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.
10 #include <linux/module.h>
12 #include <linux/slab.h>
13 #include <linux/file.h>
14 #include <linux/splice.h>
15 #include <linux/xattr.h>
16 #include <linux/security.h>
17 #include <linux/uaccess.h>
18 #include <linux/sched.h>
19 #include <linux/namei.h>
20 #include <linux/fdtable.h>
21 #include <linux/ratelimit.h>
22 #include "overlayfs.h"
24 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
26 static bool __read_mostly ovl_check_copy_up
;
27 module_param_named(check_copy_up
, ovl_check_copy_up
, bool,
29 MODULE_PARM_DESC(ovl_check_copy_up
,
30 "Warn on copy-up when causing process also has a R/O fd open");
32 static int ovl_check_fd(const void *data
, struct file
*f
, unsigned int fd
)
34 const struct dentry
*dentry
= data
;
36 if (f
->f_inode
== d_inode(dentry
))
37 pr_warn_ratelimited("overlayfs: Warning: Copying up %pD, but open R/O on fd %u which will cease to be coherent [pid=%d %s]\n",
38 f
, fd
, current
->pid
, current
->comm
);
43 * Check the fds open by this process and warn if something like the following
44 * scenario is about to occur:
46 * fd1 = open("foo", O_RDONLY);
47 * fd2 = open("foo", O_RDWR);
49 static void ovl_do_check_copy_up(struct dentry
*dentry
)
51 if (ovl_check_copy_up
)
52 iterate_fd(current
->files
, 0, ovl_check_fd
, dentry
);
55 int ovl_copy_xattr(struct dentry
*old
, struct dentry
*new)
57 ssize_t list_size
, size
, value_size
= 0;
58 char *buf
, *name
, *value
= NULL
;
59 int uninitialized_var(error
);
61 if (!(old
->d_inode
->i_opflags
& IOP_XATTR
) ||
62 !(new->d_inode
->i_opflags
& IOP_XATTR
))
65 list_size
= vfs_listxattr(old
, NULL
, 0);
67 if (list_size
== -EOPNOTSUPP
)
72 buf
= kzalloc(list_size
, GFP_KERNEL
);
76 list_size
= vfs_listxattr(old
, buf
, list_size
);
82 for (name
= buf
; name
< (buf
+ list_size
); name
+= strlen(name
) + 1) {
83 if (ovl_is_private_xattr(name
))
86 size
= vfs_getxattr(old
, name
, value
, value_size
);
88 size
= vfs_getxattr(old
, name
, NULL
, 0);
95 if (size
> value_size
) {
98 new = krealloc(value
, size
, GFP_KERNEL
);
108 error
= security_inode_copy_up_xattr(name
);
109 if (error
< 0 && error
!= -EOPNOTSUPP
)
113 continue; /* Discard */
115 error
= vfs_setxattr(new, name
, value
, size
, 0);
125 static int ovl_copy_up_data(struct path
*old
, struct path
*new, loff_t len
)
127 struct file
*old_file
;
128 struct file
*new_file
;
136 old_file
= ovl_path_open(old
, O_LARGEFILE
| O_RDONLY
);
137 if (IS_ERR(old_file
))
138 return PTR_ERR(old_file
);
140 new_file
= ovl_path_open(new, O_LARGEFILE
| O_WRONLY
);
141 if (IS_ERR(new_file
)) {
142 error
= PTR_ERR(new_file
);
146 /* FIXME: copy up sparse files efficiently */
148 size_t this_len
= OVL_COPY_UP_CHUNK_SIZE
;
154 if (signal_pending_state(TASK_KILLABLE
, current
)) {
159 bytes
= do_splice_direct(old_file
, &old_pos
,
161 this_len
, SPLICE_F_MOVE
);
166 WARN_ON(old_pos
!= new_pos
);
177 static char *ovl_read_symlink(struct dentry
*realdentry
)
181 struct inode
*inode
= realdentry
->d_inode
;
185 if (!inode
->i_op
->readlink
)
189 buf
= (char *) __get_free_page(GFP_KERNEL
);
195 /* The cast to a user pointer is valid due to the set_fs() */
196 res
= inode
->i_op
->readlink(realdentry
,
197 (char __user
*)buf
, PAGE_SIZE
- 1);
200 free_page((unsigned long) buf
);
211 static int ovl_set_timestamps(struct dentry
*upperdentry
, struct kstat
*stat
)
213 struct iattr attr
= {
215 ATTR_ATIME
| ATTR_MTIME
| ATTR_ATIME_SET
| ATTR_MTIME_SET
,
216 .ia_atime
= stat
->atime
,
217 .ia_mtime
= stat
->mtime
,
220 return notify_change(upperdentry
, &attr
, NULL
);
223 int ovl_set_attr(struct dentry
*upperdentry
, struct kstat
*stat
)
227 if (!S_ISLNK(stat
->mode
)) {
228 struct iattr attr
= {
229 .ia_valid
= ATTR_MODE
,
230 .ia_mode
= stat
->mode
,
232 err
= notify_change(upperdentry
, &attr
, NULL
);
235 struct iattr attr
= {
236 .ia_valid
= ATTR_UID
| ATTR_GID
,
240 err
= notify_change(upperdentry
, &attr
, NULL
);
243 ovl_set_timestamps(upperdentry
, stat
);
248 static int ovl_copy_up_locked(struct dentry
*workdir
, struct dentry
*upperdir
,
249 struct dentry
*dentry
, struct path
*lowerpath
,
250 struct kstat
*stat
, const char *link
)
252 struct inode
*wdir
= workdir
->d_inode
;
253 struct inode
*udir
= upperdir
->d_inode
;
254 struct dentry
*newdentry
= NULL
;
255 struct dentry
*upper
= NULL
;
256 umode_t mode
= stat
->mode
;
258 const struct cred
*old_creds
= NULL
;
259 struct cred
*new_creds
= NULL
;
261 newdentry
= ovl_lookup_temp(workdir
, dentry
);
262 err
= PTR_ERR(newdentry
);
263 if (IS_ERR(newdentry
))
266 upper
= lookup_one_len(dentry
->d_name
.name
, upperdir
,
268 err
= PTR_ERR(upper
);
272 err
= security_inode_copy_up(dentry
, &new_creds
);
277 old_creds
= override_creds(new_creds
);
279 /* Can't properly set mode on creation because of the umask */
280 stat
->mode
&= S_IFMT
;
281 err
= ovl_create_real(wdir
, newdentry
, stat
, link
, NULL
, true);
285 revert_creds(old_creds
);
292 if (S_ISREG(stat
->mode
)) {
293 struct path upperpath
;
295 ovl_path_upper(dentry
, &upperpath
);
296 BUG_ON(upperpath
.dentry
!= NULL
);
297 upperpath
.dentry
= newdentry
;
299 err
= ovl_copy_up_data(lowerpath
, &upperpath
, stat
->size
);
304 err
= ovl_copy_xattr(lowerpath
->dentry
, newdentry
);
308 inode_lock(newdentry
->d_inode
);
309 err
= ovl_set_attr(newdentry
, stat
);
310 inode_unlock(newdentry
->d_inode
);
314 err
= ovl_do_rename(wdir
, newdentry
, udir
, upper
, 0);
318 ovl_dentry_update(dentry
, newdentry
);
319 ovl_inode_update(d_inode(dentry
), d_inode(newdentry
));
323 * Non-directores become opaque when copied up.
325 if (!S_ISDIR(stat
->mode
))
326 ovl_dentry_set_opaque(dentry
, true);
335 ovl_cleanup(wdir
, newdentry
);
340 * Copy up a single dentry
342 * Directory renames only allowed on "pure upper" (already created on
343 * upper filesystem, never copied up). Directories which are on lower or
344 * are merged may not be renamed. For these -EXDEV is returned and
345 * userspace has to deal with it. This means, when copying up a
346 * directory we can rely on it and ancestors being stable.
348 * Non-directory renames start with copy up of source if necessary. The
349 * actual rename will only proceed once the copy up was successful. Copy
350 * up uses upper parent i_mutex for exclusion. Since rename can change
351 * d_parent it is possible that the copy up will lock the old parent. At
352 * that point the file will have already been copied up anyway.
354 int ovl_copy_up_one(struct dentry
*parent
, struct dentry
*dentry
,
355 struct path
*lowerpath
, struct kstat
*stat
)
357 struct dentry
*workdir
= ovl_workdir(dentry
);
360 struct path parentpath
;
361 struct dentry
*upperdir
;
362 struct dentry
*upperdentry
;
363 const struct cred
*old_cred
;
366 if (WARN_ON(!workdir
))
369 ovl_do_check_copy_up(lowerpath
->dentry
);
371 ovl_path_upper(parent
, &parentpath
);
372 upperdir
= parentpath
.dentry
;
374 err
= vfs_getattr(&parentpath
, &pstat
);
378 if (S_ISLNK(stat
->mode
)) {
379 link
= ovl_read_symlink(lowerpath
->dentry
);
381 return PTR_ERR(link
);
384 old_cred
= ovl_override_creds(dentry
->d_sb
);
387 if (lock_rename(workdir
, upperdir
) != NULL
) {
388 pr_err("overlayfs: failed to lock workdir+upperdir\n");
391 upperdentry
= ovl_dentry_upper(dentry
);
393 /* Raced with another copy-up? Nothing to do, then... */
398 err
= ovl_copy_up_locked(workdir
, upperdir
, dentry
, lowerpath
,
401 /* Restore timestamps on parent (best effort) */
402 ovl_set_timestamps(upperdir
, &pstat
);
405 unlock_rename(workdir
, upperdir
);
406 revert_creds(old_cred
);
409 free_page((unsigned long) link
);
414 int ovl_copy_up(struct dentry
*dentry
)
421 struct dentry
*parent
;
422 struct path lowerpath
;
424 enum ovl_path_type type
= ovl_path_type(dentry
);
426 if (OVL_TYPE_UPPER(type
))
430 /* find the topmost dentry not yet copied up */
432 parent
= dget_parent(next
);
434 type
= ovl_path_type(parent
);
435 if (OVL_TYPE_UPPER(type
))
442 ovl_path_lower(next
, &lowerpath
);
443 err
= vfs_getattr(&lowerpath
, &stat
);
445 err
= ovl_copy_up_one(parent
, next
, &lowerpath
, &stat
);