Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * Copyright (C) 2011 Novell Inc.
4 : * Copyright (C) 2016 Red Hat, Inc.
5 : */
6 :
7 : #include <linux/fs.h>
8 : #include <linux/cred.h>
9 : #include <linux/ctype.h>
10 : #include <linux/namei.h>
11 : #include <linux/xattr.h>
12 : #include <linux/ratelimit.h>
13 : #include <linux/mount.h>
14 : #include <linux/exportfs.h>
15 : #include "overlayfs.h"
16 :
17 : struct ovl_lookup_data {
18 : struct super_block *sb;
19 : struct qstr name;
20 : bool is_dir;
21 : bool opaque;
22 : bool stop;
23 : bool last;
24 : char *redirect;
25 : bool metacopy;
26 : };
27 :
28 4 : static int ovl_check_redirect(struct dentry *dentry, struct ovl_lookup_data *d,
29 : size_t prelen, const char *post)
30 : {
31 4 : int res;
32 4 : char *buf;
33 4 : struct ovl_fs *ofs = OVL_FS(d->sb);
34 :
35 4 : buf = ovl_get_redirect_xattr(ofs, dentry, prelen + strlen(post));
36 4 : if (IS_ERR_OR_NULL(buf))
37 4 : return PTR_ERR(buf);
38 :
39 0 : if (buf[0] == '/') {
40 : /*
41 : * One of the ancestor path elements in an absolute path
42 : * lookup in ovl_lookup_layer() could have been opaque and
43 : * that will stop further lookup in lower layers (d->stop=true)
44 : * But we have found an absolute redirect in decendant path
45 : * element and that should force continue lookup in lower
46 : * layers (reset d->stop).
47 : */
48 0 : d->stop = false;
49 : } else {
50 0 : res = strlen(buf) + 1;
51 0 : memmove(buf + prelen, buf, res);
52 0 : memcpy(buf, d->name.name, prelen);
53 : }
54 :
55 0 : strcat(buf, post);
56 0 : kfree(d->redirect);
57 0 : d->redirect = buf;
58 0 : d->name.name = d->redirect;
59 0 : d->name.len = strlen(d->redirect);
60 :
61 0 : return 0;
62 : }
63 :
64 0 : static int ovl_acceptable(void *ctx, struct dentry *dentry)
65 : {
66 : /*
67 : * A non-dir origin may be disconnected, which is fine, because
68 : * we only need it for its unique inode number.
69 : */
70 0 : if (!d_is_dir(dentry))
71 : return 1;
72 :
73 : /* Don't decode a deleted empty directory */
74 0 : if (d_unhashed(dentry))
75 : return 0;
76 :
77 : /* Check if directory belongs to the layer we are decoding from */
78 0 : return is_subdir(dentry, ((struct vfsmount *)ctx)->mnt_root);
79 : }
80 :
81 : /*
82 : * Check validity of an overlay file handle buffer.
83 : *
84 : * Return 0 for a valid file handle.
85 : * Return -ENODATA for "origin unknown".
86 : * Return <0 for an invalid file handle.
87 : */
88 0 : int ovl_check_fb_len(struct ovl_fb *fb, int fb_len)
89 : {
90 0 : if (fb_len < sizeof(struct ovl_fb) || fb_len < fb->len)
91 : return -EINVAL;
92 :
93 0 : if (fb->magic != OVL_FH_MAGIC)
94 : return -EINVAL;
95 :
96 : /* Treat larger version and unknown flags as "origin unknown" */
97 0 : if (fb->version > OVL_FH_VERSION || fb->flags & ~OVL_FH_FLAG_ALL)
98 : return -ENODATA;
99 :
100 : /* Treat endianness mismatch as "origin unknown" */
101 0 : if (!(fb->flags & OVL_FH_FLAG_ANY_ENDIAN) &&
102 : (fb->flags & OVL_FH_FLAG_BIG_ENDIAN) != OVL_FH_FLAG_CPU_ENDIAN)
103 0 : return -ENODATA;
104 :
105 : return 0;
106 : }
107 :
108 10 : static struct ovl_fh *ovl_get_fh(struct ovl_fs *ofs, struct dentry *dentry,
109 : enum ovl_xattr ox)
110 : {
111 10 : int res, err;
112 10 : struct ovl_fh *fh = NULL;
113 :
114 10 : res = ovl_do_getxattr(ofs, dentry, ox, NULL, 0);
115 10 : if (res < 0) {
116 10 : if (res == -ENODATA || res == -EOPNOTSUPP)
117 : return NULL;
118 0 : goto fail;
119 : }
120 : /* Zero size value means "copied up but origin unknown" */
121 0 : if (res == 0)
122 : return NULL;
123 :
124 0 : fh = kzalloc(res + OVL_FH_WIRE_OFFSET, GFP_KERNEL);
125 0 : if (!fh)
126 10 : return ERR_PTR(-ENOMEM);
127 :
128 0 : res = ovl_do_getxattr(ofs, dentry, ox, fh->buf, res);
129 0 : if (res < 0)
130 0 : goto fail;
131 :
132 0 : err = ovl_check_fb_len(&fh->fb, res);
133 0 : if (err < 0) {
134 0 : if (err == -ENODATA)
135 0 : goto out;
136 0 : goto invalid;
137 : }
138 :
139 : return fh;
140 :
141 0 : out:
142 0 : kfree(fh);
143 0 : return NULL;
144 :
145 0 : fail:
146 0 : pr_warn_ratelimited("failed to get origin (%i)\n", res);
147 0 : goto out;
148 0 : invalid:
149 0 : pr_warn_ratelimited("invalid origin (%*phN)\n", res, fh);
150 0 : goto out;
151 : }
152 :
153 0 : struct dentry *ovl_decode_real_fh(struct ovl_fs *ofs, struct ovl_fh *fh,
154 : struct vfsmount *mnt, bool connected)
155 : {
156 0 : struct dentry *real;
157 0 : int bytes;
158 :
159 0 : if (!capable(CAP_DAC_READ_SEARCH))
160 : return NULL;
161 :
162 : /*
163 : * Make sure that the stored uuid matches the uuid of the lower
164 : * layer where file handle will be decoded.
165 : * In case of uuid=off option just make sure that stored uuid is null.
166 : */
167 0 : if (ofs->config.uuid ? !uuid_equal(&fh->fb.uuid, &mnt->mnt_sb->s_uuid) :
168 0 : !uuid_is_null(&fh->fb.uuid))
169 : return NULL;
170 :
171 0 : bytes = (fh->fb.len - offsetof(struct ovl_fb, fid));
172 0 : real = exportfs_decode_fh(mnt, (struct fid *)fh->fb.fid,
173 0 : bytes >> 2, (int)fh->fb.type,
174 : connected ? ovl_acceptable : NULL, mnt);
175 0 : if (IS_ERR(real)) {
176 : /*
177 : * Treat stale file handle to lower file as "origin unknown".
178 : * upper file handle could become stale when upper file is
179 : * unlinked and this information is needed to handle stale
180 : * index entries correctly.
181 : */
182 0 : if (real == ERR_PTR(-ESTALE) &&
183 0 : !(fh->fb.flags & OVL_FH_FLAG_PATH_UPPER))
184 0 : real = NULL;
185 0 : return real;
186 : }
187 :
188 0 : if (ovl_dentry_weird(real)) {
189 0 : dput(real);
190 0 : return NULL;
191 : }
192 :
193 : return real;
194 : }
195 :
196 4 : static bool ovl_is_opaquedir(struct super_block *sb, struct dentry *dentry)
197 : {
198 4 : return ovl_check_dir_xattr(sb, dentry, OVL_XATTR_OPAQUE);
199 : }
200 :
201 32 : static struct dentry *ovl_lookup_positive_unlocked(const char *name,
202 : struct dentry *base, int len,
203 : bool drop_negative)
204 : {
205 32 : struct dentry *ret = lookup_one_len_unlocked(name, base, len);
206 :
207 32 : if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
208 8 : if (drop_negative && ret->d_lockref.count == 1) {
209 6 : spin_lock(&ret->d_lock);
210 : /* Recheck condition under lock */
211 6 : if (d_is_negative(ret) && ret->d_lockref.count == 1)
212 6 : __d_drop(ret);
213 6 : spin_unlock(&ret->d_lock);
214 : }
215 8 : dput(ret);
216 8 : ret = ERR_PTR(-ENOENT);
217 : }
218 32 : return ret;
219 : }
220 :
221 32 : static int ovl_lookup_single(struct dentry *base, struct ovl_lookup_data *d,
222 : const char *name, unsigned int namelen,
223 : size_t prelen, const char *post,
224 : struct dentry **ret, bool drop_negative)
225 : {
226 32 : struct dentry *this;
227 32 : int err;
228 32 : bool last_element = !post[0];
229 :
230 32 : this = ovl_lookup_positive_unlocked(name, base, namelen, drop_negative);
231 32 : if (IS_ERR(this)) {
232 8 : err = PTR_ERR(this);
233 8 : this = NULL;
234 8 : if (err == -ENOENT || err == -ENAMETOOLONG)
235 8 : goto out;
236 0 : goto out_err;
237 : }
238 :
239 24 : if (ovl_dentry_weird(this)) {
240 : /* Don't support traversing automounts and other weirdness */
241 0 : err = -EREMOTE;
242 0 : goto out_err;
243 : }
244 24 : if (ovl_is_whiteout(this)) {
245 0 : d->stop = d->opaque = true;
246 0 : goto put_and_out;
247 : }
248 : /*
249 : * This dentry should be a regular file if previous layer lookup
250 : * found a metacopy dentry.
251 : */
252 24 : if (last_element && d->metacopy && !d_is_reg(this)) {
253 0 : d->stop = true;
254 0 : goto put_and_out;
255 : }
256 24 : if (!d_can_lookup(this)) {
257 16 : if (d->is_dir || !last_element) {
258 0 : d->stop = true;
259 0 : goto put_and_out;
260 : }
261 16 : err = ovl_check_metacopy_xattr(OVL_FS(d->sb), this);
262 16 : if (err < 0)
263 0 : goto out_err;
264 :
265 16 : d->metacopy = err;
266 16 : d->stop = !d->metacopy;
267 16 : if (!d->metacopy || d->last)
268 16 : goto out;
269 : } else {
270 8 : if (ovl_lookup_trap_inode(d->sb, this)) {
271 : /* Caught in a trap of overlapping layers */
272 0 : err = -ELOOP;
273 0 : goto out_err;
274 : }
275 :
276 8 : if (last_element)
277 8 : d->is_dir = true;
278 8 : if (d->last)
279 4 : goto out;
280 :
281 4 : if (ovl_is_opaquedir(d->sb, this)) {
282 0 : d->stop = true;
283 0 : if (last_element)
284 0 : d->opaque = true;
285 0 : goto out;
286 : }
287 : }
288 4 : err = ovl_check_redirect(this, d, prelen, post);
289 4 : if (err)
290 0 : goto out_err;
291 4 : out:
292 32 : *ret = this;
293 32 : return 0;
294 :
295 0 : put_and_out:
296 0 : dput(this);
297 0 : this = NULL;
298 0 : goto out;
299 :
300 0 : out_err:
301 0 : dput(this);
302 0 : return err;
303 : }
304 :
305 32 : static int ovl_lookup_layer(struct dentry *base, struct ovl_lookup_data *d,
306 : struct dentry **ret, bool drop_negative)
307 : {
308 : /* Counting down from the end, since the prefix can change */
309 32 : size_t rem = d->name.len - 1;
310 32 : struct dentry *dentry = NULL;
311 32 : int err;
312 :
313 32 : if (d->name.name[0] != '/')
314 32 : return ovl_lookup_single(base, d, d->name.name, d->name.len,
315 : 0, "", ret, drop_negative);
316 :
317 0 : while (!IS_ERR_OR_NULL(base) && d_can_lookup(base)) {
318 0 : const char *s = d->name.name + d->name.len - rem;
319 0 : const char *next = strchrnul(s, '/');
320 0 : size_t thislen = next - s;
321 0 : bool end = !next[0];
322 :
323 : /* Verify we did not go off the rails */
324 0 : if (WARN_ON(s[-1] != '/'))
325 : return -EIO;
326 :
327 0 : err = ovl_lookup_single(base, d, s, thislen,
328 0 : d->name.len - rem, next, &base,
329 : drop_negative);
330 0 : dput(dentry);
331 0 : if (err)
332 0 : return err;
333 0 : dentry = base;
334 0 : if (end)
335 : break;
336 :
337 0 : rem -= thislen + 1;
338 :
339 0 : if (WARN_ON(rem >= d->name.len))
340 : return -EIO;
341 : }
342 0 : *ret = dentry;
343 0 : return 0;
344 : }
345 :
346 :
347 0 : int ovl_check_origin_fh(struct ovl_fs *ofs, struct ovl_fh *fh, bool connected,
348 : struct dentry *upperdentry, struct ovl_path **stackp)
349 : {
350 0 : struct dentry *origin = NULL;
351 0 : int i;
352 :
353 0 : for (i = 1; i < ofs->numlayer; i++) {
354 : /*
355 : * If lower fs uuid is not unique among lower fs we cannot match
356 : * fh->uuid to layer.
357 : */
358 0 : if (ofs->layers[i].fsid &&
359 0 : ofs->layers[i].fs->bad_uuid)
360 0 : continue;
361 :
362 0 : origin = ovl_decode_real_fh(ofs, fh, ofs->layers[i].mnt,
363 : connected);
364 0 : if (origin)
365 : break;
366 : }
367 :
368 0 : if (!origin)
369 : return -ESTALE;
370 0 : else if (IS_ERR(origin))
371 0 : return PTR_ERR(origin);
372 :
373 0 : if (upperdentry && !ovl_is_whiteout(upperdentry) &&
374 0 : ((d_inode(origin)->i_mode ^ d_inode(upperdentry)->i_mode) & S_IFMT))
375 0 : goto invalid;
376 :
377 0 : if (!*stackp)
378 0 : *stackp = kmalloc(sizeof(struct ovl_path), GFP_KERNEL);
379 0 : if (!*stackp) {
380 0 : dput(origin);
381 0 : return -ENOMEM;
382 : }
383 0 : **stackp = (struct ovl_path){
384 : .dentry = origin,
385 0 : .layer = &ofs->layers[i]
386 : };
387 :
388 0 : return 0;
389 :
390 0 : invalid:
391 0 : pr_warn_ratelimited("invalid origin (%pd2, ftype=%x, origin ftype=%x).\n",
392 : upperdentry, d_inode(upperdentry)->i_mode & S_IFMT,
393 : d_inode(origin)->i_mode & S_IFMT);
394 0 : dput(origin);
395 0 : return -EIO;
396 : }
397 :
398 10 : static int ovl_check_origin(struct ovl_fs *ofs, struct dentry *upperdentry,
399 : struct ovl_path **stackp)
400 : {
401 10 : struct ovl_fh *fh = ovl_get_fh(ofs, upperdentry, OVL_XATTR_ORIGIN);
402 10 : int err;
403 :
404 10 : if (IS_ERR_OR_NULL(fh))
405 10 : return PTR_ERR(fh);
406 :
407 0 : err = ovl_check_origin_fh(ofs, fh, false, upperdentry, stackp);
408 0 : kfree(fh);
409 :
410 0 : if (err) {
411 0 : if (err == -ESTALE)
412 : return 0;
413 0 : return err;
414 : }
415 :
416 : return 0;
417 : }
418 :
419 : /*
420 : * Verify that @fh matches the file handle stored in xattr @name.
421 : * Return 0 on match, -ESTALE on mismatch, < 0 on error.
422 : */
423 0 : static int ovl_verify_fh(struct ovl_fs *ofs, struct dentry *dentry,
424 : enum ovl_xattr ox, const struct ovl_fh *fh)
425 : {
426 0 : struct ovl_fh *ofh = ovl_get_fh(ofs, dentry, ox);
427 0 : int err = 0;
428 :
429 0 : if (!ofh)
430 : return -ENODATA;
431 :
432 0 : if (IS_ERR(ofh))
433 0 : return PTR_ERR(ofh);
434 :
435 0 : if (fh->fb.len != ofh->fb.len || memcmp(&fh->fb, &ofh->fb, fh->fb.len))
436 0 : err = -ESTALE;
437 :
438 0 : kfree(ofh);
439 0 : return err;
440 : }
441 :
442 : /*
443 : * Verify that @real dentry matches the file handle stored in xattr @name.
444 : *
445 : * If @set is true and there is no stored file handle, encode @real and store
446 : * file handle in xattr @name.
447 : *
448 : * Return 0 on match, -ESTALE on mismatch, -ENODATA on no xattr, < 0 on error.
449 : */
450 0 : int ovl_verify_set_fh(struct ovl_fs *ofs, struct dentry *dentry,
451 : enum ovl_xattr ox, struct dentry *real, bool is_upper,
452 : bool set)
453 : {
454 0 : struct inode *inode;
455 0 : struct ovl_fh *fh;
456 0 : int err;
457 :
458 0 : fh = ovl_encode_real_fh(ofs, real, is_upper);
459 0 : err = PTR_ERR(fh);
460 0 : if (IS_ERR(fh)) {
461 0 : fh = NULL;
462 0 : goto fail;
463 : }
464 :
465 0 : err = ovl_verify_fh(ofs, dentry, ox, fh);
466 0 : if (set && err == -ENODATA)
467 0 : err = ovl_do_setxattr(ofs, dentry, ox, fh->buf, fh->fb.len);
468 0 : if (err)
469 0 : goto fail;
470 :
471 0 : out:
472 0 : kfree(fh);
473 0 : return err;
474 :
475 0 : fail:
476 0 : inode = d_inode(real);
477 0 : pr_warn_ratelimited("failed to verify %s (%pd2, ino=%lu, err=%i)\n",
478 : is_upper ? "upper" : "origin", real,
479 : inode ? inode->i_ino : 0, err);
480 0 : goto out;
481 : }
482 :
483 : /* Get upper dentry from index */
484 0 : struct dentry *ovl_index_upper(struct ovl_fs *ofs, struct dentry *index)
485 : {
486 0 : struct ovl_fh *fh;
487 0 : struct dentry *upper;
488 :
489 0 : if (!d_is_dir(index))
490 0 : return dget(index);
491 :
492 0 : fh = ovl_get_fh(ofs, index, OVL_XATTR_UPPER);
493 0 : if (IS_ERR_OR_NULL(fh))
494 0 : return ERR_CAST(fh);
495 :
496 0 : upper = ovl_decode_real_fh(ofs, fh, ovl_upper_mnt(ofs), true);
497 0 : kfree(fh);
498 :
499 0 : if (IS_ERR_OR_NULL(upper))
500 0 : return upper ?: ERR_PTR(-ESTALE);
501 :
502 0 : if (!d_is_dir(upper)) {
503 0 : pr_warn_ratelimited("invalid index upper (%pd2, upper=%pd2).\n",
504 : index, upper);
505 0 : dput(upper);
506 0 : return ERR_PTR(-EIO);
507 : }
508 :
509 : return upper;
510 : }
511 :
512 : /*
513 : * Verify that an index entry name matches the origin file handle stored in
514 : * OVL_XATTR_ORIGIN and that origin file handle can be decoded to lower path.
515 : * Return 0 on match, -ESTALE on mismatch or stale origin, < 0 on error.
516 : */
517 0 : int ovl_verify_index(struct ovl_fs *ofs, struct dentry *index)
518 : {
519 0 : struct ovl_fh *fh = NULL;
520 0 : size_t len;
521 0 : struct ovl_path origin = { };
522 0 : struct ovl_path *stack = &origin;
523 0 : struct dentry *upper = NULL;
524 0 : int err;
525 :
526 0 : if (!d_inode(index))
527 : return 0;
528 :
529 0 : err = -EINVAL;
530 0 : if (index->d_name.len < sizeof(struct ovl_fb)*2)
531 0 : goto fail;
532 :
533 0 : err = -ENOMEM;
534 0 : len = index->d_name.len / 2;
535 0 : fh = kzalloc(len + OVL_FH_WIRE_OFFSET, GFP_KERNEL);
536 0 : if (!fh)
537 0 : goto fail;
538 :
539 0 : err = -EINVAL;
540 0 : if (hex2bin(fh->buf, index->d_name.name, len))
541 0 : goto fail;
542 :
543 0 : err = ovl_check_fb_len(&fh->fb, len);
544 0 : if (err)
545 0 : goto fail;
546 :
547 : /*
548 : * Whiteout index entries are used as an indication that an exported
549 : * overlay file handle should be treated as stale (i.e. after unlink
550 : * of the overlay inode). These entries contain no origin xattr.
551 : */
552 0 : if (ovl_is_whiteout(index))
553 0 : goto out;
554 :
555 : /*
556 : * Verifying directory index entries are not stale is expensive, so
557 : * only verify stale dir index if NFS export is enabled.
558 : */
559 0 : if (d_is_dir(index) && !ofs->config.nfs_export)
560 0 : goto out;
561 :
562 : /*
563 : * Directory index entries should have 'upper' xattr pointing to the
564 : * real upper dir. Non-dir index entries are hardlinks to the upper
565 : * real inode. For non-dir index, we can read the copy up origin xattr
566 : * directly from the index dentry, but for dir index we first need to
567 : * decode the upper directory.
568 : */
569 0 : upper = ovl_index_upper(ofs, index);
570 0 : if (IS_ERR_OR_NULL(upper)) {
571 0 : err = PTR_ERR(upper);
572 : /*
573 : * Directory index entries with no 'upper' xattr need to be
574 : * removed. When dir index entry has a stale 'upper' xattr,
575 : * we assume that upper dir was removed and we treat the dir
576 : * index as orphan entry that needs to be whited out.
577 : */
578 0 : if (err == -ESTALE)
579 0 : goto orphan;
580 0 : else if (!err)
581 0 : err = -ESTALE;
582 0 : goto fail;
583 : }
584 :
585 0 : err = ovl_verify_fh(ofs, upper, OVL_XATTR_ORIGIN, fh);
586 0 : dput(upper);
587 0 : if (err)
588 0 : goto fail;
589 :
590 : /* Check if non-dir index is orphan and don't warn before cleaning it */
591 0 : if (!d_is_dir(index) && d_inode(index)->i_nlink == 1) {
592 0 : err = ovl_check_origin_fh(ofs, fh, false, index, &stack);
593 0 : if (err)
594 0 : goto fail;
595 :
596 0 : if (ovl_get_nlink(ofs, origin.dentry, index, 0) == 0)
597 0 : goto orphan;
598 : }
599 :
600 0 : out:
601 0 : dput(origin.dentry);
602 0 : kfree(fh);
603 0 : return err;
604 :
605 0 : fail:
606 0 : pr_warn_ratelimited("failed to verify index (%pd2, ftype=%x, err=%i)\n",
607 : index, d_inode(index)->i_mode & S_IFMT, err);
608 0 : goto out;
609 :
610 0 : orphan:
611 0 : pr_warn_ratelimited("orphan index entry (%pd2, ftype=%x, nlink=%u)\n",
612 : index, d_inode(index)->i_mode & S_IFMT,
613 : d_inode(index)->i_nlink);
614 0 : err = -ENOENT;
615 0 : goto out;
616 : }
617 :
618 0 : static int ovl_get_index_name_fh(struct ovl_fh *fh, struct qstr *name)
619 : {
620 0 : char *n, *s;
621 :
622 0 : n = kcalloc(fh->fb.len, 2, GFP_KERNEL);
623 0 : if (!n)
624 : return -ENOMEM;
625 :
626 0 : s = bin2hex(n, fh->buf, fh->fb.len);
627 0 : *name = (struct qstr) QSTR_INIT(n, s - n);
628 :
629 0 : return 0;
630 :
631 : }
632 :
633 : /*
634 : * Lookup in indexdir for the index entry of a lower real inode or a copy up
635 : * origin inode. The index entry name is the hex representation of the lower
636 : * inode file handle.
637 : *
638 : * If the index dentry in negative, then either no lower aliases have been
639 : * copied up yet, or aliases have been copied up in older kernels and are
640 : * not indexed.
641 : *
642 : * If the index dentry for a copy up origin inode is positive, but points
643 : * to an inode different than the upper inode, then either the upper inode
644 : * has been copied up and not indexed or it was indexed, but since then
645 : * index dir was cleared. Either way, that index cannot be used to indentify
646 : * the overlay inode.
647 : */
648 0 : int ovl_get_index_name(struct ovl_fs *ofs, struct dentry *origin,
649 : struct qstr *name)
650 : {
651 0 : struct ovl_fh *fh;
652 0 : int err;
653 :
654 0 : fh = ovl_encode_real_fh(ofs, origin, false);
655 0 : if (IS_ERR(fh))
656 0 : return PTR_ERR(fh);
657 :
658 0 : err = ovl_get_index_name_fh(fh, name);
659 :
660 0 : kfree(fh);
661 0 : return err;
662 : }
663 :
664 : /* Lookup index by file handle for NFS export */
665 0 : struct dentry *ovl_get_index_fh(struct ovl_fs *ofs, struct ovl_fh *fh)
666 : {
667 0 : struct dentry *index;
668 0 : struct qstr name;
669 0 : int err;
670 :
671 0 : err = ovl_get_index_name_fh(fh, &name);
672 0 : if (err)
673 0 : return ERR_PTR(err);
674 :
675 0 : index = lookup_positive_unlocked(name.name, ofs->indexdir, name.len);
676 0 : kfree(name.name);
677 0 : if (IS_ERR(index)) {
678 0 : if (PTR_ERR(index) == -ENOENT)
679 0 : index = NULL;
680 0 : return index;
681 : }
682 :
683 0 : if (ovl_is_whiteout(index))
684 : err = -ESTALE;
685 0 : else if (ovl_dentry_weird(index))
686 : err = -EIO;
687 : else
688 : return index;
689 :
690 0 : dput(index);
691 0 : return ERR_PTR(err);
692 : }
693 :
694 0 : struct dentry *ovl_lookup_index(struct ovl_fs *ofs, struct dentry *upper,
695 : struct dentry *origin, bool verify)
696 : {
697 0 : struct dentry *index;
698 0 : struct inode *inode;
699 0 : struct qstr name;
700 0 : bool is_dir = d_is_dir(origin);
701 0 : int err;
702 :
703 0 : err = ovl_get_index_name(ofs, origin, &name);
704 0 : if (err)
705 0 : return ERR_PTR(err);
706 :
707 0 : index = lookup_positive_unlocked(name.name, ofs->indexdir, name.len);
708 0 : if (IS_ERR(index)) {
709 0 : err = PTR_ERR(index);
710 0 : if (err == -ENOENT) {
711 0 : index = NULL;
712 0 : goto out;
713 : }
714 0 : pr_warn_ratelimited("failed inode index lookup (ino=%lu, key=%.*s, err=%i);\n"
715 : "overlayfs: mount with '-o index=off' to disable inodes index.\n",
716 : d_inode(origin)->i_ino, name.len, name.name,
717 : err);
718 0 : goto out;
719 : }
720 :
721 0 : inode = d_inode(index);
722 0 : if (ovl_is_whiteout(index) && !verify) {
723 : /*
724 : * When index lookup is called with !verify for decoding an
725 : * overlay file handle, a whiteout index implies that decode
726 : * should treat file handle as stale and no need to print a
727 : * warning about it.
728 : */
729 0 : dput(index);
730 0 : index = ERR_PTR(-ESTALE);
731 0 : goto out;
732 0 : } else if (ovl_dentry_weird(index) || ovl_is_whiteout(index) ||
733 0 : ((inode->i_mode ^ d_inode(origin)->i_mode) & S_IFMT)) {
734 : /*
735 : * Index should always be of the same file type as origin
736 : * except for the case of a whiteout index. A whiteout
737 : * index should only exist if all lower aliases have been
738 : * unlinked, which means that finding a lower origin on lookup
739 : * whose index is a whiteout should be treated as an error.
740 : */
741 0 : pr_warn_ratelimited("bad index found (index=%pd2, ftype=%x, origin ftype=%x).\n",
742 : index, d_inode(index)->i_mode & S_IFMT,
743 : d_inode(origin)->i_mode & S_IFMT);
744 0 : goto fail;
745 0 : } else if (is_dir && verify) {
746 0 : if (!upper) {
747 0 : pr_warn_ratelimited("suspected uncovered redirected dir found (origin=%pd2, index=%pd2).\n",
748 : origin, index);
749 0 : goto fail;
750 : }
751 :
752 : /* Verify that dir index 'upper' xattr points to upper dir */
753 0 : err = ovl_verify_upper(ofs, index, upper, false);
754 0 : if (err) {
755 0 : if (err == -ESTALE) {
756 0 : pr_warn_ratelimited("suspected multiply redirected dir found (upper=%pd2, origin=%pd2, index=%pd2).\n",
757 : upper, origin, index);
758 : }
759 0 : goto fail;
760 : }
761 0 : } else if (upper && d_inode(upper) != inode) {
762 0 : goto out_dput;
763 : }
764 0 : out:
765 0 : kfree(name.name);
766 0 : return index;
767 :
768 0 : out_dput:
769 0 : dput(index);
770 0 : index = NULL;
771 0 : goto out;
772 :
773 0 : fail:
774 0 : dput(index);
775 0 : index = ERR_PTR(-EIO);
776 0 : goto out;
777 : }
778 :
779 : /*
780 : * Returns next layer in stack starting from top.
781 : * Returns -1 if this is the last layer.
782 : */
783 0 : int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
784 : {
785 0 : struct ovl_entry *oe = dentry->d_fsdata;
786 :
787 0 : BUG_ON(idx < 0);
788 0 : if (idx == 0) {
789 0 : ovl_path_upper(dentry, path);
790 0 : if (path->dentry)
791 0 : return oe->numlower ? 1 : -1;
792 : idx++;
793 : }
794 0 : BUG_ON(idx > oe->numlower);
795 0 : path->dentry = oe->lowerstack[idx - 1].dentry;
796 0 : path->mnt = oe->lowerstack[idx - 1].layer->mnt;
797 :
798 0 : return (idx < oe->numlower) ? idx + 1 : -1;
799 : }
800 :
801 : /* Fix missing 'origin' xattr */
802 2 : static int ovl_fix_origin(struct ovl_fs *ofs, struct dentry *dentry,
803 : struct dentry *lower, struct dentry *upper)
804 : {
805 2 : int err;
806 :
807 2 : if (ovl_check_origin_xattr(ofs, upper))
808 : return 0;
809 :
810 2 : err = ovl_want_write(dentry);
811 2 : if (err)
812 : return err;
813 :
814 2 : err = ovl_set_origin(ofs, dentry, lower, upper);
815 2 : if (!err)
816 2 : err = ovl_set_impure(dentry->d_parent, upper->d_parent);
817 :
818 2 : ovl_drop_write(dentry);
819 2 : return err;
820 : }
821 :
822 22 : struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
823 : unsigned int flags)
824 : {
825 22 : struct ovl_entry *oe;
826 22 : const struct cred *old_cred;
827 22 : struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
828 22 : struct ovl_entry *poe = dentry->d_parent->d_fsdata;
829 22 : struct ovl_entry *roe = dentry->d_sb->s_root->d_fsdata;
830 22 : struct ovl_path *stack = NULL, *origin_path = NULL;
831 22 : struct dentry *upperdir, *upperdentry = NULL;
832 22 : struct dentry *origin = NULL;
833 22 : struct dentry *index = NULL;
834 22 : unsigned int ctr = 0;
835 22 : struct inode *inode = NULL;
836 22 : bool upperopaque = false;
837 22 : char *upperredirect = NULL;
838 22 : struct dentry *this;
839 22 : unsigned int i;
840 22 : int err;
841 22 : bool uppermetacopy = false;
842 44 : struct ovl_lookup_data d = {
843 : .sb = dentry->d_sb,
844 : .name = dentry->d_name,
845 : .is_dir = false,
846 : .opaque = false,
847 : .stop = false,
848 22 : .last = ofs->config.redirect_follow ? false : !poe->numlower,
849 : .redirect = NULL,
850 : .metacopy = false,
851 : };
852 :
853 22 : if (dentry->d_name.len > ofs->namelen)
854 22 : return ERR_PTR(-ENAMETOOLONG);
855 :
856 22 : old_cred = ovl_override_creds(dentry->d_sb);
857 22 : upperdir = ovl_dentry_upper(dentry->d_parent);
858 22 : if (upperdir) {
859 20 : err = ovl_lookup_layer(upperdir, &d, &upperdentry, true);
860 20 : if (err)
861 0 : goto out;
862 :
863 20 : if (upperdentry && upperdentry->d_flags & DCACHE_OP_REAL) {
864 0 : dput(upperdentry);
865 0 : err = -EREMOTE;
866 0 : goto out;
867 : }
868 20 : if (upperdentry && !d.is_dir) {
869 : /*
870 : * Lookup copy up origin by decoding origin file handle.
871 : * We may get a disconnected dentry, which is fine,
872 : * because we only need to hold the origin inode in
873 : * cache and use its inode number. We may even get a
874 : * connected dentry, that is not under any of the lower
875 : * layers root. That is also fine for using it's inode
876 : * number - it's the same as if we held a reference
877 : * to a dentry in lower layer that was moved under us.
878 : */
879 10 : err = ovl_check_origin(ofs, upperdentry, &origin_path);
880 10 : if (err)
881 0 : goto out_put_upper;
882 :
883 10 : if (d.metacopy)
884 0 : uppermetacopy = true;
885 : }
886 :
887 20 : if (d.redirect) {
888 0 : err = -ENOMEM;
889 0 : upperredirect = kstrdup(d.redirect, GFP_KERNEL);
890 0 : if (!upperredirect)
891 0 : goto out_put_upper;
892 0 : if (d.redirect[0] == '/')
893 0 : poe = roe;
894 : }
895 20 : upperopaque = d.opaque;
896 : }
897 :
898 22 : if (!d.stop && poe->numlower) {
899 12 : err = -ENOMEM;
900 12 : stack = kcalloc(ofs->numlayer - 1, sizeof(struct ovl_path),
901 : GFP_KERNEL);
902 12 : if (!stack)
903 0 : goto out_put_upper;
904 : }
905 :
906 28 : for (i = 0; !d.stop && i < poe->numlower; i++) {
907 12 : struct ovl_path lower = poe->lowerstack[i];
908 :
909 12 : if (!ofs->config.redirect_follow)
910 12 : d.last = i == poe->numlower - 1;
911 : else
912 0 : d.last = lower.layer->idx == roe->numlower;
913 :
914 12 : err = ovl_lookup_layer(lower.dentry, &d, &this, false);
915 12 : if (err)
916 0 : goto out_put;
917 :
918 12 : if (!this)
919 2 : continue;
920 :
921 10 : if ((uppermetacopy || d.metacopy) && !ofs->config.metacopy) {
922 0 : err = -EPERM;
923 0 : pr_warn_ratelimited("refusing to follow metacopy origin for (%pd2)\n", dentry);
924 0 : goto out_put;
925 : }
926 :
927 : /*
928 : * If no origin fh is stored in upper of a merge dir, store fh
929 : * of lower dir and set upper parent "impure".
930 : */
931 10 : if (upperdentry && !ctr && !ofs->noxattr && d.is_dir) {
932 2 : err = ovl_fix_origin(ofs, dentry, this, upperdentry);
933 2 : if (err) {
934 0 : dput(this);
935 0 : goto out_put;
936 : }
937 : }
938 :
939 : /*
940 : * When "verify_lower" feature is enabled, do not merge with a
941 : * lower dir that does not match a stored origin xattr. In any
942 : * case, only verified origin is used for index lookup.
943 : *
944 : * For non-dir dentry, if index=on, then ensure origin
945 : * matches the dentry found using path based lookup,
946 : * otherwise error out.
947 : */
948 10 : if (upperdentry && !ctr &&
949 2 : ((d.is_dir && ovl_verify_lower(dentry->d_sb)) ||
950 2 : (!d.is_dir && ofs->config.index && origin_path))) {
951 0 : err = ovl_verify_origin(ofs, upperdentry, this, false);
952 0 : if (err) {
953 0 : dput(this);
954 0 : if (d.is_dir)
955 : break;
956 0 : goto out_put;
957 : }
958 0 : origin = this;
959 : }
960 :
961 10 : if (d.metacopy && ctr) {
962 : /*
963 : * Do not store intermediate metacopy dentries in
964 : * lower chain, except top most lower metacopy dentry.
965 : * Continue the loop so that if there is an absolute
966 : * redirect on this dentry, poe can be reset to roe.
967 : */
968 0 : dput(this);
969 0 : this = NULL;
970 : } else {
971 10 : stack[ctr].dentry = this;
972 10 : stack[ctr].layer = lower.layer;
973 10 : ctr++;
974 : }
975 :
976 : /*
977 : * Following redirects can have security consequences: it's like
978 : * a symlink into the lower layer without the permission checks.
979 : * This is only a problem if the upper layer is untrusted (e.g
980 : * comes from an USB drive). This can allow a non-readable file
981 : * or directory to become readable.
982 : *
983 : * Only following redirects when redirects are enabled disables
984 : * this attack vector when not necessary.
985 : */
986 10 : err = -EPERM;
987 10 : if (d.redirect && !ofs->config.redirect_follow) {
988 0 : pr_warn_ratelimited("refusing to follow redirect for (%pd2)\n",
989 : dentry);
990 0 : goto out_put;
991 : }
992 :
993 10 : if (d.stop)
994 : break;
995 :
996 4 : if (d.redirect && d.redirect[0] == '/' && poe != roe) {
997 0 : poe = roe;
998 : /* Find the current layer on the root dentry */
999 0 : i = lower.layer->idx - 1;
1000 : }
1001 : }
1002 :
1003 : /*
1004 : * For regular non-metacopy upper dentries, there is no lower
1005 : * path based lookup, hence ctr will be zero. If a dentry is found
1006 : * using ORIGIN xattr on upper, install it in stack.
1007 : *
1008 : * For metacopy dentry, path based lookup will find lower dentries.
1009 : * Just make sure a corresponding data dentry has been found.
1010 : */
1011 22 : if (d.metacopy || (uppermetacopy && !ctr)) {
1012 0 : pr_warn_ratelimited("metacopy with no lower data found - abort lookup (%pd2)\n",
1013 : dentry);
1014 0 : err = -EIO;
1015 0 : goto out_put;
1016 22 : } else if (!d.is_dir && upperdentry && !ctr && origin_path) {
1017 0 : if (WARN_ON(stack != NULL)) {
1018 0 : err = -EIO;
1019 0 : goto out_put;
1020 : }
1021 0 : stack = origin_path;
1022 0 : ctr = 1;
1023 0 : origin = origin_path->dentry;
1024 0 : origin_path = NULL;
1025 : }
1026 :
1027 : /*
1028 : * Always lookup index if there is no-upperdentry.
1029 : *
1030 : * For the case of upperdentry, we have set origin by now if it
1031 : * needed to be set. There are basically three cases.
1032 : *
1033 : * For directories, lookup index by lower inode and verify it matches
1034 : * upper inode. We only trust dir index if we verified that lower dir
1035 : * matches origin, otherwise dir index entries may be inconsistent
1036 : * and we ignore them.
1037 : *
1038 : * For regular upper, we already set origin if upper had ORIGIN
1039 : * xattr. There is no verification though as there is no path
1040 : * based dentry lookup in lower in this case.
1041 : *
1042 : * For metacopy upper, we set a verified origin already if index
1043 : * is enabled and if upper had an ORIGIN xattr.
1044 : *
1045 : */
1046 22 : if (!upperdentry && ctr)
1047 8 : origin = stack[0].dentry;
1048 :
1049 22 : if (origin && ovl_indexdir(dentry->d_sb) &&
1050 0 : (!d.is_dir || ovl_index_all(dentry->d_sb))) {
1051 0 : index = ovl_lookup_index(ofs, upperdentry, origin, true);
1052 0 : if (IS_ERR(index)) {
1053 0 : err = PTR_ERR(index);
1054 0 : index = NULL;
1055 0 : goto out_put;
1056 : }
1057 : }
1058 :
1059 22 : oe = ovl_alloc_entry(ctr);
1060 22 : err = -ENOMEM;
1061 22 : if (!oe)
1062 0 : goto out_put;
1063 :
1064 22 : memcpy(oe->lowerstack, stack, sizeof(struct ovl_path) * ctr);
1065 22 : dentry->d_fsdata = oe;
1066 :
1067 22 : if (upperopaque)
1068 0 : ovl_dentry_set_opaque(dentry);
1069 :
1070 22 : if (upperdentry)
1071 14 : ovl_dentry_set_upper_alias(dentry);
1072 8 : else if (index) {
1073 0 : upperdentry = dget(index);
1074 0 : upperredirect = ovl_get_redirect_xattr(ofs, upperdentry, 0);
1075 0 : if (IS_ERR(upperredirect)) {
1076 0 : err = PTR_ERR(upperredirect);
1077 0 : upperredirect = NULL;
1078 0 : goto out_free_oe;
1079 : }
1080 0 : err = ovl_check_metacopy_xattr(ofs, upperdentry);
1081 0 : if (err < 0)
1082 0 : goto out_free_oe;
1083 0 : uppermetacopy = err;
1084 : }
1085 :
1086 22 : if (upperdentry || ctr) {
1087 44 : struct ovl_inode_params oip = {
1088 : .upperdentry = upperdentry,
1089 : .lowerpath = stack,
1090 : .index = index,
1091 : .numlower = ctr,
1092 : .redirect = upperredirect,
1093 0 : .lowerdata = (ctr > 1 && !d.is_dir) ?
1094 22 : stack[ctr - 1].dentry : NULL,
1095 : };
1096 :
1097 22 : inode = ovl_get_inode(dentry->d_sb, &oip);
1098 22 : err = PTR_ERR(inode);
1099 22 : if (IS_ERR(inode))
1100 0 : goto out_free_oe;
1101 22 : if (upperdentry && !uppermetacopy)
1102 14 : ovl_set_flag(OVL_UPPERDATA, inode);
1103 : }
1104 :
1105 22 : ovl_dentry_update_reval(dentry, upperdentry,
1106 : DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
1107 :
1108 22 : revert_creds(old_cred);
1109 22 : if (origin_path) {
1110 0 : dput(origin_path->dentry);
1111 0 : kfree(origin_path);
1112 : }
1113 22 : dput(index);
1114 22 : kfree(stack);
1115 22 : kfree(d.redirect);
1116 22 : return d_splice_alias(inode, dentry);
1117 :
1118 0 : out_free_oe:
1119 0 : dentry->d_fsdata = NULL;
1120 0 : kfree(oe);
1121 0 : out_put:
1122 0 : dput(index);
1123 0 : for (i = 0; i < ctr; i++)
1124 0 : dput(stack[i].dentry);
1125 0 : kfree(stack);
1126 0 : out_put_upper:
1127 0 : if (origin_path) {
1128 0 : dput(origin_path->dentry);
1129 0 : kfree(origin_path);
1130 : }
1131 0 : dput(upperdentry);
1132 0 : kfree(upperredirect);
1133 0 : out:
1134 0 : kfree(d.redirect);
1135 0 : revert_creds(old_cred);
1136 0 : return ERR_PTR(err);
1137 : }
1138 :
1139 0 : bool ovl_lower_positive(struct dentry *dentry)
1140 : {
1141 0 : struct ovl_entry *poe = dentry->d_parent->d_fsdata;
1142 0 : const struct qstr *name = &dentry->d_name;
1143 0 : const struct cred *old_cred;
1144 0 : unsigned int i;
1145 0 : bool positive = false;
1146 0 : bool done = false;
1147 :
1148 : /*
1149 : * If dentry is negative, then lower is positive iff this is a
1150 : * whiteout.
1151 : */
1152 0 : if (!dentry->d_inode)
1153 0 : return ovl_dentry_is_opaque(dentry);
1154 :
1155 : /* Negative upper -> positive lower */
1156 0 : if (!ovl_dentry_upper(dentry))
1157 : return true;
1158 :
1159 0 : old_cred = ovl_override_creds(dentry->d_sb);
1160 : /* Positive upper -> have to look up lower to see whether it exists */
1161 0 : for (i = 0; !done && !positive && i < poe->numlower; i++) {
1162 0 : struct dentry *this;
1163 0 : struct dentry *lowerdir = poe->lowerstack[i].dentry;
1164 :
1165 0 : this = lookup_positive_unlocked(name->name, lowerdir,
1166 0 : name->len);
1167 0 : if (IS_ERR(this)) {
1168 0 : switch (PTR_ERR(this)) {
1169 : case -ENOENT:
1170 : case -ENAMETOOLONG:
1171 : break;
1172 :
1173 0 : default:
1174 : /*
1175 : * Assume something is there, we just couldn't
1176 : * access it.
1177 : */
1178 0 : positive = true;
1179 0 : break;
1180 : }
1181 : } else {
1182 0 : positive = !ovl_is_whiteout(this);
1183 0 : done = true;
1184 0 : dput(this);
1185 : }
1186 : }
1187 0 : revert_creds(old_cred);
1188 :
1189 0 : return positive;
1190 : }
|
