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 <linux/fiemap.h>
14 #include "overlayfs.h"
17 int ovl_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
21 bool full_copy_up = false;
22 struct dentry *upperdentry;
23 const struct cred *old_cred;
25 err = setattr_prepare(&init_user_ns, dentry, attr);
29 err = ovl_want_write(dentry);
33 if (attr->ia_valid & ATTR_SIZE) {
34 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
37 if (atomic_read(&realinode->i_writecount) < 0)
40 /* Truncate should trigger data copy up as well */
45 err = ovl_copy_up(dentry);
47 err = ovl_copy_up_with_data(dentry);
49 struct inode *winode = NULL;
51 upperdentry = ovl_dentry_upper(dentry);
53 if (attr->ia_valid & ATTR_SIZE) {
54 winode = d_inode(upperdentry);
55 err = get_write_access(winode);
60 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
61 attr->ia_valid &= ~ATTR_MODE;
64 * We might have to translate ovl file into real file object
65 * once use cases emerge. For now, simply don't let underlying
66 * filesystem rely on attr->ia_file
68 attr->ia_valid &= ~ATTR_FILE;
71 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
72 * set. Overlayfs does not pass O_TRUNC flag to underlying
73 * filesystem during open -> do not pass ATTR_OPEN. This
74 * disables optimization in fuse which assumes open(O_TRUNC)
75 * already set file size to 0. But we never passed O_TRUNC to
76 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
77 * setattr request to server.
79 attr->ia_valid &= ~ATTR_OPEN;
81 inode_lock(upperdentry->d_inode);
82 old_cred = ovl_override_creds(dentry->d_sb);
83 err = notify_change(&init_user_ns, upperdentry, attr, NULL);
84 revert_creds(old_cred);
86 ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
87 inode_unlock(upperdentry->d_inode);
90 put_write_access(winode);
93 ovl_drop_write(dentry);
98 static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
100 bool samefs = ovl_same_fs(dentry->d_sb);
101 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
102 unsigned int xinoshift = 64 - xinobits;
106 * When all layers are on the same fs, all real inode
107 * number are unique, so we use the overlay st_dev,
108 * which is friendly to du -x.
110 stat->dev = dentry->d_sb->s_dev;
112 } else if (xinobits) {
114 * All inode numbers of underlying fs should not be using the
115 * high xinobits, so we use high xinobits to partition the
116 * overlay st_ino address space. The high bits holds the fsid
117 * (upper fsid is 0). The lowest xinobit is reserved for mapping
118 * the non-peresistent inode numbers range in case of overflow.
119 * This way all overlay inode numbers are unique and use the
122 if (likely(!(stat->ino >> xinoshift))) {
123 stat->ino |= ((u64)fsid) << (xinoshift + 1);
124 stat->dev = dentry->d_sb->s_dev;
126 } else if (ovl_xino_warn(dentry->d_sb)) {
127 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
128 dentry, stat->ino, xinobits);
132 /* The inode could not be mapped to a unified st_ino address space */
133 if (S_ISDIR(dentry->d_inode->i_mode)) {
135 * Always use the overlay st_dev for directories, so 'find
136 * -xdev' will scan the entire overlay mount and won't cross the
137 * overlay mount boundaries.
139 * If not all layers are on the same fs the pair {real st_ino;
140 * overlay st_dev} is not unique, so use the non persistent
141 * overlay st_ino for directories.
143 stat->dev = dentry->d_sb->s_dev;
144 stat->ino = dentry->d_inode->i_ino;
147 * For non-samefs setup, if we cannot map all layers st_ino
148 * to a unified address space, we need to make sure that st_dev
149 * is unique per underlying fs, so we use the unique anonymous
150 * bdev assigned to the underlying fs.
152 stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
158 int ovl_getattr(struct user_namespace *mnt_userns, const struct path *path,
159 struct kstat *stat, u32 request_mask, unsigned int flags)
161 struct dentry *dentry = path->dentry;
162 enum ovl_path_type type;
163 struct path realpath;
164 const struct cred *old_cred;
165 bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
168 bool metacopy_blocks = false;
170 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
172 type = ovl_path_real(dentry, &realpath);
173 old_cred = ovl_override_creds(dentry->d_sb);
174 err = vfs_getattr(&realpath, stat, request_mask, flags);
179 * For non-dir or same fs, we use st_ino of the copy up origin.
180 * This guaranties constant st_dev/st_ino across copy up.
181 * With xino feature and non-samefs, we use st_ino of the copy up
182 * origin masked with high bits that represent the layer id.
184 * If lower filesystem supports NFS file handles, this also guaranties
185 * persistent st_ino across mount cycle.
187 if (!is_dir || ovl_same_dev(dentry->d_sb)) {
188 if (!OVL_TYPE_UPPER(type)) {
189 fsid = ovl_layer_lower(dentry)->fsid;
190 } else if (OVL_TYPE_ORIGIN(type)) {
191 struct kstat lowerstat;
192 u32 lowermask = STATX_INO | STATX_BLOCKS |
193 (!is_dir ? STATX_NLINK : 0);
195 ovl_path_lower(dentry, &realpath);
196 err = vfs_getattr(&realpath, &lowerstat,
202 * Lower hardlinks may be broken on copy up to different
203 * upper files, so we cannot use the lower origin st_ino
204 * for those different files, even for the same fs case.
206 * Similarly, several redirected dirs can point to the
207 * same dir on a lower layer. With the "verify_lower"
208 * feature, we do not use the lower origin st_ino, if
209 * we haven't verified that this redirect is unique.
211 * With inodes index enabled, it is safe to use st_ino
212 * of an indexed origin. The index validates that the
213 * upper hardlink is not broken and that a redirected
214 * dir is the only redirect to that origin.
216 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
217 (!ovl_verify_lower(dentry->d_sb) &&
218 (is_dir || lowerstat.nlink == 1))) {
219 fsid = ovl_layer_lower(dentry)->fsid;
220 stat->ino = lowerstat.ino;
224 * If we are querying a metacopy dentry and lower
225 * dentry is data dentry, then use the blocks we
226 * queried just now. We don't have to do additional
227 * vfs_getattr(). If lower itself is metacopy, then
228 * additional vfs_getattr() is unavoidable.
230 if (metacopy_blocks &&
231 realpath.dentry == ovl_dentry_lowerdata(dentry)) {
232 stat->blocks = lowerstat.blocks;
233 metacopy_blocks = false;
237 if (metacopy_blocks) {
239 * If lower is not same as lowerdata or if there was
240 * no origin on upper, we can end up here.
242 struct kstat lowerdatastat;
243 u32 lowermask = STATX_BLOCKS;
245 ovl_path_lowerdata(dentry, &realpath);
246 err = vfs_getattr(&realpath, &lowerdatastat,
250 stat->blocks = lowerdatastat.blocks;
254 err = ovl_map_dev_ino(dentry, stat, fsid);
259 * It's probably not worth it to count subdirs to get the
260 * correct link count. nlink=1 seems to pacify 'find' and
263 if (is_dir && OVL_TYPE_MERGE(type))
267 * Return the overlay inode nlinks for indexed upper inodes.
268 * Overlay inode nlink counts the union of the upper hardlinks
269 * and non-covered lower hardlinks. It does not include the upper
272 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
273 stat->nlink = dentry->d_inode->i_nlink;
276 revert_creds(old_cred);
281 int ovl_permission(struct user_namespace *mnt_userns,
282 struct inode *inode, int mask)
284 struct inode *upperinode = ovl_inode_upper(inode);
285 struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
286 const struct cred *old_cred;
289 /* Careful in RCU walk mode */
291 WARN_ON(!(mask & MAY_NOT_BLOCK));
296 * Check overlay inode with the creds of task and underlying inode
297 * with creds of mounter
299 err = generic_permission(&init_user_ns, inode, mask);
303 old_cred = ovl_override_creds(inode->i_sb);
305 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
306 mask &= ~(MAY_WRITE | MAY_APPEND);
307 /* Make sure mounter can read file for copy up later */
310 err = inode_permission(&init_user_ns, realinode, mask);
311 revert_creds(old_cred);
316 static const char *ovl_get_link(struct dentry *dentry,
318 struct delayed_call *done)
320 const struct cred *old_cred;
324 return ERR_PTR(-ECHILD);
326 old_cred = ovl_override_creds(dentry->d_sb);
327 p = vfs_get_link(ovl_dentry_real(dentry), done);
328 revert_creds(old_cred);
332 bool ovl_is_private_xattr(struct super_block *sb, const char *name)
334 struct ovl_fs *ofs = sb->s_fs_info;
336 if (ofs->config.userxattr)
337 return strncmp(name, OVL_XATTR_USER_PREFIX,
338 sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
340 return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
341 sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
344 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
345 const void *value, size_t size, int flags)
348 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
349 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
350 const struct cred *old_cred;
352 err = ovl_want_write(dentry);
356 if (!value && !upperdentry) {
357 err = vfs_getxattr(&init_user_ns, realdentry, name, NULL, 0);
363 err = ovl_copy_up(dentry);
367 realdentry = ovl_dentry_upper(dentry);
370 old_cred = ovl_override_creds(dentry->d_sb);
372 err = vfs_setxattr(&init_user_ns, realdentry, name, value, size,
375 WARN_ON(flags != XATTR_REPLACE);
376 err = vfs_removexattr(&init_user_ns, realdentry, name);
378 revert_creds(old_cred);
381 ovl_copyattr(d_inode(realdentry), inode);
384 ovl_drop_write(dentry);
389 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
390 void *value, size_t size)
393 const struct cred *old_cred;
394 struct dentry *realdentry =
395 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
397 old_cred = ovl_override_creds(dentry->d_sb);
398 res = vfs_getxattr(&init_user_ns, realdentry, name, value, size);
399 revert_creds(old_cred);
403 static bool ovl_can_list(struct super_block *sb, const char *s)
405 /* Never list private (.overlay) */
406 if (ovl_is_private_xattr(sb, s))
409 /* List all non-trusted xatts */
410 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
413 /* list other trusted for superuser only */
414 return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
417 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
419 struct dentry *realdentry = ovl_dentry_real(dentry);
423 const struct cred *old_cred;
425 old_cred = ovl_override_creds(dentry->d_sb);
426 res = vfs_listxattr(realdentry, list, size);
427 revert_creds(old_cred);
428 if (res <= 0 || size == 0)
431 /* filter out private xattrs */
432 for (s = list, len = res; len;) {
433 size_t slen = strnlen(s, len) + 1;
435 /* underlying fs providing us with an broken xattr list? */
436 if (WARN_ON(slen > len))
440 if (!ovl_can_list(dentry->d_sb, s)) {
442 memmove(s, s + slen, len);
451 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
453 struct inode *realinode = ovl_inode_real(inode);
454 const struct cred *old_cred;
455 struct posix_acl *acl;
457 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
460 old_cred = ovl_override_creds(inode->i_sb);
461 acl = get_acl(realinode, type);
462 revert_creds(old_cred);
467 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
469 if (flags & S_ATIME) {
470 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
471 struct path upperpath = {
472 .mnt = ovl_upper_mnt(ofs),
473 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
476 if (upperpath.dentry) {
477 touch_atime(&upperpath);
478 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
484 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
488 struct inode *realinode = ovl_inode_realdata(inode);
489 const struct cred *old_cred;
491 if (!realinode->i_op->fiemap)
494 old_cred = ovl_override_creds(inode->i_sb);
495 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
496 revert_creds(old_cred);
501 static const struct inode_operations ovl_file_inode_operations = {
502 .setattr = ovl_setattr,
503 .permission = ovl_permission,
504 .getattr = ovl_getattr,
505 .listxattr = ovl_listxattr,
506 .get_acl = ovl_get_acl,
507 .update_time = ovl_update_time,
508 .fiemap = ovl_fiemap,
511 static const struct inode_operations ovl_symlink_inode_operations = {
512 .setattr = ovl_setattr,
513 .get_link = ovl_get_link,
514 .getattr = ovl_getattr,
515 .listxattr = ovl_listxattr,
516 .update_time = ovl_update_time,
519 static const struct inode_operations ovl_special_inode_operations = {
520 .setattr = ovl_setattr,
521 .permission = ovl_permission,
522 .getattr = ovl_getattr,
523 .listxattr = ovl_listxattr,
524 .get_acl = ovl_get_acl,
525 .update_time = ovl_update_time,
528 static const struct address_space_operations ovl_aops = {
529 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
530 .direct_IO = noop_direct_IO,
534 * It is possible to stack overlayfs instance on top of another
535 * overlayfs instance as lower layer. We need to annotate the
536 * stackable i_mutex locks according to stack level of the super
537 * block instance. An overlayfs instance can never be in stack
538 * depth 0 (there is always a real fs below it). An overlayfs
539 * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
541 * For example, here is a snip from /proc/lockdep_chains after
542 * dir_iterate of nested overlayfs:
544 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
545 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
546 * [...] &type->i_mutex_dir_key (stack_depth=0)
548 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
550 * This chain is valid:
551 * - inode->i_rwsem (inode_lock[2])
552 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
553 * - OVL_I(inode)->lock (ovl_inode_lock[2])
554 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
556 * And this chain is valid:
557 * - inode->i_rwsem (inode_lock[2])
558 * - OVL_I(inode)->lock (ovl_inode_lock[2])
559 * - lowerinode->i_rwsem (inode_lock[1])
560 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
562 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
563 * held, because it is in reverse order of the non-nested case using the same
565 * - inode->i_rwsem (inode_lock[1])
566 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
567 * - OVL_I(inode)->lock (ovl_inode_lock[1])
569 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
571 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
573 #ifdef CONFIG_LOCKDEP
574 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
575 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
576 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
578 int depth = inode->i_sb->s_stack_depth - 1;
580 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
583 if (S_ISDIR(inode->i_mode))
584 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
586 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
588 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
592 static void ovl_next_ino(struct inode *inode)
594 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
596 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
597 if (unlikely(!inode->i_ino))
598 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
601 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
603 int xinobits = ovl_xino_bits(inode->i_sb);
604 unsigned int xinoshift = 64 - xinobits;
607 * When d_ino is consistent with st_ino (samefs or i_ino has enough
608 * bits to encode layer), set the same value used for st_ino to i_ino,
609 * so inode number exposed via /proc/locks and a like will be
610 * consistent with d_ino and st_ino values. An i_ino value inconsistent
611 * with d_ino also causes nfsd readdirplus to fail.
614 if (ovl_same_fs(inode->i_sb)) {
616 } else if (xinobits && likely(!(ino >> xinoshift))) {
617 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
622 * For directory inodes on non-samefs with xino disabled or xino
623 * overflow, we allocate a non-persistent inode number, to be used for
624 * resolving st_ino collisions in ovl_map_dev_ino().
626 * To avoid ino collision with legitimate xino values from upper
627 * layer (fsid 0), use the lowest xinobit to map the non
628 * persistent inode numbers to the unified st_ino address space.
630 if (S_ISDIR(inode->i_mode)) {
633 inode->i_ino &= ~0UL >> xinobits;
634 inode->i_ino |= 1UL << xinoshift;
639 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
640 unsigned long ino, int fsid)
642 struct inode *realinode;
644 if (oip->upperdentry)
645 OVL_I(inode)->__upperdentry = oip->upperdentry;
646 if (oip->lowerpath && oip->lowerpath->dentry)
647 OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
649 OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
651 realinode = ovl_inode_real(inode);
652 ovl_copyattr(realinode, inode);
653 ovl_copyflags(realinode, inode);
654 ovl_map_ino(inode, ino, fsid);
657 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
659 inode->i_mode = mode;
660 inode->i_flags |= S_NOCMTIME;
661 #ifdef CONFIG_FS_POSIX_ACL
662 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
665 ovl_lockdep_annotate_inode_mutex_key(inode);
667 switch (mode & S_IFMT) {
669 inode->i_op = &ovl_file_inode_operations;
670 inode->i_fop = &ovl_file_operations;
671 inode->i_mapping->a_ops = &ovl_aops;
675 inode->i_op = &ovl_dir_inode_operations;
676 inode->i_fop = &ovl_dir_operations;
680 inode->i_op = &ovl_symlink_inode_operations;
684 inode->i_op = &ovl_special_inode_operations;
685 init_special_inode(inode, mode, rdev);
691 * With inodes index enabled, an overlay inode nlink counts the union of upper
692 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
693 * upper inode, the following nlink modifying operations can happen:
695 * 1. Lower hardlink copy up
696 * 2. Upper hardlink created, unlinked or renamed over
697 * 3. Lower hardlink whiteout or renamed over
699 * For the first, copy up case, the union nlink does not change, whether the
700 * operation succeeds or fails, but the upper inode nlink may change.
701 * Therefore, before copy up, we store the union nlink value relative to the
702 * lower inode nlink in the index inode xattr .overlay.nlink.
704 * For the second, upper hardlink case, the union nlink should be incremented
705 * or decremented IFF the operation succeeds, aligned with nlink change of the
706 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
707 * value relative to the upper inode nlink in the index inode.
709 * For the last, lower cover up case, we simplify things by preceding the
710 * whiteout or cover up with copy up. This makes sure that there is an index
711 * upper inode where the nlink xattr can be stored before the copied up upper
714 #define OVL_NLINK_ADD_UPPER (1 << 0)
717 * On-disk format for indexed nlink:
719 * nlink relative to the upper inode - "U[+-]NUM"
720 * nlink relative to the lower inode - "L[+-]NUM"
723 static int ovl_set_nlink_common(struct dentry *dentry,
724 struct dentry *realdentry, const char *format)
726 struct inode *inode = d_inode(dentry);
727 struct inode *realinode = d_inode(realdentry);
731 len = snprintf(buf, sizeof(buf), format,
732 (int) (inode->i_nlink - realinode->i_nlink));
734 if (WARN_ON(len >= sizeof(buf)))
737 return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
738 OVL_XATTR_NLINK, buf, len);
741 int ovl_set_nlink_upper(struct dentry *dentry)
743 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
746 int ovl_set_nlink_lower(struct dentry *dentry)
748 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
751 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
752 struct dentry *upperdentry,
753 unsigned int fallback)
760 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
763 err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
764 &buf, sizeof(buf) - 1);
769 if ((buf[0] != 'L' && buf[0] != 'U') ||
770 (buf[1] != '+' && buf[1] != '-'))
773 err = kstrtoint(buf + 1, 10, &nlink_diff);
777 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
786 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
791 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
795 inode = new_inode(sb);
797 ovl_fill_inode(inode, mode, rdev);
802 static int ovl_inode_test(struct inode *inode, void *data)
804 return inode->i_private == data;
807 static int ovl_inode_set(struct inode *inode, void *data)
809 inode->i_private = data;
813 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
814 struct dentry *upperdentry, bool strict)
817 * For directories, @strict verify from lookup path performs consistency
818 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
819 * inode. Non @strict verify from NFS handle decode path passes NULL for
820 * 'unknown' lower/upper.
822 if (S_ISDIR(inode->i_mode) && strict) {
823 /* Real lower dir moved to upper layer under us? */
824 if (!lowerdentry && ovl_inode_lower(inode))
827 /* Lookup of an uncovered redirect origin? */
828 if (!upperdentry && ovl_inode_upper(inode))
833 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
834 * This happens when finding a copied up overlay inode for a renamed
835 * or hardlinked overlay dentry and lower dentry cannot be followed
836 * by origin because lower fs does not support file handles.
838 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
842 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
843 * This happens when finding a lower alias for a copied up hard link.
845 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
851 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
854 struct inode *inode, *key = d_inode(real);
856 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
860 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
861 is_upper ? real : NULL, false)) {
863 return ERR_PTR(-ESTALE);
869 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
871 struct inode *key = d_inode(dir);
875 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
879 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
880 !ovl_inode_lower(trap);
887 * Create an inode cache entry for layer root dir, that will intentionally
888 * fail ovl_verify_inode(), so any lookup that will find some layer root
891 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
893 struct inode *key = d_inode(dir);
897 return ERR_PTR(-ENOTDIR);
899 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
902 return ERR_PTR(-ENOMEM);
904 if (!(trap->i_state & I_NEW)) {
905 /* Conflicting layer roots? */
907 return ERR_PTR(-ELOOP);
910 trap->i_mode = S_IFDIR;
911 trap->i_flags = S_DEAD;
912 unlock_new_inode(trap);
918 * Does overlay inode need to be hashed by lower inode?
920 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
921 struct dentry *lower, bool index)
923 struct ovl_fs *ofs = sb->s_fs_info;
925 /* No, if pure upper */
929 /* Yes, if already indexed */
933 /* Yes, if won't be copied up */
934 if (!ovl_upper_mnt(ofs))
937 /* No, if lower hardlink is or will be broken on copy up */
938 if ((upper || !ovl_indexdir(sb)) &&
939 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
942 /* No, if non-indexed upper with NFS export */
943 if (sb->s_export_op && upper)
946 /* Otherwise, hash by lower inode for fsnotify */
950 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
953 return newinode ? inode_insert5(newinode, (unsigned long) key,
954 ovl_inode_test, ovl_inode_set, key) :
955 iget5_locked(sb, (unsigned long) key,
956 ovl_inode_test, ovl_inode_set, key);
959 struct inode *ovl_get_inode(struct super_block *sb,
960 struct ovl_inode_params *oip)
962 struct ovl_fs *ofs = OVL_FS(sb);
963 struct dentry *upperdentry = oip->upperdentry;
964 struct ovl_path *lowerpath = oip->lowerpath;
965 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
967 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
968 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
970 int fsid = bylower ? lowerpath->layer->fsid : 0;
972 unsigned long ino = 0;
973 int err = oip->newinode ? -EEXIST : -ENOMEM;
976 realinode = d_inode(lowerdentry);
979 * Copy up origin (lower) may exist for non-indexed upper, but we must
980 * not use lower as hash key if this is a broken hardlink.
982 is_dir = S_ISDIR(realinode->i_mode);
983 if (upperdentry || bylower) {
984 struct inode *key = d_inode(bylower ? lowerdentry :
986 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
988 inode = ovl_iget5(sb, oip->newinode, key);
991 if (!(inode->i_state & I_NEW)) {
993 * Verify that the underlying files stored in the inode
994 * match those in the dentry.
996 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1004 kfree(oip->redirect);
1008 /* Recalculate nlink for non-dir due to indexing */
1010 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1012 set_nlink(inode, nlink);
1015 /* Lower hardlink that will be broken on copy up */
1016 inode = new_inode(sb);
1021 ino = realinode->i_ino;
1022 fsid = lowerpath->layer->fsid;
1024 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1025 ovl_inode_init(inode, oip, ino, fsid);
1027 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1028 ovl_set_flag(OVL_IMPURE, inode);
1031 ovl_set_flag(OVL_INDEX, inode);
1033 OVL_I(inode)->redirect = oip->redirect;
1036 ovl_set_flag(OVL_CONST_INO, inode);
1038 /* Check for non-merge dir that may have whiteouts */
1040 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1041 ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
1042 ovl_set_flag(OVL_WHITEOUTS, inode);
1046 if (inode->i_state & I_NEW)
1047 unlock_new_inode(inode);
1052 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1053 inode = ERR_PTR(err);