEM;
23371 +               goto out;
23372 +       }
23373 +
23374 +       spin_lock(&sphl->spin);
23375 +       hlist_for_each_entry(plink, plink_hlist, hlist) {
23376 +               if (plink->inode == inode) {
23377 +                       found = 1;
23378 +                       break;
23379 +               }
23380 +       }
23381 +       if (!found)
23382 +               hlist_add_head_rcu(&tmp->hlist, plink_hlist);
23383 +       spin_unlock(&sphl->spin);
23384 +       if (!found) {
23385 +               cnt = au_sphl_count(sphl);
23386 +#define msg "unexpectedly unblanced or too many pseudo-links"
23387 +               if (cnt > AUFS_PLINK_WARN)
23388 +                       AuWarn1(msg ", %d\n", cnt);
23389 +#undef msg
23390 +               err = whplink(h_dentry, inode, bindex, au_sbr(sb, bindex));
23391 +       } else {
23392 +               do_put_plink(tmp, 0);
23393 +               return;
23394 +       }
23395 +
23396 +out:
23397 +       if (unlikely(err)) {
23398 +               pr_warn("err %d, damaged pseudo link.\n", err);
23399 +               if (tmp) {
23400 +                       au_sphl_del_rcu(&tmp->hlist, sphl);
23401 +                       call_rcu(&tmp->rcu, do_put_plink_rcu);
23402 +               }
23403 +       }
23404 +}
23405 +
23406 +/* free all plinks */
23407 +void au_plink_put(struct super_block *sb, int verbose)
23408 +{
23409 +       int i, warned;
23410 +       struct au_sbinfo *sbinfo;
23411 +       struct hlist_head *plink_hlist;
23412 +       struct hlist_node *tmp;
23413 +       struct pseudo_link *plink;
23414 +
23415 +       SiMustWriteLock(sb);
23416 +
23417 +       sbinfo = au_sbi(sb);
23418 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
23419 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
23420 +
23421 +       /* no spin_lock since sbinfo is write-locked */
23422 +       warned = 0;
23423 +       for (i = 0; i < AuPlink_NHASH; i++) {
23424 +               plink_hlist = &sbinfo->si_plink[i].head;
23425 +               if (!warned && verbose && !hlist_empty(plink_hlist)) {
23426 +                       pr_warn("pseudo-link is not flushed");
23427 +                       warned = 1;
23428 +               }
23429 +               hlist_for_each_entry_safe(plink, tmp, plink_hlist, hlist)
23430 +                       do_put_plink(plink, 0);
23431 +               INIT_HLIST_HEAD(plink_hlist);
23432 +       }
23433 +}
23434 +
23435 +void au_plink_clean(struct super_block *sb, int verbose)
23436 +{
23437 +       struct dentry *root;
23438 +
23439 +       root = sb->s_root;
23440 +       aufs_write_lock(root);
23441 +       if (au_opt_test(au_mntflags(sb), PLINK))
23442 +               au_plink_put(sb, verbose);
23443 +       aufs_write_unlock(root);
23444 +}
23445 +
23446 +static int au_plink_do_half_refresh(struct inode *inode, aufs_bindex_t br_id)
23447 +{
23448 +       int do_put;
23449 +       aufs_bindex_t bstart, bend, bindex;
23450 +
23451 +       do_put = 0;
23452 +       bstart = au_ibstart(inode);
23453 +       bend = au_ibend(inode);
23454 +       if (bstart >= 0) {
23455 +               for (bindex = bstart; bindex <= bend; bindex++) {
23456 +                       if (!au_h_iptr(inode, bindex)
23457 +                           || au_ii_br_id(inode, bindex) != br_id)
23458 +                               continue;
23459 +                       au_set_h_iptr(inode, bindex, NULL, 0);
23460 +                       do_put = 1;
23461 +                       break;
23462 +               }
23463 +               if (do_put)
23464 +                       for (bindex = bstart; bindex <= bend; bindex++)
23465 +                               if (au_h_iptr(inode, bindex)) {
23466 +                                       do_put = 0;
23467 +                                       break;
23468 +                               }
23469 +       } else
23470 +               do_put = 1;
23471 +
23472 +       return do_put;
23473 +}
23474 +
23475 +/* free the plinks on a branch specified by @br_id */
23476 +void au_plink_half_refresh(struct super_block *sb, aufs_bindex_t br_id)
23477 +{
23478 +       struct au_sbinfo *sbinfo;
23479 +       struct hlist_head *plink_hlist;
23480 +       struct hlist_node *tmp;
23481 +       struct pseudo_link *plink;
23482 +       struct inode *inode;
23483 +       int i, do_put;
23484 +
23485 +       SiMustWriteLock(sb);
23486 +
23487 +       sbinfo = au_sbi(sb);
23488 +       AuDebugOn(!au_opt_test(au_mntflags(sb), PLINK));
23489 +       AuDebugOn(au_plink_maint(sb, AuLock_NOPLM));
23490 +
23491 +       /* no spin_lock since sbinfo is write-locked */
23492 +       for (i = 0; i < AuPlink_NHASH; i++) {
23493 +               plink_hlist = &sbinfo->si_plink[i].head;
23494 +               hlist_for_each_entry_safe(plink, tmp, plink_hlist, hlist) {
23495 +                       inode = au_igrab(plink->inode);
23496 +                       ii_write_lock_child(inode);
23497 +                       do_put = au_plink_do_half_refresh(inode, br_id);
23498 +                       if (do_put)
23499 +                               do_put_plink(plink, 1);
23500 +                       ii_write_unlock(inode);
23501 +                       iput(inode);
23502 +               }
23503 +       }
23504 +}
23505 diff -urN /usr/share/empty/fs/aufs/poll.c linux/fs/aufs/poll.c
23506 --- /usr/share/empty/fs/aufs/poll.c     1970-01-01 01:00:00.000000000 +0100
23507 +++ linux/fs/aufs/poll.c        2014-04-24 22:11:10.855269116 +0200
23508 @@ -0,0 +1,55 @@
23509 +/*
23510 + * Copyright (C) 2005-2014 Junjiro R. Okajima
23511 + *
23512 + * This program, aufs is free software; you can redistribute it and/or modify
23513 + * it under the terms of the GNU General Public License as published by
23514 + * the Free Software Foundation; either version 2 of the License, or
23515 + * (at your option) any later version.
23516 + *
23517 + * This program is distributed in the hope that it will be useful,
23518 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
23519 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
23520 + * GNU General Public License for more details.
23521 + *
23522 + * You should have received a copy of the GNU General Public License
23523 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
23524 + */
23525 +
23526 +/*
23527 + * poll operation
23528 + * There is only one filesystem which implements ->poll operation, currently.
23529 + */
23530 +
23531 +#include "aufs.h"
23532 +
23533 +unsigned int aufs_poll(struct file *file, poll_table *wait)
23534 +{
23535 +       unsigned int mask;
23536 +       int err;
23537 +       struct file *h_file;
23538 +       struct dentry *dentry;
23539 +       struct super_block *sb;
23540 +
23541 +       /* We should pretend an error happened. */
23542 +       mask = POLLERR /* | POLLIN | POLLOUT */;
23543 +       dentry = file->f_dentry;
23544 +       sb = dentry->d_sb;
23545 +       si_read_lock(sb, AuLock_FLUSH | AuLock_NOPLMW);
23546 +       err = au_reval_and_lock_fdi(file, au_reopen_nondir, /*wlock*/0);
23547 +       if (unlikely(err))
23548 +               goto out;
23549 +
23550 +       /* it is not an error if h_file has no operation */
23551 +       mask = DEFAULT_POLLMASK;
23552 +       h_file = au_hf_top(file);
23553 +       if (h_file->f_op->poll)
23554 +               mask = h_file->f_op->poll(h_file, wait);
23555 +
23556 +       di_read_unlock(dentry, AuLock_IR);
23557 +       fi_read_unlock(file);
23558 +
23559 +out:
23560 +       si_read_unlock(sb);
23561 +       AuTraceErr((int)mask);
23562 +       return mask;
23563 +}
23564 diff -urN /usr/share/empty/fs/aufs/procfs.c linux/fs/aufs/procfs.c
23565 --- /usr/share/empty/fs/aufs/procfs.c   1970-01-01 01:00:00.000000000 +0100
23566 +++ linux/fs/aufs/procfs.c      2014-04-24 22:11:10.855269116 +0200
23567 @@ -0,0 +1,169 @@
23568 +/*
23569 + * Copyright (C) 2010-2014 Junjiro R. Okajima
23570 + *
23571 + * This program, aufs is free software; you can redistribute it and/or modify
23572 + * it under the terms of the GNU General Public License as published by
23573 + * the Free Software Foundation; either version 2 of the License, or
23574 + * (at your option) any later version.
23575 + *
23576 + * This program is distributed in the hope that it will be useful,
23577 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
23578 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
23579 + * GNU General Public License for more details.
23580 + *
23581 + * You should have received a copy of the GNU General Public License
23582 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
23583 + */
23584 +
23585 +/*
23586 + * procfs interfaces
23587 + */
23588 +
23589 +#include <linux/proc_fs.h>
23590 +#include "aufs.h"
23591 +
23592 +static int au_procfs_plm_release(struct inode *inode, struct file *file)
23593 +{
23594 +       struct au_sbinfo *sbinfo;
23595 +
23596 +       sbinfo = file->private_data;
23597 +       if (sbinfo) {
23598 +               au_plink_maint_leave(sbinfo);
23599 +               kobject_put(&sbinfo->si_kobj);
23600 +       }
23601 +
23602 +       return 0;
23603 +}
23604 +
23605 +static void au_procfs_plm_write_clean(struct file *file)
23606 +{
23607 +       struct au_sbinfo *sbinfo;
23608 +
23609 +       sbinfo = file->private_data;
23610 +       if (sbinfo)
23611 +               au_plink_clean(sbinfo->si_sb, /*verbose*/0);
23612 +}
23613 +
23614 +static int au_procfs_plm_write_si(struct file *file, unsigned long id)
23615 +{
23616 +       int err;
23617 +       struct super_block *sb;
23618 +       struct au_sbinfo *sbinfo;
23619 +
23620 +       err = -EBUSY;
23621 +       if (unlikely(file->private_data))
23622 +               goto out;
23623 +
23624 +       sb = NULL;
23625 +       /* don't use au_sbilist_lock() here */
23626 +       spin_lock(&au_sbilist.spin);
23627 +       list_for_each_entry(sbinfo, &au_sbilist.head, si_list)
23628 +               if (id == sysaufs_si_id(sbinfo)) {
23629 +                       kobject_get(&sbinfo->si_kobj);
23630 +                       sb = sbinfo->si_sb;
23631 +                       break;
23632 +               }
23633 +       spin_unlock(&au_sbilist.spin);
23634 +
23635 +       err = -EINVAL;
23636 +       if (unlikely(!sb))
23637 +               goto out;
23638 +
23639 +       err = au_plink_maint_enter(sb);
23640 +       if (!err)
23641 +               /* keep kobject_get() */
23642 +               file->private_data = sbinfo;
23643 +       else
23644 +               kobject_put(&sbinfo->si_kobj);
23645 +out:
23646 +       return err;
23647 +}
23648 +
23649 +/*
23650 + * Accept a valid "si=xxxx" only.
23651 + * Once it is accepted successfully, accept "clean" too.
23652 + */
23653 +static ssize_t au_procfs_plm_write(struct file *file, const char __user *ubuf,
23654 +                                  size_t count, loff_t *ppos)
23655 +{
23656 +       ssize_t err;
23657 +       unsigned long id;
23658 +       /* last newline is allowed */
23659 +       char buf[3 + sizeof(unsigned long) * 2 + 1];
23660 +
23661 +       err = -EACCES;
23662 +       if (unlikely(!capable(CAP_SYS_ADMIN)))
23663 +               goto out;
23664 +
23665 +       err = -EINVAL;
23666 +       if (unlikely(count > sizeof(buf)))
23667 +               goto out;
23668 +
23669 +       err = copy_from_user(buf, ubuf, count);
23670 +       if (unlikely(err)) {
23671 +               err = -EFAULT;
23672 +               goto out;
23673 +       }
23674 +       buf[count] = 0;
23675 +
23676 +       err = -EINVAL;
23677 +       if (!strcmp("clean", buf)) {
23678 +               au_procfs_plm_write_clean(file);
23679 +               goto out_success;
23680 +       } else if (unlikely(strncmp("si=", buf, 3)))
23681 +               goto out;
23682 +
23683 +       err = kstrtoul(buf + 3, 16, &id);
23684 +       if (unlikely(err))
23685 +               goto out;
23686 +
23687 +       err = au_procfs_plm_write_si(file, id);
23688 +       if (unlikely(err))
23689 +               goto out;
23690 +
23691 +out_success:
23692 +       err = count; /* success */
23693 +out:
23694 +       return err;
23695 +}
23696 +
23697 +static const struct file_operations au_procfs_plm_fop = {
23698 +       .write          = au_procfs_plm_write,
23699 +       .release        = au_procfs_plm_release,
23700 +       .owner          = THIS_MODULE
23701 +};
23702 +
23703 +/* ---------------------------------------------------------------------- */
23704 +
23705 +static struct proc_dir_entry *au_procfs_dir;
23706 +
23707 +void au_procfs_fin(void)
23708 +{
23709 +       remove_proc_entry(AUFS_PLINK_MAINT_NAME, au_procfs_dir);
23710 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
23711 +}
23712 +
23713 +int __init au_procfs_init(void)
23714 +{
23715 +       int err;
23716 +       struct proc_dir_entry *entry;
23717 +
23718 +       err = -ENOMEM;
23719 +       au_procfs_dir = proc_mkdir(AUFS_PLINK_MAINT_DIR, NULL);
23720 +       if (unlikely(!au_procfs_dir))
23721 +               goto out;
23722 +
23723 +       entry = proc_create(AUFS_PLINK_MAINT_NAME, S_IFREG | S_IWUSR,
23724 +                           au_procfs_dir, &au_procfs_plm_fop);
23725 +       if (unlikely(!entry))
23726 +               goto out_dir;
23727 +
23728 +       err = 0;
23729 +       goto out; /* success */
23730 +
23731 +
23732 +out_dir:
23733 +       remove_proc_entry(AUFS_PLINK_MAINT_DIR, NULL);
23734 +out:
23735 +       return err;
23736 +}
23737 diff -urN /usr/share/empty/fs/aufs/rdu.c linux/fs/aufs/rdu.c
23738 --- /usr/share/empty/fs/aufs/rdu.c      1970-01-01 01:00:00.000000000 +0100
23739 +++ linux/fs/aufs/rdu.c 2014-04-24 22:11:10.858602483 +0200
23740 @@ -0,0 +1,388 @@
23741 +/*
23742 + * Copyright (C) 2005-2014 Junjiro R. Okajima
23743 + *
23744 + * This program, aufs is free software; you can redistribute it and/or modify
23745 + * it under the terms of the GNU General Public License as published by
23746 + * the Free Software Foundation; either version 2 of the License, or
23747 + * (at your option) any later version.
23748 + *
23749 + * This program is distributed in the hope that it will be useful,
23750 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
23751 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
23752 + * GNU General Public License for more details.
23753 + *
23754 + * You should have received a copy of the GNU General Public License
23755 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
23756 + */
23757 +
23758 +/*
23759 + * readdir in userspace.
23760 + */
23761 +
23762 +#include <linux/compat.h>
23763 +#include <linux/fs_stack.h>
23764 +#include <linux/security.h>
23765 +#include "aufs.h"
23766 +
23767 +/* bits for struct aufs_rdu.flags */
23768 +#define        AuRdu_CALLED    1
23769 +#define        AuRdu_CONT      (1 << 1)
23770 +#define        AuRdu_FULL      (1 << 2)
23771 +#define au_ftest_rdu(flags, name)      ((flags) & AuRdu_##name)
23772 +#define au_fset_rdu(flags, name) \
23773 +       do { (flags) |= AuRdu_##name; } while (0)
23774 +#define au_fclr_rdu(flags, name) \
23775 +       do { (flags) &= ~AuRdu_##name; } while (0)
23776 +
23777 +struct au_rdu_arg {
23778 +       struct dir_context              ctx;
23779 +       struct aufs_rdu                 *rdu;
23780 +       union au_rdu_ent_ul             ent;
23781 +       unsigned long                   end;
23782 +
23783 +       struct super_block              *sb;
23784 +       int                             err;
23785 +};
23786 +
23787 +static int au_rdu_fill(struct dir_context *ctx, const char *name, int nlen,
23788 +                      loff_t offset, u64 h_ino, unsigned int d_type)
23789 +{
23790 +       int err, len;
23791 +       struct au_rdu_arg *arg = container_of(ctx, struct au_rdu_arg, ctx);
23792 +       struct aufs_rdu *rdu = arg->rdu;
23793 +       struct au_rdu_ent ent;
23794 +
23795 +       err = 0;
23796 +       arg->err = 0;
23797 +       au_fset_rdu(rdu->cookie.flags, CALLED);
23798 +       len = au_rdu_len(nlen);
23799 +       if (arg->ent.ul + len  < arg->end) {
23800 +               ent.ino = h_ino;
23801 +               ent.bindex = rdu->cookie.bindex;
23802 +               ent.type = d_type;
23803 +               ent.nlen = nlen;
23804 +               if (unlikely(nlen > AUFS_MAX_NAMELEN))
23805 +                       ent.type = DT_UNKNOWN;
23806 +
23807 +               /* unnecessary to support mmap_sem since this is a dir */
23808 +               err = -EFAULT;
23809 +               if (copy_to_user(arg->ent.e, &ent, sizeof(ent)))
23810 +                       goto out;
23811 +               if (copy_to_user(arg->ent.e->name, name, nlen))
23812 +                       goto out;
23813 +               /* the terminating NULL */
23814 +               if (__put_user(0, arg->ent.e->name + nlen))
23815 +                       goto out;
23816 +               err = 0;
23817 +               /* AuDbg("%p, %.*s\n", arg->ent.p, nlen, name); */
23818 +               arg->ent.ul += len;
23819 +               rdu->rent++;
23820 +       } else {
23821 +               err = -EFAULT;
23822 +               au_fset_rdu(rdu->cookie.flags, FULL);
23823 +               rdu->full = 1;
23824 +               rdu->tail = arg->ent;
23825 +       }
23826 +
23827 +out:
23828 +       /* AuTraceErr(err); */
23829 +       return err;
23830 +}
23831 +
23832 +static int au_rdu_do(struct file *h_file, struct au_rdu_arg *arg)
23833 +{
23834 +       int err;
23835 +       loff_t offset;
23836 +       struct au_rdu_cookie *cookie = &arg->rdu->cookie;
23837 +
23838 +       /* we don't have to care (FMODE_32BITHASH | FMODE_64BITHASH) for ext4 */
23839 +       offset = vfsub_llseek(h_file, cookie->h_pos, SEEK_SET);
23840 +       err = offset;
23841 +       if (unlikely(offset != cookie->h_pos))
23842 +               goto out;
23843 +
23844 +       err = 0;
23845 +       do {
23846 +               arg->err = 0;
23847 +               au_fclr_rdu(cookie->flags, CALLED);
23848 +               /* smp_mb(); */
23849 +               err = vfsub_iterate_dir(h_file, &arg->ctx);
23850 +               if (err >= 0)
23851 +                       err = arg->err;
23852 +       } while (!err
23853 +                && au_ftest_rdu(cookie->flags, CALLED)
23854 +                && !au_ftest_rdu(cookie->flags, FULL));
23855 +       cookie->h_pos = h_file->f_pos;
23856 +
23857 +out:
23858 +       AuTraceErr(err);
23859 +       return err;
23860 +}
23861 +
23862 +static int au_rdu(struct file *file, struct aufs_rdu *rdu)
23863 +{
23864 +       int err;
23865 +       aufs_bindex_t bend;
23866 +       struct au_rdu_arg arg = {
23867 +               .ctx = {
23868 +                       .actor = au_diractor(au_rdu_fill)
23869 +               }
23870 +       };
23871 +       struct dentry *dentry;
23872 +       struct inode *inode;
23873 +       struct file *h_file;
23874 +       struct au_rdu_cookie *cookie = &rdu->cookie;
23875 +
23876 +       err = !access_ok(VERIFY_WRITE, rdu->ent.e, rdu->sz);
23877 +       if (unlikely(err)) {
23878 +               err = -EFAULT;
23879 +               AuTraceErr(err);
23880 +               goto out;
23881 +       }
23882 +       rdu->rent = 0;
23883 +       rdu->tail = rdu->ent;
23884 +       rdu->full = 0;
23885 +       arg.rdu = rdu;
23886 +       arg.ent = rdu->ent;
23887 +       arg.end = arg.ent.ul;
23888 +       arg.end += rdu->sz;
23889 +
23890 +       err = -ENOTDIR;
23891 +       if (unlikely(!file->f_op->iterate))
23892 +               goto out;
23893 +
23894 +       err = security_file_permission(file, MAY_READ);
23895 +       AuTraceErr(err);
23896 +       if (unlikely(err))
23897 +               goto out;
23898 +
23899 +       dentry = file->f_dentry;
23900 +       inode = dentry->d_inode;
23901 +#if 1
23902 +       mutex_lock(&inode->i_mutex);
23903 +#else
23904 +       err = mutex_lock_killable(&inode->i_mutex);
23905 +       AuTraceErr(err);
23906 +       if (unlikely(err))
23907 +               goto out;
23908 +#endif
23909 +
23910 +       arg.sb = inode->i_sb;
23911 +       err = si_read_lock(arg.sb, AuLock_FLUSH | AuLock_NOPLM);
23912 +       if (unlikely(err))
23913 +               goto out_mtx;
23914 +       err = au_alive_dir(dentry);
23915 +       if (unlikely(err))
23916 +               goto out_si;
23917 +       /* todo: reval? */
23918 +       fi_read_lock(file);
23919 +
23920 +       err = -EAGAIN;
23921 +       if (unlikely(au_ftest_rdu(cookie->flags, CONT)
23922 +                    && cookie->generation != au_figen(file)))
23923 +               goto out_unlock;
23924 +
23925 +       err = 0;
23926 +       if (!rdu->blk) {
23927 +               rdu->blk = au_sbi(arg.sb)->si_rdblk;
23928 +               if (!rdu->blk)
23929 +                       rdu->blk = au_dir_size(file, /*dentry*/NULL);
23930 +       }
23931 +       bend = au_fbstart(file);
23932 +       if (cookie->bindex < bend)
23933 +               cookie->bindex = bend;
23934 +       bend = au_fbend_dir(file);
23935 +       /* AuDbg("b%d, b%d\n", cookie->bindex, bend); */
23936 +       for (; !err && cookie->bindex <= bend;
23937 +            cookie->bindex++, cookie->h_pos = 0) {
23938 +               h_file = au_hf_dir(file, cookie->bindex);
23939 +               if (!h_file)
23940 +                       continue;
23941 +
23942 +               au_fclr_rdu(cookie->flags, FULL);
23943 +               err = au_rdu_do(h_file, &arg);
23944 +               AuTraceErr(err);
23945 +               if (unlikely(au_ftest_rdu(cookie->flags, FULL) || err))
23946 +                       break;
23947 +       }
23948 +       AuDbg("rent %llu\n", rdu->rent);
23949 +
23950 +       if (!err && !au_ftest_rdu(cookie->flags, CONT)) {
23951 +               rdu->shwh = !!au_opt_test(au_sbi(arg.sb)->si_mntflags, SHWH);
23952 +               au_fset_rdu(cookie->flags, CONT);
23953 +               cookie->generation = au_figen(file);
23954 +       }
23955 +
23956 +       ii_read_lock_child(inode);
23957 +       fsstack_copy_attr_atime(inode, au_h_iptr(inode, au_ibstart(inode)));
23958 +       ii_read_unlock(inode);
23959 +
23960 +out_unlock:
23961 +       fi_read_unlock(file);
23962 +out_si:
23963 +       si_read_unlock(arg.sb);
23964 +out_mtx:
23965 +       mutex_unlock(&inode->i_mutex);
23966 +out:
23967 +       AuTraceErr(err);
23968 +       return err;
23969 +}
23970 +
23971 +static int au_rdu_ino(struct file *file, struct aufs_rdu *rdu)
23972 +{
23973 +       int err;
23974 +       ino_t ino;
23975 +       unsigned long long nent;
23976 +       union au_rdu_ent_ul *u;
23977 +       struct au_rdu_ent ent;
23978 +       struct super_block *sb;
23979 +
23980 +       err = 0;
23981 +       nent = rdu->nent;
23982 +       u = &rdu->ent;
23983 +       sb = file->f_dentry->d_sb;
23984 +       si_read_lock(sb, AuLock_FLUSH);
23985 +       while (nent-- > 0) {
23986 +               /* unnecessary to support mmap_sem since this is a dir */
23987 +               err = copy_from_user(&ent, u->e, sizeof(ent));
23988 +               if (!err)
23989 +                       err = !access_ok(VERIFY_WRITE, &u->e->ino, sizeof(ino));
23990 +               if (unlikely(err)) {
23991 +                       err = -EFAULT;
23992 +                       AuTraceErr(err);
23993 +                       break;
23994 +               }
23995 +
23996 +               /* AuDbg("b%d, i%llu\n", ent.bindex, ent.ino); */
23997 +               if (!ent.wh)
23998 +                       err = au_ino(sb, ent.bindex, ent.ino, ent.type, &ino);
23999 +               else
24000 +                       err = au_wh_ino(sb, ent.bindex, ent.ino, ent.type,
24001 +                                       &ino);
24002 +               if (unlikely(err)) {
24003 +                       AuTraceErr(err);
24004 +                       break;
24005 +               }
24006 +
24007 +               err = __put_user(ino, &u->e->ino);
24008 +               if (unlikely(err)) {
24009 +                       err = -EFAULT;
24010 +                       AuTraceErr(err);
24011 +                       break;
24012 +               }
24013 +               u->ul += au_rdu_len(ent.nlen);
24014 +       }
24015 +       si_read_unlock(sb);
24016 +
24017 +       return err;
24018 +}
24019 +
24020 +/* ---------------------------------------------------------------------- */
24021 +
24022 +static int au_rdu_verify(struct aufs_rdu *rdu)
24023 +{
24024 +       AuDbg("rdu{%llu, %p, %u | %u | %llu, %u, %u | "
24025 +             "%llu, b%d, 0x%x, g%u}\n",
24026 +             rdu->sz, rdu->ent.e, rdu->verify[AufsCtlRduV_SZ],
24027 +             rdu->blk,
24028 +             rdu->rent, rdu->shwh, rdu->full,
24029 +             rdu->cookie.h_pos, rdu->cookie.bindex, rdu->cookie.flags,
24030 +             rdu->cookie.generation);
24031 +
24032 +       if (rdu->verify[AufsCtlRduV_SZ] == sizeof(*rdu))
24033 +               return 0;
24034 +
24035 +       AuDbg("%u:%u\n",
24036 +             rdu->verify[AufsCtlRduV_SZ], (unsigned int)sizeof(*rdu));
24037 +       return -EINVAL;
24038 +}
24039 +
24040 +long au_rdu_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
24041 +{
24042 +       long err, e;
24043 +       struct aufs_rdu rdu;
24044 +       void __user *p = (void __user *)arg;
24045 +
24046 +       err = copy_from_user(&rdu, p, sizeof(rdu));
24047 +       if (unlikely(err)) {
24048 +               err = -EFAULT;
24049 +               AuTraceErr(err);
24050 +               goto out;
24051 +       }
24052 +       err = au_rdu_verify(&rdu);
24053 +       if (unlikely(err))
24054 +               goto out;
24055 +
24056 +       switch (cmd) {
24057 +       case AUFS_CTL_RDU:
24058 +               err = au_rdu(file, &rdu);
24059 +               if (unlikely(err))
24060 +                       break;
24061 +
24062 +               e = copy_to_user(p, &rdu, sizeof(rdu));
24063 +               if (unlikely(e)) {
24064 +                       err = -EFAULT;
24065 +                       AuTraceErr(err);
24066 +               }
24067 +               break;
24068 +       case AUFS_CTL_RDU_INO:
24069 +               err = au_rdu_ino(file, &rdu);
24070 +               break;
24071 +
24072 +       default:
24073 +               /* err = -ENOTTY; */
24074 +               err = -EINVAL;
24075 +       }
24076 +
24077 +out:
24078 +       AuTraceErr(err);
24079 +       return err;
24080 +}
24081 +
24082 +#ifdef CONFIG_COMPAT
24083 +long au_rdu_compat_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
24084 +{
24085 +       long err, e;
24086 +       struct aufs_rdu rdu;
24087 +       void __user *p = compat_ptr(arg);
24088 +
24089 +       /* todo: get_user()? */
24090 +       err = copy_from_user(&rdu, p, sizeof(rdu));
24091 +       if (unlikely(err)) {
24092 +               err = -EFAULT;
24093 +               AuTraceErr(err);
24094 +               goto out;
24095 +       }
24096 +       rdu.ent.e = compat_ptr(rdu.ent.ul);
24097 +       err = au_rdu_verify(&rdu);
24098 +       if (unlikely(err))
24099 +               goto out;
24100 +
24101 +       switch (cmd) {
24102 +       case AUFS_CTL_RDU:
24103 +               err = au_rdu(file, &rdu);
24104 +               if (unlikely(err))
24105 +                       break;
24106 +
24107 +               rdu.ent.ul = ptr_to_compat(rdu.ent.e);
24108 +               rdu.tail.ul = ptr_to_compat(rdu.tail.e);
24109 +               e = copy_to_user(p, &rdu, sizeof(rdu));
24110 +               if (unlikely(e)) {
24111 +                       err = -EFAULT;
24112 +                       AuTraceErr(err);
24113 +               }
24114 +               break;
24115 +       case AUFS_CTL_RDU_INO:
24116 +               err = au_rdu_ino(file, &rdu);
24117 +               break;
24118 +
24119 +       default:
24120 +               /* err = -ENOTTY; */
24121 +               err = -EINVAL;
24122 +       }
24123 +
24124 +out:
24125 +       AuTraceErr(err);
24126 +       return err;
24127 +}
24128 +#endif
24129 diff -urN /usr/share/empty/fs/aufs/rwsem.h linux/fs/aufs/rwsem.h
24130 --- /usr/share/empty/fs/aufs/rwsem.h    1970-01-01 01:00:00.000000000 +0100
24131 +++ linux/fs/aufs/rwsem.h       2014-04-24 22:11:10.858602483 +0200
24132 @@ -0,0 +1,187 @@
24133 +/*
24134 + * Copyright (C) 2005-2014 Junjiro R. Okajima
24135 + *
24136 + * This program, aufs is free software; you can redistribute it and/or modify
24137 + * it under the terms of the GNU General Public License as published by
24138 + * the Free Software Foundation; either version 2 of the License, or
24139 + * (at your option) any later version.
24140 + *
24141 + * This program is distributed in the hope that it will be useful,
24142 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24143 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24144 + * GNU General Public License for more details.
24145 + *
24146 + * You should have received a copy of the GNU General Public License
24147 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24148 + */
24149 +
24150 +/*
24151 + * simple read-write semaphore wrappers
24152 + */
24153 +
24154 +#ifndef __AUFS_RWSEM_H__
24155 +#define __AUFS_RWSEM_H__
24156 +
24157 +#ifdef __KERNEL__
24158 +
24159 +#include "debug.h"
24160 +
24161 +struct au_rwsem {
24162 +       struct rw_semaphore     rwsem;
24163 +#ifdef CONFIG_AUFS_DEBUG
24164 +       /* just for debugging, not almighty counter */
24165 +       atomic_t                rcnt, wcnt;
24166 +#endif
24167 +};
24168 +
24169 +#ifdef CONFIG_AUFS_DEBUG
24170 +#define AuDbgCntInit(rw) do { \
24171 +       atomic_set(&(rw)->rcnt, 0); \
24172 +       atomic_set(&(rw)->wcnt, 0); \
24173 +       smp_mb(); /* atomic set */ \
24174 +} while (0)
24175 +
24176 +#define AuDbgRcntInc(rw)       atomic_inc(&(rw)->rcnt)
24177 +#define AuDbgRcntDec(rw)       WARN_ON(atomic_dec_return(&(rw)->rcnt) < 0)
24178 +#define AuDbgWcntInc(rw)       atomic_inc(&(rw)->wcnt)
24179 +#define AuDbgWcntDec(rw)       WARN_ON(atomic_dec_return(&(rw)->wcnt) < 0)
24180 +#else
24181 +#define AuDbgCntInit(rw)       do {} while (0)
24182 +#define AuDbgRcntInc(rw)       do {} while (0)
24183 +#define AuDbgRcntDec(rw)       do {} while (0)
24184 +#define AuDbgWcntInc(rw)       do {} while (0)
24185 +#define AuDbgWcntDec(rw)       do {} while (0)
24186 +#endif /* CONFIG_AUFS_DEBUG */
24187 +
24188 +/* to debug easier, do not make them inlined functions */
24189 +#define AuRwMustNoWaiters(rw)  AuDebugOn(!list_empty(&(rw)->rwsem.wait_list))
24190 +/* rwsem_is_locked() is unusable */
24191 +#define AuRwMustReadLock(rw)   AuDebugOn(atomic_read(&(rw)->rcnt) <= 0)
24192 +#define AuRwMustWriteLock(rw)  AuDebugOn(atomic_read(&(rw)->wcnt) <= 0)
24193 +#define AuRwMustAnyLock(rw)    AuDebugOn(atomic_read(&(rw)->rcnt) <= 0 \
24194 +                                       && atomic_read(&(rw)->wcnt) <= 0)
24195 +#define AuRwDestroy(rw)                AuDebugOn(atomic_read(&(rw)->rcnt) \
24196 +                                       || atomic_read(&(rw)->wcnt))
24197 +
24198 +#define au_rw_class(rw, key)   lockdep_set_class(&(rw)->rwsem, key)
24199 +
24200 +static inline void au_rw_init(struct au_rwsem *rw)
24201 +{
24202 +       AuDbgCntInit(rw);
24203 +       init_rwsem(&rw->rwsem);
24204 +}
24205 +
24206 +static inline void au_rw_init_wlock(struct au_rwsem *rw)
24207 +{
24208 +       au_rw_init(rw);
24209 +       down_write(&rw->rwsem);
24210 +       AuDbgWcntInc(rw);
24211 +}
24212 +
24213 +static inline void au_rw_init_wlock_nested(struct au_rwsem *rw,
24214 +                                          unsigned int lsc)
24215 +{
24216 +       au_rw_init(rw);
24217 +       down_write_nested(&rw->rwsem, lsc);
24218 +       AuDbgWcntInc(rw);
24219 +}
24220 +
24221 +static inline void au_rw_read_lock(struct au_rwsem *rw)
24222 +{
24223 +       down_read(&rw->rwsem);
24224 +       AuDbgRcntInc(rw);
24225 +}
24226 +
24227 +static inline void au_rw_read_lock_nested(struct au_rwsem *rw, unsigned int lsc)
24228 +{
24229 +       down_read_nested(&rw->rwsem, lsc);
24230 +       AuDbgRcntInc(rw);
24231 +}
24232 +
24233 +static inline void au_rw_read_unlock(struct au_rwsem *rw)
24234 +{
24235 +       AuRwMustReadLock(rw);
24236 +       AuDbgRcntDec(rw);
24237 +       up_read(&rw->rwsem);
24238 +}
24239 +
24240 +static inline void au_rw_dgrade_lock(struct au_rwsem *rw)
24241 +{
24242 +       AuRwMustWriteLock(rw);
24243 +       AuDbgRcntInc(rw);
24244 +       AuDbgWcntDec(rw);
24245 +       downgrade_write(&rw->rwsem);
24246 +}
24247 +
24248 +static inline void au_rw_write_lock(struct au_rwsem *rw)
24249 +{
24250 +       down_write(&rw->rwsem);
24251 +       AuDbgWcntInc(rw);
24252 +}
24253 +
24254 +static inline void au_rw_write_lock_nested(struct au_rwsem *rw,
24255 +                                          unsigned int lsc)
24256 +{
24257 +       down_write_nested(&rw->rwsem, lsc);
24258 +       AuDbgWcntInc(rw);
24259 +}
24260 +
24261 +static inline void au_rw_write_unlock(struct au_rwsem *rw)
24262 +{
24263 +       AuRwMustWriteLock(rw);
24264 +       AuDbgWcntDec(rw);
24265 +       up_write(&rw->rwsem);
24266 +}
24267 +
24268 +/* why is not _nested version defined */
24269 +static inline int au_rw_read_trylock(struct au_rwsem *rw)
24270 +{
24271 +       int ret = down_read_trylock(&rw->rwsem);
24272 +       if (ret)
24273 +               AuDbgRcntInc(rw);
24274 +       return ret;
24275 +}
24276 +
24277 +static inline int au_rw_write_trylock(struct au_rwsem *rw)
24278 +{
24279 +       int ret = down_write_trylock(&rw->rwsem);
24280 +       if (ret)
24281 +               AuDbgWcntInc(rw);
24282 +       return ret;
24283 +}
24284 +
24285 +#undef AuDbgCntInit
24286 +#undef AuDbgRcntInc
24287 +#undef AuDbgRcntDec
24288 +#undef AuDbgWcntInc
24289 +#undef AuDbgWcntDec
24290 +
24291 +#define AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
24292 +static inline void prefix##_read_lock(param) \
24293 +{ au_rw_read_lock(rwsem); } \
24294 +static inline void prefix##_write_lock(param) \
24295 +{ au_rw_write_lock(rwsem); } \
24296 +static inline int prefix##_read_trylock(param) \
24297 +{ return au_rw_read_trylock(rwsem); } \
24298 +static inline int prefix##_write_trylock(param) \
24299 +{ return au_rw_write_trylock(rwsem); }
24300 +/* why is not _nested version defined */
24301 +/* static inline void prefix##_read_trylock_nested(param, lsc)
24302 +{ au_rw_read_trylock_nested(rwsem, lsc)); }
24303 +static inline void prefix##_write_trylock_nestd(param, lsc)
24304 +{ au_rw_write_trylock_nested(rwsem, lsc); } */
24305 +
24306 +#define AuSimpleUnlockRwsemFuncs(prefix, param, rwsem) \
24307 +static inline void prefix##_read_unlock(param) \
24308 +{ au_rw_read_unlock(rwsem); } \
24309 +static inline void prefix##_write_unlock(param) \
24310 +{ au_rw_write_unlock(rwsem); } \
24311 +static inline void prefix##_downgrade_lock(param) \
24312 +{ au_rw_dgrade_lock(rwsem); }
24313 +
24314 +#define AuSimpleRwsemFuncs(prefix, param, rwsem) \
24315 +       AuSimpleLockRwsemFuncs(prefix, param, rwsem) \
24316 +       AuSimpleUnlockRwsemFuncs(prefix, param, rwsem)
24317 +
24318 +#endif /* __KERNEL__ */
24319 +#endif /* __AUFS_RWSEM_H__ */
24320 diff -urN /usr/share/empty/fs/aufs/sbinfo.c linux/fs/aufs/sbinfo.c
24321 --- /usr/share/empty/fs/aufs/sbinfo.c   1970-01-01 01:00:00.000000000 +0100
24322 +++ linux/fs/aufs/sbinfo.c      2014-04-24 22:11:10.858602483 +0200
24323 @@ -0,0 +1,351 @@
24324 +/*
24325 + * Copyright (C) 2005-2014 Junjiro R. Okajima
24326 + *
24327 + * This program, aufs is free software; you can redistribute it and/or modify
24328 + * it under the terms of the GNU General Public License as published by
24329 + * the Free Software Foundation; either version 2 of the License, or
24330 + * (at your option) any later version.
24331 + *
24332 + * This program is distributed in the hope that it will be useful,
24333 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24334 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24335 + * GNU General Public License for more details.
24336 + *
24337 + * You should have received a copy of the GNU General Public License
24338 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24339 + */
24340 +
24341 +/*
24342 + * superblock private data
24343 + */
24344 +
24345 +#include "aufs.h"
24346 +
24347 +/*
24348 + * they are necessary regardless sysfs is disabled.
24349 + */
24350 +void au_si_free(struct kobject *kobj)
24351 +{
24352 +       int i;
24353 +       struct au_sbinfo *sbinfo;
24354 +       char *locked __maybe_unused; /* debug only */
24355 +
24356 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
24357 +       for (i = 0; i < AuPlink_NHASH; i++)
24358 +               AuDebugOn(!hlist_empty(&sbinfo->si_plink[i].head));
24359 +       AuDebugOn(atomic_read(&sbinfo->si_nowait.nw_len));
24360 +
24361 +       au_rw_write_lock(&sbinfo->si_rwsem);
24362 +       au_br_free(sbinfo);
24363 +       au_rw_write_unlock(&sbinfo->si_rwsem);
24364 +
24365 +       AuDebugOn(radix_tree_gang_lookup
24366 +                 (&sbinfo->au_si_pid.tree, (void **)&locked,
24367 +                  /*first_index*/PID_MAX_DEFAULT - 1,
24368 +                  /*max_items*/sizeof(locked)/sizeof(*locked)));
24369 +
24370 +       kfree(sbinfo->si_branch);
24371 +       kfree(sbinfo->au_si_pid.bitmap);
24372 +       mutex_destroy(&sbinfo->si_xib_mtx);
24373 +       AuRwDestroy(&sbinfo->si_rwsem);
24374 +
24375 +       kfree(sbinfo);
24376 +}
24377 +
24378 +int au_si_alloc(struct super_block *sb)
24379 +{
24380 +       int err, i;
24381 +       struct au_sbinfo *sbinfo;
24382 +       static struct lock_class_key aufs_si;
24383 +
24384 +       err = -ENOMEM;
24385 +       sbinfo = kzalloc(sizeof(*sbinfo), GFP_NOFS);
24386 +       if (unlikely(!sbinfo))
24387 +               goto out;
24388 +
24389 +       BUILD_BUG_ON(sizeof(unsigned long) !=
24390 +                    sizeof(*sbinfo->au_si_pid.bitmap));
24391 +       sbinfo->au_si_pid.bitmap = kcalloc(BITS_TO_LONGS(PID_MAX_DEFAULT),
24392 +                                       sizeof(*sbinfo->au_si_pid.bitmap),
24393 +                                       GFP_NOFS);
24394 +       if (unlikely(!sbinfo->au_si_pid.bitmap))
24395 +               goto out_sbinfo;
24396 +
24397 +       /* will be reallocated separately */
24398 +       sbinfo->si_branch = kzalloc(sizeof(*sbinfo->si_branch), GFP_NOFS);
24399 +       if (unlikely(!sbinfo->si_branch))
24400 +               goto out_pidmap;
24401 +
24402 +       err = sysaufs_si_init(sbinfo);
24403 +       if (unlikely(err))
24404 +               goto out_br;
24405 +
24406 +       au_nwt_init(&sbinfo->si_nowait);
24407 +       au_rw_init_wlock(&sbinfo->si_rwsem);
24408 +       au_rw_class(&sbinfo->si_rwsem, &aufs_si);
24409 +       spin_lock_init(&sbinfo->au_si_pid.tree_lock);
24410 +       INIT_RADIX_TREE(&sbinfo->au_si_pid.tree, GFP_ATOMIC | __GFP_NOFAIL);
24411 +
24412 +       atomic_long_set(&sbinfo->si_ninodes, 0);
24413 +       atomic_long_set(&sbinfo->si_nfiles, 0);
24414 +
24415 +       sbinfo->si_bend = -1;
24416 +       sbinfo->si_last_br_id = AUFS_BRANCH_MAX / 2;
24417 +
24418 +       sbinfo->si_wbr_copyup = AuWbrCopyup_Def;
24419 +       sbinfo->si_wbr_create = AuWbrCreate_Def;
24420 +       sbinfo->si_wbr_copyup_ops = au_wbr_copyup_ops + sbinfo->si_wbr_copyup;
24421 +       sbinfo->si_wbr_create_ops = au_wbr_create_ops + sbinfo->si_wbr_create;
24422 +
24423 +       sbinfo->si_mntflags = au_opts_plink(AuOpt_Def);
24424 +
24425 +       sbinfo->si_xino_jiffy = jiffies;
24426 +       sbinfo->si_xino_expire
24427 +               = msecs_to_jiffies(AUFS_XINO_DEF_SEC * MSEC_PER_SEC);
24428 +       mutex_init(&sbinfo->si_xib_mtx);
24429 +       sbinfo->si_xino_brid = -1;
24430 +       /* leave si_xib_last_pindex and si_xib_next_bit */
24431 +
24432 +       sbinfo->si_rdcache = msecs_to_jiffies(AUFS_RDCACHE_DEF * MSEC_PER_SEC);
24433 +       sbinfo->si_rdblk = AUFS_RDBLK_DEF;
24434 +       sbinfo->si_rdhash = AUFS_RDHASH_DEF;
24435 +       sbinfo->si_dirwh = AUFS_DIRWH_DEF;
24436 +
24437 +       for (i = 0; i < AuPlink_NHASH; i++)
24438 +               au_sphl_init(sbinfo->si_plink + i);
24439 +       init_waitqueue_head(&sbinfo->si_plink_wq);
24440 +       spin_lock_init(&sbinfo->si_plink_maint_lock);
24441 +
24442 +       au_sphl_init(&sbinfo->si_files);
24443 +
24444 +       /* leave other members for sysaufs and si_mnt. */
24445 +       sbinfo->si_sb = sb;
24446 +       sb->s_fs_info = sbinfo;
24447 +       si_pid_set(sb);
24448 +       au_debug_sbinfo_init(sbinfo);
24449 +       return 0; /* success */
24450 +
24451 +out_br:
24452 +       kfree(sbinfo->si_branch);
24453 +out_pidmap:
24454 +       kfree(sbinfo->au_si_pid.bitmap);
24455 +out_sbinfo:
24456 +       kfree(sbinfo);
24457 +out:
24458 +       return err;
24459 +}
24460 +
24461 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr)
24462 +{
24463 +       int err, sz;
24464 +       struct au_branch **brp;
24465 +
24466 +       AuRwMustWriteLock(&sbinfo->si_rwsem);
24467 +
24468 +       err = -ENOMEM;
24469 +       sz = sizeof(*brp) * (sbinfo->si_bend + 1);
24470 +       if (unlikely(!sz))
24471 +               sz = sizeof(*brp);
24472 +       brp = au_kzrealloc(sbinfo->si_branch, sz, sizeof(*brp) * nbr, GFP_NOFS);
24473 +       if (brp) {
24474 +               sbinfo->si_branch = brp;
24475 +               err = 0;
24476 +       }
24477 +
24478 +       return err;
24479 +}
24480 +
24481 +/* ---------------------------------------------------------------------- */
24482 +
24483 +unsigned int au_sigen_inc(struct super_block *sb)
24484 +{
24485 +       unsigned int gen;
24486 +
24487 +       SiMustWriteLock(sb);
24488 +
24489 +       gen = ++au_sbi(sb)->si_generation;
24490 +       au_update_digen(sb->s_root);
24491 +       au_update_iigen(sb->s_root->d_inode, /*half*/0);
24492 +       sb->s_root->d_inode->i_version++;
24493 +       return gen;
24494 +}
24495 +
24496 +aufs_bindex_t au_new_br_id(struct super_block *sb)
24497 +{
24498 +       aufs_bindex_t br_id;
24499 +       int i;
24500 +       struct au_sbinfo *sbinfo;
24501 +
24502 +       SiMustWriteLock(sb);
24503 +
24504 +       sbinfo = au_sbi(sb);
24505 +       for (i = 0; i <= AUFS_BRANCH_MAX; i++) {
24506 +               br_id = ++sbinfo->si_last_br_id;
24507 +               AuDebugOn(br_id < 0);
24508 +               if (br_id && au_br_index(sb, br_id) < 0)
24509 +                       return br_id;
24510 +       }
24511 +
24512 +       return -1;
24513 +}
24514 +
24515 +/* ---------------------------------------------------------------------- */
24516 +
24517 +/* it is ok that new 'nwt' tasks are appended while we are sleeping */
24518 +int si_read_lock(struct super_block *sb, int flags)
24519 +{
24520 +       int err;
24521 +
24522 +       err = 0;
24523 +       if (au_ftest_lock(flags, FLUSH))
24524 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
24525 +
24526 +       si_noflush_read_lock(sb);
24527 +       err = au_plink_maint(sb, flags);
24528 +       if (unlikely(err))
24529 +               si_read_unlock(sb);
24530 +
24531 +       return err;
24532 +}
24533 +
24534 +int si_write_lock(struct super_block *sb, int flags)
24535 +{
24536 +       int err;
24537 +
24538 +       if (au_ftest_lock(flags, FLUSH))
24539 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
24540 +
24541 +       si_noflush_write_lock(sb);
24542 +       err = au_plink_maint(sb, flags);
24543 +       if (unlikely(err))
24544 +               si_write_unlock(sb);
24545 +
24546 +       return err;
24547 +}
24548 +
24549 +/* dentry and super_block lock. call at entry point */
24550 +int aufs_read_lock(struct dentry *dentry, int flags)
24551 +{
24552 +       int err;
24553 +       struct super_block *sb;
24554 +
24555 +       sb = dentry->d_sb;
24556 +       err = si_read_lock(sb, flags);
24557 +       if (unlikely(err))
24558 +               goto out;
24559 +
24560 +       if (au_ftest_lock(flags, DW))
24561 +               di_write_lock_child(dentry);
24562 +       else
24563 +               di_read_lock_child(dentry, flags);
24564 +
24565 +       if (au_ftest_lock(flags, GEN)) {
24566 +               err = au_digen_test(dentry, au_sigen(sb));
24567 +               AuDebugOn(!err && au_dbrange_test(dentry));
24568 +               if (unlikely(err))
24569 +                       aufs_read_unlock(dentry, flags);
24570 +       }
24571 +
24572 +out:
24573 +       return err;
24574 +}
24575 +
24576 +void aufs_read_unlock(struct dentry *dentry, int flags)
24577 +{
24578 +       if (au_ftest_lock(flags, DW))
24579 +               di_write_unlock(dentry);
24580 +       else
24581 +               di_read_unlock(dentry, flags);
24582 +       si_read_unlock(dentry->d_sb);
24583 +}
24584 +
24585 +void aufs_write_lock(struct dentry *dentry)
24586 +{
24587 +       si_write_lock(dentry->d_sb, AuLock_FLUSH | AuLock_NOPLMW);
24588 +       di_write_lock_child(dentry);
24589 +}
24590 +
24591 +void aufs_write_unlock(struct dentry *dentry)
24592 +{
24593 +       di_write_unlock(dentry);
24594 +       si_write_unlock(dentry->d_sb);
24595 +}
24596 +
24597 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags)
24598 +{
24599 +       int err;
24600 +       unsigned int sigen;
24601 +       struct super_block *sb;
24602 +
24603 +       sb = d1->d_sb;
24604 +       err = si_read_lock(sb, flags);
24605 +       if (unlikely(err))
24606 +               goto out;
24607 +
24608 +       di_write_lock2_child(d1, d2, au_ftest_lock(flags, DIR));
24609 +
24610 +       if (au_ftest_lock(flags, GEN)) {
24611 +               sigen = au_sigen(sb);
24612 +               err = au_digen_test(d1, sigen);
24613 +               AuDebugOn(!err && au_dbrange_test(d1));
24614 +               if (!err) {
24615 +                       err = au_digen_test(d2, sigen);
24616 +                       AuDebugOn(!err && au_dbrange_test(d2));
24617 +               }
24618 +               if (unlikely(err))
24619 +                       aufs_read_and_write_unlock2(d1, d2);
24620 +       }
24621 +
24622 +out:
24623 +       return err;
24624 +}
24625 +
24626 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2)
24627 +{
24628 +       di_write_unlock2(d1, d2);
24629 +       si_read_unlock(d1->d_sb);
24630 +}
24631 +
24632 +/* ---------------------------------------------------------------------- */
24633 +
24634 +int si_pid_test_slow(struct super_block *sb)
24635 +{
24636 +       void *p;
24637 +
24638 +       rcu_read_lock();
24639 +       p = radix_tree_lookup(&au_sbi(sb)->au_si_pid.tree, current->pid);
24640 +       rcu_read_unlock();
24641 +
24642 +       return (long)!!p;
24643 +}
24644 +
24645 +void si_pid_set_slow(struct super_block *sb)
24646 +{
24647 +       int err;
24648 +       struct au_sbinfo *sbinfo;
24649 +
24650 +       AuDebugOn(si_pid_test_slow(sb));
24651 +
24652 +       sbinfo = au_sbi(sb);
24653 +       err = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
24654 +       AuDebugOn(err);
24655 +       spin_lock(&sbinfo->au_si_pid.tree_lock);
24656 +       err = radix_tree_insert(&sbinfo->au_si_pid.tree, current->pid,
24657 +                               /*any valid ptr*/sb);
24658 +       spin_unlock(&sbinfo->au_si_pid.tree_lock);
24659 +       AuDebugOn(err);
24660 +       radix_tree_preload_end();
24661 +}
24662 +
24663 +void si_pid_clr_slow(struct super_block *sb)
24664 +{
24665 +       void *p;
24666 +       struct au_sbinfo *sbinfo;
24667 +
24668 +       AuDebugOn(!si_pid_test_slow(sb));
24669 +
24670 +       sbinfo = au_sbi(sb);
24671 +       spin_lock(&sbinfo->au_si_pid.tree_lock);
24672 +       p = radix_tree_delete(&sbinfo->au_si_pid.tree, current->pid);
24673 +       spin_unlock(&sbinfo->au_si_pid.tree_lock);
24674 +}
24675 diff -urN /usr/share/empty/fs/aufs/spl.h linux/fs/aufs/spl.h
24676 --- /usr/share/empty/fs/aufs/spl.h      1970-01-01 01:00:00.000000000 +0100
24677 +++ linux/fs/aufs/spl.h 2014-04-24 22:11:10.858602483 +0200
24678 @@ -0,0 +1,111 @@
24679 +/*
24680 + * Copyright (C) 2005-2014 Junjiro R. Okajima
24681 + *
24682 + * This program, aufs is free software; you can redistribute it and/or modify
24683 + * it under the terms of the GNU General Public License as published by
24684 + * the Free Software Foundation; either version 2 of the License, or
24685 + * (at your option) any later version.
24686 + *
24687 + * This program is distributed in the hope that it will be useful,
24688 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24689 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24690 + * GNU General Public License for more details.
24691 + *
24692 + * You should have received a copy of the GNU General Public License
24693 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24694 + */
24695 +
24696 +/*
24697 + * simple list protected by a spinlock
24698 + */
24699 +
24700 +#ifndef __AUFS_SPL_H__
24701 +#define __AUFS_SPL_H__
24702 +
24703 +#ifdef __KERNEL__
24704 +
24705 +struct au_splhead {
24706 +       spinlock_t              spin;
24707 +       struct list_head        head;
24708 +};
24709 +
24710 +static inline void au_spl_init(struct au_splhead *spl)
24711 +{
24712 +       spin_lock_init(&spl->spin);
24713 +       INIT_LIST_HEAD(&spl->head);
24714 +}
24715 +
24716 +static inline void au_spl_add(struct list_head *list, struct au_splhead *spl)
24717 +{
24718 +       spin_lock(&spl->spin);
24719 +       list_add(list, &spl->head);
24720 +       spin_unlock(&spl->spin);
24721 +}
24722 +
24723 +static inline void au_spl_del(struct list_head *list, struct au_splhead *spl)
24724 +{
24725 +       spin_lock(&spl->spin);
24726 +       list_del(list);
24727 +       spin_unlock(&spl->spin);
24728 +}
24729 +
24730 +static inline void au_spl_del_rcu(struct list_head *list,
24731 +                                 struct au_splhead *spl)
24732 +{
24733 +       spin_lock(&spl->spin);
24734 +       list_del_rcu(list);
24735 +       spin_unlock(&spl->spin);
24736 +}
24737 +
24738 +/* ---------------------------------------------------------------------- */
24739 +
24740 +struct au_sphlhead {
24741 +       spinlock_t              spin;
24742 +       struct hlist_head       head;
24743 +};
24744 +
24745 +static inline void au_sphl_init(struct au_sphlhead *sphl)
24746 +{
24747 +       spin_lock_init(&sphl->spin);
24748 +       INIT_HLIST_HEAD(&sphl->head);
24749 +}
24750 +
24751 +static inline void au_sphl_add(struct hlist_node *hlist,
24752 +                              struct au_sphlhead *sphl)
24753 +{
24754 +       spin_lock(&sphl->spin);
24755 +       hlist_add_head(hlist, &sphl->head);
24756 +       spin_unlock(&sphl->spin);
24757 +}
24758 +
24759 +static inline void au_sphl_del(struct hlist_node *hlist,
24760 +                              struct au_sphlhead *sphl)
24761 +{
24762 +       spin_lock(&sphl->spin);
24763 +       hlist_del(hlist);
24764 +       spin_unlock(&sphl->spin);
24765 +}
24766 +
24767 +static inline void au_sphl_del_rcu(struct hlist_node *hlist,
24768 +                                  struct au_sphlhead *sphl)
24769 +{
24770 +       spin_lock(&sphl->spin);
24771 +       hlist_del_rcu(hlist);
24772 +       spin_unlock(&sphl->spin);
24773 +}
24774 +
24775 +static inline unsigned long au_sphl_count(struct au_sphlhead *sphl)
24776 +{
24777 +       unsigned long cnt;
24778 +       struct hlist_node *pos;
24779 +
24780 +       cnt = 0;
24781 +       spin_lock(&sphl->spin);
24782 +       hlist_for_each(pos, &sphl->head)
24783 +               cnt++;
24784 +       spin_unlock(&sphl->spin);
24785 +       return cnt;
24786 +}
24787 +
24788 +#endif /* __KERNEL__ */
24789 +#endif /* __AUFS_SPL_H__ */
24790 diff -urN /usr/share/empty/fs/aufs/super.c linux/fs/aufs/super.c
24791 --- /usr/share/empty/fs/aufs/super.c    1970-01-01 01:00:00.000000000 +0100
24792 +++ linux/fs/aufs/super.c       2014-04-24 22:11:10.858602483 +0200
24793 @@ -0,0 +1,1001 @@
24794 +/*
24795 + * Copyright (C) 2005-2014 Junjiro R. Okajima
24796 + *
24797 + * This program, aufs is free software; you can redistribute it and/or modify
24798 + * it under the terms of the GNU General Public License as published by
24799 + * the Free Software Foundation; either version 2 of the License, or
24800 + * (at your option) any later version.
24801 + *
24802 + * This program is distributed in the hope that it will be useful,
24803 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
24804 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
24805 + * GNU General Public License for more details.
24806 + *
24807 + * You should have received a copy of the GNU General Public License
24808 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
24809 + */
24810 +
24811 +/*
24812 + * mount and super_block operations
24813 + */
24814 +
24815 +#include <linux/mm.h>
24816 +#include <linux/module.h>
24817 +#include <linux/seq_file.h>
24818 +#include <linux/statfs.h>
24819 +#include <linux/vmalloc.h>
24820 +#include <linux/writeback.h>
24821 +#include "aufs.h"
24822 +
24823 +/*
24824 + * super_operations
24825 + */
24826 +static struct inode *aufs_alloc_inode(struct super_block *sb __maybe_unused)
24827 +{
24828 +       struct au_icntnr *c;
24829 +
24830 +       c = au_cache_alloc_icntnr();
24831 +       if (c) {
24832 +               au_icntnr_init(c);
24833 +               c->vfs_inode.i_version = 1; /* sigen(sb); */
24834 +               c->iinfo.ii_hinode = NULL;
24835 +               return &c->vfs_inode;
24836 +       }
24837 +       return NULL;
24838 +}
24839 +
24840 +static void aufs_destroy_inode_cb(struct rcu_head *head)
24841 +{
24842 +       struct inode *inode = container_of(head, struct inode, i_rcu);
24843 +
24844 +       INIT_HLIST_HEAD(&inode->i_dentry);
24845 +       au_cache_free_icntnr(container_of(inode, struct au_icntnr, vfs_inode));
24846 +}
24847 +
24848 +static void aufs_destroy_inode(struct inode *inode)
24849 +{
24850 +       au_iinfo_fin(inode);
24851 +       call_rcu(&inode->i_rcu, aufs_destroy_inode_cb);
24852 +}
24853 +
24854 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino)
24855 +{
24856 +       struct inode *inode;
24857 +       int err;
24858 +
24859 +       inode = iget_locked(sb, ino);
24860 +       if (unlikely(!inode)) {
24861 +               inode = ERR_PTR(-ENOMEM);
24862 +               goto out;
24863 +       }
24864 +       if (!(inode->i_state & I_NEW))
24865 +               goto out;
24866 +
24867 +       err = au_xigen_new(inode);
24868 +       if (!err)
24869 +               err = au_iinfo_init(inode);
24870 +       if (!err)
24871 +               inode->i_version++;
24872 +       else {
24873 +               iget_failed(inode);
24874 +               inode = ERR_PTR(err);
24875 +       }
24876 +
24877 +out:
24878 +       /* never return NULL */
24879 +       AuDebugOn(!inode);
24880 +       AuTraceErrPtr(inode);
24881 +       return inode;
24882 +}
24883 +
24884 +/* lock free root dinfo */
24885 +static int au_show_brs(struct seq_file *seq, struct super_block *sb)
24886 +{
24887 +       int err;
24888 +       aufs_bindex_t bindex, bend;
24889 +       struct path path;
24890 +       struct au_hdentry *hdp;
24891 +       struct au_branch *br;
24892 +       char *perm;
24893 +
24894 +       err = 0;
24895 +       bend = au_sbend(sb);
24896 +       hdp = au_di(sb->s_root)->di_hdentry;
24897 +       for (bindex = 0; !err && bindex <= bend; bindex++) {
24898 +               br = au_sbr(sb, bindex);
24899 +               path.mnt = au_br_mnt(br);
24900 +               path.dentry = hdp[bindex].hd_dentry;
24901 +               err = au_seq_path(seq, &path);
24902 +               if (err > 0) {
24903 +                       perm = au_optstr_br_perm(br->br_perm);
24904 +                       if (perm) {
24905 +                               err = seq_printf(seq, "=%s", perm);
24906 +                               kfree(perm);
24907 +                               if (err == -1)
24908 +                                       err = -E2BIG;
24909 +                       } else
24910 +                               err = -ENOMEM;
24911 +               }
24912 +               if (!err && bindex != bend)
24913 +                       err = seq_putc(seq, ':');
24914 +       }
24915 +
24916 +       return err;
24917 +}
24918 +
24919 +static void au_show_wbr_create(struct seq_file *m, int v,
24920 +                              struct au_sbinfo *sbinfo)
24921 +{
24922 +       const char *pat;
24923 +
24924 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
24925 +
24926 +       seq_puts(m, ",create=");
24927 +       pat = au_optstr_wbr_create(v);
24928 +       switch (v) {
24929 +       case AuWbrCreate_TDP:
24930 +       case AuWbrCreate_RR:
24931 +       case AuWbrCreate_MFS:
24932 +       case AuWbrCreate_PMFS:
24933 +               seq_puts(m, pat);
24934 +               break;
24935 +       case AuWbrCreate_MFSV:
24936 +               seq_printf(m, /*pat*/"mfs:%lu",
24937 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
24938 +                          / MSEC_PER_SEC);
24939 +               break;
24940 +       case AuWbrCreate_PMFSV:
24941 +               seq_printf(m, /*pat*/"pmfs:%lu",
24942 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
24943 +                          / MSEC_PER_SEC);
24944 +               break;
24945 +       case AuWbrCreate_MFSRR:
24946 +               seq_printf(m, /*pat*/"mfsrr:%llu",
24947 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
24948 +               break;
24949 +       case AuWbrCreate_MFSRRV:
24950 +               seq_printf(m, /*pat*/"mfsrr:%llu:%lu",
24951 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
24952 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
24953 +                          / MSEC_PER_SEC);
24954 +               break;
24955 +       case AuWbrCreate_PMFSRR:
24956 +               seq_printf(m, /*pat*/"pmfsrr:%llu",
24957 +                          sbinfo->si_wbr_mfs.mfsrr_watermark);
24958 +               break;
24959 +       case AuWbrCreate_PMFSRRV:
24960 +               seq_printf(m, /*pat*/"pmfsrr:%llu:%lu",
24961 +                          sbinfo->si_wbr_mfs.mfsrr_watermark,
24962 +                          jiffies_to_msecs(sbinfo->si_wbr_mfs.mfs_expire)
24963 +                          / MSEC_PER_SEC);
24964 +               break;
24965 +       }
24966 +}
24967 +
24968 +static int au_show_xino(struct seq_file *seq, struct super_block *sb)
24969 +{
24970 +#ifdef CONFIG_SYSFS
24971 +       return 0;
24972 +#else
24973 +       int err;
24974 +       const int len = sizeof(AUFS_XINO_FNAME) - 1;
24975 +       aufs_bindex_t bindex, brid;
24976 +       struct qstr *name;
24977 +       struct file *f;
24978 +       struct dentry *d, *h_root;
24979 +       struct au_hdentry *hdp;
24980 +
24981 +       AuRwMustAnyLock(&sbinfo->si_rwsem);
24982 +
24983 +       err = 0;
24984 +       f = au_sbi(sb)->si_xib;
24985 +       if (!f)
24986 +               goto out;
24987 +
24988 +       /* stop printing the default xino path on the first writable branch */
24989 +       h_root = NULL;
24990 +       brid = au_xino_brid(sb);
24991 +       if (brid >= 0) {
24992 +               bindex = au_br_index(sb, brid);
24993 +               hdp = au_di(sb->s_root)->di_hdentry;
24994 +               h_root = hdp[0 + bindex].hd_dentry;
24995 +       }
24996 +       d = f->f_dentry;
24997 +       name = &d->d_name;
24998 +       /* safe ->d_parent because the file is unlinked */
24999 +       if (d->d_parent == h_root
25000 +           && name->len == len
25001 +           && !memcmp(name->name, AUFS_XINO_FNAME, len))
25002 +               goto out;
25003 +
25004 +       seq_puts(seq, ",xino=");
25005 +       err = au_xino_path(seq, f);
25006 +
25007 +out:
25008 +       return err;
25009 +#endif
25010 +}
25011 +
25012 +/* seq_file will re-call me in case of too long string */
25013 +static int aufs_show_options(struct seq_file *m, struct dentry *dentry)
25014 +{
25015 +       int err;
25016 +       unsigned int mnt_flags, v;
25017 +       struct super_block *sb;
25018 +       struct au_sbinfo *sbinfo;
25019 +
25020 +#define AuBool(name, str) do { \
25021 +       v = au_opt_test(mnt_flags, name); \
25022 +       if (v != au_opt_test(AuOpt_Def, name)) \
25023 +               seq_printf(m, ",%s" #str, v ? "" : "no"); \
25024 +} while (0)
25025 +
25026 +#define AuStr(name, str) do { \
25027 +       v = mnt_flags & AuOptMask_##name; \
25028 +       if (v != (AuOpt_Def & AuOptMask_##name)) \
25029 +               seq_printf(m, "," #str "=%s", au_optstr_##str(v)); \
25030 +} while (0)
25031 +
25032 +#define AuUInt(name, str, val) do { \
25033 +       if (val != AUFS_##name##_DEF) \
25034 +               seq_printf(m, "," #str "=%u", val); \
25035 +} while (0)
25036 +
25037 +       /* lock free root dinfo */
25038 +       sb = dentry->d_sb;
25039 +       si_noflush_read_lock(sb);
25040 +       sbinfo = au_sbi(sb);
25041 +       seq_printf(m, ",si=%lx", sysaufs_si_id(sbinfo));
25042 +
25043 +       mnt_flags = au_mntflags(sb);
25044 +       if (au_opt_test(mnt_flags, XINO)) {
25045 +               err = au_show_xino(m, sb);
25046 +               if (unlikely(err))
25047 +                       goto out;
25048 +       } else
25049 +               seq_puts(m, ",noxino");
25050 +
25051 +       AuBool(TRUNC_XINO, trunc_xino);
25052 +       AuStr(UDBA, udba);
25053 +       AuBool(SHWH, shwh);
25054 +       AuBool(PLINK, plink);
25055 +       AuBool(DIO, dio);
25056 +       /* AuBool(DIRPERM1, dirperm1); */
25057 +       /* AuBool(REFROF, refrof); */
25058 +
25059 +       v = sbinfo->si_wbr_create;
25060 +       if (v != AuWbrCreate_Def)
25061 +               au_show_wbr_create(m, v, sbinfo);
25062 +
25063 +       v = sbinfo->si_wbr_copyup;
25064 +       if (v != AuWbrCopyup_Def)
25065 +               seq_printf(m, ",cpup=%s", au_optstr_wbr_copyup(v));
25066 +
25067 +       v = au_opt_test(mnt_flags, ALWAYS_DIROPQ);
25068 +       if (v != au_opt_test(AuOpt_Def, ALWAYS_DIROPQ))
25069 +               seq_printf(m, ",diropq=%c", v ? 'a' : 'w');
25070 +
25071 +       AuUInt(DIRWH, dirwh, sbinfo->si_dirwh);
25072 +
25073 +       v = jiffies_to_msecs(sbinfo->si_rdcache) / MSEC_PER_SEC;
25074 +       AuUInt(RDCACHE, rdcache, v);
25075 +
25076 +       AuUInt(RDBLK, rdblk, sbinfo->si_rdblk);
25077 +       AuUInt(RDHASH, rdhash, sbinfo->si_rdhash);
25078 +
25079 +       AuBool(SUM, sum);
25080 +       /* AuBool(SUM_W, wsum); */
25081 +       AuBool(WARN_PERM, warn_perm);
25082 +       AuBool(VERBOSE, verbose);
25083 +
25084 +out:
25085 +       /* be sure to print "br:" last */
25086 +       if (!sysaufs_brs) {
25087 +               seq_puts(m, ",br:");
25088 +               au_show_brs(m, sb);
25089 +       }
25090 +       si_read_unlock(sb);
25091 +       return 0;
25092 +
25093 +#undef AuBool
25094 +#undef AuStr
25095 +#undef AuUInt
25096 +}
25097 +
25098 +/* ---------------------------------------------------------------------- */
25099 +
25100 +/* sum mode which returns the summation for statfs(2) */
25101 +
25102 +static u64 au_add_till_max(u64 a, u64 b)
25103 +{
25104 +       u64 old;
25105 +
25106 +       old = a;
25107 +       a += b;
25108 +       if (old <= a)
25109 +               return a;
25110 +       return ULLONG_MAX;
25111 +}
25112 +
25113 +static u64 au_mul_till_max(u64 a, long mul)
25114 +{
25115 +       u64 old;
25116 +
25117 +       old = a;
25118 +       a *= mul;
25119 +       if (old <= a)
25120 +               return a;
25121 +       return ULLONG_MAX;
25122 +}
25123 +
25124 +static int au_statfs_sum(struct super_block *sb, struct kstatfs *buf)
25125 +{
25126 +       int err;
25127 +       long bsize, factor;
25128 +       u64 blocks, bfree, bavail, files, ffree;
25129 +       aufs_bindex_t bend, bindex, i;
25130 +       unsigned char shared;
25131 +       struct path h_path;
25132 +       struct super_block *h_sb;
25133 +
25134 +       err = 0;
25135 +       bsize = LONG_MAX;
25136 +       files = 0;
25137 +       ffree = 0;
25138 +       blocks = 0;
25139 +       bfree = 0;
25140 +       bavail = 0;
25141 +       bend = au_sbend(sb);
25142 +       for (bindex = 0; bindex <= bend; bindex++) {
25143 +               h_path.mnt = au_sbr_mnt(sb, bindex);
25144 +               h_sb = h_path.mnt->mnt_sb;
25145 +               shared = 0;
25146 +               for (i = 0; !shared && i < bindex; i++)
25147 +                       shared = (au_sbr_sb(sb, i) == h_sb);
25148 +               if (shared)
25149 +                       continue;
25150 +
25151 +               /* sb->s_root for NFS is unreliable */
25152 +               h_path.dentry = h_path.mnt->mnt_root;
25153 +               err = vfs_statfs(&h_path, buf);
25154 +               if (unlikely(err))
25155 +                       goto out;
25156 +
25157 +               if (bsize > buf->f_bsize) {
25158 +                       /*
25159 +                        * we will reduce bsize, so we have to expand blocks
25160 +                        * etc. to match them again
25161 +                        */
25162 +                       factor = (bsize / buf->f_bsize);
25163 +                       blocks = au_mul_till_max(blocks, factor);
25164 +                       bfree = au_mul_till_max(bfree, factor);
25165 +                       bavail = au_mul_till_max(bavail, factor);
25166 +                       bsize = buf->f_bsize;
25167 +               }
25168 +
25169 +               factor = (buf->f_bsize / bsize);
25170 +               blocks = au_add_till_max(blocks,
25171 +                               au_mul_till_max(buf->f_blocks, factor));
25172 +               bfree = au_add_till_max(bfree,
25173 +                               au_mul_till_max(buf->f_bfree, factor));
25174 +               bavail = au_add_till_max(bavail,
25175 +                               au_mul_till_max(buf->f_bavail, factor));
25176 +               files = au_add_till_max(files, buf->f_files);
25177 +               ffree = au_add_till_max(ffree, buf->f_ffree);
25178 +       }
25179 +
25180 +       buf->f_bsize = bsize;
25181 +       buf->f_blocks = blocks;
25182 +       buf->f_bfree = bfree;
25183 +       buf->f_bavail = bavail;
25184 +       buf->f_files = files;
25185 +       buf->f_ffree = ffree;
25186 +       buf->f_frsize = 0;
25187 +
25188 +out:
25189 +       return err;
25190 +}
25191 +
25192 +static int aufs_statfs(struct dentry *dentry, struct kstatfs *buf)
25193 +{
25194 +       int err;
25195 +       struct path h_path;
25196 +       struct super_block *sb;
25197 +
25198 +       /* lock free root dinfo */
25199 +       sb = dentry->d_sb;
25200 +       si_noflush_read_lock(sb);
25201 +       if (!au_opt_test(au_mntflags(sb), SUM)) {
25202 +               /* sb->s_root for NFS is unreliable */
25203 +               h_path.mnt = au_sbr_mnt(sb, 0);
25204 +               h_path.dentry = h_path.mnt->mnt_root;
25205 +               err = vfs_statfs(&h_path, buf);
25206 +       } else
25207 +               err = au_statfs_sum(sb, buf);
25208 +       si_read_unlock(sb);
25209 +
25210 +       if (!err) {
25211 +               buf->f_type = AUFS_SUPER_MAGIC;
25212 +               buf->f_namelen = AUFS_MAX_NAMELEN;
25213 +               memset(&buf->f_fsid, 0, sizeof(buf->f_fsid));
25214 +       }
25215 +       /* buf->f_bsize = buf->f_blocks = buf->f_bfree = buf->f_bavail = -1; */
25216 +
25217 +       return err;
25218 +}
25219 +
25220 +/* ---------------------------------------------------------------------- */
25221 +
25222 +static int aufs_sync_fs(struct super_block *sb, int wait)
25223 +{
25224 +       int err, e;
25225 +       aufs_bindex_t bend, bindex;
25226 +       struct au_branch *br;
25227 +       struct super_block *h_sb;
25228 +
25229 +       err = 0;
25230 +       si_noflush_read_lock(sb);
25231 +       bend = au_sbend(sb);
25232 +       for (bindex = 0; bindex <= bend; bindex++) {
25233 +               br = au_sbr(sb, bindex);
25234 +               if (!au_br_writable(br->br_perm))
25235 +                       continue;
25236 +
25237 +               h_sb = au_sbr_sb(sb, bindex);
25238 +               if (h_sb->s_op->sync_fs) {
25239 +                       e = h_sb->s_op->sync_fs(h_sb, wait);
25240 +                       if (unlikely(e && !err))
25241 +                               err = e;
25242 +                       /* go on even if an error happens */
25243 +               }
25244 +       }
25245 +       si_read_unlock(sb);
25246 +
25247 +       return err;
25248 +}
25249 +
25250 +/* ---------------------------------------------------------------------- */
25251 +
25252 +/* final actions when unmounting a file system */
25253 +static void aufs_put_super(struct super_block *sb)
25254 +{
25255 +       struct au_sbinfo *sbinfo;
25256 +
25257 +       sbinfo = au_sbi(sb);
25258 +       if (!sbinfo)
25259 +               return;
25260 +
25261 +       dbgaufs_si_fin(sbinfo);
25262 +       kobject_put(&sbinfo->si_kobj);
25263 +}
25264 +
25265 +/* ---------------------------------------------------------------------- */
25266 +
25267 +void au_array_free(void *array)
25268 +{
25269 +       if (array) {
25270 +               if (!is_vmalloc_addr(array))
25271 +                       kfree(array);
25272 +               else
25273 +                       vfree(array);
25274 +       }
25275 +}
25276 +
25277 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb, void *arg)
25278 +{
25279 +       void *array;
25280 +       unsigned long long n;
25281 +
25282 +       array = NULL;
25283 +       n = 0;
25284 +       if (!*hint)
25285 +               goto out;
25286 +
25287 +       if (*hint > ULLONG_MAX / sizeof(array)) {
25288 +               array = ERR_PTR(-EMFILE);
25289 +               pr_err("hint %llu\n", *hint);
25290 +               goto out;
25291 +       }
25292 +
25293 +       array = kmalloc(sizeof(array) * *hint, GFP_NOFS);
25294 +       if (unlikely(!array))
25295 +               array = vmalloc(sizeof(array) * *hint);
25296 +       if (unlikely(!array)) {
25297 +               array = ERR_PTR(-ENOMEM);
25298 +               goto out;
25299 +       }
25300 +
25301 +       n = cb(array, *hint, arg);
25302 +       AuDebugOn(n > *hint);
25303 +
25304 +out:
25305 +       *hint = n;
25306 +       return array;
25307 +}
25308 +
25309 +static unsigned long long au_iarray_cb(void *a,
25310 +                                      unsigned long long max __maybe_unused,
25311 +                                      void *arg)
25312 +{
25313 +       unsigned long long n;
25314 +       struct inode **p, *inode;
25315 +       struct list_head *head;
25316 +
25317 +       n = 0;
25318 +       p = a;
25319 +       head = arg;
25320 +       spin_lock(&inode_sb_list_lock);
25321 +       list_for_each_entry(inode, head, i_sb_list) {
25322 +               if (!is_bad_inode(inode)
25323 +                   && au_ii(inode)->ii_bstart >= 0) {
25324 +                       spin_lock(&inode->i_lock);
25325 +                       if (atomic_read(&inode->i_count)) {
25326 +                               au_igrab(inode);
25327 +                               *p++ = inode;
25328 +                               n++;
25329 +                               AuDebugOn(n > max);
25330 +                       }
25331 +                       spin_unlock(&inode->i_lock);
25332 +               }
25333 +       }
25334 +       spin_unlock(&inode_sb_list_lock);
25335 +
25336 +       return n;
25337 +}
25338 +
25339 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max)
25340 +{
25341 +       *max = atomic_long_read(&au_sbi(sb)->si_ninodes);
25342 +       return au_array_alloc(max, au_iarray_cb, &sb->s_inodes);
25343 +}
25344 +
25345 +void au_iarray_free(struct inode **a, unsigned long long max)
25346 +{
25347 +       unsigned long long ull;
25348 +
25349 +       for (ull = 0; ull < max; ull++)
25350 +               iput(a[ull]);
25351 +       au_array_free(a);
25352 +}
25353 +
25354 +/* ---------------------------------------------------------------------- */
25355 +
25356 +/*
25357 + * refresh dentry and inode at remount time.
25358 + */
25359 +/* todo: consolidate with simple_reval_dpath() and au_reval_for_attr() */
25360 +static int au_do_refresh(struct dentry *dentry, unsigned int dir_flags,
25361 +                     struct dentry *parent)
25362 +{
25363 +       int err;
25364 +
25365 +       di_write_lock_child(dentry);
25366 +       di_read_lock_parent(parent, AuLock_IR);
25367 +       err = au_refresh_dentry(dentry, parent);
25368 +       if (!err && dir_flags)
25369 +               au_hn_reset(dentry->d_inode, dir_flags);
25370 +       di_read_unlock(parent, AuLock_IR);
25371 +       di_write_unlock(dentry);
25372 +
25373 +       return err;
25374 +}
25375 +
25376 +static int au_do_refresh_d(struct dentry *dentry, unsigned int sigen,
25377 +                          struct au_sbinfo *sbinfo,
25378 +                          const unsigned int dir_flags)
25379 +{
25380 +       int err;
25381 +       struct dentry *parent;
25382 +       struct inode *inode;
25383 +
25384 +       err = 0;
25385 +       parent = dget_parent(dentry);
25386 +       if (!au_digen_test(parent, sigen) && au_digen_test(dentry, sigen)) {
25387 +               inode = dentry->d_inode;
25388 +               if (inode) {
25389 +                       if (!S_ISDIR(inode->i_mode))
25390 +                               err = au_do_refresh(dentry, /*dir_flags*/0,
25391 +                                                parent);
25392 +                       else {
25393 +                               err = au_do_refresh(dentry, dir_flags, parent);
25394 +                               if (unlikely(err))
25395 +                                       au_fset_si(sbinfo, FAILED_REFRESH_DIR);
25396 +                       }
25397 +               } else
25398 +                       err = au_do_refresh(dentry, /*dir_flags*/0, parent);
25399 +               AuDbgDentry(dentry);
25400 +       }
25401 +       dput(parent);
25402 +
25403 +       AuTraceErr(err);
25404 +       return err;
25405 +}
25406 +
25407 +static int au_refresh_d(struct super_block *sb)
25408 +{
25409 +       int err, i, j, ndentry, e;
25410 +       unsigned int sigen;
25411 +       struct au_dcsub_pages dpages;
25412 +       struct au_dpage *dpage;
25413 +       struct dentry **dentries, *d;
25414 +       struct au_sbinfo *sbinfo;
25415 +       struct dentry *root = sb->s_root;
25416 +       const unsigned int dir_flags = au_hi_flags(root->d_inode, /*isdir*/1);
25417 +
25418 +       err = au_dpages_init(&dpages, GFP_NOFS);
25419 +       if (unlikely(err))
25420 +               goto out;
25421 +       err = au_dcsub_pages(&dpages, root, NULL, NULL);
25422 +       if (unlikely(err))
25423 +               goto out_dpages;
25424 +
25425 +       sigen = au_sigen(sb);
25426 +       sbinfo = au_sbi(sb);
25427 +       for (i = 0; i < dpages.ndpage; i++) {
25428 +               dpage = dpages.dpages + i;
25429 +               dentries = dpage->dentries;
25430 +               ndentry = dpage->ndentry;
25431 +               for (j = 0; j < ndentry; j++) {
25432 +                       d = dentries[j];
25433 +                       e = au_do_refresh_d(d, sigen, sbinfo, dir_flags);
25434 +                       if (unlikely(e && !err))
25435 +                               err = e;
25436 +                       /* go on even err */
25437 +               }
25438 +       }
25439 +
25440 +out_dpages:
25441 +       au_dpages_free(&dpages);
25442 +out:
25443 +       return err;
25444 +}
25445 +
25446 +static int au_refresh_i(struct super_block *sb)
25447 +{
25448 +       int err, e;
25449 +       unsigned int sigen;
25450 +       unsigned long long max, ull;
25451 +       struct inode *inode, **array;
25452 +
25453 +       array = au_iarray_alloc(sb, &max);
25454 +       err = PTR_ERR(array);
25455 +       if (IS_ERR(array))
25456 +               goto out;
25457 +
25458 +       err = 0;
25459 +       sigen = au_sigen(sb);
25460 +       for (ull = 0; ull < max; ull++) {
25461 +               inode = array[ull];
25462 +               if (au_iigen(inode, NULL) != sigen) {
25463 +                       ii_write_lock_child(inode);
25464 +                       e = au_refresh_hinode_self(inode);
25465 +                       ii_write_unlock(inode);
25466 +                       if (unlikely(e)) {
25467 +                               pr_err("error %d, i%lu\n", e, inode->i_ino);
25468 +                               if (!err)
25469 +                                       err = e;
25470 +                               /* go on even if err */
25471 +                       }
25472 +               }
25473 +       }
25474 +
25475 +       au_iarray_free(array, max);
25476 +
25477 +out:
25478 +       return err;
25479 +}
25480 +
25481 +static void au_remount_refresh(struct super_block *sb)
25482 +{
25483 +       int err, e;
25484 +       unsigned int udba;
25485 +       aufs_bindex_t bindex, bend;
25486 +       struct dentry *root;
25487 +       struct inode *inode;
25488 +       struct au_branch *br;
25489 +
25490 +       au_sigen_inc(sb);
25491 +       au_fclr_si(au_sbi(sb), FAILED_REFRESH_DIR);
25492 +
25493 +       root = sb->s_root;
25494 +       DiMustNoWaiters(root);
25495 +       inode = root->d_inode;
25496 +       IiMustNoWaiters(inode);
25497 +
25498 +       udba = au_opt_udba(sb);
25499 +       bend = au_sbend(sb);
25500 +       for (bindex = 0; bindex <= bend; bindex++) {
25501 +               br = au_sbr(sb, bindex);
25502 +               err = au_hnotify_reset_br(udba, br, br->br_perm);
25503 +               if (unlikely(err))
25504 +                       AuIOErr("hnotify failed on br %d, %d, ignored\n",
25505 +                               bindex, err);
25506 +               /* go on even if err */
25507 +       }
25508 +       au_hn_reset(inode, au_hi_flags(inode, /*isdir*/1));
25509 +
25510 +       di_write_unlock(root);
25511 +       err = au_refresh_d(sb);
25512 +       e = au_refresh_i(sb);
25513 +       if (unlikely(e && !err))
25514 +               err = e;
25515 +       /* aufs_write_lock() calls ..._child() */
25516 +       di_write_lock_child(root);
25517 +
25518 +       au_cpup_attr_all(inode, /*force*/1);
25519 +
25520 +       if (unlikely(err))
25521 +               AuIOErr("refresh failed, ignored, %d\n", err);
25522 +}
25523 +
25524 +/* stop extra interpretation of errno in mount(8), and strange error messages */
25525 +static int cvt_err(int err)
25526 +{
25527 +       AuTraceErr(err);
25528 +
25529 +       switch (err) {
25530 +       case -ENOENT:
25531 +       case -ENOTDIR:
25532 +       case -EEXIST:
25533 +       case -EIO:
25534 +               err = -EINVAL;
25535 +       }
25536 +       return err;
25537 +}
25538 +
25539 +static int aufs_remount_fs(struct super_block *sb, int *flags, char *data)
25540 +{
25541 +       int err, do_dx;
25542 +       unsigned int mntflags;
25543 +       struct au_opts opts;
25544 +       struct dentry *root;
25545 +       struct inode *inode;
25546 +       struct au_sbinfo *sbinfo;
25547 +
25548 +       err = 0;
25549 +       root = sb->s_root;
25550 +       if (!data || !*data) {
25551 +               err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
25552 +               if (!err) {
25553 +                       di_write_lock_child(root);
25554 +                       err = au_opts_verify(sb, *flags, /*pending*/0);
25555 +                       aufs_write_unlock(root);
25556 +               }
25557 +               goto out;
25558 +       }
25559 +
25560 +       err = -ENOMEM;
25561 +       memset(&opts, 0, sizeof(opts));
25562 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
25563 +       if (unlikely(!opts.opt))
25564 +               goto out;
25565 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
25566 +       opts.flags = AuOpts_REMOUNT;
25567 +       opts.sb_flags = *flags;
25568 +
25569 +       /* parse it before aufs lock */
25570 +       err = au_opts_parse(sb, data, &opts);
25571 +       if (unlikely(err))
25572 +               goto out_opts;
25573 +
25574 +       sbinfo = au_sbi(sb);
25575 +       inode = root->d_inode;
25576 +       mutex_lock(&inode->i_mutex);
25577 +       err = si_write_lock(sb, AuLock_FLUSH | AuLock_NOPLM);
25578 +       if (unlikely(err))
25579 +               goto out_mtx;
25580 +       di_write_lock_child(root);
25581 +
25582 +       /* au_opts_remount() may return an error */
25583 +       err = au_opts_remount(sb, &opts);
25584 +       au_opts_free(&opts);
25585 +
25586 +       if (au_ftest_opts(opts.flags, REFRESH))
25587 +               au_remount_refresh(sb);
25588 +
25589 +       if (au_ftest_opts(opts.flags, REFRESH_DYAOP)) {
25590 +               mntflags = au_mntflags(sb);
25591 +               do_dx = !!au_opt_test(mntflags, DIO);
25592 +               au_dy_arefresh(do_dx);
25593 +       }
25594 +
25595 +       aufs_write_unlock(root);
25596 +
25597 +out_mtx:
25598 +       mutex_unlock(&inode->i_mutex);
25599 +out_opts:
25600 +       free_page((unsigned long)opts.opt);
25601 +out:
25602 +       err = cvt_err(err);
25603 +       AuTraceErr(err);
25604 +       return err;
25605 +}
25606 +
25607 +static const struct super_operations aufs_sop = {
25608 +       .alloc_inode    = aufs_alloc_inode,
25609 +       .destroy_inode  = aufs_destroy_inode,
25610 +       /* always deleting, no clearing */
25611 +       .drop_inode     = generic_delete_inode,
25612 +       .show_options   = aufs_show_options,
25613 +       .statfs         = aufs_statfs,
25614 +       .put_super      = aufs_put_super,
25615 +       .sync_fs        = aufs_sync_fs,
25616 +       .remount_fs     = aufs_remount_fs
25617 +};
25618 +
25619 +/* ---------------------------------------------------------------------- */
25620 +
25621 +static int alloc_root(struct super_block *sb)
25622 +{
25623 +       int err;
25624 +       struct inode *inode;
25625 +       struct dentry *root;
25626 +
25627 +       err = -ENOMEM;
25628 +       inode = au_iget_locked(sb, AUFS_ROOT_INO);
25629 +       err = PTR_ERR(inode);
25630 +       if (IS_ERR(inode))
25631 +               goto out;
25632 +
25633 +       inode->i_op = &aufs_dir_iop;
25634 +       inode->i_fop = &aufs_dir_fop;
25635 +       inode->i_mode = S_IFDIR;
25636 +       set_nlink(inode, 2);
25637 +       unlock_new_inode(inode);
25638 +
25639 +       root = d_make_root(inode);
25640 +       if (unlikely(!root))
25641 +               goto out;
25642 +       err = PTR_ERR(root);
25643 +       if (IS_ERR(root))
25644 +               goto out;
25645 +
25646 +       err = au_di_init(root);
25647 +       if (!err) {
25648 +               sb->s_root = root;
25649 +               return 0; /* success */
25650 +       }
25651 +       dput(root);
25652 +
25653 +out:
25654 +       return err;
25655 +}
25656 +
25657 +static int aufs_fill_super(struct super_block *sb, void *raw_data,
25658 +                          int silent __maybe_unused)
25659 +{
25660 +       int err;
25661 +       struct au_opts opts;
25662 +       struct dentry *root;
25663 +       struct inode *inode;
25664 +       char *arg = raw_data;
25665 +
25666 +       if (unlikely(!arg || !*arg)) {
25667 +               err = -EINVAL;
25668 +               pr_err("no arg\n");
25669 +               goto out;
25670 +       }
25671 +
25672 +       err = -ENOMEM;
25673 +       memset(&opts, 0, sizeof(opts));
25674 +       opts.opt = (void *)__get_free_page(GFP_NOFS);
25675 +       if (unlikely(!opts.opt))
25676 +               goto out;
25677 +       opts.max_opt = PAGE_SIZE / sizeof(*opts.opt);
25678 +       opts.sb_flags = sb->s_flags;
25679 +
25680 +       err = au_si_alloc(sb);
25681 +       if (unlikely(err))
25682 +               goto out_opts;
25683 +
25684 +       /* all timestamps always follow the ones on the branch */
25685 +       sb->s_flags |= MS_NOATIME | MS_NODIRATIME;
25686 +       sb->s_op = &aufs_sop;
25687 +       sb->s_d_op = &aufs_dop;
25688 +       sb->s_magic = AUFS_SUPER_MAGIC;
25689 +       sb->s_maxbytes = 0;
25690 +       au_export_init(sb);
25691 +
25692 +       err = alloc_root(sb);
25693 +       if (unlikely(err)) {
25694 +               si_write_unlock(sb);
25695 +               goto out_info;
25696 +       }
25697 +       root = sb->s_root;
25698 +       inode = root->d_inode;
25699 +
25700 +       /*
25701 +        * actually we can parse options regardless aufs lock here.
25702 +        * but at remount time, parsing must be done before aufs lock.
25703 +        * so we follow the same rule.
25704 +        */
25705 +       ii_write_lock_parent(inode);
25706 +       aufs_write_unlock(root);
25707 +       err = au_opts_parse(sb, arg, &opts);
25708 +       if (unlikely(err))
25709 +               goto out_root;
25710 +
25711 +       /* lock vfs_inode first, then aufs. */
25712 +       mutex_lock(&inode->i_mutex);
25713 +       aufs_write_lock(root);
25714 +       err = au_opts_mount(sb, &opts);
25715 +       au_opts_free(&opts);
25716 +       aufs_write_unlock(root);
25717 +       mutex_unlock(&inode->i_mutex);
25718 +       if (!err)
25719 +               goto out_opts; /* success */
25720 +
25721 +out_root:
25722 +       dput(root);
25723 +       sb->s_root = NULL;
25724 +out_info:
25725 +       dbgaufs_si_fin(au_sbi(sb));
25726 +       kobject_put(&au_sbi(sb)->si_kobj);
25727 +       sb->s_fs_info = NULL;
25728 +out_opts:
25729 +       free_page((unsigned long)opts.opt);
25730 +out:
25731 +       AuTraceErr(err);
25732 +       err = cvt_err(err);
25733 +       AuTraceErr(err);
25734 +       return err;
25735 +}
25736 +
25737 +/* ---------------------------------------------------------------------- */
25738 +
25739 +static struct dentry *aufs_mount(struct file_system_type *fs_type, int flags,
25740 +                                const char *dev_name __maybe_unused,
25741 +                                void *raw_data)
25742 +{
25743 +       struct dentry *root;
25744 +       struct super_block *sb;
25745 +
25746 +       /* all timestamps always follow the ones on the branch */
25747 +       /* mnt->mnt_flags |= MNT_NOATIME | MNT_NODIRATIME; */
25748 +       root = mount_nodev(fs_type, flags, raw_data, aufs_fill_super);
25749 +       if (IS_ERR(root))
25750 +               goto out;
25751 +
25752 +       sb = root->d_sb;
25753 +       si_write_lock(sb, !AuLock_FLUSH);
25754 +       sysaufs_brs_add(sb, 0);
25755 +       si_write_unlock(sb);
25756 +       au_sbilist_add(sb);
25757 +
25758 +out:
25759 +       return root;
25760 +}
25761 +
25762 +static void aufs_kill_sb(struct super_block *sb)
25763 +{
25764 +       struct au_sbinfo *sbinfo;
25765 +
25766 +       sbinfo = au_sbi(sb);
25767 +       if (sbinfo) {
25768 +               au_sbilist_del(sb);
25769 +               aufs_write_lock(sb->s_root);
25770 +               if (sbinfo->si_wbr_create_ops->fin)
25771 +                       sbinfo->si_wbr_create_ops->fin(sb);
25772 +               if (au_opt_test(sbinfo->si_mntflags, UDBA_HNOTIFY)) {
25773 +                       au_opt_set_udba(sbinfo->si_mntflags, UDBA_NONE);
25774 +                       au_remount_refresh(sb);
25775 +               }
25776 +               if (au_opt_test(sbinfo->si_mntflags, PLINK))
25777 +                       au_plink_put(sb, /*verbose*/1);
25778 +               au_xino_clr(sb);
25779 +               sbinfo->si_sb = NULL;
25780 +               aufs_write_unlock(sb->s_root);
25781 +               au_nwt_flush(&sbinfo->si_nowait);
25782 +       }
25783 +       kill_anon_super(sb);
25784 +}
25785 +
25786 +struct file_system_type aufs_fs_type = {
25787 +       .name           = AUFS_FSTYPE,
25788 +       /* a race between rename and others */
25789 +       .fs_flags       = FS_RENAME_DOES_D_MOVE,
25790 +       .mount          = aufs_mount,
25791 +       .kill_sb        = aufs_kill_sb,
25792 +       /* no need to __module_get() and module_put(). */
25793 +       .owner          = THIS_MODULE,
25794 +};
25795 diff -urN /usr/share/empty/fs/aufs/super.h linux/fs/aufs/super.h
25796 --- /usr/share/empty/fs/aufs/super.h    1970-01-01 01:00:00.000000000 +0100
25797 +++ linux/fs/aufs/super.h       2014-04-24 22:11:10.858602483 +0200
25798 @@ -0,0 +1,571 @@
25799 +/*
25800 + * Copyright (C) 2005-2014 Junjiro R. Okajima
25801 + *
25802 + * This program, aufs is free software; you can redistribute it and/or modify
25803 + * it under the terms of the GNU General Public License as published by
25804 + * the Free Software Foundation; either version 2 of the License, or
25805 + * (at your option) any later version.
25806 + *
25807 + * This program is distributed in the hope that it will be useful,
25808 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
25809 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
25810 + * GNU General Public License for more details.
25811 + *
25812 + * You should have received a copy of the GNU General Public License
25813 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
25814 + */
25815 +
25816 +/*
25817 + * super_block operations
25818 + */
25819 +
25820 +#ifndef __AUFS_SUPER_H__
25821 +#define __AUFS_SUPER_H__
25822 +
25823 +#ifdef __KERNEL__
25824 +
25825 +#include <linux/fs.h>
25826 +#include "rwsem.h"
25827 +#include "spl.h"
25828 +#include "wkq.h"
25829 +
25830 +typedef ssize_t (*au_readf_t)(struct file *, char __user *, size_t, loff_t *);
25831 +typedef ssize_t (*au_writef_t)(struct file *, const char __user *, size_t,
25832 +                              loff_t *);
25833 +
25834 +/* policies to select one among multiple writable branches */
25835 +struct au_wbr_copyup_operations {
25836 +       int (*copyup)(struct dentry *dentry);
25837 +};
25838 +
25839 +#define AuWbr_DIR      1               /* target is a dir */
25840 +#define AuWbr_PARENT   (1 << 1)        /* always require a parent */
25841 +
25842 +#define au_ftest_wbr(flags, name)      ((flags) & AuWbr_##name)
25843 +#define au_fset_wbr(flags, name)       { (flags) |= AuWbr_##name; }
25844 +#define au_fclr_wbr(flags, name)       { (flags) &= ~AuWbr_##name; }
25845 +
25846 +struct au_wbr_create_operations {
25847 +       int (*create)(struct dentry *dentry, unsigned int flags);
25848 +       int (*init)(struct super_block *sb);
25849 +       int (*fin)(struct super_block *sb);
25850 +};
25851 +
25852 +struct au_wbr_mfs {
25853 +       struct mutex    mfs_lock; /* protect this structure */
25854 +       unsigned long   mfs_jiffy;
25855 +       unsigned long   mfs_expire;
25856 +       aufs_bindex_t   mfs_bindex;
25857 +
25858 +       unsigned long long      mfsrr_bytes;
25859 +       unsigned long long      mfsrr_watermark;
25860 +};
25861 +
25862 +struct pseudo_link {
25863 +       union {
25864 +               struct hlist_node hlist;
25865 +               struct rcu_head rcu;
25866 +       };
25867 +       struct inode *inode;
25868 +};
25869 +
25870 +#define AuPlink_NHASH 100
25871 +static inline int au_plink_hash(ino_t ino)
25872 +{
25873 +       return ino % AuPlink_NHASH;
25874 +}
25875 +
25876 +struct au_branch;
25877 +struct au_sbinfo {
25878 +       /* nowait tasks in the system-wide workqueue */
25879 +       struct au_nowait_tasks  si_nowait;
25880 +
25881 +       /*
25882 +        * tried sb->s_umount, but failed due to the dependecy between i_mutex.
25883 +        * rwsem for au_sbinfo is necessary.
25884 +        */
25885 +       struct au_rwsem         si_rwsem;
25886 +
25887 +       /* prevent recursive locking in deleting inode */
25888 +       struct {
25889 +               unsigned long           *bitmap;
25890 +               spinlock_t              tree_lock;
25891 +               struct radix_tree_root  tree;
25892 +       } au_si_pid;
25893 +
25894 +       /*
25895 +        * dirty approach to protect sb->sb_inodes and ->s_files (gone) from
25896 +        * remount.
25897 +        */
25898 +       atomic_long_t           si_ninodes, si_nfiles;
25899 +
25900 +       /* branch management */
25901 +       unsigned int            si_generation;
25902 +
25903 +       /* see above flags */
25904 +       unsigned char           au_si_status;
25905 +
25906 +       aufs_bindex_t           si_bend;
25907 +
25908 +       /* dirty trick to keep br_id plus */
25909 +       unsigned int            si_last_br_id :
25910 +                               sizeof(aufs_bindex_t) * BITS_PER_BYTE - 1;
25911 +       struct au_branch        **si_branch;
25912 +
25913 +       /* policy to select a writable branch */
25914 +       unsigned char           si_wbr_copyup;
25915 +       unsigned char           si_wbr_create;
25916 +       struct au_wbr_copyup_operations *si_wbr_copyup_ops;
25917 +       struct au_wbr_create_operations *si_wbr_create_ops;
25918 +
25919 +       /* round robin */
25920 +       atomic_t                si_wbr_rr_next;
25921 +
25922 +       /* most free space */
25923 +       struct au_wbr_mfs       si_wbr_mfs;
25924 +
25925 +       /* mount flags */
25926 +       /* include/asm-ia64/siginfo.h defines a macro named si_flags */
25927 +       unsigned int            si_mntflags;
25928 +
25929 +       /* external inode number (bitmap and translation table) */
25930 +       au_readf_t              si_xread;
25931 +       au_writef_t             si_xwrite;
25932 +       struct file             *si_xib;
25933 +       struct mutex            si_xib_mtx; /* protect xib members */
25934 +       unsigned long           *si_xib_buf;
25935 +       unsigned long           si_xib_last_pindex;
25936 +       int                     si_xib_next_bit;
25937 +       aufs_bindex_t           si_xino_brid;
25938 +       unsigned long           si_xino_jiffy;
25939 +       unsigned long           si_xino_expire;
25940 +       /* reserved for future use */
25941 +       /* unsigned long long   si_xib_limit; */        /* Max xib file size */
25942 +
25943 +#ifdef CONFIG_AUFS_EXPORT
25944 +       /* i_generation */
25945 +       struct file             *si_xigen;
25946 +       atomic_t                si_xigen_next;
25947 +#endif
25948 +
25949 +       /* vdir parameters */
25950 +       unsigned long           si_rdcache;     /* max cache time in jiffies */
25951 +       unsigned int            si_rdblk;       /* deblk size */
25952 +       unsigned int            si_rdhash;      /* hash size */
25953 +
25954 +       /*
25955 +        * If the number of whiteouts are larger than si_dirwh, leave all of
25956 +        * them after au_whtmp_ren to reduce the cost of rmdir(2).
25957 +        * future fsck.aufs or kernel thread will remove them later.
25958 +        * Otherwise, remove all whiteouts and the dir in rmdir(2).
25959 +        */
25960 +       unsigned int            si_dirwh;
25961 +
25962 +       /*
25963 +        * rename(2) a directory with all children.
25964 +        */
25965 +       /* reserved for future use */
25966 +       /* int                  si_rendir; */
25967 +
25968 +       /* pseudo_link list */
25969 +       struct au_sphlhead      si_plink[AuPlink_NHASH];
25970 +       wait_queue_head_t       si_plink_wq;
25971 +       spinlock_t              si_plink_maint_lock;
25972 +       pid_t                   si_plink_maint_pid;
25973 +
25974 +       /* file list */
25975 +       struct au_sphlhead      si_files;
25976 +
25977 +       /*
25978 +        * sysfs and lifetime management.
25979 +        * this is not a small structure and it may be a waste of memory in case
25980 +        * of sysfs is disabled, particulary when many aufs-es are mounted.
25981 +        * but using sysfs is majority.
25982 +        */
25983 +       struct kobject          si_kobj;
25984 +#ifdef CONFIG_DEBUG_FS
25985 +       struct dentry            *si_dbgaufs;
25986 +       struct dentry            *si_dbgaufs_plink;
25987 +       struct dentry            *si_dbgaufs_xib;
25988 +#ifdef CONFIG_AUFS_EXPORT
25989 +       struct dentry            *si_dbgaufs_xigen;
25990 +#endif
25991 +#endif
25992 +
25993 +#ifdef CONFIG_AUFS_SBILIST
25994 +       struct list_head        si_list;
25995 +#endif
25996 +
25997 +       /* dirty, necessary for unmounting, sysfs and sysrq */
25998 +       struct super_block      *si_sb;
25999 +};
26000 +
26001 +/* sbinfo status flags */
26002 +/*
26003 + * set true when refresh_dirs() failed at remount time.
26004 + * then try refreshing dirs at access time again.
26005 + * if it is false, refreshing dirs at access time is unnecesary
26006 + */
26007 +#define AuSi_FAILED_REFRESH_DIR        1
26008 +static inline unsigned char au_do_ftest_si(struct au_sbinfo *sbi,
26009 +                                          unsigned int flag)
26010 +{
26011 +       AuRwMustAnyLock(&sbi->si_rwsem);
26012 +       return sbi->au_si_status & flag;
26013 +}
26014 +#define au_ftest_si(sbinfo, name)      au_do_ftest_si(sbinfo, AuSi_##name)
26015 +#define au_fset_si(sbinfo, name) do { \
26016 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
26017 +       (sbinfo)->au_si_status |= AuSi_##name; \
26018 +} while (0)
26019 +#define au_fclr_si(sbinfo, name) do { \
26020 +       AuRwMustWriteLock(&(sbinfo)->si_rwsem); \
26021 +       (sbinfo)->au_si_status &= ~AuSi_##name; \
26022 +} while (0)
26023 +
26024 +/* ---------------------------------------------------------------------- */
26025 +
26026 +/* policy to select one among writable branches */
26027 +#define AuWbrCopyup(sbinfo, ...) \
26028 +       ((sbinfo)->si_wbr_copyup_ops->copyup(__VA_ARGS__))
26029 +#define AuWbrCreate(sbinfo, ...) \
26030 +       ((sbinfo)->si_wbr_create_ops->create(__VA_ARGS__))
26031 +
26032 +/* flags for si_read_lock()/aufs_read_lock()/di_read_lock() */
26033 +#define AuLock_DW              1               /* write-lock dentry */
26034 +#define AuLock_IR              (1 << 1)        /* read-lock inode */
26035 +#define AuLock_IW              (1 << 2)        /* write-lock inode */
26036 +#define AuLock_FLUSH           (1 << 3)        /* wait for 'nowait' tasks */
26037 +#define AuLock_DIR             (1 << 4)        /* target is a dir */
26038 +#define AuLock_NOPLM           (1 << 5)        /* return err in plm mode */
26039 +#define AuLock_NOPLMW          (1 << 6)        /* wait for plm mode ends */
26040 +#define AuLock_GEN             (1 << 7)        /* test digen/iigen */
26041 +#define au_ftest_lock(flags, name)     ((flags) & AuLock_##name)
26042 +#define au_fset_lock(flags, name) \
26043 +       do { (flags) |= AuLock_##name; } while (0)
26044 +#define au_fclr_lock(flags, name) \
26045 +       do { (flags) &= ~AuLock_##name; } while (0)
26046 +
26047 +/* ---------------------------------------------------------------------- */
26048 +
26049 +/* super.c */
26050 +extern struct file_system_type aufs_fs_type;
26051 +struct inode *au_iget_locked(struct super_block *sb, ino_t ino);
26052 +typedef unsigned long long (*au_arraycb_t)(void *array, unsigned long long max,
26053 +                                          void *arg);
26054 +void au_array_free(void *array);
26055 +void *au_array_alloc(unsigned long long *hint, au_arraycb_t cb, void *arg);
26056 +struct inode **au_iarray_alloc(struct super_block *sb, unsigned long long *max);
26057 +void au_iarray_free(struct inode **a, unsigned long long max);
26058 +
26059 +/* sbinfo.c */
26060 +void au_si_free(struct kobject *kobj);
26061 +int au_si_alloc(struct super_block *sb);
26062 +int au_sbr_realloc(struct au_sbinfo *sbinfo, int nbr);
26063 +
26064 +unsigned int au_sigen_inc(struct super_block *sb);
26065 +aufs_bindex_t au_new_br_id(struct super_block *sb);
26066 +
26067 +int si_read_lock(struct super_block *sb, int flags);
26068 +int si_write_lock(struct super_block *sb, int flags);
26069 +int aufs_read_lock(struct dentry *dentry, int flags);
26070 +void aufs_read_unlock(struct dentry *dentry, int flags);
26071 +void aufs_write_lock(struct dentry *dentry);
26072 +void aufs_write_unlock(struct dentry *dentry);
26073 +int aufs_read_and_write_lock2(struct dentry *d1, struct dentry *d2, int flags);
26074 +void aufs_read_and_write_unlock2(struct dentry *d1, struct dentry *d2);
26075 +
26076 +int si_pid_test_slow(struct super_block *sb);
26077 +void si_pid_set_slow(struct super_block *sb);
26078 +void si_pid_clr_slow(struct super_block *sb);
26079 +
26080 +/* wbr_policy.c */
26081 +extern struct au_wbr_copyup_operations au_wbr_copyup_ops[];
26082 +extern struct au_wbr_create_operations au_wbr_create_ops[];
26083 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst);
26084 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex);
26085 +
26086 +/* mvdown.c */
26087 +int au_mvdown(struct dentry *dentry, struct aufs_mvdown __user *arg);
26088 +
26089 +/* ---------------------------------------------------------------------- */
26090 +
26091 +static inline struct au_sbinfo *au_sbi(struct super_block *sb)
26092 +{
26093 +       return sb->s_fs_info;
26094 +}
26095 +
26096 +/* ---------------------------------------------------------------------- */
26097 +
26098 +#ifdef CONFIG_AUFS_EXPORT
26099 +int au_test_nfsd(void);
26100 +void au_export_init(struct super_block *sb);
26101 +void au_xigen_inc(struct inode *inode);
26102 +int au_xigen_new(struct inode *inode);
26103 +int au_xigen_set(struct super_block *sb, struct file *base);
26104 +void au_xigen_clr(struct super_block *sb);
26105 +
26106 +static inline int au_busy_or_stale(void)
26107 +{
26108 +       if (!au_test_nfsd())
26109 +               return -EBUSY;
26110 +       return -ESTALE;
26111 +}
26112 +#else
26113 +AuStubInt0(au_test_nfsd, void)
26114 +AuStubVoid(au_export_init, struct super_block *sb)
26115 +AuStubVoid(au_xigen_inc, struct inode *inode)
26116 +AuStubInt0(au_xigen_new, struct inode *inode)
26117 +AuStubInt0(au_xigen_set, struct super_block *sb, struct file *base)
26118 +AuStubVoid(au_xigen_clr, struct super_block *sb)
26119 +static inline int au_busy_or_stale(void)
26120 +{
26121 +       return -EBUSY;
26122 +}
26123 +#endif /* CONFIG_AUFS_EXPORT */
26124 +
26125 +/* ---------------------------------------------------------------------- */
26126 +
26127 +#ifdef CONFIG_AUFS_SBILIST
26128 +/* module.c */
26129 +extern struct au_splhead au_sbilist;
26130 +
26131 +static inline void au_sbilist_init(void)
26132 +{
26133 +       au_spl_init(&au_sbilist);
26134 +}
26135 +
26136 +static inline void au_sbilist_add(struct super_block *sb)
26137 +{
26138 +       au_spl_add(&au_sbi(sb)->si_list, &au_sbilist);
26139 +}
26140 +
26141 +static inline void au_sbilist_del(struct super_block *sb)
26142 +{
26143 +       au_spl_del(&au_sbi(sb)->si_list, &au_sbilist);
26144 +}
26145 +
26146 +#ifdef CONFIG_AUFS_MAGIC_SYSRQ
26147 +static inline void au_sbilist_lock(void)
26148 +{
26149 +       spin_lock(&au_sbilist.spin);
26150 +}
26151 +
26152 +static inline void au_sbilist_unlock(void)
26153 +{
26154 +       spin_unlock(&au_sbilist.spin);
26155 +}
26156 +#define AuGFP_SBILIST  GFP_ATOMIC
26157 +#else
26158 +AuStubVoid(au_sbilist_lock, void)
26159 +AuStubVoid(au_sbilist_unlock, void)
26160 +#define AuGFP_SBILIST  GFP_NOFS
26161 +#endif /* CONFIG_AUFS_MAGIC_SYSRQ */
26162 +#else
26163 +AuStubVoid(au_sbilist_init, void)
26164 +AuStubVoid(au_sbilist_add, struct super_block*)
26165 +AuStubVoid(au_sbilist_del, struct super_block*)
26166 +AuStubVoid(au_sbilist_lock, void)
26167 +AuStubVoid(au_sbilist_unlock, void)
26168 +#define AuGFP_SBILIST  GFP_NOFS
26169 +#endif
26170 +
26171 +/* ---------------------------------------------------------------------- */
26172 +
26173 +static inline void dbgaufs_si_null(struct au_sbinfo *sbinfo)
26174 +{
26175 +       /*
26176 +        * This function is a dynamic '__init' fucntion actually,
26177 +        * so the tiny check for si_rwsem is unnecessary.
26178 +        */
26179 +       /* AuRwMustWriteLock(&sbinfo->si_rwsem); */
26180 +#ifdef CONFIG_DEBUG_FS
26181 +       sbinfo->si_dbgaufs = NULL;
26182 +       sbinfo->si_dbgaufs_plink = NULL;
26183 +       sbinfo->si_dbgaufs_xib = NULL;
26184 +#ifdef CONFIG_AUFS_EXPORT
26185 +       sbinfo->si_dbgaufs_xigen = NULL;
26186 +#endif
26187 +#endif
26188 +}
26189 +
26190 +/* ---------------------------------------------------------------------- */
26191 +
26192 +static inline pid_t si_pid_bit(void)
26193 +{
26194 +       /* the origin of pid is 1, but the bitmap's is 0 */
26195 +       return current->pid - 1;
26196 +}
26197 +
26198 +static inline int si_pid_test(struct super_block *sb)
26199 +{
26200 +       pid_t bit = si_pid_bit();
26201 +       if (bit < PID_MAX_DEFAULT)
26202 +               return test_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
26203 +       else
26204 +               return si_pid_test_slow(sb);
26205 +}
26206 +
26207 +static inline void si_pid_set(struct super_block *sb)
26208 +{
26209 +       pid_t bit = si_pid_bit();
26210 +       if (bit < PID_MAX_DEFAULT) {
26211 +               AuDebugOn(test_bit(bit, au_sbi(sb)->au_si_pid.bitmap));
26212 +               set_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
26213 +               /* smp_mb(); */
26214 +       } else
26215 +               si_pid_set_slow(sb);
26216 +}
26217 +
26218 +static inline void si_pid_clr(struct super_block *sb)
26219 +{
26220 +       pid_t bit = si_pid_bit();
26221 +       if (bit < PID_MAX_DEFAULT) {
26222 +               AuDebugOn(!test_bit(bit, au_sbi(sb)->au_si_pid.bitmap));
26223 +               clear_bit(bit, au_sbi(sb)->au_si_pid.bitmap);
26224 +               /* smp_mb(); */
26225 +       } else
26226 +               si_pid_clr_slow(sb);
26227 +}
26228 +
26229 +/* ---------------------------------------------------------------------- */
26230 +
26231 +/* lock superblock. mainly for entry point functions */
26232 +/*
26233 + * __si_read_lock, __si_write_lock,
26234 + * __si_read_unlock, __si_write_unlock, __si_downgrade_lock
26235 + */
26236 +AuSimpleRwsemFuncs(__si, struct super_block *sb, &au_sbi(sb)->si_rwsem);
26237 +
26238 +#define SiMustNoWaiters(sb)    AuRwMustNoWaiters(&au_sbi(sb)->si_rwsem)
26239 +#define SiMustAnyLock(sb)      AuRwMustAnyLock(&au_sbi(sb)->si_rwsem)
26240 +#define SiMustWriteLock(sb)    AuRwMustWriteLock(&au_sbi(sb)->si_rwsem)
26241 +
26242 +static inline void si_noflush_read_lock(struct super_block *sb)
26243 +{
26244 +       __si_read_lock(sb);
26245 +       si_pid_set(sb);
26246 +}
26247 +
26248 +static inline int si_noflush_read_trylock(struct super_block *sb)
26249 +{
26250 +       int locked = __si_read_trylock(sb);
26251 +       if (locked)
26252 +               si_pid_set(sb);
26253 +       return locked;
26254 +}
26255 +
26256 +static inline void si_noflush_write_lock(struct super_block *sb)
26257 +{
26258 +       __si_write_lock(sb);
26259 +       si_pid_set(sb);
26260 +}
26261 +
26262 +static inline int si_noflush_write_trylock(struct super_block *sb)
26263 +{
26264 +       int locked = __si_write_trylock(sb);
26265 +       if (locked)
26266 +               si_pid_set(sb);
26267 +       return locked;
26268 +}
26269 +
26270 +#if 0 /* unused */
26271 +static inline int si_read_trylock(struct super_block *sb, int flags)
26272 +{
26273 +       if (au_ftest_lock(flags, FLUSH))
26274 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26275 +       return si_noflush_read_trylock(sb);
26276 +}
26277 +#endif
26278 +
26279 +static inline void si_read_unlock(struct super_block *sb)
26280 +{
26281 +       si_pid_clr(sb);
26282 +       __si_read_unlock(sb);
26283 +}
26284 +
26285 +#if 0 /* unused */
26286 +static inline int si_write_trylock(struct super_block *sb, int flags)
26287 +{
26288 +       if (au_ftest_lock(flags, FLUSH))
26289 +               au_nwt_flush(&au_sbi(sb)->si_nowait);
26290 +       return si_noflush_write_trylock(sb);
26291 +}
26292 +#endif
26293 +
26294 +static inline void si_write_unlock(struct super_block *sb)
26295 +{
26296 +       si_pid_clr(sb);
26297 +       __si_write_unlock(sb);
26298 +}
26299 +
26300 +#if 0 /* unused */
26301 +static inline void si_downgrade_lock(struct super_block *sb)
26302 +{
26303 +       __si_downgrade_lock(sb);
26304 +}
26305 +#endif
26306 +
26307 +/* ---------------------------------------------------------------------- */
26308 +
26309 +static inline aufs_bindex_t au_sbend(struct super_block *sb)
26310 +{
26311 +       SiMustAnyLock(sb);
26312 +       return au_sbi(sb)->si_bend;
26313 +}
26314 +
26315 +static inline unsigned int au_mntflags(struct super_block *sb)
26316 +{
26317 +       SiMustAnyLock(sb);
26318 +       return au_sbi(sb)->si_mntflags;
26319 +}
26320 +
26321 +static inline unsigned int au_sigen(struct super_block *sb)
26322 +{
26323 +       SiMustAnyLock(sb);
26324 +       return au_sbi(sb)->si_generation;
26325 +}
26326 +
26327 +static inline void au_ninodes_inc(struct super_block *sb)
26328 +{
26329 +       atomic_long_inc(&au_sbi(sb)->si_ninodes);
26330 +}
26331 +
26332 +static inline void au_ninodes_dec(struct super_block *sb)
26333 +{
26334 +       AuDebugOn(!atomic_long_read(&au_sbi(sb)->si_ninodes));
26335 +       atomic_long_dec(&au_sbi(sb)->si_ninodes);
26336 +}
26337 +
26338 +static inline void au_nfiles_inc(struct super_block *sb)
26339 +{
26340 +       atomic_long_inc(&au_sbi(sb)->si_nfiles);
26341 +}
26342 +
26343 +static inline void au_nfiles_dec(struct super_block *sb)
26344 +{
26345 +       AuDebugOn(!atomic_long_read(&au_sbi(sb)->si_nfiles));
26346 +       atomic_long_dec(&au_sbi(sb)->si_nfiles);
26347 +}
26348 +
26349 +static inline struct au_branch *au_sbr(struct super_block *sb,
26350 +                                      aufs_bindex_t bindex)
26351 +{
26352 +       SiMustAnyLock(sb);
26353 +       return au_sbi(sb)->si_branch[0 + bindex];
26354 +}
26355 +
26356 +static inline void au_xino_brid_set(struct super_block *sb, aufs_bindex_t brid)
26357 +{
26358 +       SiMustWriteLock(sb);
26359 +       au_sbi(sb)->si_xino_brid = brid;
26360 +}
26361 +
26362 +static inline aufs_bindex_t au_xino_brid(struct super_block *sb)
26363 +{
26364 +       SiMustAnyLock(sb);
26365 +       return au_sbi(sb)->si_xino_brid;
26366 +}
26367 +
26368 +#endif /* __KERNEL__ */
26369 +#endif /* __AUFS_SUPER_H__ */
26370 diff -urN /usr/share/empty/fs/aufs/sysaufs.c linux/fs/aufs/sysaufs.c
26371 --- /usr/share/empty/fs/aufs/sysaufs.c  1970-01-01 01:00:00.000000000 +0100
26372 +++ linux/fs/aufs/sysaufs.c     2014-04-24 22:11:10.858602483 +0200
26373 @@ -0,0 +1,104 @@
26374 +/*
26375 + * Copyright (C) 2005-2014 Junjiro R. Okajima
26376 + *
26377 + * This program, aufs is free software; you can redistribute it and/or modify
26378 + * it under the terms of the GNU General Public License as published by
26379 + * the Free Software Foundation; either version 2 of the License, or
26380 + * (at your option) any later version.
26381 + *
26382 + * This program is distributed in the hope that it will be useful,
26383 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26384 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26385 + * GNU General Public License for more details.
26386 + *
26387 + * You should have received a copy of the GNU General Public License
26388 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26389 + */
26390 +
26391 +/*
26392 + * sysfs interface and lifetime management
26393 + * they are necessary regardless sysfs is disabled.
26394 + */
26395 +
26396 +#include <linux/random.h>
26397 +#include "aufs.h"
26398 +
26399 +unsigned long sysaufs_si_mask;
26400 +struct kset *sysaufs_kset;
26401 +
26402 +#define AuSiAttr(_name) { \
26403 +       .attr   = { .name = __stringify(_name), .mode = 0444 }, \
26404 +       .show   = sysaufs_si_##_name,                           \
26405 +}
26406 +
26407 +static struct sysaufs_si_attr sysaufs_si_attr_xi_path = AuSiAttr(xi_path);
26408 +struct attribute *sysaufs_si_attrs[] = {
26409 +       &sysaufs_si_attr_xi_path.attr,
26410 +       NULL,
26411 +};
26412 +
26413 +static const struct sysfs_ops au_sbi_ops = {
26414 +       .show   = sysaufs_si_show
26415 +};
26416 +
26417 +static struct kobj_type au_sbi_ktype = {
26418 +       .release        = au_si_free,
26419 +       .sysfs_ops      = &au_sbi_ops,
26420 +       .default_attrs  = sysaufs_si_attrs
26421 +};
26422 +
26423 +/* ---------------------------------------------------------------------- */
26424 +
26425 +int sysaufs_si_init(struct au_sbinfo *sbinfo)
26426 +{
26427 +       int err;
26428 +
26429 +       sbinfo->si_kobj.kset = sysaufs_kset;
26430 +       /* cf. sysaufs_name() */
26431 +       err = kobject_init_and_add
26432 +               (&sbinfo->si_kobj, &au_sbi_ktype, /*&sysaufs_kset->kobj*/NULL,
26433 +                SysaufsSiNamePrefix "%lx", sysaufs_si_id(sbinfo));
26434 +
26435 +       dbgaufs_si_null(sbinfo);
26436 +       if (!err) {
26437 +               err = dbgaufs_si_init(sbinfo);
26438 +               if (unlikely(err))
26439 +                       kobject_put(&sbinfo->si_kobj);
26440 +       }
26441 +       return err;
26442 +}
26443 +
26444 +void sysaufs_fin(void)
26445 +{
26446 +       dbgaufs_fin();
26447 +       sysfs_remove_group(&sysaufs_kset->kobj, sysaufs_attr_group);
26448 +       kset_unregister(sysaufs_kset);
26449 +}
26450 +
26451 +int __init sysaufs_init(void)
26452 +{
26453 +       int err;
26454 +
26455 +       do {
26456 +               get_random_bytes(&sysaufs_si_mask, sizeof(sysaufs_si_mask));
26457 +       } while (!sysaufs_si_mask);
26458 +
26459 +       err = -EINVAL;
26460 +       sysaufs_kset = kset_create_and_add(AUFS_NAME, NULL, fs_kobj);
26461 +       if (unlikely(!sysaufs_kset))
26462 +               goto out;
26463 +       err = PTR_ERR(sysaufs_kset);
26464 +       if (IS_ERR(sysaufs_kset))
26465 +               goto out;
26466 +       err = sysfs_create_group(&sysaufs_kset->kobj, sysaufs_attr_group);
26467 +       if (unlikely(err)) {
26468 +               kset_unregister(sysaufs_kset);
26469 +               goto out;
26470 +       }
26471 +
26472 +       err = dbgaufs_init();
26473 +       if (unlikely(err))
26474 +               sysaufs_fin();
26475 +out:
26476 +       return err;
26477 +}
26478 diff -urN /usr/share/empty/fs/aufs/sysaufs.h linux/fs/aufs/sysaufs.h
26479 --- /usr/share/empty/fs/aufs/sysaufs.h  1970-01-01 01:00:00.000000000 +0100
26480 +++ linux/fs/aufs/sysaufs.h     2014-04-24 22:11:10.858602483 +0200
26481 @@ -0,0 +1,103 @@
26482 +/*
26483 + * Copyright (C) 2005-2014 Junjiro R. Okajima
26484 + *
26485 + * This program, aufs is free software; you can redistribute it and/or modify
26486 + * it under the terms of the GNU General Public License as published by
26487 + * the Free Software Foundation; either version 2 of the License, or
26488 + * (at your option) any later version.
26489 + *
26490 + * This program is distributed in the hope that it will be useful,
26491 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26492 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26493 + * GNU General Public License for more details.
26494 + *
26495 + * You should have received a copy of the GNU General Public License
26496 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26497 + */
26498 +
26499 +/*
26500 + * sysfs interface and mount lifetime management
26501 + */
26502 +
26503 +#ifndef __SYSAUFS_H__
26504 +#define __SYSAUFS_H__
26505 +
26506 +#ifdef __KERNEL__
26507 +
26508 +#include <linux/sysfs.h>
26509 +#include "module.h"
26510 +
26511 +struct super_block;
26512 +struct au_sbinfo;
26513 +
26514 +struct sysaufs_si_attr {
26515 +       struct attribute attr;
26516 +       int (*show)(struct seq_file *seq, struct super_block *sb);
26517 +};
26518 +
26519 +/* ---------------------------------------------------------------------- */
26520 +
26521 +/* sysaufs.c */
26522 +extern unsigned long sysaufs_si_mask;
26523 +extern struct kset *sysaufs_kset;
26524 +extern struct attribute *sysaufs_si_attrs[];
26525 +int sysaufs_si_init(struct au_sbinfo *sbinfo);
26526 +int __init sysaufs_init(void);
26527 +void sysaufs_fin(void);
26528 +
26529 +/* ---------------------------------------------------------------------- */
26530 +
26531 +/* some people doesn't like to show a pointer in kernel */
26532 +static inline unsigned long sysaufs_si_id(struct au_sbinfo *sbinfo)
26533 +{
26534 +       return sysaufs_si_mask ^ (unsigned long)sbinfo;
26535 +}
26536 +
26537 +#define SysaufsSiNamePrefix    "si_"
26538 +#define SysaufsSiNameLen       (sizeof(SysaufsSiNamePrefix) + 16)
26539 +static inline void sysaufs_name(struct au_sbinfo *sbinfo, char *name)
26540 +{
26541 +       snprintf(name, SysaufsSiNameLen, SysaufsSiNamePrefix "%lx",
26542 +                sysaufs_si_id(sbinfo));
26543 +}
26544 +
26545 +struct au_branch;
26546 +#ifdef CONFIG_SYSFS
26547 +/* sysfs.c */
26548 +extern struct attribute_group *sysaufs_attr_group;
26549 +
26550 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb);
26551 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
26552 +                        char *buf);
26553 +
26554 +void sysaufs_br_init(struct au_branch *br);
26555 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex);
26556 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex);
26557 +
26558 +#define sysaufs_brs_init()     do {} while (0)
26559 +
26560 +#else
26561 +#define sysaufs_attr_group     NULL
26562 +
26563 +AuStubInt0(sysaufs_si_xi_path, struct seq_file *seq, struct super_block *sb)
26564 +
26565 +static inline
26566 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
26567 +                        char *buf)
26568 +{
26569 +       return 0;
26570 +}
26571 +
26572 +AuStubVoid(sysaufs_br_init, struct au_branch *br)
26573 +AuStubVoid(sysaufs_brs_add, struct super_block *sb, aufs_bindex_t bindex)
26574 +AuStubVoid(sysaufs_brs_del, struct super_block *sb, aufs_bindex_t bindex)
26575 +
26576 +static inline void sysaufs_brs_init(void)
26577 +{
26578 +       sysaufs_brs = 0;
26579 +}
26580 +
26581 +#endif /* CONFIG_SYSFS */
26582 +
26583 +#endif /* __KERNEL__ */
26584 +#endif /* __SYSAUFS_H__ */
26585 diff -urN /usr/share/empty/fs/aufs/sysfs.c linux/fs/aufs/sysfs.c
26586 --- /usr/share/empty/fs/aufs/sysfs.c    1970-01-01 01:00:00.000000000 +0100
26587 +++ linux/fs/aufs/sysfs.c       2014-04-24 22:11:10.858602483 +0200
26588 @@ -0,0 +1,296 @@
26589 +/*
26590 + * Copyright (C) 2005-2014 Junjiro R. Okajima
26591 + *
26592 + * This program, aufs is free software; you can redistribute it and/or modify
26593 + * it under the terms of the GNU General Public License as published by
26594 + * the Free Software Foundation; either version 2 of the License, or
26595 + * (at your option) any later version.
26596 + *
26597 + * This program is distributed in the hope that it will be useful,
26598 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26599 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26600 + * GNU General Public License for more details.
26601 + *
26602 + * You should have received a copy of the GNU General Public License
26603 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26604 + */
26605 +
26606 +/*
26607 + * sysfs interface
26608 + */
26609 +
26610 +#include <linux/seq_file.h>
26611 +#include "aufs.h"
26612 +
26613 +#ifdef CONFIG_AUFS_FS_MODULE
26614 +/* this entry violates the "one line per file" policy of sysfs */
26615 +static ssize_t config_show(struct kobject *kobj, struct kobj_attribute *attr,
26616 +                          char *buf)
26617 +{
26618 +       ssize_t err;
26619 +       static char *conf =
26620 +/* this file is generated at compiling */
26621 +#include "conf.str"
26622 +               ;
26623 +
26624 +       err = snprintf(buf, PAGE_SIZE, conf);
26625 +       if (unlikely(err >= PAGE_SIZE))
26626 +               err = -EFBIG;
26627 +       return err;
26628 +}
26629 +
26630 +static struct kobj_attribute au_config_attr = __ATTR_RO(config);
26631 +#endif
26632 +
26633 +static struct attribute *au_attr[] = {
26634 +#ifdef CONFIG_AUFS_FS_MODULE
26635 +       &au_config_attr.attr,
26636 +#endif
26637 +       NULL,   /* need to NULL terminate the list of attributes */
26638 +};
26639 +
26640 +static struct attribute_group sysaufs_attr_group_body = {
26641 +       .attrs = au_attr
26642 +};
26643 +
26644 +struct attribute_group *sysaufs_attr_group = &sysaufs_attr_group_body;
26645 +
26646 +/* ---------------------------------------------------------------------- */
26647 +
26648 +int sysaufs_si_xi_path(struct seq_file *seq, struct super_block *sb)
26649 +{
26650 +       int err;
26651 +
26652 +       SiMustAnyLock(sb);
26653 +
26654 +       err = 0;
26655 +       if (au_opt_test(au_mntflags(sb), XINO)) {
26656 +               err = au_xino_path(seq, au_sbi(sb)->si_xib);
26657 +               seq_putc(seq, '\n');
26658 +       }
26659 +       return err;
26660 +}
26661 +
26662 +/*
26663 + * the lifetime of branch is independent from the entry under sysfs.
26664 + * sysfs handles the lifetime of the entry, and never call ->show() after it is
26665 + * unlinked.
26666 + */
26667 +static int sysaufs_si_br(struct seq_file *seq, struct super_block *sb,
26668 +                        aufs_bindex_t bindex, int idx)
26669 +{
26670 +       int err;
26671 +       struct path path;
26672 +       struct dentry *root;
26673 +       struct au_branch *br;
26674 +       char *perm;
26675 +
26676 +       AuDbg("b%d\n", bindex);
26677 +
26678 +       err = 0;
26679 +       root = sb->s_root;
26680 +       di_read_lock_parent(root, !AuLock_IR);
26681 +       br = au_sbr(sb, bindex);
26682 +
26683 +       switch (idx) {
26684 +       case AuBrSysfs_BR:
26685 +               path.mnt = au_br_mnt(br);
26686 +               path.dentry = au_h_dptr(root, bindex);
26687 +               au_seq_path(seq, &path);
26688 +               di_read_unlock(root, !AuLock_IR);
26689 +               perm = au_optstr_br_perm(br->br_perm);
26690 +               if (perm) {
26691 +                       err = seq_printf(seq, "=%s\n", perm);
26692 +                       kfree(perm);
26693 +                       if (err == -1)
26694 +                               err = -E2BIG;
26695 +               } else
26696 +                       err = -ENOMEM;
26697 +               break;
26698 +       case AuBrSysfs_BRID:
26699 +               err = seq_printf(seq, "%d\n", br->br_id);
26700 +               di_read_unlock(root, !AuLock_IR);
26701 +               if (err == -1)
26702 +                       err = -E2BIG;
26703 +               break;
26704 +       }
26705 +
26706 +       return err;
26707 +}
26708 +
26709 +/* ---------------------------------------------------------------------- */
26710 +
26711 +static struct seq_file *au_seq(char *p, ssize_t len)
26712 +{
26713 +       struct seq_file *seq;
26714 +
26715 +       seq = kzalloc(sizeof(*seq), GFP_NOFS);
26716 +       if (seq) {
26717 +               /* mutex_init(&seq.lock); */
26718 +               seq->buf = p;
26719 +               seq->size = len;
26720 +               return seq; /* success */
26721 +       }
26722 +
26723 +       seq = ERR_PTR(-ENOMEM);
26724 +       return seq;
26725 +}
26726 +
26727 +#define SysaufsBr_PREFIX       "br"
26728 +#define SysaufsBrid_PREFIX     "brid"
26729 +
26730 +/* todo: file size may exceed PAGE_SIZE */
26731 +ssize_t sysaufs_si_show(struct kobject *kobj, struct attribute *attr,
26732 +                       char *buf)
26733 +{
26734 +       ssize_t err;
26735 +       int idx;
26736 +       long l;
26737 +       aufs_bindex_t bend;
26738 +       struct au_sbinfo *sbinfo;
26739 +       struct super_block *sb;
26740 +       struct seq_file *seq;
26741 +       char *name;
26742 +       struct attribute **cattr;
26743 +
26744 +       sbinfo = container_of(kobj, struct au_sbinfo, si_kobj);
26745 +       sb = sbinfo->si_sb;
26746 +
26747 +       /*
26748 +        * prevent a race condition between sysfs and aufs.
26749 +        * for instance, sysfs_file_read() calls sysfs_get_active_two() which
26750 +        * prohibits maintaining the sysfs entries.
26751 +        * hew we acquire read lock after sysfs_get_active_two().
26752 +        * on the other hand, the remount process may maintain the sysfs/aufs
26753 +        * entries after acquiring write lock.
26754 +        * it can cause a deadlock.
26755 +        * simply we gave up processing read here.
26756 +        */
26757 +       err = -EBUSY;
26758 +       if (unlikely(!si_noflush_read_trylock(sb)))
26759 +               goto out;
26760 +
26761 +       seq = au_seq(buf, PAGE_SIZE);
26762 +       err = PTR_ERR(seq);
26763 +       if (IS_ERR(seq))
26764 +               goto out_unlock;
26765 +
26766 +       name = (void *)attr->name;
26767 +       cattr = sysaufs_si_attrs;
26768 +       while (*cattr) {
26769 +               if (!strcmp(name, (*cattr)->name)) {
26770 +                       err = container_of(*cattr, struct sysaufs_si_attr, attr)
26771 +                               ->show(seq, sb);
26772 +                       goto out_seq;
26773 +               }
26774 +               cattr++;
26775 +       }
26776 +
26777 +       if (!strncmp(name, SysaufsBrid_PREFIX,
26778 +                    sizeof(SysaufsBrid_PREFIX) - 1)) {
26779 +               idx = AuBrSysfs_BRID;
26780 +               name += sizeof(SysaufsBrid_PREFIX) - 1;
26781 +       } else if (!strncmp(name, SysaufsBr_PREFIX,
26782 +                           sizeof(SysaufsBr_PREFIX) - 1)) {
26783 +               idx = AuBrSysfs_BR;
26784 +               name += sizeof(SysaufsBr_PREFIX) - 1;
26785 +       } else
26786 +                 BUG();
26787 +
26788 +       err = kstrtol(name, 10, &l);
26789 +       if (!err) {
26790 +               bend = au_sbend(sb);
26791 +               if (l <= bend)
26792 +                       err = sysaufs_si_br(seq, sb, (aufs_bindex_t)l, idx);
26793 +               else
26794 +                       err = -ENOENT;
26795 +       }
26796 +
26797 +out_seq:
26798 +       if (!err) {
26799 +               err = seq->count;
26800 +               /* sysfs limit */
26801 +               if (unlikely(err == PAGE_SIZE))
26802 +                       err = -EFBIG;
26803 +       }
26804 +       kfree(seq);
26805 +out_unlock:
26806 +       si_read_unlock(sb);
26807 +out:
26808 +       return err;
26809 +}
26810 +
26811 +/* ---------------------------------------------------------------------- */
26812 +
26813 +void sysaufs_br_init(struct au_branch *br)
26814 +{
26815 +       int i;
26816 +       struct au_brsysfs *br_sysfs;
26817 +       struct attribute *attr;
26818 +
26819 +       br_sysfs = br->br_sysfs;
26820 +       for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
26821 +               attr = &br_sysfs->attr;
26822 +               sysfs_attr_init(attr);
26823 +               attr->name = br_sysfs->name;
26824 +               attr->mode = S_IRUGO;
26825 +               br_sysfs++;
26826 +       }
26827 +}
26828 +
26829 +void sysaufs_brs_del(struct super_block *sb, aufs_bindex_t bindex)
26830 +{
26831 +       struct au_branch *br;
26832 +       struct kobject *kobj;
26833 +       struct au_brsysfs *br_sysfs;
26834 +       int i;
26835 +       aufs_bindex_t bend;
26836 +
26837 +       dbgaufs_brs_del(sb, bindex);
26838 +
26839 +       if (!sysaufs_brs)
26840 +               return;
26841 +
26842 +       kobj = &au_sbi(sb)->si_kobj;
26843 +       bend = au_sbend(sb);
26844 +       for (; bindex <= bend; bindex++) {
26845 +               br = au_sbr(sb, bindex);
26846 +               br_sysfs = br->br_sysfs;
26847 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
26848 +                       sysfs_remove_file(kobj, &br_sysfs->attr);
26849 +                       br_sysfs++;
26850 +               }
26851 +       }
26852 +}
26853 +
26854 +void sysaufs_brs_add(struct super_block *sb, aufs_bindex_t bindex)
26855 +{
26856 +       int err, i;
26857 +       aufs_bindex_t bend;
26858 +       struct kobject *kobj;
26859 +       struct au_branch *br;
26860 +       struct au_brsysfs *br_sysfs;
26861 +
26862 +       dbgaufs_brs_add(sb, bindex);
26863 +
26864 +       if (!sysaufs_brs)
26865 +               return;
26866 +
26867 +       kobj = &au_sbi(sb)->si_kobj;
26868 +       bend = au_sbend(sb);
26869 +       for (; bindex <= bend; bindex++) {
26870 +               br = au_sbr(sb, bindex);
26871 +               br_sysfs = br->br_sysfs;
26872 +               snprintf(br_sysfs[AuBrSysfs_BR].name, sizeof(br_sysfs->name),
26873 +                        SysaufsBr_PREFIX "%d", bindex);
26874 +               snprintf(br_sysfs[AuBrSysfs_BRID].name, sizeof(br_sysfs->name),
26875 +                        SysaufsBrid_PREFIX "%d", bindex);
26876 +               for (i = 0; i < ARRAY_SIZE(br->br_sysfs); i++) {
26877 +                       err = sysfs_create_file(kobj, &br_sysfs->attr);
26878 +                       if (unlikely(err))
26879 +                               pr_warn("failed %s under sysfs(%d)\n",
26880 +                                       br_sysfs->name, err);
26881 +                       br_sysfs++;
26882 +               }
26883 +       }
26884 +}
26885 diff -urN /usr/share/empty/fs/aufs/sysrq.c linux/fs/aufs/sysrq.c
26886 --- /usr/share/empty/fs/aufs/sysrq.c    1970-01-01 01:00:00.000000000 +0100
26887 +++ linux/fs/aufs/sysrq.c       2014-04-24 22:11:10.858602483 +0200
26888 @@ -0,0 +1,154 @@
26889 +/*
26890 + * Copyright (C) 2005-2014 Junjiro R. Okajima
26891 + *
26892 + * This program, aufs is free software; you can redistribute it and/or modify
26893 + * it under the terms of the GNU General Public License as published by
26894 + * the Free Software Foundation; either version 2 of the License, or
26895 + * (at your option) any later version.
26896 + *
26897 + * This program is distributed in the hope that it will be useful,
26898 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
26899 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
26900 + * GNU General Public License for more details.
26901 + *
26902 + * You should have received a copy of the GNU General Public License
26903 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
26904 + */
26905 +
26906 +/*
26907 + * magic sysrq hanlder
26908 + */
26909 +
26910 +/* #include <linux/sysrq.h> */
26911 +#include <linux/writeback.h>
26912 +#include "aufs.h"
26913 +
26914 +/* ---------------------------------------------------------------------- */
26915 +
26916 +static void sysrq_sb(struct super_block *sb)
26917 +{
26918 +       char *plevel;
26919 +       struct au_sbinfo *sbinfo;
26920 +       struct file *file;
26921 +       struct au_sphlhead *files;
26922 +       struct au_finfo *finfo;
26923 +
26924 +       plevel = au_plevel;
26925 +       au_plevel = KERN_WARNING;
26926 +
26927 +       /* since we define pr_fmt, call printk directly */
26928 +#define pr(str) printk(KERN_WARNING AUFS_NAME ": " str)
26929 +
26930 +       sbinfo = au_sbi(sb);
26931 +       printk(KERN_WARNING "si=%lx\n", sysaufs_si_id(sbinfo));
26932 +       pr("superblock\n");
26933 +       au_dpri_sb(sb);
26934 +
26935 +#if 0
26936 +       pr("root dentry\n");
26937 +       au_dpri_dentry(sb->s_root);
26938 +       pr("root inode\n");
26939 +       au_dpri_inode(sb->s_root->d_inode);
26940 +#endif
26941 +
26942 +#if 0
26943 +       do {
26944 +               int err, i, j, ndentry;
26945 +               struct au_dcsub_pages dpages;
26946 +               struct au_dpage *dpage;
26947 +
26948 +               err = au_dpages_init(&dpages, GFP_ATOMIC);
26949 +               if (unlikely(err))
26950 +                       break;
26951 +               err = au_dcsub_pages(&dpages, sb->s_root, NULL, NULL);
26952 +               if (!err)
26953 +                       for (i = 0; i < dpages.ndpage; i++) {
26954 +                               dpage = dpages.dpages + i;
26955 +                               ndentry = dpage->ndentry;
26956 +                               for (j = 0; j < ndentry; j++)
26957 +                                       au_dpri_dentry(dpage->dentries[j]);
26958 +                       }
26959 +               au_dpages_free(&dpages);
26960 +       } while (0);
26961 +#endif
26962 +
26963 +#if 1
26964 +       {
26965 +               struct inode *i;
26966 +               pr("isolated inode\n");
26967 +               spin_lock(&inode_sb_list_lock);
26968 +               list_for_each_entry(i, &sb->s_inodes, i_sb_list) {
26969 +                       spin_lock(&i->i_lock);
26970 +                       if (1 || hlist_empty(&i->i_dentry))
26971 +                               au_dpri_inode(i);
26972 +                       spin_unlock(&i->i_lock);
26973 +               }
26974 +               spin_unlock(&inode_sb_list_lock);
26975 +       }
26976 +#endif
26977 +       pr("files\n");
26978 +       files = &au_sbi(sb)->si_files;
26979 +       spin_lock(&files->spin);
26980 +       hlist_for_each_entry(finfo, &files->head, fi_hlist) {
26981 +               umode_t mode;
26982 +               file = finfo->fi_file;
26983 +               mode = file_inode(file)->i_mode;
26984 +               if (!special_file(mode) || au_special_file(mode))
26985 +                       au_dpri_file(file);
26986 +       }
26987 +       spin_unlock(&files->spin);
26988 +       pr("done\n");
26989 +
26990 +#undef pr
26991 +       au_plevel = plevel;
26992 +}
26993 +
26994 +/* ---------------------------------------------------------------------- */
26995 +
26996 +/* module parameter */
26997 +static char *aufs_sysrq_key = "a";
26998 +module_param_named(sysrq, aufs_sysrq_key, charp, S_IRUGO);
26999 +MODULE_PARM_DESC(sysrq, "MagicSysRq key for " AUFS_NAME);
27000 +
27001 +static void au_sysrq(int key __maybe_unused)
27002 +{
27003 +       struct au_sbinfo *sbinfo;
27004 +
27005 +       lockdep_off();
27006 +       au_sbilist_lock();
27007 +       list_for_each_entry(sbinfo, &au_sbilist.head, si_list)
27008 +               sysrq_sb(sbinfo->si_sb);
27009 +       au_sbilist_unlock();
27010 +       lockdep_on();
27011 +}
27012 +
27013 +static struct sysrq_key_op au_sysrq_op = {
27014 +       .handler        = au_sysrq,
27015 +       .help_msg       = "Aufs",
27016 +       .action_msg     = "Aufs",
27017 +       .enable_mask    = SYSRQ_ENABLE_DUMP
27018 +};
27019 +
27020 +/* ---------------------------------------------------------------------- */
27021 +
27022 +int __init au_sysrq_init(void)
27023 +{
27024 +       int err;
27025 +       char key;
27026 +
27027 +       err = -1;
27028 +       key = *aufs_sysrq_key;
27029 +       if ('a' <= key && key <= 'z')
27030 +               err = register_sysrq_key(key, &au_sysrq_op);
27031 +       if (unlikely(err))
27032 +               pr_err("err %d, sysrq=%c\n", err, key);
27033 +       return err;
27034 +}
27035 +
27036 +void au_sysrq_fin(void)
27037 +{
27038 +       int err;
27039 +       err = unregister_sysrq_key(*aufs_sysrq_key, &au_sysrq_op);
27040 +       if (unlikely(err))
27041 +               pr_err("err %d (ignored)\n", err);
27042 +}
27043 diff -urN /usr/share/empty/fs/aufs/vdir.c linux/fs/aufs/vdir.c
27044 --- /usr/share/empty/fs/aufs/vdir.c     1970-01-01 01:00:00.000000000 +0100
27045 +++ linux/fs/aufs/vdir.c        2014-04-24 22:11:10.858602483 +0200
27046 @@ -0,0 +1,887 @@
27047 +/*
27048 + * Copyright (C) 2005-2014 Junjiro R. Okajima
27049 + *
27050 + * This program, aufs is free software; you can redistribute it and/or modify
27051 + * it under the terms of the GNU General Public License as published by
27052 + * the Free Software Foundation; either version 2 of the License, or
27053 + * (at your option) any later version.
27054 + *
27055 + * This program is distributed in the hope that it will be useful,
27056 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27057 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27058 + * GNU General Public License for more details.
27059 + *
27060 + * You should have received a copy of the GNU General Public License
27061 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27062 + */
27063 +
27064 +/*
27065 + * virtual or vertical directory
27066 + */
27067 +
27068 +#include "aufs.h"
27069 +
27070 +static unsigned int calc_size(int nlen)
27071 +{
27072 +       return ALIGN(sizeof(struct au_vdir_de) + nlen, sizeof(ino_t));
27073 +}
27074 +
27075 +static int set_deblk_end(union au_vdir_deblk_p *p,
27076 +                        union au_vdir_deblk_p *deblk_end)
27077 +{
27078 +       if (calc_size(0) <= deblk_end->deblk - p->deblk) {
27079 +               p->de->de_str.len = 0;
27080 +               /* smp_mb(); */
27081 +               return 0;
27082 +       }
27083 +       return -1; /* error */
27084 +}
27085 +
27086 +/* returns true or false */
27087 +static int is_deblk_end(union au_vdir_deblk_p *p,
27088 +                       union au_vdir_deblk_p *deblk_end)
27089 +{
27090 +       if (calc_size(0) <= deblk_end->deblk - p->deblk)
27091 +               return !p->de->de_str.len;
27092 +       return 1;
27093 +}
27094 +
27095 +static unsigned char *last_deblk(struct au_vdir *vdir)
27096 +{
27097 +       return vdir->vd_deblk[vdir->vd_nblk - 1];
27098 +}
27099 +
27100 +/* ---------------------------------------------------------------------- */
27101 +
27102 +/* estimate the apropriate size for name hash table */
27103 +unsigned int au_rdhash_est(loff_t sz)
27104 +{
27105 +       unsigned int n;
27106 +
27107 +       n = UINT_MAX;
27108 +       sz >>= 10;
27109 +       if (sz < n)
27110 +               n = sz;
27111 +       if (sz < AUFS_RDHASH_DEF)
27112 +               n = AUFS_RDHASH_DEF;
27113 +       /* pr_info("n %u\n", n); */
27114 +       return n;
27115 +}
27116 +
27117 +/*
27118 + * the allocated memory has to be freed by
27119 + * au_nhash_wh_free() or au_nhash_de_free().
27120 + */
27121 +int au_nhash_alloc(struct au_nhash *nhash, unsigned int num_hash, gfp_t gfp)
27122 +{
27123 +       struct hlist_head *head;
27124 +       unsigned int u;
27125 +
27126 +       head = kmalloc(sizeof(*nhash->nh_head) * num_hash, gfp);
27127 +       if (head) {
27128 +               nhash->nh_num = num_hash;
27129 +               nhash->nh_head = head;
27130 +               for (u = 0; u < num_hash; u++)
27131 +                       INIT_HLIST_HEAD(head++);
27132 +               return 0; /* success */
27133 +       }
27134 +
27135 +       return -ENOMEM;
27136 +}
27137 +
27138 +static void nhash_count(struct hlist_head *head)
27139 +{
27140 +#if 0
27141 +       unsigned long n;
27142 +       struct hlist_node *pos;
27143 +
27144 +       n = 0;
27145 +       hlist_for_each(pos, head)
27146 +               n++;
27147 +       pr_info("%lu\n", n);
27148 +#endif
27149 +}
27150 +
27151 +static void au_nhash_wh_do_free(struct hlist_head *head)
27152 +{
27153 +       struct au_vdir_wh *pos;
27154 +       struct hlist_node *node;
27155 +
27156 +       hlist_for_each_entry_safe(pos, node, head, wh_hash)
27157 +               kfree(pos);
27158 +}
27159 +
27160 +static void au_nhash_de_do_free(struct hlist_head *head)
27161 +{
27162 +       struct au_vdir_dehstr *pos;
27163 +       struct hlist_node *node;
27164 +
27165 +       hlist_for_each_entry_safe(pos, node, head, hash)
27166 +               au_cache_free_vdir_dehstr(pos);
27167 +}
27168 +
27169 +static void au_nhash_do_free(struct au_nhash *nhash,
27170 +                            void (*free)(struct hlist_head *head))
27171 +{
27172 +       unsigned int n;
27173 +       struct hlist_head *head;
27174 +
27175 +       n = nhash->nh_num;
27176 +       if (!n)
27177 +               return;
27178 +
27179 +       head = nhash->nh_head;
27180 +       while (n-- > 0) {
27181 +               nhash_count(head);
27182 +               free(head++);
27183 +       }
27184 +       kfree(nhash->nh_head);
27185 +}
27186 +
27187 +void au_nhash_wh_free(struct au_nhash *whlist)
27188 +{
27189 +       au_nhash_do_free(whlist, au_nhash_wh_do_free);
27190 +}
27191 +
27192 +static void au_nhash_de_free(struct au_nhash *delist)
27193 +{
27194 +       au_nhash_do_free(delist, au_nhash_de_do_free);
27195 +}
27196 +
27197 +/* ---------------------------------------------------------------------- */
27198 +
27199 +int au_nhash_test_longer_wh(struct au_nhash *whlist, aufs_bindex_t btgt,
27200 +                           int limit)
27201 +{
27202 +       int num;
27203 +       unsigned int u, n;
27204 +       struct hlist_head *head;
27205 +       struct au_vdir_wh *pos;
27206 +
27207 +       num = 0;
27208 +       n = whlist->nh_num;
27209 +       head = whlist->nh_head;
27210 +       for (u = 0; u < n; u++, head++)
27211 +               hlist_for_each_entry(pos, head, wh_hash)
27212 +                       if (pos->wh_bindex == btgt && ++num > limit)
27213 +                               return 1;
27214 +       return 0;
27215 +}
27216 +
27217 +static struct hlist_head *au_name_hash(struct au_nhash *nhash,
27218 +                                      unsigned char *name,
27219 +                                      unsigned int len)
27220 +{
27221 +       unsigned int v;
27222 +       /* const unsigned int magic_bit = 12; */
27223 +
27224 +       AuDebugOn(!nhash->nh_num || !nhash->nh_head);
27225 +
27226 +       v = 0;
27227 +       while (len--)
27228 +               v += *name++;
27229 +       /* v = hash_long(v, magic_bit); */
27230 +       v %= nhash->nh_num;
27231 +       return nhash->nh_head + v;
27232 +}
27233 +
27234 +static int au_nhash_test_name(struct au_vdir_destr *str, const char *name,
27235 +                             int nlen)
27236 +{
27237 +       return str->len == nlen && !memcmp(str->name, name, nlen);
27238 +}
27239 +
27240 +/* returns found or not */
27241 +int au_nhash_test_known_wh(struct au_nhash *whlist, char *name, int nlen)
27242 +{
27243 +       struct hlist_head *head;
27244 +       struct au_vdir_wh *pos;
27245 +       struct au_vdir_destr *str;
27246 +
27247 +       head = au_name_hash(whlist, name, nlen);
27248 +       hlist_for_each_entry(pos, head, wh_hash) {
27249 +               str = &pos->wh_str;
27250 +               AuDbg("%.*s\n", str->len, str->name);
27251 +               if (au_nhash_test_name(str, name, nlen))
27252 +                       return 1;
27253 +       }
27254 +       return 0;
27255 +}
27256 +
27257 +/* returns found(true) or not */
27258 +static int test_known(struct au_nhash *delist, char *name, int nlen)
27259 +{
27260 +       struct hlist_head *head;
27261 +       struct au_vdir_dehstr *pos;
27262 +       struct au_vdir_destr *str;
27263 +
27264 +       head = au_name_hash(delist, name, nlen);
27265 +       hlist_for_each_entry(pos, head, hash) {
27266 +               str = pos->str;
27267 +               AuDbg("%.*s\n", str->len, str->name);
27268 +               if (au_nhash_test_name(str, name, nlen))
27269 +                       return 1;
27270 +       }
27271 +       return 0;
27272 +}
27273 +
27274 +static void au_shwh_init_wh(struct au_vdir_wh *wh, ino_t ino,
27275 +                           unsigned char d_type)
27276 +{
27277 +#ifdef CONFIG_AUFS_SHWH
27278 +       wh->wh_ino = ino;
27279 +       wh->wh_type = d_type;
27280 +#endif
27281 +}
27282 +
27283 +/* ---------------------------------------------------------------------- */
27284 +
27285 +int au_nhash_append_wh(struct au_nhash *whlist, char *name, int nlen, ino_t ino,
27286 +                      unsigned int d_type, aufs_bindex_t bindex,
27287 +                      unsigned char shwh)
27288 +{
27289 +       int err;
27290 +       struct au_vdir_destr *str;
27291 +       struct au_vdir_wh *wh;
27292 +
27293 +       AuDbg("%.*s\n", nlen, name);
27294 +       AuDebugOn(!whlist->nh_num || !whlist->nh_head);
27295 +
27296 +       err = -ENOMEM;
27297 +       wh = kmalloc(sizeof(*wh) + nlen, GFP_NOFS);
27298 +       if (unlikely(!wh))
27299 +               goto out;
27300 +
27301 +       err = 0;
27302 +       wh->wh_bindex = bindex;
27303 +       if (shwh)
27304 +               au_shwh_init_wh(wh, ino, d_type);
27305 +       str = &wh->wh_str;
27306 +       str->len = nlen;
27307 +       memcpy(str->name, name, nlen);
27308 +       hlist_add_head(&wh->wh_hash, au_name_hash(whlist, name, nlen));
27309 +       /* smp_mb(); */
27310 +
27311 +out:
27312 +       return err;
27313 +}
27314 +
27315 +static int append_deblk(struct au_vdir *vdir)
27316 +{
27317 +       int err;
27318 +       unsigned long ul;
27319 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
27320 +       union au_vdir_deblk_p p, deblk_end;
27321 +       unsigned char **o;
27322 +
27323 +       err = -ENOMEM;
27324 +       o = krealloc(vdir->vd_deblk, sizeof(*o) * (vdir->vd_nblk + 1),
27325 +                    GFP_NOFS);
27326 +       if (unlikely(!o))
27327 +               goto out;
27328 +
27329 +       vdir->vd_deblk = o;
27330 +       p.deblk = kmalloc(deblk_sz, GFP_NOFS);
27331 +       if (p.deblk) {
27332 +               ul = vdir->vd_nblk++;
27333 +               vdir->vd_deblk[ul] = p.deblk;
27334 +               vdir->vd_last.ul = ul;
27335 +               vdir->vd_last.p.deblk = p.deblk;
27336 +               deblk_end.deblk = p.deblk + deblk_sz;
27337 +               err = set_deblk_end(&p, &deblk_end);
27338 +       }
27339 +
27340 +out:
27341 +       return err;
27342 +}
27343 +
27344 +static int append_de(struct au_vdir *vdir, char *name, int nlen, ino_t ino,
27345 +                    unsigned int d_type, struct au_nhash *delist)
27346 +{
27347 +       int err;
27348 +       unsigned int sz;
27349 +       const unsigned int deblk_sz = vdir->vd_deblk_sz;
27350 +       union au_vdir_deblk_p p, *room, deblk_end;
27351 +       struct au_vdir_dehstr *dehstr;
27352 +
27353 +       p.deblk = last_deblk(vdir);
27354 +       deblk_end.deblk = p.deblk + deblk_sz;
27355 +       room = &vdir->vd_last.p;
27356 +       AuDebugOn(room->deblk < p.deblk || deblk_end.deblk <= room->deblk
27357 +                 || !is_deblk_end(room, &deblk_end));
27358 +
27359 +       sz = calc_size(nlen);
27360 +       if (unlikely(sz > deblk_end.deblk - room->deblk)) {
27361 +               err = append_deblk(vdir);
27362 +               if (unlikely(err))
27363 +                       goto out;
27364 +
27365 +               p.deblk = last_deblk(vdir);
27366 +               deblk_end.deblk = p.deblk + deblk_sz;
27367 +               /* smp_mb(); */
27368 +               AuDebugOn(room->deblk != p.deblk);
27369 +       }
27370 +
27371 +       err = -ENOMEM;
27372 +       dehstr = au_cache_alloc_vdir_dehstr();
27373 +       if (unlikely(!dehstr))
27374 +               goto out;
27375 +
27376 +       dehstr->str = &room->de->de_str;
27377 +       hlist_add_head(&dehstr->hash, au_name_hash(delist, name, nlen));
27378 +       room->de->de_ino = ino;
27379 +       room->de->de_type = d_type;
27380 +       room->de->de_str.len = nlen;
27381 +       memcpy(room->de->de_str.name, name, nlen);
27382 +
27383 +       err = 0;
27384 +       room->deblk += sz;
27385 +       if (unlikely(set_deblk_end(room, &deblk_end)))
27386 +               err = append_deblk(vdir);
27387 +       /* smp_mb(); */
27388 +
27389 +out:
27390 +       return err;
27391 +}
27392 +
27393 +/* ---------------------------------------------------------------------- */
27394 +
27395 +void au_vdir_free(struct au_vdir *vdir)
27396 +{
27397 +       unsigned char **deblk;
27398 +
27399 +       deblk = vdir->vd_deblk;
27400 +       while (vdir->vd_nblk--)
27401 +               kfree(*deblk++);
27402 +       kfree(vdir->vd_deblk);
27403 +       au_cache_free_vdir(vdir);
27404 +}
27405 +
27406 +static struct au_vdir *alloc_vdir(struct file *file)
27407 +{
27408 +       struct au_vdir *vdir;
27409 +       struct super_block *sb;
27410 +       int err;
27411 +
27412 +       sb = file->f_dentry->d_sb;
27413 +       SiMustAnyLock(sb);
27414 +
27415 +       err = -ENOMEM;
27416 +       vdir = au_cache_alloc_vdir();
27417 +       if (unlikely(!vdir))
27418 +               goto out;
27419 +
27420 +       vdir->vd_deblk = kzalloc(sizeof(*vdir->vd_deblk), GFP_NOFS);
27421 +       if (unlikely(!vdir->vd_deblk))
27422 +               goto out_free;
27423 +
27424 +       vdir->vd_deblk_sz = au_sbi(sb)->si_rdblk;
27425 +       if (!vdir->vd_deblk_sz) {
27426 +               /* estimate the apropriate size for deblk */
27427 +               vdir->vd_deblk_sz = au_dir_size(file, /*dentry*/NULL);
27428 +               /* pr_info("vd_deblk_sz %u\n", vdir->vd_deblk_sz); */
27429 +       }
27430 +       vdir->vd_nblk = 0;
27431 +       vdir->vd_version = 0;
27432 +       vdir->vd_jiffy = 0;
27433 +       err = append_deblk(vdir);
27434 +       if (!err)
27435 +               return vdir; /* success */
27436 +
27437 +       kfree(vdir->vd_deblk);
27438 +
27439 +out_free:
27440 +       au_cache_free_vdir(vdir);
27441 +out:
27442 +       vdir = ERR_PTR(err);
27443 +       return vdir;
27444 +}
27445 +
27446 +static int reinit_vdir(struct au_vdir *vdir)
27447 +{
27448 +       int err;
27449 +       union au_vdir_deblk_p p, deblk_end;
27450 +
27451 +       while (vdir->vd_nblk > 1) {
27452 +               kfree(vdir->vd_deblk[vdir->vd_nblk - 1]);
27453 +               /* vdir->vd_deblk[vdir->vd_nblk - 1] = NULL; */
27454 +               vdir->vd_nblk--;
27455 +       }
27456 +       p.deblk = vdir->vd_deblk[0];
27457 +       deblk_end.deblk = p.deblk + vdir->vd_deblk_sz;
27458 +       err = set_deblk_end(&p, &deblk_end);
27459 +       /* keep vd_dblk_sz */
27460 +       vdir->vd_last.ul = 0;
27461 +       vdir->vd_last.p.deblk = vdir->vd_deblk[0];
27462 +       vdir->vd_version = 0;
27463 +       vdir->vd_jiffy = 0;
27464 +       /* smp_mb(); */
27465 +       return err;
27466 +}
27467 +
27468 +/* ---------------------------------------------------------------------- */
27469 +
27470 +#define AuFillVdir_CALLED      1
27471 +#define AuFillVdir_WHABLE      (1 << 1)
27472 +#define AuFillVdir_SHWH                (1 << 2)
27473 +#define au_ftest_fillvdir(flags, name) ((flags) & AuFillVdir_##name)
27474 +#define au_fset_fillvdir(flags, name) \
27475 +       do { (flags) |= AuFillVdir_##name; } while (0)
27476 +#define au_fclr_fillvdir(flags, name) \
27477 +       do { (flags) &= ~AuFillVdir_##name; } while (0)
27478 +
27479 +#ifndef CONFIG_AUFS_SHWH
27480 +#undef AuFillVdir_SHWH
27481 +#define AuFillVdir_SHWH                0
27482 +#endif
27483 +
27484 +struct fillvdir_arg {
27485 +       struct dir_context      ctx;
27486 +       struct file             *file;
27487 +       struct au_vdir          *vdir;
27488 +       struct au_nhash         delist;
27489 +       struct au_nhash         whlist;
27490 +       aufs_bindex_t           bindex;
27491 +       unsigned int            flags;
27492 +       int                     err;
27493 +};
27494 +
27495 +static int fillvdir(struct dir_context *ctx, const char *__name, int nlen,
27496 +                   loff_t offset __maybe_unused, u64 h_ino,
27497 +                   unsigned int d_type)
27498 +{
27499 +       struct fillvdir_arg *arg = container_of(ctx, struct fillvdir_arg, ctx);
27500 +       char *name = (void *)__name;
27501 +       struct super_block *sb;
27502 +       ino_t ino;
27503 +       const unsigned char shwh = !!au_ftest_fillvdir(arg->flags, SHWH);
27504 +
27505 +       arg->err = 0;
27506 +       sb = arg->file->f_dentry->d_sb;
27507 +       au_fset_fillvdir(arg->flags, CALLED);
27508 +       /* smp_mb(); */
27509 +       if (nlen <= AUFS_WH_PFX_LEN
27510 +           || memcmp(name, AUFS_WH_PFX, AUFS_WH_PFX_LEN)) {
27511 +               if (test_known(&arg->delist, name, nlen)
27512 +                   || au_nhash_test_known_wh(&arg->whlist, name, nlen))
27513 +                       goto out; /* already exists or whiteouted */
27514 +
27515 +               sb = arg->file->f_dentry->d_sb;
27516 +               arg->err = au_ino(sb, arg->bindex, h_ino, d_type, &ino);
27517 +               if (!arg->err) {
27518 +                       if (unlikely(nlen > AUFS_MAX_NAMELEN))
27519 +                               d_type = DT_UNKNOWN;
27520 +                       arg->err = append_de(arg->vdir, name, nlen, ino,
27521 +                                            d_type, &arg->delist);
27522 +               }
27523 +       } else if (au_ftest_fillvdir(arg->flags, WHABLE)) {
27524 +               name += AUFS_WH_PFX_LEN;
27525 +               nlen -= AUFS_WH_PFX_LEN;
27526 +               if (au_nhash_test_known_wh(&arg->whlist, name, nlen))
27527 +                       goto out; /* already whiteouted */
27528 +
27529 +               if (shwh)
27530 +                       arg->err = au_wh_ino(sb, arg->bindex, h_ino, d_type,
27531 +                                            &ino);
27532 +               if (!arg->err) {
27533 +                       if (nlen <= AUFS_MAX_NAMELEN + AUFS_WH_PFX_LEN)
27534 +                               d_type = DT_UNKNOWN;
27535 +                       arg->err = au_nhash_append_wh
27536 +                               (&arg->whlist, name, nlen, ino, d_type,
27537 +                                arg->bindex, shwh);
27538 +               }
27539 +       }
27540 +
27541 +out:
27542 +       if (!arg->err)
27543 +               arg->vdir->vd_jiffy = jiffies;
27544 +       /* smp_mb(); */
27545 +       AuTraceErr(arg->err);
27546 +       return arg->err;
27547 +}
27548 +
27549 +static int au_handle_shwh(struct super_block *sb, struct au_vdir *vdir,
27550 +                         struct au_nhash *whlist, struct au_nhash *delist)
27551 +{
27552 +#ifdef CONFIG_AUFS_SHWH
27553 +       int err;
27554 +       unsigned int nh, u;
27555 +       struct hlist_head *head;
27556 +       struct au_vdir_wh *pos;
27557 +       struct hlist_node *n;
27558 +       char *p, *o;
27559 +       struct au_vdir_destr *destr;
27560 +
27561 +       AuDebugOn(!au_opt_test(au_mntflags(sb), SHWH));
27562 +
27563 +       err = -ENOMEM;
27564 +       o = p = (void *)__get_free_page(GFP_NOFS);
27565 +       if (unlikely(!p))
27566 +               goto out;
27567 +
27568 +       err = 0;
27569 +       nh = whlist->nh_num;
27570 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
27571 +       p += AUFS_WH_PFX_LEN;
27572 +       for (u = 0; u < nh; u++) {
27573 +               head = whlist->nh_head + u;
27574 +               hlist_for_each_entry_safe(pos, n, head, wh_hash) {
27575 +                       destr = &pos->wh_str;
27576 +                       memcpy(p, destr->name, destr->len);
27577 +                       err = append_de(vdir, o, destr->len + AUFS_WH_PFX_LEN,
27578 +                                       pos->wh_ino, pos->wh_type, delist);
27579 +                       if (unlikely(err))
27580 +                               break;
27581 +               }
27582 +       }
27583 +
27584 +       free_page((unsigned long)o);
27585 +
27586 +out:
27587 +       AuTraceErr(err);
27588 +       return err;
27589 +#else
27590 +       return 0;
27591 +#endif
27592 +}
27593 +
27594 +static int au_do_read_vdir(struct fillvdir_arg *arg)
27595 +{
27596 +       int err;
27597 +       unsigned int rdhash;
27598 +       loff_t offset;
27599 +       aufs_bindex_t bend, bindex, bstart;
27600 +       unsigned char shwh;
27601 +       struct file *hf, *file;
27602 +       struct super_block *sb;
27603 +
27604 +       file = arg->file;
27605 +       sb = file->f_dentry->d_sb;
27606 +       SiMustAnyLock(sb);
27607 +
27608 +       rdhash = au_sbi(sb)->si_rdhash;
27609 +       if (!rdhash)
27610 +               rdhash = au_rdhash_est(au_dir_size(file, /*dentry*/NULL));
27611 +       err = au_nhash_alloc(&arg->delist, rdhash, GFP_NOFS);
27612 +       if (unlikely(err))
27613 +               goto out;
27614 +       err = au_nhash_alloc(&arg->whlist, rdhash, GFP_NOFS);
27615 +       if (unlikely(err))
27616 +               goto out_delist;
27617 +
27618 +       err = 0;
27619 +       arg->flags = 0;
27620 +       shwh = 0;
27621 +       if (au_opt_test(au_mntflags(sb), SHWH)) {
27622 +               shwh = 1;
27623 +               au_fset_fillvdir(arg->flags, SHWH);
27624 +       }
27625 +       bstart = au_fbstart(file);
27626 +       bend = au_fbend_dir(file);
27627 +       for (bindex = bstart; !err && bindex <= bend; bindex++) {
27628 +               hf = au_hf_dir(file, bindex);
27629 +               if (!hf)
27630 +                       continue;
27631 +
27632 +               offset = vfsub_llseek(hf, 0, SEEK_SET);
27633 +               err = offset;
27634 +               if (unlikely(offset))
27635 +                       break;
27636 +
27637 +               arg->bindex = bindex;
27638 +               au_fclr_fillvdir(arg->flags, WHABLE);
27639 +               if (shwh
27640 +                   || (bindex != bend
27641 +                       && au_br_whable(au_sbr_perm(sb, bindex))))
27642 +                       au_fset_fillvdir(arg->flags, WHABLE);
27643 +               do {
27644 +                       arg->err = 0;
27645 +                       au_fclr_fillvdir(arg->flags, CALLED);
27646 +                       /* smp_mb(); */
27647 +                       err = vfsub_iterate_dir(hf, &arg->ctx);
27648 +                       if (err >= 0)
27649 +                               err = arg->err;
27650 +               } while (!err && au_ftest_fillvdir(arg->flags, CALLED));
27651 +
27652 +               /*
27653 +                * dir_relax() may be good for concurrency, but aufs should not
27654 +                * use it since it will cause a lockdep problem.
27655 +                */
27656 +       }
27657 +
27658 +       if (!err && shwh)
27659 +               err = au_handle_shwh(sb, arg->vdir, &arg->whlist, &arg->delist);
27660 +
27661 +       au_nhash_wh_free(&arg->whlist);
27662 +
27663 +out_delist:
27664 +       au_nhash_de_free(&arg->delist);
27665 +out:
27666 +       return err;
27667 +}
27668 +
27669 +static int read_vdir(struct file *file, int may_read)
27670 +{
27671 +       int err;
27672 +       unsigned long expire;
27673 +       unsigned char do_read;
27674 +       struct fillvdir_arg arg = {
27675 +               .ctx = {
27676 +                       .actor = au_diractor(fillvdir)
27677 +               }
27678 +       };
27679 +       struct inode *inode;
27680 +       struct au_vdir *vdir, *allocated;
27681 +
27682 +       err = 0;
27683 +       inode = file_inode(file);
27684 +       IMustLock(inode);
27685 +       SiMustAnyLock(inode->i_sb);
27686 +
27687 +       allocated = NULL;
27688 +       do_read = 0;
27689 +       expire = au_sbi(inode->i_sb)->si_rdcache;
27690 +       vdir = au_ivdir(inode);
27691 +       if (!vdir) {
27692 +               do_read = 1;
27693 +               vdir = alloc_vdir(file);
27694 +               err = PTR_ERR(vdir);
27695 +               if (IS_ERR(vdir))
27696 +                       goto out;
27697 +               err = 0;
27698 +               allocated = vdir;
27699 +       } else if (may_read
27700 +                  && (inode->i_version != vdir->vd_version
27701 +                      || time_after(jiffies, vdir->vd_jiffy + expire))) {
27702 +               do_read = 1;
27703 +               err = reinit_vdir(vdir);
27704 +               if (unlikely(err))
27705 +                       goto out;
27706 +       }
27707 +
27708 +       if (!do_read)
27709 +               return 0; /* success */
27710 +
27711 +       arg.file = file;
27712 +       arg.vdir = vdir;
27713 +       err = au_do_read_vdir(&arg);
27714 +       if (!err) {
27715 +               /* file->f_pos = 0; */ /* todo: ctx->pos? */
27716 +               vdir->vd_version = inode->i_version;
27717 +               vdir->vd_last.ul = 0;
27718 +               vdir->vd_last.p.deblk = vdir->vd_deblk[0];
27719 +               if (allocated)
27720 +                       au_set_ivdir(inode, allocated);
27721 +       } else if (allocated)
27722 +               au_vdir_free(allocated);
27723 +
27724 +out:
27725 +       return err;
27726 +}
27727 +
27728 +static int copy_vdir(struct au_vdir *tgt, struct au_vdir *src)
27729 +{
27730 +       int err, rerr;
27731 +       unsigned long ul, n;
27732 +       const unsigned int deblk_sz = src->vd_deblk_sz;
27733 +
27734 +       AuDebugOn(tgt->vd_nblk != 1);
27735 +
27736 +       err = -ENOMEM;
27737 +       if (tgt->vd_nblk < src->vd_nblk) {
27738 +               unsigned char **p;
27739 +
27740 +               p = krealloc(tgt->vd_deblk, sizeof(*p) * src->vd_nblk,
27741 +                            GFP_NOFS);
27742 +               if (unlikely(!p))
27743 +                       goto out;
27744 +               tgt->vd_deblk = p;
27745 +       }
27746 +
27747 +       if (tgt->vd_deblk_sz != deblk_sz) {
27748 +               unsigned char *p;
27749 +
27750 +               tgt->vd_deblk_sz = deblk_sz;
27751 +               p = krealloc(tgt->vd_deblk[0], deblk_sz, GFP_NOFS);
27752 +               if (unlikely(!p))
27753 +                       goto out;
27754 +               tgt->vd_deblk[0] = p;
27755 +       }
27756 +       memcpy(tgt->vd_deblk[0], src->vd_deblk[0], deblk_sz);
27757 +       tgt->vd_version = src->vd_version;
27758 +       tgt->vd_jiffy = src->vd_jiffy;
27759 +
27760 +       n = src->vd_nblk;
27761 +       for (ul = 1; ul < n; ul++) {
27762 +               tgt->vd_deblk[ul] = kmemdup(src->vd_deblk[ul], deblk_sz,
27763 +                                           GFP_NOFS);
27764 +               if (unlikely(!tgt->vd_deblk[ul]))
27765 +                       goto out;
27766 +               tgt->vd_nblk++;
27767 +       }
27768 +       tgt->vd_nblk = n;
27769 +       tgt->vd_last.ul = tgt->vd_last.ul;
27770 +       tgt->vd_last.p.deblk = tgt->vd_deblk[tgt->vd_last.ul];
27771 +       tgt->vd_last.p.deblk += src->vd_last.p.deblk
27772 +               - src->vd_deblk[src->vd_last.ul];
27773 +       /* smp_mb(); */
27774 +       return 0; /* success */
27775 +
27776 +out:
27777 +       rerr = reinit_vdir(tgt);
27778 +       BUG_ON(rerr);
27779 +       return err;
27780 +}
27781 +
27782 +int au_vdir_init(struct file *file)
27783 +{
27784 +       int err;
27785 +       struct inode *inode;
27786 +       struct au_vdir *vdir_cache, *allocated;
27787 +
27788 +       /* test file->f_pos here instead of ctx->pos */
27789 +       err = read_vdir(file, !file->f_pos);
27790 +       if (unlikely(err))
27791 +               goto out;
27792 +
27793 +       allocated = NULL;
27794 +       vdir_cache = au_fvdir_cache(file);
27795 +       if (!vdir_cache) {
27796 +               vdir_cache = alloc_vdir(file);
27797 +               err = PTR_ERR(vdir_cache);
27798 +               if (IS_ERR(vdir_cache))
27799 +                       goto out;
27800 +               allocated = vdir_cache;
27801 +       } else if (!file->f_pos && vdir_cache->vd_version != file->f_version) {
27802 +               /* test file->f_pos here instead of ctx->pos */
27803 +               err = reinit_vdir(vdir_cache);
27804 +               if (unlikely(err))
27805 +                       goto out;
27806 +       } else
27807 +               return 0; /* success */
27808 +
27809 +       inode = file_inode(file);
27810 +       err = copy_vdir(vdir_cache, au_ivdir(inode));
27811 +       if (!err) {
27812 +               file->f_version = inode->i_version;
27813 +               if (allocated)
27814 +                       au_set_fvdir_cache(file, allocated);
27815 +       } else if (allocated)
27816 +               au_vdir_free(allocated);
27817 +
27818 +out:
27819 +       return err;
27820 +}
27821 +
27822 +static loff_t calc_offset(struct au_vdir *vdir)
27823 +{
27824 +       loff_t offset;
27825 +       union au_vdir_deblk_p p;
27826 +
27827 +       p.deblk = vdir->vd_deblk[vdir->vd_last.ul];
27828 +       offset = vdir->vd_last.p.deblk - p.deblk;
27829 +       offset += vdir->vd_deblk_sz * vdir->vd_last.ul;
27830 +       return offset;
27831 +}
27832 +
27833 +/* returns true or false */
27834 +static int seek_vdir(struct file *file, struct dir_context *ctx)
27835 +{
27836 +       int valid;
27837 +       unsigned int deblk_sz;
27838 +       unsigned long ul, n;
27839 +       loff_t offset;
27840 +       union au_vdir_deblk_p p, deblk_end;
27841 +       struct au_vdir *vdir_cache;
27842 +
27843 +       valid = 1;
27844 +       vdir_cache = au_fvdir_cache(file);
27845 +       offset = calc_offset(vdir_cache);
27846 +       AuDbg("offset %lld\n", offset);
27847 +       if (ctx->pos == offset)
27848 +               goto out;
27849 +
27850 +       vdir_cache->vd_last.ul = 0;
27851 +       vdir_cache->vd_last.p.deblk = vdir_cache->vd_deblk[0];
27852 +       if (!ctx->pos)
27853 +               goto out;
27854 +
27855 +       valid = 0;
27856 +       deblk_sz = vdir_cache->vd_deblk_sz;
27857 +       ul = div64_u64(ctx->pos, deblk_sz);
27858 +       AuDbg("ul %lu\n", ul);
27859 +       if (ul >= vdir_cache->vd_nblk)
27860 +               goto out;
27861 +
27862 +       n = vdir_cache->vd_nblk;
27863 +       for (; ul < n; ul++) {
27864 +               p.deblk = vdir_cache->vd_deblk[ul];
27865 +               deblk_end.deblk = p.deblk + deblk_sz;
27866 +               offset = ul;
27867 +               offset *= deblk_sz;
27868 +               while (!is_deblk_end(&p, &deblk_end) && offset < ctx->pos) {
27869 +                       unsigned int l;
27870 +
27871 +                       l = calc_size(p.de->de_str.len);
27872 +                       offset += l;
27873 +                       p.deblk += l;
27874 +               }
27875 +               if (!is_deblk_end(&p, &deblk_end)) {
27876 +                       valid = 1;
27877 +                       vdir_cache->vd_last.ul = ul;
27878 +                       vdir_cache->vd_last.p = p;
27879 +                       break;
27880 +               }
27881 +       }
27882 +
27883 +out:
27884 +       /* smp_mb(); */
27885 +       AuTraceErr(!valid);
27886 +       return valid;
27887 +}
27888 +
27889 +int au_vdir_fill_de(struct file *file, struct dir_context *ctx)
27890 +{
27891 +       unsigned int l, deblk_sz;
27892 +       union au_vdir_deblk_p deblk_end;
27893 +       struct au_vdir *vdir_cache;
27894 +       struct au_vdir_de *de;
27895 +
27896 +       vdir_cache = au_fvdir_cache(file);
27897 +       if (!seek_vdir(file, ctx))
27898 +               return 0;
27899 +
27900 +       deblk_sz = vdir_cache->vd_deblk_sz;
27901 +       while (1) {
27902 +               deblk_end.deblk = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
27903 +               deblk_end.deblk += deblk_sz;
27904 +               while (!is_deblk_end(&vdir_cache->vd_last.p, &deblk_end)) {
27905 +                       de = vdir_cache->vd_last.p.de;
27906 +                       AuDbg("%.*s, off%lld, i%lu, dt%d\n",
27907 +                             de->de_str.len, de->de_str.name, ctx->pos,
27908 +                             (unsigned long)de->de_ino, de->de_type);
27909 +                       if (unlikely(!dir_emit(ctx, de->de_str.name,
27910 +                                              de->de_str.len, de->de_ino,
27911 +                                              de->de_type))) {
27912 +                               /* todo: ignore the error caused by udba? */
27913 +                               /* return err; */
27914 +                               return 0;
27915 +                       }
27916 +
27917 +                       l = calc_size(de->de_str.len);
27918 +                       vdir_cache->vd_last.p.deblk += l;
27919 +                       ctx->pos += l;
27920 +               }
27921 +               if (vdir_cache->vd_last.ul < vdir_cache->vd_nblk - 1) {
27922 +                       vdir_cache->vd_last.ul++;
27923 +                       vdir_cache->vd_last.p.deblk
27924 +                               = vdir_cache->vd_deblk[vdir_cache->vd_last.ul];
27925 +                       ctx->pos = deblk_sz * vdir_cache->vd_last.ul;
27926 +                       continue;
27927 +               }
27928 +               break;
27929 +       }
27930 +
27931 +       /* smp_mb(); */
27932 +       return 0;
27933 +}
27934 diff -urN /usr/share/empty/fs/aufs/vfsub.c linux/fs/aufs/vfsub.c
27935 --- /usr/share/empty/fs/aufs/vfsub.c    1970-01-01 01:00:00.000000000 +0100
27936 +++ linux/fs/aufs/vfsub.c       2014-04-24 22:11:10.861935852 +0200
27937 @@ -0,0 +1,782 @@
27938 +/*
27939 + * Copyright (C) 2005-2014 Junjiro R. Okajima
27940 + *
27941 + * This program, aufs is free software; you can redistribute it and/or modify
27942 + * it under the terms of the GNU General Public License as published by
27943 + * the Free Software Foundation; either version 2 of the License, or
27944 + * (at your option) any later version.
27945 + *
27946 + * This program is distributed in the hope that it will be useful,
27947 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
27948 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
27949 + * GNU General Public License for more details.
27950 + *
27951 + * You should have received a copy of the GNU General Public License
27952 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
27953 + */
27954 +
27955 +/*
27956 + * sub-routines for VFS
27957 + */
27958 +
27959 +#include <linux/ima.h>
27960 +#include <linux/namei.h>
27961 +#include <linux/security.h>
27962 +#include <linux/splice.h>
27963 +#include "aufs.h"
27964 +
27965 +int vfsub_update_h_iattr(struct path *h_path, int *did)
27966 +{
27967 +       int err;
27968 +       struct kstat st;
27969 +       struct super_block *h_sb;
27970 +
27971 +       /* for remote fs, leave work for its getattr or d_revalidate */
27972 +       /* for bad i_attr fs, handle them in aufs_getattr() */
27973 +       /* still some fs may acquire i_mutex. we need to skip them */
27974 +       err = 0;
27975 +       if (!did)
27976 +               did = &err;
27977 +       h_sb = h_path->dentry->d_sb;
27978 +       *did = (!au_test_fs_remote(h_sb) && au_test_fs_refresh_iattr(h_sb));
27979 +       if (*did)
27980 +               err = vfs_getattr(h_path, &st);
27981 +
27982 +       return err;
27983 +}
27984 +
27985 +/* ---------------------------------------------------------------------- */
27986 +
27987 +struct file *vfsub_dentry_open(struct path *path, int flags)
27988 +{
27989 +       struct file *file;
27990 +
27991 +       file = dentry_open(path, flags /* | __FMODE_NONOTIFY */,
27992 +                          current_cred());
27993 +       if (!IS_ERR_OR_NULL(file)
27994 +           && (file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
27995 +               i_readcount_inc(path->dentry->d_inode);
27996 +
27997 +       return file;
27998 +}
27999 +
28000 +struct file *vfsub_filp_open(const char *path, int oflags, int mode)
28001 +{
28002 +       struct file *file;
28003 +
28004 +       lockdep_off();
28005 +       file = filp_open(path,
28006 +                        oflags /* | __FMODE_NONOTIFY */,
28007 +                        mode);
28008 +       lockdep_on();
28009 +       if (IS_ERR(file))
28010 +               goto out;
28011 +       vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
28012 +
28013 +out:
28014 +       return file;
28015 +}
28016 +
28017 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path)
28018 +{
28019 +       int err;
28020 +
28021 +       err = kern_path(name, flags, path);
28022 +       if (!err && path->dentry->d_inode)
28023 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
28024 +       return err;
28025 +}
28026 +
28027 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
28028 +                                   int len)
28029 +{
28030 +       struct path path = {
28031 +               .mnt = NULL
28032 +       };
28033 +
28034 +       /* VFS checks it too, but by WARN_ON_ONCE() */
28035 +       IMustLock(parent->d_inode);
28036 +
28037 +       path.dentry = lookup_one_len(name, parent, len);
28038 +       if (IS_ERR(path.dentry))
28039 +               goto out;
28040 +       if (path.dentry->d_inode)
28041 +               vfsub_update_h_iattr(&path, /*did*/NULL); /*ignore*/
28042 +
28043 +out:
28044 +       AuTraceErrPtr(path.dentry);
28045 +       return path.dentry;
28046 +}
28047 +
28048 +void vfsub_call_lkup_one(void *args)
28049 +{
28050 +       struct vfsub_lkup_one_args *a = args;
28051 +       *a->errp = vfsub_lkup_one(a->name, a->parent);
28052 +}
28053 +
28054 +/* ---------------------------------------------------------------------- */
28055 +
28056 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
28057 +                                struct dentry *d2, struct au_hinode *hdir2)
28058 +{
28059 +       struct dentry *d;
28060 +
28061 +       lockdep_off();
28062 +       d = lock_rename(d1, d2);
28063 +       lockdep_on();
28064 +       au_hn_suspend(hdir1);
28065 +       if (hdir1 != hdir2)
28066 +               au_hn_suspend(hdir2);
28067 +
28068 +       return d;
28069 +}
28070 +
28071 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
28072 +                        struct dentry *d2, struct au_hinode *hdir2)
28073 +{
28074 +       au_hn_resume(hdir1);
28075 +       if (hdir1 != hdir2)
28076 +               au_hn_resume(hdir2);
28077 +       lockdep_off();
28078 +       unlock_rename(d1, d2);
28079 +       lockdep_on();
28080 +}
28081 +
28082 +/* ---------------------------------------------------------------------- */
28083 +
28084 +int vfsub_create(struct inode *dir, struct path *path, int mode, bool want_excl)
28085 +{
28086 +       int err;
28087 +       struct dentry *d;
28088 +
28089 +       IMustLock(dir);
28090 +
28091 +       d = path->dentry;
28092 +       path->dentry = d->d_parent;
28093 +       err = security_path_mknod(path, d, mode, 0);
28094 +       path->dentry = d;
28095 +       if (unlikely(err))
28096 +               goto out;
28097 +
28098 +       err = vfs_create(dir, path->dentry, mode, want_excl);
28099 +       if (!err) {
28100 +               struct path tmp = *path;
28101 +               int did;
28102 +
28103 +               vfsub_update_h_iattr(&tmp, &did);
28104 +               if (did) {
28105 +                       tmp.dentry = path->dentry->d_parent;
28106 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28107 +               }
28108 +               /*ignore*/
28109 +       }
28110 +
28111 +out:
28112 +       return err;
28113 +}
28114 +
28115 +int vfsub_symlink(struct inode *dir, struct path *path, const char *symname)
28116 +{
28117 +       int err;
28118 +       struct dentry *d;
28119 +
28120 +       IMustLock(dir);
28121 +
28122 +       d = path->dentry;
28123 +       path->dentry = d->d_parent;
28124 +       err = security_path_symlink(path, d, symname);
28125 +       path->dentry = d;
28126 +       if (unlikely(err))
28127 +               goto out;
28128 +
28129 +       err = vfs_symlink(dir, path->dentry, symname);
28130 +       if (!err) {
28131 +               struct path tmp = *path;
28132 +               int did;
28133 +
28134 +               vfsub_update_h_iattr(&tmp, &did);
28135 +               if (did) {
28136 +                       tmp.dentry = path->dentry->d_parent;
28137 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28138 +               }
28139 +               /*ignore*/
28140 +       }
28141 +
28142 +out:
28143 +       return err;
28144 +}
28145 +
28146 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev)
28147 +{
28148 +       int err;
28149 +       struct dentry *d;
28150 +
28151 +       IMustLock(dir);
28152 +
28153 +       d = path->dentry;
28154 +       path->dentry = d->d_parent;
28155 +       err = security_path_mknod(path, d, mode, new_encode_dev(dev));
28156 +       path->dentry = d;
28157 +       if (unlikely(err))
28158 +               goto out;
28159 +
28160 +       err = vfs_mknod(dir, path->dentry, mode, dev);
28161 +       if (!err) {
28162 +               struct path tmp = *path;
28163 +               int did;
28164 +
28165 +               vfsub_update_h_iattr(&tmp, &did);
28166 +               if (did) {
28167 +                       tmp.dentry = path->dentry->d_parent;
28168 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28169 +               }
28170 +               /*ignore*/
28171 +       }
28172 +
28173 +out:
28174 +       return err;
28175 +}
28176 +
28177 +static int au_test_nlink(struct inode *inode)
28178 +{
28179 +       const unsigned int link_max = UINT_MAX >> 1; /* rough margin */
28180 +
28181 +       if (!au_test_fs_no_limit_nlink(inode->i_sb)
28182 +           || inode->i_nlink < link_max)
28183 +               return 0;
28184 +       return -EMLINK;
28185 +}
28186 +
28187 +int vfsub_link(struct dentry *src_dentry, struct inode *dir, struct path *path,
28188 +              struct inode **delegated_inode)
28189 +{
28190 +       int err;
28191 +       struct dentry *d;
28192 +
28193 +       IMustLock(dir);
28194 +
28195 +       err = au_test_nlink(src_dentry->d_inode);
28196 +       if (unlikely(err))
28197 +               return err;
28198 +
28199 +       /* we don't call may_linkat() */
28200 +       d = path->dentry;
28201 +       path->dentry = d->d_parent;
28202 +       err = security_path_link(src_dentry, path, d);
28203 +       path->dentry = d;
28204 +       if (unlikely(err))
28205 +               goto out;
28206 +
28207 +       lockdep_off();
28208 +       err = vfs_link(src_dentry, dir, path->dentry, delegated_inode);
28209 +       lockdep_on();
28210 +       if (!err) {
28211 +               struct path tmp = *path;
28212 +               int did;
28213 +
28214 +               /* fuse has different memory inode for the same inumber */
28215 +               vfsub_update_h_iattr(&tmp, &did);
28216 +               if (did) {
28217 +                       tmp.dentry = path->dentry->d_parent;
28218 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28219 +                       tmp.dentry = src_dentry;
28220 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28221 +               }
28222 +               /*ignore*/
28223 +       }
28224 +
28225 +out:
28226 +       return err;
28227 +}
28228 +
28229 +int vfsub_rename(struct inode *src_dir, struct dentry *src_dentry,
28230 +                struct inode *dir, struct path *path,
28231 +                struct inode **delegated_inode)
28232 +{
28233 +       int err;
28234 +       struct path tmp = {
28235 +               .mnt    = path->mnt
28236 +       };
28237 +       struct dentry *d;
28238 +
28239 +       IMustLock(dir);
28240 +       IMustLock(src_dir);
28241 +
28242 +       d = path->dentry;
28243 +       path->dentry = d->d_parent;
28244 +       tmp.dentry = src_dentry->d_parent;
28245 +       err = security_path_rename(&tmp, src_dentry, path, d);
28246 +       path->dentry = d;
28247 +       if (unlikely(err))
28248 +               goto out;
28249 +
28250 +       lockdep_off();
28251 +       err = vfs_rename(src_dir, src_dentry, dir, path->dentry,
28252 +                        delegated_inode);
28253 +       lockdep_on();
28254 +       if (!err) {
28255 +               int did;
28256 +
28257 +               tmp.dentry = d->d_parent;
28258 +               vfsub_update_h_iattr(&tmp, &did);
28259 +               if (did) {
28260 +                       tmp.dentry = src_dentry;
28261 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28262 +                       tmp.dentry = src_dentry->d_parent;
28263 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28264 +               }
28265 +               /*ignore*/
28266 +       }
28267 +
28268 +out:
28269 +       return err;
28270 +}
28271 +
28272 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode)
28273 +{
28274 +       int err;
28275 +       struct dentry *d;
28276 +
28277 +       IMustLock(dir);
28278 +
28279 +       d = path->dentry;
28280 +       path->dentry = d->d_parent;
28281 +       err = security_path_mkdir(path, d, mode);
28282 +       path->dentry = d;
28283 +       if (unlikely(err))
28284 +               goto out;
28285 +
28286 +       err = vfs_mkdir(dir, path->dentry, mode);
28287 +       if (!err) {
28288 +               struct path tmp = *path;
28289 +               int did;
28290 +
28291 +               vfsub_update_h_iattr(&tmp, &did);
28292 +               if (did) {
28293 +                       tmp.dentry = path->dentry->d_parent;
28294 +                       vfsub_update_h_iattr(&tmp, /*did*/NULL);
28295 +               }
28296 +               /*ignore*/
28297 +       }
28298 +
28299 +out:
28300 +       return err;
28301 +}
28302 +
28303 +int vfsub_rmdir(struct inode *dir, struct path *path)
28304 +{
28305 +       int err;
28306 +       struct dentry *d;
28307 +
28308 +       IMustLock(dir);
28309 +
28310 +       d = path->dentry;
28311 +       path->dentry = d->d_parent;
28312 +       err = security_path_rmdir(path, d);
28313 +       path->dentry = d;
28314 +       if (unlikely(err))
28315 +               goto out;
28316 +
28317 +       lockdep_off();
28318 +       err = vfs_rmdir(dir, path->dentry);
28319 +       lockdep_on();
28320 +       if (!err) {
28321 +               struct path tmp = {
28322 +                       .dentry = path->dentry->d_parent,
28323 +                       .mnt    = path->mnt
28324 +               };
28325 +
28326 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
28327 +       }
28328 +
28329 +out:
28330 +       return err;
28331 +}
28332 +
28333 +/* ---------------------------------------------------------------------- */
28334 +
28335 +/* todo: support mmap_sem? */
28336 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
28337 +                    loff_t *ppos)
28338 +{
28339 +       ssize_t err;
28340 +
28341 +       lockdep_off();
28342 +       err = vfs_read(file, ubuf, count, ppos);
28343 +       lockdep_on();
28344 +       if (err >= 0)
28345 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
28346 +       return err;
28347 +}
28348 +
28349 +/* todo: kernel_read()? */
28350 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
28351 +                    loff_t *ppos)
28352 +{
28353 +       ssize_t err;
28354 +       mm_segment_t oldfs;
28355 +       union {
28356 +               void *k;
28357 +               char __user *u;
28358 +       } buf;
28359 +
28360 +       buf.k = kbuf;
28361 +       oldfs = get_fs();
28362 +       set_fs(KERNEL_DS);
28363 +       err = vfsub_read_u(file, buf.u, count, ppos);
28364 +       set_fs(oldfs);
28365 +       return err;
28366 +}
28367 +
28368 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
28369 +                     loff_t *ppos)
28370 +{
28371 +       ssize_t err;
28372 +
28373 +       lockdep_off();
28374 +       err = vfs_write(file, ubuf, count, ppos);
28375 +       lockdep_on();
28376 +       if (err >= 0)
28377 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
28378 +       return err;
28379 +}
28380 +
28381 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count, loff_t *ppos)
28382 +{
28383 +       ssize_t err;
28384 +       mm_segment_t oldfs;
28385 +       union {
28386 +               void *k;
28387 +               const char __user *u;
28388 +       } buf;
28389 +
28390 +       buf.k = kbuf;
28391 +       oldfs = get_fs();
28392 +       set_fs(KERNEL_DS);
28393 +       err = vfsub_write_u(file, buf.u, count, ppos);
28394 +       set_fs(oldfs);
28395 +       return err;
28396 +}
28397 +
28398 +int vfsub_flush(struct file *file, fl_owner_t id)
28399 +{
28400 +       int err;
28401 +
28402 +       err = 0;
28403 +       if (file->f_op->flush) {
28404 +               if (!au_test_nfs(file->f_dentry->d_sb))
28405 +                       err = file->f_op->flush(file, id);
28406 +               else {
28407 +                       lockdep_off();
28408 +                       err = file->f_op->flush(file, id);
28409 +                       lockdep_on();
28410 +               }
28411 +               if (!err)
28412 +                       vfsub_update_h_iattr(&file->f_path, /*did*/NULL);
28413 +               /*ignore*/
28414 +       }
28415 +       return err;
28416 +}
28417 +
28418 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx)
28419 +{
28420 +       int err;
28421 +
28422 +       AuDbg("%pD, ctx{%pf, %llu}\n", file, ctx->actor, ctx->pos);
28423 +
28424 +       lockdep_off();
28425 +       err = iterate_dir(file, ctx);
28426 +       lockdep_on();
28427 +       if (err >= 0)
28428 +               vfsub_update_h_iattr(&file->f_path, /*did*/NULL); /*ignore*/
28429 +       return err;
28430 +}
28431 +
28432 +long vfsub_splice_to(struct file *in, loff_t *ppos,
28433 +                    struct pipe_inode_info *pipe, size_t len,
28434 +                    unsigned int flags)
28435 +{
28436 +       long err;
28437 +
28438 +       lockdep_off();
28439 +       err = do_splice_to(in, ppos, pipe, len, flags);
28440 +       lockdep_on();
28441 +       file_accessed(in);
28442 +       if (err >= 0)
28443 +               vfsub_update_h_iattr(&in->f_path, /*did*/NULL); /*ignore*/
28444 +       return err;
28445 +}
28446 +
28447 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
28448 +                      loff_t *ppos, size_t len, unsigned int flags)
28449 +{
28450 +       long err;
28451 +
28452 +       lockdep_off();
28453 +       err = do_splice_from(pipe, out, ppos, len, flags);
28454 +       lockdep_on();
28455 +       if (err >= 0)
28456 +               vfsub_update_h_iattr(&out->f_path, /*did*/NULL); /*ignore*/
28457 +       return err;
28458 +}
28459 +
28460 +int vfsub_fsync(struct file *file, struct path *path, int datasync)
28461 +{
28462 +       int err;
28463 +
28464 +       /* file can be NULL */
28465 +       lockdep_off();
28466 +       err = vfs_fsync(file, datasync);
28467 +       lockdep_on();
28468 +       if (!err) {
28469 +               if (!path) {
28470 +                       AuDebugOn(!file);
28471 +                       path = &file->f_path;
28472 +               }
28473 +               vfsub_update_h_iattr(path, /*did*/NULL); /*ignore*/
28474 +       }
28475 +       return err;
28476 +}
28477 +
28478 +/* cf. open.c:do_sys_truncate() and do_sys_ftruncate() */
28479 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
28480 +               struct file *h_file)
28481 +{
28482 +       int err;
28483 +       struct inode *h_inode;
28484 +       struct super_block *h_sb;
28485 +
28486 +       if (!h_file) {
28487 +               err = vfsub_truncate(h_path, length);
28488 +               goto out;
28489 +       }
28490 +
28491 +       h_inode = h_path->dentry->d_inode;
28492 +       h_sb = h_inode->i_sb;
28493 +       lockdep_off();
28494 +       sb_start_write(h_sb);
28495 +       lockdep_on();
28496 +       err = locks_verify_truncate(h_inode, h_file, length);
28497 +       if (!err)
28498 +               err = security_path_truncate(h_path);
28499 +       if (!err) {
28500 +               lockdep_off();
28501 +               err = do_truncate(h_path->dentry, length, attr, h_file);
28502 +               lockdep_on();
28503 +       }
28504 +       lockdep_off();
28505 +       sb_end_write(h_sb);
28506 +       lockdep_on();
28507 +
28508 +out:
28509 +       return err;
28510 +}
28511 +
28512 +/* ---------------------------------------------------------------------- */
28513 +
28514 +struct au_vfsub_mkdir_args {
28515 +       int *errp;
28516 +       struct inode *dir;
28517 +       struct path *path;
28518 +       int mode;
28519 +};
28520 +
28521 +static void au_call_vfsub_mkdir(void *args)
28522 +{
28523 +       struct au_vfsub_mkdir_args *a = args;
28524 +       *a->errp = vfsub_mkdir(a->dir, a->path, a->mode);
28525 +}
28526 +
28527 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode)
28528 +{
28529 +       int err, do_sio, wkq_err;
28530 +
28531 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
28532 +       if (!do_sio)
28533 +               err = vfsub_mkdir(dir, path, mode);
28534 +       else {
28535 +               struct au_vfsub_mkdir_args args = {
28536 +                       .errp   = &err,
28537 +                       .dir    = dir,
28538 +                       .path   = path,
28539 +                       .mode   = mode
28540 +               };
28541 +               wkq_err = au_wkq_wait(au_call_vfsub_mkdir, &args);
28542 +               if (unlikely(wkq_err))
28543 +                       err = wkq_err;
28544 +       }
28545 +
28546 +       return err;
28547 +}
28548 +
28549 +struct au_vfsub_rmdir_args {
28550 +       int *errp;
28551 +       struct inode *dir;
28552 +       struct path *path;
28553 +};
28554 +
28555 +static void au_call_vfsub_rmdir(void *args)
28556 +{
28557 +       struct au_vfsub_rmdir_args *a = args;
28558 +       *a->errp = vfsub_rmdir(a->dir, a->path);
28559 +}
28560 +
28561 +int vfsub_sio_rmdir(struct inode *dir, struct path *path)
28562 +{
28563 +       int err, do_sio, wkq_err;
28564 +
28565 +       do_sio = au_test_h_perm_sio(dir, MAY_EXEC | MAY_WRITE);
28566 +       if (!do_sio)
28567 +               err = vfsub_rmdir(dir, path);
28568 +       else {
28569 +               struct au_vfsub_rmdir_args args = {
28570 +                       .errp   = &err,
28571 +                       .dir    = dir,
28572 +                       .path   = path
28573 +               };
28574 +               wkq_err = au_wkq_wait(au_call_vfsub_rmdir, &args);
28575 +               if (unlikely(wkq_err))
28576 +                       err = wkq_err;
28577 +       }
28578 +
28579 +       return err;
28580 +}
28581 +
28582 +/* ---------------------------------------------------------------------- */
28583 +
28584 +struct notify_change_args {
28585 +       int *errp;
28586 +       struct path *path;
28587 +       struct iattr *ia;
28588 +       struct inode **delegated_inode;
28589 +};
28590 +
28591 +static void call_notify_change(void *args)
28592 +{
28593 +       struct notify_change_args *a = args;
28594 +       struct inode *h_inode;
28595 +
28596 +       h_inode = a->path->dentry->d_inode;
28597 +       IMustLock(h_inode);
28598 +
28599 +       *a->errp = -EPERM;
28600 +       if (!IS_IMMUTABLE(h_inode) && !IS_APPEND(h_inode)) {
28601 +               *a->errp = notify_change(a->path->dentry, a->ia,
28602 +                                        a->delegated_inode);
28603 +               if (!*a->errp)
28604 +                       vfsub_update_h_iattr(a->path, /*did*/NULL); /*ignore*/
28605 +       }
28606 +       AuTraceErr(*a->errp);
28607 +}
28608 +
28609 +int vfsub_notify_change(struct path *path, struct iattr *ia,
28610 +                       struct inode **delegated_inode)
28611 +{
28612 +       int err;
28613 +       struct notify_change_args args = {
28614 +               .errp                   = &err,
28615 +               .path                   = path,
28616 +               .ia                     = ia,
28617 +               .delegated_inode        = delegated_inode
28618 +       };
28619 +
28620 +       call_notify_change(&args);
28621 +
28622 +       return err;
28623 +}
28624 +
28625 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
28626 +                           struct inode **delegated_inode)
28627 +{
28628 +       int err, wkq_err;
28629 +       struct notify_change_args args = {
28630 +               .errp                   = &err,
28631 +               .path                   = path,
28632 +               .ia                     = ia,
28633 +               .delegated_inode        = delegated_inode
28634 +       };
28635 +
28636 +       wkq_err = au_wkq_wait(call_notify_change, &args);
28637 +       if (unlikely(wkq_err))
28638 +               err = wkq_err;
28639 +
28640 +       return err;
28641 +}
28642 +
28643 +/* ---------------------------------------------------------------------- */
28644 +
28645 +struct unlink_args {
28646 +       int *errp;
28647 +       struct inode *dir;
28648 +       struct path *path;
28649 +       struct inode **delegated_inode;
28650 +};
28651 +
28652 +static void call_unlink(void *args)
28653 +{
28654 +       struct unlink_args *a = args;
28655 +       struct dentry *d = a->path->dentry;
28656 +       struct inode *h_inode;
28657 +       const int stop_sillyrename = (au_test_nfs(d->d_sb)
28658 +                                     && d_count(d) == 1);
28659 +
28660 +       IMustLock(a->dir);
28661 +
28662 +       a->path->dentry = d->d_parent;
28663 +       *a->errp = security_path_unlink(a->path, d);
28664 +       a->path->dentry = d;
28665 +       if (unlikely(*a->errp))
28666 +               return;
28667 +
28668 +       if (!stop_sillyrename)
28669 +               dget(d);
28670 +       h_inode = d->d_inode;
28671 +       if (h_inode)
28672 +               ihold(h_inode);
28673 +
28674 +       lockdep_off();
28675 +       *a->errp = vfs_unlink(a->dir, d, a->delegated_inode);
28676 +       lockdep_on();
28677 +       if (!*a->errp) {
28678 +               struct path tmp = {
28679 +                       .dentry = d->d_parent,
28680 +                       .mnt    = a->path->mnt
28681 +               };
28682 +               vfsub_update_h_iattr(&tmp, /*did*/NULL); /*ignore*/
28683 +       }
28684 +
28685 +       if (!stop_sillyrename)
28686 +               dput(d);
28687 +       if (h_inode)
28688 +               iput(h_inode);
28689 +
28690 +       AuTraceErr(*a->errp);
28691 +}
28692 +
28693 +/*
28694 + * @dir: must be locked.
28695 + * @dentry: target dentry.
28696 + */
28697 +int vfsub_unlink(struct inode *dir, struct path *path,
28698 +                struct inode **delegated_inode, int force)
28699 +{
28700 +       int err;
28701 +       struct unlink_args args = {
28702 +               .errp                   = &err,
28703 +               .dir                    = dir,
28704 +               .path                   = path,
28705 +               .delegated_inode        = delegated_inode
28706 +       };
28707 +
28708 +       if (!force)
28709 +               call_unlink(&args);
28710 +       else {
28711 +               int wkq_err;
28712 +
28713 +               wkq_err = au_wkq_wait(call_unlink, &args);
28714 +               if (unlikely(wkq_err))
28715 +                       err = wkq_err;
28716 +       }
28717 +
28718 +       return err;
28719 +}
28720 diff -urN /usr/share/empty/fs/aufs/vfsub.h linux/fs/aufs/vfsub.h
28721 --- /usr/share/empty/fs/aufs/vfsub.h    1970-01-01 01:00:00.000000000 +0100
28722 +++ linux/fs/aufs/vfsub.h       2014-04-24 22:11:10.861935852 +0200
28723 @@ -0,0 +1,282 @@
28724 +/*
28725 + * Copyright (C) 2005-2014 Junjiro R. Okajima
28726 + *
28727 + * This program, aufs is free software; you can redistribute it and/or modify
28728 + * it under the terms of the GNU General Public License as published by
28729 + * the Free Software Foundation; either version 2 of the License, or
28730 + * (at your option) any later version.
28731 + *
28732 + * This program is distributed in the hope that it will be useful,
28733 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
28734 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28735 + * GNU General Public License for more details.
28736 + *
28737 + * You should have received a copy of the GNU General Public License
28738 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
28739 + */
28740 +
28741 +/*
28742 + * sub-routines for VFS
28743 + */
28744 +
28745 +#ifndef __AUFS_VFSUB_H__
28746 +#define __AUFS_VFSUB_H__
28747 +
28748 +#ifdef __KERNEL__
28749 +
28750 +#include <linux/fs.h>
28751 +#include <linux/lglock.h>
28752 +#include <linux/mount.h>
28753 +#include "debug.h"
28754 +
28755 +/* copied from linux/fs/internal.h */
28756 +/* todo: BAD approach!! */
28757 +extern void __mnt_drop_write(struct vfsmount *);
28758 +extern spinlock_t inode_sb_list_lock;
28759 +
28760 +/* ---------------------------------------------------------------------- */
28761 +
28762 +/* lock subclass for lower inode */
28763 +/* default MAX_LOCKDEP_SUBCLASSES(8) is not enough */
28764 +/* reduce? gave up. */
28765 +enum {
28766 +       AuLsc_I_Begin = I_MUTEX_NONDIR2, /* 4 */
28767 +       AuLsc_I_PARENT,         /* lower inode, parent first */
28768 +       AuLsc_I_PARENT2,        /* copyup dirs */
28769 +       AuLsc_I_PARENT3,        /* copyup wh */
28770 +       AuLsc_I_CHILD,
28771 +       AuLsc_I_CHILD2,
28772 +       AuLsc_I_End
28773 +};
28774 +
28775 +/* to debug easier, do not make them inlined functions */
28776 +#define MtxMustLock(mtx)       AuDebugOn(!mutex_is_locked(mtx))
28777 +#define IMustLock(i)           MtxMustLock(&(i)->i_mutex)
28778 +
28779 +/* ---------------------------------------------------------------------- */
28780 +
28781 +static inline void vfsub_drop_nlink(struct inode *inode)
28782 +{
28783 +       AuDebugOn(!inode->i_nlink);
28784 +       drop_nlink(inode);
28785 +}
28786 +
28787 +static inline void vfsub_dead_dir(struct inode *inode)
28788 +{
28789 +       AuDebugOn(!S_ISDIR(inode->i_mode));
28790 +       inode->i_flags |= S_DEAD;
28791 +       clear_nlink(inode);
28792 +}
28793 +
28794 +static inline int vfsub_native_ro(struct inode *inode)
28795 +{
28796 +       return (inode->i_sb->s_flags & MS_RDONLY)
28797 +               || IS_RDONLY(inode)
28798 +               /* || IS_APPEND(inode) */
28799 +               || IS_IMMUTABLE(inode);
28800 +}
28801 +
28802 +/* ---------------------------------------------------------------------- */
28803 +
28804 +int vfsub_update_h_iattr(struct path *h_path, int *did);
28805 +struct file *vfsub_dentry_open(struct path *path, int flags);
28806 +struct file *vfsub_filp_open(const char *path, int oflags, int mode);
28807 +int vfsub_kern_path(const char *name, unsigned int flags, struct path *path);
28808 +
28809 +struct dentry *vfsub_lookup_one_len(const char *name, struct dentry *parent,
28810 +                                   int len);
28811 +
28812 +struct vfsub_lkup_one_args {
28813 +       struct dentry **errp;
28814 +       struct qstr *name;
28815 +       struct dentry *parent;
28816 +};
28817 +
28818 +static inline struct dentry *vfsub_lkup_one(struct qstr *name,
28819 +                                           struct dentry *parent)
28820 +{
28821 +       return vfsub_lookup_one_len(name->name, parent, name->len);
28822 +}
28823 +
28824 +void vfsub_call_lkup_one(void *args);
28825 +
28826 +/* ---------------------------------------------------------------------- */
28827 +
28828 +static inline int vfsub_mnt_want_write(struct vfsmount *mnt)
28829 +{
28830 +       int err;
28831 +       lockdep_off();
28832 +       err = mnt_want_write(mnt);
28833 +       lockdep_on();
28834 +       return err;
28835 +}
28836 +
28837 +static inline void vfsub_mnt_drop_write(struct vfsmount *mnt)
28838 +{
28839 +       lockdep_off();
28840 +       mnt_drop_write(mnt);
28841 +       lockdep_on();
28842 +}
28843 +
28844 +static inline void vfsub_mnt_drop_write_file(struct file *file)
28845 +{
28846 +       lockdep_off();
28847 +       mnt_drop_write_file(file);
28848 +       lockdep_on();
28849 +}
28850 +
28851 +/* ---------------------------------------------------------------------- */
28852 +
28853 +struct au_hinode;
28854 +struct dentry *vfsub_lock_rename(struct dentry *d1, struct au_hinode *hdir1,
28855 +                                struct dentry *d2, struct au_hinode *hdir2);
28856 +void vfsub_unlock_rename(struct dentry *d1, struct au_hinode *hdir1,
28857 +                        struct dentry *d2, struct au_hinode *hdir2);
28858 +
28859 +int vfsub_create(struct inode *dir, struct path *path, int mode,
28860 +                bool want_excl);
28861 +int vfsub_symlink(struct inode *dir, struct path *path,
28862 +                 const char *symname);
28863 +int vfsub_mknod(struct inode *dir, struct path *path, int mode, dev_t dev);
28864 +int vfsub_link(struct dentry *src_dentry, struct inode *dir,
28865 +              struct path *path, struct inode **delegated_inode);
28866 +int vfsub_rename(struct inode *src_hdir, struct dentry *src_dentry,
28867 +                struct inode *hdir, struct path *path,
28868 +                struct inode **delegated_inode);
28869 +int vfsub_mkdir(struct inode *dir, struct path *path, int mode);
28870 +int vfsub_rmdir(struct inode *dir, struct path *path);
28871 +
28872 +/* ---------------------------------------------------------------------- */
28873 +
28874 +ssize_t vfsub_read_u(struct file *file, char __user *ubuf, size_t count,
28875 +                    loff_t *ppos);
28876 +ssize_t vfsub_read_k(struct file *file, void *kbuf, size_t count,
28877 +                       loff_t *ppos);
28878 +ssize_t vfsub_write_u(struct file *file, const char __user *ubuf, size_t count,
28879 +                     loff_t *ppos);
28880 +ssize_t vfsub_write_k(struct file *file, void *kbuf, size_t count,
28881 +                     loff_t *ppos);
28882 +int vfsub_flush(struct file *file, fl_owner_t id);
28883 +int vfsub_iterate_dir(struct file *file, struct dir_context *ctx);
28884 +
28885 +/* just for type-check */
28886 +static inline filldir_t au_diractor(int (*func)(struct dir_context *,
28887 +                                               const char *, int, loff_t, u64,
28888 +                                               unsigned))
28889 +{
28890 +       return (filldir_t)func;
28891 +}
28892 +
28893 +
28894 +static inline loff_t vfsub_f_size_read(struct file *file)
28895 +{
28896 +       return i_size_read(file_inode(file));
28897 +}
28898 +
28899 +static inline unsigned int vfsub_file_flags(struct file *file)
28900 +{
28901 +       unsigned int flags;
28902 +
28903 +       spin_lock(&file->f_lock);
28904 +       flags = file->f_flags;
28905 +       spin_unlock(&file->f_lock);
28906 +
28907 +       return flags;
28908 +}
28909 +
28910 +static inline void vfsub_file_accessed(struct file *h_file)
28911 +{
28912 +       file_accessed(h_file);
28913 +       vfsub_update_h_iattr(&h_file->f_path, /*did*/NULL); /*ignore*/
28914 +}
28915 +
28916 +static inline void vfsub_touch_atime(struct vfsmount *h_mnt,
28917 +                                    struct dentry *h_dentry)
28918 +{
28919 +       struct path h_path = {
28920 +               .dentry = h_dentry,
28921 +               .mnt    = h_mnt
28922 +       };
28923 +       touch_atime(&h_path);
28924 +       vfsub_update_h_iattr(&h_path, /*did*/NULL); /*ignore*/
28925 +}
28926 +
28927 +static inline int vfsub_update_time(struct inode *h_inode, struct timespec *ts,
28928 +                                   int flags)
28929 +{
28930 +       return update_time(h_inode, ts, flags);
28931 +       /* no vfsub_update_h_iattr() since we don't have struct path */
28932 +}
28933 +
28934 +long vfsub_splice_to(struct file *in, loff_t *ppos,
28935 +                    struct pipe_inode_info *pipe, size_t len,
28936 +                    unsigned int flags);
28937 +long vfsub_splice_from(struct pipe_inode_info *pipe, struct file *out,
28938 +                      loff_t *ppos, size_t len, unsigned int flags);
28939 +
28940 +static inline long vfsub_truncate(struct path *path, loff_t length)
28941 +{
28942 +       long err;
28943 +       lockdep_off();
28944 +       err = vfs_truncate(path, length);
28945 +       lockdep_on();
28946 +       return err;
28947 +}
28948 +
28949 +int vfsub_trunc(struct path *h_path, loff_t length, unsigned int attr,
28950 +               struct file *h_file);
28951 +int vfsub_fsync(struct file *file, struct path *path, int datasync);
28952 +
28953 +/* ---------------------------------------------------------------------- */
28954 +
28955 +static inline loff_t vfsub_llseek(struct file *file, loff_t offset, int origin)
28956 +{
28957 +       loff_t err;
28958 +
28959 +       lockdep_off();
28960 +       err = vfs_llseek(file, offset, origin);
28961 +       lockdep_on();
28962 +       return err;
28963 +}
28964 +
28965 +/* ---------------------------------------------------------------------- */
28966 +
28967 +/* dirty workaround for strict type of fmode_t */
28968 +union vfsub_fmu {
28969 +       fmode_t fm;
28970 +       unsigned int ui;
28971 +};
28972 +
28973 +static inline unsigned int vfsub_fmode_to_uint(fmode_t fm)
28974 +{
28975 +       union vfsub_fmu u = {
28976 +               .fm = fm
28977 +       };
28978 +
28979 +       BUILD_BUG_ON(sizeof(u.fm) != sizeof(u.ui));
28980 +
28981 +       return u.ui;
28982 +}
28983 +
28984 +static inline fmode_t vfsub_uint_to_fmode(unsigned int ui)
28985 +{
28986 +       union vfsub_fmu u = {
28987 +               .ui = ui
28988 +       };
28989 +
28990 +       return u.fm;
28991 +}
28992 +
28993 +/* ---------------------------------------------------------------------- */
28994 +
28995 +int vfsub_sio_mkdir(struct inode *dir, struct path *path, int mode);
28996 +int vfsub_sio_rmdir(struct inode *dir, struct path *path);
28997 +int vfsub_sio_notify_change(struct path *path, struct iattr *ia,
28998 +                           struct inode **delegated_inode);
28999 +int vfsub_notify_change(struct path *path, struct iattr *ia,
29000 +                       struct inode **delegated_inode);
29001 +int vfsub_unlink(struct inode *dir, struct path *path,
29002 +                struct inode **delegated_inode, int force);
29003 +
29004 +#endif /* __KERNEL__ */
29005 +#endif /* __AUFS_VFSUB_H__ */
29006 diff -urN /usr/share/empty/fs/aufs/wbr_policy.c linux/fs/aufs/wbr_policy.c
29007 --- /usr/share/empty/fs/aufs/wbr_policy.c       1970-01-01 01:00:00.000000000 +0100
29008 +++ linux/fs/aufs/wbr_policy.c  2014-04-24 22:11:10.861935852 +0200
29009 @@ -0,0 +1,756 @@
29010 +/*
29011 + * Copyright (C) 2005-2014 Junjiro R. Okajima
29012 + *
29013 + * This program, aufs is free software; you can redistribute it and/or modify
29014 + * it under the terms of the GNU General Public License as published by
29015 + * the Free Software Foundation; either version 2 of the License, or
29016 + * (at your option) any later version.
29017 + *
29018 + * This program is distributed in the hope that it will be useful,
29019 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29020 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29021 + * GNU General Public License for more details.
29022 + *
29023 + * You should have received a copy of the GNU General Public License
29024 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29025 + */
29026 +
29027 +/*
29028 + * policies for selecting one among multiple writable branches
29029 + */
29030 +
29031 +#include <linux/statfs.h>
29032 +#include "aufs.h"
29033 +
29034 +/* subset of cpup_attr() */
29035 +static noinline_for_stack
29036 +int au_cpdown_attr(struct path *h_path, struct dentry *h_src)
29037 +{
29038 +       int err, sbits;
29039 +       struct iattr ia;
29040 +       struct inode *h_isrc;
29041 +
29042 +       h_isrc = h_src->d_inode;
29043 +       ia.ia_valid = ATTR_FORCE | ATTR_MODE | ATTR_UID | ATTR_GID;
29044 +       ia.ia_mode = h_isrc->i_mode;
29045 +       ia.ia_uid = h_isrc->i_uid;
29046 +       ia.ia_gid = h_isrc->i_gid;
29047 +       sbits = !!(ia.ia_mode & (S_ISUID | S_ISGID));
29048 +       au_cpup_attr_flags(h_path->dentry->d_inode, h_isrc->i_flags);
29049 +       /* no delegation since it is just created */
29050 +       err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
29051 +
29052 +       /* is this nfs only? */
29053 +       if (!err && sbits && au_test_nfs(h_path->dentry->d_sb)) {
29054 +               ia.ia_valid = ATTR_FORCE | ATTR_MODE;
29055 +               ia.ia_mode = h_isrc->i_mode;
29056 +               err = vfsub_sio_notify_change(h_path, &ia, /*delegated*/NULL);
29057 +       }
29058 +
29059 +       return err;
29060 +}
29061 +
29062 +#define AuCpdown_PARENT_OPQ    1
29063 +#define AuCpdown_WHED          (1 << 1)
29064 +#define AuCpdown_MADE_DIR      (1 << 2)
29065 +#define AuCpdown_DIROPQ                (1 << 3)
29066 +#define au_ftest_cpdown(flags, name)   ((flags) & AuCpdown_##name)
29067 +#define au_fset_cpdown(flags, name) \
29068 +       do { (flags) |= AuCpdown_##name; } while (0)
29069 +#define au_fclr_cpdown(flags, name) \
29070 +       do { (flags) &= ~AuCpdown_##name; } while (0)
29071 +
29072 +static int au_cpdown_dir_opq(struct dentry *dentry, aufs_bindex_t bdst,
29073 +                            unsigned int *flags)
29074 +{
29075 +       int err;
29076 +       struct dentry *opq_dentry;
29077 +
29078 +       opq_dentry = au_diropq_create(dentry, bdst);
29079 +       err = PTR_ERR(opq_dentry);
29080 +       if (IS_ERR(opq_dentry))
29081 +               goto out;
29082 +       dput(opq_dentry);
29083 +       au_fset_cpdown(*flags, DIROPQ);
29084 +
29085 +out:
29086 +       return err;
29087 +}
29088 +
29089 +static int au_cpdown_dir_wh(struct dentry *dentry, struct dentry *h_parent,
29090 +                           struct inode *dir, aufs_bindex_t bdst)
29091 +{
29092 +       int err;
29093 +       struct path h_path;
29094 +       struct au_branch *br;
29095 +
29096 +       br = au_sbr(dentry->d_sb, bdst);
29097 +       h_path.dentry = au_wh_lkup(h_parent, &dentry->d_name, br);
29098 +       err = PTR_ERR(h_path.dentry);
29099 +       if (IS_ERR(h_path.dentry))
29100 +               goto out;
29101 +
29102 +       err = 0;
29103 +       if (h_path.dentry->d_inode) {
29104 +               h_path.mnt = au_br_mnt(br);
29105 +               err = au_wh_unlink_dentry(au_h_iptr(dir, bdst), &h_path,
29106 +                                         dentry);
29107 +       }
29108 +       dput(h_path.dentry);
29109 +
29110 +out:
29111 +       return err;
29112 +}
29113 +
29114 +static int au_cpdown_dir(struct dentry *dentry, aufs_bindex_t bdst,
29115 +                        struct au_pin *pin,
29116 +                        struct dentry *h_parent, void *arg)
29117 +{
29118 +       int err, rerr;
29119 +       aufs_bindex_t bopq, bstart;
29120 +       struct path h_path;
29121 +       struct dentry *parent;
29122 +       struct inode *h_dir, *h_inode, *inode, *dir;
29123 +       unsigned int *flags = arg;
29124 +
29125 +       bstart = au_dbstart(dentry);
29126 +       /* dentry is di-locked */
29127 +       parent = dget_parent(dentry);
29128 +       dir = parent->d_inode;
29129 +       h_dir = h_parent->d_inode;
29130 +       AuDebugOn(h_dir != au_h_iptr(dir, bdst));
29131 +       IMustLock(h_dir);
29132 +
29133 +       err = au_lkup_neg(dentry, bdst, /*wh*/0);
29134 +       if (unlikely(err < 0))
29135 +               goto out;
29136 +       h_path.dentry = au_h_dptr(dentry, bdst);
29137 +       h_path.mnt = au_sbr_mnt(dentry->d_sb, bdst);
29138 +       err = vfsub_sio_mkdir(au_h_iptr(dir, bdst), &h_path,
29139 +                             S_IRWXU | S_IRUGO | S_IXUGO);
29140 +       if (unlikely(err))
29141 +               goto out_put;
29142 +       au_fset_cpdown(*flags, MADE_DIR);
29143 +
29144 +       bopq = au_dbdiropq(dentry);
29145 +       au_fclr_cpdown(*flags, WHED);
29146 +       au_fclr_cpdown(*flags, DIROPQ);
29147 +       if (au_dbwh(dentry) == bdst)
29148 +               au_fset_cpdown(*flags, WHED);
29149 +       if (!au_ftest_cpdown(*flags, PARENT_OPQ) && bopq <= bdst)
29150 +               au_fset_cpdown(*flags, PARENT_OPQ);
29151 +       h_inode = h_path.dentry->d_inode;
29152 +       mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
29153 +       if (au_ftest_cpdown(*flags, WHED)) {
29154 +               err = au_cpdown_dir_opq(dentry, bdst, flags);
29155 +               if (unlikely(err)) {
29156 +                       mutex_unlock(&h_inode->i_mutex);
29157 +                       goto out_dir;
29158 +               }
29159 +       }
29160 +
29161 +       err = au_cpdown_attr(&h_path, au_h_dptr(dentry, bstart));
29162 +       mutex_unlock(&h_inode->i_mutex);
29163 +       if (unlikely(err))
29164 +               goto out_opq;
29165 +
29166 +       if (au_ftest_cpdown(*flags, WHED)) {
29167 +               err = au_cpdown_dir_wh(dentry, h_parent, dir, bdst);
29168 +               if (unlikely(err))
29169 +                       goto out_opq;
29170 +       }
29171 +
29172 +       inode = dentry->d_inode;
29173 +       if (au_ibend(inode) < bdst)
29174 +               au_set_ibend(inode, bdst);
29175 +       au_set_h_iptr(inode, bdst, au_igrab(h_inode),
29176 +                     au_hi_flags(inode, /*isdir*/1));
29177 +       goto out; /* success */
29178 +
29179 +       /* revert */
29180 +out_opq:
29181 +       if (au_ftest_cpdown(*flags, DIROPQ)) {
29182 +               mutex_lock_nested(&h_inode->i_mutex, AuLsc_I_CHILD);
29183 +               rerr = au_diropq_remove(dentry, bdst);
29184 +               mutex_unlock(&h_inode->i_mutex);
29185 +               if (unlikely(rerr)) {
29186 +                       AuIOErr("failed removing diropq for %pd b%d (%d)\n",
29187 +                               dentry, bdst, rerr);
29188 +                       err = -EIO;
29189 +                       goto out;
29190 +               }
29191 +       }
29192 +out_dir:
29193 +       if (au_ftest_cpdown(*flags, MADE_DIR)) {
29194 +               rerr = vfsub_sio_rmdir(au_h_iptr(dir, bdst), &h_path);
29195 +               if (unlikely(rerr)) {
29196 +                       AuIOErr("failed removing %pd b%d (%d)\n",
29197 +                               dentry, bdst, rerr);
29198 +                       err = -EIO;
29199 +               }
29200 +       }
29201 +out_put:
29202 +       au_set_h_dptr(dentry, bdst, NULL);
29203 +       if (au_dbend(dentry) == bdst)
29204 +               au_update_dbend(dentry);
29205 +out:
29206 +       dput(parent);
29207 +       return err;
29208 +}
29209 +
29210 +int au_cpdown_dirs(struct dentry *dentry, aufs_bindex_t bdst)
29211 +{
29212 +       int err;
29213 +       unsigned int flags;
29214 +
29215 +       flags = 0;
29216 +       err = au_cp_dirs(dentry, bdst, au_cpdown_dir, &flags);
29217 +
29218 +       return err;
29219 +}
29220 +
29221 +/* ---------------------------------------------------------------------- */
29222 +
29223 +/* policies for create */
29224 +
29225 +int au_wbr_nonopq(struct dentry *dentry, aufs_bindex_t bindex)
29226 +{
29227 +       int err, i, j, ndentry;
29228 +       aufs_bindex_t bopq;
29229 +       struct au_dcsub_pages dpages;
29230 +       struct au_dpage *dpage;
29231 +       struct dentry **dentries, *parent, *d;
29232 +
29233 +       err = au_dpages_init(&dpages, GFP_NOFS);
29234 +       if (unlikely(err))
29235 +               goto out;
29236 +       parent = dget_parent(dentry);
29237 +       err = au_dcsub_pages_rev_aufs(&dpages, parent, /*do_include*/0);
29238 +       if (unlikely(err))
29239 +               goto out_free;
29240 +
29241 +       err = bindex;
29242 +       for (i = 0; i < dpages.ndpage; i++) {
29243 +               dpage = dpages.dpages + i;
29244 +               dentries = dpage->dentries;
29245 +               ndentry = dpage->ndentry;
29246 +               for (j = 0; j < ndentry; j++) {
29247 +                       d = dentries[j];
29248 +                       di_read_lock_parent2(d, !AuLock_IR);
29249 +                       bopq = au_dbdiropq(d);
29250 +                       di_read_unlock(d, !AuLock_IR);
29251 +                       if (bopq >= 0 && bopq < err)
29252 +                               err = bopq;
29253 +               }
29254 +       }
29255 +
29256 +out_free:
29257 +       dput(parent);
29258 +       au_dpages_free(&dpages);
29259 +out:
29260 +       return err;
29261 +}
29262 +
29263 +static int au_wbr_bu(struct super_block *sb, aufs_bindex_t bindex)
29264 +{
29265 +       for (; bindex >= 0; bindex--)
29266 +               if (!au_br_rdonly(au_sbr(sb, bindex)))
29267 +                       return bindex;
29268 +       return -EROFS;
29269 +}
29270 +
29271 +/* top down parent */
29272 +static int au_wbr_create_tdp(struct dentry *dentry,
29273 +                            unsigned int flags __maybe_unused)
29274 +{
29275 +       int err;
29276 +       aufs_bindex_t bstart, bindex;
29277 +       struct super_block *sb;
29278 +       struct dentry *parent, *h_parent;
29279 +
29280 +       sb = dentry->d_sb;
29281 +       bstart = au_dbstart(dentry);
29282 +       err = bstart;
29283 +       if (!au_br_rdonly(au_sbr(sb, bstart)))
29284 +               goto out;
29285 +
29286 +       err = -EROFS;
29287 +       parent = dget_parent(dentry);
29288 +       for (bindex = au_dbstart(parent); bindex < bstart; bindex++) {
29289 +               h_parent = au_h_dptr(parent, bindex);
29290 +               if (!h_parent || !h_parent->d_inode)
29291 +                       continue;
29292 +
29293 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
29294 +                       err = bindex;
29295 +                       break;
29296 +               }
29297 +       }
29298 +       dput(parent);
29299 +
29300 +       /* bottom up here */
29301 +       if (unlikely(err < 0)) {
29302 +               err = au_wbr_bu(sb, bstart - 1);
29303 +               if (err >= 0)
29304 +                       err = au_wbr_nonopq(dentry, err);
29305 +       }
29306 +
29307 +out:
29308 +       AuDbg("b%d\n", err);
29309 +       return err;
29310 +}
29311 +
29312 +/* ---------------------------------------------------------------------- */
29313 +
29314 +/* an exception for the policy other than tdp */
29315 +static int au_wbr_create_exp(struct dentry *dentry)
29316 +{
29317 +       int err;
29318 +       aufs_bindex_t bwh, bdiropq;
29319 +       struct dentry *parent;
29320 +
29321 +       err = -1;
29322 +       bwh = au_dbwh(dentry);
29323 +       parent = dget_parent(dentry);
29324 +       bdiropq = au_dbdiropq(parent);
29325 +       if (bwh >= 0) {
29326 +               if (bdiropq >= 0)
29327 +                       err = min(bdiropq, bwh);
29328 +               else
29329 +                       err = bwh;
29330 +               AuDbg("%d\n", err);
29331 +       } else if (bdiropq >= 0) {
29332 +               err = bdiropq;
29333 +               AuDbg("%d\n", err);
29334 +       }
29335 +       dput(parent);
29336 +
29337 +       if (err >= 0)
29338 +               err = au_wbr_nonopq(dentry, err);
29339 +
29340 +       if (err >= 0 && au_br_rdonly(au_sbr(dentry->d_sb, err)))
29341 +               err = -1;
29342 +
29343 +       AuDbg("%d\n", err);
29344 +       return err;
29345 +}
29346 +
29347 +/* ---------------------------------------------------------------------- */
29348 +
29349 +/* round robin */
29350 +static int au_wbr_create_init_rr(struct super_block *sb)
29351 +{
29352 +       int err;
29353 +
29354 +       err = au_wbr_bu(sb, au_sbend(sb));
29355 +       atomic_set(&au_sbi(sb)->si_wbr_rr_next, -err); /* less important */
29356 +       /* smp_mb(); */
29357 +
29358 +       AuDbg("b%d\n", err);
29359 +       return err;
29360 +}
29361 +
29362 +static int au_wbr_create_rr(struct dentry *dentry, unsigned int flags)
29363 +{
29364 +       int err, nbr;
29365 +       unsigned int u;
29366 +       aufs_bindex_t bindex, bend;
29367 +       struct super_block *sb;
29368 +       atomic_t *next;
29369 +
29370 +       err = au_wbr_create_exp(dentry);
29371 +       if (err >= 0)
29372 +               goto out;
29373 +
29374 +       sb = dentry->d_sb;
29375 +       next = &au_sbi(sb)->si_wbr_rr_next;
29376 +       bend = au_sbend(sb);
29377 +       nbr = bend + 1;
29378 +       for (bindex = 0; bindex <= bend; bindex++) {
29379 +               if (!au_ftest_wbr(flags, DIR)) {
29380 +                       err = atomic_dec_return(next) + 1;
29381 +                       /* modulo for 0 is meaningless */
29382 +                       if (unlikely(!err))
29383 +                               err = atomic_dec_return(next) + 1;
29384 +               } else
29385 +                       err = atomic_read(next);
29386 +               AuDbg("%d\n", err);
29387 +               u = err;
29388 +               err = u % nbr;
29389 +               AuDbg("%d\n", err);
29390 +               if (!au_br_rdonly(au_sbr(sb, err)))
29391 +                       break;
29392 +               err = -EROFS;
29393 +       }
29394 +
29395 +       if (err >= 0)
29396 +               err = au_wbr_nonopq(dentry, err);
29397 +
29398 +out:
29399 +       AuDbg("%d\n", err);
29400 +       return err;
29401 +}
29402 +
29403 +/* ---------------------------------------------------------------------- */
29404 +
29405 +/* most free space */
29406 +static void au_mfs(struct dentry *dentry, struct dentry *parent)
29407 +{
29408 +       struct super_block *sb;
29409 +       struct au_branch *br;
29410 +       struct au_wbr_mfs *mfs;
29411 +       struct dentry *h_parent;
29412 +       aufs_bindex_t bindex, bend;
29413 +       int err;
29414 +       unsigned long long b, bavail;
29415 +       struct path h_path;
29416 +       /* reduce the stack usage */
29417 +       struct kstatfs *st;
29418 +
29419 +       st = kmalloc(sizeof(*st), GFP_NOFS);
29420 +       if (unlikely(!st)) {
29421 +               AuWarn1("failed updating mfs(%d), ignored\n", -ENOMEM);
29422 +               return;
29423 +       }
29424 +
29425 +       bavail = 0;
29426 +       sb = dentry->d_sb;
29427 +       mfs = &au_sbi(sb)->si_wbr_mfs;
29428 +       MtxMustLock(&mfs->mfs_lock);
29429 +       mfs->mfs_bindex = -EROFS;
29430 +       mfs->mfsrr_bytes = 0;
29431 +       if (!parent) {
29432 +               bindex = 0;
29433 +               bend = au_sbend(sb);
29434 +       } else {
29435 +               bindex = au_dbstart(parent);
29436 +               bend = au_dbtaildir(parent);
29437 +       }
29438 +
29439 +       for (; bindex <= bend; bindex++) {
29440 +               if (parent) {
29441 +                       h_parent = au_h_dptr(parent, bindex);
29442 +                       if (!h_parent || !h_parent->d_inode)
29443 +                               continue;
29444 +               }
29445 +               br = au_sbr(sb, bindex);
29446 +               if (au_br_rdonly(br))
29447 +                       continue;
29448 +
29449 +               /* sb->s_root for NFS is unreliable */
29450 +               h_path.mnt = au_br_mnt(br);
29451 +               h_path.dentry = h_path.mnt->mnt_root;
29452 +               err = vfs_statfs(&h_path, st);
29453 +               if (unlikely(err)) {
29454 +                       AuWarn1("failed statfs, b%d, %d\n", bindex, err);
29455 +                       continue;
29456 +               }
29457 +
29458 +               /* when the available size is equal, select the lower one */
29459 +               BUILD_BUG_ON(sizeof(b) < sizeof(st->f_bavail)
29460 +                            || sizeof(b) < sizeof(st->f_bsize));
29461 +               b = st->f_bavail * st->f_bsize;
29462 +               br->br_wbr->wbr_bytes = b;
29463 +               if (b >= bavail) {
29464 +                       bavail = b;
29465 +                       mfs->mfs_bindex = bindex;
29466 +                       mfs->mfs_jiffy = jiffies;
29467 +               }
29468 +       }
29469 +
29470 +       mfs->mfsrr_bytes = bavail;
29471 +       AuDbg("b%d\n", mfs->mfs_bindex);
29472 +       kfree(st);
29473 +}
29474 +
29475 +static int au_wbr_create_mfs(struct dentry *dentry, unsigned int flags)
29476 +{
29477 +       int err;
29478 +       struct dentry *parent;
29479 +       struct super_block *sb;
29480 +       struct au_wbr_mfs *mfs;
29481 +
29482 +       err = au_wbr_create_exp(dentry);
29483 +       if (err >= 0)
29484 +               goto out;
29485 +
29486 +       sb = dentry->d_sb;
29487 +       parent = NULL;
29488 +       if (au_ftest_wbr(flags, PARENT))
29489 +               parent = dget_parent(dentry);
29490 +       mfs = &au_sbi(sb)->si_wbr_mfs;
29491 +       mutex_lock(&mfs->mfs_lock);
29492 +       if (time_after(jiffies, mfs->mfs_jiffy + mfs->mfs_expire)
29493 +           || mfs->mfs_bindex < 0
29494 +           || au_br_rdonly(au_sbr(sb, mfs->mfs_bindex)))
29495 +               au_mfs(dentry, parent);
29496 +       mutex_unlock(&mfs->mfs_lock);
29497 +       err = mfs->mfs_bindex;
29498 +       dput(parent);
29499 +
29500 +       if (err >= 0)
29501 +               err = au_wbr_nonopq(dentry, err);
29502 +
29503 +out:
29504 +       AuDbg("b%d\n", err);
29505 +       return err;
29506 +}
29507 +
29508 +static int au_wbr_create_init_mfs(struct super_block *sb)
29509 +{
29510 +       struct au_wbr_mfs *mfs;
29511 +
29512 +       mfs = &au_sbi(sb)->si_wbr_mfs;
29513 +       mutex_init(&mfs->mfs_lock);
29514 +       mfs->mfs_jiffy = 0;
29515 +       mfs->mfs_bindex = -EROFS;
29516 +
29517 +       return 0;
29518 +}
29519 +
29520 +static int au_wbr_create_fin_mfs(struct super_block *sb __maybe_unused)
29521 +{
29522 +       mutex_destroy(&au_sbi(sb)->si_wbr_mfs.mfs_lock);
29523 +       return 0;
29524 +}
29525 +
29526 +/* ---------------------------------------------------------------------- */
29527 +
29528 +/* most free space and then round robin */
29529 +static int au_wbr_create_mfsrr(struct dentry *dentry, unsigned int flags)
29530 +{
29531 +       int err;
29532 +       struct au_wbr_mfs *mfs;
29533 +
29534 +       err = au_wbr_create_mfs(dentry, flags);
29535 +       if (err >= 0) {
29536 +               mfs = &au_sbi(dentry->d_sb)->si_wbr_mfs;
29537 +               mutex_lock(&mfs->mfs_lock);
29538 +               if (mfs->mfsrr_bytes < mfs->mfsrr_watermark)
29539 +                       err = au_wbr_create_rr(dentry, flags);
29540 +               mutex_unlock(&mfs->mfs_lock);
29541 +       }
29542 +
29543 +       AuDbg("b%d\n", err);
29544 +       return err;
29545 +}
29546 +
29547 +static int au_wbr_create_init_mfsrr(struct super_block *sb)
29548 +{
29549 +       int err;
29550 +
29551 +       au_wbr_create_init_mfs(sb); /* ignore */
29552 +       err = au_wbr_create_init_rr(sb);
29553 +
29554 +       return err;
29555 +}
29556 +
29557 +/* ---------------------------------------------------------------------- */
29558 +
29559 +/* top down parent and most free space */
29560 +static int au_wbr_create_pmfs(struct dentry *dentry, unsigned int flags)
29561 +{
29562 +       int err, e2;
29563 +       unsigned long long b;
29564 +       aufs_bindex_t bindex, bstart, bend;
29565 +       struct super_block *sb;
29566 +       struct dentry *parent, *h_parent;
29567 +       struct au_branch *br;
29568 +
29569 +       err = au_wbr_create_tdp(dentry, flags);
29570 +       if (unlikely(err < 0))
29571 +               goto out;
29572 +       parent = dget_parent(dentry);
29573 +       bstart = au_dbstart(parent);
29574 +       bend = au_dbtaildir(parent);
29575 +       if (bstart == bend)
29576 +               goto out_parent; /* success */
29577 +
29578 +       e2 = au_wbr_create_mfs(dentry, flags);
29579 +       if (e2 < 0)
29580 +               goto out_parent; /* success */
29581 +
29582 +       /* when the available size is equal, select upper one */
29583 +       sb = dentry->d_sb;
29584 +       br = au_sbr(sb, err);
29585 +       b = br->br_wbr->wbr_bytes;
29586 +       AuDbg("b%d, %llu\n", err, b);
29587 +
29588 +       for (bindex = bstart; bindex <= bend; bindex++) {
29589 +               h_parent = au_h_dptr(parent, bindex);
29590 +               if (!h_parent || !h_parent->d_inode)
29591 +                       continue;
29592 +
29593 +               br = au_sbr(sb, bindex);
29594 +               if (!au_br_rdonly(br) && br->br_wbr->wbr_bytes > b) {
29595 +                       b = br->br_wbr->wbr_bytes;
29596 +                       err = bindex;
29597 +                       AuDbg("b%d, %llu\n", err, b);
29598 +               }
29599 +       }
29600 +
29601 +       if (err >= 0)
29602 +               err = au_wbr_nonopq(dentry, err);
29603 +
29604 +out_parent:
29605 +       dput(parent);
29606 +out:
29607 +       AuDbg("b%d\n", err);
29608 +       return err;
29609 +}
29610 +
29611 +/* ---------------------------------------------------------------------- */
29612 +
29613 +/*
29614 + * - top down parent
29615 + * - most free space with parent
29616 + * - most free space round-robin regardless parent
29617 + */
29618 +static int au_wbr_create_pmfsrr(struct dentry *dentry, unsigned int flags)
29619 +{
29620 +       int err;
29621 +       unsigned long long watermark;
29622 +       struct super_block *sb;
29623 +       struct au_branch *br;
29624 +       struct au_wbr_mfs *mfs;
29625 +
29626 +       err = au_wbr_create_pmfs(dentry, flags | AuWbr_PARENT);
29627 +       if (unlikely(err < 0))
29628 +               goto out;
29629 +
29630 +       sb = dentry->d_sb;
29631 +       br = au_sbr(sb, err);
29632 +       mfs = &au_sbi(sb)->si_wbr_mfs;
29633 +       mutex_lock(&mfs->mfs_lock);
29634 +       watermark = mfs->mfsrr_watermark;
29635 +       mutex_unlock(&mfs->mfs_lock);
29636 +       if (br->br_wbr->wbr_bytes < watermark)
29637 +               /* regardless the parent dir */
29638 +               err = au_wbr_create_mfsrr(dentry, flags);
29639 +
29640 +out:
29641 +       AuDbg("b%d\n", err);
29642 +       return err;
29643 +}
29644 +
29645 +/* ---------------------------------------------------------------------- */
29646 +
29647 +/* policies for copyup */
29648 +
29649 +/* top down parent */
29650 +static int au_wbr_copyup_tdp(struct dentry *dentry)
29651 +{
29652 +       return au_wbr_create_tdp(dentry, /*flags, anything is ok*/0);
29653 +}
29654 +
29655 +/* bottom up parent */
29656 +static int au_wbr_copyup_bup(struct dentry *dentry)
29657 +{
29658 +       int err;
29659 +       aufs_bindex_t bindex, bstart;
29660 +       struct dentry *parent, *h_parent;
29661 +       struct super_block *sb;
29662 +
29663 +       err = -EROFS;
29664 +       sb = dentry->d_sb;
29665 +       parent = dget_parent(dentry);
29666 +       bstart = au_dbstart(parent);
29667 +       for (bindex = au_dbstart(dentry); bindex >= bstart; bindex--) {
29668 +               h_parent = au_h_dptr(parent, bindex);
29669 +               if (!h_parent || !h_parent->d_inode)
29670 +                       continue;
29671 +
29672 +               if (!au_br_rdonly(au_sbr(sb, bindex))) {
29673 +                       err = bindex;
29674 +                       break;
29675 +               }
29676 +       }
29677 +       dput(parent);
29678 +
29679 +       /* bottom up here */
29680 +       if (unlikely(err < 0))
29681 +               err = au_wbr_bu(sb, bstart - 1);
29682 +
29683 +       AuDbg("b%d\n", err);
29684 +       return err;
29685 +}
29686 +
29687 +/* bottom up */
29688 +static int au_wbr_copyup_bu(struct dentry *dentry)
29689 +{
29690 +       int err;
29691 +       aufs_bindex_t bstart;
29692 +
29693 +       bstart = au_dbstart(dentry);
29694 +       err = au_wbr_bu(dentry->d_sb, bstart);
29695 +       AuDbg("b%d\n", err);
29696 +       if (err > bstart)
29697 +               err = au_wbr_nonopq(dentry, err);
29698 +
29699 +       AuDbg("b%d\n", err);
29700 +       return err;
29701 +}
29702 +
29703 +/* ---------------------------------------------------------------------- */
29704 +
29705 +struct au_wbr_copyup_operations au_wbr_copyup_ops[] = {
29706 +       [AuWbrCopyup_TDP] = {
29707 +               .copyup = au_wbr_copyup_tdp
29708 +       },
29709 +       [AuWbrCopyup_BUP] = {
29710 +               .copyup = au_wbr_copyup_bup
29711 +       },
29712 +       [AuWbrCopyup_BU] = {
29713 +               .copyup = au_wbr_copyup_bu
29714 +       }
29715 +};
29716 +
29717 +struct au_wbr_create_operations au_wbr_create_ops[] = {
29718 +       [AuWbrCreate_TDP] = {
29719 +               .create = au_wbr_create_tdp
29720 +       },
29721 +       [AuWbrCreate_RR] = {
29722 +               .create = au_wbr_create_rr,
29723 +               .init   = au_wbr_create_init_rr
29724 +       },
29725 +       [AuWbrCreate_MFS] = {
29726 +               .create = au_wbr_create_mfs,
29727 +               .init   = au_wbr_create_init_mfs,
29728 +               .fin    = au_wbr_create_fin_mfs
29729 +       },
29730 +       [AuWbrCreate_MFSV] = {
29731 +               .create = au_wbr_create_mfs,
29732 +               .init   = au_wbr_create_init_mfs,
29733 +               .fin    = au_wbr_create_fin_mfs
29734 +       },
29735 +       [AuWbrCreate_MFSRR] = {
29736 +               .create = au_wbr_create_mfsrr,
29737 +               .init   = au_wbr_create_init_mfsrr,
29738 +               .fin    = au_wbr_create_fin_mfs
29739 +       },
29740 +       [AuWbrCreate_MFSRRV] = {
29741 +               .create = au_wbr_create_mfsrr,
29742 +               .init   = au_wbr_create_init_mfsrr,
29743 +               .fin    = au_wbr_create_fin_mfs
29744 +       },
29745 +       [AuWbrCreate_PMFS] = {
29746 +               .create = au_wbr_create_pmfs,
29747 +               .init   = au_wbr_create_init_mfs,
29748 +               .fin    = au_wbr_create_fin_mfs
29749 +       },
29750 +       [AuWbrCreate_PMFSV] = {
29751 +               .create = au_wbr_create_pmfs,
29752 +               .init   = au_wbr_create_init_mfs,
29753 +               .fin    = au_wbr_create_fin_mfs
29754 +       },
29755 +       [AuWbrCreate_PMFSRR] = {
29756 +               .create = au_wbr_create_pmfsrr,
29757 +               .init   = au_wbr_create_init_mfsrr,
29758 +               .fin    = au_wbr_create_fin_mfs
29759 +       },
29760 +       [AuWbrCreate_PMFSRRV] = {
29761 +               .create = au_wbr_create_pmfsrr,
29762 +               .init   = au_wbr_create_init_mfsrr,
29763 +               .fin    = au_wbr_create_fin_mfs
29764 +       }
29765 +};
29766 diff -urN /usr/share/empty/fs/aufs/whout.c linux/fs/aufs/whout.c
29767 --- /usr/share/empty/fs/aufs/whout.c    1970-01-01 01:00:00.000000000 +0100
29768 +++ linux/fs/aufs/whout.c       2014-04-24 22:11:10.861935852 +0200
29769 @@ -0,0 +1,1052 @@
29770 +/*
29771 + * Copyright (C) 2005-2014 Junjiro R. Okajima
29772 + *
29773 + * This program, aufs is free software; you can redistribute it and/or modify
29774 + * it under the terms of the GNU General Public License as published by
29775 + * the Free Software Foundation; either version 2 of the License, or
29776 + * (at your option) any later version.
29777 + *
29778 + * This program is distributed in the hope that it will be useful,
29779 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
29780 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29781 + * GNU General Public License for more details.
29782 + *
29783 + * You should have received a copy of the GNU General Public License
29784 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
29785 + */
29786 +
29787 +/*
29788 + * whiteout for logical deletion and opaque directory
29789 + */
29790 +
29791 +#include "aufs.h"
29792 +
29793 +#define WH_MASK                        S_IRUGO
29794 +
29795 +/*
29796 + * If a directory contains this file, then it is opaque.  We start with the
29797 + * .wh. flag so that it is blocked by lookup.
29798 + */
29799 +static struct qstr diropq_name = QSTR_INIT(AUFS_WH_DIROPQ,
29800 +                                          sizeof(AUFS_WH_DIROPQ) - 1);
29801 +
29802 +/*
29803 + * generate whiteout name, which is NOT terminated by NULL.
29804 + * @name: original d_name.name
29805 + * @len: original d_name.len
29806 + * @wh: whiteout qstr
29807 + * returns zero when succeeds, otherwise error.
29808 + * succeeded value as wh->name should be freed by kfree().
29809 + */
29810 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name)
29811 +{
29812 +       char *p;
29813 +
29814 +       if (unlikely(name->len > PATH_MAX - AUFS_WH_PFX_LEN))
29815 +               return -ENAMETOOLONG;
29816 +
29817 +       wh->len = name->len + AUFS_WH_PFX_LEN;
29818 +       p = kmalloc(wh->len, GFP_NOFS);
29819 +       wh->name = p;
29820 +       if (p) {
29821 +               memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
29822 +               memcpy(p + AUFS_WH_PFX_LEN, name->name, name->len);
29823 +               /* smp_mb(); */
29824 +               return 0;
29825 +       }
29826 +       return -ENOMEM;
29827 +}
29828 +
29829 +/* ---------------------------------------------------------------------- */
29830 +
29831 +/*
29832 + * test if the @wh_name exists under @h_parent.
29833 + * @try_sio specifies the necessary of super-io.
29834 + */
29835 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name,
29836 +              struct au_branch *br, int try_sio)
29837 +{
29838 +       int err;
29839 +       struct dentry *wh_dentry;
29840 +
29841 +       if (!try_sio)
29842 +               wh_dentry = vfsub_lkup_one(wh_name, h_parent);
29843 +       else
29844 +               wh_dentry = au_sio_lkup_one(wh_name, h_parent, br);
29845 +       err = PTR_ERR(wh_dentry);
29846 +       if (IS_ERR(wh_dentry))
29847 +               goto out;
29848 +
29849 +       err = 0;
29850 +       if (!wh_dentry->d_inode)
29851 +               goto out_wh; /* success */
29852 +
29853 +       err = 1;
29854 +       if (S_ISREG(wh_dentry->d_inode->i_mode))
29855 +               goto out_wh; /* success */
29856 +
29857 +       err = -EIO;
29858 +       AuIOErr("%pd Invalid whiteout entry type 0%o.\n",
29859 +               wh_dentry, wh_dentry->d_inode->i_mode);
29860 +
29861 +out_wh:
29862 +       dput(wh_dentry);
29863 +out:
29864 +       return err;
29865 +}
29866 +
29867 +/*
29868 + * test if the @h_dentry sets opaque or not.
29869 + */
29870 +int au_diropq_test(struct dentry *h_dentry, struct au_branch *br)
29871 +{
29872 +       int err;
29873 +       struct inode *h_dir;
29874 +
29875 +       h_dir = h_dentry->d_inode;
29876 +       err = au_wh_test(h_dentry, &diropq_name, br,
29877 +                        au_test_h_perm_sio(h_dir, MAY_EXEC));
29878 +       return err;
29879 +}
29880 +
29881 +/*
29882 + * returns a negative dentry whose name is unique and temporary.
29883 + */
29884 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
29885 +                            struct qstr *prefix)
29886 +{
29887 +       struct dentry *dentry;
29888 +       int i;
29889 +       char defname[NAME_MAX - AUFS_MAX_NAMELEN + DNAME_INLINE_LEN + 1],
29890 +               *name, *p;
29891 +       /* strict atomic_t is unnecessary here */
29892 +       static unsigned short cnt;
29893 +       struct qstr qs;
29894 +
29895 +       BUILD_BUG_ON(sizeof(cnt) * 2 > AUFS_WH_TMP_LEN);
29896 +
29897 +       name = defname;
29898 +       qs.len = sizeof(defname) - DNAME_INLINE_LEN + prefix->len - 1;
29899 +       if (unlikely(prefix->len > DNAME_INLINE_LEN)) {
29900 +               dentry = ERR_PTR(-ENAMETOOLONG);
29901 +               if (unlikely(qs.len > NAME_MAX))
29902 +                       goto out;
29903 +               dentry = ERR_PTR(-ENOMEM);
29904 +               name = kmalloc(qs.len + 1, GFP_NOFS);
29905 +               if (unlikely(!name))
29906 +                       goto out;
29907 +       }
29908 +
29909 +       /* doubly whiteout-ed */
29910 +       memcpy(name, AUFS_WH_PFX AUFS_WH_PFX, AUFS_WH_PFX_LEN * 2);
29911 +       p = name + AUFS_WH_PFX_LEN * 2;
29912 +       memcpy(p, prefix->name, prefix->len);
29913 +       p += prefix->len;
29914 +       *p++ = '.';
29915 +       AuDebugOn(name + qs.len + 1 - p <= AUFS_WH_TMP_LEN);
29916 +
29917 +       qs.name = name;
29918 +       for (i = 0; i < 3; i++) {
29919 +               sprintf(p, "%.*x", AUFS_WH_TMP_LEN, cnt++);
29920 +               dentry = au_sio_lkup_one(&qs, h_parent, br);
29921 +               if (IS_ERR(dentry) || !dentry->d_inode)
29922 +                       goto out_name;
29923 +               dput(dentry);
29924 +       }
29925 +       /* pr_warn("could not get random name\n"); */
29926 +       dentry = ERR_PTR(-EEXIST);
29927 +       AuDbg("%.*s\n", AuLNPair(&qs));
29928 +       BUG();
29929 +
29930 +out_name:
29931 +       if (name != defname)
29932 +               kfree(name);
29933 +out:
29934 +       AuTraceErrPtr(dentry);
29935 +       return dentry;
29936 +}
29937 +
29938 +/*
29939 + * rename the @h_dentry on @br to the whiteouted temporary name.
29940 + */
29941 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br)
29942 +{
29943 +       int err;
29944 +       struct path h_path = {
29945 +               .mnt = au_br_mnt(br)
29946 +       };
29947 +       struct inode *h_dir, *delegated;
29948 +       struct dentry *h_parent;
29949 +
29950 +       h_parent = h_dentry->d_parent; /* dir inode is locked */
29951 +       h_dir = h_parent->d_inode;
29952 +       IMustLock(h_dir);
29953 +
29954 +       h_path.dentry = au_whtmp_lkup(h_parent, br, &h_dentry->d_name);
29955 +       err = PTR_ERR(h_path.dentry);
29956 +       if (IS_ERR(h_path.dentry))
29957 +               goto out;
29958 +
29959 +       /* under the same dir, no need to lock_rename() */
29960 +       delegated = NULL;
29961 +       err = vfsub_rename(h_dir, h_dentry, h_dir, &h_path, &delegated);
29962 +       AuTraceErr(err);
29963 +       if (unlikely(err == -EWOULDBLOCK)) {
29964 +               pr_warn("cannot retry for NFSv4 delegation"
29965 +                       " for an internal rename\n");
29966 +               iput(delegated);
29967 +       }
29968 +       dput(h_path.dentry);
29969 +
29970 +out:
29971 +       AuTraceErr(err);
29972 +       return err;
29973 +}
29974 +
29975 +/* ---------------------------------------------------------------------- */
29976 +/*
29977 + * functions for removing a whiteout
29978 + */
29979 +
29980 +static int do_unlink_wh(struct inode *h_dir, struct path *h_path)
29981 +{
29982 +       int err, force;
29983 +       struct inode *delegated;
29984 +
29985 +       /*
29986 +        * forces superio when the dir has a sticky bit.
29987 +        * this may be a violation of unix fs semantics.
29988 +        */
29989 +       force = (h_dir->i_mode & S_ISVTX)
29990 +               && !uid_eq(current_fsuid(), h_path->dentry->d_inode->i_uid);
29991 +       delegated = NULL;
29992 +       err = vfsub_unlink(h_dir, h_path, &delegated, force);
29993 +       if (unlikely(err == -EWOULDBLOCK)) {
29994 +               pr_warn("cannot retry for NFSv4 delegation"
29995 +                       " for an internal unlink\n");
29996 +               iput(delegated);
29997 +       }
29998 +       return err;
29999 +}
30000 +
30001 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
30002 +                       struct dentry *dentry)
30003 +{
30004 +       int err;
30005 +
30006 +       err = do_unlink_wh(h_dir, h_path);
30007 +       if (!err && dentry)
30008 +               au_set_dbwh(dentry, -1);
30009 +
30010 +       return err;
30011 +}
30012 +
30013 +static int unlink_wh_name(struct dentry *h_parent, struct qstr *wh,
30014 +                         struct au_branch *br)
30015 +{
30016 +       int err;
30017 +       struct path h_path = {
30018 +               .mnt = au_br_mnt(br)
30019 +       };
30020 +
30021 +       err = 0;
30022 +       h_path.dentry = vfsub_lkup_one(wh, h_parent);
30023 +       if (IS_ERR(h_path.dentry))
30024 +               err = PTR_ERR(h_path.dentry);
30025 +       else {
30026 +               if (h_path.dentry->d_inode
30027 +                   && S_ISREG(h_path.dentry->d_inode->i_mode))
30028 +                       err = do_unlink_wh(h_parent->d_inode, &h_path);
30029 +               dput(h_path.dentry);
30030 +       }
30031 +
30032 +       return err;
30033 +}
30034 +
30035 +/* ---------------------------------------------------------------------- */
30036 +/*
30037 + * initialize/clean whiteout for a branch
30038 + */
30039 +
30040 +static void au_wh_clean(struct inode *h_dir, struct path *whpath,
30041 +                       const int isdir)
30042 +{
30043 +       int err;
30044 +       struct inode *delegated;
30045 +
30046 +       if (!whpath->dentry->d_inode)
30047 +               return;
30048 +
30049 +       if (isdir)
30050 +               err = vfsub_rmdir(h_dir, whpath);
30051 +       else {
30052 +               delegated = NULL;
30053 +               err = vfsub_unlink(h_dir, whpath, &delegated, /*force*/0);
30054 +               if (unlikely(err == -EWOULDBLOCK)) {
30055 +                       pr_warn("cannot retry for NFSv4 delegation"
30056 +                               " for an internal unlink\n");
30057 +                       iput(delegated);
30058 +               }
30059 +       }
30060 +       if (unlikely(err))
30061 +               pr_warn("failed removing %pd (%d), ignored.\n",
30062 +                       whpath->dentry, err);
30063 +}
30064 +
30065 +static int test_linkable(struct dentry *h_root)
30066 +{
30067 +       struct inode *h_dir = h_root->d_inode;
30068 +
30069 +       if (h_dir->i_op->link)
30070 +               return 0;
30071 +
30072 +       pr_err("%pd (%s) doesn't support link(2), use noplink and rw+nolwh\n",
30073 +              h_root, au_sbtype(h_root->d_sb));
30074 +       return -ENOSYS;
30075 +}
30076 +
30077 +/* todo: should this mkdir be done in /sbin/mount.aufs helper? */
30078 +static int au_whdir(struct inode *h_dir, struct path *path)
30079 +{
30080 +       int err;
30081 +
30082 +       err = -EEXIST;
30083 +       if (!path->dentry->d_inode) {
30084 +               int mode = S_IRWXU;
30085 +
30086 +               if (au_test_nfs(path->dentry->d_sb))
30087 +                       mode |= S_IXUGO;
30088 +               err = vfsub_mkdir(h_dir, path, mode);
30089 +       } else if (S_ISDIR(path->dentry->d_inode->i_mode))
30090 +               err = 0;
30091 +       else
30092 +               pr_err("unknown %pd exists\n", path->dentry);
30093 +
30094 +       return err;
30095 +}
30096 +
30097 +struct au_wh_base {
30098 +       const struct qstr *name;
30099 +       struct dentry *dentry;
30100 +};
30101 +
30102 +static void au_wh_init_ro(struct inode *h_dir, struct au_wh_base base[],
30103 +                         struct path *h_path)
30104 +{
30105 +       h_path->dentry = base[AuBrWh_BASE].dentry;
30106 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
30107 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
30108 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
30109 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
30110 +       au_wh_clean(h_dir, h_path, /*isdir*/1);
30111 +}
30112 +
30113 +/*
30114 + * returns tri-state,
30115 + * minus: error, caller should print the mesage
30116 + * zero: succuess
30117 + * plus: error, caller should NOT print the mesage
30118 + */
30119 +static int au_wh_init_rw_nolink(struct dentry *h_root, struct au_wbr *wbr,
30120 +                               int do_plink, struct au_wh_base base[],
30121 +                               struct path *h_path)
30122 +{
30123 +       int err;
30124 +       struct inode *h_dir;
30125 +
30126 +       h_dir = h_root->d_inode;
30127 +       h_path->dentry = base[AuBrWh_BASE].dentry;
30128 +       au_wh_clean(h_dir, h_path, /*isdir*/0);
30129 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
30130 +       if (do_plink) {
30131 +               err = test_linkable(h_root);
30132 +               if (unlikely(err)) {
30133 +                       err = 1;
30134 +                       goto out;
30135 +               }
30136 +
30137 +               err = au_whdir(h_dir, h_path);
30138 +               if (unlikely(err))
30139 +                       goto out;
30140 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
30141 +       } else
30142 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
30143 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
30144 +       err = au_whdir(h_dir, h_path);
30145 +       if (unlikely(err))
30146 +               goto out;
30147 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
30148 +
30149 +out:
30150 +       return err;
30151 +}
30152 +
30153 +/*
30154 + * for the moment, aufs supports the branch filesystem which does not support
30155 + * link(2). testing on FAT which does not support i_op->setattr() fully either,
30156 + * copyup failed. finally, such filesystem will not be used as the writable
30157 + * branch.
30158 + *
30159 + * returns tri-state, see above.
30160 + */
30161 +static int au_wh_init_rw(struct dentry *h_root, struct au_wbr *wbr,
30162 +                        int do_plink, struct au_wh_base base[],
30163 +                        struct path *h_path)
30164 +{
30165 +       int err;
30166 +       struct inode *h_dir;
30167 +
30168 +       WbrWhMustWriteLock(wbr);
30169 +
30170 +       err = test_linkable(h_root);
30171 +       if (unlikely(err)) {
30172 +               err = 1;
30173 +               goto out;
30174 +       }
30175 +
30176 +       /*
30177 +        * todo: should this create be done in /sbin/mount.aufs helper?
30178 +        */
30179 +       err = -EEXIST;
30180 +       h_dir = h_root->d_inode;
30181 +       if (!base[AuBrWh_BASE].dentry->d_inode) {
30182 +               h_path->dentry = base[AuBrWh_BASE].dentry;
30183 +               err = vfsub_create(h_dir, h_path, WH_MASK, /*want_excl*/true);
30184 +       } else if (S_ISREG(base[AuBrWh_BASE].dentry->d_inode->i_mode))
30185 +               err = 0;
30186 +       else
30187 +               pr_err("unknown %pd2 exists\n", base[AuBrWh_BASE].dentry);
30188 +       if (unlikely(err))
30189 +               goto out;
30190 +
30191 +       h_path->dentry = base[AuBrWh_PLINK].dentry;
30192 +       if (do_plink) {
30193 +               err = au_whdir(h_dir, h_path);
30194 +               if (unlikely(err))
30195 +                       goto out;
30196 +               wbr->wbr_plink = dget(base[AuBrWh_PLINK].dentry);
30197 +       } else
30198 +               au_wh_clean(h_dir, h_path, /*isdir*/1);
30199 +       wbr->wbr_whbase = dget(base[AuBrWh_BASE].dentry);
30200 +
30201 +       h_path->dentry = base[AuBrWh_ORPH].dentry;
30202 +       err = au_whdir(h_dir, h_path);
30203 +       if (unlikely(err))
30204 +               goto out;
30205 +       wbr->wbr_orph = dget(base[AuBrWh_ORPH].dentry);
30206 +
30207 +out:
30208 +       return err;
30209 +}
30210 +
30211 +/*
30212 + * initialize the whiteout base file/dir for @br.
30213 + */
30214 +int au_wh_init(struct au_branch *br, struct super_block *sb)
30215 +{
30216 +       int err, i;
30217 +       const unsigned char do_plink
30218 +               = !!au_opt_test(au_mntflags(sb), PLINK);
30219 +       struct inode *h_dir;
30220 +       struct path path = br->br_path;
30221 +       struct dentry *h_root = path.dentry;
30222 +       struct au_wbr *wbr = br->br_wbr;
30223 +       static const struct qstr base_name[] = {
30224 +               [AuBrWh_BASE] = QSTR_INIT(AUFS_BASE_NAME,
30225 +                                         sizeof(AUFS_BASE_NAME) - 1),
30226 +               [AuBrWh_PLINK] = QSTR_INIT(AUFS_PLINKDIR_NAME,
30227 +                                          sizeof(AUFS_PLINKDIR_NAME) - 1),
30228 +               [AuBrWh_ORPH] = QSTR_INIT(AUFS_ORPHDIR_NAME,
30229 +                                         sizeof(AUFS_ORPHDIR_NAME) - 1)
30230 +       };
30231 +       struct au_wh_base base[] = {
30232 +               [AuBrWh_BASE] = {
30233 +                       .name   = base_name + AuBrWh_BASE,
30234 +                       .dentry = NULL
30235 +               },
30236 +               [AuBrWh_PLINK] = {
30237 +                       .name   = base_name + AuBrWh_PLINK,
30238 +                       .dentry = NULL
30239 +               },
30240 +               [AuBrWh_ORPH] = {
30241 +                       .name   = base_name + AuBrWh_ORPH,
30242 +                       .dentry = NULL
30243 +               }
30244 +       };
30245 +
30246 +       if (wbr)
30247 +               WbrWhMustWriteLock(wbr);
30248 +
30249 +       for (i = 0; i < AuBrWh_Last; i++) {
30250 +               /* doubly whiteouted */
30251 +               struct dentry *d;
30252 +
30253 +               d = au_wh_lkup(h_root, (void *)base[i].name, br);
30254 +               err = PTR_ERR(d);
30255 +               if (IS_ERR(d))
30256 +                       goto out;
30257 +
30258 +               base[i].dentry = d;
30259 +               AuDebugOn(wbr
30260 +                         && wbr->wbr_wh[i]
30261 +                         && wbr->wbr_wh[i] != base[i].dentry);
30262 +       }
30263 +
30264 +       if (wbr)
30265 +               for (i = 0; i < AuBrWh_Last; i++) {
30266 +                       dput(wbr->wbr_wh[i]);
30267 +                       wbr->wbr_wh[i] = NULL;
30268 +               }
30269 +
30270 +       err = 0;
30271 +       if (!au_br_writable(br->br_perm)) {
30272 +               h_dir = h_root->d_inode;
30273 +               au_wh_init_ro(h_dir, base, &path);
30274 +       } else if (!au_br_wh_linkable(br->br_perm)) {
30275 +               err = au_wh_init_rw_nolink(h_root, wbr, do_plink, base, &path);
30276 +               if (err > 0)
30277 +                       goto out;
30278 +               else if (err)
30279 +                       goto out_err;
30280 +       } else {
30281 +               err = au_wh_init_rw(h_root, wbr, do_plink, base, &path);
30282 +               if (err > 0)
30283 +                       goto out;
30284 +               else if (err)
30285 +                       goto out_err;
30286 +       }
30287 +       goto out; /* success */
30288 +
30289 +out_err:
30290 +       pr_err("an error(%d) on the writable branch %pd(%s)\n",
30291 +              err, h_root, au_sbtype(h_root->d_sb));
30292 +out:
30293 +       for (i = 0; i < AuBrWh_Last; i++)
30294 +               dput(base[i].dentry);
30295 +       return err;
30296 +}
30297 +
30298 +/* ---------------------------------------------------------------------- */
30299 +/*
30300 + * whiteouts are all hard-linked usually.
30301 + * when its link count reaches a ceiling, we create a new whiteout base
30302 + * asynchronously.
30303 + */
30304 +
30305 +struct reinit_br_wh {
30306 +       struct super_block *sb;
30307 +       struct au_branch *br;
30308 +};
30309 +
30310 +static void reinit_br_wh(void *arg)
30311 +{
30312 +       int err;
30313 +       aufs_bindex_t bindex;
30314 +       struct path h_path;
30315 +       struct reinit_br_wh *a = arg;
30316 +       struct au_wbr *wbr;
30317 +       struct inode *dir, *delegated;
30318 +       struct dentry *h_root;
30319 +       struct au_hinode *hdir;
30320 +
30321 +       err = 0;
30322 +       wbr = a->br->br_wbr;
30323 +       /* big aufs lock */
30324 +       si_noflush_write_lock(a->sb);
30325 +       if (!au_br_writable(a->br->br_perm))
30326 +               goto out;
30327 +       bindex = au_br_index(a->sb, a->br->br_id);
30328 +       if (unlikely(bindex < 0))
30329 +               goto out;
30330 +
30331 +       di_read_lock_parent(a->sb->s_root, AuLock_IR);
30332 +       dir = a->sb->s_root->d_inode;
30333 +       hdir = au_hi(dir, bindex);
30334 +       h_root = au_h_dptr(a->sb->s_root, bindex);
30335 +       AuDebugOn(h_root != au_br_dentry(a->br));
30336 +
30337 +       au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
30338 +       wbr_wh_write_lock(wbr);
30339 +       err = au_h_verify(wbr->wbr_whbase, au_opt_udba(a->sb), hdir->hi_inode,
30340 +                         h_root, a->br);
30341 +       if (!err) {
30342 +               h_path.dentry = wbr->wbr_whbase;
30343 +               h_path.mnt = au_br_mnt(a->br);
30344 +               delegated = NULL;
30345 +               err = vfsub_unlink(hdir->hi_inode, &h_path, &delegated,
30346 +                                  /*force*/0);
30347 +               if (unlikely(err == -EWOULDBLOCK)) {
30348 +                       pr_warn("cannot retry for NFSv4 delegation"
30349 +                               " for an internal unlink\n");
30350 +                       iput(delegated);
30351 +               }
30352 +       } else {
30353 +               pr_warn("%pd is moved, ignored\n", wbr->wbr_whbase);
30354 +               err = 0;
30355 +       }
30356 +       dput(wbr->wbr_whbase);
30357 +       wbr->wbr_whbase = NULL;
30358 +       if (!err)
30359 +               err = au_wh_init(a->br, a->sb);
30360 +       wbr_wh_write_unlock(wbr);
30361 +       au_hn_imtx_unlock(hdir);
30362 +       di_read_unlock(a->sb->s_root, AuLock_IR);
30363 +
30364 +out:
30365 +       if (wbr)
30366 +               atomic_dec(&wbr->wbr_wh_running);
30367 +       atomic_dec(&a->br->br_count);
30368 +       si_write_unlock(a->sb);
30369 +       au_nwt_done(&au_sbi(a->sb)->si_nowait);
30370 +       kfree(arg);
30371 +       if (unlikely(err))
30372 +               AuIOErr("err %d\n", err);
30373 +}
30374 +
30375 +static void kick_reinit_br_wh(struct super_block *sb, struct au_branch *br)
30376 +{
30377 +       int do_dec, wkq_err;
30378 +       struct reinit_br_wh *arg;
30379 +
30380 +       do_dec = 1;
30381 +       if (atomic_inc_return(&br->br_wbr->wbr_wh_running) != 1)
30382 +               goto out;
30383 +
30384 +       /* ignore ENOMEM */
30385 +       arg = kmalloc(sizeof(*arg), GFP_NOFS);
30386 +       if (arg) {
30387 +               /*
30388 +                * dec(wh_running), kfree(arg) and dec(br_count)
30389 +                * in reinit function
30390 +                */
30391 +               arg->sb = sb;
30392 +               arg->br = br;
30393 +               atomic_inc(&br->br_count);
30394 +               wkq_err = au_wkq_nowait(reinit_br_wh, arg, sb, /*flags*/0);
30395 +               if (unlikely(wkq_err)) {
30396 +                       atomic_dec(&br->br_wbr->wbr_wh_running);
30397 +                       atomic_dec(&br->br_count);
30398 +                       kfree(arg);
30399 +               }
30400 +               do_dec = 0;
30401 +       }
30402 +
30403 +out:
30404 +       if (do_dec)
30405 +               atomic_dec(&br->br_wbr->wbr_wh_running);
30406 +}
30407 +
30408 +/* ---------------------------------------------------------------------- */
30409 +
30410 +/*
30411 + * create the whiteout @wh.
30412 + */
30413 +static int link_or_create_wh(struct super_block *sb, aufs_bindex_t bindex,
30414 +                            struct dentry *wh)
30415 +{
30416 +       int err;
30417 +       struct path h_path = {
30418 +               .dentry = wh
30419 +       };
30420 +       struct au_branch *br;
30421 +       struct au_wbr *wbr;
30422 +       struct dentry *h_parent;
30423 +       struct inode *h_dir, *delegated;
30424 +
30425 +       h_parent = wh->d_parent; /* dir inode is locked */
30426 +       h_dir = h_parent->d_inode;
30427 +       IMustLock(h_dir);
30428 +
30429 +       br = au_sbr(sb, bindex);
30430 +       h_path.mnt = au_br_mnt(br);
30431 +       wbr = br->br_wbr;
30432 +       wbr_wh_read_lock(wbr);
30433 +       if (wbr->wbr_whbase) {
30434 +               delegated = NULL;
30435 +               err = vfsub_link(wbr->wbr_whbase, h_dir, &h_path, &delegated);
30436 +               if (unlikely(err == -EWOULDBLOCK)) {
30437 +                       pr_warn("cannot retry for NFSv4 delegation"
30438 +                               " for an internal link\n");
30439 +                       iput(delegated);
30440 +               }
30441 +               if (!err || err != -EMLINK)
30442 +                       goto out;
30443 +
30444 +               /* link count full. re-initialize br_whbase. */
30445 +               kick_reinit_br_wh(sb, br);
30446 +       }
30447 +
30448 +       /* return this error in this context */
30449 +       err = vfsub_create(h_dir, &h_path, WH_MASK, /*want_excl*/true);
30450 +
30451 +out:
30452 +       wbr_wh_read_unlock(wbr);
30453 +       return err;
30454 +}
30455 +
30456 +/* ---------------------------------------------------------------------- */
30457 +
30458 +/*
30459 + * create or remove the diropq.
30460 + */
30461 +static struct dentry *do_diropq(struct dentry *dentry, aufs_bindex_t bindex,
30462 +                               unsigned int flags)
30463 +{
30464 +       struct dentry *opq_dentry, *h_dentry;
30465 +       struct super_block *sb;
30466 +       struct au_branch *br;
30467 +       int err;
30468 +
30469 +       sb = dentry->d_sb;
30470 +       br = au_sbr(sb, bindex);
30471 +       h_dentry = au_h_dptr(dentry, bindex);
30472 +       opq_dentry = vfsub_lkup_one(&diropq_name, h_dentry);
30473 +       if (IS_ERR(opq_dentry))
30474 +               goto out;
30475 +
30476 +       if (au_ftest_diropq(flags, CREATE)) {
30477 +               err = link_or_create_wh(sb, bindex, opq_dentry);
30478 +               if (!err) {
30479 +                       au_set_dbdiropq(dentry, bindex);
30480 +                       goto out; /* success */
30481 +               }
30482 +       } else {
30483 +               struct path tmp = {
30484 +                       .dentry = opq_dentry,
30485 +                       .mnt    = au_br_mnt(br)
30486 +               };
30487 +               err = do_unlink_wh(au_h_iptr(dentry->d_inode, bindex), &tmp);
30488 +               if (!err)
30489 +                       au_set_dbdiropq(dentry, -1);
30490 +       }
30491 +       dput(opq_dentry);
30492 +       opq_dentry = ERR_PTR(err);
30493 +
30494 +out:
30495 +       return opq_dentry;
30496 +}
30497 +
30498 +struct do_diropq_args {
30499 +       struct dentry **errp;
30500 +       struct dentry *dentry;
30501 +       aufs_bindex_t bindex;
30502 +       unsigned int flags;
30503 +};
30504 +
30505 +static void call_do_diropq(void *args)
30506 +{
30507 +       struct do_diropq_args *a = args;
30508 +       *a->errp = do_diropq(a->dentry, a->bindex, a->flags);
30509 +}
30510 +
30511 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
30512 +                            unsigned int flags)
30513 +{
30514 +       struct dentry *diropq, *h_dentry;
30515 +
30516 +       h_dentry = au_h_dptr(dentry, bindex);
30517 +       if (!au_test_h_perm_sio(h_dentry->d_inode, MAY_EXEC | MAY_WRITE))
30518 +               diropq = do_diropq(dentry, bindex, flags);
30519 +       else {
30520 +               int wkq_err;
30521 +               struct do_diropq_args args = {
30522 +                       .errp           = &diropq,
30523 +                       .dentry         = dentry,
30524 +                       .bindex         = bindex,
30525 +                       .flags          = flags
30526 +               };
30527 +
30528 +               wkq_err = au_wkq_wait(call_do_diropq, &args);
30529 +               if (unlikely(wkq_err))
30530 +                       diropq = ERR_PTR(wkq_err);
30531 +       }
30532 +
30533 +       return diropq;
30534 +}
30535 +
30536 +/* ---------------------------------------------------------------------- */
30537 +
30538 +/*
30539 + * lookup whiteout dentry.
30540 + * @h_parent: lower parent dentry which must exist and be locked
30541 + * @base_name: name of dentry which will be whiteouted
30542 + * returns dentry for whiteout.
30543 + */
30544 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
30545 +                         struct au_branch *br)
30546 +{
30547 +       int err;
30548 +       struct qstr wh_name;
30549 +       struct dentry *wh_dentry;
30550 +
30551 +       err = au_wh_name_alloc(&wh_name, base_name);
30552 +       wh_dentry = ERR_PTR(err);
30553 +       if (!err) {
30554 +               wh_dentry = vfsub_lkup_one(&wh_name, h_parent);
30555 +               kfree(wh_name.name);
30556 +       }
30557 +       return wh_dentry;
30558 +}
30559 +
30560 +/*
30561 + * link/create a whiteout for @dentry on @bindex.
30562 + */
30563 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
30564 +                           struct dentry *h_parent)
30565 +{
30566 +       struct dentry *wh_dentry;
30567 +       struct super_block *sb;
30568 +       int err;
30569 +
30570 +       sb = dentry->d_sb;
30571 +       wh_dentry = au_wh_lkup(h_parent, &dentry->d_name, au_sbr(sb, bindex));
30572 +       if (!IS_ERR(wh_dentry) && !wh_dentry->d_inode) {
30573 +               err = link_or_create_wh(sb, bindex, wh_dentry);
30574 +               if (!err)
30575 +                       au_set_dbwh(dentry, bindex);
30576 +               else {
30577 +                       dput(wh_dentry);
30578 +                       wh_dentry = ERR_PTR(err);
30579 +               }
30580 +       }
30581 +
30582 +       return wh_dentry;
30583 +}
30584 +
30585 +/* ---------------------------------------------------------------------- */
30586 +
30587 +/* Delete all whiteouts in this directory on branch bindex. */
30588 +static int del_wh_children(struct dentry *h_dentry, struct au_nhash *whlist,
30589 +                          aufs_bindex_t bindex, struct au_branch *br)
30590 +{
30591 +       int err;
30592 +       unsigned long ul, n;
30593 +       struct qstr wh_name;
30594 +       char *p;
30595 +       struct hlist_head *head;
30596 +       struct au_vdir_wh *pos;
30597 +       struct au_vdir_destr *str;
30598 +
30599 +       err = -ENOMEM;
30600 +       p = (void *)__get_free_page(GFP_NOFS);
30601 +       wh_name.name = p;
30602 +       if (unlikely(!wh_name.name))
30603 +               goto out;
30604 +
30605 +       err = 0;
30606 +       memcpy(p, AUFS_WH_PFX, AUFS_WH_PFX_LEN);
30607 +       p += AUFS_WH_PFX_LEN;
30608 +       n = whlist->nh_num;
30609 +       head = whlist->nh_head;
30610 +       for (ul = 0; !err && ul < n; ul++, head++) {
30611 +               hlist_for_each_entry(pos, head, wh_hash) {
30612 +                       if (pos->wh_bindex != bindex)
30613 +                               continue;
30614 +
30615 +                       str = &pos->wh_str;
30616 +                       if (str->len + AUFS_WH_PFX_LEN <= PATH_MAX) {
30617 +                               memcpy(p, str->name, str->len);
30618 +                               wh_name.len = AUFS_WH_PFX_LEN + str->len;
30619 +                               err = unlink_wh_name(h_dentry, &wh_name, br);
30620 +                               if (!err)
30621 +                                       continue;
30622 +                               break;
30623 +                       }
30624 +                       AuIOErr("whiteout name too long %.*s\n",
30625 +                               str->len, str->name);
30626 +                       err = -EIO;
30627 +                       break;
30628 +               }
30629 +       }
30630 +       free_page((unsigned long)wh_name.name);
30631 +
30632 +out:
30633 +       return err;
30634 +}
30635 +
30636 +struct del_wh_children_args {
30637 +       int *errp;
30638 +       struct dentry *h_dentry;
30639 +       struct au_nhash *whlist;
30640 +       aufs_bindex_t bindex;
30641 +       struct au_branch *br;
30642 +};
30643 +
30644 +static void call_del_wh_children(void *args)
30645 +{
30646 +       struct del_wh_children_args *a = args;
30647 +       *a->errp = del_wh_children(a->h_dentry, a->whlist, a->bindex, a->br);
30648 +}
30649 +
30650 +/* ---------------------------------------------------------------------- */
30651 +
30652 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp)
30653 +{
30654 +       struct au_whtmp_rmdir *whtmp;
30655 +       int err;
30656 +       unsigned int rdhash;
30657 +
30658 +       SiMustAnyLock(sb);
30659 +
30660 +       whtmp = kmalloc(sizeof(*whtmp), gfp);
30661 +       if (unlikely(!whtmp)) {
30662 +               whtmp = ERR_PTR(-ENOMEM);
30663 +               goto out;
30664 +       }
30665 +
30666 +       whtmp->dir = NULL;
30667 +       whtmp->br = NULL;
30668 +       whtmp->wh_dentry = NULL;
30669 +       /* no estimation for dir size */
30670 +       rdhash = au_sbi(sb)->si_rdhash;
30671 +       if (!rdhash)
30672 +               rdhash = AUFS_RDHASH_DEF;
30673 +       err = au_nhash_alloc(&whtmp->whlist, rdhash, gfp);
30674 +       if (unlikely(err)) {
30675 +               kfree(whtmp);
30676 +               whtmp = ERR_PTR(err);
30677 +       }
30678 +
30679 +out:
30680 +       return whtmp;
30681 +}
30682 +
30683 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp)
30684 +{
30685 +       if (whtmp->br)
30686 +               atomic_dec(&whtmp->br->br_count);
30687 +       dput(whtmp->wh_dentry);
30688 +       iput(whtmp->dir);
30689 +       au_nhash_wh_free(&whtmp->whlist);
30690 +       kfree(whtmp);
30691 +}
30692 +
30693 +/*
30694 + * rmdir the whiteouted temporary named dir @h_dentry.
30695 + * @whlist: whiteouted children.
30696 + */
30697 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
30698 +                  struct dentry *wh_dentry, struct au_nhash *whlist)
30699 +{
30700 +       int err;
30701 +       struct path h_tmp;
30702 +       struct inode *wh_inode, *h_dir;
30703 +       struct au_branch *br;
30704 +
30705 +       h_dir = wh_dentry->d_parent->d_inode; /* dir inode is locked */
30706 +       IMustLock(h_dir);
30707 +
30708 +       br = au_sbr(dir->i_sb, bindex);
30709 +       wh_inode = wh_dentry->d_inode;
30710 +       mutex_lock_nested(&wh_inode->i_mutex, AuLsc_I_CHILD);
30711 +
30712 +       /*
30713 +        * someone else might change some whiteouts while we were sleeping.
30714 +        * it means this whlist may have an obsoleted entry.
30715 +        */
30716 +       if (!au_test_h_perm_sio(wh_inode, MAY_EXEC | MAY_WRITE))
30717 +               err = del_wh_children(wh_dentry, whlist, bindex, br);
30718 +       else {
30719 +               int wkq_err;
30720 +               struct del_wh_children_args args = {
30721 +                       .errp           = &err,
30722 +                       .h_dentry       = wh_dentry,
30723 +                       .whlist         = whlist,
30724 +                       .bindex         = bindex,
30725 +                       .br             = br
30726 +               };
30727 +
30728 +               wkq_err = au_wkq_wait(call_del_wh_children, &args);
30729 +               if (unlikely(wkq_err))
30730 +                       err = wkq_err;
30731 +       }
30732 +       mutex_unlock(&wh_inode->i_mutex);
30733 +
30734 +       if (!err) {
30735 +               h_tmp.dentry = wh_dentry;
30736 +               h_tmp.mnt = au_br_mnt(br);
30737 +               err = vfsub_rmdir(h_dir, &h_tmp);
30738 +       }
30739 +
30740 +       if (!err) {
30741 +               if (au_ibstart(dir) == bindex) {
30742 +                       /* todo: dir->i_mutex is necessary */
30743 +                       au_cpup_attr_timesizes(dir);
30744 +                       vfsub_drop_nlink(dir);
30745 +               }
30746 +               return 0; /* success */
30747 +       }
30748 +
30749 +       pr_warn("failed removing %pd(%d), ignored\n", wh_dentry, err);
30750 +       return err;
30751 +}
30752 +
30753 +static void call_rmdir_whtmp(void *args)
30754 +{
30755 +       int err;
30756 +       aufs_bindex_t bindex;
30757 +       struct au_whtmp_rmdir *a = args;
30758 +       struct super_block *sb;
30759 +       struct dentry *h_parent;
30760 +       struct inode *h_dir;
30761 +       struct au_hinode *hdir;
30762 +
30763 +       /* rmdir by nfsd may cause deadlock with this i_mutex */
30764 +       /* mutex_lock(&a->dir->i_mutex); */
30765 +       err = -EROFS;
30766 +       sb = a->dir->i_sb;
30767 +       si_read_lock(sb, !AuLock_FLUSH);
30768 +       if (!au_br_writable(a->br->br_perm))
30769 +               goto out;
30770 +       bindex = au_br_index(sb, a->br->br_id);
30771 +       if (unlikely(bindex < 0))
30772 +               goto out;
30773 +
30774 +       err = -EIO;
30775 +       ii_write_lock_parent(a->dir);
30776 +       h_parent = dget_parent(a->wh_dentry);
30777 +       h_dir = h_parent->d_inode;
30778 +       hdir = au_hi(a->dir, bindex);
30779 +       err = vfsub_mnt_want_write(au_br_mnt(a->br));
30780 +       if (unlikely(err))
30781 +               goto out_mnt;
30782 +       au_hn_imtx_lock_nested(hdir, AuLsc_I_PARENT);
30783 +       err = au_h_verify(a->wh_dentry, au_opt_udba(sb), h_dir, h_parent,
30784 +                         a->br);
30785 +       if (!err)
30786 +               err = au_whtmp_rmdir(a->dir, bindex, a->wh_dentry, &a->whlist);
30787 +       au_hn_imtx_unlock(hdir);
30788 +       vfsub_mnt_drop_write(au_br_mnt(a->br));
30789 +
30790 +out_mnt:
30791 +       dput(h_parent);
30792 +       ii_write_unlock(a->dir);
30793 +out:
30794 +       /* mutex_unlock(&a->dir->i_mutex); */
30795 +       au_whtmp_rmdir_free(a);
30796 +       si_read_unlock(sb);
30797 +       au_nwt_done(&au_sbi(sb)->si_nowait);
30798 +       if (unlikely(err))
30799 +               AuIOErr("err %d\n", err);
30800 +}
30801 +
30802 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
30803 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args)
30804 +{
30805 +       int wkq_err;
30806 +       struct super_block *sb;
30807 +
30808 +       IMustLock(dir);
30809 +
30810 +       /* all post-process will be done in do_rmdir_whtmp(). */
30811 +       sb = dir->i_sb;
30812 +       args->dir = au_igrab(dir);
30813 +       args->br = au_sbr(sb, bindex);
30814 +       atomic_inc(&args->br->br_count);
30815 +       args->wh_dentry = dget(wh_dentry);
30816 +       wkq_err = au_wkq_nowait(call_rmdir_whtmp, args, sb, /*flags*/0);
30817 +       if (unlikely(wkq_err)) {
30818 +               pr_warn("rmdir error %pd (%d), ignored\n", wh_dentry, wkq_err);
30819 +               au_whtmp_rmdir_free(args);
30820 +       }
30821 +}
30822 diff -urN /usr/share/empty/fs/aufs/whout.h linux/fs/aufs/whout.h
30823 --- /usr/share/empty/fs/aufs/whout.h    1970-01-01 01:00:00.000000000 +0100
30824 +++ linux/fs/aufs/whout.h       2014-04-24 22:11:10.861935852 +0200
30825 @@ -0,0 +1,86 @@
30826 +/*
30827 + * Copyright (C) 2005-2014 Junjiro R. Okajima
30828 + *
30829 + * This program, aufs is free software; you can redistribute it and/or modify
30830 + * it under the terms of the GNU General Public License as published by
30831 + * the Free Software Foundation; either version 2 of the License, or
30832 + * (at your option) any later version.
30833 + *
30834 + * This program is distributed in the hope that it will be useful,
30835 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30836 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30837 + * GNU General Public License for more details.
30838 + *
30839 + * You should have received a copy of the GNU General Public License
30840 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30841 + */
30842 +
30843 +/*
30844 + * whiteout for logical deletion and opaque directory
30845 + */
30846 +
30847 +#ifndef __AUFS_WHOUT_H__
30848 +#define __AUFS_WHOUT_H__
30849 +
30850 +#ifdef __KERNEL__
30851 +
30852 +#include "dir.h"
30853 +
30854 +/* whout.c */
30855 +int au_wh_name_alloc(struct qstr *wh, const struct qstr *name);
30856 +struct au_branch;
30857 +int au_wh_test(struct dentry *h_parent, struct qstr *wh_name,
30858 +              struct au_branch *br, int try_sio);
30859 +int au_diropq_test(struct dentry *h_dentry, struct au_branch *br);
30860 +struct dentry *au_whtmp_lkup(struct dentry *h_parent, struct au_branch *br,
30861 +                            struct qstr *prefix);
30862 +int au_whtmp_ren(struct dentry *h_dentry, struct au_branch *br);
30863 +int au_wh_unlink_dentry(struct inode *h_dir, struct path *h_path,
30864 +                       struct dentry *dentry);
30865 +int au_wh_init(struct au_branch *br, struct super_block *sb);
30866 +
30867 +/* diropq flags */
30868 +#define AuDiropq_CREATE        1
30869 +#define au_ftest_diropq(flags, name)   ((flags) & AuDiropq_##name)
30870 +#define au_fset_diropq(flags, name) \
30871 +       do { (flags) |= AuDiropq_##name; } while (0)
30872 +#define au_fclr_diropq(flags, name) \
30873 +       do { (flags) &= ~AuDiropq_##name; } while (0)
30874 +
30875 +struct dentry *au_diropq_sio(struct dentry *dentry, aufs_bindex_t bindex,
30876 +                            unsigned int flags);
30877 +struct dentry *au_wh_lkup(struct dentry *h_parent, struct qstr *base_name,
30878 +                         struct au_branch *br);
30879 +struct dentry *au_wh_create(struct dentry *dentry, aufs_bindex_t bindex,
30880 +                           struct dentry *h_parent);
30881 +
30882 +/* real rmdir for the whiteout-ed dir */
30883 +struct au_whtmp_rmdir {
30884 +       struct inode *dir;
30885 +       struct au_branch *br;
30886 +       struct dentry *wh_dentry;
30887 +       struct au_nhash whlist;
30888 +};
30889 +
30890 +struct au_whtmp_rmdir *au_whtmp_rmdir_alloc(struct super_block *sb, gfp_t gfp);
30891 +void au_whtmp_rmdir_free(struct au_whtmp_rmdir *whtmp);
30892 +int au_whtmp_rmdir(struct inode *dir, aufs_bindex_t bindex,
30893 +                  struct dentry *wh_dentry, struct au_nhash *whlist);
30894 +void au_whtmp_kick_rmdir(struct inode *dir, aufs_bindex_t bindex,
30895 +                        struct dentry *wh_dentry, struct au_whtmp_rmdir *args);
30896 +
30897 +/* ---------------------------------------------------------------------- */
30898 +
30899 +static inline struct dentry *au_diropq_create(struct dentry *dentry,
30900 +                                             aufs_bindex_t bindex)
30901 +{
30902 +       return au_diropq_sio(dentry, bindex, AuDiropq_CREATE);
30903 +}
30904 +
30905 +static inline int au_diropq_remove(struct dentry *dentry, aufs_bindex_t bindex)
30906 +{
30907 +       return PTR_ERR(au_diropq_sio(dentry, bindex, !AuDiropq_CREATE));
30908 +}
30909 +
30910 +#endif /* __KERNEL__ */
30911 +#endif /* __AUFS_WHOUT_H__ */
30912 diff -urN /usr/share/empty/fs/aufs/wkq.c linux/fs/aufs/wkq.c
30913 --- /usr/share/empty/fs/aufs/wkq.c      1970-01-01 01:00:00.000000000 +0100
30914 +++ linux/fs/aufs/wkq.c 2014-04-24 22:11:10.861935852 +0200
30915 @@ -0,0 +1,212 @@
30916 +/*
30917 + * Copyright (C) 2005-2014 Junjiro R. Okajima
30918 + *
30919 + * This program, aufs is free software; you can redistribute it and/or modify
30920 + * it under the terms of the GNU General Public License as published by
30921 + * the Free Software Foundation; either version 2 of the License, or
30922 + * (at your option) any later version.
30923 + *
30924 + * This program is distributed in the hope that it will be useful,
30925 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
30926 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
30927 + * GNU General Public License for more details.
30928 + *
30929 + * You should have received a copy of the GNU General Public License
30930 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
30931 + */
30932 +
30933 +/*
30934 + * workqueue for asynchronous/super-io operations
30935 + * todo: try new dredential scheme
30936 + */
30937 +
30938 +#include <linux/module.h>
30939 +#include "aufs.h"
30940 +
30941 +/* internal workqueue named AUFS_WKQ_NAME */
30942 +
30943 +static struct workqueue_struct *au_wkq;
30944 +
30945 +struct au_wkinfo {
30946 +       struct work_struct wk;
30947 +       struct kobject *kobj;
30948 +
30949 +       unsigned int flags; /* see wkq.h */
30950 +
30951 +       au_wkq_func_t func;
30952 +       void *args;
30953 +
30954 +       struct completion *comp;
30955 +};
30956 +
30957 +/* ---------------------------------------------------------------------- */
30958 +
30959 +static void wkq_func(struct work_struct *wk)
30960 +{
30961 +       struct au_wkinfo *wkinfo = container_of(wk, struct au_wkinfo, wk);
30962 +
30963 +       AuDebugOn(!uid_eq(current_fsuid(), GLOBAL_ROOT_UID));
30964 +       AuDebugOn(rlimit(RLIMIT_FSIZE) != RLIM_INFINITY);
30965 +
30966 +       wkinfo->func(wkinfo->args);
30967 +       if (au_ftest_wkq(wkinfo->flags, WAIT))
30968 +               complete(wkinfo->comp);
30969 +       else {
30970 +               kobject_put(wkinfo->kobj);
30971 +               module_put(THIS_MODULE); /* todo: ?? */
30972 +               kfree(wkinfo);
30973 +       }
30974 +}
30975 +
30976 +/*
30977 + * Since struct completion is large, try allocating it dynamically.
30978 + */
30979 +#if 1 /* defined(CONFIG_4KSTACKS) || defined(AuTest4KSTACKS) */
30980 +#define AuWkqCompDeclare(name) struct completion *comp = NULL
30981 +
30982 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
30983 +{
30984 +       *comp = kmalloc(sizeof(**comp), GFP_NOFS);
30985 +       if (*comp) {
30986 +               init_completion(*comp);
30987 +               wkinfo->comp = *comp;
30988 +               return 0;
30989 +       }
30990 +       return -ENOMEM;
30991 +}
30992 +
30993 +static void au_wkq_comp_free(struct completion *comp)
30994 +{
30995 +       kfree(comp);
30996 +}
30997 +
30998 +#else
30999 +
31000 +/* no braces */
31001 +#define AuWkqCompDeclare(name) \
31002 +       DECLARE_COMPLETION_ONSTACK(_ ## name); \
31003 +       struct completion *comp = &_ ## name
31004 +
31005 +static int au_wkq_comp_alloc(struct au_wkinfo *wkinfo, struct completion **comp)
31006 +{
31007 +       wkinfo->comp = *comp;
31008 +       return 0;
31009 +}
31010 +
31011 +static void au_wkq_comp_free(struct completion *comp __maybe_unused)
31012 +{
31013 +       /* empty */
31014 +}
31015 +#endif /* 4KSTACKS */
31016 +
31017 +static void au_wkq_run(struct au_wkinfo *wkinfo)
31018 +{
31019 +       if (au_ftest_wkq(wkinfo->flags, NEST)) {
31020 +               if (au_wkq_test()) {
31021 +                       AuWarn1("wkq from wkq, due to a dead dir by UDBA?\n");
31022 +                       AuDebugOn(au_ftest_wkq(wkinfo->flags, WAIT));
31023 +               }
31024 +       } else
31025 +               au_dbg_verify_kthread();
31026 +
31027 +       if (au_ftest_wkq(wkinfo->flags, WAIT)) {
31028 +               INIT_WORK_ONSTACK(&wkinfo->wk, wkq_func);
31029 +               queue_work(au_wkq, &wkinfo->wk);
31030 +       } else {
31031 +               INIT_WORK(&wkinfo->wk, wkq_func);
31032 +               schedule_work(&wkinfo->wk);
31033 +       }
31034 +}
31035 +
31036 +/*
31037 + * Be careful. It is easy to make deadlock happen.
31038 + * processA: lock, wkq and wait
31039 + * processB: wkq and wait, lock in wkq
31040 + * --> deadlock
31041 + */
31042 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args)
31043 +{
31044 +       int err;
31045 +       AuWkqCompDeclare(comp);
31046 +       struct au_wkinfo wkinfo = {
31047 +               .flags  = flags,
31048 +               .func   = func,
31049 +               .args   = args
31050 +       };
31051 +
31052 +       err = au_wkq_comp_alloc(&wkinfo, &comp);
31053 +       if (!err) {
31054 +               au_wkq_run(&wkinfo);
31055 +               /* no timeout, no interrupt */
31056 +               wait_for_completion(wkinfo.comp);
31057 +               au_wkq_comp_free(comp);
31058 +               destroy_work_on_stack(&wkinfo.wk);
31059 +       }
31060 +
31061 +       return err;
31062 +
31063 +}
31064 +
31065 +/*
31066 + * Note: dget/dput() in func for aufs dentries are not supported. It will be a
31067 + * problem in a concurrent umounting.
31068 + */
31069 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
31070 +                 unsigned int flags)
31071 +{
31072 +       int err;
31073 +       struct au_wkinfo *wkinfo;
31074 +
31075 +       atomic_inc(&au_sbi(sb)->si_nowait.nw_len);
31076 +
31077 +       /*
31078 +        * wkq_func() must free this wkinfo.
31079 +        * it highly depends upon the implementation of workqueue.
31080 +        */
31081 +       err = 0;
31082 +       wkinfo = kmalloc(sizeof(*wkinfo), GFP_NOFS);
31083 +       if (wkinfo) {
31084 +               wkinfo->kobj = &au_sbi(sb)->si_kobj;
31085 +               wkinfo->flags = flags & ~AuWkq_WAIT;
31086 +               wkinfo->func = func;
31087 +               wkinfo->args = args;
31088 +               wkinfo->comp = NULL;
31089 +               kobject_get(wkinfo->kobj);
31090 +               __module_get(THIS_MODULE); /* todo: ?? */
31091 +
31092 +               au_wkq_run(wkinfo);
31093 +       } else {
31094 +               err = -ENOMEM;
31095 +               au_nwt_done(&au_sbi(sb)->si_nowait);
31096 +       }
31097 +
31098 +       return err;
31099 +}
31100 +
31101 +/* ---------------------------------------------------------------------- */
31102 +
31103 +void au_nwt_init(struct au_nowait_tasks *nwt)
31104 +{
31105 +       atomic_set(&nwt->nw_len, 0);
31106 +       /* smp_mb(); */ /* atomic_set */
31107 +       init_waitqueue_head(&nwt->nw_wq);
31108 +}
31109 +
31110 +void au_wkq_fin(void)
31111 +{
31112 +       destroy_workqueue(au_wkq);
31113 +}
31114 +
31115 +int __init au_wkq_init(void)
31116 +{
31117 +       int err;
31118 +
31119 +       err = 0;
31120 +       au_wkq = alloc_workqueue(AUFS_WKQ_NAME, 0, WQ_DFL_ACTIVE);
31121 +       if (IS_ERR(au_wkq))
31122 +               err = PTR_ERR(au_wkq);
31123 +       else if (!au_wkq)
31124 +               err = -ENOMEM;
31125 +
31126 +       return err;
31127 +}
31128 diff -urN /usr/share/empty/fs/aufs/wkq.h linux/fs/aufs/wkq.h
31129 --- /usr/share/empty/fs/aufs/wkq.h      1970-01-01 01:00:00.000000000 +0100
31130 +++ linux/fs/aufs/wkq.h 2014-04-24 22:11:10.861935852 +0200
31131 @@ -0,0 +1,91 @@
31132 +/*
31133 + * Copyright (C) 2005-2014 Junjiro R. Okajima
31134 + *
31135 + * This program, aufs is free software; you can redistribute it and/or modify
31136 + * it under the terms of the GNU General Public License as published by
31137 + * the Free Software Foundation; either version 2 of the License, or
31138 + * (at your option) any later version.
31139 + *
31140 + * This program is distributed in the hope that it will be useful,
31141 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31142 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31143 + * GNU General Public License for more details.
31144 + *
31145 + * You should have received a copy of the GNU General Public License
31146 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31147 + */
31148 +
31149 +/*
31150 + * workqueue for asynchronous/super-io operations
31151 + * todo: try new credentials management scheme
31152 + */
31153 +
31154 +#ifndef __AUFS_WKQ_H__
31155 +#define __AUFS_WKQ_H__
31156 +
31157 +#ifdef __KERNEL__
31158 +
31159 +struct super_block;
31160 +
31161 +/* ---------------------------------------------------------------------- */
31162 +
31163 +/*
31164 + * in the next operation, wait for the 'nowait' tasks in system-wide workqueue
31165 + */
31166 +struct au_nowait_tasks {
31167 +       atomic_t                nw_len;
31168 +       wait_queue_head_t       nw_wq;
31169 +};
31170 +
31171 +/* ---------------------------------------------------------------------- */
31172 +
31173 +typedef void (*au_wkq_func_t)(void *args);
31174 +
31175 +/* wkq flags */
31176 +#define AuWkq_WAIT     1
31177 +#define AuWkq_NEST     (1 << 1)
31178 +#define au_ftest_wkq(flags, name)      ((flags) & AuWkq_##name)
31179 +#define au_fset_wkq(flags, name) \
31180 +       do { (flags) |= AuWkq_##name; } while (0)
31181 +#define au_fclr_wkq(flags, name) \
31182 +       do { (flags) &= ~AuWkq_##name; } while (0)
31183 +
31184 +#ifndef CONFIG_AUFS_HNOTIFY
31185 +#undef AuWkq_NEST
31186 +#define AuWkq_NEST     0
31187 +#endif
31188 +
31189 +/* wkq.c */
31190 +int au_wkq_do_wait(unsigned int flags, au_wkq_func_t func, void *args);
31191 +int au_wkq_nowait(au_wkq_func_t func, void *args, struct super_block *sb,
31192 +                 unsigned int flags);
31193 +void au_nwt_init(struct au_nowait_tasks *nwt);
31194 +int __init au_wkq_init(void);
31195 +void au_wkq_fin(void);
31196 +
31197 +/* ---------------------------------------------------------------------- */
31198 +
31199 +static inline int au_wkq_test(void)
31200 +{
31201 +       return current->flags & PF_WQ_WORKER;
31202 +}
31203 +
31204 +static inline int au_wkq_wait(au_wkq_func_t func, void *args)
31205 +{
31206 +       return au_wkq_do_wait(AuWkq_WAIT, func, args);
31207 +}
31208 +
31209 +static inline void au_nwt_done(struct au_nowait_tasks *nwt)
31210 +{
31211 +       if (atomic_dec_and_test(&nwt->nw_len))
31212 +               wake_up_all(&nwt->nw_wq);
31213 +}
31214 +
31215 +static inline int au_nwt_flush(struct au_nowait_tasks *nwt)
31216 +{
31217 +       wait_event(nwt->nw_wq, !atomic_read(&nwt->nw_len));
31218 +       return 0;
31219 +}
31220 +
31221 +#endif /* __KERNEL__ */
31222 +#endif /* __AUFS_WKQ_H__ */
31223 diff -urN /usr/share/empty/fs/aufs/xino.c linux/fs/aufs/xino.c
31224 --- /usr/share/empty/fs/aufs/xino.c     1970-01-01 01:00:00.000000000 +0100
31225 +++ linux/fs/aufs/xino.c        2014-04-24 22:11:10.861935852 +0200
31226 @@ -0,0 +1,1314 @@
31227 +/*
31228 + * Copyright (C) 2005-2014 Junjiro R. Okajima
31229 + *
31230 + * This program, aufs is free software; you can redistribute it and/or modify
31231 + * it under the terms of the GNU General Public License as published by
31232 + * the Free Software Foundation; either version 2 of the License, or
31233 + * (at your option) any later version.
31234 + *
31235 + * This program is distributed in the hope that it will be useful,
31236 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
31237 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
31238 + * GNU General Public License for more details.
31239 + *
31240 + * You should have received a copy of the GNU General Public License
31241 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
31242 + */
31243 +
31244 +/*
31245 + * external inode number translation table and bitmap
31246 + */
31247 +
31248 +#include <linux/seq_file.h>
31249 +#include <linux/statfs.h>
31250 +#include "aufs.h"
31251 +
31252 +/* todo: unnecessary to support mmap_sem since kernel-space? */
31253 +ssize_t xino_fread(au_readf_t func, struct file *file, void *kbuf, size_t size,
31254 +                  loff_t *pos)
31255 +{
31256 +       ssize_t err;
31257 +       mm_segment_t oldfs;
31258 +       union {
31259 +               void *k;
31260 +               char __user *u;
31261 +       } buf;
31262 +
31263 +       buf.k = kbuf;
31264 +       oldfs = get_fs();
31265 +       set_fs(KERNEL_DS);
31266 +       do {
31267 +               /* todo: signal_pending? */
31268 +               err = func(file, buf.u, size, pos);
31269 +       } while (err == -EAGAIN || err == -EINTR);
31270 +       set_fs(oldfs);
31271 +
31272 +#if 0 /* reserved for future use */
31273 +       if (err > 0)
31274 +               fsnotify_access(file->f_dentry);
31275 +#endif
31276 +
31277 +       return err;
31278 +}
31279 +
31280 +/* ---------------------------------------------------------------------- */
31281 +
31282 +static ssize_t do_xino_fwrite(au_writef_t func, struct file *file, void *kbuf,
31283 +                             size_t size, loff_t *pos)
31284 +{
31285 +       ssize_t err;
31286 +       mm_segment_t oldfs;
31287 +       union {
31288 +               void *k;
31289 +               const char __user *u;
31290 +       } buf;
31291 +
31292 +       buf.k = kbuf;
31293 +       oldfs = get_fs();
31294 +       set_fs(KERNEL_DS);
31295 +       do {
31296 +               /* todo: signal_pending? */
31297 +               err = func(file, buf.u, size, pos);
31298 +       } while (err == -EAGAIN || err == -EINTR);
31299 +       set_fs(oldfs);
31300 +
31301 +#if 0 /* reserved for future use */
31302 +       if (err > 0)
31303 +               fsnotify_modify(file->f_dentry);
31304 +#endif
31305 +
31306 +       return err;
31307 +}
31308 +
31309 +struct do_xino_fwrite_args {
31310 +       ssize_t *errp;
31311 +       au_writef_t func;
31312 +       struct file *file;
31313 +       void *buf;
31314 +       size_t size;
31315 +       loff_t *pos;
31316 +};
31317 +
31318 +static void call_do_xino_fwrite(void *args)
31319 +{
31320 +       struct do_xino_fwrite_args *a = args;
31321 +       *a->errp = do_xino_fwrite(a->func, a->file, a->buf, a->size, a->pos);
31322 +}
31323 +
31324 +ssize_t xino_fwrite(au_writef_t func, struct file *file, void *buf, size_t size,
31325 +                   loff_t *pos)
31326 +{
31327 +       ssize_t err;
31328 +
31329 +       /* todo: signal block and no wkq? */
31330 +       if (rlimit(RLIMIT_FSIZE) == RLIM_INFINITY) {
31331 +               lockdep_off();
31332 +               err = do_xino_fwrite(func, file, buf, size, pos);
31333 +               lockdep_on();
31334 +       } else {
31335 +               /*
31336 +                * it breaks RLIMIT_FSIZE and normal user's limit,
31337 +                * users should care about quota and real 'filesystem full.'
31338 +                */
31339 +               int wkq_err;
31340 +               struct do_xino_fwrite_args args = {
31341 +                       .errp   = &err,
31342 +                       .func   = func,
31343 +                       .file   = file,
31344 +                       .buf    = buf,
31345 +                       .size   = size,
31346 +                       .pos    = pos
31347 +               };
31348 +
31349 +               wkq_err = au_wkq_wait(call_do_xino_fwrite, &args);
31350 +               if (unlikely(wkq_err))
31351 +                       err = wkq_err;
31352 +       }
31353 +
31354 +       return err;
31355 +}
31356 +
31357 +/* ---------------------------------------------------------------------- */
31358 +
31359 +/*
31360 + * create a new xinofile at the same place/path as @base_file.
31361 + */
31362 +struct file *au_xino_create2(struct file *base_file, struct file *copy_src)
31363 +{
31364 +       struct file *file;
31365 +       struct dentry *base, *parent;
31366 +       struct inode *dir, *delegated;
31367 +       struct qstr *name;
31368 +       struct path path;
31369 +       int err;
31370 +
31371 +       base = base_file->f_dentry;
31372 +       parent = base->d_parent; /* dir inode is locked */
31373 +       dir = parent->d_inode;
31374 +       IMustLock(dir);
31375 +
31376 +       file = ERR_PTR(-EINVAL);
31377 +       name = &base->d_name;
31378 +       path.dentry = vfsub_lookup_one_len(name->name, parent, name->len);
31379 +       if (IS_ERR(path.dentry)) {
31380 +               file = (void *)path.dentry;
31381 +               pr_err("%pd lookup err %ld\n",
31382 +                      base, PTR_ERR(path.dentry));
31383 +               goto out;
31384 +       }
31385 +
31386 +       /* no need to mnt_want_write() since we call dentry_open() later */
31387 +       err = vfs_create(dir, path.dentry, S_IRUGO | S_IWUGO, NULL);
31388 +       if (unlikely(err)) {
31389 +               file = ERR_PTR(err);
31390 +               pr_err("%pd create err %d\n", base, err);
31391 +               goto out_dput;
31392 +       }
31393 +
31394 +       path.mnt = base_file->f_path.mnt;
31395 +       file = vfsub_dentry_open(&path,
31396 +                                O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
31397 +                                /* | __FMODE_NONOTIFY */);
31398 +       if (IS_ERR(file)) {
31399 +               pr_err("%pd open err %ld\n", base, PTR_ERR(file));
31400 +               goto out_dput;
31401 +       }
31402 +
31403 +       delegated = NULL;
31404 +       err = vfsub_unlink(dir, &file->f_path, &delegated, /*force*/0);
31405 +       if (unlikely(err == -EWOULDBLOCK)) {
31406 +               pr_warn("cannot retry for NFSv4 delegation"
31407 +                       " for an internal unlink\n");
31408 +               iput(delegated);
31409 +       }
31410 +       if (unlikely(err)) {
31411 +               pr_err("%pd unlink err %d\n", base, err);
31412 +               goto out_fput;
31413 +       }
31414 +
31415 +       if (copy_src) {
31416 +               /* no one can touch copy_src xino */
31417 +               err = au_copy_file(file, copy_src, vfsub_f_size_read(copy_src));
31418 +               if (unlikely(err)) {
31419 +                       pr_err("%pd copy err %d\n", base, err);
31420 +                       goto out_fput;
31421 +               }
31422 +       }
31423 +       goto out_dput; /* success */
31424 +
31425 +out_fput:
31426 +       fput(file);
31427 +       file = ERR_PTR(err);
31428 +out_dput:
31429 +       dput(path.dentry);
31430 +out:
31431 +       return file;
31432 +}
31433 +
31434 +struct au_xino_lock_dir {
31435 +       struct au_hinode *hdir;
31436 +       struct dentry *parent;
31437 +       struct mutex *mtx;
31438 +};
31439 +
31440 +static void au_xino_lock_dir(struct super_block *sb, struct file *xino,
31441 +                            struct au_xino_lock_dir *ldir)
31442 +{
31443 +       aufs_bindex_t brid, bindex;
31444 +
31445 +       ldir->hdir = NULL;
31446 +       bindex = -1;
31447 +       brid = au_xino_brid(sb);
31448 +       if (brid >= 0)
31449 +               bindex = au_br_index(sb, brid);
31450 +       if (bindex >= 0) {
31451 +               ldir->hdir = au_hi(sb->s_root->d_inode, bindex);
31452 +               au_hn_imtx_lock_nested(ldir->hdir, AuLsc_I_PARENT);
31453 +       } else {
31454 +               ldir->parent = dget_parent(xino->f_dentry);
31455 +               ldir->mtx = &ldir->parent->d_inode->i_mutex;
31456 +               mutex_lock_nested(ldir->mtx, AuLsc_I_PARENT);
31457 +       }
31458 +}
31459 +
31460 +static void au_xino_unlock_dir(struct au_xino_lock_dir *ldir)
31461 +{
31462 +       if (ldir->hdir)
31463 +               au_hn_imtx_unlock(ldir->hdir);
31464 +       else {
31465 +               mutex_unlock(ldir->mtx);
31466 +               dput(ldir->parent);
31467 +       }
31468 +}
31469 +
31470 +/* ---------------------------------------------------------------------- */
31471 +
31472 +/* trucate xino files asynchronously */
31473 +
31474 +int au_xino_trunc(struct super_block *sb, aufs_bindex_t bindex)
31475 +{
31476 +       int err;
31477 +       unsigned long jiffy;
31478 +       blkcnt_t blocks;
31479 +       aufs_bindex_t bi, bend;
31480 +       struct kstatfs *st;
31481 +       struct au_branch *br;
31482 +       struct file *new_xino, *file;
31483 +       struct super_block *h_sb;
31484 +       struct au_xino_lock_dir ldir;
31485 +
31486 +       err = -ENOMEM;
31487 +       st = kzalloc(sizeof(*st), GFP_NOFS);
31488 +       if (unlikely(!st))
31489 +               goto out;
31490 +
31491 +       err = -EINVAL;
31492 +       bend = au_sbend(sb);
31493 +       if (unlikely(bindex < 0 || bend < bindex))
31494 +               goto out_st;
31495 +       br = au_sbr(sb, bindex);
31496 +       file = br->br_xino.xi_file;
31497 +       if (!file)
31498 +               goto out_st;
31499 +
31500 +       err = vfs_statfs(&file->f_path, st);
31501 +       if (unlikely(err))
31502 +               AuErr1("statfs err %d, ignored\n", err);
31503 +       jiffy = jiffies;
31504 +       blocks = file_inode(file)->i_blocks;
31505 +       pr_info("begin truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
31506 +               bindex, (u64)blocks, st->f_bfree, st->f_blocks);
31507 +
31508 +       au_xino_lock_dir(sb, file, &ldir);
31509 +       /* mnt_want_write() is unnecessary here */
31510 +       new_xino = au_xino_create2(file, file);
31511 +       au_xino_unlock_dir(&ldir);
31512 +       err = PTR_ERR(new_xino);
31513 +       if (IS_ERR(new_xino)) {
31514 +               pr_err("err %d, ignored\n", err);
31515 +               goto out_st;
31516 +       }
31517 +       err = 0;
31518 +       fput(file);
31519 +       br->br_xino.xi_file = new_xino;
31520 +
31521 +       h_sb = au_br_sb(br);
31522 +       for (bi = 0; bi <= bend; bi++) {
31523 +               if (unlikely(bi == bindex))
31524 +                       continue;
31525 +               br = au_sbr(sb, bi);
31526 +               if (au_br_sb(br) != h_sb)
31527 +                       continue;
31528 +
31529 +               fput(br->br_xino.xi_file);
31530 +               br->br_xino.xi_file = new_xino;
31531 +               get_file(new_xino);
31532 +       }
31533 +
31534 +       err = vfs_statfs(&new_xino->f_path, st);
31535 +       if (!err) {
31536 +               pr_info("end truncating xino(b%d), ib%llu, %llu/%llu free blks\n",
31537 +                       bindex, (u64)file_inode(new_xino)->i_blocks,
31538 +                       st->f_bfree, st->f_blocks);
31539 +               if (file_inode(new_xino)->i_blocks < blocks)
31540 +                       au_sbi(sb)->si_xino_jiffy = jiffy;
31541 +       } else
31542 +               AuErr1("statfs err %d, ignored\n", err);
31543 +
31544 +out_st:
31545 +       kfree(st);
31546 +out:
31547 +       return err;
31548 +}
31549 +
31550 +struct xino_do_trunc_args {
31551 +       struct super_block *sb;
31552 +       struct au_branch *br;
31553 +};
31554 +
31555 +static void xino_do_trunc(void *_args)
31556 +{
31557 +       struct xino_do_trunc_args *args = _args;
31558 +       struct super_block *sb;
31559 +       struct au_branch *br;
31560 +       struct inode *dir;
31561 +       int err;
31562 +       aufs_bindex_t bindex;
31563 +
31564 +       err = 0;
31565 +       sb = args->sb;
31566 +       dir = sb->s_root->d_inode;
31567 +       br = args->br;
31568 +
31569 +       si_noflush_write_lock(sb);
31570 +       ii_read_lock_parent(dir);
31571 +       bindex = au_br_index(sb, br->br_id);
31572 +       err = au_xino_trunc(sb, bindex);
31573 +       ii_read_unlock(dir);
31574 +       if (unlikely(err))
31575 +               pr_warn("err b%d, (%d)\n", bindex, err);
31576 +       atomic_dec(&br->br_xino_running);
31577 +       atomic_dec(&br->br_count);
31578 +       si_write_unlock(sb);
31579 +       au_nwt_done(&au_sbi(sb)->si_nowait);
31580 +       kfree(args);
31581 +}
31582 +
31583 +static int xino_trunc_test(struct super_block *sb, struct au_branch *br)
31584 +{
31585 +       int err;
31586 +       struct kstatfs st;
31587 +       struct au_sbinfo *sbinfo;
31588 +
31589 +       /* todo: si_xino_expire and the ratio should be customizable */
31590 +       sbinfo = au_sbi(sb);
31591 +       if (time_before(jiffies,
31592 +                       sbinfo->si_xino_jiffy + sbinfo->si_xino_expire))
31593 +               return 0;
31594 +
31595 +       /* truncation border */
31596 +       err = vfs_statfs(&br->br_xino.xi_file->f_path, &st);
31597 +       if (unlikely(err)) {
31598 +               AuErr1("statfs err %d, ignored\n", err);
31599 +               return 0;
31600 +       }
31601 +       if (div64_u64(st.f_bfree * 100, st.f_blocks) >= AUFS_XINO_DEF_TRUNC)
31602 +               return 0;
31603 +
31604 +       return 1;
31605 +}
31606 +
31607 +static void xino_try_trunc(struct super_block *sb, struct au_branch *br)
31608 +{
31609 +       struct xino_do_trunc_args *args;
31610 +       int wkq_err;
31611 +
31612 +       if (!xino_trunc_test(sb, br))
31613 +               return;
31614 +
31615 +       if (atomic_inc_return(&br->br_xino_running) > 1)
31616 +               goto out;
31617 +
31618 +       /* lock and kfree() will be called in trunc_xino() */
31619 +       args = kmalloc(sizeof(*args), GFP_NOFS);
31620 +       if (unlikely(!args)) {
31621 +               AuErr1("no memory\n");
31622 +               goto out_args;
31623 +       }
31624 +
31625 +       atomic_inc(&br->br_count);
31626 +       args->sb = sb;
31627 +       args->br = br;
31628 +       wkq_err = au_wkq_nowait(xino_do_trunc, args, sb, /*flags*/0);
31629 +       if (!wkq_err)
31630 +               return; /* success */
31631 +
31632 +       pr_err("wkq %d\n", wkq_err);
31633 +       atomic_dec(&br->br_count);
31634 +
31635 +out_args:
31636 +       kfree(args);
31637 +out:
31638 +       atomic_dec(&br->br_xino_running);
31639 +}
31640 +
31641 +/* ---------------------------------------------------------------------- */
31642 +
31643 +static int au_xino_do_write(au_writef_t write, struct file *file,
31644 +                           ino_t h_ino, ino_t ino)
31645 +{
31646 +       loff_t pos;
31647 +       ssize_t sz;
31648 +
31649 +       pos = h_ino;
31650 +       if (unlikely(au_loff_max / sizeof(ino) - 1 < pos)) {
31651 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
31652 +               return -EFBIG;
31653 +       }
31654 +       pos *= sizeof(ino);
31655 +       sz = xino_fwrite(write, file, &ino, sizeof(ino), &pos);
31656 +       if (sz == sizeof(ino))
31657 +               return 0; /* success */
31658 +
31659 +       AuIOErr("write failed (%zd)\n", sz);
31660 +       return -EIO;
31661 +}
31662 +
31663 +/*
31664 + * write @ino to the xinofile for the specified branch{@sb, @bindex}
31665 + * at the position of @h_ino.
31666 + * even if @ino is zero, it is written to the xinofile and means no entry.
31667 + * if the size of the xino file on a specific filesystem exceeds the watermark,
31668 + * try truncating it.
31669 + */
31670 +int au_xino_write(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
31671 +                 ino_t ino)
31672 +{
31673 +       int err;
31674 +       unsigned int mnt_flags;
31675 +       struct au_branch *br;
31676 +
31677 +       BUILD_BUG_ON(sizeof(long long) != sizeof(au_loff_max)
31678 +                    || ((loff_t)-1) > 0);
31679 +       SiMustAnyLock(sb);
31680 +
31681 +       mnt_flags = au_mntflags(sb);
31682 +       if (!au_opt_test(mnt_flags, XINO))
31683 +               return 0;
31684 +
31685 +       br = au_sbr(sb, bindex);
31686 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
31687 +                              h_ino, ino);
31688 +       if (!err) {
31689 +               if (au_opt_test(mnt_flags, TRUNC_XINO)
31690 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
31691 +                       xino_try_trunc(sb, br);
31692 +               return 0; /* success */
31693 +       }
31694 +
31695 +       AuIOErr("write failed (%d)\n", err);
31696 +       return -EIO;
31697 +}
31698 +
31699 +/* ---------------------------------------------------------------------- */
31700 +
31701 +/* aufs inode number bitmap */
31702 +
31703 +static const int page_bits = (int)PAGE_SIZE * BITS_PER_BYTE;
31704 +static ino_t xib_calc_ino(unsigned long pindex, int bit)
31705 +{
31706 +       ino_t ino;
31707 +
31708 +       AuDebugOn(bit < 0 || page_bits <= bit);
31709 +       ino = AUFS_FIRST_INO + pindex * page_bits + bit;
31710 +       return ino;
31711 +}
31712 +
31713 +static void xib_calc_bit(ino_t ino, unsigned long *pindex, int *bit)
31714 +{
31715 +       AuDebugOn(ino < AUFS_FIRST_INO);
31716 +       ino -= AUFS_FIRST_INO;
31717 +       *pindex = ino / page_bits;
31718 +       *bit = ino % page_bits;
31719 +}
31720 +
31721 +static int xib_pindex(struct super_block *sb, unsigned long pindex)
31722 +{
31723 +       int err;
31724 +       loff_t pos;
31725 +       ssize_t sz;
31726 +       struct au_sbinfo *sbinfo;
31727 +       struct file *xib;
31728 +       unsigned long *p;
31729 +
31730 +       sbinfo = au_sbi(sb);
31731 +       MtxMustLock(&sbinfo->si_xib_mtx);
31732 +       AuDebugOn(pindex > ULONG_MAX / PAGE_SIZE
31733 +                 || !au_opt_test(sbinfo->si_mntflags, XINO));
31734 +
31735 +       if (pindex == sbinfo->si_xib_last_pindex)
31736 +               return 0;
31737 +
31738 +       xib = sbinfo->si_xib;
31739 +       p = sbinfo->si_xib_buf;
31740 +       pos = sbinfo->si_xib_last_pindex;
31741 +       pos *= PAGE_SIZE;
31742 +       sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
31743 +       if (unlikely(sz != PAGE_SIZE))
31744 +               goto out;
31745 +
31746 +       pos = pindex;
31747 +       pos *= PAGE_SIZE;
31748 +       if (vfsub_f_size_read(xib) >= pos + PAGE_SIZE)
31749 +               sz = xino_fread(sbinfo->si_xread, xib, p, PAGE_SIZE, &pos);
31750 +       else {
31751 +               memset(p, 0, PAGE_SIZE);
31752 +               sz = xino_fwrite(sbinfo->si_xwrite, xib, p, PAGE_SIZE, &pos);
31753 +       }
31754 +       if (sz == PAGE_SIZE) {
31755 +               sbinfo->si_xib_last_pindex = pindex;
31756 +               return 0; /* success */
31757 +       }
31758 +
31759 +out:
31760 +       AuIOErr1("write failed (%zd)\n", sz);
31761 +       err = sz;
31762 +       if (sz >= 0)
31763 +               err = -EIO;
31764 +       return err;
31765 +}
31766 +
31767 +/* ---------------------------------------------------------------------- */
31768 +
31769 +static void au_xib_clear_bit(struct inode *inode)
31770 +{
31771 +       int err, bit;
31772 +       unsigned long pindex;
31773 +       struct super_block *sb;
31774 +       struct au_sbinfo *sbinfo;
31775 +
31776 +       AuDebugOn(inode->i_nlink);
31777 +
31778 +       sb = inode->i_sb;
31779 +       xib_calc_bit(inode->i_ino, &pindex, &bit);
31780 +       AuDebugOn(page_bits <= bit);
31781 +       sbinfo = au_sbi(sb);
31782 +       mutex_lock(&sbinfo->si_xib_mtx);
31783 +       err = xib_pindex(sb, pindex);
31784 +       if (!err) {
31785 +               clear_bit(bit, sbinfo->si_xib_buf);
31786 +               sbinfo->si_xib_next_bit = bit;
31787 +       }
31788 +       mutex_unlock(&sbinfo->si_xib_mtx);
31789 +}
31790 +
31791 +/* for s_op->delete_inode() */
31792 +void au_xino_delete_inode(struct inode *inode, const int unlinked)
31793 +{
31794 +       int err;
31795 +       unsigned int mnt_flags;
31796 +       aufs_bindex_t bindex, bend, bi;
31797 +       unsigned char try_trunc;
31798 +       struct au_iinfo *iinfo;
31799 +       struct super_block *sb;
31800 +       struct au_hinode *hi;
31801 +       struct inode *h_inode;
31802 +       struct au_branch *br;
31803 +       au_writef_t xwrite;
31804 +
31805 +       sb = inode->i_sb;
31806 +       mnt_flags = au_mntflags(sb);
31807 +       if (!au_opt_test(mnt_flags, XINO)
31808 +           || inode->i_ino == AUFS_ROOT_INO)
31809 +               return;
31810 +
31811 +       if (unlinked) {
31812 +               au_xigen_inc(inode);
31813 +               au_xib_clear_bit(inode);
31814 +       }
31815 +
31816 +       iinfo = au_ii(inode);
31817 +       if (!iinfo)
31818 +               return;
31819 +
31820 +       bindex = iinfo->ii_bstart;
31821 +       if (bindex < 0)
31822 +               return;
31823 +
31824 +       xwrite = au_sbi(sb)->si_xwrite;
31825 +       try_trunc = !!au_opt_test(mnt_flags, TRUNC_XINO);
31826 +       hi = iinfo->ii_hinode + bindex;
31827 +       bend = iinfo->ii_bend;
31828 +       for (; bindex <= bend; bindex++, hi++) {
31829 +               h_inode = hi->hi_inode;
31830 +               if (!h_inode
31831 +                   || (!unlinked && h_inode->i_nlink))
31832 +                       continue;
31833 +
31834 +               /* inode may not be revalidated */
31835 +               bi = au_br_index(sb, hi->hi_id);
31836 +               if (bi < 0)
31837 +                       continue;
31838 +
31839 +               br = au_sbr(sb, bi);
31840 +               err = au_xino_do_write(xwrite, br->br_xino.xi_file,
31841 +                                      h_inode->i_ino, /*ino*/0);
31842 +               if (!err && try_trunc
31843 +                   && au_test_fs_trunc_xino(au_br_sb(br)))
31844 +                       xino_try_trunc(sb, br);
31845 +       }
31846 +}
31847 +
31848 +/* get an unused inode number from bitmap */
31849 +ino_t au_xino_new_ino(struct super_block *sb)
31850 +{
31851 +       ino_t ino;
31852 +       unsigned long *p, pindex, ul, pend;
31853 +       struct au_sbinfo *sbinfo;
31854 +       struct file *file;
31855 +       int free_bit, err;
31856 +
31857 +       if (!au_opt_test(au_mntflags(sb), XINO))
31858 +               return iunique(sb, AUFS_FIRST_INO);
31859 +
31860 +       sbinfo = au_sbi(sb);
31861 +       mutex_lock(&sbinfo->si_xib_mtx);
31862 +       p = sbinfo->si_xib_buf;
31863 +       free_bit = sbinfo->si_xib_next_bit;
31864 +       if (free_bit < page_bits && !test_bit(free_bit, p))
31865 +               goto out; /* success */
31866 +       free_bit = find_first_zero_bit(p, page_bits);
31867 +       if (free_bit < page_bits)
31868 +               goto out; /* success */
31869 +
31870 +       pindex = sbinfo->si_xib_last_pindex;
31871 +       for (ul = pindex - 1; ul < ULONG_MAX; ul--) {
31872 +               err = xib_pindex(sb, ul);
31873 +               if (unlikely(err))
31874 +                       goto out_err;
31875 +               free_bit = find_first_zero_bit(p, page_bits);
31876 +               if (free_bit < page_bits)
31877 +                       goto out; /* success */
31878 +       }
31879 +
31880 +       file = sbinfo->si_xib;
31881 +       pend = vfsub_f_size_read(file) / PAGE_SIZE;
31882 +       for (ul = pindex + 1; ul <= pend; ul++) {
31883 +               err = xib_pindex(sb, ul);
31884 +               if (unlikely(err))
31885 +                       goto out_err;
31886 +               free_bit = find_first_zero_bit(p, page_bits);
31887 +               if (free_bit < page_bits)
31888 +                       goto out; /* success */
31889 +       }
31890 +       BUG();
31891 +
31892 +out:
31893 +       set_bit(free_bit, p);
31894 +       sbinfo->si_xib_next_bit = free_bit + 1;
31895 +       pindex = sbinfo->si_xib_last_pindex;
31896 +       mutex_unlock(&sbinfo->si_xib_mtx);
31897 +       ino = xib_calc_ino(pindex, free_bit);
31898 +       AuDbg("i%lu\n", (unsigned long)ino);
31899 +       return ino;
31900 +out_err:
31901 +       mutex_unlock(&sbinfo->si_xib_mtx);
31902 +       AuDbg("i0\n");
31903 +       return 0;
31904 +}
31905 +
31906 +/*
31907 + * read @ino from xinofile for the specified branch{@sb, @bindex}
31908 + * at the position of @h_ino.
31909 + * if @ino does not exist and @do_new is true, get new one.
31910 + */
31911 +int au_xino_read(struct super_block *sb, aufs_bindex_t bindex, ino_t h_ino,
31912 +                ino_t *ino)
31913 +{
31914 +       int err;
31915 +       ssize_t sz;
31916 +       loff_t pos;
31917 +       struct file *file;
31918 +       struct au_sbinfo *sbinfo;
31919 +
31920 +       *ino = 0;
31921 +       if (!au_opt_test(au_mntflags(sb), XINO))
31922 +               return 0; /* no xino */
31923 +
31924 +       err = 0;
31925 +       sbinfo = au_sbi(sb);
31926 +       pos = h_ino;
31927 +       if (unlikely(au_loff_max / sizeof(*ino) - 1 < pos)) {
31928 +               AuIOErr1("too large hi%lu\n", (unsigned long)h_ino);
31929 +               return -EFBIG;
31930 +       }
31931 +       pos *= sizeof(*ino);
31932 +
31933 +       file = au_sbr(sb, bindex)->br_xino.xi_file;
31934 +       if (vfsub_f_size_read(file) < pos + sizeof(*ino))
31935 +               return 0; /* no ino */
31936 +
31937 +       sz = xino_fread(sbinfo->si_xread, file, ino, sizeof(*ino), &pos);
31938 +       if (sz == sizeof(*ino))
31939 +               return 0; /* success */
31940 +
31941 +       err = sz;
31942 +       if (unlikely(sz >= 0)) {
31943 +               err = -EIO;
31944 +               AuIOErr("xino read error (%zd)\n", sz);
31945 +       }
31946 +
31947 +       return err;
31948 +}
31949 +
31950 +/* ---------------------------------------------------------------------- */
31951 +
31952 +/* create and set a new xino file */
31953 +
31954 +struct file *au_xino_create(struct super_block *sb, char *fname, int silent)
31955 +{
31956 +       struct file *file;
31957 +       struct dentry *h_parent, *d;
31958 +       struct inode *h_dir;
31959 +       int err;
31960 +
31961 +       /*
31962 +        * at mount-time, and the xino file is the default path,
31963 +        * hnotify is disabled so we have no notify events to ignore.
31964 +        * when a user specified the xino, we cannot get au_hdir to be ignored.
31965 +        */
31966 +       file = vfsub_filp_open(fname, O_RDWR | O_CREAT | O_EXCL | O_LARGEFILE
31967 +                              /* | __FMODE_NONOTIFY */,
31968 +                              S_IRUGO | S_IWUGO);
31969 +       if (IS_ERR(file)) {
31970 +               if (!silent)
31971 +                       pr_err("open %s(%ld)\n", fname, PTR_ERR(file));
31972 +               return file;
31973 +       }
31974 +
31975 +       /* keep file count */
31976 +       h_parent = dget_parent(file->f_dentry);
31977 +       h_dir = h_parent->d_inode;
31978 +       mutex_lock_nested(&h_dir->i_mutex, AuLsc_I_PARENT);
31979 +       /* mnt_want_write() is unnecessary here */
31980 +       /* no delegation since it is just created */
31981 +       err = vfsub_unlink(h_dir, &file->f_path, /*delegated*/NULL, /*force*/0);
31982 +       mutex_unlock(&h_dir->i_mutex);
31983 +       dput(h_parent);
31984 +       if (unlikely(err)) {
31985 +               if (!silent)
31986 +                       pr_err("unlink %s(%d)\n", fname, err);
31987 +               goto out;
31988 +       }
31989 +
31990 +       err = -EINVAL;
31991 +       d = file->f_dentry;
31992 +       if (unlikely(sb == d->d_sb)) {
31993 +               if (!silent)
31994 +                       pr_err("%s must be outside\n", fname);
31995 +               goto out;
31996 +       }
31997 +       if (unlikely(au_test_fs_bad_xino(d->d_sb))) {
31998 +               if (!silent)
31999 +                       pr_err("xino doesn't support %s(%s)\n",
32000 +                              fname, au_sbtype(d->d_sb));
32001 +               goto out;
32002 +       }
32003 +       return file; /* success */
32004 +
32005 +out:
32006 +       fput(file);
32007 +       file = ERR_PTR(err);
32008 +       return file;
32009 +}
32010 +
32011 +/*
32012 + * find another branch who is on the same filesystem of the specified
32013 + * branch{@btgt}. search until @bend.
32014 + */
32015 +static int is_sb_shared(struct super_block *sb, aufs_bindex_t btgt,
32016 +                       aufs_bindex_t bend)
32017 +{
32018 +       aufs_bindex_t bindex;
32019 +       struct super_block *tgt_sb = au_sbr_sb(sb, btgt);
32020 +
32021 +       for (bindex = 0; bindex < btgt; bindex++)
32022 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
32023 +                       return bindex;
32024 +       for (bindex++; bindex <= bend; bindex++)
32025 +               if (unlikely(tgt_sb == au_sbr_sb(sb, bindex)))
32026 +                       return bindex;
32027 +       return -1;
32028 +}
32029 +
32030 +/* ---------------------------------------------------------------------- */
32031 +
32032 +/*
32033 + * initialize the xinofile for the specified branch @br
32034 + * at the place/path where @base_file indicates.
32035 + * test whether another branch is on the same filesystem or not,
32036 + * if @do_test is true.
32037 + */
32038 +int au_xino_br(struct super_block *sb, struct au_branch *br, ino_t h_ino,
32039 +              struct file *base_file, int do_test)
32040 +{
32041 +       int err;
32042 +       ino_t ino;
32043 +       aufs_bindex_t bend, bindex;
32044 +       struct au_branch *shared_br, *b;
32045 +       struct file *file;
32046 +       struct super_block *tgt_sb;
32047 +
32048 +       shared_br = NULL;
32049 +       bend = au_sbend(sb);
32050 +       if (do_test) {
32051 +               tgt_sb = au_br_sb(br);
32052 +               for (bindex = 0; bindex <= bend; bindex++) {
32053 +                       b = au_sbr(sb, bindex);
32054 +                       if (tgt_sb == au_br_sb(b)) {
32055 +                               shared_br = b;
32056 +                               break;
32057 +                       }
32058 +               }
32059 +       }
32060 +
32061 +       if (!shared_br || !shared_br->br_xino.xi_file) {
32062 +               struct au_xino_lock_dir ldir;
32063 +
32064 +               au_xino_lock_dir(sb, base_file, &ldir);
32065 +               /* mnt_want_write() is unnecessary here */
32066 +               file = au_xino_create2(base_file, NULL);
32067 +               au_xino_unlock_dir(&ldir);
32068 +               err = PTR_ERR(file);
32069 +               if (IS_ERR(file))
32070 +                       goto out;
32071 +               br->br_xino.xi_file = file;
32072 +       } else {
32073 +               br->br_xino.xi_file = shared_br->br_xino.xi_file;
32074 +               get_file(br->br_xino.xi_file);
32075 +       }
32076 +
32077 +       ino = AUFS_ROOT_INO;
32078 +       err = au_xino_do_write(au_sbi(sb)->si_xwrite, br->br_xino.xi_file,
32079 +                              h_ino, ino);
32080 +       if (unlikely(err)) {
32081 +               fput(br->br_xino.xi_file);
32082 +               br->br_xino.xi_file = NULL;
32083 +       }
32084 +
32085 +out:
32086 +       return err;
32087 +}
32088 +
32089 +/* ---------------------------------------------------------------------- */
32090 +
32091 +/* trucate a xino bitmap file */
32092 +
32093 +/* todo: slow */
32094 +static int do_xib_restore(struct super_block *sb, struct file *file, void *page)
32095 +{
32096 +       int err, bit;
32097 +       ssize_t sz;
32098 +       unsigned long pindex;
32099 +       loff_t pos, pend;
32100 +       struct au_sbinfo *sbinfo;
32101 +       au_readf_t func;
32102 +       ino_t *ino;
32103 +       unsigned long *p;
32104 +
32105 +       err = 0;
32106 +       sbinfo = au_sbi(sb);
32107 +       MtxMustLock(&sbinfo->si_xib_mtx);
32108 +       p = sbinfo->si_xib_buf;
32109 +       func = sbinfo->si_xread;
32110 +       pend = vfsub_f_size_read(file);
32111 +       pos = 0;
32112 +       while (pos < pend) {
32113 +               sz = xino_fread(func, file, page, PAGE_SIZE, &pos);
32114 +               err = sz;
32115 +               if (unlikely(sz <= 0))
32116 +                       goto out;
32117 +
32118 +               err = 0;
32119 +               for (ino = page; sz > 0; ino++, sz -= sizeof(ino)) {
32120 +                       if (unlikely(*ino < AUFS_FIRST_INO))
32121 +                               continue;
32122 +
32123 +                       xib_calc_bit(*ino, &pindex, &bit);
32124 +                       AuDebugOn(page_bits <= bit);
32125 +                       err = xib_pindex(sb, pindex);
32126 +                       if (!err)
32127 +                               set_bit(bit, p);
32128 +                       else
32129 +                               goto out;
32130 +               }
32131 +       }
32132 +
32133 +out:
32134 +       return err;
32135 +}
32136 +
32137 +static int xib_restore(struct super_block *sb)
32138 +{
32139 +       int err;
32140 +       aufs_bindex_t bindex, bend;
32141 +       void *page;
32142 +
32143 +       err = -ENOMEM;
32144 +       page = (void *)__get_free_page(GFP_NOFS);
32145 +       if (unlikely(!page))
32146 +               goto out;
32147 +
32148 +       err = 0;
32149 +       bend = au_sbend(sb);
32150 +       for (bindex = 0; !err && bindex <= bend; bindex++)
32151 +               if (!bindex || is_sb_shared(sb, bindex, bindex - 1) < 0)
32152 +                       err = do_xib_restore
32153 +                               (sb, au_sbr(sb, bindex)->br_xino.xi_file, page);
32154 +               else
32155 +                       AuDbg("b%d\n", bindex);
32156 +       free_page((unsigned long)page);
32157 +
32158 +out:
32159 +       return err;
32160 +}
32161 +
32162 +int au_xib_trunc(struct super_block *sb)
32163 +{
32164 +       int err;
32165 +       ssize_t sz;
32166 +       loff_t pos;
32167 +       struct au_xino_lock_dir ldir;
32168 +       struct au_sbinfo *sbinfo;
32169 +       unsigned long *p;
32170 +       struct file *file;
32171 +
32172 +       SiMustWriteLock(sb);
32173 +
32174 +       err = 0;
32175 +       sbinfo = au_sbi(sb);
32176 +       if (!au_opt_test(sbinfo->si_mntflags, XINO))
32177 +               goto out;
32178 +
32179 +       file = sbinfo->si_xib;
32180 +       if (vfsub_f_size_read(file) <= PAGE_SIZE)
32181 +               goto out;
32182 +
32183 +       au_xino_lock_dir(sb, file, &ldir);
32184 +       /* mnt_want_write() is unnecessary here */
32185 +       file = au_xino_create2(sbinfo->si_xib, NULL);
32186 +       au_xino_unlock_dir(&ldir);
32187 +       err = PTR_ERR(file);
32188 +       if (IS_ERR(file))
32189 +               goto out;
32190 +       fput(sbinfo->si_xib);
32191 +       sbinfo->si_xib = file;
32192 +
32193 +       p = sbinfo->si_xib_buf;
32194 +       memset(p, 0, PAGE_SIZE);
32195 +       pos = 0;
32196 +       sz = xino_fwrite(sbinfo->si_xwrite, sbinfo->si_xib, p, PAGE_SIZE, &pos);
32197 +       if (unlikely(sz != PAGE_SIZE)) {
32198 +               err = sz;
32199 +               AuIOErr("err %d\n", err);
32200 +               if (sz >= 0)
32201 +                       err = -EIO;
32202 +               goto out;
32203 +       }
32204 +
32205 +       mutex_lock(&sbinfo->si_xib_mtx);
32206 +       /* mnt_want_write() is unnecessary here */
32207 +       err = xib_restore(sb);
32208 +       mutex_unlock(&sbinfo->si_xib_mtx);
32209 +
32210 +out:
32211 +       return err;
32212 +}
32213 +
32214 +/* ---------------------------------------------------------------------- */
32215 +
32216 +/*
32217 + * xino mount option handlers
32218 + */
32219 +static au_readf_t find_readf(struct file *h_file)
32220 +{
32221 +       const struct file_operations *fop = h_file->f_op;
32222 +
32223 +       if (fop->read)
32224 +               return fop->read;
32225 +       if (fop->aio_read)
32226 +               return do_sync_read;
32227 +       return ERR_PTR(-ENOSYS);
32228 +}
32229 +
32230 +static au_writef_t find_writef(struct file *h_file)
32231 +{
32232 +       const struct file_operations *fop = h_file->f_op;
32233 +
32234 +       if (fop->write)
32235 +               return fop->write;
32236 +       if (fop->aio_write)
32237 +               return do_sync_write;
32238 +       return ERR_PTR(-ENOSYS);
32239 +}
32240 +
32241 +/* xino bitmap */
32242 +static void xino_clear_xib(struct super_block *sb)
32243 +{
32244 +       struct au_sbinfo *sbinfo;
32245 +
32246 +       SiMustWriteLock(sb);
32247 +
32248 +       sbinfo = au_sbi(sb);
32249 +       sbinfo->si_xread = NULL;
32250 +       sbinfo->si_xwrite = NULL;
32251 +       if (sbinfo->si_xib)
32252 +               fput(sbinfo->si_xib);
32253 +       sbinfo->si_xib = NULL;
32254 +       free_page((unsigned long)sbinfo->si_xib_buf);
32255 +       sbinfo->si_xib_buf = NULL;
32256 +}
32257 +
32258 +static int au_xino_set_xib(struct super_block *sb, struct file *base)
32259 +{
32260 +       int err;
32261 +       loff_t pos;
32262 +       struct au_sbinfo *sbinfo;
32263 +       struct file *file;
32264 +
32265 +       SiMustWriteLock(sb);
32266 +
32267 +       sbinfo = au_sbi(sb);
32268 +       file = au_xino_create2(base, sbinfo->si_xib);
32269 +       err = PTR_ERR(file);
32270 +       if (IS_ERR(file))
32271 +               goto out;
32272 +       if (sbinfo->si_xib)
32273 +               fput(sbinfo->si_xib);
32274 +       sbinfo->si_xib = file;
32275 +       sbinfo->si_xread = find_readf(file);
32276 +       sbinfo->si_xwrite = find_writef(file);
32277 +
32278 +       err = -ENOMEM;
32279 +       if (!sbinfo->si_xib_buf)
32280 +               sbinfo->si_xib_buf = (void *)get_zeroed_page(GFP_NOFS);
32281 +       if (unlikely(!sbinfo->si_xib_buf))
32282 +               goto out_unset;
32283 +
32284 +       sbinfo->si_xib_last_pindex = 0;
32285 +       sbinfo->si_xib_next_bit = 0;
32286 +       if (vfsub_f_size_read(file) < PAGE_SIZE) {
32287 +               pos = 0;
32288 +               err = xino_fwrite(sbinfo->si_xwrite, file, sbinfo->si_xib_buf,
32289 +                                 PAGE_SIZE, &pos);
32290 +               if (unlikely(err != PAGE_SIZE))
32291 +                       goto out_free;
32292 +       }
32293 +       err = 0;
32294 +       goto out; /* success */
32295 +
32296 +out_free:
32297 +       free_page((unsigned long)sbinfo->si_xib_buf);
32298 +       sbinfo->si_xib_buf = NULL;
32299 +       if (err >= 0)
32300 +               err = -EIO;
32301 +out_unset:
32302 +       fput(sbinfo->si_xib);
32303 +       sbinfo->si_xib = NULL;
32304 +       sbinfo->si_xread = NULL;
32305 +       sbinfo->si_xwrite = NULL;
32306 +out:
32307 +       return err;
32308 +}
32309 +
32310 +/* xino for each branch */
32311 +static void xino_clear_br(struct super_block *sb)
32312 +{
32313 +       aufs_bindex_t bindex, bend;
32314 +       struct au_branch *br;
32315 +
32316 +       bend = au_sbend(sb);
32317 +       for (bindex = 0; bindex <= bend; bindex++) {
32318 +               br = au_sbr(sb, bindex);
32319 +               if (!br || !br->br_xino.xi_file)
32320 +                       continue;
32321 +
32322 +               fput(br->br_xino.xi_file);
32323 +               br->br_xino.xi_file = NULL;
32324 +       }
32325 +}
32326 +
32327 +static int au_xino_set_br(struct super_block *sb, struct file *base)
32328 +{
32329 +       int err;
32330 +       ino_t ino;
32331 +       aufs_bindex_t bindex, bend, bshared;
32332 +       struct {
32333 +               struct file *old, *new;
32334 +       } *fpair, *p;
32335 +       struct au_branch *br;
32336 +       struct inode *inode;
32337 +       au_writef_t writef;
32338 +
32339 +       SiMustWriteLock(sb);
32340 +
32341 +       err = -ENOMEM;
32342 +       bend = au_sbend(sb);
32343 +       fpair = kcalloc(bend + 1, sizeof(*fpair), GFP_NOFS);
32344 +       if (unlikely(!fpair))
32345 +               goto out;
32346 +
32347 +       inode = sb->s_root->d_inode;
32348 +       ino = AUFS_ROOT_INO;
32349 +       writef = au_sbi(sb)->si_xwrite;
32350 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++) {
32351 +               br = au_sbr(sb, bindex);
32352 +               bshared = is_sb_shared(sb, bindex, bindex - 1);
32353 +               if (bshared >= 0) {
32354 +                       /* shared xino */
32355 +                       *p = fpair[bshared];
32356 +                       get_file(p->new);
32357 +               }
32358 +
32359 +               if (!p->new) {
32360 +                       /* new xino */
32361 +                       p->old = br->br_xino.xi_file;
32362 +                       p->new = au_xino_create2(base, br->br_xino.xi_file);
32363 +                       err = PTR_ERR(p->new);
32364 +                       if (IS_ERR(p->new)) {
32365 +                               p->new = NULL;
32366 +                               goto out_pair;
32367 +                       }
32368 +               }
32369 +
32370 +               err = au_xino_do_write(writef, p->new,
32371 +                                      au_h_iptr(inode, bindex)->i_ino, ino);
32372 +               if (unlikely(err))
32373 +                       goto out_pair;
32374 +       }
32375 +
32376 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++) {
32377 +               br = au_sbr(sb, bindex);
32378 +               if (br->br_xino.xi_file)
32379 +                       fput(br->br_xino.xi_file);
32380 +               get_file(p->new);
32381 +               br->br_xino.xi_file = p->new;
32382 +       }
32383 +
32384 +out_pair:
32385 +       for (bindex = 0, p = fpair; bindex <= bend; bindex++, p++)
32386 +               if (p->new)
32387 +                       fput(p->new);
32388 +               else
32389 +                       break;
32390 +       kfree(fpair);
32391 +out:
32392 +       return err;
32393 +}
32394 +
32395 +void au_xino_clr(struct super_block *sb)
32396 +{
32397 +       struct au_sbinfo *sbinfo;
32398 +
32399 +       au_xigen_clr(sb);
32400 +       xino_clear_xib(sb);
32401 +       xino_clear_br(sb);
32402 +       sbinfo = au_sbi(sb);
32403 +       /* lvalue, do not call au_mntflags() */
32404 +       au_opt_clr(sbinfo->si_mntflags, XINO);
32405 +}
32406 +
32407 +int au_xino_set(struct super_block *sb, struct au_opt_xino *xino, int remount)
32408 +{
32409 +       int err, skip;
32410 +       struct dentry *parent, *cur_parent;
32411 +       struct qstr *dname, *cur_name;
32412 +       struct file *cur_xino;
32413 +       struct inode *dir;
32414 +       struct au_sbinfo *sbinfo;
32415 +
32416 +       SiMustWriteLock(sb);
32417 +
32418 +       err = 0;
32419 +       sbinfo = au_sbi(sb);
32420 +       parent = dget_parent(xino->file->f_dentry);
32421 +       if (remount) {
32422 +               skip = 0;
32423 +               dname = &xino->file->f_dentry->d_name;
32424 +               cur_xino = sbinfo->si_xib;
32425 +               if (cur_xino) {
32426 +                       cur_parent = dget_parent(cur_xino->f_dentry);
32427 +                       cur_name = &cur_xino->f_dentry->d_name;
32428 +                       skip = (cur_parent == parent
32429 +                               && dname->len == cur_name->len
32430 +                               && !memcmp(dname->name, cur_name->name,
32431 +                                          dname->len));
32432 +                       dput(cur_parent);
32433 +               }
32434 +               if (skip)
32435 +                       goto out;
32436 +       }
32437 +
32438 +       au_opt_set(sbinfo->si_mntflags, XINO);
32439 +       dir = parent->d_inode;
32440 +       mutex_lock_nested(&dir->i_mutex, AuLsc_I_PARENT);
32441 +       /* mnt_want_write() is unnecessary here */
32442 +       err = au_xino_set_xib(sb, xino->file);
32443 +       if (!err)
32444 +               err = au_xigen_set(sb, xino->file);
32445 +       if (!err)
32446 +               err = au_xino_set_br(sb, xino->file);
32447 +       mutex_unlock(&dir->i_mutex);
32448 +       if (!err)
32449 +               goto out; /* success */
32450 +
32451 +       /* reset all */
32452 +       AuIOErr("failed creating xino(%d).\n", err);
32453 +
32454 +out:
32455 +       dput(parent);
32456 +       return err;
32457 +}
32458 +
32459 +/* ---------------------------------------------------------------------- */
32460 +
32461 +/*
32462 + * create a xinofile at the default place/path.
32463 + */
32464 +struct file *au_xino_def(struct super_block *sb)
32465 +{
32466 +       struct file *file;
32467 +       char *page, *p;
32468 +       struct au_branch *br;
32469 +       struct super_block *h_sb;
32470 +       struct path path;
32471 +       aufs_bindex_t bend, bindex, bwr;
32472 +
32473 +       br = NULL;
32474 +       bend = au_sbend(sb);
32475 +       bwr = -1;
32476 +       for (bindex = 0; bindex <= bend; bindex++) {
32477 +               br = au_sbr(sb, bindex);
32478 +               if (au_br_writable(br->br_perm)
32479 +                   && !au_test_fs_bad_xino(au_br_sb(br))) {
32480 +                       bwr = bindex;
32481 +                       break;
32482 +               }
32483 +       }
32484 +
32485 +       if (bwr >= 0) {
32486 +               file = ERR_PTR(-ENOMEM);
32487 +               page = (void *)__get_free_page(GFP_NOFS);
32488 +               if (unlikely(!page))
32489 +                       goto out;
32490 +               path.mnt = au_br_mnt(br);
32491 +               path.dentry = au_h_dptr(sb->s_root, bwr);
32492 +               p = d_path(&path, page, PATH_MAX - sizeof(AUFS_XINO_FNAME));
32493 +               file = (void *)p;
32494 +               if (!IS_ERR(p)) {
32495 +                       strcat(p, "/" AUFS_XINO_FNAME);
32496 +                       AuDbg("%s\n", p);
32497 +                       file = au_xino_create(sb, p, /*silent*/0);
32498 +                       if (!IS_ERR(file))
32499 +                               au_xino_brid_set(sb, br->br_id);
32500 +               }
32501 +               free_page((unsigned long)page);
32502 +       } else {
32503 +               file = au_xino_create(sb, AUFS_XINO_DEFPATH, /*silent*/0);
32504 +               if (IS_ERR(file))
32505 +                       goto out;
32506 +               h_sb = file->f_dentry->d_sb;
32507 +               if (unlikely(au_test_fs_bad_xino(h_sb))) {
32508 +                       pr_err("xino doesn't support %s(%s)\n",
32509 +                              AUFS_XINO_DEFPATH, au_sbtype(h_sb));
32510 +                       fput(file);
32511 +                       file = ERR_PTR(-EINVAL);
32512 +               }
32513 +               if (!IS_ERR(file))
32514 +                       au_xino_brid_set(sb, -1);
32515 +       }
32516 +
32517 +out:
32518 +       return file;
32519 +}
32520 +
32521 +/* ---------------------------------------------------------------------- */
32522 +
32523 +int au_xino_path(struct seq_file *seq, struct file *file)
32524 +{
32525 +       int err;
32526 +
32527 +       err = au_seq_path(seq, &file->f_path);
32528 +       if (unlikely(err < 0))
32529 +               goto out;
32530 +
32531 +       err = 0;
32532 +#define Deleted "\\040(deleted)"
32533 +       seq->count -= sizeof(Deleted) - 1;
32534 +       AuDebugOn(memcmp(seq->buf + seq->count, Deleted,
32535 +                        sizeof(Deleted) - 1));
32536 +#undef Deleted
32537 +
32538 +out:
32539 +       return err;
32540 +}
32541 diff -urN /usr/share/empty/include/uapi/linux/aufs_type.h linux/include/uapi/linux/aufs_type.h
32542 --- /usr/share/empty/include/uapi/linux/aufs_type.h     1970-01-01 01:00:00.000000000 +0100
32543 +++ linux/include/uapi/linux/aufs_type.h        2014-04-24 22:11:10.935269950 +0200
32544 @@ -0,0 +1,281 @@
32545 +/*
32546 + * Copyright (C) 2005-2014 Junjiro R. Okajima
32547 + *
32548 + * This program, aufs is free software; you can redistribute it and/or modify
32549 + * it under the terms of the GNU General Public License as published by
32550 + * the Free Software Foundation; either version 2 of the License, or
32551 + * (at your option) any later version.
32552 + *
32553 + * This program is distributed in the hope that it will be useful,
32554 + * but WITHOUT ANY WARRANTY; without even the implied warranty of
32555 + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
32556 + * GNU General Public License for more details.
32557 + *
32558 + * You should have received a copy of the GNU General Public License
32559 + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
32560 + */
32561 +
32562 +#ifndef __AUFS_TYPE_H__
32563 +#define __AUFS_TYPE_H__
32564 +
32565 +#define AUFS_NAME      "aufs"
32566 +
32567 +#ifdef __KERNEL__
32568 +/*
32569 + * define it before including all other headers.
32570 + * sched.h may use pr_* macros before defining "current", so define the
32571 + * no-current version first, and re-define later.
32572 + */
32573 +#define pr_fmt(fmt)    AUFS_NAME " %s:%d: " fmt, __func__, __LINE__
32574 +#include <linux/sched.h>
32575 +#undef pr_fmt
32576 +#define pr_fmt(fmt) \
32577 +               AUFS_NAME " %s:%d:%.*s[%d]: " fmt, __func__, __LINE__, \
32578 +               (int)sizeof(current->comm), current->comm, current->pid
32579 +#else
32580 +#include <stdint.h>
32581 +#include <sys/types.h>
32582 +#endif /* __KERNEL__ */
32583 +
32584 +#include <linux/limits.h>
32585 +
32586 +#define AUFS_VERSION   "3.14-20140407"
32587 +
32588 +/* todo? move this to linux-2.6.19/include/magic.h */
32589 +#define AUFS_SUPER_MAGIC       ('a' << 24 | 'u' << 16 | 'f' << 8 | 's')
32590 +
32591 +/* ---------------------------------------------------------------------- */
32592 +
32593 +#ifdef CONFIG_AUFS_BRANCH_MAX_127
32594 +typedef int8_t aufs_bindex_t;
32595 +#define AUFS_BRANCH_MAX 127
32596 +#else
32597 +typedef int16_t aufs_bindex_t;
32598 +#ifdef CONFIG_AUFS_BRANCH_MAX_511
32599 +#define AUFS_BRANCH_MAX 511
32600 +#elif defined(CONFIG_AUFS_BRANCH_MAX_1023)
32601 +#define AUFS_BRANCH_MAX 1023
32602 +#elif defined(CONFIG_AUFS_BRANCH_MAX_32767)
32603 +#define AUFS_BRANCH_MAX 32767
32604 +#endif
32605 +#endif
32606 +
32607 +#ifdef __KERNEL__
32608 +#ifndef AUFS_BRANCH_MAX
32609 +#error unknown CONFIG_AUFS_BRANCH_MAX value
32610 +#endif
32611 +#endif /* __KERNEL__ */
32612 +
32613 +/* ---------------------------------------------------------------------- */
32614 +
32615 +#define AUFS_FSTYPE            AUFS_NAME
32616 +
32617 +#define AUFS_ROOT_INO          2
32618 +#define AUFS_FIRST_INO         11
32619 +
32620 +#define AUFS_WH_PFX            ".wh."
32621 +#define AUFS_WH_PFX_LEN                ((int)sizeof(AUFS_WH_PFX) - 1)
32622 +#define AUFS_WH_TMP_LEN                4
32623 +/* a limit for rmdir/rename a dir and copyup */
32624 +#define AUFS_MAX_NAMELEN       (NAME_MAX \
32625 +                               - AUFS_WH_PFX_LEN * 2   /* doubly whiteouted */\
32626 +                               - 1                     /* dot */\
32627 +                               - AUFS_WH_TMP_LEN)      /* hex */
32628 +#define AUFS_XINO_FNAME                "." AUFS_NAME ".xino"
32629 +#define AUFS_XINO_DEFPATH      "/tmp/" AUFS_XINO_FNAME
32630 +#define AUFS_XINO_DEF_SEC      30 /* seconds */
32631 +#define AUFS_XINO_DEF_TRUNC    45 /* percentage */
32632 +#define AUFS_DIRWH_DEF         3
32633 +#define AUFS_RDCACHE_DEF       10 /* seconds */
32634 +#define AUFS_RDCACHE_MAX       3600 /* seconds */
32635 +#define AUFS_RDBLK_DEF         512 /* bytes */
32636 +#define AUFS_RDHASH_DEF                32
32637 +#define AUFS_WKQ_NAME          AUFS_NAME "d"
32638 +#define AUFS_MFS_DEF_SEC       30 /* seconds */
32639 +#define AUFS_MFS_MAX_SEC       3600 /* seconds */
32640 +#define AUFS_PLINK_WARN                50 /* number of plinks in a single bucket */
32641 +
32642 +/* pseudo-link maintenace under /proc */
32643 +#define AUFS_PLINK_MAINT_NAME  "plink_maint"
32644 +#define AUFS_PLINK_MAINT_DIR   "fs/" AUFS_NAME
32645 +#define AUFS_PLINK_MAINT_PATH  AUFS_PLINK_MAINT_DIR "/" AUFS_PLINK_MAINT_NAME
32646 +
32647 +#define AUFS_DIROPQ_NAME       AUFS_WH_PFX ".opq" /* whiteouted doubly */
32648 +#define AUFS_WH_DIROPQ         AUFS_WH_PFX AUFS_DIROPQ_NAME
32649 +
32650 +#define AUFS_BASE_NAME         AUFS_WH_PFX AUFS_NAME
32651 +#define AUFS_PLINKDIR_NAME     AUFS_WH_PFX "plnk"
32652 +#define AUFS_ORPHDIR_NAME      AUFS_WH_PFX "orph"
32653 +
32654 +/* doubly whiteouted */
32655 +#define AUFS_WH_BASE           AUFS_WH_PFX AUFS_BASE_NAME
32656 +#define AUFS_WH_PLINKDIR       AUFS_WH_PFX AUFS_PLINKDIR_NAME
32657 +#define AUFS_WH_ORPHDIR                AUFS_WH_PFX AUFS_ORPHDIR_NAME
32658 +
32659 +/* branch permissions and attributes */
32660 +#define AUFS_BRPERM_RW         "rw"
32661 +#define AUFS_BRPERM_RO         "ro"
32662 +#define AUFS_BRPERM_RR         "rr"
32663 +#define AUFS_BRRATTR_WH                "wh"
32664 +#define AUFS_BRWATTR_NLWH      "nolwh"
32665 +#define AUFS_BRATTR_UNPIN      "unpin"
32666 +
32667 +/* ---------------------------------------------------------------------- */
32668 +
32669 +/* ioctl */
32670 +enum {
32671 +       /* readdir in userspace */
32672 +       AuCtl_RDU,
32673 +       AuCtl_RDU_INO,
32674 +
32675 +       /* pathconf wrapper */
32676 +       AuCtl_WBR_FD,
32677 +
32678 +       /* busy inode */
32679 +       AuCtl_IBUSY,
32680 +
32681 +       /* move-down */
32682 +       AuCtl_MVDOWN
32683 +};
32684 +
32685 +/* borrowed from linux/include/linux/kernel.h */
32686 +#ifndef ALIGN
32687 +#define ALIGN(x, a)            __ALIGN_MASK(x, (typeof(x))(a)-1)
32688 +#define __ALIGN_MASK(x, mask)  (((x)+(mask))&~(mask))
32689 +#endif
32690 +
32691 +/* borrowed from linux/include/linux/compiler-gcc3.h */
32692 +#ifndef __aligned
32693 +#define __aligned(x)                   __attribute__((aligned(x)))
32694 +#endif
32695 +
32696 +#ifdef __KERNEL__
32697 +#ifndef __packed
32698 +#define __packed                       __attribute__((packed))
32699 +#endif
32700 +#endif
32701 +
32702 +struct au_rdu_cookie {
32703 +       uint64_t        h_pos;
32704 +       int16_t         bindex;
32705 +       uint8_t         flags;
32706 +       uint8_t         pad;
32707 +       uint32_t        generation;
32708 +} __aligned(8);
32709 +
32710 +struct au_rdu_ent {
32711 +       uint64_t        ino;
32712 +       int16_t         bindex;
32713 +       uint8_t         type;
32714 +       uint8_t         nlen;
32715 +       uint8_t         wh;
32716 +       char            name[0];
32717 +} __aligned(8);
32718 +
32719 +static inline int au_rdu_len(int nlen)
32720 +{
32721 +       /* include the terminating NULL */
32722 +       return ALIGN(sizeof(struct au_rdu_ent) + nlen + 1,
32723 +                    sizeof(uint64_t));
32724 +}
32725 +
32726 +union au_rdu_ent_ul {
32727 +       struct au_rdu_ent __user        *e;
32728 +       uint64_t                        ul;
32729 +};
32730 +
32731 +enum {
32732 +       AufsCtlRduV_SZ,
32733 +       AufsCtlRduV_End
32734 +};
32735 +
32736 +struct aufs_rdu {
32737 +       /* input */
32738 +       union {
32739 +               uint64_t        sz;     /* AuCtl_RDU */
32740 +               uint64_t        nent;   /* AuCtl_RDU_INO */
32741 +       };
32742 +       union au_rdu_ent_ul     ent;
32743 +       uint16_t                verify[AufsCtlRduV_End];
32744 +
32745 +       /* input/output */
32746 +       uint32_t                blk;
32747 +
32748 +       /* output */
32749 +       union au_rdu_ent_ul     tail;
32750 +       /* number of entries which were added in a single call */
32751 +       uint64_t                rent;
32752 +       uint8_t                 full;
32753 +       uint8_t                 shwh;
32754 +
32755 +       struct au_rdu_cookie    cookie;
32756 +} __aligned(8);
32757 +
32758 +/* ---------------------------------------------------------------------- */
32759 +
32760 +struct aufs_wbr_fd {
32761 +       uint32_t        oflags;
32762 +       int16_t         brid;
32763 +} __aligned(8);
32764 +
32765 +/* ---------------------------------------------------------------------- */
32766 +
32767 +struct aufs_ibusy {
32768 +       uint64_t        ino, h_ino;
32769 +       int16_t         bindex;
32770 +} __aligned(8);
32771 +
32772 +/* ---------------------------------------------------------------------- */
32773 +
32774 +/* error code for move-down */
32775 +/* the actual message strings are implemented in aufs-util.git */
32776 +enum {
32777 +       EAU_MVDOWN_OPAQUE = 1,
32778 +       EAU_MVDOWN_WHITEOUT,
32779 +       EAU_MVDOWN_UPPER,
32780 +       EAU_MVDOWN_BOTTOM,
32781 +       EAU_MVDOWN_NOUPPER,
32782 +       EAU_MVDOWN_NOLOWERBR,
32783 +       EAU_Last
32784 +};
32785 +
32786 +/* flags for move-down */
32787 +#define AUFS_MVDOWN_DMSG       1
32788 +#define AUFS_MVDOWN_OWLOWER    (1 << 1)        /* overwrite lower */
32789 +#define AUFS_MVDOWN_KUPPER     (1 << 2)        /* keep upper */
32790 +#define AUFS_MVDOWN_ROLOWER    (1 << 3)        /* do even if lower is RO */
32791 +#define AUFS_MVDOWN_ROLOWER_R  (1 << 4)        /* did on lower RO */
32792 +#define AUFS_MVDOWN_ROUPPER    (1 << 5)        /* do even if upper is RO */
32793 +#define AUFS_MVDOWN_ROUPPER_R  (1 << 6)        /* did on upper RO */
32794 +#define AUFS_MVDOWN_BRID_UPPER (1 << 7)        /* upper brid */
32795 +#define AUFS_MVDOWN_BRID_LOWER (1 << 8)        /* lower brid */
32796 +/* will be added more */
32797 +
32798 +enum {
32799 +       AUFS_MVDOWN_UPPER,
32800 +       AUFS_MVDOWN_LOWER,
32801 +       AUFS_MVDOWN_NARRAY
32802 +};
32803 +
32804 +struct aufs_mvdown {
32805 +       uint32_t        flags;
32806 +       struct {
32807 +               int16_t         bindex;
32808 +               int16_t         brid;
32809 +       } a[AUFS_MVDOWN_NARRAY];
32810 +       int8_t          au_errno;
32811 +       /* will be added more */
32812 +} __aligned(8);
32813 +
32814 +/* ---------------------------------------------------------------------- */
32815 +
32816 +#define AuCtlType              'A'
32817 +#define AUFS_CTL_RDU           _IOWR(AuCtlType, AuCtl_RDU, struct aufs_rdu)
32818 +#define AUFS_CTL_RDU_INO       _IOWR(AuCtlType, AuCtl_RDU_INO, struct aufs_rdu)
32819 +#define AUFS_CTL_WBR_FD                _IOW(AuCtlType, AuCtl_WBR_FD, \
32820 +                                    struct aufs_wbr_fd)
32821 +#define AUFS_CTL_IBUSY         _IOWR(AuCtlType, AuCtl_IBUSY, struct aufs_ibusy)
32822 +#define AUFS_CTL_MVDOWN                _IOWR(AuCtlType, AuCtl_MVDOWN, \
32823 +                                     struct aufs_mvdown)
32824 +
32825 +#endif /* __AUFS_TYPE_H__ */
32826 aufs3.14 loopback patch
32827
32828 diff --git a/drivers/block/loop.c b/drivers/block/loop.c
32829 index ec278ac..1894990 100644
32830 --- a/drivers/block/loop.c
32831 +++ b/drivers/block/loop.c
32832 @@ -514,7 +514,7 @@ out:
32833  }
32834  
32835  struct switch_request {
32836 -       struct file *file;
32837 +       struct file *file, *virt_file;
32838         struct completion wait;
32839  };
32840  
32841 @@ -576,7 +576,8 @@ static int loop_thread(void *data)
32842   * First it needs to flush existing IO, it does this by sending a magic
32843   * BIO down the pipe. The completion of this BIO does the actual switch.
32844   */
32845 -static int loop_switch(struct loop_device *lo, struct file *file)
32846 +static int loop_switch(struct loop_device *lo, struct file *file,
32847 +                      struct file *virt_file)
32848  {
32849         struct switch_request w;
32850         struct bio *bio = bio_alloc(GFP_KERNEL, 0);
32851 @@ -584,6 +585,7 @@ static int loop_switch(struct loop_device *lo, struct file *file)
32852                 return -ENOMEM;
32853         init_completion(&w.wait);
32854         w.file = file;
32855 +       w.virt_file = virt_file;
32856         bio->bi_private = &w;
32857         bio->bi_bdev = NULL;
32858         loop_make_request(lo->lo_queue, bio);
32859 @@ -600,7 +602,7 @@ static int loop_flush(struct loop_device *lo)
32860         if (!lo->lo_thread)
32861                 return 0;
32862  
32863 -       return loop_switch(lo, NULL);
32864 +       return loop_switch(lo, NULL, NULL);
32865  }
32866  
32867  /*
32868 @@ -619,6 +621,7 @@ static void do_loop_switch(struct loop_device *lo, struct switch_request *p)
32869         mapping = file->f_mapping;
32870         mapping_set_gfp_mask(old_file->f_mapping, lo->old_gfp_mask);
32871         lo->lo_backing_file = file;
32872 +       lo->lo_backing_virt_file = p->virt_file;
32873         lo->lo_blocksize = S_ISBLK(mapping->host->i_mode) ?
32874                 mapping->host->i_bdev->bd_block_size : PAGE_SIZE;
32875         lo->old_gfp_mask = mapping_gfp_mask(mapping);
32876 @@ -627,6 +630,13 @@ out:
32877         complete(&p->wait);
32878  }
32879  
32880 +static struct file *loop_real_file(struct file *file)
32881 +{
32882 +       struct file *f = NULL;
32883 +       if (file->f_dentry->d_sb->s_op->real_loop)
32884 +               f = file->f_dentry->d_sb->s_op->real_loop(file);
32885 +       return f;
32886 +}
32887  
32888  /*
32889   * loop_change_fd switched the backing store of a loopback device to
32890 @@ -640,6 +650,7 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
32891                           unsigned int arg)
32892  {
32893         struct file     *file, *old_file;
32894 +       struct file     *f, *virt_file = NULL, *old_virt_file;
32895         struct inode    *inode;
32896         int             error;
32897  
32898 @@ -656,9 +667,16 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
32899         file = fget(arg);
32900         if (!file)
32901                 goto out;
32902 +       f = loop_real_file(file);
32903 +       if (f) {
32904 +               virt_file = file;
32905 +               file = f;
32906 +               get_file(file);
32907 +       }
32908  
32909         inode = file->f_mapping->host;
32910         old_file = lo->lo_backing_file;
32911 +       old_virt_file = lo->lo_backing_virt_file;
32912  
32913         error = -EINVAL;
32914  
32915 @@ -670,17 +688,21 @@ static int loop_change_fd(struct loop_device *lo, struct block_device *bdev,
32916                 goto out_putf;
32917  
32918         /* and ... switch */
32919 -       error = loop_switch(lo, file);
32920 +       error = loop_switch(lo, file, virt_file);
32921         if (error)
32922                 goto out_putf;
32923  
32924         fput(old_file);
32925 +       if (old_virt_file)
32926 +               fput(old_virt_file);
32927         if (lo->lo_flags & LO_FLAGS_PARTSCAN)
32928                 ioctl_by_bdev(bdev, BLKRRPART, 0);
32929         return 0;
32930  
32931   out_putf:
32932         fput(file);
32933 +       if (virt_file)
32934 +               fput(virt_file);
32935   out:
32936         return error;
32937  }
32938 @@ -841,7 +863,7 @@ static void loop_config_discard(struct loop_device *lo)
32939  static int loop_set_fd(struct loop_device *lo, fmode_t mode,
32940                        struct block_device *bdev, unsigned int arg)
32941  {
32942 -       struct file     *file, *f;
32943 +       struct file     *file, *f, *virt_file = NULL;
32944         struct inode    *inode;
32945         struct address_space *mapping;
32946         unsigned lo_blocksize;
32947 @@ -856,6 +878,12 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
32948         file = fget(arg);
32949         if (!file)
32950                 goto out;
32951 +       f = loop_real_file(file);
32952 +       if (f) {
32953 +               virt_file = file;
32954 +               file = f;
32955 +               get_file(file);
32956 +       }
32957  
32958         error = -EBUSY;
32959         if (lo->lo_state != Lo_unbound)
32960 @@ -904,6 +932,7 @@ static int loop_set_fd(struct loop_device *lo, fmode_t mode,
32961         lo->lo_device = bdev;
32962         lo->lo_flags = lo_flags;
32963         lo->lo_backing_file = file;
32964 +       lo->lo_backing_virt_file = virt_file;
32965         lo->transfer = transfer_none;
32966         lo->ioctl = NULL;
32967         lo->lo_sizelimit = 0;
32968 @@ -948,6 +977,7 @@ out_clr:
32969         lo->lo_thread = NULL;
32970         lo->lo_device = NULL;
32971         lo->lo_backing_file = NULL;
32972 +       lo->lo_backing_virt_file = NULL;
32973         lo->lo_flags = 0;
32974         set_capacity(lo->lo_disk, 0);
32975         invalidate_bdev(bdev);
32976 @@ -957,6 +987,8 @@ out_clr:
32977         lo->lo_state = Lo_unbound;
32978   out_putf:
32979         fput(file);
32980 +       if (virt_file)
32981 +               fput(virt_file);
32982   out:
32983         /* This is safe: open() is still holding a reference. */
32984         module_put(THIS_MODULE);
32985 @@ -1003,6 +1035,7 @@ loop_init_xfer(struct loop_device *lo, struct loop_func_table *xfer,
32986  static int loop_clr_fd(struct loop_device *lo)
32987  {
32988         struct file *filp = lo->lo_backing_file;
32989 +       struct file *virt_filp = lo->lo_backing_virt_file;
32990         gfp_t gfp = lo->old_gfp_mask;
32991         struct block_device *bdev = lo->lo_device;
32992  
32993 @@ -1036,6 +1069,7 @@ static int loop_clr_fd(struct loop_device *lo)
32994  
32995         spin_lock_irq(&lo->lo_lock);
32996         lo->lo_backing_file = NULL;
32997 +       lo->lo_backing_virt_file = NULL;
32998         spin_unlock_irq(&lo->lo_lock);
32999  
33000         loop_release_xfer(lo);
33001 @@ -1078,6 +1112,8 @@ static int loop_clr_fd(struct loop_device *lo)
33002          * bd_mutex which is usually taken before lo_ctl_mutex.
33003          */
33004         fput(filp);
33005 +       if (virt_filp)
33006 +               fput(virt_filp);
33007         return 0;
33008  }
33009  
33010 diff --git a/drivers/block/loop.h b/drivers/block/loop.h
33011 index 90df5d6..cb91822 100644
33012 --- a/drivers/block/loop.h
33013 +++ b/drivers/block/loop.h
33014 @@ -44,7 +44,7 @@ struct loop_device {
33015         int             (*ioctl)(struct loop_device *, int cmd, 
33016                                  unsigned long arg); 
33017  
33018 -       struct file *   lo_backing_file;
33019 +       struct file *   lo_backing_file, *lo_backing_virt_file;
33020         struct block_device *lo_device;
33021         unsigned        lo_blocksize;
33022         void            *key_data; 
33023 diff --git a/fs/aufs/f_op.c b/fs/aufs/f_op.c
33024 index 2e0302d..b35af58 100644
33025 --- a/fs/aufs/f_op.c
33026 +++ b/fs/aufs/f_op.c
33027 @@ -337,7 +337,7 @@ static ssize_t aufs_splice_read(struct file *file, loff_t *ppos,
33028         err = -EINVAL;
33029         h_file = au_hf_top(file);
33030         get_file(h_file);
33031 -       if (au_test_loopback_kthread()) {
33032 +       if (0 && au_test_loopback_kthread()) {
33033                 au_warn_loopback(h_file->f_dentry->d_sb);
33034                 if (file->f_mapping != h_file->f_mapping) {
33035                         file->f_mapping = h_file->f_mapping;
33036 diff --git a/fs/aufs/loop.c b/fs/aufs/loop.c
33037 index 3b03b52..4ab749d 100644
33038 --- a/fs/aufs/loop.c
33039 +++ b/fs/aufs/loop.c
33040 @@ -130,3 +130,19 @@ void au_loopback_fin(void)
33041         symbol_put(loop_backing_file);
33042         kfree(au_warn_loopback_array);
33043  }
33044 +
33045 +/* ---------------------------------------------------------------------- */
33046 +
33047 +/* support the loopback block device insude aufs */
33048 +
33049 +struct file *aufs_real_loop(struct file *file)
33050 +{
33051 +       struct file *f;
33052 +
33053 +       BUG_ON(!au_test_aufs(file->f_dentry->d_sb));
33054 +       fi_read_lock(file);
33055 +       f = au_hf_top(file);
33056 +       fi_read_unlock(file);
33057 +       AuDebugOn(!f);
33058 +       return f;
33059 +}
33060 diff --git a/fs/aufs/loop.h b/fs/aufs/loop.h
33061 index da8b756..28cb7ea 100644
33062 --- a/fs/aufs/loop.h
33063 +++ b/fs/aufs/loop.h
33064 @@ -25,7 +25,11 @@ void au_warn_loopback(struct super_block *h_sb);
33065  
33066  int au_loopback_init(void);
33067  void au_loopback_fin(void);
33068 +
33069 +struct file *aufs_real_loop(struct file *file);
33070  #else
33071 +AuStub(struct file *, loop_backing_file, return NULL)
33072 +
33073  AuStubInt0(au_test_loopback_overlap, struct super_block *sb,
33074            struct dentry *h_adding)
33075  AuStubInt0(au_test_loopback_kthread, void)
33076 @@ -33,6 +37,8 @@ AuStubVoid(au_warn_loopback, struct super_block *h_sb)
33077  
33078  AuStubInt0(au_loopback_init, void)
33079  AuStubVoid(au_loopback_fin, void)
33080 +
33081 +AuStub(struct file *, aufs_real_loop, return NULL, struct file *file)
33082  #endif /* BLK_DEV_LOOP */
33083  
33084  #endif /* __KERNEL__ */
33085 diff --git a/fs/aufs/super.c b/fs/aufs/super.c
33086 index b609e5a..e3909ed 100644
33087 --- a/fs/aufs/super.c
33088 +++ b/fs/aufs/super.c
33089 @@ -807,7 +807,10 @@ static const struct super_operations aufs_sop = {
33090         .statfs         = aufs_statfs,
33091         .put_super      = aufs_put_super,
33092         .sync_fs        = aufs_sync_fs,
33093 -       .remount_fs     = aufs_remount_fs
33094 +       .remount_fs     = aufs_remount_fs,
33095 +#ifdef CONFIG_AUFS_BDEV_LOOP
33096 +       .real_loop      = aufs_real_loop
33097 +#endif
33098  };
33099  
33100  /* ---------------------------------------------------------------------- */
33101 diff --git a/include/linux/fs.h b/include/linux/fs.h
33102 index f3f635c..c4308ca 100644
33103 --- a/include/linux/fs.h
33104 +++ b/include/linux/fs.h
33105 @@ -1626,6 +1626,10 @@ struct super_operations {
33106         int (*bdev_try_to_free_page)(struct super_block*, struct page*, gfp_t);
33107         long (*nr_cached_objects)(struct super_block *, int);
33108         long (*free_cached_objects)(struct super_block *, long, int);
33109 +#if defined(CONFIG_BLK_DEV_LOOP) ||  defined(CONFIG_BLK_DEV_LOOP_MODULE)
33110 +       /* and aufs */
33111 +       struct file *(*real_loop)(struct file *);
33112 +#endif
33113  };
33114  
33115  /*
This page took 3.365434 seconds and 3 git commands to generate.