mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / fs / overlayfs / namei.c
blob8442f9839c9026a884c29f09d2a463c5339a2643
1 /*
2 * Copyright (C) 2011 Novell Inc.
3 * Copyright (C) 2016 Red Hat, 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.
8 */
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"
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;
29 static int ovl_check_redirect(struct dentry *dentry, struct ovl_lookup_data *d,
30 size_t prelen, const char *post)
32 int res;
33 char *s, *next, *buf = NULL;
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;
41 buf = kzalloc(prelen + res + strlen(post) + 1, GFP_KERNEL);
42 if (!buf)
43 return -ENOMEM;
45 if (res == 0)
46 goto invalid;
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;
60 * One of the ancestor path elements in an absolute path
61 * lookup in ovl_lookup_layer() could have been opaque and
62 * that will stop further lookup in lower layers (d->stop=true)
63 * But we have found an absolute redirect in decendant path
64 * element and that should force continue lookup in lower
65 * layers (reset d->stop).
67 d->stop = false;
68 } else {
69 if (strchr(buf, '/') != NULL)
70 goto invalid;
72 memmove(buf + prelen, buf, res);
73 memcpy(buf, d->name.name, prelen);
76 strcat(buf, post);
77 kfree(d->redirect);
78 d->redirect = buf;
79 d->name.name = d->redirect;
80 d->name.len = strlen(d->redirect);
82 return 0;
84 err_free:
85 kfree(buf);
86 return 0;
87 fail:
88 pr_warn_ratelimited("overlayfs: failed to get redirect (%i)\n", res);
89 goto err_free;
90 invalid:
91 pr_warn_ratelimited("overlayfs: invalid redirect (%s)\n", buf);
92 goto err_free;
95 static int ovl_acceptable(void *ctx, struct dentry *dentry)
97 return 1;
100 static struct ovl_fh *ovl_get_origin_fh(struct dentry *dentry)
102 int res;
103 struct ovl_fh *fh = NULL;
105 res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, NULL, 0);
106 if (res < 0) {
107 if (res == -ENODATA || res == -EOPNOTSUPP)
108 return NULL;
109 goto fail;
111 /* Zero size value means "copied up but origin unknown" */
112 if (res == 0)
113 return NULL;
115 fh = kzalloc(res, GFP_KERNEL);
116 if (!fh)
117 return ERR_PTR(-ENOMEM);
119 res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, fh, res);
120 if (res < 0)
121 goto fail;
123 if (res < sizeof(struct ovl_fh) || res < fh->len)
124 goto invalid;
126 if (fh->magic != OVL_FH_MAGIC)
127 goto invalid;
129 /* Treat larger version and unknown flags as "origin unknown" */
130 if (fh->version > OVL_FH_VERSION || fh->flags & ~OVL_FH_FLAG_ALL)
131 goto out;
133 /* Treat endianness mismatch as "origin unknown" */
134 if (!(fh->flags & OVL_FH_FLAG_ANY_ENDIAN) &&
135 (fh->flags & OVL_FH_FLAG_BIG_ENDIAN) != OVL_FH_FLAG_CPU_ENDIAN)
136 goto out;
138 return fh;
140 out:
141 kfree(fh);
142 return NULL;
144 fail:
145 pr_warn_ratelimited("overlayfs: failed to get origin (%i)\n", res);
146 goto out;
147 invalid:
148 pr_warn_ratelimited("overlayfs: invalid origin (%*phN)\n", res, fh);
149 goto out;
152 static struct dentry *ovl_get_origin(struct dentry *dentry,
153 struct vfsmount *mnt)
155 struct dentry *origin = NULL;
156 struct ovl_fh *fh = ovl_get_origin_fh(dentry);
157 int bytes;
159 if (IS_ERR_OR_NULL(fh))
160 return (struct dentry *)fh;
163 * Make sure that the stored uuid matches the uuid of the lower
164 * layer where file handle will be decoded.
166 if (!uuid_equal(&fh->uuid, &mnt->mnt_sb->s_uuid))
167 goto out;
169 bytes = (fh->len - offsetof(struct ovl_fh, fid));
170 origin = exportfs_decode_fh(mnt, (struct fid *)fh->fid,
171 bytes >> 2, (int)fh->type,
172 ovl_acceptable, NULL);
173 if (IS_ERR(origin)) {
174 /* Treat stale file handle as "origin unknown" */
175 if (origin == ERR_PTR(-ESTALE))
176 origin = NULL;
177 goto out;
180 if (ovl_dentry_weird(origin) ||
181 ((d_inode(origin)->i_mode ^ d_inode(dentry)->i_mode) & S_IFMT))
182 goto invalid;
184 out:
185 kfree(fh);
186 return origin;
188 invalid:
189 pr_warn_ratelimited("overlayfs: invalid origin (%pd2)\n", origin);
190 dput(origin);
191 origin = NULL;
192 goto out;
195 static bool ovl_is_opaquedir(struct dentry *dentry)
197 return ovl_check_dir_xattr(dentry, OVL_XATTR_OPAQUE);
200 static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
201 const char *name, unsigned int namelen,
202 size_t prelen, const char *post,
203 struct dentry **ret)
205 struct dentry *this;
206 int err;
208 this = lookup_one_len_unlocked(name, base, namelen);
209 if (IS_ERR(this)) {
210 err = PTR_ERR(this);
211 this = NULL;
212 if (err == -ENOENT || err == -ENAMETOOLONG)
213 goto out;
214 goto out_err;
216 if (!this->d_inode)
217 goto put_and_out;
219 if (ovl_dentry_weird(this)) {
220 /* Don't support traversing automounts and other weirdness */
221 err = -EREMOTE;
222 goto out_err;
224 if (ovl_is_whiteout(this)) {
225 d->stop = d->opaque = true;
226 goto put_and_out;
228 if (!d_can_lookup(this)) {
229 d->stop = true;
230 if (d->is_dir)
231 goto put_and_out;
232 goto out;
234 d->is_dir = true;
235 if (!d->last && ovl_is_opaquedir(this)) {
236 d->stop = d->opaque = true;
237 goto out;
239 err = ovl_check_redirect(this, d, prelen, post);
240 if (err)
241 goto out_err;
242 out:
243 *ret = this;
244 return 0;
246 put_and_out:
247 dput(this);
248 this = NULL;
249 goto out;
251 out_err:
252 dput(this);
253 return err;
256 static int ovl_lookup_layer(struct dentry *base, struct ovl_lookup_data *d,
257 struct dentry **ret)
259 /* Counting down from the end, since the prefix can change */
260 size_t rem = d->name.len - 1;
261 struct dentry *dentry = NULL;
262 int err;
264 if (d->name.name[0] != '/')
265 return ovl_lookup_single(base, d, d->name.name, d->name.len,
266 0, "", ret);
268 while (!IS_ERR_OR_NULL(base) && d_can_lookup(base)) {
269 const char *s = d->name.name + d->name.len - rem;
270 const char *next = strchrnul(s, '/');
271 size_t thislen = next - s;
272 bool end = !next[0];
274 /* Verify we did not go off the rails */
275 if (WARN_ON(s[-1] != '/'))
276 return -EIO;
278 err = ovl_lookup_single(base, d, s, thislen,
279 d->name.len - rem, next, &base);
280 dput(dentry);
281 if (err)
282 return err;
283 dentry = base;
284 if (end)
285 break;
287 rem -= thislen + 1;
289 if (WARN_ON(rem >= d->name.len))
290 return -EIO;
292 *ret = dentry;
293 return 0;
297 static int ovl_check_origin(struct dentry *upperdentry,
298 struct path *lowerstack, unsigned int numlower,
299 struct path **stackp, unsigned int *ctrp)
301 struct vfsmount *mnt;
302 struct dentry *origin = NULL;
303 int i;
306 for (i = 0; i < numlower; i++) {
307 mnt = lowerstack[i].mnt;
308 origin = ovl_get_origin(upperdentry, mnt);
309 if (IS_ERR(origin))
310 return PTR_ERR(origin);
312 if (origin)
313 break;
316 if (!origin)
317 return 0;
319 BUG_ON(*ctrp);
320 if (!*stackp)
321 *stackp = kmalloc(sizeof(struct path), GFP_KERNEL);
322 if (!*stackp) {
323 dput(origin);
324 return -ENOMEM;
326 **stackp = (struct path) { .dentry = origin, .mnt = mnt };
327 *ctrp = 1;
329 return 0;
333 * Verify that @fh matches the origin file handle stored in OVL_XATTR_ORIGIN.
334 * Return 0 on match, -ESTALE on mismatch, < 0 on error.
336 static int ovl_verify_origin_fh(struct dentry *dentry, const struct ovl_fh *fh)
338 struct ovl_fh *ofh = ovl_get_origin_fh(dentry);
339 int err = 0;
341 if (!ofh)
342 return -ENODATA;
344 if (IS_ERR(ofh))
345 return PTR_ERR(ofh);
347 if (fh->len != ofh->len || memcmp(fh, ofh, fh->len))
348 err = -ESTALE;
350 kfree(ofh);
351 return err;
355 * Verify that an inode matches the origin file handle stored in upper inode.
357 * If @set is true and there is no stored file handle, encode and store origin
358 * file handle in OVL_XATTR_ORIGIN.
360 * Return 0 on match, -ESTALE on mismatch, < 0 on error.
362 int ovl_verify_origin(struct dentry *dentry, struct vfsmount *mnt,
363 struct dentry *origin, bool is_upper, bool set)
365 struct inode *inode;
366 struct ovl_fh *fh;
367 int err;
369 fh = ovl_encode_fh(origin, is_upper);
370 err = PTR_ERR(fh);
371 if (IS_ERR(fh)) {
372 fh = NULL;
373 goto fail;
376 err = ovl_verify_origin_fh(dentry, fh);
377 if (set && err == -ENODATA)
378 err = ovl_do_setxattr(dentry, OVL_XATTR_ORIGIN, fh, fh->len, 0);
379 if (err)
380 goto fail;
382 out:
383 kfree(fh);
384 return err;
386 fail:
387 inode = d_inode(origin);
388 pr_warn_ratelimited("overlayfs: failed to verify origin (%pd2, ino=%lu, err=%i)\n",
389 origin, inode ? inode->i_ino : 0, err);
390 goto out;
394 * Verify that an index entry name matches the origin file handle stored in
395 * OVL_XATTR_ORIGIN and that origin file handle can be decoded to lower path.
396 * Return 0 on match, -ESTALE on mismatch or stale origin, < 0 on error.
398 int ovl_verify_index(struct dentry *index, struct path *lowerstack,
399 unsigned int numlower)
401 struct ovl_fh *fh = NULL;
402 size_t len;
403 struct path origin = { };
404 struct path *stack = &origin;
405 unsigned int ctr = 0;
406 int err;
408 if (!d_inode(index))
409 return 0;
412 * Directory index entries are going to be used for looking up
413 * redirected upper dirs by lower dir fh when decoding an overlay
414 * file handle of a merge dir. Whiteout index entries are going to be
415 * used as an indication that an exported overlay file handle should
416 * be treated as stale (i.e. after unlink of the overlay inode).
417 * We don't know the verification rules for directory and whiteout
418 * index entries, because they have not been implemented yet, so return
419 * EINVAL if those entries are found to abort the mount to avoid
420 * corrupting an index that was created by a newer kernel.
422 err = -EINVAL;
423 if (d_is_dir(index) || ovl_is_whiteout(index))
424 goto fail;
426 if (index->d_name.len < sizeof(struct ovl_fh)*2)
427 goto fail;
429 err = -ENOMEM;
430 len = index->d_name.len / 2;
431 fh = kzalloc(len, GFP_KERNEL);
432 if (!fh)
433 goto fail;
435 err = -EINVAL;
436 if (hex2bin((u8 *)fh, index->d_name.name, len) || len != fh->len)
437 goto fail;
439 err = ovl_verify_origin_fh(index, fh);
440 if (err)
441 goto fail;
443 err = ovl_check_origin(index, lowerstack, numlower, &stack, &ctr);
444 if (!err && !ctr)
445 err = -ESTALE;
446 if (err)
447 goto fail;
449 /* Check if index is orphan and don't warn before cleaning it */
450 if (d_inode(index)->i_nlink == 1 &&
451 ovl_get_nlink(origin.dentry, index, 0) == 0)
452 err = -ENOENT;
454 dput(origin.dentry);
455 out:
456 kfree(fh);
457 return err;
459 fail:
460 pr_warn_ratelimited("overlayfs: failed to verify index (%pd2, ftype=%x, err=%i)\n",
461 index, d_inode(index)->i_mode & S_IFMT, err);
462 goto out;
466 * Lookup in indexdir for the index entry of a lower real inode or a copy up
467 * origin inode. The index entry name is the hex representation of the lower
468 * inode file handle.
470 * If the index dentry in negative, then either no lower aliases have been
471 * copied up yet, or aliases have been copied up in older kernels and are
472 * not indexed.
474 * If the index dentry for a copy up origin inode is positive, but points
475 * to an inode different than the upper inode, then either the upper inode
476 * has been copied up and not indexed or it was indexed, but since then
477 * index dir was cleared. Either way, that index cannot be used to indentify
478 * the overlay inode.
480 int ovl_get_index_name(struct dentry *origin, struct qstr *name)
482 int err;
483 struct ovl_fh *fh;
484 char *n, *s;
486 fh = ovl_encode_fh(origin, false);
487 if (IS_ERR(fh))
488 return PTR_ERR(fh);
490 err = -ENOMEM;
491 n = kzalloc(fh->len * 2, GFP_KERNEL);
492 if (n) {
493 s = bin2hex(n, fh, fh->len);
494 *name = (struct qstr) QSTR_INIT(n, s - n);
495 err = 0;
497 kfree(fh);
499 return err;
503 static struct dentry *ovl_lookup_index(struct dentry *dentry,
504 struct dentry *upper,
505 struct dentry *origin)
507 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
508 struct dentry *index;
509 struct inode *inode;
510 struct qstr name;
511 int err;
513 err = ovl_get_index_name(origin, &name);
514 if (err)
515 return ERR_PTR(err);
517 index = lookup_one_len_unlocked(name.name, ofs->indexdir, name.len);
518 if (IS_ERR(index)) {
519 err = PTR_ERR(index);
520 if (err == -ENOENT) {
521 index = NULL;
522 goto out;
524 pr_warn_ratelimited("overlayfs: failed inode index lookup (ino=%lu, key=%.*s, err=%i);\n"
525 "overlayfs: mount with '-o index=off' to disable inodes index.\n",
526 d_inode(origin)->i_ino, name.len, name.name,
527 err);
528 goto out;
531 inode = d_inode(index);
532 if (d_is_negative(index)) {
533 goto out_dput;
534 } else if (upper && d_inode(upper) != inode) {
535 goto out_dput;
536 } else if (ovl_dentry_weird(index) || ovl_is_whiteout(index) ||
537 ((inode->i_mode ^ d_inode(origin)->i_mode) & S_IFMT)) {
539 * Index should always be of the same file type as origin
540 * except for the case of a whiteout index. A whiteout
541 * index should only exist if all lower aliases have been
542 * unlinked, which means that finding a lower origin on lookup
543 * whose index is a whiteout should be treated as an error.
545 pr_warn_ratelimited("overlayfs: bad index found (index=%pd2, ftype=%x, origin ftype=%x).\n",
546 index, d_inode(index)->i_mode & S_IFMT,
547 d_inode(origin)->i_mode & S_IFMT);
548 goto fail;
551 out:
552 kfree(name.name);
553 return index;
555 out_dput:
556 dput(index);
557 index = NULL;
558 goto out;
560 fail:
561 dput(index);
562 index = ERR_PTR(-EIO);
563 goto out;
567 * Returns next layer in stack starting from top.
568 * Returns -1 if this is the last layer.
570 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
572 struct ovl_entry *oe = dentry->d_fsdata;
574 BUG_ON(idx < 0);
575 if (idx == 0) {
576 ovl_path_upper(dentry, path);
577 if (path->dentry)
578 return oe->numlower ? 1 : -1;
579 idx++;
581 BUG_ON(idx > oe->numlower);
582 *path = oe->lowerstack[idx - 1];
584 return (idx < oe->numlower) ? idx + 1 : -1;
587 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
588 unsigned int flags)
590 struct ovl_entry *oe;
591 const struct cred *old_cred;
592 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
593 struct ovl_entry *poe = dentry->d_parent->d_fsdata;
594 struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
595 struct path *stack = NULL;
596 struct dentry *upperdir, *upperdentry = NULL;
597 struct dentry *index = NULL;
598 unsigned int ctr = 0;
599 struct inode *inode = NULL;
600 bool upperopaque = false;
601 char *upperredirect = NULL;
602 struct dentry *this;
603 unsigned int i;
604 int err;
605 struct ovl_lookup_data d = {
606 .name = dentry->d_name,
607 .is_dir = false,
608 .opaque = false,
609 .stop = false,
610 .last = !poe->numlower,
611 .redirect = NULL,
614 if (dentry->d_name.len > ofs->namelen)
615 return ERR_PTR(-ENAMETOOLONG);
617 old_cred = ovl_override_creds(dentry->d_sb);
618 upperdir = ovl_dentry_upper(dentry->d_parent);
619 if (upperdir) {
620 err = ovl_lookup_layer(upperdir, &d, &upperdentry);
621 if (err)
622 goto out;
624 if (upperdentry && unlikely(ovl_dentry_remote(upperdentry))) {
625 dput(upperdentry);
626 err = -EREMOTE;
627 goto out;
629 if (upperdentry && !d.is_dir) {
630 BUG_ON(!d.stop || d.redirect);
632 * Lookup copy up origin by decoding origin file handle.
633 * We may get a disconnected dentry, which is fine,
634 * because we only need to hold the origin inode in
635 * cache and use its inode number. We may even get a
636 * connected dentry, that is not under any of the lower
637 * layers root. That is also fine for using it's inode
638 * number - it's the same as if we held a reference
639 * to a dentry in lower layer that was moved under us.
641 err = ovl_check_origin(upperdentry, roe->lowerstack,
642 roe->numlower, &stack, &ctr);
643 if (err)
644 goto out_put_upper;
647 if (d.redirect) {
648 err = -ENOMEM;
649 upperredirect = kstrdup(d.redirect, GFP_KERNEL);
650 if (!upperredirect)
651 goto out_put_upper;
652 if (d.redirect[0] == '/')
653 poe = roe;
655 upperopaque = d.opaque;
658 if (!d.stop && poe->numlower) {
659 err = -ENOMEM;
660 stack = kcalloc(ofs->numlower, sizeof(struct path),
661 GFP_KERNEL);
662 if (!stack)
663 goto out_put_upper;
666 for (i = 0; !d.stop && i < poe->numlower; i++) {
667 struct path lowerpath = poe->lowerstack[i];
669 d.last = i == poe->numlower - 1;
670 err = ovl_lookup_layer(lowerpath.dentry, &d, &this);
671 if (err)
672 goto out_put;
674 if (!this)
675 continue;
677 stack[ctr].dentry = this;
678 stack[ctr].mnt = lowerpath.mnt;
679 ctr++;
681 if (d.stop)
682 break;
684 if (d.redirect && d.redirect[0] == '/' && poe != roe) {
685 poe = roe;
687 /* Find the current layer on the root dentry */
688 for (i = 0; i < poe->numlower; i++)
689 if (poe->lowerstack[i].mnt == lowerpath.mnt)
690 break;
691 if (WARN_ON(i == poe->numlower))
692 break;
696 /* Lookup index by lower inode and verify it matches upper inode */
697 if (ctr && !d.is_dir && ovl_indexdir(dentry->d_sb)) {
698 struct dentry *origin = stack[0].dentry;
700 index = ovl_lookup_index(dentry, upperdentry, origin);
701 if (IS_ERR(index)) {
702 err = PTR_ERR(index);
703 index = NULL;
704 goto out_put;
708 oe = ovl_alloc_entry(ctr);
709 err = -ENOMEM;
710 if (!oe)
711 goto out_put;
713 oe->opaque = upperopaque;
714 memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
715 dentry->d_fsdata = oe;
717 if (upperdentry)
718 ovl_dentry_set_upper_alias(dentry);
719 else if (index)
720 upperdentry = dget(index);
722 if (upperdentry || ctr) {
723 inode = ovl_get_inode(dentry, upperdentry, index);
724 err = PTR_ERR(inode);
725 if (IS_ERR(inode))
726 goto out_free_oe;
728 OVL_I(inode)->redirect = upperredirect;
729 if (index)
730 ovl_set_flag(OVL_INDEX, inode);
733 revert_creds(old_cred);
734 dput(index);
735 kfree(stack);
736 kfree(d.redirect);
737 d_add(dentry, inode);
739 return NULL;
741 out_free_oe:
742 dentry->d_fsdata = NULL;
743 kfree(oe);
744 out_put:
745 dput(index);
746 for (i = 0; i < ctr; i++)
747 dput(stack[i].dentry);
748 kfree(stack);
749 out_put_upper:
750 dput(upperdentry);
751 kfree(upperredirect);
752 out:
753 kfree(d.redirect);
754 revert_creds(old_cred);
755 return ERR_PTR(err);
758 bool ovl_lower_positive(struct dentry *dentry)
760 struct ovl_entry *oe = dentry->d_fsdata;
761 struct ovl_entry *poe = dentry->d_parent->d_fsdata;
762 const struct qstr *name = &dentry->d_name;
763 unsigned int i;
764 bool positive = false;
765 bool done = false;
768 * If dentry is negative, then lower is positive iff this is a
769 * whiteout.
771 if (!dentry->d_inode)
772 return oe->opaque;
774 /* Negative upper -> positive lower */
775 if (!ovl_dentry_upper(dentry))
776 return true;
778 /* Positive upper -> have to look up lower to see whether it exists */
779 for (i = 0; !done && !positive && i < poe->numlower; i++) {
780 struct dentry *this;
781 struct dentry *lowerdir = poe->lowerstack[i].dentry;
783 this = lookup_one_len_unlocked(name->name, lowerdir,
784 name->len);
785 if (IS_ERR(this)) {
786 switch (PTR_ERR(this)) {
787 case -ENOENT:
788 case -ENAMETOOLONG:
789 break;
791 default:
793 * Assume something is there, we just couldn't
794 * access it.
796 positive = true;
797 break;
799 } else {
800 if (this->d_inode) {
801 positive = !ovl_is_whiteout(this);
802 done = true;
804 dput(this);
808 return positive;