Merge tag 'extcon-next-for-4.20' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux/fpc-iii.git] / fs / overlayfs / namei.c
blobf28711846dd6ebad2cca1ad5fc3a4ef95f3806b7
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/ctype.h>
13 #include <linux/namei.h>
14 #include <linux/xattr.h>
15 #include <linux/ratelimit.h>
16 #include <linux/mount.h>
17 #include <linux/exportfs.h>
18 #include "overlayfs.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;
27 bool metacopy;
30 static int ovl_check_redirect(struct dentry *dentry, struct ovl_lookup_data *d,
31 size_t prelen, const char *post)
33 int res;
34 char *buf;
36 buf = ovl_get_redirect_xattr(dentry, prelen + strlen(post));
37 if (IS_ERR_OR_NULL(buf))
38 return PTR_ERR(buf);
40 if (buf[0] == '/') {
42 * One of the ancestor path elements in an absolute path
43 * lookup in ovl_lookup_layer() could have been opaque and
44 * that will stop further lookup in lower layers (d->stop=true)
45 * But we have found an absolute redirect in decendant path
46 * element and that should force continue lookup in lower
47 * layers (reset d->stop).
49 d->stop = false;
50 } else {
51 res = strlen(buf) + 1;
52 memmove(buf + prelen, buf, res);
53 memcpy(buf, d->name.name, prelen);
56 strcat(buf, post);
57 kfree(d->redirect);
58 d->redirect = buf;
59 d->name.name = d->redirect;
60 d->name.len = strlen(d->redirect);
62 return 0;
65 static int ovl_acceptable(void *ctx, struct dentry *dentry)
68 * A non-dir origin may be disconnected, which is fine, because
69 * we only need it for its unique inode number.
71 if (!d_is_dir(dentry))
72 return 1;
74 /* Don't decode a deleted empty directory */
75 if (d_unhashed(dentry))
76 return 0;
78 /* Check if directory belongs to the layer we are decoding from */
79 return is_subdir(dentry, ((struct vfsmount *)ctx)->mnt_root);
83 * Check validity of an overlay file handle buffer.
85 * Return 0 for a valid file handle.
86 * Return -ENODATA for "origin unknown".
87 * Return <0 for an invalid file handle.
89 int ovl_check_fh_len(struct ovl_fh *fh, int fh_len)
91 if (fh_len < sizeof(struct ovl_fh) || fh_len < fh->len)
92 return -EINVAL;
94 if (fh->magic != OVL_FH_MAGIC)
95 return -EINVAL;
97 /* Treat larger version and unknown flags as "origin unknown" */
98 if (fh->version > OVL_FH_VERSION || fh->flags & ~OVL_FH_FLAG_ALL)
99 return -ENODATA;
101 /* Treat endianness mismatch as "origin unknown" */
102 if (!(fh->flags & OVL_FH_FLAG_ANY_ENDIAN) &&
103 (fh->flags & OVL_FH_FLAG_BIG_ENDIAN) != OVL_FH_FLAG_CPU_ENDIAN)
104 return -ENODATA;
106 return 0;
109 static struct ovl_fh *ovl_get_fh(struct dentry *dentry, const char *name)
111 int res, err;
112 struct ovl_fh *fh = NULL;
114 res = vfs_getxattr(dentry, name, NULL, 0);
115 if (res < 0) {
116 if (res == -ENODATA || res == -EOPNOTSUPP)
117 return NULL;
118 goto fail;
120 /* Zero size value means "copied up but origin unknown" */
121 if (res == 0)
122 return NULL;
124 fh = kzalloc(res, GFP_KERNEL);
125 if (!fh)
126 return ERR_PTR(-ENOMEM);
128 res = vfs_getxattr(dentry, name, fh, res);
129 if (res < 0)
130 goto fail;
132 err = ovl_check_fh_len(fh, res);
133 if (err < 0) {
134 if (err == -ENODATA)
135 goto out;
136 goto invalid;
139 return fh;
141 out:
142 kfree(fh);
143 return NULL;
145 fail:
146 pr_warn_ratelimited("overlayfs: failed to get origin (%i)\n", res);
147 goto out;
148 invalid:
149 pr_warn_ratelimited("overlayfs: invalid origin (%*phN)\n", res, fh);
150 goto out;
153 struct dentry *ovl_decode_real_fh(struct ovl_fh *fh, struct vfsmount *mnt,
154 bool connected)
156 struct dentry *real;
157 int bytes;
160 * Make sure that the stored uuid matches the uuid of the lower
161 * layer where file handle will be decoded.
163 if (!uuid_equal(&fh->uuid, &mnt->mnt_sb->s_uuid))
164 return NULL;
166 bytes = (fh->len - offsetof(struct ovl_fh, fid));
167 real = exportfs_decode_fh(mnt, (struct fid *)fh->fid,
168 bytes >> 2, (int)fh->type,
169 connected ? ovl_acceptable : NULL, mnt);
170 if (IS_ERR(real)) {
172 * Treat stale file handle to lower file as "origin unknown".
173 * upper file handle could become stale when upper file is
174 * unlinked and this information is needed to handle stale
175 * index entries correctly.
177 if (real == ERR_PTR(-ESTALE) &&
178 !(fh->flags & OVL_FH_FLAG_PATH_UPPER))
179 real = NULL;
180 return real;
183 if (ovl_dentry_weird(real)) {
184 dput(real);
185 return NULL;
188 return real;
191 static bool ovl_is_opaquedir(struct dentry *dentry)
193 return ovl_check_dir_xattr(dentry, OVL_XATTR_OPAQUE);
196 static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
197 const char *name, unsigned int namelen,
198 size_t prelen, const char *post,
199 struct dentry **ret)
201 struct dentry *this;
202 int err;
203 bool last_element = !post[0];
205 this = lookup_one_len_unlocked(name, base, namelen);
206 if (IS_ERR(this)) {
207 err = PTR_ERR(this);
208 this = NULL;
209 if (err == -ENOENT || err == -ENAMETOOLONG)
210 goto out;
211 goto out_err;
213 if (!this->d_inode)
214 goto put_and_out;
216 if (ovl_dentry_weird(this)) {
217 /* Don't support traversing automounts and other weirdness */
218 err = -EREMOTE;
219 goto out_err;
221 if (ovl_is_whiteout(this)) {
222 d->stop = d->opaque = true;
223 goto put_and_out;
226 * This dentry should be a regular file if previous layer lookup
227 * found a metacopy dentry.
229 if (last_element && d->metacopy && !d_is_reg(this)) {
230 d->stop = true;
231 goto put_and_out;
233 if (!d_can_lookup(this)) {
234 if (d->is_dir || !last_element) {
235 d->stop = true;
236 goto put_and_out;
238 err = ovl_check_metacopy_xattr(this);
239 if (err < 0)
240 goto out_err;
242 d->metacopy = err;
243 d->stop = !d->metacopy;
244 if (!d->metacopy || d->last)
245 goto out;
246 } else {
247 if (last_element)
248 d->is_dir = true;
249 if (d->last)
250 goto out;
252 if (ovl_is_opaquedir(this)) {
253 d->stop = true;
254 if (last_element)
255 d->opaque = true;
256 goto out;
259 err = ovl_check_redirect(this, d, prelen, post);
260 if (err)
261 goto out_err;
262 out:
263 *ret = this;
264 return 0;
266 put_and_out:
267 dput(this);
268 this = NULL;
269 goto out;
271 out_err:
272 dput(this);
273 return err;
276 static int ovl_lookup_layer(struct dentry *base, struct ovl_lookup_data *d,
277 struct dentry **ret)
279 /* Counting down from the end, since the prefix can change */
280 size_t rem = d->name.len - 1;
281 struct dentry *dentry = NULL;
282 int err;
284 if (d->name.name[0] != '/')
285 return ovl_lookup_single(base, d, d->name.name, d->name.len,
286 0, "", ret);
288 while (!IS_ERR_OR_NULL(base) && d_can_lookup(base)) {
289 const char *s = d->name.name + d->name.len - rem;
290 const char *next = strchrnul(s, '/');
291 size_t thislen = next - s;
292 bool end = !next[0];
294 /* Verify we did not go off the rails */
295 if (WARN_ON(s[-1] != '/'))
296 return -EIO;
298 err = ovl_lookup_single(base, d, s, thislen,
299 d->name.len - rem, next, &base);
300 dput(dentry);
301 if (err)
302 return err;
303 dentry = base;
304 if (end)
305 break;
307 rem -= thislen + 1;
309 if (WARN_ON(rem >= d->name.len))
310 return -EIO;
312 *ret = dentry;
313 return 0;
317 int ovl_check_origin_fh(struct ovl_fs *ofs, struct ovl_fh *fh, bool connected,
318 struct dentry *upperdentry, struct ovl_path **stackp)
320 struct dentry *origin = NULL;
321 int i;
323 for (i = 0; i < ofs->numlower; i++) {
324 origin = ovl_decode_real_fh(fh, ofs->lower_layers[i].mnt,
325 connected);
326 if (origin)
327 break;
330 if (!origin)
331 return -ESTALE;
332 else if (IS_ERR(origin))
333 return PTR_ERR(origin);
335 if (upperdentry && !ovl_is_whiteout(upperdentry) &&
336 ((d_inode(origin)->i_mode ^ d_inode(upperdentry)->i_mode) & S_IFMT))
337 goto invalid;
339 if (!*stackp)
340 *stackp = kmalloc(sizeof(struct ovl_path), GFP_KERNEL);
341 if (!*stackp) {
342 dput(origin);
343 return -ENOMEM;
345 **stackp = (struct ovl_path){
346 .dentry = origin,
347 .layer = &ofs->lower_layers[i]
350 return 0;
352 invalid:
353 pr_warn_ratelimited("overlayfs: invalid origin (%pd2, ftype=%x, origin ftype=%x).\n",
354 upperdentry, d_inode(upperdentry)->i_mode & S_IFMT,
355 d_inode(origin)->i_mode & S_IFMT);
356 dput(origin);
357 return -EIO;
360 static int ovl_check_origin(struct ovl_fs *ofs, struct dentry *upperdentry,
361 struct ovl_path **stackp, unsigned int *ctrp)
363 struct ovl_fh *fh = ovl_get_fh(upperdentry, OVL_XATTR_ORIGIN);
364 int err;
366 if (IS_ERR_OR_NULL(fh))
367 return PTR_ERR(fh);
369 err = ovl_check_origin_fh(ofs, fh, false, upperdentry, stackp);
370 kfree(fh);
372 if (err) {
373 if (err == -ESTALE)
374 return 0;
375 return err;
378 if (WARN_ON(*ctrp))
379 return -EIO;
381 *ctrp = 1;
382 return 0;
386 * Verify that @fh matches the file handle stored in xattr @name.
387 * Return 0 on match, -ESTALE on mismatch, < 0 on error.
389 static int ovl_verify_fh(struct dentry *dentry, const char *name,
390 const struct ovl_fh *fh)
392 struct ovl_fh *ofh = ovl_get_fh(dentry, name);
393 int err = 0;
395 if (!ofh)
396 return -ENODATA;
398 if (IS_ERR(ofh))
399 return PTR_ERR(ofh);
401 if (fh->len != ofh->len || memcmp(fh, ofh, fh->len))
402 err = -ESTALE;
404 kfree(ofh);
405 return err;
409 * Verify that @real dentry matches the file handle stored in xattr @name.
411 * If @set is true and there is no stored file handle, encode @real and store
412 * file handle in xattr @name.
414 * Return 0 on match, -ESTALE on mismatch, -ENODATA on no xattr, < 0 on error.
416 int ovl_verify_set_fh(struct dentry *dentry, const char *name,
417 struct dentry *real, bool is_upper, bool set)
419 struct inode *inode;
420 struct ovl_fh *fh;
421 int err;
423 fh = ovl_encode_real_fh(real, is_upper);
424 err = PTR_ERR(fh);
425 if (IS_ERR(fh))
426 goto fail;
428 err = ovl_verify_fh(dentry, name, fh);
429 if (set && err == -ENODATA)
430 err = ovl_do_setxattr(dentry, name, fh, fh->len, 0);
431 if (err)
432 goto fail;
434 out:
435 kfree(fh);
436 return err;
438 fail:
439 inode = d_inode(real);
440 pr_warn_ratelimited("overlayfs: failed to verify %s (%pd2, ino=%lu, err=%i)\n",
441 is_upper ? "upper" : "origin", real,
442 inode ? inode->i_ino : 0, err);
443 goto out;
446 /* Get upper dentry from index */
447 struct dentry *ovl_index_upper(struct ovl_fs *ofs, struct dentry *index)
449 struct ovl_fh *fh;
450 struct dentry *upper;
452 if (!d_is_dir(index))
453 return dget(index);
455 fh = ovl_get_fh(index, OVL_XATTR_UPPER);
456 if (IS_ERR_OR_NULL(fh))
457 return ERR_CAST(fh);
459 upper = ovl_decode_real_fh(fh, ofs->upper_mnt, true);
460 kfree(fh);
462 if (IS_ERR_OR_NULL(upper))
463 return upper ?: ERR_PTR(-ESTALE);
465 if (!d_is_dir(upper)) {
466 pr_warn_ratelimited("overlayfs: invalid index upper (%pd2, upper=%pd2).\n",
467 index, upper);
468 dput(upper);
469 return ERR_PTR(-EIO);
472 return upper;
475 /* Is this a leftover from create/whiteout of directory index entry? */
476 static bool ovl_is_temp_index(struct dentry *index)
478 return index->d_name.name[0] == '#';
482 * Verify that an index entry name matches the origin file handle stored in
483 * OVL_XATTR_ORIGIN and that origin file handle can be decoded to lower path.
484 * Return 0 on match, -ESTALE on mismatch or stale origin, < 0 on error.
486 int ovl_verify_index(struct ovl_fs *ofs, struct dentry *index)
488 struct ovl_fh *fh = NULL;
489 size_t len;
490 struct ovl_path origin = { };
491 struct ovl_path *stack = &origin;
492 struct dentry *upper = NULL;
493 int err;
495 if (!d_inode(index))
496 return 0;
498 /* Cleanup leftover from index create/cleanup attempt */
499 err = -ESTALE;
500 if (ovl_is_temp_index(index))
501 goto fail;
503 err = -EINVAL;
504 if (index->d_name.len < sizeof(struct ovl_fh)*2)
505 goto fail;
507 err = -ENOMEM;
508 len = index->d_name.len / 2;
509 fh = kzalloc(len, GFP_KERNEL);
510 if (!fh)
511 goto fail;
513 err = -EINVAL;
514 if (hex2bin((u8 *)fh, index->d_name.name, len))
515 goto fail;
517 err = ovl_check_fh_len(fh, len);
518 if (err)
519 goto fail;
522 * Whiteout index entries are used as an indication that an exported
523 * overlay file handle should be treated as stale (i.e. after unlink
524 * of the overlay inode). These entries contain no origin xattr.
526 if (ovl_is_whiteout(index))
527 goto out;
530 * Verifying directory index entries are not stale is expensive, so
531 * only verify stale dir index if NFS export is enabled.
533 if (d_is_dir(index) && !ofs->config.nfs_export)
534 goto out;
537 * Directory index entries should have 'upper' xattr pointing to the
538 * real upper dir. Non-dir index entries are hardlinks to the upper
539 * real inode. For non-dir index, we can read the copy up origin xattr
540 * directly from the index dentry, but for dir index we first need to
541 * decode the upper directory.
543 upper = ovl_index_upper(ofs, index);
544 if (IS_ERR_OR_NULL(upper)) {
545 err = PTR_ERR(upper);
547 * Directory index entries with no 'upper' xattr need to be
548 * removed. When dir index entry has a stale 'upper' xattr,
549 * we assume that upper dir was removed and we treat the dir
550 * index as orphan entry that needs to be whited out.
552 if (err == -ESTALE)
553 goto orphan;
554 else if (!err)
555 err = -ESTALE;
556 goto fail;
559 err = ovl_verify_fh(upper, OVL_XATTR_ORIGIN, fh);
560 dput(upper);
561 if (err)
562 goto fail;
564 /* Check if non-dir index is orphan and don't warn before cleaning it */
565 if (!d_is_dir(index) && d_inode(index)->i_nlink == 1) {
566 err = ovl_check_origin_fh(ofs, fh, false, index, &stack);
567 if (err)
568 goto fail;
570 if (ovl_get_nlink(origin.dentry, index, 0) == 0)
571 goto orphan;
574 out:
575 dput(origin.dentry);
576 kfree(fh);
577 return err;
579 fail:
580 pr_warn_ratelimited("overlayfs: failed to verify index (%pd2, ftype=%x, err=%i)\n",
581 index, d_inode(index)->i_mode & S_IFMT, err);
582 goto out;
584 orphan:
585 pr_warn_ratelimited("overlayfs: orphan index entry (%pd2, ftype=%x, nlink=%u)\n",
586 index, d_inode(index)->i_mode & S_IFMT,
587 d_inode(index)->i_nlink);
588 err = -ENOENT;
589 goto out;
592 static int ovl_get_index_name_fh(struct ovl_fh *fh, struct qstr *name)
594 char *n, *s;
596 n = kcalloc(fh->len, 2, GFP_KERNEL);
597 if (!n)
598 return -ENOMEM;
600 s = bin2hex(n, fh, fh->len);
601 *name = (struct qstr) QSTR_INIT(n, s - n);
603 return 0;
608 * Lookup in indexdir for the index entry of a lower real inode or a copy up
609 * origin inode. The index entry name is the hex representation of the lower
610 * inode file handle.
612 * If the index dentry in negative, then either no lower aliases have been
613 * copied up yet, or aliases have been copied up in older kernels and are
614 * not indexed.
616 * If the index dentry for a copy up origin inode is positive, but points
617 * to an inode different than the upper inode, then either the upper inode
618 * has been copied up and not indexed or it was indexed, but since then
619 * index dir was cleared. Either way, that index cannot be used to indentify
620 * the overlay inode.
622 int ovl_get_index_name(struct dentry *origin, struct qstr *name)
624 struct ovl_fh *fh;
625 int err;
627 fh = ovl_encode_real_fh(origin, false);
628 if (IS_ERR(fh))
629 return PTR_ERR(fh);
631 err = ovl_get_index_name_fh(fh, name);
633 kfree(fh);
634 return err;
637 /* Lookup index by file handle for NFS export */
638 struct dentry *ovl_get_index_fh(struct ovl_fs *ofs, struct ovl_fh *fh)
640 struct dentry *index;
641 struct qstr name;
642 int err;
644 err = ovl_get_index_name_fh(fh, &name);
645 if (err)
646 return ERR_PTR(err);
648 index = lookup_one_len_unlocked(name.name, ofs->indexdir, name.len);
649 kfree(name.name);
650 if (IS_ERR(index)) {
651 if (PTR_ERR(index) == -ENOENT)
652 index = NULL;
653 return index;
656 if (d_is_negative(index))
657 err = 0;
658 else if (ovl_is_whiteout(index))
659 err = -ESTALE;
660 else if (ovl_dentry_weird(index))
661 err = -EIO;
662 else
663 return index;
665 dput(index);
666 return ERR_PTR(err);
669 struct dentry *ovl_lookup_index(struct ovl_fs *ofs, struct dentry *upper,
670 struct dentry *origin, bool verify)
672 struct dentry *index;
673 struct inode *inode;
674 struct qstr name;
675 bool is_dir = d_is_dir(origin);
676 int err;
678 err = ovl_get_index_name(origin, &name);
679 if (err)
680 return ERR_PTR(err);
682 index = lookup_one_len_unlocked(name.name, ofs->indexdir, name.len);
683 if (IS_ERR(index)) {
684 err = PTR_ERR(index);
685 if (err == -ENOENT) {
686 index = NULL;
687 goto out;
689 pr_warn_ratelimited("overlayfs: failed inode index lookup (ino=%lu, key=%*s, err=%i);\n"
690 "overlayfs: mount with '-o index=off' to disable inodes index.\n",
691 d_inode(origin)->i_ino, name.len, name.name,
692 err);
693 goto out;
696 inode = d_inode(index);
697 if (d_is_negative(index)) {
698 goto out_dput;
699 } else if (ovl_is_whiteout(index) && !verify) {
701 * When index lookup is called with !verify for decoding an
702 * overlay file handle, a whiteout index implies that decode
703 * should treat file handle as stale and no need to print a
704 * warning about it.
706 dput(index);
707 index = ERR_PTR(-ESTALE);
708 goto out;
709 } else if (ovl_dentry_weird(index) || ovl_is_whiteout(index) ||
710 ((inode->i_mode ^ d_inode(origin)->i_mode) & S_IFMT)) {
712 * Index should always be of the same file type as origin
713 * except for the case of a whiteout index. A whiteout
714 * index should only exist if all lower aliases have been
715 * unlinked, which means that finding a lower origin on lookup
716 * whose index is a whiteout should be treated as an error.
718 pr_warn_ratelimited("overlayfs: bad index found (index=%pd2, ftype=%x, origin ftype=%x).\n",
719 index, d_inode(index)->i_mode & S_IFMT,
720 d_inode(origin)->i_mode & S_IFMT);
721 goto fail;
722 } else if (is_dir && verify) {
723 if (!upper) {
724 pr_warn_ratelimited("overlayfs: suspected uncovered redirected dir found (origin=%pd2, index=%pd2).\n",
725 origin, index);
726 goto fail;
729 /* Verify that dir index 'upper' xattr points to upper dir */
730 err = ovl_verify_upper(index, upper, false);
731 if (err) {
732 if (err == -ESTALE) {
733 pr_warn_ratelimited("overlayfs: suspected multiply redirected dir found (upper=%pd2, origin=%pd2, index=%pd2).\n",
734 upper, origin, index);
736 goto fail;
738 } else if (upper && d_inode(upper) != inode) {
739 goto out_dput;
741 out:
742 kfree(name.name);
743 return index;
745 out_dput:
746 dput(index);
747 index = NULL;
748 goto out;
750 fail:
751 dput(index);
752 index = ERR_PTR(-EIO);
753 goto out;
757 * Returns next layer in stack starting from top.
758 * Returns -1 if this is the last layer.
760 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
762 struct ovl_entry *oe = dentry->d_fsdata;
764 BUG_ON(idx < 0);
765 if (idx == 0) {
766 ovl_path_upper(dentry, path);
767 if (path->dentry)
768 return oe->numlower ? 1 : -1;
769 idx++;
771 BUG_ON(idx > oe->numlower);
772 path->dentry = oe->lowerstack[idx - 1].dentry;
773 path->mnt = oe->lowerstack[idx - 1].layer->mnt;
775 return (idx < oe->numlower) ? idx + 1 : -1;
778 /* Fix missing 'origin' xattr */
779 static int ovl_fix_origin(struct dentry *dentry, struct dentry *lower,
780 struct dentry *upper)
782 int err;
784 if (ovl_check_origin_xattr(upper))
785 return 0;
787 err = ovl_want_write(dentry);
788 if (err)
789 return err;
791 err = ovl_set_origin(dentry, lower, upper);
792 if (!err)
793 err = ovl_set_impure(dentry->d_parent, upper->d_parent);
795 ovl_drop_write(dentry);
796 return err;
799 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
800 unsigned int flags)
802 struct ovl_entry *oe;
803 const struct cred *old_cred;
804 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
805 struct ovl_entry *poe = dentry->d_parent->d_fsdata;
806 struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
807 struct ovl_path *stack = NULL, *origin_path = NULL;
808 struct dentry *upperdir, *upperdentry = NULL;
809 struct dentry *origin = NULL;
810 struct dentry *index = NULL;
811 unsigned int ctr = 0;
812 struct inode *inode = NULL;
813 bool upperopaque = false;
814 char *upperredirect = NULL;
815 struct dentry *this;
816 unsigned int i;
817 int err;
818 bool metacopy = false;
819 struct ovl_lookup_data d = {
820 .name = dentry->d_name,
821 .is_dir = false,
822 .opaque = false,
823 .stop = false,
824 .last = ofs->config.redirect_follow ? false : !poe->numlower,
825 .redirect = NULL,
826 .metacopy = false,
829 if (dentry->d_name.len > ofs->namelen)
830 return ERR_PTR(-ENAMETOOLONG);
832 old_cred = ovl_override_creds(dentry->d_sb);
833 upperdir = ovl_dentry_upper(dentry->d_parent);
834 if (upperdir) {
835 err = ovl_lookup_layer(upperdir, &d, &upperdentry);
836 if (err)
837 goto out;
839 if (upperdentry && unlikely(ovl_dentry_remote(upperdentry))) {
840 dput(upperdentry);
841 err = -EREMOTE;
842 goto out;
844 if (upperdentry && !d.is_dir) {
845 unsigned int origin_ctr = 0;
848 * Lookup copy up origin by decoding origin file handle.
849 * We may get a disconnected dentry, which is fine,
850 * because we only need to hold the origin inode in
851 * cache and use its inode number. We may even get a
852 * connected dentry, that is not under any of the lower
853 * layers root. That is also fine for using it's inode
854 * number - it's the same as if we held a reference
855 * to a dentry in lower layer that was moved under us.
857 err = ovl_check_origin(ofs, upperdentry, &origin_path,
858 &origin_ctr);
859 if (err)
860 goto out_put_upper;
862 if (d.metacopy)
863 metacopy = true;
866 if (d.redirect) {
867 err = -ENOMEM;
868 upperredirect = kstrdup(d.redirect, GFP_KERNEL);
869 if (!upperredirect)
870 goto out_put_upper;
871 if (d.redirect[0] == '/')
872 poe = roe;
874 upperopaque = d.opaque;
877 if (!d.stop && poe->numlower) {
878 err = -ENOMEM;
879 stack = kcalloc(ofs->numlower, sizeof(struct ovl_path),
880 GFP_KERNEL);
881 if (!stack)
882 goto out_put_upper;
885 for (i = 0; !d.stop && i < poe->numlower; i++) {
886 struct ovl_path lower = poe->lowerstack[i];
888 if (!ofs->config.redirect_follow)
889 d.last = i == poe->numlower - 1;
890 else
891 d.last = lower.layer->idx == roe->numlower;
893 err = ovl_lookup_layer(lower.dentry, &d, &this);
894 if (err)
895 goto out_put;
897 if (!this)
898 continue;
901 * If no origin fh is stored in upper of a merge dir, store fh
902 * of lower dir and set upper parent "impure".
904 if (upperdentry && !ctr && !ofs->noxattr && d.is_dir) {
905 err = ovl_fix_origin(dentry, this, upperdentry);
906 if (err) {
907 dput(this);
908 goto out_put;
913 * When "verify_lower" feature is enabled, do not merge with a
914 * lower dir that does not match a stored origin xattr. In any
915 * case, only verified origin is used for index lookup.
917 * For non-dir dentry, if index=on, then ensure origin
918 * matches the dentry found using path based lookup,
919 * otherwise error out.
921 if (upperdentry && !ctr &&
922 ((d.is_dir && ovl_verify_lower(dentry->d_sb)) ||
923 (!d.is_dir && ofs->config.index && origin_path))) {
924 err = ovl_verify_origin(upperdentry, this, false);
925 if (err) {
926 dput(this);
927 if (d.is_dir)
928 break;
929 goto out_put;
931 origin = this;
934 if (d.metacopy)
935 metacopy = true;
937 * Do not store intermediate metacopy dentries in chain,
938 * except top most lower metacopy dentry
940 if (d.metacopy && ctr) {
941 dput(this);
942 continue;
945 stack[ctr].dentry = this;
946 stack[ctr].layer = lower.layer;
947 ctr++;
950 * Following redirects can have security consequences: it's like
951 * a symlink into the lower layer without the permission checks.
952 * This is only a problem if the upper layer is untrusted (e.g
953 * comes from an USB drive). This can allow a non-readable file
954 * or directory to become readable.
956 * Only following redirects when redirects are enabled disables
957 * this attack vector when not necessary.
959 err = -EPERM;
960 if (d.redirect && !ofs->config.redirect_follow) {
961 pr_warn_ratelimited("overlayfs: refusing to follow redirect for (%pd2)\n",
962 dentry);
963 goto out_put;
966 if (d.stop)
967 break;
969 if (d.redirect && d.redirect[0] == '/' && poe != roe) {
970 poe = roe;
971 /* Find the current layer on the root dentry */
972 i = lower.layer->idx - 1;
976 if (metacopy) {
978 * Found a metacopy dentry but did not find corresponding
979 * data dentry
981 if (d.metacopy) {
982 err = -EIO;
983 goto out_put;
986 err = -EPERM;
987 if (!ofs->config.metacopy) {
988 pr_warn_ratelimited("overlay: refusing to follow metacopy origin for (%pd2)\n",
989 dentry);
990 goto out_put;
992 } else if (!d.is_dir && upperdentry && !ctr && origin_path) {
993 if (WARN_ON(stack != NULL)) {
994 err = -EIO;
995 goto out_put;
997 stack = origin_path;
998 ctr = 1;
999 origin_path = NULL;
1003 * Lookup index by lower inode and verify it matches upper inode.
1004 * We only trust dir index if we verified that lower dir matches
1005 * origin, otherwise dir index entries may be inconsistent and we
1006 * ignore them.
1008 * For non-dir upper metacopy dentry, we already set "origin" if we
1009 * verified that lower matched upper origin. If upper origin was
1010 * not present (because lower layer did not support fh encode/decode),
1011 * or indexing is not enabled, do not set "origin" and skip looking up
1012 * index. This case should be handled in same way as a non-dir upper
1013 * without ORIGIN is handled.
1015 * Always lookup index of non-dir non-metacopy and non-upper.
1017 if (ctr && (!upperdentry || (!d.is_dir && !metacopy)))
1018 origin = stack[0].dentry;
1020 if (origin && ovl_indexdir(dentry->d_sb) &&
1021 (!d.is_dir || ovl_index_all(dentry->d_sb))) {
1022 index = ovl_lookup_index(ofs, upperdentry, origin, true);
1023 if (IS_ERR(index)) {
1024 err = PTR_ERR(index);
1025 index = NULL;
1026 goto out_put;
1030 oe = ovl_alloc_entry(ctr);
1031 err = -ENOMEM;
1032 if (!oe)
1033 goto out_put;
1035 memcpy(oe->lowerstack, stack, sizeof(struct ovl_path) * ctr);
1036 dentry->d_fsdata = oe;
1038 if (upperopaque)
1039 ovl_dentry_set_opaque(dentry);
1041 if (upperdentry)
1042 ovl_dentry_set_upper_alias(dentry);
1043 else if (index) {
1044 upperdentry = dget(index);
1045 upperredirect = ovl_get_redirect_xattr(upperdentry, 0);
1046 if (IS_ERR(upperredirect)) {
1047 err = PTR_ERR(upperredirect);
1048 upperredirect = NULL;
1049 goto out_free_oe;
1053 if (upperdentry || ctr) {
1054 struct ovl_inode_params oip = {
1055 .upperdentry = upperdentry,
1056 .lowerpath = stack,
1057 .index = index,
1058 .numlower = ctr,
1059 .redirect = upperredirect,
1060 .lowerdata = (ctr > 1 && !d.is_dir) ?
1061 stack[ctr - 1].dentry : NULL,
1064 inode = ovl_get_inode(dentry->d_sb, &oip);
1065 err = PTR_ERR(inode);
1066 if (IS_ERR(inode))
1067 goto out_free_oe;
1070 revert_creds(old_cred);
1071 if (origin_path) {
1072 dput(origin_path->dentry);
1073 kfree(origin_path);
1075 dput(index);
1076 kfree(stack);
1077 kfree(d.redirect);
1078 return d_splice_alias(inode, dentry);
1080 out_free_oe:
1081 dentry->d_fsdata = NULL;
1082 kfree(oe);
1083 out_put:
1084 dput(index);
1085 for (i = 0; i < ctr; i++)
1086 dput(stack[i].dentry);
1087 kfree(stack);
1088 out_put_upper:
1089 if (origin_path) {
1090 dput(origin_path->dentry);
1091 kfree(origin_path);
1093 dput(upperdentry);
1094 kfree(upperredirect);
1095 out:
1096 kfree(d.redirect);
1097 revert_creds(old_cred);
1098 return ERR_PTR(err);
1101 bool ovl_lower_positive(struct dentry *dentry)
1103 struct ovl_entry *poe = dentry->d_parent->d_fsdata;
1104 const struct qstr *name = &dentry->d_name;
1105 const struct cred *old_cred;
1106 unsigned int i;
1107 bool positive = false;
1108 bool done = false;
1111 * If dentry is negative, then lower is positive iff this is a
1112 * whiteout.
1114 if (!dentry->d_inode)
1115 return ovl_dentry_is_opaque(dentry);
1117 /* Negative upper -> positive lower */
1118 if (!ovl_dentry_upper(dentry))
1119 return true;
1121 old_cred = ovl_override_creds(dentry->d_sb);
1122 /* Positive upper -> have to look up lower to see whether it exists */
1123 for (i = 0; !done && !positive && i < poe->numlower; i++) {
1124 struct dentry *this;
1125 struct dentry *lowerdir = poe->lowerstack[i].dentry;
1127 this = lookup_one_len_unlocked(name->name, lowerdir,
1128 name->len);
1129 if (IS_ERR(this)) {
1130 switch (PTR_ERR(this)) {
1131 case -ENOENT:
1132 case -ENAMETOOLONG:
1133 break;
1135 default:
1137 * Assume something is there, we just couldn't
1138 * access it.
1140 positive = true;
1141 break;
1143 } else {
1144 if (this->d_inode) {
1145 positive = !ovl_is_whiteout(this);
1146 done = true;
1148 dput(this);
1151 revert_creds(old_cred);
1153 return positive;