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 struct inode *inode = d_inode(dentry);
166 bool is_dir = S_ISDIR(inode->i_mode);
169 bool metacopy_blocks = false;
171 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
173 type = ovl_path_real(dentry, &realpath);
174 old_cred = ovl_override_creds(dentry->d_sb);
175 err = vfs_getattr(&realpath, stat, request_mask, flags);
179 /* Report the effective immutable/append-only STATX flags */
180 generic_fill_statx_attr(inode, stat);
183 * For non-dir or same fs, we use st_ino of the copy up origin.
184 * This guaranties constant st_dev/st_ino across copy up.
185 * With xino feature and non-samefs, we use st_ino of the copy up
186 * origin masked with high bits that represent the layer id.
188 * If lower filesystem supports NFS file handles, this also guaranties
189 * persistent st_ino across mount cycle.
191 if (!is_dir || ovl_same_dev(dentry->d_sb)) {
192 if (!OVL_TYPE_UPPER(type)) {
193 fsid = ovl_layer_lower(dentry)->fsid;
194 } else if (OVL_TYPE_ORIGIN(type)) {
195 struct kstat lowerstat;
196 u32 lowermask = STATX_INO | STATX_BLOCKS |
197 (!is_dir ? STATX_NLINK : 0);
199 ovl_path_lower(dentry, &realpath);
200 err = vfs_getattr(&realpath, &lowerstat,
206 * Lower hardlinks may be broken on copy up to different
207 * upper files, so we cannot use the lower origin st_ino
208 * for those different files, even for the same fs case.
210 * Similarly, several redirected dirs can point to the
211 * same dir on a lower layer. With the "verify_lower"
212 * feature, we do not use the lower origin st_ino, if
213 * we haven't verified that this redirect is unique.
215 * With inodes index enabled, it is safe to use st_ino
216 * of an indexed origin. The index validates that the
217 * upper hardlink is not broken and that a redirected
218 * dir is the only redirect to that origin.
220 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
221 (!ovl_verify_lower(dentry->d_sb) &&
222 (is_dir || lowerstat.nlink == 1))) {
223 fsid = ovl_layer_lower(dentry)->fsid;
224 stat->ino = lowerstat.ino;
228 * If we are querying a metacopy dentry and lower
229 * dentry is data dentry, then use the blocks we
230 * queried just now. We don't have to do additional
231 * vfs_getattr(). If lower itself is metacopy, then
232 * additional vfs_getattr() is unavoidable.
234 if (metacopy_blocks &&
235 realpath.dentry == ovl_dentry_lowerdata(dentry)) {
236 stat->blocks = lowerstat.blocks;
237 metacopy_blocks = false;
241 if (metacopy_blocks) {
243 * If lower is not same as lowerdata or if there was
244 * no origin on upper, we can end up here.
246 struct kstat lowerdatastat;
247 u32 lowermask = STATX_BLOCKS;
249 ovl_path_lowerdata(dentry, &realpath);
250 err = vfs_getattr(&realpath, &lowerdatastat,
254 stat->blocks = lowerdatastat.blocks;
258 ovl_map_dev_ino(dentry, stat, fsid);
261 * It's probably not worth it to count subdirs to get the
262 * correct link count. nlink=1 seems to pacify 'find' and
265 if (is_dir && OVL_TYPE_MERGE(type))
269 * Return the overlay inode nlinks for indexed upper inodes.
270 * Overlay inode nlink counts the union of the upper hardlinks
271 * and non-covered lower hardlinks. It does not include the upper
274 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
275 stat->nlink = dentry->d_inode->i_nlink;
278 revert_creds(old_cred);
283 int ovl_permission(struct user_namespace *mnt_userns,
284 struct inode *inode, int mask)
286 struct inode *upperinode = ovl_inode_upper(inode);
287 struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
288 const struct cred *old_cred;
291 /* Careful in RCU walk mode */
293 WARN_ON(!(mask & MAY_NOT_BLOCK));
298 * Check overlay inode with the creds of task and underlying inode
299 * with creds of mounter
301 err = generic_permission(&init_user_ns, inode, mask);
305 old_cred = ovl_override_creds(inode->i_sb);
307 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
308 mask &= ~(MAY_WRITE | MAY_APPEND);
309 /* Make sure mounter can read file for copy up later */
312 err = inode_permission(&init_user_ns, realinode, mask);
313 revert_creds(old_cred);
318 static const char *ovl_get_link(struct dentry *dentry,
320 struct delayed_call *done)
322 const struct cred *old_cred;
326 return ERR_PTR(-ECHILD);
328 old_cred = ovl_override_creds(dentry->d_sb);
329 p = vfs_get_link(ovl_dentry_real(dentry), done);
330 revert_creds(old_cred);
334 bool ovl_is_private_xattr(struct super_block *sb, const char *name)
336 struct ovl_fs *ofs = sb->s_fs_info;
338 if (ofs->config.userxattr)
339 return strncmp(name, OVL_XATTR_USER_PREFIX,
340 sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
342 return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
343 sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
346 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
347 const void *value, size_t size, int flags)
350 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
351 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
352 const struct cred *old_cred;
354 err = ovl_want_write(dentry);
358 if (!value && !upperdentry) {
359 old_cred = ovl_override_creds(dentry->d_sb);
360 err = vfs_getxattr(&init_user_ns, realdentry, name, NULL, 0);
361 revert_creds(old_cred);
367 err = ovl_copy_up(dentry);
371 realdentry = ovl_dentry_upper(dentry);
374 old_cred = ovl_override_creds(dentry->d_sb);
376 err = vfs_setxattr(&init_user_ns, realdentry, name, value, size,
379 WARN_ON(flags != XATTR_REPLACE);
380 err = vfs_removexattr(&init_user_ns, realdentry, name);
382 revert_creds(old_cred);
385 ovl_copyattr(d_inode(realdentry), inode);
388 ovl_drop_write(dentry);
393 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
394 void *value, size_t size)
397 const struct cred *old_cred;
398 struct dentry *realdentry =
399 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
401 old_cred = ovl_override_creds(dentry->d_sb);
402 res = vfs_getxattr(&init_user_ns, realdentry, name, value, size);
403 revert_creds(old_cred);
407 static bool ovl_can_list(struct super_block *sb, const char *s)
409 /* Never list private (.overlay) */
410 if (ovl_is_private_xattr(sb, s))
413 /* List all non-trusted xattrs */
414 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
417 /* list other trusted for superuser only */
418 return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
421 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
423 struct dentry *realdentry = ovl_dentry_real(dentry);
427 const struct cred *old_cred;
429 old_cred = ovl_override_creds(dentry->d_sb);
430 res = vfs_listxattr(realdentry, list, size);
431 revert_creds(old_cred);
432 if (res <= 0 || size == 0)
435 /* filter out private xattrs */
436 for (s = list, len = res; len;) {
437 size_t slen = strnlen(s, len) + 1;
439 /* underlying fs providing us with an broken xattr list? */
440 if (WARN_ON(slen > len))
444 if (!ovl_can_list(dentry->d_sb, s)) {
446 memmove(s, s + slen, len);
455 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
457 struct inode *realinode = ovl_inode_real(inode);
458 const struct cred *old_cred;
459 struct posix_acl *acl;
461 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
464 old_cred = ovl_override_creds(inode->i_sb);
465 acl = get_acl(realinode, type);
466 revert_creds(old_cred);
471 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
473 if (flags & S_ATIME) {
474 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
475 struct path upperpath = {
476 .mnt = ovl_upper_mnt(ofs),
477 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
480 if (upperpath.dentry) {
481 touch_atime(&upperpath);
482 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
488 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
492 struct inode *realinode = ovl_inode_realdata(inode);
493 const struct cred *old_cred;
495 if (!realinode->i_op->fiemap)
498 old_cred = ovl_override_creds(inode->i_sb);
499 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
500 revert_creds(old_cred);
506 * Work around the fact that security_file_ioctl() takes a file argument.
507 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
510 static int ovl_security_fileattr(struct path *realpath, struct fileattr *fa,
517 file = dentry_open(realpath, O_RDONLY, current_cred());
519 return PTR_ERR(file);
522 cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
524 cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
526 err = security_file_ioctl(file, cmd, 0);
532 int ovl_real_fileattr_set(struct path *realpath, struct fileattr *fa)
536 err = ovl_security_fileattr(realpath, fa, true);
540 return vfs_fileattr_set(&init_user_ns, realpath->dentry, fa);
543 int ovl_fileattr_set(struct user_namespace *mnt_userns,
544 struct dentry *dentry, struct fileattr *fa)
546 struct inode *inode = d_inode(dentry);
547 struct path upperpath;
548 const struct cred *old_cred;
552 err = ovl_want_write(dentry);
556 err = ovl_copy_up(dentry);
558 ovl_path_real(dentry, &upperpath);
560 old_cred = ovl_override_creds(inode->i_sb);
562 * Store immutable/append-only flags in xattr and clear them
563 * in upper fileattr (in case they were set by older kernel)
564 * so children of "ovl-immutable" directories lower aliases of
565 * "ovl-immutable" hardlinks could be copied up.
566 * Clear xattr when flags are cleared.
568 err = ovl_set_protattr(inode, upperpath.dentry, fa);
570 err = ovl_real_fileattr_set(&upperpath, fa);
571 revert_creds(old_cred);
574 * Merge real inode flags with inode flags read from
575 * overlay.protattr xattr
577 flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK;
579 BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK);
580 flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK;
581 inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK);
583 ovl_drop_write(dentry);
588 /* Convert inode protection flags to fileattr flags */
589 static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa)
591 BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL);
592 BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
594 if (inode->i_flags & S_APPEND) {
595 fa->flags |= FS_APPEND_FL;
596 fa->fsx_xflags |= FS_XFLAG_APPEND;
598 if (inode->i_flags & S_IMMUTABLE) {
599 fa->flags |= FS_IMMUTABLE_FL;
600 fa->fsx_xflags |= FS_XFLAG_IMMUTABLE;
604 int ovl_real_fileattr_get(struct path *realpath, struct fileattr *fa)
608 err = ovl_security_fileattr(realpath, fa, false);
612 return vfs_fileattr_get(realpath->dentry, fa);
615 int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa)
617 struct inode *inode = d_inode(dentry);
618 struct path realpath;
619 const struct cred *old_cred;
622 ovl_path_real(dentry, &realpath);
624 old_cred = ovl_override_creds(inode->i_sb);
625 err = ovl_real_fileattr_get(&realpath, fa);
626 ovl_fileattr_prot_flags(inode, fa);
627 revert_creds(old_cred);
632 static const struct inode_operations ovl_file_inode_operations = {
633 .setattr = ovl_setattr,
634 .permission = ovl_permission,
635 .getattr = ovl_getattr,
636 .listxattr = ovl_listxattr,
637 .get_acl = ovl_get_acl,
638 .update_time = ovl_update_time,
639 .fiemap = ovl_fiemap,
640 .fileattr_get = ovl_fileattr_get,
641 .fileattr_set = ovl_fileattr_set,
644 static const struct inode_operations ovl_symlink_inode_operations = {
645 .setattr = ovl_setattr,
646 .get_link = ovl_get_link,
647 .getattr = ovl_getattr,
648 .listxattr = ovl_listxattr,
649 .update_time = ovl_update_time,
652 static const struct inode_operations ovl_special_inode_operations = {
653 .setattr = ovl_setattr,
654 .permission = ovl_permission,
655 .getattr = ovl_getattr,
656 .listxattr = ovl_listxattr,
657 .get_acl = ovl_get_acl,
658 .update_time = ovl_update_time,
661 static const struct address_space_operations ovl_aops = {
662 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
663 .direct_IO = noop_direct_IO,
667 * It is possible to stack overlayfs instance on top of another
668 * overlayfs instance as lower layer. We need to annotate the
669 * stackable i_mutex locks according to stack level of the super
670 * block instance. An overlayfs instance can never be in stack
671 * depth 0 (there is always a real fs below it). An overlayfs
672 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
674 * For example, here is a snip from /proc/lockdep_chains after
675 * dir_iterate of nested overlayfs:
677 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
678 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
679 * [...] &type->i_mutex_dir_key (stack_depth=0)
681 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
683 * This chain is valid:
684 * - inode->i_rwsem (inode_lock[2])
685 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
686 * - OVL_I(inode)->lock (ovl_inode_lock[2])
687 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
689 * And this chain is valid:
690 * - inode->i_rwsem (inode_lock[2])
691 * - OVL_I(inode)->lock (ovl_inode_lock[2])
692 * - lowerinode->i_rwsem (inode_lock[1])
693 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
695 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
696 * held, because it is in reverse order of the non-nested case using the same
698 * - inode->i_rwsem (inode_lock[1])
699 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
700 * - OVL_I(inode)->lock (ovl_inode_lock[1])
702 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
704 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
706 #ifdef CONFIG_LOCKDEP
707 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
708 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
709 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
711 int depth = inode->i_sb->s_stack_depth - 1;
713 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
716 if (S_ISDIR(inode->i_mode))
717 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
719 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
721 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
725 static void ovl_next_ino(struct inode *inode)
727 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
729 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
730 if (unlikely(!inode->i_ino))
731 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
734 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
736 int xinobits = ovl_xino_bits(inode->i_sb);
737 unsigned int xinoshift = 64 - xinobits;
740 * When d_ino is consistent with st_ino (samefs or i_ino has enough
741 * bits to encode layer), set the same value used for st_ino to i_ino,
742 * so inode number exposed via /proc/locks and a like will be
743 * consistent with d_ino and st_ino values. An i_ino value inconsistent
744 * with d_ino also causes nfsd readdirplus to fail.
747 if (ovl_same_fs(inode->i_sb)) {
749 } else if (xinobits && likely(!(ino >> xinoshift))) {
750 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
755 * For directory inodes on non-samefs with xino disabled or xino
756 * overflow, we allocate a non-persistent inode number, to be used for
757 * resolving st_ino collisions in ovl_map_dev_ino().
759 * To avoid ino collision with legitimate xino values from upper
760 * layer (fsid 0), use the lowest xinobit to map the non
761 * persistent inode numbers to the unified st_ino address space.
763 if (S_ISDIR(inode->i_mode)) {
766 inode->i_ino &= ~0UL >> xinobits;
767 inode->i_ino |= 1UL << xinoshift;
772 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
773 unsigned long ino, int fsid)
775 struct inode *realinode;
777 if (oip->upperdentry)
778 OVL_I(inode)->__upperdentry = oip->upperdentry;
779 if (oip->lowerpath && oip->lowerpath->dentry)
780 OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
782 OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
784 realinode = ovl_inode_real(inode);
785 ovl_copyattr(realinode, inode);
786 ovl_copyflags(realinode, inode);
787 ovl_map_ino(inode, ino, fsid);
790 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
792 inode->i_mode = mode;
793 inode->i_flags |= S_NOCMTIME;
794 #ifdef CONFIG_FS_POSIX_ACL
795 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
798 ovl_lockdep_annotate_inode_mutex_key(inode);
800 switch (mode & S_IFMT) {
802 inode->i_op = &ovl_file_inode_operations;
803 inode->i_fop = &ovl_file_operations;
804 inode->i_mapping->a_ops = &ovl_aops;
808 inode->i_op = &ovl_dir_inode_operations;
809 inode->i_fop = &ovl_dir_operations;
813 inode->i_op = &ovl_symlink_inode_operations;
817 inode->i_op = &ovl_special_inode_operations;
818 init_special_inode(inode, mode, rdev);
824 * With inodes index enabled, an overlay inode nlink counts the union of upper
825 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
826 * upper inode, the following nlink modifying operations can happen:
828 * 1. Lower hardlink copy up
829 * 2. Upper hardlink created, unlinked or renamed over
830 * 3. Lower hardlink whiteout or renamed over
832 * For the first, copy up case, the union nlink does not change, whether the
833 * operation succeeds or fails, but the upper inode nlink may change.
834 * Therefore, before copy up, we store the union nlink value relative to the
835 * lower inode nlink in the index inode xattr .overlay.nlink.
837 * For the second, upper hardlink case, the union nlink should be incremented
838 * or decremented IFF the operation succeeds, aligned with nlink change of the
839 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
840 * value relative to the upper inode nlink in the index inode.
842 * For the last, lower cover up case, we simplify things by preceding the
843 * whiteout or cover up with copy up. This makes sure that there is an index
844 * upper inode where the nlink xattr can be stored before the copied up upper
847 #define OVL_NLINK_ADD_UPPER (1 << 0)
850 * On-disk format for indexed nlink:
852 * nlink relative to the upper inode - "U[+-]NUM"
853 * nlink relative to the lower inode - "L[+-]NUM"
856 static int ovl_set_nlink_common(struct dentry *dentry,
857 struct dentry *realdentry, const char *format)
859 struct inode *inode = d_inode(dentry);
860 struct inode *realinode = d_inode(realdentry);
864 len = snprintf(buf, sizeof(buf), format,
865 (int) (inode->i_nlink - realinode->i_nlink));
867 if (WARN_ON(len >= sizeof(buf)))
870 return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
871 OVL_XATTR_NLINK, buf, len);
874 int ovl_set_nlink_upper(struct dentry *dentry)
876 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
879 int ovl_set_nlink_lower(struct dentry *dentry)
881 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
884 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
885 struct dentry *upperdentry,
886 unsigned int fallback)
893 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
896 err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
897 &buf, sizeof(buf) - 1);
902 if ((buf[0] != 'L' && buf[0] != 'U') ||
903 (buf[1] != '+' && buf[1] != '-'))
906 err = kstrtoint(buf + 1, 10, &nlink_diff);
910 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
919 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
924 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
928 inode = new_inode(sb);
930 ovl_fill_inode(inode, mode, rdev);
935 static int ovl_inode_test(struct inode *inode, void *data)
937 return inode->i_private == data;
940 static int ovl_inode_set(struct inode *inode, void *data)
942 inode->i_private = data;
946 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
947 struct dentry *upperdentry, bool strict)
950 * For directories, @strict verify from lookup path performs consistency
951 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
952 * inode. Non @strict verify from NFS handle decode path passes NULL for
953 * 'unknown' lower/upper.
955 if (S_ISDIR(inode->i_mode) && strict) {
956 /* Real lower dir moved to upper layer under us? */
957 if (!lowerdentry && ovl_inode_lower(inode))
960 /* Lookup of an uncovered redirect origin? */
961 if (!upperdentry && ovl_inode_upper(inode))
966 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
967 * This happens when finding a copied up overlay inode for a renamed
968 * or hardlinked overlay dentry and lower dentry cannot be followed
969 * by origin because lower fs does not support file handles.
971 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
975 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
976 * This happens when finding a lower alias for a copied up hard link.
978 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
984 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
987 struct inode *inode, *key = d_inode(real);
989 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
993 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
994 is_upper ? real : NULL, false)) {
996 return ERR_PTR(-ESTALE);
1002 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
1004 struct inode *key = d_inode(dir);
1008 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1012 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
1013 !ovl_inode_lower(trap);
1020 * Create an inode cache entry for layer root dir, that will intentionally
1021 * fail ovl_verify_inode(), so any lookup that will find some layer root
1024 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
1026 struct inode *key = d_inode(dir);
1030 return ERR_PTR(-ENOTDIR);
1032 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
1033 ovl_inode_set, key);
1035 return ERR_PTR(-ENOMEM);
1037 if (!(trap->i_state & I_NEW)) {
1038 /* Conflicting layer roots? */
1040 return ERR_PTR(-ELOOP);
1043 trap->i_mode = S_IFDIR;
1044 trap->i_flags = S_DEAD;
1045 unlock_new_inode(trap);
1051 * Does overlay inode need to be hashed by lower inode?
1053 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
1054 struct dentry *lower, bool index)
1056 struct ovl_fs *ofs = sb->s_fs_info;
1058 /* No, if pure upper */
1062 /* Yes, if already indexed */
1066 /* Yes, if won't be copied up */
1067 if (!ovl_upper_mnt(ofs))
1070 /* No, if lower hardlink is or will be broken on copy up */
1071 if ((upper || !ovl_indexdir(sb)) &&
1072 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1075 /* No, if non-indexed upper with NFS export */
1076 if (sb->s_export_op && upper)
1079 /* Otherwise, hash by lower inode for fsnotify */
1083 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1086 return newinode ? inode_insert5(newinode, (unsigned long) key,
1087 ovl_inode_test, ovl_inode_set, key) :
1088 iget5_locked(sb, (unsigned long) key,
1089 ovl_inode_test, ovl_inode_set, key);
1092 struct inode *ovl_get_inode(struct super_block *sb,
1093 struct ovl_inode_params *oip)
1095 struct ovl_fs *ofs = OVL_FS(sb);
1096 struct dentry *upperdentry = oip->upperdentry;
1097 struct ovl_path *lowerpath = oip->lowerpath;
1098 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1099 struct inode *inode;
1100 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1101 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1103 int fsid = bylower ? lowerpath->layer->fsid : 0;
1105 unsigned long ino = 0;
1106 int err = oip->newinode ? -EEXIST : -ENOMEM;
1109 realinode = d_inode(lowerdentry);
1112 * Copy up origin (lower) may exist for non-indexed upper, but we must
1113 * not use lower as hash key if this is a broken hardlink.
1115 is_dir = S_ISDIR(realinode->i_mode);
1116 if (upperdentry || bylower) {
1117 struct inode *key = d_inode(bylower ? lowerdentry :
1119 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1121 inode = ovl_iget5(sb, oip->newinode, key);
1124 if (!(inode->i_state & I_NEW)) {
1126 * Verify that the underlying files stored in the inode
1127 * match those in the dentry.
1129 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1137 kfree(oip->redirect);
1141 /* Recalculate nlink for non-dir due to indexing */
1143 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1145 set_nlink(inode, nlink);
1148 /* Lower hardlink that will be broken on copy up */
1149 inode = new_inode(sb);
1154 ino = realinode->i_ino;
1155 fsid = lowerpath->layer->fsid;
1157 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1158 ovl_inode_init(inode, oip, ino, fsid);
1160 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1161 ovl_set_flag(OVL_IMPURE, inode);
1164 ovl_set_flag(OVL_INDEX, inode);
1166 OVL_I(inode)->redirect = oip->redirect;
1169 ovl_set_flag(OVL_CONST_INO, inode);
1171 /* Check for non-merge dir that may have whiteouts */
1173 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1174 ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
1175 ovl_set_flag(OVL_WHITEOUTS, inode);
1179 /* Check for immutable/append-only inode flags in xattr */
1181 ovl_check_protattr(inode, upperdentry);
1183 if (inode->i_state & I_NEW)
1184 unlock_new_inode(inode);
1189 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1190 inode = ERR_PTR(err);