1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * Overlayfs NFS export support. |
4 | * |
5 | * Amir Goldstein <amir73il@gmail.com> |
6 | * |
7 | * Copyright (C) 2017-2018 CTERA Networks. All Rights Reserved. |
8 | */ |
9 | |
10 | #include <linux/fs.h> |
11 | #include <linux/cred.h> |
12 | #include <linux/mount.h> |
13 | #include <linux/namei.h> |
14 | #include <linux/xattr.h> |
15 | #include <linux/exportfs.h> |
16 | #include <linux/ratelimit.h> |
17 | #include "overlayfs.h" |
18 | |
19 | static int ovl_encode_maybe_copy_up(struct dentry *dentry) |
20 | { |
21 | int err; |
22 | |
23 | if (ovl_dentry_upper(dentry)) |
24 | return 0; |
25 | |
26 | err = ovl_copy_up(dentry); |
27 | if (err) { |
28 | pr_warn_ratelimited("failed to copy up on encode (%pd2, err=%i)\n" , |
29 | dentry, err); |
30 | } |
31 | |
32 | return err; |
33 | } |
34 | |
35 | /* |
36 | * Before encoding a non-upper directory file handle from real layer N, we need |
37 | * to check if it will be possible to reconnect an overlay dentry from the real |
38 | * lower decoded dentry. This is done by following the overlay ancestry up to a |
39 | * "layer N connected" ancestor and verifying that all parents along the way are |
40 | * "layer N connectable". If an ancestor that is NOT "layer N connectable" is |
41 | * found, we need to copy up an ancestor, which is "layer N connectable", thus |
42 | * making that ancestor "layer N connected". For example: |
43 | * |
44 | * layer 1: /a |
45 | * layer 2: /a/b/c |
46 | * |
47 | * The overlay dentry /a is NOT "layer 2 connectable", because if dir /a is |
48 | * copied up and renamed, upper dir /a will be indexed by lower dir /a from |
49 | * layer 1. The dir /a from layer 2 will never be indexed, so the algorithm (*) |
50 | * in ovl_lookup_real_ancestor() will not be able to lookup a connected overlay |
51 | * dentry from the connected lower dentry /a/b/c. |
52 | * |
53 | * To avoid this problem on decode time, we need to copy up an ancestor of |
54 | * /a/b/c, which is "layer 2 connectable", on encode time. That ancestor is |
55 | * /a/b. After copy up (and index) of /a/b, it will become "layer 2 connected" |
56 | * and when the time comes to decode the file handle from lower dentry /a/b/c, |
57 | * ovl_lookup_real_ancestor() will find the indexed ancestor /a/b and decoding |
58 | * a connected overlay dentry will be accomplished. |
59 | * |
60 | * (*) the algorithm in ovl_lookup_real_ancestor() can be improved to lookup an |
61 | * entry /a in the lower layers above layer N and find the indexed dir /a from |
62 | * layer 1. If that improvement is made, then the check for "layer N connected" |
63 | * will need to verify there are no redirects in lower layers above N. In the |
64 | * example above, /a will be "layer 2 connectable". However, if layer 2 dir /a |
65 | * is a target of a layer 1 redirect, then /a will NOT be "layer 2 connectable": |
66 | * |
67 | * layer 1: /A (redirect = /a) |
68 | * layer 2: /a/b/c |
69 | */ |
70 | |
71 | /* Return the lowest layer for encoding a connectable file handle */ |
72 | static int ovl_connectable_layer(struct dentry *dentry) |
73 | { |
74 | struct ovl_entry *oe = OVL_E(dentry); |
75 | |
76 | /* We can get overlay root from root of any layer */ |
77 | if (dentry == dentry->d_sb->s_root) |
78 | return ovl_numlower(oe); |
79 | |
80 | /* |
81 | * If it's an unindexed merge dir, then it's not connectable with any |
82 | * lower layer |
83 | */ |
84 | if (ovl_dentry_upper(dentry) && |
85 | !ovl_test_flag(flag: OVL_INDEX, inode: d_inode(dentry))) |
86 | return 0; |
87 | |
88 | /* We can get upper/overlay path from indexed/lower dentry */ |
89 | return ovl_lowerstack(oe)->layer->idx; |
90 | } |
91 | |
92 | /* |
93 | * @dentry is "connected" if all ancestors up to root or a "connected" ancestor |
94 | * have the same uppermost lower layer as the origin's layer. We may need to |
95 | * copy up a "connectable" ancestor to make it "connected". A "connected" dentry |
96 | * cannot become non "connected", so cache positive result in dentry flags. |
97 | * |
98 | * Return the connected origin layer or < 0 on error. |
99 | */ |
100 | static int ovl_connect_layer(struct dentry *dentry) |
101 | { |
102 | struct dentry *next, *parent = NULL; |
103 | struct ovl_entry *oe = OVL_E(dentry); |
104 | int origin_layer; |
105 | int err = 0; |
106 | |
107 | if (WARN_ON(dentry == dentry->d_sb->s_root) || |
108 | WARN_ON(!ovl_dentry_lower(dentry))) |
109 | return -EIO; |
110 | |
111 | origin_layer = ovl_lowerstack(oe)->layer->idx; |
112 | if (ovl_dentry_test_flag(flag: OVL_E_CONNECTED, dentry)) |
113 | return origin_layer; |
114 | |
115 | /* Find the topmost origin layer connectable ancestor of @dentry */ |
116 | next = dget(dentry); |
117 | for (;;) { |
118 | parent = dget_parent(dentry: next); |
119 | if (WARN_ON(parent == next)) { |
120 | err = -EIO; |
121 | break; |
122 | } |
123 | |
124 | /* |
125 | * If @parent is not origin layer connectable, then copy up |
126 | * @next which is origin layer connectable and we are done. |
127 | */ |
128 | if (ovl_connectable_layer(dentry: parent) < origin_layer) { |
129 | err = ovl_encode_maybe_copy_up(dentry: next); |
130 | break; |
131 | } |
132 | |
133 | /* If @parent is connected or indexed we are done */ |
134 | if (ovl_dentry_test_flag(flag: OVL_E_CONNECTED, dentry: parent) || |
135 | ovl_test_flag(flag: OVL_INDEX, inode: d_inode(dentry: parent))) |
136 | break; |
137 | |
138 | dput(next); |
139 | next = parent; |
140 | } |
141 | |
142 | dput(parent); |
143 | dput(next); |
144 | |
145 | if (!err) |
146 | ovl_dentry_set_flag(flag: OVL_E_CONNECTED, dentry); |
147 | |
148 | return err ?: origin_layer; |
149 | } |
150 | |
151 | /* |
152 | * We only need to encode origin if there is a chance that the same object was |
153 | * encoded pre copy up and then we need to stay consistent with the same |
154 | * encoding also after copy up. If non-pure upper is not indexed, then it was |
155 | * copied up before NFS export was enabled. In that case we don't need to worry |
156 | * about staying consistent with pre copy up encoding and we encode an upper |
157 | * file handle. Overlay root dentry is a private case of non-indexed upper. |
158 | * |
159 | * The following table summarizes the different file handle encodings used for |
160 | * different overlay object types: |
161 | * |
162 | * Object type | Encoding |
163 | * -------------------------------- |
164 | * Pure upper | U |
165 | * Non-indexed upper | U |
166 | * Indexed upper | L (*) |
167 | * Non-upper | L (*) |
168 | * |
169 | * U = upper file handle |
170 | * L = lower file handle |
171 | * |
172 | * (*) Decoding a connected overlay dir from real lower dentry is not always |
173 | * possible when there are redirects in lower layers and non-indexed merge dirs. |
174 | * To mitigate those case, we may copy up the lower dir ancestor before encode |
175 | * of a decodable file handle for non-upper dir. |
176 | * |
177 | * Return 0 for upper file handle, > 0 for lower file handle or < 0 on error. |
178 | */ |
179 | static int ovl_check_encode_origin(struct dentry *dentry) |
180 | { |
181 | struct ovl_fs *ofs = OVL_FS(sb: dentry->d_sb); |
182 | bool decodable = ofs->config.nfs_export; |
183 | |
184 | /* Lower file handle for non-upper non-decodable */ |
185 | if (!ovl_dentry_upper(dentry) && !decodable) |
186 | return 1; |
187 | |
188 | /* Upper file handle for pure upper */ |
189 | if (!ovl_dentry_lower(dentry)) |
190 | return 0; |
191 | |
192 | /* |
193 | * Root is never indexed, so if there's an upper layer, encode upper for |
194 | * root. |
195 | */ |
196 | if (dentry == dentry->d_sb->s_root) |
197 | return 0; |
198 | |
199 | /* |
200 | * Upper decodable file handle for non-indexed upper. |
201 | */ |
202 | if (ovl_dentry_upper(dentry) && decodable && |
203 | !ovl_test_flag(flag: OVL_INDEX, inode: d_inode(dentry))) |
204 | return 0; |
205 | |
206 | /* |
207 | * Decoding a merge dir, whose origin's ancestor is under a redirected |
208 | * lower dir or under a non-indexed upper is not always possible. |
209 | * ovl_connect_layer() will try to make origin's layer "connected" by |
210 | * copying up a "connectable" ancestor. |
211 | */ |
212 | if (d_is_dir(dentry) && ovl_upper_mnt(ofs) && decodable) |
213 | return ovl_connect_layer(dentry); |
214 | |
215 | /* Lower file handle for indexed and non-upper dir/non-dir */ |
216 | return 1; |
217 | } |
218 | |
219 | static int ovl_dentry_to_fid(struct ovl_fs *ofs, struct dentry *dentry, |
220 | u32 *fid, int buflen) |
221 | { |
222 | struct ovl_fh *fh = NULL; |
223 | int err, enc_lower; |
224 | int len; |
225 | |
226 | /* |
227 | * Check if we should encode a lower or upper file handle and maybe |
228 | * copy up an ancestor to make lower file handle connectable. |
229 | */ |
230 | err = enc_lower = ovl_check_encode_origin(dentry); |
231 | if (enc_lower < 0) |
232 | goto fail; |
233 | |
234 | /* Encode an upper or lower file handle */ |
235 | fh = ovl_encode_real_fh(ofs, real: enc_lower ? ovl_dentry_lower(dentry) : |
236 | ovl_dentry_upper(dentry), is_upper: !enc_lower); |
237 | if (IS_ERR(ptr: fh)) |
238 | return PTR_ERR(ptr: fh); |
239 | |
240 | len = OVL_FH_LEN(fh); |
241 | if (len <= buflen) |
242 | memcpy(fid, fh, len); |
243 | err = len; |
244 | |
245 | out: |
246 | kfree(objp: fh); |
247 | return err; |
248 | |
249 | fail: |
250 | pr_warn_ratelimited("failed to encode file handle (%pd2, err=%i)\n" , |
251 | dentry, err); |
252 | goto out; |
253 | } |
254 | |
255 | static int ovl_encode_fh(struct inode *inode, u32 *fid, int *max_len, |
256 | struct inode *parent) |
257 | { |
258 | struct ovl_fs *ofs = OVL_FS(sb: inode->i_sb); |
259 | struct dentry *dentry; |
260 | int bytes, buflen = *max_len << 2; |
261 | |
262 | /* TODO: encode connectable file handles */ |
263 | if (parent) |
264 | return FILEID_INVALID; |
265 | |
266 | dentry = d_find_any_alias(inode); |
267 | if (!dentry) |
268 | return FILEID_INVALID; |
269 | |
270 | bytes = ovl_dentry_to_fid(ofs, dentry, fid, buflen); |
271 | dput(dentry); |
272 | if (bytes <= 0) |
273 | return FILEID_INVALID; |
274 | |
275 | *max_len = bytes >> 2; |
276 | if (bytes > buflen) |
277 | return FILEID_INVALID; |
278 | |
279 | return OVL_FILEID_V1; |
280 | } |
281 | |
282 | /* |
283 | * Find or instantiate an overlay dentry from real dentries and index. |
284 | */ |
285 | static struct dentry *ovl_obtain_alias(struct super_block *sb, |
286 | struct dentry *upper_alias, |
287 | struct ovl_path *lowerpath, |
288 | struct dentry *index) |
289 | { |
290 | struct dentry *lower = lowerpath ? lowerpath->dentry : NULL; |
291 | struct dentry *upper = upper_alias ?: index; |
292 | struct dentry *dentry; |
293 | struct inode *inode = NULL; |
294 | struct ovl_entry *oe; |
295 | struct ovl_inode_params oip = { |
296 | .index = index, |
297 | }; |
298 | |
299 | /* We get overlay directory dentries with ovl_lookup_real() */ |
300 | if (d_is_dir(dentry: upper ?: lower)) |
301 | return ERR_PTR(error: -EIO); |
302 | |
303 | oe = ovl_alloc_entry(numlower: !!lower); |
304 | if (!oe) |
305 | return ERR_PTR(error: -ENOMEM); |
306 | |
307 | oip.upperdentry = dget(dentry: upper); |
308 | if (lower) { |
309 | ovl_lowerstack(oe)->dentry = dget(dentry: lower); |
310 | ovl_lowerstack(oe)->layer = lowerpath->layer; |
311 | } |
312 | oip.oe = oe; |
313 | inode = ovl_get_inode(sb, oip: &oip); |
314 | if (IS_ERR(ptr: inode)) { |
315 | ovl_free_entry(oe); |
316 | dput(upper); |
317 | return ERR_CAST(ptr: inode); |
318 | } |
319 | |
320 | if (upper) |
321 | ovl_set_flag(flag: OVL_UPPERDATA, inode); |
322 | |
323 | dentry = d_find_any_alias(inode); |
324 | if (dentry) |
325 | goto out_iput; |
326 | |
327 | dentry = d_alloc_anon(inode->i_sb); |
328 | if (unlikely(!dentry)) |
329 | goto nomem; |
330 | |
331 | if (upper_alias) |
332 | ovl_dentry_set_upper_alias(dentry); |
333 | |
334 | ovl_dentry_init_reval(dentry, upperdentry: upper, oe: OVL_I_E(inode)); |
335 | |
336 | return d_instantiate_anon(dentry, inode); |
337 | |
338 | nomem: |
339 | dput(dentry); |
340 | dentry = ERR_PTR(error: -ENOMEM); |
341 | out_iput: |
342 | iput(inode); |
343 | return dentry; |
344 | } |
345 | |
346 | /* Get the upper or lower dentry in stack whose on layer @idx */ |
347 | static struct dentry *ovl_dentry_real_at(struct dentry *dentry, int idx) |
348 | { |
349 | struct ovl_entry *oe = OVL_E(dentry); |
350 | struct ovl_path *lowerstack = ovl_lowerstack(oe); |
351 | int i; |
352 | |
353 | if (!idx) |
354 | return ovl_dentry_upper(dentry); |
355 | |
356 | for (i = 0; i < ovl_numlower(oe); i++) { |
357 | if (lowerstack[i].layer->idx == idx) |
358 | return lowerstack[i].dentry; |
359 | } |
360 | |
361 | return NULL; |
362 | } |
363 | |
364 | /* |
365 | * Lookup a child overlay dentry to get a connected overlay dentry whose real |
366 | * dentry is @real. If @real is on upper layer, we lookup a child overlay |
367 | * dentry with the same name as the real dentry. Otherwise, we need to consult |
368 | * index for lookup. |
369 | */ |
370 | static struct dentry *ovl_lookup_real_one(struct dentry *connected, |
371 | struct dentry *real, |
372 | const struct ovl_layer *layer) |
373 | { |
374 | struct inode *dir = d_inode(dentry: connected); |
375 | struct dentry *this, *parent = NULL; |
376 | struct name_snapshot name; |
377 | int err; |
378 | |
379 | /* |
380 | * Lookup child overlay dentry by real name. The dir mutex protects us |
381 | * from racing with overlay rename. If the overlay dentry that is above |
382 | * real has already been moved to a parent that is not under the |
383 | * connected overlay dir, we return -ECHILD and restart the lookup of |
384 | * connected real path from the top. |
385 | */ |
386 | inode_lock_nested(inode: dir, subclass: I_MUTEX_PARENT); |
387 | err = -ECHILD; |
388 | parent = dget_parent(dentry: real); |
389 | if (ovl_dentry_real_at(dentry: connected, idx: layer->idx) != parent) |
390 | goto fail; |
391 | |
392 | /* |
393 | * We also need to take a snapshot of real dentry name to protect us |
394 | * from racing with underlying layer rename. In this case, we don't |
395 | * care about returning ESTALE, only from dereferencing a free name |
396 | * pointer because we hold no lock on the real dentry. |
397 | */ |
398 | take_dentry_name_snapshot(&name, real); |
399 | /* |
400 | * No idmap handling here: it's an internal lookup. Could skip |
401 | * permission checking altogether, but for now just use non-idmap |
402 | * transformed ids. |
403 | */ |
404 | this = lookup_one_len(name.name.name, connected, name.name.len); |
405 | release_dentry_name_snapshot(&name); |
406 | err = PTR_ERR(ptr: this); |
407 | if (IS_ERR(ptr: this)) { |
408 | goto fail; |
409 | } else if (!this || !this->d_inode) { |
410 | dput(this); |
411 | err = -ENOENT; |
412 | goto fail; |
413 | } else if (ovl_dentry_real_at(dentry: this, idx: layer->idx) != real) { |
414 | dput(this); |
415 | err = -ESTALE; |
416 | goto fail; |
417 | } |
418 | |
419 | out: |
420 | dput(parent); |
421 | inode_unlock(inode: dir); |
422 | return this; |
423 | |
424 | fail: |
425 | pr_warn_ratelimited("failed to lookup one by real (%pd2, layer=%d, connected=%pd2, err=%i)\n" , |
426 | real, layer->idx, connected, err); |
427 | this = ERR_PTR(error: err); |
428 | goto out; |
429 | } |
430 | |
431 | static struct dentry *ovl_lookup_real(struct super_block *sb, |
432 | struct dentry *real, |
433 | const struct ovl_layer *layer); |
434 | |
435 | /* |
436 | * Lookup an indexed or hashed overlay dentry by real inode. |
437 | */ |
438 | static struct dentry *ovl_lookup_real_inode(struct super_block *sb, |
439 | struct dentry *real, |
440 | const struct ovl_layer *layer) |
441 | { |
442 | struct ovl_fs *ofs = OVL_FS(sb); |
443 | struct dentry *index = NULL; |
444 | struct dentry *this = NULL; |
445 | struct inode *inode; |
446 | |
447 | /* |
448 | * Decoding upper dir from index is expensive, so first try to lookup |
449 | * overlay dentry in inode/dcache. |
450 | */ |
451 | inode = ovl_lookup_inode(sb, real, is_upper: !layer->idx); |
452 | if (IS_ERR(ptr: inode)) |
453 | return ERR_CAST(ptr: inode); |
454 | if (inode) { |
455 | this = d_find_any_alias(inode); |
456 | iput(inode); |
457 | } |
458 | |
459 | /* |
460 | * For decoded lower dir file handle, lookup index by origin to check |
461 | * if lower dir was copied up and and/or removed. |
462 | */ |
463 | if (!this && layer->idx && ofs->indexdir && !WARN_ON(!d_is_dir(real))) { |
464 | index = ovl_lookup_index(ofs, NULL, origin: real, verify: false); |
465 | if (IS_ERR(ptr: index)) |
466 | return index; |
467 | } |
468 | |
469 | /* Get connected upper overlay dir from index */ |
470 | if (index) { |
471 | struct dentry *upper = ovl_index_upper(ofs, index, connected: true); |
472 | |
473 | dput(index); |
474 | if (IS_ERR_OR_NULL(ptr: upper)) |
475 | return upper; |
476 | |
477 | /* |
478 | * ovl_lookup_real() in lower layer may call recursively once to |
479 | * ovl_lookup_real() in upper layer. The first level call walks |
480 | * back lower parents to the topmost indexed parent. The second |
481 | * recursive call walks back from indexed upper to the topmost |
482 | * connected/hashed upper parent (or up to root). |
483 | */ |
484 | this = ovl_lookup_real(sb, real: upper, layer: &ofs->layers[0]); |
485 | dput(upper); |
486 | } |
487 | |
488 | if (IS_ERR_OR_NULL(ptr: this)) |
489 | return this; |
490 | |
491 | if (ovl_dentry_real_at(dentry: this, idx: layer->idx) != real) { |
492 | dput(this); |
493 | this = ERR_PTR(error: -EIO); |
494 | } |
495 | |
496 | return this; |
497 | } |
498 | |
499 | /* |
500 | * Lookup an indexed or hashed overlay dentry, whose real dentry is an |
501 | * ancestor of @real. |
502 | */ |
503 | static struct dentry *ovl_lookup_real_ancestor(struct super_block *sb, |
504 | struct dentry *real, |
505 | const struct ovl_layer *layer) |
506 | { |
507 | struct dentry *next, *parent = NULL; |
508 | struct dentry *ancestor = ERR_PTR(error: -EIO); |
509 | |
510 | if (real == layer->mnt->mnt_root) |
511 | return dget(dentry: sb->s_root); |
512 | |
513 | /* Find the topmost indexed or hashed ancestor */ |
514 | next = dget(dentry: real); |
515 | for (;;) { |
516 | parent = dget_parent(dentry: next); |
517 | |
518 | /* |
519 | * Lookup a matching overlay dentry in inode/dentry |
520 | * cache or in index by real inode. |
521 | */ |
522 | ancestor = ovl_lookup_real_inode(sb, real: next, layer); |
523 | if (ancestor) |
524 | break; |
525 | |
526 | if (parent == layer->mnt->mnt_root) { |
527 | ancestor = dget(dentry: sb->s_root); |
528 | break; |
529 | } |
530 | |
531 | /* |
532 | * If @real has been moved out of the layer root directory, |
533 | * we will eventully hit the real fs root. This cannot happen |
534 | * by legit overlay rename, so we return error in that case. |
535 | */ |
536 | if (parent == next) { |
537 | ancestor = ERR_PTR(error: -EXDEV); |
538 | break; |
539 | } |
540 | |
541 | dput(next); |
542 | next = parent; |
543 | } |
544 | |
545 | dput(parent); |
546 | dput(next); |
547 | |
548 | return ancestor; |
549 | } |
550 | |
551 | /* |
552 | * Lookup a connected overlay dentry whose real dentry is @real. |
553 | * If @real is on upper layer, we lookup a child overlay dentry with the same |
554 | * path the real dentry. Otherwise, we need to consult index for lookup. |
555 | */ |
556 | static struct dentry *ovl_lookup_real(struct super_block *sb, |
557 | struct dentry *real, |
558 | const struct ovl_layer *layer) |
559 | { |
560 | struct dentry *connected; |
561 | int err = 0; |
562 | |
563 | connected = ovl_lookup_real_ancestor(sb, real, layer); |
564 | if (IS_ERR(ptr: connected)) |
565 | return connected; |
566 | |
567 | while (!err) { |
568 | struct dentry *next, *this; |
569 | struct dentry *parent = NULL; |
570 | struct dentry *real_connected = ovl_dentry_real_at(dentry: connected, |
571 | idx: layer->idx); |
572 | |
573 | if (real_connected == real) |
574 | break; |
575 | |
576 | /* Find the topmost dentry not yet connected */ |
577 | next = dget(dentry: real); |
578 | for (;;) { |
579 | parent = dget_parent(dentry: next); |
580 | |
581 | if (parent == real_connected) |
582 | break; |
583 | |
584 | /* |
585 | * If real has been moved out of 'real_connected', |
586 | * we will not find 'real_connected' and hit the layer |
587 | * root. In that case, we need to restart connecting. |
588 | * This game can go on forever in the worst case. We |
589 | * may want to consider taking s_vfs_rename_mutex if |
590 | * this happens more than once. |
591 | */ |
592 | if (parent == layer->mnt->mnt_root) { |
593 | dput(connected); |
594 | connected = dget(dentry: sb->s_root); |
595 | break; |
596 | } |
597 | |
598 | /* |
599 | * If real file has been moved out of the layer root |
600 | * directory, we will eventully hit the real fs root. |
601 | * This cannot happen by legit overlay rename, so we |
602 | * return error in that case. |
603 | */ |
604 | if (parent == next) { |
605 | err = -EXDEV; |
606 | break; |
607 | } |
608 | |
609 | dput(next); |
610 | next = parent; |
611 | } |
612 | |
613 | if (!err) { |
614 | this = ovl_lookup_real_one(connected, real: next, layer); |
615 | if (IS_ERR(ptr: this)) |
616 | err = PTR_ERR(ptr: this); |
617 | |
618 | /* |
619 | * Lookup of child in overlay can fail when racing with |
620 | * overlay rename of child away from 'connected' parent. |
621 | * In this case, we need to restart the lookup from the |
622 | * top, because we cannot trust that 'real_connected' is |
623 | * still an ancestor of 'real'. There is a good chance |
624 | * that the renamed overlay ancestor is now in cache, so |
625 | * ovl_lookup_real_ancestor() will find it and we can |
626 | * continue to connect exactly from where lookup failed. |
627 | */ |
628 | if (err == -ECHILD) { |
629 | this = ovl_lookup_real_ancestor(sb, real, |
630 | layer); |
631 | err = PTR_ERR_OR_ZERO(ptr: this); |
632 | } |
633 | if (!err) { |
634 | dput(connected); |
635 | connected = this; |
636 | } |
637 | } |
638 | |
639 | dput(parent); |
640 | dput(next); |
641 | } |
642 | |
643 | if (err) |
644 | goto fail; |
645 | |
646 | return connected; |
647 | |
648 | fail: |
649 | pr_warn_ratelimited("failed to lookup by real (%pd2, layer=%d, connected=%pd2, err=%i)\n" , |
650 | real, layer->idx, connected, err); |
651 | dput(connected); |
652 | return ERR_PTR(error: err); |
653 | } |
654 | |
655 | /* |
656 | * Get an overlay dentry from upper/lower real dentries and index. |
657 | */ |
658 | static struct dentry *ovl_get_dentry(struct super_block *sb, |
659 | struct dentry *upper, |
660 | struct ovl_path *lowerpath, |
661 | struct dentry *index) |
662 | { |
663 | struct ovl_fs *ofs = OVL_FS(sb); |
664 | const struct ovl_layer *layer = upper ? &ofs->layers[0] : lowerpath->layer; |
665 | struct dentry *real = upper ?: (index ?: lowerpath->dentry); |
666 | |
667 | /* |
668 | * Obtain a disconnected overlay dentry from a non-dir real dentry |
669 | * and index. |
670 | */ |
671 | if (!d_is_dir(dentry: real)) |
672 | return ovl_obtain_alias(sb, upper_alias: upper, lowerpath, index); |
673 | |
674 | /* Removed empty directory? */ |
675 | if ((real->d_flags & DCACHE_DISCONNECTED) || d_unhashed(dentry: real)) |
676 | return ERR_PTR(error: -ENOENT); |
677 | |
678 | /* |
679 | * If real dentry is connected and hashed, get a connected overlay |
680 | * dentry whose real dentry is @real. |
681 | */ |
682 | return ovl_lookup_real(sb, real, layer); |
683 | } |
684 | |
685 | static struct dentry *ovl_upper_fh_to_d(struct super_block *sb, |
686 | struct ovl_fh *fh) |
687 | { |
688 | struct ovl_fs *ofs = OVL_FS(sb); |
689 | struct dentry *dentry; |
690 | struct dentry *upper; |
691 | |
692 | if (!ovl_upper_mnt(ofs)) |
693 | return ERR_PTR(error: -EACCES); |
694 | |
695 | upper = ovl_decode_real_fh(ofs, fh, mnt: ovl_upper_mnt(ofs), connected: true); |
696 | if (IS_ERR_OR_NULL(ptr: upper)) |
697 | return upper; |
698 | |
699 | dentry = ovl_get_dentry(sb, upper, NULL, NULL); |
700 | dput(upper); |
701 | |
702 | return dentry; |
703 | } |
704 | |
705 | static struct dentry *ovl_lower_fh_to_d(struct super_block *sb, |
706 | struct ovl_fh *fh) |
707 | { |
708 | struct ovl_fs *ofs = OVL_FS(sb); |
709 | struct ovl_path origin = { }; |
710 | struct ovl_path *stack = &origin; |
711 | struct dentry *dentry = NULL; |
712 | struct dentry *index = NULL; |
713 | struct inode *inode; |
714 | int err; |
715 | |
716 | /* First lookup overlay inode in inode cache by origin fh */ |
717 | err = ovl_check_origin_fh(ofs, fh, connected: false, NULL, stackp: &stack); |
718 | if (err) |
719 | return ERR_PTR(error: err); |
720 | |
721 | if (!d_is_dir(dentry: origin.dentry) || |
722 | !(origin.dentry->d_flags & DCACHE_DISCONNECTED)) { |
723 | inode = ovl_lookup_inode(sb, real: origin.dentry, is_upper: false); |
724 | err = PTR_ERR(ptr: inode); |
725 | if (IS_ERR(ptr: inode)) |
726 | goto out_err; |
727 | if (inode) { |
728 | dentry = d_find_any_alias(inode); |
729 | iput(inode); |
730 | if (dentry) |
731 | goto out; |
732 | } |
733 | } |
734 | |
735 | /* Then lookup indexed upper/whiteout by origin fh */ |
736 | if (ofs->indexdir) { |
737 | index = ovl_get_index_fh(ofs, fh); |
738 | err = PTR_ERR(ptr: index); |
739 | if (IS_ERR(ptr: index)) { |
740 | index = NULL; |
741 | goto out_err; |
742 | } |
743 | } |
744 | |
745 | /* Then try to get a connected upper dir by index */ |
746 | if (index && d_is_dir(dentry: index)) { |
747 | struct dentry *upper = ovl_index_upper(ofs, index, connected: true); |
748 | |
749 | err = PTR_ERR(ptr: upper); |
750 | if (IS_ERR_OR_NULL(ptr: upper)) |
751 | goto out_err; |
752 | |
753 | dentry = ovl_get_dentry(sb, upper, NULL, NULL); |
754 | dput(upper); |
755 | goto out; |
756 | } |
757 | |
758 | /* Find origin.dentry again with ovl_acceptable() layer check */ |
759 | if (d_is_dir(dentry: origin.dentry)) { |
760 | dput(origin.dentry); |
761 | origin.dentry = NULL; |
762 | err = ovl_check_origin_fh(ofs, fh, connected: true, NULL, stackp: &stack); |
763 | if (err) |
764 | goto out_err; |
765 | } |
766 | if (index) { |
767 | err = ovl_verify_origin(ofs, upper: index, origin: origin.dentry, set: false); |
768 | if (err) |
769 | goto out_err; |
770 | } |
771 | |
772 | /* Get a connected non-upper dir or disconnected non-dir */ |
773 | dentry = ovl_get_dentry(sb, NULL, lowerpath: &origin, index); |
774 | |
775 | out: |
776 | dput(origin.dentry); |
777 | dput(index); |
778 | return dentry; |
779 | |
780 | out_err: |
781 | dentry = ERR_PTR(error: err); |
782 | goto out; |
783 | } |
784 | |
785 | static struct ovl_fh *ovl_fid_to_fh(struct fid *fid, int buflen, int fh_type) |
786 | { |
787 | struct ovl_fh *fh; |
788 | |
789 | /* If on-wire inner fid is aligned - nothing to do */ |
790 | if (fh_type == OVL_FILEID_V1) |
791 | return (struct ovl_fh *)fid; |
792 | |
793 | if (fh_type != OVL_FILEID_V0) |
794 | return ERR_PTR(error: -EINVAL); |
795 | |
796 | if (buflen <= OVL_FH_WIRE_OFFSET) |
797 | return ERR_PTR(error: -EINVAL); |
798 | |
799 | fh = kzalloc(size: buflen, GFP_KERNEL); |
800 | if (!fh) |
801 | return ERR_PTR(error: -ENOMEM); |
802 | |
803 | /* Copy unaligned inner fh into aligned buffer */ |
804 | memcpy(fh->buf, fid, buflen - OVL_FH_WIRE_OFFSET); |
805 | return fh; |
806 | } |
807 | |
808 | static struct dentry *ovl_fh_to_dentry(struct super_block *sb, struct fid *fid, |
809 | int fh_len, int fh_type) |
810 | { |
811 | struct dentry *dentry = NULL; |
812 | struct ovl_fh *fh = NULL; |
813 | int len = fh_len << 2; |
814 | unsigned int flags = 0; |
815 | int err; |
816 | |
817 | fh = ovl_fid_to_fh(fid, buflen: len, fh_type); |
818 | err = PTR_ERR(ptr: fh); |
819 | if (IS_ERR(ptr: fh)) |
820 | goto out_err; |
821 | |
822 | err = ovl_check_fh_len(fh, fh_len: len); |
823 | if (err) |
824 | goto out_err; |
825 | |
826 | flags = fh->fb.flags; |
827 | dentry = (flags & OVL_FH_FLAG_PATH_UPPER) ? |
828 | ovl_upper_fh_to_d(sb, fh) : |
829 | ovl_lower_fh_to_d(sb, fh); |
830 | err = PTR_ERR(ptr: dentry); |
831 | if (IS_ERR(ptr: dentry) && err != -ESTALE) |
832 | goto out_err; |
833 | |
834 | out: |
835 | /* We may have needed to re-align OVL_FILEID_V0 */ |
836 | if (!IS_ERR_OR_NULL(ptr: fh) && fh != (void *)fid) |
837 | kfree(objp: fh); |
838 | |
839 | return dentry; |
840 | |
841 | out_err: |
842 | pr_warn_ratelimited("failed to decode file handle (len=%d, type=%d, flags=%x, err=%i)\n" , |
843 | fh_len, fh_type, flags, err); |
844 | dentry = ERR_PTR(error: err); |
845 | goto out; |
846 | } |
847 | |
848 | static struct dentry *ovl_fh_to_parent(struct super_block *sb, struct fid *fid, |
849 | int fh_len, int fh_type) |
850 | { |
851 | pr_warn_ratelimited("connectable file handles not supported; use 'no_subtree_check' exportfs option.\n" ); |
852 | return ERR_PTR(error: -EACCES); |
853 | } |
854 | |
855 | static int ovl_get_name(struct dentry *parent, char *name, |
856 | struct dentry *child) |
857 | { |
858 | /* |
859 | * ovl_fh_to_dentry() returns connected dir overlay dentries and |
860 | * ovl_fh_to_parent() is not implemented, so we should not get here. |
861 | */ |
862 | WARN_ON_ONCE(1); |
863 | return -EIO; |
864 | } |
865 | |
866 | static struct dentry *ovl_get_parent(struct dentry *dentry) |
867 | { |
868 | /* |
869 | * ovl_fh_to_dentry() returns connected dir overlay dentries, so we |
870 | * should not get here. |
871 | */ |
872 | WARN_ON_ONCE(1); |
873 | return ERR_PTR(error: -EIO); |
874 | } |
875 | |
876 | const struct export_operations ovl_export_operations = { |
877 | .encode_fh = ovl_encode_fh, |
878 | .fh_to_dentry = ovl_fh_to_dentry, |
879 | .fh_to_parent = ovl_fh_to_parent, |
880 | .get_name = ovl_get_name, |
881 | .get_parent = ovl_get_parent, |
882 | }; |
883 | |
884 | /* encode_fh() encodes non-decodable file handles with nfs_export=off */ |
885 | const struct export_operations ovl_export_fid_operations = { |
886 | .encode_fh = ovl_encode_fh, |
887 | }; |
888 | |