1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (C) 2011 Novell Inc.
8 #include <linux/slab.h>
9 #include <linux/cred.h>
10 #include <linux/xattr.h>
11 #include <linux/posix_acl.h>
12 #include <linux/ratelimit.h>
13 #include "overlayfs.h"
16 int ovl_setattr(struct dentry
*dentry
, struct iattr
*attr
)
19 bool full_copy_up
= false;
20 struct dentry
*upperdentry
;
21 const struct cred
*old_cred
;
23 err
= setattr_prepare(dentry
, attr
);
27 err
= ovl_want_write(dentry
);
31 if (attr
->ia_valid
& ATTR_SIZE
) {
32 struct inode
*realinode
= d_inode(ovl_dentry_real(dentry
));
35 if (atomic_read(&realinode
->i_writecount
) < 0)
38 /* Truncate should trigger data copy up as well */
43 err
= ovl_copy_up(dentry
);
45 err
= ovl_copy_up_with_data(dentry
);
47 struct inode
*winode
= NULL
;
49 upperdentry
= ovl_dentry_upper(dentry
);
51 if (attr
->ia_valid
& ATTR_SIZE
) {
52 winode
= d_inode(upperdentry
);
53 err
= get_write_access(winode
);
58 if (attr
->ia_valid
& (ATTR_KILL_SUID
|ATTR_KILL_SGID
))
59 attr
->ia_valid
&= ~ATTR_MODE
;
61 inode_lock(upperdentry
->d_inode
);
62 old_cred
= ovl_override_creds(dentry
->d_sb
);
63 err
= notify_change(upperdentry
, attr
, NULL
);
64 revert_creds(old_cred
);
66 ovl_copyattr(upperdentry
->d_inode
, dentry
->d_inode
);
67 inode_unlock(upperdentry
->d_inode
);
70 put_write_access(winode
);
73 ovl_drop_write(dentry
);
78 static int ovl_map_dev_ino(struct dentry
*dentry
, struct kstat
*stat
,
79 struct ovl_layer
*lower_layer
)
81 bool samefs
= ovl_same_sb(dentry
->d_sb
);
82 unsigned int xinobits
= ovl_xino_bits(dentry
->d_sb
);
86 * When all layers are on the same fs, all real inode
87 * number are unique, so we use the overlay st_dev,
88 * which is friendly to du -x.
90 stat
->dev
= dentry
->d_sb
->s_dev
;
92 } else if (xinobits
) {
93 unsigned int shift
= 64 - xinobits
;
95 * All inode numbers of underlying fs should not be using the
96 * high xinobits, so we use high xinobits to partition the
97 * overlay st_ino address space. The high bits holds the fsid
98 * (upper fsid is 0). This way overlay inode numbers are unique
99 * and all inodes use overlay st_dev. Inode numbers are also
100 * persistent for a given layer configuration.
102 if (stat
->ino
>> shift
) {
103 pr_warn_ratelimited("overlayfs: inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
104 dentry
, stat
->ino
, xinobits
);
107 stat
->ino
|= ((u64
)lower_layer
->fsid
) << shift
;
109 stat
->dev
= dentry
->d_sb
->s_dev
;
114 /* The inode could not be mapped to a unified st_ino address space */
115 if (S_ISDIR(dentry
->d_inode
->i_mode
)) {
117 * Always use the overlay st_dev for directories, so 'find
118 * -xdev' will scan the entire overlay mount and won't cross the
119 * overlay mount boundaries.
121 * If not all layers are on the same fs the pair {real st_ino;
122 * overlay st_dev} is not unique, so use the non persistent
123 * overlay st_ino for directories.
125 stat
->dev
= dentry
->d_sb
->s_dev
;
126 stat
->ino
= dentry
->d_inode
->i_ino
;
127 } else if (lower_layer
&& lower_layer
->fsid
) {
129 * For non-samefs setup, if we cannot map all layers st_ino
130 * to a unified address space, we need to make sure that st_dev
131 * is unique per lower fs. Upper layer uses real st_dev and
132 * lower layers use the unique anonymous bdev assigned to the
135 stat
->dev
= lower_layer
->fs
->pseudo_dev
;
141 int ovl_getattr(const struct path
*path
, struct kstat
*stat
,
142 u32 request_mask
, unsigned int flags
)
144 struct dentry
*dentry
= path
->dentry
;
145 enum ovl_path_type type
;
146 struct path realpath
;
147 const struct cred
*old_cred
;
148 bool is_dir
= S_ISDIR(dentry
->d_inode
->i_mode
);
149 bool samefs
= ovl_same_sb(dentry
->d_sb
);
150 struct ovl_layer
*lower_layer
= NULL
;
152 bool metacopy_blocks
= false;
154 metacopy_blocks
= ovl_is_metacopy_dentry(dentry
);
156 type
= ovl_path_real(dentry
, &realpath
);
157 old_cred
= ovl_override_creds(dentry
->d_sb
);
158 err
= vfs_getattr(&realpath
, stat
, request_mask
, flags
);
163 * For non-dir or same fs, we use st_ino of the copy up origin.
164 * This guaranties constant st_dev/st_ino across copy up.
165 * With xino feature and non-samefs, we use st_ino of the copy up
166 * origin masked with high bits that represent the layer id.
168 * If lower filesystem supports NFS file handles, this also guaranties
169 * persistent st_ino across mount cycle.
171 if (!is_dir
|| samefs
|| ovl_xino_bits(dentry
->d_sb
)) {
172 if (!OVL_TYPE_UPPER(type
)) {
173 lower_layer
= ovl_layer_lower(dentry
);
174 } else if (OVL_TYPE_ORIGIN(type
)) {
175 struct kstat lowerstat
;
176 u32 lowermask
= STATX_INO
| STATX_BLOCKS
|
177 (!is_dir
? STATX_NLINK
: 0);
179 ovl_path_lower(dentry
, &realpath
);
180 err
= vfs_getattr(&realpath
, &lowerstat
,
186 * Lower hardlinks may be broken on copy up to different
187 * upper files, so we cannot use the lower origin st_ino
188 * for those different files, even for the same fs case.
190 * Similarly, several redirected dirs can point to the
191 * same dir on a lower layer. With the "verify_lower"
192 * feature, we do not use the lower origin st_ino, if
193 * we haven't verified that this redirect is unique.
195 * With inodes index enabled, it is safe to use st_ino
196 * of an indexed origin. The index validates that the
197 * upper hardlink is not broken and that a redirected
198 * dir is the only redirect to that origin.
200 if (ovl_test_flag(OVL_INDEX
, d_inode(dentry
)) ||
201 (!ovl_verify_lower(dentry
->d_sb
) &&
202 (is_dir
|| lowerstat
.nlink
== 1))) {
203 stat
->ino
= lowerstat
.ino
;
204 lower_layer
= ovl_layer_lower(dentry
);
208 * If we are querying a metacopy dentry and lower
209 * dentry is data dentry, then use the blocks we
210 * queried just now. We don't have to do additional
211 * vfs_getattr(). If lower itself is metacopy, then
212 * additional vfs_getattr() is unavoidable.
214 if (metacopy_blocks
&&
215 realpath
.dentry
== ovl_dentry_lowerdata(dentry
)) {
216 stat
->blocks
= lowerstat
.blocks
;
217 metacopy_blocks
= false;
221 if (metacopy_blocks
) {
223 * If lower is not same as lowerdata or if there was
224 * no origin on upper, we can end up here.
226 struct kstat lowerdatastat
;
227 u32 lowermask
= STATX_BLOCKS
;
229 ovl_path_lowerdata(dentry
, &realpath
);
230 err
= vfs_getattr(&realpath
, &lowerdatastat
,
234 stat
->blocks
= lowerdatastat
.blocks
;
238 err
= ovl_map_dev_ino(dentry
, stat
, lower_layer
);
243 * It's probably not worth it to count subdirs to get the
244 * correct link count. nlink=1 seems to pacify 'find' and
247 if (is_dir
&& OVL_TYPE_MERGE(type
))
251 * Return the overlay inode nlinks for indexed upper inodes.
252 * Overlay inode nlink counts the union of the upper hardlinks
253 * and non-covered lower hardlinks. It does not include the upper
256 if (!is_dir
&& ovl_test_flag(OVL_INDEX
, d_inode(dentry
)))
257 stat
->nlink
= dentry
->d_inode
->i_nlink
;
260 revert_creds(old_cred
);
265 int ovl_permission(struct inode
*inode
, int mask
)
267 struct inode
*upperinode
= ovl_inode_upper(inode
);
268 struct inode
*realinode
= upperinode
?: ovl_inode_lower(inode
);
269 const struct cred
*old_cred
;
272 /* Careful in RCU walk mode */
274 WARN_ON(!(mask
& MAY_NOT_BLOCK
));
279 * Check overlay inode with the creds of task and underlying inode
280 * with creds of mounter
282 err
= generic_permission(inode
, mask
);
286 old_cred
= ovl_override_creds(inode
->i_sb
);
288 !special_file(realinode
->i_mode
) && mask
& MAY_WRITE
) {
289 mask
&= ~(MAY_WRITE
| MAY_APPEND
);
290 /* Make sure mounter can read file for copy up later */
293 err
= inode_permission(realinode
, mask
);
294 revert_creds(old_cred
);
299 static const char *ovl_get_link(struct dentry
*dentry
,
301 struct delayed_call
*done
)
303 const struct cred
*old_cred
;
307 return ERR_PTR(-ECHILD
);
309 old_cred
= ovl_override_creds(dentry
->d_sb
);
310 p
= vfs_get_link(ovl_dentry_real(dentry
), done
);
311 revert_creds(old_cred
);
315 bool ovl_is_private_xattr(const char *name
)
317 return strncmp(name
, OVL_XATTR_PREFIX
,
318 sizeof(OVL_XATTR_PREFIX
) - 1) == 0;
321 int ovl_xattr_set(struct dentry
*dentry
, struct inode
*inode
, const char *name
,
322 const void *value
, size_t size
, int flags
)
325 struct dentry
*upperdentry
= ovl_i_dentry_upper(inode
);
326 struct dentry
*realdentry
= upperdentry
?: ovl_dentry_lower(dentry
);
327 const struct cred
*old_cred
;
329 err
= ovl_want_write(dentry
);
333 if (!value
&& !upperdentry
) {
334 err
= vfs_getxattr(realdentry
, name
, NULL
, 0);
340 err
= ovl_copy_up(dentry
);
344 realdentry
= ovl_dentry_upper(dentry
);
347 old_cred
= ovl_override_creds(dentry
->d_sb
);
349 err
= vfs_setxattr(realdentry
, name
, value
, size
, flags
);
351 WARN_ON(flags
!= XATTR_REPLACE
);
352 err
= vfs_removexattr(realdentry
, name
);
354 revert_creds(old_cred
);
357 ovl_copyattr(d_inode(realdentry
), inode
);
360 ovl_drop_write(dentry
);
365 int ovl_xattr_get(struct dentry
*dentry
, struct inode
*inode
, const char *name
,
366 void *value
, size_t size
)
369 const struct cred
*old_cred
;
370 struct dentry
*realdentry
=
371 ovl_i_dentry_upper(inode
) ?: ovl_dentry_lower(dentry
);
373 old_cred
= ovl_override_creds(dentry
->d_sb
);
374 res
= vfs_getxattr(realdentry
, name
, value
, size
);
375 revert_creds(old_cred
);
379 static bool ovl_can_list(const char *s
)
381 /* List all non-trusted xatts */
382 if (strncmp(s
, XATTR_TRUSTED_PREFIX
, XATTR_TRUSTED_PREFIX_LEN
) != 0)
385 /* Never list trusted.overlay, list other trusted for superuser only */
386 return !ovl_is_private_xattr(s
) &&
387 ns_capable_noaudit(&init_user_ns
, CAP_SYS_ADMIN
);
390 ssize_t
ovl_listxattr(struct dentry
*dentry
, char *list
, size_t size
)
392 struct dentry
*realdentry
= ovl_dentry_real(dentry
);
396 const struct cred
*old_cred
;
398 old_cred
= ovl_override_creds(dentry
->d_sb
);
399 res
= vfs_listxattr(realdentry
, list
, size
);
400 revert_creds(old_cred
);
401 if (res
<= 0 || size
== 0)
404 /* filter out private xattrs */
405 for (s
= list
, len
= res
; len
;) {
406 size_t slen
= strnlen(s
, len
) + 1;
408 /* underlying fs providing us with an broken xattr list? */
409 if (WARN_ON(slen
> len
))
413 if (!ovl_can_list(s
)) {
415 memmove(s
, s
+ slen
, len
);
424 struct posix_acl
*ovl_get_acl(struct inode
*inode
, int type
)
426 struct inode
*realinode
= ovl_inode_real(inode
);
427 const struct cred
*old_cred
;
428 struct posix_acl
*acl
;
430 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL
) || !IS_POSIXACL(realinode
))
433 old_cred
= ovl_override_creds(inode
->i_sb
);
434 acl
= get_acl(realinode
, type
);
435 revert_creds(old_cred
);
440 int ovl_update_time(struct inode
*inode
, struct timespec64
*ts
, int flags
)
442 if (flags
& S_ATIME
) {
443 struct ovl_fs
*ofs
= inode
->i_sb
->s_fs_info
;
444 struct path upperpath
= {
445 .mnt
= ofs
->upper_mnt
,
446 .dentry
= ovl_upperdentry_dereference(OVL_I(inode
)),
449 if (upperpath
.dentry
) {
450 touch_atime(&upperpath
);
451 inode
->i_atime
= d_inode(upperpath
.dentry
)->i_atime
;
457 static int ovl_fiemap(struct inode
*inode
, struct fiemap_extent_info
*fieinfo
,
461 struct inode
*realinode
= ovl_inode_real(inode
);
462 const struct cred
*old_cred
;
464 if (!realinode
->i_op
->fiemap
)
467 old_cred
= ovl_override_creds(inode
->i_sb
);
469 if (fieinfo
->fi_flags
& FIEMAP_FLAG_SYNC
)
470 filemap_write_and_wait(realinode
->i_mapping
);
472 err
= realinode
->i_op
->fiemap(realinode
, fieinfo
, start
, len
);
473 revert_creds(old_cred
);
478 static const struct inode_operations ovl_file_inode_operations
= {
479 .setattr
= ovl_setattr
,
480 .permission
= ovl_permission
,
481 .getattr
= ovl_getattr
,
482 .listxattr
= ovl_listxattr
,
483 .get_acl
= ovl_get_acl
,
484 .update_time
= ovl_update_time
,
485 .fiemap
= ovl_fiemap
,
488 static const struct inode_operations ovl_symlink_inode_operations
= {
489 .setattr
= ovl_setattr
,
490 .get_link
= ovl_get_link
,
491 .getattr
= ovl_getattr
,
492 .listxattr
= ovl_listxattr
,
493 .update_time
= ovl_update_time
,
496 static const struct inode_operations ovl_special_inode_operations
= {
497 .setattr
= ovl_setattr
,
498 .permission
= ovl_permission
,
499 .getattr
= ovl_getattr
,
500 .listxattr
= ovl_listxattr
,
501 .get_acl
= ovl_get_acl
,
502 .update_time
= ovl_update_time
,
505 static const struct address_space_operations ovl_aops
= {
506 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
507 .direct_IO
= noop_direct_IO
,
511 * It is possible to stack overlayfs instance on top of another
512 * overlayfs instance as lower layer. We need to annonate the
513 * stackable i_mutex locks according to stack level of the super
514 * block instance. An overlayfs instance can never be in stack
515 * depth 0 (there is always a real fs below it). An overlayfs
516 * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
518 * For example, here is a snip from /proc/lockdep_chains after
519 * dir_iterate of nested overlayfs:
521 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
522 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
523 * [...] &type->i_mutex_dir_key (stack_depth=0)
525 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
527 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode
*inode
)
529 #ifdef CONFIG_LOCKDEP
530 static struct lock_class_key ovl_i_mutex_key
[OVL_MAX_NESTING
];
531 static struct lock_class_key ovl_i_mutex_dir_key
[OVL_MAX_NESTING
];
532 static struct lock_class_key ovl_i_lock_key
[OVL_MAX_NESTING
];
534 int depth
= inode
->i_sb
->s_stack_depth
- 1;
536 if (WARN_ON_ONCE(depth
< 0 || depth
>= OVL_MAX_NESTING
))
539 if (S_ISDIR(inode
->i_mode
))
540 lockdep_set_class(&inode
->i_rwsem
, &ovl_i_mutex_dir_key
[depth
]);
542 lockdep_set_class(&inode
->i_rwsem
, &ovl_i_mutex_key
[depth
]);
544 lockdep_set_class(&OVL_I(inode
)->lock
, &ovl_i_lock_key
[depth
]);
548 static void ovl_fill_inode(struct inode
*inode
, umode_t mode
, dev_t rdev
,
549 unsigned long ino
, int fsid
)
551 int xinobits
= ovl_xino_bits(inode
->i_sb
);
554 * When d_ino is consistent with st_ino (samefs or i_ino has enough
555 * bits to encode layer), set the same value used for st_ino to i_ino,
556 * so inode number exposed via /proc/locks and a like will be
557 * consistent with d_ino and st_ino values. An i_ino value inconsistent
558 * with d_ino also causes nfsd readdirplus to fail. When called from
559 * ovl_new_inode(), ino arg is 0, so i_ino will be updated to real
560 * upper inode i_ino on ovl_inode_init() or ovl_inode_update().
562 if (ovl_same_sb(inode
->i_sb
) || xinobits
) {
564 if (xinobits
&& fsid
&& !(ino
>> (64 - xinobits
)))
565 inode
->i_ino
|= (unsigned long)fsid
<< (64 - xinobits
);
567 inode
->i_ino
= get_next_ino();
569 inode
->i_mode
= mode
;
570 inode
->i_flags
|= S_NOCMTIME
;
571 #ifdef CONFIG_FS_POSIX_ACL
572 inode
->i_acl
= inode
->i_default_acl
= ACL_DONT_CACHE
;
575 ovl_lockdep_annotate_inode_mutex_key(inode
);
577 switch (mode
& S_IFMT
) {
579 inode
->i_op
= &ovl_file_inode_operations
;
580 inode
->i_fop
= &ovl_file_operations
;
581 inode
->i_mapping
->a_ops
= &ovl_aops
;
585 inode
->i_op
= &ovl_dir_inode_operations
;
586 inode
->i_fop
= &ovl_dir_operations
;
590 inode
->i_op
= &ovl_symlink_inode_operations
;
594 inode
->i_op
= &ovl_special_inode_operations
;
595 init_special_inode(inode
, mode
, rdev
);
601 * With inodes index enabled, an overlay inode nlink counts the union of upper
602 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
603 * upper inode, the following nlink modifying operations can happen:
605 * 1. Lower hardlink copy up
606 * 2. Upper hardlink created, unlinked or renamed over
607 * 3. Lower hardlink whiteout or renamed over
609 * For the first, copy up case, the union nlink does not change, whether the
610 * operation succeeds or fails, but the upper inode nlink may change.
611 * Therefore, before copy up, we store the union nlink value relative to the
612 * lower inode nlink in the index inode xattr trusted.overlay.nlink.
614 * For the second, upper hardlink case, the union nlink should be incremented
615 * or decremented IFF the operation succeeds, aligned with nlink change of the
616 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
617 * value relative to the upper inode nlink in the index inode.
619 * For the last, lower cover up case, we simplify things by preceding the
620 * whiteout or cover up with copy up. This makes sure that there is an index
621 * upper inode where the nlink xattr can be stored before the copied up upper
624 #define OVL_NLINK_ADD_UPPER (1 << 0)
627 * On-disk format for indexed nlink:
629 * nlink relative to the upper inode - "U[+-]NUM"
630 * nlink relative to the lower inode - "L[+-]NUM"
633 static int ovl_set_nlink_common(struct dentry
*dentry
,
634 struct dentry
*realdentry
, const char *format
)
636 struct inode
*inode
= d_inode(dentry
);
637 struct inode
*realinode
= d_inode(realdentry
);
641 len
= snprintf(buf
, sizeof(buf
), format
,
642 (int) (inode
->i_nlink
- realinode
->i_nlink
));
644 if (WARN_ON(len
>= sizeof(buf
)))
647 return ovl_do_setxattr(ovl_dentry_upper(dentry
),
648 OVL_XATTR_NLINK
, buf
, len
, 0);
651 int ovl_set_nlink_upper(struct dentry
*dentry
)
653 return ovl_set_nlink_common(dentry
, ovl_dentry_upper(dentry
), "U%+i");
656 int ovl_set_nlink_lower(struct dentry
*dentry
)
658 return ovl_set_nlink_common(dentry
, ovl_dentry_lower(dentry
), "L%+i");
661 unsigned int ovl_get_nlink(struct dentry
*lowerdentry
,
662 struct dentry
*upperdentry
,
663 unsigned int fallback
)
670 if (!lowerdentry
|| !upperdentry
|| d_inode(lowerdentry
)->i_nlink
== 1)
673 err
= vfs_getxattr(upperdentry
, OVL_XATTR_NLINK
, &buf
, sizeof(buf
) - 1);
678 if ((buf
[0] != 'L' && buf
[0] != 'U') ||
679 (buf
[1] != '+' && buf
[1] != '-'))
682 err
= kstrtoint(buf
+ 1, 10, &nlink_diff
);
686 nlink
= d_inode(buf
[0] == 'L' ? lowerdentry
: upperdentry
)->i_nlink
;
695 pr_warn_ratelimited("overlayfs: failed to get index nlink (%pd2, err=%i)\n",
700 struct inode
*ovl_new_inode(struct super_block
*sb
, umode_t mode
, dev_t rdev
)
704 inode
= new_inode(sb
);
706 ovl_fill_inode(inode
, mode
, rdev
, 0, 0);
711 static int ovl_inode_test(struct inode
*inode
, void *data
)
713 return inode
->i_private
== data
;
716 static int ovl_inode_set(struct inode
*inode
, void *data
)
718 inode
->i_private
= data
;
722 static bool ovl_verify_inode(struct inode
*inode
, struct dentry
*lowerdentry
,
723 struct dentry
*upperdentry
, bool strict
)
726 * For directories, @strict verify from lookup path performs consistency
727 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
728 * inode. Non @strict verify from NFS handle decode path passes NULL for
729 * 'unknown' lower/upper.
731 if (S_ISDIR(inode
->i_mode
) && strict
) {
732 /* Real lower dir moved to upper layer under us? */
733 if (!lowerdentry
&& ovl_inode_lower(inode
))
736 /* Lookup of an uncovered redirect origin? */
737 if (!upperdentry
&& ovl_inode_upper(inode
))
742 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
743 * This happens when finding a copied up overlay inode for a renamed
744 * or hardlinked overlay dentry and lower dentry cannot be followed
745 * by origin because lower fs does not support file handles.
747 if (lowerdentry
&& ovl_inode_lower(inode
) != d_inode(lowerdentry
))
751 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
752 * This happens when finding a lower alias for a copied up hard link.
754 if (upperdentry
&& ovl_inode_upper(inode
) != d_inode(upperdentry
))
760 struct inode
*ovl_lookup_inode(struct super_block
*sb
, struct dentry
*real
,
763 struct inode
*inode
, *key
= d_inode(real
);
765 inode
= ilookup5(sb
, (unsigned long) key
, ovl_inode_test
, key
);
769 if (!ovl_verify_inode(inode
, is_upper
? NULL
: real
,
770 is_upper
? real
: NULL
, false)) {
772 return ERR_PTR(-ESTALE
);
778 bool ovl_lookup_trap_inode(struct super_block
*sb
, struct dentry
*dir
)
780 struct inode
*key
= d_inode(dir
);
784 trap
= ilookup5(sb
, (unsigned long) key
, ovl_inode_test
, key
);
788 res
= IS_DEADDIR(trap
) && !ovl_inode_upper(trap
) &&
789 !ovl_inode_lower(trap
);
796 * Create an inode cache entry for layer root dir, that will intentionally
797 * fail ovl_verify_inode(), so any lookup that will find some layer root
800 struct inode
*ovl_get_trap_inode(struct super_block
*sb
, struct dentry
*dir
)
802 struct inode
*key
= d_inode(dir
);
806 return ERR_PTR(-ENOTDIR
);
808 trap
= iget5_locked(sb
, (unsigned long) key
, ovl_inode_test
,
811 return ERR_PTR(-ENOMEM
);
813 if (!(trap
->i_state
& I_NEW
)) {
814 /* Conflicting layer roots? */
816 return ERR_PTR(-ELOOP
);
819 trap
->i_mode
= S_IFDIR
;
820 trap
->i_flags
= S_DEAD
;
821 unlock_new_inode(trap
);
827 * Does overlay inode need to be hashed by lower inode?
829 static bool ovl_hash_bylower(struct super_block
*sb
, struct dentry
*upper
,
830 struct dentry
*lower
, struct dentry
*index
)
832 struct ovl_fs
*ofs
= sb
->s_fs_info
;
834 /* No, if pure upper */
838 /* Yes, if already indexed */
842 /* Yes, if won't be copied up */
846 /* No, if lower hardlink is or will be broken on copy up */
847 if ((upper
|| !ovl_indexdir(sb
)) &&
848 !d_is_dir(lower
) && d_inode(lower
)->i_nlink
> 1)
851 /* No, if non-indexed upper with NFS export */
852 if (sb
->s_export_op
&& upper
)
855 /* Otherwise, hash by lower inode for fsnotify */
859 static struct inode
*ovl_iget5(struct super_block
*sb
, struct inode
*newinode
,
862 return newinode
? inode_insert5(newinode
, (unsigned long) key
,
863 ovl_inode_test
, ovl_inode_set
, key
) :
864 iget5_locked(sb
, (unsigned long) key
,
865 ovl_inode_test
, ovl_inode_set
, key
);
868 struct inode
*ovl_get_inode(struct super_block
*sb
,
869 struct ovl_inode_params
*oip
)
871 struct dentry
*upperdentry
= oip
->upperdentry
;
872 struct ovl_path
*lowerpath
= oip
->lowerpath
;
873 struct inode
*realinode
= upperdentry
? d_inode(upperdentry
) : NULL
;
875 struct dentry
*lowerdentry
= lowerpath
? lowerpath
->dentry
: NULL
;
876 bool bylower
= ovl_hash_bylower(sb
, upperdentry
, lowerdentry
,
878 int fsid
= bylower
? oip
->lowerpath
->layer
->fsid
: 0;
879 bool is_dir
, metacopy
= false;
880 unsigned long ino
= 0;
881 int err
= oip
->newinode
? -EEXIST
: -ENOMEM
;
884 realinode
= d_inode(lowerdentry
);
887 * Copy up origin (lower) may exist for non-indexed upper, but we must
888 * not use lower as hash key if this is a broken hardlink.
890 is_dir
= S_ISDIR(realinode
->i_mode
);
891 if (upperdentry
|| bylower
) {
892 struct inode
*key
= d_inode(bylower
? lowerdentry
:
894 unsigned int nlink
= is_dir
? 1 : realinode
->i_nlink
;
896 inode
= ovl_iget5(sb
, oip
->newinode
, key
);
899 if (!(inode
->i_state
& I_NEW
)) {
901 * Verify that the underlying files stored in the inode
902 * match those in the dentry.
904 if (!ovl_verify_inode(inode
, lowerdentry
, upperdentry
,
912 kfree(oip
->redirect
);
916 /* Recalculate nlink for non-dir due to indexing */
918 nlink
= ovl_get_nlink(lowerdentry
, upperdentry
, nlink
);
919 set_nlink(inode
, nlink
);
922 /* Lower hardlink that will be broken on copy up */
923 inode
= new_inode(sb
);
929 ovl_fill_inode(inode
, realinode
->i_mode
, realinode
->i_rdev
, ino
, fsid
);
930 ovl_inode_init(inode
, upperdentry
, lowerdentry
, oip
->lowerdata
);
932 if (upperdentry
&& ovl_is_impuredir(upperdentry
))
933 ovl_set_flag(OVL_IMPURE
, inode
);
936 ovl_set_flag(OVL_INDEX
, inode
);
939 err
= ovl_check_metacopy_xattr(upperdentry
);
944 ovl_set_flag(OVL_UPPERDATA
, inode
);
947 OVL_I(inode
)->redirect
= oip
->redirect
;
950 ovl_set_flag(OVL_CONST_INO
, inode
);
952 /* Check for non-merge dir that may have whiteouts */
954 if (((upperdentry
&& lowerdentry
) || oip
->numlower
> 1) ||
955 ovl_check_origin_xattr(upperdentry
?: lowerdentry
)) {
956 ovl_set_flag(OVL_WHITEOUTS
, inode
);
960 if (inode
->i_state
& I_NEW
)
961 unlock_new_inode(inode
);
966 pr_warn_ratelimited("overlayfs: failed to get inode (%i)\n", err
);
967 inode
= ERR_PTR(err
);