3 * Copyright (C) 2011 Novell Inc.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published by
7 * the Free Software Foundation.
11 #include <linux/namei.h>
12 #include <linux/xattr.h>
13 #include <linux/security.h>
14 #include <linux/mount.h>
15 #include <linux/slab.h>
16 #include <linux/parser.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/statfs.h>
20 #include <linux/seq_file.h>
21 #include "overlayfs.h"
23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
24 MODULE_DESCRIPTION("Overlay filesystem");
25 MODULE_LICENSE("GPL");
27 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
35 /* private information held for overlayfs's superblock */
37 struct vfsmount
*upper_mnt
;
39 struct vfsmount
**lower_mnt
;
40 struct dentry
*workdir
;
42 /* pathnames of lower and upper dirs, for show_options */
43 struct ovl_config config
;
48 /* private information held for every overlayfs dentry */
50 struct dentry
*__upperdentry
;
51 struct ovl_dir_cache
*cache
;
60 struct path lowerstack
[];
63 #define OVL_MAX_STACK 500
65 static struct dentry
*__ovl_dentry_lower(struct ovl_entry
*oe
)
67 return oe
->numlower
? oe
->lowerstack
[0].dentry
: NULL
;
70 enum ovl_path_type
ovl_path_type(struct dentry
*dentry
)
72 struct ovl_entry
*oe
= dentry
->d_fsdata
;
73 enum ovl_path_type type
= 0;
75 if (oe
->__upperdentry
) {
76 type
= __OVL_PATH_UPPER
;
79 if (S_ISDIR(dentry
->d_inode
->i_mode
))
80 type
|= __OVL_PATH_MERGE
;
81 } else if (!oe
->opaque
) {
82 type
|= __OVL_PATH_PURE
;
86 type
|= __OVL_PATH_MERGE
;
91 static struct dentry
*ovl_upperdentry_dereference(struct ovl_entry
*oe
)
93 return lockless_dereference(oe
->__upperdentry
);
96 void ovl_path_upper(struct dentry
*dentry
, struct path
*path
)
98 struct ovl_fs
*ofs
= dentry
->d_sb
->s_fs_info
;
99 struct ovl_entry
*oe
= dentry
->d_fsdata
;
101 path
->mnt
= ofs
->upper_mnt
;
102 path
->dentry
= ovl_upperdentry_dereference(oe
);
105 enum ovl_path_type
ovl_path_real(struct dentry
*dentry
, struct path
*path
)
107 enum ovl_path_type type
= ovl_path_type(dentry
);
109 if (!OVL_TYPE_UPPER(type
))
110 ovl_path_lower(dentry
, path
);
112 ovl_path_upper(dentry
, path
);
117 struct dentry
*ovl_dentry_upper(struct dentry
*dentry
)
119 struct ovl_entry
*oe
= dentry
->d_fsdata
;
121 return ovl_upperdentry_dereference(oe
);
124 struct dentry
*ovl_dentry_lower(struct dentry
*dentry
)
126 struct ovl_entry
*oe
= dentry
->d_fsdata
;
128 return __ovl_dentry_lower(oe
);
131 struct dentry
*ovl_dentry_real(struct dentry
*dentry
)
133 struct ovl_entry
*oe
= dentry
->d_fsdata
;
134 struct dentry
*realdentry
;
136 realdentry
= ovl_upperdentry_dereference(oe
);
138 realdentry
= __ovl_dentry_lower(oe
);
143 struct dentry
*ovl_entry_real(struct ovl_entry
*oe
, bool *is_upper
)
145 struct dentry
*realdentry
;
147 realdentry
= ovl_upperdentry_dereference(oe
);
151 realdentry
= __ovl_dentry_lower(oe
);
157 struct ovl_dir_cache
*ovl_dir_cache(struct dentry
*dentry
)
159 struct ovl_entry
*oe
= dentry
->d_fsdata
;
164 void ovl_set_dir_cache(struct dentry
*dentry
, struct ovl_dir_cache
*cache
)
166 struct ovl_entry
*oe
= dentry
->d_fsdata
;
171 void ovl_path_lower(struct dentry
*dentry
, struct path
*path
)
173 struct ovl_entry
*oe
= dentry
->d_fsdata
;
175 *path
= oe
->numlower
? oe
->lowerstack
[0] : (struct path
) { NULL
, NULL
};
178 int ovl_want_write(struct dentry
*dentry
)
180 struct ovl_fs
*ofs
= dentry
->d_sb
->s_fs_info
;
181 return mnt_want_write(ofs
->upper_mnt
);
184 void ovl_drop_write(struct dentry
*dentry
)
186 struct ovl_fs
*ofs
= dentry
->d_sb
->s_fs_info
;
187 mnt_drop_write(ofs
->upper_mnt
);
190 struct dentry
*ovl_workdir(struct dentry
*dentry
)
192 struct ovl_fs
*ofs
= dentry
->d_sb
->s_fs_info
;
196 bool ovl_dentry_is_opaque(struct dentry
*dentry
)
198 struct ovl_entry
*oe
= dentry
->d_fsdata
;
202 void ovl_dentry_set_opaque(struct dentry
*dentry
, bool opaque
)
204 struct ovl_entry
*oe
= dentry
->d_fsdata
;
208 void ovl_dentry_update(struct dentry
*dentry
, struct dentry
*upperdentry
)
210 struct ovl_entry
*oe
= dentry
->d_fsdata
;
212 WARN_ON(!mutex_is_locked(&upperdentry
->d_parent
->d_inode
->i_mutex
));
213 WARN_ON(oe
->__upperdentry
);
214 BUG_ON(!upperdentry
->d_inode
);
216 * Make sure upperdentry is consistent before making it visible to
217 * ovl_upperdentry_dereference().
220 oe
->__upperdentry
= upperdentry
;
223 void ovl_dentry_version_inc(struct dentry
*dentry
)
225 struct ovl_entry
*oe
= dentry
->d_fsdata
;
227 WARN_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
231 u64
ovl_dentry_version_get(struct dentry
*dentry
)
233 struct ovl_entry
*oe
= dentry
->d_fsdata
;
235 WARN_ON(!mutex_is_locked(&dentry
->d_inode
->i_mutex
));
239 bool ovl_is_whiteout(struct dentry
*dentry
)
241 struct inode
*inode
= dentry
->d_inode
;
243 return inode
&& IS_WHITEOUT(inode
);
246 static bool ovl_is_opaquedir(struct dentry
*dentry
)
250 struct inode
*inode
= dentry
->d_inode
;
252 if (!S_ISDIR(inode
->i_mode
) || !inode
->i_op
->getxattr
)
255 res
= inode
->i_op
->getxattr(dentry
, OVL_XATTR_OPAQUE
, &val
, 1);
256 if (res
== 1 && val
== 'y')
262 static void ovl_dentry_release(struct dentry
*dentry
)
264 struct ovl_entry
*oe
= dentry
->d_fsdata
;
269 dput(oe
->__upperdentry
);
270 for (i
= 0; i
< oe
->numlower
; i
++)
271 dput(oe
->lowerstack
[i
].dentry
);
276 static const struct dentry_operations ovl_dentry_operations
= {
277 .d_release
= ovl_dentry_release
,
280 static struct ovl_entry
*ovl_alloc_entry(unsigned int numlower
)
282 size_t size
= offsetof(struct ovl_entry
, lowerstack
[numlower
]);
283 struct ovl_entry
*oe
= kzalloc(size
, GFP_KERNEL
);
286 oe
->numlower
= numlower
;
291 static inline struct dentry
*ovl_lookup_real(struct dentry
*dir
,
294 struct dentry
*dentry
;
296 mutex_lock(&dir
->d_inode
->i_mutex
);
297 dentry
= lookup_one_len(name
->name
, dir
, name
->len
);
298 mutex_unlock(&dir
->d_inode
->i_mutex
);
300 if (IS_ERR(dentry
)) {
301 if (PTR_ERR(dentry
) == -ENOENT
)
303 } else if (!dentry
->d_inode
) {
311 * Returns next layer in stack starting from top.
312 * Returns -1 if this is the last layer.
314 int ovl_path_next(int idx
, struct dentry
*dentry
, struct path
*path
)
316 struct ovl_entry
*oe
= dentry
->d_fsdata
;
320 ovl_path_upper(dentry
, path
);
322 return oe
->numlower
? 1 : -1;
325 BUG_ON(idx
> oe
->numlower
);
326 *path
= oe
->lowerstack
[idx
- 1];
328 return (idx
< oe
->numlower
) ? idx
+ 1 : -1;
331 struct dentry
*ovl_lookup(struct inode
*dir
, struct dentry
*dentry
,
334 struct ovl_entry
*oe
;
335 struct ovl_entry
*poe
= dentry
->d_parent
->d_fsdata
;
336 struct path
*stack
= NULL
;
337 struct dentry
*upperdir
, *upperdentry
= NULL
;
338 unsigned int ctr
= 0;
339 struct inode
*inode
= NULL
;
340 bool upperopaque
= false;
341 struct dentry
*this, *prev
= NULL
;
345 upperdir
= ovl_upperdentry_dereference(poe
);
347 this = ovl_lookup_real(upperdir
, &dentry
->d_name
);
353 if (ovl_is_whiteout(this)) {
357 } else if (poe
->numlower
&& ovl_is_opaquedir(this)) {
361 upperdentry
= prev
= this;
364 if (!upperopaque
&& poe
->numlower
) {
366 stack
= kcalloc(poe
->numlower
, sizeof(struct path
), GFP_KERNEL
);
371 for (i
= 0; !upperopaque
&& i
< poe
->numlower
; i
++) {
373 struct path lowerpath
= poe
->lowerstack
[i
];
376 this = ovl_lookup_real(lowerpath
.dentry
, &dentry
->d_name
);
380 * If it's positive, then treat ENAMETOOLONG as ENOENT.
382 if (err
== -ENAMETOOLONG
&& (upperdentry
|| ctr
))
388 if (ovl_is_whiteout(this)) {
393 * Only makes sense to check opaque dir if this is not the
396 if (i
< poe
->numlower
- 1 && ovl_is_opaquedir(this))
399 * If this is a non-directory then stop here.
401 * FIXME: check for opaqueness maybe better done in remove code.
403 if (!S_ISDIR(this->d_inode
->i_mode
)) {
405 } else if (prev
&& (!S_ISDIR(prev
->d_inode
->i_mode
) ||
406 !S_ISDIR(this->d_inode
->i_mode
))) {
407 if (prev
== upperdentry
)
412 stack
[ctr
].dentry
= this;
413 stack
[ctr
].mnt
= lowerpath
.mnt
;
420 oe
= ovl_alloc_entry(ctr
);
425 if (upperdentry
|| ctr
) {
426 struct dentry
*realdentry
;
428 realdentry
= upperdentry
? upperdentry
: stack
[0].dentry
;
431 inode
= ovl_new_inode(dentry
->d_sb
, realdentry
->d_inode
->i_mode
,
435 ovl_copyattr(realdentry
->d_inode
, inode
);
438 oe
->opaque
= upperopaque
;
439 oe
->__upperdentry
= upperdentry
;
440 memcpy(oe
->lowerstack
, stack
, sizeof(struct path
) * ctr
);
442 dentry
->d_fsdata
= oe
;
443 d_add(dentry
, inode
);
450 for (i
= 0; i
< ctr
; i
++)
451 dput(stack
[i
].dentry
);
459 struct file
*ovl_path_open(struct path
*path
, int flags
)
461 return dentry_open(path
, flags
, current_cred());
464 static void ovl_put_super(struct super_block
*sb
)
466 struct ovl_fs
*ufs
= sb
->s_fs_info
;
470 mntput(ufs
->upper_mnt
);
471 for (i
= 0; i
< ufs
->numlower
; i
++)
472 mntput(ufs
->lower_mnt
[i
]);
474 kfree(ufs
->config
.lowerdir
);
475 kfree(ufs
->config
.upperdir
);
476 kfree(ufs
->config
.workdir
);
482 * @sb: The overlayfs super block
483 * @buf: The struct kstatfs to fill in with stats
485 * Get the filesystem statistics. As writes always target the upper layer
486 * filesystem pass the statfs to the upper filesystem (if it exists)
488 static int ovl_statfs(struct dentry
*dentry
, struct kstatfs
*buf
)
490 struct ovl_fs
*ofs
= dentry
->d_sb
->s_fs_info
;
491 struct dentry
*root_dentry
= dentry
->d_sb
->s_root
;
495 ovl_path_real(root_dentry
, &path
);
497 err
= vfs_statfs(&path
, buf
);
499 buf
->f_namelen
= max(buf
->f_namelen
, ofs
->lower_namelen
);
500 buf
->f_type
= OVERLAYFS_SUPER_MAGIC
;
509 * Prints the mount options for a given superblock.
510 * Returns zero; does not fail.
512 static int ovl_show_options(struct seq_file
*m
, struct dentry
*dentry
)
514 struct super_block
*sb
= dentry
->d_sb
;
515 struct ovl_fs
*ufs
= sb
->s_fs_info
;
517 seq_printf(m
, ",lowerdir=%s", ufs
->config
.lowerdir
);
518 if (ufs
->config
.upperdir
) {
519 seq_printf(m
, ",upperdir=%s", ufs
->config
.upperdir
);
520 seq_printf(m
, ",workdir=%s", ufs
->config
.workdir
);
525 static const struct super_operations ovl_super_operations
= {
526 .put_super
= ovl_put_super
,
527 .statfs
= ovl_statfs
,
528 .show_options
= ovl_show_options
,
538 static const match_table_t ovl_tokens
= {
539 {OPT_LOWERDIR
, "lowerdir=%s"},
540 {OPT_UPPERDIR
, "upperdir=%s"},
541 {OPT_WORKDIR
, "workdir=%s"},
545 static char *ovl_next_opt(char **s
)
553 for (p
= sbegin
; *p
; p
++) {
558 } else if (*p
== ',') {
568 static int ovl_parse_opt(char *opt
, struct ovl_config
*config
)
572 while ((p
= ovl_next_opt(&opt
)) != NULL
) {
574 substring_t args
[MAX_OPT_ARGS
];
579 token
= match_token(p
, ovl_tokens
, args
);
582 kfree(config
->upperdir
);
583 config
->upperdir
= match_strdup(&args
[0]);
584 if (!config
->upperdir
)
589 kfree(config
->lowerdir
);
590 config
->lowerdir
= match_strdup(&args
[0]);
591 if (!config
->lowerdir
)
596 kfree(config
->workdir
);
597 config
->workdir
= match_strdup(&args
[0]);
598 if (!config
->workdir
)
609 #define OVL_WORKDIR_NAME "work"
611 static struct dentry
*ovl_workdir_create(struct vfsmount
*mnt
,
612 struct dentry
*dentry
)
614 struct inode
*dir
= dentry
->d_inode
;
617 bool retried
= false;
619 err
= mnt_want_write(mnt
);
623 mutex_lock_nested(&dir
->i_mutex
, I_MUTEX_PARENT
);
625 work
= lookup_one_len(OVL_WORKDIR_NAME
, dentry
,
626 strlen(OVL_WORKDIR_NAME
));
629 struct kstat stat
= {
639 ovl_cleanup(dir
, work
);
644 err
= ovl_create_real(dir
, work
, &stat
, NULL
, NULL
, true);
649 mutex_unlock(&dir
->i_mutex
);
660 static void ovl_unescape(char *s
)
673 static bool ovl_is_allowed_fs_type(struct dentry
*root
)
675 const struct dentry_operations
*dop
= root
->d_op
;
679 * - automount filesystems
680 * - filesystems with revalidate (FIXME for lower layer)
681 * - filesystems with case insensitive names
684 (dop
->d_manage
|| dop
->d_automount
||
685 dop
->d_revalidate
|| dop
->d_weak_revalidate
||
686 dop
->d_compare
|| dop
->d_hash
)) {
692 static int ovl_mount_dir_noesc(const char *name
, struct path
*path
)
697 pr_err("overlayfs: empty lowerdir\n");
700 err
= kern_path(name
, LOOKUP_FOLLOW
, path
);
702 pr_err("overlayfs: failed to resolve '%s': %i\n", name
, err
);
706 if (!ovl_is_allowed_fs_type(path
->dentry
)) {
707 pr_err("overlayfs: filesystem on '%s' not supported\n", name
);
710 if (!S_ISDIR(path
->dentry
->d_inode
->i_mode
)) {
711 pr_err("overlayfs: '%s' not a directory\n", name
);
722 static int ovl_mount_dir(const char *name
, struct path
*path
)
725 char *tmp
= kstrdup(name
, GFP_KERNEL
);
729 err
= ovl_mount_dir_noesc(tmp
, path
);
735 static int ovl_lower_dir(const char *name
, struct path
*path
, long *namelen
,
739 struct kstatfs statfs
;
741 err
= ovl_mount_dir_noesc(name
, path
);
745 err
= vfs_statfs(path
, &statfs
);
747 pr_err("overlayfs: statfs failed on '%s'\n", name
);
750 *namelen
= max(*namelen
, statfs
.f_namelen
);
751 *stack_depth
= max(*stack_depth
, path
->mnt
->mnt_sb
->s_stack_depth
);
761 /* Workdir should not be subdir of upperdir and vice versa */
762 static bool ovl_workdir_ok(struct dentry
*workdir
, struct dentry
*upperdir
)
766 if (workdir
!= upperdir
) {
767 ok
= (lock_rename(workdir
, upperdir
) == NULL
);
768 unlock_rename(workdir
, upperdir
);
773 static unsigned int ovl_split_lowerdirs(char *str
)
775 unsigned int ctr
= 1;
778 for (s
= d
= str
;; s
++, d
++) {
781 } else if (*s
== ':') {
793 static int ovl_fill_super(struct super_block
*sb
, void *data
, int silent
)
795 struct path upperpath
= { NULL
, NULL
};
796 struct path workpath
= { NULL
, NULL
};
797 struct dentry
*root_dentry
;
798 struct ovl_entry
*oe
;
800 struct path
*stack
= NULL
;
803 unsigned int numlower
;
804 unsigned int stacklen
= 0;
809 ufs
= kzalloc(sizeof(struct ovl_fs
), GFP_KERNEL
);
813 err
= ovl_parse_opt((char *) data
, &ufs
->config
);
815 goto out_free_config
;
818 if (!ufs
->config
.lowerdir
) {
819 pr_err("overlayfs: missing 'lowerdir'\n");
820 goto out_free_config
;
823 sb
->s_stack_depth
= 0;
824 if (ufs
->config
.upperdir
) {
825 /* FIXME: workdir is not needed for a R/O mount */
826 if (!ufs
->config
.workdir
) {
827 pr_err("overlayfs: missing 'workdir'\n");
828 goto out_free_config
;
831 err
= ovl_mount_dir(ufs
->config
.upperdir
, &upperpath
);
833 goto out_free_config
;
835 err
= ovl_mount_dir(ufs
->config
.workdir
, &workpath
);
837 goto out_put_upperpath
;
840 if (upperpath
.mnt
!= workpath
.mnt
) {
841 pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
842 goto out_put_workpath
;
844 if (!ovl_workdir_ok(workpath
.dentry
, upperpath
.dentry
)) {
845 pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
846 goto out_put_workpath
;
848 sb
->s_stack_depth
= upperpath
.mnt
->mnt_sb
->s_stack_depth
;
851 lowertmp
= kstrdup(ufs
->config
.lowerdir
, GFP_KERNEL
);
853 goto out_put_workpath
;
856 stacklen
= ovl_split_lowerdirs(lowertmp
);
857 if (stacklen
> OVL_MAX_STACK
)
858 goto out_free_lowertmp
;
860 stack
= kcalloc(stacklen
, sizeof(struct path
), GFP_KERNEL
);
862 goto out_free_lowertmp
;
865 for (numlower
= 0; numlower
< stacklen
; numlower
++) {
866 err
= ovl_lower_dir(lower
, &stack
[numlower
],
867 &ufs
->lower_namelen
, &sb
->s_stack_depth
);
869 goto out_put_lowerpath
;
871 lower
= strchr(lower
, '\0') + 1;
876 if (sb
->s_stack_depth
> FILESYSTEM_MAX_STACK_DEPTH
) {
877 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
878 goto out_put_lowerpath
;
881 if (ufs
->config
.upperdir
) {
882 ufs
->upper_mnt
= clone_private_mount(&upperpath
);
883 err
= PTR_ERR(ufs
->upper_mnt
);
884 if (IS_ERR(ufs
->upper_mnt
)) {
885 pr_err("overlayfs: failed to clone upperpath\n");
886 goto out_put_lowerpath
;
889 ufs
->workdir
= ovl_workdir_create(ufs
->upper_mnt
, workpath
.dentry
);
890 err
= PTR_ERR(ufs
->workdir
);
891 if (IS_ERR(ufs
->workdir
)) {
892 pr_err("overlayfs: failed to create directory %s/%s\n",
893 ufs
->config
.workdir
, OVL_WORKDIR_NAME
);
894 goto out_put_upper_mnt
;
899 ufs
->lower_mnt
= kcalloc(numlower
, sizeof(struct vfsmount
*), GFP_KERNEL
);
900 if (ufs
->lower_mnt
== NULL
)
901 goto out_put_workdir
;
902 for (i
= 0; i
< numlower
; i
++) {
903 struct vfsmount
*mnt
= clone_private_mount(&stack
[i
]);
907 pr_err("overlayfs: failed to clone lowerpath\n");
908 goto out_put_lower_mnt
;
911 * Make lower_mnt R/O. That way fchmod/fchown on lower file
912 * will fail instead of modifying lower fs.
914 mnt
->mnt_flags
|= MNT_READONLY
;
916 ufs
->lower_mnt
[ufs
->numlower
] = mnt
;
920 /* If the upper fs is r/o or nonexistent, we mark overlayfs r/o too */
921 if (!ufs
->upper_mnt
|| (ufs
->upper_mnt
->mnt_sb
->s_flags
& MS_RDONLY
))
922 sb
->s_flags
|= MS_RDONLY
;
924 sb
->s_d_op
= &ovl_dentry_operations
;
927 oe
= ovl_alloc_entry(numlower
);
929 goto out_put_lower_mnt
;
931 root_dentry
= d_make_root(ovl_new_inode(sb
, S_IFDIR
, oe
));
935 mntput(upperpath
.mnt
);
936 for (i
= 0; i
< numlower
; i
++)
937 mntput(stack
[i
].mnt
);
941 oe
->__upperdentry
= upperpath
.dentry
;
942 for (i
= 0; i
< numlower
; i
++) {
943 oe
->lowerstack
[i
].dentry
= stack
[i
].dentry
;
944 oe
->lowerstack
[i
].mnt
= ufs
->lower_mnt
[i
];
947 root_dentry
->d_fsdata
= oe
;
949 sb
->s_magic
= OVERLAYFS_SUPER_MAGIC
;
950 sb
->s_op
= &ovl_super_operations
;
951 sb
->s_root
= root_dentry
;
959 for (i
= 0; i
< ufs
->numlower
; i
++)
960 mntput(ufs
->lower_mnt
[i
]);
961 kfree(ufs
->lower_mnt
);
965 mntput(ufs
->upper_mnt
);
967 for (i
= 0; i
< numlower
; i
++)
975 path_put(&upperpath
);
977 kfree(ufs
->config
.lowerdir
);
978 kfree(ufs
->config
.upperdir
);
979 kfree(ufs
->config
.workdir
);
985 static struct dentry
*ovl_mount(struct file_system_type
*fs_type
, int flags
,
986 const char *dev_name
, void *raw_data
)
988 return mount_nodev(fs_type
, flags
, raw_data
, ovl_fill_super
);
991 static struct file_system_type ovl_fs_type
= {
992 .owner
= THIS_MODULE
,
995 .kill_sb
= kill_anon_super
,
997 MODULE_ALIAS_FS("overlay");
999 static int __init
ovl_init(void)
1001 return register_filesystem(&ovl_fs_type
);
1004 static void __exit
ovl_exit(void)
1006 unregister_filesystem(&ovl_fs_type
);
1009 module_init(ovl_init
);
1010 module_exit(ovl_exit
);