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"
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 = OVL_E(dentry);
108 struct ovl_path *lowerstack = ovl_lowerstack(oe);
109 struct inode *inode = d_inode_rcu(dentry);
110 struct dentry *upper;
114 /* Careful in RCU mode */
118 upper = ovl_i_dentry_upper(inode);
120 ret = ovl_revalidate_real(upper, flags, weak);
122 for (i = 0; ret > 0 && i < ovl_numlower(oe); i++)
123 ret = ovl_revalidate_real(lowerstack[i].dentry, flags, weak);
128 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
130 return ovl_dentry_revalidate_common(dentry, flags, false);
133 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
135 return ovl_dentry_revalidate_common(dentry, flags, true);
138 static const struct dentry_operations ovl_dentry_operations = {
139 .d_real = ovl_d_real,
140 .d_revalidate = ovl_dentry_revalidate,
141 .d_weak_revalidate = ovl_dentry_weak_revalidate,
144 static struct kmem_cache *ovl_inode_cachep;
146 static struct inode *ovl_alloc_inode(struct super_block *sb)
148 struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
157 oi->__upperdentry = NULL;
158 oi->lowerdata_redirect = NULL;
160 mutex_init(&oi->lock);
162 return &oi->vfs_inode;
165 static void ovl_free_inode(struct inode *inode)
167 struct ovl_inode *oi = OVL_I(inode);
170 mutex_destroy(&oi->lock);
171 kmem_cache_free(ovl_inode_cachep, oi);
174 static void ovl_destroy_inode(struct inode *inode)
176 struct ovl_inode *oi = OVL_I(inode);
178 dput(oi->__upperdentry);
179 ovl_free_entry(oi->oe);
180 if (S_ISDIR(inode->i_mode))
181 ovl_dir_cache_free(inode);
183 kfree(oi->lowerdata_redirect);
186 static void ovl_put_super(struct super_block *sb)
188 struct ovl_fs *ofs = OVL_FS(sb);
194 /* Sync real dirty inodes in upper filesystem (if it exists) */
195 static int ovl_sync_fs(struct super_block *sb, int wait)
197 struct ovl_fs *ofs = OVL_FS(sb);
198 struct super_block *upper_sb;
201 ret = ovl_sync_status(ofs);
203 * We have to always set the err, because the return value isn't
204 * checked in syncfs, and instead indirectly return an error via
205 * the sb's writeback errseq, which VFS inspects after this call.
208 errseq_set(&sb->s_wb_err, -EIO);
216 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
217 * All the super blocks will be iterated, including upper_sb.
219 * If this is a syncfs(2) call, then we do need to call
220 * sync_filesystem() on upper_sb, but enough if we do it when being
221 * called with wait == 1.
226 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
228 down_read(&upper_sb->s_umount);
229 ret = sync_filesystem(upper_sb);
230 up_read(&upper_sb->s_umount);
237 * @dentry: The dentry to query
238 * @buf: The struct kstatfs to fill in with stats
240 * Get the filesystem statistics. As writes always target the upper layer
241 * filesystem pass the statfs to the upper filesystem (if it exists)
243 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
245 struct super_block *sb = dentry->d_sb;
246 struct ovl_fs *ofs = OVL_FS(sb);
247 struct dentry *root_dentry = sb->s_root;
251 ovl_path_real(root_dentry, &path);
253 err = vfs_statfs(&path, buf);
255 buf->f_namelen = ofs->namelen;
256 buf->f_type = OVERLAYFS_SUPER_MAGIC;
257 if (ovl_has_fsid(ofs))
258 buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
264 static const struct super_operations ovl_super_operations = {
265 .alloc_inode = ovl_alloc_inode,
266 .free_inode = ovl_free_inode,
267 .destroy_inode = ovl_destroy_inode,
268 .drop_inode = generic_delete_inode,
269 .put_super = ovl_put_super,
270 .sync_fs = ovl_sync_fs,
271 .statfs = ovl_statfs,
272 .show_options = ovl_show_options,
275 #define OVL_WORKDIR_NAME "work"
276 #define OVL_INDEXDIR_NAME "index"
278 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
279 const char *name, bool persist)
281 struct inode *dir = ofs->workbasedir->d_inode;
282 struct vfsmount *mnt = ovl_upper_mnt(ofs);
285 bool retried = false;
287 inode_lock_nested(dir, I_MUTEX_PARENT);
289 work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
292 struct iattr attr = {
293 .ia_valid = ATTR_MODE,
294 .ia_mode = S_IFDIR | 0,
306 err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
308 if (err == -EINVAL) {
315 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
319 /* Weird filesystem returning with hashed negative (kernfs)? */
321 if (d_really_is_negative(work))
325 * Try to remove POSIX ACL xattrs from workdir. We are good if:
327 * a) success (there was a POSIX ACL xattr and was removed)
328 * b) -ENODATA (there was no POSIX ACL xattr)
329 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
331 * There are various other error values that could effectively
332 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
333 * if the xattr name is too long), but the set of filesystems
334 * allowed as upper are limited to "normal" ones, where checking
335 * for the above two errors is sufficient.
337 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);
338 if (err && err != -ENODATA && err != -EOPNOTSUPP)
341 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);
342 if (err && err != -ENODATA && err != -EOPNOTSUPP)
345 /* Clear any inherited mode bits */
346 inode_lock(work->d_inode);
347 err = ovl_do_notify_change(ofs, work, &attr);
348 inode_unlock(work->d_inode);
362 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
363 ofs->config.workdir, name, -err);
368 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
371 struct kstatfs statfs;
372 int err = vfs_statfs(path, &statfs);
375 pr_err("statfs failed on '%s'\n", name);
377 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
382 static int ovl_lower_dir(const char *name, struct path *path,
383 struct ovl_fs *ofs, int *stack_depth)
388 err = ovl_check_namelen(path, ofs, name);
392 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
395 * The inodes index feature and NFS export need to encode and decode
396 * file handles, so they require that all layers support them.
398 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
399 if ((ofs->config.nfs_export ||
400 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
401 ofs->config.index = false;
402 ofs->config.nfs_export = false;
403 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
406 ofs->nofh |= !fh_type;
408 * Decoding origin file handle is required for persistent st_ino.
409 * Without persistent st_ino, xino=auto falls back to xino=off.
411 if (ofs->config.xino == OVL_XINO_AUTO &&
412 ofs->config.upperdir && !fh_type) {
413 ofs->config.xino = OVL_XINO_OFF;
414 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
418 /* Check if lower fs has 32bit inode numbers */
419 if (fh_type != FILEID_INO32_GEN)
425 /* Workdir should not be subdir of upperdir and vice versa */
426 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
430 if (workdir != upperdir) {
431 ok = (lock_rename(workdir, upperdir) == NULL);
432 unlock_rename(workdir, upperdir);
437 static int ovl_own_xattr_get(const struct xattr_handler *handler,
438 struct dentry *dentry, struct inode *inode,
439 const char *name, void *buffer, size_t size)
444 static int ovl_own_xattr_set(const struct xattr_handler *handler,
445 struct mnt_idmap *idmap,
446 struct dentry *dentry, struct inode *inode,
447 const char *name, const void *value,
448 size_t size, int flags)
453 static int ovl_other_xattr_get(const struct xattr_handler *handler,
454 struct dentry *dentry, struct inode *inode,
455 const char *name, void *buffer, size_t size)
457 return ovl_xattr_get(dentry, inode, name, buffer, size);
460 static int ovl_other_xattr_set(const struct xattr_handler *handler,
461 struct mnt_idmap *idmap,
462 struct dentry *dentry, struct inode *inode,
463 const char *name, const void *value,
464 size_t size, int flags)
466 return ovl_xattr_set(dentry, inode, name, value, size, flags);
469 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
470 .prefix = OVL_XATTR_TRUSTED_PREFIX,
471 .get = ovl_own_xattr_get,
472 .set = ovl_own_xattr_set,
475 static const struct xattr_handler ovl_own_user_xattr_handler = {
476 .prefix = OVL_XATTR_USER_PREFIX,
477 .get = ovl_own_xattr_get,
478 .set = ovl_own_xattr_set,
481 static const struct xattr_handler ovl_other_xattr_handler = {
482 .prefix = "", /* catch all */
483 .get = ovl_other_xattr_get,
484 .set = ovl_other_xattr_set,
487 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
488 &ovl_own_trusted_xattr_handler,
489 &ovl_other_xattr_handler,
493 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
494 &ovl_own_user_xattr_handler,
495 &ovl_other_xattr_handler,
499 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
500 struct inode **ptrap, const char *name)
505 trap = ovl_get_trap_inode(sb, dir);
506 err = PTR_ERR_OR_ZERO(trap);
509 pr_err("conflicting %s path\n", name);
518 * Determine how we treat concurrent use of upperdir/workdir based on the
519 * index feature. This is papering over mount leaks of container runtimes,
520 * for example, an old overlay mount is leaked and now its upperdir is
521 * attempted to be used as a lower layer in a new overlay mount.
523 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
525 if (ofs->config.index) {
526 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
530 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
536 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
537 struct ovl_layer *upper_layer,
538 const struct path *upperpath)
540 struct vfsmount *upper_mnt;
543 /* Upperdir path should not be r/o */
544 if (__mnt_is_readonly(upperpath->mnt)) {
545 pr_err("upper fs is r/o, try multi-lower layers mount\n");
550 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
554 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
559 upper_mnt = clone_private_mount(upperpath);
560 err = PTR_ERR(upper_mnt);
561 if (IS_ERR(upper_mnt)) {
562 pr_err("failed to clone upperpath\n");
566 /* Don't inherit atime flags */
567 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
568 upper_layer->mnt = upper_mnt;
569 upper_layer->idx = 0;
570 upper_layer->fsid = 0;
573 upper_layer->name = kstrdup(ofs->config.upperdir, GFP_KERNEL);
574 if (!upper_layer->name)
578 * Inherit SB_NOSEC flag from upperdir.
580 * This optimization changes behavior when a security related attribute
581 * (suid/sgid/security.*) is changed on an underlying layer. This is
582 * okay because we don't yet have guarantees in that case, but it will
583 * need careful treatment once we want to honour changes to underlying
586 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
587 sb->s_flags |= SB_NOSEC;
589 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
590 ofs->upperdir_locked = true;
592 err = ovl_report_in_use(ofs, "upperdir");
603 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
604 * negative values if error is encountered.
606 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
608 struct dentry *workdir = ofs->workdir;
609 struct inode *dir = d_inode(workdir);
612 struct dentry *whiteout;
613 struct name_snapshot name;
616 inode_lock_nested(dir, I_MUTEX_PARENT);
618 temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
623 dest = ovl_lookup_temp(ofs, workdir);
630 /* Name is inline and stable - using snapshot as a copy helper */
631 take_dentry_name_snapshot(&name, temp);
632 err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
639 whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
640 err = PTR_ERR(whiteout);
641 if (IS_ERR(whiteout))
644 err = ovl_is_whiteout(whiteout);
646 /* Best effort cleanup of whiteout and temp file */
648 ovl_cleanup(ofs, dir, whiteout);
652 ovl_cleanup(ofs, dir, temp);
653 release_dentry_name_snapshot(&name);
663 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
664 struct dentry *parent,
665 const char *name, umode_t mode)
667 size_t len = strlen(name);
668 struct dentry *child;
670 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
671 child = ovl_lookup_upper(ofs, name, parent, len);
672 if (!IS_ERR(child) && !child->d_inode)
673 child = ovl_create_real(ofs, parent->d_inode, child,
675 inode_unlock(parent->d_inode);
682 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
685 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
688 struct dentry *d = dget(ofs->workbasedir);
689 static const char *const volatile_path[] = {
690 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
692 const char *const *name = volatile_path;
694 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
695 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
703 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
704 const struct path *workpath)
706 struct vfsmount *mnt = ovl_upper_mnt(ofs);
707 struct dentry *workdir;
708 struct file *tmpfile;
709 bool rename_whiteout;
714 err = mnt_want_write(mnt);
718 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
719 err = PTR_ERR(workdir);
720 if (IS_ERR_OR_NULL(workdir))
723 ofs->workdir = workdir;
725 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
730 * Upper should support d_type, else whiteouts are visible. Given
731 * workdir and upper are on same fs, we can do iterate_dir() on
732 * workdir. This check requires successful creation of workdir in
735 err = ovl_check_d_type_supported(workpath);
741 pr_warn("upper fs needs to support d_type.\n");
743 /* Check if upper/work fs supports O_TMPFILE */
744 tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
745 ofs->tmpfile = !IS_ERR(tmpfile);
749 pr_warn("upper fs does not support tmpfile.\n");
752 /* Check if upper/work fs supports RENAME_WHITEOUT */
753 err = ovl_check_rename_whiteout(ofs);
757 rename_whiteout = err;
758 if (!rename_whiteout)
759 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
762 * Check if upper/work fs supports (trusted|user).overlay.* xattr
764 err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
766 pr_warn("failed to set xattr on upper\n");
768 if (ovl_redirect_follow(ofs)) {
769 ofs->config.redirect_mode = OVL_REDIRECT_NOFOLLOW;
770 pr_warn("...falling back to redirect_dir=nofollow.\n");
772 if (ofs->config.metacopy) {
773 ofs->config.metacopy = false;
774 pr_warn("...falling back to metacopy=off.\n");
776 if (ofs->config.index) {
777 ofs->config.index = false;
778 pr_warn("...falling back to index=off.\n");
780 if (ovl_has_fsid(ofs)) {
781 ofs->config.uuid = OVL_UUID_NULL;
782 pr_warn("...falling back to uuid=null.\n");
785 * xattr support is required for persistent st_ino.
786 * Without persistent st_ino, xino=auto falls back to xino=off.
788 if (ofs->config.xino == OVL_XINO_AUTO) {
789 ofs->config.xino = OVL_XINO_OFF;
790 pr_warn("...falling back to xino=off.\n");
792 if (err == -EPERM && !ofs->config.userxattr)
793 pr_info("try mounting with 'userxattr' option\n");
796 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
800 * We allowed sub-optimal upper fs configuration and don't want to break
801 * users over kernel upgrade, but we never allowed remote upper fs, so
802 * we can enforce strict requirements for remote upper fs.
804 if (ovl_dentry_remote(ofs->workdir) &&
805 (!d_type || !rename_whiteout || ofs->noxattr)) {
806 pr_err("upper fs missing required features.\n");
812 * For volatile mount, create a incompat/volatile/dirty file to keep
815 if (ofs->config.ovl_volatile) {
816 err = ovl_create_volatile_dirty(ofs);
818 pr_err("Failed to create volatile/dirty file.\n");
823 /* Check if upper/work fs supports file handles */
824 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
825 if (ofs->config.index && !fh_type) {
826 ofs->config.index = false;
827 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
829 ofs->nofh |= !fh_type;
831 /* Check if upper fs has 32bit inode numbers */
832 if (fh_type != FILEID_INO32_GEN)
835 /* NFS export of r/w mount depends on index */
836 if (ofs->config.nfs_export && !ofs->config.index) {
837 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
838 ofs->config.nfs_export = false;
845 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
846 const struct path *upperpath,
847 const struct path *workpath)
852 if (upperpath->mnt != workpath->mnt) {
853 pr_err("workdir and upperdir must reside under the same mount\n");
856 if (!ovl_workdir_ok(workpath->dentry, upperpath->dentry)) {
857 pr_err("workdir and upperdir must be separate subtrees\n");
861 ofs->workbasedir = dget(workpath->dentry);
863 if (ovl_inuse_trylock(ofs->workbasedir)) {
864 ofs->workdir_locked = true;
866 err = ovl_report_in_use(ofs, "workdir");
871 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
876 return ovl_make_workdir(sb, ofs, workpath);
879 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
880 struct ovl_entry *oe, const struct path *upperpath)
882 struct vfsmount *mnt = ovl_upper_mnt(ofs);
883 struct dentry *indexdir;
886 err = mnt_want_write(mnt);
890 /* Verify lower root is upper root origin */
891 err = ovl_verify_origin(ofs, upperpath->dentry,
892 ovl_lowerstack(oe)->dentry, true);
894 pr_err("failed to verify upper root origin\n");
898 /* index dir will act also as workdir */
899 iput(ofs->workdir_trap);
900 ofs->workdir_trap = NULL;
903 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
904 if (IS_ERR(indexdir)) {
905 err = PTR_ERR(indexdir);
906 } else if (indexdir) {
907 ofs->indexdir = indexdir;
908 ofs->workdir = dget(indexdir);
910 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
916 * Verify upper root is exclusively associated with index dir.
917 * Older kernels stored upper fh in ".overlay.origin"
918 * xattr. If that xattr exists, verify that it is a match to
919 * upper dir file handle. In any case, verify or set xattr
920 * ".overlay.upper" to indicate that index may have
923 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
924 err = ovl_verify_set_fh(ofs, ofs->indexdir,
926 upperpath->dentry, true, false);
928 pr_err("failed to verify index dir 'origin' xattr\n");
930 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
933 pr_err("failed to verify index dir 'upper' xattr\n");
935 /* Cleanup bad/stale/orphan index entries */
937 err = ovl_indexdir_cleanup(ofs);
939 if (err || !ofs->indexdir)
940 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
947 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
951 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
955 * We allow using single lower with null uuid for index and nfs_export
956 * for example to support those features with single lower squashfs.
957 * To avoid regressions in setups of overlay with re-formatted lower
958 * squashfs, do not allow decoding origin with lower null uuid unless
959 * user opted-in to one of the new features that require following the
960 * lower inode of non-dir upper.
962 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
965 for (i = 0; i < ofs->numfs; i++) {
967 * We use uuid to associate an overlay lower file handle with a
968 * lower layer, so we can accept lower fs with null uuid as long
969 * as all lower layers with null uuid are on the same fs.
970 * if we detect multiple lower fs with the same uuid, we
971 * disable lower file handle decoding on all of them.
973 if (ofs->fs[i].is_lower &&
974 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
975 ofs->fs[i].bad_uuid = true;
982 /* Get a unique fsid for the layer */
983 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
985 struct super_block *sb = path->mnt->mnt_sb;
989 bool bad_uuid = false;
992 for (i = 0; i < ofs->numfs; i++) {
993 if (ofs->fs[i].sb == sb)
997 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
999 if (ofs->config.xino == OVL_XINO_AUTO) {
1000 ofs->config.xino = OVL_XINO_OFF;
1003 if (ofs->config.index || ofs->config.nfs_export) {
1004 ofs->config.index = false;
1005 ofs->config.nfs_export = false;
1009 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1010 uuid_is_null(&sb->s_uuid) ? "null" :
1012 path->dentry, ovl_xino_mode(&ofs->config));
1016 err = get_anon_bdev(&dev);
1018 pr_err("failed to get anonymous bdev for lowerpath\n");
1022 ofs->fs[ofs->numfs].sb = sb;
1023 ofs->fs[ofs->numfs].pseudo_dev = dev;
1024 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1026 return ofs->numfs++;
1030 * The fsid after the last lower fsid is used for the data layers.
1031 * It is a "null fs" with a null sb, null uuid, and no pseudo dev.
1033 static int ovl_get_data_fsid(struct ovl_fs *ofs)
1039 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1040 struct ovl_fs_context *ctx, struct ovl_layer *layers)
1044 size_t nr_merged_lower;
1046 ofs->fs = kcalloc(ctx->nr + 2, sizeof(struct ovl_sb), GFP_KERNEL);
1047 if (ofs->fs == NULL)
1051 * idx/fsid 0 are reserved for upper fs even with lower only overlay
1052 * and the last fsid is reserved for "null fs" of the data layers.
1057 * All lower layers that share the same fs as upper layer, use the same
1058 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1059 * only overlay to simplify ovl_fs_free().
1060 * is_lower will be set if upper fs is shared with a lower layer.
1062 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1064 pr_err("failed to get anonymous bdev for upper fs\n");
1068 if (ovl_upper_mnt(ofs)) {
1069 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1070 ofs->fs[0].is_lower = false;
1073 nr_merged_lower = ctx->nr - ctx->nr_data;
1074 for (i = 0; i < ctx->nr; i++) {
1075 struct ovl_fs_context_layer *l = &ctx->lower[i];
1076 struct vfsmount *mnt;
1080 if (i < nr_merged_lower)
1081 fsid = ovl_get_fsid(ofs, &l->path);
1083 fsid = ovl_get_data_fsid(ofs);
1088 * Check if lower root conflicts with this overlay layers before
1089 * checking if it is in-use as upperdir/workdir of "another"
1090 * mount, because we do not bother to check in ovl_is_inuse() if
1091 * the upperdir/workdir is in fact in-use by our
1094 err = ovl_setup_trap(sb, l->path.dentry, &trap, "lowerdir");
1098 if (ovl_is_inuse(l->path.dentry)) {
1099 err = ovl_report_in_use(ofs, "lowerdir");
1106 mnt = clone_private_mount(&l->path);
1109 pr_err("failed to clone lowerpath\n");
1115 * Make lower layers R/O. That way fchmod/fchown on lower file
1116 * will fail instead of modifying lower fs.
1118 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1120 layers[ofs->numlayer].trap = trap;
1121 layers[ofs->numlayer].mnt = mnt;
1122 layers[ofs->numlayer].idx = ofs->numlayer;
1123 layers[ofs->numlayer].fsid = fsid;
1124 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1125 layers[ofs->numlayer].name = l->name;
1128 ofs->fs[fsid].is_lower = true;
1132 * When all layers on same fs, overlay can use real inode numbers.
1133 * With mount option "xino=<on|auto>", mounter declares that there are
1134 * enough free high bits in underlying fs to hold the unique fsid.
1135 * If overlayfs does encounter underlying inodes using the high xino
1136 * bits reserved for fsid, it emits a warning and uses the original
1137 * inode number or a non persistent inode number allocated from a
1140 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1141 if (ofs->config.xino == OVL_XINO_ON)
1142 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1144 } else if (ofs->config.xino == OVL_XINO_OFF) {
1145 ofs->xino_mode = -1;
1146 } else if (ofs->xino_mode < 0) {
1148 * This is a roundup of number of bits needed for encoding
1149 * fsid, where fsid 0 is reserved for upper fs (even with
1150 * lower only overlay) +1 extra bit is reserved for the non
1151 * persistent inode number range that is used for resolving
1152 * xino lower bits overflow.
1154 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1155 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1158 if (ofs->xino_mode > 0) {
1159 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1166 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1167 struct ovl_fs_context *ctx,
1169 struct ovl_layer *layers)
1173 size_t nr_merged_lower;
1174 struct ovl_entry *oe;
1175 struct ovl_path *lowerstack;
1177 struct ovl_fs_context_layer *l;
1179 if (!ofs->config.upperdir && ctx->nr == 1) {
1180 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1181 return ERR_PTR(-EINVAL);
1185 for (i = 0; i < ctx->nr; i++) {
1188 err = ovl_lower_dir(l->name, &l->path, ofs, &sb->s_stack_depth);
1190 return ERR_PTR(err);
1194 sb->s_stack_depth++;
1195 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1196 pr_err("maximum fs stacking depth exceeded\n");
1197 return ERR_PTR(err);
1200 err = ovl_get_layers(sb, ofs, ctx, layers);
1202 return ERR_PTR(err);
1205 /* Data-only layers are not merged in root directory */
1206 nr_merged_lower = ctx->nr - ctx->nr_data;
1207 oe = ovl_alloc_entry(nr_merged_lower);
1209 return ERR_PTR(err);
1211 lowerstack = ovl_lowerstack(oe);
1212 for (i = 0; i < nr_merged_lower; i++) {
1214 lowerstack[i].dentry = dget(l->path.dentry);
1215 lowerstack[i].layer = &ofs->layers[i + 1];
1217 ofs->numdatalayer = ctx->nr_data;
1223 * Check if this layer root is a descendant of:
1224 * - another layer of this overlayfs instance
1225 * - upper/work dir of any overlayfs instance
1227 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1228 struct dentry *dentry, const char *name,
1231 struct dentry *next = dentry, *parent;
1237 parent = dget_parent(next);
1239 /* Walk back ancestors to root (inclusive) looking for traps */
1240 while (!err && parent != next) {
1241 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1243 pr_err("overlapping %s path\n", name);
1244 } else if (ovl_is_inuse(parent)) {
1245 err = ovl_report_in_use(ofs, name);
1248 parent = dget_parent(next);
1258 * Check if any of the layers or work dirs overlap.
1260 static int ovl_check_overlapping_layers(struct super_block *sb,
1265 if (ovl_upper_mnt(ofs)) {
1266 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1272 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1273 * this instance and covers overlapping work and index dirs,
1274 * unless work or index dir have been moved since created inside
1275 * workbasedir. In that case, we already have their traps in
1276 * inode cache and we will catch that case on lookup.
1278 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1284 for (i = 1; i < ofs->numlayer; i++) {
1285 err = ovl_check_layer(sb, ofs,
1286 ofs->layers[i].mnt->mnt_root,
1295 static struct dentry *ovl_get_root(struct super_block *sb,
1296 struct dentry *upperdentry,
1297 struct ovl_entry *oe)
1299 struct dentry *root;
1300 struct ovl_path *lowerpath = ovl_lowerstack(oe);
1301 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1302 int fsid = lowerpath->layer->fsid;
1303 struct ovl_inode_params oip = {
1304 .upperdentry = upperdentry,
1308 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1313 /* Root inode uses upper st_ino/i_ino */
1314 ino = d_inode(upperdentry)->i_ino;
1316 ovl_dentry_set_upper_alias(root);
1317 if (ovl_is_impuredir(sb, upperdentry))
1318 ovl_set_flag(OVL_IMPURE, d_inode(root));
1321 /* Root is always merge -> can have whiteouts */
1322 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1323 ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1324 ovl_set_upperdata(d_inode(root));
1325 ovl_inode_init(d_inode(root), &oip, ino, fsid);
1326 ovl_dentry_init_flags(root, upperdentry, oe, DCACHE_OP_WEAK_REVALIDATE);
1327 /* root keeps a reference of upperdentry */
1333 int ovl_fill_super(struct super_block *sb, struct fs_context *fc)
1335 struct ovl_fs *ofs = sb->s_fs_info;
1336 struct ovl_fs_context *ctx = fc->fs_private;
1337 struct dentry *root_dentry;
1338 struct ovl_entry *oe;
1339 struct ovl_layer *layers;
1344 if (WARN_ON(fc->user_ns != current_user_ns()))
1347 sb->s_d_op = &ovl_dentry_operations;
1350 ofs->creator_cred = cred = prepare_creds();
1354 err = ovl_fs_params_verify(ctx, &ofs->config);
1360 if (!(fc->sb_flags & SB_SILENT))
1361 pr_err("missing 'lowerdir'\n");
1366 layers = kcalloc(ctx->nr + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1370 ofs->layers = layers;
1371 /* Layer 0 is reserved for upper even if there's no upper */
1374 sb->s_stack_depth = 0;
1375 sb->s_maxbytes = MAX_LFS_FILESIZE;
1376 atomic_long_set(&ofs->last_ino, 1);
1377 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1378 if (ofs->config.xino != OVL_XINO_OFF) {
1379 ofs->xino_mode = BITS_PER_LONG - 32;
1380 if (!ofs->xino_mode) {
1381 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1382 ofs->config.xino = OVL_XINO_OFF;
1386 /* alloc/destroy_inode needed for setting up traps in inode cache */
1387 sb->s_op = &ovl_super_operations;
1389 if (ofs->config.upperdir) {
1390 struct super_block *upper_sb;
1393 if (!ofs->config.workdir) {
1394 pr_err("missing 'workdir'\n");
1398 err = ovl_get_upper(sb, ofs, &layers[0], &ctx->upper);
1402 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
1403 if (!ovl_should_sync(ofs)) {
1404 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
1405 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
1407 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1412 err = ovl_get_workdir(sb, ofs, &ctx->upper, &ctx->work);
1417 sb->s_flags |= SB_RDONLY;
1419 sb->s_stack_depth = upper_sb->s_stack_depth;
1420 sb->s_time_gran = upper_sb->s_time_gran;
1422 oe = ovl_get_lowerstack(sb, ctx, ofs, layers);
1427 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1428 if (!ovl_upper_mnt(ofs))
1429 sb->s_flags |= SB_RDONLY;
1431 if (!ovl_origin_uuid(ofs) && ofs->numfs > 1) {
1432 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=null.\n");
1433 ofs->config.uuid = OVL_UUID_NULL;
1434 } else if (ovl_has_fsid(ofs) && ovl_upper_mnt(ofs)) {
1435 /* Use per instance persistent uuid/fsid */
1436 ovl_init_uuid_xattr(sb, ofs, &ctx->upper);
1439 if (!ovl_force_readonly(ofs) && ofs->config.index) {
1440 err = ovl_get_indexdir(sb, ofs, oe, &ctx->upper);
1444 /* Force r/o mount with no index dir */
1446 sb->s_flags |= SB_RDONLY;
1449 err = ovl_check_overlapping_layers(sb, ofs);
1453 /* Show index=off in /proc/mounts for forced r/o mount */
1454 if (!ofs->indexdir) {
1455 ofs->config.index = false;
1456 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
1457 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1458 ofs->config.nfs_export = false;
1462 if (ofs->config.metacopy && ofs->config.nfs_export) {
1463 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1464 ofs->config.nfs_export = false;
1468 * Support encoding decodable file handles with nfs_export=on
1469 * and encoding non-decodable file handles with nfs_export=off
1470 * if all layers support file handles.
1472 if (ofs->config.nfs_export)
1473 sb->s_export_op = &ovl_export_operations;
1474 else if (!ofs->nofh)
1475 sb->s_export_op = &ovl_export_fid_operations;
1477 /* Never override disk quota limits or use reserved space */
1478 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1480 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1481 sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
1482 ovl_trusted_xattr_handlers;
1483 sb->s_fs_info = ofs;
1484 sb->s_flags |= SB_POSIXACL;
1485 sb->s_iflags |= SB_I_SKIP_SYNC | SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1488 root_dentry = ovl_get_root(sb, ctx->upper.dentry, oe);
1492 sb->s_root = root_dentry;
1500 sb->s_fs_info = NULL;
1504 struct file_system_type ovl_fs_type = {
1505 .owner = THIS_MODULE,
1507 .init_fs_context = ovl_init_fs_context,
1508 .parameters = ovl_parameter_spec,
1509 .fs_flags = FS_USERNS_MOUNT,
1510 .kill_sb = kill_anon_super,
1512 MODULE_ALIAS_FS("overlay");
1514 static void ovl_inode_init_once(void *foo)
1516 struct ovl_inode *oi = foo;
1518 inode_init_once(&oi->vfs_inode);
1521 static int __init ovl_init(void)
1525 ovl_inode_cachep = kmem_cache_create("ovl_inode",
1526 sizeof(struct ovl_inode), 0,
1527 (SLAB_RECLAIM_ACCOUNT|
1528 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1529 ovl_inode_init_once);
1530 if (ovl_inode_cachep == NULL)
1533 err = ovl_aio_request_cache_init();
1535 err = register_filesystem(&ovl_fs_type);
1539 ovl_aio_request_cache_destroy();
1541 kmem_cache_destroy(ovl_inode_cachep);
1546 static void __exit ovl_exit(void)
1548 unregister_filesystem(&ovl_fs_type);
1551 * Make sure all delayed rcu free inodes are flushed before we
1555 kmem_cache_destroy(ovl_inode_cachep);
1556 ovl_aio_request_cache_destroy();
1559 module_init(ovl_init);
1560 module_exit(ovl_exit);