1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (C) 2011 Novell Inc.
7 #include <uapi/linux/magic.h>
9 #include <linux/namei.h>
10 #include <linux/xattr.h>
11 #include <linux/mount.h>
12 #include <linux/parser.h>
13 #include <linux/module.h>
14 #include <linux/statfs.h>
15 #include <linux/seq_file.h>
16 #include <linux/posix_acl_xattr.h>
17 #include <linux/exportfs.h>
18 #include <linux/file.h>
19 #include <linux/fs_context.h>
20 #include <linux/fs_parser.h>
21 #include "overlayfs.h"
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Overlay filesystem");
26 MODULE_LICENSE("GPL");
31 static struct dentry *ovl_d_real(struct dentry *dentry,
32 const struct inode *inode)
34 struct dentry *real = NULL, *lower;
37 /* It's an overlay file */
38 if (inode && d_inode(dentry) == inode)
41 if (!d_is_reg(dentry)) {
42 if (!inode || inode == d_inode(dentry))
47 real = ovl_dentry_upper(dentry);
48 if (real && (inode == d_inode(real)))
51 if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
55 * Best effort lazy lookup of lowerdata for !inode case to return
56 * the real lowerdata dentry. The only current caller of d_real() with
57 * NULL inode is d_real_inode() from trace_uprobe and this caller is
58 * likely going to be followed reading from the file, before placing
59 * uprobes on offset within the file, so lowerdata should be available
60 * when setting the uprobe.
62 err = ovl_verify_lowerdata(dentry);
65 lower = ovl_dentry_lowerdata(dentry);
70 /* Handle recursion */
71 real = d_real(real, inode);
73 if (!inode || inode == d_inode(real))
76 WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
77 __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
78 inode ? inode->i_ino : 0, real,
79 real && d_inode(real) ? d_inode(real)->i_ino : 0);
83 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
91 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
92 ret = d->d_op->d_weak_revalidate(d, flags);
93 } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
94 ret = d->d_op->d_revalidate(d, flags);
96 if (!(flags & LOOKUP_RCU))
104 static int ovl_dentry_revalidate_common(struct dentry *dentry,
105 unsigned int flags, bool weak)
107 struct ovl_entry *oe;
108 struct ovl_path *lowerstack;
109 struct inode *inode = d_inode_rcu(dentry);
110 struct dentry *upper;
114 /* Careful in RCU mode */
119 lowerstack = ovl_lowerstack(oe);
120 upper = ovl_i_dentry_upper(inode);
122 ret = ovl_revalidate_real(upper, flags, weak);
124 for (i = 0; ret > 0 && i < ovl_numlower(oe); i++)
125 ret = ovl_revalidate_real(lowerstack[i].dentry, flags, weak);
130 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
132 return ovl_dentry_revalidate_common(dentry, flags, false);
135 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
137 return ovl_dentry_revalidate_common(dentry, flags, true);
140 static const struct dentry_operations ovl_dentry_operations = {
141 .d_real = ovl_d_real,
142 .d_revalidate = ovl_dentry_revalidate,
143 .d_weak_revalidate = ovl_dentry_weak_revalidate,
146 static struct kmem_cache *ovl_inode_cachep;
148 static struct inode *ovl_alloc_inode(struct super_block *sb)
150 struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
159 oi->__upperdentry = NULL;
160 oi->lowerdata_redirect = NULL;
162 mutex_init(&oi->lock);
164 return &oi->vfs_inode;
167 static void ovl_free_inode(struct inode *inode)
169 struct ovl_inode *oi = OVL_I(inode);
173 mutex_destroy(&oi->lock);
174 kmem_cache_free(ovl_inode_cachep, oi);
177 static void ovl_destroy_inode(struct inode *inode)
179 struct ovl_inode *oi = OVL_I(inode);
181 dput(oi->__upperdentry);
182 ovl_stack_put(ovl_lowerstack(oi->oe), ovl_numlower(oi->oe));
183 if (S_ISDIR(inode->i_mode))
184 ovl_dir_cache_free(inode);
186 kfree(oi->lowerdata_redirect);
189 static void ovl_put_super(struct super_block *sb)
191 struct ovl_fs *ofs = OVL_FS(sb);
197 /* Sync real dirty inodes in upper filesystem (if it exists) */
198 static int ovl_sync_fs(struct super_block *sb, int wait)
200 struct ovl_fs *ofs = OVL_FS(sb);
201 struct super_block *upper_sb;
204 ret = ovl_sync_status(ofs);
206 * We have to always set the err, because the return value isn't
207 * checked in syncfs, and instead indirectly return an error via
208 * the sb's writeback errseq, which VFS inspects after this call.
211 errseq_set(&sb->s_wb_err, -EIO);
219 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
220 * All the super blocks will be iterated, including upper_sb.
222 * If this is a syncfs(2) call, then we do need to call
223 * sync_filesystem() on upper_sb, but enough if we do it when being
224 * called with wait == 1.
229 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
231 down_read(&upper_sb->s_umount);
232 ret = sync_filesystem(upper_sb);
233 up_read(&upper_sb->s_umount);
240 * @dentry: The dentry to query
241 * @buf: The struct kstatfs to fill in with stats
243 * Get the filesystem statistics. As writes always target the upper layer
244 * filesystem pass the statfs to the upper filesystem (if it exists)
246 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
248 struct super_block *sb = dentry->d_sb;
249 struct ovl_fs *ofs = OVL_FS(sb);
250 struct dentry *root_dentry = sb->s_root;
254 ovl_path_real(root_dentry, &path);
256 err = vfs_statfs(&path, buf);
258 buf->f_namelen = ofs->namelen;
259 buf->f_type = OVERLAYFS_SUPER_MAGIC;
260 if (ovl_has_fsid(ofs))
261 buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
267 static const struct super_operations ovl_super_operations = {
268 .alloc_inode = ovl_alloc_inode,
269 .free_inode = ovl_free_inode,
270 .destroy_inode = ovl_destroy_inode,
271 .drop_inode = generic_delete_inode,
272 .put_super = ovl_put_super,
273 .sync_fs = ovl_sync_fs,
274 .statfs = ovl_statfs,
275 .show_options = ovl_show_options,
278 #define OVL_WORKDIR_NAME "work"
279 #define OVL_INDEXDIR_NAME "index"
281 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
282 const char *name, bool persist)
284 struct inode *dir = ofs->workbasedir->d_inode;
285 struct vfsmount *mnt = ovl_upper_mnt(ofs);
288 bool retried = false;
290 inode_lock_nested(dir, I_MUTEX_PARENT);
292 work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
295 struct iattr attr = {
296 .ia_valid = ATTR_MODE,
297 .ia_mode = S_IFDIR | 0,
309 err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
311 if (err == -EINVAL) {
318 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
322 /* Weird filesystem returning with hashed negative (kernfs)? */
324 if (d_really_is_negative(work))
328 * Try to remove POSIX ACL xattrs from workdir. We are good if:
330 * a) success (there was a POSIX ACL xattr and was removed)
331 * b) -ENODATA (there was no POSIX ACL xattr)
332 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
334 * There are various other error values that could effectively
335 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
336 * if the xattr name is too long), but the set of filesystems
337 * allowed as upper are limited to "normal" ones, where checking
338 * for the above two errors is sufficient.
340 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);
341 if (err && err != -ENODATA && err != -EOPNOTSUPP)
344 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);
345 if (err && err != -ENODATA && err != -EOPNOTSUPP)
348 /* Clear any inherited mode bits */
349 inode_lock(work->d_inode);
350 err = ovl_do_notify_change(ofs, work, &attr);
351 inode_unlock(work->d_inode);
365 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
366 ofs->config.workdir, name, -err);
371 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
374 struct kstatfs statfs;
375 int err = vfs_statfs(path, &statfs);
378 pr_err("statfs failed on '%s'\n", name);
380 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
385 static int ovl_lower_dir(const char *name, struct path *path,
386 struct ovl_fs *ofs, int *stack_depth)
391 err = ovl_check_namelen(path, ofs, name);
395 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
398 * The inodes index feature and NFS export need to encode and decode
399 * file handles, so they require that all layers support them.
401 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
402 if ((ofs->config.nfs_export ||
403 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
404 ofs->config.index = false;
405 ofs->config.nfs_export = false;
406 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
409 ofs->nofh |= !fh_type;
411 * Decoding origin file handle is required for persistent st_ino.
412 * Without persistent st_ino, xino=auto falls back to xino=off.
414 if (ofs->config.xino == OVL_XINO_AUTO &&
415 ofs->config.upperdir && !fh_type) {
416 ofs->config.xino = OVL_XINO_OFF;
417 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
421 /* Check if lower fs has 32bit inode numbers */
422 if (fh_type != FILEID_INO32_GEN)
428 /* Workdir should not be subdir of upperdir and vice versa */
429 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
433 if (workdir != upperdir) {
434 ok = (lock_rename(workdir, upperdir) == NULL);
435 unlock_rename(workdir, upperdir);
440 static int ovl_own_xattr_get(const struct xattr_handler *handler,
441 struct dentry *dentry, struct inode *inode,
442 const char *name, void *buffer, size_t size)
447 static int ovl_own_xattr_set(const struct xattr_handler *handler,
448 struct mnt_idmap *idmap,
449 struct dentry *dentry, struct inode *inode,
450 const char *name, const void *value,
451 size_t size, int flags)
456 static int ovl_other_xattr_get(const struct xattr_handler *handler,
457 struct dentry *dentry, struct inode *inode,
458 const char *name, void *buffer, size_t size)
460 return ovl_xattr_get(dentry, inode, name, buffer, size);
463 static int ovl_other_xattr_set(const struct xattr_handler *handler,
464 struct mnt_idmap *idmap,
465 struct dentry *dentry, struct inode *inode,
466 const char *name, const void *value,
467 size_t size, int flags)
469 return ovl_xattr_set(dentry, inode, name, value, size, flags);
472 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
473 .prefix = OVL_XATTR_TRUSTED_PREFIX,
474 .get = ovl_own_xattr_get,
475 .set = ovl_own_xattr_set,
478 static const struct xattr_handler ovl_own_user_xattr_handler = {
479 .prefix = OVL_XATTR_USER_PREFIX,
480 .get = ovl_own_xattr_get,
481 .set = ovl_own_xattr_set,
484 static const struct xattr_handler ovl_other_xattr_handler = {
485 .prefix = "", /* catch all */
486 .get = ovl_other_xattr_get,
487 .set = ovl_other_xattr_set,
490 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
491 &ovl_own_trusted_xattr_handler,
492 &ovl_other_xattr_handler,
496 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
497 &ovl_own_user_xattr_handler,
498 &ovl_other_xattr_handler,
502 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
503 struct inode **ptrap, const char *name)
508 trap = ovl_get_trap_inode(sb, dir);
509 err = PTR_ERR_OR_ZERO(trap);
512 pr_err("conflicting %s path\n", name);
521 * Determine how we treat concurrent use of upperdir/workdir based on the
522 * index feature. This is papering over mount leaks of container runtimes,
523 * for example, an old overlay mount is leaked and now its upperdir is
524 * attempted to be used as a lower layer in a new overlay mount.
526 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
528 if (ofs->config.index) {
529 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
533 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
539 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
540 struct ovl_layer *upper_layer,
541 const struct path *upperpath)
543 struct vfsmount *upper_mnt;
546 /* Upperdir path should not be r/o */
547 if (__mnt_is_readonly(upperpath->mnt)) {
548 pr_err("upper fs is r/o, try multi-lower layers mount\n");
553 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
557 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
562 upper_mnt = clone_private_mount(upperpath);
563 err = PTR_ERR(upper_mnt);
564 if (IS_ERR(upper_mnt)) {
565 pr_err("failed to clone upperpath\n");
569 /* Don't inherit atime flags */
570 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
571 upper_layer->mnt = upper_mnt;
572 upper_layer->idx = 0;
573 upper_layer->fsid = 0;
576 * Inherit SB_NOSEC flag from upperdir.
578 * This optimization changes behavior when a security related attribute
579 * (suid/sgid/security.*) is changed on an underlying layer. This is
580 * okay because we don't yet have guarantees in that case, but it will
581 * need careful treatment once we want to honour changes to underlying
584 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
585 sb->s_flags |= SB_NOSEC;
587 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
588 ofs->upperdir_locked = true;
590 err = ovl_report_in_use(ofs, "upperdir");
601 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
602 * negative values if error is encountered.
604 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
606 struct dentry *workdir = ofs->workdir;
607 struct inode *dir = d_inode(workdir);
610 struct dentry *whiteout;
611 struct name_snapshot name;
614 inode_lock_nested(dir, I_MUTEX_PARENT);
616 temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
621 dest = ovl_lookup_temp(ofs, workdir);
628 /* Name is inline and stable - using snapshot as a copy helper */
629 take_dentry_name_snapshot(&name, temp);
630 err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
637 whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
638 err = PTR_ERR(whiteout);
639 if (IS_ERR(whiteout))
642 err = ovl_is_whiteout(whiteout);
644 /* Best effort cleanup of whiteout and temp file */
646 ovl_cleanup(ofs, dir, whiteout);
650 ovl_cleanup(ofs, dir, temp);
651 release_dentry_name_snapshot(&name);
661 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
662 struct dentry *parent,
663 const char *name, umode_t mode)
665 size_t len = strlen(name);
666 struct dentry *child;
668 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
669 child = ovl_lookup_upper(ofs, name, parent, len);
670 if (!IS_ERR(child) && !child->d_inode)
671 child = ovl_create_real(ofs, parent->d_inode, child,
673 inode_unlock(parent->d_inode);
680 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
683 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
686 struct dentry *d = dget(ofs->workbasedir);
687 static const char *const volatile_path[] = {
688 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
690 const char *const *name = volatile_path;
692 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
693 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
701 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
702 const struct path *workpath)
704 struct vfsmount *mnt = ovl_upper_mnt(ofs);
705 struct dentry *workdir;
706 struct file *tmpfile;
707 bool rename_whiteout;
712 err = mnt_want_write(mnt);
716 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
717 err = PTR_ERR(workdir);
718 if (IS_ERR_OR_NULL(workdir))
721 ofs->workdir = workdir;
723 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
728 * Upper should support d_type, else whiteouts are visible. Given
729 * workdir and upper are on same fs, we can do iterate_dir() on
730 * workdir. This check requires successful creation of workdir in
733 err = ovl_check_d_type_supported(workpath);
739 pr_warn("upper fs needs to support d_type.\n");
741 /* Check if upper/work fs supports O_TMPFILE */
742 tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
743 ofs->tmpfile = !IS_ERR(tmpfile);
747 pr_warn("upper fs does not support tmpfile.\n");
750 /* Check if upper/work fs supports RENAME_WHITEOUT */
751 err = ovl_check_rename_whiteout(ofs);
755 rename_whiteout = err;
756 if (!rename_whiteout)
757 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
760 * Check if upper/work fs supports (trusted|user).overlay.* xattr
762 err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
764 pr_warn("failed to set xattr on upper\n");
766 if (ovl_redirect_follow(ofs)) {
767 ofs->config.redirect_mode = OVL_REDIRECT_NOFOLLOW;
768 pr_warn("...falling back to redirect_dir=nofollow.\n");
770 if (ofs->config.metacopy) {
771 ofs->config.metacopy = false;
772 pr_warn("...falling back to metacopy=off.\n");
774 if (ofs->config.index) {
775 ofs->config.index = false;
776 pr_warn("...falling back to index=off.\n");
778 if (ovl_has_fsid(ofs)) {
779 ofs->config.uuid = OVL_UUID_NULL;
780 pr_warn("...falling back to uuid=null.\n");
783 * xattr support is required for persistent st_ino.
784 * Without persistent st_ino, xino=auto falls back to xino=off.
786 if (ofs->config.xino == OVL_XINO_AUTO) {
787 ofs->config.xino = OVL_XINO_OFF;
788 pr_warn("...falling back to xino=off.\n");
790 if (err == -EPERM && !ofs->config.userxattr)
791 pr_info("try mounting with 'userxattr' option\n");
794 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
798 * We allowed sub-optimal upper fs configuration and don't want to break
799 * users over kernel upgrade, but we never allowed remote upper fs, so
800 * we can enforce strict requirements for remote upper fs.
802 if (ovl_dentry_remote(ofs->workdir) &&
803 (!d_type || !rename_whiteout || ofs->noxattr)) {
804 pr_err("upper fs missing required features.\n");
810 * For volatile mount, create a incompat/volatile/dirty file to keep
813 if (ofs->config.ovl_volatile) {
814 err = ovl_create_volatile_dirty(ofs);
816 pr_err("Failed to create volatile/dirty file.\n");
821 /* Check if upper/work fs supports file handles */
822 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
823 if (ofs->config.index && !fh_type) {
824 ofs->config.index = false;
825 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
827 ofs->nofh |= !fh_type;
829 /* Check if upper fs has 32bit inode numbers */
830 if (fh_type != FILEID_INO32_GEN)
833 /* NFS export of r/w mount depends on index */
834 if (ofs->config.nfs_export && !ofs->config.index) {
835 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
836 ofs->config.nfs_export = false;
843 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
844 const struct path *upperpath,
845 const struct path *workpath)
850 if (upperpath->mnt != workpath->mnt) {
851 pr_err("workdir and upperdir must reside under the same mount\n");
854 if (!ovl_workdir_ok(workpath->dentry, upperpath->dentry)) {
855 pr_err("workdir and upperdir must be separate subtrees\n");
859 ofs->workbasedir = dget(workpath->dentry);
861 if (ovl_inuse_trylock(ofs->workbasedir)) {
862 ofs->workdir_locked = true;
864 err = ovl_report_in_use(ofs, "workdir");
869 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
874 return ovl_make_workdir(sb, ofs, workpath);
877 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
878 struct ovl_entry *oe, const struct path *upperpath)
880 struct vfsmount *mnt = ovl_upper_mnt(ofs);
881 struct dentry *indexdir;
884 err = mnt_want_write(mnt);
888 /* Verify lower root is upper root origin */
889 err = ovl_verify_origin(ofs, upperpath->dentry,
890 ovl_lowerstack(oe)->dentry, true);
892 pr_err("failed to verify upper root origin\n");
896 /* index dir will act also as workdir */
897 iput(ofs->workdir_trap);
898 ofs->workdir_trap = NULL;
901 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
902 if (IS_ERR(indexdir)) {
903 err = PTR_ERR(indexdir);
904 } else if (indexdir) {
905 ofs->indexdir = indexdir;
906 ofs->workdir = dget(indexdir);
908 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
914 * Verify upper root is exclusively associated with index dir.
915 * Older kernels stored upper fh in ".overlay.origin"
916 * xattr. If that xattr exists, verify that it is a match to
917 * upper dir file handle. In any case, verify or set xattr
918 * ".overlay.upper" to indicate that index may have
921 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
922 err = ovl_verify_set_fh(ofs, ofs->indexdir,
924 upperpath->dentry, true, false);
926 pr_err("failed to verify index dir 'origin' xattr\n");
928 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
931 pr_err("failed to verify index dir 'upper' xattr\n");
933 /* Cleanup bad/stale/orphan index entries */
935 err = ovl_indexdir_cleanup(ofs);
937 if (err || !ofs->indexdir)
938 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
945 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
949 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
953 * We allow using single lower with null uuid for index and nfs_export
954 * for example to support those features with single lower squashfs.
955 * To avoid regressions in setups of overlay with re-formatted lower
956 * squashfs, do not allow decoding origin with lower null uuid unless
957 * user opted-in to one of the new features that require following the
958 * lower inode of non-dir upper.
960 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
963 for (i = 0; i < ofs->numfs; i++) {
965 * We use uuid to associate an overlay lower file handle with a
966 * lower layer, so we can accept lower fs with null uuid as long
967 * as all lower layers with null uuid are on the same fs.
968 * if we detect multiple lower fs with the same uuid, we
969 * disable lower file handle decoding on all of them.
971 if (ofs->fs[i].is_lower &&
972 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
973 ofs->fs[i].bad_uuid = true;
980 /* Get a unique fsid for the layer */
981 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
983 struct super_block *sb = path->mnt->mnt_sb;
987 bool bad_uuid = false;
990 for (i = 0; i < ofs->numfs; i++) {
991 if (ofs->fs[i].sb == sb)
995 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
997 if (ofs->config.xino == OVL_XINO_AUTO) {
998 ofs->config.xino = OVL_XINO_OFF;
1001 if (ofs->config.index || ofs->config.nfs_export) {
1002 ofs->config.index = false;
1003 ofs->config.nfs_export = false;
1007 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1008 uuid_is_null(&sb->s_uuid) ? "null" :
1010 path->dentry, ovl_xino_mode(&ofs->config));
1014 err = get_anon_bdev(&dev);
1016 pr_err("failed to get anonymous bdev for lowerpath\n");
1020 ofs->fs[ofs->numfs].sb = sb;
1021 ofs->fs[ofs->numfs].pseudo_dev = dev;
1022 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1024 return ofs->numfs++;
1028 * The fsid after the last lower fsid is used for the data layers.
1029 * It is a "null fs" with a null sb, null uuid, and no pseudo dev.
1031 static int ovl_get_data_fsid(struct ovl_fs *ofs)
1037 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1038 struct ovl_fs_context *ctx, struct ovl_layer *layers)
1042 size_t nr_merged_lower;
1044 ofs->fs = kcalloc(ctx->nr + 2, sizeof(struct ovl_sb), GFP_KERNEL);
1045 if (ofs->fs == NULL)
1049 * idx/fsid 0 are reserved for upper fs even with lower only overlay
1050 * and the last fsid is reserved for "null fs" of the data layers.
1055 * All lower layers that share the same fs as upper layer, use the same
1056 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1057 * only overlay to simplify ovl_fs_free().
1058 * is_lower will be set if upper fs is shared with a lower layer.
1060 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1062 pr_err("failed to get anonymous bdev for upper fs\n");
1066 if (ovl_upper_mnt(ofs)) {
1067 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1068 ofs->fs[0].is_lower = false;
1071 nr_merged_lower = ctx->nr - ctx->nr_data;
1072 for (i = 0; i < ctx->nr; i++) {
1073 struct ovl_fs_context_layer *l = &ctx->lower[i];
1074 struct vfsmount *mnt;
1078 if (i < nr_merged_lower)
1079 fsid = ovl_get_fsid(ofs, &l->path);
1081 fsid = ovl_get_data_fsid(ofs);
1086 * Check if lower root conflicts with this overlay layers before
1087 * checking if it is in-use as upperdir/workdir of "another"
1088 * mount, because we do not bother to check in ovl_is_inuse() if
1089 * the upperdir/workdir is in fact in-use by our
1092 err = ovl_setup_trap(sb, l->path.dentry, &trap, "lowerdir");
1096 if (ovl_is_inuse(l->path.dentry)) {
1097 err = ovl_report_in_use(ofs, "lowerdir");
1104 mnt = clone_private_mount(&l->path);
1107 pr_err("failed to clone lowerpath\n");
1113 * Make lower layers R/O. That way fchmod/fchown on lower file
1114 * will fail instead of modifying lower fs.
1116 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1118 layers[ofs->numlayer].trap = trap;
1119 layers[ofs->numlayer].mnt = mnt;
1120 layers[ofs->numlayer].idx = ofs->numlayer;
1121 layers[ofs->numlayer].fsid = fsid;
1122 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1123 /* Store for printing lowerdir=... in ovl_show_options() */
1124 ofs->config.lowerdirs[ofs->numlayer] = l->name;
1127 ofs->fs[fsid].is_lower = true;
1131 * When all layers on same fs, overlay can use real inode numbers.
1132 * With mount option "xino=<on|auto>", mounter declares that there are
1133 * enough free high bits in underlying fs to hold the unique fsid.
1134 * If overlayfs does encounter underlying inodes using the high xino
1135 * bits reserved for fsid, it emits a warning and uses the original
1136 * inode number or a non persistent inode number allocated from a
1139 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1140 if (ofs->config.xino == OVL_XINO_ON)
1141 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1143 } else if (ofs->config.xino == OVL_XINO_OFF) {
1144 ofs->xino_mode = -1;
1145 } else if (ofs->xino_mode < 0) {
1147 * This is a roundup of number of bits needed for encoding
1148 * fsid, where fsid 0 is reserved for upper fs (even with
1149 * lower only overlay) +1 extra bit is reserved for the non
1150 * persistent inode number range that is used for resolving
1151 * xino lower bits overflow.
1153 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1154 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1157 if (ofs->xino_mode > 0) {
1158 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1165 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1166 struct ovl_fs_context *ctx,
1168 struct ovl_layer *layers)
1172 size_t nr_merged_lower;
1173 struct ovl_entry *oe;
1174 struct ovl_path *lowerstack;
1176 struct ovl_fs_context_layer *l;
1178 if (!ofs->config.upperdir && ctx->nr == 1) {
1179 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1180 return ERR_PTR(-EINVAL);
1184 for (i = 0; i < ctx->nr; i++) {
1187 err = ovl_lower_dir(l->name, &l->path, ofs, &sb->s_stack_depth);
1189 return ERR_PTR(err);
1193 sb->s_stack_depth++;
1194 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1195 pr_err("maximum fs stacking depth exceeded\n");
1196 return ERR_PTR(err);
1199 err = ovl_get_layers(sb, ofs, ctx, layers);
1201 return ERR_PTR(err);
1204 /* Data-only layers are not merged in root directory */
1205 nr_merged_lower = ctx->nr - ctx->nr_data;
1206 oe = ovl_alloc_entry(nr_merged_lower);
1208 return ERR_PTR(err);
1210 lowerstack = ovl_lowerstack(oe);
1211 for (i = 0; i < nr_merged_lower; i++) {
1213 lowerstack[i].dentry = dget(l->path.dentry);
1214 lowerstack[i].layer = &ofs->layers[i + 1];
1216 ofs->numdatalayer = ctx->nr_data;
1222 * Check if this layer root is a descendant of:
1223 * - another layer of this overlayfs instance
1224 * - upper/work dir of any overlayfs instance
1226 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1227 struct dentry *dentry, const char *name,
1230 struct dentry *next = dentry, *parent;
1236 parent = dget_parent(next);
1238 /* Walk back ancestors to root (inclusive) looking for traps */
1239 while (!err && parent != next) {
1240 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1242 pr_err("overlapping %s path\n", name);
1243 } else if (ovl_is_inuse(parent)) {
1244 err = ovl_report_in_use(ofs, name);
1247 parent = dget_parent(next);
1257 * Check if any of the layers or work dirs overlap.
1259 static int ovl_check_overlapping_layers(struct super_block *sb,
1264 if (ovl_upper_mnt(ofs)) {
1265 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1271 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1272 * this instance and covers overlapping work and index dirs,
1273 * unless work or index dir have been moved since created inside
1274 * workbasedir. In that case, we already have their traps in
1275 * inode cache and we will catch that case on lookup.
1277 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1283 for (i = 1; i < ofs->numlayer; i++) {
1284 err = ovl_check_layer(sb, ofs,
1285 ofs->layers[i].mnt->mnt_root,
1294 static struct dentry *ovl_get_root(struct super_block *sb,
1295 struct dentry *upperdentry,
1296 struct ovl_entry *oe)
1298 struct dentry *root;
1299 struct ovl_path *lowerpath = ovl_lowerstack(oe);
1300 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1301 int fsid = lowerpath->layer->fsid;
1302 struct ovl_inode_params oip = {
1303 .upperdentry = upperdentry,
1307 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1312 /* Root inode uses upper st_ino/i_ino */
1313 ino = d_inode(upperdentry)->i_ino;
1315 ovl_dentry_set_upper_alias(root);
1316 if (ovl_is_impuredir(sb, upperdentry))
1317 ovl_set_flag(OVL_IMPURE, d_inode(root));
1320 /* Root is always merge -> can have whiteouts */
1321 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1322 ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1323 ovl_set_upperdata(d_inode(root));
1324 ovl_inode_init(d_inode(root), &oip, ino, fsid);
1325 ovl_dentry_init_flags(root, upperdentry, oe, DCACHE_OP_WEAK_REVALIDATE);
1326 /* root keeps a reference of upperdentry */
1332 int ovl_fill_super(struct super_block *sb, struct fs_context *fc)
1334 struct ovl_fs *ofs = sb->s_fs_info;
1335 struct ovl_fs_context *ctx = fc->fs_private;
1336 struct dentry *root_dentry;
1337 struct ovl_entry *oe;
1338 struct ovl_layer *layers;
1343 if (WARN_ON(fc->user_ns != current_user_ns()))
1346 sb->s_d_op = &ovl_dentry_operations;
1349 ofs->creator_cred = cred = prepare_creds();
1353 err = ovl_fs_params_verify(ctx, &ofs->config);
1359 if (!(fc->sb_flags & SB_SILENT))
1360 pr_err("missing 'lowerdir'\n");
1365 layers = kcalloc(ctx->nr + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1369 ofs->config.lowerdirs = kcalloc(ctx->nr + 1, sizeof(char *), GFP_KERNEL);
1370 if (!ofs->config.lowerdirs) {
1374 ofs->layers = layers;
1376 * Layer 0 is reserved for upper even if there's no upper.
1377 * For consistency, config.lowerdirs[0] is NULL.
1381 sb->s_stack_depth = 0;
1382 sb->s_maxbytes = MAX_LFS_FILESIZE;
1383 atomic_long_set(&ofs->last_ino, 1);
1384 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1385 if (ofs->config.xino != OVL_XINO_OFF) {
1386 ofs->xino_mode = BITS_PER_LONG - 32;
1387 if (!ofs->xino_mode) {
1388 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1389 ofs->config.xino = OVL_XINO_OFF;
1393 /* alloc/destroy_inode needed for setting up traps in inode cache */
1394 sb->s_op = &ovl_super_operations;
1396 if (ofs->config.upperdir) {
1397 struct super_block *upper_sb;
1400 if (!ofs->config.workdir) {
1401 pr_err("missing 'workdir'\n");
1405 err = ovl_get_upper(sb, ofs, &layers[0], &ctx->upper);
1409 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
1410 if (!ovl_should_sync(ofs)) {
1411 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
1412 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
1414 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1419 err = ovl_get_workdir(sb, ofs, &ctx->upper, &ctx->work);
1424 sb->s_flags |= SB_RDONLY;
1426 sb->s_stack_depth = upper_sb->s_stack_depth;
1427 sb->s_time_gran = upper_sb->s_time_gran;
1429 oe = ovl_get_lowerstack(sb, ctx, ofs, layers);
1434 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1435 if (!ovl_upper_mnt(ofs))
1436 sb->s_flags |= SB_RDONLY;
1438 if (!ovl_origin_uuid(ofs) && ofs->numfs > 1) {
1439 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=null.\n");
1440 ofs->config.uuid = OVL_UUID_NULL;
1441 } else if (ovl_has_fsid(ofs) && ovl_upper_mnt(ofs)) {
1442 /* Use per instance persistent uuid/fsid */
1443 ovl_init_uuid_xattr(sb, ofs, &ctx->upper);
1446 if (!ovl_force_readonly(ofs) && ofs->config.index) {
1447 err = ovl_get_indexdir(sb, ofs, oe, &ctx->upper);
1451 /* Force r/o mount with no index dir */
1453 sb->s_flags |= SB_RDONLY;
1456 err = ovl_check_overlapping_layers(sb, ofs);
1460 /* Show index=off in /proc/mounts for forced r/o mount */
1461 if (!ofs->indexdir) {
1462 ofs->config.index = false;
1463 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
1464 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1465 ofs->config.nfs_export = false;
1469 if (ofs->config.metacopy && ofs->config.nfs_export) {
1470 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1471 ofs->config.nfs_export = false;
1475 * Support encoding decodable file handles with nfs_export=on
1476 * and encoding non-decodable file handles with nfs_export=off
1477 * if all layers support file handles.
1479 if (ofs->config.nfs_export)
1480 sb->s_export_op = &ovl_export_operations;
1481 else if (!ofs->nofh)
1482 sb->s_export_op = &ovl_export_fid_operations;
1484 /* Never override disk quota limits or use reserved space */
1485 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1487 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1488 sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
1489 ovl_trusted_xattr_handlers;
1490 sb->s_fs_info = ofs;
1491 sb->s_flags |= SB_POSIXACL;
1492 sb->s_iflags |= SB_I_SKIP_SYNC | SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1495 root_dentry = ovl_get_root(sb, ctx->upper.dentry, oe);
1499 sb->s_root = root_dentry;
1507 sb->s_fs_info = NULL;
1511 struct file_system_type ovl_fs_type = {
1512 .owner = THIS_MODULE,
1514 .init_fs_context = ovl_init_fs_context,
1515 .parameters = ovl_parameter_spec,
1516 .fs_flags = FS_USERNS_MOUNT,
1517 .kill_sb = kill_anon_super,
1519 MODULE_ALIAS_FS("overlay");
1521 static void ovl_inode_init_once(void *foo)
1523 struct ovl_inode *oi = foo;
1525 inode_init_once(&oi->vfs_inode);
1528 static int __init ovl_init(void)
1532 ovl_inode_cachep = kmem_cache_create("ovl_inode",
1533 sizeof(struct ovl_inode), 0,
1534 (SLAB_RECLAIM_ACCOUNT|
1535 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1536 ovl_inode_init_once);
1537 if (ovl_inode_cachep == NULL)
1540 err = ovl_aio_request_cache_init();
1542 err = register_filesystem(&ovl_fs_type);
1546 ovl_aio_request_cache_destroy();
1548 kmem_cache_destroy(ovl_inode_cachep);
1553 static void __exit ovl_exit(void)
1555 unregister_filesystem(&ovl_fs_type);
1558 * Make sure all delayed rcu free inodes are flushed before we
1562 kmem_cache_destroy(ovl_inode_cachep);
1563 ovl_aio_request_cache_destroy();
1566 module_init(ovl_init);
1567 module_exit(ovl_exit);