]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/overlayfs/namei.c
powerpc/mm: Ensure cpumask update is ordered
[mirror_ubuntu-artful-kernel.git] / fs / overlayfs / namei.c
1 /*
2 * Copyright (C) 2011 Novell Inc.
3 * Copyright (C) 2016 Red Hat, 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/fs.h>
11 #include <linux/cred.h>
12 #include <linux/namei.h>
13 #include <linux/xattr.h>
14 #include <linux/ratelimit.h>
15 #include <linux/mount.h>
16 #include <linux/exportfs.h>
17 #include "overlayfs.h"
18 #include "ovl_entry.h"
19
20 struct ovl_lookup_data {
21 struct qstr name;
22 bool is_dir;
23 bool opaque;
24 bool stop;
25 bool last;
26 char *redirect;
27 };
28
29 static int ovl_check_redirect(struct dentry *dentry, struct ovl_lookup_data *d,
30 size_t prelen, const char *post)
31 {
32 int res;
33 char *s, *next, *buf = NULL;
34
35 res = vfs_getxattr(dentry, OVL_XATTR_REDIRECT, NULL, 0);
36 if (res < 0) {
37 if (res == -ENODATA || res == -EOPNOTSUPP)
38 return 0;
39 goto fail;
40 }
41 buf = kzalloc(prelen + res + strlen(post) + 1, GFP_TEMPORARY);
42 if (!buf)
43 return -ENOMEM;
44
45 if (res == 0)
46 goto invalid;
47
48 res = vfs_getxattr(dentry, OVL_XATTR_REDIRECT, buf, res);
49 if (res < 0)
50 goto fail;
51 if (res == 0)
52 goto invalid;
53 if (buf[0] == '/') {
54 for (s = buf; *s++ == '/'; s = next) {
55 next = strchrnul(s, '/');
56 if (s == next)
57 goto invalid;
58 }
59 } else {
60 if (strchr(buf, '/') != NULL)
61 goto invalid;
62
63 memmove(buf + prelen, buf, res);
64 memcpy(buf, d->name.name, prelen);
65 }
66
67 strcat(buf, post);
68 kfree(d->redirect);
69 d->redirect = buf;
70 d->name.name = d->redirect;
71 d->name.len = strlen(d->redirect);
72
73 return 0;
74
75 err_free:
76 kfree(buf);
77 return 0;
78 fail:
79 pr_warn_ratelimited("overlayfs: failed to get redirect (%i)\n", res);
80 goto err_free;
81 invalid:
82 pr_warn_ratelimited("overlayfs: invalid redirect (%s)\n", buf);
83 goto err_free;
84 }
85
86 static int ovl_acceptable(void *ctx, struct dentry *dentry)
87 {
88 return 1;
89 }
90
91 static struct ovl_fh *ovl_get_origin_fh(struct dentry *dentry)
92 {
93 int res;
94 struct ovl_fh *fh = NULL;
95
96 res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, NULL, 0);
97 if (res < 0) {
98 if (res == -ENODATA || res == -EOPNOTSUPP)
99 return NULL;
100 goto fail;
101 }
102 /* Zero size value means "copied up but origin unknown" */
103 if (res == 0)
104 return NULL;
105
106 fh = kzalloc(res, GFP_TEMPORARY);
107 if (!fh)
108 return ERR_PTR(-ENOMEM);
109
110 res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, fh, res);
111 if (res < 0)
112 goto fail;
113
114 if (res < sizeof(struct ovl_fh) || res < fh->len)
115 goto invalid;
116
117 if (fh->magic != OVL_FH_MAGIC)
118 goto invalid;
119
120 /* Treat larger version and unknown flags as "origin unknown" */
121 if (fh->version > OVL_FH_VERSION || fh->flags & ~OVL_FH_FLAG_ALL)
122 goto out;
123
124 /* Treat endianness mismatch as "origin unknown" */
125 if (!(fh->flags & OVL_FH_FLAG_ANY_ENDIAN) &&
126 (fh->flags & OVL_FH_FLAG_BIG_ENDIAN) != OVL_FH_FLAG_CPU_ENDIAN)
127 goto out;
128
129 return fh;
130
131 out:
132 kfree(fh);
133 return NULL;
134
135 fail:
136 pr_warn_ratelimited("overlayfs: failed to get origin (%i)\n", res);
137 goto out;
138 invalid:
139 pr_warn_ratelimited("overlayfs: invalid origin (%*phN)\n", res, fh);
140 goto out;
141 }
142
143 static struct dentry *ovl_get_origin(struct dentry *dentry,
144 struct vfsmount *mnt)
145 {
146 struct dentry *origin = NULL;
147 struct ovl_fh *fh = ovl_get_origin_fh(dentry);
148 int bytes;
149
150 if (IS_ERR_OR_NULL(fh))
151 return (struct dentry *)fh;
152
153 /*
154 * Make sure that the stored uuid matches the uuid of the lower
155 * layer where file handle will be decoded.
156 */
157 if (!uuid_equal(&fh->uuid, &mnt->mnt_sb->s_uuid))
158 goto out;
159
160 bytes = (fh->len - offsetof(struct ovl_fh, fid));
161 origin = exportfs_decode_fh(mnt, (struct fid *)fh->fid,
162 bytes >> 2, (int)fh->type,
163 ovl_acceptable, NULL);
164 if (IS_ERR(origin)) {
165 /* Treat stale file handle as "origin unknown" */
166 if (origin == ERR_PTR(-ESTALE))
167 origin = NULL;
168 goto out;
169 }
170
171 if (ovl_dentry_weird(origin) ||
172 ((d_inode(origin)->i_mode ^ d_inode(dentry)->i_mode) & S_IFMT))
173 goto invalid;
174
175 out:
176 kfree(fh);
177 return origin;
178
179 invalid:
180 pr_warn_ratelimited("overlayfs: invalid origin (%pd2)\n", origin);
181 dput(origin);
182 origin = NULL;
183 goto out;
184 }
185
186 static bool ovl_is_opaquedir(struct dentry *dentry)
187 {
188 return ovl_check_dir_xattr(dentry, OVL_XATTR_OPAQUE);
189 }
190
191 static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
192 const char *name, unsigned int namelen,
193 size_t prelen, const char *post,
194 struct dentry **ret)
195 {
196 struct dentry *this;
197 int err;
198
199 this = lookup_one_len_unlocked(name, base, namelen);
200 if (IS_ERR(this)) {
201 err = PTR_ERR(this);
202 this = NULL;
203 if (err == -ENOENT || err == -ENAMETOOLONG)
204 goto out;
205 goto out_err;
206 }
207 if (!this->d_inode)
208 goto put_and_out;
209
210 if (ovl_dentry_weird(this)) {
211 /* Don't support traversing automounts and other weirdness */
212 err = -EREMOTE;
213 goto out_err;
214 }
215 if (ovl_is_whiteout(this)) {
216 d->stop = d->opaque = true;
217 goto put_and_out;
218 }
219 if (!d_can_lookup(this)) {
220 d->stop = true;
221 if (d->is_dir)
222 goto put_and_out;
223 goto out;
224 }
225 d->is_dir = true;
226 if (!d->last && ovl_is_opaquedir(this)) {
227 d->stop = d->opaque = true;
228 goto out;
229 }
230 err = ovl_check_redirect(this, d, prelen, post);
231 if (err)
232 goto out_err;
233 out:
234 *ret = this;
235 return 0;
236
237 put_and_out:
238 dput(this);
239 this = NULL;
240 goto out;
241
242 out_err:
243 dput(this);
244 return err;
245 }
246
247 static int ovl_lookup_layer(struct dentry *base, struct ovl_lookup_data *d,
248 struct dentry **ret)
249 {
250 /* Counting down from the end, since the prefix can change */
251 size_t rem = d->name.len - 1;
252 struct dentry *dentry = NULL;
253 int err;
254
255 if (d->name.name[0] != '/')
256 return ovl_lookup_single(base, d, d->name.name, d->name.len,
257 0, "", ret);
258
259 while (!IS_ERR_OR_NULL(base) && d_can_lookup(base)) {
260 const char *s = d->name.name + d->name.len - rem;
261 const char *next = strchrnul(s, '/');
262 size_t thislen = next - s;
263 bool end = !next[0];
264
265 /* Verify we did not go off the rails */
266 if (WARN_ON(s[-1] != '/'))
267 return -EIO;
268
269 err = ovl_lookup_single(base, d, s, thislen,
270 d->name.len - rem, next, &base);
271 dput(dentry);
272 if (err)
273 return err;
274 dentry = base;
275 if (end)
276 break;
277
278 rem -= thislen + 1;
279
280 if (WARN_ON(rem >= d->name.len))
281 return -EIO;
282 }
283 *ret = dentry;
284 return 0;
285 }
286
287
288 static int ovl_check_origin(struct dentry *upperdentry,
289 struct path *lowerstack, unsigned int numlower,
290 struct path **stackp, unsigned int *ctrp)
291 {
292 struct vfsmount *mnt;
293 struct dentry *origin = NULL;
294 int i;
295
296
297 for (i = 0; i < numlower; i++) {
298 mnt = lowerstack[i].mnt;
299 origin = ovl_get_origin(upperdentry, mnt);
300 if (IS_ERR(origin))
301 return PTR_ERR(origin);
302
303 if (origin)
304 break;
305 }
306
307 if (!origin)
308 return 0;
309
310 BUG_ON(*ctrp);
311 if (!*stackp)
312 *stackp = kmalloc(sizeof(struct path), GFP_TEMPORARY);
313 if (!*stackp) {
314 dput(origin);
315 return -ENOMEM;
316 }
317 **stackp = (struct path) { .dentry = origin, .mnt = mnt };
318 *ctrp = 1;
319
320 return 0;
321 }
322
323 /*
324 * Verify that @fh matches the origin file handle stored in OVL_XATTR_ORIGIN.
325 * Return 0 on match, -ESTALE on mismatch, < 0 on error.
326 */
327 static int ovl_verify_origin_fh(struct dentry *dentry, const struct ovl_fh *fh)
328 {
329 struct ovl_fh *ofh = ovl_get_origin_fh(dentry);
330 int err = 0;
331
332 if (!ofh)
333 return -ENODATA;
334
335 if (IS_ERR(ofh))
336 return PTR_ERR(ofh);
337
338 if (fh->len != ofh->len || memcmp(fh, ofh, fh->len))
339 err = -ESTALE;
340
341 kfree(ofh);
342 return err;
343 }
344
345 /*
346 * Verify that an inode matches the origin file handle stored in upper inode.
347 *
348 * If @set is true and there is no stored file handle, encode and store origin
349 * file handle in OVL_XATTR_ORIGIN.
350 *
351 * Return 0 on match, -ESTALE on mismatch, < 0 on error.
352 */
353 int ovl_verify_origin(struct dentry *dentry, struct vfsmount *mnt,
354 struct dentry *origin, bool is_upper, bool set)
355 {
356 struct inode *inode;
357 struct ovl_fh *fh;
358 int err;
359
360 fh = ovl_encode_fh(origin, is_upper);
361 err = PTR_ERR(fh);
362 if (IS_ERR(fh))
363 goto fail;
364
365 err = ovl_verify_origin_fh(dentry, fh);
366 if (set && err == -ENODATA)
367 err = ovl_do_setxattr(dentry, OVL_XATTR_ORIGIN, fh, fh->len, 0);
368 if (err)
369 goto fail;
370
371 out:
372 kfree(fh);
373 return err;
374
375 fail:
376 inode = d_inode(origin);
377 pr_warn_ratelimited("overlayfs: failed to verify origin (%pd2, ino=%lu, err=%i)\n",
378 origin, inode ? inode->i_ino : 0, err);
379 goto out;
380 }
381
382 /*
383 * Verify that an index entry name matches the origin file handle stored in
384 * OVL_XATTR_ORIGIN and that origin file handle can be decoded to lower path.
385 * Return 0 on match, -ESTALE on mismatch or stale origin, < 0 on error.
386 */
387 int ovl_verify_index(struct dentry *index, struct path *lowerstack,
388 unsigned int numlower)
389 {
390 struct ovl_fh *fh = NULL;
391 size_t len;
392 struct path origin = { };
393 struct path *stack = &origin;
394 unsigned int ctr = 0;
395 int err;
396
397 if (!d_inode(index))
398 return 0;
399
400 err = -EISDIR;
401 if (d_is_dir(index))
402 goto fail;
403
404 err = -EINVAL;
405 if (index->d_name.len < sizeof(struct ovl_fh)*2)
406 goto fail;
407
408 err = -ENOMEM;
409 len = index->d_name.len / 2;
410 fh = kzalloc(len, GFP_TEMPORARY);
411 if (!fh)
412 goto fail;
413
414 err = -EINVAL;
415 if (hex2bin((u8 *)fh, index->d_name.name, len) || len != fh->len)
416 goto fail;
417
418 err = ovl_verify_origin_fh(index, fh);
419 if (err)
420 goto fail;
421
422 err = ovl_check_origin(index, lowerstack, numlower, &stack, &ctr);
423 if (!err && !ctr)
424 err = -ESTALE;
425 if (err)
426 goto fail;
427
428 /* Check if index is orphan and don't warn before cleaning it */
429 if (d_inode(index)->i_nlink == 1 &&
430 ovl_get_nlink(index, origin.dentry, 0) == 0)
431 err = -ENOENT;
432
433 dput(origin.dentry);
434 out:
435 kfree(fh);
436 return err;
437
438 fail:
439 pr_warn_ratelimited("overlayfs: failed to verify index (%pd2, err=%i)\n",
440 index, err);
441 goto out;
442 }
443
444 /*
445 * Lookup in indexdir for the index entry of a lower real inode or a copy up
446 * origin inode. The index entry name is the hex representation of the lower
447 * inode file handle.
448 *
449 * If the index dentry in negative, then either no lower aliases have been
450 * copied up yet, or aliases have been copied up in older kernels and are
451 * not indexed.
452 *
453 * If the index dentry for a copy up origin inode is positive, but points
454 * to an inode different than the upper inode, then either the upper inode
455 * has been copied up and not indexed or it was indexed, but since then
456 * index dir was cleared. Either way, that index cannot be used to indentify
457 * the overlay inode.
458 */
459 int ovl_get_index_name(struct dentry *origin, struct qstr *name)
460 {
461 int err;
462 struct ovl_fh *fh;
463 char *n, *s;
464
465 fh = ovl_encode_fh(origin, false);
466 if (IS_ERR(fh))
467 return PTR_ERR(fh);
468
469 err = -ENOMEM;
470 n = kzalloc(fh->len * 2, GFP_TEMPORARY);
471 if (n) {
472 s = bin2hex(n, fh, fh->len);
473 *name = (struct qstr) QSTR_INIT(n, s - n);
474 err = 0;
475 }
476 kfree(fh);
477
478 return err;
479
480 }
481
482 static struct dentry *ovl_lookup_index(struct dentry *dentry,
483 struct dentry *upper,
484 struct dentry *origin)
485 {
486 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
487 struct dentry *index;
488 struct inode *inode;
489 struct qstr name;
490 int err;
491
492 err = ovl_get_index_name(origin, &name);
493 if (err)
494 return ERR_PTR(err);
495
496 index = lookup_one_len_unlocked(name.name, ofs->indexdir, name.len);
497 if (IS_ERR(index)) {
498 pr_warn_ratelimited("overlayfs: failed inode index lookup (ino=%lu, key=%*s, err=%i);\n"
499 "overlayfs: mount with '-o index=off' to disable inodes index.\n",
500 d_inode(origin)->i_ino, name.len, name.name,
501 err);
502 goto out;
503 }
504
505 if (d_is_negative(index)) {
506 if (upper && d_inode(origin)->i_nlink > 1) {
507 pr_warn_ratelimited("overlayfs: hard link with origin but no index (ino=%lu).\n",
508 d_inode(origin)->i_ino);
509 goto fail;
510 }
511
512 dput(index);
513 index = NULL;
514 } else if (upper && d_inode(index) != d_inode(upper)) {
515 inode = d_inode(index);
516 pr_warn_ratelimited("overlayfs: wrong index found (index ino: %lu, upper ino: %lu).\n",
517 d_inode(index)->i_ino,
518 d_inode(upper)->i_ino);
519 goto fail;
520 }
521
522 out:
523 kfree(name.name);
524 return index;
525
526 fail:
527 dput(index);
528 index = ERR_PTR(-EIO);
529 goto out;
530 }
531
532 /*
533 * Returns next layer in stack starting from top.
534 * Returns -1 if this is the last layer.
535 */
536 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
537 {
538 struct ovl_entry *oe = dentry->d_fsdata;
539
540 BUG_ON(idx < 0);
541 if (idx == 0) {
542 ovl_path_upper(dentry, path);
543 if (path->dentry)
544 return oe->numlower ? 1 : -1;
545 idx++;
546 }
547 BUG_ON(idx > oe->numlower);
548 *path = oe->lowerstack[idx - 1];
549
550 return (idx < oe->numlower) ? idx + 1 : -1;
551 }
552
553 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
554 unsigned int flags)
555 {
556 struct ovl_entry *oe;
557 const struct cred *old_cred;
558 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
559 struct ovl_entry *poe = dentry->d_parent->d_fsdata;
560 struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
561 struct path *stack = NULL;
562 struct dentry *upperdir, *upperdentry = NULL;
563 struct dentry *index = NULL;
564 unsigned int ctr = 0;
565 struct inode *inode = NULL;
566 bool upperopaque = false;
567 char *upperredirect = NULL;
568 struct dentry *this;
569 unsigned int i;
570 int err;
571 struct ovl_lookup_data d = {
572 .name = dentry->d_name,
573 .is_dir = false,
574 .opaque = false,
575 .stop = false,
576 .last = !poe->numlower,
577 .redirect = NULL,
578 };
579
580 if (dentry->d_name.len > ofs->namelen)
581 return ERR_PTR(-ENAMETOOLONG);
582
583 old_cred = ovl_override_creds(dentry->d_sb);
584 upperdir = ovl_dentry_upper(dentry->d_parent);
585 if (upperdir) {
586 err = ovl_lookup_layer(upperdir, &d, &upperdentry);
587 if (err)
588 goto out;
589
590 if (upperdentry && unlikely(ovl_dentry_remote(upperdentry))) {
591 dput(upperdentry);
592 err = -EREMOTE;
593 goto out;
594 }
595 if (upperdentry && !d.is_dir) {
596 BUG_ON(!d.stop || d.redirect);
597 /*
598 * Lookup copy up origin by decoding origin file handle.
599 * We may get a disconnected dentry, which is fine,
600 * because we only need to hold the origin inode in
601 * cache and use its inode number. We may even get a
602 * connected dentry, that is not under any of the lower
603 * layers root. That is also fine for using it's inode
604 * number - it's the same as if we held a reference
605 * to a dentry in lower layer that was moved under us.
606 */
607 err = ovl_check_origin(upperdentry, roe->lowerstack,
608 roe->numlower, &stack, &ctr);
609 if (err)
610 goto out;
611 }
612
613 if (d.redirect) {
614 upperredirect = kstrdup(d.redirect, GFP_KERNEL);
615 if (!upperredirect)
616 goto out_put_upper;
617 if (d.redirect[0] == '/')
618 poe = roe;
619 }
620 upperopaque = d.opaque;
621 }
622
623 if (!d.stop && poe->numlower) {
624 err = -ENOMEM;
625 stack = kcalloc(ofs->numlower, sizeof(struct path),
626 GFP_TEMPORARY);
627 if (!stack)
628 goto out_put_upper;
629 }
630
631 for (i = 0; !d.stop && i < poe->numlower; i++) {
632 struct path lowerpath = poe->lowerstack[i];
633
634 d.last = i == poe->numlower - 1;
635 err = ovl_lookup_layer(lowerpath.dentry, &d, &this);
636 if (err)
637 goto out_put;
638
639 if (!this)
640 continue;
641
642 stack[ctr].dentry = this;
643 stack[ctr].mnt = lowerpath.mnt;
644 ctr++;
645
646 if (d.stop)
647 break;
648
649 if (d.redirect && d.redirect[0] == '/' && poe != roe) {
650 poe = roe;
651
652 /* Find the current layer on the root dentry */
653 for (i = 0; i < poe->numlower; i++)
654 if (poe->lowerstack[i].mnt == lowerpath.mnt)
655 break;
656 if (WARN_ON(i == poe->numlower))
657 break;
658 }
659 }
660
661 /* Lookup index by lower inode and verify it matches upper inode */
662 if (ctr && !d.is_dir && ovl_indexdir(dentry->d_sb)) {
663 struct dentry *origin = stack[0].dentry;
664
665 index = ovl_lookup_index(dentry, upperdentry, origin);
666 if (IS_ERR(index)) {
667 err = PTR_ERR(index);
668 index = NULL;
669 goto out_put;
670 }
671 }
672
673 oe = ovl_alloc_entry(ctr);
674 err = -ENOMEM;
675 if (!oe)
676 goto out_put;
677
678 oe->opaque = upperopaque;
679 memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
680 dentry->d_fsdata = oe;
681
682 if (upperdentry)
683 ovl_dentry_set_upper_alias(dentry);
684 else if (index)
685 upperdentry = dget(index);
686
687 if (upperdentry || ctr) {
688 inode = ovl_get_inode(dentry, upperdentry);
689 err = PTR_ERR(inode);
690 if (IS_ERR(inode))
691 goto out_free_oe;
692
693 OVL_I(inode)->redirect = upperredirect;
694 if (index)
695 ovl_set_flag(OVL_INDEX, inode);
696 }
697
698 revert_creds(old_cred);
699 dput(index);
700 kfree(stack);
701 kfree(d.redirect);
702 d_add(dentry, inode);
703
704 return NULL;
705
706 out_free_oe:
707 dentry->d_fsdata = NULL;
708 kfree(oe);
709 out_put:
710 dput(index);
711 for (i = 0; i < ctr; i++)
712 dput(stack[i].dentry);
713 kfree(stack);
714 out_put_upper:
715 dput(upperdentry);
716 kfree(upperredirect);
717 out:
718 kfree(d.redirect);
719 revert_creds(old_cred);
720 return ERR_PTR(err);
721 }
722
723 bool ovl_lower_positive(struct dentry *dentry)
724 {
725 struct ovl_entry *oe = dentry->d_fsdata;
726 struct ovl_entry *poe = dentry->d_parent->d_fsdata;
727 const struct qstr *name = &dentry->d_name;
728 unsigned int i;
729 bool positive = false;
730 bool done = false;
731
732 /*
733 * If dentry is negative, then lower is positive iff this is a
734 * whiteout.
735 */
736 if (!dentry->d_inode)
737 return oe->opaque;
738
739 /* Negative upper -> positive lower */
740 if (!ovl_dentry_upper(dentry))
741 return true;
742
743 /* Positive upper -> have to look up lower to see whether it exists */
744 for (i = 0; !done && !positive && i < poe->numlower; i++) {
745 struct dentry *this;
746 struct dentry *lowerdir = poe->lowerstack[i].dentry;
747
748 this = lookup_one_len_unlocked(name->name, lowerdir,
749 name->len);
750 if (IS_ERR(this)) {
751 switch (PTR_ERR(this)) {
752 case -ENOENT:
753 case -ENAMETOOLONG:
754 break;
755
756 default:
757 /*
758 * Assume something is there, we just couldn't
759 * access it.
760 */
761 positive = true;
762 break;
763 }
764 } else {
765 if (this->d_inode) {
766 positive = !ovl_is_whiteout(this);
767 done = true;
768 }
769 dput(this);
770 }
771 }
772
773 return positive;
774 }