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 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
25 bool full_copy_up = false;
26 struct dentry *upperdentry;
27 const struct cred *old_cred;
29 err = setattr_prepare(&init_user_ns, dentry, attr);
33 err = ovl_want_write(dentry);
37 if (attr->ia_valid & ATTR_SIZE) {
38 /* Truncate should trigger data copy up as well */
43 err = ovl_copy_up(dentry);
45 err = ovl_copy_up_with_data(dentry);
47 struct inode *winode = NULL;
49 upperdentry = ovl_dentry_upper(dentry);
51 if (attr->ia_valid & ATTR_SIZE) {
52 winode = d_inode(upperdentry);
53 err = get_write_access(winode);
58 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
59 attr->ia_valid &= ~ATTR_MODE;
62 * We might have to translate ovl file into real file object
63 * once use cases emerge. For now, simply don't let underlying
64 * filesystem rely on attr->ia_file
66 attr->ia_valid &= ~ATTR_FILE;
69 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
70 * set. Overlayfs does not pass O_TRUNC flag to underlying
71 * filesystem during open -> do not pass ATTR_OPEN. This
72 * disables optimization in fuse which assumes open(O_TRUNC)
73 * already set file size to 0. But we never passed O_TRUNC to
74 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
75 * setattr request to server.
77 attr->ia_valid &= ~ATTR_OPEN;
79 inode_lock(upperdentry->d_inode);
80 old_cred = ovl_override_creds(dentry->d_sb);
81 err = ovl_do_notify_change(ofs, upperdentry, attr);
82 revert_creds(old_cred);
84 ovl_copyattr(dentry->d_inode);
85 inode_unlock(upperdentry->d_inode);
88 put_write_access(winode);
91 ovl_drop_write(dentry);
96 static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
98 bool samefs = ovl_same_fs(dentry->d_sb);
99 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
100 unsigned int xinoshift = 64 - xinobits;
104 * When all layers are on the same fs, all real inode
105 * number are unique, so we use the overlay st_dev,
106 * which is friendly to du -x.
108 stat->dev = dentry->d_sb->s_dev;
110 } else if (xinobits) {
112 * All inode numbers of underlying fs should not be using the
113 * high xinobits, so we use high xinobits to partition the
114 * overlay st_ino address space. The high bits holds the fsid
115 * (upper fsid is 0). The lowest xinobit is reserved for mapping
116 * the non-persistent inode numbers range in case of overflow.
117 * This way all overlay inode numbers are unique and use the
120 if (likely(!(stat->ino >> xinoshift))) {
121 stat->ino |= ((u64)fsid) << (xinoshift + 1);
122 stat->dev = dentry->d_sb->s_dev;
124 } else if (ovl_xino_warn(dentry->d_sb)) {
125 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
126 dentry, stat->ino, xinobits);
130 /* The inode could not be mapped to a unified st_ino address space */
131 if (S_ISDIR(dentry->d_inode->i_mode)) {
133 * Always use the overlay st_dev for directories, so 'find
134 * -xdev' will scan the entire overlay mount and won't cross the
135 * overlay mount boundaries.
137 * If not all layers are on the same fs the pair {real st_ino;
138 * overlay st_dev} is not unique, so use the non persistent
139 * overlay st_ino for directories.
141 stat->dev = dentry->d_sb->s_dev;
142 stat->ino = dentry->d_inode->i_ino;
145 * For non-samefs setup, if we cannot map all layers st_ino
146 * to a unified address space, we need to make sure that st_dev
147 * is unique per underlying fs, so we use the unique anonymous
148 * bdev assigned to the underlying fs.
150 stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
154 int ovl_getattr(struct user_namespace *mnt_userns, const struct path *path,
155 struct kstat *stat, u32 request_mask, unsigned int flags)
157 struct dentry *dentry = path->dentry;
158 enum ovl_path_type type;
159 struct path realpath;
160 const struct cred *old_cred;
161 struct inode *inode = d_inode(dentry);
162 bool is_dir = S_ISDIR(inode->i_mode);
165 bool metacopy_blocks = false;
167 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
169 type = ovl_path_real(dentry, &realpath);
170 old_cred = ovl_override_creds(dentry->d_sb);
171 err = vfs_getattr(&realpath, stat, request_mask, flags);
175 /* Report the effective immutable/append-only STATX flags */
176 generic_fill_statx_attr(inode, stat);
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;
284 struct path realpath;
285 const struct cred *old_cred;
288 /* Careful in RCU walk mode */
289 ovl_i_path_real(inode, &realpath);
290 if (!realpath.dentry) {
291 WARN_ON(!(mask & MAY_NOT_BLOCK));
296 * Check overlay inode with the creds of task and underlying inode
297 * with creds of mounter
299 err = generic_permission(&init_user_ns, inode, mask);
303 realinode = d_inode(realpath.dentry);
304 old_cred = ovl_override_creds(inode->i_sb);
306 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
307 mask &= ~(MAY_WRITE | MAY_APPEND);
308 /* Make sure mounter can read file for copy up later */
311 err = inode_permission(mnt_user_ns(realpath.mnt), realinode, mask);
312 revert_creds(old_cred);
317 static const char *ovl_get_link(struct dentry *dentry,
319 struct delayed_call *done)
321 const struct cred *old_cred;
325 return ERR_PTR(-ECHILD);
327 old_cred = ovl_override_creds(dentry->d_sb);
328 p = vfs_get_link(ovl_dentry_real(dentry), done);
329 revert_creds(old_cred);
333 bool ovl_is_private_xattr(struct super_block *sb, const char *name)
335 struct ovl_fs *ofs = sb->s_fs_info;
337 if (ofs->config.userxattr)
338 return strncmp(name, OVL_XATTR_USER_PREFIX,
339 sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
341 return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
342 sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
345 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
346 const void *value, size_t size, int flags)
349 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
350 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
351 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
352 struct path realpath;
353 const struct cred *old_cred;
355 err = ovl_want_write(dentry);
359 if (!value && !upperdentry) {
360 ovl_path_lower(dentry, &realpath);
361 old_cred = ovl_override_creds(dentry->d_sb);
362 err = vfs_getxattr(mnt_user_ns(realpath.mnt), realdentry, name, NULL, 0);
363 revert_creds(old_cred);
369 err = ovl_copy_up(dentry);
373 realdentry = ovl_dentry_upper(dentry);
376 old_cred = ovl_override_creds(dentry->d_sb);
378 err = ovl_do_setxattr(ofs, realdentry, name, value, size,
381 WARN_ON(flags != XATTR_REPLACE);
382 err = ovl_do_removexattr(ofs, realdentry, name);
384 revert_creds(old_cred);
390 ovl_drop_write(dentry);
395 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
396 void *value, size_t size)
399 const struct cred *old_cred;
400 struct path realpath;
402 ovl_i_path_real(inode, &realpath);
403 old_cred = ovl_override_creds(dentry->d_sb);
404 res = vfs_getxattr(mnt_user_ns(realpath.mnt), realpath.dentry, name, value, size);
405 revert_creds(old_cred);
409 static bool ovl_can_list(struct super_block *sb, const char *s)
411 /* Never list private (.overlay) */
412 if (ovl_is_private_xattr(sb, s))
415 /* List all non-trusted xattrs */
416 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
419 /* list other trusted for superuser only */
420 return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
423 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
425 struct dentry *realdentry = ovl_dentry_real(dentry);
429 const struct cred *old_cred;
431 old_cred = ovl_override_creds(dentry->d_sb);
432 res = vfs_listxattr(realdentry, list, size);
433 revert_creds(old_cred);
434 if (res <= 0 || size == 0)
437 /* filter out private xattrs */
438 for (s = list, len = res; len;) {
439 size_t slen = strnlen(s, len) + 1;
441 /* underlying fs providing us with an broken xattr list? */
442 if (WARN_ON(slen > len))
446 if (!ovl_can_list(dentry->d_sb, s)) {
448 memmove(s, s + slen, len);
457 #ifdef CONFIG_FS_POSIX_ACL
459 * Apply the idmapping of the layer to POSIX ACLs. The caller must pass a clone
460 * of the POSIX ACLs retrieved from the lower layer to this function to not
461 * alter the POSIX ACLs for the underlying filesystem.
463 static void ovl_idmap_posix_acl(struct inode *realinode,
464 struct user_namespace *mnt_userns,
465 struct posix_acl *acl)
467 struct user_namespace *fs_userns = i_user_ns(realinode);
469 for (unsigned int i = 0; i < acl->a_count; i++) {
473 struct posix_acl_entry *e = &acl->a_entries[i];
476 vfsuid = make_vfsuid(mnt_userns, fs_userns, e->e_uid);
477 e->e_uid = vfsuid_into_kuid(vfsuid);
480 vfsgid = make_vfsgid(mnt_userns, fs_userns, e->e_gid);
481 e->e_gid = vfsgid_into_kgid(vfsgid);
488 * When the relevant layer is an idmapped mount we need to take the idmapping
489 * of the layer into account and translate any ACL_{GROUP,USER} values
490 * according to the idmapped mount.
492 * We cannot alter the ACLs returned from the relevant layer as that would
493 * alter the cached values filesystem wide for the lower filesystem. Instead we
494 * can clone the ACLs and then apply the relevant idmapping of the layer.
496 * This is obviously only relevant when idmapped layers are used.
498 struct posix_acl *ovl_get_acl(struct inode *inode, int type, bool rcu)
500 struct inode *realinode = ovl_inode_real(inode);
501 struct posix_acl *acl, *clone;
502 struct path realpath;
504 if (!IS_POSIXACL(realinode))
507 /* Careful in RCU walk mode */
508 ovl_i_path_real(inode, &realpath);
509 if (!realpath.dentry) {
511 return ERR_PTR(-ECHILD);
515 acl = get_cached_acl_rcu(realinode, type);
517 const struct cred *old_cred;
519 old_cred = ovl_override_creds(inode->i_sb);
520 acl = get_acl(realinode, type);
521 revert_creds(old_cred);
524 * If there are no POSIX ACLs, or we encountered an error,
525 * or the layer isn't idmapped we don't need to do anything.
527 if (!is_idmapped_mnt(realpath.mnt) || IS_ERR_OR_NULL(acl))
531 * We only get here if the layer is idmapped. So drop out of RCU path
532 * walk so we can clone the ACLs. There's no need to release the ACLs
533 * since get_cached_acl_rcu() doesn't take a reference on the ACLs.
536 return ERR_PTR(-ECHILD);
538 clone = posix_acl_clone(acl, GFP_KERNEL);
540 clone = ERR_PTR(-ENOMEM);
542 ovl_idmap_posix_acl(realinode, mnt_user_ns(realpath.mnt), clone);
544 * Since we're not in RCU path walk we always need to release the
547 posix_acl_release(acl);
552 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
554 if (flags & S_ATIME) {
555 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
556 struct path upperpath = {
557 .mnt = ovl_upper_mnt(ofs),
558 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
561 if (upperpath.dentry) {
562 touch_atime(&upperpath);
563 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
569 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
573 struct inode *realinode = ovl_inode_realdata(inode);
574 const struct cred *old_cred;
576 if (!realinode->i_op->fiemap)
579 old_cred = ovl_override_creds(inode->i_sb);
580 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
581 revert_creds(old_cred);
587 * Work around the fact that security_file_ioctl() takes a file argument.
588 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
591 static int ovl_security_fileattr(const struct path *realpath, struct fileattr *fa,
598 file = dentry_open(realpath, O_RDONLY, current_cred());
600 return PTR_ERR(file);
603 cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
605 cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
607 err = security_file_ioctl(file, cmd, 0);
613 int ovl_real_fileattr_set(const struct path *realpath, struct fileattr *fa)
617 err = ovl_security_fileattr(realpath, fa, true);
621 return vfs_fileattr_set(mnt_user_ns(realpath->mnt), realpath->dentry, fa);
624 int ovl_fileattr_set(struct user_namespace *mnt_userns,
625 struct dentry *dentry, struct fileattr *fa)
627 struct inode *inode = d_inode(dentry);
628 struct path upperpath;
629 const struct cred *old_cred;
633 err = ovl_want_write(dentry);
637 err = ovl_copy_up(dentry);
639 ovl_path_real(dentry, &upperpath);
641 old_cred = ovl_override_creds(inode->i_sb);
643 * Store immutable/append-only flags in xattr and clear them
644 * in upper fileattr (in case they were set by older kernel)
645 * so children of "ovl-immutable" directories lower aliases of
646 * "ovl-immutable" hardlinks could be copied up.
647 * Clear xattr when flags are cleared.
649 err = ovl_set_protattr(inode, upperpath.dentry, fa);
651 err = ovl_real_fileattr_set(&upperpath, fa);
652 revert_creds(old_cred);
655 * Merge real inode flags with inode flags read from
656 * overlay.protattr xattr
658 flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK;
660 BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK);
661 flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK;
662 inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK);
667 ovl_drop_write(dentry);
672 /* Convert inode protection flags to fileattr flags */
673 static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa)
675 BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL);
676 BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
678 if (inode->i_flags & S_APPEND) {
679 fa->flags |= FS_APPEND_FL;
680 fa->fsx_xflags |= FS_XFLAG_APPEND;
682 if (inode->i_flags & S_IMMUTABLE) {
683 fa->flags |= FS_IMMUTABLE_FL;
684 fa->fsx_xflags |= FS_XFLAG_IMMUTABLE;
688 int ovl_real_fileattr_get(const struct path *realpath, struct fileattr *fa)
692 err = ovl_security_fileattr(realpath, fa, false);
696 err = vfs_fileattr_get(realpath->dentry, fa);
697 if (err == -ENOIOCTLCMD)
702 int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa)
704 struct inode *inode = d_inode(dentry);
705 struct path realpath;
706 const struct cred *old_cred;
709 ovl_path_real(dentry, &realpath);
711 old_cred = ovl_override_creds(inode->i_sb);
712 err = ovl_real_fileattr_get(&realpath, fa);
713 ovl_fileattr_prot_flags(inode, fa);
714 revert_creds(old_cred);
719 static const struct inode_operations ovl_file_inode_operations = {
720 .setattr = ovl_setattr,
721 .permission = ovl_permission,
722 .getattr = ovl_getattr,
723 .listxattr = ovl_listxattr,
724 .get_acl = ovl_get_acl,
725 .update_time = ovl_update_time,
726 .fiemap = ovl_fiemap,
727 .fileattr_get = ovl_fileattr_get,
728 .fileattr_set = ovl_fileattr_set,
731 static const struct inode_operations ovl_symlink_inode_operations = {
732 .setattr = ovl_setattr,
733 .get_link = ovl_get_link,
734 .getattr = ovl_getattr,
735 .listxattr = ovl_listxattr,
736 .update_time = ovl_update_time,
739 static const struct inode_operations ovl_special_inode_operations = {
740 .setattr = ovl_setattr,
741 .permission = ovl_permission,
742 .getattr = ovl_getattr,
743 .listxattr = ovl_listxattr,
744 .get_acl = ovl_get_acl,
745 .update_time = ovl_update_time,
748 static const struct address_space_operations ovl_aops = {
749 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
750 .direct_IO = noop_direct_IO,
754 * It is possible to stack overlayfs instance on top of another
755 * overlayfs instance as lower layer. We need to annotate the
756 * stackable i_mutex locks according to stack level of the super
757 * block instance. An overlayfs instance can never be in stack
758 * depth 0 (there is always a real fs below it). An overlayfs
759 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
761 * For example, here is a snip from /proc/lockdep_chains after
762 * dir_iterate of nested overlayfs:
764 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
765 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
766 * [...] &type->i_mutex_dir_key (stack_depth=0)
768 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
770 * This chain is valid:
771 * - inode->i_rwsem (inode_lock[2])
772 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
773 * - OVL_I(inode)->lock (ovl_inode_lock[2])
774 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
776 * And this chain is valid:
777 * - inode->i_rwsem (inode_lock[2])
778 * - OVL_I(inode)->lock (ovl_inode_lock[2])
779 * - lowerinode->i_rwsem (inode_lock[1])
780 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
782 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
783 * held, because it is in reverse order of the non-nested case using the same
785 * - inode->i_rwsem (inode_lock[1])
786 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
787 * - OVL_I(inode)->lock (ovl_inode_lock[1])
789 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
791 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
793 #ifdef CONFIG_LOCKDEP
794 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
795 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
796 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
798 int depth = inode->i_sb->s_stack_depth - 1;
800 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
803 if (S_ISDIR(inode->i_mode))
804 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
806 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
808 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
812 static void ovl_next_ino(struct inode *inode)
814 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
816 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
817 if (unlikely(!inode->i_ino))
818 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
821 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
823 int xinobits = ovl_xino_bits(inode->i_sb);
824 unsigned int xinoshift = 64 - xinobits;
827 * When d_ino is consistent with st_ino (samefs or i_ino has enough
828 * bits to encode layer), set the same value used for st_ino to i_ino,
829 * so inode number exposed via /proc/locks and a like will be
830 * consistent with d_ino and st_ino values. An i_ino value inconsistent
831 * with d_ino also causes nfsd readdirplus to fail.
834 if (ovl_same_fs(inode->i_sb)) {
836 } else if (xinobits && likely(!(ino >> xinoshift))) {
837 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
842 * For directory inodes on non-samefs with xino disabled or xino
843 * overflow, we allocate a non-persistent inode number, to be used for
844 * resolving st_ino collisions in ovl_map_dev_ino().
846 * To avoid ino collision with legitimate xino values from upper
847 * layer (fsid 0), use the lowest xinobit to map the non
848 * persistent inode numbers to the unified st_ino address space.
850 if (S_ISDIR(inode->i_mode)) {
853 inode->i_ino &= ~0UL >> xinobits;
854 inode->i_ino |= 1UL << xinoshift;
859 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
860 unsigned long ino, int fsid)
862 struct inode *realinode;
863 struct ovl_inode *oi = OVL_I(inode);
865 if (oip->upperdentry)
866 oi->__upperdentry = oip->upperdentry;
867 if (oip->lowerpath && oip->lowerpath->dentry) {
868 oi->lowerpath.dentry = dget(oip->lowerpath->dentry);
869 oi->lowerpath.layer = oip->lowerpath->layer;
872 oi->lowerdata = igrab(d_inode(oip->lowerdata));
874 realinode = ovl_inode_real(inode);
876 ovl_copyflags(realinode, inode);
877 ovl_map_ino(inode, ino, fsid);
880 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
882 inode->i_mode = mode;
883 inode->i_flags |= S_NOCMTIME;
884 #ifdef CONFIG_FS_POSIX_ACL
885 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
888 ovl_lockdep_annotate_inode_mutex_key(inode);
890 switch (mode & S_IFMT) {
892 inode->i_op = &ovl_file_inode_operations;
893 inode->i_fop = &ovl_file_operations;
894 inode->i_mapping->a_ops = &ovl_aops;
898 inode->i_op = &ovl_dir_inode_operations;
899 inode->i_fop = &ovl_dir_operations;
903 inode->i_op = &ovl_symlink_inode_operations;
907 inode->i_op = &ovl_special_inode_operations;
908 init_special_inode(inode, mode, rdev);
914 * With inodes index enabled, an overlay inode nlink counts the union of upper
915 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
916 * upper inode, the following nlink modifying operations can happen:
918 * 1. Lower hardlink copy up
919 * 2. Upper hardlink created, unlinked or renamed over
920 * 3. Lower hardlink whiteout or renamed over
922 * For the first, copy up case, the union nlink does not change, whether the
923 * operation succeeds or fails, but the upper inode nlink may change.
924 * Therefore, before copy up, we store the union nlink value relative to the
925 * lower inode nlink in the index inode xattr .overlay.nlink.
927 * For the second, upper hardlink case, the union nlink should be incremented
928 * or decremented IFF the operation succeeds, aligned with nlink change of the
929 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
930 * value relative to the upper inode nlink in the index inode.
932 * For the last, lower cover up case, we simplify things by preceding the
933 * whiteout or cover up with copy up. This makes sure that there is an index
934 * upper inode where the nlink xattr can be stored before the copied up upper
937 #define OVL_NLINK_ADD_UPPER (1 << 0)
940 * On-disk format for indexed nlink:
942 * nlink relative to the upper inode - "U[+-]NUM"
943 * nlink relative to the lower inode - "L[+-]NUM"
946 static int ovl_set_nlink_common(struct dentry *dentry,
947 struct dentry *realdentry, const char *format)
949 struct inode *inode = d_inode(dentry);
950 struct inode *realinode = d_inode(realdentry);
954 len = snprintf(buf, sizeof(buf), format,
955 (int) (inode->i_nlink - realinode->i_nlink));
957 if (WARN_ON(len >= sizeof(buf)))
960 return ovl_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
961 OVL_XATTR_NLINK, buf, len);
964 int ovl_set_nlink_upper(struct dentry *dentry)
966 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
969 int ovl_set_nlink_lower(struct dentry *dentry)
971 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
974 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
975 struct dentry *upperdentry,
976 unsigned int fallback)
983 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
986 err = ovl_getxattr_upper(ofs, upperdentry, OVL_XATTR_NLINK,
987 &buf, sizeof(buf) - 1);
992 if ((buf[0] != 'L' && buf[0] != 'U') ||
993 (buf[1] != '+' && buf[1] != '-'))
996 err = kstrtoint(buf + 1, 10, &nlink_diff);
1000 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
1001 nlink += nlink_diff;
1009 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
1014 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
1016 struct inode *inode;
1018 inode = new_inode(sb);
1020 ovl_fill_inode(inode, mode, rdev);
1025 static int ovl_inode_test(struct inode *inode, void *data)
1027 return inode->i_private == data;
1030 static int ovl_inode_set(struct inode *inode, void *data)
1032 inode->i_private = data;
1036 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
1037 struct dentry *upperdentry, bool strict)
1040 * For directories, @strict verify from lookup path performs consistency
1041 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
1042 * inode. Non @strict verify from NFS handle decode path passes NULL for
1043 * 'unknown' lower/upper.
1045 if (S_ISDIR(inode->i_mode) && strict) {
1046 /* Real lower dir moved to upper layer under us? */
1047 if (!lowerdentry && ovl_inode_lower(inode))
1050 /* Lookup of an uncovered redirect origin? */
1051 if (!upperdentry && ovl_inode_upper(inode))
1056 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
1057 * This happens when finding a copied up overlay inode for a renamed
1058 * or hardlinked overlay dentry and lower dentry cannot be followed
1059 * by origin because lower fs does not support file handles.
1061 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
1065 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
1066 * This happens when finding a lower alias for a copied up hard link.
1068 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
1074 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
1077 struct inode *inode, *key = d_inode(real);
1079 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1083 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
1084 is_upper ? real : NULL, false)) {
1086 return ERR_PTR(-ESTALE);
1092 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
1094 struct inode *key = d_inode(dir);
1098 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1102 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
1103 !ovl_inode_lower(trap);
1110 * Create an inode cache entry for layer root dir, that will intentionally
1111 * fail ovl_verify_inode(), so any lookup that will find some layer root
1114 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
1116 struct inode *key = d_inode(dir);
1120 return ERR_PTR(-ENOTDIR);
1122 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
1123 ovl_inode_set, key);
1125 return ERR_PTR(-ENOMEM);
1127 if (!(trap->i_state & I_NEW)) {
1128 /* Conflicting layer roots? */
1130 return ERR_PTR(-ELOOP);
1133 trap->i_mode = S_IFDIR;
1134 trap->i_flags = S_DEAD;
1135 unlock_new_inode(trap);
1141 * Does overlay inode need to be hashed by lower inode?
1143 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
1144 struct dentry *lower, bool index)
1146 struct ovl_fs *ofs = sb->s_fs_info;
1148 /* No, if pure upper */
1152 /* Yes, if already indexed */
1156 /* Yes, if won't be copied up */
1157 if (!ovl_upper_mnt(ofs))
1160 /* No, if lower hardlink is or will be broken on copy up */
1161 if ((upper || !ovl_indexdir(sb)) &&
1162 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1165 /* No, if non-indexed upper with NFS export */
1166 if (sb->s_export_op && upper)
1169 /* Otherwise, hash by lower inode for fsnotify */
1173 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1176 return newinode ? inode_insert5(newinode, (unsigned long) key,
1177 ovl_inode_test, ovl_inode_set, key) :
1178 iget5_locked(sb, (unsigned long) key,
1179 ovl_inode_test, ovl_inode_set, key);
1182 struct inode *ovl_get_inode(struct super_block *sb,
1183 struct ovl_inode_params *oip)
1185 struct ovl_fs *ofs = OVL_FS(sb);
1186 struct dentry *upperdentry = oip->upperdentry;
1187 struct ovl_path *lowerpath = oip->lowerpath;
1188 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1189 struct inode *inode;
1190 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1191 struct path realpath = {
1192 .dentry = upperdentry ?: lowerdentry,
1193 .mnt = upperdentry ? ovl_upper_mnt(ofs) : lowerpath->layer->mnt,
1195 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1197 int fsid = bylower ? lowerpath->layer->fsid : 0;
1199 unsigned long ino = 0;
1200 int err = oip->newinode ? -EEXIST : -ENOMEM;
1203 realinode = d_inode(lowerdentry);
1206 * Copy up origin (lower) may exist for non-indexed upper, but we must
1207 * not use lower as hash key if this is a broken hardlink.
1209 is_dir = S_ISDIR(realinode->i_mode);
1210 if (upperdentry || bylower) {
1211 struct inode *key = d_inode(bylower ? lowerdentry :
1213 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1215 inode = ovl_iget5(sb, oip->newinode, key);
1218 if (!(inode->i_state & I_NEW)) {
1220 * Verify that the underlying files stored in the inode
1221 * match those in the dentry.
1223 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1231 kfree(oip->redirect);
1235 /* Recalculate nlink for non-dir due to indexing */
1237 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1239 set_nlink(inode, nlink);
1242 /* Lower hardlink that will be broken on copy up */
1243 inode = new_inode(sb);
1248 ino = realinode->i_ino;
1249 fsid = lowerpath->layer->fsid;
1251 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1252 ovl_inode_init(inode, oip, ino, fsid);
1254 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1255 ovl_set_flag(OVL_IMPURE, inode);
1258 ovl_set_flag(OVL_INDEX, inode);
1260 OVL_I(inode)->redirect = oip->redirect;
1263 ovl_set_flag(OVL_CONST_INO, inode);
1265 /* Check for non-merge dir that may have whiteouts */
1267 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1268 ovl_path_check_origin_xattr(ofs, &realpath)) {
1269 ovl_set_flag(OVL_WHITEOUTS, inode);
1273 /* Check for immutable/append-only inode flags in xattr */
1275 ovl_check_protattr(inode, upperdentry);
1277 if (inode->i_state & I_NEW)
1278 unlock_new_inode(inode);
1283 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1284 inode = ERR_PTR(err);