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
, enum d_real_type type
)
33 struct dentry
*upper
, *lower
;
44 if (!d_is_reg(dentry
)) {
45 /* d_real_inode() is only relevant for regular files */
49 upper
= ovl_dentry_upper(dentry
);
50 if (upper
&& (type
== D_REAL_METADATA
||
51 ovl_has_upperdata(d_inode(dentry
))))
54 if (type
== D_REAL_METADATA
) {
55 lower
= ovl_dentry_lower(dentry
);
60 * Best effort lazy lookup of lowerdata for D_REAL_DATA case to return
61 * the real lowerdata dentry. The only current caller of d_real() with
62 * D_REAL_DATA is d_real_inode() from trace_uprobe and this caller is
63 * likely going to be followed reading from the file, before placing
64 * uprobes on offset within the file, so lowerdata should be available
65 * when setting the uprobe.
67 err
= ovl_verify_lowerdata(dentry
);
70 lower
= ovl_dentry_lowerdata(dentry
);
75 /* Handle recursion into stacked lower fs */
76 return d_real(lower
, type
);
79 WARN(1, "%s(%pd4, %d): real dentry not found\n", __func__
, dentry
, type
);
83 static int ovl_revalidate_real(struct dentry
*d
, unsigned int flags
, bool weak
)
91 if (d
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)
92 ret
= d
->d_op
->d_weak_revalidate(d
, flags
);
93 } else if (d
->d_flags
& DCACHE_OP_REVALIDATE
) {
94 ret
= d
->d_op
->d_revalidate(d
, flags
);
96 if (!(flags
& LOOKUP_RCU
))
104 static int ovl_dentry_revalidate_common(struct dentry
*dentry
,
105 unsigned int flags
, bool weak
)
107 struct ovl_entry
*oe
;
108 struct ovl_path
*lowerstack
;
109 struct inode
*inode
= d_inode_rcu(dentry
);
110 struct dentry
*upper
;
114 /* Careful in RCU mode */
119 lowerstack
= ovl_lowerstack(oe
);
120 upper
= ovl_i_dentry_upper(inode
);
122 ret
= ovl_revalidate_real(upper
, flags
, weak
);
124 for (i
= 0; ret
> 0 && i
< ovl_numlower(oe
); i
++)
125 ret
= ovl_revalidate_real(lowerstack
[i
].dentry
, flags
, weak
);
130 static int ovl_dentry_revalidate(struct dentry
*dentry
, unsigned int flags
)
132 return ovl_dentry_revalidate_common(dentry
, flags
, false);
135 static int ovl_dentry_weak_revalidate(struct dentry
*dentry
, unsigned int flags
)
137 return ovl_dentry_revalidate_common(dentry
, flags
, true);
140 static const struct dentry_operations ovl_dentry_operations
= {
141 .d_real
= ovl_d_real
,
142 .d_revalidate
= ovl_dentry_revalidate
,
143 .d_weak_revalidate
= ovl_dentry_weak_revalidate
,
146 static struct kmem_cache
*ovl_inode_cachep
;
148 static struct inode
*ovl_alloc_inode(struct super_block
*sb
)
150 struct ovl_inode
*oi
= alloc_inode_sb(sb
, ovl_inode_cachep
, GFP_KERNEL
);
159 oi
->__upperdentry
= NULL
;
160 oi
->lowerdata_redirect
= NULL
;
162 mutex_init(&oi
->lock
);
164 return &oi
->vfs_inode
;
167 static void ovl_free_inode(struct inode
*inode
)
169 struct ovl_inode
*oi
= OVL_I(inode
);
173 mutex_destroy(&oi
->lock
);
174 kmem_cache_free(ovl_inode_cachep
, oi
);
177 static void ovl_destroy_inode(struct inode
*inode
)
179 struct ovl_inode
*oi
= OVL_I(inode
);
181 dput(oi
->__upperdentry
);
182 ovl_stack_put(ovl_lowerstack(oi
->oe
), ovl_numlower(oi
->oe
));
183 if (S_ISDIR(inode
->i_mode
))
184 ovl_dir_cache_free(inode
);
186 kfree(oi
->lowerdata_redirect
);
189 static void ovl_put_super(struct super_block
*sb
)
191 struct ovl_fs
*ofs
= OVL_FS(sb
);
197 /* Sync real dirty inodes in upper filesystem (if it exists) */
198 static int ovl_sync_fs(struct super_block
*sb
, int wait
)
200 struct ovl_fs
*ofs
= OVL_FS(sb
);
201 struct super_block
*upper_sb
;
204 ret
= ovl_sync_status(ofs
);
206 * We have to always set the err, because the return value isn't
207 * checked in syncfs, and instead indirectly return an error via
208 * the sb's writeback errseq, which VFS inspects after this call.
211 errseq_set(&sb
->s_wb_err
, -EIO
);
219 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
220 * All the super blocks will be iterated, including upper_sb.
222 * If this is a syncfs(2) call, then we do need to call
223 * sync_filesystem() on upper_sb, but enough if we do it when being
224 * called with wait == 1.
229 upper_sb
= ovl_upper_mnt(ofs
)->mnt_sb
;
231 down_read(&upper_sb
->s_umount
);
232 ret
= sync_filesystem(upper_sb
);
233 up_read(&upper_sb
->s_umount
);
240 * @dentry: The dentry to query
241 * @buf: The struct kstatfs to fill in with stats
243 * Get the filesystem statistics. As writes always target the upper layer
244 * filesystem pass the statfs to the upper filesystem (if it exists)
246 static int ovl_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
248 struct super_block
*sb
= dentry
->d_sb
;
249 struct ovl_fs
*ofs
= OVL_FS(sb
);
250 struct dentry
*root_dentry
= sb
->s_root
;
254 ovl_path_real(root_dentry
, &path
);
256 err
= vfs_statfs(&path
, buf
);
258 buf
->f_namelen
= ofs
->namelen
;
259 buf
->f_type
= OVERLAYFS_SUPER_MAGIC
;
260 if (ovl_has_fsid(ofs
))
261 buf
->f_fsid
= uuid_to_fsid(sb
->s_uuid
.b
);
267 static const struct super_operations ovl_super_operations
= {
268 .alloc_inode
= ovl_alloc_inode
,
269 .free_inode
= ovl_free_inode
,
270 .destroy_inode
= ovl_destroy_inode
,
271 .drop_inode
= generic_delete_inode
,
272 .put_super
= ovl_put_super
,
273 .sync_fs
= ovl_sync_fs
,
274 .statfs
= ovl_statfs
,
275 .show_options
= ovl_show_options
,
278 #define OVL_WORKDIR_NAME "work"
279 #define OVL_INDEXDIR_NAME "index"
281 static struct dentry
*ovl_workdir_create(struct ovl_fs
*ofs
,
282 const char *name
, bool persist
)
284 struct inode
*dir
= ofs
->workbasedir
->d_inode
;
285 struct vfsmount
*mnt
= ovl_upper_mnt(ofs
);
288 bool retried
= false;
290 inode_lock_nested(dir
, I_MUTEX_PARENT
);
292 work
= ovl_lookup_upper(ofs
, name
, ofs
->workbasedir
, strlen(name
));
295 struct iattr attr
= {
296 .ia_valid
= ATTR_MODE
,
297 .ia_mode
= S_IFDIR
| 0,
309 err
= ovl_workdir_cleanup(ofs
, dir
, mnt
, work
, 0);
311 if (err
== -EINVAL
) {
318 err
= ovl_mkdir_real(ofs
, dir
, &work
, attr
.ia_mode
);
322 /* Weird filesystem returning with hashed negative (kernfs)? */
324 if (d_really_is_negative(work
))
328 * Try to remove POSIX ACL xattrs from workdir. We are good if:
330 * a) success (there was a POSIX ACL xattr and was removed)
331 * b) -ENODATA (there was no POSIX ACL xattr)
332 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
334 * There are various other error values that could effectively
335 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
336 * if the xattr name is too long), but the set of filesystems
337 * allowed as upper are limited to "normal" ones, where checking
338 * for the above two errors is sufficient.
340 err
= ovl_do_remove_acl(ofs
, work
, XATTR_NAME_POSIX_ACL_DEFAULT
);
341 if (err
&& err
!= -ENODATA
&& err
!= -EOPNOTSUPP
)
344 err
= ovl_do_remove_acl(ofs
, work
, XATTR_NAME_POSIX_ACL_ACCESS
);
345 if (err
&& err
!= -ENODATA
&& err
!= -EOPNOTSUPP
)
348 /* Clear any inherited mode bits */
349 inode_lock(work
->d_inode
);
350 err
= ovl_do_notify_change(ofs
, work
, &attr
);
351 inode_unlock(work
->d_inode
);
365 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
366 ofs
->config
.workdir
, name
, -err
);
371 static int ovl_check_namelen(const struct path
*path
, struct ovl_fs
*ofs
,
374 struct kstatfs statfs
;
375 int err
= vfs_statfs(path
, &statfs
);
378 pr_err("statfs failed on '%s'\n", name
);
380 ofs
->namelen
= max(ofs
->namelen
, statfs
.f_namelen
);
385 static int ovl_lower_dir(const char *name
, struct path
*path
,
386 struct ovl_fs
*ofs
, int *stack_depth
)
391 err
= ovl_check_namelen(path
, ofs
, name
);
395 *stack_depth
= max(*stack_depth
, path
->mnt
->mnt_sb
->s_stack_depth
);
398 * The inodes index feature and NFS export need to encode and decode
399 * file handles, so they require that all layers support them.
401 fh_type
= ovl_can_decode_fh(path
->dentry
->d_sb
);
402 if ((ofs
->config
.nfs_export
||
403 (ofs
->config
.index
&& ofs
->config
.upperdir
)) && !fh_type
) {
404 ofs
->config
.index
= false;
405 ofs
->config
.nfs_export
= false;
406 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
409 ofs
->nofh
|= !fh_type
;
411 * Decoding origin file handle is required for persistent st_ino.
412 * Without persistent st_ino, xino=auto falls back to xino=off.
414 if (ofs
->config
.xino
== OVL_XINO_AUTO
&&
415 ofs
->config
.upperdir
&& !fh_type
) {
416 ofs
->config
.xino
= OVL_XINO_OFF
;
417 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
421 /* Check if lower fs has 32bit inode numbers */
422 if (fh_type
!= FILEID_INO32_GEN
)
428 /* Workdir should not be subdir of upperdir and vice versa */
429 static bool ovl_workdir_ok(struct dentry
*workdir
, struct dentry
*upperdir
)
433 if (workdir
!= upperdir
) {
434 struct dentry
*trap
= lock_rename(workdir
, upperdir
);
436 unlock_rename(workdir
, upperdir
);
442 static int ovl_setup_trap(struct super_block
*sb
, struct dentry
*dir
,
443 struct inode
**ptrap
, const char *name
)
448 trap
= ovl_get_trap_inode(sb
, dir
);
449 err
= PTR_ERR_OR_ZERO(trap
);
452 pr_err("conflicting %s path\n", name
);
461 * Determine how we treat concurrent use of upperdir/workdir based on the
462 * index feature. This is papering over mount leaks of container runtimes,
463 * for example, an old overlay mount is leaked and now its upperdir is
464 * attempted to be used as a lower layer in a new overlay mount.
466 static int ovl_report_in_use(struct ovl_fs
*ofs
, const char *name
)
468 if (ofs
->config
.index
) {
469 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
473 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
479 static int ovl_get_upper(struct super_block
*sb
, struct ovl_fs
*ofs
,
480 struct ovl_layer
*upper_layer
,
481 const struct path
*upperpath
)
483 struct vfsmount
*upper_mnt
;
486 /* Upperdir path should not be r/o */
487 if (__mnt_is_readonly(upperpath
->mnt
)) {
488 pr_err("upper fs is r/o, try multi-lower layers mount\n");
493 err
= ovl_check_namelen(upperpath
, ofs
, ofs
->config
.upperdir
);
497 err
= ovl_setup_trap(sb
, upperpath
->dentry
, &upper_layer
->trap
,
502 upper_mnt
= clone_private_mount(upperpath
);
503 err
= PTR_ERR(upper_mnt
);
504 if (IS_ERR(upper_mnt
)) {
505 pr_err("failed to clone upperpath\n");
509 /* Don't inherit atime flags */
510 upper_mnt
->mnt_flags
&= ~(MNT_NOATIME
| MNT_NODIRATIME
| MNT_RELATIME
);
511 upper_layer
->mnt
= upper_mnt
;
512 upper_layer
->idx
= 0;
513 upper_layer
->fsid
= 0;
516 * Inherit SB_NOSEC flag from upperdir.
518 * This optimization changes behavior when a security related attribute
519 * (suid/sgid/security.*) is changed on an underlying layer. This is
520 * okay because we don't yet have guarantees in that case, but it will
521 * need careful treatment once we want to honour changes to underlying
524 if (upper_mnt
->mnt_sb
->s_flags
& SB_NOSEC
)
525 sb
->s_flags
|= SB_NOSEC
;
527 if (ovl_inuse_trylock(ovl_upper_mnt(ofs
)->mnt_root
)) {
528 ofs
->upperdir_locked
= true;
530 err
= ovl_report_in_use(ofs
, "upperdir");
541 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
542 * negative values if error is encountered.
544 static int ovl_check_rename_whiteout(struct ovl_fs
*ofs
)
546 struct dentry
*workdir
= ofs
->workdir
;
547 struct inode
*dir
= d_inode(workdir
);
550 struct dentry
*whiteout
;
551 struct name_snapshot name
;
554 inode_lock_nested(dir
, I_MUTEX_PARENT
);
556 temp
= ovl_create_temp(ofs
, workdir
, OVL_CATTR(S_IFREG
| 0));
561 dest
= ovl_lookup_temp(ofs
, workdir
);
568 /* Name is inline and stable - using snapshot as a copy helper */
569 take_dentry_name_snapshot(&name
, temp
);
570 err
= ovl_do_rename(ofs
, dir
, temp
, dir
, dest
, RENAME_WHITEOUT
);
577 whiteout
= ovl_lookup_upper(ofs
, name
.name
.name
, workdir
, name
.name
.len
);
578 err
= PTR_ERR(whiteout
);
579 if (IS_ERR(whiteout
))
582 err
= ovl_upper_is_whiteout(ofs
, whiteout
);
584 /* Best effort cleanup of whiteout and temp file */
586 ovl_cleanup(ofs
, dir
, whiteout
);
590 ovl_cleanup(ofs
, dir
, temp
);
591 release_dentry_name_snapshot(&name
);
601 static struct dentry
*ovl_lookup_or_create(struct ovl_fs
*ofs
,
602 struct dentry
*parent
,
603 const char *name
, umode_t mode
)
605 size_t len
= strlen(name
);
606 struct dentry
*child
;
608 inode_lock_nested(parent
->d_inode
, I_MUTEX_PARENT
);
609 child
= ovl_lookup_upper(ofs
, name
, parent
, len
);
610 if (!IS_ERR(child
) && !child
->d_inode
)
611 child
= ovl_create_real(ofs
, parent
->d_inode
, child
,
613 inode_unlock(parent
->d_inode
);
620 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
623 static int ovl_create_volatile_dirty(struct ovl_fs
*ofs
)
626 struct dentry
*d
= dget(ofs
->workbasedir
);
627 static const char *const volatile_path
[] = {
628 OVL_WORKDIR_NAME
, "incompat", "volatile", "dirty"
630 const char *const *name
= volatile_path
;
632 for (ctr
= ARRAY_SIZE(volatile_path
); ctr
; ctr
--, name
++) {
633 d
= ovl_lookup_or_create(ofs
, d
, *name
, ctr
> 1 ? S_IFDIR
: S_IFREG
);
641 static int ovl_make_workdir(struct super_block
*sb
, struct ovl_fs
*ofs
,
642 const struct path
*workpath
)
644 struct vfsmount
*mnt
= ovl_upper_mnt(ofs
);
645 struct dentry
*workdir
;
646 struct file
*tmpfile
;
647 bool rename_whiteout
;
652 err
= mnt_want_write(mnt
);
656 workdir
= ovl_workdir_create(ofs
, OVL_WORKDIR_NAME
, false);
657 err
= PTR_ERR(workdir
);
658 if (IS_ERR_OR_NULL(workdir
))
661 ofs
->workdir
= workdir
;
663 err
= ovl_setup_trap(sb
, ofs
->workdir
, &ofs
->workdir_trap
, "workdir");
668 * Upper should support d_type, else whiteouts are visible. Given
669 * workdir and upper are on same fs, we can do iterate_dir() on
670 * workdir. This check requires successful creation of workdir in
673 err
= ovl_check_d_type_supported(workpath
);
679 pr_warn("upper fs needs to support d_type.\n");
681 /* Check if upper/work fs supports O_TMPFILE */
682 tmpfile
= ovl_do_tmpfile(ofs
, ofs
->workdir
, S_IFREG
| 0);
683 ofs
->tmpfile
= !IS_ERR(tmpfile
);
687 pr_warn("upper fs does not support tmpfile.\n");
690 /* Check if upper/work fs supports RENAME_WHITEOUT */
691 err
= ovl_check_rename_whiteout(ofs
);
695 rename_whiteout
= err
;
696 if (!rename_whiteout
)
697 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
700 * Check if upper/work fs supports (trusted|user).overlay.* xattr
702 err
= ovl_setxattr(ofs
, ofs
->workdir
, OVL_XATTR_OPAQUE
, "0", 1);
704 pr_warn("failed to set xattr on upper\n");
706 if (ovl_redirect_follow(ofs
)) {
707 ofs
->config
.redirect_mode
= OVL_REDIRECT_NOFOLLOW
;
708 pr_warn("...falling back to redirect_dir=nofollow.\n");
710 if (ofs
->config
.metacopy
) {
711 ofs
->config
.metacopy
= false;
712 pr_warn("...falling back to metacopy=off.\n");
714 if (ofs
->config
.index
) {
715 ofs
->config
.index
= false;
716 pr_warn("...falling back to index=off.\n");
718 if (ovl_has_fsid(ofs
)) {
719 ofs
->config
.uuid
= OVL_UUID_NULL
;
720 pr_warn("...falling back to uuid=null.\n");
723 * xattr support is required for persistent st_ino.
724 * Without persistent st_ino, xino=auto falls back to xino=off.
726 if (ofs
->config
.xino
== OVL_XINO_AUTO
) {
727 ofs
->config
.xino
= OVL_XINO_OFF
;
728 pr_warn("...falling back to xino=off.\n");
730 if (err
== -EPERM
&& !ofs
->config
.userxattr
)
731 pr_info("try mounting with 'userxattr' option\n");
734 ovl_removexattr(ofs
, ofs
->workdir
, OVL_XATTR_OPAQUE
);
738 * We allowed sub-optimal upper fs configuration and don't want to break
739 * users over kernel upgrade, but we never allowed remote upper fs, so
740 * we can enforce strict requirements for remote upper fs.
742 if (ovl_dentry_remote(ofs
->workdir
) &&
743 (!d_type
|| !rename_whiteout
|| ofs
->noxattr
)) {
744 pr_err("upper fs missing required features.\n");
750 * For volatile mount, create a incompat/volatile/dirty file to keep
753 if (ofs
->config
.ovl_volatile
) {
754 err
= ovl_create_volatile_dirty(ofs
);
756 pr_err("Failed to create volatile/dirty file.\n");
761 /* Check if upper/work fs supports file handles */
762 fh_type
= ovl_can_decode_fh(ofs
->workdir
->d_sb
);
763 if (ofs
->config
.index
&& !fh_type
) {
764 ofs
->config
.index
= false;
765 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
767 ofs
->nofh
|= !fh_type
;
769 /* Check if upper fs has 32bit inode numbers */
770 if (fh_type
!= FILEID_INO32_GEN
)
773 /* NFS export of r/w mount depends on index */
774 if (ofs
->config
.nfs_export
&& !ofs
->config
.index
) {
775 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
776 ofs
->config
.nfs_export
= false;
783 static int ovl_get_workdir(struct super_block
*sb
, struct ovl_fs
*ofs
,
784 const struct path
*upperpath
,
785 const struct path
*workpath
)
790 if (upperpath
->mnt
!= workpath
->mnt
) {
791 pr_err("workdir and upperdir must reside under the same mount\n");
794 if (!ovl_workdir_ok(workpath
->dentry
, upperpath
->dentry
)) {
795 pr_err("workdir and upperdir must be separate subtrees\n");
799 ofs
->workbasedir
= dget(workpath
->dentry
);
801 if (ovl_inuse_trylock(ofs
->workbasedir
)) {
802 ofs
->workdir_locked
= true;
804 err
= ovl_report_in_use(ofs
, "workdir");
809 err
= ovl_setup_trap(sb
, ofs
->workbasedir
, &ofs
->workbasedir_trap
,
814 return ovl_make_workdir(sb
, ofs
, workpath
);
817 static int ovl_get_indexdir(struct super_block
*sb
, struct ovl_fs
*ofs
,
818 struct ovl_entry
*oe
, const struct path
*upperpath
)
820 struct vfsmount
*mnt
= ovl_upper_mnt(ofs
);
821 struct dentry
*indexdir
;
822 struct dentry
*origin
= ovl_lowerstack(oe
)->dentry
;
823 const struct ovl_fh
*fh
;
826 fh
= ovl_get_origin_fh(ofs
, origin
);
830 err
= mnt_want_write(mnt
);
834 /* Verify lower root is upper root origin */
835 err
= ovl_verify_origin_fh(ofs
, upperpath
->dentry
, fh
, true);
837 pr_err("failed to verify upper root origin\n");
841 /* index dir will act also as workdir */
842 iput(ofs
->workdir_trap
);
843 ofs
->workdir_trap
= NULL
;
846 indexdir
= ovl_workdir_create(ofs
, OVL_INDEXDIR_NAME
, true);
847 if (IS_ERR(indexdir
)) {
848 err
= PTR_ERR(indexdir
);
849 } else if (indexdir
) {
850 ofs
->workdir
= indexdir
;
851 err
= ovl_setup_trap(sb
, indexdir
, &ofs
->workdir_trap
,
857 * Verify upper root is exclusively associated with index dir.
858 * Older kernels stored upper fh in ".overlay.origin"
859 * xattr. If that xattr exists, verify that it is a match to
860 * upper dir file handle. In any case, verify or set xattr
861 * ".overlay.upper" to indicate that index may have
864 if (ovl_check_origin_xattr(ofs
, indexdir
)) {
865 err
= ovl_verify_origin_xattr(ofs
, indexdir
,
867 upperpath
->dentry
, true,
870 pr_err("failed to verify index dir 'origin' xattr\n");
872 err
= ovl_verify_upper(ofs
, indexdir
, upperpath
->dentry
, true);
874 pr_err("failed to verify index dir 'upper' xattr\n");
876 /* Cleanup bad/stale/orphan index entries */
878 err
= ovl_indexdir_cleanup(ofs
);
880 if (err
|| !indexdir
)
881 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
890 static bool ovl_lower_uuid_ok(struct ovl_fs
*ofs
, const uuid_t
*uuid
)
894 if (!ofs
->config
.nfs_export
&& !ovl_upper_mnt(ofs
))
898 * We allow using single lower with null uuid for index and nfs_export
899 * for example to support those features with single lower squashfs.
900 * To avoid regressions in setups of overlay with re-formatted lower
901 * squashfs, do not allow decoding origin with lower null uuid unless
902 * user opted-in to one of the new features that require following the
903 * lower inode of non-dir upper.
905 if (ovl_allow_offline_changes(ofs
) && uuid_is_null(uuid
))
908 for (i
= 0; i
< ofs
->numfs
; i
++) {
910 * We use uuid to associate an overlay lower file handle with a
911 * lower layer, so we can accept lower fs with null uuid as long
912 * as all lower layers with null uuid are on the same fs.
913 * if we detect multiple lower fs with the same uuid, we
914 * disable lower file handle decoding on all of them.
916 if (ofs
->fs
[i
].is_lower
&&
917 uuid_equal(&ofs
->fs
[i
].sb
->s_uuid
, uuid
)) {
918 ofs
->fs
[i
].bad_uuid
= true;
925 /* Get a unique fsid for the layer */
926 static int ovl_get_fsid(struct ovl_fs
*ofs
, const struct path
*path
)
928 struct super_block
*sb
= path
->mnt
->mnt_sb
;
932 bool bad_uuid
= false;
935 for (i
= 0; i
< ofs
->numfs
; i
++) {
936 if (ofs
->fs
[i
].sb
== sb
)
940 if (!ovl_lower_uuid_ok(ofs
, &sb
->s_uuid
)) {
942 if (ofs
->config
.xino
== OVL_XINO_AUTO
) {
943 ofs
->config
.xino
= OVL_XINO_OFF
;
946 if (ofs
->config
.index
|| ofs
->config
.nfs_export
) {
947 ofs
->config
.index
= false;
948 ofs
->config
.nfs_export
= false;
952 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
953 uuid_is_null(&sb
->s_uuid
) ? "null" :
955 path
->dentry
, ovl_xino_mode(&ofs
->config
));
959 err
= get_anon_bdev(&dev
);
961 pr_err("failed to get anonymous bdev for lowerpath\n");
965 ofs
->fs
[ofs
->numfs
].sb
= sb
;
966 ofs
->fs
[ofs
->numfs
].pseudo_dev
= dev
;
967 ofs
->fs
[ofs
->numfs
].bad_uuid
= bad_uuid
;
973 * The fsid after the last lower fsid is used for the data layers.
974 * It is a "null fs" with a null sb, null uuid, and no pseudo dev.
976 static int ovl_get_data_fsid(struct ovl_fs
*ofs
)
982 static int ovl_get_layers(struct super_block
*sb
, struct ovl_fs
*ofs
,
983 struct ovl_fs_context
*ctx
, struct ovl_layer
*layers
)
987 size_t nr_merged_lower
;
989 ofs
->fs
= kcalloc(ctx
->nr
+ 2, sizeof(struct ovl_sb
), GFP_KERNEL
);
994 * idx/fsid 0 are reserved for upper fs even with lower only overlay
995 * and the last fsid is reserved for "null fs" of the data layers.
1000 * All lower layers that share the same fs as upper layer, use the same
1001 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1002 * only overlay to simplify ovl_fs_free().
1003 * is_lower will be set if upper fs is shared with a lower layer.
1005 err
= get_anon_bdev(&ofs
->fs
[0].pseudo_dev
);
1007 pr_err("failed to get anonymous bdev for upper fs\n");
1011 if (ovl_upper_mnt(ofs
)) {
1012 ofs
->fs
[0].sb
= ovl_upper_mnt(ofs
)->mnt_sb
;
1013 ofs
->fs
[0].is_lower
= false;
1016 nr_merged_lower
= ctx
->nr
- ctx
->nr_data
;
1017 for (i
= 0; i
< ctx
->nr
; i
++) {
1018 struct ovl_fs_context_layer
*l
= &ctx
->lower
[i
];
1019 struct vfsmount
*mnt
;
1023 if (i
< nr_merged_lower
)
1024 fsid
= ovl_get_fsid(ofs
, &l
->path
);
1026 fsid
= ovl_get_data_fsid(ofs
);
1031 * Check if lower root conflicts with this overlay layers before
1032 * checking if it is in-use as upperdir/workdir of "another"
1033 * mount, because we do not bother to check in ovl_is_inuse() if
1034 * the upperdir/workdir is in fact in-use by our
1037 err
= ovl_setup_trap(sb
, l
->path
.dentry
, &trap
, "lowerdir");
1041 if (ovl_is_inuse(l
->path
.dentry
)) {
1042 err
= ovl_report_in_use(ofs
, "lowerdir");
1049 mnt
= clone_private_mount(&l
->path
);
1052 pr_err("failed to clone lowerpath\n");
1058 * Make lower layers R/O. That way fchmod/fchown on lower file
1059 * will fail instead of modifying lower fs.
1061 mnt
->mnt_flags
|= MNT_READONLY
| MNT_NOATIME
;
1063 layers
[ofs
->numlayer
].trap
= trap
;
1064 layers
[ofs
->numlayer
].mnt
= mnt
;
1065 layers
[ofs
->numlayer
].idx
= ofs
->numlayer
;
1066 layers
[ofs
->numlayer
].fsid
= fsid
;
1067 layers
[ofs
->numlayer
].fs
= &ofs
->fs
[fsid
];
1068 /* Store for printing lowerdir=... in ovl_show_options() */
1069 ofs
->config
.lowerdirs
[ofs
->numlayer
] = l
->name
;
1072 ofs
->fs
[fsid
].is_lower
= true;
1076 * When all layers on same fs, overlay can use real inode numbers.
1077 * With mount option "xino=<on|auto>", mounter declares that there are
1078 * enough free high bits in underlying fs to hold the unique fsid.
1079 * If overlayfs does encounter underlying inodes using the high xino
1080 * bits reserved for fsid, it emits a warning and uses the original
1081 * inode number or a non persistent inode number allocated from a
1084 if (ofs
->numfs
- !ovl_upper_mnt(ofs
) == 1) {
1085 if (ofs
->config
.xino
== OVL_XINO_ON
)
1086 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1088 } else if (ofs
->config
.xino
== OVL_XINO_OFF
) {
1089 ofs
->xino_mode
= -1;
1090 } else if (ofs
->xino_mode
< 0) {
1092 * This is a roundup of number of bits needed for encoding
1093 * fsid, where fsid 0 is reserved for upper fs (even with
1094 * lower only overlay) +1 extra bit is reserved for the non
1095 * persistent inode number range that is used for resolving
1096 * xino lower bits overflow.
1098 BUILD_BUG_ON(ilog2(OVL_MAX_STACK
) > 30);
1099 ofs
->xino_mode
= ilog2(ofs
->numfs
- 1) + 2;
1102 if (ofs
->xino_mode
> 0) {
1103 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1110 static struct ovl_entry
*ovl_get_lowerstack(struct super_block
*sb
,
1111 struct ovl_fs_context
*ctx
,
1113 struct ovl_layer
*layers
)
1117 size_t nr_merged_lower
;
1118 struct ovl_entry
*oe
;
1119 struct ovl_path
*lowerstack
;
1121 struct ovl_fs_context_layer
*l
;
1123 if (!ofs
->config
.upperdir
&& ctx
->nr
== 1) {
1124 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1125 return ERR_PTR(-EINVAL
);
1129 for (i
= 0; i
< ctx
->nr
; i
++) {
1132 err
= ovl_lower_dir(l
->name
, &l
->path
, ofs
, &sb
->s_stack_depth
);
1134 return ERR_PTR(err
);
1138 sb
->s_stack_depth
++;
1139 if (sb
->s_stack_depth
> FILESYSTEM_MAX_STACK_DEPTH
) {
1140 pr_err("maximum fs stacking depth exceeded\n");
1141 return ERR_PTR(err
);
1144 err
= ovl_get_layers(sb
, ofs
, ctx
, layers
);
1146 return ERR_PTR(err
);
1149 /* Data-only layers are not merged in root directory */
1150 nr_merged_lower
= ctx
->nr
- ctx
->nr_data
;
1151 oe
= ovl_alloc_entry(nr_merged_lower
);
1153 return ERR_PTR(err
);
1155 lowerstack
= ovl_lowerstack(oe
);
1156 for (i
= 0; i
< nr_merged_lower
; i
++) {
1158 lowerstack
[i
].dentry
= dget(l
->path
.dentry
);
1159 lowerstack
[i
].layer
= &ofs
->layers
[i
+ 1];
1161 ofs
->numdatalayer
= ctx
->nr_data
;
1167 * Check if this layer root is a descendant of:
1168 * - another layer of this overlayfs instance
1169 * - upper/work dir of any overlayfs instance
1171 static int ovl_check_layer(struct super_block
*sb
, struct ovl_fs
*ofs
,
1172 struct dentry
*dentry
, const char *name
,
1175 struct dentry
*next
= dentry
, *parent
;
1181 parent
= dget_parent(next
);
1183 /* Walk back ancestors to root (inclusive) looking for traps */
1184 while (!err
&& parent
!= next
) {
1185 if (is_lower
&& ovl_lookup_trap_inode(sb
, parent
)) {
1187 pr_err("overlapping %s path\n", name
);
1188 } else if (ovl_is_inuse(parent
)) {
1189 err
= ovl_report_in_use(ofs
, name
);
1192 parent
= dget_parent(next
);
1202 * Check if any of the layers or work dirs overlap.
1204 static int ovl_check_overlapping_layers(struct super_block
*sb
,
1209 if (ovl_upper_mnt(ofs
)) {
1210 err
= ovl_check_layer(sb
, ofs
, ovl_upper_mnt(ofs
)->mnt_root
,
1216 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1217 * this instance and covers overlapping work and index dirs,
1218 * unless work or index dir have been moved since created inside
1219 * workbasedir. In that case, we already have their traps in
1220 * inode cache and we will catch that case on lookup.
1222 err
= ovl_check_layer(sb
, ofs
, ofs
->workbasedir
, "workdir",
1228 for (i
= 1; i
< ofs
->numlayer
; i
++) {
1229 err
= ovl_check_layer(sb
, ofs
,
1230 ofs
->layers
[i
].mnt
->mnt_root
,
1239 static struct dentry
*ovl_get_root(struct super_block
*sb
,
1240 struct dentry
*upperdentry
,
1241 struct ovl_entry
*oe
)
1243 struct dentry
*root
;
1244 struct ovl_fs
*ofs
= OVL_FS(sb
);
1245 struct ovl_path
*lowerpath
= ovl_lowerstack(oe
);
1246 unsigned long ino
= d_inode(lowerpath
->dentry
)->i_ino
;
1247 int fsid
= lowerpath
->layer
->fsid
;
1248 struct ovl_inode_params oip
= {
1249 .upperdentry
= upperdentry
,
1253 root
= d_make_root(ovl_new_inode(sb
, S_IFDIR
, 0));
1258 /* Root inode uses upper st_ino/i_ino */
1259 ino
= d_inode(upperdentry
)->i_ino
;
1261 ovl_dentry_set_upper_alias(root
);
1262 if (ovl_is_impuredir(sb
, upperdentry
))
1263 ovl_set_flag(OVL_IMPURE
, d_inode(root
));
1266 /* Look for xwhiteouts marker except in the lowermost layer */
1267 for (int i
= 0; i
< ovl_numlower(oe
) - 1; i
++, lowerpath
++) {
1268 struct path path
= {
1269 .mnt
= lowerpath
->layer
->mnt
,
1270 .dentry
= lowerpath
->dentry
,
1273 /* overlay.opaque=x means xwhiteouts directory */
1274 if (ovl_get_opaquedir_val(ofs
, &path
) == 'x') {
1275 ovl_layer_set_xwhiteouts(ofs
, lowerpath
->layer
);
1276 ovl_dentry_set_xwhiteouts(root
);
1280 /* Root is always merge -> can have whiteouts */
1281 ovl_set_flag(OVL_WHITEOUTS
, d_inode(root
));
1282 ovl_dentry_set_flag(OVL_E_CONNECTED
, root
);
1283 ovl_set_upperdata(d_inode(root
));
1284 ovl_inode_init(d_inode(root
), &oip
, ino
, fsid
);
1285 ovl_dentry_init_flags(root
, upperdentry
, oe
, DCACHE_OP_WEAK_REVALIDATE
);
1286 /* root keeps a reference of upperdentry */
1292 int ovl_fill_super(struct super_block
*sb
, struct fs_context
*fc
)
1294 struct ovl_fs
*ofs
= sb
->s_fs_info
;
1295 struct ovl_fs_context
*ctx
= fc
->fs_private
;
1296 struct dentry
*root_dentry
;
1297 struct ovl_entry
*oe
;
1298 struct ovl_layer
*layers
;
1303 if (WARN_ON(fc
->user_ns
!= current_user_ns()))
1306 sb
->s_d_op
= &ovl_dentry_operations
;
1309 ofs
->creator_cred
= cred
= prepare_creds();
1313 err
= ovl_fs_params_verify(ctx
, &ofs
->config
);
1319 if (!(fc
->sb_flags
& SB_SILENT
))
1320 pr_err("missing 'lowerdir'\n");
1325 layers
= kcalloc(ctx
->nr
+ 1, sizeof(struct ovl_layer
), GFP_KERNEL
);
1329 ofs
->config
.lowerdirs
= kcalloc(ctx
->nr
+ 1, sizeof(char *), GFP_KERNEL
);
1330 if (!ofs
->config
.lowerdirs
) {
1334 ofs
->layers
= layers
;
1336 * Layer 0 is reserved for upper even if there's no upper.
1337 * config.lowerdirs[0] is used for storing the user provided colon
1338 * separated lowerdir string.
1340 ofs
->config
.lowerdirs
[0] = ctx
->lowerdir_all
;
1341 ctx
->lowerdir_all
= NULL
;
1344 sb
->s_stack_depth
= 0;
1345 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1346 atomic_long_set(&ofs
->last_ino
, 1);
1347 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1348 if (ofs
->config
.xino
!= OVL_XINO_OFF
) {
1349 ofs
->xino_mode
= BITS_PER_LONG
- 32;
1350 if (!ofs
->xino_mode
) {
1351 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1352 ofs
->config
.xino
= OVL_XINO_OFF
;
1356 /* alloc/destroy_inode needed for setting up traps in inode cache */
1357 sb
->s_op
= &ovl_super_operations
;
1359 if (ofs
->config
.upperdir
) {
1360 struct super_block
*upper_sb
;
1363 if (!ofs
->config
.workdir
) {
1364 pr_err("missing 'workdir'\n");
1368 err
= ovl_get_upper(sb
, ofs
, &layers
[0], &ctx
->upper
);
1372 upper_sb
= ovl_upper_mnt(ofs
)->mnt_sb
;
1373 if (!ovl_should_sync(ofs
)) {
1374 ofs
->errseq
= errseq_sample(&upper_sb
->s_wb_err
);
1375 if (errseq_check(&upper_sb
->s_wb_err
, ofs
->errseq
)) {
1377 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1382 err
= ovl_get_workdir(sb
, ofs
, &ctx
->upper
, &ctx
->work
);
1387 sb
->s_flags
|= SB_RDONLY
;
1389 sb
->s_stack_depth
= upper_sb
->s_stack_depth
;
1390 sb
->s_time_gran
= upper_sb
->s_time_gran
;
1392 oe
= ovl_get_lowerstack(sb
, ctx
, ofs
, layers
);
1397 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1398 if (!ovl_upper_mnt(ofs
))
1399 sb
->s_flags
|= SB_RDONLY
;
1401 if (!ovl_origin_uuid(ofs
) && ofs
->numfs
> 1) {
1402 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=null.\n");
1403 ofs
->config
.uuid
= OVL_UUID_NULL
;
1404 } else if (ovl_has_fsid(ofs
) && ovl_upper_mnt(ofs
)) {
1405 /* Use per instance persistent uuid/fsid */
1406 ovl_init_uuid_xattr(sb
, ofs
, &ctx
->upper
);
1409 if (!ovl_force_readonly(ofs
) && ofs
->config
.index
) {
1410 err
= ovl_get_indexdir(sb
, ofs
, oe
, &ctx
->upper
);
1414 /* Force r/o mount with no index dir */
1416 sb
->s_flags
|= SB_RDONLY
;
1419 err
= ovl_check_overlapping_layers(sb
, ofs
);
1423 /* Show index=off in /proc/mounts for forced r/o mount */
1424 if (!ofs
->workdir
) {
1425 ofs
->config
.index
= false;
1426 if (ovl_upper_mnt(ofs
) && ofs
->config
.nfs_export
) {
1427 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1428 ofs
->config
.nfs_export
= false;
1432 if (ofs
->config
.metacopy
&& ofs
->config
.nfs_export
) {
1433 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1434 ofs
->config
.nfs_export
= false;
1438 * Support encoding decodable file handles with nfs_export=on
1439 * and encoding non-decodable file handles with nfs_export=off
1440 * if all layers support file handles.
1442 if (ofs
->config
.nfs_export
)
1443 sb
->s_export_op
= &ovl_export_operations
;
1444 else if (!ofs
->nofh
)
1445 sb
->s_export_op
= &ovl_export_fid_operations
;
1447 /* Never override disk quota limits or use reserved space */
1448 cap_lower(cred
->cap_effective
, CAP_SYS_RESOURCE
);
1450 sb
->s_magic
= OVERLAYFS_SUPER_MAGIC
;
1451 sb
->s_xattr
= ovl_xattr_handlers(ofs
);
1452 sb
->s_fs_info
= ofs
;
1453 #ifdef CONFIG_FS_POSIX_ACL
1454 sb
->s_flags
|= SB_POSIXACL
;
1456 sb
->s_iflags
|= SB_I_SKIP_SYNC
;
1458 * Ensure that umask handling is done by the filesystems used
1459 * for the the upper layer instead of overlayfs as that would
1460 * lead to unexpected results.
1462 sb
->s_iflags
|= SB_I_NOUMASK
;
1463 sb
->s_iflags
|= SB_I_EVM_UNSUPPORTED
;
1466 root_dentry
= ovl_get_root(sb
, ctx
->upper
.dentry
, oe
);
1470 sb
->s_root
= root_dentry
;
1478 sb
->s_fs_info
= NULL
;
1482 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
|
1507 ovl_inode_init_once
);
1508 if (ovl_inode_cachep
== NULL
)
1511 err
= register_filesystem(&ovl_fs_type
);
1515 kmem_cache_destroy(ovl_inode_cachep
);
1520 static void __exit
ovl_exit(void)
1522 unregister_filesystem(&ovl_fs_type
);
1525 * Make sure all delayed rcu free inodes are flushed before we
1529 kmem_cache_destroy(ovl_inode_cachep
);
1532 module_init(ovl_init
);
1533 module_exit(ovl_exit
);