ovl: create directories inside merged parent opaque
[linux-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);
8b326c61 60 size_t slen;
e9be9d5e 61
5d6c3191
AG
62 if (!(old->d_inode->i_opflags & IOP_XATTR) ||
63 !(new->d_inode->i_opflags & IOP_XATTR))
e9be9d5e
MS
64 return 0;
65
66 list_size = vfs_listxattr(old, NULL, 0);
67 if (list_size <= 0) {
68 if (list_size == -EOPNOTSUPP)
69 return 0;
70 return list_size;
71 }
72
73 buf = kzalloc(list_size, GFP_KERNEL);
74 if (!buf)
75 return -ENOMEM;
76
e9be9d5e
MS
77 list_size = vfs_listxattr(old, buf, list_size);
78 if (list_size <= 0) {
79 error = list_size;
e4ad29fa 80 goto out;
e9be9d5e
MS
81 }
82
8b326c61
MS
83 for (name = buf; list_size; name += slen) {
84 slen = strnlen(name, list_size) + 1;
85
86 /* underlying fs providing us with an broken xattr list? */
87 if (WARN_ON(slen > list_size)) {
88 error = -EIO;
89 break;
90 }
91 list_size -= slen;
92
0956254a
MS
93 if (ovl_is_private_xattr(name))
94 continue;
e4ad29fa
VC
95retry:
96 size = vfs_getxattr(old, name, value, value_size);
97 if (size == -ERANGE)
98 size = vfs_getxattr(old, name, NULL, 0);
99
97daf8b9 100 if (size < 0) {
e9be9d5e 101 error = size;
e4ad29fa 102 break;
e9be9d5e 103 }
e4ad29fa
VC
104
105 if (size > value_size) {
106 void *new;
107
108 new = krealloc(value, size, GFP_KERNEL);
109 if (!new) {
110 error = -ENOMEM;
111 break;
112 }
113 value = new;
114 value_size = size;
115 goto retry;
116 }
117
121ab822
VG
118 error = security_inode_copy_up_xattr(name);
119 if (error < 0 && error != -EOPNOTSUPP)
120 break;
121 if (error == 1) {
122 error = 0;
123 continue; /* Discard */
124 }
e9be9d5e
MS
125 error = vfs_setxattr(new, name, value, size, 0);
126 if (error)
e4ad29fa 127 break;
e9be9d5e 128 }
e9be9d5e
MS
129 kfree(value);
130out:
131 kfree(buf);
132 return error;
133}
134
135static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
136{
137 struct file *old_file;
138 struct file *new_file;
139 loff_t old_pos = 0;
140 loff_t new_pos = 0;
141 int error = 0;
142
143 if (len == 0)
144 return 0;
145
0480334f 146 old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
e9be9d5e
MS
147 if (IS_ERR(old_file))
148 return PTR_ERR(old_file);
149
0480334f 150 new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
e9be9d5e
MS
151 if (IS_ERR(new_file)) {
152 error = PTR_ERR(new_file);
153 goto out_fput;
154 }
155
2ea98466
AG
156 /* Try to use clone_file_range to clone up within the same fs */
157 error = vfs_clone_file_range(old_file, 0, new_file, 0, len);
158 if (!error)
159 goto out;
160 /* Couldn't clone, so now we try to copy the data */
161 error = 0;
162
e9be9d5e
MS
163 /* FIXME: copy up sparse files efficiently */
164 while (len) {
165 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
166 long bytes;
167
168 if (len < this_len)
169 this_len = len;
170
171 if (signal_pending_state(TASK_KILLABLE, current)) {
172 error = -EINTR;
173 break;
174 }
175
176 bytes = do_splice_direct(old_file, &old_pos,
177 new_file, &new_pos,
178 this_len, SPLICE_F_MOVE);
179 if (bytes <= 0) {
180 error = bytes;
181 break;
182 }
183 WARN_ON(old_pos != new_pos);
184
185 len -= bytes;
186 }
2ea98466 187out:
641089c1
MS
188 if (!error)
189 error = vfs_fsync(new_file, 0);
e9be9d5e
MS
190 fput(new_file);
191out_fput:
192 fput(old_file);
193 return error;
194}
195
e9be9d5e
MS
196static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
197{
198 struct iattr attr = {
199 .ia_valid =
200 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
201 .ia_atime = stat->atime,
202 .ia_mtime = stat->mtime,
203 };
204
205 return notify_change(upperdentry, &attr, NULL);
206}
207
208int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
209{
210 int err = 0;
211
212 if (!S_ISLNK(stat->mode)) {
213 struct iattr attr = {
214 .ia_valid = ATTR_MODE,
215 .ia_mode = stat->mode,
216 };
217 err = notify_change(upperdentry, &attr, NULL);
218 }
219 if (!err) {
220 struct iattr attr = {
221 .ia_valid = ATTR_UID | ATTR_GID,
222 .ia_uid = stat->uid,
223 .ia_gid = stat->gid,
224 };
225 err = notify_change(upperdentry, &attr, NULL);
226 }
227 if (!err)
228 ovl_set_timestamps(upperdentry, stat);
229
230 return err;
e9be9d5e
MS
231}
232
233static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
234 struct dentry *dentry, struct path *lowerpath,
0f7ff2da 235 struct kstat *stat, const char *link)
e9be9d5e
MS
236{
237 struct inode *wdir = workdir->d_inode;
238 struct inode *udir = upperdir->d_inode;
239 struct dentry *newdentry = NULL;
240 struct dentry *upper = NULL;
241 umode_t mode = stat->mode;
242 int err;
d8ad8b49
VG
243 const struct cred *old_creds = NULL;
244 struct cred *new_creds = NULL;
e9be9d5e
MS
245
246 newdentry = ovl_lookup_temp(workdir, dentry);
247 err = PTR_ERR(newdentry);
248 if (IS_ERR(newdentry))
249 goto out;
250
251 upper = lookup_one_len(dentry->d_name.name, upperdir,
252 dentry->d_name.len);
253 err = PTR_ERR(upper);
254 if (IS_ERR(upper))
255 goto out1;
256
d8ad8b49
VG
257 err = security_inode_copy_up(dentry, &new_creds);
258 if (err < 0)
259 goto out2;
260
261 if (new_creds)
262 old_creds = override_creds(new_creds);
263
e9be9d5e
MS
264 /* Can't properly set mode on creation because of the umask */
265 stat->mode &= S_IFMT;
266 err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
267 stat->mode = mode;
d8ad8b49
VG
268
269 if (new_creds) {
270 revert_creds(old_creds);
271 put_cred(new_creds);
272 }
273
e9be9d5e
MS
274 if (err)
275 goto out2;
276
277 if (S_ISREG(stat->mode)) {
278 struct path upperpath;
f134f244 279
e9be9d5e
MS
280 ovl_path_upper(dentry, &upperpath);
281 BUG_ON(upperpath.dentry != NULL);
282 upperpath.dentry = newdentry;
283
284 err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
285 if (err)
286 goto out_cleanup;
287 }
288
289 err = ovl_copy_xattr(lowerpath->dentry, newdentry);
290 if (err)
291 goto out_cleanup;
292
5955102c 293 inode_lock(newdentry->d_inode);
e9be9d5e 294 err = ovl_set_attr(newdentry, stat);
5955102c 295 inode_unlock(newdentry->d_inode);
e9be9d5e
MS
296 if (err)
297 goto out_cleanup;
298
299 err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
300 if (err)
301 goto out_cleanup;
302
303 ovl_dentry_update(dentry, newdentry);
39b681f8 304 ovl_inode_update(d_inode(dentry), d_inode(newdentry));
e9be9d5e 305 newdentry = NULL;
e9be9d5e
MS
306out2:
307 dput(upper);
308out1:
309 dput(newdentry);
310out:
311 return err;
312
313out_cleanup:
314 ovl_cleanup(wdir, newdentry);
ab79efab 315 goto out2;
e9be9d5e
MS
316}
317
318/*
319 * Copy up a single dentry
320 *
a6c60655
MS
321 * All renames start with copy up of source if necessary. The actual
322 * rename will only proceed once the copy up was successful. Copy up uses
323 * upper parent i_mutex for exclusion. Since rename can change d_parent it
324 * is possible that the copy up will lock the old parent. At that point
325 * the file will have already been copied up anyway.
e9be9d5e
MS
326 */
327int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
0f7ff2da 328 struct path *lowerpath, struct kstat *stat)
e9be9d5e 329{
7764235b 330 DEFINE_DELAYED_CALL(done);
e9be9d5e
MS
331 struct dentry *workdir = ovl_workdir(dentry);
332 int err;
333 struct kstat pstat;
334 struct path parentpath;
7764235b 335 struct dentry *lowerdentry = lowerpath->dentry;
e9be9d5e 336 struct dentry *upperdir;
7764235b 337 const char *link = NULL;
e9be9d5e 338
cc6f67bc
MS
339 if (WARN_ON(!workdir))
340 return -EROFS;
341
7764235b 342 ovl_do_check_copy_up(lowerdentry);
fb5bb2c3 343
e9be9d5e
MS
344 ovl_path_upper(parent, &parentpath);
345 upperdir = parentpath.dentry;
346
347 err = vfs_getattr(&parentpath, &pstat);
348 if (err)
349 return err;
350
351 if (S_ISLNK(stat->mode)) {
7764235b 352 link = vfs_get_link(lowerdentry, &done);
e9be9d5e
MS
353 if (IS_ERR(link))
354 return PTR_ERR(link);
355 }
356
e9be9d5e
MS
357 err = -EIO;
358 if (lock_rename(workdir, upperdir) != NULL) {
359 pr_err("overlayfs: failed to lock workdir+upperdir\n");
360 goto out_unlock;
361 }
a6c60655 362 if (ovl_dentry_upper(dentry)) {
0f7ff2da 363 /* Raced with another copy-up? Nothing to do, then... */
e9be9d5e 364 err = 0;
0f7ff2da 365 goto out_unlock;
e9be9d5e
MS
366 }
367
368 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
0f7ff2da 369 stat, link);
e9be9d5e
MS
370 if (!err) {
371 /* Restore timestamps on parent (best effort) */
372 ovl_set_timestamps(upperdir, &pstat);
373 }
374out_unlock:
375 unlock_rename(workdir, upperdir);
7764235b 376 do_delayed_call(&done);
e9be9d5e
MS
377
378 return err;
379}
380
381int ovl_copy_up(struct dentry *dentry)
382{
8eac98b8
VG
383 int err = 0;
384 const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
e9be9d5e 385
e9be9d5e
MS
386 while (!err) {
387 struct dentry *next;
388 struct dentry *parent;
389 struct path lowerpath;
390 struct kstat stat;
391 enum ovl_path_type type = ovl_path_type(dentry);
392
1afaba1e 393 if (OVL_TYPE_UPPER(type))
e9be9d5e
MS
394 break;
395
396 next = dget(dentry);
397 /* find the topmost dentry not yet copied up */
398 for (;;) {
399 parent = dget_parent(next);
400
401 type = ovl_path_type(parent);
1afaba1e 402 if (OVL_TYPE_UPPER(type))
e9be9d5e
MS
403 break;
404
405 dput(next);
406 next = parent;
407 }
408
409 ovl_path_lower(next, &lowerpath);
410 err = vfs_getattr(&lowerpath, &stat);
411 if (!err)
0f7ff2da 412 err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
e9be9d5e
MS
413
414 dput(parent);
415 dput(next);
416 }
8eac98b8 417 revert_creds(old_cred);
e9be9d5e
MS
418
419 return err;
420}