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/signal.h>
19 #include <linux/cred.h>
20 #include <linux/namei.h>
21 #include <linux/fdtable.h>
22 #include <linux/ratelimit.h>
23 #include <linux/exportfs.h>
24 #include "overlayfs.h"
25 #include "ovl_entry.h"
27 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
29 static bool __read_mostly ovl_check_copy_up
;
30 module_param_named(check_copy_up
, ovl_check_copy_up
, bool,
32 MODULE_PARM_DESC(ovl_check_copy_up
,
33 "Warn on copy-up when causing process also has a R/O fd open");
35 static int ovl_check_fd(const void *data
, struct file
*f
, unsigned int fd
)
37 const struct dentry
*dentry
= data
;
39 if (file_inode(f
) == d_inode(dentry
))
40 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",
41 f
, fd
, current
->pid
, current
->comm
);
46 * Check the fds open by this process and warn if something like the following
47 * scenario is about to occur:
49 * fd1 = open("foo", O_RDONLY);
50 * fd2 = open("foo", O_RDWR);
52 static void ovl_do_check_copy_up(struct dentry
*dentry
)
54 if (ovl_check_copy_up
)
55 iterate_fd(current
->files
, 0, ovl_check_fd
, dentry
);
58 int ovl_copy_xattr(struct dentry
*old
, struct dentry
*new)
60 ssize_t list_size
, size
, value_size
= 0;
61 char *buf
, *name
, *value
= NULL
;
62 int uninitialized_var(error
);
65 if (!(old
->d_inode
->i_opflags
& IOP_XATTR
) ||
66 !(new->d_inode
->i_opflags
& IOP_XATTR
))
69 list_size
= vfs_listxattr(old
, NULL
, 0);
71 if (list_size
== -EOPNOTSUPP
)
76 buf
= kzalloc(list_size
, GFP_KERNEL
);
80 list_size
= vfs_listxattr(old
, buf
, list_size
);
86 for (name
= buf
; list_size
; name
+= slen
) {
87 slen
= strnlen(name
, list_size
) + 1;
89 /* underlying fs providing us with an broken xattr list? */
90 if (WARN_ON(slen
> list_size
)) {
96 if (ovl_is_private_xattr(name
))
99 size
= vfs_getxattr(old
, name
, value
, value_size
);
101 size
= vfs_getxattr(old
, name
, NULL
, 0);
108 if (size
> value_size
) {
111 new = krealloc(value
, size
, GFP_KERNEL
);
121 error
= security_inode_copy_up_xattr(name
);
122 if (error
< 0 && error
!= -EOPNOTSUPP
)
126 continue; /* Discard */
128 error
= vfs_setxattr(new, name
, value
, size
, 0);
138 static int ovl_copy_up_data(struct path
*old
, struct path
*new, loff_t len
)
140 struct file
*old_file
;
141 struct file
*new_file
;
149 old_file
= ovl_path_open(old
, O_LARGEFILE
| O_RDONLY
);
150 if (IS_ERR(old_file
))
151 return PTR_ERR(old_file
);
153 new_file
= ovl_path_open(new, O_LARGEFILE
| O_WRONLY
);
154 if (IS_ERR(new_file
)) {
155 error
= PTR_ERR(new_file
);
159 /* Try to use clone_file_range to clone up within the same fs */
160 error
= vfs_clone_file_range(old_file
, 0, new_file
, 0, len
);
163 /* Couldn't clone, so now we try to copy the data */
166 /* FIXME: copy up sparse files efficiently */
168 size_t this_len
= OVL_COPY_UP_CHUNK_SIZE
;
174 if (signal_pending_state(TASK_KILLABLE
, current
)) {
179 bytes
= do_splice_direct(old_file
, &old_pos
,
181 this_len
, SPLICE_F_MOVE
);
186 WARN_ON(old_pos
!= new_pos
);
192 error
= vfs_fsync(new_file
, 0);
199 static int ovl_set_timestamps(struct dentry
*upperdentry
, struct kstat
*stat
)
201 struct iattr attr
= {
203 ATTR_ATIME
| ATTR_MTIME
| ATTR_ATIME_SET
| ATTR_MTIME_SET
,
204 .ia_atime
= stat
->atime
,
205 .ia_mtime
= stat
->mtime
,
208 return notify_change(upperdentry
, &attr
, NULL
);
211 int ovl_set_attr(struct dentry
*upperdentry
, struct kstat
*stat
)
215 if (!S_ISLNK(stat
->mode
)) {
216 struct iattr attr
= {
217 .ia_valid
= ATTR_MODE
,
218 .ia_mode
= stat
->mode
,
220 err
= notify_change(upperdentry
, &attr
, NULL
);
223 struct iattr attr
= {
224 .ia_valid
= ATTR_UID
| ATTR_GID
,
228 err
= notify_change(upperdentry
, &attr
, NULL
);
231 ovl_set_timestamps(upperdentry
, stat
);
236 static struct ovl_fh
*ovl_encode_fh(struct dentry
*lower
, uuid_be
*uuid
)
239 int fh_type
, fh_len
, dwords
;
241 int buflen
= MAX_HANDLE_SZ
;
243 buf
= kmalloc(buflen
, GFP_TEMPORARY
);
245 return ERR_PTR(-ENOMEM
);
248 * We encode a non-connectable file handle for non-dir, because we
249 * only need to find the lower inode number and we don't want to pay
250 * the price or reconnecting the dentry.
252 dwords
= buflen
>> 2;
253 fh_type
= exportfs_encode_fh(lower
, buf
, &dwords
, 0);
254 buflen
= (dwords
<< 2);
257 if (WARN_ON(fh_type
< 0) ||
258 WARN_ON(buflen
> MAX_HANDLE_SZ
) ||
259 WARN_ON(fh_type
== FILEID_INVALID
))
262 BUILD_BUG_ON(MAX_HANDLE_SZ
+ offsetof(struct ovl_fh
, fid
) > 255);
263 fh_len
= offsetof(struct ovl_fh
, fid
) + buflen
;
264 fh
= kmalloc(fh_len
, GFP_KERNEL
);
266 fh
= ERR_PTR(-ENOMEM
);
270 fh
->version
= OVL_FH_VERSION
;
271 fh
->magic
= OVL_FH_MAGIC
;
273 fh
->flags
= OVL_FH_FLAG_CPU_ENDIAN
;
276 memcpy(fh
->fid
, buf
, buflen
);
283 static int ovl_set_origin(struct dentry
*dentry
, struct dentry
*lower
,
284 struct dentry
*upper
)
286 struct super_block
*sb
= lower
->d_sb
;
287 uuid_be
*uuid
= (uuid_be
*) &sb
->s_uuid
;
288 const struct ovl_fh
*fh
= NULL
;
292 * When lower layer doesn't support export operations store a 'null' fh,
293 * so we can use the overlay.origin xattr to distignuish between a copy
294 * up and a pure upper inode.
296 if (sb
->s_export_op
&& sb
->s_export_op
->fh_to_dentry
&&
297 uuid_be_cmp(*uuid
, NULL_UUID_BE
)) {
298 fh
= ovl_encode_fh(lower
, uuid
);
304 * Do not fail when upper doesn't support xattrs.
306 err
= ovl_check_setxattr(dentry
, upper
, OVL_XATTR_ORIGIN
, fh
,
307 fh
? fh
->len
: 0, 0);
313 static int ovl_copy_up_locked(struct dentry
*workdir
, struct dentry
*upperdir
,
314 struct dentry
*dentry
, struct path
*lowerpath
,
315 struct kstat
*stat
, const char *link
,
316 struct kstat
*pstat
, bool tmpfile
)
318 struct inode
*wdir
= workdir
->d_inode
;
319 struct inode
*udir
= upperdir
->d_inode
;
320 struct dentry
*newdentry
= NULL
;
321 struct dentry
*upper
= NULL
;
322 struct dentry
*temp
= NULL
;
324 const struct cred
*old_creds
= NULL
;
325 struct cred
*new_creds
= NULL
;
326 struct cattr cattr
= {
327 /* Can't properly set mode on creation because of the umask */
328 .mode
= stat
->mode
& S_IFMT
,
333 err
= security_inode_copy_up(dentry
, &new_creds
);
338 old_creds
= override_creds(new_creds
);
341 temp
= ovl_do_tmpfile(upperdir
, stat
->mode
);
343 temp
= ovl_lookup_temp(workdir
);
350 if (!err
&& !tmpfile
)
351 err
= ovl_create_real(wdir
, temp
, &cattr
, NULL
, true);
354 revert_creds(old_creds
);
361 if (S_ISREG(stat
->mode
)) {
362 struct path upperpath
;
364 ovl_path_upper(dentry
, &upperpath
);
365 BUG_ON(upperpath
.dentry
!= NULL
);
366 upperpath
.dentry
= temp
;
370 err
= ovl_copy_up_data(lowerpath
, &upperpath
,
372 inode_lock_nested(udir
, I_MUTEX_PARENT
);
374 err
= ovl_copy_up_data(lowerpath
, &upperpath
,
382 err
= ovl_copy_xattr(lowerpath
->dentry
, temp
);
386 inode_lock(temp
->d_inode
);
387 err
= ovl_set_attr(temp
, stat
);
388 inode_unlock(temp
->d_inode
);
393 * Store identifier of lower inode in upper inode xattr to
394 * allow lookup of the copy up origin inode.
396 * Don't set origin when we are breaking the association with a lower
399 if (S_ISDIR(stat
->mode
) || stat
->nlink
== 1) {
400 err
= ovl_set_origin(dentry
, lowerpath
->dentry
, temp
);
405 upper
= lookup_one_len(dentry
->d_name
.name
, upperdir
,
408 err
= PTR_ERR(upper
);
414 err
= ovl_do_link(temp
, udir
, upper
, true);
416 err
= ovl_do_rename(wdir
, temp
, udir
, upper
, 0);
420 newdentry
= dget(tmpfile
? upper
: temp
);
421 ovl_dentry_update(dentry
, newdentry
);
422 ovl_inode_update(d_inode(dentry
), d_inode(newdentry
));
424 /* Restore timestamps on parent (best effort) */
425 ovl_set_timestamps(upperdir
, pstat
);
433 ovl_cleanup(wdir
, temp
);
438 * Copy up a single dentry
440 * All renames start with copy up of source if necessary. The actual
441 * rename will only proceed once the copy up was successful. Copy up uses
442 * upper parent i_mutex for exclusion. Since rename can change d_parent it
443 * is possible that the copy up will lock the old parent. At that point
444 * the file will have already been copied up anyway.
446 static int ovl_copy_up_one(struct dentry
*parent
, struct dentry
*dentry
,
447 struct path
*lowerpath
, struct kstat
*stat
)
449 DEFINE_DELAYED_CALL(done
);
450 struct dentry
*workdir
= ovl_workdir(dentry
);
453 struct path parentpath
;
454 struct dentry
*lowerdentry
= lowerpath
->dentry
;
455 struct dentry
*upperdir
;
456 const char *link
= NULL
;
457 struct ovl_fs
*ofs
= dentry
->d_sb
->s_fs_info
;
459 if (WARN_ON(!workdir
))
462 ovl_do_check_copy_up(lowerdentry
);
464 ovl_path_upper(parent
, &parentpath
);
465 upperdir
= parentpath
.dentry
;
467 /* Mark parent "impure" because it may now contain non-pure upper */
468 err
= ovl_set_impure(parent
, upperdir
);
472 err
= vfs_getattr(&parentpath
, &pstat
,
473 STATX_ATIME
| STATX_MTIME
, AT_STATX_SYNC_AS_STAT
);
477 if (S_ISLNK(stat
->mode
)) {
478 link
= vfs_get_link(lowerdentry
, &done
);
480 return PTR_ERR(link
);
483 /* Should we copyup with O_TMPFILE or with workdir? */
484 if (S_ISREG(stat
->mode
) && ofs
->tmpfile
) {
485 err
= ovl_copy_up_start(dentry
);
486 /* err < 0: interrupted, err > 0: raced with another copy-up */
488 pr_debug("ovl_copy_up_start(%pd2) = %i\n", dentry
, err
);
494 inode_lock_nested(upperdir
->d_inode
, I_MUTEX_PARENT
);
495 err
= ovl_copy_up_locked(workdir
, upperdir
, dentry
, lowerpath
,
496 stat
, link
, &pstat
, true);
497 inode_unlock(upperdir
->d_inode
);
498 ovl_copy_up_end(dentry
);
503 if (lock_rename(workdir
, upperdir
) != NULL
) {
504 pr_err("overlayfs: failed to lock workdir+upperdir\n");
507 if (ovl_dentry_upper(dentry
)) {
508 /* Raced with another copy-up? Nothing to do, then... */
513 err
= ovl_copy_up_locked(workdir
, upperdir
, dentry
, lowerpath
,
514 stat
, link
, &pstat
, false);
516 unlock_rename(workdir
, upperdir
);
518 do_delayed_call(&done
);
523 int ovl_copy_up_flags(struct dentry
*dentry
, int flags
)
526 const struct cred
*old_cred
= ovl_override_creds(dentry
->d_sb
);
530 struct dentry
*parent
;
531 struct path lowerpath
;
533 enum ovl_path_type type
= ovl_path_type(dentry
);
535 if (OVL_TYPE_UPPER(type
))
539 /* find the topmost dentry not yet copied up */
541 parent
= dget_parent(next
);
543 type
= ovl_path_type(parent
);
544 if (OVL_TYPE_UPPER(type
))
551 ovl_path_lower(next
, &lowerpath
);
552 err
= vfs_getattr(&lowerpath
, &stat
,
553 STATX_BASIC_STATS
, AT_STATX_SYNC_AS_STAT
);
554 /* maybe truncate regular file. this has no effect on dirs */
558 err
= ovl_copy_up_one(parent
, next
, &lowerpath
, &stat
);
563 revert_creds(old_cred
);
568 int ovl_copy_up(struct dentry
*dentry
)
570 return ovl_copy_up_flags(dentry
, 0);