1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (C) 2011 Novell Inc.
8 #include <linux/slab.h>
9 #include <linux/namei.h>
10 #include <linux/file.h>
11 #include <linux/xattr.h>
12 #include <linux/rbtree.h>
13 #include <linux/security.h>
14 #include <linux/cred.h>
15 #include <linux/ratelimit.h>
16 #include "overlayfs.h"
18 struct ovl_cache_entry
{
23 struct list_head l_node
;
25 struct ovl_cache_entry
*next_maybe_whiteout
;
31 struct ovl_dir_cache
{
34 struct list_head entries
;
38 struct ovl_readdir_data
{
39 struct dir_context ctx
;
40 struct dentry
*dentry
;
43 struct list_head
*list
;
44 struct list_head middle
;
45 struct ovl_cache_entry
*first_maybe_whiteout
;
49 bool d_type_supported
;
55 struct ovl_dir_cache
*cache
;
56 struct list_head
*cursor
;
57 struct file
*realfile
;
58 struct file
*upperfile
;
61 static struct ovl_cache_entry
*ovl_cache_entry_from_node(struct rb_node
*n
)
63 return rb_entry(n
, struct ovl_cache_entry
, node
);
66 static bool ovl_cache_entry_find_link(const char *name
, int len
,
67 struct rb_node
***link
,
68 struct rb_node
**parent
)
71 struct rb_node
**newp
= *link
;
73 while (!found
&& *newp
) {
75 struct ovl_cache_entry
*tmp
;
78 tmp
= ovl_cache_entry_from_node(*newp
);
79 cmp
= strncmp(name
, tmp
->name
, len
);
81 newp
= &tmp
->node
.rb_right
;
82 else if (cmp
< 0 || len
< tmp
->len
)
83 newp
= &tmp
->node
.rb_left
;
92 static struct ovl_cache_entry
*ovl_cache_entry_find(struct rb_root
*root
,
93 const char *name
, int len
)
95 struct rb_node
*node
= root
->rb_node
;
99 struct ovl_cache_entry
*p
= ovl_cache_entry_from_node(node
);
101 cmp
= strncmp(name
, p
->name
, len
);
103 node
= p
->node
.rb_right
;
104 else if (cmp
< 0 || len
< p
->len
)
105 node
= p
->node
.rb_left
;
113 static bool ovl_calc_d_ino(struct ovl_readdir_data
*rdd
,
114 struct ovl_cache_entry
*p
)
116 /* Don't care if not doing ovl_iter() */
120 /* Always recalc d_ino when remapping lower inode numbers */
121 if (ovl_xino_bits(rdd
->dentry
->d_sb
))
124 /* Always recalc d_ino for parent */
125 if (strcmp(p
->name
, "..") == 0)
128 /* If this is lower, then native d_ino will do */
133 * Recalc d_ino for '.' and for all entries if dir is impure (contains
136 if ((p
->name
[0] == '.' && p
->len
== 1) ||
137 ovl_test_flag(OVL_IMPURE
, d_inode(rdd
->dentry
)))
143 static struct ovl_cache_entry
*ovl_cache_entry_new(struct ovl_readdir_data
*rdd
,
144 const char *name
, int len
,
145 u64 ino
, unsigned int d_type
)
147 struct ovl_cache_entry
*p
;
148 size_t size
= offsetof(struct ovl_cache_entry
, name
[len
+ 1]);
150 p
= kmalloc(size
, GFP_KERNEL
);
154 memcpy(p
->name
, name
, len
);
160 /* Defer setting d_ino for upper entry to ovl_iterate() */
161 if (ovl_calc_d_ino(rdd
, p
))
163 p
->is_upper
= rdd
->is_upper
;
164 p
->is_whiteout
= false;
166 if (d_type
== DT_CHR
) {
167 p
->next_maybe_whiteout
= rdd
->first_maybe_whiteout
;
168 rdd
->first_maybe_whiteout
= p
;
173 static int ovl_cache_entry_add_rb(struct ovl_readdir_data
*rdd
,
174 const char *name
, int len
, u64 ino
,
177 struct rb_node
**newp
= &rdd
->root
->rb_node
;
178 struct rb_node
*parent
= NULL
;
179 struct ovl_cache_entry
*p
;
181 if (ovl_cache_entry_find_link(name
, len
, &newp
, &parent
))
184 p
= ovl_cache_entry_new(rdd
, name
, len
, ino
, d_type
);
190 list_add_tail(&p
->l_node
, rdd
->list
);
191 rb_link_node(&p
->node
, parent
, newp
);
192 rb_insert_color(&p
->node
, rdd
->root
);
197 static int ovl_fill_lowest(struct ovl_readdir_data
*rdd
,
198 const char *name
, int namelen
,
199 loff_t offset
, u64 ino
, unsigned int d_type
)
201 struct ovl_cache_entry
*p
;
203 p
= ovl_cache_entry_find(rdd
->root
, name
, namelen
);
205 list_move_tail(&p
->l_node
, &rdd
->middle
);
207 p
= ovl_cache_entry_new(rdd
, name
, namelen
, ino
, d_type
);
211 list_add_tail(&p
->l_node
, &rdd
->middle
);
217 void ovl_cache_free(struct list_head
*list
)
219 struct ovl_cache_entry
*p
;
220 struct ovl_cache_entry
*n
;
222 list_for_each_entry_safe(p
, n
, list
, l_node
)
225 INIT_LIST_HEAD(list
);
228 void ovl_dir_cache_free(struct inode
*inode
)
230 struct ovl_dir_cache
*cache
= ovl_dir_cache(inode
);
233 ovl_cache_free(&cache
->entries
);
238 static void ovl_cache_put(struct ovl_dir_file
*od
, struct dentry
*dentry
)
240 struct ovl_dir_cache
*cache
= od
->cache
;
242 WARN_ON(cache
->refcount
<= 0);
244 if (!cache
->refcount
) {
245 if (ovl_dir_cache(d_inode(dentry
)) == cache
)
246 ovl_set_dir_cache(d_inode(dentry
), NULL
);
248 ovl_cache_free(&cache
->entries
);
253 static int ovl_fill_merge(struct dir_context
*ctx
, const char *name
,
254 int namelen
, loff_t offset
, u64 ino
,
257 struct ovl_readdir_data
*rdd
=
258 container_of(ctx
, struct ovl_readdir_data
, ctx
);
262 return ovl_cache_entry_add_rb(rdd
, name
, namelen
, ino
, d_type
);
264 return ovl_fill_lowest(rdd
, name
, namelen
, offset
, ino
, d_type
);
267 static int ovl_check_whiteouts(struct dentry
*dir
, struct ovl_readdir_data
*rdd
)
270 struct ovl_cache_entry
*p
;
271 struct dentry
*dentry
;
272 const struct cred
*old_cred
;
274 old_cred
= ovl_override_creds(rdd
->dentry
->d_sb
);
276 err
= down_write_killable(&dir
->d_inode
->i_rwsem
);
278 while (rdd
->first_maybe_whiteout
) {
279 p
= rdd
->first_maybe_whiteout
;
280 rdd
->first_maybe_whiteout
= p
->next_maybe_whiteout
;
281 dentry
= lookup_one_len(p
->name
, dir
, p
->len
);
282 if (!IS_ERR(dentry
)) {
283 p
->is_whiteout
= ovl_is_whiteout(dentry
);
287 inode_unlock(dir
->d_inode
);
289 revert_creds(old_cred
);
294 static inline int ovl_dir_read(struct path
*realpath
,
295 struct ovl_readdir_data
*rdd
)
297 struct file
*realfile
;
300 realfile
= ovl_path_open(realpath
, O_RDONLY
| O_LARGEFILE
);
301 if (IS_ERR(realfile
))
302 return PTR_ERR(realfile
);
304 rdd
->first_maybe_whiteout
= NULL
;
309 err
= iterate_dir(realfile
, &rdd
->ctx
);
312 } while (!err
&& rdd
->count
);
314 if (!err
&& rdd
->first_maybe_whiteout
&& rdd
->dentry
)
315 err
= ovl_check_whiteouts(realpath
->dentry
, rdd
);
323 * Can we iterate real dir directly?
325 * Non-merge dir may contain whiteouts from a time it was a merge upper, before
326 * lower dir was removed under it and possibly before it was rotated from upper
329 static bool ovl_dir_is_real(struct dentry
*dir
)
331 return !ovl_test_flag(OVL_WHITEOUTS
, d_inode(dir
));
334 static void ovl_dir_reset(struct file
*file
)
336 struct ovl_dir_file
*od
= file
->private_data
;
337 struct ovl_dir_cache
*cache
= od
->cache
;
338 struct dentry
*dentry
= file
->f_path
.dentry
;
341 if (cache
&& ovl_dentry_version_get(dentry
) != cache
->version
) {
342 ovl_cache_put(od
, dentry
);
346 is_real
= ovl_dir_is_real(dentry
);
347 if (od
->is_real
!= is_real
) {
348 /* is_real can only become false when dir is copied up */
349 if (WARN_ON(is_real
))
355 static int ovl_dir_read_merged(struct dentry
*dentry
, struct list_head
*list
,
356 struct rb_root
*root
)
359 struct path realpath
;
360 struct ovl_readdir_data rdd
= {
361 .ctx
.actor
= ovl_fill_merge
,
369 for (idx
= 0; idx
!= -1; idx
= next
) {
370 next
= ovl_path_next(idx
, dentry
, &realpath
);
371 rdd
.is_upper
= ovl_dentry_upper(dentry
) == realpath
.dentry
;
374 err
= ovl_dir_read(&realpath
, &rdd
);
379 * Insert lowest layer entries before upper ones, this
380 * allows offsets to be reasonably constant
382 list_add(&rdd
.middle
, rdd
.list
);
383 rdd
.is_lowest
= true;
384 err
= ovl_dir_read(&realpath
, &rdd
);
385 list_del(&rdd
.middle
);
391 static void ovl_seek_cursor(struct ovl_dir_file
*od
, loff_t pos
)
396 list_for_each(p
, &od
->cache
->entries
) {
401 /* Cursor is safe since the cache is stable */
405 static struct ovl_dir_cache
*ovl_cache_get(struct dentry
*dentry
)
408 struct ovl_dir_cache
*cache
;
410 cache
= ovl_dir_cache(d_inode(dentry
));
411 if (cache
&& ovl_dentry_version_get(dentry
) == cache
->version
) {
412 WARN_ON(!cache
->refcount
);
416 ovl_set_dir_cache(d_inode(dentry
), NULL
);
418 cache
= kzalloc(sizeof(struct ovl_dir_cache
), GFP_KERNEL
);
420 return ERR_PTR(-ENOMEM
);
423 INIT_LIST_HEAD(&cache
->entries
);
424 cache
->root
= RB_ROOT
;
426 res
= ovl_dir_read_merged(dentry
, &cache
->entries
, &cache
->root
);
428 ovl_cache_free(&cache
->entries
);
433 cache
->version
= ovl_dentry_version_get(dentry
);
434 ovl_set_dir_cache(d_inode(dentry
), cache
);
439 /* Map inode number to lower fs unique range */
440 static u64
ovl_remap_lower_ino(u64 ino
, int xinobits
, int fsid
,
441 const char *name
, int namelen
, bool warn
)
443 unsigned int xinoshift
= 64 - xinobits
;
445 if (unlikely(ino
>> xinoshift
)) {
447 pr_warn_ratelimited("d_ino too big (%.*s, ino=%llu, xinobits=%d)\n",
448 namelen
, name
, ino
, xinobits
);
454 * The lowest xinobit is reserved for mapping the non-peresistent inode
455 * numbers range, but this range is only exposed via st_ino, not here.
457 return ino
| ((u64
)fsid
) << (xinoshift
+ 1);
461 * Set d_ino for upper entries. Non-upper entries should always report
462 * the uppermost real inode ino and should not call this function.
464 * When not all layer are on same fs, report real ino also for upper.
466 * When all layers are on the same fs, and upper has a reference to
467 * copy up origin, call vfs_getattr() on the overlay entry to make
468 * sure that d_ino will be consistent with st_ino from stat(2).
470 static int ovl_cache_update_ino(struct path
*path
, struct ovl_cache_entry
*p
)
473 struct dentry
*dir
= path
->dentry
;
474 struct dentry
*this = NULL
;
475 enum ovl_path_type type
;
476 u64 ino
= p
->real_ino
;
477 int xinobits
= ovl_xino_bits(dir
->d_sb
);
480 if (!ovl_same_dev(dir
->d_sb
))
483 if (p
->name
[0] == '.') {
488 if (p
->len
== 2 && p
->name
[1] == '.') {
489 /* we shall not be moved */
490 this = dget(dir
->d_parent
);
494 this = lookup_one_len(p
->name
, dir
, p
->len
);
495 if (IS_ERR_OR_NULL(this) || !this->d_inode
) {
505 type
= ovl_path_type(this);
506 if (OVL_TYPE_ORIGIN(type
)) {
508 struct path statpath
= *path
;
510 statpath
.dentry
= this;
511 err
= vfs_getattr(&statpath
, &stat
, STATX_INO
, 0);
516 * Directory inode is always on overlay st_dev.
517 * Non-dir with ovl_same_dev() could be on pseudo st_dev in case
518 * of xino bits overflow.
520 WARN_ON_ONCE(S_ISDIR(stat
.mode
) &&
521 dir
->d_sb
->s_dev
!= stat
.dev
);
523 } else if (xinobits
&& !OVL_TYPE_UPPER(type
)) {
524 ino
= ovl_remap_lower_ino(ino
, xinobits
,
525 ovl_layer_lower(this)->fsid
,
527 ovl_xino_warn(dir
->d_sb
));
536 pr_warn_ratelimited("failed to look up (%s) for ino (%i)\n",
541 static int ovl_fill_plain(struct dir_context
*ctx
, const char *name
,
542 int namelen
, loff_t offset
, u64 ino
,
545 struct ovl_cache_entry
*p
;
546 struct ovl_readdir_data
*rdd
=
547 container_of(ctx
, struct ovl_readdir_data
, ctx
);
550 p
= ovl_cache_entry_new(rdd
, name
, namelen
, ino
, d_type
);
555 list_add_tail(&p
->l_node
, rdd
->list
);
560 static int ovl_dir_read_impure(struct path
*path
, struct list_head
*list
,
561 struct rb_root
*root
)
564 struct path realpath
;
565 struct ovl_cache_entry
*p
, *n
;
566 struct ovl_readdir_data rdd
= {
567 .ctx
.actor
= ovl_fill_plain
,
572 INIT_LIST_HEAD(list
);
574 ovl_path_upper(path
->dentry
, &realpath
);
576 err
= ovl_dir_read(&realpath
, &rdd
);
580 list_for_each_entry_safe(p
, n
, list
, l_node
) {
581 if (strcmp(p
->name
, ".") != 0 &&
582 strcmp(p
->name
, "..") != 0) {
583 err
= ovl_cache_update_ino(path
, p
);
587 if (p
->ino
== p
->real_ino
) {
588 list_del(&p
->l_node
);
591 struct rb_node
**newp
= &root
->rb_node
;
592 struct rb_node
*parent
= NULL
;
594 if (WARN_ON(ovl_cache_entry_find_link(p
->name
, p
->len
,
598 rb_link_node(&p
->node
, parent
, newp
);
599 rb_insert_color(&p
->node
, root
);
605 static struct ovl_dir_cache
*ovl_cache_get_impure(struct path
*path
)
608 struct dentry
*dentry
= path
->dentry
;
609 struct ovl_fs
*ofs
= OVL_FS(dentry
->d_sb
);
610 struct ovl_dir_cache
*cache
;
612 cache
= ovl_dir_cache(d_inode(dentry
));
613 if (cache
&& ovl_dentry_version_get(dentry
) == cache
->version
)
616 /* Impure cache is not refcounted, free it here */
617 ovl_dir_cache_free(d_inode(dentry
));
618 ovl_set_dir_cache(d_inode(dentry
), NULL
);
620 cache
= kzalloc(sizeof(struct ovl_dir_cache
), GFP_KERNEL
);
622 return ERR_PTR(-ENOMEM
);
624 res
= ovl_dir_read_impure(path
, &cache
->entries
, &cache
->root
);
626 ovl_cache_free(&cache
->entries
);
630 if (list_empty(&cache
->entries
)) {
632 * A good opportunity to get rid of an unneeded "impure" flag.
633 * Removing the "impure" xattr is best effort.
635 if (!ovl_want_write(dentry
)) {
636 ovl_do_removexattr(ofs
, ovl_dentry_upper(dentry
),
638 ovl_drop_write(dentry
);
640 ovl_clear_flag(OVL_IMPURE
, d_inode(dentry
));
645 cache
->version
= ovl_dentry_version_get(dentry
);
646 ovl_set_dir_cache(d_inode(dentry
), cache
);
651 struct ovl_readdir_translate
{
652 struct dir_context
*orig_ctx
;
653 struct ovl_dir_cache
*cache
;
654 struct dir_context ctx
;
661 static int ovl_fill_real(struct dir_context
*ctx
, const char *name
,
662 int namelen
, loff_t offset
, u64 ino
,
665 struct ovl_readdir_translate
*rdt
=
666 container_of(ctx
, struct ovl_readdir_translate
, ctx
);
667 struct dir_context
*orig_ctx
= rdt
->orig_ctx
;
669 if (rdt
->parent_ino
&& strcmp(name
, "..") == 0) {
670 ino
= rdt
->parent_ino
;
671 } else if (rdt
->cache
) {
672 struct ovl_cache_entry
*p
;
674 p
= ovl_cache_entry_find(&rdt
->cache
->root
, name
, namelen
);
677 } else if (rdt
->xinobits
) {
678 ino
= ovl_remap_lower_ino(ino
, rdt
->xinobits
, rdt
->fsid
,
679 name
, namelen
, rdt
->xinowarn
);
682 return orig_ctx
->actor(orig_ctx
, name
, namelen
, offset
, ino
, d_type
);
685 static bool ovl_is_impure_dir(struct file
*file
)
687 struct ovl_dir_file
*od
= file
->private_data
;
688 struct inode
*dir
= d_inode(file
->f_path
.dentry
);
691 * Only upper dir can be impure, but if we are in the middle of
692 * iterating a lower real dir, dir could be copied up and marked
693 * impure. We only want the impure cache if we started iterating
694 * a real upper dir to begin with.
696 return od
->is_upper
&& ovl_test_flag(OVL_IMPURE
, dir
);
700 static int ovl_iterate_real(struct file
*file
, struct dir_context
*ctx
)
703 struct ovl_dir_file
*od
= file
->private_data
;
704 struct dentry
*dir
= file
->f_path
.dentry
;
705 const struct ovl_layer
*lower_layer
= ovl_layer_lower(dir
);
706 struct ovl_readdir_translate rdt
= {
707 .ctx
.actor
= ovl_fill_real
,
709 .xinobits
= ovl_xino_bits(dir
->d_sb
),
710 .xinowarn
= ovl_xino_warn(dir
->d_sb
),
713 if (rdt
.xinobits
&& lower_layer
)
714 rdt
.fsid
= lower_layer
->fsid
;
716 if (OVL_TYPE_MERGE(ovl_path_type(dir
->d_parent
))) {
718 struct path statpath
= file
->f_path
;
720 statpath
.dentry
= dir
->d_parent
;
721 err
= vfs_getattr(&statpath
, &stat
, STATX_INO
, 0);
725 WARN_ON_ONCE(dir
->d_sb
->s_dev
!= stat
.dev
);
726 rdt
.parent_ino
= stat
.ino
;
729 if (ovl_is_impure_dir(file
)) {
730 rdt
.cache
= ovl_cache_get_impure(&file
->f_path
);
731 if (IS_ERR(rdt
.cache
))
732 return PTR_ERR(rdt
.cache
);
735 err
= iterate_dir(od
->realfile
, &rdt
.ctx
);
736 ctx
->pos
= rdt
.ctx
.pos
;
742 static int ovl_iterate(struct file
*file
, struct dir_context
*ctx
)
744 struct ovl_dir_file
*od
= file
->private_data
;
745 struct dentry
*dentry
= file
->f_path
.dentry
;
746 struct ovl_cache_entry
*p
;
747 const struct cred
*old_cred
;
750 old_cred
= ovl_override_creds(dentry
->d_sb
);
756 * If parent is merge, then need to adjust d_ino for '..', if
757 * dir is impure then need to adjust d_ino for copied up
760 if (ovl_xino_bits(dentry
->d_sb
) ||
761 (ovl_same_fs(dentry
->d_sb
) &&
762 (ovl_is_impure_dir(file
) ||
763 OVL_TYPE_MERGE(ovl_path_type(dentry
->d_parent
))))) {
764 err
= ovl_iterate_real(file
, ctx
);
766 err
= iterate_dir(od
->realfile
, ctx
);
772 struct ovl_dir_cache
*cache
;
774 cache
= ovl_cache_get(dentry
);
775 err
= PTR_ERR(cache
);
780 ovl_seek_cursor(od
, ctx
->pos
);
783 while (od
->cursor
!= &od
->cache
->entries
) {
784 p
= list_entry(od
->cursor
, struct ovl_cache_entry
, l_node
);
785 if (!p
->is_whiteout
) {
787 err
= ovl_cache_update_ino(&file
->f_path
, p
);
791 if (!dir_emit(ctx
, p
->name
, p
->len
, p
->ino
, p
->type
))
794 od
->cursor
= p
->l_node
.next
;
799 revert_creds(old_cred
);
803 static loff_t
ovl_dir_llseek(struct file
*file
, loff_t offset
, int origin
)
806 struct ovl_dir_file
*od
= file
->private_data
;
808 inode_lock(file_inode(file
));
813 res
= vfs_llseek(od
->realfile
, offset
, origin
);
814 file
->f_pos
= od
->realfile
->f_pos
;
820 offset
+= file
->f_pos
;
830 if (offset
!= file
->f_pos
) {
831 file
->f_pos
= offset
;
833 ovl_seek_cursor(od
, offset
);
838 inode_unlock(file_inode(file
));
843 static struct file
*ovl_dir_open_realfile(const struct file
*file
,
844 struct path
*realpath
)
847 const struct cred
*old_cred
;
849 old_cred
= ovl_override_creds(file_inode(file
)->i_sb
);
850 res
= ovl_path_open(realpath
, O_RDONLY
| (file
->f_flags
& O_LARGEFILE
));
851 revert_creds(old_cred
);
857 * Like ovl_real_fdget(), returns upperfile if dir was copied up since open.
858 * Unlike ovl_real_fdget(), this caches upperfile in file->private_data.
860 * TODO: use same abstract type for file->private_data of dir and file so
861 * upperfile could also be cached for files as well.
863 struct file
*ovl_dir_real_file(const struct file
*file
, bool want_upper
)
866 struct ovl_dir_file
*od
= file
->private_data
;
867 struct dentry
*dentry
= file
->f_path
.dentry
;
868 struct file
*old
, *realfile
= od
->realfile
;
870 if (!OVL_TYPE_UPPER(ovl_path_type(dentry
)))
871 return want_upper
? NULL
: realfile
;
874 * Need to check if we started out being a lower dir, but got copied up
877 realfile
= READ_ONCE(od
->upperfile
);
879 struct path upperpath
;
881 ovl_path_upper(dentry
, &upperpath
);
882 realfile
= ovl_dir_open_realfile(file
, &upperpath
);
883 if (IS_ERR(realfile
))
886 old
= cmpxchg_release(&od
->upperfile
, NULL
, realfile
);
897 static int ovl_dir_fsync(struct file
*file
, loff_t start
, loff_t end
,
900 struct file
*realfile
;
903 err
= ovl_sync_status(OVL_FS(file
->f_path
.dentry
->d_sb
));
907 realfile
= ovl_dir_real_file(file
, true);
908 err
= PTR_ERR_OR_ZERO(realfile
);
910 /* Nothing to sync for lower */
911 if (!realfile
|| err
)
914 return vfs_fsync_range(realfile
, start
, end
, datasync
);
917 static int ovl_dir_release(struct inode
*inode
, struct file
*file
)
919 struct ovl_dir_file
*od
= file
->private_data
;
923 ovl_cache_put(od
, file
->f_path
.dentry
);
934 static int ovl_dir_open(struct inode
*inode
, struct file
*file
)
936 struct path realpath
;
937 struct file
*realfile
;
938 struct ovl_dir_file
*od
;
939 enum ovl_path_type type
;
941 od
= kzalloc(sizeof(struct ovl_dir_file
), GFP_KERNEL
);
945 type
= ovl_path_real(file
->f_path
.dentry
, &realpath
);
946 realfile
= ovl_dir_open_realfile(file
, &realpath
);
947 if (IS_ERR(realfile
)) {
949 return PTR_ERR(realfile
);
951 od
->realfile
= realfile
;
952 od
->is_real
= ovl_dir_is_real(file
->f_path
.dentry
);
953 od
->is_upper
= OVL_TYPE_UPPER(type
);
954 file
->private_data
= od
;
959 const struct file_operations ovl_dir_operations
= {
960 .read
= generic_read_dir
,
961 .open
= ovl_dir_open
,
962 .iterate
= ovl_iterate
,
963 .llseek
= ovl_dir_llseek
,
964 .fsync
= ovl_dir_fsync
,
965 .release
= ovl_dir_release
,
966 .unlocked_ioctl
= ovl_ioctl
,
968 .compat_ioctl
= ovl_compat_ioctl
,
972 int ovl_check_empty_dir(struct dentry
*dentry
, struct list_head
*list
)
975 struct ovl_cache_entry
*p
, *n
;
976 struct rb_root root
= RB_ROOT
;
977 const struct cred
*old_cred
;
979 old_cred
= ovl_override_creds(dentry
->d_sb
);
980 err
= ovl_dir_read_merged(dentry
, list
, &root
);
981 revert_creds(old_cred
);
987 list_for_each_entry_safe(p
, n
, list
, l_node
) {
989 * Select whiteouts in upperdir, they should
990 * be cleared when deleting this directory.
992 if (p
->is_whiteout
) {
998 if (p
->name
[0] == '.') {
1001 if (p
->len
== 2 && p
->name
[1] == '.')
1008 list_del(&p
->l_node
);
1015 void ovl_cleanup_whiteouts(struct dentry
*upper
, struct list_head
*list
)
1017 struct ovl_cache_entry
*p
;
1019 inode_lock_nested(upper
->d_inode
, I_MUTEX_CHILD
);
1020 list_for_each_entry(p
, list
, l_node
) {
1021 struct dentry
*dentry
;
1023 if (WARN_ON(!p
->is_whiteout
|| !p
->is_upper
))
1026 dentry
= lookup_one_len(p
->name
, upper
, p
->len
);
1027 if (IS_ERR(dentry
)) {
1028 pr_err("lookup '%s/%.*s' failed (%i)\n",
1029 upper
->d_name
.name
, p
->len
, p
->name
,
1030 (int) PTR_ERR(dentry
));
1033 if (dentry
->d_inode
)
1034 ovl_cleanup(upper
->d_inode
, dentry
);
1037 inode_unlock(upper
->d_inode
);
1040 static int ovl_check_d_type(struct dir_context
*ctx
, const char *name
,
1041 int namelen
, loff_t offset
, u64 ino
,
1042 unsigned int d_type
)
1044 struct ovl_readdir_data
*rdd
=
1045 container_of(ctx
, struct ovl_readdir_data
, ctx
);
1047 /* Even if d_type is not supported, DT_DIR is returned for . and .. */
1048 if (!strncmp(name
, ".", namelen
) || !strncmp(name
, "..", namelen
))
1051 if (d_type
!= DT_UNKNOWN
)
1052 rdd
->d_type_supported
= true;
1058 * Returns 1 if d_type is supported, 0 not supported/unknown. Negative values
1059 * if error is encountered.
1061 int ovl_check_d_type_supported(struct path
*realpath
)
1064 struct ovl_readdir_data rdd
= {
1065 .ctx
.actor
= ovl_check_d_type
,
1066 .d_type_supported
= false,
1069 err
= ovl_dir_read(realpath
, &rdd
);
1073 return rdd
.d_type_supported
;
1076 #define OVL_INCOMPATDIR_NAME "incompat"
1078 static int ovl_workdir_cleanup_recurse(struct path
*path
, int level
)
1081 struct inode
*dir
= path
->dentry
->d_inode
;
1083 struct rb_root root
= RB_ROOT
;
1084 struct ovl_cache_entry
*p
;
1085 struct ovl_readdir_data rdd
= {
1086 .ctx
.actor
= ovl_fill_merge
,
1092 bool incompat
= false;
1095 * The "work/incompat" directory is treated specially - if it is not
1096 * empty, instead of printing a generic error and mounting read-only,
1097 * we will error about incompat features and fail the mount.
1099 * When called from ovl_indexdir_cleanup(), path->dentry->d_name.name
1103 !strcmp(path
->dentry
->d_name
.name
, OVL_INCOMPATDIR_NAME
))
1106 err
= ovl_dir_read(path
, &rdd
);
1110 inode_lock_nested(dir
, I_MUTEX_PARENT
);
1111 list_for_each_entry(p
, &list
, l_node
) {
1112 struct dentry
*dentry
;
1114 if (p
->name
[0] == '.') {
1117 if (p
->len
== 2 && p
->name
[1] == '.')
1119 } else if (incompat
) {
1120 pr_err("overlay with incompat feature '%s' cannot be mounted\n",
1125 dentry
= lookup_one_len(p
->name
, path
->dentry
, p
->len
);
1128 if (dentry
->d_inode
)
1129 err
= ovl_workdir_cleanup(dir
, path
->mnt
, dentry
, level
);
1136 ovl_cache_free(&list
);
1140 int ovl_workdir_cleanup(struct inode
*dir
, struct vfsmount
*mnt
,
1141 struct dentry
*dentry
, int level
)
1145 if (!d_is_dir(dentry
) || level
> 1) {
1146 return ovl_cleanup(dir
, dentry
);
1149 err
= ovl_do_rmdir(dir
, dentry
);
1151 struct path path
= { .mnt
= mnt
, .dentry
= dentry
};
1154 err
= ovl_workdir_cleanup_recurse(&path
, level
+ 1);
1155 inode_lock_nested(dir
, I_MUTEX_PARENT
);
1157 err
= ovl_cleanup(dir
, dentry
);
1163 int ovl_indexdir_cleanup(struct ovl_fs
*ofs
)
1166 struct dentry
*indexdir
= ofs
->indexdir
;
1167 struct dentry
*index
= NULL
;
1168 struct inode
*dir
= indexdir
->d_inode
;
1169 struct path path
= { .mnt
= ovl_upper_mnt(ofs
), .dentry
= indexdir
};
1171 struct rb_root root
= RB_ROOT
;
1172 struct ovl_cache_entry
*p
;
1173 struct ovl_readdir_data rdd
= {
1174 .ctx
.actor
= ovl_fill_merge
,
1181 err
= ovl_dir_read(&path
, &rdd
);
1185 inode_lock_nested(dir
, I_MUTEX_PARENT
);
1186 list_for_each_entry(p
, &list
, l_node
) {
1187 if (p
->name
[0] == '.') {
1190 if (p
->len
== 2 && p
->name
[1] == '.')
1193 index
= lookup_one_len(p
->name
, indexdir
, p
->len
);
1194 if (IS_ERR(index
)) {
1195 err
= PTR_ERR(index
);
1199 /* Cleanup leftover from index create/cleanup attempt */
1200 if (index
->d_name
.name
[0] == '#') {
1201 err
= ovl_workdir_cleanup(dir
, path
.mnt
, index
, 1);
1206 err
= ovl_verify_index(ofs
, index
);
1209 } else if (err
== -ESTALE
) {
1210 /* Cleanup stale index entries */
1211 err
= ovl_cleanup(dir
, index
);
1212 } else if (err
!= -ENOENT
) {
1214 * Abort mount to avoid corrupting the index if
1215 * an incompatible index entry was found or on out
1219 } else if (ofs
->config
.nfs_export
) {
1221 * Whiteout orphan index to block future open by
1222 * handle after overlay nlink dropped to zero.
1224 err
= ovl_cleanup_and_whiteout(ofs
, dir
, index
);
1226 /* Cleanup orphan index entries */
1227 err
= ovl_cleanup(dir
, index
);
1240 ovl_cache_free(&list
);
1242 pr_err("failed index dir cleanup (%i)\n", err
);