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;
38 * vfs is only expected to call d_real() with NULL from d_real_inode()
39 * and with overlay inode from file_dentry() on an overlay file.
41 * TODO: remove @inode argument from d_real() API, remove code in this
42 * function that deals with non-NULL @inode and remove d_real() call
45 if (inode && d_inode(dentry) == inode)
50 if (!d_is_reg(dentry)) {
51 /* d_real_inode() is only relevant for regular files */
55 real = ovl_dentry_upper(dentry);
56 if (real && (inode == d_inode(real)))
59 if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
63 * Best effort lazy lookup of lowerdata for !inode case to return
64 * the real lowerdata dentry. The only current caller of d_real() with
65 * NULL inode is d_real_inode() from trace_uprobe and this caller is
66 * likely going to be followed reading from the file, before placing
67 * uprobes on offset within the file, so lowerdata should be available
68 * when setting the uprobe.
70 err = ovl_verify_lowerdata(dentry);
73 lower = ovl_dentry_lowerdata(dentry);
78 /* Handle recursion */
79 real = d_real(real, inode);
81 if (!inode || inode == d_inode(real))
84 WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
85 __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
86 inode ? inode->i_ino : 0, real,
87 real && d_inode(real) ? d_inode(real)->i_ino : 0);
91 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
99 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
100 ret = d->d_op->d_weak_revalidate(d, flags);
101 } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
102 ret = d->d_op->d_revalidate(d, flags);
104 if (!(flags & LOOKUP_RCU))
112 static int ovl_dentry_revalidate_common(struct dentry *dentry,
113 unsigned int flags, bool weak)
115 struct ovl_entry *oe;
116 struct ovl_path *lowerstack;
117 struct inode *inode = d_inode_rcu(dentry);
118 struct dentry *upper;
122 /* Careful in RCU mode */
127 lowerstack = ovl_lowerstack(oe);
128 upper = ovl_i_dentry_upper(inode);
130 ret = ovl_revalidate_real(upper, flags, weak);
132 for (i = 0; ret > 0 && i < ovl_numlower(oe); i++)
133 ret = ovl_revalidate_real(lowerstack[i].dentry, flags, weak);
138 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
140 return ovl_dentry_revalidate_common(dentry, flags, false);
143 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
145 return ovl_dentry_revalidate_common(dentry, flags, true);
148 static const struct dentry_operations ovl_dentry_operations = {
149 .d_real = ovl_d_real,
150 .d_revalidate = ovl_dentry_revalidate,
151 .d_weak_revalidate = ovl_dentry_weak_revalidate,
154 static struct kmem_cache *ovl_inode_cachep;
156 static struct inode *ovl_alloc_inode(struct super_block *sb)
158 struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
167 oi->__upperdentry = NULL;
168 oi->lowerdata_redirect = NULL;
170 mutex_init(&oi->lock);
172 return &oi->vfs_inode;
175 static void ovl_free_inode(struct inode *inode)
177 struct ovl_inode *oi = OVL_I(inode);
181 mutex_destroy(&oi->lock);
182 kmem_cache_free(ovl_inode_cachep, oi);
185 static void ovl_destroy_inode(struct inode *inode)
187 struct ovl_inode *oi = OVL_I(inode);
189 dput(oi->__upperdentry);
190 ovl_stack_put(ovl_lowerstack(oi->oe), ovl_numlower(oi->oe));
191 if (S_ISDIR(inode->i_mode))
192 ovl_dir_cache_free(inode);
194 kfree(oi->lowerdata_redirect);
197 static void ovl_put_super(struct super_block *sb)
199 struct ovl_fs *ofs = OVL_FS(sb);
205 /* Sync real dirty inodes in upper filesystem (if it exists) */
206 static int ovl_sync_fs(struct super_block *sb, int wait)
208 struct ovl_fs *ofs = OVL_FS(sb);
209 struct super_block *upper_sb;
212 ret = ovl_sync_status(ofs);
214 * We have to always set the err, because the return value isn't
215 * checked in syncfs, and instead indirectly return an error via
216 * the sb's writeback errseq, which VFS inspects after this call.
219 errseq_set(&sb->s_wb_err, -EIO);
227 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
228 * All the super blocks will be iterated, including upper_sb.
230 * If this is a syncfs(2) call, then we do need to call
231 * sync_filesystem() on upper_sb, but enough if we do it when being
232 * called with wait == 1.
237 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
239 down_read(&upper_sb->s_umount);
240 ret = sync_filesystem(upper_sb);
241 up_read(&upper_sb->s_umount);
248 * @dentry: The dentry to query
249 * @buf: The struct kstatfs to fill in with stats
251 * Get the filesystem statistics. As writes always target the upper layer
252 * filesystem pass the statfs to the upper filesystem (if it exists)
254 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
256 struct super_block *sb = dentry->d_sb;
257 struct ovl_fs *ofs = OVL_FS(sb);
258 struct dentry *root_dentry = sb->s_root;
262 ovl_path_real(root_dentry, &path);
264 err = vfs_statfs(&path, buf);
266 buf->f_namelen = ofs->namelen;
267 buf->f_type = OVERLAYFS_SUPER_MAGIC;
268 if (ovl_has_fsid(ofs))
269 buf->f_fsid = uuid_to_fsid(sb->s_uuid.b);
275 static const struct super_operations ovl_super_operations = {
276 .alloc_inode = ovl_alloc_inode,
277 .free_inode = ovl_free_inode,
278 .destroy_inode = ovl_destroy_inode,
279 .drop_inode = generic_delete_inode,
280 .put_super = ovl_put_super,
281 .sync_fs = ovl_sync_fs,
282 .statfs = ovl_statfs,
283 .show_options = ovl_show_options,
286 #define OVL_WORKDIR_NAME "work"
287 #define OVL_INDEXDIR_NAME "index"
289 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
290 const char *name, bool persist)
292 struct inode *dir = ofs->workbasedir->d_inode;
293 struct vfsmount *mnt = ovl_upper_mnt(ofs);
296 bool retried = false;
298 inode_lock_nested(dir, I_MUTEX_PARENT);
300 work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
303 struct iattr attr = {
304 .ia_valid = ATTR_MODE,
305 .ia_mode = S_IFDIR | 0,
317 err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
319 if (err == -EINVAL) {
326 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
330 /* Weird filesystem returning with hashed negative (kernfs)? */
332 if (d_really_is_negative(work))
336 * Try to remove POSIX ACL xattrs from workdir. We are good if:
338 * a) success (there was a POSIX ACL xattr and was removed)
339 * b) -ENODATA (there was no POSIX ACL xattr)
340 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
342 * There are various other error values that could effectively
343 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
344 * if the xattr name is too long), but the set of filesystems
345 * allowed as upper are limited to "normal" ones, where checking
346 * for the above two errors is sufficient.
348 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);
349 if (err && err != -ENODATA && err != -EOPNOTSUPP)
352 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);
353 if (err && err != -ENODATA && err != -EOPNOTSUPP)
356 /* Clear any inherited mode bits */
357 inode_lock(work->d_inode);
358 err = ovl_do_notify_change(ofs, work, &attr);
359 inode_unlock(work->d_inode);
373 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
374 ofs->config.workdir, name, -err);
379 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
382 struct kstatfs statfs;
383 int err = vfs_statfs(path, &statfs);
386 pr_err("statfs failed on '%s'\n", name);
388 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
393 static int ovl_lower_dir(const char *name, struct path *path,
394 struct ovl_fs *ofs, int *stack_depth)
399 err = ovl_check_namelen(path, ofs, name);
403 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
406 * The inodes index feature and NFS export need to encode and decode
407 * file handles, so they require that all layers support them.
409 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
410 if ((ofs->config.nfs_export ||
411 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
412 ofs->config.index = false;
413 ofs->config.nfs_export = false;
414 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
417 ofs->nofh |= !fh_type;
419 * Decoding origin file handle is required for persistent st_ino.
420 * Without persistent st_ino, xino=auto falls back to xino=off.
422 if (ofs->config.xino == OVL_XINO_AUTO &&
423 ofs->config.upperdir && !fh_type) {
424 ofs->config.xino = OVL_XINO_OFF;
425 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
429 /* Check if lower fs has 32bit inode numbers */
430 if (fh_type != FILEID_INO32_GEN)
436 /* Workdir should not be subdir of upperdir and vice versa */
437 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
441 if (workdir != upperdir) {
442 ok = (lock_rename(workdir, upperdir) == NULL);
443 unlock_rename(workdir, upperdir);
448 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
449 struct inode **ptrap, const char *name)
454 trap = ovl_get_trap_inode(sb, dir);
455 err = PTR_ERR_OR_ZERO(trap);
458 pr_err("conflicting %s path\n", name);
467 * Determine how we treat concurrent use of upperdir/workdir based on the
468 * index feature. This is papering over mount leaks of container runtimes,
469 * for example, an old overlay mount is leaked and now its upperdir is
470 * attempted to be used as a lower layer in a new overlay mount.
472 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
474 if (ofs->config.index) {
475 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
479 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
485 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
486 struct ovl_layer *upper_layer,
487 const struct path *upperpath)
489 struct vfsmount *upper_mnt;
492 /* Upperdir path should not be r/o */
493 if (__mnt_is_readonly(upperpath->mnt)) {
494 pr_err("upper fs is r/o, try multi-lower layers mount\n");
499 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
503 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
508 upper_mnt = clone_private_mount(upperpath);
509 err = PTR_ERR(upper_mnt);
510 if (IS_ERR(upper_mnt)) {
511 pr_err("failed to clone upperpath\n");
515 /* Don't inherit atime flags */
516 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
517 upper_layer->mnt = upper_mnt;
518 upper_layer->idx = 0;
519 upper_layer->fsid = 0;
522 * Inherit SB_NOSEC flag from upperdir.
524 * This optimization changes behavior when a security related attribute
525 * (suid/sgid/security.*) is changed on an underlying layer. This is
526 * okay because we don't yet have guarantees in that case, but it will
527 * need careful treatment once we want to honour changes to underlying
530 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
531 sb->s_flags |= SB_NOSEC;
533 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
534 ofs->upperdir_locked = true;
536 err = ovl_report_in_use(ofs, "upperdir");
547 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
548 * negative values if error is encountered.
550 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
552 struct dentry *workdir = ofs->workdir;
553 struct inode *dir = d_inode(workdir);
556 struct dentry *whiteout;
557 struct name_snapshot name;
560 inode_lock_nested(dir, I_MUTEX_PARENT);
562 temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
567 dest = ovl_lookup_temp(ofs, workdir);
574 /* Name is inline and stable - using snapshot as a copy helper */
575 take_dentry_name_snapshot(&name, temp);
576 err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
583 whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
584 err = PTR_ERR(whiteout);
585 if (IS_ERR(whiteout))
588 err = ovl_upper_is_whiteout(ofs, whiteout);
590 /* Best effort cleanup of whiteout and temp file */
592 ovl_cleanup(ofs, dir, whiteout);
596 ovl_cleanup(ofs, dir, temp);
597 release_dentry_name_snapshot(&name);
607 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
608 struct dentry *parent,
609 const char *name, umode_t mode)
611 size_t len = strlen(name);
612 struct dentry *child;
614 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
615 child = ovl_lookup_upper(ofs, name, parent, len);
616 if (!IS_ERR(child) && !child->d_inode)
617 child = ovl_create_real(ofs, parent->d_inode, child,
619 inode_unlock(parent->d_inode);
626 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
629 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
632 struct dentry *d = dget(ofs->workbasedir);
633 static const char *const volatile_path[] = {
634 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
636 const char *const *name = volatile_path;
638 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
639 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
647 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
648 const struct path *workpath)
650 struct vfsmount *mnt = ovl_upper_mnt(ofs);
651 struct dentry *workdir;
652 struct file *tmpfile;
653 bool rename_whiteout;
658 err = mnt_want_write(mnt);
662 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
663 err = PTR_ERR(workdir);
664 if (IS_ERR_OR_NULL(workdir))
667 ofs->workdir = workdir;
669 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
674 * Upper should support d_type, else whiteouts are visible. Given
675 * workdir and upper are on same fs, we can do iterate_dir() on
676 * workdir. This check requires successful creation of workdir in
679 err = ovl_check_d_type_supported(workpath);
685 pr_warn("upper fs needs to support d_type.\n");
687 /* Check if upper/work fs supports O_TMPFILE */
688 tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
689 ofs->tmpfile = !IS_ERR(tmpfile);
693 pr_warn("upper fs does not support tmpfile.\n");
696 /* Check if upper/work fs supports RENAME_WHITEOUT */
697 err = ovl_check_rename_whiteout(ofs);
701 rename_whiteout = err;
702 if (!rename_whiteout)
703 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
706 * Check if upper/work fs supports (trusted|user).overlay.* xattr
708 err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
710 pr_warn("failed to set xattr on upper\n");
712 if (ovl_redirect_follow(ofs)) {
713 ofs->config.redirect_mode = OVL_REDIRECT_NOFOLLOW;
714 pr_warn("...falling back to redirect_dir=nofollow.\n");
716 if (ofs->config.metacopy) {
717 ofs->config.metacopy = false;
718 pr_warn("...falling back to metacopy=off.\n");
720 if (ofs->config.index) {
721 ofs->config.index = false;
722 pr_warn("...falling back to index=off.\n");
724 if (ovl_has_fsid(ofs)) {
725 ofs->config.uuid = OVL_UUID_NULL;
726 pr_warn("...falling back to uuid=null.\n");
729 * xattr support is required for persistent st_ino.
730 * Without persistent st_ino, xino=auto falls back to xino=off.
732 if (ofs->config.xino == OVL_XINO_AUTO) {
733 ofs->config.xino = OVL_XINO_OFF;
734 pr_warn("...falling back to xino=off.\n");
736 if (err == -EPERM && !ofs->config.userxattr)
737 pr_info("try mounting with 'userxattr' option\n");
740 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
744 * We allowed sub-optimal upper fs configuration and don't want to break
745 * users over kernel upgrade, but we never allowed remote upper fs, so
746 * we can enforce strict requirements for remote upper fs.
748 if (ovl_dentry_remote(ofs->workdir) &&
749 (!d_type || !rename_whiteout || ofs->noxattr)) {
750 pr_err("upper fs missing required features.\n");
756 * For volatile mount, create a incompat/volatile/dirty file to keep
759 if (ofs->config.ovl_volatile) {
760 err = ovl_create_volatile_dirty(ofs);
762 pr_err("Failed to create volatile/dirty file.\n");
767 /* Check if upper/work fs supports file handles */
768 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
769 if (ofs->config.index && !fh_type) {
770 ofs->config.index = false;
771 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
773 ofs->nofh |= !fh_type;
775 /* Check if upper fs has 32bit inode numbers */
776 if (fh_type != FILEID_INO32_GEN)
779 /* NFS export of r/w mount depends on index */
780 if (ofs->config.nfs_export && !ofs->config.index) {
781 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
782 ofs->config.nfs_export = false;
789 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
790 const struct path *upperpath,
791 const struct path *workpath)
796 if (upperpath->mnt != workpath->mnt) {
797 pr_err("workdir and upperdir must reside under the same mount\n");
800 if (!ovl_workdir_ok(workpath->dentry, upperpath->dentry)) {
801 pr_err("workdir and upperdir must be separate subtrees\n");
805 ofs->workbasedir = dget(workpath->dentry);
807 if (ovl_inuse_trylock(ofs->workbasedir)) {
808 ofs->workdir_locked = true;
810 err = ovl_report_in_use(ofs, "workdir");
815 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
820 return ovl_make_workdir(sb, ofs, workpath);
823 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
824 struct ovl_entry *oe, const struct path *upperpath)
826 struct vfsmount *mnt = ovl_upper_mnt(ofs);
827 struct dentry *indexdir;
828 struct dentry *origin = ovl_lowerstack(oe)->dentry;
829 const struct ovl_fh *fh;
832 fh = ovl_get_origin_fh(ofs, origin);
836 err = mnt_want_write(mnt);
840 /* Verify lower root is upper root origin */
841 err = ovl_verify_origin_fh(ofs, upperpath->dentry, fh, true);
843 pr_err("failed to verify upper root origin\n");
847 /* index dir will act also as workdir */
848 iput(ofs->workdir_trap);
849 ofs->workdir_trap = NULL;
852 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
853 if (IS_ERR(indexdir)) {
854 err = PTR_ERR(indexdir);
855 } else if (indexdir) {
856 ofs->indexdir = indexdir;
857 ofs->workdir = dget(indexdir);
859 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
865 * Verify upper root is exclusively associated with index dir.
866 * Older kernels stored upper fh in ".overlay.origin"
867 * xattr. If that xattr exists, verify that it is a match to
868 * upper dir file handle. In any case, verify or set xattr
869 * ".overlay.upper" to indicate that index may have
872 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
873 err = ovl_verify_origin_xattr(ofs, ofs->indexdir,
875 upperpath->dentry, true,
878 pr_err("failed to verify index dir 'origin' xattr\n");
880 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
883 pr_err("failed to verify index dir 'upper' xattr\n");
885 /* Cleanup bad/stale/orphan index entries */
887 err = ovl_indexdir_cleanup(ofs);
889 if (err || !ofs->indexdir)
890 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
899 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
903 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
907 * We allow using single lower with null uuid for index and nfs_export
908 * for example to support those features with single lower squashfs.
909 * To avoid regressions in setups of overlay with re-formatted lower
910 * squashfs, do not allow decoding origin with lower null uuid unless
911 * user opted-in to one of the new features that require following the
912 * lower inode of non-dir upper.
914 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
917 for (i = 0; i < ofs->numfs; i++) {
919 * We use uuid to associate an overlay lower file handle with a
920 * lower layer, so we can accept lower fs with null uuid as long
921 * as all lower layers with null uuid are on the same fs.
922 * if we detect multiple lower fs with the same uuid, we
923 * disable lower file handle decoding on all of them.
925 if (ofs->fs[i].is_lower &&
926 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
927 ofs->fs[i].bad_uuid = true;
934 /* Get a unique fsid for the layer */
935 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
937 struct super_block *sb = path->mnt->mnt_sb;
941 bool bad_uuid = false;
944 for (i = 0; i < ofs->numfs; i++) {
945 if (ofs->fs[i].sb == sb)
949 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
951 if (ofs->config.xino == OVL_XINO_AUTO) {
952 ofs->config.xino = OVL_XINO_OFF;
955 if (ofs->config.index || ofs->config.nfs_export) {
956 ofs->config.index = false;
957 ofs->config.nfs_export = false;
961 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
962 uuid_is_null(&sb->s_uuid) ? "null" :
964 path->dentry, ovl_xino_mode(&ofs->config));
968 err = get_anon_bdev(&dev);
970 pr_err("failed to get anonymous bdev for lowerpath\n");
974 ofs->fs[ofs->numfs].sb = sb;
975 ofs->fs[ofs->numfs].pseudo_dev = dev;
976 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
982 * The fsid after the last lower fsid is used for the data layers.
983 * It is a "null fs" with a null sb, null uuid, and no pseudo dev.
985 static int ovl_get_data_fsid(struct ovl_fs *ofs)
991 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
992 struct ovl_fs_context *ctx, struct ovl_layer *layers)
996 size_t nr_merged_lower;
998 ofs->fs = kcalloc(ctx->nr + 2, sizeof(struct ovl_sb), GFP_KERNEL);
1003 * idx/fsid 0 are reserved for upper fs even with lower only overlay
1004 * and the last fsid is reserved for "null fs" of the data layers.
1009 * All lower layers that share the same fs as upper layer, use the same
1010 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1011 * only overlay to simplify ovl_fs_free().
1012 * is_lower will be set if upper fs is shared with a lower layer.
1014 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1016 pr_err("failed to get anonymous bdev for upper fs\n");
1020 if (ovl_upper_mnt(ofs)) {
1021 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1022 ofs->fs[0].is_lower = false;
1025 nr_merged_lower = ctx->nr - ctx->nr_data;
1026 for (i = 0; i < ctx->nr; i++) {
1027 struct ovl_fs_context_layer *l = &ctx->lower[i];
1028 struct vfsmount *mnt;
1032 if (i < nr_merged_lower)
1033 fsid = ovl_get_fsid(ofs, &l->path);
1035 fsid = ovl_get_data_fsid(ofs);
1040 * Check if lower root conflicts with this overlay layers before
1041 * checking if it is in-use as upperdir/workdir of "another"
1042 * mount, because we do not bother to check in ovl_is_inuse() if
1043 * the upperdir/workdir is in fact in-use by our
1046 err = ovl_setup_trap(sb, l->path.dentry, &trap, "lowerdir");
1050 if (ovl_is_inuse(l->path.dentry)) {
1051 err = ovl_report_in_use(ofs, "lowerdir");
1058 mnt = clone_private_mount(&l->path);
1061 pr_err("failed to clone lowerpath\n");
1067 * Make lower layers R/O. That way fchmod/fchown on lower file
1068 * will fail instead of modifying lower fs.
1070 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1072 layers[ofs->numlayer].trap = trap;
1073 layers[ofs->numlayer].mnt = mnt;
1074 layers[ofs->numlayer].idx = ofs->numlayer;
1075 layers[ofs->numlayer].fsid = fsid;
1076 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1077 /* Store for printing lowerdir=... in ovl_show_options() */
1078 ofs->config.lowerdirs[ofs->numlayer] = l->name;
1081 ofs->fs[fsid].is_lower = true;
1085 * When all layers on same fs, overlay can use real inode numbers.
1086 * With mount option "xino=<on|auto>", mounter declares that there are
1087 * enough free high bits in underlying fs to hold the unique fsid.
1088 * If overlayfs does encounter underlying inodes using the high xino
1089 * bits reserved for fsid, it emits a warning and uses the original
1090 * inode number or a non persistent inode number allocated from a
1093 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1094 if (ofs->config.xino == OVL_XINO_ON)
1095 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1097 } else if (ofs->config.xino == OVL_XINO_OFF) {
1098 ofs->xino_mode = -1;
1099 } else if (ofs->xino_mode < 0) {
1101 * This is a roundup of number of bits needed for encoding
1102 * fsid, where fsid 0 is reserved for upper fs (even with
1103 * lower only overlay) +1 extra bit is reserved for the non
1104 * persistent inode number range that is used for resolving
1105 * xino lower bits overflow.
1107 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1108 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1111 if (ofs->xino_mode > 0) {
1112 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1119 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1120 struct ovl_fs_context *ctx,
1122 struct ovl_layer *layers)
1126 size_t nr_merged_lower;
1127 struct ovl_entry *oe;
1128 struct ovl_path *lowerstack;
1130 struct ovl_fs_context_layer *l;
1132 if (!ofs->config.upperdir && ctx->nr == 1) {
1133 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1134 return ERR_PTR(-EINVAL);
1138 for (i = 0; i < ctx->nr; i++) {
1141 err = ovl_lower_dir(l->name, &l->path, ofs, &sb->s_stack_depth);
1143 return ERR_PTR(err);
1147 sb->s_stack_depth++;
1148 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1149 pr_err("maximum fs stacking depth exceeded\n");
1150 return ERR_PTR(err);
1153 err = ovl_get_layers(sb, ofs, ctx, layers);
1155 return ERR_PTR(err);
1158 /* Data-only layers are not merged in root directory */
1159 nr_merged_lower = ctx->nr - ctx->nr_data;
1160 oe = ovl_alloc_entry(nr_merged_lower);
1162 return ERR_PTR(err);
1164 lowerstack = ovl_lowerstack(oe);
1165 for (i = 0; i < nr_merged_lower; i++) {
1167 lowerstack[i].dentry = dget(l->path.dentry);
1168 lowerstack[i].layer = &ofs->layers[i + 1];
1170 ofs->numdatalayer = ctx->nr_data;
1176 * Check if this layer root is a descendant of:
1177 * - another layer of this overlayfs instance
1178 * - upper/work dir of any overlayfs instance
1180 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1181 struct dentry *dentry, const char *name,
1184 struct dentry *next = dentry, *parent;
1190 parent = dget_parent(next);
1192 /* Walk back ancestors to root (inclusive) looking for traps */
1193 while (!err && parent != next) {
1194 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1196 pr_err("overlapping %s path\n", name);
1197 } else if (ovl_is_inuse(parent)) {
1198 err = ovl_report_in_use(ofs, name);
1201 parent = dget_parent(next);
1211 * Check if any of the layers or work dirs overlap.
1213 static int ovl_check_overlapping_layers(struct super_block *sb,
1218 if (ovl_upper_mnt(ofs)) {
1219 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1225 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1226 * this instance and covers overlapping work and index dirs,
1227 * unless work or index dir have been moved since created inside
1228 * workbasedir. In that case, we already have their traps in
1229 * inode cache and we will catch that case on lookup.
1231 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1237 for (i = 1; i < ofs->numlayer; i++) {
1238 err = ovl_check_layer(sb, ofs,
1239 ofs->layers[i].mnt->mnt_root,
1248 static struct dentry *ovl_get_root(struct super_block *sb,
1249 struct dentry *upperdentry,
1250 struct ovl_entry *oe)
1252 struct dentry *root;
1253 struct ovl_path *lowerpath = ovl_lowerstack(oe);
1254 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1255 int fsid = lowerpath->layer->fsid;
1256 struct ovl_inode_params oip = {
1257 .upperdentry = upperdentry,
1261 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1266 /* Root inode uses upper st_ino/i_ino */
1267 ino = d_inode(upperdentry)->i_ino;
1269 ovl_dentry_set_upper_alias(root);
1270 if (ovl_is_impuredir(sb, upperdentry))
1271 ovl_set_flag(OVL_IMPURE, d_inode(root));
1274 /* Root is always merge -> can have whiteouts */
1275 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1276 ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1277 ovl_set_upperdata(d_inode(root));
1278 ovl_inode_init(d_inode(root), &oip, ino, fsid);
1279 ovl_dentry_init_flags(root, upperdentry, oe, DCACHE_OP_WEAK_REVALIDATE);
1280 /* root keeps a reference of upperdentry */
1286 int ovl_fill_super(struct super_block *sb, struct fs_context *fc)
1288 struct ovl_fs *ofs = sb->s_fs_info;
1289 struct ovl_fs_context *ctx = fc->fs_private;
1290 struct dentry *root_dentry;
1291 struct ovl_entry *oe;
1292 struct ovl_layer *layers;
1297 if (WARN_ON(fc->user_ns != current_user_ns()))
1300 sb->s_d_op = &ovl_dentry_operations;
1303 ofs->creator_cred = cred = prepare_creds();
1307 err = ovl_fs_params_verify(ctx, &ofs->config);
1313 if (!(fc->sb_flags & SB_SILENT))
1314 pr_err("missing 'lowerdir'\n");
1319 layers = kcalloc(ctx->nr + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1323 ofs->config.lowerdirs = kcalloc(ctx->nr + 1, sizeof(char *), GFP_KERNEL);
1324 if (!ofs->config.lowerdirs) {
1328 ofs->layers = layers;
1330 * Layer 0 is reserved for upper even if there's no upper.
1331 * config.lowerdirs[0] is used for storing the user provided colon
1332 * separated lowerdir string.
1334 ofs->config.lowerdirs[0] = ctx->lowerdir_all;
1335 ctx->lowerdir_all = NULL;
1338 sb->s_stack_depth = 0;
1339 sb->s_maxbytes = MAX_LFS_FILESIZE;
1340 atomic_long_set(&ofs->last_ino, 1);
1341 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1342 if (ofs->config.xino != OVL_XINO_OFF) {
1343 ofs->xino_mode = BITS_PER_LONG - 32;
1344 if (!ofs->xino_mode) {
1345 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1346 ofs->config.xino = OVL_XINO_OFF;
1350 /* alloc/destroy_inode needed for setting up traps in inode cache */
1351 sb->s_op = &ovl_super_operations;
1353 if (ofs->config.upperdir) {
1354 struct super_block *upper_sb;
1357 if (!ofs->config.workdir) {
1358 pr_err("missing 'workdir'\n");
1362 err = ovl_get_upper(sb, ofs, &layers[0], &ctx->upper);
1366 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
1367 if (!ovl_should_sync(ofs)) {
1368 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
1369 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
1371 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1376 err = ovl_get_workdir(sb, ofs, &ctx->upper, &ctx->work);
1381 sb->s_flags |= SB_RDONLY;
1383 sb->s_stack_depth = upper_sb->s_stack_depth;
1384 sb->s_time_gran = upper_sb->s_time_gran;
1386 oe = ovl_get_lowerstack(sb, ctx, ofs, layers);
1391 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1392 if (!ovl_upper_mnt(ofs))
1393 sb->s_flags |= SB_RDONLY;
1395 if (!ovl_origin_uuid(ofs) && ofs->numfs > 1) {
1396 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=null.\n");
1397 ofs->config.uuid = OVL_UUID_NULL;
1398 } else if (ovl_has_fsid(ofs) && ovl_upper_mnt(ofs)) {
1399 /* Use per instance persistent uuid/fsid */
1400 ovl_init_uuid_xattr(sb, ofs, &ctx->upper);
1403 if (!ovl_force_readonly(ofs) && ofs->config.index) {
1404 err = ovl_get_indexdir(sb, ofs, oe, &ctx->upper);
1408 /* Force r/o mount with no index dir */
1410 sb->s_flags |= SB_RDONLY;
1413 err = ovl_check_overlapping_layers(sb, ofs);
1417 /* Show index=off in /proc/mounts for forced r/o mount */
1418 if (!ofs->indexdir) {
1419 ofs->config.index = false;
1420 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
1421 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1422 ofs->config.nfs_export = false;
1426 if (ofs->config.metacopy && ofs->config.nfs_export) {
1427 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1428 ofs->config.nfs_export = false;
1432 * Support encoding decodable file handles with nfs_export=on
1433 * and encoding non-decodable file handles with nfs_export=off
1434 * if all layers support file handles.
1436 if (ofs->config.nfs_export)
1437 sb->s_export_op = &ovl_export_operations;
1438 else if (!ofs->nofh)
1439 sb->s_export_op = &ovl_export_fid_operations;
1441 /* Never override disk quota limits or use reserved space */
1442 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1444 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1445 sb->s_xattr = ovl_xattr_handlers(ofs);
1446 sb->s_fs_info = ofs;
1447 #ifdef CONFIG_FS_POSIX_ACL
1448 sb->s_flags |= SB_POSIXACL;
1450 sb->s_iflags |= SB_I_SKIP_SYNC;
1452 * Ensure that umask handling is done by the filesystems used
1453 * for the the upper layer instead of overlayfs as that would
1454 * lead to unexpected results.
1456 sb->s_iflags |= SB_I_NOUMASK;
1459 root_dentry = ovl_get_root(sb, ctx->upper.dentry, oe);
1463 sb->s_root = root_dentry;
1471 sb->s_fs_info = NULL;
1475 struct file_system_type ovl_fs_type = {
1476 .owner = THIS_MODULE,
1478 .init_fs_context = ovl_init_fs_context,
1479 .parameters = ovl_parameter_spec,
1480 .fs_flags = FS_USERNS_MOUNT,
1481 .kill_sb = kill_anon_super,
1483 MODULE_ALIAS_FS("overlay");
1485 static void ovl_inode_init_once(void *foo)
1487 struct ovl_inode *oi = foo;
1489 inode_init_once(&oi->vfs_inode);
1492 static int __init ovl_init(void)
1496 ovl_inode_cachep = kmem_cache_create("ovl_inode",
1497 sizeof(struct ovl_inode), 0,
1498 (SLAB_RECLAIM_ACCOUNT|
1499 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1500 ovl_inode_init_once);
1501 if (ovl_inode_cachep == NULL)
1504 err = ovl_aio_request_cache_init();
1506 err = register_filesystem(&ovl_fs_type);
1510 ovl_aio_request_cache_destroy();
1512 kmem_cache_destroy(ovl_inode_cachep);
1517 static void __exit ovl_exit(void)
1519 unregister_filesystem(&ovl_fs_type);
1522 * Make sure all delayed rcu free inodes are flushed before we
1526 kmem_cache_destroy(ovl_inode_cachep);
1527 ovl_aio_request_cache_destroy();
1530 module_init(ovl_init);
1531 module_exit(ovl_exit);