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;
36 /* It's an overlay file */
37 if (inode && d_inode(dentry) == inode)
40 if (!d_is_reg(dentry)) {
41 if (!inode || inode == d_inode(dentry))
46 real = ovl_dentry_upper(dentry);
47 if (real && (inode == d_inode(real)))
50 if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
54 * Best effort lazy lookup of lowerdata for !inode case to return
55 * the real lowerdata dentry. The only current caller of d_real() with
56 * NULL inode is d_real_inode() from trace_uprobe and this caller is
57 * likely going to be followed reading from the file, before placing
58 * uprobes on offset within the file, so lowerdata should be available
59 * when setting the uprobe.
61 ovl_maybe_lookup_lowerdata(dentry);
62 lower = ovl_dentry_lowerdata(dentry);
67 /* Handle recursion */
68 real = d_real(real, inode);
70 if (!inode || inode == d_inode(real))
73 WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
74 __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
75 inode ? inode->i_ino : 0, real,
76 real && d_inode(real) ? d_inode(real)->i_ino : 0);
80 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
88 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
89 ret = d->d_op->d_weak_revalidate(d, flags);
90 } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
91 ret = d->d_op->d_revalidate(d, flags);
93 if (!(flags & LOOKUP_RCU))
101 static int ovl_dentry_revalidate_common(struct dentry *dentry,
102 unsigned int flags, bool weak)
104 struct ovl_entry *oe = OVL_E(dentry);
105 struct ovl_path *lowerstack = ovl_lowerstack(oe);
106 struct inode *inode = d_inode_rcu(dentry);
107 struct dentry *upper;
111 /* Careful in RCU mode */
115 upper = ovl_i_dentry_upper(inode);
117 ret = ovl_revalidate_real(upper, flags, weak);
119 for (i = 0; ret > 0 && i < ovl_numlower(oe); i++)
120 ret = ovl_revalidate_real(lowerstack[i].dentry, flags, weak);
125 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
127 return ovl_dentry_revalidate_common(dentry, flags, false);
130 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
132 return ovl_dentry_revalidate_common(dentry, flags, true);
135 static const struct dentry_operations ovl_dentry_operations = {
136 .d_real = ovl_d_real,
137 .d_revalidate = ovl_dentry_revalidate,
138 .d_weak_revalidate = ovl_dentry_weak_revalidate,
141 static struct kmem_cache *ovl_inode_cachep;
143 static struct inode *ovl_alloc_inode(struct super_block *sb)
145 struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
154 oi->__upperdentry = NULL;
155 oi->lowerdata_redirect = NULL;
157 mutex_init(&oi->lock);
159 return &oi->vfs_inode;
162 static void ovl_free_inode(struct inode *inode)
164 struct ovl_inode *oi = OVL_I(inode);
167 mutex_destroy(&oi->lock);
168 kmem_cache_free(ovl_inode_cachep, oi);
171 static void ovl_destroy_inode(struct inode *inode)
173 struct ovl_inode *oi = OVL_I(inode);
175 dput(oi->__upperdentry);
176 ovl_free_entry(oi->oe);
177 if (S_ISDIR(inode->i_mode))
178 ovl_dir_cache_free(inode);
180 kfree(oi->lowerdata_redirect);
183 static void ovl_put_super(struct super_block *sb)
185 struct ovl_fs *ofs = sb->s_fs_info;
191 /* Sync real dirty inodes in upper filesystem (if it exists) */
192 static int ovl_sync_fs(struct super_block *sb, int wait)
194 struct ovl_fs *ofs = sb->s_fs_info;
195 struct super_block *upper_sb;
198 ret = ovl_sync_status(ofs);
200 * We have to always set the err, because the return value isn't
201 * checked in syncfs, and instead indirectly return an error via
202 * the sb's writeback errseq, which VFS inspects after this call.
205 errseq_set(&sb->s_wb_err, -EIO);
213 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
214 * All the super blocks will be iterated, including upper_sb.
216 * If this is a syncfs(2) call, then we do need to call
217 * sync_filesystem() on upper_sb, but enough if we do it when being
218 * called with wait == 1.
223 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
225 down_read(&upper_sb->s_umount);
226 ret = sync_filesystem(upper_sb);
227 up_read(&upper_sb->s_umount);
234 * @dentry: The dentry to query
235 * @buf: The struct kstatfs to fill in with stats
237 * Get the filesystem statistics. As writes always target the upper layer
238 * filesystem pass the statfs to the upper filesystem (if it exists)
240 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
242 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
243 struct dentry *root_dentry = dentry->d_sb->s_root;
247 ovl_path_real(root_dentry, &path);
249 err = vfs_statfs(&path, buf);
251 buf->f_namelen = ofs->namelen;
252 buf->f_type = OVERLAYFS_SUPER_MAGIC;
258 static const struct super_operations ovl_super_operations = {
259 .alloc_inode = ovl_alloc_inode,
260 .free_inode = ovl_free_inode,
261 .destroy_inode = ovl_destroy_inode,
262 .drop_inode = generic_delete_inode,
263 .put_super = ovl_put_super,
264 .sync_fs = ovl_sync_fs,
265 .statfs = ovl_statfs,
266 .show_options = ovl_show_options,
269 #define OVL_WORKDIR_NAME "work"
270 #define OVL_INDEXDIR_NAME "index"
272 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
273 const char *name, bool persist)
275 struct inode *dir = ofs->workbasedir->d_inode;
276 struct vfsmount *mnt = ovl_upper_mnt(ofs);
279 bool retried = false;
281 inode_lock_nested(dir, I_MUTEX_PARENT);
283 work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
286 struct iattr attr = {
287 .ia_valid = ATTR_MODE,
288 .ia_mode = S_IFDIR | 0,
300 err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
302 if (err == -EINVAL) {
309 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
313 /* Weird filesystem returning with hashed negative (kernfs)? */
315 if (d_really_is_negative(work))
319 * Try to remove POSIX ACL xattrs from workdir. We are good if:
321 * a) success (there was a POSIX ACL xattr and was removed)
322 * b) -ENODATA (there was no POSIX ACL xattr)
323 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
325 * There are various other error values that could effectively
326 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
327 * if the xattr name is too long), but the set of filesystems
328 * allowed as upper are limited to "normal" ones, where checking
329 * for the above two errors is sufficient.
331 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);
332 if (err && err != -ENODATA && err != -EOPNOTSUPP)
335 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);
336 if (err && err != -ENODATA && err != -EOPNOTSUPP)
339 /* Clear any inherited mode bits */
340 inode_lock(work->d_inode);
341 err = ovl_do_notify_change(ofs, work, &attr);
342 inode_unlock(work->d_inode);
356 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
357 ofs->config.workdir, name, -err);
362 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
365 struct kstatfs statfs;
366 int err = vfs_statfs(path, &statfs);
369 pr_err("statfs failed on '%s'\n", name);
371 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
376 static int ovl_lower_dir(const char *name, struct path *path,
377 struct ovl_fs *ofs, int *stack_depth)
382 err = ovl_check_namelen(path, ofs, name);
386 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
389 * The inodes index feature and NFS export need to encode and decode
390 * file handles, so they require that all layers support them.
392 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
393 if ((ofs->config.nfs_export ||
394 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
395 ofs->config.index = false;
396 ofs->config.nfs_export = false;
397 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
401 * Decoding origin file handle is required for persistent st_ino.
402 * Without persistent st_ino, xino=auto falls back to xino=off.
404 if (ofs->config.xino == OVL_XINO_AUTO &&
405 ofs->config.upperdir && !fh_type) {
406 ofs->config.xino = OVL_XINO_OFF;
407 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
411 /* Check if lower fs has 32bit inode numbers */
412 if (fh_type != FILEID_INO32_GEN)
418 /* Workdir should not be subdir of upperdir and vice versa */
419 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
423 if (workdir != upperdir) {
424 ok = (lock_rename(workdir, upperdir) == NULL);
425 unlock_rename(workdir, upperdir);
430 static int ovl_own_xattr_get(const struct xattr_handler *handler,
431 struct dentry *dentry, struct inode *inode,
432 const char *name, void *buffer, size_t size)
437 static int ovl_own_xattr_set(const struct xattr_handler *handler,
438 struct mnt_idmap *idmap,
439 struct dentry *dentry, struct inode *inode,
440 const char *name, const void *value,
441 size_t size, int flags)
446 static int ovl_other_xattr_get(const struct xattr_handler *handler,
447 struct dentry *dentry, struct inode *inode,
448 const char *name, void *buffer, size_t size)
450 return ovl_xattr_get(dentry, inode, name, buffer, size);
453 static int ovl_other_xattr_set(const struct xattr_handler *handler,
454 struct mnt_idmap *idmap,
455 struct dentry *dentry, struct inode *inode,
456 const char *name, const void *value,
457 size_t size, int flags)
459 return ovl_xattr_set(dentry, inode, name, value, size, flags);
462 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
463 .prefix = OVL_XATTR_TRUSTED_PREFIX,
464 .get = ovl_own_xattr_get,
465 .set = ovl_own_xattr_set,
468 static const struct xattr_handler ovl_own_user_xattr_handler = {
469 .prefix = OVL_XATTR_USER_PREFIX,
470 .get = ovl_own_xattr_get,
471 .set = ovl_own_xattr_set,
474 static const struct xattr_handler ovl_other_xattr_handler = {
475 .prefix = "", /* catch all */
476 .get = ovl_other_xattr_get,
477 .set = ovl_other_xattr_set,
480 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
481 &ovl_own_trusted_xattr_handler,
482 &ovl_other_xattr_handler,
486 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
487 &ovl_own_user_xattr_handler,
488 &ovl_other_xattr_handler,
492 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
493 struct inode **ptrap, const char *name)
498 trap = ovl_get_trap_inode(sb, dir);
499 err = PTR_ERR_OR_ZERO(trap);
502 pr_err("conflicting %s path\n", name);
511 * Determine how we treat concurrent use of upperdir/workdir based on the
512 * index feature. This is papering over mount leaks of container runtimes,
513 * for example, an old overlay mount is leaked and now its upperdir is
514 * attempted to be used as a lower layer in a new overlay mount.
516 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
518 if (ofs->config.index) {
519 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
523 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
529 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
530 struct ovl_layer *upper_layer,
531 const struct path *upperpath)
533 struct vfsmount *upper_mnt;
536 /* Upperdir path should not be r/o */
537 if (__mnt_is_readonly(upperpath->mnt)) {
538 pr_err("upper fs is r/o, try multi-lower layers mount\n");
543 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
547 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
552 upper_mnt = clone_private_mount(upperpath);
553 err = PTR_ERR(upper_mnt);
554 if (IS_ERR(upper_mnt)) {
555 pr_err("failed to clone upperpath\n");
559 /* Don't inherit atime flags */
560 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
561 upper_layer->mnt = upper_mnt;
562 upper_layer->idx = 0;
563 upper_layer->fsid = 0;
566 upper_layer->name = kstrdup(ofs->config.upperdir, GFP_KERNEL);
567 if (!upper_layer->name)
571 * Inherit SB_NOSEC flag from upperdir.
573 * This optimization changes behavior when a security related attribute
574 * (suid/sgid/security.*) is changed on an underlying layer. This is
575 * okay because we don't yet have guarantees in that case, but it will
576 * need careful treatment once we want to honour changes to underlying
579 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
580 sb->s_flags |= SB_NOSEC;
582 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
583 ofs->upperdir_locked = true;
585 err = ovl_report_in_use(ofs, "upperdir");
596 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
597 * negative values if error is encountered.
599 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
601 struct dentry *workdir = ofs->workdir;
602 struct inode *dir = d_inode(workdir);
605 struct dentry *whiteout;
606 struct name_snapshot name;
609 inode_lock_nested(dir, I_MUTEX_PARENT);
611 temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
616 dest = ovl_lookup_temp(ofs, workdir);
623 /* Name is inline and stable - using snapshot as a copy helper */
624 take_dentry_name_snapshot(&name, temp);
625 err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
632 whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
633 err = PTR_ERR(whiteout);
634 if (IS_ERR(whiteout))
637 err = ovl_is_whiteout(whiteout);
639 /* Best effort cleanup of whiteout and temp file */
641 ovl_cleanup(ofs, dir, whiteout);
645 ovl_cleanup(ofs, dir, temp);
646 release_dentry_name_snapshot(&name);
656 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
657 struct dentry *parent,
658 const char *name, umode_t mode)
660 size_t len = strlen(name);
661 struct dentry *child;
663 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
664 child = ovl_lookup_upper(ofs, name, parent, len);
665 if (!IS_ERR(child) && !child->d_inode)
666 child = ovl_create_real(ofs, parent->d_inode, child,
668 inode_unlock(parent->d_inode);
675 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
678 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
681 struct dentry *d = dget(ofs->workbasedir);
682 static const char *const volatile_path[] = {
683 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
685 const char *const *name = volatile_path;
687 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
688 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
696 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
697 const struct path *workpath)
699 struct vfsmount *mnt = ovl_upper_mnt(ofs);
700 struct dentry *workdir;
701 struct file *tmpfile;
702 bool rename_whiteout;
707 err = mnt_want_write(mnt);
711 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
712 err = PTR_ERR(workdir);
713 if (IS_ERR_OR_NULL(workdir))
716 ofs->workdir = workdir;
718 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
723 * Upper should support d_type, else whiteouts are visible. Given
724 * workdir and upper are on same fs, we can do iterate_dir() on
725 * workdir. This check requires successful creation of workdir in
728 err = ovl_check_d_type_supported(workpath);
734 pr_warn("upper fs needs to support d_type.\n");
736 /* Check if upper/work fs supports O_TMPFILE */
737 tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
738 ofs->tmpfile = !IS_ERR(tmpfile);
742 pr_warn("upper fs does not support tmpfile.\n");
745 /* Check if upper/work fs supports RENAME_WHITEOUT */
746 err = ovl_check_rename_whiteout(ofs);
750 rename_whiteout = err;
751 if (!rename_whiteout)
752 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
755 * Check if upper/work fs supports (trusted|user).overlay.* xattr
757 err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
759 pr_warn("failed to set xattr on upper\n");
761 if (ovl_redirect_follow(ofs)) {
762 ofs->config.redirect_mode = OVL_REDIRECT_NOFOLLOW;
763 pr_warn("...falling back to redirect_dir=nofollow.\n");
765 if (ofs->config.metacopy) {
766 ofs->config.metacopy = false;
767 pr_warn("...falling back to metacopy=off.\n");
769 if (ofs->config.index) {
770 ofs->config.index = false;
771 pr_warn("...falling back to index=off.\n");
774 * xattr support is required for persistent st_ino.
775 * Without persistent st_ino, xino=auto falls back to xino=off.
777 if (ofs->config.xino == OVL_XINO_AUTO) {
778 ofs->config.xino = OVL_XINO_OFF;
779 pr_warn("...falling back to xino=off.\n");
781 if (err == -EPERM && !ofs->config.userxattr)
782 pr_info("try mounting with 'userxattr' option\n");
785 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
789 * We allowed sub-optimal upper fs configuration and don't want to break
790 * users over kernel upgrade, but we never allowed remote upper fs, so
791 * we can enforce strict requirements for remote upper fs.
793 if (ovl_dentry_remote(ofs->workdir) &&
794 (!d_type || !rename_whiteout || ofs->noxattr)) {
795 pr_err("upper fs missing required features.\n");
801 * For volatile mount, create a incompat/volatile/dirty file to keep
804 if (ofs->config.ovl_volatile) {
805 err = ovl_create_volatile_dirty(ofs);
807 pr_err("Failed to create volatile/dirty file.\n");
812 /* Check if upper/work fs supports file handles */
813 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
814 if (ofs->config.index && !fh_type) {
815 ofs->config.index = false;
816 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
819 /* Check if upper fs has 32bit inode numbers */
820 if (fh_type != FILEID_INO32_GEN)
823 /* NFS export of r/w mount depends on index */
824 if (ofs->config.nfs_export && !ofs->config.index) {
825 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
826 ofs->config.nfs_export = false;
833 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
834 const struct path *upperpath,
835 const struct path *workpath)
840 if (upperpath->mnt != workpath->mnt) {
841 pr_err("workdir and upperdir must reside under the same mount\n");
844 if (!ovl_workdir_ok(workpath->dentry, upperpath->dentry)) {
845 pr_err("workdir and upperdir must be separate subtrees\n");
849 ofs->workbasedir = dget(workpath->dentry);
851 if (ovl_inuse_trylock(ofs->workbasedir)) {
852 ofs->workdir_locked = true;
854 err = ovl_report_in_use(ofs, "workdir");
859 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
864 return ovl_make_workdir(sb, ofs, workpath);
867 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
868 struct ovl_entry *oe, const struct path *upperpath)
870 struct vfsmount *mnt = ovl_upper_mnt(ofs);
871 struct dentry *indexdir;
874 err = mnt_want_write(mnt);
878 /* Verify lower root is upper root origin */
879 err = ovl_verify_origin(ofs, upperpath->dentry,
880 ovl_lowerstack(oe)->dentry, true);
882 pr_err("failed to verify upper root origin\n");
886 /* index dir will act also as workdir */
887 iput(ofs->workdir_trap);
888 ofs->workdir_trap = NULL;
891 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
892 if (IS_ERR(indexdir)) {
893 err = PTR_ERR(indexdir);
894 } else if (indexdir) {
895 ofs->indexdir = indexdir;
896 ofs->workdir = dget(indexdir);
898 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
904 * Verify upper root is exclusively associated with index dir.
905 * Older kernels stored upper fh in ".overlay.origin"
906 * xattr. If that xattr exists, verify that it is a match to
907 * upper dir file handle. In any case, verify or set xattr
908 * ".overlay.upper" to indicate that index may have
911 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
912 err = ovl_verify_set_fh(ofs, ofs->indexdir,
914 upperpath->dentry, true, false);
916 pr_err("failed to verify index dir 'origin' xattr\n");
918 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
921 pr_err("failed to verify index dir 'upper' xattr\n");
923 /* Cleanup bad/stale/orphan index entries */
925 err = ovl_indexdir_cleanup(ofs);
927 if (err || !ofs->indexdir)
928 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
935 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
939 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
943 * We allow using single lower with null uuid for index and nfs_export
944 * for example to support those features with single lower squashfs.
945 * To avoid regressions in setups of overlay with re-formatted lower
946 * squashfs, do not allow decoding origin with lower null uuid unless
947 * user opted-in to one of the new features that require following the
948 * lower inode of non-dir upper.
950 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
953 for (i = 0; i < ofs->numfs; i++) {
955 * We use uuid to associate an overlay lower file handle with a
956 * lower layer, so we can accept lower fs with null uuid as long
957 * as all lower layers with null uuid are on the same fs.
958 * if we detect multiple lower fs with the same uuid, we
959 * disable lower file handle decoding on all of them.
961 if (ofs->fs[i].is_lower &&
962 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
963 ofs->fs[i].bad_uuid = true;
970 /* Get a unique fsid for the layer */
971 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
973 struct super_block *sb = path->mnt->mnt_sb;
977 bool bad_uuid = false;
980 for (i = 0; i < ofs->numfs; i++) {
981 if (ofs->fs[i].sb == sb)
985 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
987 if (ofs->config.xino == OVL_XINO_AUTO) {
988 ofs->config.xino = OVL_XINO_OFF;
991 if (ofs->config.index || ofs->config.nfs_export) {
992 ofs->config.index = false;
993 ofs->config.nfs_export = false;
997 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
998 uuid_is_null(&sb->s_uuid) ? "null" :
1000 path->dentry, ovl_xino_mode(&ofs->config));
1004 err = get_anon_bdev(&dev);
1006 pr_err("failed to get anonymous bdev for lowerpath\n");
1010 ofs->fs[ofs->numfs].sb = sb;
1011 ofs->fs[ofs->numfs].pseudo_dev = dev;
1012 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1014 return ofs->numfs++;
1018 * The fsid after the last lower fsid is used for the data layers.
1019 * It is a "null fs" with a null sb, null uuid, and no pseudo dev.
1021 static int ovl_get_data_fsid(struct ovl_fs *ofs)
1027 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1028 struct ovl_fs_context *ctx, struct ovl_layer *layers)
1032 size_t nr_merged_lower;
1034 ofs->fs = kcalloc(ctx->nr + 2, sizeof(struct ovl_sb), GFP_KERNEL);
1035 if (ofs->fs == NULL)
1039 * idx/fsid 0 are reserved for upper fs even with lower only overlay
1040 * and the last fsid is reserved for "null fs" of the data layers.
1045 * All lower layers that share the same fs as upper layer, use the same
1046 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1047 * only overlay to simplify ovl_fs_free().
1048 * is_lower will be set if upper fs is shared with a lower layer.
1050 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1052 pr_err("failed to get anonymous bdev for upper fs\n");
1056 if (ovl_upper_mnt(ofs)) {
1057 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1058 ofs->fs[0].is_lower = false;
1061 nr_merged_lower = ctx->nr - ctx->nr_data;
1062 for (i = 0; i < ctx->nr; i++) {
1063 struct ovl_fs_context_layer *l = &ctx->lower[i];
1064 struct vfsmount *mnt;
1068 if (i < nr_merged_lower)
1069 fsid = ovl_get_fsid(ofs, &l->path);
1071 fsid = ovl_get_data_fsid(ofs);
1076 * Check if lower root conflicts with this overlay layers before
1077 * checking if it is in-use as upperdir/workdir of "another"
1078 * mount, because we do not bother to check in ovl_is_inuse() if
1079 * the upperdir/workdir is in fact in-use by our
1082 err = ovl_setup_trap(sb, l->path.dentry, &trap, "lowerdir");
1086 if (ovl_is_inuse(l->path.dentry)) {
1087 err = ovl_report_in_use(ofs, "lowerdir");
1094 mnt = clone_private_mount(&l->path);
1097 pr_err("failed to clone lowerpath\n");
1103 * Make lower layers R/O. That way fchmod/fchown on lower file
1104 * will fail instead of modifying lower fs.
1106 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1108 layers[ofs->numlayer].trap = trap;
1109 layers[ofs->numlayer].mnt = mnt;
1110 layers[ofs->numlayer].idx = ofs->numlayer;
1111 layers[ofs->numlayer].fsid = fsid;
1112 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1113 layers[ofs->numlayer].name = l->name;
1116 ofs->fs[fsid].is_lower = true;
1120 * When all layers on same fs, overlay can use real inode numbers.
1121 * With mount option "xino=<on|auto>", mounter declares that there are
1122 * enough free high bits in underlying fs to hold the unique fsid.
1123 * If overlayfs does encounter underlying inodes using the high xino
1124 * bits reserved for fsid, it emits a warning and uses the original
1125 * inode number or a non persistent inode number allocated from a
1128 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1129 if (ofs->config.xino == OVL_XINO_ON)
1130 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1132 } else if (ofs->config.xino == OVL_XINO_OFF) {
1133 ofs->xino_mode = -1;
1134 } else if (ofs->xino_mode < 0) {
1136 * This is a roundup of number of bits needed for encoding
1137 * fsid, where fsid 0 is reserved for upper fs (even with
1138 * lower only overlay) +1 extra bit is reserved for the non
1139 * persistent inode number range that is used for resolving
1140 * xino lower bits overflow.
1142 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1143 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1146 if (ofs->xino_mode > 0) {
1147 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1154 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1155 struct ovl_fs_context *ctx,
1157 struct ovl_layer *layers)
1161 size_t nr_merged_lower;
1162 struct ovl_entry *oe;
1163 struct ovl_path *lowerstack;
1165 struct ovl_fs_context_layer *l;
1167 if (!ofs->config.upperdir && ctx->nr == 1) {
1168 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1169 return ERR_PTR(-EINVAL);
1173 for (i = 0; i < ctx->nr; i++) {
1176 err = ovl_lower_dir(l->name, &l->path, ofs, &sb->s_stack_depth);
1178 return ERR_PTR(err);
1182 sb->s_stack_depth++;
1183 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1184 pr_err("maximum fs stacking depth exceeded\n");
1185 return ERR_PTR(err);
1188 err = ovl_get_layers(sb, ofs, ctx, layers);
1190 return ERR_PTR(err);
1193 /* Data-only layers are not merged in root directory */
1194 nr_merged_lower = ctx->nr - ctx->nr_data;
1195 oe = ovl_alloc_entry(nr_merged_lower);
1197 return ERR_PTR(err);
1199 lowerstack = ovl_lowerstack(oe);
1200 for (i = 0; i < nr_merged_lower; i++) {
1202 lowerstack[i].dentry = dget(l->path.dentry);
1203 lowerstack[i].layer = &ofs->layers[i + 1];
1205 ofs->numdatalayer = ctx->nr_data;
1211 * Check if this layer root is a descendant of:
1212 * - another layer of this overlayfs instance
1213 * - upper/work dir of any overlayfs instance
1215 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1216 struct dentry *dentry, const char *name,
1219 struct dentry *next = dentry, *parent;
1225 parent = dget_parent(next);
1227 /* Walk back ancestors to root (inclusive) looking for traps */
1228 while (!err && parent != next) {
1229 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1231 pr_err("overlapping %s path\n", name);
1232 } else if (ovl_is_inuse(parent)) {
1233 err = ovl_report_in_use(ofs, name);
1236 parent = dget_parent(next);
1246 * Check if any of the layers or work dirs overlap.
1248 static int ovl_check_overlapping_layers(struct super_block *sb,
1253 if (ovl_upper_mnt(ofs)) {
1254 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1260 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1261 * this instance and covers overlapping work and index dirs,
1262 * unless work or index dir have been moved since created inside
1263 * workbasedir. In that case, we already have their traps in
1264 * inode cache and we will catch that case on lookup.
1266 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1272 for (i = 1; i < ofs->numlayer; i++) {
1273 err = ovl_check_layer(sb, ofs,
1274 ofs->layers[i].mnt->mnt_root,
1283 static struct dentry *ovl_get_root(struct super_block *sb,
1284 struct dentry *upperdentry,
1285 struct ovl_entry *oe)
1287 struct dentry *root;
1288 struct ovl_path *lowerpath = ovl_lowerstack(oe);
1289 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1290 int fsid = lowerpath->layer->fsid;
1291 struct ovl_inode_params oip = {
1292 .upperdentry = upperdentry,
1296 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1301 /* Root inode uses upper st_ino/i_ino */
1302 ino = d_inode(upperdentry)->i_ino;
1304 ovl_dentry_set_upper_alias(root);
1305 if (ovl_is_impuredir(sb, upperdentry))
1306 ovl_set_flag(OVL_IMPURE, d_inode(root));
1309 /* Root is always merge -> can have whiteouts */
1310 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1311 ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1312 ovl_set_upperdata(d_inode(root));
1313 ovl_inode_init(d_inode(root), &oip, ino, fsid);
1314 ovl_dentry_init_flags(root, upperdentry, oe, DCACHE_OP_WEAK_REVALIDATE);
1315 /* root keeps a reference of upperdentry */
1321 int ovl_fill_super(struct super_block *sb, struct fs_context *fc)
1323 struct ovl_fs *ofs = sb->s_fs_info;
1324 struct ovl_fs_context *ctx = fc->fs_private;
1325 struct dentry *root_dentry;
1326 struct ovl_entry *oe;
1327 struct ovl_layer *layers;
1332 if (WARN_ON(fc->user_ns != current_user_ns()))
1335 sb->s_d_op = &ovl_dentry_operations;
1338 ofs->creator_cred = cred = prepare_creds();
1342 err = ovl_fs_params_verify(ctx, &ofs->config);
1348 if (!(fc->sb_flags & SB_SILENT))
1349 pr_err("missing 'lowerdir'\n");
1354 layers = kcalloc(ctx->nr + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1358 ofs->layers = layers;
1359 /* Layer 0 is reserved for upper even if there's no upper */
1362 sb->s_stack_depth = 0;
1363 sb->s_maxbytes = MAX_LFS_FILESIZE;
1364 atomic_long_set(&ofs->last_ino, 1);
1365 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1366 if (ofs->config.xino != OVL_XINO_OFF) {
1367 ofs->xino_mode = BITS_PER_LONG - 32;
1368 if (!ofs->xino_mode) {
1369 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1370 ofs->config.xino = OVL_XINO_OFF;
1374 /* alloc/destroy_inode needed for setting up traps in inode cache */
1375 sb->s_op = &ovl_super_operations;
1377 if (ofs->config.upperdir) {
1378 struct super_block *upper_sb;
1381 if (!ofs->config.workdir) {
1382 pr_err("missing 'workdir'\n");
1386 err = ovl_get_upper(sb, ofs, &layers[0], &ctx->upper);
1390 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
1391 if (!ovl_should_sync(ofs)) {
1392 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
1393 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
1395 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1400 err = ovl_get_workdir(sb, ofs, &ctx->upper, &ctx->work);
1405 sb->s_flags |= SB_RDONLY;
1407 sb->s_stack_depth = upper_sb->s_stack_depth;
1408 sb->s_time_gran = upper_sb->s_time_gran;
1410 oe = ovl_get_lowerstack(sb, ctx, ofs, layers);
1415 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1416 if (!ovl_upper_mnt(ofs))
1417 sb->s_flags |= SB_RDONLY;
1419 if (!ofs->config.uuid && ofs->numfs > 1) {
1420 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
1421 ofs->config.uuid = true;
1424 if (!ovl_force_readonly(ofs) && ofs->config.index) {
1425 err = ovl_get_indexdir(sb, ofs, oe, &ctx->upper);
1429 /* Force r/o mount with no index dir */
1431 sb->s_flags |= SB_RDONLY;
1434 err = ovl_check_overlapping_layers(sb, ofs);
1438 /* Show index=off in /proc/mounts for forced r/o mount */
1439 if (!ofs->indexdir) {
1440 ofs->config.index = false;
1441 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
1442 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1443 ofs->config.nfs_export = false;
1447 if (ofs->config.metacopy && ofs->config.nfs_export) {
1448 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1449 ofs->config.nfs_export = false;
1452 if (ofs->config.nfs_export)
1453 sb->s_export_op = &ovl_export_operations;
1455 /* Never override disk quota limits or use reserved space */
1456 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1458 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1459 sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
1460 ovl_trusted_xattr_handlers;
1461 sb->s_fs_info = ofs;
1462 sb->s_flags |= SB_POSIXACL;
1463 sb->s_iflags |= SB_I_SKIP_SYNC | SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1466 root_dentry = ovl_get_root(sb, ctx->upper.dentry, oe);
1470 sb->s_root = root_dentry;
1478 sb->s_fs_info = NULL;
1482 static struct file_system_type ovl_fs_type = {
1483 .owner = THIS_MODULE,
1485 .init_fs_context = ovl_init_fs_context,
1486 .parameters = ovl_parameter_spec,
1487 .fs_flags = FS_USERNS_MOUNT,
1488 .kill_sb = kill_anon_super,
1490 MODULE_ALIAS_FS("overlay");
1492 static void ovl_inode_init_once(void *foo)
1494 struct ovl_inode *oi = foo;
1496 inode_init_once(&oi->vfs_inode);
1499 static int __init ovl_init(void)
1503 ovl_inode_cachep = kmem_cache_create("ovl_inode",
1504 sizeof(struct ovl_inode), 0,
1505 (SLAB_RECLAIM_ACCOUNT|
1506 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1507 ovl_inode_init_once);
1508 if (ovl_inode_cachep == NULL)
1511 err = ovl_aio_request_cache_init();
1513 err = register_filesystem(&ovl_fs_type);
1517 ovl_aio_request_cache_destroy();
1519 kmem_cache_destroy(ovl_inode_cachep);
1524 static void __exit ovl_exit(void)
1526 unregister_filesystem(&ovl_fs_type);
1529 * Make sure all delayed rcu free inodes are flushed before we
1533 kmem_cache_destroy(ovl_inode_cachep);
1534 ovl_aio_request_cache_destroy();
1537 module_init(ovl_init);
1538 module_exit(ovl_exit);