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 <linux/namei.h>
17 #include "overlayfs.h"
20 int ovl_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
24 bool full_copy_up = false;
25 struct dentry *upperdentry;
26 const struct cred *old_cred;
28 err = setattr_prepare(&init_user_ns, dentry, attr);
32 err = ovl_want_write(dentry);
36 if (attr->ia_valid & ATTR_SIZE) {
37 /* Truncate should trigger data copy up as well */
42 err = ovl_copy_up(dentry);
44 err = ovl_copy_up_with_data(dentry);
46 struct inode *winode = NULL;
48 upperdentry = ovl_dentry_upper(dentry);
50 if (attr->ia_valid & ATTR_SIZE) {
51 winode = d_inode(upperdentry);
52 err = get_write_access(winode);
57 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
58 attr->ia_valid &= ~ATTR_MODE;
61 * We might have to translate ovl file into real file object
62 * once use cases emerge. For now, simply don't let underlying
63 * filesystem rely on attr->ia_file
65 attr->ia_valid &= ~ATTR_FILE;
68 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
69 * set. Overlayfs does not pass O_TRUNC flag to underlying
70 * filesystem during open -> do not pass ATTR_OPEN. This
71 * disables optimization in fuse which assumes open(O_TRUNC)
72 * already set file size to 0. But we never passed O_TRUNC to
73 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
74 * setattr request to server.
76 attr->ia_valid &= ~ATTR_OPEN;
78 inode_lock(upperdentry->d_inode);
79 old_cred = ovl_override_creds(dentry->d_sb);
80 err = notify_change(&init_user_ns, upperdentry, attr, NULL);
81 revert_creds(old_cred);
83 ovl_copyattr(dentry->d_inode);
84 inode_unlock(upperdentry->d_inode);
87 put_write_access(winode);
90 ovl_drop_write(dentry);
95 static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
97 bool samefs = ovl_same_fs(dentry->d_sb);
98 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
99 unsigned int xinoshift = 64 - xinobits;
103 * When all layers are on the same fs, all real inode
104 * number are unique, so we use the overlay st_dev,
105 * which is friendly to du -x.
107 stat->dev = dentry->d_sb->s_dev;
109 } else if (xinobits) {
111 * All inode numbers of underlying fs should not be using the
112 * high xinobits, so we use high xinobits to partition the
113 * overlay st_ino address space. The high bits holds the fsid
114 * (upper fsid is 0). The lowest xinobit is reserved for mapping
115 * the non-persistent inode numbers range in case of overflow.
116 * This way all overlay inode numbers are unique and use the
119 if (likely(!(stat->ino >> xinoshift))) {
120 stat->ino |= ((u64)fsid) << (xinoshift + 1);
121 stat->dev = dentry->d_sb->s_dev;
123 } else if (ovl_xino_warn(dentry->d_sb)) {
124 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
125 dentry, stat->ino, xinobits);
129 /* The inode could not be mapped to a unified st_ino address space */
130 if (S_ISDIR(dentry->d_inode->i_mode)) {
132 * Always use the overlay st_dev for directories, so 'find
133 * -xdev' will scan the entire overlay mount and won't cross the
134 * overlay mount boundaries.
136 * If not all layers are on the same fs the pair {real st_ino;
137 * overlay st_dev} is not unique, so use the non persistent
138 * overlay st_ino for directories.
140 stat->dev = dentry->d_sb->s_dev;
141 stat->ino = dentry->d_inode->i_ino;
144 * For non-samefs setup, if we cannot map all layers st_ino
145 * to a unified address space, we need to make sure that st_dev
146 * is unique per underlying fs, so we use the unique anonymous
147 * bdev assigned to the underlying fs.
149 stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
153 int ovl_getattr(struct user_namespace *mnt_userns, const struct path *path,
154 struct kstat *stat, u32 request_mask, unsigned int flags)
156 struct dentry *dentry = path->dentry;
157 enum ovl_path_type type;
158 struct path realpath;
159 const struct cred *old_cred;
160 struct inode *inode = d_inode(dentry);
161 bool is_dir = S_ISDIR(inode->i_mode);
164 bool metacopy_blocks = false;
166 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
168 type = ovl_path_real(dentry, &realpath);
169 old_cred = ovl_override_creds(dentry->d_sb);
170 err = vfs_getattr(&realpath, stat, request_mask, flags);
174 /* Report the effective immutable/append-only STATX flags */
175 generic_fill_statx_attr(inode, stat);
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 ovl_map_dev_ino(dentry, stat, fsid);
256 * It's probably not worth it to count subdirs to get the
257 * correct link count. nlink=1 seems to pacify 'find' and
260 if (is_dir && OVL_TYPE_MERGE(type))
264 * Return the overlay inode nlinks for indexed upper inodes.
265 * Overlay inode nlink counts the union of the upper hardlinks
266 * and non-covered lower hardlinks. It does not include the upper
269 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
270 stat->nlink = dentry->d_inode->i_nlink;
273 revert_creds(old_cred);
278 int ovl_permission(struct user_namespace *mnt_userns,
279 struct inode *inode, int mask)
281 struct inode *upperinode = ovl_inode_upper(inode);
282 struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
283 const struct cred *old_cred;
286 /* Careful in RCU walk mode */
288 WARN_ON(!(mask & MAY_NOT_BLOCK));
293 * Check overlay inode with the creds of task and underlying inode
294 * with creds of mounter
296 err = generic_permission(&init_user_ns, inode, mask);
300 old_cred = ovl_override_creds(inode->i_sb);
302 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
303 mask &= ~(MAY_WRITE | MAY_APPEND);
304 /* Make sure mounter can read file for copy up later */
307 err = inode_permission(&init_user_ns, realinode, mask);
308 revert_creds(old_cred);
313 static const char *ovl_get_link(struct dentry *dentry,
315 struct delayed_call *done)
317 const struct cred *old_cred;
321 return ERR_PTR(-ECHILD);
323 old_cred = ovl_override_creds(dentry->d_sb);
324 p = vfs_get_link(ovl_dentry_real(dentry), done);
325 revert_creds(old_cred);
329 bool ovl_is_private_xattr(struct super_block *sb, const char *name)
331 struct ovl_fs *ofs = sb->s_fs_info;
333 if (ofs->config.userxattr)
334 return strncmp(name, OVL_XATTR_USER_PREFIX,
335 sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
337 return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
338 sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
341 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
342 const void *value, size_t size, int flags)
345 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
346 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
347 const struct cred *old_cred;
349 err = ovl_want_write(dentry);
353 if (!value && !upperdentry) {
354 old_cred = ovl_override_creds(dentry->d_sb);
355 err = vfs_getxattr(&init_user_ns, realdentry, name, NULL, 0);
356 revert_creds(old_cred);
362 err = ovl_copy_up(dentry);
366 realdentry = ovl_dentry_upper(dentry);
369 old_cred = ovl_override_creds(dentry->d_sb);
371 err = vfs_setxattr(&init_user_ns, realdentry, name, value, size,
374 WARN_ON(flags != XATTR_REPLACE);
375 err = vfs_removexattr(&init_user_ns, realdentry, name);
377 revert_creds(old_cred);
383 ovl_drop_write(dentry);
388 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
389 void *value, size_t size)
392 const struct cred *old_cred;
393 struct dentry *realdentry =
394 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
396 old_cred = ovl_override_creds(dentry->d_sb);
397 res = vfs_getxattr(&init_user_ns, realdentry, name, value, size);
398 revert_creds(old_cred);
402 static bool ovl_can_list(struct super_block *sb, const char *s)
404 /* Never list private (.overlay) */
405 if (ovl_is_private_xattr(sb, s))
408 /* List all non-trusted xattrs */
409 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
412 /* list other trusted for superuser only */
413 return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
416 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
418 struct dentry *realdentry = ovl_dentry_real(dentry);
422 const struct cred *old_cred;
424 old_cred = ovl_override_creds(dentry->d_sb);
425 res = vfs_listxattr(realdentry, list, size);
426 revert_creds(old_cred);
427 if (res <= 0 || size == 0)
430 /* filter out private xattrs */
431 for (s = list, len = res; len;) {
432 size_t slen = strnlen(s, len) + 1;
434 /* underlying fs providing us with an broken xattr list? */
435 if (WARN_ON(slen > len))
439 if (!ovl_can_list(dentry->d_sb, s)) {
441 memmove(s, s + slen, len);
450 struct posix_acl *ovl_get_acl(struct inode *inode, int type, bool rcu)
452 struct inode *realinode = ovl_inode_real(inode);
453 const struct cred *old_cred;
454 struct posix_acl *acl;
456 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL))
461 return ERR_PTR(-ECHILD);
464 if (!IS_POSIXACL(realinode))
468 return get_cached_acl_rcu(realinode, type);
470 old_cred = ovl_override_creds(inode->i_sb);
471 acl = get_acl(realinode, type);
472 revert_creds(old_cred);
477 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
479 if (flags & S_ATIME) {
480 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
481 struct path upperpath = {
482 .mnt = ovl_upper_mnt(ofs),
483 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
486 if (upperpath.dentry) {
487 touch_atime(&upperpath);
488 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
494 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
498 struct inode *realinode = ovl_inode_realdata(inode);
499 const struct cred *old_cred;
501 if (!realinode->i_op->fiemap)
504 old_cred = ovl_override_creds(inode->i_sb);
505 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
506 revert_creds(old_cred);
512 * Work around the fact that security_file_ioctl() takes a file argument.
513 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
516 static int ovl_security_fileattr(struct path *realpath, struct fileattr *fa,
523 file = dentry_open(realpath, O_RDONLY, current_cred());
525 return PTR_ERR(file);
528 cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
530 cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
532 err = security_file_ioctl(file, cmd, 0);
538 int ovl_real_fileattr_set(struct path *realpath, struct fileattr *fa)
542 err = ovl_security_fileattr(realpath, fa, true);
546 return vfs_fileattr_set(&init_user_ns, realpath->dentry, fa);
549 int ovl_fileattr_set(struct user_namespace *mnt_userns,
550 struct dentry *dentry, struct fileattr *fa)
552 struct inode *inode = d_inode(dentry);
553 struct path upperpath;
554 const struct cred *old_cred;
558 err = ovl_want_write(dentry);
562 err = ovl_copy_up(dentry);
564 ovl_path_real(dentry, &upperpath);
566 old_cred = ovl_override_creds(inode->i_sb);
568 * Store immutable/append-only flags in xattr and clear them
569 * in upper fileattr (in case they were set by older kernel)
570 * so children of "ovl-immutable" directories lower aliases of
571 * "ovl-immutable" hardlinks could be copied up.
572 * Clear xattr when flags are cleared.
574 err = ovl_set_protattr(inode, upperpath.dentry, fa);
576 err = ovl_real_fileattr_set(&upperpath, fa);
577 revert_creds(old_cred);
580 * Merge real inode flags with inode flags read from
581 * overlay.protattr xattr
583 flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK;
585 BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK);
586 flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK;
587 inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK);
592 ovl_drop_write(dentry);
597 /* Convert inode protection flags to fileattr flags */
598 static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa)
600 BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL);
601 BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
603 if (inode->i_flags & S_APPEND) {
604 fa->flags |= FS_APPEND_FL;
605 fa->fsx_xflags |= FS_XFLAG_APPEND;
607 if (inode->i_flags & S_IMMUTABLE) {
608 fa->flags |= FS_IMMUTABLE_FL;
609 fa->fsx_xflags |= FS_XFLAG_IMMUTABLE;
613 int ovl_real_fileattr_get(struct path *realpath, struct fileattr *fa)
617 err = ovl_security_fileattr(realpath, fa, false);
621 err = vfs_fileattr_get(realpath->dentry, fa);
622 if (err == -ENOIOCTLCMD)
627 int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa)
629 struct inode *inode = d_inode(dentry);
630 struct path realpath;
631 const struct cred *old_cred;
634 ovl_path_real(dentry, &realpath);
636 old_cred = ovl_override_creds(inode->i_sb);
637 err = ovl_real_fileattr_get(&realpath, fa);
638 ovl_fileattr_prot_flags(inode, fa);
639 revert_creds(old_cred);
644 static const struct inode_operations ovl_file_inode_operations = {
645 .setattr = ovl_setattr,
646 .permission = ovl_permission,
647 .getattr = ovl_getattr,
648 .listxattr = ovl_listxattr,
649 .get_acl = ovl_get_acl,
650 .update_time = ovl_update_time,
651 .fiemap = ovl_fiemap,
652 .fileattr_get = ovl_fileattr_get,
653 .fileattr_set = ovl_fileattr_set,
656 static const struct inode_operations ovl_symlink_inode_operations = {
657 .setattr = ovl_setattr,
658 .get_link = ovl_get_link,
659 .getattr = ovl_getattr,
660 .listxattr = ovl_listxattr,
661 .update_time = ovl_update_time,
664 static const struct inode_operations ovl_special_inode_operations = {
665 .setattr = ovl_setattr,
666 .permission = ovl_permission,
667 .getattr = ovl_getattr,
668 .listxattr = ovl_listxattr,
669 .get_acl = ovl_get_acl,
670 .update_time = ovl_update_time,
673 static const struct address_space_operations ovl_aops = {
674 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
675 .direct_IO = noop_direct_IO,
679 * It is possible to stack overlayfs instance on top of another
680 * overlayfs instance as lower layer. We need to annotate the
681 * stackable i_mutex locks according to stack level of the super
682 * block instance. An overlayfs instance can never be in stack
683 * depth 0 (there is always a real fs below it). An overlayfs
684 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
686 * For example, here is a snip from /proc/lockdep_chains after
687 * dir_iterate of nested overlayfs:
689 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
690 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
691 * [...] &type->i_mutex_dir_key (stack_depth=0)
693 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
695 * This chain is valid:
696 * - inode->i_rwsem (inode_lock[2])
697 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
698 * - OVL_I(inode)->lock (ovl_inode_lock[2])
699 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
701 * And this chain is valid:
702 * - inode->i_rwsem (inode_lock[2])
703 * - OVL_I(inode)->lock (ovl_inode_lock[2])
704 * - lowerinode->i_rwsem (inode_lock[1])
705 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
707 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
708 * held, because it is in reverse order of the non-nested case using the same
710 * - inode->i_rwsem (inode_lock[1])
711 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
712 * - OVL_I(inode)->lock (ovl_inode_lock[1])
714 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
716 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
718 #ifdef CONFIG_LOCKDEP
719 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
720 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
721 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
723 int depth = inode->i_sb->s_stack_depth - 1;
725 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
728 if (S_ISDIR(inode->i_mode))
729 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
731 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
733 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
737 static void ovl_next_ino(struct inode *inode)
739 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
741 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
742 if (unlikely(!inode->i_ino))
743 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
746 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
748 int xinobits = ovl_xino_bits(inode->i_sb);
749 unsigned int xinoshift = 64 - xinobits;
752 * When d_ino is consistent with st_ino (samefs or i_ino has enough
753 * bits to encode layer), set the same value used for st_ino to i_ino,
754 * so inode number exposed via /proc/locks and a like will be
755 * consistent with d_ino and st_ino values. An i_ino value inconsistent
756 * with d_ino also causes nfsd readdirplus to fail.
759 if (ovl_same_fs(inode->i_sb)) {
761 } else if (xinobits && likely(!(ino >> xinoshift))) {
762 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
767 * For directory inodes on non-samefs with xino disabled or xino
768 * overflow, we allocate a non-persistent inode number, to be used for
769 * resolving st_ino collisions in ovl_map_dev_ino().
771 * To avoid ino collision with legitimate xino values from upper
772 * layer (fsid 0), use the lowest xinobit to map the non
773 * persistent inode numbers to the unified st_ino address space.
775 if (S_ISDIR(inode->i_mode)) {
778 inode->i_ino &= ~0UL >> xinobits;
779 inode->i_ino |= 1UL << xinoshift;
784 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
785 unsigned long ino, int fsid)
787 struct inode *realinode;
788 struct ovl_inode *oi = OVL_I(inode);
790 if (oip->upperdentry)
791 oi->__upperdentry = oip->upperdentry;
792 if (oip->lowerpath && oip->lowerpath->dentry) {
793 oi->lowerpath.dentry = dget(oip->lowerpath->dentry);
794 oi->lowerpath.layer = oip->lowerpath->layer;
797 oi->lowerdata = igrab(d_inode(oip->lowerdata));
799 realinode = ovl_inode_real(inode);
801 ovl_copyflags(realinode, inode);
802 ovl_map_ino(inode, ino, fsid);
805 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
807 inode->i_mode = mode;
808 inode->i_flags |= S_NOCMTIME;
809 #ifdef CONFIG_FS_POSIX_ACL
810 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
813 ovl_lockdep_annotate_inode_mutex_key(inode);
815 switch (mode & S_IFMT) {
817 inode->i_op = &ovl_file_inode_operations;
818 inode->i_fop = &ovl_file_operations;
819 inode->i_mapping->a_ops = &ovl_aops;
823 inode->i_op = &ovl_dir_inode_operations;
824 inode->i_fop = &ovl_dir_operations;
828 inode->i_op = &ovl_symlink_inode_operations;
832 inode->i_op = &ovl_special_inode_operations;
833 init_special_inode(inode, mode, rdev);
839 * With inodes index enabled, an overlay inode nlink counts the union of upper
840 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
841 * upper inode, the following nlink modifying operations can happen:
843 * 1. Lower hardlink copy up
844 * 2. Upper hardlink created, unlinked or renamed over
845 * 3. Lower hardlink whiteout or renamed over
847 * For the first, copy up case, the union nlink does not change, whether the
848 * operation succeeds or fails, but the upper inode nlink may change.
849 * Therefore, before copy up, we store the union nlink value relative to the
850 * lower inode nlink in the index inode xattr .overlay.nlink.
852 * For the second, upper hardlink case, the union nlink should be incremented
853 * or decremented IFF the operation succeeds, aligned with nlink change of the
854 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
855 * value relative to the upper inode nlink in the index inode.
857 * For the last, lower cover up case, we simplify things by preceding the
858 * whiteout or cover up with copy up. This makes sure that there is an index
859 * upper inode where the nlink xattr can be stored before the copied up upper
862 #define OVL_NLINK_ADD_UPPER (1 << 0)
865 * On-disk format for indexed nlink:
867 * nlink relative to the upper inode - "U[+-]NUM"
868 * nlink relative to the lower inode - "L[+-]NUM"
871 static int ovl_set_nlink_common(struct dentry *dentry,
872 struct dentry *realdentry, const char *format)
874 struct inode *inode = d_inode(dentry);
875 struct inode *realinode = d_inode(realdentry);
879 len = snprintf(buf, sizeof(buf), format,
880 (int) (inode->i_nlink - realinode->i_nlink));
882 if (WARN_ON(len >= sizeof(buf)))
885 return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
886 OVL_XATTR_NLINK, buf, len);
889 int ovl_set_nlink_upper(struct dentry *dentry)
891 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
894 int ovl_set_nlink_lower(struct dentry *dentry)
896 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
899 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
900 struct dentry *upperdentry,
901 unsigned int fallback)
908 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
911 err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
912 &buf, sizeof(buf) - 1);
917 if ((buf[0] != 'L' && buf[0] != 'U') ||
918 (buf[1] != '+' && buf[1] != '-'))
921 err = kstrtoint(buf + 1, 10, &nlink_diff);
925 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
934 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
939 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
943 inode = new_inode(sb);
945 ovl_fill_inode(inode, mode, rdev);
950 static int ovl_inode_test(struct inode *inode, void *data)
952 return inode->i_private == data;
955 static int ovl_inode_set(struct inode *inode, void *data)
957 inode->i_private = data;
961 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
962 struct dentry *upperdentry, bool strict)
965 * For directories, @strict verify from lookup path performs consistency
966 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
967 * inode. Non @strict verify from NFS handle decode path passes NULL for
968 * 'unknown' lower/upper.
970 if (S_ISDIR(inode->i_mode) && strict) {
971 /* Real lower dir moved to upper layer under us? */
972 if (!lowerdentry && ovl_inode_lower(inode))
975 /* Lookup of an uncovered redirect origin? */
976 if (!upperdentry && ovl_inode_upper(inode))
981 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
982 * This happens when finding a copied up overlay inode for a renamed
983 * or hardlinked overlay dentry and lower dentry cannot be followed
984 * by origin because lower fs does not support file handles.
986 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
990 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
991 * This happens when finding a lower alias for a copied up hard link.
993 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
999 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
1002 struct inode *inode, *key = d_inode(real);
1004 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1008 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
1009 is_upper ? real : NULL, false)) {
1011 return ERR_PTR(-ESTALE);
1017 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
1019 struct inode *key = d_inode(dir);
1023 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1027 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
1028 !ovl_inode_lower(trap);
1035 * Create an inode cache entry for layer root dir, that will intentionally
1036 * fail ovl_verify_inode(), so any lookup that will find some layer root
1039 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
1041 struct inode *key = d_inode(dir);
1045 return ERR_PTR(-ENOTDIR);
1047 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
1048 ovl_inode_set, key);
1050 return ERR_PTR(-ENOMEM);
1052 if (!(trap->i_state & I_NEW)) {
1053 /* Conflicting layer roots? */
1055 return ERR_PTR(-ELOOP);
1058 trap->i_mode = S_IFDIR;
1059 trap->i_flags = S_DEAD;
1060 unlock_new_inode(trap);
1066 * Does overlay inode need to be hashed by lower inode?
1068 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
1069 struct dentry *lower, bool index)
1071 struct ovl_fs *ofs = sb->s_fs_info;
1073 /* No, if pure upper */
1077 /* Yes, if already indexed */
1081 /* Yes, if won't be copied up */
1082 if (!ovl_upper_mnt(ofs))
1085 /* No, if lower hardlink is or will be broken on copy up */
1086 if ((upper || !ovl_indexdir(sb)) &&
1087 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1090 /* No, if non-indexed upper with NFS export */
1091 if (sb->s_export_op && upper)
1094 /* Otherwise, hash by lower inode for fsnotify */
1098 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1101 return newinode ? inode_insert5(newinode, (unsigned long) key,
1102 ovl_inode_test, ovl_inode_set, key) :
1103 iget5_locked(sb, (unsigned long) key,
1104 ovl_inode_test, ovl_inode_set, key);
1107 struct inode *ovl_get_inode(struct super_block *sb,
1108 struct ovl_inode_params *oip)
1110 struct ovl_fs *ofs = OVL_FS(sb);
1111 struct dentry *upperdentry = oip->upperdentry;
1112 struct ovl_path *lowerpath = oip->lowerpath;
1113 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1114 struct inode *inode;
1115 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1116 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1118 int fsid = bylower ? lowerpath->layer->fsid : 0;
1120 unsigned long ino = 0;
1121 int err = oip->newinode ? -EEXIST : -ENOMEM;
1124 realinode = d_inode(lowerdentry);
1127 * Copy up origin (lower) may exist for non-indexed upper, but we must
1128 * not use lower as hash key if this is a broken hardlink.
1130 is_dir = S_ISDIR(realinode->i_mode);
1131 if (upperdentry || bylower) {
1132 struct inode *key = d_inode(bylower ? lowerdentry :
1134 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1136 inode = ovl_iget5(sb, oip->newinode, key);
1139 if (!(inode->i_state & I_NEW)) {
1141 * Verify that the underlying files stored in the inode
1142 * match those in the dentry.
1144 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1152 kfree(oip->redirect);
1156 /* Recalculate nlink for non-dir due to indexing */
1158 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1160 set_nlink(inode, nlink);
1163 /* Lower hardlink that will be broken on copy up */
1164 inode = new_inode(sb);
1169 ino = realinode->i_ino;
1170 fsid = lowerpath->layer->fsid;
1172 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1173 ovl_inode_init(inode, oip, ino, fsid);
1175 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1176 ovl_set_flag(OVL_IMPURE, inode);
1179 ovl_set_flag(OVL_INDEX, inode);
1181 OVL_I(inode)->redirect = oip->redirect;
1184 ovl_set_flag(OVL_CONST_INO, inode);
1186 /* Check for non-merge dir that may have whiteouts */
1188 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1189 ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
1190 ovl_set_flag(OVL_WHITEOUTS, inode);
1194 /* Check for immutable/append-only inode flags in xattr */
1196 ovl_check_protattr(inode, upperdentry);
1198 if (inode->i_state & I_NEW)
1199 unlock_new_inode(inode);
1204 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1205 inode = ERR_PTR(err);