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 <linux/fileattr.h>
15 #include <linux/security.h>
16 #include "overlayfs.h"
19 int ovl_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
23 bool full_copy_up = false;
24 struct dentry *upperdentry;
25 const struct cred *old_cred;
27 err = setattr_prepare(&init_user_ns, dentry, attr);
31 err = ovl_want_write(dentry);
35 if (attr->ia_valid & ATTR_SIZE) {
36 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
39 if (atomic_read(&realinode->i_writecount) < 0)
42 /* Truncate should trigger data copy up as well */
47 err = ovl_copy_up(dentry);
49 err = ovl_copy_up_with_data(dentry);
51 struct inode *winode = NULL;
53 upperdentry = ovl_dentry_upper(dentry);
55 if (attr->ia_valid & ATTR_SIZE) {
56 winode = d_inode(upperdentry);
57 err = get_write_access(winode);
62 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
63 attr->ia_valid &= ~ATTR_MODE;
66 * We might have to translate ovl file into real file object
67 * once use cases emerge. For now, simply don't let underlying
68 * filesystem rely on attr->ia_file
70 attr->ia_valid &= ~ATTR_FILE;
73 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
74 * set. Overlayfs does not pass O_TRUNC flag to underlying
75 * filesystem during open -> do not pass ATTR_OPEN. This
76 * disables optimization in fuse which assumes open(O_TRUNC)
77 * already set file size to 0. But we never passed O_TRUNC to
78 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
79 * setattr request to server.
81 attr->ia_valid &= ~ATTR_OPEN;
83 inode_lock(upperdentry->d_inode);
84 old_cred = ovl_override_creds(dentry->d_sb);
85 err = notify_change(&init_user_ns, upperdentry, attr, NULL);
86 revert_creds(old_cred);
88 ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
89 inode_unlock(upperdentry->d_inode);
92 put_write_access(winode);
95 ovl_drop_write(dentry);
100 static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
102 bool samefs = ovl_same_fs(dentry->d_sb);
103 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
104 unsigned int xinoshift = 64 - xinobits;
108 * When all layers are on the same fs, all real inode
109 * number are unique, so we use the overlay st_dev,
110 * which is friendly to du -x.
112 stat->dev = dentry->d_sb->s_dev;
114 } else if (xinobits) {
116 * All inode numbers of underlying fs should not be using the
117 * high xinobits, so we use high xinobits to partition the
118 * overlay st_ino address space. The high bits holds the fsid
119 * (upper fsid is 0). The lowest xinobit is reserved for mapping
120 * the non-persistent inode numbers range in case of overflow.
121 * This way all overlay inode numbers are unique and use the
124 if (likely(!(stat->ino >> xinoshift))) {
125 stat->ino |= ((u64)fsid) << (xinoshift + 1);
126 stat->dev = dentry->d_sb->s_dev;
128 } else if (ovl_xino_warn(dentry->d_sb)) {
129 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
130 dentry, stat->ino, xinobits);
134 /* The inode could not be mapped to a unified st_ino address space */
135 if (S_ISDIR(dentry->d_inode->i_mode)) {
137 * Always use the overlay st_dev for directories, so 'find
138 * -xdev' will scan the entire overlay mount and won't cross the
139 * overlay mount boundaries.
141 * If not all layers are on the same fs the pair {real st_ino;
142 * overlay st_dev} is not unique, so use the non persistent
143 * overlay st_ino for directories.
145 stat->dev = dentry->d_sb->s_dev;
146 stat->ino = dentry->d_inode->i_ino;
149 * For non-samefs setup, if we cannot map all layers st_ino
150 * to a unified address space, we need to make sure that st_dev
151 * is unique per underlying fs, so we use the unique anonymous
152 * bdev assigned to the underlying fs.
154 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 ovl_map_dev_ino(dentry, stat, fsid);
257 * It's probably not worth it to count subdirs to get the
258 * correct link count. nlink=1 seems to pacify 'find' and
261 if (is_dir && OVL_TYPE_MERGE(type))
265 * Return the overlay inode nlinks for indexed upper inodes.
266 * Overlay inode nlink counts the union of the upper hardlinks
267 * and non-covered lower hardlinks. It does not include the upper
270 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
271 stat->nlink = dentry->d_inode->i_nlink;
274 revert_creds(old_cred);
279 int ovl_permission(struct user_namespace *mnt_userns,
280 struct inode *inode, int mask)
282 struct inode *upperinode = ovl_inode_upper(inode);
283 struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
284 const struct cred *old_cred;
287 /* Careful in RCU walk mode */
289 WARN_ON(!(mask & MAY_NOT_BLOCK));
294 * Check overlay inode with the creds of task and underlying inode
295 * with creds of mounter
297 err = generic_permission(&init_user_ns, inode, mask);
301 old_cred = ovl_override_creds(inode->i_sb);
303 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
304 mask &= ~(MAY_WRITE | MAY_APPEND);
305 /* Make sure mounter can read file for copy up later */
308 err = inode_permission(&init_user_ns, realinode, mask);
309 revert_creds(old_cred);
314 static const char *ovl_get_link(struct dentry *dentry,
316 struct delayed_call *done)
318 const struct cred *old_cred;
322 return ERR_PTR(-ECHILD);
324 old_cred = ovl_override_creds(dentry->d_sb);
325 p = vfs_get_link(ovl_dentry_real(dentry), done);
326 revert_creds(old_cred);
330 bool ovl_is_private_xattr(struct super_block *sb, const char *name)
332 struct ovl_fs *ofs = sb->s_fs_info;
334 if (ofs->config.userxattr)
335 return strncmp(name, OVL_XATTR_USER_PREFIX,
336 sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
338 return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
339 sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
342 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
343 const void *value, size_t size, int flags)
346 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
347 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
348 const struct cred *old_cred;
350 err = ovl_want_write(dentry);
354 if (!value && !upperdentry) {
355 old_cred = ovl_override_creds(dentry->d_sb);
356 err = vfs_getxattr(&init_user_ns, realdentry, name, NULL, 0);
357 revert_creds(old_cred);
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 xattrs */
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);
502 * Work around the fact that security_file_ioctl() takes a file argument.
503 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
506 static int ovl_security_fileattr(struct dentry *dentry, struct fileattr *fa,
509 struct path realpath;
514 ovl_path_real(dentry, &realpath);
515 file = dentry_open(&realpath, O_RDONLY, current_cred());
517 return PTR_ERR(file);
520 cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
522 cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
524 err = security_file_ioctl(file, cmd, 0);
530 int ovl_fileattr_set(struct user_namespace *mnt_userns,
531 struct dentry *dentry, struct fileattr *fa)
533 struct inode *inode = d_inode(dentry);
534 struct dentry *upperdentry;
535 const struct cred *old_cred;
538 err = ovl_want_write(dentry);
542 err = ovl_copy_up(dentry);
544 upperdentry = ovl_dentry_upper(dentry);
546 old_cred = ovl_override_creds(inode->i_sb);
547 err = ovl_security_fileattr(dentry, fa, true);
549 err = vfs_fileattr_set(&init_user_ns, upperdentry, fa);
550 revert_creds(old_cred);
551 ovl_copyflags(ovl_inode_real(inode), inode);
553 ovl_drop_write(dentry);
558 int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa)
560 struct inode *inode = d_inode(dentry);
561 struct dentry *realdentry = ovl_dentry_real(dentry);
562 const struct cred *old_cred;
565 old_cred = ovl_override_creds(inode->i_sb);
566 err = ovl_security_fileattr(dentry, fa, false);
568 err = vfs_fileattr_get(realdentry, fa);
569 revert_creds(old_cred);
574 static const struct inode_operations ovl_file_inode_operations = {
575 .setattr = ovl_setattr,
576 .permission = ovl_permission,
577 .getattr = ovl_getattr,
578 .listxattr = ovl_listxattr,
579 .get_acl = ovl_get_acl,
580 .update_time = ovl_update_time,
581 .fiemap = ovl_fiemap,
582 .fileattr_get = ovl_fileattr_get,
583 .fileattr_set = ovl_fileattr_set,
586 static const struct inode_operations ovl_symlink_inode_operations = {
587 .setattr = ovl_setattr,
588 .get_link = ovl_get_link,
589 .getattr = ovl_getattr,
590 .listxattr = ovl_listxattr,
591 .update_time = ovl_update_time,
594 static const struct inode_operations ovl_special_inode_operations = {
595 .setattr = ovl_setattr,
596 .permission = ovl_permission,
597 .getattr = ovl_getattr,
598 .listxattr = ovl_listxattr,
599 .get_acl = ovl_get_acl,
600 .update_time = ovl_update_time,
603 static const struct address_space_operations ovl_aops = {
604 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
605 .direct_IO = noop_direct_IO,
609 * It is possible to stack overlayfs instance on top of another
610 * overlayfs instance as lower layer. We need to annotate the
611 * stackable i_mutex locks according to stack level of the super
612 * block instance. An overlayfs instance can never be in stack
613 * depth 0 (there is always a real fs below it). An overlayfs
614 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
616 * For example, here is a snip from /proc/lockdep_chains after
617 * dir_iterate of nested overlayfs:
619 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
620 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
621 * [...] &type->i_mutex_dir_key (stack_depth=0)
623 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
625 * This chain is valid:
626 * - inode->i_rwsem (inode_lock[2])
627 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
628 * - OVL_I(inode)->lock (ovl_inode_lock[2])
629 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
631 * And this chain is valid:
632 * - inode->i_rwsem (inode_lock[2])
633 * - OVL_I(inode)->lock (ovl_inode_lock[2])
634 * - lowerinode->i_rwsem (inode_lock[1])
635 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
637 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
638 * held, because it is in reverse order of the non-nested case using the same
640 * - inode->i_rwsem (inode_lock[1])
641 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
642 * - OVL_I(inode)->lock (ovl_inode_lock[1])
644 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
646 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
648 #ifdef CONFIG_LOCKDEP
649 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
650 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
651 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
653 int depth = inode->i_sb->s_stack_depth - 1;
655 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
658 if (S_ISDIR(inode->i_mode))
659 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
661 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
663 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
667 static void ovl_next_ino(struct inode *inode)
669 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
671 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
672 if (unlikely(!inode->i_ino))
673 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
676 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
678 int xinobits = ovl_xino_bits(inode->i_sb);
679 unsigned int xinoshift = 64 - xinobits;
682 * When d_ino is consistent with st_ino (samefs or i_ino has enough
683 * bits to encode layer), set the same value used for st_ino to i_ino,
684 * so inode number exposed via /proc/locks and a like will be
685 * consistent with d_ino and st_ino values. An i_ino value inconsistent
686 * with d_ino also causes nfsd readdirplus to fail.
689 if (ovl_same_fs(inode->i_sb)) {
691 } else if (xinobits && likely(!(ino >> xinoshift))) {
692 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
697 * For directory inodes on non-samefs with xino disabled or xino
698 * overflow, we allocate a non-persistent inode number, to be used for
699 * resolving st_ino collisions in ovl_map_dev_ino().
701 * To avoid ino collision with legitimate xino values from upper
702 * layer (fsid 0), use the lowest xinobit to map the non
703 * persistent inode numbers to the unified st_ino address space.
705 if (S_ISDIR(inode->i_mode)) {
708 inode->i_ino &= ~0UL >> xinobits;
709 inode->i_ino |= 1UL << xinoshift;
714 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
715 unsigned long ino, int fsid)
717 struct inode *realinode;
719 if (oip->upperdentry)
720 OVL_I(inode)->__upperdentry = oip->upperdentry;
721 if (oip->lowerpath && oip->lowerpath->dentry)
722 OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
724 OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
726 realinode = ovl_inode_real(inode);
727 ovl_copyattr(realinode, inode);
728 ovl_copyflags(realinode, inode);
729 ovl_map_ino(inode, ino, fsid);
732 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
734 inode->i_mode = mode;
735 inode->i_flags |= S_NOCMTIME;
736 #ifdef CONFIG_FS_POSIX_ACL
737 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
740 ovl_lockdep_annotate_inode_mutex_key(inode);
742 switch (mode & S_IFMT) {
744 inode->i_op = &ovl_file_inode_operations;
745 inode->i_fop = &ovl_file_operations;
746 inode->i_mapping->a_ops = &ovl_aops;
750 inode->i_op = &ovl_dir_inode_operations;
751 inode->i_fop = &ovl_dir_operations;
755 inode->i_op = &ovl_symlink_inode_operations;
759 inode->i_op = &ovl_special_inode_operations;
760 init_special_inode(inode, mode, rdev);
766 * With inodes index enabled, an overlay inode nlink counts the union of upper
767 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
768 * upper inode, the following nlink modifying operations can happen:
770 * 1. Lower hardlink copy up
771 * 2. Upper hardlink created, unlinked or renamed over
772 * 3. Lower hardlink whiteout or renamed over
774 * For the first, copy up case, the union nlink does not change, whether the
775 * operation succeeds or fails, but the upper inode nlink may change.
776 * Therefore, before copy up, we store the union nlink value relative to the
777 * lower inode nlink in the index inode xattr .overlay.nlink.
779 * For the second, upper hardlink case, the union nlink should be incremented
780 * or decremented IFF the operation succeeds, aligned with nlink change of the
781 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
782 * value relative to the upper inode nlink in the index inode.
784 * For the last, lower cover up case, we simplify things by preceding the
785 * whiteout or cover up with copy up. This makes sure that there is an index
786 * upper inode where the nlink xattr can be stored before the copied up upper
789 #define OVL_NLINK_ADD_UPPER (1 << 0)
792 * On-disk format for indexed nlink:
794 * nlink relative to the upper inode - "U[+-]NUM"
795 * nlink relative to the lower inode - "L[+-]NUM"
798 static int ovl_set_nlink_common(struct dentry *dentry,
799 struct dentry *realdentry, const char *format)
801 struct inode *inode = d_inode(dentry);
802 struct inode *realinode = d_inode(realdentry);
806 len = snprintf(buf, sizeof(buf), format,
807 (int) (inode->i_nlink - realinode->i_nlink));
809 if (WARN_ON(len >= sizeof(buf)))
812 return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
813 OVL_XATTR_NLINK, buf, len);
816 int ovl_set_nlink_upper(struct dentry *dentry)
818 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
821 int ovl_set_nlink_lower(struct dentry *dentry)
823 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
826 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
827 struct dentry *upperdentry,
828 unsigned int fallback)
835 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
838 err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
839 &buf, sizeof(buf) - 1);
844 if ((buf[0] != 'L' && buf[0] != 'U') ||
845 (buf[1] != '+' && buf[1] != '-'))
848 err = kstrtoint(buf + 1, 10, &nlink_diff);
852 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
861 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
866 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
870 inode = new_inode(sb);
872 ovl_fill_inode(inode, mode, rdev);
877 static int ovl_inode_test(struct inode *inode, void *data)
879 return inode->i_private == data;
882 static int ovl_inode_set(struct inode *inode, void *data)
884 inode->i_private = data;
888 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
889 struct dentry *upperdentry, bool strict)
892 * For directories, @strict verify from lookup path performs consistency
893 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
894 * inode. Non @strict verify from NFS handle decode path passes NULL for
895 * 'unknown' lower/upper.
897 if (S_ISDIR(inode->i_mode) && strict) {
898 /* Real lower dir moved to upper layer under us? */
899 if (!lowerdentry && ovl_inode_lower(inode))
902 /* Lookup of an uncovered redirect origin? */
903 if (!upperdentry && ovl_inode_upper(inode))
908 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
909 * This happens when finding a copied up overlay inode for a renamed
910 * or hardlinked overlay dentry and lower dentry cannot be followed
911 * by origin because lower fs does not support file handles.
913 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
917 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
918 * This happens when finding a lower alias for a copied up hard link.
920 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
926 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
929 struct inode *inode, *key = d_inode(real);
931 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
935 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
936 is_upper ? real : NULL, false)) {
938 return ERR_PTR(-ESTALE);
944 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
946 struct inode *key = d_inode(dir);
950 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
954 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
955 !ovl_inode_lower(trap);
962 * Create an inode cache entry for layer root dir, that will intentionally
963 * fail ovl_verify_inode(), so any lookup that will find some layer root
966 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
968 struct inode *key = d_inode(dir);
972 return ERR_PTR(-ENOTDIR);
974 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
977 return ERR_PTR(-ENOMEM);
979 if (!(trap->i_state & I_NEW)) {
980 /* Conflicting layer roots? */
982 return ERR_PTR(-ELOOP);
985 trap->i_mode = S_IFDIR;
986 trap->i_flags = S_DEAD;
987 unlock_new_inode(trap);
993 * Does overlay inode need to be hashed by lower inode?
995 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
996 struct dentry *lower, bool index)
998 struct ovl_fs *ofs = sb->s_fs_info;
1000 /* No, if pure upper */
1004 /* Yes, if already indexed */
1008 /* Yes, if won't be copied up */
1009 if (!ovl_upper_mnt(ofs))
1012 /* No, if lower hardlink is or will be broken on copy up */
1013 if ((upper || !ovl_indexdir(sb)) &&
1014 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1017 /* No, if non-indexed upper with NFS export */
1018 if (sb->s_export_op && upper)
1021 /* Otherwise, hash by lower inode for fsnotify */
1025 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1028 return newinode ? inode_insert5(newinode, (unsigned long) key,
1029 ovl_inode_test, ovl_inode_set, key) :
1030 iget5_locked(sb, (unsigned long) key,
1031 ovl_inode_test, ovl_inode_set, key);
1034 struct inode *ovl_get_inode(struct super_block *sb,
1035 struct ovl_inode_params *oip)
1037 struct ovl_fs *ofs = OVL_FS(sb);
1038 struct dentry *upperdentry = oip->upperdentry;
1039 struct ovl_path *lowerpath = oip->lowerpath;
1040 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1041 struct inode *inode;
1042 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1043 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1045 int fsid = bylower ? lowerpath->layer->fsid : 0;
1047 unsigned long ino = 0;
1048 int err = oip->newinode ? -EEXIST : -ENOMEM;
1051 realinode = d_inode(lowerdentry);
1054 * Copy up origin (lower) may exist for non-indexed upper, but we must
1055 * not use lower as hash key if this is a broken hardlink.
1057 is_dir = S_ISDIR(realinode->i_mode);
1058 if (upperdentry || bylower) {
1059 struct inode *key = d_inode(bylower ? lowerdentry :
1061 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1063 inode = ovl_iget5(sb, oip->newinode, key);
1066 if (!(inode->i_state & I_NEW)) {
1068 * Verify that the underlying files stored in the inode
1069 * match those in the dentry.
1071 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1079 kfree(oip->redirect);
1083 /* Recalculate nlink for non-dir due to indexing */
1085 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1087 set_nlink(inode, nlink);
1090 /* Lower hardlink that will be broken on copy up */
1091 inode = new_inode(sb);
1096 ino = realinode->i_ino;
1097 fsid = lowerpath->layer->fsid;
1099 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1100 ovl_inode_init(inode, oip, ino, fsid);
1102 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1103 ovl_set_flag(OVL_IMPURE, inode);
1106 ovl_set_flag(OVL_INDEX, inode);
1108 OVL_I(inode)->redirect = oip->redirect;
1111 ovl_set_flag(OVL_CONST_INO, inode);
1113 /* Check for non-merge dir that may have whiteouts */
1115 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1116 ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
1117 ovl_set_flag(OVL_WHITEOUTS, inode);
1121 if (inode->i_state & I_NEW)
1122 unlock_new_inode(inode);
1127 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1128 inode = ERR_PTR(err);