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 struct dentry
*trap
= lock_rename(workdir
, upperdir
);
444 unlock_rename(workdir
, upperdir
);
450 static int ovl_setup_trap(struct super_block
*sb
, struct dentry
*dir
,
451 struct inode
**ptrap
, const char *name
)
456 trap
= ovl_get_trap_inode(sb
, dir
);
457 err
= PTR_ERR_OR_ZERO(trap
);
460 pr_err("conflicting %s path\n", name
);
469 * Determine how we treat concurrent use of upperdir/workdir based on the
470 * index feature. This is papering over mount leaks of container runtimes,
471 * for example, an old overlay mount is leaked and now its upperdir is
472 * attempted to be used as a lower layer in a new overlay mount.
474 static int ovl_report_in_use(struct ovl_fs
*ofs
, const char *name
)
476 if (ofs
->config
.index
) {
477 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
481 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
487 static int ovl_get_upper(struct super_block
*sb
, struct ovl_fs
*ofs
,
488 struct ovl_layer
*upper_layer
,
489 const struct path
*upperpath
)
491 struct vfsmount
*upper_mnt
;
494 /* Upperdir path should not be r/o */
495 if (__mnt_is_readonly(upperpath
->mnt
)) {
496 pr_err("upper fs is r/o, try multi-lower layers mount\n");
501 err
= ovl_check_namelen(upperpath
, ofs
, ofs
->config
.upperdir
);
505 err
= ovl_setup_trap(sb
, upperpath
->dentry
, &upper_layer
->trap
,
510 upper_mnt
= clone_private_mount(upperpath
);
511 err
= PTR_ERR(upper_mnt
);
512 if (IS_ERR(upper_mnt
)) {
513 pr_err("failed to clone upperpath\n");
517 /* Don't inherit atime flags */
518 upper_mnt
->mnt_flags
&= ~(MNT_NOATIME
| MNT_NODIRATIME
| MNT_RELATIME
);
519 upper_layer
->mnt
= upper_mnt
;
520 upper_layer
->idx
= 0;
521 upper_layer
->fsid
= 0;
524 * Inherit SB_NOSEC flag from upperdir.
526 * This optimization changes behavior when a security related attribute
527 * (suid/sgid/security.*) is changed on an underlying layer. This is
528 * okay because we don't yet have guarantees in that case, but it will
529 * need careful treatment once we want to honour changes to underlying
532 if (upper_mnt
->mnt_sb
->s_flags
& SB_NOSEC
)
533 sb
->s_flags
|= SB_NOSEC
;
535 if (ovl_inuse_trylock(ovl_upper_mnt(ofs
)->mnt_root
)) {
536 ofs
->upperdir_locked
= true;
538 err
= ovl_report_in_use(ofs
, "upperdir");
549 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
550 * negative values if error is encountered.
552 static int ovl_check_rename_whiteout(struct ovl_fs
*ofs
)
554 struct dentry
*workdir
= ofs
->workdir
;
555 struct inode
*dir
= d_inode(workdir
);
558 struct dentry
*whiteout
;
559 struct name_snapshot name
;
562 inode_lock_nested(dir
, I_MUTEX_PARENT
);
564 temp
= ovl_create_temp(ofs
, workdir
, OVL_CATTR(S_IFREG
| 0));
569 dest
= ovl_lookup_temp(ofs
, workdir
);
576 /* Name is inline and stable - using snapshot as a copy helper */
577 take_dentry_name_snapshot(&name
, temp
);
578 err
= ovl_do_rename(ofs
, dir
, temp
, dir
, dest
, RENAME_WHITEOUT
);
585 whiteout
= ovl_lookup_upper(ofs
, name
.name
.name
, workdir
, name
.name
.len
);
586 err
= PTR_ERR(whiteout
);
587 if (IS_ERR(whiteout
))
590 err
= ovl_upper_is_whiteout(ofs
, whiteout
);
592 /* Best effort cleanup of whiteout and temp file */
594 ovl_cleanup(ofs
, dir
, whiteout
);
598 ovl_cleanup(ofs
, dir
, temp
);
599 release_dentry_name_snapshot(&name
);
609 static struct dentry
*ovl_lookup_or_create(struct ovl_fs
*ofs
,
610 struct dentry
*parent
,
611 const char *name
, umode_t mode
)
613 size_t len
= strlen(name
);
614 struct dentry
*child
;
616 inode_lock_nested(parent
->d_inode
, I_MUTEX_PARENT
);
617 child
= ovl_lookup_upper(ofs
, name
, parent
, len
);
618 if (!IS_ERR(child
) && !child
->d_inode
)
619 child
= ovl_create_real(ofs
, parent
->d_inode
, child
,
621 inode_unlock(parent
->d_inode
);
628 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
631 static int ovl_create_volatile_dirty(struct ovl_fs
*ofs
)
634 struct dentry
*d
= dget(ofs
->workbasedir
);
635 static const char *const volatile_path
[] = {
636 OVL_WORKDIR_NAME
, "incompat", "volatile", "dirty"
638 const char *const *name
= volatile_path
;
640 for (ctr
= ARRAY_SIZE(volatile_path
); ctr
; ctr
--, name
++) {
641 d
= ovl_lookup_or_create(ofs
, d
, *name
, ctr
> 1 ? S_IFDIR
: S_IFREG
);
649 static int ovl_make_workdir(struct super_block
*sb
, struct ovl_fs
*ofs
,
650 const struct path
*workpath
)
652 struct vfsmount
*mnt
= ovl_upper_mnt(ofs
);
653 struct dentry
*workdir
;
654 struct file
*tmpfile
;
655 bool rename_whiteout
;
660 err
= mnt_want_write(mnt
);
664 workdir
= ovl_workdir_create(ofs
, OVL_WORKDIR_NAME
, false);
665 err
= PTR_ERR(workdir
);
666 if (IS_ERR_OR_NULL(workdir
))
669 ofs
->workdir
= workdir
;
671 err
= ovl_setup_trap(sb
, ofs
->workdir
, &ofs
->workdir_trap
, "workdir");
676 * Upper should support d_type, else whiteouts are visible. Given
677 * workdir and upper are on same fs, we can do iterate_dir() on
678 * workdir. This check requires successful creation of workdir in
681 err
= ovl_check_d_type_supported(workpath
);
687 pr_warn("upper fs needs to support d_type.\n");
689 /* Check if upper/work fs supports O_TMPFILE */
690 tmpfile
= ovl_do_tmpfile(ofs
, ofs
->workdir
, S_IFREG
| 0);
691 ofs
->tmpfile
= !IS_ERR(tmpfile
);
695 pr_warn("upper fs does not support tmpfile.\n");
698 /* Check if upper/work fs supports RENAME_WHITEOUT */
699 err
= ovl_check_rename_whiteout(ofs
);
703 rename_whiteout
= err
;
704 if (!rename_whiteout
)
705 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
708 * Check if upper/work fs supports (trusted|user).overlay.* xattr
710 err
= ovl_setxattr(ofs
, ofs
->workdir
, OVL_XATTR_OPAQUE
, "0", 1);
712 pr_warn("failed to set xattr on upper\n");
714 if (ovl_redirect_follow(ofs
)) {
715 ofs
->config
.redirect_mode
= OVL_REDIRECT_NOFOLLOW
;
716 pr_warn("...falling back to redirect_dir=nofollow.\n");
718 if (ofs
->config
.metacopy
) {
719 ofs
->config
.metacopy
= false;
720 pr_warn("...falling back to metacopy=off.\n");
722 if (ofs
->config
.index
) {
723 ofs
->config
.index
= false;
724 pr_warn("...falling back to index=off.\n");
726 if (ovl_has_fsid(ofs
)) {
727 ofs
->config
.uuid
= OVL_UUID_NULL
;
728 pr_warn("...falling back to uuid=null.\n");
731 * xattr support is required for persistent st_ino.
732 * Without persistent st_ino, xino=auto falls back to xino=off.
734 if (ofs
->config
.xino
== OVL_XINO_AUTO
) {
735 ofs
->config
.xino
= OVL_XINO_OFF
;
736 pr_warn("...falling back to xino=off.\n");
738 if (err
== -EPERM
&& !ofs
->config
.userxattr
)
739 pr_info("try mounting with 'userxattr' option\n");
742 ovl_removexattr(ofs
, ofs
->workdir
, OVL_XATTR_OPAQUE
);
746 * We allowed sub-optimal upper fs configuration and don't want to break
747 * users over kernel upgrade, but we never allowed remote upper fs, so
748 * we can enforce strict requirements for remote upper fs.
750 if (ovl_dentry_remote(ofs
->workdir
) &&
751 (!d_type
|| !rename_whiteout
|| ofs
->noxattr
)) {
752 pr_err("upper fs missing required features.\n");
758 * For volatile mount, create a incompat/volatile/dirty file to keep
761 if (ofs
->config
.ovl_volatile
) {
762 err
= ovl_create_volatile_dirty(ofs
);
764 pr_err("Failed to create volatile/dirty file.\n");
769 /* Check if upper/work fs supports file handles */
770 fh_type
= ovl_can_decode_fh(ofs
->workdir
->d_sb
);
771 if (ofs
->config
.index
&& !fh_type
) {
772 ofs
->config
.index
= false;
773 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
775 ofs
->nofh
|= !fh_type
;
777 /* Check if upper fs has 32bit inode numbers */
778 if (fh_type
!= FILEID_INO32_GEN
)
781 /* NFS export of r/w mount depends on index */
782 if (ofs
->config
.nfs_export
&& !ofs
->config
.index
) {
783 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
784 ofs
->config
.nfs_export
= false;
791 static int ovl_get_workdir(struct super_block
*sb
, struct ovl_fs
*ofs
,
792 const struct path
*upperpath
,
793 const struct path
*workpath
)
798 if (upperpath
->mnt
!= workpath
->mnt
) {
799 pr_err("workdir and upperdir must reside under the same mount\n");
802 if (!ovl_workdir_ok(workpath
->dentry
, upperpath
->dentry
)) {
803 pr_err("workdir and upperdir must be separate subtrees\n");
807 ofs
->workbasedir
= dget(workpath
->dentry
);
809 if (ovl_inuse_trylock(ofs
->workbasedir
)) {
810 ofs
->workdir_locked
= true;
812 err
= ovl_report_in_use(ofs
, "workdir");
817 err
= ovl_setup_trap(sb
, ofs
->workbasedir
, &ofs
->workbasedir_trap
,
822 return ovl_make_workdir(sb
, ofs
, workpath
);
825 static int ovl_get_indexdir(struct super_block
*sb
, struct ovl_fs
*ofs
,
826 struct ovl_entry
*oe
, const struct path
*upperpath
)
828 struct vfsmount
*mnt
= ovl_upper_mnt(ofs
);
829 struct dentry
*indexdir
;
830 struct dentry
*origin
= ovl_lowerstack(oe
)->dentry
;
831 const struct ovl_fh
*fh
;
834 fh
= ovl_get_origin_fh(ofs
, origin
);
838 err
= mnt_want_write(mnt
);
842 /* Verify lower root is upper root origin */
843 err
= ovl_verify_origin_fh(ofs
, upperpath
->dentry
, fh
, true);
845 pr_err("failed to verify upper root origin\n");
849 /* index dir will act also as workdir */
850 iput(ofs
->workdir_trap
);
851 ofs
->workdir_trap
= NULL
;
854 indexdir
= ovl_workdir_create(ofs
, OVL_INDEXDIR_NAME
, true);
855 if (IS_ERR(indexdir
)) {
856 err
= PTR_ERR(indexdir
);
857 } else if (indexdir
) {
858 ofs
->workdir
= indexdir
;
859 err
= ovl_setup_trap(sb
, indexdir
, &ofs
->workdir_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
, indexdir
)) {
873 err
= ovl_verify_origin_xattr(ofs
, indexdir
,
875 upperpath
->dentry
, true,
878 pr_err("failed to verify index dir 'origin' xattr\n");
880 err
= ovl_verify_upper(ofs
, indexdir
, upperpath
->dentry
, true);
882 pr_err("failed to verify index dir 'upper' xattr\n");
884 /* Cleanup bad/stale/orphan index entries */
886 err
= ovl_indexdir_cleanup(ofs
);
888 if (err
|| !indexdir
)
889 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
898 static bool ovl_lower_uuid_ok(struct ovl_fs
*ofs
, const uuid_t
*uuid
)
902 if (!ofs
->config
.nfs_export
&& !ovl_upper_mnt(ofs
))
906 * We allow using single lower with null uuid for index and nfs_export
907 * for example to support those features with single lower squashfs.
908 * To avoid regressions in setups of overlay with re-formatted lower
909 * squashfs, do not allow decoding origin with lower null uuid unless
910 * user opted-in to one of the new features that require following the
911 * lower inode of non-dir upper.
913 if (ovl_allow_offline_changes(ofs
) && uuid_is_null(uuid
))
916 for (i
= 0; i
< ofs
->numfs
; i
++) {
918 * We use uuid to associate an overlay lower file handle with a
919 * lower layer, so we can accept lower fs with null uuid as long
920 * as all lower layers with null uuid are on the same fs.
921 * if we detect multiple lower fs with the same uuid, we
922 * disable lower file handle decoding on all of them.
924 if (ofs
->fs
[i
].is_lower
&&
925 uuid_equal(&ofs
->fs
[i
].sb
->s_uuid
, uuid
)) {
926 ofs
->fs
[i
].bad_uuid
= true;
933 /* Get a unique fsid for the layer */
934 static int ovl_get_fsid(struct ovl_fs
*ofs
, const struct path
*path
)
936 struct super_block
*sb
= path
->mnt
->mnt_sb
;
940 bool bad_uuid
= false;
943 for (i
= 0; i
< ofs
->numfs
; i
++) {
944 if (ofs
->fs
[i
].sb
== sb
)
948 if (!ovl_lower_uuid_ok(ofs
, &sb
->s_uuid
)) {
950 if (ofs
->config
.xino
== OVL_XINO_AUTO
) {
951 ofs
->config
.xino
= OVL_XINO_OFF
;
954 if (ofs
->config
.index
|| ofs
->config
.nfs_export
) {
955 ofs
->config
.index
= false;
956 ofs
->config
.nfs_export
= false;
960 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
961 uuid_is_null(&sb
->s_uuid
) ? "null" :
963 path
->dentry
, ovl_xino_mode(&ofs
->config
));
967 err
= get_anon_bdev(&dev
);
969 pr_err("failed to get anonymous bdev for lowerpath\n");
973 ofs
->fs
[ofs
->numfs
].sb
= sb
;
974 ofs
->fs
[ofs
->numfs
].pseudo_dev
= dev
;
975 ofs
->fs
[ofs
->numfs
].bad_uuid
= bad_uuid
;
981 * The fsid after the last lower fsid is used for the data layers.
982 * It is a "null fs" with a null sb, null uuid, and no pseudo dev.
984 static int ovl_get_data_fsid(struct ovl_fs
*ofs
)
990 static int ovl_get_layers(struct super_block
*sb
, struct ovl_fs
*ofs
,
991 struct ovl_fs_context
*ctx
, struct ovl_layer
*layers
)
995 size_t nr_merged_lower
;
997 ofs
->fs
= kcalloc(ctx
->nr
+ 2, sizeof(struct ovl_sb
), GFP_KERNEL
);
1002 * idx/fsid 0 are reserved for upper fs even with lower only overlay
1003 * and the last fsid is reserved for "null fs" of the data layers.
1008 * All lower layers that share the same fs as upper layer, use the same
1009 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1010 * only overlay to simplify ovl_fs_free().
1011 * is_lower will be set if upper fs is shared with a lower layer.
1013 err
= get_anon_bdev(&ofs
->fs
[0].pseudo_dev
);
1015 pr_err("failed to get anonymous bdev for upper fs\n");
1019 if (ovl_upper_mnt(ofs
)) {
1020 ofs
->fs
[0].sb
= ovl_upper_mnt(ofs
)->mnt_sb
;
1021 ofs
->fs
[0].is_lower
= false;
1024 nr_merged_lower
= ctx
->nr
- ctx
->nr_data
;
1025 for (i
= 0; i
< ctx
->nr
; i
++) {
1026 struct ovl_fs_context_layer
*l
= &ctx
->lower
[i
];
1027 struct vfsmount
*mnt
;
1031 if (i
< nr_merged_lower
)
1032 fsid
= ovl_get_fsid(ofs
, &l
->path
);
1034 fsid
= ovl_get_data_fsid(ofs
);
1039 * Check if lower root conflicts with this overlay layers before
1040 * checking if it is in-use as upperdir/workdir of "another"
1041 * mount, because we do not bother to check in ovl_is_inuse() if
1042 * the upperdir/workdir is in fact in-use by our
1045 err
= ovl_setup_trap(sb
, l
->path
.dentry
, &trap
, "lowerdir");
1049 if (ovl_is_inuse(l
->path
.dentry
)) {
1050 err
= ovl_report_in_use(ofs
, "lowerdir");
1057 mnt
= clone_private_mount(&l
->path
);
1060 pr_err("failed to clone lowerpath\n");
1066 * Make lower layers R/O. That way fchmod/fchown on lower file
1067 * will fail instead of modifying lower fs.
1069 mnt
->mnt_flags
|= MNT_READONLY
| MNT_NOATIME
;
1071 layers
[ofs
->numlayer
].trap
= trap
;
1072 layers
[ofs
->numlayer
].mnt
= mnt
;
1073 layers
[ofs
->numlayer
].idx
= ofs
->numlayer
;
1074 layers
[ofs
->numlayer
].fsid
= fsid
;
1075 layers
[ofs
->numlayer
].fs
= &ofs
->fs
[fsid
];
1076 /* Store for printing lowerdir=... in ovl_show_options() */
1077 ofs
->config
.lowerdirs
[ofs
->numlayer
] = l
->name
;
1080 ofs
->fs
[fsid
].is_lower
= true;
1084 * When all layers on same fs, overlay can use real inode numbers.
1085 * With mount option "xino=<on|auto>", mounter declares that there are
1086 * enough free high bits in underlying fs to hold the unique fsid.
1087 * If overlayfs does encounter underlying inodes using the high xino
1088 * bits reserved for fsid, it emits a warning and uses the original
1089 * inode number or a non persistent inode number allocated from a
1092 if (ofs
->numfs
- !ovl_upper_mnt(ofs
) == 1) {
1093 if (ofs
->config
.xino
== OVL_XINO_ON
)
1094 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1096 } else if (ofs
->config
.xino
== OVL_XINO_OFF
) {
1097 ofs
->xino_mode
= -1;
1098 } else if (ofs
->xino_mode
< 0) {
1100 * This is a roundup of number of bits needed for encoding
1101 * fsid, where fsid 0 is reserved for upper fs (even with
1102 * lower only overlay) +1 extra bit is reserved for the non
1103 * persistent inode number range that is used for resolving
1104 * xino lower bits overflow.
1106 BUILD_BUG_ON(ilog2(OVL_MAX_STACK
) > 30);
1107 ofs
->xino_mode
= ilog2(ofs
->numfs
- 1) + 2;
1110 if (ofs
->xino_mode
> 0) {
1111 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1118 static struct ovl_entry
*ovl_get_lowerstack(struct super_block
*sb
,
1119 struct ovl_fs_context
*ctx
,
1121 struct ovl_layer
*layers
)
1125 size_t nr_merged_lower
;
1126 struct ovl_entry
*oe
;
1127 struct ovl_path
*lowerstack
;
1129 struct ovl_fs_context_layer
*l
;
1131 if (!ofs
->config
.upperdir
&& ctx
->nr
== 1) {
1132 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1133 return ERR_PTR(-EINVAL
);
1137 for (i
= 0; i
< ctx
->nr
; i
++) {
1140 err
= ovl_lower_dir(l
->name
, &l
->path
, ofs
, &sb
->s_stack_depth
);
1142 return ERR_PTR(err
);
1146 sb
->s_stack_depth
++;
1147 if (sb
->s_stack_depth
> FILESYSTEM_MAX_STACK_DEPTH
) {
1148 pr_err("maximum fs stacking depth exceeded\n");
1149 return ERR_PTR(err
);
1152 err
= ovl_get_layers(sb
, ofs
, ctx
, layers
);
1154 return ERR_PTR(err
);
1157 /* Data-only layers are not merged in root directory */
1158 nr_merged_lower
= ctx
->nr
- ctx
->nr_data
;
1159 oe
= ovl_alloc_entry(nr_merged_lower
);
1161 return ERR_PTR(err
);
1163 lowerstack
= ovl_lowerstack(oe
);
1164 for (i
= 0; i
< nr_merged_lower
; i
++) {
1166 lowerstack
[i
].dentry
= dget(l
->path
.dentry
);
1167 lowerstack
[i
].layer
= &ofs
->layers
[i
+ 1];
1169 ofs
->numdatalayer
= ctx
->nr_data
;
1175 * Check if this layer root is a descendant of:
1176 * - another layer of this overlayfs instance
1177 * - upper/work dir of any overlayfs instance
1179 static int ovl_check_layer(struct super_block
*sb
, struct ovl_fs
*ofs
,
1180 struct dentry
*dentry
, const char *name
,
1183 struct dentry
*next
= dentry
, *parent
;
1189 parent
= dget_parent(next
);
1191 /* Walk back ancestors to root (inclusive) looking for traps */
1192 while (!err
&& parent
!= next
) {
1193 if (is_lower
&& ovl_lookup_trap_inode(sb
, parent
)) {
1195 pr_err("overlapping %s path\n", name
);
1196 } else if (ovl_is_inuse(parent
)) {
1197 err
= ovl_report_in_use(ofs
, name
);
1200 parent
= dget_parent(next
);
1210 * Check if any of the layers or work dirs overlap.
1212 static int ovl_check_overlapping_layers(struct super_block
*sb
,
1217 if (ovl_upper_mnt(ofs
)) {
1218 err
= ovl_check_layer(sb
, ofs
, ovl_upper_mnt(ofs
)->mnt_root
,
1224 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1225 * this instance and covers overlapping work and index dirs,
1226 * unless work or index dir have been moved since created inside
1227 * workbasedir. In that case, we already have their traps in
1228 * inode cache and we will catch that case on lookup.
1230 err
= ovl_check_layer(sb
, ofs
, ofs
->workbasedir
, "workdir",
1236 for (i
= 1; i
< ofs
->numlayer
; i
++) {
1237 err
= ovl_check_layer(sb
, ofs
,
1238 ofs
->layers
[i
].mnt
->mnt_root
,
1247 static struct dentry
*ovl_get_root(struct super_block
*sb
,
1248 struct dentry
*upperdentry
,
1249 struct ovl_entry
*oe
)
1251 struct dentry
*root
;
1252 struct ovl_path
*lowerpath
= ovl_lowerstack(oe
);
1253 unsigned long ino
= d_inode(lowerpath
->dentry
)->i_ino
;
1254 int fsid
= lowerpath
->layer
->fsid
;
1255 struct ovl_inode_params oip
= {
1256 .upperdentry
= upperdentry
,
1260 root
= d_make_root(ovl_new_inode(sb
, S_IFDIR
, 0));
1265 /* Root inode uses upper st_ino/i_ino */
1266 ino
= d_inode(upperdentry
)->i_ino
;
1268 ovl_dentry_set_upper_alias(root
);
1269 if (ovl_is_impuredir(sb
, upperdentry
))
1270 ovl_set_flag(OVL_IMPURE
, d_inode(root
));
1273 /* Root is always merge -> can have whiteouts */
1274 ovl_set_flag(OVL_WHITEOUTS
, d_inode(root
));
1275 ovl_dentry_set_flag(OVL_E_CONNECTED
, root
);
1276 ovl_set_upperdata(d_inode(root
));
1277 ovl_inode_init(d_inode(root
), &oip
, ino
, fsid
);
1278 ovl_dentry_init_flags(root
, upperdentry
, oe
, DCACHE_OP_WEAK_REVALIDATE
);
1279 /* root keeps a reference of upperdentry */
1285 int ovl_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
1287 struct ovl_fs
*ofs
= sb
->s_fs_info
;
1288 struct ovl_fs_context
*ctx
= fc
->fs_private
;
1289 struct dentry
*root_dentry
;
1290 struct ovl_entry
*oe
;
1291 struct ovl_layer
*layers
;
1296 if (WARN_ON(fc
->user_ns
!= current_user_ns()))
1299 sb
->s_d_op
= &ovl_dentry_operations
;
1302 ofs
->creator_cred
= cred
= prepare_creds();
1306 err
= ovl_fs_params_verify(ctx
, &ofs
->config
);
1312 if (!(fc
->sb_flags
& SB_SILENT
))
1313 pr_err("missing 'lowerdir'\n");
1318 layers
= kcalloc(ctx
->nr
+ 1, sizeof(struct ovl_layer
), GFP_KERNEL
);
1322 ofs
->config
.lowerdirs
= kcalloc(ctx
->nr
+ 1, sizeof(char *), GFP_KERNEL
);
1323 if (!ofs
->config
.lowerdirs
) {
1327 ofs
->layers
= layers
;
1329 * Layer 0 is reserved for upper even if there's no upper.
1330 * config.lowerdirs[0] is used for storing the user provided colon
1331 * separated lowerdir string.
1333 ofs
->config
.lowerdirs
[0] = ctx
->lowerdir_all
;
1334 ctx
->lowerdir_all
= NULL
;
1337 sb
->s_stack_depth
= 0;
1338 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1339 atomic_long_set(&ofs
->last_ino
, 1);
1340 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1341 if (ofs
->config
.xino
!= OVL_XINO_OFF
) {
1342 ofs
->xino_mode
= BITS_PER_LONG
- 32;
1343 if (!ofs
->xino_mode
) {
1344 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1345 ofs
->config
.xino
= OVL_XINO_OFF
;
1349 /* alloc/destroy_inode needed for setting up traps in inode cache */
1350 sb
->s_op
= &ovl_super_operations
;
1352 if (ofs
->config
.upperdir
) {
1353 struct super_block
*upper_sb
;
1356 if (!ofs
->config
.workdir
) {
1357 pr_err("missing 'workdir'\n");
1361 err
= ovl_get_upper(sb
, ofs
, &layers
[0], &ctx
->upper
);
1365 upper_sb
= ovl_upper_mnt(ofs
)->mnt_sb
;
1366 if (!ovl_should_sync(ofs
)) {
1367 ofs
->errseq
= errseq_sample(&upper_sb
->s_wb_err
);
1368 if (errseq_check(&upper_sb
->s_wb_err
, ofs
->errseq
)) {
1370 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1375 err
= ovl_get_workdir(sb
, ofs
, &ctx
->upper
, &ctx
->work
);
1380 sb
->s_flags
|= SB_RDONLY
;
1382 sb
->s_stack_depth
= upper_sb
->s_stack_depth
;
1383 sb
->s_time_gran
= upper_sb
->s_time_gran
;
1385 oe
= ovl_get_lowerstack(sb
, ctx
, ofs
, layers
);
1390 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1391 if (!ovl_upper_mnt(ofs
))
1392 sb
->s_flags
|= SB_RDONLY
;
1394 if (!ovl_origin_uuid(ofs
) && ofs
->numfs
> 1) {
1395 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=null.\n");
1396 ofs
->config
.uuid
= OVL_UUID_NULL
;
1397 } else if (ovl_has_fsid(ofs
) && ovl_upper_mnt(ofs
)) {
1398 /* Use per instance persistent uuid/fsid */
1399 ovl_init_uuid_xattr(sb
, ofs
, &ctx
->upper
);
1402 if (!ovl_force_readonly(ofs
) && ofs
->config
.index
) {
1403 err
= ovl_get_indexdir(sb
, ofs
, oe
, &ctx
->upper
);
1407 /* Force r/o mount with no index dir */
1409 sb
->s_flags
|= SB_RDONLY
;
1412 err
= ovl_check_overlapping_layers(sb
, ofs
);
1416 /* Show index=off in /proc/mounts for forced r/o mount */
1417 if (!ofs
->workdir
) {
1418 ofs
->config
.index
= false;
1419 if (ovl_upper_mnt(ofs
) && ofs
->config
.nfs_export
) {
1420 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1421 ofs
->config
.nfs_export
= false;
1425 if (ofs
->config
.metacopy
&& ofs
->config
.nfs_export
) {
1426 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1427 ofs
->config
.nfs_export
= false;
1431 * Support encoding decodable file handles with nfs_export=on
1432 * and encoding non-decodable file handles with nfs_export=off
1433 * if all layers support file handles.
1435 if (ofs
->config
.nfs_export
)
1436 sb
->s_export_op
= &ovl_export_operations
;
1437 else if (!ofs
->nofh
)
1438 sb
->s_export_op
= &ovl_export_fid_operations
;
1440 /* Never override disk quota limits or use reserved space */
1441 cap_lower(cred
->cap_effective
, CAP_SYS_RESOURCE
);
1443 sb
->s_magic
= OVERLAYFS_SUPER_MAGIC
;
1444 sb
->s_xattr
= ovl_xattr_handlers(ofs
);
1445 sb
->s_fs_info
= ofs
;
1446 #ifdef CONFIG_FS_POSIX_ACL
1447 sb
->s_flags
|= SB_POSIXACL
;
1449 sb
->s_iflags
|= SB_I_SKIP_SYNC
;
1451 * Ensure that umask handling is done by the filesystems used
1452 * for the the upper layer instead of overlayfs as that would
1453 * lead to unexpected results.
1455 sb
->s_iflags
|= SB_I_NOUMASK
;
1456 sb
->s_iflags
|= SB_I_EVM_UNSUPPORTED
;
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
= register_filesystem(&ovl_fs_type
);
1508 kmem_cache_destroy(ovl_inode_cachep
);
1513 static void __exit
ovl_exit(void)
1515 unregister_filesystem(&ovl_fs_type
);
1518 * Make sure all delayed rcu free inodes are flushed before we
1522 kmem_cache_destroy(ovl_inode_cachep
);
1525 module_init(ovl_init
);
1526 module_exit(ovl_exit
);