Merge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target...
[linux-2.6-block.git] / fs / overlayfs / copy_up.c
CommitLineData
e9be9d5e
MS
1/*
2 *
3 * Copyright (C) 2011 Novell Inc.
4 *
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.
8 */
9
fb5bb2c3 10#include <linux/module.h>
e9be9d5e
MS
11#include <linux/fs.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>
fb5bb2c3
DH
20#include <linux/fdtable.h>
21#include <linux/ratelimit.h>
e9be9d5e
MS
22#include "overlayfs.h"
23
24#define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
25
fb5bb2c3
DH
26static bool __read_mostly ovl_check_copy_up;
27module_param_named(check_copy_up, ovl_check_copy_up, bool,
28 S_IWUSR | S_IRUGO);
29MODULE_PARM_DESC(ovl_check_copy_up,
30 "Warn on copy-up when causing process also has a R/O fd open");
31
32static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
33{
34 const struct dentry *dentry = data;
35
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);
39 return 0;
40}
41
42/*
43 * Check the fds open by this process and warn if something like the following
44 * scenario is about to occur:
45 *
46 * fd1 = open("foo", O_RDONLY);
47 * fd2 = open("foo", O_RDWR);
48 */
49static void ovl_do_check_copy_up(struct dentry *dentry)
50{
51 if (ovl_check_copy_up)
52 iterate_fd(current->files, 0, ovl_check_fd, dentry);
53}
54
e9be9d5e
MS
55int ovl_copy_xattr(struct dentry *old, struct dentry *new)
56{
e4ad29fa
VC
57 ssize_t list_size, size, value_size = 0;
58 char *buf, *name, *value = NULL;
59 int uninitialized_var(error);
e9be9d5e
MS
60
61 if (!old->d_inode->i_op->getxattr ||
62 !new->d_inode->i_op->getxattr)
63 return 0;
64
65 list_size = vfs_listxattr(old, NULL, 0);
66 if (list_size <= 0) {
67 if (list_size == -EOPNOTSUPP)
68 return 0;
69 return list_size;
70 }
71
72 buf = kzalloc(list_size, GFP_KERNEL);
73 if (!buf)
74 return -ENOMEM;
75
e9be9d5e
MS
76 list_size = vfs_listxattr(old, buf, list_size);
77 if (list_size <= 0) {
78 error = list_size;
e4ad29fa 79 goto out;
e9be9d5e
MS
80 }
81
82 for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
e4ad29fa
VC
83retry:
84 size = vfs_getxattr(old, name, value, value_size);
85 if (size == -ERANGE)
86 size = vfs_getxattr(old, name, NULL, 0);
87
97daf8b9 88 if (size < 0) {
e9be9d5e 89 error = size;
e4ad29fa 90 break;
e9be9d5e 91 }
e4ad29fa
VC
92
93 if (size > value_size) {
94 void *new;
95
96 new = krealloc(value, size, GFP_KERNEL);
97 if (!new) {
98 error = -ENOMEM;
99 break;
100 }
101 value = new;
102 value_size = size;
103 goto retry;
104 }
105
e9be9d5e
MS
106 error = vfs_setxattr(new, name, value, size, 0);
107 if (error)
e4ad29fa 108 break;
e9be9d5e 109 }
e9be9d5e
MS
110 kfree(value);
111out:
112 kfree(buf);
113 return error;
114}
115
116static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
117{
118 struct file *old_file;
119 struct file *new_file;
120 loff_t old_pos = 0;
121 loff_t new_pos = 0;
122 int error = 0;
123
124 if (len == 0)
125 return 0;
126
0480334f 127 old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
e9be9d5e
MS
128 if (IS_ERR(old_file))
129 return PTR_ERR(old_file);
130
0480334f 131 new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
e9be9d5e
MS
132 if (IS_ERR(new_file)) {
133 error = PTR_ERR(new_file);
134 goto out_fput;
135 }
136
137 /* FIXME: copy up sparse files efficiently */
138 while (len) {
139 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
140 long bytes;
141
142 if (len < this_len)
143 this_len = len;
144
145 if (signal_pending_state(TASK_KILLABLE, current)) {
146 error = -EINTR;
147 break;
148 }
149
150 bytes = do_splice_direct(old_file, &old_pos,
151 new_file, &new_pos,
152 this_len, SPLICE_F_MOVE);
153 if (bytes <= 0) {
154 error = bytes;
155 break;
156 }
157 WARN_ON(old_pos != new_pos);
158
159 len -= bytes;
160 }
161
162 fput(new_file);
163out_fput:
164 fput(old_file);
165 return error;
166}
167
168static char *ovl_read_symlink(struct dentry *realdentry)
169{
170 int res;
171 char *buf;
172 struct inode *inode = realdentry->d_inode;
173 mm_segment_t old_fs;
174
175 res = -EINVAL;
176 if (!inode->i_op->readlink)
177 goto err;
178
179 res = -ENOMEM;
180 buf = (char *) __get_free_page(GFP_KERNEL);
181 if (!buf)
182 goto err;
183
184 old_fs = get_fs();
185 set_fs(get_ds());
186 /* The cast to a user pointer is valid due to the set_fs() */
187 res = inode->i_op->readlink(realdentry,
188 (char __user *)buf, PAGE_SIZE - 1);
189 set_fs(old_fs);
190 if (res < 0) {
191 free_page((unsigned long) buf);
192 goto err;
193 }
194 buf[res] = '\0';
195
196 return buf;
197
198err:
199 return ERR_PTR(res);
200}
201
202static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
203{
204 struct iattr attr = {
205 .ia_valid =
206 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
207 .ia_atime = stat->atime,
208 .ia_mtime = stat->mtime,
209 };
210
211 return notify_change(upperdentry, &attr, NULL);
212}
213
214int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
215{
216 int err = 0;
217
218 if (!S_ISLNK(stat->mode)) {
219 struct iattr attr = {
220 .ia_valid = ATTR_MODE,
221 .ia_mode = stat->mode,
222 };
223 err = notify_change(upperdentry, &attr, NULL);
224 }
225 if (!err) {
226 struct iattr attr = {
227 .ia_valid = ATTR_UID | ATTR_GID,
228 .ia_uid = stat->uid,
229 .ia_gid = stat->gid,
230 };
231 err = notify_change(upperdentry, &attr, NULL);
232 }
233 if (!err)
234 ovl_set_timestamps(upperdentry, stat);
235
236 return err;
e9be9d5e
MS
237}
238
239static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
240 struct dentry *dentry, struct path *lowerpath,
0f7ff2da 241 struct kstat *stat, const char *link)
e9be9d5e
MS
242{
243 struct inode *wdir = workdir->d_inode;
244 struct inode *udir = upperdir->d_inode;
245 struct dentry *newdentry = NULL;
246 struct dentry *upper = NULL;
247 umode_t mode = stat->mode;
248 int err;
249
250 newdentry = ovl_lookup_temp(workdir, dentry);
251 err = PTR_ERR(newdentry);
252 if (IS_ERR(newdentry))
253 goto out;
254
255 upper = lookup_one_len(dentry->d_name.name, upperdir,
256 dentry->d_name.len);
257 err = PTR_ERR(upper);
258 if (IS_ERR(upper))
259 goto out1;
260
261 /* Can't properly set mode on creation because of the umask */
262 stat->mode &= S_IFMT;
263 err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
264 stat->mode = mode;
265 if (err)
266 goto out2;
267
268 if (S_ISREG(stat->mode)) {
269 struct path upperpath;
f134f244 270
e9be9d5e
MS
271 ovl_path_upper(dentry, &upperpath);
272 BUG_ON(upperpath.dentry != NULL);
273 upperpath.dentry = newdentry;
274
275 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
276 if (err)
277 goto out_cleanup;
278 }
279
280 err = ovl_copy_xattr(lowerpath->dentry, newdentry);
281 if (err)
282 goto out_cleanup;
283
5955102c 284 inode_lock(newdentry->d_inode);
e9be9d5e 285 err = ovl_set_attr(newdentry, stat);
5955102c 286 inode_unlock(newdentry->d_inode);
e9be9d5e
MS
287 if (err)
288 goto out_cleanup;
289
290 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
291 if (err)
292 goto out_cleanup;
293
294 ovl_dentry_update(dentry, newdentry);
295 newdentry = NULL;
296
297 /*
298 * Non-directores become opaque when copied up.
299 */
300 if (!S_ISDIR(stat->mode))
301 ovl_dentry_set_opaque(dentry, true);
302out2:
303 dput(upper);
304out1:
305 dput(newdentry);
306out:
307 return err;
308
309out_cleanup:
310 ovl_cleanup(wdir, newdentry);
ab79efab 311 goto out2;
e9be9d5e
MS
312}
313
314/*
315 * Copy up a single dentry
316 *
317 * Directory renames only allowed on "pure upper" (already created on
318 * upper filesystem, never copied up). Directories which are on lower or
319 * are merged may not be renamed. For these -EXDEV is returned and
320 * userspace has to deal with it. This means, when copying up a
321 * directory we can rely on it and ancestors being stable.
322 *
323 * Non-directory renames start with copy up of source if necessary. The
324 * actual rename will only proceed once the copy up was successful. Copy
325 * up uses upper parent i_mutex for exclusion. Since rename can change
326 * d_parent it is possible that the copy up will lock the old parent. At
327 * that point the file will have already been copied up anyway.
328 */
329int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
0f7ff2da 330 struct path *lowerpath, struct kstat *stat)
e9be9d5e
MS
331{
332 struct dentry *workdir = ovl_workdir(dentry);
333 int err;
334 struct kstat pstat;
335 struct path parentpath;
336 struct dentry *upperdir;
337 struct dentry *upperdentry;
338 const struct cred *old_cred;
e9be9d5e
MS
339 char *link = NULL;
340
cc6f67bc
MS
341 if (WARN_ON(!workdir))
342 return -EROFS;
343
fb5bb2c3
DH
344 ovl_do_check_copy_up(lowerpath->dentry);
345
e9be9d5e
MS
346 ovl_path_upper(parent, &parentpath);
347 upperdir = parentpath.dentry;
348
349 err = vfs_getattr(&parentpath, &pstat);
350 if (err)
351 return err;
352
353 if (S_ISLNK(stat->mode)) {
354 link = ovl_read_symlink(lowerpath->dentry);
355 if (IS_ERR(link))
356 return PTR_ERR(link);
357 }
358
3fe6e52f 359 old_cred = ovl_override_creds(dentry->d_sb);
e9be9d5e
MS
360
361 err = -EIO;
362 if (lock_rename(workdir, upperdir) != NULL) {
363 pr_err("overlayfs: failed to lock workdir+upperdir\n");
364 goto out_unlock;
365 }
366 upperdentry = ovl_dentry_upper(dentry);
367 if (upperdentry) {
0f7ff2da 368 /* Raced with another copy-up? Nothing to do, then... */
e9be9d5e 369 err = 0;
0f7ff2da 370 goto out_unlock;
e9be9d5e
MS
371 }
372
373 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
0f7ff2da 374 stat, link);
e9be9d5e
MS
375 if (!err) {
376 /* Restore timestamps on parent (best effort) */
377 ovl_set_timestamps(upperdir, &pstat);
378 }
379out_unlock:
380 unlock_rename(workdir, upperdir);
e9be9d5e 381 revert_creds(old_cred);
e9be9d5e 382
e9be9d5e
MS
383 if (link)
384 free_page((unsigned long) link);
385
386 return err;
387}
388
389int ovl_copy_up(struct dentry *dentry)
390{
391 int err;
392
393 err = 0;
394 while (!err) {
395 struct dentry *next;
396 struct dentry *parent;
397 struct path lowerpath;
398 struct kstat stat;
399 enum ovl_path_type type = ovl_path_type(dentry);
400
1afaba1e 401 if (OVL_TYPE_UPPER(type))
e9be9d5e
MS
402 break;
403
404 next = dget(dentry);
405 /* find the topmost dentry not yet copied up */
406 for (;;) {
407 parent = dget_parent(next);
408
409 type = ovl_path_type(parent);
1afaba1e 410 if (OVL_TYPE_UPPER(type))
e9be9d5e
MS
411 break;
412
413 dput(next);
414 next = parent;
415 }
416
417 ovl_path_lower(next, &lowerpath);
418 err = vfs_getattr(&lowerpath, &stat);
419 if (!err)
0f7ff2da 420 err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
e9be9d5e
MS
421
422 dput(parent);
423 dput(next);
424 }
425
426 return err;
427}