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 dentry *dentry, struct iattr *attr)
20 bool full_copy_up = false;
21 struct dentry *upperdentry;
22 const struct cred *old_cred;
24 err = setattr_prepare(dentry, attr);
28 err = ovl_want_write(dentry);
32 if (attr->ia_valid & ATTR_SIZE) {
33 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
36 if (atomic_read(&realinode->i_writecount) < 0)
39 /* Truncate should trigger data copy up as well */
44 err = ovl_copy_up(dentry);
46 err = ovl_copy_up_with_data(dentry);
48 struct inode *winode = NULL;
50 upperdentry = ovl_dentry_upper(dentry);
52 if (attr->ia_valid & ATTR_SIZE) {
53 winode = d_inode(upperdentry);
54 err = get_write_access(winode);
59 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
60 attr->ia_valid &= ~ATTR_MODE;
63 * We might have to translate ovl file into real file object
64 * once use cases emerge. For now, simply don't let underlying
65 * filesystem rely on attr->ia_file
67 attr->ia_valid &= ~ATTR_FILE;
70 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
71 * set. Overlayfs does not pass O_TRUNC flag to underlying
72 * filesystem during open -> do not pass ATTR_OPEN. This
73 * disables optimization in fuse which assumes open(O_TRUNC)
74 * already set file size to 0. But we never passed O_TRUNC to
75 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
76 * setattr request to server.
78 attr->ia_valid &= ~ATTR_OPEN;
80 inode_lock(upperdentry->d_inode);
81 old_cred = ovl_override_creds(dentry->d_sb);
82 err = notify_change(upperdentry, attr, NULL);
83 revert_creds(old_cred);
85 ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
86 inode_unlock(upperdentry->d_inode);
89 put_write_access(winode);
92 ovl_drop_write(dentry);
97 static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
99 bool samefs = ovl_same_fs(dentry->d_sb);
100 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
101 unsigned int xinoshift = 64 - xinobits;
105 * When all layers are on the same fs, all real inode
106 * number are unique, so we use the overlay st_dev,
107 * which is friendly to du -x.
109 stat->dev = dentry->d_sb->s_dev;
111 } else if (xinobits) {
113 * All inode numbers of underlying fs should not be using the
114 * high xinobits, so we use high xinobits to partition the
115 * overlay st_ino address space. The high bits holds the fsid
116 * (upper fsid is 0). The lowest xinobit is reserved for mapping
117 * the non-peresistent inode numbers range in case of overflow.
118 * This way all overlay inode numbers are unique and use the
121 if (likely(!(stat->ino >> xinoshift))) {
122 stat->ino |= ((u64)fsid) << (xinoshift + 1);
123 stat->dev = dentry->d_sb->s_dev;
125 } else if (ovl_xino_warn(dentry->d_sb)) {
126 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
127 dentry, stat->ino, xinobits);
131 /* The inode could not be mapped to a unified st_ino address space */
132 if (S_ISDIR(dentry->d_inode->i_mode)) {
134 * Always use the overlay st_dev for directories, so 'find
135 * -xdev' will scan the entire overlay mount and won't cross the
136 * overlay mount boundaries.
138 * If not all layers are on the same fs the pair {real st_ino;
139 * overlay st_dev} is not unique, so use the non persistent
140 * overlay st_ino for directories.
142 stat->dev = dentry->d_sb->s_dev;
143 stat->ino = dentry->d_inode->i_ino;
146 * For non-samefs setup, if we cannot map all layers st_ino
147 * to a unified address space, we need to make sure that st_dev
148 * is unique per underlying fs, so we use the unique anonymous
149 * bdev assigned to the underlying fs.
151 stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
157 int ovl_getattr(const struct path *path, struct kstat *stat,
158 u32 request_mask, unsigned int flags)
160 struct dentry *dentry = path->dentry;
161 enum ovl_path_type type;
162 struct path realpath;
163 const struct cred *old_cred;
164 bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
167 bool metacopy_blocks = false;
169 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
171 type = ovl_path_real(dentry, &realpath);
172 old_cred = ovl_override_creds(dentry->d_sb);
173 err = vfs_getattr(&realpath, stat, request_mask, flags);
178 * For non-dir or same fs, we use st_ino of the copy up origin.
179 * This guaranties constant st_dev/st_ino across copy up.
180 * With xino feature and non-samefs, we use st_ino of the copy up
181 * origin masked with high bits that represent the layer id.
183 * If lower filesystem supports NFS file handles, this also guaranties
184 * persistent st_ino across mount cycle.
186 if (!is_dir || ovl_same_dev(dentry->d_sb)) {
187 if (!OVL_TYPE_UPPER(type)) {
188 fsid = ovl_layer_lower(dentry)->fsid;
189 } else if (OVL_TYPE_ORIGIN(type)) {
190 struct kstat lowerstat;
191 u32 lowermask = STATX_INO | STATX_BLOCKS |
192 (!is_dir ? STATX_NLINK : 0);
194 ovl_path_lower(dentry, &realpath);
195 err = vfs_getattr(&realpath, &lowerstat,
201 * Lower hardlinks may be broken on copy up to different
202 * upper files, so we cannot use the lower origin st_ino
203 * for those different files, even for the same fs case.
205 * Similarly, several redirected dirs can point to the
206 * same dir on a lower layer. With the "verify_lower"
207 * feature, we do not use the lower origin st_ino, if
208 * we haven't verified that this redirect is unique.
210 * With inodes index enabled, it is safe to use st_ino
211 * of an indexed origin. The index validates that the
212 * upper hardlink is not broken and that a redirected
213 * dir is the only redirect to that origin.
215 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
216 (!ovl_verify_lower(dentry->d_sb) &&
217 (is_dir || lowerstat.nlink == 1))) {
218 fsid = ovl_layer_lower(dentry)->fsid;
219 stat->ino = lowerstat.ino;
223 * If we are querying a metacopy dentry and lower
224 * dentry is data dentry, then use the blocks we
225 * queried just now. We don't have to do additional
226 * vfs_getattr(). If lower itself is metacopy, then
227 * additional vfs_getattr() is unavoidable.
229 if (metacopy_blocks &&
230 realpath.dentry == ovl_dentry_lowerdata(dentry)) {
231 stat->blocks = lowerstat.blocks;
232 metacopy_blocks = false;
236 if (metacopy_blocks) {
238 * If lower is not same as lowerdata or if there was
239 * no origin on upper, we can end up here.
241 struct kstat lowerdatastat;
242 u32 lowermask = STATX_BLOCKS;
244 ovl_path_lowerdata(dentry, &realpath);
245 err = vfs_getattr(&realpath, &lowerdatastat,
249 stat->blocks = lowerdatastat.blocks;
253 err = ovl_map_dev_ino(dentry, stat, fsid);
258 * It's probably not worth it to count subdirs to get the
259 * correct link count. nlink=1 seems to pacify 'find' and
262 if (is_dir && OVL_TYPE_MERGE(type))
266 * Return the overlay inode nlinks for indexed upper inodes.
267 * Overlay inode nlink counts the union of the upper hardlinks
268 * and non-covered lower hardlinks. It does not include the upper
271 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
272 stat->nlink = dentry->d_inode->i_nlink;
275 revert_creds(old_cred);
280 int ovl_permission(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(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(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 err = vfs_getxattr(realdentry, name, NULL, 0);
361 err = ovl_copy_up(dentry);
365 realdentry = ovl_dentry_upper(dentry);
368 old_cred = ovl_override_creds(dentry->d_sb);
370 err = vfs_setxattr(realdentry, name, value, size, flags);
372 WARN_ON(flags != XATTR_REPLACE);
373 err = vfs_removexattr(realdentry, name);
375 revert_creds(old_cred);
378 ovl_copyattr(d_inode(realdentry), inode);
381 ovl_drop_write(dentry);
386 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
387 void *value, size_t size)
390 const struct cred *old_cred;
391 struct dentry *realdentry =
392 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
394 old_cred = ovl_override_creds(dentry->d_sb);
395 res = vfs_getxattr(realdentry, name, value, size);
396 revert_creds(old_cred);
400 static bool ovl_can_list(struct super_block *sb, const char *s)
402 /* Never list private (.overlay) */
403 if (ovl_is_private_xattr(sb, s))
406 /* List all non-trusted xatts */
407 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
410 /* list other trusted for superuser only */
411 return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
414 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
416 struct dentry *realdentry = ovl_dentry_real(dentry);
420 const struct cred *old_cred;
422 old_cred = ovl_override_creds(dentry->d_sb);
423 res = vfs_listxattr(realdentry, list, size);
424 revert_creds(old_cred);
425 if (res <= 0 || size == 0)
428 /* filter out private xattrs */
429 for (s = list, len = res; len;) {
430 size_t slen = strnlen(s, len) + 1;
432 /* underlying fs providing us with an broken xattr list? */
433 if (WARN_ON(slen > len))
437 if (!ovl_can_list(dentry->d_sb, s)) {
439 memmove(s, s + slen, len);
448 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
450 struct inode *realinode = ovl_inode_real(inode);
451 const struct cred *old_cred;
452 struct posix_acl *acl;
454 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
457 old_cred = ovl_override_creds(inode->i_sb);
458 acl = get_acl(realinode, type);
459 revert_creds(old_cred);
464 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
466 if (flags & S_ATIME) {
467 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
468 struct path upperpath = {
469 .mnt = ovl_upper_mnt(ofs),
470 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
473 if (upperpath.dentry) {
474 touch_atime(&upperpath);
475 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
481 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
485 struct inode *realinode = ovl_inode_realdata(inode);
486 const struct cred *old_cred;
488 if (!realinode->i_op->fiemap)
491 old_cred = ovl_override_creds(inode->i_sb);
492 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
493 revert_creds(old_cred);
498 static const struct inode_operations ovl_file_inode_operations = {
499 .setattr = ovl_setattr,
500 .permission = ovl_permission,
501 .getattr = ovl_getattr,
502 .listxattr = ovl_listxattr,
503 .get_acl = ovl_get_acl,
504 .update_time = ovl_update_time,
505 .fiemap = ovl_fiemap,
508 static const struct inode_operations ovl_symlink_inode_operations = {
509 .setattr = ovl_setattr,
510 .get_link = ovl_get_link,
511 .getattr = ovl_getattr,
512 .listxattr = ovl_listxattr,
513 .update_time = ovl_update_time,
516 static const struct inode_operations ovl_special_inode_operations = {
517 .setattr = ovl_setattr,
518 .permission = ovl_permission,
519 .getattr = ovl_getattr,
520 .listxattr = ovl_listxattr,
521 .get_acl = ovl_get_acl,
522 .update_time = ovl_update_time,
525 static const struct address_space_operations ovl_aops = {
526 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
527 .direct_IO = noop_direct_IO,
531 * It is possible to stack overlayfs instance on top of another
532 * overlayfs instance as lower layer. We need to annotate the
533 * stackable i_mutex locks according to stack level of the super
534 * block instance. An overlayfs instance can never be in stack
535 * depth 0 (there is always a real fs below it). An overlayfs
536 * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
538 * For example, here is a snip from /proc/lockdep_chains after
539 * dir_iterate of nested overlayfs:
541 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
542 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
543 * [...] &type->i_mutex_dir_key (stack_depth=0)
545 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
547 * This chain is valid:
548 * - inode->i_rwsem (inode_lock[2])
549 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
550 * - OVL_I(inode)->lock (ovl_inode_lock[2])
551 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
553 * And this chain is valid:
554 * - inode->i_rwsem (inode_lock[2])
555 * - OVL_I(inode)->lock (ovl_inode_lock[2])
556 * - lowerinode->i_rwsem (inode_lock[1])
557 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
559 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
560 * held, because it is in reverse order of the non-nested case using the same
562 * - inode->i_rwsem (inode_lock[1])
563 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
564 * - OVL_I(inode)->lock (ovl_inode_lock[1])
566 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
568 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
570 #ifdef CONFIG_LOCKDEP
571 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
572 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
573 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
575 int depth = inode->i_sb->s_stack_depth - 1;
577 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
580 if (S_ISDIR(inode->i_mode))
581 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
583 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
585 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
589 static void ovl_next_ino(struct inode *inode)
591 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
593 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
594 if (unlikely(!inode->i_ino))
595 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
598 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
600 int xinobits = ovl_xino_bits(inode->i_sb);
601 unsigned int xinoshift = 64 - xinobits;
604 * When d_ino is consistent with st_ino (samefs or i_ino has enough
605 * bits to encode layer), set the same value used for st_ino to i_ino,
606 * so inode number exposed via /proc/locks and a like will be
607 * consistent with d_ino and st_ino values. An i_ino value inconsistent
608 * with d_ino also causes nfsd readdirplus to fail.
611 if (ovl_same_fs(inode->i_sb)) {
613 } else if (xinobits && likely(!(ino >> xinoshift))) {
614 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
619 * For directory inodes on non-samefs with xino disabled or xino
620 * overflow, we allocate a non-persistent inode number, to be used for
621 * resolving st_ino collisions in ovl_map_dev_ino().
623 * To avoid ino collision with legitimate xino values from upper
624 * layer (fsid 0), use the lowest xinobit to map the non
625 * persistent inode numbers to the unified st_ino address space.
627 if (S_ISDIR(inode->i_mode)) {
630 inode->i_ino &= ~0UL >> xinobits;
631 inode->i_ino |= 1UL << xinoshift;
636 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
637 unsigned long ino, int fsid)
639 struct inode *realinode;
641 if (oip->upperdentry)
642 OVL_I(inode)->__upperdentry = oip->upperdentry;
643 if (oip->lowerpath && oip->lowerpath->dentry)
644 OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
646 OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
648 realinode = ovl_inode_real(inode);
649 ovl_copyattr(realinode, inode);
650 ovl_copyflags(realinode, inode);
651 ovl_map_ino(inode, ino, fsid);
654 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
656 inode->i_mode = mode;
657 inode->i_flags |= S_NOCMTIME;
658 #ifdef CONFIG_FS_POSIX_ACL
659 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
662 ovl_lockdep_annotate_inode_mutex_key(inode);
664 switch (mode & S_IFMT) {
666 inode->i_op = &ovl_file_inode_operations;
667 inode->i_fop = &ovl_file_operations;
668 inode->i_mapping->a_ops = &ovl_aops;
672 inode->i_op = &ovl_dir_inode_operations;
673 inode->i_fop = &ovl_dir_operations;
677 inode->i_op = &ovl_symlink_inode_operations;
681 inode->i_op = &ovl_special_inode_operations;
682 init_special_inode(inode, mode, rdev);
688 * With inodes index enabled, an overlay inode nlink counts the union of upper
689 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
690 * upper inode, the following nlink modifying operations can happen:
692 * 1. Lower hardlink copy up
693 * 2. Upper hardlink created, unlinked or renamed over
694 * 3. Lower hardlink whiteout or renamed over
696 * For the first, copy up case, the union nlink does not change, whether the
697 * operation succeeds or fails, but the upper inode nlink may change.
698 * Therefore, before copy up, we store the union nlink value relative to the
699 * lower inode nlink in the index inode xattr .overlay.nlink.
701 * For the second, upper hardlink case, the union nlink should be incremented
702 * or decremented IFF the operation succeeds, aligned with nlink change of the
703 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
704 * value relative to the upper inode nlink in the index inode.
706 * For the last, lower cover up case, we simplify things by preceding the
707 * whiteout or cover up with copy up. This makes sure that there is an index
708 * upper inode where the nlink xattr can be stored before the copied up upper
711 #define OVL_NLINK_ADD_UPPER (1 << 0)
714 * On-disk format for indexed nlink:
716 * nlink relative to the upper inode - "U[+-]NUM"
717 * nlink relative to the lower inode - "L[+-]NUM"
720 static int ovl_set_nlink_common(struct dentry *dentry,
721 struct dentry *realdentry, const char *format)
723 struct inode *inode = d_inode(dentry);
724 struct inode *realinode = d_inode(realdentry);
728 len = snprintf(buf, sizeof(buf), format,
729 (int) (inode->i_nlink - realinode->i_nlink));
731 if (WARN_ON(len >= sizeof(buf)))
734 return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
735 OVL_XATTR_NLINK, buf, len);
738 int ovl_set_nlink_upper(struct dentry *dentry)
740 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
743 int ovl_set_nlink_lower(struct dentry *dentry)
745 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
748 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
749 struct dentry *upperdentry,
750 unsigned int fallback)
757 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
760 err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
761 &buf, sizeof(buf) - 1);
766 if ((buf[0] != 'L' && buf[0] != 'U') ||
767 (buf[1] != '+' && buf[1] != '-'))
770 err = kstrtoint(buf + 1, 10, &nlink_diff);
774 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
783 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
788 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
792 inode = new_inode(sb);
794 ovl_fill_inode(inode, mode, rdev);
799 static int ovl_inode_test(struct inode *inode, void *data)
801 return inode->i_private == data;
804 static int ovl_inode_set(struct inode *inode, void *data)
806 inode->i_private = data;
810 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
811 struct dentry *upperdentry, bool strict)
814 * For directories, @strict verify from lookup path performs consistency
815 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
816 * inode. Non @strict verify from NFS handle decode path passes NULL for
817 * 'unknown' lower/upper.
819 if (S_ISDIR(inode->i_mode) && strict) {
820 /* Real lower dir moved to upper layer under us? */
821 if (!lowerdentry && ovl_inode_lower(inode))
824 /* Lookup of an uncovered redirect origin? */
825 if (!upperdentry && ovl_inode_upper(inode))
830 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
831 * This happens when finding a copied up overlay inode for a renamed
832 * or hardlinked overlay dentry and lower dentry cannot be followed
833 * by origin because lower fs does not support file handles.
835 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
839 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
840 * This happens when finding a lower alias for a copied up hard link.
842 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
848 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
851 struct inode *inode, *key = d_inode(real);
853 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
857 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
858 is_upper ? real : NULL, false)) {
860 return ERR_PTR(-ESTALE);
866 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
868 struct inode *key = d_inode(dir);
872 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
876 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
877 !ovl_inode_lower(trap);
884 * Create an inode cache entry for layer root dir, that will intentionally
885 * fail ovl_verify_inode(), so any lookup that will find some layer root
888 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
890 struct inode *key = d_inode(dir);
894 return ERR_PTR(-ENOTDIR);
896 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
899 return ERR_PTR(-ENOMEM);
901 if (!(trap->i_state & I_NEW)) {
902 /* Conflicting layer roots? */
904 return ERR_PTR(-ELOOP);
907 trap->i_mode = S_IFDIR;
908 trap->i_flags = S_DEAD;
909 unlock_new_inode(trap);
915 * Does overlay inode need to be hashed by lower inode?
917 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
918 struct dentry *lower, bool index)
920 struct ovl_fs *ofs = sb->s_fs_info;
922 /* No, if pure upper */
926 /* Yes, if already indexed */
930 /* Yes, if won't be copied up */
931 if (!ovl_upper_mnt(ofs))
934 /* No, if lower hardlink is or will be broken on copy up */
935 if ((upper || !ovl_indexdir(sb)) &&
936 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
939 /* No, if non-indexed upper with NFS export */
940 if (sb->s_export_op && upper)
943 /* Otherwise, hash by lower inode for fsnotify */
947 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
950 return newinode ? inode_insert5(newinode, (unsigned long) key,
951 ovl_inode_test, ovl_inode_set, key) :
952 iget5_locked(sb, (unsigned long) key,
953 ovl_inode_test, ovl_inode_set, key);
956 struct inode *ovl_get_inode(struct super_block *sb,
957 struct ovl_inode_params *oip)
959 struct ovl_fs *ofs = OVL_FS(sb);
960 struct dentry *upperdentry = oip->upperdentry;
961 struct ovl_path *lowerpath = oip->lowerpath;
962 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
964 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
965 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
967 int fsid = bylower ? lowerpath->layer->fsid : 0;
969 unsigned long ino = 0;
970 int err = oip->newinode ? -EEXIST : -ENOMEM;
973 realinode = d_inode(lowerdentry);
976 * Copy up origin (lower) may exist for non-indexed upper, but we must
977 * not use lower as hash key if this is a broken hardlink.
979 is_dir = S_ISDIR(realinode->i_mode);
980 if (upperdentry || bylower) {
981 struct inode *key = d_inode(bylower ? lowerdentry :
983 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
985 inode = ovl_iget5(sb, oip->newinode, key);
988 if (!(inode->i_state & I_NEW)) {
990 * Verify that the underlying files stored in the inode
991 * match those in the dentry.
993 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1001 kfree(oip->redirect);
1005 /* Recalculate nlink for non-dir due to indexing */
1007 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1009 set_nlink(inode, nlink);
1012 /* Lower hardlink that will be broken on copy up */
1013 inode = new_inode(sb);
1018 ino = realinode->i_ino;
1019 fsid = lowerpath->layer->fsid;
1021 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1022 ovl_inode_init(inode, oip, ino, fsid);
1024 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1025 ovl_set_flag(OVL_IMPURE, inode);
1028 ovl_set_flag(OVL_INDEX, inode);
1030 OVL_I(inode)->redirect = oip->redirect;
1033 ovl_set_flag(OVL_CONST_INO, inode);
1035 /* Check for non-merge dir that may have whiteouts */
1037 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1038 ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
1039 ovl_set_flag(OVL_WHITEOUTS, inode);
1043 if (inode->i_state & I_NEW)
1044 unlock_new_inode(inode);
1049 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1050 inode = ERR_PTR(err);