]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/overlayfs/copy_up.c
Merge remote-tracking branch 'regulator/fix/max77802' into regulator-linus
[mirror_ubuntu-artful-kernel.git] / fs / overlayfs / copy_up.c
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
10 #include <linux/module.h>
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/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"
26
27 #define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
28
29 static bool __read_mostly ovl_check_copy_up;
30 module_param_named(check_copy_up, ovl_check_copy_up, bool,
31 S_IWUSR | S_IRUGO);
32 MODULE_PARM_DESC(ovl_check_copy_up,
33 "Warn on copy-up when causing process also has a R/O fd open");
34
35 static int ovl_check_fd(const void *data, struct file *f, unsigned int fd)
36 {
37 const struct dentry *dentry = data;
38
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);
42 return 0;
43 }
44
45 /*
46 * Check the fds open by this process and warn if something like the following
47 * scenario is about to occur:
48 *
49 * fd1 = open("foo", O_RDONLY);
50 * fd2 = open("foo", O_RDWR);
51 */
52 static void ovl_do_check_copy_up(struct dentry *dentry)
53 {
54 if (ovl_check_copy_up)
55 iterate_fd(current->files, 0, ovl_check_fd, dentry);
56 }
57
58 int ovl_copy_xattr(struct dentry *old, struct dentry *new)
59 {
60 ssize_t list_size, size, value_size = 0;
61 char *buf, *name, *value = NULL;
62 int uninitialized_var(error);
63 size_t slen;
64
65 if (!(old->d_inode->i_opflags & IOP_XATTR) ||
66 !(new->d_inode->i_opflags & IOP_XATTR))
67 return 0;
68
69 list_size = vfs_listxattr(old, NULL, 0);
70 if (list_size <= 0) {
71 if (list_size == -EOPNOTSUPP)
72 return 0;
73 return list_size;
74 }
75
76 buf = kzalloc(list_size, GFP_KERNEL);
77 if (!buf)
78 return -ENOMEM;
79
80 list_size = vfs_listxattr(old, buf, list_size);
81 if (list_size <= 0) {
82 error = list_size;
83 goto out;
84 }
85
86 for (name = buf; list_size; name += slen) {
87 slen = strnlen(name, list_size) + 1;
88
89 /* underlying fs providing us with an broken xattr list? */
90 if (WARN_ON(slen > list_size)) {
91 error = -EIO;
92 break;
93 }
94 list_size -= slen;
95
96 if (ovl_is_private_xattr(name))
97 continue;
98 retry:
99 size = vfs_getxattr(old, name, value, value_size);
100 if (size == -ERANGE)
101 size = vfs_getxattr(old, name, NULL, 0);
102
103 if (size < 0) {
104 error = size;
105 break;
106 }
107
108 if (size > value_size) {
109 void *new;
110
111 new = krealloc(value, size, GFP_KERNEL);
112 if (!new) {
113 error = -ENOMEM;
114 break;
115 }
116 value = new;
117 value_size = size;
118 goto retry;
119 }
120
121 error = security_inode_copy_up_xattr(name);
122 if (error < 0 && error != -EOPNOTSUPP)
123 break;
124 if (error == 1) {
125 error = 0;
126 continue; /* Discard */
127 }
128 error = vfs_setxattr(new, name, value, size, 0);
129 if (error)
130 break;
131 }
132 kfree(value);
133 out:
134 kfree(buf);
135 return error;
136 }
137
138 static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
139 {
140 struct file *old_file;
141 struct file *new_file;
142 loff_t old_pos = 0;
143 loff_t new_pos = 0;
144 int error = 0;
145
146 if (len == 0)
147 return 0;
148
149 old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
150 if (IS_ERR(old_file))
151 return PTR_ERR(old_file);
152
153 new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
154 if (IS_ERR(new_file)) {
155 error = PTR_ERR(new_file);
156 goto out_fput;
157 }
158
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);
161 if (!error)
162 goto out;
163 /* Couldn't clone, so now we try to copy the data */
164 error = 0;
165
166 /* FIXME: copy up sparse files efficiently */
167 while (len) {
168 size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
169 long bytes;
170
171 if (len < this_len)
172 this_len = len;
173
174 if (signal_pending_state(TASK_KILLABLE, current)) {
175 error = -EINTR;
176 break;
177 }
178
179 bytes = do_splice_direct(old_file, &old_pos,
180 new_file, &new_pos,
181 this_len, SPLICE_F_MOVE);
182 if (bytes <= 0) {
183 error = bytes;
184 break;
185 }
186 WARN_ON(old_pos != new_pos);
187
188 len -= bytes;
189 }
190 out:
191 if (!error)
192 error = vfs_fsync(new_file, 0);
193 fput(new_file);
194 out_fput:
195 fput(old_file);
196 return error;
197 }
198
199 static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
200 {
201 struct iattr attr = {
202 .ia_valid =
203 ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
204 .ia_atime = stat->atime,
205 .ia_mtime = stat->mtime,
206 };
207
208 return notify_change(upperdentry, &attr, NULL);
209 }
210
211 int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
212 {
213 int err = 0;
214
215 if (!S_ISLNK(stat->mode)) {
216 struct iattr attr = {
217 .ia_valid = ATTR_MODE,
218 .ia_mode = stat->mode,
219 };
220 err = notify_change(upperdentry, &attr, NULL);
221 }
222 if (!err) {
223 struct iattr attr = {
224 .ia_valid = ATTR_UID | ATTR_GID,
225 .ia_uid = stat->uid,
226 .ia_gid = stat->gid,
227 };
228 err = notify_change(upperdentry, &attr, NULL);
229 }
230 if (!err)
231 ovl_set_timestamps(upperdentry, stat);
232
233 return err;
234 }
235
236 static struct ovl_fh *ovl_encode_fh(struct dentry *lower, uuid_be *uuid)
237 {
238 struct ovl_fh *fh;
239 int fh_type, fh_len, dwords;
240 void *buf;
241 int buflen = MAX_HANDLE_SZ;
242
243 buf = kmalloc(buflen, GFP_TEMPORARY);
244 if (!buf)
245 return ERR_PTR(-ENOMEM);
246
247 /*
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.
251 */
252 dwords = buflen >> 2;
253 fh_type = exportfs_encode_fh(lower, buf, &dwords, 0);
254 buflen = (dwords << 2);
255
256 fh = ERR_PTR(-EIO);
257 if (WARN_ON(fh_type < 0) ||
258 WARN_ON(buflen > MAX_HANDLE_SZ) ||
259 WARN_ON(fh_type == FILEID_INVALID))
260 goto out;
261
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);
265 if (!fh) {
266 fh = ERR_PTR(-ENOMEM);
267 goto out;
268 }
269
270 fh->version = OVL_FH_VERSION;
271 fh->magic = OVL_FH_MAGIC;
272 fh->type = fh_type;
273 fh->flags = OVL_FH_FLAG_CPU_ENDIAN;
274 fh->len = fh_len;
275 fh->uuid = *uuid;
276 memcpy(fh->fid, buf, buflen);
277
278 out:
279 kfree(buf);
280 return fh;
281 }
282
283 static int ovl_set_origin(struct dentry *dentry, struct dentry *lower,
284 struct dentry *upper)
285 {
286 struct super_block *sb = lower->d_sb;
287 uuid_be *uuid = (uuid_be *) &sb->s_uuid;
288 const struct ovl_fh *fh = NULL;
289 int err;
290
291 /*
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.
295 */
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);
299 if (IS_ERR(fh))
300 return PTR_ERR(fh);
301 }
302
303 /*
304 * Do not fail when upper doesn't support xattrs.
305 */
306 err = ovl_check_setxattr(dentry, upper, OVL_XATTR_ORIGIN, fh,
307 fh ? fh->len : 0, 0);
308 kfree(fh);
309
310 return err;
311 }
312
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)
317 {
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;
323 int err;
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,
329 .rdev = stat->rdev,
330 .link = link
331 };
332
333 err = security_inode_copy_up(dentry, &new_creds);
334 if (err < 0)
335 goto out;
336
337 if (new_creds)
338 old_creds = override_creds(new_creds);
339
340 if (tmpfile)
341 temp = ovl_do_tmpfile(upperdir, stat->mode);
342 else
343 temp = ovl_lookup_temp(workdir);
344 err = 0;
345 if (IS_ERR(temp)) {
346 err = PTR_ERR(temp);
347 temp = NULL;
348 }
349
350 if (!err && !tmpfile)
351 err = ovl_create_real(wdir, temp, &cattr, NULL, true);
352
353 if (new_creds) {
354 revert_creds(old_creds);
355 put_cred(new_creds);
356 }
357
358 if (err)
359 goto out;
360
361 if (S_ISREG(stat->mode)) {
362 struct path upperpath;
363
364 ovl_path_upper(dentry, &upperpath);
365 BUG_ON(upperpath.dentry != NULL);
366 upperpath.dentry = temp;
367
368 if (tmpfile) {
369 inode_unlock(udir);
370 err = ovl_copy_up_data(lowerpath, &upperpath,
371 stat->size);
372 inode_lock_nested(udir, I_MUTEX_PARENT);
373 } else {
374 err = ovl_copy_up_data(lowerpath, &upperpath,
375 stat->size);
376 }
377
378 if (err)
379 goto out_cleanup;
380 }
381
382 err = ovl_copy_xattr(lowerpath->dentry, temp);
383 if (err)
384 goto out_cleanup;
385
386 inode_lock(temp->d_inode);
387 err = ovl_set_attr(temp, stat);
388 inode_unlock(temp->d_inode);
389 if (err)
390 goto out_cleanup;
391
392 /*
393 * Store identifier of lower inode in upper inode xattr to
394 * allow lookup of the copy up origin inode.
395 *
396 * Don't set origin when we are breaking the association with a lower
397 * hard link.
398 */
399 if (S_ISDIR(stat->mode) || stat->nlink == 1) {
400 err = ovl_set_origin(dentry, lowerpath->dentry, temp);
401 if (err)
402 goto out_cleanup;
403 }
404
405 upper = lookup_one_len(dentry->d_name.name, upperdir,
406 dentry->d_name.len);
407 if (IS_ERR(upper)) {
408 err = PTR_ERR(upper);
409 upper = NULL;
410 goto out_cleanup;
411 }
412
413 if (tmpfile)
414 err = ovl_do_link(temp, udir, upper, true);
415 else
416 err = ovl_do_rename(wdir, temp, udir, upper, 0);
417 if (err)
418 goto out_cleanup;
419
420 newdentry = dget(tmpfile ? upper : temp);
421 ovl_dentry_update(dentry, newdentry);
422 ovl_inode_update(d_inode(dentry), d_inode(newdentry));
423
424 /* Restore timestamps on parent (best effort) */
425 ovl_set_timestamps(upperdir, pstat);
426 out:
427 dput(temp);
428 dput(upper);
429 return err;
430
431 out_cleanup:
432 if (!tmpfile)
433 ovl_cleanup(wdir, temp);
434 goto out;
435 }
436
437 /*
438 * Copy up a single dentry
439 *
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.
445 */
446 static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
447 struct path *lowerpath, struct kstat *stat)
448 {
449 DEFINE_DELAYED_CALL(done);
450 struct dentry *workdir = ovl_workdir(dentry);
451 int err;
452 struct kstat pstat;
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;
458
459 if (WARN_ON(!workdir))
460 return -EROFS;
461
462 ovl_do_check_copy_up(lowerdentry);
463
464 ovl_path_upper(parent, &parentpath);
465 upperdir = parentpath.dentry;
466
467 /* Mark parent "impure" because it may now contain non-pure upper */
468 err = ovl_set_impure(parent, upperdir);
469 if (err)
470 return err;
471
472 err = vfs_getattr(&parentpath, &pstat,
473 STATX_ATIME | STATX_MTIME, AT_STATX_SYNC_AS_STAT);
474 if (err)
475 return err;
476
477 if (S_ISLNK(stat->mode)) {
478 link = vfs_get_link(lowerdentry, &done);
479 if (IS_ERR(link))
480 return PTR_ERR(link);
481 }
482
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 */
487 if (unlikely(err)) {
488 pr_debug("ovl_copy_up_start(%pd2) = %i\n", dentry, err);
489 if (err > 0)
490 err = 0;
491 goto out_done;
492 }
493
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);
499 goto out_done;
500 }
501
502 err = -EIO;
503 if (lock_rename(workdir, upperdir) != NULL) {
504 pr_err("overlayfs: failed to lock workdir+upperdir\n");
505 goto out_unlock;
506 }
507 if (ovl_dentry_upper(dentry)) {
508 /* Raced with another copy-up? Nothing to do, then... */
509 err = 0;
510 goto out_unlock;
511 }
512
513 err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
514 stat, link, &pstat, false);
515 out_unlock:
516 unlock_rename(workdir, upperdir);
517 out_done:
518 do_delayed_call(&done);
519
520 return err;
521 }
522
523 int ovl_copy_up_flags(struct dentry *dentry, int flags)
524 {
525 int err = 0;
526 const struct cred *old_cred = ovl_override_creds(dentry->d_sb);
527
528 while (!err) {
529 struct dentry *next;
530 struct dentry *parent;
531 struct path lowerpath;
532 struct kstat stat;
533 enum ovl_path_type type = ovl_path_type(dentry);
534
535 if (OVL_TYPE_UPPER(type))
536 break;
537
538 next = dget(dentry);
539 /* find the topmost dentry not yet copied up */
540 for (;;) {
541 parent = dget_parent(next);
542
543 type = ovl_path_type(parent);
544 if (OVL_TYPE_UPPER(type))
545 break;
546
547 dput(next);
548 next = parent;
549 }
550
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 */
555 if (flags & O_TRUNC)
556 stat.size = 0;
557 if (!err)
558 err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
559
560 dput(parent);
561 dput(next);
562 }
563 revert_creds(old_cred);
564
565 return err;
566 }
567
568 int ovl_copy_up(struct dentry *dentry)
569 {
570 return ovl_copy_up_flags(dentry, 0);
571 }